]>
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 | ||
8f18cf13 | 6 | #include <linux/kernel.h> |
065631f6 | 7 | #include <linux/bio.h> |
55e20bd1 | 8 | #include <linux/buffer_head.h> |
f2eb0a24 | 9 | #include <linux/file.h> |
39279cc3 CM |
10 | #include <linux/fs.h> |
11 | #include <linux/pagemap.h> | |
12 | #include <linux/highmem.h> | |
13 | #include <linux/time.h> | |
14 | #include <linux/init.h> | |
15 | #include <linux/string.h> | |
39279cc3 | 16 | #include <linux/backing-dev.h> |
39279cc3 | 17 | #include <linux/writeback.h> |
39279cc3 | 18 | #include <linux/compat.h> |
5103e947 | 19 | #include <linux/xattr.h> |
33268eaf | 20 | #include <linux/posix_acl.h> |
d899e052 | 21 | #include <linux/falloc.h> |
5a0e3ad6 | 22 | #include <linux/slab.h> |
7a36ddec | 23 | #include <linux/ratelimit.h> |
55e301fd | 24 | #include <linux/btrfs.h> |
53b381b3 | 25 | #include <linux/blkdev.h> |
f23b5a59 | 26 | #include <linux/posix_acl_xattr.h> |
e2e40f2c | 27 | #include <linux/uio.h> |
69fe2d75 | 28 | #include <linux/magic.h> |
ae5e165d | 29 | #include <linux/iversion.h> |
ed46ff3d | 30 | #include <linux/swap.h> |
f8e66081 | 31 | #include <linux/migrate.h> |
b1c16ac9 | 32 | #include <linux/sched/mm.h> |
92d32170 | 33 | #include <asm/unaligned.h> |
602cbe91 | 34 | #include "misc.h" |
39279cc3 CM |
35 | #include "ctree.h" |
36 | #include "disk-io.h" | |
37 | #include "transaction.h" | |
38 | #include "btrfs_inode.h" | |
39279cc3 | 39 | #include "print-tree.h" |
e6dcd2dc | 40 | #include "ordered-data.h" |
95819c05 | 41 | #include "xattr.h" |
e02119d5 | 42 | #include "tree-log.h" |
4a54c8c1 | 43 | #include "volumes.h" |
c8b97818 | 44 | #include "compression.h" |
b4ce94de | 45 | #include "locking.h" |
dc89e982 | 46 | #include "free-space-cache.h" |
581bb050 | 47 | #include "inode-map.h" |
63541927 | 48 | #include "props.h" |
31193213 | 49 | #include "qgroup.h" |
86736342 | 50 | #include "delalloc-space.h" |
aac0023c | 51 | #include "block-group.h" |
467dc47e | 52 | #include "space-info.h" |
39279cc3 CM |
53 | |
54 | struct btrfs_iget_args { | |
0202e83f | 55 | u64 ino; |
39279cc3 CM |
56 | struct btrfs_root *root; |
57 | }; | |
58 | ||
f28a4928 | 59 | struct btrfs_dio_data { |
f28a4928 | 60 | u64 reserve; |
55e20bd1 DS |
61 | u64 unsubmitted_oe_range_start; |
62 | u64 unsubmitted_oe_range_end; | |
63 | int overwrite; | |
f28a4928 FM |
64 | }; |
65 | ||
6e1d5dcc AD |
66 | static const struct inode_operations btrfs_dir_inode_operations; |
67 | static const struct inode_operations btrfs_symlink_inode_operations; | |
6e1d5dcc AD |
68 | static const struct inode_operations btrfs_special_inode_operations; |
69 | static const struct inode_operations btrfs_file_inode_operations; | |
7f09410b | 70 | static const struct address_space_operations btrfs_aops; |
828c0950 | 71 | static const struct file_operations btrfs_dir_file_operations; |
20e5506b | 72 | static const struct extent_io_ops btrfs_extent_io_ops; |
39279cc3 CM |
73 | |
74 | static struct kmem_cache *btrfs_inode_cachep; | |
75 | struct kmem_cache *btrfs_trans_handle_cachep; | |
39279cc3 | 76 | struct kmem_cache *btrfs_path_cachep; |
dc89e982 | 77 | struct kmem_cache *btrfs_free_space_cachep; |
3acd4850 | 78 | struct kmem_cache *btrfs_free_space_bitmap_cachep; |
39279cc3 | 79 | |
3972f260 | 80 | static int btrfs_setsize(struct inode *inode, struct iattr *attr); |
213e8c55 | 81 | static int btrfs_truncate(struct inode *inode, bool skip_writeback); |
5fd02043 | 82 | static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent); |
6e26c442 | 83 | static noinline int cow_file_range(struct btrfs_inode *inode, |
771ed689 | 84 | struct page *locked_page, |
74e9194a | 85 | u64 start, u64 end, int *page_started, |
330a5827 | 86 | unsigned long *nr_written, int unlock); |
4b67c11d NB |
87 | static struct extent_map *create_io_em(struct btrfs_inode *inode, u64 start, |
88 | u64 len, u64 orig_start, u64 block_start, | |
6f9994db LB |
89 | u64 block_len, u64 orig_block_len, |
90 | u64 ram_bytes, int compress_type, | |
91 | int type); | |
7b128766 | 92 | |
b672b5c1 | 93 | static void __endio_write_update_ordered(struct btrfs_inode *inode, |
52427260 QW |
94 | const u64 offset, const u64 bytes, |
95 | const bool uptodate); | |
96 | ||
97 | /* | |
98 | * Cleanup all submitted ordered extents in specified range to handle errors | |
52042d8e | 99 | * from the btrfs_run_delalloc_range() callback. |
52427260 QW |
100 | * |
101 | * NOTE: caller must ensure that when an error happens, it can not call | |
102 | * extent_clear_unlock_delalloc() to clear both the bits EXTENT_DO_ACCOUNTING | |
103 | * and EXTENT_DELALLOC simultaneously, because that causes the reserved metadata | |
104 | * to be released, which we want to happen only when finishing the ordered | |
d1051d6e | 105 | * extent (btrfs_finish_ordered_io()). |
52427260 | 106 | */ |
64e1db56 | 107 | static inline void btrfs_cleanup_ordered_extents(struct btrfs_inode *inode, |
d1051d6e NB |
108 | struct page *locked_page, |
109 | u64 offset, u64 bytes) | |
52427260 | 110 | { |
63d71450 NA |
111 | unsigned long index = offset >> PAGE_SHIFT; |
112 | unsigned long end_index = (offset + bytes - 1) >> PAGE_SHIFT; | |
d1051d6e NB |
113 | u64 page_start = page_offset(locked_page); |
114 | u64 page_end = page_start + PAGE_SIZE - 1; | |
115 | ||
63d71450 NA |
116 | struct page *page; |
117 | ||
118 | while (index <= end_index) { | |
64e1db56 | 119 | page = find_get_page(inode->vfs_inode.i_mapping, index); |
63d71450 NA |
120 | index++; |
121 | if (!page) | |
122 | continue; | |
123 | ClearPagePrivate2(page); | |
124 | put_page(page); | |
125 | } | |
d1051d6e NB |
126 | |
127 | /* | |
128 | * In case this page belongs to the delalloc range being instantiated | |
129 | * then skip it, since the first page of a range is going to be | |
130 | * properly cleaned up by the caller of run_delalloc_range | |
131 | */ | |
132 | if (page_start >= offset && page_end <= (offset + bytes - 1)) { | |
133 | offset += PAGE_SIZE; | |
134 | bytes -= PAGE_SIZE; | |
135 | } | |
136 | ||
64e1db56 | 137 | return __endio_write_update_ordered(inode, offset, bytes, false); |
52427260 QW |
138 | } |
139 | ||
48a3b636 | 140 | static int btrfs_dirty_inode(struct inode *inode); |
7b128766 | 141 | |
6a3891c5 JB |
142 | #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS |
143 | void btrfs_test_inode_set_ops(struct inode *inode) | |
144 | { | |
145 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; | |
146 | } | |
147 | #endif | |
148 | ||
f34f57a3 | 149 | static int btrfs_init_inode_security(struct btrfs_trans_handle *trans, |
2a7dba39 EP |
150 | struct inode *inode, struct inode *dir, |
151 | const struct qstr *qstr) | |
0279b4cd JO |
152 | { |
153 | int err; | |
154 | ||
f34f57a3 | 155 | err = btrfs_init_acl(trans, inode, dir); |
0279b4cd | 156 | if (!err) |
2a7dba39 | 157 | err = btrfs_xattr_security_init(trans, inode, dir, qstr); |
0279b4cd JO |
158 | return err; |
159 | } | |
160 | ||
c8b97818 CM |
161 | /* |
162 | * this does all the hard work for inserting an inline extent into | |
163 | * the btree. The caller should have done a btrfs_drop_extents so that | |
164 | * no overlapping inline items exist in the btree | |
165 | */ | |
40f76580 | 166 | static int insert_inline_extent(struct btrfs_trans_handle *trans, |
1acae57b | 167 | struct btrfs_path *path, int extent_inserted, |
c8b97818 CM |
168 | struct btrfs_root *root, struct inode *inode, |
169 | u64 start, size_t size, size_t compressed_size, | |
fe3f566c | 170 | int compress_type, |
c8b97818 CM |
171 | struct page **compressed_pages) |
172 | { | |
c8b97818 CM |
173 | struct extent_buffer *leaf; |
174 | struct page *page = NULL; | |
175 | char *kaddr; | |
176 | unsigned long ptr; | |
177 | struct btrfs_file_extent_item *ei; | |
c8b97818 CM |
178 | int ret; |
179 | size_t cur_size = size; | |
c8b97818 | 180 | unsigned long offset; |
c8b97818 | 181 | |
982f1f5d JJB |
182 | ASSERT((compressed_size > 0 && compressed_pages) || |
183 | (compressed_size == 0 && !compressed_pages)); | |
184 | ||
fe3f566c | 185 | if (compressed_size && compressed_pages) |
c8b97818 | 186 | cur_size = compressed_size; |
c8b97818 | 187 | |
1acae57b | 188 | inode_add_bytes(inode, size); |
c8b97818 | 189 | |
1acae57b FDBM |
190 | if (!extent_inserted) { |
191 | struct btrfs_key key; | |
192 | size_t datasize; | |
c8b97818 | 193 | |
4a0cc7ca | 194 | key.objectid = btrfs_ino(BTRFS_I(inode)); |
1acae57b | 195 | key.offset = start; |
962a298f | 196 | key.type = BTRFS_EXTENT_DATA_KEY; |
c8b97818 | 197 | |
1acae57b FDBM |
198 | datasize = btrfs_file_extent_calc_inline_size(cur_size); |
199 | path->leave_spinning = 1; | |
200 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
201 | datasize); | |
79b4f4c6 | 202 | if (ret) |
1acae57b | 203 | goto fail; |
c8b97818 CM |
204 | } |
205 | leaf = path->nodes[0]; | |
206 | ei = btrfs_item_ptr(leaf, path->slots[0], | |
207 | struct btrfs_file_extent_item); | |
208 | btrfs_set_file_extent_generation(leaf, ei, trans->transid); | |
209 | btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE); | |
210 | btrfs_set_file_extent_encryption(leaf, ei, 0); | |
211 | btrfs_set_file_extent_other_encoding(leaf, ei, 0); | |
212 | btrfs_set_file_extent_ram_bytes(leaf, ei, size); | |
213 | ptr = btrfs_file_extent_inline_start(ei); | |
214 | ||
261507a0 | 215 | if (compress_type != BTRFS_COMPRESS_NONE) { |
c8b97818 CM |
216 | struct page *cpage; |
217 | int i = 0; | |
d397712b | 218 | while (compressed_size > 0) { |
c8b97818 | 219 | cpage = compressed_pages[i]; |
5b050f04 | 220 | cur_size = min_t(unsigned long, compressed_size, |
09cbfeaf | 221 | PAGE_SIZE); |
c8b97818 | 222 | |
7ac687d9 | 223 | kaddr = kmap_atomic(cpage); |
c8b97818 | 224 | write_extent_buffer(leaf, kaddr, ptr, cur_size); |
7ac687d9 | 225 | kunmap_atomic(kaddr); |
c8b97818 CM |
226 | |
227 | i++; | |
228 | ptr += cur_size; | |
229 | compressed_size -= cur_size; | |
230 | } | |
231 | btrfs_set_file_extent_compression(leaf, ei, | |
261507a0 | 232 | compress_type); |
c8b97818 CM |
233 | } else { |
234 | page = find_get_page(inode->i_mapping, | |
09cbfeaf | 235 | start >> PAGE_SHIFT); |
c8b97818 | 236 | btrfs_set_file_extent_compression(leaf, ei, 0); |
7ac687d9 | 237 | kaddr = kmap_atomic(page); |
7073017a | 238 | offset = offset_in_page(start); |
c8b97818 | 239 | write_extent_buffer(leaf, kaddr + offset, ptr, size); |
7ac687d9 | 240 | kunmap_atomic(kaddr); |
09cbfeaf | 241 | put_page(page); |
c8b97818 CM |
242 | } |
243 | btrfs_mark_buffer_dirty(leaf); | |
1acae57b | 244 | btrfs_release_path(path); |
c8b97818 | 245 | |
9ddc959e JB |
246 | /* |
247 | * We align size to sectorsize for inline extents just for simplicity | |
248 | * sake. | |
249 | */ | |
250 | size = ALIGN(size, root->fs_info->sectorsize); | |
251 | ret = btrfs_inode_set_file_extent_range(BTRFS_I(inode), start, size); | |
252 | if (ret) | |
253 | goto fail; | |
254 | ||
c2167754 YZ |
255 | /* |
256 | * we're an inline extent, so nobody can | |
257 | * extend the file past i_size without locking | |
258 | * a page we already have locked. | |
259 | * | |
260 | * We must do any isize and inode updates | |
261 | * before we unlock the pages. Otherwise we | |
262 | * could end up racing with unlink. | |
263 | */ | |
c8b97818 | 264 | BTRFS_I(inode)->disk_i_size = inode->i_size; |
79787eaa | 265 | ret = btrfs_update_inode(trans, root, inode); |
c2167754 | 266 | |
c8b97818 | 267 | fail: |
79b4f4c6 | 268 | return ret; |
c8b97818 CM |
269 | } |
270 | ||
271 | ||
272 | /* | |
273 | * conditionally insert an inline extent into the file. This | |
274 | * does the checks required to make sure the data is small enough | |
275 | * to fit as an inline extent. | |
276 | */ | |
a0349401 | 277 | static noinline int cow_file_range_inline(struct btrfs_inode *inode, u64 start, |
00361589 JB |
278 | u64 end, size_t compressed_size, |
279 | int compress_type, | |
280 | struct page **compressed_pages) | |
c8b97818 | 281 | { |
a0349401 | 282 | struct btrfs_root *root = inode->root; |
0b246afa | 283 | struct btrfs_fs_info *fs_info = root->fs_info; |
00361589 | 284 | struct btrfs_trans_handle *trans; |
a0349401 | 285 | u64 isize = i_size_read(&inode->vfs_inode); |
c8b97818 CM |
286 | u64 actual_end = min(end + 1, isize); |
287 | u64 inline_len = actual_end - start; | |
0b246afa | 288 | u64 aligned_end = ALIGN(end, fs_info->sectorsize); |
c8b97818 CM |
289 | u64 data_len = inline_len; |
290 | int ret; | |
1acae57b FDBM |
291 | struct btrfs_path *path; |
292 | int extent_inserted = 0; | |
293 | u32 extent_item_size; | |
c8b97818 CM |
294 | |
295 | if (compressed_size) | |
296 | data_len = compressed_size; | |
297 | ||
298 | if (start > 0 || | |
0b246afa JM |
299 | actual_end > fs_info->sectorsize || |
300 | data_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info) || | |
c8b97818 | 301 | (!compressed_size && |
0b246afa | 302 | (actual_end & (fs_info->sectorsize - 1)) == 0) || |
c8b97818 | 303 | end + 1 < isize || |
0b246afa | 304 | data_len > fs_info->max_inline) { |
c8b97818 CM |
305 | return 1; |
306 | } | |
307 | ||
1acae57b FDBM |
308 | path = btrfs_alloc_path(); |
309 | if (!path) | |
310 | return -ENOMEM; | |
311 | ||
00361589 | 312 | trans = btrfs_join_transaction(root); |
1acae57b FDBM |
313 | if (IS_ERR(trans)) { |
314 | btrfs_free_path(path); | |
00361589 | 315 | return PTR_ERR(trans); |
1acae57b | 316 | } |
a0349401 | 317 | trans->block_rsv = &inode->block_rsv; |
00361589 | 318 | |
1acae57b FDBM |
319 | if (compressed_size && compressed_pages) |
320 | extent_item_size = btrfs_file_extent_calc_inline_size( | |
321 | compressed_size); | |
322 | else | |
323 | extent_item_size = btrfs_file_extent_calc_inline_size( | |
324 | inline_len); | |
325 | ||
a0349401 NB |
326 | ret = __btrfs_drop_extents(trans, root, inode, path, start, aligned_end, |
327 | NULL, 1, 1, extent_item_size, | |
328 | &extent_inserted); | |
00361589 | 329 | if (ret) { |
66642832 | 330 | btrfs_abort_transaction(trans, ret); |
00361589 JB |
331 | goto out; |
332 | } | |
c8b97818 CM |
333 | |
334 | if (isize > actual_end) | |
335 | inline_len = min_t(u64, isize, actual_end); | |
1acae57b | 336 | ret = insert_inline_extent(trans, path, extent_inserted, |
a0349401 | 337 | root, &inode->vfs_inode, start, |
c8b97818 | 338 | inline_len, compressed_size, |
fe3f566c | 339 | compress_type, compressed_pages); |
2adcac1a | 340 | if (ret && ret != -ENOSPC) { |
66642832 | 341 | btrfs_abort_transaction(trans, ret); |
00361589 | 342 | goto out; |
2adcac1a | 343 | } else if (ret == -ENOSPC) { |
00361589 JB |
344 | ret = 1; |
345 | goto out; | |
79787eaa | 346 | } |
2adcac1a | 347 | |
a0349401 NB |
348 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &inode->runtime_flags); |
349 | btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0); | |
00361589 | 350 | out: |
94ed938a QW |
351 | /* |
352 | * Don't forget to free the reserved space, as for inlined extent | |
353 | * it won't count as data extent, free them directly here. | |
354 | * And at reserve time, it's always aligned to page size, so | |
355 | * just free one page here. | |
356 | */ | |
a0349401 | 357 | btrfs_qgroup_free_data(inode, NULL, 0, PAGE_SIZE); |
1acae57b | 358 | btrfs_free_path(path); |
3a45bb20 | 359 | btrfs_end_transaction(trans); |
00361589 | 360 | return ret; |
c8b97818 CM |
361 | } |
362 | ||
771ed689 CM |
363 | struct async_extent { |
364 | u64 start; | |
365 | u64 ram_size; | |
366 | u64 compressed_size; | |
367 | struct page **pages; | |
368 | unsigned long nr_pages; | |
261507a0 | 369 | int compress_type; |
771ed689 CM |
370 | struct list_head list; |
371 | }; | |
372 | ||
97db1204 | 373 | struct async_chunk { |
771ed689 | 374 | struct inode *inode; |
771ed689 CM |
375 | struct page *locked_page; |
376 | u64 start; | |
377 | u64 end; | |
f82b7359 | 378 | unsigned int write_flags; |
771ed689 | 379 | struct list_head extents; |
ec39f769 | 380 | struct cgroup_subsys_state *blkcg_css; |
771ed689 | 381 | struct btrfs_work work; |
97db1204 | 382 | atomic_t *pending; |
771ed689 CM |
383 | }; |
384 | ||
97db1204 NB |
385 | struct async_cow { |
386 | /* Number of chunks in flight; must be first in the structure */ | |
387 | atomic_t num_chunks; | |
388 | struct async_chunk chunks[]; | |
771ed689 CM |
389 | }; |
390 | ||
97db1204 | 391 | static noinline int add_async_extent(struct async_chunk *cow, |
771ed689 CM |
392 | u64 start, u64 ram_size, |
393 | u64 compressed_size, | |
394 | struct page **pages, | |
261507a0 LZ |
395 | unsigned long nr_pages, |
396 | int compress_type) | |
771ed689 CM |
397 | { |
398 | struct async_extent *async_extent; | |
399 | ||
400 | async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS); | |
79787eaa | 401 | BUG_ON(!async_extent); /* -ENOMEM */ |
771ed689 CM |
402 | async_extent->start = start; |
403 | async_extent->ram_size = ram_size; | |
404 | async_extent->compressed_size = compressed_size; | |
405 | async_extent->pages = pages; | |
406 | async_extent->nr_pages = nr_pages; | |
261507a0 | 407 | async_extent->compress_type = compress_type; |
771ed689 CM |
408 | list_add_tail(&async_extent->list, &cow->extents); |
409 | return 0; | |
410 | } | |
411 | ||
42c16da6 QW |
412 | /* |
413 | * Check if the inode has flags compatible with compression | |
414 | */ | |
99c88dc7 | 415 | static inline bool inode_can_compress(struct btrfs_inode *inode) |
42c16da6 | 416 | { |
99c88dc7 NB |
417 | if (inode->flags & BTRFS_INODE_NODATACOW || |
418 | inode->flags & BTRFS_INODE_NODATASUM) | |
42c16da6 QW |
419 | return false; |
420 | return true; | |
421 | } | |
422 | ||
423 | /* | |
424 | * Check if the inode needs to be submitted to compression, based on mount | |
425 | * options, defragmentation, properties or heuristics. | |
426 | */ | |
808a1292 NB |
427 | static inline int inode_need_compress(struct btrfs_inode *inode, u64 start, |
428 | u64 end) | |
f79707b0 | 429 | { |
808a1292 | 430 | struct btrfs_fs_info *fs_info = inode->root->fs_info; |
f79707b0 | 431 | |
808a1292 | 432 | if (!inode_can_compress(inode)) { |
42c16da6 QW |
433 | WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG), |
434 | KERN_ERR "BTRFS: unexpected compression for ino %llu\n", | |
808a1292 | 435 | btrfs_ino(inode)); |
42c16da6 QW |
436 | return 0; |
437 | } | |
f79707b0 | 438 | /* force compress */ |
0b246afa | 439 | if (btrfs_test_opt(fs_info, FORCE_COMPRESS)) |
f79707b0 | 440 | return 1; |
eec63c65 | 441 | /* defrag ioctl */ |
808a1292 | 442 | if (inode->defrag_compress) |
eec63c65 | 443 | return 1; |
f79707b0 | 444 | /* bad compression ratios */ |
808a1292 | 445 | if (inode->flags & BTRFS_INODE_NOCOMPRESS) |
f79707b0 | 446 | return 0; |
0b246afa | 447 | if (btrfs_test_opt(fs_info, COMPRESS) || |
808a1292 NB |
448 | inode->flags & BTRFS_INODE_COMPRESS || |
449 | inode->prop_compress) | |
450 | return btrfs_compress_heuristic(&inode->vfs_inode, start, end); | |
f79707b0 WS |
451 | return 0; |
452 | } | |
453 | ||
6158e1ce | 454 | static inline void inode_should_defrag(struct btrfs_inode *inode, |
26d30f85 AJ |
455 | u64 start, u64 end, u64 num_bytes, u64 small_write) |
456 | { | |
457 | /* If this is a small write inside eof, kick off a defrag */ | |
458 | if (num_bytes < small_write && | |
6158e1ce | 459 | (start > 0 || end + 1 < inode->disk_i_size)) |
26d30f85 AJ |
460 | btrfs_add_inode_defrag(NULL, inode); |
461 | } | |
462 | ||
d352ac68 | 463 | /* |
771ed689 CM |
464 | * we create compressed extents in two phases. The first |
465 | * phase compresses a range of pages that have already been | |
466 | * locked (both pages and state bits are locked). | |
c8b97818 | 467 | * |
771ed689 CM |
468 | * This is done inside an ordered work queue, and the compression |
469 | * is spread across many cpus. The actual IO submission is step | |
470 | * two, and the ordered work queue takes care of making sure that | |
471 | * happens in the same order things were put onto the queue by | |
472 | * writepages and friends. | |
c8b97818 | 473 | * |
771ed689 CM |
474 | * If this code finds it can't get good compression, it puts an |
475 | * entry onto the work queue to write the uncompressed bytes. This | |
476 | * makes sure that both compressed inodes and uncompressed inodes | |
b2570314 AB |
477 | * are written in the same order that the flusher thread sent them |
478 | * down. | |
d352ac68 | 479 | */ |
ac3e9933 | 480 | static noinline int compress_file_range(struct async_chunk *async_chunk) |
b888db2b | 481 | { |
1368c6da | 482 | struct inode *inode = async_chunk->inode; |
0b246afa | 483 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
0b246afa | 484 | u64 blocksize = fs_info->sectorsize; |
1368c6da NB |
485 | u64 start = async_chunk->start; |
486 | u64 end = async_chunk->end; | |
c8b97818 | 487 | u64 actual_end; |
d98da499 | 488 | u64 i_size; |
e6dcd2dc | 489 | int ret = 0; |
c8b97818 CM |
490 | struct page **pages = NULL; |
491 | unsigned long nr_pages; | |
c8b97818 CM |
492 | unsigned long total_compressed = 0; |
493 | unsigned long total_in = 0; | |
c8b97818 CM |
494 | int i; |
495 | int will_compress; | |
0b246afa | 496 | int compress_type = fs_info->compress_type; |
ac3e9933 | 497 | int compressed_extents = 0; |
4adaa611 | 498 | int redirty = 0; |
b888db2b | 499 | |
6158e1ce NB |
500 | inode_should_defrag(BTRFS_I(inode), start, end, end - start + 1, |
501 | SZ_16K); | |
4cb5300b | 502 | |
d98da499 JB |
503 | /* |
504 | * We need to save i_size before now because it could change in between | |
505 | * us evaluating the size and assigning it. This is because we lock and | |
506 | * unlock the page in truncate and fallocate, and then modify the i_size | |
507 | * later on. | |
508 | * | |
509 | * The barriers are to emulate READ_ONCE, remove that once i_size_read | |
510 | * does that for us. | |
511 | */ | |
512 | barrier(); | |
513 | i_size = i_size_read(inode); | |
514 | barrier(); | |
515 | actual_end = min_t(u64, i_size, end + 1); | |
c8b97818 CM |
516 | again: |
517 | will_compress = 0; | |
09cbfeaf | 518 | nr_pages = (end >> PAGE_SHIFT) - (start >> PAGE_SHIFT) + 1; |
069eac78 DS |
519 | BUILD_BUG_ON((BTRFS_MAX_COMPRESSED % PAGE_SIZE) != 0); |
520 | nr_pages = min_t(unsigned long, nr_pages, | |
521 | BTRFS_MAX_COMPRESSED / PAGE_SIZE); | |
be20aa9d | 522 | |
f03d9301 CM |
523 | /* |
524 | * we don't want to send crud past the end of i_size through | |
525 | * compression, that's just a waste of CPU time. So, if the | |
526 | * end of the file is before the start of our current | |
527 | * requested range of bytes, we bail out to the uncompressed | |
528 | * cleanup code that can deal with all of this. | |
529 | * | |
530 | * It isn't really the fastest way to fix things, but this is a | |
531 | * very uncommon corner. | |
532 | */ | |
533 | if (actual_end <= start) | |
534 | goto cleanup_and_bail_uncompressed; | |
535 | ||
c8b97818 CM |
536 | total_compressed = actual_end - start; |
537 | ||
4bcbb332 SW |
538 | /* |
539 | * skip compression for a small file range(<=blocksize) that | |
01327610 | 540 | * isn't an inline extent, since it doesn't save disk space at all. |
4bcbb332 SW |
541 | */ |
542 | if (total_compressed <= blocksize && | |
543 | (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size)) | |
544 | goto cleanup_and_bail_uncompressed; | |
545 | ||
069eac78 DS |
546 | total_compressed = min_t(unsigned long, total_compressed, |
547 | BTRFS_MAX_UNCOMPRESSED); | |
c8b97818 CM |
548 | total_in = 0; |
549 | ret = 0; | |
db94535d | 550 | |
771ed689 CM |
551 | /* |
552 | * we do compression for mount -o compress and when the | |
553 | * inode has not been flagged as nocompress. This flag can | |
554 | * change at any time if we discover bad compression ratios. | |
c8b97818 | 555 | */ |
808a1292 | 556 | if (inode_need_compress(BTRFS_I(inode), start, end)) { |
c8b97818 | 557 | WARN_ON(pages); |
31e818fe | 558 | pages = kcalloc(nr_pages, sizeof(struct page *), GFP_NOFS); |
560f7d75 LZ |
559 | if (!pages) { |
560 | /* just bail out to the uncompressed code */ | |
3527a018 | 561 | nr_pages = 0; |
560f7d75 LZ |
562 | goto cont; |
563 | } | |
c8b97818 | 564 | |
eec63c65 DS |
565 | if (BTRFS_I(inode)->defrag_compress) |
566 | compress_type = BTRFS_I(inode)->defrag_compress; | |
567 | else if (BTRFS_I(inode)->prop_compress) | |
b52aa8c9 | 568 | compress_type = BTRFS_I(inode)->prop_compress; |
261507a0 | 569 | |
4adaa611 CM |
570 | /* |
571 | * we need to call clear_page_dirty_for_io on each | |
572 | * page in the range. Otherwise applications with the file | |
573 | * mmap'd can wander in and change the page contents while | |
574 | * we are compressing them. | |
575 | * | |
576 | * If the compression fails for any reason, we set the pages | |
577 | * dirty again later on. | |
e9679de3 TT |
578 | * |
579 | * Note that the remaining part is redirtied, the start pointer | |
580 | * has moved, the end is the original one. | |
4adaa611 | 581 | */ |
e9679de3 TT |
582 | if (!redirty) { |
583 | extent_range_clear_dirty_for_io(inode, start, end); | |
584 | redirty = 1; | |
585 | } | |
f51d2b59 DS |
586 | |
587 | /* Compression level is applied here and only here */ | |
588 | ret = btrfs_compress_pages( | |
589 | compress_type | (fs_info->compress_level << 4), | |
261507a0 | 590 | inode->i_mapping, start, |
38c31464 | 591 | pages, |
4d3a800e | 592 | &nr_pages, |
261507a0 | 593 | &total_in, |
e5d74902 | 594 | &total_compressed); |
c8b97818 CM |
595 | |
596 | if (!ret) { | |
7073017a | 597 | unsigned long offset = offset_in_page(total_compressed); |
4d3a800e | 598 | struct page *page = pages[nr_pages - 1]; |
c8b97818 CM |
599 | char *kaddr; |
600 | ||
601 | /* zero the tail end of the last page, we might be | |
602 | * sending it down to disk | |
603 | */ | |
604 | if (offset) { | |
7ac687d9 | 605 | kaddr = kmap_atomic(page); |
c8b97818 | 606 | memset(kaddr + offset, 0, |
09cbfeaf | 607 | PAGE_SIZE - offset); |
7ac687d9 | 608 | kunmap_atomic(kaddr); |
c8b97818 CM |
609 | } |
610 | will_compress = 1; | |
611 | } | |
612 | } | |
560f7d75 | 613 | cont: |
c8b97818 CM |
614 | if (start == 0) { |
615 | /* lets try to make an inline extent */ | |
6018ba0a | 616 | if (ret || total_in < actual_end) { |
c8b97818 | 617 | /* we didn't compress the entire range, try |
771ed689 | 618 | * to make an uncompressed inline extent. |
c8b97818 | 619 | */ |
a0349401 NB |
620 | ret = cow_file_range_inline(BTRFS_I(inode), start, end, |
621 | 0, BTRFS_COMPRESS_NONE, | |
622 | NULL); | |
c8b97818 | 623 | } else { |
771ed689 | 624 | /* try making a compressed inline extent */ |
a0349401 | 625 | ret = cow_file_range_inline(BTRFS_I(inode), start, end, |
fe3f566c LZ |
626 | total_compressed, |
627 | compress_type, pages); | |
c8b97818 | 628 | } |
79787eaa | 629 | if (ret <= 0) { |
151a41bc | 630 | unsigned long clear_flags = EXTENT_DELALLOC | |
8b62f87b JB |
631 | EXTENT_DELALLOC_NEW | EXTENT_DEFRAG | |
632 | EXTENT_DO_ACCOUNTING; | |
e6eb4314 FM |
633 | unsigned long page_error_op; |
634 | ||
e6eb4314 | 635 | page_error_op = ret < 0 ? PAGE_SET_ERROR : 0; |
151a41bc | 636 | |
771ed689 | 637 | /* |
79787eaa JM |
638 | * inline extent creation worked or returned error, |
639 | * we don't need to create any more async work items. | |
640 | * Unlock and free up our temp pages. | |
8b62f87b JB |
641 | * |
642 | * We use DO_ACCOUNTING here because we need the | |
643 | * delalloc_release_metadata to be done _after_ we drop | |
644 | * our outstanding extent for clearing delalloc for this | |
645 | * range. | |
771ed689 | 646 | */ |
ad7ff17b NB |
647 | extent_clear_unlock_delalloc(BTRFS_I(inode), start, end, |
648 | NULL, | |
74e9194a | 649 | clear_flags, |
ba8b04c1 | 650 | PAGE_UNLOCK | |
c2790a2e JB |
651 | PAGE_CLEAR_DIRTY | |
652 | PAGE_SET_WRITEBACK | | |
e6eb4314 | 653 | page_error_op | |
c2790a2e | 654 | PAGE_END_WRITEBACK); |
cecc8d90 NB |
655 | |
656 | for (i = 0; i < nr_pages; i++) { | |
657 | WARN_ON(pages[i]->mapping); | |
658 | put_page(pages[i]); | |
659 | } | |
660 | kfree(pages); | |
661 | ||
662 | return 0; | |
c8b97818 CM |
663 | } |
664 | } | |
665 | ||
666 | if (will_compress) { | |
667 | /* | |
668 | * we aren't doing an inline extent round the compressed size | |
669 | * up to a block size boundary so the allocator does sane | |
670 | * things | |
671 | */ | |
fda2832f | 672 | total_compressed = ALIGN(total_compressed, blocksize); |
c8b97818 CM |
673 | |
674 | /* | |
675 | * one last check to make sure the compression is really a | |
170607eb TT |
676 | * win, compare the page count read with the blocks on disk, |
677 | * compression must free at least one sector size | |
c8b97818 | 678 | */ |
09cbfeaf | 679 | total_in = ALIGN(total_in, PAGE_SIZE); |
170607eb | 680 | if (total_compressed + blocksize <= total_in) { |
ac3e9933 | 681 | compressed_extents++; |
c8bb0c8b AS |
682 | |
683 | /* | |
684 | * The async work queues will take care of doing actual | |
685 | * allocation on disk for these compressed pages, and | |
686 | * will submit them to the elevator. | |
687 | */ | |
b5326271 | 688 | add_async_extent(async_chunk, start, total_in, |
4d3a800e | 689 | total_compressed, pages, nr_pages, |
c8bb0c8b AS |
690 | compress_type); |
691 | ||
1170862d TT |
692 | if (start + total_in < end) { |
693 | start += total_in; | |
c8bb0c8b AS |
694 | pages = NULL; |
695 | cond_resched(); | |
696 | goto again; | |
697 | } | |
ac3e9933 | 698 | return compressed_extents; |
c8b97818 CM |
699 | } |
700 | } | |
c8bb0c8b | 701 | if (pages) { |
c8b97818 CM |
702 | /* |
703 | * the compression code ran but failed to make things smaller, | |
704 | * free any pages it allocated and our page pointer array | |
705 | */ | |
4d3a800e | 706 | for (i = 0; i < nr_pages; i++) { |
70b99e69 | 707 | WARN_ON(pages[i]->mapping); |
09cbfeaf | 708 | put_page(pages[i]); |
c8b97818 CM |
709 | } |
710 | kfree(pages); | |
711 | pages = NULL; | |
712 | total_compressed = 0; | |
4d3a800e | 713 | nr_pages = 0; |
c8b97818 CM |
714 | |
715 | /* flag the file so we don't compress in the future */ | |
0b246afa | 716 | if (!btrfs_test_opt(fs_info, FORCE_COMPRESS) && |
b52aa8c9 | 717 | !(BTRFS_I(inode)->prop_compress)) { |
a555f810 | 718 | BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS; |
1e701a32 | 719 | } |
c8b97818 | 720 | } |
f03d9301 | 721 | cleanup_and_bail_uncompressed: |
c8bb0c8b AS |
722 | /* |
723 | * No compression, but we still need to write the pages in the file | |
724 | * we've been given so far. redirty the locked page if it corresponds | |
725 | * to our extent and set things up for the async work queue to run | |
726 | * cow_file_range to do the normal delalloc dance. | |
727 | */ | |
1d53c9e6 CM |
728 | if (async_chunk->locked_page && |
729 | (page_offset(async_chunk->locked_page) >= start && | |
730 | page_offset(async_chunk->locked_page)) <= end) { | |
1368c6da | 731 | __set_page_dirty_nobuffers(async_chunk->locked_page); |
c8bb0c8b | 732 | /* unlocked later on in the async handlers */ |
1d53c9e6 | 733 | } |
c8bb0c8b AS |
734 | |
735 | if (redirty) | |
736 | extent_range_redirty_for_io(inode, start, end); | |
b5326271 | 737 | add_async_extent(async_chunk, start, end - start + 1, 0, NULL, 0, |
c8bb0c8b | 738 | BTRFS_COMPRESS_NONE); |
ac3e9933 | 739 | compressed_extents++; |
3b951516 | 740 | |
ac3e9933 | 741 | return compressed_extents; |
771ed689 | 742 | } |
771ed689 | 743 | |
40ae837b FM |
744 | static void free_async_extent_pages(struct async_extent *async_extent) |
745 | { | |
746 | int i; | |
747 | ||
748 | if (!async_extent->pages) | |
749 | return; | |
750 | ||
751 | for (i = 0; i < async_extent->nr_pages; i++) { | |
752 | WARN_ON(async_extent->pages[i]->mapping); | |
09cbfeaf | 753 | put_page(async_extent->pages[i]); |
40ae837b FM |
754 | } |
755 | kfree(async_extent->pages); | |
756 | async_extent->nr_pages = 0; | |
757 | async_extent->pages = NULL; | |
771ed689 CM |
758 | } |
759 | ||
760 | /* | |
761 | * phase two of compressed writeback. This is the ordered portion | |
762 | * of the code, which only gets called in the order the work was | |
763 | * queued. We walk all the async extents created by compress_file_range | |
764 | * and send them down to the disk. | |
765 | */ | |
b5326271 | 766 | static noinline void submit_compressed_extents(struct async_chunk *async_chunk) |
771ed689 | 767 | { |
a0ff10dc NB |
768 | struct btrfs_inode *inode = BTRFS_I(async_chunk->inode); |
769 | struct btrfs_fs_info *fs_info = inode->root->fs_info; | |
771ed689 CM |
770 | struct async_extent *async_extent; |
771 | u64 alloc_hint = 0; | |
771ed689 CM |
772 | struct btrfs_key ins; |
773 | struct extent_map *em; | |
a0ff10dc NB |
774 | struct btrfs_root *root = inode->root; |
775 | struct extent_io_tree *io_tree = &inode->io_tree; | |
f5a84ee3 | 776 | int ret = 0; |
771ed689 | 777 | |
3e04e7f1 | 778 | again: |
b5326271 NB |
779 | while (!list_empty(&async_chunk->extents)) { |
780 | async_extent = list_entry(async_chunk->extents.next, | |
771ed689 CM |
781 | struct async_extent, list); |
782 | list_del(&async_extent->list); | |
c8b97818 | 783 | |
f5a84ee3 | 784 | retry: |
7447555f NB |
785 | lock_extent(io_tree, async_extent->start, |
786 | async_extent->start + async_extent->ram_size - 1); | |
771ed689 CM |
787 | /* did the compression code fall back to uncompressed IO? */ |
788 | if (!async_extent->pages) { | |
789 | int page_started = 0; | |
790 | unsigned long nr_written = 0; | |
791 | ||
771ed689 | 792 | /* allocate blocks */ |
a0ff10dc | 793 | ret = cow_file_range(inode, async_chunk->locked_page, |
f5a84ee3 JB |
794 | async_extent->start, |
795 | async_extent->start + | |
796 | async_extent->ram_size - 1, | |
330a5827 | 797 | &page_started, &nr_written, 0); |
771ed689 | 798 | |
79787eaa JM |
799 | /* JDM XXX */ |
800 | ||
771ed689 CM |
801 | /* |
802 | * if page_started, cow_file_range inserted an | |
803 | * inline extent and took care of all the unlocking | |
804 | * and IO for us. Otherwise, we need to submit | |
805 | * all those pages down to the drive. | |
806 | */ | |
f5a84ee3 | 807 | if (!page_started && !ret) |
a0ff10dc | 808 | extent_write_locked_range(&inode->vfs_inode, |
5e3ee236 | 809 | async_extent->start, |
d397712b | 810 | async_extent->start + |
771ed689 | 811 | async_extent->ram_size - 1, |
771ed689 | 812 | WB_SYNC_ALL); |
1d53c9e6 | 813 | else if (ret && async_chunk->locked_page) |
b5326271 | 814 | unlock_page(async_chunk->locked_page); |
771ed689 CM |
815 | kfree(async_extent); |
816 | cond_resched(); | |
817 | continue; | |
818 | } | |
819 | ||
18513091 | 820 | ret = btrfs_reserve_extent(root, async_extent->ram_size, |
771ed689 CM |
821 | async_extent->compressed_size, |
822 | async_extent->compressed_size, | |
e570fd27 | 823 | 0, alloc_hint, &ins, 1, 1); |
f5a84ee3 | 824 | if (ret) { |
40ae837b | 825 | free_async_extent_pages(async_extent); |
3e04e7f1 | 826 | |
fdf8e2ea JB |
827 | if (ret == -ENOSPC) { |
828 | unlock_extent(io_tree, async_extent->start, | |
829 | async_extent->start + | |
830 | async_extent->ram_size - 1); | |
ce62003f LB |
831 | |
832 | /* | |
833 | * we need to redirty the pages if we decide to | |
834 | * fallback to uncompressed IO, otherwise we | |
835 | * will not submit these pages down to lower | |
836 | * layers. | |
837 | */ | |
a0ff10dc | 838 | extent_range_redirty_for_io(&inode->vfs_inode, |
ce62003f LB |
839 | async_extent->start, |
840 | async_extent->start + | |
841 | async_extent->ram_size - 1); | |
842 | ||
79787eaa | 843 | goto retry; |
fdf8e2ea | 844 | } |
3e04e7f1 | 845 | goto out_free; |
f5a84ee3 | 846 | } |
c2167754 YZ |
847 | /* |
848 | * here we're doing allocation and writeback of the | |
849 | * compressed pages | |
850 | */ | |
a0ff10dc | 851 | em = create_io_em(inode, async_extent->start, |
6f9994db LB |
852 | async_extent->ram_size, /* len */ |
853 | async_extent->start, /* orig_start */ | |
854 | ins.objectid, /* block_start */ | |
855 | ins.offset, /* block_len */ | |
856 | ins.offset, /* orig_block_len */ | |
857 | async_extent->ram_size, /* ram_bytes */ | |
858 | async_extent->compress_type, | |
859 | BTRFS_ORDERED_COMPRESSED); | |
860 | if (IS_ERR(em)) | |
861 | /* ret value is not necessary due to void function */ | |
3e04e7f1 | 862 | goto out_free_reserve; |
6f9994db | 863 | free_extent_map(em); |
3e04e7f1 | 864 | |
a0ff10dc | 865 | ret = btrfs_add_ordered_extent_compress(inode, |
261507a0 LZ |
866 | async_extent->start, |
867 | ins.objectid, | |
868 | async_extent->ram_size, | |
869 | ins.offset, | |
870 | BTRFS_ORDERED_COMPRESSED, | |
871 | async_extent->compress_type); | |
d9f85963 | 872 | if (ret) { |
a0ff10dc | 873 | btrfs_drop_extent_cache(inode, async_extent->start, |
d9f85963 FM |
874 | async_extent->start + |
875 | async_extent->ram_size - 1, 0); | |
3e04e7f1 | 876 | goto out_free_reserve; |
d9f85963 | 877 | } |
0b246afa | 878 | btrfs_dec_block_group_reservations(fs_info, ins.objectid); |
771ed689 | 879 | |
771ed689 CM |
880 | /* |
881 | * clear dirty, set writeback and unlock the pages. | |
882 | */ | |
a0ff10dc | 883 | extent_clear_unlock_delalloc(inode, async_extent->start, |
a791e35e CM |
884 | async_extent->start + |
885 | async_extent->ram_size - 1, | |
151a41bc JB |
886 | NULL, EXTENT_LOCKED | EXTENT_DELALLOC, |
887 | PAGE_UNLOCK | PAGE_CLEAR_DIRTY | | |
c2790a2e | 888 | PAGE_SET_WRITEBACK); |
a0ff10dc | 889 | if (btrfs_submit_compressed_write(inode, async_extent->start, |
d397712b CM |
890 | async_extent->ram_size, |
891 | ins.objectid, | |
892 | ins.offset, async_extent->pages, | |
f82b7359 | 893 | async_extent->nr_pages, |
ec39f769 CM |
894 | async_chunk->write_flags, |
895 | async_chunk->blkcg_css)) { | |
fce2a4e6 FM |
896 | struct page *p = async_extent->pages[0]; |
897 | const u64 start = async_extent->start; | |
898 | const u64 end = start + async_extent->ram_size - 1; | |
899 | ||
a0ff10dc | 900 | p->mapping = inode->vfs_inode.i_mapping; |
c629732d | 901 | btrfs_writepage_endio_finish_ordered(p, start, end, 0); |
7087a9d8 | 902 | |
fce2a4e6 | 903 | p->mapping = NULL; |
a0ff10dc | 904 | extent_clear_unlock_delalloc(inode, start, end, NULL, 0, |
fce2a4e6 FM |
905 | PAGE_END_WRITEBACK | |
906 | PAGE_SET_ERROR); | |
40ae837b | 907 | free_async_extent_pages(async_extent); |
fce2a4e6 | 908 | } |
771ed689 CM |
909 | alloc_hint = ins.objectid + ins.offset; |
910 | kfree(async_extent); | |
911 | cond_resched(); | |
912 | } | |
dec8f175 | 913 | return; |
3e04e7f1 | 914 | out_free_reserve: |
0b246afa | 915 | btrfs_dec_block_group_reservations(fs_info, ins.objectid); |
2ff7e61e | 916 | btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1); |
79787eaa | 917 | out_free: |
a0ff10dc | 918 | extent_clear_unlock_delalloc(inode, async_extent->start, |
3e04e7f1 JB |
919 | async_extent->start + |
920 | async_extent->ram_size - 1, | |
c2790a2e | 921 | NULL, EXTENT_LOCKED | EXTENT_DELALLOC | |
a7e3b975 | 922 | EXTENT_DELALLOC_NEW | |
151a41bc JB |
923 | EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING, |
924 | PAGE_UNLOCK | PAGE_CLEAR_DIRTY | | |
704de49d FM |
925 | PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK | |
926 | PAGE_SET_ERROR); | |
40ae837b | 927 | free_async_extent_pages(async_extent); |
79787eaa | 928 | kfree(async_extent); |
3e04e7f1 | 929 | goto again; |
771ed689 CM |
930 | } |
931 | ||
43c69849 | 932 | static u64 get_extent_allocation_hint(struct btrfs_inode *inode, u64 start, |
4b46fce2 JB |
933 | u64 num_bytes) |
934 | { | |
43c69849 | 935 | struct extent_map_tree *em_tree = &inode->extent_tree; |
4b46fce2 JB |
936 | struct extent_map *em; |
937 | u64 alloc_hint = 0; | |
938 | ||
939 | read_lock(&em_tree->lock); | |
940 | em = search_extent_mapping(em_tree, start, num_bytes); | |
941 | if (em) { | |
942 | /* | |
943 | * if block start isn't an actual block number then find the | |
944 | * first block in this inode and use that as a hint. If that | |
945 | * block is also bogus then just don't worry about it. | |
946 | */ | |
947 | if (em->block_start >= EXTENT_MAP_LAST_BYTE) { | |
948 | free_extent_map(em); | |
949 | em = search_extent_mapping(em_tree, 0, 0); | |
950 | if (em && em->block_start < EXTENT_MAP_LAST_BYTE) | |
951 | alloc_hint = em->block_start; | |
952 | if (em) | |
953 | free_extent_map(em); | |
954 | } else { | |
955 | alloc_hint = em->block_start; | |
956 | free_extent_map(em); | |
957 | } | |
958 | } | |
959 | read_unlock(&em_tree->lock); | |
960 | ||
961 | return alloc_hint; | |
962 | } | |
963 | ||
771ed689 CM |
964 | /* |
965 | * when extent_io.c finds a delayed allocation range in the file, | |
966 | * the call backs end up in this code. The basic idea is to | |
967 | * allocate extents on disk for the range, and create ordered data structs | |
968 | * in ram to track those extents. | |
969 | * | |
970 | * locked_page is the page that writepage had locked already. We use | |
971 | * it to make sure we don't do extra locks or unlocks. | |
972 | * | |
973 | * *page_started is set to one if we unlock locked_page and do everything | |
974 | * required to start IO on it. It may be clean and already done with | |
975 | * IO when we return. | |
976 | */ | |
6e26c442 | 977 | static noinline int cow_file_range(struct btrfs_inode *inode, |
00361589 | 978 | struct page *locked_page, |
74e9194a | 979 | u64 start, u64 end, int *page_started, |
330a5827 | 980 | unsigned long *nr_written, int unlock) |
771ed689 | 981 | { |
6e26c442 NB |
982 | struct btrfs_root *root = inode->root; |
983 | struct btrfs_fs_info *fs_info = root->fs_info; | |
771ed689 CM |
984 | u64 alloc_hint = 0; |
985 | u64 num_bytes; | |
986 | unsigned long ram_size; | |
a315e68f | 987 | u64 cur_alloc_size = 0; |
432cd2a1 | 988 | u64 min_alloc_size; |
0b246afa | 989 | u64 blocksize = fs_info->sectorsize; |
771ed689 CM |
990 | struct btrfs_key ins; |
991 | struct extent_map *em; | |
a315e68f FM |
992 | unsigned clear_bits; |
993 | unsigned long page_ops; | |
994 | bool extent_reserved = false; | |
771ed689 CM |
995 | int ret = 0; |
996 | ||
6e26c442 | 997 | if (btrfs_is_free_space_inode(inode)) { |
02ecd2c2 | 998 | WARN_ON_ONCE(1); |
29bce2f3 JB |
999 | ret = -EINVAL; |
1000 | goto out_unlock; | |
02ecd2c2 | 1001 | } |
771ed689 | 1002 | |
fda2832f | 1003 | num_bytes = ALIGN(end - start + 1, blocksize); |
771ed689 | 1004 | num_bytes = max(blocksize, num_bytes); |
566b1760 | 1005 | ASSERT(num_bytes <= btrfs_super_total_bytes(fs_info->super_copy)); |
771ed689 | 1006 | |
6e26c442 | 1007 | inode_should_defrag(inode, start, end, num_bytes, SZ_64K); |
4cb5300b | 1008 | |
771ed689 CM |
1009 | if (start == 0) { |
1010 | /* lets try to make an inline extent */ | |
6e26c442 | 1011 | ret = cow_file_range_inline(inode, start, end, 0, |
d02c0e20 | 1012 | BTRFS_COMPRESS_NONE, NULL); |
771ed689 | 1013 | if (ret == 0) { |
8b62f87b JB |
1014 | /* |
1015 | * We use DO_ACCOUNTING here because we need the | |
1016 | * delalloc_release_metadata to be run _after_ we drop | |
1017 | * our outstanding extent for clearing delalloc for this | |
1018 | * range. | |
1019 | */ | |
6e26c442 | 1020 | extent_clear_unlock_delalloc(inode, start, end, NULL, |
c2790a2e | 1021 | EXTENT_LOCKED | EXTENT_DELALLOC | |
8b62f87b JB |
1022 | EXTENT_DELALLOC_NEW | EXTENT_DEFRAG | |
1023 | EXTENT_DO_ACCOUNTING, PAGE_UNLOCK | | |
c2790a2e JB |
1024 | PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK | |
1025 | PAGE_END_WRITEBACK); | |
771ed689 | 1026 | *nr_written = *nr_written + |
09cbfeaf | 1027 | (end - start + PAGE_SIZE) / PAGE_SIZE; |
771ed689 | 1028 | *page_started = 1; |
771ed689 | 1029 | goto out; |
79787eaa | 1030 | } else if (ret < 0) { |
79787eaa | 1031 | goto out_unlock; |
771ed689 CM |
1032 | } |
1033 | } | |
1034 | ||
6e26c442 NB |
1035 | alloc_hint = get_extent_allocation_hint(inode, start, num_bytes); |
1036 | btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0); | |
771ed689 | 1037 | |
432cd2a1 FM |
1038 | /* |
1039 | * Relocation relies on the relocated extents to have exactly the same | |
1040 | * size as the original extents. Normally writeback for relocation data | |
1041 | * extents follows a NOCOW path because relocation preallocates the | |
1042 | * extents. However, due to an operation such as scrub turning a block | |
1043 | * group to RO mode, it may fallback to COW mode, so we must make sure | |
1044 | * an extent allocated during COW has exactly the requested size and can | |
1045 | * not be split into smaller extents, otherwise relocation breaks and | |
1046 | * fails during the stage where it updates the bytenr of file extent | |
1047 | * items. | |
1048 | */ | |
1049 | if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID) | |
1050 | min_alloc_size = num_bytes; | |
1051 | else | |
1052 | min_alloc_size = fs_info->sectorsize; | |
1053 | ||
3752d22f AJ |
1054 | while (num_bytes > 0) { |
1055 | cur_alloc_size = num_bytes; | |
18513091 | 1056 | ret = btrfs_reserve_extent(root, cur_alloc_size, cur_alloc_size, |
432cd2a1 | 1057 | min_alloc_size, 0, alloc_hint, |
e570fd27 | 1058 | &ins, 1, 1); |
00361589 | 1059 | if (ret < 0) |
79787eaa | 1060 | goto out_unlock; |
a315e68f FM |
1061 | cur_alloc_size = ins.offset; |
1062 | extent_reserved = true; | |
d397712b | 1063 | |
771ed689 | 1064 | ram_size = ins.offset; |
6e26c442 | 1065 | em = create_io_em(inode, start, ins.offset, /* len */ |
6f9994db LB |
1066 | start, /* orig_start */ |
1067 | ins.objectid, /* block_start */ | |
1068 | ins.offset, /* block_len */ | |
1069 | ins.offset, /* orig_block_len */ | |
1070 | ram_size, /* ram_bytes */ | |
1071 | BTRFS_COMPRESS_NONE, /* compress_type */ | |
1af4a0aa | 1072 | BTRFS_ORDERED_REGULAR /* type */); |
090a127a SY |
1073 | if (IS_ERR(em)) { |
1074 | ret = PTR_ERR(em); | |
ace68bac | 1075 | goto out_reserve; |
090a127a | 1076 | } |
6f9994db | 1077 | free_extent_map(em); |
e6dcd2dc | 1078 | |
6e26c442 NB |
1079 | ret = btrfs_add_ordered_extent(inode, start, ins.objectid, |
1080 | ram_size, cur_alloc_size, 0); | |
ace68bac | 1081 | if (ret) |
d9f85963 | 1082 | goto out_drop_extent_cache; |
c8b97818 | 1083 | |
17d217fe YZ |
1084 | if (root->root_key.objectid == |
1085 | BTRFS_DATA_RELOC_TREE_OBJECTID) { | |
6e26c442 | 1086 | ret = btrfs_reloc_clone_csums(inode, start, |
17d217fe | 1087 | cur_alloc_size); |
4dbd80fb QW |
1088 | /* |
1089 | * Only drop cache here, and process as normal. | |
1090 | * | |
1091 | * We must not allow extent_clear_unlock_delalloc() | |
1092 | * at out_unlock label to free meta of this ordered | |
1093 | * extent, as its meta should be freed by | |
1094 | * btrfs_finish_ordered_io(). | |
1095 | * | |
1096 | * So we must continue until @start is increased to | |
1097 | * skip current ordered extent. | |
1098 | */ | |
00361589 | 1099 | if (ret) |
6e26c442 | 1100 | btrfs_drop_extent_cache(inode, start, |
4dbd80fb | 1101 | start + ram_size - 1, 0); |
17d217fe YZ |
1102 | } |
1103 | ||
0b246afa | 1104 | btrfs_dec_block_group_reservations(fs_info, ins.objectid); |
9cfa3e34 | 1105 | |
c8b97818 CM |
1106 | /* we're not doing compressed IO, don't unlock the first |
1107 | * page (which the caller expects to stay locked), don't | |
1108 | * clear any dirty bits and don't set any writeback bits | |
8b62b72b CM |
1109 | * |
1110 | * Do set the Private2 bit so we know this page was properly | |
1111 | * setup for writepage | |
c8b97818 | 1112 | */ |
a315e68f FM |
1113 | page_ops = unlock ? PAGE_UNLOCK : 0; |
1114 | page_ops |= PAGE_SET_PRIVATE2; | |
a791e35e | 1115 | |
6e26c442 | 1116 | extent_clear_unlock_delalloc(inode, start, start + ram_size - 1, |
74e9194a | 1117 | locked_page, |
c2790a2e | 1118 | EXTENT_LOCKED | EXTENT_DELALLOC, |
a315e68f | 1119 | page_ops); |
3752d22f AJ |
1120 | if (num_bytes < cur_alloc_size) |
1121 | num_bytes = 0; | |
4dbd80fb | 1122 | else |
3752d22f | 1123 | num_bytes -= cur_alloc_size; |
c59f8951 CM |
1124 | alloc_hint = ins.objectid + ins.offset; |
1125 | start += cur_alloc_size; | |
a315e68f | 1126 | extent_reserved = false; |
4dbd80fb QW |
1127 | |
1128 | /* | |
1129 | * btrfs_reloc_clone_csums() error, since start is increased | |
1130 | * extent_clear_unlock_delalloc() at out_unlock label won't | |
1131 | * free metadata of current ordered extent, we're OK to exit. | |
1132 | */ | |
1133 | if (ret) | |
1134 | goto out_unlock; | |
b888db2b | 1135 | } |
79787eaa | 1136 | out: |
be20aa9d | 1137 | return ret; |
b7d5b0a8 | 1138 | |
d9f85963 | 1139 | out_drop_extent_cache: |
6e26c442 | 1140 | btrfs_drop_extent_cache(inode, start, start + ram_size - 1, 0); |
ace68bac | 1141 | out_reserve: |
0b246afa | 1142 | btrfs_dec_block_group_reservations(fs_info, ins.objectid); |
2ff7e61e | 1143 | btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1); |
79787eaa | 1144 | out_unlock: |
a7e3b975 FM |
1145 | clear_bits = EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DELALLOC_NEW | |
1146 | EXTENT_DEFRAG | EXTENT_CLEAR_META_RESV; | |
a315e68f FM |
1147 | page_ops = PAGE_UNLOCK | PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK | |
1148 | PAGE_END_WRITEBACK; | |
1149 | /* | |
1150 | * If we reserved an extent for our delalloc range (or a subrange) and | |
1151 | * failed to create the respective ordered extent, then it means that | |
1152 | * when we reserved the extent we decremented the extent's size from | |
1153 | * the data space_info's bytes_may_use counter and incremented the | |
1154 | * space_info's bytes_reserved counter by the same amount. We must make | |
1155 | * sure extent_clear_unlock_delalloc() does not try to decrement again | |
1156 | * the data space_info's bytes_may_use counter, therefore we do not pass | |
1157 | * it the flag EXTENT_CLEAR_DATA_RESV. | |
1158 | */ | |
1159 | if (extent_reserved) { | |
6e26c442 | 1160 | extent_clear_unlock_delalloc(inode, start, |
e2c8e92d | 1161 | start + cur_alloc_size - 1, |
a315e68f FM |
1162 | locked_page, |
1163 | clear_bits, | |
1164 | page_ops); | |
1165 | start += cur_alloc_size; | |
1166 | if (start >= end) | |
1167 | goto out; | |
1168 | } | |
6e26c442 | 1169 | extent_clear_unlock_delalloc(inode, start, end, locked_page, |
a315e68f FM |
1170 | clear_bits | EXTENT_CLEAR_DATA_RESV, |
1171 | page_ops); | |
79787eaa | 1172 | goto out; |
771ed689 | 1173 | } |
c8b97818 | 1174 | |
771ed689 CM |
1175 | /* |
1176 | * work queue call back to started compression on a file and pages | |
1177 | */ | |
1178 | static noinline void async_cow_start(struct btrfs_work *work) | |
1179 | { | |
b5326271 | 1180 | struct async_chunk *async_chunk; |
ac3e9933 | 1181 | int compressed_extents; |
771ed689 | 1182 | |
b5326271 | 1183 | async_chunk = container_of(work, struct async_chunk, work); |
771ed689 | 1184 | |
ac3e9933 NB |
1185 | compressed_extents = compress_file_range(async_chunk); |
1186 | if (compressed_extents == 0) { | |
b5326271 NB |
1187 | btrfs_add_delayed_iput(async_chunk->inode); |
1188 | async_chunk->inode = NULL; | |
8180ef88 | 1189 | } |
771ed689 CM |
1190 | } |
1191 | ||
1192 | /* | |
1193 | * work queue call back to submit previously compressed pages | |
1194 | */ | |
1195 | static noinline void async_cow_submit(struct btrfs_work *work) | |
1196 | { | |
c5a68aec NB |
1197 | struct async_chunk *async_chunk = container_of(work, struct async_chunk, |
1198 | work); | |
1199 | struct btrfs_fs_info *fs_info = btrfs_work_owner(work); | |
771ed689 CM |
1200 | unsigned long nr_pages; |
1201 | ||
b5326271 | 1202 | nr_pages = (async_chunk->end - async_chunk->start + PAGE_SIZE) >> |
09cbfeaf | 1203 | PAGE_SHIFT; |
771ed689 | 1204 | |
093258e6 | 1205 | /* atomic_sub_return implies a barrier */ |
0b246afa | 1206 | if (atomic_sub_return(nr_pages, &fs_info->async_delalloc_pages) < |
093258e6 DS |
1207 | 5 * SZ_1M) |
1208 | cond_wake_up_nomb(&fs_info->async_submit_wait); | |
771ed689 | 1209 | |
4546d178 | 1210 | /* |
b5326271 | 1211 | * ->inode could be NULL if async_chunk_start has failed to compress, |
4546d178 NB |
1212 | * in which case we don't have anything to submit, yet we need to |
1213 | * always adjust ->async_delalloc_pages as its paired with the init | |
1214 | * happening in cow_file_range_async | |
1215 | */ | |
b5326271 NB |
1216 | if (async_chunk->inode) |
1217 | submit_compressed_extents(async_chunk); | |
771ed689 | 1218 | } |
c8b97818 | 1219 | |
771ed689 CM |
1220 | static noinline void async_cow_free(struct btrfs_work *work) |
1221 | { | |
b5326271 | 1222 | struct async_chunk *async_chunk; |
97db1204 | 1223 | |
b5326271 NB |
1224 | async_chunk = container_of(work, struct async_chunk, work); |
1225 | if (async_chunk->inode) | |
1226 | btrfs_add_delayed_iput(async_chunk->inode); | |
ec39f769 CM |
1227 | if (async_chunk->blkcg_css) |
1228 | css_put(async_chunk->blkcg_css); | |
97db1204 NB |
1229 | /* |
1230 | * Since the pointer to 'pending' is at the beginning of the array of | |
b5326271 | 1231 | * async_chunk's, freeing it ensures the whole array has been freed. |
97db1204 | 1232 | */ |
b5326271 | 1233 | if (atomic_dec_and_test(async_chunk->pending)) |
b1c16ac9 | 1234 | kvfree(async_chunk->pending); |
771ed689 CM |
1235 | } |
1236 | ||
751b6431 | 1237 | static int cow_file_range_async(struct btrfs_inode *inode, |
ec39f769 CM |
1238 | struct writeback_control *wbc, |
1239 | struct page *locked_page, | |
771ed689 | 1240 | u64 start, u64 end, int *page_started, |
fac07d2b | 1241 | unsigned long *nr_written) |
771ed689 | 1242 | { |
751b6431 | 1243 | struct btrfs_fs_info *fs_info = inode->root->fs_info; |
ec39f769 | 1244 | struct cgroup_subsys_state *blkcg_css = wbc_blkcg_css(wbc); |
97db1204 NB |
1245 | struct async_cow *ctx; |
1246 | struct async_chunk *async_chunk; | |
771ed689 CM |
1247 | unsigned long nr_pages; |
1248 | u64 cur_end; | |
97db1204 NB |
1249 | u64 num_chunks = DIV_ROUND_UP(end - start, SZ_512K); |
1250 | int i; | |
1251 | bool should_compress; | |
b1c16ac9 | 1252 | unsigned nofs_flag; |
fac07d2b | 1253 | const unsigned int write_flags = wbc_to_write_flags(wbc); |
771ed689 | 1254 | |
751b6431 | 1255 | unlock_extent(&inode->io_tree, start, end); |
97db1204 | 1256 | |
751b6431 | 1257 | if (inode->flags & BTRFS_INODE_NOCOMPRESS && |
97db1204 NB |
1258 | !btrfs_test_opt(fs_info, FORCE_COMPRESS)) { |
1259 | num_chunks = 1; | |
1260 | should_compress = false; | |
1261 | } else { | |
1262 | should_compress = true; | |
1263 | } | |
1264 | ||
b1c16ac9 NB |
1265 | nofs_flag = memalloc_nofs_save(); |
