Commit | Line | Data |
---|---|---|
c1d7c514 | 1 | // SPDX-License-Identifier: GPL-2.0 |
6cbd5570 CM |
2 | /* |
3 | * Copyright (C) 2007 Oracle. All rights reserved. | |
6cbd5570 CM |
4 | */ |
5 | ||
7999096f | 6 | #include <crypto/hash.h> |
8f18cf13 | 7 | #include <linux/kernel.h> |
065631f6 | 8 | #include <linux/bio.h> |
348332e0 | 9 | #include <linux/blk-cgroup.h> |
f2eb0a24 | 10 | #include <linux/file.h> |
39279cc3 CM |
11 | #include <linux/fs.h> |
12 | #include <linux/pagemap.h> | |
13 | #include <linux/highmem.h> | |
14 | #include <linux/time.h> | |
15 | #include <linux/init.h> | |
16 | #include <linux/string.h> | |
39279cc3 | 17 | #include <linux/backing-dev.h> |
39279cc3 | 18 | #include <linux/writeback.h> |
39279cc3 | 19 | #include <linux/compat.h> |
5103e947 | 20 | #include <linux/xattr.h> |
33268eaf | 21 | #include <linux/posix_acl.h> |
d899e052 | 22 | #include <linux/falloc.h> |
5a0e3ad6 | 23 | #include <linux/slab.h> |
7a36ddec | 24 | #include <linux/ratelimit.h> |
55e301fd | 25 | #include <linux/btrfs.h> |
53b381b3 | 26 | #include <linux/blkdev.h> |
f23b5a59 | 27 | #include <linux/posix_acl_xattr.h> |
e2e40f2c | 28 | #include <linux/uio.h> |
69fe2d75 | 29 | #include <linux/magic.h> |
ae5e165d | 30 | #include <linux/iversion.h> |
ed46ff3d | 31 | #include <linux/swap.h> |
f8e66081 | 32 | #include <linux/migrate.h> |
b1c16ac9 | 33 | #include <linux/sched/mm.h> |
f85781fb | 34 | #include <linux/iomap.h> |
92d32170 | 35 | #include <asm/unaligned.h> |
14605409 | 36 | #include <linux/fsverity.h> |
602cbe91 | 37 | #include "misc.h" |
39279cc3 CM |
38 | #include "ctree.h" |
39 | #include "disk-io.h" | |
40 | #include "transaction.h" | |
41 | #include "btrfs_inode.h" | |
39279cc3 | 42 | #include "print-tree.h" |
e6dcd2dc | 43 | #include "ordered-data.h" |
95819c05 | 44 | #include "xattr.h" |
e02119d5 | 45 | #include "tree-log.h" |
103c1972 | 46 | #include "bio.h" |
c8b97818 | 47 | #include "compression.h" |
b4ce94de | 48 | #include "locking.h" |
dc89e982 | 49 | #include "free-space-cache.h" |
63541927 | 50 | #include "props.h" |
31193213 | 51 | #include "qgroup.h" |
86736342 | 52 | #include "delalloc-space.h" |
aac0023c | 53 | #include "block-group.h" |
467dc47e | 54 | #include "space-info.h" |
d8e3fb10 | 55 | #include "zoned.h" |
b945a463 | 56 | #include "subpage.h" |
26c2c454 | 57 | #include "inode-item.h" |
c7f13d42 | 58 | #include "fs.h" |
ad1ac501 | 59 | #include "accessors.h" |
a0231804 | 60 | #include "extent-tree.h" |
45c40c8f | 61 | #include "root-tree.h" |
59b818e0 | 62 | #include "defrag.h" |
f2b39277 | 63 | #include "dir-item.h" |
7c8ede16 | 64 | #include "file-item.h" |
c7a03b52 | 65 | #include "uuid-tree.h" |
7572dec8 | 66 | #include "ioctl.h" |
af142b6f | 67 | #include "file.h" |
33cf97a7 | 68 | #include "acl.h" |
67707479 | 69 | #include "relocation.h" |
5c11adcc | 70 | #include "verity.h" |
7f0add25 | 71 | #include "super.h" |
aa5d3003 | 72 | #include "orphan.h" |
b9a9a850 | 73 | #include "backref.h" |
39279cc3 CM |
74 | |
75 | struct btrfs_iget_args { | |
0202e83f | 76 | u64 ino; |
39279cc3 CM |
77 | struct btrfs_root *root; |
78 | }; | |
79 | ||
f28a4928 | 80 | struct btrfs_dio_data { |
f85781fb GR |
81 | ssize_t submitted; |
82 | struct extent_changeset *data_reserved; | |
53f2c206 | 83 | struct btrfs_ordered_extent *ordered; |
f5585f4f FM |
84 | bool data_space_reserved; |
85 | bool nocow_done; | |
f28a4928 FM |
86 | }; |
87 | ||
a3e171a0 | 88 | struct btrfs_dio_private { |
67d66982 | 89 | /* Range of I/O */ |
a3e171a0 | 90 | u64 file_offset; |
a3e171a0 CH |
91 | u32 bytes; |
92 | ||
642c5d34 | 93 | /* This must be last */ |
67d66982 | 94 | struct btrfs_bio bbio; |
a3e171a0 CH |
95 | }; |
96 | ||
642c5d34 CH |
97 | static struct bio_set btrfs_dio_bioset; |
98 | ||
88d2beec FM |
99 | struct btrfs_rename_ctx { |
100 | /* Output field. Stores the index number of the old directory entry. */ | |
101 | u64 index; | |
102 | }; | |
103 | ||
b9a9a850 QW |
104 | /* |
105 | * Used by data_reloc_print_warning_inode() to pass needed info for filename | |
106 | * resolution and output of error message. | |
107 | */ | |
108 | struct data_reloc_warn { | |
109 | struct btrfs_path path; | |
110 | struct btrfs_fs_info *fs_info; | |
111 | u64 extent_item_size; | |
112 | u64 logical; | |
113 | int mirror_num; | |
114 | }; | |
115 | ||
6e1d5dcc AD |
116 | static const struct inode_operations btrfs_dir_inode_operations; |
117 | static const struct inode_operations btrfs_symlink_inode_operations; | |
6e1d5dcc AD |
118 | static const struct inode_operations btrfs_special_inode_operations; |
119 | static const struct inode_operations btrfs_file_inode_operations; | |
7f09410b | 120 | static const struct address_space_operations btrfs_aops; |
828c0950 | 121 | static const struct file_operations btrfs_dir_file_operations; |
39279cc3 CM |
122 | |
123 | static struct kmem_cache *btrfs_inode_cachep; | |
39279cc3 | 124 | |
3972f260 | 125 | static int btrfs_setsize(struct inode *inode, struct iattr *attr); |
d9dcae67 | 126 | static int btrfs_truncate(struct btrfs_inode *inode, bool skip_writeback); |
ba9145ad | 127 | |
256b0cf9 CH |
128 | static noinline int run_delalloc_cow(struct btrfs_inode *inode, |
129 | struct page *locked_page, u64 start, | |
130 | u64 end, struct writeback_control *wbc, | |
131 | bool pages_dirty); | |
4b67c11d NB |
132 | static struct extent_map *create_io_em(struct btrfs_inode *inode, u64 start, |
133 | u64 len, u64 orig_start, u64 block_start, | |
6f9994db LB |
134 | u64 block_len, u64 orig_block_len, |
135 | u64 ram_bytes, int compress_type, | |
136 | int type); | |
7b128766 | 137 | |
b9a9a850 QW |
138 | static int data_reloc_print_warning_inode(u64 inum, u64 offset, u64 num_bytes, |
139 | u64 root, void *warn_ctx) | |
140 | { | |
141 | struct data_reloc_warn *warn = warn_ctx; | |
142 | struct btrfs_fs_info *fs_info = warn->fs_info; | |
143 | struct extent_buffer *eb; | |
144 | struct btrfs_inode_item *inode_item; | |
145 | struct inode_fs_paths *ipath = NULL; | |
146 | struct btrfs_root *local_root; | |
147 | struct btrfs_key key; | |
148 | unsigned int nofs_flag; | |
149 | u32 nlink; | |
150 | int ret; | |
151 | ||
152 | local_root = btrfs_get_fs_root(fs_info, root, true); | |
153 | if (IS_ERR(local_root)) { | |
154 | ret = PTR_ERR(local_root); | |
155 | goto err; | |
156 | } | |
157 | ||
158 | /* This makes the path point to (inum INODE_ITEM ioff). */ | |
159 | key.objectid = inum; | |
160 | key.type = BTRFS_INODE_ITEM_KEY; | |
161 | key.offset = 0; | |
162 | ||
163 | ret = btrfs_search_slot(NULL, local_root, &key, &warn->path, 0, 0); | |
164 | if (ret) { | |
165 | btrfs_put_root(local_root); | |
166 | btrfs_release_path(&warn->path); | |
167 | goto err; | |
168 | } | |
169 | ||
170 | eb = warn->path.nodes[0]; | |
171 | inode_item = btrfs_item_ptr(eb, warn->path.slots[0], struct btrfs_inode_item); | |
172 | nlink = btrfs_inode_nlink(eb, inode_item); | |
173 | btrfs_release_path(&warn->path); | |
174 | ||
175 | nofs_flag = memalloc_nofs_save(); | |
176 | ipath = init_ipath(4096, local_root, &warn->path); | |
177 | memalloc_nofs_restore(nofs_flag); | |
178 | if (IS_ERR(ipath)) { | |
179 | btrfs_put_root(local_root); | |
180 | ret = PTR_ERR(ipath); | |
181 | ipath = NULL; | |
182 | /* | |
183 | * -ENOMEM, not a critical error, just output an generic error | |
184 | * without filename. | |
185 | */ | |
186 | btrfs_warn(fs_info, | |
187 | "checksum error at logical %llu mirror %u root %llu, inode %llu offset %llu", | |
188 | warn->logical, warn->mirror_num, root, inum, offset); | |
189 | return ret; | |
190 | } | |
191 | ret = paths_from_inode(inum, ipath); | |
192 | if (ret < 0) | |
193 | goto err; | |
194 | ||
195 | /* | |
196 | * We deliberately ignore the bit ipath might have been too small to | |
197 | * hold all of the paths here | |
198 | */ | |
199 | for (int i = 0; i < ipath->fspath->elem_cnt; i++) { | |
200 | btrfs_warn(fs_info, | |
201 | "checksum error at logical %llu mirror %u root %llu inode %llu offset %llu length %u links %u (path: %s)", | |
202 | warn->logical, warn->mirror_num, root, inum, offset, | |
203 | fs_info->sectorsize, nlink, | |
204 | (char *)(unsigned long)ipath->fspath->val[i]); | |
205 | } | |
206 | ||
207 | btrfs_put_root(local_root); | |
208 | free_ipath(ipath); | |
209 | return 0; | |
210 | ||
211 | err: | |
212 | btrfs_warn(fs_info, | |
213 | "checksum error at logical %llu mirror %u root %llu inode %llu offset %llu, path resolving failed with ret=%d", | |
214 | warn->logical, warn->mirror_num, root, inum, offset, ret); | |
215 | ||
216 | free_ipath(ipath); | |
217 | return ret; | |
218 | } | |
219 | ||
220 | /* | |
221 | * Do extra user-friendly error output (e.g. lookup all the affected files). | |
222 | * | |
223 | * Return true if we succeeded doing the backref lookup. | |
224 | * Return false if such lookup failed, and has to fallback to the old error message. | |
225 | */ | |
226 | static void print_data_reloc_error(const struct btrfs_inode *inode, u64 file_off, | |
227 | const u8 *csum, const u8 *csum_expected, | |
228 | int mirror_num) | |
229 | { | |
230 | struct btrfs_fs_info *fs_info = inode->root->fs_info; | |
231 | struct btrfs_path path = { 0 }; | |
232 | struct btrfs_key found_key = { 0 }; | |
233 | struct extent_buffer *eb; | |
234 | struct btrfs_extent_item *ei; | |
235 | const u32 csum_size = fs_info->csum_size; | |
236 | u64 logical; | |
237 | u64 flags; | |
238 | u32 item_size; | |
239 | int ret; | |
240 | ||
241 | mutex_lock(&fs_info->reloc_mutex); | |
242 | logical = btrfs_get_reloc_bg_bytenr(fs_info); | |
243 | mutex_unlock(&fs_info->reloc_mutex); | |
244 | ||
245 | if (logical == U64_MAX) { | |
246 | btrfs_warn_rl(fs_info, "has data reloc tree but no running relocation"); | |
247 | btrfs_warn_rl(fs_info, | |
248 | "csum failed root %lld ino %llu off %llu csum " CSUM_FMT " expected csum " CSUM_FMT " mirror %d", | |
249 | inode->root->root_key.objectid, btrfs_ino(inode), file_off, | |
250 | CSUM_FMT_VALUE(csum_size, csum), | |
251 | CSUM_FMT_VALUE(csum_size, csum_expected), | |
252 | mirror_num); | |
253 | return; | |
254 | } | |
255 | ||
256 | logical += file_off; | |
257 | btrfs_warn_rl(fs_info, | |
258 | "csum failed root %lld ino %llu off %llu logical %llu csum " CSUM_FMT " expected csum " CSUM_FMT " mirror %d", | |
259 | inode->root->root_key.objectid, | |
260 | btrfs_ino(inode), file_off, logical, | |
261 | CSUM_FMT_VALUE(csum_size, csum), | |
262 | CSUM_FMT_VALUE(csum_size, csum_expected), | |
263 | mirror_num); | |
264 | ||
265 | ret = extent_from_logical(fs_info, logical, &path, &found_key, &flags); | |
266 | if (ret < 0) { | |
267 | btrfs_err_rl(fs_info, "failed to lookup extent item for logical %llu: %d", | |
268 | logical, ret); | |
269 | return; | |
270 | } | |
271 | eb = path.nodes[0]; | |
272 | ei = btrfs_item_ptr(eb, path.slots[0], struct btrfs_extent_item); | |
273 | item_size = btrfs_item_size(eb, path.slots[0]); | |
274 | if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) { | |
275 | unsigned long ptr = 0; | |
276 | u64 ref_root; | |
277 | u8 ref_level; | |
278 | ||
b7f9945a | 279 | while (true) { |
b9a9a850 QW |
280 | ret = tree_backref_for_extent(&ptr, eb, &found_key, ei, |
281 | item_size, &ref_root, | |
282 | &ref_level); | |
b7f9945a QW |
283 | if (ret < 0) { |
284 | btrfs_warn_rl(fs_info, | |
285 | "failed to resolve tree backref for logical %llu: %d", | |
286 | logical, ret); | |
287 | break; | |
288 | } | |
289 | if (ret > 0) | |
290 | break; | |
291 | ||
b9a9a850 QW |
292 | btrfs_warn_rl(fs_info, |
293 | "csum error at logical %llu mirror %u: metadata %s (level %d) in tree %llu", | |
294 | logical, mirror_num, | |
295 | (ref_level ? "node" : "leaf"), | |
b7f9945a QW |
296 | ref_level, ref_root); |
297 | } | |
b9a9a850 QW |
298 | btrfs_release_path(&path); |
299 | } else { | |
300 | struct btrfs_backref_walk_ctx ctx = { 0 }; | |
301 | struct data_reloc_warn reloc_warn = { 0 }; | |
302 | ||
303 | btrfs_release_path(&path); | |
304 | ||
305 | ctx.bytenr = found_key.objectid; | |
306 | ctx.extent_item_pos = logical - found_key.objectid; | |
307 | ctx.fs_info = fs_info; | |
308 | ||
309 | reloc_warn.logical = logical; | |
310 | reloc_warn.extent_item_size = found_key.offset; | |
311 | reloc_warn.mirror_num = mirror_num; | |
312 | reloc_warn.fs_info = fs_info; | |
313 | ||
314 | iterate_extent_inodes(&ctx, true, | |
315 | data_reloc_print_warning_inode, &reloc_warn); | |
316 | } | |
317 | } | |
318 | ||
f60acad3 JB |
319 | static void __cold btrfs_print_data_csum_error(struct btrfs_inode *inode, |
320 | u64 logical_start, u8 *csum, u8 *csum_expected, int mirror_num) | |
321 | { | |
322 | struct btrfs_root *root = inode->root; | |
323 | const u32 csum_size = root->fs_info->csum_size; | |
324 | ||
b9a9a850 QW |
325 | /* For data reloc tree, it's better to do a backref lookup instead. */ |
326 | if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID) | |
327 | return print_data_reloc_error(inode, logical_start, csum, | |
328 | csum_expected, mirror_num); | |
329 | ||
f60acad3 JB |
330 | /* Output without objectid, which is more meaningful */ |
331 | if (root->root_key.objectid >= BTRFS_LAST_FREE_OBJECTID) { | |
332 | btrfs_warn_rl(root->fs_info, | |
333 | "csum failed root %lld ino %lld off %llu csum " CSUM_FMT " expected csum " CSUM_FMT " mirror %d", | |
334 | root->root_key.objectid, btrfs_ino(inode), | |
335 | logical_start, | |
336 | CSUM_FMT_VALUE(csum_size, csum), | |
337 | CSUM_FMT_VALUE(csum_size, csum_expected), | |
338 | mirror_num); | |
339 | } else { | |
340 | btrfs_warn_rl(root->fs_info, | |
341 | "csum failed root %llu ino %llu off %llu csum " CSUM_FMT " expected csum " CSUM_FMT " mirror %d", | |
342 | root->root_key.objectid, btrfs_ino(inode), | |
343 | logical_start, | |
344 | CSUM_FMT_VALUE(csum_size, csum), | |
345 | CSUM_FMT_VALUE(csum_size, csum_expected), | |
346 | mirror_num); | |
347 | } | |
348 | } | |
349 | ||
a14b78ad GR |
350 | /* |
351 | * btrfs_inode_lock - lock inode i_rwsem based on arguments passed | |
352 | * | |
353 | * ilock_flags can have the following bit set: | |
354 | * | |
355 | * BTRFS_ILOCK_SHARED - acquire a shared lock on the inode | |
356 | * BTRFS_ILOCK_TRY - try to acquire the lock, if fails on first attempt | |
357 | * return -EAGAIN | |
8318ba79 | 358 | * BTRFS_ILOCK_MMAP - acquire a write lock on the i_mmap_lock |
a14b78ad | 359 | */ |
29b6352b | 360 | int btrfs_inode_lock(struct btrfs_inode *inode, unsigned int ilock_flags) |
a14b78ad GR |
361 | { |
362 | if (ilock_flags & BTRFS_ILOCK_SHARED) { | |
363 | if (ilock_flags & BTRFS_ILOCK_TRY) { | |
29b6352b | 364 | if (!inode_trylock_shared(&inode->vfs_inode)) |
a14b78ad GR |
365 | return -EAGAIN; |
366 | else | |
367 | return 0; | |
368 | } | |
29b6352b | 369 | inode_lock_shared(&inode->vfs_inode); |
a14b78ad GR |
370 | } else { |
371 | if (ilock_flags & BTRFS_ILOCK_TRY) { | |
29b6352b | 372 | if (!inode_trylock(&inode->vfs_inode)) |
a14b78ad GR |
373 | return -EAGAIN; |
374 | else | |
375 | return 0; | |
376 | } | |
29b6352b | 377 | inode_lock(&inode->vfs_inode); |
a14b78ad | 378 | } |
8318ba79 | 379 | if (ilock_flags & BTRFS_ILOCK_MMAP) |
29b6352b | 380 | down_write(&inode->i_mmap_lock); |
a14b78ad GR |
381 | return 0; |
382 | } | |
383 | ||
384 | /* | |
385 | * btrfs_inode_unlock - unock inode i_rwsem | |
386 | * | |
387 | * ilock_flags should contain the same bits set as passed to btrfs_inode_lock() | |
388 | * to decide whether the lock acquired is shared or exclusive. | |
389 | */ | |
e5d4d75b | 390 | void btrfs_inode_unlock(struct btrfs_inode *inode, unsigned int ilock_flags) |
a14b78ad | 391 | { |
8318ba79 | 392 | if (ilock_flags & BTRFS_ILOCK_MMAP) |
e5d4d75b | 393 | up_write(&inode->i_mmap_lock); |
a14b78ad | 394 | if (ilock_flags & BTRFS_ILOCK_SHARED) |
e5d4d75b | 395 | inode_unlock_shared(&inode->vfs_inode); |
a14b78ad | 396 | else |
e5d4d75b | 397 | inode_unlock(&inode->vfs_inode); |
a14b78ad GR |
398 | } |
399 | ||
52427260 QW |
400 | /* |
401 | * Cleanup all submitted ordered extents in specified range to handle errors | |
52042d8e | 402 | * from the btrfs_run_delalloc_range() callback. |
52427260 QW |
403 | * |
404 | * NOTE: caller must ensure that when an error happens, it can not call | |
405 | * extent_clear_unlock_delalloc() to clear both the bits EXTENT_DO_ACCOUNTING | |
406 | * and EXTENT_DELALLOC simultaneously, because that causes the reserved metadata | |
407 | * to be released, which we want to happen only when finishing the ordered | |
d1051d6e | 408 | * extent (btrfs_finish_ordered_io()). |
52427260 | 409 | */ |
64e1db56 | 410 | static inline void btrfs_cleanup_ordered_extents(struct btrfs_inode *inode, |
d1051d6e NB |
411 | struct page *locked_page, |
412 | u64 offset, u64 bytes) | |
52427260 | 413 | { |
63d71450 NA |
414 | unsigned long index = offset >> PAGE_SHIFT; |
415 | unsigned long end_index = (offset + bytes - 1) >> PAGE_SHIFT; | |
0e47b25c | 416 | u64 page_start = 0, page_end = 0; |
63d71450 NA |
417 | struct page *page; |
418 | ||
99826e4c NA |
419 | if (locked_page) { |
420 | page_start = page_offset(locked_page); | |
421 | page_end = page_start + PAGE_SIZE - 1; | |
422 | } | |
423 | ||
63d71450 | 424 | while (index <= end_index) { |
968f2566 | 425 | /* |
9783e4de CH |
426 | * For locked page, we will call btrfs_mark_ordered_io_finished |
427 | * through btrfs_mark_ordered_io_finished() on it | |
428 | * in run_delalloc_range() for the error handling, which will | |
429 | * clear page Ordered and run the ordered extent accounting. | |
968f2566 QW |
430 | * |
431 | * Here we can't just clear the Ordered bit, or | |
432 | * btrfs_mark_ordered_io_finished() would skip the accounting | |
433 | * for the page range, and the ordered extent will never finish. | |
434 | */ | |
99826e4c | 435 | if (locked_page && index == (page_start >> PAGE_SHIFT)) { |
968f2566 QW |
436 | index++; |
437 | continue; | |
438 | } | |
64e1db56 | 439 | page = find_get_page(inode->vfs_inode.i_mapping, index); |
63d71450 NA |
440 | index++; |
441 | if (!page) | |
442 | continue; | |
968f2566 QW |
443 | |
444 | /* | |
445 | * Here we just clear all Ordered bits for every page in the | |
711f447b | 446 | * range, then btrfs_mark_ordered_io_finished() will handle |
968f2566 QW |
447 | * the ordered extent accounting for the range. |
448 | */ | |
b945a463 QW |
449 | btrfs_page_clamp_clear_ordered(inode->root->fs_info, page, |
450 | offset, bytes); | |
63d71450 NA |
451 | put_page(page); |
452 | } | |
d1051d6e | 453 | |
99826e4c NA |
454 | if (locked_page) { |
455 | /* The locked page covers the full range, nothing needs to be done */ | |
456 | if (bytes + offset <= page_start + PAGE_SIZE) | |
457 | return; | |
458 | /* | |
459 | * In case this page belongs to the delalloc range being | |
460 | * instantiated then skip it, since the first page of a range is | |
461 | * going to be properly cleaned up by the caller of | |
462 | * run_delalloc_range | |
463 | */ | |
464 | if (page_start >= offset && page_end <= (offset + bytes - 1)) { | |
465 | bytes = offset + bytes - page_offset(locked_page) - PAGE_SIZE; | |
466 | offset = page_offset(locked_page) + PAGE_SIZE; | |
467 | } | |
d1051d6e NB |
468 | } |
469 | ||
711f447b | 470 | return btrfs_mark_ordered_io_finished(inode, NULL, offset, bytes, false); |
52427260 QW |
471 | } |
472 | ||
7152b425 | 473 | static int btrfs_dirty_inode(struct btrfs_inode *inode); |
7b128766 | 474 | |
f34f57a3 | 475 | static int btrfs_init_inode_security(struct btrfs_trans_handle *trans, |
3538d68d | 476 | struct btrfs_new_inode_args *args) |
0279b4cd JO |
477 | { |
478 | int err; | |
479 | ||
3538d68d OS |
480 | if (args->default_acl) { |
481 | err = __btrfs_set_acl(trans, args->inode, args->default_acl, | |
482 | ACL_TYPE_DEFAULT); | |
483 | if (err) | |
484 | return err; | |
485 | } | |
486 | if (args->acl) { | |
487 | err = __btrfs_set_acl(trans, args->inode, args->acl, ACL_TYPE_ACCESS); | |
488 | if (err) | |
489 | return err; | |
490 | } | |
491 | if (!args->default_acl && !args->acl) | |
492 | cache_no_acl(args->inode); | |
493 | return btrfs_xattr_security_init(trans, args->inode, args->dir, | |
494 | &args->dentry->d_name); | |
0279b4cd JO |
495 | } |
496 | ||
c8b97818 CM |
497 | /* |
498 | * this does all the hard work for inserting an inline extent into | |
499 | * the btree. The caller should have done a btrfs_drop_extents so that | |
500 | * no overlapping inline items exist in the btree | |
501 | */ | |
40f76580 | 502 | static int insert_inline_extent(struct btrfs_trans_handle *trans, |
8dd9872d OS |
503 | struct btrfs_path *path, |
504 | struct btrfs_inode *inode, bool extent_inserted, | |
505 | size_t size, size_t compressed_size, | |
fe3f566c | 506 | int compress_type, |
d9496e8a OS |
507 | struct page **compressed_pages, |
508 | bool update_i_size) | |
c8b97818 | 509 | { |
8dd9872d | 510 | struct btrfs_root *root = inode->root; |
c8b97818 CM |
511 | struct extent_buffer *leaf; |
512 | struct page *page = NULL; | |
513 | char *kaddr; | |
514 | unsigned long ptr; | |
515 | struct btrfs_file_extent_item *ei; | |
c8b97818 CM |
516 | int ret; |
517 | size_t cur_size = size; | |
d9496e8a | 518 | u64 i_size; |
c8b97818 | 519 | |
982f1f5d JJB |
520 | ASSERT((compressed_size > 0 && compressed_pages) || |
521 | (compressed_size == 0 && !compressed_pages)); | |
522 | ||
fe3f566c | 523 | if (compressed_size && compressed_pages) |
c8b97818 | 524 | cur_size = compressed_size; |
c8b97818 | 525 | |
1acae57b FDBM |
526 | if (!extent_inserted) { |
527 | struct btrfs_key key; | |
528 | size_t datasize; | |
c8b97818 | 529 | |
8dd9872d OS |
530 | key.objectid = btrfs_ino(inode); |
531 | key.offset = 0; | |
962a298f | 532 | key.type = BTRFS_EXTENT_DATA_KEY; |
c8b97818 | 533 | |
1acae57b | 534 | datasize = btrfs_file_extent_calc_inline_size(cur_size); |
1acae57b FDBM |
535 | ret = btrfs_insert_empty_item(trans, root, path, &key, |
536 | datasize); | |
79b4f4c6 | 537 | if (ret) |
1acae57b | 538 | goto fail; |
c8b97818 CM |
539 | } |
540 | leaf = path->nodes[0]; | |
541 | ei = btrfs_item_ptr(leaf, path->slots[0], | |
542 | struct btrfs_file_extent_item); | |
543 | btrfs_set_file_extent_generation(leaf, ei, trans->transid); | |
544 | btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE); | |
545 | btrfs_set_file_extent_encryption(leaf, ei, 0); | |
546 | btrfs_set_file_extent_other_encoding(leaf, ei, 0); | |
547 | btrfs_set_file_extent_ram_bytes(leaf, ei, size); | |
548 | ptr = btrfs_file_extent_inline_start(ei); | |
549 | ||
261507a0 | 550 | if (compress_type != BTRFS_COMPRESS_NONE) { |
c8b97818 CM |
551 | struct page *cpage; |
552 | int i = 0; | |
d397712b | 553 | while (compressed_size > 0) { |
c8b97818 | 554 | cpage = compressed_pages[i]; |
5b050f04 | 555 | cur_size = min_t(unsigned long, compressed_size, |
09cbfeaf | 556 | PAGE_SIZE); |
c8b97818 | 557 | |
4cb2e5e8 | 558 | kaddr = kmap_local_page(cpage); |
c8b97818 | 559 | write_extent_buffer(leaf, kaddr, ptr, cur_size); |
4cb2e5e8 | 560 | kunmap_local(kaddr); |
c8b97818 CM |
561 | |
562 | i++; | |
563 | ptr += cur_size; | |
564 | compressed_size -= cur_size; | |
565 | } | |
566 | btrfs_set_file_extent_compression(leaf, ei, | |
261507a0 | 567 | compress_type); |
c8b97818 | 568 | } else { |
8dd9872d | 569 | page = find_get_page(inode->vfs_inode.i_mapping, 0); |
c8b97818 | 570 | btrfs_set_file_extent_compression(leaf, ei, 0); |
4cb2e5e8 | 571 | kaddr = kmap_local_page(page); |
8dd9872d | 572 | write_extent_buffer(leaf, kaddr, ptr, size); |
4cb2e5e8 | 573 | kunmap_local(kaddr); |
09cbfeaf | 574 | put_page(page); |
c8b97818 CM |
575 | } |
576 | btrfs_mark_buffer_dirty(leaf); | |
1acae57b | 577 | btrfs_release_path(path); |
c8b97818 | 578 | |
9ddc959e JB |
579 | /* |
580 | * We align size to sectorsize for inline extents just for simplicity | |
581 | * sake. | |
582 | */ | |
8dd9872d OS |
583 | ret = btrfs_inode_set_file_extent_range(inode, 0, |
584 | ALIGN(size, root->fs_info->sectorsize)); | |
9ddc959e JB |
585 | if (ret) |
586 | goto fail; | |
587 | ||
c2167754 | 588 | /* |
d9496e8a OS |
589 | * We're an inline extent, so nobody can extend the file past i_size |
590 | * without locking a page we already have locked. | |
c2167754 | 591 | * |
d9496e8a OS |
592 | * We must do any i_size and inode updates before we unlock the pages. |
593 | * Otherwise we could end up racing with unlink. | |
c2167754 | 594 | */ |
d9496e8a OS |
595 | i_size = i_size_read(&inode->vfs_inode); |
596 | if (update_i_size && size > i_size) { | |
597 | i_size_write(&inode->vfs_inode, size); | |
598 | i_size = size; | |
599 | } | |
600 | inode->disk_i_size = i_size; | |
8dd9872d | 601 | |
c8b97818 | 602 | fail: |
79b4f4c6 | 603 | return ret; |
c8b97818 CM |
604 | } |
605 | ||
606 | ||
607 | /* | |
608 | * conditionally insert an inline extent into the file. This | |
609 | * does the checks required to make sure the data is small enough | |
610 | * to fit as an inline extent. | |
611 | */ | |
8dd9872d OS |
612 | static noinline int cow_file_range_inline(struct btrfs_inode *inode, u64 size, |
613 | size_t compressed_size, | |
00361589 | 614 | int compress_type, |
d9496e8a OS |
615 | struct page **compressed_pages, |
616 | bool update_i_size) | |
c8b97818 | 617 | { |
5893dfb9 | 618 | struct btrfs_drop_extents_args drop_args = { 0 }; |
a0349401 | 619 | struct btrfs_root *root = inode->root; |
0b246afa | 620 | struct btrfs_fs_info *fs_info = root->fs_info; |
00361589 | 621 | struct btrfs_trans_handle *trans; |
8dd9872d | 622 | u64 data_len = (compressed_size ?: size); |
c8b97818 | 623 | int ret; |
1acae57b | 624 | struct btrfs_path *path; |
c8b97818 | 625 | |
8dd9872d OS |
626 | /* |
627 | * We can create an inline extent if it ends at or beyond the current | |
628 | * i_size, is no larger than a sector (decompressed), and the (possibly | |
629 | * compressed) data fits in a leaf and the configured maximum inline | |
630 | * size. | |
631 | */ | |
632 | if (size < i_size_read(&inode->vfs_inode) || | |
633 | size > fs_info->sectorsize || | |
0b246afa | 634 | data_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info) || |
8dd9872d | 635 | data_len > fs_info->max_inline) |
c8b97818 | 636 | return 1; |
c8b97818 | 637 | |
1acae57b FDBM |
638 | path = btrfs_alloc_path(); |
639 | if (!path) | |
640 | return -ENOMEM; | |
641 | ||
00361589 | 642 | trans = btrfs_join_transaction(root); |
1acae57b FDBM |
643 | if (IS_ERR(trans)) { |
644 | btrfs_free_path(path); | |
00361589 | 645 | return PTR_ERR(trans); |
1acae57b | 646 | } |
a0349401 | 647 | trans->block_rsv = &inode->block_rsv; |
00361589 | 648 | |
5893dfb9 | 649 | drop_args.path = path; |
8dd9872d OS |
650 | drop_args.start = 0; |
651 | drop_args.end = fs_info->sectorsize; | |
5893dfb9 FM |
652 | drop_args.drop_cache = true; |
653 | drop_args.replace_extent = true; | |
8dd9872d | 654 | drop_args.extent_item_size = btrfs_file_extent_calc_inline_size(data_len); |
5893dfb9 | 655 | ret = btrfs_drop_extents(trans, root, inode, &drop_args); |
00361589 | 656 | if (ret) { |
66642832 | 657 | btrfs_abort_transaction(trans, ret); |
00361589 JB |
658 | goto out; |
659 | } | |
c8b97818 | 660 | |
8dd9872d OS |
661 | ret = insert_inline_extent(trans, path, inode, drop_args.extent_inserted, |
662 | size, compressed_size, compress_type, | |
d9496e8a | 663 | compressed_pages, update_i_size); |
2adcac1a | 664 | if (ret && ret != -ENOSPC) { |
66642832 | 665 | btrfs_abort_transaction(trans, ret); |
00361589 | 666 | goto out; |
2adcac1a | 667 | } else if (ret == -ENOSPC) { |
00361589 JB |
668 | ret = 1; |
669 | goto out; | |
79787eaa | 670 | } |
2adcac1a | 671 | |
8dd9872d | 672 | btrfs_update_inode_bytes(inode, size, drop_args.bytes_found); |
9a56fcd1 | 673 | ret = btrfs_update_inode(trans, root, inode); |
2766ff61 FM |
674 | if (ret && ret != -ENOSPC) { |
675 | btrfs_abort_transaction(trans, ret); | |
676 | goto out; | |
677 | } else if (ret == -ENOSPC) { | |
678 | ret = 1; | |
679 | goto out; | |
680 | } | |
681 | ||
23e3337f | 682 | btrfs_set_inode_full_sync(inode); |
00361589 | 683 | out: |
94ed938a QW |
684 | /* |
685 | * Don't forget to free the reserved space, as for inlined extent | |
686 | * it won't count as data extent, free them directly here. | |
687 | * And at reserve time, it's always aligned to page size, so | |
688 | * just free one page here. | |
689 | */ | |
a0349401 | 690 | btrfs_qgroup_free_data(inode, NULL, 0, PAGE_SIZE); |
1acae57b | 691 | btrfs_free_path(path); |
3a45bb20 | 692 | btrfs_end_transaction(trans); |
00361589 | 693 | return ret; |
c8b97818 CM |
694 | } |
695 | ||
771ed689 CM |
696 | struct async_extent { |
697 | u64 start; | |
698 | u64 ram_size; | |
699 | u64 compressed_size; | |
700 | struct page **pages; | |
701 | unsigned long nr_pages; | |
261507a0 | 702 | int compress_type; |
771ed689 CM |
703 | struct list_head list; |
704 | }; | |
705 | ||
97db1204 | 706 | struct async_chunk { |
99a81a44 | 707 | struct btrfs_inode *inode; |
771ed689 CM |
708 | struct page *locked_page; |
709 | u64 start; | |
710 | u64 end; | |
bf9486d6 | 711 | blk_opf_t write_flags; |
771ed689 | 712 | struct list_head extents; |
ec39f769 | 713 | struct cgroup_subsys_state *blkcg_css; |
771ed689 | 714 | struct btrfs_work work; |
9e895a8f | 715 | struct async_cow *async_cow; |
771ed689 CM |
716 | }; |
717 | ||
97db1204 | 718 | struct async_cow { |
97db1204 NB |
719 | atomic_t num_chunks; |
720 | struct async_chunk chunks[]; | |
771ed689 CM |
721 | }; |
722 | ||
97db1204 | 723 | static noinline int add_async_extent(struct async_chunk *cow, |
771ed689 CM |
724 | u64 start, u64 ram_size, |
725 | u64 compressed_size, | |
726 | struct page **pages, | |
261507a0 LZ |
727 | unsigned long nr_pages, |
728 | int compress_type) | |
771ed689 CM |
729 | { |
730 | struct async_extent *async_extent; | |
731 | ||
732 | async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS); | |
79787eaa | 733 | BUG_ON(!async_extent); /* -ENOMEM */ |
771ed689 CM |
734 | async_extent->start = start; |
735 | async_extent->ram_size = ram_size; | |
736 | async_extent->compressed_size = compressed_size; | |
737 | async_extent->pages = pages; | |
738 | async_extent->nr_pages = nr_pages; | |
261507a0 | 739 | async_extent->compress_type = compress_type; |
771ed689 CM |
740 | list_add_tail(&async_extent->list, &cow->extents); |
741 | return 0; | |
742 | } | |
743 | ||
42c16da6 QW |
744 | /* |
745 | * Check if the inode needs to be submitted to compression, based on mount | |
746 | * options, defragmentation, properties or heuristics. | |
747 | */ | |
808a1292 NB |
748 | static inline int inode_need_compress(struct btrfs_inode *inode, u64 start, |
749 | u64 end) | |
f79707b0 | 750 | { |
808a1292 | 751 | struct btrfs_fs_info *fs_info = inode->root->fs_info; |
f79707b0 | 752 | |
e6f9d696 | 753 | if (!btrfs_inode_can_compress(inode)) { |
42c16da6 QW |
754 | WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG), |
755 | KERN_ERR "BTRFS: unexpected compression for ino %llu\n", | |
808a1292 | 756 | btrfs_ino(inode)); |
42c16da6 QW |
757 | return 0; |
758 | } | |
0cf9b244 QW |
759 | /* |
760 | * Special check for subpage. | |
761 | * | |
762 | * We lock the full page then run each delalloc range in the page, thus | |
763 | * for the following case, we will hit some subpage specific corner case: | |
764 | * | |
765 | * 0 32K 64K | |
766 | * | |///////| |///////| | |
767 | * \- A \- B | |
768 | * | |
769 | * In above case, both range A and range B will try to unlock the full | |
770 | * page [0, 64K), causing the one finished later will have page | |
771 | * unlocked already, triggering various page lock requirement BUG_ON()s. | |
772 | * | |
773 | * So here we add an artificial limit that subpage compression can only | |
774 | * if the range is fully page aligned. | |
775 | * | |
776 | * In theory we only need to ensure the first page is fully covered, but | |
777 | * the tailing partial page will be locked until the full compression | |
778 | * finishes, delaying the write of other range. | |
779 | * | |
780 | * TODO: Make btrfs_run_delalloc_range() to lock all delalloc range | |
781 | * first to prevent any submitted async extent to unlock the full page. | |
782 | * By this, we can ensure for subpage case that only the last async_cow | |
783 | * will unlock the full page. | |
784 | */ | |
785 | if (fs_info->sectorsize < PAGE_SIZE) { | |
1280d2d1 FK |
786 | if (!PAGE_ALIGNED(start) || |
787 | !PAGE_ALIGNED(end + 1)) | |
0cf9b244 QW |
788 | return 0; |
789 | } | |
790 | ||
f79707b0 | 791 | /* force compress */ |
0b246afa | 792 | if (btrfs_test_opt(fs_info, FORCE_COMPRESS)) |
f79707b0 | 793 | return 1; |
eec63c65 | 794 | /* defrag ioctl */ |
808a1292 | 795 | if (inode->defrag_compress) |
eec63c65 | 796 | return 1; |
f79707b0 | 797 | /* bad compression ratios */ |
808a1292 | 798 | if (inode->flags & BTRFS_INODE_NOCOMPRESS) |
f79707b0 | 799 | return 0; |
0b246afa | 800 | if (btrfs_test_opt(fs_info, COMPRESS) || |
808a1292 NB |
801 | inode->flags & BTRFS_INODE_COMPRESS || |
802 | inode->prop_compress) | |
803 | return btrfs_compress_heuristic(&inode->vfs_inode, start, end); | |
f79707b0 WS |
804 | return 0; |
805 | } | |
806 | ||
6158e1ce | 807 | static inline void inode_should_defrag(struct btrfs_inode *inode, |
558732df | 808 | u64 start, u64 end, u64 num_bytes, u32 small_write) |
26d30f85 AJ |
809 | { |
810 | /* If this is a small write inside eof, kick off a defrag */ | |
811 | if (num_bytes < small_write && | |
6158e1ce | 812 | (start > 0 || end + 1 < inode->disk_i_size)) |
558732df | 813 | btrfs_add_inode_defrag(NULL, inode, small_write); |
26d30f85 AJ |
814 | } |
815 | ||
d352ac68 | 816 | /* |
c15d8cf2 | 817 | * Work queue call back to started compression on a file and pages. |
c8b97818 | 818 | * |
c15d8cf2 CH |
819 | * This is done inside an ordered work queue, and the compression is spread |
820 | * across many cpus. The actual IO submission is step two, and the ordered work | |
821 | * queue takes care of making sure that happens in the same order things were | |
822 | * put onto the queue by writepages and friends. | |
c8b97818 | 823 | * |
c15d8cf2 CH |
824 | * If this code finds it can't get good compression, it puts an entry onto the |
825 | * work queue to write the uncompressed bytes. This makes sure that both | |
826 | * compressed inodes and uncompressed inodes are written in the same order that | |
827 | * the flusher thread sent them down. | |
d352ac68 | 828 | */ |
c15d8cf2 | 829 | static void compress_file_range(struct btrfs_work *work) |
b888db2b | 830 | { |
c15d8cf2 CH |
831 | struct async_chunk *async_chunk = |
832 | container_of(work, struct async_chunk, work); | |
99a01bd6 DS |
833 | struct btrfs_inode *inode = async_chunk->inode; |
834 | struct btrfs_fs_info *fs_info = inode->root->fs_info; | |
a994310a | 835 | struct address_space *mapping = inode->vfs_inode.i_mapping; |
0b246afa | 836 | u64 blocksize = fs_info->sectorsize; |
1368c6da NB |
837 | u64 start = async_chunk->start; |
838 | u64 end = async_chunk->end; | |
c8b97818 | 839 | u64 actual_end; |
d98da499 | 840 | u64 i_size; |
e6dcd2dc | 841 | int ret = 0; |
e94e54e8 | 842 | struct page **pages; |
c8b97818 | 843 | unsigned long nr_pages; |
c8b97818 CM |
844 | unsigned long total_compressed = 0; |
845 | unsigned long total_in = 0; | |
e94e54e8 | 846 | unsigned int poff; |
c8b97818 | 847 | int i; |
0b246afa | 848 | int compress_type = fs_info->compress_type; |
b888db2b | 849 | |
99a01bd6 | 850 | inode_should_defrag(inode, start, end, end - start + 1, SZ_16K); |
4cb5300b | 851 | |
44962ca3 CH |
852 | /* |
853 | * We need to call clear_page_dirty_for_io on each page in the range. | |
854 | * Otherwise applications with the file mmap'd can wander in and change | |
855 | * the page contents while we are compressing them. | |
856 | */ | |
857 | extent_range_clear_dirty_for_io(&inode->vfs_inode, start, end); | |
858 | ||
d98da499 JB |
859 | /* |
860 | * We need to save i_size before now because it could change in between | |
861 | * us evaluating the size and assigning it. This is because we lock and | |
862 | * unlock the page in truncate and fallocate, and then modify the i_size | |
863 | * later on. | |
864 | * | |
865 | * The barriers are to emulate READ_ONCE, remove that once i_size_read | |
866 | * does that for us. | |
867 | */ | |
868 | barrier(); | |
99a01bd6 | 869 | i_size = i_size_read(&inode->vfs_inode); |
d98da499 JB |
870 | barrier(); |
871 | actual_end = min_t(u64, i_size, end + 1); | |
c8b97818 | 872 | again: |
e94e54e8 | 873 | pages = NULL; |
09cbfeaf | 874 | nr_pages = (end >> PAGE_SHIFT) - (start >> PAGE_SHIFT) + 1; |
544fe4a9 | 875 | nr_pages = min_t(unsigned long, nr_pages, BTRFS_MAX_COMPRESSED_PAGES); |
be20aa9d | 876 | |
f03d9301 CM |
877 | /* |
878 | * we don't want to send crud past the end of i_size through | |
879 | * compression, that's just a waste of CPU time. So, if the | |
880 | * end of the file is before the start of our current | |
881 | * requested range of bytes, we bail out to the uncompressed | |
882 | * cleanup code that can deal with all of this. | |
883 | * | |
884 | * It isn't really the fastest way to fix things, but this is a | |
885 | * very uncommon corner. | |
886 | */ | |
887 | if (actual_end <= start) | |
888 | goto cleanup_and_bail_uncompressed; | |
889 | ||
c8b97818 CM |
890 | total_compressed = actual_end - start; |
891 | ||
4bcbb332 | 892 | /* |
0cf9b244 | 893 | * Skip compression for a small file range(<=blocksize) that |
01327610 | 894 | * isn't an inline extent, since it doesn't save disk space at all. |
4bcbb332 SW |
895 | */ |
896 | if (total_compressed <= blocksize && | |
99a01bd6 | 897 | (start > 0 || end + 1 < inode->disk_i_size)) |
4bcbb332 SW |
898 | goto cleanup_and_bail_uncompressed; |
899 | ||
0cf9b244 QW |
900 | /* |
901 | * For subpage case, we require full page alignment for the sector | |
902 | * aligned range. | |
903 | * Thus we must also check against @actual_end, not just @end. | |
904 | */ | |
905 | if (blocksize < PAGE_SIZE) { | |
1280d2d1 FK |
906 | if (!PAGE_ALIGNED(start) || |
907 | !PAGE_ALIGNED(round_up(actual_end, blocksize))) | |
0cf9b244 QW |
908 | goto cleanup_and_bail_uncompressed; |
909 | } | |
910 | ||
069eac78 DS |
911 | total_compressed = min_t(unsigned long, total_compressed, |
912 | BTRFS_MAX_UNCOMPRESSED); | |
c8b97818 CM |
913 | total_in = 0; |
914 | ret = 0; | |
db94535d | 915 | |
771ed689 | 916 | /* |
e94e54e8 CH |
917 | * We do compression for mount -o compress and when the inode has not |
918 | * been flagged as NOCOMPRESS. This flag can change at any time if we | |
919 | * discover bad compression ratios. | |
c8b97818 | 920 | */ |
e94e54e8 | 921 | if (!inode_need_compress(inode, start, end)) |
6a7167bf | 922 | goto cleanup_and_bail_uncompressed; |
261507a0 | 923 | |
e94e54e8 CH |
924 | pages = kcalloc(nr_pages, sizeof(struct page *), GFP_NOFS); |
925 | if (!pages) { | |
4adaa611 | 926 | /* |
e94e54e8 CH |
927 | * Memory allocation failure is not a fatal error, we can fall |
928 | * back to uncompressed code. | |
4adaa611 | 929 | */ |
6a7167bf | 930 | goto cleanup_and_bail_uncompressed; |
e94e54e8 | 931 | } |
f51d2b59 | 932 | |
e94e54e8 CH |
933 | if (inode->defrag_compress) |
934 | compress_type = inode->defrag_compress; | |
935 | else if (inode->prop_compress) | |
936 | compress_type = inode->prop_compress; | |
937 | ||
e94e54e8 CH |
938 | /* Compression level is applied here. */ |
939 | ret = btrfs_compress_pages(compress_type | (fs_info->compress_level << 4), | |
940 | mapping, start, pages, &nr_pages, &total_in, | |
941 | &total_compressed); | |
942 | if (ret) | |
184aa1ff | 943 | goto mark_incompressible; |
c8b97818 | 944 | |
e94e54e8 CH |
945 | /* |
946 | * Zero the tail end of the last page, as we might be sending it down | |
947 | * to disk. | |
948 | */ | |
949 | poff = offset_in_page(total_compressed); | |
950 | if (poff) | |
951 | memzero_page(pages[nr_pages - 1], poff, PAGE_SIZE - poff); | |
c8b97818 | 952 | |
7367253a | 953 | /* |
6a7167bf CH |
954 | * Try to create an inline extent. |
955 | * | |
956 | * If we didn't compress the entire range, try to create an uncompressed | |
957 | * inline extent, else a compressed one. | |
958 | * | |
7367253a | 959 | * Check cow_file_range() for why we don't even try to create inline |
e94e54e8 | 960 | * extent for the subpage case. |
7367253a QW |
961 | */ |
962 | if (start == 0 && fs_info->sectorsize == PAGE_SIZE) { | |
6a7167bf CH |
963 | if (total_in < actual_end) { |
964 | ret = cow_file_range_inline(inode, actual_end, 0, | |
965 | BTRFS_COMPRESS_NONE, NULL, | |
966 | false); | |
c8b97818 | 967 | } else { |
99a01bd6 | 968 | ret = cow_file_range_inline(inode, actual_end, |
fe3f566c | 969 | total_compressed, |
d9496e8a OS |
970 | compress_type, pages, |
971 | false); | |
c8b97818 | 972 | } |
79787eaa | 973 | if (ret <= 0) { |
151a41bc | 974 | unsigned long clear_flags = EXTENT_DELALLOC | |
8b62f87b JB |
975 | EXTENT_DELALLOC_NEW | EXTENT_DEFRAG | |
976 | EXTENT_DO_ACCOUNTING; | |
e6eb4314 | 977 | |
a994310a CH |
978 | if (ret < 0) |
979 | mapping_set_error(mapping, -EIO); | |
151a41bc | 980 | |
771ed689 | 981 | /* |
79787eaa JM |
982 | * inline extent creation worked or returned error, |
983 | * we don't need to create any more async work items. | |
984 | * Unlock and free up our temp pages. | |
8b62f87b JB |
985 | * |
986 | * We use DO_ACCOUNTING here because we need the | |
987 | * delalloc_release_metadata to be done _after_ we drop | |
988 | * our outstanding extent for clearing delalloc for this | |
989 | * range. | |
771ed689 | 990 | */ |
99a01bd6 | 991 | extent_clear_unlock_delalloc(inode, start, end, |
ad7ff17b | 992 | NULL, |
74e9194a | 993 | clear_flags, |
ba8b04c1 | 994 | PAGE_UNLOCK | |
6869b0a8 | 995 | PAGE_START_WRITEBACK | |
c2790a2e | 996 | PAGE_END_WRITEBACK); |
f778b6b8 | 997 | goto free_pages; |
c8b97818 CM |
998 | } |
999 | } | |
1000 | ||
e94e54e8 CH |
1001 | /* |
1002 | * We aren't doing an inline extent. Round the compressed size up to a | |
1003 | * block size boundary so the allocator does sane things. | |
1004 | */ | |
1005 | total_compressed = ALIGN(total_compressed, blocksize); | |
c8b97818 | 1006 | |
e94e54e8 CH |
1007 | /* |
1008 | * One last check to make sure the compression is really a win, compare | |
1009 | * the page count read with the blocks on disk, compression must free at | |
1010 | * least one sector. | |
1011 | */ | |
1012 | total_in = round_up(total_in, fs_info->sectorsize); | |
1013 | if (total_compressed + blocksize > total_in) | |
184aa1ff | 1014 | goto mark_incompressible; |
c8bb0c8b | 1015 | |
e94e54e8 CH |
1016 | /* |
1017 | * The async work queues will take care of doing actual allocation on | |
1018 | * disk for these compressed pages, and will submit the bios. | |
1019 | */ | |
1020 | add_async_extent(async_chunk, start, total_in, total_compressed, pages, | |
1021 | nr_pages, compress_type); | |
1022 | if (start + total_in < end) { | |
1023 | start += total_in; | |
1024 | cond_resched(); | |
1025 | goto again; | |
c8b97818 | 1026 | } |
e94e54e8 CH |
1027 | return; |
1028 | ||
184aa1ff CH |
1029 | mark_incompressible: |
1030 | if (!btrfs_test_opt(fs_info, FORCE_COMPRESS) && !inode->prop_compress) | |
1031 | inode->flags |= BTRFS_INODE_NOCOMPRESS; | |
e94e54e8 | 1032 | cleanup_and_bail_uncompressed: |
b5326271 | 1033 | add_async_extent(async_chunk, start, end - start + 1, 0, NULL, 0, |
c8bb0c8b | 1034 | BTRFS_COMPRESS_NONE); |
f778b6b8 | 1035 | free_pages: |
c8bb0c8b | 1036 | if (pages) { |
4d3a800e | 1037 | for (i = 0; i < nr_pages; i++) { |
70b99e69 | 1038 | WARN_ON(pages[i]->mapping); |
09cbfeaf | 1039 | put_page(pages[i]); |
c8b97818 CM |
1040 | } |
1041 | kfree(pages); | |
c8b97818 | 1042 | } |
771ed689 | 1043 | } |
771ed689 | 1044 | |
40ae837b FM |
1045 | static void free_async_extent_pages(struct async_extent *async_extent) |
1046 | { | |
1047 | int i; | |
1048 | ||
1049 | if (!async_extent->pages) | |
1050 | return; | |
1051 | ||
1052 | for (i = 0; i < async_extent->nr_pages; i++) { | |
1053 | WARN_ON(async_extent->pages[i]->mapping); | |
09cbfeaf | 1054 | put_page(async_extent->pages[i]); |
40ae837b FM |
1055 | } |
1056 | kfree(async_extent->pages); | |
1057 | async_extent->nr_pages = 0; | |
1058 | async_extent->pages = NULL; | |
771ed689 CM |
1059 | } |
1060 | ||
ff20d6a4 CH |
1061 | static void submit_uncompressed_range(struct btrfs_inode *inode, |
1062 | struct async_extent *async_extent, | |
1063 | struct page *locked_page) | |
771ed689 | 1064 | { |
2b83a0ee QW |
1065 | u64 start = async_extent->start; |
1066 | u64 end = async_extent->start + async_extent->ram_size - 1; | |
2b83a0ee | 1067 | int ret; |
7027f871 CH |
1068 | struct writeback_control wbc = { |
1069 | .sync_mode = WB_SYNC_ALL, | |
1070 | .range_start = start, | |
1071 | .range_end = end, | |
1072 | .no_cgroup_owner = 1, | |
1073 | }; | |
771ed689 | 1074 | |
256b0cf9 CH |
1075 | wbc_attach_fdatawrite_inode(&wbc, &inode->vfs_inode); |
1076 | ret = run_delalloc_cow(inode, locked_page, start, end, &wbc, false); | |
1077 | wbc_detach_inode(&wbc); | |
2b83a0ee | 1078 | if (ret < 0) { |
71aa147b NA |
1079 | btrfs_cleanup_ordered_extents(inode, locked_page, start, end - start + 1); |
1080 | if (locked_page) { | |
1081 | const u64 page_start = page_offset(locked_page); | |
71aa147b | 1082 | |
71aa147b NA |
1083 | set_page_writeback(locked_page); |
1084 | end_page_writeback(locked_page); | |
9783e4de CH |
1085 | btrfs_mark_ordered_io_finished(inode, locked_page, |
1086 | page_start, PAGE_SIZE, | |
1087 | !ret); | |
1088 | btrfs_page_clear_uptodate(inode->root->fs_info, | |
1089 | locked_page, page_start, | |
1090 | PAGE_SIZE); | |
1091 | mapping_set_error(locked_page->mapping, ret); | |
2b83a0ee | 1092 | unlock_page(locked_page); |
71aa147b | 1093 | } |
2b83a0ee | 1094 | } |
2b83a0ee | 1095 | } |
79787eaa | 1096 | |
84f262f0 CH |
1097 | static void submit_one_async_extent(struct async_chunk *async_chunk, |
1098 | struct async_extent *async_extent, | |
1099 | u64 *alloc_hint) | |
771ed689 | 1100 | { |
84f262f0 | 1101 | struct btrfs_inode *inode = async_chunk->inode; |
b4ccace8 QW |
1102 | struct extent_io_tree *io_tree = &inode->io_tree; |
1103 | struct btrfs_root *root = inode->root; | |
1104 | struct btrfs_fs_info *fs_info = root->fs_info; | |
d611935b | 1105 | struct btrfs_ordered_extent *ordered; |
771ed689 | 1106 | struct btrfs_key ins; |
2b83a0ee | 1107 | struct page *locked_page = NULL; |
771ed689 | 1108 | struct extent_map *em; |
f5a84ee3 | 1109 | int ret = 0; |
b4ccace8 QW |
1110 | u64 start = async_extent->start; |
1111 | u64 end = async_extent->start + async_extent->ram_size - 1; | |
771ed689 | 1112 | |
896d7c1a CH |
1113 | if (async_chunk->blkcg_css) |
1114 | kthread_associate_blkcg(async_chunk->blkcg_css); | |
1115 | ||
2b83a0ee QW |
1116 | /* |
1117 | * If async_chunk->locked_page is in the async_extent range, we need to | |
1118 | * handle it. | |
1119 | */ | |
1120 | if (async_chunk->locked_page) { | |
1121 | u64 locked_page_start = page_offset(async_chunk->locked_page); | |
1122 | u64 locked_page_end = locked_page_start + PAGE_SIZE - 1; | |
3e04e7f1 | 1123 | |
2b83a0ee QW |
1124 | if (!(start >= locked_page_end || end <= locked_page_start)) |
1125 | locked_page = async_chunk->locked_page; | |
b4ccace8 | 1126 | } |
570eb97b | 1127 | lock_extent(io_tree, start, end, NULL); |
ce62003f | 1128 | |
67583468 | 1129 | if (async_extent->compress_type == BTRFS_COMPRESS_NONE) { |
ff20d6a4 | 1130 | submit_uncompressed_range(inode, async_extent, locked_page); |
e43a6210 CH |
1131 | goto done; |
1132 | } | |
ce62003f | 1133 | |
b4ccace8 QW |
1134 | ret = btrfs_reserve_extent(root, async_extent->ram_size, |
1135 | async_extent->compressed_size, | |
1136 | async_extent->compressed_size, | |
1137 | 0, *alloc_hint, &ins, 1, 1); | |
1138 | if (ret) { | |
c2167754 | 1139 | /* |
b4ccace8 QW |
1140 | * Here we used to try again by going back to non-compressed |
1141 | * path for ENOSPC. But we can't reserve space even for | |
1142 | * compressed size, how could it work for uncompressed size | |
1143 | * which requires larger size? So here we directly go error | |
1144 | * path. | |
c2167754 | 1145 | */ |
b4ccace8 QW |
1146 | goto out_free; |
1147 | } | |
1148 | ||
1149 | /* Here we're doing allocation and writeback of the compressed pages */ | |
1150 | em = create_io_em(inode, start, | |
1151 | async_extent->ram_size, /* len */ | |
1152 | start, /* orig_start */ | |
1153 | ins.objectid, /* block_start */ | |
1154 | ins.offset, /* block_len */ | |
1155 | ins.offset, /* orig_block_len */ | |
1156 | async_extent->ram_size, /* ram_bytes */ | |
1157 | async_extent->compress_type, | |
1158 | BTRFS_ORDERED_COMPRESSED); | |
1159 | if (IS_ERR(em)) { | |
1160 | ret = PTR_ERR(em); | |
1161 | goto out_free_reserve; | |
1162 | } | |
1163 | free_extent_map(em); | |
771ed689 | 1164 | |
d611935b | 1165 | ordered = btrfs_alloc_ordered_extent(inode, start, /* file_offset */ |
cb36a9bb OS |
1166 | async_extent->ram_size, /* num_bytes */ |
1167 | async_extent->ram_size, /* ram_bytes */ | |
1168 | ins.objectid, /* disk_bytenr */ | |
1169 | ins.offset, /* disk_num_bytes */ | |
1170 | 0, /* offset */ | |
1171 | 1 << BTRFS_ORDERED_COMPRESSED, | |
1172 | async_extent->compress_type); | |
d611935b | 1173 | if (IS_ERR(ordered)) { |
4c0c8cfc | 1174 | btrfs_drop_extent_map_range(inode, start, end, false); |
d611935b | 1175 | ret = PTR_ERR(ordered); |
b4ccace8 | 1176 | goto out_free_reserve; |
771ed689 | 1177 | } |
b4ccace8 QW |
1178 | btrfs_dec_block_group_reservations(fs_info, ins.objectid); |
1179 | ||
1180 | /* Clear dirty, set writeback and unlock the pages. */ | |
1181 | extent_clear_unlock_delalloc(inode, start, end, | |
1182 | NULL, EXTENT_LOCKED | EXTENT_DELALLOC, | |
1183 | PAGE_UNLOCK | PAGE_START_WRITEBACK); | |
d611935b | 1184 | btrfs_submit_compressed_write(ordered, |
b4ccace8 QW |
1185 | async_extent->pages, /* compressed_pages */ |
1186 | async_extent->nr_pages, | |
05d06a5c | 1187 | async_chunk->write_flags, true); |
b4ccace8 | 1188 | *alloc_hint = ins.objectid + ins.offset; |
e43a6210 | 1189 | done: |
896d7c1a CH |
1190 | if (async_chunk->blkcg_css) |
1191 | kthread_associate_blkcg(NULL); | |
b4ccace8 | 1192 | kfree(async_extent); |
84f262f0 | 1193 | return; |
b4ccace8 | 1194 | |
3e04e7f1 | 1195 | out_free_reserve: |
0b246afa | 1196 | btrfs_dec_block_group_reservations(fs_info, ins.objectid); |
2ff7e61e | 1197 | btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1); |
79787eaa | 1198 | out_free: |
a994310a | 1199 | mapping_set_error(inode->vfs_inode.i_mapping, -EIO); |
b4ccace8 | 1200 | extent_clear_unlock_delalloc(inode, start, end, |
c2790a2e | 1201 | NULL, EXTENT_LOCKED | EXTENT_DELALLOC | |
a7e3b975 | 1202 | EXTENT_DELALLOC_NEW | |
151a41bc | 1203 | EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING, |
6869b0a8 | 1204 | PAGE_UNLOCK | PAGE_START_WRITEBACK | |
a994310a | 1205 | PAGE_END_WRITEBACK); |
40ae837b | 1206 | free_async_extent_pages(async_extent); |
84f262f0 CH |
1207 | if (async_chunk->blkcg_css) |
1208 | kthread_associate_blkcg(NULL); | |
1209 | btrfs_debug(fs_info, | |
b4ccace8 | 1210 | "async extent submission failed root=%lld inode=%llu start=%llu len=%llu ret=%d", |
84f262f0 CH |
1211 | root->root_key.objectid, btrfs_ino(inode), start, |
1212 | async_extent->ram_size, ret); | |
1213 | kfree(async_extent); | |
771ed689 CM |
1214 | } |
1215 | ||
43c69849 | 1216 | static u64 get_extent_allocation_hint(struct btrfs_inode *inode, u64 start, |
4b46fce2 JB |
1217 | u64 num_bytes) |
1218 | { | |
43c69849 | 1219 | struct extent_map_tree *em_tree = &inode->extent_tree; |
4b46fce2 JB |
1220 | struct extent_map *em; |
1221 | u64 alloc_hint = 0; | |
1222 | ||
1223 | read_lock(&em_tree->lock); | |
1224 | em = search_extent_mapping(em_tree, start, num_bytes); | |
1225 | if (em) { | |
1226 | /* | |
1227 | * if block start isn't an actual block number then find the | |
1228 | * first block in this inode and use that as a hint. If that | |
1229 | * block is also bogus then just don't worry about it. | |
1230 | */ | |
1231 | if (em->block_start >= EXTENT_MAP_LAST_BYTE) { | |
1232 | free_extent_map(em); | |
1233 | em = search_extent_mapping(em_tree, 0, 0); | |
1234 | if (em && em->block_start < EXTENT_MAP_LAST_BYTE) | |
1235 | alloc_hint = em->block_start; | |
1236 | if (em) | |
1237 | free_extent_map(em); | |
1238 | } else { | |
1239 | alloc_hint = em->block_start; | |
1240 | free_extent_map(em); | |
1241 | } | |
1242 | } | |
1243 | read_unlock(&em_tree->lock); | |
1244 | ||
1245 | return alloc_hint; | |
1246 | } | |
1247 | ||
771ed689 CM |
1248 | /* |
1249 | * when extent_io.c finds a delayed allocation range in the file, | |
1250 | * the call backs end up in this code. The basic idea is to | |
1251 | * allocate extents on disk for the range, and create ordered data structs | |
1252 | * in ram to track those extents. | |
1253 | * | |
1254 | * locked_page is the page that writepage had locked already. We use | |
1255 | * it to make sure we don't do extra locks or unlocks. | |
1256 | * | |
ba9145ad | 1257 | * When this function fails, it unlocks all pages except @locked_page. |
9ce7466f | 1258 | * |
c56cbe90 CH |
1259 | * When this function successfully creates an inline extent, it returns 1 and |
1260 | * unlocks all pages including locked_page and starts I/O on them. | |
ba9145ad CH |
1261 | * (In reality inline extents are limited to a single page, so locked_page is |
1262 | * the only page handled anyway). | |
9ce7466f | 1263 | * |
ba9145ad CH |
1264 | * When this function succeed and creates a normal extent, the page locking |
1265 | * status depends on the passed in flags: | |
9ce7466f | 1266 | * |
ba9145ad CH |
1267 | * - If @keep_locked is set, all pages are kept locked. |
1268 | * - Else all pages except for @locked_page are unlocked. | |
9ce7466f NA |
1269 | * |
1270 | * When a failure happens in the second or later iteration of the | |
1271 | * while-loop, the ordered extents created in previous iterations are kept | |
1272 | * intact. So, the caller must clean them up by calling | |
1273 | * btrfs_cleanup_ordered_extents(). See btrfs_run_delalloc_range() for | |
1274 | * example. | |
771ed689 | 1275 | */ |
6e26c442 | 1276 | static noinline int cow_file_range(struct btrfs_inode *inode, |
c56cbe90 CH |
1277 | struct page *locked_page, u64 start, u64 end, |
1278 | u64 *done_offset, | |
53ffb30a | 1279 | bool keep_locked, bool no_inline) |
771ed689 | 1280 | { |
6e26c442 NB |
1281 | struct btrfs_root *root = inode->root; |
1282 | struct btrfs_fs_info *fs_info = root->fs_info; | |
771ed689 | 1283 | u64 alloc_hint = 0; |
9ce7466f | 1284 | u64 orig_start = start; |
771ed689 CM |
1285 | u64 num_bytes; |
1286 | unsigned long ram_size; | |
a315e68f | 1287 | u64 cur_alloc_size = 0; |
432cd2a1 | 1288 | u64 min_alloc_size; |
0b246afa | 1289 | u64 blocksize = fs_info->sectorsize; |
771ed689 CM |
1290 | struct btrfs_key ins; |
1291 | struct extent_map *em; | |
a315e68f FM |
1292 | unsigned clear_bits; |
1293 | unsigned long page_ops; | |
1294 | bool extent_reserved = false; | |
771ed689 CM |
