]>
Commit | Line | Data |
---|---|---|
c1d7c514 | 1 | // SPDX-License-Identifier: GPL-2.0 |
6cbd5570 CM |
2 | /* |
3 | * Copyright (C) 2007 Oracle. All rights reserved. | |
6cbd5570 CM |
4 | */ |
5 | ||
7999096f | 6 | #include <crypto/hash.h> |
8f18cf13 | 7 | #include <linux/kernel.h> |
065631f6 | 8 | #include <linux/bio.h> |
348332e0 | 9 | #include <linux/blk-cgroup.h> |
f2eb0a24 | 10 | #include <linux/file.h> |
39279cc3 CM |
11 | #include <linux/fs.h> |
12 | #include <linux/pagemap.h> | |
13 | #include <linux/highmem.h> | |
14 | #include <linux/time.h> | |
15 | #include <linux/init.h> | |
16 | #include <linux/string.h> | |
39279cc3 | 17 | #include <linux/backing-dev.h> |
39279cc3 | 18 | #include <linux/writeback.h> |
39279cc3 | 19 | #include <linux/compat.h> |
5103e947 | 20 | #include <linux/xattr.h> |
33268eaf | 21 | #include <linux/posix_acl.h> |
d899e052 | 22 | #include <linux/falloc.h> |
5a0e3ad6 | 23 | #include <linux/slab.h> |
7a36ddec | 24 | #include <linux/ratelimit.h> |
55e301fd | 25 | #include <linux/btrfs.h> |
53b381b3 | 26 | #include <linux/blkdev.h> |
f23b5a59 | 27 | #include <linux/posix_acl_xattr.h> |
e2e40f2c | 28 | #include <linux/uio.h> |
69fe2d75 | 29 | #include <linux/magic.h> |
ae5e165d | 30 | #include <linux/iversion.h> |
ed46ff3d | 31 | #include <linux/swap.h> |
f8e66081 | 32 | #include <linux/migrate.h> |
b1c16ac9 | 33 | #include <linux/sched/mm.h> |
f85781fb | 34 | #include <linux/iomap.h> |
92d32170 | 35 | #include <asm/unaligned.h> |
14605409 | 36 | #include <linux/fsverity.h> |
602cbe91 | 37 | #include "misc.h" |
39279cc3 CM |
38 | #include "ctree.h" |
39 | #include "disk-io.h" | |
40 | #include "transaction.h" | |
41 | #include "btrfs_inode.h" | |
39279cc3 | 42 | #include "print-tree.h" |
e6dcd2dc | 43 | #include "ordered-data.h" |
95819c05 | 44 | #include "xattr.h" |
e02119d5 | 45 | #include "tree-log.h" |
4a54c8c1 | 46 | #include "volumes.h" |
c8b97818 | 47 | #include "compression.h" |
b4ce94de | 48 | #include "locking.h" |
dc89e982 | 49 | #include "free-space-cache.h" |
63541927 | 50 | #include "props.h" |
31193213 | 51 | #include "qgroup.h" |
86736342 | 52 | #include "delalloc-space.h" |
aac0023c | 53 | #include "block-group.h" |
467dc47e | 54 | #include "space-info.h" |
d8e3fb10 | 55 | #include "zoned.h" |
b945a463 | 56 | #include "subpage.h" |
26c2c454 | 57 | #include "inode-item.h" |
39279cc3 CM |
58 | |
59 | struct btrfs_iget_args { | |
0202e83f | 60 | u64 ino; |
39279cc3 CM |
61 | struct btrfs_root *root; |
62 | }; | |
63 | ||
f28a4928 | 64 | struct btrfs_dio_data { |
f85781fb GR |
65 | ssize_t submitted; |
66 | struct extent_changeset *data_reserved; | |
f5585f4f FM |
67 | bool data_space_reserved; |
68 | bool nocow_done; | |
f28a4928 FM |
69 | }; |
70 | ||
a3e171a0 CH |
71 | struct btrfs_dio_private { |
72 | struct inode *inode; | |
73 | ||
74 | /* | |
75 | * Since DIO can use anonymous page, we cannot use page_offset() to | |
76 | * grab the file offset, thus need a dedicated member for file offset. | |
77 | */ | |
78 | u64 file_offset; | |
79 | /* Used for bio::bi_size */ | |
80 | u32 bytes; | |
81 | ||
82 | /* | |
83 | * References to this structure. There is one reference per in-flight | |
84 | * bio plus one while we're still setting up. | |
85 | */ | |
86 | refcount_t refs; | |
87 | ||
a3e171a0 | 88 | /* Array of checksums */ |
642c5d34 CH |
89 | u8 *csums; |
90 | ||
91 | /* This must be last */ | |
92 | struct bio bio; | |
a3e171a0 CH |
93 | }; |
94 | ||
642c5d34 CH |
95 | static struct bio_set btrfs_dio_bioset; |
96 | ||
88d2beec FM |
97 | struct btrfs_rename_ctx { |
98 | /* Output field. Stores the index number of the old directory entry. */ | |
99 | u64 index; | |
100 | }; | |
101 | ||
6e1d5dcc AD |
102 | static const struct inode_operations btrfs_dir_inode_operations; |
103 | static const struct inode_operations btrfs_symlink_inode_operations; | |
6e1d5dcc AD |
104 | static const struct inode_operations btrfs_special_inode_operations; |
105 | static const struct inode_operations btrfs_file_inode_operations; | |
7f09410b | 106 | static const struct address_space_operations btrfs_aops; |
828c0950 | 107 | static const struct file_operations btrfs_dir_file_operations; |
39279cc3 CM |
108 | |
109 | static struct kmem_cache *btrfs_inode_cachep; | |
110 | struct kmem_cache *btrfs_trans_handle_cachep; | |
39279cc3 | 111 | struct kmem_cache *btrfs_path_cachep; |
dc89e982 | 112 | struct kmem_cache *btrfs_free_space_cachep; |
3acd4850 | 113 | struct kmem_cache *btrfs_free_space_bitmap_cachep; |
39279cc3 | 114 | |
3972f260 | 115 | static int btrfs_setsize(struct inode *inode, struct iattr *attr); |
213e8c55 | 116 | static int btrfs_truncate(struct inode *inode, bool skip_writeback); |
6e26c442 | 117 | static noinline int cow_file_range(struct btrfs_inode *inode, |
771ed689 | 118 | struct page *locked_page, |
74e9194a | 119 | u64 start, u64 end, int *page_started, |
898793d9 NA |
120 | unsigned long *nr_written, int unlock, |
121 | u64 *done_offset); | |
4b67c11d NB |
122 | static struct extent_map *create_io_em(struct btrfs_inode *inode, u64 start, |
123 | u64 len, u64 orig_start, u64 block_start, | |
6f9994db LB |
124 | u64 block_len, u64 orig_block_len, |
125 | u64 ram_bytes, int compress_type, | |
126 | int type); | |
7b128766 | 127 | |
a14b78ad GR |
128 | /* |
129 | * btrfs_inode_lock - lock inode i_rwsem based on arguments passed | |
130 | * | |
131 | * ilock_flags can have the following bit set: | |
132 | * | |
133 | * BTRFS_ILOCK_SHARED - acquire a shared lock on the inode | |
134 | * BTRFS_ILOCK_TRY - try to acquire the lock, if fails on first attempt | |
135 | * return -EAGAIN | |
8318ba79 | 136 | * BTRFS_ILOCK_MMAP - acquire a write lock on the i_mmap_lock |
a14b78ad GR |
137 | */ |
138 | int btrfs_inode_lock(struct inode *inode, unsigned int ilock_flags) | |
139 | { | |
140 | if (ilock_flags & BTRFS_ILOCK_SHARED) { | |
141 | if (ilock_flags & BTRFS_ILOCK_TRY) { | |
142 | if (!inode_trylock_shared(inode)) | |
143 | return -EAGAIN; | |
144 | else | |
145 | return 0; | |
146 | } | |
147 | inode_lock_shared(inode); | |
148 | } else { | |
149 | if (ilock_flags & BTRFS_ILOCK_TRY) { | |
150 | if (!inode_trylock(inode)) | |
151 | return -EAGAIN; | |
152 | else | |
153 | return 0; | |
154 | } | |
155 | inode_lock(inode); | |
156 | } | |
8318ba79 JB |
157 | if (ilock_flags & BTRFS_ILOCK_MMAP) |
158 | down_write(&BTRFS_I(inode)->i_mmap_lock); | |
a14b78ad GR |
159 | return 0; |
160 | } | |
161 | ||
162 | /* | |
163 | * btrfs_inode_unlock - unock inode i_rwsem | |
164 | * | |
165 | * ilock_flags should contain the same bits set as passed to btrfs_inode_lock() | |
166 | * to decide whether the lock acquired is shared or exclusive. | |
167 | */ | |
168 | void btrfs_inode_unlock(struct inode *inode, unsigned int ilock_flags) | |
169 | { | |
8318ba79 JB |
170 | if (ilock_flags & BTRFS_ILOCK_MMAP) |
171 | up_write(&BTRFS_I(inode)->i_mmap_lock); | |
a14b78ad GR |
172 | if (ilock_flags & BTRFS_ILOCK_SHARED) |
173 | inode_unlock_shared(inode); | |
174 | else | |
175 | inode_unlock(inode); | |
176 | } | |
177 | ||
52427260 QW |
178 | /* |
179 | * Cleanup all submitted ordered extents in specified range to handle errors | |
52042d8e | 180 | * from the btrfs_run_delalloc_range() callback. |
52427260 QW |
181 | * |
182 | * NOTE: caller must ensure that when an error happens, it can not call | |
183 | * extent_clear_unlock_delalloc() to clear both the bits EXTENT_DO_ACCOUNTING | |
184 | * and EXTENT_DELALLOC simultaneously, because that causes the reserved metadata | |
185 | * to be released, which we want to happen only when finishing the ordered | |
d1051d6e | 186 | * extent (btrfs_finish_ordered_io()). |
52427260 | 187 | */ |
64e1db56 | 188 | static inline void btrfs_cleanup_ordered_extents(struct btrfs_inode *inode, |
d1051d6e NB |
189 | struct page *locked_page, |
190 | u64 offset, u64 bytes) | |
52427260 | 191 | { |
63d71450 NA |
192 | unsigned long index = offset >> PAGE_SHIFT; |
193 | unsigned long end_index = (offset + bytes - 1) >> PAGE_SHIFT; | |
99826e4c | 194 | u64 page_start, page_end; |
63d71450 NA |
195 | struct page *page; |
196 | ||
99826e4c NA |
197 | if (locked_page) { |
198 | page_start = page_offset(locked_page); | |
199 | page_end = page_start + PAGE_SIZE - 1; | |
200 | } | |
201 | ||
63d71450 | 202 | while (index <= end_index) { |
968f2566 QW |
203 | /* |
204 | * For locked page, we will call end_extent_writepage() on it | |
205 | * in run_delalloc_range() for the error handling. That | |
206 | * end_extent_writepage() function will call | |
207 | * btrfs_mark_ordered_io_finished() to clear page Ordered and | |
208 | * run the ordered extent accounting. | |
209 | * | |
210 | * Here we can't just clear the Ordered bit, or | |
211 | * btrfs_mark_ordered_io_finished() would skip the accounting | |
212 | * for the page range, and the ordered extent will never finish. | |
213 | */ | |
99826e4c | 214 | if (locked_page && index == (page_start >> PAGE_SHIFT)) { |
968f2566 QW |
215 | index++; |
216 | continue; | |
217 | } | |
64e1db56 | 218 | page = find_get_page(inode->vfs_inode.i_mapping, index); |
63d71450 NA |
219 | index++; |
220 | if (!page) | |
221 | continue; | |
968f2566 QW |
222 | |
223 | /* | |
224 | * Here we just clear all Ordered bits for every page in the | |
711f447b | 225 | * range, then btrfs_mark_ordered_io_finished() will handle |
968f2566 QW |
226 | * the ordered extent accounting for the range. |
227 | */ | |
b945a463 QW |
228 | btrfs_page_clamp_clear_ordered(inode->root->fs_info, page, |
229 | offset, bytes); | |
63d71450 NA |
230 | put_page(page); |
231 | } | |
d1051d6e | 232 | |
99826e4c NA |
233 | if (locked_page) { |
234 | /* The locked page covers the full range, nothing needs to be done */ | |
235 | if (bytes + offset <= page_start + PAGE_SIZE) | |
236 | return; | |
237 | /* | |
238 | * In case this page belongs to the delalloc range being | |
239 | * instantiated then skip it, since the first page of a range is | |
240 | * going to be properly cleaned up by the caller of | |
241 | * run_delalloc_range | |
242 | */ | |
243 | if (page_start >= offset && page_end <= (offset + bytes - 1)) { | |
244 | bytes = offset + bytes - page_offset(locked_page) - PAGE_SIZE; | |
245 | offset = page_offset(locked_page) + PAGE_SIZE; | |
246 | } | |
d1051d6e NB |
247 | } |
248 | ||
711f447b | 249 | return btrfs_mark_ordered_io_finished(inode, NULL, offset, bytes, false); |
52427260 QW |
250 | } |
251 | ||
48a3b636 | 252 | static int btrfs_dirty_inode(struct inode *inode); |
7b128766 | 253 | |
f34f57a3 | 254 | static int btrfs_init_inode_security(struct btrfs_trans_handle *trans, |
3538d68d | 255 | struct btrfs_new_inode_args *args) |
0279b4cd JO |
256 | { |
257 | int err; | |
258 | ||
3538d68d OS |
259 | if (args->default_acl) { |
260 | err = __btrfs_set_acl(trans, args->inode, args->default_acl, | |
261 | ACL_TYPE_DEFAULT); | |
262 | if (err) | |
263 | return err; | |
264 | } | |
265 | if (args->acl) { | |
266 | err = __btrfs_set_acl(trans, args->inode, args->acl, ACL_TYPE_ACCESS); | |
267 | if (err) | |
268 | return err; | |
269 | } | |
270 | if (!args->default_acl && !args->acl) | |
271 | cache_no_acl(args->inode); | |
272 | return btrfs_xattr_security_init(trans, args->inode, args->dir, | |
273 | &args->dentry->d_name); | |
0279b4cd JO |
274 | } |
275 | ||
c8b97818 CM |
276 | /* |
277 | * this does all the hard work for inserting an inline extent into | |
278 | * the btree. The caller should have done a btrfs_drop_extents so that | |
279 | * no overlapping inline items exist in the btree | |
280 | */ | |
40f76580 | 281 | static int insert_inline_extent(struct btrfs_trans_handle *trans, |
8dd9872d OS |
282 | struct btrfs_path *path, |
283 | struct btrfs_inode *inode, bool extent_inserted, | |
284 | size_t size, size_t compressed_size, | |
fe3f566c | 285 | int compress_type, |
d9496e8a OS |
286 | struct page **compressed_pages, |
287 | bool update_i_size) | |
c8b97818 | 288 | { |
8dd9872d | 289 | struct btrfs_root *root = inode->root; |
c8b97818 CM |
290 | struct extent_buffer *leaf; |
291 | struct page *page = NULL; | |
292 | char *kaddr; | |
293 | unsigned long ptr; | |
294 | struct btrfs_file_extent_item *ei; | |
c8b97818 CM |
295 | int ret; |
296 | size_t cur_size = size; | |
d9496e8a | 297 | u64 i_size; |
c8b97818 | 298 | |
982f1f5d JJB |
299 | ASSERT((compressed_size > 0 && compressed_pages) || |
300 | (compressed_size == 0 && !compressed_pages)); | |
301 | ||
fe3f566c | 302 | if (compressed_size && compressed_pages) |
c8b97818 | 303 | cur_size = compressed_size; |
c8b97818 | 304 | |
1acae57b FDBM |
305 | if (!extent_inserted) { |
306 | struct btrfs_key key; | |
307 | size_t datasize; | |
c8b97818 | 308 | |
8dd9872d OS |
309 | key.objectid = btrfs_ino(inode); |
310 | key.offset = 0; | |
962a298f | 311 | key.type = BTRFS_EXTENT_DATA_KEY; |
c8b97818 | 312 | |
1acae57b | 313 | datasize = btrfs_file_extent_calc_inline_size(cur_size); |
1acae57b FDBM |
314 | ret = btrfs_insert_empty_item(trans, root, path, &key, |
315 | datasize); | |
79b4f4c6 | 316 | if (ret) |
1acae57b | 317 | goto fail; |
c8b97818 CM |
318 | } |
319 | leaf = path->nodes[0]; | |
320 | ei = btrfs_item_ptr(leaf, path->slots[0], | |
321 | struct btrfs_file_extent_item); | |
322 | btrfs_set_file_extent_generation(leaf, ei, trans->transid); | |
323 | btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE); | |
324 | btrfs_set_file_extent_encryption(leaf, ei, 0); | |
325 | btrfs_set_file_extent_other_encoding(leaf, ei, 0); | |
326 | btrfs_set_file_extent_ram_bytes(leaf, ei, size); | |
327 | ptr = btrfs_file_extent_inline_start(ei); | |
328 | ||
261507a0 | 329 | if (compress_type != BTRFS_COMPRESS_NONE) { |
c8b97818 CM |
330 | struct page *cpage; |
331 | int i = 0; | |
d397712b | 332 | while (compressed_size > 0) { |
c8b97818 | 333 | cpage = compressed_pages[i]; |
5b050f04 | 334 | cur_size = min_t(unsigned long, compressed_size, |
09cbfeaf | 335 | PAGE_SIZE); |
c8b97818 | 336 | |
4cb2e5e8 | 337 | kaddr = kmap_local_page(cpage); |
c8b97818 | 338 | write_extent_buffer(leaf, kaddr, ptr, cur_size); |
4cb2e5e8 | 339 | kunmap_local(kaddr); |
c8b97818 CM |
340 | |
341 | i++; | |
342 | ptr += cur_size; | |
343 | compressed_size -= cur_size; | |
344 | } | |
345 | btrfs_set_file_extent_compression(leaf, ei, | |
261507a0 | 346 | compress_type); |
c8b97818 | 347 | } else { |
8dd9872d | 348 | page = find_get_page(inode->vfs_inode.i_mapping, 0); |
c8b97818 | 349 | btrfs_set_file_extent_compression(leaf, ei, 0); |
4cb2e5e8 | 350 | kaddr = kmap_local_page(page); |
8dd9872d | 351 | write_extent_buffer(leaf, kaddr, ptr, size); |
4cb2e5e8 | 352 | kunmap_local(kaddr); |
09cbfeaf | 353 | put_page(page); |
c8b97818 CM |
354 | } |
355 | btrfs_mark_buffer_dirty(leaf); | |
1acae57b | 356 | btrfs_release_path(path); |
c8b97818 | 357 | |
9ddc959e JB |
358 | /* |
359 | * We align size to sectorsize for inline extents just for simplicity | |
360 | * sake. | |
361 | */ | |
8dd9872d OS |
362 | ret = btrfs_inode_set_file_extent_range(inode, 0, |
363 | ALIGN(size, root->fs_info->sectorsize)); | |
9ddc959e JB |
364 | if (ret) |
365 | goto fail; | |
366 | ||
c2167754 | 367 | /* |
d9496e8a OS |
368 | * We're an inline extent, so nobody can extend the file past i_size |
369 | * without locking a page we already have locked. | |
c2167754 | 370 | * |
d9496e8a OS |
371 | * We must do any i_size and inode updates before we unlock the pages. |
372 | * Otherwise we could end up racing with unlink. | |
c2167754 | 373 | */ |
d9496e8a OS |
374 | i_size = i_size_read(&inode->vfs_inode); |
375 | if (update_i_size && size > i_size) { | |
376 | i_size_write(&inode->vfs_inode, size); | |
377 | i_size = size; | |
378 | } | |
379 | inode->disk_i_size = i_size; | |
8dd9872d | 380 | |
c8b97818 | 381 | fail: |
79b4f4c6 | 382 | return ret; |
c8b97818 CM |
383 | } |
384 | ||
385 | ||
386 | /* | |
387 | * conditionally insert an inline extent into the file. This | |
388 | * does the checks required to make sure the data is small enough | |
389 | * to fit as an inline extent. | |
390 | */ | |
8dd9872d OS |
391 | static noinline int cow_file_range_inline(struct btrfs_inode *inode, u64 size, |
392 | size_t compressed_size, | |
00361589 | 393 | int compress_type, |
d9496e8a OS |
394 | struct page **compressed_pages, |
395 | bool update_i_size) | |
c8b97818 | 396 | { |
5893dfb9 | 397 | struct btrfs_drop_extents_args drop_args = { 0 }; |
a0349401 | 398 | struct btrfs_root *root = inode->root; |
0b246afa | 399 | struct btrfs_fs_info *fs_info = root->fs_info; |
00361589 | 400 | struct btrfs_trans_handle *trans; |
8dd9872d | 401 | u64 data_len = (compressed_size ?: size); |
c8b97818 | 402 | int ret; |
1acae57b | 403 | struct btrfs_path *path; |
c8b97818 | 404 | |
8dd9872d OS |
405 | /* |
406 | * We can create an inline extent if it ends at or beyond the current | |
407 | * i_size, is no larger than a sector (decompressed), and the (possibly | |
408 | * compressed) data fits in a leaf and the configured maximum inline | |
409 | * size. | |
410 | */ | |
411 | if (size < i_size_read(&inode->vfs_inode) || | |
412 | size > fs_info->sectorsize || | |
0b246afa | 413 | data_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info) || |
8dd9872d | 414 | data_len > fs_info->max_inline) |
c8b97818 | 415 | return 1; |
c8b97818 | 416 | |
1acae57b FDBM |
417 | path = btrfs_alloc_path(); |
418 | if (!path) | |
419 | return -ENOMEM; | |
420 | ||
00361589 | 421 | trans = btrfs_join_transaction(root); |
1acae57b FDBM |
422 | if (IS_ERR(trans)) { |
423 | btrfs_free_path(path); | |
00361589 | 424 | return PTR_ERR(trans); |
1acae57b | 425 | } |
a0349401 | 426 | trans->block_rsv = &inode->block_rsv; |
00361589 | 427 | |
5893dfb9 | 428 | drop_args.path = path; |
8dd9872d OS |
429 | drop_args.start = 0; |
430 | drop_args.end = fs_info->sectorsize; | |
5893dfb9 FM |
431 | drop_args.drop_cache = true; |
432 | drop_args.replace_extent = true; | |
8dd9872d | 433 | drop_args.extent_item_size = btrfs_file_extent_calc_inline_size(data_len); |
5893dfb9 | 434 | ret = btrfs_drop_extents(trans, root, inode, &drop_args); |
00361589 | 435 | if (ret) { |
66642832 | 436 | btrfs_abort_transaction(trans, ret); |
00361589 JB |
437 | goto out; |
438 | } | |
c8b97818 | 439 | |
8dd9872d OS |
440 | ret = insert_inline_extent(trans, path, inode, drop_args.extent_inserted, |
441 | size, compressed_size, compress_type, | |
d9496e8a | 442 | compressed_pages, update_i_size); |
2adcac1a | 443 | if (ret && ret != -ENOSPC) { |
66642832 | 444 | btrfs_abort_transaction(trans, ret); |
00361589 | 445 | goto out; |
2adcac1a | 446 | } else if (ret == -ENOSPC) { |
00361589 JB |
447 | ret = 1; |
448 | goto out; | |
79787eaa | 449 | } |
2adcac1a | 450 | |
8dd9872d | 451 | btrfs_update_inode_bytes(inode, size, drop_args.bytes_found); |
9a56fcd1 | 452 | ret = btrfs_update_inode(trans, root, inode); |
2766ff61 FM |
453 | if (ret && ret != -ENOSPC) { |
454 | btrfs_abort_transaction(trans, ret); | |
455 | goto out; | |
456 | } else if (ret == -ENOSPC) { | |
457 | ret = 1; | |
458 | goto out; | |
459 | } | |
460 | ||
23e3337f | 461 | btrfs_set_inode_full_sync(inode); |
00361589 | 462 | out: |
94ed938a QW |
463 | /* |
464 | * Don't forget to free the reserved space, as for inlined extent | |
465 | * it won't count as data extent, free them directly here. | |
466 | * And at reserve time, it's always aligned to page size, so | |
467 | * just free one page here. | |
468 | */ | |
a0349401 | 469 | btrfs_qgroup_free_data(inode, NULL, 0, PAGE_SIZE); |
1acae57b | 470 | btrfs_free_path(path); |
3a45bb20 | 471 | btrfs_end_transaction(trans); |
00361589 | 472 | return ret; |
c8b97818 CM |
473 | } |
474 | ||
771ed689 CM |
475 | struct async_extent { |
476 | u64 start; | |
477 | u64 ram_size; | |
478 | u64 compressed_size; | |
479 | struct page **pages; | |
480 | unsigned long nr_pages; | |
261507a0 | 481 | int compress_type; |
771ed689 CM |
482 | struct list_head list; |
483 | }; | |
484 | ||
97db1204 | 485 | struct async_chunk { |
771ed689 | 486 | struct inode *inode; |
771ed689 CM |
487 | struct page *locked_page; |
488 | u64 start; | |
489 | u64 end; | |
bf9486d6 | 490 | blk_opf_t write_flags; |
771ed689 | 491 | struct list_head extents; |
ec39f769 | 492 | struct cgroup_subsys_state *blkcg_css; |
771ed689 | 493 | struct btrfs_work work; |
9e895a8f | 494 | struct async_cow *async_cow; |
771ed689 CM |
495 | }; |
496 | ||
97db1204 | 497 | struct async_cow { |
97db1204 NB |
498 | atomic_t num_chunks; |
499 | struct async_chunk chunks[]; | |
771ed689 CM |
500 | }; |
501 | ||
97db1204 | 502 | static noinline int add_async_extent(struct async_chunk *cow, |
771ed689 CM |
503 | u64 start, u64 ram_size, |
504 | u64 compressed_size, | |
505 | struct page **pages, | |
261507a0 LZ |
506 | unsigned long nr_pages, |
507 | int compress_type) | |
771ed689 CM |
508 | { |
509 | struct async_extent *async_extent; | |
510 | ||
511 | async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS); | |
79787eaa | 512 | BUG_ON(!async_extent); /* -ENOMEM */ |
771ed689 CM |
513 | async_extent->start = start; |
514 | async_extent->ram_size = ram_size; | |
515 | async_extent->compressed_size = compressed_size; | |
516 | async_extent->pages = pages; | |
517 | async_extent->nr_pages = nr_pages; | |
261507a0 | 518 | async_extent->compress_type = compress_type; |
771ed689 CM |
519 | list_add_tail(&async_extent->list, &cow->extents); |
520 | return 0; | |
521 | } | |
522 | ||
42c16da6 QW |
523 | /* |
524 | * Check if the inode needs to be submitted to compression, based on mount | |
525 | * options, defragmentation, properties or heuristics. | |
526 | */ | |
808a1292 NB |
527 | static inline int inode_need_compress(struct btrfs_inode *inode, u64 start, |
528 | u64 end) | |
f79707b0 | 529 | { |
808a1292 | 530 | struct btrfs_fs_info *fs_info = inode->root->fs_info; |
f79707b0 | 531 | |
e6f9d696 | 532 | if (!btrfs_inode_can_compress(inode)) { |
42c16da6 QW |
533 | WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG), |
534 | KERN_ERR "BTRFS: unexpected compression for ino %llu\n", | |
808a1292 | 535 | btrfs_ino(inode)); |
42c16da6 QW |
536 | return 0; |
537 | } | |
0cf9b244 QW |
538 | /* |
539 | * Special check for subpage. | |
540 | * | |
541 | * We lock the full page then run each delalloc range in the page, thus | |
542 | * for the following case, we will hit some subpage specific corner case: | |
543 | * | |
544 | * 0 32K 64K | |
545 | * | |///////| |///////| | |
546 | * \- A \- B | |
547 | * | |
548 | * In above case, both range A and range B will try to unlock the full | |
549 | * page [0, 64K), causing the one finished later will have page | |
550 | * unlocked already, triggering various page lock requirement BUG_ON()s. | |
551 | * | |
552 | * So here we add an artificial limit that subpage compression can only | |
553 | * if the range is fully page aligned. | |
554 | * | |
555 | * In theory we only need to ensure the first page is fully covered, but | |
556 | * the tailing partial page will be locked until the full compression | |
557 | * finishes, delaying the write of other range. | |
558 | * | |
559 | * TODO: Make btrfs_run_delalloc_range() to lock all delalloc range | |
560 | * first to prevent any submitted async extent to unlock the full page. | |
561 | * By this, we can ensure for subpage case that only the last async_cow | |
562 | * will unlock the full page. | |
563 | */ | |
564 | if (fs_info->sectorsize < PAGE_SIZE) { | |
1280d2d1 FK |
565 | if (!PAGE_ALIGNED(start) || |
566 | !PAGE_ALIGNED(end + 1)) | |
0cf9b244 QW |
567 | return 0; |
568 | } | |
569 | ||
f79707b0 | 570 | /* force compress */ |
0b246afa | 571 | if (btrfs_test_opt(fs_info, FORCE_COMPRESS)) |
f79707b0 | 572 | return 1; |
eec63c65 | 573 | /* defrag ioctl */ |
808a1292 | 574 | if (inode->defrag_compress) |
eec63c65 | 575 | return 1; |
f79707b0 | 576 | /* bad compression ratios */ |
808a1292 | 577 | if (inode->flags & BTRFS_INODE_NOCOMPRESS) |
f79707b0 | 578 | return 0; |
0b246afa | 579 | if (btrfs_test_opt(fs_info, COMPRESS) || |
808a1292 NB |
580 | inode->flags & BTRFS_INODE_COMPRESS || |
581 | inode->prop_compress) | |
582 | return btrfs_compress_heuristic(&inode->vfs_inode, start, end); | |
f79707b0 WS |
583 | return 0; |
584 | } | |
585 | ||
6158e1ce | 586 | static inline void inode_should_defrag(struct btrfs_inode *inode, |
558732df | 587 | u64 start, u64 end, u64 num_bytes, u32 small_write) |
26d30f85 AJ |
588 | { |
589 | /* If this is a small write inside eof, kick off a defrag */ | |
590 | if (num_bytes < small_write && | |
6158e1ce | 591 | (start > 0 || end + 1 < inode->disk_i_size)) |
558732df | 592 | btrfs_add_inode_defrag(NULL, inode, small_write); |
26d30f85 AJ |
593 | } |
594 | ||
d352ac68 | 595 | /* |
771ed689 CM |
596 | * we create compressed extents in two phases. The first |
597 | * phase compresses a range of pages that have already been | |
598 | * locked (both pages and state bits are locked). | |
c8b97818 | 599 | * |
771ed689 CM |
600 | * This is done inside an ordered work queue, and the compression |
601 | * is spread across many cpus. The actual IO submission is step | |
602 | * two, and the ordered work queue takes care of making sure that | |
603 | * happens in the same order things were put onto the queue by | |
604 | * writepages and friends. | |
c8b97818 | 605 | * |
771ed689 CM |
606 | * If this code finds it can't get good compression, it puts an |
607 | * entry onto the work queue to write the uncompressed bytes. This | |
608 | * makes sure that both compressed inodes and uncompressed inodes | |
b2570314 AB |
609 | * are written in the same order that the flusher thread sent them |
610 | * down. | |
d352ac68 | 611 | */ |
ac3e9933 | 612 | static noinline int compress_file_range(struct async_chunk *async_chunk) |
b888db2b | 613 | { |
1368c6da | 614 | struct inode *inode = async_chunk->inode; |
0b246afa | 615 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
0b246afa | 616 | u64 blocksize = fs_info->sectorsize; |
1368c6da NB |
617 | u64 start = async_chunk->start; |
618 | u64 end = async_chunk->end; | |
c8b97818 | 619 | u64 actual_end; |
d98da499 | 620 | u64 i_size; |
e6dcd2dc | 621 | int ret = 0; |
c8b97818 CM |
622 | struct page **pages = NULL; |
623 | unsigned long nr_pages; | |
c8b97818 CM |
624 | unsigned long total_compressed = 0; |
625 | unsigned long total_in = 0; | |
c8b97818 CM |
626 | int i; |
627 | int will_compress; | |
0b246afa | 628 | int compress_type = fs_info->compress_type; |
ac3e9933 | 629 | int compressed_extents = 0; |
4adaa611 | 630 | int redirty = 0; |
b888db2b | 631 | |
6158e1ce NB |
632 | inode_should_defrag(BTRFS_I(inode), start, end, end - start + 1, |
633 | SZ_16K); | |
4cb5300b | 634 | |
d98da499 JB |
635 | /* |
636 | * We need to save i_size before now because it could change in between | |
637 | * us evaluating the size and assigning it. This is because we lock and | |
638 | * unlock the page in truncate and fallocate, and then modify the i_size | |
639 | * later on. | |
640 | * | |
641 | * The barriers are to emulate READ_ONCE, remove that once i_size_read | |
642 | * does that for us. | |
643 | */ | |
644 | barrier(); | |
645 | i_size = i_size_read(inode); | |
646 | barrier(); | |
647 | actual_end = min_t(u64, i_size, end + 1); | |
c8b97818 CM |
648 | again: |
649 | will_compress = 0; | |
09cbfeaf | 650 | nr_pages = (end >> PAGE_SHIFT) - (start >> PAGE_SHIFT) + 1; |
069eac78 DS |
651 | nr_pages = min_t(unsigned long, nr_pages, |
652 | BTRFS_MAX_COMPRESSED / PAGE_SIZE); | |
be20aa9d | 653 | |
f03d9301 CM |
654 | /* |
655 | * we don't want to send crud past the end of i_size through | |
656 | * compression, that's just a waste of CPU time. So, if the | |
657 | * end of the file is before the start of our current | |
658 | * requested range of bytes, we bail out to the uncompressed | |
659 | * cleanup code that can deal with all of this. | |
660 | * | |
661 | * It isn't really the fastest way to fix things, but this is a | |
662 | * very uncommon corner. | |
663 | */ | |
664 | if (actual_end <= start) | |
665 | goto cleanup_and_bail_uncompressed; | |
666 | ||
c8b97818 CM |
667 | total_compressed = actual_end - start; |
668 | ||
4bcbb332 | 669 | /* |
0cf9b244 | 670 | * Skip compression for a small file range(<=blocksize) that |
01327610 | 671 | * isn't an inline extent, since it doesn't save disk space at all. |
4bcbb332 SW |
672 | */ |
673 | if (total_compressed <= blocksize && | |
674 | (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size)) | |
675 | goto cleanup_and_bail_uncompressed; | |
676 | ||
0cf9b244 QW |
677 | /* |
678 | * For subpage case, we require full page alignment for the sector | |
679 | * aligned range. | |
680 | * Thus we must also check against @actual_end, not just @end. | |
681 | */ | |
682 | if (blocksize < PAGE_SIZE) { | |
1280d2d1 FK |
683 | if (!PAGE_ALIGNED(start) || |
684 | !PAGE_ALIGNED(round_up(actual_end, blocksize))) | |
0cf9b244 QW |
685 | goto cleanup_and_bail_uncompressed; |
686 | } | |
687 | ||
069eac78 DS |
688 | total_compressed = min_t(unsigned long, total_compressed, |
689 | BTRFS_MAX_UNCOMPRESSED); | |
c8b97818 CM |
690 | total_in = 0; |
691 | ret = 0; | |
db94535d | 692 | |
771ed689 CM |
693 | /* |
694 | * we do compression for mount -o compress and when the | |
695 | * inode has not been flagged as nocompress. This flag can | |
696 | * change at any time if we discover bad compression ratios. | |
c8b97818 | 697 | */ |
4e965576 | 698 | if (inode_need_compress(BTRFS_I(inode), start, end)) { |
c8b97818 | 699 | WARN_ON(pages); |
31e818fe | 700 | pages = kcalloc(nr_pages, sizeof(struct page *), GFP_NOFS); |
560f7d75 LZ |
701 | if (!pages) { |
702 | /* just bail out to the uncompressed code */ | |
3527a018 | 703 | nr_pages = 0; |
560f7d75 LZ |
704 | goto cont; |
705 | } | |
c8b97818 | 706 | |
eec63c65 DS |
707 | if (BTRFS_I(inode)->defrag_compress) |
708 | compress_type = BTRFS_I(inode)->defrag_compress; | |
709 | else if (BTRFS_I(inode)->prop_compress) | |
b52aa8c9 | 710 | compress_type = BTRFS_I(inode)->prop_compress; |
261507a0 | 711 | |
4adaa611 CM |
712 | /* |
713 | * we need to call clear_page_dirty_for_io on each | |
714 | * page in the range. Otherwise applications with the file | |
715 | * mmap'd can wander in and change the page contents while | |
716 | * we are compressing them. | |
717 | * | |
718 | * If the compression fails for any reason, we set the pages | |
719 | * dirty again later on. | |
e9679de3 TT |
720 | * |
721 | * Note that the remaining part is redirtied, the start pointer | |
722 | * has moved, the end is the original one. | |
4adaa611 | 723 | */ |
e9679de3 TT |
724 | if (!redirty) { |
725 | extent_range_clear_dirty_for_io(inode, start, end); | |
726 | redirty = 1; | |
727 | } | |
f51d2b59 DS |
728 | |
729 | /* Compression level is applied here and only here */ | |
730 | ret = btrfs_compress_pages( | |
731 | compress_type | (fs_info->compress_level << 4), | |
261507a0 | 732 | inode->i_mapping, start, |
38c31464 | 733 | pages, |
4d3a800e | 734 | &nr_pages, |
261507a0 | 735 | &total_in, |
e5d74902 | 736 | &total_compressed); |
c8b97818 CM |
737 | |
738 | if (!ret) { | |
7073017a | 739 | unsigned long offset = offset_in_page(total_compressed); |
4d3a800e | 740 | struct page *page = pages[nr_pages - 1]; |
c8b97818 CM |
741 | |
742 | /* zero the tail end of the last page, we might be | |
743 | * sending it down to disk | |
744 | */ | |
d048b9c2 IW |
745 | if (offset) |
746 | memzero_page(page, offset, PAGE_SIZE - offset); | |
c8b97818 CM |
747 | will_compress = 1; |
748 | } | |
749 | } | |
560f7d75 | 750 | cont: |
7367253a QW |
751 | /* |
752 | * Check cow_file_range() for why we don't even try to create inline | |
753 | * extent for subpage case. | |
754 | */ | |
755 | if (start == 0 && fs_info->sectorsize == PAGE_SIZE) { | |
c8b97818 | 756 | /* lets try to make an inline extent */ |
6018ba0a | 757 | if (ret || total_in < actual_end) { |
c8b97818 | 758 | /* we didn't compress the entire range, try |
771ed689 | 759 | * to make an uncompressed inline extent. |
c8b97818 | 760 | */ |
8dd9872d | 761 | ret = cow_file_range_inline(BTRFS_I(inode), actual_end, |
a0349401 | 762 | 0, BTRFS_COMPRESS_NONE, |
d9496e8a | 763 | NULL, false); |
c8b97818 | 764 | } else { |
771ed689 | 765 | /* try making a compressed inline extent */ |
8dd9872d | 766 | ret = cow_file_range_inline(BTRFS_I(inode), actual_end, |
fe3f566c | 767 | total_compressed, |
d9496e8a OS |
768 | compress_type, pages, |
769 | false); | |
c8b97818 | 770 | } |
79787eaa | 771 | if (ret <= 0) { |
151a41bc | 772 | unsigned long clear_flags = EXTENT_DELALLOC | |
8b62f87b JB |
773 | EXTENT_DELALLOC_NEW | EXTENT_DEFRAG | |
774 | EXTENT_DO_ACCOUNTING; | |
e6eb4314 FM |
775 | unsigned long page_error_op; |
776 | ||
e6eb4314 | 777 | page_error_op = ret < 0 ? PAGE_SET_ERROR : 0; |
151a41bc | 778 | |
771ed689 | 779 | /* |
79787eaa JM |
780 | * inline extent creation worked or returned error, |
781 | * we don't need to create any more async work items. | |
782 | * Unlock and free up our temp pages. | |
8b62f87b JB |
783 | * |
784 | * We use DO_ACCOUNTING here because we need the | |
785 | * delalloc_release_metadata to be done _after_ we drop | |
786 | * our outstanding extent for clearing delalloc for this | |
787 | * range. | |
771ed689 | 788 | */ |
ad7ff17b NB |
789 | extent_clear_unlock_delalloc(BTRFS_I(inode), start, end, |
790 | NULL, | |
74e9194a | 791 | clear_flags, |
ba8b04c1 | 792 | PAGE_UNLOCK | |
6869b0a8 | 793 | PAGE_START_WRITEBACK | |
e6eb4314 | 794 | page_error_op | |
c2790a2e | 795 | PAGE_END_WRITEBACK); |
cecc8d90 | 796 | |
1e6e238c QW |
797 | /* |
798 | * Ensure we only free the compressed pages if we have | |
799 | * them allocated, as we can still reach here with | |
800 | * inode_need_compress() == false. | |
801 | */ | |
802 | if (pages) { | |
803 | for (i = 0; i < nr_pages; i++) { | |
804 | WARN_ON(pages[i]->mapping); | |
805 | put_page(pages[i]); | |
806 | } | |
807 | kfree(pages); | |
cecc8d90 | 808 | } |
cecc8d90 | 809 | return 0; |
c8b97818 CM |
810 | } |
811 | } | |
812 | ||
813 | if (will_compress) { | |
814 | /* | |
815 | * we aren't doing an inline extent round the compressed size | |
816 | * up to a block size boundary so the allocator does sane | |
817 | * things | |
818 | */ | |
fda2832f | 819 | total_compressed = ALIGN(total_compressed, blocksize); |
c8b97818 CM |
820 | |
821 | /* | |
822 | * one last check to make sure the compression is really a | |
170607eb TT |
823 | * win, compare the page count read with the blocks on disk, |
824 | * compression must free at least one sector size | |
c8b97818 | 825 | */ |
4c162778 | 826 | total_in = round_up(total_in, fs_info->sectorsize); |
170607eb | 827 | if (total_compressed + blocksize <= total_in) { |
ac3e9933 | 828 | compressed_extents++; |
c8bb0c8b AS |
829 | |
830 | /* | |
831 | * The async work queues will take care of doing actual | |
832 | * allocation on disk for these compressed pages, and | |
833 | * will submit them to the elevator. | |
834 | */ | |
b5326271 | 835 | add_async_extent(async_chunk, start, total_in, |
4d3a800e | 836 | total_compressed, pages, nr_pages, |
c8bb0c8b AS |
837 | compress_type); |
838 | ||
1170862d TT |
839 | if (start + total_in < end) { |
840 | start += total_in; | |
c8bb0c8b AS |
841 | pages = NULL; |
842 | cond_resched(); | |
843 | goto again; | |
844 | } | |
ac3e9933 | 845 | return compressed_extents; |
c8b97818 CM |
846 | } |
847 | } | |
c8bb0c8b | 848 | if (pages) { |
c8b97818 CM |
849 | /* |
850 | * the compression code ran but failed to make things smaller, | |
851 | * free any pages it allocated and our page pointer array | |
852 | */ | |
4d3a800e | 853 | for (i = 0; i < nr_pages; i++) { |
70b99e69 | 854 | WARN_ON(pages[i]->mapping); |
09cbfeaf | 855 | put_page(pages[i]); |
c8b97818 CM |
856 | } |
857 | kfree(pages); | |
858 | pages = NULL; | |
859 | total_compressed = 0; | |
4d3a800e | 860 | nr_pages = 0; |
c8b97818 CM |
861 | |
862 | /* flag the file so we don't compress in the future */ | |
0b246afa | 863 | if (!btrfs_test_opt(fs_info, FORCE_COMPRESS) && |
b52aa8c9 | 864 | !(BTRFS_I(inode)->prop_compress)) { |
a555f810 | 865 | BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS; |
1e701a32 | 866 | } |
c8b97818 | 867 | } |
f03d9301 | 868 | cleanup_and_bail_uncompressed: |
c8bb0c8b AS |
869 | /* |
870 | * No compression, but we still need to write the pages in the file | |
871 | * we've been given so far. redirty the locked page if it corresponds | |
872 | * to our extent and set things up for the async work queue to run | |
873 | * cow_file_range to do the normal delalloc dance. | |
874 | */ | |
1d53c9e6 CM |
875 | if (async_chunk->locked_page && |
876 | (page_offset(async_chunk->locked_page) >= start && | |
877 | page_offset(async_chunk->locked_page)) <= end) { | |
1368c6da | 878 | __set_page_dirty_nobuffers(async_chunk->locked_page); |
c8bb0c8b | 879 | /* unlocked later on in the async handlers */ |
1d53c9e6 | 880 | } |
c8bb0c8b AS |
881 | |
882 | if (redirty) | |
883 | extent_range_redirty_for_io(inode, start, end); | |
b5326271 | 884 | add_async_extent(async_chunk, start, end - start + 1, 0, NULL, 0, |
c8bb0c8b | 885 | BTRFS_COMPRESS_NONE); |
ac3e9933 | 886 | compressed_extents++; |
3b951516 | 887 | |
ac3e9933 | 888 | return compressed_extents; |
771ed689 | 889 | } |
771ed689 | 890 | |
40ae837b FM |
891 | static void free_async_extent_pages(struct async_extent *async_extent) |
892 | { | |
893 | int i; | |
894 | ||
895 | if (!async_extent->pages) | |
896 | return; | |
897 | ||
898 | for (i = 0; i < async_extent->nr_pages; i++) { | |
899 | WARN_ON(async_extent->pages[i]->mapping); | |
09cbfeaf | 900 | put_page(async_extent->pages[i]); |
40ae837b FM |
901 | } |
902 | kfree(async_extent->pages); | |
903 | async_extent->nr_pages = 0; | |
904 | async_extent->pages = NULL; | |
771ed689 CM |
905 | } |
906 | ||
2b83a0ee QW |
907 | static int submit_uncompressed_range(struct btrfs_inode *inode, |
908 | struct async_extent *async_extent, | |
909 | struct page *locked_page) | |
771ed689 | 910 | { |
2b83a0ee QW |
911 | u64 start = async_extent->start; |
912 | u64 end = async_extent->start + async_extent->ram_size - 1; | |
913 | unsigned long nr_written = 0; | |
914 | int page_started = 0; | |
915 | int ret; | |
771ed689 | 916 | |
2b83a0ee QW |
917 | /* |
918 | * Call cow_file_range() to run the delalloc range directly, since we | |
919 | * won't go to NOCOW or async path again. | |
920 | * | |
921 | * Also we call cow_file_range() with @unlock_page == 0, so that we | |
922 | * can directly submit them without interruption. | |
923 | */ | |
924 | ret = cow_file_range(inode, locked_page, start, end, &page_started, | |
898793d9 | 925 | &nr_written, 0, NULL); |
2b83a0ee QW |
926 | /* Inline extent inserted, page gets unlocked and everything is done */ |
927 | if (page_started) { | |
928 | ret = 0; | |
929 | goto out; | |
930 | } | |
931 | if (ret < 0) { | |
71aa147b NA |
932 | btrfs_cleanup_ordered_extents(inode, locked_page, start, end - start + 1); |
933 | if (locked_page) { | |
934 | const u64 page_start = page_offset(locked_page); | |
935 | const u64 page_end = page_start + PAGE_SIZE - 1; | |
936 | ||
937 | btrfs_page_set_error(inode->root->fs_info, locked_page, | |
938 | page_start, PAGE_SIZE); | |
939 | set_page_writeback(locked_page); | |
940 | end_page_writeback(locked_page); | |
941 | end_extent_writepage(locked_page, ret, page_start, page_end); | |
2b83a0ee | 942 | unlock_page(locked_page); |
71aa147b | 943 | } |
2b83a0ee QW |
944 | goto out; |
945 | } | |
771ed689 | 946 | |
2b83a0ee QW |
947 | ret = extent_write_locked_range(&inode->vfs_inode, start, end); |
948 | /* All pages will be unlocked, including @locked_page */ | |
949 | out: | |
950 | kfree(async_extent); | |
951 | return ret; | |
952 | } | |
79787eaa | 953 | |
b4ccace8 QW |
954 | static int submit_one_async_extent(struct btrfs_inode *inode, |
955 | struct async_chunk *async_chunk, | |
956 | struct async_extent *async_extent, | |
957 | u64 *alloc_hint) | |
771ed689 | 958 | { |
b4ccace8 QW |
959 | struct extent_io_tree *io_tree = &inode->io_tree; |
960 | struct btrfs_root *root = inode->root; | |
961 | struct btrfs_fs_info *fs_info = root->fs_info; | |
771ed689 | 962 | struct btrfs_key ins; |
2b83a0ee | 963 | struct page *locked_page = NULL; |
771ed689 | 964 | struct extent_map *em; |
f5a84ee3 | 965 | int ret = 0; |
b4ccace8 QW |
966 | u64 start = async_extent->start; |
967 | u64 end = async_extent->start + async_extent->ram_size - 1; | |
771ed689 | 968 | |
2b83a0ee QW |
969 | /* |
970 | * If async_chunk->locked_page is in the async_extent range, we need to | |
971 | * handle it. | |
972 | */ | |
973 | if (async_chunk->locked_page) { | |
974 | u64 locked_page_start = page_offset(async_chunk->locked_page); | |
975 | u64 locked_page_end = locked_page_start + PAGE_SIZE - 1; | |
3e04e7f1 | 976 | |
2b83a0ee QW |
977 | if (!(start >= locked_page_end || end <= locked_page_start)) |
978 | locked_page = async_chunk->locked_page; | |
b4ccace8 | 979 | } |
570eb97b | 980 | lock_extent(io_tree, start, end, NULL); |
ce62003f | 981 | |
2b83a0ee QW |
982 | /* We have fall back to uncompressed write */ |
983 | if (!async_extent->pages) | |
984 | return submit_uncompressed_range(inode, async_extent, locked_page); | |
ce62003f | 985 | |
b4ccace8 QW |
986 | ret = btrfs_reserve_extent(root, async_extent->ram_size, |
987 | async_extent->compressed_size, | |
988 | async_extent->compressed_size, | |
989 | 0, *alloc_hint, &ins, 1, 1); | |
990 | if (ret) { | |
991 | free_async_extent_pages(async_extent); | |
c2167754 | 992 | /* |
b4ccace8 QW |
993 | * Here we used to try again by going back to non-compressed |
994 | * path for ENOSPC. But we can't reserve space even for | |
995 | * compressed size, how could it work for uncompressed size | |
996 | * which requires larger size? So here we directly go error | |
997 | * path. | |
c2167754 | 998 | */ |
b4ccace8 QW |
999 | goto out_free; |
1000 | } | |
1001 | ||
1002 | /* Here we're doing allocation and writeback of the compressed pages */ | |
1003 | em = create_io_em(inode, start, | |
1004 | async_extent->ram_size, /* len */ | |
1005 | start, /* orig_start */ | |
1006 | ins.objectid, /* block_start */ | |
1007 | ins.offset, /* block_len */ | |
1008 | ins.offset, /* orig_block_len */ | |
1009 | async_extent->ram_size, /* ram_bytes */ | |
1010 | async_extent->compress_type, | |
1011 | BTRFS_ORDERED_COMPRESSED); | |
1012 | if (IS_ERR(em)) { | |
1013 | ret = PTR_ERR(em); | |
1014 | goto out_free_reserve; | |
1015 | } | |
1016 | free_extent_map(em); | |
771ed689 | 1017 | |
cb36a9bb OS |
1018 | ret = btrfs_add_ordered_extent(inode, start, /* file_offset */ |
1019 | async_extent->ram_size, /* num_bytes */ | |
1020 | async_extent->ram_size, /* ram_bytes */ | |
1021 | ins.objectid, /* disk_bytenr */ | |
1022 | ins.offset, /* disk_num_bytes */ | |
1023 | 0, /* offset */ | |
1024 | 1 << BTRFS_ORDERED_COMPRESSED, | |
1025 | async_extent->compress_type); | |
b4ccace8 QW |
1026 | if (ret) { |
1027 | btrfs_drop_extent_cache(inode, start, end, 0); | |
1028 | goto out_free_reserve; | |
771ed689 | 1029 | } |
b4ccace8 QW |
1030 | btrfs_dec_block_group_reservations(fs_info, ins.objectid); |
1031 | ||
1032 | /* Clear dirty, set writeback and unlock the pages. */ | |
1033 | extent_clear_unlock_delalloc(inode, start, end, | |
1034 | NULL, EXTENT_LOCKED | EXTENT_DELALLOC, | |
1035 | PAGE_UNLOCK | PAGE_START_WRITEBACK); | |
1036 | if (btrfs_submit_compressed_write(inode, start, /* file_offset */ | |
1037 | async_extent->ram_size, /* num_bytes */ | |
1038 | ins.objectid, /* disk_bytenr */ | |
1039 | ins.offset, /* compressed_len */ | |
1040 | async_extent->pages, /* compressed_pages */ | |
1041 | async_extent->nr_pages, | |
1042 | async_chunk->write_flags, | |
7c0c7269 | 1043 | async_chunk->blkcg_css, true)) { |
b4ccace8 QW |
1044 | const u64 start = async_extent->start; |
1045 | const u64 end = start + async_extent->ram_size - 1; | |
1046 | ||
1047 | btrfs_writepage_endio_finish_ordered(inode, NULL, start, end, 0); | |
1048 | ||
1049 | extent_clear_unlock_delalloc(inode, start, end, NULL, 0, | |
1050 | PAGE_END_WRITEBACK | PAGE_SET_ERROR); | |
1051 | free_async_extent_pages(async_extent); | |
771ed689 | 1052 | } |
b4ccace8 QW |
1053 | *alloc_hint = ins.objectid + ins.offset; |
1054 | kfree(async_extent); | |
1055 | return ret; | |
1056 | ||
3e04e7f1 | 1057 | out_free_reserve: |
0b246afa | 1058 | btrfs_dec_block_group_reservations(fs_info, ins.objectid); |
2ff7e61e | 1059 | btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1); |
79787eaa | 1060 | out_free: |
b4ccace8 | 1061 | extent_clear_unlock_delalloc(inode, start, end, |
c2790a2e | 1062 | NULL, EXTENT_LOCKED | EXTENT_DELALLOC | |
a7e3b975 | 1063 | EXTENT_DELALLOC_NEW | |
151a41bc | 1064 | EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING, |
6869b0a8 QW |
1065 | PAGE_UNLOCK | PAGE_START_WRITEBACK | |
1066 | PAGE_END_WRITEBACK | PAGE_SET_ERROR); | |
40ae837b | 1067 | free_async_extent_pages(async_extent); |
79787eaa | 1068 | kfree(async_extent); |
b4ccace8 QW |
1069 | return ret; |
1070 | } | |
1071 | ||
1072 | /* | |
1073 | * Phase two of compressed writeback. This is the ordered portion of the code, | |
1074 | * which only gets called in the order the work was queued. We walk all the | |
1075 | * async extents created by compress_file_range and send them down to the disk. | |
1076 | */ | |
1077 | static noinline void submit_compressed_extents(struct async_chunk *async_chunk) | |
1078 | { | |
1079 | struct btrfs_inode *inode = BTRFS_I(async_chunk->inode); | |
1080 | struct btrfs_fs_info *fs_info = inode->root->fs_info; | |
1081 | struct async_extent *async_extent; | |
1082 | u64 alloc_hint = 0; | |
1083 | int ret = 0; | |
1084 | ||
1085 | while (!list_empty(&async_chunk->extents)) { | |
1086 | u64 extent_start; | |
1087 | u64 ram_size; | |
1088 | ||
1089 | async_extent = list_entry(async_chunk->extents.next, | |
1090 | struct async_extent, list); | |
1091 | list_del(&async_extent->list); | |
1092 | extent_start = async_extent->start; | |
1093 | ram_size = async_extent->ram_size; | |
1094 | ||
1095 | ret = submit_one_async_extent(inode, async_chunk, async_extent, | |
1096 | &alloc_hint); | |
1097 | btrfs_debug(fs_info, | |
1098 | "async extent submission failed root=%lld inode=%llu start=%llu len=%llu ret=%d", | |
1099 | inode->root->root_key.objectid, | |
1100 | btrfs_ino(inode), extent_start, ram_size, ret); | |
1101 | } | |
771ed689 CM |
1102 | } |
1103 | ||
43c69849 | 1104 | static u64 get_extent_allocation_hint(struct btrfs_inode *inode, u64 start, |
4b46fce2 JB |
1105 | u64 num_bytes) |
1106 | { | |
43c69849 | 1107 | struct extent_map_tree *em_tree = &inode->extent_tree; |
4b46fce2 JB |
1108 | struct extent_map *em; |
1109 | u64 alloc_hint = 0; | |
1110 | ||
1111 | read_lock(&em_tree->lock); | |
1112 | em = search_extent_mapping(em_tree, start, num_bytes); | |
1113 | if (em) { | |
1114 | /* | |
1115 | * if block start isn't an actual block number then find the | |
1116 | * first block in this inode and use that as a hint. If that | |
1117 | * block is also bogus then just don't worry about it. | |
1118 | */ | |
1119 | if (em->block_start >= EXTENT_MAP_LAST_BYTE) { | |
1120 | free_extent_map(em); | |
1121 | em = search_extent_mapping(em_tree, 0, 0); | |
1122 | if (em && em->block_start < EXTENT_MAP_LAST_BYTE) | |
1123 | alloc_hint = em->block_start; | |
1124 | if (em) | |
1125 | free_extent_map(em); | |
1126 | } else { | |
1127 | alloc_hint = em->block_start; | |
1128 | free_extent_map(em); | |
1129 | } | |
1130 | } | |
1131 | read_unlock(&em_tree->lock); | |
1132 | ||
1133 | return alloc_hint; | |
1134 | } | |
1135 | ||
771ed689 CM |
1136 | /* |
1137 | * when extent_io.c finds a delayed allocation range in the file, | |
1138 | * the call backs end up in this code. The basic idea is to | |
1139 | * allocate extents on disk for the range, and create ordered data structs | |
1140 | * in ram to track those extents. | |
1141 | * | |
1142 | * locked_page is the page that writepage had locked already. We use | |
1143 | * it to make sure we don't do extra locks or unlocks. | |
1144 | * | |
1145 | * *page_started is set to one if we unlock locked_page and do everything | |
1146 | * required to start IO on it. It may be clean and already done with | |
1147 | * IO when we return. | |
9ce7466f NA |
1148 | * |
1149 | * When unlock == 1, we unlock the pages in successfully allocated regions. | |
1150 | * When unlock == 0, we leave them locked for writing them out. | |
1151 | * | |
1152 | * However, we unlock all the pages except @locked_page in case of failure. | |
1153 | * | |
1154 | * In summary, page locking state will be as follow: | |
1155 | * | |
1156 | * - page_started == 1 (return value) | |
1157 | * - All the pages are unlocked. IO is started. | |
1158 | * - Note that this can happen only on success | |
1159 | * - unlock == 1 | |
1160 | * - All the pages except @locked_page are unlocked in any case | |
1161 | * - unlock == 0 | |
1162 | * - On success, all the pages are locked for writing out them | |
1163 | * - On failure, all the pages except @locked_page are unlocked | |
1164 | * | |
1165 | * When a failure happens in the second or later iteration of the | |
1166 | * while-loop, the ordered extents created in previous iterations are kept | |
1167 | * intact. So, the caller must clean them up by calling | |
1168 | * btrfs_cleanup_ordered_extents(). See btrfs_run_delalloc_range() for | |
1169 | * example. | |
771ed689 | 1170 | */ |
6e26c442 | 1171 | static noinline int cow_file_range(struct btrfs_inode *inode, |
00361589 | 1172 | struct page *locked_page, |
74e9194a | 1173 | u64 start, u64 end, int *page_started, |
898793d9 NA |
1174 | unsigned long *nr_written, int unlock, |
1175 | u64 *done_offset) | |
771ed689 | 1176 | { |
6e26c442 NB |
1177 | struct btrfs_root *root = inode->root; |
1178 | struct btrfs_fs_info *fs_info = root->fs_info; | |
771ed689 | 1179 | u64 alloc_hint = 0; |
9ce7466f | 1180 | u64 orig_start = start; |
771ed689 CM |
1181 | u64 num_bytes; |
1182 | unsigned long ram_size; | |
a315e68f | 1183 | u64 cur_alloc_size = 0; |
432cd2a1 | 1184 | u64 min_alloc_size; |
0b246afa | 1185 | u64 blocksize = fs_info->sectorsize; |
771ed689 CM |
1186 | struct btrfs_key ins; |
1187 | struct extent_map *em; | |
a315e68f FM |
1188 | unsigned clear_bits; |
1189 | unsigned long page_ops; | |
1190 | bool extent_reserved = false; | |
771ed689 CM |
1191 | int ret = 0; |
1192 | ||
6e26c442 | 1193 | if (btrfs_is_free_space_inode(inode)) { |
29bce2f3 JB |
1194 | ret = -EINVAL; |
1195 | goto out_unlock; | |
02ecd2c2 | 1196 | } |
771ed689 | 1197 | |
fda2832f | 1198 | num_bytes = ALIGN(end - start + 1, blocksize); |
771ed689 | 1199 | num_bytes = max(blocksize, num_bytes); |
566b1760 | 1200 | ASSERT(num_bytes <= btrfs_super_total_bytes(fs_info->super_copy)); |
771ed689 | 1201 | |
6e26c442 | 1202 | inode_should_defrag(inode, start, end, num_bytes, SZ_64K); |
4cb5300b | 1203 | |
7367253a QW |
1204 | /* |
1205 | * Due to the page size limit, for subpage we can only trigger the | |
1206 | * writeback for the dirty sectors of page, that means data writeback | |
1207 | * is doing more writeback than what we want. | |
1208 | * | |
1209 | * This is especially unexpected for some call sites like fallocate, | |
1210 | * where we only increase i_size after everything is done. | |
1211 | * This means we can trigger inline extent even if we didn't want to. | |
1212 | * So here we skip inline extent creation completely. | |
1213 | */ | |
1214 | if (start == 0 && fs_info->sectorsize == PAGE_SIZE) { | |
8dd9872d OS |
1215 | u64 actual_end = min_t(u64, i_size_read(&inode->vfs_inode), |
1216 | end + 1); | |
1217 | ||
771ed689 | 1218 | /* lets try to make an inline extent */ |
8dd9872d | 1219 | ret = cow_file_range_inline(inode, actual_end, 0, |
d9496e8a | 1220 | BTRFS_COMPRESS_NONE, NULL, false); |
771ed689 | 1221 | if (ret == 0) { |
8b62f87b JB |
1222 | /* |
1223 | * We use DO_ACCOUNTING here because we need the | |
1224 | * delalloc_release_metadata to be run _after_ we drop | |
1225 | * our outstanding extent for clearing delalloc for this | |
1226 | * range. | |
1227 | */ | |
4750af3b QW |
1228 | extent_clear_unlock_delalloc(inode, start, end, |
1229 | locked_page, | |
c2790a2e | 1230 | EXTENT_LOCKED | EXTENT_DELALLOC | |
8b62f87b JB |
1231 | EXTENT_DELALLOC_NEW | EXTENT_DEFRAG | |
1232 | EXTENT_DO_ACCOUNTING, PAGE_UNLOCK | | |
6869b0a8 | 1233 | PAGE_START_WRITEBACK | PAGE_END_WRITEBACK); |
771ed689 | 1234 | *nr_written = *nr_written + |
09cbfeaf | 1235 | (end - start + PAGE_SIZE) / PAGE_SIZE; |
771ed689 | 1236 | *page_started = 1; |
4750af3b QW |
1237 | /* |
1238 | * locked_page is locked by the caller of | |
1239 | * writepage_delalloc(), not locked by | |
1240 | * __process_pages_contig(). | |
1241 | * | |
1242 | * We can't let __process_pages_contig() to unlock it, | |
1243 | * as it doesn't have any subpage::writers recorded. | |
1244 | * | |
1245 | * Here we manually unlock the page, since the caller | |
1246 | * can't use page_started to determine if it's an | |
1247 | * inline extent or a compressed extent. | |
1248 | */ | |
1249 | unlock_page(locked_page); | |
771ed689 | 1250 | goto out; |
79787eaa | 1251 | } else if (ret < 0) { |
79787eaa | 1252 | goto out_unlock; |
771ed689 CM |
1253 | } |
1254 | } | |
1255 | ||
6e26c442 NB |
1256 | alloc_hint = get_extent_allocation_hint(inode, start, num_bytes); |
1257 | btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0); | |
771ed689 | 1258 | |
432cd2a1 FM |
1259 | /* |
1260 | * Relocation relies on the relocated extents to have exactly the same | |
1261 | * size as the original extents. Normally writeback for relocation data | |
1262 | * extents follows a NOCOW path because relocation preallocates the | |
1263 | * extents. However, due to an operation such as scrub turning a block | |
1264 | * group to RO mode, it may fallback to COW mode, so we must make sure | |
1265 | * an extent allocated during COW has exactly the requested size and can | |
1266 | * not be split into smaller extents, otherwise relocation breaks and | |
1267 | * fails during the stage where it updates the bytenr of file extent | |
1268 | * items. | |
1269 | */ | |
37f00a6d | 1270 | if (btrfs_is_data_reloc_root(root)) |
432cd2a1 FM |
1271 | min_alloc_size = num_bytes; |
1272 | else | |
1273 | min_alloc_size = fs_info->sectorsize; | |
1274 | ||
3752d22f AJ |
1275 | while (num_bytes > 0) { |
1276 | cur_alloc_size = num_bytes; | |
18513091 | 1277 | ret = btrfs_reserve_extent(root, cur_alloc_size, cur_alloc_size, |
432cd2a1 | 1278 | min_alloc_size, 0, alloc_hint, |
e570fd27 | 1279 | &ins, 1, 1); |
00361589 | 1280 | if (ret < 0) |
79787eaa | 1281 | goto out_unlock; |
a315e68f FM |
1282 | cur_alloc_size = ins.offset; |
1283 | extent_reserved = true; | |
d397712b | 1284 | |
771ed689 | 1285 | ram_size = ins.offset; |
6e26c442 | 1286 | em = create_io_em(inode, start, ins.offset, /* len */ |
6f9994db LB |
1287 | start, /* orig_start */ |
1288 | ins.objectid, /* block_start */ | |
1289 | ins.offset, /* block_len */ | |
1290 | ins.offset, /* orig_block_len */ | |
1291 | ram_size, /* ram_bytes */ | |
1292 | BTRFS_COMPRESS_NONE, /* compress_type */ | |
1af4a0aa | 1293 | BTRFS_ORDERED_REGULAR /* type */); |
090a127a SY |
1294 | if (IS_ERR(em)) { |
1295 | ret = PTR_ERR(em); | |
ace68bac | 1296 | goto out_reserve; |
090a127a | 1297 | } |
6f9994db | 1298 | free_extent_map(em); |
e6dcd2dc | 1299 | |
cb36a9bb OS |
1300 | ret = btrfs_add_ordered_extent(inode, start, ram_size, ram_size, |
1301 | ins.objectid, cur_alloc_size, 0, | |
1302 | 1 << BTRFS_ORDERED_REGULAR, | |
1303 | BTRFS_COMPRESS_NONE); | |
ace68bac | 1304 | if (ret) |
d9f85963 | 1305 | goto out_drop_extent_cache; |
c8b97818 | 1306 | |
37f00a6d | 1307 | if (btrfs_is_data_reloc_root(root)) { |
6e26c442 | 1308 | ret = btrfs_reloc_clone_csums(inode, start, |
17d217fe | 1309 | cur_alloc_size); |
4dbd80fb QW |
1310 | /* |
1311 | * Only drop cache here, and process as normal. | |
1312 | * | |
1313 | * We must not allow extent_clear_unlock_delalloc() | |
1314 | * at out_unlock label to free meta of this ordered | |
1315 | * extent, as its meta should be freed by | |
1316 | * btrfs_finish_ordered_io(). | |
1317 | * | |
1318 | * So we must continue until @start is increased to | |
1319 | * skip current ordered extent. | |
1320 | */ | |
00361589 | 1321 | if (ret) |
6e26c442 | 1322 | btrfs_drop_extent_cache(inode, start, |
4dbd80fb | 1323 | start + ram_size - 1, 0); |
17d217fe YZ |
1324 | } |
1325 | ||
0b246afa | 1326 | btrfs_dec_block_group_reservations(fs_info, ins.objectid); |
9cfa3e34 | 1327 | |
f57ad937 QW |
1328 | /* |
1329 | * We're not doing compressed IO, don't unlock the first page | |
1330 | * (which the caller expects to stay locked), don't clear any | |
1331 | * dirty bits and don't set any writeback bits | |
8b62b72b | 1332 | * |
f57ad937 QW |
1333 | * Do set the Ordered (Private2) bit so we know this page was |
1334 | * properly setup for writepage. | |
c8b97818 | 1335 | */ |
a315e68f | 1336 | page_ops = unlock ? PAGE_UNLOCK : 0; |
f57ad937 | 1337 | page_ops |= PAGE_SET_ORDERED; |
a791e35e | 1338 | |
6e26c442 | 1339 | extent_clear_unlock_delalloc(inode, start, start + ram_size - 1, |
74e9194a | 1340 | locked_page, |
c2790a2e | 1341 | EXTENT_LOCKED | EXTENT_DELALLOC, |
a315e68f | 1342 | page_ops); |
3752d22f AJ |
1343 | if (num_bytes < cur_alloc_size) |
1344 | num_bytes = 0; | |
4dbd80fb | 1345 | else |
3752d22f | 1346 | num_bytes -= cur_alloc_size; |
c59f8951 CM |
1347 | alloc_hint = ins.objectid + ins.offset; |
1348 | start += cur_alloc_size; | |
a315e68f | 1349 | extent_reserved = false; |
4dbd80fb QW |
1350 | |
1351 | /* | |
1352 | * btrfs_reloc_clone_csums() error, since start is increased | |
1353 | * extent_clear_unlock_delalloc() at out_unlock label won't | |
1354 | * free metadata of current ordered extent, we're OK to exit. | |
1355 | */ | |
1356 | if (ret) | |
1357 | goto out_unlock; | |
b888db2b | 1358 | } |
79787eaa | 1359 | out: |
be20aa9d | 1360 | return ret; |
b7d5b0a8 | 1361 | |
d9f85963 | 1362 | out_drop_extent_cache: |
6e26c442 | 1363 | btrfs_drop_extent_cache(inode, start, start + ram_size - 1, 0); |
ace68bac | 1364 | out_reserve: |
0b246afa | 1365 | btrfs_dec_block_group_reservations(fs_info, ins.objectid); |
2ff7e61e | 1366 | btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1); |
79787eaa | 1367 | out_unlock: |
898793d9 NA |
1368 | /* |
1369 | * If done_offset is non-NULL and ret == -EAGAIN, we expect the | |
1370 | * caller to write out the successfully allocated region and retry. | |
1371 | */ | |
1372 | if (done_offset && ret == -EAGAIN) { | |
1373 | if (orig_start < start) | |
1374 | *done_offset = start - 1; | |
1375 | else | |
1376 | *done_offset = start; | |
1377 | return ret; | |
1378 | } else if (ret == -EAGAIN) { | |
1379 | /* Convert to -ENOSPC since the caller cannot retry. */ | |
1380 | ret = -ENOSPC; | |
1381 | } | |
1382 | ||
9ce7466f NA |
1383 | /* |
1384 | * Now, we have three regions to clean up: | |
1385 | * | |
1386 | * |-------(1)----|---(2)---|-------------(3)----------| | |
1387 | * `- orig_start `- start `- start + cur_alloc_size `- end | |
1388 | * | |
1389 | * We process each region below. | |
1390 | */ | |
1391 | ||
a7e3b975 FM |
1392 | clear_bits = EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DELALLOC_NEW | |
1393 | EXTENT_DEFRAG | EXTENT_CLEAR_META_RESV; | |
6869b0a8 | 1394 | page_ops = PAGE_UNLOCK | PAGE_START_WRITEBACK | PAGE_END_WRITEBACK; |
9ce7466f | 1395 | |
a315e68f | 1396 | /* |
9ce7466f NA |
1397 | * For the range (1). We have already instantiated the ordered extents |
1398 | * for this region. They are cleaned up by | |
1399 | * btrfs_cleanup_ordered_extents() in e.g, | |
1400 | * btrfs_run_delalloc_range(). EXTENT_LOCKED | EXTENT_DELALLOC are | |
1401 | * already cleared in the above loop. And, EXTENT_DELALLOC_NEW | | |
1402 | * EXTENT_DEFRAG | EXTENT_CLEAR_META_RESV are handled by the cleanup | |
1403 | * function. | |
1404 | * | |
1405 | * However, in case of unlock == 0, we still need to unlock the pages | |
1406 | * (except @locked_page) to ensure all the pages are unlocked. | |
1407 | */ | |
71aa147b NA |
1408 | if (!unlock && orig_start < start) { |
1409 | if (!locked_page) | |
1410 | mapping_set_error(inode->vfs_inode.i_mapping, ret); | |
9ce7466f NA |
1411 | extent_clear_unlock_delalloc(inode, orig_start, start - 1, |
1412 | locked_page, 0, page_ops); | |
71aa147b | 1413 | } |
9ce7466f | 1414 | |
a315e68f | 1415 | /* |
9ce7466f NA |