1266 | ctx = kvmalloc(struct_size(ctx, chunks, num_chunks), GFP_KERNEL); | |
1267 | memalloc_nofs_restore(nofs_flag); | |
1268 | ||
97db1204 NB |
1269 | if (!ctx) { |
1270 | unsigned clear_bits = EXTENT_LOCKED | EXTENT_DELALLOC | | |
1271 | EXTENT_DELALLOC_NEW | EXTENT_DEFRAG | | |
1272 | EXTENT_DO_ACCOUNTING; | |
1273 | unsigned long page_ops = PAGE_UNLOCK | PAGE_CLEAR_DIRTY | | |
1274 | PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK | | |
1275 | PAGE_SET_ERROR; | |
1276 | ||
751b6431 NB |
1277 | extent_clear_unlock_delalloc(inode, start, end, locked_page, |
1278 | clear_bits, page_ops); | |
97db1204 NB |
1279 | return -ENOMEM; |
1280 | } | |
1281 | ||
1282 | async_chunk = ctx->chunks; | |
1283 | atomic_set(&ctx->num_chunks, num_chunks); | |
1284 | ||
1285 | for (i = 0; i < num_chunks; i++) { | |
1286 | if (should_compress) | |
1287 | cur_end = min(end, start + SZ_512K - 1); | |
1288 | else | |
1289 | cur_end = end; | |
771ed689 | 1290 | |
bd4691a0 NB |
1291 | /* |
1292 | * igrab is called higher up in the call chain, take only the | |
1293 | * lightweight reference for the callback lifetime | |
1294 | */ | |
751b6431 | 1295 | ihold(&inode->vfs_inode); |
97db1204 | 1296 | async_chunk[i].pending = &ctx->num_chunks; |
751b6431 | 1297 | async_chunk[i].inode = &inode->vfs_inode; |
97db1204 NB |
1298 | async_chunk[i].start = start; |
1299 | async_chunk[i].end = cur_end; | |
97db1204 NB |
1300 | async_chunk[i].write_flags = write_flags; |
1301 | INIT_LIST_HEAD(&async_chunk[i].extents); | |
1302 | ||
1d53c9e6 CM |
1303 | /* |
1304 | * The locked_page comes all the way from writepage and its | |
1305 | * the original page we were actually given. As we spread | |
1306 | * this large delalloc region across multiple async_chunk | |
1307 | * structs, only the first struct needs a pointer to locked_page | |
1308 | * | |
1309 | * This way we don't need racey decisions about who is supposed | |
1310 | * to unlock it. | |
1311 | */ | |
1312 | if (locked_page) { | |
ec39f769 CM |
1313 | /* |
1314 | * Depending on the compressibility, the pages might or | |
1315 | * might not go through async. We want all of them to | |
1316 | * be accounted against wbc once. Let's do it here | |
1317 | * before the paths diverge. wbc accounting is used | |
1318 | * only for foreign writeback detection and doesn't | |
1319 | * need full accuracy. Just account the whole thing | |
1320 | * against the first page. | |
1321 | */ | |
1322 | wbc_account_cgroup_owner(wbc, locked_page, | |
1323 | cur_end - start); | |
1d53c9e6 CM |
1324 | async_chunk[i].locked_page = locked_page; |
1325 | locked_page = NULL; | |
1326 | } else { | |
1327 | async_chunk[i].locked_page = NULL; | |
1328 | } | |
1329 | ||
ec39f769 CM |
1330 | if (blkcg_css != blkcg_root_css) { |
1331 | css_get(blkcg_css); | |
1332 | async_chunk[i].blkcg_css = blkcg_css; | |
1333 | } else { | |
1334 | async_chunk[i].blkcg_css = NULL; | |
1335 | } | |
1336 | ||
a0cac0ec OS |
1337 | btrfs_init_work(&async_chunk[i].work, async_cow_start, |
1338 | async_cow_submit, async_cow_free); | |
771ed689 | 1339 | |
97db1204 | 1340 | nr_pages = DIV_ROUND_UP(cur_end - start, PAGE_SIZE); |
0b246afa | 1341 | atomic_add(nr_pages, &fs_info->async_delalloc_pages); |
771ed689 | 1342 | |
97db1204 | 1343 | btrfs_queue_work(fs_info->delalloc_workers, &async_chunk[i].work); |
771ed689 | 1344 | |
771ed689 CM |
1345 | *nr_written += nr_pages; |
1346 | start = cur_end + 1; | |
1347 | } | |
1348 | *page_started = 1; | |
1349 | return 0; | |
be20aa9d CM |
1350 | } |
1351 | ||
2ff7e61e | 1352 | static noinline int csum_exist_in_range(struct btrfs_fs_info *fs_info, |
17d217fe YZ |
1353 | u64 bytenr, u64 num_bytes) |
1354 | { | |
1355 | int ret; | |
1356 | struct btrfs_ordered_sum *sums; | |
1357 | LIST_HEAD(list); | |
1358 | ||
0b246afa | 1359 | ret = btrfs_lookup_csums_range(fs_info->csum_root, bytenr, |
a2de733c | 1360 | bytenr + num_bytes - 1, &list, 0); |
17d217fe YZ |
1361 | if (ret == 0 && list_empty(&list)) |
1362 | return 0; | |
1363 | ||
1364 | while (!list_empty(&list)) { | |
1365 | sums = list_entry(list.next, struct btrfs_ordered_sum, list); | |
1366 | list_del(&sums->list); | |
1367 | kfree(sums); | |
1368 | } | |
58113753 LB |
1369 | if (ret < 0) |
1370 | return ret; | |
17d217fe YZ |
1371 | return 1; |
1372 | } | |
1373 | ||
8ba96f3d | 1374 | static int fallback_to_cow(struct btrfs_inode *inode, struct page *locked_page, |
467dc47e FM |
1375 | const u64 start, const u64 end, |
1376 | int *page_started, unsigned long *nr_written) | |
1377 | { | |
8ba96f3d NB |
1378 | const bool is_space_ino = btrfs_is_free_space_inode(inode); |
1379 | const bool is_reloc_ino = (inode->root->root_key.objectid == | |
6bd335b4 | 1380 | BTRFS_DATA_RELOC_TREE_OBJECTID); |
2166e5ed | 1381 | const u64 range_bytes = end + 1 - start; |
8ba96f3d | 1382 | struct extent_io_tree *io_tree = &inode->io_tree; |
467dc47e FM |
1383 | u64 range_start = start; |
1384 | u64 count; | |
1385 | ||
1386 | /* | |
1387 | * If EXTENT_NORESERVE is set it means that when the buffered write was | |
1388 | * made we had not enough available data space and therefore we did not | |
1389 | * reserve data space for it, since we though we could do NOCOW for the | |
1390 | * respective file range (either there is prealloc extent or the inode | |
1391 | * has the NOCOW bit set). | |
1392 | * | |
1393 | * However when we need to fallback to COW mode (because for example the | |
1394 | * block group for the corresponding extent was turned to RO mode by a | |
1395 | * scrub or relocation) we need to do the following: | |
1396 | * | |
1397 | * 1) We increment the bytes_may_use counter of the data space info. | |
1398 | * If COW succeeds, it allocates a new data extent and after doing | |
1399 | * that it decrements the space info's bytes_may_use counter and | |
1400 | * increments its bytes_reserved counter by the same amount (we do | |
1401 | * this at btrfs_add_reserved_bytes()). So we need to increment the | |
1402 | * bytes_may_use counter to compensate (when space is reserved at | |
1403 | * buffered write time, the bytes_may_use counter is incremented); | |
1404 | * | |
1405 | * 2) We clear the EXTENT_NORESERVE bit from the range. We do this so | |
1406 | * that if the COW path fails for any reason, it decrements (through | |
1407 | * extent_clear_unlock_delalloc()) the bytes_may_use counter of the | |
1408 | * data space info, which we incremented in the step above. | |
2166e5ed FM |
1409 | * |
1410 | * If we need to fallback to cow and the inode corresponds to a free | |
6bd335b4 FM |
1411 | * space cache inode or an inode of the data relocation tree, we must |
1412 | * also increment bytes_may_use of the data space_info for the same | |
1413 | * reason. Space caches and relocated data extents always get a prealloc | |
2166e5ed | 1414 | * extent for them, however scrub or balance may have set the block |
6bd335b4 FM |
1415 | * group that contains that extent to RO mode and therefore force COW |
1416 | * when starting writeback. | |
467dc47e | 1417 | */ |
2166e5ed | 1418 | count = count_range_bits(io_tree, &range_start, end, range_bytes, |
467dc47e | 1419 | EXTENT_NORESERVE, 0); |
6bd335b4 FM |
1420 | if (count > 0 || is_space_ino || is_reloc_ino) { |
1421 | u64 bytes = count; | |
8ba96f3d | 1422 | struct btrfs_fs_info *fs_info = inode->root->fs_info; |
467dc47e FM |
1423 | struct btrfs_space_info *sinfo = fs_info->data_sinfo; |
1424 | ||
6bd335b4 FM |
1425 | if (is_space_ino || is_reloc_ino) |
1426 | bytes = range_bytes; | |
1427 | ||
467dc47e | 1428 | spin_lock(&sinfo->lock); |
2166e5ed | 1429 | btrfs_space_info_update_bytes_may_use(fs_info, sinfo, bytes); |
467dc47e FM |
1430 | spin_unlock(&sinfo->lock); |
1431 | ||
2166e5ed FM |
1432 | if (count > 0) |
1433 | clear_extent_bit(io_tree, start, end, EXTENT_NORESERVE, | |
1434 | 0, 0, NULL); | |
467dc47e FM |
1435 | } |
1436 | ||
8ba96f3d NB |
1437 | return cow_file_range(inode, locked_page, start, end, page_started, |
1438 | nr_written, 1); | |
467dc47e FM |
1439 | } |
1440 | ||
d352ac68 CM |
1441 | /* |
1442 | * when nowcow writeback call back. This checks for snapshots or COW copies | |
1443 | * of the extents that exist in the file, and COWs the file as required. | |
1444 | * | |
1445 | * If no cow copies or snapshots exist, we write directly to the existing | |
1446 | * blocks on disk | |
1447 | */ | |
968322c8 | 1448 | static noinline int run_delalloc_nocow(struct btrfs_inode *inode, |
7f366cfe | 1449 | struct page *locked_page, |
3e024846 NB |
1450 | const u64 start, const u64 end, |
1451 | int *page_started, int force, | |
1452 | unsigned long *nr_written) | |
be20aa9d | 1453 | { |
968322c8 NB |
1454 | struct btrfs_fs_info *fs_info = inode->root->fs_info; |
1455 | struct btrfs_root *root = inode->root; | |
be20aa9d | 1456 | struct btrfs_path *path; |
3e024846 NB |
1457 | u64 cow_start = (u64)-1; |
1458 | u64 cur_offset = start; | |
8ecebf4d | 1459 | int ret; |
3e024846 | 1460 | bool check_prev = true; |
968322c8 NB |
1461 | const bool freespace_inode = btrfs_is_free_space_inode(inode); |
1462 | u64 ino = btrfs_ino(inode); | |
762bf098 NB |
1463 | bool nocow = false; |
1464 | u64 disk_bytenr = 0; | |
be20aa9d CM |
1465 | |
1466 | path = btrfs_alloc_path(); | |
17ca04af | 1467 | if (!path) { |
968322c8 | 1468 | extent_clear_unlock_delalloc(inode, start, end, locked_page, |
c2790a2e | 1469 | EXTENT_LOCKED | EXTENT_DELALLOC | |
151a41bc JB |
1470 | EXTENT_DO_ACCOUNTING | |
1471 | EXTENT_DEFRAG, PAGE_UNLOCK | | |
c2790a2e JB |
1472 | PAGE_CLEAR_DIRTY | |
1473 | PAGE_SET_WRITEBACK | | |
1474 | PAGE_END_WRITEBACK); | |
d8926bb3 | 1475 | return -ENOMEM; |
17ca04af | 1476 | } |
82d5902d | 1477 | |
80ff3856 | 1478 | while (1) { |
3e024846 NB |
1479 | struct btrfs_key found_key; |
1480 | struct btrfs_file_extent_item *fi; | |
1481 | struct extent_buffer *leaf; | |
1482 | u64 extent_end; | |
1483 | u64 extent_offset; | |
3e024846 NB |
1484 | u64 num_bytes = 0; |
1485 | u64 disk_num_bytes; | |
3e024846 NB |
1486 | u64 ram_bytes; |
1487 | int extent_type; | |
762bf098 NB |
1488 | |
1489 | nocow = false; | |
3e024846 | 1490 | |
e4c3b2dc | 1491 | ret = btrfs_lookup_file_extent(NULL, root, path, ino, |
80ff3856 | 1492 | cur_offset, 0); |
d788a349 | 1493 | if (ret < 0) |
79787eaa | 1494 | goto error; |
a6bd9cd1 NB |
1495 | |
1496 | /* | |
1497 | * If there is no extent for our range when doing the initial | |
1498 | * search, then go back to the previous slot as it will be the | |
1499 | * one containing the search offset | |
1500 | */ | |
80ff3856 YZ |
1501 | if (ret > 0 && path->slots[0] > 0 && check_prev) { |
1502 | leaf = path->nodes[0]; | |
1503 | btrfs_item_key_to_cpu(leaf, &found_key, | |
1504 | path->slots[0] - 1); | |
33345d01 | 1505 | if (found_key.objectid == ino && |
80ff3856 YZ |
1506 | found_key.type == BTRFS_EXTENT_DATA_KEY) |
1507 | path->slots[0]--; | |
1508 | } | |
3e024846 | 1509 | check_prev = false; |
80ff3856 | 1510 | next_slot: |
a6bd9cd1 | 1511 | /* Go to next leaf if we have exhausted the current one */ |
80ff3856 YZ |
1512 | leaf = path->nodes[0]; |
1513 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
1514 | ret = btrfs_next_leaf(root, path); | |
e8916699 LB |
1515 | if (ret < 0) { |
1516 | if (cow_start != (u64)-1) | |
1517 | cur_offset = cow_start; | |
79787eaa | 1518 | goto error; |
e8916699 | 1519 | } |
80ff3856 YZ |
1520 | if (ret > 0) |
1521 | break; | |
1522 | leaf = path->nodes[0]; | |
1523 | } | |
be20aa9d | 1524 | |
80ff3856 YZ |
1525 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
1526 | ||
a6bd9cd1 | 1527 | /* Didn't find anything for our INO */ |
1d512cb7 FM |
1528 | if (found_key.objectid > ino) |
1529 | break; | |
a6bd9cd1 NB |
1530 | /* |
1531 | * Keep searching until we find an EXTENT_ITEM or there are no | |
1532 | * more extents for this inode | |
1533 | */ | |
1d512cb7 FM |
1534 | if (WARN_ON_ONCE(found_key.objectid < ino) || |
1535 | found_key.type < BTRFS_EXTENT_DATA_KEY) { | |
1536 | path->slots[0]++; | |
1537 | goto next_slot; | |
1538 | } | |
a6bd9cd1 NB |
1539 | |
1540 | /* Found key is not EXTENT_DATA_KEY or starts after req range */ | |
1d512cb7 | 1541 | if (found_key.type > BTRFS_EXTENT_DATA_KEY || |
80ff3856 YZ |
1542 | found_key.offset > end) |
1543 | break; | |
1544 | ||
a6bd9cd1 NB |
1545 | /* |
1546 | * If the found extent starts after requested offset, then | |
1547 | * adjust extent_end to be right before this extent begins | |
1548 | */ | |
80ff3856 YZ |
1549 | if (found_key.offset > cur_offset) { |
1550 | extent_end = found_key.offset; | |
e9061e21 | 1551 | extent_type = 0; |
80ff3856 YZ |
1552 | goto out_check; |
1553 | } | |
1554 | ||
a6bd9cd1 NB |
1555 | /* |
1556 | * Found extent which begins before our range and potentially | |
1557 | * intersect it | |
1558 | */ | |
80ff3856 YZ |
1559 | fi = btrfs_item_ptr(leaf, path->slots[0], |
1560 | struct btrfs_file_extent_item); | |
1561 | extent_type = btrfs_file_extent_type(leaf, fi); | |
1562 | ||
cc95bef6 | 1563 | ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi); |
d899e052 YZ |
1564 | if (extent_type == BTRFS_FILE_EXTENT_REG || |
1565 | extent_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
80ff3856 | 1566 | disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); |
5d4f98a2 | 1567 | extent_offset = btrfs_file_extent_offset(leaf, fi); |
80ff3856 YZ |
1568 | extent_end = found_key.offset + |
1569 | btrfs_file_extent_num_bytes(leaf, fi); | |
b4939680 JB |
1570 | disk_num_bytes = |
1571 | btrfs_file_extent_disk_num_bytes(leaf, fi); | |
a6bd9cd1 | 1572 | /* |
de7999af FM |
1573 | * If the extent we got ends before our current offset, |
1574 | * skip to the next extent. | |
a6bd9cd1 | 1575 | */ |
de7999af | 1576 | if (extent_end <= cur_offset) { |
80ff3856 YZ |
1577 | path->slots[0]++; |
1578 | goto next_slot; | |
1579 | } | |
a6bd9cd1 | 1580 | /* Skip holes */ |
17d217fe YZ |
1581 | if (disk_bytenr == 0) |
1582 | goto out_check; | |
a6bd9cd1 | 1583 | /* Skip compressed/encrypted/encoded extents */ |
80ff3856 YZ |
1584 | if (btrfs_file_extent_compression(leaf, fi) || |
1585 | btrfs_file_extent_encryption(leaf, fi) || | |
1586 | btrfs_file_extent_other_encoding(leaf, fi)) | |
1587 | goto out_check; | |
78d4295b | 1588 | /* |
a6bd9cd1 NB |
1589 | * If extent is created before the last volume's snapshot |
1590 | * this implies the extent is shared, hence we can't do | |
1591 | * nocow. This is the same check as in | |
1592 | * btrfs_cross_ref_exist but without calling | |
1593 | * btrfs_search_slot. | |
78d4295b | 1594 | */ |
3e024846 | 1595 | if (!freespace_inode && |
27a7ff55 | 1596 | btrfs_file_extent_generation(leaf, fi) <= |
78d4295b EL |
1597 | btrfs_root_last_snapshot(&root->root_item)) |
1598 | goto out_check; | |
d899e052 YZ |
1599 | if (extent_type == BTRFS_FILE_EXTENT_REG && !force) |
1600 | goto out_check; | |
a6bd9cd1 | 1601 | /* If extent is RO, we must COW it */ |
2ff7e61e | 1602 | if (btrfs_extent_readonly(fs_info, disk_bytenr)) |
80ff3856 | 1603 | goto out_check; |
58113753 LB |
1604 | ret = btrfs_cross_ref_exist(root, ino, |
1605 | found_key.offset - | |
1606 | extent_offset, disk_bytenr); | |
1607 | if (ret) { | |
1608 | /* | |
1609 | * ret could be -EIO if the above fails to read | |
1610 | * metadata. | |
1611 | */ | |
1612 | if (ret < 0) { | |
1613 | if (cow_start != (u64)-1) | |
1614 | cur_offset = cow_start; | |
1615 | goto error; | |
1616 | } | |
1617 | ||
3e024846 | 1618 | WARN_ON_ONCE(freespace_inode); |
17d217fe | 1619 | goto out_check; |
58113753 | 1620 | } |
5d4f98a2 | 1621 | disk_bytenr += extent_offset; |
17d217fe YZ |
1622 | disk_bytenr += cur_offset - found_key.offset; |
1623 | num_bytes = min(end + 1, extent_end) - cur_offset; | |
e9894fd3 | 1624 | /* |
a6bd9cd1 NB |
1625 | * If there are pending snapshots for this root, we |
1626 | * fall into common COW way | |
e9894fd3 | 1627 | */ |
3e024846 | 1628 | if (!freespace_inode && atomic_read(&root->snapshot_force_cow)) |
8ecebf4d | 1629 | goto out_check; |
17d217fe YZ |
1630 | /* |
1631 | * force cow if csum exists in the range. | |
1632 | * this ensure that csum for a given extent are | |
1633 | * either valid or do not exist. | |
1634 | */ | |
58113753 LB |
1635 | ret = csum_exist_in_range(fs_info, disk_bytenr, |
1636 | num_bytes); | |
1637 | if (ret) { | |
58113753 LB |
1638 | /* |
1639 | * ret could be -EIO if the above fails to read | |
1640 | * metadata. | |
1641 | */ | |
1642 | if (ret < 0) { | |
1643 | if (cow_start != (u64)-1) | |
1644 | cur_offset = cow_start; | |
1645 | goto error; | |
1646 | } | |
3e024846 | 1647 | WARN_ON_ONCE(freespace_inode); |
17d217fe | 1648 | goto out_check; |
91e1f56a | 1649 | } |
8ecebf4d | 1650 | if (!btrfs_inc_nocow_writers(fs_info, disk_bytenr)) |
f78c436c | 1651 | goto out_check; |
3e024846 | 1652 | nocow = true; |
80ff3856 | 1653 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
e8e21007 NB |
1654 | extent_end = found_key.offset + ram_bytes; |
1655 | extent_end = ALIGN(extent_end, fs_info->sectorsize); | |
922f0518 NB |
1656 | /* Skip extents outside of our requested range */ |
1657 | if (extent_end <= start) { | |
1658 | path->slots[0]++; | |
1659 | goto next_slot; | |
1660 | } | |
80ff3856 | 1661 | } else { |
e8e21007 | 1662 | /* If this triggers then we have a memory corruption */ |
290342f6 | 1663 | BUG(); |
80ff3856 YZ |
1664 | } |
1665 | out_check: | |
a6bd9cd1 NB |
1666 | /* |
1667 | * If nocow is false then record the beginning of the range | |
1668 | * that needs to be COWed | |
1669 | */ | |
80ff3856 YZ |
1670 | if (!nocow) { |
1671 | if (cow_start == (u64)-1) | |
1672 | cow_start = cur_offset; | |
1673 | cur_offset = extent_end; | |
1674 | if (cur_offset > end) | |
1675 | break; | |
1676 | path->slots[0]++; | |
1677 | goto next_slot; | |
7ea394f1 YZ |
1678 | } |
1679 | ||
b3b4aa74 | 1680 | btrfs_release_path(path); |
a6bd9cd1 NB |
1681 | |
1682 | /* | |
1683 | * COW range from cow_start to found_key.offset - 1. As the key | |
1684 | * will contain the beginning of the first extent that can be | |
1685 | * NOCOW, following one which needs to be COW'ed | |
1686 | */ | |
80ff3856 | 1687 | if (cow_start != (u64)-1) { |
968322c8 | 1688 | ret = fallback_to_cow(inode, locked_page, |
8ba96f3d | 1689 | cow_start, found_key.offset - 1, |
467dc47e | 1690 | page_started, nr_written); |
230ed397 | 1691 | if (ret) |
79787eaa | 1692 | goto error; |
80ff3856 | 1693 | cow_start = (u64)-1; |
7ea394f1 | 1694 | } |
80ff3856 | 1695 | |
d899e052 | 1696 | if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) { |
6f9994db | 1697 | u64 orig_start = found_key.offset - extent_offset; |
3e024846 | 1698 | struct extent_map *em; |
6f9994db | 1699 | |
968322c8 | 1700 | em = create_io_em(inode, cur_offset, num_bytes, |
6f9994db LB |
1701 | orig_start, |
1702 | disk_bytenr, /* block_start */ | |
1703 | num_bytes, /* block_len */ | |
1704 | disk_num_bytes, /* orig_block_len */ | |
1705 | ram_bytes, BTRFS_COMPRESS_NONE, | |
1706 | BTRFS_ORDERED_PREALLOC); | |
1707 | if (IS_ERR(em)) { | |
6f9994db LB |
1708 | ret = PTR_ERR(em); |
1709 | goto error; | |
d899e052 | 1710 | } |
6f9994db | 1711 | free_extent_map(em); |
968322c8 | 1712 | ret = btrfs_add_ordered_extent(inode, cur_offset, |
bb55f626 NB |
1713 | disk_bytenr, num_bytes, |
1714 | num_bytes, | |
1715 | BTRFS_ORDERED_PREALLOC); | |
762bf098 | 1716 | if (ret) { |
968322c8 | 1717 | btrfs_drop_extent_cache(inode, cur_offset, |
762bf098 NB |
1718 | cur_offset + num_bytes - 1, |
1719 | 0); | |
1720 | goto error; | |
1721 | } | |
d899e052 | 1722 | } else { |
968322c8 | 1723 | ret = btrfs_add_ordered_extent(inode, cur_offset, |
bb55f626 NB |
1724 | disk_bytenr, num_bytes, |
1725 | num_bytes, | |
1726 | BTRFS_ORDERED_NOCOW); | |
762bf098 NB |
1727 | if (ret) |
1728 | goto error; | |
d899e052 | 1729 | } |
80ff3856 | 1730 | |
f78c436c | 1731 | if (nocow) |
0b246afa | 1732 | btrfs_dec_nocow_writers(fs_info, disk_bytenr); |
762bf098 | 1733 | nocow = false; |
771ed689 | 1734 | |
efa56464 | 1735 | if (root->root_key.objectid == |
4dbd80fb QW |
1736 | BTRFS_DATA_RELOC_TREE_OBJECTID) |
1737 | /* | |
1738 | * Error handled later, as we must prevent | |
1739 | * extent_clear_unlock_delalloc() in error handler | |
1740 | * from freeing metadata of created ordered extent. | |
1741 | */ | |
968322c8 | 1742 | ret = btrfs_reloc_clone_csums(inode, cur_offset, |
efa56464 | 1743 | num_bytes); |
efa56464 | 1744 | |
968322c8 | 1745 | extent_clear_unlock_delalloc(inode, cur_offset, |
74e9194a | 1746 | cur_offset + num_bytes - 1, |
c2790a2e | 1747 | locked_page, EXTENT_LOCKED | |
18513091 WX |
1748 | EXTENT_DELALLOC | |
1749 | EXTENT_CLEAR_DATA_RESV, | |
1750 | PAGE_UNLOCK | PAGE_SET_PRIVATE2); | |
1751 | ||
80ff3856 | 1752 | cur_offset = extent_end; |
4dbd80fb QW |
1753 | |
1754 | /* | |
1755 | * btrfs_reloc_clone_csums() error, now we're OK to call error | |
1756 | * handler, as metadata for created ordered extent will only | |
1757 | * be freed by btrfs_finish_ordered_io(). | |
1758 | */ | |
1759 | if (ret) | |
1760 | goto error; | |
80ff3856 YZ |
1761 | if (cur_offset > end) |
1762 | break; | |
be20aa9d | 1763 | } |
b3b4aa74 | 1764 | btrfs_release_path(path); |
80ff3856 | 1765 | |
506481b2 | 1766 | if (cur_offset <= end && cow_start == (u64)-1) |
80ff3856 | 1767 | cow_start = cur_offset; |
17ca04af | 1768 | |
80ff3856 | 1769 | if (cow_start != (u64)-1) { |
506481b2 | 1770 | cur_offset = end; |
968322c8 NB |
1771 | ret = fallback_to_cow(inode, locked_page, cow_start, end, |
1772 | page_started, nr_written); | |
d788a349 | 1773 | if (ret) |
79787eaa | 1774 | goto error; |
80ff3856 YZ |
1775 | } |
1776 | ||
79787eaa | 1777 | error: |
762bf098 NB |
1778 | if (nocow) |
1779 | btrfs_dec_nocow_writers(fs_info, disk_bytenr); | |
1780 | ||
17ca04af | 1781 | if (ret && cur_offset < end) |
968322c8 | 1782 | extent_clear_unlock_delalloc(inode, cur_offset, end, |
c2790a2e | 1783 | locked_page, EXTENT_LOCKED | |
151a41bc JB |
1784 | EXTENT_DELALLOC | EXTENT_DEFRAG | |
1785 | EXTENT_DO_ACCOUNTING, PAGE_UNLOCK | | |
1786 | PAGE_CLEAR_DIRTY | | |
c2790a2e JB |
1787 | PAGE_SET_WRITEBACK | |
1788 | PAGE_END_WRITEBACK); | |
7ea394f1 | 1789 | btrfs_free_path(path); |
79787eaa | 1790 | return ret; |
be20aa9d CM |
1791 | } |
1792 | ||
0c494225 | 1793 | static inline int need_force_cow(struct btrfs_inode *inode, u64 start, u64 end) |
47059d93 WS |
1794 | { |
1795 | ||
0c494225 NB |
1796 | if (!(inode->flags & BTRFS_INODE_NODATACOW) && |
1797 | !(inode->flags & BTRFS_INODE_PREALLOC)) | |
47059d93 WS |
1798 | return 0; |
1799 | ||
1800 | /* | |
1801 | * @defrag_bytes is a hint value, no spinlock held here, | |
1802 | * if is not zero, it means the file is defragging. | |
1803 | * Force cow if given extent needs to be defragged. | |
1804 | */ | |
0c494225 NB |
1805 | if (inode->defrag_bytes && |
1806 | test_range_bit(&inode->io_tree, start, end, EXTENT_DEFRAG, 0, NULL)) | |
47059d93 WS |
1807 | return 1; |
1808 | ||
1809 | return 0; | |
1810 | } | |
1811 | ||
d352ac68 | 1812 | /* |
5eaad97a NB |
1813 | * Function to process delayed allocation (create CoW) for ranges which are |
1814 | * being touched for the first time. | |
d352ac68 | 1815 | */ |
98456b9c | 1816 | int btrfs_run_delalloc_range(struct btrfs_inode *inode, struct page *locked_page, |
5eaad97a NB |
1817 | u64 start, u64 end, int *page_started, unsigned long *nr_written, |
1818 | struct writeback_control *wbc) | |
be20aa9d | 1819 | { |
be20aa9d | 1820 | int ret; |
98456b9c | 1821 | int force_cow = need_force_cow(inode, start, end); |
a2135011 | 1822 | |
98456b9c NB |
1823 | if (inode->flags & BTRFS_INODE_NODATACOW && !force_cow) { |
1824 | ret = run_delalloc_nocow(inode, locked_page, start, end, | |
d397712b | 1825 | page_started, 1, nr_written); |
98456b9c NB |
1826 | } else if (inode->flags & BTRFS_INODE_PREALLOC && !force_cow) { |
1827 | ret = run_delalloc_nocow(inode, locked_page, start, end, | |
d397712b | 1828 | page_started, 0, nr_written); |
98456b9c NB |
1829 | } else if (!inode_can_compress(inode) || |
1830 | !inode_need_compress(inode, start, end)) { | |
1831 | ret = cow_file_range(inode, locked_page, start, end, | |
1832 | page_started, nr_written, 1); | |
7ddf5a42 | 1833 | } else { |
98456b9c NB |
1834 | set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT, &inode->runtime_flags); |
1835 | ret = cow_file_range_async(inode, wbc, locked_page, start, end, | |
fac07d2b | 1836 | page_started, nr_written); |
7ddf5a42 | 1837 | } |
52427260 | 1838 | if (ret) |
98456b9c | 1839 | btrfs_cleanup_ordered_extents(inode, locked_page, start, |
d1051d6e | 1840 | end - start + 1); |
b888db2b CM |
1841 | return ret; |
1842 | } | |
1843 | ||
abbb55f4 NB |
1844 | void btrfs_split_delalloc_extent(struct inode *inode, |
1845 | struct extent_state *orig, u64 split) | |
9ed74f2d | 1846 | { |
dcab6a3b JB |
1847 | u64 size; |
1848 | ||
0ca1f7ce | 1849 | /* not delalloc, ignore it */ |
9ed74f2d | 1850 | if (!(orig->state & EXTENT_DELALLOC)) |
1bf85046 | 1851 | return; |
9ed74f2d | 1852 | |
dcab6a3b JB |
1853 | size = orig->end - orig->start + 1; |
1854 | if (size > BTRFS_MAX_EXTENT_SIZE) { | |
823bb20a | 1855 | u32 num_extents; |
dcab6a3b JB |
1856 | u64 new_size; |
1857 | ||
1858 | /* | |
5c848198 | 1859 | * See the explanation in btrfs_merge_delalloc_extent, the same |
ba117213 | 1860 | * applies here, just in reverse. |
dcab6a3b JB |
1861 | */ |
1862 | new_size = orig->end - split + 1; | |
823bb20a | 1863 | num_extents = count_max_extents(new_size); |
ba117213 | 1864 | new_size = split - orig->start; |
823bb20a DS |
1865 | num_extents += count_max_extents(new_size); |
1866 | if (count_max_extents(size) >= num_extents) | |
dcab6a3b JB |
1867 | return; |
1868 | } | |
1869 | ||
9e0baf60 | 1870 | spin_lock(&BTRFS_I(inode)->lock); |
8b62f87b | 1871 | btrfs_mod_outstanding_extents(BTRFS_I(inode), 1); |
9e0baf60 | 1872 | spin_unlock(&BTRFS_I(inode)->lock); |
9ed74f2d JB |
1873 | } |
1874 | ||
1875 | /* | |
5c848198 NB |
1876 | * Handle merged delayed allocation extents so we can keep track of new extents |
1877 | * that are just merged onto old extents, such as when we are doing sequential | |
1878 | * writes, so we can properly account for the metadata space we'll need. | |
9ed74f2d | 1879 | */ |
5c848198 NB |
1880 | void btrfs_merge_delalloc_extent(struct inode *inode, struct extent_state *new, |
1881 | struct extent_state *other) | |
9ed74f2d | 1882 | { |
dcab6a3b | 1883 | u64 new_size, old_size; |
823bb20a | 1884 | u32 num_extents; |
dcab6a3b | 1885 | |
9ed74f2d JB |
1886 | /* not delalloc, ignore it */ |
1887 | if (!(other->state & EXTENT_DELALLOC)) | |
1bf85046 | 1888 | return; |
9ed74f2d | 1889 | |
8461a3de JB |
1890 | if (new->start > other->start) |
1891 | new_size = new->end - other->start + 1; | |
1892 | else | |
1893 | new_size = other->end - new->start + 1; | |
dcab6a3b JB |
1894 | |
1895 | /* we're not bigger than the max, unreserve the space and go */ | |
1896 | if (new_size <= BTRFS_MAX_EXTENT_SIZE) { | |
1897 | spin_lock(&BTRFS_I(inode)->lock); | |
8b62f87b | 1898 | btrfs_mod_outstanding_extents(BTRFS_I(inode), -1); |
dcab6a3b JB |
1899 | spin_unlock(&BTRFS_I(inode)->lock); |
1900 | return; | |
1901 | } | |
1902 | ||
1903 | /* | |
ba117213 JB |
1904 | * We have to add up either side to figure out how many extents were |
1905 | * accounted for before we merged into one big extent. If the number of | |
1906 | * extents we accounted for is <= the amount we need for the new range | |
1907 | * then we can return, otherwise drop. Think of it like this | |
1908 | * | |
1909 | * [ 4k][MAX_SIZE] | |
1910 | * | |
1911 | * So we've grown the extent by a MAX_SIZE extent, this would mean we | |
1912 | * need 2 outstanding extents, on one side we have 1 and the other side | |
1913 | * we have 1 so they are == and we can return. But in this case | |
1914 | * | |
1915 | * [MAX_SIZE+4k][MAX_SIZE+4k] | |
1916 | * | |
1917 | * Each range on their own accounts for 2 extents, but merged together | |
1918 | * they are only 3 extents worth of accounting, so we need to drop in | |
1919 | * this case. | |
dcab6a3b | 1920 | */ |
ba117213 | 1921 | old_size = other->end - other->start + 1; |
823bb20a | 1922 | num_extents = count_max_extents(old_size); |
ba117213 | 1923 | old_size = new->end - new->start + 1; |
823bb20a DS |
1924 | num_extents += count_max_extents(old_size); |
1925 | if (count_max_extents(new_size) >= num_extents) | |
dcab6a3b JB |
1926 | return; |
1927 | ||
9e0baf60 | 1928 | spin_lock(&BTRFS_I(inode)->lock); |
8b62f87b | 1929 | btrfs_mod_outstanding_extents(BTRFS_I(inode), -1); |
9e0baf60 | 1930 | spin_unlock(&BTRFS_I(inode)->lock); |
9ed74f2d JB |
1931 | } |
1932 | ||
eb73c1b7 MX |
1933 | static void btrfs_add_delalloc_inodes(struct btrfs_root *root, |
1934 | struct inode *inode) | |
1935 | { | |
0b246afa JM |
1936 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
1937 | ||
eb73c1b7 MX |
1938 | spin_lock(&root->delalloc_lock); |
1939 | if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) { | |
1940 | list_add_tail(&BTRFS_I(inode)->delalloc_inodes, | |
1941 | &root->delalloc_inodes); | |
1942 | set_bit(BTRFS_INODE_IN_DELALLOC_LIST, | |
1943 | &BTRFS_I(inode)->runtime_flags); | |
1944 | root->nr_delalloc_inodes++; | |
1945 | if (root->nr_delalloc_inodes == 1) { | |
0b246afa | 1946 | spin_lock(&fs_info->delalloc_root_lock); |
eb73c1b7 MX |
1947 | BUG_ON(!list_empty(&root->delalloc_root)); |
1948 | list_add_tail(&root->delalloc_root, | |
0b246afa JM |
1949 | &fs_info->delalloc_roots); |
1950 | spin_unlock(&fs_info->delalloc_root_lock); | |
eb73c1b7 MX |
1951 | } |
1952 | } | |
1953 | spin_unlock(&root->delalloc_lock); | |
1954 | } | |
1955 | ||
2b877331 NB |
1956 | |
1957 | void __btrfs_del_delalloc_inode(struct btrfs_root *root, | |
1958 | struct btrfs_inode *inode) | |
eb73c1b7 | 1959 | { |
3ffbd68c | 1960 | struct btrfs_fs_info *fs_info = root->fs_info; |
0b246afa | 1961 | |
9e3e97f4 NB |
1962 | if (!list_empty(&inode->delalloc_inodes)) { |
1963 | list_del_init(&inode->delalloc_inodes); | |
eb73c1b7 | 1964 | clear_bit(BTRFS_INODE_IN_DELALLOC_LIST, |
9e3e97f4 | 1965 | &inode->runtime_flags); |
eb73c1b7 MX |
1966 | root->nr_delalloc_inodes--; |
1967 | if (!root->nr_delalloc_inodes) { | |
7c8a0d36 | 1968 | ASSERT(list_empty(&root->delalloc_inodes)); |
0b246afa | 1969 | spin_lock(&fs_info->delalloc_root_lock); |
eb73c1b7 MX |
1970 | BUG_ON(list_empty(&root->delalloc_root)); |
1971 | list_del_init(&root->delalloc_root); | |
0b246afa | 1972 | spin_unlock(&fs_info->delalloc_root_lock); |
eb73c1b7 MX |
1973 | } |
1974 | } | |
2b877331 NB |
1975 | } |
1976 | ||
1977 | static void btrfs_del_delalloc_inode(struct btrfs_root *root, | |
1978 | struct btrfs_inode *inode) | |
1979 | { | |
1980 | spin_lock(&root->delalloc_lock); | |
1981 | __btrfs_del_delalloc_inode(root, inode); | |
eb73c1b7 MX |
1982 | spin_unlock(&root->delalloc_lock); |
1983 | } | |
1984 | ||
d352ac68 | 1985 | /* |
e06a1fc9 NB |
1986 | * Properly track delayed allocation bytes in the inode and to maintain the |
1987 | * list of inodes that have pending delalloc work to be done. | |
d352ac68 | 1988 | */ |
e06a1fc9 NB |
1989 | void btrfs_set_delalloc_extent(struct inode *inode, struct extent_state *state, |
1990 | unsigned *bits) | |
291d673e | 1991 | { |
0b246afa JM |
1992 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
1993 | ||
47059d93 WS |
1994 | if ((*bits & EXTENT_DEFRAG) && !(*bits & EXTENT_DELALLOC)) |
1995 | WARN_ON(1); | |
75eff68e CM |
1996 | /* |
1997 | * set_bit and clear bit hooks normally require _irqsave/restore | |
27160b6b | 1998 | * but in this case, we are only testing for the DELALLOC |
75eff68e CM |
1999 | * bit, which is only set or cleared with irqs on |
2000 | */ | |
0ca1f7ce | 2001 | if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) { |
291d673e | 2002 | struct btrfs_root *root = BTRFS_I(inode)->root; |
0ca1f7ce | 2003 | u64 len = state->end + 1 - state->start; |
8b62f87b | 2004 | u32 num_extents = count_max_extents(len); |
70ddc553 | 2005 | bool do_list = !btrfs_is_free_space_inode(BTRFS_I(inode)); |
9ed74f2d | 2006 | |
8b62f87b JB |
2007 | spin_lock(&BTRFS_I(inode)->lock); |
2008 | btrfs_mod_outstanding_extents(BTRFS_I(inode), num_extents); | |
2009 | spin_unlock(&BTRFS_I(inode)->lock); | |
287a0ab9 | 2010 | |
6a3891c5 | 2011 | /* For sanity tests */ |
0b246afa | 2012 | if (btrfs_is_testing(fs_info)) |
6a3891c5 JB |
2013 | return; |
2014 | ||
104b4e51 NB |
2015 | percpu_counter_add_batch(&fs_info->delalloc_bytes, len, |
2016 | fs_info->delalloc_batch); | |
df0af1a5 | 2017 | spin_lock(&BTRFS_I(inode)->lock); |
0ca1f7ce | 2018 | BTRFS_I(inode)->delalloc_bytes += len; |
47059d93 WS |
2019 | if (*bits & EXTENT_DEFRAG) |
2020 | BTRFS_I(inode)->defrag_bytes += len; | |
df0af1a5 | 2021 | if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST, |
eb73c1b7 MX |
2022 | &BTRFS_I(inode)->runtime_flags)) |
2023 | btrfs_add_delalloc_inodes(root, inode); | |
df0af1a5 | 2024 | spin_unlock(&BTRFS_I(inode)->lock); |
291d673e | 2025 | } |
a7e3b975 FM |
2026 | |
2027 | if (!(state->state & EXTENT_DELALLOC_NEW) && | |
2028 | (*bits & EXTENT_DELALLOC_NEW)) { | |
2029 | spin_lock(&BTRFS_I(inode)->lock); | |
2030 | BTRFS_I(inode)->new_delalloc_bytes += state->end + 1 - | |
2031 | state->start; | |
2032 | spin_unlock(&BTRFS_I(inode)->lock); | |
2033 | } | |
291d673e CM |
2034 | } |
2035 | ||
d352ac68 | 2036 | /* |
a36bb5f9 NB |
2037 | * Once a range is no longer delalloc this function ensures that proper |
2038 | * accounting happens. | |
d352ac68 | 2039 | */ |
a36bb5f9 NB |
2040 | void btrfs_clear_delalloc_extent(struct inode *vfs_inode, |
2041 | struct extent_state *state, unsigned *bits) | |
291d673e | 2042 | { |
a36bb5f9 NB |
2043 | struct btrfs_inode *inode = BTRFS_I(vfs_inode); |
2044 | struct btrfs_fs_info *fs_info = btrfs_sb(vfs_inode->i_sb); | |
47059d93 | 2045 | u64 len = state->end + 1 - state->start; |
823bb20a | 2046 | u32 num_extents = count_max_extents(len); |
47059d93 | 2047 | |
4a4b964f FM |
2048 | if ((state->state & EXTENT_DEFRAG) && (*bits & EXTENT_DEFRAG)) { |
2049 | spin_lock(&inode->lock); | |
6fc0ef68 | 2050 | inode->defrag_bytes -= len; |
4a4b964f FM |
2051 | spin_unlock(&inode->lock); |
2052 | } | |
47059d93 | 2053 | |
75eff68e CM |
2054 | /* |
2055 | * set_bit and clear bit hooks normally require _irqsave/restore | |
27160b6b | 2056 | * but in this case, we are only testing for the DELALLOC |
75eff68e CM |
2057 | * bit, which is only set or cleared with irqs on |
2058 | */ | |
0ca1f7ce | 2059 | if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) { |
6fc0ef68 | 2060 | struct btrfs_root *root = inode->root; |
83eea1f1 | 2061 | bool do_list = !btrfs_is_free_space_inode(inode); |
bcbfce8a | 2062 | |
8b62f87b JB |
2063 | spin_lock(&inode->lock); |
2064 | btrfs_mod_outstanding_extents(inode, -num_extents); | |
2065 | spin_unlock(&inode->lock); | |
0ca1f7ce | 2066 | |
b6d08f06 JB |
2067 | /* |
2068 | * We don't reserve metadata space for space cache inodes so we | |
52042d8e | 2069 | * don't need to call delalloc_release_metadata if there is an |
b6d08f06 JB |
2070 | * error. |
2071 | */ | |
a315e68f | 2072 | if (*bits & EXTENT_CLEAR_META_RESV && |
0b246afa | 2073 | root != fs_info->tree_root) |
43b18595 | 2074 | btrfs_delalloc_release_metadata(inode, len, false); |
0ca1f7ce | 2075 | |
6a3891c5 | 2076 | /* For sanity tests. */ |
0b246afa | 2077 | if (btrfs_is_testing(fs_info)) |
6a3891c5 JB |
2078 | return; |
2079 | ||
a315e68f FM |
2080 | if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID && |
2081 | do_list && !(state->state & EXTENT_NORESERVE) && | |
2082 | (*bits & EXTENT_CLEAR_DATA_RESV)) | |
9db5d510 | 2083 | btrfs_free_reserved_data_space_noquota(fs_info, len); |
9ed74f2d | 2084 | |
104b4e51 NB |
2085 | percpu_counter_add_batch(&fs_info->delalloc_bytes, -len, |
2086 | fs_info->delalloc_batch); | |
6fc0ef68 NB |
2087 | spin_lock(&inode->lock); |
2088 | inode->delalloc_bytes -= len; | |
2089 | if (do_list && inode->delalloc_bytes == 0 && | |
df0af1a5 | 2090 | test_bit(BTRFS_INODE_IN_DELALLOC_LIST, |
9e3e97f4 | 2091 | &inode->runtime_flags)) |
eb73c1b7 | 2092 | btrfs_del_delalloc_inode(root, inode); |
6fc0ef68 | 2093 | spin_unlock(&inode->lock); |
291d673e | 2094 | } |
a7e3b975 FM |
2095 | |
2096 | if ((state->state & EXTENT_DELALLOC_NEW) && | |
2097 | (*bits & EXTENT_DELALLOC_NEW)) { | |
2098 | spin_lock(&inode->lock); | |
2099 | ASSERT(inode->new_delalloc_bytes >= len); | |
2100 | inode->new_delalloc_bytes -= len; | |
2101 | spin_unlock(&inode->lock); | |
2102 | } | |
291d673e CM |
2103 | } |
2104 | ||
d352ac68 | 2105 | /* |
da12fe54 NB |
2106 | * btrfs_bio_fits_in_stripe - Checks whether the size of the given bio will fit |
2107 | * in a chunk's stripe. This function ensures that bios do not span a | |
2108 | * stripe/chunk | |
6f034ece | 2109 | * |
da12fe54 NB |
2110 | * @page - The page we are about to add to the bio |
2111 | * @size - size we want to add to the bio | |
2112 | * @bio - bio we want to ensure is smaller than a stripe | |
2113 | * @bio_flags - flags of the bio | |
2114 | * | |
2115 | * return 1 if page cannot be added to the bio | |
2116 | * return 0 if page can be added to the bio | |
6f034ece | 2117 | * return error otherwise |
d352ac68 | 2118 | */ |
da12fe54 NB |
2119 | int btrfs_bio_fits_in_stripe(struct page *page, size_t size, struct bio *bio, |
2120 | unsigned long bio_flags) | |
239b14b3 | 2121 | { |
0b246afa JM |
2122 | struct inode *inode = page->mapping->host; |
2123 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); | |
4f024f37 | 2124 | u64 logical = (u64)bio->bi_iter.bi_sector << 9; |
239b14b3 CM |
2125 | u64 length = 0; |
2126 | u64 map_length; | |
239b14b3 | 2127 | int ret; |
89b798ad | 2128 | struct btrfs_io_geometry geom; |
239b14b3 | 2129 | |
771ed689 CM |
2130 | if (bio_flags & EXTENT_BIO_COMPRESSED) |
2131 | return 0; | |
2132 | ||
4f024f37 | 2133 | length = bio->bi_iter.bi_size; |
239b14b3 | 2134 | map_length = length; |
89b798ad NB |
2135 | ret = btrfs_get_io_geometry(fs_info, btrfs_op(bio), logical, map_length, |
2136 | &geom); | |
6f034ece LB |
2137 | if (ret < 0) |
2138 | return ret; | |
89b798ad NB |
2139 | |
2140 | if (geom.len < length + size) | |
239b14b3 | 2141 | return 1; |
3444a972 | 2142 | return 0; |
239b14b3 CM |
2143 | } |
2144 | ||
d352ac68 CM |
2145 | /* |
2146 | * in order to insert checksums into the metadata in large chunks, | |
2147 | * we wait until bio submission time. All the pages in the bio are | |
2148 | * checksummed and sums are attached onto the ordered extent record. | |
2149 | * | |
2150 | * At IO completion time the cums attached on the ordered extent record | |
2151 | * are inserted into the btree | |
2152 | */ | |
d0ee3934 | 2153 | static blk_status_t btrfs_submit_bio_start(void *private_data, struct bio *bio, |
eaf25d93 | 2154 | u64 bio_offset) |
065631f6 | 2155 | { |
c6100a4b | 2156 | struct inode *inode = private_data; |
4e4cbee9 | 2157 | blk_status_t ret = 0; |
e015640f | 2158 | |
bd242a08 | 2159 | ret = btrfs_csum_one_bio(BTRFS_I(inode), bio, 0, 0); |
79787eaa | 2160 | BUG_ON(ret); /* -ENOMEM */ |
4a69a410 CM |
2161 | return 0; |
2162 | } | |
e015640f | 2163 | |
d352ac68 | 2164 | /* |
cad321ad | 2165 | * extent_io.c submission hook. This does the right thing for csum calculation |
4c274bc6 LB |
2166 | * on write, or reading the csums from the tree before a read. |
2167 | * | |
2168 | * Rules about async/sync submit, | |
2169 | * a) read: sync submit | |
2170 | * | |
2171 | * b) write without checksum: sync submit | |
2172 | * | |
2173 | * c) write with checksum: | |
2174 | * c-1) if bio is issued by fsync: sync submit | |
2175 | * (sync_writers != 0) | |
2176 | * | |
2177 | * c-2) if root is reloc root: sync submit | |
2178 | * (only in case of buffered IO) | |
2179 | * | |
2180 | * c-3) otherwise: async submit | |
d352ac68 | 2181 | */ |
a56b1c7b | 2182 | static blk_status_t btrfs_submit_bio_hook(struct inode *inode, struct bio *bio, |
50489a57 NB |
2183 | int mirror_num, |
2184 | unsigned long bio_flags) | |
2185 | ||
44b8bd7e | 2186 | { |
0b246afa | 2187 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
44b8bd7e | 2188 | struct btrfs_root *root = BTRFS_I(inode)->root; |
0d51e28a | 2189 | enum btrfs_wq_endio_type metadata = BTRFS_WQ_ENDIO_DATA; |
4e4cbee9 | 2190 | blk_status_t ret = 0; |
19b9bdb0 | 2191 | int skip_sum; |
b812ce28 | 2192 | int async = !atomic_read(&BTRFS_I(inode)->sync_writers); |
44b8bd7e | 2193 | |
6cbff00f | 2194 | skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM; |
cad321ad | 2195 | |
70ddc553 | 2196 | if (btrfs_is_free_space_inode(BTRFS_I(inode))) |
0d51e28a | 2197 | metadata = BTRFS_WQ_ENDIO_FREE_SPACE; |
0417341e | 2198 | |
37226b21 | 2199 | if (bio_op(bio) != REQ_OP_WRITE) { |
0b246afa | 2200 | ret = btrfs_bio_wq_end_io(fs_info, bio, metadata); |
5fd02043 | 2201 | if (ret) |
61891923 | 2202 | goto out; |
5fd02043 | 2203 | |
d20f7043 | 2204 | if (bio_flags & EXTENT_BIO_COMPRESSED) { |
61891923 SB |
2205 | ret = btrfs_submit_compressed_read(inode, bio, |
2206 | mirror_num, | |
2207 | bio_flags); | |
2208 | goto out; | |
c2db1073 | 2209 | } else if (!skip_sum) { |
db72e47f | 2210 | ret = btrfs_lookup_bio_sums(inode, bio, (u64)-1, NULL); |
c2db1073 | 2211 | if (ret) |
61891923 | 2212 | goto out; |
c2db1073 | 2213 | } |
4d1b5fb4 | 2214 | goto mapit; |
b812ce28 | 2215 | } else if (async && !skip_sum) { |
17d217fe YZ |
2216 | /* csum items have already been cloned */ |
2217 | if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID) | |
2218 | goto mapit; | |
19b9bdb0 | 2219 | /* we're doing a write, do the async checksumming */ |
c6100a4b | 2220 | ret = btrfs_wq_submit_bio(fs_info, bio, mirror_num, bio_flags, |
e7681167 | 2221 | 0, inode, btrfs_submit_bio_start); |
61891923 | 2222 | goto out; |
b812ce28 | 2223 | } else if (!skip_sum) { |
bd242a08 | 2224 | ret = btrfs_csum_one_bio(BTRFS_I(inode), bio, 0, 0); |
b812ce28 JB |
2225 | if (ret) |
2226 | goto out; | |
19b9bdb0 CM |
2227 | } |
2228 | ||
0b86a832 | 2229 | mapit: |
08635bae | 2230 | ret = btrfs_map_bio(fs_info, bio, mirror_num); |
61891923 SB |
2231 | |
2232 | out: | |
4e4cbee9 CH |
2233 | if (ret) { |
2234 | bio->bi_status = ret; | |
4246a0b6 CH |
2235 | bio_endio(bio); |
2236 | } | |
61891923 | 2237 | return ret; |
065631f6 | 2238 | } |
6885f308 | 2239 | |
d352ac68 CM |
2240 | /* |
2241 | * given a list of ordered sums record them in the inode. This happens | |
2242 | * at IO completion time based on sums calculated at bio submission time. | |
2243 | */ | |
ba1da2f4 | 2244 | static noinline int add_pending_csums(struct btrfs_trans_handle *trans, |
df9f628e | 2245 | struct inode *inode, struct list_head *list) |
e6dcd2dc | 2246 | { |
e6dcd2dc | 2247 | struct btrfs_ordered_sum *sum; |
ac01f26a | 2248 | int ret; |
e6dcd2dc | 2249 | |
c6e30871 | 2250 | list_for_each_entry(sum, list, list) { |
7c2871a2 | 2251 | trans->adding_csums = true; |
ac01f26a | 2252 | ret = btrfs_csum_file_blocks(trans, |
d20f7043 | 2253 | BTRFS_I(inode)->root->fs_info->csum_root, sum); |
7c2871a2 | 2254 | trans->adding_csums = false; |
ac01f26a NB |
2255 | if (ret) |
2256 | return ret; | |
e6dcd2dc CM |
2257 | } |
2258 | return 0; | |
2259 | } | |
2260 | ||
c2566f22 | 2261 | int btrfs_set_extent_delalloc(struct btrfs_inode *inode, u64 start, u64 end, |
e3b8a485 | 2262 | unsigned int extra_bits, |
330a5827 | 2263 | struct extent_state **cached_state) |
ea8c2819 | 2264 | { |
fdb1e121 | 2265 | WARN_ON(PAGE_ALIGNED(end)); |
c2566f22 NB |
2266 | return set_extent_delalloc(&inode->io_tree, start, end, extra_bits, |
2267 | cached_state); | |
ea8c2819 CM |
2268 | } |
2269 | ||
d352ac68 | 2270 | /* see btrfs_writepage_start_hook for details on why this is required */ |
247e743c CM |
2271 | struct btrfs_writepage_fixup { |
2272 | struct page *page; | |
f4b1363c | 2273 | struct inode *inode; |
247e743c CM |
2274 | struct btrfs_work work; |
2275 | }; | |
2276 | ||
b2950863 | 2277 | static void btrfs_writepage_fixup_worker(struct btrfs_work *work) |
247e743c CM |
2278 | { |
2279 | struct btrfs_writepage_fixup *fixup; | |
2280 | struct btrfs_ordered_extent *ordered; | |
2ac55d41 | 2281 | struct extent_state *cached_state = NULL; |
364ecf36 | 2282 | struct extent_changeset *data_reserved = NULL; |
247e743c | 2283 | struct page *page; |
65d87f79 | 2284 | struct btrfs_inode *inode; |
247e743c CM |
2285 | u64 page_start; |
2286 | u64 page_end; | |
25f3c502 | 2287 | int ret = 0; |
f4b1363c | 2288 | bool free_delalloc_space = true; |
247e743c CM |
2289 | |
2290 | fixup = container_of(work, struct btrfs_writepage_fixup, work); | |
2291 | page = fixup->page; | |
65d87f79 | 2292 | inode = BTRFS_I(fixup->inode); |
f4b1363c JB |
2293 | page_start = page_offset(page); |
2294 | page_end = page_offset(page) + PAGE_SIZE - 1; | |
2295 | ||
2296 | /* | |
2297 | * This is similar to page_mkwrite, we need to reserve the space before | |
2298 | * we take the page lock. | |
2299 | */ | |
65d87f79 NB |
2300 | ret = btrfs_delalloc_reserve_space(inode, &data_reserved, page_start, |
2301 | PAGE_SIZE); | |
4a096752 | 2302 | again: |
247e743c | 2303 | lock_page(page); |
25f3c502 CM |
2304 | |
2305 | /* | |
2306 | * Before we queued this fixup, we took a reference on the page. | |
2307 | * page->mapping may go NULL, but it shouldn't be moved to a different | |
2308 | * address space. | |
2309 | */ | |
f4b1363c JB |
2310 | if (!page->mapping || !PageDirty(page) || !PageChecked(page)) { |
2311 | /* | |
2312 | * Unfortunately this is a little tricky, either | |
2313 | * | |
2314 | * 1) We got here and our page had already been dealt with and | |
2315 | * we reserved our space, thus ret == 0, so we need to just | |
2316 | * drop our space reservation and bail. This can happen the | |
2317 | * first time we come into the fixup worker, or could happen | |
2318 | * while waiting for the ordered extent. | |
2319 | * 2) Our page was already dealt with, but we happened to get an | |
2320 | * ENOSPC above from the btrfs_delalloc_reserve_space. In | |
2321 | * this case we obviously don't have anything to release, but | |
2322 | * because the page was already dealt with we don't want to | |
2323 | * mark the page with an error, so make sure we're resetting | |
2324 | * ret to 0. This is why we have this check _before_ the ret | |
2325 | * check, because we do not want to have a surprise ENOSPC | |
2326 | * when the page was already properly dealt with. | |
2327 | */ | |
2328 | if (!ret) { | |
65d87f79 NB |
2329 | btrfs_delalloc_release_extents(inode, PAGE_SIZE); |
2330 | btrfs_delalloc_release_space(inode, data_reserved, | |
f4b1363c JB |
2331 | page_start, PAGE_SIZE, |
2332 | true); | |
2333 | } | |
2334 | ret = 0; | |
247e743c | 2335 | goto out_page; |
f4b1363c | 2336 | } |
247e743c | 2337 | |
25f3c502 | 2338 | /* |
f4b1363c JB |
2339 | * We can't mess with the page state unless it is locked, so now that |
2340 | * it is locked bail if we failed to make our space reservation. | |
25f3c502 | 2341 | */ |
f4b1363c JB |
2342 | if (ret) |
2343 | goto out_page; | |
247e743c | 2344 | |
65d87f79 | 2345 | lock_extent_bits(&inode->io_tree, page_start, page_end, &cached_state); |
4a096752 CM |
2346 | |
2347 | /* already ordered? We're done */ | |
8b62b72b | 2348 | if (PagePrivate2(page)) |
f4b1363c | 2349 | goto out_reserved; |
4a096752 | 2350 | |
65d87f79 | 2351 | ordered = btrfs_lookup_ordered_range(inode, page_start, PAGE_SIZE); |
4a096752 | 2352 | if (ordered) { |
65d87f79 NB |
2353 | unlock_extent_cached(&inode->io_tree, page_start, page_end, |
2354 | &cached_state); | |
4a096752 | 2355 | unlock_page(page); |
65d87f79 | 2356 | btrfs_start_ordered_extent(&inode->vfs_inode, ordered, 1); |
87826df0 | 2357 | btrfs_put_ordered_extent(ordered); |
4a096752 CM |
2358 | goto again; |
2359 | } | |
247e743c | 2360 | |
65d87f79 | 2361 | ret = btrfs_set_extent_delalloc(inode, page_start, page_end, 0, |
330a5827 | 2362 | &cached_state); |
25f3c502 | 2363 | if (ret) |
53687007 | 2364 | goto out_reserved; |
f3038ee3 | 2365 | |
25f3c502 CM |
2366 | /* |
2367 | * Everything went as planned, we're now the owner of a dirty page with | |
2368 | * delayed allocation bits set and space reserved for our COW | |
2369 | * destination. | |
2370 | * | |
2371 | * The page was dirty when we started, nothing should have cleaned it. | |
2372 | */ | |
2373 | BUG_ON(!PageDirty(page)); | |
f4b1363c | 2374 | free_delalloc_space = false; |
53687007 | 2375 | out_reserved: |
65d87f79 | 2376 | btrfs_delalloc_release_extents(inode, PAGE_SIZE); |
f4b1363c | 2377 | if (free_delalloc_space) |
65d87f79 NB |
2378 | btrfs_delalloc_release_space(inode, data_reserved, page_start, |
2379 | PAGE_SIZE, true); | |
2380 | unlock_extent_cached(&inode->io_tree, page_start, page_end, | |
e43bbe5e | 2381 | &cached_state); |
247e743c | 2382 | out_page: |
25f3c502 CM |
2383 | if (ret) { |
2384 | /* | |
2385 | * We hit ENOSPC or other errors. Update the mapping and page | |
2386 | * to reflect the errors and clean the page. | |
2387 | */ | |
2388 | mapping_set_error(page->mapping, ret); | |
2389 | end_extent_writepage(page, ret, page_start, page_end); | |
2390 | clear_page_dirty_for_io(page); | |
2391 | SetPageError(page); | |
2392 | } | |
2393 | ClearPageChecked(page); | |
247e743c | 2394 | unlock_page(page); |
09cbfeaf | 2395 | put_page(page); |
b897abec | 2396 | kfree(fixup); |
364ecf36 | 2397 | extent_changeset_free(data_reserved); |
f4b1363c JB |
2398 | /* |
2399 | * As a precaution, do a delayed iput in case it would be the last iput | |
2400 | * that could need flushing space. Recursing back to fixup worker would | |
2401 | * deadlock. | |
2402 | */ | |
65d87f79 | 2403 | btrfs_add_delayed_iput(&inode->vfs_inode); |
247e743c CM |
2404 | } |
2405 | ||
2406 | /* | |
2407 | * There are a few paths in the higher layers of the kernel that directly | |
2408 | * set the page dirty bit without asking the filesystem if it is a | |
2409 | * good idea. This causes problems because we want to make sure COW | |
2410 | * properly happens and the data=ordered rules are followed. | |
2411 | * | |
c8b97818 | 2412 | * In our case any range that doesn't have the ORDERED bit set |
247e743c CM |
2413 | * hasn't been properly setup for IO. We kick off an async process |
2414 | * to fix it up. The async helper will wait for ordered extents, set | |
2415 | * the delalloc bit and make it safe to write the page. | |
2416 | */ | |
d75855b4 | 2417 | int btrfs_writepage_cow_fixup(struct page *page, u64 start, u64 end) |
247e743c CM |
2418 | { |
2419 | struct inode *inode = page->mapping->host; | |
0b246afa | 2420 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
247e743c | 2421 | struct btrfs_writepage_fixup *fixup; |
247e743c | 2422 | |
8b62b72b CM |
2423 | /* this page is properly in the ordered list */ |
2424 | if (TestClearPagePrivate2(page)) | |
247e743c CM |
2425 | return 0; |
2426 | ||
25f3c502 CM |
2427 | /* |
2428 | * PageChecked is set below when we create a fixup worker for this page, | |
2429 | * don't try to create another one if we're already PageChecked() | |
2430 | * | |
2431 | * The extent_io writepage code will redirty the page if we send back | |
2432 | * EAGAIN. | |
2433 | */ | |
247e743c CM |
2434 | if (PageChecked(page)) |
2435 | return -EAGAIN; | |
2436 | ||
2437 | fixup = kzalloc(sizeof(*fixup), GFP_NOFS); | |
2438 | if (!fixup) | |
2439 | return -EAGAIN; | |
f421950f | 2440 | |
f4b1363c JB |
2441 | /* |
2442 | * We are already holding a reference to this inode from | |
2443 | * write_cache_pages. We need to hold it because the space reservation | |
2444 | * takes place outside of the page lock, and we can't trust | |
2445 | * page->mapping outside of the page lock. | |
2446 | */ | |
2447 | ihold(inode); | |
247e743c | 2448 | SetPageChecked(page); |
09cbfeaf | 2449 | get_page(page); |
a0cac0ec | 2450 | btrfs_init_work(&fixup->work, btrfs_writepage_fixup_worker, NULL, NULL); |
247e743c | 2451 | fixup->page = page; |
f4b1363c | 2452 | fixup->inode = inode; |
0b246afa | 2453 | btrfs_queue_work(fs_info->fixup_workers, &fixup->work); |
25f3c502 CM |
2454 | |
2455 | return -EAGAIN; | |
247e743c CM |
2456 | } |
2457 | ||
d899e052 | 2458 | static int insert_reserved_file_extent(struct btrfs_trans_handle *trans, |
c553f94d | 2459 | struct btrfs_inode *inode, u64 file_pos, |
9729f10a QW |
2460 | struct btrfs_file_extent_item *stack_fi, |
2461 | u64 qgroup_reserved) | |
d899e052 | 2462 | { |
c553f94d | 2463 | struct btrfs_root *root = inode->root; |
d899e052 YZ |
2464 | struct btrfs_path *path; |
2465 | struct extent_buffer *leaf; | |
2466 | struct btrfs_key ins; | |
203f44c5 QW |
2467 | u64 disk_num_bytes = btrfs_stack_file_extent_disk_num_bytes(stack_fi); |
2468 | u64 disk_bytenr = btrfs_stack_file_extent_disk_bytenr(stack_fi); | |
2469 | u64 num_bytes = btrfs_stack_file_extent_num_bytes(stack_fi); | |
2470 | u64 ram_bytes = btrfs_stack_file_extent_ram_bytes(stack_fi); | |
1acae57b | 2471 | int extent_inserted = 0; |
d899e052 YZ |
2472 | int ret; |
2473 | ||
2474 | path = btrfs_alloc_path(); | |
d8926bb3 MF |
2475 | if (!path) |
2476 | return -ENOMEM; | |
d899e052 | 2477 | |
a1ed835e CM |
2478 | /* |
2479 | * we may be replacing one extent in the tree with another. | |
2480 | * The new extent is pinned in the extent map, and we don't want | |
2481 | * to drop it from the cache until it is completely in the btree. | |
2482 | * | |
2483 | * So, tell btrfs_drop_extents to leave this extent in the cache. | |
2484 | * the caller is expected to unpin it and allow it to be merged | |
2485 | * with the others. | |
2486 | */ | |
c553f94d | 2487 | ret = __btrfs_drop_extents(trans, root, inode, path, file_pos, |
1acae57b | 2488 | file_pos + num_bytes, NULL, 0, |
203f44c5 | 2489 | 1, sizeof(*stack_fi), &extent_inserted); |
79787eaa JM |
2490 | if (ret) |
2491 | goto out; | |
d899e052 | 2492 | |
1acae57b | 2493 | if (!extent_inserted) { |
c553f94d | 2494 | ins.objectid = btrfs_ino(inode); |
1acae57b FDBM |
2495 | ins.offset = file_pos; |
2496 | ins.type = BTRFS_EXTENT_DATA_KEY; | |
2497 | ||
2498 | path->leave_spinning = 1; | |
2499 | ret = btrfs_insert_empty_item(trans, root, path, &ins, | |
203f44c5 | 2500 | sizeof(*stack_fi)); |
1acae57b FDBM |
2501 | if (ret) |
2502 | goto out; | |
2503 | } | |