1295 | int ret = 0; |
1296 | ||
6e26c442 | 1297 | if (btrfs_is_free_space_inode(inode)) { |
29bce2f3 JB |
1298 | ret = -EINVAL; |
1299 | goto out_unlock; | |
02ecd2c2 | 1300 | } |
771ed689 | 1301 | |
fda2832f | 1302 | num_bytes = ALIGN(end - start + 1, blocksize); |
771ed689 | 1303 | num_bytes = max(blocksize, num_bytes); |
566b1760 | 1304 | ASSERT(num_bytes <= btrfs_super_total_bytes(fs_info->super_copy)); |
771ed689 | 1305 | |
6e26c442 | 1306 | inode_should_defrag(inode, start, end, num_bytes, SZ_64K); |
4cb5300b | 1307 | |
7367253a QW |
1308 | /* |
1309 | * Due to the page size limit, for subpage we can only trigger the | |
1310 | * writeback for the dirty sectors of page, that means data writeback | |
1311 | * is doing more writeback than what we want. | |
1312 | * | |
1313 | * This is especially unexpected for some call sites like fallocate, | |
1314 | * where we only increase i_size after everything is done. | |
1315 | * This means we can trigger inline extent even if we didn't want to. | |
1316 | * So here we skip inline extent creation completely. | |
1317 | */ | |
53ffb30a | 1318 | if (start == 0 && fs_info->sectorsize == PAGE_SIZE && !no_inline) { |
8dd9872d OS |
1319 | u64 actual_end = min_t(u64, i_size_read(&inode->vfs_inode), |
1320 | end + 1); | |
1321 | ||
771ed689 | 1322 | /* lets try to make an inline extent */ |
8dd9872d | 1323 | ret = cow_file_range_inline(inode, actual_end, 0, |
d9496e8a | 1324 | BTRFS_COMPRESS_NONE, NULL, false); |
771ed689 | 1325 | if (ret == 0) { |
8b62f87b JB |
1326 | /* |
1327 | * We use DO_ACCOUNTING here because we need the | |
1328 | * delalloc_release_metadata to be run _after_ we drop | |
1329 | * our outstanding extent for clearing delalloc for this | |
1330 | * range. | |
1331 | */ | |
4750af3b QW |
1332 | extent_clear_unlock_delalloc(inode, start, end, |
1333 | locked_page, | |
c2790a2e | 1334 | EXTENT_LOCKED | EXTENT_DELALLOC | |
8b62f87b JB |
1335 | EXTENT_DELALLOC_NEW | EXTENT_DEFRAG | |
1336 | EXTENT_DO_ACCOUNTING, PAGE_UNLOCK | | |
6869b0a8 | 1337 | PAGE_START_WRITEBACK | PAGE_END_WRITEBACK); |
4750af3b QW |
1338 | /* |
1339 | * locked_page is locked by the caller of | |
1340 | * writepage_delalloc(), not locked by | |
1341 | * __process_pages_contig(). | |
1342 | * | |
1343 | * We can't let __process_pages_contig() to unlock it, | |
1344 | * as it doesn't have any subpage::writers recorded. | |
1345 | * | |
1346 | * Here we manually unlock the page, since the caller | |
c56cbe90 CH |
1347 | * can't determine if it's an inline extent or a |
1348 | * compressed extent. | |
4750af3b QW |
1349 | */ |
1350 | unlock_page(locked_page); | |
6e144bf1 CH |
1351 | ret = 1; |
1352 | goto done; | |
79787eaa | 1353 | } else if (ret < 0) { |
79787eaa | 1354 | goto out_unlock; |
771ed689 CM |
1355 | } |
1356 | } | |
1357 | ||
6e26c442 | 1358 | alloc_hint = get_extent_allocation_hint(inode, start, num_bytes); |
771ed689 | 1359 | |
432cd2a1 FM |
1360 | /* |
1361 | * Relocation relies on the relocated extents to have exactly the same | |
1362 | * size as the original extents. Normally writeback for relocation data | |
1363 | * extents follows a NOCOW path because relocation preallocates the | |
1364 | * extents. However, due to an operation such as scrub turning a block | |
1365 | * group to RO mode, it may fallback to COW mode, so we must make sure | |
1366 | * an extent allocated during COW has exactly the requested size and can | |
1367 | * not be split into smaller extents, otherwise relocation breaks and | |
1368 | * fails during the stage where it updates the bytenr of file extent | |
1369 | * items. | |
1370 | */ | |
37f00a6d | 1371 | if (btrfs_is_data_reloc_root(root)) |
432cd2a1 FM |
1372 | min_alloc_size = num_bytes; |
1373 | else | |
1374 | min_alloc_size = fs_info->sectorsize; | |
1375 | ||
3752d22f | 1376 | while (num_bytes > 0) { |
34bfaf15 CH |
1377 | struct btrfs_ordered_extent *ordered; |
1378 | ||
3752d22f | 1379 | cur_alloc_size = num_bytes; |
18513091 | 1380 | ret = btrfs_reserve_extent(root, cur_alloc_size, cur_alloc_size, |
432cd2a1 | 1381 | min_alloc_size, 0, alloc_hint, |
e570fd27 | 1382 | &ins, 1, 1); |
6e144bf1 CH |
1383 | if (ret == -EAGAIN) { |
1384 | /* | |
1385 | * btrfs_reserve_extent only returns -EAGAIN for zoned | |
1386 | * file systems, which is an indication that there are | |
1387 | * no active zones to allocate from at the moment. | |
1388 | * | |
1389 | * If this is the first loop iteration, wait for at | |
1390 | * least one zone to finish before retrying the | |
1391 | * allocation. Otherwise ask the caller to write out | |
1392 | * the already allocated blocks before coming back to | |
1393 | * us, or return -ENOSPC if it can't handle retries. | |
1394 | */ | |
1395 | ASSERT(btrfs_is_zoned(fs_info)); | |
1396 | if (start == orig_start) { | |
1397 | wait_on_bit_io(&inode->root->fs_info->flags, | |
1398 | BTRFS_FS_NEED_ZONE_FINISH, | |
1399 | TASK_UNINTERRUPTIBLE); | |
1400 | continue; | |
1401 | } | |
1402 | if (done_offset) { | |
1403 | *done_offset = start - 1; | |
1404 | return 0; | |
1405 | } | |
1406 | ret = -ENOSPC; | |
1407 | } | |
00361589 | 1408 | if (ret < 0) |
79787eaa | 1409 | goto out_unlock; |
a315e68f FM |
1410 | cur_alloc_size = ins.offset; |
1411 | extent_reserved = true; | |
d397712b | 1412 | |
771ed689 | 1413 | ram_size = ins.offset; |
6e26c442 | 1414 | em = create_io_em(inode, start, ins.offset, /* len */ |
6f9994db LB |
1415 | start, /* orig_start */ |
1416 | ins.objectid, /* block_start */ | |
1417 | ins.offset, /* block_len */ | |
1418 | ins.offset, /* orig_block_len */ | |
1419 | ram_size, /* ram_bytes */ | |
1420 | BTRFS_COMPRESS_NONE, /* compress_type */ | |
1af4a0aa | 1421 | BTRFS_ORDERED_REGULAR /* type */); |
090a127a SY |
1422 | if (IS_ERR(em)) { |
1423 | ret = PTR_ERR(em); | |
ace68bac | 1424 | goto out_reserve; |
090a127a | 1425 | } |
6f9994db | 1426 | free_extent_map(em); |
e6dcd2dc | 1427 | |
34bfaf15 CH |
1428 | ordered = btrfs_alloc_ordered_extent(inode, start, ram_size, |
1429 | ram_size, ins.objectid, cur_alloc_size, | |
1430 | 0, 1 << BTRFS_ORDERED_REGULAR, | |
1431 | BTRFS_COMPRESS_NONE); | |
1432 | if (IS_ERR(ordered)) { | |
1433 | ret = PTR_ERR(ordered); | |
d9f85963 | 1434 | goto out_drop_extent_cache; |
34bfaf15 | 1435 | } |
c8b97818 | 1436 | |
37f00a6d | 1437 | if (btrfs_is_data_reloc_root(root)) { |
34bfaf15 CH |
1438 | ret = btrfs_reloc_clone_csums(ordered); |
1439 | ||
4dbd80fb QW |
1440 | /* |
1441 | * Only drop cache here, and process as normal. | |
1442 | * | |
1443 | * We must not allow extent_clear_unlock_delalloc() | |
1444 | * at out_unlock label to free meta of this ordered | |
1445 | * extent, as its meta should be freed by | |
1446 | * btrfs_finish_ordered_io(). | |
1447 | * | |
1448 | * So we must continue until @start is increased to | |
1449 | * skip current ordered extent. | |
1450 | */ | |
00361589 | 1451 | if (ret) |
4c0c8cfc FM |
1452 | btrfs_drop_extent_map_range(inode, start, |
1453 | start + ram_size - 1, | |
1454 | false); | |
17d217fe | 1455 | } |
34bfaf15 | 1456 | btrfs_put_ordered_extent(ordered); |
17d217fe | 1457 | |
0b246afa | 1458 | btrfs_dec_block_group_reservations(fs_info, ins.objectid); |
9cfa3e34 | 1459 | |
f57ad937 QW |
1460 | /* |
1461 | * We're not doing compressed IO, don't unlock the first page | |
1462 | * (which the caller expects to stay locked), don't clear any | |
1463 | * dirty bits and don't set any writeback bits | |
8b62b72b | 1464 | * |
f57ad937 QW |
1465 | * Do set the Ordered (Private2) bit so we know this page was |
1466 | * properly setup for writepage. | |
c8b97818 | 1467 | */ |
ba9145ad | 1468 | page_ops = (keep_locked ? 0 : PAGE_UNLOCK); |
f57ad937 | 1469 | page_ops |= PAGE_SET_ORDERED; |
a791e35e | 1470 | |
6e26c442 | 1471 | extent_clear_unlock_delalloc(inode, start, start + ram_size - 1, |
74e9194a | 1472 | locked_page, |
c2790a2e | 1473 | EXTENT_LOCKED | EXTENT_DELALLOC, |
a315e68f | 1474 | page_ops); |
3752d22f AJ |
1475 | if (num_bytes < cur_alloc_size) |
1476 | num_bytes = 0; | |
4dbd80fb | 1477 | else |
3752d22f | 1478 | num_bytes -= cur_alloc_size; |
c59f8951 CM |
1479 | alloc_hint = ins.objectid + ins.offset; |
1480 | start += cur_alloc_size; | |
a315e68f | 1481 | extent_reserved = false; |
4dbd80fb QW |
1482 | |
1483 | /* | |
1484 | * btrfs_reloc_clone_csums() error, since start is increased | |
1485 | * extent_clear_unlock_delalloc() at out_unlock label won't | |
1486 | * free metadata of current ordered extent, we're OK to exit. | |
1487 | */ | |
1488 | if (ret) | |
1489 | goto out_unlock; | |
b888db2b | 1490 | } |
6e144bf1 CH |
1491 | done: |
1492 | if (done_offset) | |
1493 | *done_offset = end; | |
be20aa9d | 1494 | return ret; |
b7d5b0a8 | 1495 | |
d9f85963 | 1496 | out_drop_extent_cache: |
4c0c8cfc | 1497 | btrfs_drop_extent_map_range(inode, start, start + ram_size - 1, false); |
ace68bac | 1498 | out_reserve: |
0b246afa | 1499 | btrfs_dec_block_group_reservations(fs_info, ins.objectid); |
2ff7e61e | 1500 | btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1); |
79787eaa | 1501 | out_unlock: |
9ce7466f NA |
1502 | /* |
1503 | * Now, we have three regions to clean up: | |
1504 | * | |
1505 | * |-------(1)----|---(2)---|-------------(3)----------| | |
1506 | * `- orig_start `- start `- start + cur_alloc_size `- end | |
1507 | * | |
1508 | * We process each region below. | |
1509 | */ | |
1510 | ||
a7e3b975 FM |
1511 | clear_bits = EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DELALLOC_NEW | |
1512 | EXTENT_DEFRAG | EXTENT_CLEAR_META_RESV; | |
6869b0a8 | 1513 | page_ops = PAGE_UNLOCK | PAGE_START_WRITEBACK | PAGE_END_WRITEBACK; |
9ce7466f | 1514 | |
a315e68f | 1515 | /* |
9ce7466f NA |
1516 | * For the range (1). We have already instantiated the ordered extents |
1517 | * for this region. They are cleaned up by | |
1518 | * btrfs_cleanup_ordered_extents() in e.g, | |
1519 | * btrfs_run_delalloc_range(). EXTENT_LOCKED | EXTENT_DELALLOC are | |
1520 | * already cleared in the above loop. And, EXTENT_DELALLOC_NEW | | |
1521 | * EXTENT_DEFRAG | EXTENT_CLEAR_META_RESV are handled by the cleanup | |
1522 | * function. | |
1523 | * | |
ba9145ad | 1524 | * However, in case of @keep_locked, we still need to unlock the pages |
9ce7466f NA |
1525 | * (except @locked_page) to ensure all the pages are unlocked. |
1526 | */ | |
ba9145ad | 1527 | if (keep_locked && orig_start < start) { |
71aa147b NA |
1528 | if (!locked_page) |
1529 | mapping_set_error(inode->vfs_inode.i_mapping, ret); | |
9ce7466f NA |
1530 | extent_clear_unlock_delalloc(inode, orig_start, start - 1, |
1531 | locked_page, 0, page_ops); | |
71aa147b | 1532 | } |
9ce7466f | 1533 | |
a315e68f | 1534 | /* |
9ce7466f NA |
1535 | * For the range (2). If we reserved an extent for our delalloc range |
1536 | * (or a subrange) and failed to create the respective ordered extent, | |
1537 | * then it means that when we reserved the extent we decremented the | |
1538 | * extent's size from the data space_info's bytes_may_use counter and | |
1539 | * incremented the space_info's bytes_reserved counter by the same | |
1540 | * amount. We must make sure extent_clear_unlock_delalloc() does not try | |
1541 | * to decrement again the data space_info's bytes_may_use counter, | |
1542 | * therefore we do not pass it the flag EXTENT_CLEAR_DATA_RESV. | |
a315e68f FM |
1543 | */ |
1544 | if (extent_reserved) { | |
6e26c442 | 1545 | extent_clear_unlock_delalloc(inode, start, |
e2c8e92d | 1546 | start + cur_alloc_size - 1, |
a315e68f FM |
1547 | locked_page, |
1548 | clear_bits, | |
1549 | page_ops); | |
1550 | start += cur_alloc_size; | |
a315e68f | 1551 | } |
9ce7466f NA |
1552 | |
1553 | /* | |
1554 | * For the range (3). We never touched the region. In addition to the | |
1555 | * clear_bits above, we add EXTENT_CLEAR_DATA_RESV to release the data | |
1556 | * space_info's bytes_may_use counter, reserved in | |
1557 | * btrfs_check_data_free_space(). | |
1558 | */ | |
12b2d64e CH |
1559 | if (start < end) { |
1560 | clear_bits |= EXTENT_CLEAR_DATA_RESV; | |
1561 | extent_clear_unlock_delalloc(inode, start, end, locked_page, | |
1562 | clear_bits, page_ops); | |
1563 | } | |
aaafa1eb | 1564 | return ret; |
771ed689 | 1565 | } |
c8b97818 | 1566 | |
771ed689 | 1567 | /* |
c15d8cf2 CH |
1568 | * Phase two of compressed writeback. This is the ordered portion of the code, |
1569 | * which only gets called in the order the work was queued. We walk all the | |
1570 | * async extents created by compress_file_range and send them down to the disk. | |
771ed689 | 1571 | */ |
00d31d17 | 1572 | static noinline void submit_compressed_extents(struct btrfs_work *work) |
771ed689 | 1573 | { |
c5a68aec NB |
1574 | struct async_chunk *async_chunk = container_of(work, struct async_chunk, |
1575 | work); | |
1576 | struct btrfs_fs_info *fs_info = btrfs_work_owner(work); | |
00d31d17 | 1577 | struct async_extent *async_extent; |
771ed689 | 1578 | unsigned long nr_pages; |
00d31d17 | 1579 | u64 alloc_hint = 0; |
771ed689 | 1580 | |
b5326271 | 1581 | nr_pages = (async_chunk->end - async_chunk->start + PAGE_SIZE) >> |
09cbfeaf | 1582 | PAGE_SHIFT; |
771ed689 | 1583 | |
00d31d17 CH |
1584 | while (!list_empty(&async_chunk->extents)) { |
1585 | async_extent = list_entry(async_chunk->extents.next, | |
1586 | struct async_extent, list); | |
1587 | list_del(&async_extent->list); | |
1588 | submit_one_async_extent(async_chunk, async_extent, &alloc_hint); | |
1589 | } | |
ac98141d JB |
1590 | |
1591 | /* atomic_sub_return implies a barrier */ | |
1592 | if (atomic_sub_return(nr_pages, &fs_info->async_delalloc_pages) < | |
1593 | 5 * SZ_1M) | |
1594 | cond_wake_up_nomb(&fs_info->async_submit_wait); | |
771ed689 | 1595 | } |
c8b97818 | 1596 | |
771ed689 CM |
1597 | static noinline void async_cow_free(struct btrfs_work *work) |
1598 | { | |
b5326271 | 1599 | struct async_chunk *async_chunk; |
9e895a8f | 1600 | struct async_cow *async_cow; |
97db1204 | 1601 | |
b5326271 | 1602 | async_chunk = container_of(work, struct async_chunk, work); |
3134508e | 1603 | btrfs_add_delayed_iput(async_chunk->inode); |
ec39f769 CM |
1604 | if (async_chunk->blkcg_css) |
1605 | css_put(async_chunk->blkcg_css); | |
9e895a8f QW |
1606 | |
1607 | async_cow = async_chunk->async_cow; | |
1608 | if (atomic_dec_and_test(&async_cow->num_chunks)) | |
1609 | kvfree(async_cow); | |
771ed689 CM |
1610 | } |
1611 | ||
bb7b05fe | 1612 | static bool run_delalloc_compressed(struct btrfs_inode *inode, |
c56cbe90 CH |
1613 | struct page *locked_page, u64 start, |
1614 | u64 end, struct writeback_control *wbc) | |
771ed689 | 1615 | { |
751b6431 | 1616 | struct btrfs_fs_info *fs_info = inode->root->fs_info; |
ec39f769 | 1617 | struct cgroup_subsys_state *blkcg_css = wbc_blkcg_css(wbc); |
97db1204 NB |
1618 | struct async_cow *ctx; |
1619 | struct async_chunk *async_chunk; | |
771ed689 | 1620 | unsigned long nr_pages; |
97db1204 NB |
1621 | u64 num_chunks = DIV_ROUND_UP(end - start, SZ_512K); |
1622 | int i; | |
b1c16ac9 | 1623 | unsigned nofs_flag; |
bf9486d6 | 1624 | const blk_opf_t write_flags = wbc_to_write_flags(wbc); |
771ed689 | 1625 | |
b1c16ac9 NB |
1626 | nofs_flag = memalloc_nofs_save(); |
1627 | ctx = kvmalloc(struct_size(ctx, chunks, num_chunks), GFP_KERNEL); | |
1628 | memalloc_nofs_restore(nofs_flag); | |
973fb26e CH |
1629 | if (!ctx) |
1630 | return false; | |
b1c16ac9 | 1631 | |
973fb26e CH |
1632 | unlock_extent(&inode->io_tree, start, end, NULL); |
1633 | set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT, &inode->runtime_flags); | |
97db1204 NB |
1634 | |
1635 | async_chunk = ctx->chunks; | |
1636 | atomic_set(&ctx->num_chunks, num_chunks); | |
1637 | ||
1638 | for (i = 0; i < num_chunks; i++) { | |
973fb26e | 1639 | u64 cur_end = min(end, start + SZ_512K - 1); |
771ed689 | 1640 | |
bd4691a0 NB |
1641 | /* |
1642 | * igrab is called higher up in the call chain, take only the | |
1643 | * lightweight reference for the callback lifetime | |
1644 | */ | |
751b6431 | 1645 | ihold(&inode->vfs_inode); |
9e895a8f | 1646 | async_chunk[i].async_cow = ctx; |
99a81a44 | 1647 | async_chunk[i].inode = inode; |
97db1204 NB |
1648 | async_chunk[i].start = start; |
1649 | async_chunk[i].end = cur_end; | |
97db1204 NB |
1650 | async_chunk[i].write_flags = write_flags; |
1651 | INIT_LIST_HEAD(&async_chunk[i].extents); | |
1652 | ||
1d53c9e6 CM |
1653 | /* |
1654 | * The locked_page comes all the way from writepage and its | |
1655 | * the original page we were actually given. As we spread | |
1656 | * this large delalloc region across multiple async_chunk | |
1657 | * structs, only the first struct needs a pointer to locked_page | |
1658 | * | |
1659 | * This way we don't need racey decisions about who is supposed | |
1660 | * to unlock it. | |
1661 | */ | |
1662 | if (locked_page) { | |
ec39f769 CM |
1663 | /* |
1664 | * Depending on the compressibility, the pages might or | |
1665 | * might not go through async. We want all of them to | |
1666 | * be accounted against wbc once. Let's do it here | |
1667 | * before the paths diverge. wbc accounting is used | |
1668 | * only for foreign writeback detection and doesn't | |
1669 | * need full accuracy. Just account the whole thing | |
1670 | * against the first page. | |
1671 | */ | |
1672 | wbc_account_cgroup_owner(wbc, locked_page, | |
1673 | cur_end - start); | |
1d53c9e6 CM |
1674 | async_chunk[i].locked_page = locked_page; |
1675 | locked_page = NULL; | |
1676 | } else { | |
1677 | async_chunk[i].locked_page = NULL; | |
1678 | } | |
1679 | ||
ec39f769 CM |
1680 | if (blkcg_css != blkcg_root_css) { |
1681 | css_get(blkcg_css); | |
1682 | async_chunk[i].blkcg_css = blkcg_css; | |
3480373e | 1683 | async_chunk[i].write_flags |= REQ_BTRFS_CGROUP_PUNT; |
ec39f769 CM |
1684 | } else { |
1685 | async_chunk[i].blkcg_css = NULL; | |
1686 | } | |
1687 | ||
c15d8cf2 | 1688 | btrfs_init_work(&async_chunk[i].work, compress_file_range, |
00d31d17 | 1689 | submit_compressed_extents, async_cow_free); |
771ed689 | 1690 | |
97db1204 | 1691 | nr_pages = DIV_ROUND_UP(cur_end - start, PAGE_SIZE); |
0b246afa | 1692 | atomic_add(nr_pages, &fs_info->async_delalloc_pages); |
771ed689 | 1693 | |
97db1204 | 1694 | btrfs_queue_work(fs_info->delalloc_workers, &async_chunk[i].work); |
771ed689 | 1695 | |
771ed689 CM |
1696 | start = cur_end + 1; |
1697 | } | |
973fb26e | 1698 | return true; |
be20aa9d CM |
1699 | } |
1700 | ||
256b0cf9 CH |
1701 | /* |
1702 | * Run the delalloc range from start to end, and write back any dirty pages | |
1703 | * covered by the range. | |
1704 | */ | |
1705 | static noinline int run_delalloc_cow(struct btrfs_inode *inode, | |
1706 | struct page *locked_page, u64 start, | |
1707 | u64 end, struct writeback_control *wbc, | |
1708 | bool pages_dirty) | |
42c01100 | 1709 | { |
898793d9 | 1710 | u64 done_offset = end; |
42c01100 NA |
1711 | int ret; |
1712 | ||
898793d9 | 1713 | while (start <= end) { |
c56cbe90 CH |
1714 | ret = cow_file_range(inode, locked_page, start, end, &done_offset, |
1715 | true, false); | |
6e144bf1 | 1716 | if (ret) |
898793d9 | 1717 | return ret; |
778b8785 | 1718 | extent_write_locked_range(&inode->vfs_inode, locked_page, start, |
256b0cf9 | 1719 | done_offset, wbc, pages_dirty); |
898793d9 NA |
1720 | start = done_offset + 1; |
1721 | } | |
42c01100 | 1722 | |
c56cbe90 | 1723 | return 1; |
42c01100 NA |
1724 | } |
1725 | ||
2ff7e61e | 1726 | static noinline int csum_exist_in_range(struct btrfs_fs_info *fs_info, |
26ce9114 | 1727 | u64 bytenr, u64 num_bytes, bool nowait) |
17d217fe | 1728 | { |
fc28b25e | 1729 | struct btrfs_root *csum_root = btrfs_csum_root(fs_info, bytenr); |
17d217fe | 1730 | struct btrfs_ordered_sum *sums; |
fc28b25e | 1731 | int ret; |
17d217fe YZ |
1732 | LIST_HEAD(list); |
1733 | ||
97e38239 QW |
1734 | ret = btrfs_lookup_csums_list(csum_root, bytenr, bytenr + num_bytes - 1, |
1735 | &list, 0, nowait); | |
17d217fe YZ |
1736 | if (ret == 0 && list_empty(&list)) |
1737 | return 0; | |
1738 | ||
1739 | while (!list_empty(&list)) { | |
1740 | sums = list_entry(list.next, struct btrfs_ordered_sum, list); | |
1741 | list_del(&sums->list); | |
1742 | kfree(sums); | |
1743 | } | |
58113753 LB |
1744 | if (ret < 0) |
1745 | return ret; | |
17d217fe YZ |
1746 | return 1; |
1747 | } | |
1748 | ||
8ba96f3d | 1749 | static int fallback_to_cow(struct btrfs_inode *inode, struct page *locked_page, |
53ffb30a | 1750 | const u64 start, const u64 end) |
467dc47e | 1751 | { |
8ba96f3d | 1752 | const bool is_space_ino = btrfs_is_free_space_inode(inode); |
37f00a6d | 1753 | const bool is_reloc_ino = btrfs_is_data_reloc_root(inode->root); |
2166e5ed | 1754 | const u64 range_bytes = end + 1 - start; |
8ba96f3d | 1755 | struct extent_io_tree *io_tree = &inode->io_tree; |
467dc47e FM |
1756 | u64 range_start = start; |
1757 | u64 count; | |
53ffb30a | 1758 | int ret; |
467dc47e FM |
1759 | |
1760 | /* | |
1761 | * If EXTENT_NORESERVE is set it means that when the buffered write was | |
1762 | * made we had not enough available data space and therefore we did not | |
1763 | * reserve data space for it, since we though we could do NOCOW for the | |
1764 | * respective file range (either there is prealloc extent or the inode | |
1765 | * has the NOCOW bit set). | |
1766 | * | |
1767 | * However when we need to fallback to COW mode (because for example the | |
1768 | * block group for the corresponding extent was turned to RO mode by a | |
1769 | * scrub or relocation) we need to do the following: | |
1770 | * | |
1771 | * 1) We increment the bytes_may_use counter of the data space info. | |
1772 | * If COW succeeds, it allocates a new data extent and after doing | |
1773 | * that it decrements the space info's bytes_may_use counter and | |
1774 | * increments its bytes_reserved counter by the same amount (we do | |
1775 | * this at btrfs_add_reserved_bytes()). So we need to increment the | |
1776 | * bytes_may_use counter to compensate (when space is reserved at | |
1777 | * buffered write time, the bytes_may_use counter is incremented); | |
1778 | * | |
1779 | * 2) We clear the EXTENT_NORESERVE bit from the range. We do this so | |
1780 | * that if the COW path fails for any reason, it decrements (through | |
1781 | * extent_clear_unlock_delalloc()) the bytes_may_use counter of the | |
1782 | * data space info, which we incremented in the step above. | |
2166e5ed FM |
1783 | * |
1784 | * If we need to fallback to cow and the inode corresponds to a free | |
6bd335b4 FM |
1785 | * space cache inode or an inode of the data relocation tree, we must |
1786 | * also increment bytes_may_use of the data space_info for the same | |
1787 | * reason. Space caches and relocated data extents always get a prealloc | |
2166e5ed | 1788 | * extent for them, however scrub or balance may have set the block |
6bd335b4 FM |
1789 | * group that contains that extent to RO mode and therefore force COW |
1790 | * when starting writeback. | |
467dc47e | 1791 | */ |
2166e5ed | 1792 | count = count_range_bits(io_tree, &range_start, end, range_bytes, |
8c6e53a7 | 1793 | EXTENT_NORESERVE, 0, NULL); |
6bd335b4 FM |
1794 | if (count > 0 || is_space_ino || is_reloc_ino) { |
1795 | u64 bytes = count; | |
8ba96f3d | 1796 | struct btrfs_fs_info *fs_info = inode->root->fs_info; |
467dc47e FM |
1797 | struct btrfs_space_info *sinfo = fs_info->data_sinfo; |
1798 | ||
6bd335b4 FM |
1799 | if (is_space_ino || is_reloc_ino) |
1800 | bytes = range_bytes; | |
1801 | ||
467dc47e | 1802 | spin_lock(&sinfo->lock); |
2166e5ed | 1803 | btrfs_space_info_update_bytes_may_use(fs_info, sinfo, bytes); |
467dc47e FM |
1804 | spin_unlock(&sinfo->lock); |
1805 | ||
2166e5ed FM |
1806 | if (count > 0) |
1807 | clear_extent_bit(io_tree, start, end, EXTENT_NORESERVE, | |
bd015294 | 1808 | NULL); |
467dc47e FM |
1809 | } |
1810 | ||
53ffb30a CH |
1811 | /* |
1812 | * Don't try to create inline extents, as a mix of inline extent that | |
1813 | * is written out and unlocked directly and a normal NOCOW extent | |
1814 | * doesn't work. | |
1815 | */ | |
c56cbe90 CH |
1816 | ret = cow_file_range(inode, locked_page, start, end, NULL, false, true); |
1817 | ASSERT(ret != 1); | |
53ffb30a | 1818 | return ret; |
467dc47e FM |
1819 | } |
1820 | ||
619104ba FM |
1821 | struct can_nocow_file_extent_args { |
1822 | /* Input fields. */ | |
1823 | ||
1824 | /* Start file offset of the range we want to NOCOW. */ | |
1825 | u64 start; | |
1826 | /* End file offset (inclusive) of the range we want to NOCOW. */ | |
1827 | u64 end; | |
1828 | bool writeback_path; | |
1829 | bool strict; | |
1830 | /* | |
1831 | * Free the path passed to can_nocow_file_extent() once it's not needed | |
1832 | * anymore. | |
1833 | */ | |
1834 | bool free_path; | |
1835 | ||
1836 | /* Output fields. Only set when can_nocow_file_extent() returns 1. */ | |
1837 | ||
1838 | u64 disk_bytenr; | |
1839 | u64 disk_num_bytes; | |
1840 | u64 extent_offset; | |
1841 | /* Number of bytes that can be written to in NOCOW mode. */ | |
1842 | u64 num_bytes; | |
1843 | }; | |
1844 | ||
1845 | /* | |
1846 | * Check if we can NOCOW the file extent that the path points to. | |
1847 | * This function may return with the path released, so the caller should check | |
1848 | * if path->nodes[0] is NULL or not if it needs to use the path afterwards. | |
1849 | * | |
1850 | * Returns: < 0 on error | |
1851 | * 0 if we can not NOCOW | |
1852 | * 1 if we can NOCOW | |
1853 | */ | |
1854 | static int can_nocow_file_extent(struct btrfs_path *path, | |
1855 | struct btrfs_key *key, | |
1856 | struct btrfs_inode *inode, | |
1857 | struct can_nocow_file_extent_args *args) | |
1858 | { | |
1859 | const bool is_freespace_inode = btrfs_is_free_space_inode(inode); | |
1860 | struct extent_buffer *leaf = path->nodes[0]; | |
1861 | struct btrfs_root *root = inode->root; | |
1862 | struct btrfs_file_extent_item *fi; | |
1863 | u64 extent_end; | |
1864 | u8 extent_type; | |
1865 | int can_nocow = 0; | |
1866 | int ret = 0; | |
26ce9114 | 1867 | bool nowait = path->nowait; |
619104ba FM |
1868 | |
1869 | fi = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_file_extent_item); | |
1870 | extent_type = btrfs_file_extent_type(leaf, fi); | |
1871 | ||
1872 | if (extent_type == BTRFS_FILE_EXTENT_INLINE) | |
1873 | goto out; | |
1874 | ||
1875 | /* Can't access these fields unless we know it's not an inline extent. */ | |
1876 | args->disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); | |
1877 | args->disk_num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi); | |
1878 | args->extent_offset = btrfs_file_extent_offset(leaf, fi); | |
1879 | ||
1880 | if (!(inode->flags & BTRFS_INODE_NODATACOW) && | |
1881 | extent_type == BTRFS_FILE_EXTENT_REG) | |
1882 | goto out; | |
1883 | ||
1884 | /* | |
1885 | * If the extent was created before the generation where the last snapshot | |
1886 | * for its subvolume was created, then this implies the extent is shared, | |
1887 | * hence we must COW. | |
1888 | */ | |
a7bb6bd4 | 1889 | if (!args->strict && |
619104ba FM |
1890 | btrfs_file_extent_generation(leaf, fi) <= |
1891 | btrfs_root_last_snapshot(&root->root_item)) | |
1892 | goto out; | |
1893 | ||
1894 | /* An explicit hole, must COW. */ | |
1895 | if (args->disk_bytenr == 0) | |
1896 | goto out; | |
1897 | ||
1898 | /* Compressed/encrypted/encoded extents must be COWed. */ | |
1899 | if (btrfs_file_extent_compression(leaf, fi) || | |
1900 | btrfs_file_extent_encryption(leaf, fi) || | |
1901 | btrfs_file_extent_other_encoding(leaf, fi)) | |
1902 | goto out; | |
1903 | ||
1904 | extent_end = btrfs_file_extent_end(path); | |
1905 | ||
1906 | /* | |
1907 | * The following checks can be expensive, as they need to take other | |
1908 | * locks and do btree or rbtree searches, so release the path to avoid | |
1909 | * blocking other tasks for too long. | |
1910 | */ | |
1911 | btrfs_release_path(path); | |
1912 | ||
1913 | ret = btrfs_cross_ref_exist(root, btrfs_ino(inode), | |
1914 | key->offset - args->extent_offset, | |
deccae40 | 1915 | args->disk_bytenr, args->strict, path); |
619104ba FM |
1916 | WARN_ON_ONCE(ret > 0 && is_freespace_inode); |
1917 | if (ret != 0) | |
1918 | goto out; | |
1919 | ||
1920 | if (args->free_path) { | |
1921 | /* | |
1922 | * We don't need the path anymore, plus through the | |
1923 | * csum_exist_in_range() call below we will end up allocating | |
1924 | * another path. So free the path to avoid unnecessary extra | |
1925 | * memory usage. | |
1926 | */ | |
1927 | btrfs_free_path(path); | |
1928 | path = NULL; | |
1929 | } | |
1930 | ||
1931 | /* If there are pending snapshots for this root, we must COW. */ | |
1932 | if (args->writeback_path && !is_freespace_inode && | |
1933 | atomic_read(&root->snapshot_force_cow)) | |
1934 | goto out; | |
1935 | ||
1936 | args->disk_bytenr += args->extent_offset; | |
1937 | args->disk_bytenr += args->start - key->offset; | |
1938 | args->num_bytes = min(args->end + 1, extent_end) - args->start; | |
1939 | ||
1940 | /* | |
1941 | * Force COW if csums exist in the range. This ensures that csums for a | |
1942 | * given extent are either valid or do not exist. | |
1943 | */ | |
26ce9114 JB |
1944 | ret = csum_exist_in_range(root->fs_info, args->disk_bytenr, args->num_bytes, |
1945 | nowait); | |
619104ba FM |
1946 | WARN_ON_ONCE(ret > 0 && is_freespace_inode); |
1947 | if (ret != 0) | |
1948 | goto out; | |
1949 | ||
1950 | can_nocow = 1; | |
1951 | out: | |
1952 | if (args->free_path && path) | |
1953 | btrfs_free_path(path); | |
1954 | ||
1955 | return ret < 0 ? ret : can_nocow; | |
1956 | } | |
1957 | ||
d352ac68 CM |
1958 | /* |
1959 | * when nowcow writeback call back. This checks for snapshots or COW copies | |
1960 | * of the extents that exist in the file, and COWs the file as required. | |
1961 | * | |
1962 | * If no cow copies or snapshots exist, we write directly to the existing | |
1963 | * blocks on disk | |
1964 | */ | |
968322c8 | 1965 | static noinline int run_delalloc_nocow(struct btrfs_inode *inode, |
7f366cfe | 1966 | struct page *locked_page, |
53ffb30a | 1967 | const u64 start, const u64 end) |
be20aa9d | 1968 | { |
968322c8 NB |
1969 | struct btrfs_fs_info *fs_info = inode->root->fs_info; |
1970 | struct btrfs_root *root = inode->root; | |
be20aa9d | 1971 | struct btrfs_path *path; |
3e024846 NB |
1972 | u64 cow_start = (u64)-1; |
1973 | u64 cur_offset = start; | |
8ecebf4d | 1974 | int ret; |
3e024846 | 1975 | bool check_prev = true; |
968322c8 | 1976 | u64 ino = btrfs_ino(inode); |
619104ba | 1977 | struct can_nocow_file_extent_args nocow_args = { 0 }; |
be20aa9d | 1978 | |
76c5126e CH |
1979 | /* |
1980 | * Normally on a zoned device we're only doing COW writes, but in case | |
1981 | * of relocation on a zoned filesystem serializes I/O so that we're only | |
1982 | * writing sequentially and can end up here as well. | |
1983 | */ | |
1984 | ASSERT(!btrfs_is_zoned(fs_info) || btrfs_is_data_reloc_root(root)); | |
1985 | ||
be20aa9d | 1986 | path = btrfs_alloc_path(); |
17ca04af | 1987 | if (!path) { |
38dc8889 CH |
1988 | ret = -ENOMEM; |
1989 | goto error; | |
17ca04af | 1990 | } |
82d5902d | 1991 | |
619104ba FM |
1992 | nocow_args.end = end; |
1993 | nocow_args.writeback_path = true; | |
1994 | ||
80ff3856 | 1995 | while (1) { |
18f62b86 | 1996 | struct btrfs_block_group *nocow_bg = NULL; |
34bfaf15 | 1997 | struct btrfs_ordered_extent *ordered; |
3e024846 NB |
1998 | struct btrfs_key found_key; |
1999 | struct btrfs_file_extent_item *fi; | |
2000 | struct extent_buffer *leaf; | |
2001 | u64 extent_end; | |
3e024846 | 2002 | u64 ram_bytes; |
619104ba | 2003 | u64 nocow_end; |
3e024846 | 2004 | int extent_type; |
3daea5fd | 2005 | bool is_prealloc; |
762bf098 | 2006 | |
e4c3b2dc | 2007 | ret = btrfs_lookup_file_extent(NULL, root, path, ino, |
80ff3856 | 2008 | cur_offset, 0); |
d788a349 | 2009 | if (ret < 0) |
79787eaa | 2010 | goto error; |
a6bd9cd1 NB |
2011 | |
2012 | /* | |
2013 | * If there is no extent for our range when doing the initial | |
2014 | * search, then go back to the previous slot as it will be the | |
2015 | * one containing the search offset | |
2016 | */ | |
80ff3856 YZ |
2017 | if (ret > 0 && path->slots[0] > 0 && check_prev) { |
2018 | leaf = path->nodes[0]; | |
2019 | btrfs_item_key_to_cpu(leaf, &found_key, | |
2020 | path->slots[0] - 1); | |
33345d01 | 2021 | if (found_key.objectid == ino && |
80ff3856 YZ |
2022 | found_key.type == BTRFS_EXTENT_DATA_KEY) |
2023 | path->slots[0]--; | |
2024 | } | |
3e024846 | 2025 | check_prev = false; |
80ff3856 | 2026 | next_slot: |
a6bd9cd1 | 2027 | /* Go to next leaf if we have exhausted the current one */ |
80ff3856 YZ |
2028 | leaf = path->nodes[0]; |
2029 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
2030 | ret = btrfs_next_leaf(root, path); | |
953fa5ce | 2031 | if (ret < 0) |
79787eaa | 2032 | goto error; |
80ff3856 YZ |
2033 | if (ret > 0) |
2034 | break; | |
2035 | leaf = path->nodes[0]; | |
2036 | } | |
be20aa9d | 2037 | |
80ff3856 YZ |
2038 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
2039 | ||
a6bd9cd1 | 2040 | /* Didn't find anything for our INO */ |
1d512cb7 FM |
2041 | if (found_key.objectid > ino) |
2042 | break; | |
a6bd9cd1 NB |
2043 | /* |
2044 | * Keep searching until we find an EXTENT_ITEM or there are no | |
2045 | * more extents for this inode | |
2046 | */ | |
1d512cb7 FM |
2047 | if (WARN_ON_ONCE(found_key.objectid < ino) || |
2048 | found_key.type < BTRFS_EXTENT_DATA_KEY) { | |
2049 | path->slots[0]++; | |
2050 | goto next_slot; | |
2051 | } | |
a6bd9cd1 NB |
2052 | |
2053 | /* Found key is not EXTENT_DATA_KEY or starts after req range */ | |
1d512cb7 | 2054 | if (found_key.type > BTRFS_EXTENT_DATA_KEY || |
80ff3856 YZ |
2055 | found_key.offset > end) |
2056 | break; | |
2057 | ||
a6bd9cd1 NB |
2058 | /* |
2059 | * If the found extent starts after requested offset, then | |
2060 | * adjust extent_end to be right before this extent begins | |
2061 | */ | |
80ff3856 YZ |
2062 | if (found_key.offset > cur_offset) { |
2063 | extent_end = found_key.offset; | |
e9061e21 | 2064 | extent_type = 0; |
18f62b86 | 2065 | goto must_cow; |
80ff3856 YZ |
2066 | } |
2067 | ||
a6bd9cd1 NB |
2068 | /* |
2069 | * Found extent which begins before our range and potentially | |
2070 | * intersect it | |
2071 | */ | |
80ff3856 YZ |
2072 | fi = btrfs_item_ptr(leaf, path->slots[0], |
2073 | struct btrfs_file_extent_item); | |
2074 | extent_type = btrfs_file_extent_type(leaf, fi); | |
619104ba FM |
2075 | /* If this is triggered then we have a memory corruption. */ |
2076 | ASSERT(extent_type < BTRFS_NR_FILE_EXTENT_TYPES); | |
2077 | if (WARN_ON(extent_type >= BTRFS_NR_FILE_EXTENT_TYPES)) { | |
2078 | ret = -EUCLEAN; | |
2079 | goto error; | |
2080 | } | |
cc95bef6 | 2081 | ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi); |
619104ba | 2082 | extent_end = btrfs_file_extent_end(path); |
c65ca98f | 2083 | |
619104ba FM |
2084 | /* |
2085 | * If the extent we got ends before our current offset, skip to | |
2086 | * the next extent. | |
2087 | */ | |
2088 | if (extent_end <= cur_offset) { | |
2089 | path->slots[0]++; | |
2090 | goto next_slot; | |
2091 | } | |
c65ca98f | 2092 | |
619104ba FM |
2093 | nocow_args.start = cur_offset; |
2094 | ret = can_nocow_file_extent(path, &found_key, inode, &nocow_args); | |
953fa5ce | 2095 | if (ret < 0) |
619104ba | 2096 | goto error; |
953fa5ce | 2097 | if (ret == 0) |
18f62b86 | 2098 | goto must_cow; |
58113753 | 2099 | |
619104ba | 2100 | ret = 0; |
18f62b86 CH |
2101 | nocow_bg = btrfs_inc_nocow_writers(fs_info, nocow_args.disk_bytenr); |
2102 | if (!nocow_bg) { | |
2103 | must_cow: | |
2104 | /* | |
2105 | * If we can't perform NOCOW writeback for the range, | |
2106 | * then record the beginning of the range that needs to | |
2107 | * be COWed. It will be written out before the next | |
2108 | * NOCOW range if we find one, or when exiting this | |
2109 | * loop. | |
2110 | */ | |
80ff3856 YZ |
2111 | if (cow_start == (u64)-1) |
2112 | cow_start = cur_offset; | |
2113 | cur_offset = extent_end; | |
2114 | if (cur_offset > end) | |
2115 | break; | |
c65ca98f FM |
2116 | if (!path->nodes[0]) |
2117 | continue; | |
80ff3856 YZ |
2118 | path->slots[0]++; |
2119 | goto next_slot; | |
7ea394f1 YZ |
2120 | } |
2121 | ||
a6bd9cd1 NB |
2122 | /* |
2123 | * COW range from cow_start to found_key.offset - 1. As the key | |
2124 | * will contain the beginning of the first extent that can be | |
2125 | * NOCOW, following one which needs to be COW'ed | |
2126 | */ | |
80ff3856 | 2127 | if (cow_start != (u64)-1) { |
968322c8 | 2128 | ret = fallback_to_cow(inode, locked_page, |
53ffb30a | 2129 | cow_start, found_key.offset - 1); |
80ff3856 | 2130 | cow_start = (u64)-1; |
18f62b86 CH |
2131 | if (ret) { |
2132 | btrfs_dec_nocow_writers(nocow_bg); | |
79787eaa | 2133 | goto error; |
18f62b86 | 2134 | } |
7ea394f1 | 2135 | } |
80ff3856 | 2136 | |
619104ba | 2137 | nocow_end = cur_offset + nocow_args.num_bytes - 1; |
3daea5fd CH |
2138 | is_prealloc = extent_type == BTRFS_FILE_EXTENT_PREALLOC; |
2139 | if (is_prealloc) { | |
619104ba | 2140 | u64 orig_start = found_key.offset - nocow_args.extent_offset; |
3e024846 | 2141 | struct extent_map *em; |
6f9994db | 2142 | |
619104ba | 2143 | em = create_io_em(inode, cur_offset, nocow_args.num_bytes, |
6f9994db | 2144 | orig_start, |
619104ba FM |
2145 | nocow_args.disk_bytenr, /* block_start */ |
2146 | nocow_args.num_bytes, /* block_len */ | |
2147 | nocow_args.disk_num_bytes, /* orig_block_len */ | |
6f9994db LB |
2148 | ram_bytes, BTRFS_COMPRESS_NONE, |
2149 | BTRFS_ORDERED_PREALLOC); | |
2150 | if (IS_ERR(em)) { | |
18f62b86 | 2151 | btrfs_dec_nocow_writers(nocow_bg); |
6f9994db LB |
2152 | ret = PTR_ERR(em); |
2153 | goto error; | |
d899e052 | 2154 | } |
6f9994db | 2155 | free_extent_map(em); |
3daea5fd CH |
2156 | } |
2157 | ||
34bfaf15 | 2158 | ordered = btrfs_alloc_ordered_extent(inode, cur_offset, |
3daea5fd CH |
2159 | nocow_args.num_bytes, nocow_args.num_bytes, |
2160 | nocow_args.disk_bytenr, nocow_args.num_bytes, 0, | |
2161 | is_prealloc | |
2162 | ? (1 << BTRFS_ORDERED_PREALLOC) | |
2163 | : (1 << BTRFS_ORDERED_NOCOW), | |
2164 | BTRFS_COMPRESS_NONE); | |
18f62b86 | 2165 | btrfs_dec_nocow_writers(nocow_bg); |
34bfaf15 | 2166 | if (IS_ERR(ordered)) { |
3daea5fd | 2167 | if (is_prealloc) { |
4c0c8cfc FM |
2168 | btrfs_drop_extent_map_range(inode, cur_offset, |
2169 | nocow_end, false); | |
762bf098 | 2170 | } |
34bfaf15 | 2171 | ret = PTR_ERR(ordered); |
3daea5fd | 2172 | goto error; |
d899e052 | 2173 | } |
80ff3856 | 2174 | |
37f00a6d | 2175 | if (btrfs_is_data_reloc_root(root)) |
4dbd80fb QW |
2176 | /* |
2177 | * Error handled later, as we must prevent | |
2178 | * extent_clear_unlock_delalloc() in error handler | |
2179 | * from freeing metadata of created ordered extent. | |
2180 | */ | |
34bfaf15 CH |
2181 | ret = btrfs_reloc_clone_csums(ordered); |
2182 | btrfs_put_ordered_extent(ordered); | |
efa56464 | 2183 | |
619104ba | 2184 | extent_clear_unlock_delalloc(inode, cur_offset, nocow_end, |
c2790a2e | 2185 | locked_page, EXTENT_LOCKED | |
18513091 WX |
2186 | EXTENT_DELALLOC | |
2187 | EXTENT_CLEAR_DATA_RESV, | |
f57ad937 | 2188 | PAGE_UNLOCK | PAGE_SET_ORDERED); |
18513091 | 2189 | |
80ff3856 | 2190 | cur_offset = extent_end; |
4dbd80fb QW |
2191 | |
2192 | /* | |
2193 | * btrfs_reloc_clone_csums() error, now we're OK to call error | |
2194 | * handler, as metadata for created ordered extent will only | |
2195 | * be freed by btrfs_finish_ordered_io(). | |
2196 | */ | |
2197 | if (ret) | |
2198 | goto error; | |
80ff3856 YZ |
2199 | if (cur_offset > end) |
2200 | break; | |
be20aa9d | 2201 | } |
b3b4aa74 | 2202 | btrfs_release_path(path); |
80ff3856 | 2203 | |
506481b2 | 2204 | if (cur_offset <= end && cow_start == (u64)-1) |
80ff3856 | 2205 | cow_start = cur_offset; |
17ca04af | 2206 | |
80ff3856 | 2207 | if (cow_start != (u64)-1) { |
506481b2 | 2208 | cur_offset = end; |
53ffb30a | 2209 | ret = fallback_to_cow(inode, locked_page, cow_start, end); |
953fa5ce | 2210 | cow_start = (u64)-1; |
d788a349 | 2211 | if (ret) |
79787eaa | 2212 | goto error; |
80ff3856 YZ |
2213 | } |
2214 | ||
18f62b86 CH |
2215 | btrfs_free_path(path); |
2216 | return 0; | |
762bf098 | 2217 | |
18f62b86 | 2218 | error: |
953fa5ce CH |
2219 | /* |
2220 | * If an error happened while a COW region is outstanding, cur_offset | |
2221 | * needs to be reset to cow_start to ensure the COW region is unlocked | |
2222 | * as well. | |
2223 | */ | |
2224 | if (cow_start != (u64)-1) | |
2225 | cur_offset = cow_start; | |
18f62b86 | 2226 | if (cur_offset < end) |
968322c8 | 2227 | extent_clear_unlock_delalloc(inode, cur_offset, end, |
c2790a2e | 2228 | locked_page, EXTENT_LOCKED | |
151a41bc JB |
2229 | EXTENT_DELALLOC | EXTENT_DEFRAG | |
2230 | EXTENT_DO_ACCOUNTING, PAGE_UNLOCK | | |
6869b0a8 | 2231 | PAGE_START_WRITEBACK | |
c2790a2e | 2232 | PAGE_END_WRITEBACK); |
7ea394f1 | 2233 | btrfs_free_path(path); |
79787eaa | 2234 | return ret; |
be20aa9d CM |
2235 | } |
2236 | ||
6e65ae76 | 2237 | static bool should_nocow(struct btrfs_inode *inode, u64 start, u64 end) |
47059d93 | 2238 | { |
6e65ae76 GR |
2239 | if (inode->flags & (BTRFS_INODE_NODATACOW | BTRFS_INODE_PREALLOC)) { |
2240 | if (inode->defrag_bytes && | |
2241 | test_range_bit(&inode->io_tree, start, end, EXTENT_DEFRAG, | |
2242 | 0, NULL)) | |
2243 | return false; | |
2244 | return true; | |
2245 | } | |
2246 | return false; | |
47059d93 WS |
2247 | } |
2248 | ||
d352ac68 | 2249 | /* |
5eaad97a NB |
2250 | * Function to process delayed allocation (create CoW) for ranges which are |
2251 | * being touched for the first time. | |
d352ac68 | 2252 | */ |
98456b9c | 2253 | int btrfs_run_delalloc_range(struct btrfs_inode *inode, struct page *locked_page, |
c56cbe90 | 2254 | u64 start, u64 end, struct writeback_control *wbc) |
be20aa9d | 2255 | { |
42c01100 | 2256 | const bool zoned = btrfs_is_zoned(inode->root->fs_info); |
c56cbe90 | 2257 | int ret; |
a2135011 | 2258 | |
2749f7ef | 2259 | /* |
c56cbe90 CH |
2260 | * The range must cover part of the @locked_page, or a return of 1 |
2261 | * can confuse the caller. | |
2749f7ef QW |
2262 | */ |
2263 | ASSERT(!(end <= page_offset(locked_page) || | |
2264 | start >= page_offset(locked_page) + PAGE_SIZE)); | |
2265 | ||
6e65ae76 | 2266 | if (should_nocow(inode, start, end)) { |
53ffb30a | 2267 | ret = run_delalloc_nocow(inode, locked_page, start, end); |
973fb26e | 2268 | goto out; |
7ddf5a42 | 2269 | } |
973fb26e CH |
2270 | |
2271 | if (btrfs_inode_can_compress(inode) && | |
2272 | inode_need_compress(inode, start, end) && | |
c56cbe90 CH |
2273 | run_delalloc_compressed(inode, locked_page, start, end, wbc)) |
2274 | return 1; | |
973fb26e CH |
2275 | |
2276 | if (zoned) | |
256b0cf9 CH |
2277 | ret = run_delalloc_cow(inode, locked_page, start, end, wbc, |
2278 | true); | |
973fb26e | 2279 | else |
c56cbe90 CH |
2280 | ret = cow_file_range(inode, locked_page, start, end, NULL, |
2281 | false, false); | |
973fb26e CH |
2282 | |
2283 | out: | |
c56cbe90 | 2284 | if (ret < 0) |
98456b9c | 2285 | btrfs_cleanup_ordered_extents(inode, locked_page, start, |
d1051d6e | 2286 | end - start + 1); |
b888db2b CM |
2287 | return ret; |
2288 | } | |
2289 | ||
62798a49 | 2290 | void btrfs_split_delalloc_extent(struct btrfs_inode *inode, |
abbb55f4 | 2291 | struct extent_state *orig, u64 split) |
9ed74f2d | 2292 | { |
62798a49 | 2293 | struct btrfs_fs_info *fs_info = inode->root->fs_info; |
dcab6a3b JB |
2294 | u64 size; |
2295 | ||
0ca1f7ce | 2296 | /* not delalloc, ignore it */ |
9ed74f2d | 2297 | if (!(orig->state & EXTENT_DELALLOC)) |
1bf85046 | 2298 | return; |
9ed74f2d | 2299 | |
dcab6a3b | 2300 | size = orig->end - orig->start + 1; |
f7b12a62 | 2301 | if (size > fs_info->max_extent_size) { |
823bb20a | 2302 | u32 num_extents; |
dcab6a3b JB |
2303 | u64 new_size; |
2304 | ||
2305 | /* | |
5c848198 | 2306 | * See the explanation in btrfs_merge_delalloc_extent, the same |
ba117213 | 2307 | * applies here, just in reverse. |
dcab6a3b JB |
2308 | */ |
2309 | new_size = orig->end - split + 1; | |
7d7672bc | 2310 | num_extents = count_max_extents(fs_info, new_size); |
ba117213 | 2311 | new_size = split - orig->start; |
7d7672bc NA |
2312 | num_extents += count_max_extents(fs_info, new_size); |
2313 | if (count_max_extents(fs_info, size) >= num_extents) | |
dcab6a3b JB |
2314 | return; |
2315 | } | |
2316 | ||
62798a49 DS |
2317 | spin_lock(&inode->lock); |
2318 | btrfs_mod_outstanding_extents(inode, 1); | |
2319 | spin_unlock(&inode->lock); | |
9ed74f2d JB |
2320 | } |
2321 | ||
2322 | /* | |
5c848198 NB |
2323 | * Handle merged delayed allocation extents so we can keep track of new extents |
2324 | * that are just merged onto old extents, such as when we are doing sequential | |
2325 | * writes, so we can properly account for the metadata space we'll need. | |
9ed74f2d | 2326 | */ |
2454151c | 2327 | void btrfs_merge_delalloc_extent(struct btrfs_inode *inode, struct extent_state *new, |
5c848198 | 2328 | struct extent_state *other) |
9ed74f2d | 2329 | { |
2454151c | 2330 | struct btrfs_fs_info *fs_info = inode->root->fs_info; |
dcab6a3b | 2331 | u64 new_size, old_size; |
823bb20a | 2332 | u32 num_extents; |
dcab6a3b | 2333 | |
9ed74f2d JB |
2334 | /* not delalloc, ignore it */ |
2335 | if (!(other->state & EXTENT_DELALLOC)) | |
1bf85046 | 2336 | return; |
9ed74f2d | 2337 | |
8461a3de JB |
2338 | if (new->start > other->start) |
2339 | new_size = new->end - other->start + 1; | |
2340 | else | |
2341 | new_size = other->end - new->start + 1; | |
dcab6a3b JB |
2342 | |
2343 | /* we're not bigger than the max, unreserve the space and go */ | |
f7b12a62 | 2344 | if (new_size <= fs_info->max_extent_size) { |
2454151c DS |
2345 | spin_lock(&inode->lock); |
2346 | btrfs_mod_outstanding_extents(inode, -1); | |
2347 | spin_unlock(&inode->lock); | |
dcab6a3b JB |
2348 | return; |
2349 | } | |
2350 | ||
2351 | /* | |
ba117213 JB |
2352 | * We have to add up either side to figure out how many extents were |
2353 | * accounted for before we merged into one big extent. If the number of | |
2354 | * extents we accounted for is <= the amount we need for the new range | |
2355 | * then we can return, otherwise drop. Think of it like this | |
2356 | * | |
2357 | * [ 4k][MAX_SIZE] | |
2358 | * | |
2359 | * So we've grown the extent by a MAX_SIZE extent, this would mean we | |
2360 | * need 2 outstanding extents, on one side we have 1 and the other side | |
2361 | * we have 1 so they are == and we can return. But in this case | |
2362 | * | |
2363 | * [MAX_SIZE+4k][MAX_SIZE+4k] | |
2364 | * | |
2365 | * Each range on their own accounts for 2 extents, but merged together | |
2366 | * they are only 3 extents worth of accounting, so we need to drop in | |
2367 | * this case. | |
dcab6a3b | 2368 | */ |
ba117213 | 2369 | old_size = other->end - other->start + 1; |
7d7672bc | 2370 | num_extents = count_max_extents(fs_info, old_size); |
ba117213 | 2371 | old_size = new->end - new->start + 1; |
7d7672bc NA |
2372 | num_extents += count_max_extents(fs_info, old_size); |
2373 | if (count_max_extents(fs_info, new_size) >= num_extents) | |
dcab6a3b JB |
2374 | return; |
2375 | ||
2454151c DS |
2376 | spin_lock(&inode->lock); |
2377 | btrfs_mod_outstanding_extents(inode, -1); | |
2378 | spin_unlock(&inode->lock); | |
9ed74f2d JB |
2379 | } |
2380 | ||
eb73c1b7 | 2381 | static void btrfs_add_delalloc_inodes(struct btrfs_root *root, |
82ca5a04 | 2382 | struct btrfs_inode *inode) |
eb73c1b7 | 2383 | { |
82ca5a04 | 2384 | struct btrfs_fs_info *fs_info = inode->root->fs_info; |
0b246afa | 2385 | |
eb73c1b7 | 2386 | spin_lock(&root->delalloc_lock); |
82ca5a04 DS |
2387 | if (list_empty(&inode->delalloc_inodes)) { |
2388 | list_add_tail(&inode->delalloc_inodes, &root->delalloc_inodes); | |
2389 | set_bit(BTRFS_INODE_IN_DELALLOC_LIST, &inode->runtime_flags); | |
eb73c1b7 MX |
2390 | root->nr_delalloc_inodes++; |
2391 | if (root->nr_delalloc_inodes == 1) { | |
0b246afa | 2392 | spin_lock(&fs_info->delalloc_root_lock); |
eb73c1b7 MX |
2393 | BUG_ON(!list_empty(&root->delalloc_root)); |
2394 | list_add_tail(&root->delalloc_root, | |
0b246afa JM |
2395 | &fs_info->delalloc_roots); |
2396 | spin_unlock(&fs_info->delalloc_root_lock); | |
eb73c1b7 MX |
2397 | } |
2398 | } | |
2399 | spin_unlock(&root->delalloc_lock); | |
2400 | } | |
2401 | ||
2b877331 NB |
2402 | void __btrfs_del_delalloc_inode(struct btrfs_root *root, |
2403 | struct btrfs_inode *inode) | |
eb73c1b7 | 2404 | { |
3ffbd68c | 2405 | struct btrfs_fs_info *fs_info = root->fs_info; |
0b246afa | 2406 | |
9e3e97f4 NB |
2407 | if (!list_empty(&inode->delalloc_inodes)) { |
2408 | list_del_init(&inode->delalloc_inodes); | |
eb73c1b7 | 2409 | clear_bit(BTRFS_INODE_IN_DELALLOC_LIST, |
9e3e97f4 | 2410 | &inode->runtime_flags); |
eb73c1b7 MX |
2411 | root->nr_delalloc_inodes--; |
2412 | if (!root->nr_delalloc_inodes) { | |
7c8a0d36 | 2413 | ASSERT(list_empty(&root->delalloc_inodes)); |
0b246afa | 2414 | spin_lock(&fs_info->delalloc_root_lock); |
eb73c1b7 MX |
2415 | BUG_ON(list_empty(&root->delalloc_root)); |
2416 | list_del_init(&root->delalloc_root); | |
0b246afa | 2417 | spin_unlock(&fs_info->delalloc_root_lock); |
eb73c1b7 MX |
2418 | } |
2419 | } | |
2b877331 NB |
2420 | } |
2421 | ||
2422 | static void btrfs_del_delalloc_inode(struct btrfs_root *root, | |
2423 | struct btrfs_inode *inode) | |
2424 | { | |
2425 | spin_lock(&root->delalloc_lock); | |
2426 | __btrfs_del_delalloc_inode(root, inode); | |
eb73c1b7 MX |
2427 | spin_unlock(&root->delalloc_lock); |
2428 | } | |
2429 | ||
d352ac68 | 2430 | /* |
e06a1fc9 NB |
2431 | * Properly track delayed allocation bytes in the inode and to maintain the |
2432 | * list of inodes that have pending delalloc work to be done. | |
d352ac68 | 2433 | */ |
4c5d166f | 2434 | void btrfs_set_delalloc_extent(struct btrfs_inode *inode, struct extent_state *state, |
6d92b304 | 2435 | u32 bits) |
291d673e | 2436 | { |
4c5d166f | 2437 | struct btrfs_fs_info *fs_info = inode->root->fs_info; |
0b246afa | 2438 | |
6d92b304 | 2439 | if ((bits & EXTENT_DEFRAG) && !(bits & EXTENT_DELALLOC)) |
47059d93 | 2440 | WARN_ON(1); |
75eff68e CM |
2441 | /* |
2442 | * set_bit and clear bit hooks normally require _irqsave/restore | |
27160b6b | 2443 | * but in this case, we are only testing for the DELALLOC |
75eff68e CM |
2444 | * bit, which is only set or cleared with irqs on |
2445 | */ | |
6d92b304 | 2446 | if (!(state->state & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) { |
4c5d166f | 2447 | struct btrfs_root *root = inode->root; |
0ca1f7ce | 2448 | u64 len = state->end + 1 - state->start; |
7d7672bc | 2449 | u32 num_extents = count_max_extents(fs_info, len); |
4c5d166f | 2450 | bool do_list = !btrfs_is_free_space_inode(inode); |
9ed74f2d | 2451 | |
4c5d166f DS |
2452 | spin_lock(&inode->lock); |
2453 | btrfs_mod_outstanding_extents(inode, num_extents); | |
2454 | spin_unlock(&inode->lock); | |
287a0ab9 | 2455 | |
6a3891c5 | 2456 | /* For sanity tests */ |
0b246afa | 2457 | if (btrfs_is_testing(fs_info)) |
6a3891c5 JB |
2458 | return; |
2459 | ||
104b4e51 NB |
2460 | percpu_counter_add_batch(&fs_info->delalloc_bytes, len, |
2461 | fs_info->delalloc_batch); | |
4c5d166f DS |
2462 | spin_lock(&inode->lock); |
2463 | inode->delalloc_bytes += len; | |
6d92b304 | 2464 | if (bits & EXTENT_DEFRAG) |
4c5d166f | 2465 | inode->defrag_bytes += len; |
df0af1a5 | 2466 | if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST, |
4c5d166f | 2467 | &inode->runtime_flags)) |
eb73c1b7 | 2468 | btrfs_add_delalloc_inodes(root, inode); |
4c5d166f | 2469 | spin_unlock(&inode->lock); |
291d673e | 2470 | } |
a7e3b975 FM |
2471 | |
2472 | if (!(state->state & EXTENT_DELALLOC_NEW) && | |
6d92b304 | 2473 | (bits & EXTENT_DELALLOC_NEW)) { |
4c5d166f DS |
2474 | spin_lock(&inode->lock); |
2475 | inode->new_delalloc_bytes += state->end + 1 - state->start; | |
2476 | spin_unlock(&inode->lock); | |
a7e3b975 | 2477 | } |
291d673e CM |
2478 | } |
2479 | ||
d352ac68 | 2480 | /* |
a36bb5f9 NB |
2481 | * Once a range is no longer delalloc this function ensures that proper |
2482 | * accounting happens. | |
d352ac68 | 2483 | */ |
bd54766e | 2484 | void btrfs_clear_delalloc_extent(struct btrfs_inode *inode, |
6d92b304 | 2485 | struct extent_state *state, u32 bits) |
291d673e | 2486 | { |
bd54766e | 2487 | struct btrfs_fs_info *fs_info = inode->root->fs_info; |
47059d93 | 2488 | u64 len = state->end + 1 - state->start; |
7d7672bc | 2489 | u32 num_extents = count_max_extents(fs_info, len); |
47059d93 | 2490 | |
6d92b304 | 2491 | if ((state->state & EXTENT_DEFRAG) && (bits & EXTENT_DEFRAG)) { |
4a4b964f | 2492 | spin_lock(&inode->lock); |
6fc0ef68 | 2493 | inode->defrag_bytes -= len; |
4a4b964f FM |
2494 | spin_unlock(&inode->lock); |
2495 | } | |
47059d93 | 2496 | |
75eff68e CM |
2497 | /* |
2498 | * set_bit and clear bit hooks normally require _irqsave/restore | |
27160b6b | 2499 | * but in this case, we are only testing for the DELALLOC |
75eff68e CM |
2500 | * bit, which is only set or cleared with irqs on |
2501 | */ | |
6d92b304 | 2502 | if ((state->state & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) { |
6fc0ef68 | 2503 | struct btrfs_root *root = inode->root; |
83eea1f1 | 2504 | bool do_list = !btrfs_is_free_space_inode(inode); |
bcbfce8a | 2505 | |
8b62f87b JB |
2506 | spin_lock(&inode->lock); |
2507 | btrfs_mod_outstanding_extents(inode, -num_extents); | |
2508 | spin_unlock(&inode->lock); | |
0ca1f7ce | 2509 | |
b6d08f06 JB |
2510 | /* |
2511 | * We don't reserve metadata space for space cache inodes so we | |
52042d8e | 2512 | * don't need to call delalloc_release_metadata if there is an |
b6d08f06 JB |
2513 | * error. |
2514 | */ | |
6d92b304 | 2515 | if (bits & EXTENT_CLEAR_META_RESV && |
0b246afa | 2516 | root != fs_info->tree_root) |
43b18595 | 2517 | btrfs_delalloc_release_metadata(inode, len, false); |
0ca1f7ce | 2518 | |
6a3891c5 | 2519 | /* For sanity tests. */ |
0b246afa | 2520 | if (btrfs_is_testing(fs_info)) |
6a3891c5 JB |
2521 | return; |
2522 | ||
37f00a6d | 2523 | if (!btrfs_is_data_reloc_root(root) && |
a315e68f | 2524 | do_list && !(state->state & EXTENT_NORESERVE) && |
6d92b304 | 2525 | (bits & EXTENT_CLEAR_DATA_RESV)) |
9db5d510 | 2526 | btrfs_free_reserved_data_space_noquota(fs_info, len); |
9ed74f2d | 2527 | |
104b4e51 NB |
2528 | percpu_counter_add_batch(&fs_info->delalloc_bytes, -len, |
2529 | fs_info->delalloc_batch); | |
6fc0ef68 NB |
2530 | spin_lock(&inode->lock); |
2531 | inode->delalloc_bytes -= len; | |
2532 | if (do_list && inode->delalloc_bytes == 0 && | |
df0af1a5 | 2533 | test_bit(BTRFS_INODE_IN_DELALLOC_LIST, |
9e3e97f4 | 2534 | &inode->runtime_flags)) |
eb73c1b7 | 2535 | btrfs_del_delalloc_inode(root, inode); |
6fc0ef68 | 2536 | spin_unlock(&inode->lock); |
291d673e | 2537 | } |
a7e3b975 FM |
2538 | |
2539 | if ((state->state & EXTENT_DELALLOC_NEW) && | |
6d92b304 | 2540 | (bits & EXTENT_DELALLOC_NEW)) { |
a7e3b975 FM |
2541 | spin_lock(&inode->lock); |
2542 | ASSERT(inode->new_delalloc_bytes >= len); | |