1416 | * For the range (2). If we reserved an extent for our delalloc range |
1417 | * (or a subrange) and failed to create the respective ordered extent, | |
1418 | * then it means that when we reserved the extent we decremented the | |
1419 | * extent's size from the data space_info's bytes_may_use counter and | |
1420 | * incremented the space_info's bytes_reserved counter by the same | |
1421 | * amount. We must make sure extent_clear_unlock_delalloc() does not try | |
1422 | * to decrement again the data space_info's bytes_may_use counter, | |
1423 | * therefore we do not pass it the flag EXTENT_CLEAR_DATA_RESV. | |
a315e68f FM |
1424 | */ |
1425 | if (extent_reserved) { | |
6e26c442 | 1426 | extent_clear_unlock_delalloc(inode, start, |
e2c8e92d | 1427 | start + cur_alloc_size - 1, |
a315e68f FM |
1428 | locked_page, |
1429 | clear_bits, | |
1430 | page_ops); | |
1431 | start += cur_alloc_size; | |
1432 | if (start >= end) | |
aaafa1eb | 1433 | return ret; |
a315e68f | 1434 | } |
9ce7466f NA |
1435 | |
1436 | /* | |
1437 | * For the range (3). We never touched the region. In addition to the | |
1438 | * clear_bits above, we add EXTENT_CLEAR_DATA_RESV to release the data | |
1439 | * space_info's bytes_may_use counter, reserved in | |
1440 | * btrfs_check_data_free_space(). | |
1441 | */ | |
6e26c442 | 1442 | extent_clear_unlock_delalloc(inode, start, end, locked_page, |
a315e68f FM |
1443 | clear_bits | EXTENT_CLEAR_DATA_RESV, |
1444 | page_ops); | |
aaafa1eb | 1445 | return ret; |
771ed689 | 1446 | } |
c8b97818 | 1447 | |
771ed689 CM |
1448 | /* |
1449 | * work queue call back to started compression on a file and pages | |
1450 | */ | |
1451 | static noinline void async_cow_start(struct btrfs_work *work) | |
1452 | { | |
b5326271 | 1453 | struct async_chunk *async_chunk; |
ac3e9933 | 1454 | int compressed_extents; |
771ed689 | 1455 | |
b5326271 | 1456 | async_chunk = container_of(work, struct async_chunk, work); |
771ed689 | 1457 | |
ac3e9933 NB |
1458 | compressed_extents = compress_file_range(async_chunk); |
1459 | if (compressed_extents == 0) { | |
b5326271 NB |
1460 | btrfs_add_delayed_iput(async_chunk->inode); |
1461 | async_chunk->inode = NULL; | |
8180ef88 | 1462 | } |
771ed689 CM |
1463 | } |
1464 | ||
1465 | /* | |
1466 | * work queue call back to submit previously compressed pages | |
1467 | */ | |
1468 | static noinline void async_cow_submit(struct btrfs_work *work) | |
1469 | { | |
c5a68aec NB |
1470 | struct async_chunk *async_chunk = container_of(work, struct async_chunk, |
1471 | work); | |
1472 | struct btrfs_fs_info *fs_info = btrfs_work_owner(work); | |
771ed689 CM |
1473 | unsigned long nr_pages; |
1474 | ||
b5326271 | 1475 | nr_pages = (async_chunk->end - async_chunk->start + PAGE_SIZE) >> |
09cbfeaf | 1476 | PAGE_SHIFT; |
771ed689 | 1477 | |
4546d178 | 1478 | /* |
b5326271 | 1479 | * ->inode could be NULL if async_chunk_start has failed to compress, |
4546d178 NB |
1480 | * in which case we don't have anything to submit, yet we need to |
1481 | * always adjust ->async_delalloc_pages as its paired with the init | |
1482 | * happening in cow_file_range_async | |
1483 | */ | |
b5326271 NB |
1484 | if (async_chunk->inode) |
1485 | submit_compressed_extents(async_chunk); | |
ac98141d JB |
1486 | |
1487 | /* atomic_sub_return implies a barrier */ | |
1488 | if (atomic_sub_return(nr_pages, &fs_info->async_delalloc_pages) < | |
1489 | 5 * SZ_1M) | |
1490 | cond_wake_up_nomb(&fs_info->async_submit_wait); | |
771ed689 | 1491 | } |
c8b97818 | 1492 | |
771ed689 CM |
1493 | static noinline void async_cow_free(struct btrfs_work *work) |
1494 | { | |
b5326271 | 1495 | struct async_chunk *async_chunk; |
9e895a8f | 1496 | struct async_cow *async_cow; |
97db1204 | 1497 | |
b5326271 NB |
1498 | async_chunk = container_of(work, struct async_chunk, work); |
1499 | if (async_chunk->inode) | |
1500 | btrfs_add_delayed_iput(async_chunk->inode); | |
ec39f769 CM |
1501 | if (async_chunk->blkcg_css) |
1502 | css_put(async_chunk->blkcg_css); | |
9e895a8f QW |
1503 | |
1504 | async_cow = async_chunk->async_cow; | |
1505 | if (atomic_dec_and_test(&async_cow->num_chunks)) | |
1506 | kvfree(async_cow); | |
771ed689 CM |
1507 | } |
1508 | ||
751b6431 | 1509 | static int cow_file_range_async(struct btrfs_inode *inode, |
ec39f769 CM |
1510 | struct writeback_control *wbc, |
1511 | struct page *locked_page, | |
771ed689 | 1512 | u64 start, u64 end, int *page_started, |
fac07d2b | 1513 | unsigned long *nr_written) |
771ed689 | 1514 | { |
751b6431 | 1515 | struct btrfs_fs_info *fs_info = inode->root->fs_info; |
ec39f769 | 1516 | struct cgroup_subsys_state *blkcg_css = wbc_blkcg_css(wbc); |
97db1204 NB |
1517 | struct async_cow *ctx; |
1518 | struct async_chunk *async_chunk; | |
771ed689 CM |
1519 | unsigned long nr_pages; |
1520 | u64 cur_end; | |
97db1204 NB |
1521 | u64 num_chunks = DIV_ROUND_UP(end - start, SZ_512K); |
1522 | int i; | |
1523 | bool should_compress; | |
b1c16ac9 | 1524 | unsigned nofs_flag; |
bf9486d6 | 1525 | const blk_opf_t write_flags = wbc_to_write_flags(wbc); |
771ed689 | 1526 | |
570eb97b | 1527 | unlock_extent(&inode->io_tree, start, end, NULL); |
97db1204 | 1528 | |
751b6431 | 1529 | if (inode->flags & BTRFS_INODE_NOCOMPRESS && |
97db1204 NB |
1530 | !btrfs_test_opt(fs_info, FORCE_COMPRESS)) { |
1531 | num_chunks = 1; | |
1532 | should_compress = false; | |
1533 | } else { | |
1534 | should_compress = true; | |
1535 | } | |
1536 | ||
b1c16ac9 NB |
1537 | nofs_flag = memalloc_nofs_save(); |
1538 | ctx = kvmalloc(struct_size(ctx, chunks, num_chunks), GFP_KERNEL); | |
1539 | memalloc_nofs_restore(nofs_flag); | |
1540 | ||
97db1204 NB |
1541 | if (!ctx) { |
1542 | unsigned clear_bits = EXTENT_LOCKED | EXTENT_DELALLOC | | |
1543 | EXTENT_DELALLOC_NEW | EXTENT_DEFRAG | | |
1544 | EXTENT_DO_ACCOUNTING; | |
6869b0a8 QW |
1545 | unsigned long page_ops = PAGE_UNLOCK | PAGE_START_WRITEBACK | |
1546 | PAGE_END_WRITEBACK | PAGE_SET_ERROR; | |
97db1204 | 1547 | |
751b6431 NB |
1548 | extent_clear_unlock_delalloc(inode, start, end, locked_page, |
1549 | clear_bits, page_ops); | |
97db1204 NB |
1550 | return -ENOMEM; |
1551 | } | |
1552 | ||
1553 | async_chunk = ctx->chunks; | |
1554 | atomic_set(&ctx->num_chunks, num_chunks); | |
1555 | ||
1556 | for (i = 0; i < num_chunks; i++) { | |
1557 | if (should_compress) | |
1558 | cur_end = min(end, start + SZ_512K - 1); | |
1559 | else | |
1560 | cur_end = end; | |
771ed689 | 1561 | |
bd4691a0 NB |
1562 | /* |
1563 | * igrab is called higher up in the call chain, take only the | |
1564 | * lightweight reference for the callback lifetime | |
1565 | */ | |
751b6431 | 1566 | ihold(&inode->vfs_inode); |
9e895a8f | 1567 | async_chunk[i].async_cow = ctx; |
751b6431 | 1568 | async_chunk[i].inode = &inode->vfs_inode; |
97db1204 NB |
1569 | async_chunk[i].start = start; |
1570 | async_chunk[i].end = cur_end; | |
97db1204 NB |
1571 | async_chunk[i].write_flags = write_flags; |
1572 | INIT_LIST_HEAD(&async_chunk[i].extents); | |
1573 | ||
1d53c9e6 CM |
1574 | /* |
1575 | * The locked_page comes all the way from writepage and its | |
1576 | * the original page we were actually given. As we spread | |
1577 | * this large delalloc region across multiple async_chunk | |
1578 | * structs, only the first struct needs a pointer to locked_page | |
1579 | * | |
1580 | * This way we don't need racey decisions about who is supposed | |
1581 | * to unlock it. | |
1582 | */ | |
1583 | if (locked_page) { | |
ec39f769 CM |
1584 | /* |
1585 | * Depending on the compressibility, the pages might or | |
1586 | * might not go through async. We want all of them to | |
1587 | * be accounted against wbc once. Let's do it here | |
1588 | * before the paths diverge. wbc accounting is used | |
1589 | * only for foreign writeback detection and doesn't | |
1590 | * need full accuracy. Just account the whole thing | |
1591 | * against the first page. | |
1592 | */ | |
1593 | wbc_account_cgroup_owner(wbc, locked_page, | |
1594 | cur_end - start); | |
1d53c9e6 CM |
1595 | async_chunk[i].locked_page = locked_page; |
1596 | locked_page = NULL; | |
1597 | } else { | |
1598 | async_chunk[i].locked_page = NULL; | |
1599 | } | |
1600 | ||
ec39f769 CM |
1601 | if (blkcg_css != blkcg_root_css) { |
1602 | css_get(blkcg_css); | |
1603 | async_chunk[i].blkcg_css = blkcg_css; | |
1604 | } else { | |
1605 | async_chunk[i].blkcg_css = NULL; | |
1606 | } | |
1607 | ||
a0cac0ec OS |
1608 | btrfs_init_work(&async_chunk[i].work, async_cow_start, |
1609 | async_cow_submit, async_cow_free); | |
771ed689 | 1610 | |
97db1204 | 1611 | nr_pages = DIV_ROUND_UP(cur_end - start, PAGE_SIZE); |
0b246afa | 1612 | atomic_add(nr_pages, &fs_info->async_delalloc_pages); |
771ed689 | 1613 | |
97db1204 | 1614 | btrfs_queue_work(fs_info->delalloc_workers, &async_chunk[i].work); |
771ed689 | 1615 | |
771ed689 CM |
1616 | *nr_written += nr_pages; |
1617 | start = cur_end + 1; | |
1618 | } | |
1619 | *page_started = 1; | |
1620 | return 0; | |
be20aa9d CM |
1621 | } |
1622 | ||
42c01100 NA |
1623 | static noinline int run_delalloc_zoned(struct btrfs_inode *inode, |
1624 | struct page *locked_page, u64 start, | |
1625 | u64 end, int *page_started, | |
1626 | unsigned long *nr_written) | |
1627 | { | |
898793d9 | 1628 | u64 done_offset = end; |
42c01100 | 1629 | int ret; |
898793d9 | 1630 | bool locked_page_done = false; |
42c01100 | 1631 | |
898793d9 NA |
1632 | while (start <= end) { |
1633 | ret = cow_file_range(inode, locked_page, start, end, page_started, | |
1634 | nr_written, 0, &done_offset); | |
1635 | if (ret && ret != -EAGAIN) | |
1636 | return ret; | |
42c01100 | 1637 | |
898793d9 NA |
1638 | if (*page_started) { |
1639 | ASSERT(ret == 0); | |
1640 | return 0; | |
1641 | } | |
1642 | ||
1643 | if (ret == 0) | |
1644 | done_offset = end; | |
1645 | ||
2ce543f4 | 1646 | if (done_offset == start) { |
d5b81ced NA |
1647 | wait_on_bit_io(&inode->root->fs_info->flags, |
1648 | BTRFS_FS_NEED_ZONE_FINISH, | |
1649 | TASK_UNINTERRUPTIBLE); | |
2ce543f4 NA |
1650 | continue; |
1651 | } | |
898793d9 NA |
1652 | |
1653 | if (!locked_page_done) { | |
1654 | __set_page_dirty_nobuffers(locked_page); | |
1655 | account_page_redirty(locked_page); | |
1656 | } | |
1657 | locked_page_done = true; | |
1658 | extent_write_locked_range(&inode->vfs_inode, start, done_offset); | |
1659 | ||
1660 | start = done_offset + 1; | |
1661 | } | |
42c01100 | 1662 | |
42c01100 NA |
1663 | *page_started = 1; |
1664 | ||
1665 | return 0; | |
1666 | } | |
1667 | ||
2ff7e61e | 1668 | static noinline int csum_exist_in_range(struct btrfs_fs_info *fs_info, |
26ce9114 | 1669 | u64 bytenr, u64 num_bytes, bool nowait) |
17d217fe | 1670 | { |
fc28b25e | 1671 | struct btrfs_root *csum_root = btrfs_csum_root(fs_info, bytenr); |
17d217fe | 1672 | struct btrfs_ordered_sum *sums; |
fc28b25e | 1673 | int ret; |
17d217fe YZ |
1674 | LIST_HEAD(list); |
1675 | ||
fc28b25e | 1676 | ret = btrfs_lookup_csums_range(csum_root, bytenr, |
26ce9114 JB |
1677 | bytenr + num_bytes - 1, &list, 0, |
1678 | nowait); | |
17d217fe YZ |
1679 | if (ret == 0 && list_empty(&list)) |
1680 | return 0; | |
1681 | ||
1682 | while (!list_empty(&list)) { | |
1683 | sums = list_entry(list.next, struct btrfs_ordered_sum, list); | |
1684 | list_del(&sums->list); | |
1685 | kfree(sums); | |
1686 | } | |
58113753 LB |
1687 | if (ret < 0) |
1688 | return ret; | |
17d217fe YZ |
1689 | return 1; |
1690 | } | |
1691 | ||
8ba96f3d | 1692 | static int fallback_to_cow(struct btrfs_inode *inode, struct page *locked_page, |
467dc47e FM |
1693 | const u64 start, const u64 end, |
1694 | int *page_started, unsigned long *nr_written) | |
1695 | { | |
8ba96f3d | 1696 | const bool is_space_ino = btrfs_is_free_space_inode(inode); |
37f00a6d | 1697 | const bool is_reloc_ino = btrfs_is_data_reloc_root(inode->root); |
2166e5ed | 1698 | const u64 range_bytes = end + 1 - start; |
8ba96f3d | 1699 | struct extent_io_tree *io_tree = &inode->io_tree; |
467dc47e FM |
1700 | u64 range_start = start; |
1701 | u64 count; | |
1702 | ||
1703 | /* | |
1704 | * If EXTENT_NORESERVE is set it means that when the buffered write was | |
1705 | * made we had not enough available data space and therefore we did not | |
1706 | * reserve data space for it, since we though we could do NOCOW for the | |
1707 | * respective file range (either there is prealloc extent or the inode | |
1708 | * has the NOCOW bit set). | |
1709 | * | |
1710 | * However when we need to fallback to COW mode (because for example the | |
1711 | * block group for the corresponding extent was turned to RO mode by a | |
1712 | * scrub or relocation) we need to do the following: | |
1713 | * | |
1714 | * 1) We increment the bytes_may_use counter of the data space info. | |
1715 | * If COW succeeds, it allocates a new data extent and after doing | |
1716 | * that it decrements the space info's bytes_may_use counter and | |
1717 | * increments its bytes_reserved counter by the same amount (we do | |
1718 | * this at btrfs_add_reserved_bytes()). So we need to increment the | |
1719 | * bytes_may_use counter to compensate (when space is reserved at | |
1720 | * buffered write time, the bytes_may_use counter is incremented); | |
1721 | * | |
1722 | * 2) We clear the EXTENT_NORESERVE bit from the range. We do this so | |
1723 | * that if the COW path fails for any reason, it decrements (through | |
1724 | * extent_clear_unlock_delalloc()) the bytes_may_use counter of the | |
1725 | * data space info, which we incremented in the step above. | |
2166e5ed FM |
1726 | * |
1727 | * If we need to fallback to cow and the inode corresponds to a free | |
6bd335b4 FM |
1728 | * space cache inode or an inode of the data relocation tree, we must |
1729 | * also increment bytes_may_use of the data space_info for the same | |
1730 | * reason. Space caches and relocated data extents always get a prealloc | |
2166e5ed | 1731 | * extent for them, however scrub or balance may have set the block |
6bd335b4 FM |
1732 | * group that contains that extent to RO mode and therefore force COW |
1733 | * when starting writeback. | |
467dc47e | 1734 | */ |
2166e5ed | 1735 | count = count_range_bits(io_tree, &range_start, end, range_bytes, |
467dc47e | 1736 | EXTENT_NORESERVE, 0); |
6bd335b4 FM |
1737 | if (count > 0 || is_space_ino || is_reloc_ino) { |
1738 | u64 bytes = count; | |
8ba96f3d | 1739 | struct btrfs_fs_info *fs_info = inode->root->fs_info; |
467dc47e FM |
1740 | struct btrfs_space_info *sinfo = fs_info->data_sinfo; |
1741 | ||
6bd335b4 FM |
1742 | if (is_space_ino || is_reloc_ino) |
1743 | bytes = range_bytes; | |
1744 | ||
467dc47e | 1745 | spin_lock(&sinfo->lock); |
2166e5ed | 1746 | btrfs_space_info_update_bytes_may_use(fs_info, sinfo, bytes); |
467dc47e FM |
1747 | spin_unlock(&sinfo->lock); |
1748 | ||
2166e5ed FM |
1749 | if (count > 0) |
1750 | clear_extent_bit(io_tree, start, end, EXTENT_NORESERVE, | |
bd015294 | 1751 | NULL); |
467dc47e FM |
1752 | } |
1753 | ||
8ba96f3d | 1754 | return cow_file_range(inode, locked_page, start, end, page_started, |
898793d9 | 1755 | nr_written, 1, NULL); |
467dc47e FM |
1756 | } |
1757 | ||
619104ba FM |
1758 | struct can_nocow_file_extent_args { |
1759 | /* Input fields. */ | |
1760 | ||
1761 | /* Start file offset of the range we want to NOCOW. */ | |
1762 | u64 start; | |
1763 | /* End file offset (inclusive) of the range we want to NOCOW. */ | |
1764 | u64 end; | |
1765 | bool writeback_path; | |
1766 | bool strict; | |
1767 | /* | |
1768 | * Free the path passed to can_nocow_file_extent() once it's not needed | |
1769 | * anymore. | |
1770 | */ | |
1771 | bool free_path; | |
1772 | ||
1773 | /* Output fields. Only set when can_nocow_file_extent() returns 1. */ | |
1774 | ||
1775 | u64 disk_bytenr; | |
1776 | u64 disk_num_bytes; | |
1777 | u64 extent_offset; | |
1778 | /* Number of bytes that can be written to in NOCOW mode. */ | |
1779 | u64 num_bytes; | |
1780 | }; | |
1781 | ||
1782 | /* | |
1783 | * Check if we can NOCOW the file extent that the path points to. | |
1784 | * This function may return with the path released, so the caller should check | |
1785 | * if path->nodes[0] is NULL or not if it needs to use the path afterwards. | |
1786 | * | |
1787 | * Returns: < 0 on error | |
1788 | * 0 if we can not NOCOW | |
1789 | * 1 if we can NOCOW | |
1790 | */ | |
1791 | static int can_nocow_file_extent(struct btrfs_path *path, | |
1792 | struct btrfs_key *key, | |
1793 | struct btrfs_inode *inode, | |
1794 | struct can_nocow_file_extent_args *args) | |
1795 | { | |
1796 | const bool is_freespace_inode = btrfs_is_free_space_inode(inode); | |
1797 | struct extent_buffer *leaf = path->nodes[0]; | |
1798 | struct btrfs_root *root = inode->root; | |
1799 | struct btrfs_file_extent_item *fi; | |
1800 | u64 extent_end; | |
1801 | u8 extent_type; | |
1802 | int can_nocow = 0; | |
1803 | int ret = 0; | |
26ce9114 | 1804 | bool nowait = path->nowait; |
619104ba FM |
1805 | |
1806 | fi = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_file_extent_item); | |
1807 | extent_type = btrfs_file_extent_type(leaf, fi); | |
1808 | ||
1809 | if (extent_type == BTRFS_FILE_EXTENT_INLINE) | |
1810 | goto out; | |
1811 | ||
1812 | /* Can't access these fields unless we know it's not an inline extent. */ | |
1813 | args->disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); | |
1814 | args->disk_num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi); | |
1815 | args->extent_offset = btrfs_file_extent_offset(leaf, fi); | |
1816 | ||
1817 | if (!(inode->flags & BTRFS_INODE_NODATACOW) && | |
1818 | extent_type == BTRFS_FILE_EXTENT_REG) | |
1819 | goto out; | |
1820 | ||
1821 | /* | |
1822 | * If the extent was created before the generation where the last snapshot | |
1823 | * for its subvolume was created, then this implies the extent is shared, | |
1824 | * hence we must COW. | |
1825 | */ | |
a7bb6bd4 | 1826 | if (!args->strict && |
619104ba FM |
1827 | btrfs_file_extent_generation(leaf, fi) <= |
1828 | btrfs_root_last_snapshot(&root->root_item)) | |
1829 | goto out; | |
1830 | ||
1831 | /* An explicit hole, must COW. */ | |
1832 | if (args->disk_bytenr == 0) | |
1833 | goto out; | |
1834 | ||
1835 | /* Compressed/encrypted/encoded extents must be COWed. */ | |
1836 | if (btrfs_file_extent_compression(leaf, fi) || | |
1837 | btrfs_file_extent_encryption(leaf, fi) || | |
1838 | btrfs_file_extent_other_encoding(leaf, fi)) | |
1839 | goto out; | |
1840 | ||
1841 | extent_end = btrfs_file_extent_end(path); | |
1842 | ||
1843 | /* | |
1844 | * The following checks can be expensive, as they need to take other | |
1845 | * locks and do btree or rbtree searches, so release the path to avoid | |
1846 | * blocking other tasks for too long. | |
1847 | */ | |
1848 | btrfs_release_path(path); | |
1849 | ||
1850 | ret = btrfs_cross_ref_exist(root, btrfs_ino(inode), | |
1851 | key->offset - args->extent_offset, | |
1852 | args->disk_bytenr, false, path); | |
1853 | WARN_ON_ONCE(ret > 0 && is_freespace_inode); | |
1854 | if (ret != 0) | |
1855 | goto out; | |
1856 | ||
1857 | if (args->free_path) { | |
1858 | /* | |
1859 | * We don't need the path anymore, plus through the | |
1860 | * csum_exist_in_range() call below we will end up allocating | |
1861 | * another path. So free the path to avoid unnecessary extra | |
1862 | * memory usage. | |
1863 | */ | |
1864 | btrfs_free_path(path); | |
1865 | path = NULL; | |
1866 | } | |
1867 | ||
1868 | /* If there are pending snapshots for this root, we must COW. */ | |
1869 | if (args->writeback_path && !is_freespace_inode && | |
1870 | atomic_read(&root->snapshot_force_cow)) | |
1871 | goto out; | |
1872 | ||
1873 | args->disk_bytenr += args->extent_offset; | |
1874 | args->disk_bytenr += args->start - key->offset; | |
1875 | args->num_bytes = min(args->end + 1, extent_end) - args->start; | |
1876 | ||
1877 | /* | |
1878 | * Force COW if csums exist in the range. This ensures that csums for a | |
1879 | * given extent are either valid or do not exist. | |
1880 | */ | |
26ce9114 JB |
1881 | ret = csum_exist_in_range(root->fs_info, args->disk_bytenr, args->num_bytes, |
1882 | nowait); | |
619104ba FM |
1883 | WARN_ON_ONCE(ret > 0 && is_freespace_inode); |
1884 | if (ret != 0) | |
1885 | goto out; | |
1886 | ||
1887 | can_nocow = 1; | |
1888 | out: | |
1889 | if (args->free_path && path) | |
1890 | btrfs_free_path(path); | |
1891 | ||
1892 | return ret < 0 ? ret : can_nocow; | |
1893 | } | |
1894 | ||
d352ac68 CM |
1895 | /* |
1896 | * when nowcow writeback call back. This checks for snapshots or COW copies | |
1897 | * of the extents that exist in the file, and COWs the file as required. | |
1898 | * | |
1899 | * If no cow copies or snapshots exist, we write directly to the existing | |
1900 | * blocks on disk | |
1901 | */ | |
968322c8 | 1902 | static noinline int run_delalloc_nocow(struct btrfs_inode *inode, |
7f366cfe | 1903 | struct page *locked_page, |
3e024846 | 1904 | const u64 start, const u64 end, |
6e65ae76 | 1905 | int *page_started, |
3e024846 | 1906 | unsigned long *nr_written) |
be20aa9d | 1907 | { |
968322c8 NB |
1908 | struct btrfs_fs_info *fs_info = inode->root->fs_info; |
1909 | struct btrfs_root *root = inode->root; | |
be20aa9d | 1910 | struct btrfs_path *path; |
3e024846 NB |
1911 | u64 cow_start = (u64)-1; |
1912 | u64 cur_offset = start; | |
8ecebf4d | 1913 | int ret; |
3e024846 | 1914 | bool check_prev = true; |
968322c8 | 1915 | u64 ino = btrfs_ino(inode); |
2306e83e | 1916 | struct btrfs_block_group *bg; |
762bf098 | 1917 | bool nocow = false; |
619104ba | 1918 | struct can_nocow_file_extent_args nocow_args = { 0 }; |
be20aa9d CM |
1919 | |
1920 | path = btrfs_alloc_path(); | |
17ca04af | 1921 | if (!path) { |
968322c8 | 1922 | extent_clear_unlock_delalloc(inode, start, end, locked_page, |
c2790a2e | 1923 | EXTENT_LOCKED | EXTENT_DELALLOC | |
151a41bc JB |
1924 | EXTENT_DO_ACCOUNTING | |
1925 | EXTENT_DEFRAG, PAGE_UNLOCK | | |
6869b0a8 | 1926 | PAGE_START_WRITEBACK | |
c2790a2e | 1927 | PAGE_END_WRITEBACK); |
d8926bb3 | 1928 | return -ENOMEM; |
17ca04af | 1929 | } |
82d5902d | 1930 | |
619104ba FM |
1931 | nocow_args.end = end; |
1932 | nocow_args.writeback_path = true; | |
1933 | ||
80ff3856 | 1934 | while (1) { |
3e024846 NB |
1935 | struct btrfs_key found_key; |
1936 | struct btrfs_file_extent_item *fi; | |
1937 | struct extent_buffer *leaf; | |
1938 | u64 extent_end; | |
3e024846 | 1939 | u64 ram_bytes; |
619104ba | 1940 | u64 nocow_end; |
3e024846 | 1941 | int extent_type; |
762bf098 NB |
1942 | |
1943 | nocow = false; | |
3e024846 | 1944 | |
e4c3b2dc | 1945 | ret = btrfs_lookup_file_extent(NULL, root, path, ino, |
80ff3856 | 1946 | cur_offset, 0); |
d788a349 | 1947 | if (ret < 0) |
79787eaa | 1948 | goto error; |
a6bd9cd1 NB |
1949 | |
1950 | /* | |
1951 | * If there is no extent for our range when doing the initial | |
1952 | * search, then go back to the previous slot as it will be the | |
1953 | * one containing the search offset | |
1954 | */ | |
80ff3856 YZ |
1955 | if (ret > 0 && path->slots[0] > 0 && check_prev) { |
1956 | leaf = path->nodes[0]; | |
1957 | btrfs_item_key_to_cpu(leaf, &found_key, | |
1958 | path->slots[0] - 1); | |
33345d01 | 1959 | if (found_key.objectid == ino && |
80ff3856 YZ |
1960 | found_key.type == BTRFS_EXTENT_DATA_KEY) |
1961 | path->slots[0]--; | |
1962 | } | |
3e024846 | 1963 | check_prev = false; |
80ff3856 | 1964 | next_slot: |
a6bd9cd1 | 1965 | /* Go to next leaf if we have exhausted the current one */ |
80ff3856 YZ |
1966 | leaf = path->nodes[0]; |
1967 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
1968 | ret = btrfs_next_leaf(root, path); | |
e8916699 LB |
1969 | if (ret < 0) { |
1970 | if (cow_start != (u64)-1) | |
1971 | cur_offset = cow_start; | |
79787eaa | 1972 | goto error; |
e8916699 | 1973 | } |
80ff3856 YZ |
1974 | if (ret > 0) |
1975 | break; | |
1976 | leaf = path->nodes[0]; | |
1977 | } | |
be20aa9d | 1978 | |
80ff3856 YZ |
1979 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
1980 | ||
a6bd9cd1 | 1981 | /* Didn't find anything for our INO */ |
1d512cb7 FM |
1982 | if (found_key.objectid > ino) |
1983 | break; | |
a6bd9cd1 NB |
1984 | /* |
1985 | * Keep searching until we find an EXTENT_ITEM or there are no | |
1986 | * more extents for this inode | |
1987 | */ | |
1d512cb7 FM |
1988 | if (WARN_ON_ONCE(found_key.objectid < ino) || |
1989 | found_key.type < BTRFS_EXTENT_DATA_KEY) { | |
1990 | path->slots[0]++; | |
1991 | goto next_slot; | |
1992 | } | |
a6bd9cd1 NB |
1993 | |
1994 | /* Found key is not EXTENT_DATA_KEY or starts after req range */ | |
1d512cb7 | 1995 | if (found_key.type > BTRFS_EXTENT_DATA_KEY || |
80ff3856 YZ |
1996 | found_key.offset > end) |
1997 | break; | |
1998 | ||
a6bd9cd1 NB |
1999 | /* |
2000 | * If the found extent starts after requested offset, then | |
2001 | * adjust extent_end to be right before this extent begins | |
2002 | */ | |
80ff3856 YZ |
2003 | if (found_key.offset > cur_offset) { |
2004 | extent_end = found_key.offset; | |
e9061e21 | 2005 | extent_type = 0; |
80ff3856 YZ |
2006 | goto out_check; |
2007 | } | |
2008 | ||
a6bd9cd1 NB |
2009 | /* |
2010 | * Found extent which begins before our range and potentially | |
2011 | * intersect it | |
2012 | */ | |
80ff3856 YZ |
2013 | fi = btrfs_item_ptr(leaf, path->slots[0], |
2014 | struct btrfs_file_extent_item); | |
2015 | extent_type = btrfs_file_extent_type(leaf, fi); | |
619104ba FM |
2016 | /* If this is triggered then we have a memory corruption. */ |
2017 | ASSERT(extent_type < BTRFS_NR_FILE_EXTENT_TYPES); | |
2018 | if (WARN_ON(extent_type >= BTRFS_NR_FILE_EXTENT_TYPES)) { | |
2019 | ret = -EUCLEAN; | |
2020 | goto error; | |
2021 | } | |
cc95bef6 | 2022 | ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi); |
619104ba | 2023 | extent_end = btrfs_file_extent_end(path); |
c65ca98f | 2024 | |
619104ba FM |
2025 | /* |
2026 | * If the extent we got ends before our current offset, skip to | |
2027 | * the next extent. | |
2028 | */ | |
2029 | if (extent_end <= cur_offset) { | |
2030 | path->slots[0]++; | |
2031 | goto next_slot; | |
2032 | } | |
c65ca98f | 2033 | |
619104ba FM |
2034 | nocow_args.start = cur_offset; |
2035 | ret = can_nocow_file_extent(path, &found_key, inode, &nocow_args); | |
2036 | if (ret < 0) { | |
2037 | if (cow_start != (u64)-1) | |
2038 | cur_offset = cow_start; | |
2039 | goto error; | |
2040 | } else if (ret == 0) { | |
2041 | goto out_check; | |
2042 | } | |
58113753 | 2043 | |
619104ba | 2044 | ret = 0; |
2306e83e FM |
2045 | bg = btrfs_inc_nocow_writers(fs_info, nocow_args.disk_bytenr); |
2046 | if (bg) | |
3e024846 | 2047 | nocow = true; |
80ff3856 | 2048 | out_check: |
a6bd9cd1 NB |
2049 | /* |
2050 | * If nocow is false then record the beginning of the range | |
2051 | * that needs to be COWed | |
2052 | */ | |
80ff3856 YZ |
2053 | if (!nocow) { |
2054 | if (cow_start == (u64)-1) | |
2055 | cow_start = cur_offset; | |
2056 | cur_offset = extent_end; | |
2057 | if (cur_offset > end) | |
2058 | break; | |
c65ca98f FM |
2059 | if (!path->nodes[0]) |
2060 | continue; | |
80ff3856 YZ |
2061 | path->slots[0]++; |
2062 | goto next_slot; | |
7ea394f1 YZ |
2063 | } |
2064 | ||
a6bd9cd1 NB |
2065 | /* |
2066 | * COW range from cow_start to found_key.offset - 1. As the key | |
2067 | * will contain the beginning of the first extent that can be | |
2068 | * NOCOW, following one which needs to be COW'ed | |
2069 | */ | |
80ff3856 | 2070 | if (cow_start != (u64)-1) { |
968322c8 | 2071 | ret = fallback_to_cow(inode, locked_page, |
8ba96f3d | 2072 | cow_start, found_key.offset - 1, |
467dc47e | 2073 | page_started, nr_written); |
230ed397 | 2074 | if (ret) |
79787eaa | 2075 | goto error; |
80ff3856 | 2076 | cow_start = (u64)-1; |
7ea394f1 | 2077 | } |
80ff3856 | 2078 | |
619104ba FM |
2079 | nocow_end = cur_offset + nocow_args.num_bytes - 1; |
2080 | ||
d899e052 | 2081 | if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) { |
619104ba | 2082 | u64 orig_start = found_key.offset - nocow_args.extent_offset; |
3e024846 | 2083 | struct extent_map *em; |
6f9994db | 2084 | |
619104ba | 2085 | em = create_io_em(inode, cur_offset, nocow_args.num_bytes, |
6f9994db | 2086 | orig_start, |
619104ba FM |
2087 | nocow_args.disk_bytenr, /* block_start */ |
2088 | nocow_args.num_bytes, /* block_len */ | |
2089 | nocow_args.disk_num_bytes, /* orig_block_len */ | |
6f9994db LB |
2090 | ram_bytes, BTRFS_COMPRESS_NONE, |
2091 | BTRFS_ORDERED_PREALLOC); | |
2092 | if (IS_ERR(em)) { | |
6f9994db LB |
2093 | ret = PTR_ERR(em); |
2094 | goto error; | |
d899e052 | 2095 | } |
6f9994db | 2096 | free_extent_map(em); |
cb36a9bb | 2097 | ret = btrfs_add_ordered_extent(inode, |
619104ba FM |
2098 | cur_offset, nocow_args.num_bytes, |
2099 | nocow_args.num_bytes, | |
2100 | nocow_args.disk_bytenr, | |
2101 | nocow_args.num_bytes, 0, | |
cb36a9bb OS |
2102 | 1 << BTRFS_ORDERED_PREALLOC, |
2103 | BTRFS_COMPRESS_NONE); | |
762bf098 | 2104 | if (ret) { |
968322c8 | 2105 | btrfs_drop_extent_cache(inode, cur_offset, |
619104ba | 2106 | nocow_end, 0); |
762bf098 NB |
2107 | goto error; |
2108 | } | |
d899e052 | 2109 | } else { |
968322c8 | 2110 | ret = btrfs_add_ordered_extent(inode, cur_offset, |
619104ba FM |
2111 | nocow_args.num_bytes, |
2112 | nocow_args.num_bytes, | |
2113 | nocow_args.disk_bytenr, | |
2114 | nocow_args.num_bytes, | |
cb36a9bb OS |
2115 | 0, |
2116 | 1 << BTRFS_ORDERED_NOCOW, | |
2117 | BTRFS_COMPRESS_NONE); | |
762bf098 NB |
2118 | if (ret) |
2119 | goto error; | |
d899e052 | 2120 | } |
80ff3856 | 2121 | |
2306e83e FM |
2122 | if (nocow) { |
2123 | btrfs_dec_nocow_writers(bg); | |
2124 | nocow = false; | |
2125 | } | |
771ed689 | 2126 | |
37f00a6d | 2127 | if (btrfs_is_data_reloc_root(root)) |
4dbd80fb QW |
2128 | /* |
2129 | * Error handled later, as we must prevent | |
2130 | * extent_clear_unlock_delalloc() in error handler | |
2131 | * from freeing metadata of created ordered extent. | |
2132 | */ | |
968322c8 | 2133 | ret = btrfs_reloc_clone_csums(inode, cur_offset, |
619104ba | 2134 | nocow_args.num_bytes); |
efa56464 | 2135 | |
619104ba | 2136 | extent_clear_unlock_delalloc(inode, cur_offset, nocow_end, |
c2790a2e | 2137 | locked_page, EXTENT_LOCKED | |
18513091 WX |
2138 | EXTENT_DELALLOC | |
2139 | EXTENT_CLEAR_DATA_RESV, | |
f57ad937 | 2140 | PAGE_UNLOCK | PAGE_SET_ORDERED); |
18513091 | 2141 | |
80ff3856 | 2142 | cur_offset = extent_end; |
4dbd80fb QW |
2143 | |
2144 | /* | |
2145 | * btrfs_reloc_clone_csums() error, now we're OK to call error | |
2146 | * handler, as metadata for created ordered extent will only | |
2147 | * be freed by btrfs_finish_ordered_io(). | |
2148 | */ | |
2149 | if (ret) | |
2150 | goto error; | |
80ff3856 YZ |
2151 | if (cur_offset > end) |
2152 | break; | |
be20aa9d | 2153 | } |
b3b4aa74 | 2154 | btrfs_release_path(path); |
80ff3856 | 2155 | |
506481b2 | 2156 | if (cur_offset <= end && cow_start == (u64)-1) |
80ff3856 | 2157 | cow_start = cur_offset; |
17ca04af | 2158 | |
80ff3856 | 2159 | if (cow_start != (u64)-1) { |
506481b2 | 2160 | cur_offset = end; |
968322c8 NB |
2161 | ret = fallback_to_cow(inode, locked_page, cow_start, end, |
2162 | page_started, nr_written); | |
d788a349 | 2163 | if (ret) |
79787eaa | 2164 | goto error; |
80ff3856 YZ |
2165 | } |
2166 | ||
79787eaa | 2167 | error: |
762bf098 | 2168 | if (nocow) |
2306e83e | 2169 | btrfs_dec_nocow_writers(bg); |
762bf098 | 2170 | |
17ca04af | 2171 | if (ret && cur_offset < end) |
968322c8 | 2172 | extent_clear_unlock_delalloc(inode, cur_offset, end, |
c2790a2e | 2173 | locked_page, EXTENT_LOCKED | |
151a41bc JB |
2174 | EXTENT_DELALLOC | EXTENT_DEFRAG | |
2175 | EXTENT_DO_ACCOUNTING, PAGE_UNLOCK | | |
6869b0a8 | 2176 | PAGE_START_WRITEBACK | |
c2790a2e | 2177 | PAGE_END_WRITEBACK); |
7ea394f1 | 2178 | btrfs_free_path(path); |
79787eaa | 2179 | return ret; |
be20aa9d CM |
2180 | } |
2181 | ||
6e65ae76 | 2182 | static bool should_nocow(struct btrfs_inode *inode, u64 start, u64 end) |
47059d93 | 2183 | { |
6e65ae76 GR |
2184 | if (inode->flags & (BTRFS_INODE_NODATACOW | BTRFS_INODE_PREALLOC)) { |
2185 | if (inode->defrag_bytes && | |
2186 | test_range_bit(&inode->io_tree, start, end, EXTENT_DEFRAG, | |
2187 | 0, NULL)) | |
2188 | return false; | |
2189 | return true; | |
2190 | } | |
2191 | return false; | |
47059d93 WS |
2192 | } |
2193 | ||
d352ac68 | 2194 | /* |
5eaad97a NB |
2195 | * Function to process delayed allocation (create CoW) for ranges which are |
2196 | * being touched for the first time. | |
d352ac68 | 2197 | */ |
98456b9c | 2198 | int btrfs_run_delalloc_range(struct btrfs_inode *inode, struct page *locked_page, |
5eaad97a NB |
2199 | u64 start, u64 end, int *page_started, unsigned long *nr_written, |
2200 | struct writeback_control *wbc) | |
be20aa9d | 2201 | { |
be20aa9d | 2202 | int ret; |
42c01100 | 2203 | const bool zoned = btrfs_is_zoned(inode->root->fs_info); |
a2135011 | 2204 | |
2749f7ef QW |
2205 | /* |
2206 | * The range must cover part of the @locked_page, or the returned | |
2207 | * @page_started can confuse the caller. | |
2208 | */ | |
2209 | ASSERT(!(end <= page_offset(locked_page) || | |
2210 | start >= page_offset(locked_page) + PAGE_SIZE)); | |
2211 | ||
6e65ae76 | 2212 | if (should_nocow(inode, start, end)) { |
2adada88 JT |
2213 | /* |
2214 | * Normally on a zoned device we're only doing COW writes, but | |
2215 | * in case of relocation on a zoned filesystem we have taken | |
2216 | * precaution, that we're only writing sequentially. It's safe | |
2217 | * to use run_delalloc_nocow() here, like for regular | |
2218 | * preallocated inodes. | |
2219 | */ | |
9435be73 | 2220 | ASSERT(!zoned || btrfs_is_data_reloc_root(inode->root)); |
98456b9c | 2221 | ret = run_delalloc_nocow(inode, locked_page, start, end, |
6e65ae76 | 2222 | page_started, nr_written); |
e6f9d696 | 2223 | } else if (!btrfs_inode_can_compress(inode) || |
98456b9c | 2224 | !inode_need_compress(inode, start, end)) { |
42c01100 NA |
2225 | if (zoned) |
2226 | ret = run_delalloc_zoned(inode, locked_page, start, end, | |
2227 | page_started, nr_written); | |
2228 | else | |
2229 | ret = cow_file_range(inode, locked_page, start, end, | |
898793d9 | 2230 | page_started, nr_written, 1, NULL); |
7ddf5a42 | 2231 | } else { |
98456b9c NB |
2232 | set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT, &inode->runtime_flags); |
2233 | ret = cow_file_range_async(inode, wbc, locked_page, start, end, | |
fac07d2b | 2234 | page_started, nr_written); |
7ddf5a42 | 2235 | } |
7361b4ae | 2236 | ASSERT(ret <= 0); |
52427260 | 2237 | if (ret) |
98456b9c | 2238 | btrfs_cleanup_ordered_extents(inode, locked_page, start, |
d1051d6e | 2239 | end - start + 1); |
b888db2b CM |
2240 | return ret; |
2241 | } | |
2242 | ||
abbb55f4 NB |
2243 | void btrfs_split_delalloc_extent(struct inode *inode, |
2244 | struct extent_state *orig, u64 split) | |
9ed74f2d | 2245 | { |
f7b12a62 | 2246 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
dcab6a3b JB |
2247 | u64 size; |
2248 | ||
0ca1f7ce | 2249 | /* not delalloc, ignore it */ |
9ed74f2d | 2250 | if (!(orig->state & EXTENT_DELALLOC)) |
1bf85046 | 2251 | return; |
9ed74f2d | 2252 | |
dcab6a3b | 2253 | size = orig->end - orig->start + 1; |
f7b12a62 | 2254 | if (size > fs_info->max_extent_size) { |
823bb20a | 2255 | u32 num_extents; |
dcab6a3b JB |
2256 | u64 new_size; |
2257 | ||
2258 | /* | |
5c848198 | 2259 | * See the explanation in btrfs_merge_delalloc_extent, the same |
ba117213 | 2260 | * applies here, just in reverse. |
dcab6a3b JB |
2261 | */ |
2262 | new_size = orig->end - split + 1; | |
7d7672bc | 2263 | num_extents = count_max_extents(fs_info, new_size); |
ba117213 | 2264 | new_size = split - orig->start; |
7d7672bc NA |
2265 | num_extents += count_max_extents(fs_info, new_size); |
2266 | if (count_max_extents(fs_info, size) >= num_extents) | |
dcab6a3b JB |
2267 | return; |
2268 | } | |
2269 | ||
9e0baf60 | 2270 | spin_lock(&BTRFS_I(inode)->lock); |
8b62f87b | 2271 | btrfs_mod_outstanding_extents(BTRFS_I(inode), 1); |
9e0baf60 | 2272 | spin_unlock(&BTRFS_I(inode)->lock); |
9ed74f2d JB |
2273 | } |
2274 | ||
2275 | /* | |
5c848198 NB |
2276 | * Handle merged delayed allocation extents so we can keep track of new extents |
2277 | * that are just merged onto old extents, such as when we are doing sequential | |
2278 | * writes, so we can properly account for the metadata space we'll need. | |
9ed74f2d | 2279 | */ |
5c848198 NB |
2280 | void btrfs_merge_delalloc_extent(struct inode *inode, struct extent_state *new, |
2281 | struct extent_state *other) | |
9ed74f2d | 2282 | { |
f7b12a62 | 2283 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
dcab6a3b | 2284 | u64 new_size, old_size; |
823bb20a | 2285 | u32 num_extents; |
dcab6a3b | 2286 | |
9ed74f2d JB |
2287 | /* not delalloc, ignore it */ |
2288 | if (!(other->state & EXTENT_DELALLOC)) | |
1bf85046 | 2289 | return; |
9ed74f2d | 2290 | |
8461a3de JB |
2291 | if (new->start > other->start) |
2292 | new_size = new->end - other->start + 1; | |
2293 | else | |
2294 | new_size = other->end - new->start + 1; | |
dcab6a3b JB |
2295 | |
2296 | /* we're not bigger than the max, unreserve the space and go */ | |
f7b12a62 | 2297 | if (new_size <= fs_info->max_extent_size) { |
dcab6a3b | 2298 | spin_lock(&BTRFS_I(inode)->lock); |
8b62f87b | 2299 | btrfs_mod_outstanding_extents(BTRFS_I(inode), -1); |
dcab6a3b JB |
2300 | spin_unlock(&BTRFS_I(inode)->lock); |
2301 | return; | |
2302 | } | |
2303 | ||
2304 | /* | |
ba117213 JB |
2305 | * We have to add up either side to figure out how many extents were |
2306 | * accounted for before we merged into one big extent. If the number of | |
2307 | * extents we accounted for is <= the amount we need for the new range | |
2308 | * then we can return, otherwise drop. Think of it like this | |
2309 | * | |
2310 | * [ 4k][MAX_SIZE] | |
2311 | * | |
2312 | * So we've grown the extent by a MAX_SIZE extent, this would mean we | |
2313 | * need 2 outstanding extents, on one side we have 1 and the other side | |
2314 | * we have 1 so they are == and we can return. But in this case | |
2315 | * | |
2316 | * [MAX_SIZE+4k][MAX_SIZE+4k] | |
2317 | * | |
2318 | * Each range on their own accounts for 2 extents, but merged together | |
2319 | * they are only 3 extents worth of accounting, so we need to drop in | |
2320 | * this case. | |
dcab6a3b | 2321 | */ |
ba117213 | 2322 | old_size = other->end - other->start + 1; |
7d7672bc | 2323 | num_extents = count_max_extents(fs_info, old_size); |
ba117213 | 2324 | old_size = new->end - new->start + 1; |
7d7672bc NA |
2325 | num_extents += count_max_extents(fs_info, old_size); |
2326 | if (count_max_extents(fs_info, new_size) >= num_extents) | |
dcab6a3b JB |
2327 | return; |
2328 | ||
9e0baf60 | 2329 | spin_lock(&BTRFS_I(inode)->lock); |
8b62f87b | 2330 | btrfs_mod_outstanding_extents(BTRFS_I(inode), -1); |
9e0baf60 | 2331 | spin_unlock(&BTRFS_I(inode)->lock); |
9ed74f2d JB |
2332 | } |
2333 | ||
eb73c1b7 MX |
2334 | static void btrfs_add_delalloc_inodes(struct btrfs_root *root, |
2335 | struct inode *inode) | |
2336 | { | |
0b246afa JM |
2337 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
2338 | ||
eb73c1b7 MX |
2339 | spin_lock(&root->delalloc_lock); |
2340 | if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) { | |
2341 | list_add_tail(&BTRFS_I(inode)->delalloc_inodes, | |
2342 | &root->delalloc_inodes); | |
2343 | set_bit(BTRFS_INODE_IN_DELALLOC_LIST, | |
2344 | &BTRFS_I(inode)->runtime_flags); | |
2345 | root->nr_delalloc_inodes++; | |
2346 | if (root->nr_delalloc_inodes == 1) { | |
0b246afa | 2347 | spin_lock(&fs_info->delalloc_root_lock); |
eb73c1b7 MX |
2348 | BUG_ON(!list_empty(&root->delalloc_root)); |
2349 | list_add_tail(&root->delalloc_root, | |
0b246afa JM |
2350 | &fs_info->delalloc_roots); |
2351 | spin_unlock(&fs_info->delalloc_root_lock); | |
eb73c1b7 MX |
2352 | } |
2353 | } | |
2354 | spin_unlock(&root->delalloc_lock); | |
2355 | } | |
2356 | ||
2b877331 NB |
2357 | |
2358 | void __btrfs_del_delalloc_inode(struct btrfs_root *root, | |
2359 | struct btrfs_inode *inode) | |
eb73c1b7 | 2360 | { |
3ffbd68c | 2361 | struct btrfs_fs_info *fs_info = root->fs_info; |
0b246afa | 2362 | |
9e3e97f4 NB |
2363 | if (!list_empty(&inode->delalloc_inodes)) { |
2364 | list_del_init(&inode->delalloc_inodes); | |
eb73c1b7 | 2365 | clear_bit(BTRFS_INODE_IN_DELALLOC_LIST, |
9e3e97f4 | 2366 | &inode->runtime_flags); |
eb73c1b7 MX |
2367 | root->nr_delalloc_inodes--; |
2368 | if (!root->nr_delalloc_inodes) { | |
7c8a0d36 | 2369 | ASSERT(list_empty(&root->delalloc_inodes)); |
0b246afa | 2370 | spin_lock(&fs_info->delalloc_root_lock); |
eb73c1b7 MX |
2371 | BUG_ON(list_empty(&root->delalloc_root)); |
2372 | list_del_init(&root->delalloc_root); | |
0b246afa | 2373 | spin_unlock(&fs_info->delalloc_root_lock); |
eb73c1b7 MX |
2374 | } |
2375 | } | |
2b877331 NB |
2376 | } |
2377 | ||
2378 | static void btrfs_del_delalloc_inode(struct btrfs_root *root, | |
2379 | struct btrfs_inode *inode) | |
2380 | { | |
2381 | spin_lock(&root->delalloc_lock); | |
2382 | __btrfs_del_delalloc_inode(root, inode); | |
eb73c1b7 MX |
2383 | spin_unlock(&root->delalloc_lock); |
2384 | } | |
2385 | ||
d352ac68 | 2386 | /* |
e06a1fc9 NB |
2387 | * Properly track delayed allocation bytes in the inode and to maintain the |
2388 | * list of inodes that have pending delalloc work to be done. | |
d352ac68 | 2389 | */ |
e06a1fc9 | 2390 | void btrfs_set_delalloc_extent(struct inode *inode, struct extent_state *state, |
6d92b304 | 2391 | u32 bits) |
291d673e | 2392 | { |
0b246afa JM |
2393 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
2394 | ||
6d92b304 | 2395 | if ((bits & EXTENT_DEFRAG) && !(bits & EXTENT_DELALLOC)) |
47059d93 | 2396 | WARN_ON(1); |
75eff68e CM |
2397 | /* |
2398 | * set_bit and clear bit hooks normally require _irqsave/restore | |
27160b6b | 2399 | * but in this case, we are only testing for the DELALLOC |
75eff68e CM |
2400 | * bit, which is only set or cleared with irqs on |
2401 | */ | |
6d92b304 | 2402 | if (!(state->state & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) { |
291d673e | 2403 | struct btrfs_root *root = BTRFS_I(inode)->root; |
0ca1f7ce | 2404 | u64 len = state->end + 1 - state->start; |
7d7672bc | 2405 | u32 num_extents = count_max_extents(fs_info, len); |
70ddc553 | 2406 | bool do_list = !btrfs_is_free_space_inode(BTRFS_I(inode)); |
9ed74f2d | 2407 | |
8b62f87b JB |
2408 | spin_lock(&BTRFS_I(inode)->lock); |
2409 | btrfs_mod_outstanding_extents(BTRFS_I(inode), num_extents); | |
2410 | spin_unlock(&BTRFS_I(inode)->lock); | |
287a0ab9 | 2411 | |
6a3891c5 | 2412 | /* For sanity tests */ |
0b246afa | 2413 | if (btrfs_is_testing(fs_info)) |
6a3891c5 JB |
2414 | return; |
2415 | ||
104b4e51 NB |
2416 | percpu_counter_add_batch(&fs_info->delalloc_bytes, len, |
2417 | fs_info->delalloc_batch); | |
df0af1a5 | 2418 | spin_lock(&BTRFS_I(inode)->lock); |
0ca1f7ce | 2419 | BTRFS_I(inode)->delalloc_bytes += len; |
6d92b304 | 2420 | if (bits & EXTENT_DEFRAG) |
47059d93 | 2421 | BTRFS_I(inode)->defrag_bytes += len; |
df0af1a5 | 2422 | if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST, |
eb73c1b7 MX |
2423 | &BTRFS_I(inode)->runtime_flags)) |
2424 | btrfs_add_delalloc_inodes(root, inode); | |
df0af1a5 | 2425 | spin_unlock(&BTRFS_I(inode)->lock); |
291d673e | 2426 | } |
a7e3b975 FM |
2427 | |
2428 | if (!(state->state & EXTENT_DELALLOC_NEW) && | |
6d92b304 | 2429 | (bits & EXTENT_DELALLOC_NEW)) { |
a7e3b975 FM |
2430 | spin_lock(&BTRFS_I(inode)->lock); |
2431 | BTRFS_I(inode)->new_delalloc_bytes += state->end + 1 - | |
2432 | state->start; | |
2433 | spin_unlock(&BTRFS_I(inode)->lock); | |
2434 | } | |
291d673e CM |
2435 | } |
2436 | ||
d352ac68 | 2437 | /* |
a36bb5f9 NB |
2438 | * Once a range is no longer delalloc this function ensures that proper |
2439 | * accounting happens. | |
d352ac68 | 2440 | */ |
a36bb5f9 | 2441 | void btrfs_clear_delalloc_extent(struct inode *vfs_inode, |
6d92b304 | 2442 | struct extent_state *state, u32 bits) |
291d673e | 2443 | { |
a36bb5f9 NB |
2444 | struct btrfs_inode *inode = BTRFS_I(vfs_inode); |
2445 | struct btrfs_fs_info *fs_info = btrfs_sb(vfs_inode->i_sb); | |
47059d93 | 2446 | u64 len = state->end + 1 - state->start; |
7d7672bc | 2447 | u32 num_extents = count_max_extents(fs_info, len); |
47059d93 | 2448 | |
6d92b304 | 2449 | if ((state->state & EXTENT_DEFRAG) && (bits & EXTENT_DEFRAG)) { |
4a4b964f | 2450 | spin_lock(&inode->lock); |
6fc0ef68 | 2451 | inode->defrag_bytes -= len; |
4a4b964f FM |
2452 | spin_unlock(&inode->lock); |
2453 | } | |
47059d93 | 2454 | |
75eff68e CM |
2455 | /* |
2456 | * set_bit and clear bit hooks normally require _irqsave/restore | |
27160b6b | 2457 | * but in this case, we are only testing for the DELALLOC |
75eff68e CM |
2458 | * bit, which is only set or cleared with irqs on |
2459 | */ | |
6d92b304 | 2460 | if ((state->state & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) { |
6fc0ef68 | 2461 | struct btrfs_root *root = inode->root; |
83eea1f1 | 2462 | bool do_list = !btrfs_is_free_space_inode(inode); |
bcbfce8a | 2463 | |
8b62f87b JB |
2464 | spin_lock(&inode->lock); |
2465 | btrfs_mod_outstanding_extents(inode, -num_extents); | |
2466 | spin_unlock(&inode->lock); | |
0ca1f7ce | 2467 | |
b6d08f06 JB |
2468 | /* |
2469 | * We don't reserve metadata space for space cache inodes so we | |
52042d8e | 2470 | * don't need to call delalloc_release_metadata if there is an |
b6d08f06 JB |
2471 | * error. |
2472 | */ | |
6d92b304 | 2473 | if (bits & EXTENT_CLEAR_META_RESV && |
0b246afa | 2474 | root != fs_info->tree_root) |
43b18595 | 2475 | btrfs_delalloc_release_metadata(inode, len, false); |
0ca1f7ce | 2476 | |
6a3891c5 | 2477 | /* For sanity tests. */ |
0b246afa | 2478 | if (btrfs_is_testing(fs_info)) |
6a3891c5 JB |
2479 | return; |
2480 | ||
37f00a6d | 2481 | if (!btrfs_is_data_reloc_root(root) && |
a315e68f | 2482 | do_list && !(state->state & EXTENT_NORESERVE) && |
6d92b304 | 2483 | (bits & EXTENT_CLEAR_DATA_RESV)) |
9db5d510 | 2484 | btrfs_free_reserved_data_space_noquota(fs_info, len); |
9ed74f2d | 2485 | |
104b4e51 NB |
2486 | percpu_counter_add_batch(&fs_info->delalloc_bytes, -len, |
2487 | fs_info->delalloc_batch); | |
6fc0ef68 NB |
2488 | spin_lock(&inode->lock); |
2489 | inode->delalloc_bytes -= len; | |
2490 | if (do_list && inode->delalloc_bytes == 0 && | |
df0af1a5 | 2491 | test_bit(BTRFS_INODE_IN_DELALLOC_LIST, |
9e3e97f4 | 2492 | &inode->runtime_flags)) |
eb73c1b7 | 2493 | btrfs_del_delalloc_inode(root, inode); |
6fc0ef68 | 2494 | spin_unlock(&inode->lock); |
291d673e | 2495 | } |
a7e3b975 FM |
2496 | |
2497 | if ((state->state & EXTENT_DELALLOC_NEW) && | |
6d92b304 | 2498 | (bits & EXTENT_DELALLOC_NEW)) { |
a7e3b975 FM |
2499 | spin_lock(&inode->lock); |
2500 | ASSERT(inode->new_delalloc_bytes >= len); | |
2501 | inode->new_delalloc_bytes -= len; | |
6d92b304 | 2502 | if (bits & EXTENT_ADD_INODE_BYTES) |
2766ff61 | 2503 | inode_add_bytes(&inode->vfs_inode, len); |
a7e3b975 FM |
2504 | spin_unlock(&inode->lock); |
2505 | } | |
291d673e CM |
2506 | } |
2507 | ||
d352ac68 CM |
2508 | /* |
2509 | * in order to insert checksums into the metadata in large chunks, | |
2510 | * we wait until bio submission time. All the pages in the bio are | |
2511 | * checksummed and sums are attached onto the ordered extent record. | |
2512 | * | |
2513 | * At IO completion time the cums attached on the ordered extent record | |
2514 | * are inserted into the btree | |
2515 | */ | |
8896a08d | 2516 | static blk_status_t btrfs_submit_bio_start(struct inode *inode, struct bio *bio, |
1941b64b | 2517 | u64 dio_file_offset) |
065631f6 | 2518 | { |
e331f6b1 | 2519 | return btrfs_csum_one_bio(BTRFS_I(inode), bio, (u64)-1, false); |
4a69a410 | 2520 | } |
e015640f | 2521 | |
abb99cfd NA |
2522 | /* |
2523 | * Split an extent_map at [start, start + len] | |
2524 | * | |
2525 | * This function is intended to be used only for extract_ordered_extent(). | |
2526 | */ | |
2527 | static int split_zoned_em(struct btrfs_inode *inode, u64 start, u64 len, | |
2528 | u64 pre, u64 post) | |
2529 | { | |
2530 | struct extent_map_tree *em_tree = &inode->extent_tree; | |
2531 | struct extent_map *em; | |
2532 | struct extent_map *split_pre = NULL; | |
2533 | struct extent_map *split_mid = NULL; | |
2534 | struct extent_map *split_post = NULL; | |
2535 | int ret = 0; | |
abb99cfd NA |
2536 | unsigned long flags; |
2537 | ||
2538 | /* Sanity check */ | |
2539 | if (pre == 0 && post == 0) | |
2540 | return 0; | |
2541 | ||
2542 | split_pre = alloc_extent_map(); | |
2543 | if (pre) | |
2544 | split_mid = alloc_extent_map(); | |
2545 | if (post) | |
2546 | split_post = alloc_extent_map(); | |
2547 | if (!split_pre || (pre && !split_mid) || (post && !split_post)) { | |
2548 | ret = -ENOMEM; | |
2549 | goto out; | |
2550 | } | |
2551 | ||
2552 | ASSERT(pre + post < len); | |
2553 | ||
570eb97b | 2554 | lock_extent(&inode->io_tree, start, start + len - 1, NULL); |
abb99cfd NA |
2555 | write_lock(&em_tree->lock); |
2556 | em = lookup_extent_mapping(em_tree, start, len); | |
2557 | if (!em) { | |
2558 | ret = -EIO; | |
2559 | goto out_unlock; | |
2560 | } | |
2561 | ||
2562 | ASSERT(em->len == len); | |
2563 | ASSERT(!test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)); | |
2564 | ASSERT(em->block_start < EXTENT_MAP_LAST_BYTE); | |
63fb5879 NA |
2565 | ASSERT(test_bit(EXTENT_FLAG_PINNED, &em->flags)); |
2566 | ASSERT(!test_bit(EXTENT_FLAG_LOGGING, &em->flags)); | |
2567 | ASSERT(!list_empty(&em->list)); | |
abb99cfd NA |
2568 | |
2569 | flags = em->flags; | |
2570 | clear_bit(EXTENT_FLAG_PINNED, &em->flags); | |
abb99cfd NA |
2571 | |
2572 | /* First, replace the em with a new extent_map starting from * em->start */ | |
2573 | split_pre->start = em->start; | |
2574 | split_pre->len = (pre ? pre : em->len - post); | |
2575 | split_pre->orig_start = split_pre->start; | |
2576 | split_pre->block_start = em->block_start; | |
2577 | split_pre->block_len = split_pre->len; | |
2578 | split_pre->orig_block_len = split_pre->block_len; | |
2579 | split_pre->ram_bytes = split_pre->len; | |
2580 | split_pre->flags = flags; | |
2581 | split_pre->compress_type = em->compress_type; | |
2582 | split_pre->generation = em->generation; | |
2583 | ||
63fb5879 | 2584 | replace_extent_mapping(em_tree, em, split_pre, 1); |
abb99cfd NA |
2585 | |
2586 | /* | |
2587 | * Now we only have an extent_map at: | |
2588 | * [em->start, em->start + pre] if pre != 0 | |
2589 | * [em->start, em->start + em->len - post] if pre == 0 | |
2590 | */ | |
2591 | ||
2592 | if (pre) { | |
2593 | /* Insert the middle extent_map */ | |
2594 | split_mid->start = em->start + pre; | |
2595 | split_mid->len = em->len - pre - post; | |
2596 | split_mid->orig_start = split_mid->start; | |
2597 | split_mid->block_start = em->block_start + pre; | |
2598 | split_mid->block_len = split_mid->len; | |
2599 | split_mid->orig_block_len = split_mid->block_len; | |
2600 | split_mid->ram_bytes = split_mid->len; | |
2601 | split_mid->flags = flags; | |
2602 | split_mid->compress_type = em->compress_type; | |
2603 | split_mid->generation = em->generation; | |
63fb5879 | 2604 | add_extent_mapping(em_tree, split_mid, 1); |
abb99cfd NA |
2605 | } |
2606 | ||
2607 | if (post) { | |
2608 | split_post->start = em->start + em->len - post; | |
2609 | split_post->len = post; | |
2610 | split_post->orig_start = split_post->start; | |
2611 | split_post->block_start = em->block_start + em->len - post; | |
2612 | split_post->block_len = split_post->len; | |
2613 | split_post->orig_block_len = split_post->block_len; | |
2614 | split_post->ram_bytes = split_post->len; | |
2615 | split_post->flags = flags; | |
2616 | split_post->compress_type = em->compress_type; | |
2617 | split_post->generation = em->generation; | |
63fb5879 | 2618 | add_extent_mapping(em_tree, split_post, 1); |
abb99cfd NA |
2619 | } |
2620 | ||
2621 | /* Once for us */ | |
2622 | free_extent_map(em); | |
2623 | /* Once for the tree */ | |
2624 | free_extent_map(em); | |
2625 | ||
2626 | out_unlock: | |
2627 | write_unlock(&em_tree->lock); | |
570eb97b | 2628 | unlock_extent(&inode->io_tree, start, start + len - 1, NULL); |
abb99cfd NA |
2629 | out: |
2630 | free_extent_map(split_pre); | |
2631 | free_extent_map(split_mid); | |
2632 | free_extent_map(split_post); | |
2633 | ||
2634 | return ret; | |
2635 | } | |
2636 | ||
d22002fd NA |
2637 | static blk_status_t extract_ordered_extent(struct btrfs_inode *inode, |
2638 | struct bio *bio, loff_t file_offset) | |
2639 | { | |
2640 | struct btrfs_ordered_extent *ordered; | |
d22002fd | 2641 | u64 start = (u64)bio->bi_iter.bi_sector << SECTOR_SHIFT; |
abb99cfd | 2642 | u64 file_len; |
d22002fd NA |
2643 | u64 len = bio->bi_iter.bi_size; |
2644 | u64 end = start + len; | |
2645 | u64 ordered_end; | |
2646 | u64 pre, post; | |
2647 | int ret = 0; | |
2648 | ||
2649 | ordered = btrfs_lookup_ordered_extent(inode, file_offset); | |
2650 | if (WARN_ON_ONCE(!ordered)) | |
2651 | return BLK_STS_IOERR; | |
2652 | ||
2653 | /* No need to split */ | |
2654 | if (ordered->disk_num_bytes == len) | |
2655 | goto out; | |
2656 | ||
2657 | /* We cannot split once end_bio'd ordered extent */ | |
2658 | if (WARN_ON_ONCE(ordered->bytes_left != ordered->disk_num_bytes)) { | |
2659 | ret = -EINVAL; | |
2660 | goto out; | |
2661 | } | |
2662 | ||
2663 | /* We cannot split a compressed ordered extent */ | |
2664 | if (WARN_ON_ONCE(ordered->disk_num_bytes != ordered->num_bytes)) { | |
2665 | ret = -EINVAL; | |
2666 | goto out; | |
2667 | } | |
2668 | ||
2669 | ordered_end = ordered->disk_bytenr + ordered->disk_num_bytes; | |
2670 | /* bio must be in one ordered extent */ | |
2671 | if (WARN_ON_ONCE(start < ordered->disk_bytenr || end > ordered_end)) { | |
2672 | ret = -EINVAL; | |
2673 | goto out; | |
2674 | } | |
2675 | ||
2676 | /* Checksum list should be empty */ | |
2677 | if (WARN_ON_ONCE(!list_empty(&ordered->list))) { | |
2678 | ret = -EINVAL; | |
2679 | goto out; | |
2680 | } | |
2681 | ||
abb99cfd | 2682 | file_len = ordered->num_bytes; |
d22002fd NA |
2683 | pre = start - ordered->disk_bytenr; |
2684 | post = ordered_end - end; | |
2685 | ||
2686 | ret = btrfs_split_ordered_extent(ordered, pre, post); | |
2687 | if (ret) | |
2688 | goto out; | |
abb99cfd | 2689 | ret = split_zoned_em(inode, file_offset, file_len, pre, post); |
d22002fd NA |
2690 | |
2691 | out: | |
d22002fd NA |
2692 | btrfs_put_ordered_extent(ordered); |
2693 | ||
2694 | return errno_to_blk_status(ret); | |
2695 | } | |
2696 | ||
c93104e7 | 2697 | void btrfs_submit_data_write_bio(struct inode *inode, struct bio *bio, int mirror_num) |
44b8bd7e | 2698 | { |
0b246afa | 2699 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
c93104e7 CH |
2700 | struct btrfs_inode *bi = BTRFS_I(inode); |
2701 | blk_status_t ret; | |
0417341e | 2702 | |
d22002fd | 2703 | if (bio_op(bio) == REQ_OP_ZONE_APPEND) { |
c93104e7 CH |
2704 | ret = extract_ordered_extent(bi, bio, |
2705 | page_offset(bio_first_bvec_all(bio)->bv_page)); | |
917f32a2 CH |
2706 | if (ret) { |
2707 | btrfs_bio_end_io(btrfs_bio(bio), ret); | |
2708 | return; | |
2709 | } | |
d22002fd NA |
2710 | } |
2711 | ||
c93104e7 | 2712 | /* |
82443fd5 CH |
2713 | * If we need to checksum, and the I/O is not issued by fsync and |
2714 | * friends, that is ->sync_writers != 0, defer the submission to a | |
2715 | * workqueue to parallelize it. | |
2716 | * | |
2717 | * Csum items for reloc roots have already been cloned at this point, | |
2718 | * so they are handled as part of the no-checksum case. | |
c93104e7 CH |
2719 | */ |
2720 | if (!(bi->flags & BTRFS_INODE_NODATASUM) && | |
82443fd5 CH |
2721 | !test_bit(BTRFS_FS_STATE_NO_CSUMS, &fs_info->fs_state) && |
2722 | !btrfs_is_data_reloc_root(bi->root)) { | |
ea1f0ced CH |
2723 | if (!atomic_read(&bi->sync_writers) && |
2724 | btrfs_wq_submit_bio(inode, bio, mirror_num, 0, | |
2725 | btrfs_submit_bio_start)) | |
ad357938 | 2726 | return; |
82443fd5 CH |
2727 | |
2728 | ret = btrfs_csum_one_bio(bi, bio, (u64)-1, false); | |
917f32a2 CH |
2729 | if (ret) { |
2730 | btrfs_bio_end_io(btrfs_bio(bio), ret); | |
2731 | return; | |
2732 | } | |
19b9bdb0 | 2733 | } |
1a722d8f | 2734 | btrfs_submit_bio(fs_info, bio, mirror_num); |
c93104e7 | 2735 | } |
19b9bdb0 | 2736 | |
c93104e7 CH |
2737 | void btrfs_submit_data_read_bio(struct inode *inode, struct bio *bio, |
2738 | int mirror_num, enum btrfs_compression_type compress_type) | |
2739 | { | |
2740 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); | |
2741 | blk_status_t ret; | |
61891923 | 2742 | |
c93104e7 CH |
2743 | if (compress_type != BTRFS_COMPRESS_NONE) { |
2744 | /* | |
2745 | * btrfs_submit_compressed_read will handle completing the bio | |
2746 | * if there were any errors, so just return here. | |
2747 | */ | |
2748 | btrfs_submit_compressed_read(inode, bio, mirror_num); | |
2749 | return; | |
19b9bdb0 CM |
2750 | } |
2751 | ||
81bd9328 CH |
2752 | /* Save the original iter for read repair */ |
2753 | btrfs_bio(bio)->iter = bio->bi_iter; | |
2754 | ||
c93104e7 CH |
2755 | /* |
2756 | * Lookup bio sums does extra checks around whether we need to csum or | |
2757 | * not, which is why we ignore skip_sum here. | |
2758 | */ | |
2759 | ret = btrfs_lookup_bio_sums(inode, bio, NULL); | |
4e4cbee9 | 2760 | if (ret) { |
917f32a2 | 2761 | btrfs_bio_end_io(btrfs_bio(bio), ret); |
1a722d8f | 2762 | return; |
4246a0b6 | 2763 | } |
1a722d8f CH |
2764 | |
2765 | btrfs_submit_bio(fs_info, bio, mirror_num); | |
065631f6 | 2766 | } |
6885f308 | 2767 | |
d352ac68 CM |
2768 | /* |
2769 | * given a list of ordered sums record them in the inode. This happens | |
2770 | * at IO completion time based on sums calculated at bio submission time. | |
2771 | */ | |
510f85ed NB |
2772 | static int add_pending_csums(struct btrfs_trans_handle *trans, |
2773 | struct list_head *list) | |
e6dcd2dc | 2774 | { |
e6dcd2dc | 2775 | struct btrfs_ordered_sum *sum; |
fc28b25e | 2776 | struct btrfs_root *csum_root = NULL; |
ac01f26a | 2777 | int ret; |
e6dcd2dc | 2778 | |
c6e30871 | 2779 | list_for_each_entry(sum, list, list) { |
7c2871a2 | 2780 | trans->adding_csums = true; |
fc28b25e JB |
2781 | if (!csum_root) |
2782 | csum_root = btrfs_csum_root(trans->fs_info, | |
2783 | sum->bytenr); | |
2784 | ret = btrfs_csum_file_blocks(trans, csum_root, sum); | |
7c2871a2 | 2785 | trans->adding_csums = false; |
ac01f26a NB |
2786 | if (ret) |
2787 | return ret; | |
e6dcd2dc CM |
2788 | } |
2789 | return 0; | |
2790 | } | |
2791 | ||
c3347309 FM |
2792 | static int btrfs_find_new_delalloc_bytes(struct btrfs_inode *inode, |
2793 | const u64 start, | |
2794 | const u64 len, | |
2795 | struct extent_state **cached_state) | |
2796 | { | |
2797 | u64 search_start = start; | |
2798 | const u64 end = start + len - 1; | |
2799 | ||
2800 | while (search_start < end) { | |
2801 | const u64 search_len = end - search_start + 1; | |
2802 | struct extent_map *em; | |
2803 | u64 em_len; | |
2804 | int ret = 0; | |
2805 | ||
2806 | em = btrfs_get_extent(inode, NULL, 0, search_start, search_len); | |
2807 | if (IS_ERR(em)) | |
2808 | return PTR_ERR(em); | |
2809 | ||
2810 | if (em->block_start != EXTENT_MAP_HOLE) | |
2811 | goto next; | |
2812 | ||
2813 | em_len = em->len; | |
2814 | if (em->start < search_start) | |
2815 | em_len -= search_start - em->start; | |
2816 | if (em_len > search_len) | |
2817 | em_len = search_len; | |
2818 | ||
2819 | ret = set_extent_bit(&inode->io_tree, search_start, | |
2820 | search_start + em_len - 1, | |
994bcd1e | 2821 | EXTENT_DELALLOC_NEW, cached_state, |
291bbb1e | 2822 | GFP_NOFS); |
c3347309 FM |
2823 | next: |
2824 | search_start = extent_map_end(em); | |
2825 | free_extent_map(em); | |
2826 | if (ret) | |
2827 | return ret; | |
2828 | } | |
2829 | return 0; | |
2830 | } | |
2831 | ||
c2566f22 | 2832 | int btrfs_set_extent_delalloc(struct btrfs_inode *inode, u64 start, u64 end, |
e3b8a485 | 2833 | unsigned int extra_bits, |
330a5827 | 2834 | struct extent_state **cached_state) |
ea8c2819 | 2835 | { |
fdb1e121 | 2836 | WARN_ON(PAGE_ALIGNED(end)); |
c3347309 FM |
2837 | |
2838 | if (start >= i_size_read(&inode->vfs_inode) && | |
2839 | !(inode->flags & BTRFS_INODE_PREALLOC)) { | |
2840 | /* | |
2841 | * There can't be any extents following eof in this case so just | |
2842 | * set the delalloc new bit for the range directly. | |
2843 | */ | |
2844 | extra_bits |= EXTENT_DELALLOC_NEW; | |
2845 | } else { | |
2846 | int ret; | |
2847 | ||
2848 | ret = btrfs_find_new_delalloc_bytes(inode, start, | |
2849 | end + 1 - start, | |
2850 | cached_state); | |
