]>
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; | |
f28a4928 FM |
67 | }; |
68 | ||
88d2beec FM |
69 | struct btrfs_rename_ctx { |
70 | /* Output field. Stores the index number of the old directory entry. */ | |
71 | u64 index; | |
72 | }; | |
73 | ||
6e1d5dcc AD |
74 | static const struct inode_operations btrfs_dir_inode_operations; |
75 | static const struct inode_operations btrfs_symlink_inode_operations; | |
6e1d5dcc AD |
76 | static const struct inode_operations btrfs_special_inode_operations; |
77 | static const struct inode_operations btrfs_file_inode_operations; | |
7f09410b | 78 | static const struct address_space_operations btrfs_aops; |
828c0950 | 79 | static const struct file_operations btrfs_dir_file_operations; |
39279cc3 CM |
80 | |
81 | static struct kmem_cache *btrfs_inode_cachep; | |
82 | struct kmem_cache *btrfs_trans_handle_cachep; | |
39279cc3 | 83 | struct kmem_cache *btrfs_path_cachep; |
dc89e982 | 84 | struct kmem_cache *btrfs_free_space_cachep; |
3acd4850 | 85 | struct kmem_cache *btrfs_free_space_bitmap_cachep; |
39279cc3 | 86 | |
3972f260 | 87 | static int btrfs_setsize(struct inode *inode, struct iattr *attr); |
213e8c55 | 88 | static int btrfs_truncate(struct inode *inode, bool skip_writeback); |
5fd02043 | 89 | static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent); |
6e26c442 | 90 | static noinline int cow_file_range(struct btrfs_inode *inode, |
771ed689 | 91 | struct page *locked_page, |
74e9194a | 92 | u64 start, u64 end, int *page_started, |
330a5827 | 93 | unsigned long *nr_written, int unlock); |
4b67c11d NB |
94 | static struct extent_map *create_io_em(struct btrfs_inode *inode, u64 start, |
95 | u64 len, u64 orig_start, u64 block_start, | |
6f9994db LB |
96 | u64 block_len, u64 orig_block_len, |
97 | u64 ram_bytes, int compress_type, | |
98 | int type); | |
7b128766 | 99 | |
b672b5c1 | 100 | static void __endio_write_update_ordered(struct btrfs_inode *inode, |
52427260 QW |
101 | const u64 offset, const u64 bytes, |
102 | const bool uptodate); | |
103 | ||
a14b78ad GR |
104 | /* |
105 | * btrfs_inode_lock - lock inode i_rwsem based on arguments passed | |
106 | * | |
107 | * ilock_flags can have the following bit set: | |
108 | * | |
109 | * BTRFS_ILOCK_SHARED - acquire a shared lock on the inode | |
110 | * BTRFS_ILOCK_TRY - try to acquire the lock, if fails on first attempt | |
111 | * return -EAGAIN | |
8318ba79 | 112 | * BTRFS_ILOCK_MMAP - acquire a write lock on the i_mmap_lock |
a14b78ad GR |
113 | */ |
114 | int btrfs_inode_lock(struct inode *inode, unsigned int ilock_flags) | |
115 | { | |
116 | if (ilock_flags & BTRFS_ILOCK_SHARED) { | |
117 | if (ilock_flags & BTRFS_ILOCK_TRY) { | |
118 | if (!inode_trylock_shared(inode)) | |
119 | return -EAGAIN; | |
120 | else | |
121 | return 0; | |
122 | } | |
123 | inode_lock_shared(inode); | |
124 | } else { | |
125 | if (ilock_flags & BTRFS_ILOCK_TRY) { | |
126 | if (!inode_trylock(inode)) | |
127 | return -EAGAIN; | |
128 | else | |
129 | return 0; | |
130 | } | |
131 | inode_lock(inode); | |
132 | } | |
8318ba79 JB |
133 | if (ilock_flags & BTRFS_ILOCK_MMAP) |
134 | down_write(&BTRFS_I(inode)->i_mmap_lock); | |
a14b78ad GR |
135 | return 0; |
136 | } | |
137 | ||
138 | /* | |
139 | * btrfs_inode_unlock - unock inode i_rwsem | |
140 | * | |
141 | * ilock_flags should contain the same bits set as passed to btrfs_inode_lock() | |
142 | * to decide whether the lock acquired is shared or exclusive. | |
143 | */ | |
144 | void btrfs_inode_unlock(struct inode *inode, unsigned int ilock_flags) | |
145 | { | |
8318ba79 JB |
146 | if (ilock_flags & BTRFS_ILOCK_MMAP) |
147 | up_write(&BTRFS_I(inode)->i_mmap_lock); | |
a14b78ad GR |
148 | if (ilock_flags & BTRFS_ILOCK_SHARED) |
149 | inode_unlock_shared(inode); | |
150 | else | |
151 | inode_unlock(inode); | |
152 | } | |
153 | ||
52427260 QW |
154 | /* |
155 | * Cleanup all submitted ordered extents in specified range to handle errors | |
52042d8e | 156 | * from the btrfs_run_delalloc_range() callback. |
52427260 QW |
157 | * |
158 | * NOTE: caller must ensure that when an error happens, it can not call | |
159 | * extent_clear_unlock_delalloc() to clear both the bits EXTENT_DO_ACCOUNTING | |
160 | * and EXTENT_DELALLOC simultaneously, because that causes the reserved metadata | |
161 | * to be released, which we want to happen only when finishing the ordered | |
d1051d6e | 162 | * extent (btrfs_finish_ordered_io()). |
52427260 | 163 | */ |
64e1db56 | 164 | static inline void btrfs_cleanup_ordered_extents(struct btrfs_inode *inode, |
d1051d6e NB |
165 | struct page *locked_page, |
166 | u64 offset, u64 bytes) | |
52427260 | 167 | { |
63d71450 NA |
168 | unsigned long index = offset >> PAGE_SHIFT; |
169 | unsigned long end_index = (offset + bytes - 1) >> PAGE_SHIFT; | |
d1051d6e NB |
170 | u64 page_start = page_offset(locked_page); |
171 | u64 page_end = page_start + PAGE_SIZE - 1; | |
172 | ||
63d71450 NA |
173 | struct page *page; |
174 | ||
175 | while (index <= end_index) { | |
968f2566 QW |
176 | /* |
177 | * For locked page, we will call end_extent_writepage() on it | |
178 | * in run_delalloc_range() for the error handling. That | |
179 | * end_extent_writepage() function will call | |
180 | * btrfs_mark_ordered_io_finished() to clear page Ordered and | |
181 | * run the ordered extent accounting. | |
182 | * | |
183 | * Here we can't just clear the Ordered bit, or | |
184 | * btrfs_mark_ordered_io_finished() would skip the accounting | |
185 | * for the page range, and the ordered extent will never finish. | |
186 | */ | |
187 | if (index == (page_offset(locked_page) >> PAGE_SHIFT)) { | |
188 | index++; | |
189 | continue; | |
190 | } | |
64e1db56 | 191 | page = find_get_page(inode->vfs_inode.i_mapping, index); |
63d71450 NA |
192 | index++; |
193 | if (!page) | |
194 | continue; | |
968f2566 QW |
195 | |
196 | /* | |
197 | * Here we just clear all Ordered bits for every page in the | |
198 | * range, then __endio_write_update_ordered() will handle | |
199 | * the ordered extent accounting for the range. | |
200 | */ | |
b945a463 QW |
201 | btrfs_page_clamp_clear_ordered(inode->root->fs_info, page, |
202 | offset, bytes); | |
63d71450 NA |
203 | put_page(page); |
204 | } | |
d1051d6e | 205 | |
b945a463 QW |
206 | /* The locked page covers the full range, nothing needs to be done */ |
207 | if (bytes + offset <= page_offset(locked_page) + PAGE_SIZE) | |
208 | return; | |
d1051d6e NB |
209 | /* |
210 | * In case this page belongs to the delalloc range being instantiated | |
211 | * then skip it, since the first page of a range is going to be | |
212 | * properly cleaned up by the caller of run_delalloc_range | |
213 | */ | |
214 | if (page_start >= offset && page_end <= (offset + bytes - 1)) { | |
b945a463 QW |
215 | bytes = offset + bytes - page_offset(locked_page) - PAGE_SIZE; |
216 | offset = page_offset(locked_page) + PAGE_SIZE; | |
d1051d6e NB |
217 | } |
218 | ||
64e1db56 | 219 | return __endio_write_update_ordered(inode, offset, bytes, false); |
52427260 QW |
220 | } |
221 | ||
48a3b636 | 222 | static int btrfs_dirty_inode(struct inode *inode); |
7b128766 | 223 | |
f34f57a3 | 224 | static int btrfs_init_inode_security(struct btrfs_trans_handle *trans, |
2a7dba39 EP |
225 | struct inode *inode, struct inode *dir, |
226 | const struct qstr *qstr) | |
0279b4cd JO |
227 | { |
228 | int err; | |
229 | ||
f34f57a3 | 230 | err = btrfs_init_acl(trans, inode, dir); |
0279b4cd | 231 | if (!err) |
2a7dba39 | 232 | err = btrfs_xattr_security_init(trans, inode, dir, qstr); |
0279b4cd JO |
233 | return err; |
234 | } | |
235 | ||
c8b97818 CM |
236 | /* |
237 | * this does all the hard work for inserting an inline extent into | |
238 | * the btree. The caller should have done a btrfs_drop_extents so that | |
239 | * no overlapping inline items exist in the btree | |
240 | */ | |
40f76580 | 241 | static int insert_inline_extent(struct btrfs_trans_handle *trans, |
8dd9872d OS |
242 | struct btrfs_path *path, |
243 | struct btrfs_inode *inode, bool extent_inserted, | |
244 | size_t size, size_t compressed_size, | |
fe3f566c | 245 | int compress_type, |
d9496e8a OS |
246 | struct page **compressed_pages, |
247 | bool update_i_size) | |
c8b97818 | 248 | { |
8dd9872d | 249 | struct btrfs_root *root = inode->root; |
c8b97818 CM |
250 | struct extent_buffer *leaf; |
251 | struct page *page = NULL; | |
252 | char *kaddr; | |
253 | unsigned long ptr; | |
254 | struct btrfs_file_extent_item *ei; | |
c8b97818 CM |
255 | int ret; |
256 | size_t cur_size = size; | |
d9496e8a | 257 | u64 i_size; |
c8b97818 | 258 | |
982f1f5d JJB |
259 | ASSERT((compressed_size > 0 && compressed_pages) || |
260 | (compressed_size == 0 && !compressed_pages)); | |
261 | ||
fe3f566c | 262 | if (compressed_size && compressed_pages) |
c8b97818 | 263 | cur_size = compressed_size; |
c8b97818 | 264 | |
1acae57b FDBM |
265 | if (!extent_inserted) { |
266 | struct btrfs_key key; | |
267 | size_t datasize; | |
c8b97818 | 268 | |
8dd9872d OS |
269 | key.objectid = btrfs_ino(inode); |
270 | key.offset = 0; | |
962a298f | 271 | key.type = BTRFS_EXTENT_DATA_KEY; |
c8b97818 | 272 | |
1acae57b | 273 | datasize = btrfs_file_extent_calc_inline_size(cur_size); |
1acae57b FDBM |
274 | ret = btrfs_insert_empty_item(trans, root, path, &key, |
275 | datasize); | |
79b4f4c6 | 276 | if (ret) |
1acae57b | 277 | goto fail; |
c8b97818 CM |
278 | } |
279 | leaf = path->nodes[0]; | |
280 | ei = btrfs_item_ptr(leaf, path->slots[0], | |
281 | struct btrfs_file_extent_item); | |
282 | btrfs_set_file_extent_generation(leaf, ei, trans->transid); | |
283 | btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE); | |
284 | btrfs_set_file_extent_encryption(leaf, ei, 0); | |
285 | btrfs_set_file_extent_other_encoding(leaf, ei, 0); | |
286 | btrfs_set_file_extent_ram_bytes(leaf, ei, size); | |
287 | ptr = btrfs_file_extent_inline_start(ei); | |
288 | ||
261507a0 | 289 | if (compress_type != BTRFS_COMPRESS_NONE) { |
c8b97818 CM |
290 | struct page *cpage; |
291 | int i = 0; | |
d397712b | 292 | while (compressed_size > 0) { |
c8b97818 | 293 | cpage = compressed_pages[i]; |
5b050f04 | 294 | cur_size = min_t(unsigned long, compressed_size, |
09cbfeaf | 295 | PAGE_SIZE); |
c8b97818 | 296 | |
3a60f653 | 297 | kaddr = kmap_atomic(cpage); |
c8b97818 | 298 | write_extent_buffer(leaf, kaddr, ptr, cur_size); |
3a60f653 | 299 | kunmap_atomic(kaddr); |
c8b97818 CM |
300 | |
301 | i++; | |
302 | ptr += cur_size; | |
303 | compressed_size -= cur_size; | |
304 | } | |
305 | btrfs_set_file_extent_compression(leaf, ei, | |
261507a0 | 306 | compress_type); |
c8b97818 | 307 | } else { |
8dd9872d | 308 | page = find_get_page(inode->vfs_inode.i_mapping, 0); |
c8b97818 | 309 | btrfs_set_file_extent_compression(leaf, ei, 0); |
7ac687d9 | 310 | kaddr = kmap_atomic(page); |
8dd9872d | 311 | write_extent_buffer(leaf, kaddr, ptr, size); |
7ac687d9 | 312 | kunmap_atomic(kaddr); |
09cbfeaf | 313 | put_page(page); |
c8b97818 CM |
314 | } |
315 | btrfs_mark_buffer_dirty(leaf); | |
1acae57b | 316 | btrfs_release_path(path); |
c8b97818 | 317 | |
9ddc959e JB |
318 | /* |
319 | * We align size to sectorsize for inline extents just for simplicity | |
320 | * sake. | |
321 | */ | |
8dd9872d OS |
322 | ret = btrfs_inode_set_file_extent_range(inode, 0, |
323 | ALIGN(size, root->fs_info->sectorsize)); | |
9ddc959e JB |
324 | if (ret) |
325 | goto fail; | |
326 | ||
c2167754 | 327 | /* |
d9496e8a OS |
328 | * We're an inline extent, so nobody can extend the file past i_size |
329 | * without locking a page we already have locked. | |
c2167754 | 330 | * |
d9496e8a OS |
331 | * We must do any i_size and inode updates before we unlock the pages. |
332 | * Otherwise we could end up racing with unlink. | |
c2167754 | 333 | */ |
d9496e8a OS |
334 | i_size = i_size_read(&inode->vfs_inode); |
335 | if (update_i_size && size > i_size) { | |
336 | i_size_write(&inode->vfs_inode, size); | |
337 | i_size = size; | |
338 | } | |
339 | inode->disk_i_size = i_size; | |
8dd9872d | 340 | |
c8b97818 | 341 | fail: |
79b4f4c6 | 342 | return ret; |
c8b97818 CM |
343 | } |
344 | ||
345 | ||
346 | /* | |
347 | * conditionally insert an inline extent into the file. This | |
348 | * does the checks required to make sure the data is small enough | |
349 | * to fit as an inline extent. | |
350 | */ | |
8dd9872d OS |
351 | static noinline int cow_file_range_inline(struct btrfs_inode *inode, u64 size, |
352 | size_t compressed_size, | |
00361589 | 353 | int compress_type, |
d9496e8a OS |
354 | struct page **compressed_pages, |
355 | bool update_i_size) | |
c8b97818 | 356 | { |
5893dfb9 | 357 | struct btrfs_drop_extents_args drop_args = { 0 }; |
a0349401 | 358 | struct btrfs_root *root = inode->root; |
0b246afa | 359 | struct btrfs_fs_info *fs_info = root->fs_info; |
00361589 | 360 | struct btrfs_trans_handle *trans; |
8dd9872d | 361 | u64 data_len = (compressed_size ?: size); |
c8b97818 | 362 | int ret; |
1acae57b | 363 | struct btrfs_path *path; |
c8b97818 | 364 | |
8dd9872d OS |
365 | /* |
366 | * We can create an inline extent if it ends at or beyond the current | |
367 | * i_size, is no larger than a sector (decompressed), and the (possibly | |
368 | * compressed) data fits in a leaf and the configured maximum inline | |
369 | * size. | |
370 | */ | |
371 | if (size < i_size_read(&inode->vfs_inode) || | |
372 | size > fs_info->sectorsize || | |
0b246afa | 373 | data_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info) || |
8dd9872d | 374 | data_len > fs_info->max_inline) |
c8b97818 | 375 | return 1; |
c8b97818 | 376 | |
1acae57b FDBM |
377 | path = btrfs_alloc_path(); |
378 | if (!path) | |
379 | return -ENOMEM; | |
380 | ||
00361589 | 381 | trans = btrfs_join_transaction(root); |
1acae57b FDBM |
382 | if (IS_ERR(trans)) { |
383 | btrfs_free_path(path); | |
00361589 | 384 | return PTR_ERR(trans); |
1acae57b | 385 | } |
a0349401 | 386 | trans->block_rsv = &inode->block_rsv; |
00361589 | 387 | |
5893dfb9 | 388 | drop_args.path = path; |
8dd9872d OS |
389 | drop_args.start = 0; |
390 | drop_args.end = fs_info->sectorsize; | |
5893dfb9 FM |
391 | drop_args.drop_cache = true; |
392 | drop_args.replace_extent = true; | |
8dd9872d | 393 | drop_args.extent_item_size = btrfs_file_extent_calc_inline_size(data_len); |
5893dfb9 | 394 | ret = btrfs_drop_extents(trans, root, inode, &drop_args); |
00361589 | 395 | if (ret) { |
66642832 | 396 | btrfs_abort_transaction(trans, ret); |
00361589 JB |
397 | goto out; |
398 | } | |
c8b97818 | 399 | |
8dd9872d OS |
400 | ret = insert_inline_extent(trans, path, inode, drop_args.extent_inserted, |
401 | size, compressed_size, compress_type, | |
d9496e8a | 402 | compressed_pages, update_i_size); |
2adcac1a | 403 | if (ret && ret != -ENOSPC) { |
66642832 | 404 | btrfs_abort_transaction(trans, ret); |
00361589 | 405 | goto out; |
2adcac1a | 406 | } else if (ret == -ENOSPC) { |
00361589 JB |
407 | ret = 1; |
408 | goto out; | |
79787eaa | 409 | } |
2adcac1a | 410 | |
8dd9872d | 411 | btrfs_update_inode_bytes(inode, size, drop_args.bytes_found); |
9a56fcd1 | 412 | ret = btrfs_update_inode(trans, root, inode); |
2766ff61 FM |
413 | if (ret && ret != -ENOSPC) { |
414 | btrfs_abort_transaction(trans, ret); | |
415 | goto out; | |
416 | } else if (ret == -ENOSPC) { | |
417 | ret = 1; | |
418 | goto out; | |
419 | } | |
420 | ||
23e3337f | 421 | btrfs_set_inode_full_sync(inode); |
00361589 | 422 | out: |
94ed938a QW |
423 | /* |
424 | * Don't forget to free the reserved space, as for inlined extent | |
425 | * it won't count as data extent, free them directly here. | |
426 | * And at reserve time, it's always aligned to page size, so | |
427 | * just free one page here. | |
428 | */ | |
a0349401 | 429 | btrfs_qgroup_free_data(inode, NULL, 0, PAGE_SIZE); |
1acae57b | 430 | btrfs_free_path(path); |
3a45bb20 | 431 | btrfs_end_transaction(trans); |
00361589 | 432 | return ret; |
c8b97818 CM |
433 | } |
434 | ||
771ed689 CM |
435 | struct async_extent { |
436 | u64 start; | |
437 | u64 ram_size; | |
438 | u64 compressed_size; | |
439 | struct page **pages; | |
440 | unsigned long nr_pages; | |
261507a0 | 441 | int compress_type; |
771ed689 CM |
442 | struct list_head list; |
443 | }; | |
444 | ||
97db1204 | 445 | struct async_chunk { |
771ed689 | 446 | struct inode *inode; |
771ed689 CM |
447 | struct page *locked_page; |
448 | u64 start; | |
449 | u64 end; | |
f82b7359 | 450 | unsigned int write_flags; |
771ed689 | 451 | struct list_head extents; |
ec39f769 | 452 | struct cgroup_subsys_state *blkcg_css; |
771ed689 | 453 | struct btrfs_work work; |
9e895a8f | 454 | struct async_cow *async_cow; |
771ed689 CM |
455 | }; |
456 | ||
97db1204 | 457 | struct async_cow { |
97db1204 NB |
458 | atomic_t num_chunks; |
459 | struct async_chunk chunks[]; | |
771ed689 CM |
460 | }; |
461 | ||
97db1204 | 462 | static noinline int add_async_extent(struct async_chunk *cow, |
771ed689 CM |
463 | u64 start, u64 ram_size, |
464 | u64 compressed_size, | |
465 | struct page **pages, | |
261507a0 LZ |
466 | unsigned long nr_pages, |
467 | int compress_type) | |
771ed689 CM |
468 | { |
469 | struct async_extent *async_extent; | |
470 | ||
471 | async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS); | |
79787eaa | 472 | BUG_ON(!async_extent); /* -ENOMEM */ |
771ed689 CM |
473 | async_extent->start = start; |
474 | async_extent->ram_size = ram_size; | |
475 | async_extent->compressed_size = compressed_size; | |
476 | async_extent->pages = pages; | |
477 | async_extent->nr_pages = nr_pages; | |
261507a0 | 478 | async_extent->compress_type = compress_type; |
771ed689 CM |
479 | list_add_tail(&async_extent->list, &cow->extents); |
480 | return 0; | |
481 | } | |
482 | ||
42c16da6 QW |
483 | /* |
484 | * Check if the inode needs to be submitted to compression, based on mount | |
485 | * options, defragmentation, properties or heuristics. | |
486 | */ | |
808a1292 NB |
487 | static inline int inode_need_compress(struct btrfs_inode *inode, u64 start, |
488 | u64 end) | |
f79707b0 | 489 | { |
808a1292 | 490 | struct btrfs_fs_info *fs_info = inode->root->fs_info; |
f79707b0 | 491 | |
e6f9d696 | 492 | if (!btrfs_inode_can_compress(inode)) { |
42c16da6 QW |
493 | WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG), |
494 | KERN_ERR "BTRFS: unexpected compression for ino %llu\n", | |
808a1292 | 495 | btrfs_ino(inode)); |
42c16da6 QW |
496 | return 0; |
497 | } | |
0cf9b244 QW |
498 | /* |
499 | * Special check for subpage. | |
500 | * | |
501 | * We lock the full page then run each delalloc range in the page, thus | |
502 | * for the following case, we will hit some subpage specific corner case: | |
503 | * | |
504 | * 0 32K 64K | |
505 | * | |///////| |///////| | |
506 | * \- A \- B | |
507 | * | |
508 | * In above case, both range A and range B will try to unlock the full | |
509 | * page [0, 64K), causing the one finished later will have page | |
510 | * unlocked already, triggering various page lock requirement BUG_ON()s. | |
511 | * | |
512 | * So here we add an artificial limit that subpage compression can only | |
513 | * if the range is fully page aligned. | |
514 | * | |
515 | * In theory we only need to ensure the first page is fully covered, but | |
516 | * the tailing partial page will be locked until the full compression | |
517 | * finishes, delaying the write of other range. | |
518 | * | |
519 | * TODO: Make btrfs_run_delalloc_range() to lock all delalloc range | |
520 | * first to prevent any submitted async extent to unlock the full page. | |
521 | * By this, we can ensure for subpage case that only the last async_cow | |
522 | * will unlock the full page. | |
523 | */ | |
524 | if (fs_info->sectorsize < PAGE_SIZE) { | |
525 | if (!IS_ALIGNED(start, PAGE_SIZE) || | |
526 | !IS_ALIGNED(end + 1, PAGE_SIZE)) | |
527 | return 0; | |
528 | } | |
529 | ||
f79707b0 | 530 | /* force compress */ |
0b246afa | 531 | if (btrfs_test_opt(fs_info, FORCE_COMPRESS)) |
f79707b0 | 532 | return 1; |
eec63c65 | 533 | /* defrag ioctl */ |
808a1292 | 534 | if (inode->defrag_compress) |
eec63c65 | 535 | return 1; |
f79707b0 | 536 | /* bad compression ratios */ |
808a1292 | 537 | if (inode->flags & BTRFS_INODE_NOCOMPRESS) |
f79707b0 | 538 | return 0; |
0b246afa | 539 | if (btrfs_test_opt(fs_info, COMPRESS) || |
808a1292 NB |
540 | inode->flags & BTRFS_INODE_COMPRESS || |
541 | inode->prop_compress) | |
542 | return btrfs_compress_heuristic(&inode->vfs_inode, start, end); | |
f79707b0 WS |
543 | return 0; |
544 | } | |
545 | ||
6158e1ce | 546 | static inline void inode_should_defrag(struct btrfs_inode *inode, |
558732df | 547 | u64 start, u64 end, u64 num_bytes, u32 small_write) |
26d30f85 AJ |
548 | { |
549 | /* If this is a small write inside eof, kick off a defrag */ | |
550 | if (num_bytes < small_write && | |
6158e1ce | 551 | (start > 0 || end + 1 < inode->disk_i_size)) |
558732df | 552 | btrfs_add_inode_defrag(NULL, inode, small_write); |
26d30f85 AJ |
553 | } |
554 | ||
d352ac68 | 555 | /* |
771ed689 CM |
556 | * we create compressed extents in two phases. The first |
557 | * phase compresses a range of pages that have already been | |
558 | * locked (both pages and state bits are locked). | |
c8b97818 | 559 | * |
771ed689 CM |
560 | * This is done inside an ordered work queue, and the compression |
561 | * is spread across many cpus. The actual IO submission is step | |
562 | * two, and the ordered work queue takes care of making sure that | |
563 | * happens in the same order things were put onto the queue by | |
564 | * writepages and friends. | |
c8b97818 | 565 | * |
771ed689 CM |
566 | * If this code finds it can't get good compression, it puts an |
567 | * entry onto the work queue to write the uncompressed bytes. This | |
568 | * makes sure that both compressed inodes and uncompressed inodes | |
b2570314 AB |
569 | * are written in the same order that the flusher thread sent them |
570 | * down. | |
d352ac68 | 571 | */ |
ac3e9933 | 572 | static noinline int compress_file_range(struct async_chunk *async_chunk) |
b888db2b | 573 | { |
1368c6da | 574 | struct inode *inode = async_chunk->inode; |
0b246afa | 575 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
0b246afa | 576 | u64 blocksize = fs_info->sectorsize; |
1368c6da NB |
577 | u64 start = async_chunk->start; |
578 | u64 end = async_chunk->end; | |
c8b97818 | 579 | u64 actual_end; |
d98da499 | 580 | u64 i_size; |
e6dcd2dc | 581 | int ret = 0; |
c8b97818 CM |
582 | struct page **pages = NULL; |
583 | unsigned long nr_pages; | |
c8b97818 CM |
584 | unsigned long total_compressed = 0; |
585 | unsigned long total_in = 0; | |
c8b97818 CM |
586 | int i; |
587 | int will_compress; | |
0b246afa | 588 | int compress_type = fs_info->compress_type; |
ac3e9933 | 589 | int compressed_extents = 0; |
4adaa611 | 590 | int redirty = 0; |
b888db2b | 591 | |
6158e1ce NB |
592 | inode_should_defrag(BTRFS_I(inode), start, end, end - start + 1, |
593 | SZ_16K); | |
4cb5300b | 594 | |
d98da499 JB |
595 | /* |
596 | * We need to save i_size before now because it could change in between | |
597 | * us evaluating the size and assigning it. This is because we lock and | |
598 | * unlock the page in truncate and fallocate, and then modify the i_size | |
599 | * later on. | |
600 | * | |
601 | * The barriers are to emulate READ_ONCE, remove that once i_size_read | |
602 | * does that for us. | |
603 | */ | |
604 | barrier(); | |
605 | i_size = i_size_read(inode); | |
606 | barrier(); | |
607 | actual_end = min_t(u64, i_size, end + 1); | |
c8b97818 CM |
608 | again: |
609 | will_compress = 0; | |
09cbfeaf | 610 | nr_pages = (end >> PAGE_SHIFT) - (start >> PAGE_SHIFT) + 1; |
069eac78 DS |
611 | nr_pages = min_t(unsigned long, nr_pages, |
612 | BTRFS_MAX_COMPRESSED / PAGE_SIZE); | |
be20aa9d | 613 | |
f03d9301 CM |
614 | /* |
615 | * we don't want to send crud past the end of i_size through | |
616 | * compression, that's just a waste of CPU time. So, if the | |
617 | * end of the file is before the start of our current | |
618 | * requested range of bytes, we bail out to the uncompressed | |
619 | * cleanup code that can deal with all of this. | |
620 | * | |
621 | * It isn't really the fastest way to fix things, but this is a | |
622 | * very uncommon corner. | |
623 | */ | |
624 | if (actual_end <= start) | |
625 | goto cleanup_and_bail_uncompressed; | |
626 | ||
c8b97818 CM |
627 | total_compressed = actual_end - start; |
628 | ||
4bcbb332 | 629 | /* |
0cf9b244 | 630 | * Skip compression for a small file range(<=blocksize) that |
01327610 | 631 | * isn't an inline extent, since it doesn't save disk space at all. |
4bcbb332 SW |
632 | */ |
633 | if (total_compressed <= blocksize && | |
634 | (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size)) | |
635 | goto cleanup_and_bail_uncompressed; | |
636 | ||
0cf9b244 QW |
637 | /* |
638 | * For subpage case, we require full page alignment for the sector | |
639 | * aligned range. | |
640 | * Thus we must also check against @actual_end, not just @end. | |
641 | */ | |
642 | if (blocksize < PAGE_SIZE) { | |
643 | if (!IS_ALIGNED(start, PAGE_SIZE) || | |
644 | !IS_ALIGNED(round_up(actual_end, blocksize), PAGE_SIZE)) | |
645 | goto cleanup_and_bail_uncompressed; | |
646 | } | |
647 | ||
069eac78 DS |
648 | total_compressed = min_t(unsigned long, total_compressed, |
649 | BTRFS_MAX_UNCOMPRESSED); | |
c8b97818 CM |
650 | total_in = 0; |
651 | ret = 0; | |
db94535d | 652 | |
771ed689 CM |
653 | /* |
654 | * we do compression for mount -o compress and when the | |
655 | * inode has not been flagged as nocompress. This flag can | |
656 | * change at any time if we discover bad compression ratios. | |
c8b97818 | 657 | */ |
4e965576 | 658 | if (inode_need_compress(BTRFS_I(inode), start, end)) { |
c8b97818 | 659 | WARN_ON(pages); |
31e818fe | 660 | pages = kcalloc(nr_pages, sizeof(struct page *), GFP_NOFS); |
560f7d75 LZ |
661 | if (!pages) { |
662 | /* just bail out to the uncompressed code */ | |
3527a018 | 663 | nr_pages = 0; |
560f7d75 LZ |
664 | goto cont; |
665 | } | |
c8b97818 | 666 | |
eec63c65 DS |
667 | if (BTRFS_I(inode)->defrag_compress) |
668 | compress_type = BTRFS_I(inode)->defrag_compress; | |
669 | else if (BTRFS_I(inode)->prop_compress) | |
b52aa8c9 | 670 | compress_type = BTRFS_I(inode)->prop_compress; |
261507a0 | 671 | |
4adaa611 CM |
672 | /* |
673 | * we need to call clear_page_dirty_for_io on each | |
674 | * page in the range. Otherwise applications with the file | |
675 | * mmap'd can wander in and change the page contents while | |
676 | * we are compressing them. | |
677 | * | |
678 | * If the compression fails for any reason, we set the pages | |
679 | * dirty again later on. | |
e9679de3 TT |
680 | * |
681 | * Note that the remaining part is redirtied, the start pointer | |
682 | * has moved, the end is the original one. | |
4adaa611 | 683 | */ |
e9679de3 TT |
684 | if (!redirty) { |
685 | extent_range_clear_dirty_for_io(inode, start, end); | |
686 | redirty = 1; | |
687 | } | |
f51d2b59 DS |
688 | |
689 | /* Compression level is applied here and only here */ | |
690 | ret = btrfs_compress_pages( | |
691 | compress_type | (fs_info->compress_level << 4), | |
261507a0 | 692 | inode->i_mapping, start, |
38c31464 | 693 | pages, |
4d3a800e | 694 | &nr_pages, |
261507a0 | 695 | &total_in, |
e5d74902 | 696 | &total_compressed); |
c8b97818 CM |
697 | |
698 | if (!ret) { | |
7073017a | 699 | unsigned long offset = offset_in_page(total_compressed); |
4d3a800e | 700 | struct page *page = pages[nr_pages - 1]; |
c8b97818 CM |
701 | |
702 | /* zero the tail end of the last page, we might be | |
703 | * sending it down to disk | |
704 | */ | |
d048b9c2 IW |
705 | if (offset) |
706 | memzero_page(page, offset, PAGE_SIZE - offset); | |
c8b97818 CM |
707 | will_compress = 1; |
708 | } | |
709 | } | |
560f7d75 | 710 | cont: |
7367253a QW |
711 | /* |
712 | * Check cow_file_range() for why we don't even try to create inline | |
713 | * extent for subpage case. | |
714 | */ | |
715 | if (start == 0 && fs_info->sectorsize == PAGE_SIZE) { | |
c8b97818 | 716 | /* lets try to make an inline extent */ |
6018ba0a | 717 | if (ret || total_in < actual_end) { |
c8b97818 | 718 | /* we didn't compress the entire range, try |
771ed689 | 719 | * to make an uncompressed inline extent. |
c8b97818 | 720 | */ |
8dd9872d | 721 | ret = cow_file_range_inline(BTRFS_I(inode), actual_end, |
a0349401 | 722 | 0, BTRFS_COMPRESS_NONE, |
d9496e8a | 723 | NULL, false); |
c8b97818 | 724 | } else { |
771ed689 | 725 | /* try making a compressed inline extent */ |
8dd9872d | 726 | ret = cow_file_range_inline(BTRFS_I(inode), actual_end, |
fe3f566c | 727 | total_compressed, |
d9496e8a OS |
728 | compress_type, pages, |
729 | false); | |
c8b97818 | 730 | } |
79787eaa | 731 | if (ret <= 0) { |
151a41bc | 732 | unsigned long clear_flags = EXTENT_DELALLOC | |
8b62f87b JB |
733 | EXTENT_DELALLOC_NEW | EXTENT_DEFRAG | |
734 | EXTENT_DO_ACCOUNTING; | |
e6eb4314 FM |
735 | unsigned long page_error_op; |
736 | ||
e6eb4314 | 737 | page_error_op = ret < 0 ? PAGE_SET_ERROR : 0; |
151a41bc | 738 | |
771ed689 | 739 | /* |
79787eaa JM |
740 | * inline extent creation worked or returned error, |
741 | * we don't need to create any more async work items. | |
742 | * Unlock and free up our temp pages. | |
8b62f87b JB |
743 | * |
744 | * We use DO_ACCOUNTING here because we need the | |
745 | * delalloc_release_metadata to be done _after_ we drop | |
746 | * our outstanding extent for clearing delalloc for this | |
747 | * range. | |
771ed689 | 748 | */ |
ad7ff17b NB |
749 | extent_clear_unlock_delalloc(BTRFS_I(inode), start, end, |
750 | NULL, | |
74e9194a | 751 | clear_flags, |
ba8b04c1 | 752 | PAGE_UNLOCK | |
6869b0a8 | 753 | PAGE_START_WRITEBACK | |
e6eb4314 | 754 | page_error_op | |
c2790a2e | 755 | PAGE_END_WRITEBACK); |
cecc8d90 | 756 | |
1e6e238c QW |
757 | /* |
758 | * Ensure we only free the compressed pages if we have | |
759 | * them allocated, as we can still reach here with | |
760 | * inode_need_compress() == false. | |
761 | */ | |
762 | if (pages) { | |
763 | for (i = 0; i < nr_pages; i++) { | |
764 | WARN_ON(pages[i]->mapping); | |
765 | put_page(pages[i]); | |
766 | } | |
767 | kfree(pages); | |
cecc8d90 | 768 | } |
cecc8d90 | 769 | return 0; |
c8b97818 CM |
770 | } |
771 | } | |
772 | ||
773 | if (will_compress) { | |
774 | /* | |
775 | * we aren't doing an inline extent round the compressed size | |
776 | * up to a block size boundary so the allocator does sane | |
777 | * things | |
778 | */ | |
fda2832f | 779 | total_compressed = ALIGN(total_compressed, blocksize); |
c8b97818 CM |
780 | |
781 | /* | |
782 | * one last check to make sure the compression is really a | |
170607eb TT |
783 | * win, compare the page count read with the blocks on disk, |
784 | * compression must free at least one sector size | |
c8b97818 | 785 | */ |
4c162778 | 786 | total_in = round_up(total_in, fs_info->sectorsize); |
170607eb | 787 | if (total_compressed + blocksize <= total_in) { |
ac3e9933 | 788 | compressed_extents++; |
c8bb0c8b AS |
789 | |
790 | /* | |
791 | * The async work queues will take care of doing actual | |
792 | * allocation on disk for these compressed pages, and | |
793 | * will submit them to the elevator. | |
794 | */ | |
b5326271 | 795 | add_async_extent(async_chunk, start, total_in, |
4d3a800e | 796 | total_compressed, pages, nr_pages, |
c8bb0c8b AS |
797 | compress_type); |
798 | ||
1170862d TT |
799 | if (start + total_in < end) { |
800 | start += total_in; | |
c8bb0c8b AS |
801 | pages = NULL; |
802 | cond_resched(); | |
803 | goto again; | |
804 | } | |
ac3e9933 | 805 | return compressed_extents; |
c8b97818 CM |
806 | } |
807 | } | |
c8bb0c8b | 808 | if (pages) { |
c8b97818 CM |
809 | /* |
810 | * the compression code ran but failed to make things smaller, | |
811 | * free any pages it allocated and our page pointer array | |
812 | */ | |
4d3a800e | 813 | for (i = 0; i < nr_pages; i++) { |
70b99e69 | 814 | WARN_ON(pages[i]->mapping); |
09cbfeaf | 815 | put_page(pages[i]); |
c8b97818 CM |
816 | } |
817 | kfree(pages); | |
818 | pages = NULL; | |
819 | total_compressed = 0; | |
4d3a800e | 820 | nr_pages = 0; |
c8b97818 CM |
821 | |
822 | /* flag the file so we don't compress in the future */ | |
0b246afa | 823 | if (!btrfs_test_opt(fs_info, FORCE_COMPRESS) && |
b52aa8c9 | 824 | !(BTRFS_I(inode)->prop_compress)) { |
a555f810 | 825 | BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS; |
1e701a32 | 826 | } |
c8b97818 | 827 | } |
f03d9301 | 828 | cleanup_and_bail_uncompressed: |
c8bb0c8b AS |
829 | /* |
830 | * No compression, but we still need to write the pages in the file | |
831 | * we've been given so far. redirty the locked page if it corresponds | |
832 | * to our extent and set things up for the async work queue to run | |
833 | * cow_file_range to do the normal delalloc dance. | |
834 | */ | |
1d53c9e6 CM |
835 | if (async_chunk->locked_page && |
836 | (page_offset(async_chunk->locked_page) >= start && | |
837 | page_offset(async_chunk->locked_page)) <= end) { | |
1368c6da | 838 | __set_page_dirty_nobuffers(async_chunk->locked_page); |
c8bb0c8b | 839 | /* unlocked later on in the async handlers */ |
1d53c9e6 | 840 | } |
c8bb0c8b AS |
841 | |
842 | if (redirty) | |
843 | extent_range_redirty_for_io(inode, start, end); | |
b5326271 | 844 | add_async_extent(async_chunk, start, end - start + 1, 0, NULL, 0, |
c8bb0c8b | 845 | BTRFS_COMPRESS_NONE); |
ac3e9933 | 846 | compressed_extents++; |
3b951516 | 847 | |
ac3e9933 | 848 | return compressed_extents; |
771ed689 | 849 | } |
771ed689 | 850 | |
40ae837b FM |
851 | static void free_async_extent_pages(struct async_extent *async_extent) |
852 | { | |
853 | int i; | |
854 | ||
855 | if (!async_extent->pages) | |
856 | return; | |
857 | ||
858 | for (i = 0; i < async_extent->nr_pages; i++) { | |
859 | WARN_ON(async_extent->pages[i]->mapping); | |
09cbfeaf | 860 | put_page(async_extent->pages[i]); |
40ae837b FM |
861 | } |
862 | kfree(async_extent->pages); | |
863 | async_extent->nr_pages = 0; | |
864 | async_extent->pages = NULL; | |
771ed689 CM |
865 | } |
866 | ||
2b83a0ee QW |
867 | static int submit_uncompressed_range(struct btrfs_inode *inode, |
868 | struct async_extent *async_extent, | |
869 | struct page *locked_page) | |
771ed689 | 870 | { |
2b83a0ee QW |
871 | u64 start = async_extent->start; |
872 | u64 end = async_extent->start + async_extent->ram_size - 1; | |
873 | unsigned long nr_written = 0; | |
874 | int page_started = 0; | |
875 | int ret; | |
771ed689 | 876 | |
2b83a0ee QW |
877 | /* |
878 | * Call cow_file_range() to run the delalloc range directly, since we | |
879 | * won't go to NOCOW or async path again. | |
880 | * | |
881 | * Also we call cow_file_range() with @unlock_page == 0, so that we | |
882 | * can directly submit them without interruption. | |
883 | */ | |
884 | ret = cow_file_range(inode, locked_page, start, end, &page_started, | |
885 | &nr_written, 0); | |
886 | /* Inline extent inserted, page gets unlocked and everything is done */ | |
887 | if (page_started) { | |
888 | ret = 0; | |
889 | goto out; | |
890 | } | |
891 | if (ret < 0) { | |
892 | if (locked_page) | |
893 | unlock_page(locked_page); | |
894 | goto out; | |
895 | } | |
771ed689 | 896 | |
2b83a0ee QW |
897 | ret = extent_write_locked_range(&inode->vfs_inode, start, end); |
898 | /* All pages will be unlocked, including @locked_page */ | |
899 | out: | |
900 | kfree(async_extent); | |
901 | return ret; | |
902 | } | |
79787eaa | 903 | |
b4ccace8 QW |
904 | static int submit_one_async_extent(struct btrfs_inode *inode, |
905 | struct async_chunk *async_chunk, | |
906 | struct async_extent *async_extent, | |
907 | u64 *alloc_hint) | |
771ed689 | 908 | { |
b4ccace8 QW |
909 | struct extent_io_tree *io_tree = &inode->io_tree; |
910 | struct btrfs_root *root = inode->root; | |
911 | struct btrfs_fs_info *fs_info = root->fs_info; | |
771ed689 | 912 | struct btrfs_key ins; |
2b83a0ee | 913 | struct page *locked_page = NULL; |
771ed689 | 914 | struct extent_map *em; |
f5a84ee3 | 915 | int ret = 0; |
b4ccace8 QW |
916 | u64 start = async_extent->start; |
917 | u64 end = async_extent->start + async_extent->ram_size - 1; | |
771ed689 | 918 | |
2b83a0ee QW |
919 | /* |
920 | * If async_chunk->locked_page is in the async_extent range, we need to | |
921 | * handle it. | |
922 | */ | |
923 | if (async_chunk->locked_page) { | |
924 | u64 locked_page_start = page_offset(async_chunk->locked_page); | |
925 | u64 locked_page_end = locked_page_start + PAGE_SIZE - 1; | |
3e04e7f1 | 926 | |
2b83a0ee QW |
927 | if (!(start >= locked_page_end || end <= locked_page_start)) |
928 | locked_page = async_chunk->locked_page; | |
b4ccace8 | 929 | } |
2b83a0ee | 930 | lock_extent(io_tree, start, end); |
ce62003f | 931 | |
2b83a0ee QW |
932 | /* We have fall back to uncompressed write */ |
933 | if (!async_extent->pages) | |
934 | return submit_uncompressed_range(inode, async_extent, locked_page); | |
ce62003f | 935 | |
b4ccace8 QW |
936 | ret = btrfs_reserve_extent(root, async_extent->ram_size, |
937 | async_extent->compressed_size, | |
938 | async_extent->compressed_size, | |
939 | 0, *alloc_hint, &ins, 1, 1); | |
940 | if (ret) { | |
941 | free_async_extent_pages(async_extent); | |
c2167754 | 942 | /* |
b4ccace8 QW |
943 | * Here we used to try again by going back to non-compressed |
944 | * path for ENOSPC. But we can't reserve space even for | |
945 | * compressed size, how could it work for uncompressed size | |
946 | * which requires larger size? So here we directly go error | |
947 | * path. | |
c2167754 | 948 | */ |
b4ccace8 QW |
949 | goto out_free; |
950 | } | |
951 | ||
952 | /* Here we're doing allocation and writeback of the compressed pages */ | |
953 | em = create_io_em(inode, start, | |
954 | async_extent->ram_size, /* len */ | |
955 | start, /* orig_start */ | |
956 | ins.objectid, /* block_start */ | |
957 | ins.offset, /* block_len */ | |
958 | ins.offset, /* orig_block_len */ | |
959 | async_extent->ram_size, /* ram_bytes */ | |
960 | async_extent->compress_type, | |
961 | BTRFS_ORDERED_COMPRESSED); | |
962 | if (IS_ERR(em)) { | |
963 | ret = PTR_ERR(em); | |
964 | goto out_free_reserve; | |
965 | } | |
966 | free_extent_map(em); | |
771ed689 | 967 | |
cb36a9bb OS |
968 | ret = btrfs_add_ordered_extent(inode, start, /* file_offset */ |
969 | async_extent->ram_size, /* num_bytes */ | |
970 | async_extent->ram_size, /* ram_bytes */ | |
971 | ins.objectid, /* disk_bytenr */ | |
972 | ins.offset, /* disk_num_bytes */ | |
973 | 0, /* offset */ | |
974 | 1 << BTRFS_ORDERED_COMPRESSED, | |
975 | async_extent->compress_type); | |
b4ccace8 QW |
976 | if (ret) { |
977 | btrfs_drop_extent_cache(inode, start, end, 0); | |
978 | goto out_free_reserve; | |
771ed689 | 979 | } |
b4ccace8 QW |
980 | btrfs_dec_block_group_reservations(fs_info, ins.objectid); |
981 | ||
982 | /* Clear dirty, set writeback and unlock the pages. */ | |
983 | extent_clear_unlock_delalloc(inode, start, end, | |
984 | NULL, EXTENT_LOCKED | EXTENT_DELALLOC, | |
985 | PAGE_UNLOCK | PAGE_START_WRITEBACK); | |
986 | if (btrfs_submit_compressed_write(inode, start, /* file_offset */ | |
987 | async_extent->ram_size, /* num_bytes */ | |
988 | ins.objectid, /* disk_bytenr */ | |
989 | ins.offset, /* compressed_len */ | |
990 | async_extent->pages, /* compressed_pages */ | |
991 | async_extent->nr_pages, | |
992 | async_chunk->write_flags, | |
7c0c7269 | 993 | async_chunk->blkcg_css, true)) { |
b4ccace8 QW |
994 | const u64 start = async_extent->start; |
995 | const u64 end = start + async_extent->ram_size - 1; | |
996 | ||
997 | btrfs_writepage_endio_finish_ordered(inode, NULL, start, end, 0); | |
998 | ||
999 | extent_clear_unlock_delalloc(inode, start, end, NULL, 0, | |
1000 | PAGE_END_WRITEBACK | PAGE_SET_ERROR); | |
1001 | free_async_extent_pages(async_extent); | |
771ed689 | 1002 | } |
b4ccace8 QW |
1003 | *alloc_hint = ins.objectid + ins.offset; |
1004 | kfree(async_extent); | |
1005 | return ret; | |
1006 | ||
3e04e7f1 | 1007 | out_free_reserve: |
0b246afa | 1008 | btrfs_dec_block_group_reservations(fs_info, ins.objectid); |
2ff7e61e | 1009 | btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1); |
79787eaa | 1010 | out_free: |
b4ccace8 | 1011 | extent_clear_unlock_delalloc(inode, start, end, |
c2790a2e | 1012 | NULL, EXTENT_LOCKED | EXTENT_DELALLOC | |
a7e3b975 | 1013 | EXTENT_DELALLOC_NEW | |
151a41bc | 1014 | EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING, |
6869b0a8 QW |
1015 | PAGE_UNLOCK | PAGE_START_WRITEBACK | |
1016 | PAGE_END_WRITEBACK | PAGE_SET_ERROR); | |
40ae837b | 1017 | free_async_extent_pages(async_extent); |
79787eaa | 1018 | kfree(async_extent); |
b4ccace8 QW |
1019 | return ret; |
1020 | } | |
1021 | ||
1022 | /* | |
1023 | * Phase two of compressed writeback. This is the ordered portion of the code, | |
1024 | * which only gets called in the order the work was queued. We walk all the | |
1025 | * async extents created by compress_file_range and send them down to the disk. | |
1026 | */ | |
1027 | static noinline void submit_compressed_extents(struct async_chunk *async_chunk) | |
1028 | { | |
1029 | struct btrfs_inode *inode = BTRFS_I(async_chunk->inode); | |
1030 | struct btrfs_fs_info *fs_info = inode->root->fs_info; | |
1031 | struct async_extent *async_extent; | |
1032 | u64 alloc_hint = 0; | |
1033 | int ret = 0; | |
1034 | ||
1035 | while (!list_empty(&async_chunk->extents)) { | |
1036 | u64 extent_start; | |
1037 | u64 ram_size; | |
1038 | ||
1039 | async_extent = list_entry(async_chunk->extents.next, | |
1040 | struct async_extent, list); | |
1041 | list_del(&async_extent->list); | |
1042 | extent_start = async_extent->start; | |
1043 | ram_size = async_extent->ram_size; | |
1044 | ||
1045 | ret = submit_one_async_extent(inode, async_chunk, async_extent, | |
1046 | &alloc_hint); | |
1047 | btrfs_debug(fs_info, | |
1048 | "async extent submission failed root=%lld inode=%llu start=%llu len=%llu ret=%d", | |
1049 | inode->root->root_key.objectid, | |
1050 | btrfs_ino(inode), extent_start, ram_size, ret); | |
1051 | } | |
771ed689 CM |
1052 | } |
1053 | ||
43c69849 | 1054 | static u64 get_extent_allocation_hint(struct btrfs_inode *inode, u64 start, |
4b46fce2 JB |
1055 | u64 num_bytes) |
1056 | { | |
43c69849 | 1057 | struct extent_map_tree *em_tree = &inode->extent_tree; |
4b46fce2 JB |
1058 | struct extent_map *em; |
1059 | u64 alloc_hint = 0; | |
1060 | ||
1061 | read_lock(&em_tree->lock); | |
1062 | em = search_extent_mapping(em_tree, start, num_bytes); | |
1063 | if (em) { | |
1064 | /* | |
1065 | * if block start isn't an actual block number then find the | |
1066 | * first block in this inode and use that as a hint. If that | |
1067 | * block is also bogus then just don't worry about it. | |
1068 | */ | |
1069 | if (em->block_start >= EXTENT_MAP_LAST_BYTE) { | |
1070 | free_extent_map(em); | |
1071 | em = search_extent_mapping(em_tree, 0, 0); | |
1072 | if (em && em->block_start < EXTENT_MAP_LAST_BYTE) | |
1073 | alloc_hint = em->block_start; | |
1074 | if (em) | |
1075 | free_extent_map(em); | |
1076 | } else { | |
1077 | alloc_hint = em->block_start; | |
1078 | free_extent_map(em); | |
1079 | } | |
1080 | } | |
1081 | read_unlock(&em_tree->lock); | |
1082 | ||
1083 | return alloc_hint; | |
1084 | } | |
1085 | ||
771ed689 CM |
1086 | /* |
1087 | * when extent_io.c finds a delayed allocation range in the file, | |
1088 | * the call backs end up in this code. The basic idea is to | |
1089 | * allocate extents on disk for the range, and create ordered data structs | |
1090 | * in ram to track those extents. | |
1091 | * | |
1092 | * locked_page is the page that writepage had locked already. We use | |
1093 | * it to make sure we don't do extra locks or unlocks. | |
1094 | * | |
1095 | * *page_started is set to one if we unlock locked_page and do everything | |
1096 | * required to start IO on it. It may be clean and already done with | |
1097 | * IO when we return. | |
1098 | */ | |
6e26c442 | 1099 | static noinline int cow_file_range(struct btrfs_inode *inode, |
00361589 | 1100 | struct page *locked_page, |
74e9194a | 1101 | u64 start, u64 end, int *page_started, |
330a5827 | 1102 | unsigned long *nr_written, int unlock) |
771ed689 | 1103 | { |
6e26c442 NB |
1104 | struct btrfs_root *root = inode->root; |
1105 | struct btrfs_fs_info *fs_info = root->fs_info; | |
771ed689 CM |
1106 | u64 alloc_hint = 0; |
1107 | u64 num_bytes; | |
1108 | unsigned long ram_size; | |
a315e68f | 1109 | u64 cur_alloc_size = 0; |
432cd2a1 | 1110 | u64 min_alloc_size; |
0b246afa | 1111 | u64 blocksize = fs_info->sectorsize; |
771ed689 CM |
1112 | struct btrfs_key ins; |
1113 | struct extent_map *em; | |
a315e68f FM |
1114 | unsigned clear_bits; |
1115 | unsigned long page_ops; | |
1116 | bool extent_reserved = false; | |
771ed689 CM |
1117 | int ret = 0; |
1118 | ||
6e26c442 | 1119 | if (btrfs_is_free_space_inode(inode)) { |
29bce2f3 JB |
1120 | ret = -EINVAL; |
1121 | goto out_unlock; | |
02ecd2c2 | 1122 | } |
771ed689 | 1123 | |
fda2832f | 1124 | num_bytes = ALIGN(end - start + 1, blocksize); |
771ed689 | 1125 | num_bytes = max(blocksize, num_bytes); |
566b1760 | 1126 | ASSERT(num_bytes <= btrfs_super_total_bytes(fs_info->super_copy)); |
771ed689 | 1127 | |
6e26c442 | 1128 | inode_should_defrag(inode, start, end, num_bytes, SZ_64K); |
4cb5300b | 1129 | |
7367253a QW |
1130 | /* |
1131 | * Due to the page size limit, for subpage we can only trigger the | |
1132 | * writeback for the dirty sectors of page, that means data writeback | |
1133 | * is doing more writeback than what we want. | |
1134 | * | |
1135 | * This is especially unexpected for some call sites like fallocate, | |
1136 | * where we only increase i_size after everything is done. | |
1137 | * This means we can trigger inline extent even if we didn't want to. | |
1138 | * So here we skip inline extent creation completely. | |
1139 | */ | |
1140 | if (start == 0 && fs_info->sectorsize == PAGE_SIZE) { | |
8dd9872d OS |
1141 | u64 actual_end = min_t(u64, i_size_read(&inode->vfs_inode), |
1142 | end + 1); | |
1143 | ||
771ed689 | 1144 | /* lets try to make an inline extent */ |
8dd9872d | 1145 | ret = cow_file_range_inline(inode, actual_end, 0, |
d9496e8a | 1146 | BTRFS_COMPRESS_NONE, NULL, false); |
771ed689 | 1147 | if (ret == 0) { |
8b62f87b JB |
1148 | /* |
1149 | * We use DO_ACCOUNTING here because we need the | |
1150 | * delalloc_release_metadata to be run _after_ we drop | |
1151 | * our outstanding extent for clearing delalloc for this | |
1152 | * range. | |
1153 | */ | |
4750af3b QW |
1154 | extent_clear_unlock_delalloc(inode, start, end, |
1155 | locked_page, | |
c2790a2e | 1156 | EXTENT_LOCKED | EXTENT_DELALLOC | |
8b62f87b JB |
1157 | EXTENT_DELALLOC_NEW | EXTENT_DEFRAG | |
1158 | EXTENT_DO_ACCOUNTING, PAGE_UNLOCK | | |
6869b0a8 | 1159 | PAGE_START_WRITEBACK | PAGE_END_WRITEBACK); |
771ed689 | 1160 | *nr_written = *nr_written + |
09cbfeaf | 1161 | (end - start + PAGE_SIZE) / PAGE_SIZE; |
771ed689 | 1162 | *page_started = 1; |
4750af3b QW |
1163 | /* |
1164 | * locked_page is locked by the caller of | |
1165 | * writepage_delalloc(), not locked by | |
1166 | * __process_pages_contig(). | |
1167 | * | |
1168 | * We can't let __process_pages_contig() to unlock it, | |
1169 | * as it doesn't have any subpage::writers recorded. | |
1170 | * | |
1171 | * Here we manually unlock the page, since the caller | |
1172 | * can't use page_started to determine if it's an | |
1173 | * inline extent or a compressed extent. | |
1174 | */ | |
1175 | unlock_page(locked_page); | |
771ed689 | 1176 | goto out; |
79787eaa | 1177 | } else if (ret < 0) { |
79787eaa | 1178 | goto out_unlock; |
771ed689 CM |
1179 | } |
1180 | } | |
1181 | ||
6e26c442 NB |
1182 | alloc_hint = get_extent_allocation_hint(inode, start, num_bytes); |
1183 | btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0); | |
771ed689 | 1184 | |
432cd2a1 FM |
1185 | /* |
1186 | * Relocation relies on the relocated extents to have exactly the same | |
1187 | * size as the original extents. Normally writeback for relocation data | |
1188 | * extents follows a NOCOW path because relocation preallocates the | |
1189 | * extents. However, due to an operation such as scrub turning a block | |
1190 | * group to RO mode, it may fallback to COW mode, so we must make sure | |
1191 | * an extent allocated during COW has exactly the requested size and can | |
1192 | * not be split into smaller extents, otherwise relocation breaks and | |
1193 | * fails during the stage where it updates the bytenr of file extent | |
1194 | * items. | |
1195 | */ | |
37f00a6d | 1196 | if (btrfs_is_data_reloc_root(root)) |
432cd2a1 FM |
1197 | min_alloc_size = num_bytes; |
1198 | else | |
1199 | min_alloc_size = fs_info->sectorsize; | |
1200 | ||
3752d22f AJ |
1201 | while (num_bytes > 0) { |
1202 | cur_alloc_size = num_bytes; | |
18513091 | 1203 | ret = btrfs_reserve_extent(root, cur_alloc_size, cur_alloc_size, |
432cd2a1 | 1204 | min_alloc_size, 0, alloc_hint, |
e570fd27 | 1205 | &ins, 1, 1); |
00361589 | 1206 | if (ret < 0) |
79787eaa | 1207 | goto out_unlock; |
a315e68f FM |
1208 | cur_alloc_size = ins.offset; |
1209 | extent_reserved = true; | |
d397712b | 1210 | |
771ed689 | 1211 | ram_size = ins.offset; |
6e26c442 | 1212 | em = create_io_em(inode, start, ins.offset, /* len */ |
6f9994db LB |
1213 | start, /* orig_start */ |
1214 | ins.objectid, /* block_start */ | |
1215 | ins.offset, /* block_len */ | |
1216 | ins.offset, /* orig_block_len */ | |
1217 | ram_size, /* ram_bytes */ | |
1218 | BTRFS_COMPRESS_NONE, /* compress_type */ | |
1af4a0aa | 1219 | BTRFS_ORDERED_REGULAR /* type */); |
090a127a SY |
1220 | if (IS_ERR(em)) { |
1221 | ret = PTR_ERR(em); | |
ace68bac | 1222 | goto out_reserve; |
090a127a | 1223 | } |
6f9994db | 1224 | free_extent_map(em); |
e6dcd2dc | 1225 | |
cb36a9bb OS |
1226 | ret = btrfs_add_ordered_extent(inode, start, ram_size, ram_size, |
1227 | ins.objectid, cur_alloc_size, 0, | |
1228 | 1 << BTRFS_ORDERED_REGULAR, | |
1229 | BTRFS_COMPRESS_NONE); | |
ace68bac | 1230 | if (ret) |
d9f85963 | 1231 | goto out_drop_extent_cache; |
c8b97818 | 1232 | |
37f00a6d | 1233 | if (btrfs_is_data_reloc_root(root)) { |
6e26c442 | 1234 | ret = btrfs_reloc_clone_csums(inode, start, |
17d217fe | 1235 | cur_alloc_size); |
4dbd80fb QW |
1236 | /* |
1237 | * Only drop cache here, and process as normal. | |
1238 | * | |
1239 | * We must not allow extent_clear_unlock_delalloc() | |
1240 | * at out_unlock label to free meta of this ordered | |
1241 | * extent, as its meta should be freed by | |
1242 | * btrfs_finish_ordered_io(). | |
1243 | * | |
1244 | * So we must continue until @start is increased to | |
1245 | * skip current ordered extent. | |
1246 | */ | |
00361589 | 1247 | if (ret) |
6e26c442 | 1248 | btrfs_drop_extent_cache(inode, start, |
4dbd80fb | 1249 | start + ram_size - 1, 0); |
17d217fe YZ |
1250 | } |
1251 | ||
0b246afa | 1252 | btrfs_dec_block_group_reservations(fs_info, ins.objectid); |
9cfa3e34 | 1253 | |
f57ad937 QW |
1254 | /* |
1255 | * We're not doing compressed IO, don't unlock the first page | |
1256 | * (which the caller expects to stay locked), don't clear any | |
1257 | * dirty bits and don't set any writeback bits | |
8b62b72b | 1258 | * |
f57ad937 QW |
1259 | * Do set the Ordered (Private2) bit so we know this page was |
1260 | * properly setup for writepage. | |
c8b97818 | 1261 | */ |
a315e68f | 1262 | page_ops = unlock ? PAGE_UNLOCK : 0; |
f57ad937 | 1263 | page_ops |= PAGE_SET_ORDERED; |
a791e35e | 1264 | |
6e26c442 | 1265 | extent_clear_unlock_delalloc(inode, start, start + ram_size - 1, |
74e9194a | 1266 | locked_page, |
c2790a2e | 1267 | EXTENT_LOCKED | EXTENT_DELALLOC, |
a315e68f | 1268 | page_ops); |
3752d22f AJ |
1269 | if (num_bytes < cur_alloc_size) |
1270 | num_bytes = 0; | |
4dbd80fb | 1271 | else |
3752d22f | 1272 | num_bytes -= cur_alloc_size; |
c59f8951 CM |
1273 | alloc_hint = ins.objectid + ins.offset; |
1274 | start += cur_alloc_size; | |
a315e68f | 1275 | extent_reserved = false; |
4dbd80fb QW |
1276 | |
1277 | /* | |
1278 | * btrfs_reloc_clone_csums() error, since start is increased | |
1279 | * extent_clear_unlock_delalloc() at out_unlock label won't | |
1280 | * free metadata of current ordered extent, we're OK to exit. | |
1281 | */ | |
1282 | if (ret) | |
1283 | goto out_unlock; | |
b888db2b | 1284 | } |
79787eaa | 1285 | out: |
be20aa9d | 1286 | return ret; |
b7d5b0a8 | 1287 | |
d9f85963 | 1288 | out_drop_extent_cache: |
6e26c442 | 1289 | btrfs_drop_extent_cache(inode, start, start + ram_size - 1, 0); |
ace68bac | 1290 | out_reserve: |
0b246afa | 1291 | btrfs_dec_block_group_reservations(fs_info, ins.objectid); |
2ff7e61e | 1292 | btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1); |
79787eaa | 1293 | out_unlock: |
a7e3b975 FM |
1294 | clear_bits = EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DELALLOC_NEW | |
1295 | EXTENT_DEFRAG | EXTENT_CLEAR_META_RESV; | |
6869b0a8 | 1296 | page_ops = PAGE_UNLOCK | PAGE_START_WRITEBACK | PAGE_END_WRITEBACK; |
a315e68f FM |
1297 | /* |
1298 | * If we reserved an extent for our delalloc range (or a subrange) and | |
1299 | * failed to create the respective ordered extent, then it means that | |
1300 | * when we reserved the extent we decremented the extent's size from | |
1301 | * the data space_info's bytes_may_use counter and incremented the | |
1302 | * space_info's bytes_reserved counter by the same amount. We must make | |
1303 | * sure extent_clear_unlock_delalloc() does not try to decrement again | |
1304 | * the data space_info's bytes_may_use counter, therefore we do not pass | |
1305 | * it the flag EXTENT_CLEAR_DATA_RESV. | |
1306 | */ | |
1307 | if (extent_reserved) { | |
6e26c442 | 1308 | extent_clear_unlock_delalloc(inode, start, |
e2c8e92d | 1309 | start + cur_alloc_size - 1, |
a315e68f FM |
1310 | locked_page, |
1311 | clear_bits, | |
1312 | page_ops); | |
1313 | start += cur_alloc_size; | |
1314 | if (start >= end) | |
1315 | goto out; | |
1316 | } | |
6e26c442 | 1317 | extent_clear_unlock_delalloc(inode, start, end, locked_page, |
a315e68f FM |
1318 | clear_bits | EXTENT_CLEAR_DATA_RESV, |
1319 | page_ops); | |
79787eaa | 1320 | goto out; |
771ed689 | 1321 | } |
c8b97818 | 1322 | |
771ed689 CM |
1323 | /* |
1324 | * work queue call back to started compression on a file and pages | |
1325 | */ | |
1326 | static noinline void async_cow_start(struct btrfs_work *work) | |
1327 | { | |
b5326271 | 1328 | struct async_chunk *async_chunk; |
ac3e9933 | 1329 | int compressed_extents; |
771ed689 | 1330 | |
b5326271 | 1331 | async_chunk = container_of(work, struct async_chunk, work); |
771ed689 | 1332 | |
ac3e9933 NB |
1333 | compressed_extents = compress_file_range(async_chunk); |
1334 | if (compressed_extents == 0) { | |
b5326271 NB |
1335 | btrfs_add_delayed_iput(async_chunk->inode); |
1336 | async_chunk->inode = NULL; | |
8180ef88 | 1337 | } |
771ed689 CM |
1338 | } |
1339 | ||
1340 | /* | |
1341 | * work queue call back to submit previously compressed pages | |
1342 | */ | |
1343 | static noinline void async_cow_submit(struct btrfs_work *work) | |
1344 | { | |
c5a68aec NB |
1345 | struct async_chunk *async_chunk = container_of(work, struct async_chunk, |
1346 | work); | |
1347 | struct btrfs_fs_info *fs_info = btrfs_work_owner(work); | |
771ed689 CM |
1348 | unsigned long nr_pages; |
1349 | ||
b5326271 | 1350 | nr_pages = (async_chunk->end - async_chunk->start + PAGE_SIZE) >> |
09cbfeaf | 1351 | PAGE_SHIFT; |
771ed689 | 1352 | |
4546d178 | 1353 | /* |
b5326271 | 1354 | * ->inode could be NULL if async_chunk_start has failed to compress, |
4546d178 NB |
1355 | * in which case we don't have anything to submit, yet we need to |
1356 | * always adjust ->async_delalloc_pages as its paired with the init | |
1357 | * happening in cow_file_range_async | |
1358 | */ | |
b5326271 NB |
1359 | if (async_chunk->inode) |
1360 | submit_compressed_extents(async_chunk); | |
ac98141d JB |
1361 | |
1362 | /* atomic_sub_return implies a barrier */ | |
1363 | if (atomic_sub_return(nr_pages, &fs_info->async_delalloc_pages) < | |
1364 | 5 * SZ_1M) | |
1365 | cond_wake_up_nomb(&fs_info->async_submit_wait); | |
771ed689 | 1366 | } |
c8b97818 | 1367 | |
771ed689 CM |
1368 | static noinline void async_cow_free(struct btrfs_work *work) |
1369 | { | |
b5326271 | 1370 | struct async_chunk *async_chunk; |
9e895a8f | 1371 | struct async_cow *async_cow; |
97db1204 | 1372 | |
b5326271 NB |
1373 | async_chunk = container_of(work, struct async_chunk, work); |
1374 | if (async_chunk->inode) | |
1375 | btrfs_add_delayed_iput(async_chunk->inode); | |
ec39f769 CM |
1376 | if (async_chunk->blkcg_css) |
1377 | css_put(async_chunk->blkcg_css); | |
9e895a8f QW |
1378 | |
1379 | async_cow = async_chunk->async_cow; | |
1380 | if (atomic_dec_and_test(&async_cow->num_chunks)) | |
1381 | kvfree(async_cow); | |
771ed689 CM |
1382 | } |
1383 | ||
751b6431 | 1384 | static int cow_file_range_async(struct btrfs_inode *inode, |
ec39f769 CM |
1385 | struct writeback_control *wbc, |
1386 | struct page *locked_page, | |
771ed689 | 1387 | u64 start, u64 end, int *page_started, |
fac07d2b | 1388 | unsigned long *nr_written) |
771ed689 | 1389 | { |
751b6431 | 1390 | struct btrfs_fs_info *fs_info = inode->root->fs_info; |
ec39f769 | 1391 | struct cgroup_subsys_state *blkcg_css = wbc_blkcg_css(wbc); |
97db1204 NB |
1392 | struct async_cow *ctx; |
1393 | struct async_chunk *async_chunk; | |
771ed689 CM |
1394 | unsigned long nr_pages; |
1395 | u64 cur_end; | |
97db1204 NB |
1396 | u64 num_chunks = DIV_ROUND_UP(end - start, SZ_512K); |
1397 | int i; | |
1398 | bool should_compress; | |
b1c16ac9 | 1399 | unsigned nofs_flag; |
fac07d2b | 1400 | const unsigned int write_flags = wbc_to_write_flags(wbc); |
771ed689 | 1401 | |
751b6431 | 1402 | unlock_extent(&inode->io_tree, start, end); |
97db1204 | 1403 | |
751b6431 | 1404 | if (inode->flags & BTRFS_INODE_NOCOMPRESS && |
97db1204 NB |
1405 | !btrfs_test_opt(fs_info, FORCE_COMPRESS)) { |
1406 | num_chunks = 1; | |
1407 | should_compress = false; | |
1408 | } else { | |
1409 | should_compress = true; | |
1410 | } | |
1411 | ||
b1c16ac9 NB |
1412 | nofs_flag = memalloc_nofs_save(); |
1413 | ctx = kvmalloc(struct_size(ctx, chunks, num_chunks), GFP_KERNEL); | |
1414 | memalloc_nofs_restore(nofs_flag); | |
1415 | ||
97db1204 NB |
1416 | if (!ctx) { |
1417 | unsigned clear_bits = EXTENT_LOCKED | EXTENT_DELALLOC | | |
1418 | EXTENT_DELALLOC_NEW | EXTENT_DEFRAG | | |
1419 | EXTENT_DO_ACCOUNTING; | |
6869b0a8 QW |
1420 | unsigned long page_ops = PAGE_UNLOCK | PAGE_START_WRITEBACK | |
1421 | PAGE_END_WRITEBACK | PAGE_SET_ERROR; | |
97db1204 | 1422 | |
751b6431 NB |
1423 | extent_clear_unlock_delalloc(inode, start, end, locked_page, |
1424 | clear_bits, page_ops); | |
97db1204 NB |
1425 | return -ENOMEM; |
1426 | } | |
1427 | ||
1428 | async_chunk = ctx->chunks; | |
1429 | atomic_set(&ctx->num_chunks, num_chunks); | |
1430 | ||
1431 | for (i = 0; i < num_chunks; i++) { | |
1432 | if (should_compress) | |
1433 | cur_end = min(end, start + SZ_512K - 1); | |
1434 | else | |
1435 | cur_end = end; | |
771ed689 | 1436 | |
bd4691a0 NB |
1437 | /* |
1438 | * igrab is called higher up in the call chain, take only the | |
1439 | * lightweight reference for the callback lifetime | |
1440 | */ | |
751b6431 | 1441 | ihold(&inode->vfs_inode); |
9e895a8f | 1442 | async_chunk[i].async_cow = ctx; |
751b6431 | 1443 | async_chunk[i].inode = &inode->vfs_inode; |
97db1204 NB |
1444 | async_chunk[i].start = start; |
1445 | async_chunk[i].end = cur_end; | |