2543 | inode->new_delalloc_bytes -= len; | |
6d92b304 | 2544 | if (bits & EXTENT_ADD_INODE_BYTES) |
2766ff61 | 2545 | inode_add_bytes(&inode->vfs_inode, len); |
a7e3b975 FM |
2546 | spin_unlock(&inode->lock); |
2547 | } | |
291d673e CM |
2548 | } |
2549 | ||
71df088c CH |
2550 | static int btrfs_extract_ordered_extent(struct btrfs_bio *bbio, |
2551 | struct btrfs_ordered_extent *ordered) | |
d22002fd | 2552 | { |
69ccf3f4 CH |
2553 | u64 start = (u64)bbio->bio.bi_iter.bi_sector << SECTOR_SHIFT; |
2554 | u64 len = bbio->bio.bi_iter.bi_size; | |
b0307e28 | 2555 | struct btrfs_ordered_extent *new; |
ebdb44a0 | 2556 | int ret; |
d22002fd | 2557 | |
11d33ab6 | 2558 | /* Must always be called for the beginning of an ordered extent. */ |
7edd339c CH |
2559 | if (WARN_ON_ONCE(start != ordered->disk_bytenr)) |
2560 | return -EINVAL; | |
d22002fd | 2561 | |
11d33ab6 | 2562 | /* No need to split if the ordered extent covers the entire bio. */ |
ec63b84d CH |
2563 | if (ordered->disk_num_bytes == len) { |
2564 | refcount_inc(&ordered->refs); | |
2565 | bbio->ordered = ordered; | |
7edd339c | 2566 | return 0; |
ec63b84d | 2567 | } |
d22002fd | 2568 | |
f0f5329a BB |
2569 | /* |
2570 | * Don't split the extent_map for NOCOW extents, as we're writing into | |
2571 | * a pre-existing one. | |
2572 | */ | |
ebdb44a0 CH |
2573 | if (!test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags)) { |
2574 | ret = split_extent_map(bbio->inode, bbio->file_offset, | |
f000bc6f CH |
2575 | ordered->num_bytes, len, |
2576 | ordered->disk_bytenr); | |
ebdb44a0 CH |
2577 | if (ret) |
2578 | return ret; | |
2579 | } | |
f0f5329a | 2580 | |
b0307e28 CH |
2581 | new = btrfs_split_ordered_extent(ordered, len); |
2582 | if (IS_ERR(new)) | |
2583 | return PTR_ERR(new); | |
ec63b84d | 2584 | bbio->ordered = new; |
b0307e28 | 2585 | return 0; |
d22002fd NA |
2586 | } |
2587 | ||
d352ac68 CM |
2588 | /* |
2589 | * given a list of ordered sums record them in the inode. This happens | |
2590 | * at IO completion time based on sums calculated at bio submission time. | |
2591 | */ | |
510f85ed NB |
2592 | static int add_pending_csums(struct btrfs_trans_handle *trans, |
2593 | struct list_head *list) | |
e6dcd2dc | 2594 | { |
e6dcd2dc | 2595 | struct btrfs_ordered_sum *sum; |
fc28b25e | 2596 | struct btrfs_root *csum_root = NULL; |
ac01f26a | 2597 | int ret; |
e6dcd2dc | 2598 | |
c6e30871 | 2599 | list_for_each_entry(sum, list, list) { |
7c2871a2 | 2600 | trans->adding_csums = true; |
fc28b25e JB |
2601 | if (!csum_root) |
2602 | csum_root = btrfs_csum_root(trans->fs_info, | |
5cfe76f8 | 2603 | sum->logical); |
fc28b25e | 2604 | ret = btrfs_csum_file_blocks(trans, csum_root, sum); |
7c2871a2 | 2605 | trans->adding_csums = false; |
ac01f26a NB |
2606 | if (ret) |
2607 | return ret; | |
e6dcd2dc CM |
2608 | } |
2609 | return 0; | |
2610 | } | |
2611 | ||
c3347309 FM |
2612 | static int btrfs_find_new_delalloc_bytes(struct btrfs_inode *inode, |
2613 | const u64 start, | |
2614 | const u64 len, | |
2615 | struct extent_state **cached_state) | |
2616 | { | |
2617 | u64 search_start = start; | |
2618 | const u64 end = start + len - 1; | |
2619 | ||
2620 | while (search_start < end) { | |
2621 | const u64 search_len = end - search_start + 1; | |
2622 | struct extent_map *em; | |
2623 | u64 em_len; | |
2624 | int ret = 0; | |
2625 | ||
2626 | em = btrfs_get_extent(inode, NULL, 0, search_start, search_len); | |
2627 | if (IS_ERR(em)) | |
2628 | return PTR_ERR(em); | |
2629 | ||
2630 | if (em->block_start != EXTENT_MAP_HOLE) | |
2631 | goto next; | |
2632 | ||
2633 | em_len = em->len; | |
2634 | if (em->start < search_start) | |
2635 | em_len -= search_start - em->start; | |
2636 | if (em_len > search_len) | |
2637 | em_len = search_len; | |
2638 | ||
2639 | ret = set_extent_bit(&inode->io_tree, search_start, | |
2640 | search_start + em_len - 1, | |
1d126800 | 2641 | EXTENT_DELALLOC_NEW, cached_state); |
c3347309 FM |
2642 | next: |
2643 | search_start = extent_map_end(em); | |
2644 | free_extent_map(em); | |
2645 | if (ret) | |
2646 | return ret; | |
2647 | } | |
2648 | return 0; | |
2649 | } | |
2650 | ||
c2566f22 | 2651 | int btrfs_set_extent_delalloc(struct btrfs_inode *inode, u64 start, u64 end, |
e3b8a485 | 2652 | unsigned int extra_bits, |
330a5827 | 2653 | struct extent_state **cached_state) |
ea8c2819 | 2654 | { |
fdb1e121 | 2655 | WARN_ON(PAGE_ALIGNED(end)); |
c3347309 FM |
2656 | |
2657 | if (start >= i_size_read(&inode->vfs_inode) && | |
2658 | !(inode->flags & BTRFS_INODE_PREALLOC)) { | |
2659 | /* | |
2660 | * There can't be any extents following eof in this case so just | |
2661 | * set the delalloc new bit for the range directly. | |
2662 | */ | |
2663 | extra_bits |= EXTENT_DELALLOC_NEW; | |
2664 | } else { | |
2665 | int ret; | |
2666 | ||
2667 | ret = btrfs_find_new_delalloc_bytes(inode, start, | |
2668 | end + 1 - start, | |
2669 | cached_state); | |
2670 | if (ret) | |
2671 | return ret; | |
2672 | } | |
2673 | ||
66240ab1 | 2674 | return set_extent_bit(&inode->io_tree, start, end, |
1d126800 | 2675 | EXTENT_DELALLOC | extra_bits, cached_state); |
ea8c2819 CM |
2676 | } |
2677 | ||
d352ac68 | 2678 | /* see btrfs_writepage_start_hook for details on why this is required */ |
247e743c CM |
2679 | struct btrfs_writepage_fixup { |
2680 | struct page *page; | |
36eeaef5 | 2681 | struct btrfs_inode *inode; |
247e743c CM |
2682 | struct btrfs_work work; |
2683 | }; | |
2684 | ||
b2950863 | 2685 | static void btrfs_writepage_fixup_worker(struct btrfs_work *work) |
247e743c | 2686 | { |
9783e4de CH |
2687 | struct btrfs_writepage_fixup *fixup = |
2688 | container_of(work, struct btrfs_writepage_fixup, work); | |
247e743c | 2689 | struct btrfs_ordered_extent *ordered; |
2ac55d41 | 2690 | struct extent_state *cached_state = NULL; |
364ecf36 | 2691 | struct extent_changeset *data_reserved = NULL; |
9783e4de CH |
2692 | struct page *page = fixup->page; |
2693 | struct btrfs_inode *inode = fixup->inode; | |
2694 | struct btrfs_fs_info *fs_info = inode->root->fs_info; | |
2695 | u64 page_start = page_offset(page); | |
2696 | u64 page_end = page_offset(page) + PAGE_SIZE - 1; | |
25f3c502 | 2697 | int ret = 0; |
f4b1363c | 2698 | bool free_delalloc_space = true; |
247e743c | 2699 | |
f4b1363c JB |
2700 | /* |
2701 | * This is similar to page_mkwrite, we need to reserve the space before | |
2702 | * we take the page lock. | |
2703 | */ | |
65d87f79 NB |
2704 | ret = btrfs_delalloc_reserve_space(inode, &data_reserved, page_start, |
2705 | PAGE_SIZE); | |
4a096752 | 2706 | again: |
247e743c | 2707 | lock_page(page); |
25f3c502 CM |
2708 | |
2709 | /* | |
2710 | * Before we queued this fixup, we took a reference on the page. | |
2711 | * page->mapping may go NULL, but it shouldn't be moved to a different | |
2712 | * address space. | |
2713 | */ | |
f4b1363c JB |
2714 | if (!page->mapping || !PageDirty(page) || !PageChecked(page)) { |
2715 | /* | |
2716 | * Unfortunately this is a little tricky, either | |
2717 | * | |
2718 | * 1) We got here and our page had already been dealt with and | |
2719 | * we reserved our space, thus ret == 0, so we need to just | |
2720 | * drop our space reservation and bail. This can happen the | |
2721 | * first time we come into the fixup worker, or could happen | |
2722 | * while waiting for the ordered extent. | |
2723 | * 2) Our page was already dealt with, but we happened to get an | |
2724 | * ENOSPC above from the btrfs_delalloc_reserve_space. In | |
2725 | * this case we obviously don't have anything to release, but | |
2726 | * because the page was already dealt with we don't want to | |
2727 | * mark the page with an error, so make sure we're resetting | |
2728 | * ret to 0. This is why we have this check _before_ the ret | |
2729 | * check, because we do not want to have a surprise ENOSPC | |
2730 | * when the page was already properly dealt with. | |
2731 | */ | |
2732 | if (!ret) { | |
65d87f79 NB |
2733 | btrfs_delalloc_release_extents(inode, PAGE_SIZE); |
2734 | btrfs_delalloc_release_space(inode, data_reserved, | |
f4b1363c JB |
2735 | page_start, PAGE_SIZE, |
2736 | true); | |
2737 | } | |
2738 | ret = 0; | |
247e743c | 2739 | goto out_page; |
f4b1363c | 2740 | } |
247e743c | 2741 | |
25f3c502 | 2742 | /* |
f4b1363c JB |
2743 | * We can't mess with the page state unless it is locked, so now that |
2744 | * it is locked bail if we failed to make our space reservation. | |
25f3c502 | 2745 | */ |
f4b1363c JB |
2746 | if (ret) |
2747 | goto out_page; | |
247e743c | 2748 | |
570eb97b | 2749 | lock_extent(&inode->io_tree, page_start, page_end, &cached_state); |
4a096752 CM |
2750 | |
2751 | /* already ordered? We're done */ | |
f57ad937 | 2752 | if (PageOrdered(page)) |
f4b1363c | 2753 | goto out_reserved; |
4a096752 | 2754 | |
65d87f79 | 2755 | ordered = btrfs_lookup_ordered_range(inode, page_start, PAGE_SIZE); |
4a096752 | 2756 | if (ordered) { |
570eb97b JB |
2757 | unlock_extent(&inode->io_tree, page_start, page_end, |
2758 | &cached_state); | |
4a096752 | 2759 | unlock_page(page); |
36d45567 | 2760 | btrfs_start_ordered_extent(ordered); |
87826df0 | 2761 | btrfs_put_ordered_extent(ordered); |
4a096752 CM |
2762 | goto again; |
2763 | } | |
247e743c | 2764 | |
65d87f79 | 2765 | ret = btrfs_set_extent_delalloc(inode, page_start, page_end, 0, |
330a5827 | 2766 | &cached_state); |
25f3c502 | 2767 | if (ret) |
53687007 | 2768 | goto out_reserved; |
f3038ee3 | 2769 | |
25f3c502 CM |
2770 | /* |
2771 | * Everything went as planned, we're now the owner of a dirty page with | |
2772 | * delayed allocation bits set and space reserved for our COW | |
2773 | * destination. | |
2774 | * | |
2775 | * The page was dirty when we started, nothing should have cleaned it. | |
2776 | */ | |
2777 | BUG_ON(!PageDirty(page)); | |
f4b1363c | 2778 | free_delalloc_space = false; |
53687007 | 2779 | out_reserved: |
65d87f79 | 2780 | btrfs_delalloc_release_extents(inode, PAGE_SIZE); |
f4b1363c | 2781 | if (free_delalloc_space) |
65d87f79 NB |
2782 | btrfs_delalloc_release_space(inode, data_reserved, page_start, |
2783 | PAGE_SIZE, true); | |
570eb97b | 2784 | unlock_extent(&inode->io_tree, page_start, page_end, &cached_state); |
247e743c | 2785 | out_page: |
25f3c502 CM |
2786 | if (ret) { |
2787 | /* | |
2788 | * We hit ENOSPC or other errors. Update the mapping and page | |
2789 | * to reflect the errors and clean the page. | |
2790 | */ | |
2791 | mapping_set_error(page->mapping, ret); | |
9783e4de CH |
2792 | btrfs_mark_ordered_io_finished(inode, page, page_start, |
2793 | PAGE_SIZE, !ret); | |
2794 | btrfs_page_clear_uptodate(fs_info, page, page_start, PAGE_SIZE); | |
25f3c502 | 2795 | clear_page_dirty_for_io(page); |
25f3c502 | 2796 | } |
9783e4de | 2797 | btrfs_page_clear_checked(fs_info, page, page_start, PAGE_SIZE); |
247e743c | 2798 | unlock_page(page); |
09cbfeaf | 2799 | put_page(page); |
b897abec | 2800 | kfree(fixup); |
364ecf36 | 2801 | extent_changeset_free(data_reserved); |
f4b1363c JB |
2802 | /* |
2803 | * As a precaution, do a delayed iput in case it would be the last iput | |
2804 | * that could need flushing space. Recursing back to fixup worker would | |
2805 | * deadlock. | |
2806 | */ | |
e55cf7ca | 2807 | btrfs_add_delayed_iput(inode); |
247e743c CM |
2808 | } |
2809 | ||
2810 | /* | |
2811 | * There are a few paths in the higher layers of the kernel that directly | |
2812 | * set the page dirty bit without asking the filesystem if it is a | |
2813 | * good idea. This causes problems because we want to make sure COW | |
2814 | * properly happens and the data=ordered rules are followed. | |
2815 | * | |
c8b97818 | 2816 | * In our case any range that doesn't have the ORDERED bit set |
247e743c CM |
2817 | * hasn't been properly setup for IO. We kick off an async process |
2818 | * to fix it up. The async helper will wait for ordered extents, set | |
2819 | * the delalloc bit and make it safe to write the page. | |
2820 | */ | |
a129ffb8 | 2821 | int btrfs_writepage_cow_fixup(struct page *page) |
247e743c CM |
2822 | { |
2823 | struct inode *inode = page->mapping->host; | |
0b246afa | 2824 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
247e743c | 2825 | struct btrfs_writepage_fixup *fixup; |
247e743c | 2826 | |
f57ad937 QW |
2827 | /* This page has ordered extent covering it already */ |
2828 | if (PageOrdered(page)) | |
247e743c CM |
2829 | return 0; |
2830 | ||
25f3c502 CM |
2831 | /* |
2832 | * PageChecked is set below when we create a fixup worker for this page, | |
2833 | * don't try to create another one if we're already PageChecked() | |
2834 | * | |
2835 | * The extent_io writepage code will redirty the page if we send back | |
2836 | * EAGAIN. | |
2837 | */ | |
247e743c CM |
2838 | if (PageChecked(page)) |
2839 | return -EAGAIN; | |
2840 | ||
2841 | fixup = kzalloc(sizeof(*fixup), GFP_NOFS); | |
2842 | if (!fixup) | |
2843 | return -EAGAIN; | |
f421950f | 2844 | |
f4b1363c JB |
2845 | /* |
2846 | * We are already holding a reference to this inode from | |
2847 | * write_cache_pages. We need to hold it because the space reservation | |
2848 | * takes place outside of the page lock, and we can't trust | |
2849 | * page->mapping outside of the page lock. | |
2850 | */ | |
2851 | ihold(inode); | |
e4f94347 | 2852 | btrfs_page_set_checked(fs_info, page, page_offset(page), PAGE_SIZE); |
09cbfeaf | 2853 | get_page(page); |
a0cac0ec | 2854 | btrfs_init_work(&fixup->work, btrfs_writepage_fixup_worker, NULL, NULL); |
247e743c | 2855 | fixup->page = page; |
36eeaef5 | 2856 | fixup->inode = BTRFS_I(inode); |
0b246afa | 2857 | btrfs_queue_work(fs_info->fixup_workers, &fixup->work); |
25f3c502 CM |
2858 | |
2859 | return -EAGAIN; | |
247e743c CM |
2860 | } |
2861 | ||
d899e052 | 2862 | static int insert_reserved_file_extent(struct btrfs_trans_handle *trans, |
c553f94d | 2863 | struct btrfs_inode *inode, u64 file_pos, |
9729f10a | 2864 | struct btrfs_file_extent_item *stack_fi, |
2766ff61 | 2865 | const bool update_inode_bytes, |
9729f10a | 2866 | u64 qgroup_reserved) |
d899e052 | 2867 | { |
c553f94d | 2868 | struct btrfs_root *root = inode->root; |
2766ff61 | 2869 | const u64 sectorsize = root->fs_info->sectorsize; |
d899e052 YZ |
2870 | struct btrfs_path *path; |
2871 | struct extent_buffer *leaf; | |
2872 | struct btrfs_key ins; | |
203f44c5 QW |
2873 | u64 disk_num_bytes = btrfs_stack_file_extent_disk_num_bytes(stack_fi); |
2874 | u64 disk_bytenr = btrfs_stack_file_extent_disk_bytenr(stack_fi); | |
cb36a9bb | 2875 | u64 offset = btrfs_stack_file_extent_offset(stack_fi); |
203f44c5 QW |
2876 | u64 num_bytes = btrfs_stack_file_extent_num_bytes(stack_fi); |
2877 | u64 ram_bytes = btrfs_stack_file_extent_ram_bytes(stack_fi); | |
5893dfb9 | 2878 | struct btrfs_drop_extents_args drop_args = { 0 }; |
d899e052 YZ |
2879 | int ret; |
2880 | ||
2881 | path = btrfs_alloc_path(); | |
d8926bb3 MF |
2882 | if (!path) |
2883 | return -ENOMEM; | |
d899e052 | 2884 | |
a1ed835e CM |
2885 | /* |
2886 | * we may be replacing one extent in the tree with another. | |
2887 | * The new extent is pinned in the extent map, and we don't want | |
2888 | * to drop it from the cache until it is completely in the btree. | |
2889 | * | |
2890 | * So, tell btrfs_drop_extents to leave this extent in the cache. | |
2891 | * the caller is expected to unpin it and allow it to be merged | |
2892 | * with the others. | |
2893 | */ | |
5893dfb9 FM |
2894 | drop_args.path = path; |
2895 | drop_args.start = file_pos; | |
2896 | drop_args.end = file_pos + num_bytes; | |
2897 | drop_args.replace_extent = true; | |
2898 | drop_args.extent_item_size = sizeof(*stack_fi); | |
2899 | ret = btrfs_drop_extents(trans, root, inode, &drop_args); | |
79787eaa JM |
2900 | if (ret) |
2901 | goto out; | |
d899e052 | 2902 | |
5893dfb9 | 2903 | if (!drop_args.extent_inserted) { |
c553f94d | 2904 | ins.objectid = btrfs_ino(inode); |
1acae57b FDBM |
2905 | ins.offset = file_pos; |
2906 | ins.type = BTRFS_EXTENT_DATA_KEY; | |
2907 | ||
1acae57b | 2908 | ret = btrfs_insert_empty_item(trans, root, path, &ins, |
203f44c5 | 2909 | sizeof(*stack_fi)); |
1acae57b FDBM |
2910 | if (ret) |
2911 | goto out; | |
2912 | } | |
d899e052 | 2913 | leaf = path->nodes[0]; |
203f44c5 QW |
2914 | btrfs_set_stack_file_extent_generation(stack_fi, trans->transid); |
2915 | write_extent_buffer(leaf, stack_fi, | |
2916 | btrfs_item_ptr_offset(leaf, path->slots[0]), | |
2917 | sizeof(struct btrfs_file_extent_item)); | |
b9473439 | 2918 | |
d899e052 | 2919 | btrfs_mark_buffer_dirty(leaf); |
ce195332 | 2920 | btrfs_release_path(path); |
d899e052 | 2921 | |
2766ff61 FM |
2922 | /* |
2923 | * If we dropped an inline extent here, we know the range where it is | |
2924 | * was not marked with the EXTENT_DELALLOC_NEW bit, so we update the | |
1a9fd417 | 2925 | * number of bytes only for that range containing the inline extent. |
2766ff61 FM |
2926 | * The remaining of the range will be processed when clearning the |
2927 | * EXTENT_DELALLOC_BIT bit through the ordered extent completion. | |
2928 | */ | |
2929 | if (file_pos == 0 && !IS_ALIGNED(drop_args.bytes_found, sectorsize)) { | |
2930 | u64 inline_size = round_down(drop_args.bytes_found, sectorsize); | |
2931 | ||
2932 | inline_size = drop_args.bytes_found - inline_size; | |
2933 | btrfs_update_inode_bytes(inode, sectorsize, inline_size); | |
2934 | drop_args.bytes_found -= inline_size; | |
2935 | num_bytes -= sectorsize; | |
2936 | } | |
2937 | ||
2938 | if (update_inode_bytes) | |
2939 | btrfs_update_inode_bytes(inode, num_bytes, drop_args.bytes_found); | |
d899e052 YZ |
2940 | |
2941 | ins.objectid = disk_bytenr; | |
2942 | ins.offset = disk_num_bytes; | |
2943 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
a12b877b | 2944 | |
c553f94d | 2945 | ret = btrfs_inode_set_file_extent_range(inode, file_pos, ram_bytes); |
9ddc959e JB |
2946 | if (ret) |
2947 | goto out; | |
2948 | ||
c553f94d | 2949 | ret = btrfs_alloc_reserved_file_extent(trans, root, btrfs_ino(inode), |
cb36a9bb OS |
2950 | file_pos - offset, |
2951 | qgroup_reserved, &ins); | |
79787eaa | 2952 | out: |
d899e052 | 2953 | btrfs_free_path(path); |
b9473439 | 2954 | |
79787eaa | 2955 | return ret; |
d899e052 YZ |
2956 | } |
2957 | ||
2ff7e61e | 2958 | static void btrfs_release_delalloc_bytes(struct btrfs_fs_info *fs_info, |
e570fd27 MX |
2959 | u64 start, u64 len) |
2960 | { | |
32da5386 | 2961 | struct btrfs_block_group *cache; |
e570fd27 | 2962 | |
0b246afa | 2963 | cache = btrfs_lookup_block_group(fs_info, start); |
e570fd27 MX |
2964 | ASSERT(cache); |
2965 | ||
2966 | spin_lock(&cache->lock); | |
2967 | cache->delalloc_bytes -= len; | |
2968 | spin_unlock(&cache->lock); | |
2969 | ||
2970 | btrfs_put_block_group(cache); | |
2971 | } | |
2972 | ||
203f44c5 | 2973 | static int insert_ordered_extent_file_extent(struct btrfs_trans_handle *trans, |
203f44c5 QW |
2974 | struct btrfs_ordered_extent *oe) |
2975 | { | |
2976 | struct btrfs_file_extent_item stack_fi; | |
2766ff61 | 2977 | bool update_inode_bytes; |
cb36a9bb OS |
2978 | u64 num_bytes = oe->num_bytes; |
2979 | u64 ram_bytes = oe->ram_bytes; | |
203f44c5 QW |
2980 | |
2981 | memset(&stack_fi, 0, sizeof(stack_fi)); | |
2982 | btrfs_set_stack_file_extent_type(&stack_fi, BTRFS_FILE_EXTENT_REG); | |
2983 | btrfs_set_stack_file_extent_disk_bytenr(&stack_fi, oe->disk_bytenr); | |
2984 | btrfs_set_stack_file_extent_disk_num_bytes(&stack_fi, | |
2985 | oe->disk_num_bytes); | |
cb36a9bb | 2986 | btrfs_set_stack_file_extent_offset(&stack_fi, oe->offset); |
c1867eb3 DS |
2987 | if (test_bit(BTRFS_ORDERED_TRUNCATED, &oe->flags)) { |
2988 | num_bytes = oe->truncated_len; | |
2989 | ram_bytes = num_bytes; | |
2990 | } | |
cb36a9bb OS |
2991 | btrfs_set_stack_file_extent_num_bytes(&stack_fi, num_bytes); |
2992 | btrfs_set_stack_file_extent_ram_bytes(&stack_fi, ram_bytes); | |
203f44c5 QW |
2993 | btrfs_set_stack_file_extent_compression(&stack_fi, oe->compress_type); |
2994 | /* Encryption and other encoding is reserved and all 0 */ | |
2995 | ||
2766ff61 FM |
2996 | /* |
2997 | * For delalloc, when completing an ordered extent we update the inode's | |
2998 | * bytes when clearing the range in the inode's io tree, so pass false | |
2999 | * as the argument 'update_inode_bytes' to insert_reserved_file_extent(), | |
3000 | * except if the ordered extent was truncated. | |
3001 | */ | |
3002 | update_inode_bytes = test_bit(BTRFS_ORDERED_DIRECT, &oe->flags) || | |
7c0c7269 | 3003 | test_bit(BTRFS_ORDERED_ENCODED, &oe->flags) || |
2766ff61 FM |
3004 | test_bit(BTRFS_ORDERED_TRUNCATED, &oe->flags); |
3005 | ||
3c38c877 NB |
3006 | return insert_reserved_file_extent(trans, BTRFS_I(oe->inode), |
3007 | oe->file_offset, &stack_fi, | |
2766ff61 | 3008 | update_inode_bytes, oe->qgroup_rsv); |
203f44c5 QW |
3009 | } |
3010 | ||
3011 | /* | |
3012 | * As ordered data IO finishes, this gets called so we can finish | |
d352ac68 CM |
3013 | * an ordered extent if the range of bytes in the file it covers are |
3014 | * fully written. | |
3015 | */ | |
71df088c | 3016 | int btrfs_finish_one_ordered(struct btrfs_ordered_extent *ordered_extent) |
e6dcd2dc | 3017 | { |
72e7e6ed NB |
3018 | struct btrfs_inode *inode = BTRFS_I(ordered_extent->inode); |
3019 | struct btrfs_root *root = inode->root; | |
3020 | struct btrfs_fs_info *fs_info = root->fs_info; | |
0ca1f7ce | 3021 | struct btrfs_trans_handle *trans = NULL; |
72e7e6ed | 3022 | struct extent_io_tree *io_tree = &inode->io_tree; |
2ac55d41 | 3023 | struct extent_state *cached_state = NULL; |
bffe633e | 3024 | u64 start, end; |
261507a0 | 3025 | int compress_type = 0; |
77cef2ec | 3026 | int ret = 0; |
bffe633e | 3027 | u64 logical_len = ordered_extent->num_bytes; |
8d510121 | 3028 | bool freespace_inode; |
77cef2ec | 3029 | bool truncated = false; |
49940bdd | 3030 | bool clear_reserved_extent = true; |
2766ff61 | 3031 | unsigned int clear_bits = EXTENT_DEFRAG; |
a7e3b975 | 3032 | |
bffe633e OS |
3033 | start = ordered_extent->file_offset; |
3034 | end = start + ordered_extent->num_bytes - 1; | |
3035 | ||
a7e3b975 FM |
3036 | if (!test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) && |
3037 | !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags) && | |
7c0c7269 OS |
3038 | !test_bit(BTRFS_ORDERED_DIRECT, &ordered_extent->flags) && |
3039 | !test_bit(BTRFS_ORDERED_ENCODED, &ordered_extent->flags)) | |
2766ff61 | 3040 | clear_bits |= EXTENT_DELALLOC_NEW; |
e6dcd2dc | 3041 | |
72e7e6ed | 3042 | freespace_inode = btrfs_is_free_space_inode(inode); |
5f4403e1 IA |
3043 | if (!freespace_inode) |
3044 | btrfs_lockdep_acquire(fs_info, btrfs_ordered_extent); | |
0cb59c99 | 3045 | |
5fd02043 JB |
3046 | if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) { |
3047 | ret = -EIO; | |
3048 | goto out; | |
3049 | } | |
3050 | ||
71df088c | 3051 | if (btrfs_is_zoned(fs_info)) |
be1a1d7a NA |
3052 | btrfs_zone_finish_endio(fs_info, ordered_extent->disk_bytenr, |
3053 | ordered_extent->disk_num_bytes); | |
d8e3fb10 | 3054 | |
77cef2ec JB |
3055 | if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) { |
3056 | truncated = true; | |
3057 | logical_len = ordered_extent->truncated_len; | |
3058 | /* Truncated the entire extent, don't bother adding */ | |
3059 | if (!logical_len) | |
3060 | goto out; | |
3061 | } | |
3062 | ||
c2167754 | 3063 | if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) { |
79787eaa | 3064 | BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */ |
94ed938a | 3065 | |
72e7e6ed | 3066 | btrfs_inode_safe_disk_i_size_write(inode, 0); |
8d510121 NB |
3067 | if (freespace_inode) |
3068 | trans = btrfs_join_transaction_spacecache(root); | |
6c760c07 JB |
3069 | else |
3070 | trans = btrfs_join_transaction(root); | |
3071 | if (IS_ERR(trans)) { | |
3072 | ret = PTR_ERR(trans); | |
3073 | trans = NULL; | |
3074 | goto out; | |
c2167754 | 3075 | } |
72e7e6ed | 3076 | trans->block_rsv = &inode->block_rsv; |
729f7961 | 3077 | ret = btrfs_update_inode_fallback(trans, root, inode); |
6c760c07 | 3078 | if (ret) /* -ENOMEM or corruption */ |
66642832 | 3079 | btrfs_abort_transaction(trans, ret); |
c2167754 YZ |
3080 | goto out; |
3081 | } | |
e6dcd2dc | 3082 | |
2766ff61 | 3083 | clear_bits |= EXTENT_LOCKED; |
570eb97b | 3084 | lock_extent(io_tree, start, end, &cached_state); |
e6dcd2dc | 3085 | |
8d510121 NB |
3086 | if (freespace_inode) |
3087 | trans = btrfs_join_transaction_spacecache(root); | |
0cb59c99 | 3088 | else |
7a7eaa40 | 3089 | trans = btrfs_join_transaction(root); |
79787eaa JM |
3090 | if (IS_ERR(trans)) { |
3091 | ret = PTR_ERR(trans); | |
3092 | trans = NULL; | |
a7e3b975 | 3093 | goto out; |
79787eaa | 3094 | } |
a79b7d4b | 3095 | |
72e7e6ed | 3096 | trans->block_rsv = &inode->block_rsv; |
c2167754 | 3097 | |
c8b97818 | 3098 | if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags)) |
261507a0 | 3099 | compress_type = ordered_extent->compress_type; |
d899e052 | 3100 | if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) { |
261507a0 | 3101 | BUG_ON(compress_type); |
72e7e6ed | 3102 | ret = btrfs_mark_extent_written(trans, inode, |
d899e052 YZ |
3103 | ordered_extent->file_offset, |
3104 | ordered_extent->file_offset + | |
77cef2ec | 3105 | logical_len); |
343d8a30 NA |
3106 | btrfs_zoned_release_data_reloc_bg(fs_info, ordered_extent->disk_bytenr, |
3107 | ordered_extent->disk_num_bytes); | |
d899e052 | 3108 | } else { |
0b246afa | 3109 | BUG_ON(root == fs_info->tree_root); |
3c38c877 | 3110 | ret = insert_ordered_extent_file_extent(trans, ordered_extent); |
49940bdd JB |
3111 | if (!ret) { |
3112 | clear_reserved_extent = false; | |
2ff7e61e | 3113 | btrfs_release_delalloc_bytes(fs_info, |
bffe633e OS |
3114 | ordered_extent->disk_bytenr, |
3115 | ordered_extent->disk_num_bytes); | |
49940bdd | 3116 | } |
d899e052 | 3117 | } |
72e7e6ed | 3118 | unpin_extent_cache(&inode->extent_tree, ordered_extent->file_offset, |
bffe633e | 3119 | ordered_extent->num_bytes, trans->transid); |
79787eaa | 3120 | if (ret < 0) { |
66642832 | 3121 | btrfs_abort_transaction(trans, ret); |
a7e3b975 | 3122 | goto out; |
79787eaa | 3123 | } |
2ac55d41 | 3124 | |
510f85ed | 3125 | ret = add_pending_csums(trans, &ordered_extent->list); |
ac01f26a NB |
3126 | if (ret) { |
3127 | btrfs_abort_transaction(trans, ret); | |
3128 | goto out; | |
3129 | } | |
e6dcd2dc | 3130 | |
2766ff61 FM |
3131 | /* |
3132 | * If this is a new delalloc range, clear its new delalloc flag to | |
3133 | * update the inode's number of bytes. This needs to be done first | |
3134 | * before updating the inode item. | |
3135 | */ | |
3136 | if ((clear_bits & EXTENT_DELALLOC_NEW) && | |
3137 | !test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) | |
72e7e6ed | 3138 | clear_extent_bit(&inode->io_tree, start, end, |
2766ff61 | 3139 | EXTENT_DELALLOC_NEW | EXTENT_ADD_INODE_BYTES, |
bd015294 | 3140 | &cached_state); |
2766ff61 | 3141 | |
72e7e6ed | 3142 | btrfs_inode_safe_disk_i_size_write(inode, 0); |
729f7961 | 3143 | ret = btrfs_update_inode_fallback(trans, root, inode); |
6c760c07 | 3144 | if (ret) { /* -ENOMEM or corruption */ |
66642832 | 3145 | btrfs_abort_transaction(trans, ret); |
a7e3b975 | 3146 | goto out; |
1ef30be1 JB |
3147 | } |
3148 | ret = 0; | |
c2167754 | 3149 | out: |
bd015294 | 3150 | clear_extent_bit(&inode->io_tree, start, end, clear_bits, |
313facc5 | 3151 | &cached_state); |
a7e3b975 | 3152 | |
a698d075 | 3153 | if (trans) |
3a45bb20 | 3154 | btrfs_end_transaction(trans); |
0cb59c99 | 3155 | |
77cef2ec | 3156 | if (ret || truncated) { |
bffe633e | 3157 | u64 unwritten_start = start; |
77cef2ec | 3158 | |
d61bec08 JB |
3159 | /* |
3160 | * If we failed to finish this ordered extent for any reason we | |
3161 | * need to make sure BTRFS_ORDERED_IOERR is set on the ordered | |
3162 | * extent, and mark the inode with the error if it wasn't | |
3163 | * already set. Any error during writeback would have already | |
3164 | * set the mapping error, so we need to set it if we're the ones | |
3165 | * marking this ordered extent as failed. | |
3166 | */ | |
3167 | if (ret && !test_and_set_bit(BTRFS_ORDERED_IOERR, | |
3168 | &ordered_extent->flags)) | |
3169 | mapping_set_error(ordered_extent->inode->i_mapping, -EIO); | |
3170 | ||
77cef2ec | 3171 | if (truncated) |
bffe633e OS |
3172 | unwritten_start += logical_len; |
3173 | clear_extent_uptodate(io_tree, unwritten_start, end, NULL); | |
77cef2ec | 3174 | |
4c0c8cfc FM |
3175 | /* Drop extent maps for the part of the extent we didn't write. */ |
3176 | btrfs_drop_extent_map_range(inode, unwritten_start, end, false); | |
5fd02043 | 3177 | |
0bec9ef5 JB |
3178 | /* |
3179 | * If the ordered extent had an IOERR or something else went | |
3180 | * wrong we need to return the space for this ordered extent | |
77cef2ec JB |
3181 | * back to the allocator. We only free the extent in the |
3182 | * truncated case if we didn't write out the extent at all. | |
49940bdd JB |
3183 | * |
3184 | * If we made it past insert_reserved_file_extent before we | |
3185 | * errored out then we don't need to do this as the accounting | |
3186 | * has already been done. | |
0bec9ef5 | 3187 | */ |
77cef2ec | 3188 | if ((ret || !logical_len) && |
49940bdd | 3189 | clear_reserved_extent && |
77cef2ec | 3190 | !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) && |
4eaaec24 NB |
3191 | !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) { |
3192 | /* | |
3193 | * Discard the range before returning it back to the | |
3194 | * free space pool | |
3195 | */ | |
46b27f50 | 3196 | if (ret && btrfs_test_opt(fs_info, DISCARD_SYNC)) |
4eaaec24 | 3197 | btrfs_discard_extent(fs_info, |
bffe633e OS |
3198 | ordered_extent->disk_bytenr, |
3199 | ordered_extent->disk_num_bytes, | |
3200 | NULL); | |
2ff7e61e | 3201 | btrfs_free_reserved_extent(fs_info, |
bffe633e OS |
3202 | ordered_extent->disk_bytenr, |
3203 | ordered_extent->disk_num_bytes, 1); | |
e28b0211 BB |
3204 | /* |
3205 | * Actually free the qgroup rsv which was released when | |
3206 | * the ordered extent was created. | |
3207 | */ | |
3208 | btrfs_qgroup_free_refroot(fs_info, inode->root->root_key.objectid, | |
3209 | ordered_extent->qgroup_rsv, | |
3210 | BTRFS_QGROUP_RSV_DATA); | |
4eaaec24 | 3211 | } |
0bec9ef5 JB |
3212 | } |
3213 | ||
5fd02043 | 3214 | /* |
8bad3c02 LB |
3215 | * This needs to be done to make sure anybody waiting knows we are done |
3216 | * updating everything for this ordered extent. | |
5fd02043 | 3217 | */ |
72e7e6ed | 3218 | btrfs_remove_ordered_extent(inode, ordered_extent); |
5fd02043 | 3219 | |
e6dcd2dc CM |
3220 | /* once for us */ |
3221 | btrfs_put_ordered_extent(ordered_extent); | |
3222 | /* once for the tree */ | |
3223 | btrfs_put_ordered_extent(ordered_extent); | |
3224 | ||
5fd02043 JB |
3225 | return ret; |
3226 | } | |
3227 | ||
71df088c CH |
3228 | int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered) |
3229 | { | |
3230 | if (btrfs_is_zoned(btrfs_sb(ordered->inode->i_sb)) && | |
3231 | !test_bit(BTRFS_ORDERED_IOERR, &ordered->flags)) | |
3232 | btrfs_finish_ordered_zoned(ordered); | |
3233 | return btrfs_finish_one_ordered(ordered); | |
3234 | } | |
3235 | ||
ae643a74 QW |
3236 | /* |
3237 | * Verify the checksum for a single sector without any extra action that depend | |
3238 | * on the type of I/O. | |
3239 | */ | |
3240 | int btrfs_check_sector_csum(struct btrfs_fs_info *fs_info, struct page *page, | |
3241 | u32 pgoff, u8 *csum, const u8 * const csum_expected) | |
3242 | { | |
3243 | SHASH_DESC_ON_STACK(shash, fs_info->csum_shash); | |
3244 | char *kaddr; | |
3245 | ||
3246 | ASSERT(pgoff + fs_info->sectorsize <= PAGE_SIZE); | |
3247 | ||
3248 | shash->tfm = fs_info->csum_shash; | |
3249 | ||
3250 | kaddr = kmap_local_page(page) + pgoff; | |
3251 | crypto_shash_digest(shash, kaddr, fs_info->sectorsize, csum); | |
3252 | kunmap_local(kaddr); | |
3253 | ||
3254 | if (memcmp(csum, csum_expected, fs_info->csum_size)) | |
3255 | return -EIO; | |
3256 | return 0; | |
211f90e6 CM |
3257 | } |
3258 | ||
265d4ac0 | 3259 | /* |
e5219044 CH |
3260 | * Verify the checksum of a single data sector. |
3261 | * | |
3262 | * @bbio: btrfs_io_bio which contains the csum | |
3263 | * @dev: device the sector is on | |
7ffd27e3 | 3264 | * @bio_offset: offset to the beginning of the bio (in bytes) |
e5219044 | 3265 | * @bv: bio_vec to check |
265d4ac0 | 3266 | * |
e5219044 CH |
3267 | * Check if the checksum on a data block is valid. When a checksum mismatch is |
3268 | * detected, report the error and fill the corrupted range with zero. | |
ae643a74 | 3269 | * |
e5219044 | 3270 | * Return %true if the sector is ok or had no checksum to start with, else %false. |
265d4ac0 | 3271 | */ |
e5219044 CH |
3272 | bool btrfs_data_csum_ok(struct btrfs_bio *bbio, struct btrfs_device *dev, |
3273 | u32 bio_offset, struct bio_vec *bv) | |
dc380aea | 3274 | { |
e5219044 | 3275 | struct btrfs_inode *inode = bbio->inode; |
621af94a | 3276 | struct btrfs_fs_info *fs_info = inode->root->fs_info; |
e5219044 CH |
3277 | u64 file_offset = bbio->file_offset + bio_offset; |
3278 | u64 end = file_offset + bv->bv_len - 1; | |
d5178578 JT |
3279 | u8 *csum_expected; |
3280 | u8 csum[BTRFS_CSUM_SIZE]; | |
dc380aea | 3281 | |
3d49d0d3 | 3282 | ASSERT(bv->bv_len == fs_info->sectorsize); |
265d4ac0 | 3283 | |
e5219044 CH |
3284 | if (!bbio->csum) |
3285 | return true; | |
d5178578 | 3286 | |
e5219044 CH |
3287 | if (btrfs_is_data_reloc_root(inode->root) && |
3288 | test_range_bit(&inode->io_tree, file_offset, end, EXTENT_NODATASUM, | |
3289 | 1, NULL)) { | |
3290 | /* Skip the range without csum for data reloc inode */ | |
3291 | clear_extent_bits(&inode->io_tree, file_offset, end, | |
3292 | EXTENT_NODATASUM); | |
3293 | return true; | |
3294 | } | |
3295 | ||
fa13661c JT |
3296 | csum_expected = bbio->csum + (bio_offset >> fs_info->sectorsize_bits) * |
3297 | fs_info->csum_size; | |
3d49d0d3 CH |
3298 | if (btrfs_check_sector_csum(fs_info, bv->bv_page, bv->bv_offset, csum, |
3299 | csum_expected)) | |
dc380aea | 3300 | goto zeroit; |
e5219044 | 3301 | return true; |
ae643a74 | 3302 | |
dc380aea | 3303 | zeroit: |
3d49d0d3 CH |
3304 | btrfs_print_data_csum_error(inode, file_offset, csum, csum_expected, |
3305 | bbio->mirror_num); | |
3306 | if (dev) | |
3307 | btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS); | |
3308 | memzero_bvec(bv); | |
3309 | return false; | |
07157aac | 3310 | } |
b888db2b | 3311 | |
c1c3fac2 NB |
3312 | /* |
3313 | * btrfs_add_delayed_iput - perform a delayed iput on @inode | |
3314 | * | |
3315 | * @inode: The inode we want to perform iput on | |
3316 | * | |
3317 | * This function uses the generic vfs_inode::i_count to track whether we should | |
3318 | * just decrement it (in case it's > 1) or if this is the last iput then link | |
3319 | * the inode to the delayed iput machinery. Delayed iputs are processed at | |
3320 | * transaction commit time/superblock commit/cleaner kthread. | |
3321 | */ | |
e55cf7ca | 3322 | void btrfs_add_delayed_iput(struct btrfs_inode *inode) |
24bbcf04 | 3323 | { |
e55cf7ca | 3324 | struct btrfs_fs_info *fs_info = inode->root->fs_info; |
866e98a4 | 3325 | unsigned long flags; |
24bbcf04 | 3326 | |
e55cf7ca | 3327 | if (atomic_add_unless(&inode->vfs_inode.i_count, -1, 1)) |
24bbcf04 YZ |
3328 | return; |
3329 | ||
034f784d | 3330 | atomic_inc(&fs_info->nr_delayed_iputs); |
866e98a4 FM |
3331 | /* |
3332 | * Need to be irq safe here because we can be called from either an irq | |
3333 | * context (see bio.c and btrfs_put_ordered_extent()) or a non-irq | |
3334 | * context. | |
3335 | */ | |
3336 | spin_lock_irqsave(&fs_info->delayed_iput_lock, flags); | |
e55cf7ca DS |
3337 | ASSERT(list_empty(&inode->delayed_iput)); |
3338 | list_add_tail(&inode->delayed_iput, &fs_info->delayed_iputs); | |
866e98a4 | 3339 | spin_unlock_irqrestore(&fs_info->delayed_iput_lock, flags); |
fd340d0f JB |
3340 | if (!test_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags)) |
3341 | wake_up_process(fs_info->cleaner_kthread); | |
24bbcf04 YZ |
3342 | } |
3343 | ||
63611e73 JB |
3344 | static void run_delayed_iput_locked(struct btrfs_fs_info *fs_info, |
3345 | struct btrfs_inode *inode) | |
3346 | { | |
3347 | list_del_init(&inode->delayed_iput); | |
866e98a4 | 3348 | spin_unlock_irq(&fs_info->delayed_iput_lock); |
63611e73 JB |
3349 | iput(&inode->vfs_inode); |
3350 | if (atomic_dec_and_test(&fs_info->nr_delayed_iputs)) | |
3351 | wake_up(&fs_info->delayed_iputs_wait); | |
866e98a4 | 3352 | spin_lock_irq(&fs_info->delayed_iput_lock); |
63611e73 JB |
3353 | } |
3354 | ||
3355 | static void btrfs_run_delayed_iput(struct btrfs_fs_info *fs_info, | |
3356 | struct btrfs_inode *inode) | |
3357 | { | |
3358 | if (!list_empty(&inode->delayed_iput)) { | |
866e98a4 | 3359 | spin_lock_irq(&fs_info->delayed_iput_lock); |
63611e73 JB |
3360 | if (!list_empty(&inode->delayed_iput)) |
3361 | run_delayed_iput_locked(fs_info, inode); | |
866e98a4 | 3362 | spin_unlock_irq(&fs_info->delayed_iput_lock); |
63611e73 JB |
3363 | } |
3364 | } | |
3365 | ||
2ff7e61e | 3366 | void btrfs_run_delayed_iputs(struct btrfs_fs_info *fs_info) |
24bbcf04 | 3367 | { |
866e98a4 FM |
3368 | /* |
3369 | * btrfs_put_ordered_extent() can run in irq context (see bio.c), which | |
3370 | * calls btrfs_add_delayed_iput() and that needs to lock | |
3371 | * fs_info->delayed_iput_lock. So we need to disable irqs here to | |
3372 | * prevent a deadlock. | |
3373 | */ | |
3374 | spin_lock_irq(&fs_info->delayed_iput_lock); | |
8089fe62 DS |
3375 | while (!list_empty(&fs_info->delayed_iputs)) { |
3376 | struct btrfs_inode *inode; | |
3377 | ||
3378 | inode = list_first_entry(&fs_info->delayed_iputs, | |
3379 | struct btrfs_inode, delayed_iput); | |
63611e73 | 3380 | run_delayed_iput_locked(fs_info, inode); |
866e98a4 FM |
3381 | if (need_resched()) { |
3382 | spin_unlock_irq(&fs_info->delayed_iput_lock); | |
3383 | cond_resched(); | |
3384 | spin_lock_irq(&fs_info->delayed_iput_lock); | |
3385 | } | |
24bbcf04 | 3386 | } |
866e98a4 | 3387 | spin_unlock_irq(&fs_info->delayed_iput_lock); |
24bbcf04 YZ |
3388 | } |
3389 | ||
e43eec81 | 3390 | /* |
2639631d NB |
3391 | * Wait for flushing all delayed iputs |
3392 | * | |
3393 | * @fs_info: the filesystem | |
034f784d JB |
3394 | * |
3395 | * This will wait on any delayed iputs that are currently running with KILLABLE | |
3396 | * set. Once they are all done running we will return, unless we are killed in | |
3397 | * which case we return EINTR. This helps in user operations like fallocate etc | |
3398 | * that might get blocked on the iputs. | |
2639631d NB |
3399 | * |
3400 | * Return EINTR if we were killed, 0 if nothing's pending | |
034f784d JB |
3401 | */ |
3402 | int btrfs_wait_on_delayed_iputs(struct btrfs_fs_info *fs_info) | |
3403 | { | |
3404 | int ret = wait_event_killable(fs_info->delayed_iputs_wait, | |
3405 | atomic_read(&fs_info->nr_delayed_iputs) == 0); | |
3406 | if (ret) | |
3407 | return -EINTR; | |
3408 | return 0; | |
3409 | } | |
3410 | ||
7b128766 | 3411 | /* |
f7e9e8fc OS |
3412 | * This creates an orphan entry for the given inode in case something goes wrong |
3413 | * in the middle of an unlink. | |
7b128766 | 3414 | */ |
73f2e545 | 3415 | int btrfs_orphan_add(struct btrfs_trans_handle *trans, |
27919067 | 3416 | struct btrfs_inode *inode) |
7b128766 | 3417 | { |
d68fc57b | 3418 | int ret; |
7b128766 | 3419 | |
27919067 OS |
3420 | ret = btrfs_insert_orphan_item(trans, inode->root, btrfs_ino(inode)); |
3421 | if (ret && ret != -EEXIST) { | |
3422 | btrfs_abort_transaction(trans, ret); | |
3423 | return ret; | |
d68fc57b YZ |
3424 | } |
3425 | ||
d68fc57b | 3426 | return 0; |
7b128766 JB |
3427 | } |
3428 | ||
3429 | /* | |
f7e9e8fc OS |
3430 | * We have done the delete so we can go ahead and remove the orphan item for |
3431 | * this particular inode. | |
7b128766 | 3432 | */ |
48a3b636 | 3433 | static int btrfs_orphan_del(struct btrfs_trans_handle *trans, |
3d6ae7bb | 3434 | struct btrfs_inode *inode) |
7b128766 | 3435 | { |
27919067 | 3436 | return btrfs_del_orphan_item(trans, inode->root, btrfs_ino(inode)); |
7b128766 JB |
3437 | } |
3438 | ||
3439 | /* | |
3440 | * this cleans up any orphans that may be left on the list from the last use | |
3441 | * of this root. | |
3442 | */ | |
66b4ffd1 | 3443 | int btrfs_orphan_cleanup(struct btrfs_root *root) |
7b128766 | 3444 | { |
0b246afa | 3445 | struct btrfs_fs_info *fs_info = root->fs_info; |
7b128766 JB |
3446 | struct btrfs_path *path; |
3447 | struct extent_buffer *leaf; | |
7b128766 JB |
3448 | struct btrfs_key key, found_key; |
3449 | struct btrfs_trans_handle *trans; | |
3450 | struct inode *inode; | |
8f6d7f4f | 3451 | u64 last_objectid = 0; |
f7e9e8fc | 3452 | int ret = 0, nr_unlink = 0; |
7b128766 | 3453 | |
54230013 | 3454 | if (test_and_set_bit(BTRFS_ROOT_ORPHAN_CLEANUP, &root->state)) |
66b4ffd1 | 3455 | return 0; |
c71bf099 YZ |
3456 | |
3457 | path = btrfs_alloc_path(); | |
66b4ffd1 JB |
3458 | if (!path) { |
3459 | ret = -ENOMEM; | |
3460 | goto out; | |
3461 | } | |
e4058b54 | 3462 | path->reada = READA_BACK; |
7b128766 JB |
3463 | |
3464 | key.objectid = BTRFS_ORPHAN_OBJECTID; | |
962a298f | 3465 | key.type = BTRFS_ORPHAN_ITEM_KEY; |
7b128766 JB |
3466 | key.offset = (u64)-1; |
3467 | ||
7b128766 JB |
3468 | while (1) { |
3469 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
66b4ffd1 JB |
3470 | if (ret < 0) |
3471 | goto out; | |
7b128766 JB |
3472 | |
3473 | /* | |
3474 | * if ret == 0 means we found what we were searching for, which | |
25985edc | 3475 | * is weird, but possible, so only screw with path if we didn't |
7b128766 JB |
3476 | * find the key and see if we have stuff that matches |
3477 | */ | |
3478 | if (ret > 0) { | |
66b4ffd1 | 3479 | ret = 0; |
7b128766 JB |
3480 | if (path->slots[0] == 0) |
3481 | break; | |
3482 | path->slots[0]--; | |
3483 | } | |
3484 | ||
3485 | /* pull out the item */ | |
3486 | leaf = path->nodes[0]; | |
7b128766 JB |
3487 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
3488 | ||
3489 | /* make sure the item matches what we want */ | |
3490 | if (found_key.objectid != BTRFS_ORPHAN_OBJECTID) | |
3491 | break; | |
962a298f | 3492 | if (found_key.type != BTRFS_ORPHAN_ITEM_KEY) |
7b128766 JB |
3493 | break; |
3494 | ||
3495 | /* release the path since we're done with it */ | |
b3b4aa74 | 3496 | btrfs_release_path(path); |
7b128766 JB |
3497 | |
3498 | /* | |
3499 | * this is where we are basically btrfs_lookup, without the | |
3500 | * crossing root thing. we store the inode number in the | |
3501 | * offset of the orphan item. | |
3502 | */ | |
8f6d7f4f JB |
3503 | |
3504 | if (found_key.offset == last_objectid) { | |
a7f8de50 FM |
3505 | /* |
3506 | * We found the same inode as before. This means we were | |
3507 | * not able to remove its items via eviction triggered | |
3508 | * by an iput(). A transaction abort may have happened, | |
3509 | * due to -ENOSPC for example, so try to grab the error | |
3510 | * that lead to a transaction abort, if any. | |
3511 | */ | |
0b246afa JM |
3512 | btrfs_err(fs_info, |
3513 | "Error removing orphan entry, stopping orphan cleanup"); | |
a7f8de50 | 3514 | ret = BTRFS_FS_ERROR(fs_info) ?: -EINVAL; |
8f6d7f4f JB |
3515 | goto out; |
3516 | } | |
3517 | ||
3518 | last_objectid = found_key.offset; | |
3519 | ||
5d4f98a2 YZ |
3520 | found_key.objectid = found_key.offset; |
3521 | found_key.type = BTRFS_INODE_ITEM_KEY; | |
3522 | found_key.offset = 0; | |
0202e83f | 3523 | inode = btrfs_iget(fs_info->sb, last_objectid, root); |
cbaee87f FM |
3524 | if (IS_ERR(inode)) { |
3525 | ret = PTR_ERR(inode); | |
3526 | inode = NULL; | |
3527 | if (ret != -ENOENT) | |
3528 | goto out; | |
3529 | } | |
7b128766 | 3530 | |
cbaee87f | 3531 | if (!inode && root == fs_info->tree_root) { |
f8e9e0b0 | 3532 | struct btrfs_root *dead_root; |
f8e9e0b0 AJ |
3533 | int is_dead_root = 0; |
3534 | ||
3535 | /* | |
0c0218e9 | 3536 | * This is an orphan in the tree root. Currently these |
f8e9e0b0 | 3537 | * could come from 2 sources: |
0c0218e9 | 3538 | * a) a root (snapshot/subvolume) deletion in progress |
f8e9e0b0 | 3539 | * b) a free space cache inode |
0c0218e9 FM |
3540 | * We need to distinguish those two, as the orphan item |
3541 | * for a root must not get deleted before the deletion | |
3542 | * of the snapshot/subvolume's tree completes. | |
3543 | * | |
3544 | * btrfs_find_orphan_roots() ran before us, which has | |
3545 | * found all deleted roots and loaded them into | |
fc7cbcd4 | 3546 | * fs_info->fs_roots_radix. So here we can find if an |
0c0218e9 | 3547 | * orphan item corresponds to a deleted root by looking |
fc7cbcd4 | 3548 | * up the root from that radix tree. |
f8e9e0b0 | 3549 | */ |
a619b3c7 | 3550 | |
fc7cbcd4 DS |
3551 | spin_lock(&fs_info->fs_roots_radix_lock); |
3552 | dead_root = radix_tree_lookup(&fs_info->fs_roots_radix, | |
3553 | (unsigned long)found_key.objectid); | |
a619b3c7 RK |
3554 | if (dead_root && btrfs_root_refs(&dead_root->root_item) == 0) |
3555 | is_dead_root = 1; | |
fc7cbcd4 | 3556 | spin_unlock(&fs_info->fs_roots_radix_lock); |
a619b3c7 | 3557 | |
f8e9e0b0 AJ |
3558 | if (is_dead_root) { |
3559 | /* prevent this orphan from being found again */ | |
3560 | key.offset = found_key.objectid - 1; | |
3561 | continue; | |
3562 | } | |
f7e9e8fc | 3563 | |
f8e9e0b0 | 3564 | } |
f7e9e8fc | 3565 | |
7b128766 | 3566 | /* |
f7e9e8fc | 3567 | * If we have an inode with links, there are a couple of |
70524253 BB |
3568 | * possibilities: |
3569 | * | |
3570 | * 1. We were halfway through creating fsverity metadata for the | |
3571 | * file. In that case, the orphan item represents incomplete | |
3572 | * fsverity metadata which must be cleaned up with | |
3573 | * btrfs_drop_verity_items and deleting the orphan item. | |
3574 | ||
3575 | * 2. Old kernels (before v3.12) used to create an | |
f7e9e8fc OS |
3576 | * orphan item for truncate indicating that there were possibly |
3577 | * extent items past i_size that needed to be deleted. In v3.12, | |
3578 | * truncate was changed to update i_size in sync with the extent | |
3579 | * items, but the (useless) orphan item was still created. Since | |
3580 | * v4.18, we don't create the orphan item for truncate at all. | |
3581 | * | |
3582 | * So, this item could mean that we need to do a truncate, but | |
3583 | * only if this filesystem was last used on a pre-v3.12 kernel | |
3584 | * and was not cleanly unmounted. The odds of that are quite | |
3585 | * slim, and it's a pain to do the truncate now, so just delete | |
3586 | * the orphan item. | |
3587 | * | |
3588 | * It's also possible that this orphan item was supposed to be | |
3589 | * deleted but wasn't. The inode number may have been reused, | |
3590 | * but either way, we can delete the orphan item. | |
7b128766 | 3591 | */ |
cbaee87f FM |
3592 | if (!inode || inode->i_nlink) { |
3593 | if (inode) { | |
70524253 | 3594 | ret = btrfs_drop_verity_items(BTRFS_I(inode)); |
f7e9e8fc | 3595 | iput(inode); |
b777d279 | 3596 | inode = NULL; |
70524253 BB |
3597 | if (ret) |
3598 | goto out; | |
3599 | } | |
a8c9e576 | 3600 | trans = btrfs_start_transaction(root, 1); |
66b4ffd1 JB |
3601 | if (IS_ERR(trans)) { |
3602 | ret = PTR_ERR(trans); | |
3603 | goto out; | |
3604 | } | |
0b246afa JM |
3605 | btrfs_debug(fs_info, "auto deleting %Lu", |
3606 | found_key.objectid); | |
a8c9e576 JB |
3607 | ret = btrfs_del_orphan_item(trans, root, |
3608 | found_key.objectid); | |
3a45bb20 | 3609 | btrfs_end_transaction(trans); |
cbaee87f | 3610 | if (ret) |
4ef31a45 | 3611 | goto out; |
7b128766 JB |
3612 | continue; |
3613 | } | |
3614 | ||
f7e9e8fc | 3615 | nr_unlink++; |
7b128766 JB |
3616 | |
3617 | /* this will do delete_inode and everything for us */ | |
3618 | iput(inode); | |
3619 | } | |
3254c876 MX |
3620 | /* release the path since we're done with it */ |
3621 | btrfs_release_path(path); | |
3622 | ||
a575ceeb | 3623 | if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state)) { |
7a7eaa40 | 3624 | trans = btrfs_join_transaction(root); |
66b4ffd1 | 3625 | if (!IS_ERR(trans)) |
3a45bb20 | 3626 | btrfs_end_transaction(trans); |
d68fc57b | 3627 | } |
7b128766 JB |
3628 | |
3629 | if (nr_unlink) | |
0b246afa | 3630 | btrfs_debug(fs_info, "unlinked %d orphans", nr_unlink); |
66b4ffd1 JB |
3631 | |
3632 | out: | |
3633 | if (ret) | |
0b246afa | 3634 | btrfs_err(fs_info, "could not do orphan cleanup %d", ret); |
66b4ffd1 JB |
3635 | btrfs_free_path(path); |
3636 | return ret; | |
7b128766 JB |
3637 | } |
3638 | ||
46a53cca CM |
3639 | /* |
3640 | * very simple check to peek ahead in the leaf looking for xattrs. If we | |
3641 | * don't find any xattrs, we know there can't be any acls. | |
3642 | * | |
3643 | * slot is the slot the inode is in, objectid is the objectid of the inode | |
3644 | */ | |
3645 | static noinline int acls_after_inode_item(struct extent_buffer *leaf, | |
63541927 FDBM |
3646 | int slot, u64 objectid, |
3647 | int *first_xattr_slot) | |
46a53cca CM |
3648 | { |
3649 | u32 nritems = btrfs_header_nritems(leaf); | |
3650 | struct btrfs_key found_key; | |
f23b5a59 JB |
3651 | static u64 xattr_access = 0; |
3652 | static u64 xattr_default = 0; | |
46a53cca CM |
3653 | int scanned = 0; |
3654 | ||
f23b5a59 | 3655 | if (!xattr_access) { |
97d79299 AG |
3656 | xattr_access = btrfs_name_hash(XATTR_NAME_POSIX_ACL_ACCESS, |
3657 | strlen(XATTR_NAME_POSIX_ACL_ACCESS)); | |
3658 | xattr_default = btrfs_name_hash(XATTR_NAME_POSIX_ACL_DEFAULT, | |
3659 | strlen(XATTR_NAME_POSIX_ACL_DEFAULT)); | |
f23b5a59 JB |
3660 | } |
3661 | ||
46a53cca | 3662 | slot++; |
63541927 | 3663 | *first_xattr_slot = -1; |
46a53cca CM |
3664 | while (slot < nritems) { |
3665 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
3666 | ||
3667 | /* we found a different objectid, there must not be acls */ | |
3668 | if (found_key.objectid != objectid) | |
3669 | return 0; | |
3670 | ||
3671 | /* we found an xattr, assume we've got an acl */ | |
f23b5a59 | 3672 | if (found_key.type == BTRFS_XATTR_ITEM_KEY) { |
63541927 FDBM |
3673 | if (*first_xattr_slot == -1) |
3674 | *first_xattr_slot = slot; | |
f23b5a59 JB |
3675 | if (found_key.offset == xattr_access || |
3676 | found_key.offset == xattr_default) | |
3677 | return 1; | |
3678 | } | |
46a53cca CM |
3679 | |
3680 | /* | |
3681 | * we found a key greater than an xattr key, there can't | |
3682 | * be any acls later on | |
3683 | */ | |
3684 | if (found_key.type > BTRFS_XATTR_ITEM_KEY) | |
3685 | return 0; | |
3686 | ||
3687 | slot++; | |
3688 | scanned++; | |
3689 | ||
3690 | /* | |
3691 | * it goes inode, inode backrefs, xattrs, extents, | |
3692 | * so if there are a ton of hard links to an inode there can | |
3693 | * be a lot of backrefs. Don't waste time searching too hard, | |
3694 | * this is just an optimization | |
3695 | */ | |
3696 | if (scanned >= 8) | |
3697 | break; | |
3698 | } | |
3699 | /* we hit the end of the leaf before we found an xattr or | |
3700 | * something larger than an xattr. We have to assume the inode | |
3701 | * has acls | |
3702 | */ | |
63541927 FDBM |
3703 | if (*first_xattr_slot == -1) |
3704 | *first_xattr_slot = slot; | |
46a53cca CM |
3705 | return 1; |
3706 | } | |
3707 | ||
d352ac68 CM |
3708 | /* |
3709 | * read an inode from the btree into the in-memory inode | |
3710 | */ | |
4222ea71 FM |
3711 | static int btrfs_read_locked_inode(struct inode *inode, |
3712 | struct btrfs_path *in_path) | |
39279cc3 | 3713 | { |
0b246afa | 3714 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
4222ea71 | 3715 | struct btrfs_path *path = in_path; |
5f39d397 | 3716 | struct extent_buffer *leaf; |
39279cc3 CM |
3717 | struct btrfs_inode_item *inode_item; |
3718 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
3719 | struct btrfs_key location; | |
67de1176 | 3720 | unsigned long ptr; |
46a53cca | 3721 | int maybe_acls; |
618e21d5 | 3722 | u32 rdev; |
39279cc3 | 3723 | int ret; |
2f7e33d4 | 3724 | bool filled = false; |
63541927 | 3725 | int first_xattr_slot; |
2f7e33d4 MX |
3726 | |
3727 | ret = btrfs_fill_inode(inode, &rdev); | |
3728 | if (!ret) | |
3729 | filled = true; | |
39279cc3 | 3730 | |
4222ea71 FM |
3731 | if (!path) { |
3732 | path = btrfs_alloc_path(); | |
3733 | if (!path) | |
3734 | return -ENOMEM; | |
3735 | } | |
1748f843 | 3736 | |
39279cc3 | 3737 | memcpy(&location, &BTRFS_I(inode)->location, sizeof(location)); |
dc17ff8f | 3738 | |
39279cc3 | 3739 | ret = btrfs_lookup_inode(NULL, root, path, &location, 0); |
67710892 | 3740 | if (ret) { |
4222ea71 FM |
3741 | if (path != in_path) |
3742 | btrfs_free_path(path); | |
f5b3a417 | 3743 | return ret; |
67710892 | 3744 | } |
39279cc3 | 3745 | |
5f39d397 | 3746 | leaf = path->nodes[0]; |
2f7e33d4 MX |
3747 | |
3748 | if (filled) | |
67de1176 | 3749 | goto cache_index; |
2f7e33d4 | 3750 | |
5f39d397 CM |
3751 | inode_item = btrfs_item_ptr(leaf, path->slots[0], |
3752 | struct btrfs_inode_item); | |
5f39d397 | 3753 | inode->i_mode = btrfs_inode_mode(leaf, inode_item); |
bfe86848 | 3754 | set_nlink(inode, btrfs_inode_nlink(leaf, inode_item)); |
2f2f43d3 EB |
3755 | i_uid_write(inode, btrfs_inode_uid(leaf, inode_item)); |
3756 | i_gid_write(inode, btrfs_inode_gid(leaf, inode_item)); | |
6ef06d27 | 3757 | btrfs_i_size_write(BTRFS_I(inode), btrfs_inode_size(leaf, inode_item)); |
41a2ee75 JB |
3758 | btrfs_inode_set_file_extent_range(BTRFS_I(inode), 0, |
3759 | round_up(i_size_read(inode), fs_info->sectorsize)); | |
5f39d397 | 3760 | |
a937b979 DS |
3761 | inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->atime); |
3762 | inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->atime); | |
5f39d397 | 3763 | |
a937b979 DS |
3764 | inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->mtime); |
3765 | inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->mtime); | |
5f39d397 | 3766 | |
2a9462de JL |
3767 | inode_set_ctime(inode, btrfs_timespec_sec(leaf, &inode_item->ctime), |
3768 | btrfs_timespec_nsec(leaf, &inode_item->ctime)); | |
5f39d397 | 3769 | |
9cc97d64 | 3770 | BTRFS_I(inode)->i_otime.tv_sec = |
3771 | btrfs_timespec_sec(leaf, &inode_item->otime); | |
3772 | BTRFS_I(inode)->i_otime.tv_nsec = | |
3773 | btrfs_timespec_nsec(leaf, &inode_item->otime); | |
5f39d397 | 3774 | |
a76a3cd4 | 3775 | inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item)); |
e02119d5 | 3776 | BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item); |
5dc562c5 JB |
3777 | BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item); |
3778 | ||
c7f88c4e JL |
3779 | inode_set_iversion_queried(inode, |
3780 | btrfs_inode_sequence(leaf, inode_item)); | |
6e17d30b YD |
3781 | inode->i_generation = BTRFS_I(inode)->generation; |
3782 | inode->i_rdev = 0; | |
3783 | rdev = btrfs_inode_rdev(leaf, inode_item); | |
3784 | ||
3785 | BTRFS_I(inode)->index_cnt = (u64)-1; | |
77eea05e BB |
3786 | btrfs_inode_split_flags(btrfs_inode_flags(leaf, inode_item), |
3787 | &BTRFS_I(inode)->flags, &BTRFS_I(inode)->ro_flags); | |
6e17d30b YD |
3788 | |
3789 | cache_index: | |
5dc562c5 JB |
3790 | /* |