2851 | if (ret) | |
2852 | return ret; | |
2853 | } | |
2854 | ||
c2566f22 NB |
2855 | return set_extent_delalloc(&inode->io_tree, start, end, extra_bits, |
2856 | cached_state); | |
ea8c2819 CM |
2857 | } |
2858 | ||
d352ac68 | 2859 | /* see btrfs_writepage_start_hook for details on why this is required */ |
247e743c CM |
2860 | struct btrfs_writepage_fixup { |
2861 | struct page *page; | |
f4b1363c | 2862 | struct inode *inode; |
247e743c CM |
2863 | struct btrfs_work work; |
2864 | }; | |
2865 | ||
b2950863 | 2866 | static void btrfs_writepage_fixup_worker(struct btrfs_work *work) |
247e743c CM |
2867 | { |
2868 | struct btrfs_writepage_fixup *fixup; | |
2869 | struct btrfs_ordered_extent *ordered; | |
2ac55d41 | 2870 | struct extent_state *cached_state = NULL; |
364ecf36 | 2871 | struct extent_changeset *data_reserved = NULL; |
247e743c | 2872 | struct page *page; |
65d87f79 | 2873 | struct btrfs_inode *inode; |
247e743c CM |
2874 | u64 page_start; |
2875 | u64 page_end; | |
25f3c502 | 2876 | int ret = 0; |
f4b1363c | 2877 | bool free_delalloc_space = true; |
247e743c CM |
2878 | |
2879 | fixup = container_of(work, struct btrfs_writepage_fixup, work); | |
2880 | page = fixup->page; | |
65d87f79 | 2881 | inode = BTRFS_I(fixup->inode); |
f4b1363c JB |
2882 | page_start = page_offset(page); |
2883 | page_end = page_offset(page) + PAGE_SIZE - 1; | |
2884 | ||
2885 | /* | |
2886 | * This is similar to page_mkwrite, we need to reserve the space before | |
2887 | * we take the page lock. | |
2888 | */ | |
65d87f79 NB |
2889 | ret = btrfs_delalloc_reserve_space(inode, &data_reserved, page_start, |
2890 | PAGE_SIZE); | |
4a096752 | 2891 | again: |
247e743c | 2892 | lock_page(page); |
25f3c502 CM |
2893 | |
2894 | /* | |
2895 | * Before we queued this fixup, we took a reference on the page. | |
2896 | * page->mapping may go NULL, but it shouldn't be moved to a different | |
2897 | * address space. | |
2898 | */ | |
f4b1363c JB |
2899 | if (!page->mapping || !PageDirty(page) || !PageChecked(page)) { |
2900 | /* | |
2901 | * Unfortunately this is a little tricky, either | |
2902 | * | |
2903 | * 1) We got here and our page had already been dealt with and | |
2904 | * we reserved our space, thus ret == 0, so we need to just | |
2905 | * drop our space reservation and bail. This can happen the | |
2906 | * first time we come into the fixup worker, or could happen | |
2907 | * while waiting for the ordered extent. | |
2908 | * 2) Our page was already dealt with, but we happened to get an | |
2909 | * ENOSPC above from the btrfs_delalloc_reserve_space. In | |
2910 | * this case we obviously don't have anything to release, but | |
2911 | * because the page was already dealt with we don't want to | |
2912 | * mark the page with an error, so make sure we're resetting | |
2913 | * ret to 0. This is why we have this check _before_ the ret | |
2914 | * check, because we do not want to have a surprise ENOSPC | |
2915 | * when the page was already properly dealt with. | |
2916 | */ | |
2917 | if (!ret) { | |
65d87f79 NB |
2918 | btrfs_delalloc_release_extents(inode, PAGE_SIZE); |
2919 | btrfs_delalloc_release_space(inode, data_reserved, | |
f4b1363c JB |
2920 | page_start, PAGE_SIZE, |
2921 | true); | |
2922 | } | |
2923 | ret = 0; | |
247e743c | 2924 | goto out_page; |
f4b1363c | 2925 | } |
247e743c | 2926 | |
25f3c502 | 2927 | /* |
f4b1363c JB |
2928 | * We can't mess with the page state unless it is locked, so now that |
2929 | * it is locked bail if we failed to make our space reservation. | |
25f3c502 | 2930 | */ |
f4b1363c JB |
2931 | if (ret) |
2932 | goto out_page; | |
247e743c | 2933 | |
570eb97b | 2934 | lock_extent(&inode->io_tree, page_start, page_end, &cached_state); |
4a096752 CM |
2935 | |
2936 | /* already ordered? We're done */ | |
f57ad937 | 2937 | if (PageOrdered(page)) |
f4b1363c | 2938 | goto out_reserved; |
4a096752 | 2939 | |
65d87f79 | 2940 | ordered = btrfs_lookup_ordered_range(inode, page_start, PAGE_SIZE); |
4a096752 | 2941 | if (ordered) { |
570eb97b JB |
2942 | unlock_extent(&inode->io_tree, page_start, page_end, |
2943 | &cached_state); | |
4a096752 | 2944 | unlock_page(page); |
c0a43603 | 2945 | btrfs_start_ordered_extent(ordered, 1); |
87826df0 | 2946 | btrfs_put_ordered_extent(ordered); |
4a096752 CM |
2947 | goto again; |
2948 | } | |
247e743c | 2949 | |
65d87f79 | 2950 | ret = btrfs_set_extent_delalloc(inode, page_start, page_end, 0, |
330a5827 | 2951 | &cached_state); |
25f3c502 | 2952 | if (ret) |
53687007 | 2953 | goto out_reserved; |
f3038ee3 | 2954 | |
25f3c502 CM |
2955 | /* |
2956 | * Everything went as planned, we're now the owner of a dirty page with | |
2957 | * delayed allocation bits set and space reserved for our COW | |
2958 | * destination. | |
2959 | * | |
2960 | * The page was dirty when we started, nothing should have cleaned it. | |
2961 | */ | |
2962 | BUG_ON(!PageDirty(page)); | |
f4b1363c | 2963 | free_delalloc_space = false; |
53687007 | 2964 | out_reserved: |
65d87f79 | 2965 | btrfs_delalloc_release_extents(inode, PAGE_SIZE); |
f4b1363c | 2966 | if (free_delalloc_space) |
65d87f79 NB |
2967 | btrfs_delalloc_release_space(inode, data_reserved, page_start, |
2968 | PAGE_SIZE, true); | |
570eb97b | 2969 | unlock_extent(&inode->io_tree, page_start, page_end, &cached_state); |
247e743c | 2970 | out_page: |
25f3c502 CM |
2971 | if (ret) { |
2972 | /* | |
2973 | * We hit ENOSPC or other errors. Update the mapping and page | |
2974 | * to reflect the errors and clean the page. | |
2975 | */ | |
2976 | mapping_set_error(page->mapping, ret); | |
2977 | end_extent_writepage(page, ret, page_start, page_end); | |
2978 | clear_page_dirty_for_io(page); | |
2979 | SetPageError(page); | |
2980 | } | |
e4f94347 | 2981 | btrfs_page_clear_checked(inode->root->fs_info, page, page_start, PAGE_SIZE); |
247e743c | 2982 | unlock_page(page); |
09cbfeaf | 2983 | put_page(page); |
b897abec | 2984 | kfree(fixup); |
364ecf36 | 2985 | extent_changeset_free(data_reserved); |
f4b1363c JB |
2986 | /* |
2987 | * As a precaution, do a delayed iput in case it would be the last iput | |
2988 | * that could need flushing space. Recursing back to fixup worker would | |
2989 | * deadlock. | |
2990 | */ | |
65d87f79 | 2991 | btrfs_add_delayed_iput(&inode->vfs_inode); |
247e743c CM |
2992 | } |
2993 | ||
2994 | /* | |
2995 | * There are a few paths in the higher layers of the kernel that directly | |
2996 | * set the page dirty bit without asking the filesystem if it is a | |
2997 | * good idea. This causes problems because we want to make sure COW | |
2998 | * properly happens and the data=ordered rules are followed. | |
2999 | * | |
c8b97818 | 3000 | * In our case any range that doesn't have the ORDERED bit set |
247e743c CM |
3001 | * hasn't been properly setup for IO. We kick off an async process |
3002 | * to fix it up. The async helper will wait for ordered extents, set | |
3003 | * the delalloc bit and make it safe to write the page. | |
3004 | */ | |
a129ffb8 | 3005 | int btrfs_writepage_cow_fixup(struct page *page) |
247e743c CM |
3006 | { |
3007 | struct inode *inode = page->mapping->host; | |
0b246afa | 3008 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
247e743c | 3009 | struct btrfs_writepage_fixup *fixup; |
247e743c | 3010 | |
f57ad937 QW |
3011 | /* This page has ordered extent covering it already */ |
3012 | if (PageOrdered(page)) | |
247e743c CM |
3013 | return 0; |
3014 | ||
25f3c502 CM |
3015 | /* |
3016 | * PageChecked is set below when we create a fixup worker for this page, | |
3017 | * don't try to create another one if we're already PageChecked() | |
3018 | * | |
3019 | * The extent_io writepage code will redirty the page if we send back | |
3020 | * EAGAIN. | |
3021 | */ | |
247e743c CM |
3022 | if (PageChecked(page)) |
3023 | return -EAGAIN; | |
3024 | ||
3025 | fixup = kzalloc(sizeof(*fixup), GFP_NOFS); | |
3026 | if (!fixup) | |
3027 | return -EAGAIN; | |
f421950f | 3028 | |
f4b1363c JB |
3029 | /* |
3030 | * We are already holding a reference to this inode from | |
3031 | * write_cache_pages. We need to hold it because the space reservation | |
3032 | * takes place outside of the page lock, and we can't trust | |
3033 | * page->mapping outside of the page lock. | |
3034 | */ | |
3035 | ihold(inode); | |
e4f94347 | 3036 | btrfs_page_set_checked(fs_info, page, page_offset(page), PAGE_SIZE); |
09cbfeaf | 3037 | get_page(page); |
a0cac0ec | 3038 | btrfs_init_work(&fixup->work, btrfs_writepage_fixup_worker, NULL, NULL); |
247e743c | 3039 | fixup->page = page; |
f4b1363c | 3040 | fixup->inode = inode; |
0b246afa | 3041 | btrfs_queue_work(fs_info->fixup_workers, &fixup->work); |
25f3c502 CM |
3042 | |
3043 | return -EAGAIN; | |
247e743c CM |
3044 | } |
3045 | ||
d899e052 | 3046 | static int insert_reserved_file_extent(struct btrfs_trans_handle *trans, |
c553f94d | 3047 | struct btrfs_inode *inode, u64 file_pos, |
9729f10a | 3048 | struct btrfs_file_extent_item *stack_fi, |
2766ff61 | 3049 | const bool update_inode_bytes, |
9729f10a | 3050 | u64 qgroup_reserved) |
d899e052 | 3051 | { |
c553f94d | 3052 | struct btrfs_root *root = inode->root; |
2766ff61 | 3053 | const u64 sectorsize = root->fs_info->sectorsize; |
d899e052 YZ |
3054 | struct btrfs_path *path; |
3055 | struct extent_buffer *leaf; | |
3056 | struct btrfs_key ins; | |
203f44c5 QW |
3057 | u64 disk_num_bytes = btrfs_stack_file_extent_disk_num_bytes(stack_fi); |
3058 | u64 disk_bytenr = btrfs_stack_file_extent_disk_bytenr(stack_fi); | |
cb36a9bb | 3059 | u64 offset = btrfs_stack_file_extent_offset(stack_fi); |
203f44c5 QW |
3060 | u64 num_bytes = btrfs_stack_file_extent_num_bytes(stack_fi); |
3061 | u64 ram_bytes = btrfs_stack_file_extent_ram_bytes(stack_fi); | |
5893dfb9 | 3062 | struct btrfs_drop_extents_args drop_args = { 0 }; |
d899e052 YZ |
3063 | int ret; |
3064 | ||
3065 | path = btrfs_alloc_path(); | |
d8926bb3 MF |
3066 | if (!path) |
3067 | return -ENOMEM; | |
d899e052 | 3068 | |
a1ed835e CM |
3069 | /* |
3070 | * we may be replacing one extent in the tree with another. | |
3071 | * The new extent is pinned in the extent map, and we don't want | |
3072 | * to drop it from the cache until it is completely in the btree. | |
3073 | * | |
3074 | * So, tell btrfs_drop_extents to leave this extent in the cache. | |
3075 | * the caller is expected to unpin it and allow it to be merged | |
3076 | * with the others. | |
3077 | */ | |
5893dfb9 FM |
3078 | drop_args.path = path; |
3079 | drop_args.start = file_pos; | |
3080 | drop_args.end = file_pos + num_bytes; | |
3081 | drop_args.replace_extent = true; | |
3082 | drop_args.extent_item_size = sizeof(*stack_fi); | |
3083 | ret = btrfs_drop_extents(trans, root, inode, &drop_args); | |
79787eaa JM |
3084 | if (ret) |
3085 | goto out; | |
d899e052 | 3086 | |
5893dfb9 | 3087 | if (!drop_args.extent_inserted) { |
c553f94d | 3088 | ins.objectid = btrfs_ino(inode); |
1acae57b FDBM |
3089 | ins.offset = file_pos; |
3090 | ins.type = BTRFS_EXTENT_DATA_KEY; | |
3091 | ||
1acae57b | 3092 | ret = btrfs_insert_empty_item(trans, root, path, &ins, |
203f44c5 | 3093 | sizeof(*stack_fi)); |
1acae57b FDBM |
3094 | if (ret) |
3095 | goto out; | |
3096 | } | |
d899e052 | 3097 | leaf = path->nodes[0]; |
203f44c5 QW |
3098 | btrfs_set_stack_file_extent_generation(stack_fi, trans->transid); |
3099 | write_extent_buffer(leaf, stack_fi, | |
3100 | btrfs_item_ptr_offset(leaf, path->slots[0]), | |
3101 | sizeof(struct btrfs_file_extent_item)); | |
b9473439 | 3102 | |
d899e052 | 3103 | btrfs_mark_buffer_dirty(leaf); |
ce195332 | 3104 | btrfs_release_path(path); |
d899e052 | 3105 | |
2766ff61 FM |
3106 | /* |
3107 | * If we dropped an inline extent here, we know the range where it is | |
3108 | * was not marked with the EXTENT_DELALLOC_NEW bit, so we update the | |
1a9fd417 | 3109 | * number of bytes only for that range containing the inline extent. |
2766ff61 FM |
3110 | * The remaining of the range will be processed when clearning the |
3111 | * EXTENT_DELALLOC_BIT bit through the ordered extent completion. | |
3112 | */ | |
3113 | if (file_pos == 0 && !IS_ALIGNED(drop_args.bytes_found, sectorsize)) { | |
3114 | u64 inline_size = round_down(drop_args.bytes_found, sectorsize); | |
3115 | ||
3116 | inline_size = drop_args.bytes_found - inline_size; | |
3117 | btrfs_update_inode_bytes(inode, sectorsize, inline_size); | |
3118 | drop_args.bytes_found -= inline_size; | |
3119 | num_bytes -= sectorsize; | |
3120 | } | |
3121 | ||
3122 | if (update_inode_bytes) | |
3123 | btrfs_update_inode_bytes(inode, num_bytes, drop_args.bytes_found); | |
d899e052 YZ |
3124 | |
3125 | ins.objectid = disk_bytenr; | |
3126 | ins.offset = disk_num_bytes; | |
3127 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
a12b877b | 3128 | |
c553f94d | 3129 | ret = btrfs_inode_set_file_extent_range(inode, file_pos, ram_bytes); |
9ddc959e JB |
3130 | if (ret) |
3131 | goto out; | |
3132 | ||
c553f94d | 3133 | ret = btrfs_alloc_reserved_file_extent(trans, root, btrfs_ino(inode), |
cb36a9bb OS |
3134 | file_pos - offset, |
3135 | qgroup_reserved, &ins); | |
79787eaa | 3136 | out: |
d899e052 | 3137 | btrfs_free_path(path); |
b9473439 | 3138 | |
79787eaa | 3139 | return ret; |
d899e052 YZ |
3140 | } |
3141 | ||
2ff7e61e | 3142 | static void btrfs_release_delalloc_bytes(struct btrfs_fs_info *fs_info, |
e570fd27 MX |
3143 | u64 start, u64 len) |
3144 | { | |
32da5386 | 3145 | struct btrfs_block_group *cache; |
e570fd27 | 3146 | |
0b246afa | 3147 | cache = btrfs_lookup_block_group(fs_info, start); |
e570fd27 MX |
3148 | ASSERT(cache); |
3149 | ||
3150 | spin_lock(&cache->lock); | |
3151 | cache->delalloc_bytes -= len; | |
3152 | spin_unlock(&cache->lock); | |
3153 | ||
3154 | btrfs_put_block_group(cache); | |
3155 | } | |
3156 | ||
203f44c5 | 3157 | static int insert_ordered_extent_file_extent(struct btrfs_trans_handle *trans, |
203f44c5 QW |
3158 | struct btrfs_ordered_extent *oe) |
3159 | { | |
3160 | struct btrfs_file_extent_item stack_fi; | |
2766ff61 | 3161 | bool update_inode_bytes; |
cb36a9bb OS |
3162 | u64 num_bytes = oe->num_bytes; |
3163 | u64 ram_bytes = oe->ram_bytes; | |
203f44c5 QW |
3164 | |
3165 | memset(&stack_fi, 0, sizeof(stack_fi)); | |
3166 | btrfs_set_stack_file_extent_type(&stack_fi, BTRFS_FILE_EXTENT_REG); | |
3167 | btrfs_set_stack_file_extent_disk_bytenr(&stack_fi, oe->disk_bytenr); | |
3168 | btrfs_set_stack_file_extent_disk_num_bytes(&stack_fi, | |
3169 | oe->disk_num_bytes); | |
cb36a9bb | 3170 | btrfs_set_stack_file_extent_offset(&stack_fi, oe->offset); |
c1867eb3 DS |
3171 | if (test_bit(BTRFS_ORDERED_TRUNCATED, &oe->flags)) { |
3172 | num_bytes = oe->truncated_len; | |
3173 | ram_bytes = num_bytes; | |
3174 | } | |
cb36a9bb OS |
3175 | btrfs_set_stack_file_extent_num_bytes(&stack_fi, num_bytes); |
3176 | btrfs_set_stack_file_extent_ram_bytes(&stack_fi, ram_bytes); | |
203f44c5 QW |
3177 | btrfs_set_stack_file_extent_compression(&stack_fi, oe->compress_type); |
3178 | /* Encryption and other encoding is reserved and all 0 */ | |
3179 | ||
2766ff61 FM |
3180 | /* |
3181 | * For delalloc, when completing an ordered extent we update the inode's | |
3182 | * bytes when clearing the range in the inode's io tree, so pass false | |
3183 | * as the argument 'update_inode_bytes' to insert_reserved_file_extent(), | |
3184 | * except if the ordered extent was truncated. | |
3185 | */ | |
3186 | update_inode_bytes = test_bit(BTRFS_ORDERED_DIRECT, &oe->flags) || | |
7c0c7269 | 3187 | test_bit(BTRFS_ORDERED_ENCODED, &oe->flags) || |
2766ff61 FM |
3188 | test_bit(BTRFS_ORDERED_TRUNCATED, &oe->flags); |
3189 | ||
3c38c877 NB |
3190 | return insert_reserved_file_extent(trans, BTRFS_I(oe->inode), |
3191 | oe->file_offset, &stack_fi, | |
2766ff61 | 3192 | update_inode_bytes, oe->qgroup_rsv); |
203f44c5 QW |
3193 | } |
3194 | ||
3195 | /* | |
3196 | * As ordered data IO finishes, this gets called so we can finish | |
d352ac68 CM |
3197 | * an ordered extent if the range of bytes in the file it covers are |
3198 | * fully written. | |
3199 | */ | |
711f447b | 3200 | int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent) |
e6dcd2dc | 3201 | { |
72e7e6ed NB |
3202 | struct btrfs_inode *inode = BTRFS_I(ordered_extent->inode); |
3203 | struct btrfs_root *root = inode->root; | |
3204 | struct btrfs_fs_info *fs_info = root->fs_info; | |
0ca1f7ce | 3205 | struct btrfs_trans_handle *trans = NULL; |
72e7e6ed | 3206 | struct extent_io_tree *io_tree = &inode->io_tree; |
2ac55d41 | 3207 | struct extent_state *cached_state = NULL; |
bffe633e | 3208 | u64 start, end; |
261507a0 | 3209 | int compress_type = 0; |
77cef2ec | 3210 | int ret = 0; |
bffe633e | 3211 | u64 logical_len = ordered_extent->num_bytes; |
8d510121 | 3212 | bool freespace_inode; |
77cef2ec | 3213 | bool truncated = false; |
49940bdd | 3214 | bool clear_reserved_extent = true; |
2766ff61 | 3215 | unsigned int clear_bits = EXTENT_DEFRAG; |
a7e3b975 | 3216 | |
bffe633e OS |
3217 | start = ordered_extent->file_offset; |
3218 | end = start + ordered_extent->num_bytes - 1; | |
3219 | ||
a7e3b975 FM |
3220 | if (!test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) && |
3221 | !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags) && | |
7c0c7269 OS |
3222 | !test_bit(BTRFS_ORDERED_DIRECT, &ordered_extent->flags) && |
3223 | !test_bit(BTRFS_ORDERED_ENCODED, &ordered_extent->flags)) | |
2766ff61 | 3224 | clear_bits |= EXTENT_DELALLOC_NEW; |
e6dcd2dc | 3225 | |
72e7e6ed | 3226 | freespace_inode = btrfs_is_free_space_inode(inode); |
5f4403e1 IA |
3227 | if (!freespace_inode) |
3228 | btrfs_lockdep_acquire(fs_info, btrfs_ordered_extent); | |
0cb59c99 | 3229 | |
5fd02043 JB |
3230 | if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) { |
3231 | ret = -EIO; | |
3232 | goto out; | |
3233 | } | |
3234 | ||
be1a1d7a NA |
3235 | /* A valid bdev implies a write on a sequential zone */ |
3236 | if (ordered_extent->bdev) { | |
d8e3fb10 | 3237 | btrfs_rewrite_logical_zoned(ordered_extent); |
be1a1d7a NA |
3238 | btrfs_zone_finish_endio(fs_info, ordered_extent->disk_bytenr, |
3239 | ordered_extent->disk_num_bytes); | |
3240 | } | |
d8e3fb10 | 3241 | |
72e7e6ed | 3242 | btrfs_free_io_failure_record(inode, start, end); |
f612496b | 3243 | |
77cef2ec JB |
3244 | if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) { |
3245 | truncated = true; | |
3246 | logical_len = ordered_extent->truncated_len; | |
3247 | /* Truncated the entire extent, don't bother adding */ | |
3248 | if (!logical_len) | |
3249 | goto out; | |
3250 | } | |
3251 | ||
c2167754 | 3252 | if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) { |
79787eaa | 3253 | BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */ |
94ed938a | 3254 | |
72e7e6ed | 3255 | btrfs_inode_safe_disk_i_size_write(inode, 0); |
8d510121 NB |
3256 | if (freespace_inode) |
3257 | trans = btrfs_join_transaction_spacecache(root); | |
6c760c07 JB |
3258 | else |
3259 | trans = btrfs_join_transaction(root); | |
3260 | if (IS_ERR(trans)) { | |
3261 | ret = PTR_ERR(trans); | |
3262 | trans = NULL; | |
3263 | goto out; | |
c2167754 | 3264 | } |
72e7e6ed | 3265 | trans->block_rsv = &inode->block_rsv; |
729f7961 | 3266 | ret = btrfs_update_inode_fallback(trans, root, inode); |
6c760c07 | 3267 | if (ret) /* -ENOMEM or corruption */ |
66642832 | 3268 | btrfs_abort_transaction(trans, ret); |
c2167754 YZ |
3269 | goto out; |
3270 | } | |
e6dcd2dc | 3271 | |
2766ff61 | 3272 | clear_bits |= EXTENT_LOCKED; |
570eb97b | 3273 | lock_extent(io_tree, start, end, &cached_state); |
e6dcd2dc | 3274 | |
8d510121 NB |
3275 | if (freespace_inode) |
3276 | trans = btrfs_join_transaction_spacecache(root); | |
0cb59c99 | 3277 | else |
7a7eaa40 | 3278 | trans = btrfs_join_transaction(root); |
79787eaa JM |
3279 | if (IS_ERR(trans)) { |
3280 | ret = PTR_ERR(trans); | |
3281 | trans = NULL; | |
a7e3b975 | 3282 | goto out; |
79787eaa | 3283 | } |
a79b7d4b | 3284 | |
72e7e6ed | 3285 | trans->block_rsv = &inode->block_rsv; |
c2167754 | 3286 | |
c8b97818 | 3287 | if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags)) |
261507a0 | 3288 | compress_type = ordered_extent->compress_type; |
d899e052 | 3289 | if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) { |
261507a0 | 3290 | BUG_ON(compress_type); |
72e7e6ed | 3291 | ret = btrfs_mark_extent_written(trans, inode, |
d899e052 YZ |
3292 | ordered_extent->file_offset, |
3293 | ordered_extent->file_offset + | |
77cef2ec | 3294 | logical_len); |
343d8a30 NA |
3295 | btrfs_zoned_release_data_reloc_bg(fs_info, ordered_extent->disk_bytenr, |
3296 | ordered_extent->disk_num_bytes); | |
d899e052 | 3297 | } else { |
0b246afa | 3298 | BUG_ON(root == fs_info->tree_root); |
3c38c877 | 3299 | ret = insert_ordered_extent_file_extent(trans, ordered_extent); |
49940bdd JB |
3300 | if (!ret) { |
3301 | clear_reserved_extent = false; | |
2ff7e61e | 3302 | btrfs_release_delalloc_bytes(fs_info, |
bffe633e OS |
3303 | ordered_extent->disk_bytenr, |
3304 | ordered_extent->disk_num_bytes); | |
49940bdd | 3305 | } |
d899e052 | 3306 | } |
72e7e6ed | 3307 | unpin_extent_cache(&inode->extent_tree, ordered_extent->file_offset, |
bffe633e | 3308 | ordered_extent->num_bytes, trans->transid); |
79787eaa | 3309 | if (ret < 0) { |
66642832 | 3310 | btrfs_abort_transaction(trans, ret); |
a7e3b975 | 3311 | goto out; |
79787eaa | 3312 | } |
2ac55d41 | 3313 | |
510f85ed | 3314 | ret = add_pending_csums(trans, &ordered_extent->list); |
ac01f26a NB |
3315 | if (ret) { |
3316 | btrfs_abort_transaction(trans, ret); | |
3317 | goto out; | |
3318 | } | |
e6dcd2dc | 3319 | |
2766ff61 FM |
3320 | /* |
3321 | * If this is a new delalloc range, clear its new delalloc flag to | |
3322 | * update the inode's number of bytes. This needs to be done first | |
3323 | * before updating the inode item. | |
3324 | */ | |
3325 | if ((clear_bits & EXTENT_DELALLOC_NEW) && | |
3326 | !test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) | |
72e7e6ed | 3327 | clear_extent_bit(&inode->io_tree, start, end, |
2766ff61 | 3328 | EXTENT_DELALLOC_NEW | EXTENT_ADD_INODE_BYTES, |
bd015294 | 3329 | &cached_state); |
2766ff61 | 3330 | |
72e7e6ed | 3331 | btrfs_inode_safe_disk_i_size_write(inode, 0); |
729f7961 | 3332 | ret = btrfs_update_inode_fallback(trans, root, inode); |
6c760c07 | 3333 | if (ret) { /* -ENOMEM or corruption */ |
66642832 | 3334 | btrfs_abort_transaction(trans, ret); |
a7e3b975 | 3335 | goto out; |
1ef30be1 JB |
3336 | } |
3337 | ret = 0; | |
c2167754 | 3338 | out: |
bd015294 | 3339 | clear_extent_bit(&inode->io_tree, start, end, clear_bits, |
313facc5 | 3340 | &cached_state); |
a7e3b975 | 3341 | |
a698d075 | 3342 | if (trans) |
3a45bb20 | 3343 | btrfs_end_transaction(trans); |
0cb59c99 | 3344 | |
77cef2ec | 3345 | if (ret || truncated) { |
bffe633e | 3346 | u64 unwritten_start = start; |
77cef2ec | 3347 | |
d61bec08 JB |
3348 | /* |
3349 | * If we failed to finish this ordered extent for any reason we | |
3350 | * need to make sure BTRFS_ORDERED_IOERR is set on the ordered | |
3351 | * extent, and mark the inode with the error if it wasn't | |
3352 | * already set. Any error during writeback would have already | |
3353 | * set the mapping error, so we need to set it if we're the ones | |
3354 | * marking this ordered extent as failed. | |
3355 | */ | |
3356 | if (ret && !test_and_set_bit(BTRFS_ORDERED_IOERR, | |
3357 | &ordered_extent->flags)) | |
3358 | mapping_set_error(ordered_extent->inode->i_mapping, -EIO); | |
3359 | ||
77cef2ec | 3360 | if (truncated) |
bffe633e OS |
3361 | unwritten_start += logical_len; |
3362 | clear_extent_uptodate(io_tree, unwritten_start, end, NULL); | |
77cef2ec JB |
3363 | |
3364 | /* Drop the cache for the part of the extent we didn't write. */ | |
72e7e6ed | 3365 | btrfs_drop_extent_cache(inode, unwritten_start, end, 0); |
5fd02043 | 3366 | |
0bec9ef5 JB |
3367 | /* |
3368 | * If the ordered extent had an IOERR or something else went | |
3369 | * wrong we need to return the space for this ordered extent | |
77cef2ec JB |
3370 | * back to the allocator. We only free the extent in the |
3371 | * truncated case if we didn't write out the extent at all. | |
49940bdd JB |
3372 | * |
3373 | * If we made it past insert_reserved_file_extent before we | |
3374 | * errored out then we don't need to do this as the accounting | |
3375 | * has already been done. | |
0bec9ef5 | 3376 | */ |
77cef2ec | 3377 | if ((ret || !logical_len) && |
49940bdd | 3378 | clear_reserved_extent && |
77cef2ec | 3379 | !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) && |
4eaaec24 NB |
3380 | !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) { |
3381 | /* | |
3382 | * Discard the range before returning it back to the | |
3383 | * free space pool | |
3384 | */ | |
46b27f50 | 3385 | if (ret && btrfs_test_opt(fs_info, DISCARD_SYNC)) |
4eaaec24 | 3386 | btrfs_discard_extent(fs_info, |
bffe633e OS |
3387 | ordered_extent->disk_bytenr, |
3388 | ordered_extent->disk_num_bytes, | |
3389 | NULL); | |
2ff7e61e | 3390 | btrfs_free_reserved_extent(fs_info, |
bffe633e OS |
3391 | ordered_extent->disk_bytenr, |
3392 | ordered_extent->disk_num_bytes, 1); | |
4eaaec24 | 3393 | } |
0bec9ef5 JB |
3394 | } |
3395 | ||
5fd02043 | 3396 | /* |
8bad3c02 LB |
3397 | * This needs to be done to make sure anybody waiting knows we are done |
3398 | * updating everything for this ordered extent. | |
5fd02043 | 3399 | */ |
72e7e6ed | 3400 | btrfs_remove_ordered_extent(inode, ordered_extent); |
5fd02043 | 3401 | |
e6dcd2dc CM |
3402 | /* once for us */ |
3403 | btrfs_put_ordered_extent(ordered_extent); | |
3404 | /* once for the tree */ | |
3405 | btrfs_put_ordered_extent(ordered_extent); | |
3406 | ||
5fd02043 JB |
3407 | return ret; |
3408 | } | |
3409 | ||
38a39ac7 QW |
3410 | void btrfs_writepage_endio_finish_ordered(struct btrfs_inode *inode, |
3411 | struct page *page, u64 start, | |
25c1252a | 3412 | u64 end, bool uptodate) |
211f90e6 | 3413 | { |
38a39ac7 | 3414 | trace_btrfs_writepage_end_io_hook(inode, start, end, uptodate); |
1abe9b8a | 3415 | |
711f447b | 3416 | btrfs_mark_ordered_io_finished(inode, page, start, end + 1 - start, uptodate); |
211f90e6 CM |
3417 | } |
3418 | ||
ae643a74 QW |
3419 | /* |
3420 | * Verify the checksum for a single sector without any extra action that depend | |
3421 | * on the type of I/O. | |
3422 | */ | |
3423 | int btrfs_check_sector_csum(struct btrfs_fs_info *fs_info, struct page *page, | |
3424 | u32 pgoff, u8 *csum, const u8 * const csum_expected) | |
3425 | { | |
3426 | SHASH_DESC_ON_STACK(shash, fs_info->csum_shash); | |
3427 | char *kaddr; | |
3428 | ||
3429 | ASSERT(pgoff + fs_info->sectorsize <= PAGE_SIZE); | |
3430 | ||
3431 | shash->tfm = fs_info->csum_shash; | |
3432 | ||
3433 | kaddr = kmap_local_page(page) + pgoff; | |
3434 | crypto_shash_digest(shash, kaddr, fs_info->sectorsize, csum); | |
3435 | kunmap_local(kaddr); | |
3436 | ||
3437 | if (memcmp(csum, csum_expected, fs_info->csum_size)) | |
3438 | return -EIO; | |
3439 | return 0; | |
211f90e6 CM |
3440 | } |
3441 | ||
f119553f JB |
3442 | static u8 *btrfs_csum_ptr(const struct btrfs_fs_info *fs_info, u8 *csums, u64 offset) |
3443 | { | |
3444 | u64 offset_in_sectors = offset >> fs_info->sectorsize_bits; | |
3445 | ||
3446 | return csums + offset_in_sectors * fs_info->csum_size; | |
3447 | } | |
3448 | ||
265d4ac0 QW |
3449 | /* |
3450 | * check_data_csum - verify checksum of one sector of uncompressed data | |
7ffd27e3 | 3451 | * @inode: inode |
7959bd44 | 3452 | * @bbio: btrfs_bio which contains the csum |
7ffd27e3 | 3453 | * @bio_offset: offset to the beginning of the bio (in bytes) |
265d4ac0 QW |
3454 | * @page: page where is the data to be verified |
3455 | * @pgoff: offset inside the page | |
3456 | * | |
3457 | * The length of such check is always one sector size. | |
ae643a74 QW |
3458 | * |
3459 | * When csum mismatch is detected, we will also report the error and fill the | |
3460 | * corrupted range with zero. (Thus it needs the extra parameters) | |
265d4ac0 | 3461 | */ |
7959bd44 CH |
3462 | int btrfs_check_data_csum(struct inode *inode, struct btrfs_bio *bbio, |
3463 | u32 bio_offset, struct page *page, u32 pgoff) | |
dc380aea | 3464 | { |
d5178578 | 3465 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
265d4ac0 | 3466 | u32 len = fs_info->sectorsize; |
d5178578 JT |
3467 | u8 *csum_expected; |
3468 | u8 csum[BTRFS_CSUM_SIZE]; | |
dc380aea | 3469 | |
265d4ac0 QW |
3470 | ASSERT(pgoff + len <= PAGE_SIZE); |
3471 | ||
a89ce08c | 3472 | csum_expected = btrfs_csum_ptr(fs_info, bbio->csum, bio_offset); |
d5178578 | 3473 | |
ae643a74 | 3474 | if (btrfs_check_sector_csum(fs_info, page, pgoff, csum, csum_expected)) |
dc380aea | 3475 | goto zeroit; |
dc380aea | 3476 | return 0; |
ae643a74 | 3477 | |
dc380aea | 3478 | zeroit: |
7959bd44 CH |
3479 | btrfs_print_data_csum_error(BTRFS_I(inode), |
3480 | bbio->file_offset + bio_offset, | |
3481 | csum, csum_expected, bbio->mirror_num); | |
c3a3b19b QW |
3482 | if (bbio->device) |
3483 | btrfs_dev_stat_inc_and_print(bbio->device, | |
814723e0 | 3484 | BTRFS_DEV_STAT_CORRUPTION_ERRS); |
b06660b5 | 3485 | memzero_page(page, pgoff, len); |
dc380aea MX |
3486 | return -EIO; |
3487 | } | |
3488 | ||
d352ac68 | 3489 | /* |
7ffd27e3 | 3490 | * When reads are done, we need to check csums to verify the data is correct. |
4a54c8c1 JS |
3491 | * if there's a match, we allow the bio to finish. If not, the code in |
3492 | * extent_io.c will try to find good copies for us. | |
7ffd27e3 QW |
3493 | * |
3494 | * @bio_offset: offset to the beginning of the bio (in bytes) | |
3495 | * @start: file offset of the range start | |
3496 | * @end: file offset of the range end (inclusive) | |
08508fea QW |
3497 | * |
3498 | * Return a bitmap where bit set means a csum mismatch, and bit not set means | |
3499 | * csum match. | |
d352ac68 | 3500 | */ |
c3a3b19b QW |
3501 | unsigned int btrfs_verify_data_csum(struct btrfs_bio *bbio, |
3502 | u32 bio_offset, struct page *page, | |
3503 | u64 start, u64 end) | |
07157aac | 3504 | { |
07157aac | 3505 | struct inode *inode = page->mapping->host; |
e4f94347 | 3506 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
d1310b2e | 3507 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
ff79f819 | 3508 | struct btrfs_root *root = BTRFS_I(inode)->root; |
f44cf410 QW |
3509 | const u32 sectorsize = root->fs_info->sectorsize; |
3510 | u32 pg_off; | |
08508fea | 3511 | unsigned int result = 0; |
d1310b2e | 3512 | |
3670e645 | 3513 | /* |
e4f94347 QW |
3514 | * This only happens for NODATASUM or compressed read. |
3515 | * Normally this should be covered by above check for compressed read | |
3516 | * or the next check for NODATASUM. Just do a quicker exit here. | |
3670e645 | 3517 | */ |
c3a3b19b | 3518 | if (bbio->csum == NULL) |
dc380aea | 3519 | return 0; |
17d217fe | 3520 | |
6cbff00f | 3521 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM) |
42437a63 JB |
3522 | return 0; |
3523 | ||
056c8311 | 3524 | if (unlikely(test_bit(BTRFS_FS_STATE_NO_CSUMS, &fs_info->fs_state))) |
b6cda9bc | 3525 | return 0; |
d20f7043 | 3526 | |
f44cf410 QW |
3527 | ASSERT(page_offset(page) <= start && |
3528 | end <= page_offset(page) + PAGE_SIZE - 1); | |
3529 | for (pg_off = offset_in_page(start); | |
3530 | pg_off < offset_in_page(end); | |
3531 | pg_off += sectorsize, bio_offset += sectorsize) { | |
e3c62324 | 3532 | u64 file_offset = pg_off + page_offset(page); |
f44cf410 QW |
3533 | int ret; |
3534 | ||
37f00a6d | 3535 | if (btrfs_is_data_reloc_root(root) && |
e3c62324 QW |
3536 | test_range_bit(io_tree, file_offset, |
3537 | file_offset + sectorsize - 1, | |
3538 | EXTENT_NODATASUM, 1, NULL)) { | |
3539 | /* Skip the range without csum for data reloc inode */ | |
3540 | clear_extent_bits(io_tree, file_offset, | |
3541 | file_offset + sectorsize - 1, | |
3542 | EXTENT_NODATASUM); | |
3543 | continue; | |
3544 | } | |
7959bd44 | 3545 | ret = btrfs_check_data_csum(inode, bbio, bio_offset, page, pg_off); |
08508fea QW |
3546 | if (ret < 0) { |
3547 | const int nr_bit = (pg_off - offset_in_page(start)) >> | |
3548 | root->fs_info->sectorsize_bits; | |
3549 | ||
3550 | result |= (1U << nr_bit); | |
3551 | } | |
f44cf410 | 3552 | } |
08508fea | 3553 | return result; |
07157aac | 3554 | } |
b888db2b | 3555 | |
c1c3fac2 NB |
3556 | /* |
3557 | * btrfs_add_delayed_iput - perform a delayed iput on @inode | |
3558 | * | |
3559 | * @inode: The inode we want to perform iput on | |
3560 | * | |
3561 | * This function uses the generic vfs_inode::i_count to track whether we should | |
3562 | * just decrement it (in case it's > 1) or if this is the last iput then link | |
3563 | * the inode to the delayed iput machinery. Delayed iputs are processed at | |
3564 | * transaction commit time/superblock commit/cleaner kthread. | |
3565 | */ | |
24bbcf04 YZ |
3566 | void btrfs_add_delayed_iput(struct inode *inode) |
3567 | { | |
0b246afa | 3568 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
8089fe62 | 3569 | struct btrfs_inode *binode = BTRFS_I(inode); |
24bbcf04 YZ |
3570 | |
3571 | if (atomic_add_unless(&inode->i_count, -1, 1)) | |
3572 | return; | |
3573 | ||
034f784d | 3574 | atomic_inc(&fs_info->nr_delayed_iputs); |
24bbcf04 | 3575 | spin_lock(&fs_info->delayed_iput_lock); |
c1c3fac2 NB |
3576 | ASSERT(list_empty(&binode->delayed_iput)); |
3577 | list_add_tail(&binode->delayed_iput, &fs_info->delayed_iputs); | |
24bbcf04 | 3578 | spin_unlock(&fs_info->delayed_iput_lock); |
fd340d0f JB |
3579 | if (!test_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags)) |
3580 | wake_up_process(fs_info->cleaner_kthread); | |
24bbcf04 YZ |
3581 | } |
3582 | ||
63611e73 JB |
3583 | static void run_delayed_iput_locked(struct btrfs_fs_info *fs_info, |
3584 | struct btrfs_inode *inode) | |
3585 | { | |
3586 | list_del_init(&inode->delayed_iput); | |
3587 | spin_unlock(&fs_info->delayed_iput_lock); | |
3588 | iput(&inode->vfs_inode); | |
3589 | if (atomic_dec_and_test(&fs_info->nr_delayed_iputs)) | |
3590 | wake_up(&fs_info->delayed_iputs_wait); | |
3591 | spin_lock(&fs_info->delayed_iput_lock); | |
3592 | } | |
3593 | ||
3594 | static void btrfs_run_delayed_iput(struct btrfs_fs_info *fs_info, | |
3595 | struct btrfs_inode *inode) | |
3596 | { | |
3597 | if (!list_empty(&inode->delayed_iput)) { | |
3598 | spin_lock(&fs_info->delayed_iput_lock); | |
3599 | if (!list_empty(&inode->delayed_iput)) | |
3600 | run_delayed_iput_locked(fs_info, inode); | |
3601 | spin_unlock(&fs_info->delayed_iput_lock); | |
3602 | } | |
3603 | } | |
3604 | ||
2ff7e61e | 3605 | void btrfs_run_delayed_iputs(struct btrfs_fs_info *fs_info) |
24bbcf04 | 3606 | { |
24bbcf04 | 3607 | |
24bbcf04 | 3608 | spin_lock(&fs_info->delayed_iput_lock); |
8089fe62 DS |
3609 | while (!list_empty(&fs_info->delayed_iputs)) { |
3610 | struct btrfs_inode *inode; | |
3611 | ||
3612 | inode = list_first_entry(&fs_info->delayed_iputs, | |
3613 | struct btrfs_inode, delayed_iput); | |
63611e73 | 3614 | run_delayed_iput_locked(fs_info, inode); |
71795ee5 | 3615 | cond_resched_lock(&fs_info->delayed_iput_lock); |
24bbcf04 | 3616 | } |
8089fe62 | 3617 | spin_unlock(&fs_info->delayed_iput_lock); |
24bbcf04 YZ |
3618 | } |
3619 | ||
034f784d | 3620 | /** |
2639631d NB |
3621 | * Wait for flushing all delayed iputs |
3622 | * | |
3623 | * @fs_info: the filesystem | |
034f784d JB |
3624 | * |
3625 | * This will wait on any delayed iputs that are currently running with KILLABLE | |
3626 | * set. Once they are all done running we will return, unless we are killed in | |
3627 | * which case we return EINTR. This helps in user operations like fallocate etc | |
3628 | * that might get blocked on the iputs. | |
2639631d NB |
3629 | * |
3630 | * Return EINTR if we were killed, 0 if nothing's pending | |
034f784d JB |
3631 | */ |
3632 | int btrfs_wait_on_delayed_iputs(struct btrfs_fs_info *fs_info) | |
3633 | { | |
3634 | int ret = wait_event_killable(fs_info->delayed_iputs_wait, | |
3635 | atomic_read(&fs_info->nr_delayed_iputs) == 0); | |
3636 | if (ret) | |
3637 | return -EINTR; | |
3638 | return 0; | |
3639 | } | |
3640 | ||
7b128766 | 3641 | /* |
f7e9e8fc OS |
3642 | * This creates an orphan entry for the given inode in case something goes wrong |
3643 | * in the middle of an unlink. | |
7b128766 | 3644 | */ |
73f2e545 | 3645 | int btrfs_orphan_add(struct btrfs_trans_handle *trans, |
27919067 | 3646 | struct btrfs_inode *inode) |
7b128766 | 3647 | { |
d68fc57b | 3648 | int ret; |
7b128766 | 3649 | |
27919067 OS |
3650 | ret = btrfs_insert_orphan_item(trans, inode->root, btrfs_ino(inode)); |
3651 | if (ret && ret != -EEXIST) { | |
3652 | btrfs_abort_transaction(trans, ret); | |
3653 | return ret; | |
d68fc57b YZ |
3654 | } |
3655 | ||
d68fc57b | 3656 | return 0; |
7b128766 JB |
3657 | } |
3658 | ||
3659 | /* | |
f7e9e8fc OS |
3660 | * We have done the delete so we can go ahead and remove the orphan item for |
3661 | * this particular inode. | |
7b128766 | 3662 | */ |
48a3b636 | 3663 | static int btrfs_orphan_del(struct btrfs_trans_handle *trans, |
3d6ae7bb | 3664 | struct btrfs_inode *inode) |
7b128766 | 3665 | { |
27919067 | 3666 | return btrfs_del_orphan_item(trans, inode->root, btrfs_ino(inode)); |
7b128766 JB |
3667 | } |
3668 | ||
3669 | /* | |
3670 | * this cleans up any orphans that may be left on the list from the last use | |
3671 | * of this root. | |
3672 | */ | |
66b4ffd1 | 3673 | int btrfs_orphan_cleanup(struct btrfs_root *root) |
7b128766 | 3674 | { |
0b246afa | 3675 | struct btrfs_fs_info *fs_info = root->fs_info; |
7b128766 JB |
3676 | struct btrfs_path *path; |
3677 | struct extent_buffer *leaf; | |
7b128766 JB |
3678 | struct btrfs_key key, found_key; |
3679 | struct btrfs_trans_handle *trans; | |
3680 | struct inode *inode; | |
8f6d7f4f | 3681 | u64 last_objectid = 0; |
f7e9e8fc | 3682 | int ret = 0, nr_unlink = 0; |
7b128766 | 3683 | |
54230013 | 3684 | if (test_and_set_bit(BTRFS_ROOT_ORPHAN_CLEANUP, &root->state)) |
66b4ffd1 | 3685 | return 0; |
c71bf099 YZ |
3686 | |
3687 | path = btrfs_alloc_path(); | |
66b4ffd1 JB |
3688 | if (!path) { |
3689 | ret = -ENOMEM; | |
3690 | goto out; | |
3691 | } | |
e4058b54 | 3692 | path->reada = READA_BACK; |
7b128766 JB |
3693 | |
3694 | key.objectid = BTRFS_ORPHAN_OBJECTID; | |
962a298f | 3695 | key.type = BTRFS_ORPHAN_ITEM_KEY; |
7b128766 JB |
3696 | key.offset = (u64)-1; |
3697 | ||
7b128766 JB |
3698 | while (1) { |
3699 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
66b4ffd1 JB |
3700 | if (ret < 0) |
3701 | goto out; | |
7b128766 JB |
3702 | |
3703 | /* | |
3704 | * if ret == 0 means we found what we were searching for, which | |
25985edc | 3705 | * is weird, but possible, so only screw with path if we didn't |
7b128766 JB |
3706 | * find the key and see if we have stuff that matches |
3707 | */ | |
3708 | if (ret > 0) { | |
66b4ffd1 | 3709 | ret = 0; |
7b128766 JB |
3710 | if (path->slots[0] == 0) |
3711 | break; | |
3712 | path->slots[0]--; | |
3713 | } | |
3714 | ||
3715 | /* pull out the item */ | |
3716 | leaf = path->nodes[0]; | |
7b128766 JB |
3717 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
3718 | ||
3719 | /* make sure the item matches what we want */ | |
3720 | if (found_key.objectid != BTRFS_ORPHAN_OBJECTID) | |
3721 | break; | |
962a298f | 3722 | if (found_key.type != BTRFS_ORPHAN_ITEM_KEY) |
7b128766 JB |
3723 | break; |
3724 | ||
3725 | /* release the path since we're done with it */ | |
b3b4aa74 | 3726 | btrfs_release_path(path); |
7b128766 JB |
3727 | |
3728 | /* | |
3729 | * this is where we are basically btrfs_lookup, without the | |
3730 | * crossing root thing. we store the inode number in the | |
3731 | * offset of the orphan item. | |
3732 | */ | |
8f6d7f4f JB |
3733 | |
3734 | if (found_key.offset == last_objectid) { | |
0b246afa JM |
3735 | btrfs_err(fs_info, |
3736 | "Error removing orphan entry, stopping orphan cleanup"); | |
8f6d7f4f JB |
3737 | ret = -EINVAL; |
3738 | goto out; | |
3739 | } | |
3740 | ||
3741 | last_objectid = found_key.offset; | |
3742 | ||
5d4f98a2 YZ |
3743 | found_key.objectid = found_key.offset; |
3744 | found_key.type = BTRFS_INODE_ITEM_KEY; | |
3745 | found_key.offset = 0; | |
0202e83f | 3746 | inode = btrfs_iget(fs_info->sb, last_objectid, root); |
8c6ffba0 | 3747 | ret = PTR_ERR_OR_ZERO(inode); |
67710892 | 3748 | if (ret && ret != -ENOENT) |
66b4ffd1 | 3749 | goto out; |
7b128766 | 3750 | |
0b246afa | 3751 | if (ret == -ENOENT && root == fs_info->tree_root) { |
f8e9e0b0 | 3752 | struct btrfs_root *dead_root; |
f8e9e0b0 AJ |
3753 | int is_dead_root = 0; |
3754 | ||
3755 | /* | |
0c0218e9 | 3756 | * This is an orphan in the tree root. Currently these |
f8e9e0b0 | 3757 | * could come from 2 sources: |
0c0218e9 | 3758 | * a) a root (snapshot/subvolume) deletion in progress |
f8e9e0b0 | 3759 | * b) a free space cache inode |
0c0218e9 FM |
3760 | * We need to distinguish those two, as the orphan item |
3761 | * for a root must not get deleted before the deletion | |
3762 | * of the snapshot/subvolume's tree completes. | |
3763 | * | |
3764 | * btrfs_find_orphan_roots() ran before us, which has | |
3765 | * found all deleted roots and loaded them into | |
fc7cbcd4 | 3766 | * fs_info->fs_roots_radix. So here we can find if an |
0c0218e9 | 3767 | * orphan item corresponds to a deleted root by looking |
fc7cbcd4 | 3768 | * up the root from that radix tree. |
f8e9e0b0 | 3769 | */ |
a619b3c7 | 3770 | |
fc7cbcd4 DS |
3771 | spin_lock(&fs_info->fs_roots_radix_lock); |
3772 | dead_root = radix_tree_lookup(&fs_info->fs_roots_radix, | |
3773 | (unsigned long)found_key.objectid); | |
a619b3c7 RK |
3774 | if (dead_root && btrfs_root_refs(&dead_root->root_item) == 0) |
3775 | is_dead_root = 1; | |
fc7cbcd4 | 3776 | spin_unlock(&fs_info->fs_roots_radix_lock); |
a619b3c7 | 3777 | |
f8e9e0b0 AJ |
3778 | if (is_dead_root) { |
3779 | /* prevent this orphan from being found again */ | |
3780 | key.offset = found_key.objectid - 1; | |
3781 | continue; | |
3782 | } | |
f7e9e8fc | 3783 | |
f8e9e0b0 | 3784 | } |
f7e9e8fc | 3785 | |
7b128766 | 3786 | /* |
f7e9e8fc | 3787 | * If we have an inode with links, there are a couple of |
70524253 BB |
3788 | * possibilities: |
3789 | * | |
3790 | * 1. We were halfway through creating fsverity metadata for the | |
3791 | * file. In that case, the orphan item represents incomplete | |
3792 | * fsverity metadata which must be cleaned up with | |
3793 | * btrfs_drop_verity_items and deleting the orphan item. | |
3794 | ||
3795 | * 2. Old kernels (before v3.12) used to create an | |
f7e9e8fc OS |
3796 | * orphan item for truncate indicating that there were possibly |
3797 | * extent items past i_size that needed to be deleted. In v3.12, | |
3798 | * truncate was changed to update i_size in sync with the extent | |
3799 | * items, but the (useless) orphan item was still created. Since | |
3800 | * v4.18, we don't create the orphan item for truncate at all. | |
3801 | * | |
3802 | * So, this item could mean that we need to do a truncate, but | |
3803 | * only if this filesystem was last used on a pre-v3.12 kernel | |
3804 | * and was not cleanly unmounted. The odds of that are quite | |
3805 | * slim, and it's a pain to do the truncate now, so just delete | |
3806 | * the orphan item. | |
3807 | * | |
3808 | * It's also possible that this orphan item was supposed to be | |
3809 | * deleted but wasn't. The inode number may have been reused, | |
3810 | * but either way, we can delete the orphan item. | |
7b128766 | 3811 | */ |
f7e9e8fc | 3812 | if (ret == -ENOENT || inode->i_nlink) { |
70524253 BB |
3813 | if (!ret) { |
3814 | ret = btrfs_drop_verity_items(BTRFS_I(inode)); | |
f7e9e8fc | 3815 | iput(inode); |
70524253 BB |
3816 | if (ret) |
3817 | goto out; | |
3818 | } | |
a8c9e576 | 3819 | trans = btrfs_start_transaction(root, 1); |
66b4ffd1 JB |
3820 | if (IS_ERR(trans)) { |
3821 | ret = PTR_ERR(trans); | |
3822 | goto out; | |
3823 | } | |
0b246afa JM |
3824 | btrfs_debug(fs_info, "auto deleting %Lu", |
3825 | found_key.objectid); | |
a8c9e576 JB |
3826 | ret = btrfs_del_orphan_item(trans, root, |
3827 | found_key.objectid); | |
3a45bb20 | 3828 | btrfs_end_transaction(trans); |
4ef31a45 JB |
3829 | if (ret) |
3830 | goto out; | |
7b128766 JB |
3831 | continue; |
3832 | } | |
3833 | ||
f7e9e8fc | 3834 | nr_unlink++; |
7b128766 JB |
3835 | |
3836 | /* this will do delete_inode and everything for us */ | |
3837 | iput(inode); | |
3838 | } | |
3254c876 MX |
3839 | /* release the path since we're done with it */ |
3840 | btrfs_release_path(path); | |
3841 | ||
a575ceeb | 3842 | if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state)) { |
7a7eaa40 | 3843 | trans = btrfs_join_transaction(root); |
66b4ffd1 | 3844 | if (!IS_ERR(trans)) |
3a45bb20 | 3845 | btrfs_end_transaction(trans); |
d68fc57b | 3846 | } |
7b128766 JB |
3847 | |
3848 | if (nr_unlink) | |
0b246afa | 3849 | btrfs_debug(fs_info, "unlinked %d orphans", nr_unlink); |
66b4ffd1 JB |
3850 | |
3851 | out: | |
3852 | if (ret) | |
0b246afa | 3853 | btrfs_err(fs_info, "could not do orphan cleanup %d", ret); |
66b4ffd1 JB |
3854 | btrfs_free_path(path); |
3855 | return ret; | |
7b128766 JB |
3856 | } |
3857 | ||
46a53cca CM |
3858 | /* |
3859 | * very simple check to peek ahead in the leaf looking for xattrs. If we | |
3860 | * don't find any xattrs, we know there can't be any acls. | |
3861 | * | |
3862 | * slot is the slot the inode is in, objectid is the objectid of the inode | |
3863 | */ | |
3864 | static noinline int acls_after_inode_item(struct extent_buffer *leaf, | |
63541927 FDBM |
3865 | int slot, u64 objectid, |
3866 | int *first_xattr_slot) | |
46a53cca CM |
3867 | { |
3868 | u32 nritems = btrfs_header_nritems(leaf); | |
3869 | struct btrfs_key found_key; | |
f23b5a59 JB |
3870 | static u64 xattr_access = 0; |
3871 | static u64 xattr_default = 0; | |
46a53cca CM |
3872 | int scanned = 0; |
3873 | ||
f23b5a59 | 3874 | if (!xattr_access) { |
97d79299 AG |
3875 | xattr_access = btrfs_name_hash(XATTR_NAME_POSIX_ACL_ACCESS, |
3876 | strlen(XATTR_NAME_POSIX_ACL_ACCESS)); | |
3877 | xattr_default = btrfs_name_hash(XATTR_NAME_POSIX_ACL_DEFAULT, | |
3878 | strlen(XATTR_NAME_POSIX_ACL_DEFAULT)); | |
f23b5a59 JB |
3879 | } |
3880 | ||
46a53cca | 3881 | slot++; |
63541927 | 3882 | *first_xattr_slot = -1; |
46a53cca CM |
3883 | while (slot < nritems) { |
3884 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
3885 | ||
3886 | /* we found a different objectid, there must not be acls */ | |
3887 | if (found_key.objectid != objectid) | |
3888 | return 0; | |
3889 | ||
3890 | /* we found an xattr, assume we've got an acl */ | |
f23b5a59 | 3891 | if (found_key.type == BTRFS_XATTR_ITEM_KEY) { |
63541927 FDBM |
3892 | if (*first_xattr_slot == -1) |
3893 | *first_xattr_slot = slot; | |
f23b5a59 JB |
3894 | if (found_key.offset == xattr_access || |
3895 | found_key.offset == xattr_default) | |
3896 | return 1; | |
3897 | } | |
46a53cca CM |
3898 | |
3899 | /* | |
3900 | * we found a key greater than an xattr key, there can't | |
3901 | * be any acls later on | |
3902 | */ | |
3903 | if (found_key.type > BTRFS_XATTR_ITEM_KEY) | |
3904 | return 0; | |
3905 | ||
3906 | slot++; | |
3907 | scanned++; | |
3908 | ||
3909 | /* | |
3910 | * it goes inode, inode backrefs, xattrs, extents, | |
3911 | * so if there are a ton of hard links to an inode there can | |
3912 | * be a lot of backrefs. Don't waste time searching too hard, | |
3913 | * this is just an optimization | |
3914 | */ | |
3915 | if (scanned >= 8) | |
3916 | break; | |
3917 | } | |
3918 | /* we hit the end of the leaf before we found an xattr or | |
3919 | * something larger than an xattr. We have to assume the inode | |
3920 | * has acls | |
3921 | */ | |
63541927 FDBM |
3922 | if (*first_xattr_slot == -1) |
3923 | *first_xattr_slot = slot; | |
46a53cca CM |
3924 | return 1; |
3925 | } | |
3926 | ||
d352ac68 CM |
3927 | /* |
3928 | * read an inode from the btree into the in-memory inode | |
3929 | */ | |
4222ea71 FM |
3930 | static int btrfs_read_locked_inode(struct inode *inode, |
3931 | struct btrfs_path *in_path) | |
39279cc3 | 3932 | { |
0b246afa | 3933 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
4222ea71 | 3934 | struct btrfs_path *path = in_path; |
5f39d397 | 3935 | struct extent_buffer *leaf; |
39279cc3 CM |
3936 | struct btrfs_inode_item *inode_item; |
3937 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
3938 | struct btrfs_key location; | |
67de1176 | 3939 | unsigned long ptr; |
46a53cca | 3940 | int maybe_acls; |
618e21d5 | 3941 | u32 rdev; |
39279cc3 | 3942 | int ret; |
2f7e33d4 | 3943 | bool filled = false; |
63541927 | 3944 | int first_xattr_slot; |
2f7e33d4 MX |
3945 | |
3946 | ret = btrfs_fill_inode(inode, &rdev); | |
3947 | if (!ret) | |
3948 | filled = true; | |
39279cc3 | 3949 | |
4222ea71 FM |
3950 | if (!path) { |
3951 | path = btrfs_alloc_path(); | |
3952 | if (!path) | |
3953 | return -ENOMEM; | |
3954 | } | |
1748f843 | 3955 | |
39279cc3 | 3956 | memcpy(&location, &BTRFS_I(inode)->location, sizeof(location)); |
dc17ff8f | 3957 | |
39279cc3 | 3958 | ret = btrfs_lookup_inode(NULL, root, path, &location, 0); |
67710892 | 3959 | if (ret) { |
4222ea71 FM |
3960 | if (path != in_path) |
3961 | btrfs_free_path(path); | |
f5b3a417 | 3962 | return ret; |
67710892 | 3963 | } |
39279cc3 | 3964 | |
5f39d397 | 3965 | leaf = path->nodes[0]; |
2f7e33d4 MX |
3966 | |
3967 | if (filled) | |
67de1176 | 3968 | goto cache_index; |
2f7e33d4 | 3969 | |
5f39d397 CM |
3970 | inode_item = btrfs_item_ptr(leaf, path->slots[0], |
3971 | struct btrfs_inode_item); | |
5f39d397 | 3972 | inode->i_mode = btrfs_inode_mode(leaf, inode_item); |
bfe86848 | 3973 | set_nlink(inode, btrfs_inode_nlink(leaf, inode_item)); |
2f2f43d3 EB |
3974 | i_uid_write(inode, btrfs_inode_uid(leaf, inode_item)); |
3975 | i_gid_write(inode, btrfs_inode_gid(leaf, inode_item)); | |
6ef06d27 | 3976 | btrfs_i_size_write(BTRFS_I(inode), btrfs_inode_size(leaf, inode_item)); |
41a2ee75 JB |
3977 | btrfs_inode_set_file_extent_range(BTRFS_I(inode), 0, |
3978 | round_up(i_size_read(inode), fs_info->sectorsize)); | |
5f39d397 | 3979 | |
a937b979 DS |
3980 | inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->atime); |
3981 | inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->atime); | |
5f39d397 | 3982 | |
a937b979 DS |
3983 | inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->mtime); |
3984 | inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->mtime); | |
5f39d397 | 3985 | |
a937b979 DS |
3986 | inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->ctime); |
3987 | inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->ctime); | |
5f39d397 | 3988 | |
9cc97d64 | 3989 | BTRFS_I(inode)->i_otime.tv_sec = |
3990 | btrfs_timespec_sec(leaf, &inode_item->otime); | |
3991 | BTRFS_I(inode)->i_otime.tv_nsec = | |
3992 | btrfs_timespec_nsec(leaf, &inode_item->otime); | |
5f39d397 | 3993 | |
a76a3cd4 | 3994 | inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item)); |
e02119d5 | 3995 | BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item); |
5dc562c5 JB |
3996 | BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item); |
3997 | ||
c7f88c4e JL |
3998 | inode_set_iversion_queried(inode, |
3999 | btrfs_inode_sequence(leaf, inode_item)); | |
6e17d30b YD |
4000 | inode->i_generation = BTRFS_I(inode)->generation; |
4001 | inode->i_rdev = 0; | |
4002 | rdev = btrfs_inode_rdev(leaf, inode_item); | |
4003 | ||
4004 | BTRFS_I(inode)->index_cnt = (u64)-1; | |
77eea05e BB |
4005 | btrfs_inode_split_flags(btrfs_inode_flags(leaf, inode_item), |
4006 | &BTRFS_I(inode)->flags, &BTRFS_I(inode)->ro_flags); | |
6e17d30b YD |
4007 | |
4008 | cache_index: | |
5dc562c5 JB |
4009 | /* |
4010 | * If we were modified in the current generation and evicted from memory | |
4011 | * and then re-read we need to do a full sync since we don't have any | |
4012 | * idea about which extents were modified before we were evicted from | |
4013 | * cache. | |
6e17d30b YD |
4014 | * |
4015 | * This is required for both inode re-read from disk and delayed inode | |
088aea3b | 4016 | * in delayed_nodes_tree. |
5dc562c5 | 4017 | */ |
0b246afa | 4018 | if (BTRFS_I(inode)->last_trans == fs_info->generation) |
5dc562c5 JB |
4019 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, |
4020 | &BTRFS_I(inode)->runtime_flags); | |
4021 | ||
bde6c242 FM |
4022 | /* |
4023 | * We don't persist the id of the transaction where an unlink operation | |
4024 | * against the inode was last made. So here we assume the inode might | |
4025 | * have been evicted, and therefore the exact value of last_unlink_trans | |
4026 | * lost, and set it to last_trans to avoid metadata inconsistencies | |
4027 | * between the inode and its parent if the inode is fsync'ed and the log | |
4028 | * replayed. For example, in the scenario: | |
4029 | * | |
4030 | * touch mydir/foo | |
4031 | * ln mydir/foo mydir/bar | |
4032 | * sync | |
4033 | * unlink mydir/bar | |
4034 | * echo 2 > /proc/sys/vm/drop_caches # evicts inode | |
4035 | * xfs_io -c fsync mydir/foo | |
4036 | * <power failure> | |
4037 | * mount fs, triggers fsync log replay | |
4038 | * | |
4039 | * We must make sure that when we fsync our inode foo we also log its | |
4040 | * parent inode, otherwise after log replay the parent still has the | |
4041 | * dentry with the "bar" name but our inode foo has a link count of 1 | |
4042 | * and doesn't have an inode ref with the name "bar" anymore. | |
4043 | * | |
4044 | * Setting last_unlink_trans to last_trans is a pessimistic approach, | |
01327610 | 4045 | * but it guarantees correctness at the expense of occasional full |
bde6c242 FM |
4046 | * transaction commits on fsync if our inode is a directory, or if our |
4047 | * inode is not a directory, logging its parent unnecessarily. | |
4048 | */ | |
4049 | BTRFS_I(inode)->last_unlink_trans = BTRFS_I(inode)->last_trans; | |
4050 | ||
3ebac17c FM |
4051 | /* |
4052 | * Same logic as for last_unlink_trans. We don't persist the generation | |
4053 | * of the last transaction where this inode was used for a reflink | |
4054 | * operation, so after eviction and reloading the inode we must be | |
4055 | * pessimistic and assume the last transaction that modified the inode. | |
4056 | */ | |
4057 | BTRFS_I(inode)->last_reflink_trans = BTRFS_I(inode)->last_trans; | |
4058 | ||
67de1176 MX |
4059 | path->slots[0]++; |
4060 | if (inode->i_nlink != 1 || | |
4061 | path->slots[0] >= btrfs_header_nritems(leaf)) | |
4062 | goto cache_acl; | |
4063 | ||
4064 | btrfs_item_key_to_cpu(leaf, &location, path->slots[0]); | |
4a0cc7ca | 4065 | if (location.objectid != btrfs_ino(BTRFS_I(inode))) |
67de1176 MX |
4066 | goto cache_acl; |
4067 | ||
4068 | ptr = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
4069 | if (location.type == BTRFS_INODE_REF_KEY) { | |
4070 | struct btrfs_inode_ref *ref; | |
4071 | ||
4072 | ref = (struct btrfs_inode_ref *)ptr; | |
4073 | BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref); | |
4074 | } else if (location.type == BTRFS_INODE_EXTREF_KEY) { | |
4075 | struct btrfs_inode_extref *extref; | |
4076 | ||
4077 | extref = (struct btrfs_inode_extref *)ptr; | |
4078 | BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf, | |
4079 | extref); | |
4080 | } | |
2f7e33d4 | 4081 | cache_acl: |
46a53cca CM |
4082 | /* |
4083 | * try to precache a NULL acl entry for files that don't have | |
4084 | * any xattrs or acls | |
4085 | */ | |
33345d01 | 4086 | maybe_acls = acls_after_inode_item(leaf, path->slots[0], |
f85b7379 | 4087 | btrfs_ino(BTRFS_I(inode)), &first_xattr_slot); |
63541927 FDBM |
4088 | if (first_xattr_slot != -1) { |
4089 | path->slots[0] = first_xattr_slot; | |
4090 | ret = btrfs_load_inode_props(inode, path); | |
4091 | if (ret) | |
0b246afa | 4092 | btrfs_err(fs_info, |
351fd353 | 4093 | "error loading props for ino %llu (root %llu): %d", |
4a0cc7ca | 4094 | btrfs_ino(BTRFS_I(inode)), |
63541927 FDBM |
4095 | root->root_key.objectid, ret); |
4096 | } | |
4222ea71 FM |
4097 | if (path != in_path) |
4098 | btrfs_free_path(path); | |
63541927 | 4099 | |
72c04902 AV |
4100 | if (!maybe_acls) |
4101 | cache_no_acl(inode); | |
46a53cca | 4102 | |
39279cc3 | 4103 | switch (inode->i_mode & S_IFMT) { |
39279cc3 CM |
4104 | case S_IFREG: |
4105 | inode->i_mapping->a_ops = &btrfs_aops; | |
4106 | inode->i_fop = &btrfs_file_operations; | |
4107 | inode->i_op = &btrfs_file_inode_operations; | |
4108 | break; | |
4109 | case S_IFDIR: | |
4110 | inode->i_fop = &btrfs_dir_file_operations; | |
67ade058 | 4111 | inode->i_op = &btrfs_dir_inode_operations; |
39279cc3 CM |
4112 | break; |
4113 | case S_IFLNK: | |
4114 | inode->i_op = &btrfs_symlink_inode_operations; | |
21fc61c7 | 4115 | inode_nohighmem(inode); |
4779cc04 | 4116 | inode->i_mapping->a_ops = &btrfs_aops; |
39279cc3 | 4117 | break; |
618e21d5 | 4118 | default: |
0279b4cd | 4119 | inode->i_op = &btrfs_special_inode_operations; |
618e21d5 JB |
4120 | init_special_inode(inode, inode->i_mode, rdev); |
4121 | break; | |
39279cc3 | 4122 | } |
6cbff00f | 4123 | |
7b6a221e | 4124 | btrfs_sync_inode_flags_to_i_flags(inode); |
67710892 | 4125 | return 0; |
39279cc3 CM |
4126 | } |
4127 | ||
d352ac68 CM |
4128 | /* |
4129 | * given a leaf and an inode, copy the inode fields into the leaf | |
4130 | */ | |
e02119d5 CM |
4131 | static void fill_inode_item(struct btrfs_trans_handle *trans, |
4132 | struct extent_buffer *leaf, | |
5f39d397 | 4133 | struct btrfs_inode_item *item, |
39279cc3 CM |
4134 | struct inode *inode) |
4135 | { | |
51fab693 | 4136 | struct btrfs_map_token token; |
77eea05e | 4137 | u64 flags; |
51fab693 | 4138 | |
c82f823c | 4139 | btrfs_init_map_token(&token, leaf); |
5f39d397 | 4140 | |
cc4c13d5 DS |
4141 | btrfs_set_token_inode_uid(&token, item, i_uid_read(inode)); |
4142 | btrfs_set_token_inode_gid(&token, item, i_gid_read(inode)); | |
4143 | btrfs_set_token_inode_size(&token, item, BTRFS_I(inode)->disk_i_size); | |
4144 | btrfs_set_token_inode_mode(&token, item, inode->i_mode); | |
4145 | btrfs_set_token_inode_nlink(&token, item, inode->i_nlink); | |
4146 | ||
4147 | btrfs_set_token_timespec_sec(&token, &item->atime, | |
4148 | inode->i_atime.tv_sec); | |
4149 | btrfs_set_token_timespec_nsec(&token, &item->atime, | |
4150 | inode->i_atime.tv_nsec); | |
4151 | ||
4152 | btrfs_set_token_timespec_sec(&token, &item->mtime, | |
4153 | inode->i_mtime.tv_sec); | |
4154 | btrfs_set_token_timespec_nsec(&token, &item->mtime, | |
4155 | inode->i_mtime.tv_nsec); | |
4156 | ||
4157 | btrfs_set_token_timespec_sec(&token, &item->ctime, | |
4158 | inode->i_ctime.tv_sec); | |
4159 | btrfs_set_token_timespec_nsec(&token, &item->ctime, | |
4160 | inode->i_ctime.tv_nsec); | |
4161 | ||
4162 | btrfs_set_token_timespec_sec(&token, &item->otime, | |
4163 | BTRFS_I(inode)->i_otime.tv_sec); | |
4164 | btrfs_set_token_timespec_nsec(&token, &item->otime, | |
4165 | BTRFS_I(inode)->i_otime.tv_nsec); | |
4166 | ||
4167 | btrfs_set_token_inode_nbytes(&token, item, inode_get_bytes(inode)); | |
4168 | btrfs_set_token_inode_generation(&token, item, | |
4169 | BTRFS_I(inode)->generation); | |
4170 | btrfs_set_token_inode_sequence(&token, item, inode_peek_iversion(inode)); | |
4171 | btrfs_set_token_inode_transid(&token, item, trans->transid); | |
4172 | btrfs_set_token_inode_rdev(&token, item, inode->i_rdev); | |
77eea05e BB |
4173 | flags = btrfs_inode_combine_flags(BTRFS_I(inode)->flags, |
4174 | BTRFS_I(inode)->ro_flags); | |
4175 | btrfs_set_token_inode_flags(&token, item, flags); | |
cc4c13d5 | 4176 | btrfs_set_token_inode_block_group(&token, item, 0); |
39279cc3 CM |
4177 | } |
4178 | ||
d352ac68 CM |
4179 | /* |
4180 | * copy everything in the in-memory inode into the btree. | |
4181 | */ | |
2115133f | 4182 | static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans, |
dfeb9e7c NB |
4183 | struct btrfs_root *root, |
4184 | struct btrfs_inode *inode) | |
39279cc3 CM |
4185 | { |
4186 | struct btrfs_inode_item *inode_item; | |
4187 | struct btrfs_path *path; | |
5f39d397 | 4188 | struct extent_buffer *leaf; |
39279cc3 CM |
4189 | int ret; |
4190 | ||
4191 | path = btrfs_alloc_path(); | |
16cdcec7 MX |
4192 | if (!path) |
4193 | return -ENOMEM; | |
4194 | ||
dfeb9e7c | 4195 | ret = btrfs_lookup_inode(trans, root, path, &inode->location, 1); |
39279cc3 CM |
4196 | if (ret) { |
4197 | if (ret > 0) | |
4198 | ret = -ENOENT; | |
4199 | goto failed; | |
4200 | } | |
4201 | ||
5f39d397 CM |
4202 | leaf = path->nodes[0]; |
4203 | inode_item = btrfs_item_ptr(leaf, path->slots[0], | |
16cdcec7 | 4204 | struct btrfs_inode_item); |
39279cc3 | 4205 | |
dfeb9e7c | 4206 | fill_inode_item(trans, leaf, inode_item, &inode->vfs_inode); |