97db1204 NB |
1446 | async_chunk[i].write_flags = write_flags; |
1447 | INIT_LIST_HEAD(&async_chunk[i].extents); | |
1448 | ||
1d53c9e6 CM |
1449 | /* |
1450 | * The locked_page comes all the way from writepage and its | |
1451 | * the original page we were actually given. As we spread | |
1452 | * this large delalloc region across multiple async_chunk | |
1453 | * structs, only the first struct needs a pointer to locked_page | |
1454 | * | |
1455 | * This way we don't need racey decisions about who is supposed | |
1456 | * to unlock it. | |
1457 | */ | |
1458 | if (locked_page) { | |
ec39f769 CM |
1459 | /* |
1460 | * Depending on the compressibility, the pages might or | |
1461 | * might not go through async. We want all of them to | |
1462 | * be accounted against wbc once. Let's do it here | |
1463 | * before the paths diverge. wbc accounting is used | |
1464 | * only for foreign writeback detection and doesn't | |
1465 | * need full accuracy. Just account the whole thing | |
1466 | * against the first page. | |
1467 | */ | |
1468 | wbc_account_cgroup_owner(wbc, locked_page, | |
1469 | cur_end - start); | |
1d53c9e6 CM |
1470 | async_chunk[i].locked_page = locked_page; |
1471 | locked_page = NULL; | |
1472 | } else { | |
1473 | async_chunk[i].locked_page = NULL; | |
1474 | } | |
1475 | ||
ec39f769 CM |
1476 | if (blkcg_css != blkcg_root_css) { |
1477 | css_get(blkcg_css); | |
1478 | async_chunk[i].blkcg_css = blkcg_css; | |
1479 | } else { | |
1480 | async_chunk[i].blkcg_css = NULL; | |
1481 | } | |
1482 | ||
a0cac0ec OS |
1483 | btrfs_init_work(&async_chunk[i].work, async_cow_start, |
1484 | async_cow_submit, async_cow_free); | |
771ed689 | 1485 | |
97db1204 | 1486 | nr_pages = DIV_ROUND_UP(cur_end - start, PAGE_SIZE); |
0b246afa | 1487 | atomic_add(nr_pages, &fs_info->async_delalloc_pages); |
771ed689 | 1488 | |
97db1204 | 1489 | btrfs_queue_work(fs_info->delalloc_workers, &async_chunk[i].work); |
771ed689 | 1490 | |
771ed689 CM |
1491 | *nr_written += nr_pages; |
1492 | start = cur_end + 1; | |
1493 | } | |
1494 | *page_started = 1; | |
1495 | return 0; | |
be20aa9d CM |
1496 | } |
1497 | ||
42c01100 NA |
1498 | static noinline int run_delalloc_zoned(struct btrfs_inode *inode, |
1499 | struct page *locked_page, u64 start, | |
1500 | u64 end, int *page_started, | |
1501 | unsigned long *nr_written) | |
1502 | { | |
1503 | int ret; | |
1504 | ||
1505 | ret = cow_file_range(inode, locked_page, start, end, page_started, | |
1506 | nr_written, 0); | |
1507 | if (ret) | |
1508 | return ret; | |
1509 | ||
1510 | if (*page_started) | |
1511 | return 0; | |
1512 | ||
1513 | __set_page_dirty_nobuffers(locked_page); | |
1514 | account_page_redirty(locked_page); | |
2bd0fc93 | 1515 | extent_write_locked_range(&inode->vfs_inode, start, end); |
42c01100 NA |
1516 | *page_started = 1; |
1517 | ||
1518 | return 0; | |
1519 | } | |
1520 | ||
2ff7e61e | 1521 | static noinline int csum_exist_in_range(struct btrfs_fs_info *fs_info, |
17d217fe YZ |
1522 | u64 bytenr, u64 num_bytes) |
1523 | { | |
fc28b25e | 1524 | struct btrfs_root *csum_root = btrfs_csum_root(fs_info, bytenr); |
17d217fe | 1525 | struct btrfs_ordered_sum *sums; |
fc28b25e | 1526 | int ret; |
17d217fe YZ |
1527 | LIST_HEAD(list); |
1528 | ||
fc28b25e | 1529 | ret = btrfs_lookup_csums_range(csum_root, bytenr, |
a2de733c | 1530 | bytenr + num_bytes - 1, &list, 0); |
17d217fe YZ |
1531 | if (ret == 0 && list_empty(&list)) |
1532 | return 0; | |
1533 | ||
1534 | while (!list_empty(&list)) { | |
1535 | sums = list_entry(list.next, struct btrfs_ordered_sum, list); | |
1536 | list_del(&sums->list); | |
1537 | kfree(sums); | |
1538 | } | |
58113753 LB |
1539 | if (ret < 0) |
1540 | return ret; | |
17d217fe YZ |
1541 | return 1; |
1542 | } | |
1543 | ||
8ba96f3d | 1544 | static int fallback_to_cow(struct btrfs_inode *inode, struct page *locked_page, |
467dc47e FM |
1545 | const u64 start, const u64 end, |
1546 | int *page_started, unsigned long *nr_written) | |
1547 | { | |
8ba96f3d | 1548 | const bool is_space_ino = btrfs_is_free_space_inode(inode); |
37f00a6d | 1549 | const bool is_reloc_ino = btrfs_is_data_reloc_root(inode->root); |
2166e5ed | 1550 | const u64 range_bytes = end + 1 - start; |
8ba96f3d | 1551 | struct extent_io_tree *io_tree = &inode->io_tree; |
467dc47e FM |
1552 | u64 range_start = start; |
1553 | u64 count; | |
1554 | ||
1555 | /* | |
1556 | * If EXTENT_NORESERVE is set it means that when the buffered write was | |
1557 | * made we had not enough available data space and therefore we did not | |
1558 | * reserve data space for it, since we though we could do NOCOW for the | |
1559 | * respective file range (either there is prealloc extent or the inode | |
1560 | * has the NOCOW bit set). | |
1561 | * | |
1562 | * However when we need to fallback to COW mode (because for example the | |
1563 | * block group for the corresponding extent was turned to RO mode by a | |
1564 | * scrub or relocation) we need to do the following: | |
1565 | * | |
1566 | * 1) We increment the bytes_may_use counter of the data space info. | |
1567 | * If COW succeeds, it allocates a new data extent and after doing | |
1568 | * that it decrements the space info's bytes_may_use counter and | |
1569 | * increments its bytes_reserved counter by the same amount (we do | |
1570 | * this at btrfs_add_reserved_bytes()). So we need to increment the | |
1571 | * bytes_may_use counter to compensate (when space is reserved at | |
1572 | * buffered write time, the bytes_may_use counter is incremented); | |
1573 | * | |
1574 | * 2) We clear the EXTENT_NORESERVE bit from the range. We do this so | |
1575 | * that if the COW path fails for any reason, it decrements (through | |
1576 | * extent_clear_unlock_delalloc()) the bytes_may_use counter of the | |
1577 | * data space info, which we incremented in the step above. | |
2166e5ed FM |
1578 | * |
1579 | * If we need to fallback to cow and the inode corresponds to a free | |
6bd335b4 FM |
1580 | * space cache inode or an inode of the data relocation tree, we must |
1581 | * also increment bytes_may_use of the data space_info for the same | |
1582 | * reason. Space caches and relocated data extents always get a prealloc | |
2166e5ed | 1583 | * extent for them, however scrub or balance may have set the block |
6bd335b4 FM |
1584 | * group that contains that extent to RO mode and therefore force COW |
1585 | * when starting writeback. | |
467dc47e | 1586 | */ |
2166e5ed | 1587 | count = count_range_bits(io_tree, &range_start, end, range_bytes, |
467dc47e | 1588 | EXTENT_NORESERVE, 0); |
6bd335b4 FM |
1589 | if (count > 0 || is_space_ino || is_reloc_ino) { |
1590 | u64 bytes = count; | |
8ba96f3d | 1591 | struct btrfs_fs_info *fs_info = inode->root->fs_info; |
467dc47e FM |
1592 | struct btrfs_space_info *sinfo = fs_info->data_sinfo; |
1593 | ||
6bd335b4 FM |
1594 | if (is_space_ino || is_reloc_ino) |
1595 | bytes = range_bytes; | |
1596 | ||
467dc47e | 1597 | spin_lock(&sinfo->lock); |
2166e5ed | 1598 | btrfs_space_info_update_bytes_may_use(fs_info, sinfo, bytes); |
467dc47e FM |
1599 | spin_unlock(&sinfo->lock); |
1600 | ||
2166e5ed FM |
1601 | if (count > 0) |
1602 | clear_extent_bit(io_tree, start, end, EXTENT_NORESERVE, | |
1603 | 0, 0, NULL); | |
467dc47e FM |
1604 | } |
1605 | ||
8ba96f3d NB |
1606 | return cow_file_range(inode, locked_page, start, end, page_started, |
1607 | nr_written, 1); | |
467dc47e FM |
1608 | } |
1609 | ||
d352ac68 CM |
1610 | /* |
1611 | * when nowcow writeback call back. This checks for snapshots or COW copies | |
1612 | * of the extents that exist in the file, and COWs the file as required. | |
1613 | * | |
1614 | * If no cow copies or snapshots exist, we write directly to the existing | |
1615 | * blocks on disk | |
1616 | */ | |
968322c8 | 1617 | static noinline int run_delalloc_nocow(struct btrfs_inode *inode, |
7f366cfe | 1618 | struct page *locked_page, |
3e024846 | 1619 | const u64 start, const u64 end, |
6e65ae76 | 1620 | int *page_started, |
3e024846 | 1621 | unsigned long *nr_written) |
be20aa9d | 1622 | { |
968322c8 NB |
1623 | struct btrfs_fs_info *fs_info = inode->root->fs_info; |
1624 | struct btrfs_root *root = inode->root; | |
be20aa9d | 1625 | struct btrfs_path *path; |
3e024846 NB |
1626 | u64 cow_start = (u64)-1; |
1627 | u64 cur_offset = start; | |
8ecebf4d | 1628 | int ret; |
3e024846 | 1629 | bool check_prev = true; |
968322c8 NB |
1630 | const bool freespace_inode = btrfs_is_free_space_inode(inode); |
1631 | u64 ino = btrfs_ino(inode); | |
762bf098 NB |
1632 | bool nocow = false; |
1633 | u64 disk_bytenr = 0; | |
6e65ae76 | 1634 | const bool force = inode->flags & BTRFS_INODE_NODATACOW; |
be20aa9d CM |
1635 | |
1636 | path = btrfs_alloc_path(); | |
17ca04af | 1637 | if (!path) { |
968322c8 | 1638 | extent_clear_unlock_delalloc(inode, start, end, locked_page, |
c2790a2e | 1639 | EXTENT_LOCKED | EXTENT_DELALLOC | |
151a41bc JB |
1640 | EXTENT_DO_ACCOUNTING | |
1641 | EXTENT_DEFRAG, PAGE_UNLOCK | | |
6869b0a8 | 1642 | PAGE_START_WRITEBACK | |
c2790a2e | 1643 | PAGE_END_WRITEBACK); |
d8926bb3 | 1644 | return -ENOMEM; |
17ca04af | 1645 | } |
82d5902d | 1646 | |
80ff3856 | 1647 | while (1) { |
3e024846 NB |
1648 | struct btrfs_key found_key; |
1649 | struct btrfs_file_extent_item *fi; | |
1650 | struct extent_buffer *leaf; | |
1651 | u64 extent_end; | |
1652 | u64 extent_offset; | |
3e024846 NB |
1653 | u64 num_bytes = 0; |
1654 | u64 disk_num_bytes; | |
3e024846 NB |
1655 | u64 ram_bytes; |
1656 | int extent_type; | |
762bf098 NB |
1657 | |
1658 | nocow = false; | |
3e024846 | 1659 | |
e4c3b2dc | 1660 | ret = btrfs_lookup_file_extent(NULL, root, path, ino, |
80ff3856 | 1661 | cur_offset, 0); |
d788a349 | 1662 | if (ret < 0) |
79787eaa | 1663 | goto error; |
a6bd9cd1 NB |
1664 | |
1665 | /* | |
1666 | * If there is no extent for our range when doing the initial | |
1667 | * search, then go back to the previous slot as it will be the | |
1668 | * one containing the search offset | |
1669 | */ | |
80ff3856 YZ |
1670 | if (ret > 0 && path->slots[0] > 0 && check_prev) { |
1671 | leaf = path->nodes[0]; | |
1672 | btrfs_item_key_to_cpu(leaf, &found_key, | |
1673 | path->slots[0] - 1); | |
33345d01 | 1674 | if (found_key.objectid == ino && |
80ff3856 YZ |
1675 | found_key.type == BTRFS_EXTENT_DATA_KEY) |
1676 | path->slots[0]--; | |
1677 | } | |
3e024846 | 1678 | check_prev = false; |
80ff3856 | 1679 | next_slot: |
a6bd9cd1 | 1680 | /* Go to next leaf if we have exhausted the current one */ |
80ff3856 YZ |
1681 | leaf = path->nodes[0]; |
1682 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
1683 | ret = btrfs_next_leaf(root, path); | |
e8916699 LB |
1684 | if (ret < 0) { |
1685 | if (cow_start != (u64)-1) | |
1686 | cur_offset = cow_start; | |
79787eaa | 1687 | goto error; |
e8916699 | 1688 | } |
80ff3856 YZ |
1689 | if (ret > 0) |
1690 | break; | |
1691 | leaf = path->nodes[0]; | |
1692 | } | |
be20aa9d | 1693 | |
80ff3856 YZ |
1694 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
1695 | ||
a6bd9cd1 | 1696 | /* Didn't find anything for our INO */ |
1d512cb7 FM |
1697 | if (found_key.objectid > ino) |
1698 | break; | |
a6bd9cd1 NB |
1699 | /* |
1700 | * Keep searching until we find an EXTENT_ITEM or there are no | |
1701 | * more extents for this inode | |
1702 | */ | |
1d512cb7 FM |
1703 | if (WARN_ON_ONCE(found_key.objectid < ino) || |
1704 | found_key.type < BTRFS_EXTENT_DATA_KEY) { | |
1705 | path->slots[0]++; | |
1706 | goto next_slot; | |
1707 | } | |
a6bd9cd1 NB |
1708 | |
1709 | /* Found key is not EXTENT_DATA_KEY or starts after req range */ | |
1d512cb7 | 1710 | if (found_key.type > BTRFS_EXTENT_DATA_KEY || |
80ff3856 YZ |
1711 | found_key.offset > end) |
1712 | break; | |
1713 | ||
a6bd9cd1 NB |
1714 | /* |
1715 | * If the found extent starts after requested offset, then | |
1716 | * adjust extent_end to be right before this extent begins | |
1717 | */ | |
80ff3856 YZ |
1718 | if (found_key.offset > cur_offset) { |
1719 | extent_end = found_key.offset; | |
e9061e21 | 1720 | extent_type = 0; |
80ff3856 YZ |
1721 | goto out_check; |
1722 | } | |
1723 | ||
a6bd9cd1 NB |
1724 | /* |
1725 | * Found extent which begins before our range and potentially | |
1726 | * intersect it | |
1727 | */ | |
80ff3856 YZ |
1728 | fi = btrfs_item_ptr(leaf, path->slots[0], |
1729 | struct btrfs_file_extent_item); | |
1730 | extent_type = btrfs_file_extent_type(leaf, fi); | |
1731 | ||
cc95bef6 | 1732 | ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi); |
d899e052 YZ |
1733 | if (extent_type == BTRFS_FILE_EXTENT_REG || |
1734 | extent_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
80ff3856 | 1735 | disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); |
5d4f98a2 | 1736 | extent_offset = btrfs_file_extent_offset(leaf, fi); |
80ff3856 YZ |
1737 | extent_end = found_key.offset + |
1738 | btrfs_file_extent_num_bytes(leaf, fi); | |
b4939680 JB |
1739 | disk_num_bytes = |
1740 | btrfs_file_extent_disk_num_bytes(leaf, fi); | |
a6bd9cd1 | 1741 | /* |
de7999af FM |
1742 | * If the extent we got ends before our current offset, |
1743 | * skip to the next extent. | |
a6bd9cd1 | 1744 | */ |
de7999af | 1745 | if (extent_end <= cur_offset) { |
80ff3856 YZ |
1746 | path->slots[0]++; |
1747 | goto next_slot; | |
1748 | } | |
a6bd9cd1 | 1749 | /* Skip holes */ |
17d217fe YZ |
1750 | if (disk_bytenr == 0) |
1751 | goto out_check; | |
a6bd9cd1 | 1752 | /* Skip compressed/encrypted/encoded extents */ |
80ff3856 YZ |
1753 | if (btrfs_file_extent_compression(leaf, fi) || |
1754 | btrfs_file_extent_encryption(leaf, fi) || | |
1755 | btrfs_file_extent_other_encoding(leaf, fi)) | |
1756 | goto out_check; | |
78d4295b | 1757 | /* |
a6bd9cd1 NB |
1758 | * If extent is created before the last volume's snapshot |
1759 | * this implies the extent is shared, hence we can't do | |
1760 | * nocow. This is the same check as in | |
1761 | * btrfs_cross_ref_exist but without calling | |
1762 | * btrfs_search_slot. | |
78d4295b | 1763 | */ |
3e024846 | 1764 | if (!freespace_inode && |
27a7ff55 | 1765 | btrfs_file_extent_generation(leaf, fi) <= |
78d4295b EL |
1766 | btrfs_root_last_snapshot(&root->root_item)) |
1767 | goto out_check; | |
d899e052 YZ |
1768 | if (extent_type == BTRFS_FILE_EXTENT_REG && !force) |
1769 | goto out_check; | |
c65ca98f FM |
1770 | |
1771 | /* | |
1772 | * The following checks can be expensive, as they need to | |
1773 | * take other locks and do btree or rbtree searches, so | |
1774 | * release the path to avoid blocking other tasks for too | |
1775 | * long. | |
1776 | */ | |
1777 | btrfs_release_path(path); | |
1778 | ||
58113753 LB |
1779 | ret = btrfs_cross_ref_exist(root, ino, |
1780 | found_key.offset - | |
a84d5d42 | 1781 | extent_offset, disk_bytenr, false); |
58113753 LB |
1782 | if (ret) { |
1783 | /* | |
1784 | * ret could be -EIO if the above fails to read | |
1785 | * metadata. | |
1786 | */ | |
1787 | if (ret < 0) { | |
1788 | if (cow_start != (u64)-1) | |
1789 | cur_offset = cow_start; | |
1790 | goto error; | |
1791 | } | |
1792 | ||
3e024846 | 1793 | WARN_ON_ONCE(freespace_inode); |
17d217fe | 1794 | goto out_check; |
58113753 | 1795 | } |
5d4f98a2 | 1796 | disk_bytenr += extent_offset; |
17d217fe YZ |
1797 | disk_bytenr += cur_offset - found_key.offset; |
1798 | num_bytes = min(end + 1, extent_end) - cur_offset; | |
e9894fd3 | 1799 | /* |
a6bd9cd1 NB |
1800 | * If there are pending snapshots for this root, we |
1801 | * fall into common COW way | |
e9894fd3 | 1802 | */ |
3e024846 | 1803 | if (!freespace_inode && atomic_read(&root->snapshot_force_cow)) |
8ecebf4d | 1804 | goto out_check; |
17d217fe YZ |
1805 | /* |
1806 | * force cow if csum exists in the range. | |
1807 | * this ensure that csum for a given extent are | |
1808 | * either valid or do not exist. | |
1809 | */ | |
58113753 LB |
1810 | ret = csum_exist_in_range(fs_info, disk_bytenr, |
1811 | num_bytes); | |
1812 | if (ret) { | |
58113753 LB |
1813 | /* |
1814 | * ret could be -EIO if the above fails to read | |
1815 | * metadata. | |
1816 | */ | |
1817 | if (ret < 0) { | |
1818 | if (cow_start != (u64)-1) | |
1819 | cur_offset = cow_start; | |
1820 | goto error; | |
1821 | } | |
3e024846 | 1822 | WARN_ON_ONCE(freespace_inode); |
17d217fe | 1823 | goto out_check; |
91e1f56a | 1824 | } |
20903032 | 1825 | /* If the extent's block group is RO, we must COW */ |
8ecebf4d | 1826 | if (!btrfs_inc_nocow_writers(fs_info, disk_bytenr)) |
f78c436c | 1827 | goto out_check; |
3e024846 | 1828 | nocow = true; |
80ff3856 | 1829 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
e8e21007 NB |
1830 | extent_end = found_key.offset + ram_bytes; |
1831 | extent_end = ALIGN(extent_end, fs_info->sectorsize); | |
922f0518 NB |
1832 | /* Skip extents outside of our requested range */ |
1833 | if (extent_end <= start) { | |
1834 | path->slots[0]++; | |
1835 | goto next_slot; | |
1836 | } | |
80ff3856 | 1837 | } else { |
e8e21007 | 1838 | /* If this triggers then we have a memory corruption */ |
290342f6 | 1839 | BUG(); |
80ff3856 YZ |
1840 | } |
1841 | out_check: | |
a6bd9cd1 NB |
1842 | /* |
1843 | * If nocow is false then record the beginning of the range | |
1844 | * that needs to be COWed | |
1845 | */ | |
80ff3856 YZ |
1846 | if (!nocow) { |
1847 | if (cow_start == (u64)-1) | |
1848 | cow_start = cur_offset; | |
1849 | cur_offset = extent_end; | |
1850 | if (cur_offset > end) | |
1851 | break; | |
c65ca98f FM |
1852 | if (!path->nodes[0]) |
1853 | continue; | |
80ff3856 YZ |
1854 | path->slots[0]++; |
1855 | goto next_slot; | |
7ea394f1 YZ |
1856 | } |
1857 | ||
a6bd9cd1 NB |
1858 | /* |
1859 | * COW range from cow_start to found_key.offset - 1. As the key | |
1860 | * will contain the beginning of the first extent that can be | |
1861 | * NOCOW, following one which needs to be COW'ed | |
1862 | */ | |
80ff3856 | 1863 | if (cow_start != (u64)-1) { |
968322c8 | 1864 | ret = fallback_to_cow(inode, locked_page, |
8ba96f3d | 1865 | cow_start, found_key.offset - 1, |
467dc47e | 1866 | page_started, nr_written); |
230ed397 | 1867 | if (ret) |
79787eaa | 1868 | goto error; |
80ff3856 | 1869 | cow_start = (u64)-1; |
7ea394f1 | 1870 | } |
80ff3856 | 1871 | |
d899e052 | 1872 | if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) { |
6f9994db | 1873 | u64 orig_start = found_key.offset - extent_offset; |
3e024846 | 1874 | struct extent_map *em; |
6f9994db | 1875 | |
968322c8 | 1876 | em = create_io_em(inode, cur_offset, num_bytes, |
6f9994db LB |
1877 | orig_start, |
1878 | disk_bytenr, /* block_start */ | |
1879 | num_bytes, /* block_len */ | |
1880 | disk_num_bytes, /* orig_block_len */ | |
1881 | ram_bytes, BTRFS_COMPRESS_NONE, | |
1882 | BTRFS_ORDERED_PREALLOC); | |
1883 | if (IS_ERR(em)) { | |
6f9994db LB |
1884 | ret = PTR_ERR(em); |
1885 | goto error; | |
d899e052 | 1886 | } |
6f9994db | 1887 | free_extent_map(em); |
cb36a9bb OS |
1888 | ret = btrfs_add_ordered_extent(inode, |
1889 | cur_offset, num_bytes, num_bytes, | |
1890 | disk_bytenr, num_bytes, 0, | |
1891 | 1 << BTRFS_ORDERED_PREALLOC, | |
1892 | BTRFS_COMPRESS_NONE); | |
762bf098 | 1893 | if (ret) { |
968322c8 | 1894 | btrfs_drop_extent_cache(inode, cur_offset, |
762bf098 NB |
1895 | cur_offset + num_bytes - 1, |
1896 | 0); | |
1897 | goto error; | |
1898 | } | |
d899e052 | 1899 | } else { |
968322c8 | 1900 | ret = btrfs_add_ordered_extent(inode, cur_offset, |
cb36a9bb | 1901 | num_bytes, num_bytes, |
bb55f626 | 1902 | disk_bytenr, num_bytes, |
cb36a9bb OS |
1903 | 0, |
1904 | 1 << BTRFS_ORDERED_NOCOW, | |
1905 | BTRFS_COMPRESS_NONE); | |
762bf098 NB |
1906 | if (ret) |
1907 | goto error; | |
d899e052 | 1908 | } |
80ff3856 | 1909 | |
f78c436c | 1910 | if (nocow) |
0b246afa | 1911 | btrfs_dec_nocow_writers(fs_info, disk_bytenr); |
762bf098 | 1912 | nocow = false; |
771ed689 | 1913 | |
37f00a6d | 1914 | if (btrfs_is_data_reloc_root(root)) |
4dbd80fb QW |
1915 | /* |
1916 | * Error handled later, as we must prevent | |
1917 | * extent_clear_unlock_delalloc() in error handler | |
1918 | * from freeing metadata of created ordered extent. | |
1919 | */ | |
968322c8 | 1920 | ret = btrfs_reloc_clone_csums(inode, cur_offset, |
efa56464 | 1921 | num_bytes); |
efa56464 | 1922 | |
968322c8 | 1923 | extent_clear_unlock_delalloc(inode, cur_offset, |
74e9194a | 1924 | cur_offset + num_bytes - 1, |
c2790a2e | 1925 | locked_page, EXTENT_LOCKED | |
18513091 WX |
1926 | EXTENT_DELALLOC | |
1927 | EXTENT_CLEAR_DATA_RESV, | |
f57ad937 | 1928 | PAGE_UNLOCK | PAGE_SET_ORDERED); |
18513091 | 1929 | |
80ff3856 | 1930 | cur_offset = extent_end; |
4dbd80fb QW |
1931 | |
1932 | /* | |
1933 | * btrfs_reloc_clone_csums() error, now we're OK to call error | |
1934 | * handler, as metadata for created ordered extent will only | |
1935 | * be freed by btrfs_finish_ordered_io(). | |
1936 | */ | |
1937 | if (ret) | |
1938 | goto error; | |
80ff3856 YZ |
1939 | if (cur_offset > end) |
1940 | break; | |
be20aa9d | 1941 | } |
b3b4aa74 | 1942 | btrfs_release_path(path); |
80ff3856 | 1943 | |
506481b2 | 1944 | if (cur_offset <= end && cow_start == (u64)-1) |
80ff3856 | 1945 | cow_start = cur_offset; |
17ca04af | 1946 | |
80ff3856 | 1947 | if (cow_start != (u64)-1) { |
506481b2 | 1948 | cur_offset = end; |
968322c8 NB |
1949 | ret = fallback_to_cow(inode, locked_page, cow_start, end, |
1950 | page_started, nr_written); | |
d788a349 | 1951 | if (ret) |
79787eaa | 1952 | goto error; |
80ff3856 YZ |
1953 | } |
1954 | ||
79787eaa | 1955 | error: |
762bf098 NB |
1956 | if (nocow) |
1957 | btrfs_dec_nocow_writers(fs_info, disk_bytenr); | |
1958 | ||
17ca04af | 1959 | if (ret && cur_offset < end) |
968322c8 | 1960 | extent_clear_unlock_delalloc(inode, cur_offset, end, |
c2790a2e | 1961 | locked_page, EXTENT_LOCKED | |
151a41bc JB |
1962 | EXTENT_DELALLOC | EXTENT_DEFRAG | |
1963 | EXTENT_DO_ACCOUNTING, PAGE_UNLOCK | | |
6869b0a8 | 1964 | PAGE_START_WRITEBACK | |
c2790a2e | 1965 | PAGE_END_WRITEBACK); |
7ea394f1 | 1966 | btrfs_free_path(path); |
79787eaa | 1967 | return ret; |
be20aa9d CM |
1968 | } |
1969 | ||
6e65ae76 | 1970 | static bool should_nocow(struct btrfs_inode *inode, u64 start, u64 end) |
47059d93 | 1971 | { |
6e65ae76 GR |
1972 | if (inode->flags & (BTRFS_INODE_NODATACOW | BTRFS_INODE_PREALLOC)) { |
1973 | if (inode->defrag_bytes && | |
1974 | test_range_bit(&inode->io_tree, start, end, EXTENT_DEFRAG, | |
1975 | 0, NULL)) | |
1976 | return false; | |
1977 | return true; | |
1978 | } | |
1979 | return false; | |
47059d93 WS |
1980 | } |
1981 | ||
d352ac68 | 1982 | /* |
5eaad97a NB |
1983 | * Function to process delayed allocation (create CoW) for ranges which are |
1984 | * being touched for the first time. | |
d352ac68 | 1985 | */ |
98456b9c | 1986 | int btrfs_run_delalloc_range(struct btrfs_inode *inode, struct page *locked_page, |
5eaad97a NB |
1987 | u64 start, u64 end, int *page_started, unsigned long *nr_written, |
1988 | struct writeback_control *wbc) | |
be20aa9d | 1989 | { |
be20aa9d | 1990 | int ret; |
42c01100 | 1991 | const bool zoned = btrfs_is_zoned(inode->root->fs_info); |
a2135011 | 1992 | |
2749f7ef QW |
1993 | /* |
1994 | * The range must cover part of the @locked_page, or the returned | |
1995 | * @page_started can confuse the caller. | |
1996 | */ | |
1997 | ASSERT(!(end <= page_offset(locked_page) || | |
1998 | start >= page_offset(locked_page) + PAGE_SIZE)); | |
1999 | ||
6e65ae76 | 2000 | if (should_nocow(inode, start, end)) { |
2adada88 JT |
2001 | /* |
2002 | * Normally on a zoned device we're only doing COW writes, but | |
2003 | * in case of relocation on a zoned filesystem we have taken | |
2004 | * precaution, that we're only writing sequentially. It's safe | |
2005 | * to use run_delalloc_nocow() here, like for regular | |
2006 | * preallocated inodes. | |
2007 | */ | |
9435be73 | 2008 | ASSERT(!zoned || btrfs_is_data_reloc_root(inode->root)); |
98456b9c | 2009 | ret = run_delalloc_nocow(inode, locked_page, start, end, |
6e65ae76 | 2010 | page_started, nr_written); |
e6f9d696 | 2011 | } else if (!btrfs_inode_can_compress(inode) || |
98456b9c | 2012 | !inode_need_compress(inode, start, end)) { |
42c01100 NA |
2013 | if (zoned) |
2014 | ret = run_delalloc_zoned(inode, locked_page, start, end, | |
2015 | page_started, nr_written); | |
2016 | else | |
2017 | ret = cow_file_range(inode, locked_page, start, end, | |
2018 | page_started, nr_written, 1); | |
7ddf5a42 | 2019 | } else { |
98456b9c NB |
2020 | set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT, &inode->runtime_flags); |
2021 | ret = cow_file_range_async(inode, wbc, locked_page, start, end, | |
fac07d2b | 2022 | page_started, nr_written); |
7ddf5a42 | 2023 | } |
7361b4ae | 2024 | ASSERT(ret <= 0); |
52427260 | 2025 | if (ret) |
98456b9c | 2026 | btrfs_cleanup_ordered_extents(inode, locked_page, start, |
d1051d6e | 2027 | end - start + 1); |
b888db2b CM |
2028 | return ret; |
2029 | } | |
2030 | ||
abbb55f4 NB |
2031 | void btrfs_split_delalloc_extent(struct inode *inode, |
2032 | struct extent_state *orig, u64 split) | |
9ed74f2d | 2033 | { |
dcab6a3b JB |
2034 | u64 size; |
2035 | ||
0ca1f7ce | 2036 | /* not delalloc, ignore it */ |
9ed74f2d | 2037 | if (!(orig->state & EXTENT_DELALLOC)) |
1bf85046 | 2038 | return; |
9ed74f2d | 2039 | |
dcab6a3b JB |
2040 | size = orig->end - orig->start + 1; |
2041 | if (size > BTRFS_MAX_EXTENT_SIZE) { | |
823bb20a | 2042 | u32 num_extents; |
dcab6a3b JB |
2043 | u64 new_size; |
2044 | ||
2045 | /* | |
5c848198 | 2046 | * See the explanation in btrfs_merge_delalloc_extent, the same |
ba117213 | 2047 | * applies here, just in reverse. |
dcab6a3b JB |
2048 | */ |
2049 | new_size = orig->end - split + 1; | |
823bb20a | 2050 | num_extents = count_max_extents(new_size); |
ba117213 | 2051 | new_size = split - orig->start; |
823bb20a DS |
2052 | num_extents += count_max_extents(new_size); |
2053 | if (count_max_extents(size) >= num_extents) | |
dcab6a3b JB |
2054 | return; |
2055 | } | |
2056 | ||
9e0baf60 | 2057 | spin_lock(&BTRFS_I(inode)->lock); |
8b62f87b | 2058 | btrfs_mod_outstanding_extents(BTRFS_I(inode), 1); |
9e0baf60 | 2059 | spin_unlock(&BTRFS_I(inode)->lock); |
9ed74f2d JB |
2060 | } |
2061 | ||
2062 | /* | |
5c848198 NB |
2063 | * Handle merged delayed allocation extents so we can keep track of new extents |
2064 | * that are just merged onto old extents, such as when we are doing sequential | |
2065 | * writes, so we can properly account for the metadata space we'll need. | |
9ed74f2d | 2066 | */ |
5c848198 NB |
2067 | void btrfs_merge_delalloc_extent(struct inode *inode, struct extent_state *new, |
2068 | struct extent_state *other) | |
9ed74f2d | 2069 | { |
dcab6a3b | 2070 | u64 new_size, old_size; |
823bb20a | 2071 | u32 num_extents; |
dcab6a3b | 2072 | |
9ed74f2d JB |
2073 | /* not delalloc, ignore it */ |
2074 | if (!(other->state & EXTENT_DELALLOC)) | |
1bf85046 | 2075 | return; |
9ed74f2d | 2076 | |
8461a3de JB |
2077 | if (new->start > other->start) |
2078 | new_size = new->end - other->start + 1; | |
2079 | else | |
2080 | new_size = other->end - new->start + 1; | |
dcab6a3b JB |
2081 | |
2082 | /* we're not bigger than the max, unreserve the space and go */ | |
2083 | if (new_size <= BTRFS_MAX_EXTENT_SIZE) { | |
2084 | spin_lock(&BTRFS_I(inode)->lock); | |
8b62f87b | 2085 | btrfs_mod_outstanding_extents(BTRFS_I(inode), -1); |
dcab6a3b JB |
2086 | spin_unlock(&BTRFS_I(inode)->lock); |
2087 | return; | |
2088 | } | |
2089 | ||
2090 | /* | |
ba117213 JB |
2091 | * We have to add up either side to figure out how many extents were |
2092 | * accounted for before we merged into one big extent. If the number of | |
2093 | * extents we accounted for is <= the amount we need for the new range | |
2094 | * then we can return, otherwise drop. Think of it like this | |
2095 | * | |
2096 | * [ 4k][MAX_SIZE] | |
2097 | * | |
2098 | * So we've grown the extent by a MAX_SIZE extent, this would mean we | |
2099 | * need 2 outstanding extents, on one side we have 1 and the other side | |
2100 | * we have 1 so they are == and we can return. But in this case | |
2101 | * | |
2102 | * [MAX_SIZE+4k][MAX_SIZE+4k] | |
2103 | * | |
2104 | * Each range on their own accounts for 2 extents, but merged together | |
2105 | * they are only 3 extents worth of accounting, so we need to drop in | |
2106 | * this case. | |
dcab6a3b | 2107 | */ |
ba117213 | 2108 | old_size = other->end - other->start + 1; |
823bb20a | 2109 | num_extents = count_max_extents(old_size); |
ba117213 | 2110 | old_size = new->end - new->start + 1; |
823bb20a DS |
2111 | num_extents += count_max_extents(old_size); |
2112 | if (count_max_extents(new_size) >= num_extents) | |
dcab6a3b JB |
2113 | return; |
2114 | ||
9e0baf60 | 2115 | spin_lock(&BTRFS_I(inode)->lock); |
8b62f87b | 2116 | btrfs_mod_outstanding_extents(BTRFS_I(inode), -1); |
9e0baf60 | 2117 | spin_unlock(&BTRFS_I(inode)->lock); |
9ed74f2d JB |
2118 | } |
2119 | ||
eb73c1b7 MX |
2120 | static void btrfs_add_delalloc_inodes(struct btrfs_root *root, |
2121 | struct inode *inode) | |
2122 | { | |
0b246afa JM |
2123 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
2124 | ||
eb73c1b7 MX |
2125 | spin_lock(&root->delalloc_lock); |
2126 | if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) { | |
2127 | list_add_tail(&BTRFS_I(inode)->delalloc_inodes, | |
2128 | &root->delalloc_inodes); | |
2129 | set_bit(BTRFS_INODE_IN_DELALLOC_LIST, | |
2130 | &BTRFS_I(inode)->runtime_flags); | |
2131 | root->nr_delalloc_inodes++; | |
2132 | if (root->nr_delalloc_inodes == 1) { | |
0b246afa | 2133 | spin_lock(&fs_info->delalloc_root_lock); |
eb73c1b7 MX |
2134 | BUG_ON(!list_empty(&root->delalloc_root)); |
2135 | list_add_tail(&root->delalloc_root, | |
0b246afa JM |
2136 | &fs_info->delalloc_roots); |
2137 | spin_unlock(&fs_info->delalloc_root_lock); | |
eb73c1b7 MX |
2138 | } |
2139 | } | |
2140 | spin_unlock(&root->delalloc_lock); | |
2141 | } | |
2142 | ||
2b877331 NB |
2143 | |
2144 | void __btrfs_del_delalloc_inode(struct btrfs_root *root, | |
2145 | struct btrfs_inode *inode) | |
eb73c1b7 | 2146 | { |
3ffbd68c | 2147 | struct btrfs_fs_info *fs_info = root->fs_info; |
0b246afa | 2148 | |
9e3e97f4 NB |
2149 | if (!list_empty(&inode->delalloc_inodes)) { |
2150 | list_del_init(&inode->delalloc_inodes); | |
eb73c1b7 | 2151 | clear_bit(BTRFS_INODE_IN_DELALLOC_LIST, |
9e3e97f4 | 2152 | &inode->runtime_flags); |
eb73c1b7 MX |
2153 | root->nr_delalloc_inodes--; |
2154 | if (!root->nr_delalloc_inodes) { | |
7c8a0d36 | 2155 | ASSERT(list_empty(&root->delalloc_inodes)); |
0b246afa | 2156 | spin_lock(&fs_info->delalloc_root_lock); |
eb73c1b7 MX |
2157 | BUG_ON(list_empty(&root->delalloc_root)); |
2158 | list_del_init(&root->delalloc_root); | |
0b246afa | 2159 | spin_unlock(&fs_info->delalloc_root_lock); |
eb73c1b7 MX |
2160 | } |
2161 | } | |
2b877331 NB |
2162 | } |
2163 | ||
2164 | static void btrfs_del_delalloc_inode(struct btrfs_root *root, | |
2165 | struct btrfs_inode *inode) | |
2166 | { | |
2167 | spin_lock(&root->delalloc_lock); | |
2168 | __btrfs_del_delalloc_inode(root, inode); | |
eb73c1b7 MX |
2169 | spin_unlock(&root->delalloc_lock); |
2170 | } | |
2171 | ||
d352ac68 | 2172 | /* |
e06a1fc9 NB |
2173 | * Properly track delayed allocation bytes in the inode and to maintain the |
2174 | * list of inodes that have pending delalloc work to be done. | |
d352ac68 | 2175 | */ |
e06a1fc9 NB |
2176 | void btrfs_set_delalloc_extent(struct inode *inode, struct extent_state *state, |
2177 | unsigned *bits) | |
291d673e | 2178 | { |
0b246afa JM |
2179 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
2180 | ||
47059d93 WS |
2181 | if ((*bits & EXTENT_DEFRAG) && !(*bits & EXTENT_DELALLOC)) |
2182 | WARN_ON(1); | |
75eff68e CM |
2183 | /* |
2184 | * set_bit and clear bit hooks normally require _irqsave/restore | |
27160b6b | 2185 | * but in this case, we are only testing for the DELALLOC |
75eff68e CM |
2186 | * bit, which is only set or cleared with irqs on |
2187 | */ | |
0ca1f7ce | 2188 | if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) { |
291d673e | 2189 | struct btrfs_root *root = BTRFS_I(inode)->root; |
0ca1f7ce | 2190 | u64 len = state->end + 1 - state->start; |
8b62f87b | 2191 | u32 num_extents = count_max_extents(len); |
70ddc553 | 2192 | bool do_list = !btrfs_is_free_space_inode(BTRFS_I(inode)); |
9ed74f2d | 2193 | |
8b62f87b JB |
2194 | spin_lock(&BTRFS_I(inode)->lock); |
2195 | btrfs_mod_outstanding_extents(BTRFS_I(inode), num_extents); | |
2196 | spin_unlock(&BTRFS_I(inode)->lock); | |
287a0ab9 | 2197 | |
6a3891c5 | 2198 | /* For sanity tests */ |
0b246afa | 2199 | if (btrfs_is_testing(fs_info)) |
6a3891c5 JB |
2200 | return; |
2201 | ||
104b4e51 NB |
2202 | percpu_counter_add_batch(&fs_info->delalloc_bytes, len, |
2203 | fs_info->delalloc_batch); | |
df0af1a5 | 2204 | spin_lock(&BTRFS_I(inode)->lock); |
0ca1f7ce | 2205 | BTRFS_I(inode)->delalloc_bytes += len; |
47059d93 WS |
2206 | if (*bits & EXTENT_DEFRAG) |
2207 | BTRFS_I(inode)->defrag_bytes += len; | |
df0af1a5 | 2208 | if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST, |
eb73c1b7 MX |
2209 | &BTRFS_I(inode)->runtime_flags)) |
2210 | btrfs_add_delalloc_inodes(root, inode); | |
df0af1a5 | 2211 | spin_unlock(&BTRFS_I(inode)->lock); |
291d673e | 2212 | } |
a7e3b975 FM |
2213 | |
2214 | if (!(state->state & EXTENT_DELALLOC_NEW) && | |
2215 | (*bits & EXTENT_DELALLOC_NEW)) { | |
2216 | spin_lock(&BTRFS_I(inode)->lock); | |
2217 | BTRFS_I(inode)->new_delalloc_bytes += state->end + 1 - | |
2218 | state->start; | |
2219 | spin_unlock(&BTRFS_I(inode)->lock); | |
2220 | } | |
291d673e CM |
2221 | } |
2222 | ||
d352ac68 | 2223 | /* |
a36bb5f9 NB |
2224 | * Once a range is no longer delalloc this function ensures that proper |
2225 | * accounting happens. | |
d352ac68 | 2226 | */ |
a36bb5f9 NB |
2227 | void btrfs_clear_delalloc_extent(struct inode *vfs_inode, |
2228 | struct extent_state *state, unsigned *bits) | |
291d673e | 2229 | { |
a36bb5f9 NB |
2230 | struct btrfs_inode *inode = BTRFS_I(vfs_inode); |
2231 | struct btrfs_fs_info *fs_info = btrfs_sb(vfs_inode->i_sb); | |
47059d93 | 2232 | u64 len = state->end + 1 - state->start; |
823bb20a | 2233 | u32 num_extents = count_max_extents(len); |
47059d93 | 2234 | |
4a4b964f FM |
2235 | if ((state->state & EXTENT_DEFRAG) && (*bits & EXTENT_DEFRAG)) { |
2236 | spin_lock(&inode->lock); | |
6fc0ef68 | 2237 | inode->defrag_bytes -= len; |
4a4b964f FM |
2238 | spin_unlock(&inode->lock); |
2239 | } | |
47059d93 | 2240 | |
75eff68e CM |
2241 | /* |
2242 | * set_bit and clear bit hooks normally require _irqsave/restore | |
27160b6b | 2243 | * but in this case, we are only testing for the DELALLOC |
75eff68e CM |
2244 | * bit, which is only set or cleared with irqs on |
2245 | */ | |
0ca1f7ce | 2246 | if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) { |
6fc0ef68 | 2247 | struct btrfs_root *root = inode->root; |
83eea1f1 | 2248 | bool do_list = !btrfs_is_free_space_inode(inode); |
bcbfce8a | 2249 | |
8b62f87b JB |
2250 | spin_lock(&inode->lock); |
2251 | btrfs_mod_outstanding_extents(inode, -num_extents); | |
2252 | spin_unlock(&inode->lock); | |
0ca1f7ce | 2253 | |
b6d08f06 JB |
2254 | /* |
2255 | * We don't reserve metadata space for space cache inodes so we | |
52042d8e | 2256 | * don't need to call delalloc_release_metadata if there is an |
b6d08f06 JB |
2257 | * error. |
2258 | */ | |
a315e68f | 2259 | if (*bits & EXTENT_CLEAR_META_RESV && |
0b246afa | 2260 | root != fs_info->tree_root) |
43b18595 | 2261 | btrfs_delalloc_release_metadata(inode, len, false); |
0ca1f7ce | 2262 | |
6a3891c5 | 2263 | /* For sanity tests. */ |
0b246afa | 2264 | if (btrfs_is_testing(fs_info)) |
6a3891c5 JB |
2265 | return; |
2266 | ||
37f00a6d | 2267 | if (!btrfs_is_data_reloc_root(root) && |
a315e68f FM |
2268 | do_list && !(state->state & EXTENT_NORESERVE) && |
2269 | (*bits & EXTENT_CLEAR_DATA_RESV)) | |
9db5d510 | 2270 | btrfs_free_reserved_data_space_noquota(fs_info, len); |
9ed74f2d | 2271 | |
104b4e51 NB |
2272 | percpu_counter_add_batch(&fs_info->delalloc_bytes, -len, |
2273 | fs_info->delalloc_batch); | |
6fc0ef68 NB |
2274 | spin_lock(&inode->lock); |
2275 | inode->delalloc_bytes -= len; | |
2276 | if (do_list && inode->delalloc_bytes == 0 && | |
df0af1a5 | 2277 | test_bit(BTRFS_INODE_IN_DELALLOC_LIST, |
9e3e97f4 | 2278 | &inode->runtime_flags)) |
eb73c1b7 | 2279 | btrfs_del_delalloc_inode(root, inode); |
6fc0ef68 | 2280 | spin_unlock(&inode->lock); |
291d673e | 2281 | } |
a7e3b975 FM |
2282 | |
2283 | if ((state->state & EXTENT_DELALLOC_NEW) && | |
2284 | (*bits & EXTENT_DELALLOC_NEW)) { | |
2285 | spin_lock(&inode->lock); | |
2286 | ASSERT(inode->new_delalloc_bytes >= len); | |
2287 | inode->new_delalloc_bytes -= len; | |
2766ff61 FM |
2288 | if (*bits & EXTENT_ADD_INODE_BYTES) |
2289 | inode_add_bytes(&inode->vfs_inode, len); | |
a7e3b975 FM |
2290 | spin_unlock(&inode->lock); |
2291 | } | |
291d673e CM |
2292 | } |
2293 | ||
d352ac68 CM |
2294 | /* |
2295 | * in order to insert checksums into the metadata in large chunks, | |
2296 | * we wait until bio submission time. All the pages in the bio are | |
2297 | * checksummed and sums are attached onto the ordered extent record. | |
2298 | * | |
2299 | * At IO completion time the cums attached on the ordered extent record | |
2300 | * are inserted into the btree | |
2301 | */ | |
8896a08d | 2302 | static blk_status_t btrfs_submit_bio_start(struct inode *inode, struct bio *bio, |
1941b64b | 2303 | u64 dio_file_offset) |
065631f6 | 2304 | { |
e331f6b1 | 2305 | return btrfs_csum_one_bio(BTRFS_I(inode), bio, (u64)-1, false); |
4a69a410 | 2306 | } |
e015640f | 2307 | |
abb99cfd NA |
2308 | /* |
2309 | * Split an extent_map at [start, start + len] | |
2310 | * | |
2311 | * This function is intended to be used only for extract_ordered_extent(). | |
2312 | */ | |
2313 | static int split_zoned_em(struct btrfs_inode *inode, u64 start, u64 len, | |
2314 | u64 pre, u64 post) | |
2315 | { | |
2316 | struct extent_map_tree *em_tree = &inode->extent_tree; | |
2317 | struct extent_map *em; | |
2318 | struct extent_map *split_pre = NULL; | |
2319 | struct extent_map *split_mid = NULL; | |
2320 | struct extent_map *split_post = NULL; | |
2321 | int ret = 0; | |
abb99cfd NA |
2322 | unsigned long flags; |
2323 | ||
2324 | /* Sanity check */ | |
2325 | if (pre == 0 && post == 0) | |
2326 | return 0; | |
2327 | ||
2328 | split_pre = alloc_extent_map(); | |
2329 | if (pre) | |
2330 | split_mid = alloc_extent_map(); | |
2331 | if (post) | |
2332 | split_post = alloc_extent_map(); | |
2333 | if (!split_pre || (pre && !split_mid) || (post && !split_post)) { | |
2334 | ret = -ENOMEM; | |
2335 | goto out; | |
2336 | } | |
2337 | ||
2338 | ASSERT(pre + post < len); | |
2339 | ||
2340 | lock_extent(&inode->io_tree, start, start + len - 1); | |
2341 | write_lock(&em_tree->lock); | |
2342 | em = lookup_extent_mapping(em_tree, start, len); | |
2343 | if (!em) { | |
2344 | ret = -EIO; | |
2345 | goto out_unlock; | |
2346 | } | |
2347 | ||
2348 | ASSERT(em->len == len); | |
2349 | ASSERT(!test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)); | |
2350 | ASSERT(em->block_start < EXTENT_MAP_LAST_BYTE); | |
63fb5879 NA |
2351 | ASSERT(test_bit(EXTENT_FLAG_PINNED, &em->flags)); |
2352 | ASSERT(!test_bit(EXTENT_FLAG_LOGGING, &em->flags)); | |
2353 | ASSERT(!list_empty(&em->list)); | |
abb99cfd NA |
2354 | |
2355 | flags = em->flags; | |
2356 | clear_bit(EXTENT_FLAG_PINNED, &em->flags); | |
abb99cfd NA |
2357 | |
2358 | /* First, replace the em with a new extent_map starting from * em->start */ | |
2359 | split_pre->start = em->start; | |
2360 | split_pre->len = (pre ? pre : em->len - post); | |
2361 | split_pre->orig_start = split_pre->start; | |
2362 | split_pre->block_start = em->block_start; | |
2363 | split_pre->block_len = split_pre->len; | |
2364 | split_pre->orig_block_len = split_pre->block_len; | |
2365 | split_pre->ram_bytes = split_pre->len; | |
2366 | split_pre->flags = flags; | |
2367 | split_pre->compress_type = em->compress_type; | |
2368 | split_pre->generation = em->generation; | |
2369 | ||
63fb5879 | 2370 | replace_extent_mapping(em_tree, em, split_pre, 1); |
abb99cfd NA |
2371 | |
2372 | /* | |
2373 | * Now we only have an extent_map at: | |
2374 | * [em->start, em->start + pre] if pre != 0 | |
2375 | * [em->start, em->start + em->len - post] if pre == 0 | |
2376 | */ | |
2377 | ||
2378 | if (pre) { | |
2379 | /* Insert the middle extent_map */ | |
2380 | split_mid->start = em->start + pre; | |
2381 | split_mid->len = em->len - pre - post; | |
2382 | split_mid->orig_start = split_mid->start; | |
2383 | split_mid->block_start = em->block_start + pre; | |
2384 | split_mid->block_len = split_mid->len; | |
2385 | split_mid->orig_block_len = split_mid->block_len; | |
2386 | split_mid->ram_bytes = split_mid->len; | |
2387 | split_mid->flags = flags; | |
2388 | split_mid->compress_type = em->compress_type; | |
2389 | split_mid->generation = em->generation; | |
63fb5879 | 2390 | add_extent_mapping(em_tree, split_mid, 1); |
abb99cfd NA |
2391 | } |
2392 | ||
2393 | if (post) { | |
2394 | split_post->start = em->start + em->len - post; | |
2395 | split_post->len = post; | |
2396 | split_post->orig_start = split_post->start; | |
2397 | split_post->block_start = em->block_start + em->len - post; | |
2398 | split_post->block_len = split_post->len; | |
2399 | split_post->orig_block_len = split_post->block_len; | |
2400 | split_post->ram_bytes = split_post->len; | |
2401 | split_post->flags = flags; | |
2402 | split_post->compress_type = em->compress_type; | |
2403 | split_post->generation = em->generation; | |
63fb5879 | 2404 | add_extent_mapping(em_tree, split_post, 1); |
abb99cfd NA |
2405 | } |
2406 | ||
2407 | /* Once for us */ | |
2408 | free_extent_map(em); | |
2409 | /* Once for the tree */ | |
2410 | free_extent_map(em); | |
2411 | ||
2412 | out_unlock: | |
2413 | write_unlock(&em_tree->lock); | |
2414 | unlock_extent(&inode->io_tree, start, start + len - 1); | |
2415 | out: | |
2416 | free_extent_map(split_pre); | |
2417 | free_extent_map(split_mid); | |
2418 | free_extent_map(split_post); | |
2419 | ||
2420 | return ret; | |
2421 | } | |
2422 | ||
d22002fd NA |
2423 | static blk_status_t extract_ordered_extent(struct btrfs_inode *inode, |
2424 | struct bio *bio, loff_t file_offset) | |
2425 | { | |
2426 | struct btrfs_ordered_extent *ordered; | |
d22002fd | 2427 | u64 start = (u64)bio->bi_iter.bi_sector << SECTOR_SHIFT; |
abb99cfd | 2428 | u64 file_len; |
d22002fd NA |
2429 | u64 len = bio->bi_iter.bi_size; |
2430 | u64 end = start + len; | |
2431 | u64 ordered_end; | |
2432 | u64 pre, post; | |
2433 | int ret = 0; | |
2434 | ||
2435 | ordered = btrfs_lookup_ordered_extent(inode, file_offset); | |
2436 | if (WARN_ON_ONCE(!ordered)) | |
2437 | return BLK_STS_IOERR; | |
2438 | ||
2439 | /* No need to split */ | |
2440 | if (ordered->disk_num_bytes == len) | |
2441 | goto out; | |
2442 | ||
2443 | /* We cannot split once end_bio'd ordered extent */ | |
2444 | if (WARN_ON_ONCE(ordered->bytes_left != ordered->disk_num_bytes)) { | |
2445 | ret = -EINVAL; | |
2446 | goto out; | |
2447 | } | |
2448 | ||
2449 | /* We cannot split a compressed ordered extent */ | |
2450 | if (WARN_ON_ONCE(ordered->disk_num_bytes != ordered->num_bytes)) { | |
2451 | ret = -EINVAL; | |
2452 | goto out; | |
2453 | } | |
2454 | ||
2455 | ordered_end = ordered->disk_bytenr + ordered->disk_num_bytes; | |
2456 | /* bio must be in one ordered extent */ | |
2457 | if (WARN_ON_ONCE(start < ordered->disk_bytenr || end > ordered_end)) { | |
2458 | ret = -EINVAL; | |
2459 | goto out; | |
2460 | } | |
2461 | ||
2462 | /* Checksum list should be empty */ | |
2463 | if (WARN_ON_ONCE(!list_empty(&ordered->list))) { | |
2464 | ret = -EINVAL; | |
2465 | goto out; | |
2466 | } | |
2467 | ||
abb99cfd | 2468 | file_len = ordered->num_bytes; |
d22002fd NA |
2469 | pre = start - ordered->disk_bytenr; |
2470 | post = ordered_end - end; | |
2471 | ||
2472 | ret = btrfs_split_ordered_extent(ordered, pre, post); | |
2473 | if (ret) | |
2474 | goto out; | |
abb99cfd | 2475 | ret = split_zoned_em(inode, file_offset, file_len, pre, post); |
d22002fd NA |
2476 | |
2477 | out: | |
d22002fd NA |
2478 | btrfs_put_ordered_extent(ordered); |
2479 | ||
2480 | return errno_to_blk_status(ret); | |
2481 | } | |
2482 | ||
d352ac68 | 2483 | /* |
cad321ad | 2484 | * extent_io.c submission hook. This does the right thing for csum calculation |
4c274bc6 LB |
2485 | * on write, or reading the csums from the tree before a read. |
2486 | * | |
2487 | * Rules about async/sync submit, | |
2488 | * a) read: sync submit | |
2489 | * | |
2490 | * b) write without checksum: sync submit | |
2491 | * | |
2492 | * c) write with checksum: | |
2493 | * c-1) if bio is issued by fsync: sync submit | |
2494 | * (sync_writers != 0) | |
2495 | * | |
2496 | * c-2) if root is reloc root: sync submit | |
2497 | * (only in case of buffered IO) | |
2498 | * | |
2499 | * c-3) otherwise: async submit | |
d352ac68 | 2500 | */ |
908930f3 NB |
2501 | blk_status_t btrfs_submit_data_bio(struct inode *inode, struct bio *bio, |
2502 | int mirror_num, unsigned long bio_flags) | |
50489a57 | 2503 | |
44b8bd7e | 2504 | { |
0b246afa | 2505 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
44b8bd7e | 2506 | struct btrfs_root *root = BTRFS_I(inode)->root; |
0d51e28a | 2507 | enum btrfs_wq_endio_type metadata = BTRFS_WQ_ENDIO_DATA; |
4e4cbee9 | 2508 | blk_status_t ret = 0; |
19b9bdb0 | 2509 | int skip_sum; |
b812ce28 | 2510 | int async = !atomic_read(&BTRFS_I(inode)->sync_writers); |
44b8bd7e | 2511 | |
42437a63 | 2512 | skip_sum = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM) || |
056c8311 | 2513 | test_bit(BTRFS_FS_STATE_NO_CSUMS, &fs_info->fs_state); |
cad321ad | 2514 | |
70ddc553 | 2515 | if (btrfs_is_free_space_inode(BTRFS_I(inode))) |
0d51e28a | 2516 | metadata = BTRFS_WQ_ENDIO_FREE_SPACE; |
0417341e | 2517 | |
d22002fd NA |
2518 | if (bio_op(bio) == REQ_OP_ZONE_APPEND) { |
2519 | struct page *page = bio_first_bvec_all(bio)->bv_page; | |
2520 | loff_t file_offset = page_offset(page); | |
2521 | ||
2522 | ret = extract_ordered_extent(BTRFS_I(inode), bio, file_offset); | |
2523 | if (ret) | |
2524 | goto out; | |
2525 | } | |
2526 | ||
cfe94440 | 2527 | if (btrfs_op(bio) != BTRFS_MAP_WRITE) { |
0b246afa | 2528 | ret = btrfs_bio_wq_end_io(fs_info, bio, metadata); |
5fd02043 | 2529 | if (ret) |
61891923 | 2530 | goto out; |
5fd02043 | 2531 | |
d20f7043 | 2532 | if (bio_flags & EXTENT_BIO_COMPRESSED) { |
f9f15de8 JB |
2533 | /* |
2534 | * btrfs_submit_compressed_read will handle completing | |
2535 | * the bio if there were any errors, so just return | |
2536 | * here. | |
2537 | */ | |
61891923 SB |
2538 | ret = btrfs_submit_compressed_read(inode, bio, |
2539 | mirror_num, | |
2540 | bio_flags); | |
f9f15de8 | 2541 | goto out_no_endio; |
334c16d8 JB |
2542 | } else { |
2543 | /* | |
2544 | * Lookup bio sums does extra checks around whether we | |
2545 | * need to csum or not, which is why we ignore skip_sum | |
2546 | * here. | |
2547 | */ | |
6275193e | 2548 | ret = btrfs_lookup_bio_sums(inode, bio, NULL); |
c2db1073 | 2549 | if (ret) |
61891923 | 2550 | goto out; |
c2db1073 | 2551 | } |
4d1b5fb4 | 2552 | goto mapit; |
b812ce28 | 2553 | } else if (async && !skip_sum) { |
17d217fe | 2554 | /* csum items have already been cloned */ |
37f00a6d | 2555 | if (btrfs_is_data_reloc_root(root)) |
17d217fe | 2556 | goto mapit; |
19b9bdb0 | 2557 | /* we're doing a write, do the async checksumming */ |
8896a08d QW |
2558 | ret = btrfs_wq_submit_bio(inode, bio, mirror_num, bio_flags, |
2559 | 0, btrfs_submit_bio_start); | |
61891923 | 2560 | goto out; |
b812ce28 | 2561 | } else if (!skip_sum) { |
e331f6b1 | 2562 | ret = btrfs_csum_one_bio(BTRFS_I(inode), bio, (u64)-1, false); |
b812ce28 JB |
2563 | if (ret) |
2564 | goto out; | |
19b9bdb0 CM |
2565 | } |
2566 | ||
0b86a832 | 2567 | mapit: |
08635bae | 2568 | ret = btrfs_map_bio(fs_info, bio, mirror_num); |
61891923 SB |
2569 | |
2570 | out: | |
4e4cbee9 CH |
2571 | if (ret) { |
2572 | bio->bi_status = ret; | |
4246a0b6 CH |
2573 | bio_endio(bio); |
2574 | } | |
f9f15de8 | 2575 | out_no_endio: |
61891923 | 2576 | return ret; |
065631f6 | 2577 | } |
6885f308 | 2578 | |
d352ac68 CM |
2579 | /* |
2580 | * given a list of ordered sums record them in the inode. This happens | |
2581 | * at IO completion time based on sums calculated at bio submission time. | |
2582 | */ | |
510f85ed NB |
2583 | static int add_pending_csums(struct btrfs_trans_handle *trans, |
2584 | struct list_head *list) | |
e6dcd2dc | 2585 | { |
e6dcd2dc | 2586 | struct btrfs_ordered_sum *sum; |
fc28b25e | 2587 | struct btrfs_root *csum_root = NULL; |
ac01f26a | 2588 | int ret; |
e6dcd2dc | 2589 | |
c6e30871 | 2590 | list_for_each_entry(sum, list, list) { |
7c2871a2 | 2591 | trans->adding_csums = true; |
fc28b25e JB |
2592 | if (!csum_root) |
2593 | csum_root = btrfs_csum_root(trans->fs_info, | |
2594 | sum->bytenr); | |
2595 | ret = btrfs_csum_file_blocks(trans, csum_root, sum); | |
7c2871a2 | 2596 | trans->adding_csums = false; |
ac01f26a NB |
2597 | if (ret) |
2598 | return ret; | |
e6dcd2dc CM |
2599 | } |
2600 | return 0; | |
2601 | } | |
2602 | ||
c3347309 FM |
2603 | static int btrfs_find_new_delalloc_bytes(struct btrfs_inode *inode, |
2604 | const u64 start, | |
2605 | const u64 len, | |
2606 | struct extent_state **cached_state) | |
2607 | { | |
2608 | u64 search_start = start; | |
2609 | const u64 end = start + len - 1; | |
2610 | ||
2611 | while (search_start < end) { | |
2612 | const u64 search_len = end - search_start + 1; | |
2613 | struct extent_map *em; | |
2614 | u64 em_len; | |
2615 | int ret = 0; | |
2616 | ||
2617 | em = btrfs_get_extent(inode, NULL, 0, search_start, search_len); | |
2618 | if (IS_ERR(em)) | |
2619 | return PTR_ERR(em); | |
2620 | ||
2621 | if (em->block_start != EXTENT_MAP_HOLE) | |
2622 | goto next; | |
2623 | ||
2624 | em_len = em->len; | |
2625 | if (em->start < search_start) | |
2626 | em_len -= search_start - em->start; | |
2627 | if (em_len > search_len) | |
2628 | em_len = search_len; | |
2629 | ||
2630 | ret = set_extent_bit(&inode->io_tree, search_start, | |
2631 | search_start + em_len - 1, | |
1cab5e72 NB |
2632 | EXTENT_DELALLOC_NEW, 0, NULL, cached_state, |
2633 | GFP_NOFS, NULL); | |
c3347309 FM |
2634 | next: |
2635 | search_start = extent_map_end(em); | |
2636 | free_extent_map(em); | |
2637 | if (ret) | |
2638 | return ret; | |
2639 | } | |
2640 | return 0; | |
2641 | } | |
2642 | ||
c2566f22 | 2643 | int btrfs_set_extent_delalloc(struct btrfs_inode *inode, u64 start, u64 end, |
e3b8a485 | 2644 | unsigned int extra_bits, |
330a5827 | 2645 | struct extent_state **cached_state) |
ea8c2819 | 2646 | { |
fdb1e121 | 2647 | WARN_ON(PAGE_ALIGNED(end)); |
c3347309 FM |
2648 | |
2649 | if (start >= i_size_read(&inode->vfs_inode) && | |
2650 | !(inode->flags & BTRFS_INODE_PREALLOC)) { | |
2651 | /* | |
2652 | * There can't be any extents following eof in this case so just | |
2653 | * set the delalloc new bit for the range directly. | |
2654 | */ | |
2655 | extra_bits |= EXTENT_DELALLOC_NEW; | |
2656 | } else { | |
2657 | int ret; | |
2658 | ||
2659 | ret = btrfs_find_new_delalloc_bytes(inode, start, | |
2660 | end + 1 - start, | |
2661 | cached_state); | |
2662 | if (ret) | |
2663 | return ret; | |
2664 | } | |
2665 | ||
c2566f22 NB |
2666 | return set_extent_delalloc(&inode->io_tree, start, end, extra_bits, |
2667 | cached_state); | |
ea8c2819 CM |
2668 | } |
2669 | ||
d352ac68 | 2670 | /* see btrfs_writepage_start_hook for details on why this is required */ |
247e743c CM |
2671 | struct btrfs_writepage_fixup { |
2672 | struct page *page; | |
f4b1363c | 2673 | struct inode *inode; |
247e743c CM |
2674 | struct btrfs_work work; |
2675 | }; | |
2676 | ||
b2950863 | 2677 | static void btrfs_writepage_fixup_worker(struct btrfs_work *work) |
247e743c CM |
2678 | { |
2679 | struct btrfs_writepage_fixup *fixup; | |
2680 | struct btrfs_ordered_extent *ordered; | |
2ac55d41 | 2681 | struct extent_state *cached_state = NULL; |
364ecf36 | 2682 | struct extent_changeset *data_reserved = NULL; |
247e743c | 2683 | struct page *page; |
65d87f79 | 2684 | struct btrfs_inode *inode; |
247e743c CM |
2685 | u64 page_start; |
2686 | u64 page_end; | |
25f3c502 | 2687 | int ret = 0; |
f4b1363c | 2688 | bool free_delalloc_space = true; |
247e743c CM |
2689 | |
2690 | fixup = container_of(work, struct btrfs_writepage_fixup, work); | |
2691 | page = fixup->page; | |
65d87f79 | 2692 | inode = BTRFS_I(fixup->inode); |
f4b1363c JB |
2693 | page_start = page_offset(page); |
2694 | page_end = page_offset(page) + PAGE_SIZE - 1; | |
2695 | ||
2696 | /* | |
2697 | * This is similar to page_mkwrite, we need to reserve the space before | |
2698 | * we take the page lock. | |
2699 | */ | |
65d87f79 NB |
2700 | ret = btrfs_delalloc_reserve_space(inode, &data_reserved, page_start, |
2701 | PAGE_SIZE); | |
4a096752 | 2702 | again: |
247e743c | 2703 | lock_page(page); |
25f3c502 CM |
2704 | |
2705 | /* | |
2706 | * Before we queued this fixup, we took a reference on the page. | |
2707 | * page->mapping may go NULL, but it shouldn't be moved to a different | |
2708 | * address space. | |
2709 | */ | |
f4b1363c JB |
2710 | if (!page->mapping || !PageDirty(page) || !PageChecked(page)) { |
2711 | /* | |
2712 | * Unfortunately this is a little tricky, either | |
2713 | * | |
2714 | * 1) We got here and our page had already been dealt with and | |
2715 | * we reserved our space, thus ret == 0, so we need to just | |
2716 | * drop our space reservation and bail. This can happen the | |
2717 | * first time we come into the fixup worker, or could happen | |
2718 | * while waiting for the ordered extent. | |
2719 | * 2) Our page was already dealt with, but we happened to get an | |
2720 | * ENOSPC above from the btrfs_delalloc_reserve_space. In | |
2721 | * this case we obviously don't have anything to release, but | |
2722 | * because the page was already dealt with we don't want to | |
2723 | * mark the page with an error, so make sure we're resetting | |
2724 | * ret to 0. This is why we have this check _before_ the ret | |
2725 | * check, because we do not want to have a surprise ENOSPC | |
2726 | * when the page was already properly dealt with. | |
2727 | */ | |
2728 | if (!ret) { | |
65d87f79 NB |
2729 | btrfs_delalloc_release_extents(inode, PAGE_SIZE); |
2730 | btrfs_delalloc_release_space(inode, data_reserved, | |
f4b1363c JB |
2731 | page_start, PAGE_SIZE, |
2732 | true); | |
2733 | } | |
2734 | ret = 0; | |
247e743c | 2735 | goto out_page; |
f4b1363c | 2736 | } |
247e743c | 2737 | |
25f3c502 | 2738 | /* |
f4b1363c JB |
2739 | * We can't mess with the page state unless it is locked, so now that |
2740 | * it is locked bail if we failed to make our space reservation. | |
25f3c502 | 2741 | */ |
f4b1363c JB |
2742 | if (ret) |
2743 | goto out_page; | |
247e743c | 2744 | |
65d87f79 | 2745 | lock_extent_bits(&inode->io_tree, page_start, page_end, &cached_state); |
4a096752 CM |
2746 | |
2747 | /* already ordered? We're done */ | |
f57ad937 | 2748 | if (PageOrdered(page)) |
f4b1363c | 2749 | goto out_reserved; |
4a096752 | 2750 | |
65d87f79 | 2751 | ordered = btrfs_lookup_ordered_range(inode, page_start, PAGE_SIZE); |
4a096752 | 2752 | if (ordered) { |
65d87f79 NB |
2753 | unlock_extent_cached(&inode->io_tree, page_start, page_end, |
2754 | &cached_state); | |
4a096752 | 2755 | unlock_page(page); |
c0a43603 | 2756 | btrfs_start_ordered_extent(ordered, 1); |
87826df0 | 2757 | btrfs_put_ordered_extent(ordered); |
4a096752 CM |
2758 | goto again; |
2759 | } | |
247e743c | 2760 | |
65d87f79 | 2761 | ret = btrfs_set_extent_delalloc(inode, page_start, page_end, 0, |
330a5827 | 2762 | &cached_state); |
25f3c502 | 2763 | if (ret) |
53687007 | 2764 | goto out_reserved; |
f3038ee3 | 2765 | |
25f3c502 CM |
2766 | /* |
2767 | * Everything went as planned, we're now the owner of a dirty page with | |
2768 | * delayed allocation bits set and space reserved for our COW | |
2769 | * destination. | |
2770 | * | |
2771 | * The page was dirty when we started, nothing should have cleaned it. | |
2772 | */ | |
2773 | BUG_ON(!PageDirty(page)); | |
f4b1363c | 2774 | free_delalloc_space = false; |
53687007 | 2775 | out_reserved: |
65d87f79 | 2776 | btrfs_delalloc_release_extents(inode, PAGE_SIZE); |
f4b1363c | 2777 | if (free_delalloc_space) |
65d87f79 NB |
2778 | btrfs_delalloc_release_space(inode, data_reserved, page_start, |
2779 | PAGE_SIZE, true); | |
2780 | unlock_extent_cached(&inode->io_tree, page_start, page_end, | |
e43bbe5e | 2781 | &cached_state); |
247e743c | 2782 | out_page: |
25f3c502 CM |
2783 | if (ret) { |
2784 | /* | |
2785 | * We hit ENOSPC or other errors. Update the mapping and page | |
2786 | * to reflect the errors and clean the page. | |
2787 | */ | |
2788 | mapping_set_error(page->mapping, ret); | |
2789 | end_extent_writepage(page, ret, page_start, page_end); | |
2790 | clear_page_dirty_for_io(page); | |
2791 | SetPageError(page); | |
2792 | } | |
e4f94347 | 2793 | btrfs_page_clear_checked(inode->root->fs_info, page, page_start, PAGE_SIZE); |
247e743c | 2794 | unlock_page(page); |
09cbfeaf | 2795 | put_page(page); |
b897abec | 2796 | kfree(fixup); |
364ecf36 | 2797 | extent_changeset_free(data_reserved); |
f4b1363c JB |
2798 | /* |
2799 | * As a precaution, do a delayed iput in case it would be the last iput | |
2800 | * that could need flushing space. Recursing back to fixup worker would | |
2801 | * deadlock. | |
2802 | */ | |
65d87f79 | 2803 | btrfs_add_delayed_iput(&inode->vfs_inode); |
247e743c CM |
2804 | } |
2805 | ||
2806 | /* | |
2807 | * There are a few paths in the higher layers of the kernel that directly | |
2808 | * set the page dirty bit without asking the filesystem if it is a | |
2809 | * good idea. This causes problems because we want to make sure COW | |
2810 | * properly happens and the data=ordered rules are followed. | |
2811 | * | |
c8b97818 | 2812 | * In our case any range that doesn't have the ORDERED bit set |
247e743c CM |
2813 | * hasn't been properly setup for IO. We kick off an async process |
2814 | * to fix it up. The async helper will wait for ordered extents, set | |
2815 | * the delalloc bit and make it safe to write the page. | |
2816 | */ | |
a129ffb8 | 2817 | int btrfs_writepage_cow_fixup(struct page *page) |
247e743c CM |
2818 | { |
2819 | struct inode *inode = page->mapping->host; | |
0b246afa | 2820 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
247e743c | 2821 | struct btrfs_writepage_fixup *fixup; |
247e743c | 2822 | |
f57ad937 QW |
2823 | /* This page has ordered extent covering it already */ |
2824 | if (PageOrdered(page)) | |
247e743c CM |
2825 | return 0; |
2826 | ||
25f3c502 CM |
2827 | /* |
2828 | * PageChecked is set below when we create a fixup worker for this page, | |
2829 | * don't try to create another one if we're already PageChecked() | |
2830 | * | |
2831 | * The extent_io writepage code will redirty the page if we send back | |
2832 | * EAGAIN. | |
2833 | */ | |
247e743c CM |
2834 | if (PageChecked(page)) |
2835 | return -EAGAIN; | |
2836 | ||
2837 | fixup = kzalloc(sizeof(*fixup), GFP_NOFS); | |
2838 | if (!fixup) | |
2839 | return -EAGAIN; | |
f421950f | 2840 | |
f4b1363c JB |
2841 | /* |
2842 | * We are already holding a reference to this inode from | |
2843 | * write_cache_pages. We need to hold it because the space reservation | |
2844 | * takes place outside of the page lock, and we can't trust | |
2845 | * page->mapping outside of the page lock. | |
2846 | */ | |
2847 | ihold(inode); | |