3791 | * If we were modified in the current generation and evicted from memory | |
3792 | * and then re-read we need to do a full sync since we don't have any | |
3793 | * idea about which extents were modified before we were evicted from | |
3794 | * cache. | |
6e17d30b YD |
3795 | * |
3796 | * This is required for both inode re-read from disk and delayed inode | |
088aea3b | 3797 | * in delayed_nodes_tree. |
5dc562c5 | 3798 | */ |
0b246afa | 3799 | if (BTRFS_I(inode)->last_trans == fs_info->generation) |
5dc562c5 JB |
3800 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, |
3801 | &BTRFS_I(inode)->runtime_flags); | |
3802 | ||
bde6c242 FM |
3803 | /* |
3804 | * We don't persist the id of the transaction where an unlink operation | |
3805 | * against the inode was last made. So here we assume the inode might | |
3806 | * have been evicted, and therefore the exact value of last_unlink_trans | |
3807 | * lost, and set it to last_trans to avoid metadata inconsistencies | |
3808 | * between the inode and its parent if the inode is fsync'ed and the log | |
3809 | * replayed. For example, in the scenario: | |
3810 | * | |
3811 | * touch mydir/foo | |
3812 | * ln mydir/foo mydir/bar | |
3813 | * sync | |
3814 | * unlink mydir/bar | |
3815 | * echo 2 > /proc/sys/vm/drop_caches # evicts inode | |
3816 | * xfs_io -c fsync mydir/foo | |
3817 | * <power failure> | |
3818 | * mount fs, triggers fsync log replay | |
3819 | * | |
3820 | * We must make sure that when we fsync our inode foo we also log its | |
3821 | * parent inode, otherwise after log replay the parent still has the | |
3822 | * dentry with the "bar" name but our inode foo has a link count of 1 | |
3823 | * and doesn't have an inode ref with the name "bar" anymore. | |
3824 | * | |
3825 | * Setting last_unlink_trans to last_trans is a pessimistic approach, | |
01327610 | 3826 | * but it guarantees correctness at the expense of occasional full |
bde6c242 FM |
3827 | * transaction commits on fsync if our inode is a directory, or if our |
3828 | * inode is not a directory, logging its parent unnecessarily. | |
3829 | */ | |
3830 | BTRFS_I(inode)->last_unlink_trans = BTRFS_I(inode)->last_trans; | |
3831 | ||
3ebac17c FM |
3832 | /* |
3833 | * Same logic as for last_unlink_trans. We don't persist the generation | |
3834 | * of the last transaction where this inode was used for a reflink | |
3835 | * operation, so after eviction and reloading the inode we must be | |
3836 | * pessimistic and assume the last transaction that modified the inode. | |
3837 | */ | |
3838 | BTRFS_I(inode)->last_reflink_trans = BTRFS_I(inode)->last_trans; | |
3839 | ||
67de1176 MX |
3840 | path->slots[0]++; |
3841 | if (inode->i_nlink != 1 || | |
3842 | path->slots[0] >= btrfs_header_nritems(leaf)) | |
3843 | goto cache_acl; | |
3844 | ||
3845 | btrfs_item_key_to_cpu(leaf, &location, path->slots[0]); | |
4a0cc7ca | 3846 | if (location.objectid != btrfs_ino(BTRFS_I(inode))) |
67de1176 MX |
3847 | goto cache_acl; |
3848 | ||
3849 | ptr = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
3850 | if (location.type == BTRFS_INODE_REF_KEY) { | |
3851 | struct btrfs_inode_ref *ref; | |
3852 | ||
3853 | ref = (struct btrfs_inode_ref *)ptr; | |
3854 | BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref); | |
3855 | } else if (location.type == BTRFS_INODE_EXTREF_KEY) { | |
3856 | struct btrfs_inode_extref *extref; | |
3857 | ||
3858 | extref = (struct btrfs_inode_extref *)ptr; | |
3859 | BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf, | |
3860 | extref); | |
3861 | } | |
2f7e33d4 | 3862 | cache_acl: |
46a53cca CM |
3863 | /* |
3864 | * try to precache a NULL acl entry for files that don't have | |
3865 | * any xattrs or acls | |
3866 | */ | |
33345d01 | 3867 | maybe_acls = acls_after_inode_item(leaf, path->slots[0], |
f85b7379 | 3868 | btrfs_ino(BTRFS_I(inode)), &first_xattr_slot); |
63541927 FDBM |
3869 | if (first_xattr_slot != -1) { |
3870 | path->slots[0] = first_xattr_slot; | |
3871 | ret = btrfs_load_inode_props(inode, path); | |
3872 | if (ret) | |
0b246afa | 3873 | btrfs_err(fs_info, |
351fd353 | 3874 | "error loading props for ino %llu (root %llu): %d", |
4a0cc7ca | 3875 | btrfs_ino(BTRFS_I(inode)), |
63541927 FDBM |
3876 | root->root_key.objectid, ret); |
3877 | } | |
4222ea71 FM |
3878 | if (path != in_path) |
3879 | btrfs_free_path(path); | |
63541927 | 3880 | |
72c04902 AV |
3881 | if (!maybe_acls) |
3882 | cache_no_acl(inode); | |
46a53cca | 3883 | |
39279cc3 | 3884 | switch (inode->i_mode & S_IFMT) { |
39279cc3 CM |
3885 | case S_IFREG: |
3886 | inode->i_mapping->a_ops = &btrfs_aops; | |
3887 | inode->i_fop = &btrfs_file_operations; | |
3888 | inode->i_op = &btrfs_file_inode_operations; | |
3889 | break; | |
3890 | case S_IFDIR: | |
3891 | inode->i_fop = &btrfs_dir_file_operations; | |
67ade058 | 3892 | inode->i_op = &btrfs_dir_inode_operations; |
39279cc3 CM |
3893 | break; |
3894 | case S_IFLNK: | |
3895 | inode->i_op = &btrfs_symlink_inode_operations; | |
21fc61c7 | 3896 | inode_nohighmem(inode); |
4779cc04 | 3897 | inode->i_mapping->a_ops = &btrfs_aops; |
39279cc3 | 3898 | break; |
618e21d5 | 3899 | default: |
0279b4cd | 3900 | inode->i_op = &btrfs_special_inode_operations; |
618e21d5 JB |
3901 | init_special_inode(inode, inode->i_mode, rdev); |
3902 | break; | |
39279cc3 | 3903 | } |
6cbff00f | 3904 | |
7b6a221e | 3905 | btrfs_sync_inode_flags_to_i_flags(inode); |
67710892 | 3906 | return 0; |
39279cc3 CM |
3907 | } |
3908 | ||
d352ac68 CM |
3909 | /* |
3910 | * given a leaf and an inode, copy the inode fields into the leaf | |
3911 | */ | |
e02119d5 CM |
3912 | static void fill_inode_item(struct btrfs_trans_handle *trans, |
3913 | struct extent_buffer *leaf, | |
5f39d397 | 3914 | struct btrfs_inode_item *item, |
39279cc3 CM |
3915 | struct inode *inode) |
3916 | { | |
51fab693 | 3917 | struct btrfs_map_token token; |
77eea05e | 3918 | u64 flags; |
51fab693 | 3919 | |
c82f823c | 3920 | btrfs_init_map_token(&token, leaf); |
5f39d397 | 3921 | |
cc4c13d5 DS |
3922 | btrfs_set_token_inode_uid(&token, item, i_uid_read(inode)); |
3923 | btrfs_set_token_inode_gid(&token, item, i_gid_read(inode)); | |
3924 | btrfs_set_token_inode_size(&token, item, BTRFS_I(inode)->disk_i_size); | |
3925 | btrfs_set_token_inode_mode(&token, item, inode->i_mode); | |
3926 | btrfs_set_token_inode_nlink(&token, item, inode->i_nlink); | |
3927 | ||
3928 | btrfs_set_token_timespec_sec(&token, &item->atime, | |
3929 | inode->i_atime.tv_sec); | |
3930 | btrfs_set_token_timespec_nsec(&token, &item->atime, | |
3931 | inode->i_atime.tv_nsec); | |
3932 | ||
3933 | btrfs_set_token_timespec_sec(&token, &item->mtime, | |
3934 | inode->i_mtime.tv_sec); | |
3935 | btrfs_set_token_timespec_nsec(&token, &item->mtime, | |
3936 | inode->i_mtime.tv_nsec); | |
3937 | ||
3938 | btrfs_set_token_timespec_sec(&token, &item->ctime, | |
2a9462de | 3939 | inode_get_ctime(inode).tv_sec); |
cc4c13d5 | 3940 | btrfs_set_token_timespec_nsec(&token, &item->ctime, |
2a9462de | 3941 | inode_get_ctime(inode).tv_nsec); |
cc4c13d5 DS |
3942 | |
3943 | btrfs_set_token_timespec_sec(&token, &item->otime, | |
3944 | BTRFS_I(inode)->i_otime.tv_sec); | |
3945 | btrfs_set_token_timespec_nsec(&token, &item->otime, | |
3946 | BTRFS_I(inode)->i_otime.tv_nsec); | |
3947 | ||
3948 | btrfs_set_token_inode_nbytes(&token, item, inode_get_bytes(inode)); | |
3949 | btrfs_set_token_inode_generation(&token, item, | |
3950 | BTRFS_I(inode)->generation); | |
3951 | btrfs_set_token_inode_sequence(&token, item, inode_peek_iversion(inode)); | |
3952 | btrfs_set_token_inode_transid(&token, item, trans->transid); | |
3953 | btrfs_set_token_inode_rdev(&token, item, inode->i_rdev); | |
77eea05e BB |
3954 | flags = btrfs_inode_combine_flags(BTRFS_I(inode)->flags, |
3955 | BTRFS_I(inode)->ro_flags); | |
3956 | btrfs_set_token_inode_flags(&token, item, flags); | |
cc4c13d5 | 3957 | btrfs_set_token_inode_block_group(&token, item, 0); |
39279cc3 CM |
3958 | } |
3959 | ||
d352ac68 CM |
3960 | /* |
3961 | * copy everything in the in-memory inode into the btree. | |
3962 | */ | |
2115133f | 3963 | static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans, |
dfeb9e7c NB |
3964 | struct btrfs_root *root, |
3965 | struct btrfs_inode *inode) | |
39279cc3 CM |
3966 | { |
3967 | struct btrfs_inode_item *inode_item; | |
3968 | struct btrfs_path *path; | |
5f39d397 | 3969 | struct extent_buffer *leaf; |
39279cc3 CM |
3970 | int ret; |
3971 | ||
3972 | path = btrfs_alloc_path(); | |
16cdcec7 MX |
3973 | if (!path) |
3974 | return -ENOMEM; | |
3975 | ||
dfeb9e7c | 3976 | ret = btrfs_lookup_inode(trans, root, path, &inode->location, 1); |
39279cc3 CM |
3977 | if (ret) { |
3978 | if (ret > 0) | |
3979 | ret = -ENOENT; | |
3980 | goto failed; | |
3981 | } | |
3982 | ||
5f39d397 CM |
3983 | leaf = path->nodes[0]; |
3984 | inode_item = btrfs_item_ptr(leaf, path->slots[0], | |
16cdcec7 | 3985 | struct btrfs_inode_item); |
39279cc3 | 3986 | |
dfeb9e7c | 3987 | fill_inode_item(trans, leaf, inode_item, &inode->vfs_inode); |
5f39d397 | 3988 | btrfs_mark_buffer_dirty(leaf); |
dfeb9e7c | 3989 | btrfs_set_inode_last_trans(trans, inode); |
39279cc3 CM |
3990 | ret = 0; |
3991 | failed: | |
39279cc3 CM |
3992 | btrfs_free_path(path); |
3993 | return ret; | |
3994 | } | |
3995 | ||
2115133f CM |
3996 | /* |
3997 | * copy everything in the in-memory inode into the btree. | |
3998 | */ | |
3999 | noinline int btrfs_update_inode(struct btrfs_trans_handle *trans, | |
9a56fcd1 NB |
4000 | struct btrfs_root *root, |
4001 | struct btrfs_inode *inode) | |
2115133f | 4002 | { |
0b246afa | 4003 | struct btrfs_fs_info *fs_info = root->fs_info; |
2115133f CM |
4004 | int ret; |
4005 | ||
4006 | /* | |
4007 | * If the inode is a free space inode, we can deadlock during commit | |
4008 | * if we put it into the delayed code. | |
4009 | * | |
4010 | * The data relocation inode should also be directly updated | |
4011 | * without delay | |
4012 | */ | |
9a56fcd1 | 4013 | if (!btrfs_is_free_space_inode(inode) |
37f00a6d | 4014 | && !btrfs_is_data_reloc_root(root) |
0b246afa | 4015 | && !test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) { |
8ea05e3a AB |
4016 | btrfs_update_root_times(trans, root); |
4017 | ||
9a56fcd1 | 4018 | ret = btrfs_delayed_update_inode(trans, root, inode); |
2115133f | 4019 | if (!ret) |
9a56fcd1 | 4020 | btrfs_set_inode_last_trans(trans, inode); |
2115133f CM |
4021 | return ret; |
4022 | } | |
4023 | ||
9a56fcd1 | 4024 | return btrfs_update_inode_item(trans, root, inode); |
2115133f CM |
4025 | } |
4026 | ||
729f7961 NB |
4027 | int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans, |
4028 | struct btrfs_root *root, struct btrfs_inode *inode) | |
2115133f CM |
4029 | { |
4030 | int ret; | |
4031 | ||
729f7961 | 4032 | ret = btrfs_update_inode(trans, root, inode); |
2115133f | 4033 | if (ret == -ENOSPC) |
729f7961 | 4034 | return btrfs_update_inode_item(trans, root, inode); |
2115133f CM |
4035 | return ret; |
4036 | } | |
4037 | ||
d352ac68 CM |
4038 | /* |
4039 | * unlink helper that gets used here in inode.c and in the tree logging | |
4040 | * recovery code. It remove a link in a directory with a given name, and | |
4041 | * also drops the back refs in the inode to the directory | |
4042 | */ | |
92986796 | 4043 | static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans, |
4ec5934e NB |
4044 | struct btrfs_inode *dir, |
4045 | struct btrfs_inode *inode, | |
6db75318 | 4046 | const struct fscrypt_str *name, |
88d2beec | 4047 | struct btrfs_rename_ctx *rename_ctx) |
39279cc3 | 4048 | { |
4467af88 | 4049 | struct btrfs_root *root = dir->root; |
0b246afa | 4050 | struct btrfs_fs_info *fs_info = root->fs_info; |
39279cc3 | 4051 | struct btrfs_path *path; |
39279cc3 | 4052 | int ret = 0; |
39279cc3 | 4053 | struct btrfs_dir_item *di; |
aec7477b | 4054 | u64 index; |
33345d01 LZ |
4055 | u64 ino = btrfs_ino(inode); |
4056 | u64 dir_ino = btrfs_ino(dir); | |
39279cc3 CM |
4057 | |
4058 | path = btrfs_alloc_path(); | |
54aa1f4d CM |
4059 | if (!path) { |
4060 | ret = -ENOMEM; | |
554233a6 | 4061 | goto out; |
54aa1f4d CM |
4062 | } |
4063 | ||
e43eec81 | 4064 | di = btrfs_lookup_dir_item(trans, root, path, dir_ino, name, -1); |
3cf5068f LB |
4065 | if (IS_ERR_OR_NULL(di)) { |
4066 | ret = di ? PTR_ERR(di) : -ENOENT; | |
39279cc3 CM |
4067 | goto err; |
4068 | } | |
39279cc3 | 4069 | ret = btrfs_delete_one_dir_name(trans, root, path, di); |
54aa1f4d CM |
4070 | if (ret) |
4071 | goto err; | |
b3b4aa74 | 4072 | btrfs_release_path(path); |
39279cc3 | 4073 | |
67de1176 MX |
4074 | /* |
4075 | * If we don't have dir index, we have to get it by looking up | |
4076 | * the inode ref, since we get the inode ref, remove it directly, | |
4077 | * it is unnecessary to do delayed deletion. | |
4078 | * | |
4079 | * But if we have dir index, needn't search inode ref to get it. | |
4080 | * Since the inode ref is close to the inode item, it is better | |
4081 | * that we delay to delete it, and just do this deletion when | |
4082 | * we update the inode item. | |
4083 | */ | |
4ec5934e | 4084 | if (inode->dir_index) { |
67de1176 MX |
4085 | ret = btrfs_delayed_delete_inode_ref(inode); |
4086 | if (!ret) { | |
4ec5934e | 4087 | index = inode->dir_index; |
67de1176 MX |
4088 | goto skip_backref; |
4089 | } | |
4090 | } | |
4091 | ||
e43eec81 | 4092 | ret = btrfs_del_inode_ref(trans, root, name, ino, dir_ino, &index); |
aec7477b | 4093 | if (ret) { |
0b246afa | 4094 | btrfs_info(fs_info, |
c2cf52eb | 4095 | "failed to delete reference to %.*s, inode %llu parent %llu", |
e43eec81 | 4096 | name->len, name->name, ino, dir_ino); |
66642832 | 4097 | btrfs_abort_transaction(trans, ret); |
aec7477b JB |
4098 | goto err; |
4099 | } | |
67de1176 | 4100 | skip_backref: |
88d2beec FM |
4101 | if (rename_ctx) |
4102 | rename_ctx->index = index; | |
4103 | ||
9add2945 | 4104 | ret = btrfs_delete_delayed_dir_index(trans, dir, index); |
79787eaa | 4105 | if (ret) { |
66642832 | 4106 | btrfs_abort_transaction(trans, ret); |
39279cc3 | 4107 | goto err; |
79787eaa | 4108 | } |
39279cc3 | 4109 | |
259c4b96 FM |
4110 | /* |
4111 | * If we are in a rename context, we don't need to update anything in the | |
4112 | * log. That will be done later during the rename by btrfs_log_new_name(). | |
143823cf | 4113 | * Besides that, doing it here would only cause extra unnecessary btree |
259c4b96 FM |
4114 | * operations on the log tree, increasing latency for applications. |
4115 | */ | |
4116 | if (!rename_ctx) { | |
e43eec81 STD |
4117 | btrfs_del_inode_ref_in_log(trans, root, name, inode, dir_ino); |
4118 | btrfs_del_dir_entries_in_log(trans, root, name, dir, index); | |
259c4b96 | 4119 | } |
63611e73 JB |
4120 | |
4121 | /* | |
4122 | * If we have a pending delayed iput we could end up with the final iput | |
4123 | * being run in btrfs-cleaner context. If we have enough of these built | |
4124 | * up we can end up burning a lot of time in btrfs-cleaner without any | |
4125 | * way to throttle the unlinks. Since we're currently holding a ref on | |
4126 | * the inode we can run the delayed iput here without any issues as the | |
4127 | * final iput won't be done until after we drop the ref we're currently | |
4128 | * holding. | |
4129 | */ | |
4130 | btrfs_run_delayed_iput(fs_info, inode); | |
39279cc3 CM |
4131 | err: |
4132 | btrfs_free_path(path); | |
e02119d5 CM |
4133 | if (ret) |
4134 | goto out; | |
4135 | ||
e43eec81 | 4136 | btrfs_i_size_write(dir, dir->vfs_inode.i_size - name->len * 2); |
4ec5934e NB |
4137 | inode_inc_iversion(&inode->vfs_inode); |
4138 | inode_inc_iversion(&dir->vfs_inode); | |
2a9462de JL |
4139 | inode_set_ctime_current(&inode->vfs_inode); |
4140 | dir->vfs_inode.i_mtime = inode_set_ctime_current(&dir->vfs_inode); | |
9a56fcd1 | 4141 | ret = btrfs_update_inode(trans, root, dir); |
e02119d5 | 4142 | out: |
39279cc3 CM |
4143 | return ret; |
4144 | } | |
4145 | ||
92986796 | 4146 | int btrfs_unlink_inode(struct btrfs_trans_handle *trans, |
4ec5934e | 4147 | struct btrfs_inode *dir, struct btrfs_inode *inode, |
6db75318 | 4148 | const struct fscrypt_str *name) |
92986796 AV |
4149 | { |
4150 | int ret; | |
e43eec81 STD |
4151 | |
4152 | ret = __btrfs_unlink_inode(trans, dir, inode, name, NULL); | |
92986796 | 4153 | if (!ret) { |
4ec5934e | 4154 | drop_nlink(&inode->vfs_inode); |
4467af88 | 4155 | ret = btrfs_update_inode(trans, inode->root, inode); |
92986796 AV |
4156 | } |
4157 | return ret; | |
4158 | } | |
39279cc3 | 4159 | |
a22285a6 YZ |
4160 | /* |
4161 | * helper to start transaction for unlink and rmdir. | |
4162 | * | |
d52be818 JB |
4163 | * unlink and rmdir are special in btrfs, they do not always free space, so |
4164 | * if we cannot make our reservations the normal way try and see if there is | |
4165 | * plenty of slack room in the global reserve to migrate, otherwise we cannot | |
4166 | * allow the unlink to occur. | |
a22285a6 | 4167 | */ |
e569b1d5 | 4168 | static struct btrfs_trans_handle *__unlink_start_trans(struct btrfs_inode *dir) |
4df27c4d | 4169 | { |
e569b1d5 | 4170 | struct btrfs_root *root = dir->root; |
4df27c4d | 4171 | |
5630e2bc FM |
4172 | return btrfs_start_transaction_fallback_global_rsv(root, |
4173 | BTRFS_UNLINK_METADATA_UNITS); | |
a22285a6 YZ |
4174 | } |
4175 | ||
4176 | static int btrfs_unlink(struct inode *dir, struct dentry *dentry) | |
4177 | { | |
a22285a6 | 4178 | struct btrfs_trans_handle *trans; |
2b0143b5 | 4179 | struct inode *inode = d_inode(dentry); |
a22285a6 | 4180 | int ret; |
ab3c5c18 | 4181 | struct fscrypt_name fname; |
a22285a6 | 4182 | |
ab3c5c18 STD |
4183 | ret = fscrypt_setup_filename(dir, &dentry->d_name, 1, &fname); |
4184 | if (ret) | |
4185 | return ret; | |
ab3c5c18 STD |
4186 | |
4187 | /* This needs to handle no-key deletions later on */ | |
a22285a6 | 4188 | |
e569b1d5 | 4189 | trans = __unlink_start_trans(BTRFS_I(dir)); |
ab3c5c18 STD |
4190 | if (IS_ERR(trans)) { |
4191 | ret = PTR_ERR(trans); | |
4192 | goto fscrypt_free; | |
4193 | } | |
5f39d397 | 4194 | |
4ec5934e | 4195 | btrfs_record_unlink_dir(trans, BTRFS_I(dir), BTRFS_I(d_inode(dentry)), |
59fcf388 | 4196 | false); |
12fcfd22 | 4197 | |
e43eec81 | 4198 | ret = btrfs_unlink_inode(trans, BTRFS_I(dir), BTRFS_I(d_inode(dentry)), |
6db75318 | 4199 | &fname.disk_name); |
b532402e | 4200 | if (ret) |
ab3c5c18 | 4201 | goto end_trans; |
7b128766 | 4202 | |
a22285a6 | 4203 | if (inode->i_nlink == 0) { |
73f2e545 | 4204 | ret = btrfs_orphan_add(trans, BTRFS_I(inode)); |
b532402e | 4205 | if (ret) |
ab3c5c18 | 4206 | goto end_trans; |
a22285a6 | 4207 | } |
7b128766 | 4208 | |
ab3c5c18 | 4209 | end_trans: |
3a45bb20 | 4210 | btrfs_end_transaction(trans); |
4467af88 | 4211 | btrfs_btree_balance_dirty(BTRFS_I(dir)->root->fs_info); |
ab3c5c18 STD |
4212 | fscrypt_free: |
4213 | fscrypt_free_filename(&fname); | |
39279cc3 CM |
4214 | return ret; |
4215 | } | |
4216 | ||
f60a2364 | 4217 | static int btrfs_unlink_subvol(struct btrfs_trans_handle *trans, |
5b7544cb | 4218 | struct btrfs_inode *dir, struct dentry *dentry) |
4df27c4d | 4219 | { |
5b7544cb | 4220 | struct btrfs_root *root = dir->root; |
045d3967 | 4221 | struct btrfs_inode *inode = BTRFS_I(d_inode(dentry)); |
4df27c4d YZ |
4222 | struct btrfs_path *path; |
4223 | struct extent_buffer *leaf; | |
4224 | struct btrfs_dir_item *di; | |
4225 | struct btrfs_key key; | |
4226 | u64 index; | |
4227 | int ret; | |
045d3967 | 4228 | u64 objectid; |
5b7544cb | 4229 | u64 dir_ino = btrfs_ino(dir); |
ab3c5c18 STD |
4230 | struct fscrypt_name fname; |
4231 | ||
5b7544cb | 4232 | ret = fscrypt_setup_filename(&dir->vfs_inode, &dentry->d_name, 1, &fname); |
ab3c5c18 STD |
4233 | if (ret) |
4234 | return ret; | |
ab3c5c18 STD |
4235 | |
4236 | /* This needs to handle no-key deletions later on */ | |
4df27c4d | 4237 | |
045d3967 JB |
4238 | if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID) { |
4239 | objectid = inode->root->root_key.objectid; | |
4240 | } else if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) { | |
4241 | objectid = inode->location.objectid; | |
4242 | } else { | |
4243 | WARN_ON(1); | |
ab3c5c18 | 4244 | fscrypt_free_filename(&fname); |
045d3967 JB |
4245 | return -EINVAL; |
4246 | } | |
4247 | ||
4df27c4d | 4248 | path = btrfs_alloc_path(); |
ab3c5c18 STD |
4249 | if (!path) { |
4250 | ret = -ENOMEM; | |
4251 | goto out; | |
4252 | } | |
4df27c4d | 4253 | |
33345d01 | 4254 | di = btrfs_lookup_dir_item(trans, root, path, dir_ino, |
6db75318 | 4255 | &fname.disk_name, -1); |
79787eaa | 4256 | if (IS_ERR_OR_NULL(di)) { |
3cf5068f | 4257 | ret = di ? PTR_ERR(di) : -ENOENT; |
79787eaa JM |
4258 | goto out; |
4259 | } | |
4df27c4d YZ |
4260 | |
4261 | leaf = path->nodes[0]; | |
4262 | btrfs_dir_item_key_to_cpu(leaf, di, &key); | |
4263 | WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid); | |
4264 | ret = btrfs_delete_one_dir_name(trans, root, path, di); | |
79787eaa | 4265 | if (ret) { |
66642832 | 4266 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
4267 | goto out; |
4268 | } | |
b3b4aa74 | 4269 | btrfs_release_path(path); |
4df27c4d | 4270 | |
d49d3287 JB |
4271 | /* |
4272 | * This is a placeholder inode for a subvolume we didn't have a | |
4273 | * reference to at the time of the snapshot creation. In the meantime | |
4274 | * we could have renamed the real subvol link into our snapshot, so | |
1a9fd417 | 4275 | * depending on btrfs_del_root_ref to return -ENOENT here is incorrect. |
d49d3287 JB |
4276 | * Instead simply lookup the dir_index_item for this entry so we can |
4277 | * remove it. Otherwise we know we have a ref to the root and we can | |
4278 | * call btrfs_del_root_ref, and it _shouldn't_ fail. | |
4279 | */ | |
4280 | if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) { | |
6db75318 | 4281 | di = btrfs_search_dir_index_item(root, path, dir_ino, &fname.disk_name); |
79787eaa JM |
4282 | if (IS_ERR_OR_NULL(di)) { |
4283 | if (!di) | |
4284 | ret = -ENOENT; | |
4285 | else | |
4286 | ret = PTR_ERR(di); | |
66642832 | 4287 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
4288 | goto out; |
4289 | } | |
4df27c4d YZ |
4290 | |
4291 | leaf = path->nodes[0]; | |
4292 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
4df27c4d | 4293 | index = key.offset; |
d49d3287 JB |
4294 | btrfs_release_path(path); |
4295 | } else { | |
4296 | ret = btrfs_del_root_ref(trans, objectid, | |
4297 | root->root_key.objectid, dir_ino, | |
6db75318 | 4298 | &index, &fname.disk_name); |
d49d3287 JB |
4299 | if (ret) { |
4300 | btrfs_abort_transaction(trans, ret); | |
4301 | goto out; | |
4302 | } | |
4df27c4d YZ |
4303 | } |
4304 | ||
5b7544cb | 4305 | ret = btrfs_delete_delayed_dir_index(trans, dir, index); |
79787eaa | 4306 | if (ret) { |
66642832 | 4307 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
4308 | goto out; |
4309 | } | |
4df27c4d | 4310 | |
5b7544cb DS |
4311 | btrfs_i_size_write(dir, dir->vfs_inode.i_size - fname.disk_name.len * 2); |
4312 | inode_inc_iversion(&dir->vfs_inode); | |
2a9462de | 4313 | dir->vfs_inode.i_mtime = inode_set_ctime_current(&dir->vfs_inode); |
5b7544cb | 4314 | ret = btrfs_update_inode_fallback(trans, root, dir); |
79787eaa | 4315 | if (ret) |
66642832 | 4316 | btrfs_abort_transaction(trans, ret); |
79787eaa | 4317 | out: |
71d7aed0 | 4318 | btrfs_free_path(path); |
ab3c5c18 | 4319 | fscrypt_free_filename(&fname); |
79787eaa | 4320 | return ret; |
4df27c4d YZ |
4321 | } |
4322 | ||
ec42f167 MT |
4323 | /* |
4324 | * Helper to check if the subvolume references other subvolumes or if it's | |
4325 | * default. | |
4326 | */ | |
f60a2364 | 4327 | static noinline int may_destroy_subvol(struct btrfs_root *root) |
ec42f167 MT |
4328 | { |
4329 | struct btrfs_fs_info *fs_info = root->fs_info; | |
4330 | struct btrfs_path *path; | |
4331 | struct btrfs_dir_item *di; | |
4332 | struct btrfs_key key; | |
6db75318 | 4333 | struct fscrypt_str name = FSTR_INIT("default", 7); |
ec42f167 MT |
4334 | u64 dir_id; |
4335 | int ret; | |
4336 | ||
4337 | path = btrfs_alloc_path(); | |
4338 | if (!path) | |
4339 | return -ENOMEM; | |
4340 | ||
4341 | /* Make sure this root isn't set as the default subvol */ | |
4342 | dir_id = btrfs_super_root_dir(fs_info->super_copy); | |
4343 | di = btrfs_lookup_dir_item(NULL, fs_info->tree_root, path, | |
e43eec81 | 4344 | dir_id, &name, 0); |
ec42f167 MT |
4345 | if (di && !IS_ERR(di)) { |
4346 | btrfs_dir_item_key_to_cpu(path->nodes[0], di, &key); | |
4347 | if (key.objectid == root->root_key.objectid) { | |
4348 | ret = -EPERM; | |
4349 | btrfs_err(fs_info, | |
4350 | "deleting default subvolume %llu is not allowed", | |
4351 | key.objectid); | |
4352 | goto out; | |
4353 | } | |
4354 | btrfs_release_path(path); | |
4355 | } | |
4356 | ||
4357 | key.objectid = root->root_key.objectid; | |
4358 | key.type = BTRFS_ROOT_REF_KEY; | |
4359 | key.offset = (u64)-1; | |
4360 | ||
4361 | ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0); | |
4362 | if (ret < 0) | |
4363 | goto out; | |
4364 | BUG_ON(ret == 0); | |
4365 | ||
4366 | ret = 0; | |
4367 | if (path->slots[0] > 0) { | |
4368 | path->slots[0]--; | |
4369 | btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]); | |
4370 | if (key.objectid == root->root_key.objectid && | |
4371 | key.type == BTRFS_ROOT_REF_KEY) | |
4372 | ret = -ENOTEMPTY; | |
4373 | } | |
4374 | out: | |
4375 | btrfs_free_path(path); | |
4376 | return ret; | |
4377 | } | |
4378 | ||
20a68004 NB |
4379 | /* Delete all dentries for inodes belonging to the root */ |
4380 | static void btrfs_prune_dentries(struct btrfs_root *root) | |
4381 | { | |
4382 | struct btrfs_fs_info *fs_info = root->fs_info; | |
4383 | struct rb_node *node; | |
4384 | struct rb_node *prev; | |
4385 | struct btrfs_inode *entry; | |
4386 | struct inode *inode; | |
4387 | u64 objectid = 0; | |
4388 | ||
84961539 | 4389 | if (!BTRFS_FS_ERROR(fs_info)) |
20a68004 NB |
4390 | WARN_ON(btrfs_root_refs(&root->root_item) != 0); |
4391 | ||
4392 | spin_lock(&root->inode_lock); | |
4393 | again: | |
4394 | node = root->inode_tree.rb_node; | |
4395 | prev = NULL; | |
4396 | while (node) { | |
4397 | prev = node; | |
4398 | entry = rb_entry(node, struct btrfs_inode, rb_node); | |
4399 | ||
37508515 | 4400 | if (objectid < btrfs_ino(entry)) |
20a68004 | 4401 | node = node->rb_left; |
37508515 | 4402 | else if (objectid > btrfs_ino(entry)) |
20a68004 NB |
4403 | node = node->rb_right; |
4404 | else | |
4405 | break; | |
4406 | } | |
4407 | if (!node) { | |
4408 | while (prev) { | |
4409 | entry = rb_entry(prev, struct btrfs_inode, rb_node); | |
37508515 | 4410 | if (objectid <= btrfs_ino(entry)) { |
20a68004 NB |
4411 | node = prev; |
4412 | break; | |
4413 | } | |
4414 | prev = rb_next(prev); | |
4415 | } | |
4416 | } | |
4417 | while (node) { | |
4418 | entry = rb_entry(node, struct btrfs_inode, rb_node); | |
37508515 | 4419 | objectid = btrfs_ino(entry) + 1; |
20a68004 NB |
4420 | inode = igrab(&entry->vfs_inode); |
4421 | if (inode) { | |
4422 | spin_unlock(&root->inode_lock); | |
4423 | if (atomic_read(&inode->i_count) > 1) | |
4424 | d_prune_aliases(inode); | |
4425 | /* | |
4426 | * btrfs_drop_inode will have it removed from the inode | |
4427 | * cache when its usage count hits zero. | |
4428 | */ | |
4429 | iput(inode); | |
4430 | cond_resched(); | |
4431 | spin_lock(&root->inode_lock); | |
4432 | goto again; | |
4433 | } | |
4434 | ||
4435 | if (cond_resched_lock(&root->inode_lock)) | |
4436 | goto again; | |
4437 | ||
4438 | node = rb_next(node); | |
4439 | } | |
4440 | spin_unlock(&root->inode_lock); | |
4441 | } | |
4442 | ||
3c4f91e2 | 4443 | int btrfs_delete_subvolume(struct btrfs_inode *dir, struct dentry *dentry) |
f60a2364 MT |
4444 | { |
4445 | struct btrfs_fs_info *fs_info = btrfs_sb(dentry->d_sb); | |
3c4f91e2 | 4446 | struct btrfs_root *root = dir->root; |
f60a2364 MT |
4447 | struct inode *inode = d_inode(dentry); |
4448 | struct btrfs_root *dest = BTRFS_I(inode)->root; | |
4449 | struct btrfs_trans_handle *trans; | |
4450 | struct btrfs_block_rsv block_rsv; | |
4451 | u64 root_flags; | |
f60a2364 | 4452 | int ret; |
f60a2364 MT |
4453 | |
4454 | /* | |
4455 | * Don't allow to delete a subvolume with send in progress. This is | |
4456 | * inside the inode lock so the error handling that has to drop the bit | |
4457 | * again is not run concurrently. | |
4458 | */ | |
4459 | spin_lock(&dest->root_item_lock); | |
a7176f74 | 4460 | if (dest->send_in_progress) { |
f60a2364 MT |
4461 | spin_unlock(&dest->root_item_lock); |
4462 | btrfs_warn(fs_info, | |
4463 | "attempt to delete subvolume %llu during send", | |
4464 | dest->root_key.objectid); | |
4465 | return -EPERM; | |
4466 | } | |
60021bd7 KH |
4467 | if (atomic_read(&dest->nr_swapfiles)) { |
4468 | spin_unlock(&dest->root_item_lock); | |
4469 | btrfs_warn(fs_info, | |
4470 | "attempt to delete subvolume %llu with active swapfile", | |
4471 | root->root_key.objectid); | |
4472 | return -EPERM; | |
4473 | } | |
a7176f74 LF |
4474 | root_flags = btrfs_root_flags(&dest->root_item); |
4475 | btrfs_set_root_flags(&dest->root_item, | |
4476 | root_flags | BTRFS_ROOT_SUBVOL_DEAD); | |
4477 | spin_unlock(&dest->root_item_lock); | |
f60a2364 MT |
4478 | |
4479 | down_write(&fs_info->subvol_sem); | |
4480 | ||
ee0d904f NB |
4481 | ret = may_destroy_subvol(dest); |
4482 | if (ret) | |
f60a2364 MT |
4483 | goto out_up_write; |
4484 | ||
4485 | btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP); | |
4486 | /* | |
4487 | * One for dir inode, | |
4488 | * two for dir entries, | |
4489 | * two for root ref/backref. | |
4490 | */ | |
ee0d904f NB |
4491 | ret = btrfs_subvolume_reserve_metadata(root, &block_rsv, 5, true); |
4492 | if (ret) | |
f60a2364 MT |
4493 | goto out_up_write; |
4494 | ||
4495 | trans = btrfs_start_transaction(root, 0); | |
4496 | if (IS_ERR(trans)) { | |
ee0d904f | 4497 | ret = PTR_ERR(trans); |
f60a2364 MT |
4498 | goto out_release; |
4499 | } | |
4500 | trans->block_rsv = &block_rsv; | |
4501 | trans->bytes_reserved = block_rsv.size; | |
4502 | ||
3c4f91e2 | 4503 | btrfs_record_snapshot_destroy(trans, dir); |
f60a2364 | 4504 | |
045d3967 | 4505 | ret = btrfs_unlink_subvol(trans, dir, dentry); |
f60a2364 | 4506 | if (ret) { |
f60a2364 MT |
4507 | btrfs_abort_transaction(trans, ret); |
4508 | goto out_end_trans; | |
4509 | } | |
4510 | ||
2731f518 JB |
4511 | ret = btrfs_record_root_in_trans(trans, dest); |
4512 | if (ret) { | |
4513 | btrfs_abort_transaction(trans, ret); | |
4514 | goto out_end_trans; | |
4515 | } | |
f60a2364 MT |
4516 | |
4517 | memset(&dest->root_item.drop_progress, 0, | |
4518 | sizeof(dest->root_item.drop_progress)); | |
c8422684 | 4519 | btrfs_set_root_drop_level(&dest->root_item, 0); |
f60a2364 MT |
4520 | btrfs_set_root_refs(&dest->root_item, 0); |
4521 | ||
4522 | if (!test_and_set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &dest->state)) { | |
4523 | ret = btrfs_insert_orphan_item(trans, | |
4524 | fs_info->tree_root, | |
4525 | dest->root_key.objectid); | |
4526 | if (ret) { | |
4527 | btrfs_abort_transaction(trans, ret); | |
f60a2364 MT |
4528 | goto out_end_trans; |
4529 | } | |
4530 | } | |
4531 | ||
d1957791 | 4532 | ret = btrfs_uuid_tree_remove(trans, dest->root_item.uuid, |
f60a2364 MT |
4533 | BTRFS_UUID_KEY_SUBVOL, |
4534 | dest->root_key.objectid); | |
4535 | if (ret && ret != -ENOENT) { | |
4536 | btrfs_abort_transaction(trans, ret); | |
f60a2364 MT |
4537 | goto out_end_trans; |
4538 | } | |
4539 | if (!btrfs_is_empty_uuid(dest->root_item.received_uuid)) { | |
d1957791 | 4540 | ret = btrfs_uuid_tree_remove(trans, |
f60a2364 MT |
4541 | dest->root_item.received_uuid, |
4542 | BTRFS_UUID_KEY_RECEIVED_SUBVOL, | |
4543 | dest->root_key.objectid); | |
4544 | if (ret && ret != -ENOENT) { | |
4545 | btrfs_abort_transaction(trans, ret); | |
f60a2364 MT |
4546 | goto out_end_trans; |
4547 | } | |
4548 | } | |
4549 | ||
082b6c97 QW |
4550 | free_anon_bdev(dest->anon_dev); |
4551 | dest->anon_dev = 0; | |
f60a2364 MT |
4552 | out_end_trans: |
4553 | trans->block_rsv = NULL; | |
4554 | trans->bytes_reserved = 0; | |
4555 | ret = btrfs_end_transaction(trans); | |
f60a2364 MT |
4556 | inode->i_flags |= S_DEAD; |
4557 | out_release: | |
e85fde51 | 4558 | btrfs_subvolume_release_metadata(root, &block_rsv); |
f60a2364 MT |
4559 | out_up_write: |
4560 | up_write(&fs_info->subvol_sem); | |
ee0d904f | 4561 | if (ret) { |
f60a2364 MT |
4562 | spin_lock(&dest->root_item_lock); |
4563 | root_flags = btrfs_root_flags(&dest->root_item); | |
4564 | btrfs_set_root_flags(&dest->root_item, | |
4565 | root_flags & ~BTRFS_ROOT_SUBVOL_DEAD); | |
4566 | spin_unlock(&dest->root_item_lock); | |
4567 | } else { | |
4568 | d_invalidate(dentry); | |
20a68004 | 4569 | btrfs_prune_dentries(dest); |
f60a2364 | 4570 | ASSERT(dest->send_in_progress == 0); |
f60a2364 MT |
4571 | } |
4572 | ||
ee0d904f | 4573 | return ret; |
f60a2364 MT |
4574 | } |
4575 | ||
39279cc3 CM |
4576 | static int btrfs_rmdir(struct inode *dir, struct dentry *dentry) |
4577 | { | |
2b0143b5 | 4578 | struct inode *inode = d_inode(dentry); |
813febdb | 4579 | struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info; |
1832a6d5 | 4580 | int err = 0; |
39279cc3 | 4581 | struct btrfs_trans_handle *trans; |
44f714da | 4582 | u64 last_unlink_trans; |
ab3c5c18 | 4583 | struct fscrypt_name fname; |
39279cc3 | 4584 | |
b3ae244e | 4585 | if (inode->i_size > BTRFS_EMPTY_DIR_SIZE) |
134d4512 | 4586 | return -ENOTEMPTY; |
813febdb JB |
4587 | if (btrfs_ino(BTRFS_I(inode)) == BTRFS_FIRST_FREE_OBJECTID) { |
4588 | if (unlikely(btrfs_fs_incompat(fs_info, EXTENT_TREE_V2))) { | |
4589 | btrfs_err(fs_info, | |
4590 | "extent tree v2 doesn't support snapshot deletion yet"); | |
4591 | return -EOPNOTSUPP; | |
4592 | } | |
3c4f91e2 | 4593 | return btrfs_delete_subvolume(BTRFS_I(dir), dentry); |
813febdb | 4594 | } |
134d4512 | 4595 | |
ab3c5c18 STD |
4596 | err = fscrypt_setup_filename(dir, &dentry->d_name, 1, &fname); |
4597 | if (err) | |
4598 | return err; | |
ab3c5c18 STD |
4599 | |
4600 | /* This needs to handle no-key deletions later on */ | |
4601 | ||
e569b1d5 | 4602 | trans = __unlink_start_trans(BTRFS_I(dir)); |
ab3c5c18 STD |
4603 | if (IS_ERR(trans)) { |
4604 | err = PTR_ERR(trans); | |
4605 | goto out_notrans; | |
4606 | } | |
5df6a9f6 | 4607 | |
4a0cc7ca | 4608 | if (unlikely(btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) { |
5b7544cb | 4609 | err = btrfs_unlink_subvol(trans, BTRFS_I(dir), dentry); |
4df27c4d YZ |
4610 | goto out; |
4611 | } | |
4612 | ||
73f2e545 | 4613 | err = btrfs_orphan_add(trans, BTRFS_I(inode)); |
7b128766 | 4614 | if (err) |
4df27c4d | 4615 | goto out; |
7b128766 | 4616 | |
44f714da FM |
4617 | last_unlink_trans = BTRFS_I(inode)->last_unlink_trans; |
4618 | ||
39279cc3 | 4619 | /* now the directory is empty */ |
e43eec81 | 4620 | err = btrfs_unlink_inode(trans, BTRFS_I(dir), BTRFS_I(d_inode(dentry)), |
6db75318 | 4621 | &fname.disk_name); |
44f714da | 4622 | if (!err) { |
6ef06d27 | 4623 | btrfs_i_size_write(BTRFS_I(inode), 0); |
44f714da FM |
4624 | /* |
4625 | * Propagate the last_unlink_trans value of the deleted dir to | |
4626 | * its parent directory. This is to prevent an unrecoverable | |
4627 | * log tree in the case we do something like this: | |
4628 | * 1) create dir foo | |
4629 | * 2) create snapshot under dir foo | |
4630 | * 3) delete the snapshot | |
4631 | * 4) rmdir foo | |
4632 | * 5) mkdir foo | |
4633 | * 6) fsync foo or some file inside foo | |
4634 | */ | |
4635 | if (last_unlink_trans >= trans->transid) | |
4636 | BTRFS_I(dir)->last_unlink_trans = last_unlink_trans; | |
4637 | } | |
4df27c4d | 4638 | out: |
3a45bb20 | 4639 | btrfs_end_transaction(trans); |
ab3c5c18 | 4640 | out_notrans: |
813febdb | 4641 | btrfs_btree_balance_dirty(fs_info); |
ab3c5c18 | 4642 | fscrypt_free_filename(&fname); |
3954401f | 4643 | |
39279cc3 CM |
4644 | return err; |
4645 | } | |
4646 | ||
39279cc3 | 4647 | /* |
9703fefe | 4648 | * btrfs_truncate_block - read, zero a chunk and write a block |
2aaa6655 JB |
4649 | * @inode - inode that we're zeroing |
4650 | * @from - the offset to start zeroing | |
4651 | * @len - the length to zero, 0 to zero the entire range respective to the | |
4652 | * offset | |
4653 | * @front - zero up to the offset instead of from the offset on | |
4654 | * | |
9703fefe | 4655 | * This will find the block for the "from" offset and cow the block and zero the |
2aaa6655 | 4656 | * part we want to zero. This is used with truncate and hole punching. |
39279cc3 | 4657 | */ |
217f42eb NB |
4658 | int btrfs_truncate_block(struct btrfs_inode *inode, loff_t from, loff_t len, |
4659 | int front) | |
39279cc3 | 4660 | { |
217f42eb NB |
4661 | struct btrfs_fs_info *fs_info = inode->root->fs_info; |
4662 | struct address_space *mapping = inode->vfs_inode.i_mapping; | |
4663 | struct extent_io_tree *io_tree = &inode->io_tree; | |
e6dcd2dc | 4664 | struct btrfs_ordered_extent *ordered; |
2ac55d41 | 4665 | struct extent_state *cached_state = NULL; |
364ecf36 | 4666 | struct extent_changeset *data_reserved = NULL; |
6d4572a9 | 4667 | bool only_release_metadata = false; |
0b246afa | 4668 | u32 blocksize = fs_info->sectorsize; |
09cbfeaf | 4669 | pgoff_t index = from >> PAGE_SHIFT; |
9703fefe | 4670 | unsigned offset = from & (blocksize - 1); |
39279cc3 | 4671 | struct page *page; |
3b16a4e3 | 4672 | gfp_t mask = btrfs_alloc_write_mask(mapping); |
6d4572a9 | 4673 | size_t write_bytes = blocksize; |
39279cc3 | 4674 | int ret = 0; |
9703fefe CR |
4675 | u64 block_start; |
4676 | u64 block_end; | |
39279cc3 | 4677 | |
b03ebd99 NB |
4678 | if (IS_ALIGNED(offset, blocksize) && |
4679 | (!len || IS_ALIGNED(len, blocksize))) | |
39279cc3 | 4680 | goto out; |
9703fefe | 4681 | |
8b62f87b JB |
4682 | block_start = round_down(from, blocksize); |
4683 | block_end = block_start + blocksize - 1; | |
4684 | ||
217f42eb | 4685 | ret = btrfs_check_data_free_space(inode, &data_reserved, block_start, |
1daedb1d | 4686 | blocksize, false); |
6d4572a9 | 4687 | if (ret < 0) { |
80f9d241 | 4688 | if (btrfs_check_nocow_lock(inode, block_start, &write_bytes, false) > 0) { |
6d4572a9 QW |
4689 | /* For nocow case, no need to reserve data space */ |
4690 | only_release_metadata = true; | |
4691 | } else { | |
4692 | goto out; | |
4693 | } | |
4694 | } | |
d4135134 | 4695 | ret = btrfs_delalloc_reserve_metadata(inode, blocksize, blocksize, false); |
6d4572a9 QW |
4696 | if (ret < 0) { |
4697 | if (!only_release_metadata) | |
217f42eb NB |
4698 | btrfs_free_reserved_data_space(inode, data_reserved, |
4699 | block_start, blocksize); | |
6d4572a9 QW |
4700 | goto out; |
4701 | } | |
211c17f5 | 4702 | again: |
3b16a4e3 | 4703 | page = find_or_create_page(mapping, index, mask); |
5d5e103a | 4704 | if (!page) { |
217f42eb NB |
4705 | btrfs_delalloc_release_space(inode, data_reserved, block_start, |
4706 | blocksize, true); | |
4707 | btrfs_delalloc_release_extents(inode, blocksize); | |
ac6a2b36 | 4708 | ret = -ENOMEM; |
39279cc3 | 4709 | goto out; |
5d5e103a | 4710 | } |
e6dcd2dc | 4711 | |
39279cc3 | 4712 | if (!PageUptodate(page)) { |
fb12489b | 4713 | ret = btrfs_read_folio(NULL, page_folio(page)); |
39279cc3 | 4714 | lock_page(page); |
211c17f5 CM |
4715 | if (page->mapping != mapping) { |
4716 | unlock_page(page); | |
09cbfeaf | 4717 | put_page(page); |
211c17f5 CM |
4718 | goto again; |
4719 | } | |
39279cc3 CM |
4720 | if (!PageUptodate(page)) { |
4721 | ret = -EIO; | |
89642229 | 4722 | goto out_unlock; |
39279cc3 CM |
4723 | } |
4724 | } | |
17b17fcd JB |
4725 | |
4726 | /* | |
4727 | * We unlock the page after the io is completed and then re-lock it | |
4728 | * above. release_folio() could have come in between that and cleared | |
4729 | * PagePrivate(), but left the page in the mapping. Set the page mapped | |
4730 | * here to make sure it's properly set for the subpage stuff. | |
4731 | */ | |
4732 | ret = set_page_extent_mapped(page); | |
4733 | if (ret < 0) | |
4734 | goto out_unlock; | |
4735 | ||
211c17f5 | 4736 | wait_on_page_writeback(page); |
e6dcd2dc | 4737 | |
570eb97b | 4738 | lock_extent(io_tree, block_start, block_end, &cached_state); |
e6dcd2dc | 4739 | |
217f42eb | 4740 | ordered = btrfs_lookup_ordered_extent(inode, block_start); |
e6dcd2dc | 4741 | if (ordered) { |
570eb97b | 4742 | unlock_extent(io_tree, block_start, block_end, &cached_state); |
e6dcd2dc | 4743 | unlock_page(page); |
09cbfeaf | 4744 | put_page(page); |
36d45567 | 4745 | btrfs_start_ordered_extent(ordered); |
e6dcd2dc CM |
4746 | btrfs_put_ordered_extent(ordered); |
4747 | goto again; | |
4748 | } | |
4749 | ||
217f42eb | 4750 | clear_extent_bit(&inode->io_tree, block_start, block_end, |
e182163d | 4751 | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, |
bd015294 | 4752 | &cached_state); |
5d5e103a | 4753 | |
217f42eb | 4754 | ret = btrfs_set_extent_delalloc(inode, block_start, block_end, 0, |
330a5827 | 4755 | &cached_state); |
9ed74f2d | 4756 | if (ret) { |
570eb97b | 4757 | unlock_extent(io_tree, block_start, block_end, &cached_state); |
9ed74f2d JB |
4758 | goto out_unlock; |
4759 | } | |
4760 | ||
9703fefe | 4761 | if (offset != blocksize) { |
2aaa6655 | 4762 | if (!len) |
9703fefe | 4763 | len = blocksize - offset; |
2aaa6655 | 4764 | if (front) |
d048b9c2 IW |
4765 | memzero_page(page, (block_start - page_offset(page)), |
4766 | offset); | |
2aaa6655 | 4767 | else |
d048b9c2 IW |
4768 | memzero_page(page, (block_start - page_offset(page)) + offset, |
4769 | len); | |
e6dcd2dc | 4770 | } |
e4f94347 QW |
4771 | btrfs_page_clear_checked(fs_info, page, block_start, |
4772 | block_end + 1 - block_start); | |
6c9ac8be | 4773 | btrfs_page_set_dirty(fs_info, page, block_start, block_end + 1 - block_start); |
570eb97b | 4774 | unlock_extent(io_tree, block_start, block_end, &cached_state); |
39279cc3 | 4775 | |
6d4572a9 | 4776 | if (only_release_metadata) |
217f42eb | 4777 | set_extent_bit(&inode->io_tree, block_start, block_end, |
1d126800 | 4778 | EXTENT_NORESERVE, NULL); |
6d4572a9 | 4779 | |
89642229 | 4780 | out_unlock: |
6d4572a9 QW |
4781 | if (ret) { |
4782 | if (only_release_metadata) | |
217f42eb | 4783 | btrfs_delalloc_release_metadata(inode, blocksize, true); |
6d4572a9 | 4784 | else |
217f42eb | 4785 | btrfs_delalloc_release_space(inode, data_reserved, |
6d4572a9 QW |
4786 | block_start, blocksize, true); |
4787 | } | |
217f42eb | 4788 | btrfs_delalloc_release_extents(inode, blocksize); |
39279cc3 | 4789 | unlock_page(page); |
09cbfeaf | 4790 | put_page(page); |
39279cc3 | 4791 | out: |
6d4572a9 | 4792 | if (only_release_metadata) |
217f42eb | 4793 | btrfs_check_nocow_unlock(inode); |
364ecf36 | 4794 | extent_changeset_free(data_reserved); |
39279cc3 CM |
4795 | return ret; |
4796 | } | |
4797 | ||
a4ba6cc0 | 4798 | static int maybe_insert_hole(struct btrfs_root *root, struct btrfs_inode *inode, |
16e7549f JB |
4799 | u64 offset, u64 len) |
4800 | { | |
a4ba6cc0 | 4801 | struct btrfs_fs_info *fs_info = root->fs_info; |
16e7549f | 4802 | struct btrfs_trans_handle *trans; |
5893dfb9 | 4803 | struct btrfs_drop_extents_args drop_args = { 0 }; |
16e7549f JB |
4804 | int ret; |
4805 | ||
4806 | /* | |
cceaa89f FM |
4807 | * If NO_HOLES is enabled, we don't need to do anything. |
4808 | * Later, up in the call chain, either btrfs_set_inode_last_sub_trans() | |
4809 | * or btrfs_update_inode() will be called, which guarantee that the next | |
4810 | * fsync will know this inode was changed and needs to be logged. | |
16e7549f | 4811 | */ |
cceaa89f | 4812 | if (btrfs_fs_incompat(fs_info, NO_HOLES)) |
16e7549f | 4813 | return 0; |
16e7549f JB |
4814 | |
4815 | /* | |
4816 | * 1 - for the one we're dropping | |
4817 | * 1 - for the one we're adding | |
4818 | * 1 - for updating the inode. | |
4819 | */ | |
4820 | trans = btrfs_start_transaction(root, 3); | |
4821 | if (IS_ERR(trans)) | |
4822 | return PTR_ERR(trans); | |
4823 | ||
5893dfb9 FM |
4824 | drop_args.start = offset; |
4825 | drop_args.end = offset + len; | |
4826 | drop_args.drop_cache = true; | |
4827 | ||
a4ba6cc0 | 4828 | ret = btrfs_drop_extents(trans, root, inode, &drop_args); |
16e7549f | 4829 | if (ret) { |
66642832 | 4830 | btrfs_abort_transaction(trans, ret); |
3a45bb20 | 4831 | btrfs_end_transaction(trans); |
16e7549f JB |
4832 | return ret; |
4833 | } | |
4834 | ||
d1f68ba0 | 4835 | ret = btrfs_insert_hole_extent(trans, root, btrfs_ino(inode), offset, len); |
2766ff61 | 4836 | if (ret) { |
66642832 | 4837 | btrfs_abort_transaction(trans, ret); |
2766ff61 | 4838 | } else { |
a4ba6cc0 NB |
4839 | btrfs_update_inode_bytes(inode, 0, drop_args.bytes_found); |
4840 | btrfs_update_inode(trans, root, inode); | |
2766ff61 | 4841 | } |
3a45bb20 | 4842 | btrfs_end_transaction(trans); |
16e7549f JB |
4843 | return ret; |
4844 | } | |
4845 | ||
695a0d0d JB |
4846 | /* |
4847 | * This function puts in dummy file extents for the area we're creating a hole | |
4848 | * for. So if we are truncating this file to a larger size we need to insert | |
4849 | * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for | |
4850 | * the range between oldsize and size | |
4851 | */ | |
b06359a3 | 4852 | int btrfs_cont_expand(struct btrfs_inode *inode, loff_t oldsize, loff_t size) |
39279cc3 | 4853 | { |
b06359a3 NB |
4854 | struct btrfs_root *root = inode->root; |
4855 | struct btrfs_fs_info *fs_info = root->fs_info; | |
4856 | struct extent_io_tree *io_tree = &inode->io_tree; | |
a22285a6 | 4857 | struct extent_map *em = NULL; |
2ac55d41 | 4858 | struct extent_state *cached_state = NULL; |
0b246afa JM |
4859 | u64 hole_start = ALIGN(oldsize, fs_info->sectorsize); |
4860 | u64 block_end = ALIGN(size, fs_info->sectorsize); | |
9036c102 YZ |
4861 | u64 last_byte; |
4862 | u64 cur_offset; | |
4863 | u64 hole_size; | |
9ed74f2d | 4864 | int err = 0; |
39279cc3 | 4865 | |
a71754fc | 4866 | /* |
9703fefe CR |
4867 | * If our size started in the middle of a block we need to zero out the |
4868 | * rest of the block before we expand the i_size, otherwise we could | |
a71754fc JB |
4869 | * expose stale data. |
4870 | */ | |
b06359a3 | 4871 | err = btrfs_truncate_block(inode, oldsize, 0, 0); |
a71754fc JB |
4872 | if (err) |
4873 | return err; | |
4874 | ||
9036c102 YZ |
4875 | if (size <= hole_start) |
4876 | return 0; | |
4877 | ||
b06359a3 NB |
4878 | btrfs_lock_and_flush_ordered_range(inode, hole_start, block_end - 1, |
4879 | &cached_state); | |
9036c102 YZ |
4880 | cur_offset = hole_start; |
4881 | while (1) { | |
b06359a3 | 4882 | em = btrfs_get_extent(inode, NULL, 0, cur_offset, |
39b07b5d | 4883 | block_end - cur_offset); |
79787eaa JM |
4884 | if (IS_ERR(em)) { |
4885 | err = PTR_ERR(em); | |
f2767956 | 4886 | em = NULL; |
79787eaa JM |
4887 | break; |
4888 | } | |
9036c102 | 4889 | last_byte = min(extent_map_end(em), block_end); |
0b246afa | 4890 | last_byte = ALIGN(last_byte, fs_info->sectorsize); |
9ddc959e JB |
4891 | hole_size = last_byte - cur_offset; |
4892 | ||
8082510e | 4893 | if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) { |
5dc562c5 | 4894 | struct extent_map *hole_em; |
9ed74f2d | 4895 | |
b06359a3 NB |
4896 | err = maybe_insert_hole(root, inode, cur_offset, |
4897 | hole_size); | |
16e7549f | 4898 | if (err) |
3893e33b | 4899 | break; |
9ddc959e | 4900 | |
b06359a3 | 4901 | err = btrfs_inode_set_file_extent_range(inode, |
9ddc959e JB |
4902 | cur_offset, hole_size); |
4903 | if (err) | |
4904 | break; | |
4905 | ||
5dc562c5 JB |
4906 | hole_em = alloc_extent_map(); |
4907 | if (!hole_em) { | |
a1ba4c08 FM |
4908 | btrfs_drop_extent_map_range(inode, cur_offset, |
4909 | cur_offset + hole_size - 1, | |
4910 | false); | |
23e3337f | 4911 | btrfs_set_inode_full_sync(inode); |
5dc562c5 JB |
4912 | goto next; |
4913 | } | |
4914 | hole_em->start = cur_offset; | |
4915 | hole_em->len = hole_size; | |
4916 | hole_em->orig_start = cur_offset; | |
8082510e | 4917 | |
5dc562c5 JB |
4918 | hole_em->block_start = EXTENT_MAP_HOLE; |
4919 | hole_em->block_len = 0; | |
b4939680 | 4920 | hole_em->orig_block_len = 0; |
cc95bef6 | 4921 | hole_em->ram_bytes = hole_size; |
5dc562c5 | 4922 | hole_em->compress_type = BTRFS_COMPRESS_NONE; |
0b246afa | 4923 | hole_em->generation = fs_info->generation; |
8082510e | 4924 | |
a1ba4c08 | 4925 | err = btrfs_replace_extent_map_range(inode, hole_em, true); |
5dc562c5 | 4926 | free_extent_map(hole_em); |
9ddc959e | 4927 | } else { |
b06359a3 | 4928 | err = btrfs_inode_set_file_extent_range(inode, |
9ddc959e JB |
4929 | cur_offset, hole_size); |
4930 | if (err) | |
4931 | break; | |
9036c102 | 4932 | } |
16e7549f | 4933 | next: |
9036c102 | 4934 | free_extent_map(em); |
a22285a6 | 4935 | em = NULL; |
9036c102 | 4936 | cur_offset = last_byte; |
8082510e | 4937 | if (cur_offset >= block_end) |
9036c102 YZ |
4938 | break; |
4939 | } | |
a22285a6 | 4940 | free_extent_map(em); |
570eb97b | 4941 | unlock_extent(io_tree, hole_start, block_end - 1, &cached_state); |
9036c102 YZ |
4942 | return err; |
4943 | } | |
39279cc3 | 4944 | |
3972f260 | 4945 | static int btrfs_setsize(struct inode *inode, struct iattr *attr) |
8082510e | 4946 | { |
f4a2f4c5 MX |
4947 | struct btrfs_root *root = BTRFS_I(inode)->root; |
4948 | struct btrfs_trans_handle *trans; | |
a41ad394 | 4949 | loff_t oldsize = i_size_read(inode); |
3972f260 ES |
4950 | loff_t newsize = attr->ia_size; |
4951 | int mask = attr->ia_valid; | |
8082510e YZ |
4952 | int ret; |
4953 | ||
3972f260 ES |
4954 | /* |
4955 | * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a | |
4956 | * special case where we need to update the times despite not having | |
4957 | * these flags set. For all other operations the VFS set these flags | |
4958 | * explicitly if it wants a timestamp update. | |
4959 | */ | |
dff6efc3 CH |
4960 | if (newsize != oldsize) { |
4961 | inode_inc_iversion(inode); | |
c1867eb3 | 4962 | if (!(mask & (ATTR_CTIME | ATTR_MTIME))) { |
2a9462de | 4963 | inode->i_mtime = inode_set_ctime_current(inode); |
c1867eb3 | 4964 | } |
dff6efc3 | 4965 | } |
3972f260 | 4966 | |
a41ad394 | 4967 | if (newsize > oldsize) { |
9ea24bbe | 4968 | /* |
ea14b57f | 4969 | * Don't do an expanding truncate while snapshotting is ongoing. |
9ea24bbe FM |
4970 | * This is to ensure the snapshot captures a fully consistent |
4971 | * state of this file - if the snapshot captures this expanding | |
4972 | * truncation, it must capture all writes that happened before | |
4973 | * this truncation. | |
4974 | */ | |
dcc3eb96 | 4975 | btrfs_drew_write_lock(&root->snapshot_lock); |
b06359a3 | 4976 | ret = btrfs_cont_expand(BTRFS_I(inode), oldsize, newsize); |
9ea24bbe | 4977 | if (ret) { |
dcc3eb96 | 4978 | btrfs_drew_write_unlock(&root->snapshot_lock); |
8082510e | 4979 | return ret; |
9ea24bbe | 4980 | } |
8082510e | 4981 | |
f4a2f4c5 | 4982 | trans = btrfs_start_transaction(root, 1); |
9ea24bbe | 4983 | if (IS_ERR(trans)) { |
dcc3eb96 | 4984 | btrfs_drew_write_unlock(&root->snapshot_lock); |
f4a2f4c5 | 4985 | return PTR_ERR(trans); |
9ea24bbe | 4986 | } |
f4a2f4c5 MX |
4987 | |
4988 | i_size_write(inode, newsize); | |
76aea537 | 4989 | btrfs_inode_safe_disk_i_size_write(BTRFS_I(inode), 0); |
27772b68 | 4990 | pagecache_isize_extended(inode, oldsize, newsize); |
9a56fcd1 | 4991 | ret = btrfs_update_inode(trans, root, BTRFS_I(inode)); |
dcc3eb96 | 4992 | btrfs_drew_write_unlock(&root->snapshot_lock); |
3a45bb20 | 4993 | btrfs_end_transaction(trans); |
a41ad394 | 4994 | } else { |
24c0a722 NA |
4995 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
4996 | ||
4997 | if (btrfs_is_zoned(fs_info)) { | |
4998 | ret = btrfs_wait_ordered_range(inode, | |
4999 | ALIGN(newsize, fs_info->sectorsize), | |
5000 | (u64)-1); | |
5001 | if (ret) | |
5002 | return ret; | |
5003 | } | |
8082510e | 5004 | |
a41ad394 JB |
5005 | /* |
5006 | * We're truncating a file that used to have good data down to | |
1fd4033d NB |
5007 | * zero. Make sure any new writes to the file get on disk |
5008 | * on close. | |
a41ad394 JB |
5009 | */ |
5010 | if (newsize == 0) | |
1fd4033d | 5011 | set_bit(BTRFS_INODE_FLUSH_ON_CLOSE, |
72ac3c0d | 5012 | &BTRFS_I(inode)->runtime_flags); |
8082510e | 5013 | |
a41ad394 | 5014 | truncate_setsize(inode, newsize); |
2e60a51e | 5015 | |
2e60a51e | 5016 | inode_dio_wait(inode); |
2e60a51e | 5017 | |
d9dcae67 | 5018 | ret = btrfs_truncate(BTRFS_I(inode), newsize == oldsize); |
7f4f6e0a JB |
5019 | if (ret && inode->i_nlink) { |
5020 | int err; | |
5021 | ||
5022 | /* | |
f7e9e8fc OS |
5023 | * Truncate failed, so fix up the in-memory size. We |
5024 | * adjusted disk_i_size down as we removed extents, so | |
5025 | * wait for disk_i_size to be stable and then update the | |
5026 | * in-memory size to match. | |
7f4f6e0a | 5027 | */ |
f7e9e8fc | 5028 | err = btrfs_wait_ordered_range(inode, 0, (u64)-1); |
7f4f6e0a | 5029 | if (err) |
f7e9e8fc OS |
5030 | return err; |
5031 | i_size_write(inode, BTRFS_I(inode)->disk_i_size); | |
7f4f6e0a | 5032 | } |
8082510e YZ |
5033 | } |
5034 | ||
a41ad394 | 5035 | return ret; |
8082510e YZ |
5036 | } |
5037 | ||
c1632a0f | 5038 | static int btrfs_setattr(struct mnt_idmap *idmap, struct dentry *dentry, |
549c7297 | 5039 | struct iattr *attr) |
9036c102 | 5040 | { |
2b0143b5 | 5041 | struct inode *inode = d_inode(dentry); |
b83cc969 | 5042 | struct btrfs_root *root = BTRFS_I(inode)->root; |
9036c102 | 5043 | int err; |
39279cc3 | 5044 | |
b83cc969 LZ |
5045 | if (btrfs_root_readonly(root)) |
5046 | return -EROFS; | |
5047 | ||
c1632a0f | 5048 | err = setattr_prepare(idmap, dentry, attr); |
9036c102 YZ |
5049 | if (err) |
5050 | return err; | |
2bf5a725 | 5051 | |
5a3f23d5 | 5052 | if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) { |
3972f260 | 5053 | err = btrfs_setsize(inode, attr); |
8082510e YZ |
5054 | if (err) |
5055 | return err; | |
39279cc3 | 5056 | } |
9036c102 | 5057 | |
1025774c | 5058 | if (attr->ia_valid) { |
c1632a0f | 5059 | setattr_copy(idmap, inode, attr); |
0c4d2d95 | 5060 | inode_inc_iversion(inode); |
7152b425 | 5061 | err = btrfs_dirty_inode(BTRFS_I(inode)); |
1025774c | 5062 | |
22c44fe6 | 5063 | if (!err && attr->ia_valid & ATTR_MODE) |
13e83a49 | 5064 | err = posix_acl_chmod(idmap, dentry, inode->i_mode); |
1025774c | 5065 | } |
33268eaf | 5066 | |
39279cc3 CM |
5067 | return err; |
5068 | } | |
61295eb8 | 5069 | |
131e404a | 5070 | /* |
895586eb MWO |
5071 | * While truncating the inode pages during eviction, we get the VFS |
5072 | * calling btrfs_invalidate_folio() against each folio of the inode. This | |
5073 | * is slow because the calls to btrfs_invalidate_folio() result in a | |
570eb97b | 5074 | * huge amount of calls to lock_extent() and clear_extent_bit(), |
895586eb MWO |
5075 | * which keep merging and splitting extent_state structures over and over, |
5076 | * wasting lots of time. | |
131e404a | 5077 | * |
895586eb MWO |
5078 | * Therefore if the inode is being evicted, let btrfs_invalidate_folio() |
5079 | * skip all those expensive operations on a per folio basis and do only | |
5080 | * the ordered io finishing, while we release here the extent_map and | |