5f39d397 | 4207 | btrfs_mark_buffer_dirty(leaf); |
dfeb9e7c | 4208 | btrfs_set_inode_last_trans(trans, inode); |
39279cc3 CM |
4209 | ret = 0; |
4210 | failed: | |
39279cc3 CM |
4211 | btrfs_free_path(path); |
4212 | return ret; | |
4213 | } | |
4214 | ||
2115133f CM |
4215 | /* |
4216 | * copy everything in the in-memory inode into the btree. | |
4217 | */ | |
4218 | noinline int btrfs_update_inode(struct btrfs_trans_handle *trans, | |
9a56fcd1 NB |
4219 | struct btrfs_root *root, |
4220 | struct btrfs_inode *inode) | |
2115133f | 4221 | { |
0b246afa | 4222 | struct btrfs_fs_info *fs_info = root->fs_info; |
2115133f CM |
4223 | int ret; |
4224 | ||
4225 | /* | |
4226 | * If the inode is a free space inode, we can deadlock during commit | |
4227 | * if we put it into the delayed code. | |
4228 | * | |
4229 | * The data relocation inode should also be directly updated | |
4230 | * without delay | |
4231 | */ | |
9a56fcd1 | 4232 | if (!btrfs_is_free_space_inode(inode) |
37f00a6d | 4233 | && !btrfs_is_data_reloc_root(root) |
0b246afa | 4234 | && !test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) { |
8ea05e3a AB |
4235 | btrfs_update_root_times(trans, root); |
4236 | ||
9a56fcd1 | 4237 | ret = btrfs_delayed_update_inode(trans, root, inode); |
2115133f | 4238 | if (!ret) |
9a56fcd1 | 4239 | btrfs_set_inode_last_trans(trans, inode); |
2115133f CM |
4240 | return ret; |
4241 | } | |
4242 | ||
9a56fcd1 | 4243 | return btrfs_update_inode_item(trans, root, inode); |
2115133f CM |
4244 | } |
4245 | ||
729f7961 NB |
4246 | int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans, |
4247 | struct btrfs_root *root, struct btrfs_inode *inode) | |
2115133f CM |
4248 | { |
4249 | int ret; | |
4250 | ||
729f7961 | 4251 | ret = btrfs_update_inode(trans, root, inode); |
2115133f | 4252 | if (ret == -ENOSPC) |
729f7961 | 4253 | return btrfs_update_inode_item(trans, root, inode); |
2115133f CM |
4254 | return ret; |
4255 | } | |
4256 | ||
d352ac68 CM |
4257 | /* |
4258 | * unlink helper that gets used here in inode.c and in the tree logging | |
4259 | * recovery code. It remove a link in a directory with a given name, and | |
4260 | * also drops the back refs in the inode to the directory | |
4261 | */ | |
92986796 | 4262 | static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans, |
4ec5934e NB |
4263 | struct btrfs_inode *dir, |
4264 | struct btrfs_inode *inode, | |
88d2beec FM |
4265 | const char *name, int name_len, |
4266 | struct btrfs_rename_ctx *rename_ctx) | |
39279cc3 | 4267 | { |
4467af88 | 4268 | struct btrfs_root *root = dir->root; |
0b246afa | 4269 | struct btrfs_fs_info *fs_info = root->fs_info; |
39279cc3 | 4270 | struct btrfs_path *path; |
39279cc3 | 4271 | int ret = 0; |
39279cc3 | 4272 | struct btrfs_dir_item *di; |
aec7477b | 4273 | u64 index; |
33345d01 LZ |
4274 | u64 ino = btrfs_ino(inode); |
4275 | u64 dir_ino = btrfs_ino(dir); | |
39279cc3 CM |
4276 | |
4277 | path = btrfs_alloc_path(); | |
54aa1f4d CM |
4278 | if (!path) { |
4279 | ret = -ENOMEM; | |
554233a6 | 4280 | goto out; |
54aa1f4d CM |
4281 | } |
4282 | ||
33345d01 | 4283 | di = btrfs_lookup_dir_item(trans, root, path, dir_ino, |
39279cc3 | 4284 | name, name_len, -1); |
3cf5068f LB |
4285 | if (IS_ERR_OR_NULL(di)) { |
4286 | ret = di ? PTR_ERR(di) : -ENOENT; | |
39279cc3 CM |
4287 | goto err; |
4288 | } | |
39279cc3 | 4289 | ret = btrfs_delete_one_dir_name(trans, root, path, di); |
54aa1f4d CM |
4290 | if (ret) |
4291 | goto err; | |
b3b4aa74 | 4292 | btrfs_release_path(path); |
39279cc3 | 4293 | |
67de1176 MX |
4294 | /* |
4295 | * If we don't have dir index, we have to get it by looking up | |
4296 | * the inode ref, since we get the inode ref, remove it directly, | |
4297 | * it is unnecessary to do delayed deletion. | |
4298 | * | |
4299 | * But if we have dir index, needn't search inode ref to get it. | |
4300 | * Since the inode ref is close to the inode item, it is better | |
4301 | * that we delay to delete it, and just do this deletion when | |
4302 | * we update the inode item. | |
4303 | */ | |
4ec5934e | 4304 | if (inode->dir_index) { |
67de1176 MX |
4305 | ret = btrfs_delayed_delete_inode_ref(inode); |
4306 | if (!ret) { | |
4ec5934e | 4307 | index = inode->dir_index; |
67de1176 MX |
4308 | goto skip_backref; |
4309 | } | |
4310 | } | |
4311 | ||
33345d01 LZ |
4312 | ret = btrfs_del_inode_ref(trans, root, name, name_len, ino, |
4313 | dir_ino, &index); | |
aec7477b | 4314 | if (ret) { |
0b246afa | 4315 | btrfs_info(fs_info, |
c2cf52eb | 4316 | "failed to delete reference to %.*s, inode %llu parent %llu", |
c1c9ff7c | 4317 | name_len, name, ino, dir_ino); |
66642832 | 4318 | btrfs_abort_transaction(trans, ret); |
aec7477b JB |
4319 | goto err; |
4320 | } | |
67de1176 | 4321 | skip_backref: |
88d2beec FM |
4322 | if (rename_ctx) |
4323 | rename_ctx->index = index; | |
4324 | ||
9add2945 | 4325 | ret = btrfs_delete_delayed_dir_index(trans, dir, index); |
79787eaa | 4326 | if (ret) { |
66642832 | 4327 | btrfs_abort_transaction(trans, ret); |
39279cc3 | 4328 | goto err; |
79787eaa | 4329 | } |
39279cc3 | 4330 | |
259c4b96 FM |
4331 | /* |
4332 | * If we are in a rename context, we don't need to update anything in the | |
4333 | * log. That will be done later during the rename by btrfs_log_new_name(). | |
143823cf | 4334 | * Besides that, doing it here would only cause extra unnecessary btree |
259c4b96 FM |
4335 | * operations on the log tree, increasing latency for applications. |
4336 | */ | |
4337 | if (!rename_ctx) { | |
4338 | btrfs_del_inode_ref_in_log(trans, root, name, name_len, inode, | |
4339 | dir_ino); | |
4340 | btrfs_del_dir_entries_in_log(trans, root, name, name_len, dir, | |
4341 | index); | |
4342 | } | |
63611e73 JB |
4343 | |
4344 | /* | |
4345 | * If we have a pending delayed iput we could end up with the final iput | |
4346 | * being run in btrfs-cleaner context. If we have enough of these built | |
4347 | * up we can end up burning a lot of time in btrfs-cleaner without any | |
4348 | * way to throttle the unlinks. Since we're currently holding a ref on | |
4349 | * the inode we can run the delayed iput here without any issues as the | |
4350 | * final iput won't be done until after we drop the ref we're currently | |
4351 | * holding. | |
4352 | */ | |
4353 | btrfs_run_delayed_iput(fs_info, inode); | |
39279cc3 CM |
4354 | err: |
4355 | btrfs_free_path(path); | |
e02119d5 CM |
4356 | if (ret) |
4357 | goto out; | |
4358 | ||
6ef06d27 | 4359 | btrfs_i_size_write(dir, dir->vfs_inode.i_size - name_len * 2); |
4ec5934e NB |
4360 | inode_inc_iversion(&inode->vfs_inode); |
4361 | inode_inc_iversion(&dir->vfs_inode); | |
c1867eb3 DS |
4362 | inode->vfs_inode.i_ctime = current_time(&inode->vfs_inode); |
4363 | dir->vfs_inode.i_mtime = inode->vfs_inode.i_ctime; | |
4364 | dir->vfs_inode.i_ctime = inode->vfs_inode.i_ctime; | |
9a56fcd1 | 4365 | ret = btrfs_update_inode(trans, root, dir); |
e02119d5 | 4366 | out: |
39279cc3 CM |
4367 | return ret; |
4368 | } | |
4369 | ||
92986796 | 4370 | int btrfs_unlink_inode(struct btrfs_trans_handle *trans, |
4ec5934e | 4371 | struct btrfs_inode *dir, struct btrfs_inode *inode, |
92986796 AV |
4372 | const char *name, int name_len) |
4373 | { | |
4374 | int ret; | |
88d2beec | 4375 | ret = __btrfs_unlink_inode(trans, dir, inode, name, name_len, NULL); |
92986796 | 4376 | if (!ret) { |
4ec5934e | 4377 | drop_nlink(&inode->vfs_inode); |
4467af88 | 4378 | ret = btrfs_update_inode(trans, inode->root, inode); |
92986796 AV |
4379 | } |
4380 | return ret; | |
4381 | } | |
39279cc3 | 4382 | |
a22285a6 YZ |
4383 | /* |
4384 | * helper to start transaction for unlink and rmdir. | |
4385 | * | |
d52be818 JB |
4386 | * unlink and rmdir are special in btrfs, they do not always free space, so |
4387 | * if we cannot make our reservations the normal way try and see if there is | |
4388 | * plenty of slack room in the global reserve to migrate, otherwise we cannot | |
4389 | * allow the unlink to occur. | |
a22285a6 | 4390 | */ |
d52be818 | 4391 | static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir) |
4df27c4d | 4392 | { |
a22285a6 | 4393 | struct btrfs_root *root = BTRFS_I(dir)->root; |
4df27c4d | 4394 | |
e70bea5f JB |
4395 | /* |
4396 | * 1 for the possible orphan item | |
4397 | * 1 for the dir item | |
4398 | * 1 for the dir index | |
4399 | * 1 for the inode ref | |
e70bea5f | 4400 | * 1 for the inode |
bca4ad7c | 4401 | * 1 for the parent inode |
e70bea5f | 4402 | */ |
bca4ad7c | 4403 | return btrfs_start_transaction_fallback_global_rsv(root, 6); |
a22285a6 YZ |
4404 | } |
4405 | ||
4406 | static int btrfs_unlink(struct inode *dir, struct dentry *dentry) | |
4407 | { | |
a22285a6 | 4408 | struct btrfs_trans_handle *trans; |
2b0143b5 | 4409 | struct inode *inode = d_inode(dentry); |
a22285a6 | 4410 | int ret; |
a22285a6 | 4411 | |
d52be818 | 4412 | trans = __unlink_start_trans(dir); |
a22285a6 YZ |
4413 | if (IS_ERR(trans)) |
4414 | return PTR_ERR(trans); | |
5f39d397 | 4415 | |
4ec5934e NB |
4416 | btrfs_record_unlink_dir(trans, BTRFS_I(dir), BTRFS_I(d_inode(dentry)), |
4417 | 0); | |
12fcfd22 | 4418 | |
4467af88 | 4419 | ret = btrfs_unlink_inode(trans, BTRFS_I(dir), |
4ec5934e NB |
4420 | BTRFS_I(d_inode(dentry)), dentry->d_name.name, |
4421 | dentry->d_name.len); | |
b532402e TI |
4422 | if (ret) |
4423 | goto out; | |
7b128766 | 4424 | |
a22285a6 | 4425 | if (inode->i_nlink == 0) { |
73f2e545 | 4426 | ret = btrfs_orphan_add(trans, BTRFS_I(inode)); |
b532402e TI |
4427 | if (ret) |
4428 | goto out; | |
a22285a6 | 4429 | } |
7b128766 | 4430 | |
b532402e | 4431 | out: |
3a45bb20 | 4432 | btrfs_end_transaction(trans); |
4467af88 | 4433 | btrfs_btree_balance_dirty(BTRFS_I(dir)->root->fs_info); |
39279cc3 CM |
4434 | return ret; |
4435 | } | |
4436 | ||
f60a2364 | 4437 | static int btrfs_unlink_subvol(struct btrfs_trans_handle *trans, |
045d3967 | 4438 | struct inode *dir, struct dentry *dentry) |
4df27c4d | 4439 | { |
401b3b19 | 4440 | struct btrfs_root *root = BTRFS_I(dir)->root; |
045d3967 | 4441 | struct btrfs_inode *inode = BTRFS_I(d_inode(dentry)); |
4df27c4d YZ |
4442 | struct btrfs_path *path; |
4443 | struct extent_buffer *leaf; | |
4444 | struct btrfs_dir_item *di; | |
4445 | struct btrfs_key key; | |
045d3967 JB |
4446 | const char *name = dentry->d_name.name; |
4447 | int name_len = dentry->d_name.len; | |
4df27c4d YZ |
4448 | u64 index; |
4449 | int ret; | |
045d3967 | 4450 | u64 objectid; |
4a0cc7ca | 4451 | u64 dir_ino = btrfs_ino(BTRFS_I(dir)); |
4df27c4d | 4452 | |
045d3967 JB |
4453 | if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID) { |
4454 | objectid = inode->root->root_key.objectid; | |
4455 | } else if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) { | |
4456 | objectid = inode->location.objectid; | |
4457 | } else { | |
4458 | WARN_ON(1); | |
4459 | return -EINVAL; | |
4460 | } | |
4461 | ||
4df27c4d YZ |
4462 | path = btrfs_alloc_path(); |
4463 | if (!path) | |
4464 | return -ENOMEM; | |
4465 | ||
33345d01 | 4466 | di = btrfs_lookup_dir_item(trans, root, path, dir_ino, |
4df27c4d | 4467 | name, name_len, -1); |
79787eaa | 4468 | if (IS_ERR_OR_NULL(di)) { |
3cf5068f | 4469 | ret = di ? PTR_ERR(di) : -ENOENT; |
79787eaa JM |
4470 | goto out; |
4471 | } | |
4df27c4d YZ |
4472 | |
4473 | leaf = path->nodes[0]; | |
4474 | btrfs_dir_item_key_to_cpu(leaf, di, &key); | |
4475 | WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid); | |
4476 | ret = btrfs_delete_one_dir_name(trans, root, path, di); | |
79787eaa | 4477 | if (ret) { |
66642832 | 4478 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
4479 | goto out; |
4480 | } | |
b3b4aa74 | 4481 | btrfs_release_path(path); |
4df27c4d | 4482 | |
d49d3287 JB |
4483 | /* |
4484 | * This is a placeholder inode for a subvolume we didn't have a | |
4485 | * reference to at the time of the snapshot creation. In the meantime | |
4486 | * we could have renamed the real subvol link into our snapshot, so | |
1a9fd417 | 4487 | * depending on btrfs_del_root_ref to return -ENOENT here is incorrect. |
d49d3287 JB |
4488 | * Instead simply lookup the dir_index_item for this entry so we can |
4489 | * remove it. Otherwise we know we have a ref to the root and we can | |
4490 | * call btrfs_del_root_ref, and it _shouldn't_ fail. | |
4491 | */ | |
4492 | if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) { | |
33345d01 | 4493 | di = btrfs_search_dir_index_item(root, path, dir_ino, |
4df27c4d | 4494 | name, name_len); |
79787eaa JM |
4495 | if (IS_ERR_OR_NULL(di)) { |
4496 | if (!di) | |
4497 | ret = -ENOENT; | |
4498 | else | |
4499 | ret = PTR_ERR(di); | |
66642832 | 4500 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
4501 | goto out; |
4502 | } | |
4df27c4d YZ |
4503 | |
4504 | leaf = path->nodes[0]; | |
4505 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
4df27c4d | 4506 | index = key.offset; |
d49d3287 JB |
4507 | btrfs_release_path(path); |
4508 | } else { | |
4509 | ret = btrfs_del_root_ref(trans, objectid, | |
4510 | root->root_key.objectid, dir_ino, | |
4511 | &index, name, name_len); | |
4512 | if (ret) { | |
4513 | btrfs_abort_transaction(trans, ret); | |
4514 | goto out; | |
4515 | } | |
4df27c4d YZ |
4516 | } |
4517 | ||
9add2945 | 4518 | ret = btrfs_delete_delayed_dir_index(trans, BTRFS_I(dir), index); |
79787eaa | 4519 | if (ret) { |
66642832 | 4520 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
4521 | goto out; |
4522 | } | |
4df27c4d | 4523 | |
6ef06d27 | 4524 | btrfs_i_size_write(BTRFS_I(dir), dir->i_size - name_len * 2); |
0c4d2d95 | 4525 | inode_inc_iversion(dir); |
c1867eb3 DS |
4526 | dir->i_mtime = current_time(dir); |
4527 | dir->i_ctime = dir->i_mtime; | |
729f7961 | 4528 | ret = btrfs_update_inode_fallback(trans, root, BTRFS_I(dir)); |
79787eaa | 4529 | if (ret) |
66642832 | 4530 | btrfs_abort_transaction(trans, ret); |
79787eaa | 4531 | out: |
71d7aed0 | 4532 | btrfs_free_path(path); |
79787eaa | 4533 | return ret; |
4df27c4d YZ |
4534 | } |
4535 | ||
ec42f167 MT |
4536 | /* |
4537 | * Helper to check if the subvolume references other subvolumes or if it's | |
4538 | * default. | |
4539 | */ | |
f60a2364 | 4540 | static noinline int may_destroy_subvol(struct btrfs_root *root) |
ec42f167 MT |
4541 | { |
4542 | struct btrfs_fs_info *fs_info = root->fs_info; | |
4543 | struct btrfs_path *path; | |
4544 | struct btrfs_dir_item *di; | |
4545 | struct btrfs_key key; | |
4546 | u64 dir_id; | |
4547 | int ret; | |
4548 | ||
4549 | path = btrfs_alloc_path(); | |
4550 | if (!path) | |
4551 | return -ENOMEM; | |
4552 | ||
4553 | /* Make sure this root isn't set as the default subvol */ | |
4554 | dir_id = btrfs_super_root_dir(fs_info->super_copy); | |
4555 | di = btrfs_lookup_dir_item(NULL, fs_info->tree_root, path, | |
4556 | dir_id, "default", 7, 0); | |
4557 | if (di && !IS_ERR(di)) { | |
4558 | btrfs_dir_item_key_to_cpu(path->nodes[0], di, &key); | |
4559 | if (key.objectid == root->root_key.objectid) { | |
4560 | ret = -EPERM; | |
4561 | btrfs_err(fs_info, | |
4562 | "deleting default subvolume %llu is not allowed", | |
4563 | key.objectid); | |
4564 | goto out; | |
4565 | } | |
4566 | btrfs_release_path(path); | |
4567 | } | |
4568 | ||
4569 | key.objectid = root->root_key.objectid; | |
4570 | key.type = BTRFS_ROOT_REF_KEY; | |
4571 | key.offset = (u64)-1; | |
4572 | ||
4573 | ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0); | |
4574 | if (ret < 0) | |
4575 | goto out; | |
4576 | BUG_ON(ret == 0); | |
4577 | ||
4578 | ret = 0; | |
4579 | if (path->slots[0] > 0) { | |
4580 | path->slots[0]--; | |
4581 | btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]); | |
4582 | if (key.objectid == root->root_key.objectid && | |
4583 | key.type == BTRFS_ROOT_REF_KEY) | |
4584 | ret = -ENOTEMPTY; | |
4585 | } | |
4586 | out: | |
4587 | btrfs_free_path(path); | |
4588 | return ret; | |
4589 | } | |
4590 | ||
20a68004 NB |
4591 | /* Delete all dentries for inodes belonging to the root */ |
4592 | static void btrfs_prune_dentries(struct btrfs_root *root) | |
4593 | { | |
4594 | struct btrfs_fs_info *fs_info = root->fs_info; | |
4595 | struct rb_node *node; | |
4596 | struct rb_node *prev; | |
4597 | struct btrfs_inode *entry; | |
4598 | struct inode *inode; | |
4599 | u64 objectid = 0; | |
4600 | ||
84961539 | 4601 | if (!BTRFS_FS_ERROR(fs_info)) |
20a68004 NB |
4602 | WARN_ON(btrfs_root_refs(&root->root_item) != 0); |
4603 | ||
4604 | spin_lock(&root->inode_lock); | |
4605 | again: | |
4606 | node = root->inode_tree.rb_node; | |
4607 | prev = NULL; | |
4608 | while (node) { | |
4609 | prev = node; | |
4610 | entry = rb_entry(node, struct btrfs_inode, rb_node); | |
4611 | ||
37508515 | 4612 | if (objectid < btrfs_ino(entry)) |
20a68004 | 4613 | node = node->rb_left; |
37508515 | 4614 | else if (objectid > btrfs_ino(entry)) |
20a68004 NB |
4615 | node = node->rb_right; |
4616 | else | |
4617 | break; | |
4618 | } | |
4619 | if (!node) { | |
4620 | while (prev) { | |
4621 | entry = rb_entry(prev, struct btrfs_inode, rb_node); | |
37508515 | 4622 | if (objectid <= btrfs_ino(entry)) { |
20a68004 NB |
4623 | node = prev; |
4624 | break; | |
4625 | } | |
4626 | prev = rb_next(prev); | |
4627 | } | |
4628 | } | |
4629 | while (node) { | |
4630 | entry = rb_entry(node, struct btrfs_inode, rb_node); | |
37508515 | 4631 | objectid = btrfs_ino(entry) + 1; |
20a68004 NB |
4632 | inode = igrab(&entry->vfs_inode); |
4633 | if (inode) { | |
4634 | spin_unlock(&root->inode_lock); | |
4635 | if (atomic_read(&inode->i_count) > 1) | |
4636 | d_prune_aliases(inode); | |
4637 | /* | |
4638 | * btrfs_drop_inode will have it removed from the inode | |
4639 | * cache when its usage count hits zero. | |
4640 | */ | |
4641 | iput(inode); | |
4642 | cond_resched(); | |
4643 | spin_lock(&root->inode_lock); | |
4644 | goto again; | |
4645 | } | |
4646 | ||
4647 | if (cond_resched_lock(&root->inode_lock)) | |
4648 | goto again; | |
4649 | ||
4650 | node = rb_next(node); | |
4651 | } | |
4652 | spin_unlock(&root->inode_lock); | |
4653 | } | |
4654 | ||
f60a2364 MT |
4655 | int btrfs_delete_subvolume(struct inode *dir, struct dentry *dentry) |
4656 | { | |
4657 | struct btrfs_fs_info *fs_info = btrfs_sb(dentry->d_sb); | |
4658 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
4659 | struct inode *inode = d_inode(dentry); | |
4660 | struct btrfs_root *dest = BTRFS_I(inode)->root; | |
4661 | struct btrfs_trans_handle *trans; | |
4662 | struct btrfs_block_rsv block_rsv; | |
4663 | u64 root_flags; | |
f60a2364 | 4664 | int ret; |
f60a2364 MT |
4665 | |
4666 | /* | |
4667 | * Don't allow to delete a subvolume with send in progress. This is | |
4668 | * inside the inode lock so the error handling that has to drop the bit | |
4669 | * again is not run concurrently. | |
4670 | */ | |
4671 | spin_lock(&dest->root_item_lock); | |
a7176f74 | 4672 | if (dest->send_in_progress) { |
f60a2364 MT |
4673 | spin_unlock(&dest->root_item_lock); |
4674 | btrfs_warn(fs_info, | |
4675 | "attempt to delete subvolume %llu during send", | |
4676 | dest->root_key.objectid); | |
4677 | return -EPERM; | |
4678 | } | |
60021bd7 KH |
4679 | if (atomic_read(&dest->nr_swapfiles)) { |
4680 | spin_unlock(&dest->root_item_lock); | |
4681 | btrfs_warn(fs_info, | |
4682 | "attempt to delete subvolume %llu with active swapfile", | |
4683 | root->root_key.objectid); | |
4684 | return -EPERM; | |
4685 | } | |
a7176f74 LF |
4686 | root_flags = btrfs_root_flags(&dest->root_item); |
4687 | btrfs_set_root_flags(&dest->root_item, | |
4688 | root_flags | BTRFS_ROOT_SUBVOL_DEAD); | |
4689 | spin_unlock(&dest->root_item_lock); | |
f60a2364 MT |
4690 | |
4691 | down_write(&fs_info->subvol_sem); | |
4692 | ||
ee0d904f NB |
4693 | ret = may_destroy_subvol(dest); |
4694 | if (ret) | |
f60a2364 MT |
4695 | goto out_up_write; |
4696 | ||
4697 | btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP); | |
4698 | /* | |
4699 | * One for dir inode, | |
4700 | * two for dir entries, | |
4701 | * two for root ref/backref. | |
4702 | */ | |
ee0d904f NB |
4703 | ret = btrfs_subvolume_reserve_metadata(root, &block_rsv, 5, true); |
4704 | if (ret) | |
f60a2364 MT |
4705 | goto out_up_write; |
4706 | ||
4707 | trans = btrfs_start_transaction(root, 0); | |
4708 | if (IS_ERR(trans)) { | |
ee0d904f | 4709 | ret = PTR_ERR(trans); |
f60a2364 MT |
4710 | goto out_release; |
4711 | } | |
4712 | trans->block_rsv = &block_rsv; | |
4713 | trans->bytes_reserved = block_rsv.size; | |
4714 | ||
4715 | btrfs_record_snapshot_destroy(trans, BTRFS_I(dir)); | |
4716 | ||
045d3967 | 4717 | ret = btrfs_unlink_subvol(trans, dir, dentry); |
f60a2364 | 4718 | if (ret) { |
f60a2364 MT |
4719 | btrfs_abort_transaction(trans, ret); |
4720 | goto out_end_trans; | |
4721 | } | |
4722 | ||
2731f518 JB |
4723 | ret = btrfs_record_root_in_trans(trans, dest); |
4724 | if (ret) { | |
4725 | btrfs_abort_transaction(trans, ret); | |
4726 | goto out_end_trans; | |
4727 | } | |
f60a2364 MT |
4728 | |
4729 | memset(&dest->root_item.drop_progress, 0, | |
4730 | sizeof(dest->root_item.drop_progress)); | |
c8422684 | 4731 | btrfs_set_root_drop_level(&dest->root_item, 0); |
f60a2364 MT |
4732 | btrfs_set_root_refs(&dest->root_item, 0); |
4733 | ||
4734 | if (!test_and_set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &dest->state)) { | |
4735 | ret = btrfs_insert_orphan_item(trans, | |
4736 | fs_info->tree_root, | |
4737 | dest->root_key.objectid); | |
4738 | if (ret) { | |
4739 | btrfs_abort_transaction(trans, ret); | |
f60a2364 MT |
4740 | goto out_end_trans; |
4741 | } | |
4742 | } | |
4743 | ||
d1957791 | 4744 | ret = btrfs_uuid_tree_remove(trans, dest->root_item.uuid, |
f60a2364 MT |
4745 | BTRFS_UUID_KEY_SUBVOL, |
4746 | dest->root_key.objectid); | |
4747 | if (ret && ret != -ENOENT) { | |
4748 | btrfs_abort_transaction(trans, ret); | |
f60a2364 MT |
4749 | goto out_end_trans; |
4750 | } | |
4751 | if (!btrfs_is_empty_uuid(dest->root_item.received_uuid)) { | |
d1957791 | 4752 | ret = btrfs_uuid_tree_remove(trans, |
f60a2364 MT |
4753 | dest->root_item.received_uuid, |
4754 | BTRFS_UUID_KEY_RECEIVED_SUBVOL, | |
4755 | dest->root_key.objectid); | |
4756 | if (ret && ret != -ENOENT) { | |
4757 | btrfs_abort_transaction(trans, ret); | |
f60a2364 MT |
4758 | goto out_end_trans; |
4759 | } | |
4760 | } | |
4761 | ||
082b6c97 QW |
4762 | free_anon_bdev(dest->anon_dev); |
4763 | dest->anon_dev = 0; | |
f60a2364 MT |
4764 | out_end_trans: |
4765 | trans->block_rsv = NULL; | |
4766 | trans->bytes_reserved = 0; | |
4767 | ret = btrfs_end_transaction(trans); | |
f60a2364 MT |
4768 | inode->i_flags |= S_DEAD; |
4769 | out_release: | |
e85fde51 | 4770 | btrfs_subvolume_release_metadata(root, &block_rsv); |
f60a2364 MT |
4771 | out_up_write: |
4772 | up_write(&fs_info->subvol_sem); | |
ee0d904f | 4773 | if (ret) { |
f60a2364 MT |
4774 | spin_lock(&dest->root_item_lock); |
4775 | root_flags = btrfs_root_flags(&dest->root_item); | |
4776 | btrfs_set_root_flags(&dest->root_item, | |
4777 | root_flags & ~BTRFS_ROOT_SUBVOL_DEAD); | |
4778 | spin_unlock(&dest->root_item_lock); | |
4779 | } else { | |
4780 | d_invalidate(dentry); | |
20a68004 | 4781 | btrfs_prune_dentries(dest); |
f60a2364 | 4782 | ASSERT(dest->send_in_progress == 0); |
f60a2364 MT |
4783 | } |
4784 | ||
ee0d904f | 4785 | return ret; |
f60a2364 MT |
4786 | } |
4787 | ||
39279cc3 CM |
4788 | static int btrfs_rmdir(struct inode *dir, struct dentry *dentry) |
4789 | { | |
2b0143b5 | 4790 | struct inode *inode = d_inode(dentry); |
813febdb | 4791 | struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info; |
1832a6d5 | 4792 | int err = 0; |
39279cc3 | 4793 | struct btrfs_trans_handle *trans; |
44f714da | 4794 | u64 last_unlink_trans; |
39279cc3 | 4795 | |
b3ae244e | 4796 | if (inode->i_size > BTRFS_EMPTY_DIR_SIZE) |
134d4512 | 4797 | return -ENOTEMPTY; |
813febdb JB |
4798 | if (btrfs_ino(BTRFS_I(inode)) == BTRFS_FIRST_FREE_OBJECTID) { |
4799 | if (unlikely(btrfs_fs_incompat(fs_info, EXTENT_TREE_V2))) { | |
4800 | btrfs_err(fs_info, | |
4801 | "extent tree v2 doesn't support snapshot deletion yet"); | |
4802 | return -EOPNOTSUPP; | |
4803 | } | |
a79a464d | 4804 | return btrfs_delete_subvolume(dir, dentry); |
813febdb | 4805 | } |
134d4512 | 4806 | |
d52be818 | 4807 | trans = __unlink_start_trans(dir); |
a22285a6 | 4808 | if (IS_ERR(trans)) |
5df6a9f6 | 4809 | return PTR_ERR(trans); |
5df6a9f6 | 4810 | |
4a0cc7ca | 4811 | if (unlikely(btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) { |
045d3967 | 4812 | err = btrfs_unlink_subvol(trans, dir, dentry); |
4df27c4d YZ |
4813 | goto out; |
4814 | } | |
4815 | ||
73f2e545 | 4816 | err = btrfs_orphan_add(trans, BTRFS_I(inode)); |
7b128766 | 4817 | if (err) |
4df27c4d | 4818 | goto out; |
7b128766 | 4819 | |
44f714da FM |
4820 | last_unlink_trans = BTRFS_I(inode)->last_unlink_trans; |
4821 | ||
39279cc3 | 4822 | /* now the directory is empty */ |
4467af88 | 4823 | err = btrfs_unlink_inode(trans, BTRFS_I(dir), |
4ec5934e NB |
4824 | BTRFS_I(d_inode(dentry)), dentry->d_name.name, |
4825 | dentry->d_name.len); | |
44f714da | 4826 | if (!err) { |
6ef06d27 | 4827 | btrfs_i_size_write(BTRFS_I(inode), 0); |
44f714da FM |
4828 | /* |
4829 | * Propagate the last_unlink_trans value of the deleted dir to | |
4830 | * its parent directory. This is to prevent an unrecoverable | |
4831 | * log tree in the case we do something like this: | |
4832 | * 1) create dir foo | |
4833 | * 2) create snapshot under dir foo | |
4834 | * 3) delete the snapshot | |
4835 | * 4) rmdir foo | |
4836 | * 5) mkdir foo | |
4837 | * 6) fsync foo or some file inside foo | |
4838 | */ | |
4839 | if (last_unlink_trans >= trans->transid) | |
4840 | BTRFS_I(dir)->last_unlink_trans = last_unlink_trans; | |
4841 | } | |
4df27c4d | 4842 | out: |
3a45bb20 | 4843 | btrfs_end_transaction(trans); |
813febdb | 4844 | btrfs_btree_balance_dirty(fs_info); |
3954401f | 4845 | |
39279cc3 CM |
4846 | return err; |
4847 | } | |
4848 | ||
39279cc3 | 4849 | /* |
9703fefe | 4850 | * btrfs_truncate_block - read, zero a chunk and write a block |
2aaa6655 JB |
4851 | * @inode - inode that we're zeroing |
4852 | * @from - the offset to start zeroing | |
4853 | * @len - the length to zero, 0 to zero the entire range respective to the | |
4854 | * offset | |
4855 | * @front - zero up to the offset instead of from the offset on | |
4856 | * | |
9703fefe | 4857 | * This will find the block for the "from" offset and cow the block and zero the |
2aaa6655 | 4858 | * part we want to zero. This is used with truncate and hole punching. |
39279cc3 | 4859 | */ |
217f42eb NB |
4860 | int btrfs_truncate_block(struct btrfs_inode *inode, loff_t from, loff_t len, |
4861 | int front) | |
39279cc3 | 4862 | { |
217f42eb NB |
4863 | struct btrfs_fs_info *fs_info = inode->root->fs_info; |
4864 | struct address_space *mapping = inode->vfs_inode.i_mapping; | |
4865 | struct extent_io_tree *io_tree = &inode->io_tree; | |
e6dcd2dc | 4866 | struct btrfs_ordered_extent *ordered; |
2ac55d41 | 4867 | struct extent_state *cached_state = NULL; |
364ecf36 | 4868 | struct extent_changeset *data_reserved = NULL; |
6d4572a9 | 4869 | bool only_release_metadata = false; |
0b246afa | 4870 | u32 blocksize = fs_info->sectorsize; |
09cbfeaf | 4871 | pgoff_t index = from >> PAGE_SHIFT; |
9703fefe | 4872 | unsigned offset = from & (blocksize - 1); |
39279cc3 | 4873 | struct page *page; |
3b16a4e3 | 4874 | gfp_t mask = btrfs_alloc_write_mask(mapping); |
6d4572a9 | 4875 | size_t write_bytes = blocksize; |
39279cc3 | 4876 | int ret = 0; |
9703fefe CR |
4877 | u64 block_start; |
4878 | u64 block_end; | |
39279cc3 | 4879 | |
b03ebd99 NB |
4880 | if (IS_ALIGNED(offset, blocksize) && |
4881 | (!len || IS_ALIGNED(len, blocksize))) | |
39279cc3 | 4882 | goto out; |
9703fefe | 4883 | |
8b62f87b JB |
4884 | block_start = round_down(from, blocksize); |
4885 | block_end = block_start + blocksize - 1; | |
4886 | ||
217f42eb NB |
4887 | ret = btrfs_check_data_free_space(inode, &data_reserved, block_start, |
4888 | blocksize); | |
6d4572a9 | 4889 | if (ret < 0) { |
217f42eb | 4890 | if (btrfs_check_nocow_lock(inode, block_start, &write_bytes) > 0) { |
6d4572a9 QW |
4891 | /* For nocow case, no need to reserve data space */ |
4892 | only_release_metadata = true; | |
4893 | } else { | |
4894 | goto out; | |
4895 | } | |
4896 | } | |
d4135134 | 4897 | ret = btrfs_delalloc_reserve_metadata(inode, blocksize, blocksize, false); |
6d4572a9 QW |
4898 | if (ret < 0) { |
4899 | if (!only_release_metadata) | |
217f42eb NB |
4900 | btrfs_free_reserved_data_space(inode, data_reserved, |
4901 | block_start, blocksize); | |
6d4572a9 QW |
4902 | goto out; |
4903 | } | |
211c17f5 | 4904 | again: |
3b16a4e3 | 4905 | page = find_or_create_page(mapping, index, mask); |
5d5e103a | 4906 | if (!page) { |
217f42eb NB |
4907 | btrfs_delalloc_release_space(inode, data_reserved, block_start, |
4908 | blocksize, true); | |
4909 | btrfs_delalloc_release_extents(inode, blocksize); | |
ac6a2b36 | 4910 | ret = -ENOMEM; |
39279cc3 | 4911 | goto out; |
5d5e103a | 4912 | } |
32443de3 QW |
4913 | ret = set_page_extent_mapped(page); |
4914 | if (ret < 0) | |
4915 | goto out_unlock; | |
e6dcd2dc | 4916 | |
39279cc3 | 4917 | if (!PageUptodate(page)) { |
fb12489b | 4918 | ret = btrfs_read_folio(NULL, page_folio(page)); |
39279cc3 | 4919 | lock_page(page); |
211c17f5 CM |
4920 | if (page->mapping != mapping) { |
4921 | unlock_page(page); | |
09cbfeaf | 4922 | put_page(page); |
211c17f5 CM |
4923 | goto again; |
4924 | } | |
39279cc3 CM |
4925 | if (!PageUptodate(page)) { |
4926 | ret = -EIO; | |
89642229 | 4927 | goto out_unlock; |
39279cc3 CM |
4928 | } |
4929 | } | |
211c17f5 | 4930 | wait_on_page_writeback(page); |
e6dcd2dc | 4931 | |
570eb97b | 4932 | lock_extent(io_tree, block_start, block_end, &cached_state); |
e6dcd2dc | 4933 | |
217f42eb | 4934 | ordered = btrfs_lookup_ordered_extent(inode, block_start); |
e6dcd2dc | 4935 | if (ordered) { |
570eb97b | 4936 | unlock_extent(io_tree, block_start, block_end, &cached_state); |
e6dcd2dc | 4937 | unlock_page(page); |
09cbfeaf | 4938 | put_page(page); |
c0a43603 | 4939 | btrfs_start_ordered_extent(ordered, 1); |
e6dcd2dc CM |
4940 | btrfs_put_ordered_extent(ordered); |
4941 | goto again; | |
4942 | } | |
4943 | ||
217f42eb | 4944 | clear_extent_bit(&inode->io_tree, block_start, block_end, |
e182163d | 4945 | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, |
bd015294 | 4946 | &cached_state); |
5d5e103a | 4947 | |
217f42eb | 4948 | ret = btrfs_set_extent_delalloc(inode, block_start, block_end, 0, |
330a5827 | 4949 | &cached_state); |
9ed74f2d | 4950 | if (ret) { |
570eb97b | 4951 | unlock_extent(io_tree, block_start, block_end, &cached_state); |
9ed74f2d JB |
4952 | goto out_unlock; |
4953 | } | |
4954 | ||
9703fefe | 4955 | if (offset != blocksize) { |
2aaa6655 | 4956 | if (!len) |
9703fefe | 4957 | len = blocksize - offset; |
2aaa6655 | 4958 | if (front) |
d048b9c2 IW |
4959 | memzero_page(page, (block_start - page_offset(page)), |
4960 | offset); | |
2aaa6655 | 4961 | else |
d048b9c2 IW |
4962 | memzero_page(page, (block_start - page_offset(page)) + offset, |
4963 | len); | |
e6dcd2dc | 4964 | } |
e4f94347 QW |
4965 | btrfs_page_clear_checked(fs_info, page, block_start, |
4966 | block_end + 1 - block_start); | |
6c9ac8be | 4967 | btrfs_page_set_dirty(fs_info, page, block_start, block_end + 1 - block_start); |
570eb97b | 4968 | unlock_extent(io_tree, block_start, block_end, &cached_state); |
39279cc3 | 4969 | |
6d4572a9 | 4970 | if (only_release_metadata) |
217f42eb | 4971 | set_extent_bit(&inode->io_tree, block_start, block_end, |
291bbb1e | 4972 | EXTENT_NORESERVE, NULL, GFP_NOFS); |
6d4572a9 | 4973 | |
89642229 | 4974 | out_unlock: |
6d4572a9 QW |
4975 | if (ret) { |
4976 | if (only_release_metadata) | |
217f42eb | 4977 | btrfs_delalloc_release_metadata(inode, blocksize, true); |
6d4572a9 | 4978 | else |
217f42eb | 4979 | btrfs_delalloc_release_space(inode, data_reserved, |
6d4572a9 QW |
4980 | block_start, blocksize, true); |
4981 | } | |
217f42eb | 4982 | btrfs_delalloc_release_extents(inode, blocksize); |
39279cc3 | 4983 | unlock_page(page); |
09cbfeaf | 4984 | put_page(page); |
39279cc3 | 4985 | out: |
6d4572a9 | 4986 | if (only_release_metadata) |
217f42eb | 4987 | btrfs_check_nocow_unlock(inode); |
364ecf36 | 4988 | extent_changeset_free(data_reserved); |
39279cc3 CM |
4989 | return ret; |
4990 | } | |
4991 | ||
a4ba6cc0 | 4992 | static int maybe_insert_hole(struct btrfs_root *root, struct btrfs_inode *inode, |
16e7549f JB |
4993 | u64 offset, u64 len) |
4994 | { | |
a4ba6cc0 | 4995 | struct btrfs_fs_info *fs_info = root->fs_info; |
16e7549f | 4996 | struct btrfs_trans_handle *trans; |
5893dfb9 | 4997 | struct btrfs_drop_extents_args drop_args = { 0 }; |
16e7549f JB |
4998 | int ret; |
4999 | ||
5000 | /* | |
cceaa89f FM |
5001 | * If NO_HOLES is enabled, we don't need to do anything. |
5002 | * Later, up in the call chain, either btrfs_set_inode_last_sub_trans() | |
5003 | * or btrfs_update_inode() will be called, which guarantee that the next | |
5004 | * fsync will know this inode was changed and needs to be logged. | |
16e7549f | 5005 | */ |
cceaa89f | 5006 | if (btrfs_fs_incompat(fs_info, NO_HOLES)) |
16e7549f | 5007 | return 0; |
16e7549f JB |
5008 | |
5009 | /* | |
5010 | * 1 - for the one we're dropping | |
5011 | * 1 - for the one we're adding | |
5012 | * 1 - for updating the inode. | |
5013 | */ | |
5014 | trans = btrfs_start_transaction(root, 3); | |
5015 | if (IS_ERR(trans)) | |
5016 | return PTR_ERR(trans); | |
5017 | ||
5893dfb9 FM |
5018 | drop_args.start = offset; |
5019 | drop_args.end = offset + len; | |
5020 | drop_args.drop_cache = true; | |
5021 | ||
a4ba6cc0 | 5022 | ret = btrfs_drop_extents(trans, root, inode, &drop_args); |
16e7549f | 5023 | if (ret) { |
66642832 | 5024 | btrfs_abort_transaction(trans, ret); |
3a45bb20 | 5025 | btrfs_end_transaction(trans); |
16e7549f JB |
5026 | return ret; |
5027 | } | |
5028 | ||
d1f68ba0 | 5029 | ret = btrfs_insert_hole_extent(trans, root, btrfs_ino(inode), offset, len); |
2766ff61 | 5030 | if (ret) { |
66642832 | 5031 | btrfs_abort_transaction(trans, ret); |
2766ff61 | 5032 | } else { |
a4ba6cc0 NB |
5033 | btrfs_update_inode_bytes(inode, 0, drop_args.bytes_found); |
5034 | btrfs_update_inode(trans, root, inode); | |
2766ff61 | 5035 | } |
3a45bb20 | 5036 | btrfs_end_transaction(trans); |
16e7549f JB |
5037 | return ret; |
5038 | } | |
5039 | ||
695a0d0d JB |
5040 | /* |
5041 | * This function puts in dummy file extents for the area we're creating a hole | |
5042 | * for. So if we are truncating this file to a larger size we need to insert | |
5043 | * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for | |
5044 | * the range between oldsize and size | |
5045 | */ | |
b06359a3 | 5046 | int btrfs_cont_expand(struct btrfs_inode *inode, loff_t oldsize, loff_t size) |
39279cc3 | 5047 | { |
b06359a3 NB |
5048 | struct btrfs_root *root = inode->root; |
5049 | struct btrfs_fs_info *fs_info = root->fs_info; | |
5050 | struct extent_io_tree *io_tree = &inode->io_tree; | |
a22285a6 | 5051 | struct extent_map *em = NULL; |
2ac55d41 | 5052 | struct extent_state *cached_state = NULL; |
b06359a3 | 5053 | struct extent_map_tree *em_tree = &inode->extent_tree; |
0b246afa JM |
5054 | u64 hole_start = ALIGN(oldsize, fs_info->sectorsize); |
5055 | u64 block_end = ALIGN(size, fs_info->sectorsize); | |
9036c102 YZ |
5056 | u64 last_byte; |
5057 | u64 cur_offset; | |
5058 | u64 hole_size; | |
9ed74f2d | 5059 | int err = 0; |
39279cc3 | 5060 | |
a71754fc | 5061 | /* |
9703fefe CR |
5062 | * If our size started in the middle of a block we need to zero out the |
5063 | * rest of the block before we expand the i_size, otherwise we could | |
a71754fc JB |
5064 | * expose stale data. |
5065 | */ | |
b06359a3 | 5066 | err = btrfs_truncate_block(inode, oldsize, 0, 0); |
a71754fc JB |
5067 | if (err) |
5068 | return err; | |
5069 | ||
9036c102 YZ |
5070 | if (size <= hole_start) |
5071 | return 0; | |
5072 | ||
b06359a3 NB |
5073 | btrfs_lock_and_flush_ordered_range(inode, hole_start, block_end - 1, |
5074 | &cached_state); | |
9036c102 YZ |
5075 | cur_offset = hole_start; |
5076 | while (1) { | |
b06359a3 | 5077 | em = btrfs_get_extent(inode, NULL, 0, cur_offset, |
39b07b5d | 5078 | block_end - cur_offset); |
79787eaa JM |
5079 | if (IS_ERR(em)) { |
5080 | err = PTR_ERR(em); | |
f2767956 | 5081 | em = NULL; |
79787eaa JM |
5082 | break; |
5083 | } | |
9036c102 | 5084 | last_byte = min(extent_map_end(em), block_end); |
0b246afa | 5085 | last_byte = ALIGN(last_byte, fs_info->sectorsize); |
9ddc959e JB |
5086 | hole_size = last_byte - cur_offset; |
5087 | ||
8082510e | 5088 | if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) { |
5dc562c5 | 5089 | struct extent_map *hole_em; |
9ed74f2d | 5090 | |
b06359a3 NB |
5091 | err = maybe_insert_hole(root, inode, cur_offset, |
5092 | hole_size); | |
16e7549f | 5093 | if (err) |
3893e33b | 5094 | break; |
9ddc959e | 5095 | |
b06359a3 | 5096 | err = btrfs_inode_set_file_extent_range(inode, |
9ddc959e JB |
5097 | cur_offset, hole_size); |
5098 | if (err) | |
5099 | break; | |
5100 | ||
b06359a3 | 5101 | btrfs_drop_extent_cache(inode, cur_offset, |
5dc562c5 JB |
5102 | cur_offset + hole_size - 1, 0); |
5103 | hole_em = alloc_extent_map(); | |
5104 | if (!hole_em) { | |
23e3337f | 5105 | btrfs_set_inode_full_sync(inode); |
5dc562c5 JB |
5106 | goto next; |
5107 | } | |
5108 | hole_em->start = cur_offset; | |
5109 | hole_em->len = hole_size; | |
5110 | hole_em->orig_start = cur_offset; | |
8082510e | 5111 | |
5dc562c5 JB |
5112 | hole_em->block_start = EXTENT_MAP_HOLE; |
5113 | hole_em->block_len = 0; | |
b4939680 | 5114 | hole_em->orig_block_len = 0; |
cc95bef6 | 5115 | hole_em->ram_bytes = hole_size; |
5dc562c5 | 5116 | hole_em->compress_type = BTRFS_COMPRESS_NONE; |
0b246afa | 5117 | hole_em->generation = fs_info->generation; |
8082510e | 5118 | |
5dc562c5 JB |
5119 | while (1) { |
5120 | write_lock(&em_tree->lock); | |
09a2a8f9 | 5121 | err = add_extent_mapping(em_tree, hole_em, 1); |
5dc562c5 JB |
5122 | write_unlock(&em_tree->lock); |
5123 | if (err != -EEXIST) | |
5124 | break; | |
b06359a3 | 5125 | btrfs_drop_extent_cache(inode, cur_offset, |
5dc562c5 JB |
5126 | cur_offset + |
5127 | hole_size - 1, 0); | |
5128 | } | |
5129 | free_extent_map(hole_em); | |
9ddc959e | 5130 | } else { |
b06359a3 | 5131 | err = btrfs_inode_set_file_extent_range(inode, |
9ddc959e JB |
5132 | cur_offset, hole_size); |
5133 | if (err) | |
5134 | break; | |
9036c102 | 5135 | } |
16e7549f | 5136 | next: |
9036c102 | 5137 | free_extent_map(em); |
a22285a6 | 5138 | em = NULL; |
9036c102 | 5139 | cur_offset = last_byte; |
8082510e | 5140 | if (cur_offset >= block_end) |
9036c102 YZ |
5141 | break; |
5142 | } | |
a22285a6 | 5143 | free_extent_map(em); |
570eb97b | 5144 | unlock_extent(io_tree, hole_start, block_end - 1, &cached_state); |
9036c102 YZ |
5145 | return err; |
5146 | } | |
39279cc3 | 5147 | |
3972f260 | 5148 | static int btrfs_setsize(struct inode *inode, struct iattr *attr) |
8082510e | 5149 | { |
f4a2f4c5 MX |
5150 | struct btrfs_root *root = BTRFS_I(inode)->root; |
5151 | struct btrfs_trans_handle *trans; | |
a41ad394 | 5152 | loff_t oldsize = i_size_read(inode); |
3972f260 ES |
5153 | loff_t newsize = attr->ia_size; |
5154 | int mask = attr->ia_valid; | |
8082510e YZ |
5155 | int ret; |
5156 | ||
3972f260 ES |
5157 | /* |
5158 | * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a | |
5159 | * special case where we need to update the times despite not having | |
5160 | * these flags set. For all other operations the VFS set these flags | |
5161 | * explicitly if it wants a timestamp update. | |
5162 | */ | |
dff6efc3 CH |
5163 | if (newsize != oldsize) { |
5164 | inode_inc_iversion(inode); | |
c1867eb3 DS |
5165 | if (!(mask & (ATTR_CTIME | ATTR_MTIME))) { |
5166 | inode->i_mtime = current_time(inode); | |
5167 | inode->i_ctime = inode->i_mtime; | |
5168 | } | |
dff6efc3 | 5169 | } |
3972f260 | 5170 | |
a41ad394 | 5171 | if (newsize > oldsize) { |
9ea24bbe | 5172 | /* |
ea14b57f | 5173 | * Don't do an expanding truncate while snapshotting is ongoing. |
9ea24bbe FM |
5174 | * This is to ensure the snapshot captures a fully consistent |
5175 | * state of this file - if the snapshot captures this expanding | |
5176 | * truncation, it must capture all writes that happened before | |
5177 | * this truncation. | |
5178 | */ | |
dcc3eb96 | 5179 | btrfs_drew_write_lock(&root->snapshot_lock); |
b06359a3 | 5180 | ret = btrfs_cont_expand(BTRFS_I(inode), oldsize, newsize); |
9ea24bbe | 5181 | if (ret) { |
dcc3eb96 | 5182 | btrfs_drew_write_unlock(&root->snapshot_lock); |
8082510e | 5183 | return ret; |
9ea24bbe | 5184 | } |
8082510e | 5185 | |
f4a2f4c5 | 5186 | trans = btrfs_start_transaction(root, 1); |
9ea24bbe | 5187 | if (IS_ERR(trans)) { |
dcc3eb96 | 5188 | btrfs_drew_write_unlock(&root->snapshot_lock); |
f4a2f4c5 | 5189 | return PTR_ERR(trans); |
9ea24bbe | 5190 | } |
f4a2f4c5 MX |
5191 | |
5192 | i_size_write(inode, newsize); | |
76aea537 | 5193 | btrfs_inode_safe_disk_i_size_write(BTRFS_I(inode), 0); |
27772b68 | 5194 | pagecache_isize_extended(inode, oldsize, newsize); |
9a56fcd1 | 5195 | ret = btrfs_update_inode(trans, root, BTRFS_I(inode)); |
dcc3eb96 | 5196 | btrfs_drew_write_unlock(&root->snapshot_lock); |
3a45bb20 | 5197 | btrfs_end_transaction(trans); |
a41ad394 | 5198 | } else { |
24c0a722 NA |
5199 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
5200 | ||
5201 | if (btrfs_is_zoned(fs_info)) { | |
5202 | ret = btrfs_wait_ordered_range(inode, | |
5203 | ALIGN(newsize, fs_info->sectorsize), | |
5204 | (u64)-1); | |
5205 | if (ret) | |
5206 | return ret; | |
5207 | } | |
8082510e | 5208 | |
a41ad394 JB |
5209 | /* |
5210 | * We're truncating a file that used to have good data down to | |
1fd4033d NB |
5211 | * zero. Make sure any new writes to the file get on disk |
5212 | * on close. | |
a41ad394 JB |
5213 | */ |
5214 | if (newsize == 0) | |
1fd4033d | 5215 | set_bit(BTRFS_INODE_FLUSH_ON_CLOSE, |
72ac3c0d | 5216 | &BTRFS_I(inode)->runtime_flags); |
8082510e | 5217 | |
a41ad394 | 5218 | truncate_setsize(inode, newsize); |
2e60a51e | 5219 | |
2e60a51e | 5220 | inode_dio_wait(inode); |
2e60a51e | 5221 | |
213e8c55 | 5222 | ret = btrfs_truncate(inode, newsize == oldsize); |
7f4f6e0a JB |
5223 | if (ret && inode->i_nlink) { |
5224 | int err; | |
5225 | ||
5226 | /* | |
f7e9e8fc OS |
5227 | * Truncate failed, so fix up the in-memory size. We |
5228 | * adjusted disk_i_size down as we removed extents, so | |
5229 | * wait for disk_i_size to be stable and then update the | |
5230 | * in-memory size to match. | |
7f4f6e0a | 5231 | */ |
f7e9e8fc | 5232 | err = btrfs_wait_ordered_range(inode, 0, (u64)-1); |
7f4f6e0a | 5233 | if (err) |
f7e9e8fc OS |
5234 | return err; |
5235 | i_size_write(inode, BTRFS_I(inode)->disk_i_size); | |
7f4f6e0a | 5236 | } |
8082510e YZ |
5237 | } |
5238 | ||
a41ad394 | 5239 | return ret; |
8082510e YZ |
5240 | } |
5241 | ||
549c7297 CB |
5242 | static int btrfs_setattr(struct user_namespace *mnt_userns, struct dentry *dentry, |
5243 | struct iattr *attr) | |
9036c102 | 5244 | { |
2b0143b5 | 5245 | struct inode *inode = d_inode(dentry); |
b83cc969 | 5246 | struct btrfs_root *root = BTRFS_I(inode)->root; |
9036c102 | 5247 | int err; |
39279cc3 | 5248 | |
b83cc969 LZ |
5249 | if (btrfs_root_readonly(root)) |
5250 | return -EROFS; | |
5251 | ||
d4d09464 | 5252 | err = setattr_prepare(mnt_userns, dentry, attr); |
9036c102 YZ |
5253 | if (err) |
5254 | return err; | |
2bf5a725 | 5255 | |
5a3f23d5 | 5256 | if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) { |
3972f260 | 5257 | err = btrfs_setsize(inode, attr); |
8082510e YZ |
5258 | if (err) |
5259 | return err; | |
39279cc3 | 5260 | } |
9036c102 | 5261 | |
1025774c | 5262 | if (attr->ia_valid) { |
d4d09464 | 5263 | setattr_copy(mnt_userns, inode, attr); |
0c4d2d95 | 5264 | inode_inc_iversion(inode); |
22c44fe6 | 5265 | err = btrfs_dirty_inode(inode); |
1025774c | 5266 | |
22c44fe6 | 5267 | if (!err && attr->ia_valid & ATTR_MODE) |
d4d09464 | 5268 | err = posix_acl_chmod(mnt_userns, inode, inode->i_mode); |
1025774c | 5269 | } |
33268eaf | 5270 | |
39279cc3 CM |
5271 | return err; |
5272 | } | |
61295eb8 | 5273 | |
131e404a | 5274 | /* |
895586eb MWO |
5275 | * While truncating the inode pages during eviction, we get the VFS |
5276 | * calling btrfs_invalidate_folio() against each folio of the inode. This | |
5277 | * is slow because the calls to btrfs_invalidate_folio() result in a | |
570eb97b | 5278 | * huge amount of calls to lock_extent() and clear_extent_bit(), |
895586eb MWO |
5279 | * which keep merging and splitting extent_state structures over and over, |
5280 | * wasting lots of time. | |
131e404a | 5281 | * |
895586eb MWO |
5282 | * Therefore if the inode is being evicted, let btrfs_invalidate_folio() |
5283 | * skip all those expensive operations on a per folio basis and do only | |
5284 | * the ordered io finishing, while we release here the extent_map and | |
5285 | * extent_state structures, without the excessive merging and splitting. | |
131e404a FDBM |
5286 | */ |
5287 | static void evict_inode_truncate_pages(struct inode *inode) | |
5288 | { | |
5289 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; | |
5290 | struct extent_map_tree *map_tree = &BTRFS_I(inode)->extent_tree; | |
5291 | struct rb_node *node; | |
5292 | ||
5293 | ASSERT(inode->i_state & I_FREEING); | |
91b0abe3 | 5294 | truncate_inode_pages_final(&inode->i_data); |
131e404a FDBM |
5295 | |
5296 | write_lock(&map_tree->lock); | |
07e1ce09 | 5297 | while (!RB_EMPTY_ROOT(&map_tree->map.rb_root)) { |
131e404a FDBM |
5298 | struct extent_map *em; |
5299 | ||
07e1ce09 | 5300 | node = rb_first_cached(&map_tree->map); |
131e404a | 5301 | em = rb_entry(node, struct extent_map, rb_node); |
180589ef WS |
5302 | clear_bit(EXTENT_FLAG_PINNED, &em->flags); |
5303 | clear_bit(EXTENT_FLAG_LOGGING, &em->flags); | |
131e404a FDBM |
5304 | remove_extent_mapping(map_tree, em); |
5305 | free_extent_map(em); | |
7064dd5c FM |
5306 | if (need_resched()) { |
5307 | write_unlock(&map_tree->lock); | |
5308 | cond_resched(); | |
5309 | write_lock(&map_tree->lock); | |
5310 | } | |
131e404a FDBM |
5311 | } |
5312 | write_unlock(&map_tree->lock); | |
5313 | ||
6ca07097 FM |
5314 | /* |
5315 | * Keep looping until we have no more ranges in the io tree. | |
ba206a02 MWO |
5316 | * We can have ongoing bios started by readahead that have |
5317 | * their endio callback (extent_io.c:end_bio_extent_readpage) | |
9c6429d9 FM |
5318 | * still in progress (unlocked the pages in the bio but did not yet |
5319 | * unlocked the ranges in the io tree). Therefore this means some | |
6ca07097 FM |
5320 | * ranges can still be locked and eviction started because before |
5321 | * submitting those bios, which are executed by a separate task (work | |
5322 | * queue kthread), inode references (inode->i_count) were not taken | |
5323 | * (which would be dropped in the end io callback of each bio). | |
5324 | * Therefore here we effectively end up waiting for those bios and | |
5325 | * anyone else holding locked ranges without having bumped the inode's | |
5326 | * reference count - if we don't do it, when they access the inode's | |
5327 | * io_tree to unlock a range it may be too late, leading to an | |
5328 | * use-after-free issue. | |
5329 | */ | |
131e404a FDBM |
5330 | spin_lock(&io_tree->lock); |
5331 | while (!RB_EMPTY_ROOT(&io_tree->state)) { | |
5332 | struct extent_state *state; | |
5333 | struct extent_state *cached_state = NULL; | |
6ca07097 FM |
5334 | u64 start; |
5335 | u64 end; | |
421f0922 | 5336 | unsigned state_flags; |
131e404a FDBM |
5337 | |
5338 | node = rb_first(&io_tree->state); | |
5339 | state = rb_entry(node, struct extent_state, rb_node); | |
6ca07097 FM |
5340 | start = state->start; |
5341 | end = state->end; | |
421f0922 | 5342 | state_flags = state->state; |
131e404a FDBM |
5343 | spin_unlock(&io_tree->lock); |
5344 | ||
570eb97b | 5345 | lock_extent(io_tree, start, end, &cached_state); |
b9d0b389 QW |
5346 | |
5347 | /* | |
5348 | * If still has DELALLOC flag, the extent didn't reach disk, | |
5349 | * and its reserved space won't be freed by delayed_ref. | |
5350 | * So we need to free its reserved space here. | |
895586eb | 5351 | * (Refer to comment in btrfs_invalidate_folio, case 2) |
b9d0b389 QW |
5352 | * |
5353 | * Note, end is the bytenr of last byte, so we need + 1 here. | |
5354 | */ | |
421f0922 | 5355 | if (state_flags & EXTENT_DELALLOC) |
8b8a979f NB |
5356 | btrfs_qgroup_free_data(BTRFS_I(inode), NULL, start, |
5357 | end - start + 1); | |
b9d0b389 | 5358 | |
6ca07097 | 5359 | clear_extent_bit(io_tree, start, end, |
bd015294 | 5360 | EXTENT_CLEAR_ALL_BITS | EXTENT_DO_ACCOUNTING, |
e182163d | 5361 | &cached_state); |
131e404a | 5362 | |
7064dd5c | 5363 | cond_resched(); |
131e404a FDBM |
5364 | spin_lock(&io_tree->lock); |
5365 | } | |
5366 | spin_unlock(&io_tree->lock); | |
5367 | } | |
5368 | ||
4b9d7b59 | 5369 | static struct btrfs_trans_handle *evict_refill_and_join(struct btrfs_root *root, |
ad80cf50 | 5370 | struct btrfs_block_rsv *rsv) |
4b9d7b59 OS |
5371 | { |
5372 | struct btrfs_fs_info *fs_info = root->fs_info; | |
d3984c90 | 5373 | struct btrfs_trans_handle *trans; |
2bd36e7b | 5374 | u64 delayed_refs_extra = btrfs_calc_insert_metadata_size(fs_info, 1); |
d3984c90 | 5375 | int ret; |
4b9d7b59 | 5376 | |
d3984c90 JB |
5377 | /* |
5378 | * Eviction should be taking place at some place safe because of our | |
5379 | * delayed iputs. However the normal flushing code will run delayed | |
5380 | * iputs, so we cannot use FLUSH_ALL otherwise we'll deadlock. | |
5381 | * | |
5382 | * We reserve the delayed_refs_extra here again because we can't use | |
5383 | * btrfs_start_transaction(root, 0) for the same deadlocky reason as | |
5384 | * above. We reserve our extra bit here because we generate a ton of | |
5385 | * delayed refs activity by truncating. | |
5386 | * | |
ee6adbfd JB |
5387 | * BTRFS_RESERVE_FLUSH_EVICT will steal from the global_rsv if it can, |
5388 | * if we fail to make this reservation we can re-try without the | |
5389 | * delayed_refs_extra so we can make some forward progress. | |
d3984c90 | 5390 | */ |
9270501c | 5391 | ret = btrfs_block_rsv_refill(fs_info, rsv, rsv->size + delayed_refs_extra, |
d3984c90 JB |
5392 | BTRFS_RESERVE_FLUSH_EVICT); |
5393 | if (ret) { | |
9270501c | 5394 | ret = btrfs_block_rsv_refill(fs_info, rsv, rsv->size, |
ee6adbfd JB |
5395 | BTRFS_RESERVE_FLUSH_EVICT); |
5396 | if (ret) { | |
d3984c90 JB |
5397 | btrfs_warn(fs_info, |
5398 | "could not allocate space for delete; will truncate on mount"); | |
5399 | return ERR_PTR(-ENOSPC); | |
5400 | } | |
5401 | delayed_refs_extra = 0; | |
5402 | } | |
4b9d7b59 | 5403 | |
d3984c90 JB |
5404 | trans = btrfs_join_transaction(root); |
5405 | if (IS_ERR(trans)) | |
5406 | return trans; | |
5407 | ||
5408 | if (delayed_refs_extra) { | |
5409 | trans->block_rsv = &fs_info->trans_block_rsv; | |
5410 | trans->bytes_reserved = delayed_refs_extra; | |
5411 | btrfs_block_rsv_migrate(rsv, trans->block_rsv, | |
5412 | delayed_refs_extra, 1); | |
4b9d7b59 | 5413 | } |
d3984c90 | 5414 | return trans; |
4b9d7b59 OS |
5415 | } |
5416 | ||
bd555975 | 5417 | void btrfs_evict_inode(struct inode *inode) |
39279cc3 | 5418 | { |
0b246afa | 5419 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
39279cc3 CM |
5420 | struct btrfs_trans_handle *trans; |
5421 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
4b9d7b59 | 5422 | struct btrfs_block_rsv *rsv; |
39279cc3 CM |
5423 | int ret; |
5424 | ||
1abe9b8a | 5425 | trace_btrfs_inode_evict(inode); |
5426 | ||
3d48d981 | 5427 | if (!root) { |
14605409 | 5428 | fsverity_cleanup_inode(inode); |
e8f1bc14 | 5429 | clear_inode(inode); |
3d48d981 NB |
5430 | return; |
5431 | } | |
5432 | ||
131e404a FDBM |
5433 | evict_inode_truncate_pages(inode); |
5434 | ||
69e9c6c6 SB |
5435 | if (inode->i_nlink && |
5436 | ((btrfs_root_refs(&root->root_item) != 0 && | |
5437 | root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) || | |
70ddc553 | 5438 | btrfs_is_free_space_inode(BTRFS_I(inode)))) |
bd555975 AV |
5439 | goto no_delete; |
5440 | ||
27919067 | 5441 | if (is_bad_inode(inode)) |
39279cc3 | 5442 | goto no_delete; |
5f39d397 | 5443 | |
7ab7956e | 5444 | btrfs_free_io_failure_record(BTRFS_I(inode), 0, (u64)-1); |
f612496b | 5445 | |
7b40b695 | 5446 | if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) |
c71bf099 | 5447 | goto no_delete; |
c71bf099 | 5448 | |
76dda93c | 5449 | if (inode->i_nlink > 0) { |
69e9c6c6 SB |
5450 | BUG_ON(btrfs_root_refs(&root->root_item) != 0 && |
5451 | root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID); | |
76dda93c YZ |
5452 | goto no_delete; |
5453 | } | |
5454 | ||
2adc75d6 JB |
5455 | /* |
5456 | * This makes sure the inode item in tree is uptodate and the space for | |
5457 | * the inode update is released. | |
5458 | */ | |
aa79021f | 5459 | ret = btrfs_commit_inode_delayed_inode(BTRFS_I(inode)); |
27919067 | 5460 | if (ret) |
0e8c36a9 | 5461 | goto no_delete; |
0e8c36a9 | 5462 | |
2adc75d6 JB |
5463 | /* |
5464 | * This drops any pending insert or delete operations we have for this | |
5465 | * inode. We could have a delayed dir index deletion queued up, but | |
5466 | * we're removing the inode completely so that'll be taken care of in | |
5467 | * the truncate. | |
5468 | */ | |
5469 | btrfs_kill_delayed_inode_items(BTRFS_I(inode)); | |
5470 | ||
2ff7e61e | 5471 | rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP); |
27919067 | 5472 | if (!rsv) |
4289a667 | 5473 | goto no_delete; |
2bd36e7b | 5474 | rsv->size = btrfs_calc_metadata_size(fs_info, 1); |
710d5921 | 5475 | rsv->failfast = true; |
4289a667 | 5476 | |
6ef06d27 | 5477 | btrfs_i_size_write(BTRFS_I(inode), 0); |
5f39d397 | 5478 | |
8082510e | 5479 | while (1) { |
d9ac19c3 | 5480 | struct btrfs_truncate_control control = { |
71d18b53 | 5481 | .inode = BTRFS_I(inode), |
487e81d2 | 5482 | .ino = btrfs_ino(BTRFS_I(inode)), |
d9ac19c3 JB |
5483 | .new_size = 0, |
5484 | .min_type = 0, | |
5485 | }; | |
5486 | ||
ad80cf50 | 5487 | trans = evict_refill_and_join(root, rsv); |
27919067 OS |
5488 | if (IS_ERR(trans)) |
5489 | goto free_rsv; | |
7b128766 | 5490 | |
4289a667 JB |
5491 | trans->block_rsv = rsv; |
5492 | ||
71d18b53 | 5493 | ret = btrfs_truncate_inode_items(trans, root, &control); |
27919067 OS |
5494 | trans->block_rsv = &fs_info->trans_block_rsv; |
5495 | btrfs_end_transaction(trans); | |
5496 | btrfs_btree_balance_dirty(fs_info); | |
5497 | if (ret && ret != -ENOSPC && ret != -EAGAIN) | |
5498 | goto free_rsv; | |
5499 | else if (!ret) | |
8082510e | 5500 | break; |
8082510e | 5501 | } |
5f39d397 | 5502 | |
4ef31a45 | 5503 | /* |
27919067 OS |
5504 | * Errors here aren't a big deal, it just means we leave orphan items in |
5505 | * the tree. They will be cleaned up on the next mount. If the inode | |
5506 | * number gets reused, cleanup deletes the orphan item without doing | |
5507 | * anything, and unlink reuses the existing orphan item. | |
5508 | * | |
5509 | * If it turns out that we are dropping too many of these, we might want | |
5510 | * to add a mechanism for retrying these after a commit. | |
4ef31a45 | 5511 | */ |
ad80cf50 | 5512 | trans = evict_refill_and_join(root, rsv); |
27919067 OS |
5513 | if (!IS_ERR(trans)) { |
5514 | trans->block_rsv = rsv; | |
5515 | btrfs_orphan_del(trans, BTRFS_I(inode)); | |
5516 | trans->block_rsv = &fs_info->trans_block_rsv; | |
5517 | btrfs_end_transaction(trans); | |
5518 | } | |
54aa1f4d | 5519 | |
27919067 OS |
5520 | free_rsv: |
5521 | btrfs_free_block_rsv(fs_info, rsv); | |
39279cc3 | 5522 | no_delete: |
27919067 OS |
5523 | /* |
5524 | * If we didn't successfully delete, the orphan item will still be in | |
5525 | * the tree and we'll retry on the next mount. Again, we might also want | |
5526 | * to retry these periodically in the future. | |
5527 | */ | |
f48d1cf5 | 5528 | btrfs_remove_delayed_node(BTRFS_I(inode)); |
14605409 | 5529 | fsverity_cleanup_inode(inode); |
dbd5768f | 5530 | clear_inode(inode); |
39279cc3 CM |
5531 | } |
5532 | ||
5533 | /* | |
6bf9e4bd QW |
5534 | * Return the key found in the dir entry in the location pointer, fill @type |
5535 | * with BTRFS_FT_*, and return 0. | |
5536 | * | |
005d6712 SY |
5537 | * If no dir entries were found, returns -ENOENT. |
5538 | * If found a corrupted location in dir entry, returns -EUCLEAN. | |
39279cc3 CM |
5539 | */ |
5540 | static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry, | |
6bf9e4bd | 5541 | struct btrfs_key *location, u8 *type) |
39279cc3 CM |
5542 | { |
5543 | const char *name = dentry->d_name.name; | |
5544 | int namelen = dentry->d_name.len; | |
5545 | struct btrfs_dir_item *di; | |
5546 | struct btrfs_path *path; | |
5547 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
0d9f7f3e | 5548 | int ret = 0; |
39279cc3 CM |
5549 | |
5550 | path = btrfs_alloc_path(); | |
d8926bb3 MF |
5551 | if (!path) |
5552 | return -ENOMEM; | |
3954401f | 5553 | |
f85b7379 DS |
5554 | di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(BTRFS_I(dir)), |
5555 | name, namelen, 0); | |
3cf5068f LB |
5556 | if (IS_ERR_OR_NULL(di)) { |
5557 | ret = di ? PTR_ERR(di) : -ENOENT; | |
005d6712 SY |
5558 | goto out; |
5559 | } | |
d397712b | 5560 | |
5f39d397 | 5561 | btrfs_dir_item_key_to_cpu(path->nodes[0], di, location); |
56a0e706 LB |
5562 | if (location->type != BTRFS_INODE_ITEM_KEY && |
5563 | location->type != BTRFS_ROOT_ITEM_KEY) { | |
005d6712 | 5564 | ret = -EUCLEAN; |
56a0e706 LB |
5565 | btrfs_warn(root->fs_info, |
5566 | "%s gets something invalid in DIR_ITEM (name %s, directory ino %llu, location(%llu %u %llu))", | |
5567 | __func__, name, btrfs_ino(BTRFS_I(dir)), | |
5568 | location->objectid, location->type, location->offset); | |
56a0e706 | 5569 | } |
6bf9e4bd QW |
5570 | if (!ret) |
5571 | *type = btrfs_dir_type(path->nodes[0], di); | |
39279cc3 | 5572 | out: |
39279cc3 CM |
5573 | btrfs_free_path(path); |
5574 | return ret; | |
5575 | } | |
5576 | ||
5577 | /* | |
5578 | * when we hit a tree root in a directory, the btrfs part of the inode | |
5579 | * needs to be changed to reflect the root directory of the tree root. This | |
5580 | * is kind of like crossing a mount point. | |
5581 | */ | |
2ff7e61e | 5582 | static int fixup_tree_root_location(struct btrfs_fs_info *fs_info, |
4df27c4d YZ |
5583 | struct inode *dir, |
5584 | struct dentry *dentry, | |
5585 | struct btrfs_key *location, | |
5586 | struct btrfs_root **sub_root) | |
39279cc3 | 5587 | { |
4df27c4d YZ |
5588 | struct btrfs_path *path; |
5589 | struct btrfs_root *new_root; | |
5590 | struct btrfs_root_ref *ref; | |
5591 | struct extent_buffer *leaf; | |
1d4c08e0 | 5592 | struct btrfs_key key; |
4df27c4d YZ |
5593 | int ret; |
5594 | int err = 0; | |
39279cc3 | 5595 | |
4df27c4d YZ |
5596 | path = btrfs_alloc_path(); |
5597 | if (!path) { | |
5598 | err = -ENOMEM; | |
5599 | goto out; | |
5600 | } | |
39279cc3 | 5601 | |
4df27c4d | 5602 | err = -ENOENT; |
1d4c08e0 DS |
5603 | key.objectid = BTRFS_I(dir)->root->root_key.objectid; |
5604 | key.type = BTRFS_ROOT_REF_KEY; | |
5605 | key.offset = location->objectid; | |
5606 | ||
0b246afa | 5607 | ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0); |
4df27c4d YZ |
5608 | if (ret) { |
5609 | if (ret < 0) | |
5610 | err = ret; | |
5611 | goto out; | |
5612 | } | |
39279cc3 | 5613 | |
4df27c4d YZ |
5614 | leaf = path->nodes[0]; |
5615 | ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref); | |
4a0cc7ca | 5616 | if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(BTRFS_I(dir)) || |
4df27c4d YZ |
5617 | btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len) |
5618 | goto out; | |
39279cc3 | 5619 | |
4df27c4d YZ |
5620 | ret = memcmp_extent_buffer(leaf, dentry->d_name.name, |
5621 | (unsigned long)(ref + 1), | |
5622 | dentry->d_name.len); | |
5623 | if (ret) | |
5624 | goto out; | |
5625 | ||
b3b4aa74 | 5626 | btrfs_release_path(path); |
4df27c4d | 5627 | |
56e9357a | 5628 | new_root = btrfs_get_fs_root(fs_info, location->objectid, true); |
4df27c4d YZ |
5629 | if (IS_ERR(new_root)) { |
5630 | err = PTR_ERR(new_root); | |
5631 | goto out; | |
5632 | } | |
5633 | ||
4df27c4d YZ |
5634 | *sub_root = new_root; |
5635 | location->objectid = btrfs_root_dirid(&new_root->root_item); | |
5636 | location->type = BTRFS_INODE_ITEM_KEY; | |
5637 | location->offset = 0; | |
5638 | err = 0; | |
5639 | out: | |
5640 | btrfs_free_path(path); | |