e4f94347 | 2848 | btrfs_page_set_checked(fs_info, page, page_offset(page), PAGE_SIZE); |
09cbfeaf | 2849 | get_page(page); |
a0cac0ec | 2850 | btrfs_init_work(&fixup->work, btrfs_writepage_fixup_worker, NULL, NULL); |
247e743c | 2851 | fixup->page = page; |
f4b1363c | 2852 | fixup->inode = inode; |
0b246afa | 2853 | btrfs_queue_work(fs_info->fixup_workers, &fixup->work); |
25f3c502 CM |
2854 | |
2855 | return -EAGAIN; | |
247e743c CM |
2856 | } |
2857 | ||
d899e052 | 2858 | static int insert_reserved_file_extent(struct btrfs_trans_handle *trans, |
c553f94d | 2859 | struct btrfs_inode *inode, u64 file_pos, |
9729f10a | 2860 | struct btrfs_file_extent_item *stack_fi, |
2766ff61 | 2861 | const bool update_inode_bytes, |
9729f10a | 2862 | u64 qgroup_reserved) |
d899e052 | 2863 | { |
c553f94d | 2864 | struct btrfs_root *root = inode->root; |
2766ff61 | 2865 | const u64 sectorsize = root->fs_info->sectorsize; |
d899e052 YZ |
2866 | struct btrfs_path *path; |
2867 | struct extent_buffer *leaf; | |
2868 | struct btrfs_key ins; | |
203f44c5 QW |
2869 | u64 disk_num_bytes = btrfs_stack_file_extent_disk_num_bytes(stack_fi); |
2870 | u64 disk_bytenr = btrfs_stack_file_extent_disk_bytenr(stack_fi); | |
cb36a9bb | 2871 | u64 offset = btrfs_stack_file_extent_offset(stack_fi); |
203f44c5 QW |
2872 | u64 num_bytes = btrfs_stack_file_extent_num_bytes(stack_fi); |
2873 | u64 ram_bytes = btrfs_stack_file_extent_ram_bytes(stack_fi); | |
5893dfb9 | 2874 | struct btrfs_drop_extents_args drop_args = { 0 }; |
d899e052 YZ |
2875 | int ret; |
2876 | ||
2877 | path = btrfs_alloc_path(); | |
d8926bb3 MF |
2878 | if (!path) |
2879 | return -ENOMEM; | |
d899e052 | 2880 | |
a1ed835e CM |
2881 | /* |
2882 | * we may be replacing one extent in the tree with another. | |
2883 | * The new extent is pinned in the extent map, and we don't want | |
2884 | * to drop it from the cache until it is completely in the btree. | |
2885 | * | |
2886 | * So, tell btrfs_drop_extents to leave this extent in the cache. | |
2887 | * the caller is expected to unpin it and allow it to be merged | |
2888 | * with the others. | |
2889 | */ | |
5893dfb9 FM |
2890 | drop_args.path = path; |
2891 | drop_args.start = file_pos; | |
2892 | drop_args.end = file_pos + num_bytes; | |
2893 | drop_args.replace_extent = true; | |
2894 | drop_args.extent_item_size = sizeof(*stack_fi); | |
2895 | ret = btrfs_drop_extents(trans, root, inode, &drop_args); | |
79787eaa JM |
2896 | if (ret) |
2897 | goto out; | |
d899e052 | 2898 | |
5893dfb9 | 2899 | if (!drop_args.extent_inserted) { |
c553f94d | 2900 | ins.objectid = btrfs_ino(inode); |
1acae57b FDBM |
2901 | ins.offset = file_pos; |
2902 | ins.type = BTRFS_EXTENT_DATA_KEY; | |
2903 | ||
1acae57b | 2904 | ret = btrfs_insert_empty_item(trans, root, path, &ins, |
203f44c5 | 2905 | sizeof(*stack_fi)); |
1acae57b FDBM |
2906 | if (ret) |
2907 | goto out; | |
2908 | } | |
d899e052 | 2909 | leaf = path->nodes[0]; |
203f44c5 QW |
2910 | btrfs_set_stack_file_extent_generation(stack_fi, trans->transid); |
2911 | write_extent_buffer(leaf, stack_fi, | |
2912 | btrfs_item_ptr_offset(leaf, path->slots[0]), | |
2913 | sizeof(struct btrfs_file_extent_item)); | |
b9473439 | 2914 | |
d899e052 | 2915 | btrfs_mark_buffer_dirty(leaf); |
ce195332 | 2916 | btrfs_release_path(path); |
d899e052 | 2917 | |
2766ff61 FM |
2918 | /* |
2919 | * If we dropped an inline extent here, we know the range where it is | |
2920 | * was not marked with the EXTENT_DELALLOC_NEW bit, so we update the | |
1a9fd417 | 2921 | * number of bytes only for that range containing the inline extent. |
2766ff61 FM |
2922 | * The remaining of the range will be processed when clearning the |
2923 | * EXTENT_DELALLOC_BIT bit through the ordered extent completion. | |
2924 | */ | |
2925 | if (file_pos == 0 && !IS_ALIGNED(drop_args.bytes_found, sectorsize)) { | |
2926 | u64 inline_size = round_down(drop_args.bytes_found, sectorsize); | |
2927 | ||
2928 | inline_size = drop_args.bytes_found - inline_size; | |
2929 | btrfs_update_inode_bytes(inode, sectorsize, inline_size); | |
2930 | drop_args.bytes_found -= inline_size; | |
2931 | num_bytes -= sectorsize; | |
2932 | } | |
2933 | ||
2934 | if (update_inode_bytes) | |
2935 | btrfs_update_inode_bytes(inode, num_bytes, drop_args.bytes_found); | |
d899e052 YZ |
2936 | |
2937 | ins.objectid = disk_bytenr; | |
2938 | ins.offset = disk_num_bytes; | |
2939 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
a12b877b | 2940 | |
c553f94d | 2941 | ret = btrfs_inode_set_file_extent_range(inode, file_pos, ram_bytes); |
9ddc959e JB |
2942 | if (ret) |
2943 | goto out; | |
2944 | ||
c553f94d | 2945 | ret = btrfs_alloc_reserved_file_extent(trans, root, btrfs_ino(inode), |
cb36a9bb OS |
2946 | file_pos - offset, |
2947 | qgroup_reserved, &ins); | |
79787eaa | 2948 | out: |
d899e052 | 2949 | btrfs_free_path(path); |
b9473439 | 2950 | |
79787eaa | 2951 | return ret; |
d899e052 YZ |
2952 | } |
2953 | ||
2ff7e61e | 2954 | static void btrfs_release_delalloc_bytes(struct btrfs_fs_info *fs_info, |
e570fd27 MX |
2955 | u64 start, u64 len) |
2956 | { | |
32da5386 | 2957 | struct btrfs_block_group *cache; |
e570fd27 | 2958 | |
0b246afa | 2959 | cache = btrfs_lookup_block_group(fs_info, start); |
e570fd27 MX |
2960 | ASSERT(cache); |
2961 | ||
2962 | spin_lock(&cache->lock); | |
2963 | cache->delalloc_bytes -= len; | |
2964 | spin_unlock(&cache->lock); | |
2965 | ||
2966 | btrfs_put_block_group(cache); | |
2967 | } | |
2968 | ||
203f44c5 | 2969 | static int insert_ordered_extent_file_extent(struct btrfs_trans_handle *trans, |
203f44c5 QW |
2970 | struct btrfs_ordered_extent *oe) |
2971 | { | |
2972 | struct btrfs_file_extent_item stack_fi; | |
2766ff61 | 2973 | bool update_inode_bytes; |
cb36a9bb OS |
2974 | u64 num_bytes = oe->num_bytes; |
2975 | u64 ram_bytes = oe->ram_bytes; | |
203f44c5 QW |
2976 | |
2977 | memset(&stack_fi, 0, sizeof(stack_fi)); | |
2978 | btrfs_set_stack_file_extent_type(&stack_fi, BTRFS_FILE_EXTENT_REG); | |
2979 | btrfs_set_stack_file_extent_disk_bytenr(&stack_fi, oe->disk_bytenr); | |
2980 | btrfs_set_stack_file_extent_disk_num_bytes(&stack_fi, | |
2981 | oe->disk_num_bytes); | |
cb36a9bb | 2982 | btrfs_set_stack_file_extent_offset(&stack_fi, oe->offset); |
203f44c5 | 2983 | if (test_bit(BTRFS_ORDERED_TRUNCATED, &oe->flags)) |
cb36a9bb OS |
2984 | num_bytes = ram_bytes = oe->truncated_len; |
2985 | btrfs_set_stack_file_extent_num_bytes(&stack_fi, num_bytes); | |
2986 | btrfs_set_stack_file_extent_ram_bytes(&stack_fi, ram_bytes); | |
203f44c5 QW |
2987 | btrfs_set_stack_file_extent_compression(&stack_fi, oe->compress_type); |
2988 | /* Encryption and other encoding is reserved and all 0 */ | |
2989 | ||
2766ff61 FM |
2990 | /* |
2991 | * For delalloc, when completing an ordered extent we update the inode's | |
2992 | * bytes when clearing the range in the inode's io tree, so pass false | |
2993 | * as the argument 'update_inode_bytes' to insert_reserved_file_extent(), | |
2994 | * except if the ordered extent was truncated. | |
2995 | */ | |
2996 | update_inode_bytes = test_bit(BTRFS_ORDERED_DIRECT, &oe->flags) || | |
7c0c7269 | 2997 | test_bit(BTRFS_ORDERED_ENCODED, &oe->flags) || |
2766ff61 FM |
2998 | test_bit(BTRFS_ORDERED_TRUNCATED, &oe->flags); |
2999 | ||
3c38c877 NB |
3000 | return insert_reserved_file_extent(trans, BTRFS_I(oe->inode), |
3001 | oe->file_offset, &stack_fi, | |
2766ff61 | 3002 | update_inode_bytes, oe->qgroup_rsv); |
203f44c5 QW |
3003 | } |
3004 | ||
3005 | /* | |
3006 | * As ordered data IO finishes, this gets called so we can finish | |
d352ac68 CM |
3007 | * an ordered extent if the range of bytes in the file it covers are |
3008 | * fully written. | |
3009 | */ | |
5fd02043 | 3010 | static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent) |
e6dcd2dc | 3011 | { |
72e7e6ed NB |
3012 | struct btrfs_inode *inode = BTRFS_I(ordered_extent->inode); |
3013 | struct btrfs_root *root = inode->root; | |
3014 | struct btrfs_fs_info *fs_info = root->fs_info; | |
0ca1f7ce | 3015 | struct btrfs_trans_handle *trans = NULL; |
72e7e6ed | 3016 | struct extent_io_tree *io_tree = &inode->io_tree; |
2ac55d41 | 3017 | struct extent_state *cached_state = NULL; |
bffe633e | 3018 | u64 start, end; |
261507a0 | 3019 | int compress_type = 0; |
77cef2ec | 3020 | int ret = 0; |
bffe633e | 3021 | u64 logical_len = ordered_extent->num_bytes; |
8d510121 | 3022 | bool freespace_inode; |
77cef2ec | 3023 | bool truncated = false; |
49940bdd | 3024 | bool clear_reserved_extent = true; |
2766ff61 | 3025 | unsigned int clear_bits = EXTENT_DEFRAG; |
a7e3b975 | 3026 | |
bffe633e OS |
3027 | start = ordered_extent->file_offset; |
3028 | end = start + ordered_extent->num_bytes - 1; | |
3029 | ||
a7e3b975 FM |
3030 | if (!test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) && |
3031 | !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags) && | |
7c0c7269 OS |
3032 | !test_bit(BTRFS_ORDERED_DIRECT, &ordered_extent->flags) && |
3033 | !test_bit(BTRFS_ORDERED_ENCODED, &ordered_extent->flags)) | |
2766ff61 | 3034 | clear_bits |= EXTENT_DELALLOC_NEW; |
e6dcd2dc | 3035 | |
72e7e6ed | 3036 | freespace_inode = btrfs_is_free_space_inode(inode); |
0cb59c99 | 3037 | |
5fd02043 JB |
3038 | if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) { |
3039 | ret = -EIO; | |
3040 | goto out; | |
3041 | } | |
3042 | ||
be1a1d7a NA |
3043 | /* A valid bdev implies a write on a sequential zone */ |
3044 | if (ordered_extent->bdev) { | |
d8e3fb10 | 3045 | btrfs_rewrite_logical_zoned(ordered_extent); |
be1a1d7a NA |
3046 | btrfs_zone_finish_endio(fs_info, ordered_extent->disk_bytenr, |
3047 | ordered_extent->disk_num_bytes); | |
3048 | } | |
d8e3fb10 | 3049 | |
72e7e6ed | 3050 | btrfs_free_io_failure_record(inode, start, end); |
f612496b | 3051 | |
77cef2ec JB |
3052 | if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) { |
3053 | truncated = true; | |
3054 | logical_len = ordered_extent->truncated_len; | |
3055 | /* Truncated the entire extent, don't bother adding */ | |
3056 | if (!logical_len) | |
3057 | goto out; | |
3058 | } | |
3059 | ||
c2167754 | 3060 | if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) { |
79787eaa | 3061 | BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */ |
94ed938a | 3062 | |
72e7e6ed | 3063 | btrfs_inode_safe_disk_i_size_write(inode, 0); |
8d510121 NB |
3064 | if (freespace_inode) |
3065 | trans = btrfs_join_transaction_spacecache(root); | |
6c760c07 JB |
3066 | else |
3067 | trans = btrfs_join_transaction(root); | |
3068 | if (IS_ERR(trans)) { | |
3069 | ret = PTR_ERR(trans); | |
3070 | trans = NULL; | |
3071 | goto out; | |
c2167754 | 3072 | } |
72e7e6ed | 3073 | trans->block_rsv = &inode->block_rsv; |
729f7961 | 3074 | ret = btrfs_update_inode_fallback(trans, root, inode); |
6c760c07 | 3075 | if (ret) /* -ENOMEM or corruption */ |
66642832 | 3076 | btrfs_abort_transaction(trans, ret); |
c2167754 YZ |
3077 | goto out; |
3078 | } | |
e6dcd2dc | 3079 | |
2766ff61 | 3080 | clear_bits |= EXTENT_LOCKED; |
bffe633e | 3081 | lock_extent_bits(io_tree, start, end, &cached_state); |
e6dcd2dc | 3082 | |
8d510121 NB |
3083 | if (freespace_inode) |
3084 | trans = btrfs_join_transaction_spacecache(root); | |
0cb59c99 | 3085 | else |
7a7eaa40 | 3086 | trans = btrfs_join_transaction(root); |
79787eaa JM |
3087 | if (IS_ERR(trans)) { |
3088 | ret = PTR_ERR(trans); | |
3089 | trans = NULL; | |
a7e3b975 | 3090 | goto out; |
79787eaa | 3091 | } |
a79b7d4b | 3092 | |
72e7e6ed | 3093 | trans->block_rsv = &inode->block_rsv; |
c2167754 | 3094 | |
c8b97818 | 3095 | if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags)) |
261507a0 | 3096 | compress_type = ordered_extent->compress_type; |
d899e052 | 3097 | if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) { |
261507a0 | 3098 | BUG_ON(compress_type); |
72e7e6ed | 3099 | ret = btrfs_mark_extent_written(trans, inode, |
d899e052 YZ |
3100 | ordered_extent->file_offset, |
3101 | ordered_extent->file_offset + | |
77cef2ec | 3102 | logical_len); |
d899e052 | 3103 | } else { |
0b246afa | 3104 | BUG_ON(root == fs_info->tree_root); |
3c38c877 | 3105 | ret = insert_ordered_extent_file_extent(trans, ordered_extent); |
49940bdd JB |
3106 | if (!ret) { |
3107 | clear_reserved_extent = false; | |
2ff7e61e | 3108 | btrfs_release_delalloc_bytes(fs_info, |
bffe633e OS |
3109 | ordered_extent->disk_bytenr, |
3110 | ordered_extent->disk_num_bytes); | |
49940bdd | 3111 | } |
d899e052 | 3112 | } |
72e7e6ed | 3113 | unpin_extent_cache(&inode->extent_tree, ordered_extent->file_offset, |
bffe633e | 3114 | ordered_extent->num_bytes, trans->transid); |
79787eaa | 3115 | if (ret < 0) { |
66642832 | 3116 | btrfs_abort_transaction(trans, ret); |
a7e3b975 | 3117 | goto out; |
79787eaa | 3118 | } |
2ac55d41 | 3119 | |
510f85ed | 3120 | ret = add_pending_csums(trans, &ordered_extent->list); |
ac01f26a NB |
3121 | if (ret) { |
3122 | btrfs_abort_transaction(trans, ret); | |
3123 | goto out; | |
3124 | } | |
e6dcd2dc | 3125 | |
2766ff61 FM |
3126 | /* |
3127 | * If this is a new delalloc range, clear its new delalloc flag to | |
3128 | * update the inode's number of bytes. This needs to be done first | |
3129 | * before updating the inode item. | |
3130 | */ | |
3131 | if ((clear_bits & EXTENT_DELALLOC_NEW) && | |
3132 | !test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) | |
72e7e6ed | 3133 | clear_extent_bit(&inode->io_tree, start, end, |
2766ff61 FM |
3134 | EXTENT_DELALLOC_NEW | EXTENT_ADD_INODE_BYTES, |
3135 | 0, 0, &cached_state); | |
3136 | ||
72e7e6ed | 3137 | btrfs_inode_safe_disk_i_size_write(inode, 0); |
729f7961 | 3138 | ret = btrfs_update_inode_fallback(trans, root, inode); |
6c760c07 | 3139 | if (ret) { /* -ENOMEM or corruption */ |
66642832 | 3140 | btrfs_abort_transaction(trans, ret); |
a7e3b975 | 3141 | goto out; |
1ef30be1 JB |
3142 | } |
3143 | ret = 0; | |
c2167754 | 3144 | out: |
72e7e6ed | 3145 | clear_extent_bit(&inode->io_tree, start, end, clear_bits, |
bffe633e | 3146 | (clear_bits & EXTENT_LOCKED) ? 1 : 0, 0, |
313facc5 | 3147 | &cached_state); |
a7e3b975 | 3148 | |
a698d075 | 3149 | if (trans) |
3a45bb20 | 3150 | btrfs_end_transaction(trans); |
0cb59c99 | 3151 | |
77cef2ec | 3152 | if (ret || truncated) { |
bffe633e | 3153 | u64 unwritten_start = start; |
77cef2ec | 3154 | |
d61bec08 JB |
3155 | /* |
3156 | * If we failed to finish this ordered extent for any reason we | |
3157 | * need to make sure BTRFS_ORDERED_IOERR is set on the ordered | |
3158 | * extent, and mark the inode with the error if it wasn't | |
3159 | * already set. Any error during writeback would have already | |
3160 | * set the mapping error, so we need to set it if we're the ones | |
3161 | * marking this ordered extent as failed. | |
3162 | */ | |
3163 | if (ret && !test_and_set_bit(BTRFS_ORDERED_IOERR, | |
3164 | &ordered_extent->flags)) | |
3165 | mapping_set_error(ordered_extent->inode->i_mapping, -EIO); | |
3166 | ||
77cef2ec | 3167 | if (truncated) |
bffe633e OS |
3168 | unwritten_start += logical_len; |
3169 | clear_extent_uptodate(io_tree, unwritten_start, end, NULL); | |
77cef2ec JB |
3170 | |
3171 | /* Drop the cache for the part of the extent we didn't write. */ | |
72e7e6ed | 3172 | btrfs_drop_extent_cache(inode, unwritten_start, end, 0); |
5fd02043 | 3173 | |
0bec9ef5 JB |
3174 | /* |
3175 | * If the ordered extent had an IOERR or something else went | |
3176 | * wrong we need to return the space for this ordered extent | |
77cef2ec JB |
3177 | * back to the allocator. We only free the extent in the |
3178 | * truncated case if we didn't write out the extent at all. | |
49940bdd JB |
3179 | * |
3180 | * If we made it past insert_reserved_file_extent before we | |
3181 | * errored out then we don't need to do this as the accounting | |
3182 | * has already been done. | |
0bec9ef5 | 3183 | */ |
77cef2ec | 3184 | if ((ret || !logical_len) && |
49940bdd | 3185 | clear_reserved_extent && |
77cef2ec | 3186 | !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) && |
4eaaec24 NB |
3187 | !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) { |
3188 | /* | |
3189 | * Discard the range before returning it back to the | |
3190 | * free space pool | |
3191 | */ | |
46b27f50 | 3192 | if (ret && btrfs_test_opt(fs_info, DISCARD_SYNC)) |
4eaaec24 | 3193 | btrfs_discard_extent(fs_info, |
bffe633e OS |
3194 | ordered_extent->disk_bytenr, |
3195 | ordered_extent->disk_num_bytes, | |
3196 | NULL); | |
2ff7e61e | 3197 | btrfs_free_reserved_extent(fs_info, |
bffe633e OS |
3198 | ordered_extent->disk_bytenr, |
3199 | ordered_extent->disk_num_bytes, 1); | |
4eaaec24 | 3200 | } |
0bec9ef5 JB |
3201 | } |
3202 | ||
5fd02043 | 3203 | /* |
8bad3c02 LB |
3204 | * This needs to be done to make sure anybody waiting knows we are done |
3205 | * updating everything for this ordered extent. | |
5fd02043 | 3206 | */ |
72e7e6ed | 3207 | btrfs_remove_ordered_extent(inode, ordered_extent); |
5fd02043 | 3208 | |
e6dcd2dc CM |
3209 | /* once for us */ |
3210 | btrfs_put_ordered_extent(ordered_extent); | |
3211 | /* once for the tree */ | |
3212 | btrfs_put_ordered_extent(ordered_extent); | |
3213 | ||
5fd02043 JB |
3214 | return ret; |
3215 | } | |
3216 | ||
3217 | static void finish_ordered_fn(struct btrfs_work *work) | |
3218 | { | |
3219 | struct btrfs_ordered_extent *ordered_extent; | |
3220 | ordered_extent = container_of(work, struct btrfs_ordered_extent, work); | |
3221 | btrfs_finish_ordered_io(ordered_extent); | |
e6dcd2dc CM |
3222 | } |
3223 | ||
38a39ac7 QW |
3224 | void btrfs_writepage_endio_finish_ordered(struct btrfs_inode *inode, |
3225 | struct page *page, u64 start, | |
25c1252a | 3226 | u64 end, bool uptodate) |
211f90e6 | 3227 | { |
38a39ac7 | 3228 | trace_btrfs_writepage_end_io_hook(inode, start, end, uptodate); |
1abe9b8a | 3229 | |
e65f152e QW |
3230 | btrfs_mark_ordered_io_finished(inode, page, start, end + 1 - start, |
3231 | finish_ordered_fn, uptodate); | |
211f90e6 CM |
3232 | } |
3233 | ||
265d4ac0 QW |
3234 | /* |
3235 | * check_data_csum - verify checksum of one sector of uncompressed data | |
7ffd27e3 | 3236 | * @inode: inode |
265d4ac0 | 3237 | * @io_bio: btrfs_io_bio which contains the csum |
7ffd27e3 | 3238 | * @bio_offset: offset to the beginning of the bio (in bytes) |
265d4ac0 QW |
3239 | * @page: page where is the data to be verified |
3240 | * @pgoff: offset inside the page | |
c1d6abda | 3241 | * @start: logical offset in the file |
265d4ac0 QW |
3242 | * |
3243 | * The length of such check is always one sector size. | |
3244 | */ | |
c3a3b19b | 3245 | static int check_data_csum(struct inode *inode, struct btrfs_bio *bbio, |
c1d6abda OS |
3246 | u32 bio_offset, struct page *page, u32 pgoff, |
3247 | u64 start) | |
dc380aea | 3248 | { |
d5178578 JT |
3249 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
3250 | SHASH_DESC_ON_STACK(shash, fs_info->csum_shash); | |
dc380aea | 3251 | char *kaddr; |
265d4ac0 | 3252 | u32 len = fs_info->sectorsize; |
223486c2 | 3253 | const u32 csum_size = fs_info->csum_size; |
7ffd27e3 | 3254 | unsigned int offset_sectors; |
d5178578 JT |
3255 | u8 *csum_expected; |
3256 | u8 csum[BTRFS_CSUM_SIZE]; | |
dc380aea | 3257 | |
265d4ac0 QW |
3258 | ASSERT(pgoff + len <= PAGE_SIZE); |
3259 | ||
7ffd27e3 | 3260 | offset_sectors = bio_offset >> fs_info->sectorsize_bits; |
c3a3b19b | 3261 | csum_expected = ((u8 *)bbio->csum) + offset_sectors * csum_size; |
dc380aea MX |
3262 | |
3263 | kaddr = kmap_atomic(page); | |
d5178578 JT |
3264 | shash->tfm = fs_info->csum_shash; |
3265 | ||
fd08001f | 3266 | crypto_shash_digest(shash, kaddr + pgoff, len, csum); |
d5178578 JT |
3267 | |
3268 | if (memcmp(csum, csum_expected, csum_size)) | |
dc380aea MX |
3269 | goto zeroit; |
3270 | ||
3271 | kunmap_atomic(kaddr); | |
3272 | return 0; | |
3273 | zeroit: | |
c1d6abda | 3274 | btrfs_print_data_csum_error(BTRFS_I(inode), start, csum, csum_expected, |
c3a3b19b QW |
3275 | bbio->mirror_num); |
3276 | if (bbio->device) | |
3277 | btrfs_dev_stat_inc_and_print(bbio->device, | |
814723e0 | 3278 | BTRFS_DEV_STAT_CORRUPTION_ERRS); |
dc380aea MX |
3279 | memset(kaddr + pgoff, 1, len); |
3280 | flush_dcache_page(page); | |
3281 | kunmap_atomic(kaddr); | |
dc380aea MX |
3282 | return -EIO; |
3283 | } | |
3284 | ||
d352ac68 | 3285 | /* |
7ffd27e3 | 3286 | * When reads are done, we need to check csums to verify the data is correct. |
4a54c8c1 JS |
3287 | * if there's a match, we allow the bio to finish. If not, the code in |
3288 | * extent_io.c will try to find good copies for us. | |
7ffd27e3 QW |
3289 | * |
3290 | * @bio_offset: offset to the beginning of the bio (in bytes) | |
3291 | * @start: file offset of the range start | |
3292 | * @end: file offset of the range end (inclusive) | |
08508fea QW |
3293 | * |
3294 | * Return a bitmap where bit set means a csum mismatch, and bit not set means | |
3295 | * csum match. | |
d352ac68 | 3296 | */ |
c3a3b19b QW |
3297 | unsigned int btrfs_verify_data_csum(struct btrfs_bio *bbio, |
3298 | u32 bio_offset, struct page *page, | |
3299 | u64 start, u64 end) | |
07157aac | 3300 | { |
07157aac | 3301 | struct inode *inode = page->mapping->host; |
e4f94347 | 3302 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
d1310b2e | 3303 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
ff79f819 | 3304 | struct btrfs_root *root = BTRFS_I(inode)->root; |
f44cf410 QW |
3305 | const u32 sectorsize = root->fs_info->sectorsize; |
3306 | u32 pg_off; | |
08508fea | 3307 | unsigned int result = 0; |
d1310b2e | 3308 | |
e4f94347 QW |
3309 | if (btrfs_page_test_checked(fs_info, page, start, end + 1 - start)) { |
3310 | btrfs_page_clear_checked(fs_info, page, start, end + 1 - start); | |
dc380aea | 3311 | return 0; |
d20f7043 | 3312 | } |
6cbff00f | 3313 | |
3670e645 | 3314 | /* |
e4f94347 QW |
3315 | * This only happens for NODATASUM or compressed read. |
3316 | * Normally this should be covered by above check for compressed read | |
3317 | * or the next check for NODATASUM. Just do a quicker exit here. | |
3670e645 | 3318 | */ |
c3a3b19b | 3319 | if (bbio->csum == NULL) |
dc380aea | 3320 | return 0; |
17d217fe | 3321 | |
6cbff00f | 3322 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM) |
42437a63 JB |
3323 | return 0; |
3324 | ||
056c8311 | 3325 | if (unlikely(test_bit(BTRFS_FS_STATE_NO_CSUMS, &fs_info->fs_state))) |
b6cda9bc | 3326 | return 0; |
d20f7043 | 3327 | |
f44cf410 QW |
3328 | ASSERT(page_offset(page) <= start && |
3329 | end <= page_offset(page) + PAGE_SIZE - 1); | |
3330 | for (pg_off = offset_in_page(start); | |
3331 | pg_off < offset_in_page(end); | |
3332 | pg_off += sectorsize, bio_offset += sectorsize) { | |
e3c62324 | 3333 | u64 file_offset = pg_off + page_offset(page); |
f44cf410 QW |
3334 | int ret; |
3335 | ||
37f00a6d | 3336 | if (btrfs_is_data_reloc_root(root) && |
e3c62324 QW |
3337 | test_range_bit(io_tree, file_offset, |
3338 | file_offset + sectorsize - 1, | |
3339 | EXTENT_NODATASUM, 1, NULL)) { | |
3340 | /* Skip the range without csum for data reloc inode */ | |
3341 | clear_extent_bits(io_tree, file_offset, | |
3342 | file_offset + sectorsize - 1, | |
3343 | EXTENT_NODATASUM); | |
3344 | continue; | |
3345 | } | |
c3a3b19b | 3346 | ret = check_data_csum(inode, bbio, bio_offset, page, pg_off, |
c1d6abda | 3347 | page_offset(page) + pg_off); |
08508fea QW |
3348 | if (ret < 0) { |
3349 | const int nr_bit = (pg_off - offset_in_page(start)) >> | |
3350 | root->fs_info->sectorsize_bits; | |
3351 | ||
3352 | result |= (1U << nr_bit); | |
3353 | } | |
f44cf410 | 3354 | } |
08508fea | 3355 | return result; |
07157aac | 3356 | } |
b888db2b | 3357 | |
c1c3fac2 NB |
3358 | /* |
3359 | * btrfs_add_delayed_iput - perform a delayed iput on @inode | |
3360 | * | |
3361 | * @inode: The inode we want to perform iput on | |
3362 | * | |
3363 | * This function uses the generic vfs_inode::i_count to track whether we should | |
3364 | * just decrement it (in case it's > 1) or if this is the last iput then link | |
3365 | * the inode to the delayed iput machinery. Delayed iputs are processed at | |
3366 | * transaction commit time/superblock commit/cleaner kthread. | |
3367 | */ | |
24bbcf04 YZ |
3368 | void btrfs_add_delayed_iput(struct inode *inode) |
3369 | { | |
0b246afa | 3370 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
8089fe62 | 3371 | struct btrfs_inode *binode = BTRFS_I(inode); |
24bbcf04 YZ |
3372 | |
3373 | if (atomic_add_unless(&inode->i_count, -1, 1)) | |
3374 | return; | |
3375 | ||
034f784d | 3376 | atomic_inc(&fs_info->nr_delayed_iputs); |
24bbcf04 | 3377 | spin_lock(&fs_info->delayed_iput_lock); |
c1c3fac2 NB |
3378 | ASSERT(list_empty(&binode->delayed_iput)); |
3379 | list_add_tail(&binode->delayed_iput, &fs_info->delayed_iputs); | |
24bbcf04 | 3380 | spin_unlock(&fs_info->delayed_iput_lock); |
fd340d0f JB |
3381 | if (!test_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags)) |
3382 | wake_up_process(fs_info->cleaner_kthread); | |
24bbcf04 YZ |
3383 | } |
3384 | ||
63611e73 JB |
3385 | static void run_delayed_iput_locked(struct btrfs_fs_info *fs_info, |
3386 | struct btrfs_inode *inode) | |
3387 | { | |
3388 | list_del_init(&inode->delayed_iput); | |
3389 | spin_unlock(&fs_info->delayed_iput_lock); | |
3390 | iput(&inode->vfs_inode); | |
3391 | if (atomic_dec_and_test(&fs_info->nr_delayed_iputs)) | |
3392 | wake_up(&fs_info->delayed_iputs_wait); | |
3393 | spin_lock(&fs_info->delayed_iput_lock); | |
3394 | } | |
3395 | ||
3396 | static void btrfs_run_delayed_iput(struct btrfs_fs_info *fs_info, | |
3397 | struct btrfs_inode *inode) | |
3398 | { | |
3399 | if (!list_empty(&inode->delayed_iput)) { | |
3400 | spin_lock(&fs_info->delayed_iput_lock); | |
3401 | if (!list_empty(&inode->delayed_iput)) | |
3402 | run_delayed_iput_locked(fs_info, inode); | |
3403 | spin_unlock(&fs_info->delayed_iput_lock); | |
3404 | } | |
3405 | } | |
3406 | ||
2ff7e61e | 3407 | void btrfs_run_delayed_iputs(struct btrfs_fs_info *fs_info) |
24bbcf04 | 3408 | { |
24bbcf04 | 3409 | |
24bbcf04 | 3410 | spin_lock(&fs_info->delayed_iput_lock); |
8089fe62 DS |
3411 | while (!list_empty(&fs_info->delayed_iputs)) { |
3412 | struct btrfs_inode *inode; | |
3413 | ||
3414 | inode = list_first_entry(&fs_info->delayed_iputs, | |
3415 | struct btrfs_inode, delayed_iput); | |
63611e73 | 3416 | run_delayed_iput_locked(fs_info, inode); |
71795ee5 | 3417 | cond_resched_lock(&fs_info->delayed_iput_lock); |
24bbcf04 | 3418 | } |
8089fe62 | 3419 | spin_unlock(&fs_info->delayed_iput_lock); |
24bbcf04 YZ |
3420 | } |
3421 | ||
034f784d | 3422 | /** |
2639631d NB |
3423 | * Wait for flushing all delayed iputs |
3424 | * | |
3425 | * @fs_info: the filesystem | |
034f784d JB |
3426 | * |
3427 | * This will wait on any delayed iputs that are currently running with KILLABLE | |
3428 | * set. Once they are all done running we will return, unless we are killed in | |
3429 | * which case we return EINTR. This helps in user operations like fallocate etc | |
3430 | * that might get blocked on the iputs. | |
2639631d NB |
3431 | * |
3432 | * Return EINTR if we were killed, 0 if nothing's pending | |
034f784d JB |
3433 | */ |
3434 | int btrfs_wait_on_delayed_iputs(struct btrfs_fs_info *fs_info) | |
3435 | { | |
3436 | int ret = wait_event_killable(fs_info->delayed_iputs_wait, | |
3437 | atomic_read(&fs_info->nr_delayed_iputs) == 0); | |
3438 | if (ret) | |
3439 | return -EINTR; | |
3440 | return 0; | |
3441 | } | |
3442 | ||
7b128766 | 3443 | /* |
f7e9e8fc OS |
3444 | * This creates an orphan entry for the given inode in case something goes wrong |
3445 | * in the middle of an unlink. | |
7b128766 | 3446 | */ |
73f2e545 | 3447 | int btrfs_orphan_add(struct btrfs_trans_handle *trans, |
27919067 | 3448 | struct btrfs_inode *inode) |
7b128766 | 3449 | { |
d68fc57b | 3450 | int ret; |
7b128766 | 3451 | |
27919067 OS |
3452 | ret = btrfs_insert_orphan_item(trans, inode->root, btrfs_ino(inode)); |
3453 | if (ret && ret != -EEXIST) { | |
3454 | btrfs_abort_transaction(trans, ret); | |
3455 | return ret; | |
d68fc57b YZ |
3456 | } |
3457 | ||
d68fc57b | 3458 | return 0; |
7b128766 JB |
3459 | } |
3460 | ||
3461 | /* | |
f7e9e8fc OS |
3462 | * We have done the delete so we can go ahead and remove the orphan item for |
3463 | * this particular inode. | |
7b128766 | 3464 | */ |
48a3b636 | 3465 | static int btrfs_orphan_del(struct btrfs_trans_handle *trans, |
3d6ae7bb | 3466 | struct btrfs_inode *inode) |
7b128766 | 3467 | { |
27919067 | 3468 | return btrfs_del_orphan_item(trans, inode->root, btrfs_ino(inode)); |
7b128766 JB |
3469 | } |
3470 | ||
3471 | /* | |
3472 | * this cleans up any orphans that may be left on the list from the last use | |
3473 | * of this root. | |
3474 | */ | |
66b4ffd1 | 3475 | int btrfs_orphan_cleanup(struct btrfs_root *root) |
7b128766 | 3476 | { |
0b246afa | 3477 | struct btrfs_fs_info *fs_info = root->fs_info; |
7b128766 JB |
3478 | struct btrfs_path *path; |
3479 | struct extent_buffer *leaf; | |
7b128766 JB |
3480 | struct btrfs_key key, found_key; |
3481 | struct btrfs_trans_handle *trans; | |
3482 | struct inode *inode; | |
8f6d7f4f | 3483 | u64 last_objectid = 0; |
f7e9e8fc | 3484 | int ret = 0, nr_unlink = 0; |
7b128766 | 3485 | |
54230013 | 3486 | if (test_and_set_bit(BTRFS_ROOT_ORPHAN_CLEANUP, &root->state)) |
66b4ffd1 | 3487 | return 0; |
c71bf099 YZ |
3488 | |
3489 | path = btrfs_alloc_path(); | |
66b4ffd1 JB |
3490 | if (!path) { |
3491 | ret = -ENOMEM; | |
3492 | goto out; | |
3493 | } | |
e4058b54 | 3494 | path->reada = READA_BACK; |
7b128766 JB |
3495 | |
3496 | key.objectid = BTRFS_ORPHAN_OBJECTID; | |
962a298f | 3497 | key.type = BTRFS_ORPHAN_ITEM_KEY; |
7b128766 JB |
3498 | key.offset = (u64)-1; |
3499 | ||
7b128766 JB |
3500 | while (1) { |
3501 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
66b4ffd1 JB |
3502 | if (ret < 0) |
3503 | goto out; | |
7b128766 JB |
3504 | |
3505 | /* | |
3506 | * if ret == 0 means we found what we were searching for, which | |
25985edc | 3507 | * is weird, but possible, so only screw with path if we didn't |
7b128766 JB |
3508 | * find the key and see if we have stuff that matches |
3509 | */ | |
3510 | if (ret > 0) { | |
66b4ffd1 | 3511 | ret = 0; |
7b128766 JB |
3512 | if (path->slots[0] == 0) |
3513 | break; | |
3514 | path->slots[0]--; | |
3515 | } | |
3516 | ||
3517 | /* pull out the item */ | |
3518 | leaf = path->nodes[0]; | |
7b128766 JB |
3519 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
3520 | ||
3521 | /* make sure the item matches what we want */ | |
3522 | if (found_key.objectid != BTRFS_ORPHAN_OBJECTID) | |
3523 | break; | |
962a298f | 3524 | if (found_key.type != BTRFS_ORPHAN_ITEM_KEY) |
7b128766 JB |
3525 | break; |
3526 | ||
3527 | /* release the path since we're done with it */ | |
b3b4aa74 | 3528 | btrfs_release_path(path); |
7b128766 JB |
3529 | |
3530 | /* | |
3531 | * this is where we are basically btrfs_lookup, without the | |
3532 | * crossing root thing. we store the inode number in the | |
3533 | * offset of the orphan item. | |
3534 | */ | |
8f6d7f4f JB |
3535 | |
3536 | if (found_key.offset == last_objectid) { | |
0b246afa JM |
3537 | btrfs_err(fs_info, |
3538 | "Error removing orphan entry, stopping orphan cleanup"); | |
8f6d7f4f JB |
3539 | ret = -EINVAL; |
3540 | goto out; | |
3541 | } | |
3542 | ||
3543 | last_objectid = found_key.offset; | |
3544 | ||
5d4f98a2 YZ |
3545 | found_key.objectid = found_key.offset; |
3546 | found_key.type = BTRFS_INODE_ITEM_KEY; | |
3547 | found_key.offset = 0; | |
0202e83f | 3548 | inode = btrfs_iget(fs_info->sb, last_objectid, root); |
8c6ffba0 | 3549 | ret = PTR_ERR_OR_ZERO(inode); |
67710892 | 3550 | if (ret && ret != -ENOENT) |
66b4ffd1 | 3551 | goto out; |
7b128766 | 3552 | |
0b246afa | 3553 | if (ret == -ENOENT && root == fs_info->tree_root) { |
f8e9e0b0 | 3554 | struct btrfs_root *dead_root; |
f8e9e0b0 AJ |
3555 | int is_dead_root = 0; |
3556 | ||
3557 | /* | |
0c0218e9 | 3558 | * This is an orphan in the tree root. Currently these |
f8e9e0b0 | 3559 | * could come from 2 sources: |
0c0218e9 | 3560 | * a) a root (snapshot/subvolume) deletion in progress |
f8e9e0b0 | 3561 | * b) a free space cache inode |
0c0218e9 FM |
3562 | * We need to distinguish those two, as the orphan item |
3563 | * for a root must not get deleted before the deletion | |
3564 | * of the snapshot/subvolume's tree completes. | |
3565 | * | |
3566 | * btrfs_find_orphan_roots() ran before us, which has | |
3567 | * found all deleted roots and loaded them into | |
3568 | * fs_info->fs_roots_radix. So here we can find if an | |
3569 | * orphan item corresponds to a deleted root by looking | |
3570 | * up the root from that radix tree. | |
f8e9e0b0 | 3571 | */ |
a619b3c7 RK |
3572 | |
3573 | spin_lock(&fs_info->fs_roots_radix_lock); | |
3574 | dead_root = radix_tree_lookup(&fs_info->fs_roots_radix, | |
3575 | (unsigned long)found_key.objectid); | |
3576 | if (dead_root && btrfs_root_refs(&dead_root->root_item) == 0) | |
3577 | is_dead_root = 1; | |
3578 | spin_unlock(&fs_info->fs_roots_radix_lock); | |
3579 | ||
f8e9e0b0 AJ |
3580 | if (is_dead_root) { |
3581 | /* prevent this orphan from being found again */ | |
3582 | key.offset = found_key.objectid - 1; | |
3583 | continue; | |
3584 | } | |
f7e9e8fc | 3585 | |
f8e9e0b0 | 3586 | } |
f7e9e8fc | 3587 | |
7b128766 | 3588 | /* |
f7e9e8fc | 3589 | * If we have an inode with links, there are a couple of |
70524253 BB |
3590 | * possibilities: |
3591 | * | |
3592 | * 1. We were halfway through creating fsverity metadata for the | |
3593 | * file. In that case, the orphan item represents incomplete | |
3594 | * fsverity metadata which must be cleaned up with | |
3595 | * btrfs_drop_verity_items and deleting the orphan item. | |
3596 | ||
3597 | * 2. Old kernels (before v3.12) used to create an | |
f7e9e8fc OS |
3598 | * orphan item for truncate indicating that there were possibly |
3599 | * extent items past i_size that needed to be deleted. In v3.12, | |
3600 | * truncate was changed to update i_size in sync with the extent | |
3601 | * items, but the (useless) orphan item was still created. Since | |
3602 | * v4.18, we don't create the orphan item for truncate at all. | |
3603 | * | |
3604 | * So, this item could mean that we need to do a truncate, but | |
3605 | * only if this filesystem was last used on a pre-v3.12 kernel | |
3606 | * and was not cleanly unmounted. The odds of that are quite | |
3607 | * slim, and it's a pain to do the truncate now, so just delete | |
3608 | * the orphan item. | |
3609 | * | |
3610 | * It's also possible that this orphan item was supposed to be | |
3611 | * deleted but wasn't. The inode number may have been reused, | |
3612 | * but either way, we can delete the orphan item. | |
7b128766 | 3613 | */ |
f7e9e8fc | 3614 | if (ret == -ENOENT || inode->i_nlink) { |
70524253 BB |
3615 | if (!ret) { |
3616 | ret = btrfs_drop_verity_items(BTRFS_I(inode)); | |
f7e9e8fc | 3617 | iput(inode); |
70524253 BB |
3618 | if (ret) |
3619 | goto out; | |
3620 | } | |
a8c9e576 | 3621 | trans = btrfs_start_transaction(root, 1); |
66b4ffd1 JB |
3622 | if (IS_ERR(trans)) { |
3623 | ret = PTR_ERR(trans); | |
3624 | goto out; | |
3625 | } | |
0b246afa JM |
3626 | btrfs_debug(fs_info, "auto deleting %Lu", |
3627 | found_key.objectid); | |
a8c9e576 JB |
3628 | ret = btrfs_del_orphan_item(trans, root, |
3629 | found_key.objectid); | |
3a45bb20 | 3630 | btrfs_end_transaction(trans); |
4ef31a45 JB |
3631 | if (ret) |
3632 | goto out; | |
7b128766 JB |
3633 | continue; |
3634 | } | |
3635 | ||
f7e9e8fc | 3636 | nr_unlink++; |
7b128766 JB |
3637 | |
3638 | /* this will do delete_inode and everything for us */ | |
3639 | iput(inode); | |
3640 | } | |
3254c876 MX |
3641 | /* release the path since we're done with it */ |
3642 | btrfs_release_path(path); | |
3643 | ||
a575ceeb | 3644 | if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state)) { |
7a7eaa40 | 3645 | trans = btrfs_join_transaction(root); |
66b4ffd1 | 3646 | if (!IS_ERR(trans)) |
3a45bb20 | 3647 | btrfs_end_transaction(trans); |
d68fc57b | 3648 | } |
7b128766 JB |
3649 | |
3650 | if (nr_unlink) | |
0b246afa | 3651 | btrfs_debug(fs_info, "unlinked %d orphans", nr_unlink); |
66b4ffd1 JB |
3652 | |
3653 | out: | |
3654 | if (ret) | |
0b246afa | 3655 | btrfs_err(fs_info, "could not do orphan cleanup %d", ret); |
66b4ffd1 JB |
3656 | btrfs_free_path(path); |
3657 | return ret; | |
7b128766 JB |
3658 | } |
3659 | ||
46a53cca CM |
3660 | /* |
3661 | * very simple check to peek ahead in the leaf looking for xattrs. If we | |
3662 | * don't find any xattrs, we know there can't be any acls. | |
3663 | * | |
3664 | * slot is the slot the inode is in, objectid is the objectid of the inode | |
3665 | */ | |
3666 | static noinline int acls_after_inode_item(struct extent_buffer *leaf, | |
63541927 FDBM |
3667 | int slot, u64 objectid, |
3668 | int *first_xattr_slot) | |
46a53cca CM |
3669 | { |
3670 | u32 nritems = btrfs_header_nritems(leaf); | |
3671 | struct btrfs_key found_key; | |
f23b5a59 JB |
3672 | static u64 xattr_access = 0; |
3673 | static u64 xattr_default = 0; | |
46a53cca CM |
3674 | int scanned = 0; |
3675 | ||
f23b5a59 | 3676 | if (!xattr_access) { |
97d79299 AG |
3677 | xattr_access = btrfs_name_hash(XATTR_NAME_POSIX_ACL_ACCESS, |
3678 | strlen(XATTR_NAME_POSIX_ACL_ACCESS)); | |
3679 | xattr_default = btrfs_name_hash(XATTR_NAME_POSIX_ACL_DEFAULT, | |
3680 | strlen(XATTR_NAME_POSIX_ACL_DEFAULT)); | |
f23b5a59 JB |
3681 | } |
3682 | ||
46a53cca | 3683 | slot++; |
63541927 | 3684 | *first_xattr_slot = -1; |
46a53cca CM |
3685 | while (slot < nritems) { |
3686 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
3687 | ||
3688 | /* we found a different objectid, there must not be acls */ | |
3689 | if (found_key.objectid != objectid) | |
3690 | return 0; | |
3691 | ||
3692 | /* we found an xattr, assume we've got an acl */ | |
f23b5a59 | 3693 | if (found_key.type == BTRFS_XATTR_ITEM_KEY) { |
63541927 FDBM |
3694 | if (*first_xattr_slot == -1) |
3695 | *first_xattr_slot = slot; | |
f23b5a59 JB |
3696 | if (found_key.offset == xattr_access || |
3697 | found_key.offset == xattr_default) | |
3698 | return 1; | |
3699 | } | |
46a53cca CM |
3700 | |
3701 | /* | |
3702 | * we found a key greater than an xattr key, there can't | |
3703 | * be any acls later on | |
3704 | */ | |
3705 | if (found_key.type > BTRFS_XATTR_ITEM_KEY) | |
3706 | return 0; | |
3707 | ||
3708 | slot++; | |
3709 | scanned++; | |
3710 | ||
3711 | /* | |
3712 | * it goes inode, inode backrefs, xattrs, extents, | |
3713 | * so if there are a ton of hard links to an inode there can | |
3714 | * be a lot of backrefs. Don't waste time searching too hard, | |
3715 | * this is just an optimization | |
3716 | */ | |
3717 | if (scanned >= 8) | |
3718 | break; | |
3719 | } | |
3720 | /* we hit the end of the leaf before we found an xattr or | |
3721 | * something larger than an xattr. We have to assume the inode | |
3722 | * has acls | |
3723 | */ | |
63541927 FDBM |
3724 | if (*first_xattr_slot == -1) |
3725 | *first_xattr_slot = slot; | |
46a53cca CM |
3726 | return 1; |
3727 | } | |
3728 | ||
d352ac68 CM |
3729 | /* |
3730 | * read an inode from the btree into the in-memory inode | |
3731 | */ | |
4222ea71 FM |
3732 | static int btrfs_read_locked_inode(struct inode *inode, |
3733 | struct btrfs_path *in_path) | |
39279cc3 | 3734 | { |
0b246afa | 3735 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
4222ea71 | 3736 | struct btrfs_path *path = in_path; |
5f39d397 | 3737 | struct extent_buffer *leaf; |
39279cc3 CM |
3738 | struct btrfs_inode_item *inode_item; |
3739 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
3740 | struct btrfs_key location; | |
67de1176 | 3741 | unsigned long ptr; |
46a53cca | 3742 | int maybe_acls; |
618e21d5 | 3743 | u32 rdev; |
39279cc3 | 3744 | int ret; |
2f7e33d4 | 3745 | bool filled = false; |
63541927 | 3746 | int first_xattr_slot; |
2f7e33d4 MX |
3747 | |
3748 | ret = btrfs_fill_inode(inode, &rdev); | |
3749 | if (!ret) | |
3750 | filled = true; | |
39279cc3 | 3751 | |
4222ea71 FM |
3752 | if (!path) { |
3753 | path = btrfs_alloc_path(); | |
3754 | if (!path) | |
3755 | return -ENOMEM; | |
3756 | } | |
1748f843 | 3757 | |
39279cc3 | 3758 | memcpy(&location, &BTRFS_I(inode)->location, sizeof(location)); |
dc17ff8f | 3759 | |
39279cc3 | 3760 | ret = btrfs_lookup_inode(NULL, root, path, &location, 0); |
67710892 | 3761 | if (ret) { |
4222ea71 FM |
3762 | if (path != in_path) |
3763 | btrfs_free_path(path); | |
f5b3a417 | 3764 | return ret; |
67710892 | 3765 | } |
39279cc3 | 3766 | |
5f39d397 | 3767 | leaf = path->nodes[0]; |
2f7e33d4 MX |
3768 | |
3769 | if (filled) | |
67de1176 | 3770 | goto cache_index; |
2f7e33d4 | 3771 | |
5f39d397 CM |
3772 | inode_item = btrfs_item_ptr(leaf, path->slots[0], |
3773 | struct btrfs_inode_item); | |
5f39d397 | 3774 | inode->i_mode = btrfs_inode_mode(leaf, inode_item); |
bfe86848 | 3775 | set_nlink(inode, btrfs_inode_nlink(leaf, inode_item)); |
2f2f43d3 EB |
3776 | i_uid_write(inode, btrfs_inode_uid(leaf, inode_item)); |
3777 | i_gid_write(inode, btrfs_inode_gid(leaf, inode_item)); | |
6ef06d27 | 3778 | btrfs_i_size_write(BTRFS_I(inode), btrfs_inode_size(leaf, inode_item)); |
41a2ee75 JB |
3779 | btrfs_inode_set_file_extent_range(BTRFS_I(inode), 0, |
3780 | round_up(i_size_read(inode), fs_info->sectorsize)); | |
5f39d397 | 3781 | |
a937b979 DS |
3782 | inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->atime); |
3783 | inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->atime); | |
5f39d397 | 3784 | |
a937b979 DS |
3785 | inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->mtime); |
3786 | inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->mtime); | |
5f39d397 | 3787 | |
a937b979 DS |
3788 | inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->ctime); |
3789 | inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->ctime); | |
5f39d397 | 3790 | |
9cc97d64 | 3791 | BTRFS_I(inode)->i_otime.tv_sec = |
3792 | btrfs_timespec_sec(leaf, &inode_item->otime); | |
3793 | BTRFS_I(inode)->i_otime.tv_nsec = | |
3794 | btrfs_timespec_nsec(leaf, &inode_item->otime); | |
5f39d397 | 3795 | |
a76a3cd4 | 3796 | inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item)); |
e02119d5 | 3797 | BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item); |
5dc562c5 JB |
3798 | BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item); |
3799 | ||
c7f88c4e JL |
3800 | inode_set_iversion_queried(inode, |
3801 | btrfs_inode_sequence(leaf, inode_item)); | |
6e17d30b YD |
3802 | inode->i_generation = BTRFS_I(inode)->generation; |
3803 | inode->i_rdev = 0; | |
3804 | rdev = btrfs_inode_rdev(leaf, inode_item); | |
3805 | ||
3806 | BTRFS_I(inode)->index_cnt = (u64)-1; | |
77eea05e BB |
3807 | btrfs_inode_split_flags(btrfs_inode_flags(leaf, inode_item), |
3808 | &BTRFS_I(inode)->flags, &BTRFS_I(inode)->ro_flags); | |
6e17d30b YD |
3809 | |
3810 | cache_index: | |
5dc562c5 JB |
3811 | /* |
3812 | * If we were modified in the current generation and evicted from memory | |
3813 | * and then re-read we need to do a full sync since we don't have any | |
3814 | * idea about which extents were modified before we were evicted from | |
3815 | * cache. | |
6e17d30b YD |
3816 | * |
3817 | * This is required for both inode re-read from disk and delayed inode | |
3818 | * in delayed_nodes_tree. | |
5dc562c5 | 3819 | */ |
0b246afa | 3820 | if (BTRFS_I(inode)->last_trans == fs_info->generation) |
5dc562c5 JB |
3821 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, |
3822 | &BTRFS_I(inode)->runtime_flags); | |
3823 | ||
bde6c242 FM |
3824 | /* |
3825 | * We don't persist the id of the transaction where an unlink operation | |
3826 | * against the inode was last made. So here we assume the inode might | |
3827 | * have been evicted, and therefore the exact value of last_unlink_trans | |
3828 | * lost, and set it to last_trans to avoid metadata inconsistencies | |
3829 | * between the inode and its parent if the inode is fsync'ed and the log | |
3830 | * replayed. For example, in the scenario: | |
3831 | * | |
3832 | * touch mydir/foo | |
3833 | * ln mydir/foo mydir/bar | |
3834 | * sync | |
3835 | * unlink mydir/bar | |
3836 | * echo 2 > /proc/sys/vm/drop_caches # evicts inode | |
3837 | * xfs_io -c fsync mydir/foo | |
3838 | * <power failure> | |
3839 | * mount fs, triggers fsync log replay | |
3840 | * | |
3841 | * We must make sure that when we fsync our inode foo we also log its | |
3842 | * parent inode, otherwise after log replay the parent still has the | |
3843 | * dentry with the "bar" name but our inode foo has a link count of 1 | |
3844 | * and doesn't have an inode ref with the name "bar" anymore. | |
3845 | * | |
3846 | * Setting last_unlink_trans to last_trans is a pessimistic approach, | |
01327610 | 3847 | * but it guarantees correctness at the expense of occasional full |
bde6c242 FM |
3848 | * transaction commits on fsync if our inode is a directory, or if our |
3849 | * inode is not a directory, logging its parent unnecessarily. | |
3850 | */ | |
3851 | BTRFS_I(inode)->last_unlink_trans = BTRFS_I(inode)->last_trans; | |
3852 | ||
3ebac17c FM |
3853 | /* |
3854 | * Same logic as for last_unlink_trans. We don't persist the generation | |
3855 | * of the last transaction where this inode was used for a reflink | |
3856 | * operation, so after eviction and reloading the inode we must be | |
3857 | * pessimistic and assume the last transaction that modified the inode. | |
3858 | */ | |
3859 | BTRFS_I(inode)->last_reflink_trans = BTRFS_I(inode)->last_trans; | |
3860 | ||
67de1176 MX |
3861 | path->slots[0]++; |
3862 | if (inode->i_nlink != 1 || | |
3863 | path->slots[0] >= btrfs_header_nritems(leaf)) | |
3864 | goto cache_acl; | |
3865 | ||
3866 | btrfs_item_key_to_cpu(leaf, &location, path->slots[0]); | |
4a0cc7ca | 3867 | if (location.objectid != btrfs_ino(BTRFS_I(inode))) |
67de1176 MX |
3868 | goto cache_acl; |
3869 | ||
3870 | ptr = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
3871 | if (location.type == BTRFS_INODE_REF_KEY) { | |
3872 | struct btrfs_inode_ref *ref; | |
3873 | ||
3874 | ref = (struct btrfs_inode_ref *)ptr; | |
3875 | BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref); | |
3876 | } else if (location.type == BTRFS_INODE_EXTREF_KEY) { | |
3877 | struct btrfs_inode_extref *extref; | |
3878 | ||
3879 | extref = (struct btrfs_inode_extref *)ptr; | |
3880 | BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf, | |
3881 | extref); | |
3882 | } | |
2f7e33d4 | 3883 | cache_acl: |
46a53cca CM |
3884 | /* |
3885 | * try to precache a NULL acl entry for files that don't have | |
3886 | * any xattrs or acls | |
3887 | */ | |
33345d01 | 3888 | maybe_acls = acls_after_inode_item(leaf, path->slots[0], |
f85b7379 | 3889 | btrfs_ino(BTRFS_I(inode)), &first_xattr_slot); |
63541927 FDBM |
3890 | if (first_xattr_slot != -1) { |
3891 | path->slots[0] = first_xattr_slot; | |
3892 | ret = btrfs_load_inode_props(inode, path); | |
3893 | if (ret) | |
0b246afa | 3894 | btrfs_err(fs_info, |
351fd353 | 3895 | "error loading props for ino %llu (root %llu): %d", |
4a0cc7ca | 3896 | btrfs_ino(BTRFS_I(inode)), |
63541927 FDBM |
3897 | root->root_key.objectid, ret); |
3898 | } | |
4222ea71 FM |
3899 | if (path != in_path) |
3900 | btrfs_free_path(path); | |
63541927 | 3901 | |
72c04902 AV |
3902 | if (!maybe_acls) |
3903 | cache_no_acl(inode); | |
46a53cca | 3904 | |
39279cc3 | 3905 | switch (inode->i_mode & S_IFMT) { |
39279cc3 CM |
3906 | case S_IFREG: |
3907 | inode->i_mapping->a_ops = &btrfs_aops; | |
3908 | inode->i_fop = &btrfs_file_operations; | |
3909 | inode->i_op = &btrfs_file_inode_operations; | |
3910 | break; | |
3911 | case S_IFDIR: | |
3912 | inode->i_fop = &btrfs_dir_file_operations; | |
67ade058 | 3913 | inode->i_op = &btrfs_dir_inode_operations; |
39279cc3 CM |
3914 | break; |
3915 | case S_IFLNK: | |
3916 | inode->i_op = &btrfs_symlink_inode_operations; | |
21fc61c7 | 3917 | inode_nohighmem(inode); |
4779cc04 | 3918 | inode->i_mapping->a_ops = &btrfs_aops; |
39279cc3 | 3919 | break; |
618e21d5 | 3920 | default: |
0279b4cd | 3921 | inode->i_op = &btrfs_special_inode_operations; |
618e21d5 JB |
3922 | init_special_inode(inode, inode->i_mode, rdev); |
3923 | break; | |
39279cc3 | 3924 | } |
6cbff00f | 3925 | |
7b6a221e | 3926 | btrfs_sync_inode_flags_to_i_flags(inode); |
67710892 | 3927 | return 0; |
39279cc3 CM |
3928 | } |
3929 | ||
d352ac68 CM |
3930 | /* |
3931 | * given a leaf and an inode, copy the inode fields into the leaf | |
3932 | */ | |
e02119d5 CM |
3933 | static void fill_inode_item(struct btrfs_trans_handle *trans, |
3934 | struct extent_buffer *leaf, | |
5f39d397 | 3935 | struct btrfs_inode_item *item, |
39279cc3 CM |
3936 | struct inode *inode) |
3937 | { | |
51fab693 | 3938 | struct btrfs_map_token token; |
77eea05e | 3939 | u64 flags; |
51fab693 | 3940 | |
c82f823c | 3941 | btrfs_init_map_token(&token, leaf); |
5f39d397 | 3942 | |
cc4c13d5 DS |
3943 | btrfs_set_token_inode_uid(&token, item, i_uid_read(inode)); |
3944 | btrfs_set_token_inode_gid(&token, item, i_gid_read(inode)); | |
3945 | btrfs_set_token_inode_size(&token, item, BTRFS_I(inode)->disk_i_size); | |
3946 | btrfs_set_token_inode_mode(&token, item, inode->i_mode); | |
3947 | btrfs_set_token_inode_nlink(&token, item, inode->i_nlink); | |
3948 | ||
3949 | btrfs_set_token_timespec_sec(&token, &item->atime, | |
3950 | inode->i_atime.tv_sec); | |
3951 | btrfs_set_token_timespec_nsec(&token, &item->atime, | |
3952 | inode->i_atime.tv_nsec); | |
3953 | ||
3954 | btrfs_set_token_timespec_sec(&token, &item->mtime, | |
3955 | inode->i_mtime.tv_sec); | |
3956 | btrfs_set_token_timespec_nsec(&token, &item->mtime, | |
3957 | inode->i_mtime.tv_nsec); | |
3958 | ||
3959 | btrfs_set_token_timespec_sec(&token, &item->ctime, | |
3960 | inode->i_ctime.tv_sec); | |
3961 | btrfs_set_token_timespec_nsec(&token, &item->ctime, | |
3962 | inode->i_ctime.tv_nsec); | |
3963 | ||
3964 | btrfs_set_token_timespec_sec(&token, &item->otime, | |
3965 | BTRFS_I(inode)->i_otime.tv_sec); | |
3966 | btrfs_set_token_timespec_nsec(&token, &item->otime, | |
3967 | BTRFS_I(inode)->i_otime.tv_nsec); | |
3968 | ||
3969 | btrfs_set_token_inode_nbytes(&token, item, inode_get_bytes(inode)); | |
3970 | btrfs_set_token_inode_generation(&token, item, | |
3971 | BTRFS_I(inode)->generation); | |
3972 | btrfs_set_token_inode_sequence(&token, item, inode_peek_iversion(inode)); | |
3973 | btrfs_set_token_inode_transid(&token, item, trans->transid); | |
3974 | btrfs_set_token_inode_rdev(&token, item, inode->i_rdev); | |
77eea05e BB |
3975 | flags = btrfs_inode_combine_flags(BTRFS_I(inode)->flags, |
3976 | BTRFS_I(inode)->ro_flags); | |
3977 | btrfs_set_token_inode_flags(&token, item, flags); | |
cc4c13d5 | 3978 | btrfs_set_token_inode_block_group(&token, item, 0); |
39279cc3 CM |
3979 | } |
3980 | ||
d352ac68 CM |
3981 | /* |
3982 | * copy everything in the in-memory inode into the btree. | |
3983 | */ | |
2115133f | 3984 | static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans, |
dfeb9e7c NB |
3985 | struct btrfs_root *root, |
3986 | struct btrfs_inode *inode) | |
39279cc3 CM |
3987 | { |
3988 | struct btrfs_inode_item *inode_item; | |
3989 | struct btrfs_path *path; | |
5f39d397 | 3990 | struct extent_buffer *leaf; |
39279cc3 CM |
3991 | int ret; |
3992 | ||
3993 | path = btrfs_alloc_path(); | |
16cdcec7 MX |
3994 | if (!path) |
3995 | return -ENOMEM; | |
3996 | ||
dfeb9e7c | 3997 | ret = btrfs_lookup_inode(trans, root, path, &inode->location, 1); |
39279cc3 CM |
3998 | if (ret) { |
3999 | if (ret > 0) | |
4000 | ret = -ENOENT; | |
4001 | goto failed; | |
4002 | } | |
4003 | ||
5f39d397 CM |
4004 | leaf = path->nodes[0]; |
4005 | inode_item = btrfs_item_ptr(leaf, path->slots[0], | |
16cdcec7 | 4006 | struct btrfs_inode_item); |
39279cc3 | 4007 | |
dfeb9e7c | 4008 | fill_inode_item(trans, leaf, inode_item, &inode->vfs_inode); |
5f39d397 | 4009 | btrfs_mark_buffer_dirty(leaf); |
dfeb9e7c | 4010 | btrfs_set_inode_last_trans(trans, inode); |
39279cc3 CM |
4011 | ret = 0; |
4012 | failed: | |
39279cc3 CM |
4013 | btrfs_free_path(path); |
4014 | return ret; | |
4015 | } | |
4016 | ||
2115133f CM |
4017 | /* |
4018 | * copy everything in the in-memory inode into the btree. | |
4019 | */ | |
4020 | noinline int btrfs_update_inode(struct btrfs_trans_handle *trans, | |
9a56fcd1 NB |
4021 | struct btrfs_root *root, |
4022 | struct btrfs_inode *inode) | |
2115133f | 4023 | { |
0b246afa | 4024 | struct btrfs_fs_info *fs_info = root->fs_info; |
2115133f CM |
4025 | int ret; |
4026 | ||
4027 | /* | |
4028 | * If the inode is a free space inode, we can deadlock during commit | |
4029 | * if we put it into the delayed code. | |
4030 | * | |
4031 | * The data relocation inode should also be directly updated | |
4032 | * without delay | |
4033 | */ | |
9a56fcd1 | 4034 | if (!btrfs_is_free_space_inode(inode) |
37f00a6d | 4035 | && !btrfs_is_data_reloc_root(root) |
0b246afa | 4036 | && !test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) { |
8ea05e3a AB |
4037 | btrfs_update_root_times(trans, root); |
4038 | ||
9a56fcd1 | 4039 | ret = btrfs_delayed_update_inode(trans, root, inode); |
2115133f | 4040 | if (!ret) |
9a56fcd1 | 4041 | btrfs_set_inode_last_trans(trans, inode); |
2115133f CM |
4042 | return ret; |
4043 | } | |
4044 | ||
9a56fcd1 | 4045 | return btrfs_update_inode_item(trans, root, inode); |
2115133f CM |
4046 | } |
4047 | ||
729f7961 NB |
4048 | int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans, |
4049 | struct btrfs_root *root, struct btrfs_inode *inode) | |
2115133f CM |
4050 | { |
4051 | int ret; | |
4052 | ||
729f7961 | 4053 | ret = btrfs_update_inode(trans, root, inode); |
2115133f | 4054 | if (ret == -ENOSPC) |
729f7961 | 4055 | return btrfs_update_inode_item(trans, root, inode); |
2115133f CM |
4056 | return ret; |
4057 | } | |
4058 | ||
d352ac68 CM |
4059 | /* |
4060 | * unlink helper that gets used here in inode.c and in the tree logging | |
4061 | * recovery code. It remove a link in a directory with a given name, and | |
4062 | * also drops the back refs in the inode to the directory | |
4063 | */ | |
92986796 | 4064 | static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans, |
4ec5934e NB |
4065 | struct btrfs_inode *dir, |
4066 | struct btrfs_inode *inode, | |
88d2beec FM |
4067 | const char *name, int name_len, |
4068 | struct btrfs_rename_ctx *rename_ctx) | |
39279cc3 | 4069 | { |
4467af88 | 4070 | struct btrfs_root *root = dir->root; |
0b246afa | 4071 | struct btrfs_fs_info *fs_info = root->fs_info; |
39279cc3 | 4072 | struct btrfs_path *path; |
39279cc3 | 4073 | int ret = 0; |
39279cc3 | 4074 | struct btrfs_dir_item *di; |
aec7477b | 4075 | u64 index; |
33345d01 LZ |
4076 | u64 ino = btrfs_ino(inode); |
4077 | u64 dir_ino = btrfs_ino(dir); | |
39279cc3 CM |
4078 | |
4079 | path = btrfs_alloc_path(); | |
54aa1f4d CM |
4080 | if (!path) { |
4081 | ret = -ENOMEM; | |
554233a6 | 4082 | goto out; |
54aa1f4d CM |
4083 | } |
4084 | ||
33345d01 | 4085 | di = btrfs_lookup_dir_item(trans, root, path, dir_ino, |
39279cc3 | 4086 | name, name_len, -1); |
3cf5068f LB |
4087 | if (IS_ERR_OR_NULL(di)) { |
4088 | ret = di ? PTR_ERR(di) : -ENOENT; | |
39279cc3 CM |
4089 | goto err; |
4090 | } | |
39279cc3 | 4091 | ret = btrfs_delete_one_dir_name(trans, root, path, di); |
54aa1f4d CM |
4092 | if (ret) |
4093 | goto err; | |
b3b4aa74 | 4094 | btrfs_release_path(path); |
39279cc3 | 4095 | |
67de1176 MX |
4096 | /* |
4097 | * If we don't have dir index, we have to get it by looking up | |
4098 | * the inode ref, since we get the inode ref, remove it directly, | |
4099 | * it is unnecessary to do delayed deletion. | |
4100 | * | |
4101 | * But if we have dir index, needn't search inode ref to get it. | |
4102 | * Since the inode ref is close to the inode item, it is better | |
4103 | * that we delay to delete it, and just do this deletion when | |
4104 | * we update the inode item. | |
4105 | */ | |
4ec5934e | 4106 | if (inode->dir_index) { |
67de1176 MX |
4107 | ret = btrfs_delayed_delete_inode_ref(inode); |
4108 | if (!ret) { | |
4ec5934e | 4109 | index = inode->dir_index; |
67de1176 MX |
4110 | goto skip_backref; |
4111 | } | |
4112 | } | |
4113 | ||
33345d01 LZ |
4114 | ret = btrfs_del_inode_ref(trans, root, name, name_len, ino, |
4115 | dir_ino, &index); | |
aec7477b | 4116 | if (ret) { |
0b246afa | 4117 | btrfs_info(fs_info, |
c2cf52eb | 4118 | "failed to delete reference to %.*s, inode %llu parent %llu", |
c1c9ff7c | 4119 | name_len, name, ino, dir_ino); |
66642832 | 4120 | btrfs_abort_transaction(trans, ret); |
aec7477b JB |
4121 | goto err; |
4122 | } | |
67de1176 | 4123 | skip_backref: |
88d2beec FM |
4124 | if (rename_ctx) |
4125 | rename_ctx->index = index; | |
4126 | ||
9add2945 | 4127 | ret = btrfs_delete_delayed_dir_index(trans, dir, index); |
79787eaa | 4128 | if (ret) { |
66642832 | 4129 | btrfs_abort_transaction(trans, ret); |
39279cc3 | 4130 | goto err; |
79787eaa | 4131 | } |
39279cc3 | 4132 | |
259c4b96 FM |
4133 | /* |
4134 | * If we are in a rename context, we don't need to update anything in the | |
4135 | * log. That will be done later during the rename by btrfs_log_new_name(). | |
4136 | * Besides that, doing it here would only cause extra unncessary btree | |
4137 | * operations on the log tree, increasing latency for applications. | |
4138 | */ | |
4139 | if (!rename_ctx) { | |
4140 | btrfs_del_inode_ref_in_log(trans, root, name, name_len, inode, | |
4141 | dir_ino); | |
4142 | btrfs_del_dir_entries_in_log(trans, root, name, name_len, dir, | |
4143 | index); | |
4144 | } | |
63611e73 JB |
4145 | |
4146 | /* | |
4147 | * If we have a pending delayed iput we could end up with the final iput | |
4148 | * being run in btrfs-cleaner context. If we have enough of these built | |
4149 | * up we can end up burning a lot of time in btrfs-cleaner without any | |
4150 | * way to throttle the unlinks. Since we're currently holding a ref on | |
4151 | * the inode we can run the delayed iput here without any issues as the | |
4152 | * final iput won't be done until after we drop the ref we're currently | |
4153 | * holding. | |
4154 | */ | |
4155 | btrfs_run_delayed_iput(fs_info, inode); | |
39279cc3 CM |
4156 | err: |
4157 | btrfs_free_path(path); | |
e02119d5 CM |
4158 | if (ret) |
4159 | goto out; | |
4160 | ||
6ef06d27 | 4161 | btrfs_i_size_write(dir, dir->vfs_inode.i_size - name_len * 2); |
4ec5934e NB |
4162 | inode_inc_iversion(&inode->vfs_inode); |
4163 | inode_inc_iversion(&dir->vfs_inode); | |
4164 | inode->vfs_inode.i_ctime = dir->vfs_inode.i_mtime = | |
4165 | dir->vfs_inode.i_ctime = current_time(&inode->vfs_inode); | |
9a56fcd1 | 4166 | ret = btrfs_update_inode(trans, root, dir); |
e02119d5 | 4167 | out: |
39279cc3 CM |
4168 | return ret; |
4169 | } | |
4170 | ||
92986796 | 4171 | int btrfs_unlink_inode(struct btrfs_trans_handle *trans, |
4ec5934e | 4172 | struct btrfs_inode *dir, struct btrfs_inode *inode, |
92986796 AV |
4173 | const char *name, int name_len) |
4174 | { | |
4175 | int ret; | |
88d2beec | 4176 | ret = __btrfs_unlink_inode(trans, dir, inode, name, name_len, NULL); |
92986796 | 4177 | if (!ret) { |
4ec5934e | 4178 | drop_nlink(&inode->vfs_inode); |
4467af88 | 4179 | ret = btrfs_update_inode(trans, inode->root, inode); |
92986796 AV |
4180 | } |
4181 | return ret; | |
4182 | } | |
39279cc3 | 4183 | |
a22285a6 YZ |
4184 | /* |
4185 | * helper to start transaction for unlink and rmdir. | |
4186 | * | |
d52be818 JB |
4187 | * unlink and rmdir are special in btrfs, they do not always free space, so |
4188 | * if we cannot make our reservations the normal way try and see if there is | |