5081 | * extent_state structures, without the excessive merging and splitting. | |
131e404a FDBM |
5082 | */ |
5083 | static void evict_inode_truncate_pages(struct inode *inode) | |
5084 | { | |
5085 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; | |
131e404a FDBM |
5086 | struct rb_node *node; |
5087 | ||
5088 | ASSERT(inode->i_state & I_FREEING); | |
91b0abe3 | 5089 | truncate_inode_pages_final(&inode->i_data); |
131e404a | 5090 | |
9c9d1b4f | 5091 | btrfs_drop_extent_map_range(BTRFS_I(inode), 0, (u64)-1, false); |
131e404a | 5092 | |
6ca07097 FM |
5093 | /* |
5094 | * Keep looping until we have no more ranges in the io tree. | |
ba206a02 MWO |
5095 | * We can have ongoing bios started by readahead that have |
5096 | * their endio callback (extent_io.c:end_bio_extent_readpage) | |
9c6429d9 FM |
5097 | * still in progress (unlocked the pages in the bio but did not yet |
5098 | * unlocked the ranges in the io tree). Therefore this means some | |
6ca07097 FM |
5099 | * ranges can still be locked and eviction started because before |
5100 | * submitting those bios, which are executed by a separate task (work | |
5101 | * queue kthread), inode references (inode->i_count) were not taken | |
5102 | * (which would be dropped in the end io callback of each bio). | |
5103 | * Therefore here we effectively end up waiting for those bios and | |
5104 | * anyone else holding locked ranges without having bumped the inode's | |
5105 | * reference count - if we don't do it, when they access the inode's | |
5106 | * io_tree to unlock a range it may be too late, leading to an | |
5107 | * use-after-free issue. | |
5108 | */ | |
131e404a FDBM |
5109 | spin_lock(&io_tree->lock); |
5110 | while (!RB_EMPTY_ROOT(&io_tree->state)) { | |
5111 | struct extent_state *state; | |
5112 | struct extent_state *cached_state = NULL; | |
6ca07097 FM |
5113 | u64 start; |
5114 | u64 end; | |
421f0922 | 5115 | unsigned state_flags; |
131e404a FDBM |
5116 | |
5117 | node = rb_first(&io_tree->state); | |
5118 | state = rb_entry(node, struct extent_state, rb_node); | |
6ca07097 FM |
5119 | start = state->start; |
5120 | end = state->end; | |
421f0922 | 5121 | state_flags = state->state; |
131e404a FDBM |
5122 | spin_unlock(&io_tree->lock); |
5123 | ||
570eb97b | 5124 | lock_extent(io_tree, start, end, &cached_state); |
b9d0b389 QW |
5125 | |
5126 | /* | |
5127 | * If still has DELALLOC flag, the extent didn't reach disk, | |
5128 | * and its reserved space won't be freed by delayed_ref. | |
5129 | * So we need to free its reserved space here. | |
895586eb | 5130 | * (Refer to comment in btrfs_invalidate_folio, case 2) |
b9d0b389 QW |
5131 | * |
5132 | * Note, end is the bytenr of last byte, so we need + 1 here. | |
5133 | */ | |
421f0922 | 5134 | if (state_flags & EXTENT_DELALLOC) |
8b8a979f NB |
5135 | btrfs_qgroup_free_data(BTRFS_I(inode), NULL, start, |
5136 | end - start + 1); | |
b9d0b389 | 5137 | |
6ca07097 | 5138 | clear_extent_bit(io_tree, start, end, |
bd015294 | 5139 | EXTENT_CLEAR_ALL_BITS | EXTENT_DO_ACCOUNTING, |
e182163d | 5140 | &cached_state); |
131e404a | 5141 | |
7064dd5c | 5142 | cond_resched(); |
131e404a FDBM |
5143 | spin_lock(&io_tree->lock); |
5144 | } | |
5145 | spin_unlock(&io_tree->lock); | |
5146 | } | |
5147 | ||
4b9d7b59 | 5148 | static struct btrfs_trans_handle *evict_refill_and_join(struct btrfs_root *root, |
ad80cf50 | 5149 | struct btrfs_block_rsv *rsv) |
4b9d7b59 OS |
5150 | { |
5151 | struct btrfs_fs_info *fs_info = root->fs_info; | |
d3984c90 | 5152 | struct btrfs_trans_handle *trans; |
b13d57db | 5153 | u64 delayed_refs_extra = btrfs_calc_delayed_ref_bytes(fs_info, 1); |
d3984c90 | 5154 | int ret; |
4b9d7b59 | 5155 | |
d3984c90 JB |
5156 | /* |
5157 | * Eviction should be taking place at some place safe because of our | |
5158 | * delayed iputs. However the normal flushing code will run delayed | |
5159 | * iputs, so we cannot use FLUSH_ALL otherwise we'll deadlock. | |
5160 | * | |
5161 | * We reserve the delayed_refs_extra here again because we can't use | |
5162 | * btrfs_start_transaction(root, 0) for the same deadlocky reason as | |
5163 | * above. We reserve our extra bit here because we generate a ton of | |
5164 | * delayed refs activity by truncating. | |
5165 | * | |
ee6adbfd JB |
5166 | * BTRFS_RESERVE_FLUSH_EVICT will steal from the global_rsv if it can, |
5167 | * if we fail to make this reservation we can re-try without the | |
5168 | * delayed_refs_extra so we can make some forward progress. | |
d3984c90 | 5169 | */ |
9270501c | 5170 | ret = btrfs_block_rsv_refill(fs_info, rsv, rsv->size + delayed_refs_extra, |
d3984c90 JB |
5171 | BTRFS_RESERVE_FLUSH_EVICT); |
5172 | if (ret) { | |
9270501c | 5173 | ret = btrfs_block_rsv_refill(fs_info, rsv, rsv->size, |
ee6adbfd JB |
5174 | BTRFS_RESERVE_FLUSH_EVICT); |
5175 | if (ret) { | |
d3984c90 JB |
5176 | btrfs_warn(fs_info, |
5177 | "could not allocate space for delete; will truncate on mount"); | |
5178 | return ERR_PTR(-ENOSPC); | |
5179 | } | |
5180 | delayed_refs_extra = 0; | |
5181 | } | |
4b9d7b59 | 5182 | |
d3984c90 JB |
5183 | trans = btrfs_join_transaction(root); |
5184 | if (IS_ERR(trans)) | |
5185 | return trans; | |
5186 | ||
5187 | if (delayed_refs_extra) { | |
5188 | trans->block_rsv = &fs_info->trans_block_rsv; | |
5189 | trans->bytes_reserved = delayed_refs_extra; | |
5190 | btrfs_block_rsv_migrate(rsv, trans->block_rsv, | |
4e0527de | 5191 | delayed_refs_extra, true); |
4b9d7b59 | 5192 | } |
d3984c90 | 5193 | return trans; |
4b9d7b59 OS |
5194 | } |
5195 | ||
bd555975 | 5196 | void btrfs_evict_inode(struct inode *inode) |
39279cc3 | 5197 | { |
0b246afa | 5198 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
39279cc3 CM |
5199 | struct btrfs_trans_handle *trans; |
5200 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
b7b1167c | 5201 | struct btrfs_block_rsv *rsv = NULL; |
39279cc3 CM |
5202 | int ret; |
5203 | ||
1abe9b8a | 5204 | trace_btrfs_inode_evict(inode); |
5205 | ||
3d48d981 | 5206 | if (!root) { |
14605409 | 5207 | fsverity_cleanup_inode(inode); |
e8f1bc14 | 5208 | clear_inode(inode); |
3d48d981 NB |
5209 | return; |
5210 | } | |
5211 | ||
131e404a FDBM |
5212 | evict_inode_truncate_pages(inode); |
5213 | ||
69e9c6c6 SB |
5214 | if (inode->i_nlink && |
5215 | ((btrfs_root_refs(&root->root_item) != 0 && | |
5216 | root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) || | |
70ddc553 | 5217 | btrfs_is_free_space_inode(BTRFS_I(inode)))) |
b7b1167c | 5218 | goto out; |
bd555975 | 5219 | |
27919067 | 5220 | if (is_bad_inode(inode)) |
b7b1167c | 5221 | goto out; |
5f39d397 | 5222 | |
7b40b695 | 5223 | if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) |
b7b1167c | 5224 | goto out; |
c71bf099 | 5225 | |
76dda93c | 5226 | if (inode->i_nlink > 0) { |
69e9c6c6 SB |
5227 | BUG_ON(btrfs_root_refs(&root->root_item) != 0 && |
5228 | root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID); | |
b7b1167c | 5229 | goto out; |
76dda93c YZ |
5230 | } |
5231 | ||
2adc75d6 JB |
5232 | /* |
5233 | * This makes sure the inode item in tree is uptodate and the space for | |
5234 | * the inode update is released. | |
5235 | */ | |
aa79021f | 5236 | ret = btrfs_commit_inode_delayed_inode(BTRFS_I(inode)); |
27919067 | 5237 | if (ret) |
b7b1167c | 5238 | goto out; |
0e8c36a9 | 5239 | |
2adc75d6 JB |
5240 | /* |
5241 | * This drops any pending insert or delete operations we have for this | |
5242 | * inode. We could have a delayed dir index deletion queued up, but | |
5243 | * we're removing the inode completely so that'll be taken care of in | |
5244 | * the truncate. | |
5245 | */ | |
5246 | btrfs_kill_delayed_inode_items(BTRFS_I(inode)); | |
5247 | ||
2ff7e61e | 5248 | rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP); |
27919067 | 5249 | if (!rsv) |
b7b1167c | 5250 | goto out; |
2bd36e7b | 5251 | rsv->size = btrfs_calc_metadata_size(fs_info, 1); |
710d5921 | 5252 | rsv->failfast = true; |
4289a667 | 5253 | |
6ef06d27 | 5254 | btrfs_i_size_write(BTRFS_I(inode), 0); |
5f39d397 | 5255 | |
8082510e | 5256 | while (1) { |
d9ac19c3 | 5257 | struct btrfs_truncate_control control = { |
71d18b53 | 5258 | .inode = BTRFS_I(inode), |
487e81d2 | 5259 | .ino = btrfs_ino(BTRFS_I(inode)), |
d9ac19c3 JB |
5260 | .new_size = 0, |
5261 | .min_type = 0, | |
5262 | }; | |
5263 | ||
ad80cf50 | 5264 | trans = evict_refill_and_join(root, rsv); |
27919067 | 5265 | if (IS_ERR(trans)) |
b7b1167c | 5266 | goto out; |
7b128766 | 5267 | |
4289a667 JB |
5268 | trans->block_rsv = rsv; |
5269 | ||
71d18b53 | 5270 | ret = btrfs_truncate_inode_items(trans, root, &control); |
27919067 OS |
5271 | trans->block_rsv = &fs_info->trans_block_rsv; |
5272 | btrfs_end_transaction(trans); | |
afa4b0af FM |
5273 | /* |
5274 | * We have not added new delayed items for our inode after we | |
5275 | * have flushed its delayed items, so no need to throttle on | |
5276 | * delayed items. However we have modified extent buffers. | |
5277 | */ | |
5278 | btrfs_btree_balance_dirty_nodelay(fs_info); | |
27919067 | 5279 | if (ret && ret != -ENOSPC && ret != -EAGAIN) |
b7b1167c | 5280 | goto out; |
27919067 | 5281 | else if (!ret) |
8082510e | 5282 | break; |
8082510e | 5283 | } |
5f39d397 | 5284 | |
4ef31a45 | 5285 | /* |
27919067 OS |
5286 | * Errors here aren't a big deal, it just means we leave orphan items in |
5287 | * the tree. They will be cleaned up on the next mount. If the inode | |
5288 | * number gets reused, cleanup deletes the orphan item without doing | |
5289 | * anything, and unlink reuses the existing orphan item. | |
5290 | * | |
5291 | * If it turns out that we are dropping too many of these, we might want | |
5292 | * to add a mechanism for retrying these after a commit. | |
4ef31a45 | 5293 | */ |
ad80cf50 | 5294 | trans = evict_refill_and_join(root, rsv); |
27919067 OS |
5295 | if (!IS_ERR(trans)) { |
5296 | trans->block_rsv = rsv; | |
5297 | btrfs_orphan_del(trans, BTRFS_I(inode)); | |
5298 | trans->block_rsv = &fs_info->trans_block_rsv; | |
5299 | btrfs_end_transaction(trans); | |
5300 | } | |
54aa1f4d | 5301 | |
b7b1167c | 5302 | out: |
27919067 | 5303 | btrfs_free_block_rsv(fs_info, rsv); |
27919067 OS |
5304 | /* |
5305 | * If we didn't successfully delete, the orphan item will still be in | |
5306 | * the tree and we'll retry on the next mount. Again, we might also want | |
5307 | * to retry these periodically in the future. | |
5308 | */ | |
f48d1cf5 | 5309 | btrfs_remove_delayed_node(BTRFS_I(inode)); |
14605409 | 5310 | fsverity_cleanup_inode(inode); |
dbd5768f | 5311 | clear_inode(inode); |
39279cc3 CM |
5312 | } |
5313 | ||
5314 | /* | |
6bf9e4bd QW |
5315 | * Return the key found in the dir entry in the location pointer, fill @type |
5316 | * with BTRFS_FT_*, and return 0. | |
5317 | * | |
005d6712 SY |
5318 | * If no dir entries were found, returns -ENOENT. |
5319 | * If found a corrupted location in dir entry, returns -EUCLEAN. | |
39279cc3 | 5320 | */ |
d1de429b | 5321 | static int btrfs_inode_by_name(struct btrfs_inode *dir, struct dentry *dentry, |
6bf9e4bd | 5322 | struct btrfs_key *location, u8 *type) |
39279cc3 | 5323 | { |
39279cc3 CM |
5324 | struct btrfs_dir_item *di; |
5325 | struct btrfs_path *path; | |
d1de429b | 5326 | struct btrfs_root *root = dir->root; |
0d9f7f3e | 5327 | int ret = 0; |
ab3c5c18 | 5328 | struct fscrypt_name fname; |
39279cc3 CM |
5329 | |
5330 | path = btrfs_alloc_path(); | |
d8926bb3 MF |
5331 | if (!path) |
5332 | return -ENOMEM; | |
3954401f | 5333 | |
d1de429b | 5334 | ret = fscrypt_setup_filename(&dir->vfs_inode, &dentry->d_name, 1, &fname); |
10a8857a | 5335 | if (ret < 0) |
ab3c5c18 | 5336 | goto out; |
10a8857a STD |
5337 | /* |
5338 | * fscrypt_setup_filename() should never return a positive value, but | |
5339 | * gcc on sparc/parisc thinks it can, so assert that doesn't happen. | |
5340 | */ | |
5341 | ASSERT(ret == 0); | |
ab3c5c18 | 5342 | |
ab3c5c18 STD |
5343 | /* This needs to handle no-key deletions later on */ |
5344 | ||
d1de429b | 5345 | di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), |
6db75318 | 5346 | &fname.disk_name, 0); |
3cf5068f LB |
5347 | if (IS_ERR_OR_NULL(di)) { |
5348 | ret = di ? PTR_ERR(di) : -ENOENT; | |
005d6712 SY |
5349 | goto out; |
5350 | } | |
d397712b | 5351 | |
5f39d397 | 5352 | btrfs_dir_item_key_to_cpu(path->nodes[0], di, location); |
56a0e706 LB |
5353 | if (location->type != BTRFS_INODE_ITEM_KEY && |
5354 | location->type != BTRFS_ROOT_ITEM_KEY) { | |
005d6712 | 5355 | ret = -EUCLEAN; |
56a0e706 LB |
5356 | btrfs_warn(root->fs_info, |
5357 | "%s gets something invalid in DIR_ITEM (name %s, directory ino %llu, location(%llu %u %llu))", | |
d1de429b | 5358 | __func__, fname.disk_name.name, btrfs_ino(dir), |
56a0e706 | 5359 | location->objectid, location->type, location->offset); |
56a0e706 | 5360 | } |
6bf9e4bd | 5361 | if (!ret) |
94a48aef | 5362 | *type = btrfs_dir_ftype(path->nodes[0], di); |
39279cc3 | 5363 | out: |
ab3c5c18 | 5364 | fscrypt_free_filename(&fname); |
39279cc3 CM |
5365 | btrfs_free_path(path); |
5366 | return ret; | |
5367 | } | |
5368 | ||
5369 | /* | |
5370 | * when we hit a tree root in a directory, the btrfs part of the inode | |
5371 | * needs to be changed to reflect the root directory of the tree root. This | |
5372 | * is kind of like crossing a mount point. | |
5373 | */ | |
2ff7e61e | 5374 | static int fixup_tree_root_location(struct btrfs_fs_info *fs_info, |
3c1b1c4c | 5375 | struct btrfs_inode *dir, |
4df27c4d YZ |
5376 | struct dentry *dentry, |
5377 | struct btrfs_key *location, | |
5378 | struct btrfs_root **sub_root) | |
39279cc3 | 5379 | { |
4df27c4d YZ |
5380 | struct btrfs_path *path; |
5381 | struct btrfs_root *new_root; | |
5382 | struct btrfs_root_ref *ref; | |
5383 | struct extent_buffer *leaf; | |
1d4c08e0 | 5384 | struct btrfs_key key; |
4df27c4d YZ |
5385 | int ret; |
5386 | int err = 0; | |
ab3c5c18 | 5387 | struct fscrypt_name fname; |
ab3c5c18 | 5388 | |
3c1b1c4c | 5389 | ret = fscrypt_setup_filename(&dir->vfs_inode, &dentry->d_name, 0, &fname); |
ab3c5c18 STD |
5390 | if (ret) |
5391 | return ret; | |
39279cc3 | 5392 | |
4df27c4d YZ |
5393 | path = btrfs_alloc_path(); |
5394 | if (!path) { | |
5395 | err = -ENOMEM; | |
5396 | goto out; | |
5397 | } | |
39279cc3 | 5398 | |
4df27c4d | 5399 | err = -ENOENT; |
3c1b1c4c | 5400 | key.objectid = dir->root->root_key.objectid; |
1d4c08e0 DS |
5401 | key.type = BTRFS_ROOT_REF_KEY; |
5402 | key.offset = location->objectid; | |
5403 | ||
0b246afa | 5404 | ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0); |
4df27c4d YZ |
5405 | if (ret) { |
5406 | if (ret < 0) | |
5407 | err = ret; | |
5408 | goto out; | |
5409 | } | |
39279cc3 | 5410 | |
4df27c4d YZ |
5411 | leaf = path->nodes[0]; |
5412 | ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref); | |
3c1b1c4c | 5413 | if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) || |
6db75318 | 5414 | btrfs_root_ref_name_len(leaf, ref) != fname.disk_name.len) |
4df27c4d | 5415 | goto out; |
39279cc3 | 5416 | |
6db75318 STD |
5417 | ret = memcmp_extent_buffer(leaf, fname.disk_name.name, |
5418 | (unsigned long)(ref + 1), fname.disk_name.len); | |
4df27c4d YZ |
5419 | if (ret) |
5420 | goto out; | |
5421 | ||
b3b4aa74 | 5422 | btrfs_release_path(path); |
4df27c4d | 5423 | |
56e9357a | 5424 | new_root = btrfs_get_fs_root(fs_info, location->objectid, true); |
4df27c4d YZ |
5425 | if (IS_ERR(new_root)) { |
5426 | err = PTR_ERR(new_root); | |
5427 | goto out; | |
5428 | } | |
5429 | ||
4df27c4d YZ |
5430 | *sub_root = new_root; |
5431 | location->objectid = btrfs_root_dirid(&new_root->root_item); | |
5432 | location->type = BTRFS_INODE_ITEM_KEY; | |
5433 | location->offset = 0; | |
5434 | err = 0; | |
5435 | out: | |
5436 | btrfs_free_path(path); | |
ab3c5c18 | 5437 | fscrypt_free_filename(&fname); |
4df27c4d | 5438 | return err; |
39279cc3 CM |
5439 | } |
5440 | ||
4c45a4f4 | 5441 | static void inode_tree_add(struct btrfs_inode *inode) |
5d4f98a2 | 5442 | { |
4c45a4f4 | 5443 | struct btrfs_root *root = inode->root; |
5d4f98a2 | 5444 | struct btrfs_inode *entry; |
03e860bd FNP |
5445 | struct rb_node **p; |
5446 | struct rb_node *parent; | |
4c45a4f4 DS |
5447 | struct rb_node *new = &inode->rb_node; |
5448 | u64 ino = btrfs_ino(inode); | |
5d4f98a2 | 5449 | |
4c45a4f4 | 5450 | if (inode_unhashed(&inode->vfs_inode)) |
76dda93c | 5451 | return; |
e1409cef | 5452 | parent = NULL; |
5d4f98a2 | 5453 | spin_lock(&root->inode_lock); |
e1409cef | 5454 | p = &root->inode_tree.rb_node; |
5d4f98a2 YZ |
5455 | while (*p) { |
5456 | parent = *p; | |
5457 | entry = rb_entry(parent, struct btrfs_inode, rb_node); | |
5458 | ||
37508515 | 5459 | if (ino < btrfs_ino(entry)) |
03e860bd | 5460 | p = &parent->rb_left; |
37508515 | 5461 | else if (ino > btrfs_ino(entry)) |
03e860bd | 5462 | p = &parent->rb_right; |
5d4f98a2 YZ |
5463 | else { |
5464 | WARN_ON(!(entry->vfs_inode.i_state & | |
a4ffdde6 | 5465 | (I_WILL_FREE | I_FREEING))); |
cef21937 | 5466 | rb_replace_node(parent, new, &root->inode_tree); |
03e860bd FNP |
5467 | RB_CLEAR_NODE(parent); |
5468 | spin_unlock(&root->inode_lock); | |
cef21937 | 5469 | return; |
5d4f98a2 YZ |
5470 | } |
5471 | } | |
cef21937 FDBM |
5472 | rb_link_node(new, parent, p); |
5473 | rb_insert_color(new, &root->inode_tree); | |
5d4f98a2 YZ |
5474 | spin_unlock(&root->inode_lock); |
5475 | } | |
5476 | ||
b79b7249 | 5477 | static void inode_tree_del(struct btrfs_inode *inode) |
5d4f98a2 | 5478 | { |
b79b7249 | 5479 | struct btrfs_root *root = inode->root; |
76dda93c | 5480 | int empty = 0; |
5d4f98a2 | 5481 | |
03e860bd | 5482 | spin_lock(&root->inode_lock); |
b79b7249 NB |
5483 | if (!RB_EMPTY_NODE(&inode->rb_node)) { |
5484 | rb_erase(&inode->rb_node, &root->inode_tree); | |
5485 | RB_CLEAR_NODE(&inode->rb_node); | |
76dda93c | 5486 | empty = RB_EMPTY_ROOT(&root->inode_tree); |
5d4f98a2 | 5487 | } |
03e860bd | 5488 | spin_unlock(&root->inode_lock); |
76dda93c | 5489 | |
69e9c6c6 | 5490 | if (empty && btrfs_root_refs(&root->root_item) == 0) { |
76dda93c YZ |
5491 | spin_lock(&root->inode_lock); |
5492 | empty = RB_EMPTY_ROOT(&root->inode_tree); | |
5493 | spin_unlock(&root->inode_lock); | |
5494 | if (empty) | |
5495 | btrfs_add_dead_root(root); | |
5496 | } | |
5497 | } | |
5498 | ||
5d4f98a2 | 5499 | |
e02119d5 CM |
5500 | static int btrfs_init_locked_inode(struct inode *inode, void *p) |
5501 | { | |
5502 | struct btrfs_iget_args *args = p; | |
0202e83f DS |
5503 | |
5504 | inode->i_ino = args->ino; | |
5505 | BTRFS_I(inode)->location.objectid = args->ino; | |
5506 | BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY; | |
5507 | BTRFS_I(inode)->location.offset = 0; | |
5c8fd99f JB |
5508 | BTRFS_I(inode)->root = btrfs_grab_root(args->root); |
5509 | BUG_ON(args->root && !BTRFS_I(inode)->root); | |
9b9b8854 JB |
5510 | |
5511 | if (args->root && args->root == args->root->fs_info->tree_root && | |
5512 | args->ino != BTRFS_BTREE_INODE_OBJECTID) | |
5513 | set_bit(BTRFS_INODE_FREE_SPACE_INODE, | |
5514 | &BTRFS_I(inode)->runtime_flags); | |
39279cc3 CM |
5515 | return 0; |
5516 | } | |
5517 | ||
5518 | static int btrfs_find_actor(struct inode *inode, void *opaque) | |
5519 | { | |
5520 | struct btrfs_iget_args *args = opaque; | |
0202e83f DS |
5521 | |
5522 | return args->ino == BTRFS_I(inode)->location.objectid && | |
d397712b | 5523 | args->root == BTRFS_I(inode)->root; |
39279cc3 CM |
5524 | } |
5525 | ||
0202e83f | 5526 | static struct inode *btrfs_iget_locked(struct super_block *s, u64 ino, |
5d4f98a2 | 5527 | struct btrfs_root *root) |
39279cc3 CM |
5528 | { |
5529 | struct inode *inode; | |
5530 | struct btrfs_iget_args args; | |
0202e83f | 5531 | unsigned long hashval = btrfs_inode_hash(ino, root); |
778ba82b | 5532 | |
0202e83f | 5533 | args.ino = ino; |
39279cc3 CM |
5534 | args.root = root; |
5535 | ||
778ba82b | 5536 | inode = iget5_locked(s, hashval, btrfs_find_actor, |
39279cc3 CM |
5537 | btrfs_init_locked_inode, |
5538 | (void *)&args); | |
5539 | return inode; | |
5540 | } | |
5541 | ||
4c66e0d4 | 5542 | /* |
0202e83f | 5543 | * Get an inode object given its inode number and corresponding root. |
4c66e0d4 DS |
5544 | * Path can be preallocated to prevent recursing back to iget through |
5545 | * allocator. NULL is also valid but may require an additional allocation | |
5546 | * later. | |
1a54ef8c | 5547 | */ |
0202e83f | 5548 | struct inode *btrfs_iget_path(struct super_block *s, u64 ino, |
4c66e0d4 | 5549 | struct btrfs_root *root, struct btrfs_path *path) |
1a54ef8c BR |
5550 | { |
5551 | struct inode *inode; | |
5552 | ||
0202e83f | 5553 | inode = btrfs_iget_locked(s, ino, root); |
1a54ef8c | 5554 | if (!inode) |
5d4f98a2 | 5555 | return ERR_PTR(-ENOMEM); |
1a54ef8c BR |
5556 | |
5557 | if (inode->i_state & I_NEW) { | |
67710892 FM |
5558 | int ret; |
5559 | ||
4222ea71 | 5560 | ret = btrfs_read_locked_inode(inode, path); |
9bc2ceff | 5561 | if (!ret) { |
4c45a4f4 | 5562 | inode_tree_add(BTRFS_I(inode)); |
1748f843 | 5563 | unlock_new_inode(inode); |
1748f843 | 5564 | } else { |
f5b3a417 AV |
5565 | iget_failed(inode); |
5566 | /* | |
5567 | * ret > 0 can come from btrfs_search_slot called by | |
5568 | * btrfs_read_locked_inode, this means the inode item | |
5569 | * was not found. | |
5570 | */ | |
5571 | if (ret > 0) | |
5572 | ret = -ENOENT; | |
5573 | inode = ERR_PTR(ret); | |
1748f843 MF |
5574 | } |
5575 | } | |
5576 | ||
1a54ef8c BR |
5577 | return inode; |
5578 | } | |
5579 | ||
0202e83f | 5580 | struct inode *btrfs_iget(struct super_block *s, u64 ino, struct btrfs_root *root) |
4222ea71 | 5581 | { |
0202e83f | 5582 | return btrfs_iget_path(s, ino, root, NULL); |
4222ea71 FM |
5583 | } |
5584 | ||
94628ad9 | 5585 | static struct inode *new_simple_dir(struct inode *dir, |
4df27c4d YZ |
5586 | struct btrfs_key *key, |
5587 | struct btrfs_root *root) | |
5588 | { | |
94628ad9 | 5589 | struct inode *inode = new_inode(dir->i_sb); |
4df27c4d YZ |
5590 | |
5591 | if (!inode) | |
5592 | return ERR_PTR(-ENOMEM); | |
5593 | ||
5c8fd99f | 5594 | BTRFS_I(inode)->root = btrfs_grab_root(root); |
4df27c4d | 5595 | memcpy(&BTRFS_I(inode)->location, key, sizeof(*key)); |
72ac3c0d | 5596 | set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags); |
4df27c4d YZ |
5597 | |
5598 | inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID; | |
6bb6b514 OS |
5599 | /* |
5600 | * We only need lookup, the rest is read-only and there's no inode | |
5601 | * associated with the dentry | |
5602 | */ | |
5603 | inode->i_op = &simple_dir_inode_operations; | |
1fdf4194 | 5604 | inode->i_opflags &= ~IOP_XATTR; |
4df27c4d YZ |
5605 | inode->i_fop = &simple_dir_operations; |
5606 | inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO; | |
2a9462de | 5607 | inode->i_mtime = inode_set_ctime_current(inode); |
94628ad9 | 5608 | inode->i_atime = dir->i_atime; |
d3c6be6f | 5609 | BTRFS_I(inode)->i_otime = inode->i_mtime; |
94628ad9 LT |
5610 | inode->i_uid = dir->i_uid; |
5611 | inode->i_gid = dir->i_gid; | |
4df27c4d YZ |
5612 | |
5613 | return inode; | |
5614 | } | |
5615 | ||
a55e65b8 DS |
5616 | static_assert(BTRFS_FT_UNKNOWN == FT_UNKNOWN); |
5617 | static_assert(BTRFS_FT_REG_FILE == FT_REG_FILE); | |
5618 | static_assert(BTRFS_FT_DIR == FT_DIR); | |
5619 | static_assert(BTRFS_FT_CHRDEV == FT_CHRDEV); | |
5620 | static_assert(BTRFS_FT_BLKDEV == FT_BLKDEV); | |
5621 | static_assert(BTRFS_FT_FIFO == FT_FIFO); | |
5622 | static_assert(BTRFS_FT_SOCK == FT_SOCK); | |
5623 | static_assert(BTRFS_FT_SYMLINK == FT_SYMLINK); | |
5624 | ||
6bf9e4bd QW |
5625 | static inline u8 btrfs_inode_type(struct inode *inode) |
5626 | { | |
6bf9e4bd QW |
5627 | return fs_umode_to_ftype(inode->i_mode); |
5628 | } | |
5629 | ||
3de4586c | 5630 | struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry) |
39279cc3 | 5631 | { |
0b246afa | 5632 | struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb); |
d397712b | 5633 | struct inode *inode; |
4df27c4d | 5634 | struct btrfs_root *root = BTRFS_I(dir)->root; |
39279cc3 CM |
5635 | struct btrfs_root *sub_root = root; |
5636 | struct btrfs_key location; | |
6bf9e4bd | 5637 | u8 di_type = 0; |
b4aff1f8 | 5638 | int ret = 0; |
39279cc3 CM |
5639 | |
5640 | if (dentry->d_name.len > BTRFS_NAME_LEN) | |
5641 | return ERR_PTR(-ENAMETOOLONG); | |
5f39d397 | 5642 | |
d1de429b | 5643 | ret = btrfs_inode_by_name(BTRFS_I(dir), dentry, &location, &di_type); |
39279cc3 CM |
5644 | if (ret < 0) |
5645 | return ERR_PTR(ret); | |
5f39d397 | 5646 | |
4df27c4d | 5647 | if (location.type == BTRFS_INODE_ITEM_KEY) { |
0202e83f | 5648 | inode = btrfs_iget(dir->i_sb, location.objectid, root); |
6bf9e4bd QW |
5649 | if (IS_ERR(inode)) |
5650 | return inode; | |
5651 | ||
5652 | /* Do extra check against inode mode with di_type */ | |
5653 | if (btrfs_inode_type(inode) != di_type) { | |
5654 | btrfs_crit(fs_info, | |
5655 | "inode mode mismatch with dir: inode mode=0%o btrfs type=%u dir type=%u", | |
5656 | inode->i_mode, btrfs_inode_type(inode), | |
5657 | di_type); | |
5658 | iput(inode); | |
5659 | return ERR_PTR(-EUCLEAN); | |
5660 | } | |
4df27c4d YZ |
5661 | return inode; |
5662 | } | |
5663 | ||
3c1b1c4c | 5664 | ret = fixup_tree_root_location(fs_info, BTRFS_I(dir), dentry, |
4df27c4d YZ |
5665 | &location, &sub_root); |
5666 | if (ret < 0) { | |
5667 | if (ret != -ENOENT) | |
5668 | inode = ERR_PTR(ret); | |
5669 | else | |
94628ad9 | 5670 | inode = new_simple_dir(dir, &location, root); |
4df27c4d | 5671 | } else { |
0202e83f | 5672 | inode = btrfs_iget(dir->i_sb, location.objectid, sub_root); |
00246528 | 5673 | btrfs_put_root(sub_root); |
76dda93c | 5674 | |
fc8b235f NB |
5675 | if (IS_ERR(inode)) |
5676 | return inode; | |
5677 | ||
0b246afa | 5678 | down_read(&fs_info->cleanup_work_sem); |
bc98a42c | 5679 | if (!sb_rdonly(inode->i_sb)) |
66b4ffd1 | 5680 | ret = btrfs_orphan_cleanup(sub_root); |
0b246afa | 5681 | up_read(&fs_info->cleanup_work_sem); |
01cd3367 JB |
5682 | if (ret) { |
5683 | iput(inode); | |
66b4ffd1 | 5684 | inode = ERR_PTR(ret); |
01cd3367 | 5685 | } |
c71bf099 YZ |
5686 | } |
5687 | ||
3de4586c CM |
5688 | return inode; |
5689 | } | |
5690 | ||
fe15ce44 | 5691 | static int btrfs_dentry_delete(const struct dentry *dentry) |
76dda93c YZ |
5692 | { |
5693 | struct btrfs_root *root; | |
2b0143b5 | 5694 | struct inode *inode = d_inode(dentry); |
76dda93c | 5695 | |
848cce0d | 5696 | if (!inode && !IS_ROOT(dentry)) |
2b0143b5 | 5697 | inode = d_inode(dentry->d_parent); |
76dda93c | 5698 | |
848cce0d LZ |
5699 | if (inode) { |
5700 | root = BTRFS_I(inode)->root; | |
efefb143 YZ |
5701 | if (btrfs_root_refs(&root->root_item) == 0) |
5702 | return 1; | |
848cce0d | 5703 | |
4a0cc7ca | 5704 | if (btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) |
848cce0d | 5705 | return 1; |
efefb143 | 5706 | } |
76dda93c YZ |
5707 | return 0; |
5708 | } | |
5709 | ||
3de4586c | 5710 | static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry, |
00cd8dd3 | 5711 | unsigned int flags) |
3de4586c | 5712 | { |
3837d208 | 5713 | struct inode *inode = btrfs_lookup_dentry(dir, dentry); |
5662344b | 5714 | |
3837d208 AV |
5715 | if (inode == ERR_PTR(-ENOENT)) |
5716 | inode = NULL; | |
41d28bca | 5717 | return d_splice_alias(inode, dentry); |
39279cc3 CM |
5718 | } |
5719 | ||
9b378f6a FM |
5720 | /* |
5721 | * Find the highest existing sequence number in a directory and then set the | |
5722 | * in-memory index_cnt variable to the first free sequence number. | |
5723 | */ | |
5724 | static int btrfs_set_inode_index_count(struct btrfs_inode *inode) | |
5725 | { | |
5726 | struct btrfs_root *root = inode->root; | |
5727 | struct btrfs_key key, found_key; | |
5728 | struct btrfs_path *path; | |
5729 | struct extent_buffer *leaf; | |
5730 | int ret; | |
5731 | ||
5732 | key.objectid = btrfs_ino(inode); | |
5733 | key.type = BTRFS_DIR_INDEX_KEY; | |
5734 | key.offset = (u64)-1; | |
5735 | ||
5736 | path = btrfs_alloc_path(); | |
5737 | if (!path) | |
5738 | return -ENOMEM; | |
5739 | ||
5740 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
5741 | if (ret < 0) | |
5742 | goto out; | |
5743 | /* FIXME: we should be able to handle this */ | |
5744 | if (ret == 0) | |
5745 | goto out; | |
5746 | ret = 0; | |
5747 | ||
5748 | if (path->slots[0] == 0) { | |
5749 | inode->index_cnt = BTRFS_DIR_START_INDEX; | |
5750 | goto out; | |
5751 | } | |
5752 | ||
5753 | path->slots[0]--; | |
5754 | ||
5755 | leaf = path->nodes[0]; | |
5756 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
5757 | ||
5758 | if (found_key.objectid != btrfs_ino(inode) || | |
5759 | found_key.type != BTRFS_DIR_INDEX_KEY) { | |
5760 | inode->index_cnt = BTRFS_DIR_START_INDEX; | |
5761 | goto out; | |
5762 | } | |
5763 | ||
5764 | inode->index_cnt = found_key.offset + 1; | |
5765 | out: | |
5766 | btrfs_free_path(path); | |
5767 | return ret; | |
5768 | } | |
5769 | ||
5770 | static int btrfs_get_dir_last_index(struct btrfs_inode *dir, u64 *index) | |
5771 | { | |
5772 | if (dir->index_cnt == (u64)-1) { | |
5773 | int ret; | |
5774 | ||
5775 | ret = btrfs_inode_delayed_dir_index_count(dir); | |
5776 | if (ret) { | |
5777 | ret = btrfs_set_inode_index_count(dir); | |
5778 | if (ret) | |
5779 | return ret; | |
5780 | } | |
5781 | } | |
5782 | ||
5783 | *index = dir->index_cnt; | |
5784 | ||
5785 | return 0; | |
5786 | } | |
5787 | ||
23b5ec74 JB |
5788 | /* |
5789 | * All this infrastructure exists because dir_emit can fault, and we are holding | |
5790 | * the tree lock when doing readdir. For now just allocate a buffer and copy | |
5791 | * our information into that, and then dir_emit from the buffer. This is | |
5792 | * similar to what NFS does, only we don't keep the buffer around in pagecache | |
5793 | * because I'm afraid I'll mess that up. Long term we need to make filldir do | |
5794 | * copy_to_user_inatomic so we don't have to worry about page faulting under the | |
5795 | * tree lock. | |
5796 | */ | |
5797 | static int btrfs_opendir(struct inode *inode, struct file *file) | |
5798 | { | |
5799 | struct btrfs_file_private *private; | |
9b378f6a FM |
5800 | u64 last_index; |
5801 | int ret; | |
5802 | ||
5803 | ret = btrfs_get_dir_last_index(BTRFS_I(inode), &last_index); | |
5804 | if (ret) | |
5805 | return ret; | |
23b5ec74 JB |
5806 | |
5807 | private = kzalloc(sizeof(struct btrfs_file_private), GFP_KERNEL); | |
5808 | if (!private) | |
5809 | return -ENOMEM; | |
9b378f6a | 5810 | private->last_index = last_index; |
23b5ec74 JB |
5811 | private->filldir_buf = kzalloc(PAGE_SIZE, GFP_KERNEL); |
5812 | if (!private->filldir_buf) { | |
5813 | kfree(private); | |
5814 | return -ENOMEM; | |
5815 | } | |
5816 | file->private_data = private; | |
5817 | return 0; | |
5818 | } | |
5819 | ||
5820 | struct dir_entry { | |
5821 | u64 ino; | |
5822 | u64 offset; | |
5823 | unsigned type; | |
5824 | int name_len; | |
5825 | }; | |
5826 | ||
5827 | static int btrfs_filldir(void *addr, int entries, struct dir_context *ctx) | |
5828 | { | |
5829 | while (entries--) { | |
5830 | struct dir_entry *entry = addr; | |
5831 | char *name = (char *)(entry + 1); | |
5832 | ||
92d32170 DS |
5833 | ctx->pos = get_unaligned(&entry->offset); |
5834 | if (!dir_emit(ctx, name, get_unaligned(&entry->name_len), | |
5835 | get_unaligned(&entry->ino), | |
5836 | get_unaligned(&entry->type))) | |
23b5ec74 | 5837 | return 1; |
92d32170 DS |
5838 | addr += sizeof(struct dir_entry) + |
5839 | get_unaligned(&entry->name_len); | |
23b5ec74 JB |
5840 | ctx->pos++; |
5841 | } | |
5842 | return 0; | |
5843 | } | |
5844 | ||
9cdda8d3 | 5845 | static int btrfs_real_readdir(struct file *file, struct dir_context *ctx) |
39279cc3 | 5846 | { |
9cdda8d3 | 5847 | struct inode *inode = file_inode(file); |
39279cc3 | 5848 | struct btrfs_root *root = BTRFS_I(inode)->root; |
23b5ec74 | 5849 | struct btrfs_file_private *private = file->private_data; |
39279cc3 CM |
5850 | struct btrfs_dir_item *di; |
5851 | struct btrfs_key key; | |
5f39d397 | 5852 | struct btrfs_key found_key; |
39279cc3 | 5853 | struct btrfs_path *path; |
23b5ec74 | 5854 | void *addr; |
84af994b RJ |
5855 | LIST_HEAD(ins_list); |
5856 | LIST_HEAD(del_list); | |
39279cc3 | 5857 | int ret; |
5f39d397 CM |
5858 | char *name_ptr; |
5859 | int name_len; | |
23b5ec74 JB |
5860 | int entries = 0; |
5861 | int total_len = 0; | |
02dbfc99 | 5862 | bool put = false; |
c2951f32 | 5863 | struct btrfs_key location; |
5f39d397 | 5864 | |
9cdda8d3 AV |
5865 | if (!dir_emit_dots(file, ctx)) |
5866 | return 0; | |
5867 | ||
49593bfa | 5868 | path = btrfs_alloc_path(); |
16cdcec7 MX |
5869 | if (!path) |
5870 | return -ENOMEM; | |
ff5714cc | 5871 | |
23b5ec74 | 5872 | addr = private->filldir_buf; |
e4058b54 | 5873 | path->reada = READA_FORWARD; |
49593bfa | 5874 | |
9b378f6a FM |
5875 | put = btrfs_readdir_get_delayed_items(inode, private->last_index, |
5876 | &ins_list, &del_list); | |
16cdcec7 | 5877 | |
23b5ec74 | 5878 | again: |
c2951f32 | 5879 | key.type = BTRFS_DIR_INDEX_KEY; |
9cdda8d3 | 5880 | key.offset = ctx->pos; |
4a0cc7ca | 5881 | key.objectid = btrfs_ino(BTRFS_I(inode)); |
5f39d397 | 5882 | |
a8ce68fd | 5883 | btrfs_for_each_slot(root, &key, &found_key, path, ret) { |
23b5ec74 | 5884 | struct dir_entry *entry; |
a8ce68fd | 5885 | struct extent_buffer *leaf = path->nodes[0]; |
94a48aef | 5886 | u8 ftype; |
5f39d397 CM |
5887 | |
5888 | if (found_key.objectid != key.objectid) | |
39279cc3 | 5889 | break; |
c2951f32 | 5890 | if (found_key.type != BTRFS_DIR_INDEX_KEY) |
39279cc3 | 5891 | break; |
9cdda8d3 | 5892 | if (found_key.offset < ctx->pos) |
a8ce68fd | 5893 | continue; |
9b378f6a FM |
5894 | if (found_key.offset > private->last_index) |
5895 | break; | |
c2951f32 | 5896 | if (btrfs_should_delete_dir_index(&del_list, found_key.offset)) |
a8ce68fd GN |
5897 | continue; |
5898 | di = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dir_item); | |
c2951f32 | 5899 | name_len = btrfs_dir_name_len(leaf, di); |
23b5ec74 JB |
5900 | if ((total_len + sizeof(struct dir_entry) + name_len) >= |
5901 | PAGE_SIZE) { | |
5902 | btrfs_release_path(path); | |
5903 | ret = btrfs_filldir(private->filldir_buf, entries, ctx); | |
5904 | if (ret) | |
5905 | goto nopos; | |
5906 | addr = private->filldir_buf; | |
5907 | entries = 0; | |
5908 | total_len = 0; | |
5909 | goto again; | |
c2951f32 | 5910 | } |
23b5ec74 | 5911 | |
94a48aef | 5912 | ftype = btrfs_dir_flags_to_ftype(btrfs_dir_flags(leaf, di)); |
23b5ec74 | 5913 | entry = addr; |
23b5ec74 | 5914 | name_ptr = (char *)(entry + 1); |
94a48aef OS |
5915 | read_extent_buffer(leaf, name_ptr, |
5916 | (unsigned long)(di + 1), name_len); | |
5917 | put_unaligned(name_len, &entry->name_len); | |
5918 | put_unaligned(fs_ftype_to_dtype(ftype), &entry->type); | |
c2951f32 | 5919 | btrfs_dir_item_key_to_cpu(leaf, di, &location); |
92d32170 DS |
5920 | put_unaligned(location.objectid, &entry->ino); |
5921 | put_unaligned(found_key.offset, &entry->offset); | |
23b5ec74 JB |
5922 | entries++; |
5923 | addr += sizeof(struct dir_entry) + name_len; | |
5924 | total_len += sizeof(struct dir_entry) + name_len; | |
39279cc3 | 5925 | } |
a8ce68fd GN |
5926 | /* Catch error encountered during iteration */ |
5927 | if (ret < 0) | |
5928 | goto err; | |
5929 | ||
23b5ec74 JB |
5930 | btrfs_release_path(path); |
5931 | ||
5932 | ret = btrfs_filldir(private->filldir_buf, entries, ctx); | |
5933 | if (ret) | |
5934 | goto nopos; | |
49593bfa | 5935 | |
d2fbb2b5 | 5936 | ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list); |
c2951f32 | 5937 | if (ret) |
bc4ef759 DS |
5938 | goto nopos; |
5939 | ||
db62efbb ZB |
5940 | /* |
5941 | * Stop new entries from being returned after we return the last | |
5942 | * entry. | |
5943 | * | |
5944 | * New directory entries are assigned a strictly increasing | |
5945 | * offset. This means that new entries created during readdir | |
5946 | * are *guaranteed* to be seen in the future by that readdir. | |
5947 | * This has broken buggy programs which operate on names as | |
5948 | * they're returned by readdir. Until we re-use freed offsets | |
5949 | * we have this hack to stop new entries from being returned | |
5950 | * under the assumption that they'll never reach this huge | |
5951 | * offset. | |
5952 | * | |
5953 | * This is being careful not to overflow 32bit loff_t unless the | |
5954 | * last entry requires it because doing so has broken 32bit apps | |
5955 | * in the past. | |
5956 | */ | |
c2951f32 JM |
5957 | if (ctx->pos >= INT_MAX) |
5958 | ctx->pos = LLONG_MAX; | |
5959 | else | |
5960 | ctx->pos = INT_MAX; | |
39279cc3 CM |
5961 | nopos: |
5962 | ret = 0; | |
5963 | err: | |
02dbfc99 OS |
5964 | if (put) |
5965 | btrfs_readdir_put_delayed_items(inode, &ins_list, &del_list); | |
39279cc3 | 5966 | btrfs_free_path(path); |
39279cc3 CM |
5967 | return ret; |
5968 | } | |
5969 | ||
39279cc3 | 5970 | /* |
54aa1f4d | 5971 | * This is somewhat expensive, updating the tree every time the |
39279cc3 CM |
5972 | * inode changes. But, it is most likely to find the inode in cache. |
5973 | * FIXME, needs more benchmarking...there are no reasons other than performance | |
5974 | * to keep or drop this code. | |
5975 | */ | |
7152b425 | 5976 | static int btrfs_dirty_inode(struct btrfs_inode *inode) |
39279cc3 | 5977 | { |
7152b425 DS |
5978 | struct btrfs_root *root = inode->root; |
5979 | struct btrfs_fs_info *fs_info = root->fs_info; | |
39279cc3 | 5980 | struct btrfs_trans_handle *trans; |
8929ecfa YZ |
5981 | int ret; |
5982 | ||
7152b425 | 5983 | if (test_bit(BTRFS_INODE_DUMMY, &inode->runtime_flags)) |
22c44fe6 | 5984 | return 0; |
39279cc3 | 5985 | |
7a7eaa40 | 5986 | trans = btrfs_join_transaction(root); |
22c44fe6 JB |
5987 | if (IS_ERR(trans)) |
5988 | return PTR_ERR(trans); | |
8929ecfa | 5989 | |
7152b425 | 5990 | ret = btrfs_update_inode(trans, root, inode); |
4d14c5cd | 5991 | if (ret && (ret == -ENOSPC || ret == -EDQUOT)) { |
94b60442 | 5992 | /* whoops, lets try again with the full transaction */ |
3a45bb20 | 5993 | btrfs_end_transaction(trans); |
94b60442 | 5994 | trans = btrfs_start_transaction(root, 1); |
22c44fe6 JB |
5995 | if (IS_ERR(trans)) |
5996 | return PTR_ERR(trans); | |
8929ecfa | 5997 | |
7152b425 | 5998 | ret = btrfs_update_inode(trans, root, inode); |
94b60442 | 5999 | } |
3a45bb20 | 6000 | btrfs_end_transaction(trans); |
7152b425 | 6001 | if (inode->delayed_node) |
2ff7e61e | 6002 | btrfs_balance_delayed_items(fs_info); |
22c44fe6 JB |
6003 | |
6004 | return ret; | |
6005 | } | |
6006 | ||
6007 | /* | |
6008 | * This is a copy of file_update_time. We need this so we can return error on | |
6009 | * ENOSPC for updating the inode in the case of file write and mmap writes. | |
6010 | */ | |
913e9928 | 6011 | static int btrfs_update_time(struct inode *inode, int flags) |
22c44fe6 | 6012 | { |
2bc55652 | 6013 | struct btrfs_root *root = BTRFS_I(inode)->root; |
3a8c7231 | 6014 | bool dirty = flags & ~S_VERSION; |
2bc55652 AB |
6015 | |
6016 | if (btrfs_root_readonly(root)) | |
6017 | return -EROFS; | |
6018 | ||
bb7cc0a6 | 6019 | dirty = inode_update_timestamps(inode, flags); |
7152b425 | 6020 | return dirty ? btrfs_dirty_inode(BTRFS_I(inode)) : 0; |
39279cc3 CM |
6021 | } |
6022 | ||
d352ac68 CM |
6023 | /* |
6024 | * helper to find a free sequence number in a given directory. This current | |
6025 | * code is very simple, later versions will do smarter things in the btree | |
6026 | */ | |
877574e2 | 6027 | int btrfs_set_inode_index(struct btrfs_inode *dir, u64 *index) |
aec7477b JB |
6028 | { |
6029 | int ret = 0; | |
6030 | ||
877574e2 NB |
6031 | if (dir->index_cnt == (u64)-1) { |
6032 | ret = btrfs_inode_delayed_dir_index_count(dir); | |
16cdcec7 MX |
6033 | if (ret) { |
6034 | ret = btrfs_set_inode_index_count(dir); | |
6035 | if (ret) | |
6036 | return ret; | |
6037 | } | |
aec7477b JB |
6038 | } |
6039 | ||
877574e2 NB |
6040 | *index = dir->index_cnt; |
6041 | dir->index_cnt++; | |
aec7477b JB |
6042 | |
6043 | return ret; | |
6044 | } | |
6045 | ||
b0d5d10f CM |
6046 | static int btrfs_insert_inode_locked(struct inode *inode) |
6047 | { | |
6048 | struct btrfs_iget_args args; | |
0202e83f DS |
6049 | |
6050 | args.ino = BTRFS_I(inode)->location.objectid; | |
b0d5d10f CM |
6051 | args.root = BTRFS_I(inode)->root; |
6052 | ||
6053 | return insert_inode_locked4(inode, | |
6054 | btrfs_inode_hash(inode->i_ino, BTRFS_I(inode)->root), | |
6055 | btrfs_find_actor, &args); | |
6056 | } | |
6057 | ||
3538d68d OS |
6058 | int btrfs_new_inode_prepare(struct btrfs_new_inode_args *args, |
6059 | unsigned int *trans_num_items) | |
6060 | { | |
6061 | struct inode *dir = args->dir; | |
6062 | struct inode *inode = args->inode; | |
6063 | int ret; | |
6064 | ||
ab3c5c18 STD |
6065 | if (!args->orphan) { |
6066 | ret = fscrypt_setup_filename(dir, &args->dentry->d_name, 0, | |
6067 | &args->fname); | |
6068 | if (ret) | |
6069 | return ret; | |
ab3c5c18 STD |
6070 | } |
6071 | ||
3538d68d | 6072 | ret = posix_acl_create(dir, &inode->i_mode, &args->default_acl, &args->acl); |
ab3c5c18 STD |
6073 | if (ret) { |
6074 | fscrypt_free_filename(&args->fname); | |
3538d68d | 6075 | return ret; |
ab3c5c18 | 6076 | } |
3538d68d OS |
6077 | |
6078 | /* 1 to add inode item */ | |
6079 | *trans_num_items = 1; | |
6080 | /* 1 to add compression property */ | |
6081 | if (BTRFS_I(dir)->prop_compress) | |
6082 | (*trans_num_items)++; | |
6083 | /* 1 to add default ACL xattr */ | |
6084 | if (args->default_acl) | |
6085 | (*trans_num_items)++; | |
6086 | /* 1 to add access ACL xattr */ | |
6087 | if (args->acl) | |
6088 | (*trans_num_items)++; | |
6089 | #ifdef CONFIG_SECURITY | |
6090 | /* 1 to add LSM xattr */ | |
6091 | if (dir->i_security) | |
6092 | (*trans_num_items)++; | |
6093 | #endif | |
6094 | if (args->orphan) { | |
6095 | /* 1 to add orphan item */ | |
6096 | (*trans_num_items)++; | |
6097 | } else { | |
6098 | /* | |
3538d68d OS |
6099 | * 1 to add dir item |
6100 | * 1 to add dir index | |
6101 | * 1 to update parent inode item | |
97bdf1a9 FM |
6102 | * |
6103 | * No need for 1 unit for the inode ref item because it is | |
6104 | * inserted in a batch together with the inode item at | |
6105 | * btrfs_create_new_inode(). | |
3538d68d | 6106 | */ |
97bdf1a9 | 6107 | *trans_num_items += 3; |
3538d68d OS |
6108 | } |
6109 | return 0; | |
6110 | } | |
6111 | ||
6112 | void btrfs_new_inode_args_destroy(struct btrfs_new_inode_args *args) | |
6113 | { | |
6114 | posix_acl_release(args->acl); | |
6115 | posix_acl_release(args->default_acl); | |
ab3c5c18 | 6116 | fscrypt_free_filename(&args->fname); |
3538d68d OS |
6117 | } |
6118 | ||
19aee8de AJ |
6119 | /* |
6120 | * Inherit flags from the parent inode. | |
6121 | * | |
6122 | * Currently only the compression flags and the cow flags are inherited. | |
6123 | */ | |
7a0443f0 | 6124 | static void btrfs_inherit_iflags(struct btrfs_inode *inode, struct btrfs_inode *dir) |
19aee8de AJ |
6125 | { |
6126 | unsigned int flags; | |
6127 | ||
7a0443f0 | 6128 | flags = dir->flags; |
19aee8de AJ |
6129 | |
6130 | if (flags & BTRFS_INODE_NOCOMPRESS) { | |
7a0443f0 DS |
6131 | inode->flags &= ~BTRFS_INODE_COMPRESS; |
6132 | inode->flags |= BTRFS_INODE_NOCOMPRESS; | |
19aee8de | 6133 | } else if (flags & BTRFS_INODE_COMPRESS) { |
7a0443f0 DS |
6134 | inode->flags &= ~BTRFS_INODE_NOCOMPRESS; |
6135 | inode->flags |= BTRFS_INODE_COMPRESS; | |
19aee8de AJ |
6136 | } |
6137 | ||
6138 | if (flags & BTRFS_INODE_NODATACOW) { | |
7a0443f0 DS |
6139 | inode->flags |= BTRFS_INODE_NODATACOW; |
6140 | if (S_ISREG(inode->vfs_inode.i_mode)) | |
6141 | inode->flags |= BTRFS_INODE_NODATASUM; | |
19aee8de AJ |
6142 | } |
6143 | ||
7a0443f0 | 6144 | btrfs_sync_inode_flags_to_i_flags(&inode->vfs_inode); |
19aee8de AJ |
6145 | } |
6146 | ||
3538d68d | 6147 | int btrfs_create_new_inode(struct btrfs_trans_handle *trans, |
caae78e0 | 6148 | struct btrfs_new_inode_args *args) |
39279cc3 | 6149 | { |
caae78e0 | 6150 | struct inode *dir = args->dir; |
3538d68d | 6151 | struct inode *inode = args->inode; |
6db75318 | 6152 | const struct fscrypt_str *name = args->orphan ? NULL : &args->fname.disk_name; |
caae78e0 | 6153 | struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb); |
3538d68d | 6154 | struct btrfs_root *root; |
5f39d397 | 6155 | struct btrfs_inode_item *inode_item; |
39279cc3 | 6156 | struct btrfs_key *location; |
5f39d397 | 6157 | struct btrfs_path *path; |
6437d458 | 6158 | u64 objectid; |
9c58309d CM |
6159 | struct btrfs_inode_ref *ref; |
6160 | struct btrfs_key key[2]; | |
6161 | u32 sizes[2]; | |
b7ef5f3a | 6162 | struct btrfs_item_batch batch; |
9c58309d | 6163 | unsigned long ptr; |
39279cc3 | 6164 | int ret; |
39279cc3 | 6165 | |
5f39d397 | 6166 | path = btrfs_alloc_path(); |
d8926bb3 | 6167 | if (!path) |
a1fd0c35 | 6168 | return -ENOMEM; |
39279cc3 | 6169 | |
3538d68d OS |
6170 | if (!args->subvol) |
6171 | BTRFS_I(inode)->root = btrfs_grab_root(BTRFS_I(dir)->root); | |
6172 | root = BTRFS_I(inode)->root; | |
6173 | ||
6437d458 | 6174 | ret = btrfs_get_free_objectid(root, &objectid); |
caae78e0 OS |
6175 | if (ret) |
6176 | goto out; | |
581bb050 LZ |
6177 | inode->i_ino = objectid; |
6178 | ||
caae78e0 OS |
6179 | if (args->orphan) { |
6180 | /* | |
6181 | * O_TMPFILE, set link count to 0, so that after this point, we | |
6182 | * fill in an inode item with the correct link count. | |
6183 | */ | |
6184 | set_nlink(inode, 0); | |
6185 | } else { | |
1abe9b8a | 6186 | trace_btrfs_inode_request(dir); |
6187 | ||
caae78e0 OS |
6188 | ret = btrfs_set_inode_index(BTRFS_I(dir), &BTRFS_I(inode)->dir_index); |
6189 | if (ret) | |
6190 | goto out; | |
aec7477b | 6191 | } |
49024388 FM |
6192 | /* index_cnt is ignored for everything but a dir. */ |
6193 | BTRFS_I(inode)->index_cnt = BTRFS_DIR_START_INDEX; | |
e02119d5 | 6194 | BTRFS_I(inode)->generation = trans->transid; |
76195853 | 6195 | inode->i_generation = BTRFS_I(inode)->generation; |
b888db2b | 6196 | |
caae78e0 OS |
6197 | /* |
6198 | * Subvolumes don't inherit flags from their parent directory. | |
6199 | * Originally this was probably by accident, but we probably can't | |
6200 | * change it now without compatibility issues. | |
6201 | */ | |
6202 | if (!args->subvol) | |
7a0443f0 | 6203 | btrfs_inherit_iflags(BTRFS_I(inode), BTRFS_I(dir)); |
305eaac0 | 6204 | |
a1fd0c35 | 6205 | if (S_ISREG(inode->i_mode)) { |
305eaac0 OS |
6206 | if (btrfs_test_opt(fs_info, NODATASUM)) |
6207 | BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM; | |
6208 | if (btrfs_test_opt(fs_info, NODATACOW)) | |
6209 | BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW | | |
6210 | BTRFS_INODE_NODATASUM; | |
6211 | } | |
6212 | ||
caae78e0 OS |
6213 | location = &BTRFS_I(inode)->location; |
6214 | location->objectid = objectid; | |
6215 | location->offset = 0; | |
6216 | location->type = BTRFS_INODE_ITEM_KEY; | |
6217 | ||
6218 | ret = btrfs_insert_inode_locked(inode); | |
6219 | if (ret < 0) { | |
6220 | if (!args->orphan) | |
6221 | BTRFS_I(dir)->index_cnt--; | |
6222 | goto out; | |
6223 | } | |
6224 | ||
5dc562c5 JB |
6225 | /* |
6226 | * We could have gotten an inode number from somebody who was fsynced | |
6227 | * and then removed in this same transaction, so let's just set full | |
6228 | * sync since it will be a full sync anyway and this will blow away the | |
6229 | * old info in the log. | |
6230 | */ | |
23e3337f | 6231 | btrfs_set_inode_full_sync(BTRFS_I(inode)); |
5dc562c5 | 6232 | |
9c58309d | 6233 | key[0].objectid = objectid; |
962a298f | 6234 | key[0].type = BTRFS_INODE_ITEM_KEY; |
9c58309d CM |
6235 | key[0].offset = 0; |
6236 | ||
9c58309d | 6237 | sizes[0] = sizeof(struct btrfs_inode_item); |
ef3b9af5 | 6238 | |
caae78e0 | 6239 | if (!args->orphan) { |
ef3b9af5 FM |
6240 | /* |
6241 | * Start new inodes with an inode_ref. This is slightly more | |
6242 | * efficient for small numbers of hard links since they will | |
6243 | * be packed into one item. Extended refs will kick in if we | |
6244 | * add more hard links than can fit in the ref item. | |
6245 | */ | |
6246 | key[1].objectid = objectid; | |
962a298f | 6247 | key[1].type = BTRFS_INODE_REF_KEY; |
caae78e0 | 6248 | if (args->subvol) { |
23c24ef8 | 6249 | key[1].offset = objectid; |
caae78e0 OS |
6250 | sizes[1] = 2 + sizeof(*ref); |
6251 | } else { | |
6252 | key[1].offset = btrfs_ino(BTRFS_I(dir)); | |
e43eec81 | 6253 | sizes[1] = name->len + sizeof(*ref); |
caae78e0 | 6254 | } |
ef3b9af5 | 6255 | } |
9c58309d | 6256 | |
b7ef5f3a FM |
6257 | batch.keys = &key[0]; |
6258 | batch.data_sizes = &sizes[0]; | |
caae78e0 OS |
6259 | batch.total_data_size = sizes[0] + (args->orphan ? 0 : sizes[1]); |
6260 | batch.nr = args->orphan ? 1 : 2; | |
b7ef5f3a | 6261 | ret = btrfs_insert_empty_items(trans, root, path, &batch); |
caae78e0 OS |
6262 | if (ret != 0) { |
6263 | btrfs_abort_transaction(trans, ret); | |
6264 | goto discard; | |
6265 | } | |
5f39d397 | 6266 | |
2a9462de | 6267 | inode->i_mtime = inode_set_ctime_current(inode); |
9cc97d64 | 6268 | inode->i_atime = inode->i_mtime; |
d3c6be6f | 6269 | BTRFS_I(inode)->i_otime = inode->i_mtime; |
9cc97d64 | 6270 | |
caae78e0 OS |
6271 | /* |
6272 | * We're going to fill the inode item now, so at this point the inode | |
6273 | * must be fully initialized. | |
6274 | */ | |
6275 | ||
5f39d397 CM |
6276 | inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0], |
6277 | struct btrfs_inode_item); | |
b159fa28 | 6278 | memzero_extent_buffer(path->nodes[0], (unsigned long)inode_item, |
293f7e07 | 6279 | sizeof(*inode_item)); |
e02119d5 | 6280 | fill_inode_item(trans, path->nodes[0], inode_item, inode); |
9c58309d | 6281 | |
caae78e0 | 6282 | if (!args->orphan) { |
ef3b9af5 FM |
6283 | ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1, |
6284 | struct btrfs_inode_ref); | |
ef3b9af5 | 6285 | ptr = (unsigned long)(ref + 1); |
caae78e0 OS |
6286 | if (args->subvol) { |
6287 | btrfs_set_inode_ref_name_len(path->nodes[0], ref, 2); | |
6288 | btrfs_set_inode_ref_index(path->nodes[0], ref, 0); | |
6289 | write_extent_buffer(path->nodes[0], "..", ptr, 2); | |
6290 | } else { | |
e43eec81 STD |
6291 | btrfs_set_inode_ref_name_len(path->nodes[0], ref, |
6292 | name->len); | |
caae78e0 OS |
6293 | btrfs_set_inode_ref_index(path->nodes[0], ref, |
6294 | BTRFS_I(inode)->dir_index); | |
e43eec81 STD |
6295 | write_extent_buffer(path->nodes[0], name->name, ptr, |
6296 | name->len); | |
caae78e0 | 6297 | } |
ef3b9af5 | 6298 | } |
9c58309d | 6299 | |
5f39d397 | 6300 | btrfs_mark_buffer_dirty(path->nodes[0]); |
814e7718 FM |
6301 | /* |
6302 | * We don't need the path anymore, plus inheriting properties, adding | |
6303 | * ACLs, security xattrs, orphan item or adding the link, will result in | |
6304 | * allocating yet another path. So just free our path. | |
6305 | */ | |
6306 | btrfs_free_path(path); | |
6307 | path = NULL; | |
5f39d397 | 6308 | |
6c3636eb STD |
6309 | if (args->subvol) { |
6310 | struct inode *parent; | |
6311 | ||
6312 | /* | |
6313 | * Subvolumes inherit properties from their parent subvolume, | |
6314 | * not the directory they were created in. | |
6315 | */ | |
6316 | parent = btrfs_iget(fs_info->sb, BTRFS_FIRST_FREE_OBJECTID, | |
6317 | BTRFS_I(dir)->root); | |
6318 | if (IS_ERR(parent)) { | |
6319 | ret = PTR_ERR(parent); | |
6320 | } else { | |
6321 | ret = btrfs_inode_inherit_props(trans, inode, parent); | |
6322 | iput(parent); | |
6323 | } | |
6324 | } else { | |
6325 | ret = btrfs_inode_inherit_props(trans, inode, dir); | |
6326 | } | |
6327 | if (ret) { | |
6328 | btrfs_err(fs_info, | |
6329 | "error inheriting props for ino %llu (root %llu): %d", | |
6330 | btrfs_ino(BTRFS_I(inode)), root->root_key.objectid, | |
6331 | ret); | |
6332 | } | |
6333 | ||
6334 | /* | |
6335 | * Subvolumes don't inherit ACLs or get passed to the LSM. This is | |
6336 | * probably a bug. | |
6337 | */ | |
6338 | if (!args->subvol) { | |
6339 | ret = btrfs_init_inode_security(trans, args); | |
6340 | if (ret) { | |
6341 | btrfs_abort_transaction(trans, ret); | |
6342 | goto discard; | |
6343 | } | |
6344 | } | |
6345 | ||
4c45a4f4 | 6346 | inode_tree_add(BTRFS_I(inode)); |
1abe9b8a | 6347 | |
6348 | trace_btrfs_inode_new(inode); | |
d9094414 | 6349 | btrfs_set_inode_last_trans(trans, BTRFS_I(inode)); |
1abe9b8a | 6350 | |
8ea05e3a AB |
6351 | btrfs_update_root_times(trans, root); |
6352 | ||
caae78e0 OS |
6353 | if (args->orphan) { |
6354 | ret = btrfs_orphan_add(trans, BTRFS_I(inode)); | |
6355 | } else { | |
6356 | ret = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode), name, | |
e43eec81 | 6357 | 0, BTRFS_I(inode)->dir_index); |
caae78e0 OS |
6358 | } |
6359 | if (ret) { | |
6360 | btrfs_abort_transaction(trans, ret); | |
6361 | goto discard; | |
6362 | } | |
63541927 | 6363 | |
814e7718 | 6364 | return 0; |
b0d5d10f | 6365 | |
caae78e0 | 6366 | discard: |
a1fd0c35 OS |
6367 | /* |
6368 | * discard_new_inode() calls iput(), but the caller owns the reference | |
6369 | * to the inode. | |
6370 | */ | |
6371 | ihold(inode); | |
32955c54 | 6372 | discard_new_inode(inode); |
caae78e0 | 6373 | out: |
5f39d397 | 6374 | btrfs_free_path(path); |
a1fd0c35 | 6375 | return ret; |
39279cc3 CM |
6376 | } |
6377 | ||
d352ac68 CM |
6378 | /* |
6379 | * utility function to add 'inode' into 'parent_inode' with | |
6380 | * a give name and a given sequence number. | |
6381 | * if 'add_backref' is true, also insert a backref from the | |
6382 | * inode to the parent directory. | |
6383 | */ | |
e02119d5 | 6384 | int btrfs_add_link(struct btrfs_trans_handle *trans, |
db0a669f | 6385 | struct btrfs_inode *parent_inode, struct btrfs_inode *inode, |
6db75318 | 6386 | const struct fscrypt_str *name, int add_backref, u64 index) |
39279cc3 | 6387 | { |
4df27c4d | 6388 | int ret = 0; |
39279cc3 | 6389 | struct btrfs_key key; |
db0a669f NB |
6390 | struct btrfs_root *root = parent_inode->root; |
6391 | u64 ino = btrfs_ino(inode); | |
6392 | u64 parent_ino = btrfs_ino(parent_inode); | |
5f39d397 | 6393 | |
33345d01 | 6394 | if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) { |
db0a669f | 6395 | memcpy(&key, &inode->root->root_key, sizeof(key)); |
4df27c4d | 6396 | } else { |
33345d01 | 6397 | key.objectid = ino; |
962a298f | 6398 | key.type = BTRFS_INODE_ITEM_KEY; |
4df27c4d YZ |
6399 | key.offset = 0; |
6400 | } | |
6401 | ||
33345d01 | 6402 | if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) { |
6025c19f | 6403 | ret = btrfs_add_root_ref(trans, key.objectid, |
0b246afa | 6404 | root->root_key.objectid, parent_ino, |
e43eec81 | 6405 | index, name); |
4df27c4d | 6406 | } else if (add_backref) { |
e43eec81 STD |
6407 | ret = btrfs_insert_inode_ref(trans, root, name, |
6408 | ino, parent_ino, index); | |
4df27c4d | 6409 | } |
39279cc3 | 6410 | |
79787eaa JM |
6411 | /* Nothing to clean up yet */ |
6412 | if (ret) | |
6413 | return ret; | |
4df27c4d | 6414 | |
e43eec81 | 6415 | ret = btrfs_insert_dir_item(trans, name, parent_inode, &key, |
db0a669f | 6416 | btrfs_inode_type(&inode->vfs_inode), index); |
9c52057c | 6417 | if (ret == -EEXIST || ret == -EOVERFLOW) |
79787eaa JM |
6418 | goto fail_dir_item; |
6419 | else if (ret) { | |