5641 | return err; | |
39279cc3 CM |
5642 | } |
5643 | ||
5d4f98a2 YZ |
5644 | static void inode_tree_add(struct inode *inode) |
5645 | { | |
5646 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5647 | struct btrfs_inode *entry; | |
03e860bd NP |
5648 | struct rb_node **p; |
5649 | struct rb_node *parent; | |
cef21937 | 5650 | struct rb_node *new = &BTRFS_I(inode)->rb_node; |
4a0cc7ca | 5651 | u64 ino = btrfs_ino(BTRFS_I(inode)); |
5d4f98a2 | 5652 | |
1d3382cb | 5653 | if (inode_unhashed(inode)) |
76dda93c | 5654 | return; |
e1409cef | 5655 | parent = NULL; |
5d4f98a2 | 5656 | spin_lock(&root->inode_lock); |
e1409cef | 5657 | p = &root->inode_tree.rb_node; |
5d4f98a2 YZ |
5658 | while (*p) { |
5659 | parent = *p; | |
5660 | entry = rb_entry(parent, struct btrfs_inode, rb_node); | |
5661 | ||
37508515 | 5662 | if (ino < btrfs_ino(entry)) |
03e860bd | 5663 | p = &parent->rb_left; |
37508515 | 5664 | else if (ino > btrfs_ino(entry)) |
03e860bd | 5665 | p = &parent->rb_right; |
5d4f98a2 YZ |
5666 | else { |
5667 | WARN_ON(!(entry->vfs_inode.i_state & | |
a4ffdde6 | 5668 | (I_WILL_FREE | I_FREEING))); |
cef21937 | 5669 | rb_replace_node(parent, new, &root->inode_tree); |
03e860bd NP |
5670 | RB_CLEAR_NODE(parent); |
5671 | spin_unlock(&root->inode_lock); | |
cef21937 | 5672 | return; |
5d4f98a2 YZ |
5673 | } |
5674 | } | |
cef21937 FDBM |
5675 | rb_link_node(new, parent, p); |
5676 | rb_insert_color(new, &root->inode_tree); | |
5d4f98a2 YZ |
5677 | spin_unlock(&root->inode_lock); |
5678 | } | |
5679 | ||
b79b7249 | 5680 | static void inode_tree_del(struct btrfs_inode *inode) |
5d4f98a2 | 5681 | { |
b79b7249 | 5682 | struct btrfs_root *root = inode->root; |
76dda93c | 5683 | int empty = 0; |
5d4f98a2 | 5684 | |
03e860bd | 5685 | spin_lock(&root->inode_lock); |
b79b7249 NB |
5686 | if (!RB_EMPTY_NODE(&inode->rb_node)) { |
5687 | rb_erase(&inode->rb_node, &root->inode_tree); | |
5688 | RB_CLEAR_NODE(&inode->rb_node); | |
76dda93c | 5689 | empty = RB_EMPTY_ROOT(&root->inode_tree); |
5d4f98a2 | 5690 | } |
03e860bd | 5691 | spin_unlock(&root->inode_lock); |
76dda93c | 5692 | |
69e9c6c6 | 5693 | if (empty && btrfs_root_refs(&root->root_item) == 0) { |
76dda93c YZ |
5694 | spin_lock(&root->inode_lock); |
5695 | empty = RB_EMPTY_ROOT(&root->inode_tree); | |
5696 | spin_unlock(&root->inode_lock); | |
5697 | if (empty) | |
5698 | btrfs_add_dead_root(root); | |
5699 | } | |
5700 | } | |
5701 | ||
5d4f98a2 | 5702 | |
e02119d5 CM |
5703 | static int btrfs_init_locked_inode(struct inode *inode, void *p) |
5704 | { | |
5705 | struct btrfs_iget_args *args = p; | |
0202e83f DS |
5706 | |
5707 | inode->i_ino = args->ino; | |
5708 | BTRFS_I(inode)->location.objectid = args->ino; | |
5709 | BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY; | |
5710 | BTRFS_I(inode)->location.offset = 0; | |
5c8fd99f JB |
5711 | BTRFS_I(inode)->root = btrfs_grab_root(args->root); |
5712 | BUG_ON(args->root && !BTRFS_I(inode)->root); | |
9b9b8854 JB |
5713 | |
5714 | if (args->root && args->root == args->root->fs_info->tree_root && | |
5715 | args->ino != BTRFS_BTREE_INODE_OBJECTID) | |
5716 | set_bit(BTRFS_INODE_FREE_SPACE_INODE, | |
5717 | &BTRFS_I(inode)->runtime_flags); | |
39279cc3 CM |
5718 | return 0; |
5719 | } | |
5720 | ||
5721 | static int btrfs_find_actor(struct inode *inode, void *opaque) | |
5722 | { | |
5723 | struct btrfs_iget_args *args = opaque; | |
0202e83f DS |
5724 | |
5725 | return args->ino == BTRFS_I(inode)->location.objectid && | |
d397712b | 5726 | args->root == BTRFS_I(inode)->root; |
39279cc3 CM |
5727 | } |
5728 | ||
0202e83f | 5729 | static struct inode *btrfs_iget_locked(struct super_block *s, u64 ino, |
5d4f98a2 | 5730 | struct btrfs_root *root) |
39279cc3 CM |
5731 | { |
5732 | struct inode *inode; | |
5733 | struct btrfs_iget_args args; | |
0202e83f | 5734 | unsigned long hashval = btrfs_inode_hash(ino, root); |
778ba82b | 5735 | |
0202e83f | 5736 | args.ino = ino; |
39279cc3 CM |
5737 | args.root = root; |
5738 | ||
778ba82b | 5739 | inode = iget5_locked(s, hashval, btrfs_find_actor, |
39279cc3 CM |
5740 | btrfs_init_locked_inode, |
5741 | (void *)&args); | |
5742 | return inode; | |
5743 | } | |
5744 | ||
4c66e0d4 | 5745 | /* |
0202e83f | 5746 | * Get an inode object given its inode number and corresponding root. |
4c66e0d4 DS |
5747 | * Path can be preallocated to prevent recursing back to iget through |
5748 | * allocator. NULL is also valid but may require an additional allocation | |
5749 | * later. | |
1a54ef8c | 5750 | */ |
0202e83f | 5751 | struct inode *btrfs_iget_path(struct super_block *s, u64 ino, |
4c66e0d4 | 5752 | struct btrfs_root *root, struct btrfs_path *path) |
1a54ef8c BR |
5753 | { |
5754 | struct inode *inode; | |
5755 | ||
0202e83f | 5756 | inode = btrfs_iget_locked(s, ino, root); |
1a54ef8c | 5757 | if (!inode) |
5d4f98a2 | 5758 | return ERR_PTR(-ENOMEM); |
1a54ef8c BR |
5759 | |
5760 | if (inode->i_state & I_NEW) { | |
67710892 FM |
5761 | int ret; |
5762 | ||
4222ea71 | 5763 | ret = btrfs_read_locked_inode(inode, path); |
9bc2ceff | 5764 | if (!ret) { |
1748f843 MF |
5765 | inode_tree_add(inode); |
5766 | unlock_new_inode(inode); | |
1748f843 | 5767 | } else { |
f5b3a417 AV |
5768 | iget_failed(inode); |
5769 | /* | |
5770 | * ret > 0 can come from btrfs_search_slot called by | |
5771 | * btrfs_read_locked_inode, this means the inode item | |
5772 | * was not found. | |
5773 | */ | |
5774 | if (ret > 0) | |
5775 | ret = -ENOENT; | |
5776 | inode = ERR_PTR(ret); | |
1748f843 MF |
5777 | } |
5778 | } | |
5779 | ||
1a54ef8c BR |
5780 | return inode; |
5781 | } | |
5782 | ||
0202e83f | 5783 | struct inode *btrfs_iget(struct super_block *s, u64 ino, struct btrfs_root *root) |
4222ea71 | 5784 | { |
0202e83f | 5785 | return btrfs_iget_path(s, ino, root, NULL); |
4222ea71 FM |
5786 | } |
5787 | ||
4df27c4d YZ |
5788 | static struct inode *new_simple_dir(struct super_block *s, |
5789 | struct btrfs_key *key, | |
5790 | struct btrfs_root *root) | |
5791 | { | |
5792 | struct inode *inode = new_inode(s); | |
5793 | ||
5794 | if (!inode) | |
5795 | return ERR_PTR(-ENOMEM); | |
5796 | ||
5c8fd99f | 5797 | BTRFS_I(inode)->root = btrfs_grab_root(root); |
4df27c4d | 5798 | memcpy(&BTRFS_I(inode)->location, key, sizeof(*key)); |
72ac3c0d | 5799 | set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags); |
4df27c4d YZ |
5800 | |
5801 | inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID; | |
6bb6b514 OS |
5802 | /* |
5803 | * We only need lookup, the rest is read-only and there's no inode | |
5804 | * associated with the dentry | |
5805 | */ | |
5806 | inode->i_op = &simple_dir_inode_operations; | |
1fdf4194 | 5807 | inode->i_opflags &= ~IOP_XATTR; |
4df27c4d YZ |
5808 | inode->i_fop = &simple_dir_operations; |
5809 | inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO; | |
c2050a45 | 5810 | inode->i_mtime = current_time(inode); |
9cc97d64 | 5811 | inode->i_atime = inode->i_mtime; |
5812 | inode->i_ctime = inode->i_mtime; | |
d3c6be6f | 5813 | BTRFS_I(inode)->i_otime = inode->i_mtime; |
4df27c4d YZ |
5814 | |
5815 | return inode; | |
5816 | } | |
5817 | ||
a55e65b8 DS |
5818 | static_assert(BTRFS_FT_UNKNOWN == FT_UNKNOWN); |
5819 | static_assert(BTRFS_FT_REG_FILE == FT_REG_FILE); | |
5820 | static_assert(BTRFS_FT_DIR == FT_DIR); | |
5821 | static_assert(BTRFS_FT_CHRDEV == FT_CHRDEV); | |
5822 | static_assert(BTRFS_FT_BLKDEV == FT_BLKDEV); | |
5823 | static_assert(BTRFS_FT_FIFO == FT_FIFO); | |
5824 | static_assert(BTRFS_FT_SOCK == FT_SOCK); | |
5825 | static_assert(BTRFS_FT_SYMLINK == FT_SYMLINK); | |
5826 | ||
6bf9e4bd QW |
5827 | static inline u8 btrfs_inode_type(struct inode *inode) |
5828 | { | |
6bf9e4bd QW |
5829 | return fs_umode_to_ftype(inode->i_mode); |
5830 | } | |
5831 | ||
3de4586c | 5832 | struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry) |
39279cc3 | 5833 | { |
0b246afa | 5834 | struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb); |
d397712b | 5835 | struct inode *inode; |
4df27c4d | 5836 | struct btrfs_root *root = BTRFS_I(dir)->root; |
39279cc3 CM |
5837 | struct btrfs_root *sub_root = root; |
5838 | struct btrfs_key location; | |
6bf9e4bd | 5839 | u8 di_type = 0; |
b4aff1f8 | 5840 | int ret = 0; |
39279cc3 CM |
5841 | |
5842 | if (dentry->d_name.len > BTRFS_NAME_LEN) | |
5843 | return ERR_PTR(-ENAMETOOLONG); | |
5f39d397 | 5844 | |
6bf9e4bd | 5845 | ret = btrfs_inode_by_name(dir, dentry, &location, &di_type); |
39279cc3 CM |
5846 | if (ret < 0) |
5847 | return ERR_PTR(ret); | |
5f39d397 | 5848 | |
4df27c4d | 5849 | if (location.type == BTRFS_INODE_ITEM_KEY) { |
0202e83f | 5850 | inode = btrfs_iget(dir->i_sb, location.objectid, root); |
6bf9e4bd QW |
5851 | if (IS_ERR(inode)) |
5852 | return inode; | |
5853 | ||
5854 | /* Do extra check against inode mode with di_type */ | |
5855 | if (btrfs_inode_type(inode) != di_type) { | |
5856 | btrfs_crit(fs_info, | |
5857 | "inode mode mismatch with dir: inode mode=0%o btrfs type=%u dir type=%u", | |
5858 | inode->i_mode, btrfs_inode_type(inode), | |
5859 | di_type); | |
5860 | iput(inode); | |
5861 | return ERR_PTR(-EUCLEAN); | |
5862 | } | |
4df27c4d YZ |
5863 | return inode; |
5864 | } | |
5865 | ||
2ff7e61e | 5866 | ret = fixup_tree_root_location(fs_info, dir, dentry, |
4df27c4d YZ |
5867 | &location, &sub_root); |
5868 | if (ret < 0) { | |
5869 | if (ret != -ENOENT) | |
5870 | inode = ERR_PTR(ret); | |
5871 | else | |
fc8b235f | 5872 | inode = new_simple_dir(dir->i_sb, &location, root); |
4df27c4d | 5873 | } else { |
0202e83f | 5874 | inode = btrfs_iget(dir->i_sb, location.objectid, sub_root); |
00246528 | 5875 | btrfs_put_root(sub_root); |
76dda93c | 5876 | |
fc8b235f NB |
5877 | if (IS_ERR(inode)) |
5878 | return inode; | |
5879 | ||
0b246afa | 5880 | down_read(&fs_info->cleanup_work_sem); |
bc98a42c | 5881 | if (!sb_rdonly(inode->i_sb)) |
66b4ffd1 | 5882 | ret = btrfs_orphan_cleanup(sub_root); |
0b246afa | 5883 | up_read(&fs_info->cleanup_work_sem); |
01cd3367 JB |
5884 | if (ret) { |
5885 | iput(inode); | |
66b4ffd1 | 5886 | inode = ERR_PTR(ret); |
01cd3367 | 5887 | } |
c71bf099 YZ |
5888 | } |
5889 | ||
3de4586c CM |
5890 | return inode; |
5891 | } | |
5892 | ||
fe15ce44 | 5893 | static int btrfs_dentry_delete(const struct dentry *dentry) |
76dda93c YZ |
5894 | { |
5895 | struct btrfs_root *root; | |
2b0143b5 | 5896 | struct inode *inode = d_inode(dentry); |
76dda93c | 5897 | |
848cce0d | 5898 | if (!inode && !IS_ROOT(dentry)) |
2b0143b5 | 5899 | inode = d_inode(dentry->d_parent); |
76dda93c | 5900 | |
848cce0d LZ |
5901 | if (inode) { |
5902 | root = BTRFS_I(inode)->root; | |
efefb143 YZ |
5903 | if (btrfs_root_refs(&root->root_item) == 0) |
5904 | return 1; | |
848cce0d | 5905 | |
4a0cc7ca | 5906 | if (btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) |
848cce0d | 5907 | return 1; |
efefb143 | 5908 | } |
76dda93c YZ |
5909 | return 0; |
5910 | } | |
5911 | ||
3de4586c | 5912 | static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry, |
00cd8dd3 | 5913 | unsigned int flags) |
3de4586c | 5914 | { |
3837d208 | 5915 | struct inode *inode = btrfs_lookup_dentry(dir, dentry); |
5662344b | 5916 | |
3837d208 AV |
5917 | if (inode == ERR_PTR(-ENOENT)) |
5918 | inode = NULL; | |
41d28bca | 5919 | return d_splice_alias(inode, dentry); |
39279cc3 CM |
5920 | } |
5921 | ||
23b5ec74 JB |
5922 | /* |
5923 | * All this infrastructure exists because dir_emit can fault, and we are holding | |
5924 | * the tree lock when doing readdir. For now just allocate a buffer and copy | |
5925 | * our information into that, and then dir_emit from the buffer. This is | |
5926 | * similar to what NFS does, only we don't keep the buffer around in pagecache | |
5927 | * because I'm afraid I'll mess that up. Long term we need to make filldir do | |
5928 | * copy_to_user_inatomic so we don't have to worry about page faulting under the | |
5929 | * tree lock. | |
5930 | */ | |
5931 | static int btrfs_opendir(struct inode *inode, struct file *file) | |
5932 | { | |
5933 | struct btrfs_file_private *private; | |
5934 | ||
5935 | private = kzalloc(sizeof(struct btrfs_file_private), GFP_KERNEL); | |
5936 | if (!private) | |
5937 | return -ENOMEM; | |
5938 | private->filldir_buf = kzalloc(PAGE_SIZE, GFP_KERNEL); | |
5939 | if (!private->filldir_buf) { | |
5940 | kfree(private); | |
5941 | return -ENOMEM; | |
5942 | } | |
5943 | file->private_data = private; | |
5944 | return 0; | |
5945 | } | |
5946 | ||
5947 | struct dir_entry { | |
5948 | u64 ino; | |
5949 | u64 offset; | |
5950 | unsigned type; | |
5951 | int name_len; | |
5952 | }; | |
5953 | ||
5954 | static int btrfs_filldir(void *addr, int entries, struct dir_context *ctx) | |
5955 | { | |
5956 | while (entries--) { | |
5957 | struct dir_entry *entry = addr; | |
5958 | char *name = (char *)(entry + 1); | |
5959 | ||
92d32170 DS |
5960 | ctx->pos = get_unaligned(&entry->offset); |
5961 | if (!dir_emit(ctx, name, get_unaligned(&entry->name_len), | |
5962 | get_unaligned(&entry->ino), | |
5963 | get_unaligned(&entry->type))) | |
23b5ec74 | 5964 | return 1; |
92d32170 DS |
5965 | addr += sizeof(struct dir_entry) + |
5966 | get_unaligned(&entry->name_len); | |
23b5ec74 JB |
5967 | ctx->pos++; |
5968 | } | |
5969 | return 0; | |
5970 | } | |
5971 | ||
9cdda8d3 | 5972 | static int btrfs_real_readdir(struct file *file, struct dir_context *ctx) |
39279cc3 | 5973 | { |
9cdda8d3 | 5974 | struct inode *inode = file_inode(file); |
39279cc3 | 5975 | struct btrfs_root *root = BTRFS_I(inode)->root; |
23b5ec74 | 5976 | struct btrfs_file_private *private = file->private_data; |
39279cc3 CM |
5977 | struct btrfs_dir_item *di; |
5978 | struct btrfs_key key; | |
5f39d397 | 5979 | struct btrfs_key found_key; |
39279cc3 | 5980 | struct btrfs_path *path; |
23b5ec74 | 5981 | void *addr; |
16cdcec7 MX |
5982 | struct list_head ins_list; |
5983 | struct list_head del_list; | |
39279cc3 | 5984 | int ret; |
5f39d397 CM |
5985 | char *name_ptr; |
5986 | int name_len; | |
23b5ec74 JB |
5987 | int entries = 0; |
5988 | int total_len = 0; | |
02dbfc99 | 5989 | bool put = false; |
c2951f32 | 5990 | struct btrfs_key location; |
5f39d397 | 5991 | |
9cdda8d3 AV |
5992 | if (!dir_emit_dots(file, ctx)) |
5993 | return 0; | |
5994 | ||
49593bfa | 5995 | path = btrfs_alloc_path(); |
16cdcec7 MX |
5996 | if (!path) |
5997 | return -ENOMEM; | |
ff5714cc | 5998 | |
23b5ec74 | 5999 | addr = private->filldir_buf; |
e4058b54 | 6000 | path->reada = READA_FORWARD; |
49593bfa | 6001 | |
c2951f32 JM |
6002 | INIT_LIST_HEAD(&ins_list); |
6003 | INIT_LIST_HEAD(&del_list); | |
6004 | put = btrfs_readdir_get_delayed_items(inode, &ins_list, &del_list); | |
16cdcec7 | 6005 | |
23b5ec74 | 6006 | again: |
c2951f32 | 6007 | key.type = BTRFS_DIR_INDEX_KEY; |
9cdda8d3 | 6008 | key.offset = ctx->pos; |
4a0cc7ca | 6009 | key.objectid = btrfs_ino(BTRFS_I(inode)); |
5f39d397 | 6010 | |
a8ce68fd | 6011 | btrfs_for_each_slot(root, &key, &found_key, path, ret) { |
23b5ec74 | 6012 | struct dir_entry *entry; |
a8ce68fd | 6013 | struct extent_buffer *leaf = path->nodes[0]; |
5f39d397 CM |
6014 | |
6015 | if (found_key.objectid != key.objectid) | |
39279cc3 | 6016 | break; |
c2951f32 | 6017 | if (found_key.type != BTRFS_DIR_INDEX_KEY) |
39279cc3 | 6018 | break; |
9cdda8d3 | 6019 | if (found_key.offset < ctx->pos) |
a8ce68fd | 6020 | continue; |
c2951f32 | 6021 | if (btrfs_should_delete_dir_index(&del_list, found_key.offset)) |
a8ce68fd GN |
6022 | continue; |
6023 | di = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dir_item); | |
c2951f32 | 6024 | name_len = btrfs_dir_name_len(leaf, di); |
23b5ec74 JB |
6025 | if ((total_len + sizeof(struct dir_entry) + name_len) >= |
6026 | PAGE_SIZE) { | |
6027 | btrfs_release_path(path); | |
6028 | ret = btrfs_filldir(private->filldir_buf, entries, ctx); | |
6029 | if (ret) | |
6030 | goto nopos; | |
6031 | addr = private->filldir_buf; | |
6032 | entries = 0; | |
6033 | total_len = 0; | |
6034 | goto again; | |
c2951f32 | 6035 | } |
23b5ec74 JB |
6036 | |
6037 | entry = addr; | |
92d32170 | 6038 | put_unaligned(name_len, &entry->name_len); |
23b5ec74 | 6039 | name_ptr = (char *)(entry + 1); |
c2951f32 JM |
6040 | read_extent_buffer(leaf, name_ptr, (unsigned long)(di + 1), |
6041 | name_len); | |
7d157c3d | 6042 | put_unaligned(fs_ftype_to_dtype(btrfs_dir_type(leaf, di)), |
92d32170 | 6043 | &entry->type); |
c2951f32 | 6044 | btrfs_dir_item_key_to_cpu(leaf, di, &location); |
92d32170 DS |
6045 | put_unaligned(location.objectid, &entry->ino); |
6046 | put_unaligned(found_key.offset, &entry->offset); | |
23b5ec74 JB |
6047 | entries++; |
6048 | addr += sizeof(struct dir_entry) + name_len; | |
6049 | total_len += sizeof(struct dir_entry) + name_len; | |
39279cc3 | 6050 | } |
a8ce68fd GN |
6051 | /* Catch error encountered during iteration */ |
6052 | if (ret < 0) | |
6053 | goto err; | |
6054 | ||
23b5ec74 JB |
6055 | btrfs_release_path(path); |
6056 | ||
6057 | ret = btrfs_filldir(private->filldir_buf, entries, ctx); | |
6058 | if (ret) | |
6059 | goto nopos; | |
49593bfa | 6060 | |
d2fbb2b5 | 6061 | ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list); |
c2951f32 | 6062 | if (ret) |
bc4ef759 DS |
6063 | goto nopos; |
6064 | ||
db62efbb ZB |
6065 | /* |
6066 | * Stop new entries from being returned after we return the last | |
6067 | * entry. | |
6068 | * | |
6069 | * New directory entries are assigned a strictly increasing | |
6070 | * offset. This means that new entries created during readdir | |
6071 | * are *guaranteed* to be seen in the future by that readdir. | |
6072 | * This has broken buggy programs which operate on names as | |
6073 | * they're returned by readdir. Until we re-use freed offsets | |
6074 | * we have this hack to stop new entries from being returned | |
6075 | * under the assumption that they'll never reach this huge | |
6076 | * offset. | |
6077 | * | |
6078 | * This is being careful not to overflow 32bit loff_t unless the | |
6079 | * last entry requires it because doing so has broken 32bit apps | |
6080 | * in the past. | |
6081 | */ | |
c2951f32 JM |
6082 | if (ctx->pos >= INT_MAX) |
6083 | ctx->pos = LLONG_MAX; | |
6084 | else | |
6085 | ctx->pos = INT_MAX; | |
39279cc3 CM |
6086 | nopos: |
6087 | ret = 0; | |
6088 | err: | |
02dbfc99 OS |
6089 | if (put) |
6090 | btrfs_readdir_put_delayed_items(inode, &ins_list, &del_list); | |
39279cc3 | 6091 | btrfs_free_path(path); |
39279cc3 CM |
6092 | return ret; |
6093 | } | |
6094 | ||
39279cc3 | 6095 | /* |
54aa1f4d | 6096 | * This is somewhat expensive, updating the tree every time the |
39279cc3 CM |
6097 | * inode changes. But, it is most likely to find the inode in cache. |
6098 | * FIXME, needs more benchmarking...there are no reasons other than performance | |
6099 | * to keep or drop this code. | |
6100 | */ | |
48a3b636 | 6101 | static int btrfs_dirty_inode(struct inode *inode) |
39279cc3 | 6102 | { |
2ff7e61e | 6103 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
39279cc3 CM |
6104 | struct btrfs_root *root = BTRFS_I(inode)->root; |
6105 | struct btrfs_trans_handle *trans; | |
8929ecfa YZ |
6106 | int ret; |
6107 | ||
72ac3c0d | 6108 | if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags)) |
22c44fe6 | 6109 | return 0; |
39279cc3 | 6110 | |
7a7eaa40 | 6111 | trans = btrfs_join_transaction(root); |
22c44fe6 JB |
6112 | if (IS_ERR(trans)) |
6113 | return PTR_ERR(trans); | |
8929ecfa | 6114 | |
9a56fcd1 | 6115 | ret = btrfs_update_inode(trans, root, BTRFS_I(inode)); |
4d14c5cd | 6116 | if (ret && (ret == -ENOSPC || ret == -EDQUOT)) { |
94b60442 | 6117 | /* whoops, lets try again with the full transaction */ |
3a45bb20 | 6118 | btrfs_end_transaction(trans); |
94b60442 | 6119 | trans = btrfs_start_transaction(root, 1); |
22c44fe6 JB |
6120 | if (IS_ERR(trans)) |
6121 | return PTR_ERR(trans); | |
8929ecfa | 6122 | |
9a56fcd1 | 6123 | ret = btrfs_update_inode(trans, root, BTRFS_I(inode)); |
94b60442 | 6124 | } |
3a45bb20 | 6125 | btrfs_end_transaction(trans); |
16cdcec7 | 6126 | if (BTRFS_I(inode)->delayed_node) |
2ff7e61e | 6127 | btrfs_balance_delayed_items(fs_info); |
22c44fe6 JB |
6128 | |
6129 | return ret; | |
6130 | } | |
6131 | ||
6132 | /* | |
6133 | * This is a copy of file_update_time. We need this so we can return error on | |
6134 | * ENOSPC for updating the inode in the case of file write and mmap writes. | |
6135 | */ | |
95582b00 | 6136 | static int btrfs_update_time(struct inode *inode, struct timespec64 *now, |
e41f941a | 6137 | int flags) |
22c44fe6 | 6138 | { |
2bc55652 | 6139 | struct btrfs_root *root = BTRFS_I(inode)->root; |
3a8c7231 | 6140 | bool dirty = flags & ~S_VERSION; |
2bc55652 AB |
6141 | |
6142 | if (btrfs_root_readonly(root)) | |
6143 | return -EROFS; | |
6144 | ||
e41f941a | 6145 | if (flags & S_VERSION) |
3a8c7231 | 6146 | dirty |= inode_maybe_inc_iversion(inode, dirty); |
e41f941a JB |
6147 | if (flags & S_CTIME) |
6148 | inode->i_ctime = *now; | |
6149 | if (flags & S_MTIME) | |
6150 | inode->i_mtime = *now; | |
6151 | if (flags & S_ATIME) | |
6152 | inode->i_atime = *now; | |
3a8c7231 | 6153 | return dirty ? btrfs_dirty_inode(inode) : 0; |
39279cc3 CM |
6154 | } |
6155 | ||
d352ac68 CM |
6156 | /* |
6157 | * find the highest existing sequence number in a directory | |
6158 | * and then set the in-memory index_cnt variable to reflect | |
6159 | * free sequence numbers | |
6160 | */ | |
4c570655 | 6161 | static int btrfs_set_inode_index_count(struct btrfs_inode *inode) |
aec7477b | 6162 | { |
4c570655 | 6163 | struct btrfs_root *root = inode->root; |
aec7477b JB |
6164 | struct btrfs_key key, found_key; |
6165 | struct btrfs_path *path; | |
6166 | struct extent_buffer *leaf; | |
6167 | int ret; | |
6168 | ||
4c570655 | 6169 | key.objectid = btrfs_ino(inode); |
962a298f | 6170 | key.type = BTRFS_DIR_INDEX_KEY; |
aec7477b JB |
6171 | key.offset = (u64)-1; |
6172 | ||
6173 | path = btrfs_alloc_path(); | |
6174 | if (!path) | |
6175 | return -ENOMEM; | |
6176 | ||
6177 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
6178 | if (ret < 0) | |
6179 | goto out; | |
6180 | /* FIXME: we should be able to handle this */ | |
6181 | if (ret == 0) | |
6182 | goto out; | |
6183 | ret = 0; | |
6184 | ||
aec7477b | 6185 | if (path->slots[0] == 0) { |
528ee697 | 6186 | inode->index_cnt = BTRFS_DIR_START_INDEX; |
aec7477b JB |
6187 | goto out; |
6188 | } | |
6189 | ||
6190 | path->slots[0]--; | |
6191 | ||
6192 | leaf = path->nodes[0]; | |
6193 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
6194 | ||
4c570655 | 6195 | if (found_key.objectid != btrfs_ino(inode) || |
962a298f | 6196 | found_key.type != BTRFS_DIR_INDEX_KEY) { |
528ee697 | 6197 | inode->index_cnt = BTRFS_DIR_START_INDEX; |
aec7477b JB |
6198 | goto out; |
6199 | } | |
6200 | ||
4c570655 | 6201 | inode->index_cnt = found_key.offset + 1; |
aec7477b JB |
6202 | out: |
6203 | btrfs_free_path(path); | |
6204 | return ret; | |
6205 | } | |
6206 | ||
d352ac68 CM |
6207 | /* |
6208 | * helper to find a free sequence number in a given directory. This current | |
6209 | * code is very simple, later versions will do smarter things in the btree | |
6210 | */ | |
877574e2 | 6211 | int btrfs_set_inode_index(struct btrfs_inode *dir, u64 *index) |
aec7477b JB |
6212 | { |
6213 | int ret = 0; | |
6214 | ||
877574e2 NB |
6215 | if (dir->index_cnt == (u64)-1) { |
6216 | ret = btrfs_inode_delayed_dir_index_count(dir); | |
16cdcec7 MX |
6217 | if (ret) { |
6218 | ret = btrfs_set_inode_index_count(dir); | |
6219 | if (ret) | |
6220 | return ret; | |
6221 | } | |
aec7477b JB |
6222 | } |
6223 | ||
877574e2 NB |
6224 | *index = dir->index_cnt; |
6225 | dir->index_cnt++; | |
aec7477b JB |
6226 | |
6227 | return ret; | |
6228 | } | |
6229 | ||
b0d5d10f CM |
6230 | static int btrfs_insert_inode_locked(struct inode *inode) |
6231 | { | |
6232 | struct btrfs_iget_args args; | |
0202e83f DS |
6233 | |
6234 | args.ino = BTRFS_I(inode)->location.objectid; | |
b0d5d10f CM |
6235 | args.root = BTRFS_I(inode)->root; |
6236 | ||
6237 | return insert_inode_locked4(inode, | |
6238 | btrfs_inode_hash(inode->i_ino, BTRFS_I(inode)->root), | |
6239 | btrfs_find_actor, &args); | |
6240 | } | |
6241 | ||
3538d68d OS |
6242 | int btrfs_new_inode_prepare(struct btrfs_new_inode_args *args, |
6243 | unsigned int *trans_num_items) | |
6244 | { | |
6245 | struct inode *dir = args->dir; | |
6246 | struct inode *inode = args->inode; | |
6247 | int ret; | |
6248 | ||
6249 | ret = posix_acl_create(dir, &inode->i_mode, &args->default_acl, &args->acl); | |
6250 | if (ret) | |
6251 | return ret; | |
6252 | ||
6253 | /* 1 to add inode item */ | |
6254 | *trans_num_items = 1; | |
6255 | /* 1 to add compression property */ | |
6256 | if (BTRFS_I(dir)->prop_compress) | |
6257 | (*trans_num_items)++; | |
6258 | /* 1 to add default ACL xattr */ | |
6259 | if (args->default_acl) | |
6260 | (*trans_num_items)++; | |
6261 | /* 1 to add access ACL xattr */ | |
6262 | if (args->acl) | |
6263 | (*trans_num_items)++; | |
6264 | #ifdef CONFIG_SECURITY | |
6265 | /* 1 to add LSM xattr */ | |
6266 | if (dir->i_security) | |
6267 | (*trans_num_items)++; | |
6268 | #endif | |
6269 | if (args->orphan) { | |
6270 | /* 1 to add orphan item */ | |
6271 | (*trans_num_items)++; | |
6272 | } else { | |
6273 | /* | |
3538d68d OS |
6274 | * 1 to add dir item |
6275 | * 1 to add dir index | |
6276 | * 1 to update parent inode item | |
97bdf1a9 FM |
6277 | * |
6278 | * No need for 1 unit for the inode ref item because it is | |
6279 | * inserted in a batch together with the inode item at | |
6280 | * btrfs_create_new_inode(). | |
3538d68d | 6281 | */ |
97bdf1a9 | 6282 | *trans_num_items += 3; |
3538d68d OS |
6283 | } |
6284 | return 0; | |
6285 | } | |
6286 | ||
6287 | void btrfs_new_inode_args_destroy(struct btrfs_new_inode_args *args) | |
6288 | { | |
6289 | posix_acl_release(args->acl); | |
6290 | posix_acl_release(args->default_acl); | |
6291 | } | |
6292 | ||
19aee8de AJ |
6293 | /* |
6294 | * Inherit flags from the parent inode. | |
6295 | * | |
6296 | * Currently only the compression flags and the cow flags are inherited. | |
6297 | */ | |
6298 | static void btrfs_inherit_iflags(struct inode *inode, struct inode *dir) | |
6299 | { | |
6300 | unsigned int flags; | |
6301 | ||
19aee8de AJ |
6302 | flags = BTRFS_I(dir)->flags; |
6303 | ||
6304 | if (flags & BTRFS_INODE_NOCOMPRESS) { | |
6305 | BTRFS_I(inode)->flags &= ~BTRFS_INODE_COMPRESS; | |
6306 | BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS; | |
6307 | } else if (flags & BTRFS_INODE_COMPRESS) { | |
6308 | BTRFS_I(inode)->flags &= ~BTRFS_INODE_NOCOMPRESS; | |
6309 | BTRFS_I(inode)->flags |= BTRFS_INODE_COMPRESS; | |
6310 | } | |
6311 | ||
6312 | if (flags & BTRFS_INODE_NODATACOW) { | |
6313 | BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW; | |
6314 | if (S_ISREG(inode->i_mode)) | |
6315 | BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM; | |
6316 | } | |
6317 | ||
7b6a221e | 6318 | btrfs_sync_inode_flags_to_i_flags(inode); |
19aee8de AJ |
6319 | } |
6320 | ||
3538d68d | 6321 | int btrfs_create_new_inode(struct btrfs_trans_handle *trans, |
caae78e0 | 6322 | struct btrfs_new_inode_args *args) |
39279cc3 | 6323 | { |
caae78e0 | 6324 | struct inode *dir = args->dir; |
3538d68d | 6325 | struct inode *inode = args->inode; |
caae78e0 OS |
6326 | const char *name = args->orphan ? NULL : args->dentry->d_name.name; |
6327 | int name_len = args->orphan ? 0 : args->dentry->d_name.len; | |
6328 | struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb); | |
3538d68d | 6329 | struct btrfs_root *root; |
5f39d397 | 6330 | struct btrfs_inode_item *inode_item; |
39279cc3 | 6331 | struct btrfs_key *location; |
5f39d397 | 6332 | struct btrfs_path *path; |
6437d458 | 6333 | u64 objectid; |
9c58309d CM |
6334 | struct btrfs_inode_ref *ref; |
6335 | struct btrfs_key key[2]; | |
6336 | u32 sizes[2]; | |
b7ef5f3a | 6337 | struct btrfs_item_batch batch; |
9c58309d | 6338 | unsigned long ptr; |
39279cc3 | 6339 | int ret; |
39279cc3 | 6340 | |
5f39d397 | 6341 | path = btrfs_alloc_path(); |
d8926bb3 | 6342 | if (!path) |
a1fd0c35 | 6343 | return -ENOMEM; |
39279cc3 | 6344 | |
3538d68d OS |
6345 | if (!args->subvol) |
6346 | BTRFS_I(inode)->root = btrfs_grab_root(BTRFS_I(dir)->root); | |
6347 | root = BTRFS_I(inode)->root; | |
6348 | ||
6437d458 | 6349 | ret = btrfs_get_free_objectid(root, &objectid); |
caae78e0 OS |
6350 | if (ret) |
6351 | goto out; | |
581bb050 LZ |
6352 | inode->i_ino = objectid; |
6353 | ||
caae78e0 OS |
6354 | if (args->orphan) { |
6355 | /* | |
6356 | * O_TMPFILE, set link count to 0, so that after this point, we | |
6357 | * fill in an inode item with the correct link count. | |
6358 | */ | |
6359 | set_nlink(inode, 0); | |
6360 | } else { | |
1abe9b8a | 6361 | trace_btrfs_inode_request(dir); |
6362 | ||
caae78e0 OS |
6363 | ret = btrfs_set_inode_index(BTRFS_I(dir), &BTRFS_I(inode)->dir_index); |
6364 | if (ret) | |
6365 | goto out; | |
aec7477b | 6366 | } |
49024388 FM |
6367 | /* index_cnt is ignored for everything but a dir. */ |
6368 | BTRFS_I(inode)->index_cnt = BTRFS_DIR_START_INDEX; | |
e02119d5 | 6369 | BTRFS_I(inode)->generation = trans->transid; |
76195853 | 6370 | inode->i_generation = BTRFS_I(inode)->generation; |
b888db2b | 6371 | |
caae78e0 OS |
6372 | /* |
6373 | * Subvolumes don't inherit flags from their parent directory. | |
6374 | * Originally this was probably by accident, but we probably can't | |
6375 | * change it now without compatibility issues. | |
6376 | */ | |
6377 | if (!args->subvol) | |
6378 | btrfs_inherit_iflags(inode, dir); | |
305eaac0 | 6379 | |
a1fd0c35 | 6380 | if (S_ISREG(inode->i_mode)) { |
305eaac0 OS |
6381 | if (btrfs_test_opt(fs_info, NODATASUM)) |
6382 | BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM; | |
6383 | if (btrfs_test_opt(fs_info, NODATACOW)) | |
6384 | BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW | | |
6385 | BTRFS_INODE_NODATASUM; | |
6386 | } | |
6387 | ||
caae78e0 OS |
6388 | location = &BTRFS_I(inode)->location; |
6389 | location->objectid = objectid; | |
6390 | location->offset = 0; | |
6391 | location->type = BTRFS_INODE_ITEM_KEY; | |
6392 | ||
6393 | ret = btrfs_insert_inode_locked(inode); | |
6394 | if (ret < 0) { | |
6395 | if (!args->orphan) | |
6396 | BTRFS_I(dir)->index_cnt--; | |
6397 | goto out; | |
6398 | } | |
6399 | ||
5dc562c5 JB |
6400 | /* |
6401 | * We could have gotten an inode number from somebody who was fsynced | |
6402 | * and then removed in this same transaction, so let's just set full | |
6403 | * sync since it will be a full sync anyway and this will blow away the | |
6404 | * old info in the log. | |
6405 | */ | |
23e3337f | 6406 | btrfs_set_inode_full_sync(BTRFS_I(inode)); |
5dc562c5 | 6407 | |
9c58309d | 6408 | key[0].objectid = objectid; |
962a298f | 6409 | key[0].type = BTRFS_INODE_ITEM_KEY; |
9c58309d CM |
6410 | key[0].offset = 0; |
6411 | ||
9c58309d | 6412 | sizes[0] = sizeof(struct btrfs_inode_item); |
ef3b9af5 | 6413 | |
caae78e0 | 6414 | if (!args->orphan) { |
ef3b9af5 FM |
6415 | /* |
6416 | * Start new inodes with an inode_ref. This is slightly more | |
6417 | * efficient for small numbers of hard links since they will | |
6418 | * be packed into one item. Extended refs will kick in if we | |
6419 | * add more hard links than can fit in the ref item. | |
6420 | */ | |
6421 | key[1].objectid = objectid; | |
962a298f | 6422 | key[1].type = BTRFS_INODE_REF_KEY; |
caae78e0 | 6423 | if (args->subvol) { |
23c24ef8 | 6424 | key[1].offset = objectid; |
caae78e0 OS |
6425 | sizes[1] = 2 + sizeof(*ref); |
6426 | } else { | |
6427 | key[1].offset = btrfs_ino(BTRFS_I(dir)); | |
6428 | sizes[1] = name_len + sizeof(*ref); | |
6429 | } | |
ef3b9af5 | 6430 | } |
9c58309d | 6431 | |
b7ef5f3a FM |
6432 | batch.keys = &key[0]; |
6433 | batch.data_sizes = &sizes[0]; | |
caae78e0 OS |
6434 | batch.total_data_size = sizes[0] + (args->orphan ? 0 : sizes[1]); |
6435 | batch.nr = args->orphan ? 1 : 2; | |
b7ef5f3a | 6436 | ret = btrfs_insert_empty_items(trans, root, path, &batch); |
caae78e0 OS |
6437 | if (ret != 0) { |
6438 | btrfs_abort_transaction(trans, ret); | |
6439 | goto discard; | |
6440 | } | |
5f39d397 | 6441 | |
c2050a45 | 6442 | inode->i_mtime = current_time(inode); |
9cc97d64 | 6443 | inode->i_atime = inode->i_mtime; |
6444 | inode->i_ctime = inode->i_mtime; | |
d3c6be6f | 6445 | BTRFS_I(inode)->i_otime = inode->i_mtime; |
9cc97d64 | 6446 | |
caae78e0 OS |
6447 | /* |
6448 | * We're going to fill the inode item now, so at this point the inode | |
6449 | * must be fully initialized. | |
6450 | */ | |
6451 | ||
5f39d397 CM |
6452 | inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0], |
6453 | struct btrfs_inode_item); | |
b159fa28 | 6454 | memzero_extent_buffer(path->nodes[0], (unsigned long)inode_item, |
293f7e07 | 6455 | sizeof(*inode_item)); |
e02119d5 | 6456 | fill_inode_item(trans, path->nodes[0], inode_item, inode); |
9c58309d | 6457 | |
caae78e0 | 6458 | if (!args->orphan) { |
ef3b9af5 FM |
6459 | ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1, |
6460 | struct btrfs_inode_ref); | |
ef3b9af5 | 6461 | ptr = (unsigned long)(ref + 1); |
caae78e0 OS |
6462 | if (args->subvol) { |
6463 | btrfs_set_inode_ref_name_len(path->nodes[0], ref, 2); | |
6464 | btrfs_set_inode_ref_index(path->nodes[0], ref, 0); | |
6465 | write_extent_buffer(path->nodes[0], "..", ptr, 2); | |
6466 | } else { | |
6467 | btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len); | |
6468 | btrfs_set_inode_ref_index(path->nodes[0], ref, | |
6469 | BTRFS_I(inode)->dir_index); | |
6470 | write_extent_buffer(path->nodes[0], name, ptr, name_len); | |
6471 | } | |
ef3b9af5 | 6472 | } |
9c58309d | 6473 | |
5f39d397 | 6474 | btrfs_mark_buffer_dirty(path->nodes[0]); |
814e7718 FM |
6475 | /* |
6476 | * We don't need the path anymore, plus inheriting properties, adding | |
6477 | * ACLs, security xattrs, orphan item or adding the link, will result in | |
6478 | * allocating yet another path. So just free our path. | |
6479 | */ | |
6480 | btrfs_free_path(path); | |
6481 | path = NULL; | |
5f39d397 | 6482 | |
6c3636eb STD |
6483 | if (args->subvol) { |
6484 | struct inode *parent; | |
6485 | ||
6486 | /* | |
6487 | * Subvolumes inherit properties from their parent subvolume, | |
6488 | * not the directory they were created in. | |
6489 | */ | |
6490 | parent = btrfs_iget(fs_info->sb, BTRFS_FIRST_FREE_OBJECTID, | |
6491 | BTRFS_I(dir)->root); | |
6492 | if (IS_ERR(parent)) { | |
6493 | ret = PTR_ERR(parent); | |
6494 | } else { | |
6495 | ret = btrfs_inode_inherit_props(trans, inode, parent); | |
6496 | iput(parent); | |
6497 | } | |
6498 | } else { | |
6499 | ret = btrfs_inode_inherit_props(trans, inode, dir); | |
6500 | } | |
6501 | if (ret) { | |
6502 | btrfs_err(fs_info, | |
6503 | "error inheriting props for ino %llu (root %llu): %d", | |
6504 | btrfs_ino(BTRFS_I(inode)), root->root_key.objectid, | |
6505 | ret); | |
6506 | } | |
6507 | ||
6508 | /* | |
6509 | * Subvolumes don't inherit ACLs or get passed to the LSM. This is | |
6510 | * probably a bug. | |
6511 | */ | |
6512 | if (!args->subvol) { | |
6513 | ret = btrfs_init_inode_security(trans, args); | |
6514 | if (ret) { | |
6515 | btrfs_abort_transaction(trans, ret); | |
6516 | goto discard; | |
6517 | } | |
6518 | } | |
6519 | ||
5d4f98a2 | 6520 | inode_tree_add(inode); |
1abe9b8a | 6521 | |
6522 | trace_btrfs_inode_new(inode); | |
d9094414 | 6523 | btrfs_set_inode_last_trans(trans, BTRFS_I(inode)); |
1abe9b8a | 6524 | |
8ea05e3a AB |
6525 | btrfs_update_root_times(trans, root); |
6526 | ||
caae78e0 OS |
6527 | if (args->orphan) { |
6528 | ret = btrfs_orphan_add(trans, BTRFS_I(inode)); | |
6529 | } else { | |
6530 | ret = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode), name, | |
6531 | name_len, 0, BTRFS_I(inode)->dir_index); | |
6532 | } | |
6533 | if (ret) { | |
6534 | btrfs_abort_transaction(trans, ret); | |
6535 | goto discard; | |
6536 | } | |
63541927 | 6537 | |
814e7718 | 6538 | return 0; |
b0d5d10f | 6539 | |
caae78e0 | 6540 | discard: |
a1fd0c35 OS |
6541 | /* |
6542 | * discard_new_inode() calls iput(), but the caller owns the reference | |
6543 | * to the inode. | |
6544 | */ | |
6545 | ihold(inode); | |
32955c54 | 6546 | discard_new_inode(inode); |
caae78e0 | 6547 | out: |
5f39d397 | 6548 | btrfs_free_path(path); |
a1fd0c35 | 6549 | return ret; |
39279cc3 CM |
6550 | } |
6551 | ||
d352ac68 CM |
6552 | /* |
6553 | * utility function to add 'inode' into 'parent_inode' with | |
6554 | * a give name and a given sequence number. | |
6555 | * if 'add_backref' is true, also insert a backref from the | |
6556 | * inode to the parent directory. | |
6557 | */ | |
e02119d5 | 6558 | int btrfs_add_link(struct btrfs_trans_handle *trans, |
db0a669f | 6559 | struct btrfs_inode *parent_inode, struct btrfs_inode *inode, |
e02119d5 | 6560 | const char *name, int name_len, int add_backref, u64 index) |
39279cc3 | 6561 | { |
4df27c4d | 6562 | int ret = 0; |
39279cc3 | 6563 | struct btrfs_key key; |
db0a669f NB |
6564 | struct btrfs_root *root = parent_inode->root; |
6565 | u64 ino = btrfs_ino(inode); | |
6566 | u64 parent_ino = btrfs_ino(parent_inode); | |
5f39d397 | 6567 | |
33345d01 | 6568 | if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) { |
db0a669f | 6569 | memcpy(&key, &inode->root->root_key, sizeof(key)); |
4df27c4d | 6570 | } else { |
33345d01 | 6571 | key.objectid = ino; |
962a298f | 6572 | key.type = BTRFS_INODE_ITEM_KEY; |
4df27c4d YZ |
6573 | key.offset = 0; |
6574 | } | |
6575 | ||
33345d01 | 6576 | if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) { |
6025c19f | 6577 | ret = btrfs_add_root_ref(trans, key.objectid, |
0b246afa JM |
6578 | root->root_key.objectid, parent_ino, |
6579 | index, name, name_len); | |
4df27c4d | 6580 | } else if (add_backref) { |
33345d01 LZ |
6581 | ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino, |
6582 | parent_ino, index); | |
4df27c4d | 6583 | } |
39279cc3 | 6584 | |
79787eaa JM |
6585 | /* Nothing to clean up yet */ |
6586 | if (ret) | |
6587 | return ret; | |
4df27c4d | 6588 | |
684572df | 6589 | ret = btrfs_insert_dir_item(trans, name, name_len, parent_inode, &key, |
db0a669f | 6590 | btrfs_inode_type(&inode->vfs_inode), index); |
9c52057c | 6591 | if (ret == -EEXIST || ret == -EOVERFLOW) |
79787eaa JM |
6592 | goto fail_dir_item; |
6593 | else if (ret) { | |
66642832 | 6594 | btrfs_abort_transaction(trans, ret); |
79787eaa | 6595 | return ret; |
39279cc3 | 6596 | } |
79787eaa | 6597 | |
db0a669f | 6598 | btrfs_i_size_write(parent_inode, parent_inode->vfs_inode.i_size + |
79787eaa | 6599 | name_len * 2); |
db0a669f | 6600 | inode_inc_iversion(&parent_inode->vfs_inode); |
5338e43a FM |
6601 | /* |
6602 | * If we are replaying a log tree, we do not want to update the mtime | |
6603 | * and ctime of the parent directory with the current time, since the | |
6604 | * log replay procedure is responsible for setting them to their correct | |
6605 | * values (the ones it had when the fsync was done). | |
6606 | */ | |
6607 | if (!test_bit(BTRFS_FS_LOG_RECOVERING, &root->fs_info->flags)) { | |
6608 | struct timespec64 now = current_time(&parent_inode->vfs_inode); | |
6609 | ||
6610 | parent_inode->vfs_inode.i_mtime = now; | |
6611 | parent_inode->vfs_inode.i_ctime = now; | |
6612 | } | |
9a56fcd1 | 6613 | ret = btrfs_update_inode(trans, root, parent_inode); |
79787eaa | 6614 | if (ret) |
66642832 | 6615 | btrfs_abort_transaction(trans, ret); |
39279cc3 | 6616 | return ret; |
fe66a05a CM |
6617 | |
6618 | fail_dir_item: | |
6619 | if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) { | |
6620 | u64 local_index; | |
6621 | int err; | |
3ee1c553 | 6622 | err = btrfs_del_root_ref(trans, key.objectid, |
0b246afa JM |
6623 | root->root_key.objectid, parent_ino, |
6624 | &local_index, name, name_len); | |
1690dd41 JT |
6625 | if (err) |
6626 | btrfs_abort_transaction(trans, err); | |
fe66a05a CM |
6627 | } else if (add_backref) { |
6628 | u64 local_index; | |
6629 | int err; | |
6630 | ||
6631 | err = btrfs_del_inode_ref(trans, root, name, name_len, | |
6632 | ino, parent_ino, &local_index); | |
1690dd41 JT |
6633 | if (err) |
6634 | btrfs_abort_transaction(trans, err); | |
fe66a05a | 6635 | } |
1690dd41 JT |
6636 | |
6637 | /* Return the original error code */ | |
fe66a05a | 6638 | return ret; |
39279cc3 CM |
6639 | } |
6640 | ||
5f465bf1 OS |
6641 | static int btrfs_create_common(struct inode *dir, struct dentry *dentry, |
6642 | struct inode *inode) | |
618e21d5 | 6643 | { |
2ff7e61e | 6644 | struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb); |
618e21d5 | 6645 | struct btrfs_root *root = BTRFS_I(dir)->root; |
3538d68d OS |
6646 | struct btrfs_new_inode_args new_inode_args = { |
6647 | .dir = dir, | |
6648 | .dentry = dentry, | |
6649 | .inode = inode, | |
6650 | }; | |
6651 | unsigned int trans_num_items; | |
5f465bf1 | 6652 | struct btrfs_trans_handle *trans; |
618e21d5 | 6653 | int err; |
618e21d5 | 6654 | |
3538d68d | 6655 | err = btrfs_new_inode_prepare(&new_inode_args, &trans_num_items); |
caae78e0 OS |
6656 | if (err) |
6657 | goto out_inode; | |
3538d68d OS |
6658 | |
6659 | trans = btrfs_start_transaction(root, trans_num_items); | |
a1fd0c35 | 6660 | if (IS_ERR(trans)) { |
3538d68d OS |
6661 | err = PTR_ERR(trans); |
6662 | goto out_new_inode_args; | |
a1fd0c35 | 6663 | } |
1832a6d5 | 6664 | |
caae78e0 OS |
6665 | err = btrfs_create_new_inode(trans, &new_inode_args); |
6666 | if (!err) | |
6667 | d_instantiate_new(dentry, inode); | |
b0d5d10f | 6668 | |
3a45bb20 | 6669 | btrfs_end_transaction(trans); |
5f465bf1 | 6670 | btrfs_btree_balance_dirty(fs_info); |
3538d68d OS |
6671 | out_new_inode_args: |
6672 | btrfs_new_inode_args_destroy(&new_inode_args); | |
caae78e0 OS |
6673 | out_inode: |
6674 | if (err) | |
6675 | iput(inode); | |
618e21d5 JB |
6676 | return err; |
6677 | } | |
6678 | ||
5f465bf1 OS |
6679 | static int btrfs_mknod(struct user_namespace *mnt_userns, struct inode *dir, |
6680 | struct dentry *dentry, umode_t mode, dev_t rdev) | |
6681 | { | |
6682 | struct inode *inode; | |
6683 | ||
6684 | inode = new_inode(dir->i_sb); | |
6685 | if (!inode) | |
6686 | return -ENOMEM; | |
6687 | inode_init_owner(mnt_userns, inode, dir, mode); | |
6688 | inode->i_op = &btrfs_special_inode_operations; | |
6689 | init_special_inode(inode, inode->i_mode, rdev); | |
6690 | return btrfs_create_common(dir, dentry, inode); | |
6691 | } | |
6692 | ||
549c7297 CB |
6693 | static int btrfs_create(struct user_namespace *mnt_userns, struct inode *dir, |
6694 | struct dentry *dentry, umode_t mode, bool excl) | |
39279cc3 | 6695 | { |
a1fd0c35 | 6696 | struct inode *inode; |
39279cc3 | 6697 | |
a1fd0c35 OS |
6698 | inode = new_inode(dir->i_sb); |
6699 | if (!inode) | |
6700 | return -ENOMEM; | |
6701 | inode_init_owner(mnt_userns, inode, dir, mode); | |
6702 | inode->i_fop = &btrfs_file_operations; | |
6703 | inode->i_op = &btrfs_file_inode_operations; | |
6704 | inode->i_mapping->a_ops = &btrfs_aops; | |
5f465bf1 | 6705 | return btrfs_create_common(dir, dentry, inode); |
39279cc3 CM |
6706 | } |
6707 | ||
6708 | static int btrfs_link(struct dentry *old_dentry, struct inode *dir, | |
6709 | struct dentry *dentry) | |
6710 | { | |
271dba45 | 6711 | struct btrfs_trans_handle *trans = NULL; |
39279cc3 | 6712 | struct btrfs_root *root = BTRFS_I(dir)->root; |
2b0143b5 | 6713 | struct inode *inode = d_inode(old_dentry); |
2ff7e61e | 6714 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
00e4e6b3 | 6715 | u64 index; |
39279cc3 CM |
6716 | int err; |
6717 | int drop_inode = 0; | |
6718 | ||
4a8be425 | 6719 | /* do not allow sys_link's with other subvols of the same device */ |
4fd786e6 | 6720 | if (root->root_key.objectid != BTRFS_I(inode)->root->root_key.objectid) |
3ab3564f | 6721 | return -EXDEV; |
4a8be425 | 6722 | |
f186373f | 6723 | if (inode->i_nlink >= BTRFS_LINK_MAX) |
c055e99e | 6724 | return -EMLINK; |
4a8be425 | 6725 | |
877574e2 | 6726 | err = btrfs_set_inode_index(BTRFS_I(dir), &index); |
aec7477b JB |
6727 | if (err) |
6728 | goto fail; | |
6729 | ||
a22285a6 | 6730 | /* |
7e6b6465 | 6731 | * 2 items for inode and inode ref |
a22285a6 | 6732 | * 2 items for dir items |
7e6b6465 | 6733 | * 1 item for parent inode |
399b0bbf | 6734 | * 1 item for orphan item deletion if O_TMPFILE |
a22285a6 | 6735 | */ |
399b0bbf | 6736 | trans = btrfs_start_transaction(root, inode->i_nlink ? 5 : 6); |
a22285a6 YZ |
6737 | if (IS_ERR(trans)) { |
6738 | err = PTR_ERR(trans); | |
271dba45 | 6739 | trans = NULL; |
a22285a6 YZ |
6740 | goto fail; |
6741 | } | |
5f39d397 | 6742 | |
67de1176 MX |
6743 | /* There are several dir indexes for this inode, clear the cache. */ |
6744 | BTRFS_I(inode)->dir_index = 0ULL; | |
8b558c5f | 6745 | inc_nlink(inode); |
0c4d2d95 | 6746 | inode_inc_iversion(inode); |
c2050a45 | 6747 | inode->i_ctime = current_time(inode); |
7de9c6ee | 6748 | ihold(inode); |
e9976151 | 6749 | set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags); |
aec7477b | 6750 | |
81512e89 OS |
6751 | err = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode), |
6752 | dentry->d_name.name, dentry->d_name.len, 1, index); | |
5f39d397 | 6753 | |
a5719521 | 6754 | if (err) { |
54aa1f4d | 6755 | drop_inode = 1; |
a5719521 | 6756 | } else { |
10d9f309 | 6757 | struct dentry *parent = dentry->d_parent; |
d4682ba0 | 6758 | |
9a56fcd1 | 6759 | err = btrfs_update_inode(trans, root, BTRFS_I(inode)); |
79787eaa JM |
6760 | if (err) |
6761 | goto fail; | |
ef3b9af5 FM |
6762 | if (inode->i_nlink == 1) { |
6763 | /* | |
6764 | * If new hard link count is 1, it's a file created | |
6765 | * with open(2) O_TMPFILE flag. | |
6766 | */ | |
3d6ae7bb | 6767 | err = btrfs_orphan_del(trans, BTRFS_I(inode)); |
ef3b9af5 FM |
6768 | if (err) |
6769 | goto fail; | |
6770 | } | |
08c422c2 | 6771 | d_instantiate(dentry, inode); |
88d2beec | 6772 | btrfs_log_new_name(trans, old_dentry, NULL, 0, parent); |
a5719521 | 6773 | } |
39279cc3 | 6774 | |
1832a6d5 | 6775 | fail: |
271dba45 | 6776 | if (trans) |
3a45bb20 | 6777 | btrfs_end_transaction(trans); |
39279cc3 CM |
6778 | if (drop_inode) { |
6779 | inode_dec_link_count(inode); | |
6780 | iput(inode); | |
6781 | } | |
2ff7e61e | 6782 | btrfs_btree_balance_dirty(fs_info); |
39279cc3 CM |
6783 | return err; |
6784 | } | |
6785 | ||
549c7297 CB |
6786 | static int btrfs_mkdir(struct user_namespace *mnt_userns, struct inode *dir, |
6787 | struct dentry *dentry, umode_t mode) | |
39279cc3 | 6788 | { |
a1fd0c35 | 6789 | struct inode *inode; |
39279cc3 | 6790 | |
a1fd0c35 OS |
6791 | inode = new_inode(dir->i_sb); |
6792 | if (!inode) | |
6793 | return -ENOMEM; | |
6794 | inode_init_owner(mnt_userns, inode, dir, S_IFDIR | mode); | |
6795 | inode->i_op = &btrfs_dir_inode_operations; | |
6796 | inode->i_fop = &btrfs_dir_file_operations; | |
5f465bf1 | 6797 | return btrfs_create_common(dir, dentry, inode); |
39279cc3 CM |
6798 | } |
6799 | ||
c8b97818 | 6800 | static noinline int uncompress_inline(struct btrfs_path *path, |
e40da0e5 | 6801 | struct page *page, |
c8b97818 CM |
6802 | size_t pg_offset, u64 extent_offset, |
6803 | struct btrfs_file_extent_item *item) | |
6804 | { | |
6805 | int ret; | |
6806 | struct extent_buffer *leaf = path->nodes[0]; | |
6807 | char *tmp; | |
6808 | size_t max_size; | |
6809 | unsigned long inline_size; | |
6810 | unsigned long ptr; | |
261507a0 | 6811 | int compress_type; |
c8b97818 CM |
6812 | |
6813 | WARN_ON(pg_offset != 0); | |
261507a0 | 6814 | compress_type = btrfs_file_extent_compression(leaf, item); |
c8b97818 | 6815 | max_size = btrfs_file_extent_ram_bytes(leaf, item); |
437bd07e | 6816 | inline_size = btrfs_file_extent_inline_item_len(leaf, path->slots[0]); |
c8b97818 | 6817 | tmp = kmalloc(inline_size, GFP_NOFS); |
8d413713 TI |
6818 | if (!tmp) |
6819 | return -ENOMEM; | |
c8b97818 CM |
6820 | ptr = btrfs_file_extent_inline_start(item); |
6821 | ||
6822 | read_extent_buffer(leaf, tmp, ptr, inline_size); | |
6823 | ||
09cbfeaf | 6824 | max_size = min_t(unsigned long, PAGE_SIZE, max_size); |
261507a0 LZ |
6825 | ret = btrfs_decompress(compress_type, tmp, page, |
6826 | extent_offset, inline_size, max_size); | |
e1699d2d ZB |
6827 | |
6828 | /* | |
6829 | * decompression code contains a memset to fill in any space between the end | |
6830 | * of the uncompressed data and the end of max_size in case the decompressed | |
6831 | * data ends up shorter than ram_bytes. That doesn't cover the hole between | |
6832 | * the end of an inline extent and the beginning of the next block, so we | |
6833 | * cover that region here. | |
6834 | */ | |
6835 | ||
d048b9c2 IW |
6836 | if (max_size + pg_offset < PAGE_SIZE) |
6837 | memzero_page(page, pg_offset + max_size, | |
6838 | PAGE_SIZE - max_size - pg_offset); | |
c8b97818 | 6839 | kfree(tmp); |
166ae5a4 | 6840 | return ret; |
c8b97818 CM |
6841 | } |
6842 | ||
39b07b5d OS |
6843 | /** |
6844 | * btrfs_get_extent - Lookup the first extent overlapping a range in a file. | |
6845 | * @inode: file to search in | |
6846 | * @page: page to read extent data into if the extent is inline | |
6847 | * @pg_offset: offset into @page to copy to | |
6848 | * @start: file offset | |
6849 | * @len: length of range starting at @start | |
6850 | * | |
6851 | * This returns the first &struct extent_map which overlaps with the given | |
6852 | * range, reading it from the B-tree and caching it if necessary. Note that | |
6853 | * there may be more extents which overlap the given range after the returned | |
6854 | * extent_map. | |
d352ac68 | 6855 | * |
39b07b5d OS |
6856 | * If @page is not NULL and the extent is inline, this also reads the extent |
6857 | * data directly into the page and marks the extent up to date in the io_tree. | |
6858 | * | |
6859 | * Return: ERR_PTR on error, non-NULL extent_map on success. | |
d352ac68 | 6860 | */ |
fc4f21b1 | 6861 | struct extent_map *btrfs_get_extent(struct btrfs_inode *inode, |
39b07b5d OS |
6862 | struct page *page, size_t pg_offset, |
6863 | u64 start, u64 len) | |
a52d9a80 | 6864 | { |
3ffbd68c | 6865 | struct btrfs_fs_info *fs_info = inode->root->fs_info; |
1028d1c4 | 6866 | int ret = 0; |
a52d9a80 CM |
6867 | u64 extent_start = 0; |
6868 | u64 extent_end = 0; | |
fc4f21b1 | 6869 | u64 objectid = btrfs_ino(inode); |
7e74e235 | 6870 | int extent_type = -1; |
f421950f | 6871 | struct btrfs_path *path = NULL; |
fc4f21b1 | 6872 | struct btrfs_root *root = inode->root; |
a52d9a80 | 6873 | struct btrfs_file_extent_item *item; |
5f39d397 CM |
6874 | struct extent_buffer *leaf; |
6875 | struct btrfs_key found_key; | |
a52d9a80 | 6876 | struct extent_map *em = NULL; |
fc4f21b1 | 6877 | struct extent_map_tree *em_tree = &inode->extent_tree; |
a52d9a80 | 6878 | |
890871be | 6879 | read_lock(&em_tree->lock); |
d1310b2e | 6880 | em = lookup_extent_mapping(em_tree, start, len); |
890871be | 6881 | read_unlock(&em_tree->lock); |
d1310b2e | 6882 | |
a52d9a80 | 6883 | if (em) { |
e1c4b745 CM |
6884 | if (em->start > start || em->start + em->len <= start) |
6885 | free_extent_map(em); | |
6886 | else if (em->block_start == EXTENT_MAP_INLINE && page) | |
70dec807 CM |
6887 | free_extent_map(em); |
6888 | else | |
6889 | goto out; | |
a52d9a80 | 6890 | } |
172ddd60 | 6891 | em = alloc_extent_map(); |
a52d9a80 | 6892 | if (!em) { |
1028d1c4 | 6893 | ret = -ENOMEM; |
d1310b2e | 6894 | goto out; |
a52d9a80 | 6895 | } |
d1310b2e | 6896 | em->start = EXTENT_MAP_HOLE; |
445a6944 | 6897 | em->orig_start = EXTENT_MAP_HOLE; |
d1310b2e | 6898 | em->len = (u64)-1; |
c8b97818 | 6899 | em->block_len = (u64)-1; |
f421950f | 6900 | |
bee6ec82 | 6901 | path = btrfs_alloc_path(); |
f421950f | 6902 | if (!path) { |
1028d1c4 | 6903 | ret = -ENOMEM; |
bee6ec82 | 6904 | goto out; |
f421950f CM |
6905 | } |
6906 | ||
bee6ec82 LB |
6907 | /* Chances are we'll be called again, so go ahead and do readahead */ |
6908 | path->reada = READA_FORWARD; | |
4d7240f0 JB |
6909 | |
6910 | /* | |
6911 | * The same explanation in load_free_space_cache applies here as well, | |
6912 | * we only read when we're loading the free space cache, and at that | |
6913 | * point the commit_root has everything we need. | |
6914 | */ | |
6915 | if (btrfs_is_free_space_inode(inode)) { | |
6916 | path->search_commit_root = 1; | |
6917 | path->skip_locking = 1; | |
6918 | } | |
51899412 | 6919 | |
5c9a702e | 6920 | ret = btrfs_lookup_file_extent(NULL, root, path, objectid, start, 0); |
a52d9a80 | 6921 | if (ret < 0) { |
a52d9a80 | 6922 | goto out; |
b8eeab7f | 6923 | } else if (ret > 0) { |
a52d9a80 CM |
6924 | if (path->slots[0] == 0) |
6925 | goto not_found; | |
6926 | path->slots[0]--; | |
1028d1c4 | 6927 | ret = 0; |
a52d9a80 CM |
6928 | } |
6929 | ||
5f39d397 CM |
6930 | leaf = path->nodes[0]; |
6931 | item = btrfs_item_ptr(leaf, path->slots[0], | |
a52d9a80 | 6932 | struct btrfs_file_extent_item); |
5f39d397 | 6933 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
5f39d397 | 6934 | if (found_key.objectid != objectid || |
694c12ed | 6935 | found_key.type != BTRFS_EXTENT_DATA_KEY) { |
25a50341 JB |
6936 | /* |
6937 | * If we backup past the first extent we want to move forward | |
6938 | * and see if there is an extent in front of us, otherwise we'll | |
6939 | * say there is a hole for our whole search range which can | |
6940 | * cause problems. | |
6941 | */ | |
6942 | extent_end = start; | |
6943 | goto next; | |
a52d9a80 CM |
6944 | } |
6945 | ||
694c12ed | 6946 | extent_type = btrfs_file_extent_type(leaf, item); |
5f39d397 | 6947 | extent_start = found_key.offset; |
a5eeb3d1 | 6948 | extent_end = btrfs_file_extent_end(path); |
694c12ed NB |
6949 | if (extent_type == BTRFS_FILE_EXTENT_REG || |
6950 | extent_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
6bf9e4bd QW |
6951 | /* Only regular file could have regular/prealloc extent */ |
6952 | if (!S_ISREG(inode->vfs_inode.i_mode)) { | |
1028d1c4 | 6953 | ret = -EUCLEAN; |
6bf9e4bd QW |
6954 | btrfs_crit(fs_info, |
6955 | "regular/prealloc extent found for non-regular inode %llu", | |
6956 | btrfs_ino(inode)); | |
6957 | goto out; | |
6958 | } | |
09ed2f16 LB |
6959 | trace_btrfs_get_extent_show_fi_regular(inode, leaf, item, |
6960 | extent_start); | |
694c12ed | 6961 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
09ed2f16 LB |
6962 | trace_btrfs_get_extent_show_fi_inline(inode, leaf, item, |
6963 | path->slots[0], | |
6964 | extent_start); | |
9036c102 | 6965 | } |
25a50341 | 6966 | next: |
9036c102 YZ |
6967 | if (start >= extent_end) { |
6968 | path->slots[0]++; | |
6969 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
6970 | ret = btrfs_next_leaf(root, path); | |
1028d1c4 | 6971 | if (ret < 0) |
9036c102 | 6972 | goto out; |
1028d1c4 | 6973 | else if (ret > 0) |
9036c102 | 6974 | goto not_found; |
1028d1c4 | 6975 | |
9036c102 | 6976 | leaf = path->nodes[0]; |
a52d9a80 | 6977 | } |
9036c102 YZ |
6978 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
6979 | if (found_key.objectid != objectid || | |
6980 | found_key.type != BTRFS_EXTENT_DATA_KEY) | |
6981 | goto not_found; | |
6982 | if (start + len <= found_key.offset) | |
6983 | goto not_found; | |
e2eca69d WS |
6984 | if (start > found_key.offset) |
6985 | goto next; | |
02a033df NB |
6986 | |
6987 | /* New extent overlaps with existing one */ | |
9036c102 | 6988 | em->start = start; |
70c8a91c | 6989 | em->orig_start = start; |
9036c102 | 6990 | em->len = found_key.offset - start; |
02a033df NB |
6991 | em->block_start = EXTENT_MAP_HOLE; |
6992 | goto insert; | |
9036c102 YZ |
6993 | } |
6994 | ||
39b07b5d | 6995 | btrfs_extent_item_to_extent_map(inode, path, item, !page, em); |
7ffbb598 | 6996 | |
694c12ed NB |
6997 | if (extent_type == BTRFS_FILE_EXTENT_REG || |
6998 | extent_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
a52d9a80 | 6999 | goto insert; |
694c12ed | 7000 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
5f39d397 | 7001 | unsigned long ptr; |
a52d9a80 | 7002 | char *map; |
3326d1b0 CM |
7003 | size_t size; |
7004 | size_t extent_offset; | |
7005 | size_t copy_size; | |
a52d9a80 | 7006 | |
39b07b5d | 7007 | if (!page) |
689f9346 | 7008 | goto out; |
5f39d397 | 7009 | |
e41ca589 | 7010 | size = btrfs_file_extent_ram_bytes(leaf, item); |
9036c102 | 7011 | extent_offset = page_offset(page) + pg_offset - extent_start; |
09cbfeaf KS |
7012 | copy_size = min_t(u64, PAGE_SIZE - pg_offset, |
7013 | size - extent_offset); | |
3326d1b0 | 7014 | em->start = extent_start + extent_offset; |
0b246afa | 7015 | em->len = ALIGN(copy_size, fs_info->sectorsize); |
b4939680 | 7016 | em->orig_block_len = em->len; |
70c8a91c | 7017 | em->orig_start = em->start; |
689f9346 | 7018 | ptr = btrfs_file_extent_inline_start(item) + extent_offset; |
e49aabd9 | 7019 | |
bf46f52d | 7020 | if (!PageUptodate(page)) { |
261507a0 LZ |
7021 | if (btrfs_file_extent_compression(leaf, item) != |
7022 | BTRFS_COMPRESS_NONE) { | |
e40da0e5 | 7023 | ret = uncompress_inline(path, page, pg_offset, |
c8b97818 | 7024 | extent_offset, item); |
1028d1c4 | 7025 | if (ret) |
166ae5a4 | 7026 | goto out; |
c8b97818 | 7027 | } else { |
58c1a35c | 7028 | map = kmap_local_page(page); |
c8b97818 CM |
7029 | read_extent_buffer(leaf, map + pg_offset, ptr, |
7030 | copy_size); | |
09cbfeaf | 7031 | if (pg_offset + copy_size < PAGE_SIZE) { |
93c82d57 | 7032 | memset(map + pg_offset + copy_size, 0, |
09cbfeaf | 7033 | PAGE_SIZE - pg_offset - |
93c82d57 CM |
7034 | copy_size); |
7035 | } | |
58c1a35c | 7036 | kunmap_local(map); |
c8b97818 | 7037 | } |
179e29e4 | 7038 | flush_dcache_page(page); |
a52d9a80 | 7039 | } |
a52d9a80 | 7040 | goto insert; |
a52d9a80 CM |
7041 | } |
7042 | not_found: | |
7043 | em->start = start; | |
70c8a91c | 7044 | em->orig_start = start; |
d1310b2e | 7045 | em->len = len; |
5f39d397 | 7046 | em->block_start = EXTENT_MAP_HOLE; |
a52d9a80 | 7047 | insert: |
1028d1c4 | 7048 | ret = 0; |
b3b4aa74 | 7049 | btrfs_release_path(path); |
d1310b2e | 7050 | if (em->start > start || extent_map_end(em) <= start) { |
0b246afa | 7051 | btrfs_err(fs_info, |
5d163e0e JM |
7052 | "bad extent! em: [%llu %llu] passed [%llu %llu]", |
7053 | em->start, em->len, start, len); | |
1028d1c4 | 7054 | ret = -EIO; |
a52d9a80 CM |
7055 | goto out; |
7056 | } | |
d1310b2e | 7057 | |
890871be | 7058 | write_lock(&em_tree->lock); |
1028d1c4 | 7059 | ret = btrfs_add_extent_mapping(fs_info, em_tree, &em, start, len); |
890871be | 7060 | write_unlock(&em_tree->lock); |
a52d9a80 | 7061 | out: |
c6414280 | 7062 | btrfs_free_path(path); |
1abe9b8a | 7063 | |
fc4f21b1 | 7064 | trace_btrfs_get_extent(root, inode, em); |
1abe9b8a | 7065 | |
1028d1c4 | 7066 | if (ret) { |
a52d9a80 | 7067 | free_extent_map(em); |
1028d1c4 | 7068 | return ERR_PTR(ret); |
a52d9a80 CM |
7069 | } |
7070 | return em; | |
7071 | } | |
7072 | ||
64f54188 | 7073 | static struct extent_map *btrfs_create_dio_extent(struct btrfs_inode *inode, |
5f9a8a51 FM |
7074 | const u64 start, |
7075 | const u64 len, | |
7076 | const u64 orig_start, | |
7077 | const u64 block_start, | |
7078 | const u64 block_len, | |
7079 | const u64 orig_block_len, | |
7080 | const u64 ram_bytes, | |
7081 | const int type) | |
7082 | { | |
7083 | struct extent_map *em = NULL; | |
7084 | int ret; | |
7085 | ||
5f9a8a51 | 7086 | if (type != BTRFS_ORDERED_NOCOW) { |
64f54188 NB |
7087 | em = create_io_em(inode, start, len, orig_start, block_start, |
7088 | block_len, orig_block_len, ram_bytes, | |
6f9994db LB |
7089 | BTRFS_COMPRESS_NONE, /* compress_type */ |
7090 | type); | |
5f9a8a51 FM |
7091 | if (IS_ERR(em)) |
7092 | goto out; | |
7093 | } | |
cb36a9bb OS |
7094 | ret = btrfs_add_ordered_extent(inode, start, len, len, block_start, |
7095 | block_len, 0, | |
7096 | (1 << type) | | |
7097 | (1 << BTRFS_ORDERED_DIRECT), | |
7098 | BTRFS_COMPRESS_NONE); | |
5f9a8a51 FM |
7099 | if (ret) { |
7100 | if (em) { | |
7101 | free_extent_map(em); | |
64f54188 | 7102 | btrfs_drop_extent_cache(inode, start, start + len - 1, 0); |
5f9a8a51 FM |
7103 | } |
7104 | em = ERR_PTR(ret); | |
7105 | } | |
7106 | out: | |
5f9a8a51 FM |
7107 | |
7108 | return em; | |
7109 | } | |
7110 | ||
9fc6f911 | 7111 | static struct extent_map *btrfs_new_extent_direct(struct btrfs_inode *inode, |
4b46fce2 JB |
7112 | u64 start, u64 len) |
7113 | { | |
9fc6f911 NB |
7114 | struct btrfs_root *root = inode->root; |
7115 | struct btrfs_fs_info *fs_info = root->fs_info; | |
70c8a91c | 7116 | struct extent_map *em; |
4b46fce2 JB |
7117 | struct btrfs_key ins; |
7118 | u64 alloc_hint; | |
7119 | int ret; | |
4b46fce2 | 7120 | |
9fc6f911 | 7121 | alloc_hint = get_extent_allocation_hint(inode, start, len); |
0b246afa | 7122 | ret = btrfs_reserve_extent(root, len, len, fs_info->sectorsize, |