4189 | * plenty of slack room in the global reserve to migrate, otherwise we cannot | |
4190 | * allow the unlink to occur. | |
a22285a6 | 4191 | */ |
d52be818 | 4192 | static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir) |
4df27c4d | 4193 | { |
a22285a6 | 4194 | struct btrfs_root *root = BTRFS_I(dir)->root; |
4df27c4d | 4195 | |
e70bea5f JB |
4196 | /* |
4197 | * 1 for the possible orphan item | |
4198 | * 1 for the dir item | |
4199 | * 1 for the dir index | |
4200 | * 1 for the inode ref | |
e70bea5f | 4201 | * 1 for the inode |
bca4ad7c | 4202 | * 1 for the parent inode |
e70bea5f | 4203 | */ |
bca4ad7c | 4204 | return btrfs_start_transaction_fallback_global_rsv(root, 6); |
a22285a6 YZ |
4205 | } |
4206 | ||
4207 | static int btrfs_unlink(struct inode *dir, struct dentry *dentry) | |
4208 | { | |
a22285a6 | 4209 | struct btrfs_trans_handle *trans; |
2b0143b5 | 4210 | struct inode *inode = d_inode(dentry); |
a22285a6 | 4211 | int ret; |
a22285a6 | 4212 | |
d52be818 | 4213 | trans = __unlink_start_trans(dir); |
a22285a6 YZ |
4214 | if (IS_ERR(trans)) |
4215 | return PTR_ERR(trans); | |
5f39d397 | 4216 | |
4ec5934e NB |
4217 | btrfs_record_unlink_dir(trans, BTRFS_I(dir), BTRFS_I(d_inode(dentry)), |
4218 | 0); | |
12fcfd22 | 4219 | |
4467af88 | 4220 | ret = btrfs_unlink_inode(trans, BTRFS_I(dir), |
4ec5934e NB |
4221 | BTRFS_I(d_inode(dentry)), dentry->d_name.name, |
4222 | dentry->d_name.len); | |
b532402e TI |
4223 | if (ret) |
4224 | goto out; | |
7b128766 | 4225 | |
a22285a6 | 4226 | if (inode->i_nlink == 0) { |
73f2e545 | 4227 | ret = btrfs_orphan_add(trans, BTRFS_I(inode)); |
b532402e TI |
4228 | if (ret) |
4229 | goto out; | |
a22285a6 | 4230 | } |
7b128766 | 4231 | |
b532402e | 4232 | out: |
3a45bb20 | 4233 | btrfs_end_transaction(trans); |
4467af88 | 4234 | btrfs_btree_balance_dirty(BTRFS_I(dir)->root->fs_info); |
39279cc3 CM |
4235 | return ret; |
4236 | } | |
4237 | ||
f60a2364 | 4238 | static int btrfs_unlink_subvol(struct btrfs_trans_handle *trans, |
045d3967 | 4239 | struct inode *dir, struct dentry *dentry) |
4df27c4d | 4240 | { |
401b3b19 | 4241 | struct btrfs_root *root = BTRFS_I(dir)->root; |
045d3967 | 4242 | struct btrfs_inode *inode = BTRFS_I(d_inode(dentry)); |
4df27c4d YZ |
4243 | struct btrfs_path *path; |
4244 | struct extent_buffer *leaf; | |
4245 | struct btrfs_dir_item *di; | |
4246 | struct btrfs_key key; | |
045d3967 JB |
4247 | const char *name = dentry->d_name.name; |
4248 | int name_len = dentry->d_name.len; | |
4df27c4d YZ |
4249 | u64 index; |
4250 | int ret; | |
045d3967 | 4251 | u64 objectid; |
4a0cc7ca | 4252 | u64 dir_ino = btrfs_ino(BTRFS_I(dir)); |
4df27c4d | 4253 | |
045d3967 JB |
4254 | if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID) { |
4255 | objectid = inode->root->root_key.objectid; | |
4256 | } else if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) { | |
4257 | objectid = inode->location.objectid; | |
4258 | } else { | |
4259 | WARN_ON(1); | |
4260 | return -EINVAL; | |
4261 | } | |
4262 | ||
4df27c4d YZ |
4263 | path = btrfs_alloc_path(); |
4264 | if (!path) | |
4265 | return -ENOMEM; | |
4266 | ||
33345d01 | 4267 | di = btrfs_lookup_dir_item(trans, root, path, dir_ino, |
4df27c4d | 4268 | name, name_len, -1); |
79787eaa | 4269 | if (IS_ERR_OR_NULL(di)) { |
3cf5068f | 4270 | ret = di ? PTR_ERR(di) : -ENOENT; |
79787eaa JM |
4271 | goto out; |
4272 | } | |
4df27c4d YZ |
4273 | |
4274 | leaf = path->nodes[0]; | |
4275 | btrfs_dir_item_key_to_cpu(leaf, di, &key); | |
4276 | WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid); | |
4277 | ret = btrfs_delete_one_dir_name(trans, root, path, di); | |
79787eaa | 4278 | if (ret) { |
66642832 | 4279 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
4280 | goto out; |
4281 | } | |
b3b4aa74 | 4282 | btrfs_release_path(path); |
4df27c4d | 4283 | |
d49d3287 JB |
4284 | /* |
4285 | * This is a placeholder inode for a subvolume we didn't have a | |
4286 | * reference to at the time of the snapshot creation. In the meantime | |
4287 | * we could have renamed the real subvol link into our snapshot, so | |
1a9fd417 | 4288 | * depending on btrfs_del_root_ref to return -ENOENT here is incorrect. |
d49d3287 JB |
4289 | * Instead simply lookup the dir_index_item for this entry so we can |
4290 | * remove it. Otherwise we know we have a ref to the root and we can | |
4291 | * call btrfs_del_root_ref, and it _shouldn't_ fail. | |
4292 | */ | |
4293 | if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) { | |
33345d01 | 4294 | di = btrfs_search_dir_index_item(root, path, dir_ino, |
4df27c4d | 4295 | name, name_len); |
79787eaa JM |
4296 | if (IS_ERR_OR_NULL(di)) { |
4297 | if (!di) | |
4298 | ret = -ENOENT; | |
4299 | else | |
4300 | ret = PTR_ERR(di); | |
66642832 | 4301 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
4302 | goto out; |
4303 | } | |
4df27c4d YZ |
4304 | |
4305 | leaf = path->nodes[0]; | |
4306 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
4df27c4d | 4307 | index = key.offset; |
d49d3287 JB |
4308 | btrfs_release_path(path); |
4309 | } else { | |
4310 | ret = btrfs_del_root_ref(trans, objectid, | |
4311 | root->root_key.objectid, dir_ino, | |
4312 | &index, name, name_len); | |
4313 | if (ret) { | |
4314 | btrfs_abort_transaction(trans, ret); | |
4315 | goto out; | |
4316 | } | |
4df27c4d YZ |
4317 | } |
4318 | ||
9add2945 | 4319 | ret = btrfs_delete_delayed_dir_index(trans, BTRFS_I(dir), index); |
79787eaa | 4320 | if (ret) { |
66642832 | 4321 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
4322 | goto out; |
4323 | } | |
4df27c4d | 4324 | |
6ef06d27 | 4325 | btrfs_i_size_write(BTRFS_I(dir), dir->i_size - name_len * 2); |
0c4d2d95 | 4326 | inode_inc_iversion(dir); |
c2050a45 | 4327 | dir->i_mtime = dir->i_ctime = current_time(dir); |
729f7961 | 4328 | ret = btrfs_update_inode_fallback(trans, root, BTRFS_I(dir)); |
79787eaa | 4329 | if (ret) |
66642832 | 4330 | btrfs_abort_transaction(trans, ret); |
79787eaa | 4331 | out: |
71d7aed0 | 4332 | btrfs_free_path(path); |
79787eaa | 4333 | return ret; |
4df27c4d YZ |
4334 | } |
4335 | ||
ec42f167 MT |
4336 | /* |
4337 | * Helper to check if the subvolume references other subvolumes or if it's | |
4338 | * default. | |
4339 | */ | |
f60a2364 | 4340 | static noinline int may_destroy_subvol(struct btrfs_root *root) |
ec42f167 MT |
4341 | { |
4342 | struct btrfs_fs_info *fs_info = root->fs_info; | |
4343 | struct btrfs_path *path; | |
4344 | struct btrfs_dir_item *di; | |
4345 | struct btrfs_key key; | |
4346 | u64 dir_id; | |
4347 | int ret; | |
4348 | ||
4349 | path = btrfs_alloc_path(); | |
4350 | if (!path) | |
4351 | return -ENOMEM; | |
4352 | ||
4353 | /* Make sure this root isn't set as the default subvol */ | |
4354 | dir_id = btrfs_super_root_dir(fs_info->super_copy); | |
4355 | di = btrfs_lookup_dir_item(NULL, fs_info->tree_root, path, | |
4356 | dir_id, "default", 7, 0); | |
4357 | if (di && !IS_ERR(di)) { | |
4358 | btrfs_dir_item_key_to_cpu(path->nodes[0], di, &key); | |
4359 | if (key.objectid == root->root_key.objectid) { | |
4360 | ret = -EPERM; | |
4361 | btrfs_err(fs_info, | |
4362 | "deleting default subvolume %llu is not allowed", | |
4363 | key.objectid); | |
4364 | goto out; | |
4365 | } | |
4366 | btrfs_release_path(path); | |
4367 | } | |
4368 | ||
4369 | key.objectid = root->root_key.objectid; | |
4370 | key.type = BTRFS_ROOT_REF_KEY; | |
4371 | key.offset = (u64)-1; | |
4372 | ||
4373 | ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0); | |
4374 | if (ret < 0) | |
4375 | goto out; | |
4376 | BUG_ON(ret == 0); | |
4377 | ||
4378 | ret = 0; | |
4379 | if (path->slots[0] > 0) { | |
4380 | path->slots[0]--; | |
4381 | btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]); | |
4382 | if (key.objectid == root->root_key.objectid && | |
4383 | key.type == BTRFS_ROOT_REF_KEY) | |
4384 | ret = -ENOTEMPTY; | |
4385 | } | |
4386 | out: | |
4387 | btrfs_free_path(path); | |
4388 | return ret; | |
4389 | } | |
4390 | ||
20a68004 NB |
4391 | /* Delete all dentries for inodes belonging to the root */ |
4392 | static void btrfs_prune_dentries(struct btrfs_root *root) | |
4393 | { | |
4394 | struct btrfs_fs_info *fs_info = root->fs_info; | |
4395 | struct rb_node *node; | |
4396 | struct rb_node *prev; | |
4397 | struct btrfs_inode *entry; | |
4398 | struct inode *inode; | |
4399 | u64 objectid = 0; | |
4400 | ||
84961539 | 4401 | if (!BTRFS_FS_ERROR(fs_info)) |
20a68004 NB |
4402 | WARN_ON(btrfs_root_refs(&root->root_item) != 0); |
4403 | ||
4404 | spin_lock(&root->inode_lock); | |
4405 | again: | |
4406 | node = root->inode_tree.rb_node; | |
4407 | prev = NULL; | |
4408 | while (node) { | |
4409 | prev = node; | |
4410 | entry = rb_entry(node, struct btrfs_inode, rb_node); | |
4411 | ||
37508515 | 4412 | if (objectid < btrfs_ino(entry)) |
20a68004 | 4413 | node = node->rb_left; |
37508515 | 4414 | else if (objectid > btrfs_ino(entry)) |
20a68004 NB |
4415 | node = node->rb_right; |
4416 | else | |
4417 | break; | |
4418 | } | |
4419 | if (!node) { | |
4420 | while (prev) { | |
4421 | entry = rb_entry(prev, struct btrfs_inode, rb_node); | |
37508515 | 4422 | if (objectid <= btrfs_ino(entry)) { |
20a68004 NB |
4423 | node = prev; |
4424 | break; | |
4425 | } | |
4426 | prev = rb_next(prev); | |
4427 | } | |
4428 | } | |
4429 | while (node) { | |
4430 | entry = rb_entry(node, struct btrfs_inode, rb_node); | |
37508515 | 4431 | objectid = btrfs_ino(entry) + 1; |
20a68004 NB |
4432 | inode = igrab(&entry->vfs_inode); |
4433 | if (inode) { | |
4434 | spin_unlock(&root->inode_lock); | |
4435 | if (atomic_read(&inode->i_count) > 1) | |
4436 | d_prune_aliases(inode); | |
4437 | /* | |
4438 | * btrfs_drop_inode will have it removed from the inode | |
4439 | * cache when its usage count hits zero. | |
4440 | */ | |
4441 | iput(inode); | |
4442 | cond_resched(); | |
4443 | spin_lock(&root->inode_lock); | |
4444 | goto again; | |
4445 | } | |
4446 | ||
4447 | if (cond_resched_lock(&root->inode_lock)) | |
4448 | goto again; | |
4449 | ||
4450 | node = rb_next(node); | |
4451 | } | |
4452 | spin_unlock(&root->inode_lock); | |
4453 | } | |
4454 | ||
f60a2364 MT |
4455 | int btrfs_delete_subvolume(struct inode *dir, struct dentry *dentry) |
4456 | { | |
4457 | struct btrfs_fs_info *fs_info = btrfs_sb(dentry->d_sb); | |
4458 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
4459 | struct inode *inode = d_inode(dentry); | |
4460 | struct btrfs_root *dest = BTRFS_I(inode)->root; | |
4461 | struct btrfs_trans_handle *trans; | |
4462 | struct btrfs_block_rsv block_rsv; | |
4463 | u64 root_flags; | |
f60a2364 | 4464 | int ret; |
f60a2364 MT |
4465 | |
4466 | /* | |
4467 | * Don't allow to delete a subvolume with send in progress. This is | |
4468 | * inside the inode lock so the error handling that has to drop the bit | |
4469 | * again is not run concurrently. | |
4470 | */ | |
4471 | spin_lock(&dest->root_item_lock); | |
a7176f74 | 4472 | if (dest->send_in_progress) { |
f60a2364 MT |
4473 | spin_unlock(&dest->root_item_lock); |
4474 | btrfs_warn(fs_info, | |
4475 | "attempt to delete subvolume %llu during send", | |
4476 | dest->root_key.objectid); | |
4477 | return -EPERM; | |
4478 | } | |
60021bd7 KH |
4479 | if (atomic_read(&dest->nr_swapfiles)) { |
4480 | spin_unlock(&dest->root_item_lock); | |
4481 | btrfs_warn(fs_info, | |
4482 | "attempt to delete subvolume %llu with active swapfile", | |
4483 | root->root_key.objectid); | |
4484 | return -EPERM; | |
4485 | } | |
a7176f74 LF |
4486 | root_flags = btrfs_root_flags(&dest->root_item); |
4487 | btrfs_set_root_flags(&dest->root_item, | |
4488 | root_flags | BTRFS_ROOT_SUBVOL_DEAD); | |
4489 | spin_unlock(&dest->root_item_lock); | |
f60a2364 MT |
4490 | |
4491 | down_write(&fs_info->subvol_sem); | |
4492 | ||
ee0d904f NB |
4493 | ret = may_destroy_subvol(dest); |
4494 | if (ret) | |
f60a2364 MT |
4495 | goto out_up_write; |
4496 | ||
4497 | btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP); | |
4498 | /* | |
4499 | * One for dir inode, | |
4500 | * two for dir entries, | |
4501 | * two for root ref/backref. | |
4502 | */ | |
ee0d904f NB |
4503 | ret = btrfs_subvolume_reserve_metadata(root, &block_rsv, 5, true); |
4504 | if (ret) | |
f60a2364 MT |
4505 | goto out_up_write; |
4506 | ||
4507 | trans = btrfs_start_transaction(root, 0); | |
4508 | if (IS_ERR(trans)) { | |
ee0d904f | 4509 | ret = PTR_ERR(trans); |
f60a2364 MT |
4510 | goto out_release; |
4511 | } | |
4512 | trans->block_rsv = &block_rsv; | |
4513 | trans->bytes_reserved = block_rsv.size; | |
4514 | ||
4515 | btrfs_record_snapshot_destroy(trans, BTRFS_I(dir)); | |
4516 | ||
045d3967 | 4517 | ret = btrfs_unlink_subvol(trans, dir, dentry); |
f60a2364 | 4518 | if (ret) { |
f60a2364 MT |
4519 | btrfs_abort_transaction(trans, ret); |
4520 | goto out_end_trans; | |
4521 | } | |
4522 | ||
2731f518 JB |
4523 | ret = btrfs_record_root_in_trans(trans, dest); |
4524 | if (ret) { | |
4525 | btrfs_abort_transaction(trans, ret); | |
4526 | goto out_end_trans; | |
4527 | } | |
f60a2364 MT |
4528 | |
4529 | memset(&dest->root_item.drop_progress, 0, | |
4530 | sizeof(dest->root_item.drop_progress)); | |
c8422684 | 4531 | btrfs_set_root_drop_level(&dest->root_item, 0); |
f60a2364 MT |
4532 | btrfs_set_root_refs(&dest->root_item, 0); |
4533 | ||
4534 | if (!test_and_set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &dest->state)) { | |
4535 | ret = btrfs_insert_orphan_item(trans, | |
4536 | fs_info->tree_root, | |
4537 | dest->root_key.objectid); | |
4538 | if (ret) { | |
4539 | btrfs_abort_transaction(trans, ret); | |
f60a2364 MT |
4540 | goto out_end_trans; |
4541 | } | |
4542 | } | |
4543 | ||
d1957791 | 4544 | ret = btrfs_uuid_tree_remove(trans, dest->root_item.uuid, |
f60a2364 MT |
4545 | BTRFS_UUID_KEY_SUBVOL, |
4546 | dest->root_key.objectid); | |
4547 | if (ret && ret != -ENOENT) { | |
4548 | btrfs_abort_transaction(trans, ret); | |
f60a2364 MT |
4549 | goto out_end_trans; |
4550 | } | |
4551 | if (!btrfs_is_empty_uuid(dest->root_item.received_uuid)) { | |
d1957791 | 4552 | ret = btrfs_uuid_tree_remove(trans, |
f60a2364 MT |
4553 | dest->root_item.received_uuid, |
4554 | BTRFS_UUID_KEY_RECEIVED_SUBVOL, | |
4555 | dest->root_key.objectid); | |
4556 | if (ret && ret != -ENOENT) { | |
4557 | btrfs_abort_transaction(trans, ret); | |
f60a2364 MT |
4558 | goto out_end_trans; |
4559 | } | |
4560 | } | |
4561 | ||
082b6c97 QW |
4562 | free_anon_bdev(dest->anon_dev); |
4563 | dest->anon_dev = 0; | |
f60a2364 MT |
4564 | out_end_trans: |
4565 | trans->block_rsv = NULL; | |
4566 | trans->bytes_reserved = 0; | |
4567 | ret = btrfs_end_transaction(trans); | |
f60a2364 MT |
4568 | inode->i_flags |= S_DEAD; |
4569 | out_release: | |
e85fde51 | 4570 | btrfs_subvolume_release_metadata(root, &block_rsv); |
f60a2364 MT |
4571 | out_up_write: |
4572 | up_write(&fs_info->subvol_sem); | |
ee0d904f | 4573 | if (ret) { |
f60a2364 MT |
4574 | spin_lock(&dest->root_item_lock); |
4575 | root_flags = btrfs_root_flags(&dest->root_item); | |
4576 | btrfs_set_root_flags(&dest->root_item, | |
4577 | root_flags & ~BTRFS_ROOT_SUBVOL_DEAD); | |
4578 | spin_unlock(&dest->root_item_lock); | |
4579 | } else { | |
4580 | d_invalidate(dentry); | |
20a68004 | 4581 | btrfs_prune_dentries(dest); |
f60a2364 | 4582 | ASSERT(dest->send_in_progress == 0); |
f60a2364 MT |
4583 | } |
4584 | ||
ee0d904f | 4585 | return ret; |
f60a2364 MT |
4586 | } |
4587 | ||
39279cc3 CM |
4588 | static int btrfs_rmdir(struct inode *dir, struct dentry *dentry) |
4589 | { | |
2b0143b5 | 4590 | struct inode *inode = d_inode(dentry); |
813febdb | 4591 | struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info; |
1832a6d5 | 4592 | int err = 0; |
39279cc3 | 4593 | struct btrfs_trans_handle *trans; |
44f714da | 4594 | u64 last_unlink_trans; |
39279cc3 | 4595 | |
b3ae244e | 4596 | if (inode->i_size > BTRFS_EMPTY_DIR_SIZE) |
134d4512 | 4597 | return -ENOTEMPTY; |
813febdb JB |
4598 | if (btrfs_ino(BTRFS_I(inode)) == BTRFS_FIRST_FREE_OBJECTID) { |
4599 | if (unlikely(btrfs_fs_incompat(fs_info, EXTENT_TREE_V2))) { | |
4600 | btrfs_err(fs_info, | |
4601 | "extent tree v2 doesn't support snapshot deletion yet"); | |
4602 | return -EOPNOTSUPP; | |
4603 | } | |
a79a464d | 4604 | return btrfs_delete_subvolume(dir, dentry); |
813febdb | 4605 | } |
134d4512 | 4606 | |
d52be818 | 4607 | trans = __unlink_start_trans(dir); |
a22285a6 | 4608 | if (IS_ERR(trans)) |
5df6a9f6 | 4609 | return PTR_ERR(trans); |
5df6a9f6 | 4610 | |
4a0cc7ca | 4611 | if (unlikely(btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) { |
045d3967 | 4612 | err = btrfs_unlink_subvol(trans, dir, dentry); |
4df27c4d YZ |
4613 | goto out; |
4614 | } | |
4615 | ||
73f2e545 | 4616 | err = btrfs_orphan_add(trans, BTRFS_I(inode)); |
7b128766 | 4617 | if (err) |
4df27c4d | 4618 | goto out; |
7b128766 | 4619 | |
44f714da FM |
4620 | last_unlink_trans = BTRFS_I(inode)->last_unlink_trans; |
4621 | ||
39279cc3 | 4622 | /* now the directory is empty */ |
4467af88 | 4623 | err = btrfs_unlink_inode(trans, BTRFS_I(dir), |
4ec5934e NB |
4624 | BTRFS_I(d_inode(dentry)), dentry->d_name.name, |
4625 | dentry->d_name.len); | |
44f714da | 4626 | if (!err) { |
6ef06d27 | 4627 | btrfs_i_size_write(BTRFS_I(inode), 0); |
44f714da FM |
4628 | /* |
4629 | * Propagate the last_unlink_trans value of the deleted dir to | |
4630 | * its parent directory. This is to prevent an unrecoverable | |
4631 | * log tree in the case we do something like this: | |
4632 | * 1) create dir foo | |
4633 | * 2) create snapshot under dir foo | |
4634 | * 3) delete the snapshot | |
4635 | * 4) rmdir foo | |
4636 | * 5) mkdir foo | |
4637 | * 6) fsync foo or some file inside foo | |
4638 | */ | |
4639 | if (last_unlink_trans >= trans->transid) | |
4640 | BTRFS_I(dir)->last_unlink_trans = last_unlink_trans; | |
4641 | } | |
4df27c4d | 4642 | out: |
3a45bb20 | 4643 | btrfs_end_transaction(trans); |
813febdb | 4644 | btrfs_btree_balance_dirty(fs_info); |
3954401f | 4645 | |
39279cc3 CM |
4646 | return err; |
4647 | } | |
4648 | ||
39279cc3 | 4649 | /* |
9703fefe | 4650 | * btrfs_truncate_block - read, zero a chunk and write a block |
2aaa6655 JB |
4651 | * @inode - inode that we're zeroing |
4652 | * @from - the offset to start zeroing | |
4653 | * @len - the length to zero, 0 to zero the entire range respective to the | |
4654 | * offset | |
4655 | * @front - zero up to the offset instead of from the offset on | |
4656 | * | |
9703fefe | 4657 | * This will find the block for the "from" offset and cow the block and zero the |
2aaa6655 | 4658 | * part we want to zero. This is used with truncate and hole punching. |
39279cc3 | 4659 | */ |
217f42eb NB |
4660 | int btrfs_truncate_block(struct btrfs_inode *inode, loff_t from, loff_t len, |
4661 | int front) | |
39279cc3 | 4662 | { |
217f42eb NB |
4663 | struct btrfs_fs_info *fs_info = inode->root->fs_info; |
4664 | struct address_space *mapping = inode->vfs_inode.i_mapping; | |
4665 | struct extent_io_tree *io_tree = &inode->io_tree; | |
e6dcd2dc | 4666 | struct btrfs_ordered_extent *ordered; |
2ac55d41 | 4667 | struct extent_state *cached_state = NULL; |
364ecf36 | 4668 | struct extent_changeset *data_reserved = NULL; |
6d4572a9 | 4669 | bool only_release_metadata = false; |
0b246afa | 4670 | u32 blocksize = fs_info->sectorsize; |
09cbfeaf | 4671 | pgoff_t index = from >> PAGE_SHIFT; |
9703fefe | 4672 | unsigned offset = from & (blocksize - 1); |
39279cc3 | 4673 | struct page *page; |
3b16a4e3 | 4674 | gfp_t mask = btrfs_alloc_write_mask(mapping); |
6d4572a9 | 4675 | size_t write_bytes = blocksize; |
39279cc3 | 4676 | int ret = 0; |
9703fefe CR |
4677 | u64 block_start; |
4678 | u64 block_end; | |
39279cc3 | 4679 | |
b03ebd99 NB |
4680 | if (IS_ALIGNED(offset, blocksize) && |
4681 | (!len || IS_ALIGNED(len, blocksize))) | |
39279cc3 | 4682 | goto out; |
9703fefe | 4683 | |
8b62f87b JB |
4684 | block_start = round_down(from, blocksize); |
4685 | block_end = block_start + blocksize - 1; | |
4686 | ||
217f42eb NB |
4687 | ret = btrfs_check_data_free_space(inode, &data_reserved, block_start, |
4688 | blocksize); | |
6d4572a9 | 4689 | if (ret < 0) { |
217f42eb | 4690 | if (btrfs_check_nocow_lock(inode, block_start, &write_bytes) > 0) { |
6d4572a9 QW |
4691 | /* For nocow case, no need to reserve data space */ |
4692 | only_release_metadata = true; | |
4693 | } else { | |
4694 | goto out; | |
4695 | } | |
4696 | } | |
28c9b1e7 | 4697 | ret = btrfs_delalloc_reserve_metadata(inode, blocksize, blocksize); |
6d4572a9 QW |
4698 | if (ret < 0) { |
4699 | if (!only_release_metadata) | |
217f42eb NB |
4700 | btrfs_free_reserved_data_space(inode, data_reserved, |
4701 | block_start, blocksize); | |
6d4572a9 QW |
4702 | goto out; |
4703 | } | |
211c17f5 | 4704 | again: |
3b16a4e3 | 4705 | page = find_or_create_page(mapping, index, mask); |
5d5e103a | 4706 | if (!page) { |
217f42eb NB |
4707 | btrfs_delalloc_release_space(inode, data_reserved, block_start, |
4708 | blocksize, true); | |
4709 | btrfs_delalloc_release_extents(inode, blocksize); | |
ac6a2b36 | 4710 | ret = -ENOMEM; |
39279cc3 | 4711 | goto out; |
5d5e103a | 4712 | } |
32443de3 QW |
4713 | ret = set_page_extent_mapped(page); |
4714 | if (ret < 0) | |
4715 | goto out_unlock; | |
e6dcd2dc | 4716 | |
39279cc3 | 4717 | if (!PageUptodate(page)) { |
9ebefb18 | 4718 | ret = btrfs_readpage(NULL, page); |
39279cc3 | 4719 | lock_page(page); |
211c17f5 CM |
4720 | if (page->mapping != mapping) { |
4721 | unlock_page(page); | |
09cbfeaf | 4722 | put_page(page); |
211c17f5 CM |
4723 | goto again; |
4724 | } | |
39279cc3 CM |
4725 | if (!PageUptodate(page)) { |
4726 | ret = -EIO; | |
89642229 | 4727 | goto out_unlock; |
39279cc3 CM |
4728 | } |
4729 | } | |
211c17f5 | 4730 | wait_on_page_writeback(page); |
e6dcd2dc | 4731 | |
9703fefe | 4732 | lock_extent_bits(io_tree, block_start, block_end, &cached_state); |
e6dcd2dc | 4733 | |
217f42eb | 4734 | ordered = btrfs_lookup_ordered_extent(inode, block_start); |
e6dcd2dc | 4735 | if (ordered) { |
9703fefe | 4736 | unlock_extent_cached(io_tree, block_start, block_end, |
e43bbe5e | 4737 | &cached_state); |
e6dcd2dc | 4738 | unlock_page(page); |
09cbfeaf | 4739 | put_page(page); |
c0a43603 | 4740 | btrfs_start_ordered_extent(ordered, 1); |
e6dcd2dc CM |
4741 | btrfs_put_ordered_extent(ordered); |
4742 | goto again; | |
4743 | } | |
4744 | ||
217f42eb | 4745 | clear_extent_bit(&inode->io_tree, block_start, block_end, |
e182163d OS |
4746 | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, |
4747 | 0, 0, &cached_state); | |
5d5e103a | 4748 | |
217f42eb | 4749 | ret = btrfs_set_extent_delalloc(inode, block_start, block_end, 0, |
330a5827 | 4750 | &cached_state); |
9ed74f2d | 4751 | if (ret) { |
9703fefe | 4752 | unlock_extent_cached(io_tree, block_start, block_end, |
e43bbe5e | 4753 | &cached_state); |
9ed74f2d JB |
4754 | goto out_unlock; |
4755 | } | |
4756 | ||
9703fefe | 4757 | if (offset != blocksize) { |
2aaa6655 | 4758 | if (!len) |
9703fefe | 4759 | len = blocksize - offset; |
2aaa6655 | 4760 | if (front) |
d048b9c2 IW |
4761 | memzero_page(page, (block_start - page_offset(page)), |
4762 | offset); | |
2aaa6655 | 4763 | else |
d048b9c2 IW |
4764 | memzero_page(page, (block_start - page_offset(page)) + offset, |
4765 | len); | |
e6dcd2dc | 4766 | flush_dcache_page(page); |
e6dcd2dc | 4767 | } |
e4f94347 QW |
4768 | btrfs_page_clear_checked(fs_info, page, block_start, |
4769 | block_end + 1 - block_start); | |
6c9ac8be | 4770 | btrfs_page_set_dirty(fs_info, page, block_start, block_end + 1 - block_start); |
e43bbe5e | 4771 | unlock_extent_cached(io_tree, block_start, block_end, &cached_state); |
39279cc3 | 4772 | |
6d4572a9 | 4773 | if (only_release_metadata) |
217f42eb | 4774 | set_extent_bit(&inode->io_tree, block_start, block_end, |
1cab5e72 | 4775 | EXTENT_NORESERVE, 0, NULL, NULL, GFP_NOFS, NULL); |
6d4572a9 | 4776 | |
89642229 | 4777 | out_unlock: |
6d4572a9 QW |
4778 | if (ret) { |
4779 | if (only_release_metadata) | |
217f42eb | 4780 | btrfs_delalloc_release_metadata(inode, blocksize, true); |
6d4572a9 | 4781 | else |
217f42eb | 4782 | btrfs_delalloc_release_space(inode, data_reserved, |
6d4572a9 QW |
4783 | block_start, blocksize, true); |
4784 | } | |
217f42eb | 4785 | btrfs_delalloc_release_extents(inode, blocksize); |
39279cc3 | 4786 | unlock_page(page); |
09cbfeaf | 4787 | put_page(page); |
39279cc3 | 4788 | out: |
6d4572a9 | 4789 | if (only_release_metadata) |
217f42eb | 4790 | btrfs_check_nocow_unlock(inode); |
364ecf36 | 4791 | extent_changeset_free(data_reserved); |
39279cc3 CM |
4792 | return ret; |
4793 | } | |
4794 | ||
a4ba6cc0 | 4795 | static int maybe_insert_hole(struct btrfs_root *root, struct btrfs_inode *inode, |
16e7549f JB |
4796 | u64 offset, u64 len) |
4797 | { | |
a4ba6cc0 | 4798 | struct btrfs_fs_info *fs_info = root->fs_info; |
16e7549f | 4799 | struct btrfs_trans_handle *trans; |
5893dfb9 | 4800 | struct btrfs_drop_extents_args drop_args = { 0 }; |
16e7549f JB |
4801 | int ret; |
4802 | ||
4803 | /* | |
cceaa89f FM |
4804 | * If NO_HOLES is enabled, we don't need to do anything. |
4805 | * Later, up in the call chain, either btrfs_set_inode_last_sub_trans() | |
4806 | * or btrfs_update_inode() will be called, which guarantee that the next | |
4807 | * fsync will know this inode was changed and needs to be logged. | |
16e7549f | 4808 | */ |
cceaa89f | 4809 | if (btrfs_fs_incompat(fs_info, NO_HOLES)) |
16e7549f | 4810 | return 0; |
16e7549f JB |
4811 | |
4812 | /* | |
4813 | * 1 - for the one we're dropping | |
4814 | * 1 - for the one we're adding | |
4815 | * 1 - for updating the inode. | |
4816 | */ | |
4817 | trans = btrfs_start_transaction(root, 3); | |
4818 | if (IS_ERR(trans)) | |
4819 | return PTR_ERR(trans); | |
4820 | ||
5893dfb9 FM |
4821 | drop_args.start = offset; |
4822 | drop_args.end = offset + len; | |
4823 | drop_args.drop_cache = true; | |
4824 | ||
a4ba6cc0 | 4825 | ret = btrfs_drop_extents(trans, root, inode, &drop_args); |
16e7549f | 4826 | if (ret) { |
66642832 | 4827 | btrfs_abort_transaction(trans, ret); |
3a45bb20 | 4828 | btrfs_end_transaction(trans); |
16e7549f JB |
4829 | return ret; |
4830 | } | |
4831 | ||
a4ba6cc0 | 4832 | ret = btrfs_insert_file_extent(trans, root, btrfs_ino(inode), |
f85b7379 | 4833 | offset, 0, 0, len, 0, len, 0, 0, 0); |
2766ff61 | 4834 | if (ret) { |
66642832 | 4835 | btrfs_abort_transaction(trans, ret); |
2766ff61 | 4836 | } else { |
a4ba6cc0 NB |
4837 | btrfs_update_inode_bytes(inode, 0, drop_args.bytes_found); |
4838 | btrfs_update_inode(trans, root, inode); | |
2766ff61 | 4839 | } |
3a45bb20 | 4840 | btrfs_end_transaction(trans); |
16e7549f JB |
4841 | return ret; |
4842 | } | |
4843 | ||
695a0d0d JB |
4844 | /* |
4845 | * This function puts in dummy file extents for the area we're creating a hole | |
4846 | * for. So if we are truncating this file to a larger size we need to insert | |
4847 | * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for | |
4848 | * the range between oldsize and size | |
4849 | */ | |
b06359a3 | 4850 | int btrfs_cont_expand(struct btrfs_inode *inode, loff_t oldsize, loff_t size) |
39279cc3 | 4851 | { |
b06359a3 NB |
4852 | struct btrfs_root *root = inode->root; |
4853 | struct btrfs_fs_info *fs_info = root->fs_info; | |
4854 | struct extent_io_tree *io_tree = &inode->io_tree; | |
a22285a6 | 4855 | struct extent_map *em = NULL; |
2ac55d41 | 4856 | struct extent_state *cached_state = NULL; |
b06359a3 | 4857 | struct extent_map_tree *em_tree = &inode->extent_tree; |
0b246afa JM |
4858 | u64 hole_start = ALIGN(oldsize, fs_info->sectorsize); |
4859 | u64 block_end = ALIGN(size, fs_info->sectorsize); | |
9036c102 YZ |
4860 | u64 last_byte; |
4861 | u64 cur_offset; | |
4862 | u64 hole_size; | |
9ed74f2d | 4863 | int err = 0; |
39279cc3 | 4864 | |
a71754fc | 4865 | /* |
9703fefe CR |
4866 | * If our size started in the middle of a block we need to zero out the |
4867 | * rest of the block before we expand the i_size, otherwise we could | |
a71754fc JB |
4868 | * expose stale data. |
4869 | */ | |
b06359a3 | 4870 | err = btrfs_truncate_block(inode, oldsize, 0, 0); |
a71754fc JB |
4871 | if (err) |
4872 | return err; | |
4873 | ||
9036c102 YZ |
4874 | if (size <= hole_start) |
4875 | return 0; | |
4876 | ||
b06359a3 NB |
4877 | btrfs_lock_and_flush_ordered_range(inode, hole_start, block_end - 1, |
4878 | &cached_state); | |
9036c102 YZ |
4879 | cur_offset = hole_start; |
4880 | while (1) { | |
b06359a3 | 4881 | em = btrfs_get_extent(inode, NULL, 0, cur_offset, |
39b07b5d | 4882 | block_end - cur_offset); |
79787eaa JM |
4883 | if (IS_ERR(em)) { |
4884 | err = PTR_ERR(em); | |
f2767956 | 4885 | em = NULL; |
79787eaa JM |
4886 | break; |
4887 | } | |
9036c102 | 4888 | last_byte = min(extent_map_end(em), block_end); |
0b246afa | 4889 | last_byte = ALIGN(last_byte, fs_info->sectorsize); |
9ddc959e JB |
4890 | hole_size = last_byte - cur_offset; |
4891 | ||
8082510e | 4892 | if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) { |
5dc562c5 | 4893 | struct extent_map *hole_em; |
9ed74f2d | 4894 | |
b06359a3 NB |
4895 | err = maybe_insert_hole(root, inode, cur_offset, |
4896 | hole_size); | |
16e7549f | 4897 | if (err) |
3893e33b | 4898 | break; |
9ddc959e | 4899 | |
b06359a3 | 4900 | err = btrfs_inode_set_file_extent_range(inode, |
9ddc959e JB |
4901 | cur_offset, hole_size); |
4902 | if (err) | |
4903 | break; | |
4904 | ||
b06359a3 | 4905 | btrfs_drop_extent_cache(inode, cur_offset, |
5dc562c5 JB |
4906 | cur_offset + hole_size - 1, 0); |
4907 | hole_em = alloc_extent_map(); | |
4908 | if (!hole_em) { | |
23e3337f | 4909 | btrfs_set_inode_full_sync(inode); |
5dc562c5 JB |
4910 | goto next; |
4911 | } | |
4912 | hole_em->start = cur_offset; | |
4913 | hole_em->len = hole_size; | |
4914 | hole_em->orig_start = cur_offset; | |
8082510e | 4915 | |
5dc562c5 JB |
4916 | hole_em->block_start = EXTENT_MAP_HOLE; |
4917 | hole_em->block_len = 0; | |
b4939680 | 4918 | hole_em->orig_block_len = 0; |
cc95bef6 | 4919 | hole_em->ram_bytes = hole_size; |
5dc562c5 | 4920 | hole_em->compress_type = BTRFS_COMPRESS_NONE; |
0b246afa | 4921 | hole_em->generation = fs_info->generation; |
8082510e | 4922 | |
5dc562c5 JB |
4923 | while (1) { |
4924 | write_lock(&em_tree->lock); | |
09a2a8f9 | 4925 | err = add_extent_mapping(em_tree, hole_em, 1); |
5dc562c5 JB |
4926 | write_unlock(&em_tree->lock); |
4927 | if (err != -EEXIST) | |
4928 | break; | |
b06359a3 | 4929 | btrfs_drop_extent_cache(inode, cur_offset, |
5dc562c5 JB |
4930 | cur_offset + |
4931 | hole_size - 1, 0); | |
4932 | } | |
4933 | free_extent_map(hole_em); | |
9ddc959e | 4934 | } else { |
b06359a3 | 4935 | err = btrfs_inode_set_file_extent_range(inode, |
9ddc959e JB |
4936 | cur_offset, hole_size); |
4937 | if (err) | |
4938 | break; | |
9036c102 | 4939 | } |
16e7549f | 4940 | next: |
9036c102 | 4941 | free_extent_map(em); |
a22285a6 | 4942 | em = NULL; |
9036c102 | 4943 | cur_offset = last_byte; |
8082510e | 4944 | if (cur_offset >= block_end) |
9036c102 YZ |
4945 | break; |
4946 | } | |
a22285a6 | 4947 | free_extent_map(em); |
e43bbe5e | 4948 | unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state); |
9036c102 YZ |
4949 | return err; |
4950 | } | |
39279cc3 | 4951 | |
3972f260 | 4952 | static int btrfs_setsize(struct inode *inode, struct iattr *attr) |
8082510e | 4953 | { |
f4a2f4c5 MX |
4954 | struct btrfs_root *root = BTRFS_I(inode)->root; |
4955 | struct btrfs_trans_handle *trans; | |
a41ad394 | 4956 | loff_t oldsize = i_size_read(inode); |
3972f260 ES |
4957 | loff_t newsize = attr->ia_size; |
4958 | int mask = attr->ia_valid; | |
8082510e YZ |
4959 | int ret; |
4960 | ||
3972f260 ES |
4961 | /* |
4962 | * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a | |
4963 | * special case where we need to update the times despite not having | |
4964 | * these flags set. For all other operations the VFS set these flags | |
4965 | * explicitly if it wants a timestamp update. | |
4966 | */ | |
dff6efc3 CH |
4967 | if (newsize != oldsize) { |
4968 | inode_inc_iversion(inode); | |
4969 | if (!(mask & (ATTR_CTIME | ATTR_MTIME))) | |
4970 | inode->i_ctime = inode->i_mtime = | |
c2050a45 | 4971 | current_time(inode); |
dff6efc3 | 4972 | } |
3972f260 | 4973 | |
a41ad394 | 4974 | if (newsize > oldsize) { |
9ea24bbe | 4975 | /* |
ea14b57f | 4976 | * Don't do an expanding truncate while snapshotting is ongoing. |
9ea24bbe FM |
4977 | * This is to ensure the snapshot captures a fully consistent |
4978 | * state of this file - if the snapshot captures this expanding | |
4979 | * truncation, it must capture all writes that happened before | |
4980 | * this truncation. | |
4981 | */ | |
dcc3eb96 | 4982 | btrfs_drew_write_lock(&root->snapshot_lock); |
b06359a3 | 4983 | ret = btrfs_cont_expand(BTRFS_I(inode), oldsize, newsize); |
9ea24bbe | 4984 | if (ret) { |
dcc3eb96 | 4985 | btrfs_drew_write_unlock(&root->snapshot_lock); |
8082510e | 4986 | return ret; |
9ea24bbe | 4987 | } |
8082510e | 4988 | |
f4a2f4c5 | 4989 | trans = btrfs_start_transaction(root, 1); |
9ea24bbe | 4990 | if (IS_ERR(trans)) { |
dcc3eb96 | 4991 | btrfs_drew_write_unlock(&root->snapshot_lock); |
f4a2f4c5 | 4992 | return PTR_ERR(trans); |
9ea24bbe | 4993 | } |
f4a2f4c5 MX |
4994 | |
4995 | i_size_write(inode, newsize); | |
76aea537 | 4996 | btrfs_inode_safe_disk_i_size_write(BTRFS_I(inode), 0); |
27772b68 | 4997 | pagecache_isize_extended(inode, oldsize, newsize); |
9a56fcd1 | 4998 | ret = btrfs_update_inode(trans, root, BTRFS_I(inode)); |
dcc3eb96 | 4999 | btrfs_drew_write_unlock(&root->snapshot_lock); |
3a45bb20 | 5000 | btrfs_end_transaction(trans); |
a41ad394 | 5001 | } else { |
24c0a722 NA |
5002 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
5003 | ||
5004 | if (btrfs_is_zoned(fs_info)) { | |
5005 | ret = btrfs_wait_ordered_range(inode, | |
5006 | ALIGN(newsize, fs_info->sectorsize), | |
5007 | (u64)-1); | |
5008 | if (ret) | |
5009 | return ret; | |
5010 | } | |
8082510e | 5011 | |
a41ad394 JB |
5012 | /* |
5013 | * We're truncating a file that used to have good data down to | |
1fd4033d NB |
5014 | * zero. Make sure any new writes to the file get on disk |
5015 | * on close. | |
a41ad394 JB |
5016 | */ |
5017 | if (newsize == 0) | |
1fd4033d | 5018 | set_bit(BTRFS_INODE_FLUSH_ON_CLOSE, |
72ac3c0d | 5019 | &BTRFS_I(inode)->runtime_flags); |
8082510e | 5020 | |
a41ad394 | 5021 | truncate_setsize(inode, newsize); |
2e60a51e | 5022 | |
2e60a51e | 5023 | inode_dio_wait(inode); |
2e60a51e | 5024 | |
213e8c55 | 5025 | ret = btrfs_truncate(inode, newsize == oldsize); |
7f4f6e0a JB |
5026 | if (ret && inode->i_nlink) { |
5027 | int err; | |
5028 | ||
5029 | /* | |
f7e9e8fc OS |
5030 | * Truncate failed, so fix up the in-memory size. We |
5031 | * adjusted disk_i_size down as we removed extents, so | |
5032 | * wait for disk_i_size to be stable and then update the | |
5033 | * in-memory size to match. | |
7f4f6e0a | 5034 | */ |
f7e9e8fc | 5035 | err = btrfs_wait_ordered_range(inode, 0, (u64)-1); |
7f4f6e0a | 5036 | if (err) |
f7e9e8fc OS |
5037 | return err; |
5038 | i_size_write(inode, BTRFS_I(inode)->disk_i_size); | |
7f4f6e0a | 5039 | } |
8082510e YZ |
5040 | } |
5041 | ||
a41ad394 | 5042 | return ret; |
8082510e YZ |
5043 | } |
5044 | ||
549c7297 CB |
5045 | static int btrfs_setattr(struct user_namespace *mnt_userns, struct dentry *dentry, |
5046 | struct iattr *attr) | |
9036c102 | 5047 | { |
2b0143b5 | 5048 | struct inode *inode = d_inode(dentry); |
b83cc969 | 5049 | struct btrfs_root *root = BTRFS_I(inode)->root; |
9036c102 | 5050 | int err; |
39279cc3 | 5051 | |
b83cc969 LZ |
5052 | if (btrfs_root_readonly(root)) |
5053 | return -EROFS; | |
5054 | ||
d4d09464 | 5055 | err = setattr_prepare(mnt_userns, dentry, attr); |
9036c102 YZ |
5056 | if (err) |
5057 | return err; | |
2bf5a725 | 5058 | |
5a3f23d5 | 5059 | if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) { |
3972f260 | 5060 | err = btrfs_setsize(inode, attr); |
8082510e YZ |
5061 | if (err) |
5062 | return err; | |
39279cc3 | 5063 | } |
9036c102 | 5064 | |
1025774c | 5065 | if (attr->ia_valid) { |
d4d09464 | 5066 | setattr_copy(mnt_userns, inode, attr); |
0c4d2d95 | 5067 | inode_inc_iversion(inode); |
22c44fe6 | 5068 | err = btrfs_dirty_inode(inode); |
1025774c | 5069 | |
22c44fe6 | 5070 | if (!err && attr->ia_valid & ATTR_MODE) |
d4d09464 | 5071 | err = posix_acl_chmod(mnt_userns, inode, inode->i_mode); |
1025774c | 5072 | } |
33268eaf | 5073 | |
39279cc3 CM |
5074 | return err; |
5075 | } | |
61295eb8 | 5076 | |
131e404a | 5077 | /* |
895586eb MWO |
5078 | * While truncating the inode pages during eviction, we get the VFS |
5079 | * calling btrfs_invalidate_folio() against each folio of the inode. This | |
5080 | * is slow because the calls to btrfs_invalidate_folio() result in a | |
5081 | * huge amount of calls to lock_extent_bits() and clear_extent_bit(), | |
5082 | * which keep merging and splitting extent_state structures over and over, | |
5083 | * wasting lots of time. | |
131e404a | 5084 | * |
895586eb MWO |
5085 | * Therefore if the inode is being evicted, let btrfs_invalidate_folio() |
5086 | * skip all those expensive operations on a per folio basis and do only | |
5087 | * the ordered io finishing, while we release here the extent_map and | |
5088 | * extent_state structures, without the excessive merging and splitting. | |
131e404a FDBM |
5089 | */ |
5090 | static void evict_inode_truncate_pages(struct inode *inode) | |
5091 | { | |
5092 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; | |
5093 | struct extent_map_tree *map_tree = &BTRFS_I(inode)->extent_tree; | |
5094 | struct rb_node *node; | |
5095 | ||
5096 | ASSERT(inode->i_state & I_FREEING); | |
91b0abe3 | 5097 | truncate_inode_pages_final(&inode->i_data); |
131e404a FDBM |
5098 | |
5099 | write_lock(&map_tree->lock); | |
07e1ce09 | 5100 | while (!RB_EMPTY_ROOT(&map_tree->map.rb_root)) { |
131e404a FDBM |
5101 | struct extent_map *em; |
5102 | ||
07e1ce09 | 5103 | node = rb_first_cached(&map_tree->map); |
131e404a | 5104 | em = rb_entry(node, struct extent_map, rb_node); |
180589ef WS |
5105 | clear_bit(EXTENT_FLAG_PINNED, &em->flags); |
5106 | clear_bit(EXTENT_FLAG_LOGGING, &em->flags); | |
131e404a FDBM |
5107 | remove_extent_mapping(map_tree, em); |
5108 | free_extent_map(em); | |
7064dd5c FM |
5109 | if (need_resched()) { |
5110 | write_unlock(&map_tree->lock); | |
5111 | cond_resched(); | |
5112 | write_lock(&map_tree->lock); | |
5113 | } | |
131e404a FDBM |
5114 | } |
5115 | write_unlock(&map_tree->lock); | |
5116 | ||
6ca07097 FM |
5117 | /* |
5118 | * Keep looping until we have no more ranges in the io tree. | |
ba206a02 MWO |
5119 | * We can have ongoing bios started by readahead that have |
5120 | * their endio callback (extent_io.c:end_bio_extent_readpage) | |
9c6429d9 FM |
5121 | * still in progress (unlocked the pages in the bio but did not yet |
5122 | * unlocked the ranges in the io tree). Therefore this means some | |
6ca07097 FM |
5123 | * ranges can still be locked and eviction started because before |
5124 | * submitting those bios, which are executed by a separate task (work | |
5125 | * queue kthread), inode references (inode->i_count) were not taken | |
5126 | * (which would be dropped in the end io callback of each bio). | |
5127 | * Therefore here we effectively end up waiting for those bios and | |
5128 | * anyone else holding locked ranges without having bumped the inode's | |
5129 | * reference count - if we don't do it, when they access the inode's | |
5130 | * io_tree to unlock a range it may be too late, leading to an | |
5131 | * use-after-free issue. | |
5132 | */ | |
131e404a FDBM |
5133 | spin_lock(&io_tree->lock); |
5134 | while (!RB_EMPTY_ROOT(&io_tree->state)) { | |
5135 | struct extent_state *state; | |
5136 | struct extent_state *cached_state = NULL; | |
6ca07097 FM |
5137 | u64 start; |
5138 | u64 end; | |
421f0922 | 5139 | unsigned state_flags; |
131e404a FDBM |
5140 | |
5141 | node = rb_first(&io_tree->state); | |
5142 | state = rb_entry(node, struct extent_state, rb_node); | |
6ca07097 FM |
5143 | start = state->start; |
5144 | end = state->end; | |
421f0922 | 5145 | state_flags = state->state; |
131e404a FDBM |
5146 | spin_unlock(&io_tree->lock); |
5147 | ||
ff13db41 | 5148 | lock_extent_bits(io_tree, start, end, &cached_state); |
b9d0b389 QW |
5149 | |
5150 | /* | |
5151 | * If still has DELALLOC flag, the extent didn't reach disk, | |
5152 | * and its reserved space won't be freed by delayed_ref. | |
5153 | * So we need to free its reserved space here. | |
895586eb | 5154 | * (Refer to comment in btrfs_invalidate_folio, case 2) |
b9d0b389 QW |
5155 | * |
5156 | * Note, end is the bytenr of last byte, so we need + 1 here. | |
5157 | */ | |
421f0922 | 5158 | if (state_flags & EXTENT_DELALLOC) |
8b8a979f NB |
5159 | btrfs_qgroup_free_data(BTRFS_I(inode), NULL, start, |
5160 | end - start + 1); | |
b9d0b389 | 5161 | |
6ca07097 | 5162 | clear_extent_bit(io_tree, start, end, |
e182163d OS |
5163 | EXTENT_LOCKED | EXTENT_DELALLOC | |
5164 | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 1, 1, | |
5165 | &cached_state); | |
131e404a | 5166 | |
7064dd5c | 5167 | cond_resched(); |
131e404a FDBM |
5168 | spin_lock(&io_tree->lock); |
5169 | } | |
5170 | spin_unlock(&io_tree->lock); | |
5171 | } | |
5172 | ||
4b9d7b59 | 5173 | static struct btrfs_trans_handle *evict_refill_and_join(struct btrfs_root *root, |
ad80cf50 | 5174 | struct btrfs_block_rsv *rsv) |
4b9d7b59 OS |
5175 | { |
5176 | struct btrfs_fs_info *fs_info = root->fs_info; | |
d3984c90 | 5177 | struct btrfs_trans_handle *trans; |
2bd36e7b | 5178 | u64 delayed_refs_extra = btrfs_calc_insert_metadata_size(fs_info, 1); |
d3984c90 | 5179 | int ret; |
4b9d7b59 | 5180 | |
d3984c90 JB |
5181 | /* |
5182 | * Eviction should be taking place at some place safe because of our | |
5183 | * delayed iputs. However the normal flushing code will run delayed | |
5184 | * iputs, so we cannot use FLUSH_ALL otherwise we'll deadlock. | |
5185 | * | |
5186 | * We reserve the delayed_refs_extra here again because we can't use | |
5187 | * btrfs_start_transaction(root, 0) for the same deadlocky reason as | |
5188 | * above. We reserve our extra bit here because we generate a ton of | |
5189 | * delayed refs activity by truncating. | |
5190 | * | |
ee6adbfd JB |
5191 | * BTRFS_RESERVE_FLUSH_EVICT will steal from the global_rsv if it can, |
5192 | * if we fail to make this reservation we can re-try without the | |
5193 | * delayed_refs_extra so we can make some forward progress. | |
d3984c90 | 5194 | */ |
9270501c | 5195 | ret = btrfs_block_rsv_refill(fs_info, rsv, rsv->size + delayed_refs_extra, |
d3984c90 JB |
5196 | BTRFS_RESERVE_FLUSH_EVICT); |
5197 | if (ret) { | |
9270501c | 5198 | ret = btrfs_block_rsv_refill(fs_info, rsv, rsv->size, |
ee6adbfd JB |
5199 | BTRFS_RESERVE_FLUSH_EVICT); |
5200 | if (ret) { | |
d3984c90 JB |
5201 | btrfs_warn(fs_info, |
5202 | "could not allocate space for delete; will truncate on mount"); | |
5203 | return ERR_PTR(-ENOSPC); | |
5204 | } | |
5205 | delayed_refs_extra = 0; | |
5206 | } | |
4b9d7b59 | 5207 | |
d3984c90 JB |
5208 | trans = btrfs_join_transaction(root); |
5209 | if (IS_ERR(trans)) | |
5210 | return trans; | |
5211 | ||
5212 | if (delayed_refs_extra) { | |
5213 | trans->block_rsv = &fs_info->trans_block_rsv; | |
5214 | trans->bytes_reserved = delayed_refs_extra; | |
5215 | btrfs_block_rsv_migrate(rsv, trans->block_rsv, | |
5216 | delayed_refs_extra, 1); | |
4b9d7b59 | 5217 | } |
d3984c90 | 5218 | return trans; |
4b9d7b59 OS |
5219 | } |
5220 | ||
bd555975 | 5221 | void btrfs_evict_inode(struct inode *inode) |
39279cc3 | 5222 | { |
0b246afa | 5223 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
39279cc3 CM |
5224 | struct btrfs_trans_handle *trans; |
5225 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
4b9d7b59 | 5226 | struct btrfs_block_rsv *rsv; |
39279cc3 CM |
5227 | int ret; |
5228 | ||
1abe9b8a | 5229 | trace_btrfs_inode_evict(inode); |
5230 | ||
3d48d981 | 5231 | if (!root) { |
14605409 | 5232 | fsverity_cleanup_inode(inode); |
e8f1bc14 | 5233 | clear_inode(inode); |
3d48d981 NB |
5234 | return; |
5235 | } | |
5236 | ||
131e404a FDBM |
5237 | evict_inode_truncate_pages(inode); |
5238 | ||
69e9c6c6 SB |
5239 | if (inode->i_nlink && |
5240 | ((btrfs_root_refs(&root->root_item) != 0 && | |
5241 | root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) || | |
70ddc553 | 5242 | btrfs_is_free_space_inode(BTRFS_I(inode)))) |
bd555975 AV |
5243 | goto no_delete; |
5244 | ||
27919067 | 5245 | if (is_bad_inode(inode)) |
39279cc3 | 5246 | goto no_delete; |
5f39d397 | 5247 | |
7ab7956e | 5248 | btrfs_free_io_failure_record(BTRFS_I(inode), 0, (u64)-1); |
f612496b | 5249 | |
7b40b695 | 5250 | if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) |
c71bf099 | 5251 | goto no_delete; |
c71bf099 | 5252 | |
76dda93c | 5253 | if (inode->i_nlink > 0) { |
69e9c6c6 SB |
5254 | BUG_ON(btrfs_root_refs(&root->root_item) != 0 && |
5255 | root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID); | |
76dda93c YZ |
5256 | goto no_delete; |
5257 | } | |
5258 | ||
2adc75d6 JB |
5259 | /* |
5260 | * This makes sure the inode item in tree is uptodate and the space for | |
5261 | * the inode update is released. | |
5262 | */ | |
aa79021f | 5263 | ret = btrfs_commit_inode_delayed_inode(BTRFS_I(inode)); |
27919067 | 5264 | if (ret) |
0e8c36a9 | 5265 | goto no_delete; |
0e8c36a9 | 5266 | |
2adc75d6 JB |
5267 | /* |
5268 | * This drops any pending insert or delete operations we have for this | |
5269 | * inode. We could have a delayed dir index deletion queued up, but | |
5270 | * we're removing the inode completely so that'll be taken care of in | |
5271 | * the truncate. | |
5272 | */ | |
5273 | btrfs_kill_delayed_inode_items(BTRFS_I(inode)); | |
5274 | ||
2ff7e61e | 5275 | rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP); |
27919067 | 5276 | if (!rsv) |
4289a667 | 5277 | goto no_delete; |
2bd36e7b | 5278 | rsv->size = btrfs_calc_metadata_size(fs_info, 1); |
ca7e70f5 | 5279 | rsv->failfast = 1; |
4289a667 | 5280 | |
6ef06d27 | 5281 | btrfs_i_size_write(BTRFS_I(inode), 0); |
5f39d397 | 5282 | |
8082510e | 5283 | while (1) { |
d9ac19c3 | 5284 | struct btrfs_truncate_control control = { |
71d18b53 | 5285 | .inode = BTRFS_I(inode), |
487e81d2 | 5286 | .ino = btrfs_ino(BTRFS_I(inode)), |
d9ac19c3 JB |
5287 | .new_size = 0, |
5288 | .min_type = 0, | |
5289 | }; | |
5290 | ||
ad80cf50 | 5291 | trans = evict_refill_and_join(root, rsv); |
27919067 OS |
5292 | if (IS_ERR(trans)) |
5293 | goto free_rsv; | |
7b128766 | 5294 | |
4289a667 JB |
5295 | trans->block_rsv = rsv; |
5296 | ||
71d18b53 | 5297 | ret = btrfs_truncate_inode_items(trans, root, &control); |
27919067 OS |
5298 | trans->block_rsv = &fs_info->trans_block_rsv; |
5299 | btrfs_end_transaction(trans); | |
5300 | btrfs_btree_balance_dirty(fs_info); | |
5301 | if (ret && ret != -ENOSPC && ret != -EAGAIN) | |
5302 | goto free_rsv; | |
5303 | else if (!ret) | |
8082510e | 5304 | break; |
8082510e | 5305 | } |
5f39d397 | 5306 | |
4ef31a45 | 5307 | /* |
27919067 OS |
5308 | * Errors here aren't a big deal, it just means we leave orphan items in |
5309 | * the tree. They will be cleaned up on the next mount. If the inode | |
5310 | * number gets reused, cleanup deletes the orphan item without doing | |
5311 | * anything, and unlink reuses the existing orphan item. | |
5312 | * | |
5313 | * If it turns out that we are dropping too many of these, we might want | |
5314 | * to add a mechanism for retrying these after a commit. | |
4ef31a45 | 5315 | */ |
ad80cf50 | 5316 | trans = evict_refill_and_join(root, rsv); |
27919067 OS |
5317 | if (!IS_ERR(trans)) { |
5318 | trans->block_rsv = rsv; | |
5319 | btrfs_orphan_del(trans, BTRFS_I(inode)); | |
5320 | trans->block_rsv = &fs_info->trans_block_rsv; | |
5321 | btrfs_end_transaction(trans); | |
5322 | } | |
54aa1f4d | 5323 | |
27919067 OS |
5324 | free_rsv: |
5325 | btrfs_free_block_rsv(fs_info, rsv); | |
39279cc3 | 5326 | no_delete: |
27919067 OS |
5327 | /* |
5328 | * If we didn't successfully delete, the orphan item will still be in | |
5329 | * the tree and we'll retry on the next mount. Again, we might also want | |
5330 | * to retry these periodically in the future. | |
5331 | */ | |
f48d1cf5 | 5332 | btrfs_remove_delayed_node(BTRFS_I(inode)); |
14605409 | 5333 | fsverity_cleanup_inode(inode); |
dbd5768f | 5334 | clear_inode(inode); |
39279cc3 CM |
5335 | } |
5336 | ||
5337 | /* | |
6bf9e4bd QW |
5338 | * Return the key found in the dir entry in the location pointer, fill @type |
5339 | * with BTRFS_FT_*, and return 0. | |
5340 | * | |
005d6712 SY |
5341 | * If no dir entries were found, returns -ENOENT. |
5342 | * If found a corrupted location in dir entry, returns -EUCLEAN. | |
39279cc3 CM |
5343 | */ |
5344 | static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry, | |
6bf9e4bd | 5345 | struct btrfs_key *location, u8 *type) |
39279cc3 CM |
5346 | { |
5347 | const char *name = dentry->d_name.name; | |
5348 | int namelen = dentry->d_name.len; | |
5349 | struct btrfs_dir_item *di; | |
5350 | struct btrfs_path *path; | |
5351 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
0d9f7f3e | 5352 | int ret = 0; |
39279cc3 CM |
5353 | |
5354 | path = btrfs_alloc_path(); | |
d8926bb3 MF |
5355 | if (!path) |
5356 | return -ENOMEM; | |
3954401f | 5357 | |
f85b7379 DS |
5358 | di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(BTRFS_I(dir)), |
5359 | name, namelen, 0); | |
3cf5068f LB |
5360 | if (IS_ERR_OR_NULL(di)) { |
5361 | ret = di ? PTR_ERR(di) : -ENOENT; | |
005d6712 SY |
5362 | goto out; |
5363 | } | |
d397712b | 5364 | |
5f39d397 | 5365 | btrfs_dir_item_key_to_cpu(path->nodes[0], di, location); |
56a0e706 LB |
5366 | if (location->type != BTRFS_INODE_ITEM_KEY && |
5367 | location->type != BTRFS_ROOT_ITEM_KEY) { | |
005d6712 | 5368 | ret = -EUCLEAN; |
56a0e706 LB |
5369 | btrfs_warn(root->fs_info, |
5370 | "%s gets something invalid in DIR_ITEM (name %s, directory ino %llu, location(%llu %u %llu))", | |
5371 | __func__, name, btrfs_ino(BTRFS_I(dir)), | |
5372 | location->objectid, location->type, location->offset); | |
56a0e706 | 5373 | } |
6bf9e4bd QW |
5374 | if (!ret) |
5375 | *type = btrfs_dir_type(path->nodes[0], di); | |
39279cc3 | 5376 | out: |
39279cc3 CM |
5377 | btrfs_free_path(path); |
5378 | return ret; | |
5379 | } | |
5380 | ||
5381 | /* | |
5382 | * when we hit a tree root in a directory, the btrfs part of the inode | |
5383 | * needs to be changed to reflect the root directory of the tree root. This | |
5384 | * is kind of like crossing a mount point. | |
5385 | */ | |
2ff7e61e | 5386 | static int fixup_tree_root_location(struct btrfs_fs_info *fs_info, |
4df27c4d YZ |
5387 | struct inode *dir, |
5388 | struct dentry *dentry, | |
5389 | struct btrfs_key *location, | |
5390 | struct btrfs_root **sub_root) | |
39279cc3 | 5391 | { |
4df27c4d YZ |
5392 | struct btrfs_path *path; |
5393 | struct btrfs_root *new_root; | |
5394 | struct btrfs_root_ref *ref; | |
5395 | struct extent_buffer *leaf; | |
1d4c08e0 | 5396 | struct btrfs_key key; |
4df27c4d YZ |
5397 | int ret; |
5398 | int err = 0; | |
39279cc3 | 5399 | |
4df27c4d YZ |
5400 | path = btrfs_alloc_path(); |
5401 | if (!path) { | |
5402 | err = -ENOMEM; | |
5403 | goto out; | |
5404 | } | |
39279cc3 | 5405 | |
4df27c4d | 5406 | err = -ENOENT; |
1d4c08e0 DS |
5407 | key.objectid = BTRFS_I(dir)->root->root_key.objectid; |
5408 | key.type = BTRFS_ROOT_REF_KEY; | |
5409 | key.offset = location->objectid; | |
5410 | ||
0b246afa | 5411 | ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0); |
4df27c4d YZ |
5412 | if (ret) { |
5413 | if (ret < 0) | |
5414 | err = ret; | |
5415 | goto out; | |
5416 | } | |
39279cc3 | 5417 | |
4df27c4d YZ |
5418 | leaf = path->nodes[0]; |
5419 | ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref); | |
4a0cc7ca | 5420 | if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(BTRFS_I(dir)) || |
4df27c4d YZ |
5421 | btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len) |
5422 | goto out; | |
39279cc3 | 5423 | |
4df27c4d YZ |
5424 | ret = memcmp_extent_buffer(leaf, dentry->d_name.name, |
5425 | (unsigned long)(ref + 1), | |
5426 | dentry->d_name.len); | |
5427 | if (ret) | |
5428 | goto out; | |
5429 | ||
b3b4aa74 | 5430 | btrfs_release_path(path); |
4df27c4d | 5431 | |
56e9357a | 5432 | new_root = btrfs_get_fs_root(fs_info, location->objectid, true); |
4df27c4d YZ |
5433 | if (IS_ERR(new_root)) { |
5434 | err = PTR_ERR(new_root); | |
5435 | goto out; | |
5436 | } | |
5437 | ||
4df27c4d YZ |
5438 | *sub_root = new_root; |
5439 | location->objectid = btrfs_root_dirid(&new_root->root_item); | |
5440 | location->type = BTRFS_INODE_ITEM_KEY; | |
5441 | location->offset = 0; | |
5442 | err = 0; | |
5443 | out: | |
5444 | btrfs_free_path(path); | |
5445 | return err; | |
39279cc3 CM |
5446 | } |
5447 | ||
5d4f98a2 YZ |
5448 | static void inode_tree_add(struct inode *inode) |
5449 | { | |
5450 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5451 | struct btrfs_inode *entry; | |
03e860bd NP |
5452 | struct rb_node **p; |
5453 | struct rb_node *parent; | |
cef21937 | 5454 | struct rb_node *new = &BTRFS_I(inode)->rb_node; |
4a0cc7ca | 5455 | u64 ino = btrfs_ino(BTRFS_I(inode)); |
5d4f98a2 | 5456 | |
1d3382cb | 5457 | if (inode_unhashed(inode)) |
76dda93c | 5458 | return; |
e1409cef | 5459 | parent = NULL; |
5d4f98a2 | 5460 | spin_lock(&root->inode_lock); |
e1409cef | 5461 | p = &root->inode_tree.rb_node; |
5d4f98a2 YZ |
5462 | while (*p) { |
5463 | parent = *p; | |
5464 | entry = rb_entry(parent, struct btrfs_inode, rb_node); | |
5465 | ||
37508515 | 5466 | if (ino < btrfs_ino(entry)) |
03e860bd | 5467 | p = &parent->rb_left; |
37508515 | 5468 | else if (ino > btrfs_ino(entry)) |
03e860bd | 5469 | p = &parent->rb_right; |
5d4f98a2 YZ |
5470 | else { |
5471 | WARN_ON(!(entry->vfs_inode.i_state & | |
a4ffdde6 | 5472 | (I_WILL_FREE | I_FREEING))); |
cef21937 | 5473 | rb_replace_node(parent, new, &root->inode_tree); |
03e860bd NP |
5474 | RB_CLEAR_NODE(parent); |
5475 | spin_unlock(&root->inode_lock); | |
cef21937 | 5476 | return; |
5d4f98a2 YZ |
5477 | } |
5478 | } | |
cef21937 FDBM |
5479 | rb_link_node(new, parent, p); |
5480 | rb_insert_color(new, &root->inode_tree); | |
5d4f98a2 YZ |
5481 | spin_unlock(&root->inode_lock); |
5482 | } | |
5483 | ||
b79b7249 | 5484 | static void inode_tree_del(struct btrfs_inode *inode) |
5d4f98a2 | 5485 | { |
b79b7249 | 5486 | struct btrfs_root *root = inode->root; |
76dda93c | 5487 | int empty = 0; |
5d4f98a2 | 5488 | |
03e860bd | 5489 | spin_lock(&root->inode_lock); |
b79b7249 NB |
5490 | if (!RB_EMPTY_NODE(&inode->rb_node)) { |
5491 | rb_erase(&inode->rb_node, &root->inode_tree); | |
5492 | RB_CLEAR_NODE(&inode->rb_node); | |
76dda93c | 5493 | empty = RB_EMPTY_ROOT(&root->inode_tree); |
5d4f98a2 | 5494 | } |
03e860bd | 5495 | spin_unlock(&root->inode_lock); |
76dda93c | 5496 | |
69e9c6c6 | 5497 | if (empty && btrfs_root_refs(&root->root_item) == 0) { |
76dda93c YZ |
5498 | spin_lock(&root->inode_lock); |
5499 | empty = RB_EMPTY_ROOT(&root->inode_tree); | |
5500 | spin_unlock(&root->inode_lock); | |
5501 | if (empty) | |
5502 | btrfs_add_dead_root(root); | |
5503 | } | |
5504 | } | |
5505 | ||
5d4f98a2 | 5506 | |
e02119d5 CM |
5507 | static int btrfs_init_locked_inode(struct inode *inode, void *p) |
5508 | { | |
5509 | struct btrfs_iget_args *args = p; | |
0202e83f DS |
5510 | |
5511 | inode->i_ino = args->ino; | |
5512 | BTRFS_I(inode)->location.objectid = args->ino; | |
5513 | BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY; | |
5514 | BTRFS_I(inode)->location.offset = 0; | |
5c8fd99f JB |
5515 | BTRFS_I(inode)->root = btrfs_grab_root(args->root); |
5516 | BUG_ON(args->root && !BTRFS_I(inode)->root); | |
39279cc3 CM |
5517 | return 0; |
5518 | } | |
5519 | ||
5520 | static int btrfs_find_actor(struct inode *inode, void *opaque) | |
5521 | { | |
5522 | struct btrfs_iget_args *args = opaque; | |
0202e83f DS |
5523 | |
5524 | return args->ino == BTRFS_I(inode)->location.objectid && | |
d397712b | 5525 | args->root == BTRFS_I(inode)->root; |
39279cc3 CM |
5526 | } |
5527 | ||
0202e83f | 5528 | static struct inode *btrfs_iget_locked(struct super_block *s, u64 ino, |
5d4f98a2 | 5529 | struct btrfs_root *root) |
39279cc3 CM |
5530 | { |
5531 | struct inode *inode; | |
5532 | struct btrfs_iget_args args; | |
0202e83f | 5533 | unsigned long hashval = btrfs_inode_hash(ino, root); |
778ba82b | 5534 | |
0202e83f | 5535 | args.ino = ino; |
39279cc3 CM |
5536 | args.root = root; |
5537 | ||
778ba82b | 5538 | inode = iget5_locked(s, hashval, btrfs_find_actor, |
39279cc3 CM |
5539 | btrfs_init_locked_inode, |
5540 | (void *)&args); | |
5541 | return inode; | |
5542 | } | |
5543 | ||
4c66e0d4 | 5544 | /* |
0202e83f | 5545 | * Get an inode object given its inode number and corresponding root. |
4c66e0d4 DS |
5546 | * Path can be preallocated to prevent recursing back to iget through |
5547 | * allocator. NULL is also valid but may require an additional allocation | |
5548 | * later. | |
1a54ef8c | 5549 | */ |
0202e83f | 5550 | struct inode *btrfs_iget_path(struct super_block *s, u64 ino, |
4c66e0d4 | 5551 | struct btrfs_root *root, struct btrfs_path *path) |
1a54ef8c BR |
5552 | { |
5553 | struct inode *inode; | |
5554 | ||
0202e83f | 5555 | inode = btrfs_iget_locked(s, ino, root); |
1a54ef8c | 5556 | if (!inode) |
5d4f98a2 | 5557 | return ERR_PTR(-ENOMEM); |
1a54ef8c BR |
5558 | |
5559 | if (inode->i_state & I_NEW) { | |
67710892 FM |
5560 | int ret; |
5561 | ||
4222ea71 | 5562 | ret = btrfs_read_locked_inode(inode, path); |
9bc2ceff | 5563 | if (!ret) { |
1748f843 MF |
5564 | inode_tree_add(inode); |
5565 | unlock_new_inode(inode); | |
1748f843 | 5566 | } else { |
f5b3a417 AV |
5567 | iget_failed(inode); |
5568 | /* | |
5569 | * ret > 0 can come from btrfs_search_slot called by | |
5570 | * btrfs_read_locked_inode, this means the inode item | |
5571 | * was not found. | |
5572 | */ | |
5573 | if (ret > 0) | |
5574 | ret = -ENOENT; | |
5575 | inode = ERR_PTR(ret); | |
1748f843 MF |
5576 | } |
5577 | } | |
5578 | ||
1a54ef8c BR |
5579 | return inode; |
5580 | } | |
5581 | ||
0202e83f | 5582 | struct inode *btrfs_iget(struct super_block *s, u64 ino, struct btrfs_root *root) |
4222ea71 | 5583 | { |
0202e83f | 5584 | return btrfs_iget_path(s, ino, root, NULL); |
4222ea71 FM |
5585 | } |
5586 | ||
4df27c4d YZ |
5587 | static struct inode *new_simple_dir(struct super_block *s, |
5588 | struct btrfs_key *key, | |
5589 | struct btrfs_root *root) | |
5590 | { | |
5591 | struct inode *inode = new_inode(s); | |
5592 | ||
5593 | if (!inode) | |
5594 | return ERR_PTR(-ENOMEM); | |
5595 | ||
5c8fd99f | 5596 | BTRFS_I(inode)->root = btrfs_grab_root(root); |
4df27c4d | 5597 | memcpy(&BTRFS_I(inode)->location, key, sizeof(*key)); |
72ac3c0d | 5598 | set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags); |
4df27c4d YZ |
5599 | |
5600 | inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID; | |
6bb6b514 OS |
5601 | /* |
5602 | * We only need lookup, the rest is read-only and there's no inode | |
5603 | * associated with the dentry | |
5604 | */ | |
5605 | inode->i_op = &simple_dir_inode_operations; | |
1fdf4194 | 5606 | inode->i_opflags &= ~IOP_XATTR; |
4df27c4d YZ |
5607 | inode->i_fop = &simple_dir_operations; |
5608 | inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO; | |
c2050a45 | 5609 | inode->i_mtime = current_time(inode); |
9cc97d64 | 5610 | inode->i_atime = inode->i_mtime; |
5611 | inode->i_ctime = inode->i_mtime; | |
d3c6be6f | 5612 | BTRFS_I(inode)->i_otime = inode->i_mtime; |
4df27c4d YZ |
5613 | |
5614 | return inode; | |