66642832 | 6420 | btrfs_abort_transaction(trans, ret); |
79787eaa | 6421 | return ret; |
39279cc3 | 6422 | } |
79787eaa | 6423 | |
db0a669f | 6424 | btrfs_i_size_write(parent_inode, parent_inode->vfs_inode.i_size + |
e43eec81 | 6425 | name->len * 2); |
db0a669f | 6426 | inode_inc_iversion(&parent_inode->vfs_inode); |
5338e43a FM |
6427 | /* |
6428 | * If we are replaying a log tree, we do not want to update the mtime | |
6429 | * and ctime of the parent directory with the current time, since the | |
6430 | * log replay procedure is responsible for setting them to their correct | |
6431 | * values (the ones it had when the fsync was done). | |
6432 | */ | |
2a9462de JL |
6433 | if (!test_bit(BTRFS_FS_LOG_RECOVERING, &root->fs_info->flags)) |
6434 | parent_inode->vfs_inode.i_mtime = | |
6435 | inode_set_ctime_current(&parent_inode->vfs_inode); | |
5338e43a | 6436 | |
9a56fcd1 | 6437 | ret = btrfs_update_inode(trans, root, parent_inode); |
79787eaa | 6438 | if (ret) |
66642832 | 6439 | btrfs_abort_transaction(trans, ret); |
39279cc3 | 6440 | return ret; |
fe66a05a CM |
6441 | |
6442 | fail_dir_item: | |
6443 | if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) { | |
6444 | u64 local_index; | |
6445 | int err; | |
3ee1c553 | 6446 | err = btrfs_del_root_ref(trans, key.objectid, |
0b246afa | 6447 | root->root_key.objectid, parent_ino, |
e43eec81 | 6448 | &local_index, name); |
1690dd41 JT |
6449 | if (err) |
6450 | btrfs_abort_transaction(trans, err); | |
fe66a05a CM |
6451 | } else if (add_backref) { |
6452 | u64 local_index; | |
6453 | int err; | |
6454 | ||
e43eec81 STD |
6455 | err = btrfs_del_inode_ref(trans, root, name, ino, parent_ino, |
6456 | &local_index); | |
1690dd41 JT |
6457 | if (err) |
6458 | btrfs_abort_transaction(trans, err); | |
fe66a05a | 6459 | } |
1690dd41 JT |
6460 | |
6461 | /* Return the original error code */ | |
fe66a05a | 6462 | return ret; |
39279cc3 CM |
6463 | } |
6464 | ||
5f465bf1 OS |
6465 | static int btrfs_create_common(struct inode *dir, struct dentry *dentry, |
6466 | struct inode *inode) | |
618e21d5 | 6467 | { |
2ff7e61e | 6468 | struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb); |
618e21d5 | 6469 | struct btrfs_root *root = BTRFS_I(dir)->root; |
3538d68d OS |
6470 | struct btrfs_new_inode_args new_inode_args = { |
6471 | .dir = dir, | |
6472 | .dentry = dentry, | |
6473 | .inode = inode, | |
6474 | }; | |
6475 | unsigned int trans_num_items; | |
5f465bf1 | 6476 | struct btrfs_trans_handle *trans; |
618e21d5 | 6477 | int err; |
618e21d5 | 6478 | |
3538d68d | 6479 | err = btrfs_new_inode_prepare(&new_inode_args, &trans_num_items); |
caae78e0 OS |
6480 | if (err) |
6481 | goto out_inode; | |
3538d68d OS |
6482 | |
6483 | trans = btrfs_start_transaction(root, trans_num_items); | |
a1fd0c35 | 6484 | if (IS_ERR(trans)) { |
3538d68d OS |
6485 | err = PTR_ERR(trans); |
6486 | goto out_new_inode_args; | |
a1fd0c35 | 6487 | } |
1832a6d5 | 6488 | |
caae78e0 OS |
6489 | err = btrfs_create_new_inode(trans, &new_inode_args); |
6490 | if (!err) | |
6491 | d_instantiate_new(dentry, inode); | |
b0d5d10f | 6492 | |
3a45bb20 | 6493 | btrfs_end_transaction(trans); |
5f465bf1 | 6494 | btrfs_btree_balance_dirty(fs_info); |
3538d68d OS |
6495 | out_new_inode_args: |
6496 | btrfs_new_inode_args_destroy(&new_inode_args); | |
caae78e0 OS |
6497 | out_inode: |
6498 | if (err) | |
6499 | iput(inode); | |
618e21d5 JB |
6500 | return err; |
6501 | } | |
6502 | ||
5ebb29be | 6503 | static int btrfs_mknod(struct mnt_idmap *idmap, struct inode *dir, |
5f465bf1 OS |
6504 | struct dentry *dentry, umode_t mode, dev_t rdev) |
6505 | { | |
6506 | struct inode *inode; | |
6507 | ||
6508 | inode = new_inode(dir->i_sb); | |
6509 | if (!inode) | |
6510 | return -ENOMEM; | |
f2d40141 | 6511 | inode_init_owner(idmap, inode, dir, mode); |
5f465bf1 OS |
6512 | inode->i_op = &btrfs_special_inode_operations; |
6513 | init_special_inode(inode, inode->i_mode, rdev); | |
6514 | return btrfs_create_common(dir, dentry, inode); | |
6515 | } | |
6516 | ||
6c960e68 | 6517 | static int btrfs_create(struct mnt_idmap *idmap, struct inode *dir, |
549c7297 | 6518 | struct dentry *dentry, umode_t mode, bool excl) |
39279cc3 | 6519 | { |
a1fd0c35 | 6520 | struct inode *inode; |
39279cc3 | 6521 | |
a1fd0c35 OS |
6522 | inode = new_inode(dir->i_sb); |
6523 | if (!inode) | |
6524 | return -ENOMEM; | |
f2d40141 | 6525 | inode_init_owner(idmap, inode, dir, mode); |
a1fd0c35 OS |
6526 | inode->i_fop = &btrfs_file_operations; |
6527 | inode->i_op = &btrfs_file_inode_operations; | |
6528 | inode->i_mapping->a_ops = &btrfs_aops; | |
5f465bf1 | 6529 | return btrfs_create_common(dir, dentry, inode); |
39279cc3 CM |
6530 | } |
6531 | ||
6532 | static int btrfs_link(struct dentry *old_dentry, struct inode *dir, | |
6533 | struct dentry *dentry) | |
6534 | { | |
271dba45 | 6535 | struct btrfs_trans_handle *trans = NULL; |
39279cc3 | 6536 | struct btrfs_root *root = BTRFS_I(dir)->root; |
2b0143b5 | 6537 | struct inode *inode = d_inode(old_dentry); |
2ff7e61e | 6538 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
ab3c5c18 | 6539 | struct fscrypt_name fname; |
00e4e6b3 | 6540 | u64 index; |
39279cc3 CM |
6541 | int err; |
6542 | int drop_inode = 0; | |
6543 | ||
4a8be425 | 6544 | /* do not allow sys_link's with other subvols of the same device */ |
4fd786e6 | 6545 | if (root->root_key.objectid != BTRFS_I(inode)->root->root_key.objectid) |
3ab3564f | 6546 | return -EXDEV; |
4a8be425 | 6547 | |
f186373f | 6548 | if (inode->i_nlink >= BTRFS_LINK_MAX) |
c055e99e | 6549 | return -EMLINK; |
4a8be425 | 6550 | |
ab3c5c18 STD |
6551 | err = fscrypt_setup_filename(dir, &dentry->d_name, 0, &fname); |
6552 | if (err) | |
6553 | goto fail; | |
6554 | ||
877574e2 | 6555 | err = btrfs_set_inode_index(BTRFS_I(dir), &index); |
aec7477b JB |
6556 | if (err) |
6557 | goto fail; | |
6558 | ||
a22285a6 | 6559 | /* |
7e6b6465 | 6560 | * 2 items for inode and inode ref |
a22285a6 | 6561 | * 2 items for dir items |
7e6b6465 | 6562 | * 1 item for parent inode |
399b0bbf | 6563 | * 1 item for orphan item deletion if O_TMPFILE |
a22285a6 | 6564 | */ |
399b0bbf | 6565 | trans = btrfs_start_transaction(root, inode->i_nlink ? 5 : 6); |
a22285a6 YZ |
6566 | if (IS_ERR(trans)) { |
6567 | err = PTR_ERR(trans); | |
271dba45 | 6568 | trans = NULL; |
a22285a6 YZ |
6569 | goto fail; |
6570 | } | |
5f39d397 | 6571 | |
67de1176 MX |
6572 | /* There are several dir indexes for this inode, clear the cache. */ |
6573 | BTRFS_I(inode)->dir_index = 0ULL; | |
8b558c5f | 6574 | inc_nlink(inode); |
0c4d2d95 | 6575 | inode_inc_iversion(inode); |
2a9462de | 6576 | inode_set_ctime_current(inode); |
7de9c6ee | 6577 | ihold(inode); |
e9976151 | 6578 | set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags); |
aec7477b | 6579 | |
81512e89 | 6580 | err = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode), |
6db75318 | 6581 | &fname.disk_name, 1, index); |
5f39d397 | 6582 | |
a5719521 | 6583 | if (err) { |
54aa1f4d | 6584 | drop_inode = 1; |
a5719521 | 6585 | } else { |
10d9f309 | 6586 | struct dentry *parent = dentry->d_parent; |
d4682ba0 | 6587 | |
9a56fcd1 | 6588 | err = btrfs_update_inode(trans, root, BTRFS_I(inode)); |
79787eaa JM |
6589 | if (err) |
6590 | goto fail; | |
ef3b9af5 FM |
6591 | if (inode->i_nlink == 1) { |
6592 | /* | |
6593 | * If new hard link count is 1, it's a file created | |
6594 | * with open(2) O_TMPFILE flag. | |
6595 | */ | |
3d6ae7bb | 6596 | err = btrfs_orphan_del(trans, BTRFS_I(inode)); |
ef3b9af5 FM |
6597 | if (err) |
6598 | goto fail; | |
6599 | } | |
08c422c2 | 6600 | d_instantiate(dentry, inode); |
88d2beec | 6601 | btrfs_log_new_name(trans, old_dentry, NULL, 0, parent); |
a5719521 | 6602 | } |
39279cc3 | 6603 | |
1832a6d5 | 6604 | fail: |
ab3c5c18 | 6605 | fscrypt_free_filename(&fname); |
271dba45 | 6606 | if (trans) |
3a45bb20 | 6607 | btrfs_end_transaction(trans); |
39279cc3 CM |
6608 | if (drop_inode) { |
6609 | inode_dec_link_count(inode); | |
6610 | iput(inode); | |
6611 | } | |
2ff7e61e | 6612 | btrfs_btree_balance_dirty(fs_info); |
39279cc3 CM |
6613 | return err; |
6614 | } | |
6615 | ||
c54bd91e | 6616 | static int btrfs_mkdir(struct mnt_idmap *idmap, struct inode *dir, |
549c7297 | 6617 | struct dentry *dentry, umode_t mode) |
39279cc3 | 6618 | { |
a1fd0c35 | 6619 | struct inode *inode; |
39279cc3 | 6620 | |
a1fd0c35 OS |
6621 | inode = new_inode(dir->i_sb); |
6622 | if (!inode) | |
6623 | return -ENOMEM; | |
f2d40141 | 6624 | inode_init_owner(idmap, inode, dir, S_IFDIR | mode); |
a1fd0c35 OS |
6625 | inode->i_op = &btrfs_dir_inode_operations; |
6626 | inode->i_fop = &btrfs_dir_file_operations; | |
5f465bf1 | 6627 | return btrfs_create_common(dir, dentry, inode); |
39279cc3 CM |
6628 | } |
6629 | ||
c8b97818 | 6630 | static noinline int uncompress_inline(struct btrfs_path *path, |
e40da0e5 | 6631 | struct page *page, |
c8b97818 CM |
6632 | struct btrfs_file_extent_item *item) |
6633 | { | |
6634 | int ret; | |
6635 | struct extent_buffer *leaf = path->nodes[0]; | |
6636 | char *tmp; | |
6637 | size_t max_size; | |
6638 | unsigned long inline_size; | |
6639 | unsigned long ptr; | |
261507a0 | 6640 | int compress_type; |
c8b97818 | 6641 | |
261507a0 | 6642 | compress_type = btrfs_file_extent_compression(leaf, item); |
c8b97818 | 6643 | max_size = btrfs_file_extent_ram_bytes(leaf, item); |
437bd07e | 6644 | inline_size = btrfs_file_extent_inline_item_len(leaf, path->slots[0]); |
c8b97818 | 6645 | tmp = kmalloc(inline_size, GFP_NOFS); |
8d413713 TI |
6646 | if (!tmp) |
6647 | return -ENOMEM; | |
c8b97818 CM |
6648 | ptr = btrfs_file_extent_inline_start(item); |
6649 | ||
6650 | read_extent_buffer(leaf, tmp, ptr, inline_size); | |
6651 | ||
09cbfeaf | 6652 | max_size = min_t(unsigned long, PAGE_SIZE, max_size); |
a982fc82 | 6653 | ret = btrfs_decompress(compress_type, tmp, page, 0, inline_size, max_size); |
e1699d2d ZB |
6654 | |
6655 | /* | |
6656 | * decompression code contains a memset to fill in any space between the end | |
6657 | * of the uncompressed data and the end of max_size in case the decompressed | |
6658 | * data ends up shorter than ram_bytes. That doesn't cover the hole between | |
6659 | * the end of an inline extent and the beginning of the next block, so we | |
6660 | * cover that region here. | |
6661 | */ | |
6662 | ||
a982fc82 QW |
6663 | if (max_size < PAGE_SIZE) |
6664 | memzero_page(page, max_size, PAGE_SIZE - max_size); | |
c8b97818 | 6665 | kfree(tmp); |
166ae5a4 | 6666 | return ret; |
c8b97818 CM |
6667 | } |
6668 | ||
a982fc82 QW |
6669 | static int read_inline_extent(struct btrfs_inode *inode, struct btrfs_path *path, |
6670 | struct page *page) | |
6671 | { | |
6672 | struct btrfs_file_extent_item *fi; | |
6673 | void *kaddr; | |
6674 | size_t copy_size; | |
6675 | ||
6676 | if (!page || PageUptodate(page)) | |
6677 | return 0; | |
6678 | ||
6679 | ASSERT(page_offset(page) == 0); | |
6680 | ||
6681 | fi = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
6682 | struct btrfs_file_extent_item); | |
6683 | if (btrfs_file_extent_compression(path->nodes[0], fi) != BTRFS_COMPRESS_NONE) | |
6684 | return uncompress_inline(path, page, fi); | |
6685 | ||
6686 | copy_size = min_t(u64, PAGE_SIZE, | |
6687 | btrfs_file_extent_ram_bytes(path->nodes[0], fi)); | |
6688 | kaddr = kmap_local_page(page); | |
6689 | read_extent_buffer(path->nodes[0], kaddr, | |
6690 | btrfs_file_extent_inline_start(fi), copy_size); | |
6691 | kunmap_local(kaddr); | |
6692 | if (copy_size < PAGE_SIZE) | |
6693 | memzero_page(page, copy_size, PAGE_SIZE - copy_size); | |
6694 | return 0; | |
6695 | } | |
6696 | ||
43dd529a DS |
6697 | /* |
6698 | * Lookup the first extent overlapping a range in a file. | |
6699 | * | |
39b07b5d OS |
6700 | * @inode: file to search in |
6701 | * @page: page to read extent data into if the extent is inline | |
6702 | * @pg_offset: offset into @page to copy to | |
6703 | * @start: file offset | |
6704 | * @len: length of range starting at @start | |
6705 | * | |
43dd529a DS |
6706 | * Return the first &struct extent_map which overlaps the given range, reading |
6707 | * it from the B-tree and caching it if necessary. Note that there may be more | |
6708 | * extents which overlap the given range after the returned extent_map. | |
d352ac68 | 6709 | * |
39b07b5d OS |
6710 | * If @page is not NULL and the extent is inline, this also reads the extent |
6711 | * data directly into the page and marks the extent up to date in the io_tree. | |
6712 | * | |
6713 | * Return: ERR_PTR on error, non-NULL extent_map on success. | |
d352ac68 | 6714 | */ |
fc4f21b1 | 6715 | struct extent_map *btrfs_get_extent(struct btrfs_inode *inode, |
39b07b5d OS |
6716 | struct page *page, size_t pg_offset, |
6717 | u64 start, u64 len) | |
a52d9a80 | 6718 | { |
3ffbd68c | 6719 | struct btrfs_fs_info *fs_info = inode->root->fs_info; |
1028d1c4 | 6720 | int ret = 0; |
a52d9a80 CM |
6721 | u64 extent_start = 0; |
6722 | u64 extent_end = 0; | |
fc4f21b1 | 6723 | u64 objectid = btrfs_ino(inode); |
7e74e235 | 6724 | int extent_type = -1; |
f421950f | 6725 | struct btrfs_path *path = NULL; |
fc4f21b1 | 6726 | struct btrfs_root *root = inode->root; |
a52d9a80 | 6727 | struct btrfs_file_extent_item *item; |
5f39d397 CM |
6728 | struct extent_buffer *leaf; |
6729 | struct btrfs_key found_key; | |
a52d9a80 | 6730 | struct extent_map *em = NULL; |
fc4f21b1 | 6731 | struct extent_map_tree *em_tree = &inode->extent_tree; |
a52d9a80 | 6732 | |
890871be | 6733 | read_lock(&em_tree->lock); |
d1310b2e | 6734 | em = lookup_extent_mapping(em_tree, start, len); |
890871be | 6735 | read_unlock(&em_tree->lock); |
d1310b2e | 6736 | |
a52d9a80 | 6737 | if (em) { |
e1c4b745 CM |
6738 | if (em->start > start || em->start + em->len <= start) |
6739 | free_extent_map(em); | |
6740 | else if (em->block_start == EXTENT_MAP_INLINE && page) | |
70dec807 CM |
6741 | free_extent_map(em); |
6742 | else | |
6743 | goto out; | |
a52d9a80 | 6744 | } |
172ddd60 | 6745 | em = alloc_extent_map(); |
a52d9a80 | 6746 | if (!em) { |
1028d1c4 | 6747 | ret = -ENOMEM; |
d1310b2e | 6748 | goto out; |
a52d9a80 | 6749 | } |
d1310b2e | 6750 | em->start = EXTENT_MAP_HOLE; |
445a6944 | 6751 | em->orig_start = EXTENT_MAP_HOLE; |
d1310b2e | 6752 | em->len = (u64)-1; |
c8b97818 | 6753 | em->block_len = (u64)-1; |
f421950f | 6754 | |
bee6ec82 | 6755 | path = btrfs_alloc_path(); |
f421950f | 6756 | if (!path) { |
1028d1c4 | 6757 | ret = -ENOMEM; |
bee6ec82 | 6758 | goto out; |
f421950f CM |
6759 | } |
6760 | ||
bee6ec82 LB |
6761 | /* Chances are we'll be called again, so go ahead and do readahead */ |
6762 | path->reada = READA_FORWARD; | |
4d7240f0 JB |
6763 | |
6764 | /* | |
6765 | * The same explanation in load_free_space_cache applies here as well, | |
6766 | * we only read when we're loading the free space cache, and at that | |
6767 | * point the commit_root has everything we need. | |
6768 | */ | |
6769 | if (btrfs_is_free_space_inode(inode)) { | |
6770 | path->search_commit_root = 1; | |
6771 | path->skip_locking = 1; | |
6772 | } | |
51899412 | 6773 | |
5c9a702e | 6774 | ret = btrfs_lookup_file_extent(NULL, root, path, objectid, start, 0); |
a52d9a80 | 6775 | if (ret < 0) { |
a52d9a80 | 6776 | goto out; |
b8eeab7f | 6777 | } else if (ret > 0) { |
a52d9a80 CM |
6778 | if (path->slots[0] == 0) |
6779 | goto not_found; | |
6780 | path->slots[0]--; | |
1028d1c4 | 6781 | ret = 0; |
a52d9a80 CM |
6782 | } |
6783 | ||
5f39d397 CM |
6784 | leaf = path->nodes[0]; |
6785 | item = btrfs_item_ptr(leaf, path->slots[0], | |
a52d9a80 | 6786 | struct btrfs_file_extent_item); |
5f39d397 | 6787 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
5f39d397 | 6788 | if (found_key.objectid != objectid || |
694c12ed | 6789 | found_key.type != BTRFS_EXTENT_DATA_KEY) { |
25a50341 JB |
6790 | /* |
6791 | * If we backup past the first extent we want to move forward | |
6792 | * and see if there is an extent in front of us, otherwise we'll | |
6793 | * say there is a hole for our whole search range which can | |
6794 | * cause problems. | |
6795 | */ | |
6796 | extent_end = start; | |
6797 | goto next; | |
a52d9a80 CM |
6798 | } |
6799 | ||
694c12ed | 6800 | extent_type = btrfs_file_extent_type(leaf, item); |
5f39d397 | 6801 | extent_start = found_key.offset; |
a5eeb3d1 | 6802 | extent_end = btrfs_file_extent_end(path); |
694c12ed NB |
6803 | if (extent_type == BTRFS_FILE_EXTENT_REG || |
6804 | extent_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
6bf9e4bd QW |
6805 | /* Only regular file could have regular/prealloc extent */ |
6806 | if (!S_ISREG(inode->vfs_inode.i_mode)) { | |
1028d1c4 | 6807 | ret = -EUCLEAN; |
6bf9e4bd QW |
6808 | btrfs_crit(fs_info, |
6809 | "regular/prealloc extent found for non-regular inode %llu", | |
6810 | btrfs_ino(inode)); | |
6811 | goto out; | |
6812 | } | |
09ed2f16 LB |
6813 | trace_btrfs_get_extent_show_fi_regular(inode, leaf, item, |
6814 | extent_start); | |
694c12ed | 6815 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
09ed2f16 LB |
6816 | trace_btrfs_get_extent_show_fi_inline(inode, leaf, item, |
6817 | path->slots[0], | |
6818 | extent_start); | |
9036c102 | 6819 | } |
25a50341 | 6820 | next: |
9036c102 YZ |
6821 | if (start >= extent_end) { |
6822 | path->slots[0]++; | |
6823 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
6824 | ret = btrfs_next_leaf(root, path); | |
1028d1c4 | 6825 | if (ret < 0) |
9036c102 | 6826 | goto out; |
1028d1c4 | 6827 | else if (ret > 0) |
9036c102 | 6828 | goto not_found; |
1028d1c4 | 6829 | |
9036c102 | 6830 | leaf = path->nodes[0]; |
a52d9a80 | 6831 | } |
9036c102 YZ |
6832 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
6833 | if (found_key.objectid != objectid || | |
6834 | found_key.type != BTRFS_EXTENT_DATA_KEY) | |
6835 | goto not_found; | |
6836 | if (start + len <= found_key.offset) | |
6837 | goto not_found; | |
e2eca69d WS |
6838 | if (start > found_key.offset) |
6839 | goto next; | |
02a033df NB |
6840 | |
6841 | /* New extent overlaps with existing one */ | |
9036c102 | 6842 | em->start = start; |
70c8a91c | 6843 | em->orig_start = start; |
9036c102 | 6844 | em->len = found_key.offset - start; |
02a033df NB |
6845 | em->block_start = EXTENT_MAP_HOLE; |
6846 | goto insert; | |
9036c102 YZ |
6847 | } |
6848 | ||
280f15cb | 6849 | btrfs_extent_item_to_extent_map(inode, path, item, em); |
7ffbb598 | 6850 | |
694c12ed NB |
6851 | if (extent_type == BTRFS_FILE_EXTENT_REG || |
6852 | extent_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
a52d9a80 | 6853 | goto insert; |
694c12ed | 6854 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
affc5424 QW |
6855 | /* |
6856 | * Inline extent can only exist at file offset 0. This is | |
6857 | * ensured by tree-checker and inline extent creation path. | |
6858 | * Thus all members representing file offsets should be zero. | |
6859 | */ | |
affc5424 QW |
6860 | ASSERT(pg_offset == 0); |
6861 | ASSERT(extent_start == 0); | |
6862 | ASSERT(em->start == 0); | |
5f39d397 | 6863 | |
a196a894 QW |
6864 | /* |
6865 | * btrfs_extent_item_to_extent_map() should have properly | |
6866 | * initialized em members already. | |
6867 | * | |
6868 | * Other members are not utilized for inline extents. | |
6869 | */ | |
6870 | ASSERT(em->block_start == EXTENT_MAP_INLINE); | |
946c2923 | 6871 | ASSERT(em->len == fs_info->sectorsize); |
e49aabd9 | 6872 | |
a982fc82 QW |
6873 | ret = read_inline_extent(inode, path, page); |
6874 | if (ret < 0) | |
6875 | goto out; | |
a52d9a80 | 6876 | goto insert; |
a52d9a80 CM |
6877 | } |
6878 | not_found: | |
6879 | em->start = start; | |
70c8a91c | 6880 | em->orig_start = start; |
d1310b2e | 6881 | em->len = len; |
5f39d397 | 6882 | em->block_start = EXTENT_MAP_HOLE; |
a52d9a80 | 6883 | insert: |
1028d1c4 | 6884 | ret = 0; |
b3b4aa74 | 6885 | btrfs_release_path(path); |
d1310b2e | 6886 | if (em->start > start || extent_map_end(em) <= start) { |
0b246afa | 6887 | btrfs_err(fs_info, |
5d163e0e JM |
6888 | "bad extent! em: [%llu %llu] passed [%llu %llu]", |
6889 | em->start, em->len, start, len); | |
1028d1c4 | 6890 | ret = -EIO; |
a52d9a80 CM |
6891 | goto out; |
6892 | } | |
d1310b2e | 6893 | |
890871be | 6894 | write_lock(&em_tree->lock); |
1028d1c4 | 6895 | ret = btrfs_add_extent_mapping(fs_info, em_tree, &em, start, len); |
890871be | 6896 | write_unlock(&em_tree->lock); |
a52d9a80 | 6897 | out: |
c6414280 | 6898 | btrfs_free_path(path); |
1abe9b8a | 6899 | |
fc4f21b1 | 6900 | trace_btrfs_get_extent(root, inode, em); |
1abe9b8a | 6901 | |
1028d1c4 | 6902 | if (ret) { |
a52d9a80 | 6903 | free_extent_map(em); |
1028d1c4 | 6904 | return ERR_PTR(ret); |
a52d9a80 CM |
6905 | } |
6906 | return em; | |
6907 | } | |
6908 | ||
64f54188 | 6909 | static struct extent_map *btrfs_create_dio_extent(struct btrfs_inode *inode, |
53f2c206 | 6910 | struct btrfs_dio_data *dio_data, |
5f9a8a51 FM |
6911 | const u64 start, |
6912 | const u64 len, | |
6913 | const u64 orig_start, | |
6914 | const u64 block_start, | |
6915 | const u64 block_len, | |
6916 | const u64 orig_block_len, | |
6917 | const u64 ram_bytes, | |
6918 | const int type) | |
6919 | { | |
6920 | struct extent_map *em = NULL; | |
53f2c206 | 6921 | struct btrfs_ordered_extent *ordered; |
5f9a8a51 | 6922 | |
5f9a8a51 | 6923 | if (type != BTRFS_ORDERED_NOCOW) { |
64f54188 NB |
6924 | em = create_io_em(inode, start, len, orig_start, block_start, |
6925 | block_len, orig_block_len, ram_bytes, | |
6f9994db LB |
6926 | BTRFS_COMPRESS_NONE, /* compress_type */ |
6927 | type); | |
5f9a8a51 FM |
6928 | if (IS_ERR(em)) |
6929 | goto out; | |
6930 | } | |
53f2c206 BB |
6931 | ordered = btrfs_alloc_ordered_extent(inode, start, len, len, |
6932 | block_start, block_len, 0, | |
6933 | (1 << type) | | |
6934 | (1 << BTRFS_ORDERED_DIRECT), | |
6935 | BTRFS_COMPRESS_NONE); | |
6936 | if (IS_ERR(ordered)) { | |
5f9a8a51 FM |
6937 | if (em) { |
6938 | free_extent_map(em); | |
4c0c8cfc FM |
6939 | btrfs_drop_extent_map_range(inode, start, |
6940 | start + len - 1, false); | |
5f9a8a51 | 6941 | } |
53f2c206 BB |
6942 | em = ERR_CAST(ordered); |
6943 | } else { | |
6944 | ASSERT(!dio_data->ordered); | |
6945 | dio_data->ordered = ordered; | |
5f9a8a51 FM |
6946 | } |
6947 | out: | |
5f9a8a51 FM |
6948 | |
6949 | return em; | |
6950 | } | |
6951 | ||
9fc6f911 | 6952 | static struct extent_map *btrfs_new_extent_direct(struct btrfs_inode *inode, |
53f2c206 | 6953 | struct btrfs_dio_data *dio_data, |
4b46fce2 JB |
6954 | u64 start, u64 len) |
6955 | { | |
9fc6f911 NB |
6956 | struct btrfs_root *root = inode->root; |
6957 | struct btrfs_fs_info *fs_info = root->fs_info; | |
70c8a91c | 6958 | struct extent_map *em; |
4b46fce2 JB |
6959 | struct btrfs_key ins; |
6960 | u64 alloc_hint; | |
6961 | int ret; | |
4b46fce2 | 6962 | |
9fc6f911 | 6963 | alloc_hint = get_extent_allocation_hint(inode, start, len); |
0b246afa | 6964 | ret = btrfs_reserve_extent(root, len, len, fs_info->sectorsize, |
da17066c | 6965 | 0, alloc_hint, &ins, 1, 1); |
00361589 JB |
6966 | if (ret) |
6967 | return ERR_PTR(ret); | |
4b46fce2 | 6968 | |
53f2c206 | 6969 | em = btrfs_create_dio_extent(inode, dio_data, start, ins.offset, start, |
5f9a8a51 | 6970 | ins.objectid, ins.offset, ins.offset, |
6288d6ea | 6971 | ins.offset, BTRFS_ORDERED_REGULAR); |
0b246afa | 6972 | btrfs_dec_block_group_reservations(fs_info, ins.objectid); |
5f9a8a51 | 6973 | if (IS_ERR(em)) |
9fc6f911 NB |
6974 | btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, |
6975 | 1); | |
de0ee0ed | 6976 | |
4b46fce2 JB |
6977 | return em; |
6978 | } | |
6979 | ||
f4639636 | 6980 | static bool btrfs_extent_readonly(struct btrfs_fs_info *fs_info, u64 bytenr) |
05947ae1 AJ |
6981 | { |
6982 | struct btrfs_block_group *block_group; | |
f4639636 | 6983 | bool readonly = false; |
05947ae1 AJ |
6984 | |
6985 | block_group = btrfs_lookup_block_group(fs_info, bytenr); | |
6986 | if (!block_group || block_group->ro) | |
f4639636 | 6987 | readonly = true; |
05947ae1 AJ |
6988 | if (block_group) |
6989 | btrfs_put_block_group(block_group); | |
6990 | return readonly; | |
6991 | } | |
6992 | ||
46bfbb5c | 6993 | /* |
e4ecaf90 QW |
6994 | * Check if we can do nocow write into the range [@offset, @offset + @len) |
6995 | * | |
6996 | * @offset: File offset | |
6997 | * @len: The length to write, will be updated to the nocow writeable | |
6998 | * range | |
6999 | * @orig_start: (optional) Return the original file offset of the file extent | |
7000 | * @orig_len: (optional) Return the original on-disk length of the file extent | |
7001 | * @ram_bytes: (optional) Return the ram_bytes of the file extent | |
a84d5d42 BB |
7002 | * @strict: if true, omit optimizations that might force us into unnecessary |
7003 | * cow. e.g., don't trust generation number. | |
e4ecaf90 | 7004 | * |
e4ecaf90 QW |
7005 | * Return: |
7006 | * >0 and update @len if we can do nocow write | |
7007 | * 0 if we can't do nocow write | |
7008 | * <0 if error happened | |
7009 | * | |
7010 | * NOTE: This only checks the file extents, caller is responsible to wait for | |
7011 | * any ordered extents. | |
46bfbb5c | 7012 | */ |
00361589 | 7013 | noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len, |
7ee9e440 | 7014 | u64 *orig_start, u64 *orig_block_len, |
26ce9114 | 7015 | u64 *ram_bytes, bool nowait, bool strict) |
46bfbb5c | 7016 | { |
2ff7e61e | 7017 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
619104ba | 7018 | struct can_nocow_file_extent_args nocow_args = { 0 }; |
46bfbb5c CM |
7019 | struct btrfs_path *path; |
7020 | int ret; | |
7021 | struct extent_buffer *leaf; | |
7022 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
7b2b7085 | 7023 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
46bfbb5c CM |
7024 | struct btrfs_file_extent_item *fi; |
7025 | struct btrfs_key key; | |
46bfbb5c | 7026 | int found_type; |
e77751aa | 7027 | |
46bfbb5c CM |
7028 | path = btrfs_alloc_path(); |
7029 | if (!path) | |
7030 | return -ENOMEM; | |
26ce9114 | 7031 | path->nowait = nowait; |
46bfbb5c | 7032 | |
f85b7379 DS |
7033 | ret = btrfs_lookup_file_extent(NULL, root, path, |
7034 | btrfs_ino(BTRFS_I(inode)), offset, 0); | |
46bfbb5c CM |
7035 | if (ret < 0) |
7036 | goto out; | |
7037 | ||
46bfbb5c | 7038 | if (ret == 1) { |
619104ba | 7039 | if (path->slots[0] == 0) { |
46bfbb5c CM |
7040 | /* can't find the item, must cow */ |
7041 | ret = 0; | |
7042 | goto out; | |
7043 | } | |
619104ba | 7044 | path->slots[0]--; |
46bfbb5c CM |
7045 | } |
7046 | ret = 0; | |
7047 | leaf = path->nodes[0]; | |
619104ba | 7048 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
4a0cc7ca | 7049 | if (key.objectid != btrfs_ino(BTRFS_I(inode)) || |
46bfbb5c CM |
7050 | key.type != BTRFS_EXTENT_DATA_KEY) { |
7051 | /* not our file or wrong item type, must cow */ | |
7052 | goto out; | |
7053 | } | |
7054 | ||
7055 | if (key.offset > offset) { | |
7056 | /* Wrong offset, must cow */ | |
7057 | goto out; | |
7058 | } | |
7059 | ||
619104ba | 7060 | if (btrfs_file_extent_end(path) <= offset) |
7ee9e440 JB |
7061 | goto out; |
7062 | ||
619104ba FM |
7063 | fi = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_file_extent_item); |
7064 | found_type = btrfs_file_extent_type(leaf, fi); | |
7065 | if (ram_bytes) | |
7066 | *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi); | |
e77751aa | 7067 | |
619104ba FM |
7068 | nocow_args.start = offset; |
7069 | nocow_args.end = offset + *len - 1; | |
7070 | nocow_args.strict = strict; | |
7071 | nocow_args.free_path = true; | |
7ee9e440 | 7072 | |
619104ba FM |
7073 | ret = can_nocow_file_extent(path, &key, BTRFS_I(inode), &nocow_args); |
7074 | /* can_nocow_file_extent() has freed the path. */ | |
7075 | path = NULL; | |
7ee9e440 | 7076 | |
619104ba FM |
7077 | if (ret != 1) { |
7078 | /* Treat errors as not being able to NOCOW. */ | |
7079 | ret = 0; | |
78d4295b | 7080 | goto out; |
7ee9e440 | 7081 | } |
eb384b55 | 7082 | |
619104ba FM |
7083 | ret = 0; |
7084 | if (btrfs_extent_readonly(fs_info, nocow_args.disk_bytenr)) | |
46bfbb5c | 7085 | goto out; |
7b2b7085 | 7086 | |
619104ba FM |
7087 | if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) && |
7088 | found_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
7b2b7085 MX |
7089 | u64 range_end; |
7090 | ||
619104ba | 7091 | range_end = round_up(offset + nocow_args.num_bytes, |
da17066c | 7092 | root->fs_info->sectorsize) - 1; |
7b2b7085 MX |
7093 | ret = test_range_bit(io_tree, offset, range_end, |
7094 | EXTENT_DELALLOC, 0, NULL); | |
7095 | if (ret) { | |
7096 | ret = -EAGAIN; | |
7097 | goto out; | |
7098 | } | |
7099 | } | |
7100 | ||
619104ba FM |
7101 | if (orig_start) |
7102 | *orig_start = key.offset - nocow_args.extent_offset; | |
7103 | if (orig_block_len) | |
7104 | *orig_block_len = nocow_args.disk_num_bytes; | |
00361589 | 7105 | |
619104ba | 7106 | *len = nocow_args.num_bytes; |
46bfbb5c CM |
7107 | ret = 1; |
7108 | out: | |
7109 | btrfs_free_path(path); | |
7110 | return ret; | |
7111 | } | |
7112 | ||
eb838e73 | 7113 | static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend, |
59094403 FM |
7114 | struct extent_state **cached_state, |
7115 | unsigned int iomap_flags) | |
eb838e73 | 7116 | { |
59094403 FM |
7117 | const bool writing = (iomap_flags & IOMAP_WRITE); |
7118 | const bool nowait = (iomap_flags & IOMAP_NOWAIT); | |
7119 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; | |
eb838e73 JB |
7120 | struct btrfs_ordered_extent *ordered; |
7121 | int ret = 0; | |
7122 | ||
7123 | while (1) { | |
59094403 | 7124 | if (nowait) { |
83ae4133 JB |
7125 | if (!try_lock_extent(io_tree, lockstart, lockend, |
7126 | cached_state)) | |
59094403 FM |
7127 | return -EAGAIN; |
7128 | } else { | |
570eb97b | 7129 | lock_extent(io_tree, lockstart, lockend, cached_state); |
59094403 | 7130 | } |
eb838e73 JB |
7131 | /* |
7132 | * We're concerned with the entire range that we're going to be | |
01327610 | 7133 | * doing DIO to, so we need to make sure there's no ordered |
eb838e73 JB |
7134 | * extents in this range. |
7135 | */ | |
a776c6fa | 7136 | ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), lockstart, |
eb838e73 JB |
7137 | lockend - lockstart + 1); |
7138 | ||
7139 | /* | |
7140 | * We need to make sure there are no buffered pages in this | |
7141 | * range either, we could have raced between the invalidate in | |
7142 | * generic_file_direct_write and locking the extent. The | |
7143 | * invalidate needs to happen so that reads after a write do not | |
7144 | * get stale data. | |
7145 | */ | |
fc4adbff | 7146 | if (!ordered && |
051c98eb DS |
7147 | (!writing || !filemap_range_has_page(inode->i_mapping, |
7148 | lockstart, lockend))) | |
eb838e73 JB |
7149 | break; |
7150 | ||
570eb97b | 7151 | unlock_extent(io_tree, lockstart, lockend, cached_state); |
eb838e73 JB |
7152 | |
7153 | if (ordered) { | |
59094403 FM |
7154 | if (nowait) { |
7155 | btrfs_put_ordered_extent(ordered); | |
7156 | ret = -EAGAIN; | |
7157 | break; | |
7158 | } | |
ade77029 FM |
7159 | /* |
7160 | * If we are doing a DIO read and the ordered extent we | |
7161 | * found is for a buffered write, we can not wait for it | |
7162 | * to complete and retry, because if we do so we can | |
7163 | * deadlock with concurrent buffered writes on page | |
7164 | * locks. This happens only if our DIO read covers more | |
7165 | * than one extent map, if at this point has already | |
7166 | * created an ordered extent for a previous extent map | |
7167 | * and locked its range in the inode's io tree, and a | |
7168 | * concurrent write against that previous extent map's | |
7169 | * range and this range started (we unlock the ranges | |
7170 | * in the io tree only when the bios complete and | |
7171 | * buffered writes always lock pages before attempting | |
7172 | * to lock range in the io tree). | |
7173 | */ | |
7174 | if (writing || | |
7175 | test_bit(BTRFS_ORDERED_DIRECT, &ordered->flags)) | |
36d45567 | 7176 | btrfs_start_ordered_extent(ordered); |
ade77029 | 7177 | else |
59094403 | 7178 | ret = nowait ? -EAGAIN : -ENOTBLK; |
eb838e73 JB |
7179 | btrfs_put_ordered_extent(ordered); |
7180 | } else { | |
eb838e73 | 7181 | /* |
b850ae14 FM |
7182 | * We could trigger writeback for this range (and wait |
7183 | * for it to complete) and then invalidate the pages for | |
7184 | * this range (through invalidate_inode_pages2_range()), | |
7185 | * but that can lead us to a deadlock with a concurrent | |
ba206a02 | 7186 | * call to readahead (a buffered read or a defrag call |
b850ae14 FM |
7187 | * triggered a readahead) on a page lock due to an |
7188 | * ordered dio extent we created before but did not have | |
7189 | * yet a corresponding bio submitted (whence it can not | |
ba206a02 | 7190 | * complete), which makes readahead wait for that |
b850ae14 FM |
7191 | * ordered extent to complete while holding a lock on |
7192 | * that page. | |
eb838e73 | 7193 | */ |
59094403 | 7194 | ret = nowait ? -EAGAIN : -ENOTBLK; |
eb838e73 JB |
7195 | } |
7196 | ||
ade77029 FM |
7197 | if (ret) |
7198 | break; | |
7199 | ||
eb838e73 JB |
7200 | cond_resched(); |
7201 | } | |
7202 | ||
7203 | return ret; | |
7204 | } | |
7205 | ||
6f9994db | 7206 | /* The callers of this must take lock_extent() */ |
4b67c11d NB |
7207 | static struct extent_map *create_io_em(struct btrfs_inode *inode, u64 start, |
7208 | u64 len, u64 orig_start, u64 block_start, | |
6f9994db LB |
7209 | u64 block_len, u64 orig_block_len, |
7210 | u64 ram_bytes, int compress_type, | |
7211 | int type) | |
69ffb543 | 7212 | { |
69ffb543 | 7213 | struct extent_map *em; |
69ffb543 JB |
7214 | int ret; |
7215 | ||
6f9994db LB |
7216 | ASSERT(type == BTRFS_ORDERED_PREALLOC || |
7217 | type == BTRFS_ORDERED_COMPRESSED || | |
7218 | type == BTRFS_ORDERED_NOCOW || | |
1af4a0aa | 7219 | type == BTRFS_ORDERED_REGULAR); |
6f9994db | 7220 | |
69ffb543 JB |
7221 | em = alloc_extent_map(); |
7222 | if (!em) | |
7223 | return ERR_PTR(-ENOMEM); | |
7224 | ||
7225 | em->start = start; | |
7226 | em->orig_start = orig_start; | |
7227 | em->len = len; | |
7228 | em->block_len = block_len; | |
7229 | em->block_start = block_start; | |
b4939680 | 7230 | em->orig_block_len = orig_block_len; |
cc95bef6 | 7231 | em->ram_bytes = ram_bytes; |
70c8a91c | 7232 | em->generation = -1; |
69ffb543 | 7233 | set_bit(EXTENT_FLAG_PINNED, &em->flags); |
1af4a0aa | 7234 | if (type == BTRFS_ORDERED_PREALLOC) { |
b11e234d | 7235 | set_bit(EXTENT_FLAG_FILLING, &em->flags); |
1af4a0aa | 7236 | } else if (type == BTRFS_ORDERED_COMPRESSED) { |
6f9994db LB |
7237 | set_bit(EXTENT_FLAG_COMPRESSED, &em->flags); |
7238 | em->compress_type = compress_type; | |
7239 | } | |
69ffb543 | 7240 | |
a1ba4c08 | 7241 | ret = btrfs_replace_extent_map_range(inode, em, true); |
69ffb543 JB |
7242 | if (ret) { |
7243 | free_extent_map(em); | |
7244 | return ERR_PTR(ret); | |
7245 | } | |
7246 | ||
6f9994db | 7247 | /* em got 2 refs now, callers needs to do free_extent_map once. */ |
69ffb543 JB |
7248 | return em; |
7249 | } | |
7250 | ||
1c8d0175 | 7251 | |
c5794e51 | 7252 | static int btrfs_get_blocks_direct_write(struct extent_map **map, |
c5794e51 NB |
7253 | struct inode *inode, |
7254 | struct btrfs_dio_data *dio_data, | |
7833b865 | 7255 | u64 start, u64 *lenp, |
d7a8ab4e | 7256 | unsigned int iomap_flags) |
c5794e51 | 7257 | { |
d4135134 | 7258 | const bool nowait = (iomap_flags & IOMAP_NOWAIT); |
c5794e51 NB |
7259 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
7260 | struct extent_map *em = *map; | |
f0bfa76a FM |
7261 | int type; |
7262 | u64 block_start, orig_start, orig_block_len, ram_bytes; | |
2306e83e | 7263 | struct btrfs_block_group *bg; |
f0bfa76a FM |
7264 | bool can_nocow = false; |
7265 | bool space_reserved = false; | |
7833b865 | 7266 | u64 len = *lenp; |
6d82ad13 | 7267 | u64 prev_len; |
c5794e51 NB |
7268 | int ret = 0; |
7269 | ||
7270 | /* | |
7271 | * We don't allocate a new extent in the following cases | |
7272 | * | |
7273 | * 1) The inode is marked as NODATACOW. In this case we'll just use the | |
7274 | * existing extent. | |
7275 | * 2) The extent is marked as PREALLOC. We're good to go here and can | |
7276 | * just use the extent. | |
7277 | * | |
7278 | */ | |
7279 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) || | |
7280 | ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) && | |
7281 | em->block_start != EXTENT_MAP_HOLE)) { | |
c5794e51 NB |
7282 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) |
7283 | type = BTRFS_ORDERED_PREALLOC; | |
7284 | else | |
7285 | type = BTRFS_ORDERED_NOCOW; | |
7286 | len = min(len, em->len - (start - em->start)); | |
7287 | block_start = em->block_start + (start - em->start); | |
7288 | ||
7289 | if (can_nocow_extent(inode, start, &len, &orig_start, | |
26ce9114 | 7290 | &orig_block_len, &ram_bytes, false, false) == 1) { |
2306e83e FM |
7291 | bg = btrfs_inc_nocow_writers(fs_info, block_start); |
7292 | if (bg) | |
7293 | can_nocow = true; | |
7294 | } | |
f0bfa76a | 7295 | } |
c5794e51 | 7296 | |
6d82ad13 | 7297 | prev_len = len; |
f0bfa76a FM |
7298 | if (can_nocow) { |
7299 | struct extent_map *em2; | |
7300 | ||
7301 | /* We can NOCOW, so only need to reserve metadata space. */ | |
d4135134 FM |
7302 | ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode), len, len, |
7303 | nowait); | |
f0bfa76a FM |
7304 | if (ret < 0) { |
7305 | /* Our caller expects us to free the input extent map. */ | |
7306 | free_extent_map(em); | |
7307 | *map = NULL; | |
2306e83e | 7308 | btrfs_dec_nocow_writers(bg); |
d4135134 FM |
7309 | if (nowait && (ret == -ENOSPC || ret == -EDQUOT)) |
7310 | ret = -EAGAIN; | |
f0bfa76a FM |
7311 | goto out; |
7312 | } | |
7313 | space_reserved = true; | |
7314 | ||
53f2c206 | 7315 | em2 = btrfs_create_dio_extent(BTRFS_I(inode), dio_data, start, len, |
f0bfa76a FM |
7316 | orig_start, block_start, |
7317 | len, orig_block_len, | |
7318 | ram_bytes, type); | |
2306e83e | 7319 | btrfs_dec_nocow_writers(bg); |
f0bfa76a FM |
7320 | if (type == BTRFS_ORDERED_PREALLOC) { |
7321 | free_extent_map(em); | |
c1867eb3 DS |
7322 | *map = em2; |
7323 | em = em2; | |
f0bfa76a | 7324 | } |
c5794e51 | 7325 | |
f0bfa76a FM |
7326 | if (IS_ERR(em2)) { |
7327 | ret = PTR_ERR(em2); | |
7328 | goto out; | |
c5794e51 | 7329 | } |
f5585f4f FM |
7330 | |
7331 | dio_data->nocow_done = true; | |
f0bfa76a | 7332 | } else { |
f0bfa76a FM |
7333 | /* Our caller expects us to free the input extent map. */ |
7334 | free_extent_map(em); | |
7335 | *map = NULL; | |
7336 | ||
7833b865 CH |
7337 | if (nowait) { |
7338 | ret = -EAGAIN; | |
7339 | goto out; | |
7340 | } | |
d7a8ab4e | 7341 | |
f5585f4f FM |
7342 | /* |
7343 | * If we could not allocate data space before locking the file | |
7344 | * range and we can't do a NOCOW write, then we have to fail. | |
7345 | */ | |
7833b865 CH |
7346 | if (!dio_data->data_space_reserved) { |
7347 | ret = -ENOSPC; | |
7348 | goto out; | |
7349 | } | |
f5585f4f FM |
7350 | |
7351 | /* | |
7352 | * We have to COW and we have already reserved data space before, | |
7353 | * so now we reserve only metadata. | |
7354 | */ | |
7355 | ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode), len, len, | |
7356 | false); | |
f0bfa76a FM |
7357 | if (ret < 0) |
7358 | goto out; | |
7359 | space_reserved = true; | |
7360 | ||
53f2c206 | 7361 | em = btrfs_new_extent_direct(BTRFS_I(inode), dio_data, start, len); |
f0bfa76a FM |
7362 | if (IS_ERR(em)) { |
7363 | ret = PTR_ERR(em); | |
7364 | goto out; | |
7365 | } | |
7366 | *map = em; | |
7367 | len = min(len, em->len - (start - em->start)); | |
7368 | if (len < prev_len) | |
f5585f4f FM |
7369 | btrfs_delalloc_release_metadata(BTRFS_I(inode), |
7370 | prev_len - len, true); | |
c5794e51 NB |
7371 | } |
7372 | ||
f0bfa76a FM |
7373 | /* |
7374 | * We have created our ordered extent, so we can now release our reservation | |
7375 | * for an outstanding extent. | |
7376 | */ | |
6d82ad13 | 7377 | btrfs_delalloc_release_extents(BTRFS_I(inode), prev_len); |
c5794e51 | 7378 | |
c5794e51 NB |
7379 | /* |
7380 | * Need to update the i_size under the extent lock so buffered | |
7381 | * readers will get the updated i_size when we unlock. | |
7382 | */ | |
f85781fb | 7383 | if (start + len > i_size_read(inode)) |
c5794e51 | 7384 | i_size_write(inode, start + len); |
c5794e51 | 7385 | out: |
f0bfa76a FM |
7386 | if (ret && space_reserved) { |
7387 | btrfs_delalloc_release_extents(BTRFS_I(inode), len); | |
f5585f4f | 7388 | btrfs_delalloc_release_metadata(BTRFS_I(inode), len, true); |
f0bfa76a | 7389 | } |
7833b865 | 7390 | *lenp = len; |
c5794e51 NB |
7391 | return ret; |
7392 | } | |
7393 | ||
f85781fb GR |
7394 | static int btrfs_dio_iomap_begin(struct inode *inode, loff_t start, |
7395 | loff_t length, unsigned int flags, struct iomap *iomap, | |
7396 | struct iomap *srcmap) | |
4b46fce2 | 7397 | { |
491a6d01 | 7398 | struct iomap_iter *iter = container_of(iomap, struct iomap_iter, iomap); |
0b246afa | 7399 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
4b46fce2 | 7400 | struct extent_map *em; |
eb838e73 | 7401 | struct extent_state *cached_state = NULL; |
491a6d01 | 7402 | struct btrfs_dio_data *dio_data = iter->private; |
eb838e73 | 7403 | u64 lockstart, lockend; |
f85781fb | 7404 | const bool write = !!(flags & IOMAP_WRITE); |
0934856d | 7405 | int ret = 0; |
f85781fb | 7406 | u64 len = length; |
f5585f4f | 7407 | const u64 data_alloc_len = length; |
f85781fb | 7408 | bool unlock_extents = false; |
eb838e73 | 7409 | |
79d3d1d1 JB |
7410 | /* |
7411 | * We could potentially fault if we have a buffer > PAGE_SIZE, and if | |
7412 | * we're NOWAIT we may submit a bio for a partial range and return | |
7413 | * EIOCBQUEUED, which would result in an errant short read. | |
7414 | * | |
7415 | * The best way to handle this would be to allow for partial completions | |
7416 | * of iocb's, so we could submit the partial bio, return and fault in | |
7417 | * the rest of the pages, and then submit the io for the rest of the | |
7418 | * range. However we don't have that currently, so simply return | |
7419 | * -EAGAIN at this point so that the normal path is used. | |
7420 | */ | |
7421 | if (!write && (flags & IOMAP_NOWAIT) && length > PAGE_SIZE) | |
7422 | return -EAGAIN; | |
7423 | ||
ee5b46a3 CH |
7424 | /* |
7425 | * Cap the size of reads to that usually seen in buffered I/O as we need | |
7426 | * to allocate a contiguous array for the checksums. | |
7427 | */ | |
f85781fb | 7428 | if (!write) |
ee5b46a3 | 7429 | len = min_t(u64, len, fs_info->sectorsize * BTRFS_MAX_BIO_SECTORS); |
eb838e73 | 7430 | |
c329861d JB |
7431 | lockstart = start; |
7432 | lockend = start + len - 1; | |
7433 | ||
f85781fb | 7434 | /* |
b023e675 FM |
7435 | * iomap_dio_rw() only does filemap_write_and_wait_range(), which isn't |
7436 | * enough if we've written compressed pages to this area, so we need to | |
7437 | * flush the dirty pages again to make absolutely sure that any | |
7438 | * outstanding dirty pages are on disk - the first flush only starts | |
7439 | * compression on the data, while keeping the pages locked, so by the | |
7440 | * time the second flush returns we know bios for the compressed pages | |
7441 | * were submitted and finished, and the pages no longer under writeback. | |
7442 | * | |
7443 | * If we have a NOWAIT request and we have any pages in the range that | |
7444 | * are locked, likely due to compression still in progress, we don't want | |
7445 | * to block on page locks. We also don't want to block on pages marked as | |
7446 | * dirty or under writeback (same as for the non-compression case). | |
7447 | * iomap_dio_rw() did the same check, but after that and before we got | |
7448 | * here, mmap'ed writes may have happened or buffered reads started | |
7449 | * (readpage() and readahead(), which lock pages), as we haven't locked | |
7450 | * the file range yet. | |
f85781fb GR |
7451 | */ |
7452 | if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT, | |
7453 | &BTRFS_I(inode)->runtime_flags)) { | |
b023e675 FM |
7454 | if (flags & IOMAP_NOWAIT) { |
7455 | if (filemap_range_needs_writeback(inode->i_mapping, | |
7456 | lockstart, lockend)) | |
7457 | return -EAGAIN; | |
7458 | } else { | |
7459 | ret = filemap_fdatawrite_range(inode->i_mapping, start, | |
7460 | start + length - 1); | |
7461 | if (ret) | |
7462 | return ret; | |
7463 | } | |
f85781fb GR |
7464 | } |
7465 | ||
491a6d01 | 7466 | memset(dio_data, 0, sizeof(*dio_data)); |
f85781fb | 7467 | |
f5585f4f FM |
7468 | /* |
7469 | * We always try to allocate data space and must do it before locking | |
7470 | * the file range, to avoid deadlocks with concurrent writes to the same | |
7471 | * range if the range has several extents and the writes don't expand the | |
7472 | * current i_size (the inode lock is taken in shared mode). If we fail to | |
7473 | * allocate data space here we continue and later, after locking the | |
7474 | * file range, we fail with ENOSPC only if we figure out we can not do a | |
7475 | * NOCOW write. | |
7476 | */ | |
7477 | if (write && !(flags & IOMAP_NOWAIT)) { | |
7478 | ret = btrfs_check_data_free_space(BTRFS_I(inode), | |
7479 | &dio_data->data_reserved, | |
1daedb1d | 7480 | start, data_alloc_len, false); |
f5585f4f FM |
7481 | if (!ret) |
7482 | dio_data->data_space_reserved = true; | |
7483 | else if (ret && !(BTRFS_I(inode)->flags & | |
7484 | (BTRFS_INODE_NODATACOW | BTRFS_INODE_PREALLOC))) | |
7485 | goto err; | |
7486 | } | |
e1cbbfa5 | 7487 | |
eb838e73 JB |
7488 | /* |
7489 | * If this errors out it's because we couldn't invalidate pagecache for | |
59094403 FM |
7490 | * this range and we need to fallback to buffered IO, or we are doing a |
7491 | * NOWAIT read/write and we need to block. | |
eb838e73 | 7492 | */ |
59094403 FM |
7493 | ret = lock_extent_direct(inode, lockstart, lockend, &cached_state, flags); |
7494 | if (ret < 0) | |
9c9464cc | 7495 | goto err; |
eb838e73 | 7496 | |
39b07b5d | 7497 | em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len); |
eb838e73 JB |
7498 | if (IS_ERR(em)) { |
7499 | ret = PTR_ERR(em); | |
7500 | goto unlock_err; | |
7501 | } | |
4b46fce2 JB |
7502 | |
7503 | /* | |
7504 | * Ok for INLINE and COMPRESSED extents we need to fallback on buffered | |
7505 | * io. INLINE is special, and we could probably kludge it in here, but | |
7506 | * it's still buffered so for safety lets just fall back to the generic | |
7507 | * buffered path. | |
7508 | * | |
7509 | * For COMPRESSED we _have_ to read the entire extent in so we can | |
7510 | * decompress it, so there will be buffering required no matter what we | |
7511 | * do, so go ahead and fallback to buffered. | |
7512 | * | |
01327610 | 7513 | * We return -ENOTBLK because that's what makes DIO go ahead and go back |
4b46fce2 JB |
7514 | * to buffered IO. Don't blame me, this is the price we pay for using |
7515 | * the generic code. | |
7516 | */ | |
7517 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) || | |
7518 | em->block_start == EXTENT_MAP_INLINE) { | |
7519 | free_extent_map(em); | |
a4527e18 FM |
7520 | /* |
7521 | * If we are in a NOWAIT context, return -EAGAIN in order to | |
7522 | * fallback to buffered IO. This is not only because we can | |
7523 | * block with buffered IO (no support for NOWAIT semantics at | |
7524 | * the moment) but also to avoid returning short reads to user | |
7525 | * space - this happens if we were able to read some data from | |
7526 | * previous non-compressed extents and then when we fallback to | |
7527 | * buffered IO, at btrfs_file_read_iter() by calling | |
7528 | * filemap_read(), we fail to fault in pages for the read buffer, | |
7529 | * in which case filemap_read() returns a short read (the number | |
7530 | * of bytes previously read is > 0, so it does not return -EFAULT). | |
7531 | */ | |
7532 | ret = (flags & IOMAP_NOWAIT) ? -EAGAIN : -ENOTBLK; | |
eb838e73 | 7533 | goto unlock_err; |
4b46fce2 JB |
7534 | } |
7535 | ||
f85781fb | 7536 | len = min(len, em->len - (start - em->start)); |
ca93e44b FM |
7537 | |
7538 | /* | |
7539 | * If we have a NOWAIT request and the range contains multiple extents | |
7540 | * (or a mix of extents and holes), then we return -EAGAIN to make the | |
7541 | * caller fallback to a context where it can do a blocking (without | |
7542 | * NOWAIT) request. This way we avoid doing partial IO and returning | |
7543 | * success to the caller, which is not optimal for writes and for reads | |
7544 | * it can result in unexpected behaviour for an application. | |
7545 | * | |
7546 | * When doing a read, because we use IOMAP_DIO_PARTIAL when calling | |
7547 | * iomap_dio_rw(), we can end up returning less data then what the caller | |
7548 | * asked for, resulting in an unexpected, and incorrect, short read. | |
7549 | * That is, the caller asked to read N bytes and we return less than that, | |
7550 | * which is wrong unless we are crossing EOF. This happens if we get a | |
7551 | * page fault error when trying to fault in pages for the buffer that is | |
7552 | * associated to the struct iov_iter passed to iomap_dio_rw(), and we | |
7553 | * have previously submitted bios for other extents in the range, in | |
7554 | * which case iomap_dio_rw() may return us EIOCBQUEUED if not all of | |
7555 | * those bios have completed by the time we get the page fault error, | |
7556 | * which we return back to our caller - we should only return EIOCBQUEUED | |
7557 | * after we have submitted bios for all the extents in the range. | |
7558 | */ | |
7559 | if ((flags & IOMAP_NOWAIT) && len < length) { | |
7560 | free_extent_map(em); | |
7561 | ret = -EAGAIN; | |
7562 | goto unlock_err; | |
7563 | } | |
7564 | ||
f85781fb GR |
7565 | if (write) { |
7566 | ret = btrfs_get_blocks_direct_write(&em, inode, dio_data, | |
7833b865 | 7567 | start, &len, flags); |
c5794e51 NB |
7568 | if (ret < 0) |
7569 | goto unlock_err; | |
f85781fb GR |
7570 | unlock_extents = true; |
7571 | /* Recalc len in case the new em is smaller than requested */ | |
7572 | len = min(len, em->len - (start - em->start)); | |
f5585f4f FM |
7573 | if (dio_data->data_space_reserved) { |
7574 | u64 release_offset; | |
7575 | u64 release_len = 0; | |
7576 | ||
7577 | if (dio_data->nocow_done) { | |
7578 | release_offset = start; | |
7579 | release_len = data_alloc_len; | |
7580 | } else if (len < data_alloc_len) { | |
7581 | release_offset = start + len; | |
7582 | release_len = data_alloc_len - len; | |
7583 | } | |
7584 | ||
7585 | if (release_len > 0) | |
7586 | btrfs_free_reserved_data_space(BTRFS_I(inode), | |
7587 | dio_data->data_reserved, | |
7588 | release_offset, | |
7589 | release_len); | |
7590 | } | |
c5794e51 | 7591 | } else { |
1c8d0175 NB |
7592 | /* |
7593 | * We need to unlock only the end area that we aren't using. | |
7594 | * The rest is going to be unlocked by the endio routine. | |
7595 | */ | |
f85781fb GR |
7596 | lockstart = start + len; |
7597 | if (lockstart < lockend) | |
7598 | unlock_extents = true; | |
7599 | } | |
7600 | ||
7601 | if (unlock_extents) | |
570eb97b JB |
7602 | unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend, |
7603 | &cached_state); | |
f85781fb GR |
7604 | else |
7605 | free_extent_state(cached_state); | |
7606 | ||
7607 | /* | |
7608 | * Translate extent map information to iomap. | |
7609 | * We trim the extents (and move the addr) even though iomap code does | |
7610 | * that, since we have locked only the parts we are performing I/O in. | |
7611 | */ | |
7612 | if ((em->block_start == EXTENT_MAP_HOLE) || | |
7613 | (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) && !write)) { | |
7614 | iomap->addr = IOMAP_NULL_ADDR; | |
7615 | iomap->type = IOMAP_HOLE; | |
7616 | } else { | |
7617 | iomap->addr = em->block_start + (start - em->start); | |
7618 | iomap->type = IOMAP_MAPPED; | |
a43a67a2 | 7619 | } |
f85781fb | 7620 | iomap->offset = start; |
d24fa5c1 | 7621 | iomap->bdev = fs_info->fs_devices->latest_dev->bdev; |
f85781fb | 7622 | iomap->length = len; |
4b46fce2 JB |
7623 | free_extent_map(em); |
7624 | ||
7625 | return 0; | |
eb838e73 JB |
7626 | |
7627 | unlock_err: | |
570eb97b JB |
7628 | unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend, |
7629 | &cached_state); | |
9c9464cc | 7630 | err: |
f5585f4f FM |
7631 | if (dio_data->data_space_reserved) { |
7632 | btrfs_free_reserved_data_space(BTRFS_I(inode), | |
7633 | dio_data->data_reserved, | |
7634 | start, data_alloc_len); | |
7635 | extent_changeset_free(dio_data->data_reserved); | |
7636 | } | |
7637 | ||
f85781fb GR |
7638 | return ret; |
7639 | } | |
7640 | ||
7641 | static int btrfs_dio_iomap_end(struct inode *inode, loff_t pos, loff_t length, | |
7642 | ssize_t written, unsigned int flags, struct iomap *iomap) | |
7643 | { | |
491a6d01 CH |
7644 | struct iomap_iter *iter = container_of(iomap, struct iomap_iter, iomap); |
7645 | struct btrfs_dio_data *dio_data = iter->private; | |
f85781fb GR |
7646 | size_t submitted = dio_data->submitted; |
7647 | const bool write = !!(flags & IOMAP_WRITE); | |
491a6d01 | 7648 | int ret = 0; |
f85781fb GR |
7649 | |
7650 | if (!write && (iomap->type == IOMAP_HOLE)) { | |
7651 | /* If reading from a hole, unlock and return */ | |
570eb97b JB |
7652 | unlock_extent(&BTRFS_I(inode)->io_tree, pos, pos + length - 1, |
7653 | NULL); | |
491a6d01 | 7654 | return 0; |
f85781fb GR |
7655 | } |
7656 | ||
7657 | if (submitted < length) { | |
7658 | pos += submitted; | |
7659 | length -= submitted; | |
7660 | if (write) | |
b41b6f69 CH |
7661 | btrfs_finish_ordered_extent(dio_data->ordered, NULL, |
7662 | pos, length, false); | |
f85781fb GR |
7663 | else |
7664 | unlock_extent(&BTRFS_I(inode)->io_tree, pos, | |
570eb97b | 7665 | pos + length - 1, NULL); |
f85781fb GR |
7666 | ret = -ENOTBLK; |
7667 | } | |
53f2c206 BB |
7668 | if (write) { |
7669 | btrfs_put_ordered_extent(dio_data->ordered); | |
7670 | dio_data->ordered = NULL; | |
7671 | } | |
f85781fb | 7672 | |
f0bfa76a | 7673 | if (write) |
f85781fb | 7674 | extent_changeset_free(dio_data->data_reserved); |
8b110e39 MX |
7675 | return ret; |
7676 | } | |
7677 | ||
67d66982 | 7678 | static void btrfs_dio_end_io(struct btrfs_bio *bbio) |
8b110e39 | 7679 | { |
67d66982 CH |
7680 | struct btrfs_dio_private *dip = |
7681 | container_of(bbio, struct btrfs_dio_private, bbio); | |
7682 | struct btrfs_inode *inode = bbio->inode; | |
917f32a2 | 7683 | struct bio *bio = &bbio->bio; |
e65e1535 | 7684 | |
67d66982 CH |
7685 | if (bio->bi_status) { |
7686 | btrfs_warn(inode->root->fs_info, | |
7687 | "direct IO failed ino %llu op 0x%0x offset %#llx len %u err no %d", | |
7688 | btrfs_ino(inode), bio->bi_opf, | |
7689 | dip->file_offset, dip->bytes, bio->bi_status); | |
7609afac | 7690 | } |
1ae39938 | 7691 | |
b41b6f69 CH |
7692 | if (btrfs_op(bio) == BTRFS_MAP_WRITE) { |
7693 | btrfs_finish_ordered_extent(bbio->ordered, NULL, | |
7694 | dip->file_offset, dip->bytes, | |
7695 | !bio->bi_status); | |
7696 | } else { | |
67d66982 CH |
7697 | unlock_extent(&inode->io_tree, dip->file_offset, |