d899e052 | 2504 | leaf = path->nodes[0]; |
203f44c5 QW |
2505 | btrfs_set_stack_file_extent_generation(stack_fi, trans->transid); |
2506 | write_extent_buffer(leaf, stack_fi, | |
2507 | btrfs_item_ptr_offset(leaf, path->slots[0]), | |
2508 | sizeof(struct btrfs_file_extent_item)); | |
b9473439 | 2509 | |
d899e052 | 2510 | btrfs_mark_buffer_dirty(leaf); |
ce195332 | 2511 | btrfs_release_path(path); |
d899e052 | 2512 | |
c553f94d | 2513 | inode_add_bytes(&inode->vfs_inode, num_bytes); |
d899e052 YZ |
2514 | |
2515 | ins.objectid = disk_bytenr; | |
2516 | ins.offset = disk_num_bytes; | |
2517 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
a12b877b | 2518 | |
c553f94d | 2519 | ret = btrfs_inode_set_file_extent_range(inode, file_pos, ram_bytes); |
9ddc959e JB |
2520 | if (ret) |
2521 | goto out; | |
2522 | ||
c553f94d | 2523 | ret = btrfs_alloc_reserved_file_extent(trans, root, btrfs_ino(inode), |
9729f10a | 2524 | file_pos, qgroup_reserved, &ins); |
79787eaa | 2525 | out: |
d899e052 | 2526 | btrfs_free_path(path); |
b9473439 | 2527 | |
79787eaa | 2528 | return ret; |
d899e052 YZ |
2529 | } |
2530 | ||
2ff7e61e | 2531 | static void btrfs_release_delalloc_bytes(struct btrfs_fs_info *fs_info, |
e570fd27 MX |
2532 | u64 start, u64 len) |
2533 | { | |
32da5386 | 2534 | struct btrfs_block_group *cache; |
e570fd27 | 2535 | |
0b246afa | 2536 | cache = btrfs_lookup_block_group(fs_info, start); |
e570fd27 MX |
2537 | ASSERT(cache); |
2538 | ||
2539 | spin_lock(&cache->lock); | |
2540 | cache->delalloc_bytes -= len; | |
2541 | spin_unlock(&cache->lock); | |
2542 | ||
2543 | btrfs_put_block_group(cache); | |
2544 | } | |
2545 | ||
203f44c5 QW |
2546 | static int insert_ordered_extent_file_extent(struct btrfs_trans_handle *trans, |
2547 | struct inode *inode, | |
2548 | struct btrfs_ordered_extent *oe) | |
2549 | { | |
2550 | struct btrfs_file_extent_item stack_fi; | |
2551 | u64 logical_len; | |
2552 | ||
2553 | memset(&stack_fi, 0, sizeof(stack_fi)); | |
2554 | btrfs_set_stack_file_extent_type(&stack_fi, BTRFS_FILE_EXTENT_REG); | |
2555 | btrfs_set_stack_file_extent_disk_bytenr(&stack_fi, oe->disk_bytenr); | |
2556 | btrfs_set_stack_file_extent_disk_num_bytes(&stack_fi, | |
2557 | oe->disk_num_bytes); | |
2558 | if (test_bit(BTRFS_ORDERED_TRUNCATED, &oe->flags)) | |
2559 | logical_len = oe->truncated_len; | |
2560 | else | |
2561 | logical_len = oe->num_bytes; | |
2562 | btrfs_set_stack_file_extent_num_bytes(&stack_fi, logical_len); | |
2563 | btrfs_set_stack_file_extent_ram_bytes(&stack_fi, logical_len); | |
2564 | btrfs_set_stack_file_extent_compression(&stack_fi, oe->compress_type); | |
2565 | /* Encryption and other encoding is reserved and all 0 */ | |
2566 | ||
c553f94d | 2567 | return insert_reserved_file_extent(trans, BTRFS_I(inode), oe->file_offset, |
7dbeaad0 | 2568 | &stack_fi, oe->qgroup_rsv); |
203f44c5 QW |
2569 | } |
2570 | ||
2571 | /* | |
2572 | * As ordered data IO finishes, this gets called so we can finish | |
d352ac68 CM |
2573 | * an ordered extent if the range of bytes in the file it covers are |
2574 | * fully written. | |
2575 | */ | |
5fd02043 | 2576 | static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent) |
e6dcd2dc | 2577 | { |
5fd02043 | 2578 | struct inode *inode = ordered_extent->inode; |
0b246afa | 2579 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
e6dcd2dc | 2580 | struct btrfs_root *root = BTRFS_I(inode)->root; |
0ca1f7ce | 2581 | struct btrfs_trans_handle *trans = NULL; |
e6dcd2dc | 2582 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
2ac55d41 | 2583 | struct extent_state *cached_state = NULL; |
bffe633e | 2584 | u64 start, end; |
261507a0 | 2585 | int compress_type = 0; |
77cef2ec | 2586 | int ret = 0; |
bffe633e | 2587 | u64 logical_len = ordered_extent->num_bytes; |
8d510121 | 2588 | bool freespace_inode; |
77cef2ec | 2589 | bool truncated = false; |
a7e3b975 FM |
2590 | bool range_locked = false; |
2591 | bool clear_new_delalloc_bytes = false; | |
49940bdd | 2592 | bool clear_reserved_extent = true; |
313facc5 | 2593 | unsigned int clear_bits; |
a7e3b975 | 2594 | |
bffe633e OS |
2595 | start = ordered_extent->file_offset; |
2596 | end = start + ordered_extent->num_bytes - 1; | |
2597 | ||
a7e3b975 FM |
2598 | if (!test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) && |
2599 | !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags) && | |
2600 | !test_bit(BTRFS_ORDERED_DIRECT, &ordered_extent->flags)) | |
2601 | clear_new_delalloc_bytes = true; | |
e6dcd2dc | 2602 | |
8d510121 | 2603 | freespace_inode = btrfs_is_free_space_inode(BTRFS_I(inode)); |
0cb59c99 | 2604 | |
5fd02043 JB |
2605 | if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) { |
2606 | ret = -EIO; | |
2607 | goto out; | |
2608 | } | |
2609 | ||
bffe633e | 2610 | btrfs_free_io_failure_record(BTRFS_I(inode), start, end); |
f612496b | 2611 | |
77cef2ec JB |
2612 | if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) { |
2613 | truncated = true; | |
2614 | logical_len = ordered_extent->truncated_len; | |
2615 | /* Truncated the entire extent, don't bother adding */ | |
2616 | if (!logical_len) | |
2617 | goto out; | |
2618 | } | |
2619 | ||
c2167754 | 2620 | if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) { |
79787eaa | 2621 | BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */ |
94ed938a | 2622 | |
d923afe9 | 2623 | btrfs_inode_safe_disk_i_size_write(inode, 0); |
8d510121 NB |
2624 | if (freespace_inode) |
2625 | trans = btrfs_join_transaction_spacecache(root); | |
6c760c07 JB |
2626 | else |
2627 | trans = btrfs_join_transaction(root); | |
2628 | if (IS_ERR(trans)) { | |
2629 | ret = PTR_ERR(trans); | |
2630 | trans = NULL; | |
2631 | goto out; | |
c2167754 | 2632 | } |
69fe2d75 | 2633 | trans->block_rsv = &BTRFS_I(inode)->block_rsv; |
6c760c07 JB |
2634 | ret = btrfs_update_inode_fallback(trans, root, inode); |
2635 | if (ret) /* -ENOMEM or corruption */ | |
66642832 | 2636 | btrfs_abort_transaction(trans, ret); |
c2167754 YZ |
2637 | goto out; |
2638 | } | |
e6dcd2dc | 2639 | |
a7e3b975 | 2640 | range_locked = true; |
bffe633e | 2641 | lock_extent_bits(io_tree, start, end, &cached_state); |
e6dcd2dc | 2642 | |
8d510121 NB |
2643 | if (freespace_inode) |
2644 | trans = btrfs_join_transaction_spacecache(root); | |
0cb59c99 | 2645 | else |
7a7eaa40 | 2646 | trans = btrfs_join_transaction(root); |
79787eaa JM |
2647 | if (IS_ERR(trans)) { |
2648 | ret = PTR_ERR(trans); | |
2649 | trans = NULL; | |
a7e3b975 | 2650 | goto out; |
79787eaa | 2651 | } |
a79b7d4b | 2652 | |
69fe2d75 | 2653 | trans->block_rsv = &BTRFS_I(inode)->block_rsv; |
c2167754 | 2654 | |
c8b97818 | 2655 | if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags)) |
261507a0 | 2656 | compress_type = ordered_extent->compress_type; |
d899e052 | 2657 | if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) { |
261507a0 | 2658 | BUG_ON(compress_type); |
7a6d7067 | 2659 | ret = btrfs_mark_extent_written(trans, BTRFS_I(inode), |
d899e052 YZ |
2660 | ordered_extent->file_offset, |
2661 | ordered_extent->file_offset + | |
77cef2ec | 2662 | logical_len); |
d899e052 | 2663 | } else { |
0b246afa | 2664 | BUG_ON(root == fs_info->tree_root); |
203f44c5 QW |
2665 | ret = insert_ordered_extent_file_extent(trans, inode, |
2666 | ordered_extent); | |
49940bdd JB |
2667 | if (!ret) { |
2668 | clear_reserved_extent = false; | |
2ff7e61e | 2669 | btrfs_release_delalloc_bytes(fs_info, |
bffe633e OS |
2670 | ordered_extent->disk_bytenr, |
2671 | ordered_extent->disk_num_bytes); | |
49940bdd | 2672 | } |
d899e052 | 2673 | } |
5dc562c5 | 2674 | unpin_extent_cache(&BTRFS_I(inode)->extent_tree, |
bffe633e OS |
2675 | ordered_extent->file_offset, |
2676 | ordered_extent->num_bytes, trans->transid); | |
79787eaa | 2677 | if (ret < 0) { |
66642832 | 2678 | btrfs_abort_transaction(trans, ret); |
a7e3b975 | 2679 | goto out; |
79787eaa | 2680 | } |
2ac55d41 | 2681 | |
ac01f26a NB |
2682 | ret = add_pending_csums(trans, inode, &ordered_extent->list); |
2683 | if (ret) { | |
2684 | btrfs_abort_transaction(trans, ret); | |
2685 | goto out; | |
2686 | } | |
e6dcd2dc | 2687 | |
d923afe9 | 2688 | btrfs_inode_safe_disk_i_size_write(inode, 0); |
6c760c07 JB |
2689 | ret = btrfs_update_inode_fallback(trans, root, inode); |
2690 | if (ret) { /* -ENOMEM or corruption */ | |
66642832 | 2691 | btrfs_abort_transaction(trans, ret); |
a7e3b975 | 2692 | goto out; |
1ef30be1 JB |
2693 | } |
2694 | ret = 0; | |
c2167754 | 2695 | out: |
313facc5 OS |
2696 | clear_bits = EXTENT_DEFRAG; |
2697 | if (range_locked) | |
2698 | clear_bits |= EXTENT_LOCKED; | |
2699 | if (clear_new_delalloc_bytes) | |
2700 | clear_bits |= EXTENT_DELALLOC_NEW; | |
bffe633e OS |
2701 | clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, clear_bits, |
2702 | (clear_bits & EXTENT_LOCKED) ? 1 : 0, 0, | |
313facc5 | 2703 | &cached_state); |
a7e3b975 | 2704 | |
a698d075 | 2705 | if (trans) |
3a45bb20 | 2706 | btrfs_end_transaction(trans); |
0cb59c99 | 2707 | |
77cef2ec | 2708 | if (ret || truncated) { |
bffe633e | 2709 | u64 unwritten_start = start; |
77cef2ec JB |
2710 | |
2711 | if (truncated) | |
bffe633e OS |
2712 | unwritten_start += logical_len; |
2713 | clear_extent_uptodate(io_tree, unwritten_start, end, NULL); | |
77cef2ec JB |
2714 | |
2715 | /* Drop the cache for the part of the extent we didn't write. */ | |
bffe633e | 2716 | btrfs_drop_extent_cache(BTRFS_I(inode), unwritten_start, end, 0); |
5fd02043 | 2717 | |
0bec9ef5 JB |
2718 | /* |
2719 | * If the ordered extent had an IOERR or something else went | |
2720 | * wrong we need to return the space for this ordered extent | |
77cef2ec JB |
2721 | * back to the allocator. We only free the extent in the |
2722 | * truncated case if we didn't write out the extent at all. | |
49940bdd JB |
2723 | * |
2724 | * If we made it past insert_reserved_file_extent before we | |
2725 | * errored out then we don't need to do this as the accounting | |
2726 | * has already been done. | |
0bec9ef5 | 2727 | */ |
77cef2ec | 2728 | if ((ret || !logical_len) && |
49940bdd | 2729 | clear_reserved_extent && |
77cef2ec | 2730 | !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) && |
4eaaec24 NB |
2731 | !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) { |
2732 | /* | |
2733 | * Discard the range before returning it back to the | |
2734 | * free space pool | |
2735 | */ | |
46b27f50 | 2736 | if (ret && btrfs_test_opt(fs_info, DISCARD_SYNC)) |
4eaaec24 | 2737 | btrfs_discard_extent(fs_info, |
bffe633e OS |
2738 | ordered_extent->disk_bytenr, |
2739 | ordered_extent->disk_num_bytes, | |
2740 | NULL); | |
2ff7e61e | 2741 | btrfs_free_reserved_extent(fs_info, |
bffe633e OS |
2742 | ordered_extent->disk_bytenr, |
2743 | ordered_extent->disk_num_bytes, 1); | |
4eaaec24 | 2744 | } |
0bec9ef5 JB |
2745 | } |
2746 | ||
5fd02043 | 2747 | /* |
8bad3c02 LB |
2748 | * This needs to be done to make sure anybody waiting knows we are done |
2749 | * updating everything for this ordered extent. | |
5fd02043 JB |
2750 | */ |
2751 | btrfs_remove_ordered_extent(inode, ordered_extent); | |
2752 | ||
e6dcd2dc CM |
2753 | /* once for us */ |
2754 | btrfs_put_ordered_extent(ordered_extent); | |
2755 | /* once for the tree */ | |
2756 | btrfs_put_ordered_extent(ordered_extent); | |
2757 | ||
5fd02043 JB |
2758 | return ret; |
2759 | } | |
2760 | ||
2761 | static void finish_ordered_fn(struct btrfs_work *work) | |
2762 | { | |
2763 | struct btrfs_ordered_extent *ordered_extent; | |
2764 | ordered_extent = container_of(work, struct btrfs_ordered_extent, work); | |
2765 | btrfs_finish_ordered_io(ordered_extent); | |
e6dcd2dc CM |
2766 | } |
2767 | ||
c629732d NB |
2768 | void btrfs_writepage_endio_finish_ordered(struct page *page, u64 start, |
2769 | u64 end, int uptodate) | |
211f90e6 | 2770 | { |
5fd02043 | 2771 | struct inode *inode = page->mapping->host; |
0b246afa | 2772 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
5fd02043 | 2773 | struct btrfs_ordered_extent *ordered_extent = NULL; |
9e0af237 | 2774 | struct btrfs_workqueue *wq; |
5fd02043 | 2775 | |
1abe9b8a | 2776 | trace_btrfs_writepage_end_io_hook(page, start, end, uptodate); |
2777 | ||
8b62b72b | 2778 | ClearPagePrivate2(page); |
5fd02043 JB |
2779 | if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start, |
2780 | end - start + 1, uptodate)) | |
c3988d63 | 2781 | return; |
5fd02043 | 2782 | |
a0cac0ec | 2783 | if (btrfs_is_free_space_inode(BTRFS_I(inode))) |
0b246afa | 2784 | wq = fs_info->endio_freespace_worker; |
a0cac0ec | 2785 | else |
0b246afa | 2786 | wq = fs_info->endio_write_workers; |
5fd02043 | 2787 | |
a0cac0ec | 2788 | btrfs_init_work(&ordered_extent->work, finish_ordered_fn, NULL, NULL); |
9e0af237 | 2789 | btrfs_queue_work(wq, &ordered_extent->work); |
211f90e6 CM |
2790 | } |
2791 | ||
47df7765 OS |
2792 | static int check_data_csum(struct inode *inode, struct btrfs_io_bio *io_bio, |
2793 | int icsum, struct page *page, int pgoff, u64 start, | |
2794 | size_t len) | |
dc380aea | 2795 | { |
d5178578 JT |
2796 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
2797 | SHASH_DESC_ON_STACK(shash, fs_info->csum_shash); | |
dc380aea | 2798 | char *kaddr; |
d5178578 JT |
2799 | u16 csum_size = btrfs_super_csum_size(fs_info->super_copy); |
2800 | u8 *csum_expected; | |
2801 | u8 csum[BTRFS_CSUM_SIZE]; | |
dc380aea | 2802 | |
d5178578 | 2803 | csum_expected = ((u8 *)io_bio->csum) + icsum * csum_size; |
dc380aea MX |
2804 | |
2805 | kaddr = kmap_atomic(page); | |
d5178578 JT |
2806 | shash->tfm = fs_info->csum_shash; |
2807 | ||
fd08001f | 2808 | crypto_shash_digest(shash, kaddr + pgoff, len, csum); |
d5178578 JT |
2809 | |
2810 | if (memcmp(csum, csum_expected, csum_size)) | |
dc380aea MX |
2811 | goto zeroit; |
2812 | ||
2813 | kunmap_atomic(kaddr); | |
2814 | return 0; | |
2815 | zeroit: | |
ea41d6b2 JT |
2816 | btrfs_print_data_csum_error(BTRFS_I(inode), start, csum, csum_expected, |
2817 | io_bio->mirror_num); | |
dc380aea MX |
2818 | memset(kaddr + pgoff, 1, len); |
2819 | flush_dcache_page(page); | |
2820 | kunmap_atomic(kaddr); | |
dc380aea MX |
2821 | return -EIO; |
2822 | } | |
2823 | ||
d352ac68 CM |
2824 | /* |
2825 | * when reads are done, we need to check csums to verify the data is correct | |
4a54c8c1 JS |
2826 | * if there's a match, we allow the bio to finish. If not, the code in |
2827 | * extent_io.c will try to find good copies for us. | |
d352ac68 | 2828 | */ |
facc8a22 MX |
2829 | static int btrfs_readpage_end_io_hook(struct btrfs_io_bio *io_bio, |
2830 | u64 phy_offset, struct page *page, | |
2831 | u64 start, u64 end, int mirror) | |
07157aac | 2832 | { |
4eee4fa4 | 2833 | size_t offset = start - page_offset(page); |
07157aac | 2834 | struct inode *inode = page->mapping->host; |
d1310b2e | 2835 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
ff79f819 | 2836 | struct btrfs_root *root = BTRFS_I(inode)->root; |
d1310b2e | 2837 | |
d20f7043 CM |
2838 | if (PageChecked(page)) { |
2839 | ClearPageChecked(page); | |
dc380aea | 2840 | return 0; |
d20f7043 | 2841 | } |
6cbff00f CH |
2842 | |
2843 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM) | |
dc380aea | 2844 | return 0; |
17d217fe YZ |
2845 | |
2846 | if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID && | |
9655d298 | 2847 | test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) { |
91166212 | 2848 | clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM); |
b6cda9bc | 2849 | return 0; |
17d217fe | 2850 | } |
d20f7043 | 2851 | |
facc8a22 | 2852 | phy_offset >>= inode->i_sb->s_blocksize_bits; |
47df7765 OS |
2853 | return check_data_csum(inode, io_bio, phy_offset, page, offset, start, |
2854 | (size_t)(end - start + 1)); | |
07157aac | 2855 | } |
b888db2b | 2856 | |
c1c3fac2 NB |
2857 | /* |
2858 | * btrfs_add_delayed_iput - perform a delayed iput on @inode | |
2859 | * | |
2860 | * @inode: The inode we want to perform iput on | |
2861 | * | |
2862 | * This function uses the generic vfs_inode::i_count to track whether we should | |
2863 | * just decrement it (in case it's > 1) or if this is the last iput then link | |
2864 | * the inode to the delayed iput machinery. Delayed iputs are processed at | |
2865 | * transaction commit time/superblock commit/cleaner kthread. | |
2866 | */ | |
24bbcf04 YZ |
2867 | void btrfs_add_delayed_iput(struct inode *inode) |
2868 | { | |
0b246afa | 2869 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
8089fe62 | 2870 | struct btrfs_inode *binode = BTRFS_I(inode); |
24bbcf04 YZ |
2871 | |
2872 | if (atomic_add_unless(&inode->i_count, -1, 1)) | |
2873 | return; | |
2874 | ||
034f784d | 2875 | atomic_inc(&fs_info->nr_delayed_iputs); |
24bbcf04 | 2876 | spin_lock(&fs_info->delayed_iput_lock); |
c1c3fac2 NB |
2877 | ASSERT(list_empty(&binode->delayed_iput)); |
2878 | list_add_tail(&binode->delayed_iput, &fs_info->delayed_iputs); | |
24bbcf04 | 2879 | spin_unlock(&fs_info->delayed_iput_lock); |
fd340d0f JB |
2880 | if (!test_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags)) |
2881 | wake_up_process(fs_info->cleaner_kthread); | |
24bbcf04 YZ |
2882 | } |
2883 | ||
63611e73 JB |
2884 | static void run_delayed_iput_locked(struct btrfs_fs_info *fs_info, |
2885 | struct btrfs_inode *inode) | |
2886 | { | |
2887 | list_del_init(&inode->delayed_iput); | |
2888 | spin_unlock(&fs_info->delayed_iput_lock); | |
2889 | iput(&inode->vfs_inode); | |
2890 | if (atomic_dec_and_test(&fs_info->nr_delayed_iputs)) | |
2891 | wake_up(&fs_info->delayed_iputs_wait); | |
2892 | spin_lock(&fs_info->delayed_iput_lock); | |
2893 | } | |
2894 | ||
2895 | static void btrfs_run_delayed_iput(struct btrfs_fs_info *fs_info, | |
2896 | struct btrfs_inode *inode) | |
2897 | { | |
2898 | if (!list_empty(&inode->delayed_iput)) { | |
2899 | spin_lock(&fs_info->delayed_iput_lock); | |
2900 | if (!list_empty(&inode->delayed_iput)) | |
2901 | run_delayed_iput_locked(fs_info, inode); | |
2902 | spin_unlock(&fs_info->delayed_iput_lock); | |
2903 | } | |
2904 | } | |
2905 | ||
2ff7e61e | 2906 | void btrfs_run_delayed_iputs(struct btrfs_fs_info *fs_info) |
24bbcf04 | 2907 | { |
24bbcf04 | 2908 | |
24bbcf04 | 2909 | spin_lock(&fs_info->delayed_iput_lock); |
8089fe62 DS |
2910 | while (!list_empty(&fs_info->delayed_iputs)) { |
2911 | struct btrfs_inode *inode; | |
2912 | ||
2913 | inode = list_first_entry(&fs_info->delayed_iputs, | |
2914 | struct btrfs_inode, delayed_iput); | |
63611e73 | 2915 | run_delayed_iput_locked(fs_info, inode); |
24bbcf04 | 2916 | } |
8089fe62 | 2917 | spin_unlock(&fs_info->delayed_iput_lock); |
24bbcf04 YZ |
2918 | } |
2919 | ||
034f784d JB |
2920 | /** |
2921 | * btrfs_wait_on_delayed_iputs - wait on the delayed iputs to be done running | |
2922 | * @fs_info - the fs_info for this fs | |
2923 | * @return - EINTR if we were killed, 0 if nothing's pending | |
2924 | * | |
2925 | * This will wait on any delayed iputs that are currently running with KILLABLE | |
2926 | * set. Once they are all done running we will return, unless we are killed in | |
2927 | * which case we return EINTR. This helps in user operations like fallocate etc | |
2928 | * that might get blocked on the iputs. | |
2929 | */ | |
2930 | int btrfs_wait_on_delayed_iputs(struct btrfs_fs_info *fs_info) | |
2931 | { | |
2932 | int ret = wait_event_killable(fs_info->delayed_iputs_wait, | |
2933 | atomic_read(&fs_info->nr_delayed_iputs) == 0); | |
2934 | if (ret) | |
2935 | return -EINTR; | |
2936 | return 0; | |
2937 | } | |
2938 | ||
7b128766 | 2939 | /* |
f7e9e8fc OS |
2940 | * This creates an orphan entry for the given inode in case something goes wrong |
2941 | * in the middle of an unlink. | |
7b128766 | 2942 | */ |
73f2e545 | 2943 | int btrfs_orphan_add(struct btrfs_trans_handle *trans, |
27919067 | 2944 | struct btrfs_inode *inode) |
7b128766 | 2945 | { |
d68fc57b | 2946 | int ret; |
7b128766 | 2947 | |
27919067 OS |
2948 | ret = btrfs_insert_orphan_item(trans, inode->root, btrfs_ino(inode)); |
2949 | if (ret && ret != -EEXIST) { | |
2950 | btrfs_abort_transaction(trans, ret); | |
2951 | return ret; | |
d68fc57b YZ |
2952 | } |
2953 | ||
d68fc57b | 2954 | return 0; |
7b128766 JB |
2955 | } |
2956 | ||
2957 | /* | |
f7e9e8fc OS |
2958 | * We have done the delete so we can go ahead and remove the orphan item for |
2959 | * this particular inode. | |
7b128766 | 2960 | */ |
48a3b636 | 2961 | static int btrfs_orphan_del(struct btrfs_trans_handle *trans, |
3d6ae7bb | 2962 | struct btrfs_inode *inode) |
7b128766 | 2963 | { |
27919067 | 2964 | return btrfs_del_orphan_item(trans, inode->root, btrfs_ino(inode)); |
7b128766 JB |
2965 | } |
2966 | ||
2967 | /* | |
2968 | * this cleans up any orphans that may be left on the list from the last use | |
2969 | * of this root. | |
2970 | */ | |
66b4ffd1 | 2971 | int btrfs_orphan_cleanup(struct btrfs_root *root) |
7b128766 | 2972 | { |
0b246afa | 2973 | struct btrfs_fs_info *fs_info = root->fs_info; |
7b128766 JB |
2974 | struct btrfs_path *path; |
2975 | struct extent_buffer *leaf; | |
7b128766 JB |
2976 | struct btrfs_key key, found_key; |
2977 | struct btrfs_trans_handle *trans; | |
2978 | struct inode *inode; | |
8f6d7f4f | 2979 | u64 last_objectid = 0; |
f7e9e8fc | 2980 | int ret = 0, nr_unlink = 0; |
7b128766 | 2981 | |
d68fc57b | 2982 | if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED)) |
66b4ffd1 | 2983 | return 0; |
c71bf099 YZ |
2984 | |
2985 | path = btrfs_alloc_path(); | |
66b4ffd1 JB |
2986 | if (!path) { |
2987 | ret = -ENOMEM; | |
2988 | goto out; | |
2989 | } | |
e4058b54 | 2990 | path->reada = READA_BACK; |
7b128766 JB |
2991 | |
2992 | key.objectid = BTRFS_ORPHAN_OBJECTID; | |
962a298f | 2993 | key.type = BTRFS_ORPHAN_ITEM_KEY; |
7b128766 JB |
2994 | key.offset = (u64)-1; |
2995 | ||
7b128766 JB |
2996 | while (1) { |
2997 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
66b4ffd1 JB |
2998 | if (ret < 0) |
2999 | goto out; | |
7b128766 JB |
3000 | |
3001 | /* | |
3002 | * if ret == 0 means we found what we were searching for, which | |
25985edc | 3003 | * is weird, but possible, so only screw with path if we didn't |
7b128766 JB |
3004 | * find the key and see if we have stuff that matches |
3005 | */ | |
3006 | if (ret > 0) { | |
66b4ffd1 | 3007 | ret = 0; |
7b128766 JB |
3008 | if (path->slots[0] == 0) |
3009 | break; | |
3010 | path->slots[0]--; | |
3011 | } | |
3012 | ||
3013 | /* pull out the item */ | |
3014 | leaf = path->nodes[0]; | |
7b128766 JB |
3015 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
3016 | ||
3017 | /* make sure the item matches what we want */ | |
3018 | if (found_key.objectid != BTRFS_ORPHAN_OBJECTID) | |
3019 | break; | |
962a298f | 3020 | if (found_key.type != BTRFS_ORPHAN_ITEM_KEY) |
7b128766 JB |
3021 | break; |
3022 | ||
3023 | /* release the path since we're done with it */ | |
b3b4aa74 | 3024 | btrfs_release_path(path); |
7b128766 JB |
3025 | |
3026 | /* | |
3027 | * this is where we are basically btrfs_lookup, without the | |
3028 | * crossing root thing. we store the inode number in the | |
3029 | * offset of the orphan item. | |
3030 | */ | |
8f6d7f4f JB |
3031 | |
3032 | if (found_key.offset == last_objectid) { | |
0b246afa JM |
3033 | btrfs_err(fs_info, |
3034 | "Error removing orphan entry, stopping orphan cleanup"); | |
8f6d7f4f JB |
3035 | ret = -EINVAL; |
3036 | goto out; | |
3037 | } | |
3038 | ||
3039 | last_objectid = found_key.offset; | |
3040 | ||
5d4f98a2 YZ |
3041 | found_key.objectid = found_key.offset; |
3042 | found_key.type = BTRFS_INODE_ITEM_KEY; | |
3043 | found_key.offset = 0; | |
0202e83f | 3044 | inode = btrfs_iget(fs_info->sb, last_objectid, root); |
8c6ffba0 | 3045 | ret = PTR_ERR_OR_ZERO(inode); |
67710892 | 3046 | if (ret && ret != -ENOENT) |
66b4ffd1 | 3047 | goto out; |
7b128766 | 3048 | |
0b246afa | 3049 | if (ret == -ENOENT && root == fs_info->tree_root) { |
f8e9e0b0 AJ |
3050 | struct btrfs_root *dead_root; |
3051 | struct btrfs_fs_info *fs_info = root->fs_info; | |
3052 | int is_dead_root = 0; | |
3053 | ||
3054 | /* | |
3055 | * this is an orphan in the tree root. Currently these | |
3056 | * could come from 2 sources: | |
3057 | * a) a snapshot deletion in progress | |
3058 | * b) a free space cache inode | |
3059 | * We need to distinguish those two, as the snapshot | |
3060 | * orphan must not get deleted. | |
3061 | * find_dead_roots already ran before us, so if this | |
3062 | * is a snapshot deletion, we should find the root | |
a619b3c7 | 3063 | * in the fs_roots radix tree. |
f8e9e0b0 | 3064 | */ |
a619b3c7 RK |
3065 | |
3066 | spin_lock(&fs_info->fs_roots_radix_lock); | |
3067 | dead_root = radix_tree_lookup(&fs_info->fs_roots_radix, | |
3068 | (unsigned long)found_key.objectid); | |
3069 | if (dead_root && btrfs_root_refs(&dead_root->root_item) == 0) | |
3070 | is_dead_root = 1; | |
3071 | spin_unlock(&fs_info->fs_roots_radix_lock); | |
3072 | ||
f8e9e0b0 AJ |
3073 | if (is_dead_root) { |
3074 | /* prevent this orphan from being found again */ | |
3075 | key.offset = found_key.objectid - 1; | |
3076 | continue; | |
3077 | } | |
f7e9e8fc | 3078 | |
f8e9e0b0 | 3079 | } |
f7e9e8fc | 3080 | |
7b128766 | 3081 | /* |
f7e9e8fc OS |
3082 | * If we have an inode with links, there are a couple of |
3083 | * possibilities. Old kernels (before v3.12) used to create an | |
3084 | * orphan item for truncate indicating that there were possibly | |
3085 | * extent items past i_size that needed to be deleted. In v3.12, | |
3086 | * truncate was changed to update i_size in sync with the extent | |
3087 | * items, but the (useless) orphan item was still created. Since | |
3088 | * v4.18, we don't create the orphan item for truncate at all. | |
3089 | * | |
3090 | * So, this item could mean that we need to do a truncate, but | |
3091 | * only if this filesystem was last used on a pre-v3.12 kernel | |
3092 | * and was not cleanly unmounted. The odds of that are quite | |
3093 | * slim, and it's a pain to do the truncate now, so just delete | |
3094 | * the orphan item. | |
3095 | * | |
3096 | * It's also possible that this orphan item was supposed to be | |
3097 | * deleted but wasn't. The inode number may have been reused, | |
3098 | * but either way, we can delete the orphan item. | |
7b128766 | 3099 | */ |
f7e9e8fc OS |
3100 | if (ret == -ENOENT || inode->i_nlink) { |
3101 | if (!ret) | |
3102 | iput(inode); | |
a8c9e576 | 3103 | trans = btrfs_start_transaction(root, 1); |
66b4ffd1 JB |
3104 | if (IS_ERR(trans)) { |
3105 | ret = PTR_ERR(trans); | |
3106 | goto out; | |
3107 | } | |
0b246afa JM |
3108 | btrfs_debug(fs_info, "auto deleting %Lu", |
3109 | found_key.objectid); | |
a8c9e576 JB |
3110 | ret = btrfs_del_orphan_item(trans, root, |
3111 | found_key.objectid); | |
3a45bb20 | 3112 | btrfs_end_transaction(trans); |
4ef31a45 JB |
3113 | if (ret) |
3114 | goto out; | |
7b128766 JB |
3115 | continue; |
3116 | } | |
3117 | ||
f7e9e8fc | 3118 | nr_unlink++; |
7b128766 JB |
3119 | |
3120 | /* this will do delete_inode and everything for us */ | |
3121 | iput(inode); | |
3122 | } | |
3254c876 MX |
3123 | /* release the path since we're done with it */ |
3124 | btrfs_release_path(path); | |
3125 | ||
d68fc57b YZ |
3126 | root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE; |
3127 | ||
a575ceeb | 3128 | if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state)) { |
7a7eaa40 | 3129 | trans = btrfs_join_transaction(root); |
66b4ffd1 | 3130 | if (!IS_ERR(trans)) |
3a45bb20 | 3131 | btrfs_end_transaction(trans); |
d68fc57b | 3132 | } |
7b128766 JB |
3133 | |
3134 | if (nr_unlink) | |
0b246afa | 3135 | btrfs_debug(fs_info, "unlinked %d orphans", nr_unlink); |
66b4ffd1 JB |
3136 | |
3137 | out: | |
3138 | if (ret) | |
0b246afa | 3139 | btrfs_err(fs_info, "could not do orphan cleanup %d", ret); |
66b4ffd1 JB |
3140 | btrfs_free_path(path); |
3141 | return ret; | |
7b128766 JB |
3142 | } |
3143 | ||
46a53cca CM |
3144 | /* |
3145 | * very simple check to peek ahead in the leaf looking for xattrs. If we | |
3146 | * don't find any xattrs, we know there can't be any acls. | |
3147 | * | |
3148 | * slot is the slot the inode is in, objectid is the objectid of the inode | |
3149 | */ | |
3150 | static noinline int acls_after_inode_item(struct extent_buffer *leaf, | |
63541927 FDBM |
3151 | int slot, u64 objectid, |
3152 | int *first_xattr_slot) | |
46a53cca CM |
3153 | { |
3154 | u32 nritems = btrfs_header_nritems(leaf); | |
3155 | struct btrfs_key found_key; | |
f23b5a59 JB |
3156 | static u64 xattr_access = 0; |
3157 | static u64 xattr_default = 0; | |
46a53cca CM |
3158 | int scanned = 0; |
3159 | ||
f23b5a59 | 3160 | if (!xattr_access) { |
97d79299 AG |
3161 | xattr_access = btrfs_name_hash(XATTR_NAME_POSIX_ACL_ACCESS, |
3162 | strlen(XATTR_NAME_POSIX_ACL_ACCESS)); | |
3163 | xattr_default = btrfs_name_hash(XATTR_NAME_POSIX_ACL_DEFAULT, | |
3164 | strlen(XATTR_NAME_POSIX_ACL_DEFAULT)); | |
f23b5a59 JB |
3165 | } |
3166 | ||
46a53cca | 3167 | slot++; |
63541927 | 3168 | *first_xattr_slot = -1; |
46a53cca CM |
3169 | while (slot < nritems) { |
3170 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
3171 | ||
3172 | /* we found a different objectid, there must not be acls */ | |
3173 | if (found_key.objectid != objectid) | |
3174 | return 0; | |
3175 | ||
3176 | /* we found an xattr, assume we've got an acl */ | |
f23b5a59 | 3177 | if (found_key.type == BTRFS_XATTR_ITEM_KEY) { |
63541927 FDBM |
3178 | if (*first_xattr_slot == -1) |
3179 | *first_xattr_slot = slot; | |
f23b5a59 JB |
3180 | if (found_key.offset == xattr_access || |
3181 | found_key.offset == xattr_default) | |
3182 | return 1; | |
3183 | } | |
46a53cca CM |
3184 | |
3185 | /* | |
3186 | * we found a key greater than an xattr key, there can't | |
3187 | * be any acls later on | |
3188 | */ | |
3189 | if (found_key.type > BTRFS_XATTR_ITEM_KEY) | |
3190 | return 0; | |
3191 | ||
3192 | slot++; | |
3193 | scanned++; | |
3194 | ||
3195 | /* | |
3196 | * it goes inode, inode backrefs, xattrs, extents, | |
3197 | * so if there are a ton of hard links to an inode there can | |
3198 | * be a lot of backrefs. Don't waste time searching too hard, | |
3199 | * this is just an optimization | |
3200 | */ | |
3201 | if (scanned >= 8) | |
3202 | break; | |
3203 | } | |
3204 | /* we hit the end of the leaf before we found an xattr or | |
3205 | * something larger than an xattr. We have to assume the inode | |
3206 | * has acls | |
3207 | */ | |
63541927 FDBM |
3208 | if (*first_xattr_slot == -1) |
3209 | *first_xattr_slot = slot; | |
46a53cca CM |
3210 | return 1; |
3211 | } | |
3212 | ||
d352ac68 CM |
3213 | /* |
3214 | * read an inode from the btree into the in-memory inode | |
3215 | */ | |
4222ea71 FM |
3216 | static int btrfs_read_locked_inode(struct inode *inode, |
3217 | struct btrfs_path *in_path) | |
39279cc3 | 3218 | { |
0b246afa | 3219 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
4222ea71 | 3220 | struct btrfs_path *path = in_path; |
5f39d397 | 3221 | struct extent_buffer *leaf; |
39279cc3 CM |
3222 | struct btrfs_inode_item *inode_item; |
3223 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
3224 | struct btrfs_key location; | |
67de1176 | 3225 | unsigned long ptr; |
46a53cca | 3226 | int maybe_acls; |
618e21d5 | 3227 | u32 rdev; |
39279cc3 | 3228 | int ret; |
2f7e33d4 | 3229 | bool filled = false; |
63541927 | 3230 | int first_xattr_slot; |
2f7e33d4 MX |
3231 | |
3232 | ret = btrfs_fill_inode(inode, &rdev); | |
3233 | if (!ret) | |
3234 | filled = true; | |
39279cc3 | 3235 | |
4222ea71 FM |
3236 | if (!path) { |
3237 | path = btrfs_alloc_path(); | |
3238 | if (!path) | |
3239 | return -ENOMEM; | |
3240 | } | |
1748f843 | 3241 | |
39279cc3 | 3242 | memcpy(&location, &BTRFS_I(inode)->location, sizeof(location)); |
dc17ff8f | 3243 | |
39279cc3 | 3244 | ret = btrfs_lookup_inode(NULL, root, path, &location, 0); |
67710892 | 3245 | if (ret) { |
4222ea71 FM |
3246 | if (path != in_path) |
3247 | btrfs_free_path(path); | |
f5b3a417 | 3248 | return ret; |
67710892 | 3249 | } |
39279cc3 | 3250 | |
5f39d397 | 3251 | leaf = path->nodes[0]; |
2f7e33d4 MX |
3252 | |
3253 | if (filled) | |
67de1176 | 3254 | goto cache_index; |
2f7e33d4 | 3255 | |
5f39d397 CM |
3256 | inode_item = btrfs_item_ptr(leaf, path->slots[0], |
3257 | struct btrfs_inode_item); | |
5f39d397 | 3258 | inode->i_mode = btrfs_inode_mode(leaf, inode_item); |
bfe86848 | 3259 | set_nlink(inode, btrfs_inode_nlink(leaf, inode_item)); |
2f2f43d3 EB |
3260 | i_uid_write(inode, btrfs_inode_uid(leaf, inode_item)); |
3261 | i_gid_write(inode, btrfs_inode_gid(leaf, inode_item)); | |
6ef06d27 | 3262 | btrfs_i_size_write(BTRFS_I(inode), btrfs_inode_size(leaf, inode_item)); |
41a2ee75 JB |
3263 | btrfs_inode_set_file_extent_range(BTRFS_I(inode), 0, |
3264 | round_up(i_size_read(inode), fs_info->sectorsize)); | |
5f39d397 | 3265 | |
a937b979 DS |
3266 | inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->atime); |
3267 | inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->atime); | |
5f39d397 | 3268 | |
a937b979 DS |
3269 | inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->mtime); |
3270 | inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->mtime); | |
5f39d397 | 3271 | |
a937b979 DS |
3272 | inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->ctime); |
3273 | inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->ctime); | |
5f39d397 | 3274 | |
9cc97d64 | 3275 | BTRFS_I(inode)->i_otime.tv_sec = |
3276 | btrfs_timespec_sec(leaf, &inode_item->otime); | |
3277 | BTRFS_I(inode)->i_otime.tv_nsec = | |
3278 | btrfs_timespec_nsec(leaf, &inode_item->otime); | |
5f39d397 | 3279 | |
a76a3cd4 | 3280 | inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item)); |
e02119d5 | 3281 | BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item); |
5dc562c5 JB |
3282 | BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item); |
3283 | ||
c7f88c4e JL |
3284 | inode_set_iversion_queried(inode, |
3285 | btrfs_inode_sequence(leaf, inode_item)); | |
6e17d30b YD |
3286 | inode->i_generation = BTRFS_I(inode)->generation; |
3287 | inode->i_rdev = 0; | |
3288 | rdev = btrfs_inode_rdev(leaf, inode_item); | |
3289 | ||
3290 | BTRFS_I(inode)->index_cnt = (u64)-1; | |
3291 | BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item); | |
3292 | ||
3293 | cache_index: | |
5dc562c5 JB |
3294 | /* |
3295 | * If we were modified in the current generation and evicted from memory | |
3296 | * and then re-read we need to do a full sync since we don't have any | |
3297 | * idea about which extents were modified before we were evicted from | |
3298 | * cache. | |
6e17d30b YD |
3299 | * |
3300 | * This is required for both inode re-read from disk and delayed inode | |
3301 | * in delayed_nodes_tree. | |
5dc562c5 | 3302 | */ |
0b246afa | 3303 | if (BTRFS_I(inode)->last_trans == fs_info->generation) |
5dc562c5 JB |
3304 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, |
3305 | &BTRFS_I(inode)->runtime_flags); | |
3306 | ||
bde6c242 FM |
3307 | /* |
3308 | * We don't persist the id of the transaction where an unlink operation | |
3309 | * against the inode was last made. So here we assume the inode might | |
3310 | * have been evicted, and therefore the exact value of last_unlink_trans | |
3311 | * lost, and set it to last_trans to avoid metadata inconsistencies | |
3312 | * between the inode and its parent if the inode is fsync'ed and the log | |
3313 | * replayed. For example, in the scenario: | |
3314 | * | |
3315 | * touch mydir/foo | |
3316 | * ln mydir/foo mydir/bar | |
3317 | * sync | |
3318 | * unlink mydir/bar | |
3319 | * echo 2 > /proc/sys/vm/drop_caches # evicts inode | |
3320 | * xfs_io -c fsync mydir/foo | |
3321 | * <power failure> | |
3322 | * mount fs, triggers fsync log replay | |
3323 | * | |
3324 | * We must make sure that when we fsync our inode foo we also log its | |
3325 | * parent inode, otherwise after log replay the parent still has the | |
3326 | * dentry with the "bar" name but our inode foo has a link count of 1 | |
3327 | * and doesn't have an inode ref with the name "bar" anymore. | |
3328 | * | |
3329 | * Setting last_unlink_trans to last_trans is a pessimistic approach, | |
01327610 | 3330 | * but it guarantees correctness at the expense of occasional full |
bde6c242 FM |
3331 | * transaction commits on fsync if our inode is a directory, or if our |
3332 | * inode is not a directory, logging its parent unnecessarily. | |
3333 | */ | |
3334 | BTRFS_I(inode)->last_unlink_trans = BTRFS_I(inode)->last_trans; | |
3335 | ||
67de1176 MX |
3336 | path->slots[0]++; |
3337 | if (inode->i_nlink != 1 || | |
3338 | path->slots[0] >= btrfs_header_nritems(leaf)) | |
3339 | goto cache_acl; | |
3340 | ||
3341 | btrfs_item_key_to_cpu(leaf, &location, path->slots[0]); | |
4a0cc7ca | 3342 | if (location.objectid != btrfs_ino(BTRFS_I(inode))) |
67de1176 MX |
3343 | goto cache_acl; |
3344 | ||
3345 | ptr = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
3346 | if (location.type == BTRFS_INODE_REF_KEY) { | |
3347 | struct btrfs_inode_ref *ref; | |
3348 | ||
3349 | ref = (struct btrfs_inode_ref *)ptr; | |
3350 | BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref); | |
3351 | } else if (location.type == BTRFS_INODE_EXTREF_KEY) { | |
3352 | struct btrfs_inode_extref *extref; | |
3353 | ||
3354 | extref = (struct btrfs_inode_extref *)ptr; | |
3355 | BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf, | |
3356 | extref); | |
3357 | } | |
2f7e33d4 | 3358 | cache_acl: |
46a53cca CM |
3359 | /* |
3360 | * try to precache a NULL acl entry for files that don't have | |
3361 | * any xattrs or acls | |
3362 | */ | |
33345d01 | 3363 | maybe_acls = acls_after_inode_item(leaf, path->slots[0], |
f85b7379 | 3364 | btrfs_ino(BTRFS_I(inode)), &first_xattr_slot); |
63541927 FDBM |
3365 | if (first_xattr_slot != -1) { |
3366 | path->slots[0] = first_xattr_slot; | |
3367 | ret = btrfs_load_inode_props(inode, path); | |
3368 | if (ret) | |
0b246afa | 3369 | btrfs_err(fs_info, |
351fd353 | 3370 | "error loading props for ino %llu (root %llu): %d", |
4a0cc7ca | 3371 | btrfs_ino(BTRFS_I(inode)), |
63541927 FDBM |
3372 | root->root_key.objectid, ret); |
3373 | } | |
4222ea71 FM |
3374 | if (path != in_path) |
3375 | btrfs_free_path(path); | |
63541927 | 3376 | |
72c04902 AV |
3377 | if (!maybe_acls) |
3378 | cache_no_acl(inode); | |
46a53cca | 3379 | |
39279cc3 | 3380 | switch (inode->i_mode & S_IFMT) { |
39279cc3 CM |
3381 | case S_IFREG: |
3382 | inode->i_mapping->a_ops = &btrfs_aops; | |
d1310b2e | 3383 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; |
39279cc3 CM |
3384 | inode->i_fop = &btrfs_file_operations; |
3385 | inode->i_op = &btrfs_file_inode_operations; | |
3386 | break; | |
3387 | case S_IFDIR: | |
3388 | inode->i_fop = &btrfs_dir_file_operations; | |
67ade058 | 3389 | inode->i_op = &btrfs_dir_inode_operations; |
39279cc3 CM |
3390 | break; |
3391 | case S_IFLNK: | |
3392 | inode->i_op = &btrfs_symlink_inode_operations; | |
21fc61c7 | 3393 | inode_nohighmem(inode); |
4779cc04 | 3394 | inode->i_mapping->a_ops = &btrfs_aops; |
39279cc3 | 3395 | break; |
618e21d5 | 3396 | default: |
0279b4cd | 3397 | inode->i_op = &btrfs_special_inode_operations; |
618e21d5 JB |
3398 | init_special_inode(inode, inode->i_mode, rdev); |
3399 | break; | |
39279cc3 | 3400 | } |
6cbff00f | 3401 | |
7b6a221e | 3402 | btrfs_sync_inode_flags_to_i_flags(inode); |
67710892 | 3403 | return 0; |
39279cc3 CM |
3404 | } |
3405 | ||
d352ac68 CM |
3406 | /* |
3407 | * given a leaf and an inode, copy the inode fields into the leaf | |
3408 | */ | |
e02119d5 CM |
3409 | static void fill_inode_item(struct btrfs_trans_handle *trans, |
3410 | struct extent_buffer *leaf, | |
5f39d397 | 3411 | struct btrfs_inode_item *item, |
39279cc3 CM |
3412 | struct inode *inode) |
3413 | { | |
51fab693 LB |
3414 | struct btrfs_map_token token; |
3415 | ||
c82f823c | 3416 | btrfs_init_map_token(&token, leaf); |
5f39d397 | 3417 | |
cc4c13d5 DS |
3418 | btrfs_set_token_inode_uid(&token, item, i_uid_read(inode)); |
3419 | btrfs_set_token_inode_gid(&token, item, i_gid_read(inode)); | |
3420 | btrfs_set_token_inode_size(&token, item, BTRFS_I(inode)->disk_i_size); | |
3421 | btrfs_set_token_inode_mode(&token, item, inode->i_mode); | |
3422 | btrfs_set_token_inode_nlink(&token, item, inode->i_nlink); | |
3423 | ||
3424 | btrfs_set_token_timespec_sec(&token, &item->atime, | |
3425 | inode->i_atime.tv_sec); | |
3426 | btrfs_set_token_timespec_nsec(&token, &item->atime, | |
3427 | inode->i_atime.tv_nsec); | |
3428 | ||
3429 | btrfs_set_token_timespec_sec(&token, &item->mtime, | |
3430 | inode->i_mtime.tv_sec); | |
3431 | btrfs_set_token_timespec_nsec(&token, &item->mtime, | |
3432 | inode->i_mtime.tv_nsec); | |
3433 | ||
3434 | btrfs_set_token_timespec_sec(&token, &item->ctime, | |
3435 | inode->i_ctime.tv_sec); | |
3436 | btrfs_set_token_timespec_nsec(&token, &item->ctime, | |
3437 | inode->i_ctime.tv_nsec); | |
3438 | ||
3439 | btrfs_set_token_timespec_sec(&token, &item->otime, | |
3440 | BTRFS_I(inode)->i_otime.tv_sec); | |
3441 | btrfs_set_token_timespec_nsec(&token, &item->otime, | |
3442 | BTRFS_I(inode)->i_otime.tv_nsec); | |
3443 | ||
3444 | btrfs_set_token_inode_nbytes(&token, item, inode_get_bytes(inode)); | |
3445 | btrfs_set_token_inode_generation(&token, item, | |
3446 | BTRFS_I(inode)->generation); | |
3447 | btrfs_set_token_inode_sequence(&token, item, inode_peek_iversion(inode)); | |
3448 | btrfs_set_token_inode_transid(&token, item, trans->transid); | |
3449 | btrfs_set_token_inode_rdev(&token, item, inode->i_rdev); | |
3450 | btrfs_set_token_inode_flags(&token, item, BTRFS_I(inode)->flags); | |
3451 | btrfs_set_token_inode_block_group(&token, item, 0); | |
39279cc3 CM |
3452 | } |
3453 | ||
d352ac68 CM |
3454 | /* |
3455 | * copy everything in the in-memory inode into the btree. | |
3456 | */ | |
2115133f | 3457 | static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans, |
d397712b | 3458 | struct btrfs_root *root, struct inode *inode) |
39279cc3 CM |
3459 | { |
3460 | struct btrfs_inode_item *inode_item; | |
3461 | struct btrfs_path *path; | |
5f39d397 | 3462 | struct extent_buffer *leaf; |
39279cc3 CM |
3463 | int ret; |
3464 | ||
3465 | path = btrfs_alloc_path(); | |
16cdcec7 MX |
3466 | if (!path) |
3467 | return -ENOMEM; | |
3468 | ||
b9473439 | 3469 | path->leave_spinning = 1; |
16cdcec7 MX |
3470 | ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location, |
3471 | 1); | |
39279cc3 CM |
3472 | if (ret) { |
3473 | if (ret > 0) | |
3474 | ret = -ENOENT; | |
3475 | goto failed; | |
3476 | } | |
3477 | ||
5f39d397 CM |
3478 | leaf = path->nodes[0]; |
3479 | inode_item = btrfs_item_ptr(leaf, path->slots[0], | |
16cdcec7 | 3480 | struct btrfs_inode_item); |
39279cc3 | 3481 | |
e02119d5 | 3482 | fill_inode_item(trans, leaf, inode_item, inode); |
5f39d397 | 3483 | btrfs_mark_buffer_dirty(leaf); |
d9094414 | 3484 | btrfs_set_inode_last_trans(trans, BTRFS_I(inode)); |
39279cc3 CM |
3485 | ret = 0; |
3486 | failed: | |
39279cc3 CM |
3487 | btrfs_free_path(path); |
3488 | return ret; | |
3489 | } | |
3490 | ||
2115133f CM |
3491 | /* |
3492 | * copy everything in the in-memory inode into the btree. | |
3493 | */ | |
3494 | noinline int btrfs_update_inode(struct btrfs_trans_handle *trans, | |
3495 | struct btrfs_root *root, struct inode *inode) | |
3496 | { | |
0b246afa | 3497 | struct btrfs_fs_info *fs_info = root->fs_info; |
2115133f CM |
3498 | int ret; |
3499 | ||
3500 | /* | |
3501 | * If the inode is a free space inode, we can deadlock during commit | |
3502 | * if we put it into the delayed code. | |
3503 | * | |
3504 | * The data relocation inode should also be directly updated | |
3505 | * without delay | |
3506 | */ | |
70ddc553 | 3507 | if (!btrfs_is_free_space_inode(BTRFS_I(inode)) |
1d52c78a | 3508 | && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID |
0b246afa | 3509 | && !test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) { |
8ea05e3a AB |
3510 | btrfs_update_root_times(trans, root); |
3511 | ||
2115133f CM |
3512 | ret = btrfs_delayed_update_inode(trans, root, inode); |
3513 | if (!ret) | |
d9094414 | 3514 | btrfs_set_inode_last_trans(trans, BTRFS_I(inode)); |
2115133f CM |
3515 | return ret; |
3516 | } | |
3517 | ||
3518 | return btrfs_update_inode_item(trans, root, inode); | |
3519 | } | |
3520 | ||
be6aef60 JB |
3521 | noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans, |
3522 | struct btrfs_root *root, | |
3523 | struct inode *inode) | |
2115133f CM |
3524 | { |
3525 | int ret; | |
3526 | ||
3527 | ret = btrfs_update_inode(trans, root, inode); | |
3528 | if (ret == -ENOSPC) | |
3529 | return btrfs_update_inode_item(trans, root, inode); | |
3530 | return ret; | |
3531 | } | |
3532 | ||
d352ac68 CM |
3533 | /* |
3534 | * unlink helper that gets used here in inode.c and in the tree logging | |
3535 | * recovery code. It remove a link in a directory with a given name, and | |
3536 | * also drops the back refs in the inode to the directory | |
3537 | */ | |
92986796 AV |
3538 | static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans, |
3539 | struct btrfs_root *root, | |
4ec5934e NB |
3540 | struct btrfs_inode *dir, |
3541 | struct btrfs_inode *inode, | |
92986796 | 3542 | const char *name, int name_len) |
39279cc3 | 3543 | { |
0b246afa | 3544 | struct btrfs_fs_info *fs_info = root->fs_info; |
39279cc3 | 3545 | struct btrfs_path *path; |
39279cc3 | 3546 | int ret = 0; |
39279cc3 | 3547 | struct btrfs_dir_item *di; |
aec7477b | 3548 | u64 index; |
33345d01 LZ |
3549 | u64 ino = btrfs_ino(inode); |
3550 | u64 dir_ino = btrfs_ino(dir); | |
39279cc3 CM |
3551 | |
3552 | path = btrfs_alloc_path(); | |
54aa1f4d CM |
3553 | if (!path) { |
3554 | ret = -ENOMEM; | |
554233a6 | 3555 | goto out; |
54aa1f4d CM |
3556 | } |
3557 | ||
b9473439 | 3558 | path->leave_spinning = 1; |
33345d01 | 3559 | di = btrfs_lookup_dir_item(trans, root, path, dir_ino, |
39279cc3 | 3560 | name, name_len, -1); |
3cf5068f LB |
3561 | if (IS_ERR_OR_NULL(di)) { |
3562 | ret = di ? PTR_ERR(di) : -ENOENT; | |
39279cc3 CM |
3563 | goto err; |
3564 | } | |
39279cc3 | 3565 | ret = btrfs_delete_one_dir_name(trans, root, path, di); |
54aa1f4d CM |
3566 | if (ret) |
3567 | goto err; | |
b3b4aa74 | 3568 | btrfs_release_path(path); |
39279cc3 | 3569 | |
67de1176 MX |
3570 | /* |
3571 | * If we don't have dir index, we have to get it by looking up | |
3572 | * the inode ref, since we get the inode ref, remove it directly, | |
3573 | * it is unnecessary to do delayed deletion. | |
3574 | * | |
3575 | * But if we have dir index, needn't search inode ref to get it. | |
3576 | * Since the inode ref is close to the inode item, it is better | |
3577 | * that we delay to delete it, and just do this deletion when | |
3578 | * we update the inode item. | |
3579 | */ | |
4ec5934e | 3580 | if (inode->dir_index) { |
67de1176 MX |
3581 | ret = btrfs_delayed_delete_inode_ref(inode); |
3582 | if (!ret) { | |
4ec5934e | 3583 | index = inode->dir_index; |
67de1176 MX |
3584 | goto skip_backref; |
3585 | } | |
3586 | } | |
3587 | ||
33345d01 LZ |
3588 | ret = btrfs_del_inode_ref(trans, root, name, name_len, ino, |
3589 | dir_ino, &index); | |
aec7477b | 3590 | if (ret) { |
0b246afa | 3591 | btrfs_info(fs_info, |
c2cf52eb | 3592 | "failed to delete reference to %.*s, inode %llu parent %llu", |
c1c9ff7c | 3593 | name_len, name, ino, dir_ino); |
66642832 | 3594 | btrfs_abort_transaction(trans, ret); |
aec7477b JB |
3595 | goto err; |
3596 | } | |
67de1176 | 3597 | skip_backref: |
9add2945 | 3598 | ret = btrfs_delete_delayed_dir_index(trans, dir, index); |
79787eaa | 3599 | if (ret) { |
66642832 | 3600 | btrfs_abort_transaction(trans, ret); |
39279cc3 | 3601 | goto err; |
79787eaa | 3602 | } |
39279cc3 | 3603 | |
4ec5934e NB |
3604 | ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len, inode, |
3605 | dir_ino); | |
79787eaa | 3606 | if (ret != 0 && ret != -ENOENT) { |
66642832 | 3607 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
3608 | goto err; |
3609 | } | |
e02119d5 | 3610 | |
4ec5934e NB |
3611 | ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len, dir, |
3612 | index); | |
6418c961 CM |
3613 | if (ret == -ENOENT) |
3614 | ret = 0; | |
d4e3991b | 3615 | else if (ret) |
66642832 | 3616 | btrfs_abort_transaction(trans, ret); |
63611e73 JB |
3617 | |
3618 | /* | |
3619 | * If we have a pending delayed iput we could end up with the final iput | |
3620 | * being run in btrfs-cleaner context. If we have enough of these built | |
3621 | * up we can end up burning a lot of time in btrfs-cleaner without any | |
3622 | * way to throttle the unlinks. Since we're currently holding a ref on | |
3623 | * the inode we can run the delayed iput here without any issues as the | |
3624 | * final iput won't be done until after we drop the ref we're currently | |
3625 | * holding. | |
3626 | */ | |
3627 | btrfs_run_delayed_iput(fs_info, inode); | |
39279cc3 CM |
3628 | err: |
3629 | btrfs_free_path(path); | |
e02119d5 CM |
3630 | if (ret) |
3631 | goto out; | |
3632 | ||
6ef06d27 | 3633 | btrfs_i_size_write(dir, dir->vfs_inode.i_size - name_len * 2); |
4ec5934e NB |
3634 | inode_inc_iversion(&inode->vfs_inode); |
3635 | inode_inc_iversion(&dir->vfs_inode); | |
3636 | inode->vfs_inode.i_ctime = dir->vfs_inode.i_mtime = | |
3637 | dir->vfs_inode.i_ctime = current_time(&inode->vfs_inode); | |
3638 | ret = btrfs_update_inode(trans, root, &dir->vfs_inode); | |
e02119d5 | 3639 | out: |
39279cc3 CM |
3640 | return ret; |
3641 | } | |
3642 | ||
92986796 AV |
3643 | int btrfs_unlink_inode(struct btrfs_trans_handle *trans, |
3644 | struct btrfs_root *root, | |
4ec5934e | 3645 | struct btrfs_inode *dir, struct btrfs_inode *inode, |
92986796 AV |
3646 | const char *name, int name_len) |
3647 | { | |
3648 | int ret; | |
3649 | ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len); | |
3650 | if (!ret) { | |
4ec5934e NB |
3651 | drop_nlink(&inode->vfs_inode); |
3652 | ret = btrfs_update_inode(trans, root, &inode->vfs_inode); | |
92986796 AV |
3653 | } |
3654 | return ret; | |
3655 | } | |
39279cc3 | 3656 | |
a22285a6 YZ |
3657 | /* |
3658 | * helper to start transaction for unlink and rmdir. | |
3659 | * | |
d52be818 JB |
3660 | * unlink and rmdir are special in btrfs, they do not always free space, so |
3661 | * if we cannot make our reservations the normal way try and see if there is | |
3662 | * plenty of slack room in the global reserve to migrate, otherwise we cannot | |
3663 | * allow the unlink to occur. | |
a22285a6 | 3664 | */ |
d52be818 | 3665 | static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir) |
4df27c4d | 3666 | { |
a22285a6 | 3667 | struct btrfs_root *root = BTRFS_I(dir)->root; |
4df27c4d | 3668 | |
e70bea5f JB |
3669 | /* |
3670 | * 1 for the possible orphan item | |
3671 | * 1 for the dir item | |
3672 | * 1 for the dir index | |
3673 | * 1 for the inode ref | |
e70bea5f JB |
3674 | * 1 for the inode |
3675 | */ | |
7f9fe614 | 3676 | return btrfs_start_transaction_fallback_global_rsv(root, 5); |
a22285a6 YZ |
3677 | } |
3678 | ||
3679 | static int btrfs_unlink(struct inode *dir, struct dentry *dentry) | |
3680 | { | |
3681 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
3682 | struct btrfs_trans_handle *trans; | |
2b0143b5 | 3683 | struct inode *inode = d_inode(dentry); |
a22285a6 | 3684 | int ret; |
a22285a6 | 3685 | |
d52be818 | 3686 | trans = __unlink_start_trans(dir); |
a22285a6 YZ |
3687 | if (IS_ERR(trans)) |
3688 | return PTR_ERR(trans); | |
5f39d397 | 3689 | |
4ec5934e NB |
3690 | btrfs_record_unlink_dir(trans, BTRFS_I(dir), BTRFS_I(d_inode(dentry)), |
3691 | 0); | |
12fcfd22 | 3692 | |
4ec5934e NB |
3693 | ret = btrfs_unlink_inode(trans, root, BTRFS_I(dir), |
3694 | BTRFS_I(d_inode(dentry)), dentry->d_name.name, | |
3695 | dentry->d_name.len); | |
b532402e TI |
3696 | if (ret) |
3697 | goto out; | |
7b128766 | 3698 | |
a22285a6 | 3699 | if (inode->i_nlink == 0) { |
73f2e545 | 3700 | ret = btrfs_orphan_add(trans, BTRFS_I(inode)); |
b532402e TI |
3701 | if (ret) |
3702 | goto out; | |
a22285a6 | 3703 | } |
7b128766 | 3704 | |
b532402e | 3705 | out: |
3a45bb20 | 3706 | btrfs_end_transaction(trans); |
2ff7e61e | 3707 | btrfs_btree_balance_dirty(root->fs_info); |
39279cc3 CM |
3708 | return ret; |
3709 | } | |
3710 | ||
f60a2364 | 3711 | static int btrfs_unlink_subvol(struct btrfs_trans_handle *trans, |
045d3967 | 3712 | struct inode *dir, struct dentry *dentry) |
4df27c4d | 3713 | { |
401b3b19 | 3714 | struct btrfs_root *root = BTRFS_I(dir)->root; |
045d3967 | 3715 | struct btrfs_inode *inode = BTRFS_I(d_inode(dentry)); |
4df27c4d YZ |
3716 | struct btrfs_path *path; |
3717 | struct extent_buffer *leaf; | |
3718 | struct btrfs_dir_item *di; | |
3719 | struct btrfs_key key; | |
045d3967 JB |
3720 | const char *name = dentry->d_name.name; |
3721 | int name_len = dentry->d_name.len; | |
4df27c4d YZ |
3722 | u64 index; |
3723 | int ret; | |
045d3967 | 3724 | u64 objectid; |
4a0cc7ca | 3725 | u64 dir_ino = btrfs_ino(BTRFS_I(dir)); |
4df27c4d | 3726 | |
045d3967 JB |
3727 | if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID) { |
3728 | objectid = inode->root->root_key.objectid; | |
3729 | } else if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) { | |
3730 | objectid = inode->location.objectid; | |
3731 | } else { | |
3732 | WARN_ON(1); | |
3733 | return -EINVAL; | |
3734 | } | |
3735 | ||
4df27c4d YZ |
3736 | path = btrfs_alloc_path(); |
3737 | if (!path) | |
3738 | return -ENOMEM; | |
3739 | ||
33345d01 | 3740 | di = btrfs_lookup_dir_item(trans, root, path, dir_ino, |
4df27c4d | 3741 | name, name_len, -1); |
79787eaa | 3742 | if (IS_ERR_OR_NULL(di)) { |
3cf5068f | 3743 | ret = di ? PTR_ERR(di) : -ENOENT; |
79787eaa JM |
3744 | goto out; |
3745 | } | |
4df27c4d YZ |
3746 | |
3747 | leaf = path->nodes[0]; | |
3748 | btrfs_dir_item_key_to_cpu(leaf, di, &key); | |
3749 | WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid); | |
3750 | ret = btrfs_delete_one_dir_name(trans, root, path, di); | |
79787eaa | 3751 | if (ret) { |
66642832 | 3752 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
3753 | goto out; |
3754 | } | |
b3b4aa74 | 3755 | btrfs_release_path(path); |
4df27c4d | 3756 | |
d49d3287 JB |
3757 | /* |