5615 | } | |
5616 | ||
a55e65b8 DS |
5617 | static_assert(BTRFS_FT_UNKNOWN == FT_UNKNOWN); |
5618 | static_assert(BTRFS_FT_REG_FILE == FT_REG_FILE); | |
5619 | static_assert(BTRFS_FT_DIR == FT_DIR); | |
5620 | static_assert(BTRFS_FT_CHRDEV == FT_CHRDEV); | |
5621 | static_assert(BTRFS_FT_BLKDEV == FT_BLKDEV); | |
5622 | static_assert(BTRFS_FT_FIFO == FT_FIFO); | |
5623 | static_assert(BTRFS_FT_SOCK == FT_SOCK); | |
5624 | static_assert(BTRFS_FT_SYMLINK == FT_SYMLINK); | |
5625 | ||
6bf9e4bd QW |
5626 | static inline u8 btrfs_inode_type(struct inode *inode) |
5627 | { | |
6bf9e4bd QW |
5628 | return fs_umode_to_ftype(inode->i_mode); |
5629 | } | |
5630 | ||
3de4586c | 5631 | struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry) |
39279cc3 | 5632 | { |
0b246afa | 5633 | struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb); |
d397712b | 5634 | struct inode *inode; |
4df27c4d | 5635 | struct btrfs_root *root = BTRFS_I(dir)->root; |
39279cc3 CM |
5636 | struct btrfs_root *sub_root = root; |
5637 | struct btrfs_key location; | |
6bf9e4bd | 5638 | u8 di_type = 0; |
b4aff1f8 | 5639 | int ret = 0; |
39279cc3 CM |
5640 | |
5641 | if (dentry->d_name.len > BTRFS_NAME_LEN) | |
5642 | return ERR_PTR(-ENAMETOOLONG); | |
5f39d397 | 5643 | |
6bf9e4bd | 5644 | ret = btrfs_inode_by_name(dir, dentry, &location, &di_type); |
39279cc3 CM |
5645 | if (ret < 0) |
5646 | return ERR_PTR(ret); | |
5f39d397 | 5647 | |
4df27c4d | 5648 | if (location.type == BTRFS_INODE_ITEM_KEY) { |
0202e83f | 5649 | inode = btrfs_iget(dir->i_sb, location.objectid, root); |
6bf9e4bd QW |
5650 | if (IS_ERR(inode)) |
5651 | return inode; | |
5652 | ||
5653 | /* Do extra check against inode mode with di_type */ | |
5654 | if (btrfs_inode_type(inode) != di_type) { | |
5655 | btrfs_crit(fs_info, | |
5656 | "inode mode mismatch with dir: inode mode=0%o btrfs type=%u dir type=%u", | |
5657 | inode->i_mode, btrfs_inode_type(inode), | |
5658 | di_type); | |
5659 | iput(inode); | |
5660 | return ERR_PTR(-EUCLEAN); | |
5661 | } | |
4df27c4d YZ |
5662 | return inode; |
5663 | } | |
5664 | ||
2ff7e61e | 5665 | ret = fixup_tree_root_location(fs_info, dir, dentry, |
4df27c4d YZ |
5666 | &location, &sub_root); |
5667 | if (ret < 0) { | |
5668 | if (ret != -ENOENT) | |
5669 | inode = ERR_PTR(ret); | |
5670 | else | |
5671 | inode = new_simple_dir(dir->i_sb, &location, sub_root); | |
5672 | } else { | |
0202e83f | 5673 | inode = btrfs_iget(dir->i_sb, location.objectid, sub_root); |
39279cc3 | 5674 | } |
8727002f | 5675 | if (root != sub_root) |
00246528 | 5676 | btrfs_put_root(sub_root); |
76dda93c | 5677 | |
34d19bad | 5678 | if (!IS_ERR(inode) && root != sub_root) { |
0b246afa | 5679 | down_read(&fs_info->cleanup_work_sem); |
bc98a42c | 5680 | if (!sb_rdonly(inode->i_sb)) |
66b4ffd1 | 5681 | ret = btrfs_orphan_cleanup(sub_root); |
0b246afa | 5682 | up_read(&fs_info->cleanup_work_sem); |
01cd3367 JB |
5683 | if (ret) { |
5684 | iput(inode); | |
66b4ffd1 | 5685 | inode = ERR_PTR(ret); |
01cd3367 | 5686 | } |
c71bf099 YZ |
5687 | } |
5688 | ||
3de4586c CM |
5689 | return inode; |
5690 | } | |
5691 | ||
fe15ce44 | 5692 | static int btrfs_dentry_delete(const struct dentry *dentry) |
76dda93c YZ |
5693 | { |
5694 | struct btrfs_root *root; | |
2b0143b5 | 5695 | struct inode *inode = d_inode(dentry); |
76dda93c | 5696 | |
848cce0d | 5697 | if (!inode && !IS_ROOT(dentry)) |
2b0143b5 | 5698 | inode = d_inode(dentry->d_parent); |
76dda93c | 5699 | |
848cce0d LZ |
5700 | if (inode) { |
5701 | root = BTRFS_I(inode)->root; | |
efefb143 YZ |
5702 | if (btrfs_root_refs(&root->root_item) == 0) |
5703 | return 1; | |
848cce0d | 5704 | |
4a0cc7ca | 5705 | if (btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) |
848cce0d | 5706 | return 1; |
efefb143 | 5707 | } |
76dda93c YZ |
5708 | return 0; |
5709 | } | |
5710 | ||
3de4586c | 5711 | static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry, |
00cd8dd3 | 5712 | unsigned int flags) |
3de4586c | 5713 | { |
3837d208 | 5714 | struct inode *inode = btrfs_lookup_dentry(dir, dentry); |
5662344b | 5715 | |
3837d208 AV |
5716 | if (inode == ERR_PTR(-ENOENT)) |
5717 | inode = NULL; | |
41d28bca | 5718 | return d_splice_alias(inode, dentry); |
39279cc3 CM |
5719 | } |
5720 | ||
23b5ec74 JB |
5721 | /* |
5722 | * All this infrastructure exists because dir_emit can fault, and we are holding | |
5723 | * the tree lock when doing readdir. For now just allocate a buffer and copy | |
5724 | * our information into that, and then dir_emit from the buffer. This is | |
5725 | * similar to what NFS does, only we don't keep the buffer around in pagecache | |
5726 | * because I'm afraid I'll mess that up. Long term we need to make filldir do | |
5727 | * copy_to_user_inatomic so we don't have to worry about page faulting under the | |
5728 | * tree lock. | |
5729 | */ | |
5730 | static int btrfs_opendir(struct inode *inode, struct file *file) | |
5731 | { | |
5732 | struct btrfs_file_private *private; | |
5733 | ||
5734 | private = kzalloc(sizeof(struct btrfs_file_private), GFP_KERNEL); | |
5735 | if (!private) | |
5736 | return -ENOMEM; | |
5737 | private->filldir_buf = kzalloc(PAGE_SIZE, GFP_KERNEL); | |
5738 | if (!private->filldir_buf) { | |
5739 | kfree(private); | |
5740 | return -ENOMEM; | |
5741 | } | |
5742 | file->private_data = private; | |
5743 | return 0; | |
5744 | } | |
5745 | ||
5746 | struct dir_entry { | |
5747 | u64 ino; | |
5748 | u64 offset; | |
5749 | unsigned type; | |
5750 | int name_len; | |
5751 | }; | |
5752 | ||
5753 | static int btrfs_filldir(void *addr, int entries, struct dir_context *ctx) | |
5754 | { | |
5755 | while (entries--) { | |
5756 | struct dir_entry *entry = addr; | |
5757 | char *name = (char *)(entry + 1); | |
5758 | ||
92d32170 DS |
5759 | ctx->pos = get_unaligned(&entry->offset); |
5760 | if (!dir_emit(ctx, name, get_unaligned(&entry->name_len), | |
5761 | get_unaligned(&entry->ino), | |
5762 | get_unaligned(&entry->type))) | |
23b5ec74 | 5763 | return 1; |
92d32170 DS |
5764 | addr += sizeof(struct dir_entry) + |
5765 | get_unaligned(&entry->name_len); | |
23b5ec74 JB |
5766 | ctx->pos++; |
5767 | } | |
5768 | return 0; | |
5769 | } | |
5770 | ||
9cdda8d3 | 5771 | static int btrfs_real_readdir(struct file *file, struct dir_context *ctx) |
39279cc3 | 5772 | { |
9cdda8d3 | 5773 | struct inode *inode = file_inode(file); |
39279cc3 | 5774 | struct btrfs_root *root = BTRFS_I(inode)->root; |
23b5ec74 | 5775 | struct btrfs_file_private *private = file->private_data; |
39279cc3 CM |
5776 | struct btrfs_dir_item *di; |
5777 | struct btrfs_key key; | |
5f39d397 | 5778 | struct btrfs_key found_key; |
39279cc3 | 5779 | struct btrfs_path *path; |
23b5ec74 | 5780 | void *addr; |
16cdcec7 MX |
5781 | struct list_head ins_list; |
5782 | struct list_head del_list; | |
39279cc3 | 5783 | int ret; |
5f39d397 CM |
5784 | char *name_ptr; |
5785 | int name_len; | |
23b5ec74 JB |
5786 | int entries = 0; |
5787 | int total_len = 0; | |
02dbfc99 | 5788 | bool put = false; |
c2951f32 | 5789 | struct btrfs_key location; |
5f39d397 | 5790 | |
9cdda8d3 AV |
5791 | if (!dir_emit_dots(file, ctx)) |
5792 | return 0; | |
5793 | ||
49593bfa | 5794 | path = btrfs_alloc_path(); |
16cdcec7 MX |
5795 | if (!path) |
5796 | return -ENOMEM; | |
ff5714cc | 5797 | |
23b5ec74 | 5798 | addr = private->filldir_buf; |
e4058b54 | 5799 | path->reada = READA_FORWARD; |
49593bfa | 5800 | |
c2951f32 JM |
5801 | INIT_LIST_HEAD(&ins_list); |
5802 | INIT_LIST_HEAD(&del_list); | |
5803 | put = btrfs_readdir_get_delayed_items(inode, &ins_list, &del_list); | |
16cdcec7 | 5804 | |
23b5ec74 | 5805 | again: |
c2951f32 | 5806 | key.type = BTRFS_DIR_INDEX_KEY; |
9cdda8d3 | 5807 | key.offset = ctx->pos; |
4a0cc7ca | 5808 | key.objectid = btrfs_ino(BTRFS_I(inode)); |
5f39d397 | 5809 | |
a8ce68fd | 5810 | btrfs_for_each_slot(root, &key, &found_key, path, ret) { |
23b5ec74 | 5811 | struct dir_entry *entry; |
a8ce68fd | 5812 | struct extent_buffer *leaf = path->nodes[0]; |
5f39d397 CM |
5813 | |
5814 | if (found_key.objectid != key.objectid) | |
39279cc3 | 5815 | break; |
c2951f32 | 5816 | if (found_key.type != BTRFS_DIR_INDEX_KEY) |
39279cc3 | 5817 | break; |
9cdda8d3 | 5818 | if (found_key.offset < ctx->pos) |
a8ce68fd | 5819 | continue; |
c2951f32 | 5820 | if (btrfs_should_delete_dir_index(&del_list, found_key.offset)) |
a8ce68fd GN |
5821 | continue; |
5822 | di = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dir_item); | |
c2951f32 | 5823 | name_len = btrfs_dir_name_len(leaf, di); |
23b5ec74 JB |
5824 | if ((total_len + sizeof(struct dir_entry) + name_len) >= |
5825 | PAGE_SIZE) { | |
5826 | btrfs_release_path(path); | |
5827 | ret = btrfs_filldir(private->filldir_buf, entries, ctx); | |
5828 | if (ret) | |
5829 | goto nopos; | |
5830 | addr = private->filldir_buf; | |
5831 | entries = 0; | |
5832 | total_len = 0; | |
5833 | goto again; | |
c2951f32 | 5834 | } |
23b5ec74 JB |
5835 | |
5836 | entry = addr; | |
92d32170 | 5837 | put_unaligned(name_len, &entry->name_len); |
23b5ec74 | 5838 | name_ptr = (char *)(entry + 1); |
c2951f32 JM |
5839 | read_extent_buffer(leaf, name_ptr, (unsigned long)(di + 1), |
5840 | name_len); | |
7d157c3d | 5841 | put_unaligned(fs_ftype_to_dtype(btrfs_dir_type(leaf, di)), |
92d32170 | 5842 | &entry->type); |
c2951f32 | 5843 | btrfs_dir_item_key_to_cpu(leaf, di, &location); |
92d32170 DS |
5844 | put_unaligned(location.objectid, &entry->ino); |
5845 | put_unaligned(found_key.offset, &entry->offset); | |
23b5ec74 JB |
5846 | entries++; |
5847 | addr += sizeof(struct dir_entry) + name_len; | |
5848 | total_len += sizeof(struct dir_entry) + name_len; | |
39279cc3 | 5849 | } |
a8ce68fd GN |
5850 | /* Catch error encountered during iteration */ |
5851 | if (ret < 0) | |
5852 | goto err; | |
5853 | ||
23b5ec74 JB |
5854 | btrfs_release_path(path); |
5855 | ||
5856 | ret = btrfs_filldir(private->filldir_buf, entries, ctx); | |
5857 | if (ret) | |
5858 | goto nopos; | |
49593bfa | 5859 | |
d2fbb2b5 | 5860 | ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list); |
c2951f32 | 5861 | if (ret) |
bc4ef759 DS |
5862 | goto nopos; |
5863 | ||
db62efbb ZB |
5864 | /* |
5865 | * Stop new entries from being returned after we return the last | |
5866 | * entry. | |
5867 | * | |
5868 | * New directory entries are assigned a strictly increasing | |
5869 | * offset. This means that new entries created during readdir | |
5870 | * are *guaranteed* to be seen in the future by that readdir. | |
5871 | * This has broken buggy programs which operate on names as | |
5872 | * they're returned by readdir. Until we re-use freed offsets | |
5873 | * we have this hack to stop new entries from being returned | |
5874 | * under the assumption that they'll never reach this huge | |
5875 | * offset. | |
5876 | * | |
5877 | * This is being careful not to overflow 32bit loff_t unless the | |
5878 | * last entry requires it because doing so has broken 32bit apps | |
5879 | * in the past. | |
5880 | */ | |
c2951f32 JM |
5881 | if (ctx->pos >= INT_MAX) |
5882 | ctx->pos = LLONG_MAX; | |
5883 | else | |
5884 | ctx->pos = INT_MAX; | |
39279cc3 CM |
5885 | nopos: |
5886 | ret = 0; | |
5887 | err: | |
02dbfc99 OS |
5888 | if (put) |
5889 | btrfs_readdir_put_delayed_items(inode, &ins_list, &del_list); | |
39279cc3 | 5890 | btrfs_free_path(path); |
39279cc3 CM |
5891 | return ret; |
5892 | } | |
5893 | ||
39279cc3 | 5894 | /* |
54aa1f4d | 5895 | * This is somewhat expensive, updating the tree every time the |
39279cc3 CM |
5896 | * inode changes. But, it is most likely to find the inode in cache. |
5897 | * FIXME, needs more benchmarking...there are no reasons other than performance | |
5898 | * to keep or drop this code. | |
5899 | */ | |
48a3b636 | 5900 | static int btrfs_dirty_inode(struct inode *inode) |
39279cc3 | 5901 | { |
2ff7e61e | 5902 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
39279cc3 CM |
5903 | struct btrfs_root *root = BTRFS_I(inode)->root; |
5904 | struct btrfs_trans_handle *trans; | |
8929ecfa YZ |
5905 | int ret; |
5906 | ||
72ac3c0d | 5907 | if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags)) |
22c44fe6 | 5908 | return 0; |
39279cc3 | 5909 | |
7a7eaa40 | 5910 | trans = btrfs_join_transaction(root); |
22c44fe6 JB |
5911 | if (IS_ERR(trans)) |
5912 | return PTR_ERR(trans); | |
8929ecfa | 5913 | |
9a56fcd1 | 5914 | ret = btrfs_update_inode(trans, root, BTRFS_I(inode)); |
4d14c5cd | 5915 | if (ret && (ret == -ENOSPC || ret == -EDQUOT)) { |
94b60442 | 5916 | /* whoops, lets try again with the full transaction */ |
3a45bb20 | 5917 | btrfs_end_transaction(trans); |
94b60442 | 5918 | trans = btrfs_start_transaction(root, 1); |
22c44fe6 JB |
5919 | if (IS_ERR(trans)) |
5920 | return PTR_ERR(trans); | |
8929ecfa | 5921 | |
9a56fcd1 | 5922 | ret = btrfs_update_inode(trans, root, BTRFS_I(inode)); |
94b60442 | 5923 | } |
3a45bb20 | 5924 | btrfs_end_transaction(trans); |
16cdcec7 | 5925 | if (BTRFS_I(inode)->delayed_node) |
2ff7e61e | 5926 | btrfs_balance_delayed_items(fs_info); |
22c44fe6 JB |
5927 | |
5928 | return ret; | |
5929 | } | |
5930 | ||
5931 | /* | |
5932 | * This is a copy of file_update_time. We need this so we can return error on | |
5933 | * ENOSPC for updating the inode in the case of file write and mmap writes. | |
5934 | */ | |
95582b00 | 5935 | static int btrfs_update_time(struct inode *inode, struct timespec64 *now, |
e41f941a | 5936 | int flags) |
22c44fe6 | 5937 | { |
2bc55652 | 5938 | struct btrfs_root *root = BTRFS_I(inode)->root; |
3a8c7231 | 5939 | bool dirty = flags & ~S_VERSION; |
2bc55652 AB |
5940 | |
5941 | if (btrfs_root_readonly(root)) | |
5942 | return -EROFS; | |
5943 | ||
e41f941a | 5944 | if (flags & S_VERSION) |
3a8c7231 | 5945 | dirty |= inode_maybe_inc_iversion(inode, dirty); |
e41f941a JB |
5946 | if (flags & S_CTIME) |
5947 | inode->i_ctime = *now; | |
5948 | if (flags & S_MTIME) | |
5949 | inode->i_mtime = *now; | |
5950 | if (flags & S_ATIME) | |
5951 | inode->i_atime = *now; | |
3a8c7231 | 5952 | return dirty ? btrfs_dirty_inode(inode) : 0; |
39279cc3 CM |
5953 | } |
5954 | ||
d352ac68 CM |
5955 | /* |
5956 | * find the highest existing sequence number in a directory | |
5957 | * and then set the in-memory index_cnt variable to reflect | |
5958 | * free sequence numbers | |
5959 | */ | |
4c570655 | 5960 | static int btrfs_set_inode_index_count(struct btrfs_inode *inode) |
aec7477b | 5961 | { |
4c570655 | 5962 | struct btrfs_root *root = inode->root; |
aec7477b JB |
5963 | struct btrfs_key key, found_key; |
5964 | struct btrfs_path *path; | |
5965 | struct extent_buffer *leaf; | |
5966 | int ret; | |
5967 | ||
4c570655 | 5968 | key.objectid = btrfs_ino(inode); |
962a298f | 5969 | key.type = BTRFS_DIR_INDEX_KEY; |
aec7477b JB |
5970 | key.offset = (u64)-1; |
5971 | ||
5972 | path = btrfs_alloc_path(); | |
5973 | if (!path) | |
5974 | return -ENOMEM; | |
5975 | ||
5976 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
5977 | if (ret < 0) | |
5978 | goto out; | |
5979 | /* FIXME: we should be able to handle this */ | |
5980 | if (ret == 0) | |
5981 | goto out; | |
5982 | ret = 0; | |
5983 | ||
aec7477b | 5984 | if (path->slots[0] == 0) { |
528ee697 | 5985 | inode->index_cnt = BTRFS_DIR_START_INDEX; |
aec7477b JB |
5986 | goto out; |
5987 | } | |
5988 | ||
5989 | path->slots[0]--; | |
5990 | ||
5991 | leaf = path->nodes[0]; | |
5992 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
5993 | ||
4c570655 | 5994 | if (found_key.objectid != btrfs_ino(inode) || |
962a298f | 5995 | found_key.type != BTRFS_DIR_INDEX_KEY) { |
528ee697 | 5996 | inode->index_cnt = BTRFS_DIR_START_INDEX; |
aec7477b JB |
5997 | goto out; |
5998 | } | |
5999 | ||
4c570655 | 6000 | inode->index_cnt = found_key.offset + 1; |
aec7477b JB |
6001 | out: |
6002 | btrfs_free_path(path); | |
6003 | return ret; | |
6004 | } | |
6005 | ||
d352ac68 CM |
6006 | /* |
6007 | * helper to find a free sequence number in a given directory. This current | |
6008 | * code is very simple, later versions will do smarter things in the btree | |
6009 | */ | |
877574e2 | 6010 | int btrfs_set_inode_index(struct btrfs_inode *dir, u64 *index) |
aec7477b JB |
6011 | { |
6012 | int ret = 0; | |
6013 | ||
877574e2 NB |
6014 | if (dir->index_cnt == (u64)-1) { |
6015 | ret = btrfs_inode_delayed_dir_index_count(dir); | |
16cdcec7 MX |
6016 | if (ret) { |
6017 | ret = btrfs_set_inode_index_count(dir); | |
6018 | if (ret) | |
6019 | return ret; | |
6020 | } | |
aec7477b JB |
6021 | } |
6022 | ||
877574e2 NB |
6023 | *index = dir->index_cnt; |
6024 | dir->index_cnt++; | |
aec7477b JB |
6025 | |
6026 | return ret; | |
6027 | } | |
6028 | ||
b0d5d10f CM |
6029 | static int btrfs_insert_inode_locked(struct inode *inode) |
6030 | { | |
6031 | struct btrfs_iget_args args; | |
0202e83f DS |
6032 | |
6033 | args.ino = BTRFS_I(inode)->location.objectid; | |
b0d5d10f CM |
6034 | args.root = BTRFS_I(inode)->root; |
6035 | ||
6036 | return insert_inode_locked4(inode, | |
6037 | btrfs_inode_hash(inode->i_ino, BTRFS_I(inode)->root), | |
6038 | btrfs_find_actor, &args); | |
6039 | } | |
6040 | ||
19aee8de AJ |
6041 | /* |
6042 | * Inherit flags from the parent inode. | |
6043 | * | |
6044 | * Currently only the compression flags and the cow flags are inherited. | |
6045 | */ | |
6046 | static void btrfs_inherit_iflags(struct inode *inode, struct inode *dir) | |
6047 | { | |
6048 | unsigned int flags; | |
6049 | ||
6050 | if (!dir) | |
6051 | return; | |
6052 | ||
6053 | flags = BTRFS_I(dir)->flags; | |
6054 | ||
6055 | if (flags & BTRFS_INODE_NOCOMPRESS) { | |
6056 | BTRFS_I(inode)->flags &= ~BTRFS_INODE_COMPRESS; | |
6057 | BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS; | |
6058 | } else if (flags & BTRFS_INODE_COMPRESS) { | |
6059 | BTRFS_I(inode)->flags &= ~BTRFS_INODE_NOCOMPRESS; | |
6060 | BTRFS_I(inode)->flags |= BTRFS_INODE_COMPRESS; | |
6061 | } | |
6062 | ||
6063 | if (flags & BTRFS_INODE_NODATACOW) { | |
6064 | BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW; | |
6065 | if (S_ISREG(inode->i_mode)) | |
6066 | BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM; | |
6067 | } | |
6068 | ||
7b6a221e | 6069 | btrfs_sync_inode_flags_to_i_flags(inode); |
19aee8de AJ |
6070 | } |
6071 | ||
39279cc3 CM |
6072 | static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans, |
6073 | struct btrfs_root *root, | |
b3b6f5b9 | 6074 | struct user_namespace *mnt_userns, |
aec7477b | 6075 | struct inode *dir, |
9c58309d | 6076 | const char *name, int name_len, |
6437d458 | 6077 | umode_t mode, u64 *index) |
39279cc3 | 6078 | { |
0b246afa | 6079 | struct btrfs_fs_info *fs_info = root->fs_info; |
39279cc3 | 6080 | struct inode *inode; |
5f39d397 | 6081 | struct btrfs_inode_item *inode_item; |
39279cc3 | 6082 | struct btrfs_key *location; |
5f39d397 | 6083 | struct btrfs_path *path; |
6437d458 | 6084 | u64 objectid; |
9c58309d CM |
6085 | struct btrfs_inode_ref *ref; |
6086 | struct btrfs_key key[2]; | |
6087 | u32 sizes[2]; | |
b7ef5f3a | 6088 | struct btrfs_item_batch batch; |
9c58309d | 6089 | unsigned long ptr; |
11a19a90 | 6090 | unsigned int nofs_flag; |
39279cc3 | 6091 | int ret; |
39279cc3 | 6092 | |
5f39d397 | 6093 | path = btrfs_alloc_path(); |
d8926bb3 MF |
6094 | if (!path) |
6095 | return ERR_PTR(-ENOMEM); | |
5f39d397 | 6096 | |
11a19a90 | 6097 | nofs_flag = memalloc_nofs_save(); |
0b246afa | 6098 | inode = new_inode(fs_info->sb); |
11a19a90 | 6099 | memalloc_nofs_restore(nofs_flag); |
8fb27640 YS |
6100 | if (!inode) { |
6101 | btrfs_free_path(path); | |
39279cc3 | 6102 | return ERR_PTR(-ENOMEM); |
8fb27640 | 6103 | } |
39279cc3 | 6104 | |
5762b5c9 FM |
6105 | /* |
6106 | * O_TMPFILE, set link count to 0, so that after this point, | |
6107 | * we fill in an inode item with the correct link count. | |
6108 | */ | |
6109 | if (!name) | |
6110 | set_nlink(inode, 0); | |
6111 | ||
6437d458 OS |
6112 | ret = btrfs_get_free_objectid(root, &objectid); |
6113 | if (ret) { | |
6114 | btrfs_free_path(path); | |
6115 | iput(inode); | |
6116 | return ERR_PTR(ret); | |
6117 | } | |
581bb050 LZ |
6118 | inode->i_ino = objectid; |
6119 | ||
ef3b9af5 | 6120 | if (dir && name) { |
1abe9b8a | 6121 | trace_btrfs_inode_request(dir); |
6122 | ||
877574e2 | 6123 | ret = btrfs_set_inode_index(BTRFS_I(dir), index); |
09771430 | 6124 | if (ret) { |
8fb27640 | 6125 | btrfs_free_path(path); |
09771430 | 6126 | iput(inode); |
aec7477b | 6127 | return ERR_PTR(ret); |
09771430 | 6128 | } |
ef3b9af5 FM |
6129 | } else if (dir) { |
6130 | *index = 0; | |
aec7477b JB |
6131 | } |
6132 | /* | |
6133 | * index_cnt is ignored for everything but a dir, | |
df6703e1 | 6134 | * btrfs_set_inode_index_count has an explanation for the magic |
aec7477b JB |
6135 | * number |
6136 | */ | |
6137 | BTRFS_I(inode)->index_cnt = 2; | |
67de1176 | 6138 | BTRFS_I(inode)->dir_index = *index; |
5c8fd99f | 6139 | BTRFS_I(inode)->root = btrfs_grab_root(root); |
e02119d5 | 6140 | BTRFS_I(inode)->generation = trans->transid; |
76195853 | 6141 | inode->i_generation = BTRFS_I(inode)->generation; |
b888db2b | 6142 | |
305eaac0 OS |
6143 | btrfs_inherit_iflags(inode, dir); |
6144 | ||
6145 | if (S_ISREG(mode)) { | |
6146 | if (btrfs_test_opt(fs_info, NODATASUM)) | |
6147 | BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM; | |
6148 | if (btrfs_test_opt(fs_info, NODATACOW)) | |
6149 | BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW | | |
6150 | BTRFS_INODE_NODATASUM; | |
6151 | } | |
6152 | ||
5dc562c5 JB |
6153 | /* |
6154 | * We could have gotten an inode number from somebody who was fsynced | |
6155 | * and then removed in this same transaction, so let's just set full | |
6156 | * sync since it will be a full sync anyway and this will blow away the | |
6157 | * old info in the log. | |
6158 | */ | |
23e3337f | 6159 | btrfs_set_inode_full_sync(BTRFS_I(inode)); |
5dc562c5 | 6160 | |
9c58309d | 6161 | key[0].objectid = objectid; |
962a298f | 6162 | key[0].type = BTRFS_INODE_ITEM_KEY; |
9c58309d CM |
6163 | key[0].offset = 0; |
6164 | ||
9c58309d | 6165 | sizes[0] = sizeof(struct btrfs_inode_item); |
ef3b9af5 FM |
6166 | |
6167 | if (name) { | |
6168 | /* | |
6169 | * Start new inodes with an inode_ref. This is slightly more | |
6170 | * efficient for small numbers of hard links since they will | |
6171 | * be packed into one item. Extended refs will kick in if we | |
6172 | * add more hard links than can fit in the ref item. | |
6173 | */ | |
6174 | key[1].objectid = objectid; | |
962a298f | 6175 | key[1].type = BTRFS_INODE_REF_KEY; |
23c24ef8 OS |
6176 | if (dir) |
6177 | key[1].offset = btrfs_ino(BTRFS_I(dir)); | |
6178 | else | |
6179 | key[1].offset = objectid; | |
ef3b9af5 FM |
6180 | |
6181 | sizes[1] = name_len + sizeof(*ref); | |
6182 | } | |
9c58309d | 6183 | |
b0d5d10f CM |
6184 | location = &BTRFS_I(inode)->location; |
6185 | location->objectid = objectid; | |
6186 | location->offset = 0; | |
962a298f | 6187 | location->type = BTRFS_INODE_ITEM_KEY; |
b0d5d10f CM |
6188 | |
6189 | ret = btrfs_insert_inode_locked(inode); | |
32955c54 AV |
6190 | if (ret < 0) { |
6191 | iput(inode); | |
b0d5d10f | 6192 | goto fail; |
32955c54 | 6193 | } |
b0d5d10f | 6194 | |
b7ef5f3a FM |
6195 | batch.keys = &key[0]; |
6196 | batch.data_sizes = &sizes[0]; | |
6197 | batch.total_data_size = sizes[0] + (name ? sizes[1] : 0); | |
6198 | batch.nr = name ? 2 : 1; | |
6199 | ret = btrfs_insert_empty_items(trans, root, path, &batch); | |
9c58309d | 6200 | if (ret != 0) |
b0d5d10f | 6201 | goto fail_unlock; |
5f39d397 | 6202 | |
b3b6f5b9 | 6203 | inode_init_owner(mnt_userns, inode, dir, mode); |
9cc97d64 | 6204 | |
c2050a45 | 6205 | inode->i_mtime = current_time(inode); |
9cc97d64 | 6206 | inode->i_atime = inode->i_mtime; |
6207 | inode->i_ctime = inode->i_mtime; | |
d3c6be6f | 6208 | BTRFS_I(inode)->i_otime = inode->i_mtime; |
9cc97d64 | 6209 | |
5f39d397 CM |
6210 | inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0], |
6211 | struct btrfs_inode_item); | |
b159fa28 | 6212 | memzero_extent_buffer(path->nodes[0], (unsigned long)inode_item, |
293f7e07 | 6213 | sizeof(*inode_item)); |
e02119d5 | 6214 | fill_inode_item(trans, path->nodes[0], inode_item, inode); |
9c58309d | 6215 | |
ef3b9af5 FM |
6216 | if (name) { |
6217 | ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1, | |
6218 | struct btrfs_inode_ref); | |
6219 | btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len); | |
6220 | btrfs_set_inode_ref_index(path->nodes[0], ref, *index); | |
6221 | ptr = (unsigned long)(ref + 1); | |
6222 | write_extent_buffer(path->nodes[0], name, ptr, name_len); | |
6223 | } | |
9c58309d | 6224 | |
5f39d397 CM |
6225 | btrfs_mark_buffer_dirty(path->nodes[0]); |
6226 | btrfs_free_path(path); | |
6227 | ||
5d4f98a2 | 6228 | inode_tree_add(inode); |
1abe9b8a | 6229 | |
6230 | trace_btrfs_inode_new(inode); | |
d9094414 | 6231 | btrfs_set_inode_last_trans(trans, BTRFS_I(inode)); |
1abe9b8a | 6232 | |
8ea05e3a AB |
6233 | btrfs_update_root_times(trans, root); |
6234 | ||
63541927 FDBM |
6235 | ret = btrfs_inode_inherit_props(trans, inode, dir); |
6236 | if (ret) | |
0b246afa | 6237 | btrfs_err(fs_info, |
63541927 | 6238 | "error inheriting props for ino %llu (root %llu): %d", |
f85b7379 | 6239 | btrfs_ino(BTRFS_I(inode)), root->root_key.objectid, ret); |
63541927 | 6240 | |
39279cc3 | 6241 | return inode; |
b0d5d10f CM |
6242 | |
6243 | fail_unlock: | |
32955c54 | 6244 | discard_new_inode(inode); |
5f39d397 | 6245 | fail: |
ef3b9af5 | 6246 | if (dir && name) |
aec7477b | 6247 | BTRFS_I(dir)->index_cnt--; |
5f39d397 CM |
6248 | btrfs_free_path(path); |
6249 | return ERR_PTR(ret); | |
39279cc3 CM |
6250 | } |
6251 | ||
d352ac68 CM |
6252 | /* |
6253 | * utility function to add 'inode' into 'parent_inode' with | |
6254 | * a give name and a given sequence number. | |
6255 | * if 'add_backref' is true, also insert a backref from the | |
6256 | * inode to the parent directory. | |
6257 | */ | |
e02119d5 | 6258 | int btrfs_add_link(struct btrfs_trans_handle *trans, |
db0a669f | 6259 | struct btrfs_inode *parent_inode, struct btrfs_inode *inode, |
e02119d5 | 6260 | const char *name, int name_len, int add_backref, u64 index) |
39279cc3 | 6261 | { |
4df27c4d | 6262 | int ret = 0; |
39279cc3 | 6263 | struct btrfs_key key; |
db0a669f NB |
6264 | struct btrfs_root *root = parent_inode->root; |
6265 | u64 ino = btrfs_ino(inode); | |
6266 | u64 parent_ino = btrfs_ino(parent_inode); | |
5f39d397 | 6267 | |
33345d01 | 6268 | if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) { |
db0a669f | 6269 | memcpy(&key, &inode->root->root_key, sizeof(key)); |
4df27c4d | 6270 | } else { |
33345d01 | 6271 | key.objectid = ino; |
962a298f | 6272 | key.type = BTRFS_INODE_ITEM_KEY; |
4df27c4d YZ |
6273 | key.offset = 0; |
6274 | } | |
6275 | ||
33345d01 | 6276 | if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) { |
6025c19f | 6277 | ret = btrfs_add_root_ref(trans, key.objectid, |
0b246afa JM |
6278 | root->root_key.objectid, parent_ino, |
6279 | index, name, name_len); | |
4df27c4d | 6280 | } else if (add_backref) { |
33345d01 LZ |
6281 | ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino, |
6282 | parent_ino, index); | |
4df27c4d | 6283 | } |
39279cc3 | 6284 | |
79787eaa JM |
6285 | /* Nothing to clean up yet */ |
6286 | if (ret) | |
6287 | return ret; | |
4df27c4d | 6288 | |
684572df | 6289 | ret = btrfs_insert_dir_item(trans, name, name_len, parent_inode, &key, |
db0a669f | 6290 | btrfs_inode_type(&inode->vfs_inode), index); |
9c52057c | 6291 | if (ret == -EEXIST || ret == -EOVERFLOW) |
79787eaa JM |
6292 | goto fail_dir_item; |
6293 | else if (ret) { | |
66642832 | 6294 | btrfs_abort_transaction(trans, ret); |
79787eaa | 6295 | return ret; |
39279cc3 | 6296 | } |
79787eaa | 6297 | |
db0a669f | 6298 | btrfs_i_size_write(parent_inode, parent_inode->vfs_inode.i_size + |
79787eaa | 6299 | name_len * 2); |
db0a669f | 6300 | inode_inc_iversion(&parent_inode->vfs_inode); |
5338e43a FM |
6301 | /* |
6302 | * If we are replaying a log tree, we do not want to update the mtime | |
6303 | * and ctime of the parent directory with the current time, since the | |
6304 | * log replay procedure is responsible for setting them to their correct | |
6305 | * values (the ones it had when the fsync was done). | |
6306 | */ | |
6307 | if (!test_bit(BTRFS_FS_LOG_RECOVERING, &root->fs_info->flags)) { | |
6308 | struct timespec64 now = current_time(&parent_inode->vfs_inode); | |
6309 | ||
6310 | parent_inode->vfs_inode.i_mtime = now; | |
6311 | parent_inode->vfs_inode.i_ctime = now; | |
6312 | } | |
9a56fcd1 | 6313 | ret = btrfs_update_inode(trans, root, parent_inode); |
79787eaa | 6314 | if (ret) |
66642832 | 6315 | btrfs_abort_transaction(trans, ret); |
39279cc3 | 6316 | return ret; |
fe66a05a CM |
6317 | |
6318 | fail_dir_item: | |
6319 | if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) { | |
6320 | u64 local_index; | |
6321 | int err; | |
3ee1c553 | 6322 | err = btrfs_del_root_ref(trans, key.objectid, |
0b246afa JM |
6323 | root->root_key.objectid, parent_ino, |
6324 | &local_index, name, name_len); | |
1690dd41 JT |
6325 | if (err) |
6326 | btrfs_abort_transaction(trans, err); | |
fe66a05a CM |
6327 | } else if (add_backref) { |
6328 | u64 local_index; | |
6329 | int err; | |
6330 | ||
6331 | err = btrfs_del_inode_ref(trans, root, name, name_len, | |
6332 | ino, parent_ino, &local_index); | |
1690dd41 JT |
6333 | if (err) |
6334 | btrfs_abort_transaction(trans, err); | |
fe66a05a | 6335 | } |
1690dd41 JT |
6336 | |
6337 | /* Return the original error code */ | |
fe66a05a | 6338 | return ret; |
39279cc3 CM |
6339 | } |
6340 | ||
549c7297 CB |
6341 | static int btrfs_mknod(struct user_namespace *mnt_userns, struct inode *dir, |
6342 | struct dentry *dentry, umode_t mode, dev_t rdev) | |
618e21d5 | 6343 | { |
2ff7e61e | 6344 | struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb); |
618e21d5 JB |
6345 | struct btrfs_trans_handle *trans; |
6346 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
1832a6d5 | 6347 | struct inode *inode = NULL; |
618e21d5 | 6348 | int err; |
00e4e6b3 | 6349 | u64 index = 0; |
618e21d5 | 6350 | |
9ed74f2d JB |
6351 | /* |
6352 | * 2 for inode item and ref | |
6353 | * 2 for dir items | |
6354 | * 1 for xattr if selinux is on | |
6355 | */ | |
a22285a6 YZ |
6356 | trans = btrfs_start_transaction(root, 5); |
6357 | if (IS_ERR(trans)) | |
6358 | return PTR_ERR(trans); | |
1832a6d5 | 6359 | |
72105277 | 6360 | inode = btrfs_new_inode(trans, root, mnt_userns, dir, |
b3b6f5b9 | 6361 | dentry->d_name.name, dentry->d_name.len, |
6437d458 | 6362 | mode, &index); |
7cf96da3 TI |
6363 | if (IS_ERR(inode)) { |
6364 | err = PTR_ERR(inode); | |
32955c54 | 6365 | inode = NULL; |
618e21d5 | 6366 | goto out_unlock; |
7cf96da3 | 6367 | } |
618e21d5 | 6368 | |
ad19db71 CS |
6369 | /* |
6370 | * If the active LSM wants to access the inode during | |
6371 | * d_instantiate it needs these. Smack checks to see | |
6372 | * if the filesystem supports xattrs by looking at the | |
6373 | * ops vector. | |
6374 | */ | |
ad19db71 | 6375 | inode->i_op = &btrfs_special_inode_operations; |
b0d5d10f CM |
6376 | init_special_inode(inode, inode->i_mode, rdev); |
6377 | ||
6378 | err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name); | |
618e21d5 | 6379 | if (err) |
32955c54 | 6380 | goto out_unlock; |
b0d5d10f | 6381 | |
81512e89 OS |
6382 | err = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode), |
6383 | dentry->d_name.name, dentry->d_name.len, 0, index); | |
32955c54 AV |
6384 | if (err) |
6385 | goto out_unlock; | |
6386 | ||
9a56fcd1 | 6387 | btrfs_update_inode(trans, root, BTRFS_I(inode)); |
32955c54 | 6388 | d_instantiate_new(dentry, inode); |
b0d5d10f | 6389 | |
618e21d5 | 6390 | out_unlock: |
3a45bb20 | 6391 | btrfs_end_transaction(trans); |
2ff7e61e | 6392 | btrfs_btree_balance_dirty(fs_info); |
32955c54 | 6393 | if (err && inode) { |
618e21d5 | 6394 | inode_dec_link_count(inode); |
32955c54 | 6395 | discard_new_inode(inode); |
618e21d5 | 6396 | } |
618e21d5 JB |
6397 | return err; |
6398 | } | |
6399 | ||
549c7297 CB |
6400 | static int btrfs_create(struct user_namespace *mnt_userns, struct inode *dir, |
6401 | struct dentry *dentry, umode_t mode, bool excl) | |
39279cc3 | 6402 | { |
2ff7e61e | 6403 | struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb); |
39279cc3 CM |
6404 | struct btrfs_trans_handle *trans; |
6405 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
1832a6d5 | 6406 | struct inode *inode = NULL; |
a22285a6 | 6407 | int err; |
00e4e6b3 | 6408 | u64 index = 0; |
39279cc3 | 6409 | |
9ed74f2d JB |
6410 | /* |
6411 | * 2 for inode item and ref | |
6412 | * 2 for dir items | |
6413 | * 1 for xattr if selinux is on | |
6414 | */ | |
a22285a6 YZ |
6415 | trans = btrfs_start_transaction(root, 5); |
6416 | if (IS_ERR(trans)) | |
6417 | return PTR_ERR(trans); | |
9ed74f2d | 6418 | |
e93ca491 | 6419 | inode = btrfs_new_inode(trans, root, mnt_userns, dir, |
b3b6f5b9 | 6420 | dentry->d_name.name, dentry->d_name.len, |
6437d458 | 6421 | mode, &index); |
7cf96da3 TI |
6422 | if (IS_ERR(inode)) { |
6423 | err = PTR_ERR(inode); | |
32955c54 | 6424 | inode = NULL; |
39279cc3 | 6425 | goto out_unlock; |
7cf96da3 | 6426 | } |
ad19db71 CS |
6427 | /* |
6428 | * If the active LSM wants to access the inode during | |
6429 | * d_instantiate it needs these. Smack checks to see | |
6430 | * if the filesystem supports xattrs by looking at the | |
6431 | * ops vector. | |
6432 | */ | |
6433 | inode->i_fop = &btrfs_file_operations; | |
6434 | inode->i_op = &btrfs_file_inode_operations; | |
b0d5d10f | 6435 | inode->i_mapping->a_ops = &btrfs_aops; |
b0d5d10f CM |
6436 | |
6437 | err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name); | |
6438 | if (err) | |
32955c54 | 6439 | goto out_unlock; |
b0d5d10f | 6440 | |
9a56fcd1 | 6441 | err = btrfs_update_inode(trans, root, BTRFS_I(inode)); |
b0d5d10f | 6442 | if (err) |
32955c54 | 6443 | goto out_unlock; |
ad19db71 | 6444 | |
81512e89 OS |
6445 | err = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode), |
6446 | dentry->d_name.name, dentry->d_name.len, 0, index); | |
39279cc3 | 6447 | if (err) |
32955c54 | 6448 | goto out_unlock; |
43baa579 | 6449 | |
1e2e547a | 6450 | d_instantiate_new(dentry, inode); |
43baa579 | 6451 | |
39279cc3 | 6452 | out_unlock: |
3a45bb20 | 6453 | btrfs_end_transaction(trans); |
32955c54 | 6454 | if (err && inode) { |
39279cc3 | 6455 | inode_dec_link_count(inode); |
32955c54 | 6456 | discard_new_inode(inode); |
39279cc3 | 6457 | } |
2ff7e61e | 6458 | btrfs_btree_balance_dirty(fs_info); |
39279cc3 CM |
6459 | return err; |
6460 | } | |
6461 | ||
6462 | static int btrfs_link(struct dentry *old_dentry, struct inode *dir, | |
6463 | struct dentry *dentry) | |
6464 | { | |
271dba45 | 6465 | struct btrfs_trans_handle *trans = NULL; |
39279cc3 | 6466 | struct btrfs_root *root = BTRFS_I(dir)->root; |
2b0143b5 | 6467 | struct inode *inode = d_inode(old_dentry); |
2ff7e61e | 6468 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
00e4e6b3 | 6469 | u64 index; |
39279cc3 CM |
6470 | int err; |
6471 | int drop_inode = 0; | |
6472 | ||
4a8be425 | 6473 | /* do not allow sys_link's with other subvols of the same device */ |
4fd786e6 | 6474 | if (root->root_key.objectid != BTRFS_I(inode)->root->root_key.objectid) |
3ab3564f | 6475 | return -EXDEV; |
4a8be425 | 6476 | |
f186373f | 6477 | if (inode->i_nlink >= BTRFS_LINK_MAX) |
c055e99e | 6478 | return -EMLINK; |
4a8be425 | 6479 | |
877574e2 | 6480 | err = btrfs_set_inode_index(BTRFS_I(dir), &index); |
aec7477b JB |
6481 | if (err) |
6482 | goto fail; | |
6483 | ||
a22285a6 | 6484 | /* |
7e6b6465 | 6485 | * 2 items for inode and inode ref |
a22285a6 | 6486 | * 2 items for dir items |
7e6b6465 | 6487 | * 1 item for parent inode |
399b0bbf | 6488 | * 1 item for orphan item deletion if O_TMPFILE |
a22285a6 | 6489 | */ |
399b0bbf | 6490 | trans = btrfs_start_transaction(root, inode->i_nlink ? 5 : 6); |
a22285a6 YZ |
6491 | if (IS_ERR(trans)) { |
6492 | err = PTR_ERR(trans); | |
271dba45 | 6493 | trans = NULL; |
a22285a6 YZ |
6494 | goto fail; |
6495 | } | |
5f39d397 | 6496 | |
67de1176 MX |
6497 | /* There are several dir indexes for this inode, clear the cache. */ |
6498 | BTRFS_I(inode)->dir_index = 0ULL; | |
8b558c5f | 6499 | inc_nlink(inode); |
0c4d2d95 | 6500 | inode_inc_iversion(inode); |
c2050a45 | 6501 | inode->i_ctime = current_time(inode); |
7de9c6ee | 6502 | ihold(inode); |
e9976151 | 6503 | set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags); |
aec7477b | 6504 | |
81512e89 OS |
6505 | err = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode), |
6506 | dentry->d_name.name, dentry->d_name.len, 1, index); | |
5f39d397 | 6507 | |
a5719521 | 6508 | if (err) { |
54aa1f4d | 6509 | drop_inode = 1; |
a5719521 | 6510 | } else { |
10d9f309 | 6511 | struct dentry *parent = dentry->d_parent; |
d4682ba0 | 6512 | |
9a56fcd1 | 6513 | err = btrfs_update_inode(trans, root, BTRFS_I(inode)); |
79787eaa JM |
6514 | if (err) |
6515 | goto fail; | |
ef3b9af5 FM |
6516 | if (inode->i_nlink == 1) { |
6517 | /* | |
6518 | * If new hard link count is 1, it's a file created | |
6519 | * with open(2) O_TMPFILE flag. | |
6520 | */ | |
3d6ae7bb | 6521 | err = btrfs_orphan_del(trans, BTRFS_I(inode)); |
ef3b9af5 FM |
6522 | if (err) |
6523 | goto fail; | |
6524 | } | |
08c422c2 | 6525 | d_instantiate(dentry, inode); |
88d2beec | 6526 | btrfs_log_new_name(trans, old_dentry, NULL, 0, parent); |
a5719521 | 6527 | } |
39279cc3 | 6528 | |
1832a6d5 | 6529 | fail: |
271dba45 | 6530 | if (trans) |
3a45bb20 | 6531 | btrfs_end_transaction(trans); |
39279cc3 CM |
6532 | if (drop_inode) { |
6533 | inode_dec_link_count(inode); | |
6534 | iput(inode); | |
6535 | } | |
2ff7e61e | 6536 | btrfs_btree_balance_dirty(fs_info); |
39279cc3 CM |
6537 | return err; |
6538 | } | |
6539 | ||
549c7297 CB |
6540 | static int btrfs_mkdir(struct user_namespace *mnt_userns, struct inode *dir, |
6541 | struct dentry *dentry, umode_t mode) | |
39279cc3 | 6542 | { |
2ff7e61e | 6543 | struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb); |
b9d86667 | 6544 | struct inode *inode = NULL; |
39279cc3 CM |
6545 | struct btrfs_trans_handle *trans; |
6546 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
6547 | int err = 0; | |
00e4e6b3 | 6548 | u64 index = 0; |
39279cc3 | 6549 | |
9ed74f2d JB |
6550 | /* |
6551 | * 2 items for inode and ref | |
6552 | * 2 items for dir items | |
6553 | * 1 for xattr if selinux is on | |
6554 | */ | |
a22285a6 YZ |
6555 | trans = btrfs_start_transaction(root, 5); |
6556 | if (IS_ERR(trans)) | |
6557 | return PTR_ERR(trans); | |
39279cc3 | 6558 | |
b0b3e44d | 6559 | inode = btrfs_new_inode(trans, root, mnt_userns, dir, |
b3b6f5b9 | 6560 | dentry->d_name.name, dentry->d_name.len, |
f85b7379 | 6561 | S_IFDIR | mode, &index); |
39279cc3 CM |
6562 | if (IS_ERR(inode)) { |
6563 | err = PTR_ERR(inode); | |
32955c54 | 6564 | inode = NULL; |
39279cc3 CM |
6565 | goto out_fail; |
6566 | } | |
5f39d397 | 6567 | |
b0d5d10f CM |
6568 | /* these must be set before we unlock the inode */ |
6569 | inode->i_op = &btrfs_dir_inode_operations; | |
6570 | inode->i_fop = &btrfs_dir_file_operations; | |
33268eaf | 6571 | |
2a7dba39 | 6572 | err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name); |
33268eaf | 6573 | if (err) |
32955c54 | 6574 | goto out_fail; |
39279cc3 | 6575 | |
9a56fcd1 | 6576 | err = btrfs_update_inode(trans, root, BTRFS_I(inode)); |
39279cc3 | 6577 | if (err) |
32955c54 | 6578 | goto out_fail; |
5f39d397 | 6579 | |
db0a669f NB |
6580 | err = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode), |
6581 | dentry->d_name.name, | |
6582 | dentry->d_name.len, 0, index); | |
39279cc3 | 6583 | if (err) |
32955c54 | 6584 | goto out_fail; |
5f39d397 | 6585 | |
1e2e547a | 6586 | d_instantiate_new(dentry, inode); |
39279cc3 CM |
6587 | |
6588 | out_fail: | |
3a45bb20 | 6589 | btrfs_end_transaction(trans); |
32955c54 | 6590 | if (err && inode) { |
c7cfb8a5 | 6591 | inode_dec_link_count(inode); |
32955c54 | 6592 | discard_new_inode(inode); |
c7cfb8a5 | 6593 | } |
2ff7e61e | 6594 | btrfs_btree_balance_dirty(fs_info); |
39279cc3 CM |
6595 | return err; |
6596 | } | |
6597 | ||
c8b97818 | 6598 | static noinline int uncompress_inline(struct btrfs_path *path, |
e40da0e5 | 6599 | struct page *page, |
c8b97818 CM |
6600 | size_t pg_offset, u64 extent_offset, |
6601 | struct btrfs_file_extent_item *item) | |
6602 | { | |
6603 | int ret; | |
6604 | struct extent_buffer *leaf = path->nodes[0]; | |
6605 | char *tmp; | |
6606 | size_t max_size; | |
6607 | unsigned long inline_size; | |
6608 | unsigned long ptr; | |
261507a0 | 6609 | int compress_type; |
c8b97818 CM |
6610 | |
6611 | WARN_ON(pg_offset != 0); | |
261507a0 | 6612 | compress_type = btrfs_file_extent_compression(leaf, item); |
c8b97818 | 6613 | max_size = btrfs_file_extent_ram_bytes(leaf, item); |
437bd07e | 6614 | inline_size = btrfs_file_extent_inline_item_len(leaf, path->slots[0]); |
c8b97818 | 6615 | tmp = kmalloc(inline_size, GFP_NOFS); |
8d413713 TI |
6616 | if (!tmp) |
6617 | return -ENOMEM; | |
c8b97818 CM |
6618 | ptr = btrfs_file_extent_inline_start(item); |
6619 | ||
6620 | read_extent_buffer(leaf, tmp, ptr, inline_size); | |
6621 | ||
09cbfeaf | 6622 | max_size = min_t(unsigned long, PAGE_SIZE, max_size); |
261507a0 LZ |
6623 | ret = btrfs_decompress(compress_type, tmp, page, |
6624 | extent_offset, inline_size, max_size); | |
e1699d2d ZB |
6625 | |
6626 | /* | |
6627 | * decompression code contains a memset to fill in any space between the end | |
6628 | * of the uncompressed data and the end of max_size in case the decompressed | |
6629 | * data ends up shorter than ram_bytes. That doesn't cover the hole between | |
6630 | * the end of an inline extent and the beginning of the next block, so we | |
6631 | * cover that region here. | |
6632 | */ | |
6633 | ||
d048b9c2 IW |
6634 | if (max_size + pg_offset < PAGE_SIZE) |
6635 | memzero_page(page, pg_offset + max_size, | |
6636 | PAGE_SIZE - max_size - pg_offset); | |
c8b97818 | 6637 | kfree(tmp); |
166ae5a4 | 6638 | return ret; |
c8b97818 CM |
6639 | } |
6640 | ||
39b07b5d OS |
6641 | /** |
6642 | * btrfs_get_extent - Lookup the first extent overlapping a range in a file. | |
6643 | * @inode: file to search in | |
6644 | * @page: page to read extent data into if the extent is inline | |
6645 | * @pg_offset: offset into @page to copy to | |
6646 | * @start: file offset | |
6647 | * @len: length of range starting at @start | |
6648 | * | |
6649 | * This returns the first &struct extent_map which overlaps with the given | |
6650 | * range, reading it from the B-tree and caching it if necessary. Note that | |
6651 | * there may be more extents which overlap the given range after the returned | |
6652 | * extent_map. | |
d352ac68 | 6653 | * |
39b07b5d OS |
6654 | * If @page is not NULL and the extent is inline, this also reads the extent |
6655 | * data directly into the page and marks the extent up to date in the io_tree. | |
6656 | * | |
6657 | * Return: ERR_PTR on error, non-NULL extent_map on success. | |
d352ac68 | 6658 | */ |
fc4f21b1 | 6659 | struct extent_map *btrfs_get_extent(struct btrfs_inode *inode, |
39b07b5d OS |
6660 | struct page *page, size_t pg_offset, |
6661 | u64 start, u64 len) | |
a52d9a80 | 6662 | { |
3ffbd68c | 6663 | struct btrfs_fs_info *fs_info = inode->root->fs_info; |
1028d1c4 | 6664 | int ret = 0; |
a52d9a80 CM |
6665 | u64 extent_start = 0; |
6666 | u64 extent_end = 0; | |
fc4f21b1 | 6667 | u64 objectid = btrfs_ino(inode); |
7e74e235 | 6668 | int extent_type = -1; |
f421950f | 6669 | struct btrfs_path *path = NULL; |
fc4f21b1 | 6670 | struct btrfs_root *root = inode->root; |
a52d9a80 | 6671 | struct btrfs_file_extent_item *item; |
5f39d397 CM |
6672 | struct extent_buffer *leaf; |
6673 | struct btrfs_key found_key; | |
a52d9a80 | 6674 | struct extent_map *em = NULL; |
fc4f21b1 NB |
6675 | struct extent_map_tree *em_tree = &inode->extent_tree; |
6676 | struct extent_io_tree *io_tree = &inode->io_tree; | |
a52d9a80 | 6677 | |
890871be | 6678 | read_lock(&em_tree->lock); |
d1310b2e | 6679 | em = lookup_extent_mapping(em_tree, start, len); |
890871be | 6680 | read_unlock(&em_tree->lock); |
d1310b2e | 6681 | |
a52d9a80 | 6682 | if (em) { |
e1c4b745 CM |
6683 | if (em->start > start || em->start + em->len <= start) |
6684 | free_extent_map(em); | |
6685 | else if (em->block_start == EXTENT_MAP_INLINE && page) | |
70dec807 CM |
6686 | free_extent_map(em); |
6687 | else | |
6688 | goto out; | |
a52d9a80 | 6689 | } |
172ddd60 | 6690 | em = alloc_extent_map(); |
a52d9a80 | 6691 | if (!em) { |
1028d1c4 | 6692 | ret = -ENOMEM; |
d1310b2e | 6693 | goto out; |
a52d9a80 | 6694 | } |
d1310b2e | 6695 | em->start = EXTENT_MAP_HOLE; |
445a6944 | 6696 | em->orig_start = EXTENT_MAP_HOLE; |
d1310b2e | 6697 | em->len = (u64)-1; |
c8b97818 | 6698 | em->block_len = (u64)-1; |
f421950f | 6699 | |
bee6ec82 | 6700 | path = btrfs_alloc_path(); |
f421950f | 6701 | if (!path) { |
1028d1c4 | 6702 | ret = -ENOMEM; |
bee6ec82 | 6703 | goto out; |
f421950f CM |
6704 | } |
6705 | ||
bee6ec82 LB |
6706 | /* Chances are we'll be called again, so go ahead and do readahead */ |
6707 | path->reada = READA_FORWARD; | |
4d7240f0 JB |
6708 | |
6709 | /* | |
6710 | * The same explanation in load_free_space_cache applies here as well, | |
6711 | * we only read when we're loading the free space cache, and at that | |
6712 | * point the commit_root has everything we need. | |
6713 | */ | |
6714 | if (btrfs_is_free_space_inode(inode)) { | |
6715 | path->search_commit_root = 1; | |
6716 | path->skip_locking = 1; | |
6717 | } | |
51899412 | 6718 | |
5c9a702e | 6719 | ret = btrfs_lookup_file_extent(NULL, root, path, objectid, start, 0); |
a52d9a80 | 6720 | if (ret < 0) { |
a52d9a80 | 6721 | goto out; |
b8eeab7f | 6722 | } else if (ret > 0) { |
a52d9a80 CM |
6723 | if (path->slots[0] == 0) |
6724 | goto not_found; | |
6725 | path->slots[0]--; | |
1028d1c4 | 6726 | ret = 0; |
a52d9a80 CM |
6727 | } |
6728 | ||
5f39d397 CM |
6729 | leaf = path->nodes[0]; |
6730 | item = btrfs_item_ptr(leaf, path->slots[0], | |
a52d9a80 | 6731 | struct btrfs_file_extent_item); |
5f39d397 | 6732 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
5f39d397 | 6733 | if (found_key.objectid != objectid || |
694c12ed | 6734 | found_key.type != BTRFS_EXTENT_DATA_KEY) { |
25a50341 JB |
6735 | /* |
6736 | * If we backup past the first extent we want to move forward | |
6737 | * and see if there is an extent in front of us, otherwise we'll | |
6738 | * say there is a hole for our whole search range which can | |
6739 | * cause problems. | |
6740 | */ | |
6741 | extent_end = start; | |
6742 | goto next; | |
a52d9a80 CM |
6743 | } |
6744 | ||
694c12ed | 6745 | extent_type = btrfs_file_extent_type(leaf, item); |
5f39d397 | 6746 | extent_start = found_key.offset; |
a5eeb3d1 | 6747 | extent_end = btrfs_file_extent_end(path); |
694c12ed NB |
6748 | if (extent_type == BTRFS_FILE_EXTENT_REG || |
6749 | extent_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
6bf9e4bd QW |
6750 | /* Only regular file could have regular/prealloc extent */ |
6751 | if (!S_ISREG(inode->vfs_inode.i_mode)) { | |
1028d1c4 | 6752 | ret = -EUCLEAN; |
6bf9e4bd QW |
6753 | btrfs_crit(fs_info, |
6754 | "regular/prealloc extent found for non-regular inode %llu", | |
6755 | btrfs_ino(inode)); | |
6756 | goto out; | |
6757 | } | |
09ed2f16 LB |
6758 | trace_btrfs_get_extent_show_fi_regular(inode, leaf, item, |
6759 | extent_start); | |
694c12ed | 6760 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
09ed2f16 LB |
6761 | trace_btrfs_get_extent_show_fi_inline(inode, leaf, item, |
6762 | path->slots[0], | |
6763 | extent_start); | |
9036c102 | 6764 | } |
25a50341 | 6765 | next: |
9036c102 YZ |
6766 | if (start >= extent_end) { |
6767 | path->slots[0]++; | |
6768 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
6769 | ret = btrfs_next_leaf(root, path); | |
1028d1c4 | 6770 | if (ret < 0) |
9036c102 | 6771 | goto out; |
1028d1c4 | 6772 | else if (ret > 0) |
9036c102 | 6773 | goto not_found; |
1028d1c4 | 6774 | |
9036c102 | 6775 | leaf = path->nodes[0]; |
a52d9a80 | 6776 | } |
9036c102 YZ |
6777 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
6778 | if (found_key.objectid != objectid || | |
6779 | found_key.type != BTRFS_EXTENT_DATA_KEY) | |
6780 | goto not_found; | |
6781 | if (start + len <= found_key.offset) | |
6782 | goto not_found; | |
e2eca69d WS |
6783 | if (start > found_key.offset) |
6784 | goto next; | |
02a033df NB |
6785 | |
6786 | /* New extent overlaps with existing one */ | |
9036c102 | 6787 | em->start = start; |
70c8a91c | 6788 | em->orig_start = start; |
9036c102 | 6789 | em->len = found_key.offset - start; |
02a033df NB |
6790 | em->block_start = EXTENT_MAP_HOLE; |
6791 | goto insert; | |
9036c102 YZ |
6792 | } |
6793 | ||
39b07b5d | 6794 | btrfs_extent_item_to_extent_map(inode, path, item, !page, em); |
7ffbb598 | 6795 | |
694c12ed NB |
6796 | if (extent_type == BTRFS_FILE_EXTENT_REG || |
6797 | extent_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
a52d9a80 | 6798 | goto insert; |
694c12ed | 6799 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
5f39d397 | 6800 | unsigned long ptr; |
a52d9a80 | 6801 | char *map; |
3326d1b0 CM |
6802 | size_t size; |
6803 | size_t extent_offset; | |
6804 | size_t copy_size; | |
a52d9a80 | 6805 | |
39b07b5d | 6806 | if (!page) |
689f9346 | 6807 | goto out; |
5f39d397 | 6808 | |
e41ca589 | 6809 | size = btrfs_file_extent_ram_bytes(leaf, item); |
9036c102 | 6810 | extent_offset = page_offset(page) + pg_offset - extent_start; |
09cbfeaf KS |
6811 | copy_size = min_t(u64, PAGE_SIZE - pg_offset, |
6812 | size - extent_offset); | |
3326d1b0 | 6813 | em->start = extent_start + extent_offset; |
0b246afa | 6814 | em->len = ALIGN(copy_size, fs_info->sectorsize); |
b4939680 | 6815 | em->orig_block_len = em->len; |
70c8a91c | 6816 | em->orig_start = em->start; |
689f9346 | 6817 | ptr = btrfs_file_extent_inline_start(item) + extent_offset; |
e49aabd9 | 6818 | |
bf46f52d | 6819 | if (!PageUptodate(page)) { |
261507a0 LZ |
6820 | if (btrfs_file_extent_compression(leaf, item) != |
6821 | BTRFS_COMPRESS_NONE) { | |
e40da0e5 | 6822 | ret = uncompress_inline(path, page, pg_offset, |
c8b97818 | 6823 | extent_offset, item); |
1028d1c4 | 6824 | if (ret) |
166ae5a4 | 6825 | goto out; |
c8b97818 | 6826 | } else { |
58c1a35c | 6827 | map = kmap_local_page(page); |
c8b97818 CM |
6828 | read_extent_buffer(leaf, map + pg_offset, ptr, |
6829 | copy_size); | |
09cbfeaf | 6830 | if (pg_offset + copy_size < PAGE_SIZE) { |
93c82d57 | 6831 | memset(map + pg_offset + copy_size, 0, |
09cbfeaf | 6832 | PAGE_SIZE - pg_offset - |
93c82d57 CM |
6833 | copy_size); |
6834 | } | |
58c1a35c | 6835 | kunmap_local(map); |
c8b97818 | 6836 | } |
179e29e4 | 6837 | flush_dcache_page(page); |
a52d9a80 | 6838 | } |
d1310b2e | 6839 | set_extent_uptodate(io_tree, em->start, |
507903b8 | 6840 | extent_map_end(em) - 1, NULL, GFP_NOFS); |
a52d9a80 | 6841 | goto insert; |
a52d9a80 CM |
6842 | } |
6843 | not_found: | |
6844 | em->start = start; | |
70c8a91c | 6845 | em->orig_start = start; |
d1310b2e | 6846 | em->len = len; |
5f39d397 | 6847 | em->block_start = EXTENT_MAP_HOLE; |
a52d9a80 | 6848 | insert: |
1028d1c4 | 6849 | ret = 0; |
b3b4aa74 | 6850 | btrfs_release_path(path); |
d1310b2e | 6851 | if (em->start > start || extent_map_end(em) <= start) { |
0b246afa | 6852 | btrfs_err(fs_info, |
5d163e0e JM |
6853 | "bad extent! em: [%llu %llu] passed [%llu %llu]", |
6854 | em->start, em->len, start, len); | |
1028d1c4 | 6855 | ret = -EIO; |
a52d9a80 CM |
6856 | goto out; |
6857 | } | |
d1310b2e | 6858 | |
890871be | 6859 | write_lock(&em_tree->lock); |
1028d1c4 | 6860 | ret = btrfs_add_extent_mapping(fs_info, em_tree, &em, start, len); |
890871be | 6861 | write_unlock(&em_tree->lock); |
a52d9a80 | 6862 | out: |
c6414280 | 6863 | btrfs_free_path(path); |
1abe9b8a | 6864 | |
fc4f21b1 | 6865 | trace_btrfs_get_extent(root, inode, em); |
1abe9b8a | 6866 | |
1028d1c4 | 6867 | if (ret) { |
a52d9a80 | 6868 | free_extent_map(em); |
1028d1c4 | 6869 | return ERR_PTR(ret); |
a52d9a80 CM |
6870 | } |
6871 | return em; | |
6872 | } | |
6873 | ||
fc4f21b1 | 6874 | struct extent_map *btrfs_get_extent_fiemap(struct btrfs_inode *inode, |
4ab47a8d | 6875 | u64 start, u64 len) |
ec29ed5b CM |
6876 | { |
6877 | struct extent_map *em; | |
6878 | struct extent_map *hole_em = NULL; | |
f3714ef4 | 6879 | u64 delalloc_start = start; |
ec29ed5b | 6880 | u64 end; |
f3714ef4 NB |
6881 | u64 delalloc_len; |
6882 | u64 delalloc_end; | |
ec29ed5b CM |
6883 | int err = 0; |
6884 | ||
39b07b5d | 6885 | em = btrfs_get_extent(inode, NULL, 0, start, len); |
ec29ed5b CM |
6886 | if (IS_ERR(em)) |
6887 | return em; | |
9986277e DC |
6888 | /* |
6889 | * If our em maps to: | |
6890 | * - a hole or | |
6891 | * - a pre-alloc extent, | |
6892 | * there might actually be delalloc bytes behind it. | |
6893 | */ | |
6894 | if (em->block_start != EXTENT_MAP_HOLE && | |
6895 | !test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) | |
6896 | return em; | |
6897 | else | |
6898 | hole_em = em; | |
ec29ed5b CM |
6899 | |
6900 | /* check to see if we've wrapped (len == -1 or similar) */ | |
6901 | end = start + len; | |
6902 | if (end < start) | |
6903 | end = (u64)-1; | |
6904 | else | |
6905 | end -= 1; | |
6906 | ||
6907 | em = NULL; | |
6908 | ||
6909 | /* ok, we didn't find anything, lets look for delalloc */ | |
f3714ef4 | 6910 | delalloc_len = count_range_bits(&inode->io_tree, &delalloc_start, |
ec29ed5b | 6911 | end, len, EXTENT_DELALLOC, 1); |
f3714ef4 NB |
6912 | delalloc_end = delalloc_start + delalloc_len; |
6913 | if (delalloc_end < delalloc_start) | |
6914 | delalloc_end = (u64)-1; | |
ec29ed5b CM |
6915 | |
6916 | /* | |
f3714ef4 NB |
6917 | * We didn't find anything useful, return the original results from |
6918 | * get_extent() | |
ec29ed5b | 6919 | */ |
f3714ef4 | 6920 | if (delalloc_start > end || delalloc_end <= start) { |
ec29ed5b CM |
6921 | em = hole_em; |
6922 | hole_em = NULL; | |
6923 | goto out; | |
6924 | } | |
6925 | ||
f3714ef4 NB |
6926 | /* |
6927 | * Adjust the delalloc_start to make sure it doesn't go backwards from | |
6928 | * the start they passed in | |
ec29ed5b | 6929 | */ |
f3714ef4 NB |
6930 | delalloc_start = max(start, delalloc_start); |
6931 | delalloc_len = delalloc_end - delalloc_start; | |
ec29ed5b | 6932 | |
f3714ef4 NB |
6933 | if (delalloc_len > 0) { |
6934 | u64 hole_start; | |
02950af4 | 6935 | u64 hole_len; |
f3714ef4 | 6936 | const u64 hole_end = extent_map_end(hole_em); |
ec29ed5b | 6937 | |
172ddd60 | 6938 | em = alloc_extent_map(); |
ec29ed5b CM |
6939 | if (!em) { |
6940 | err = -ENOMEM; | |
6941 | goto out; | |
6942 | } | |
f3714ef4 NB |
6943 | |
6944 | ASSERT(hole_em); | |
ec29ed5b | 6945 | /* |
f3714ef4 NB |
6946 | * When btrfs_get_extent can't find anything it returns one |
6947 | * huge hole | |
ec29ed5b | 6948 | * |
f3714ef4 NB |
6949 | * Make sure what it found really fits our range, and adjust to |
6950 | * make sure it is based on the start from the caller | |
ec29ed5b | 6951 | */ |
f3714ef4 NB |
6952 | if (hole_end <= start || hole_em->start > end) { |
6953 | free_extent_map(hole_em); | |
6954 | hole_em = NULL; | |
6955 | } else { | |
6956 | hole_start = max(hole_em->start, start); | |
6957 | hole_len = hole_end - hole_start; | |
ec29ed5b | 6958 | } |
f3714ef4 NB |
6959 | |
6960 | if (hole_em && delalloc_start > hole_start) { | |
6961 | /* | |
6962 | * Our hole starts before our delalloc, so we have to | |
6963 | * return just the parts of the hole that go until the | |
6964 | * delalloc starts | |
ec29ed5b | 6965 | */ |
f3714ef4 | 6966 | em->len = min(hole_len, delalloc_start - hole_start); |
ec29ed5b CM |
6967 | em->start = hole_start; |
6968 | em->orig_start = hole_start; | |
6969 | /* | |
f3714ef4 NB |
6970 | * Don't adjust block start at all, it is fixed at |
6971 | * EXTENT_MAP_HOLE | |
ec29ed5b CM |
6972 | */ |
6973 | em->block_start = hole_em->block_start; | |
6974 | em->block_len = hole_len; | |
f9e4fb53 LB |
6975 | if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags)) |
6976 | set_bit(EXTENT_FLAG_PREALLOC, &em->flags); | |
ec29ed5b | 6977 | } else { |
f3714ef4 NB |
6978 | /* |
6979 | * Hole is out of passed range or it starts after | |
6980 | * delalloc range | |
6981 | */ | |
6982 | em->start = delalloc_start; | |
6983 | em->len = delalloc_len; | |
6984 | em->orig_start = delalloc_start; | |
ec29ed5b | 6985 | em->block_start = EXTENT_MAP_DELALLOC; |
f3714ef4 | 6986 | em->block_len = delalloc_len; |
ec29ed5b | 6987 | } |
bf8d32b9 | 6988 | } else { |
ec29ed5b CM |
6989 | return hole_em; |
6990 | } | |
6991 | out: | |
6992 | ||
6993 | free_extent_map(hole_em); | |
6994 | if (err) { | |
6995 | free_extent_map(em); | |
6996 | return ERR_PTR(err); | |
6997 | } | |
6998 | return em; | |
6999 | } | |
7000 | ||
64f54188 | 7001 | static struct extent_map *btrfs_create_dio_extent(struct btrfs_inode *inode, |
5f9a8a51 FM |
7002 | const u64 start, |
7003 | const u64 len, | |
7004 | const u64 orig_start, | |
7005 | const u64 block_start, | |
7006 | const u64 block_len, | |
7007 | const u64 orig_block_len, | |
7008 | const u64 ram_bytes, | |
7009 | const int type) | |
7010 | { | |
7011 | struct extent_map *em = NULL; | |
7012 | int ret; | |
7013 | ||
5f9a8a51 | 7014 | if (type != BTRFS_ORDERED_NOCOW) { |
64f54188 NB |
7015 | em = create_io_em(inode, start, len, orig_start, block_start, |
7016 | block_len, orig_block_len, ram_bytes, | |
6f9994db LB |
7017 | BTRFS_COMPRESS_NONE, /* compress_type */ |
7018 | type); | |
5f9a8a51 FM |
7019 | if (IS_ERR(em)) |
7020 | goto out; | |
7021 | } | |
cb36a9bb OS |
7022 | ret = btrfs_add_ordered_extent(inode, start, len, len, block_start, |
7023 | block_len, 0, | |
7024 | (1 << type) | | |
7025 | (1 << BTRFS_ORDERED_DIRECT), | |
7026 | BTRFS_COMPRESS_NONE); | |
5f9a8a51 FM |
7027 | if (ret) { |
7028 | if (em) { | |
7029 | free_extent_map(em); | |
64f54188 | 7030 | btrfs_drop_extent_cache(inode, start, start + len - 1, 0); |
5f9a8a51 FM |
7031 | } |
7032 | em = ERR_PTR(ret); | |
7033 | } | |
7034 | out: | |
5f9a8a51 FM |
7035 | |
7036 | return em; | |
7037 | } | |
7038 | ||
9fc6f911 | 7039 | static struct extent_map *btrfs_new_extent_direct(struct btrfs_inode *inode, |
4b46fce2 JB |
7040 | u64 start, u64 len) |
7041 | { | |
9fc6f911 NB |
7042 | struct btrfs_root *root = inode->root; |
7043 | struct btrfs_fs_info *fs_info = root->fs_info; | |
70c8a91c | 7044 | struct extent_map *em; |
4b46fce2 JB |
7045 | struct btrfs_key ins; |
7046 | u64 alloc_hint; | |
7047 | int ret; | |
4b46fce2 | 7048 | |
9fc6f911 | 7049 | alloc_hint = get_extent_allocation_hint(inode, start, len); |
0b246afa | 7050 | ret = btrfs_reserve_extent(root, len, len, fs_info->sectorsize, |
da17066c | 7051 | 0, alloc_hint, &ins, 1, 1); |
00361589 JB |
7052 | if (ret) |
7053 | return ERR_PTR(ret); | |
4b46fce2 | 7054 | |
9fc6f911 | 7055 | em = btrfs_create_dio_extent(inode, start, ins.offset, start, |
5f9a8a51 | 7056 | ins.objectid, ins.offset, ins.offset, |
6288d6ea | 7057 | ins.offset, BTRFS_ORDERED_REGULAR); |
0b246afa | 7058 | btrfs_dec_block_group_reservations(fs_info, ins.objectid); |
5f9a8a51 | 7059 | if (IS_ERR(em)) |
9fc6f911 NB |
7060 | btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, |
7061 | 1); | |
de0ee0ed | 7062 | |
4b46fce2 JB |
7063 | return em; |
7064 | } | |
7065 | ||
f4639636 | 7066 | static bool btrfs_extent_readonly(struct btrfs_fs_info *fs_info, u64 bytenr) |
05947ae1 AJ |
7067 | { |
7068 | struct btrfs_block_group *block_group; | |
f4639636 | 7069 | bool readonly = false; |
05947ae1 AJ |
7070 | |