da17066c | 7123 | 0, alloc_hint, &ins, 1, 1); |
00361589 JB |
7124 | if (ret) |
7125 | return ERR_PTR(ret); | |
4b46fce2 | 7126 | |
9fc6f911 | 7127 | em = btrfs_create_dio_extent(inode, start, ins.offset, start, |
5f9a8a51 | 7128 | ins.objectid, ins.offset, ins.offset, |
6288d6ea | 7129 | ins.offset, BTRFS_ORDERED_REGULAR); |
0b246afa | 7130 | btrfs_dec_block_group_reservations(fs_info, ins.objectid); |
5f9a8a51 | 7131 | if (IS_ERR(em)) |
9fc6f911 NB |
7132 | btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, |
7133 | 1); | |
de0ee0ed | 7134 | |
4b46fce2 JB |
7135 | return em; |
7136 | } | |
7137 | ||
f4639636 | 7138 | static bool btrfs_extent_readonly(struct btrfs_fs_info *fs_info, u64 bytenr) |
05947ae1 AJ |
7139 | { |
7140 | struct btrfs_block_group *block_group; | |
f4639636 | 7141 | bool readonly = false; |
05947ae1 AJ |
7142 | |
7143 | block_group = btrfs_lookup_block_group(fs_info, bytenr); | |
7144 | if (!block_group || block_group->ro) | |
f4639636 | 7145 | readonly = true; |
05947ae1 AJ |
7146 | if (block_group) |
7147 | btrfs_put_block_group(block_group); | |
7148 | return readonly; | |
7149 | } | |
7150 | ||
46bfbb5c | 7151 | /* |
e4ecaf90 QW |
7152 | * Check if we can do nocow write into the range [@offset, @offset + @len) |
7153 | * | |
7154 | * @offset: File offset | |
7155 | * @len: The length to write, will be updated to the nocow writeable | |
7156 | * range | |
7157 | * @orig_start: (optional) Return the original file offset of the file extent | |
7158 | * @orig_len: (optional) Return the original on-disk length of the file extent | |
7159 | * @ram_bytes: (optional) Return the ram_bytes of the file extent | |
a84d5d42 BB |
7160 | * @strict: if true, omit optimizations that might force us into unnecessary |
7161 | * cow. e.g., don't trust generation number. | |
e4ecaf90 | 7162 | * |
e4ecaf90 QW |
7163 | * Return: |
7164 | * >0 and update @len if we can do nocow write | |
7165 | * 0 if we can't do nocow write | |
7166 | * <0 if error happened | |
7167 | * | |
7168 | * NOTE: This only checks the file extents, caller is responsible to wait for | |
7169 | * any ordered extents. | |
46bfbb5c | 7170 | */ |
00361589 | 7171 | noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len, |
7ee9e440 | 7172 | u64 *orig_start, u64 *orig_block_len, |
26ce9114 | 7173 | u64 *ram_bytes, bool nowait, bool strict) |
46bfbb5c | 7174 | { |
2ff7e61e | 7175 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
619104ba | 7176 | struct can_nocow_file_extent_args nocow_args = { 0 }; |
46bfbb5c CM |
7177 | struct btrfs_path *path; |
7178 | int ret; | |
7179 | struct extent_buffer *leaf; | |
7180 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
7b2b7085 | 7181 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
46bfbb5c CM |
7182 | struct btrfs_file_extent_item *fi; |
7183 | struct btrfs_key key; | |
46bfbb5c | 7184 | int found_type; |
e77751aa | 7185 | |
46bfbb5c CM |
7186 | path = btrfs_alloc_path(); |
7187 | if (!path) | |
7188 | return -ENOMEM; | |
26ce9114 | 7189 | path->nowait = nowait; |
46bfbb5c | 7190 | |
f85b7379 DS |
7191 | ret = btrfs_lookup_file_extent(NULL, root, path, |
7192 | btrfs_ino(BTRFS_I(inode)), offset, 0); | |
46bfbb5c CM |
7193 | if (ret < 0) |
7194 | goto out; | |
7195 | ||
46bfbb5c | 7196 | if (ret == 1) { |
619104ba | 7197 | if (path->slots[0] == 0) { |
46bfbb5c CM |
7198 | /* can't find the item, must cow */ |
7199 | ret = 0; | |
7200 | goto out; | |
7201 | } | |
619104ba | 7202 | path->slots[0]--; |
46bfbb5c CM |
7203 | } |
7204 | ret = 0; | |
7205 | leaf = path->nodes[0]; | |
619104ba | 7206 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
4a0cc7ca | 7207 | if (key.objectid != btrfs_ino(BTRFS_I(inode)) || |
46bfbb5c CM |
7208 | key.type != BTRFS_EXTENT_DATA_KEY) { |
7209 | /* not our file or wrong item type, must cow */ | |
7210 | goto out; | |
7211 | } | |
7212 | ||
7213 | if (key.offset > offset) { | |
7214 | /* Wrong offset, must cow */ | |
7215 | goto out; | |
7216 | } | |
7217 | ||
619104ba | 7218 | if (btrfs_file_extent_end(path) <= offset) |
7ee9e440 JB |
7219 | goto out; |
7220 | ||
619104ba FM |
7221 | fi = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_file_extent_item); |
7222 | found_type = btrfs_file_extent_type(leaf, fi); | |
7223 | if (ram_bytes) | |
7224 | *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi); | |
e77751aa | 7225 | |
619104ba FM |
7226 | nocow_args.start = offset; |
7227 | nocow_args.end = offset + *len - 1; | |
7228 | nocow_args.strict = strict; | |
7229 | nocow_args.free_path = true; | |
7ee9e440 | 7230 | |
619104ba FM |
7231 | ret = can_nocow_file_extent(path, &key, BTRFS_I(inode), &nocow_args); |
7232 | /* can_nocow_file_extent() has freed the path. */ | |
7233 | path = NULL; | |
7ee9e440 | 7234 | |
619104ba FM |
7235 | if (ret != 1) { |
7236 | /* Treat errors as not being able to NOCOW. */ | |
7237 | ret = 0; | |
78d4295b | 7238 | goto out; |
7ee9e440 | 7239 | } |
eb384b55 | 7240 | |
619104ba FM |
7241 | ret = 0; |
7242 | if (btrfs_extent_readonly(fs_info, nocow_args.disk_bytenr)) | |
46bfbb5c | 7243 | goto out; |
7b2b7085 | 7244 | |
619104ba FM |
7245 | if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) && |
7246 | found_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
7b2b7085 MX |
7247 | u64 range_end; |
7248 | ||
619104ba | 7249 | range_end = round_up(offset + nocow_args.num_bytes, |
da17066c | 7250 | root->fs_info->sectorsize) - 1; |
7b2b7085 MX |
7251 | ret = test_range_bit(io_tree, offset, range_end, |
7252 | EXTENT_DELALLOC, 0, NULL); | |
7253 | if (ret) { | |
7254 | ret = -EAGAIN; | |
7255 | goto out; | |
7256 | } | |
7257 | } | |
7258 | ||
619104ba FM |
7259 | if (orig_start) |
7260 | *orig_start = key.offset - nocow_args.extent_offset; | |
7261 | if (orig_block_len) | |
7262 | *orig_block_len = nocow_args.disk_num_bytes; | |
00361589 | 7263 | |
619104ba | 7264 | *len = nocow_args.num_bytes; |
46bfbb5c CM |
7265 | ret = 1; |
7266 | out: | |
7267 | btrfs_free_path(path); | |
7268 | return ret; | |
7269 | } | |
7270 | ||
eb838e73 | 7271 | static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend, |
59094403 FM |
7272 | struct extent_state **cached_state, |
7273 | unsigned int iomap_flags) | |
eb838e73 | 7274 | { |
59094403 FM |
7275 | const bool writing = (iomap_flags & IOMAP_WRITE); |
7276 | const bool nowait = (iomap_flags & IOMAP_NOWAIT); | |
7277 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; | |
eb838e73 JB |
7278 | struct btrfs_ordered_extent *ordered; |
7279 | int ret = 0; | |
7280 | ||
7281 | while (1) { | |
59094403 FM |
7282 | if (nowait) { |
7283 | if (!try_lock_extent(io_tree, lockstart, lockend)) | |
7284 | return -EAGAIN; | |
7285 | } else { | |
570eb97b | 7286 | lock_extent(io_tree, lockstart, lockend, cached_state); |
59094403 | 7287 | } |
eb838e73 JB |
7288 | /* |
7289 | * We're concerned with the entire range that we're going to be | |
01327610 | 7290 | * doing DIO to, so we need to make sure there's no ordered |
eb838e73 JB |
7291 | * extents in this range. |
7292 | */ | |
a776c6fa | 7293 | ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), lockstart, |
eb838e73 JB |
7294 | lockend - lockstart + 1); |
7295 | ||
7296 | /* | |
7297 | * We need to make sure there are no buffered pages in this | |
7298 | * range either, we could have raced between the invalidate in | |
7299 | * generic_file_direct_write and locking the extent. The | |
7300 | * invalidate needs to happen so that reads after a write do not | |
7301 | * get stale data. | |
7302 | */ | |
fc4adbff | 7303 | if (!ordered && |
051c98eb DS |
7304 | (!writing || !filemap_range_has_page(inode->i_mapping, |
7305 | lockstart, lockend))) | |
eb838e73 JB |
7306 | break; |
7307 | ||
570eb97b | 7308 | unlock_extent(io_tree, lockstart, lockend, cached_state); |
eb838e73 JB |
7309 | |
7310 | if (ordered) { | |
59094403 FM |
7311 | if (nowait) { |
7312 | btrfs_put_ordered_extent(ordered); | |
7313 | ret = -EAGAIN; | |
7314 | break; | |
7315 | } | |
ade77029 FM |
7316 | /* |
7317 | * If we are doing a DIO read and the ordered extent we | |
7318 | * found is for a buffered write, we can not wait for it | |
7319 | * to complete and retry, because if we do so we can | |
7320 | * deadlock with concurrent buffered writes on page | |
7321 | * locks. This happens only if our DIO read covers more | |
7322 | * than one extent map, if at this point has already | |
7323 | * created an ordered extent for a previous extent map | |
7324 | * and locked its range in the inode's io tree, and a | |
7325 | * concurrent write against that previous extent map's | |
7326 | * range and this range started (we unlock the ranges | |
7327 | * in the io tree only when the bios complete and | |
7328 | * buffered writes always lock pages before attempting | |
7329 | * to lock range in the io tree). | |
7330 | */ | |
7331 | if (writing || | |
7332 | test_bit(BTRFS_ORDERED_DIRECT, &ordered->flags)) | |
c0a43603 | 7333 | btrfs_start_ordered_extent(ordered, 1); |
ade77029 | 7334 | else |
59094403 | 7335 | ret = nowait ? -EAGAIN : -ENOTBLK; |
eb838e73 JB |
7336 | btrfs_put_ordered_extent(ordered); |
7337 | } else { | |
eb838e73 | 7338 | /* |
b850ae14 FM |
7339 | * We could trigger writeback for this range (and wait |
7340 | * for it to complete) and then invalidate the pages for | |
7341 | * this range (through invalidate_inode_pages2_range()), | |
7342 | * but that can lead us to a deadlock with a concurrent | |
ba206a02 | 7343 | * call to readahead (a buffered read or a defrag call |
b850ae14 FM |
7344 | * triggered a readahead) on a page lock due to an |
7345 | * ordered dio extent we created before but did not have | |
7346 | * yet a corresponding bio submitted (whence it can not | |
ba206a02 | 7347 | * complete), which makes readahead wait for that |
b850ae14 FM |
7348 | * ordered extent to complete while holding a lock on |
7349 | * that page. | |
eb838e73 | 7350 | */ |
59094403 | 7351 | ret = nowait ? -EAGAIN : -ENOTBLK; |
eb838e73 JB |
7352 | } |
7353 | ||
ade77029 FM |
7354 | if (ret) |
7355 | break; | |
7356 | ||
eb838e73 JB |
7357 | cond_resched(); |
7358 | } | |
7359 | ||
7360 | return ret; | |
7361 | } | |
7362 | ||
6f9994db | 7363 | /* The callers of this must take lock_extent() */ |
4b67c11d NB |
7364 | static struct extent_map *create_io_em(struct btrfs_inode *inode, u64 start, |
7365 | u64 len, u64 orig_start, u64 block_start, | |
6f9994db LB |
7366 | u64 block_len, u64 orig_block_len, |
7367 | u64 ram_bytes, int compress_type, | |
7368 | int type) | |
69ffb543 JB |
7369 | { |
7370 | struct extent_map_tree *em_tree; | |
7371 | struct extent_map *em; | |
69ffb543 JB |
7372 | int ret; |
7373 | ||
6f9994db LB |
7374 | ASSERT(type == BTRFS_ORDERED_PREALLOC || |
7375 | type == BTRFS_ORDERED_COMPRESSED || | |
7376 | type == BTRFS_ORDERED_NOCOW || | |
1af4a0aa | 7377 | type == BTRFS_ORDERED_REGULAR); |
6f9994db | 7378 | |
4b67c11d | 7379 | em_tree = &inode->extent_tree; |
69ffb543 JB |
7380 | em = alloc_extent_map(); |
7381 | if (!em) | |
7382 | return ERR_PTR(-ENOMEM); | |
7383 | ||
7384 | em->start = start; | |
7385 | em->orig_start = orig_start; | |
7386 | em->len = len; | |
7387 | em->block_len = block_len; | |
7388 | em->block_start = block_start; | |
b4939680 | 7389 | em->orig_block_len = orig_block_len; |
cc95bef6 | 7390 | em->ram_bytes = ram_bytes; |
70c8a91c | 7391 | em->generation = -1; |
69ffb543 | 7392 | set_bit(EXTENT_FLAG_PINNED, &em->flags); |
1af4a0aa | 7393 | if (type == BTRFS_ORDERED_PREALLOC) { |
b11e234d | 7394 | set_bit(EXTENT_FLAG_FILLING, &em->flags); |
1af4a0aa | 7395 | } else if (type == BTRFS_ORDERED_COMPRESSED) { |
6f9994db LB |
7396 | set_bit(EXTENT_FLAG_COMPRESSED, &em->flags); |
7397 | em->compress_type = compress_type; | |
7398 | } | |
69ffb543 JB |
7399 | |
7400 | do { | |
4b67c11d NB |
7401 | btrfs_drop_extent_cache(inode, em->start, |
7402 | em->start + em->len - 1, 0); | |
69ffb543 | 7403 | write_lock(&em_tree->lock); |
09a2a8f9 | 7404 | ret = add_extent_mapping(em_tree, em, 1); |
69ffb543 | 7405 | write_unlock(&em_tree->lock); |
6f9994db LB |
7406 | /* |
7407 | * The caller has taken lock_extent(), who could race with us | |
7408 | * to add em? | |
7409 | */ | |
69ffb543 JB |
7410 | } while (ret == -EEXIST); |
7411 | ||
7412 | if (ret) { | |
7413 | free_extent_map(em); | |
7414 | return ERR_PTR(ret); | |
7415 | } | |
7416 | ||
6f9994db | 7417 | /* em got 2 refs now, callers needs to do free_extent_map once. */ |
69ffb543 JB |
7418 | return em; |
7419 | } | |
7420 | ||
1c8d0175 | 7421 | |
c5794e51 | 7422 | static int btrfs_get_blocks_direct_write(struct extent_map **map, |
c5794e51 NB |
7423 | struct inode *inode, |
7424 | struct btrfs_dio_data *dio_data, | |
d7a8ab4e FM |
7425 | u64 start, u64 len, |
7426 | unsigned int iomap_flags) | |
c5794e51 | 7427 | { |
d4135134 | 7428 | const bool nowait = (iomap_flags & IOMAP_NOWAIT); |
c5794e51 NB |
7429 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
7430 | struct extent_map *em = *map; | |
f0bfa76a FM |
7431 | int type; |
7432 | u64 block_start, orig_start, orig_block_len, ram_bytes; | |
2306e83e | 7433 | struct btrfs_block_group *bg; |
f0bfa76a FM |
7434 | bool can_nocow = false; |
7435 | bool space_reserved = false; | |
6d82ad13 | 7436 | u64 prev_len; |
c5794e51 NB |
7437 | int ret = 0; |
7438 | ||
7439 | /* | |
7440 | * We don't allocate a new extent in the following cases | |
7441 | * | |
7442 | * 1) The inode is marked as NODATACOW. In this case we'll just use the | |
7443 | * existing extent. | |
7444 | * 2) The extent is marked as PREALLOC. We're good to go here and can | |
7445 | * just use the extent. | |
7446 | * | |
7447 | */ | |
7448 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) || | |
7449 | ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) && | |
7450 | em->block_start != EXTENT_MAP_HOLE)) { | |
c5794e51 NB |
7451 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) |
7452 | type = BTRFS_ORDERED_PREALLOC; | |
7453 | else | |
7454 | type = BTRFS_ORDERED_NOCOW; | |
7455 | len = min(len, em->len - (start - em->start)); | |
7456 | block_start = em->block_start + (start - em->start); | |
7457 | ||
7458 | if (can_nocow_extent(inode, start, &len, &orig_start, | |
26ce9114 | 7459 | &orig_block_len, &ram_bytes, false, false) == 1) { |
2306e83e FM |
7460 | bg = btrfs_inc_nocow_writers(fs_info, block_start); |
7461 | if (bg) | |
7462 | can_nocow = true; | |
7463 | } | |
f0bfa76a | 7464 | } |
c5794e51 | 7465 | |
6d82ad13 | 7466 | prev_len = len; |
f0bfa76a FM |
7467 | if (can_nocow) { |
7468 | struct extent_map *em2; | |
7469 | ||
7470 | /* We can NOCOW, so only need to reserve metadata space. */ | |
d4135134 FM |
7471 | ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode), len, len, |
7472 | nowait); | |
f0bfa76a FM |
7473 | if (ret < 0) { |
7474 | /* Our caller expects us to free the input extent map. */ | |
7475 | free_extent_map(em); | |
7476 | *map = NULL; | |
2306e83e | 7477 | btrfs_dec_nocow_writers(bg); |
d4135134 FM |
7478 | if (nowait && (ret == -ENOSPC || ret == -EDQUOT)) |
7479 | ret = -EAGAIN; | |
f0bfa76a FM |
7480 | goto out; |
7481 | } | |
7482 | space_reserved = true; | |
7483 | ||
7484 | em2 = btrfs_create_dio_extent(BTRFS_I(inode), start, len, | |
7485 | orig_start, block_start, | |
7486 | len, orig_block_len, | |
7487 | ram_bytes, type); | |
2306e83e | 7488 | btrfs_dec_nocow_writers(bg); |
f0bfa76a FM |
7489 | if (type == BTRFS_ORDERED_PREALLOC) { |
7490 | free_extent_map(em); | |
c1867eb3 DS |
7491 | *map = em2; |
7492 | em = em2; | |
f0bfa76a | 7493 | } |
c5794e51 | 7494 | |
f0bfa76a FM |
7495 | if (IS_ERR(em2)) { |
7496 | ret = PTR_ERR(em2); | |
7497 | goto out; | |
c5794e51 | 7498 | } |
f5585f4f FM |
7499 | |
7500 | dio_data->nocow_done = true; | |
f0bfa76a | 7501 | } else { |
f0bfa76a FM |
7502 | /* Our caller expects us to free the input extent map. */ |
7503 | free_extent_map(em); | |
7504 | *map = NULL; | |
7505 | ||
d4135134 | 7506 | if (nowait) |
d7a8ab4e FM |
7507 | return -EAGAIN; |
7508 | ||
f5585f4f FM |
7509 | /* |
7510 | * If we could not allocate data space before locking the file | |
7511 | * range and we can't do a NOCOW write, then we have to fail. | |
7512 | */ | |
7513 | if (!dio_data->data_space_reserved) | |
7514 | return -ENOSPC; | |
7515 | ||
7516 | /* | |
7517 | * We have to COW and we have already reserved data space before, | |
7518 | * so now we reserve only metadata. | |
7519 | */ | |
7520 | ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode), len, len, | |
7521 | false); | |
f0bfa76a FM |
7522 | if (ret < 0) |
7523 | goto out; | |
7524 | space_reserved = true; | |
7525 | ||
7526 | em = btrfs_new_extent_direct(BTRFS_I(inode), start, len); | |
7527 | if (IS_ERR(em)) { | |
7528 | ret = PTR_ERR(em); | |
7529 | goto out; | |
7530 | } | |
7531 | *map = em; | |
7532 | len = min(len, em->len - (start - em->start)); | |
7533 | if (len < prev_len) | |
f5585f4f FM |
7534 | btrfs_delalloc_release_metadata(BTRFS_I(inode), |
7535 | prev_len - len, true); | |
c5794e51 NB |
7536 | } |
7537 | ||
f0bfa76a FM |
7538 | /* |
7539 | * We have created our ordered extent, so we can now release our reservation | |
7540 | * for an outstanding extent. | |
7541 | */ | |
6d82ad13 | 7542 | btrfs_delalloc_release_extents(BTRFS_I(inode), prev_len); |
c5794e51 | 7543 | |
c5794e51 NB |
7544 | /* |
7545 | * Need to update the i_size under the extent lock so buffered | |
7546 | * readers will get the updated i_size when we unlock. | |
7547 | */ | |
f85781fb | 7548 | if (start + len > i_size_read(inode)) |
c5794e51 | 7549 | i_size_write(inode, start + len); |
c5794e51 | 7550 | out: |
f0bfa76a FM |
7551 | if (ret && space_reserved) { |
7552 | btrfs_delalloc_release_extents(BTRFS_I(inode), len); | |
f5585f4f | 7553 | btrfs_delalloc_release_metadata(BTRFS_I(inode), len, true); |
f0bfa76a | 7554 | } |
c5794e51 NB |
7555 | return ret; |
7556 | } | |
7557 | ||
f85781fb GR |
7558 | static int btrfs_dio_iomap_begin(struct inode *inode, loff_t start, |
7559 | loff_t length, unsigned int flags, struct iomap *iomap, | |
7560 | struct iomap *srcmap) | |
4b46fce2 | 7561 | { |
491a6d01 | 7562 | struct iomap_iter *iter = container_of(iomap, struct iomap_iter, iomap); |
0b246afa | 7563 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
4b46fce2 | 7564 | struct extent_map *em; |
eb838e73 | 7565 | struct extent_state *cached_state = NULL; |
491a6d01 | 7566 | struct btrfs_dio_data *dio_data = iter->private; |
eb838e73 | 7567 | u64 lockstart, lockend; |
f85781fb | 7568 | const bool write = !!(flags & IOMAP_WRITE); |
0934856d | 7569 | int ret = 0; |
f85781fb | 7570 | u64 len = length; |
f5585f4f | 7571 | const u64 data_alloc_len = length; |
f85781fb | 7572 | bool unlock_extents = false; |
eb838e73 | 7573 | |
79d3d1d1 JB |
7574 | /* |
7575 | * We could potentially fault if we have a buffer > PAGE_SIZE, and if | |
7576 | * we're NOWAIT we may submit a bio for a partial range and return | |
7577 | * EIOCBQUEUED, which would result in an errant short read. | |
7578 | * | |
7579 | * The best way to handle this would be to allow for partial completions | |
7580 | * of iocb's, so we could submit the partial bio, return and fault in | |
7581 | * the rest of the pages, and then submit the io for the rest of the | |
7582 | * range. However we don't have that currently, so simply return | |
7583 | * -EAGAIN at this point so that the normal path is used. | |
7584 | */ | |
7585 | if (!write && (flags & IOMAP_NOWAIT) && length > PAGE_SIZE) | |
7586 | return -EAGAIN; | |
7587 | ||
ee5b46a3 CH |
7588 | /* |
7589 | * Cap the size of reads to that usually seen in buffered I/O as we need | |
7590 | * to allocate a contiguous array for the checksums. | |
7591 | */ | |
f85781fb | 7592 | if (!write) |
ee5b46a3 | 7593 | len = min_t(u64, len, fs_info->sectorsize * BTRFS_MAX_BIO_SECTORS); |
eb838e73 | 7594 | |
c329861d JB |
7595 | lockstart = start; |
7596 | lockend = start + len - 1; | |
7597 | ||
f85781fb | 7598 | /* |
b023e675 FM |
7599 | * iomap_dio_rw() only does filemap_write_and_wait_range(), which isn't |
7600 | * enough if we've written compressed pages to this area, so we need to | |
7601 | * flush the dirty pages again to make absolutely sure that any | |
7602 | * outstanding dirty pages are on disk - the first flush only starts | |
7603 | * compression on the data, while keeping the pages locked, so by the | |
7604 | * time the second flush returns we know bios for the compressed pages | |
7605 | * were submitted and finished, and the pages no longer under writeback. | |
7606 | * | |
7607 | * If we have a NOWAIT request and we have any pages in the range that | |
7608 | * are locked, likely due to compression still in progress, we don't want | |
7609 | * to block on page locks. We also don't want to block on pages marked as | |
7610 | * dirty or under writeback (same as for the non-compression case). | |
7611 | * iomap_dio_rw() did the same check, but after that and before we got | |
7612 | * here, mmap'ed writes may have happened or buffered reads started | |
7613 | * (readpage() and readahead(), which lock pages), as we haven't locked | |
7614 | * the file range yet. | |
f85781fb GR |
7615 | */ |
7616 | if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT, | |
7617 | &BTRFS_I(inode)->runtime_flags)) { | |
b023e675 FM |
7618 | if (flags & IOMAP_NOWAIT) { |
7619 | if (filemap_range_needs_writeback(inode->i_mapping, | |
7620 | lockstart, lockend)) | |
7621 | return -EAGAIN; | |
7622 | } else { | |
7623 | ret = filemap_fdatawrite_range(inode->i_mapping, start, | |
7624 | start + length - 1); | |
7625 | if (ret) | |
7626 | return ret; | |
7627 | } | |
f85781fb GR |
7628 | } |
7629 | ||
491a6d01 | 7630 | memset(dio_data, 0, sizeof(*dio_data)); |
f85781fb | 7631 | |
f5585f4f FM |
7632 | /* |
7633 | * We always try to allocate data space and must do it before locking | |
7634 | * the file range, to avoid deadlocks with concurrent writes to the same | |
7635 | * range if the range has several extents and the writes don't expand the | |
7636 | * current i_size (the inode lock is taken in shared mode). If we fail to | |
7637 | * allocate data space here we continue and later, after locking the | |
7638 | * file range, we fail with ENOSPC only if we figure out we can not do a | |
7639 | * NOCOW write. | |
7640 | */ | |
7641 | if (write && !(flags & IOMAP_NOWAIT)) { | |
7642 | ret = btrfs_check_data_free_space(BTRFS_I(inode), | |
7643 | &dio_data->data_reserved, | |
7644 | start, data_alloc_len); | |
7645 | if (!ret) | |
7646 | dio_data->data_space_reserved = true; | |
7647 | else if (ret && !(BTRFS_I(inode)->flags & | |
7648 | (BTRFS_INODE_NODATACOW | BTRFS_INODE_PREALLOC))) | |
7649 | goto err; | |
7650 | } | |
e1cbbfa5 | 7651 | |
eb838e73 JB |
7652 | /* |
7653 | * If this errors out it's because we couldn't invalidate pagecache for | |
59094403 FM |
7654 | * this range and we need to fallback to buffered IO, or we are doing a |
7655 | * NOWAIT read/write and we need to block. | |
eb838e73 | 7656 | */ |
59094403 FM |
7657 | ret = lock_extent_direct(inode, lockstart, lockend, &cached_state, flags); |
7658 | if (ret < 0) | |
9c9464cc | 7659 | goto err; |
eb838e73 | 7660 | |
39b07b5d | 7661 | em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len); |
eb838e73 JB |
7662 | if (IS_ERR(em)) { |
7663 | ret = PTR_ERR(em); | |
7664 | goto unlock_err; | |
7665 | } | |
4b46fce2 JB |
7666 | |
7667 | /* | |
7668 | * Ok for INLINE and COMPRESSED extents we need to fallback on buffered | |
7669 | * io. INLINE is special, and we could probably kludge it in here, but | |
7670 | * it's still buffered so for safety lets just fall back to the generic | |
7671 | * buffered path. | |
7672 | * | |
7673 | * For COMPRESSED we _have_ to read the entire extent in so we can | |
7674 | * decompress it, so there will be buffering required no matter what we | |
7675 | * do, so go ahead and fallback to buffered. | |
7676 | * | |
01327610 | 7677 | * We return -ENOTBLK because that's what makes DIO go ahead and go back |
4b46fce2 JB |
7678 | * to buffered IO. Don't blame me, this is the price we pay for using |
7679 | * the generic code. | |
7680 | */ | |
7681 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) || | |
7682 | em->block_start == EXTENT_MAP_INLINE) { | |
7683 | free_extent_map(em); | |
a4527e18 FM |
7684 | /* |
7685 | * If we are in a NOWAIT context, return -EAGAIN in order to | |
7686 | * fallback to buffered IO. This is not only because we can | |
7687 | * block with buffered IO (no support for NOWAIT semantics at | |
7688 | * the moment) but also to avoid returning short reads to user | |
7689 | * space - this happens if we were able to read some data from | |
7690 | * previous non-compressed extents and then when we fallback to | |
7691 | * buffered IO, at btrfs_file_read_iter() by calling | |
7692 | * filemap_read(), we fail to fault in pages for the read buffer, | |
7693 | * in which case filemap_read() returns a short read (the number | |
7694 | * of bytes previously read is > 0, so it does not return -EFAULT). | |
7695 | */ | |
7696 | ret = (flags & IOMAP_NOWAIT) ? -EAGAIN : -ENOTBLK; | |
eb838e73 | 7697 | goto unlock_err; |
4b46fce2 JB |
7698 | } |
7699 | ||
f85781fb | 7700 | len = min(len, em->len - (start - em->start)); |
ca93e44b FM |
7701 | |
7702 | /* | |
7703 | * If we have a NOWAIT request and the range contains multiple extents | |
7704 | * (or a mix of extents and holes), then we return -EAGAIN to make the | |
7705 | * caller fallback to a context where it can do a blocking (without | |
7706 | * NOWAIT) request. This way we avoid doing partial IO and returning | |
7707 | * success to the caller, which is not optimal for writes and for reads | |
7708 | * it can result in unexpected behaviour for an application. | |
7709 | * | |
7710 | * When doing a read, because we use IOMAP_DIO_PARTIAL when calling | |
7711 | * iomap_dio_rw(), we can end up returning less data then what the caller | |
7712 | * asked for, resulting in an unexpected, and incorrect, short read. | |
7713 | * That is, the caller asked to read N bytes and we return less than that, | |
7714 | * which is wrong unless we are crossing EOF. This happens if we get a | |
7715 | * page fault error when trying to fault in pages for the buffer that is | |
7716 | * associated to the struct iov_iter passed to iomap_dio_rw(), and we | |
7717 | * have previously submitted bios for other extents in the range, in | |
7718 | * which case iomap_dio_rw() may return us EIOCBQUEUED if not all of | |
7719 | * those bios have completed by the time we get the page fault error, | |
7720 | * which we return back to our caller - we should only return EIOCBQUEUED | |
7721 | * after we have submitted bios for all the extents in the range. | |
7722 | */ | |
7723 | if ((flags & IOMAP_NOWAIT) && len < length) { | |
7724 | free_extent_map(em); | |
7725 | ret = -EAGAIN; | |
7726 | goto unlock_err; | |
7727 | } | |
7728 | ||
f85781fb GR |
7729 | if (write) { |
7730 | ret = btrfs_get_blocks_direct_write(&em, inode, dio_data, | |
d7a8ab4e | 7731 | start, len, flags); |
c5794e51 NB |
7732 | if (ret < 0) |
7733 | goto unlock_err; | |
f85781fb GR |
7734 | unlock_extents = true; |
7735 | /* Recalc len in case the new em is smaller than requested */ | |
7736 | len = min(len, em->len - (start - em->start)); | |
f5585f4f FM |
7737 | if (dio_data->data_space_reserved) { |
7738 | u64 release_offset; | |
7739 | u64 release_len = 0; | |
7740 | ||
7741 | if (dio_data->nocow_done) { | |
7742 | release_offset = start; | |
7743 | release_len = data_alloc_len; | |
7744 | } else if (len < data_alloc_len) { | |
7745 | release_offset = start + len; | |
7746 | release_len = data_alloc_len - len; | |
7747 | } | |
7748 | ||
7749 | if (release_len > 0) | |
7750 | btrfs_free_reserved_data_space(BTRFS_I(inode), | |
7751 | dio_data->data_reserved, | |
7752 | release_offset, | |
7753 | release_len); | |
7754 | } | |
c5794e51 | 7755 | } else { |
1c8d0175 NB |
7756 | /* |
7757 | * We need to unlock only the end area that we aren't using. | |
7758 | * The rest is going to be unlocked by the endio routine. | |
7759 | */ | |
f85781fb GR |
7760 | lockstart = start + len; |
7761 | if (lockstart < lockend) | |
7762 | unlock_extents = true; | |
7763 | } | |
7764 | ||
7765 | if (unlock_extents) | |
570eb97b JB |
7766 | unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend, |
7767 | &cached_state); | |
f85781fb GR |
7768 | else |
7769 | free_extent_state(cached_state); | |
7770 | ||
7771 | /* | |
7772 | * Translate extent map information to iomap. | |
7773 | * We trim the extents (and move the addr) even though iomap code does | |
7774 | * that, since we have locked only the parts we are performing I/O in. | |
7775 | */ | |
7776 | if ((em->block_start == EXTENT_MAP_HOLE) || | |
7777 | (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) && !write)) { | |
7778 | iomap->addr = IOMAP_NULL_ADDR; | |
7779 | iomap->type = IOMAP_HOLE; | |
7780 | } else { | |
7781 | iomap->addr = em->block_start + (start - em->start); | |
7782 | iomap->type = IOMAP_MAPPED; | |
a43a67a2 | 7783 | } |
f85781fb | 7784 | iomap->offset = start; |
d24fa5c1 | 7785 | iomap->bdev = fs_info->fs_devices->latest_dev->bdev; |
f85781fb | 7786 | iomap->length = len; |
a43a67a2 | 7787 | |
e380adfc | 7788 | if (write && btrfs_use_zone_append(BTRFS_I(inode), em->block_start)) |
544d24f9 NA |
7789 | iomap->flags |= IOMAP_F_ZONE_APPEND; |
7790 | ||
4b46fce2 JB |
7791 | free_extent_map(em); |
7792 | ||
7793 | return 0; | |
eb838e73 JB |
7794 | |
7795 | unlock_err: | |
570eb97b JB |
7796 | unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend, |
7797 | &cached_state); | |
9c9464cc | 7798 | err: |
f5585f4f FM |
7799 | if (dio_data->data_space_reserved) { |
7800 | btrfs_free_reserved_data_space(BTRFS_I(inode), | |
7801 | dio_data->data_reserved, | |
7802 | start, data_alloc_len); | |
7803 | extent_changeset_free(dio_data->data_reserved); | |
7804 | } | |
7805 | ||
f85781fb GR |
7806 | return ret; |
7807 | } | |
7808 | ||
7809 | static int btrfs_dio_iomap_end(struct inode *inode, loff_t pos, loff_t length, | |
7810 | ssize_t written, unsigned int flags, struct iomap *iomap) | |
7811 | { | |
491a6d01 CH |
7812 | struct iomap_iter *iter = container_of(iomap, struct iomap_iter, iomap); |
7813 | struct btrfs_dio_data *dio_data = iter->private; | |
f85781fb GR |
7814 | size_t submitted = dio_data->submitted; |
7815 | const bool write = !!(flags & IOMAP_WRITE); | |
491a6d01 | 7816 | int ret = 0; |
f85781fb GR |
7817 | |
7818 | if (!write && (iomap->type == IOMAP_HOLE)) { | |
7819 | /* If reading from a hole, unlock and return */ | |
570eb97b JB |
7820 | unlock_extent(&BTRFS_I(inode)->io_tree, pos, pos + length - 1, |
7821 | NULL); | |
491a6d01 | 7822 | return 0; |
f85781fb GR |
7823 | } |
7824 | ||
7825 | if (submitted < length) { | |
7826 | pos += submitted; | |
7827 | length -= submitted; | |
7828 | if (write) | |
711f447b CH |
7829 | btrfs_mark_ordered_io_finished(BTRFS_I(inode), NULL, |
7830 | pos, length, false); | |
f85781fb GR |
7831 | else |
7832 | unlock_extent(&BTRFS_I(inode)->io_tree, pos, | |
570eb97b | 7833 | pos + length - 1, NULL); |
f85781fb GR |
7834 | ret = -ENOTBLK; |
7835 | } | |
7836 | ||
f0bfa76a | 7837 | if (write) |
f85781fb | 7838 | extent_changeset_free(dio_data->data_reserved); |
8b110e39 MX |
7839 | return ret; |
7840 | } | |
7841 | ||
769b4f24 | 7842 | static void btrfs_dio_private_put(struct btrfs_dio_private *dip) |
8b110e39 | 7843 | { |
769b4f24 OS |
7844 | /* |
7845 | * This implies a barrier so that stores to dio_bio->bi_status before | |
7846 | * this and loads of dio_bio->bi_status after this are fully ordered. | |
7847 | */ | |
7848 | if (!refcount_dec_and_test(&dip->refs)) | |
7849 | return; | |
8b110e39 | 7850 | |
642c5d34 | 7851 | if (btrfs_op(&dip->bio) == BTRFS_MAP_WRITE) { |
711f447b CH |
7852 | btrfs_mark_ordered_io_finished(BTRFS_I(dip->inode), NULL, |
7853 | dip->file_offset, dip->bytes, | |
7854 | !dip->bio.bi_status); | |
769b4f24 OS |
7855 | } else { |
7856 | unlock_extent(&BTRFS_I(dip->inode)->io_tree, | |
47926ab5 | 7857 | dip->file_offset, |
570eb97b | 7858 | dip->file_offset + dip->bytes - 1, NULL); |
8b110e39 MX |
7859 | } |
7860 | ||
642c5d34 CH |
7861 | kfree(dip->csums); |
7862 | bio_endio(&dip->bio); | |
8b110e39 MX |
7863 | } |
7864 | ||
ad357938 | 7865 | static void submit_dio_repair_bio(struct inode *inode, struct bio *bio, |
cb3a12d9 DS |
7866 | int mirror_num, |
7867 | enum btrfs_compression_type compress_type) | |
8b110e39 | 7868 | { |
917f32a2 | 7869 | struct btrfs_dio_private *dip = btrfs_bio(bio)->private; |
2ff7e61e | 7870 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
8b110e39 | 7871 | |
37226b21 | 7872 | BUG_ON(bio_op(bio) == REQ_OP_WRITE); |
8b110e39 | 7873 | |
77d5d689 | 7874 | refcount_inc(&dip->refs); |
1a722d8f | 7875 | btrfs_submit_bio(fs_info, bio, mirror_num); |
8b110e39 MX |
7876 | } |
7877 | ||
f4f39fc5 | 7878 | static blk_status_t btrfs_check_read_dio_bio(struct btrfs_dio_private *dip, |
c3a3b19b | 7879 | struct btrfs_bio *bbio, |
fd9d6670 | 7880 | const bool uptodate) |
4b46fce2 | 7881 | { |
f4f39fc5 | 7882 | struct inode *inode = dip->inode; |
fd9d6670 | 7883 | struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info; |
fd9d6670 | 7884 | const bool csum = !(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM); |
58efbc9f | 7885 | blk_status_t err = BLK_STS_OK; |
1e87770c CH |
7886 | struct bvec_iter iter; |
7887 | struct bio_vec bv; | |
7888 | u32 offset; | |
7889 | ||
7890 | btrfs_bio_for_each_sector(fs_info, bv, bbio, iter, offset) { | |
7891 | u64 start = bbio->file_offset + offset; | |
7892 | ||
7893 | if (uptodate && | |
7959bd44 CH |
7894 | (!csum || !btrfs_check_data_csum(inode, bbio, offset, bv.bv_page, |
7895 | bv.bv_offset))) { | |
0d0a762c JB |
7896 | btrfs_clean_io_failure(BTRFS_I(inode), start, |
7897 | bv.bv_page, bv.bv_offset); | |
1e87770c CH |
7898 | } else { |
7899 | int ret; | |
4b46fce2 | 7900 | |
7aa51232 CH |
7901 | ret = btrfs_repair_one_sector(inode, bbio, offset, |
7902 | bv.bv_page, bv.bv_offset, | |
1e87770c CH |
7903 | submit_dio_repair_bio); |
7904 | if (ret) | |
7905 | err = errno_to_blk_status(ret); | |
2dabb324 | 7906 | } |
2c30c71b | 7907 | } |
c1dc0896 | 7908 | |
c1dc0896 | 7909 | return err; |
14543774 FM |
7910 | } |
7911 | ||
8896a08d | 7912 | static blk_status_t btrfs_submit_bio_start_direct_io(struct inode *inode, |
1941b64b QW |
7913 | struct bio *bio, |
7914 | u64 dio_file_offset) | |
eaf25d93 | 7915 | { |
e331f6b1 | 7916 | return btrfs_csum_one_bio(BTRFS_I(inode), bio, dio_file_offset, false); |
eaf25d93 CM |
7917 | } |
7918 | ||
917f32a2 | 7919 | static void btrfs_end_dio_bio(struct btrfs_bio *bbio) |
e65e1535 | 7920 | { |
917f32a2 CH |
7921 | struct btrfs_dio_private *dip = bbio->private; |
7922 | struct bio *bio = &bbio->bio; | |
4e4cbee9 | 7923 | blk_status_t err = bio->bi_status; |
e65e1535 | 7924 | |
8b110e39 MX |
7925 | if (err) |
7926 | btrfs_warn(BTRFS_I(dip->inode)->root->fs_info, | |
6296b960 | 7927 | "direct IO failed ino %llu rw %d,%u sector %#Lx len %u err no %d", |
f85b7379 | 7928 | btrfs_ino(BTRFS_I(dip->inode)), bio_op(bio), |
1201b58b | 7929 | bio->bi_opf, bio->bi_iter.bi_sector, |
8b110e39 MX |
7930 | bio->bi_iter.bi_size, err); |
7931 | ||
f4f39fc5 | 7932 | if (bio_op(bio) == REQ_OP_READ) |
0fdf977d | 7933 | err = btrfs_check_read_dio_bio(dip, bbio, !err); |
e65e1535 | 7934 | |
769b4f24 | 7935 | if (err) |
642c5d34 | 7936 | dip->bio.bi_status = err; |
e65e1535 | 7937 | |
0fdf977d | 7938 | btrfs_record_physical_zoned(dip->inode, bbio->file_offset, bio); |
544d24f9 | 7939 | |
e65e1535 | 7940 | bio_put(bio); |
769b4f24 | 7941 | btrfs_dio_private_put(dip); |
c1dc0896 MX |
7942 | } |
7943 | ||
37899117 CH |
7944 | static void btrfs_submit_dio_bio(struct bio *bio, struct inode *inode, |
7945 | u64 file_offset, int async_submit) | |
e65e1535 | 7946 | { |
0b246afa | 7947 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
917f32a2 | 7948 | struct btrfs_dio_private *dip = btrfs_bio(bio)->private; |
4e4cbee9 | 7949 | blk_status_t ret; |
e65e1535 | 7950 | |
81bd9328 CH |
7951 | /* Save the original iter for read repair */ |
7952 | if (btrfs_op(bio) == BTRFS_MAP_READ) | |
7953 | btrfs_bio(bio)->iter = bio->bi_iter; | |
e65e1535 | 7954 | |
e6961cac | 7955 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM) |
1ae39938 JB |
7956 | goto map; |
7957 | ||
d7b9416f | 7958 | if (btrfs_op(bio) == BTRFS_MAP_WRITE) { |
c93104e7 | 7959 | /* Check btrfs_submit_data_write_bio() for async submit rules */ |
ea1f0ced CH |
7960 | if (async_submit && !atomic_read(&BTRFS_I(inode)->sync_writers) && |
7961 | btrfs_wq_submit_bio(inode, bio, 0, file_offset, | |
7962 | btrfs_submit_bio_start_direct_io)) | |
37899117 | 7963 | return; |
ea1f0ced | 7964 | |
1ae39938 JB |
7965 | /* |
7966 | * If we aren't doing async submit, calculate the csum of the | |
7967 | * bio now. | |
7968 | */ | |
e331f6b1 | 7969 | ret = btrfs_csum_one_bio(BTRFS_I(inode), bio, file_offset, false); |
37899117 | 7970 | if (ret) { |
917f32a2 | 7971 | btrfs_bio_end_io(btrfs_bio(bio), ret); |
37899117 CH |
7972 | return; |
7973 | } | |
23ea8e5a | 7974 | } else { |
a89ce08c CH |
7975 | btrfs_bio(bio)->csum = btrfs_csum_ptr(fs_info, dip->csums, |
7976 | file_offset - dip->file_offset); | |
c2db1073 | 7977 | } |
1ae39938 | 7978 | map: |
1a722d8f | 7979 | btrfs_submit_bio(fs_info, bio, 0); |
e65e1535 MX |
7980 | } |
7981 | ||
3e08773c | 7982 | static void btrfs_submit_direct(const struct iomap_iter *iter, |
f85781fb | 7983 | struct bio *dio_bio, loff_t file_offset) |
c36cac28 | 7984 | { |
642c5d34 CH |
7985 | struct btrfs_dio_private *dip = |
7986 | container_of(dio_bio, struct btrfs_dio_private, bio); | |
a6d3d495 | 7987 | struct inode *inode = iter->inode; |
cfe94440 | 7988 | const bool write = (btrfs_op(dio_bio) == BTRFS_MAP_WRITE); |
0b246afa | 7989 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
769b4f24 OS |
7990 | const bool raid56 = (btrfs_data_alloc_profile(fs_info) & |
7991 | BTRFS_BLOCK_GROUP_RAID56_MASK); | |
e65e1535 | 7992 | struct bio *bio; |
c36cac28 | 7993 | u64 start_sector; |
1ae39938 | 7994 | int async_submit = 0; |
725130ba | 7995 | u64 submit_len; |
42b5d73b NA |
7996 | u64 clone_offset = 0; |
7997 | u64 clone_len; | |
42034313 | 7998 | u64 logical; |
5f4dc8fc | 7999 | int ret; |
58efbc9f | 8000 | blk_status_t status; |
89b798ad | 8001 | struct btrfs_io_geometry geom; |
491a6d01 | 8002 | struct btrfs_dio_data *dio_data = iter->private; |
42034313 | 8003 | struct extent_map *em = NULL; |
e65e1535 | 8004 | |
642c5d34 CH |
8005 | dip->inode = inode; |
8006 | dip->file_offset = file_offset; | |
8007 | dip->bytes = dio_bio->bi_iter.bi_size; | |
8008 | refcount_set(&dip->refs, 1); | |
8009 | dip->csums = NULL; | |
8010 | ||
8011 | if (!write && !(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) { | |
8012 | unsigned int nr_sectors = | |
8013 | (dio_bio->bi_iter.bi_size >> fs_info->sectorsize_bits); | |
facc8a22 | 8014 | |
85879573 OS |
8015 | /* |
8016 | * Load the csums up front to reduce csum tree searches and | |
8017 | * contention when submitting bios. | |
8018 | */ | |
642c5d34 CH |
8019 | status = BLK_STS_RESOURCE; |
8020 | dip->csums = kcalloc(nr_sectors, fs_info->csum_size, GFP_NOFS); | |
8021 | if (!dip) | |
8022 | goto out_err; | |
8023 | ||
6275193e | 8024 | status = btrfs_lookup_bio_sums(inode, dio_bio, dip->csums); |
85879573 OS |
8025 | if (status != BLK_STS_OK) |
8026 | goto out_err; | |
02f57c7a JB |
8027 | } |
8028 | ||
769b4f24 OS |
8029 | start_sector = dio_bio->bi_iter.bi_sector; |
8030 | submit_len = dio_bio->bi_iter.bi_size; | |
53b381b3 | 8031 | |
3c91ee69 | 8032 | do { |
42034313 MR |
8033 | logical = start_sector << 9; |
8034 | em = btrfs_get_chunk_map(fs_info, logical, submit_len); | |
8035 | if (IS_ERR(em)) { | |
8036 | status = errno_to_blk_status(PTR_ERR(em)); | |
8037 | em = NULL; | |
8038 | goto out_err_em; | |
8039 | } | |
8040 | ret = btrfs_get_io_geometry(fs_info, em, btrfs_op(dio_bio), | |
43c0d1a5 | 8041 | logical, &geom); |
769b4f24 OS |
8042 | if (ret) { |
8043 | status = errno_to_blk_status(ret); | |
42034313 | 8044 | goto out_err_em; |
769b4f24 | 8045 | } |
769b4f24 | 8046 | |
42b5d73b NA |
8047 | clone_len = min(submit_len, geom.len); |
8048 | ASSERT(clone_len <= UINT_MAX); | |
02f57c7a | 8049 | |
725130ba LB |
8050 | /* |
8051 | * This will never fail as it's passing GPF_NOFS and | |
8052 | * the allocation is backed by btrfs_bioset. | |
8053 | */ | |
917f32a2 CH |
8054 | bio = btrfs_bio_clone_partial(dio_bio, clone_offset, clone_len, |
8055 | btrfs_end_dio_bio, dip); | |
00d82525 | 8056 | btrfs_bio(bio)->file_offset = file_offset; |
725130ba | 8057 | |
544d24f9 NA |
8058 | if (bio_op(bio) == REQ_OP_ZONE_APPEND) { |
8059 | status = extract_ordered_extent(BTRFS_I(inode), bio, | |
8060 | file_offset); | |
8061 | if (status) { | |
8062 | bio_put(bio); | |
8063 | goto out_err; | |
8064 | } | |
8065 | } | |
8066 | ||
725130ba LB |
8067 | ASSERT(submit_len >= clone_len); |
8068 | submit_len -= clone_len; | |
e65e1535 | 8069 | |
725130ba LB |
8070 | /* |
8071 | * Increase the count before we submit the bio so we know | |
8072 | * the end IO handler won't happen before we increase the | |
8073 | * count. Otherwise, the dip might get freed before we're | |
8074 | * done setting it up. | |
769b4f24 OS |
8075 | * |
8076 | * We transfer the initial reference to the last bio, so we | |
8077 | * don't need to increment the reference count for the last one. | |
725130ba | 8078 | */ |
769b4f24 OS |
8079 | if (submit_len > 0) { |
8080 | refcount_inc(&dip->refs); | |
8081 | /* | |
8082 | * If we are submitting more than one bio, submit them | |
8083 | * all asynchronously. The exception is RAID 5 or 6, as | |
8084 | * asynchronous checksums make it difficult to collect | |
8085 | * full stripe writes. | |
8086 | */ | |
8087 | if (!raid56) | |
8088 | async_submit = 1; | |
8089 | } | |
e65e1535 | 8090 | |
37899117 | 8091 | btrfs_submit_dio_bio(bio, inode, file_offset, async_submit); |
e65e1535 | 8092 | |
f85781fb | 8093 | dio_data->submitted += clone_len; |
725130ba LB |
8094 | clone_offset += clone_len; |
8095 | start_sector += clone_len >> 9; | |
8096 | file_offset += clone_len; | |
42034313 MR |
8097 | |
8098 | free_extent_map(em); | |
3c91ee69 | 8099 | } while (submit_len > 0); |
3e08773c | 8100 | return; |
e65e1535 | 8101 | |
42034313 MR |
8102 | out_err_em: |
8103 | free_extent_map(em); | |
e65e1535 | 8104 | out_err: |
642c5d34 | 8105 | dio_bio->bi_status = status; |
769b4f24 | 8106 | btrfs_dio_private_put(dip); |
4b46fce2 JB |
8107 | } |
8108 | ||
36e8c622 | 8109 | static const struct iomap_ops btrfs_dio_iomap_ops = { |
f85781fb GR |
8110 | .iomap_begin = btrfs_dio_iomap_begin, |
8111 | .iomap_end = btrfs_dio_iomap_end, | |
8112 | }; | |
8113 | ||
36e8c622 | 8114 | static const struct iomap_dio_ops btrfs_dio_ops = { |
f85781fb | 8115 | .submit_io = btrfs_submit_direct, |
642c5d34 | 8116 | .bio_set = &btrfs_dio_bioset, |
f85781fb GR |
8117 | }; |
8118 | ||
36e8c622 CH |
8119 | ssize_t btrfs_dio_rw(struct kiocb *iocb, struct iov_iter *iter, size_t done_before) |
8120 | { | |
491a6d01 CH |
8121 | struct btrfs_dio_data data; |
8122 | ||
36e8c622 | 8123 | return iomap_dio_rw(iocb, iter, &btrfs_dio_iomap_ops, &btrfs_dio_ops, |
eacdf4ea AV |
8124 | IOMAP_DIO_PARTIAL | IOMAP_DIO_NOSYNC, |
8125 | &data, done_before); | |
36e8c622 CH |
8126 | } |
8127 | ||
1506fcc8 | 8128 | static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, |
bab16e21 | 8129 | u64 start, u64 len) |
1506fcc8 | 8130 | { |
05dadc09 TI |
8131 | int ret; |
8132 | ||
45dd052e | 8133 | ret = fiemap_prep(inode, fieinfo, start, &len, 0); |
05dadc09 TI |
8134 | if (ret) |
8135 | return ret; | |
8136 | ||
33a86cfa FM |
8137 | /* |
8138 | * fiemap_prep() called filemap_write_and_wait() for the whole possible | |
8139 | * file range (0 to LLONG_MAX), but that is not enough if we have | |
8140 | * compression enabled. The first filemap_fdatawrite_range() only kicks | |
8141 | * in the compression of data (in an async thread) and will return | |
8142 | * before the compression is done and writeback is started. A second | |
8143 | * filemap_fdatawrite_range() is needed to wait for the compression to | |
ac3c0d36 FM |
8144 | * complete and writeback to start. We also need to wait for ordered |
8145 | * extents to complete, because our fiemap implementation uses mainly | |
8146 | * file extent items to list the extents, searching for extent maps | |
8147 | * only for file ranges with holes or prealloc extents to figure out | |
8148 | * if we have delalloc in those ranges. | |
33a86cfa FM |
8149 | */ |
8150 | if (fieinfo->fi_flags & FIEMAP_FLAG_SYNC) { | |
ac3c0d36 | 8151 | ret = btrfs_wait_ordered_range(inode, 0, LLONG_MAX); |
33a86cfa FM |
8152 | if (ret) |
8153 | return ret; | |
8154 | } | |
8155 | ||
facee0a0 | 8156 | return extent_fiemap(BTRFS_I(inode), fieinfo, start, len); |
1506fcc8 YS |
8157 | } |
8158 | ||
48a3b636 ES |
8159 | static int btrfs_writepages(struct address_space *mapping, |
8160 | struct writeback_control *wbc) | |
b293f02e | 8161 | { |
8ae225a8 | 8162 | return extent_writepages(mapping, wbc); |
b293f02e CM |
8163 | } |
8164 | ||
ba206a02 | 8165 | static void btrfs_readahead(struct readahead_control *rac) |
3ab2fb5a | 8166 | { |
ba206a02 | 8167 | extent_readahead(rac); |
3ab2fb5a | 8168 | } |
2a3ff0ad | 8169 | |
7c11d0ae | 8170 | /* |
f913cff3 | 8171 | * For release_folio() and invalidate_folio() we have a race window where |
895586eb | 8172 | * folio_end_writeback() is called but the subpage spinlock is not yet released. |
7c11d0ae QW |
8173 | * If we continue to release/invalidate the page, we could cause use-after-free |
8174 | * for subpage spinlock. So this function is to spin and wait for subpage | |
8175 | * spinlock. | |
8176 | */ | |
8177 | static void wait_subpage_spinlock(struct page *page) | |
8178 | { | |
8179 | struct btrfs_fs_info *fs_info = btrfs_sb(page->mapping->host->i_sb); | |
8180 | struct btrfs_subpage *subpage; | |
8181 | ||
fbca46eb | 8182 | if (!btrfs_is_subpage(fs_info, page)) |
7c11d0ae QW |
8183 | return; |
8184 | ||
8185 | ASSERT(PagePrivate(page) && page->private); | |
8186 | subpage = (struct btrfs_subpage *)page->private; | |
8187 | ||
8188 | /* | |
8189 | * This may look insane as we just acquire the spinlock and release it, | |
8190 | * without doing anything. But we just want to make sure no one is | |
8191 | * still holding the subpage spinlock. | |
8192 | * And since the page is not dirty nor writeback, and we have page | |
8193 | * locked, the only possible way to hold a spinlock is from the endio | |
8194 | * function to clear page writeback. | |
8195 | * | |
8196 | * Here we just acquire the spinlock so that all existing callers | |
8197 | * should exit and we're safe to release/invalidate the page. | |
8198 | */ | |
8199 | spin_lock_irq(&subpage->lock); | |
8200 | spin_unlock_irq(&subpage->lock); | |
8201 | } | |
8202 | ||
f913cff3 | 8203 | static bool __btrfs_release_folio(struct folio *folio, gfp_t gfp_flags) |
9ebefb18 | 8204 | { |
f913cff3 | 8205 | int ret = try_release_extent_mapping(&folio->page, gfp_flags); |
7c11d0ae QW |
8206 | |
8207 | if (ret == 1) { | |
f913cff3 MWO |
8208 | wait_subpage_spinlock(&folio->page); |
8209 | clear_page_extent_mapped(&folio->page); | |
7c11d0ae | 8210 | } |
a52d9a80 | 8211 | return ret; |
39279cc3 CM |
8212 | } |
8213 | ||
f913cff3 | 8214 | static bool btrfs_release_folio(struct folio *folio, gfp_t gfp_flags) |
e6dcd2dc | 8215 | { |
f913cff3 MWO |
8216 | if (folio_test_writeback(folio) || folio_test_dirty(folio)) |
8217 | return false; | |
8218 | return __btrfs_release_folio(folio, gfp_flags); | |
e6dcd2dc CM |
8219 | } |
8220 | ||
f8e66081 | 8221 | #ifdef CONFIG_MIGRATION |
e7a60a17 MWO |
8222 | static int btrfs_migrate_folio(struct address_space *mapping, |
8223 | struct folio *dst, struct folio *src, | |
f8e66081 RG |
8224 | enum migrate_mode mode) |
8225 | { | |
e7a60a17 | 8226 | int ret = filemap_migrate_folio(mapping, dst, src, mode); |
f8e66081 | 8227 | |
f8e66081 RG |
8228 | if (ret != MIGRATEPAGE_SUCCESS) |
8229 | return ret; | |
8230 | ||
e7a60a17 MWO |
8231 | if (folio_test_ordered(src)) { |
8232 | folio_clear_ordered(src); | |
8233 | folio_set_ordered(dst); | |
f8e66081 RG |
8234 | } |
8235 | ||
f8e66081 RG |
8236 | return MIGRATEPAGE_SUCCESS; |
8237 | } | |
e7a60a17 MWO |
8238 | #else |
8239 | #define btrfs_migrate_folio NULL | |
f8e66081 RG |
8240 | #endif |
8241 | ||
895586eb MWO |
8242 | static void btrfs_invalidate_folio(struct folio *folio, size_t offset, |
8243 | size_t length) | |
39279cc3 | 8244 | { |
895586eb | 8245 | struct btrfs_inode *inode = BTRFS_I(folio->mapping->host); |
b945a463 | 8246 | struct btrfs_fs_info *fs_info = inode->root->fs_info; |
53ac7ead | 8247 | struct extent_io_tree *tree = &inode->io_tree; |
2ac55d41 | 8248 | struct extent_state *cached_state = NULL; |
895586eb MWO |
8249 | u64 page_start = folio_pos(folio); |
8250 | u64 page_end = page_start + folio_size(folio) - 1; | |
3b835840 | 8251 | u64 cur; |
53ac7ead | 8252 | int inode_evicting = inode->vfs_inode.i_state & I_FREEING; |
39279cc3 | 8253 | |
8b62b72b | 8254 | /* |
895586eb MWO |
8255 | * We have folio locked so no new ordered extent can be created on this |
8256 | * page, nor bio can be submitted for this folio. | |
8b62b72b | 8257 | * |
895586eb MWO |
8258 | * But already submitted bio can still be finished on this folio. |
8259 | * Furthermore, endio function won't skip folio which has Ordered | |
f57ad937 | 8260 | * (Private2) already cleared, so it's possible for endio and |
895586eb MWO |
8261 | * invalidate_folio to do the same ordered extent accounting twice |
8262 | * on one folio. | |
266a2586 QW |
8263 | * |
8264 | * So here we wait for any submitted bios to finish, so that we won't | |
895586eb | 8265 | * do double ordered extent accounting on the same folio. |
8b62b72b | 8266 | */ |
895586eb MWO |
8267 | folio_wait_writeback(folio); |
8268 | wait_subpage_spinlock(&folio->page); | |
8b62b72b | 8269 | |
bcd77455 QW |
8270 | /* |
8271 | * For subpage case, we have call sites like | |
8272 | * btrfs_punch_hole_lock_range() which passes range not aligned to | |
8273 | * sectorsize. | |
895586eb MWO |
8274 | * If the range doesn't cover the full folio, we don't need to and |
8275 | * shouldn't clear page extent mapped, as folio->private can still | |
bcd77455 QW |
8276 | * record subpage dirty bits for other part of the range. |
8277 | * | |
895586eb MWO |
8278 | * For cases that invalidate the full folio even the range doesn't |
8279 | * cover the full folio, like invalidating the last folio, we're | |
bcd77455 QW |
8280 | * still safe to wait for ordered extent to finish. |
8281 | */ | |
5a60542c | 8282 | if (!(offset == 0 && length == folio_size(folio))) { |
f913cff3 | 8283 | btrfs_release_folio(folio, GFP_NOFS); |
e6dcd2dc CM |
8284 | return; |
8285 | } | |
131e404a FDBM |
8286 | |
8287 | if (!inode_evicting) | |
570eb97b | 8288 | lock_extent(tree, page_start, page_end, &cached_state); |
951c80f8 | 8289 | |
3b835840 QW |
8290 | cur = page_start; |
8291 | while (cur < page_end) { | |
8292 | struct btrfs_ordered_extent *ordered; | |
3b835840 | 8293 | u64 range_end; |
b945a463 | 8294 | u32 range_len; |
bd015294 | 8295 | u32 extra_flags = 0; |
3b835840 QW |
8296 | |
8297 | ordered = btrfs_lookup_first_ordered_range(inode, cur, | |
8298 | page_end + 1 - cur); | |
8299 | if (!ordered) { | |
8300 | range_end = page_end; | |
8301 | /* | |
8302 | * No ordered extent covering this range, we are safe | |
8303 | * to delete all extent states in the range. | |
8304 | */ | |
bd015294 | 8305 | extra_flags = EXTENT_CLEAR_ALL_BITS; |
3b835840 QW |
8306 | goto next; |
8307 | } | |
8308 | if (ordered->file_offset > cur) { | |
8309 | /* | |
8310 | * There is a range between [cur, oe->file_offset) not | |
8311 | * covered by any ordered extent. | |
8312 | * We are safe to delete all extent states, and handle | |
8313 | * the ordered extent in the next iteration. | |
8314 | */ | |
8315 | range_end = ordered->file_offset - 1; | |
bd015294 | 8316 | extra_flags = EXTENT_CLEAR_ALL_BITS; |
3b835840 QW |
8317 | goto next; |
8318 | } | |
8319 | ||
8320 | range_end = min(ordered->file_offset + ordered->num_bytes - 1, | |
8321 | page_end); | |
b945a463 QW |
8322 | ASSERT(range_end + 1 - cur < U32_MAX); |
8323 | range_len = range_end + 1 - cur; | |
895586eb | 8324 | if (!btrfs_page_test_ordered(fs_info, &folio->page, cur, range_len)) { |
3b835840 | 8325 | /* |
f57ad937 QW |
8326 | * If Ordered (Private2) is cleared, it means endio has |
8327 | * already been executed for the range. | |
3b835840 QW |
8328 | * We can't delete the extent states as |
8329 | * btrfs_finish_ordered_io() may still use some of them. | |
8330 | */ | |
3b835840 QW |
8331 | goto next; |
8332 | } | |
895586eb | 8333 | btrfs_page_clear_ordered(fs_info, &folio->page, cur, range_len); |
3b835840 | 8334 | |
eb84ae03 | 8335 | /* |
2766ff61 FM |
8336 | * IO on this page will never be started, so we need to account |
8337 | * for any ordered extents now. Don't clear EXTENT_DELALLOC_NEW | |
8338 | * here, must leave that up for the ordered extent completion. | |
3b835840 QW |
8339 | * |
8340 | * This will also unlock the range for incoming | |
8341 | * btrfs_finish_ordered_io(). | |
eb84ae03 | 8342 | */ |
131e404a | 8343 | if (!inode_evicting) |
3b835840 | 8344 | clear_extent_bit(tree, cur, range_end, |
2766ff61 | 8345 | EXTENT_DELALLOC | |
131e404a | 8346 | EXTENT_LOCKED | EXTENT_DO_ACCOUNTING | |
bd015294 | 8347 | EXTENT_DEFRAG, &cached_state); |
3b835840 QW |
8348 | |
8349 | spin_lock_irq(&inode->ordered_tree.lock); | |
8350 | set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags); | |
8351 | ordered->truncated_len = min(ordered->truncated_len, | |
8352 | cur - ordered->file_offset); | |
8353 | spin_unlock_irq(&inode->ordered_tree.lock); | |
8354 | ||
bd015294 JB |
8355 | /* |
8356 | * If the ordered extent has finished, we're safe to delete all | |
8357 | * the extent states of the range, otherwise | |
8358 | * btrfs_finish_ordered_io() will get executed by endio for | |
8359 | * other pages, so we can't delete extent states. | |
8360 | */ | |
3b835840 | 8361 | if (btrfs_dec_test_ordered_pending(inode, &ordered, |
f41b6ba9 | 8362 | cur, range_end + 1 - cur)) { |
3b835840 QW |
8363 | btrfs_finish_ordered_io(ordered); |
8364 | /* | |
8365 | * The ordered extent has finished, now we're again | |
8366 | * safe to delete all extent states of the range. | |
8367 | */ | |
bd015294 | 8368 | extra_flags = EXTENT_CLEAR_ALL_BITS; |
3b835840 QW |
8369 | } |
8370 | next: | |
8371 | if (ordered) | |
8372 | btrfs_put_ordered_extent(ordered); | |
8b62b72b | 8373 | /* |
3b835840 QW |
8374 | * Qgroup reserved space handler |
8375 | * Sector(s) here will be either: | |
266a2586 | 8376 | * |
3b835840 QW |
8377 | * 1) Already written to disk or bio already finished |
8378 | * Then its QGROUP_RESERVED bit in io_tree is already cleared. | |
8379 | * Qgroup will be handled by its qgroup_record then. | |
8380 | * btrfs_qgroup_free_data() call will do nothing here. | |
8381 | * | |
8382 | * 2) Not written to disk yet | |
8383 | * Then btrfs_qgroup_free_data() call will clear the | |
8384 | * QGROUP_RESERVED bit of its io_tree, and free the qgroup | |
8385 | * reserved data space. | |
8386 | * Since the IO will never happen for this page. | |
8b62b72b | 8387 | */ |
3b835840 | 8388 | btrfs_qgroup_free_data(inode, NULL, cur, range_end + 1 - cur); |
131e404a | 8389 | if (!inode_evicting) { |
3b835840 QW |
8390 | clear_extent_bit(tree, cur, range_end, EXTENT_LOCKED | |
8391 | EXTENT_DELALLOC | EXTENT_UPTODATE | | |
bd015294 JB |
8392 | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG | |
8393 | extra_flags, &cached_state); | |
131e404a | 8394 | } |
3b835840 | 8395 | cur = range_end + 1; |
131e404a | 8396 | } |
b9d0b389 | 8397 | /* |
3b835840 | 8398 | * We have iterated through all ordered extents of the page, the page |
f57ad937 QW |
8399 | * should not have Ordered (Private2) anymore, or the above iteration |
8400 | * did something wrong. | |
b9d0b389 | 8401 | */ |
895586eb MWO |
8402 | ASSERT(!folio_test_ordered(folio)); |
8403 | btrfs_page_clear_checked(fs_info, &folio->page, folio_pos(folio), folio_size(folio)); | |
3b835840 | 8404 | if (!inode_evicting) |
f913cff3 | 8405 | __btrfs_release_folio(folio, GFP_NOFS); |
895586eb | 8406 | clear_page_extent_mapped(&folio->page); |
39279cc3 CM |
8407 | } |
8408 | ||
9ebefb18 CM |
8409 | /* |
8410 | * btrfs_page_mkwrite() is not allowed to change the file size as it gets | |
8411 | * called from a page fault handler when a page is first dirtied. Hence we must | |
8412 | * be careful to check for EOF conditions here. We set the page up correctly | |
8413 | * for a written page which means we get ENOSPC checking when writing into | |
8414 | * holes and correct delalloc and unwritten extent mapping on filesystems that | |
8415 | * support these features. | |
8416 | * | |
8417 | * We are not allowed to take the i_mutex here so we have to play games to | |
8418 | * protect against truncate races as the page could now be beyond EOF. Because | |
d1342aad OS |
8419 | * truncate_setsize() writes the inode size before removing pages, once we have |
8420 | * the page lock we can determine safely if the page is beyond EOF. If it is not | |
9ebefb18 CM |
8421 | * beyond EOF, then the page is guaranteed safe against truncation until we |
8422 | * unlock the page. | |
8423 | */ | |
a528a241 | 8424 | vm_fault_t btrfs_page_mkwrite(struct vm_fault *vmf) |
9ebefb18 | 8425 | { |
c2ec175c | 8426 | struct page *page = vmf->page; |
11bac800 | 8427 | struct inode *inode = file_inode(vmf->vma->vm_file); |
0b246afa | 8428 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
e6dcd2dc CM |
8429 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
8430 | struct btrfs_ordered_extent *ordered; | |
2ac55d41 | 8431 | struct extent_state *cached_state = NULL; |
364ecf36 | 8432 | struct extent_changeset *data_reserved = NULL; |
e6dcd2dc | 8433 | unsigned long zero_start; |
9ebefb18 | 8434 | loff_t size; |
a528a241 SJ |
8435 | vm_fault_t ret; |
8436 | int ret2; | |
9998eb70 | 8437 | int reserved = 0; |
d0b7da88 | 8438 | u64 reserved_space; |
a52d9a80 | 8439 | u64 page_start; |
e6dcd2dc | 8440 | u64 page_end; |
d0b7da88 CR |
8441 | u64 end; |
8442 | ||
09cbfeaf | 8443 | reserved_space = PAGE_SIZE; |
9ebefb18 | 8444 | |
b2b5ef5c | 8445 | sb_start_pagefault(inode->i_sb); |
df480633 | 8446 | page_start = page_offset(page); |
09cbfeaf | 8447 | page_end = page_start + PAGE_SIZE - 1; |
d0b7da88 | 8448 | end = page_end; |
df480633 | 8449 | |
d0b7da88 CR |
8450 | /* |
8451 | * Reserving delalloc space after obtaining the page lock can lead to | |
8452 | * deadlock. For example, if a dirty page is locked by this function | |
8453 | * and the call to btrfs_delalloc_reserve_space() ends up triggering | |
f3e90c1c | 8454 | * dirty page write out, then the btrfs_writepages() function could |
d0b7da88 CR |
8455 | * end up waiting indefinitely to get a lock on the page currently |
8456 | * being processed by btrfs_page_mkwrite() function. | |
8457 | */ | |
e5b7231e NB |
8458 | ret2 = btrfs_delalloc_reserve_space(BTRFS_I(inode), &data_reserved, |
8459 | page_start, reserved_space); | |
a528a241 SJ |
8460 | if (!ret2) { |
8461 | ret2 = file_update_time(vmf->vma->vm_file); | |
9998eb70 CM |
8462 | reserved = 1; |
8463 | } | |
a528a241 SJ |
8464 | if (ret2) { |
8465 | ret = vmf_error(ret2); | |
9998eb70 CM |
8466 | if (reserved) |
8467 | goto out; | |
8468 | goto out_noreserve; | |
56a76f82 | 8469 | } |
1832a6d5 | 8470 | |
56a76f82 | 8471 | ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */ |
e6dcd2dc | 8472 | again: |
8318ba79 | 8473 | down_read(&BTRFS_I(inode)->i_mmap_lock); |
9ebefb18 | 8474 | lock_page(page); |
9ebefb18 | 8475 | size = i_size_read(inode); |
a52d9a80 | 8476 | |
9ebefb18 | 8477 | if ((page->mapping != inode->i_mapping) || |
e6dcd2dc | 8478 | (page_start >= size)) { |
9ebefb18 CM |
8479 | /* page got truncated out from underneath us */ |
8480 | goto out_unlock; | |
8481 | } | |
e6dcd2dc CM |
8482 | wait_on_page_writeback(page); |
8483 | ||
570eb97b | 8484 | lock_extent(io_tree, page_start, page_end, &cached_state); |
32443de3 QW |
8485 | ret2 = set_page_extent_mapped(page); |
8486 | if (ret2 < 0) { | |
8487 | ret = vmf_error(ret2); | |
570eb97b | 8488 | unlock_extent(io_tree, page_start, page_end, &cached_state); |
32443de3 QW |
8489 | goto out_unlock; |