7698 | dip->file_offset + dip->bytes - 1, NULL); | |
b41b6f69 | 7699 | } |
ea1f0ced | 7700 | |
67d66982 CH |
7701 | bbio->bio.bi_private = bbio->private; |
7702 | iomap_dio_bio_end_io(bio); | |
e65e1535 MX |
7703 | } |
7704 | ||
67d66982 CH |
7705 | static void btrfs_dio_submit_io(const struct iomap_iter *iter, struct bio *bio, |
7706 | loff_t file_offset) | |
c36cac28 | 7707 | { |
67d66982 | 7708 | struct btrfs_bio *bbio = btrfs_bio(bio); |
642c5d34 | 7709 | struct btrfs_dio_private *dip = |
67d66982 | 7710 | container_of(bbio, struct btrfs_dio_private, bbio); |
491a6d01 | 7711 | struct btrfs_dio_data *dio_data = iter->private; |
544d24f9 | 7712 | |
4317ff00 QW |
7713 | btrfs_bio_init(bbio, BTRFS_I(iter->inode)->root->fs_info, |
7714 | btrfs_dio_end_io, bio->bi_private); | |
7715 | bbio->inode = BTRFS_I(iter->inode); | |
67d66982 | 7716 | bbio->file_offset = file_offset; |
e65e1535 | 7717 | |
67d66982 CH |
7718 | dip->file_offset = file_offset; |
7719 | dip->bytes = bio->bi_iter.bi_size; | |
e65e1535 | 7720 | |
67d66982 | 7721 | dio_data->submitted += bio->bi_iter.bi_size; |
b73a6fd1 BB |
7722 | |
7723 | /* | |
7724 | * Check if we are doing a partial write. If we are, we need to split | |
7725 | * the ordered extent to match the submitted bio. Hang on to the | |
7726 | * remaining unfinishable ordered_extent in dio_data so that it can be | |
7727 | * cancelled in iomap_end to avoid a deadlock wherein faulting the | |
7728 | * remaining pages is blocked on the outstanding ordered extent. | |
7729 | */ | |
7730 | if (iter->flags & IOMAP_WRITE) { | |
7731 | int ret; | |
7732 | ||
7733 | ret = btrfs_extract_ordered_extent(bbio, dio_data->ordered); | |
7734 | if (ret) { | |
7cad645e CH |
7735 | btrfs_finish_ordered_extent(dio_data->ordered, NULL, |
7736 | file_offset, dip->bytes, | |
7737 | !ret); | |
7738 | bio->bi_status = errno_to_blk_status(ret); | |
7739 | iomap_dio_bio_end_io(bio); | |
b73a6fd1 BB |
7740 | return; |
7741 | } | |
7742 | } | |
7743 | ||
ae42a154 | 7744 | btrfs_submit_bio(bbio, 0); |
4b46fce2 JB |
7745 | } |
7746 | ||
36e8c622 | 7747 | static const struct iomap_ops btrfs_dio_iomap_ops = { |
f85781fb GR |
7748 | .iomap_begin = btrfs_dio_iomap_begin, |
7749 | .iomap_end = btrfs_dio_iomap_end, | |
7750 | }; | |
7751 | ||
36e8c622 | 7752 | static const struct iomap_dio_ops btrfs_dio_ops = { |
67d66982 | 7753 | .submit_io = btrfs_dio_submit_io, |
642c5d34 | 7754 | .bio_set = &btrfs_dio_bioset, |
f85781fb GR |
7755 | }; |
7756 | ||
8184620a | 7757 | ssize_t btrfs_dio_read(struct kiocb *iocb, struct iov_iter *iter, size_t done_before) |
36e8c622 | 7758 | { |
53f2c206 | 7759 | struct btrfs_dio_data data = { 0 }; |
491a6d01 | 7760 | |
36e8c622 | 7761 | return iomap_dio_rw(iocb, iter, &btrfs_dio_iomap_ops, &btrfs_dio_ops, |
8184620a FM |
7762 | IOMAP_DIO_PARTIAL, &data, done_before); |
7763 | } | |
7764 | ||
7765 | struct iomap_dio *btrfs_dio_write(struct kiocb *iocb, struct iov_iter *iter, | |
7766 | size_t done_before) | |
7767 | { | |
53f2c206 | 7768 | struct btrfs_dio_data data = { 0 }; |
8184620a FM |
7769 | |
7770 | return __iomap_dio_rw(iocb, iter, &btrfs_dio_iomap_ops, &btrfs_dio_ops, | |
7771 | IOMAP_DIO_PARTIAL, &data, done_before); | |
36e8c622 CH |
7772 | } |
7773 | ||
1506fcc8 | 7774 | static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, |
bab16e21 | 7775 | u64 start, u64 len) |
1506fcc8 | 7776 | { |
05dadc09 TI |
7777 | int ret; |
7778 | ||
45dd052e | 7779 | ret = fiemap_prep(inode, fieinfo, start, &len, 0); |
05dadc09 TI |
7780 | if (ret) |
7781 | return ret; | |
7782 | ||
33a86cfa FM |
7783 | /* |
7784 | * fiemap_prep() called filemap_write_and_wait() for the whole possible | |
7785 | * file range (0 to LLONG_MAX), but that is not enough if we have | |
7786 | * compression enabled. The first filemap_fdatawrite_range() only kicks | |
7787 | * in the compression of data (in an async thread) and will return | |
7788 | * before the compression is done and writeback is started. A second | |
7789 | * filemap_fdatawrite_range() is needed to wait for the compression to | |
ac3c0d36 FM |
7790 | * complete and writeback to start. We also need to wait for ordered |
7791 | * extents to complete, because our fiemap implementation uses mainly | |
7792 | * file extent items to list the extents, searching for extent maps | |
7793 | * only for file ranges with holes or prealloc extents to figure out | |
7794 | * if we have delalloc in those ranges. | |
33a86cfa FM |
7795 | */ |
7796 | if (fieinfo->fi_flags & FIEMAP_FLAG_SYNC) { | |
ac3c0d36 | 7797 | ret = btrfs_wait_ordered_range(inode, 0, LLONG_MAX); |
33a86cfa FM |
7798 | if (ret) |
7799 | return ret; | |
7800 | } | |
7801 | ||
facee0a0 | 7802 | return extent_fiemap(BTRFS_I(inode), fieinfo, start, len); |
1506fcc8 YS |
7803 | } |
7804 | ||
48a3b636 ES |
7805 | static int btrfs_writepages(struct address_space *mapping, |
7806 | struct writeback_control *wbc) | |
b293f02e | 7807 | { |
8ae225a8 | 7808 | return extent_writepages(mapping, wbc); |
b293f02e CM |
7809 | } |
7810 | ||
ba206a02 | 7811 | static void btrfs_readahead(struct readahead_control *rac) |
3ab2fb5a | 7812 | { |
ba206a02 | 7813 | extent_readahead(rac); |
3ab2fb5a | 7814 | } |
2a3ff0ad | 7815 | |
7c11d0ae | 7816 | /* |
f913cff3 | 7817 | * For release_folio() and invalidate_folio() we have a race window where |
895586eb | 7818 | * folio_end_writeback() is called but the subpage spinlock is not yet released. |
7c11d0ae QW |
7819 | * If we continue to release/invalidate the page, we could cause use-after-free |
7820 | * for subpage spinlock. So this function is to spin and wait for subpage | |
7821 | * spinlock. | |
7822 | */ | |
7823 | static void wait_subpage_spinlock(struct page *page) | |
7824 | { | |
7825 | struct btrfs_fs_info *fs_info = btrfs_sb(page->mapping->host->i_sb); | |
7826 | struct btrfs_subpage *subpage; | |
7827 | ||
fbca46eb | 7828 | if (!btrfs_is_subpage(fs_info, page)) |
7c11d0ae QW |
7829 | return; |
7830 | ||
7831 | ASSERT(PagePrivate(page) && page->private); | |
7832 | subpage = (struct btrfs_subpage *)page->private; | |
7833 | ||
7834 | /* | |
7835 | * This may look insane as we just acquire the spinlock and release it, | |
7836 | * without doing anything. But we just want to make sure no one is | |
7837 | * still holding the subpage spinlock. | |
7838 | * And since the page is not dirty nor writeback, and we have page | |
7839 | * locked, the only possible way to hold a spinlock is from the endio | |
7840 | * function to clear page writeback. | |
7841 | * | |
7842 | * Here we just acquire the spinlock so that all existing callers | |
7843 | * should exit and we're safe to release/invalidate the page. | |
7844 | */ | |
7845 | spin_lock_irq(&subpage->lock); | |
7846 | spin_unlock_irq(&subpage->lock); | |
7847 | } | |
7848 | ||
f913cff3 | 7849 | static bool __btrfs_release_folio(struct folio *folio, gfp_t gfp_flags) |
9ebefb18 | 7850 | { |
f913cff3 | 7851 | int ret = try_release_extent_mapping(&folio->page, gfp_flags); |
7c11d0ae QW |
7852 | |
7853 | if (ret == 1) { | |
f913cff3 MWO |
7854 | wait_subpage_spinlock(&folio->page); |
7855 | clear_page_extent_mapped(&folio->page); | |
7c11d0ae | 7856 | } |
a52d9a80 | 7857 | return ret; |
39279cc3 CM |
7858 | } |
7859 | ||
f913cff3 | 7860 | static bool btrfs_release_folio(struct folio *folio, gfp_t gfp_flags) |
e6dcd2dc | 7861 | { |
f913cff3 MWO |
7862 | if (folio_test_writeback(folio) || folio_test_dirty(folio)) |
7863 | return false; | |
7864 | return __btrfs_release_folio(folio, gfp_flags); | |
e6dcd2dc CM |
7865 | } |
7866 | ||
f8e66081 | 7867 | #ifdef CONFIG_MIGRATION |
e7a60a17 MWO |
7868 | static int btrfs_migrate_folio(struct address_space *mapping, |
7869 | struct folio *dst, struct folio *src, | |
f8e66081 RG |
7870 | enum migrate_mode mode) |
7871 | { | |
e7a60a17 | 7872 | int ret = filemap_migrate_folio(mapping, dst, src, mode); |
f8e66081 | 7873 | |
f8e66081 RG |
7874 | if (ret != MIGRATEPAGE_SUCCESS) |
7875 | return ret; | |
7876 | ||
e7a60a17 MWO |
7877 | if (folio_test_ordered(src)) { |
7878 | folio_clear_ordered(src); | |
7879 | folio_set_ordered(dst); | |
f8e66081 RG |
7880 | } |
7881 | ||
f8e66081 RG |
7882 | return MIGRATEPAGE_SUCCESS; |
7883 | } | |
e7a60a17 MWO |
7884 | #else |
7885 | #define btrfs_migrate_folio NULL | |
f8e66081 RG |
7886 | #endif |
7887 | ||
895586eb MWO |
7888 | static void btrfs_invalidate_folio(struct folio *folio, size_t offset, |
7889 | size_t length) | |
39279cc3 | 7890 | { |
895586eb | 7891 | struct btrfs_inode *inode = BTRFS_I(folio->mapping->host); |
b945a463 | 7892 | struct btrfs_fs_info *fs_info = inode->root->fs_info; |
53ac7ead | 7893 | struct extent_io_tree *tree = &inode->io_tree; |
2ac55d41 | 7894 | struct extent_state *cached_state = NULL; |
895586eb MWO |
7895 | u64 page_start = folio_pos(folio); |
7896 | u64 page_end = page_start + folio_size(folio) - 1; | |
3b835840 | 7897 | u64 cur; |
53ac7ead | 7898 | int inode_evicting = inode->vfs_inode.i_state & I_FREEING; |
39279cc3 | 7899 | |
8b62b72b | 7900 | /* |
895586eb MWO |
7901 | * We have folio locked so no new ordered extent can be created on this |
7902 | * page, nor bio can be submitted for this folio. | |
8b62b72b | 7903 | * |
895586eb MWO |
7904 | * But already submitted bio can still be finished on this folio. |
7905 | * Furthermore, endio function won't skip folio which has Ordered | |
f57ad937 | 7906 | * (Private2) already cleared, so it's possible for endio and |
895586eb MWO |
7907 | * invalidate_folio to do the same ordered extent accounting twice |
7908 | * on one folio. | |
266a2586 QW |
7909 | * |
7910 | * So here we wait for any submitted bios to finish, so that we won't | |
895586eb | 7911 | * do double ordered extent accounting on the same folio. |
8b62b72b | 7912 | */ |
895586eb MWO |
7913 | folio_wait_writeback(folio); |
7914 | wait_subpage_spinlock(&folio->page); | |
8b62b72b | 7915 | |
bcd77455 QW |
7916 | /* |
7917 | * For subpage case, we have call sites like | |
7918 | * btrfs_punch_hole_lock_range() which passes range not aligned to | |
7919 | * sectorsize. | |
895586eb MWO |
7920 | * If the range doesn't cover the full folio, we don't need to and |
7921 | * shouldn't clear page extent mapped, as folio->private can still | |
bcd77455 QW |
7922 | * record subpage dirty bits for other part of the range. |
7923 | * | |
895586eb MWO |
7924 | * For cases that invalidate the full folio even the range doesn't |
7925 | * cover the full folio, like invalidating the last folio, we're | |
bcd77455 QW |
7926 | * still safe to wait for ordered extent to finish. |
7927 | */ | |
5a60542c | 7928 | if (!(offset == 0 && length == folio_size(folio))) { |
f913cff3 | 7929 | btrfs_release_folio(folio, GFP_NOFS); |
e6dcd2dc CM |
7930 | return; |
7931 | } | |
131e404a FDBM |
7932 | |
7933 | if (!inode_evicting) | |
570eb97b | 7934 | lock_extent(tree, page_start, page_end, &cached_state); |
951c80f8 | 7935 | |
3b835840 QW |
7936 | cur = page_start; |
7937 | while (cur < page_end) { | |
7938 | struct btrfs_ordered_extent *ordered; | |
3b835840 | 7939 | u64 range_end; |
b945a463 | 7940 | u32 range_len; |
bd015294 | 7941 | u32 extra_flags = 0; |
3b835840 QW |
7942 | |
7943 | ordered = btrfs_lookup_first_ordered_range(inode, cur, | |
7944 | page_end + 1 - cur); | |
7945 | if (!ordered) { | |
7946 | range_end = page_end; | |
7947 | /* | |
7948 | * No ordered extent covering this range, we are safe | |
7949 | * to delete all extent states in the range. | |
7950 | */ | |
bd015294 | 7951 | extra_flags = EXTENT_CLEAR_ALL_BITS; |
3b835840 QW |
7952 | goto next; |
7953 | } | |
7954 | if (ordered->file_offset > cur) { | |
7955 | /* | |
7956 | * There is a range between [cur, oe->file_offset) not | |
7957 | * covered by any ordered extent. | |
7958 | * We are safe to delete all extent states, and handle | |
7959 | * the ordered extent in the next iteration. | |
7960 | */ | |
7961 | range_end = ordered->file_offset - 1; | |
bd015294 | 7962 | extra_flags = EXTENT_CLEAR_ALL_BITS; |
3b835840 QW |
7963 | goto next; |
7964 | } | |
7965 | ||
7966 | range_end = min(ordered->file_offset + ordered->num_bytes - 1, | |
7967 | page_end); | |
b945a463 QW |
7968 | ASSERT(range_end + 1 - cur < U32_MAX); |
7969 | range_len = range_end + 1 - cur; | |
895586eb | 7970 | if (!btrfs_page_test_ordered(fs_info, &folio->page, cur, range_len)) { |
3b835840 | 7971 | /* |
f57ad937 QW |
7972 | * If Ordered (Private2) is cleared, it means endio has |
7973 | * already been executed for the range. | |
3b835840 QW |
7974 | * We can't delete the extent states as |
7975 | * btrfs_finish_ordered_io() may still use some of them. | |
7976 | */ | |
3b835840 QW |
7977 | goto next; |
7978 | } | |
895586eb | 7979 | btrfs_page_clear_ordered(fs_info, &folio->page, cur, range_len); |
3b835840 | 7980 | |
eb84ae03 | 7981 | /* |
2766ff61 FM |
7982 | * IO on this page will never be started, so we need to account |
7983 | * for any ordered extents now. Don't clear EXTENT_DELALLOC_NEW | |
7984 | * here, must leave that up for the ordered extent completion. | |
3b835840 QW |
7985 | * |
7986 | * This will also unlock the range for incoming | |
7987 | * btrfs_finish_ordered_io(). | |
eb84ae03 | 7988 | */ |
131e404a | 7989 | if (!inode_evicting) |
3b835840 | 7990 | clear_extent_bit(tree, cur, range_end, |
2766ff61 | 7991 | EXTENT_DELALLOC | |
131e404a | 7992 | EXTENT_LOCKED | EXTENT_DO_ACCOUNTING | |
bd015294 | 7993 | EXTENT_DEFRAG, &cached_state); |
3b835840 QW |
7994 | |
7995 | spin_lock_irq(&inode->ordered_tree.lock); | |
7996 | set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags); | |
7997 | ordered->truncated_len = min(ordered->truncated_len, | |
7998 | cur - ordered->file_offset); | |
7999 | spin_unlock_irq(&inode->ordered_tree.lock); | |
8000 | ||
bd015294 JB |
8001 | /* |
8002 | * If the ordered extent has finished, we're safe to delete all | |
8003 | * the extent states of the range, otherwise | |
8004 | * btrfs_finish_ordered_io() will get executed by endio for | |
8005 | * other pages, so we can't delete extent states. | |
8006 | */ | |
3b835840 | 8007 | if (btrfs_dec_test_ordered_pending(inode, &ordered, |
f41b6ba9 | 8008 | cur, range_end + 1 - cur)) { |
3b835840 QW |
8009 | btrfs_finish_ordered_io(ordered); |
8010 | /* | |
8011 | * The ordered extent has finished, now we're again | |
8012 | * safe to delete all extent states of the range. | |
8013 | */ | |
bd015294 | 8014 | extra_flags = EXTENT_CLEAR_ALL_BITS; |
3b835840 QW |
8015 | } |
8016 | next: | |
8017 | if (ordered) | |
8018 | btrfs_put_ordered_extent(ordered); | |
8b62b72b | 8019 | /* |
3b835840 QW |
8020 | * Qgroup reserved space handler |
8021 | * Sector(s) here will be either: | |
266a2586 | 8022 | * |
3b835840 QW |
8023 | * 1) Already written to disk or bio already finished |
8024 | * Then its QGROUP_RESERVED bit in io_tree is already cleared. | |
8025 | * Qgroup will be handled by its qgroup_record then. | |
8026 | * btrfs_qgroup_free_data() call will do nothing here. | |
8027 | * | |
8028 | * 2) Not written to disk yet | |
8029 | * Then btrfs_qgroup_free_data() call will clear the | |
8030 | * QGROUP_RESERVED bit of its io_tree, and free the qgroup | |
8031 | * reserved data space. | |
8032 | * Since the IO will never happen for this page. | |
8b62b72b | 8033 | */ |
3b835840 | 8034 | btrfs_qgroup_free_data(inode, NULL, cur, range_end + 1 - cur); |
131e404a | 8035 | if (!inode_evicting) { |
3b835840 QW |
8036 | clear_extent_bit(tree, cur, range_end, EXTENT_LOCKED | |
8037 | EXTENT_DELALLOC | EXTENT_UPTODATE | | |
bd015294 JB |
8038 | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG | |
8039 | extra_flags, &cached_state); | |
131e404a | 8040 | } |
3b835840 | 8041 | cur = range_end + 1; |
131e404a | 8042 | } |
b9d0b389 | 8043 | /* |
3b835840 | 8044 | * We have iterated through all ordered extents of the page, the page |
f57ad937 QW |
8045 | * should not have Ordered (Private2) anymore, or the above iteration |
8046 | * did something wrong. | |
b9d0b389 | 8047 | */ |
895586eb MWO |
8048 | ASSERT(!folio_test_ordered(folio)); |
8049 | btrfs_page_clear_checked(fs_info, &folio->page, folio_pos(folio), folio_size(folio)); | |
3b835840 | 8050 | if (!inode_evicting) |
f913cff3 | 8051 | __btrfs_release_folio(folio, GFP_NOFS); |
895586eb | 8052 | clear_page_extent_mapped(&folio->page); |
39279cc3 CM |
8053 | } |
8054 | ||
9ebefb18 CM |
8055 | /* |
8056 | * btrfs_page_mkwrite() is not allowed to change the file size as it gets | |
8057 | * called from a page fault handler when a page is first dirtied. Hence we must | |
8058 | * be careful to check for EOF conditions here. We set the page up correctly | |
8059 | * for a written page which means we get ENOSPC checking when writing into | |
8060 | * holes and correct delalloc and unwritten extent mapping on filesystems that | |
8061 | * support these features. | |
8062 | * | |
8063 | * We are not allowed to take the i_mutex here so we have to play games to | |
8064 | * protect against truncate races as the page could now be beyond EOF. Because | |
d1342aad OS |
8065 | * truncate_setsize() writes the inode size before removing pages, once we have |
8066 | * the page lock we can determine safely if the page is beyond EOF. If it is not | |
9ebefb18 CM |
8067 | * beyond EOF, then the page is guaranteed safe against truncation until we |
8068 | * unlock the page. | |
8069 | */ | |
a528a241 | 8070 | vm_fault_t btrfs_page_mkwrite(struct vm_fault *vmf) |
9ebefb18 | 8071 | { |
c2ec175c | 8072 | struct page *page = vmf->page; |
11bac800 | 8073 | struct inode *inode = file_inode(vmf->vma->vm_file); |
0b246afa | 8074 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
e6dcd2dc CM |
8075 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
8076 | struct btrfs_ordered_extent *ordered; | |
2ac55d41 | 8077 | struct extent_state *cached_state = NULL; |
364ecf36 | 8078 | struct extent_changeset *data_reserved = NULL; |
e6dcd2dc | 8079 | unsigned long zero_start; |
9ebefb18 | 8080 | loff_t size; |
a528a241 SJ |
8081 | vm_fault_t ret; |
8082 | int ret2; | |
9998eb70 | 8083 | int reserved = 0; |
d0b7da88 | 8084 | u64 reserved_space; |
a52d9a80 | 8085 | u64 page_start; |
e6dcd2dc | 8086 | u64 page_end; |
d0b7da88 CR |
8087 | u64 end; |
8088 | ||
09cbfeaf | 8089 | reserved_space = PAGE_SIZE; |
9ebefb18 | 8090 | |
b2b5ef5c | 8091 | sb_start_pagefault(inode->i_sb); |
df480633 | 8092 | page_start = page_offset(page); |
09cbfeaf | 8093 | page_end = page_start + PAGE_SIZE - 1; |
d0b7da88 | 8094 | end = page_end; |
df480633 | 8095 | |
d0b7da88 CR |
8096 | /* |
8097 | * Reserving delalloc space after obtaining the page lock can lead to | |
8098 | * deadlock. For example, if a dirty page is locked by this function | |
8099 | * and the call to btrfs_delalloc_reserve_space() ends up triggering | |
f3e90c1c | 8100 | * dirty page write out, then the btrfs_writepages() function could |
d0b7da88 CR |
8101 | * end up waiting indefinitely to get a lock on the page currently |
8102 | * being processed by btrfs_page_mkwrite() function. | |
8103 | */ | |
e5b7231e NB |
8104 | ret2 = btrfs_delalloc_reserve_space(BTRFS_I(inode), &data_reserved, |
8105 | page_start, reserved_space); | |
a528a241 SJ |
8106 | if (!ret2) { |
8107 | ret2 = file_update_time(vmf->vma->vm_file); | |
9998eb70 CM |
8108 | reserved = 1; |
8109 | } | |
a528a241 SJ |
8110 | if (ret2) { |
8111 | ret = vmf_error(ret2); | |
9998eb70 CM |
8112 | if (reserved) |
8113 | goto out; | |
8114 | goto out_noreserve; | |
56a76f82 | 8115 | } |
1832a6d5 | 8116 | |
56a76f82 | 8117 | ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */ |
e6dcd2dc | 8118 | again: |
8318ba79 | 8119 | down_read(&BTRFS_I(inode)->i_mmap_lock); |
9ebefb18 | 8120 | lock_page(page); |
9ebefb18 | 8121 | size = i_size_read(inode); |
a52d9a80 | 8122 | |
9ebefb18 | 8123 | if ((page->mapping != inode->i_mapping) || |
e6dcd2dc | 8124 | (page_start >= size)) { |
9ebefb18 CM |
8125 | /* page got truncated out from underneath us */ |
8126 | goto out_unlock; | |
8127 | } | |
e6dcd2dc CM |
8128 | wait_on_page_writeback(page); |
8129 | ||
570eb97b | 8130 | lock_extent(io_tree, page_start, page_end, &cached_state); |
32443de3 QW |
8131 | ret2 = set_page_extent_mapped(page); |
8132 | if (ret2 < 0) { | |
8133 | ret = vmf_error(ret2); | |
570eb97b | 8134 | unlock_extent(io_tree, page_start, page_end, &cached_state); |
32443de3 QW |
8135 | goto out_unlock; |
8136 | } | |
e6dcd2dc | 8137 | |
eb84ae03 CM |
8138 | /* |
8139 | * we can't set the delalloc bits if there are pending ordered | |
8140 | * extents. Drop our locks and wait for them to finish | |
8141 | */ | |
a776c6fa NB |
8142 | ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), page_start, |
8143 | PAGE_SIZE); | |
e6dcd2dc | 8144 | if (ordered) { |
570eb97b | 8145 | unlock_extent(io_tree, page_start, page_end, &cached_state); |
e6dcd2dc | 8146 | unlock_page(page); |
8318ba79 | 8147 | up_read(&BTRFS_I(inode)->i_mmap_lock); |
36d45567 | 8148 | btrfs_start_ordered_extent(ordered); |
e6dcd2dc CM |
8149 | btrfs_put_ordered_extent(ordered); |
8150 | goto again; | |
8151 | } | |
8152 | ||
09cbfeaf | 8153 | if (page->index == ((size - 1) >> PAGE_SHIFT)) { |
da17066c | 8154 | reserved_space = round_up(size - page_start, |
0b246afa | 8155 | fs_info->sectorsize); |
09cbfeaf | 8156 | if (reserved_space < PAGE_SIZE) { |
d0b7da88 | 8157 | end = page_start + reserved_space - 1; |
86d52921 NB |
8158 | btrfs_delalloc_release_space(BTRFS_I(inode), |
8159 | data_reserved, page_start, | |
8160 | PAGE_SIZE - reserved_space, true); | |
d0b7da88 CR |
8161 | } |
8162 | } | |
8163 | ||
fbf19087 | 8164 | /* |
5416034f LB |
8165 | * page_mkwrite gets called when the page is firstly dirtied after it's |
8166 | * faulted in, but write(2) could also dirty a page and set delalloc | |
8167 | * bits, thus in this case for space account reason, we still need to | |
8168 | * clear any delalloc bits within this page range since we have to | |
8169 | * reserve data&meta space before lock_page() (see above comments). | |
fbf19087 | 8170 | */ |
d0b7da88 | 8171 | clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, end, |
e182163d | 8172 | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING | |
bd015294 | 8173 | EXTENT_DEFRAG, &cached_state); |
fbf19087 | 8174 | |
c2566f22 | 8175 | ret2 = btrfs_set_extent_delalloc(BTRFS_I(inode), page_start, end, 0, |
330a5827 | 8176 | &cached_state); |
a528a241 | 8177 | if (ret2) { |
570eb97b | 8178 | unlock_extent(io_tree, page_start, page_end, &cached_state); |
9ed74f2d JB |
8179 | ret = VM_FAULT_SIGBUS; |
8180 | goto out_unlock; | |
8181 | } | |
9ebefb18 CM |
8182 | |
8183 | /* page is wholly or partially inside EOF */ | |
09cbfeaf | 8184 | if (page_start + PAGE_SIZE > size) |
7073017a | 8185 | zero_start = offset_in_page(size); |
9ebefb18 | 8186 | else |
09cbfeaf | 8187 | zero_start = PAGE_SIZE; |
9ebefb18 | 8188 | |
21a8935e | 8189 | if (zero_start != PAGE_SIZE) |
d048b9c2 | 8190 | memzero_page(page, zero_start, PAGE_SIZE - zero_start); |
21a8935e | 8191 | |
e4f94347 | 8192 | btrfs_page_clear_checked(fs_info, page, page_start, PAGE_SIZE); |
2d8ec40e QW |
8193 | btrfs_page_set_dirty(fs_info, page, page_start, end + 1 - page_start); |
8194 | btrfs_page_set_uptodate(fs_info, page, page_start, end + 1 - page_start); | |
5a3f23d5 | 8195 | |
bc0939fc | 8196 | btrfs_set_inode_last_sub_trans(BTRFS_I(inode)); |
257c62e1 | 8197 | |
570eb97b | 8198 | unlock_extent(io_tree, page_start, page_end, &cached_state); |
8318ba79 | 8199 | up_read(&BTRFS_I(inode)->i_mmap_lock); |
9ebefb18 | 8200 | |
76de60ed YY |
8201 | btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE); |
8202 | sb_end_pagefault(inode->i_sb); | |
8203 | extent_changeset_free(data_reserved); | |
8204 | return VM_FAULT_LOCKED; | |
717beb96 CM |
8205 | |
8206 | out_unlock: | |
9ebefb18 | 8207 | unlock_page(page); |
8318ba79 | 8208 | up_read(&BTRFS_I(inode)->i_mmap_lock); |
1832a6d5 | 8209 | out: |
8702ba93 | 8210 | btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE); |
86d52921 | 8211 | btrfs_delalloc_release_space(BTRFS_I(inode), data_reserved, page_start, |
43b18595 | 8212 | reserved_space, (ret != 0)); |
9998eb70 | 8213 | out_noreserve: |
b2b5ef5c | 8214 | sb_end_pagefault(inode->i_sb); |
364ecf36 | 8215 | extent_changeset_free(data_reserved); |
9ebefb18 CM |
8216 | return ret; |
8217 | } | |
8218 | ||
d9dcae67 | 8219 | static int btrfs_truncate(struct btrfs_inode *inode, bool skip_writeback) |
39279cc3 | 8220 | { |
d9ac19c3 | 8221 | struct btrfs_truncate_control control = { |
d9dcae67 DS |
8222 | .inode = inode, |
8223 | .ino = btrfs_ino(inode), | |
d9ac19c3 | 8224 | .min_type = BTRFS_EXTENT_DATA_KEY, |
655807b8 | 8225 | .clear_extent_range = true, |
d9ac19c3 | 8226 | }; |
d9dcae67 DS |
8227 | struct btrfs_root *root = inode->root; |
8228 | struct btrfs_fs_info *fs_info = root->fs_info; | |
fcb80c2a | 8229 | struct btrfs_block_rsv *rsv; |
ad7e1a74 | 8230 | int ret; |
39279cc3 | 8231 | struct btrfs_trans_handle *trans; |
0b246afa | 8232 | u64 mask = fs_info->sectorsize - 1; |
6822b3f7 | 8233 | const u64 min_size = btrfs_calc_metadata_size(fs_info, 1); |
39279cc3 | 8234 | |
213e8c55 | 8235 | if (!skip_writeback) { |
d9dcae67 DS |
8236 | ret = btrfs_wait_ordered_range(&inode->vfs_inode, |
8237 | inode->vfs_inode.i_size & (~mask), | |
213e8c55 FM |
8238 | (u64)-1); |
8239 | if (ret) | |
8240 | return ret; | |
8241 | } | |
39279cc3 | 8242 | |
fcb80c2a | 8243 | /* |
f7e9e8fc OS |
8244 | * Yes ladies and gentlemen, this is indeed ugly. We have a couple of |
8245 | * things going on here: | |
fcb80c2a | 8246 | * |
f7e9e8fc | 8247 | * 1) We need to reserve space to update our inode. |
fcb80c2a | 8248 | * |
f7e9e8fc | 8249 | * 2) We need to have something to cache all the space that is going to |
fcb80c2a JB |
8250 | * be free'd up by the truncate operation, but also have some slack |
8251 | * space reserved in case it uses space during the truncate (thank you | |
8252 | * very much snapshotting). | |
8253 | * | |
f7e9e8fc | 8254 | * And we need these to be separate. The fact is we can use a lot of |
fcb80c2a | 8255 | * space doing the truncate, and we have no earthly idea how much space |
01327610 | 8256 | * we will use, so we need the truncate reservation to be separate so it |
f7e9e8fc OS |
8257 | * doesn't end up using space reserved for updating the inode. We also |
8258 | * need to be able to stop the transaction and start a new one, which | |
8259 | * means we need to be able to update the inode several times, and we | |
8260 | * have no idea of knowing how many times that will be, so we can't just | |
8261 | * reserve 1 item for the entirety of the operation, so that has to be | |
8262 | * done separately as well. | |
fcb80c2a JB |
8263 | * |
8264 | * So that leaves us with | |
8265 | * | |
f7e9e8fc | 8266 | * 1) rsv - for the truncate reservation, which we will steal from the |
fcb80c2a | 8267 | * transaction reservation. |
f7e9e8fc | 8268 | * 2) fs_info->trans_block_rsv - this will have 1 items worth left for |
fcb80c2a JB |
8269 | * updating the inode. |
8270 | */ | |
2ff7e61e | 8271 | rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP); |
fcb80c2a JB |
8272 | if (!rsv) |
8273 | return -ENOMEM; | |
4a338542 | 8274 | rsv->size = min_size; |
710d5921 | 8275 | rsv->failfast = true; |
f0cd846e | 8276 | |
907cbceb | 8277 | /* |
07127184 | 8278 | * 1 for the truncate slack space |
907cbceb JB |
8279 | * 1 for updating the inode. |
8280 | */ | |
f3fe820c | 8281 | trans = btrfs_start_transaction(root, 2); |
fcb80c2a | 8282 | if (IS_ERR(trans)) { |
ad7e1a74 | 8283 | ret = PTR_ERR(trans); |
fcb80c2a JB |
8284 | goto out; |
8285 | } | |
f0cd846e | 8286 | |
907cbceb | 8287 | /* Migrate the slack space for the truncate to our reserve */ |
0b246afa | 8288 | ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv, rsv, |
3a584174 | 8289 | min_size, false); |
6822b3f7 FM |
8290 | /* |
8291 | * We have reserved 2 metadata units when we started the transaction and | |
8292 | * min_size matches 1 unit, so this should never fail, but if it does, | |
8293 | * it's not critical we just fail truncation. | |
8294 | */ | |
8295 | if (WARN_ON(ret)) { | |
8296 | btrfs_end_transaction(trans); | |
8297 | goto out; | |
8298 | } | |
f0cd846e | 8299 | |
ca7e70f5 | 8300 | trans->block_rsv = rsv; |
907cbceb | 8301 | |
8082510e | 8302 | while (1) { |
9a4a1429 | 8303 | struct extent_state *cached_state = NULL; |
d9dcae67 | 8304 | const u64 new_size = inode->vfs_inode.i_size; |
9a4a1429 JB |
8305 | const u64 lock_start = ALIGN_DOWN(new_size, fs_info->sectorsize); |
8306 | ||
d9ac19c3 | 8307 | control.new_size = new_size; |
d9dcae67 | 8308 | lock_extent(&inode->io_tree, lock_start, (u64)-1, &cached_state); |
9a4a1429 JB |
8309 | /* |
8310 | * We want to drop from the next block forward in case this new | |
8311 | * size is not block aligned since we will be keeping the last | |
8312 | * block of the extent just the way it is. | |
8313 | */ | |
d9dcae67 | 8314 | btrfs_drop_extent_map_range(inode, |
4c0c8cfc FM |
8315 | ALIGN(new_size, fs_info->sectorsize), |
8316 | (u64)-1, false); | |
9a4a1429 | 8317 | |
71d18b53 | 8318 | ret = btrfs_truncate_inode_items(trans, root, &control); |
c2ddb612 | 8319 | |
d9dcae67 DS |
8320 | inode_sub_bytes(&inode->vfs_inode, control.sub_bytes); |
8321 | btrfs_inode_safe_disk_i_size_write(inode, control.last_size); | |
c2ddb612 | 8322 | |
d9dcae67 | 8323 | unlock_extent(&inode->io_tree, lock_start, (u64)-1, &cached_state); |
9a4a1429 | 8324 | |
ddfae63c | 8325 | trans->block_rsv = &fs_info->trans_block_rsv; |
ad7e1a74 | 8326 | if (ret != -ENOSPC && ret != -EAGAIN) |
8082510e | 8327 | break; |
39279cc3 | 8328 | |
d9dcae67 | 8329 | ret = btrfs_update_inode(trans, root, inode); |
ad7e1a74 | 8330 | if (ret) |
3893e33b | 8331 | break; |
ca7e70f5 | 8332 | |
3a45bb20 | 8333 | btrfs_end_transaction(trans); |
2ff7e61e | 8334 | btrfs_btree_balance_dirty(fs_info); |
ca7e70f5 JB |
8335 | |
8336 | trans = btrfs_start_transaction(root, 2); | |
8337 | if (IS_ERR(trans)) { | |
ad7e1a74 | 8338 | ret = PTR_ERR(trans); |
ca7e70f5 JB |
8339 | trans = NULL; |
8340 | break; | |
8341 | } | |
8342 | ||
63f018be | 8343 | btrfs_block_rsv_release(fs_info, rsv, -1, NULL); |
0b246afa | 8344 | ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv, |
3a584174 | 8345 | rsv, min_size, false); |
6822b3f7 FM |
8346 | /* |
8347 | * We have reserved 2 metadata units when we started the | |
8348 | * transaction and min_size matches 1 unit, so this should never | |
8349 | * fail, but if it does, it's not critical we just fail truncation. | |
8350 | */ | |
8351 | if (WARN_ON(ret)) | |
8352 | break; | |
8353 | ||
ca7e70f5 | 8354 | trans->block_rsv = rsv; |
8082510e YZ |
8355 | } |
8356 | ||
ddfae63c JB |
8357 | /* |
8358 | * We can't call btrfs_truncate_block inside a trans handle as we could | |
54f03ab1 JB |
8359 | * deadlock with freeze, if we got BTRFS_NEED_TRUNCATE_BLOCK then we |
8360 | * know we've truncated everything except the last little bit, and can | |
8361 | * do btrfs_truncate_block and then update the disk_i_size. | |
ddfae63c | 8362 | */ |
54f03ab1 | 8363 | if (ret == BTRFS_NEED_TRUNCATE_BLOCK) { |
ddfae63c JB |
8364 | btrfs_end_transaction(trans); |
8365 | btrfs_btree_balance_dirty(fs_info); | |
8366 | ||
d9dcae67 | 8367 | ret = btrfs_truncate_block(inode, inode->vfs_inode.i_size, 0, 0); |
ddfae63c JB |
8368 | if (ret) |
8369 | goto out; | |
8370 | trans = btrfs_start_transaction(root, 1); | |
8371 | if (IS_ERR(trans)) { | |
8372 | ret = PTR_ERR(trans); | |
8373 | goto out; | |
8374 | } | |
d9dcae67 | 8375 | btrfs_inode_safe_disk_i_size_write(inode, 0); |
ddfae63c JB |
8376 | } |
8377 | ||
917c16b2 | 8378 | if (trans) { |
ad7e1a74 OS |
8379 | int ret2; |
8380 | ||
0b246afa | 8381 | trans->block_rsv = &fs_info->trans_block_rsv; |
d9dcae67 | 8382 | ret2 = btrfs_update_inode(trans, root, inode); |
ad7e1a74 OS |
8383 | if (ret2 && !ret) |
8384 | ret = ret2; | |
7b128766 | 8385 | |
ad7e1a74 OS |
8386 | ret2 = btrfs_end_transaction(trans); |
8387 | if (ret2 && !ret) | |
8388 | ret = ret2; | |
2ff7e61e | 8389 | btrfs_btree_balance_dirty(fs_info); |
917c16b2 | 8390 | } |
fcb80c2a | 8391 | out: |
2ff7e61e | 8392 | btrfs_free_block_rsv(fs_info, rsv); |
0d7d3165 FM |
8393 | /* |
8394 | * So if we truncate and then write and fsync we normally would just | |
8395 | * write the extents that changed, which is a problem if we need to | |
8396 | * first truncate that entire inode. So set this flag so we write out | |
8397 | * all of the extents in the inode to the sync log so we're completely | |
8398 | * safe. | |
8399 | * | |
8400 | * If no extents were dropped or trimmed we don't need to force the next | |
8401 | * fsync to truncate all the inode's items from the log and re-log them | |
8402 | * all. This means the truncate operation did not change the file size, | |
8403 | * or changed it to a smaller size but there was only an implicit hole | |
8404 | * between the old i_size and the new i_size, and there were no prealloc | |
8405 | * extents beyond i_size to drop. | |
8406 | */ | |
d9ac19c3 | 8407 | if (control.extents_found > 0) |
d9dcae67 | 8408 | btrfs_set_inode_full_sync(inode); |
fcb80c2a | 8409 | |
ad7e1a74 | 8410 | return ret; |
39279cc3 CM |
8411 | } |
8412 | ||
f2d40141 | 8413 | struct inode *btrfs_new_subvol_inode(struct mnt_idmap *idmap, |
a1fd0c35 OS |
8414 | struct inode *dir) |
8415 | { | |
8416 | struct inode *inode; | |
8417 | ||
8418 | inode = new_inode(dir->i_sb); | |
8419 | if (inode) { | |
8420 | /* | |
8421 | * Subvolumes don't inherit the sgid bit or the parent's gid if | |
8422 | * the parent's sgid bit is set. This is probably a bug. | |
8423 | */ | |
f2d40141 | 8424 | inode_init_owner(idmap, inode, NULL, |
a1fd0c35 OS |
8425 | S_IFDIR | (~current_umask() & S_IRWXUGO)); |
8426 | inode->i_op = &btrfs_dir_inode_operations; | |
8427 | inode->i_fop = &btrfs_dir_file_operations; | |
8428 | } | |
8429 | return inode; | |
8430 | } | |
8431 | ||
39279cc3 CM |
8432 | struct inode *btrfs_alloc_inode(struct super_block *sb) |
8433 | { | |
69fe2d75 | 8434 | struct btrfs_fs_info *fs_info = btrfs_sb(sb); |
39279cc3 | 8435 | struct btrfs_inode *ei; |
2ead6ae7 | 8436 | struct inode *inode; |
39279cc3 | 8437 | |
fd60b288 | 8438 | ei = alloc_inode_sb(sb, btrfs_inode_cachep, GFP_KERNEL); |
39279cc3 CM |
8439 | if (!ei) |
8440 | return NULL; | |
2ead6ae7 YZ |
8441 | |
8442 | ei->root = NULL; | |
2ead6ae7 | 8443 | ei->generation = 0; |
15ee9bc7 | 8444 | ei->last_trans = 0; |
257c62e1 | 8445 | ei->last_sub_trans = 0; |
e02119d5 | 8446 | ei->logged_trans = 0; |
2ead6ae7 | 8447 | ei->delalloc_bytes = 0; |
a7e3b975 | 8448 | ei->new_delalloc_bytes = 0; |
47059d93 | 8449 | ei->defrag_bytes = 0; |
2ead6ae7 YZ |
8450 | ei->disk_i_size = 0; |
8451 | ei->flags = 0; | |
77eea05e | 8452 | ei->ro_flags = 0; |
7709cde3 | 8453 | ei->csum_bytes = 0; |
2ead6ae7 | 8454 | ei->index_cnt = (u64)-1; |
67de1176 | 8455 | ei->dir_index = 0; |
2ead6ae7 | 8456 | ei->last_unlink_trans = 0; |
3ebac17c | 8457 | ei->last_reflink_trans = 0; |
46d8bc34 | 8458 | ei->last_log_commit = 0; |
2ead6ae7 | 8459 | |
9e0baf60 JB |
8460 | spin_lock_init(&ei->lock); |
8461 | ei->outstanding_extents = 0; | |
69fe2d75 JB |
8462 | if (sb->s_magic != BTRFS_TEST_MAGIC) |
8463 | btrfs_init_metadata_block_rsv(fs_info, &ei->block_rsv, | |
8464 | BTRFS_BLOCK_RSV_DELALLOC); | |
72ac3c0d | 8465 | ei->runtime_flags = 0; |
b52aa8c9 | 8466 | ei->prop_compress = BTRFS_COMPRESS_NONE; |
eec63c65 | 8467 | ei->defrag_compress = BTRFS_COMPRESS_NONE; |
2ead6ae7 | 8468 | |
16cdcec7 MX |
8469 | ei->delayed_node = NULL; |
8470 | ||
9cc97d64 | 8471 | ei->i_otime.tv_sec = 0; |
8472 | ei->i_otime.tv_nsec = 0; | |
8473 | ||
2ead6ae7 | 8474 | inode = &ei->vfs_inode; |
a8067e02 | 8475 | extent_map_tree_init(&ei->extent_tree); |
35da5a7e | 8476 | extent_io_tree_init(fs_info, &ei->io_tree, IO_TREE_INODE_IO); |
0988fc7b | 8477 | ei->io_tree.inode = ei; |
41a2ee75 | 8478 | extent_io_tree_init(fs_info, &ei->file_extent_tree, |
35da5a7e | 8479 | IO_TREE_INODE_FILE_EXTENT); |
2ead6ae7 | 8480 | mutex_init(&ei->log_mutex); |
e6dcd2dc | 8481 | btrfs_ordered_inode_tree_init(&ei->ordered_tree); |
2ead6ae7 | 8482 | INIT_LIST_HEAD(&ei->delalloc_inodes); |
8089fe62 | 8483 | INIT_LIST_HEAD(&ei->delayed_iput); |
2ead6ae7 | 8484 | RB_CLEAR_NODE(&ei->rb_node); |
8318ba79 | 8485 | init_rwsem(&ei->i_mmap_lock); |
2ead6ae7 YZ |
8486 | |
8487 | return inode; | |
39279cc3 CM |
8488 | } |
8489 | ||
aaedb55b JB |
8490 | #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS |
8491 | void btrfs_test_destroy_inode(struct inode *inode) | |
8492 | { | |
4c0c8cfc | 8493 | btrfs_drop_extent_map_range(BTRFS_I(inode), 0, (u64)-1, false); |
aaedb55b JB |
8494 | kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode)); |
8495 | } | |
8496 | #endif | |
8497 | ||
26602cab | 8498 | void btrfs_free_inode(struct inode *inode) |
fa0d7e3d | 8499 | { |
fa0d7e3d NP |
8500 | kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode)); |
8501 | } | |
8502 | ||
633cc816 | 8503 | void btrfs_destroy_inode(struct inode *vfs_inode) |
39279cc3 | 8504 | { |
e6dcd2dc | 8505 | struct btrfs_ordered_extent *ordered; |
633cc816 NB |
8506 | struct btrfs_inode *inode = BTRFS_I(vfs_inode); |
8507 | struct btrfs_root *root = inode->root; | |
5f4403e1 | 8508 | bool freespace_inode; |
5a3f23d5 | 8509 | |
633cc816 NB |
8510 | WARN_ON(!hlist_empty(&vfs_inode->i_dentry)); |
8511 | WARN_ON(vfs_inode->i_data.nrpages); | |
8512 | WARN_ON(inode->block_rsv.reserved); | |
8513 | WARN_ON(inode->block_rsv.size); | |
8514 | WARN_ON(inode->outstanding_extents); | |
dc287224 FM |
8515 | if (!S_ISDIR(vfs_inode->i_mode)) { |
8516 | WARN_ON(inode->delalloc_bytes); | |
8517 | WARN_ON(inode->new_delalloc_bytes); | |
8518 | } | |
633cc816 NB |
8519 | WARN_ON(inode->csum_bytes); |
8520 | WARN_ON(inode->defrag_bytes); | |
39279cc3 | 8521 | |
a6dbd429 JB |
8522 | /* |
8523 | * This can happen where we create an inode, but somebody else also | |
8524 | * created the same inode and we need to destroy the one we already | |
8525 | * created. | |
8526 | */ | |
8527 | if (!root) | |
26602cab | 8528 | return; |
a6dbd429 | 8529 | |
5f4403e1 IA |
8530 | /* |
8531 | * If this is a free space inode do not take the ordered extents lockdep | |
8532 | * map. | |
8533 | */ | |
8534 | freespace_inode = btrfs_is_free_space_inode(inode); | |
8535 | ||
d397712b | 8536 | while (1) { |
633cc816 | 8537 | ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1); |
e6dcd2dc CM |
8538 | if (!ordered) |
8539 | break; | |
8540 | else { | |
633cc816 | 8541 | btrfs_err(root->fs_info, |
5d163e0e | 8542 | "found ordered extent %llu %llu on inode cleanup", |
bffe633e | 8543 | ordered->file_offset, ordered->num_bytes); |
5f4403e1 IA |
8544 | |
8545 | if (!freespace_inode) | |
8546 | btrfs_lockdep_acquire(root->fs_info, btrfs_ordered_extent); | |
8547 | ||
71fe0a55 | 8548 | btrfs_remove_ordered_extent(inode, ordered); |
e6dcd2dc CM |
8549 | btrfs_put_ordered_extent(ordered); |
8550 | btrfs_put_ordered_extent(ordered); | |
8551 | } | |
8552 | } | |
633cc816 NB |
8553 | btrfs_qgroup_check_reserved_leak(inode); |
8554 | inode_tree_del(inode); | |
4c0c8cfc | 8555 | btrfs_drop_extent_map_range(inode, 0, (u64)-1, false); |
633cc816 NB |
8556 | btrfs_inode_clear_file_extent_range(inode, 0, (u64)-1); |
8557 | btrfs_put_root(inode->root); | |
39279cc3 CM |
8558 | } |
8559 | ||
45321ac5 | 8560 | int btrfs_drop_inode(struct inode *inode) |
76dda93c YZ |
8561 | { |
8562 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
45321ac5 | 8563 | |
6379ef9f NA |
8564 | if (root == NULL) |
8565 | return 1; | |
8566 | ||
fa6ac876 | 8567 | /* the snap/subvol tree is on deleting */ |
69e9c6c6 | 8568 | if (btrfs_root_refs(&root->root_item) == 0) |
45321ac5 | 8569 | return 1; |
76dda93c | 8570 | else |
45321ac5 | 8571 | return generic_drop_inode(inode); |
76dda93c YZ |
8572 | } |
8573 | ||
0ee0fda0 | 8574 | static void init_once(void *foo) |
39279cc3 | 8575 | { |
0d031dc4 | 8576 | struct btrfs_inode *ei = foo; |
39279cc3 CM |
8577 | |
8578 | inode_init_once(&ei->vfs_inode); | |
8579 | } | |
8580 | ||
e67c718b | 8581 | void __cold btrfs_destroy_cachep(void) |
39279cc3 | 8582 | { |
8c0a8537 KS |
8583 | /* |
8584 | * Make sure all delayed rcu free inodes are flushed before we | |
8585 | * destroy cache. | |
8586 | */ | |
8587 | rcu_barrier(); | |
642c5d34 | 8588 | bioset_exit(&btrfs_dio_bioset); |
5598e900 | 8589 | kmem_cache_destroy(btrfs_inode_cachep); |
39279cc3 CM |
8590 | } |
8591 | ||
f5c29bd9 | 8592 | int __init btrfs_init_cachep(void) |
39279cc3 | 8593 | { |
837e1972 | 8594 | btrfs_inode_cachep = kmem_cache_create("btrfs_inode", |
9601e3f6 | 8595 | sizeof(struct btrfs_inode), 0, |
5d097056 VD |
8596 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD | SLAB_ACCOUNT, |
8597 | init_once); | |
39279cc3 CM |
8598 | if (!btrfs_inode_cachep) |
8599 | goto fail; | |
9601e3f6 | 8600 | |
642c5d34 | 8601 | if (bioset_init(&btrfs_dio_bioset, BIO_POOL_SIZE, |
67d66982 | 8602 | offsetof(struct btrfs_dio_private, bbio.bio), |
642c5d34 CH |
8603 | BIOSET_NEED_BVECS)) |
8604 | goto fail; | |
8605 | ||
39279cc3 CM |
8606 | return 0; |
8607 | fail: | |
8608 | btrfs_destroy_cachep(); | |
8609 | return -ENOMEM; | |
8610 | } | |
8611 | ||
b74d24f7 | 8612 | static int btrfs_getattr(struct mnt_idmap *idmap, |
549c7297 | 8613 | const struct path *path, struct kstat *stat, |
a528d35e | 8614 | u32 request_mask, unsigned int flags) |
39279cc3 | 8615 | { |
df0af1a5 | 8616 | u64 delalloc_bytes; |
2766ff61 | 8617 | u64 inode_bytes; |
a528d35e | 8618 | struct inode *inode = d_inode(path->dentry); |
fadc0d8b | 8619 | u32 blocksize = inode->i_sb->s_blocksize; |
04a87e34 | 8620 | u32 bi_flags = BTRFS_I(inode)->flags; |
14605409 | 8621 | u32 bi_ro_flags = BTRFS_I(inode)->ro_flags; |
04a87e34 YS |
8622 | |
8623 | stat->result_mask |= STATX_BTIME; | |
8624 | stat->btime.tv_sec = BTRFS_I(inode)->i_otime.tv_sec; | |
8625 | stat->btime.tv_nsec = BTRFS_I(inode)->i_otime.tv_nsec; | |
8626 | if (bi_flags & BTRFS_INODE_APPEND) | |
8627 | stat->attributes |= STATX_ATTR_APPEND; | |
8628 | if (bi_flags & BTRFS_INODE_COMPRESS) | |
8629 | stat->attributes |= STATX_ATTR_COMPRESSED; | |
8630 | if (bi_flags & BTRFS_INODE_IMMUTABLE) | |
8631 | stat->attributes |= STATX_ATTR_IMMUTABLE; | |
8632 | if (bi_flags & BTRFS_INODE_NODUMP) | |
8633 | stat->attributes |= STATX_ATTR_NODUMP; | |
14605409 BB |
8634 | if (bi_ro_flags & BTRFS_INODE_RO_VERITY) |
8635 | stat->attributes |= STATX_ATTR_VERITY; | |
04a87e34 YS |
8636 | |
8637 | stat->attributes_mask |= (STATX_ATTR_APPEND | | |
8638 | STATX_ATTR_COMPRESSED | | |
8639 | STATX_ATTR_IMMUTABLE | | |
8640 | STATX_ATTR_NODUMP); | |
fadc0d8b | 8641 | |
0d72b928 | 8642 | generic_fillattr(idmap, request_mask, inode, stat); |
0ee5dc67 | 8643 | stat->dev = BTRFS_I(inode)->root->anon_dev; |
df0af1a5 MX |
8644 | |
8645 | spin_lock(&BTRFS_I(inode)->lock); | |
a7e3b975 | 8646 | delalloc_bytes = BTRFS_I(inode)->new_delalloc_bytes; |
2766ff61 | 8647 | inode_bytes = inode_get_bytes(inode); |
df0af1a5 | 8648 | spin_unlock(&BTRFS_I(inode)->lock); |
2766ff61 | 8649 | stat->blocks = (ALIGN(inode_bytes, blocksize) + |
29e70be2 | 8650 | ALIGN(delalloc_bytes, blocksize)) >> SECTOR_SHIFT; |
39279cc3 CM |
8651 | return 0; |
8652 | } | |
8653 | ||
cdd1fedf DF |
8654 | static int btrfs_rename_exchange(struct inode *old_dir, |
8655 | struct dentry *old_dentry, | |
8656 | struct inode *new_dir, | |
8657 | struct dentry *new_dentry) | |
8658 | { | |
0b246afa | 8659 | struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb); |
cdd1fedf | 8660 | struct btrfs_trans_handle *trans; |
c1621871 | 8661 | unsigned int trans_num_items; |
cdd1fedf DF |
8662 | struct btrfs_root *root = BTRFS_I(old_dir)->root; |
8663 | struct btrfs_root *dest = BTRFS_I(new_dir)->root; | |
8664 | struct inode *new_inode = new_dentry->d_inode; | |
8665 | struct inode *old_inode = old_dentry->d_inode; | |
88d2beec FM |
8666 | struct btrfs_rename_ctx old_rename_ctx; |
8667 | struct btrfs_rename_ctx new_rename_ctx; | |
4a0cc7ca NB |
8668 | u64 old_ino = btrfs_ino(BTRFS_I(old_inode)); |
8669 | u64 new_ino = btrfs_ino(BTRFS_I(new_inode)); | |
cdd1fedf DF |
8670 | u64 old_idx = 0; |
8671 | u64 new_idx = 0; | |
cdd1fedf | 8672 | int ret; |
75b463d2 | 8673 | int ret2; |
dc09ef35 | 8674 | bool need_abort = false; |
ab3c5c18 | 8675 | struct fscrypt_name old_fname, new_fname; |
6db75318 | 8676 | struct fscrypt_str *old_name, *new_name; |
cdd1fedf | 8677 | |
3f79f6f6 N |
8678 | /* |
8679 | * For non-subvolumes allow exchange only within one subvolume, in the | |
8680 | * same inode namespace. Two subvolumes (represented as directory) can | |
8681 | * be exchanged as they're a logical link and have a fixed inode number. | |
8682 | */ | |
8683 | if (root != dest && | |
8684 | (old_ino != BTRFS_FIRST_FREE_OBJECTID || | |
8685 | new_ino != BTRFS_FIRST_FREE_OBJECTID)) | |
cdd1fedf DF |
8686 | return -EXDEV; |
8687 | ||
ab3c5c18 STD |
8688 | ret = fscrypt_setup_filename(old_dir, &old_dentry->d_name, 0, &old_fname); |
8689 | if (ret) | |
8690 | return ret; | |
8691 | ||
8692 | ret = fscrypt_setup_filename(new_dir, &new_dentry->d_name, 0, &new_fname); | |
8693 | if (ret) { | |
8694 | fscrypt_free_filename(&old_fname); | |
8695 | return ret; | |
8696 | } | |
8697 | ||
6db75318 STD |
8698 | old_name = &old_fname.disk_name; |
8699 | new_name = &new_fname.disk_name; | |
ab3c5c18 | 8700 | |
cdd1fedf | 8701 | /* close the race window with snapshot create/destroy ioctl */ |
943eb3bf JB |
8702 | if (old_ino == BTRFS_FIRST_FREE_OBJECTID || |
8703 | new_ino == BTRFS_FIRST_FREE_OBJECTID) | |
0b246afa | 8704 | down_read(&fs_info->subvol_sem); |
cdd1fedf DF |
8705 | |
8706 | /* | |
c1621871 OS |
8707 | * For each inode: |
8708 | * 1 to remove old dir item | |
8709 | * 1 to remove old dir index | |
8710 | * 1 to add new dir item | |
8711 | * 1 to add new dir index | |
8712 | * 1 to update parent inode | |
8713 | * | |
8714 | * If the parents are the same, we only need to account for one | |
cdd1fedf | 8715 | */ |
c1621871 OS |
8716 | trans_num_items = (old_dir == new_dir ? 9 : 10); |
8717 | if (old_ino == BTRFS_FIRST_FREE_OBJECTID) { | |
8718 | /* | |
8719 | * 1 to remove old root ref | |
8720 | * 1 to remove old root backref | |
8721 | * 1 to add new root ref | |
8722 | * 1 to add new root backref | |
8723 | */ | |
8724 | trans_num_items += 4; | |
8725 | } else { | |
8726 | /* | |
8727 | * 1 to update inode item | |
8728 | * 1 to remove old inode ref | |
8729 | * 1 to add new inode ref | |
8730 | */ | |
8731 | trans_num_items += 3; | |
8732 | } | |
8733 | if (new_ino == BTRFS_FIRST_FREE_OBJECTID) | |
8734 | trans_num_items += 4; | |
8735 | else | |
8736 | trans_num_items += 3; | |
8737 | trans = btrfs_start_transaction(root, trans_num_items); | |
cdd1fedf DF |
8738 | if (IS_ERR(trans)) { |
8739 | ret = PTR_ERR(trans); | |
8740 | goto out_notrans; | |
8741 | } | |
8742 | ||
00aa8e87 JB |
8743 | if (dest != root) { |
8744 | ret = btrfs_record_root_in_trans(trans, dest); | |
8745 | if (ret) | |
8746 | goto out_fail; | |
8747 | } | |
3e174099 | 8748 | |
cdd1fedf DF |
8749 | /* |
8750 | * We need to find a free sequence number both in the source and | |
8751 | * in the destination directory for the exchange. | |
8752 | */ | |
877574e2 | 8753 | ret = btrfs_set_inode_index(BTRFS_I(new_dir), &old_idx); |
cdd1fedf DF |
8754 | if (ret) |
8755 | goto out_fail; | |
877574e2 | 8756 | ret = btrfs_set_inode_index(BTRFS_I(old_dir), &new_idx); |
cdd1fedf DF |
8757 | if (ret) |
8758 | goto out_fail; | |
8759 | ||
8760 | BTRFS_I(old_inode)->dir_index = 0ULL; | |
8761 | BTRFS_I(new_inode)->dir_index = 0ULL; | |
8762 | ||
8763 | /* Reference for the source. */ | |
8764 | if (old_ino == BTRFS_FIRST_FREE_OBJECTID) { | |
8765 | /* force full log commit if subvolume involved. */ | |
90787766 | 8766 | btrfs_set_log_full_commit(trans); |
cdd1fedf | 8767 | } else { |
6db75318 | 8768 | ret = btrfs_insert_inode_ref(trans, dest, new_name, old_ino, |
f85b7379 DS |
8769 | btrfs_ino(BTRFS_I(new_dir)), |
8770 | old_idx); | |
cdd1fedf DF |
8771 | if (ret) |
8772 | goto out_fail; | |
dc09ef35 | 8773 | need_abort = true; |
cdd1fedf DF |
8774 | } |
8775 | ||
8776 | /* And now for the dest. */ | |
8777 | if (new_ino == BTRFS_FIRST_FREE_OBJECTID) { | |
8778 | /* force full log commit if subvolume involved. */ | |
90787766 | 8779 | btrfs_set_log_full_commit(trans); |
cdd1fedf | 8780 | } else { |
6db75318 | 8781 | ret = btrfs_insert_inode_ref(trans, root, old_name, new_ino, |
f85b7379 DS |
8782 | btrfs_ino(BTRFS_I(old_dir)), |
8783 | new_idx); | |
dc09ef35 JB |
8784 | if (ret) { |
8785 | if (need_abort) | |
8786 | btrfs_abort_transaction(trans, ret); | |
cdd1fedf | 8787 | goto out_fail; |
dc09ef35 | 8788 | } |
cdd1fedf DF |
8789 | } |
8790 | ||
8791 | /* Update inode version and ctime/mtime. */ | |
8792 | inode_inc_iversion(old_dir); | |
8793 | inode_inc_iversion(new_dir); | |
8794 | inode_inc_iversion(old_inode); | |
8795 | inode_inc_iversion(new_inode); | |
130f1eca | 8796 | simple_rename_timestamp(old_dir, old_dentry, new_dir, new_dentry); |
cdd1fedf DF |
8797 | |
8798 | if (old_dentry->d_parent != new_dentry->d_parent) { | |
f85b7379 | 8799 | btrfs_record_unlink_dir(trans, BTRFS_I(old_dir), |
59fcf388 | 8800 | BTRFS_I(old_inode), true); |
f85b7379 | 8801 | btrfs_record_unlink_dir(trans, BTRFS_I(new_dir), |
59fcf388 | 8802 | BTRFS_I(new_inode), true); |
cdd1fedf DF |
8803 | } |
8804 | ||
8805 | /* src is a subvolume */ | |
8806 | if (old_ino == BTRFS_FIRST_FREE_OBJECTID) { | |
5b7544cb | 8807 | ret = btrfs_unlink_subvol(trans, BTRFS_I(old_dir), old_dentry); |
cdd1fedf | 8808 | } else { /* src is an inode */ |
4467af88 | 8809 | ret = __btrfs_unlink_inode(trans, BTRFS_I(old_dir), |
4ec5934e | 8810 | BTRFS_I(old_dentry->d_inode), |
6db75318 | 8811 | old_name, &old_rename_ctx); |
cdd1fedf | 8812 | if (!ret) |
9a56fcd1 | 8813 | ret = btrfs_update_inode(trans, root, BTRFS_I(old_inode)); |
cdd1fedf DF |
8814 | } |
8815 | if (ret) { | |
66642832 | 8816 | btrfs_abort_transaction(trans, ret); |
cdd1fedf DF |
8817 | goto out_fail; |
8818 | } | |
8819 | ||
8820 | /* dest is a subvolume */ | |
8821 | if (new_ino == BTRFS_FIRST_FREE_OBJECTID) { | |
5b7544cb | 8822 | ret = btrfs_unlink_subvol(trans, BTRFS_I(new_dir), new_dentry); |
cdd1fedf | 8823 | } else { /* dest is an inode */ |
4467af88 | 8824 | ret = __btrfs_unlink_inode(trans, BTRFS_I(new_dir), |
4ec5934e | 8825 | BTRFS_I(new_dentry->d_inode), |
6db75318 | 8826 | new_name, &new_rename_ctx); |
cdd1fedf | 8827 | if (!ret) |
9a56fcd1 | 8828 | ret = btrfs_update_inode(trans, dest, BTRFS_I(new_inode)); |
cdd1fedf DF |
8829 | } |
8830 | if (ret) { | |
66642832 | 8831 | btrfs_abort_transaction(trans, ret); |
cdd1fedf DF |
8832 | goto out_fail; |
8833 | } | |
8834 | ||
db0a669f | 8835 | ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode), |
6db75318 | 8836 | new_name, 0, old_idx); |
cdd1fedf | 8837 | if (ret) { |
66642832 | 8838 | btrfs_abort_transaction(trans, ret); |
cdd1fedf DF |
8839 | goto out_fail; |
8840 | } | |
8841 | ||
db0a669f | 8842 | ret = btrfs_add_link(trans, BTRFS_I(old_dir), BTRFS_I(new_inode), |
6db75318 | 8843 | old_name, 0, new_idx); |
cdd1fedf | 8844 | if (ret) { |
66642832 | 8845 | btrfs_abort_transaction(trans, ret); |
cdd1fedf DF |
8846 | goto out_fail; |
8847 | } | |
8848 | ||
8849 | if (old_inode->i_nlink == 1) | |
8850 | BTRFS_I(old_inode)->dir_index = old_idx; | |
8851 | if (new_inode->i_nlink == 1) | |
8852 | BTRFS_I(new_inode)->dir_index = new_idx; | |
8853 | ||
259c4b96 FM |
8854 | /* |
8855 | * Now pin the logs of the roots. We do it to ensure that no other task | |
8856 | * can sync the logs while we are in progress with the rename, because | |
8857 | * that could result in an inconsistency in case any of the inodes that | |
8858 | * are part of this rename operation were logged before. | |
8859 | */ | |
8860 | if (old_ino != BTRFS_FIRST_FREE_OBJECTID) | |
8861 | btrfs_pin_log_trans(root); | |
8862 | if (new_ino != BTRFS_FIRST_FREE_OBJECTID) | |
8863 | btrfs_pin_log_trans(dest); | |
8864 | ||
8865 | /* Do the log updates for all inodes. */ | |
8866 | if (old_ino != BTRFS_FIRST_FREE_OBJECTID) | |
d5f5bd54 | 8867 | btrfs_log_new_name(trans, old_dentry, BTRFS_I(old_dir), |
88d2beec | 8868 | old_rename_ctx.index, new_dentry->d_parent); |
259c4b96 | 8869 | if (new_ino != BTRFS_FIRST_FREE_OBJECTID) |
d5f5bd54 | 8870 | btrfs_log_new_name(trans, new_dentry, BTRFS_I(new_dir), |
88d2beec | 8871 | new_rename_ctx.index, old_dentry->d_parent); |
259c4b96 FM |
8872 | |
8873 | /* Now unpin the logs. */ | |
8874 | if (old_ino != BTRFS_FIRST_FREE_OBJECTID) | |
8875 | btrfs_end_log_trans(root); | |
8876 | if (new_ino != BTRFS_FIRST_FREE_OBJECTID) | |
cdd1fedf | 8877 | btrfs_end_log_trans(dest); |
cdd1fedf | 8878 | out_fail: |
75b463d2 FM |
8879 | ret2 = btrfs_end_transaction(trans); |
8880 | ret = ret ? ret : ret2; | |
cdd1fedf | 8881 | out_notrans: |
943eb3bf JB |
8882 | if (new_ino == BTRFS_FIRST_FREE_OBJECTID || |
8883 | old_ino == BTRFS_FIRST_FREE_OBJECTID) | |
0b246afa | 8884 | up_read(&fs_info->subvol_sem); |
cdd1fedf | 8885 | |
ab3c5c18 STD |
8886 | fscrypt_free_filename(&new_fname); |
8887 | fscrypt_free_filename(&old_fname); | |
cdd1fedf DF |
8888 | return ret; |
8889 | } | |
8890 | ||
f2d40141 | 8891 | static struct inode *new_whiteout_inode(struct mnt_idmap *idmap, |
a1fd0c35 OS |
8892 | struct inode *dir) |
8893 | { | |
8894 | struct inode *inode; | |
8895 | ||
8896 | inode = new_inode(dir->i_sb); | |
8897 | if (inode) { | |
f2d40141 | 8898 | inode_init_owner(idmap, inode, dir, |
a1fd0c35 OS |
8899 | S_IFCHR | WHITEOUT_MODE); |
8900 | inode->i_op = &btrfs_special_inode_operations; | |
8901 | init_special_inode(inode, inode->i_mode, WHITEOUT_DEV); | |
8902 | } | |
8903 | return inode; | |
8904 | } | |
8905 | ||
f2d40141 | 8906 | static int btrfs_rename(struct mnt_idmap *idmap, |
ca07274c CB |
8907 | struct inode *old_dir, struct dentry *old_dentry, |
8908 | struct inode *new_dir, struct dentry *new_dentry, | |
8909 | unsigned int flags) | |
39279cc3 | 8910 | { |
0b246afa | 8911 | struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb); |
3538d68d OS |
8912 | struct btrfs_new_inode_args whiteout_args = { |
8913 | .dir = old_dir, | |
8914 | .dentry = old_dentry, | |
8915 | }; | |
39279cc3 | 8916 | struct btrfs_trans_handle *trans; |
5062af35 | 8917 | unsigned int trans_num_items; |
39279cc3 | 8918 | struct btrfs_root *root = BTRFS_I(old_dir)->root; |
4df27c4d | 8919 | struct btrfs_root *dest = BTRFS_I(new_dir)->root; |
2b0143b5 DH |
8920 | struct inode *new_inode = d_inode(new_dentry); |
8921 | struct inode *old_inode = d_inode(old_dentry); | |
88d2beec | 8922 | struct btrfs_rename_ctx rename_ctx; |
00e4e6b3 | 8923 | u64 index = 0; |
39279cc3 | 8924 | int ret; |
75b463d2 | 8925 | int ret2; |
4a0cc7ca | 8926 | u64 old_ino = btrfs_ino(BTRFS_I(old_inode)); |
ab3c5c18 | 8927 | struct fscrypt_name old_fname, new_fname; |
39279cc3 | 8928 | |
4a0cc7ca | 8929 | if (btrfs_ino(BTRFS_I(new_dir)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) |