3758 | * This is a placeholder inode for a subvolume we didn't have a | |
3759 | * reference to at the time of the snapshot creation. In the meantime | |
3760 | * we could have renamed the real subvol link into our snapshot, so | |
3761 | * depending on btrfs_del_root_ref to return -ENOENT here is incorret. | |
3762 | * Instead simply lookup the dir_index_item for this entry so we can | |
3763 | * remove it. Otherwise we know we have a ref to the root and we can | |
3764 | * call btrfs_del_root_ref, and it _shouldn't_ fail. | |
3765 | */ | |
3766 | if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) { | |
33345d01 | 3767 | di = btrfs_search_dir_index_item(root, path, dir_ino, |
4df27c4d | 3768 | name, name_len); |
79787eaa JM |
3769 | if (IS_ERR_OR_NULL(di)) { |
3770 | if (!di) | |
3771 | ret = -ENOENT; | |
3772 | else | |
3773 | ret = PTR_ERR(di); | |
66642832 | 3774 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
3775 | goto out; |
3776 | } | |
4df27c4d YZ |
3777 | |
3778 | leaf = path->nodes[0]; | |
3779 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
4df27c4d | 3780 | index = key.offset; |
d49d3287 JB |
3781 | btrfs_release_path(path); |
3782 | } else { | |
3783 | ret = btrfs_del_root_ref(trans, objectid, | |
3784 | root->root_key.objectid, dir_ino, | |
3785 | &index, name, name_len); | |
3786 | if (ret) { | |
3787 | btrfs_abort_transaction(trans, ret); | |
3788 | goto out; | |
3789 | } | |
4df27c4d YZ |
3790 | } |
3791 | ||
9add2945 | 3792 | ret = btrfs_delete_delayed_dir_index(trans, BTRFS_I(dir), index); |
79787eaa | 3793 | if (ret) { |
66642832 | 3794 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
3795 | goto out; |
3796 | } | |
4df27c4d | 3797 | |
6ef06d27 | 3798 | btrfs_i_size_write(BTRFS_I(dir), dir->i_size - name_len * 2); |
0c4d2d95 | 3799 | inode_inc_iversion(dir); |
c2050a45 | 3800 | dir->i_mtime = dir->i_ctime = current_time(dir); |
5a24e84c | 3801 | ret = btrfs_update_inode_fallback(trans, root, dir); |
79787eaa | 3802 | if (ret) |
66642832 | 3803 | btrfs_abort_transaction(trans, ret); |
79787eaa | 3804 | out: |
71d7aed0 | 3805 | btrfs_free_path(path); |
79787eaa | 3806 | return ret; |
4df27c4d YZ |
3807 | } |
3808 | ||
ec42f167 MT |
3809 | /* |
3810 | * Helper to check if the subvolume references other subvolumes or if it's | |
3811 | * default. | |
3812 | */ | |
f60a2364 | 3813 | static noinline int may_destroy_subvol(struct btrfs_root *root) |
ec42f167 MT |
3814 | { |
3815 | struct btrfs_fs_info *fs_info = root->fs_info; | |
3816 | struct btrfs_path *path; | |
3817 | struct btrfs_dir_item *di; | |
3818 | struct btrfs_key key; | |
3819 | u64 dir_id; | |
3820 | int ret; | |
3821 | ||
3822 | path = btrfs_alloc_path(); | |
3823 | if (!path) | |
3824 | return -ENOMEM; | |
3825 | ||
3826 | /* Make sure this root isn't set as the default subvol */ | |
3827 | dir_id = btrfs_super_root_dir(fs_info->super_copy); | |
3828 | di = btrfs_lookup_dir_item(NULL, fs_info->tree_root, path, | |
3829 | dir_id, "default", 7, 0); | |
3830 | if (di && !IS_ERR(di)) { | |
3831 | btrfs_dir_item_key_to_cpu(path->nodes[0], di, &key); | |
3832 | if (key.objectid == root->root_key.objectid) { | |
3833 | ret = -EPERM; | |
3834 | btrfs_err(fs_info, | |
3835 | "deleting default subvolume %llu is not allowed", | |
3836 | key.objectid); | |
3837 | goto out; | |
3838 | } | |
3839 | btrfs_release_path(path); | |
3840 | } | |
3841 | ||
3842 | key.objectid = root->root_key.objectid; | |
3843 | key.type = BTRFS_ROOT_REF_KEY; | |
3844 | key.offset = (u64)-1; | |
3845 | ||
3846 | ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0); | |
3847 | if (ret < 0) | |
3848 | goto out; | |
3849 | BUG_ON(ret == 0); | |
3850 | ||
3851 | ret = 0; | |
3852 | if (path->slots[0] > 0) { | |
3853 | path->slots[0]--; | |
3854 | btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]); | |
3855 | if (key.objectid == root->root_key.objectid && | |
3856 | key.type == BTRFS_ROOT_REF_KEY) | |
3857 | ret = -ENOTEMPTY; | |
3858 | } | |
3859 | out: | |
3860 | btrfs_free_path(path); | |
3861 | return ret; | |
3862 | } | |
3863 | ||
20a68004 NB |
3864 | /* Delete all dentries for inodes belonging to the root */ |
3865 | static void btrfs_prune_dentries(struct btrfs_root *root) | |
3866 | { | |
3867 | struct btrfs_fs_info *fs_info = root->fs_info; | |
3868 | struct rb_node *node; | |
3869 | struct rb_node *prev; | |
3870 | struct btrfs_inode *entry; | |
3871 | struct inode *inode; | |
3872 | u64 objectid = 0; | |
3873 | ||
3874 | if (!test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) | |
3875 | WARN_ON(btrfs_root_refs(&root->root_item) != 0); | |
3876 | ||
3877 | spin_lock(&root->inode_lock); | |
3878 | again: | |
3879 | node = root->inode_tree.rb_node; | |
3880 | prev = NULL; | |
3881 | while (node) { | |
3882 | prev = node; | |
3883 | entry = rb_entry(node, struct btrfs_inode, rb_node); | |
3884 | ||
37508515 | 3885 | if (objectid < btrfs_ino(entry)) |
20a68004 | 3886 | node = node->rb_left; |
37508515 | 3887 | else if (objectid > btrfs_ino(entry)) |
20a68004 NB |
3888 | node = node->rb_right; |
3889 | else | |
3890 | break; | |
3891 | } | |
3892 | if (!node) { | |
3893 | while (prev) { | |
3894 | entry = rb_entry(prev, struct btrfs_inode, rb_node); | |
37508515 | 3895 | if (objectid <= btrfs_ino(entry)) { |
20a68004 NB |
3896 | node = prev; |
3897 | break; | |
3898 | } | |
3899 | prev = rb_next(prev); | |
3900 | } | |
3901 | } | |
3902 | while (node) { | |
3903 | entry = rb_entry(node, struct btrfs_inode, rb_node); | |
37508515 | 3904 | objectid = btrfs_ino(entry) + 1; |
20a68004 NB |
3905 | inode = igrab(&entry->vfs_inode); |
3906 | if (inode) { | |
3907 | spin_unlock(&root->inode_lock); | |
3908 | if (atomic_read(&inode->i_count) > 1) | |
3909 | d_prune_aliases(inode); | |
3910 | /* | |
3911 | * btrfs_drop_inode will have it removed from the inode | |
3912 | * cache when its usage count hits zero. | |
3913 | */ | |
3914 | iput(inode); | |
3915 | cond_resched(); | |
3916 | spin_lock(&root->inode_lock); | |
3917 | goto again; | |
3918 | } | |
3919 | ||
3920 | if (cond_resched_lock(&root->inode_lock)) | |
3921 | goto again; | |
3922 | ||
3923 | node = rb_next(node); | |
3924 | } | |
3925 | spin_unlock(&root->inode_lock); | |
3926 | } | |
3927 | ||
f60a2364 MT |
3928 | int btrfs_delete_subvolume(struct inode *dir, struct dentry *dentry) |
3929 | { | |
3930 | struct btrfs_fs_info *fs_info = btrfs_sb(dentry->d_sb); | |
3931 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
3932 | struct inode *inode = d_inode(dentry); | |
3933 | struct btrfs_root *dest = BTRFS_I(inode)->root; | |
3934 | struct btrfs_trans_handle *trans; | |
3935 | struct btrfs_block_rsv block_rsv; | |
3936 | u64 root_flags; | |
f60a2364 MT |
3937 | int ret; |
3938 | int err; | |
3939 | ||
3940 | /* | |
3941 | * Don't allow to delete a subvolume with send in progress. This is | |
3942 | * inside the inode lock so the error handling that has to drop the bit | |
3943 | * again is not run concurrently. | |
3944 | */ | |
3945 | spin_lock(&dest->root_item_lock); | |
a7176f74 | 3946 | if (dest->send_in_progress) { |
f60a2364 MT |
3947 | spin_unlock(&dest->root_item_lock); |
3948 | btrfs_warn(fs_info, | |
3949 | "attempt to delete subvolume %llu during send", | |
3950 | dest->root_key.objectid); | |
3951 | return -EPERM; | |
3952 | } | |
a7176f74 LF |
3953 | root_flags = btrfs_root_flags(&dest->root_item); |
3954 | btrfs_set_root_flags(&dest->root_item, | |
3955 | root_flags | BTRFS_ROOT_SUBVOL_DEAD); | |
3956 | spin_unlock(&dest->root_item_lock); | |
f60a2364 MT |
3957 | |
3958 | down_write(&fs_info->subvol_sem); | |
3959 | ||
3960 | err = may_destroy_subvol(dest); | |
3961 | if (err) | |
3962 | goto out_up_write; | |
3963 | ||
3964 | btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP); | |
3965 | /* | |
3966 | * One for dir inode, | |
3967 | * two for dir entries, | |
3968 | * two for root ref/backref. | |
3969 | */ | |
c4c129db | 3970 | err = btrfs_subvolume_reserve_metadata(root, &block_rsv, 5, true); |
f60a2364 MT |
3971 | if (err) |
3972 | goto out_up_write; | |
3973 | ||
3974 | trans = btrfs_start_transaction(root, 0); | |
3975 | if (IS_ERR(trans)) { | |
3976 | err = PTR_ERR(trans); | |
3977 | goto out_release; | |
3978 | } | |
3979 | trans->block_rsv = &block_rsv; | |
3980 | trans->bytes_reserved = block_rsv.size; | |
3981 | ||
3982 | btrfs_record_snapshot_destroy(trans, BTRFS_I(dir)); | |
3983 | ||
045d3967 | 3984 | ret = btrfs_unlink_subvol(trans, dir, dentry); |
f60a2364 MT |
3985 | if (ret) { |
3986 | err = ret; | |
3987 | btrfs_abort_transaction(trans, ret); | |
3988 | goto out_end_trans; | |
3989 | } | |
3990 | ||
3991 | btrfs_record_root_in_trans(trans, dest); | |
3992 | ||
3993 | memset(&dest->root_item.drop_progress, 0, | |
3994 | sizeof(dest->root_item.drop_progress)); | |
3995 | dest->root_item.drop_level = 0; | |
3996 | btrfs_set_root_refs(&dest->root_item, 0); | |
3997 | ||
3998 | if (!test_and_set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &dest->state)) { | |
3999 | ret = btrfs_insert_orphan_item(trans, | |
4000 | fs_info->tree_root, | |
4001 | dest->root_key.objectid); | |
4002 | if (ret) { | |
4003 | btrfs_abort_transaction(trans, ret); | |
4004 | err = ret; | |
4005 | goto out_end_trans; | |
4006 | } | |
4007 | } | |
4008 | ||
d1957791 | 4009 | ret = btrfs_uuid_tree_remove(trans, dest->root_item.uuid, |
f60a2364 MT |
4010 | BTRFS_UUID_KEY_SUBVOL, |
4011 | dest->root_key.objectid); | |
4012 | if (ret && ret != -ENOENT) { | |
4013 | btrfs_abort_transaction(trans, ret); | |
4014 | err = ret; | |
4015 | goto out_end_trans; | |
4016 | } | |
4017 | if (!btrfs_is_empty_uuid(dest->root_item.received_uuid)) { | |
d1957791 | 4018 | ret = btrfs_uuid_tree_remove(trans, |
f60a2364 MT |
4019 | dest->root_item.received_uuid, |
4020 | BTRFS_UUID_KEY_RECEIVED_SUBVOL, | |
4021 | dest->root_key.objectid); | |
4022 | if (ret && ret != -ENOENT) { | |
4023 | btrfs_abort_transaction(trans, ret); | |
4024 | err = ret; | |
4025 | goto out_end_trans; | |
4026 | } | |
4027 | } | |
4028 | ||
4029 | out_end_trans: | |
4030 | trans->block_rsv = NULL; | |
4031 | trans->bytes_reserved = 0; | |
4032 | ret = btrfs_end_transaction(trans); | |
4033 | if (ret && !err) | |
4034 | err = ret; | |
4035 | inode->i_flags |= S_DEAD; | |
4036 | out_release: | |
4037 | btrfs_subvolume_release_metadata(fs_info, &block_rsv); | |
4038 | out_up_write: | |
4039 | up_write(&fs_info->subvol_sem); | |
4040 | if (err) { | |
4041 | spin_lock(&dest->root_item_lock); | |
4042 | root_flags = btrfs_root_flags(&dest->root_item); | |
4043 | btrfs_set_root_flags(&dest->root_item, | |
4044 | root_flags & ~BTRFS_ROOT_SUBVOL_DEAD); | |
4045 | spin_unlock(&dest->root_item_lock); | |
4046 | } else { | |
4047 | d_invalidate(dentry); | |
20a68004 | 4048 | btrfs_prune_dentries(dest); |
f60a2364 MT |
4049 | ASSERT(dest->send_in_progress == 0); |
4050 | ||
4051 | /* the last ref */ | |
4052 | if (dest->ino_cache_inode) { | |
4053 | iput(dest->ino_cache_inode); | |
4054 | dest->ino_cache_inode = NULL; | |
4055 | } | |
4056 | } | |
4057 | ||
4058 | return err; | |
4059 | } | |
4060 | ||
39279cc3 CM |
4061 | static int btrfs_rmdir(struct inode *dir, struct dentry *dentry) |
4062 | { | |
2b0143b5 | 4063 | struct inode *inode = d_inode(dentry); |
1832a6d5 | 4064 | int err = 0; |
39279cc3 | 4065 | struct btrfs_root *root = BTRFS_I(dir)->root; |
39279cc3 | 4066 | struct btrfs_trans_handle *trans; |
44f714da | 4067 | u64 last_unlink_trans; |
39279cc3 | 4068 | |
b3ae244e | 4069 | if (inode->i_size > BTRFS_EMPTY_DIR_SIZE) |
134d4512 | 4070 | return -ENOTEMPTY; |
4a0cc7ca | 4071 | if (btrfs_ino(BTRFS_I(inode)) == BTRFS_FIRST_FREE_OBJECTID) |
a79a464d | 4072 | return btrfs_delete_subvolume(dir, dentry); |
134d4512 | 4073 | |
d52be818 | 4074 | trans = __unlink_start_trans(dir); |
a22285a6 | 4075 | if (IS_ERR(trans)) |
5df6a9f6 | 4076 | return PTR_ERR(trans); |
5df6a9f6 | 4077 | |
4a0cc7ca | 4078 | if (unlikely(btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) { |
045d3967 | 4079 | err = btrfs_unlink_subvol(trans, dir, dentry); |
4df27c4d YZ |
4080 | goto out; |
4081 | } | |
4082 | ||
73f2e545 | 4083 | err = btrfs_orphan_add(trans, BTRFS_I(inode)); |
7b128766 | 4084 | if (err) |
4df27c4d | 4085 | goto out; |
7b128766 | 4086 | |
44f714da FM |
4087 | last_unlink_trans = BTRFS_I(inode)->last_unlink_trans; |
4088 | ||
39279cc3 | 4089 | /* now the directory is empty */ |
4ec5934e NB |
4090 | err = btrfs_unlink_inode(trans, root, BTRFS_I(dir), |
4091 | BTRFS_I(d_inode(dentry)), dentry->d_name.name, | |
4092 | dentry->d_name.len); | |
44f714da | 4093 | if (!err) { |
6ef06d27 | 4094 | btrfs_i_size_write(BTRFS_I(inode), 0); |
44f714da FM |
4095 | /* |
4096 | * Propagate the last_unlink_trans value of the deleted dir to | |
4097 | * its parent directory. This is to prevent an unrecoverable | |
4098 | * log tree in the case we do something like this: | |
4099 | * 1) create dir foo | |
4100 | * 2) create snapshot under dir foo | |
4101 | * 3) delete the snapshot | |
4102 | * 4) rmdir foo | |
4103 | * 5) mkdir foo | |
4104 | * 6) fsync foo or some file inside foo | |
4105 | */ | |
4106 | if (last_unlink_trans >= trans->transid) | |
4107 | BTRFS_I(dir)->last_unlink_trans = last_unlink_trans; | |
4108 | } | |
4df27c4d | 4109 | out: |
3a45bb20 | 4110 | btrfs_end_transaction(trans); |
2ff7e61e | 4111 | btrfs_btree_balance_dirty(root->fs_info); |
3954401f | 4112 | |
39279cc3 CM |
4113 | return err; |
4114 | } | |
4115 | ||
ddfae63c JB |
4116 | /* |
4117 | * Return this if we need to call truncate_block for the last bit of the | |
4118 | * truncate. | |
4119 | */ | |
4120 | #define NEED_TRUNCATE_BLOCK 1 | |
0305cd5f | 4121 | |
39279cc3 CM |
4122 | /* |
4123 | * this can truncate away extent items, csum items and directory items. | |
4124 | * It starts at a high offset and removes keys until it can't find | |
d352ac68 | 4125 | * any higher than new_size |
39279cc3 CM |
4126 | * |
4127 | * csum items that cross the new i_size are truncated to the new size | |
4128 | * as well. | |
7b128766 JB |
4129 | * |
4130 | * min_type is the minimum key type to truncate down to. If set to 0, this | |
4131 | * will kill all the items on this inode, including the INODE_ITEM_KEY. | |
39279cc3 | 4132 | */ |
8082510e YZ |
4133 | int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans, |
4134 | struct btrfs_root *root, | |
4135 | struct inode *inode, | |
4136 | u64 new_size, u32 min_type) | |
39279cc3 | 4137 | { |
0b246afa | 4138 | struct btrfs_fs_info *fs_info = root->fs_info; |
39279cc3 | 4139 | struct btrfs_path *path; |
5f39d397 | 4140 | struct extent_buffer *leaf; |
39279cc3 | 4141 | struct btrfs_file_extent_item *fi; |
8082510e YZ |
4142 | struct btrfs_key key; |
4143 | struct btrfs_key found_key; | |
39279cc3 | 4144 | u64 extent_start = 0; |
db94535d | 4145 | u64 extent_num_bytes = 0; |
5d4f98a2 | 4146 | u64 extent_offset = 0; |
39279cc3 | 4147 | u64 item_end = 0; |
c1aa4575 | 4148 | u64 last_size = new_size; |
8082510e | 4149 | u32 found_type = (u8)-1; |
39279cc3 CM |
4150 | int found_extent; |
4151 | int del_item; | |
85e21bac CM |
4152 | int pending_del_nr = 0; |
4153 | int pending_del_slot = 0; | |
179e29e4 | 4154 | int extent_type = -1; |
8082510e | 4155 | int ret; |
4a0cc7ca | 4156 | u64 ino = btrfs_ino(BTRFS_I(inode)); |
28ed1345 | 4157 | u64 bytes_deleted = 0; |
897ca819 TM |
4158 | bool be_nice = false; |
4159 | bool should_throttle = false; | |
28553fa9 FM |
4160 | const u64 lock_start = ALIGN_DOWN(new_size, fs_info->sectorsize); |
4161 | struct extent_state *cached_state = NULL; | |
8082510e YZ |
4162 | |
4163 | BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY); | |
39279cc3 | 4164 | |
28ed1345 | 4165 | /* |
92a7cc42 QW |
4166 | * For non-free space inodes and non-shareable roots, we want to back |
4167 | * off from time to time. This means all inodes in subvolume roots, | |
4168 | * reloc roots, and data reloc roots. | |
28ed1345 | 4169 | */ |
70ddc553 | 4170 | if (!btrfs_is_free_space_inode(BTRFS_I(inode)) && |
92a7cc42 | 4171 | test_bit(BTRFS_ROOT_SHAREABLE, &root->state)) |
897ca819 | 4172 | be_nice = true; |
28ed1345 | 4173 | |
0eb0e19c MF |
4174 | path = btrfs_alloc_path(); |
4175 | if (!path) | |
4176 | return -ENOMEM; | |
e4058b54 | 4177 | path->reada = READA_BACK; |
0eb0e19c | 4178 | |
82028e0a | 4179 | if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) { |
a5ae50de FM |
4180 | lock_extent_bits(&BTRFS_I(inode)->io_tree, lock_start, (u64)-1, |
4181 | &cached_state); | |
28553fa9 | 4182 | |
82028e0a QW |
4183 | /* |
4184 | * We want to drop from the next block forward in case this | |
4185 | * new size is not block aligned since we will be keeping the | |
4186 | * last block of the extent just the way it is. | |
4187 | */ | |
dcdbc059 | 4188 | btrfs_drop_extent_cache(BTRFS_I(inode), ALIGN(new_size, |
0b246afa | 4189 | fs_info->sectorsize), |
da17066c | 4190 | (u64)-1, 0); |
82028e0a | 4191 | } |
8082510e | 4192 | |
16cdcec7 MX |
4193 | /* |
4194 | * This function is also used to drop the items in the log tree before | |
4195 | * we relog the inode, so if root != BTRFS_I(inode)->root, it means | |
52042d8e | 4196 | * it is used to drop the logged items. So we shouldn't kill the delayed |
16cdcec7 MX |
4197 | * items. |
4198 | */ | |
4199 | if (min_type == 0 && root == BTRFS_I(inode)->root) | |
4ccb5c72 | 4200 | btrfs_kill_delayed_inode_items(BTRFS_I(inode)); |
16cdcec7 | 4201 | |
33345d01 | 4202 | key.objectid = ino; |
39279cc3 | 4203 | key.offset = (u64)-1; |
5f39d397 CM |
4204 | key.type = (u8)-1; |
4205 | ||
85e21bac | 4206 | search_again: |
28ed1345 CM |
4207 | /* |
4208 | * with a 16K leaf size and 128MB extents, you can actually queue | |
4209 | * up a huge file in a single leaf. Most of the time that | |
4210 | * bytes_deleted is > 0, it will be huge by the time we get here | |
4211 | */ | |
fd86a3a3 OS |
4212 | if (be_nice && bytes_deleted > SZ_32M && |
4213 | btrfs_should_end_transaction(trans)) { | |
4214 | ret = -EAGAIN; | |
4215 | goto out; | |
28ed1345 CM |
4216 | } |
4217 | ||
85e21bac | 4218 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
fd86a3a3 | 4219 | if (ret < 0) |
8082510e | 4220 | goto out; |
d397712b | 4221 | |
85e21bac | 4222 | if (ret > 0) { |
fd86a3a3 | 4223 | ret = 0; |
e02119d5 CM |
4224 | /* there are no items in the tree for us to truncate, we're |
4225 | * done | |
4226 | */ | |
8082510e YZ |
4227 | if (path->slots[0] == 0) |
4228 | goto out; | |
85e21bac CM |
4229 | path->slots[0]--; |
4230 | } | |
4231 | ||
d397712b | 4232 | while (1) { |
9ddc959e JB |
4233 | u64 clear_start = 0, clear_len = 0; |
4234 | ||
39279cc3 | 4235 | fi = NULL; |
5f39d397 CM |
4236 | leaf = path->nodes[0]; |
4237 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
962a298f | 4238 | found_type = found_key.type; |
39279cc3 | 4239 | |
33345d01 | 4240 | if (found_key.objectid != ino) |
39279cc3 | 4241 | break; |
5f39d397 | 4242 | |
85e21bac | 4243 | if (found_type < min_type) |
39279cc3 CM |
4244 | break; |
4245 | ||
5f39d397 | 4246 | item_end = found_key.offset; |
39279cc3 | 4247 | if (found_type == BTRFS_EXTENT_DATA_KEY) { |
5f39d397 | 4248 | fi = btrfs_item_ptr(leaf, path->slots[0], |
39279cc3 | 4249 | struct btrfs_file_extent_item); |
179e29e4 CM |
4250 | extent_type = btrfs_file_extent_type(leaf, fi); |
4251 | if (extent_type != BTRFS_FILE_EXTENT_INLINE) { | |
5f39d397 | 4252 | item_end += |
db94535d | 4253 | btrfs_file_extent_num_bytes(leaf, fi); |
09ed2f16 LB |
4254 | |
4255 | trace_btrfs_truncate_show_fi_regular( | |
4256 | BTRFS_I(inode), leaf, fi, | |
4257 | found_key.offset); | |
179e29e4 | 4258 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
e41ca589 QW |
4259 | item_end += btrfs_file_extent_ram_bytes(leaf, |
4260 | fi); | |
09ed2f16 LB |
4261 | |
4262 | trace_btrfs_truncate_show_fi_inline( | |
4263 | BTRFS_I(inode), leaf, fi, path->slots[0], | |
4264 | found_key.offset); | |
39279cc3 | 4265 | } |
008630c1 | 4266 | item_end--; |
39279cc3 | 4267 | } |
8082510e YZ |
4268 | if (found_type > min_type) { |
4269 | del_item = 1; | |
4270 | } else { | |
76b42abb | 4271 | if (item_end < new_size) |
b888db2b | 4272 | break; |
8082510e YZ |
4273 | if (found_key.offset >= new_size) |
4274 | del_item = 1; | |
4275 | else | |
4276 | del_item = 0; | |
39279cc3 | 4277 | } |
39279cc3 | 4278 | found_extent = 0; |
39279cc3 | 4279 | /* FIXME, shrink the extent if the ref count is only 1 */ |
179e29e4 CM |
4280 | if (found_type != BTRFS_EXTENT_DATA_KEY) |
4281 | goto delete; | |
4282 | ||
4283 | if (extent_type != BTRFS_FILE_EXTENT_INLINE) { | |
39279cc3 | 4284 | u64 num_dec; |
9ddc959e JB |
4285 | |
4286 | clear_start = found_key.offset; | |
db94535d | 4287 | extent_start = btrfs_file_extent_disk_bytenr(leaf, fi); |
f70a9a6b | 4288 | if (!del_item) { |
db94535d CM |
4289 | u64 orig_num_bytes = |
4290 | btrfs_file_extent_num_bytes(leaf, fi); | |
fda2832f QW |
4291 | extent_num_bytes = ALIGN(new_size - |
4292 | found_key.offset, | |
0b246afa | 4293 | fs_info->sectorsize); |
9ddc959e | 4294 | clear_start = ALIGN(new_size, fs_info->sectorsize); |
db94535d CM |
4295 | btrfs_set_file_extent_num_bytes(leaf, fi, |
4296 | extent_num_bytes); | |
4297 | num_dec = (orig_num_bytes - | |
9069218d | 4298 | extent_num_bytes); |
92a7cc42 | 4299 | if (test_bit(BTRFS_ROOT_SHAREABLE, |
27cdeb70 MX |
4300 | &root->state) && |
4301 | extent_start != 0) | |
a76a3cd4 | 4302 | inode_sub_bytes(inode, num_dec); |
5f39d397 | 4303 | btrfs_mark_buffer_dirty(leaf); |
39279cc3 | 4304 | } else { |
db94535d CM |
4305 | extent_num_bytes = |
4306 | btrfs_file_extent_disk_num_bytes(leaf, | |
4307 | fi); | |
5d4f98a2 YZ |
4308 | extent_offset = found_key.offset - |
4309 | btrfs_file_extent_offset(leaf, fi); | |
4310 | ||
39279cc3 | 4311 | /* FIXME blocksize != 4096 */ |
9069218d | 4312 | num_dec = btrfs_file_extent_num_bytes(leaf, fi); |
39279cc3 CM |
4313 | if (extent_start != 0) { |
4314 | found_extent = 1; | |
92a7cc42 | 4315 | if (test_bit(BTRFS_ROOT_SHAREABLE, |
27cdeb70 | 4316 | &root->state)) |
a76a3cd4 | 4317 | inode_sub_bytes(inode, num_dec); |
e02119d5 | 4318 | } |
39279cc3 | 4319 | } |
9ddc959e | 4320 | clear_len = num_dec; |
9069218d | 4321 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
c8b97818 CM |
4322 | /* |
4323 | * we can't truncate inline items that have had | |
4324 | * special encodings | |
4325 | */ | |
4326 | if (!del_item && | |
c8b97818 | 4327 | btrfs_file_extent_encryption(leaf, fi) == 0 && |
ddfae63c JB |
4328 | btrfs_file_extent_other_encoding(leaf, fi) == 0 && |
4329 | btrfs_file_extent_compression(leaf, fi) == 0) { | |
4330 | u32 size = (u32)(new_size - found_key.offset); | |
4331 | ||
4332 | btrfs_set_file_extent_ram_bytes(leaf, fi, size); | |
4333 | size = btrfs_file_extent_calc_inline_size(size); | |
78ac4f9e | 4334 | btrfs_truncate_item(path, size, 1); |
ddfae63c | 4335 | } else if (!del_item) { |
514ac8ad | 4336 | /* |
ddfae63c JB |
4337 | * We have to bail so the last_size is set to |
4338 | * just before this extent. | |
514ac8ad | 4339 | */ |
fd86a3a3 | 4340 | ret = NEED_TRUNCATE_BLOCK; |
ddfae63c | 4341 | break; |
9ddc959e JB |
4342 | } else { |
4343 | /* | |
4344 | * Inline extents are special, we just treat | |
4345 | * them as a full sector worth in the file | |
4346 | * extent tree just for simplicity sake. | |
4347 | */ | |
4348 | clear_len = fs_info->sectorsize; | |
ddfae63c | 4349 | } |
0305cd5f | 4350 | |
92a7cc42 | 4351 | if (test_bit(BTRFS_ROOT_SHAREABLE, &root->state)) |
0305cd5f | 4352 | inode_sub_bytes(inode, item_end + 1 - new_size); |
39279cc3 | 4353 | } |
179e29e4 | 4354 | delete: |
9ddc959e JB |
4355 | /* |
4356 | * We use btrfs_truncate_inode_items() to clean up log trees for | |
4357 | * multiple fsyncs, and in this case we don't want to clear the | |
4358 | * file extent range because it's just the log. | |
4359 | */ | |
4360 | if (root == BTRFS_I(inode)->root) { | |
4361 | ret = btrfs_inode_clear_file_extent_range(BTRFS_I(inode), | |
4362 | clear_start, clear_len); | |
4363 | if (ret) { | |
4364 | btrfs_abort_transaction(trans, ret); | |
4365 | break; | |
4366 | } | |
4367 | } | |
4368 | ||
ddfae63c JB |
4369 | if (del_item) |
4370 | last_size = found_key.offset; | |
4371 | else | |
4372 | last_size = new_size; | |
39279cc3 | 4373 | if (del_item) { |
85e21bac CM |
4374 | if (!pending_del_nr) { |
4375 | /* no pending yet, add ourselves */ | |
4376 | pending_del_slot = path->slots[0]; | |
4377 | pending_del_nr = 1; | |
4378 | } else if (pending_del_nr && | |
4379 | path->slots[0] + 1 == pending_del_slot) { | |
4380 | /* hop on the pending chunk */ | |
4381 | pending_del_nr++; | |
4382 | pending_del_slot = path->slots[0]; | |
4383 | } else { | |
d397712b | 4384 | BUG(); |
85e21bac | 4385 | } |
39279cc3 CM |
4386 | } else { |
4387 | break; | |
4388 | } | |
897ca819 | 4389 | should_throttle = false; |
28f75a0e | 4390 | |
27cdeb70 | 4391 | if (found_extent && |
82028e0a | 4392 | root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) { |
ffd4bb2a QW |
4393 | struct btrfs_ref ref = { 0 }; |
4394 | ||
28ed1345 | 4395 | bytes_deleted += extent_num_bytes; |
ffd4bb2a QW |
4396 | |
4397 | btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF, | |
4398 | extent_start, extent_num_bytes, 0); | |
4399 | ref.real_root = root->root_key.objectid; | |
4400 | btrfs_init_data_ref(&ref, btrfs_header_owner(leaf), | |
4401 | ino, extent_offset); | |
4402 | ret = btrfs_free_extent(trans, &ref); | |
05522109 OS |
4403 | if (ret) { |
4404 | btrfs_abort_transaction(trans, ret); | |
4405 | break; | |
4406 | } | |
28f75a0e | 4407 | if (be_nice) { |
7c861627 | 4408 | if (btrfs_should_throttle_delayed_refs(trans)) |
897ca819 | 4409 | should_throttle = true; |
28f75a0e | 4410 | } |
39279cc3 | 4411 | } |
85e21bac | 4412 | |
8082510e YZ |
4413 | if (found_type == BTRFS_INODE_ITEM_KEY) |
4414 | break; | |
4415 | ||
4416 | if (path->slots[0] == 0 || | |
1262133b | 4417 | path->slots[0] != pending_del_slot || |
28bad212 | 4418 | should_throttle) { |
8082510e YZ |
4419 | if (pending_del_nr) { |
4420 | ret = btrfs_del_items(trans, root, path, | |
4421 | pending_del_slot, | |
4422 | pending_del_nr); | |
79787eaa | 4423 | if (ret) { |
66642832 | 4424 | btrfs_abort_transaction(trans, ret); |
fd86a3a3 | 4425 | break; |
79787eaa | 4426 | } |
8082510e YZ |
4427 | pending_del_nr = 0; |
4428 | } | |
b3b4aa74 | 4429 | btrfs_release_path(path); |
28bad212 | 4430 | |
28f75a0e | 4431 | /* |
28bad212 JB |
4432 | * We can generate a lot of delayed refs, so we need to |
4433 | * throttle every once and a while and make sure we're | |
4434 | * adding enough space to keep up with the work we are | |
4435 | * generating. Since we hold a transaction here we | |
4436 | * can't flush, and we don't want to FLUSH_LIMIT because | |
4437 | * we could have generated too many delayed refs to | |
4438 | * actually allocate, so just bail if we're short and | |
4439 | * let the normal reservation dance happen higher up. | |
28f75a0e | 4440 | */ |
28bad212 JB |
4441 | if (should_throttle) { |
4442 | ret = btrfs_delayed_refs_rsv_refill(fs_info, | |
4443 | BTRFS_RESERVE_NO_FLUSH); | |
4444 | if (ret) { | |
4445 | ret = -EAGAIN; | |
4446 | break; | |
4447 | } | |
28f75a0e | 4448 | } |
85e21bac | 4449 | goto search_again; |
8082510e YZ |
4450 | } else { |
4451 | path->slots[0]--; | |
85e21bac | 4452 | } |
39279cc3 | 4453 | } |
8082510e | 4454 | out: |
fd86a3a3 OS |
4455 | if (ret >= 0 && pending_del_nr) { |
4456 | int err; | |
4457 | ||
4458 | err = btrfs_del_items(trans, root, path, pending_del_slot, | |
85e21bac | 4459 | pending_del_nr); |
fd86a3a3 OS |
4460 | if (err) { |
4461 | btrfs_abort_transaction(trans, err); | |
4462 | ret = err; | |
4463 | } | |
85e21bac | 4464 | } |
76b42abb FM |
4465 | if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) { |
4466 | ASSERT(last_size >= new_size); | |
fd86a3a3 | 4467 | if (!ret && last_size > new_size) |
76b42abb | 4468 | last_size = new_size; |
d923afe9 | 4469 | btrfs_inode_safe_disk_i_size_write(inode, last_size); |
a5ae50de FM |
4470 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, lock_start, |
4471 | (u64)-1, &cached_state); | |
76b42abb | 4472 | } |
28ed1345 | 4473 | |
39279cc3 | 4474 | btrfs_free_path(path); |
fd86a3a3 | 4475 | return ret; |
39279cc3 CM |
4476 | } |
4477 | ||
4478 | /* | |
9703fefe | 4479 | * btrfs_truncate_block - read, zero a chunk and write a block |
2aaa6655 JB |
4480 | * @inode - inode that we're zeroing |
4481 | * @from - the offset to start zeroing | |
4482 | * @len - the length to zero, 0 to zero the entire range respective to the | |
4483 | * offset | |
4484 | * @front - zero up to the offset instead of from the offset on | |
4485 | * | |
9703fefe | 4486 | * This will find the block for the "from" offset and cow the block and zero the |
2aaa6655 | 4487 | * part we want to zero. This is used with truncate and hole punching. |
39279cc3 | 4488 | */ |
9703fefe | 4489 | int btrfs_truncate_block(struct inode *inode, loff_t from, loff_t len, |
2aaa6655 | 4490 | int front) |
39279cc3 | 4491 | { |
0b246afa | 4492 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
2aaa6655 | 4493 | struct address_space *mapping = inode->i_mapping; |
e6dcd2dc CM |
4494 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
4495 | struct btrfs_ordered_extent *ordered; | |
2ac55d41 | 4496 | struct extent_state *cached_state = NULL; |
364ecf36 | 4497 | struct extent_changeset *data_reserved = NULL; |
e6dcd2dc | 4498 | char *kaddr; |
6d4572a9 | 4499 | bool only_release_metadata = false; |
0b246afa | 4500 | u32 blocksize = fs_info->sectorsize; |
09cbfeaf | 4501 | pgoff_t index = from >> PAGE_SHIFT; |
9703fefe | 4502 | unsigned offset = from & (blocksize - 1); |
39279cc3 | 4503 | struct page *page; |
3b16a4e3 | 4504 | gfp_t mask = btrfs_alloc_write_mask(mapping); |
6d4572a9 | 4505 | size_t write_bytes = blocksize; |
39279cc3 | 4506 | int ret = 0; |
9703fefe CR |
4507 | u64 block_start; |
4508 | u64 block_end; | |
39279cc3 | 4509 | |
b03ebd99 NB |
4510 | if (IS_ALIGNED(offset, blocksize) && |
4511 | (!len || IS_ALIGNED(len, blocksize))) | |
39279cc3 | 4512 | goto out; |
9703fefe | 4513 | |
8b62f87b JB |
4514 | block_start = round_down(from, blocksize); |
4515 | block_end = block_start + blocksize - 1; | |
4516 | ||
36ea6f3e NB |
4517 | ret = btrfs_check_data_free_space(BTRFS_I(inode), &data_reserved, |
4518 | block_start, blocksize); | |
6d4572a9 | 4519 | if (ret < 0) { |
38d37aa9 QW |
4520 | if (btrfs_check_nocow_lock(BTRFS_I(inode), block_start, |
4521 | &write_bytes) > 0) { | |
6d4572a9 QW |
4522 | /* For nocow case, no need to reserve data space */ |
4523 | only_release_metadata = true; | |
4524 | } else { | |
4525 | goto out; | |
4526 | } | |
4527 | } | |
4528 | ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode), blocksize); | |
4529 | if (ret < 0) { | |
4530 | if (!only_release_metadata) | |
25ce28ca NB |
4531 | btrfs_free_reserved_data_space(BTRFS_I(inode), |
4532 | data_reserved, block_start, blocksize); | |
6d4572a9 QW |
4533 | goto out; |
4534 | } | |
211c17f5 | 4535 | again: |
3b16a4e3 | 4536 | page = find_or_create_page(mapping, index, mask); |
5d5e103a | 4537 | if (!page) { |
86d52921 | 4538 | btrfs_delalloc_release_space(BTRFS_I(inode), data_reserved, |
43b18595 | 4539 | block_start, blocksize, true); |
8702ba93 | 4540 | btrfs_delalloc_release_extents(BTRFS_I(inode), blocksize); |
ac6a2b36 | 4541 | ret = -ENOMEM; |
39279cc3 | 4542 | goto out; |
5d5e103a | 4543 | } |
e6dcd2dc | 4544 | |
39279cc3 | 4545 | if (!PageUptodate(page)) { |
9ebefb18 | 4546 | ret = btrfs_readpage(NULL, page); |
39279cc3 | 4547 | lock_page(page); |
211c17f5 CM |
4548 | if (page->mapping != mapping) { |
4549 | unlock_page(page); | |
09cbfeaf | 4550 | put_page(page); |
211c17f5 CM |
4551 | goto again; |
4552 | } | |
39279cc3 CM |
4553 | if (!PageUptodate(page)) { |
4554 | ret = -EIO; | |
89642229 | 4555 | goto out_unlock; |
39279cc3 CM |
4556 | } |
4557 | } | |
211c17f5 | 4558 | wait_on_page_writeback(page); |
e6dcd2dc | 4559 | |
9703fefe | 4560 | lock_extent_bits(io_tree, block_start, block_end, &cached_state); |
e6dcd2dc CM |
4561 | set_page_extent_mapped(page); |
4562 | ||
c3504372 | 4563 | ordered = btrfs_lookup_ordered_extent(BTRFS_I(inode), block_start); |
e6dcd2dc | 4564 | if (ordered) { |
9703fefe | 4565 | unlock_extent_cached(io_tree, block_start, block_end, |
e43bbe5e | 4566 | &cached_state); |
e6dcd2dc | 4567 | unlock_page(page); |
09cbfeaf | 4568 | put_page(page); |
eb84ae03 | 4569 | btrfs_start_ordered_extent(inode, ordered, 1); |
e6dcd2dc CM |
4570 | btrfs_put_ordered_extent(ordered); |
4571 | goto again; | |
4572 | } | |
4573 | ||
9703fefe | 4574 | clear_extent_bit(&BTRFS_I(inode)->io_tree, block_start, block_end, |
e182163d OS |
4575 | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, |
4576 | 0, 0, &cached_state); | |
5d5e103a | 4577 | |
c2566f22 | 4578 | ret = btrfs_set_extent_delalloc(BTRFS_I(inode), block_start, block_end, 0, |
330a5827 | 4579 | &cached_state); |
9ed74f2d | 4580 | if (ret) { |
9703fefe | 4581 | unlock_extent_cached(io_tree, block_start, block_end, |
e43bbe5e | 4582 | &cached_state); |
9ed74f2d JB |
4583 | goto out_unlock; |
4584 | } | |
4585 | ||
9703fefe | 4586 | if (offset != blocksize) { |
2aaa6655 | 4587 | if (!len) |
9703fefe | 4588 | len = blocksize - offset; |
e6dcd2dc | 4589 | kaddr = kmap(page); |
2aaa6655 | 4590 | if (front) |
9703fefe CR |
4591 | memset(kaddr + (block_start - page_offset(page)), |
4592 | 0, offset); | |
2aaa6655 | 4593 | else |
9703fefe CR |
4594 | memset(kaddr + (block_start - page_offset(page)) + offset, |
4595 | 0, len); | |
e6dcd2dc CM |
4596 | flush_dcache_page(page); |
4597 | kunmap(page); | |
4598 | } | |
247e743c | 4599 | ClearPageChecked(page); |
e6dcd2dc | 4600 | set_page_dirty(page); |
e43bbe5e | 4601 | unlock_extent_cached(io_tree, block_start, block_end, &cached_state); |
39279cc3 | 4602 | |
6d4572a9 QW |
4603 | if (only_release_metadata) |
4604 | set_extent_bit(&BTRFS_I(inode)->io_tree, block_start, | |
4605 | block_end, EXTENT_NORESERVE, NULL, NULL, | |
4606 | GFP_NOFS); | |
4607 | ||
89642229 | 4608 | out_unlock: |
6d4572a9 QW |
4609 | if (ret) { |
4610 | if (only_release_metadata) | |
4611 | btrfs_delalloc_release_metadata(BTRFS_I(inode), | |
4612 | blocksize, true); | |
4613 | else | |
86d52921 | 4614 | btrfs_delalloc_release_space(BTRFS_I(inode), data_reserved, |
6d4572a9 QW |
4615 | block_start, blocksize, true); |
4616 | } | |
8702ba93 | 4617 | btrfs_delalloc_release_extents(BTRFS_I(inode), blocksize); |
39279cc3 | 4618 | unlock_page(page); |
09cbfeaf | 4619 | put_page(page); |
39279cc3 | 4620 | out: |
6d4572a9 | 4621 | if (only_release_metadata) |
38d37aa9 | 4622 | btrfs_check_nocow_unlock(BTRFS_I(inode)); |
364ecf36 | 4623 | extent_changeset_free(data_reserved); |
39279cc3 CM |
4624 | return ret; |
4625 | } | |
4626 | ||
16e7549f JB |
4627 | static int maybe_insert_hole(struct btrfs_root *root, struct inode *inode, |
4628 | u64 offset, u64 len) | |
4629 | { | |
0b246afa | 4630 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
16e7549f JB |
4631 | struct btrfs_trans_handle *trans; |
4632 | int ret; | |
4633 | ||
4634 | /* | |
4635 | * Still need to make sure the inode looks like it's been updated so | |
4636 | * that any holes get logged if we fsync. | |
4637 | */ | |
0b246afa JM |
4638 | if (btrfs_fs_incompat(fs_info, NO_HOLES)) { |
4639 | BTRFS_I(inode)->last_trans = fs_info->generation; | |
16e7549f JB |
4640 | BTRFS_I(inode)->last_sub_trans = root->log_transid; |
4641 | BTRFS_I(inode)->last_log_commit = root->last_log_commit; | |
4642 | return 0; | |
4643 | } | |
4644 | ||
4645 | /* | |
4646 | * 1 - for the one we're dropping | |
4647 | * 1 - for the one we're adding | |
4648 | * 1 - for updating the inode. | |
4649 | */ | |
4650 | trans = btrfs_start_transaction(root, 3); | |
4651 | if (IS_ERR(trans)) | |
4652 | return PTR_ERR(trans); | |
4653 | ||
4654 | ret = btrfs_drop_extents(trans, root, inode, offset, offset + len, 1); | |
4655 | if (ret) { | |
66642832 | 4656 | btrfs_abort_transaction(trans, ret); |
3a45bb20 | 4657 | btrfs_end_transaction(trans); |
16e7549f JB |
4658 | return ret; |
4659 | } | |
4660 | ||
f85b7379 DS |
4661 | ret = btrfs_insert_file_extent(trans, root, btrfs_ino(BTRFS_I(inode)), |
4662 | offset, 0, 0, len, 0, len, 0, 0, 0); | |
16e7549f | 4663 | if (ret) |
66642832 | 4664 | btrfs_abort_transaction(trans, ret); |
16e7549f JB |
4665 | else |
4666 | btrfs_update_inode(trans, root, inode); | |
3a45bb20 | 4667 | btrfs_end_transaction(trans); |
16e7549f JB |
4668 | return ret; |
4669 | } | |
4670 | ||
695a0d0d JB |
4671 | /* |
4672 | * This function puts in dummy file extents for the area we're creating a hole | |
4673 | * for. So if we are truncating this file to a larger size we need to insert | |
4674 | * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for | |
4675 | * the range between oldsize and size | |
4676 | */ | |
a41ad394 | 4677 | int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size) |
39279cc3 | 4678 | { |
0b246afa | 4679 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
9036c102 YZ |
4680 | struct btrfs_root *root = BTRFS_I(inode)->root; |
4681 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; | |
a22285a6 | 4682 | struct extent_map *em = NULL; |
2ac55d41 | 4683 | struct extent_state *cached_state = NULL; |
5dc562c5 | 4684 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; |
0b246afa JM |
4685 | u64 hole_start = ALIGN(oldsize, fs_info->sectorsize); |
4686 | u64 block_end = ALIGN(size, fs_info->sectorsize); | |
9036c102 YZ |
4687 | u64 last_byte; |
4688 | u64 cur_offset; | |
4689 | u64 hole_size; | |
9ed74f2d | 4690 | int err = 0; |
39279cc3 | 4691 | |
a71754fc | 4692 | /* |
9703fefe CR |
4693 | * If our size started in the middle of a block we need to zero out the |
4694 | * rest of the block before we expand the i_size, otherwise we could | |
a71754fc JB |
4695 | * expose stale data. |
4696 | */ | |
9703fefe | 4697 | err = btrfs_truncate_block(inode, oldsize, 0, 0); |
a71754fc JB |
4698 | if (err) |
4699 | return err; | |
4700 | ||
9036c102 YZ |
4701 | if (size <= hole_start) |
4702 | return 0; | |
4703 | ||
b272ae22 | 4704 | btrfs_lock_and_flush_ordered_range(BTRFS_I(inode), hole_start, |
23d31bd4 | 4705 | block_end - 1, &cached_state); |
9036c102 YZ |
4706 | cur_offset = hole_start; |
4707 | while (1) { | |
fc4f21b1 | 4708 | em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, cur_offset, |
39b07b5d | 4709 | block_end - cur_offset); |
79787eaa JM |
4710 | if (IS_ERR(em)) { |
4711 | err = PTR_ERR(em); | |
f2767956 | 4712 | em = NULL; |
79787eaa JM |
4713 | break; |
4714 | } | |
9036c102 | 4715 | last_byte = min(extent_map_end(em), block_end); |
0b246afa | 4716 | last_byte = ALIGN(last_byte, fs_info->sectorsize); |
9ddc959e JB |
4717 | hole_size = last_byte - cur_offset; |
4718 | ||
8082510e | 4719 | if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) { |
5dc562c5 | 4720 | struct extent_map *hole_em; |
9ed74f2d | 4721 | |
16e7549f JB |
4722 | err = maybe_insert_hole(root, inode, cur_offset, |
4723 | hole_size); | |
4724 | if (err) | |
3893e33b | 4725 | break; |
9ddc959e JB |
4726 | |
4727 | err = btrfs_inode_set_file_extent_range(BTRFS_I(inode), | |
4728 | cur_offset, hole_size); | |
4729 | if (err) | |
4730 | break; | |
4731 | ||
dcdbc059 | 4732 | btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset, |
5dc562c5 JB |
4733 | cur_offset + hole_size - 1, 0); |
4734 | hole_em = alloc_extent_map(); | |
4735 | if (!hole_em) { | |
4736 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, | |
4737 | &BTRFS_I(inode)->runtime_flags); | |
4738 | goto next; | |
4739 | } | |
4740 | hole_em->start = cur_offset; | |
4741 | hole_em->len = hole_size; | |
4742 | hole_em->orig_start = cur_offset; | |
8082510e | 4743 | |
5dc562c5 JB |
4744 | hole_em->block_start = EXTENT_MAP_HOLE; |
4745 | hole_em->block_len = 0; | |
b4939680 | 4746 | hole_em->orig_block_len = 0; |
cc95bef6 | 4747 | hole_em->ram_bytes = hole_size; |
5dc562c5 | 4748 | hole_em->compress_type = BTRFS_COMPRESS_NONE; |
0b246afa | 4749 | hole_em->generation = fs_info->generation; |
8082510e | 4750 | |
5dc562c5 JB |
4751 | while (1) { |
4752 | write_lock(&em_tree->lock); | |
09a2a8f9 | 4753 | err = add_extent_mapping(em_tree, hole_em, 1); |
5dc562c5 JB |
4754 | write_unlock(&em_tree->lock); |
4755 | if (err != -EEXIST) | |
4756 | break; | |
dcdbc059 NB |
4757 | btrfs_drop_extent_cache(BTRFS_I(inode), |
4758 | cur_offset, | |
5dc562c5 JB |
4759 | cur_offset + |
4760 | hole_size - 1, 0); | |
4761 | } | |
4762 | free_extent_map(hole_em); | |
9ddc959e JB |
4763 | } else { |
4764 | err = btrfs_inode_set_file_extent_range(BTRFS_I(inode), | |
4765 | cur_offset, hole_size); | |
4766 | if (err) | |
4767 | break; | |
9036c102 | 4768 | } |
16e7549f | 4769 | next: |
9036c102 | 4770 | free_extent_map(em); |
a22285a6 | 4771 | em = NULL; |
9036c102 | 4772 | cur_offset = last_byte; |
8082510e | 4773 | if (cur_offset >= block_end) |
9036c102 YZ |
4774 | break; |
4775 | } | |
a22285a6 | 4776 | free_extent_map(em); |
e43bbe5e | 4777 | unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state); |
9036c102 YZ |
4778 | return err; |
4779 | } | |
39279cc3 | 4780 | |
3972f260 | 4781 | static int btrfs_setsize(struct inode *inode, struct iattr *attr) |
8082510e | 4782 | { |
f4a2f4c5 MX |
4783 | struct btrfs_root *root = BTRFS_I(inode)->root; |
4784 | struct btrfs_trans_handle *trans; | |
a41ad394 | 4785 | loff_t oldsize = i_size_read(inode); |
3972f260 ES |
4786 | loff_t newsize = attr->ia_size; |
4787 | int mask = attr->ia_valid; | |
8082510e YZ |
4788 | int ret; |
4789 | ||
3972f260 ES |
4790 | /* |
4791 | * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a | |
4792 | * special case where we need to update the times despite not having | |
4793 | * these flags set. For all other operations the VFS set these flags | |
4794 | * explicitly if it wants a timestamp update. | |
4795 | */ | |
dff6efc3 CH |
4796 | if (newsize != oldsize) { |
4797 | inode_inc_iversion(inode); | |
4798 | if (!(mask & (ATTR_CTIME | ATTR_MTIME))) | |
4799 | inode->i_ctime = inode->i_mtime = | |
c2050a45 | 4800 | current_time(inode); |
dff6efc3 | 4801 | } |
3972f260 | 4802 | |
a41ad394 | 4803 | if (newsize > oldsize) { |
9ea24bbe | 4804 | /* |
ea14b57f | 4805 | * Don't do an expanding truncate while snapshotting is ongoing. |
9ea24bbe FM |
4806 | * This is to ensure the snapshot captures a fully consistent |
4807 | * state of this file - if the snapshot captures this expanding | |
4808 | * truncation, it must capture all writes that happened before | |
4809 | * this truncation. | |
4810 | */ | |
dcc3eb96 | 4811 | btrfs_drew_write_lock(&root->snapshot_lock); |
a41ad394 | 4812 | ret = btrfs_cont_expand(inode, oldsize, newsize); |
9ea24bbe | 4813 | if (ret) { |
dcc3eb96 | 4814 | btrfs_drew_write_unlock(&root->snapshot_lock); |
8082510e | 4815 | return ret; |
9ea24bbe | 4816 | } |
8082510e | 4817 | |
f4a2f4c5 | 4818 | trans = btrfs_start_transaction(root, 1); |
9ea24bbe | 4819 | if (IS_ERR(trans)) { |
dcc3eb96 | 4820 | btrfs_drew_write_unlock(&root->snapshot_lock); |
f4a2f4c5 | 4821 | return PTR_ERR(trans); |
9ea24bbe | 4822 | } |
f4a2f4c5 MX |
4823 | |
4824 | i_size_write(inode, newsize); | |
d923afe9 | 4825 | btrfs_inode_safe_disk_i_size_write(inode, 0); |
27772b68 | 4826 | pagecache_isize_extended(inode, oldsize, newsize); |
f4a2f4c5 | 4827 | ret = btrfs_update_inode(trans, root, inode); |
dcc3eb96 | 4828 | btrfs_drew_write_unlock(&root->snapshot_lock); |
3a45bb20 | 4829 | btrfs_end_transaction(trans); |
a41ad394 | 4830 | } else { |
8082510e | 4831 | |
a41ad394 JB |
4832 | /* |
4833 | * We're truncating a file that used to have good data down to | |
4834 | * zero. Make sure it gets into the ordered flush list so that | |
4835 | * any new writes get down to disk quickly. | |
4836 | */ | |
4837 | if (newsize == 0) | |
72ac3c0d JB |
4838 | set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE, |
4839 | &BTRFS_I(inode)->runtime_flags); | |
8082510e | 4840 | |
a41ad394 | 4841 | truncate_setsize(inode, newsize); |
2e60a51e | 4842 | |
8e0fa5d7 DS |
4843 | /* Disable nonlocked read DIO to avoid the endless truncate */ |
4844 | btrfs_inode_block_unlocked_dio(BTRFS_I(inode)); | |
2e60a51e | 4845 | inode_dio_wait(inode); |
8e0fa5d7 | 4846 | btrfs_inode_resume_unlocked_dio(BTRFS_I(inode)); |
2e60a51e | 4847 | |
213e8c55 | 4848 | ret = btrfs_truncate(inode, newsize == oldsize); |
7f4f6e0a JB |
4849 | if (ret && inode->i_nlink) { |
4850 | int err; | |
4851 | ||
4852 | /* | |
f7e9e8fc OS |
4853 | * Truncate failed, so fix up the in-memory size. We |
4854 | * adjusted disk_i_size down as we removed extents, so | |
4855 | * wait for disk_i_size to be stable and then update the | |
4856 | * in-memory size to match. | |
7f4f6e0a | 4857 | */ |
f7e9e8fc | 4858 | err = btrfs_wait_ordered_range(inode, 0, (u64)-1); |
7f4f6e0a | 4859 | if (err) |
f7e9e8fc OS |
4860 | return err; |
4861 | i_size_write(inode, BTRFS_I(inode)->disk_i_size); | |
7f4f6e0a | 4862 | } |
8082510e YZ |
4863 | } |
4864 | ||
a41ad394 | 4865 | return ret; |
8082510e YZ |
4866 | } |
4867 | ||
9036c102 YZ |
4868 | static int btrfs_setattr(struct dentry *dentry, struct iattr *attr) |
4869 | { | |
2b0143b5 | 4870 | struct inode *inode = d_inode(dentry); |
b83cc969 | 4871 | struct btrfs_root *root = BTRFS_I(inode)->root; |
9036c102 | 4872 | int err; |
39279cc3 | 4873 | |
b83cc969 LZ |
4874 | if (btrfs_root_readonly(root)) |
4875 | return -EROFS; | |
4876 | ||
31051c85 | 4877 | err = setattr_prepare(dentry, attr); |
9036c102 YZ |
4878 | if (err) |
4879 | return err; | |
2bf5a725 | 4880 | |
5a3f23d5 | 4881 | if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) { |
3972f260 | 4882 | err = btrfs_setsize(inode, attr); |
8082510e YZ |
4883 | if (err) |
4884 | return err; | |
39279cc3 | 4885 | } |
9036c102 | 4886 | |
1025774c CH |
4887 | if (attr->ia_valid) { |
4888 | setattr_copy(inode, attr); | |
0c4d2d95 | 4889 | inode_inc_iversion(inode); |
22c44fe6 | 4890 | err = btrfs_dirty_inode(inode); |
1025774c | 4891 | |
22c44fe6 | 4892 | if (!err && attr->ia_valid & ATTR_MODE) |
996a710d | 4893 | err = posix_acl_chmod(inode, inode->i_mode); |
1025774c | 4894 | } |
33268eaf | 4895 | |
39279cc3 CM |
4896 | return err; |
4897 | } | |
61295eb8 | 4898 | |
131e404a FDBM |
4899 | /* |
4900 | * While truncating the inode pages during eviction, we get the VFS calling | |
4901 | * btrfs_invalidatepage() against each page of the inode. This is slow because | |
4902 | * the calls to btrfs_invalidatepage() result in a huge amount of calls to | |
4903 | * lock_extent_bits() and clear_extent_bit(), which keep merging and splitting | |
4904 | * extent_state structures over and over, wasting lots of time. | |
4905 | * | |
4906 | * Therefore if the inode is being evicted, let btrfs_invalidatepage() skip all | |
4907 | * those expensive operations on a per page basis and do only the ordered io | |
4908 | * finishing, while we release here the extent_map and extent_state structures, | |
4909 | * without the excessive merging and splitting. | |
4910 | */ | |
4911 | static void evict_inode_truncate_pages(struct inode *inode) | |
4912 | { | |
4913 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; | |
4914 | struct extent_map_tree *map_tree = &BTRFS_I(inode)->extent_tree; | |
4915 | struct rb_node *node; | |
4916 | ||
4917 | ASSERT(inode->i_state & I_FREEING); | |
91b0abe3 | 4918 | truncate_inode_pages_final(&inode->i_data); |
131e404a FDBM |
4919 | |
4920 | write_lock(&map_tree->lock); | |
07e1ce09 | 4921 | while (!RB_EMPTY_ROOT(&map_tree->map.rb_root)) { |
131e404a FDBM |
4922 | struct extent_map *em; |
4923 | ||
07e1ce09 | 4924 | node = rb_first_cached(&map_tree->map); |
131e404a | 4925 | em = rb_entry(node, struct extent_map, rb_node); |
180589ef WS |
4926 | clear_bit(EXTENT_FLAG_PINNED, &em->flags); |
4927 | clear_bit(EXTENT_FLAG_LOGGING, &em->flags); | |
131e404a FDBM |
4928 | remove_extent_mapping(map_tree, em); |
4929 | free_extent_map(em); | |
7064dd5c FM |
4930 | if (need_resched()) { |
4931 | write_unlock(&map_tree->lock); | |
4932 | cond_resched(); | |
4933 | write_lock(&map_tree->lock); | |
4934 | } | |
131e404a FDBM |
4935 | } |
4936 | write_unlock(&map_tree->lock); | |
4937 | ||
6ca07097 FM |
4938 | /* |
4939 | * Keep looping until we have no more ranges in the io tree. | |
ba206a02 MWO |
4940 | * We can have ongoing bios started by readahead that have |
4941 | * their endio callback (extent_io.c:end_bio_extent_readpage) | |
9c6429d9 FM |
4942 | * still in progress (unlocked the pages in the bio but did not yet |
4943 | * unlocked the ranges in the io tree). Therefore this means some | |
6ca07097 FM |
4944 | * ranges can still be locked and eviction started because before |
4945 | * submitting those bios, which are executed by a separate task (work | |
4946 | * queue kthread), inode references (inode->i_count) were not taken | |
4947 | * (which would be dropped in the end io callback of each bio). | |
4948 | * Therefore here we effectively end up waiting for those bios and | |
4949 | * anyone else holding locked ranges without having bumped the inode's | |
4950 | * reference count - if we don't do it, when they access the inode's | |
4951 | * io_tree to unlock a range it may be too late, leading to an | |
4952 | * use-after-free issue. | |
4953 | */ | |
131e404a FDBM |
4954 | spin_lock(&io_tree->lock); |
4955 | while (!RB_EMPTY_ROOT(&io_tree->state)) { | |
4956 | struct extent_state *state; | |
4957 | struct extent_state *cached_state = NULL; | |
6ca07097 FM |
4958 | u64 start; |
4959 | u64 end; | |
421f0922 | 4960 | unsigned state_flags; |
131e404a FDBM |
4961 | |
4962 | node = rb_first(&io_tree->state); | |
4963 | state = rb_entry(node, struct extent_state, rb_node); | |
6ca07097 FM |
4964 | start = state->start; |
4965 | end = state->end; | |
421f0922 | 4966 | state_flags = state->state; |
131e404a FDBM |
4967 | spin_unlock(&io_tree->lock); |
4968 | ||
ff13db41 | 4969 | lock_extent_bits(io_tree, start, end, &cached_state); |
b9d0b389 QW |
4970 | |
4971 | /* | |
4972 | * If still has DELALLOC flag, the extent didn't reach disk, | |
4973 | * and its reserved space won't be freed by delayed_ref. | |
4974 | * So we need to free its reserved space here. | |
4975 | * (Refer to comment in btrfs_invalidatepage, case 2) | |
4976 | * | |
4977 | * Note, end is the bytenr of last byte, so we need + 1 here. | |
4978 | */ | |
421f0922 | 4979 | if (state_flags & EXTENT_DELALLOC) |
8b8a979f NB |
4980 | btrfs_qgroup_free_data(BTRFS_I(inode), NULL, start, |
4981 | end - start + 1); | |
b9d0b389 | 4982 | |
6ca07097 | 4983 | clear_extent_bit(io_tree, start, end, |
e182163d OS |
4984 | EXTENT_LOCKED | EXTENT_DELALLOC | |
4985 | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 1, 1, | |
4986 | &cached_state); | |
131e404a | 4987 | |
7064dd5c | 4988 | cond_resched(); |
131e404a FDBM |
4989 | spin_lock(&io_tree->lock); |
4990 | } | |
4991 | spin_unlock(&io_tree->lock); | |
4992 | } | |
4993 | ||
4b9d7b59 | 4994 | static struct btrfs_trans_handle *evict_refill_and_join(struct btrfs_root *root, |
ad80cf50 | 4995 | struct btrfs_block_rsv *rsv) |
4b9d7b59 OS |
4996 | { |
4997 | struct btrfs_fs_info *fs_info = root->fs_info; | |
4998 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; | |
d3984c90 | 4999 | struct btrfs_trans_handle *trans; |
2bd36e7b | 5000 | u64 delayed_refs_extra = btrfs_calc_insert_metadata_size(fs_info, 1); |
d3984c90 | 5001 | int ret; |
4b9d7b59 | 5002 | |
d3984c90 JB |
5003 | /* |
5004 | * Eviction should be taking place at some place safe because of our | |
5005 | * delayed iputs. However the normal flushing code will run delayed | |
5006 | * iputs, so we cannot use FLUSH_ALL otherwise we'll deadlock. | |
5007 | * | |
5008 | * We reserve the delayed_refs_extra here again because we can't use | |
5009 | * btrfs_start_transaction(root, 0) for the same deadlocky reason as | |
5010 | * above. We reserve our extra bit here because we generate a ton of | |
5011 | * delayed refs activity by truncating. | |
5012 | * | |
5013 | * If we cannot make our reservation we'll attempt to steal from the | |
5014 | * global reserve, because we really want to be able to free up space. | |
5015 | */ | |
5016 | ret = btrfs_block_rsv_refill(root, rsv, rsv->size + delayed_refs_extra, | |
5017 | BTRFS_RESERVE_FLUSH_EVICT); | |
5018 | if (ret) { | |
4b9d7b59 OS |
5019 | /* |
5020 | * Try to steal from the global reserve if there is space for | |
5021 | * it. | |
5022 | */ | |
d3984c90 JB |
5023 | if (btrfs_check_space_for_delayed_refs(fs_info) || |
5024 | btrfs_block_rsv_migrate(global_rsv, rsv, rsv->size, 0)) { | |
5025 | btrfs_warn(fs_info, | |
5026 | "could not allocate space for delete; will truncate on mount"); | |
5027 | return ERR_PTR(-ENOSPC); | |
5028 | } | |
5029 | delayed_refs_extra = 0; | |
5030 | } | |
4b9d7b59 | 5031 | |
d3984c90 JB |
5032 | trans = btrfs_join_transaction(root); |
5033 | if (IS_ERR(trans)) | |
5034 | return trans; | |
5035 | ||
5036 | if (delayed_refs_extra) { | |
5037 | trans->block_rsv = &fs_info->trans_block_rsv; | |
5038 | trans->bytes_reserved = delayed_refs_extra; | |
5039 | btrfs_block_rsv_migrate(rsv, trans->block_rsv, | |
5040 | delayed_refs_extra, 1); | |
4b9d7b59 | 5041 | } |
d3984c90 | 5042 | return trans; |
4b9d7b59 OS |
5043 | } |
5044 | ||
bd555975 | 5045 | void btrfs_evict_inode(struct inode *inode) |
39279cc3 | 5046 | { |
0b246afa | 5047 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
39279cc3 CM |
5048 | struct btrfs_trans_handle *trans; |
5049 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
4b9d7b59 | 5050 | struct btrfs_block_rsv *rsv; |
39279cc3 CM |
5051 | int ret; |
5052 | ||
1abe9b8a | 5053 | trace_btrfs_inode_evict(inode); |
5054 | ||
3d48d981 | 5055 | if (!root) { |
e8f1bc14 | 5056 | clear_inode(inode); |
3d48d981 NB |
5057 | return; |
5058 | } | |
5059 | ||
131e404a FDBM |
5060 | evict_inode_truncate_pages(inode); |
5061 | ||
69e9c6c6 SB |
5062 | if (inode->i_nlink && |
5063 | ((btrfs_root_refs(&root->root_item) != 0 && | |
5064 | root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) || | |