7071 | block_group = btrfs_lookup_block_group(fs_info, bytenr); | |
7072 | if (!block_group || block_group->ro) | |
f4639636 | 7073 | readonly = true; |
05947ae1 AJ |
7074 | if (block_group) |
7075 | btrfs_put_block_group(block_group); | |
7076 | return readonly; | |
7077 | } | |
7078 | ||
46bfbb5c | 7079 | /* |
e4ecaf90 QW |
7080 | * Check if we can do nocow write into the range [@offset, @offset + @len) |
7081 | * | |
7082 | * @offset: File offset | |
7083 | * @len: The length to write, will be updated to the nocow writeable | |
7084 | * range | |
7085 | * @orig_start: (optional) Return the original file offset of the file extent | |
7086 | * @orig_len: (optional) Return the original on-disk length of the file extent | |
7087 | * @ram_bytes: (optional) Return the ram_bytes of the file extent | |
a84d5d42 BB |
7088 | * @strict: if true, omit optimizations that might force us into unnecessary |
7089 | * cow. e.g., don't trust generation number. | |
e4ecaf90 | 7090 | * |
e4ecaf90 QW |
7091 | * Return: |
7092 | * >0 and update @len if we can do nocow write | |
7093 | * 0 if we can't do nocow write | |
7094 | * <0 if error happened | |
7095 | * | |
7096 | * NOTE: This only checks the file extents, caller is responsible to wait for | |
7097 | * any ordered extents. | |
46bfbb5c | 7098 | */ |
00361589 | 7099 | noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len, |
7ee9e440 | 7100 | u64 *orig_start, u64 *orig_block_len, |
a84d5d42 | 7101 | u64 *ram_bytes, bool strict) |
46bfbb5c | 7102 | { |
2ff7e61e | 7103 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
46bfbb5c CM |
7104 | struct btrfs_path *path; |
7105 | int ret; | |
7106 | struct extent_buffer *leaf; | |
7107 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
7b2b7085 | 7108 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
46bfbb5c CM |
7109 | struct btrfs_file_extent_item *fi; |
7110 | struct btrfs_key key; | |
7111 | u64 disk_bytenr; | |
7112 | u64 backref_offset; | |
7113 | u64 extent_end; | |
7114 | u64 num_bytes; | |
7115 | int slot; | |
7116 | int found_type; | |
7ee9e440 | 7117 | bool nocow = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW); |
e77751aa | 7118 | |
46bfbb5c CM |
7119 | path = btrfs_alloc_path(); |
7120 | if (!path) | |
7121 | return -ENOMEM; | |
7122 | ||
f85b7379 DS |
7123 | ret = btrfs_lookup_file_extent(NULL, root, path, |
7124 | btrfs_ino(BTRFS_I(inode)), offset, 0); | |
46bfbb5c CM |
7125 | if (ret < 0) |
7126 | goto out; | |
7127 | ||
7128 | slot = path->slots[0]; | |
7129 | if (ret == 1) { | |
7130 | if (slot == 0) { | |
7131 | /* can't find the item, must cow */ | |
7132 | ret = 0; | |
7133 | goto out; | |
7134 | } | |
7135 | slot--; | |
7136 | } | |
7137 | ret = 0; | |
7138 | leaf = path->nodes[0]; | |
7139 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
4a0cc7ca | 7140 | if (key.objectid != btrfs_ino(BTRFS_I(inode)) || |
46bfbb5c CM |
7141 | key.type != BTRFS_EXTENT_DATA_KEY) { |
7142 | /* not our file or wrong item type, must cow */ | |
7143 | goto out; | |
7144 | } | |
7145 | ||
7146 | if (key.offset > offset) { | |
7147 | /* Wrong offset, must cow */ | |
7148 | goto out; | |
7149 | } | |
7150 | ||
7151 | fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item); | |
7152 | found_type = btrfs_file_extent_type(leaf, fi); | |
7153 | if (found_type != BTRFS_FILE_EXTENT_REG && | |
7154 | found_type != BTRFS_FILE_EXTENT_PREALLOC) { | |
7155 | /* not a regular extent, must cow */ | |
7156 | goto out; | |
7157 | } | |
7ee9e440 JB |
7158 | |
7159 | if (!nocow && found_type == BTRFS_FILE_EXTENT_REG) | |
7160 | goto out; | |
7161 | ||
e77751aa MX |
7162 | extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi); |
7163 | if (extent_end <= offset) | |
7164 | goto out; | |
7165 | ||
46bfbb5c | 7166 | disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); |
7ee9e440 JB |
7167 | if (disk_bytenr == 0) |
7168 | goto out; | |
7169 | ||
7170 | if (btrfs_file_extent_compression(leaf, fi) || | |
7171 | btrfs_file_extent_encryption(leaf, fi) || | |
7172 | btrfs_file_extent_other_encoding(leaf, fi)) | |
7173 | goto out; | |
7174 | ||
78d4295b EL |
7175 | /* |
7176 | * Do the same check as in btrfs_cross_ref_exist but without the | |
7177 | * unnecessary search. | |
7178 | */ | |
a84d5d42 BB |
7179 | if (!strict && |
7180 | (btrfs_file_extent_generation(leaf, fi) <= | |
7181 | btrfs_root_last_snapshot(&root->root_item))) | |
78d4295b EL |
7182 | goto out; |
7183 | ||
46bfbb5c CM |
7184 | backref_offset = btrfs_file_extent_offset(leaf, fi); |
7185 | ||
7ee9e440 JB |
7186 | if (orig_start) { |
7187 | *orig_start = key.offset - backref_offset; | |
7188 | *orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi); | |
7189 | *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi); | |
7190 | } | |
eb384b55 | 7191 | |
2ff7e61e | 7192 | if (btrfs_extent_readonly(fs_info, disk_bytenr)) |
46bfbb5c | 7193 | goto out; |
7b2b7085 MX |
7194 | |
7195 | num_bytes = min(offset + *len, extent_end) - offset; | |
7196 | if (!nocow && found_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
7197 | u64 range_end; | |
7198 | ||
da17066c JM |
7199 | range_end = round_up(offset + num_bytes, |
7200 | root->fs_info->sectorsize) - 1; | |
7b2b7085 MX |
7201 | ret = test_range_bit(io_tree, offset, range_end, |
7202 | EXTENT_DELALLOC, 0, NULL); | |
7203 | if (ret) { | |
7204 | ret = -EAGAIN; | |
7205 | goto out; | |
7206 | } | |
7207 | } | |
7208 | ||
1bda19eb | 7209 | btrfs_release_path(path); |
46bfbb5c CM |
7210 | |
7211 | /* | |
7212 | * look for other files referencing this extent, if we | |
7213 | * find any we must cow | |
7214 | */ | |
00361589 | 7215 | |
e4c3b2dc | 7216 | ret = btrfs_cross_ref_exist(root, btrfs_ino(BTRFS_I(inode)), |
a84d5d42 BB |
7217 | key.offset - backref_offset, disk_bytenr, |
7218 | strict); | |
00361589 JB |
7219 | if (ret) { |
7220 | ret = 0; | |
7221 | goto out; | |
7222 | } | |
46bfbb5c CM |
7223 | |
7224 | /* | |
7225 | * adjust disk_bytenr and num_bytes to cover just the bytes | |
7226 | * in this extent we are about to write. If there | |
7227 | * are any csums in that range we have to cow in order | |
7228 | * to keep the csums correct | |
7229 | */ | |
7230 | disk_bytenr += backref_offset; | |
7231 | disk_bytenr += offset - key.offset; | |
2ff7e61e JM |
7232 | if (csum_exist_in_range(fs_info, disk_bytenr, num_bytes)) |
7233 | goto out; | |
46bfbb5c CM |
7234 | /* |
7235 | * all of the above have passed, it is safe to overwrite this extent | |
7236 | * without cow | |
7237 | */ | |
eb384b55 | 7238 | *len = num_bytes; |
46bfbb5c CM |
7239 | ret = 1; |
7240 | out: | |
7241 | btrfs_free_path(path); | |
7242 | return ret; | |
7243 | } | |
7244 | ||
eb838e73 | 7245 | static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend, |
f85781fb | 7246 | struct extent_state **cached_state, bool writing) |
eb838e73 JB |
7247 | { |
7248 | struct btrfs_ordered_extent *ordered; | |
7249 | int ret = 0; | |
7250 | ||
7251 | while (1) { | |
7252 | lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend, | |
ff13db41 | 7253 | cached_state); |
eb838e73 JB |
7254 | /* |
7255 | * We're concerned with the entire range that we're going to be | |
01327610 | 7256 | * doing DIO to, so we need to make sure there's no ordered |
eb838e73 JB |
7257 | * extents in this range. |
7258 | */ | |
a776c6fa | 7259 | ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), lockstart, |
eb838e73 JB |
7260 | lockend - lockstart + 1); |
7261 | ||
7262 | /* | |
7263 | * We need to make sure there are no buffered pages in this | |
7264 | * range either, we could have raced between the invalidate in | |
7265 | * generic_file_direct_write and locking the extent. The | |
7266 | * invalidate needs to happen so that reads after a write do not | |
7267 | * get stale data. | |
7268 | */ | |
fc4adbff | 7269 | if (!ordered && |
051c98eb DS |
7270 | (!writing || !filemap_range_has_page(inode->i_mapping, |
7271 | lockstart, lockend))) | |
eb838e73 JB |
7272 | break; |
7273 | ||
7274 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend, | |
e43bbe5e | 7275 | cached_state); |
eb838e73 JB |
7276 | |
7277 | if (ordered) { | |
ade77029 FM |
7278 | /* |
7279 | * If we are doing a DIO read and the ordered extent we | |
7280 | * found is for a buffered write, we can not wait for it | |
7281 | * to complete and retry, because if we do so we can | |
7282 | * deadlock with concurrent buffered writes on page | |
7283 | * locks. This happens only if our DIO read covers more | |
7284 | * than one extent map, if at this point has already | |
7285 | * created an ordered extent for a previous extent map | |
7286 | * and locked its range in the inode's io tree, and a | |
7287 | * concurrent write against that previous extent map's | |
7288 | * range and this range started (we unlock the ranges | |
7289 | * in the io tree only when the bios complete and | |
7290 | * buffered writes always lock pages before attempting | |
7291 | * to lock range in the io tree). | |
7292 | */ | |
7293 | if (writing || | |
7294 | test_bit(BTRFS_ORDERED_DIRECT, &ordered->flags)) | |
c0a43603 | 7295 | btrfs_start_ordered_extent(ordered, 1); |
ade77029 FM |
7296 | else |
7297 | ret = -ENOTBLK; | |
eb838e73 JB |
7298 | btrfs_put_ordered_extent(ordered); |
7299 | } else { | |
eb838e73 | 7300 | /* |
b850ae14 FM |
7301 | * We could trigger writeback for this range (and wait |
7302 | * for it to complete) and then invalidate the pages for | |
7303 | * this range (through invalidate_inode_pages2_range()), | |
7304 | * but that can lead us to a deadlock with a concurrent | |
ba206a02 | 7305 | * call to readahead (a buffered read or a defrag call |
b850ae14 FM |
7306 | * triggered a readahead) on a page lock due to an |
7307 | * ordered dio extent we created before but did not have | |
7308 | * yet a corresponding bio submitted (whence it can not | |
ba206a02 | 7309 | * complete), which makes readahead wait for that |
b850ae14 FM |
7310 | * ordered extent to complete while holding a lock on |
7311 | * that page. | |
eb838e73 | 7312 | */ |
b850ae14 | 7313 | ret = -ENOTBLK; |
eb838e73 JB |
7314 | } |
7315 | ||
ade77029 FM |
7316 | if (ret) |
7317 | break; | |
7318 | ||
eb838e73 JB |
7319 | cond_resched(); |
7320 | } | |
7321 | ||
7322 | return ret; | |
7323 | } | |
7324 | ||
6f9994db | 7325 | /* The callers of this must take lock_extent() */ |
4b67c11d NB |
7326 | static struct extent_map *create_io_em(struct btrfs_inode *inode, u64 start, |
7327 | u64 len, u64 orig_start, u64 block_start, | |
6f9994db LB |
7328 | u64 block_len, u64 orig_block_len, |
7329 | u64 ram_bytes, int compress_type, | |
7330 | int type) | |
69ffb543 JB |
7331 | { |
7332 | struct extent_map_tree *em_tree; | |
7333 | struct extent_map *em; | |
69ffb543 JB |
7334 | int ret; |
7335 | ||
6f9994db LB |
7336 | ASSERT(type == BTRFS_ORDERED_PREALLOC || |
7337 | type == BTRFS_ORDERED_COMPRESSED || | |
7338 | type == BTRFS_ORDERED_NOCOW || | |
1af4a0aa | 7339 | type == BTRFS_ORDERED_REGULAR); |
6f9994db | 7340 | |
4b67c11d | 7341 | em_tree = &inode->extent_tree; |
69ffb543 JB |
7342 | em = alloc_extent_map(); |
7343 | if (!em) | |
7344 | return ERR_PTR(-ENOMEM); | |
7345 | ||
7346 | em->start = start; | |
7347 | em->orig_start = orig_start; | |
7348 | em->len = len; | |
7349 | em->block_len = block_len; | |
7350 | em->block_start = block_start; | |
b4939680 | 7351 | em->orig_block_len = orig_block_len; |
cc95bef6 | 7352 | em->ram_bytes = ram_bytes; |
70c8a91c | 7353 | em->generation = -1; |
69ffb543 | 7354 | set_bit(EXTENT_FLAG_PINNED, &em->flags); |
1af4a0aa | 7355 | if (type == BTRFS_ORDERED_PREALLOC) { |
b11e234d | 7356 | set_bit(EXTENT_FLAG_FILLING, &em->flags); |
1af4a0aa | 7357 | } else if (type == BTRFS_ORDERED_COMPRESSED) { |
6f9994db LB |
7358 | set_bit(EXTENT_FLAG_COMPRESSED, &em->flags); |
7359 | em->compress_type = compress_type; | |
7360 | } | |
69ffb543 JB |
7361 | |
7362 | do { | |
4b67c11d NB |
7363 | btrfs_drop_extent_cache(inode, em->start, |
7364 | em->start + em->len - 1, 0); | |
69ffb543 | 7365 | write_lock(&em_tree->lock); |
09a2a8f9 | 7366 | ret = add_extent_mapping(em_tree, em, 1); |
69ffb543 | 7367 | write_unlock(&em_tree->lock); |
6f9994db LB |
7368 | /* |
7369 | * The caller has taken lock_extent(), who could race with us | |
7370 | * to add em? | |
7371 | */ | |
69ffb543 JB |
7372 | } while (ret == -EEXIST); |
7373 | ||
7374 | if (ret) { | |
7375 | free_extent_map(em); | |
7376 | return ERR_PTR(ret); | |
7377 | } | |
7378 | ||
6f9994db | 7379 | /* em got 2 refs now, callers needs to do free_extent_map once. */ |
69ffb543 JB |
7380 | return em; |
7381 | } | |
7382 | ||
1c8d0175 | 7383 | |
c5794e51 | 7384 | static int btrfs_get_blocks_direct_write(struct extent_map **map, |
c5794e51 NB |
7385 | struct inode *inode, |
7386 | struct btrfs_dio_data *dio_data, | |
7387 | u64 start, u64 len) | |
7388 | { | |
7389 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); | |
7390 | struct extent_map *em = *map; | |
f0bfa76a FM |
7391 | int type; |
7392 | u64 block_start, orig_start, orig_block_len, ram_bytes; | |
7393 | bool can_nocow = false; | |
7394 | bool space_reserved = false; | |
6d82ad13 | 7395 | u64 prev_len; |
c5794e51 NB |
7396 | int ret = 0; |
7397 | ||
7398 | /* | |
7399 | * We don't allocate a new extent in the following cases | |
7400 | * | |
7401 | * 1) The inode is marked as NODATACOW. In this case we'll just use the | |
7402 | * existing extent. | |
7403 | * 2) The extent is marked as PREALLOC. We're good to go here and can | |
7404 | * just use the extent. | |
7405 | * | |
7406 | */ | |
7407 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) || | |
7408 | ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) && | |
7409 | em->block_start != EXTENT_MAP_HOLE)) { | |
c5794e51 NB |
7410 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) |
7411 | type = BTRFS_ORDERED_PREALLOC; | |
7412 | else | |
7413 | type = BTRFS_ORDERED_NOCOW; | |
7414 | len = min(len, em->len - (start - em->start)); | |
7415 | block_start = em->block_start + (start - em->start); | |
7416 | ||
7417 | if (can_nocow_extent(inode, start, &len, &orig_start, | |
a84d5d42 | 7418 | &orig_block_len, &ram_bytes, false) == 1 && |
f0bfa76a FM |
7419 | btrfs_inc_nocow_writers(fs_info, block_start)) |
7420 | can_nocow = true; | |
7421 | } | |
c5794e51 | 7422 | |
6d82ad13 | 7423 | prev_len = len; |
f0bfa76a FM |
7424 | if (can_nocow) { |
7425 | struct extent_map *em2; | |
7426 | ||
7427 | /* We can NOCOW, so only need to reserve metadata space. */ | |
28c9b1e7 | 7428 | ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode), len, len); |
f0bfa76a FM |
7429 | if (ret < 0) { |
7430 | /* Our caller expects us to free the input extent map. */ | |
7431 | free_extent_map(em); | |
7432 | *map = NULL; | |
c5794e51 | 7433 | btrfs_dec_nocow_writers(fs_info, block_start); |
f0bfa76a FM |
7434 | goto out; |
7435 | } | |
7436 | space_reserved = true; | |
7437 | ||
7438 | em2 = btrfs_create_dio_extent(BTRFS_I(inode), start, len, | |
7439 | orig_start, block_start, | |
7440 | len, orig_block_len, | |
7441 | ram_bytes, type); | |
7442 | btrfs_dec_nocow_writers(fs_info, block_start); | |
7443 | if (type == BTRFS_ORDERED_PREALLOC) { | |
7444 | free_extent_map(em); | |
7445 | *map = em = em2; | |
7446 | } | |
c5794e51 | 7447 | |
f0bfa76a FM |
7448 | if (IS_ERR(em2)) { |
7449 | ret = PTR_ERR(em2); | |
7450 | goto out; | |
c5794e51 | 7451 | } |
f0bfa76a | 7452 | } else { |
f0bfa76a FM |
7453 | /* Our caller expects us to free the input extent map. */ |
7454 | free_extent_map(em); | |
7455 | *map = NULL; | |
7456 | ||
7457 | /* We have to COW, so need to reserve metadata and data space. */ | |
7458 | ret = btrfs_delalloc_reserve_space(BTRFS_I(inode), | |
7459 | &dio_data->data_reserved, | |
7460 | start, len); | |
7461 | if (ret < 0) | |
7462 | goto out; | |
7463 | space_reserved = true; | |
7464 | ||
7465 | em = btrfs_new_extent_direct(BTRFS_I(inode), start, len); | |
7466 | if (IS_ERR(em)) { | |
7467 | ret = PTR_ERR(em); | |
7468 | goto out; | |
7469 | } | |
7470 | *map = em; | |
7471 | len = min(len, em->len - (start - em->start)); | |
7472 | if (len < prev_len) | |
7473 | btrfs_delalloc_release_space(BTRFS_I(inode), | |
7474 | dio_data->data_reserved, | |
7475 | start + len, prev_len - len, | |
7476 | true); | |
c5794e51 NB |
7477 | } |
7478 | ||
f0bfa76a FM |
7479 | /* |
7480 | * We have created our ordered extent, so we can now release our reservation | |
7481 | * for an outstanding extent. | |
7482 | */ | |
6d82ad13 | 7483 | btrfs_delalloc_release_extents(BTRFS_I(inode), prev_len); |
c5794e51 | 7484 | |
c5794e51 NB |
7485 | /* |
7486 | * Need to update the i_size under the extent lock so buffered | |
7487 | * readers will get the updated i_size when we unlock. | |
7488 | */ | |
f85781fb | 7489 | if (start + len > i_size_read(inode)) |
c5794e51 | 7490 | i_size_write(inode, start + len); |
c5794e51 | 7491 | out: |
f0bfa76a FM |
7492 | if (ret && space_reserved) { |
7493 | btrfs_delalloc_release_extents(BTRFS_I(inode), len); | |
7494 | if (can_nocow) { | |
7495 | btrfs_delalloc_release_metadata(BTRFS_I(inode), len, true); | |
7496 | } else { | |
7497 | btrfs_delalloc_release_space(BTRFS_I(inode), | |
7498 | dio_data->data_reserved, | |
7499 | start, len, true); | |
7500 | extent_changeset_free(dio_data->data_reserved); | |
7501 | dio_data->data_reserved = NULL; | |
7502 | } | |
7503 | } | |
c5794e51 NB |
7504 | return ret; |
7505 | } | |
7506 | ||
f85781fb GR |
7507 | static int btrfs_dio_iomap_begin(struct inode *inode, loff_t start, |
7508 | loff_t length, unsigned int flags, struct iomap *iomap, | |
7509 | struct iomap *srcmap) | |
4b46fce2 | 7510 | { |
0b246afa | 7511 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
4b46fce2 | 7512 | struct extent_map *em; |
eb838e73 | 7513 | struct extent_state *cached_state = NULL; |
50745b0a | 7514 | struct btrfs_dio_data *dio_data = NULL; |
eb838e73 | 7515 | u64 lockstart, lockend; |
f85781fb | 7516 | const bool write = !!(flags & IOMAP_WRITE); |
0934856d | 7517 | int ret = 0; |
f85781fb GR |
7518 | u64 len = length; |
7519 | bool unlock_extents = false; | |
eb838e73 | 7520 | |
f85781fb | 7521 | if (!write) |
0b246afa | 7522 | len = min_t(u64, len, fs_info->sectorsize); |
eb838e73 | 7523 | |
c329861d JB |
7524 | lockstart = start; |
7525 | lockend = start + len - 1; | |
7526 | ||
f85781fb GR |
7527 | /* |
7528 | * The generic stuff only does filemap_write_and_wait_range, which | |
7529 | * isn't enough if we've written compressed pages to this area, so we | |
7530 | * need to flush the dirty pages again to make absolutely sure that any | |
7531 | * outstanding dirty pages are on disk. | |
7532 | */ | |
7533 | if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT, | |
7534 | &BTRFS_I(inode)->runtime_flags)) { | |
7535 | ret = filemap_fdatawrite_range(inode->i_mapping, start, | |
7536 | start + length - 1); | |
7537 | if (ret) | |
7538 | return ret; | |
7539 | } | |
7540 | ||
7541 | dio_data = kzalloc(sizeof(*dio_data), GFP_NOFS); | |
7542 | if (!dio_data) | |
7543 | return -ENOMEM; | |
7544 | ||
f85781fb GR |
7545 | iomap->private = dio_data; |
7546 | ||
e1cbbfa5 | 7547 | |
eb838e73 JB |
7548 | /* |
7549 | * If this errors out it's because we couldn't invalidate pagecache for | |
7550 | * this range and we need to fallback to buffered. | |
7551 | */ | |
f85781fb | 7552 | if (lock_extent_direct(inode, lockstart, lockend, &cached_state, write)) { |
9c9464cc FM |
7553 | ret = -ENOTBLK; |
7554 | goto err; | |
7555 | } | |
eb838e73 | 7556 | |
39b07b5d | 7557 | em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len); |
eb838e73 JB |
7558 | if (IS_ERR(em)) { |
7559 | ret = PTR_ERR(em); | |
7560 | goto unlock_err; | |
7561 | } | |
4b46fce2 JB |
7562 | |
7563 | /* | |
7564 | * Ok for INLINE and COMPRESSED extents we need to fallback on buffered | |
7565 | * io. INLINE is special, and we could probably kludge it in here, but | |
7566 | * it's still buffered so for safety lets just fall back to the generic | |
7567 | * buffered path. | |
7568 | * | |
7569 | * For COMPRESSED we _have_ to read the entire extent in so we can | |
7570 | * decompress it, so there will be buffering required no matter what we | |
7571 | * do, so go ahead and fallback to buffered. | |
7572 | * | |
01327610 | 7573 | * We return -ENOTBLK because that's what makes DIO go ahead and go back |
4b46fce2 JB |
7574 | * to buffered IO. Don't blame me, this is the price we pay for using |
7575 | * the generic code. | |
7576 | */ | |
7577 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) || | |
7578 | em->block_start == EXTENT_MAP_INLINE) { | |
7579 | free_extent_map(em); | |
eb838e73 JB |
7580 | ret = -ENOTBLK; |
7581 | goto unlock_err; | |
4b46fce2 JB |
7582 | } |
7583 | ||
f85781fb | 7584 | len = min(len, em->len - (start - em->start)); |
ca93e44b FM |
7585 | |
7586 | /* | |
7587 | * If we have a NOWAIT request and the range contains multiple extents | |
7588 | * (or a mix of extents and holes), then we return -EAGAIN to make the | |
7589 | * caller fallback to a context where it can do a blocking (without | |
7590 | * NOWAIT) request. This way we avoid doing partial IO and returning | |
7591 | * success to the caller, which is not optimal for writes and for reads | |
7592 | * it can result in unexpected behaviour for an application. | |
7593 | * | |
7594 | * When doing a read, because we use IOMAP_DIO_PARTIAL when calling | |
7595 | * iomap_dio_rw(), we can end up returning less data then what the caller | |
7596 | * asked for, resulting in an unexpected, and incorrect, short read. | |
7597 | * That is, the caller asked to read N bytes and we return less than that, | |
7598 | * which is wrong unless we are crossing EOF. This happens if we get a | |
7599 | * page fault error when trying to fault in pages for the buffer that is | |
7600 | * associated to the struct iov_iter passed to iomap_dio_rw(), and we | |
7601 | * have previously submitted bios for other extents in the range, in | |
7602 | * which case iomap_dio_rw() may return us EIOCBQUEUED if not all of | |
7603 | * those bios have completed by the time we get the page fault error, | |
7604 | * which we return back to our caller - we should only return EIOCBQUEUED | |
7605 | * after we have submitted bios for all the extents in the range. | |
7606 | */ | |
7607 | if ((flags & IOMAP_NOWAIT) && len < length) { | |
7608 | free_extent_map(em); | |
7609 | ret = -EAGAIN; | |
7610 | goto unlock_err; | |
7611 | } | |
7612 | ||
f85781fb GR |
7613 | if (write) { |
7614 | ret = btrfs_get_blocks_direct_write(&em, inode, dio_data, | |
7615 | start, len); | |
c5794e51 NB |
7616 | if (ret < 0) |
7617 | goto unlock_err; | |
f85781fb GR |
7618 | unlock_extents = true; |
7619 | /* Recalc len in case the new em is smaller than requested */ | |
7620 | len = min(len, em->len - (start - em->start)); | |
c5794e51 | 7621 | } else { |
1c8d0175 NB |
7622 | /* |
7623 | * We need to unlock only the end area that we aren't using. | |
7624 | * The rest is going to be unlocked by the endio routine. | |
7625 | */ | |
f85781fb GR |
7626 | lockstart = start + len; |
7627 | if (lockstart < lockend) | |
7628 | unlock_extents = true; | |
7629 | } | |
7630 | ||
7631 | if (unlock_extents) | |
7632 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, | |
7633 | lockstart, lockend, &cached_state); | |
7634 | else | |
7635 | free_extent_state(cached_state); | |
7636 | ||
7637 | /* | |
7638 | * Translate extent map information to iomap. | |
7639 | * We trim the extents (and move the addr) even though iomap code does | |
7640 | * that, since we have locked only the parts we are performing I/O in. | |
7641 | */ | |
7642 | if ((em->block_start == EXTENT_MAP_HOLE) || | |
7643 | (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) && !write)) { | |
7644 | iomap->addr = IOMAP_NULL_ADDR; | |
7645 | iomap->type = IOMAP_HOLE; | |
7646 | } else { | |
7647 | iomap->addr = em->block_start + (start - em->start); | |
7648 | iomap->type = IOMAP_MAPPED; | |
a43a67a2 | 7649 | } |
f85781fb | 7650 | iomap->offset = start; |
d24fa5c1 | 7651 | iomap->bdev = fs_info->fs_devices->latest_dev->bdev; |
f85781fb | 7652 | iomap->length = len; |
a43a67a2 | 7653 | |
e380adfc | 7654 | if (write && btrfs_use_zone_append(BTRFS_I(inode), em->block_start)) |
544d24f9 NA |
7655 | iomap->flags |= IOMAP_F_ZONE_APPEND; |
7656 | ||
4b46fce2 JB |
7657 | free_extent_map(em); |
7658 | ||
7659 | return 0; | |
eb838e73 JB |
7660 | |
7661 | unlock_err: | |
e182163d OS |
7662 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend, |
7663 | &cached_state); | |
9c9464cc | 7664 | err: |
f0bfa76a FM |
7665 | kfree(dio_data); |
7666 | ||
f85781fb GR |
7667 | return ret; |
7668 | } | |
7669 | ||
7670 | static int btrfs_dio_iomap_end(struct inode *inode, loff_t pos, loff_t length, | |
7671 | ssize_t written, unsigned int flags, struct iomap *iomap) | |
7672 | { | |
7673 | int ret = 0; | |
7674 | struct btrfs_dio_data *dio_data = iomap->private; | |
7675 | size_t submitted = dio_data->submitted; | |
7676 | const bool write = !!(flags & IOMAP_WRITE); | |
7677 | ||
7678 | if (!write && (iomap->type == IOMAP_HOLE)) { | |
7679 | /* If reading from a hole, unlock and return */ | |
7680 | unlock_extent(&BTRFS_I(inode)->io_tree, pos, pos + length - 1); | |
7681 | goto out; | |
7682 | } | |
7683 | ||
7684 | if (submitted < length) { | |
7685 | pos += submitted; | |
7686 | length -= submitted; | |
7687 | if (write) | |
7688 | __endio_write_update_ordered(BTRFS_I(inode), pos, | |
7689 | length, false); | |
7690 | else | |
7691 | unlock_extent(&BTRFS_I(inode)->io_tree, pos, | |
7692 | pos + length - 1); | |
7693 | ret = -ENOTBLK; | |
7694 | } | |
7695 | ||
f0bfa76a | 7696 | if (write) |
f85781fb | 7697 | extent_changeset_free(dio_data->data_reserved); |
f85781fb GR |
7698 | out: |
7699 | kfree(dio_data); | |
7700 | iomap->private = NULL; | |
7701 | ||
8b110e39 MX |
7702 | return ret; |
7703 | } | |
7704 | ||
769b4f24 | 7705 | static void btrfs_dio_private_put(struct btrfs_dio_private *dip) |
8b110e39 | 7706 | { |
769b4f24 OS |
7707 | /* |
7708 | * This implies a barrier so that stores to dio_bio->bi_status before | |
7709 | * this and loads of dio_bio->bi_status after this are fully ordered. | |
7710 | */ | |
7711 | if (!refcount_dec_and_test(&dip->refs)) | |
7712 | return; | |
8b110e39 | 7713 | |
cfe94440 | 7714 | if (btrfs_op(dip->dio_bio) == BTRFS_MAP_WRITE) { |
b672b5c1 | 7715 | __endio_write_update_ordered(BTRFS_I(dip->inode), |
47926ab5 | 7716 | dip->file_offset, |
769b4f24 OS |
7717 | dip->bytes, |
7718 | !dip->dio_bio->bi_status); | |
7719 | } else { | |
7720 | unlock_extent(&BTRFS_I(dip->inode)->io_tree, | |
47926ab5 QW |
7721 | dip->file_offset, |
7722 | dip->file_offset + dip->bytes - 1); | |
8b110e39 MX |
7723 | } |
7724 | ||
f85781fb | 7725 | bio_endio(dip->dio_bio); |
769b4f24 | 7726 | kfree(dip); |
8b110e39 MX |
7727 | } |
7728 | ||
77d5d689 OS |
7729 | static blk_status_t submit_dio_repair_bio(struct inode *inode, struct bio *bio, |
7730 | int mirror_num, | |
7731 | unsigned long bio_flags) | |
8b110e39 | 7732 | { |
77d5d689 | 7733 | struct btrfs_dio_private *dip = bio->bi_private; |
2ff7e61e | 7734 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
58efbc9f | 7735 | blk_status_t ret; |
8b110e39 | 7736 | |
37226b21 | 7737 | BUG_ON(bio_op(bio) == REQ_OP_WRITE); |
8b110e39 | 7738 | |
5c047a69 | 7739 | ret = btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DATA); |
8b110e39 | 7740 | if (ret) |
ea057f6d | 7741 | return ret; |
8b110e39 | 7742 | |
77d5d689 | 7743 | refcount_inc(&dip->refs); |
08635bae | 7744 | ret = btrfs_map_bio(fs_info, bio, mirror_num); |
8b110e39 | 7745 | if (ret) |
fd9d6670 | 7746 | refcount_dec(&dip->refs); |
77d5d689 | 7747 | return ret; |
8b110e39 MX |
7748 | } |
7749 | ||
f4f39fc5 | 7750 | static blk_status_t btrfs_check_read_dio_bio(struct btrfs_dio_private *dip, |
c3a3b19b | 7751 | struct btrfs_bio *bbio, |
fd9d6670 | 7752 | const bool uptodate) |
4b46fce2 | 7753 | { |
f4f39fc5 | 7754 | struct inode *inode = dip->inode; |
fd9d6670 OS |
7755 | struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info; |
7756 | const u32 sectorsize = fs_info->sectorsize; | |
7757 | struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree; | |
7758 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; | |
7759 | const bool csum = !(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM); | |
17347cec LB |
7760 | struct bio_vec bvec; |
7761 | struct bvec_iter iter; | |
7ffd27e3 | 7762 | u32 bio_offset = 0; |
58efbc9f | 7763 | blk_status_t err = BLK_STS_OK; |
4b46fce2 | 7764 | |
c3a3b19b | 7765 | __bio_for_each_segment(bvec, &bbio->bio, iter, bbio->iter) { |
fd9d6670 | 7766 | unsigned int i, nr_sectors, pgoff; |
8b110e39 | 7767 | |
17347cec LB |
7768 | nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec.bv_len); |
7769 | pgoff = bvec.bv_offset; | |
fd9d6670 | 7770 | for (i = 0; i < nr_sectors; i++) { |
00d82525 CH |
7771 | u64 start = bbio->file_offset + bio_offset; |
7772 | ||
97bf5a55 | 7773 | ASSERT(pgoff < PAGE_SIZE); |
fd9d6670 | 7774 | if (uptodate && |
c3a3b19b | 7775 | (!csum || !check_data_csum(inode, bbio, |
c1d6abda OS |
7776 | bio_offset, bvec.bv_page, |
7777 | pgoff, start))) { | |
fd9d6670 OS |
7778 | clean_io_failure(fs_info, failure_tree, io_tree, |
7779 | start, bvec.bv_page, | |
7780 | btrfs_ino(BTRFS_I(inode)), | |
7781 | pgoff); | |
7782 | } else { | |
150e4b05 | 7783 | int ret; |
fd9d6670 | 7784 | |
00d82525 CH |
7785 | ret = btrfs_repair_one_sector(inode, &bbio->bio, |
7786 | bio_offset, bvec.bv_page, pgoff, | |
c3a3b19b | 7787 | start, bbio->mirror_num, |
150e4b05 QW |
7788 | submit_dio_repair_bio); |
7789 | if (ret) | |
7790 | err = errno_to_blk_status(ret); | |
fd9d6670 | 7791 | } |
7ffd27e3 QW |
7792 | ASSERT(bio_offset + sectorsize > bio_offset); |
7793 | bio_offset += sectorsize; | |
2dabb324 | 7794 | pgoff += sectorsize; |
2dabb324 | 7795 | } |
2c30c71b | 7796 | } |
c1dc0896 MX |
7797 | return err; |
7798 | } | |
7799 | ||
b672b5c1 | 7800 | static void __endio_write_update_ordered(struct btrfs_inode *inode, |
52427260 QW |
7801 | const u64 offset, const u64 bytes, |
7802 | const bool uptodate) | |
4b46fce2 | 7803 | { |
e65f152e QW |
7804 | btrfs_mark_ordered_io_finished(inode, NULL, offset, bytes, |
7805 | finish_ordered_fn, uptodate); | |
14543774 FM |
7806 | } |
7807 | ||
8896a08d | 7808 | static blk_status_t btrfs_submit_bio_start_direct_io(struct inode *inode, |
1941b64b QW |
7809 | struct bio *bio, |
7810 | u64 dio_file_offset) | |
eaf25d93 | 7811 | { |
e331f6b1 | 7812 | return btrfs_csum_one_bio(BTRFS_I(inode), bio, dio_file_offset, false); |
eaf25d93 CM |
7813 | } |
7814 | ||
4246a0b6 | 7815 | static void btrfs_end_dio_bio(struct bio *bio) |
e65e1535 MX |
7816 | { |
7817 | struct btrfs_dio_private *dip = bio->bi_private; | |
0fdf977d | 7818 | struct btrfs_bio *bbio = btrfs_bio(bio); |
4e4cbee9 | 7819 | blk_status_t err = bio->bi_status; |
e65e1535 | 7820 | |
8b110e39 MX |
7821 | if (err) |
7822 | btrfs_warn(BTRFS_I(dip->inode)->root->fs_info, | |
6296b960 | 7823 | "direct IO failed ino %llu rw %d,%u sector %#Lx len %u err no %d", |
f85b7379 | 7824 | btrfs_ino(BTRFS_I(dip->inode)), bio_op(bio), |
1201b58b | 7825 | bio->bi_opf, bio->bi_iter.bi_sector, |
8b110e39 MX |
7826 | bio->bi_iter.bi_size, err); |
7827 | ||
f4f39fc5 | 7828 | if (bio_op(bio) == REQ_OP_READ) |
0fdf977d | 7829 | err = btrfs_check_read_dio_bio(dip, bbio, !err); |
e65e1535 | 7830 | |
769b4f24 OS |
7831 | if (err) |
7832 | dip->dio_bio->bi_status = err; | |
e65e1535 | 7833 | |
0fdf977d | 7834 | btrfs_record_physical_zoned(dip->inode, bbio->file_offset, bio); |
544d24f9 | 7835 | |
e65e1535 | 7836 | bio_put(bio); |
769b4f24 | 7837 | btrfs_dio_private_put(dip); |
c1dc0896 MX |
7838 | } |
7839 | ||
d0ee3934 DS |
7840 | static inline blk_status_t btrfs_submit_dio_bio(struct bio *bio, |
7841 | struct inode *inode, u64 file_offset, int async_submit) | |
e65e1535 | 7842 | { |
0b246afa | 7843 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
facc8a22 | 7844 | struct btrfs_dio_private *dip = bio->bi_private; |
cfe94440 | 7845 | bool write = btrfs_op(bio) == BTRFS_MAP_WRITE; |
4e4cbee9 | 7846 | blk_status_t ret; |
e65e1535 | 7847 | |
4c274bc6 | 7848 | /* Check btrfs_submit_bio_hook() for rules about async submit. */ |
b812ce28 JB |
7849 | if (async_submit) |
7850 | async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers); | |
7851 | ||
5fd02043 | 7852 | if (!write) { |
0b246afa | 7853 | ret = btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DATA); |
5fd02043 JB |
7854 | if (ret) |
7855 | goto err; | |
7856 | } | |
e65e1535 | 7857 | |
e6961cac | 7858 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM) |
1ae39938 JB |
7859 | goto map; |
7860 | ||
7861 | if (write && async_submit) { | |
1941b64b | 7862 | ret = btrfs_wq_submit_bio(inode, bio, 0, 0, file_offset, |
e288c080 | 7863 | btrfs_submit_bio_start_direct_io); |
e65e1535 | 7864 | goto err; |
1ae39938 JB |
7865 | } else if (write) { |
7866 | /* | |
7867 | * If we aren't doing async submit, calculate the csum of the | |
7868 | * bio now. | |
7869 | */ | |
e331f6b1 | 7870 | ret = btrfs_csum_one_bio(BTRFS_I(inode), bio, file_offset, false); |
1ae39938 JB |
7871 | if (ret) |
7872 | goto err; | |
23ea8e5a | 7873 | } else { |
85879573 OS |
7874 | u64 csum_offset; |
7875 | ||
47926ab5 | 7876 | csum_offset = file_offset - dip->file_offset; |
265fdfa6 | 7877 | csum_offset >>= fs_info->sectorsize_bits; |
55fc29be | 7878 | csum_offset *= fs_info->csum_size; |
c3a3b19b | 7879 | btrfs_bio(bio)->csum = dip->csums + csum_offset; |
c2db1073 | 7880 | } |
1ae39938 | 7881 | map: |
08635bae | 7882 | ret = btrfs_map_bio(fs_info, bio, 0); |
e65e1535 | 7883 | err: |
e65e1535 MX |
7884 | return ret; |
7885 | } | |
7886 | ||
c36cac28 OS |
7887 | /* |
7888 | * If this succeeds, the btrfs_dio_private is responsible for cleaning up locked | |
7889 | * or ordered extents whether or not we submit any bios. | |
7890 | */ | |
7891 | static struct btrfs_dio_private *btrfs_create_dio_private(struct bio *dio_bio, | |
7892 | struct inode *inode, | |
7893 | loff_t file_offset) | |
e65e1535 | 7894 | { |
cfe94440 | 7895 | const bool write = (btrfs_op(dio_bio) == BTRFS_MAP_WRITE); |
85879573 OS |
7896 | const bool csum = !(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM); |
7897 | size_t dip_size; | |
c36cac28 | 7898 | struct btrfs_dio_private *dip; |
c36cac28 | 7899 | |
85879573 OS |
7900 | dip_size = sizeof(*dip); |
7901 | if (!write && csum) { | |
7902 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); | |
85879573 OS |
7903 | size_t nblocks; |
7904 | ||
265fdfa6 | 7905 | nblocks = dio_bio->bi_iter.bi_size >> fs_info->sectorsize_bits; |
223486c2 | 7906 | dip_size += fs_info->csum_size * nblocks; |
85879573 OS |
7907 | } |
7908 | ||
7909 | dip = kzalloc(dip_size, GFP_NOFS); | |
c36cac28 OS |
7910 | if (!dip) |
7911 | return NULL; | |
7912 | ||
c36cac28 | 7913 | dip->inode = inode; |
47926ab5 | 7914 | dip->file_offset = file_offset; |
c36cac28 | 7915 | dip->bytes = dio_bio->bi_iter.bi_size; |
1201b58b | 7916 | dip->disk_bytenr = dio_bio->bi_iter.bi_sector << 9; |
c36cac28 | 7917 | dip->dio_bio = dio_bio; |
e3b318d1 | 7918 | refcount_set(&dip->refs, 1); |
c36cac28 OS |
7919 | return dip; |
7920 | } | |
7921 | ||
3e08773c | 7922 | static void btrfs_submit_direct(const struct iomap_iter *iter, |
f85781fb | 7923 | struct bio *dio_bio, loff_t file_offset) |
c36cac28 | 7924 | { |
a6d3d495 | 7925 | struct inode *inode = iter->inode; |
cfe94440 | 7926 | const bool write = (btrfs_op(dio_bio) == BTRFS_MAP_WRITE); |
0b246afa | 7927 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
769b4f24 OS |
7928 | const bool raid56 = (btrfs_data_alloc_profile(fs_info) & |
7929 | BTRFS_BLOCK_GROUP_RAID56_MASK); | |
c36cac28 | 7930 | struct btrfs_dio_private *dip; |
e65e1535 | 7931 | struct bio *bio; |
c36cac28 | 7932 | u64 start_sector; |
1ae39938 | 7933 | int async_submit = 0; |
725130ba | 7934 | u64 submit_len; |
42b5d73b NA |
7935 | u64 clone_offset = 0; |
7936 | u64 clone_len; | |
42034313 | 7937 | u64 logical; |
5f4dc8fc | 7938 | int ret; |
58efbc9f | 7939 | blk_status_t status; |
89b798ad | 7940 | struct btrfs_io_geometry geom; |
a6d3d495 | 7941 | struct btrfs_dio_data *dio_data = iter->iomap.private; |
42034313 | 7942 | struct extent_map *em = NULL; |
e65e1535 | 7943 | |
c36cac28 OS |
7944 | dip = btrfs_create_dio_private(dio_bio, inode, file_offset); |
7945 | if (!dip) { | |
7946 | if (!write) { | |
7947 | unlock_extent(&BTRFS_I(inode)->io_tree, file_offset, | |
7948 | file_offset + dio_bio->bi_iter.bi_size - 1); | |
7949 | } | |
7950 | dio_bio->bi_status = BLK_STS_RESOURCE; | |
f85781fb | 7951 | bio_endio(dio_bio); |
3e08773c | 7952 | return; |
c36cac28 | 7953 | } |
facc8a22 | 7954 | |
334c16d8 | 7955 | if (!write) { |
85879573 OS |
7956 | /* |
7957 | * Load the csums up front to reduce csum tree searches and | |
7958 | * contention when submitting bios. | |
334c16d8 JB |
7959 | * |
7960 | * If we have csums disabled this will do nothing. | |
85879573 | 7961 | */ |
6275193e | 7962 | status = btrfs_lookup_bio_sums(inode, dio_bio, dip->csums); |
85879573 OS |
7963 | if (status != BLK_STS_OK) |
7964 | goto out_err; | |
02f57c7a JB |
7965 | } |
7966 | ||
769b4f24 OS |
7967 | start_sector = dio_bio->bi_iter.bi_sector; |
7968 | submit_len = dio_bio->bi_iter.bi_size; | |
53b381b3 | 7969 | |
3c91ee69 | 7970 | do { |
42034313 MR |
7971 | logical = start_sector << 9; |
7972 | em = btrfs_get_chunk_map(fs_info, logical, submit_len); | |
7973 | if (IS_ERR(em)) { | |
7974 | status = errno_to_blk_status(PTR_ERR(em)); | |
7975 | em = NULL; | |
7976 | goto out_err_em; | |
7977 | } | |
7978 | ret = btrfs_get_io_geometry(fs_info, em, btrfs_op(dio_bio), | |
43c0d1a5 | 7979 | logical, &geom); |
769b4f24 OS |
7980 | if (ret) { |
7981 | status = errno_to_blk_status(ret); | |
42034313 | 7982 | goto out_err_em; |
769b4f24 | 7983 | } |
769b4f24 | 7984 | |
42b5d73b NA |
7985 | clone_len = min(submit_len, geom.len); |
7986 | ASSERT(clone_len <= UINT_MAX); | |
02f57c7a | 7987 | |
725130ba LB |
7988 | /* |
7989 | * This will never fail as it's passing GPF_NOFS and | |
7990 | * the allocation is backed by btrfs_bioset. | |
7991 | */ | |
769b4f24 | 7992 | bio = btrfs_bio_clone_partial(dio_bio, clone_offset, clone_len); |
725130ba LB |
7993 | bio->bi_private = dip; |
7994 | bio->bi_end_io = btrfs_end_dio_bio; | |
00d82525 | 7995 | btrfs_bio(bio)->file_offset = file_offset; |
725130ba | 7996 | |
544d24f9 NA |
7997 | if (bio_op(bio) == REQ_OP_ZONE_APPEND) { |
7998 | status = extract_ordered_extent(BTRFS_I(inode), bio, | |
7999 | file_offset); | |
8000 | if (status) { | |
8001 | bio_put(bio); | |
8002 | goto out_err; | |
8003 | } | |
8004 | } | |
8005 | ||
725130ba LB |
8006 | ASSERT(submit_len >= clone_len); |
8007 | submit_len -= clone_len; | |
e65e1535 | 8008 | |
725130ba LB |
8009 | /* |
8010 | * Increase the count before we submit the bio so we know | |
8011 | * the end IO handler won't happen before we increase the | |
8012 | * count. Otherwise, the dip might get freed before we're | |
8013 | * done setting it up. | |
769b4f24 OS |
8014 | * |
8015 | * We transfer the initial reference to the last bio, so we | |
8016 | * don't need to increment the reference count for the last one. | |
725130ba | 8017 | */ |
769b4f24 OS |
8018 | if (submit_len > 0) { |
8019 | refcount_inc(&dip->refs); | |
8020 | /* | |
8021 | * If we are submitting more than one bio, submit them | |
8022 | * all asynchronously. The exception is RAID 5 or 6, as | |
8023 | * asynchronous checksums make it difficult to collect | |
8024 | * full stripe writes. | |
8025 | */ | |
8026 | if (!raid56) | |
8027 | async_submit = 1; | |
8028 | } | |
e65e1535 | 8029 | |
d0ee3934 | 8030 | status = btrfs_submit_dio_bio(bio, inode, file_offset, |
58efbc9f OS |
8031 | async_submit); |
8032 | if (status) { | |
725130ba | 8033 | bio_put(bio); |
769b4f24 OS |
8034 | if (submit_len > 0) |
8035 | refcount_dec(&dip->refs); | |
42034313 | 8036 | goto out_err_em; |
725130ba | 8037 | } |
e65e1535 | 8038 | |
f85781fb | 8039 | dio_data->submitted += clone_len; |
725130ba LB |
8040 | clone_offset += clone_len; |
8041 | start_sector += clone_len >> 9; | |
8042 | file_offset += clone_len; | |
42034313 MR |
8043 | |
8044 | free_extent_map(em); | |
3c91ee69 | 8045 | } while (submit_len > 0); |
3e08773c | 8046 | return; |
e65e1535 | 8047 | |
42034313 MR |
8048 | out_err_em: |
8049 | free_extent_map(em); | |
e65e1535 | 8050 | out_err: |
769b4f24 OS |
8051 | dip->dio_bio->bi_status = status; |
8052 | btrfs_dio_private_put(dip); | |
4b46fce2 JB |
8053 | } |
8054 | ||
4e4cabec | 8055 | const struct iomap_ops btrfs_dio_iomap_ops = { |
f85781fb GR |
8056 | .iomap_begin = btrfs_dio_iomap_begin, |
8057 | .iomap_end = btrfs_dio_iomap_end, | |
8058 | }; | |
8059 | ||
4e4cabec | 8060 | const struct iomap_dio_ops btrfs_dio_ops = { |
f85781fb GR |
8061 | .submit_io = btrfs_submit_direct, |
8062 | }; | |
8063 | ||
1506fcc8 | 8064 | static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, |
bab16e21 | 8065 | u64 start, u64 len) |
1506fcc8 | 8066 | { |
05dadc09 TI |
8067 | int ret; |
8068 | ||
45dd052e | 8069 | ret = fiemap_prep(inode, fieinfo, start, &len, 0); |
05dadc09 TI |
8070 | if (ret) |
8071 | return ret; | |
8072 | ||
facee0a0 | 8073 | return extent_fiemap(BTRFS_I(inode), fieinfo, start, len); |
1506fcc8 YS |
8074 | } |
8075 | ||
a52d9a80 | 8076 | int btrfs_readpage(struct file *file, struct page *page) |
9ebefb18 | 8077 | { |
0f208812 NB |
8078 | struct btrfs_inode *inode = BTRFS_I(page->mapping->host); |
8079 | u64 start = page_offset(page); | |
8080 | u64 end = start + PAGE_SIZE - 1; | |
390ed29b | 8081 | struct btrfs_bio_ctrl bio_ctrl = { 0 }; |
c1be9c1a NB |
8082 | int ret; |
8083 | ||
0f208812 NB |
8084 | btrfs_lock_and_flush_ordered_range(inode, start, end, NULL); |
8085 | ||
390ed29b | 8086 | ret = btrfs_do_readpage(page, NULL, &bio_ctrl, 0, NULL); |
bbf0ea7e FM |
8087 | if (bio_ctrl.bio) { |
8088 | int ret2; | |
8089 | ||
8090 | ret2 = submit_one_bio(bio_ctrl.bio, 0, bio_ctrl.bio_flags); | |
8091 | if (ret == 0) | |
8092 | ret = ret2; | |
8093 | } | |
c1be9c1a | 8094 | return ret; |
9ebefb18 | 8095 | } |
1832a6d5 | 8096 | |
a52d9a80 | 8097 | static int btrfs_writepage(struct page *page, struct writeback_control *wbc) |
39279cc3 | 8098 | { |
be7bd730 JB |
8099 | struct inode *inode = page->mapping->host; |
8100 | int ret; | |
b888db2b CM |
8101 | |
8102 | if (current->flags & PF_MEMALLOC) { | |
8103 | redirty_page_for_writepage(wbc, page); | |
8104 | unlock_page(page); | |
8105 | return 0; | |
8106 | } | |
be7bd730 JB |
8107 | |
8108 | /* | |
8109 | * If we are under memory pressure we will call this directly from the | |
8110 | * VM, we need to make sure we have the inode referenced for the ordered | |
8111 | * extent. If not just return like we didn't do anything. | |
8112 | */ | |
8113 | if (!igrab(inode)) { | |
8114 | redirty_page_for_writepage(wbc, page); | |
8115 | return AOP_WRITEPAGE_ACTIVATE; | |
8116 | } | |
0a9b0e53 | 8117 | ret = extent_write_full_page(page, wbc); |
be7bd730 JB |
8118 | btrfs_add_delayed_iput(inode); |
8119 | return ret; | |
9ebefb18 CM |
8120 | } |
8121 | ||
48a3b636 ES |
8122 | static int btrfs_writepages(struct address_space *mapping, |
8123 | struct writeback_control *wbc) | |
b293f02e | 8124 | { |
8ae225a8 | 8125 | return extent_writepages(mapping, wbc); |
b293f02e CM |
8126 | } |
8127 | ||
ba206a02 | 8128 | static void btrfs_readahead(struct readahead_control *rac) |
3ab2fb5a | 8129 | { |
ba206a02 | 8130 | extent_readahead(rac); |
3ab2fb5a | 8131 | } |
2a3ff0ad | 8132 | |
7c11d0ae | 8133 | /* |
895586eb MWO |
8134 | * For releasepage() and invalidate_folio() we have a race window where |
8135 | * folio_end_writeback() is called but the subpage spinlock is not yet released. | |
7c11d0ae QW |
8136 | * If we continue to release/invalidate the page, we could cause use-after-free |
8137 | * for subpage spinlock. So this function is to spin and wait for subpage | |
8138 | * spinlock. | |
8139 | */ | |
8140 | static void wait_subpage_spinlock(struct page *page) | |
8141 | { | |
8142 | struct btrfs_fs_info *fs_info = btrfs_sb(page->mapping->host->i_sb); | |
8143 | struct btrfs_subpage *subpage; | |
8144 | ||
8145 | if (fs_info->sectorsize == PAGE_SIZE) | |
8146 | return; | |
8147 | ||
8148 | ASSERT(PagePrivate(page) && page->private); | |
8149 | subpage = (struct btrfs_subpage *)page->private; | |
8150 | ||
8151 | /* | |
8152 | * This may look insane as we just acquire the spinlock and release it, | |
8153 | * without doing anything. But we just want to make sure no one is | |
8154 | * still holding the subpage spinlock. | |
8155 | * And since the page is not dirty nor writeback, and we have page | |
8156 | * locked, the only possible way to hold a spinlock is from the endio | |
8157 | * function to clear page writeback. | |
8158 | * | |
8159 | * Here we just acquire the spinlock so that all existing callers | |
8160 | * should exit and we're safe to release/invalidate the page. | |
8161 | */ | |
8162 | spin_lock_irq(&subpage->lock); | |
8163 | spin_unlock_irq(&subpage->lock); | |
8164 | } | |
8165 | ||
e6dcd2dc | 8166 | static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags) |
9ebefb18 | 8167 | { |
477a30ba | 8168 | int ret = try_release_extent_mapping(page, gfp_flags); |
7c11d0ae QW |
8169 | |
8170 | if (ret == 1) { | |
8171 | wait_subpage_spinlock(page); | |
32443de3 | 8172 | clear_page_extent_mapped(page); |
7c11d0ae | 8173 | } |
a52d9a80 | 8174 | return ret; |
39279cc3 CM |
8175 | } |
8176 | ||
e6dcd2dc CM |
8177 | static int btrfs_releasepage(struct page *page, gfp_t gfp_flags) |
8178 | { | |
98509cfc CM |
8179 | if (PageWriteback(page) || PageDirty(page)) |
8180 | return 0; | |
3ba7ab22 | 8181 | return __btrfs_releasepage(page, gfp_flags); |
e6dcd2dc CM |
8182 | } |
8183 | ||
f8e66081 RG |
8184 | #ifdef CONFIG_MIGRATION |
8185 | static int btrfs_migratepage(struct address_space *mapping, | |
8186 | struct page *newpage, struct page *page, | |
8187 | enum migrate_mode mode) | |
8188 | { | |
8189 | int ret; | |
8190 | ||
8191 | ret = migrate_page_move_mapping(mapping, newpage, page, 0); | |
8192 | if (ret != MIGRATEPAGE_SUCCESS) | |
8193 | return ret; | |
8194 | ||
d1b89bc0 GJ |
8195 | if (page_has_private(page)) |
8196 | attach_page_private(newpage, detach_page_private(page)); | |
f8e66081 | 8197 | |
f57ad937 QW |
8198 | if (PageOrdered(page)) { |
8199 | ClearPageOrdered(page); | |
8200 | SetPageOrdered(newpage); | |
f8e66081 RG |
8201 | } |
8202 | ||
8203 | if (mode != MIGRATE_SYNC_NO_COPY) | |
8204 | migrate_page_copy(newpage, page); | |
8205 | else | |
8206 | migrate_page_states(newpage, page); | |
8207 | return MIGRATEPAGE_SUCCESS; | |
8208 | } | |
8209 | #endif | |
8210 | ||
895586eb MWO |
8211 | static void btrfs_invalidate_folio(struct folio *folio, size_t offset, |
8212 | size_t length) | |
39279cc3 | 8213 | { |
895586eb | 8214 | struct btrfs_inode *inode = BTRFS_I(folio->mapping->host); |
b945a463 | 8215 | struct btrfs_fs_info *fs_info = inode->root->fs_info; |
53ac7ead | 8216 | struct extent_io_tree *tree = &inode->io_tree; |
2ac55d41 | 8217 | struct extent_state *cached_state = NULL; |
895586eb MWO |
8218 | u64 page_start = folio_pos(folio); |
8219 | u64 page_end = page_start + folio_size(folio) - 1; | |
3b835840 | 8220 | u64 cur; |
53ac7ead | 8221 | int inode_evicting = inode->vfs_inode.i_state & I_FREEING; |
39279cc3 | 8222 | |
8b62b72b | 8223 | /* |
895586eb MWO |
8224 | * We have folio locked so no new ordered extent can be created on this |
8225 | * page, nor bio can be submitted for this folio. | |
8b62b72b | 8226 | * |
895586eb MWO |
8227 | * But already submitted bio can still be finished on this folio. |
8228 | * Furthermore, endio function won't skip folio which has Ordered | |
f57ad937 | 8229 | * (Private2) already cleared, so it's possible for endio and |
895586eb MWO |
8230 | * invalidate_folio to do the same ordered extent accounting twice |
8231 | * on one folio. | |
266a2586 QW |
8232 | * |
8233 | * So here we wait for any submitted bios to finish, so that we won't | |
895586eb | 8234 | * do double ordered extent accounting on the same folio. |
8b62b72b | 8235 | */ |
895586eb MWO |
8236 | folio_wait_writeback(folio); |
8237 | wait_subpage_spinlock(&folio->page); | |
8b62b72b | 8238 | |
bcd77455 QW |
8239 | /* |
8240 | * For subpage case, we have call sites like | |
8241 | * btrfs_punch_hole_lock_range() which passes range not aligned to | |
8242 | * sectorsize. | |
895586eb MWO |
8243 | * If the range doesn't cover the full folio, we don't need to and |
8244 | * shouldn't clear page extent mapped, as folio->private can still | |
bcd77455 QW |
8245 | * record subpage dirty bits for other part of the range. |
8246 | * | |
895586eb MWO |
8247 | * For cases that invalidate the full folio even the range doesn't |
8248 | * cover the full folio, like invalidating the last folio, we're | |
bcd77455 QW |
8249 | * still safe to wait for ordered extent to finish. |
8250 | */ | |
5a60542c | 8251 | if (!(offset == 0 && length == folio_size(folio))) { |
895586eb | 8252 | btrfs_releasepage(&folio->page, GFP_NOFS); |
e6dcd2dc CM |
8253 | return; |
8254 | } | |
131e404a FDBM |
8255 | |
8256 | if (!inode_evicting) | |
ff13db41 | 8257 | lock_extent_bits(tree, page_start, page_end, &cached_state); |
951c80f8 | 8258 | |
3b835840 QW |
8259 | cur = page_start; |
8260 | while (cur < page_end) { | |
8261 | struct btrfs_ordered_extent *ordered; | |
8262 | bool delete_states; | |
8263 | u64 range_end; | |
b945a463 | 8264 | u32 range_len; |
3b835840 QW |
8265 | |
8266 | ordered = btrfs_lookup_first_ordered_range(inode, cur, | |
8267 | page_end + 1 - cur); | |
8268 | if (!ordered) { | |
8269 | range_end = page_end; | |
8270 | /* | |
8271 | * No ordered extent covering this range, we are safe | |
8272 | * to delete all extent states in the range. | |
8273 | */ | |
8274 | delete_states = true; | |
8275 | goto next; | |
8276 | } | |
8277 | if (ordered->file_offset > cur) { | |
8278 | /* | |
8279 | * There is a range between [cur, oe->file_offset) not | |
8280 | * covered by any ordered extent. | |
8281 | * We are safe to delete all extent states, and handle | |
8282 | * the ordered extent in the next iteration. | |
8283 | */ | |
8284 | range_end = ordered->file_offset - 1; | |
8285 | delete_states = true; | |
8286 | goto next; | |
8287 | } | |
8288 | ||
8289 | range_end = min(ordered->file_offset + ordered->num_bytes - 1, | |
8290 | page_end); | |
b945a463 QW |
8291 | ASSERT(range_end + 1 - cur < U32_MAX); |
8292 | range_len = range_end + 1 - cur; | |
895586eb | 8293 | if (!btrfs_page_test_ordered(fs_info, &folio->page, cur, range_len)) { |
3b835840 | 8294 | /* |
f57ad937 QW |
8295 | * If Ordered (Private2) is cleared, it means endio has |
8296 | * already been executed for the range. | |
3b835840 QW |
8297 | * We can't delete the extent states as |
8298 | * btrfs_finish_ordered_io() may still use some of them. | |
8299 | */ | |
8300 | delete_states = false; | |
8301 | goto next; | |
8302 | } | |
895586eb | 8303 | btrfs_page_clear_ordered(fs_info, &folio->page, cur, range_len); |
3b835840 | 8304 | |
eb84ae03 | 8305 | /* |
2766ff61 FM |
8306 | * IO on this page will never be started, so we need to account |
8307 | * for any ordered extents now. Don't clear EXTENT_DELALLOC_NEW | |
8308 | * here, must leave that up for the ordered extent completion. | |
3b835840 QW |
8309 | * |
8310 | * This will also unlock the range for incoming | |
8311 | * btrfs_finish_ordered_io(). | |
eb84ae03 | 8312 | */ |
131e404a | 8313 | if (!inode_evicting) |
3b835840 | 8314 | clear_extent_bit(tree, cur, range_end, |
2766ff61 | 8315 | EXTENT_DELALLOC | |
131e404a | 8316 | EXTENT_LOCKED | EXTENT_DO_ACCOUNTING | |
ae0f1625 | 8317 | EXTENT_DEFRAG, 1, 0, &cached_state); |
3b835840 QW |
8318 | |
8319 | spin_lock_irq(&inode->ordered_tree.lock); | |
8320 | set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags); | |
8321 | ordered->truncated_len = min(ordered->truncated_len, | |
8322 | cur - ordered->file_offset); | |
8323 | spin_unlock_irq(&inode->ordered_tree.lock); | |
8324 | ||
8325 | if (btrfs_dec_test_ordered_pending(inode, &ordered, | |
f41b6ba9 | 8326 | cur, range_end + 1 - cur)) { |
3b835840 QW |
8327 | btrfs_finish_ordered_io(ordered); |
8328 | /* | |
8329 | * The ordered extent has finished, now we're again | |
8330 | * safe to delete all extent states of the range. | |
8331 | */ | |
8332 | delete_states = true; | |
8333 | } else { | |
8334 | /* | |
8335 | * btrfs_finish_ordered_io() will get executed by endio | |
8336 | * of other pages, thus we can't delete extent states | |
8337 | * anymore | |
8338 | */ | |
8339 | delete_states = false; | |
8340 | } | |
8341 | next: | |
8342 | if (ordered) | |
8343 | btrfs_put_ordered_extent(ordered); | |
8b62b72b | 8344 | /* |
3b835840 QW |
8345 | * Qgroup reserved space handler |
8346 | * Sector(s) here will be either: | |
266a2586 | 8347 | * |
3b835840 QW |
8348 | * 1) Already written to disk or bio already finished |
8349 | * Then its QGROUP_RESERVED bit in io_tree is already cleared. | |
8350 | * Qgroup will be handled by its qgroup_record then. | |
8351 | * btrfs_qgroup_free_data() call will do nothing here. | |
8352 | * | |
8353 | * 2) Not written to disk yet | |
8354 | * Then btrfs_qgroup_free_data() call will clear the | |
8355 | * QGROUP_RESERVED bit of its io_tree, and free the qgroup | |
8356 | * reserved data space. | |
8357 | * Since the IO will never happen for this page. | |
8b62b72b | 8358 | */ |
3b835840 | 8359 | btrfs_qgroup_free_data(inode, NULL, cur, range_end + 1 - cur); |
131e404a | 8360 | if (!inode_evicting) { |
3b835840 QW |
8361 | clear_extent_bit(tree, cur, range_end, EXTENT_LOCKED | |
8362 | EXTENT_DELALLOC | EXTENT_UPTODATE | | |
8363 | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 1, | |
8364 | delete_states, &cached_state); | |
131e404a | 8365 | } |
3b835840 | 8366 | cur = range_end + 1; |
131e404a | 8367 | } |
b9d0b389 | 8368 | /* |
3b835840 | 8369 | * We have iterated through all ordered extents of the page, the page |
f57ad937 QW |
8370 | * should not have Ordered (Private2) anymore, or the above iteration |
8371 | * did something wrong. | |
b9d0b389 | 8372 | */ |
895586eb MWO |
8373 | ASSERT(!folio_test_ordered(folio)); |
8374 | btrfs_page_clear_checked(fs_info, &folio->page, folio_pos(folio), folio_size(folio)); | |
3b835840 | 8375 | if (!inode_evicting) |
895586eb MWO |
8376 | __btrfs_releasepage(&folio->page, GFP_NOFS); |
8377 | clear_page_extent_mapped(&folio->page); | |
39279cc3 CM |
8378 | } |
8379 | ||
9ebefb18 CM |
8380 | /* |
8381 | * btrfs_page_mkwrite() is not allowed to change the file size as it gets | |
8382 | * called from a page fault handler when a page is first dirtied. Hence we must | |
8383 | * be careful to check for EOF conditions here. We set the page up correctly | |
8384 | * for a written page which means we get ENOSPC checking when writing into | |
8385 | * holes and correct delalloc and unwritten extent mapping on filesystems that | |
8386 | * support these features. | |
8387 | * | |
8388 | * We are not allowed to take the i_mutex here so we have to play games to | |
8389 | * protect against truncate races as the page could now be beyond EOF. Because | |
d1342aad OS |
8390 | * truncate_setsize() writes the inode size before removing pages, once we have |
8391 | * the page lock we can determine safely if the page is beyond EOF. If it is not | |
9ebefb18 CM |
8392 | * beyond EOF, then the page is guaranteed safe against truncation until we |
8393 | * unlock the page. | |
8394 | */ | |
a528a241 | 8395 | vm_fault_t btrfs_page_mkwrite(struct vm_fault *vmf) |
9ebefb18 | 8396 | { |
c2ec175c | 8397 | struct page *page = vmf->page; |
11bac800 | 8398 | struct inode *inode = file_inode(vmf->vma->vm_file); |
0b246afa | 8399 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
e6dcd2dc CM |
8400 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
8401 | struct btrfs_ordered_extent *ordered; | |
2ac55d41 | 8402 | struct extent_state *cached_state = NULL; |
364ecf36 | 8403 | struct extent_changeset *data_reserved = NULL; |
e6dcd2dc | 8404 | unsigned long zero_start; |
9ebefb18 | 8405 | loff_t size; |
a528a241 SJ |
8406 | vm_fault_t ret; |
8407 | int ret2; | |
9998eb70 | 8408 | int reserved = 0; |
d0b7da88 | 8409 | u64 reserved_space; |
a52d9a80 | 8410 | u64 page_start; |
e6dcd2dc | 8411 | u64 page_end; |
d0b7da88 CR |
8412 | u64 end; |
8413 | ||
09cbfeaf | 8414 | reserved_space = PAGE_SIZE; |
9ebefb18 | 8415 | |
b2b5ef5c | 8416 | sb_start_pagefault(inode->i_sb); |
df480633 | 8417 | page_start = page_offset(page); |
09cbfeaf | 8418 | page_end = page_start + PAGE_SIZE - 1; |
d0b7da88 | 8419 | end = page_end; |
df480633 | 8420 | |
d0b7da88 CR |
8421 | /* |
8422 | * Reserving delalloc space after obtaining the page lock can lead to | |
8423 | * deadlock. For example, if a dirty page is locked by this function | |
8424 | * and the call to btrfs_delalloc_reserve_space() ends up triggering | |
8425 | * dirty page write out, then the btrfs_writepage() function could | |
8426 | * end up waiting indefinitely to get a lock on the page currently | |
8427 | * being processed by btrfs_page_mkwrite() function. | |
8428 | */ | |
e5b7231e NB |
8429 | ret2 = btrfs_delalloc_reserve_space(BTRFS_I(inode), &data_reserved, |
8430 | page_start, reserved_space); | |
a528a241 SJ |
8431 | if (!ret2) { |
8432 | ret2 = file_update_time(vmf->vma->vm_file); | |
9998eb70 CM |
8433 | reserved = 1; |
8434 | } | |
a528a241 SJ |
8435 | if (ret2) { |
8436 | ret = vmf_error(ret2); | |
9998eb70 CM |
8437 | if (reserved) |
8438 | goto out; | |
8439 | goto out_noreserve; | |
56a76f82 | 8440 | } |
1832a6d5 | 8441 | |
56a76f82 | 8442 | ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */ |
e6dcd2dc | 8443 | again: |
8318ba79 | 8444 | down_read(&BTRFS_I(inode)->i_mmap_lock); |
9ebefb18 | 8445 | lock_page(page); |
9ebefb18 | 8446 | size = i_size_read(inode); |
a52d9a80 | 8447 | |
9ebefb18 | 8448 | if ((page->mapping != inode->i_mapping) || |
e6dcd2dc | 8449 | (page_start >= size)) { |
9ebefb18 CM |
8450 | /* page got truncated out from underneath us */ |
8451 | goto out_unlock; | |
8452 | } | |
e6dcd2dc CM |
8453 | wait_on_page_writeback(page); |
8454 | ||
ff13db41 | 8455 | lock_extent_bits(io_tree, page_start, page_end, &cached_state); |
32443de3 QW |
8456 | ret2 = set_page_extent_mapped(page); |
8457 | if (ret2 < 0) { | |
8458 | ret = vmf_error(ret2); | |
8459 | unlock_extent_cached(io_tree, page_start, page_end, &cached_state); | |
8460 | goto out_unlock; | |
8461 | } | |
e6dcd2dc | 8462 | |
eb84ae03 CM |
8463 | /* |
8464 | * we can't set the delalloc bits if there are pending ordered | |
8465 | * extents. Drop our locks and wait for them to finish | |
8466 | */ | |
a776c6fa NB |
8467 | ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), page_start, |
8468 | PAGE_SIZE); | |
e6dcd2dc | 8469 | if (ordered) { |
2ac55d41 | 8470 | unlock_extent_cached(io_tree, page_start, page_end, |
e43bbe5e | 8471 | &cached_state); |
e6dcd2dc | 8472 | unlock_page(page); |
8318ba79 | 8473 | up_read(&BTRFS_I(inode)->i_mmap_lock); |
c0a43603 | 8474 | btrfs_start_ordered_extent(ordered, 1); |
e6dcd2dc CM |
8475 | btrfs_put_ordered_extent(ordered); |
8476 | goto again; | |
8477 | } | |
8478 | ||
09cbfeaf | 8479 | if (page->index == ((size - 1) >> PAGE_SHIFT)) { |
da17066c | 8480 | reserved_space = round_up(size - page_start, |
0b246afa | 8481 | fs_info->sectorsize); |
09cbfeaf | 8482 | if (reserved_space < PAGE_SIZE) { |