8490 | } | |
e6dcd2dc | 8491 | |
eb84ae03 CM |
8492 | /* |
8493 | * we can't set the delalloc bits if there are pending ordered | |
8494 | * extents. Drop our locks and wait for them to finish | |
8495 | */ | |
a776c6fa NB |
8496 | ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), page_start, |
8497 | PAGE_SIZE); | |
e6dcd2dc | 8498 | if (ordered) { |
570eb97b | 8499 | unlock_extent(io_tree, page_start, page_end, &cached_state); |
e6dcd2dc | 8500 | unlock_page(page); |
8318ba79 | 8501 | up_read(&BTRFS_I(inode)->i_mmap_lock); |
c0a43603 | 8502 | btrfs_start_ordered_extent(ordered, 1); |
e6dcd2dc CM |
8503 | btrfs_put_ordered_extent(ordered); |
8504 | goto again; | |
8505 | } | |
8506 | ||
09cbfeaf | 8507 | if (page->index == ((size - 1) >> PAGE_SHIFT)) { |
da17066c | 8508 | reserved_space = round_up(size - page_start, |
0b246afa | 8509 | fs_info->sectorsize); |
09cbfeaf | 8510 | if (reserved_space < PAGE_SIZE) { |
d0b7da88 | 8511 | end = page_start + reserved_space - 1; |
86d52921 NB |
8512 | btrfs_delalloc_release_space(BTRFS_I(inode), |
8513 | data_reserved, page_start, | |
8514 | PAGE_SIZE - reserved_space, true); | |
d0b7da88 CR |
8515 | } |
8516 | } | |
8517 | ||
fbf19087 | 8518 | /* |
5416034f LB |
8519 | * page_mkwrite gets called when the page is firstly dirtied after it's |
8520 | * faulted in, but write(2) could also dirty a page and set delalloc | |
8521 | * bits, thus in this case for space account reason, we still need to | |
8522 | * clear any delalloc bits within this page range since we have to | |
8523 | * reserve data&meta space before lock_page() (see above comments). | |
fbf19087 | 8524 | */ |
d0b7da88 | 8525 | clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, end, |
e182163d | 8526 | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING | |
bd015294 | 8527 | EXTENT_DEFRAG, &cached_state); |
fbf19087 | 8528 | |
c2566f22 | 8529 | ret2 = btrfs_set_extent_delalloc(BTRFS_I(inode), page_start, end, 0, |
330a5827 | 8530 | &cached_state); |
a528a241 | 8531 | if (ret2) { |
570eb97b | 8532 | unlock_extent(io_tree, page_start, page_end, &cached_state); |
9ed74f2d JB |
8533 | ret = VM_FAULT_SIGBUS; |
8534 | goto out_unlock; | |
8535 | } | |
9ebefb18 CM |
8536 | |
8537 | /* page is wholly or partially inside EOF */ | |
09cbfeaf | 8538 | if (page_start + PAGE_SIZE > size) |
7073017a | 8539 | zero_start = offset_in_page(size); |
9ebefb18 | 8540 | else |
09cbfeaf | 8541 | zero_start = PAGE_SIZE; |
9ebefb18 | 8542 | |
21a8935e | 8543 | if (zero_start != PAGE_SIZE) |
d048b9c2 | 8544 | memzero_page(page, zero_start, PAGE_SIZE - zero_start); |
21a8935e | 8545 | |
e4f94347 | 8546 | btrfs_page_clear_checked(fs_info, page, page_start, PAGE_SIZE); |
2d8ec40e QW |
8547 | btrfs_page_set_dirty(fs_info, page, page_start, end + 1 - page_start); |
8548 | btrfs_page_set_uptodate(fs_info, page, page_start, end + 1 - page_start); | |
5a3f23d5 | 8549 | |
bc0939fc | 8550 | btrfs_set_inode_last_sub_trans(BTRFS_I(inode)); |
257c62e1 | 8551 | |
570eb97b | 8552 | unlock_extent(io_tree, page_start, page_end, &cached_state); |
8318ba79 | 8553 | up_read(&BTRFS_I(inode)->i_mmap_lock); |
9ebefb18 | 8554 | |
76de60ed YY |
8555 | btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE); |
8556 | sb_end_pagefault(inode->i_sb); | |
8557 | extent_changeset_free(data_reserved); | |
8558 | return VM_FAULT_LOCKED; | |
717beb96 CM |
8559 | |
8560 | out_unlock: | |
9ebefb18 | 8561 | unlock_page(page); |
8318ba79 | 8562 | up_read(&BTRFS_I(inode)->i_mmap_lock); |
1832a6d5 | 8563 | out: |
8702ba93 | 8564 | btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE); |
86d52921 | 8565 | btrfs_delalloc_release_space(BTRFS_I(inode), data_reserved, page_start, |
43b18595 | 8566 | reserved_space, (ret != 0)); |
9998eb70 | 8567 | out_noreserve: |
b2b5ef5c | 8568 | sb_end_pagefault(inode->i_sb); |
364ecf36 | 8569 | extent_changeset_free(data_reserved); |
9ebefb18 CM |
8570 | return ret; |
8571 | } | |
8572 | ||
213e8c55 | 8573 | static int btrfs_truncate(struct inode *inode, bool skip_writeback) |
39279cc3 | 8574 | { |
d9ac19c3 | 8575 | struct btrfs_truncate_control control = { |
71d18b53 | 8576 | .inode = BTRFS_I(inode), |
487e81d2 | 8577 | .ino = btrfs_ino(BTRFS_I(inode)), |
d9ac19c3 | 8578 | .min_type = BTRFS_EXTENT_DATA_KEY, |
655807b8 | 8579 | .clear_extent_range = true, |
d9ac19c3 | 8580 | }; |
0b246afa | 8581 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
39279cc3 | 8582 | struct btrfs_root *root = BTRFS_I(inode)->root; |
fcb80c2a | 8583 | struct btrfs_block_rsv *rsv; |
ad7e1a74 | 8584 | int ret; |
39279cc3 | 8585 | struct btrfs_trans_handle *trans; |
0b246afa | 8586 | u64 mask = fs_info->sectorsize - 1; |
2bd36e7b | 8587 | u64 min_size = btrfs_calc_metadata_size(fs_info, 1); |
39279cc3 | 8588 | |
213e8c55 FM |
8589 | if (!skip_writeback) { |
8590 | ret = btrfs_wait_ordered_range(inode, inode->i_size & (~mask), | |
8591 | (u64)-1); | |
8592 | if (ret) | |
8593 | return ret; | |
8594 | } | |
39279cc3 | 8595 | |
fcb80c2a | 8596 | /* |
f7e9e8fc OS |
8597 | * Yes ladies and gentlemen, this is indeed ugly. We have a couple of |
8598 | * things going on here: | |
fcb80c2a | 8599 | * |
f7e9e8fc | 8600 | * 1) We need to reserve space to update our inode. |
fcb80c2a | 8601 | * |
f7e9e8fc | 8602 | * 2) We need to have something to cache all the space that is going to |
fcb80c2a JB |
8603 | * be free'd up by the truncate operation, but also have some slack |
8604 | * space reserved in case it uses space during the truncate (thank you | |
8605 | * very much snapshotting). | |
8606 | * | |
f7e9e8fc | 8607 | * And we need these to be separate. The fact is we can use a lot of |
fcb80c2a | 8608 | * space doing the truncate, and we have no earthly idea how much space |
01327610 | 8609 | * we will use, so we need the truncate reservation to be separate so it |
f7e9e8fc OS |
8610 | * doesn't end up using space reserved for updating the inode. We also |
8611 | * need to be able to stop the transaction and start a new one, which | |
8612 | * means we need to be able to update the inode several times, and we | |
8613 | * have no idea of knowing how many times that will be, so we can't just | |
8614 | * reserve 1 item for the entirety of the operation, so that has to be | |
8615 | * done separately as well. | |
fcb80c2a JB |
8616 | * |
8617 | * So that leaves us with | |
8618 | * | |
f7e9e8fc | 8619 | * 1) rsv - for the truncate reservation, which we will steal from the |
fcb80c2a | 8620 | * transaction reservation. |
f7e9e8fc | 8621 | * 2) fs_info->trans_block_rsv - this will have 1 items worth left for |
fcb80c2a JB |
8622 | * updating the inode. |
8623 | */ | |
2ff7e61e | 8624 | rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP); |
fcb80c2a JB |
8625 | if (!rsv) |
8626 | return -ENOMEM; | |
4a338542 | 8627 | rsv->size = min_size; |
710d5921 | 8628 | rsv->failfast = true; |
f0cd846e | 8629 | |
907cbceb | 8630 | /* |
07127184 | 8631 | * 1 for the truncate slack space |
907cbceb JB |
8632 | * 1 for updating the inode. |
8633 | */ | |
f3fe820c | 8634 | trans = btrfs_start_transaction(root, 2); |
fcb80c2a | 8635 | if (IS_ERR(trans)) { |
ad7e1a74 | 8636 | ret = PTR_ERR(trans); |
fcb80c2a JB |
8637 | goto out; |
8638 | } | |
f0cd846e | 8639 | |
907cbceb | 8640 | /* Migrate the slack space for the truncate to our reserve */ |
0b246afa | 8641 | ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv, rsv, |
3a584174 | 8642 | min_size, false); |
fcb80c2a | 8643 | BUG_ON(ret); |
f0cd846e | 8644 | |
ca7e70f5 | 8645 | trans->block_rsv = rsv; |
907cbceb | 8646 | |
8082510e | 8647 | while (1) { |
9a4a1429 JB |
8648 | struct extent_state *cached_state = NULL; |
8649 | const u64 new_size = inode->i_size; | |
8650 | const u64 lock_start = ALIGN_DOWN(new_size, fs_info->sectorsize); | |
8651 | ||
d9ac19c3 | 8652 | control.new_size = new_size; |
570eb97b | 8653 | lock_extent(&BTRFS_I(inode)->io_tree, lock_start, (u64)-1, |
9a4a1429 JB |
8654 | &cached_state); |
8655 | /* | |
8656 | * We want to drop from the next block forward in case this new | |
8657 | * size is not block aligned since we will be keeping the last | |
8658 | * block of the extent just the way it is. | |
8659 | */ | |
8660 | btrfs_drop_extent_cache(BTRFS_I(inode), | |
8661 | ALIGN(new_size, fs_info->sectorsize), | |
8662 | (u64)-1, 0); | |
8663 | ||
71d18b53 | 8664 | ret = btrfs_truncate_inode_items(trans, root, &control); |
c2ddb612 | 8665 | |
462b728e | 8666 | inode_sub_bytes(inode, control.sub_bytes); |
c2ddb612 JB |
8667 | btrfs_inode_safe_disk_i_size_write(BTRFS_I(inode), control.last_size); |
8668 | ||
570eb97b JB |
8669 | unlock_extent(&BTRFS_I(inode)->io_tree, lock_start, (u64)-1, |
8670 | &cached_state); | |
9a4a1429 | 8671 | |
ddfae63c | 8672 | trans->block_rsv = &fs_info->trans_block_rsv; |
ad7e1a74 | 8673 | if (ret != -ENOSPC && ret != -EAGAIN) |
8082510e | 8674 | break; |
39279cc3 | 8675 | |
9a56fcd1 | 8676 | ret = btrfs_update_inode(trans, root, BTRFS_I(inode)); |
ad7e1a74 | 8677 | if (ret) |
3893e33b | 8678 | break; |
ca7e70f5 | 8679 | |
3a45bb20 | 8680 | btrfs_end_transaction(trans); |
2ff7e61e | 8681 | btrfs_btree_balance_dirty(fs_info); |
ca7e70f5 JB |
8682 | |
8683 | trans = btrfs_start_transaction(root, 2); | |
8684 | if (IS_ERR(trans)) { | |
ad7e1a74 | 8685 | ret = PTR_ERR(trans); |
ca7e70f5 JB |
8686 | trans = NULL; |
8687 | break; | |
8688 | } | |
8689 | ||
63f018be | 8690 | btrfs_block_rsv_release(fs_info, rsv, -1, NULL); |
0b246afa | 8691 | ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv, |
3a584174 | 8692 | rsv, min_size, false); |
ca7e70f5 JB |
8693 | BUG_ON(ret); /* shouldn't happen */ |
8694 | trans->block_rsv = rsv; | |
8082510e YZ |
8695 | } |
8696 | ||
ddfae63c JB |
8697 | /* |
8698 | * We can't call btrfs_truncate_block inside a trans handle as we could | |
54f03ab1 JB |
8699 | * deadlock with freeze, if we got BTRFS_NEED_TRUNCATE_BLOCK then we |
8700 | * know we've truncated everything except the last little bit, and can | |
8701 | * do btrfs_truncate_block and then update the disk_i_size. | |
ddfae63c | 8702 | */ |
54f03ab1 | 8703 | if (ret == BTRFS_NEED_TRUNCATE_BLOCK) { |
ddfae63c JB |
8704 | btrfs_end_transaction(trans); |
8705 | btrfs_btree_balance_dirty(fs_info); | |
8706 | ||
217f42eb | 8707 | ret = btrfs_truncate_block(BTRFS_I(inode), inode->i_size, 0, 0); |
ddfae63c JB |
8708 | if (ret) |
8709 | goto out; | |
8710 | trans = btrfs_start_transaction(root, 1); | |
8711 | if (IS_ERR(trans)) { | |
8712 | ret = PTR_ERR(trans); | |
8713 | goto out; | |
8714 | } | |
76aea537 | 8715 | btrfs_inode_safe_disk_i_size_write(BTRFS_I(inode), 0); |
ddfae63c JB |
8716 | } |
8717 | ||
917c16b2 | 8718 | if (trans) { |
ad7e1a74 OS |
8719 | int ret2; |
8720 | ||
0b246afa | 8721 | trans->block_rsv = &fs_info->trans_block_rsv; |
9a56fcd1 | 8722 | ret2 = btrfs_update_inode(trans, root, BTRFS_I(inode)); |
ad7e1a74 OS |
8723 | if (ret2 && !ret) |
8724 | ret = ret2; | |
7b128766 | 8725 | |
ad7e1a74 OS |
8726 | ret2 = btrfs_end_transaction(trans); |
8727 | if (ret2 && !ret) | |
8728 | ret = ret2; | |
2ff7e61e | 8729 | btrfs_btree_balance_dirty(fs_info); |
917c16b2 | 8730 | } |
fcb80c2a | 8731 | out: |
2ff7e61e | 8732 | btrfs_free_block_rsv(fs_info, rsv); |
0d7d3165 FM |
8733 | /* |
8734 | * So if we truncate and then write and fsync we normally would just | |
8735 | * write the extents that changed, which is a problem if we need to | |
8736 | * first truncate that entire inode. So set this flag so we write out | |
8737 | * all of the extents in the inode to the sync log so we're completely | |
8738 | * safe. | |
8739 | * | |
8740 | * If no extents were dropped or trimmed we don't need to force the next | |
8741 | * fsync to truncate all the inode's items from the log and re-log them | |
8742 | * all. This means the truncate operation did not change the file size, | |
8743 | * or changed it to a smaller size but there was only an implicit hole | |
8744 | * between the old i_size and the new i_size, and there were no prealloc | |
8745 | * extents beyond i_size to drop. | |
8746 | */ | |
d9ac19c3 | 8747 | if (control.extents_found > 0) |
23e3337f | 8748 | btrfs_set_inode_full_sync(BTRFS_I(inode)); |
fcb80c2a | 8749 | |
ad7e1a74 | 8750 | return ret; |
39279cc3 CM |
8751 | } |
8752 | ||
a1fd0c35 OS |
8753 | struct inode *btrfs_new_subvol_inode(struct user_namespace *mnt_userns, |
8754 | struct inode *dir) | |
8755 | { | |
8756 | struct inode *inode; | |
8757 | ||
8758 | inode = new_inode(dir->i_sb); | |
8759 | if (inode) { | |
8760 | /* | |
8761 | * Subvolumes don't inherit the sgid bit or the parent's gid if | |
8762 | * the parent's sgid bit is set. This is probably a bug. | |
8763 | */ | |
8764 | inode_init_owner(mnt_userns, inode, NULL, | |
8765 | S_IFDIR | (~current_umask() & S_IRWXUGO)); | |
8766 | inode->i_op = &btrfs_dir_inode_operations; | |
8767 | inode->i_fop = &btrfs_dir_file_operations; | |
8768 | } | |
8769 | return inode; | |
8770 | } | |
8771 | ||
39279cc3 CM |
8772 | struct inode *btrfs_alloc_inode(struct super_block *sb) |
8773 | { | |
69fe2d75 | 8774 | struct btrfs_fs_info *fs_info = btrfs_sb(sb); |
39279cc3 | 8775 | struct btrfs_inode *ei; |
2ead6ae7 | 8776 | struct inode *inode; |
39279cc3 | 8777 | |
fd60b288 | 8778 | ei = alloc_inode_sb(sb, btrfs_inode_cachep, GFP_KERNEL); |
39279cc3 CM |
8779 | if (!ei) |
8780 | return NULL; | |
2ead6ae7 YZ |
8781 | |
8782 | ei->root = NULL; | |
2ead6ae7 | 8783 | ei->generation = 0; |
15ee9bc7 | 8784 | ei->last_trans = 0; |
257c62e1 | 8785 | ei->last_sub_trans = 0; |
e02119d5 | 8786 | ei->logged_trans = 0; |
2ead6ae7 | 8787 | ei->delalloc_bytes = 0; |
a7e3b975 | 8788 | ei->new_delalloc_bytes = 0; |
47059d93 | 8789 | ei->defrag_bytes = 0; |
2ead6ae7 YZ |
8790 | ei->disk_i_size = 0; |
8791 | ei->flags = 0; | |
77eea05e | 8792 | ei->ro_flags = 0; |
7709cde3 | 8793 | ei->csum_bytes = 0; |
2ead6ae7 | 8794 | ei->index_cnt = (u64)-1; |
67de1176 | 8795 | ei->dir_index = 0; |
2ead6ae7 | 8796 | ei->last_unlink_trans = 0; |
3ebac17c | 8797 | ei->last_reflink_trans = 0; |
46d8bc34 | 8798 | ei->last_log_commit = 0; |
2ead6ae7 | 8799 | |
9e0baf60 | 8800 | spin_lock_init(&ei->lock); |
87c11705 | 8801 | spin_lock_init(&ei->io_failure_lock); |
9e0baf60 | 8802 | ei->outstanding_extents = 0; |
69fe2d75 JB |
8803 | if (sb->s_magic != BTRFS_TEST_MAGIC) |
8804 | btrfs_init_metadata_block_rsv(fs_info, &ei->block_rsv, | |
8805 | BTRFS_BLOCK_RSV_DELALLOC); | |
72ac3c0d | 8806 | ei->runtime_flags = 0; |
b52aa8c9 | 8807 | ei->prop_compress = BTRFS_COMPRESS_NONE; |
eec63c65 | 8808 | ei->defrag_compress = BTRFS_COMPRESS_NONE; |
2ead6ae7 | 8809 | |
16cdcec7 MX |
8810 | ei->delayed_node = NULL; |
8811 | ||
9cc97d64 | 8812 | ei->i_otime.tv_sec = 0; |
8813 | ei->i_otime.tv_nsec = 0; | |
8814 | ||
2ead6ae7 | 8815 | inode = &ei->vfs_inode; |
a8067e02 | 8816 | extent_map_tree_init(&ei->extent_tree); |
43eb5f29 | 8817 | extent_io_tree_init(fs_info, &ei->io_tree, IO_TREE_INODE_IO, inode); |
41a2ee75 | 8818 | extent_io_tree_init(fs_info, &ei->file_extent_tree, |
efb0645b | 8819 | IO_TREE_INODE_FILE_EXTENT, NULL); |
87c11705 | 8820 | ei->io_failure_tree = RB_ROOT; |
b812ce28 | 8821 | atomic_set(&ei->sync_writers, 0); |
2ead6ae7 | 8822 | mutex_init(&ei->log_mutex); |
e6dcd2dc | 8823 | btrfs_ordered_inode_tree_init(&ei->ordered_tree); |
2ead6ae7 | 8824 | INIT_LIST_HEAD(&ei->delalloc_inodes); |
8089fe62 | 8825 | INIT_LIST_HEAD(&ei->delayed_iput); |
2ead6ae7 | 8826 | RB_CLEAR_NODE(&ei->rb_node); |
8318ba79 | 8827 | init_rwsem(&ei->i_mmap_lock); |
2ead6ae7 YZ |
8828 | |
8829 | return inode; | |
39279cc3 CM |
8830 | } |
8831 | ||
aaedb55b JB |
8832 | #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS |
8833 | void btrfs_test_destroy_inode(struct inode *inode) | |
8834 | { | |
dcdbc059 | 8835 | btrfs_drop_extent_cache(BTRFS_I(inode), 0, (u64)-1, 0); |
aaedb55b JB |
8836 | kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode)); |
8837 | } | |
8838 | #endif | |
8839 | ||
26602cab | 8840 | void btrfs_free_inode(struct inode *inode) |
fa0d7e3d | 8841 | { |
fa0d7e3d NP |
8842 | kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode)); |
8843 | } | |
8844 | ||
633cc816 | 8845 | void btrfs_destroy_inode(struct inode *vfs_inode) |
39279cc3 | 8846 | { |
e6dcd2dc | 8847 | struct btrfs_ordered_extent *ordered; |
633cc816 NB |
8848 | struct btrfs_inode *inode = BTRFS_I(vfs_inode); |
8849 | struct btrfs_root *root = inode->root; | |
5f4403e1 | 8850 | bool freespace_inode; |
5a3f23d5 | 8851 | |
633cc816 NB |
8852 | WARN_ON(!hlist_empty(&vfs_inode->i_dentry)); |
8853 | WARN_ON(vfs_inode->i_data.nrpages); | |
8854 | WARN_ON(inode->block_rsv.reserved); | |
8855 | WARN_ON(inode->block_rsv.size); | |
8856 | WARN_ON(inode->outstanding_extents); | |
dc287224 FM |
8857 | if (!S_ISDIR(vfs_inode->i_mode)) { |
8858 | WARN_ON(inode->delalloc_bytes); | |
8859 | WARN_ON(inode->new_delalloc_bytes); | |
8860 | } | |
633cc816 NB |
8861 | WARN_ON(inode->csum_bytes); |
8862 | WARN_ON(inode->defrag_bytes); | |
39279cc3 | 8863 | |
a6dbd429 JB |
8864 | /* |
8865 | * This can happen where we create an inode, but somebody else also | |
8866 | * created the same inode and we need to destroy the one we already | |
8867 | * created. | |
8868 | */ | |
8869 | if (!root) | |
26602cab | 8870 | return; |
a6dbd429 | 8871 | |
5f4403e1 IA |
8872 | /* |
8873 | * If this is a free space inode do not take the ordered extents lockdep | |
8874 | * map. | |
8875 | */ | |
8876 | freespace_inode = btrfs_is_free_space_inode(inode); | |
8877 | ||
d397712b | 8878 | while (1) { |
633cc816 | 8879 | ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1); |
e6dcd2dc CM |
8880 | if (!ordered) |
8881 | break; | |
8882 | else { | |
633cc816 | 8883 | btrfs_err(root->fs_info, |
5d163e0e | 8884 | "found ordered extent %llu %llu on inode cleanup", |
bffe633e | 8885 | ordered->file_offset, ordered->num_bytes); |
5f4403e1 IA |
8886 | |
8887 | if (!freespace_inode) | |
8888 | btrfs_lockdep_acquire(root->fs_info, btrfs_ordered_extent); | |
8889 | ||
71fe0a55 | 8890 | btrfs_remove_ordered_extent(inode, ordered); |
e6dcd2dc CM |
8891 | btrfs_put_ordered_extent(ordered); |
8892 | btrfs_put_ordered_extent(ordered); | |
8893 | } | |
8894 | } | |
633cc816 NB |
8895 | btrfs_qgroup_check_reserved_leak(inode); |
8896 | inode_tree_del(inode); | |
8897 | btrfs_drop_extent_cache(inode, 0, (u64)-1, 0); | |
8898 | btrfs_inode_clear_file_extent_range(inode, 0, (u64)-1); | |
8899 | btrfs_put_root(inode->root); | |
39279cc3 CM |
8900 | } |
8901 | ||
45321ac5 | 8902 | int btrfs_drop_inode(struct inode *inode) |
76dda93c YZ |
8903 | { |
8904 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
45321ac5 | 8905 | |
6379ef9f NA |
8906 | if (root == NULL) |
8907 | return 1; | |
8908 | ||
fa6ac876 | 8909 | /* the snap/subvol tree is on deleting */ |
69e9c6c6 | 8910 | if (btrfs_root_refs(&root->root_item) == 0) |
45321ac5 | 8911 | return 1; |
76dda93c | 8912 | else |
45321ac5 | 8913 | return generic_drop_inode(inode); |
76dda93c YZ |
8914 | } |
8915 | ||
0ee0fda0 | 8916 | static void init_once(void *foo) |
39279cc3 | 8917 | { |
0d031dc4 | 8918 | struct btrfs_inode *ei = foo; |
39279cc3 CM |
8919 | |
8920 | inode_init_once(&ei->vfs_inode); | |
8921 | } | |
8922 | ||
e67c718b | 8923 | void __cold btrfs_destroy_cachep(void) |
39279cc3 | 8924 | { |
8c0a8537 KS |
8925 | /* |
8926 | * Make sure all delayed rcu free inodes are flushed before we | |
8927 | * destroy cache. | |
8928 | */ | |
8929 | rcu_barrier(); | |
642c5d34 | 8930 | bioset_exit(&btrfs_dio_bioset); |
5598e900 KM |
8931 | kmem_cache_destroy(btrfs_inode_cachep); |
8932 | kmem_cache_destroy(btrfs_trans_handle_cachep); | |
5598e900 KM |
8933 | kmem_cache_destroy(btrfs_path_cachep); |
8934 | kmem_cache_destroy(btrfs_free_space_cachep); | |
3acd4850 | 8935 | kmem_cache_destroy(btrfs_free_space_bitmap_cachep); |
39279cc3 CM |
8936 | } |
8937 | ||
f5c29bd9 | 8938 | int __init btrfs_init_cachep(void) |
39279cc3 | 8939 | { |
837e1972 | 8940 | btrfs_inode_cachep = kmem_cache_create("btrfs_inode", |
9601e3f6 | 8941 | sizeof(struct btrfs_inode), 0, |
5d097056 VD |
8942 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD | SLAB_ACCOUNT, |
8943 | init_once); | |
39279cc3 CM |
8944 | if (!btrfs_inode_cachep) |
8945 | goto fail; | |
9601e3f6 | 8946 | |
837e1972 | 8947 | btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle", |
9601e3f6 | 8948 | sizeof(struct btrfs_trans_handle), 0, |
fba4b697 | 8949 | SLAB_TEMPORARY | SLAB_MEM_SPREAD, NULL); |
39279cc3 CM |
8950 | if (!btrfs_trans_handle_cachep) |
8951 | goto fail; | |
9601e3f6 | 8952 | |
837e1972 | 8953 | btrfs_path_cachep = kmem_cache_create("btrfs_path", |
9601e3f6 | 8954 | sizeof(struct btrfs_path), 0, |
fba4b697 | 8955 | SLAB_MEM_SPREAD, NULL); |
39279cc3 CM |
8956 | if (!btrfs_path_cachep) |
8957 | goto fail; | |
9601e3f6 | 8958 | |
837e1972 | 8959 | btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space", |
dc89e982 | 8960 | sizeof(struct btrfs_free_space), 0, |
fba4b697 | 8961 | SLAB_MEM_SPREAD, NULL); |
dc89e982 JB |
8962 | if (!btrfs_free_space_cachep) |
8963 | goto fail; | |
8964 | ||
3acd4850 CL |
8965 | btrfs_free_space_bitmap_cachep = kmem_cache_create("btrfs_free_space_bitmap", |
8966 | PAGE_SIZE, PAGE_SIZE, | |
34e49994 | 8967 | SLAB_MEM_SPREAD, NULL); |
3acd4850 CL |
8968 | if (!btrfs_free_space_bitmap_cachep) |
8969 | goto fail; | |
8970 | ||
642c5d34 CH |
8971 | if (bioset_init(&btrfs_dio_bioset, BIO_POOL_SIZE, |
8972 | offsetof(struct btrfs_dio_private, bio), | |
8973 | BIOSET_NEED_BVECS)) | |
8974 | goto fail; | |
8975 | ||
39279cc3 CM |
8976 | return 0; |
8977 | fail: | |
8978 | btrfs_destroy_cachep(); | |
8979 | return -ENOMEM; | |
8980 | } | |
8981 | ||
549c7297 CB |
8982 | static int btrfs_getattr(struct user_namespace *mnt_userns, |
8983 | const struct path *path, struct kstat *stat, | |
a528d35e | 8984 | u32 request_mask, unsigned int flags) |
39279cc3 | 8985 | { |
df0af1a5 | 8986 | u64 delalloc_bytes; |
2766ff61 | 8987 | u64 inode_bytes; |
a528d35e | 8988 | struct inode *inode = d_inode(path->dentry); |
fadc0d8b | 8989 | u32 blocksize = inode->i_sb->s_blocksize; |
04a87e34 | 8990 | u32 bi_flags = BTRFS_I(inode)->flags; |
14605409 | 8991 | u32 bi_ro_flags = BTRFS_I(inode)->ro_flags; |
04a87e34 YS |
8992 | |
8993 | stat->result_mask |= STATX_BTIME; | |
8994 | stat->btime.tv_sec = BTRFS_I(inode)->i_otime.tv_sec; | |
8995 | stat->btime.tv_nsec = BTRFS_I(inode)->i_otime.tv_nsec; | |
8996 | if (bi_flags & BTRFS_INODE_APPEND) | |
8997 | stat->attributes |= STATX_ATTR_APPEND; | |
8998 | if (bi_flags & BTRFS_INODE_COMPRESS) | |
8999 | stat->attributes |= STATX_ATTR_COMPRESSED; | |
9000 | if (bi_flags & BTRFS_INODE_IMMUTABLE) | |
9001 | stat->attributes |= STATX_ATTR_IMMUTABLE; | |
9002 | if (bi_flags & BTRFS_INODE_NODUMP) | |
9003 | stat->attributes |= STATX_ATTR_NODUMP; | |
14605409 BB |
9004 | if (bi_ro_flags & BTRFS_INODE_RO_VERITY) |
9005 | stat->attributes |= STATX_ATTR_VERITY; | |
04a87e34 YS |
9006 | |
9007 | stat->attributes_mask |= (STATX_ATTR_APPEND | | |
9008 | STATX_ATTR_COMPRESSED | | |
9009 | STATX_ATTR_IMMUTABLE | | |
9010 | STATX_ATTR_NODUMP); | |
fadc0d8b | 9011 | |
c020d2ea | 9012 | generic_fillattr(mnt_userns, inode, stat); |
0ee5dc67 | 9013 | stat->dev = BTRFS_I(inode)->root->anon_dev; |
df0af1a5 MX |
9014 | |
9015 | spin_lock(&BTRFS_I(inode)->lock); | |
a7e3b975 | 9016 | delalloc_bytes = BTRFS_I(inode)->new_delalloc_bytes; |
2766ff61 | 9017 | inode_bytes = inode_get_bytes(inode); |
df0af1a5 | 9018 | spin_unlock(&BTRFS_I(inode)->lock); |
2766ff61 | 9019 | stat->blocks = (ALIGN(inode_bytes, blocksize) + |
df0af1a5 | 9020 | ALIGN(delalloc_bytes, blocksize)) >> 9; |
39279cc3 CM |
9021 | return 0; |
9022 | } | |
9023 | ||
cdd1fedf DF |
9024 | static int btrfs_rename_exchange(struct inode *old_dir, |
9025 | struct dentry *old_dentry, | |
9026 | struct inode *new_dir, | |
9027 | struct dentry *new_dentry) | |
9028 | { | |
0b246afa | 9029 | struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb); |
cdd1fedf | 9030 | struct btrfs_trans_handle *trans; |
c1621871 | 9031 | unsigned int trans_num_items; |
cdd1fedf DF |
9032 | struct btrfs_root *root = BTRFS_I(old_dir)->root; |
9033 | struct btrfs_root *dest = BTRFS_I(new_dir)->root; | |
9034 | struct inode *new_inode = new_dentry->d_inode; | |
9035 | struct inode *old_inode = old_dentry->d_inode; | |
95582b00 | 9036 | struct timespec64 ctime = current_time(old_inode); |
88d2beec FM |
9037 | struct btrfs_rename_ctx old_rename_ctx; |
9038 | struct btrfs_rename_ctx new_rename_ctx; | |
4a0cc7ca NB |
9039 | u64 old_ino = btrfs_ino(BTRFS_I(old_inode)); |
9040 | u64 new_ino = btrfs_ino(BTRFS_I(new_inode)); | |
cdd1fedf DF |
9041 | u64 old_idx = 0; |
9042 | u64 new_idx = 0; | |
cdd1fedf | 9043 | int ret; |
75b463d2 | 9044 | int ret2; |
dc09ef35 | 9045 | bool need_abort = false; |
cdd1fedf | 9046 | |
3f79f6f6 N |
9047 | /* |
9048 | * For non-subvolumes allow exchange only within one subvolume, in the | |
9049 | * same inode namespace. Two subvolumes (represented as directory) can | |
9050 | * be exchanged as they're a logical link and have a fixed inode number. | |
9051 | */ | |
9052 | if (root != dest && | |
9053 | (old_ino != BTRFS_FIRST_FREE_OBJECTID || | |
9054 | new_ino != BTRFS_FIRST_FREE_OBJECTID)) | |
cdd1fedf DF |
9055 | return -EXDEV; |
9056 | ||
9057 | /* close the race window with snapshot create/destroy ioctl */ | |
943eb3bf JB |
9058 | if (old_ino == BTRFS_FIRST_FREE_OBJECTID || |
9059 | new_ino == BTRFS_FIRST_FREE_OBJECTID) | |
0b246afa | 9060 | down_read(&fs_info->subvol_sem); |
cdd1fedf DF |
9061 | |
9062 | /* | |
c1621871 OS |
9063 | * For each inode: |
9064 | * 1 to remove old dir item | |
9065 | * 1 to remove old dir index | |
9066 | * 1 to add new dir item | |
9067 | * 1 to add new dir index | |
9068 | * 1 to update parent inode | |
9069 | * | |
9070 | * If the parents are the same, we only need to account for one | |
cdd1fedf | 9071 | */ |
c1621871 OS |
9072 | trans_num_items = (old_dir == new_dir ? 9 : 10); |
9073 | if (old_ino == BTRFS_FIRST_FREE_OBJECTID) { | |
9074 | /* | |
9075 | * 1 to remove old root ref | |
9076 | * 1 to remove old root backref | |
9077 | * 1 to add new root ref | |
9078 | * 1 to add new root backref | |
9079 | */ | |
9080 | trans_num_items += 4; | |
9081 | } else { | |
9082 | /* | |
9083 | * 1 to update inode item | |
9084 | * 1 to remove old inode ref | |
9085 | * 1 to add new inode ref | |
9086 | */ | |
9087 | trans_num_items += 3; | |
9088 | } | |
9089 | if (new_ino == BTRFS_FIRST_FREE_OBJECTID) | |
9090 | trans_num_items += 4; | |
9091 | else | |
9092 | trans_num_items += 3; | |
9093 | trans = btrfs_start_transaction(root, trans_num_items); | |
cdd1fedf DF |
9094 | if (IS_ERR(trans)) { |
9095 | ret = PTR_ERR(trans); | |
9096 | goto out_notrans; | |
9097 | } | |
9098 | ||
00aa8e87 JB |
9099 | if (dest != root) { |
9100 | ret = btrfs_record_root_in_trans(trans, dest); | |
9101 | if (ret) | |
9102 | goto out_fail; | |
9103 | } | |
3e174099 | 9104 | |
cdd1fedf DF |
9105 | /* |
9106 | * We need to find a free sequence number both in the source and | |
9107 | * in the destination directory for the exchange. | |
9108 | */ | |
877574e2 | 9109 | ret = btrfs_set_inode_index(BTRFS_I(new_dir), &old_idx); |
cdd1fedf DF |
9110 | if (ret) |
9111 | goto out_fail; | |
877574e2 | 9112 | ret = btrfs_set_inode_index(BTRFS_I(old_dir), &new_idx); |
cdd1fedf DF |
9113 | if (ret) |
9114 | goto out_fail; | |
9115 | ||
9116 | BTRFS_I(old_inode)->dir_index = 0ULL; | |
9117 | BTRFS_I(new_inode)->dir_index = 0ULL; | |
9118 | ||
9119 | /* Reference for the source. */ | |
9120 | if (old_ino == BTRFS_FIRST_FREE_OBJECTID) { | |
9121 | /* force full log commit if subvolume involved. */ | |
90787766 | 9122 | btrfs_set_log_full_commit(trans); |
cdd1fedf DF |
9123 | } else { |
9124 | ret = btrfs_insert_inode_ref(trans, dest, | |
9125 | new_dentry->d_name.name, | |
9126 | new_dentry->d_name.len, | |
9127 | old_ino, | |
f85b7379 DS |
9128 | btrfs_ino(BTRFS_I(new_dir)), |
9129 | old_idx); | |
cdd1fedf DF |
9130 | if (ret) |
9131 | goto out_fail; | |
dc09ef35 | 9132 | need_abort = true; |
cdd1fedf DF |
9133 | } |
9134 | ||
9135 | /* And now for the dest. */ | |
9136 | if (new_ino == BTRFS_FIRST_FREE_OBJECTID) { | |
9137 | /* force full log commit if subvolume involved. */ | |
90787766 | 9138 | btrfs_set_log_full_commit(trans); |
cdd1fedf DF |
9139 | } else { |
9140 | ret = btrfs_insert_inode_ref(trans, root, | |
9141 | old_dentry->d_name.name, | |
9142 | old_dentry->d_name.len, | |
9143 | new_ino, | |
f85b7379 DS |
9144 | btrfs_ino(BTRFS_I(old_dir)), |
9145 | new_idx); | |
dc09ef35 JB |
9146 | if (ret) { |
9147 | if (need_abort) | |
9148 | btrfs_abort_transaction(trans, ret); | |
cdd1fedf | 9149 | goto out_fail; |
dc09ef35 | 9150 | } |
cdd1fedf DF |
9151 | } |
9152 | ||
9153 | /* Update inode version and ctime/mtime. */ | |
9154 | inode_inc_iversion(old_dir); | |
9155 | inode_inc_iversion(new_dir); | |
9156 | inode_inc_iversion(old_inode); | |
9157 | inode_inc_iversion(new_inode); | |
c1867eb3 DS |
9158 | old_dir->i_mtime = ctime; |
9159 | old_dir->i_ctime = ctime; | |
9160 | new_dir->i_mtime = ctime; | |
9161 | new_dir->i_ctime = ctime; | |
cdd1fedf DF |
9162 | old_inode->i_ctime = ctime; |
9163 | new_inode->i_ctime = ctime; | |
9164 | ||
9165 | if (old_dentry->d_parent != new_dentry->d_parent) { | |
f85b7379 DS |
9166 | btrfs_record_unlink_dir(trans, BTRFS_I(old_dir), |
9167 | BTRFS_I(old_inode), 1); | |
9168 | btrfs_record_unlink_dir(trans, BTRFS_I(new_dir), | |
9169 | BTRFS_I(new_inode), 1); | |
cdd1fedf DF |
9170 | } |
9171 | ||
9172 | /* src is a subvolume */ | |
9173 | if (old_ino == BTRFS_FIRST_FREE_OBJECTID) { | |
045d3967 | 9174 | ret = btrfs_unlink_subvol(trans, old_dir, old_dentry); |
cdd1fedf | 9175 | } else { /* src is an inode */ |
4467af88 | 9176 | ret = __btrfs_unlink_inode(trans, BTRFS_I(old_dir), |
4ec5934e | 9177 | BTRFS_I(old_dentry->d_inode), |
cdd1fedf | 9178 | old_dentry->d_name.name, |
88d2beec FM |
9179 | old_dentry->d_name.len, |
9180 | &old_rename_ctx); | |
cdd1fedf | 9181 | if (!ret) |
9a56fcd1 | 9182 | ret = btrfs_update_inode(trans, root, BTRFS_I(old_inode)); |
cdd1fedf DF |
9183 | } |
9184 | if (ret) { | |
66642832 | 9185 | btrfs_abort_transaction(trans, ret); |
cdd1fedf DF |
9186 | goto out_fail; |
9187 | } | |
9188 | ||
9189 | /* dest is a subvolume */ | |
9190 | if (new_ino == BTRFS_FIRST_FREE_OBJECTID) { | |
045d3967 | 9191 | ret = btrfs_unlink_subvol(trans, new_dir, new_dentry); |
cdd1fedf | 9192 | } else { /* dest is an inode */ |
4467af88 | 9193 | ret = __btrfs_unlink_inode(trans, BTRFS_I(new_dir), |
4ec5934e | 9194 | BTRFS_I(new_dentry->d_inode), |
cdd1fedf | 9195 | new_dentry->d_name.name, |
88d2beec FM |
9196 | new_dentry->d_name.len, |
9197 | &new_rename_ctx); | |
cdd1fedf | 9198 | if (!ret) |
9a56fcd1 | 9199 | ret = btrfs_update_inode(trans, dest, BTRFS_I(new_inode)); |
cdd1fedf DF |
9200 | } |
9201 | if (ret) { | |
66642832 | 9202 | btrfs_abort_transaction(trans, ret); |
cdd1fedf DF |
9203 | goto out_fail; |
9204 | } | |
9205 | ||
db0a669f | 9206 | ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode), |
cdd1fedf DF |
9207 | new_dentry->d_name.name, |
9208 | new_dentry->d_name.len, 0, old_idx); | |
9209 | if (ret) { | |
66642832 | 9210 | btrfs_abort_transaction(trans, ret); |
cdd1fedf DF |
9211 | goto out_fail; |
9212 | } | |
9213 | ||
db0a669f | 9214 | ret = btrfs_add_link(trans, BTRFS_I(old_dir), BTRFS_I(new_inode), |
cdd1fedf DF |
9215 | old_dentry->d_name.name, |
9216 | old_dentry->d_name.len, 0, new_idx); | |
9217 | if (ret) { | |
66642832 | 9218 | btrfs_abort_transaction(trans, ret); |
cdd1fedf DF |
9219 | goto out_fail; |
9220 | } | |
9221 | ||
9222 | if (old_inode->i_nlink == 1) | |
9223 | BTRFS_I(old_inode)->dir_index = old_idx; | |
9224 | if (new_inode->i_nlink == 1) | |
9225 | BTRFS_I(new_inode)->dir_index = new_idx; | |
9226 | ||
259c4b96 FM |
9227 | /* |
9228 | * Now pin the logs of the roots. We do it to ensure that no other task | |
9229 | * can sync the logs while we are in progress with the rename, because | |
9230 | * that could result in an inconsistency in case any of the inodes that | |
9231 | * are part of this rename operation were logged before. | |
9232 | */ | |
9233 | if (old_ino != BTRFS_FIRST_FREE_OBJECTID) | |
9234 | btrfs_pin_log_trans(root); | |
9235 | if (new_ino != BTRFS_FIRST_FREE_OBJECTID) | |
9236 | btrfs_pin_log_trans(dest); | |
9237 | ||
9238 | /* Do the log updates for all inodes. */ | |
9239 | if (old_ino != BTRFS_FIRST_FREE_OBJECTID) | |
d5f5bd54 | 9240 | btrfs_log_new_name(trans, old_dentry, BTRFS_I(old_dir), |
88d2beec | 9241 | old_rename_ctx.index, new_dentry->d_parent); |
259c4b96 | 9242 | if (new_ino != BTRFS_FIRST_FREE_OBJECTID) |
d5f5bd54 | 9243 | btrfs_log_new_name(trans, new_dentry, BTRFS_I(new_dir), |
88d2beec | 9244 | new_rename_ctx.index, old_dentry->d_parent); |
259c4b96 FM |
9245 | |
9246 | /* Now unpin the logs. */ | |
9247 | if (old_ino != BTRFS_FIRST_FREE_OBJECTID) | |
9248 | btrfs_end_log_trans(root); | |
9249 | if (new_ino != BTRFS_FIRST_FREE_OBJECTID) | |
cdd1fedf | 9250 | btrfs_end_log_trans(dest); |
cdd1fedf | 9251 | out_fail: |
75b463d2 FM |
9252 | ret2 = btrfs_end_transaction(trans); |
9253 | ret = ret ? ret : ret2; | |
cdd1fedf | 9254 | out_notrans: |
943eb3bf JB |
9255 | if (new_ino == BTRFS_FIRST_FREE_OBJECTID || |
9256 | old_ino == BTRFS_FIRST_FREE_OBJECTID) | |
0b246afa | 9257 | up_read(&fs_info->subvol_sem); |
cdd1fedf DF |
9258 | |
9259 | return ret; | |
9260 | } | |
9261 | ||
a1fd0c35 OS |
9262 | static struct inode *new_whiteout_inode(struct user_namespace *mnt_userns, |
9263 | struct inode *dir) | |
9264 | { | |
9265 | struct inode *inode; | |
9266 | ||
9267 | inode = new_inode(dir->i_sb); | |
9268 | if (inode) { | |
9269 | inode_init_owner(mnt_userns, inode, dir, | |
9270 | S_IFCHR | WHITEOUT_MODE); | |
9271 | inode->i_op = &btrfs_special_inode_operations; | |
9272 | init_special_inode(inode, inode->i_mode, WHITEOUT_DEV); | |
9273 | } | |
9274 | return inode; | |
9275 | } | |
9276 | ||
ca07274c CB |
9277 | static int btrfs_rename(struct user_namespace *mnt_userns, |
9278 | struct inode *old_dir, struct dentry *old_dentry, | |
9279 | struct inode *new_dir, struct dentry *new_dentry, | |
9280 | unsigned int flags) | |
39279cc3 | 9281 | { |
0b246afa | 9282 | struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb); |
3538d68d OS |
9283 | struct btrfs_new_inode_args whiteout_args = { |
9284 | .dir = old_dir, | |
9285 | .dentry = old_dentry, | |
9286 | }; | |
39279cc3 | 9287 | struct btrfs_trans_handle *trans; |
5062af35 | 9288 | unsigned int trans_num_items; |
39279cc3 | 9289 | struct btrfs_root *root = BTRFS_I(old_dir)->root; |
4df27c4d | 9290 | struct btrfs_root *dest = BTRFS_I(new_dir)->root; |
2b0143b5 DH |
9291 | struct inode *new_inode = d_inode(new_dentry); |
9292 | struct inode *old_inode = d_inode(old_dentry); | |
88d2beec | 9293 | struct btrfs_rename_ctx rename_ctx; |
00e4e6b3 | 9294 | u64 index = 0; |
39279cc3 | 9295 | int ret; |
75b463d2 | 9296 | int ret2; |
4a0cc7ca | 9297 | u64 old_ino = btrfs_ino(BTRFS_I(old_inode)); |
39279cc3 | 9298 | |
4a0cc7ca | 9299 | if (btrfs_ino(BTRFS_I(new_dir)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) |
f679a840 YZ |
9300 | return -EPERM; |
9301 | ||
4df27c4d | 9302 | /* we only allow rename subvolume link between subvolumes */ |
33345d01 | 9303 | if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest) |
3394e160 CM |
9304 | return -EXDEV; |
9305 | ||
33345d01 | 9306 | if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID || |
4a0cc7ca | 9307 | (new_inode && btrfs_ino(BTRFS_I(new_inode)) == BTRFS_FIRST_FREE_OBJECTID)) |
39279cc3 | 9308 | return -ENOTEMPTY; |
5f39d397 | 9309 | |
4df27c4d YZ |
9310 | if (S_ISDIR(old_inode->i_mode) && new_inode && |
9311 | new_inode->i_size > BTRFS_EMPTY_DIR_SIZE) | |
9312 | return -ENOTEMPTY; | |
9c52057c CM |
9313 | |
9314 | ||
9315 | /* check for collisions, even if the name isn't there */ | |
4871c158 | 9316 | ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino, |
9c52057c CM |
9317 | new_dentry->d_name.name, |
9318 | new_dentry->d_name.len); | |
9319 | ||
9320 | if (ret) { | |
9321 | if (ret == -EEXIST) { | |
9322 | /* we shouldn't get | |
9323 | * eexist without a new_inode */ | |
fae7f21c | 9324 | if (WARN_ON(!new_inode)) { |
9c52057c CM |
9325 | return ret; |
9326 | } | |
9327 | } else { | |
9328 | /* maybe -EOVERFLOW */ | |
9329 | return ret; | |
9330 | } | |
9331 | } | |
9332 | ret = 0; | |
9333 | ||
5a3f23d5 | 9334 | /* |
8d875f95 CM |
9335 | * we're using rename to replace one file with another. Start IO on it |
9336 | * now so we don't add too much work to the end of the transaction | |
5a3f23d5 | 9337 | */ |
8d875f95 | 9338 | if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size) |
5a3f23d5 CM |
9339 | filemap_flush(old_inode->i_mapping); |
9340 | ||
a1fd0c35 | 9341 | if (flags & RENAME_WHITEOUT) { |
3538d68d OS |
9342 | whiteout_args.inode = new_whiteout_inode(mnt_userns, old_dir); |
9343 | if (!whiteout_args.inode) | |
a1fd0c35 | 9344 | return -ENOMEM; |
3538d68d OS |
9345 | ret = btrfs_new_inode_prepare(&whiteout_args, &trans_num_items); |
9346 | if (ret) | |
9347 | goto out_whiteout_inode; | |
9348 | } else { | |
9349 | /* 1 to update the old parent inode. */ | |
9350 | trans_num_items = 1; | |
a1fd0c35 OS |
9351 | } |
9352 | ||
c1621871 OS |
9353 | if (old_ino == BTRFS_FIRST_FREE_OBJECTID) { |
9354 | /* Close the race window with snapshot create/destroy ioctl */ | |
0b246afa | 9355 | down_read(&fs_info->subvol_sem); |
c1621871 OS |
9356 | /* |
9357 | * 1 to remove old root ref | |
9358 | * 1 to remove old root backref | |
9359 | * 1 to add new root ref | |
9360 | * 1 to add new root backref | |
9361 | */ | |
3538d68d | 9362 | trans_num_items += 4; |
c1621871 OS |
9363 | } else { |
9364 | /* | |
9365 | * 1 to update inode | |
9366 | * 1 to remove old inode ref | |
9367 | * 1 to add new inode ref | |
9368 | */ | |
3538d68d | 9369 | trans_num_items += 3; |
c1621871 | 9370 | } |
a22285a6 | 9371 | /* |
c1621871 OS |
9372 | * 1 to remove old dir item |
9373 | * 1 to remove old dir index | |
c1621871 OS |
9374 | * 1 to add new dir item |
9375 | * 1 to add new dir index | |
a22285a6 | 9376 | */ |
3538d68d OS |
9377 | trans_num_items += 4; |
9378 | /* 1 to update new parent inode if it's not the same as the old parent */ | |
c1621871 OS |
9379 | if (new_dir != old_dir) |
9380 | trans_num_items++; | |
9381 | if (new_inode) { | |
9382 | /* | |
9383 | * 1 to update inode | |
9384 | * 1 to remove inode ref | |
9385 | * 1 to remove dir item | |
9386 | * 1 to remove dir index | |
9387 | * 1 to possibly add orphan item | |
9388 | */ | |
9389 | trans_num_items += 5; | |
9390 | } | |
5062af35 | 9391 | trans = btrfs_start_transaction(root, trans_num_items); |
b44c59a8 | 9392 | if (IS_ERR(trans)) { |
cdd1fedf DF |
9393 | ret = PTR_ERR(trans); |
9394 | goto out_notrans; | |
9395 | } | |
76dda93c | 9396 | |
b0fec6fd JB |
9397 | if (dest != root) { |
9398 | ret = btrfs_record_root_in_trans(trans, dest); | |
9399 | if (ret) | |
9400 | goto out_fail; | |
9401 | } | |
5f39d397 | 9402 | |
877574e2 | 9403 | ret = btrfs_set_inode_index(BTRFS_I(new_dir), &index); |
a5719521 YZ |
9404 | if (ret) |
9405 | goto out_fail; | |
5a3f23d5 | 9406 | |
67de1176 | 9407 | BTRFS_I(old_inode)->dir_index = 0ULL; |
33345d01 | 9408 | if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) { |
4df27c4d | 9409 | /* force full log commit if subvolume involved. */ |
90787766 | 9410 | btrfs_set_log_full_commit(trans); |
4df27c4d | 9411 | } else { |
a5719521 YZ |
9412 | ret = btrfs_insert_inode_ref(trans, dest, |
9413 | new_dentry->d_name.name, | |
9414 | new_dentry->d_name.len, | |
33345d01 | 9415 | old_ino, |
4a0cc7ca | 9416 | btrfs_ino(BTRFS_I(new_dir)), index); |
a5719521 YZ |
9417 | if (ret) |
9418 | goto out_fail; | |
4df27c4d | 9419 | } |
5a3f23d5 | 9420 | |
0c4d2d95 JB |
9421 | inode_inc_iversion(old_dir); |
9422 | inode_inc_iversion(new_dir); | |
9423 | inode_inc_iversion(old_inode); | |
c1867eb3 DS |
9424 | old_dir->i_mtime = current_time(old_dir); |
9425 | old_dir->i_ctime = old_dir->i_mtime; | |
9426 | new_dir->i_mtime = old_dir->i_mtime; | |
9427 | new_dir->i_ctime = old_dir->i_mtime; | |
9428 | old_inode->i_ctime = old_dir->i_mtime; | |
5f39d397 | 9429 | |
12fcfd22 | 9430 | if (old_dentry->d_parent != new_dentry->d_parent) |
f85b7379 DS |
9431 | btrfs_record_unlink_dir(trans, BTRFS_I(old_dir), |
9432 | BTRFS_I(old_inode), 1); | |
12fcfd22 | 9433 | |
33345d01 | 9434 | if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) { |
045d3967 | 9435 | ret = btrfs_unlink_subvol(trans, old_dir, old_dentry); |
4df27c4d | 9436 | } else { |
4467af88 | 9437 | ret = __btrfs_unlink_inode(trans, BTRFS_I(old_dir), |
4ec5934e | 9438 | BTRFS_I(d_inode(old_dentry)), |
92986796 | 9439 | old_dentry->d_name.name, |
88d2beec FM |
9440 | old_dentry->d_name.len, |
9441 | &rename_ctx); | |
92986796 | 9442 | if (!ret) |
9a56fcd1 | 9443 | ret = btrfs_update_inode(trans, root, BTRFS_I(old_inode)); |
4df27c4d | 9444 | } |
79787eaa | 9445 | if (ret) { |
66642832 | 9446 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
9447 | goto out_fail; |
9448 | } | |
39279cc3 CM |
9449 | |
9450 | if (new_inode) { | |
0c4d2d95 | 9451 | inode_inc_iversion(new_inode); |
c2050a45 | 9452 | new_inode->i_ctime = current_time(new_inode); |
4a0cc7ca | 9453 | if (unlikely(btrfs_ino(BTRFS_I(new_inode)) == |
4df27c4d | 9454 | BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) { |
045d3967 | 9455 | ret = btrfs_unlink_subvol(trans, new_dir, new_dentry); |
4df27c4d YZ |
9456 | BUG_ON(new_inode->i_nlink == 0); |
9457 | } else { | |
4467af88 | 9458 | ret = btrfs_unlink_inode(trans, BTRFS_I(new_dir), |
4ec5934e | 9459 | BTRFS_I(d_inode(new_dentry)), |
4df27c4d YZ |
9460 | new_dentry->d_name.name, |
9461 | new_dentry->d_name.len); | |
9462 | } | |
4ef31a45 | 9463 | if (!ret && new_inode->i_nlink == 0) |
73f2e545 NB |
9464 | ret = btrfs_orphan_add(trans, |
9465 | BTRFS_I(d_inode(new_dentry))); | |
79787eaa | 9466 | if (ret) { |
66642832 | 9467 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
9468 | goto out_fail; |
9469 | } | |
39279cc3 | 9470 | } |
aec7477b | 9471 | |
db0a669f | 9472 | ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode), |
4df27c4d | 9473 | new_dentry->d_name.name, |
a5719521 | 9474 | new_dentry->d_name.len, 0, index); |
79787eaa | 9475 | if (ret) { |
66642832 | 9476 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
9477 | goto out_fail; |
9478 | } | |
39279cc3 | 9479 | |
67de1176 MX |
9480 | if (old_inode->i_nlink == 1) |
9481 | BTRFS_I(old_inode)->dir_index = index; | |
9482 | ||
259c4b96 | 9483 | if (old_ino != BTRFS_FIRST_FREE_OBJECTID) |
d5f5bd54 | 9484 | btrfs_log_new_name(trans, old_dentry, BTRFS_I(old_dir), |
88d2beec | 9485 | rename_ctx.index, new_dentry->d_parent); |
cdd1fedf DF |
9486 | |
9487 | if (flags & RENAME_WHITEOUT) { | |
caae78e0 | 9488 | ret = btrfs_create_new_inode(trans, &whiteout_args); |
cdd1fedf | 9489 | if (ret) { |
66642832 | 9490 | btrfs_abort_transaction(trans, ret); |
cdd1fedf | 9491 | goto out_fail; |
caae78e0 OS |
9492 | } else { |
9493 | unlock_new_inode(whiteout_args.inode); | |
9494 | iput(whiteout_args.inode); | |
9495 | whiteout_args.inode = NULL; | |
cdd1fedf | 9496 | } |
4df27c4d | 9497 | } |
39279cc3 | 9498 | out_fail: |
75b463d2 FM |
9499 | ret2 = btrfs_end_transaction(trans); |
9500 | ret = ret ? ret : ret2; | |
b44c59a8 | 9501 | out_notrans: |
33345d01 | 9502 | if (old_ino == BTRFS_FIRST_FREE_OBJECTID) |
0b246afa | 9503 | up_read(&fs_info->subvol_sem); |
a1fd0c35 | 9504 | if (flags & RENAME_WHITEOUT) |
3538d68d OS |
9505 | btrfs_new_inode_args_destroy(&whiteout_args); |
9506 | out_whiteout_inode: | |
9507 | if (flags & RENAME_WHITEOUT) | |
9508 | iput(whiteout_args.inode); | |
39279cc3 CM |
9509 | return ret; |
9510 | } | |
9511 | ||
549c7297 CB |
9512 | static int btrfs_rename2(struct user_namespace *mnt_userns, struct inode *old_dir, |
9513 | struct dentry *old_dentry, struct inode *new_dir, | |
9514 | struct dentry *new_dentry, unsigned int flags) | |
80ace85c | 9515 | { |
ca6dee6b FM |
9516 | int ret; |
9517 | ||
cdd1fedf | 9518 | if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT)) |
80ace85c MS |
9519 | return -EINVAL; |
9520 | ||
cdd1fedf | 9521 | if (flags & RENAME_EXCHANGE) |
ca6dee6b FM |
9522 | ret = btrfs_rename_exchange(old_dir, old_dentry, new_dir, |
9523 | new_dentry); | |
9524 | else | |
9525 | ret = btrfs_rename(mnt_userns, old_dir, old_dentry, new_dir, | |
9526 | new_dentry, flags); | |
cdd1fedf | 9527 | |
ca6dee6b FM |
9528 | btrfs_btree_balance_dirty(BTRFS_I(new_dir)->root->fs_info); |
9529 | ||
9530 | return ret; | |
80ace85c MS |
9531 | } |
9532 | ||
3a2f8c07 NB |
9533 | struct btrfs_delalloc_work { |
9534 | struct inode *inode; | |
9535 | struct completion completion; | |
9536 | struct list_head list; | |
9537 | struct btrfs_work work; | |
9538 | }; | |
9539 | ||
8ccf6f19 MX |
9540 | static void btrfs_run_delalloc_work(struct btrfs_work *work) |
9541 | { | |
9542 | struct btrfs_delalloc_work *delalloc_work; | |
9f23e289 | 9543 | struct inode *inode; |
8ccf6f19 MX |
9544 | |
9545 | delalloc_work = container_of(work, struct btrfs_delalloc_work, | |
9546 | work); | |
9f23e289 | 9547 | inode = delalloc_work->inode; |
30424601 DS |
9548 | filemap_flush(inode->i_mapping); |
9549 | if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT, | |
9550 | &BTRFS_I(inode)->runtime_flags)) | |
9f23e289 | 9551 | filemap_flush(inode->i_mapping); |
8ccf6f19 | 9552 | |
076da91c | 9553 | iput(inode); |
8ccf6f19 MX |
9554 | complete(&delalloc_work->completion); |
9555 | } | |
9556 | ||
3a2f8c07 | 9557 | static struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode) |
8ccf6f19 MX |
9558 | { |
9559 | struct btrfs_delalloc_work *work; | |
9560 | ||
100d5702 | 9561 | work = kmalloc(sizeof(*work), GFP_NOFS); |
8ccf6f19 MX |
9562 | if (!work) |
9563 | return NULL; | |
9564 | ||
9565 | init_completion(&work->completion); | |
9566 | INIT_LIST_HEAD(&work->list); | |
9567 | work->inode = inode; | |
a0cac0ec | 9568 | btrfs_init_work(&work->work, btrfs_run_delalloc_work, NULL, NULL); |
8ccf6f19 MX |
9569 | |
9570 | return work; | |
9571 | } | |
9572 | ||
d352ac68 CM |
9573 | /* |
9574 | * some fairly slow code that needs optimization. This walks the list | |
9575 | * of all the inodes with pending delalloc and forces them to disk. | |
9576 | */ | |
e076ab2a JB |
9577 | static int start_delalloc_inodes(struct btrfs_root *root, |
9578 | struct writeback_control *wbc, bool snapshot, | |
3d45f221 | 9579 | bool in_reclaim_context) |
ea8c2819 | 9580 | { |
ea8c2819 | 9581 | struct btrfs_inode *binode; |
5b21f2ed | 9582 | struct inode *inode; |
8ccf6f19 MX |
9583 | struct btrfs_delalloc_work *work, *next; |
9584 | struct list_head works; | |
1eafa6c7 | 9585 | struct list_head splice; |
8ccf6f19 | 9586 | int ret = 0; |
e076ab2a | 9587 | bool full_flush = wbc->nr_to_write == LONG_MAX; |
ea8c2819 | 9588 | |
8ccf6f19 | 9589 | INIT_LIST_HEAD(&works); |
1eafa6c7 | 9590 | INIT_LIST_HEAD(&splice); |
63607cc8 | 9591 | |
573bfb72 | 9592 | mutex_lock(&root->delalloc_mutex); |
eb73c1b7 MX |
9593 | spin_lock(&root->delalloc_lock); |
9594 | list_splice_init(&root->delalloc_inodes, &splice); | |
1eafa6c7 MX |
9595 | while (!list_empty(&splice)) { |
9596 | binode = list_entry(splice.next, struct btrfs_inode, | |
ea8c2819 | 9597 | delalloc_inodes); |
1eafa6c7 | 9598 | |
eb73c1b7 MX |
9599 | list_move_tail(&binode->delalloc_inodes, |
9600 | &root->delalloc_inodes); | |
3d45f221 FM |
9601 | |
9602 | if (in_reclaim_context && | |
9603 | test_bit(BTRFS_INODE_NO_DELALLOC_FLUSH, &binode->runtime_flags)) | |
9604 | continue; | |
9605 | ||
5b21f2ed | 9606 | inode = igrab(&binode->vfs_inode); |
df0af1a5 | 9607 | if (!inode) { |
eb73c1b7 | 9608 | cond_resched_lock(&root->delalloc_lock); |
1eafa6c7 | 9609 | continue; |
df0af1a5 | 9610 | } |
eb73c1b7 | 9611 | spin_unlock(&root->delalloc_lock); |
1eafa6c7 | 9612 | |
3cd24c69 EL |
9613 | if (snapshot) |
9614 | set_bit(BTRFS_INODE_SNAPSHOT_FLUSH, | |
9615 | &binode->runtime_flags); | |
e076ab2a JB |
9616 | if (full_flush) { |
9617 | work = btrfs_alloc_delalloc_work(inode); | |
9618 | if (!work) { | |
9619 | iput(inode); | |
9620 | ret = -ENOMEM; | |
9621 | goto out; | |
9622 | } | |
9623 | list_add_tail(&work->list, &works); | |
9624 | btrfs_queue_work(root->fs_info->flush_workers, | |
9625 | &work->work); | |
9626 | } else { | |
b3776305 | 9627 | ret = filemap_fdatawrite_wbc(inode->i_mapping, wbc); |
e076ab2a JB |
9628 | btrfs_add_delayed_iput(inode); |
9629 | if (ret || wbc->nr_to_write <= 0) | |
b4912139 JB |
9630 | goto out; |
9631 | } | |
5b21f2ed | 9632 | cond_resched(); |
eb73c1b7 | 9633 | spin_lock(&root->delalloc_lock); |
ea8c2819 | 9634 | } |
eb73c1b7 | 9635 | spin_unlock(&root->delalloc_lock); |
8c8bee1d | 9636 | |
a1ecaabb | 9637 | out: |
eb73c1b7 MX |
9638 | list_for_each_entry_safe(work, next, &works, list) { |
9639 | list_del_init(&work->list); | |
40012f96 NB |
9640 | wait_for_completion(&work->completion); |
9641 | kfree(work); | |
eb73c1b7 MX |
9642 | } |
9643 | ||
81f1d390 | 9644 | if (!list_empty(&splice)) { |
eb73c1b7 MX |
9645 | spin_lock(&root->delalloc_lock); |
9646 | list_splice_tail(&splice, &root->delalloc_inodes); | |
9647 | spin_unlock(&root->delalloc_lock); | |
9648 | } | |
573bfb72 | 9649 | mutex_unlock(&root->delalloc_mutex); |
eb73c1b7 MX |
9650 | return ret; |
9651 | } | |
1eafa6c7 | 9652 | |
f9baa501 | 9653 | int btrfs_start_delalloc_snapshot(struct btrfs_root *root, bool in_reclaim_context) |
eb73c1b7 | 9654 | { |
e076ab2a JB |
9655 | struct writeback_control wbc = { |
9656 | .nr_to_write = LONG_MAX, | |
9657 | .sync_mode = WB_SYNC_NONE, | |
9658 | .range_start = 0, | |
9659 | .range_end = LLONG_MAX, | |
9660 | }; | |
0b246afa | 9661 | struct btrfs_fs_info *fs_info = root->fs_info; |
1eafa6c7 | 9662 | |
84961539 | 9663 | if (BTRFS_FS_ERROR(fs_info)) |
eb73c1b7 MX |
9664 | return -EROFS; |
9665 | ||
f9baa501 | 9666 | return start_delalloc_inodes(root, &wbc, true, in_reclaim_context); |
eb73c1b7 MX |
9667 | } |
9668 | ||
9db4dc24 | 9669 | int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, long nr, |
3d45f221 | 9670 | bool in_reclaim_context) |
eb73c1b7 | 9671 | { |
e076ab2a | 9672 | struct writeback_control wbc = { |
9db4dc24 | 9673 | .nr_to_write = nr, |
e076ab2a JB |
9674 | .sync_mode = WB_SYNC_NONE, |
9675 | .range_start = 0, | |
9676 | .range_end = LLONG_MAX, | |
9677 | }; | |
eb73c1b7 MX |
9678 | struct btrfs_root *root; |
9679 | struct list_head splice; | |
9680 | int ret; | |
9681 | ||
84961539 | 9682 | if (BTRFS_FS_ERROR(fs_info)) |
eb73c1b7 MX |
9683 | return -EROFS; |
9684 | ||
9685 | INIT_LIST_HEAD(&splice); | |
9686 | ||
573bfb72 | 9687 | mutex_lock(&fs_info->delalloc_root_mutex); |
eb73c1b7 MX |
9688 | spin_lock(&fs_info->delalloc_root_lock); |
9689 | list_splice_init(&fs_info->delalloc_roots, &splice); | |
d7830b71 | 9690 | while (!list_empty(&splice)) { |
e076ab2a JB |
9691 | /* |
9692 | * Reset nr_to_write here so we know that we're doing a full | |
9693 | * flush. | |
9694 | */ | |
9db4dc24 | 9695 | if (nr == LONG_MAX) |
e076ab2a JB |
9696 | wbc.nr_to_write = LONG_MAX; |
9697 | ||
eb73c1b7 MX |
9698 | root = list_first_entry(&splice, struct btrfs_root, |
9699 | delalloc_root); | |
00246528 | 9700 | root = btrfs_grab_root(root); |
eb73c1b7 MX |
9701 | BUG_ON(!root); |
9702 | list_move_tail(&root->delalloc_root, | |
9703 | &fs_info->delalloc_roots); | |
9704 | spin_unlock(&fs_info->delalloc_root_lock); | |
9705 | ||
e076ab2a | 9706 | ret = start_delalloc_inodes(root, &wbc, false, in_reclaim_context); |
00246528 | 9707 | btrfs_put_root(root); |
e076ab2a | 9708 | if (ret < 0 || wbc.nr_to_write <= 0) |
eb73c1b7 | 9709 | goto out; |
eb73c1b7 | 9710 | spin_lock(&fs_info->delalloc_root_lock); |
8ccf6f19 | 9711 | } |
eb73c1b7 | 9712 | spin_unlock(&fs_info->delalloc_root_lock); |
1eafa6c7 | 9713 | |
6c255e67 | 9714 | ret = 0; |
eb73c1b7 | 9715 | out: |
81f1d390 | 9716 | if (!list_empty(&splice)) { |
eb73c1b7 MX |
9717 | spin_lock(&fs_info->delalloc_root_lock); |
9718 | list_splice_tail(&splice, &fs_info->delalloc_roots); | |
9719 | spin_unlock(&fs_info->delalloc_root_lock); | |
1eafa6c7 | 9720 | } |
573bfb72 | 9721 | mutex_unlock(&fs_info->delalloc_root_mutex); |
8ccf6f19 | 9722 | return ret; |
ea8c2819 CM |
9723 | } |
9724 | ||
549c7297 CB |
9725 | static int btrfs_symlink(struct user_namespace *mnt_userns, struct inode *dir, |
9726 | struct dentry *dentry, const char *symname) | |
39279cc3 | 9727 | { |
0b246afa | 9728 | struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb); |
39279cc3 CM |
9729 | struct btrfs_trans_handle *trans; |
9730 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
9731 | struct btrfs_path *path; | |
9732 | struct btrfs_key key; | |
a1fd0c35 | 9733 | struct inode *inode; |
3538d68d OS |
9734 | struct btrfs_new_inode_args new_inode_args = { |
9735 | .dir = dir, | |
9736 | .dentry = dentry, | |
9737 | }; | |
9738 | unsigned int trans_num_items; | |
39279cc3 | 9739 | int err; |
39279cc3 CM |
9740 | int name_len; |
9741 | int datasize; | |
5f39d397 | 9742 | unsigned long ptr; |
39279cc3 | 9743 | struct btrfs_file_extent_item *ei; |
5f39d397 | 9744 | struct extent_buffer *leaf; |
39279cc3 | 9745 | |
f06becc4 | 9746 | name_len = strlen(symname); |
0b246afa | 9747 | if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info)) |
39279cc3 | 9748 | return -ENAMETOOLONG; |
1832a6d5 | 9749 | |
a1fd0c35 OS |
9750 | inode = new_inode(dir->i_sb); |
9751 | if (!inode) | |
9752 | return -ENOMEM; | |
9753 | inode_init_owner(mnt_userns, inode, dir, S_IFLNK | S_IRWXUGO); | |
9754 | inode->i_op = &btrfs_symlink_inode_operations; | |
9755 | inode_nohighmem(inode); | |
9756 | inode->i_mapping->a_ops = &btrfs_aops; | |
caae78e0 OS |
9757 | btrfs_i_size_write(BTRFS_I(inode), name_len); |
9758 | inode_set_bytes(inode, name_len); | |
a1fd0c35 | 9759 | |
3538d68d OS |
9760 | new_inode_args.inode = inode; |
9761 | err = btrfs_new_inode_prepare(&new_inode_args, &trans_num_items); | |
caae78e0 OS |
9762 | if (err) |
9763 | goto out_inode; | |
3538d68d OS |
9764 | /* 1 additional item for the inline extent */ |
9765 | trans_num_items++; | |
9766 | ||
9767 | trans = btrfs_start_transaction(root, trans_num_items); | |
a1fd0c35 | 9768 | if (IS_ERR(trans)) { |
3538d68d OS |
9769 | err = PTR_ERR(trans); |
9770 | goto out_new_inode_args; | |
a1fd0c35 | 9771 | } |
1832a6d5 | 9772 | |
caae78e0 | 9773 | err = btrfs_create_new_inode(trans, &new_inode_args); |
b0d5d10f | 9774 | if (err) |
caae78e0 | 9775 | goto out; |
ad19db71 | 9776 | |
39279cc3 | 9777 | path = btrfs_alloc_path(); |
d8926bb3 MF |
9778 | if (!path) { |
9779 | err = -ENOMEM; | |
caae78e0 OS |
9780 | btrfs_abort_transaction(trans, err); |
9781 | discard_new_inode(inode); | |
9782 | inode = NULL; | |
9783 | goto out; | |
d8926bb3 | 9784 | } |
4a0cc7ca | 9785 | key.objectid = btrfs_ino(BTRFS_I(inode)); |
39279cc3 | 9786 | key.offset = 0; |
962a298f | 9787 | key.type = BTRFS_EXTENT_DATA_KEY; |
39279cc3 CM |
9788 | datasize = btrfs_file_extent_calc_inline_size(name_len); |
9789 | err = btrfs_insert_empty_item(trans, root, path, &key, | |
9790 | datasize); | |
54aa1f4d | 9791 | if (err) { |
caae78e0 | 9792 | btrfs_abort_transaction(trans, err); |
b0839166 | 9793 | btrfs_free_path(path); |
caae78e0 OS |
9794 | discard_new_inode(inode); |
9795 | inode = NULL; | |
9796 | goto out; | |
54aa1f4d | 9797 | } |
5f39d397 CM |
9798 | leaf = path->nodes[0]; |
9799 | ei = btrfs_item_ptr(leaf, path->slots[0], | |
9800 | struct btrfs_file_extent_item); | |
9801 | btrfs_set_file_extent_generation(leaf, ei, trans->transid); | |
9802 | btrfs_set_file_extent_type(leaf, ei, | |
39279cc3 | 9803 | BTRFS_FILE_EXTENT_INLINE); |
c8b97818 CM |
9804 | btrfs_set_file_extent_encryption(leaf, ei, 0); |
9805 | btrfs_set_file_extent_compression(leaf, ei, 0); | |
9806 | btrfs_set_file_extent_other_encoding(leaf, ei, 0); | |
9807 | btrfs_set_file_extent_ram_bytes(leaf, ei, name_len); | |
9808 | ||
39279cc3 | 9809 | ptr = btrfs_file_extent_inline_start(ei); |
5f39d397 CM |
9810 | write_extent_buffer(leaf, symname, ptr, name_len); |
9811 | btrfs_mark_buffer_dirty(leaf); | |
39279cc3 | 9812 | btrfs_free_path(path); |
5f39d397 | 9813 | |
1e2e547a | 9814 | d_instantiate_new(dentry, inode); |
caae78e0 OS |
9815 | err = 0; |
9816 | out: | |
3a45bb20 | 9817 | btrfs_end_transaction(trans); |
2ff7e61e | 9818 | btrfs_btree_balance_dirty(fs_info); |
3538d68d OS |
9819 | out_new_inode_args: |
9820 | btrfs_new_inode_args_destroy(&new_inode_args); | |
caae78e0 OS |
9821 | out_inode: |
9822 | if (err) | |
9823 | iput(inode); | |
39279cc3 CM |
9824 | return err; |
9825 | } | |
16432985 | 9826 | |
8fccebfa FM |
9827 | static struct btrfs_trans_handle *insert_prealloc_file_extent( |
9828 | struct btrfs_trans_handle *trans_in, | |
90dffd0c NB |
9829 | struct btrfs_inode *inode, |
9830 | struct btrfs_key *ins, | |
203f44c5 QW |
9831 | u64 file_offset) |
9832 | { | |
9833 | struct btrfs_file_extent_item stack_fi; | |
bf385648 | 9834 | struct btrfs_replace_extent_info extent_info; |
8fccebfa FM |
9835 | struct btrfs_trans_handle *trans = trans_in; |
9836 | struct btrfs_path *path; | |
203f44c5 QW |
9837 | u64 start = ins->objectid; |
9838 | u64 len = ins->offset; | |
fbf48bb0 | 9839 | int qgroup_released; |
9729f10a | 9840 | int ret; |
203f44c5 QW |
9841 | |
9842 | memset(&stack_fi, 0, sizeof(stack_fi)); | |
9843 | ||
9844 | btrfs_set_stack_file_extent_type(&stack_fi, BTRFS_FILE_EXTENT_PREALLOC); | |
9845 | btrfs_set_stack_file_extent_disk_bytenr(&stack_fi, start); | |
9846 | btrfs_set_stack_file_extent_disk_num_bytes(&stack_fi, len); | |
9847 | btrfs_set_stack_file_extent_num_bytes(&stack_fi, len); | |
9848 | btrfs_set_stack_file_extent_ram_bytes(&stack_fi, len); | |
9849 | btrfs_set_stack_file_extent_compression(&stack_fi, BTRFS_COMPRESS_NONE); | |
9850 | /* Encryption and other encoding is reserved and all 0 */ | |
9851 | ||
fbf48bb0 QW |
9852 | qgroup_released = btrfs_qgroup_release_data(inode, file_offset, len); |
9853 | if (qgroup_released < 0) | |
9854 | return ERR_PTR(qgroup_released); | |
8fccebfa FM |
9855 | |
9856 | if (trans) { | |
90dffd0c | 9857 | ret = insert_reserved_file_extent(trans, inode, |
2766ff61 | 9858 | file_offset, &stack_fi, |
fbf48bb0 | 9859 | true, qgroup_released); |
8fccebfa | 9860 | if (ret) |
a3ee79bd | 9861 | goto free_qgroup; |
8fccebfa FM |
9862 | return trans; |
9863 | } | |
9864 | ||
9865 | extent_info.disk_offset = start; | |
9866 | extent_info.disk_len = len; | |
9867 | extent_info.data_offset = 0; | |
9868 | extent_info.data_len = len; | |
9869 | extent_info.file_offset = file_offset; | |
9870 | extent_info.extent_buf = (char *)&stack_fi; | |
8fccebfa | 9871 | extent_info.is_new_extent = true; |
983d8209 | 9872 | extent_info.update_times = true; |
fbf48bb0 | 9873 | extent_info.qgroup_reserved = qgroup_released; |
8fccebfa FM |
9874 | extent_info.insertions = 0; |
9875 | ||
9876 | path = btrfs_alloc_path(); | |
a3ee79bd QW |
9877 | if (!path) { |
9878 | ret = -ENOMEM; | |
9879 | goto free_qgroup; | |
9880 | } | |
8fccebfa | 9881 | |
bfc78479 | 9882 | ret = btrfs_replace_file_extents(inode, path, file_offset, |
8fccebfa FM |
9883 | file_offset + len - 1, &extent_info, |
9884 | &trans); | |
9885 | btrfs_free_path(path); | |
9886 | if (ret) | |
a3ee79bd | 9887 | goto free_qgroup; |
8fccebfa | 9888 | return trans; |
a3ee79bd QW |
9889 | |
9890 | free_qgroup: | |
9891 | /* | |
9892 | * We have released qgroup data range at the beginning of the function, | |
9893 | * and normally qgroup_released bytes will be freed when committing | |
9894 | * transaction. | |
9895 | * But if we error out early, we have to free what we have released | |
9896 | * or we leak qgroup data reservation. | |
9897 | */ | |
9898 | btrfs_qgroup_free_refroot(inode->root->fs_info, | |
9899 | inode->root->root_key.objectid, qgroup_released, | |
9900 | BTRFS_QGROUP_RSV_DATA); | |
9901 | return ERR_PTR(ret); | |
203f44c5 | 9902 | } |
8fccebfa | 9903 | |
0af3d00b JB |
9904 | static int __btrfs_prealloc_file_range(struct inode *inode, int mode, |
9905 | u64 start, u64 num_bytes, u64 min_size, | |
9906 | loff_t actual_len, u64 *alloc_hint, | |
9907 | struct btrfs_trans_handle *trans) | |
d899e052 | 9908 | { |
0b246afa | 9909 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
5dc562c5 JB |
9910 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; |
9911 | struct extent_map *em; | |
d899e052 YZ |
9912 | struct btrfs_root *root = BTRFS_I(inode)->root; |