70ddc553 | 5065 | btrfs_is_free_space_inode(BTRFS_I(inode)))) |
bd555975 AV |
5066 | goto no_delete; |
5067 | ||
27919067 | 5068 | if (is_bad_inode(inode)) |
39279cc3 | 5069 | goto no_delete; |
5f39d397 | 5070 | |
7ab7956e | 5071 | btrfs_free_io_failure_record(BTRFS_I(inode), 0, (u64)-1); |
f612496b | 5072 | |
7b40b695 | 5073 | if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) |
c71bf099 | 5074 | goto no_delete; |
c71bf099 | 5075 | |
76dda93c | 5076 | if (inode->i_nlink > 0) { |
69e9c6c6 SB |
5077 | BUG_ON(btrfs_root_refs(&root->root_item) != 0 && |
5078 | root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID); | |
76dda93c YZ |
5079 | goto no_delete; |
5080 | } | |
5081 | ||
aa79021f | 5082 | ret = btrfs_commit_inode_delayed_inode(BTRFS_I(inode)); |
27919067 | 5083 | if (ret) |
0e8c36a9 | 5084 | goto no_delete; |
0e8c36a9 | 5085 | |
2ff7e61e | 5086 | rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP); |
27919067 | 5087 | if (!rsv) |
4289a667 | 5088 | goto no_delete; |
2bd36e7b | 5089 | rsv->size = btrfs_calc_metadata_size(fs_info, 1); |
ca7e70f5 | 5090 | rsv->failfast = 1; |
4289a667 | 5091 | |
6ef06d27 | 5092 | btrfs_i_size_write(BTRFS_I(inode), 0); |
5f39d397 | 5093 | |
8082510e | 5094 | while (1) { |
ad80cf50 | 5095 | trans = evict_refill_and_join(root, rsv); |
27919067 OS |
5096 | if (IS_ERR(trans)) |
5097 | goto free_rsv; | |
7b128766 | 5098 | |
4289a667 JB |
5099 | trans->block_rsv = rsv; |
5100 | ||
d68fc57b | 5101 | ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0); |
27919067 OS |
5102 | trans->block_rsv = &fs_info->trans_block_rsv; |
5103 | btrfs_end_transaction(trans); | |
5104 | btrfs_btree_balance_dirty(fs_info); | |
5105 | if (ret && ret != -ENOSPC && ret != -EAGAIN) | |
5106 | goto free_rsv; | |
5107 | else if (!ret) | |
8082510e | 5108 | break; |
8082510e | 5109 | } |
5f39d397 | 5110 | |
4ef31a45 | 5111 | /* |
27919067 OS |
5112 | * Errors here aren't a big deal, it just means we leave orphan items in |
5113 | * the tree. They will be cleaned up on the next mount. If the inode | |
5114 | * number gets reused, cleanup deletes the orphan item without doing | |
5115 | * anything, and unlink reuses the existing orphan item. | |
5116 | * | |
5117 | * If it turns out that we are dropping too many of these, we might want | |
5118 | * to add a mechanism for retrying these after a commit. | |
4ef31a45 | 5119 | */ |
ad80cf50 | 5120 | trans = evict_refill_and_join(root, rsv); |
27919067 OS |
5121 | if (!IS_ERR(trans)) { |
5122 | trans->block_rsv = rsv; | |
5123 | btrfs_orphan_del(trans, BTRFS_I(inode)); | |
5124 | trans->block_rsv = &fs_info->trans_block_rsv; | |
5125 | btrfs_end_transaction(trans); | |
5126 | } | |
54aa1f4d | 5127 | |
0b246afa | 5128 | if (!(root == fs_info->tree_root || |
581bb050 | 5129 | root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)) |
4a0cc7ca | 5130 | btrfs_return_ino(root, btrfs_ino(BTRFS_I(inode))); |
581bb050 | 5131 | |
27919067 OS |
5132 | free_rsv: |
5133 | btrfs_free_block_rsv(fs_info, rsv); | |
39279cc3 | 5134 | no_delete: |
27919067 OS |
5135 | /* |
5136 | * If we didn't successfully delete, the orphan item will still be in | |
5137 | * the tree and we'll retry on the next mount. Again, we might also want | |
5138 | * to retry these periodically in the future. | |
5139 | */ | |
f48d1cf5 | 5140 | btrfs_remove_delayed_node(BTRFS_I(inode)); |
dbd5768f | 5141 | clear_inode(inode); |
39279cc3 CM |
5142 | } |
5143 | ||
5144 | /* | |
6bf9e4bd QW |
5145 | * Return the key found in the dir entry in the location pointer, fill @type |
5146 | * with BTRFS_FT_*, and return 0. | |
5147 | * | |
005d6712 SY |
5148 | * If no dir entries were found, returns -ENOENT. |
5149 | * If found a corrupted location in dir entry, returns -EUCLEAN. | |
39279cc3 CM |
5150 | */ |
5151 | static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry, | |
6bf9e4bd | 5152 | struct btrfs_key *location, u8 *type) |
39279cc3 CM |
5153 | { |
5154 | const char *name = dentry->d_name.name; | |
5155 | int namelen = dentry->d_name.len; | |
5156 | struct btrfs_dir_item *di; | |
5157 | struct btrfs_path *path; | |
5158 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
0d9f7f3e | 5159 | int ret = 0; |
39279cc3 CM |
5160 | |
5161 | path = btrfs_alloc_path(); | |
d8926bb3 MF |
5162 | if (!path) |
5163 | return -ENOMEM; | |
3954401f | 5164 | |
f85b7379 DS |
5165 | di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(BTRFS_I(dir)), |
5166 | name, namelen, 0); | |
3cf5068f LB |
5167 | if (IS_ERR_OR_NULL(di)) { |
5168 | ret = di ? PTR_ERR(di) : -ENOENT; | |
005d6712 SY |
5169 | goto out; |
5170 | } | |
d397712b | 5171 | |
5f39d397 | 5172 | btrfs_dir_item_key_to_cpu(path->nodes[0], di, location); |
56a0e706 LB |
5173 | if (location->type != BTRFS_INODE_ITEM_KEY && |
5174 | location->type != BTRFS_ROOT_ITEM_KEY) { | |
005d6712 | 5175 | ret = -EUCLEAN; |
56a0e706 LB |
5176 | btrfs_warn(root->fs_info, |
5177 | "%s gets something invalid in DIR_ITEM (name %s, directory ino %llu, location(%llu %u %llu))", | |
5178 | __func__, name, btrfs_ino(BTRFS_I(dir)), | |
5179 | location->objectid, location->type, location->offset); | |
56a0e706 | 5180 | } |
6bf9e4bd QW |
5181 | if (!ret) |
5182 | *type = btrfs_dir_type(path->nodes[0], di); | |
39279cc3 | 5183 | out: |
39279cc3 CM |
5184 | btrfs_free_path(path); |
5185 | return ret; | |
5186 | } | |
5187 | ||
5188 | /* | |
5189 | * when we hit a tree root in a directory, the btrfs part of the inode | |
5190 | * needs to be changed to reflect the root directory of the tree root. This | |
5191 | * is kind of like crossing a mount point. | |
5192 | */ | |
2ff7e61e | 5193 | static int fixup_tree_root_location(struct btrfs_fs_info *fs_info, |
4df27c4d YZ |
5194 | struct inode *dir, |
5195 | struct dentry *dentry, | |
5196 | struct btrfs_key *location, | |
5197 | struct btrfs_root **sub_root) | |
39279cc3 | 5198 | { |
4df27c4d YZ |
5199 | struct btrfs_path *path; |
5200 | struct btrfs_root *new_root; | |
5201 | struct btrfs_root_ref *ref; | |
5202 | struct extent_buffer *leaf; | |
1d4c08e0 | 5203 | struct btrfs_key key; |
4df27c4d YZ |
5204 | int ret; |
5205 | int err = 0; | |
39279cc3 | 5206 | |
4df27c4d YZ |
5207 | path = btrfs_alloc_path(); |
5208 | if (!path) { | |
5209 | err = -ENOMEM; | |
5210 | goto out; | |
5211 | } | |
39279cc3 | 5212 | |
4df27c4d | 5213 | err = -ENOENT; |
1d4c08e0 DS |
5214 | key.objectid = BTRFS_I(dir)->root->root_key.objectid; |
5215 | key.type = BTRFS_ROOT_REF_KEY; | |
5216 | key.offset = location->objectid; | |
5217 | ||
0b246afa | 5218 | ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0); |
4df27c4d YZ |
5219 | if (ret) { |
5220 | if (ret < 0) | |
5221 | err = ret; | |
5222 | goto out; | |
5223 | } | |
39279cc3 | 5224 | |
4df27c4d YZ |
5225 | leaf = path->nodes[0]; |
5226 | ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref); | |
4a0cc7ca | 5227 | if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(BTRFS_I(dir)) || |
4df27c4d YZ |
5228 | btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len) |
5229 | goto out; | |
39279cc3 | 5230 | |
4df27c4d YZ |
5231 | ret = memcmp_extent_buffer(leaf, dentry->d_name.name, |
5232 | (unsigned long)(ref + 1), | |
5233 | dentry->d_name.len); | |
5234 | if (ret) | |
5235 | goto out; | |
5236 | ||
b3b4aa74 | 5237 | btrfs_release_path(path); |
4df27c4d | 5238 | |
56e9357a | 5239 | new_root = btrfs_get_fs_root(fs_info, location->objectid, true); |
4df27c4d YZ |
5240 | if (IS_ERR(new_root)) { |
5241 | err = PTR_ERR(new_root); | |
5242 | goto out; | |
5243 | } | |
5244 | ||
4df27c4d YZ |
5245 | *sub_root = new_root; |
5246 | location->objectid = btrfs_root_dirid(&new_root->root_item); | |
5247 | location->type = BTRFS_INODE_ITEM_KEY; | |
5248 | location->offset = 0; | |
5249 | err = 0; | |
5250 | out: | |
5251 | btrfs_free_path(path); | |
5252 | return err; | |
39279cc3 CM |
5253 | } |
5254 | ||
5d4f98a2 YZ |
5255 | static void inode_tree_add(struct inode *inode) |
5256 | { | |
5257 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5258 | struct btrfs_inode *entry; | |
03e860bd NP |
5259 | struct rb_node **p; |
5260 | struct rb_node *parent; | |
cef21937 | 5261 | struct rb_node *new = &BTRFS_I(inode)->rb_node; |
4a0cc7ca | 5262 | u64 ino = btrfs_ino(BTRFS_I(inode)); |
5d4f98a2 | 5263 | |
1d3382cb | 5264 | if (inode_unhashed(inode)) |
76dda93c | 5265 | return; |
e1409cef | 5266 | parent = NULL; |
5d4f98a2 | 5267 | spin_lock(&root->inode_lock); |
e1409cef | 5268 | p = &root->inode_tree.rb_node; |
5d4f98a2 YZ |
5269 | while (*p) { |
5270 | parent = *p; | |
5271 | entry = rb_entry(parent, struct btrfs_inode, rb_node); | |
5272 | ||
37508515 | 5273 | if (ino < btrfs_ino(entry)) |
03e860bd | 5274 | p = &parent->rb_left; |
37508515 | 5275 | else if (ino > btrfs_ino(entry)) |
03e860bd | 5276 | p = &parent->rb_right; |
5d4f98a2 YZ |
5277 | else { |
5278 | WARN_ON(!(entry->vfs_inode.i_state & | |
a4ffdde6 | 5279 | (I_WILL_FREE | I_FREEING))); |
cef21937 | 5280 | rb_replace_node(parent, new, &root->inode_tree); |
03e860bd NP |
5281 | RB_CLEAR_NODE(parent); |
5282 | spin_unlock(&root->inode_lock); | |
cef21937 | 5283 | return; |
5d4f98a2 YZ |
5284 | } |
5285 | } | |
cef21937 FDBM |
5286 | rb_link_node(new, parent, p); |
5287 | rb_insert_color(new, &root->inode_tree); | |
5d4f98a2 YZ |
5288 | spin_unlock(&root->inode_lock); |
5289 | } | |
5290 | ||
5291 | static void inode_tree_del(struct inode *inode) | |
5292 | { | |
5293 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
76dda93c | 5294 | int empty = 0; |
5d4f98a2 | 5295 | |
03e860bd | 5296 | spin_lock(&root->inode_lock); |
5d4f98a2 | 5297 | if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) { |
5d4f98a2 | 5298 | rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree); |
5d4f98a2 | 5299 | RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node); |
76dda93c | 5300 | empty = RB_EMPTY_ROOT(&root->inode_tree); |
5d4f98a2 | 5301 | } |
03e860bd | 5302 | spin_unlock(&root->inode_lock); |
76dda93c | 5303 | |
69e9c6c6 | 5304 | if (empty && btrfs_root_refs(&root->root_item) == 0) { |
76dda93c YZ |
5305 | spin_lock(&root->inode_lock); |
5306 | empty = RB_EMPTY_ROOT(&root->inode_tree); | |
5307 | spin_unlock(&root->inode_lock); | |
5308 | if (empty) | |
5309 | btrfs_add_dead_root(root); | |
5310 | } | |
5311 | } | |
5312 | ||
5d4f98a2 | 5313 | |
e02119d5 CM |
5314 | static int btrfs_init_locked_inode(struct inode *inode, void *p) |
5315 | { | |
5316 | struct btrfs_iget_args *args = p; | |
0202e83f DS |
5317 | |
5318 | inode->i_ino = args->ino; | |
5319 | BTRFS_I(inode)->location.objectid = args->ino; | |
5320 | BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY; | |
5321 | BTRFS_I(inode)->location.offset = 0; | |
5c8fd99f JB |
5322 | BTRFS_I(inode)->root = btrfs_grab_root(args->root); |
5323 | BUG_ON(args->root && !BTRFS_I(inode)->root); | |
39279cc3 CM |
5324 | return 0; |
5325 | } | |
5326 | ||
5327 | static int btrfs_find_actor(struct inode *inode, void *opaque) | |
5328 | { | |
5329 | struct btrfs_iget_args *args = opaque; | |
0202e83f DS |
5330 | |
5331 | return args->ino == BTRFS_I(inode)->location.objectid && | |
d397712b | 5332 | args->root == BTRFS_I(inode)->root; |
39279cc3 CM |
5333 | } |
5334 | ||
0202e83f | 5335 | static struct inode *btrfs_iget_locked(struct super_block *s, u64 ino, |
5d4f98a2 | 5336 | struct btrfs_root *root) |
39279cc3 CM |
5337 | { |
5338 | struct inode *inode; | |
5339 | struct btrfs_iget_args args; | |
0202e83f | 5340 | unsigned long hashval = btrfs_inode_hash(ino, root); |
778ba82b | 5341 | |
0202e83f | 5342 | args.ino = ino; |
39279cc3 CM |
5343 | args.root = root; |
5344 | ||
778ba82b | 5345 | inode = iget5_locked(s, hashval, btrfs_find_actor, |
39279cc3 CM |
5346 | btrfs_init_locked_inode, |
5347 | (void *)&args); | |
5348 | return inode; | |
5349 | } | |
5350 | ||
4c66e0d4 | 5351 | /* |
0202e83f | 5352 | * Get an inode object given its inode number and corresponding root. |
4c66e0d4 DS |
5353 | * Path can be preallocated to prevent recursing back to iget through |
5354 | * allocator. NULL is also valid but may require an additional allocation | |
5355 | * later. | |
1a54ef8c | 5356 | */ |
0202e83f | 5357 | struct inode *btrfs_iget_path(struct super_block *s, u64 ino, |
4c66e0d4 | 5358 | struct btrfs_root *root, struct btrfs_path *path) |
1a54ef8c BR |
5359 | { |
5360 | struct inode *inode; | |
5361 | ||
0202e83f | 5362 | inode = btrfs_iget_locked(s, ino, root); |
1a54ef8c | 5363 | if (!inode) |
5d4f98a2 | 5364 | return ERR_PTR(-ENOMEM); |
1a54ef8c BR |
5365 | |
5366 | if (inode->i_state & I_NEW) { | |
67710892 FM |
5367 | int ret; |
5368 | ||
4222ea71 | 5369 | ret = btrfs_read_locked_inode(inode, path); |
9bc2ceff | 5370 | if (!ret) { |
1748f843 MF |
5371 | inode_tree_add(inode); |
5372 | unlock_new_inode(inode); | |
1748f843 | 5373 | } else { |
f5b3a417 AV |
5374 | iget_failed(inode); |
5375 | /* | |
5376 | * ret > 0 can come from btrfs_search_slot called by | |
5377 | * btrfs_read_locked_inode, this means the inode item | |
5378 | * was not found. | |
5379 | */ | |
5380 | if (ret > 0) | |
5381 | ret = -ENOENT; | |
5382 | inode = ERR_PTR(ret); | |
1748f843 MF |
5383 | } |
5384 | } | |
5385 | ||
1a54ef8c BR |
5386 | return inode; |
5387 | } | |
5388 | ||
0202e83f | 5389 | struct inode *btrfs_iget(struct super_block *s, u64 ino, struct btrfs_root *root) |
4222ea71 | 5390 | { |
0202e83f | 5391 | return btrfs_iget_path(s, ino, root, NULL); |
4222ea71 FM |
5392 | } |
5393 | ||
4df27c4d YZ |
5394 | static struct inode *new_simple_dir(struct super_block *s, |
5395 | struct btrfs_key *key, | |
5396 | struct btrfs_root *root) | |
5397 | { | |
5398 | struct inode *inode = new_inode(s); | |
5399 | ||
5400 | if (!inode) | |
5401 | return ERR_PTR(-ENOMEM); | |
5402 | ||
5c8fd99f | 5403 | BTRFS_I(inode)->root = btrfs_grab_root(root); |
4df27c4d | 5404 | memcpy(&BTRFS_I(inode)->location, key, sizeof(*key)); |
72ac3c0d | 5405 | set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags); |
4df27c4d YZ |
5406 | |
5407 | inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID; | |
6bb6b514 OS |
5408 | /* |
5409 | * We only need lookup, the rest is read-only and there's no inode | |
5410 | * associated with the dentry | |
5411 | */ | |
5412 | inode->i_op = &simple_dir_inode_operations; | |
1fdf4194 | 5413 | inode->i_opflags &= ~IOP_XATTR; |
4df27c4d YZ |
5414 | inode->i_fop = &simple_dir_operations; |
5415 | inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO; | |
c2050a45 | 5416 | inode->i_mtime = current_time(inode); |
9cc97d64 | 5417 | inode->i_atime = inode->i_mtime; |
5418 | inode->i_ctime = inode->i_mtime; | |
d3c6be6f | 5419 | BTRFS_I(inode)->i_otime = inode->i_mtime; |
4df27c4d YZ |
5420 | |
5421 | return inode; | |
5422 | } | |
5423 | ||
6bf9e4bd QW |
5424 | static inline u8 btrfs_inode_type(struct inode *inode) |
5425 | { | |
5426 | /* | |
5427 | * Compile-time asserts that generic FT_* types still match | |
5428 | * BTRFS_FT_* types | |
5429 | */ | |
5430 | BUILD_BUG_ON(BTRFS_FT_UNKNOWN != FT_UNKNOWN); | |
5431 | BUILD_BUG_ON(BTRFS_FT_REG_FILE != FT_REG_FILE); | |
5432 | BUILD_BUG_ON(BTRFS_FT_DIR != FT_DIR); | |
5433 | BUILD_BUG_ON(BTRFS_FT_CHRDEV != FT_CHRDEV); | |
5434 | BUILD_BUG_ON(BTRFS_FT_BLKDEV != FT_BLKDEV); | |
5435 | BUILD_BUG_ON(BTRFS_FT_FIFO != FT_FIFO); | |
5436 | BUILD_BUG_ON(BTRFS_FT_SOCK != FT_SOCK); | |
5437 | BUILD_BUG_ON(BTRFS_FT_SYMLINK != FT_SYMLINK); | |
5438 | ||
5439 | return fs_umode_to_ftype(inode->i_mode); | |
5440 | } | |
5441 | ||
3de4586c | 5442 | struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry) |
39279cc3 | 5443 | { |
0b246afa | 5444 | struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb); |
d397712b | 5445 | struct inode *inode; |
4df27c4d | 5446 | struct btrfs_root *root = BTRFS_I(dir)->root; |
39279cc3 CM |
5447 | struct btrfs_root *sub_root = root; |
5448 | struct btrfs_key location; | |
6bf9e4bd | 5449 | u8 di_type = 0; |
b4aff1f8 | 5450 | int ret = 0; |
39279cc3 CM |
5451 | |
5452 | if (dentry->d_name.len > BTRFS_NAME_LEN) | |
5453 | return ERR_PTR(-ENAMETOOLONG); | |
5f39d397 | 5454 | |
6bf9e4bd | 5455 | ret = btrfs_inode_by_name(dir, dentry, &location, &di_type); |
39279cc3 CM |
5456 | if (ret < 0) |
5457 | return ERR_PTR(ret); | |
5f39d397 | 5458 | |
4df27c4d | 5459 | if (location.type == BTRFS_INODE_ITEM_KEY) { |
0202e83f | 5460 | inode = btrfs_iget(dir->i_sb, location.objectid, root); |
6bf9e4bd QW |
5461 | if (IS_ERR(inode)) |
5462 | return inode; | |
5463 | ||
5464 | /* Do extra check against inode mode with di_type */ | |
5465 | if (btrfs_inode_type(inode) != di_type) { | |
5466 | btrfs_crit(fs_info, | |
5467 | "inode mode mismatch with dir: inode mode=0%o btrfs type=%u dir type=%u", | |
5468 | inode->i_mode, btrfs_inode_type(inode), | |
5469 | di_type); | |
5470 | iput(inode); | |
5471 | return ERR_PTR(-EUCLEAN); | |
5472 | } | |
4df27c4d YZ |
5473 | return inode; |
5474 | } | |
5475 | ||
2ff7e61e | 5476 | ret = fixup_tree_root_location(fs_info, dir, dentry, |
4df27c4d YZ |
5477 | &location, &sub_root); |
5478 | if (ret < 0) { | |
5479 | if (ret != -ENOENT) | |
5480 | inode = ERR_PTR(ret); | |
5481 | else | |
5482 | inode = new_simple_dir(dir->i_sb, &location, sub_root); | |
5483 | } else { | |
0202e83f | 5484 | inode = btrfs_iget(dir->i_sb, location.objectid, sub_root); |
39279cc3 | 5485 | } |
8727002f | 5486 | if (root != sub_root) |
00246528 | 5487 | btrfs_put_root(sub_root); |
76dda93c | 5488 | |
34d19bad | 5489 | if (!IS_ERR(inode) && root != sub_root) { |
0b246afa | 5490 | down_read(&fs_info->cleanup_work_sem); |
bc98a42c | 5491 | if (!sb_rdonly(inode->i_sb)) |
66b4ffd1 | 5492 | ret = btrfs_orphan_cleanup(sub_root); |
0b246afa | 5493 | up_read(&fs_info->cleanup_work_sem); |
01cd3367 JB |
5494 | if (ret) { |
5495 | iput(inode); | |
66b4ffd1 | 5496 | inode = ERR_PTR(ret); |
01cd3367 | 5497 | } |
c71bf099 YZ |
5498 | } |
5499 | ||
3de4586c CM |
5500 | return inode; |
5501 | } | |
5502 | ||
fe15ce44 | 5503 | static int btrfs_dentry_delete(const struct dentry *dentry) |
76dda93c YZ |
5504 | { |
5505 | struct btrfs_root *root; | |
2b0143b5 | 5506 | struct inode *inode = d_inode(dentry); |
76dda93c | 5507 | |
848cce0d | 5508 | if (!inode && !IS_ROOT(dentry)) |
2b0143b5 | 5509 | inode = d_inode(dentry->d_parent); |
76dda93c | 5510 | |
848cce0d LZ |
5511 | if (inode) { |
5512 | root = BTRFS_I(inode)->root; | |
efefb143 YZ |
5513 | if (btrfs_root_refs(&root->root_item) == 0) |
5514 | return 1; | |
848cce0d | 5515 | |
4a0cc7ca | 5516 | if (btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) |
848cce0d | 5517 | return 1; |
efefb143 | 5518 | } |
76dda93c YZ |
5519 | return 0; |
5520 | } | |
5521 | ||
3de4586c | 5522 | static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry, |
00cd8dd3 | 5523 | unsigned int flags) |
3de4586c | 5524 | { |
3837d208 | 5525 | struct inode *inode = btrfs_lookup_dentry(dir, dentry); |
5662344b | 5526 | |
3837d208 AV |
5527 | if (inode == ERR_PTR(-ENOENT)) |
5528 | inode = NULL; | |
41d28bca | 5529 | return d_splice_alias(inode, dentry); |
39279cc3 CM |
5530 | } |
5531 | ||
23b5ec74 JB |
5532 | /* |
5533 | * All this infrastructure exists because dir_emit can fault, and we are holding | |
5534 | * the tree lock when doing readdir. For now just allocate a buffer and copy | |
5535 | * our information into that, and then dir_emit from the buffer. This is | |
5536 | * similar to what NFS does, only we don't keep the buffer around in pagecache | |
5537 | * because I'm afraid I'll mess that up. Long term we need to make filldir do | |
5538 | * copy_to_user_inatomic so we don't have to worry about page faulting under the | |
5539 | * tree lock. | |
5540 | */ | |
5541 | static int btrfs_opendir(struct inode *inode, struct file *file) | |
5542 | { | |
5543 | struct btrfs_file_private *private; | |
5544 | ||
5545 | private = kzalloc(sizeof(struct btrfs_file_private), GFP_KERNEL); | |
5546 | if (!private) | |
5547 | return -ENOMEM; | |
5548 | private->filldir_buf = kzalloc(PAGE_SIZE, GFP_KERNEL); | |
5549 | if (!private->filldir_buf) { | |
5550 | kfree(private); | |
5551 | return -ENOMEM; | |
5552 | } | |
5553 | file->private_data = private; | |
5554 | return 0; | |
5555 | } | |
5556 | ||
5557 | struct dir_entry { | |
5558 | u64 ino; | |
5559 | u64 offset; | |
5560 | unsigned type; | |
5561 | int name_len; | |
5562 | }; | |
5563 | ||
5564 | static int btrfs_filldir(void *addr, int entries, struct dir_context *ctx) | |
5565 | { | |
5566 | while (entries--) { | |
5567 | struct dir_entry *entry = addr; | |
5568 | char *name = (char *)(entry + 1); | |
5569 | ||
92d32170 DS |
5570 | ctx->pos = get_unaligned(&entry->offset); |
5571 | if (!dir_emit(ctx, name, get_unaligned(&entry->name_len), | |
5572 | get_unaligned(&entry->ino), | |
5573 | get_unaligned(&entry->type))) | |
23b5ec74 | 5574 | return 1; |
92d32170 DS |
5575 | addr += sizeof(struct dir_entry) + |
5576 | get_unaligned(&entry->name_len); | |
23b5ec74 JB |
5577 | ctx->pos++; |
5578 | } | |
5579 | return 0; | |
5580 | } | |
5581 | ||
9cdda8d3 | 5582 | static int btrfs_real_readdir(struct file *file, struct dir_context *ctx) |
39279cc3 | 5583 | { |
9cdda8d3 | 5584 | struct inode *inode = file_inode(file); |
39279cc3 | 5585 | struct btrfs_root *root = BTRFS_I(inode)->root; |
23b5ec74 | 5586 | struct btrfs_file_private *private = file->private_data; |
39279cc3 CM |
5587 | struct btrfs_dir_item *di; |
5588 | struct btrfs_key key; | |
5f39d397 | 5589 | struct btrfs_key found_key; |
39279cc3 | 5590 | struct btrfs_path *path; |
23b5ec74 | 5591 | void *addr; |
16cdcec7 MX |
5592 | struct list_head ins_list; |
5593 | struct list_head del_list; | |
39279cc3 | 5594 | int ret; |
5f39d397 | 5595 | struct extent_buffer *leaf; |
39279cc3 | 5596 | int slot; |
5f39d397 CM |
5597 | char *name_ptr; |
5598 | int name_len; | |
23b5ec74 JB |
5599 | int entries = 0; |
5600 | int total_len = 0; | |
02dbfc99 | 5601 | bool put = false; |
c2951f32 | 5602 | struct btrfs_key location; |
5f39d397 | 5603 | |
9cdda8d3 AV |
5604 | if (!dir_emit_dots(file, ctx)) |
5605 | return 0; | |
5606 | ||
49593bfa | 5607 | path = btrfs_alloc_path(); |
16cdcec7 MX |
5608 | if (!path) |
5609 | return -ENOMEM; | |
ff5714cc | 5610 | |
23b5ec74 | 5611 | addr = private->filldir_buf; |
e4058b54 | 5612 | path->reada = READA_FORWARD; |
49593bfa | 5613 | |
c2951f32 JM |
5614 | INIT_LIST_HEAD(&ins_list); |
5615 | INIT_LIST_HEAD(&del_list); | |
5616 | put = btrfs_readdir_get_delayed_items(inode, &ins_list, &del_list); | |
16cdcec7 | 5617 | |
23b5ec74 | 5618 | again: |
c2951f32 | 5619 | key.type = BTRFS_DIR_INDEX_KEY; |
9cdda8d3 | 5620 | key.offset = ctx->pos; |
4a0cc7ca | 5621 | key.objectid = btrfs_ino(BTRFS_I(inode)); |
5f39d397 | 5622 | |
39279cc3 CM |
5623 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
5624 | if (ret < 0) | |
5625 | goto err; | |
49593bfa DW |
5626 | |
5627 | while (1) { | |
23b5ec74 JB |
5628 | struct dir_entry *entry; |
5629 | ||
5f39d397 | 5630 | leaf = path->nodes[0]; |
39279cc3 | 5631 | slot = path->slots[0]; |
b9e03af0 LZ |
5632 | if (slot >= btrfs_header_nritems(leaf)) { |
5633 | ret = btrfs_next_leaf(root, path); | |
5634 | if (ret < 0) | |
5635 | goto err; | |
5636 | else if (ret > 0) | |
5637 | break; | |
5638 | continue; | |
39279cc3 | 5639 | } |
3de4586c | 5640 | |
5f39d397 CM |
5641 | btrfs_item_key_to_cpu(leaf, &found_key, slot); |
5642 | ||
5643 | if (found_key.objectid != key.objectid) | |
39279cc3 | 5644 | break; |
c2951f32 | 5645 | if (found_key.type != BTRFS_DIR_INDEX_KEY) |
39279cc3 | 5646 | break; |
9cdda8d3 | 5647 | if (found_key.offset < ctx->pos) |
b9e03af0 | 5648 | goto next; |
c2951f32 | 5649 | if (btrfs_should_delete_dir_index(&del_list, found_key.offset)) |
16cdcec7 | 5650 | goto next; |
39279cc3 | 5651 | di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item); |
c2951f32 | 5652 | name_len = btrfs_dir_name_len(leaf, di); |
23b5ec74 JB |
5653 | if ((total_len + sizeof(struct dir_entry) + name_len) >= |
5654 | PAGE_SIZE) { | |
5655 | btrfs_release_path(path); | |
5656 | ret = btrfs_filldir(private->filldir_buf, entries, ctx); | |
5657 | if (ret) | |
5658 | goto nopos; | |
5659 | addr = private->filldir_buf; | |
5660 | entries = 0; | |
5661 | total_len = 0; | |
5662 | goto again; | |
c2951f32 | 5663 | } |
23b5ec74 JB |
5664 | |
5665 | entry = addr; | |
92d32170 | 5666 | put_unaligned(name_len, &entry->name_len); |
23b5ec74 | 5667 | name_ptr = (char *)(entry + 1); |
c2951f32 JM |
5668 | read_extent_buffer(leaf, name_ptr, (unsigned long)(di + 1), |
5669 | name_len); | |
7d157c3d | 5670 | put_unaligned(fs_ftype_to_dtype(btrfs_dir_type(leaf, di)), |
92d32170 | 5671 | &entry->type); |
c2951f32 | 5672 | btrfs_dir_item_key_to_cpu(leaf, di, &location); |
92d32170 DS |
5673 | put_unaligned(location.objectid, &entry->ino); |
5674 | put_unaligned(found_key.offset, &entry->offset); | |
23b5ec74 JB |
5675 | entries++; |
5676 | addr += sizeof(struct dir_entry) + name_len; | |
5677 | total_len += sizeof(struct dir_entry) + name_len; | |
b9e03af0 LZ |
5678 | next: |
5679 | path->slots[0]++; | |
39279cc3 | 5680 | } |
23b5ec74 JB |
5681 | btrfs_release_path(path); |
5682 | ||
5683 | ret = btrfs_filldir(private->filldir_buf, entries, ctx); | |
5684 | if (ret) | |
5685 | goto nopos; | |
49593bfa | 5686 | |
d2fbb2b5 | 5687 | ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list); |
c2951f32 | 5688 | if (ret) |
bc4ef759 DS |
5689 | goto nopos; |
5690 | ||
db62efbb ZB |
5691 | /* |
5692 | * Stop new entries from being returned after we return the last | |
5693 | * entry. | |
5694 | * | |
5695 | * New directory entries are assigned a strictly increasing | |
5696 | * offset. This means that new entries created during readdir | |
5697 | * are *guaranteed* to be seen in the future by that readdir. | |
5698 | * This has broken buggy programs which operate on names as | |
5699 | * they're returned by readdir. Until we re-use freed offsets | |
5700 | * we have this hack to stop new entries from being returned | |
5701 | * under the assumption that they'll never reach this huge | |
5702 | * offset. | |
5703 | * | |
5704 | * This is being careful not to overflow 32bit loff_t unless the | |
5705 | * last entry requires it because doing so has broken 32bit apps | |
5706 | * in the past. | |
5707 | */ | |
c2951f32 JM |
5708 | if (ctx->pos >= INT_MAX) |
5709 | ctx->pos = LLONG_MAX; | |
5710 | else | |
5711 | ctx->pos = INT_MAX; | |
39279cc3 CM |
5712 | nopos: |
5713 | ret = 0; | |
5714 | err: | |
02dbfc99 OS |
5715 | if (put) |
5716 | btrfs_readdir_put_delayed_items(inode, &ins_list, &del_list); | |
39279cc3 | 5717 | btrfs_free_path(path); |
39279cc3 CM |
5718 | return ret; |
5719 | } | |
5720 | ||
39279cc3 | 5721 | /* |
54aa1f4d | 5722 | * This is somewhat expensive, updating the tree every time the |
39279cc3 CM |
5723 | * inode changes. But, it is most likely to find the inode in cache. |
5724 | * FIXME, needs more benchmarking...there are no reasons other than performance | |
5725 | * to keep or drop this code. | |
5726 | */ | |
48a3b636 | 5727 | static int btrfs_dirty_inode(struct inode *inode) |
39279cc3 | 5728 | { |
2ff7e61e | 5729 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
39279cc3 CM |
5730 | struct btrfs_root *root = BTRFS_I(inode)->root; |
5731 | struct btrfs_trans_handle *trans; | |
8929ecfa YZ |
5732 | int ret; |
5733 | ||
72ac3c0d | 5734 | if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags)) |
22c44fe6 | 5735 | return 0; |
39279cc3 | 5736 | |
7a7eaa40 | 5737 | trans = btrfs_join_transaction(root); |
22c44fe6 JB |
5738 | if (IS_ERR(trans)) |
5739 | return PTR_ERR(trans); | |
8929ecfa YZ |
5740 | |
5741 | ret = btrfs_update_inode(trans, root, inode); | |
94b60442 CM |
5742 | if (ret && ret == -ENOSPC) { |
5743 | /* whoops, lets try again with the full transaction */ | |
3a45bb20 | 5744 | btrfs_end_transaction(trans); |
94b60442 | 5745 | trans = btrfs_start_transaction(root, 1); |
22c44fe6 JB |
5746 | if (IS_ERR(trans)) |
5747 | return PTR_ERR(trans); | |
8929ecfa | 5748 | |
94b60442 | 5749 | ret = btrfs_update_inode(trans, root, inode); |
94b60442 | 5750 | } |
3a45bb20 | 5751 | btrfs_end_transaction(trans); |
16cdcec7 | 5752 | if (BTRFS_I(inode)->delayed_node) |
2ff7e61e | 5753 | btrfs_balance_delayed_items(fs_info); |
22c44fe6 JB |
5754 | |
5755 | return ret; | |
5756 | } | |
5757 | ||
5758 | /* | |
5759 | * This is a copy of file_update_time. We need this so we can return error on | |
5760 | * ENOSPC for updating the inode in the case of file write and mmap writes. | |
5761 | */ | |
95582b00 | 5762 | static int btrfs_update_time(struct inode *inode, struct timespec64 *now, |
e41f941a | 5763 | int flags) |
22c44fe6 | 5764 | { |
2bc55652 | 5765 | struct btrfs_root *root = BTRFS_I(inode)->root; |
3a8c7231 | 5766 | bool dirty = flags & ~S_VERSION; |
2bc55652 AB |
5767 | |
5768 | if (btrfs_root_readonly(root)) | |
5769 | return -EROFS; | |
5770 | ||
e41f941a | 5771 | if (flags & S_VERSION) |
3a8c7231 | 5772 | dirty |= inode_maybe_inc_iversion(inode, dirty); |
e41f941a JB |
5773 | if (flags & S_CTIME) |
5774 | inode->i_ctime = *now; | |
5775 | if (flags & S_MTIME) | |
5776 | inode->i_mtime = *now; | |
5777 | if (flags & S_ATIME) | |
5778 | inode->i_atime = *now; | |
3a8c7231 | 5779 | return dirty ? btrfs_dirty_inode(inode) : 0; |
39279cc3 CM |
5780 | } |
5781 | ||
d352ac68 CM |
5782 | /* |
5783 | * find the highest existing sequence number in a directory | |
5784 | * and then set the in-memory index_cnt variable to reflect | |
5785 | * free sequence numbers | |
5786 | */ | |
4c570655 | 5787 | static int btrfs_set_inode_index_count(struct btrfs_inode *inode) |
aec7477b | 5788 | { |
4c570655 | 5789 | struct btrfs_root *root = inode->root; |
aec7477b JB |
5790 | struct btrfs_key key, found_key; |
5791 | struct btrfs_path *path; | |
5792 | struct extent_buffer *leaf; | |
5793 | int ret; | |
5794 | ||
4c570655 | 5795 | key.objectid = btrfs_ino(inode); |
962a298f | 5796 | key.type = BTRFS_DIR_INDEX_KEY; |
aec7477b JB |
5797 | key.offset = (u64)-1; |
5798 | ||
5799 | path = btrfs_alloc_path(); | |
5800 | if (!path) | |
5801 | return -ENOMEM; | |
5802 | ||
5803 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
5804 | if (ret < 0) | |
5805 | goto out; | |
5806 | /* FIXME: we should be able to handle this */ | |
5807 | if (ret == 0) | |
5808 | goto out; | |
5809 | ret = 0; | |
5810 | ||
5811 | /* | |
5812 | * MAGIC NUMBER EXPLANATION: | |
5813 | * since we search a directory based on f_pos we have to start at 2 | |
5814 | * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody | |
5815 | * else has to start at 2 | |
5816 | */ | |
5817 | if (path->slots[0] == 0) { | |
4c570655 | 5818 | inode->index_cnt = 2; |
aec7477b JB |
5819 | goto out; |
5820 | } | |
5821 | ||
5822 | path->slots[0]--; | |
5823 | ||
5824 | leaf = path->nodes[0]; | |
5825 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
5826 | ||
4c570655 | 5827 | if (found_key.objectid != btrfs_ino(inode) || |
962a298f | 5828 | found_key.type != BTRFS_DIR_INDEX_KEY) { |
4c570655 | 5829 | inode->index_cnt = 2; |
aec7477b JB |
5830 | goto out; |
5831 | } | |
5832 | ||
4c570655 | 5833 | inode->index_cnt = found_key.offset + 1; |
aec7477b JB |
5834 | out: |
5835 | btrfs_free_path(path); | |
5836 | return ret; | |
5837 | } | |
5838 | ||
d352ac68 CM |
5839 | /* |
5840 | * helper to find a free sequence number in a given directory. This current | |
5841 | * code is very simple, later versions will do smarter things in the btree | |
5842 | */ | |
877574e2 | 5843 | int btrfs_set_inode_index(struct btrfs_inode *dir, u64 *index) |
aec7477b JB |
5844 | { |
5845 | int ret = 0; | |
5846 | ||
877574e2 NB |
5847 | if (dir->index_cnt == (u64)-1) { |
5848 | ret = btrfs_inode_delayed_dir_index_count(dir); | |
16cdcec7 MX |
5849 | if (ret) { |
5850 | ret = btrfs_set_inode_index_count(dir); | |
5851 | if (ret) | |
5852 | return ret; | |
5853 | } | |
aec7477b JB |
5854 | } |
5855 | ||
877574e2 NB |
5856 | *index = dir->index_cnt; |
5857 | dir->index_cnt++; | |
aec7477b JB |
5858 | |
5859 | return ret; | |
5860 | } | |
5861 | ||
b0d5d10f CM |
5862 | static int btrfs_insert_inode_locked(struct inode *inode) |
5863 | { | |
5864 | struct btrfs_iget_args args; | |
0202e83f DS |
5865 | |
5866 | args.ino = BTRFS_I(inode)->location.objectid; | |
b0d5d10f CM |
5867 | args.root = BTRFS_I(inode)->root; |
5868 | ||
5869 | return insert_inode_locked4(inode, | |
5870 | btrfs_inode_hash(inode->i_ino, BTRFS_I(inode)->root), | |
5871 | btrfs_find_actor, &args); | |
5872 | } | |
5873 | ||
19aee8de AJ |
5874 | /* |
5875 | * Inherit flags from the parent inode. | |
5876 | * | |
5877 | * Currently only the compression flags and the cow flags are inherited. | |
5878 | */ | |
5879 | static void btrfs_inherit_iflags(struct inode *inode, struct inode *dir) | |
5880 | { | |
5881 | unsigned int flags; | |
5882 | ||
5883 | if (!dir) | |
5884 | return; | |
5885 | ||
5886 | flags = BTRFS_I(dir)->flags; | |
5887 | ||
5888 | if (flags & BTRFS_INODE_NOCOMPRESS) { | |
5889 | BTRFS_I(inode)->flags &= ~BTRFS_INODE_COMPRESS; | |
5890 | BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS; | |
5891 | } else if (flags & BTRFS_INODE_COMPRESS) { | |
5892 | BTRFS_I(inode)->flags &= ~BTRFS_INODE_NOCOMPRESS; | |
5893 | BTRFS_I(inode)->flags |= BTRFS_INODE_COMPRESS; | |
5894 | } | |
5895 | ||
5896 | if (flags & BTRFS_INODE_NODATACOW) { | |
5897 | BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW; | |
5898 | if (S_ISREG(inode->i_mode)) | |
5899 | BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM; | |
5900 | } | |
5901 | ||
7b6a221e | 5902 | btrfs_sync_inode_flags_to_i_flags(inode); |
19aee8de AJ |
5903 | } |
5904 | ||
39279cc3 CM |
5905 | static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans, |
5906 | struct btrfs_root *root, | |
aec7477b | 5907 | struct inode *dir, |
9c58309d | 5908 | const char *name, int name_len, |
175a4eb7 AV |
5909 | u64 ref_objectid, u64 objectid, |
5910 | umode_t mode, u64 *index) | |
39279cc3 | 5911 | { |
0b246afa | 5912 | struct btrfs_fs_info *fs_info = root->fs_info; |
39279cc3 | 5913 | struct inode *inode; |
5f39d397 | 5914 | struct btrfs_inode_item *inode_item; |
39279cc3 | 5915 | struct btrfs_key *location; |
5f39d397 | 5916 | struct btrfs_path *path; |
9c58309d CM |
5917 | struct btrfs_inode_ref *ref; |
5918 | struct btrfs_key key[2]; | |
5919 | u32 sizes[2]; | |
ef3b9af5 | 5920 | int nitems = name ? 2 : 1; |
9c58309d | 5921 | unsigned long ptr; |
11a19a90 | 5922 | unsigned int nofs_flag; |
39279cc3 | 5923 | int ret; |
39279cc3 | 5924 | |
5f39d397 | 5925 | path = btrfs_alloc_path(); |
d8926bb3 MF |
5926 | if (!path) |
5927 | return ERR_PTR(-ENOMEM); | |
5f39d397 | 5928 | |
11a19a90 | 5929 | nofs_flag = memalloc_nofs_save(); |
0b246afa | 5930 | inode = new_inode(fs_info->sb); |
11a19a90 | 5931 | memalloc_nofs_restore(nofs_flag); |
8fb27640 YS |
5932 | if (!inode) { |
5933 | btrfs_free_path(path); | |
39279cc3 | 5934 | return ERR_PTR(-ENOMEM); |
8fb27640 | 5935 | } |
39279cc3 | 5936 | |
5762b5c9 FM |
5937 | /* |
5938 | * O_TMPFILE, set link count to 0, so that after this point, | |
5939 | * we fill in an inode item with the correct link count. | |
5940 | */ | |
5941 | if (!name) | |
5942 | set_nlink(inode, 0); | |
5943 | ||
581bb050 LZ |
5944 | /* |
5945 | * we have to initialize this early, so we can reclaim the inode | |
5946 | * number if we fail afterwards in this function. | |
5947 | */ | |
5948 | inode->i_ino = objectid; | |
5949 | ||
ef3b9af5 | 5950 | if (dir && name) { |
1abe9b8a | 5951 | trace_btrfs_inode_request(dir); |
5952 | ||
877574e2 | 5953 | ret = btrfs_set_inode_index(BTRFS_I(dir), index); |
09771430 | 5954 | if (ret) { |
8fb27640 | 5955 | btrfs_free_path(path); |
09771430 | 5956 | iput(inode); |
aec7477b | 5957 | return ERR_PTR(ret); |
09771430 | 5958 | } |
ef3b9af5 FM |
5959 | } else if (dir) { |
5960 | *index = 0; | |
aec7477b JB |
5961 | } |
5962 | /* | |
5963 | * index_cnt is ignored for everything but a dir, | |
df6703e1 | 5964 | * btrfs_set_inode_index_count has an explanation for the magic |
aec7477b JB |
5965 | * number |
5966 | */ | |
5967 | BTRFS_I(inode)->index_cnt = 2; | |
67de1176 | 5968 | BTRFS_I(inode)->dir_index = *index; |
5c8fd99f | 5969 | BTRFS_I(inode)->root = btrfs_grab_root(root); |
e02119d5 | 5970 | BTRFS_I(inode)->generation = trans->transid; |
76195853 | 5971 | inode->i_generation = BTRFS_I(inode)->generation; |
b888db2b | 5972 | |
5dc562c5 JB |
5973 | /* |
5974 | * We could have gotten an inode number from somebody who was fsynced | |
5975 | * and then removed in this same transaction, so let's just set full | |
5976 | * sync since it will be a full sync anyway and this will blow away the | |
5977 | * old info in the log. | |
5978 | */ | |
5979 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags); | |
5980 | ||
9c58309d | 5981 | key[0].objectid = objectid; |
962a298f | 5982 | key[0].type = BTRFS_INODE_ITEM_KEY; |
9c58309d CM |
5983 | key[0].offset = 0; |
5984 | ||
9c58309d | 5985 | sizes[0] = sizeof(struct btrfs_inode_item); |
ef3b9af5 FM |
5986 | |
5987 | if (name) { | |
5988 | /* | |
5989 | * Start new inodes with an inode_ref. This is slightly more | |
5990 | * efficient for small numbers of hard links since they will | |
5991 | * be packed into one item. Extended refs will kick in if we | |
5992 | * add more hard links than can fit in the ref item. | |
5993 | */ | |
5994 | key[1].objectid = objectid; | |
962a298f | 5995 | key[1].type = BTRFS_INODE_REF_KEY; |
ef3b9af5 FM |
5996 | key[1].offset = ref_objectid; |
5997 | ||
5998 | sizes[1] = name_len + sizeof(*ref); | |
5999 | } | |
9c58309d | 6000 | |
b0d5d10f CM |
6001 | location = &BTRFS_I(inode)->location; |
6002 | location->objectid = objectid; | |
6003 | location->offset = 0; | |
962a298f | 6004 | location->type = BTRFS_INODE_ITEM_KEY; |
b0d5d10f CM |
6005 | |
6006 | ret = btrfs_insert_inode_locked(inode); | |
32955c54 AV |
6007 | if (ret < 0) { |
6008 | iput(inode); | |
b0d5d10f | 6009 | goto fail; |
32955c54 | 6010 | } |
b0d5d10f | 6011 | |
b9473439 | 6012 | path->leave_spinning = 1; |
ef3b9af5 | 6013 | ret = btrfs_insert_empty_items(trans, root, path, key, sizes, nitems); |
9c58309d | 6014 | if (ret != 0) |
b0d5d10f | 6015 | goto fail_unlock; |
5f39d397 | 6016 | |
ecc11fab | 6017 | inode_init_owner(inode, dir, mode); |
a76a3cd4 | 6018 | inode_set_bytes(inode, 0); |
9cc97d64 | 6019 | |
c2050a45 | 6020 | inode->i_mtime = current_time(inode); |
9cc97d64 | 6021 | inode->i_atime = inode->i_mtime; |
6022 | inode->i_ctime = inode->i_mtime; | |
d3c6be6f | 6023 | BTRFS_I(inode)->i_otime = inode->i_mtime; |
9cc97d64 | 6024 | |
5f39d397 CM |
6025 | inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0], |
6026 | struct btrfs_inode_item); | |
b159fa28 | 6027 | memzero_extent_buffer(path->nodes[0], (unsigned long)inode_item, |
293f7e07 | 6028 | sizeof(*inode_item)); |
e02119d5 | 6029 | fill_inode_item(trans, path->nodes[0], inode_item, inode); |
9c58309d | 6030 | |
ef3b9af5 FM |
6031 | if (name) { |
6032 | ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1, | |
6033 | struct btrfs_inode_ref); | |
6034 | btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len); | |
6035 | btrfs_set_inode_ref_index(path->nodes[0], ref, *index); | |
6036 | ptr = (unsigned long)(ref + 1); | |
6037 | write_extent_buffer(path->nodes[0], name, ptr, name_len); | |
6038 | } | |
9c58309d | 6039 | |
5f39d397 CM |
6040 | btrfs_mark_buffer_dirty(path->nodes[0]); |
6041 | btrfs_free_path(path); | |
6042 | ||
6cbff00f CH |
6043 | btrfs_inherit_iflags(inode, dir); |
6044 | ||
569254b0 | 6045 | if (S_ISREG(mode)) { |
0b246afa | 6046 | if (btrfs_test_opt(fs_info, NODATASUM)) |
94272164 | 6047 | BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM; |
0b246afa | 6048 | if (btrfs_test_opt(fs_info, NODATACOW)) |
f2bdf9a8 JB |
6049 | BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW | |
6050 | BTRFS_INODE_NODATASUM; | |
94272164 CM |
6051 | } |
6052 | ||
5d4f98a2 | 6053 | inode_tree_add(inode); |
1abe9b8a | 6054 | |
6055 | trace_btrfs_inode_new(inode); | |
d9094414 | 6056 | btrfs_set_inode_last_trans(trans, BTRFS_I(inode)); |
1abe9b8a | 6057 | |
8ea05e3a AB |
6058 | btrfs_update_root_times(trans, root); |
6059 | ||
63541927 FDBM |
6060 | ret = btrfs_inode_inherit_props(trans, inode, dir); |
6061 | if (ret) | |
0b246afa | 6062 | btrfs_err(fs_info, |
63541927 | 6063 | "error inheriting props for ino %llu (root %llu): %d", |
f85b7379 | 6064 | btrfs_ino(BTRFS_I(inode)), root->root_key.objectid, ret); |
63541927 | 6065 | |
39279cc3 | 6066 | return inode; |
b0d5d10f CM |
6067 | |
6068 | fail_unlock: | |
32955c54 | 6069 | discard_new_inode(inode); |
5f39d397 | 6070 | fail: |
ef3b9af5 | 6071 | if (dir && name) |
aec7477b | 6072 | BTRFS_I(dir)->index_cnt--; |
5f39d397 CM |
6073 | btrfs_free_path(path); |
6074 | return ERR_PTR(ret); | |
39279cc3 CM |
6075 | } |
6076 | ||
d352ac68 CM |
6077 | /* |
6078 | * utility function to add 'inode' into 'parent_inode' with | |
6079 | * a give name and a given sequence number. | |
6080 | * if 'add_backref' is true, also insert a backref from the | |
6081 | * inode to the parent directory. | |
6082 | */ | |
e02119d5 | 6083 | int btrfs_add_link(struct btrfs_trans_handle *trans, |
db0a669f | 6084 | struct btrfs_inode *parent_inode, struct btrfs_inode *inode, |
e02119d5 | 6085 | const char *name, int name_len, int add_backref, u64 index) |
39279cc3 | 6086 | { |
4df27c4d | 6087 | int ret = 0; |
39279cc3 | 6088 | struct btrfs_key key; |
db0a669f NB |
6089 | struct btrfs_root *root = parent_inode->root; |
6090 | u64 ino = btrfs_ino(inode); | |
6091 | u64 parent_ino = btrfs_ino(parent_inode); | |
5f39d397 | 6092 | |
33345d01 | 6093 | if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) { |
db0a669f | 6094 | memcpy(&key, &inode->root->root_key, sizeof(key)); |
4df27c4d | 6095 | } else { |
33345d01 | 6096 | key.objectid = ino; |
962a298f | 6097 | key.type = BTRFS_INODE_ITEM_KEY; |
4df27c4d YZ |
6098 | key.offset = 0; |
6099 | } | |
6100 | ||
33345d01 | 6101 | if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) { |
6025c19f | 6102 | ret = btrfs_add_root_ref(trans, key.objectid, |
0b246afa JM |
6103 | root->root_key.objectid, parent_ino, |
6104 | index, name, name_len); | |
4df27c4d | 6105 | } else if (add_backref) { |
33345d01 LZ |
6106 | ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino, |
6107 | parent_ino, index); | |
4df27c4d | 6108 | } |
39279cc3 | 6109 | |
79787eaa JM |
6110 | /* Nothing to clean up yet */ |
6111 | if (ret) | |
6112 | return ret; | |
4df27c4d | 6113 | |
684572df | 6114 | ret = btrfs_insert_dir_item(trans, name, name_len, parent_inode, &key, |
db0a669f | 6115 | btrfs_inode_type(&inode->vfs_inode), index); |
9c52057c | 6116 | if (ret == -EEXIST || ret == -EOVERFLOW) |
79787eaa JM |
6117 | goto fail_dir_item; |
6118 | else if (ret) { | |
66642832 | 6119 | btrfs_abort_transaction(trans, ret); |
79787eaa | 6120 | return ret; |
39279cc3 | 6121 | } |
79787eaa | 6122 | |
db0a669f | 6123 | btrfs_i_size_write(parent_inode, parent_inode->vfs_inode.i_size + |
79787eaa | 6124 | name_len * 2); |
db0a669f | 6125 | inode_inc_iversion(&parent_inode->vfs_inode); |
5338e43a FM |
6126 | /* |
6127 | * If we are replaying a log tree, we do not want to update the mtime | |
6128 | * and ctime of the parent directory with the current time, since the | |
6129 | * log replay procedure is responsible for setting them to their correct | |
6130 | * values (the ones it had when the fsync was done). | |
6131 | */ | |
6132 | if (!test_bit(BTRFS_FS_LOG_RECOVERING, &root->fs_info->flags)) { | |
6133 | struct timespec64 now = current_time(&parent_inode->vfs_inode); | |
6134 | ||
6135 | parent_inode->vfs_inode.i_mtime = now; | |
6136 | parent_inode->vfs_inode.i_ctime = now; | |
6137 | } | |
db0a669f | 6138 | ret = btrfs_update_inode(trans, root, &parent_inode->vfs_inode); |
79787eaa | 6139 | if (ret) |
66642832 | 6140 | btrfs_abort_transaction(trans, ret); |
39279cc3 | 6141 | return ret; |
fe66a05a CM |
6142 | |
6143 | fail_dir_item: | |
6144 | if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) { | |
6145 | u64 local_index; | |
6146 | int err; | |
3ee1c553 | 6147 | err = btrfs_del_root_ref(trans, key.objectid, |
0b246afa JM |
6148 | root->root_key.objectid, parent_ino, |
6149 | &local_index, name, name_len); | |
1690dd41 JT |
6150 | if (err) |
6151 | btrfs_abort_transaction(trans, err); | |
fe66a05a CM |
6152 | } else if (add_backref) { |
6153 | u64 local_index; | |
6154 | int err; | |
6155 | ||
6156 | err = btrfs_del_inode_ref(trans, root, name, name_len, | |
6157 | ino, parent_ino, &local_index); | |
1690dd41 JT |
6158 | if (err) |
6159 | btrfs_abort_transaction(trans, err); | |
fe66a05a | 6160 | } |
1690dd41 JT |
6161 | |
6162 | /* Return the original error code */ | |
fe66a05a | 6163 | return ret; |
39279cc3 CM |
6164 | } |
6165 | ||
6166 | static int btrfs_add_nondir(struct btrfs_trans_handle *trans, | |
cef415af NB |
6167 | struct btrfs_inode *dir, struct dentry *dentry, |
6168 | struct btrfs_inode *inode, int backref, u64 index) | |
39279cc3 | 6169 | { |
a1b075d2 JB |
6170 | int err = btrfs_add_link(trans, dir, inode, |
6171 | dentry->d_name.name, dentry->d_name.len, | |
6172 | backref, index); | |
39279cc3 CM |
6173 | if (err > 0) |
6174 | err = -EEXIST; | |
6175 | return err; | |
6176 | } | |
6177 | ||
618e21d5 | 6178 | static int btrfs_mknod(struct inode *dir, struct dentry *dentry, |
1a67aafb | 6179 | umode_t mode, dev_t rdev) |
618e21d5 | 6180 | { |
2ff7e61e | 6181 | struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb); |
618e21d5 JB |
6182 | struct btrfs_trans_handle *trans; |
6183 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
1832a6d5 | 6184 | struct inode *inode = NULL; |
618e21d5 | 6185 | int err; |
618e21d5 | 6186 | u64 objectid; |
00e4e6b3 | 6187 | u64 index = 0; |
618e21d5 | 6188 | |
9ed74f2d JB |
6189 | /* |
6190 | * 2 for inode item and ref | |
6191 | * 2 for dir items | |
6192 | * 1 for xattr if selinux is on | |
6193 | */ | |
a22285a6 YZ |
6194 | trans = btrfs_start_transaction(root, 5); |
6195 | if (IS_ERR(trans)) | |
6196 | return PTR_ERR(trans); | |
1832a6d5 | 6197 | |
581bb050 LZ |
6198 | err = btrfs_find_free_ino(root, &objectid); |
6199 | if (err) | |
6200 | goto out_unlock; | |
6201 | ||
aec7477b | 6202 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
f85b7379 DS |
6203 | dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid, |
6204 | mode, &index); | |
7cf96da3 TI |
6205 | if (IS_ERR(inode)) { |
6206 | err = PTR_ERR(inode); | |
32955c54 | 6207 | inode = NULL; |
618e21d5 | 6208 | goto out_unlock; |
7cf96da3 | 6209 | } |
618e21d5 | 6210 | |
ad19db71 CS |
6211 | /* |
6212 | * If the active LSM wants to access the inode during | |
6213 | * d_instantiate it needs these. Smack checks to see | |
6214 | * if the filesystem supports xattrs by looking at the | |
6215 | * ops vector. | |
6216 | */ | |
ad19db71 | 6217 | inode->i_op = &btrfs_special_inode_operations; |
b0d5d10f CM |
6218 | init_special_inode(inode, inode->i_mode, rdev); |
6219 | ||
6220 | err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name); | |
618e21d5 | 6221 | if (err) |
32955c54 | 6222 | goto out_unlock; |
b0d5d10f | 6223 | |
cef415af NB |
6224 | err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode), |
6225 | 0, index); | |
32955c54 AV |
6226 | if (err) |
6227 | goto out_unlock; | |
6228 | ||
6229 | btrfs_update_inode(trans, root, inode); | |
6230 | d_instantiate_new(dentry, inode); | |
b0d5d10f | 6231 | |
618e21d5 | 6232 | out_unlock: |
3a45bb20 | 6233 | btrfs_end_transaction(trans); |
2ff7e61e | 6234 | btrfs_btree_balance_dirty(fs_info); |
32955c54 | 6235 | if (err && inode) { |
618e21d5 | 6236 | inode_dec_link_count(inode); |
32955c54 | 6237 | discard_new_inode(inode); |
618e21d5 | 6238 | } |
618e21d5 JB |
6239 | return err; |
6240 | } | |
6241 | ||
39279cc3 | 6242 | static int btrfs_create(struct inode *dir, struct dentry *dentry, |
ebfc3b49 | 6243 | umode_t mode, bool excl) |
39279cc3 | 6244 | { |
2ff7e61e | 6245 | struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb); |
39279cc3 CM |
6246 | struct btrfs_trans_handle *trans; |
6247 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
1832a6d5 | 6248 | struct inode *inode = NULL; |
a22285a6 | 6249 | int err; |
39279cc3 | 6250 | u64 objectid; |
00e4e6b3 | 6251 | u64 index = 0; |
39279cc3 | 6252 | |
9ed74f2d JB |
6253 | /* |
6254 | * 2 for inode item and ref | |
6255 | * 2 for dir items | |
6256 | * 1 for xattr if selinux is on | |
6257 | */ | |
a22285a6 YZ |
6258 | trans = btrfs_start_transaction(root, 5); |
6259 | if (IS_ERR(trans)) | |
6260 | return PTR_ERR(trans); | |
9ed74f2d | 6261 | |
581bb050 LZ |
6262 | err = btrfs_find_free_ino(root, &objectid); |
6263 | if (err) | |
6264 | goto out_unlock; | |
6265 | ||
aec7477b | 6266 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
f85b7379 DS |
6267 | dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid, |
6268 | mode, &index); | |
7cf96da3 TI |
6269 | if (IS_ERR(inode)) { |
6270 | err = PTR_ERR(inode); | |
32955c54 | 6271 | inode = NULL; |
39279cc3 | 6272 | goto out_unlock; |
7cf96da3 | 6273 | } |
ad19db71 CS |
6274 | /* |
6275 | * If the active LSM wants to access the inode during | |
6276 | * d_instantiate it needs these. Smack checks to see | |
6277 | * if the filesystem supports xattrs by looking at the | |
6278 | * ops vector. | |
6279 | */ | |
6280 | inode->i_fop = &btrfs_file_operations; | |
6281 | inode->i_op = &btrfs_file_inode_operations; | |
b0d5d10f | 6282 | inode->i_mapping->a_ops = &btrfs_aops; |
b0d5d10f CM |
6283 | |
6284 | err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name); | |
6285 | if (err) | |
32955c54 | 6286 | goto out_unlock; |
b0d5d10f CM |
6287 | |
6288 | err = btrfs_update_inode(trans, root, inode); | |
6289 | if (err) | |
32955c54 | 6290 | goto out_unlock; |
ad19db71 | 6291 | |
cef415af NB |
6292 | err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode), |
6293 | 0, index); | |
39279cc3 | 6294 | if (err) |
32955c54 | 6295 | goto out_unlock; |
43baa579 | 6296 | |
43baa579 | 6297 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; |
1e2e547a | 6298 | d_instantiate_new(dentry, inode); |
43baa579 | 6299 | |
39279cc3 | 6300 | out_unlock: |
3a45bb20 | 6301 | btrfs_end_transaction(trans); |
32955c54 | 6302 | if (err && inode) { |
39279cc3 | 6303 | inode_dec_link_count(inode); |
32955c54 | 6304 | discard_new_inode(inode); |
39279cc3 | 6305 | } |
2ff7e61e | 6306 | btrfs_btree_balance_dirty(fs_info); |
39279cc3 CM |
6307 | return err; |
6308 | } | |
6309 | ||
6310 | static int btrfs_link(struct dentry *old_dentry, struct inode *dir, | |
6311 | struct dentry *dentry) | |
6312 | { | |
271dba45 | 6313 | struct btrfs_trans_handle *trans = NULL; |
39279cc3 | 6314 | struct btrfs_root *root = BTRFS_I(dir)->root; |
2b0143b5 | 6315 | struct inode *inode = d_inode(old_dentry); |
2ff7e61e | 6316 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
00e4e6b3 | 6317 | u64 index; |
39279cc3 CM |
6318 | int err; |
6319 | int drop_inode = 0; | |
6320 | ||
4a8be425 | 6321 | /* do not allow sys_link's with other subvols of the same device */ |
4fd786e6 | 6322 | if (root->root_key.objectid != BTRFS_I(inode)->root->root_key.objectid) |
3ab3564f | 6323 | return -EXDEV; |
4a8be425 | 6324 | |
f186373f | 6325 | if (inode->i_nlink >= BTRFS_LINK_MAX) |
c055e99e | 6326 | return -EMLINK; |
4a8be425 | 6327 | |
877574e2 | 6328 | err = btrfs_set_inode_index(BTRFS_I(dir), &index); |
aec7477b JB |
6329 | if (err) |
6330 | goto fail; | |
6331 | ||
a22285a6 | 6332 | /* |
7e6b6465 | 6333 | * 2 items for inode and inode ref |
a22285a6 | 6334 | * 2 items for dir items |
7e6b6465 | 6335 | * 1 item for parent inode |
399b0bbf | 6336 | * 1 item for orphan item deletion if O_TMPFILE |
a22285a6 | 6337 | */ |
399b0bbf | 6338 | trans = btrfs_start_transaction(root, inode->i_nlink ? 5 : 6); |
a22285a6 YZ |
6339 | if (IS_ERR(trans)) { |
6340 | err = PTR_ERR(trans); | |
271dba45 | 6341 | trans = NULL; |
a22285a6 YZ |
6342 | goto fail; |
6343 | } | |
5f39d397 | 6344 | |
67de1176 MX |
6345 | /* There are several dir indexes for this inode, clear the cache. */ |
6346 | BTRFS_I(inode)->dir_index = 0ULL; | |
8b558c5f | 6347 | inc_nlink(inode); |
0c4d2d95 | 6348 | inode_inc_iversion(inode); |
c2050a45 | 6349 | inode->i_ctime = current_time(inode); |
7de9c6ee | 6350 | ihold(inode); |
e9976151 | 6351 | set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags); |
aec7477b | 6352 | |
cef415af NB |
6353 | err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode), |
6354 | 1, index); | |
5f39d397 | 6355 | |
a5719521 | 6356 | if (err) { |
54aa1f4d | 6357 | drop_inode = 1; |
a5719521 | 6358 | } else { |
10d9f309 | 6359 | struct dentry *parent = dentry->d_parent; |
d4682ba0 FM |
6360 | int ret; |
6361 | ||
a5719521 | 6362 | err = btrfs_update_inode(trans, root, inode); |
79787eaa JM |
6363 | if (err) |
6364 | goto fail; | |
ef3b9af5 FM |
6365 | if (inode->i_nlink == 1) { |
6366 | /* | |
6367 | * If new hard link count is 1, it's a file created | |
6368 | * with open(2) O_TMPFILE flag. | |
6369 | */ | |
3d6ae7bb | 6370 | err = btrfs_orphan_del(trans, BTRFS_I(inode)); |
ef3b9af5 FM |
6371 | if (err) |
6372 | goto fail; | |
6373 | } | |
08c422c2 | 6374 | d_instantiate(dentry, inode); |
d4682ba0 FM |
6375 | ret = btrfs_log_new_name(trans, BTRFS_I(inode), NULL, parent, |
6376 | true, NULL); | |
6377 | if (ret == BTRFS_NEED_TRANS_COMMIT) { | |
6378 | err = btrfs_commit_transaction(trans); | |
6379 | trans = NULL; | |
6380 | } | |
a5719521 | 6381 | } |
39279cc3 | 6382 | |
1832a6d5 | 6383 | fail: |
271dba45 | 6384 | if (trans) |
3a45bb20 | 6385 | btrfs_end_transaction(trans); |
39279cc3 CM |
6386 | if (drop_inode) { |
6387 | inode_dec_link_count(inode); | |
6388 | iput(inode); | |
6389 | } | |
2ff7e61e | 6390 | btrfs_btree_balance_dirty(fs_info); |
39279cc3 CM |
6391 | return err; |
6392 | } | |
6393 | ||
18bb1db3 | 6394 | static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) |
39279cc3 | 6395 | { |
2ff7e61e | 6396 | struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb); |
b9d86667 | 6397 | struct inode *inode = NULL; |
39279cc3 CM |
6398 | struct btrfs_trans_handle *trans; |