d0b7da88 | 8483 | end = page_start + reserved_space - 1; |
86d52921 NB |
8484 | btrfs_delalloc_release_space(BTRFS_I(inode), |
8485 | data_reserved, page_start, | |
8486 | PAGE_SIZE - reserved_space, true); | |
d0b7da88 CR |
8487 | } |
8488 | } | |
8489 | ||
fbf19087 | 8490 | /* |
5416034f LB |
8491 | * page_mkwrite gets called when the page is firstly dirtied after it's |
8492 | * faulted in, but write(2) could also dirty a page and set delalloc | |
8493 | * bits, thus in this case for space account reason, we still need to | |
8494 | * clear any delalloc bits within this page range since we have to | |
8495 | * reserve data&meta space before lock_page() (see above comments). | |
fbf19087 | 8496 | */ |
d0b7da88 | 8497 | clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, end, |
e182163d OS |
8498 | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING | |
8499 | EXTENT_DEFRAG, 0, 0, &cached_state); | |
fbf19087 | 8500 | |
c2566f22 | 8501 | ret2 = btrfs_set_extent_delalloc(BTRFS_I(inode), page_start, end, 0, |
330a5827 | 8502 | &cached_state); |
a528a241 | 8503 | if (ret2) { |
2ac55d41 | 8504 | unlock_extent_cached(io_tree, page_start, page_end, |
e43bbe5e | 8505 | &cached_state); |
9ed74f2d JB |
8506 | ret = VM_FAULT_SIGBUS; |
8507 | goto out_unlock; | |
8508 | } | |
9ebefb18 CM |
8509 | |
8510 | /* page is wholly or partially inside EOF */ | |
09cbfeaf | 8511 | if (page_start + PAGE_SIZE > size) |
7073017a | 8512 | zero_start = offset_in_page(size); |
9ebefb18 | 8513 | else |
09cbfeaf | 8514 | zero_start = PAGE_SIZE; |
9ebefb18 | 8515 | |
09cbfeaf | 8516 | if (zero_start != PAGE_SIZE) { |
d048b9c2 | 8517 | memzero_page(page, zero_start, PAGE_SIZE - zero_start); |
e6dcd2dc | 8518 | flush_dcache_page(page); |
e6dcd2dc | 8519 | } |
e4f94347 | 8520 | btrfs_page_clear_checked(fs_info, page, page_start, PAGE_SIZE); |
2d8ec40e QW |
8521 | btrfs_page_set_dirty(fs_info, page, page_start, end + 1 - page_start); |
8522 | btrfs_page_set_uptodate(fs_info, page, page_start, end + 1 - page_start); | |
5a3f23d5 | 8523 | |
bc0939fc | 8524 | btrfs_set_inode_last_sub_trans(BTRFS_I(inode)); |
257c62e1 | 8525 | |
e43bbe5e | 8526 | unlock_extent_cached(io_tree, page_start, page_end, &cached_state); |
8318ba79 | 8527 | up_read(&BTRFS_I(inode)->i_mmap_lock); |
9ebefb18 | 8528 | |
76de60ed YY |
8529 | btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE); |
8530 | sb_end_pagefault(inode->i_sb); | |
8531 | extent_changeset_free(data_reserved); | |
8532 | return VM_FAULT_LOCKED; | |
717beb96 CM |
8533 | |
8534 | out_unlock: | |
9ebefb18 | 8535 | unlock_page(page); |
8318ba79 | 8536 | up_read(&BTRFS_I(inode)->i_mmap_lock); |
1832a6d5 | 8537 | out: |
8702ba93 | 8538 | btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE); |
86d52921 | 8539 | btrfs_delalloc_release_space(BTRFS_I(inode), data_reserved, page_start, |
43b18595 | 8540 | reserved_space, (ret != 0)); |
9998eb70 | 8541 | out_noreserve: |
b2b5ef5c | 8542 | sb_end_pagefault(inode->i_sb); |
364ecf36 | 8543 | extent_changeset_free(data_reserved); |
9ebefb18 CM |
8544 | return ret; |
8545 | } | |
8546 | ||
213e8c55 | 8547 | static int btrfs_truncate(struct inode *inode, bool skip_writeback) |
39279cc3 | 8548 | { |
d9ac19c3 | 8549 | struct btrfs_truncate_control control = { |
71d18b53 | 8550 | .inode = BTRFS_I(inode), |
487e81d2 | 8551 | .ino = btrfs_ino(BTRFS_I(inode)), |
d9ac19c3 | 8552 | .min_type = BTRFS_EXTENT_DATA_KEY, |
655807b8 | 8553 | .clear_extent_range = true, |
d9ac19c3 | 8554 | }; |
0b246afa | 8555 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
39279cc3 | 8556 | struct btrfs_root *root = BTRFS_I(inode)->root; |
fcb80c2a | 8557 | struct btrfs_block_rsv *rsv; |
ad7e1a74 | 8558 | int ret; |
39279cc3 | 8559 | struct btrfs_trans_handle *trans; |
0b246afa | 8560 | u64 mask = fs_info->sectorsize - 1; |
2bd36e7b | 8561 | u64 min_size = btrfs_calc_metadata_size(fs_info, 1); |
39279cc3 | 8562 | |
213e8c55 FM |
8563 | if (!skip_writeback) { |
8564 | ret = btrfs_wait_ordered_range(inode, inode->i_size & (~mask), | |
8565 | (u64)-1); | |
8566 | if (ret) | |
8567 | return ret; | |
8568 | } | |
39279cc3 | 8569 | |
fcb80c2a | 8570 | /* |
f7e9e8fc OS |
8571 | * Yes ladies and gentlemen, this is indeed ugly. We have a couple of |
8572 | * things going on here: | |
fcb80c2a | 8573 | * |
f7e9e8fc | 8574 | * 1) We need to reserve space to update our inode. |
fcb80c2a | 8575 | * |
f7e9e8fc | 8576 | * 2) We need to have something to cache all the space that is going to |
fcb80c2a JB |
8577 | * be free'd up by the truncate operation, but also have some slack |
8578 | * space reserved in case it uses space during the truncate (thank you | |
8579 | * very much snapshotting). | |
8580 | * | |
f7e9e8fc | 8581 | * And we need these to be separate. The fact is we can use a lot of |
fcb80c2a | 8582 | * space doing the truncate, and we have no earthly idea how much space |
01327610 | 8583 | * we will use, so we need the truncate reservation to be separate so it |
f7e9e8fc OS |
8584 | * doesn't end up using space reserved for updating the inode. We also |
8585 | * need to be able to stop the transaction and start a new one, which | |
8586 | * means we need to be able to update the inode several times, and we | |
8587 | * have no idea of knowing how many times that will be, so we can't just | |
8588 | * reserve 1 item for the entirety of the operation, so that has to be | |
8589 | * done separately as well. | |
fcb80c2a JB |
8590 | * |
8591 | * So that leaves us with | |
8592 | * | |
f7e9e8fc | 8593 | * 1) rsv - for the truncate reservation, which we will steal from the |
fcb80c2a | 8594 | * transaction reservation. |
f7e9e8fc | 8595 | * 2) fs_info->trans_block_rsv - this will have 1 items worth left for |
fcb80c2a JB |
8596 | * updating the inode. |
8597 | */ | |
2ff7e61e | 8598 | rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP); |
fcb80c2a JB |
8599 | if (!rsv) |
8600 | return -ENOMEM; | |
4a338542 | 8601 | rsv->size = min_size; |
ca7e70f5 | 8602 | rsv->failfast = 1; |
f0cd846e | 8603 | |
907cbceb | 8604 | /* |
07127184 | 8605 | * 1 for the truncate slack space |
907cbceb JB |
8606 | * 1 for updating the inode. |
8607 | */ | |
f3fe820c | 8608 | trans = btrfs_start_transaction(root, 2); |
fcb80c2a | 8609 | if (IS_ERR(trans)) { |
ad7e1a74 | 8610 | ret = PTR_ERR(trans); |
fcb80c2a JB |
8611 | goto out; |
8612 | } | |
f0cd846e | 8613 | |
907cbceb | 8614 | /* Migrate the slack space for the truncate to our reserve */ |
0b246afa | 8615 | ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv, rsv, |
3a584174 | 8616 | min_size, false); |
fcb80c2a | 8617 | BUG_ON(ret); |
f0cd846e | 8618 | |
ca7e70f5 | 8619 | trans->block_rsv = rsv; |
907cbceb | 8620 | |
8082510e | 8621 | while (1) { |
9a4a1429 JB |
8622 | struct extent_state *cached_state = NULL; |
8623 | const u64 new_size = inode->i_size; | |
8624 | const u64 lock_start = ALIGN_DOWN(new_size, fs_info->sectorsize); | |
8625 | ||
d9ac19c3 | 8626 | control.new_size = new_size; |
9a4a1429 JB |
8627 | lock_extent_bits(&BTRFS_I(inode)->io_tree, lock_start, (u64)-1, |
8628 | &cached_state); | |
8629 | /* | |
8630 | * We want to drop from the next block forward in case this new | |
8631 | * size is not block aligned since we will be keeping the last | |
8632 | * block of the extent just the way it is. | |
8633 | */ | |
8634 | btrfs_drop_extent_cache(BTRFS_I(inode), | |
8635 | ALIGN(new_size, fs_info->sectorsize), | |
8636 | (u64)-1, 0); | |
8637 | ||
71d18b53 | 8638 | ret = btrfs_truncate_inode_items(trans, root, &control); |
c2ddb612 | 8639 | |
462b728e | 8640 | inode_sub_bytes(inode, control.sub_bytes); |
c2ddb612 JB |
8641 | btrfs_inode_safe_disk_i_size_write(BTRFS_I(inode), control.last_size); |
8642 | ||
9a4a1429 JB |
8643 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, lock_start, |
8644 | (u64)-1, &cached_state); | |
8645 | ||
ddfae63c | 8646 | trans->block_rsv = &fs_info->trans_block_rsv; |
ad7e1a74 | 8647 | if (ret != -ENOSPC && ret != -EAGAIN) |
8082510e | 8648 | break; |
39279cc3 | 8649 | |
9a56fcd1 | 8650 | ret = btrfs_update_inode(trans, root, BTRFS_I(inode)); |
ad7e1a74 | 8651 | if (ret) |
3893e33b | 8652 | break; |
ca7e70f5 | 8653 | |
3a45bb20 | 8654 | btrfs_end_transaction(trans); |
2ff7e61e | 8655 | btrfs_btree_balance_dirty(fs_info); |
ca7e70f5 JB |
8656 | |
8657 | trans = btrfs_start_transaction(root, 2); | |
8658 | if (IS_ERR(trans)) { | |
ad7e1a74 | 8659 | ret = PTR_ERR(trans); |
ca7e70f5 JB |
8660 | trans = NULL; |
8661 | break; | |
8662 | } | |
8663 | ||
63f018be | 8664 | btrfs_block_rsv_release(fs_info, rsv, -1, NULL); |
0b246afa | 8665 | ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv, |
3a584174 | 8666 | rsv, min_size, false); |
ca7e70f5 JB |
8667 | BUG_ON(ret); /* shouldn't happen */ |
8668 | trans->block_rsv = rsv; | |
8082510e YZ |
8669 | } |
8670 | ||
ddfae63c JB |
8671 | /* |
8672 | * We can't call btrfs_truncate_block inside a trans handle as we could | |
54f03ab1 JB |
8673 | * deadlock with freeze, if we got BTRFS_NEED_TRUNCATE_BLOCK then we |
8674 | * know we've truncated everything except the last little bit, and can | |
8675 | * do btrfs_truncate_block and then update the disk_i_size. | |
ddfae63c | 8676 | */ |
54f03ab1 | 8677 | if (ret == BTRFS_NEED_TRUNCATE_BLOCK) { |
ddfae63c JB |
8678 | btrfs_end_transaction(trans); |
8679 | btrfs_btree_balance_dirty(fs_info); | |
8680 | ||
217f42eb | 8681 | ret = btrfs_truncate_block(BTRFS_I(inode), inode->i_size, 0, 0); |
ddfae63c JB |
8682 | if (ret) |
8683 | goto out; | |
8684 | trans = btrfs_start_transaction(root, 1); | |
8685 | if (IS_ERR(trans)) { | |
8686 | ret = PTR_ERR(trans); | |
8687 | goto out; | |
8688 | } | |
76aea537 | 8689 | btrfs_inode_safe_disk_i_size_write(BTRFS_I(inode), 0); |
ddfae63c JB |
8690 | } |
8691 | ||
917c16b2 | 8692 | if (trans) { |
ad7e1a74 OS |
8693 | int ret2; |
8694 | ||
0b246afa | 8695 | trans->block_rsv = &fs_info->trans_block_rsv; |
9a56fcd1 | 8696 | ret2 = btrfs_update_inode(trans, root, BTRFS_I(inode)); |
ad7e1a74 OS |
8697 | if (ret2 && !ret) |
8698 | ret = ret2; | |
7b128766 | 8699 | |
ad7e1a74 OS |
8700 | ret2 = btrfs_end_transaction(trans); |
8701 | if (ret2 && !ret) | |
8702 | ret = ret2; | |
2ff7e61e | 8703 | btrfs_btree_balance_dirty(fs_info); |
917c16b2 | 8704 | } |
fcb80c2a | 8705 | out: |
2ff7e61e | 8706 | btrfs_free_block_rsv(fs_info, rsv); |
0d7d3165 FM |
8707 | /* |
8708 | * So if we truncate and then write and fsync we normally would just | |
8709 | * write the extents that changed, which is a problem if we need to | |
8710 | * first truncate that entire inode. So set this flag so we write out | |
8711 | * all of the extents in the inode to the sync log so we're completely | |
8712 | * safe. | |
8713 | * | |
8714 | * If no extents were dropped or trimmed we don't need to force the next | |
8715 | * fsync to truncate all the inode's items from the log and re-log them | |
8716 | * all. This means the truncate operation did not change the file size, | |
8717 | * or changed it to a smaller size but there was only an implicit hole | |
8718 | * between the old i_size and the new i_size, and there were no prealloc | |
8719 | * extents beyond i_size to drop. | |
8720 | */ | |
d9ac19c3 | 8721 | if (control.extents_found > 0) |
23e3337f | 8722 | btrfs_set_inode_full_sync(BTRFS_I(inode)); |
fcb80c2a | 8723 | |
ad7e1a74 | 8724 | return ret; |
39279cc3 CM |
8725 | } |
8726 | ||
d352ac68 CM |
8727 | /* |
8728 | * create a new subvolume directory/inode (helper for the ioctl). | |
8729 | */ | |
d2fb3437 | 8730 | int btrfs_create_subvol_root(struct btrfs_trans_handle *trans, |
63541927 | 8731 | struct btrfs_root *new_root, |
4d4340c9 CB |
8732 | struct btrfs_root *parent_root, |
8733 | struct user_namespace *mnt_userns) | |
39279cc3 | 8734 | { |
39279cc3 | 8735 | struct inode *inode; |
76dda93c | 8736 | int err; |
00e4e6b3 | 8737 | u64 index = 0; |
39279cc3 | 8738 | |
4d4340c9 | 8739 | inode = btrfs_new_inode(trans, new_root, mnt_userns, NULL, "..", 2, |
12fc9d09 FA |
8740 | S_IFDIR | (~current_umask() & S_IRWXUGO), |
8741 | &index); | |
54aa1f4d | 8742 | if (IS_ERR(inode)) |
f46b5a66 | 8743 | return PTR_ERR(inode); |
39279cc3 CM |
8744 | inode->i_op = &btrfs_dir_inode_operations; |
8745 | inode->i_fop = &btrfs_dir_file_operations; | |
8746 | ||
b0d5d10f | 8747 | unlock_new_inode(inode); |
3b96362c | 8748 | |
63541927 FDBM |
8749 | err = btrfs_subvol_inherit_props(trans, new_root, parent_root); |
8750 | if (err) | |
8751 | btrfs_err(new_root->fs_info, | |
351fd353 | 8752 | "error inheriting subvolume %llu properties: %d", |
63541927 FDBM |
8753 | new_root->root_key.objectid, err); |
8754 | ||
9a56fcd1 | 8755 | err = btrfs_update_inode(trans, new_root, BTRFS_I(inode)); |
cb8e7090 | 8756 | |
76dda93c | 8757 | iput(inode); |
ce598979 | 8758 | return err; |
39279cc3 CM |
8759 | } |
8760 | ||
39279cc3 CM |
8761 | struct inode *btrfs_alloc_inode(struct super_block *sb) |
8762 | { | |
69fe2d75 | 8763 | struct btrfs_fs_info *fs_info = btrfs_sb(sb); |
39279cc3 | 8764 | struct btrfs_inode *ei; |
2ead6ae7 | 8765 | struct inode *inode; |
39279cc3 | 8766 | |
fd60b288 | 8767 | ei = alloc_inode_sb(sb, btrfs_inode_cachep, GFP_KERNEL); |
39279cc3 CM |
8768 | if (!ei) |
8769 | return NULL; | |
2ead6ae7 YZ |
8770 | |
8771 | ei->root = NULL; | |
2ead6ae7 | 8772 | ei->generation = 0; |
15ee9bc7 | 8773 | ei->last_trans = 0; |
257c62e1 | 8774 | ei->last_sub_trans = 0; |
e02119d5 | 8775 | ei->logged_trans = 0; |
2ead6ae7 | 8776 | ei->delalloc_bytes = 0; |
a7e3b975 | 8777 | ei->new_delalloc_bytes = 0; |
47059d93 | 8778 | ei->defrag_bytes = 0; |
2ead6ae7 YZ |
8779 | ei->disk_i_size = 0; |
8780 | ei->flags = 0; | |
77eea05e | 8781 | ei->ro_flags = 0; |
7709cde3 | 8782 | ei->csum_bytes = 0; |
2ead6ae7 | 8783 | ei->index_cnt = (u64)-1; |
67de1176 | 8784 | ei->dir_index = 0; |
2ead6ae7 | 8785 | ei->last_unlink_trans = 0; |
3ebac17c | 8786 | ei->last_reflink_trans = 0; |
46d8bc34 | 8787 | ei->last_log_commit = 0; |
2ead6ae7 | 8788 | |
9e0baf60 JB |
8789 | spin_lock_init(&ei->lock); |
8790 | ei->outstanding_extents = 0; | |
69fe2d75 JB |
8791 | if (sb->s_magic != BTRFS_TEST_MAGIC) |
8792 | btrfs_init_metadata_block_rsv(fs_info, &ei->block_rsv, | |
8793 | BTRFS_BLOCK_RSV_DELALLOC); | |
72ac3c0d | 8794 | ei->runtime_flags = 0; |
b52aa8c9 | 8795 | ei->prop_compress = BTRFS_COMPRESS_NONE; |
eec63c65 | 8796 | ei->defrag_compress = BTRFS_COMPRESS_NONE; |
2ead6ae7 | 8797 | |
16cdcec7 MX |
8798 | ei->delayed_node = NULL; |
8799 | ||
9cc97d64 | 8800 | ei->i_otime.tv_sec = 0; |
8801 | ei->i_otime.tv_nsec = 0; | |
8802 | ||
2ead6ae7 | 8803 | inode = &ei->vfs_inode; |
a8067e02 | 8804 | extent_map_tree_init(&ei->extent_tree); |
43eb5f29 QW |
8805 | extent_io_tree_init(fs_info, &ei->io_tree, IO_TREE_INODE_IO, inode); |
8806 | extent_io_tree_init(fs_info, &ei->io_failure_tree, | |
8807 | IO_TREE_INODE_IO_FAILURE, inode); | |
41a2ee75 JB |
8808 | extent_io_tree_init(fs_info, &ei->file_extent_tree, |
8809 | IO_TREE_INODE_FILE_EXTENT, inode); | |
7b439738 DS |
8810 | ei->io_tree.track_uptodate = true; |
8811 | ei->io_failure_tree.track_uptodate = true; | |
b812ce28 | 8812 | atomic_set(&ei->sync_writers, 0); |
2ead6ae7 | 8813 | mutex_init(&ei->log_mutex); |
e6dcd2dc | 8814 | btrfs_ordered_inode_tree_init(&ei->ordered_tree); |
2ead6ae7 | 8815 | INIT_LIST_HEAD(&ei->delalloc_inodes); |
8089fe62 | 8816 | INIT_LIST_HEAD(&ei->delayed_iput); |
2ead6ae7 | 8817 | RB_CLEAR_NODE(&ei->rb_node); |
8318ba79 | 8818 | init_rwsem(&ei->i_mmap_lock); |
2ead6ae7 YZ |
8819 | |
8820 | return inode; | |
39279cc3 CM |
8821 | } |
8822 | ||
aaedb55b JB |
8823 | #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS |
8824 | void btrfs_test_destroy_inode(struct inode *inode) | |
8825 | { | |
dcdbc059 | 8826 | btrfs_drop_extent_cache(BTRFS_I(inode), 0, (u64)-1, 0); |
aaedb55b JB |
8827 | kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode)); |
8828 | } | |
8829 | #endif | |
8830 | ||
26602cab | 8831 | void btrfs_free_inode(struct inode *inode) |
fa0d7e3d | 8832 | { |
fa0d7e3d NP |
8833 | kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode)); |
8834 | } | |
8835 | ||
633cc816 | 8836 | void btrfs_destroy_inode(struct inode *vfs_inode) |
39279cc3 | 8837 | { |
e6dcd2dc | 8838 | struct btrfs_ordered_extent *ordered; |
633cc816 NB |
8839 | struct btrfs_inode *inode = BTRFS_I(vfs_inode); |
8840 | struct btrfs_root *root = inode->root; | |
5a3f23d5 | 8841 | |
633cc816 NB |
8842 | WARN_ON(!hlist_empty(&vfs_inode->i_dentry)); |
8843 | WARN_ON(vfs_inode->i_data.nrpages); | |
8844 | WARN_ON(inode->block_rsv.reserved); | |
8845 | WARN_ON(inode->block_rsv.size); | |
8846 | WARN_ON(inode->outstanding_extents); | |
dc287224 FM |
8847 | if (!S_ISDIR(vfs_inode->i_mode)) { |
8848 | WARN_ON(inode->delalloc_bytes); | |
8849 | WARN_ON(inode->new_delalloc_bytes); | |
8850 | } | |
633cc816 NB |
8851 | WARN_ON(inode->csum_bytes); |
8852 | WARN_ON(inode->defrag_bytes); | |
39279cc3 | 8853 | |
a6dbd429 JB |
8854 | /* |
8855 | * This can happen where we create an inode, but somebody else also | |
8856 | * created the same inode and we need to destroy the one we already | |
8857 | * created. | |
8858 | */ | |
8859 | if (!root) | |
26602cab | 8860 | return; |
a6dbd429 | 8861 | |
d397712b | 8862 | while (1) { |
633cc816 | 8863 | ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1); |
e6dcd2dc CM |
8864 | if (!ordered) |
8865 | break; | |
8866 | else { | |
633cc816 | 8867 | btrfs_err(root->fs_info, |
5d163e0e | 8868 | "found ordered extent %llu %llu on inode cleanup", |
bffe633e | 8869 | ordered->file_offset, ordered->num_bytes); |
71fe0a55 | 8870 | btrfs_remove_ordered_extent(inode, ordered); |
e6dcd2dc CM |
8871 | btrfs_put_ordered_extent(ordered); |
8872 | btrfs_put_ordered_extent(ordered); | |
8873 | } | |
8874 | } | |
633cc816 NB |
8875 | btrfs_qgroup_check_reserved_leak(inode); |
8876 | inode_tree_del(inode); | |
8877 | btrfs_drop_extent_cache(inode, 0, (u64)-1, 0); | |
8878 | btrfs_inode_clear_file_extent_range(inode, 0, (u64)-1); | |
8879 | btrfs_put_root(inode->root); | |
39279cc3 CM |
8880 | } |
8881 | ||
45321ac5 | 8882 | int btrfs_drop_inode(struct inode *inode) |
76dda93c YZ |
8883 | { |
8884 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
45321ac5 | 8885 | |
6379ef9f NA |
8886 | if (root == NULL) |
8887 | return 1; | |
8888 | ||
fa6ac876 | 8889 | /* the snap/subvol tree is on deleting */ |
69e9c6c6 | 8890 | if (btrfs_root_refs(&root->root_item) == 0) |
45321ac5 | 8891 | return 1; |
76dda93c | 8892 | else |
45321ac5 | 8893 | return generic_drop_inode(inode); |
76dda93c YZ |
8894 | } |
8895 | ||
0ee0fda0 | 8896 | static void init_once(void *foo) |
39279cc3 CM |
8897 | { |
8898 | struct btrfs_inode *ei = (struct btrfs_inode *) foo; | |
8899 | ||
8900 | inode_init_once(&ei->vfs_inode); | |
8901 | } | |
8902 | ||
e67c718b | 8903 | void __cold btrfs_destroy_cachep(void) |
39279cc3 | 8904 | { |
8c0a8537 KS |
8905 | /* |
8906 | * Make sure all delayed rcu free inodes are flushed before we | |
8907 | * destroy cache. | |
8908 | */ | |
8909 | rcu_barrier(); | |
5598e900 KM |
8910 | kmem_cache_destroy(btrfs_inode_cachep); |
8911 | kmem_cache_destroy(btrfs_trans_handle_cachep); | |
5598e900 KM |
8912 | kmem_cache_destroy(btrfs_path_cachep); |
8913 | kmem_cache_destroy(btrfs_free_space_cachep); | |
3acd4850 | 8914 | kmem_cache_destroy(btrfs_free_space_bitmap_cachep); |
39279cc3 CM |
8915 | } |
8916 | ||
f5c29bd9 | 8917 | int __init btrfs_init_cachep(void) |
39279cc3 | 8918 | { |
837e1972 | 8919 | btrfs_inode_cachep = kmem_cache_create("btrfs_inode", |
9601e3f6 | 8920 | sizeof(struct btrfs_inode), 0, |
5d097056 VD |
8921 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD | SLAB_ACCOUNT, |
8922 | init_once); | |
39279cc3 CM |
8923 | if (!btrfs_inode_cachep) |
8924 | goto fail; | |
9601e3f6 | 8925 | |
837e1972 | 8926 | btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle", |
9601e3f6 | 8927 | sizeof(struct btrfs_trans_handle), 0, |
fba4b697 | 8928 | SLAB_TEMPORARY | SLAB_MEM_SPREAD, NULL); |
39279cc3 CM |
8929 | if (!btrfs_trans_handle_cachep) |
8930 | goto fail; | |
9601e3f6 | 8931 | |
837e1972 | 8932 | btrfs_path_cachep = kmem_cache_create("btrfs_path", |
9601e3f6 | 8933 | sizeof(struct btrfs_path), 0, |
fba4b697 | 8934 | SLAB_MEM_SPREAD, NULL); |
39279cc3 CM |
8935 | if (!btrfs_path_cachep) |
8936 | goto fail; | |
9601e3f6 | 8937 | |
837e1972 | 8938 | btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space", |
dc89e982 | 8939 | sizeof(struct btrfs_free_space), 0, |
fba4b697 | 8940 | SLAB_MEM_SPREAD, NULL); |
dc89e982 JB |
8941 | if (!btrfs_free_space_cachep) |
8942 | goto fail; | |
8943 | ||
3acd4850 CL |
8944 | btrfs_free_space_bitmap_cachep = kmem_cache_create("btrfs_free_space_bitmap", |
8945 | PAGE_SIZE, PAGE_SIZE, | |
34e49994 | 8946 | SLAB_MEM_SPREAD, NULL); |
3acd4850 CL |
8947 | if (!btrfs_free_space_bitmap_cachep) |
8948 | goto fail; | |
8949 | ||
39279cc3 CM |
8950 | return 0; |
8951 | fail: | |
8952 | btrfs_destroy_cachep(); | |
8953 | return -ENOMEM; | |
8954 | } | |
8955 | ||
549c7297 CB |
8956 | static int btrfs_getattr(struct user_namespace *mnt_userns, |
8957 | const struct path *path, struct kstat *stat, | |
a528d35e | 8958 | u32 request_mask, unsigned int flags) |
39279cc3 | 8959 | { |
df0af1a5 | 8960 | u64 delalloc_bytes; |
2766ff61 | 8961 | u64 inode_bytes; |
a528d35e | 8962 | struct inode *inode = d_inode(path->dentry); |
fadc0d8b | 8963 | u32 blocksize = inode->i_sb->s_blocksize; |
04a87e34 | 8964 | u32 bi_flags = BTRFS_I(inode)->flags; |
14605409 | 8965 | u32 bi_ro_flags = BTRFS_I(inode)->ro_flags; |
04a87e34 YS |
8966 | |
8967 | stat->result_mask |= STATX_BTIME; | |
8968 | stat->btime.tv_sec = BTRFS_I(inode)->i_otime.tv_sec; | |
8969 | stat->btime.tv_nsec = BTRFS_I(inode)->i_otime.tv_nsec; | |
8970 | if (bi_flags & BTRFS_INODE_APPEND) | |
8971 | stat->attributes |= STATX_ATTR_APPEND; | |
8972 | if (bi_flags & BTRFS_INODE_COMPRESS) | |
8973 | stat->attributes |= STATX_ATTR_COMPRESSED; | |
8974 | if (bi_flags & BTRFS_INODE_IMMUTABLE) | |
8975 | stat->attributes |= STATX_ATTR_IMMUTABLE; | |
8976 | if (bi_flags & BTRFS_INODE_NODUMP) | |
8977 | stat->attributes |= STATX_ATTR_NODUMP; | |
14605409 BB |
8978 | if (bi_ro_flags & BTRFS_INODE_RO_VERITY) |
8979 | stat->attributes |= STATX_ATTR_VERITY; | |
04a87e34 YS |
8980 | |
8981 | stat->attributes_mask |= (STATX_ATTR_APPEND | | |
8982 | STATX_ATTR_COMPRESSED | | |
8983 | STATX_ATTR_IMMUTABLE | | |
8984 | STATX_ATTR_NODUMP); | |
fadc0d8b | 8985 | |
c020d2ea | 8986 | generic_fillattr(mnt_userns, inode, stat); |
0ee5dc67 | 8987 | stat->dev = BTRFS_I(inode)->root->anon_dev; |
df0af1a5 MX |
8988 | |
8989 | spin_lock(&BTRFS_I(inode)->lock); | |
a7e3b975 | 8990 | delalloc_bytes = BTRFS_I(inode)->new_delalloc_bytes; |
2766ff61 | 8991 | inode_bytes = inode_get_bytes(inode); |
df0af1a5 | 8992 | spin_unlock(&BTRFS_I(inode)->lock); |
2766ff61 | 8993 | stat->blocks = (ALIGN(inode_bytes, blocksize) + |
df0af1a5 | 8994 | ALIGN(delalloc_bytes, blocksize)) >> 9; |
39279cc3 CM |
8995 | return 0; |
8996 | } | |
8997 | ||
cdd1fedf DF |
8998 | static int btrfs_rename_exchange(struct inode *old_dir, |
8999 | struct dentry *old_dentry, | |
9000 | struct inode *new_dir, | |
9001 | struct dentry *new_dentry) | |
9002 | { | |
0b246afa | 9003 | struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb); |
cdd1fedf | 9004 | struct btrfs_trans_handle *trans; |
c1621871 | 9005 | unsigned int trans_num_items; |
cdd1fedf DF |
9006 | struct btrfs_root *root = BTRFS_I(old_dir)->root; |
9007 | struct btrfs_root *dest = BTRFS_I(new_dir)->root; | |
9008 | struct inode *new_inode = new_dentry->d_inode; | |
9009 | struct inode *old_inode = old_dentry->d_inode; | |
95582b00 | 9010 | struct timespec64 ctime = current_time(old_inode); |
88d2beec FM |
9011 | struct btrfs_rename_ctx old_rename_ctx; |
9012 | struct btrfs_rename_ctx new_rename_ctx; | |
4a0cc7ca NB |
9013 | u64 old_ino = btrfs_ino(BTRFS_I(old_inode)); |
9014 | u64 new_ino = btrfs_ino(BTRFS_I(new_inode)); | |
cdd1fedf DF |
9015 | u64 old_idx = 0; |
9016 | u64 new_idx = 0; | |
cdd1fedf | 9017 | int ret; |
75b463d2 | 9018 | int ret2; |
dc09ef35 | 9019 | bool need_abort = false; |
cdd1fedf | 9020 | |
3f79f6f6 N |
9021 | /* |
9022 | * For non-subvolumes allow exchange only within one subvolume, in the | |
9023 | * same inode namespace. Two subvolumes (represented as directory) can | |
9024 | * be exchanged as they're a logical link and have a fixed inode number. | |
9025 | */ | |
9026 | if (root != dest && | |
9027 | (old_ino != BTRFS_FIRST_FREE_OBJECTID || | |
9028 | new_ino != BTRFS_FIRST_FREE_OBJECTID)) | |
cdd1fedf DF |
9029 | return -EXDEV; |
9030 | ||
9031 | /* close the race window with snapshot create/destroy ioctl */ | |
943eb3bf JB |
9032 | if (old_ino == BTRFS_FIRST_FREE_OBJECTID || |
9033 | new_ino == BTRFS_FIRST_FREE_OBJECTID) | |
0b246afa | 9034 | down_read(&fs_info->subvol_sem); |
cdd1fedf DF |
9035 | |
9036 | /* | |
c1621871 OS |
9037 | * For each inode: |
9038 | * 1 to remove old dir item | |
9039 | * 1 to remove old dir index | |
9040 | * 1 to add new dir item | |
9041 | * 1 to add new dir index | |
9042 | * 1 to update parent inode | |
9043 | * | |
9044 | * If the parents are the same, we only need to account for one | |
cdd1fedf | 9045 | */ |
c1621871 OS |
9046 | trans_num_items = (old_dir == new_dir ? 9 : 10); |
9047 | if (old_ino == BTRFS_FIRST_FREE_OBJECTID) { | |
9048 | /* | |
9049 | * 1 to remove old root ref | |
9050 | * 1 to remove old root backref | |
9051 | * 1 to add new root ref | |
9052 | * 1 to add new root backref | |
9053 | */ | |
9054 | trans_num_items += 4; | |
9055 | } else { | |
9056 | /* | |
9057 | * 1 to update inode item | |
9058 | * 1 to remove old inode ref | |
9059 | * 1 to add new inode ref | |
9060 | */ | |
9061 | trans_num_items += 3; | |
9062 | } | |
9063 | if (new_ino == BTRFS_FIRST_FREE_OBJECTID) | |
9064 | trans_num_items += 4; | |
9065 | else | |
9066 | trans_num_items += 3; | |
9067 | trans = btrfs_start_transaction(root, trans_num_items); | |
cdd1fedf DF |
9068 | if (IS_ERR(trans)) { |
9069 | ret = PTR_ERR(trans); | |
9070 | goto out_notrans; | |
9071 | } | |
9072 | ||
00aa8e87 JB |
9073 | if (dest != root) { |
9074 | ret = btrfs_record_root_in_trans(trans, dest); | |
9075 | if (ret) | |
9076 | goto out_fail; | |
9077 | } | |
3e174099 | 9078 | |
cdd1fedf DF |
9079 | /* |
9080 | * We need to find a free sequence number both in the source and | |
9081 | * in the destination directory for the exchange. | |
9082 | */ | |
877574e2 | 9083 | ret = btrfs_set_inode_index(BTRFS_I(new_dir), &old_idx); |
cdd1fedf DF |
9084 | if (ret) |
9085 | goto out_fail; | |
877574e2 | 9086 | ret = btrfs_set_inode_index(BTRFS_I(old_dir), &new_idx); |
cdd1fedf DF |
9087 | if (ret) |
9088 | goto out_fail; | |
9089 | ||
9090 | BTRFS_I(old_inode)->dir_index = 0ULL; | |
9091 | BTRFS_I(new_inode)->dir_index = 0ULL; | |
9092 | ||
9093 | /* Reference for the source. */ | |
9094 | if (old_ino == BTRFS_FIRST_FREE_OBJECTID) { | |
9095 | /* force full log commit if subvolume involved. */ | |
90787766 | 9096 | btrfs_set_log_full_commit(trans); |
cdd1fedf DF |
9097 | } else { |
9098 | ret = btrfs_insert_inode_ref(trans, dest, | |
9099 | new_dentry->d_name.name, | |
9100 | new_dentry->d_name.len, | |
9101 | old_ino, | |
f85b7379 DS |
9102 | btrfs_ino(BTRFS_I(new_dir)), |
9103 | old_idx); | |
cdd1fedf DF |
9104 | if (ret) |
9105 | goto out_fail; | |
dc09ef35 | 9106 | need_abort = true; |
cdd1fedf DF |
9107 | } |
9108 | ||
9109 | /* And now for the dest. */ | |
9110 | if (new_ino == BTRFS_FIRST_FREE_OBJECTID) { | |
9111 | /* force full log commit if subvolume involved. */ | |
90787766 | 9112 | btrfs_set_log_full_commit(trans); |
cdd1fedf DF |
9113 | } else { |
9114 | ret = btrfs_insert_inode_ref(trans, root, | |
9115 | old_dentry->d_name.name, | |
9116 | old_dentry->d_name.len, | |
9117 | new_ino, | |
f85b7379 DS |
9118 | btrfs_ino(BTRFS_I(old_dir)), |
9119 | new_idx); | |
dc09ef35 JB |
9120 | if (ret) { |
9121 | if (need_abort) | |
9122 | btrfs_abort_transaction(trans, ret); | |
cdd1fedf | 9123 | goto out_fail; |
dc09ef35 | 9124 | } |
cdd1fedf DF |
9125 | } |
9126 | ||
9127 | /* Update inode version and ctime/mtime. */ | |
9128 | inode_inc_iversion(old_dir); | |
9129 | inode_inc_iversion(new_dir); | |
9130 | inode_inc_iversion(old_inode); | |
9131 | inode_inc_iversion(new_inode); | |
9132 | old_dir->i_ctime = old_dir->i_mtime = ctime; | |
9133 | new_dir->i_ctime = new_dir->i_mtime = ctime; | |
9134 | old_inode->i_ctime = ctime; | |
9135 | new_inode->i_ctime = ctime; | |
9136 | ||
9137 | if (old_dentry->d_parent != new_dentry->d_parent) { | |
f85b7379 DS |
9138 | btrfs_record_unlink_dir(trans, BTRFS_I(old_dir), |
9139 | BTRFS_I(old_inode), 1); | |
9140 | btrfs_record_unlink_dir(trans, BTRFS_I(new_dir), | |
9141 | BTRFS_I(new_inode), 1); | |
cdd1fedf DF |
9142 | } |
9143 | ||
9144 | /* src is a subvolume */ | |
9145 | if (old_ino == BTRFS_FIRST_FREE_OBJECTID) { | |
045d3967 | 9146 | ret = btrfs_unlink_subvol(trans, old_dir, old_dentry); |
cdd1fedf | 9147 | } else { /* src is an inode */ |
4467af88 | 9148 | ret = __btrfs_unlink_inode(trans, BTRFS_I(old_dir), |
4ec5934e | 9149 | BTRFS_I(old_dentry->d_inode), |
cdd1fedf | 9150 | old_dentry->d_name.name, |
88d2beec FM |
9151 | old_dentry->d_name.len, |
9152 | &old_rename_ctx); | |
cdd1fedf | 9153 | if (!ret) |
9a56fcd1 | 9154 | ret = btrfs_update_inode(trans, root, BTRFS_I(old_inode)); |
cdd1fedf DF |
9155 | } |
9156 | if (ret) { | |
66642832 | 9157 | btrfs_abort_transaction(trans, ret); |
cdd1fedf DF |
9158 | goto out_fail; |
9159 | } | |
9160 | ||
9161 | /* dest is a subvolume */ | |
9162 | if (new_ino == BTRFS_FIRST_FREE_OBJECTID) { | |
045d3967 | 9163 | ret = btrfs_unlink_subvol(trans, new_dir, new_dentry); |
cdd1fedf | 9164 | } else { /* dest is an inode */ |
4467af88 | 9165 | ret = __btrfs_unlink_inode(trans, BTRFS_I(new_dir), |
4ec5934e | 9166 | BTRFS_I(new_dentry->d_inode), |
cdd1fedf | 9167 | new_dentry->d_name.name, |
88d2beec FM |
9168 | new_dentry->d_name.len, |
9169 | &new_rename_ctx); | |
cdd1fedf | 9170 | if (!ret) |
9a56fcd1 | 9171 | ret = btrfs_update_inode(trans, dest, BTRFS_I(new_inode)); |
cdd1fedf DF |
9172 | } |
9173 | if (ret) { | |
66642832 | 9174 | btrfs_abort_transaction(trans, ret); |
cdd1fedf DF |
9175 | goto out_fail; |
9176 | } | |
9177 | ||
db0a669f | 9178 | ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode), |
cdd1fedf DF |
9179 | new_dentry->d_name.name, |
9180 | new_dentry->d_name.len, 0, old_idx); | |
9181 | if (ret) { | |
66642832 | 9182 | btrfs_abort_transaction(trans, ret); |
cdd1fedf DF |
9183 | goto out_fail; |
9184 | } | |
9185 | ||
db0a669f | 9186 | ret = btrfs_add_link(trans, BTRFS_I(old_dir), BTRFS_I(new_inode), |
cdd1fedf DF |
9187 | old_dentry->d_name.name, |
9188 | old_dentry->d_name.len, 0, new_idx); | |
9189 | if (ret) { | |
66642832 | 9190 | btrfs_abort_transaction(trans, ret); |
cdd1fedf DF |
9191 | goto out_fail; |
9192 | } | |
9193 | ||
9194 | if (old_inode->i_nlink == 1) | |
9195 | BTRFS_I(old_inode)->dir_index = old_idx; | |
9196 | if (new_inode->i_nlink == 1) | |
9197 | BTRFS_I(new_inode)->dir_index = new_idx; | |
9198 | ||
259c4b96 FM |
9199 | /* |
9200 | * Now pin the logs of the roots. We do it to ensure that no other task | |
9201 | * can sync the logs while we are in progress with the rename, because | |
9202 | * that could result in an inconsistency in case any of the inodes that | |
9203 | * are part of this rename operation were logged before. | |
9204 | */ | |
9205 | if (old_ino != BTRFS_FIRST_FREE_OBJECTID) | |
9206 | btrfs_pin_log_trans(root); | |
9207 | if (new_ino != BTRFS_FIRST_FREE_OBJECTID) | |
9208 | btrfs_pin_log_trans(dest); | |
9209 | ||
9210 | /* Do the log updates for all inodes. */ | |
9211 | if (old_ino != BTRFS_FIRST_FREE_OBJECTID) | |
d5f5bd54 | 9212 | btrfs_log_new_name(trans, old_dentry, BTRFS_I(old_dir), |
88d2beec | 9213 | old_rename_ctx.index, new_dentry->d_parent); |
259c4b96 | 9214 | if (new_ino != BTRFS_FIRST_FREE_OBJECTID) |
d5f5bd54 | 9215 | btrfs_log_new_name(trans, new_dentry, BTRFS_I(new_dir), |
88d2beec | 9216 | new_rename_ctx.index, old_dentry->d_parent); |
259c4b96 FM |
9217 | |
9218 | /* Now unpin the logs. */ | |
9219 | if (old_ino != BTRFS_FIRST_FREE_OBJECTID) | |
9220 | btrfs_end_log_trans(root); | |
9221 | if (new_ino != BTRFS_FIRST_FREE_OBJECTID) | |
cdd1fedf | 9222 | btrfs_end_log_trans(dest); |
cdd1fedf | 9223 | out_fail: |
75b463d2 FM |
9224 | ret2 = btrfs_end_transaction(trans); |
9225 | ret = ret ? ret : ret2; | |
cdd1fedf | 9226 | out_notrans: |
943eb3bf JB |
9227 | if (new_ino == BTRFS_FIRST_FREE_OBJECTID || |
9228 | old_ino == BTRFS_FIRST_FREE_OBJECTID) | |
0b246afa | 9229 | up_read(&fs_info->subvol_sem); |
cdd1fedf DF |
9230 | |
9231 | return ret; | |
9232 | } | |
9233 | ||
9234 | static int btrfs_whiteout_for_rename(struct btrfs_trans_handle *trans, | |
9235 | struct btrfs_root *root, | |
ca07274c | 9236 | struct user_namespace *mnt_userns, |
cdd1fedf DF |
9237 | struct inode *dir, |
9238 | struct dentry *dentry) | |
9239 | { | |
9240 | int ret; | |
9241 | struct inode *inode; | |
cdd1fedf DF |
9242 | u64 index; |
9243 | ||
ca07274c | 9244 | inode = btrfs_new_inode(trans, root, mnt_userns, dir, |
cdd1fedf DF |
9245 | dentry->d_name.name, |
9246 | dentry->d_name.len, | |
cdd1fedf DF |
9247 | S_IFCHR | WHITEOUT_MODE, |
9248 | &index); | |
9249 | ||
9250 | if (IS_ERR(inode)) { | |
9251 | ret = PTR_ERR(inode); | |
9252 | return ret; | |
9253 | } | |
9254 | ||
9255 | inode->i_op = &btrfs_special_inode_operations; | |
9256 | init_special_inode(inode, inode->i_mode, | |
9257 | WHITEOUT_DEV); | |
9258 | ||
9259 | ret = btrfs_init_inode_security(trans, inode, dir, | |
9260 | &dentry->d_name); | |
9261 | if (ret) | |
c9901618 | 9262 | goto out; |
cdd1fedf | 9263 | |
81512e89 OS |
9264 | ret = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode), |
9265 | dentry->d_name.name, dentry->d_name.len, 0, index); | |
cdd1fedf | 9266 | if (ret) |
c9901618 | 9267 | goto out; |
cdd1fedf | 9268 | |
9a56fcd1 | 9269 | ret = btrfs_update_inode(trans, root, BTRFS_I(inode)); |
c9901618 | 9270 | out: |
cdd1fedf | 9271 | unlock_new_inode(inode); |
c9901618 FM |
9272 | if (ret) |
9273 | inode_dec_link_count(inode); | |
cdd1fedf DF |
9274 | iput(inode); |
9275 | ||
c9901618 | 9276 | return ret; |
cdd1fedf DF |
9277 | } |
9278 | ||
ca07274c CB |
9279 | static int btrfs_rename(struct user_namespace *mnt_userns, |
9280 | struct inode *old_dir, struct dentry *old_dentry, | |
9281 | struct inode *new_dir, struct dentry *new_dentry, | |
9282 | unsigned int flags) | |
39279cc3 | 9283 | { |
0b246afa | 9284 | struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb); |
39279cc3 | 9285 | struct btrfs_trans_handle *trans; |
5062af35 | 9286 | unsigned int trans_num_items; |
39279cc3 | 9287 | struct btrfs_root *root = BTRFS_I(old_dir)->root; |
4df27c4d | 9288 | struct btrfs_root *dest = BTRFS_I(new_dir)->root; |
2b0143b5 DH |
9289 | struct inode *new_inode = d_inode(new_dentry); |
9290 | struct inode *old_inode = d_inode(old_dentry); | |
88d2beec | 9291 | struct btrfs_rename_ctx rename_ctx; |
00e4e6b3 | 9292 | u64 index = 0; |
39279cc3 | 9293 | int ret; |
75b463d2 | 9294 | int ret2; |
4a0cc7ca | 9295 | u64 old_ino = btrfs_ino(BTRFS_I(old_inode)); |
39279cc3 | 9296 | |
4a0cc7ca | 9297 | if (btrfs_ino(BTRFS_I(new_dir)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) |
f679a840 YZ |
9298 | return -EPERM; |
9299 | ||
4df27c4d | 9300 | /* we only allow rename subvolume link between subvolumes */ |
33345d01 | 9301 | if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest) |
3394e160 CM |
9302 | return -EXDEV; |
9303 | ||
33345d01 | 9304 | if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID || |
4a0cc7ca | 9305 | (new_inode && btrfs_ino(BTRFS_I(new_inode)) == BTRFS_FIRST_FREE_OBJECTID)) |
39279cc3 | 9306 | return -ENOTEMPTY; |
5f39d397 | 9307 | |
4df27c4d YZ |
9308 | if (S_ISDIR(old_inode->i_mode) && new_inode && |
9309 | new_inode->i_size > BTRFS_EMPTY_DIR_SIZE) | |
9310 | return -ENOTEMPTY; | |
9c52057c CM |
9311 | |
9312 | ||
9313 | /* check for collisions, even if the name isn't there */ | |
4871c158 | 9314 | ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino, |
9c52057c CM |
9315 | new_dentry->d_name.name, |
9316 | new_dentry->d_name.len); | |
9317 | ||
9318 | if (ret) { | |
9319 | if (ret == -EEXIST) { | |
9320 | /* we shouldn't get | |
9321 | * eexist without a new_inode */ | |
fae7f21c | 9322 | if (WARN_ON(!new_inode)) { |
9c52057c CM |
9323 | return ret; |
9324 | } | |
9325 | } else { | |
9326 | /* maybe -EOVERFLOW */ | |
9327 | return ret; | |
9328 | } | |
9329 | } | |
9330 | ret = 0; | |
9331 | ||
5a3f23d5 | 9332 | /* |
8d875f95 CM |
9333 | * we're using rename to replace one file with another. Start IO on it |
9334 | * now so we don't add too much work to the end of the transaction | |
5a3f23d5 | 9335 | */ |
8d875f95 | 9336 | if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size) |
5a3f23d5 CM |
9337 | filemap_flush(old_inode->i_mapping); |
9338 | ||
c1621871 OS |
9339 | if (old_ino == BTRFS_FIRST_FREE_OBJECTID) { |
9340 | /* Close the race window with snapshot create/destroy ioctl */ | |
0b246afa | 9341 | down_read(&fs_info->subvol_sem); |
c1621871 OS |
9342 | /* |
9343 | * 1 to remove old root ref | |
9344 | * 1 to remove old root backref | |
9345 | * 1 to add new root ref | |
9346 | * 1 to add new root backref | |
9347 | */ | |
9348 | trans_num_items = 4; | |
9349 | } else { | |
9350 | /* | |
9351 | * 1 to update inode | |
9352 | * 1 to remove old inode ref | |
9353 | * 1 to add new inode ref | |
9354 | */ | |
9355 | trans_num_items = 3; | |
9356 | } | |
a22285a6 | 9357 | /* |
c1621871 OS |
9358 | * 1 to remove old dir item |
9359 | * 1 to remove old dir index | |
9360 | * 1 to update old parent inode | |
9361 | * 1 to add new dir item | |
9362 | * 1 to add new dir index | |
9363 | * 1 to update new parent inode (if it's not the same as the old parent) | |
a22285a6 | 9364 | */ |
c1621871 OS |
9365 | trans_num_items += 6; |
9366 | if (new_dir != old_dir) | |
9367 | trans_num_items++; | |
9368 | if (new_inode) { | |
9369 | /* | |
9370 | * 1 to update inode | |
9371 | * 1 to remove inode ref | |
9372 | * 1 to remove dir item | |
9373 | * 1 to remove dir index | |
9374 | * 1 to possibly add orphan item | |
9375 | */ | |
9376 | trans_num_items += 5; | |
9377 | } | |
5062af35 FM |
9378 | if (flags & RENAME_WHITEOUT) |
9379 | trans_num_items += 5; | |
9380 | trans = btrfs_start_transaction(root, trans_num_items); | |
b44c59a8 | 9381 | if (IS_ERR(trans)) { |
cdd1fedf DF |
9382 | ret = PTR_ERR(trans); |
9383 | goto out_notrans; | |
9384 | } | |
76dda93c | 9385 | |
b0fec6fd JB |
9386 | if (dest != root) { |
9387 | ret = btrfs_record_root_in_trans(trans, dest); | |
9388 | if (ret) | |
9389 | goto out_fail; | |
9390 | } | |
5f39d397 | 9391 | |
877574e2 | 9392 | ret = btrfs_set_inode_index(BTRFS_I(new_dir), &index); |
a5719521 YZ |
9393 | if (ret) |
9394 | goto out_fail; | |
5a3f23d5 | 9395 | |
67de1176 | 9396 | BTRFS_I(old_inode)->dir_index = 0ULL; |
33345d01 | 9397 | if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) { |
4df27c4d | 9398 | /* force full log commit if subvolume involved. */ |
90787766 | 9399 | btrfs_set_log_full_commit(trans); |
4df27c4d | 9400 | } else { |
a5719521 YZ |
9401 | ret = btrfs_insert_inode_ref(trans, dest, |
9402 | new_dentry->d_name.name, | |
9403 | new_dentry->d_name.len, | |
33345d01 | 9404 | old_ino, |
4a0cc7ca | 9405 | btrfs_ino(BTRFS_I(new_dir)), index); |
a5719521 YZ |
9406 | if (ret) |
9407 | goto out_fail; | |
4df27c4d | 9408 | } |
5a3f23d5 | 9409 | |
0c4d2d95 JB |
9410 | inode_inc_iversion(old_dir); |
9411 | inode_inc_iversion(new_dir); | |
9412 | inode_inc_iversion(old_inode); | |
04b285f3 DD |
9413 | old_dir->i_ctime = old_dir->i_mtime = |
9414 | new_dir->i_ctime = new_dir->i_mtime = | |
c2050a45 | 9415 | old_inode->i_ctime = current_time(old_dir); |
5f39d397 | 9416 | |
12fcfd22 | 9417 | if (old_dentry->d_parent != new_dentry->d_parent) |
f85b7379 DS |
9418 | btrfs_record_unlink_dir(trans, BTRFS_I(old_dir), |
9419 | BTRFS_I(old_inode), 1); | |
12fcfd22 | 9420 | |
33345d01 | 9421 | if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) { |
045d3967 | 9422 | ret = btrfs_unlink_subvol(trans, old_dir, old_dentry); |
4df27c4d | 9423 | } else { |
4467af88 | 9424 | ret = __btrfs_unlink_inode(trans, BTRFS_I(old_dir), |
4ec5934e | 9425 | BTRFS_I(d_inode(old_dentry)), |
92986796 | 9426 | old_dentry->d_name.name, |
88d2beec FM |
9427 | old_dentry->d_name.len, |
9428 | &rename_ctx); | |
92986796 | 9429 | if (!ret) |
9a56fcd1 | 9430 | ret = btrfs_update_inode(trans, root, BTRFS_I(old_inode)); |
4df27c4d | 9431 | } |
79787eaa | 9432 | if (ret) { |
66642832 | 9433 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
9434 | goto out_fail; |
9435 | } | |
39279cc3 CM |
9436 | |
9437 | if (new_inode) { | |
0c4d2d95 | 9438 | inode_inc_iversion(new_inode); |
c2050a45 | 9439 | new_inode->i_ctime = current_time(new_inode); |
4a0cc7ca | 9440 | if (unlikely(btrfs_ino(BTRFS_I(new_inode)) == |
4df27c4d | 9441 | BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) { |
045d3967 | 9442 | ret = btrfs_unlink_subvol(trans, new_dir, new_dentry); |
4df27c4d YZ |
9443 | BUG_ON(new_inode->i_nlink == 0); |
9444 | } else { | |
4467af88 | 9445 | ret = btrfs_unlink_inode(trans, BTRFS_I(new_dir), |
4ec5934e | 9446 | BTRFS_I(d_inode(new_dentry)), |
4df27c4d YZ |
9447 | new_dentry->d_name.name, |
9448 | new_dentry->d_name.len); | |
9449 | } | |
4ef31a45 | 9450 | if (!ret && new_inode->i_nlink == 0) |
73f2e545 NB |
9451 | ret = btrfs_orphan_add(trans, |
9452 | BTRFS_I(d_inode(new_dentry))); | |
79787eaa | 9453 | if (ret) { |
66642832 | 9454 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
9455 | goto out_fail; |
9456 | } | |
39279cc3 | 9457 | } |
aec7477b | 9458 | |
db0a669f | 9459 | ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode), |
4df27c4d | 9460 | new_dentry->d_name.name, |
a5719521 | 9461 | new_dentry->d_name.len, 0, index); |
79787eaa | 9462 | if (ret) { |
66642832 | 9463 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
9464 | goto out_fail; |
9465 | } | |
39279cc3 | 9466 | |
67de1176 MX |
9467 | if (old_inode->i_nlink == 1) |
9468 | BTRFS_I(old_inode)->dir_index = index; | |
9469 | ||
259c4b96 | 9470 | if (old_ino != BTRFS_FIRST_FREE_OBJECTID) |
d5f5bd54 | 9471 | btrfs_log_new_name(trans, old_dentry, BTRFS_I(old_dir), |
88d2beec | 9472 | rename_ctx.index, new_dentry->d_parent); |
cdd1fedf DF |
9473 | |
9474 | if (flags & RENAME_WHITEOUT) { | |
ca07274c CB |
9475 | ret = btrfs_whiteout_for_rename(trans, root, mnt_userns, |
9476 | old_dir, old_dentry); | |
cdd1fedf DF |
9477 | |
9478 | if (ret) { | |
66642832 | 9479 | btrfs_abort_transaction(trans, ret); |
cdd1fedf DF |
9480 | goto out_fail; |
9481 | } | |
4df27c4d | 9482 | } |
39279cc3 | 9483 | out_fail: |
75b463d2 FM |
9484 | ret2 = btrfs_end_transaction(trans); |
9485 | ret = ret ? ret : ret2; | |
b44c59a8 | 9486 | out_notrans: |
33345d01 | 9487 | if (old_ino == BTRFS_FIRST_FREE_OBJECTID) |
0b246afa | 9488 | up_read(&fs_info->subvol_sem); |
9ed74f2d | 9489 | |
39279cc3 CM |
9490 | return ret; |
9491 | } | |
9492 | ||
549c7297 CB |
9493 | static int btrfs_rename2(struct user_namespace *mnt_userns, struct inode *old_dir, |
9494 | struct dentry *old_dentry, struct inode *new_dir, | |
9495 | struct dentry *new_dentry, unsigned int flags) | |
80ace85c | 9496 | { |
cdd1fedf | 9497 | if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT)) |
80ace85c MS |
9498 | return -EINVAL; |
9499 | ||
cdd1fedf DF |
9500 | if (flags & RENAME_EXCHANGE) |
9501 | return btrfs_rename_exchange(old_dir, old_dentry, new_dir, | |
9502 | new_dentry); | |
9503 | ||
ca07274c CB |
9504 | return btrfs_rename(mnt_userns, old_dir, old_dentry, new_dir, |
9505 | new_dentry, flags); | |
80ace85c MS |
9506 | } |
9507 | ||
3a2f8c07 NB |
9508 | struct btrfs_delalloc_work { |
9509 | struct inode *inode; | |
9510 | struct completion completion; | |
9511 | struct list_head list; | |
9512 | struct btrfs_work work; | |
9513 | }; | |
9514 | ||
8ccf6f19 MX |
9515 | static void btrfs_run_delalloc_work(struct btrfs_work *work) |
9516 | { | |
9517 | struct btrfs_delalloc_work *delalloc_work; | |
9f23e289 | 9518 | struct inode *inode; |
8ccf6f19 MX |
9519 | |
9520 | delalloc_work = container_of(work, struct btrfs_delalloc_work, | |
9521 | work); | |
9f23e289 | 9522 | inode = delalloc_work->inode; |
30424601 DS |
9523 | filemap_flush(inode->i_mapping); |
9524 | if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT, | |
9525 | &BTRFS_I(inode)->runtime_flags)) | |
9f23e289 | 9526 | filemap_flush(inode->i_mapping); |
8ccf6f19 | 9527 | |
076da91c | 9528 | iput(inode); |
8ccf6f19 MX |
9529 | complete(&delalloc_work->completion); |
9530 | } | |
9531 | ||
3a2f8c07 | 9532 | static struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode) |
8ccf6f19 MX |
9533 | { |
9534 | struct btrfs_delalloc_work *work; | |
9535 | ||
100d5702 | 9536 | work = kmalloc(sizeof(*work), GFP_NOFS); |
8ccf6f19 MX |
9537 | if (!work) |
9538 | return NULL; | |
9539 | ||
9540 | init_completion(&work->completion); | |
9541 | INIT_LIST_HEAD(&work->list); | |
9542 | work->inode = inode; | |
a0cac0ec | 9543 | btrfs_init_work(&work->work, btrfs_run_delalloc_work, NULL, NULL); |
8ccf6f19 MX |
9544 | |
9545 | return work; | |
9546 | } | |
9547 | ||
d352ac68 CM |
9548 | /* |
9549 | * some fairly slow code that needs optimization. This walks the list | |
9550 | * of all the inodes with pending delalloc and forces them to disk. | |
9551 | */ | |
e076ab2a JB |
9552 | static int start_delalloc_inodes(struct btrfs_root *root, |
9553 | struct writeback_control *wbc, bool snapshot, | |
3d45f221 | 9554 | bool in_reclaim_context) |
ea8c2819 | 9555 | { |
ea8c2819 | 9556 | struct btrfs_inode *binode; |
5b21f2ed | 9557 | struct inode *inode; |
8ccf6f19 MX |
9558 | struct btrfs_delalloc_work *work, *next; |
9559 | struct list_head works; | |
1eafa6c7 | 9560 | struct list_head splice; |
8ccf6f19 | 9561 | int ret = 0; |
e076ab2a | 9562 | bool full_flush = wbc->nr_to_write == LONG_MAX; |
ea8c2819 | 9563 | |
8ccf6f19 | 9564 | INIT_LIST_HEAD(&works); |
1eafa6c7 | 9565 | INIT_LIST_HEAD(&splice); |
63607cc8 | 9566 | |
573bfb72 | 9567 | mutex_lock(&root->delalloc_mutex); |
eb73c1b7 MX |
9568 | spin_lock(&root->delalloc_lock); |
9569 | list_splice_init(&root->delalloc_inodes, &splice); | |
1eafa6c7 MX |
9570 | while (!list_empty(&splice)) { |
9571 | binode = list_entry(splice.next, struct btrfs_inode, | |
ea8c2819 | 9572 | delalloc_inodes); |
1eafa6c7 | 9573 | |
eb73c1b7 MX |
9574 | list_move_tail(&binode->delalloc_inodes, |
9575 | &root->delalloc_inodes); | |
3d45f221 FM |
9576 | |
9577 | if (in_reclaim_context && | |
9578 | test_bit(BTRFS_INODE_NO_DELALLOC_FLUSH, &binode->runtime_flags)) | |
9579 | continue; | |
9580 | ||
5b21f2ed | 9581 | inode = igrab(&binode->vfs_inode); |
df0af1a5 | 9582 | if (!inode) { |
eb73c1b7 | 9583 | cond_resched_lock(&root->delalloc_lock); |
1eafa6c7 | 9584 | continue; |
df0af1a5 | 9585 | } |
eb73c1b7 | 9586 | spin_unlock(&root->delalloc_lock); |
1eafa6c7 | 9587 | |
3cd24c69 EL |
9588 | if (snapshot) |
9589 | set_bit(BTRFS_INODE_SNAPSHOT_FLUSH, | |
9590 | &binode->runtime_flags); | |
e076ab2a JB |
9591 | if (full_flush) { |
9592 | work = btrfs_alloc_delalloc_work(inode); | |
9593 | if (!work) { | |
9594 | iput(inode); | |
9595 | ret = -ENOMEM; | |
9596 | goto out; | |
9597 | } | |
9598 | list_add_tail(&work->list, &works); | |
9599 | btrfs_queue_work(root->fs_info->flush_workers, | |
9600 | &work->work); | |
9601 | } else { | |
b3776305 | 9602 | ret = filemap_fdatawrite_wbc(inode->i_mapping, wbc); |
e076ab2a JB |
9603 | btrfs_add_delayed_iput(inode); |
9604 | if (ret || wbc->nr_to_write <= 0) | |
b4912139 JB |
9605 | goto out; |
9606 | } | |
5b21f2ed | 9607 | cond_resched(); |
eb73c1b7 | 9608 | spin_lock(&root->delalloc_lock); |
ea8c2819 | 9609 | } |
eb73c1b7 | 9610 | spin_unlock(&root->delalloc_lock); |
8c8bee1d | 9611 | |
a1ecaabb | 9612 | out: |
eb73c1b7 MX |
9613 | list_for_each_entry_safe(work, next, &works, list) { |
9614 | list_del_init(&work->list); | |
40012f96 NB |
9615 | wait_for_completion(&work->completion); |
9616 | kfree(work); | |
eb73c1b7 MX |
9617 | } |
9618 | ||
81f1d390 | 9619 | if (!list_empty(&splice)) { |
eb73c1b7 MX |
9620 | spin_lock(&root->delalloc_lock); |
9621 | list_splice_tail(&splice, &root->delalloc_inodes); | |
9622 | spin_unlock(&root->delalloc_lock); | |
9623 | } | |
573bfb72 | 9624 | mutex_unlock(&root->delalloc_mutex); |
eb73c1b7 MX |
9625 | return ret; |
9626 | } | |
1eafa6c7 | 9627 | |
f9baa501 | 9628 | int btrfs_start_delalloc_snapshot(struct btrfs_root *root, bool in_reclaim_context) |
eb73c1b7 | 9629 | { |
e076ab2a JB |
9630 | struct writeback_control wbc = { |
9631 | .nr_to_write = LONG_MAX, | |
9632 | .sync_mode = WB_SYNC_NONE, | |
9633 | .range_start = 0, | |
9634 | .range_end = LLONG_MAX, | |
9635 | }; | |
0b246afa | 9636 | struct btrfs_fs_info *fs_info = root->fs_info; |
1eafa6c7 | 9637 | |
84961539 | 9638 | if (BTRFS_FS_ERROR(fs_info)) |
eb73c1b7 MX |
9639 | return -EROFS; |
9640 | ||
f9baa501 | 9641 | return start_delalloc_inodes(root, &wbc, true, in_reclaim_context); |
eb73c1b7 MX |
9642 | } |
9643 | ||
9db4dc24 | 9644 | int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, long nr, |
3d45f221 | 9645 | bool in_reclaim_context) |
eb73c1b7 | 9646 | { |
e076ab2a | 9647 | struct writeback_control wbc = { |
9db4dc24 | 9648 | .nr_to_write = nr, |
e076ab2a JB |
9649 | .sync_mode = WB_SYNC_NONE, |
9650 | .range_start = 0, | |
9651 | .range_end = LLONG_MAX, | |
9652 | }; | |
eb73c1b7 MX |
9653 | struct btrfs_root *root; |
9654 | struct list_head splice; | |
9655 | int ret; | |
9656 | ||
84961539 | 9657 | if (BTRFS_FS_ERROR(fs_info)) |
eb73c1b7 MX |
9658 | return -EROFS; |
9659 | ||
9660 | INIT_LIST_HEAD(&splice); | |
9661 | ||
573bfb72 | 9662 | mutex_lock(&fs_info->delalloc_root_mutex); |
eb73c1b7 MX |
9663 | spin_lock(&fs_info->delalloc_root_lock); |
9664 | list_splice_init(&fs_info->delalloc_roots, &splice); | |
d7830b71 | 9665 | while (!list_empty(&splice)) { |
e076ab2a JB |
9666 | /* |
9667 | * Reset nr_to_write here so we know that we're doing a full | |
9668 | * flush. | |
9669 | */ | |
9db4dc24 | 9670 | if (nr == LONG_MAX) |
e076ab2a JB |
9671 | wbc.nr_to_write = LONG_MAX; |
9672 | ||
eb73c1b7 MX |
9673 | root = list_first_entry(&splice, struct btrfs_root, |
9674 | delalloc_root); | |
00246528 | 9675 | root = btrfs_grab_root(root); |
eb73c1b7 MX |
9676 | BUG_ON(!root); |
9677 | list_move_tail(&root->delalloc_root, | |
9678 | &fs_info->delalloc_roots); | |
9679 | spin_unlock(&fs_info->delalloc_root_lock); | |
9680 | ||
e076ab2a | 9681 | ret = start_delalloc_inodes(root, &wbc, false, in_reclaim_context); |
00246528 | 9682 | btrfs_put_root(root); |
e076ab2a | 9683 | if (ret < 0 || wbc.nr_to_write <= 0) |
eb73c1b7 | 9684 | goto out; |
eb73c1b7 | 9685 | spin_lock(&fs_info->delalloc_root_lock); |
8ccf6f19 | 9686 | } |
eb73c1b7 | 9687 | spin_unlock(&fs_info->delalloc_root_lock); |
1eafa6c7 | 9688 | |
6c255e67 | 9689 | ret = 0; |
eb73c1b7 | 9690 | out: |
81f1d390 | 9691 | if (!list_empty(&splice)) { |
eb73c1b7 MX |
9692 | spin_lock(&fs_info->delalloc_root_lock); |
9693 | list_splice_tail(&splice, &fs_info->delalloc_roots); | |
9694 | spin_unlock(&fs_info->delalloc_root_lock); | |
1eafa6c7 | 9695 | } |
573bfb72 | 9696 | mutex_unlock(&fs_info->delalloc_root_mutex); |
8ccf6f19 | 9697 | return ret; |
ea8c2819 CM |
9698 | } |
9699 | ||
549c7297 CB |
9700 | static int btrfs_symlink(struct user_namespace *mnt_userns, struct inode *dir, |
9701 | struct dentry *dentry, const char *symname) | |
39279cc3 | 9702 | { |
0b246afa | 9703 | struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb); |
39279cc3 CM |
9704 | struct btrfs_trans_handle *trans; |
9705 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
9706 | struct btrfs_path *path; | |
9707 | struct btrfs_key key; | |
1832a6d5 | 9708 | struct inode *inode = NULL; |
39279cc3 | 9709 | int err; |
67871254 | 9710 | u64 index = 0; |
39279cc3 CM |
9711 | int name_len; |
9712 | int datasize; | |
5f39d397 | 9713 | unsigned long ptr; |
39279cc3 | 9714 | struct btrfs_file_extent_item *ei; |
5f39d397 | 9715 | struct extent_buffer *leaf; |
39279cc3 | 9716 | |
f06becc4 | 9717 | name_len = strlen(symname); |
0b246afa | 9718 | if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info)) |
39279cc3 | 9719 | return -ENAMETOOLONG; |
1832a6d5 | 9720 | |
9ed74f2d JB |
9721 | /* |
9722 | * 2 items for inode item and ref | |
9723 | * 2 items for dir items | |
9269d12b FM |
9724 | * 1 item for updating parent inode item |
9725 | * 1 item for the inline extent item | |
9ed74f2d JB |
9726 | * 1 item for xattr if selinux is on |
9727 | */ | |
9269d12b | 9728 | trans = btrfs_start_transaction(root, 7); |
a22285a6 YZ |
9729 | if (IS_ERR(trans)) |
9730 | return PTR_ERR(trans); | |
1832a6d5 | 9731 | |
5a052108 | 9732 | inode = btrfs_new_inode(trans, root, mnt_userns, dir, |
b3b6f5b9 | 9733 | dentry->d_name.name, dentry->d_name.len, |
b3b6f5b9 | 9734 | S_IFLNK | S_IRWXUGO, &index); |
7cf96da3 TI |
9735 | if (IS_ERR(inode)) { |
9736 | err = PTR_ERR(inode); | |
32955c54 | 9737 | inode = NULL; |
39279cc3 | 9738 | goto out_unlock; |
7cf96da3 | 9739 | } |
39279cc3 | 9740 | |
ad19db71 CS |
9741 | /* |
9742 | * If the active LSM wants to access the inode during | |
9743 | * d_instantiate it needs these. Smack checks to see | |
9744 | * if the filesystem supports xattrs by looking at the | |
9745 | * ops vector. | |
9746 | */ | |
9747 | inode->i_fop = &btrfs_file_operations; | |
9748 | inode->i_op = &btrfs_file_inode_operations; | |
b0d5d10f | 9749 | inode->i_mapping->a_ops = &btrfs_aops; |
b0d5d10f CM |
9750 | |
9751 | err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name); | |
9752 | if (err) | |
32955c54 | 9753 | goto out_unlock; |
ad19db71 | 9754 | |
39279cc3 | 9755 | path = btrfs_alloc_path(); |
d8926bb3 MF |
9756 | if (!path) { |
9757 | err = -ENOMEM; | |
32955c54 | 9758 | goto out_unlock; |
d8926bb3 | 9759 | } |
4a0cc7ca | 9760 | key.objectid = btrfs_ino(BTRFS_I(inode)); |
39279cc3 | 9761 | key.offset = 0; |
962a298f | 9762 | key.type = BTRFS_EXTENT_DATA_KEY; |
39279cc3 CM |
9763 | datasize = btrfs_file_extent_calc_inline_size(name_len); |
9764 | err = btrfs_insert_empty_item(trans, root, path, &key, | |
9765 | datasize); | |
54aa1f4d | 9766 | if (err) { |
b0839166 | 9767 | btrfs_free_path(path); |
32955c54 | 9768 | goto out_unlock; |
54aa1f4d | 9769 | } |
5f39d397 CM |
9770 | leaf = path->nodes[0]; |
9771 | ei = btrfs_item_ptr(leaf, path->slots[0], | |
9772 | struct btrfs_file_extent_item); | |
9773 | btrfs_set_file_extent_generation(leaf, ei, trans->transid); | |
9774 | btrfs_set_file_extent_type(leaf, ei, | |
39279cc3 | 9775 | BTRFS_FILE_EXTENT_INLINE); |
c8b97818 CM |
9776 | btrfs_set_file_extent_encryption(leaf, ei, 0); |
9777 | btrfs_set_file_extent_compression(leaf, ei, 0); | |
9778 | btrfs_set_file_extent_other_encoding(leaf, ei, 0); | |
9779 | btrfs_set_file_extent_ram_bytes(leaf, ei, name_len); | |
9780 | ||
39279cc3 | 9781 | ptr = btrfs_file_extent_inline_start(ei); |
5f39d397 CM |
9782 | write_extent_buffer(leaf, symname, ptr, name_len); |
9783 | btrfs_mark_buffer_dirty(leaf); | |
39279cc3 | 9784 | btrfs_free_path(path); |
5f39d397 | 9785 | |
39279cc3 | 9786 | inode->i_op = &btrfs_symlink_inode_operations; |
21fc61c7 | 9787 | inode_nohighmem(inode); |
d899e052 | 9788 | inode_set_bytes(inode, name_len); |
6ef06d27 | 9789 | btrfs_i_size_write(BTRFS_I(inode), name_len); |
9a56fcd1 | 9790 | err = btrfs_update_inode(trans, root, BTRFS_I(inode)); |
d50866d0 FM |
9791 | /* |
9792 | * Last step, add directory indexes for our symlink inode. This is the | |
9793 | * last step to avoid extra cleanup of these indexes if an error happens | |
9794 | * elsewhere above. | |
9795 | */ | |
9796 | if (!err) | |
81512e89 OS |
9797 | err = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode), |
9798 | dentry->d_name.name, dentry->d_name.len, 0, | |
9799 | index); | |
32955c54 AV |
9800 | if (err) |
9801 | goto out_unlock; | |
b0d5d10f | 9802 | |
1e2e547a | 9803 | d_instantiate_new(dentry, inode); |
39279cc3 CM |
9804 | |
9805 | out_unlock: | |
3a45bb20 | 9806 | btrfs_end_transaction(trans); |
32955c54 | 9807 | if (err && inode) { |
39279cc3 | 9808 | inode_dec_link_count(inode); |
32955c54 | 9809 | discard_new_inode(inode); |
39279cc3 | 9810 | } |
2ff7e61e | 9811 | btrfs_btree_balance_dirty(fs_info); |
39279cc3 CM |
9812 | return err; |
9813 | } | |
16432985 | 9814 | |
8fccebfa FM |
9815 | static struct btrfs_trans_handle *insert_prealloc_file_extent( |
9816 | struct btrfs_trans_handle *trans_in, | |
90dffd0c NB |
9817 | struct btrfs_inode *inode, |
9818 | struct btrfs_key *ins, | |
203f44c5 QW |
9819 | u64 file_offset) |
9820 | { | |
9821 | struct btrfs_file_extent_item stack_fi; | |
bf385648 | 9822 | struct btrfs_replace_extent_info extent_info; |
8fccebfa FM |
9823 | struct btrfs_trans_handle *trans = trans_in; |
9824 | struct btrfs_path *path; | |
203f44c5 QW |
9825 | u64 start = ins->objectid; |
9826 | u64 len = ins->offset; | |
fbf48bb0 | 9827 | int qgroup_released; |
9729f10a | 9828 | int ret; |
203f44c5 QW |
9829 | |
9830 | memset(&stack_fi, 0, sizeof(stack_fi)); | |
9831 | ||
9832 | btrfs_set_stack_file_extent_type(&stack_fi, BTRFS_FILE_EXTENT_PREALLOC); | |
9833 | btrfs_set_stack_file_extent_disk_bytenr(&stack_fi, start); | |
9834 | btrfs_set_stack_file_extent_disk_num_bytes(&stack_fi, len); | |
9835 | btrfs_set_stack_file_extent_num_bytes(&stack_fi, len); | |
9836 | btrfs_set_stack_file_extent_ram_bytes(&stack_fi, len); | |
9837 | btrfs_set_stack_file_extent_compression(&stack_fi, BTRFS_COMPRESS_NONE); | |
9838 | /* Encryption and other encoding is reserved and all 0 */ | |
9839 | ||
fbf48bb0 QW |
9840 | qgroup_released = btrfs_qgroup_release_data(inode, file_offset, len); |
9841 | if (qgroup_released < 0) | |
9842 | return ERR_PTR(qgroup_released); | |
8fccebfa FM |
9843 | |
9844 | if (trans) { | |
90dffd0c | 9845 | ret = insert_reserved_file_extent(trans, inode, |
2766ff61 | 9846 | file_offset, &stack_fi, |
fbf48bb0 | 9847 | true, qgroup_released); |
8fccebfa | 9848 | if (ret) |
a3ee79bd | 9849 | goto free_qgroup; |
8fccebfa FM |
9850 | return trans; |
9851 | } | |
9852 | ||
9853 | extent_info.disk_offset = start; | |
9854 | extent_info.disk_len = len; | |
9855 | extent_info.data_offset = 0; | |
9856 | extent_info.data_len = len; | |
9857 | extent_info.file_offset = file_offset; | |
9858 | extent_info.extent_buf = (char *)&stack_fi; | |
8fccebfa | 9859 | extent_info.is_new_extent = true; |
fbf48bb0 | 9860 | extent_info.qgroup_reserved = qgroup_released; |
8fccebfa FM |
9861 | extent_info.insertions = 0; |
9862 | ||
9863 | path = btrfs_alloc_path(); | |
a3ee79bd QW |
9864 | if (!path) { |
9865 | ret = -ENOMEM; | |
9866 | goto free_qgroup; | |
9867 | } | |
8fccebfa | 9868 | |
bfc78479 | 9869 | ret = btrfs_replace_file_extents(inode, path, file_offset, |
8fccebfa FM |
9870 | file_offset + len - 1, &extent_info, |
9871 | &trans); | |
9872 | btrfs_free_path(path); | |
9873 | if (ret) | |
a3ee79bd | 9874 | goto free_qgroup; |
8fccebfa | 9875 | return trans; |
a3ee79bd QW |
9876 | |
9877 | free_qgroup: | |
9878 | /* | |
9879 | * We have released qgroup data range at the beginning of the function, | |
9880 | * and normally qgroup_released bytes will be freed when committing | |
9881 | * transaction. | |
9882 | * But if we error out early, we have to free what we have released | |
9883 | * or we leak qgroup data reservation. | |
9884 | */ | |