9913 | struct btrfs_key ins; | |
d899e052 | 9914 | u64 cur_offset = start; |
b778cf96 | 9915 | u64 clear_offset = start; |
55a61d1d | 9916 | u64 i_size; |
154ea289 | 9917 | u64 cur_bytes; |
0b670dc4 | 9918 | u64 last_alloc = (u64)-1; |
d899e052 | 9919 | int ret = 0; |
0af3d00b | 9920 | bool own_trans = true; |
18513091 | 9921 | u64 end = start + num_bytes - 1; |
d899e052 | 9922 | |
0af3d00b JB |
9923 | if (trans) |
9924 | own_trans = false; | |
d899e052 | 9925 | while (num_bytes > 0) { |
ee22184b | 9926 | cur_bytes = min_t(u64, num_bytes, SZ_256M); |
154ea289 | 9927 | cur_bytes = max(cur_bytes, min_size); |
0b670dc4 JB |
9928 | /* |
9929 | * If we are severely fragmented we could end up with really | |
9930 | * small allocations, so if the allocator is returning small | |
9931 | * chunks lets make its job easier by only searching for those | |
9932 | * sized chunks. | |
9933 | */ | |
9934 | cur_bytes = min(cur_bytes, last_alloc); | |
18513091 WX |
9935 | ret = btrfs_reserve_extent(root, cur_bytes, cur_bytes, |
9936 | min_size, 0, *alloc_hint, &ins, 1, 0); | |
8fccebfa | 9937 | if (ret) |
a22285a6 | 9938 | break; |
b778cf96 JB |
9939 | |
9940 | /* | |
9941 | * We've reserved this space, and thus converted it from | |
9942 | * ->bytes_may_use to ->bytes_reserved. Any error that happens | |
9943 | * from here on out we will only need to clear our reservation | |
9944 | * for the remaining unreserved area, so advance our | |
9945 | * clear_offset by our extent size. | |
9946 | */ | |
9947 | clear_offset += ins.offset; | |
5a303d5d | 9948 | |
0b670dc4 | 9949 | last_alloc = ins.offset; |
90dffd0c NB |
9950 | trans = insert_prealloc_file_extent(trans, BTRFS_I(inode), |
9951 | &ins, cur_offset); | |
1afc708d FM |
9952 | /* |
9953 | * Now that we inserted the prealloc extent we can finally | |
9954 | * decrement the number of reservations in the block group. | |
9955 | * If we did it before, we could race with relocation and have | |
9956 | * relocation miss the reserved extent, making it fail later. | |
9957 | */ | |
9958 | btrfs_dec_block_group_reservations(fs_info, ins.objectid); | |
8fccebfa FM |
9959 | if (IS_ERR(trans)) { |
9960 | ret = PTR_ERR(trans); | |
2ff7e61e | 9961 | btrfs_free_reserved_extent(fs_info, ins.objectid, |
e570fd27 | 9962 | ins.offset, 0); |
79787eaa JM |
9963 | break; |
9964 | } | |
31193213 | 9965 | |
dcdbc059 | 9966 | btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset, |
a1ed835e | 9967 | cur_offset + ins.offset -1, 0); |
5a303d5d | 9968 | |
5dc562c5 JB |
9969 | em = alloc_extent_map(); |
9970 | if (!em) { | |
23e3337f | 9971 | btrfs_set_inode_full_sync(BTRFS_I(inode)); |
5dc562c5 JB |
9972 | goto next; |
9973 | } | |
9974 | ||
9975 | em->start = cur_offset; | |
9976 | em->orig_start = cur_offset; | |
9977 | em->len = ins.offset; | |
9978 | em->block_start = ins.objectid; | |
9979 | em->block_len = ins.offset; | |
b4939680 | 9980 | em->orig_block_len = ins.offset; |
cc95bef6 | 9981 | em->ram_bytes = ins.offset; |
5dc562c5 JB |
9982 | set_bit(EXTENT_FLAG_PREALLOC, &em->flags); |
9983 | em->generation = trans->transid; | |
9984 | ||
9985 | while (1) { | |
9986 | write_lock(&em_tree->lock); | |
09a2a8f9 | 9987 | ret = add_extent_mapping(em_tree, em, 1); |
5dc562c5 JB |
9988 | write_unlock(&em_tree->lock); |
9989 | if (ret != -EEXIST) | |
9990 | break; | |
dcdbc059 | 9991 | btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset, |
5dc562c5 JB |
9992 | cur_offset + ins.offset - 1, |
9993 | 0); | |
9994 | } | |
9995 | free_extent_map(em); | |
9996 | next: | |
d899e052 YZ |
9997 | num_bytes -= ins.offset; |
9998 | cur_offset += ins.offset; | |
efa56464 | 9999 | *alloc_hint = ins.objectid + ins.offset; |
5a303d5d | 10000 | |
0c4d2d95 | 10001 | inode_inc_iversion(inode); |
c2050a45 | 10002 | inode->i_ctime = current_time(inode); |
6cbff00f | 10003 | BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC; |
d899e052 | 10004 | if (!(mode & FALLOC_FL_KEEP_SIZE) && |
efa56464 YZ |
10005 | (actual_len > inode->i_size) && |
10006 | (cur_offset > inode->i_size)) { | |
d1ea6a61 | 10007 | if (cur_offset > actual_len) |
55a61d1d | 10008 | i_size = actual_len; |
d1ea6a61 | 10009 | else |
55a61d1d JB |
10010 | i_size = cur_offset; |
10011 | i_size_write(inode, i_size); | |
76aea537 | 10012 | btrfs_inode_safe_disk_i_size_write(BTRFS_I(inode), 0); |
5a303d5d YZ |
10013 | } |
10014 | ||
9a56fcd1 | 10015 | ret = btrfs_update_inode(trans, root, BTRFS_I(inode)); |
79787eaa JM |
10016 | |
10017 | if (ret) { | |
66642832 | 10018 | btrfs_abort_transaction(trans, ret); |
79787eaa | 10019 | if (own_trans) |
3a45bb20 | 10020 | btrfs_end_transaction(trans); |
79787eaa JM |
10021 | break; |
10022 | } | |
d899e052 | 10023 | |
8fccebfa | 10024 | if (own_trans) { |
3a45bb20 | 10025 | btrfs_end_transaction(trans); |
8fccebfa FM |
10026 | trans = NULL; |
10027 | } | |
5a303d5d | 10028 | } |
b778cf96 | 10029 | if (clear_offset < end) |
25ce28ca | 10030 | btrfs_free_reserved_data_space(BTRFS_I(inode), NULL, clear_offset, |
b778cf96 | 10031 | end - clear_offset + 1); |
d899e052 YZ |
10032 | return ret; |
10033 | } | |
10034 | ||
0af3d00b JB |
10035 | int btrfs_prealloc_file_range(struct inode *inode, int mode, |
10036 | u64 start, u64 num_bytes, u64 min_size, | |
10037 | loff_t actual_len, u64 *alloc_hint) | |
10038 | { | |
10039 | return __btrfs_prealloc_file_range(inode, mode, start, num_bytes, | |
10040 | min_size, actual_len, alloc_hint, | |
10041 | NULL); | |
10042 | } | |
10043 | ||
10044 | int btrfs_prealloc_file_range_trans(struct inode *inode, | |
10045 | struct btrfs_trans_handle *trans, int mode, | |
10046 | u64 start, u64 num_bytes, u64 min_size, | |
10047 | loff_t actual_len, u64 *alloc_hint) | |
10048 | { | |
10049 | return __btrfs_prealloc_file_range(inode, mode, start, num_bytes, | |
10050 | min_size, actual_len, alloc_hint, trans); | |
10051 | } | |
10052 | ||
549c7297 CB |
10053 | static int btrfs_permission(struct user_namespace *mnt_userns, |
10054 | struct inode *inode, int mask) | |
fdebe2bd | 10055 | { |
b83cc969 | 10056 | struct btrfs_root *root = BTRFS_I(inode)->root; |
cb6db4e5 | 10057 | umode_t mode = inode->i_mode; |
b83cc969 | 10058 | |
cb6db4e5 JM |
10059 | if (mask & MAY_WRITE && |
10060 | (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) { | |
10061 | if (btrfs_root_readonly(root)) | |
10062 | return -EROFS; | |
10063 | if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY) | |
10064 | return -EACCES; | |
10065 | } | |
3bc71ba0 | 10066 | return generic_permission(mnt_userns, inode, mask); |
fdebe2bd | 10067 | } |
39279cc3 | 10068 | |
549c7297 CB |
10069 | static int btrfs_tmpfile(struct user_namespace *mnt_userns, struct inode *dir, |
10070 | struct dentry *dentry, umode_t mode) | |
ef3b9af5 | 10071 | { |
2ff7e61e | 10072 | struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb); |
ef3b9af5 FM |
10073 | struct btrfs_trans_handle *trans; |
10074 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
a1fd0c35 | 10075 | struct inode *inode; |
3538d68d OS |
10076 | struct btrfs_new_inode_args new_inode_args = { |
10077 | .dir = dir, | |
10078 | .dentry = dentry, | |
10079 | .orphan = true, | |
10080 | }; | |
10081 | unsigned int trans_num_items; | |
a1fd0c35 OS |
10082 | int ret; |
10083 | ||
10084 | inode = new_inode(dir->i_sb); | |
10085 | if (!inode) | |
10086 | return -ENOMEM; | |
10087 | inode_init_owner(mnt_userns, inode, dir, mode); | |
10088 | inode->i_fop = &btrfs_file_operations; | |
10089 | inode->i_op = &btrfs_file_inode_operations; | |
10090 | inode->i_mapping->a_ops = &btrfs_aops; | |
ef3b9af5 | 10091 | |
3538d68d OS |
10092 | new_inode_args.inode = inode; |
10093 | ret = btrfs_new_inode_prepare(&new_inode_args, &trans_num_items); | |
caae78e0 OS |
10094 | if (ret) |
10095 | goto out_inode; | |
3538d68d OS |
10096 | |
10097 | trans = btrfs_start_transaction(root, trans_num_items); | |
a1fd0c35 | 10098 | if (IS_ERR(trans)) { |
3538d68d OS |
10099 | ret = PTR_ERR(trans); |
10100 | goto out_new_inode_args; | |
a1fd0c35 | 10101 | } |
ef3b9af5 | 10102 | |
caae78e0 | 10103 | ret = btrfs_create_new_inode(trans, &new_inode_args); |
ef3b9af5 | 10104 | |
5762b5c9 | 10105 | /* |
3538d68d OS |
10106 | * We set number of links to 0 in btrfs_create_new_inode(), and here we |
10107 | * set it to 1 because d_tmpfile() will issue a warning if the count is | |
10108 | * 0, through: | |
5762b5c9 FM |
10109 | * |
10110 | * d_tmpfile() -> inode_dec_link_count() -> drop_nlink() | |
10111 | */ | |
10112 | set_nlink(inode, 1); | |
caae78e0 OS |
10113 | |
10114 | if (!ret) { | |
10115 | d_tmpfile(dentry, inode); | |
10116 | unlock_new_inode(inode); | |
10117 | mark_inode_dirty(inode); | |
10118 | } | |
10119 | ||
3a45bb20 | 10120 | btrfs_end_transaction(trans); |
2ff7e61e | 10121 | btrfs_btree_balance_dirty(fs_info); |
3538d68d OS |
10122 | out_new_inode_args: |
10123 | btrfs_new_inode_args_destroy(&new_inode_args); | |
caae78e0 OS |
10124 | out_inode: |
10125 | if (ret) | |
10126 | iput(inode); | |
ef3b9af5 FM |
10127 | return ret; |
10128 | } | |
10129 | ||
d2a91064 | 10130 | void btrfs_set_range_writeback(struct btrfs_inode *inode, u64 start, u64 end) |
c6100a4b | 10131 | { |
d2a91064 | 10132 | struct btrfs_fs_info *fs_info = inode->root->fs_info; |
c6100a4b JB |
10133 | unsigned long index = start >> PAGE_SHIFT; |
10134 | unsigned long end_index = end >> PAGE_SHIFT; | |
10135 | struct page *page; | |
d2a91064 | 10136 | u32 len; |
c6100a4b | 10137 | |
d2a91064 QW |
10138 | ASSERT(end + 1 - start <= U32_MAX); |
10139 | len = end + 1 - start; | |
c6100a4b | 10140 | while (index <= end_index) { |
d2a91064 | 10141 | page = find_get_page(inode->vfs_inode.i_mapping, index); |
c6100a4b | 10142 | ASSERT(page); /* Pages should be in the extent_io_tree */ |
d2a91064 QW |
10143 | |
10144 | btrfs_page_set_writeback(fs_info, page, start, len); | |
c6100a4b JB |
10145 | put_page(page); |
10146 | index++; | |
10147 | } | |
10148 | } | |
10149 | ||
3ea4dc5b OS |
10150 | int btrfs_encoded_io_compression_from_extent(struct btrfs_fs_info *fs_info, |
10151 | int compress_type) | |
1881fba8 OS |
10152 | { |
10153 | switch (compress_type) { | |
10154 | case BTRFS_COMPRESS_NONE: | |
10155 | return BTRFS_ENCODED_IO_COMPRESSION_NONE; | |
10156 | case BTRFS_COMPRESS_ZLIB: | |
10157 | return BTRFS_ENCODED_IO_COMPRESSION_ZLIB; | |
10158 | case BTRFS_COMPRESS_LZO: | |
10159 | /* | |
10160 | * The LZO format depends on the sector size. 64K is the maximum | |
10161 | * sector size that we support. | |
10162 | */ | |
10163 | if (fs_info->sectorsize < SZ_4K || fs_info->sectorsize > SZ_64K) | |
10164 | return -EINVAL; | |
10165 | return BTRFS_ENCODED_IO_COMPRESSION_LZO_4K + | |
10166 | (fs_info->sectorsize_bits - 12); | |
10167 | case BTRFS_COMPRESS_ZSTD: | |
10168 | return BTRFS_ENCODED_IO_COMPRESSION_ZSTD; | |
10169 | default: | |
10170 | return -EUCLEAN; | |
10171 | } | |
10172 | } | |
10173 | ||
10174 | static ssize_t btrfs_encoded_read_inline( | |
10175 | struct kiocb *iocb, | |
10176 | struct iov_iter *iter, u64 start, | |
10177 | u64 lockend, | |
10178 | struct extent_state **cached_state, | |
10179 | u64 extent_start, size_t count, | |
10180 | struct btrfs_ioctl_encoded_io_args *encoded, | |
10181 | bool *unlocked) | |
10182 | { | |
10183 | struct btrfs_inode *inode = BTRFS_I(file_inode(iocb->ki_filp)); | |
10184 | struct btrfs_root *root = inode->root; | |
10185 | struct btrfs_fs_info *fs_info = root->fs_info; | |
10186 | struct extent_io_tree *io_tree = &inode->io_tree; | |
10187 | struct btrfs_path *path; | |
10188 | struct extent_buffer *leaf; | |
10189 | struct btrfs_file_extent_item *item; | |
10190 | u64 ram_bytes; | |
10191 | unsigned long ptr; | |
10192 | void *tmp; | |
10193 | ssize_t ret; | |
10194 | ||
10195 | path = btrfs_alloc_path(); | |
10196 | if (!path) { | |
10197 | ret = -ENOMEM; | |
10198 | goto out; | |
10199 | } | |
10200 | ret = btrfs_lookup_file_extent(NULL, root, path, btrfs_ino(inode), | |
10201 | extent_start, 0); | |
10202 | if (ret) { | |
10203 | if (ret > 0) { | |
10204 | /* The extent item disappeared? */ | |
10205 | ret = -EIO; | |
10206 | } | |
10207 | goto out; | |
10208 | } | |
10209 | leaf = path->nodes[0]; | |
10210 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_file_extent_item); | |
10211 | ||
10212 | ram_bytes = btrfs_file_extent_ram_bytes(leaf, item); | |
10213 | ptr = btrfs_file_extent_inline_start(item); | |
10214 | ||
10215 | encoded->len = min_t(u64, extent_start + ram_bytes, | |
10216 | inode->vfs_inode.i_size) - iocb->ki_pos; | |
10217 | ret = btrfs_encoded_io_compression_from_extent(fs_info, | |
10218 | btrfs_file_extent_compression(leaf, item)); | |
10219 | if (ret < 0) | |
10220 | goto out; | |
10221 | encoded->compression = ret; | |
10222 | if (encoded->compression) { | |
10223 | size_t inline_size; | |
10224 | ||
10225 | inline_size = btrfs_file_extent_inline_item_len(leaf, | |
10226 | path->slots[0]); | |
10227 | if (inline_size > count) { | |
10228 | ret = -ENOBUFS; | |
10229 | goto out; | |
10230 | } | |
10231 | count = inline_size; | |
10232 | encoded->unencoded_len = ram_bytes; | |
10233 | encoded->unencoded_offset = iocb->ki_pos - extent_start; | |
10234 | } else { | |
10235 | count = min_t(u64, count, encoded->len); | |
10236 | encoded->len = count; | |
10237 | encoded->unencoded_len = count; | |
10238 | ptr += iocb->ki_pos - extent_start; | |
10239 | } | |
10240 | ||
10241 | tmp = kmalloc(count, GFP_NOFS); | |
10242 | if (!tmp) { | |
10243 | ret = -ENOMEM; | |
10244 | goto out; | |
10245 | } | |
10246 | read_extent_buffer(leaf, tmp, ptr, count); | |
10247 | btrfs_release_path(path); | |
570eb97b | 10248 | unlock_extent(io_tree, start, lockend, cached_state); |
1881fba8 OS |
10249 | btrfs_inode_unlock(&inode->vfs_inode, BTRFS_ILOCK_SHARED); |
10250 | *unlocked = true; | |
10251 | ||
10252 | ret = copy_to_iter(tmp, count, iter); | |
10253 | if (ret != count) | |
10254 | ret = -EFAULT; | |
10255 | kfree(tmp); | |
10256 | out: | |
10257 | btrfs_free_path(path); | |
10258 | return ret; | |
10259 | } | |
10260 | ||
10261 | struct btrfs_encoded_read_private { | |
10262 | struct btrfs_inode *inode; | |
10263 | u64 file_offset; | |
10264 | wait_queue_head_t wait; | |
10265 | atomic_t pending; | |
10266 | blk_status_t status; | |
10267 | bool skip_csum; | |
10268 | }; | |
10269 | ||
10270 | static blk_status_t submit_encoded_read_bio(struct btrfs_inode *inode, | |
10271 | struct bio *bio, int mirror_num) | |
10272 | { | |
917f32a2 | 10273 | struct btrfs_encoded_read_private *priv = btrfs_bio(bio)->private; |
1881fba8 OS |
10274 | struct btrfs_fs_info *fs_info = inode->root->fs_info; |
10275 | blk_status_t ret; | |
10276 | ||
10277 | if (!priv->skip_csum) { | |
10278 | ret = btrfs_lookup_bio_sums(&inode->vfs_inode, bio, NULL); | |
10279 | if (ret) | |
10280 | return ret; | |
10281 | } | |
10282 | ||
1881fba8 | 10283 | atomic_inc(&priv->pending); |
1a722d8f CH |
10284 | btrfs_submit_bio(fs_info, bio, mirror_num); |
10285 | return BLK_STS_OK; | |
1881fba8 OS |
10286 | } |
10287 | ||
10288 | static blk_status_t btrfs_encoded_read_verify_csum(struct btrfs_bio *bbio) | |
10289 | { | |
10290 | const bool uptodate = (bbio->bio.bi_status == BLK_STS_OK); | |
917f32a2 | 10291 | struct btrfs_encoded_read_private *priv = bbio->private; |
1881fba8 OS |
10292 | struct btrfs_inode *inode = priv->inode; |
10293 | struct btrfs_fs_info *fs_info = inode->root->fs_info; | |
10294 | u32 sectorsize = fs_info->sectorsize; | |
10295 | struct bio_vec *bvec; | |
10296 | struct bvec_iter_all iter_all; | |
1881fba8 OS |
10297 | u32 bio_offset = 0; |
10298 | ||
10299 | if (priv->skip_csum || !uptodate) | |
10300 | return bbio->bio.bi_status; | |
10301 | ||
10302 | bio_for_each_segment_all(bvec, &bbio->bio, iter_all) { | |
10303 | unsigned int i, nr_sectors, pgoff; | |
10304 | ||
10305 | nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec->bv_len); | |
10306 | pgoff = bvec->bv_offset; | |
10307 | for (i = 0; i < nr_sectors; i++) { | |
10308 | ASSERT(pgoff < PAGE_SIZE); | |
7959bd44 CH |
10309 | if (btrfs_check_data_csum(&inode->vfs_inode, bbio, bio_offset, |
10310 | bvec->bv_page, pgoff)) | |
1881fba8 | 10311 | return BLK_STS_IOERR; |
1881fba8 OS |
10312 | bio_offset += sectorsize; |
10313 | pgoff += sectorsize; | |
10314 | } | |
10315 | } | |
10316 | return BLK_STS_OK; | |
10317 | } | |
10318 | ||
917f32a2 | 10319 | static void btrfs_encoded_read_endio(struct btrfs_bio *bbio) |
1881fba8 | 10320 | { |
917f32a2 | 10321 | struct btrfs_encoded_read_private *priv = bbio->private; |
1881fba8 OS |
10322 | blk_status_t status; |
10323 | ||
10324 | status = btrfs_encoded_read_verify_csum(bbio); | |
10325 | if (status) { | |
10326 | /* | |
10327 | * The memory barrier implied by the atomic_dec_return() here | |
10328 | * pairs with the memory barrier implied by the | |
10329 | * atomic_dec_return() or io_wait_event() in | |
10330 | * btrfs_encoded_read_regular_fill_pages() to ensure that this | |
10331 | * write is observed before the load of status in | |
10332 | * btrfs_encoded_read_regular_fill_pages(). | |
10333 | */ | |
10334 | WRITE_ONCE(priv->status, status); | |
10335 | } | |
10336 | if (!atomic_dec_return(&priv->pending)) | |
10337 | wake_up(&priv->wait); | |
10338 | btrfs_bio_free_csum(bbio); | |
917f32a2 | 10339 | bio_put(&bbio->bio); |
1881fba8 OS |
10340 | } |
10341 | ||
3ea4dc5b OS |
10342 | int btrfs_encoded_read_regular_fill_pages(struct btrfs_inode *inode, |
10343 | u64 file_offset, u64 disk_bytenr, | |
10344 | u64 disk_io_size, struct page **pages) | |
1881fba8 OS |
10345 | { |
10346 | struct btrfs_fs_info *fs_info = inode->root->fs_info; | |
10347 | struct btrfs_encoded_read_private priv = { | |
10348 | .inode = inode, | |
10349 | .file_offset = file_offset, | |
10350 | .pending = ATOMIC_INIT(1), | |
10351 | .skip_csum = (inode->flags & BTRFS_INODE_NODATASUM), | |
10352 | }; | |
10353 | unsigned long i = 0; | |
10354 | u64 cur = 0; | |
10355 | int ret; | |
10356 | ||
10357 | init_waitqueue_head(&priv.wait); | |
10358 | /* | |
10359 | * Submit bios for the extent, splitting due to bio or stripe limits as | |
10360 | * necessary. | |
10361 | */ | |
10362 | while (cur < disk_io_size) { | |
10363 | struct extent_map *em; | |
10364 | struct btrfs_io_geometry geom; | |
10365 | struct bio *bio = NULL; | |
10366 | u64 remaining; | |
10367 | ||
10368 | em = btrfs_get_chunk_map(fs_info, disk_bytenr + cur, | |
10369 | disk_io_size - cur); | |
10370 | if (IS_ERR(em)) { | |
10371 | ret = PTR_ERR(em); | |
10372 | } else { | |
10373 | ret = btrfs_get_io_geometry(fs_info, em, BTRFS_MAP_READ, | |
10374 | disk_bytenr + cur, &geom); | |
10375 | free_extent_map(em); | |
10376 | } | |
10377 | if (ret) { | |
10378 | WRITE_ONCE(priv.status, errno_to_blk_status(ret)); | |
10379 | break; | |
10380 | } | |
10381 | remaining = min(geom.len, disk_io_size - cur); | |
10382 | while (bio || remaining) { | |
10383 | size_t bytes = min_t(u64, remaining, PAGE_SIZE); | |
10384 | ||
10385 | if (!bio) { | |
917f32a2 CH |
10386 | bio = btrfs_bio_alloc(BIO_MAX_VECS, REQ_OP_READ, |
10387 | btrfs_encoded_read_endio, | |
10388 | &priv); | |
1881fba8 OS |
10389 | bio->bi_iter.bi_sector = |
10390 | (disk_bytenr + cur) >> SECTOR_SHIFT; | |
1881fba8 OS |
10391 | } |
10392 | ||
10393 | if (!bytes || | |
10394 | bio_add_page(bio, pages[i], bytes, 0) < bytes) { | |
10395 | blk_status_t status; | |
10396 | ||
10397 | status = submit_encoded_read_bio(inode, bio, 0); | |
10398 | if (status) { | |
10399 | WRITE_ONCE(priv.status, status); | |
10400 | bio_put(bio); | |
10401 | goto out; | |
10402 | } | |
10403 | bio = NULL; | |
10404 | continue; | |
10405 | } | |
10406 | ||
10407 | i++; | |
10408 | cur += bytes; | |
10409 | remaining -= bytes; | |
10410 | } | |
10411 | } | |
10412 | ||
10413 | out: | |
10414 | if (atomic_dec_return(&priv.pending)) | |
10415 | io_wait_event(priv.wait, !atomic_read(&priv.pending)); | |
10416 | /* See btrfs_encoded_read_endio() for ordering. */ | |
10417 | return blk_status_to_errno(READ_ONCE(priv.status)); | |
10418 | } | |
10419 | ||
10420 | static ssize_t btrfs_encoded_read_regular(struct kiocb *iocb, | |
10421 | struct iov_iter *iter, | |
10422 | u64 start, u64 lockend, | |
10423 | struct extent_state **cached_state, | |
10424 | u64 disk_bytenr, u64 disk_io_size, | |
10425 | size_t count, bool compressed, | |
10426 | bool *unlocked) | |
10427 | { | |
10428 | struct btrfs_inode *inode = BTRFS_I(file_inode(iocb->ki_filp)); | |
10429 | struct extent_io_tree *io_tree = &inode->io_tree; | |
10430 | struct page **pages; | |
10431 | unsigned long nr_pages, i; | |
10432 | u64 cur; | |
10433 | size_t page_offset; | |
10434 | ssize_t ret; | |
10435 | ||
10436 | nr_pages = DIV_ROUND_UP(disk_io_size, PAGE_SIZE); | |
10437 | pages = kcalloc(nr_pages, sizeof(struct page *), GFP_NOFS); | |
10438 | if (!pages) | |
10439 | return -ENOMEM; | |
dd137dd1 STD |
10440 | ret = btrfs_alloc_page_array(nr_pages, pages); |
10441 | if (ret) { | |
10442 | ret = -ENOMEM; | |
10443 | goto out; | |
1881fba8 | 10444 | } |
1881fba8 OS |
10445 | |
10446 | ret = btrfs_encoded_read_regular_fill_pages(inode, start, disk_bytenr, | |
10447 | disk_io_size, pages); | |
10448 | if (ret) | |
10449 | goto out; | |
10450 | ||
570eb97b | 10451 | unlock_extent(io_tree, start, lockend, cached_state); |
1881fba8 OS |
10452 | btrfs_inode_unlock(&inode->vfs_inode, BTRFS_ILOCK_SHARED); |
10453 | *unlocked = true; | |
10454 | ||
10455 | if (compressed) { | |
10456 | i = 0; | |
10457 | page_offset = 0; | |
10458 | } else { | |
10459 | i = (iocb->ki_pos - start) >> PAGE_SHIFT; | |
10460 | page_offset = (iocb->ki_pos - start) & (PAGE_SIZE - 1); | |
10461 | } | |
10462 | cur = 0; | |
10463 | while (cur < count) { | |
10464 | size_t bytes = min_t(size_t, count - cur, | |
10465 | PAGE_SIZE - page_offset); | |
10466 | ||
10467 | if (copy_page_to_iter(pages[i], page_offset, bytes, | |
10468 | iter) != bytes) { | |
10469 | ret = -EFAULT; | |
10470 | goto out; | |
10471 | } | |
10472 | i++; | |
10473 | cur += bytes; | |
10474 | page_offset = 0; | |
10475 | } | |
10476 | ret = count; | |
10477 | out: | |
10478 | for (i = 0; i < nr_pages; i++) { | |
10479 | if (pages[i]) | |
10480 | __free_page(pages[i]); | |
10481 | } | |
10482 | kfree(pages); | |
10483 | return ret; | |
10484 | } | |
10485 | ||
10486 | ssize_t btrfs_encoded_read(struct kiocb *iocb, struct iov_iter *iter, | |
10487 | struct btrfs_ioctl_encoded_io_args *encoded) | |
10488 | { | |
10489 | struct btrfs_inode *inode = BTRFS_I(file_inode(iocb->ki_filp)); | |
10490 | struct btrfs_fs_info *fs_info = inode->root->fs_info; | |
10491 | struct extent_io_tree *io_tree = &inode->io_tree; | |
10492 | ssize_t ret; | |
10493 | size_t count = iov_iter_count(iter); | |
10494 | u64 start, lockend, disk_bytenr, disk_io_size; | |
10495 | struct extent_state *cached_state = NULL; | |
10496 | struct extent_map *em; | |
10497 | bool unlocked = false; | |
10498 | ||
10499 | file_accessed(iocb->ki_filp); | |
10500 | ||
10501 | btrfs_inode_lock(&inode->vfs_inode, BTRFS_ILOCK_SHARED); | |
10502 | ||
10503 | if (iocb->ki_pos >= inode->vfs_inode.i_size) { | |
10504 | btrfs_inode_unlock(&inode->vfs_inode, BTRFS_ILOCK_SHARED); | |
10505 | return 0; | |
10506 | } | |
10507 | start = ALIGN_DOWN(iocb->ki_pos, fs_info->sectorsize); | |
10508 | /* | |
10509 | * We don't know how long the extent containing iocb->ki_pos is, but if | |
10510 | * it's compressed we know that it won't be longer than this. | |
10511 | */ | |
10512 | lockend = start + BTRFS_MAX_UNCOMPRESSED - 1; | |
10513 | ||
10514 | for (;;) { | |
10515 | struct btrfs_ordered_extent *ordered; | |
10516 | ||
10517 | ret = btrfs_wait_ordered_range(&inode->vfs_inode, start, | |
10518 | lockend - start + 1); | |
10519 | if (ret) | |
10520 | goto out_unlock_inode; | |
570eb97b | 10521 | lock_extent(io_tree, start, lockend, &cached_state); |
1881fba8 OS |
10522 | ordered = btrfs_lookup_ordered_range(inode, start, |
10523 | lockend - start + 1); | |
10524 | if (!ordered) | |
10525 | break; | |
10526 | btrfs_put_ordered_extent(ordered); | |
570eb97b | 10527 | unlock_extent(io_tree, start, lockend, &cached_state); |
1881fba8 OS |
10528 | cond_resched(); |
10529 | } | |
10530 | ||
10531 | em = btrfs_get_extent(inode, NULL, 0, start, lockend - start + 1); | |
10532 | if (IS_ERR(em)) { | |
10533 | ret = PTR_ERR(em); | |
10534 | goto out_unlock_extent; | |
10535 | } | |
10536 | ||
10537 | if (em->block_start == EXTENT_MAP_INLINE) { | |
10538 | u64 extent_start = em->start; | |
10539 | ||
10540 | /* | |
10541 | * For inline extents we get everything we need out of the | |
10542 | * extent item. | |
10543 | */ | |
10544 | free_extent_map(em); | |
10545 | em = NULL; | |
10546 | ret = btrfs_encoded_read_inline(iocb, iter, start, lockend, | |
10547 | &cached_state, extent_start, | |
10548 | count, encoded, &unlocked); | |
10549 | goto out; | |
10550 | } | |
10551 | ||
10552 | /* | |
10553 | * We only want to return up to EOF even if the extent extends beyond | |
10554 | * that. | |
10555 | */ | |
10556 | encoded->len = min_t(u64, extent_map_end(em), | |
10557 | inode->vfs_inode.i_size) - iocb->ki_pos; | |
10558 | if (em->block_start == EXTENT_MAP_HOLE || | |
10559 | test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) { | |
10560 | disk_bytenr = EXTENT_MAP_HOLE; | |
10561 | count = min_t(u64, count, encoded->len); | |
10562 | encoded->len = count; | |
10563 | encoded->unencoded_len = count; | |
10564 | } else if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) { | |
10565 | disk_bytenr = em->block_start; | |
10566 | /* | |
10567 | * Bail if the buffer isn't large enough to return the whole | |
10568 | * compressed extent. | |
10569 | */ | |
10570 | if (em->block_len > count) { | |
10571 | ret = -ENOBUFS; | |
10572 | goto out_em; | |
10573 | } | |
c1867eb3 DS |
10574 | disk_io_size = em->block_len; |
10575 | count = em->block_len; | |
1881fba8 OS |
10576 | encoded->unencoded_len = em->ram_bytes; |
10577 | encoded->unencoded_offset = iocb->ki_pos - em->orig_start; | |
10578 | ret = btrfs_encoded_io_compression_from_extent(fs_info, | |
10579 | em->compress_type); | |
10580 | if (ret < 0) | |
10581 | goto out_em; | |
10582 | encoded->compression = ret; | |
10583 | } else { | |
10584 | disk_bytenr = em->block_start + (start - em->start); | |
10585 | if (encoded->len > count) | |
10586 | encoded->len = count; | |
10587 | /* | |
10588 | * Don't read beyond what we locked. This also limits the page | |
10589 | * allocations that we'll do. | |
10590 | */ | |
10591 | disk_io_size = min(lockend + 1, iocb->ki_pos + encoded->len) - start; | |
10592 | count = start + disk_io_size - iocb->ki_pos; | |
10593 | encoded->len = count; | |
10594 | encoded->unencoded_len = count; | |
10595 | disk_io_size = ALIGN(disk_io_size, fs_info->sectorsize); | |
10596 | } | |
10597 | free_extent_map(em); | |
10598 | em = NULL; | |
10599 | ||
10600 | if (disk_bytenr == EXTENT_MAP_HOLE) { | |
570eb97b | 10601 | unlock_extent(io_tree, start, lockend, &cached_state); |
1881fba8 OS |
10602 | btrfs_inode_unlock(&inode->vfs_inode, BTRFS_ILOCK_SHARED); |
10603 | unlocked = true; | |
10604 | ret = iov_iter_zero(count, iter); | |
10605 | if (ret != count) | |
10606 | ret = -EFAULT; | |
10607 | } else { | |
10608 | ret = btrfs_encoded_read_regular(iocb, iter, start, lockend, | |
10609 | &cached_state, disk_bytenr, | |
10610 | disk_io_size, count, | |
10611 | encoded->compression, | |
10612 | &unlocked); | |
10613 | } | |
10614 | ||
10615 | out: | |
10616 | if (ret >= 0) | |
10617 | iocb->ki_pos += encoded->len; | |
10618 | out_em: | |
10619 | free_extent_map(em); | |
10620 | out_unlock_extent: | |
10621 | if (!unlocked) | |
570eb97b | 10622 | unlock_extent(io_tree, start, lockend, &cached_state); |
1881fba8 OS |
10623 | out_unlock_inode: |
10624 | if (!unlocked) | |
10625 | btrfs_inode_unlock(&inode->vfs_inode, BTRFS_ILOCK_SHARED); | |
10626 | return ret; | |
10627 | } | |
10628 | ||
7c0c7269 OS |
10629 | ssize_t btrfs_do_encoded_write(struct kiocb *iocb, struct iov_iter *from, |
10630 | const struct btrfs_ioctl_encoded_io_args *encoded) | |
10631 | { | |
10632 | struct btrfs_inode *inode = BTRFS_I(file_inode(iocb->ki_filp)); | |
10633 | struct btrfs_root *root = inode->root; | |
10634 | struct btrfs_fs_info *fs_info = root->fs_info; | |
10635 | struct extent_io_tree *io_tree = &inode->io_tree; | |
10636 | struct extent_changeset *data_reserved = NULL; | |
10637 | struct extent_state *cached_state = NULL; | |
10638 | int compression; | |
10639 | size_t orig_count; | |
10640 | u64 start, end; | |
10641 | u64 num_bytes, ram_bytes, disk_num_bytes; | |
10642 | unsigned long nr_pages, i; | |
10643 | struct page **pages; | |
10644 | struct btrfs_key ins; | |
10645 | bool extent_reserved = false; | |
10646 | struct extent_map *em; | |
10647 | ssize_t ret; | |
10648 | ||
10649 | switch (encoded->compression) { | |
10650 | case BTRFS_ENCODED_IO_COMPRESSION_ZLIB: | |
10651 | compression = BTRFS_COMPRESS_ZLIB; | |
10652 | break; | |
10653 | case BTRFS_ENCODED_IO_COMPRESSION_ZSTD: | |
10654 | compression = BTRFS_COMPRESS_ZSTD; | |
10655 | break; | |
10656 | case BTRFS_ENCODED_IO_COMPRESSION_LZO_4K: | |
10657 | case BTRFS_ENCODED_IO_COMPRESSION_LZO_8K: | |
10658 | case BTRFS_ENCODED_IO_COMPRESSION_LZO_16K: | |
10659 | case BTRFS_ENCODED_IO_COMPRESSION_LZO_32K: | |
10660 | case BTRFS_ENCODED_IO_COMPRESSION_LZO_64K: | |
10661 | /* The sector size must match for LZO. */ | |
10662 | if (encoded->compression - | |
10663 | BTRFS_ENCODED_IO_COMPRESSION_LZO_4K + 12 != | |
10664 | fs_info->sectorsize_bits) | |
10665 | return -EINVAL; | |
10666 | compression = BTRFS_COMPRESS_LZO; | |
10667 | break; | |
10668 | default: | |
10669 | return -EINVAL; | |
10670 | } | |
10671 | if (encoded->encryption != BTRFS_ENCODED_IO_ENCRYPTION_NONE) | |
10672 | return -EINVAL; | |
10673 | ||
10674 | orig_count = iov_iter_count(from); | |
10675 | ||
10676 | /* The extent size must be sane. */ | |
10677 | if (encoded->unencoded_len > BTRFS_MAX_UNCOMPRESSED || | |
10678 | orig_count > BTRFS_MAX_COMPRESSED || orig_count == 0) | |
10679 | return -EINVAL; | |
10680 | ||
10681 | /* | |
10682 | * The compressed data must be smaller than the decompressed data. | |
10683 | * | |
10684 | * It's of course possible for data to compress to larger or the same | |
10685 | * size, but the buffered I/O path falls back to no compression for such | |
10686 | * data, and we don't want to break any assumptions by creating these | |
10687 | * extents. | |
10688 | * | |
10689 | * Note that this is less strict than the current check we have that the | |
10690 | * compressed data must be at least one sector smaller than the | |
10691 | * decompressed data. We only want to enforce the weaker requirement | |
10692 | * from old kernels that it is at least one byte smaller. | |
10693 | */ | |
10694 | if (orig_count >= encoded->unencoded_len) | |
10695 | return -EINVAL; | |
10696 | ||
10697 | /* The extent must start on a sector boundary. */ | |
10698 | start = iocb->ki_pos; | |
10699 | if (!IS_ALIGNED(start, fs_info->sectorsize)) | |
10700 | return -EINVAL; | |
10701 | ||
10702 | /* | |
10703 | * The extent must end on a sector boundary. However, we allow a write | |
10704 | * which ends at or extends i_size to have an unaligned length; we round | |
10705 | * up the extent size and set i_size to the unaligned end. | |
10706 | */ | |
10707 | if (start + encoded->len < inode->vfs_inode.i_size && | |
10708 | !IS_ALIGNED(start + encoded->len, fs_info->sectorsize)) | |
10709 | return -EINVAL; | |
10710 | ||
10711 | /* Finally, the offset in the unencoded data must be sector-aligned. */ | |
10712 | if (!IS_ALIGNED(encoded->unencoded_offset, fs_info->sectorsize)) | |
10713 | return -EINVAL; | |
10714 | ||
10715 | num_bytes = ALIGN(encoded->len, fs_info->sectorsize); | |
10716 | ram_bytes = ALIGN(encoded->unencoded_len, fs_info->sectorsize); | |
10717 | end = start + num_bytes - 1; | |
10718 | ||
10719 | /* | |
10720 | * If the extent cannot be inline, the compressed data on disk must be | |
10721 | * sector-aligned. For convenience, we extend it with zeroes if it | |
10722 | * isn't. | |
10723 | */ | |
10724 | disk_num_bytes = ALIGN(orig_count, fs_info->sectorsize); | |
10725 | nr_pages = DIV_ROUND_UP(disk_num_bytes, PAGE_SIZE); | |
10726 | pages = kvcalloc(nr_pages, sizeof(struct page *), GFP_KERNEL_ACCOUNT); | |
10727 | if (!pages) | |
10728 | return -ENOMEM; | |
10729 | for (i = 0; i < nr_pages; i++) { | |
10730 | size_t bytes = min_t(size_t, PAGE_SIZE, iov_iter_count(from)); | |
10731 | char *kaddr; | |
10732 | ||
10733 | pages[i] = alloc_page(GFP_KERNEL_ACCOUNT); | |
10734 | if (!pages[i]) { | |
10735 | ret = -ENOMEM; | |
10736 | goto out_pages; | |
10737 | } | |
70826b6b | 10738 | kaddr = kmap_local_page(pages[i]); |
7c0c7269 | 10739 | if (copy_from_iter(kaddr, bytes, from) != bytes) { |
70826b6b | 10740 | kunmap_local(kaddr); |
7c0c7269 OS |
10741 | ret = -EFAULT; |
10742 | goto out_pages; | |
10743 | } | |
10744 | if (bytes < PAGE_SIZE) | |
10745 | memset(kaddr + bytes, 0, PAGE_SIZE - bytes); | |
70826b6b | 10746 | kunmap_local(kaddr); |
7c0c7269 OS |
10747 | } |
10748 | ||
10749 | for (;;) { | |
10750 | struct btrfs_ordered_extent *ordered; | |
10751 | ||
10752 | ret = btrfs_wait_ordered_range(&inode->vfs_inode, start, num_bytes); | |
10753 | if (ret) | |
10754 | goto out_pages; | |
10755 | ret = invalidate_inode_pages2_range(inode->vfs_inode.i_mapping, | |
10756 | start >> PAGE_SHIFT, | |
10757 | end >> PAGE_SHIFT); | |
10758 | if (ret) | |
10759 | goto out_pages; | |
570eb97b | 10760 | lock_extent(io_tree, start, end, &cached_state); |
7c0c7269 OS |
10761 | ordered = btrfs_lookup_ordered_range(inode, start, num_bytes); |
10762 | if (!ordered && | |
10763 | !filemap_range_has_page(inode->vfs_inode.i_mapping, start, end)) | |
10764 | break; | |
10765 | if (ordered) | |
10766 | btrfs_put_ordered_extent(ordered); | |
570eb97b | 10767 | unlock_extent(io_tree, start, end, &cached_state); |
7c0c7269 OS |
10768 | cond_resched(); |
10769 | } | |
10770 | ||
10771 | /* | |
10772 | * We don't use the higher-level delalloc space functions because our | |
10773 | * num_bytes and disk_num_bytes are different. | |
10774 | */ | |
10775 | ret = btrfs_alloc_data_chunk_ondemand(inode, disk_num_bytes); | |
10776 | if (ret) | |
10777 | goto out_unlock; | |
10778 | ret = btrfs_qgroup_reserve_data(inode, &data_reserved, start, num_bytes); | |
10779 | if (ret) | |
10780 | goto out_free_data_space; | |
d4135134 FM |
10781 | ret = btrfs_delalloc_reserve_metadata(inode, num_bytes, disk_num_bytes, |
10782 | false); | |
7c0c7269 OS |
10783 | if (ret) |
10784 | goto out_qgroup_free_data; | |
10785 | ||
10786 | /* Try an inline extent first. */ | |
10787 | if (start == 0 && encoded->unencoded_len == encoded->len && | |
10788 | encoded->unencoded_offset == 0) { | |
10789 | ret = cow_file_range_inline(inode, encoded->len, orig_count, | |
10790 | compression, pages, true); | |
10791 | if (ret <= 0) { | |
10792 | if (ret == 0) | |
10793 | ret = orig_count; | |
10794 | goto out_delalloc_release; | |
10795 | } | |
10796 | } | |
10797 | ||
10798 | ret = btrfs_reserve_extent(root, disk_num_bytes, disk_num_bytes, | |
10799 | disk_num_bytes, 0, 0, &ins, 1, 1); | |
10800 | if (ret) | |
10801 | goto out_delalloc_release; | |
10802 | extent_reserved = true; | |
10803 | ||
10804 | em = create_io_em(inode, start, num_bytes, | |
10805 | start - encoded->unencoded_offset, ins.objectid, | |
10806 | ins.offset, ins.offset, ram_bytes, compression, | |
10807 | BTRFS_ORDERED_COMPRESSED); | |
10808 | if (IS_ERR(em)) { | |
10809 | ret = PTR_ERR(em); | |
10810 | goto out_free_reserved; | |
10811 | } | |
10812 | free_extent_map(em); | |
10813 | ||
10814 | ret = btrfs_add_ordered_extent(inode, start, num_bytes, ram_bytes, | |
10815 | ins.objectid, ins.offset, | |
10816 | encoded->unencoded_offset, | |
10817 | (1 << BTRFS_ORDERED_ENCODED) | | |
10818 | (1 << BTRFS_ORDERED_COMPRESSED), | |
10819 | compression); | |
10820 | if (ret) { | |
10821 | btrfs_drop_extent_cache(inode, start, end, 0); | |
10822 | goto out_free_reserved; | |
10823 | } | |
10824 | btrfs_dec_block_group_reservations(fs_info, ins.objectid); | |
10825 | ||
10826 | if (start + encoded->len > inode->vfs_inode.i_size) | |
10827 | i_size_write(&inode->vfs_inode, start + encoded->len); | |
10828 | ||
570eb97b | 10829 | unlock_extent(io_tree, start, end, &cached_state); |
7c0c7269 OS |
10830 | |
10831 | btrfs_delalloc_release_extents(inode, num_bytes); | |
10832 | ||
10833 | if (btrfs_submit_compressed_write(inode, start, num_bytes, ins.objectid, | |
10834 | ins.offset, pages, nr_pages, 0, NULL, | |
10835 | false)) { | |
10836 | btrfs_writepage_endio_finish_ordered(inode, pages[0], start, end, 0); | |
10837 | ret = -EIO; | |
10838 | goto out_pages; | |
10839 | } | |
10840 | ret = orig_count; | |
10841 | goto out; | |
10842 | ||
10843 | out_free_reserved: | |
10844 | btrfs_dec_block_group_reservations(fs_info, ins.objectid); | |
10845 | btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1); | |
10846 | out_delalloc_release: | |
10847 | btrfs_delalloc_release_extents(inode, num_bytes); | |
10848 | btrfs_delalloc_release_metadata(inode, disk_num_bytes, ret < 0); | |
10849 | out_qgroup_free_data: | |
10850 | if (ret < 0) | |
10851 | btrfs_qgroup_free_data(inode, data_reserved, start, num_bytes); | |
10852 | out_free_data_space: | |
10853 | /* | |
10854 | * If btrfs_reserve_extent() succeeded, then we already decremented | |
10855 | * bytes_may_use. | |
10856 | */ | |
10857 | if (!extent_reserved) | |
10858 | btrfs_free_reserved_data_space_noquota(fs_info, disk_num_bytes); | |
10859 | out_unlock: | |
570eb97b | 10860 | unlock_extent(io_tree, start, end, &cached_state); |
7c0c7269 OS |
10861 | out_pages: |
10862 | for (i = 0; i < nr_pages; i++) { | |
10863 | if (pages[i]) | |
10864 | __free_page(pages[i]); | |
10865 | } | |
10866 | kvfree(pages); | |
10867 | out: | |
10868 | if (ret >= 0) | |
10869 | iocb->ki_pos += encoded->len; | |
10870 | return ret; | |
10871 | } | |
10872 | ||
ed46ff3d OS |
10873 | #ifdef CONFIG_SWAP |
10874 | /* | |
10875 | * Add an entry indicating a block group or device which is pinned by a | |
10876 | * swapfile. Returns 0 on success, 1 if there is already an entry for it, or a | |
10877 | * negative errno on failure. | |
10878 | */ | |
10879 | static int btrfs_add_swapfile_pin(struct inode *inode, void *ptr, | |
10880 | bool is_block_group) | |
10881 | { | |
10882 | struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info; | |
10883 | struct btrfs_swapfile_pin *sp, *entry; | |
10884 | struct rb_node **p; | |
10885 | struct rb_node *parent = NULL; | |
10886 | ||
10887 | sp = kmalloc(sizeof(*sp), GFP_NOFS); | |
10888 | if (!sp) | |
10889 | return -ENOMEM; | |
10890 | sp->ptr = ptr; | |
10891 | sp->inode = inode; | |
10892 | sp->is_block_group = is_block_group; | |
195a49ea | 10893 | sp->bg_extent_count = 1; |
ed46ff3d OS |
10894 | |
10895 | spin_lock(&fs_info->swapfile_pins_lock); | |
10896 | p = &fs_info->swapfile_pins.rb_node; | |
10897 | while (*p) { | |
10898 | parent = *p; | |
10899 | entry = rb_entry(parent, struct btrfs_swapfile_pin, node); | |
10900 | if (sp->ptr < entry->ptr || | |
10901 | (sp->ptr == entry->ptr && sp->inode < entry->inode)) { | |
10902 | p = &(*p)->rb_left; | |
10903 | } else if (sp->ptr > entry->ptr || | |
10904 | (sp->ptr == entry->ptr && sp->inode > entry->inode)) { | |
10905 | p = &(*p)->rb_right; | |
10906 | } else { | |
195a49ea FM |
10907 | if (is_block_group) |
10908 | entry->bg_extent_count++; | |
ed46ff3d OS |
10909 | spin_unlock(&fs_info->swapfile_pins_lock); |
10910 | kfree(sp); | |
10911 | return 1; | |
10912 | } | |
10913 | } | |
10914 | rb_link_node(&sp->node, parent, p); | |
10915 | rb_insert_color(&sp->node, &fs_info->swapfile_pins); | |
10916 | spin_unlock(&fs_info->swapfile_pins_lock); | |
10917 | return 0; | |
10918 | } | |
10919 | ||
10920 | /* Free all of the entries pinned by this swapfile. */ | |
10921 | static void btrfs_free_swapfile_pins(struct inode *inode) | |
10922 | { | |
10923 | struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info; | |
10924 | struct btrfs_swapfile_pin *sp; | |
10925 | struct rb_node *node, *next; | |
10926 | ||
10927 | spin_lock(&fs_info->swapfile_pins_lock); | |
10928 | node = rb_first(&fs_info->swapfile_pins); | |
10929 | while (node) { | |
10930 | next = rb_next(node); | |
10931 | sp = rb_entry(node, struct btrfs_swapfile_pin, node); | |
10932 | if (sp->inode == inode) { | |
10933 | rb_erase(&sp->node, &fs_info->swapfile_pins); | |
195a49ea FM |
10934 | if (sp->is_block_group) { |
10935 | btrfs_dec_block_group_swap_extents(sp->ptr, | |
10936 | sp->bg_extent_count); | |
ed46ff3d | 10937 | btrfs_put_block_group(sp->ptr); |
195a49ea | 10938 | } |
ed46ff3d OS |
10939 | kfree(sp); |
10940 | } | |
10941 | node = next; | |
10942 | } | |
10943 | spin_unlock(&fs_info->swapfile_pins_lock); | |
10944 | } | |
10945 | ||
10946 | struct btrfs_swap_info { | |
10947 | u64 start; | |
10948 | u64 block_start; | |
10949 | u64 block_len; | |
10950 | u64 lowest_ppage; | |
10951 | u64 highest_ppage; | |
10952 | unsigned long nr_pages; | |
10953 | int nr_extents; | |
10954 | }; | |
10955 | ||
10956 | static int btrfs_add_swap_extent(struct swap_info_struct *sis, | |
10957 | struct btrfs_swap_info *bsi) | |
10958 | { | |
10959 | unsigned long nr_pages; | |
c2f82263 | 10960 | unsigned long max_pages; |
ed46ff3d OS |
10961 | u64 first_ppage, first_ppage_reported, next_ppage; |
10962 | int ret; | |
10963 | ||
c2f82263 FM |
10964 | /* |
10965 | * Our swapfile may have had its size extended after the swap header was | |
10966 | * written. In that case activating the swapfile should not go beyond | |
10967 | * the max size set in the swap header. | |
10968 | */ | |
10969 | if (bsi->nr_pages >= sis->max) | |
10970 | return 0; | |
10971 | ||
10972 | max_pages = sis->max - bsi->nr_pages; | |
ed46ff3d OS |
10973 | first_ppage = ALIGN(bsi->block_start, PAGE_SIZE) >> PAGE_SHIFT; |
10974 | next_ppage = ALIGN_DOWN(bsi->block_start + bsi->block_len, | |
10975 | PAGE_SIZE) >> PAGE_SHIFT; | |
10976 | ||
10977 | if (first_ppage >= next_ppage) | |
10978 | return 0; | |
10979 | nr_pages = next_ppage - first_ppage; | |
c2f82263 | 10980 | nr_pages = min(nr_pages, max_pages); |
ed46ff3d OS |
10981 | |
10982 | first_ppage_reported = first_ppage; | |
10983 | if (bsi->start == 0) | |
10984 | first_ppage_reported++; | |
10985 | if (bsi->lowest_ppage > first_ppage_reported) | |
10986 | bsi->lowest_ppage = first_ppage_reported; | |
10987 | if (bsi->highest_ppage < (next_ppage - 1)) | |
10988 | bsi->highest_ppage = next_ppage - 1; | |
10989 | ||
10990 | ret = add_swap_extent(sis, bsi->nr_pages, nr_pages, first_ppage); | |
10991 | if (ret < 0) | |
10992 | return ret; | |
10993 | bsi->nr_extents += ret; | |
10994 | bsi->nr_pages += nr_pages; | |
10995 | return 0; | |
10996 | } | |
10997 | ||
10998 | static void btrfs_swap_deactivate(struct file *file) | |
10999 | { | |
11000 | struct inode *inode = file_inode(file); | |
11001 | ||
11002 | btrfs_free_swapfile_pins(inode); | |
11003 | atomic_dec(&BTRFS_I(inode)->root->nr_swapfiles); | |
11004 | } | |
11005 | ||
11006 | static int btrfs_swap_activate(struct swap_info_struct *sis, struct file *file, | |
11007 | sector_t *span) | |
11008 | { | |
11009 | struct inode *inode = file_inode(file); | |
dd0734f2 FM |
11010 | struct btrfs_root *root = BTRFS_I(inode)->root; |
11011 | struct btrfs_fs_info *fs_info = root->fs_info; | |
ed46ff3d OS |
11012 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
11013 | struct extent_state *cached_state = NULL; | |
11014 | struct extent_map *em = NULL; | |
11015 | struct btrfs_device *device = NULL; | |
11016 | struct btrfs_swap_info bsi = { | |
11017 | .lowest_ppage = (sector_t)-1ULL, | |
11018 | }; | |
11019 | int ret = 0; | |
11020 | u64 isize; | |
11021 | u64 start; | |
11022 | ||
11023 | /* | |
11024 | * If the swap file was just created, make sure delalloc is done. If the | |
11025 | * file changes again after this, the user is doing something stupid and | |
11026 | * we don't really care. | |
11027 | */ | |
11028 | ret = btrfs_wait_ordered_range(inode, 0, (u64)-1); | |
11029 | if (ret) | |
11030 | return ret; | |
11031 | ||
11032 | /* | |
11033 | * The inode is locked, so these flags won't change after we check them. | |
11034 | */ | |
11035 | if (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS) { | |
11036 | btrfs_warn(fs_info, "swapfile must not be compressed"); | |
11037 | return -EINVAL; | |
11038 | } | |
11039 | if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW)) { | |
11040 | btrfs_warn(fs_info, "swapfile must not be copy-on-write"); | |
11041 | return -EINVAL; | |
11042 | } | |
11043 | if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) { | |
11044 | btrfs_warn(fs_info, "swapfile must not be checksummed"); | |
11045 | return -EINVAL; | |
11046 | } | |
11047 | ||
11048 | /* | |
11049 | * Balance or device remove/replace/resize can move stuff around from | |
c3e1f96c GR |
11050 | * under us. The exclop protection makes sure they aren't running/won't |
11051 | * run concurrently while we are mapping the swap extents, and | |
11052 | * fs_info->swapfile_pins prevents them from running while the swap | |
11053 | * file is active and moving the extents. Note that this also prevents | |
11054 | * a concurrent device add which isn't actually necessary, but it's not | |
ed46ff3d OS |
11055 | * really worth the trouble to allow it. |
11056 | */ | |
c3e1f96c | 11057 | if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_SWAP_ACTIVATE)) { |
ed46ff3d OS |
11058 | btrfs_warn(fs_info, |
11059 | "cannot activate swapfile while exclusive operation is running"); | |
11060 | return -EBUSY; | |
11061 | } | |
dd0734f2 FM |
11062 | |
11063 | /* | |
11064 | * Prevent snapshot creation while we are activating the swap file. | |
11065 | * We do not want to race with snapshot creation. If snapshot creation | |
11066 | * already started before we bumped nr_swapfiles from 0 to 1 and | |
11067 | * completes before the first write into the swap file after it is | |
11068 | * activated, than that write would fallback to COW. | |
11069 | */ | |
11070 | if (!btrfs_drew_try_write_lock(&root->snapshot_lock)) { | |
11071 | btrfs_exclop_finish(fs_info); | |
11072 | btrfs_warn(fs_info, | |
11073 | "cannot activate swapfile because snapshot creation is in progress"); | |
11074 | return -EINVAL; | |
11075 | } | |
ed46ff3d OS |
11076 | /* |
11077 | * Snapshots can create extents which require COW even if NODATACOW is | |
11078 | * set. We use this counter to prevent snapshots. We must increment it | |
11079 | * before walking the extents because we don't want a concurrent | |
11080 | * snapshot to run after we've already checked the extents. | |
60021bd7 KH |
11081 | * |
11082 | * It is possible that subvolume is marked for deletion but still not | |
11083 | * removed yet. To prevent this race, we check the root status before | |
11084 | * activating the swapfile. | |
ed46ff3d | 11085 | */ |
60021bd7 KH |
11086 | spin_lock(&root->root_item_lock); |
11087 | if (btrfs_root_dead(root)) { | |
11088 | spin_unlock(&root->root_item_lock); | |
11089 | ||
11090 | btrfs_exclop_finish(fs_info); | |
11091 | btrfs_warn(fs_info, | |
11092 | "cannot activate swapfile because subvolume %llu is being deleted", | |
11093 | root->root_key.objectid); | |
11094 | return -EPERM; | |
11095 | } | |
dd0734f2 | 11096 | atomic_inc(&root->nr_swapfiles); |
60021bd7 | 11097 | spin_unlock(&root->root_item_lock); |
ed46ff3d OS |
11098 | |
11099 | isize = ALIGN_DOWN(inode->i_size, fs_info->sectorsize); | |
11100 | ||
570eb97b | 11101 | lock_extent(io_tree, 0, isize - 1, &cached_state); |
ed46ff3d OS |
11102 | start = 0; |
11103 | while (start < isize) { | |
11104 | u64 logical_block_start, physical_block_start; | |
32da5386 | 11105 | struct btrfs_block_group *bg; |
ed46ff3d OS |
11106 | u64 len = isize - start; |
11107 | ||
39b07b5d | 11108 | em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len); |
ed46ff3d OS |
11109 | if (IS_ERR(em)) { |
11110 | ret = PTR_ERR(em); | |
11111 | goto out; | |
11112 | } | |
11113 | ||
11114 | if (em->block_start == EXTENT_MAP_HOLE) { | |
11115 | btrfs_warn(fs_info, "swapfile must not have holes"); | |
11116 | ret = -EINVAL; | |
11117 | goto out; | |
11118 | } | |
11119 | if (em->block_start == EXTENT_MAP_INLINE) { | |
11120 | /* | |
11121 | * It's unlikely we'll ever actually find ourselves | |
11122 | * here, as a file small enough to fit inline won't be | |
11123 | * big enough to store more than the swap header, but in | |
11124 | * case something changes in the future, let's catch it | |
11125 | * here rather than later. | |
11126 | */ | |
11127 | btrfs_warn(fs_info, "swapfile must not be inline"); | |
11128 | ret = -EINVAL; | |
11129 | goto out; | |
11130 | } | |
11131 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) { | |
11132 | btrfs_warn(fs_info, "swapfile must not be compressed"); | |
11133 | ret = -EINVAL; | |
11134 | goto out; | |
11135 | } | |
11136 | ||
11137 | logical_block_start = em->block_start + (start - em->start); | |
11138 | len = min(len, em->len - (start - em->start)); | |
11139 | free_extent_map(em); | |
11140 | em = NULL; | |
11141 | ||
26ce9114 | 11142 | ret = can_nocow_extent(inode, start, &len, NULL, NULL, NULL, false, true); |
ed46ff3d OS |
11143 | if (ret < 0) { |
11144 | goto out; | |
11145 | } else if (ret) { | |
11146 | ret = 0; | |
11147 | } else { | |
11148 | btrfs_warn(fs_info, | |
11149 | "swapfile must not be copy-on-write"); | |
11150 | ret = -EINVAL; | |
11151 | goto out; | |
11152 | } | |
11153 | ||
11154 | em = btrfs_get_chunk_map(fs_info, logical_block_start, len); | |
11155 | if (IS_ERR(em)) { | |
11156 | ret = PTR_ERR(em); | |
11157 | goto out; | |
11158 | } | |
11159 | ||
11160 | if (em->map_lookup->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) { | |
11161 | btrfs_warn(fs_info, | |
11162 | "swapfile must have single data profile"); | |
11163 | ret = -EINVAL; | |
11164 | goto out; | |
11165 | } | |
11166 | ||
11167 | if (device == NULL) { | |
11168 | device = em->map_lookup->stripes[0].dev; | |
11169 | ret = btrfs_add_swapfile_pin(inode, device, false); | |
11170 | if (ret == 1) | |
11171 | ret = 0; | |
11172 | else if (ret) | |
11173 | goto out; | |
11174 | } else if (device != em->map_lookup->stripes[0].dev) { | |
11175 | btrfs_warn(fs_info, "swapfile must be on one device"); | |
11176 | ret = -EINVAL; | |
11177 | goto out; | |
11178 | } | |
11179 | ||
11180 | physical_block_start = (em->map_lookup->stripes[0].physical + | |
11181 | (logical_block_start - em->start)); | |
11182 | len = min(len, em->len - (logical_block_start - em->start)); | |
11183 | free_extent_map(em); | |
11184 | em = NULL; | |
11185 | ||
11186 | bg = btrfs_lookup_block_group(fs_info, logical_block_start); | |
11187 | if (!bg) { | |
11188 | btrfs_warn(fs_info, | |
11189 | "could not find block group containing swapfile"); | |
11190 | ret = -EINVAL; | |
11191 | goto out; | |
11192 | } | |
11193 | ||
195a49ea FM |
11194 | if (!btrfs_inc_block_group_swap_extents(bg)) { |
11195 | btrfs_warn(fs_info, | |
11196 | "block group for swapfile at %llu is read-only%s", | |
11197 | bg->start, | |
11198 | atomic_read(&fs_info->scrubs_running) ? | |
11199 | " (scrub running)" : ""); | |
11200 | btrfs_put_block_group(bg); | |
11201 | ret = -EINVAL; | |
11202 | goto out; | |
11203 | } | |
11204 | ||
ed46ff3d OS |
11205 | ret = btrfs_add_swapfile_pin(inode, bg, true); |
11206 | if (ret) { | |
11207 | btrfs_put_block_group(bg); | |
11208 | if (ret == 1) | |
11209 | ret = 0; | |
11210 | else | |
11211 | goto out; | |
11212 | } | |
11213 | ||
11214 | if (bsi.block_len && | |
11215 | bsi.block_start + bsi.block_len == physical_block_start) { | |
11216 | bsi.block_len += len; | |
11217 | } else { | |
11218 | if (bsi.block_len) { | |
11219 | ret = btrfs_add_swap_extent(sis, &bsi); | |
11220 | if (ret) | |
11221 | goto out; | |
11222 | } | |
11223 | bsi.start = start; | |
11224 | bsi.block_start = physical_block_start; | |
11225 | bsi.block_len = len; | |
11226 | } | |
11227 | ||
11228 | start += len; | |
11229 | } | |
11230 | ||
11231 | if (bsi.block_len) | |
11232 | ret = btrfs_add_swap_extent(sis, &bsi); | |
11233 | ||
11234 | out: | |
11235 | if (!IS_ERR_OR_NULL(em)) | |
11236 | free_extent_map(em); | |
11237 | ||
570eb97b | 11238 | unlock_extent(io_tree, 0, isize - 1, &cached_state); |
ed46ff3d OS |
11239 | |
11240 | if (ret) | |
11241 | btrfs_swap_deactivate(file); | |
11242 | ||
dd0734f2 FM |
11243 | btrfs_drew_write_unlock(&root->snapshot_lock); |
11244 | ||
c3e1f96c | 11245 | btrfs_exclop_finish(fs_info); |
ed46ff3d OS |
11246 | |
11247 | if (ret) | |
11248 | return ret; | |
11249 | ||
11250 | if (device) | |
11251 | sis->bdev = device->bdev; | |
11252 | *span = bsi.highest_ppage - bsi.lowest_ppage + 1; | |
11253 | sis->max = bsi.nr_pages; | |
11254 | sis->pages = bsi.nr_pages - 1; | |
11255 | sis->highest_bit = bsi.nr_pages - 1; | |
11256 | return bsi.nr_extents; | |
11257 | } | |
11258 | #else | |
11259 | static void btrfs_swap_deactivate(struct file *file) | |
11260 | { | |
11261 | } | |
11262 | ||
11263 | static int btrfs_swap_activate(struct swap_info_struct *sis, struct file *file, | |
11264 | sector_t *span) | |
11265 | { | |
11266 | return -EOPNOTSUPP; | |
11267 | } | |
11268 | #endif | |
11269 | ||
2766ff61 FM |
11270 | /* |
11271 | * Update the number of bytes used in the VFS' inode. When we replace extents in | |
11272 | * a range (clone, dedupe, fallocate's zero range), we must update the number of | |
11273 | * bytes used by the inode in an atomic manner, so that concurrent stat(2) calls | |
11274 | * always get a correct value. | |
11275 | */ | |
11276 | void btrfs_update_inode_bytes(struct btrfs_inode *inode, | |
11277 | const u64 add_bytes, | |
11278 | const u64 del_bytes) | |
11279 | { | |
11280 | if (add_bytes == del_bytes) | |
11281 | return; | |
11282 | ||
11283 | spin_lock(&inode->lock); | |
11284 | if (del_bytes > 0) | |
11285 | inode_sub_bytes(&inode->vfs_inode, del_bytes); | |
11286 | if (add_bytes > 0) | |
11287 | inode_add_bytes(&inode->vfs_inode, add_bytes); | |
11288 | spin_unlock(&inode->lock); | |
11289 | } | |
11290 | ||
63c34cb4 FM |
11291 | /** |
11292 | * Verify that there are no ordered extents for a given file range. | |
11293 | * | |
11294 | * @inode: The target inode. | |
11295 | * @start: Start offset of the file range, should be sector size aligned. | |
11296 | * @end: End offset (inclusive) of the file range, its value +1 should be | |
11297 | * sector size aligned. | |
11298 | * | |
11299 | * This should typically be used for cases where we locked an inode's VFS lock in | |
11300 | * exclusive mode, we have also locked the inode's i_mmap_lock in exclusive mode, | |
11301 | * we have flushed all delalloc in the range, we have waited for all ordered | |
11302 | * extents in the range to complete and finally we have locked the file range in | |
11303 | * the inode's io_tree. | |
11304 | */ | |
11305 | void btrfs_assert_inode_range_clean(struct btrfs_inode *inode, u64 start, u64 end) | |
11306 | { | |
11307 | struct btrfs_root *root = inode->root; | |
11308 | struct btrfs_ordered_extent *ordered; | |
11309 | ||
11310 | if (!IS_ENABLED(CONFIG_BTRFS_ASSERT)) | |
11311 | return; | |
11312 | ||
11313 | ordered = btrfs_lookup_first_ordered_range(inode, start, end + 1 - start); | |
11314 | if (ordered) { | |
11315 | btrfs_err(root->fs_info, | |
11316 | "found unexpected ordered extent in file range [%llu, %llu] for inode %llu root %llu (ordered range [%llu, %llu])", | |
11317 | start, end, btrfs_ino(inode), root->root_key.objectid, | |
11318 | ordered->file_offset, | |
11319 | ordered->file_offset + ordered->num_bytes - 1); | |
11320 | btrfs_put_ordered_extent(ordered); | |
11321 | } | |
11322 | ||
11323 | ASSERT(ordered == NULL); | |
11324 | } | |
11325 | ||
6e1d5dcc | 11326 | static const struct inode_operations btrfs_dir_inode_operations = { |
3394e160 | 11327 | .getattr = btrfs_getattr, |
39279cc3 CM |
11328 | .lookup = btrfs_lookup, |
11329 | .create = btrfs_create, | |
11330 | .unlink = btrfs_unlink, | |
11331 | .link = btrfs_link, | |
11332 | .mkdir = btrfs_mkdir, | |
11333 | .rmdir = btrfs_rmdir, | |
2773bf00 | 11334 | .rename = btrfs_rename2, |
39279cc3 CM |
11335 | .symlink = btrfs_symlink, |
11336 | .setattr = btrfs_setattr, | |
618e21d5 | 11337 | .mknod = btrfs_mknod, |
5103e947 | 11338 | .listxattr = btrfs_listxattr, |
fdebe2bd | 11339 | .permission = btrfs_permission, |
4e34e719 | 11340 | .get_acl = btrfs_get_acl, |
996a710d | 11341 | .set_acl = btrfs_set_acl, |
93fd63c2 | 11342 | .update_time = btrfs_update_time, |
ef3b9af5 | 11343 | .tmpfile = btrfs_tmpfile, |
97fc2977 MS |
11344 | .fileattr_get = btrfs_fileattr_get, |
11345 | .fileattr_set = btrfs_fileattr_set, | |
39279cc3 | 11346 | }; |
76dda93c | 11347 | |
828c0950 | 11348 | static const struct file_operations btrfs_dir_file_operations = { |
39279cc3 CM |
11349 | .llseek = generic_file_llseek, |
11350 | .read = generic_read_dir, | |
02dbfc99 | 11351 | .iterate_shared = btrfs_real_readdir, |
23b5ec74 | 11352 | .open = btrfs_opendir, |
34287aa3 | 11353 | .unlocked_ioctl = btrfs_ioctl, |
39279cc3 | 11354 | #ifdef CONFIG_COMPAT |
4c63c245 | 11355 | .compat_ioctl = btrfs_compat_ioctl, |
39279cc3 | 11356 | #endif |
6bf13c0c | 11357 | .release = btrfs_release_file, |
e02119d5 | 11358 | .fsync = btrfs_sync_file, |
39279cc3 CM |
11359 | }; |
11360 | ||
35054394 CM |
11361 | /* |
11362 | * btrfs doesn't support the bmap operation because swapfiles | |
11363 | * use bmap to make a mapping of extents in the file. They assume | |
11364 | * these extents won't change over the life of the file and they | |
11365 | * use the bmap result to do IO directly to the drive. | |
11366 | * | |
11367 | * the btrfs bmap call would return logical addresses that aren't | |
11368 | * suitable for IO and they also will change frequently as COW | |
11369 | * operations happen. So, swapfile + btrfs == corruption. | |
11370 | * | |
11371 | * For now we're avoiding this by dropping bmap. | |
11372 | */ | |
7f09410b | 11373 | static const struct address_space_operations btrfs_aops = { |
fb12489b | 11374 | .read_folio = btrfs_read_folio, |
b293f02e | 11375 | .writepages = btrfs_writepages, |
ba206a02 | 11376 | .readahead = btrfs_readahead, |
f85781fb | 11377 | .direct_IO = noop_direct_IO, |
895586eb | 11378 | .invalidate_folio = btrfs_invalidate_folio, |
f913cff3 | 11379 | .release_folio = btrfs_release_folio, |
e7a60a17 | 11380 | .migrate_folio = btrfs_migrate_folio, |
187c82cb | 11381 | .dirty_folio = filemap_dirty_folio, |
465fdd97 | 11382 | .error_remove_page = generic_error_remove_page, |
ed46ff3d OS |
11383 | .swap_activate = btrfs_swap_activate, |
11384 | .swap_deactivate = btrfs_swap_deactivate, | |
39279cc3 CM |
11385 | }; |
11386 | ||
6e1d5dcc | 11387 | static const struct inode_operations btrfs_file_inode_operations = { |
39279cc3 CM |
11388 | .getattr = btrfs_getattr, |
11389 | .setattr = btrfs_setattr, | |
5103e947 | 11390 | .listxattr = btrfs_listxattr, |
fdebe2bd | 11391 | .permission = btrfs_permission, |
1506fcc8 | 11392 | .fiemap = btrfs_fiemap, |
4e34e719 | 11393 | .get_acl = btrfs_get_acl, |
996a710d | 11394 | .set_acl = btrfs_set_acl, |
e41f941a | 11395 | .update_time = btrfs_update_time, |
97fc2977 MS |
11396 | .fileattr_get = btrfs_fileattr_get, |
11397 | .fileattr_set = btrfs_fileattr_set, | |
39279cc3 | 11398 | }; |
6e1d5dcc | 11399 | static const struct inode_operations btrfs_special_inode_operations = { |
618e21d5 JB |
11400 | .getattr = btrfs_getattr, |
11401 | .setattr = btrfs_setattr, | |
fdebe2bd | 11402 | .permission = btrfs_permission, |
33268eaf | 11403 | .listxattr = btrfs_listxattr, |
4e34e719 | 11404 | .get_acl = btrfs_get_acl, |
996a710d | 11405 | .set_acl = btrfs_set_acl, |
e41f941a | 11406 | .update_time = btrfs_update_time, |
618e21d5 | 11407 | }; |
6e1d5dcc | 11408 | static const struct inode_operations btrfs_symlink_inode_operations = { |
6b255391 | 11409 | .get_link = page_get_link, |
f209561a | 11410 | .getattr = btrfs_getattr, |
22c44fe6 | 11411 | .setattr = btrfs_setattr, |
fdebe2bd | 11412 | .permission = btrfs_permission, |
0279b4cd | 11413 | .listxattr = btrfs_listxattr, |
e41f941a | 11414 | .update_time = btrfs_update_time, |
39279cc3 | 11415 | }; |
76dda93c | 11416 | |
82d339d9 | 11417 | const struct dentry_operations btrfs_dentry_operations = { |
76dda93c YZ |
11418 | .d_delete = btrfs_dentry_delete, |
11419 | }; |