6399 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
6400 | int err = 0; | |
b9d86667 | 6401 | u64 objectid = 0; |
00e4e6b3 | 6402 | u64 index = 0; |
39279cc3 | 6403 | |
9ed74f2d JB |
6404 | /* |
6405 | * 2 items for inode and ref | |
6406 | * 2 items for dir items | |
6407 | * 1 for xattr if selinux is on | |
6408 | */ | |
a22285a6 YZ |
6409 | trans = btrfs_start_transaction(root, 5); |
6410 | if (IS_ERR(trans)) | |
6411 | return PTR_ERR(trans); | |
39279cc3 | 6412 | |
581bb050 LZ |
6413 | err = btrfs_find_free_ino(root, &objectid); |
6414 | if (err) | |
6415 | goto out_fail; | |
6416 | ||
aec7477b | 6417 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
f85b7379 DS |
6418 | dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid, |
6419 | S_IFDIR | mode, &index); | |
39279cc3 CM |
6420 | if (IS_ERR(inode)) { |
6421 | err = PTR_ERR(inode); | |
32955c54 | 6422 | inode = NULL; |
39279cc3 CM |
6423 | goto out_fail; |
6424 | } | |
5f39d397 | 6425 | |
b0d5d10f CM |
6426 | /* these must be set before we unlock the inode */ |
6427 | inode->i_op = &btrfs_dir_inode_operations; | |
6428 | inode->i_fop = &btrfs_dir_file_operations; | |
33268eaf | 6429 | |
2a7dba39 | 6430 | err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name); |
33268eaf | 6431 | if (err) |
32955c54 | 6432 | goto out_fail; |
39279cc3 | 6433 | |
6ef06d27 | 6434 | btrfs_i_size_write(BTRFS_I(inode), 0); |
39279cc3 CM |
6435 | err = btrfs_update_inode(trans, root, inode); |
6436 | if (err) | |
32955c54 | 6437 | goto out_fail; |
5f39d397 | 6438 | |
db0a669f NB |
6439 | err = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode), |
6440 | dentry->d_name.name, | |
6441 | dentry->d_name.len, 0, index); | |
39279cc3 | 6442 | if (err) |
32955c54 | 6443 | goto out_fail; |
5f39d397 | 6444 | |
1e2e547a | 6445 | d_instantiate_new(dentry, inode); |
39279cc3 CM |
6446 | |
6447 | out_fail: | |
3a45bb20 | 6448 | btrfs_end_transaction(trans); |
32955c54 | 6449 | if (err && inode) { |
c7cfb8a5 | 6450 | inode_dec_link_count(inode); |
32955c54 | 6451 | discard_new_inode(inode); |
c7cfb8a5 | 6452 | } |
2ff7e61e | 6453 | btrfs_btree_balance_dirty(fs_info); |
39279cc3 CM |
6454 | return err; |
6455 | } | |
6456 | ||
c8b97818 | 6457 | static noinline int uncompress_inline(struct btrfs_path *path, |
e40da0e5 | 6458 | struct page *page, |
c8b97818 CM |
6459 | size_t pg_offset, u64 extent_offset, |
6460 | struct btrfs_file_extent_item *item) | |
6461 | { | |
6462 | int ret; | |
6463 | struct extent_buffer *leaf = path->nodes[0]; | |
6464 | char *tmp; | |
6465 | size_t max_size; | |
6466 | unsigned long inline_size; | |
6467 | unsigned long ptr; | |
261507a0 | 6468 | int compress_type; |
c8b97818 CM |
6469 | |
6470 | WARN_ON(pg_offset != 0); | |
261507a0 | 6471 | compress_type = btrfs_file_extent_compression(leaf, item); |
c8b97818 CM |
6472 | max_size = btrfs_file_extent_ram_bytes(leaf, item); |
6473 | inline_size = btrfs_file_extent_inline_item_len(leaf, | |
dd3cc16b | 6474 | btrfs_item_nr(path->slots[0])); |
c8b97818 | 6475 | tmp = kmalloc(inline_size, GFP_NOFS); |
8d413713 TI |
6476 | if (!tmp) |
6477 | return -ENOMEM; | |
c8b97818 CM |
6478 | ptr = btrfs_file_extent_inline_start(item); |
6479 | ||
6480 | read_extent_buffer(leaf, tmp, ptr, inline_size); | |
6481 | ||
09cbfeaf | 6482 | max_size = min_t(unsigned long, PAGE_SIZE, max_size); |
261507a0 LZ |
6483 | ret = btrfs_decompress(compress_type, tmp, page, |
6484 | extent_offset, inline_size, max_size); | |
e1699d2d ZB |
6485 | |
6486 | /* | |
6487 | * decompression code contains a memset to fill in any space between the end | |
6488 | * of the uncompressed data and the end of max_size in case the decompressed | |
6489 | * data ends up shorter than ram_bytes. That doesn't cover the hole between | |
6490 | * the end of an inline extent and the beginning of the next block, so we | |
6491 | * cover that region here. | |
6492 | */ | |
6493 | ||
6494 | if (max_size + pg_offset < PAGE_SIZE) { | |
6495 | char *map = kmap(page); | |
6496 | memset(map + pg_offset + max_size, 0, PAGE_SIZE - max_size - pg_offset); | |
6497 | kunmap(page); | |
6498 | } | |
c8b97818 | 6499 | kfree(tmp); |
166ae5a4 | 6500 | return ret; |
c8b97818 CM |
6501 | } |
6502 | ||
39b07b5d OS |
6503 | /** |
6504 | * btrfs_get_extent - Lookup the first extent overlapping a range in a file. | |
6505 | * @inode: file to search in | |
6506 | * @page: page to read extent data into if the extent is inline | |
6507 | * @pg_offset: offset into @page to copy to | |
6508 | * @start: file offset | |
6509 | * @len: length of range starting at @start | |
6510 | * | |
6511 | * This returns the first &struct extent_map which overlaps with the given | |
6512 | * range, reading it from the B-tree and caching it if necessary. Note that | |
6513 | * there may be more extents which overlap the given range after the returned | |
6514 | * extent_map. | |
d352ac68 | 6515 | * |
39b07b5d OS |
6516 | * If @page is not NULL and the extent is inline, this also reads the extent |
6517 | * data directly into the page and marks the extent up to date in the io_tree. | |
6518 | * | |
6519 | * Return: ERR_PTR on error, non-NULL extent_map on success. | |
d352ac68 | 6520 | */ |
fc4f21b1 | 6521 | struct extent_map *btrfs_get_extent(struct btrfs_inode *inode, |
39b07b5d OS |
6522 | struct page *page, size_t pg_offset, |
6523 | u64 start, u64 len) | |
a52d9a80 | 6524 | { |
3ffbd68c | 6525 | struct btrfs_fs_info *fs_info = inode->root->fs_info; |
a52d9a80 CM |
6526 | int ret; |
6527 | int err = 0; | |
a52d9a80 CM |
6528 | u64 extent_start = 0; |
6529 | u64 extent_end = 0; | |
fc4f21b1 | 6530 | u64 objectid = btrfs_ino(inode); |
7e74e235 | 6531 | int extent_type = -1; |
f421950f | 6532 | struct btrfs_path *path = NULL; |
fc4f21b1 | 6533 | struct btrfs_root *root = inode->root; |
a52d9a80 | 6534 | struct btrfs_file_extent_item *item; |
5f39d397 CM |
6535 | struct extent_buffer *leaf; |
6536 | struct btrfs_key found_key; | |
a52d9a80 | 6537 | struct extent_map *em = NULL; |
fc4f21b1 NB |
6538 | struct extent_map_tree *em_tree = &inode->extent_tree; |
6539 | struct extent_io_tree *io_tree = &inode->io_tree; | |
a52d9a80 | 6540 | |
890871be | 6541 | read_lock(&em_tree->lock); |
d1310b2e | 6542 | em = lookup_extent_mapping(em_tree, start, len); |
890871be | 6543 | read_unlock(&em_tree->lock); |
d1310b2e | 6544 | |
a52d9a80 | 6545 | if (em) { |
e1c4b745 CM |
6546 | if (em->start > start || em->start + em->len <= start) |
6547 | free_extent_map(em); | |
6548 | else if (em->block_start == EXTENT_MAP_INLINE && page) | |
70dec807 CM |
6549 | free_extent_map(em); |
6550 | else | |
6551 | goto out; | |
a52d9a80 | 6552 | } |
172ddd60 | 6553 | em = alloc_extent_map(); |
a52d9a80 | 6554 | if (!em) { |
d1310b2e CM |
6555 | err = -ENOMEM; |
6556 | goto out; | |
a52d9a80 | 6557 | } |
d1310b2e | 6558 | em->start = EXTENT_MAP_HOLE; |
445a6944 | 6559 | em->orig_start = EXTENT_MAP_HOLE; |
d1310b2e | 6560 | em->len = (u64)-1; |
c8b97818 | 6561 | em->block_len = (u64)-1; |
f421950f | 6562 | |
bee6ec82 | 6563 | path = btrfs_alloc_path(); |
f421950f | 6564 | if (!path) { |
bee6ec82 LB |
6565 | err = -ENOMEM; |
6566 | goto out; | |
f421950f CM |
6567 | } |
6568 | ||
bee6ec82 LB |
6569 | /* Chances are we'll be called again, so go ahead and do readahead */ |
6570 | path->reada = READA_FORWARD; | |
6571 | ||
e49aabd9 LB |
6572 | /* |
6573 | * Unless we're going to uncompress the inline extent, no sleep would | |
6574 | * happen. | |
6575 | */ | |
6576 | path->leave_spinning = 1; | |
6577 | ||
5c9a702e | 6578 | ret = btrfs_lookup_file_extent(NULL, root, path, objectid, start, 0); |
a52d9a80 CM |
6579 | if (ret < 0) { |
6580 | err = ret; | |
6581 | goto out; | |
b8eeab7f | 6582 | } else if (ret > 0) { |
a52d9a80 CM |
6583 | if (path->slots[0] == 0) |
6584 | goto not_found; | |
6585 | path->slots[0]--; | |
6586 | } | |
6587 | ||
5f39d397 CM |
6588 | leaf = path->nodes[0]; |
6589 | item = btrfs_item_ptr(leaf, path->slots[0], | |
a52d9a80 | 6590 | struct btrfs_file_extent_item); |
5f39d397 | 6591 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
5f39d397 | 6592 | if (found_key.objectid != objectid || |
694c12ed | 6593 | found_key.type != BTRFS_EXTENT_DATA_KEY) { |
25a50341 JB |
6594 | /* |
6595 | * If we backup past the first extent we want to move forward | |
6596 | * and see if there is an extent in front of us, otherwise we'll | |
6597 | * say there is a hole for our whole search range which can | |
6598 | * cause problems. | |
6599 | */ | |
6600 | extent_end = start; | |
6601 | goto next; | |
a52d9a80 CM |
6602 | } |
6603 | ||
694c12ed | 6604 | extent_type = btrfs_file_extent_type(leaf, item); |
5f39d397 | 6605 | extent_start = found_key.offset; |
a5eeb3d1 | 6606 | extent_end = btrfs_file_extent_end(path); |
694c12ed NB |
6607 | if (extent_type == BTRFS_FILE_EXTENT_REG || |
6608 | extent_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
6bf9e4bd QW |
6609 | /* Only regular file could have regular/prealloc extent */ |
6610 | if (!S_ISREG(inode->vfs_inode.i_mode)) { | |
6611 | ret = -EUCLEAN; | |
6612 | btrfs_crit(fs_info, | |
6613 | "regular/prealloc extent found for non-regular inode %llu", | |
6614 | btrfs_ino(inode)); | |
6615 | goto out; | |
6616 | } | |
09ed2f16 LB |
6617 | trace_btrfs_get_extent_show_fi_regular(inode, leaf, item, |
6618 | extent_start); | |
694c12ed | 6619 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
09ed2f16 LB |
6620 | trace_btrfs_get_extent_show_fi_inline(inode, leaf, item, |
6621 | path->slots[0], | |
6622 | extent_start); | |
9036c102 | 6623 | } |
25a50341 | 6624 | next: |
9036c102 YZ |
6625 | if (start >= extent_end) { |
6626 | path->slots[0]++; | |
6627 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
6628 | ret = btrfs_next_leaf(root, path); | |
6629 | if (ret < 0) { | |
6630 | err = ret; | |
6631 | goto out; | |
b8eeab7f | 6632 | } else if (ret > 0) { |
9036c102 | 6633 | goto not_found; |
b8eeab7f | 6634 | } |
9036c102 | 6635 | leaf = path->nodes[0]; |
a52d9a80 | 6636 | } |
9036c102 YZ |
6637 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
6638 | if (found_key.objectid != objectid || | |
6639 | found_key.type != BTRFS_EXTENT_DATA_KEY) | |
6640 | goto not_found; | |
6641 | if (start + len <= found_key.offset) | |
6642 | goto not_found; | |
e2eca69d WS |
6643 | if (start > found_key.offset) |
6644 | goto next; | |
02a033df NB |
6645 | |
6646 | /* New extent overlaps with existing one */ | |
9036c102 | 6647 | em->start = start; |
70c8a91c | 6648 | em->orig_start = start; |
9036c102 | 6649 | em->len = found_key.offset - start; |
02a033df NB |
6650 | em->block_start = EXTENT_MAP_HOLE; |
6651 | goto insert; | |
9036c102 YZ |
6652 | } |
6653 | ||
39b07b5d | 6654 | btrfs_extent_item_to_extent_map(inode, path, item, !page, em); |
7ffbb598 | 6655 | |
694c12ed NB |
6656 | if (extent_type == BTRFS_FILE_EXTENT_REG || |
6657 | extent_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
a52d9a80 | 6658 | goto insert; |
694c12ed | 6659 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
5f39d397 | 6660 | unsigned long ptr; |
a52d9a80 | 6661 | char *map; |
3326d1b0 CM |
6662 | size_t size; |
6663 | size_t extent_offset; | |
6664 | size_t copy_size; | |
a52d9a80 | 6665 | |
39b07b5d | 6666 | if (!page) |
689f9346 | 6667 | goto out; |
5f39d397 | 6668 | |
e41ca589 | 6669 | size = btrfs_file_extent_ram_bytes(leaf, item); |
9036c102 | 6670 | extent_offset = page_offset(page) + pg_offset - extent_start; |
09cbfeaf KS |
6671 | copy_size = min_t(u64, PAGE_SIZE - pg_offset, |
6672 | size - extent_offset); | |
3326d1b0 | 6673 | em->start = extent_start + extent_offset; |
0b246afa | 6674 | em->len = ALIGN(copy_size, fs_info->sectorsize); |
b4939680 | 6675 | em->orig_block_len = em->len; |
70c8a91c | 6676 | em->orig_start = em->start; |
689f9346 | 6677 | ptr = btrfs_file_extent_inline_start(item) + extent_offset; |
e49aabd9 LB |
6678 | |
6679 | btrfs_set_path_blocking(path); | |
bf46f52d | 6680 | if (!PageUptodate(page)) { |
261507a0 LZ |
6681 | if (btrfs_file_extent_compression(leaf, item) != |
6682 | BTRFS_COMPRESS_NONE) { | |
e40da0e5 | 6683 | ret = uncompress_inline(path, page, pg_offset, |
c8b97818 | 6684 | extent_offset, item); |
166ae5a4 ZB |
6685 | if (ret) { |
6686 | err = ret; | |
6687 | goto out; | |
6688 | } | |
c8b97818 CM |
6689 | } else { |
6690 | map = kmap(page); | |
6691 | read_extent_buffer(leaf, map + pg_offset, ptr, | |
6692 | copy_size); | |
09cbfeaf | 6693 | if (pg_offset + copy_size < PAGE_SIZE) { |
93c82d57 | 6694 | memset(map + pg_offset + copy_size, 0, |
09cbfeaf | 6695 | PAGE_SIZE - pg_offset - |
93c82d57 CM |
6696 | copy_size); |
6697 | } | |
c8b97818 CM |
6698 | kunmap(page); |
6699 | } | |
179e29e4 | 6700 | flush_dcache_page(page); |
a52d9a80 | 6701 | } |
d1310b2e | 6702 | set_extent_uptodate(io_tree, em->start, |
507903b8 | 6703 | extent_map_end(em) - 1, NULL, GFP_NOFS); |
a52d9a80 | 6704 | goto insert; |
a52d9a80 CM |
6705 | } |
6706 | not_found: | |
6707 | em->start = start; | |
70c8a91c | 6708 | em->orig_start = start; |
d1310b2e | 6709 | em->len = len; |
5f39d397 | 6710 | em->block_start = EXTENT_MAP_HOLE; |
a52d9a80 | 6711 | insert: |
b3b4aa74 | 6712 | btrfs_release_path(path); |
d1310b2e | 6713 | if (em->start > start || extent_map_end(em) <= start) { |
0b246afa | 6714 | btrfs_err(fs_info, |
5d163e0e JM |
6715 | "bad extent! em: [%llu %llu] passed [%llu %llu]", |
6716 | em->start, em->len, start, len); | |
a52d9a80 CM |
6717 | err = -EIO; |
6718 | goto out; | |
6719 | } | |
d1310b2e CM |
6720 | |
6721 | err = 0; | |
890871be | 6722 | write_lock(&em_tree->lock); |
f46b24c9 | 6723 | err = btrfs_add_extent_mapping(fs_info, em_tree, &em, start, len); |
890871be | 6724 | write_unlock(&em_tree->lock); |
a52d9a80 | 6725 | out: |
c6414280 | 6726 | btrfs_free_path(path); |
1abe9b8a | 6727 | |
fc4f21b1 | 6728 | trace_btrfs_get_extent(root, inode, em); |
1abe9b8a | 6729 | |
a52d9a80 CM |
6730 | if (err) { |
6731 | free_extent_map(em); | |
a52d9a80 CM |
6732 | return ERR_PTR(err); |
6733 | } | |
79787eaa | 6734 | BUG_ON(!em); /* Error is always set */ |
a52d9a80 CM |
6735 | return em; |
6736 | } | |
6737 | ||
fc4f21b1 | 6738 | struct extent_map *btrfs_get_extent_fiemap(struct btrfs_inode *inode, |
4ab47a8d | 6739 | u64 start, u64 len) |
ec29ed5b CM |
6740 | { |
6741 | struct extent_map *em; | |
6742 | struct extent_map *hole_em = NULL; | |
f3714ef4 | 6743 | u64 delalloc_start = start; |
ec29ed5b | 6744 | u64 end; |
f3714ef4 NB |
6745 | u64 delalloc_len; |
6746 | u64 delalloc_end; | |
ec29ed5b CM |
6747 | int err = 0; |
6748 | ||
39b07b5d | 6749 | em = btrfs_get_extent(inode, NULL, 0, start, len); |
ec29ed5b CM |
6750 | if (IS_ERR(em)) |
6751 | return em; | |
9986277e DC |
6752 | /* |
6753 | * If our em maps to: | |
6754 | * - a hole or | |
6755 | * - a pre-alloc extent, | |
6756 | * there might actually be delalloc bytes behind it. | |
6757 | */ | |
6758 | if (em->block_start != EXTENT_MAP_HOLE && | |
6759 | !test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) | |
6760 | return em; | |
6761 | else | |
6762 | hole_em = em; | |
ec29ed5b CM |
6763 | |
6764 | /* check to see if we've wrapped (len == -1 or similar) */ | |
6765 | end = start + len; | |
6766 | if (end < start) | |
6767 | end = (u64)-1; | |
6768 | else | |
6769 | end -= 1; | |
6770 | ||
6771 | em = NULL; | |
6772 | ||
6773 | /* ok, we didn't find anything, lets look for delalloc */ | |
f3714ef4 | 6774 | delalloc_len = count_range_bits(&inode->io_tree, &delalloc_start, |
ec29ed5b | 6775 | end, len, EXTENT_DELALLOC, 1); |
f3714ef4 NB |
6776 | delalloc_end = delalloc_start + delalloc_len; |
6777 | if (delalloc_end < delalloc_start) | |
6778 | delalloc_end = (u64)-1; | |
ec29ed5b CM |
6779 | |
6780 | /* | |
f3714ef4 NB |
6781 | * We didn't find anything useful, return the original results from |
6782 | * get_extent() | |
ec29ed5b | 6783 | */ |
f3714ef4 | 6784 | if (delalloc_start > end || delalloc_end <= start) { |
ec29ed5b CM |
6785 | em = hole_em; |
6786 | hole_em = NULL; | |
6787 | goto out; | |
6788 | } | |
6789 | ||
f3714ef4 NB |
6790 | /* |
6791 | * Adjust the delalloc_start to make sure it doesn't go backwards from | |
6792 | * the start they passed in | |
ec29ed5b | 6793 | */ |
f3714ef4 NB |
6794 | delalloc_start = max(start, delalloc_start); |
6795 | delalloc_len = delalloc_end - delalloc_start; | |
ec29ed5b | 6796 | |
f3714ef4 NB |
6797 | if (delalloc_len > 0) { |
6798 | u64 hole_start; | |
02950af4 | 6799 | u64 hole_len; |
f3714ef4 | 6800 | const u64 hole_end = extent_map_end(hole_em); |
ec29ed5b | 6801 | |
172ddd60 | 6802 | em = alloc_extent_map(); |
ec29ed5b CM |
6803 | if (!em) { |
6804 | err = -ENOMEM; | |
6805 | goto out; | |
6806 | } | |
f3714ef4 NB |
6807 | |
6808 | ASSERT(hole_em); | |
ec29ed5b | 6809 | /* |
f3714ef4 NB |
6810 | * When btrfs_get_extent can't find anything it returns one |
6811 | * huge hole | |
ec29ed5b | 6812 | * |
f3714ef4 NB |
6813 | * Make sure what it found really fits our range, and adjust to |
6814 | * make sure it is based on the start from the caller | |
ec29ed5b | 6815 | */ |
f3714ef4 NB |
6816 | if (hole_end <= start || hole_em->start > end) { |
6817 | free_extent_map(hole_em); | |
6818 | hole_em = NULL; | |
6819 | } else { | |
6820 | hole_start = max(hole_em->start, start); | |
6821 | hole_len = hole_end - hole_start; | |
ec29ed5b | 6822 | } |
f3714ef4 NB |
6823 | |
6824 | if (hole_em && delalloc_start > hole_start) { | |
6825 | /* | |
6826 | * Our hole starts before our delalloc, so we have to | |
6827 | * return just the parts of the hole that go until the | |
6828 | * delalloc starts | |
ec29ed5b | 6829 | */ |
f3714ef4 | 6830 | em->len = min(hole_len, delalloc_start - hole_start); |
ec29ed5b CM |
6831 | em->start = hole_start; |
6832 | em->orig_start = hole_start; | |
6833 | /* | |
f3714ef4 NB |
6834 | * Don't adjust block start at all, it is fixed at |
6835 | * EXTENT_MAP_HOLE | |
ec29ed5b CM |
6836 | */ |
6837 | em->block_start = hole_em->block_start; | |
6838 | em->block_len = hole_len; | |
f9e4fb53 LB |
6839 | if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags)) |
6840 | set_bit(EXTENT_FLAG_PREALLOC, &em->flags); | |
ec29ed5b | 6841 | } else { |
f3714ef4 NB |
6842 | /* |
6843 | * Hole is out of passed range or it starts after | |
6844 | * delalloc range | |
6845 | */ | |
6846 | em->start = delalloc_start; | |
6847 | em->len = delalloc_len; | |
6848 | em->orig_start = delalloc_start; | |
ec29ed5b | 6849 | em->block_start = EXTENT_MAP_DELALLOC; |
f3714ef4 | 6850 | em->block_len = delalloc_len; |
ec29ed5b | 6851 | } |
bf8d32b9 | 6852 | } else { |
ec29ed5b CM |
6853 | return hole_em; |
6854 | } | |
6855 | out: | |
6856 | ||
6857 | free_extent_map(hole_em); | |
6858 | if (err) { | |
6859 | free_extent_map(em); | |
6860 | return ERR_PTR(err); | |
6861 | } | |
6862 | return em; | |
6863 | } | |
6864 | ||
64f54188 | 6865 | static struct extent_map *btrfs_create_dio_extent(struct btrfs_inode *inode, |
5f9a8a51 FM |
6866 | const u64 start, |
6867 | const u64 len, | |
6868 | const u64 orig_start, | |
6869 | const u64 block_start, | |
6870 | const u64 block_len, | |
6871 | const u64 orig_block_len, | |
6872 | const u64 ram_bytes, | |
6873 | const int type) | |
6874 | { | |
6875 | struct extent_map *em = NULL; | |
6876 | int ret; | |
6877 | ||
5f9a8a51 | 6878 | if (type != BTRFS_ORDERED_NOCOW) { |
64f54188 NB |
6879 | em = create_io_em(inode, start, len, orig_start, block_start, |
6880 | block_len, orig_block_len, ram_bytes, | |
6f9994db LB |
6881 | BTRFS_COMPRESS_NONE, /* compress_type */ |
6882 | type); | |
5f9a8a51 FM |
6883 | if (IS_ERR(em)) |
6884 | goto out; | |
6885 | } | |
64f54188 NB |
6886 | ret = btrfs_add_ordered_extent_dio(inode, start, block_start, len, |
6887 | block_len, type); | |
5f9a8a51 FM |
6888 | if (ret) { |
6889 | if (em) { | |
6890 | free_extent_map(em); | |
64f54188 | 6891 | btrfs_drop_extent_cache(inode, start, start + len - 1, 0); |
5f9a8a51 FM |
6892 | } |
6893 | em = ERR_PTR(ret); | |
6894 | } | |
6895 | out: | |
5f9a8a51 FM |
6896 | |
6897 | return em; | |
6898 | } | |
6899 | ||
9fc6f911 | 6900 | static struct extent_map *btrfs_new_extent_direct(struct btrfs_inode *inode, |
4b46fce2 JB |
6901 | u64 start, u64 len) |
6902 | { | |
9fc6f911 NB |
6903 | struct btrfs_root *root = inode->root; |
6904 | struct btrfs_fs_info *fs_info = root->fs_info; | |
70c8a91c | 6905 | struct extent_map *em; |
4b46fce2 JB |
6906 | struct btrfs_key ins; |
6907 | u64 alloc_hint; | |
6908 | int ret; | |
4b46fce2 | 6909 | |
9fc6f911 | 6910 | alloc_hint = get_extent_allocation_hint(inode, start, len); |
0b246afa | 6911 | ret = btrfs_reserve_extent(root, len, len, fs_info->sectorsize, |
da17066c | 6912 | 0, alloc_hint, &ins, 1, 1); |
00361589 JB |
6913 | if (ret) |
6914 | return ERR_PTR(ret); | |
4b46fce2 | 6915 | |
9fc6f911 | 6916 | em = btrfs_create_dio_extent(inode, start, ins.offset, start, |
5f9a8a51 | 6917 | ins.objectid, ins.offset, ins.offset, |
6288d6ea | 6918 | ins.offset, BTRFS_ORDERED_REGULAR); |
0b246afa | 6919 | btrfs_dec_block_group_reservations(fs_info, ins.objectid); |
5f9a8a51 | 6920 | if (IS_ERR(em)) |
9fc6f911 NB |
6921 | btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, |
6922 | 1); | |
de0ee0ed | 6923 | |
4b46fce2 JB |
6924 | return em; |
6925 | } | |
6926 | ||
46bfbb5c | 6927 | /* |
e4ecaf90 QW |
6928 | * Check if we can do nocow write into the range [@offset, @offset + @len) |
6929 | * | |
6930 | * @offset: File offset | |
6931 | * @len: The length to write, will be updated to the nocow writeable | |
6932 | * range | |
6933 | * @orig_start: (optional) Return the original file offset of the file extent | |
6934 | * @orig_len: (optional) Return the original on-disk length of the file extent | |
6935 | * @ram_bytes: (optional) Return the ram_bytes of the file extent | |
6936 | * | |
6937 | * This function will flush ordered extents in the range to ensure proper | |
6938 | * nocow checks for (nowait == false) case. | |
6939 | * | |
6940 | * Return: | |
6941 | * >0 and update @len if we can do nocow write | |
6942 | * 0 if we can't do nocow write | |
6943 | * <0 if error happened | |
6944 | * | |
6945 | * NOTE: This only checks the file extents, caller is responsible to wait for | |
6946 | * any ordered extents. | |
46bfbb5c | 6947 | */ |
00361589 | 6948 | noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len, |
7ee9e440 JB |
6949 | u64 *orig_start, u64 *orig_block_len, |
6950 | u64 *ram_bytes) | |
46bfbb5c | 6951 | { |
2ff7e61e | 6952 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
46bfbb5c CM |
6953 | struct btrfs_path *path; |
6954 | int ret; | |
6955 | struct extent_buffer *leaf; | |
6956 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
7b2b7085 | 6957 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
46bfbb5c CM |
6958 | struct btrfs_file_extent_item *fi; |
6959 | struct btrfs_key key; | |
6960 | u64 disk_bytenr; | |
6961 | u64 backref_offset; | |
6962 | u64 extent_end; | |
6963 | u64 num_bytes; | |
6964 | int slot; | |
6965 | int found_type; | |
7ee9e440 | 6966 | bool nocow = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW); |
e77751aa | 6967 | |
46bfbb5c CM |
6968 | path = btrfs_alloc_path(); |
6969 | if (!path) | |
6970 | return -ENOMEM; | |
6971 | ||
f85b7379 DS |
6972 | ret = btrfs_lookup_file_extent(NULL, root, path, |
6973 | btrfs_ino(BTRFS_I(inode)), offset, 0); | |
46bfbb5c CM |
6974 | if (ret < 0) |
6975 | goto out; | |
6976 | ||
6977 | slot = path->slots[0]; | |
6978 | if (ret == 1) { | |
6979 | if (slot == 0) { | |
6980 | /* can't find the item, must cow */ | |
6981 | ret = 0; | |
6982 | goto out; | |
6983 | } | |
6984 | slot--; | |
6985 | } | |
6986 | ret = 0; | |
6987 | leaf = path->nodes[0]; | |
6988 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
4a0cc7ca | 6989 | if (key.objectid != btrfs_ino(BTRFS_I(inode)) || |
46bfbb5c CM |
6990 | key.type != BTRFS_EXTENT_DATA_KEY) { |
6991 | /* not our file or wrong item type, must cow */ | |
6992 | goto out; | |
6993 | } | |
6994 | ||
6995 | if (key.offset > offset) { | |
6996 | /* Wrong offset, must cow */ | |
6997 | goto out; | |
6998 | } | |
6999 | ||
7000 | fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item); | |
7001 | found_type = btrfs_file_extent_type(leaf, fi); | |
7002 | if (found_type != BTRFS_FILE_EXTENT_REG && | |
7003 | found_type != BTRFS_FILE_EXTENT_PREALLOC) { | |
7004 | /* not a regular extent, must cow */ | |
7005 | goto out; | |
7006 | } | |
7ee9e440 JB |
7007 | |
7008 | if (!nocow && found_type == BTRFS_FILE_EXTENT_REG) | |
7009 | goto out; | |
7010 | ||
e77751aa MX |
7011 | extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi); |
7012 | if (extent_end <= offset) | |
7013 | goto out; | |
7014 | ||
46bfbb5c | 7015 | disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); |
7ee9e440 JB |
7016 | if (disk_bytenr == 0) |
7017 | goto out; | |
7018 | ||
7019 | if (btrfs_file_extent_compression(leaf, fi) || | |
7020 | btrfs_file_extent_encryption(leaf, fi) || | |
7021 | btrfs_file_extent_other_encoding(leaf, fi)) | |
7022 | goto out; | |
7023 | ||
78d4295b EL |
7024 | /* |
7025 | * Do the same check as in btrfs_cross_ref_exist but without the | |
7026 | * unnecessary search. | |
7027 | */ | |
7028 | if (btrfs_file_extent_generation(leaf, fi) <= | |
7029 | btrfs_root_last_snapshot(&root->root_item)) | |
7030 | goto out; | |
7031 | ||
46bfbb5c CM |
7032 | backref_offset = btrfs_file_extent_offset(leaf, fi); |
7033 | ||
7ee9e440 JB |
7034 | if (orig_start) { |
7035 | *orig_start = key.offset - backref_offset; | |
7036 | *orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi); | |
7037 | *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi); | |
7038 | } | |
eb384b55 | 7039 | |
2ff7e61e | 7040 | if (btrfs_extent_readonly(fs_info, disk_bytenr)) |
46bfbb5c | 7041 | goto out; |
7b2b7085 MX |
7042 | |
7043 | num_bytes = min(offset + *len, extent_end) - offset; | |
7044 | if (!nocow && found_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
7045 | u64 range_end; | |
7046 | ||
da17066c JM |
7047 | range_end = round_up(offset + num_bytes, |
7048 | root->fs_info->sectorsize) - 1; | |
7b2b7085 MX |
7049 | ret = test_range_bit(io_tree, offset, range_end, |
7050 | EXTENT_DELALLOC, 0, NULL); | |
7051 | if (ret) { | |
7052 | ret = -EAGAIN; | |
7053 | goto out; | |
7054 | } | |
7055 | } | |
7056 | ||
1bda19eb | 7057 | btrfs_release_path(path); |
46bfbb5c CM |
7058 | |
7059 | /* | |
7060 | * look for other files referencing this extent, if we | |
7061 | * find any we must cow | |
7062 | */ | |
00361589 | 7063 | |
e4c3b2dc | 7064 | ret = btrfs_cross_ref_exist(root, btrfs_ino(BTRFS_I(inode)), |
00361589 | 7065 | key.offset - backref_offset, disk_bytenr); |
00361589 JB |
7066 | if (ret) { |
7067 | ret = 0; | |
7068 | goto out; | |
7069 | } | |
46bfbb5c CM |
7070 | |
7071 | /* | |
7072 | * adjust disk_bytenr and num_bytes to cover just the bytes | |
7073 | * in this extent we are about to write. If there | |
7074 | * are any csums in that range we have to cow in order | |
7075 | * to keep the csums correct | |
7076 | */ | |
7077 | disk_bytenr += backref_offset; | |
7078 | disk_bytenr += offset - key.offset; | |
2ff7e61e JM |
7079 | if (csum_exist_in_range(fs_info, disk_bytenr, num_bytes)) |
7080 | goto out; | |
46bfbb5c CM |
7081 | /* |
7082 | * all of the above have passed, it is safe to overwrite this extent | |
7083 | * without cow | |
7084 | */ | |
eb384b55 | 7085 | *len = num_bytes; |
46bfbb5c CM |
7086 | ret = 1; |
7087 | out: | |
7088 | btrfs_free_path(path); | |
7089 | return ret; | |
7090 | } | |
7091 | ||
eb838e73 | 7092 | static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend, |
55e20bd1 | 7093 | struct extent_state **cached_state, int writing) |
eb838e73 JB |
7094 | { |
7095 | struct btrfs_ordered_extent *ordered; | |
7096 | int ret = 0; | |
7097 | ||
7098 | while (1) { | |
7099 | lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend, | |
ff13db41 | 7100 | cached_state); |
eb838e73 JB |
7101 | /* |
7102 | * We're concerned with the entire range that we're going to be | |
01327610 | 7103 | * doing DIO to, so we need to make sure there's no ordered |
eb838e73 JB |
7104 | * extents in this range. |
7105 | */ | |
a776c6fa | 7106 | ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), lockstart, |
eb838e73 JB |
7107 | lockend - lockstart + 1); |
7108 | ||
7109 | /* | |
7110 | * We need to make sure there are no buffered pages in this | |
7111 | * range either, we could have raced between the invalidate in | |
7112 | * generic_file_direct_write and locking the extent. The | |
7113 | * invalidate needs to happen so that reads after a write do not | |
7114 | * get stale data. | |
7115 | */ | |
fc4adbff | 7116 | if (!ordered && |
051c98eb DS |
7117 | (!writing || !filemap_range_has_page(inode->i_mapping, |
7118 | lockstart, lockend))) | |
eb838e73 JB |
7119 | break; |
7120 | ||
7121 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend, | |
e43bbe5e | 7122 | cached_state); |
eb838e73 JB |
7123 | |
7124 | if (ordered) { | |
ade77029 FM |
7125 | /* |
7126 | * If we are doing a DIO read and the ordered extent we | |
7127 | * found is for a buffered write, we can not wait for it | |
7128 | * to complete and retry, because if we do so we can | |
7129 | * deadlock with concurrent buffered writes on page | |
7130 | * locks. This happens only if our DIO read covers more | |
7131 | * than one extent map, if at this point has already | |
7132 | * created an ordered extent for a previous extent map | |
7133 | * and locked its range in the inode's io tree, and a | |
7134 | * concurrent write against that previous extent map's | |
7135 | * range and this range started (we unlock the ranges | |
7136 | * in the io tree only when the bios complete and | |
7137 | * buffered writes always lock pages before attempting | |
7138 | * to lock range in the io tree). | |
7139 | */ | |
7140 | if (writing || | |
7141 | test_bit(BTRFS_ORDERED_DIRECT, &ordered->flags)) | |
7142 | btrfs_start_ordered_extent(inode, ordered, 1); | |
7143 | else | |
7144 | ret = -ENOTBLK; | |
eb838e73 JB |
7145 | btrfs_put_ordered_extent(ordered); |
7146 | } else { | |
eb838e73 | 7147 | /* |
b850ae14 FM |
7148 | * We could trigger writeback for this range (and wait |
7149 | * for it to complete) and then invalidate the pages for | |
7150 | * this range (through invalidate_inode_pages2_range()), | |
7151 | * but that can lead us to a deadlock with a concurrent | |
ba206a02 | 7152 | * call to readahead (a buffered read or a defrag call |
b850ae14 FM |
7153 | * triggered a readahead) on a page lock due to an |
7154 | * ordered dio extent we created before but did not have | |
7155 | * yet a corresponding bio submitted (whence it can not | |
ba206a02 | 7156 | * complete), which makes readahead wait for that |
b850ae14 FM |
7157 | * ordered extent to complete while holding a lock on |
7158 | * that page. | |
eb838e73 | 7159 | */ |
b850ae14 | 7160 | ret = -ENOTBLK; |
eb838e73 JB |
7161 | } |
7162 | ||
ade77029 FM |
7163 | if (ret) |
7164 | break; | |
7165 | ||
eb838e73 JB |
7166 | cond_resched(); |
7167 | } | |
7168 | ||
7169 | return ret; | |
7170 | } | |
7171 | ||
6f9994db | 7172 | /* The callers of this must take lock_extent() */ |
4b67c11d NB |
7173 | static struct extent_map *create_io_em(struct btrfs_inode *inode, u64 start, |
7174 | u64 len, u64 orig_start, u64 block_start, | |
6f9994db LB |
7175 | u64 block_len, u64 orig_block_len, |
7176 | u64 ram_bytes, int compress_type, | |
7177 | int type) | |
69ffb543 JB |
7178 | { |
7179 | struct extent_map_tree *em_tree; | |
7180 | struct extent_map *em; | |
69ffb543 JB |
7181 | int ret; |
7182 | ||
6f9994db LB |
7183 | ASSERT(type == BTRFS_ORDERED_PREALLOC || |
7184 | type == BTRFS_ORDERED_COMPRESSED || | |
7185 | type == BTRFS_ORDERED_NOCOW || | |
1af4a0aa | 7186 | type == BTRFS_ORDERED_REGULAR); |
6f9994db | 7187 | |
4b67c11d | 7188 | em_tree = &inode->extent_tree; |
69ffb543 JB |
7189 | em = alloc_extent_map(); |
7190 | if (!em) | |
7191 | return ERR_PTR(-ENOMEM); | |
7192 | ||
7193 | em->start = start; | |
7194 | em->orig_start = orig_start; | |
7195 | em->len = len; | |
7196 | em->block_len = block_len; | |
7197 | em->block_start = block_start; | |
b4939680 | 7198 | em->orig_block_len = orig_block_len; |
cc95bef6 | 7199 | em->ram_bytes = ram_bytes; |
70c8a91c | 7200 | em->generation = -1; |
69ffb543 | 7201 | set_bit(EXTENT_FLAG_PINNED, &em->flags); |
1af4a0aa | 7202 | if (type == BTRFS_ORDERED_PREALLOC) { |
b11e234d | 7203 | set_bit(EXTENT_FLAG_FILLING, &em->flags); |
1af4a0aa | 7204 | } else if (type == BTRFS_ORDERED_COMPRESSED) { |
6f9994db LB |
7205 | set_bit(EXTENT_FLAG_COMPRESSED, &em->flags); |
7206 | em->compress_type = compress_type; | |
7207 | } | |
69ffb543 JB |
7208 | |
7209 | do { | |
4b67c11d NB |
7210 | btrfs_drop_extent_cache(inode, em->start, |
7211 | em->start + em->len - 1, 0); | |
69ffb543 | 7212 | write_lock(&em_tree->lock); |
09a2a8f9 | 7213 | ret = add_extent_mapping(em_tree, em, 1); |
69ffb543 | 7214 | write_unlock(&em_tree->lock); |
6f9994db LB |
7215 | /* |
7216 | * The caller has taken lock_extent(), who could race with us | |
7217 | * to add em? | |
7218 | */ | |
69ffb543 JB |
7219 | } while (ret == -EEXIST); |
7220 | ||
7221 | if (ret) { | |
7222 | free_extent_map(em); | |
7223 | return ERR_PTR(ret); | |
7224 | } | |
7225 | ||
6f9994db | 7226 | /* em got 2 refs now, callers needs to do free_extent_map once. */ |
69ffb543 JB |
7227 | return em; |
7228 | } | |
7229 | ||
1c8d0175 | 7230 | |
55e20bd1 DS |
7231 | static int btrfs_get_blocks_direct_read(struct extent_map *em, |
7232 | struct buffer_head *bh_result, | |
7233 | struct inode *inode, | |
7234 | u64 start, u64 len) | |
7235 | { | |
7236 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); | |
7237 | ||
7238 | if (em->block_start == EXTENT_MAP_HOLE || | |
7239 | test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) | |
7240 | return -ENOENT; | |
7241 | ||
7242 | len = min(len, em->len - (start - em->start)); | |
7243 | ||
7244 | bh_result->b_blocknr = (em->block_start + (start - em->start)) >> | |
7245 | inode->i_blkbits; | |
7246 | bh_result->b_size = len; | |
7247 | bh_result->b_bdev = fs_info->fs_devices->latest_bdev; | |
7248 | set_buffer_mapped(bh_result); | |
7249 | ||
7250 | return 0; | |
7251 | } | |
7252 | ||
c5794e51 | 7253 | static int btrfs_get_blocks_direct_write(struct extent_map **map, |
55e20bd1 | 7254 | struct buffer_head *bh_result, |
c5794e51 NB |
7255 | struct inode *inode, |
7256 | struct btrfs_dio_data *dio_data, | |
7257 | u64 start, u64 len) | |
7258 | { | |
7259 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); | |
7260 | struct extent_map *em = *map; | |
7261 | int ret = 0; | |
7262 | ||
7263 | /* | |
7264 | * We don't allocate a new extent in the following cases | |
7265 | * | |
7266 | * 1) The inode is marked as NODATACOW. In this case we'll just use the | |
7267 | * existing extent. | |
7268 | * 2) The extent is marked as PREALLOC. We're good to go here and can | |
7269 | * just use the extent. | |
7270 | * | |
7271 | */ | |
7272 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) || | |
7273 | ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) && | |
7274 | em->block_start != EXTENT_MAP_HOLE)) { | |
7275 | int type; | |
7276 | u64 block_start, orig_start, orig_block_len, ram_bytes; | |
7277 | ||
7278 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) | |
7279 | type = BTRFS_ORDERED_PREALLOC; | |
7280 | else | |
7281 | type = BTRFS_ORDERED_NOCOW; | |
7282 | len = min(len, em->len - (start - em->start)); | |
7283 | block_start = em->block_start + (start - em->start); | |
7284 | ||
7285 | if (can_nocow_extent(inode, start, &len, &orig_start, | |
7286 | &orig_block_len, &ram_bytes) == 1 && | |
7287 | btrfs_inc_nocow_writers(fs_info, block_start)) { | |
7288 | struct extent_map *em2; | |
7289 | ||
64f54188 | 7290 | em2 = btrfs_create_dio_extent(BTRFS_I(inode), start, len, |
c5794e51 NB |
7291 | orig_start, block_start, |
7292 | len, orig_block_len, | |
7293 | ram_bytes, type); | |
7294 | btrfs_dec_nocow_writers(fs_info, block_start); | |
7295 | if (type == BTRFS_ORDERED_PREALLOC) { | |
7296 | free_extent_map(em); | |
7297 | *map = em = em2; | |
7298 | } | |
7299 | ||
7300 | if (em2 && IS_ERR(em2)) { | |
7301 | ret = PTR_ERR(em2); | |
7302 | goto out; | |
7303 | } | |
7304 | /* | |
7305 | * For inode marked NODATACOW or extent marked PREALLOC, | |
7306 | * use the existing or preallocated extent, so does not | |
7307 | * need to adjust btrfs_space_info's bytes_may_use. | |
7308 | */ | |
9db5d510 | 7309 | btrfs_free_reserved_data_space_noquota(fs_info, len); |
c5794e51 NB |
7310 | goto skip_cow; |
7311 | } | |
7312 | } | |
7313 | ||
7314 | /* this will cow the extent */ | |
55e20bd1 | 7315 | len = bh_result->b_size; |
c5794e51 | 7316 | free_extent_map(em); |
9fc6f911 | 7317 | *map = em = btrfs_new_extent_direct(BTRFS_I(inode), start, len); |
c5794e51 NB |
7318 | if (IS_ERR(em)) { |
7319 | ret = PTR_ERR(em); | |
7320 | goto out; | |
7321 | } | |
7322 | ||
7323 | len = min(len, em->len - (start - em->start)); | |
7324 | ||
7325 | skip_cow: | |
55e20bd1 DS |
7326 | bh_result->b_blocknr = (em->block_start + (start - em->start)) >> |
7327 | inode->i_blkbits; | |
7328 | bh_result->b_size = len; | |
7329 | bh_result->b_bdev = fs_info->fs_devices->latest_bdev; | |
7330 | set_buffer_mapped(bh_result); | |
7331 | ||
7332 | if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) | |
7333 | set_buffer_new(bh_result); | |
7334 | ||
c5794e51 NB |
7335 | /* |
7336 | * Need to update the i_size under the extent lock so buffered | |
7337 | * readers will get the updated i_size when we unlock. | |
7338 | */ | |
55e20bd1 | 7339 | if (!dio_data->overwrite && start + len > i_size_read(inode)) |
c5794e51 NB |
7340 | i_size_write(inode, start + len); |
7341 | ||
55e20bd1 | 7342 | WARN_ON(dio_data->reserve < len); |
c5794e51 | 7343 | dio_data->reserve -= len; |
55e20bd1 DS |
7344 | dio_data->unsubmitted_oe_range_end = start + len; |
7345 | current->journal_info = dio_data; | |
c5794e51 NB |
7346 | out: |
7347 | return ret; | |
7348 | } | |
7349 | ||
55e20bd1 DS |
7350 | static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock, |
7351 | struct buffer_head *bh_result, int create) | |
4b46fce2 | 7352 | { |
0b246afa | 7353 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
4b46fce2 | 7354 | struct extent_map *em; |
eb838e73 | 7355 | struct extent_state *cached_state = NULL; |
50745b0a | 7356 | struct btrfs_dio_data *dio_data = NULL; |
55e20bd1 | 7357 | u64 start = iblock << inode->i_blkbits; |
eb838e73 | 7358 | u64 lockstart, lockend; |
55e20bd1 | 7359 | u64 len = bh_result->b_size; |
0934856d | 7360 | int ret = 0; |
eb838e73 | 7361 | |
55e20bd1 | 7362 | if (!create) |
0b246afa | 7363 | len = min_t(u64, len, fs_info->sectorsize); |
eb838e73 | 7364 | |
c329861d JB |
7365 | lockstart = start; |
7366 | lockend = start + len - 1; | |
7367 | ||
55e20bd1 DS |
7368 | if (current->journal_info) { |
7369 | /* | |
7370 | * Need to pull our outstanding extents and set journal_info to NULL so | |
7371 | * that anything that needs to check if there's a transaction doesn't get | |
7372 | * confused. | |
7373 | */ | |
7374 | dio_data = current->journal_info; | |
7375 | current->journal_info = NULL; | |
e1cbbfa5 JB |
7376 | } |
7377 | ||
eb838e73 JB |
7378 | /* |
7379 | * If this errors out it's because we couldn't invalidate pagecache for | |
7380 | * this range and we need to fallback to buffered. | |
7381 | */ | |
55e20bd1 DS |
7382 | if (lock_extent_direct(inode, lockstart, lockend, &cached_state, |
7383 | create)) { | |
9c9464cc FM |
7384 | ret = -ENOTBLK; |
7385 | goto err; | |
7386 | } | |
eb838e73 | 7387 | |
39b07b5d | 7388 | em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len); |
eb838e73 JB |
7389 | if (IS_ERR(em)) { |
7390 | ret = PTR_ERR(em); | |
7391 | goto unlock_err; | |
7392 | } | |
4b46fce2 JB |
7393 | |
7394 | /* | |
7395 | * Ok for INLINE and COMPRESSED extents we need to fallback on buffered | |
7396 | * io. INLINE is special, and we could probably kludge it in here, but | |
7397 | * it's still buffered so for safety lets just fall back to the generic | |
7398 | * buffered path. | |
7399 | * | |
7400 | * For COMPRESSED we _have_ to read the entire extent in so we can | |
7401 | * decompress it, so there will be buffering required no matter what we | |
7402 | * do, so go ahead and fallback to buffered. | |
7403 | * | |
01327610 | 7404 | * We return -ENOTBLK because that's what makes DIO go ahead and go back |
4b46fce2 JB |
7405 | * to buffered IO. Don't blame me, this is the price we pay for using |
7406 | * the generic code. | |
7407 | */ | |
7408 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) || | |
7409 | em->block_start == EXTENT_MAP_INLINE) { | |
7410 | free_extent_map(em); | |
eb838e73 JB |
7411 | ret = -ENOTBLK; |
7412 | goto unlock_err; | |
4b46fce2 JB |
7413 | } |
7414 | ||
55e20bd1 DS |
7415 | if (create) { |
7416 | ret = btrfs_get_blocks_direct_write(&em, bh_result, inode, | |
7417 | dio_data, start, len); | |
c5794e51 NB |
7418 | if (ret < 0) |
7419 | goto unlock_err; | |
55e20bd1 DS |
7420 | |
7421 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, | |
7422 | lockend, &cached_state); | |
c5794e51 | 7423 | } else { |
55e20bd1 DS |
7424 | ret = btrfs_get_blocks_direct_read(em, bh_result, inode, |
7425 | start, len); | |
7426 | /* Can be negative only if we read from a hole */ | |
7427 | if (ret < 0) { | |
7428 | ret = 0; | |
7429 | free_extent_map(em); | |
7430 | goto unlock_err; | |
7431 | } | |
1c8d0175 NB |
7432 | /* |
7433 | * We need to unlock only the end area that we aren't using. | |
7434 | * The rest is going to be unlocked by the endio routine. | |
7435 | */ | |
55e20bd1 DS |
7436 | lockstart = start + bh_result->b_size; |
7437 | if (lockstart < lockend) { | |
7438 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, | |
7439 | lockstart, lockend, &cached_state); | |
7440 | } else { | |
7441 | free_extent_state(cached_state); | |
7442 | } | |
a43a67a2 | 7443 | } |
a43a67a2 | 7444 | |
4b46fce2 JB |
7445 | free_extent_map(em); |
7446 | ||
7447 | return 0; | |
eb838e73 JB |
7448 | |
7449 | unlock_err: | |
e182163d OS |
7450 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend, |
7451 | &cached_state); | |
9c9464cc | 7452 | err: |
55e20bd1 DS |
7453 | if (dio_data) |
7454 | current->journal_info = dio_data; | |
8b110e39 MX |
7455 | return ret; |
7456 | } | |
7457 | ||
769b4f24 | 7458 | static void btrfs_dio_private_put(struct btrfs_dio_private *dip) |
8b110e39 | 7459 | { |
769b4f24 OS |
7460 | /* |
7461 | * This implies a barrier so that stores to dio_bio->bi_status before | |
7462 | * this and loads of dio_bio->bi_status after this are fully ordered. | |
7463 | */ | |
7464 | if (!refcount_dec_and_test(&dip->refs)) | |
7465 | return; | |
8b110e39 | 7466 | |
769b4f24 | 7467 | if (bio_op(dip->dio_bio) == REQ_OP_WRITE) { |
b672b5c1 NB |
7468 | __endio_write_update_ordered(BTRFS_I(dip->inode), |
7469 | dip->logical_offset, | |
769b4f24 OS |
7470 | dip->bytes, |
7471 | !dip->dio_bio->bi_status); | |
7472 | } else { | |
7473 | unlock_extent(&BTRFS_I(dip->inode)->io_tree, | |
7474 | dip->logical_offset, | |
7475 | dip->logical_offset + dip->bytes - 1); | |
8b110e39 MX |
7476 | } |
7477 | ||
55e20bd1 | 7478 | dio_end_io(dip->dio_bio); |
769b4f24 | 7479 | kfree(dip); |
8b110e39 MX |
7480 | } |
7481 | ||
77d5d689 OS |
7482 | static blk_status_t submit_dio_repair_bio(struct inode *inode, struct bio *bio, |
7483 | int mirror_num, | |
7484 | unsigned long bio_flags) | |
8b110e39 | 7485 | { |
77d5d689 | 7486 | struct btrfs_dio_private *dip = bio->bi_private; |
2ff7e61e | 7487 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
58efbc9f | 7488 | blk_status_t ret; |
8b110e39 | 7489 | |
37226b21 | 7490 | BUG_ON(bio_op(bio) == REQ_OP_WRITE); |
8b110e39 | 7491 | |
5c047a69 | 7492 | ret = btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DATA); |
8b110e39 | 7493 | if (ret) |
ea057f6d | 7494 | return ret; |
8b110e39 | 7495 | |
77d5d689 | 7496 | refcount_inc(&dip->refs); |
08635bae | 7497 | ret = btrfs_map_bio(fs_info, bio, mirror_num); |
8b110e39 | 7498 | if (ret) |
fd9d6670 | 7499 | refcount_dec(&dip->refs); |
77d5d689 | 7500 | return ret; |
8b110e39 MX |
7501 | } |
7502 | ||
fd9d6670 OS |
7503 | static blk_status_t btrfs_check_read_dio_bio(struct inode *inode, |
7504 | struct btrfs_io_bio *io_bio, | |
7505 | const bool uptodate) | |
4b46fce2 | 7506 | { |
fd9d6670 OS |
7507 | struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info; |
7508 | const u32 sectorsize = fs_info->sectorsize; | |
7509 | struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree; | |
7510 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; | |
7511 | const bool csum = !(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM); | |
17347cec LB |
7512 | struct bio_vec bvec; |
7513 | struct bvec_iter iter; | |
fd9d6670 OS |
7514 | u64 start = io_bio->logical; |
7515 | int icsum = 0; | |
58efbc9f | 7516 | blk_status_t err = BLK_STS_OK; |
4b46fce2 | 7517 | |
fd9d6670 OS |
7518 | __bio_for_each_segment(bvec, &io_bio->bio, iter, io_bio->iter) { |
7519 | unsigned int i, nr_sectors, pgoff; | |
8b110e39 | 7520 | |
17347cec LB |
7521 | nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec.bv_len); |
7522 | pgoff = bvec.bv_offset; | |
fd9d6670 | 7523 | for (i = 0; i < nr_sectors; i++) { |
97bf5a55 | 7524 | ASSERT(pgoff < PAGE_SIZE); |
fd9d6670 OS |
7525 | if (uptodate && |
7526 | (!csum || !check_data_csum(inode, io_bio, icsum, | |
7527 | bvec.bv_page, pgoff, | |
7528 | start, sectorsize))) { | |
7529 | clean_io_failure(fs_info, failure_tree, io_tree, | |
7530 | start, bvec.bv_page, | |
7531 | btrfs_ino(BTRFS_I(inode)), | |
7532 | pgoff); | |
7533 | } else { | |
7534 | blk_status_t status; | |
7535 | ||
77d5d689 OS |
7536 | status = btrfs_submit_read_repair(inode, |
7537 | &io_bio->bio, | |
7538 | start - io_bio->logical, | |
fd9d6670 OS |
7539 | bvec.bv_page, pgoff, |
7540 | start, | |
7541 | start + sectorsize - 1, | |
77d5d689 OS |
7542 | io_bio->mirror_num, |
7543 | submit_dio_repair_bio); | |
fd9d6670 OS |
7544 | if (status) |
7545 | err = status; | |
7546 | } | |
7547 | start += sectorsize; | |
7548 | icsum++; | |
2dabb324 | 7549 | pgoff += sectorsize; |
2dabb324 | 7550 | } |
2c30c71b | 7551 | } |
c1dc0896 MX |
7552 | return err; |
7553 | } | |
7554 | ||
b672b5c1 | 7555 | static void __endio_write_update_ordered(struct btrfs_inode *inode, |
52427260 QW |
7556 | const u64 offset, const u64 bytes, |
7557 | const bool uptodate) | |
4b46fce2 | 7558 | { |
b672b5c1 | 7559 | struct btrfs_fs_info *fs_info = inode->root->fs_info; |
4b46fce2 | 7560 | struct btrfs_ordered_extent *ordered = NULL; |
52427260 | 7561 | struct btrfs_workqueue *wq; |
14543774 FM |
7562 | u64 ordered_offset = offset; |
7563 | u64 ordered_bytes = bytes; | |
67c003f9 | 7564 | u64 last_offset; |
4b46fce2 | 7565 | |
b672b5c1 | 7566 | if (btrfs_is_free_space_inode(inode)) |
52427260 | 7567 | wq = fs_info->endio_freespace_worker; |
a0cac0ec | 7568 | else |
52427260 | 7569 | wq = fs_info->endio_write_workers; |
52427260 | 7570 | |
b25f0d00 NB |
7571 | while (ordered_offset < offset + bytes) { |
7572 | last_offset = ordered_offset; | |
b672b5c1 | 7573 | if (btrfs_dec_test_first_ordered_pending(inode, &ordered, |
7095821e NB |
7574 | &ordered_offset, |
7575 | ordered_bytes, | |
7576 | uptodate)) { | |
a0cac0ec OS |
7577 | btrfs_init_work(&ordered->work, finish_ordered_fn, NULL, |
7578 | NULL); | |
b25f0d00 NB |
7579 | btrfs_queue_work(wq, &ordered->work); |
7580 | } | |
7581 | /* | |
7582 | * If btrfs_dec_test_ordered_pending does not find any ordered | |
7583 | * extent in the range, we can exit. | |
7584 | */ | |
7585 | if (ordered_offset == last_offset) | |
7586 | return; | |
7587 | /* | |
7588 | * Our bio might span multiple ordered extents. In this case | |
52042d8e | 7589 | * we keep going until we have accounted the whole dio. |
b25f0d00 NB |
7590 | */ |
7591 | if (ordered_offset < offset + bytes) { | |
7592 | ordered_bytes = offset + bytes - ordered_offset; | |
7593 | ordered = NULL; | |
7594 | } | |
163cf09c | 7595 | } |
14543774 FM |
7596 | } |
7597 | ||
d0ee3934 | 7598 | static blk_status_t btrfs_submit_bio_start_direct_io(void *private_data, |
d0779291 | 7599 | struct bio *bio, u64 offset) |
eaf25d93 | 7600 | { |
c6100a4b | 7601 | struct inode *inode = private_data; |
4e4cbee9 | 7602 | blk_status_t ret; |
bd242a08 | 7603 | ret = btrfs_csum_one_bio(BTRFS_I(inode), bio, offset, 1); |
79787eaa | 7604 | BUG_ON(ret); /* -ENOMEM */ |
eaf25d93 CM |
7605 | return 0; |
7606 | } | |
7607 | ||
4246a0b6 | 7608 | static void btrfs_end_dio_bio(struct bio *bio) |
e65e1535 MX |
7609 | { |
7610 | struct btrfs_dio_private *dip = bio->bi_private; | |
4e4cbee9 | 7611 | blk_status_t err = bio->bi_status; |
e65e1535 | 7612 | |
8b110e39 MX |
7613 | if (err) |
7614 | btrfs_warn(BTRFS_I(dip->inode)->root->fs_info, | |
6296b960 | 7615 | "direct IO failed ino %llu rw %d,%u sector %#Lx len %u err no %d", |
f85b7379 DS |
7616 | btrfs_ino(BTRFS_I(dip->inode)), bio_op(bio), |
7617 | bio->bi_opf, | |
8b110e39 MX |
7618 | (unsigned long long)bio->bi_iter.bi_sector, |
7619 | bio->bi_iter.bi_size, err); | |
7620 | ||
769b4f24 OS |
7621 | if (bio_op(bio) == REQ_OP_READ) { |
7622 | err = btrfs_check_read_dio_bio(dip->inode, btrfs_io_bio(bio), | |
fd9d6670 | 7623 | !err); |
e65e1535 MX |
7624 | } |
7625 | ||
769b4f24 OS |
7626 | if (err) |
7627 | dip->dio_bio->bi_status = err; | |
e65e1535 | 7628 | |
e65e1535 | 7629 | bio_put(bio); |
769b4f24 | 7630 | btrfs_dio_private_put(dip); |
c1dc0896 MX |
7631 | } |
7632 | ||
d0ee3934 DS |
7633 | static inline blk_status_t btrfs_submit_dio_bio(struct bio *bio, |
7634 | struct inode *inode, u64 file_offset, int async_submit) | |
e65e1535 | 7635 | { |
0b246afa | 7636 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
facc8a22 | 7637 | struct btrfs_dio_private *dip = bio->bi_private; |
37226b21 | 7638 | bool write = bio_op(bio) == REQ_OP_WRITE; |
4e4cbee9 | 7639 | blk_status_t ret; |
e65e1535 | 7640 | |
4c274bc6 | 7641 | /* Check btrfs_submit_bio_hook() for rules about async submit. */ |
b812ce28 JB |
7642 | if (async_submit) |
7643 | async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers); | |
7644 | ||
5fd02043 | 7645 | if (!write) { |
0b246afa | 7646 | ret = btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DATA); |
5fd02043 JB |
7647 | if (ret) |
7648 | goto err; | |
7649 | } | |
e65e1535 | 7650 | |
e6961cac | 7651 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM) |
1ae39938 JB |
7652 | goto map; |
7653 | ||
7654 | if (write && async_submit) { | |
c6100a4b JB |
7655 | ret = btrfs_wq_submit_bio(fs_info, bio, 0, 0, |
7656 | file_offset, inode, | |
e288c080 | 7657 | btrfs_submit_bio_start_direct_io); |
e65e1535 | 7658 | goto err; |
1ae39938 JB |
7659 | } else if (write) { |
7660 | /* | |
7661 | * If we aren't doing async submit, calculate the csum of the | |
7662 | * bio now. | |
7663 | */ | |
bd242a08 | 7664 | ret = btrfs_csum_one_bio(BTRFS_I(inode), bio, file_offset, 1); |
1ae39938 JB |
7665 | if (ret) |
7666 | goto err; | |
23ea8e5a | 7667 | } else { |
85879573 OS |
7668 | u64 csum_offset; |
7669 | ||
7670 | csum_offset = file_offset - dip->logical_offset; | |
7671 | csum_offset >>= inode->i_sb->s_blocksize_bits; | |
7672 | csum_offset *= btrfs_super_csum_size(fs_info->super_copy); | |
7673 | btrfs_io_bio(bio)->csum = dip->csums + csum_offset; | |
c2db1073 | 7674 | } |
1ae39938 | 7675 | map: |
08635bae | 7676 | ret = btrfs_map_bio(fs_info, bio, 0); |
e65e1535 | 7677 | err: |
e65e1535 MX |
7678 | return ret; |
7679 | } | |
7680 | ||
c36cac28 OS |
7681 | /* |
7682 | * If this succeeds, the btrfs_dio_private is responsible for cleaning up locked | |
7683 | * or ordered extents whether or not we submit any bios. | |
7684 | */ | |
7685 | static struct btrfs_dio_private *btrfs_create_dio_private(struct bio *dio_bio, | |
7686 | struct inode *inode, | |
7687 | loff_t file_offset) | |
e65e1535 | 7688 | { |
c36cac28 | 7689 | const bool write = (bio_op(dio_bio) == REQ_OP_WRITE); |
85879573 OS |
7690 | const bool csum = !(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM); |
7691 | size_t dip_size; | |
c36cac28 | 7692 | struct btrfs_dio_private *dip; |
c36cac28 | 7693 | |
85879573 OS |
7694 | dip_size = sizeof(*dip); |
7695 | if (!write && csum) { | |
7696 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); | |
7697 | const u16 csum_size = btrfs_super_csum_size(fs_info->super_copy); | |
7698 | size_t nblocks; | |
7699 | ||
7700 | nblocks = dio_bio->bi_iter.bi_size >> inode->i_sb->s_blocksize_bits; | |
7701 | dip_size += csum_size * nblocks; | |
7702 | } | |
7703 | ||
7704 | dip = kzalloc(dip_size, GFP_NOFS); | |
c36cac28 OS |
7705 | if (!dip) |
7706 | return NULL; | |
7707 | ||
c36cac28 OS |
7708 | dip->inode = inode; |
7709 | dip->logical_offset = file_offset; | |
7710 | dip->bytes = dio_bio->bi_iter.bi_size; | |
7711 | dip->disk_bytenr = (u64)dio_bio->bi_iter.bi_sector << 9; | |
c36cac28 | 7712 | dip->dio_bio = dio_bio; |
e3b318d1 | 7713 | refcount_set(&dip->refs, 1); |
55e20bd1 DS |
7714 | |
7715 | if (write) { | |
7716 | struct btrfs_dio_data *dio_data = current->journal_info; | |
7717 | ||
7718 | /* | |
7719 | * Setting range start and end to the same value means that | |
7720 | * no cleanup will happen in btrfs_direct_IO | |
7721 | */ | |
7722 | dio_data->unsubmitted_oe_range_end = dip->logical_offset + | |
7723 | dip->bytes; | |
7724 | dio_data->unsubmitted_oe_range_start = | |
7725 | dio_data->unsubmitted_oe_range_end; | |
7726 | } | |
c36cac28 OS |
7727 | return dip; |
7728 | } | |
7729 | ||
55e20bd1 DS |
7730 | static void btrfs_submit_direct(struct bio *dio_bio, struct inode *inode, |
7731 | loff_t file_offset) | |
c36cac28 OS |
7732 | { |
7733 | const bool write = (bio_op(dio_bio) == REQ_OP_WRITE); | |
85879573 | 7734 | const bool csum = !(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM); |
0b246afa | 7735 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
769b4f24 OS |
7736 | const bool raid56 = (btrfs_data_alloc_profile(fs_info) & |