9885 | btrfs_qgroup_free_refroot(inode->root->fs_info, | |
9886 | inode->root->root_key.objectid, qgroup_released, | |
9887 | BTRFS_QGROUP_RSV_DATA); | |
9888 | return ERR_PTR(ret); | |
203f44c5 | 9889 | } |
8fccebfa | 9890 | |
0af3d00b JB |
9891 | static int __btrfs_prealloc_file_range(struct inode *inode, int mode, |
9892 | u64 start, u64 num_bytes, u64 min_size, | |
9893 | loff_t actual_len, u64 *alloc_hint, | |
9894 | struct btrfs_trans_handle *trans) | |
d899e052 | 9895 | { |
0b246afa | 9896 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
5dc562c5 JB |
9897 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; |
9898 | struct extent_map *em; | |
d899e052 YZ |
9899 | struct btrfs_root *root = BTRFS_I(inode)->root; |
9900 | struct btrfs_key ins; | |
d899e052 | 9901 | u64 cur_offset = start; |
b778cf96 | 9902 | u64 clear_offset = start; |
55a61d1d | 9903 | u64 i_size; |
154ea289 | 9904 | u64 cur_bytes; |
0b670dc4 | 9905 | u64 last_alloc = (u64)-1; |
d899e052 | 9906 | int ret = 0; |
0af3d00b | 9907 | bool own_trans = true; |
18513091 | 9908 | u64 end = start + num_bytes - 1; |
d899e052 | 9909 | |
0af3d00b JB |
9910 | if (trans) |
9911 | own_trans = false; | |
d899e052 | 9912 | while (num_bytes > 0) { |
ee22184b | 9913 | cur_bytes = min_t(u64, num_bytes, SZ_256M); |
154ea289 | 9914 | cur_bytes = max(cur_bytes, min_size); |
0b670dc4 JB |
9915 | /* |
9916 | * If we are severely fragmented we could end up with really | |
9917 | * small allocations, so if the allocator is returning small | |
9918 | * chunks lets make its job easier by only searching for those | |
9919 | * sized chunks. | |
9920 | */ | |
9921 | cur_bytes = min(cur_bytes, last_alloc); | |
18513091 WX |
9922 | ret = btrfs_reserve_extent(root, cur_bytes, cur_bytes, |
9923 | min_size, 0, *alloc_hint, &ins, 1, 0); | |
8fccebfa | 9924 | if (ret) |
a22285a6 | 9925 | break; |
b778cf96 JB |
9926 | |
9927 | /* | |
9928 | * We've reserved this space, and thus converted it from | |
9929 | * ->bytes_may_use to ->bytes_reserved. Any error that happens | |
9930 | * from here on out we will only need to clear our reservation | |
9931 | * for the remaining unreserved area, so advance our | |
9932 | * clear_offset by our extent size. | |
9933 | */ | |
9934 | clear_offset += ins.offset; | |
5a303d5d | 9935 | |
0b670dc4 | 9936 | last_alloc = ins.offset; |
90dffd0c NB |
9937 | trans = insert_prealloc_file_extent(trans, BTRFS_I(inode), |
9938 | &ins, cur_offset); | |
1afc708d FM |
9939 | /* |
9940 | * Now that we inserted the prealloc extent we can finally | |
9941 | * decrement the number of reservations in the block group. | |
9942 | * If we did it before, we could race with relocation and have | |
9943 | * relocation miss the reserved extent, making it fail later. | |
9944 | */ | |
9945 | btrfs_dec_block_group_reservations(fs_info, ins.objectid); | |
8fccebfa FM |
9946 | if (IS_ERR(trans)) { |
9947 | ret = PTR_ERR(trans); | |
2ff7e61e | 9948 | btrfs_free_reserved_extent(fs_info, ins.objectid, |
e570fd27 | 9949 | ins.offset, 0); |
79787eaa JM |
9950 | break; |
9951 | } | |
31193213 | 9952 | |
dcdbc059 | 9953 | btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset, |
a1ed835e | 9954 | cur_offset + ins.offset -1, 0); |
5a303d5d | 9955 | |
5dc562c5 JB |
9956 | em = alloc_extent_map(); |
9957 | if (!em) { | |
23e3337f | 9958 | btrfs_set_inode_full_sync(BTRFS_I(inode)); |
5dc562c5 JB |
9959 | goto next; |
9960 | } | |
9961 | ||
9962 | em->start = cur_offset; | |
9963 | em->orig_start = cur_offset; | |
9964 | em->len = ins.offset; | |
9965 | em->block_start = ins.objectid; | |
9966 | em->block_len = ins.offset; | |
b4939680 | 9967 | em->orig_block_len = ins.offset; |
cc95bef6 | 9968 | em->ram_bytes = ins.offset; |
5dc562c5 JB |
9969 | set_bit(EXTENT_FLAG_PREALLOC, &em->flags); |
9970 | em->generation = trans->transid; | |
9971 | ||
9972 | while (1) { | |
9973 | write_lock(&em_tree->lock); | |
09a2a8f9 | 9974 | ret = add_extent_mapping(em_tree, em, 1); |
5dc562c5 JB |
9975 | write_unlock(&em_tree->lock); |
9976 | if (ret != -EEXIST) | |
9977 | break; | |
dcdbc059 | 9978 | btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset, |
5dc562c5 JB |
9979 | cur_offset + ins.offset - 1, |
9980 | 0); | |
9981 | } | |
9982 | free_extent_map(em); | |
9983 | next: | |
d899e052 YZ |
9984 | num_bytes -= ins.offset; |
9985 | cur_offset += ins.offset; | |
efa56464 | 9986 | *alloc_hint = ins.objectid + ins.offset; |
5a303d5d | 9987 | |
0c4d2d95 | 9988 | inode_inc_iversion(inode); |
c2050a45 | 9989 | inode->i_ctime = current_time(inode); |
6cbff00f | 9990 | BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC; |
d899e052 | 9991 | if (!(mode & FALLOC_FL_KEEP_SIZE) && |
efa56464 YZ |
9992 | (actual_len > inode->i_size) && |
9993 | (cur_offset > inode->i_size)) { | |
d1ea6a61 | 9994 | if (cur_offset > actual_len) |
55a61d1d | 9995 | i_size = actual_len; |
d1ea6a61 | 9996 | else |
55a61d1d JB |
9997 | i_size = cur_offset; |
9998 | i_size_write(inode, i_size); | |
76aea537 | 9999 | btrfs_inode_safe_disk_i_size_write(BTRFS_I(inode), 0); |
5a303d5d YZ |
10000 | } |
10001 | ||
9a56fcd1 | 10002 | ret = btrfs_update_inode(trans, root, BTRFS_I(inode)); |
79787eaa JM |
10003 | |
10004 | if (ret) { | |
66642832 | 10005 | btrfs_abort_transaction(trans, ret); |
79787eaa | 10006 | if (own_trans) |
3a45bb20 | 10007 | btrfs_end_transaction(trans); |
79787eaa JM |
10008 | break; |
10009 | } | |
d899e052 | 10010 | |
8fccebfa | 10011 | if (own_trans) { |
3a45bb20 | 10012 | btrfs_end_transaction(trans); |
8fccebfa FM |
10013 | trans = NULL; |
10014 | } | |
5a303d5d | 10015 | } |
b778cf96 | 10016 | if (clear_offset < end) |
25ce28ca | 10017 | btrfs_free_reserved_data_space(BTRFS_I(inode), NULL, clear_offset, |
b778cf96 | 10018 | end - clear_offset + 1); |
d899e052 YZ |
10019 | return ret; |
10020 | } | |
10021 | ||
0af3d00b JB |
10022 | int btrfs_prealloc_file_range(struct inode *inode, int mode, |
10023 | u64 start, u64 num_bytes, u64 min_size, | |
10024 | loff_t actual_len, u64 *alloc_hint) | |
10025 | { | |
10026 | return __btrfs_prealloc_file_range(inode, mode, start, num_bytes, | |
10027 | min_size, actual_len, alloc_hint, | |
10028 | NULL); | |
10029 | } | |
10030 | ||
10031 | int btrfs_prealloc_file_range_trans(struct inode *inode, | |
10032 | struct btrfs_trans_handle *trans, int mode, | |
10033 | u64 start, u64 num_bytes, u64 min_size, | |
10034 | loff_t actual_len, u64 *alloc_hint) | |
10035 | { | |
10036 | return __btrfs_prealloc_file_range(inode, mode, start, num_bytes, | |
10037 | min_size, actual_len, alloc_hint, trans); | |
10038 | } | |
10039 | ||
549c7297 CB |
10040 | static int btrfs_permission(struct user_namespace *mnt_userns, |
10041 | struct inode *inode, int mask) | |
fdebe2bd | 10042 | { |
b83cc969 | 10043 | struct btrfs_root *root = BTRFS_I(inode)->root; |
cb6db4e5 | 10044 | umode_t mode = inode->i_mode; |
b83cc969 | 10045 | |
cb6db4e5 JM |
10046 | if (mask & MAY_WRITE && |
10047 | (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) { | |
10048 | if (btrfs_root_readonly(root)) | |
10049 | return -EROFS; | |
10050 | if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY) | |
10051 | return -EACCES; | |
10052 | } | |
3bc71ba0 | 10053 | return generic_permission(mnt_userns, inode, mask); |
fdebe2bd | 10054 | } |
39279cc3 | 10055 | |
549c7297 CB |
10056 | static int btrfs_tmpfile(struct user_namespace *mnt_userns, struct inode *dir, |
10057 | struct dentry *dentry, umode_t mode) | |
ef3b9af5 | 10058 | { |
2ff7e61e | 10059 | struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb); |
ef3b9af5 FM |
10060 | struct btrfs_trans_handle *trans; |
10061 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
10062 | struct inode *inode = NULL; | |
ef3b9af5 FM |
10063 | u64 index; |
10064 | int ret = 0; | |
10065 | ||
10066 | /* | |
10067 | * 5 units required for adding orphan entry | |
10068 | */ | |
10069 | trans = btrfs_start_transaction(root, 5); | |
10070 | if (IS_ERR(trans)) | |
10071 | return PTR_ERR(trans); | |
10072 | ||
98b6ab5f | 10073 | inode = btrfs_new_inode(trans, root, mnt_userns, dir, NULL, 0, |
6437d458 | 10074 | mode, &index); |
ef3b9af5 FM |
10075 | if (IS_ERR(inode)) { |
10076 | ret = PTR_ERR(inode); | |
10077 | inode = NULL; | |
10078 | goto out; | |
10079 | } | |
10080 | ||
ef3b9af5 FM |
10081 | inode->i_fop = &btrfs_file_operations; |
10082 | inode->i_op = &btrfs_file_inode_operations; | |
10083 | ||
10084 | inode->i_mapping->a_ops = &btrfs_aops; | |
ef3b9af5 | 10085 | |
b0d5d10f CM |
10086 | ret = btrfs_init_inode_security(trans, inode, dir, NULL); |
10087 | if (ret) | |
32955c54 | 10088 | goto out; |
b0d5d10f | 10089 | |
9a56fcd1 | 10090 | ret = btrfs_update_inode(trans, root, BTRFS_I(inode)); |
b0d5d10f | 10091 | if (ret) |
32955c54 | 10092 | goto out; |
73f2e545 | 10093 | ret = btrfs_orphan_add(trans, BTRFS_I(inode)); |
ef3b9af5 | 10094 | if (ret) |
32955c54 | 10095 | goto out; |
ef3b9af5 | 10096 | |
5762b5c9 FM |
10097 | /* |
10098 | * We set number of links to 0 in btrfs_new_inode(), and here we set | |
10099 | * it to 1 because d_tmpfile() will issue a warning if the count is 0, | |
10100 | * through: | |
10101 | * | |
10102 | * d_tmpfile() -> inode_dec_link_count() -> drop_nlink() | |
10103 | */ | |
10104 | set_nlink(inode, 1); | |
ef3b9af5 | 10105 | d_tmpfile(dentry, inode); |
32955c54 | 10106 | unlock_new_inode(inode); |
ef3b9af5 | 10107 | mark_inode_dirty(inode); |
ef3b9af5 | 10108 | out: |
3a45bb20 | 10109 | btrfs_end_transaction(trans); |
32955c54 AV |
10110 | if (ret && inode) |
10111 | discard_new_inode(inode); | |
2ff7e61e | 10112 | btrfs_btree_balance_dirty(fs_info); |
ef3b9af5 FM |
10113 | return ret; |
10114 | } | |
10115 | ||
d2a91064 | 10116 | void btrfs_set_range_writeback(struct btrfs_inode *inode, u64 start, u64 end) |
c6100a4b | 10117 | { |
d2a91064 | 10118 | struct btrfs_fs_info *fs_info = inode->root->fs_info; |
c6100a4b JB |
10119 | unsigned long index = start >> PAGE_SHIFT; |
10120 | unsigned long end_index = end >> PAGE_SHIFT; | |
10121 | struct page *page; | |
d2a91064 | 10122 | u32 len; |
c6100a4b | 10123 | |
d2a91064 QW |
10124 | ASSERT(end + 1 - start <= U32_MAX); |
10125 | len = end + 1 - start; | |
c6100a4b | 10126 | while (index <= end_index) { |
d2a91064 | 10127 | page = find_get_page(inode->vfs_inode.i_mapping, index); |
c6100a4b | 10128 | ASSERT(page); /* Pages should be in the extent_io_tree */ |
d2a91064 QW |
10129 | |
10130 | btrfs_page_set_writeback(fs_info, page, start, len); | |
c6100a4b JB |
10131 | put_page(page); |
10132 | index++; | |
10133 | } | |
10134 | } | |
10135 | ||
1881fba8 OS |
10136 | static int btrfs_encoded_io_compression_from_extent( |
10137 | struct btrfs_fs_info *fs_info, | |
10138 | int compress_type) | |
10139 | { | |
10140 | switch (compress_type) { | |
10141 | case BTRFS_COMPRESS_NONE: | |
10142 | return BTRFS_ENCODED_IO_COMPRESSION_NONE; | |
10143 | case BTRFS_COMPRESS_ZLIB: | |
10144 | return BTRFS_ENCODED_IO_COMPRESSION_ZLIB; | |
10145 | case BTRFS_COMPRESS_LZO: | |
10146 | /* | |
10147 | * The LZO format depends on the sector size. 64K is the maximum | |
10148 | * sector size that we support. | |
10149 | */ | |
10150 | if (fs_info->sectorsize < SZ_4K || fs_info->sectorsize > SZ_64K) | |
10151 | return -EINVAL; | |
10152 | return BTRFS_ENCODED_IO_COMPRESSION_LZO_4K + | |
10153 | (fs_info->sectorsize_bits - 12); | |
10154 | case BTRFS_COMPRESS_ZSTD: | |
10155 | return BTRFS_ENCODED_IO_COMPRESSION_ZSTD; | |
10156 | default: | |
10157 | return -EUCLEAN; | |
10158 | } | |
10159 | } | |
10160 | ||
10161 | static ssize_t btrfs_encoded_read_inline( | |
10162 | struct kiocb *iocb, | |
10163 | struct iov_iter *iter, u64 start, | |
10164 | u64 lockend, | |
10165 | struct extent_state **cached_state, | |
10166 | u64 extent_start, size_t count, | |
10167 | struct btrfs_ioctl_encoded_io_args *encoded, | |
10168 | bool *unlocked) | |
10169 | { | |
10170 | struct btrfs_inode *inode = BTRFS_I(file_inode(iocb->ki_filp)); | |
10171 | struct btrfs_root *root = inode->root; | |
10172 | struct btrfs_fs_info *fs_info = root->fs_info; | |
10173 | struct extent_io_tree *io_tree = &inode->io_tree; | |
10174 | struct btrfs_path *path; | |
10175 | struct extent_buffer *leaf; | |
10176 | struct btrfs_file_extent_item *item; | |
10177 | u64 ram_bytes; | |
10178 | unsigned long ptr; | |
10179 | void *tmp; | |
10180 | ssize_t ret; | |
10181 | ||
10182 | path = btrfs_alloc_path(); | |
10183 | if (!path) { | |
10184 | ret = -ENOMEM; | |
10185 | goto out; | |
10186 | } | |
10187 | ret = btrfs_lookup_file_extent(NULL, root, path, btrfs_ino(inode), | |
10188 | extent_start, 0); | |
10189 | if (ret) { | |
10190 | if (ret > 0) { | |
10191 | /* The extent item disappeared? */ | |
10192 | ret = -EIO; | |
10193 | } | |
10194 | goto out; | |
10195 | } | |
10196 | leaf = path->nodes[0]; | |
10197 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_file_extent_item); | |
10198 | ||
10199 | ram_bytes = btrfs_file_extent_ram_bytes(leaf, item); | |
10200 | ptr = btrfs_file_extent_inline_start(item); | |
10201 | ||
10202 | encoded->len = min_t(u64, extent_start + ram_bytes, | |
10203 | inode->vfs_inode.i_size) - iocb->ki_pos; | |
10204 | ret = btrfs_encoded_io_compression_from_extent(fs_info, | |
10205 | btrfs_file_extent_compression(leaf, item)); | |
10206 | if (ret < 0) | |
10207 | goto out; | |
10208 | encoded->compression = ret; | |
10209 | if (encoded->compression) { | |
10210 | size_t inline_size; | |
10211 | ||
10212 | inline_size = btrfs_file_extent_inline_item_len(leaf, | |
10213 | path->slots[0]); | |
10214 | if (inline_size > count) { | |
10215 | ret = -ENOBUFS; | |
10216 | goto out; | |
10217 | } | |
10218 | count = inline_size; | |
10219 | encoded->unencoded_len = ram_bytes; | |
10220 | encoded->unencoded_offset = iocb->ki_pos - extent_start; | |
10221 | } else { | |
10222 | count = min_t(u64, count, encoded->len); | |
10223 | encoded->len = count; | |
10224 | encoded->unencoded_len = count; | |
10225 | ptr += iocb->ki_pos - extent_start; | |
10226 | } | |
10227 | ||
10228 | tmp = kmalloc(count, GFP_NOFS); | |
10229 | if (!tmp) { | |
10230 | ret = -ENOMEM; | |
10231 | goto out; | |
10232 | } | |
10233 | read_extent_buffer(leaf, tmp, ptr, count); | |
10234 | btrfs_release_path(path); | |
10235 | unlock_extent_cached(io_tree, start, lockend, cached_state); | |
10236 | btrfs_inode_unlock(&inode->vfs_inode, BTRFS_ILOCK_SHARED); | |
10237 | *unlocked = true; | |
10238 | ||
10239 | ret = copy_to_iter(tmp, count, iter); | |
10240 | if (ret != count) | |
10241 | ret = -EFAULT; | |
10242 | kfree(tmp); | |
10243 | out: | |
10244 | btrfs_free_path(path); | |
10245 | return ret; | |
10246 | } | |
10247 | ||
10248 | struct btrfs_encoded_read_private { | |
10249 | struct btrfs_inode *inode; | |
10250 | u64 file_offset; | |
10251 | wait_queue_head_t wait; | |
10252 | atomic_t pending; | |
10253 | blk_status_t status; | |
10254 | bool skip_csum; | |
10255 | }; | |
10256 | ||
10257 | static blk_status_t submit_encoded_read_bio(struct btrfs_inode *inode, | |
10258 | struct bio *bio, int mirror_num) | |
10259 | { | |
10260 | struct btrfs_encoded_read_private *priv = bio->bi_private; | |
10261 | struct btrfs_bio *bbio = btrfs_bio(bio); | |
10262 | struct btrfs_fs_info *fs_info = inode->root->fs_info; | |
10263 | blk_status_t ret; | |
10264 | ||
10265 | if (!priv->skip_csum) { | |
10266 | ret = btrfs_lookup_bio_sums(&inode->vfs_inode, bio, NULL); | |
10267 | if (ret) | |
10268 | return ret; | |
10269 | } | |
10270 | ||
10271 | ret = btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DATA); | |
10272 | if (ret) { | |
10273 | btrfs_bio_free_csum(bbio); | |
10274 | return ret; | |
10275 | } | |
10276 | ||
10277 | atomic_inc(&priv->pending); | |
10278 | ret = btrfs_map_bio(fs_info, bio, mirror_num); | |
10279 | if (ret) { | |
10280 | atomic_dec(&priv->pending); | |
10281 | btrfs_bio_free_csum(bbio); | |
10282 | } | |
10283 | return ret; | |
10284 | } | |
10285 | ||
10286 | static blk_status_t btrfs_encoded_read_verify_csum(struct btrfs_bio *bbio) | |
10287 | { | |
10288 | const bool uptodate = (bbio->bio.bi_status == BLK_STS_OK); | |
10289 | struct btrfs_encoded_read_private *priv = bbio->bio.bi_private; | |
10290 | struct btrfs_inode *inode = priv->inode; | |
10291 | struct btrfs_fs_info *fs_info = inode->root->fs_info; | |
10292 | u32 sectorsize = fs_info->sectorsize; | |
10293 | struct bio_vec *bvec; | |
10294 | struct bvec_iter_all iter_all; | |
10295 | u64 start = priv->file_offset; | |
10296 | u32 bio_offset = 0; | |
10297 | ||
10298 | if (priv->skip_csum || !uptodate) | |
10299 | return bbio->bio.bi_status; | |
10300 | ||
10301 | bio_for_each_segment_all(bvec, &bbio->bio, iter_all) { | |
10302 | unsigned int i, nr_sectors, pgoff; | |
10303 | ||
10304 | nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec->bv_len); | |
10305 | pgoff = bvec->bv_offset; | |
10306 | for (i = 0; i < nr_sectors; i++) { | |
10307 | ASSERT(pgoff < PAGE_SIZE); | |
10308 | if (check_data_csum(&inode->vfs_inode, bbio, bio_offset, | |
10309 | bvec->bv_page, pgoff, start)) | |
10310 | return BLK_STS_IOERR; | |
10311 | start += sectorsize; | |
10312 | bio_offset += sectorsize; | |
10313 | pgoff += sectorsize; | |
10314 | } | |
10315 | } | |
10316 | return BLK_STS_OK; | |
10317 | } | |
10318 | ||
10319 | static void btrfs_encoded_read_endio(struct bio *bio) | |
10320 | { | |
10321 | struct btrfs_encoded_read_private *priv = bio->bi_private; | |
10322 | struct btrfs_bio *bbio = btrfs_bio(bio); | |
10323 | blk_status_t status; | |
10324 | ||
10325 | status = btrfs_encoded_read_verify_csum(bbio); | |
10326 | if (status) { | |
10327 | /* | |
10328 | * The memory barrier implied by the atomic_dec_return() here | |
10329 | * pairs with the memory barrier implied by the | |
10330 | * atomic_dec_return() or io_wait_event() in | |
10331 | * btrfs_encoded_read_regular_fill_pages() to ensure that this | |
10332 | * write is observed before the load of status in | |
10333 | * btrfs_encoded_read_regular_fill_pages(). | |
10334 | */ | |
10335 | WRITE_ONCE(priv->status, status); | |
10336 | } | |
10337 | if (!atomic_dec_return(&priv->pending)) | |
10338 | wake_up(&priv->wait); | |
10339 | btrfs_bio_free_csum(bbio); | |
10340 | bio_put(bio); | |
10341 | } | |
10342 | ||
10343 | static int btrfs_encoded_read_regular_fill_pages(struct btrfs_inode *inode, | |
10344 | u64 file_offset, | |
10345 | u64 disk_bytenr, | |
10346 | u64 disk_io_size, | |
10347 | struct page **pages) | |
10348 | { | |
10349 | struct btrfs_fs_info *fs_info = inode->root->fs_info; | |
10350 | struct btrfs_encoded_read_private priv = { | |
10351 | .inode = inode, | |
10352 | .file_offset = file_offset, | |
10353 | .pending = ATOMIC_INIT(1), | |
10354 | .skip_csum = (inode->flags & BTRFS_INODE_NODATASUM), | |
10355 | }; | |
10356 | unsigned long i = 0; | |
10357 | u64 cur = 0; | |
10358 | int ret; | |
10359 | ||
10360 | init_waitqueue_head(&priv.wait); | |
10361 | /* | |
10362 | * Submit bios for the extent, splitting due to bio or stripe limits as | |
10363 | * necessary. | |
10364 | */ | |
10365 | while (cur < disk_io_size) { | |
10366 | struct extent_map *em; | |
10367 | struct btrfs_io_geometry geom; | |
10368 | struct bio *bio = NULL; | |
10369 | u64 remaining; | |
10370 | ||
10371 | em = btrfs_get_chunk_map(fs_info, disk_bytenr + cur, | |
10372 | disk_io_size - cur); | |
10373 | if (IS_ERR(em)) { | |
10374 | ret = PTR_ERR(em); | |
10375 | } else { | |
10376 | ret = btrfs_get_io_geometry(fs_info, em, BTRFS_MAP_READ, | |
10377 | disk_bytenr + cur, &geom); | |
10378 | free_extent_map(em); | |
10379 | } | |
10380 | if (ret) { | |
10381 | WRITE_ONCE(priv.status, errno_to_blk_status(ret)); | |
10382 | break; | |
10383 | } | |
10384 | remaining = min(geom.len, disk_io_size - cur); | |
10385 | while (bio || remaining) { | |
10386 | size_t bytes = min_t(u64, remaining, PAGE_SIZE); | |
10387 | ||
10388 | if (!bio) { | |
10389 | bio = btrfs_bio_alloc(BIO_MAX_VECS); | |
10390 | bio->bi_iter.bi_sector = | |
10391 | (disk_bytenr + cur) >> SECTOR_SHIFT; | |
10392 | bio->bi_end_io = btrfs_encoded_read_endio; | |
10393 | bio->bi_private = &priv; | |
10394 | bio->bi_opf = REQ_OP_READ; | |
10395 | } | |
10396 | ||
10397 | if (!bytes || | |
10398 | bio_add_page(bio, pages[i], bytes, 0) < bytes) { | |
10399 | blk_status_t status; | |
10400 | ||
10401 | status = submit_encoded_read_bio(inode, bio, 0); | |
10402 | if (status) { | |
10403 | WRITE_ONCE(priv.status, status); | |
10404 | bio_put(bio); | |
10405 | goto out; | |
10406 | } | |
10407 | bio = NULL; | |
10408 | continue; | |
10409 | } | |
10410 | ||
10411 | i++; | |
10412 | cur += bytes; | |
10413 | remaining -= bytes; | |
10414 | } | |
10415 | } | |
10416 | ||
10417 | out: | |
10418 | if (atomic_dec_return(&priv.pending)) | |
10419 | io_wait_event(priv.wait, !atomic_read(&priv.pending)); | |
10420 | /* See btrfs_encoded_read_endio() for ordering. */ | |
10421 | return blk_status_to_errno(READ_ONCE(priv.status)); | |
10422 | } | |
10423 | ||
10424 | static ssize_t btrfs_encoded_read_regular(struct kiocb *iocb, | |
10425 | struct iov_iter *iter, | |
10426 | u64 start, u64 lockend, | |
10427 | struct extent_state **cached_state, | |
10428 | u64 disk_bytenr, u64 disk_io_size, | |
10429 | size_t count, bool compressed, | |
10430 | bool *unlocked) | |
10431 | { | |
10432 | struct btrfs_inode *inode = BTRFS_I(file_inode(iocb->ki_filp)); | |
10433 | struct extent_io_tree *io_tree = &inode->io_tree; | |
10434 | struct page **pages; | |
10435 | unsigned long nr_pages, i; | |
10436 | u64 cur; | |
10437 | size_t page_offset; | |
10438 | ssize_t ret; | |
10439 | ||
10440 | nr_pages = DIV_ROUND_UP(disk_io_size, PAGE_SIZE); | |
10441 | pages = kcalloc(nr_pages, sizeof(struct page *), GFP_NOFS); | |
10442 | if (!pages) | |
10443 | return -ENOMEM; | |
10444 | for (i = 0; i < nr_pages; i++) { | |
10445 | pages[i] = alloc_page(GFP_NOFS); | |
10446 | if (!pages[i]) { | |
10447 | ret = -ENOMEM; | |
10448 | goto out; | |
10449 | } | |
10450 | } | |
10451 | ||
10452 | ret = btrfs_encoded_read_regular_fill_pages(inode, start, disk_bytenr, | |
10453 | disk_io_size, pages); | |
10454 | if (ret) | |
10455 | goto out; | |
10456 | ||
10457 | unlock_extent_cached(io_tree, start, lockend, cached_state); | |
10458 | btrfs_inode_unlock(&inode->vfs_inode, BTRFS_ILOCK_SHARED); | |
10459 | *unlocked = true; | |
10460 | ||
10461 | if (compressed) { | |
10462 | i = 0; | |
10463 | page_offset = 0; | |
10464 | } else { | |
10465 | i = (iocb->ki_pos - start) >> PAGE_SHIFT; | |
10466 | page_offset = (iocb->ki_pos - start) & (PAGE_SIZE - 1); | |
10467 | } | |
10468 | cur = 0; | |
10469 | while (cur < count) { | |
10470 | size_t bytes = min_t(size_t, count - cur, | |
10471 | PAGE_SIZE - page_offset); | |
10472 | ||
10473 | if (copy_page_to_iter(pages[i], page_offset, bytes, | |
10474 | iter) != bytes) { | |
10475 | ret = -EFAULT; | |
10476 | goto out; | |
10477 | } | |
10478 | i++; | |
10479 | cur += bytes; | |
10480 | page_offset = 0; | |
10481 | } | |
10482 | ret = count; | |
10483 | out: | |
10484 | for (i = 0; i < nr_pages; i++) { | |
10485 | if (pages[i]) | |
10486 | __free_page(pages[i]); | |
10487 | } | |
10488 | kfree(pages); | |
10489 | return ret; | |
10490 | } | |
10491 | ||
10492 | ssize_t btrfs_encoded_read(struct kiocb *iocb, struct iov_iter *iter, | |
10493 | struct btrfs_ioctl_encoded_io_args *encoded) | |
10494 | { | |
10495 | struct btrfs_inode *inode = BTRFS_I(file_inode(iocb->ki_filp)); | |
10496 | struct btrfs_fs_info *fs_info = inode->root->fs_info; | |
10497 | struct extent_io_tree *io_tree = &inode->io_tree; | |
10498 | ssize_t ret; | |
10499 | size_t count = iov_iter_count(iter); | |
10500 | u64 start, lockend, disk_bytenr, disk_io_size; | |
10501 | struct extent_state *cached_state = NULL; | |
10502 | struct extent_map *em; | |
10503 | bool unlocked = false; | |
10504 | ||
10505 | file_accessed(iocb->ki_filp); | |
10506 | ||
10507 | btrfs_inode_lock(&inode->vfs_inode, BTRFS_ILOCK_SHARED); | |
10508 | ||
10509 | if (iocb->ki_pos >= inode->vfs_inode.i_size) { | |
10510 | btrfs_inode_unlock(&inode->vfs_inode, BTRFS_ILOCK_SHARED); | |
10511 | return 0; | |
10512 | } | |
10513 | start = ALIGN_DOWN(iocb->ki_pos, fs_info->sectorsize); | |
10514 | /* | |
10515 | * We don't know how long the extent containing iocb->ki_pos is, but if | |
10516 | * it's compressed we know that it won't be longer than this. | |
10517 | */ | |
10518 | lockend = start + BTRFS_MAX_UNCOMPRESSED - 1; | |
10519 | ||
10520 | for (;;) { | |
10521 | struct btrfs_ordered_extent *ordered; | |
10522 | ||
10523 | ret = btrfs_wait_ordered_range(&inode->vfs_inode, start, | |
10524 | lockend - start + 1); | |
10525 | if (ret) | |
10526 | goto out_unlock_inode; | |
10527 | lock_extent_bits(io_tree, start, lockend, &cached_state); | |
10528 | ordered = btrfs_lookup_ordered_range(inode, start, | |
10529 | lockend - start + 1); | |
10530 | if (!ordered) | |
10531 | break; | |
10532 | btrfs_put_ordered_extent(ordered); | |
10533 | unlock_extent_cached(io_tree, start, lockend, &cached_state); | |
10534 | cond_resched(); | |
10535 | } | |
10536 | ||
10537 | em = btrfs_get_extent(inode, NULL, 0, start, lockend - start + 1); | |
10538 | if (IS_ERR(em)) { | |
10539 | ret = PTR_ERR(em); | |
10540 | goto out_unlock_extent; | |
10541 | } | |
10542 | ||
10543 | if (em->block_start == EXTENT_MAP_INLINE) { | |
10544 | u64 extent_start = em->start; | |
10545 | ||
10546 | /* | |
10547 | * For inline extents we get everything we need out of the | |
10548 | * extent item. | |
10549 | */ | |
10550 | free_extent_map(em); | |
10551 | em = NULL; | |
10552 | ret = btrfs_encoded_read_inline(iocb, iter, start, lockend, | |
10553 | &cached_state, extent_start, | |
10554 | count, encoded, &unlocked); | |
10555 | goto out; | |
10556 | } | |
10557 | ||
10558 | /* | |
10559 | * We only want to return up to EOF even if the extent extends beyond | |
10560 | * that. | |
10561 | */ | |
10562 | encoded->len = min_t(u64, extent_map_end(em), | |
10563 | inode->vfs_inode.i_size) - iocb->ki_pos; | |
10564 | if (em->block_start == EXTENT_MAP_HOLE || | |
10565 | test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) { | |
10566 | disk_bytenr = EXTENT_MAP_HOLE; | |
10567 | count = min_t(u64, count, encoded->len); | |
10568 | encoded->len = count; | |
10569 | encoded->unencoded_len = count; | |
10570 | } else if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) { | |
10571 | disk_bytenr = em->block_start; | |
10572 | /* | |
10573 | * Bail if the buffer isn't large enough to return the whole | |
10574 | * compressed extent. | |
10575 | */ | |
10576 | if (em->block_len > count) { | |
10577 | ret = -ENOBUFS; | |
10578 | goto out_em; | |
10579 | } | |
10580 | disk_io_size = count = em->block_len; | |
10581 | encoded->unencoded_len = em->ram_bytes; | |
10582 | encoded->unencoded_offset = iocb->ki_pos - em->orig_start; | |
10583 | ret = btrfs_encoded_io_compression_from_extent(fs_info, | |
10584 | em->compress_type); | |
10585 | if (ret < 0) | |
10586 | goto out_em; | |
10587 | encoded->compression = ret; | |
10588 | } else { | |
10589 | disk_bytenr = em->block_start + (start - em->start); | |
10590 | if (encoded->len > count) | |
10591 | encoded->len = count; | |
10592 | /* | |
10593 | * Don't read beyond what we locked. This also limits the page | |
10594 | * allocations that we'll do. | |
10595 | */ | |
10596 | disk_io_size = min(lockend + 1, iocb->ki_pos + encoded->len) - start; | |
10597 | count = start + disk_io_size - iocb->ki_pos; | |
10598 | encoded->len = count; | |
10599 | encoded->unencoded_len = count; | |
10600 | disk_io_size = ALIGN(disk_io_size, fs_info->sectorsize); | |
10601 | } | |
10602 | free_extent_map(em); | |
10603 | em = NULL; | |
10604 | ||
10605 | if (disk_bytenr == EXTENT_MAP_HOLE) { | |
10606 | unlock_extent_cached(io_tree, start, lockend, &cached_state); | |
10607 | btrfs_inode_unlock(&inode->vfs_inode, BTRFS_ILOCK_SHARED); | |
10608 | unlocked = true; | |
10609 | ret = iov_iter_zero(count, iter); | |
10610 | if (ret != count) | |
10611 | ret = -EFAULT; | |
10612 | } else { | |
10613 | ret = btrfs_encoded_read_regular(iocb, iter, start, lockend, | |
10614 | &cached_state, disk_bytenr, | |
10615 | disk_io_size, count, | |
10616 | encoded->compression, | |
10617 | &unlocked); | |
10618 | } | |
10619 | ||
10620 | out: | |
10621 | if (ret >= 0) | |
10622 | iocb->ki_pos += encoded->len; | |
10623 | out_em: | |
10624 | free_extent_map(em); | |
10625 | out_unlock_extent: | |
10626 | if (!unlocked) | |
10627 | unlock_extent_cached(io_tree, start, lockend, &cached_state); | |
10628 | out_unlock_inode: | |
10629 | if (!unlocked) | |
10630 | btrfs_inode_unlock(&inode->vfs_inode, BTRFS_ILOCK_SHARED); | |
10631 | return ret; | |
10632 | } | |
10633 | ||
7c0c7269 OS |
10634 | ssize_t btrfs_do_encoded_write(struct kiocb *iocb, struct iov_iter *from, |
10635 | const struct btrfs_ioctl_encoded_io_args *encoded) | |
10636 | { | |
10637 | struct btrfs_inode *inode = BTRFS_I(file_inode(iocb->ki_filp)); | |
10638 | struct btrfs_root *root = inode->root; | |
10639 | struct btrfs_fs_info *fs_info = root->fs_info; | |
10640 | struct extent_io_tree *io_tree = &inode->io_tree; | |
10641 | struct extent_changeset *data_reserved = NULL; | |
10642 | struct extent_state *cached_state = NULL; | |
10643 | int compression; | |
10644 | size_t orig_count; | |
10645 | u64 start, end; | |
10646 | u64 num_bytes, ram_bytes, disk_num_bytes; | |
10647 | unsigned long nr_pages, i; | |
10648 | struct page **pages; | |
10649 | struct btrfs_key ins; | |
10650 | bool extent_reserved = false; | |
10651 | struct extent_map *em; | |
10652 | ssize_t ret; | |
10653 | ||
10654 | switch (encoded->compression) { | |
10655 | case BTRFS_ENCODED_IO_COMPRESSION_ZLIB: | |
10656 | compression = BTRFS_COMPRESS_ZLIB; | |
10657 | break; | |
10658 | case BTRFS_ENCODED_IO_COMPRESSION_ZSTD: | |
10659 | compression = BTRFS_COMPRESS_ZSTD; | |
10660 | break; | |
10661 | case BTRFS_ENCODED_IO_COMPRESSION_LZO_4K: | |
10662 | case BTRFS_ENCODED_IO_COMPRESSION_LZO_8K: | |
10663 | case BTRFS_ENCODED_IO_COMPRESSION_LZO_16K: | |
10664 | case BTRFS_ENCODED_IO_COMPRESSION_LZO_32K: | |
10665 | case BTRFS_ENCODED_IO_COMPRESSION_LZO_64K: | |
10666 | /* The sector size must match for LZO. */ | |
10667 | if (encoded->compression - | |
10668 | BTRFS_ENCODED_IO_COMPRESSION_LZO_4K + 12 != | |
10669 | fs_info->sectorsize_bits) | |
10670 | return -EINVAL; | |
10671 | compression = BTRFS_COMPRESS_LZO; | |
10672 | break; | |
10673 | default: | |
10674 | return -EINVAL; | |
10675 | } | |
10676 | if (encoded->encryption != BTRFS_ENCODED_IO_ENCRYPTION_NONE) | |
10677 | return -EINVAL; | |
10678 | ||
10679 | orig_count = iov_iter_count(from); | |
10680 | ||
10681 | /* The extent size must be sane. */ | |
10682 | if (encoded->unencoded_len > BTRFS_MAX_UNCOMPRESSED || | |
10683 | orig_count > BTRFS_MAX_COMPRESSED || orig_count == 0) | |
10684 | return -EINVAL; | |
10685 | ||
10686 | /* | |
10687 | * The compressed data must be smaller than the decompressed data. | |
10688 | * | |
10689 | * It's of course possible for data to compress to larger or the same | |
10690 | * size, but the buffered I/O path falls back to no compression for such | |
10691 | * data, and we don't want to break any assumptions by creating these | |
10692 | * extents. | |
10693 | * | |
10694 | * Note that this is less strict than the current check we have that the | |
10695 | * compressed data must be at least one sector smaller than the | |
10696 | * decompressed data. We only want to enforce the weaker requirement | |
10697 | * from old kernels that it is at least one byte smaller. | |
10698 | */ | |
10699 | if (orig_count >= encoded->unencoded_len) | |
10700 | return -EINVAL; | |
10701 | ||
10702 | /* The extent must start on a sector boundary. */ | |
10703 | start = iocb->ki_pos; | |
10704 | if (!IS_ALIGNED(start, fs_info->sectorsize)) | |
10705 | return -EINVAL; | |
10706 | ||
10707 | /* | |
10708 | * The extent must end on a sector boundary. However, we allow a write | |
10709 | * which ends at or extends i_size to have an unaligned length; we round | |
10710 | * up the extent size and set i_size to the unaligned end. | |
10711 | */ | |
10712 | if (start + encoded->len < inode->vfs_inode.i_size && | |
10713 | !IS_ALIGNED(start + encoded->len, fs_info->sectorsize)) | |
10714 | return -EINVAL; | |
10715 | ||
10716 | /* Finally, the offset in the unencoded data must be sector-aligned. */ | |
10717 | if (!IS_ALIGNED(encoded->unencoded_offset, fs_info->sectorsize)) | |
10718 | return -EINVAL; | |
10719 | ||
10720 | num_bytes = ALIGN(encoded->len, fs_info->sectorsize); | |
10721 | ram_bytes = ALIGN(encoded->unencoded_len, fs_info->sectorsize); | |
10722 | end = start + num_bytes - 1; | |
10723 | ||
10724 | /* | |
10725 | * If the extent cannot be inline, the compressed data on disk must be | |
10726 | * sector-aligned. For convenience, we extend it with zeroes if it | |
10727 | * isn't. | |
10728 | */ | |
10729 | disk_num_bytes = ALIGN(orig_count, fs_info->sectorsize); | |
10730 | nr_pages = DIV_ROUND_UP(disk_num_bytes, PAGE_SIZE); | |
10731 | pages = kvcalloc(nr_pages, sizeof(struct page *), GFP_KERNEL_ACCOUNT); | |
10732 | if (!pages) | |
10733 | return -ENOMEM; | |
10734 | for (i = 0; i < nr_pages; i++) { | |
10735 | size_t bytes = min_t(size_t, PAGE_SIZE, iov_iter_count(from)); | |
10736 | char *kaddr; | |
10737 | ||
10738 | pages[i] = alloc_page(GFP_KERNEL_ACCOUNT); | |
10739 | if (!pages[i]) { | |
10740 | ret = -ENOMEM; | |
10741 | goto out_pages; | |
10742 | } | |
10743 | kaddr = kmap(pages[i]); | |
10744 | if (copy_from_iter(kaddr, bytes, from) != bytes) { | |
10745 | kunmap(pages[i]); | |
10746 | ret = -EFAULT; | |
10747 | goto out_pages; | |
10748 | } | |
10749 | if (bytes < PAGE_SIZE) | |
10750 | memset(kaddr + bytes, 0, PAGE_SIZE - bytes); | |
10751 | kunmap(pages[i]); | |
10752 | } | |
10753 | ||
10754 | for (;;) { | |
10755 | struct btrfs_ordered_extent *ordered; | |
10756 | ||
10757 | ret = btrfs_wait_ordered_range(&inode->vfs_inode, start, num_bytes); | |
10758 | if (ret) | |
10759 | goto out_pages; | |
10760 | ret = invalidate_inode_pages2_range(inode->vfs_inode.i_mapping, | |
10761 | start >> PAGE_SHIFT, | |
10762 | end >> PAGE_SHIFT); | |
10763 | if (ret) | |
10764 | goto out_pages; | |
10765 | lock_extent_bits(io_tree, start, end, &cached_state); | |
10766 | ordered = btrfs_lookup_ordered_range(inode, start, num_bytes); | |
10767 | if (!ordered && | |
10768 | !filemap_range_has_page(inode->vfs_inode.i_mapping, start, end)) | |
10769 | break; | |
10770 | if (ordered) | |
10771 | btrfs_put_ordered_extent(ordered); | |
10772 | unlock_extent_cached(io_tree, start, end, &cached_state); | |
10773 | cond_resched(); | |
10774 | } | |
10775 | ||
10776 | /* | |
10777 | * We don't use the higher-level delalloc space functions because our | |
10778 | * num_bytes and disk_num_bytes are different. | |
10779 | */ | |
10780 | ret = btrfs_alloc_data_chunk_ondemand(inode, disk_num_bytes); | |
10781 | if (ret) | |
10782 | goto out_unlock; | |
10783 | ret = btrfs_qgroup_reserve_data(inode, &data_reserved, start, num_bytes); | |
10784 | if (ret) | |
10785 | goto out_free_data_space; | |
10786 | ret = btrfs_delalloc_reserve_metadata(inode, num_bytes, disk_num_bytes); | |
10787 | if (ret) | |
10788 | goto out_qgroup_free_data; | |
10789 | ||
10790 | /* Try an inline extent first. */ | |
10791 | if (start == 0 && encoded->unencoded_len == encoded->len && | |
10792 | encoded->unencoded_offset == 0) { | |
10793 | ret = cow_file_range_inline(inode, encoded->len, orig_count, | |
10794 | compression, pages, true); | |
10795 | if (ret <= 0) { | |
10796 | if (ret == 0) | |
10797 | ret = orig_count; | |
10798 | goto out_delalloc_release; | |
10799 | } | |
10800 | } | |
10801 | ||
10802 | ret = btrfs_reserve_extent(root, disk_num_bytes, disk_num_bytes, | |
10803 | disk_num_bytes, 0, 0, &ins, 1, 1); | |
10804 | if (ret) | |
10805 | goto out_delalloc_release; | |
10806 | extent_reserved = true; | |
10807 | ||
10808 | em = create_io_em(inode, start, num_bytes, | |
10809 | start - encoded->unencoded_offset, ins.objectid, | |
10810 | ins.offset, ins.offset, ram_bytes, compression, | |
10811 | BTRFS_ORDERED_COMPRESSED); | |
10812 | if (IS_ERR(em)) { | |
10813 | ret = PTR_ERR(em); | |
10814 | goto out_free_reserved; | |
10815 | } | |
10816 | free_extent_map(em); | |
10817 | ||
10818 | ret = btrfs_add_ordered_extent(inode, start, num_bytes, ram_bytes, | |
10819 | ins.objectid, ins.offset, | |
10820 | encoded->unencoded_offset, | |
10821 | (1 << BTRFS_ORDERED_ENCODED) | | |
10822 | (1 << BTRFS_ORDERED_COMPRESSED), | |
10823 | compression); | |
10824 | if (ret) { | |
10825 | btrfs_drop_extent_cache(inode, start, end, 0); | |
10826 | goto out_free_reserved; | |
10827 | } | |
10828 | btrfs_dec_block_group_reservations(fs_info, ins.objectid); | |
10829 | ||
10830 | if (start + encoded->len > inode->vfs_inode.i_size) | |
10831 | i_size_write(&inode->vfs_inode, start + encoded->len); | |
10832 | ||
10833 | unlock_extent_cached(io_tree, start, end, &cached_state); | |
10834 | ||
10835 | btrfs_delalloc_release_extents(inode, num_bytes); | |
10836 | ||
10837 | if (btrfs_submit_compressed_write(inode, start, num_bytes, ins.objectid, | |
10838 | ins.offset, pages, nr_pages, 0, NULL, | |
10839 | false)) { | |
10840 | btrfs_writepage_endio_finish_ordered(inode, pages[0], start, end, 0); | |
10841 | ret = -EIO; | |
10842 | goto out_pages; | |
10843 | } | |
10844 | ret = orig_count; | |
10845 | goto out; | |
10846 | ||
10847 | out_free_reserved: | |
10848 | btrfs_dec_block_group_reservations(fs_info, ins.objectid); | |
10849 | btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1); | |
10850 | out_delalloc_release: | |
10851 | btrfs_delalloc_release_extents(inode, num_bytes); | |
10852 | btrfs_delalloc_release_metadata(inode, disk_num_bytes, ret < 0); | |
10853 | out_qgroup_free_data: | |
10854 | if (ret < 0) | |
10855 | btrfs_qgroup_free_data(inode, data_reserved, start, num_bytes); | |
10856 | out_free_data_space: | |
10857 | /* | |
10858 | * If btrfs_reserve_extent() succeeded, then we already decremented | |
10859 | * bytes_may_use. | |
10860 | */ | |
10861 | if (!extent_reserved) | |
10862 | btrfs_free_reserved_data_space_noquota(fs_info, disk_num_bytes); | |
10863 | out_unlock: | |
10864 | unlock_extent_cached(io_tree, start, end, &cached_state); | |
10865 | out_pages: | |
10866 | for (i = 0; i < nr_pages; i++) { | |
10867 | if (pages[i]) | |
10868 | __free_page(pages[i]); | |
10869 | } | |
10870 | kvfree(pages); | |
10871 | out: | |
10872 | if (ret >= 0) | |
10873 | iocb->ki_pos += encoded->len; | |
10874 | return ret; | |
10875 | } | |
10876 | ||
ed46ff3d OS |
10877 | #ifdef CONFIG_SWAP |
10878 | /* | |
10879 | * Add an entry indicating a block group or device which is pinned by a | |
10880 | * swapfile. Returns 0 on success, 1 if there is already an entry for it, or a | |
10881 | * negative errno on failure. | |
10882 | */ | |
10883 | static int btrfs_add_swapfile_pin(struct inode *inode, void *ptr, | |
10884 | bool is_block_group) | |
10885 | { | |
10886 | struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info; | |
10887 | struct btrfs_swapfile_pin *sp, *entry; | |
10888 | struct rb_node **p; | |
10889 | struct rb_node *parent = NULL; | |
10890 | ||
10891 | sp = kmalloc(sizeof(*sp), GFP_NOFS); | |
10892 | if (!sp) | |
10893 | return -ENOMEM; | |
10894 | sp->ptr = ptr; | |
10895 | sp->inode = inode; | |
10896 | sp->is_block_group = is_block_group; | |
195a49ea | 10897 | sp->bg_extent_count = 1; |
ed46ff3d OS |
10898 | |
10899 | spin_lock(&fs_info->swapfile_pins_lock); | |
10900 | p = &fs_info->swapfile_pins.rb_node; | |
10901 | while (*p) { | |
10902 | parent = *p; | |
10903 | entry = rb_entry(parent, struct btrfs_swapfile_pin, node); | |
10904 | if (sp->ptr < entry->ptr || | |
10905 | (sp->ptr == entry->ptr && sp->inode < entry->inode)) { | |
10906 | p = &(*p)->rb_left; | |
10907 | } else if (sp->ptr > entry->ptr || | |
10908 | (sp->ptr == entry->ptr && sp->inode > entry->inode)) { | |
10909 | p = &(*p)->rb_right; | |
10910 | } else { | |
195a49ea FM |
10911 | if (is_block_group) |
10912 | entry->bg_extent_count++; | |
ed46ff3d OS |
10913 | spin_unlock(&fs_info->swapfile_pins_lock); |
10914 | kfree(sp); | |
10915 | return 1; | |
10916 | } | |
10917 | } | |
10918 | rb_link_node(&sp->node, parent, p); | |
10919 | rb_insert_color(&sp->node, &fs_info->swapfile_pins); | |
10920 | spin_unlock(&fs_info->swapfile_pins_lock); | |
10921 | return 0; | |
10922 | } | |
10923 | ||
10924 | /* Free all of the entries pinned by this swapfile. */ | |
10925 | static void btrfs_free_swapfile_pins(struct inode *inode) | |
10926 | { | |
10927 | struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info; | |
10928 | struct btrfs_swapfile_pin *sp; | |
10929 | struct rb_node *node, *next; | |
10930 | ||
10931 | spin_lock(&fs_info->swapfile_pins_lock); | |
10932 | node = rb_first(&fs_info->swapfile_pins); | |
10933 | while (node) { | |
10934 | next = rb_next(node); | |
10935 | sp = rb_entry(node, struct btrfs_swapfile_pin, node); | |
10936 | if (sp->inode == inode) { | |
10937 | rb_erase(&sp->node, &fs_info->swapfile_pins); | |
195a49ea FM |
10938 | if (sp->is_block_group) { |
10939 | btrfs_dec_block_group_swap_extents(sp->ptr, | |
10940 | sp->bg_extent_count); | |
ed46ff3d | 10941 | btrfs_put_block_group(sp->ptr); |
195a49ea | 10942 | } |
ed46ff3d OS |
10943 | kfree(sp); |
10944 | } | |
10945 | node = next; | |
10946 | } | |
10947 | spin_unlock(&fs_info->swapfile_pins_lock); | |
10948 | } | |
10949 | ||
10950 | struct btrfs_swap_info { | |
10951 | u64 start; | |
10952 | u64 block_start; | |
10953 | u64 block_len; | |
10954 | u64 lowest_ppage; | |
10955 | u64 highest_ppage; | |
10956 | unsigned long nr_pages; | |
10957 | int nr_extents; | |
10958 | }; | |
10959 | ||
10960 | static int btrfs_add_swap_extent(struct swap_info_struct *sis, | |
10961 | struct btrfs_swap_info *bsi) | |
10962 | { | |
10963 | unsigned long nr_pages; | |
c2f82263 | 10964 | unsigned long max_pages; |
ed46ff3d OS |
10965 | u64 first_ppage, first_ppage_reported, next_ppage; |
10966 | int ret; | |
10967 | ||
c2f82263 FM |
10968 | /* |
10969 | * Our swapfile may have had its size extended after the swap header was | |
10970 | * written. In that case activating the swapfile should not go beyond | |
10971 | * the max size set in the swap header. | |
10972 | */ | |
10973 | if (bsi->nr_pages >= sis->max) | |
10974 | return 0; | |
10975 | ||
10976 | max_pages = sis->max - bsi->nr_pages; | |
ed46ff3d OS |
10977 | first_ppage = ALIGN(bsi->block_start, PAGE_SIZE) >> PAGE_SHIFT; |
10978 | next_ppage = ALIGN_DOWN(bsi->block_start + bsi->block_len, | |
10979 | PAGE_SIZE) >> PAGE_SHIFT; | |
10980 | ||
10981 | if (first_ppage >= next_ppage) | |
10982 | return 0; | |
10983 | nr_pages = next_ppage - first_ppage; | |
c2f82263 | 10984 | nr_pages = min(nr_pages, max_pages); |
ed46ff3d OS |
10985 | |
10986 | first_ppage_reported = first_ppage; | |
10987 | if (bsi->start == 0) | |
10988 | first_ppage_reported++; | |
10989 | if (bsi->lowest_ppage > first_ppage_reported) | |
10990 | bsi->lowest_ppage = first_ppage_reported; | |
10991 | if (bsi->highest_ppage < (next_ppage - 1)) | |
10992 | bsi->highest_ppage = next_ppage - 1; | |
10993 | ||
10994 | ret = add_swap_extent(sis, bsi->nr_pages, nr_pages, first_ppage); | |
10995 | if (ret < 0) | |
10996 | return ret; | |
10997 | bsi->nr_extents += ret; | |
10998 | bsi->nr_pages += nr_pages; | |
10999 | return 0; | |
11000 | } | |
11001 | ||
11002 | static void btrfs_swap_deactivate(struct file *file) | |
11003 | { | |
11004 | struct inode *inode = file_inode(file); | |
11005 | ||
11006 | btrfs_free_swapfile_pins(inode); | |
11007 | atomic_dec(&BTRFS_I(inode)->root->nr_swapfiles); | |
11008 | } | |
11009 | ||
11010 | static int btrfs_swap_activate(struct swap_info_struct *sis, struct file *file, | |
11011 | sector_t *span) | |
11012 | { | |
11013 | struct inode *inode = file_inode(file); | |
dd0734f2 FM |
11014 | struct btrfs_root *root = BTRFS_I(inode)->root; |
11015 | struct btrfs_fs_info *fs_info = root->fs_info; | |
ed46ff3d OS |
11016 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
11017 | struct extent_state *cached_state = NULL; | |
11018 | struct extent_map *em = NULL; | |
11019 | struct btrfs_device *device = NULL; | |
11020 | struct btrfs_swap_info bsi = { | |
11021 | .lowest_ppage = (sector_t)-1ULL, | |
11022 | }; | |
11023 | int ret = 0; | |
11024 | u64 isize; | |
11025 | u64 start; | |
11026 | ||
11027 | /* | |
11028 | * If the swap file was just created, make sure delalloc is done. If the | |
11029 | * file changes again after this, the user is doing something stupid and | |
11030 | * we don't really care. | |
11031 | */ | |
11032 | ret = btrfs_wait_ordered_range(inode, 0, (u64)-1); | |
11033 | if (ret) | |
11034 | return ret; | |
11035 | ||
11036 | /* | |
11037 | * The inode is locked, so these flags won't change after we check them. | |
11038 | */ | |
11039 | if (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS) { | |
11040 | btrfs_warn(fs_info, "swapfile must not be compressed"); | |
11041 | return -EINVAL; | |
11042 | } | |
11043 | if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW)) { | |
11044 | btrfs_warn(fs_info, "swapfile must not be copy-on-write"); | |
11045 | return -EINVAL; | |
11046 | } | |
11047 | if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) { | |
11048 | btrfs_warn(fs_info, "swapfile must not be checksummed"); | |
11049 | return -EINVAL; | |
11050 | } | |
11051 | ||
11052 | /* | |
11053 | * Balance or device remove/replace/resize can move stuff around from | |
c3e1f96c GR |
11054 | * under us. The exclop protection makes sure they aren't running/won't |
11055 | * run concurrently while we are mapping the swap extents, and | |
11056 | * fs_info->swapfile_pins prevents them from running while the swap | |
11057 | * file is active and moving the extents. Note that this also prevents | |
11058 | * a concurrent device add which isn't actually necessary, but it's not | |
ed46ff3d OS |
11059 | * really worth the trouble to allow it. |
11060 | */ | |
c3e1f96c | 11061 | if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_SWAP_ACTIVATE)) { |
ed46ff3d OS |
11062 | btrfs_warn(fs_info, |
11063 | "cannot activate swapfile while exclusive operation is running"); | |
11064 | return -EBUSY; | |
11065 | } | |
dd0734f2 FM |
11066 | |
11067 | /* | |
11068 | * Prevent snapshot creation while we are activating the swap file. | |
11069 | * We do not want to race with snapshot creation. If snapshot creation | |
11070 | * already started before we bumped nr_swapfiles from 0 to 1 and | |
11071 | * completes before the first write into the swap file after it is | |
11072 | * activated, than that write would fallback to COW. | |
11073 | */ | |
11074 | if (!btrfs_drew_try_write_lock(&root->snapshot_lock)) { | |
11075 | btrfs_exclop_finish(fs_info); | |
11076 | btrfs_warn(fs_info, | |
11077 | "cannot activate swapfile because snapshot creation is in progress"); | |
11078 | return -EINVAL; | |
11079 | } | |
ed46ff3d OS |
11080 | /* |
11081 | * Snapshots can create extents which require COW even if NODATACOW is | |
11082 | * set. We use this counter to prevent snapshots. We must increment it | |
11083 | * before walking the extents because we don't want a concurrent | |
11084 | * snapshot to run after we've already checked the extents. | |
60021bd7 KH |
11085 | * |
11086 | * It is possible that subvolume is marked for deletion but still not | |
11087 | * removed yet. To prevent this race, we check the root status before | |
11088 | * activating the swapfile. | |
ed46ff3d | 11089 | */ |
60021bd7 KH |
11090 | spin_lock(&root->root_item_lock); |
11091 | if (btrfs_root_dead(root)) { | |
11092 | spin_unlock(&root->root_item_lock); | |
11093 | ||
11094 | btrfs_exclop_finish(fs_info); | |
11095 | btrfs_warn(fs_info, | |
11096 | "cannot activate swapfile because subvolume %llu is being deleted", | |
11097 | root->root_key.objectid); | |
11098 | return -EPERM; | |
11099 | } | |
dd0734f2 | 11100 | atomic_inc(&root->nr_swapfiles); |
60021bd7 | 11101 | spin_unlock(&root->root_item_lock); |
ed46ff3d OS |
11102 | |
11103 | isize = ALIGN_DOWN(inode->i_size, fs_info->sectorsize); | |
11104 | ||
11105 | lock_extent_bits(io_tree, 0, isize - 1, &cached_state); | |
11106 | start = 0; | |
11107 | while (start < isize) { | |
11108 | u64 logical_block_start, physical_block_start; | |
32da5386 | 11109 | struct btrfs_block_group *bg; |
ed46ff3d OS |
11110 | u64 len = isize - start; |
11111 | ||
39b07b5d | 11112 | em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len); |
ed46ff3d OS |
11113 | if (IS_ERR(em)) { |
11114 | ret = PTR_ERR(em); | |
11115 | goto out; | |
11116 | } | |
11117 | ||
11118 | if (em->block_start == EXTENT_MAP_HOLE) { | |
11119 | btrfs_warn(fs_info, "swapfile must not have holes"); | |
11120 | ret = -EINVAL; | |
11121 | goto out; | |
11122 | } | |
11123 | if (em->block_start == EXTENT_MAP_INLINE) { | |
11124 | /* | |
11125 | * It's unlikely we'll ever actually find ourselves | |
11126 | * here, as a file small enough to fit inline won't be | |
11127 | * big enough to store more than the swap header, but in | |
11128 | * case something changes in the future, let's catch it | |
11129 | * here rather than later. | |
11130 | */ | |
11131 | btrfs_warn(fs_info, "swapfile must not be inline"); | |
11132 | ret = -EINVAL; | |
11133 | goto out; | |
11134 | } | |
11135 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) { | |
11136 | btrfs_warn(fs_info, "swapfile must not be compressed"); | |
11137 | ret = -EINVAL; | |
11138 | goto out; | |
11139 | } | |
11140 | ||
11141 | logical_block_start = em->block_start + (start - em->start); | |
11142 | len = min(len, em->len - (start - em->start)); | |
11143 | free_extent_map(em); | |
11144 | em = NULL; | |
11145 | ||
a84d5d42 | 11146 | ret = can_nocow_extent(inode, start, &len, NULL, NULL, NULL, true); |
ed46ff3d OS |
11147 | if (ret < 0) { |
11148 | goto out; | |
11149 | } else if (ret) { | |
11150 | ret = 0; | |
11151 | } else { | |
11152 | btrfs_warn(fs_info, | |
11153 | "swapfile must not be copy-on-write"); | |
11154 | ret = -EINVAL; | |
11155 | goto out; | |
11156 | } | |
11157 | ||
11158 | em = btrfs_get_chunk_map(fs_info, logical_block_start, len); | |
11159 | if (IS_ERR(em)) { | |
11160 | ret = PTR_ERR(em); | |
11161 | goto out; | |
11162 | } | |
11163 | ||
11164 | if (em->map_lookup->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) { | |
11165 | btrfs_warn(fs_info, | |
11166 | "swapfile must have single data profile"); | |
11167 | ret = -EINVAL; | |
11168 | goto out; | |
11169 | } | |
11170 | ||
11171 | if (device == NULL) { | |
11172 | device = em->map_lookup->stripes[0].dev; | |
11173 | ret = btrfs_add_swapfile_pin(inode, device, false); | |
11174 | if (ret == 1) | |
11175 | ret = 0; | |
11176 | else if (ret) | |
11177 | goto out; | |
11178 | } else if (device != em->map_lookup->stripes[0].dev) { | |
11179 | btrfs_warn(fs_info, "swapfile must be on one device"); | |
11180 | ret = -EINVAL; | |
11181 | goto out; | |
11182 | } | |
11183 | ||
11184 | physical_block_start = (em->map_lookup->stripes[0].physical + | |
11185 | (logical_block_start - em->start)); | |
11186 | len = min(len, em->len - (logical_block_start - em->start)); | |
11187 | free_extent_map(em); | |
11188 | em = NULL; | |
11189 | ||
11190 | bg = btrfs_lookup_block_group(fs_info, logical_block_start); | |
11191 | if (!bg) { | |
11192 | btrfs_warn(fs_info, | |
11193 | "could not find block group containing swapfile"); | |
11194 | ret = -EINVAL; | |
11195 | goto out; | |
11196 | } | |
11197 | ||
195a49ea FM |
11198 | if (!btrfs_inc_block_group_swap_extents(bg)) { |
11199 | btrfs_warn(fs_info, | |
11200 | "block group for swapfile at %llu is read-only%s", | |
11201 | bg->start, | |
11202 | atomic_read(&fs_info->scrubs_running) ? | |
11203 | " (scrub running)" : ""); | |
11204 | btrfs_put_block_group(bg); | |
11205 | ret = -EINVAL; | |
11206 | goto out; | |
11207 | } | |
11208 | ||
ed46ff3d OS |
11209 | ret = btrfs_add_swapfile_pin(inode, bg, true); |
11210 | if (ret) { | |
11211 | btrfs_put_block_group(bg); | |
11212 | if (ret == 1) | |
11213 | ret = 0; | |
11214 | else | |
11215 | goto out; | |
11216 | } | |
11217 | ||
11218 | if (bsi.block_len && | |
11219 | bsi.block_start + bsi.block_len == physical_block_start) { | |
11220 | bsi.block_len += len; | |
11221 | } else { | |
11222 | if (bsi.block_len) { | |
11223 | ret = btrfs_add_swap_extent(sis, &bsi); | |
11224 | if (ret) | |
11225 | goto out; | |
11226 | } | |
11227 | bsi.start = start; | |
11228 | bsi.block_start = physical_block_start; | |
11229 | bsi.block_len = len; | |
11230 | } | |
11231 | ||
11232 | start += len; | |
11233 | } | |
11234 | ||
11235 | if (bsi.block_len) | |
11236 | ret = btrfs_add_swap_extent(sis, &bsi); | |
11237 | ||
11238 | out: | |
11239 | if (!IS_ERR_OR_NULL(em)) | |
11240 | free_extent_map(em); | |
11241 | ||
11242 | unlock_extent_cached(io_tree, 0, isize - 1, &cached_state); | |
11243 | ||
11244 | if (ret) | |
11245 | btrfs_swap_deactivate(file); | |
11246 | ||
dd0734f2 FM |
11247 | btrfs_drew_write_unlock(&root->snapshot_lock); |
11248 | ||
c3e1f96c | 11249 | btrfs_exclop_finish(fs_info); |
ed46ff3d OS |
11250 | |
11251 | if (ret) | |
11252 | return ret; | |
11253 | ||
11254 | if (device) | |
11255 | sis->bdev = device->bdev; | |
11256 | *span = bsi.highest_ppage - bsi.lowest_ppage + 1; | |
11257 | sis->max = bsi.nr_pages; | |
11258 | sis->pages = bsi.nr_pages - 1; | |
11259 | sis->highest_bit = bsi.nr_pages - 1; | |
11260 | return bsi.nr_extents; | |
11261 | } | |
11262 | #else | |
11263 | static void btrfs_swap_deactivate(struct file *file) | |
11264 | { | |
11265 | } | |
11266 | ||
11267 | static int btrfs_swap_activate(struct swap_info_struct *sis, struct file *file, | |
11268 | sector_t *span) | |
11269 | { | |
11270 | return -EOPNOTSUPP; | |
11271 | } | |
11272 | #endif | |
11273 | ||
2766ff61 FM |
11274 | /* |
11275 | * Update the number of bytes used in the VFS' inode. When we replace extents in | |
11276 | * a range (clone, dedupe, fallocate's zero range), we must update the number of | |
11277 | * bytes used by the inode in an atomic manner, so that concurrent stat(2) calls | |
11278 | * always get a correct value. | |
11279 | */ | |
11280 | void btrfs_update_inode_bytes(struct btrfs_inode *inode, | |
11281 | const u64 add_bytes, | |
11282 | const u64 del_bytes) | |
11283 | { | |
11284 | if (add_bytes == del_bytes) | |
11285 | return; | |
11286 | ||
11287 | spin_lock(&inode->lock); | |
11288 | if (del_bytes > 0) | |
11289 | inode_sub_bytes(&inode->vfs_inode, del_bytes); | |
11290 | if (add_bytes > 0) | |
11291 | inode_add_bytes(&inode->vfs_inode, add_bytes); | |
11292 | spin_unlock(&inode->lock); | |
11293 | } | |
11294 | ||
6e1d5dcc | 11295 | static const struct inode_operations btrfs_dir_inode_operations = { |
3394e160 | 11296 | .getattr = btrfs_getattr, |
39279cc3 CM |
11297 | .lookup = btrfs_lookup, |
11298 | .create = btrfs_create, | |
11299 | .unlink = btrfs_unlink, | |
11300 | .link = btrfs_link, | |
11301 | .mkdir = btrfs_mkdir, | |
11302 | .rmdir = btrfs_rmdir, | |
2773bf00 | 11303 | .rename = btrfs_rename2, |
39279cc3 CM |
11304 | .symlink = btrfs_symlink, |
11305 | .setattr = btrfs_setattr, | |
618e21d5 | 11306 | .mknod = btrfs_mknod, |
5103e947 | 11307 | .listxattr = btrfs_listxattr, |
fdebe2bd | 11308 | .permission = btrfs_permission, |
4e34e719 | 11309 | .get_acl = btrfs_get_acl, |
996a710d | 11310 | .set_acl = btrfs_set_acl, |
93fd63c2 | 11311 | .update_time = btrfs_update_time, |
ef3b9af5 | 11312 | .tmpfile = btrfs_tmpfile, |
97fc2977 MS |
11313 | .fileattr_get = btrfs_fileattr_get, |
11314 | .fileattr_set = btrfs_fileattr_set, | |
39279cc3 | 11315 | }; |
76dda93c | 11316 | |
828c0950 | 11317 | static const struct file_operations btrfs_dir_file_operations = { |
39279cc3 CM |
11318 | .llseek = generic_file_llseek, |
11319 | .read = generic_read_dir, | |
02dbfc99 | 11320 | .iterate_shared = btrfs_real_readdir, |
23b5ec74 | 11321 | .open = btrfs_opendir, |
34287aa3 | 11322 | .unlocked_ioctl = btrfs_ioctl, |
39279cc3 | 11323 | #ifdef CONFIG_COMPAT |
4c63c245 | 11324 | .compat_ioctl = btrfs_compat_ioctl, |
39279cc3 | 11325 | #endif |
6bf13c0c | 11326 | .release = btrfs_release_file, |
e02119d5 | 11327 | .fsync = btrfs_sync_file, |
39279cc3 CM |
11328 | }; |
11329 | ||
35054394 CM |
11330 | /* |
11331 | * btrfs doesn't support the bmap operation because swapfiles | |
11332 | * use bmap to make a mapping of extents in the file. They assume | |
11333 | * these extents won't change over the life of the file and they | |
11334 | * use the bmap result to do IO directly to the drive. | |
11335 | * | |
11336 | * the btrfs bmap call would return logical addresses that aren't | |
11337 | * suitable for IO and they also will change frequently as COW | |
11338 | * operations happen. So, swapfile + btrfs == corruption. | |
11339 | * | |
11340 | * For now we're avoiding this by dropping bmap. | |
11341 | */ | |
7f09410b | 11342 | static const struct address_space_operations btrfs_aops = { |
39279cc3 CM |
11343 | .readpage = btrfs_readpage, |
11344 | .writepage = btrfs_writepage, | |
b293f02e | 11345 | .writepages = btrfs_writepages, |
ba206a02 | 11346 | .readahead = btrfs_readahead, |
f85781fb | 11347 | .direct_IO = noop_direct_IO, |
895586eb | 11348 | .invalidate_folio = btrfs_invalidate_folio, |
a52d9a80 | 11349 | .releasepage = btrfs_releasepage, |
f8e66081 RG |
11350 | #ifdef CONFIG_MIGRATION |
11351 | .migratepage = btrfs_migratepage, | |
11352 | #endif | |
187c82cb | 11353 | .dirty_folio = filemap_dirty_folio, |
465fdd97 | 11354 | .error_remove_page = generic_error_remove_page, |
ed46ff3d OS |
11355 | .swap_activate = btrfs_swap_activate, |
11356 | .swap_deactivate = btrfs_swap_deactivate, | |
39279cc3 CM |
11357 | }; |
11358 | ||
6e1d5dcc | 11359 | static const struct inode_operations btrfs_file_inode_operations = { |
39279cc3 CM |
11360 | .getattr = btrfs_getattr, |
11361 | .setattr = btrfs_setattr, | |
5103e947 | 11362 | .listxattr = btrfs_listxattr, |
fdebe2bd | 11363 | .permission = btrfs_permission, |
1506fcc8 | 11364 | .fiemap = btrfs_fiemap, |
4e34e719 | 11365 | .get_acl = btrfs_get_acl, |
996a710d | 11366 | .set_acl = btrfs_set_acl, |
e41f941a | 11367 | .update_time = btrfs_update_time, |
97fc2977 MS |
11368 | .fileattr_get = btrfs_fileattr_get, |
11369 | .fileattr_set = btrfs_fileattr_set, | |
39279cc3 | 11370 | }; |
6e1d5dcc | 11371 | static const struct inode_operations btrfs_special_inode_operations = { |
618e21d5 JB |
11372 | .getattr = btrfs_getattr, |
11373 | .setattr = btrfs_setattr, | |
fdebe2bd | 11374 | .permission = btrfs_permission, |
33268eaf | 11375 | .listxattr = btrfs_listxattr, |
4e34e719 | 11376 | .get_acl = btrfs_get_acl, |
996a710d | 11377 | .set_acl = btrfs_set_acl, |
e41f941a | 11378 | .update_time = btrfs_update_time, |
618e21d5 | 11379 | }; |
6e1d5dcc | 11380 | static const struct inode_operations btrfs_symlink_inode_operations = { |
6b255391 | 11381 | .get_link = page_get_link, |
f209561a | 11382 | .getattr = btrfs_getattr, |
22c44fe6 | 11383 | .setattr = btrfs_setattr, |
fdebe2bd | 11384 | .permission = btrfs_permission, |
0279b4cd | 11385 | .listxattr = btrfs_listxattr, |
e41f941a | 11386 | .update_time = btrfs_update_time, |
39279cc3 | 11387 | }; |
76dda93c | 11388 | |
82d339d9 | 11389 | const struct dentry_operations btrfs_dentry_operations = { |
76dda93c YZ |
11390 | .d_delete = btrfs_dentry_delete, |
11391 | }; |