7737 | BTRFS_BLOCK_GROUP_RAID56_MASK); | |
c36cac28 | 7738 | struct btrfs_dio_private *dip; |
e65e1535 | 7739 | struct bio *bio; |
c36cac28 | 7740 | u64 start_sector; |
1ae39938 | 7741 | int async_submit = 0; |
725130ba LB |
7742 | u64 submit_len; |
7743 | int clone_offset = 0; | |
7744 | int clone_len; | |
5f4dc8fc | 7745 | int ret; |
58efbc9f | 7746 | blk_status_t status; |
89b798ad | 7747 | struct btrfs_io_geometry geom; |
e65e1535 | 7748 | |
c36cac28 OS |
7749 | dip = btrfs_create_dio_private(dio_bio, inode, file_offset); |
7750 | if (!dip) { | |
7751 | if (!write) { | |
7752 | unlock_extent(&BTRFS_I(inode)->io_tree, file_offset, | |
7753 | file_offset + dio_bio->bi_iter.bi_size - 1); | |
7754 | } | |
7755 | dio_bio->bi_status = BLK_STS_RESOURCE; | |
55e20bd1 DS |
7756 | dio_end_io(dio_bio); |
7757 | return; | |
c36cac28 | 7758 | } |
facc8a22 | 7759 | |
85879573 OS |
7760 | if (!write && csum) { |
7761 | /* | |
7762 | * Load the csums up front to reduce csum tree searches and | |
7763 | * contention when submitting bios. | |
7764 | */ | |
7765 | status = btrfs_lookup_bio_sums(inode, dio_bio, file_offset, | |
7766 | dip->csums); | |
7767 | if (status != BLK_STS_OK) | |
7768 | goto out_err; | |
02f57c7a JB |
7769 | } |
7770 | ||
769b4f24 OS |
7771 | start_sector = dio_bio->bi_iter.bi_sector; |
7772 | submit_len = dio_bio->bi_iter.bi_size; | |
53b381b3 | 7773 | |
3c91ee69 | 7774 | do { |
769b4f24 OS |
7775 | ret = btrfs_get_io_geometry(fs_info, btrfs_op(dio_bio), |
7776 | start_sector << 9, submit_len, | |
7777 | &geom); | |
7778 | if (ret) { | |
7779 | status = errno_to_blk_status(ret); | |
7780 | goto out_err; | |
7781 | } | |
7782 | ASSERT(geom.len <= INT_MAX); | |
7783 | ||
89b798ad | 7784 | clone_len = min_t(int, submit_len, geom.len); |
02f57c7a | 7785 | |
725130ba LB |
7786 | /* |
7787 | * This will never fail as it's passing GPF_NOFS and | |
7788 | * the allocation is backed by btrfs_bioset. | |
7789 | */ | |
769b4f24 | 7790 | bio = btrfs_bio_clone_partial(dio_bio, clone_offset, clone_len); |
725130ba LB |
7791 | bio->bi_private = dip; |
7792 | bio->bi_end_io = btrfs_end_dio_bio; | |
7793 | btrfs_io_bio(bio)->logical = file_offset; | |
7794 | ||
7795 | ASSERT(submit_len >= clone_len); | |
7796 | submit_len -= clone_len; | |
e65e1535 | 7797 | |
725130ba LB |
7798 | /* |
7799 | * Increase the count before we submit the bio so we know | |
7800 | * the end IO handler won't happen before we increase the | |
7801 | * count. Otherwise, the dip might get freed before we're | |
7802 | * done setting it up. | |
769b4f24 OS |
7803 | * |
7804 | * We transfer the initial reference to the last bio, so we | |
7805 | * don't need to increment the reference count for the last one. | |
725130ba | 7806 | */ |
769b4f24 OS |
7807 | if (submit_len > 0) { |
7808 | refcount_inc(&dip->refs); | |
7809 | /* | |
7810 | * If we are submitting more than one bio, submit them | |
7811 | * all asynchronously. The exception is RAID 5 or 6, as | |
7812 | * asynchronous checksums make it difficult to collect | |
7813 | * full stripe writes. | |
7814 | */ | |
7815 | if (!raid56) | |
7816 | async_submit = 1; | |
7817 | } | |
e65e1535 | 7818 | |
d0ee3934 | 7819 | status = btrfs_submit_dio_bio(bio, inode, file_offset, |
58efbc9f OS |
7820 | async_submit); |
7821 | if (status) { | |
725130ba | 7822 | bio_put(bio); |
769b4f24 OS |
7823 | if (submit_len > 0) |
7824 | refcount_dec(&dip->refs); | |
725130ba LB |
7825 | goto out_err; |
7826 | } | |
e65e1535 | 7827 | |
725130ba LB |
7828 | clone_offset += clone_len; |
7829 | start_sector += clone_len >> 9; | |
7830 | file_offset += clone_len; | |
3c91ee69 | 7831 | } while (submit_len > 0); |
55e20bd1 | 7832 | return; |
e65e1535 | 7833 | |
e65e1535 | 7834 | out_err: |
769b4f24 OS |
7835 | dip->dio_bio->bi_status = status; |
7836 | btrfs_dio_private_put(dip); | |
4b46fce2 JB |
7837 | } |
7838 | ||
f4c48b44 DS |
7839 | static ssize_t check_direct_IO(struct btrfs_fs_info *fs_info, |
7840 | const struct iov_iter *iter, loff_t offset) | |
7841 | { | |
7842 | int seg; | |
7843 | int i; | |
7844 | unsigned int blocksize_mask = fs_info->sectorsize - 1; | |
7845 | ssize_t retval = -EINVAL; | |
0934856d | 7846 | |
f4c48b44 DS |
7847 | if (offset & blocksize_mask) |
7848 | goto out; | |
7849 | ||
7850 | if (iov_iter_alignment(iter) & blocksize_mask) | |
7851 | goto out; | |
7852 | ||
7853 | /* If this is a write we don't need to check anymore */ | |
7854 | if (iov_iter_rw(iter) != READ || !iter_is_iovec(iter)) | |
7855 | return 0; | |
7856 | /* | |
7857 | * Check to make sure we don't have duplicate iov_base's in this | |
7858 | * iovec, if so return EINVAL, otherwise we'll get csum errors | |
7859 | * when reading back. | |
7860 | */ | |
7861 | for (seg = 0; seg < iter->nr_segs; seg++) { | |
7862 | for (i = seg + 1; i < iter->nr_segs; i++) { | |
7863 | if (iter->iov[seg].iov_base == iter->iov[i].iov_base) | |
7864 | goto out; | |
7865 | } | |
7866 | } | |
7867 | retval = 0; | |
7868 | out: | |
7869 | return retval; | |
7870 | } | |
7871 | ||
55e20bd1 | 7872 | static ssize_t btrfs_direct_IO(struct kiocb *iocb, struct iov_iter *iter) |
f4c48b44 DS |
7873 | { |
7874 | struct file *file = iocb->ki_filp; | |
7875 | struct inode *inode = file->f_mapping->host; | |
7876 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); | |
55e20bd1 | 7877 | struct btrfs_dio_data dio_data = { 0 }; |
f4c48b44 DS |
7878 | struct extent_changeset *data_reserved = NULL; |
7879 | loff_t offset = iocb->ki_pos; | |
7880 | size_t count = 0; | |
55e20bd1 DS |
7881 | int flags = 0; |
7882 | bool wakeup = true; | |
f4c48b44 DS |
7883 | bool relock = false; |
7884 | ssize_t ret; | |
7885 | ||
7886 | if (check_direct_IO(fs_info, iter, offset)) | |
7887 | return 0; | |
7888 | ||
55e20bd1 DS |
7889 | inode_dio_begin(inode); |
7890 | ||
7891 | /* | |
7892 | * The generic stuff only does filemap_write_and_wait_range, which | |
7893 | * isn't enough if we've written compressed pages to this area, so | |
7894 | * we need to flush the dirty pages again to make absolutely sure | |
7895 | * that any outstanding dirty pages are on disk. | |
7896 | */ | |
f4c48b44 | 7897 | count = iov_iter_count(iter); |
55e20bd1 DS |
7898 | if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT, |
7899 | &BTRFS_I(inode)->runtime_flags)) | |
7900 | filemap_fdatawrite_range(inode->i_mapping, offset, | |
7901 | offset + count - 1); | |
7902 | ||
f4c48b44 DS |
7903 | if (iov_iter_rw(iter) == WRITE) { |
7904 | /* | |
7905 | * If the write DIO is beyond the EOF, we need update | |
7906 | * the isize, but it is protected by i_mutex. So we can | |
7907 | * not unlock the i_mutex at this case. | |
7908 | */ | |
7909 | if (offset + count <= inode->i_size) { | |
55e20bd1 | 7910 | dio_data.overwrite = 1; |
f4c48b44 DS |
7911 | inode_unlock(inode); |
7912 | relock = true; | |
f4c48b44 | 7913 | } |
e5b7231e | 7914 | ret = btrfs_delalloc_reserve_space(BTRFS_I(inode), &data_reserved, |
55e20bd1 DS |
7915 | offset, count); |
7916 | if (ret) | |
7917 | goto out; | |
7918 | ||
7919 | /* | |
7920 | * We need to know how many extents we reserved so that we can | |
7921 | * do the accounting properly if we go over the number we | |
7922 | * originally calculated. Abuse current->journal_info for this. | |
7923 | */ | |
7924 | dio_data.reserve = round_up(count, | |
7925 | fs_info->sectorsize); | |
7926 | dio_data.unsubmitted_oe_range_start = (u64)offset; | |
7927 | dio_data.unsubmitted_oe_range_end = (u64)offset; | |
7928 | current->journal_info = &dio_data; | |
f4c48b44 | 7929 | down_read(&BTRFS_I(inode)->dio_sem); |
55e20bd1 DS |
7930 | } else if (test_bit(BTRFS_INODE_READDIO_NEED_LOCK, |
7931 | &BTRFS_I(inode)->runtime_flags)) { | |
7932 | inode_dio_end(inode); | |
7933 | flags = DIO_LOCKING | DIO_SKIP_HOLES; | |
7934 | wakeup = false; | |
f4c48b44 DS |
7935 | } |
7936 | ||
55e20bd1 DS |
7937 | ret = __blockdev_direct_IO(iocb, inode, |
7938 | fs_info->fs_devices->latest_bdev, | |
7939 | iter, btrfs_get_blocks_direct, NULL, | |
7940 | btrfs_submit_direct, flags); | |
f4c48b44 DS |
7941 | if (iov_iter_rw(iter) == WRITE) { |
7942 | up_read(&BTRFS_I(inode)->dio_sem); | |
55e20bd1 DS |
7943 | current->journal_info = NULL; |
7944 | if (ret < 0 && ret != -EIOCBQUEUED) { | |
7945 | if (dio_data.reserve) | |
86d52921 NB |
7946 | btrfs_delalloc_release_space(BTRFS_I(inode), |
7947 | data_reserved, offset, dio_data.reserve, | |
7948 | true); | |
55e20bd1 DS |
7949 | /* |
7950 | * On error we might have left some ordered extents | |
7951 | * without submitting corresponding bios for them, so | |
7952 | * cleanup them up to avoid other tasks getting them | |
7953 | * and waiting for them to complete forever. | |
7954 | */ | |
7955 | if (dio_data.unsubmitted_oe_range_start < | |
7956 | dio_data.unsubmitted_oe_range_end) | |
b672b5c1 | 7957 | __endio_write_update_ordered(BTRFS_I(inode), |
55e20bd1 DS |
7958 | dio_data.unsubmitted_oe_range_start, |
7959 | dio_data.unsubmitted_oe_range_end - | |
7960 | dio_data.unsubmitted_oe_range_start, | |
7961 | false); | |
7962 | } else if (ret >= 0 && (size_t)ret < count) | |
86d52921 | 7963 | btrfs_delalloc_release_space(BTRFS_I(inode), data_reserved, |
55e20bd1 DS |
7964 | offset, count - (size_t)ret, true); |
7965 | btrfs_delalloc_release_extents(BTRFS_I(inode), count); | |
f4c48b44 DS |
7966 | } |
7967 | out: | |
55e20bd1 DS |
7968 | if (wakeup) |
7969 | inode_dio_end(inode); | |
f4c48b44 DS |
7970 | if (relock) |
7971 | inode_lock(inode); | |
55e20bd1 | 7972 | |
f4c48b44 DS |
7973 | extent_changeset_free(data_reserved); |
7974 | return ret; | |
7975 | } | |
16432985 | 7976 | |
1506fcc8 | 7977 | static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, |
bab16e21 | 7978 | u64 start, u64 len) |
1506fcc8 | 7979 | { |
05dadc09 TI |
7980 | int ret; |
7981 | ||
45dd052e | 7982 | ret = fiemap_prep(inode, fieinfo, start, &len, 0); |
05dadc09 TI |
7983 | if (ret) |
7984 | return ret; | |
7985 | ||
2135fb9b | 7986 | return extent_fiemap(inode, fieinfo, start, len); |
1506fcc8 YS |
7987 | } |
7988 | ||
a52d9a80 | 7989 | int btrfs_readpage(struct file *file, struct page *page) |
9ebefb18 | 7990 | { |
71ad38b4 | 7991 | return extent_read_full_page(page, btrfs_get_extent, 0); |
9ebefb18 | 7992 | } |
1832a6d5 | 7993 | |
a52d9a80 | 7994 | static int btrfs_writepage(struct page *page, struct writeback_control *wbc) |
39279cc3 | 7995 | { |
be7bd730 JB |
7996 | struct inode *inode = page->mapping->host; |
7997 | int ret; | |
b888db2b CM |
7998 | |
7999 | if (current->flags & PF_MEMALLOC) { | |
8000 | redirty_page_for_writepage(wbc, page); | |
8001 | unlock_page(page); | |
8002 | return 0; | |
8003 | } | |
be7bd730 JB |
8004 | |
8005 | /* | |
8006 | * If we are under memory pressure we will call this directly from the | |
8007 | * VM, we need to make sure we have the inode referenced for the ordered | |
8008 | * extent. If not just return like we didn't do anything. | |
8009 | */ | |
8010 | if (!igrab(inode)) { | |
8011 | redirty_page_for_writepage(wbc, page); | |
8012 | return AOP_WRITEPAGE_ACTIVATE; | |
8013 | } | |
0a9b0e53 | 8014 | ret = extent_write_full_page(page, wbc); |
be7bd730 JB |
8015 | btrfs_add_delayed_iput(inode); |
8016 | return ret; | |
9ebefb18 CM |
8017 | } |
8018 | ||
48a3b636 ES |
8019 | static int btrfs_writepages(struct address_space *mapping, |
8020 | struct writeback_control *wbc) | |
b293f02e | 8021 | { |
8ae225a8 | 8022 | return extent_writepages(mapping, wbc); |
b293f02e CM |
8023 | } |
8024 | ||
ba206a02 | 8025 | static void btrfs_readahead(struct readahead_control *rac) |
3ab2fb5a | 8026 | { |
ba206a02 | 8027 | extent_readahead(rac); |
3ab2fb5a | 8028 | } |
2a3ff0ad | 8029 | |
e6dcd2dc | 8030 | static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags) |
9ebefb18 | 8031 | { |
477a30ba | 8032 | int ret = try_release_extent_mapping(page, gfp_flags); |
d1b89bc0 GJ |
8033 | if (ret == 1) |
8034 | detach_page_private(page); | |
a52d9a80 | 8035 | return ret; |
39279cc3 CM |
8036 | } |
8037 | ||
e6dcd2dc CM |
8038 | static int btrfs_releasepage(struct page *page, gfp_t gfp_flags) |
8039 | { | |
98509cfc CM |
8040 | if (PageWriteback(page) || PageDirty(page)) |
8041 | return 0; | |
3ba7ab22 | 8042 | return __btrfs_releasepage(page, gfp_flags); |
e6dcd2dc CM |
8043 | } |
8044 | ||
f8e66081 RG |
8045 | #ifdef CONFIG_MIGRATION |
8046 | static int btrfs_migratepage(struct address_space *mapping, | |
8047 | struct page *newpage, struct page *page, | |
8048 | enum migrate_mode mode) | |
8049 | { | |
8050 | int ret; | |
8051 | ||
8052 | ret = migrate_page_move_mapping(mapping, newpage, page, 0); | |
8053 | if (ret != MIGRATEPAGE_SUCCESS) | |
8054 | return ret; | |
8055 | ||
d1b89bc0 GJ |
8056 | if (page_has_private(page)) |
8057 | attach_page_private(newpage, detach_page_private(page)); | |
f8e66081 RG |
8058 | |
8059 | if (PagePrivate2(page)) { | |
8060 | ClearPagePrivate2(page); | |
8061 | SetPagePrivate2(newpage); | |
8062 | } | |
8063 | ||
8064 | if (mode != MIGRATE_SYNC_NO_COPY) | |
8065 | migrate_page_copy(newpage, page); | |
8066 | else | |
8067 | migrate_page_states(newpage, page); | |
8068 | return MIGRATEPAGE_SUCCESS; | |
8069 | } | |
8070 | #endif | |
8071 | ||
d47992f8 LC |
8072 | static void btrfs_invalidatepage(struct page *page, unsigned int offset, |
8073 | unsigned int length) | |
39279cc3 | 8074 | { |
5fd02043 | 8075 | struct inode *inode = page->mapping->host; |
d1310b2e | 8076 | struct extent_io_tree *tree; |
e6dcd2dc | 8077 | struct btrfs_ordered_extent *ordered; |
2ac55d41 | 8078 | struct extent_state *cached_state = NULL; |
e6dcd2dc | 8079 | u64 page_start = page_offset(page); |
09cbfeaf | 8080 | u64 page_end = page_start + PAGE_SIZE - 1; |
dbfdb6d1 CR |
8081 | u64 start; |
8082 | u64 end; | |
131e404a | 8083 | int inode_evicting = inode->i_state & I_FREEING; |
39279cc3 | 8084 | |
8b62b72b CM |
8085 | /* |
8086 | * we have the page locked, so new writeback can't start, | |
8087 | * and the dirty bit won't be cleared while we are here. | |
8088 | * | |
8089 | * Wait for IO on this page so that we can safely clear | |
8090 | * the PagePrivate2 bit and do ordered accounting | |
8091 | */ | |
e6dcd2dc | 8092 | wait_on_page_writeback(page); |
8b62b72b | 8093 | |
5fd02043 | 8094 | tree = &BTRFS_I(inode)->io_tree; |
e6dcd2dc CM |
8095 | if (offset) { |
8096 | btrfs_releasepage(page, GFP_NOFS); | |
8097 | return; | |
8098 | } | |
131e404a FDBM |
8099 | |
8100 | if (!inode_evicting) | |
ff13db41 | 8101 | lock_extent_bits(tree, page_start, page_end, &cached_state); |
dbfdb6d1 CR |
8102 | again: |
8103 | start = page_start; | |
a776c6fa | 8104 | ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), start, |
dbfdb6d1 | 8105 | page_end - start + 1); |
e6dcd2dc | 8106 | if (ordered) { |
bffe633e OS |
8107 | end = min(page_end, |
8108 | ordered->file_offset + ordered->num_bytes - 1); | |
eb84ae03 CM |
8109 | /* |
8110 | * IO on this page will never be started, so we need | |
8111 | * to account for any ordered extents now | |
8112 | */ | |
131e404a | 8113 | if (!inode_evicting) |
dbfdb6d1 | 8114 | clear_extent_bit(tree, start, end, |
e182163d | 8115 | EXTENT_DELALLOC | EXTENT_DELALLOC_NEW | |
131e404a | 8116 | EXTENT_LOCKED | EXTENT_DO_ACCOUNTING | |
ae0f1625 | 8117 | EXTENT_DEFRAG, 1, 0, &cached_state); |
8b62b72b CM |
8118 | /* |
8119 | * whoever cleared the private bit is responsible | |
8120 | * for the finish_ordered_io | |
8121 | */ | |
77cef2ec JB |
8122 | if (TestClearPagePrivate2(page)) { |
8123 | struct btrfs_ordered_inode_tree *tree; | |
8124 | u64 new_len; | |
8125 | ||
8126 | tree = &BTRFS_I(inode)->ordered_tree; | |
8127 | ||
8128 | spin_lock_irq(&tree->lock); | |
8129 | set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags); | |
dbfdb6d1 | 8130 | new_len = start - ordered->file_offset; |
77cef2ec JB |
8131 | if (new_len < ordered->truncated_len) |
8132 | ordered->truncated_len = new_len; | |
8133 | spin_unlock_irq(&tree->lock); | |
8134 | ||
8135 | if (btrfs_dec_test_ordered_pending(inode, &ordered, | |
dbfdb6d1 CR |
8136 | start, |
8137 | end - start + 1, 1)) | |
77cef2ec | 8138 | btrfs_finish_ordered_io(ordered); |
8b62b72b | 8139 | } |
e6dcd2dc | 8140 | btrfs_put_ordered_extent(ordered); |
131e404a FDBM |
8141 | if (!inode_evicting) { |
8142 | cached_state = NULL; | |
dbfdb6d1 | 8143 | lock_extent_bits(tree, start, end, |
131e404a FDBM |
8144 | &cached_state); |
8145 | } | |
dbfdb6d1 CR |
8146 | |
8147 | start = end + 1; | |
8148 | if (start < page_end) | |
8149 | goto again; | |
131e404a FDBM |
8150 | } |
8151 | ||
b9d0b389 QW |
8152 | /* |
8153 | * Qgroup reserved space handler | |
8154 | * Page here will be either | |
fa91e4aa QW |
8155 | * 1) Already written to disk or ordered extent already submitted |
8156 | * Then its QGROUP_RESERVED bit in io_tree is already cleaned. | |
8157 | * Qgroup will be handled by its qgroup_record then. | |
8158 | * btrfs_qgroup_free_data() call will do nothing here. | |
8159 | * | |
8160 | * 2) Not written to disk yet | |
8161 | * Then btrfs_qgroup_free_data() call will clear the QGROUP_RESERVED | |
8162 | * bit of its io_tree, and free the qgroup reserved data space. | |
8163 | * Since the IO will never happen for this page. | |
b9d0b389 | 8164 | */ |
8b8a979f | 8165 | btrfs_qgroup_free_data(BTRFS_I(inode), NULL, page_start, PAGE_SIZE); |
131e404a | 8166 | if (!inode_evicting) { |
e182163d | 8167 | clear_extent_bit(tree, page_start, page_end, EXTENT_LOCKED | |
a7e3b975 FM |
8168 | EXTENT_DELALLOC | EXTENT_DELALLOC_NEW | |
8169 | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 1, 1, | |
ae0f1625 | 8170 | &cached_state); |
131e404a FDBM |
8171 | |
8172 | __btrfs_releasepage(page, GFP_NOFS); | |
e6dcd2dc | 8173 | } |
e6dcd2dc | 8174 | |
4a096752 | 8175 | ClearPageChecked(page); |
d1b89bc0 | 8176 | detach_page_private(page); |
39279cc3 CM |
8177 | } |
8178 | ||
9ebefb18 CM |
8179 | /* |
8180 | * btrfs_page_mkwrite() is not allowed to change the file size as it gets | |
8181 | * called from a page fault handler when a page is first dirtied. Hence we must | |
8182 | * be careful to check for EOF conditions here. We set the page up correctly | |
8183 | * for a written page which means we get ENOSPC checking when writing into | |
8184 | * holes and correct delalloc and unwritten extent mapping on filesystems that | |
8185 | * support these features. | |
8186 | * | |
8187 | * We are not allowed to take the i_mutex here so we have to play games to | |
8188 | * protect against truncate races as the page could now be beyond EOF. Because | |
d1342aad OS |
8189 | * truncate_setsize() writes the inode size before removing pages, once we have |
8190 | * the page lock we can determine safely if the page is beyond EOF. If it is not | |
9ebefb18 CM |
8191 | * beyond EOF, then the page is guaranteed safe against truncation until we |
8192 | * unlock the page. | |
8193 | */ | |
a528a241 | 8194 | vm_fault_t btrfs_page_mkwrite(struct vm_fault *vmf) |
9ebefb18 | 8195 | { |
c2ec175c | 8196 | struct page *page = vmf->page; |
11bac800 | 8197 | struct inode *inode = file_inode(vmf->vma->vm_file); |
0b246afa | 8198 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
e6dcd2dc CM |
8199 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
8200 | struct btrfs_ordered_extent *ordered; | |
2ac55d41 | 8201 | struct extent_state *cached_state = NULL; |
364ecf36 | 8202 | struct extent_changeset *data_reserved = NULL; |
e6dcd2dc CM |
8203 | char *kaddr; |
8204 | unsigned long zero_start; | |
9ebefb18 | 8205 | loff_t size; |
a528a241 SJ |
8206 | vm_fault_t ret; |
8207 | int ret2; | |
9998eb70 | 8208 | int reserved = 0; |
d0b7da88 | 8209 | u64 reserved_space; |
a52d9a80 | 8210 | u64 page_start; |
e6dcd2dc | 8211 | u64 page_end; |
d0b7da88 CR |
8212 | u64 end; |
8213 | ||
09cbfeaf | 8214 | reserved_space = PAGE_SIZE; |
9ebefb18 | 8215 | |
b2b5ef5c | 8216 | sb_start_pagefault(inode->i_sb); |
df480633 | 8217 | page_start = page_offset(page); |
09cbfeaf | 8218 | page_end = page_start + PAGE_SIZE - 1; |
d0b7da88 | 8219 | end = page_end; |
df480633 | 8220 | |
d0b7da88 CR |
8221 | /* |
8222 | * Reserving delalloc space after obtaining the page lock can lead to | |
8223 | * deadlock. For example, if a dirty page is locked by this function | |
8224 | * and the call to btrfs_delalloc_reserve_space() ends up triggering | |
8225 | * dirty page write out, then the btrfs_writepage() function could | |
8226 | * end up waiting indefinitely to get a lock on the page currently | |
8227 | * being processed by btrfs_page_mkwrite() function. | |
8228 | */ | |
e5b7231e NB |
8229 | ret2 = btrfs_delalloc_reserve_space(BTRFS_I(inode), &data_reserved, |
8230 | page_start, reserved_space); | |
a528a241 SJ |
8231 | if (!ret2) { |
8232 | ret2 = file_update_time(vmf->vma->vm_file); | |
9998eb70 CM |
8233 | reserved = 1; |
8234 | } | |
a528a241 SJ |
8235 | if (ret2) { |
8236 | ret = vmf_error(ret2); | |
9998eb70 CM |
8237 | if (reserved) |
8238 | goto out; | |
8239 | goto out_noreserve; | |
56a76f82 | 8240 | } |
1832a6d5 | 8241 | |
56a76f82 | 8242 | ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */ |
e6dcd2dc | 8243 | again: |
9ebefb18 | 8244 | lock_page(page); |
9ebefb18 | 8245 | size = i_size_read(inode); |
a52d9a80 | 8246 | |
9ebefb18 | 8247 | if ((page->mapping != inode->i_mapping) || |
e6dcd2dc | 8248 | (page_start >= size)) { |
9ebefb18 CM |
8249 | /* page got truncated out from underneath us */ |
8250 | goto out_unlock; | |
8251 | } | |
e6dcd2dc CM |
8252 | wait_on_page_writeback(page); |
8253 | ||
ff13db41 | 8254 | lock_extent_bits(io_tree, page_start, page_end, &cached_state); |
e6dcd2dc CM |
8255 | set_page_extent_mapped(page); |
8256 | ||
eb84ae03 CM |
8257 | /* |
8258 | * we can't set the delalloc bits if there are pending ordered | |
8259 | * extents. Drop our locks and wait for them to finish | |
8260 | */ | |
a776c6fa NB |
8261 | ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), page_start, |
8262 | PAGE_SIZE); | |
e6dcd2dc | 8263 | if (ordered) { |
2ac55d41 | 8264 | unlock_extent_cached(io_tree, page_start, page_end, |
e43bbe5e | 8265 | &cached_state); |
e6dcd2dc | 8266 | unlock_page(page); |
eb84ae03 | 8267 | btrfs_start_ordered_extent(inode, ordered, 1); |
e6dcd2dc CM |
8268 | btrfs_put_ordered_extent(ordered); |
8269 | goto again; | |
8270 | } | |
8271 | ||
09cbfeaf | 8272 | if (page->index == ((size - 1) >> PAGE_SHIFT)) { |
da17066c | 8273 | reserved_space = round_up(size - page_start, |
0b246afa | 8274 | fs_info->sectorsize); |
09cbfeaf | 8275 | if (reserved_space < PAGE_SIZE) { |
d0b7da88 | 8276 | end = page_start + reserved_space - 1; |
86d52921 NB |
8277 | btrfs_delalloc_release_space(BTRFS_I(inode), |
8278 | data_reserved, page_start, | |
8279 | PAGE_SIZE - reserved_space, true); | |
d0b7da88 CR |
8280 | } |
8281 | } | |
8282 | ||
fbf19087 | 8283 | /* |
5416034f LB |
8284 | * page_mkwrite gets called when the page is firstly dirtied after it's |
8285 | * faulted in, but write(2) could also dirty a page and set delalloc | |
8286 | * bits, thus in this case for space account reason, we still need to | |
8287 | * clear any delalloc bits within this page range since we have to | |
8288 | * reserve data&meta space before lock_page() (see above comments). | |
fbf19087 | 8289 | */ |
d0b7da88 | 8290 | clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, end, |
e182163d OS |
8291 | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING | |
8292 | EXTENT_DEFRAG, 0, 0, &cached_state); | |
fbf19087 | 8293 | |
c2566f22 | 8294 | ret2 = btrfs_set_extent_delalloc(BTRFS_I(inode), page_start, end, 0, |
330a5827 | 8295 | &cached_state); |
a528a241 | 8296 | if (ret2) { |
2ac55d41 | 8297 | unlock_extent_cached(io_tree, page_start, page_end, |
e43bbe5e | 8298 | &cached_state); |
9ed74f2d JB |
8299 | ret = VM_FAULT_SIGBUS; |
8300 | goto out_unlock; | |
8301 | } | |
9ebefb18 CM |
8302 | |
8303 | /* page is wholly or partially inside EOF */ | |
09cbfeaf | 8304 | if (page_start + PAGE_SIZE > size) |
7073017a | 8305 | zero_start = offset_in_page(size); |
9ebefb18 | 8306 | else |
09cbfeaf | 8307 | zero_start = PAGE_SIZE; |
9ebefb18 | 8308 | |
09cbfeaf | 8309 | if (zero_start != PAGE_SIZE) { |
e6dcd2dc | 8310 | kaddr = kmap(page); |
09cbfeaf | 8311 | memset(kaddr + zero_start, 0, PAGE_SIZE - zero_start); |
e6dcd2dc CM |
8312 | flush_dcache_page(page); |
8313 | kunmap(page); | |
8314 | } | |
247e743c | 8315 | ClearPageChecked(page); |
e6dcd2dc | 8316 | set_page_dirty(page); |
50a9b214 | 8317 | SetPageUptodate(page); |
5a3f23d5 | 8318 | |
0b246afa | 8319 | BTRFS_I(inode)->last_trans = fs_info->generation; |
257c62e1 | 8320 | BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid; |
46d8bc34 | 8321 | BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit; |
257c62e1 | 8322 | |
e43bbe5e | 8323 | unlock_extent_cached(io_tree, page_start, page_end, &cached_state); |
9ebefb18 | 8324 | |
76de60ed YY |
8325 | btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE); |
8326 | sb_end_pagefault(inode->i_sb); | |
8327 | extent_changeset_free(data_reserved); | |
8328 | return VM_FAULT_LOCKED; | |
717beb96 CM |
8329 | |
8330 | out_unlock: | |
9ebefb18 | 8331 | unlock_page(page); |
1832a6d5 | 8332 | out: |
8702ba93 | 8333 | btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE); |
86d52921 | 8334 | btrfs_delalloc_release_space(BTRFS_I(inode), data_reserved, page_start, |
43b18595 | 8335 | reserved_space, (ret != 0)); |
9998eb70 | 8336 | out_noreserve: |
b2b5ef5c | 8337 | sb_end_pagefault(inode->i_sb); |
364ecf36 | 8338 | extent_changeset_free(data_reserved); |
9ebefb18 CM |
8339 | return ret; |
8340 | } | |
8341 | ||
213e8c55 | 8342 | static int btrfs_truncate(struct inode *inode, bool skip_writeback) |
39279cc3 | 8343 | { |
0b246afa | 8344 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
39279cc3 | 8345 | struct btrfs_root *root = BTRFS_I(inode)->root; |
fcb80c2a | 8346 | struct btrfs_block_rsv *rsv; |
ad7e1a74 | 8347 | int ret; |
39279cc3 | 8348 | struct btrfs_trans_handle *trans; |
0b246afa | 8349 | u64 mask = fs_info->sectorsize - 1; |
2bd36e7b | 8350 | u64 min_size = btrfs_calc_metadata_size(fs_info, 1); |
39279cc3 | 8351 | |
213e8c55 FM |
8352 | if (!skip_writeback) { |
8353 | ret = btrfs_wait_ordered_range(inode, inode->i_size & (~mask), | |
8354 | (u64)-1); | |
8355 | if (ret) | |
8356 | return ret; | |
8357 | } | |
39279cc3 | 8358 | |
fcb80c2a | 8359 | /* |
f7e9e8fc OS |
8360 | * Yes ladies and gentlemen, this is indeed ugly. We have a couple of |
8361 | * things going on here: | |
fcb80c2a | 8362 | * |
f7e9e8fc | 8363 | * 1) We need to reserve space to update our inode. |
fcb80c2a | 8364 | * |
f7e9e8fc | 8365 | * 2) We need to have something to cache all the space that is going to |
fcb80c2a JB |
8366 | * be free'd up by the truncate operation, but also have some slack |
8367 | * space reserved in case it uses space during the truncate (thank you | |
8368 | * very much snapshotting). | |
8369 | * | |
f7e9e8fc | 8370 | * And we need these to be separate. The fact is we can use a lot of |
fcb80c2a | 8371 | * space doing the truncate, and we have no earthly idea how much space |
01327610 | 8372 | * we will use, so we need the truncate reservation to be separate so it |
f7e9e8fc OS |
8373 | * doesn't end up using space reserved for updating the inode. We also |
8374 | * need to be able to stop the transaction and start a new one, which | |
8375 | * means we need to be able to update the inode several times, and we | |
8376 | * have no idea of knowing how many times that will be, so we can't just | |
8377 | * reserve 1 item for the entirety of the operation, so that has to be | |
8378 | * done separately as well. | |
fcb80c2a JB |
8379 | * |
8380 | * So that leaves us with | |
8381 | * | |
f7e9e8fc | 8382 | * 1) rsv - for the truncate reservation, which we will steal from the |
fcb80c2a | 8383 | * transaction reservation. |
f7e9e8fc | 8384 | * 2) fs_info->trans_block_rsv - this will have 1 items worth left for |
fcb80c2a JB |
8385 | * updating the inode. |
8386 | */ | |
2ff7e61e | 8387 | rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP); |
fcb80c2a JB |
8388 | if (!rsv) |
8389 | return -ENOMEM; | |
4a338542 | 8390 | rsv->size = min_size; |
ca7e70f5 | 8391 | rsv->failfast = 1; |
f0cd846e | 8392 | |
907cbceb | 8393 | /* |
07127184 | 8394 | * 1 for the truncate slack space |
907cbceb JB |
8395 | * 1 for updating the inode. |
8396 | */ | |
f3fe820c | 8397 | trans = btrfs_start_transaction(root, 2); |
fcb80c2a | 8398 | if (IS_ERR(trans)) { |
ad7e1a74 | 8399 | ret = PTR_ERR(trans); |
fcb80c2a JB |
8400 | goto out; |
8401 | } | |
f0cd846e | 8402 | |
907cbceb | 8403 | /* Migrate the slack space for the truncate to our reserve */ |
0b246afa | 8404 | ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv, rsv, |
3a584174 | 8405 | min_size, false); |
fcb80c2a | 8406 | BUG_ON(ret); |
f0cd846e | 8407 | |
5dc562c5 JB |
8408 | /* |
8409 | * So if we truncate and then write and fsync we normally would just | |
8410 | * write the extents that changed, which is a problem if we need to | |
8411 | * first truncate that entire inode. So set this flag so we write out | |
8412 | * all of the extents in the inode to the sync log so we're completely | |
8413 | * safe. | |
8414 | */ | |
8415 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags); | |
ca7e70f5 | 8416 | trans->block_rsv = rsv; |
907cbceb | 8417 | |
8082510e YZ |
8418 | while (1) { |
8419 | ret = btrfs_truncate_inode_items(trans, root, inode, | |
8420 | inode->i_size, | |
8421 | BTRFS_EXTENT_DATA_KEY); | |
ddfae63c | 8422 | trans->block_rsv = &fs_info->trans_block_rsv; |
ad7e1a74 | 8423 | if (ret != -ENOSPC && ret != -EAGAIN) |
8082510e | 8424 | break; |
39279cc3 | 8425 | |
8082510e | 8426 | ret = btrfs_update_inode(trans, root, inode); |
ad7e1a74 | 8427 | if (ret) |
3893e33b | 8428 | break; |
ca7e70f5 | 8429 | |
3a45bb20 | 8430 | btrfs_end_transaction(trans); |
2ff7e61e | 8431 | btrfs_btree_balance_dirty(fs_info); |
ca7e70f5 JB |
8432 | |
8433 | trans = btrfs_start_transaction(root, 2); | |
8434 | if (IS_ERR(trans)) { | |
ad7e1a74 | 8435 | ret = PTR_ERR(trans); |
ca7e70f5 JB |
8436 | trans = NULL; |
8437 | break; | |
8438 | } | |
8439 | ||
63f018be | 8440 | btrfs_block_rsv_release(fs_info, rsv, -1, NULL); |
0b246afa | 8441 | ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv, |
3a584174 | 8442 | rsv, min_size, false); |
ca7e70f5 JB |
8443 | BUG_ON(ret); /* shouldn't happen */ |
8444 | trans->block_rsv = rsv; | |
8082510e YZ |
8445 | } |
8446 | ||
ddfae63c JB |
8447 | /* |
8448 | * We can't call btrfs_truncate_block inside a trans handle as we could | |
8449 | * deadlock with freeze, if we got NEED_TRUNCATE_BLOCK then we know | |
8450 | * we've truncated everything except the last little bit, and can do | |
8451 | * btrfs_truncate_block and then update the disk_i_size. | |
8452 | */ | |
8453 | if (ret == NEED_TRUNCATE_BLOCK) { | |
8454 | btrfs_end_transaction(trans); | |
8455 | btrfs_btree_balance_dirty(fs_info); | |
8456 | ||
8457 | ret = btrfs_truncate_block(inode, inode->i_size, 0, 0); | |
8458 | if (ret) | |
8459 | goto out; | |
8460 | trans = btrfs_start_transaction(root, 1); | |
8461 | if (IS_ERR(trans)) { | |
8462 | ret = PTR_ERR(trans); | |
8463 | goto out; | |
8464 | } | |
d923afe9 | 8465 | btrfs_inode_safe_disk_i_size_write(inode, 0); |
ddfae63c JB |
8466 | } |
8467 | ||
917c16b2 | 8468 | if (trans) { |
ad7e1a74 OS |
8469 | int ret2; |
8470 | ||
0b246afa | 8471 | trans->block_rsv = &fs_info->trans_block_rsv; |
ad7e1a74 OS |
8472 | ret2 = btrfs_update_inode(trans, root, inode); |
8473 | if (ret2 && !ret) | |
8474 | ret = ret2; | |
7b128766 | 8475 | |
ad7e1a74 OS |
8476 | ret2 = btrfs_end_transaction(trans); |
8477 | if (ret2 && !ret) | |
8478 | ret = ret2; | |
2ff7e61e | 8479 | btrfs_btree_balance_dirty(fs_info); |
917c16b2 | 8480 | } |
fcb80c2a | 8481 | out: |
2ff7e61e | 8482 | btrfs_free_block_rsv(fs_info, rsv); |
fcb80c2a | 8483 | |
ad7e1a74 | 8484 | return ret; |
39279cc3 CM |
8485 | } |
8486 | ||
d352ac68 CM |
8487 | /* |
8488 | * create a new subvolume directory/inode (helper for the ioctl). | |
8489 | */ | |
d2fb3437 | 8490 | int btrfs_create_subvol_root(struct btrfs_trans_handle *trans, |
63541927 FDBM |
8491 | struct btrfs_root *new_root, |
8492 | struct btrfs_root *parent_root, | |
8493 | u64 new_dirid) | |
39279cc3 | 8494 | { |
39279cc3 | 8495 | struct inode *inode; |
76dda93c | 8496 | int err; |
00e4e6b3 | 8497 | u64 index = 0; |
39279cc3 | 8498 | |
12fc9d09 FA |
8499 | inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, |
8500 | new_dirid, new_dirid, | |
8501 | S_IFDIR | (~current_umask() & S_IRWXUGO), | |
8502 | &index); | |
54aa1f4d | 8503 | if (IS_ERR(inode)) |
f46b5a66 | 8504 | return PTR_ERR(inode); |
39279cc3 CM |
8505 | inode->i_op = &btrfs_dir_inode_operations; |
8506 | inode->i_fop = &btrfs_dir_file_operations; | |
8507 | ||
bfe86848 | 8508 | set_nlink(inode, 1); |
6ef06d27 | 8509 | btrfs_i_size_write(BTRFS_I(inode), 0); |
b0d5d10f | 8510 | unlock_new_inode(inode); |
3b96362c | 8511 | |
63541927 FDBM |
8512 | err = btrfs_subvol_inherit_props(trans, new_root, parent_root); |
8513 | if (err) | |
8514 | btrfs_err(new_root->fs_info, | |
351fd353 | 8515 | "error inheriting subvolume %llu properties: %d", |
63541927 FDBM |
8516 | new_root->root_key.objectid, err); |
8517 | ||
76dda93c | 8518 | err = btrfs_update_inode(trans, new_root, inode); |
cb8e7090 | 8519 | |
76dda93c | 8520 | iput(inode); |
ce598979 | 8521 | return err; |
39279cc3 CM |
8522 | } |
8523 | ||
39279cc3 CM |
8524 | struct inode *btrfs_alloc_inode(struct super_block *sb) |
8525 | { | |
69fe2d75 | 8526 | struct btrfs_fs_info *fs_info = btrfs_sb(sb); |
39279cc3 | 8527 | struct btrfs_inode *ei; |
2ead6ae7 | 8528 | struct inode *inode; |
39279cc3 | 8529 | |
712e36c5 | 8530 | ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_KERNEL); |
39279cc3 CM |
8531 | if (!ei) |
8532 | return NULL; | |
2ead6ae7 YZ |
8533 | |
8534 | ei->root = NULL; | |
2ead6ae7 | 8535 | ei->generation = 0; |
15ee9bc7 | 8536 | ei->last_trans = 0; |
257c62e1 | 8537 | ei->last_sub_trans = 0; |
e02119d5 | 8538 | ei->logged_trans = 0; |
2ead6ae7 | 8539 | ei->delalloc_bytes = 0; |
a7e3b975 | 8540 | ei->new_delalloc_bytes = 0; |
47059d93 | 8541 | ei->defrag_bytes = 0; |
2ead6ae7 YZ |
8542 | ei->disk_i_size = 0; |
8543 | ei->flags = 0; | |
7709cde3 | 8544 | ei->csum_bytes = 0; |
2ead6ae7 | 8545 | ei->index_cnt = (u64)-1; |
67de1176 | 8546 | ei->dir_index = 0; |
2ead6ae7 | 8547 | ei->last_unlink_trans = 0; |
46d8bc34 | 8548 | ei->last_log_commit = 0; |
2ead6ae7 | 8549 | |
9e0baf60 JB |
8550 | spin_lock_init(&ei->lock); |
8551 | ei->outstanding_extents = 0; | |
69fe2d75 JB |
8552 | if (sb->s_magic != BTRFS_TEST_MAGIC) |
8553 | btrfs_init_metadata_block_rsv(fs_info, &ei->block_rsv, | |
8554 | BTRFS_BLOCK_RSV_DELALLOC); | |
72ac3c0d | 8555 | ei->runtime_flags = 0; |
b52aa8c9 | 8556 | ei->prop_compress = BTRFS_COMPRESS_NONE; |
eec63c65 | 8557 | ei->defrag_compress = BTRFS_COMPRESS_NONE; |
2ead6ae7 | 8558 | |
16cdcec7 MX |
8559 | ei->delayed_node = NULL; |
8560 | ||
9cc97d64 | 8561 | ei->i_otime.tv_sec = 0; |
8562 | ei->i_otime.tv_nsec = 0; | |
8563 | ||
2ead6ae7 | 8564 | inode = &ei->vfs_inode; |
a8067e02 | 8565 | extent_map_tree_init(&ei->extent_tree); |
43eb5f29 QW |
8566 | extent_io_tree_init(fs_info, &ei->io_tree, IO_TREE_INODE_IO, inode); |
8567 | extent_io_tree_init(fs_info, &ei->io_failure_tree, | |
8568 | IO_TREE_INODE_IO_FAILURE, inode); | |
41a2ee75 JB |
8569 | extent_io_tree_init(fs_info, &ei->file_extent_tree, |
8570 | IO_TREE_INODE_FILE_EXTENT, inode); | |
7b439738 DS |
8571 | ei->io_tree.track_uptodate = true; |
8572 | ei->io_failure_tree.track_uptodate = true; | |
b812ce28 | 8573 | atomic_set(&ei->sync_writers, 0); |
2ead6ae7 | 8574 | mutex_init(&ei->log_mutex); |
e6dcd2dc | 8575 | btrfs_ordered_inode_tree_init(&ei->ordered_tree); |
2ead6ae7 | 8576 | INIT_LIST_HEAD(&ei->delalloc_inodes); |
8089fe62 | 8577 | INIT_LIST_HEAD(&ei->delayed_iput); |
2ead6ae7 | 8578 | RB_CLEAR_NODE(&ei->rb_node); |
5f9a8a51 | 8579 | init_rwsem(&ei->dio_sem); |
2ead6ae7 YZ |
8580 | |
8581 | return inode; | |
39279cc3 CM |
8582 | } |
8583 | ||
aaedb55b JB |
8584 | #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS |
8585 | void btrfs_test_destroy_inode(struct inode *inode) | |
8586 | { | |
dcdbc059 | 8587 | btrfs_drop_extent_cache(BTRFS_I(inode), 0, (u64)-1, 0); |
aaedb55b JB |
8588 | kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode)); |
8589 | } | |
8590 | #endif | |
8591 | ||
26602cab | 8592 | void btrfs_free_inode(struct inode *inode) |
fa0d7e3d | 8593 | { |
fa0d7e3d NP |
8594 | kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode)); |
8595 | } | |
8596 | ||
39279cc3 CM |
8597 | void btrfs_destroy_inode(struct inode *inode) |
8598 | { | |
0b246afa | 8599 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
e6dcd2dc | 8600 | struct btrfs_ordered_extent *ordered; |
5a3f23d5 CM |
8601 | struct btrfs_root *root = BTRFS_I(inode)->root; |
8602 | ||
b3d9b7a3 | 8603 | WARN_ON(!hlist_empty(&inode->i_dentry)); |
39279cc3 | 8604 | WARN_ON(inode->i_data.nrpages); |
69fe2d75 JB |
8605 | WARN_ON(BTRFS_I(inode)->block_rsv.reserved); |
8606 | WARN_ON(BTRFS_I(inode)->block_rsv.size); | |
9e0baf60 | 8607 | WARN_ON(BTRFS_I(inode)->outstanding_extents); |
7709cde3 | 8608 | WARN_ON(BTRFS_I(inode)->delalloc_bytes); |
a7e3b975 | 8609 | WARN_ON(BTRFS_I(inode)->new_delalloc_bytes); |
7709cde3 | 8610 | WARN_ON(BTRFS_I(inode)->csum_bytes); |
47059d93 | 8611 | WARN_ON(BTRFS_I(inode)->defrag_bytes); |
39279cc3 | 8612 | |
a6dbd429 JB |
8613 | /* |
8614 | * This can happen where we create an inode, but somebody else also | |
8615 | * created the same inode and we need to destroy the one we already | |
8616 | * created. | |
8617 | */ | |
8618 | if (!root) | |
26602cab | 8619 | return; |
a6dbd429 | 8620 | |
d397712b | 8621 | while (1) { |
e6dcd2dc CM |
8622 | ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1); |
8623 | if (!ordered) | |
8624 | break; | |
8625 | else { | |
0b246afa | 8626 | btrfs_err(fs_info, |
5d163e0e | 8627 | "found ordered extent %llu %llu on inode cleanup", |
bffe633e | 8628 | ordered->file_offset, ordered->num_bytes); |
e6dcd2dc CM |
8629 | btrfs_remove_ordered_extent(inode, ordered); |
8630 | btrfs_put_ordered_extent(ordered); | |
8631 | btrfs_put_ordered_extent(ordered); | |
8632 | } | |
8633 | } | |
cfdd4592 | 8634 | btrfs_qgroup_check_reserved_leak(BTRFS_I(inode)); |
5d4f98a2 | 8635 | inode_tree_del(inode); |
dcdbc059 | 8636 | btrfs_drop_extent_cache(BTRFS_I(inode), 0, (u64)-1, 0); |
41a2ee75 | 8637 | btrfs_inode_clear_file_extent_range(BTRFS_I(inode), 0, (u64)-1); |
5c8fd99f | 8638 | btrfs_put_root(BTRFS_I(inode)->root); |
39279cc3 CM |
8639 | } |
8640 | ||
45321ac5 | 8641 | int btrfs_drop_inode(struct inode *inode) |
76dda93c YZ |
8642 | { |
8643 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
45321ac5 | 8644 | |
6379ef9f NA |
8645 | if (root == NULL) |
8646 | return 1; | |
8647 | ||
fa6ac876 | 8648 | /* the snap/subvol tree is on deleting */ |
69e9c6c6 | 8649 | if (btrfs_root_refs(&root->root_item) == 0) |
45321ac5 | 8650 | return 1; |
76dda93c | 8651 | else |
45321ac5 | 8652 | return generic_drop_inode(inode); |
76dda93c YZ |
8653 | } |
8654 | ||
0ee0fda0 | 8655 | static void init_once(void *foo) |
39279cc3 CM |
8656 | { |
8657 | struct btrfs_inode *ei = (struct btrfs_inode *) foo; | |
8658 | ||
8659 | inode_init_once(&ei->vfs_inode); | |
8660 | } | |
8661 | ||
e67c718b | 8662 | void __cold btrfs_destroy_cachep(void) |
39279cc3 | 8663 | { |
8c0a8537 KS |
8664 | /* |
8665 | * Make sure all delayed rcu free inodes are flushed before we | |
8666 | * destroy cache. | |
8667 | */ | |
8668 | rcu_barrier(); | |
5598e900 KM |
8669 | kmem_cache_destroy(btrfs_inode_cachep); |
8670 | kmem_cache_destroy(btrfs_trans_handle_cachep); | |
5598e900 KM |
8671 | kmem_cache_destroy(btrfs_path_cachep); |
8672 | kmem_cache_destroy(btrfs_free_space_cachep); | |
3acd4850 | 8673 | kmem_cache_destroy(btrfs_free_space_bitmap_cachep); |
39279cc3 CM |
8674 | } |
8675 | ||
f5c29bd9 | 8676 | int __init btrfs_init_cachep(void) |
39279cc3 | 8677 | { |
837e1972 | 8678 | btrfs_inode_cachep = kmem_cache_create("btrfs_inode", |
9601e3f6 | 8679 | sizeof(struct btrfs_inode), 0, |
5d097056 VD |
8680 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD | SLAB_ACCOUNT, |
8681 | init_once); | |
39279cc3 CM |
8682 | if (!btrfs_inode_cachep) |
8683 | goto fail; | |
9601e3f6 | 8684 | |
837e1972 | 8685 | btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle", |
9601e3f6 | 8686 | sizeof(struct btrfs_trans_handle), 0, |
fba4b697 | 8687 | SLAB_TEMPORARY | SLAB_MEM_SPREAD, NULL); |
39279cc3 CM |
8688 | if (!btrfs_trans_handle_cachep) |
8689 | goto fail; | |
9601e3f6 | 8690 | |
837e1972 | 8691 | btrfs_path_cachep = kmem_cache_create("btrfs_path", |
9601e3f6 | 8692 | sizeof(struct btrfs_path), 0, |
fba4b697 | 8693 | SLAB_MEM_SPREAD, NULL); |
39279cc3 CM |
8694 | if (!btrfs_path_cachep) |
8695 | goto fail; | |
9601e3f6 | 8696 | |
837e1972 | 8697 | btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space", |
dc89e982 | 8698 | sizeof(struct btrfs_free_space), 0, |
fba4b697 | 8699 | SLAB_MEM_SPREAD, NULL); |
dc89e982 JB |
8700 | if (!btrfs_free_space_cachep) |
8701 | goto fail; | |
8702 | ||
3acd4850 CL |
8703 | btrfs_free_space_bitmap_cachep = kmem_cache_create("btrfs_free_space_bitmap", |
8704 | PAGE_SIZE, PAGE_SIZE, | |
8705 | SLAB_RED_ZONE, NULL); | |
8706 | if (!btrfs_free_space_bitmap_cachep) | |
8707 | goto fail; | |
8708 | ||
39279cc3 CM |
8709 | return 0; |
8710 | fail: | |
8711 | btrfs_destroy_cachep(); | |
8712 | return -ENOMEM; | |
8713 | } | |
8714 | ||
a528d35e DH |
8715 | static int btrfs_getattr(const struct path *path, struct kstat *stat, |
8716 | u32 request_mask, unsigned int flags) | |
39279cc3 | 8717 | { |
df0af1a5 | 8718 | u64 delalloc_bytes; |
a528d35e | 8719 | struct inode *inode = d_inode(path->dentry); |
fadc0d8b | 8720 | u32 blocksize = inode->i_sb->s_blocksize; |
04a87e34 YS |
8721 | u32 bi_flags = BTRFS_I(inode)->flags; |
8722 | ||
8723 | stat->result_mask |= STATX_BTIME; | |
8724 | stat->btime.tv_sec = BTRFS_I(inode)->i_otime.tv_sec; | |
8725 | stat->btime.tv_nsec = BTRFS_I(inode)->i_otime.tv_nsec; | |
8726 | if (bi_flags & BTRFS_INODE_APPEND) | |
8727 | stat->attributes |= STATX_ATTR_APPEND; | |
8728 | if (bi_flags & BTRFS_INODE_COMPRESS) | |
8729 | stat->attributes |= STATX_ATTR_COMPRESSED; | |
8730 | if (bi_flags & BTRFS_INODE_IMMUTABLE) | |
8731 | stat->attributes |= STATX_ATTR_IMMUTABLE; | |
8732 | if (bi_flags & BTRFS_INODE_NODUMP) | |
8733 | stat->attributes |= STATX_ATTR_NODUMP; | |
8734 | ||
8735 | stat->attributes_mask |= (STATX_ATTR_APPEND | | |
8736 | STATX_ATTR_COMPRESSED | | |
8737 | STATX_ATTR_IMMUTABLE | | |
8738 | STATX_ATTR_NODUMP); | |
fadc0d8b | 8739 | |
39279cc3 | 8740 | generic_fillattr(inode, stat); |
0ee5dc67 | 8741 | stat->dev = BTRFS_I(inode)->root->anon_dev; |
df0af1a5 MX |
8742 | |
8743 | spin_lock(&BTRFS_I(inode)->lock); | |
a7e3b975 | 8744 | delalloc_bytes = BTRFS_I(inode)->new_delalloc_bytes; |
df0af1a5 | 8745 | spin_unlock(&BTRFS_I(inode)->lock); |
fadc0d8b | 8746 | stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) + |
df0af1a5 | 8747 | ALIGN(delalloc_bytes, blocksize)) >> 9; |
39279cc3 CM |
8748 | return 0; |
8749 | } | |
8750 | ||
cdd1fedf DF |
8751 | static int btrfs_rename_exchange(struct inode *old_dir, |
8752 | struct dentry *old_dentry, | |
8753 | struct inode *new_dir, | |
8754 | struct dentry *new_dentry) | |
8755 | { | |
0b246afa | 8756 | struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb); |
cdd1fedf DF |
8757 | struct btrfs_trans_handle *trans; |
8758 | struct btrfs_root *root = BTRFS_I(old_dir)->root; | |
8759 | struct btrfs_root *dest = BTRFS_I(new_dir)->root; | |
8760 | struct inode *new_inode = new_dentry->d_inode; | |
8761 | struct inode *old_inode = old_dentry->d_inode; | |
95582b00 | 8762 | struct timespec64 ctime = current_time(old_inode); |
cdd1fedf | 8763 | struct dentry *parent; |
4a0cc7ca NB |
8764 | u64 old_ino = btrfs_ino(BTRFS_I(old_inode)); |
8765 | u64 new_ino = btrfs_ino(BTRFS_I(new_inode)); | |
cdd1fedf DF |
8766 | u64 old_idx = 0; |
8767 | u64 new_idx = 0; | |
cdd1fedf | 8768 | int ret; |
86e8aa0e FM |
8769 | bool root_log_pinned = false; |
8770 | bool dest_log_pinned = false; | |
d4682ba0 FM |
8771 | struct btrfs_log_ctx ctx_root; |
8772 | struct btrfs_log_ctx ctx_dest; | |
8773 | bool sync_log_root = false; | |
8774 | bool sync_log_dest = false; | |
8775 | bool commit_transaction = false; | |
cdd1fedf DF |
8776 | |
8777 | /* we only allow rename subvolume link between subvolumes */ | |
8778 | if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest) | |
8779 | return -EXDEV; | |
8780 | ||
d4682ba0 FM |
8781 | btrfs_init_log_ctx(&ctx_root, old_inode); |
8782 | btrfs_init_log_ctx(&ctx_dest, new_inode); | |
8783 | ||
cdd1fedf | 8784 | /* close the race window with snapshot create/destroy ioctl */ |
943eb3bf JB |
8785 | if (old_ino == BTRFS_FIRST_FREE_OBJECTID || |
8786 | new_ino == BTRFS_FIRST_FREE_OBJECTID) | |
0b246afa | 8787 | down_read(&fs_info->subvol_sem); |
cdd1fedf DF |
8788 | |
8789 | /* | |
8790 | * We want to reserve the absolute worst case amount of items. So if | |
8791 | * both inodes are subvols and we need to unlink them then that would | |
8792 | * require 4 item modifications, but if they are both normal inodes it | |
8793 | * would require 5 item modifications, so we'll assume their normal | |
8794 | * inodes. So 5 * 2 is 10, plus 2 for the new links, so 12 total items | |
8795 | * should cover the worst case number of items we'll modify. | |
8796 | */ | |
8797 | trans = btrfs_start_transaction(root, 12); | |
8798 | if (IS_ERR(trans)) { | |
8799 | ret = PTR_ERR(trans); | |
8800 | goto out_notrans; | |
8801 | } | |
8802 | ||
3e174099 JB |
8803 | if (dest != root) |
8804 | btrfs_record_root_in_trans(trans, dest); | |
8805 | ||
cdd1fedf DF |
8806 | /* |
8807 | * We need to find a free sequence number both in the source and | |
8808 | * in the destination directory for the exchange. | |
8809 | */ | |
877574e2 | 8810 | ret = btrfs_set_inode_index(BTRFS_I(new_dir), &old_idx); |
cdd1fedf DF |
8811 | if (ret) |
8812 | goto out_fail; | |
877574e2 | 8813 | ret = btrfs_set_inode_index(BTRFS_I(old_dir), &new_idx); |
cdd1fedf DF |
8814 | if (ret) |
8815 | goto out_fail; | |
8816 | ||
8817 | BTRFS_I(old_inode)->dir_index = 0ULL; | |
8818 | BTRFS_I(new_inode)->dir_index = 0ULL; | |
8819 | ||
8820 | /* Reference for the source. */ | |
8821 | if (old_ino == BTRFS_FIRST_FREE_OBJECTID) { | |
8822 | /* force full log commit if subvolume involved. */ | |
90787766 | 8823 | btrfs_set_log_full_commit(trans); |
cdd1fedf | 8824 | } else { |
376e5a57 FM |
8825 | btrfs_pin_log_trans(root); |
8826 | root_log_pinned = true; | |
cdd1fedf DF |
8827 | ret = btrfs_insert_inode_ref(trans, dest, |
8828 | new_dentry->d_name.name, | |
8829 | new_dentry->d_name.len, | |
8830 | old_ino, | |
f85b7379 DS |
8831 | btrfs_ino(BTRFS_I(new_dir)), |
8832 | old_idx); | |
cdd1fedf DF |
8833 | if (ret) |
8834 | goto out_fail; | |
cdd1fedf DF |
8835 | } |
8836 | ||
8837 | /* And now for the dest. */ | |
8838 | if (new_ino == BTRFS_FIRST_FREE_OBJECTID) { | |
8839 | /* force full log commit if subvolume involved. */ | |
90787766 | 8840 | btrfs_set_log_full_commit(trans); |
cdd1fedf | 8841 | } else { |
376e5a57 FM |
8842 | btrfs_pin_log_trans(dest); |
8843 | dest_log_pinned = true; | |
cdd1fedf DF |
8844 | ret = btrfs_insert_inode_ref(trans, root, |
8845 | old_dentry->d_name.name, | |
8846 | old_dentry->d_name.len, | |
8847 | new_ino, | |
f85b7379 DS |
8848 | btrfs_ino(BTRFS_I(old_dir)), |
8849 | new_idx); | |
cdd1fedf DF |
8850 | if (ret) |
8851 | goto out_fail; | |
cdd1fedf DF |
8852 | } |
8853 | ||
8854 | /* Update inode version and ctime/mtime. */ | |
8855 | inode_inc_iversion(old_dir); | |
8856 | inode_inc_iversion(new_dir); | |
8857 | inode_inc_iversion(old_inode); | |
8858 | inode_inc_iversion(new_inode); | |
8859 | old_dir->i_ctime = old_dir->i_mtime = ctime; | |
8860 | new_dir->i_ctime = new_dir->i_mtime = ctime; | |
8861 | old_inode->i_ctime = ctime; | |
8862 | new_inode->i_ctime = ctime; | |
8863 | ||
8864 | if (old_dentry->d_parent != new_dentry->d_parent) { | |
f85b7379 DS |
8865 | btrfs_record_unlink_dir(trans, BTRFS_I(old_dir), |
8866 | BTRFS_I(old_inode), 1); | |
8867 | btrfs_record_unlink_dir(trans, BTRFS_I(new_dir), | |
8868 | BTRFS_I(new_inode), 1); | |
cdd1fedf DF |
8869 | } |
8870 | ||
8871 | /* src is a subvolume */ | |
8872 | if (old_ino == BTRFS_FIRST_FREE_OBJECTID) { | |
045d3967 | 8873 | ret = btrfs_unlink_subvol(trans, old_dir, old_dentry); |
cdd1fedf | 8874 | } else { /* src is an inode */ |
4ec5934e NB |
8875 | ret = __btrfs_unlink_inode(trans, root, BTRFS_I(old_dir), |
8876 | BTRFS_I(old_dentry->d_inode), | |
cdd1fedf DF |
8877 | old_dentry->d_name.name, |
8878 | old_dentry->d_name.len); | |
8879 | if (!ret) | |
8880 | ret = btrfs_update_inode(trans, root, old_inode); | |
8881 | } | |
8882 | if (ret) { | |
66642832 | 8883 | btrfs_abort_transaction(trans, ret); |
cdd1fedf DF |
8884 | goto out_fail; |
8885 | } | |
8886 | ||
8887 | /* dest is a subvolume */ | |
8888 | if (new_ino == BTRFS_FIRST_FREE_OBJECTID) { | |
045d3967 | 8889 | ret = btrfs_unlink_subvol(trans, new_dir, new_dentry); |
cdd1fedf | 8890 | } else { /* dest is an inode */ |
4ec5934e NB |
8891 | ret = __btrfs_unlink_inode(trans, dest, BTRFS_I(new_dir), |
8892 | BTRFS_I(new_dentry->d_inode), | |
cdd1fedf DF |
8893 | new_dentry->d_name.name, |
8894 | new_dentry->d_name.len); | |
8895 | if (!ret) | |
8896 | ret = btrfs_update_inode(trans, dest, new_inode); | |
8897 | } | |
8898 | if (ret) { | |
66642832 | 8899 | btrfs_abort_transaction(trans, ret); |
cdd1fedf DF |
8900 | goto out_fail; |
8901 | } | |
8902 | ||
db0a669f | 8903 | ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode), |
cdd1fedf DF |
8904 | new_dentry->d_name.name, |
8905 | new_dentry->d_name.len, 0, old_idx); | |
8906 | if (ret) { | |
66642832 | 8907 | btrfs_abort_transaction(trans, ret); |
cdd1fedf DF |
8908 | goto out_fail; |
8909 | } | |
8910 | ||
db0a669f | 8911 | ret = btrfs_add_link(trans, BTRFS_I(old_dir), BTRFS_I(new_inode), |
cdd1fedf DF |
8912 | old_dentry->d_name.name, |
8913 | old_dentry->d_name.len, 0, new_idx); | |
8914 | if (ret) { | |
66642832 | 8915 | btrfs_abort_transaction(trans, ret); |
cdd1fedf DF |
8916 | goto out_fail; |
8917 | } | |
8918 | ||
8919 | if (old_inode->i_nlink == 1) | |
8920 | BTRFS_I(old_inode)->dir_index = old_idx; | |
8921 | if (new_inode->i_nlink == 1) | |
8922 | BTRFS_I(new_inode)->dir_index = new_idx; | |
8923 | ||
86e8aa0e | 8924 | if (root_log_pinned) { |
cdd1fedf | 8925 | parent = new_dentry->d_parent; |
d4682ba0 FM |
8926 | ret = btrfs_log_new_name(trans, BTRFS_I(old_inode), |
8927 | BTRFS_I(old_dir), parent, | |
8928 | false, &ctx_root); | |
8929 | if (ret == BTRFS_NEED_LOG_SYNC) | |
8930 | sync_log_root = true; | |
8931 | else if (ret == BTRFS_NEED_TRANS_COMMIT) | |
8932 | commit_transaction = true; | |
8933 | ret = 0; | |
cdd1fedf | 8934 | btrfs_end_log_trans(root); |
86e8aa0e | 8935 | root_log_pinned = false; |
cdd1fedf | 8936 | } |
86e8aa0e | 8937 | if (dest_log_pinned) { |
d4682ba0 FM |
8938 | if (!commit_transaction) { |
8939 | parent = old_dentry->d_parent; | |
8940 | ret = btrfs_log_new_name(trans, BTRFS_I(new_inode), | |
8941 | BTRFS_I(new_dir), parent, | |
8942 | false, &ctx_dest); | |
8943 | if (ret == BTRFS_NEED_LOG_SYNC) | |
8944 | sync_log_dest = true; | |
8945 | else if (ret == BTRFS_NEED_TRANS_COMMIT) | |
8946 | commit_transaction = true; | |
8947 | ret = 0; | |
8948 | } | |
cdd1fedf | 8949 | btrfs_end_log_trans(dest); |
86e8aa0e | 8950 | dest_log_pinned = false; |
cdd1fedf DF |
8951 | } |
8952 | out_fail: | |
86e8aa0e FM |
8953 | /* |
8954 | * If we have pinned a log and an error happened, we unpin tasks | |
8955 | * trying to sync the log and force them to fallback to a transaction | |
8956 | * commit if the log currently contains any of the inodes involved in | |
8957 | * this rename operation (to ensure we do not persist a log with an | |
8958 | * inconsistent state for any of these inodes or leading to any | |
8959 | * inconsistencies when replayed). If the transaction was aborted, the | |
8960 | * abortion reason is propagated to userspace when attempting to commit | |
8961 | * the transaction. If the log does not contain any of these inodes, we | |
8962 | * allow the tasks to sync it. | |
8963 | */ | |
8964 | if (ret && (root_log_pinned || dest_log_pinned)) { | |
0f8939b8 NB |
8965 | if (btrfs_inode_in_log(BTRFS_I(old_dir), fs_info->generation) || |
8966 | btrfs_inode_in_log(BTRFS_I(new_dir), fs_info->generation) || | |
8967 | btrfs_inode_in_log(BTRFS_I(old_inode), fs_info->generation) || | |
86e8aa0e | 8968 | (new_inode && |
0f8939b8 | 8969 | btrfs_inode_in_log(BTRFS_I(new_inode), fs_info->generation))) |
90787766 | 8970 | btrfs_set_log_full_commit(trans); |
86e8aa0e FM |
8971 | |
8972 | if (root_log_pinned) { | |
8973 | btrfs_end_log_trans(root); | |
8974 | root_log_pinned = false; | |
8975 | } | |
8976 | if (dest_log_pinned) { | |
8977 | btrfs_end_log_trans(dest); | |
8978 | dest_log_pinned = false; | |
8979 | } | |
8980 | } | |
d4682ba0 FM |
8981 | if (!ret && sync_log_root && !commit_transaction) { |
8982 | ret = btrfs_sync_log(trans, BTRFS_I(old_inode)->root, | |
8983 | &ctx_root); | |
8984 | if (ret) | |
8985 | commit_transaction = true; | |
8986 | } | |
8987 | if (!ret && sync_log_dest && !commit_transaction) { | |
8988 | ret = btrfs_sync_log(trans, BTRFS_I(new_inode)->root, | |
8989 | &ctx_dest); | |
8990 | if (ret) | |
8991 | commit_transaction = true; | |
8992 | } | |
8993 | if (commit_transaction) { | |
e6c61710 FM |
8994 | /* |
8995 | * We may have set commit_transaction when logging the new name | |
8996 | * in the destination root, in which case we left the source | |
8997 | * root context in the list of log contextes. So make sure we | |
8998 | * remove it to avoid invalid memory accesses, since the context | |
8999 | * was allocated in our stack frame. | |
9000 | */ | |
9001 | if (sync_log_root) { | |
9002 | mutex_lock(&root->log_mutex); | |
9003 | list_del_init(&ctx_root.list); | |
9004 | mutex_unlock(&root->log_mutex); | |
9005 | } | |
d4682ba0 FM |
9006 | ret = btrfs_commit_transaction(trans); |
9007 | } else { | |
9008 | int ret2; | |
9009 | ||
9010 | ret2 = btrfs_end_transaction(trans); | |
9011 | ret = ret ? ret : ret2; | |
9012 | } | |
cdd1fedf | 9013 | out_notrans: |
943eb3bf JB |
9014 | if (new_ino == BTRFS_FIRST_FREE_OBJECTID || |
9015 | old_ino == BTRFS_FIRST_FREE_OBJECTID) | |
0b246afa | 9016 | up_read(&fs_info->subvol_sem); |
cdd1fedf | 9017 | |
e6c61710 FM |
9018 | ASSERT(list_empty(&ctx_root.list)); |
9019 | ASSERT(list_empty(&ctx_dest.list)); | |
9020 | ||
cdd1fedf DF |
9021 | return ret; |
9022 | } | |
9023 | ||
9024 | static int btrfs_whiteout_for_rename(struct btrfs_trans_handle *trans, | |
9025 | struct btrfs_root *root, | |
9026 | struct inode *dir, | |
9027 | struct dentry *dentry) | |
9028 | { | |
9029 | int ret; | |
9030 | struct inode *inode; | |
9031 | u64 objectid; | |
9032 | u64 index; | |
9033 | ||
9034 | ret = btrfs_find_free_ino(root, &objectid); | |
9035 | if (ret) | |
9036 | return ret; | |
9037 | ||
9038 | inode = btrfs_new_inode(trans, root, dir, | |
9039 | dentry->d_name.name, | |
9040 | dentry->d_name.len, | |
4a0cc7ca | 9041 | btrfs_ino(BTRFS_I(dir)), |
cdd1fedf DF |
9042 | objectid, |
9043 | S_IFCHR | WHITEOUT_MODE, | |
9044 | &index); | |
9045 | ||
9046 | if (IS_ERR(inode)) { | |
9047 | ret = PTR_ERR(inode); | |
9048 | return ret; | |
9049 | } | |
9050 | ||
9051 | inode->i_op = &btrfs_special_inode_operations; | |
9052 | init_special_inode(inode, inode->i_mode, | |
9053 | WHITEOUT_DEV); | |
9054 | ||
9055 | ret = btrfs_init_inode_security(trans, inode, dir, | |
9056 | &dentry->d_name); | |
9057 | if (ret) | |
c9901618 | 9058 | goto out; |
cdd1fedf | 9059 | |
cef415af NB |
9060 | ret = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, |
9061 | BTRFS_I(inode), 0, index); | |
cdd1fedf | 9062 | if (ret) |
c9901618 | 9063 | goto out; |
cdd1fedf DF |
9064 | |
9065 | ret = btrfs_update_inode(trans, root, inode); | |
c9901618 | 9066 | out: |
cdd1fedf | 9067 | unlock_new_inode(inode); |
c9901618 FM |
9068 | if (ret) |
9069 | inode_dec_link_count(inode); | |
cdd1fedf DF |
9070 | iput(inode); |
9071 | ||
c9901618 | 9072 | return ret; |
cdd1fedf DF |
9073 | } |
9074 | ||
d397712b | 9075 | static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry, |
cdd1fedf DF |
9076 | struct inode *new_dir, struct dentry *new_dentry, |
9077 | unsigned int flags) | |
39279cc3 | 9078 | { |
0b246afa | 9079 | struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb); |
39279cc3 | 9080 | struct btrfs_trans_handle *trans; |
5062af35 | 9081 | unsigned int trans_num_items; |
39279cc3 | 9082 | struct btrfs_root *root = BTRFS_I(old_dir)->root; |
4df27c4d | 9083 | struct btrfs_root *dest = BTRFS_I(new_dir)->root; |
2b0143b5 DH |
9084 | struct inode *new_inode = d_inode(new_dentry); |
9085 | struct inode *old_inode = d_inode(old_dentry); | |
00e4e6b3 | 9086 | u64 index = 0; |
39279cc3 | 9087 | int ret; |
4a0cc7ca | 9088 | u64 old_ino = btrfs_ino(BTRFS_I(old_inode)); |
3dc9e8f7 | 9089 | bool log_pinned = false; |
d4682ba0 FM |
9090 | struct btrfs_log_ctx ctx; |
9091 | bool sync_log = false; | |
9092 | bool commit_transaction = false; | |
39279cc3 | 9093 | |
4a0cc7ca | 9094 | if (btrfs_ino(BTRFS_I(new_dir)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) |
f679a840 YZ |
9095 | return -EPERM; |
9096 | ||
4df27c4d | 9097 | /* we only allow rename subvolume link between subvolumes */ |
33345d01 | 9098 | if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest) |
3394e160 CM |
9099 | return -EXDEV; |
9100 | ||
33345d01 | 9101 | if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID || |
4a0cc7ca | 9102 | (new_inode && btrfs_ino(BTRFS_I(new_inode)) == BTRFS_FIRST_FREE_OBJECTID)) |
39279cc3 | 9103 | return -ENOTEMPTY; |
5f39d397 | 9104 | |
4df27c4d YZ |
9105 | if (S_ISDIR(old_inode->i_mode) && new_inode && |
9106 | new_inode->i_size > BTRFS_EMPTY_DIR_SIZE) | |
9107 | return -ENOTEMPTY; | |
9c52057c CM |
9108 | |
9109 | ||
9110 | /* check for collisions, even if the name isn't there */ | |
4871c158 | 9111 | ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino, |
9c52057c CM |
9112 | new_dentry->d_name.name, |
9113 | new_dentry->d_name.len); | |
9114 | ||
9115 | if (ret) { | |
9116 | if (ret == -EEXIST) { | |
9117 | /* we shouldn't get | |
9118 | * eexist without a new_inode */ | |
fae7f21c | 9119 | if (WARN_ON(!new_inode)) { |
9c52057c CM |
9120 | return ret; |
9121 | } | |
9122 | } else { | |
9123 | /* maybe -EOVERFLOW */ | |
9124 | return ret; | |
9125 | } | |
9126 | } | |
9127 | ret = 0; | |
9128 | ||
5a3f23d5 | 9129 | /* |
8d875f95 CM |
9130 | * we're using rename to replace one file with another. Start IO on it |
9131 | * now so we don't add too much work to the end of the transaction | |
5a3f23d5 | 9132 | */ |
8d875f95 | 9133 | if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size) |
5a3f23d5 CM |
9134 | filemap_flush(old_inode->i_mapping); |
9135 | ||
76dda93c | 9136 | /* close the racy window with snapshot create/destroy ioctl */ |
33345d01 | 9137 | if (old_ino == BTRFS_FIRST_FREE_OBJECTID) |
0b246afa | 9138 | down_read(&fs_info->subvol_sem); |
a22285a6 YZ |
9139 | /* |
9140 | * We want to reserve the absolute worst case amount of items. So if | |
9141 | * both inodes are subvols and we need to unlink them then that would | |
9142 | * require 4 item modifications, but if they are both normal inodes it | |
cdd1fedf | 9143 | * would require 5 item modifications, so we'll assume they are normal |
a22285a6 YZ |
9144 | * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items |
9145 | * should cover the worst case number of items we'll modify. | |
5062af35 FM |
9146 | * If our rename has the whiteout flag, we need more 5 units for the |
9147 | * new inode (1 inode item, 1 inode ref, 2 dir items and 1 xattr item | |
9148 | * when selinux is enabled). | |
a22285a6 | 9149 | */ |
5062af35 FM |
9150 | trans_num_items = 11; |
9151 | if (flags & RENAME_WHITEOUT) | |
9152 | trans_num_items += 5; | |
9153 | trans = btrfs_start_transaction(root, trans_num_items); | |
b44c59a8 | 9154 | if (IS_ERR(trans)) { |
cdd1fedf DF |
9155 | ret = PTR_ERR(trans); |
9156 | goto out_notrans; | |
9157 | } | |
76dda93c | 9158 | |
4df27c4d YZ |
9159 | if (dest != root) |
9160 | btrfs_record_root_in_trans(trans, dest); | |
5f39d397 | 9161 | |
877574e2 | 9162 | ret = btrfs_set_inode_index(BTRFS_I(new_dir), &index); |
a5719521 YZ |
9163 | if (ret) |
9164 | goto out_fail; | |
5a3f23d5 | 9165 | |
67de1176 | 9166 | BTRFS_I(old_inode)->dir_index = 0ULL; |
33345d01 | 9167 | if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) { |
4df27c4d | 9168 | /* force full log commit if subvolume involved. */ |
90787766 | 9169 | btrfs_set_log_full_commit(trans); |
4df27c4d | 9170 | } else { |
c4aba954 FM |
9171 | btrfs_pin_log_trans(root); |
9172 | log_pinned = true; | |
a5719521 YZ |
9173 | ret = btrfs_insert_inode_ref(trans, dest, |
9174 | new_dentry->d_name.name, | |
9175 | new_dentry->d_name.len, | |
33345d01 | 9176 | old_ino, |
4a0cc7ca | 9177 | btrfs_ino(BTRFS_I(new_dir)), index); |
a5719521 YZ |
9178 | if (ret) |
9179 | goto out_fail; | |
4df27c4d | 9180 | } |
5a3f23d5 | 9181 | |
0c4d2d95 JB |
9182 | inode_inc_iversion(old_dir); |
9183 | inode_inc_iversion(new_dir); | |
9184 | inode_inc_iversion(old_inode); | |
04b285f3 DD |
9185 | old_dir->i_ctime = old_dir->i_mtime = |
9186 | new_dir->i_ctime = new_dir->i_mtime = | |
c2050a45 | 9187 | old_inode->i_ctime = current_time(old_dir); |
5f39d397 | 9188 | |
12fcfd22 | 9189 | if (old_dentry->d_parent != new_dentry->d_parent) |
f85b7379 DS |
9190 | btrfs_record_unlink_dir(trans, BTRFS_I(old_dir), |
9191 | BTRFS_I(old_inode), 1); | |
12fcfd22 | 9192 | |
33345d01 | 9193 | if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) { |
045d3967 | 9194 | ret = btrfs_unlink_subvol(trans, old_dir, old_dentry); |
4df27c4d | 9195 | } else { |
4ec5934e NB |
9196 | ret = __btrfs_unlink_inode(trans, root, BTRFS_I(old_dir), |
9197 | BTRFS_I(d_inode(old_dentry)), | |
92986796 AV |
9198 | old_dentry->d_name.name, |
9199 | old_dentry->d_name.len); | |
9200 | if (!ret) | |
9201 | ret = btrfs_update_inode(trans, root, old_inode); | |
4df27c4d | 9202 | } |
79787eaa | 9203 | if (ret) { |
66642832 | 9204 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
9205 | goto out_fail; |
9206 | } | |
39279cc3 CM |
9207 | |
9208 | if (new_inode) { | |
0c4d2d95 | 9209 | inode_inc_iversion(new_inode); |
c2050a45 | 9210 | new_inode->i_ctime = current_time(new_inode); |
4a0cc7ca | 9211 | if (unlikely(btrfs_ino(BTRFS_I(new_inode)) == |
4df27c4d | 9212 | BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) { |
045d3967 | 9213 | ret = btrfs_unlink_subvol(trans, new_dir, new_dentry); |
4df27c4d YZ |
9214 | BUG_ON(new_inode->i_nlink == 0); |
9215 | } else { | |
4ec5934e NB |
9216 | ret = btrfs_unlink_inode(trans, dest, BTRFS_I(new_dir), |
9217 | BTRFS_I(d_inode(new_dentry)), | |
4df27c4d YZ |
9218 | new_dentry->d_name.name, |
9219 | new_dentry->d_name.len); | |
9220 | } | |
4ef31a45 | 9221 | if (!ret && new_inode->i_nlink == 0) |
73f2e545 NB |
9222 | ret = btrfs_orphan_add(trans, |
9223 | BTRFS_I(d_inode(new_dentry))); | |
79787eaa | 9224 | if (ret) { |
66642832 | 9225 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
9226 | goto out_fail; |
9227 | } | |
39279cc3 | 9228 | } |
aec7477b | 9229 | |
db0a669f | 9230 | ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode), |
4df27c4d | 9231 | new_dentry->d_name.name, |
a5719521 | 9232 | new_dentry->d_name.len, 0, index); |
79787eaa | 9233 | if (ret) { |
66642832 | 9234 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
9235 | goto out_fail; |
9236 | } | |
39279cc3 | 9237 | |
67de1176 MX |
9238 | if (old_inode->i_nlink == 1) |
9239 | BTRFS_I(old_inode)->dir_index = index; | |
9240 | ||
3dc9e8f7 | 9241 | if (log_pinned) { |
10d9f309 | 9242 | struct dentry *parent = new_dentry->d_parent; |
3dc9e8f7 | 9243 | |
d4682ba0 FM |
9244 | btrfs_init_log_ctx(&ctx, old_inode); |
9245 | ret = btrfs_log_new_name(trans, BTRFS_I(old_inode), | |
9246 | BTRFS_I(old_dir), parent, | |
9247 | false, &ctx); | |
9248 | if (ret == BTRFS_NEED_LOG_SYNC) | |
9249 | sync_log = true; | |
9250 | else if (ret == BTRFS_NEED_TRANS_COMMIT) | |
9251 | commit_transaction = true; | |
9252 | ret = 0; | |
4df27c4d | 9253 | btrfs_end_log_trans(root); |
3dc9e8f7 | 9254 | log_pinned = false; |
4df27c4d | 9255 | } |
cdd1fedf DF |
9256 | |
9257 | if (flags & RENAME_WHITEOUT) { | |
9258 | ret = btrfs_whiteout_for_rename(trans, root, old_dir, | |
9259 | old_dentry); | |
9260 | ||
9261 | if (ret) { | |
66642832 | 9262 | btrfs_abort_transaction(trans, ret); |
cdd1fedf DF |
9263 | goto out_fail; |
9264 | } | |
4df27c4d | 9265 | } |
39279cc3 | 9266 | out_fail: |
3dc9e8f7 FM |
9267 | /* |
9268 | * If we have pinned the log and an error happened, we unpin tasks | |
9269 | * trying to sync the log and force them to fallback to a transaction | |
9270 | * commit if the log currently contains any of the inodes involved in | |
9271 | * this rename operation (to ensure we do not persist a log with an | |
9272 | * inconsistent state for any of these inodes or leading to any | |
9273 | * inconsistencies when replayed). If the transaction was aborted, the | |
9274 | * abortion reason is propagated to userspace when attempting to commit | |
9275 | * the transaction. If the log does not contain any of these inodes, we | |
9276 | * allow the tasks to sync it. | |
9277 | */ | |
9278 | if (ret && log_pinned) { | |
0f8939b8 NB |
9279 | if (btrfs_inode_in_log(BTRFS_I(old_dir), fs_info->generation) || |
9280 | btrfs_inode_in_log(BTRFS_I(new_dir), fs_info->generation) || | |
9281 | btrfs_inode_in_log(BTRFS_I(old_inode), fs_info->generation) || | |
3dc9e8f7 | 9282 | (new_inode && |
0f8939b8 | 9283 | btrfs_inode_in_log(BTRFS_I(new_inode), fs_info->generation))) |
90787766 | 9284 | btrfs_set_log_full_commit(trans); |
3dc9e8f7 FM |
9285 | |
9286 | btrfs_end_log_trans(root); | |
9287 | log_pinned = false; | |
9288 | } | |
d4682ba0 FM |
9289 | if (!ret && sync_log) { |
9290 | ret = btrfs_sync_log(trans, BTRFS_I(old_inode)->root, &ctx); | |
9291 | if (ret) | |
9292 | commit_transaction = true; | |
236ebc20 FM |
9293 | } else if (sync_log) { |
9294 | mutex_lock(&root->log_mutex); | |
9295 | list_del(&ctx.list); | |
9296 | mutex_unlock(&root->log_mutex); | |
d4682ba0 FM |
9297 | } |
9298 | if (commit_transaction) { | |
9299 | ret = btrfs_commit_transaction(trans); | |
9300 | } else { | |
9301 | int ret2; | |
9302 | ||
9303 | ret2 = btrfs_end_transaction(trans); | |
9304 | ret = ret ? ret : ret2; | |
9305 | } | |
b44c59a8 | 9306 | out_notrans: |
33345d01 | 9307 | if (old_ino == BTRFS_FIRST_FREE_OBJECTID) |
0b246afa | 9308 | up_read(&fs_info->subvol_sem); |
9ed74f2d | 9309 | |
39279cc3 CM |
9310 | return ret; |
9311 | } | |
9312 | ||
80ace85c MS |
9313 | static int btrfs_rename2(struct inode *old_dir, struct dentry *old_dentry, |
9314 | struct inode *new_dir, struct dentry *new_dentry, | |
9315 | unsigned int flags) | |
9316 | { | |
cdd1fedf | 9317 | if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT)) |
80ace85c MS |
9318 | return -EINVAL; |
9319 | ||
cdd1fedf DF |
9320 | if (flags & RENAME_EXCHANGE) |
9321 | return btrfs_rename_exchange(old_dir, old_dentry, new_dir, | |
9322 | new_dentry); | |
9323 | ||
9324 | return btrfs_rename(old_dir, old_dentry, new_dir, new_dentry, flags); | |
80ace85c MS |
9325 | } |
9326 | ||
3a2f8c07 NB |
9327 | struct btrfs_delalloc_work { |
9328 | struct inode *inode; | |
9329 | struct completion completion; | |
9330 | struct list_head list; | |
9331 | struct btrfs_work work; | |
9332 | }; | |
9333 | ||
8ccf6f19 MX |
9334 | static void btrfs_run_delalloc_work(struct btrfs_work *work) |
9335 | { | |
9336 | struct btrfs_delalloc_work *delalloc_work; | |
9f23e289 | 9337 | struct inode *inode; |
8ccf6f19 MX |
9338 | |
9339 | delalloc_work = container_of(work, struct btrfs_delalloc_work, | |
9340 | work); | |
9f23e289 | 9341 | inode = delalloc_work->inode; |
30424601 DS |
9342 | filemap_flush(inode->i_mapping); |
9343 | if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT, | |
9344 | &BTRFS_I(inode)->runtime_flags)) | |
9f23e289 | 9345 | filemap_flush(inode->i_mapping); |
8ccf6f19 | 9346 | |
076da91c | 9347 | iput(inode); |
8ccf6f19 MX |
9348 | complete(&delalloc_work->completion); |
9349 | } | |
9350 | ||
3a2f8c07 | 9351 | static struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode) |
8ccf6f19 MX |
9352 | { |
9353 | struct btrfs_delalloc_work *work; | |
9354 | ||
100d5702 | 9355 | work = kmalloc(sizeof(*work), GFP_NOFS); |
8ccf6f19 MX |
9356 | if (!work) |
9357 | return NULL; | |
9358 | ||
9359 | init_completion(&work->completion); | |
9360 | INIT_LIST_HEAD(&work->list); | |
9361 | work->inode = inode; | |
a0cac0ec | 9362 | btrfs_init_work(&work->work, btrfs_run_delalloc_work, NULL, NULL); |
8ccf6f19 MX |
9363 | |
9364 | return work; | |
9365 | } | |
9366 | ||
d352ac68 CM |
9367 | /* |
9368 | * some fairly slow code that needs optimization. This walks the list | |
9369 | * of all the inodes with pending delalloc and forces them to disk. | |
9370 | */ | |
3cd24c69 | 9371 | static int start_delalloc_inodes(struct btrfs_root *root, int nr, bool snapshot) |
ea8c2819 | 9372 | { |
ea8c2819 | 9373 | struct btrfs_inode *binode; |
5b21f2ed | 9374 | struct inode *inode; |
8ccf6f19 MX |
9375 | struct btrfs_delalloc_work *work, *next; |
9376 | struct list_head works; | |
1eafa6c7 | 9377 | struct list_head splice; |
8ccf6f19 | 9378 | int ret = 0; |
ea8c2819 | 9379 | |
8ccf6f19 | 9380 | INIT_LIST_HEAD(&works); |
1eafa6c7 | 9381 | INIT_LIST_HEAD(&splice); |
63607cc8 | 9382 | |
573bfb72 | 9383 | mutex_lock(&root->delalloc_mutex); |
eb73c1b7 MX |
9384 | spin_lock(&root->delalloc_lock); |
9385 | list_splice_init(&root->delalloc_inodes, &splice); | |
1eafa6c7 MX |
9386 | while (!list_empty(&splice)) { |
9387 | binode = list_entry(splice.next, struct btrfs_inode, | |
ea8c2819 | 9388 | delalloc_inodes); |
1eafa6c7 | 9389 | |
eb73c1b7 MX |
9390 | list_move_tail(&binode->delalloc_inodes, |
9391 | &root->delalloc_inodes); | |
5b21f2ed | 9392 | inode = igrab(&binode->vfs_inode); |
df0af1a5 | 9393 | if (!inode) { |
eb73c1b7 | 9394 | cond_resched_lock(&root->delalloc_lock); |
1eafa6c7 | 9395 | continue; |
df0af1a5 | 9396 | } |
eb73c1b7 | 9397 | spin_unlock(&root->delalloc_lock); |
1eafa6c7 | 9398 | |
3cd24c69 EL |
9399 | if (snapshot) |
9400 | set_bit(BTRFS_INODE_SNAPSHOT_FLUSH, | |
9401 | &binode->runtime_flags); | |
076da91c | 9402 | work = btrfs_alloc_delalloc_work(inode); |
5d99a998 | 9403 | if (!work) { |
4fbb5147 | 9404 | iput(inode); |
1eafa6c7 | 9405 | ret = -ENOMEM; |
a1ecaabb | 9406 | goto out; |
5b21f2ed | 9407 | } |
1eafa6c7 | 9408 | list_add_tail(&work->list, &works); |
a44903ab QW |
9409 | btrfs_queue_work(root->fs_info->flush_workers, |
9410 | &work->work); | |
6c255e67 MX |
9411 | ret++; |
9412 | if (nr != -1 && ret >= nr) | |
a1ecaabb | 9413 | goto out; |
5b21f2ed | 9414 | cond_resched(); |
eb73c1b7 | 9415 | spin_lock(&root->delalloc_lock); |
ea8c2819 | 9416 | } |
eb73c1b7 | 9417 | spin_unlock(&root->delalloc_lock); |
8c8bee1d | 9418 | |
a1ecaabb | 9419 | out: |
eb73c1b7 MX |
9420 | list_for_each_entry_safe(work, next, &works, list) { |
9421 | list_del_init(&work->list); | |
40012f96 NB |
9422 | wait_for_completion(&work->completion); |
9423 | kfree(work); | |
eb73c1b7 MX |
9424 | } |
9425 | ||
81f1d390 | 9426 | if (!list_empty(&splice)) { |
eb73c1b7 MX |
9427 | spin_lock(&root->delalloc_lock); |
9428 | list_splice_tail(&splice, &root->delalloc_inodes); | |
9429 | spin_unlock(&root->delalloc_lock); | |
9430 | } | |
573bfb72 | 9431 | mutex_unlock(&root->delalloc_mutex); |
eb73c1b7 MX |
9432 | return ret; |
9433 | } | |
1eafa6c7 | 9434 | |
3cd24c69 | 9435 | int btrfs_start_delalloc_snapshot(struct btrfs_root *root) |
eb73c1b7 | 9436 | { |
0b246afa | 9437 | struct btrfs_fs_info *fs_info = root->fs_info; |
eb73c1b7 | 9438 | int ret; |
1eafa6c7 | 9439 | |
0b246afa | 9440 | if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) |
eb73c1b7 MX |
9441 | return -EROFS; |
9442 | ||
3cd24c69 | 9443 | ret = start_delalloc_inodes(root, -1, true); |
6c255e67 MX |
9444 | if (ret > 0) |
9445 | ret = 0; | |
eb73c1b7 MX |
9446 | return ret; |
9447 | } | |
9448 | ||
82b3e53b | 9449 | int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, int nr) |
eb73c1b7 MX |
9450 | { |
9451 | struct btrfs_root *root; | |
9452 | struct list_head splice; | |
9453 | int ret; | |
9454 | ||
2c21b4d7 | 9455 | if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) |
eb73c1b7 MX |
9456 | return -EROFS; |
9457 | ||
9458 | INIT_LIST_HEAD(&splice); | |
9459 | ||
573bfb72 | 9460 | mutex_lock(&fs_info->delalloc_root_mutex); |
eb73c1b7 MX |
9461 | spin_lock(&fs_info->delalloc_root_lock); |
9462 | list_splice_init(&fs_info->delalloc_roots, &splice); | |
6c255e67 | 9463 | while (!list_empty(&splice) && nr) { |
eb73c1b7 MX |
9464 | root = list_first_entry(&splice, struct btrfs_root, |
9465 | delalloc_root); | |
00246528 | 9466 | root = btrfs_grab_root(root); |
eb73c1b7 MX |
9467 | BUG_ON(!root); |
9468 | list_move_tail(&root->delalloc_root, | |
9469 | &fs_info->delalloc_roots); | |
9470 | spin_unlock(&fs_info->delalloc_root_lock); | |
9471 | ||
3cd24c69 | 9472 | ret = start_delalloc_inodes(root, nr, false); |
00246528 | 9473 | btrfs_put_root(root); |
6c255e67 | 9474 | if (ret < 0) |
eb73c1b7 MX |
9475 | goto out; |
9476 | ||
6c255e67 MX |
9477 | if (nr != -1) { |
9478 | nr -= ret; | |
9479 | WARN_ON(nr < 0); | |
9480 | } | |
eb73c1b7 | 9481 | spin_lock(&fs_info->delalloc_root_lock); |
8ccf6f19 | 9482 | } |
eb73c1b7 | 9483 | spin_unlock(&fs_info->delalloc_root_lock); |
1eafa6c7 | 9484 | |
6c255e67 | 9485 | ret = 0; |
eb73c1b7 | 9486 | out: |
81f1d390 | 9487 | if (!list_empty(&splice)) { |
eb73c1b7 MX |
9488 | spin_lock(&fs_info->delalloc_root_lock); |
9489 | list_splice_tail(&splice, &fs_info->delalloc_roots); | |
9490 | spin_unlock(&fs_info->delalloc_root_lock); | |
1eafa6c7 | 9491 | } |
573bfb72 | 9492 | mutex_unlock(&fs_info->delalloc_root_mutex); |
8ccf6f19 | 9493 | return ret; |
ea8c2819 CM |
9494 | } |
9495 | ||
39279cc3 CM |
9496 | static int btrfs_symlink(struct inode *dir, struct dentry *dentry, |
9497 | const char *symname) | |
9498 | { | |
0b246afa | 9499 | struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb); |
39279cc3 CM |
9500 | struct btrfs_trans_handle *trans; |
9501 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
9502 | struct btrfs_path *path; | |
9503 | struct btrfs_key key; | |
1832a6d5 | 9504 | struct inode *inode = NULL; |
39279cc3 | 9505 | int err; |
39279cc3 | 9506 | u64 objectid; |
67871254 | 9507 | u64 index = 0; |
39279cc3 CM |
9508 | int name_len; |
9509 | int datasize; | |
5f39d397 | 9510 | unsigned long ptr; |
39279cc3 | 9511 | struct btrfs_file_extent_item *ei; |
5f39d397 | 9512 | struct extent_buffer *leaf; |
39279cc3 | 9513 | |
f06becc4 | 9514 | name_len = strlen(symname); |
0b246afa | 9515 | if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info)) |
39279cc3 | 9516 | return -ENAMETOOLONG; |
1832a6d5 | 9517 | |
9ed74f2d JB |
9518 | /* |
9519 | * 2 items for inode item and ref | |
9520 | * 2 items for dir items | |
9269d12b FM |
9521 | * 1 item for updating parent inode item |
9522 | * 1 item for the inline extent item | |
9ed74f2d JB |
9523 | * 1 item for xattr if selinux is on |
9524 | */ | |
9269d12b | 9525 | trans = btrfs_start_transaction(root, 7); |
a22285a6 YZ |
9526 | if (IS_ERR(trans)) |
9527 | return PTR_ERR(trans); | |
1832a6d5 | 9528 | |
581bb050 LZ |
9529 | err = btrfs_find_free_ino(root, &objectid); |
9530 | if (err) | |
9531 | goto out_unlock; | |
9532 | ||
aec7477b | 9533 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
f85b7379 DS |
9534 | dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), |
9535 | objectid, S_IFLNK|S_IRWXUGO, &index); | |
7cf96da3 TI |
9536 | if (IS_ERR(inode)) { |
9537 | err = PTR_ERR(inode); | |
32955c54 | 9538 | inode = NULL; |
39279cc3 | 9539 | goto out_unlock; |
7cf96da3 | 9540 | } |
39279cc3 | 9541 | |
ad19db71 CS |
9542 | /* |
9543 | * If the active LSM wants to access the inode during | |
9544 | * d_instantiate it needs these. Smack checks to see | |
9545 | * if the filesystem supports xattrs by looking at the | |
9546 | * ops vector. | |
9547 | */ | |
9548 | inode->i_fop = &btrfs_file_operations; | |
9549 | inode->i_op = &btrfs_file_inode_operations; | |
b0d5d10f | 9550 | inode->i_mapping->a_ops = &btrfs_aops; |
b0d5d10f CM |
9551 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; |
9552 | ||
9553 | err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name); | |
9554 | if (err) | |
32955c54 | 9555 | goto out_unlock; |
ad19db71 | 9556 | |
39279cc3 | 9557 | path = btrfs_alloc_path(); |
d8926bb3 MF |
9558 | if (!path) { |
9559 | err = -ENOMEM; | |
32955c54 | 9560 | goto out_unlock; |
d8926bb3 | 9561 | } |
4a0cc7ca | 9562 | key.objectid = btrfs_ino(BTRFS_I(inode)); |
39279cc3 | 9563 | key.offset = 0; |
962a298f | 9564 | key.type = BTRFS_EXTENT_DATA_KEY; |
39279cc3 CM |
9565 | datasize = btrfs_file_extent_calc_inline_size(name_len); |
9566 | err = btrfs_insert_empty_item(trans, root, path, &key, | |
9567 | datasize); | |
54aa1f4d | 9568 | if (err) { |
b0839166 | 9569 | btrfs_free_path(path); |
32955c54 | 9570 | goto out_unlock; |
54aa1f4d | 9571 | } |
5f39d397 CM |
9572 | leaf = path->nodes[0]; |
9573 | ei = btrfs_item_ptr(leaf, path->slots[0], | |
9574 | struct btrfs_file_extent_item); | |
9575 | btrfs_set_file_extent_generation(leaf, ei, trans->transid); | |
9576 | btrfs_set_file_extent_type(leaf, ei, | |
39279cc3 | 9577 | BTRFS_FILE_EXTENT_INLINE); |
c8b97818 CM |
9578 | btrfs_set_file_extent_encryption(leaf, ei, 0); |
9579 | btrfs_set_file_extent_compression(leaf, ei, 0); | |
9580 | btrfs_set_file_extent_other_encoding(leaf, ei, 0); | |
9581 | btrfs_set_file_extent_ram_bytes(leaf, ei, name_len); | |
9582 | ||
39279cc3 | 9583 | ptr = btrfs_file_extent_inline_start(ei); |
5f39d397 CM |
9584 | write_extent_buffer(leaf, symname, ptr, name_len); |
9585 | btrfs_mark_buffer_dirty(leaf); | |
39279cc3 | 9586 | btrfs_free_path(path); |
5f39d397 | 9587 | |
39279cc3 | 9588 | inode->i_op = &btrfs_symlink_inode_operations; |
21fc61c7 | 9589 | inode_nohighmem(inode); |
d899e052 | 9590 | inode_set_bytes(inode, name_len); |
6ef06d27 | 9591 | btrfs_i_size_write(BTRFS_I(inode), name_len); |
54aa1f4d | 9592 | err = btrfs_update_inode(trans, root, inode); |
d50866d0 FM |
9593 | /* |
9594 | * Last step, add directory indexes for our symlink inode. This is the | |
9595 | * last step to avoid extra cleanup of these indexes if an error happens | |
9596 | * elsewhere above. | |
9597 | */ | |
9598 | if (!err) | |
cef415af NB |
9599 | err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, |
9600 | BTRFS_I(inode), 0, index); | |
32955c54 AV |
9601 | if (err) |
9602 | goto out_unlock; | |
b0d5d10f | 9603 | |
1e2e547a | 9604 | d_instantiate_new(dentry, inode); |
39279cc3 CM |
9605 | |
9606 | out_unlock: | |
3a45bb20 | 9607 | btrfs_end_transaction(trans); |
32955c54 | 9608 | if (err && inode) { |
39279cc3 | 9609 | inode_dec_link_count(inode); |
32955c54 | 9610 | discard_new_inode(inode); |
39279cc3 | 9611 | } |
2ff7e61e | 9612 | btrfs_btree_balance_dirty(fs_info); |
39279cc3 CM |
9613 | return err; |
9614 | } | |
16432985 | 9615 | |
203f44c5 QW |
9616 | static int insert_prealloc_file_extent(struct btrfs_trans_handle *trans, |
9617 | struct inode *inode, struct btrfs_key *ins, | |
9618 | u64 file_offset) | |
9619 | { | |
9620 | struct btrfs_file_extent_item stack_fi; | |
9621 | u64 start = ins->objectid; | |
9622 | u64 len = ins->offset; | |
9729f10a | 9623 | int ret; |
203f44c5 QW |
9624 | |
9625 | memset(&stack_fi, 0, sizeof(stack_fi)); | |
9626 | ||
9627 | btrfs_set_stack_file_extent_type(&stack_fi, BTRFS_FILE_EXTENT_PREALLOC); | |
9628 | btrfs_set_stack_file_extent_disk_bytenr(&stack_fi, start); | |
9629 | btrfs_set_stack_file_extent_disk_num_bytes(&stack_fi, len); | |
9630 | btrfs_set_stack_file_extent_num_bytes(&stack_fi, len); | |
9631 | btrfs_set_stack_file_extent_ram_bytes(&stack_fi, len); | |
9632 | btrfs_set_stack_file_extent_compression(&stack_fi, BTRFS_COMPRESS_NONE); | |
9633 | /* Encryption and other encoding is reserved and all 0 */ | |
9634 | ||
72b7d15b | 9635 | ret = btrfs_qgroup_release_data(BTRFS_I(inode), file_offset, len); |
9729f10a QW |
9636 | if (ret < 0) |
9637 | return ret; | |
c553f94d | 9638 | return insert_reserved_file_extent(trans, BTRFS_I(inode), file_offset, |
9729f10a | 9639 | &stack_fi, ret); |
203f44c5 | 9640 | } |
0af3d00b JB |
9641 | static int __btrfs_prealloc_file_range(struct inode *inode, int mode, |
9642 | u64 start, u64 num_bytes, u64 min_size, | |
9643 | loff_t actual_len, u64 *alloc_hint, | |
9644 | struct btrfs_trans_handle *trans) | |
d899e052 | 9645 | { |
0b246afa | 9646 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
5dc562c5 JB |
9647 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; |
9648 | struct extent_map *em; | |
d899e052 YZ |
9649 | struct btrfs_root *root = BTRFS_I(inode)->root; |
9650 | struct btrfs_key ins; | |
d899e052 | 9651 | u64 cur_offset = start; |
b778cf96 | 9652 | u64 clear_offset = start; |
55a61d1d | 9653 | u64 i_size; |
154ea289 | 9654 | u64 cur_bytes; |
0b670dc4 | 9655 | u64 last_alloc = (u64)-1; |
d899e052 | 9656 | int ret = 0; |
0af3d00b | 9657 | bool own_trans = true; |
18513091 | 9658 | u64 end = start + num_bytes - 1; |
d899e052 | 9659 | |
0af3d00b JB |
9660 | if (trans) |
9661 | own_trans = false; | |
d899e052 | 9662 | while (num_bytes > 0) { |
0af3d00b JB |
9663 | if (own_trans) { |
9664 | trans = btrfs_start_transaction(root, 3); | |
9665 | if (IS_ERR(trans)) { | |
9666 | ret = PTR_ERR(trans); | |
9667 | break; | |
9668 | } | |
5a303d5d YZ |
9669 | } |
9670 | ||
ee22184b | 9671 | cur_bytes = min_t(u64, num_bytes, SZ_256M); |
154ea289 | 9672 | cur_bytes = max(cur_bytes, min_size); |
0b670dc4 JB |
9673 | /* |
9674 | * If we are severely fragmented we could end up with really | |
9675 | * small allocations, so if the allocator is returning small | |
9676 | * chunks lets make its job easier by only searching for those | |
9677 | * sized chunks. | |
9678 | */ | |
9679 | cur_bytes = min(cur_bytes, last_alloc); | |
18513091 WX |
9680 | ret = btrfs_reserve_extent(root, cur_bytes, cur_bytes, |
9681 | min_size, 0, *alloc_hint, &ins, 1, 0); | |
5a303d5d | 9682 | if (ret) { |
0af3d00b | 9683 | if (own_trans) |
3a45bb20 | 9684 | btrfs_end_transaction(trans); |
a22285a6 | 9685 | break; |
d899e052 | 9686 | } |
b778cf96 JB |
9687 | |
9688 | /* | |
9689 | * We've reserved this space, and thus converted it from | |
9690 | * ->bytes_may_use to ->bytes_reserved. Any error that happens | |
9691 | * from here on out we will only need to clear our reservation | |
9692 | * for the remaining unreserved area, so advance our | |
9693 | * clear_offset by our extent size. | |
9694 | */ | |
9695 | clear_offset += ins.offset; | |
0b246afa | 9696 | btrfs_dec_block_group_reservations(fs_info, ins.objectid); |
5a303d5d | 9697 | |
0b670dc4 | 9698 | last_alloc = ins.offset; |
c553f94d | 9699 | ret = insert_prealloc_file_extent(trans, inode, &ins, cur_offset); |
79787eaa | 9700 | if (ret) { |
2ff7e61e | 9701 | btrfs_free_reserved_extent(fs_info, ins.objectid, |
e570fd27 | 9702 | ins.offset, 0); |
66642832 | 9703 | btrfs_abort_transaction(trans, ret); |
79787eaa | 9704 | if (own_trans) |
3a45bb20 | 9705 | btrfs_end_transaction(trans); |
79787eaa JM |
9706 | break; |
9707 | } | |
31193213 | 9708 | |
dcdbc059 | 9709 | btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset, |
a1ed835e | 9710 | cur_offset + ins.offset -1, 0); |
5a303d5d | 9711 | |
5dc562c5 JB |
9712 | em = alloc_extent_map(); |
9713 | if (!em) { | |
9714 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, | |
9715 | &BTRFS_I(inode)->runtime_flags); | |
9716 | goto next; | |
9717 | } | |
9718 | ||
9719 | em->start = cur_offset; | |
9720 | em->orig_start = cur_offset; | |
9721 | em->len = ins.offset; | |
9722 | em->block_start = ins.objectid; | |
9723 | em->block_len = ins.offset; | |
b4939680 | 9724 | em->orig_block_len = ins.offset; |
cc95bef6 | 9725 | em->ram_bytes = ins.offset; |
5dc562c5 JB |
9726 | set_bit(EXTENT_FLAG_PREALLOC, &em->flags); |
9727 | em->generation = trans->transid; | |
9728 | ||
9729 | while (1) { | |
9730 | write_lock(&em_tree->lock); | |
09a2a8f9 | 9731 | ret = add_extent_mapping(em_tree, em, 1); |
5dc562c5 JB |
9732 | write_unlock(&em_tree->lock); |
9733 | if (ret != -EEXIST) | |
9734 | break; | |
dcdbc059 | 9735 | btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset, |
5dc562c5 JB |
9736 | cur_offset + ins.offset - 1, |
9737 | 0); | |
9738 | } | |
9739 | free_extent_map(em); | |
9740 | next: | |
d899e052 YZ |
9741 | num_bytes -= ins.offset; |
9742 | cur_offset += ins.offset; | |
efa56464 | 9743 | *alloc_hint = ins.objectid + ins.offset; |
5a303d5d | 9744 | |
0c4d2d95 | 9745 | inode_inc_iversion(inode); |
c2050a45 | 9746 | inode->i_ctime = current_time(inode); |
6cbff00f | 9747 | BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC; |
d899e052 | 9748 | if (!(mode & FALLOC_FL_KEEP_SIZE) && |
efa56464 YZ |
9749 | (actual_len > inode->i_size) && |
9750 | (cur_offset > inode->i_size)) { | |
d1ea6a61 | 9751 | if (cur_offset > actual_len) |
55a61d1d | 9752 | i_size = actual_len; |
d1ea6a61 | 9753 | else |
55a61d1d JB |
9754 | i_size = cur_offset; |
9755 | i_size_write(inode, i_size); | |
d923afe9 | 9756 | btrfs_inode_safe_disk_i_size_write(inode, 0); |
5a303d5d YZ |
9757 | } |
9758 | ||
d899e052 | 9759 | ret = btrfs_update_inode(trans, root, inode); |
79787eaa JM |
9760 | |
9761 | if (ret) { | |
66642832 | 9762 | btrfs_abort_transaction(trans, ret); |
79787eaa | 9763 | if (own_trans) |
3a45bb20 | 9764 | btrfs_end_transaction(trans); |
79787eaa JM |
9765 | break; |
9766 | } | |
d899e052 | 9767 | |
0af3d00b | 9768 | if (own_trans) |
3a45bb20 | 9769 | btrfs_end_transaction(trans); |
5a303d5d | 9770 | } |
b778cf96 | 9771 | if (clear_offset < end) |
25ce28ca | 9772 | btrfs_free_reserved_data_space(BTRFS_I(inode), NULL, clear_offset, |
b778cf96 | 9773 | end - clear_offset + 1); |
d899e052 YZ |
9774 | return ret; |
9775 | } | |
9776 | ||
0af3d00b JB |
9777 | int btrfs_prealloc_file_range(struct inode *inode, int mode, |
9778 | u64 start, u64 num_bytes, u64 min_size, | |
9779 | loff_t actual_len, u64 *alloc_hint) | |
9780 | { | |
9781 | return __btrfs_prealloc_file_range(inode, mode, start, num_bytes, | |
9782 | min_size, actual_len, alloc_hint, | |
9783 | NULL); | |
9784 | } | |
9785 | ||
9786 | int btrfs_prealloc_file_range_trans(struct inode *inode, | |
9787 | struct btrfs_trans_handle *trans, int mode, | |
9788 | u64 start, u64 num_bytes, u64 min_size, | |
9789 | loff_t actual_len, u64 *alloc_hint) | |
9790 | { | |
9791 | return __btrfs_prealloc_file_range(inode, mode, start, num_bytes, | |
9792 | min_size, actual_len, alloc_hint, trans); | |
9793 | } | |
9794 | ||
e6dcd2dc CM |
9795 | static int btrfs_set_page_dirty(struct page *page) |
9796 | { | |
e6dcd2dc CM |
9797 | return __set_page_dirty_nobuffers(page); |
9798 | } | |
9799 | ||
10556cb2 | 9800 | static int btrfs_permission(struct inode *inode, int mask) |
fdebe2bd | 9801 | { |
b83cc969 | 9802 | struct btrfs_root *root = BTRFS_I(inode)->root; |
cb6db4e5 | 9803 | umode_t mode = inode->i_mode; |
b83cc969 | 9804 | |
cb6db4e5 JM |
9805 | if (mask & MAY_WRITE && |
9806 | (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) { | |
9807 | if (btrfs_root_readonly(root)) | |
9808 | return -EROFS; | |
9809 | if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY) | |
9810 | return -EACCES; | |
9811 | } | |
2830ba7f | 9812 | return generic_permission(inode, mask); |
fdebe2bd | 9813 | } |
39279cc3 | 9814 | |
ef3b9af5 FM |
9815 | static int btrfs_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode) |
9816 | { | |
2ff7e61e | 9817 | struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb); |
ef3b9af5 FM |
9818 | struct btrfs_trans_handle *trans; |
9819 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
9820 | struct inode *inode = NULL; | |
9821 | u64 objectid; | |
9822 | u64 index; | |
9823 | int ret = 0; | |
9824 | ||
9825 | /* | |
9826 | * 5 units required for adding orphan entry | |
9827 | */ | |
9828 | trans = btrfs_start_transaction(root, 5); | |
9829 | if (IS_ERR(trans)) | |
9830 | return PTR_ERR(trans); | |
9831 | ||
9832 | ret = btrfs_find_free_ino(root, &objectid); | |
9833 | if (ret) | |
9834 | goto out; | |
9835 | ||
9836 | inode = btrfs_new_inode(trans, root, dir, NULL, 0, | |
f85b7379 | 9837 | btrfs_ino(BTRFS_I(dir)), objectid, mode, &index); |
ef3b9af5 FM |
9838 | if (IS_ERR(inode)) { |
9839 | ret = PTR_ERR(inode); | |
9840 | inode = NULL; | |
9841 | goto out; | |
9842 | } | |
9843 | ||
ef3b9af5 FM |
9844 | inode->i_fop = &btrfs_file_operations; |
9845 | inode->i_op = &btrfs_file_inode_operations; | |
9846 | ||
9847 | inode->i_mapping->a_ops = &btrfs_aops; | |
ef3b9af5 FM |
9848 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; |
9849 | ||
b0d5d10f CM |
9850 | ret = btrfs_init_inode_security(trans, inode, dir, NULL); |
9851 | if (ret) | |
32955c54 | 9852 | goto out; |
b0d5d10f CM |
9853 | |
9854 | ret = btrfs_update_inode(trans, root, inode); | |
9855 | if (ret) | |
32955c54 | 9856 | goto out; |
73f2e545 | 9857 | ret = btrfs_orphan_add(trans, BTRFS_I(inode)); |
ef3b9af5 | 9858 | if (ret) |
32955c54 | 9859 | goto out; |
ef3b9af5 | 9860 | |
5762b5c9 FM |
9861 | /* |
9862 | * We set number of links to 0 in btrfs_new_inode(), and here we set | |
9863 | * it to 1 because d_tmpfile() will issue a warning if the count is 0, | |
9864 | * through: | |
9865 | * | |
9866 | * d_tmpfile() -> inode_dec_link_count() -> drop_nlink() | |
9867 | */ | |
9868 | set_nlink(inode, 1); | |
ef3b9af5 | 9869 | d_tmpfile(dentry, inode); |
32955c54 | 9870 | unlock_new_inode(inode); |
ef3b9af5 | 9871 | mark_inode_dirty(inode); |
ef3b9af5 | 9872 | out: |
3a45bb20 | 9873 | btrfs_end_transaction(trans); |
32955c54 AV |
9874 | if (ret && inode) |
9875 | discard_new_inode(inode); | |
2ff7e61e | 9876 | btrfs_btree_balance_dirty(fs_info); |
ef3b9af5 FM |
9877 | return ret; |
9878 | } | |
9879 | ||
5cdc84bf | 9880 | void btrfs_set_range_writeback(struct extent_io_tree *tree, u64 start, u64 end) |
c6100a4b | 9881 | { |
5cdc84bf | 9882 | struct inode *inode = tree->private_data; |
c6100a4b JB |
9883 | unsigned long index = start >> PAGE_SHIFT; |
9884 | unsigned long end_index = end >> PAGE_SHIFT; | |
9885 | struct page *page; | |
9886 | ||
9887 | while (index <= end_index) { | |
9888 | page = find_get_page(inode->i_mapping, index); | |
9889 | ASSERT(page); /* Pages should be in the extent_io_tree */ | |
9890 | set_page_writeback(page); | |
9891 | put_page(page); | |
9892 | index++; | |
9893 | } | |
9894 | } | |
9895 | ||
ed46ff3d OS |
9896 | #ifdef CONFIG_SWAP |
9897 | /* | |
9898 | * Add an entry indicating a block group or device which is pinned by a | |
9899 | * swapfile. Returns 0 on success, 1 if there is already an entry for it, or a | |
9900 | * negative errno on failure. | |
9901 | */ | |
9902 | static int btrfs_add_swapfile_pin(struct inode *inode, void *ptr, | |
9903 | bool is_block_group) | |
9904 | { | |
9905 | struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info; | |
9906 | struct btrfs_swapfile_pin *sp, *entry; | |
9907 | struct rb_node **p; | |
9908 | struct rb_node *parent = NULL; | |
9909 | ||
9910 | sp = kmalloc(sizeof(*sp), GFP_NOFS); | |
9911 | if (!sp) | |
9912 | return -ENOMEM; | |
9913 | sp->ptr = ptr; | |
9914 | sp->inode = inode; | |
9915 | sp->is_block_group = is_block_group; | |
9916 | ||
9917 | spin_lock(&fs_info->swapfile_pins_lock); | |
9918 | p = &fs_info->swapfile_pins.rb_node; | |
9919 | while (*p) { | |
9920 | parent = *p; | |
9921 | entry = rb_entry(parent, struct btrfs_swapfile_pin, node); | |
9922 | if (sp->ptr < entry->ptr || | |
9923 | (sp->ptr == entry->ptr && sp->inode < entry->inode)) { | |
9924 | p = &(*p)->rb_left; | |
9925 | } else if (sp->ptr > entry->ptr || | |
9926 | (sp->ptr == entry->ptr && sp->inode > entry->inode)) { | |
9927 | p = &(*p)->rb_right; | |
9928 | } else { | |
9929 | spin_unlock(&fs_info->swapfile_pins_lock); | |
9930 | kfree(sp); | |
9931 | return 1; | |
9932 | } | |
9933 | } | |
9934 | rb_link_node(&sp->node, parent, p); | |
9935 | rb_insert_color(&sp->node, &fs_info->swapfile_pins); | |
9936 | spin_unlock(&fs_info->swapfile_pins_lock); | |
9937 | return 0; | |
9938 | } | |
9939 | ||
9940 | /* Free all of the entries pinned by this swapfile. */ | |
9941 | static void btrfs_free_swapfile_pins(struct inode *inode) | |
9942 | { | |
9943 | struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info; | |
9944 | struct btrfs_swapfile_pin *sp; | |
9945 | struct rb_node *node, *next; | |
9946 | ||
9947 | spin_lock(&fs_info->swapfile_pins_lock); | |
9948 | node = rb_first(&fs_info->swapfile_pins); | |
9949 | while (node) { | |
9950 | next = rb_next(node); | |
9951 | sp = rb_entry(node, struct btrfs_swapfile_pin, node); | |
9952 | if (sp->inode == inode) { | |
9953 | rb_erase(&sp->node, &fs_info->swapfile_pins); | |
9954 | if (sp->is_block_group) | |
9955 | btrfs_put_block_group(sp->ptr); | |
9956 | kfree(sp); | |
9957 | } | |
9958 | node = next; | |
9959 | } | |
9960 | spin_unlock(&fs_info->swapfile_pins_lock); | |
9961 | } | |
9962 | ||
9963 | struct btrfs_swap_info { | |
9964 | u64 start; | |
9965 | u64 block_start; | |
9966 | u64 block_len; | |
9967 | u64 lowest_ppage; | |
9968 | u64 highest_ppage; | |
9969 | unsigned long nr_pages; | |
9970 | int nr_extents; | |
9971 | }; | |
9972 | ||
9973 | static int btrfs_add_swap_extent(struct swap_info_struct *sis, | |
9974 | struct btrfs_swap_info *bsi) | |
9975 | { | |
9976 | unsigned long nr_pages; | |
9977 | u64 first_ppage, first_ppage_reported, next_ppage; | |
9978 | int ret; | |
9979 | ||
9980 | first_ppage = ALIGN(bsi->block_start, PAGE_SIZE) >> PAGE_SHIFT; | |
9981 | next_ppage = ALIGN_DOWN(bsi->block_start + bsi->block_len, | |
9982 | PAGE_SIZE) >> PAGE_SHIFT; | |
9983 | ||
9984 | if (first_ppage >= next_ppage) | |
9985 | return 0; | |
9986 | nr_pages = next_ppage - first_ppage; | |
9987 | ||
9988 | first_ppage_reported = first_ppage; | |
9989 | if (bsi->start == 0) | |
9990 | first_ppage_reported++; | |
9991 | if (bsi->lowest_ppage > first_ppage_reported) | |
9992 | bsi->lowest_ppage = first_ppage_reported; | |
9993 | if (bsi->highest_ppage < (next_ppage - 1)) | |
9994 | bsi->highest_ppage = next_ppage - 1; | |
9995 | ||
9996 | ret = add_swap_extent(sis, bsi->nr_pages, nr_pages, first_ppage); | |
9997 | if (ret < 0) | |
9998 | return ret; | |
9999 | bsi->nr_extents += ret; | |
10000 | bsi->nr_pages += nr_pages; | |
10001 | return 0; | |
10002 | } | |
10003 | ||
10004 | static void btrfs_swap_deactivate(struct file *file) | |
10005 | { | |
10006 | struct inode *inode = file_inode(file); | |
10007 | ||
10008 | btrfs_free_swapfile_pins(inode); | |
10009 | atomic_dec(&BTRFS_I(inode)->root->nr_swapfiles); | |
10010 | } | |
10011 | ||
10012 | static int btrfs_swap_activate(struct swap_info_struct *sis, struct file *file, | |
10013 | sector_t *span) | |
10014 | { | |
10015 | struct inode *inode = file_inode(file); | |
10016 | struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info; | |
10017 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; | |
10018 | struct extent_state *cached_state = NULL; | |
10019 | struct extent_map *em = NULL; | |
10020 | struct btrfs_device *device = NULL; | |
10021 | struct btrfs_swap_info bsi = { | |
10022 | .lowest_ppage = (sector_t)-1ULL, | |
10023 | }; | |
10024 | int ret = 0; | |
10025 | u64 isize; | |
10026 | u64 start; | |
10027 | ||
10028 | /* | |
10029 | * If the swap file was just created, make sure delalloc is done. If the | |
10030 | * file changes again after this, the user is doing something stupid and | |
10031 | * we don't really care. | |
10032 | */ | |
10033 | ret = btrfs_wait_ordered_range(inode, 0, (u64)-1); | |
10034 | if (ret) | |
10035 | return ret; | |
10036 | ||
10037 | /* | |
10038 | * The inode is locked, so these flags won't change after we check them. | |
10039 | */ | |
10040 | if (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS) { | |
10041 | btrfs_warn(fs_info, "swapfile must not be compressed"); | |
10042 | return -EINVAL; | |
10043 | } | |
10044 | if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW)) { | |
10045 | btrfs_warn(fs_info, "swapfile must not be copy-on-write"); | |
10046 | return -EINVAL; | |
10047 | } | |
10048 | if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) { | |
10049 | btrfs_warn(fs_info, "swapfile must not be checksummed"); | |
10050 | return -EINVAL; | |
10051 | } | |
10052 | ||
10053 | /* | |
10054 | * Balance or device remove/replace/resize can move stuff around from | |
10055 | * under us. The EXCL_OP flag makes sure they aren't running/won't run | |
10056 | * concurrently while we are mapping the swap extents, and | |
10057 | * fs_info->swapfile_pins prevents them from running while the swap file | |
10058 | * is active and moving the extents. Note that this also prevents a | |
10059 | * concurrent device add which isn't actually necessary, but it's not | |
10060 | * really worth the trouble to allow it. | |
10061 | */ | |
10062 | if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) { | |
10063 | btrfs_warn(fs_info, | |
10064 | "cannot activate swapfile while exclusive operation is running"); | |
10065 | return -EBUSY; | |
10066 | } | |
10067 | /* | |
10068 | * Snapshots can create extents which require COW even if NODATACOW is | |
10069 | * set. We use this counter to prevent snapshots. We must increment it | |
10070 | * before walking the extents because we don't want a concurrent | |
10071 | * snapshot to run after we've already checked the extents. | |
10072 | */ | |
10073 | atomic_inc(&BTRFS_I(inode)->root->nr_swapfiles); | |
10074 | ||
10075 | isize = ALIGN_DOWN(inode->i_size, fs_info->sectorsize); | |
10076 | ||
10077 | lock_extent_bits(io_tree, 0, isize - 1, &cached_state); | |
10078 | start = 0; | |
10079 | while (start < isize) { | |
10080 | u64 logical_block_start, physical_block_start; | |
32da5386 | 10081 | struct btrfs_block_group *bg; |
ed46ff3d OS |
10082 | u64 len = isize - start; |
10083 | ||
39b07b5d | 10084 | em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len); |
ed46ff3d OS |
10085 | if (IS_ERR(em)) { |
10086 | ret = PTR_ERR(em); | |
10087 | goto out; | |
10088 | } | |
10089 | ||
10090 | if (em->block_start == EXTENT_MAP_HOLE) { | |
10091 | btrfs_warn(fs_info, "swapfile must not have holes"); | |
10092 | ret = -EINVAL; | |
10093 | goto out; | |
10094 | } | |
10095 | if (em->block_start == EXTENT_MAP_INLINE) { | |
10096 | /* | |
10097 | * It's unlikely we'll ever actually find ourselves | |
10098 | * here, as a file small enough to fit inline won't be | |
10099 | * big enough to store more than the swap header, but in | |
10100 | * case something changes in the future, let's catch it | |
10101 | * here rather than later. | |
10102 | */ | |
10103 | btrfs_warn(fs_info, "swapfile must not be inline"); | |
10104 | ret = -EINVAL; | |
10105 | goto out; | |
10106 | } | |
10107 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) { | |
10108 | btrfs_warn(fs_info, "swapfile must not be compressed"); | |
10109 | ret = -EINVAL; | |
10110 | goto out; | |
10111 | } | |
10112 | ||
10113 | logical_block_start = em->block_start + (start - em->start); | |
10114 | len = min(len, em->len - (start - em->start)); | |
10115 | free_extent_map(em); | |
10116 | em = NULL; | |
10117 | ||
10118 | ret = can_nocow_extent(inode, start, &len, NULL, NULL, NULL); | |
10119 | if (ret < 0) { | |
10120 | goto out; | |
10121 | } else if (ret) { | |
10122 | ret = 0; | |
10123 | } else { | |
10124 | btrfs_warn(fs_info, | |
10125 | "swapfile must not be copy-on-write"); | |
10126 | ret = -EINVAL; | |
10127 | goto out; | |
10128 | } | |
10129 | ||
10130 | em = btrfs_get_chunk_map(fs_info, logical_block_start, len); | |
10131 | if (IS_ERR(em)) { | |
10132 | ret = PTR_ERR(em); | |
10133 | goto out; | |
10134 | } | |
10135 | ||
10136 | if (em->map_lookup->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) { | |
10137 | btrfs_warn(fs_info, | |
10138 | "swapfile must have single data profile"); | |
10139 | ret = -EINVAL; | |
10140 | goto out; | |
10141 | } | |
10142 | ||
10143 | if (device == NULL) { | |
10144 | device = em->map_lookup->stripes[0].dev; | |
10145 | ret = btrfs_add_swapfile_pin(inode, device, false); | |
10146 | if (ret == 1) | |
10147 | ret = 0; | |
10148 | else if (ret) | |
10149 | goto out; | |
10150 | } else if (device != em->map_lookup->stripes[0].dev) { | |
10151 | btrfs_warn(fs_info, "swapfile must be on one device"); | |
10152 | ret = -EINVAL; | |
10153 | goto out; | |
10154 | } | |
10155 | ||
10156 | physical_block_start = (em->map_lookup->stripes[0].physical + | |
10157 | (logical_block_start - em->start)); | |
10158 | len = min(len, em->len - (logical_block_start - em->start)); | |
10159 | free_extent_map(em); | |
10160 | em = NULL; | |
10161 | ||
10162 | bg = btrfs_lookup_block_group(fs_info, logical_block_start); | |
10163 | if (!bg) { | |
10164 | btrfs_warn(fs_info, | |
10165 | "could not find block group containing swapfile"); | |
10166 | ret = -EINVAL; | |
10167 | goto out; | |
10168 | } | |
10169 | ||
10170 | ret = btrfs_add_swapfile_pin(inode, bg, true); | |
10171 | if (ret) { | |
10172 | btrfs_put_block_group(bg); | |
10173 | if (ret == 1) | |
10174 | ret = 0; | |
10175 | else | |
10176 | goto out; | |
10177 | } | |
10178 | ||
10179 | if (bsi.block_len && | |
10180 | bsi.block_start + bsi.block_len == physical_block_start) { | |
10181 | bsi.block_len += len; | |
10182 | } else { | |
10183 | if (bsi.block_len) { | |
10184 | ret = btrfs_add_swap_extent(sis, &bsi); | |
10185 | if (ret) | |
10186 | goto out; | |
10187 | } | |
10188 | bsi.start = start; | |
10189 | bsi.block_start = physical_block_start; | |
10190 | bsi.block_len = len; | |
10191 | } | |
10192 | ||
10193 | start += len; | |
10194 | } | |
10195 | ||
10196 | if (bsi.block_len) | |
10197 | ret = btrfs_add_swap_extent(sis, &bsi); | |
10198 | ||
10199 | out: | |
10200 | if (!IS_ERR_OR_NULL(em)) | |
10201 | free_extent_map(em); | |
10202 | ||
10203 | unlock_extent_cached(io_tree, 0, isize - 1, &cached_state); | |
10204 | ||
10205 | if (ret) | |
10206 | btrfs_swap_deactivate(file); | |
10207 | ||
10208 | clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags); | |
10209 | ||
10210 | if (ret) | |
10211 | return ret; | |
10212 | ||
10213 | if (device) | |
10214 | sis->bdev = device->bdev; | |
10215 | *span = bsi.highest_ppage - bsi.lowest_ppage + 1; | |
10216 | sis->max = bsi.nr_pages; | |
10217 | sis->pages = bsi.nr_pages - 1; | |
10218 | sis->highest_bit = bsi.nr_pages - 1; | |
10219 | return bsi.nr_extents; | |
10220 | } | |
10221 | #else | |
10222 | static void btrfs_swap_deactivate(struct file *file) | |
10223 | { | |
10224 | } | |
10225 | ||
10226 | static int btrfs_swap_activate(struct swap_info_struct *sis, struct file *file, | |
10227 | sector_t *span) | |
10228 | { | |
10229 | return -EOPNOTSUPP; | |
10230 | } | |
10231 | #endif | |
10232 | ||
6e1d5dcc | 10233 | static const struct inode_operations btrfs_dir_inode_operations = { |
3394e160 | 10234 | .getattr = btrfs_getattr, |
39279cc3 CM |
10235 | .lookup = btrfs_lookup, |
10236 | .create = btrfs_create, | |
10237 | .unlink = btrfs_unlink, | |
10238 | .link = btrfs_link, | |
10239 | .mkdir = btrfs_mkdir, | |
10240 | .rmdir = btrfs_rmdir, | |
2773bf00 | 10241 | .rename = btrfs_rename2, |
39279cc3 CM |
10242 | .symlink = btrfs_symlink, |
10243 | .setattr = btrfs_setattr, | |
618e21d5 | 10244 | .mknod = btrfs_mknod, |
5103e947 | 10245 | .listxattr = btrfs_listxattr, |
fdebe2bd | 10246 | .permission = btrfs_permission, |
4e34e719 | 10247 | .get_acl = btrfs_get_acl, |
996a710d | 10248 | .set_acl = btrfs_set_acl, |
93fd63c2 | 10249 | .update_time = btrfs_update_time, |
ef3b9af5 | 10250 | .tmpfile = btrfs_tmpfile, |
39279cc3 | 10251 | }; |
76dda93c | 10252 | |
828c0950 | 10253 | static const struct file_operations btrfs_dir_file_operations = { |
39279cc3 CM |
10254 | .llseek = generic_file_llseek, |
10255 | .read = generic_read_dir, | |
02dbfc99 | 10256 | .iterate_shared = btrfs_real_readdir, |
23b5ec74 | 10257 | .open = btrfs_opendir, |
34287aa3 | 10258 | .unlocked_ioctl = btrfs_ioctl, |
39279cc3 | 10259 | #ifdef CONFIG_COMPAT |
4c63c245 | 10260 | .compat_ioctl = btrfs_compat_ioctl, |
39279cc3 | 10261 | #endif |
6bf13c0c | 10262 | .release = btrfs_release_file, |
e02119d5 | 10263 | .fsync = btrfs_sync_file, |
39279cc3 CM |
10264 | }; |
10265 | ||
20e5506b | 10266 | static const struct extent_io_ops btrfs_extent_io_ops = { |
4d53dddb | 10267 | /* mandatory callbacks */ |
065631f6 | 10268 | .submit_bio_hook = btrfs_submit_bio_hook, |
07157aac CM |
10269 | .readpage_end_io_hook = btrfs_readpage_end_io_hook, |
10270 | }; | |
10271 | ||
35054394 CM |
10272 | /* |
10273 | * btrfs doesn't support the bmap operation because swapfiles | |
10274 | * use bmap to make a mapping of extents in the file. They assume | |
10275 | * these extents won't change over the life of the file and they | |
10276 | * use the bmap result to do IO directly to the drive. | |
10277 | * | |
10278 | * the btrfs bmap call would return logical addresses that aren't | |
10279 | * suitable for IO and they also will change frequently as COW | |
10280 | * operations happen. So, swapfile + btrfs == corruption. | |
10281 | * | |
10282 | * For now we're avoiding this by dropping bmap. | |
10283 | */ | |
7f09410b | 10284 | static const struct address_space_operations btrfs_aops = { |
39279cc3 CM |
10285 | .readpage = btrfs_readpage, |
10286 | .writepage = btrfs_writepage, | |
b293f02e | 10287 | .writepages = btrfs_writepages, |
ba206a02 | 10288 | .readahead = btrfs_readahead, |
55e20bd1 | 10289 | .direct_IO = btrfs_direct_IO, |
a52d9a80 CM |
10290 | .invalidatepage = btrfs_invalidatepage, |
10291 | .releasepage = btrfs_releasepage, | |
f8e66081 RG |
10292 | #ifdef CONFIG_MIGRATION |
10293 | .migratepage = btrfs_migratepage, | |
10294 | #endif | |
e6dcd2dc | 10295 | .set_page_dirty = btrfs_set_page_dirty, |
465fdd97 | 10296 | .error_remove_page = generic_error_remove_page, |
ed46ff3d OS |
10297 | .swap_activate = btrfs_swap_activate, |
10298 | .swap_deactivate = btrfs_swap_deactivate, | |
39279cc3 CM |
10299 | }; |
10300 | ||
6e1d5dcc | 10301 | static const struct inode_operations btrfs_file_inode_operations = { |
39279cc3 CM |
10302 | .getattr = btrfs_getattr, |
10303 | .setattr = btrfs_setattr, | |
5103e947 | 10304 | .listxattr = btrfs_listxattr, |
fdebe2bd | 10305 | .permission = btrfs_permission, |
1506fcc8 | 10306 | .fiemap = btrfs_fiemap, |
4e34e719 | 10307 | .get_acl = btrfs_get_acl, |
996a710d | 10308 | .set_acl = btrfs_set_acl, |
e41f941a | 10309 | .update_time = btrfs_update_time, |
39279cc3 | 10310 | }; |
6e1d5dcc | 10311 | static const struct inode_operations btrfs_special_inode_operations = { |
618e21d5 JB |
10312 | .getattr = btrfs_getattr, |
10313 | .setattr = btrfs_setattr, | |
fdebe2bd | 10314 | .permission = btrfs_permission, |
33268eaf | 10315 | .listxattr = btrfs_listxattr, |
4e34e719 | 10316 | .get_acl = btrfs_get_acl, |
996a710d | 10317 | .set_acl = btrfs_set_acl, |
e41f941a | 10318 | .update_time = btrfs_update_time, |
618e21d5 | 10319 | }; |
6e1d5dcc | 10320 | static const struct inode_operations btrfs_symlink_inode_operations = { |
6b255391 | 10321 | .get_link = page_get_link, |
f209561a | 10322 | .getattr = btrfs_getattr, |
22c44fe6 | 10323 | .setattr = btrfs_setattr, |
fdebe2bd | 10324 | .permission = btrfs_permission, |
0279b4cd | 10325 | .listxattr = btrfs_listxattr, |
e41f941a | 10326 | .update_time = btrfs_update_time, |
39279cc3 | 10327 | }; |
76dda93c | 10328 | |
82d339d9 | 10329 | const struct dentry_operations btrfs_dentry_operations = { |
76dda93c YZ |
10330 | .d_delete = btrfs_dentry_delete, |
10331 | }; |