f679a840 YZ |
8930 | return -EPERM; |
8931 | ||
4df27c4d | 8932 | /* we only allow rename subvolume link between subvolumes */ |
33345d01 | 8933 | if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest) |
3394e160 CM |
8934 | return -EXDEV; |
8935 | ||
33345d01 | 8936 | if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID || |
4a0cc7ca | 8937 | (new_inode && btrfs_ino(BTRFS_I(new_inode)) == BTRFS_FIRST_FREE_OBJECTID)) |
39279cc3 | 8938 | return -ENOTEMPTY; |
5f39d397 | 8939 | |
4df27c4d YZ |
8940 | if (S_ISDIR(old_inode->i_mode) && new_inode && |
8941 | new_inode->i_size > BTRFS_EMPTY_DIR_SIZE) | |
8942 | return -ENOTEMPTY; | |
9c52057c | 8943 | |
ab3c5c18 STD |
8944 | ret = fscrypt_setup_filename(old_dir, &old_dentry->d_name, 0, &old_fname); |
8945 | if (ret) | |
8946 | return ret; | |
9c52057c | 8947 | |
ab3c5c18 STD |
8948 | ret = fscrypt_setup_filename(new_dir, &new_dentry->d_name, 0, &new_fname); |
8949 | if (ret) { | |
8950 | fscrypt_free_filename(&old_fname); | |
8951 | return ret; | |
8952 | } | |
9c52057c | 8953 | |
9c52057c | 8954 | /* check for collisions, even if the name isn't there */ |
6db75318 | 8955 | ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino, &new_fname.disk_name); |
9c52057c CM |
8956 | if (ret) { |
8957 | if (ret == -EEXIST) { | |
8958 | /* we shouldn't get | |
8959 | * eexist without a new_inode */ | |
fae7f21c | 8960 | if (WARN_ON(!new_inode)) { |
ab3c5c18 | 8961 | goto out_fscrypt_names; |
9c52057c CM |
8962 | } |
8963 | } else { | |
8964 | /* maybe -EOVERFLOW */ | |
ab3c5c18 | 8965 | goto out_fscrypt_names; |
9c52057c CM |
8966 | } |
8967 | } | |
8968 | ret = 0; | |
8969 | ||
5a3f23d5 | 8970 | /* |
8d875f95 CM |
8971 | * we're using rename to replace one file with another. Start IO on it |
8972 | * now so we don't add too much work to the end of the transaction | |
5a3f23d5 | 8973 | */ |
8d875f95 | 8974 | if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size) |
5a3f23d5 CM |
8975 | filemap_flush(old_inode->i_mapping); |
8976 | ||
a1fd0c35 | 8977 | if (flags & RENAME_WHITEOUT) { |
f2d40141 | 8978 | whiteout_args.inode = new_whiteout_inode(idmap, old_dir); |
abe3bf74 CJ |
8979 | if (!whiteout_args.inode) { |
8980 | ret = -ENOMEM; | |
8981 | goto out_fscrypt_names; | |
8982 | } | |
3538d68d OS |
8983 | ret = btrfs_new_inode_prepare(&whiteout_args, &trans_num_items); |
8984 | if (ret) | |
8985 | goto out_whiteout_inode; | |
8986 | } else { | |
8987 | /* 1 to update the old parent inode. */ | |
8988 | trans_num_items = 1; | |
a1fd0c35 OS |
8989 | } |
8990 | ||
c1621871 OS |
8991 | if (old_ino == BTRFS_FIRST_FREE_OBJECTID) { |
8992 | /* Close the race window with snapshot create/destroy ioctl */ | |
0b246afa | 8993 | down_read(&fs_info->subvol_sem); |
c1621871 OS |
8994 | /* |
8995 | * 1 to remove old root ref | |
8996 | * 1 to remove old root backref | |
8997 | * 1 to add new root ref | |
8998 | * 1 to add new root backref | |
8999 | */ | |
3538d68d | 9000 | trans_num_items += 4; |
c1621871 OS |
9001 | } else { |
9002 | /* | |
9003 | * 1 to update inode | |
9004 | * 1 to remove old inode ref | |
9005 | * 1 to add new inode ref | |
9006 | */ | |
3538d68d | 9007 | trans_num_items += 3; |
c1621871 | 9008 | } |
a22285a6 | 9009 | /* |
c1621871 OS |
9010 | * 1 to remove old dir item |
9011 | * 1 to remove old dir index | |
c1621871 OS |
9012 | * 1 to add new dir item |
9013 | * 1 to add new dir index | |
a22285a6 | 9014 | */ |
3538d68d OS |
9015 | trans_num_items += 4; |
9016 | /* 1 to update new parent inode if it's not the same as the old parent */ | |
c1621871 OS |
9017 | if (new_dir != old_dir) |
9018 | trans_num_items++; | |
9019 | if (new_inode) { | |
9020 | /* | |
9021 | * 1 to update inode | |
9022 | * 1 to remove inode ref | |
9023 | * 1 to remove dir item | |
9024 | * 1 to remove dir index | |
9025 | * 1 to possibly add orphan item | |
9026 | */ | |
9027 | trans_num_items += 5; | |
9028 | } | |
5062af35 | 9029 | trans = btrfs_start_transaction(root, trans_num_items); |
b44c59a8 | 9030 | if (IS_ERR(trans)) { |
cdd1fedf DF |
9031 | ret = PTR_ERR(trans); |
9032 | goto out_notrans; | |
9033 | } | |
76dda93c | 9034 | |
b0fec6fd JB |
9035 | if (dest != root) { |
9036 | ret = btrfs_record_root_in_trans(trans, dest); | |
9037 | if (ret) | |
9038 | goto out_fail; | |
9039 | } | |
5f39d397 | 9040 | |
877574e2 | 9041 | ret = btrfs_set_inode_index(BTRFS_I(new_dir), &index); |
a5719521 YZ |
9042 | if (ret) |
9043 | goto out_fail; | |
5a3f23d5 | 9044 | |
67de1176 | 9045 | BTRFS_I(old_inode)->dir_index = 0ULL; |
33345d01 | 9046 | if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) { |
4df27c4d | 9047 | /* force full log commit if subvolume involved. */ |
90787766 | 9048 | btrfs_set_log_full_commit(trans); |
4df27c4d | 9049 | } else { |
6db75318 STD |
9050 | ret = btrfs_insert_inode_ref(trans, dest, &new_fname.disk_name, |
9051 | old_ino, btrfs_ino(BTRFS_I(new_dir)), | |
9052 | index); | |
a5719521 YZ |
9053 | if (ret) |
9054 | goto out_fail; | |
4df27c4d | 9055 | } |
5a3f23d5 | 9056 | |
0c4d2d95 JB |
9057 | inode_inc_iversion(old_dir); |
9058 | inode_inc_iversion(new_dir); | |
9059 | inode_inc_iversion(old_inode); | |
130f1eca | 9060 | simple_rename_timestamp(old_dir, old_dentry, new_dir, new_dentry); |
5f39d397 | 9061 | |
12fcfd22 | 9062 | if (old_dentry->d_parent != new_dentry->d_parent) |
f85b7379 | 9063 | btrfs_record_unlink_dir(trans, BTRFS_I(old_dir), |
59fcf388 | 9064 | BTRFS_I(old_inode), true); |
12fcfd22 | 9065 | |
33345d01 | 9066 | if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) { |
5b7544cb | 9067 | ret = btrfs_unlink_subvol(trans, BTRFS_I(old_dir), old_dentry); |
4df27c4d | 9068 | } else { |
4467af88 | 9069 | ret = __btrfs_unlink_inode(trans, BTRFS_I(old_dir), |
e43eec81 | 9070 | BTRFS_I(d_inode(old_dentry)), |
6db75318 | 9071 | &old_fname.disk_name, &rename_ctx); |
92986796 | 9072 | if (!ret) |
9a56fcd1 | 9073 | ret = btrfs_update_inode(trans, root, BTRFS_I(old_inode)); |
4df27c4d | 9074 | } |
79787eaa | 9075 | if (ret) { |
66642832 | 9076 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
9077 | goto out_fail; |
9078 | } | |
39279cc3 CM |
9079 | |
9080 | if (new_inode) { | |
0c4d2d95 | 9081 | inode_inc_iversion(new_inode); |
4a0cc7ca | 9082 | if (unlikely(btrfs_ino(BTRFS_I(new_inode)) == |
4df27c4d | 9083 | BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) { |
5b7544cb | 9084 | ret = btrfs_unlink_subvol(trans, BTRFS_I(new_dir), new_dentry); |
4df27c4d YZ |
9085 | BUG_ON(new_inode->i_nlink == 0); |
9086 | } else { | |
4467af88 | 9087 | ret = btrfs_unlink_inode(trans, BTRFS_I(new_dir), |
4ec5934e | 9088 | BTRFS_I(d_inode(new_dentry)), |
6db75318 | 9089 | &new_fname.disk_name); |
4df27c4d | 9090 | } |
4ef31a45 | 9091 | if (!ret && new_inode->i_nlink == 0) |
73f2e545 NB |
9092 | ret = btrfs_orphan_add(trans, |
9093 | BTRFS_I(d_inode(new_dentry))); | |
79787eaa | 9094 | if (ret) { |
66642832 | 9095 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
9096 | goto out_fail; |
9097 | } | |
39279cc3 | 9098 | } |
aec7477b | 9099 | |
db0a669f | 9100 | ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode), |
6db75318 | 9101 | &new_fname.disk_name, 0, index); |
79787eaa | 9102 | if (ret) { |
66642832 | 9103 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
9104 | goto out_fail; |
9105 | } | |
39279cc3 | 9106 | |
67de1176 MX |
9107 | if (old_inode->i_nlink == 1) |
9108 | BTRFS_I(old_inode)->dir_index = index; | |
9109 | ||
259c4b96 | 9110 | if (old_ino != BTRFS_FIRST_FREE_OBJECTID) |
d5f5bd54 | 9111 | btrfs_log_new_name(trans, old_dentry, BTRFS_I(old_dir), |
88d2beec | 9112 | rename_ctx.index, new_dentry->d_parent); |
cdd1fedf DF |
9113 | |
9114 | if (flags & RENAME_WHITEOUT) { | |
caae78e0 | 9115 | ret = btrfs_create_new_inode(trans, &whiteout_args); |
cdd1fedf | 9116 | if (ret) { |
66642832 | 9117 | btrfs_abort_transaction(trans, ret); |
cdd1fedf | 9118 | goto out_fail; |
caae78e0 OS |
9119 | } else { |
9120 | unlock_new_inode(whiteout_args.inode); | |
9121 | iput(whiteout_args.inode); | |
9122 | whiteout_args.inode = NULL; | |
cdd1fedf | 9123 | } |
4df27c4d | 9124 | } |
39279cc3 | 9125 | out_fail: |
75b463d2 FM |
9126 | ret2 = btrfs_end_transaction(trans); |
9127 | ret = ret ? ret : ret2; | |
b44c59a8 | 9128 | out_notrans: |
33345d01 | 9129 | if (old_ino == BTRFS_FIRST_FREE_OBJECTID) |
0b246afa | 9130 | up_read(&fs_info->subvol_sem); |
a1fd0c35 | 9131 | if (flags & RENAME_WHITEOUT) |
3538d68d OS |
9132 | btrfs_new_inode_args_destroy(&whiteout_args); |
9133 | out_whiteout_inode: | |
9134 | if (flags & RENAME_WHITEOUT) | |
9135 | iput(whiteout_args.inode); | |
ab3c5c18 STD |
9136 | out_fscrypt_names: |
9137 | fscrypt_free_filename(&old_fname); | |
9138 | fscrypt_free_filename(&new_fname); | |
39279cc3 CM |
9139 | return ret; |
9140 | } | |
9141 | ||
e18275ae | 9142 | static int btrfs_rename2(struct mnt_idmap *idmap, struct inode *old_dir, |
549c7297 CB |
9143 | struct dentry *old_dentry, struct inode *new_dir, |
9144 | struct dentry *new_dentry, unsigned int flags) | |
80ace85c | 9145 | { |
ca6dee6b FM |
9146 | int ret; |
9147 | ||
cdd1fedf | 9148 | if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT)) |
80ace85c MS |
9149 | return -EINVAL; |
9150 | ||
cdd1fedf | 9151 | if (flags & RENAME_EXCHANGE) |
ca6dee6b FM |
9152 | ret = btrfs_rename_exchange(old_dir, old_dentry, new_dir, |
9153 | new_dentry); | |
9154 | else | |
f2d40141 | 9155 | ret = btrfs_rename(idmap, old_dir, old_dentry, new_dir, |
ca6dee6b | 9156 | new_dentry, flags); |
cdd1fedf | 9157 | |
ca6dee6b FM |
9158 | btrfs_btree_balance_dirty(BTRFS_I(new_dir)->root->fs_info); |
9159 | ||
9160 | return ret; | |
80ace85c MS |
9161 | } |
9162 | ||
3a2f8c07 NB |
9163 | struct btrfs_delalloc_work { |
9164 | struct inode *inode; | |
9165 | struct completion completion; | |
9166 | struct list_head list; | |
9167 | struct btrfs_work work; | |
9168 | }; | |
9169 | ||
8ccf6f19 MX |
9170 | static void btrfs_run_delalloc_work(struct btrfs_work *work) |
9171 | { | |
9172 | struct btrfs_delalloc_work *delalloc_work; | |
9f23e289 | 9173 | struct inode *inode; |
8ccf6f19 MX |
9174 | |
9175 | delalloc_work = container_of(work, struct btrfs_delalloc_work, | |
9176 | work); | |
9f23e289 | 9177 | inode = delalloc_work->inode; |
30424601 DS |
9178 | filemap_flush(inode->i_mapping); |
9179 | if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT, | |
9180 | &BTRFS_I(inode)->runtime_flags)) | |
9f23e289 | 9181 | filemap_flush(inode->i_mapping); |
8ccf6f19 | 9182 | |
076da91c | 9183 | iput(inode); |
8ccf6f19 MX |
9184 | complete(&delalloc_work->completion); |
9185 | } | |
9186 | ||
3a2f8c07 | 9187 | static struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode) |
8ccf6f19 MX |
9188 | { |
9189 | struct btrfs_delalloc_work *work; | |
9190 | ||
100d5702 | 9191 | work = kmalloc(sizeof(*work), GFP_NOFS); |
8ccf6f19 MX |
9192 | if (!work) |
9193 | return NULL; | |
9194 | ||
9195 | init_completion(&work->completion); | |
9196 | INIT_LIST_HEAD(&work->list); | |
9197 | work->inode = inode; | |
a0cac0ec | 9198 | btrfs_init_work(&work->work, btrfs_run_delalloc_work, NULL, NULL); |
8ccf6f19 MX |
9199 | |
9200 | return work; | |
9201 | } | |
9202 | ||
d352ac68 CM |
9203 | /* |
9204 | * some fairly slow code that needs optimization. This walks the list | |
9205 | * of all the inodes with pending delalloc and forces them to disk. | |
9206 | */ | |
e076ab2a JB |
9207 | static int start_delalloc_inodes(struct btrfs_root *root, |
9208 | struct writeback_control *wbc, bool snapshot, | |
3d45f221 | 9209 | bool in_reclaim_context) |
ea8c2819 | 9210 | { |
ea8c2819 | 9211 | struct btrfs_inode *binode; |
5b21f2ed | 9212 | struct inode *inode; |
8ccf6f19 | 9213 | struct btrfs_delalloc_work *work, *next; |
84af994b RJ |
9214 | LIST_HEAD(works); |
9215 | LIST_HEAD(splice); | |
8ccf6f19 | 9216 | int ret = 0; |
e076ab2a | 9217 | bool full_flush = wbc->nr_to_write == LONG_MAX; |
ea8c2819 | 9218 | |
573bfb72 | 9219 | mutex_lock(&root->delalloc_mutex); |
eb73c1b7 MX |
9220 | spin_lock(&root->delalloc_lock); |
9221 | list_splice_init(&root->delalloc_inodes, &splice); | |
1eafa6c7 MX |
9222 | while (!list_empty(&splice)) { |
9223 | binode = list_entry(splice.next, struct btrfs_inode, | |
ea8c2819 | 9224 | delalloc_inodes); |
1eafa6c7 | 9225 | |
eb73c1b7 MX |
9226 | list_move_tail(&binode->delalloc_inodes, |
9227 | &root->delalloc_inodes); | |
3d45f221 FM |
9228 | |
9229 | if (in_reclaim_context && | |
9230 | test_bit(BTRFS_INODE_NO_DELALLOC_FLUSH, &binode->runtime_flags)) | |
9231 | continue; | |
9232 | ||
5b21f2ed | 9233 | inode = igrab(&binode->vfs_inode); |
df0af1a5 | 9234 | if (!inode) { |
eb73c1b7 | 9235 | cond_resched_lock(&root->delalloc_lock); |
1eafa6c7 | 9236 | continue; |
df0af1a5 | 9237 | } |
eb73c1b7 | 9238 | spin_unlock(&root->delalloc_lock); |
1eafa6c7 | 9239 | |
3cd24c69 EL |
9240 | if (snapshot) |
9241 | set_bit(BTRFS_INODE_SNAPSHOT_FLUSH, | |
9242 | &binode->runtime_flags); | |
e076ab2a JB |
9243 | if (full_flush) { |
9244 | work = btrfs_alloc_delalloc_work(inode); | |
9245 | if (!work) { | |
9246 | iput(inode); | |
9247 | ret = -ENOMEM; | |
9248 | goto out; | |
9249 | } | |
9250 | list_add_tail(&work->list, &works); | |
9251 | btrfs_queue_work(root->fs_info->flush_workers, | |
9252 | &work->work); | |
9253 | } else { | |
b3776305 | 9254 | ret = filemap_fdatawrite_wbc(inode->i_mapping, wbc); |
e55cf7ca | 9255 | btrfs_add_delayed_iput(BTRFS_I(inode)); |
e076ab2a | 9256 | if (ret || wbc->nr_to_write <= 0) |
b4912139 JB |
9257 | goto out; |
9258 | } | |
5b21f2ed | 9259 | cond_resched(); |
eb73c1b7 | 9260 | spin_lock(&root->delalloc_lock); |
ea8c2819 | 9261 | } |
eb73c1b7 | 9262 | spin_unlock(&root->delalloc_lock); |
8c8bee1d | 9263 | |
a1ecaabb | 9264 | out: |
eb73c1b7 MX |
9265 | list_for_each_entry_safe(work, next, &works, list) { |
9266 | list_del_init(&work->list); | |
40012f96 NB |
9267 | wait_for_completion(&work->completion); |
9268 | kfree(work); | |
eb73c1b7 MX |
9269 | } |
9270 | ||
81f1d390 | 9271 | if (!list_empty(&splice)) { |
eb73c1b7 MX |
9272 | spin_lock(&root->delalloc_lock); |
9273 | list_splice_tail(&splice, &root->delalloc_inodes); | |
9274 | spin_unlock(&root->delalloc_lock); | |
9275 | } | |
573bfb72 | 9276 | mutex_unlock(&root->delalloc_mutex); |
eb73c1b7 MX |
9277 | return ret; |
9278 | } | |
1eafa6c7 | 9279 | |
f9baa501 | 9280 | int btrfs_start_delalloc_snapshot(struct btrfs_root *root, bool in_reclaim_context) |
eb73c1b7 | 9281 | { |
e076ab2a JB |
9282 | struct writeback_control wbc = { |
9283 | .nr_to_write = LONG_MAX, | |
9284 | .sync_mode = WB_SYNC_NONE, | |
9285 | .range_start = 0, | |
9286 | .range_end = LLONG_MAX, | |
9287 | }; | |
0b246afa | 9288 | struct btrfs_fs_info *fs_info = root->fs_info; |
1eafa6c7 | 9289 | |
84961539 | 9290 | if (BTRFS_FS_ERROR(fs_info)) |
eb73c1b7 MX |
9291 | return -EROFS; |
9292 | ||
f9baa501 | 9293 | return start_delalloc_inodes(root, &wbc, true, in_reclaim_context); |
eb73c1b7 MX |
9294 | } |
9295 | ||
9db4dc24 | 9296 | int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, long nr, |
3d45f221 | 9297 | bool in_reclaim_context) |
eb73c1b7 | 9298 | { |
e076ab2a | 9299 | struct writeback_control wbc = { |
9db4dc24 | 9300 | .nr_to_write = nr, |
e076ab2a JB |
9301 | .sync_mode = WB_SYNC_NONE, |
9302 | .range_start = 0, | |
9303 | .range_end = LLONG_MAX, | |
9304 | }; | |
eb73c1b7 | 9305 | struct btrfs_root *root; |
84af994b | 9306 | LIST_HEAD(splice); |
eb73c1b7 MX |
9307 | int ret; |
9308 | ||
84961539 | 9309 | if (BTRFS_FS_ERROR(fs_info)) |
eb73c1b7 MX |
9310 | return -EROFS; |
9311 | ||
573bfb72 | 9312 | mutex_lock(&fs_info->delalloc_root_mutex); |
eb73c1b7 MX |
9313 | spin_lock(&fs_info->delalloc_root_lock); |
9314 | list_splice_init(&fs_info->delalloc_roots, &splice); | |
d7830b71 | 9315 | while (!list_empty(&splice)) { |
e076ab2a JB |
9316 | /* |
9317 | * Reset nr_to_write here so we know that we're doing a full | |
9318 | * flush. | |
9319 | */ | |
9db4dc24 | 9320 | if (nr == LONG_MAX) |
e076ab2a JB |
9321 | wbc.nr_to_write = LONG_MAX; |
9322 | ||
eb73c1b7 MX |
9323 | root = list_first_entry(&splice, struct btrfs_root, |
9324 | delalloc_root); | |
00246528 | 9325 | root = btrfs_grab_root(root); |
eb73c1b7 MX |
9326 | BUG_ON(!root); |
9327 | list_move_tail(&root->delalloc_root, | |
9328 | &fs_info->delalloc_roots); | |
9329 | spin_unlock(&fs_info->delalloc_root_lock); | |
9330 | ||
e076ab2a | 9331 | ret = start_delalloc_inodes(root, &wbc, false, in_reclaim_context); |
00246528 | 9332 | btrfs_put_root(root); |
e076ab2a | 9333 | if (ret < 0 || wbc.nr_to_write <= 0) |
eb73c1b7 | 9334 | goto out; |
eb73c1b7 | 9335 | spin_lock(&fs_info->delalloc_root_lock); |
8ccf6f19 | 9336 | } |
eb73c1b7 | 9337 | spin_unlock(&fs_info->delalloc_root_lock); |
1eafa6c7 | 9338 | |
6c255e67 | 9339 | ret = 0; |
eb73c1b7 | 9340 | out: |
81f1d390 | 9341 | if (!list_empty(&splice)) { |
eb73c1b7 MX |
9342 | spin_lock(&fs_info->delalloc_root_lock); |
9343 | list_splice_tail(&splice, &fs_info->delalloc_roots); | |
9344 | spin_unlock(&fs_info->delalloc_root_lock); | |
1eafa6c7 | 9345 | } |
573bfb72 | 9346 | mutex_unlock(&fs_info->delalloc_root_mutex); |
8ccf6f19 | 9347 | return ret; |
ea8c2819 CM |
9348 | } |
9349 | ||
7a77db95 | 9350 | static int btrfs_symlink(struct mnt_idmap *idmap, struct inode *dir, |
549c7297 | 9351 | struct dentry *dentry, const char *symname) |
39279cc3 | 9352 | { |
0b246afa | 9353 | struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb); |
39279cc3 CM |
9354 | struct btrfs_trans_handle *trans; |
9355 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
9356 | struct btrfs_path *path; | |
9357 | struct btrfs_key key; | |
a1fd0c35 | 9358 | struct inode *inode; |
3538d68d OS |
9359 | struct btrfs_new_inode_args new_inode_args = { |
9360 | .dir = dir, | |
9361 | .dentry = dentry, | |
9362 | }; | |
9363 | unsigned int trans_num_items; | |
39279cc3 | 9364 | int err; |
39279cc3 CM |
9365 | int name_len; |
9366 | int datasize; | |
5f39d397 | 9367 | unsigned long ptr; |
39279cc3 | 9368 | struct btrfs_file_extent_item *ei; |
5f39d397 | 9369 | struct extent_buffer *leaf; |
39279cc3 | 9370 | |
f06becc4 | 9371 | name_len = strlen(symname); |
0b246afa | 9372 | if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info)) |
39279cc3 | 9373 | return -ENAMETOOLONG; |
1832a6d5 | 9374 | |
a1fd0c35 OS |
9375 | inode = new_inode(dir->i_sb); |
9376 | if (!inode) | |
9377 | return -ENOMEM; | |
f2d40141 | 9378 | inode_init_owner(idmap, inode, dir, S_IFLNK | S_IRWXUGO); |
a1fd0c35 OS |
9379 | inode->i_op = &btrfs_symlink_inode_operations; |
9380 | inode_nohighmem(inode); | |
9381 | inode->i_mapping->a_ops = &btrfs_aops; | |
caae78e0 OS |
9382 | btrfs_i_size_write(BTRFS_I(inode), name_len); |
9383 | inode_set_bytes(inode, name_len); | |
a1fd0c35 | 9384 | |
3538d68d OS |
9385 | new_inode_args.inode = inode; |
9386 | err = btrfs_new_inode_prepare(&new_inode_args, &trans_num_items); | |
caae78e0 OS |
9387 | if (err) |
9388 | goto out_inode; | |
3538d68d OS |
9389 | /* 1 additional item for the inline extent */ |
9390 | trans_num_items++; | |
9391 | ||
9392 | trans = btrfs_start_transaction(root, trans_num_items); | |
a1fd0c35 | 9393 | if (IS_ERR(trans)) { |
3538d68d OS |
9394 | err = PTR_ERR(trans); |
9395 | goto out_new_inode_args; | |
a1fd0c35 | 9396 | } |
1832a6d5 | 9397 | |
caae78e0 | 9398 | err = btrfs_create_new_inode(trans, &new_inode_args); |
b0d5d10f | 9399 | if (err) |
caae78e0 | 9400 | goto out; |
ad19db71 | 9401 | |
39279cc3 | 9402 | path = btrfs_alloc_path(); |
d8926bb3 MF |
9403 | if (!path) { |
9404 | err = -ENOMEM; | |
caae78e0 OS |
9405 | btrfs_abort_transaction(trans, err); |
9406 | discard_new_inode(inode); | |
9407 | inode = NULL; | |
9408 | goto out; | |
d8926bb3 | 9409 | } |
4a0cc7ca | 9410 | key.objectid = btrfs_ino(BTRFS_I(inode)); |
39279cc3 | 9411 | key.offset = 0; |
962a298f | 9412 | key.type = BTRFS_EXTENT_DATA_KEY; |
39279cc3 CM |
9413 | datasize = btrfs_file_extent_calc_inline_size(name_len); |
9414 | err = btrfs_insert_empty_item(trans, root, path, &key, | |
9415 | datasize); | |
54aa1f4d | 9416 | if (err) { |
caae78e0 | 9417 | btrfs_abort_transaction(trans, err); |
b0839166 | 9418 | btrfs_free_path(path); |
caae78e0 OS |
9419 | discard_new_inode(inode); |
9420 | inode = NULL; | |
9421 | goto out; | |
54aa1f4d | 9422 | } |
5f39d397 CM |
9423 | leaf = path->nodes[0]; |
9424 | ei = btrfs_item_ptr(leaf, path->slots[0], | |
9425 | struct btrfs_file_extent_item); | |
9426 | btrfs_set_file_extent_generation(leaf, ei, trans->transid); | |
9427 | btrfs_set_file_extent_type(leaf, ei, | |
39279cc3 | 9428 | BTRFS_FILE_EXTENT_INLINE); |
c8b97818 CM |
9429 | btrfs_set_file_extent_encryption(leaf, ei, 0); |
9430 | btrfs_set_file_extent_compression(leaf, ei, 0); | |
9431 | btrfs_set_file_extent_other_encoding(leaf, ei, 0); | |
9432 | btrfs_set_file_extent_ram_bytes(leaf, ei, name_len); | |
9433 | ||
39279cc3 | 9434 | ptr = btrfs_file_extent_inline_start(ei); |
5f39d397 CM |
9435 | write_extent_buffer(leaf, symname, ptr, name_len); |
9436 | btrfs_mark_buffer_dirty(leaf); | |
39279cc3 | 9437 | btrfs_free_path(path); |
5f39d397 | 9438 | |
1e2e547a | 9439 | d_instantiate_new(dentry, inode); |
caae78e0 OS |
9440 | err = 0; |
9441 | out: | |
3a45bb20 | 9442 | btrfs_end_transaction(trans); |
2ff7e61e | 9443 | btrfs_btree_balance_dirty(fs_info); |
3538d68d OS |
9444 | out_new_inode_args: |
9445 | btrfs_new_inode_args_destroy(&new_inode_args); | |
caae78e0 OS |
9446 | out_inode: |
9447 | if (err) | |
9448 | iput(inode); | |
39279cc3 CM |
9449 | return err; |
9450 | } | |
16432985 | 9451 | |
8fccebfa FM |
9452 | static struct btrfs_trans_handle *insert_prealloc_file_extent( |
9453 | struct btrfs_trans_handle *trans_in, | |
90dffd0c NB |
9454 | struct btrfs_inode *inode, |
9455 | struct btrfs_key *ins, | |
203f44c5 QW |
9456 | u64 file_offset) |
9457 | { | |
9458 | struct btrfs_file_extent_item stack_fi; | |
bf385648 | 9459 | struct btrfs_replace_extent_info extent_info; |
8fccebfa FM |
9460 | struct btrfs_trans_handle *trans = trans_in; |
9461 | struct btrfs_path *path; | |
203f44c5 QW |
9462 | u64 start = ins->objectid; |
9463 | u64 len = ins->offset; | |
fbf48bb0 | 9464 | int qgroup_released; |
9729f10a | 9465 | int ret; |
203f44c5 QW |
9466 | |
9467 | memset(&stack_fi, 0, sizeof(stack_fi)); | |
9468 | ||
9469 | btrfs_set_stack_file_extent_type(&stack_fi, BTRFS_FILE_EXTENT_PREALLOC); | |
9470 | btrfs_set_stack_file_extent_disk_bytenr(&stack_fi, start); | |
9471 | btrfs_set_stack_file_extent_disk_num_bytes(&stack_fi, len); | |
9472 | btrfs_set_stack_file_extent_num_bytes(&stack_fi, len); | |
9473 | btrfs_set_stack_file_extent_ram_bytes(&stack_fi, len); | |
9474 | btrfs_set_stack_file_extent_compression(&stack_fi, BTRFS_COMPRESS_NONE); | |
9475 | /* Encryption and other encoding is reserved and all 0 */ | |
9476 | ||
fbf48bb0 QW |
9477 | qgroup_released = btrfs_qgroup_release_data(inode, file_offset, len); |
9478 | if (qgroup_released < 0) | |
9479 | return ERR_PTR(qgroup_released); | |
8fccebfa FM |
9480 | |
9481 | if (trans) { | |
90dffd0c | 9482 | ret = insert_reserved_file_extent(trans, inode, |
2766ff61 | 9483 | file_offset, &stack_fi, |
fbf48bb0 | 9484 | true, qgroup_released); |
8fccebfa | 9485 | if (ret) |
a3ee79bd | 9486 | goto free_qgroup; |
8fccebfa FM |
9487 | return trans; |
9488 | } | |
9489 | ||
9490 | extent_info.disk_offset = start; | |
9491 | extent_info.disk_len = len; | |
9492 | extent_info.data_offset = 0; | |
9493 | extent_info.data_len = len; | |
9494 | extent_info.file_offset = file_offset; | |
9495 | extent_info.extent_buf = (char *)&stack_fi; | |
8fccebfa | 9496 | extent_info.is_new_extent = true; |
983d8209 | 9497 | extent_info.update_times = true; |
fbf48bb0 | 9498 | extent_info.qgroup_reserved = qgroup_released; |
8fccebfa FM |
9499 | extent_info.insertions = 0; |
9500 | ||
9501 | path = btrfs_alloc_path(); | |
a3ee79bd QW |
9502 | if (!path) { |
9503 | ret = -ENOMEM; | |
9504 | goto free_qgroup; | |
9505 | } | |
8fccebfa | 9506 | |
bfc78479 | 9507 | ret = btrfs_replace_file_extents(inode, path, file_offset, |
8fccebfa FM |
9508 | file_offset + len - 1, &extent_info, |
9509 | &trans); | |
9510 | btrfs_free_path(path); | |
9511 | if (ret) | |
a3ee79bd | 9512 | goto free_qgroup; |
8fccebfa | 9513 | return trans; |
a3ee79bd QW |
9514 | |
9515 | free_qgroup: | |
9516 | /* | |
9517 | * We have released qgroup data range at the beginning of the function, | |
9518 | * and normally qgroup_released bytes will be freed when committing | |
9519 | * transaction. | |
9520 | * But if we error out early, we have to free what we have released | |
9521 | * or we leak qgroup data reservation. | |
9522 | */ | |
9523 | btrfs_qgroup_free_refroot(inode->root->fs_info, | |
9524 | inode->root->root_key.objectid, qgroup_released, | |
9525 | BTRFS_QGROUP_RSV_DATA); | |
9526 | return ERR_PTR(ret); | |
203f44c5 | 9527 | } |
8fccebfa | 9528 | |
0af3d00b JB |
9529 | static int __btrfs_prealloc_file_range(struct inode *inode, int mode, |
9530 | u64 start, u64 num_bytes, u64 min_size, | |
9531 | loff_t actual_len, u64 *alloc_hint, | |
9532 | struct btrfs_trans_handle *trans) | |
d899e052 | 9533 | { |
0b246afa | 9534 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
5dc562c5 | 9535 | struct extent_map *em; |
d899e052 YZ |
9536 | struct btrfs_root *root = BTRFS_I(inode)->root; |
9537 | struct btrfs_key ins; | |
d899e052 | 9538 | u64 cur_offset = start; |
b778cf96 | 9539 | u64 clear_offset = start; |
55a61d1d | 9540 | u64 i_size; |
154ea289 | 9541 | u64 cur_bytes; |
0b670dc4 | 9542 | u64 last_alloc = (u64)-1; |
d899e052 | 9543 | int ret = 0; |
0af3d00b | 9544 | bool own_trans = true; |
18513091 | 9545 | u64 end = start + num_bytes - 1; |
d899e052 | 9546 | |
0af3d00b JB |
9547 | if (trans) |
9548 | own_trans = false; | |
d899e052 | 9549 | while (num_bytes > 0) { |
ee22184b | 9550 | cur_bytes = min_t(u64, num_bytes, SZ_256M); |
154ea289 | 9551 | cur_bytes = max(cur_bytes, min_size); |
0b670dc4 JB |
9552 | /* |
9553 | * If we are severely fragmented we could end up with really | |
9554 | * small allocations, so if the allocator is returning small | |
9555 | * chunks lets make its job easier by only searching for those | |
9556 | * sized chunks. | |
9557 | */ | |
9558 | cur_bytes = min(cur_bytes, last_alloc); | |
18513091 WX |
9559 | ret = btrfs_reserve_extent(root, cur_bytes, cur_bytes, |
9560 | min_size, 0, *alloc_hint, &ins, 1, 0); | |
8fccebfa | 9561 | if (ret) |
a22285a6 | 9562 | break; |
b778cf96 JB |
9563 | |
9564 | /* | |
9565 | * We've reserved this space, and thus converted it from | |
9566 | * ->bytes_may_use to ->bytes_reserved. Any error that happens | |
9567 | * from here on out we will only need to clear our reservation | |
9568 | * for the remaining unreserved area, so advance our | |
9569 | * clear_offset by our extent size. | |
9570 | */ | |
9571 | clear_offset += ins.offset; | |
5a303d5d | 9572 | |
0b670dc4 | 9573 | last_alloc = ins.offset; |
90dffd0c NB |
9574 | trans = insert_prealloc_file_extent(trans, BTRFS_I(inode), |
9575 | &ins, cur_offset); | |
1afc708d FM |
9576 | /* |
9577 | * Now that we inserted the prealloc extent we can finally | |
9578 | * decrement the number of reservations in the block group. | |
9579 | * If we did it before, we could race with relocation and have | |
9580 | * relocation miss the reserved extent, making it fail later. | |
9581 | */ | |
9582 | btrfs_dec_block_group_reservations(fs_info, ins.objectid); | |
8fccebfa FM |
9583 | if (IS_ERR(trans)) { |
9584 | ret = PTR_ERR(trans); | |
2ff7e61e | 9585 | btrfs_free_reserved_extent(fs_info, ins.objectid, |
e570fd27 | 9586 | ins.offset, 0); |
79787eaa JM |
9587 | break; |
9588 | } | |
31193213 | 9589 | |
5dc562c5 JB |
9590 | em = alloc_extent_map(); |
9591 | if (!em) { | |
a1ba4c08 FM |
9592 | btrfs_drop_extent_map_range(BTRFS_I(inode), cur_offset, |
9593 | cur_offset + ins.offset - 1, false); | |
23e3337f | 9594 | btrfs_set_inode_full_sync(BTRFS_I(inode)); |
5dc562c5 JB |
9595 | goto next; |
9596 | } | |
9597 | ||
9598 | em->start = cur_offset; | |
9599 | em->orig_start = cur_offset; | |
9600 | em->len = ins.offset; | |
9601 | em->block_start = ins.objectid; | |
9602 | em->block_len = ins.offset; | |
b4939680 | 9603 | em->orig_block_len = ins.offset; |
cc95bef6 | 9604 | em->ram_bytes = ins.offset; |
5dc562c5 JB |
9605 | set_bit(EXTENT_FLAG_PREALLOC, &em->flags); |
9606 | em->generation = trans->transid; | |
9607 | ||
a1ba4c08 | 9608 | ret = btrfs_replace_extent_map_range(BTRFS_I(inode), em, true); |
5dc562c5 JB |
9609 | free_extent_map(em); |
9610 | next: | |
d899e052 YZ |
9611 | num_bytes -= ins.offset; |
9612 | cur_offset += ins.offset; | |
efa56464 | 9613 | *alloc_hint = ins.objectid + ins.offset; |
5a303d5d | 9614 | |
0c4d2d95 | 9615 | inode_inc_iversion(inode); |
2a9462de | 9616 | inode_set_ctime_current(inode); |
6cbff00f | 9617 | BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC; |
d899e052 | 9618 | if (!(mode & FALLOC_FL_KEEP_SIZE) && |
efa56464 YZ |
9619 | (actual_len > inode->i_size) && |
9620 | (cur_offset > inode->i_size)) { | |
d1ea6a61 | 9621 | if (cur_offset > actual_len) |
55a61d1d | 9622 | i_size = actual_len; |
d1ea6a61 | 9623 | else |
55a61d1d JB |
9624 | i_size = cur_offset; |
9625 | i_size_write(inode, i_size); | |
76aea537 | 9626 | btrfs_inode_safe_disk_i_size_write(BTRFS_I(inode), 0); |
5a303d5d YZ |
9627 | } |
9628 | ||
9a56fcd1 | 9629 | ret = btrfs_update_inode(trans, root, BTRFS_I(inode)); |
79787eaa JM |
9630 | |
9631 | if (ret) { | |
66642832 | 9632 | btrfs_abort_transaction(trans, ret); |
79787eaa | 9633 | if (own_trans) |
3a45bb20 | 9634 | btrfs_end_transaction(trans); |
79787eaa JM |
9635 | break; |
9636 | } | |
d899e052 | 9637 | |
8fccebfa | 9638 | if (own_trans) { |
3a45bb20 | 9639 | btrfs_end_transaction(trans); |
8fccebfa FM |
9640 | trans = NULL; |
9641 | } | |
5a303d5d | 9642 | } |
b778cf96 | 9643 | if (clear_offset < end) |
25ce28ca | 9644 | btrfs_free_reserved_data_space(BTRFS_I(inode), NULL, clear_offset, |
b778cf96 | 9645 | end - clear_offset + 1); |
d899e052 YZ |
9646 | return ret; |
9647 | } | |
9648 | ||
0af3d00b JB |
9649 | int btrfs_prealloc_file_range(struct inode *inode, int mode, |
9650 | u64 start, u64 num_bytes, u64 min_size, | |
9651 | loff_t actual_len, u64 *alloc_hint) | |
9652 | { | |
9653 | return __btrfs_prealloc_file_range(inode, mode, start, num_bytes, | |
9654 | min_size, actual_len, alloc_hint, | |
9655 | NULL); | |
9656 | } | |
9657 | ||
9658 | int btrfs_prealloc_file_range_trans(struct inode *inode, | |
9659 | struct btrfs_trans_handle *trans, int mode, | |
9660 | u64 start, u64 num_bytes, u64 min_size, | |
9661 | loff_t actual_len, u64 *alloc_hint) | |
9662 | { | |
9663 | return __btrfs_prealloc_file_range(inode, mode, start, num_bytes, | |
9664 | min_size, actual_len, alloc_hint, trans); | |
9665 | } | |
9666 | ||
4609e1f1 | 9667 | static int btrfs_permission(struct mnt_idmap *idmap, |
549c7297 | 9668 | struct inode *inode, int mask) |
fdebe2bd | 9669 | { |
b83cc969 | 9670 | struct btrfs_root *root = BTRFS_I(inode)->root; |
cb6db4e5 | 9671 | umode_t mode = inode->i_mode; |
b83cc969 | 9672 | |
cb6db4e5 JM |
9673 | if (mask & MAY_WRITE && |
9674 | (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) { | |
9675 | if (btrfs_root_readonly(root)) | |
9676 | return -EROFS; | |
9677 | if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY) | |
9678 | return -EACCES; | |
9679 | } | |
4609e1f1 | 9680 | return generic_permission(idmap, inode, mask); |
fdebe2bd | 9681 | } |
39279cc3 | 9682 | |
011e2b71 | 9683 | static int btrfs_tmpfile(struct mnt_idmap *idmap, struct inode *dir, |
863f144f | 9684 | struct file *file, umode_t mode) |
ef3b9af5 | 9685 | { |
2ff7e61e | 9686 | struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb); |
ef3b9af5 FM |
9687 | struct btrfs_trans_handle *trans; |
9688 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
a1fd0c35 | 9689 | struct inode *inode; |
3538d68d OS |
9690 | struct btrfs_new_inode_args new_inode_args = { |
9691 | .dir = dir, | |
863f144f | 9692 | .dentry = file->f_path.dentry, |
3538d68d OS |
9693 | .orphan = true, |
9694 | }; | |
9695 | unsigned int trans_num_items; | |
a1fd0c35 OS |
9696 | int ret; |
9697 | ||
9698 | inode = new_inode(dir->i_sb); | |
9699 | if (!inode) | |
9700 | return -ENOMEM; | |
f2d40141 | 9701 | inode_init_owner(idmap, inode, dir, mode); |
a1fd0c35 OS |
9702 | inode->i_fop = &btrfs_file_operations; |
9703 | inode->i_op = &btrfs_file_inode_operations; | |
9704 | inode->i_mapping->a_ops = &btrfs_aops; | |
ef3b9af5 | 9705 | |
3538d68d OS |
9706 | new_inode_args.inode = inode; |
9707 | ret = btrfs_new_inode_prepare(&new_inode_args, &trans_num_items); | |
caae78e0 OS |
9708 | if (ret) |
9709 | goto out_inode; | |
3538d68d OS |
9710 | |
9711 | trans = btrfs_start_transaction(root, trans_num_items); | |
a1fd0c35 | 9712 | if (IS_ERR(trans)) { |
3538d68d OS |
9713 | ret = PTR_ERR(trans); |
9714 | goto out_new_inode_args; | |
a1fd0c35 | 9715 | } |
ef3b9af5 | 9716 | |
caae78e0 | 9717 | ret = btrfs_create_new_inode(trans, &new_inode_args); |
ef3b9af5 | 9718 | |
5762b5c9 | 9719 | /* |
3538d68d OS |
9720 | * We set number of links to 0 in btrfs_create_new_inode(), and here we |
9721 | * set it to 1 because d_tmpfile() will issue a warning if the count is | |
9722 | * 0, through: | |
5762b5c9 FM |
9723 | * |
9724 | * d_tmpfile() -> inode_dec_link_count() -> drop_nlink() | |
9725 | */ | |
9726 | set_nlink(inode, 1); | |
caae78e0 OS |
9727 | |
9728 | if (!ret) { | |
863f144f | 9729 | d_tmpfile(file, inode); |
caae78e0 OS |
9730 | unlock_new_inode(inode); |
9731 | mark_inode_dirty(inode); | |
9732 | } | |
9733 | ||
3a45bb20 | 9734 | btrfs_end_transaction(trans); |
2ff7e61e | 9735 | btrfs_btree_balance_dirty(fs_info); |
3538d68d OS |
9736 | out_new_inode_args: |
9737 | btrfs_new_inode_args_destroy(&new_inode_args); | |
caae78e0 OS |
9738 | out_inode: |
9739 | if (ret) | |
9740 | iput(inode); | |
863f144f | 9741 | return finish_open_simple(file, ret); |
ef3b9af5 FM |
9742 | } |
9743 | ||
d2a91064 | 9744 | void btrfs_set_range_writeback(struct btrfs_inode *inode, u64 start, u64 end) |
c6100a4b | 9745 | { |
d2a91064 | 9746 | struct btrfs_fs_info *fs_info = inode->root->fs_info; |
c6100a4b JB |
9747 | unsigned long index = start >> PAGE_SHIFT; |
9748 | unsigned long end_index = end >> PAGE_SHIFT; | |
9749 | struct page *page; | |
d2a91064 | 9750 | u32 len; |
c6100a4b | 9751 | |
d2a91064 QW |
9752 | ASSERT(end + 1 - start <= U32_MAX); |
9753 | len = end + 1 - start; | |
c6100a4b | 9754 | while (index <= end_index) { |
d2a91064 | 9755 | page = find_get_page(inode->vfs_inode.i_mapping, index); |
c6100a4b | 9756 | ASSERT(page); /* Pages should be in the extent_io_tree */ |
d2a91064 QW |
9757 | |
9758 | btrfs_page_set_writeback(fs_info, page, start, len); | |
c6100a4b JB |
9759 | put_page(page); |
9760 | index++; | |
9761 | } | |
9762 | } | |
9763 | ||
3ea4dc5b OS |
9764 | int btrfs_encoded_io_compression_from_extent(struct btrfs_fs_info *fs_info, |
9765 | int compress_type) | |
1881fba8 OS |
9766 | { |
9767 | switch (compress_type) { | |
9768 | case BTRFS_COMPRESS_NONE: | |
9769 | return BTRFS_ENCODED_IO_COMPRESSION_NONE; | |
9770 | case BTRFS_COMPRESS_ZLIB: | |
9771 | return BTRFS_ENCODED_IO_COMPRESSION_ZLIB; | |
9772 | case BTRFS_COMPRESS_LZO: | |
9773 | /* | |
9774 | * The LZO format depends on the sector size. 64K is the maximum | |
9775 | * sector size that we support. | |
9776 | */ | |
9777 | if (fs_info->sectorsize < SZ_4K || fs_info->sectorsize > SZ_64K) | |
9778 | return -EINVAL; | |
9779 | return BTRFS_ENCODED_IO_COMPRESSION_LZO_4K + | |
9780 | (fs_info->sectorsize_bits - 12); | |
9781 | case BTRFS_COMPRESS_ZSTD: | |
9782 | return BTRFS_ENCODED_IO_COMPRESSION_ZSTD; | |
9783 | default: | |
9784 | return -EUCLEAN; | |
9785 | } | |
9786 | } | |
9787 | ||
9788 | static ssize_t btrfs_encoded_read_inline( | |
9789 | struct kiocb *iocb, | |
9790 | struct iov_iter *iter, u64 start, | |
9791 | u64 lockend, | |
9792 | struct extent_state **cached_state, | |
9793 | u64 extent_start, size_t count, | |
9794 | struct btrfs_ioctl_encoded_io_args *encoded, | |
9795 | bool *unlocked) | |
9796 | { | |
9797 | struct btrfs_inode *inode = BTRFS_I(file_inode(iocb->ki_filp)); | |
9798 | struct btrfs_root *root = inode->root; | |
9799 | struct btrfs_fs_info *fs_info = root->fs_info; | |
9800 | struct extent_io_tree *io_tree = &inode->io_tree; | |
9801 | struct btrfs_path *path; | |
9802 | struct extent_buffer *leaf; | |
9803 | struct btrfs_file_extent_item *item; | |
9804 | u64 ram_bytes; | |
9805 | unsigned long ptr; | |
9806 | void *tmp; | |
9807 | ssize_t ret; | |
9808 | ||
9809 | path = btrfs_alloc_path(); | |
9810 | if (!path) { | |
9811 | ret = -ENOMEM; | |
9812 | goto out; | |
9813 | } | |
9814 | ret = btrfs_lookup_file_extent(NULL, root, path, btrfs_ino(inode), | |
9815 | extent_start, 0); | |
9816 | if (ret) { | |
9817 | if (ret > 0) { | |
9818 | /* The extent item disappeared? */ | |
9819 | ret = -EIO; | |
9820 | } | |
9821 | goto out; | |
9822 | } | |
9823 | leaf = path->nodes[0]; | |
9824 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_file_extent_item); | |
9825 | ||
9826 | ram_bytes = btrfs_file_extent_ram_bytes(leaf, item); | |
9827 | ptr = btrfs_file_extent_inline_start(item); | |
9828 | ||
9829 | encoded->len = min_t(u64, extent_start + ram_bytes, | |
9830 | inode->vfs_inode.i_size) - iocb->ki_pos; | |
9831 | ret = btrfs_encoded_io_compression_from_extent(fs_info, | |
9832 | btrfs_file_extent_compression(leaf, item)); | |
9833 | if (ret < 0) | |
9834 | goto out; | |
9835 | encoded->compression = ret; | |
9836 | if (encoded->compression) { | |
9837 | size_t inline_size; | |
9838 | ||
9839 | inline_size = btrfs_file_extent_inline_item_len(leaf, | |
9840 | path->slots[0]); | |
9841 | if (inline_size > count) { | |
9842 | ret = -ENOBUFS; | |
9843 | goto out; | |
9844 | } | |
9845 | count = inline_size; | |
9846 | encoded->unencoded_len = ram_bytes; | |
9847 | encoded->unencoded_offset = iocb->ki_pos - extent_start; | |
9848 | } else { | |
9849 | count = min_t(u64, count, encoded->len); | |
9850 | encoded->len = count; | |
9851 | encoded->unencoded_len = count; | |
9852 | ptr += iocb->ki_pos - extent_start; | |
9853 | } | |
9854 | ||
9855 | tmp = kmalloc(count, GFP_NOFS); | |
9856 | if (!tmp) { | |
9857 | ret = -ENOMEM; | |
9858 | goto out; | |
9859 | } | |
9860 | read_extent_buffer(leaf, tmp, ptr, count); | |
9861 | btrfs_release_path(path); | |
570eb97b | 9862 | unlock_extent(io_tree, start, lockend, cached_state); |
e5d4d75b | 9863 | btrfs_inode_unlock(inode, BTRFS_ILOCK_SHARED); |
1881fba8 OS |
9864 | *unlocked = true; |
9865 | ||
9866 | ret = copy_to_iter(tmp, count, iter); | |
9867 | if (ret != count) | |
9868 | ret = -EFAULT; | |
9869 | kfree(tmp); | |
9870 | out: | |
9871 | btrfs_free_path(path); | |
9872 | return ret; | |
9873 | } | |
9874 | ||
9875 | struct btrfs_encoded_read_private { | |
1881fba8 OS |
9876 | wait_queue_head_t wait; |
9877 | atomic_t pending; | |
9878 | blk_status_t status; | |
1881fba8 OS |
9879 | }; |
9880 | ||
917f32a2 | 9881 | static void btrfs_encoded_read_endio(struct btrfs_bio *bbio) |
1881fba8 | 9882 | { |
917f32a2 | 9883 | struct btrfs_encoded_read_private *priv = bbio->private; |
1881fba8 | 9884 | |
7609afac | 9885 | if (bbio->bio.bi_status) { |
1881fba8 OS |
9886 | /* |
9887 | * The memory barrier implied by the atomic_dec_return() here | |
9888 | * pairs with the memory barrier implied by the | |
9889 | * atomic_dec_return() or io_wait_event() in | |
9890 | * btrfs_encoded_read_regular_fill_pages() to ensure that this | |
9891 | * write is observed before the load of status in | |
9892 | * btrfs_encoded_read_regular_fill_pages(). | |
9893 | */ | |
7609afac | 9894 | WRITE_ONCE(priv->status, bbio->bio.bi_status); |
1881fba8 OS |
9895 | } |
9896 | if (!atomic_dec_return(&priv->pending)) | |
9897 | wake_up(&priv->wait); | |
917f32a2 | 9898 | bio_put(&bbio->bio); |
1881fba8 OS |
9899 | } |
9900 | ||
3ea4dc5b OS |
9901 | int btrfs_encoded_read_regular_fill_pages(struct btrfs_inode *inode, |
9902 | u64 file_offset, u64 disk_bytenr, | |
9903 | u64 disk_io_size, struct page **pages) | |
1881fba8 | 9904 | { |
4317ff00 | 9905 | struct btrfs_fs_info *fs_info = inode->root->fs_info; |
1881fba8 | 9906 | struct btrfs_encoded_read_private priv = { |
1881fba8 | 9907 | .pending = ATOMIC_INIT(1), |
1881fba8 OS |
9908 | }; |
9909 | unsigned long i = 0; | |
b41bbd29 | 9910 | struct btrfs_bio *bbio; |
1881fba8 OS |
9911 | |
9912 | init_waitqueue_head(&priv.wait); | |
1881fba8 | 9913 | |
4317ff00 QW |
9914 | bbio = btrfs_bio_alloc(BIO_MAX_VECS, REQ_OP_READ, fs_info, |
9915 | btrfs_encoded_read_endio, &priv); | |
b41bbd29 | 9916 | bbio->bio.bi_iter.bi_sector = disk_bytenr >> SECTOR_SHIFT; |
4317ff00 | 9917 | bbio->inode = inode; |
1881fba8 | 9918 | |
34f888ce CH |
9919 | do { |
9920 | size_t bytes = min_t(u64, disk_io_size, PAGE_SIZE); | |
9921 | ||
b41bbd29 | 9922 | if (bio_add_page(&bbio->bio, pages[i], bytes, 0) < bytes) { |
34f888ce | 9923 | atomic_inc(&priv.pending); |
b41bbd29 | 9924 | btrfs_submit_bio(bbio, 0); |
34f888ce | 9925 | |
4317ff00 | 9926 | bbio = btrfs_bio_alloc(BIO_MAX_VECS, REQ_OP_READ, fs_info, |
b41bbd29 CH |
9927 | btrfs_encoded_read_endio, &priv); |
9928 | bbio->bio.bi_iter.bi_sector = disk_bytenr >> SECTOR_SHIFT; | |
4317ff00 | 9929 | bbio->inode = inode; |
34f888ce | 9930 | continue; |
1881fba8 | 9931 | } |
34f888ce CH |
9932 | |
9933 | i++; | |
9934 | disk_bytenr += bytes; | |
9935 | disk_io_size -= bytes; | |
9936 | } while (disk_io_size); | |
9937 | ||
9938 | atomic_inc(&priv.pending); | |
b41bbd29 | 9939 | btrfs_submit_bio(bbio, 0); |
1881fba8 | 9940 | |
1881fba8 OS |
9941 | if (atomic_dec_return(&priv.pending)) |
9942 | io_wait_event(priv.wait, !atomic_read(&priv.pending)); | |
9943 | /* See btrfs_encoded_read_endio() for ordering. */ | |
9944 | return blk_status_to_errno(READ_ONCE(priv.status)); | |
9945 | } | |
9946 | ||
9947 | static ssize_t btrfs_encoded_read_regular(struct kiocb *iocb, | |
9948 | struct iov_iter *iter, | |
9949 | u64 start, u64 lockend, | |
9950 | struct extent_state **cached_state, | |
9951 | u64 disk_bytenr, u64 disk_io_size, | |
9952 | size_t count, bool compressed, | |
9953 | bool *unlocked) | |
9954 | { | |
9955 | struct btrfs_inode *inode = BTRFS_I(file_inode(iocb->ki_filp)); | |
9956 | struct extent_io_tree *io_tree = &inode->io_tree; | |
9957 | struct page **pages; | |
9958 | unsigned long nr_pages, i; | |
9959 | u64 cur; | |
9960 | size_t page_offset; | |
9961 | ssize_t ret; | |
9962 | ||
9963 | nr_pages = DIV_ROUND_UP(disk_io_size, PAGE_SIZE); | |
9964 | pages = kcalloc(nr_pages, sizeof(struct page *), GFP_NOFS); | |
9965 | if (!pages) | |
9966 | return -ENOMEM; | |
dd137dd1 STD |
9967 | ret = btrfs_alloc_page_array(nr_pages, pages); |
9968 | if (ret) { | |
9969 | ret = -ENOMEM; | |
9970 | goto out; | |
1881fba8 | 9971 | } |
1881fba8 OS |
9972 | |
9973 | ret = btrfs_encoded_read_regular_fill_pages(inode, start, disk_bytenr, | |
9974 | disk_io_size, pages); | |
9975 | if (ret) | |
9976 | goto out; | |
9977 | ||
570eb97b | 9978 | unlock_extent(io_tree, start, lockend, cached_state); |
e5d4d75b | 9979 | btrfs_inode_unlock(inode, BTRFS_ILOCK_SHARED); |
1881fba8 OS |
9980 | *unlocked = true; |
9981 | ||
9982 | if (compressed) { | |
9983 | i = 0; | |
9984 | page_offset = 0; | |
9985 | } else { | |
9986 | i = (iocb->ki_pos - start) >> PAGE_SHIFT; | |
9987 | page_offset = (iocb->ki_pos - start) & (PAGE_SIZE - 1); | |
9988 | } | |
9989 | cur = 0; | |
9990 | while (cur < count) { | |
9991 | size_t bytes = min_t(size_t, count - cur, | |
9992 | PAGE_SIZE - page_offset); | |
9993 | ||
9994 | if (copy_page_to_iter(pages[i], page_offset, bytes, | |
9995 | iter) != bytes) { | |
9996 | ret = -EFAULT; | |
9997 | goto out; | |
9998 | } | |
9999 | i++; | |
10000 | cur += bytes; | |
10001 | page_offset = 0; | |
10002 | } | |
10003 | ret = count; | |
10004 | out: | |
10005 | for (i = 0; i < nr_pages; i++) { | |
10006 | if (pages[i]) | |
10007 | __free_page(pages[i]); | |
10008 | } | |
10009 | kfree(pages); | |
10010 | return ret; | |
10011 | } | |
10012 | ||
10013 | ssize_t btrfs_encoded_read(struct kiocb *iocb, struct iov_iter *iter, | |
10014 | struct btrfs_ioctl_encoded_io_args *encoded) | |
10015 | { | |
10016 | struct btrfs_inode *inode = BTRFS_I(file_inode(iocb->ki_filp)); | |
10017 | struct btrfs_fs_info *fs_info = inode->root->fs_info; | |
10018 | struct extent_io_tree *io_tree = &inode->io_tree; | |
10019 | ssize_t ret; | |
10020 | size_t count = iov_iter_count(iter); | |
10021 | u64 start, lockend, disk_bytenr, disk_io_size; | |
10022 | struct extent_state *cached_state = NULL; | |
10023 | struct extent_map *em; | |
10024 | bool unlocked = false; | |
10025 | ||
10026 | file_accessed(iocb->ki_filp); | |
10027 | ||
29b6352b | 10028 | btrfs_inode_lock(inode, BTRFS_ILOCK_SHARED); |
1881fba8 OS |
10029 | |
10030 | if (iocb->ki_pos >= inode->vfs_inode.i_size) { | |
e5d4d75b | 10031 | btrfs_inode_unlock(inode, BTRFS_ILOCK_SHARED); |
1881fba8 OS |
10032 | return 0; |
10033 | } | |
10034 | start = ALIGN_DOWN(iocb->ki_pos, fs_info->sectorsize); | |
10035 | /* | |
10036 | * We don't know how long the extent containing iocb->ki_pos is, but if | |
10037 | * it's compressed we know that it won't be longer than this. | |
10038 | */ | |
10039 | lockend = start + BTRFS_MAX_UNCOMPRESSED - 1; | |
10040 | ||
10041 | for (;;) { | |
10042 | struct btrfs_ordered_extent *ordered; | |
10043 | ||
10044 | ret = btrfs_wait_ordered_range(&inode->vfs_inode, start, | |
10045 | lockend - start + 1); | |
10046 | if (ret) | |
10047 | goto out_unlock_inode; | |
570eb97b | 10048 | lock_extent(io_tree, start, lockend, &cached_state); |
1881fba8 OS |
10049 | ordered = btrfs_lookup_ordered_range(inode, start, |
10050 | lockend - start + 1); | |
10051 | if (!ordered) | |
10052 | break; | |
10053 | btrfs_put_ordered_extent(ordered); | |
570eb97b | 10054 | unlock_extent(io_tree, start, lockend, &cached_state); |
1881fba8 OS |
10055 | cond_resched(); |
10056 | } | |
10057 | ||
10058 | em = btrfs_get_extent(inode, NULL, 0, start, lockend - start + 1); | |
10059 | if (IS_ERR(em)) { | |
10060 | ret = PTR_ERR(em); | |
10061 | goto out_unlock_extent; | |
10062 | } | |
10063 | ||
10064 | if (em->block_start == EXTENT_MAP_INLINE) { | |
10065 | u64 extent_start = em->start; | |
10066 | ||
10067 | /* | |
10068 | * For inline extents we get everything we need out of the | |
10069 | * extent item. | |
10070 | */ | |
10071 | free_extent_map(em); | |
10072 | em = NULL; | |
10073 | ret = btrfs_encoded_read_inline(iocb, iter, start, lockend, | |
10074 | &cached_state, extent_start, | |
10075 | count, encoded, &unlocked); | |
10076 | goto out; | |
10077 | } | |
10078 | ||
10079 | /* | |
10080 | * We only want to return up to EOF even if the extent extends beyond | |
10081 | * that. | |
10082 | */ | |
10083 | encoded->len = min_t(u64, extent_map_end(em), | |
10084 | inode->vfs_inode.i_size) - iocb->ki_pos; | |
10085 | if (em->block_start == EXTENT_MAP_HOLE || | |
10086 | test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) { | |
10087 | disk_bytenr = EXTENT_MAP_HOLE; | |
10088 | count = min_t(u64, count, encoded->len); | |
10089 | encoded->len = count; | |
10090 | encoded->unencoded_len = count; | |
10091 | } else if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) { | |
10092 | disk_bytenr = em->block_start; | |
10093 | /* | |
10094 | * Bail if the buffer isn't large enough to return the whole | |
10095 | * compressed extent. | |
10096 | */ | |
10097 | if (em->block_len > count) { | |
10098 | ret = -ENOBUFS; | |
10099 | goto out_em; | |
10100 | } | |
c1867eb3 DS |
10101 | disk_io_size = em->block_len; |
10102 | count = em->block_len; | |
1881fba8 OS |
10103 | encoded->unencoded_len = em->ram_bytes; |
10104 | encoded->unencoded_offset = iocb->ki_pos - em->orig_start; | |
10105 | ret = btrfs_encoded_io_compression_from_extent(fs_info, | |
10106 | em->compress_type); | |
10107 | if (ret < 0) | |
10108 | goto out_em; | |
10109 | encoded->compression = ret; | |
10110 | } else { | |
10111 | disk_bytenr = em->block_start + (start - em->start); | |
10112 | if (encoded->len > count) | |
10113 | encoded->len = count; | |
10114 | /* | |
10115 | * Don't read beyond what we locked. This also limits the page | |
10116 | * allocations that we'll do. | |
10117 | */ | |
10118 | disk_io_size = min(lockend + 1, iocb->ki_pos + encoded->len) - start; | |
10119 | count = start + disk_io_size - iocb->ki_pos; | |
10120 | encoded->len = count; | |
10121 | encoded->unencoded_len = count; | |
10122 | disk_io_size = ALIGN(disk_io_size, fs_info->sectorsize); | |
10123 | } | |
10124 | free_extent_map(em); | |
10125 | em = NULL; | |
10126 | ||
10127 | if (disk_bytenr == EXTENT_MAP_HOLE) { | |
570eb97b | 10128 | unlock_extent(io_tree, start, lockend, &cached_state); |
e5d4d75b | 10129 | btrfs_inode_unlock(inode, BTRFS_ILOCK_SHARED); |
1881fba8 OS |
10130 | unlocked = true; |
10131 | ret = iov_iter_zero(count, iter); | |
10132 | if (ret != count) | |
10133 | ret = -EFAULT; | |
10134 | } else { | |
10135 | ret = btrfs_encoded_read_regular(iocb, iter, start, lockend, | |
10136 | &cached_state, disk_bytenr, | |
10137 | disk_io_size, count, | |
10138 | encoded->compression, | |
10139 | &unlocked); | |
10140 | } | |
10141 | ||
10142 | out: | |
10143 | if (ret >= 0) | |
10144 | iocb->ki_pos += encoded->len; | |
10145 | out_em: | |
10146 | free_extent_map(em); | |
10147 | out_unlock_extent: | |
10148 | if (!unlocked) | |
570eb97b | 10149 | unlock_extent(io_tree, start, lockend, &cached_state); |
1881fba8 OS |
10150 | out_unlock_inode: |
10151 | if (!unlocked) | |
e5d4d75b | 10152 | btrfs_inode_unlock(inode, BTRFS_ILOCK_SHARED); |
1881fba8 OS |
10153 | return ret; |
10154 | } | |
10155 | ||
7c0c7269 OS |
10156 | ssize_t btrfs_do_encoded_write(struct kiocb *iocb, struct iov_iter *from, |
10157 | const struct btrfs_ioctl_encoded_io_args *encoded) | |
10158 | { | |
10159 | struct btrfs_inode *inode = BTRFS_I(file_inode(iocb->ki_filp)); | |
10160 | struct btrfs_root *root = inode->root; | |
10161 | struct btrfs_fs_info *fs_info = root->fs_info; | |
10162 | struct extent_io_tree *io_tree = &inode->io_tree; | |
10163 | struct extent_changeset *data_reserved = NULL; | |
10164 | struct extent_state *cached_state = NULL; | |
d611935b | 10165 | struct btrfs_ordered_extent *ordered; |
7c0c7269 OS |
10166 | int compression; |
10167 | size_t orig_count; | |
10168 | u64 start, end; | |
10169 | u64 num_bytes, ram_bytes, disk_num_bytes; | |
10170 | unsigned long nr_pages, i; | |
10171 | struct page **pages; | |
10172 | struct btrfs_key ins; | |
10173 | bool extent_reserved = false; | |
10174 | struct extent_map *em; | |
10175 | ssize_t ret; | |
10176 | ||
10177 | switch (encoded->compression) { | |
10178 | case BTRFS_ENCODED_IO_COMPRESSION_ZLIB: | |
10179 | compression = BTRFS_COMPRESS_ZLIB; | |
10180 | break; | |
10181 | case BTRFS_ENCODED_IO_COMPRESSION_ZSTD: | |
10182 | compression = BTRFS_COMPRESS_ZSTD; | |
10183 | break; | |
10184 | case BTRFS_ENCODED_IO_COMPRESSION_LZO_4K: | |
10185 | case BTRFS_ENCODED_IO_COMPRESSION_LZO_8K: | |
10186 | case BTRFS_ENCODED_IO_COMPRESSION_LZO_16K: | |
10187 | case BTRFS_ENCODED_IO_COMPRESSION_LZO_32K: | |
10188 | case BTRFS_ENCODED_IO_COMPRESSION_LZO_64K: | |
10189 | /* The sector size must match for LZO. */ | |
10190 | if (encoded->compression - | |
10191 | BTRFS_ENCODED_IO_COMPRESSION_LZO_4K + 12 != | |
10192 | fs_info->sectorsize_bits) | |
10193 | return -EINVAL; | |
10194 | compression = BTRFS_COMPRESS_LZO; | |
10195 | break; | |
10196 | default: | |
10197 | return -EINVAL; | |
10198 | } | |
10199 | if (encoded->encryption != BTRFS_ENCODED_IO_ENCRYPTION_NONE) | |
10200 | return -EINVAL; | |
10201 | ||
10202 | orig_count = iov_iter_count(from); | |
10203 | ||
10204 | /* The extent size must be sane. */ | |
10205 | if (encoded->unencoded_len > BTRFS_MAX_UNCOMPRESSED || | |
10206 | orig_count > BTRFS_MAX_COMPRESSED || orig_count == 0) | |
10207 | return -EINVAL; | |
10208 | ||
10209 | /* | |
10210 | * The compressed data must be smaller than the decompressed data. | |
10211 | * | |
10212 | * It's of course possible for data to compress to larger or the same | |
10213 | * size, but the buffered I/O path falls back to no compression for such | |
10214 | * data, and we don't want to break any assumptions by creating these | |
10215 | * extents. | |
10216 | * | |
10217 | * Note that this is less strict than the current check we have that the | |
10218 | * compressed data must be at least one sector smaller than the | |
10219 | * decompressed data. We only want to enforce the weaker requirement | |
10220 | * from old kernels that it is at least one byte smaller. | |
10221 | */ | |
10222 | if (orig_count >= encoded->unencoded_len) | |
10223 | return -EINVAL; | |
10224 | ||
10225 | /* The extent must start on a sector boundary. */ | |
10226 | start = iocb->ki_pos; | |
10227 | if (!IS_ALIGNED(start, fs_info->sectorsize)) | |
10228 | return -EINVAL; | |
10229 | ||
10230 | /* | |
10231 | * The extent must end on a sector boundary. However, we allow a write | |
10232 | * which ends at or extends i_size to have an unaligned length; we round | |
10233 | * up the extent size and set i_size to the unaligned end. | |
10234 | */ | |
10235 | if (start + encoded->len < inode->vfs_inode.i_size && | |
10236 | !IS_ALIGNED(start + encoded->len, fs_info->sectorsize)) | |
10237 | return -EINVAL; | |
10238 | ||
10239 | /* Finally, the offset in the unencoded data must be sector-aligned. */ | |
10240 | if (!IS_ALIGNED(encoded->unencoded_offset, fs_info->sectorsize)) | |
10241 | return -EINVAL; | |
10242 | ||
10243 | num_bytes = ALIGN(encoded->len, fs_info->sectorsize); | |
10244 | ram_bytes = ALIGN(encoded->unencoded_len, fs_info->sectorsize); | |
10245 | end = start + num_bytes - 1; | |
10246 | ||
10247 | /* | |
10248 | * If the extent cannot be inline, the compressed data on disk must be | |
10249 | * sector-aligned. For convenience, we extend it with zeroes if it | |
10250 | * isn't. | |
10251 | */ | |
10252 | disk_num_bytes = ALIGN(orig_count, fs_info->sectorsize); | |
10253 | nr_pages = DIV_ROUND_UP(disk_num_bytes, PAGE_SIZE); | |
10254 | pages = kvcalloc(nr_pages, sizeof(struct page *), GFP_KERNEL_ACCOUNT); | |
10255 | if (!pages) | |
10256 | return -ENOMEM; | |
10257 | for (i = 0; i < nr_pages; i++) { | |
10258 | size_t bytes = min_t(size_t, PAGE_SIZE, iov_iter_count(from)); | |
10259 | char *kaddr; | |
10260 | ||
10261 | pages[i] = alloc_page(GFP_KERNEL_ACCOUNT); | |
10262 | if (!pages[i]) { | |
10263 | ret = -ENOMEM; | |
10264 | goto out_pages; | |
10265 | } | |
70826b6b | 10266 | kaddr = kmap_local_page(pages[i]); |
7c0c7269 | 10267 | if (copy_from_iter(kaddr, bytes, from) != bytes) { |
70826b6b | 10268 | kunmap_local(kaddr); |
7c0c7269 OS |
10269 | ret = -EFAULT; |
10270 | goto out_pages; | |
10271 | } | |
10272 | if (bytes < PAGE_SIZE) | |
10273 | memset(kaddr + bytes, 0, PAGE_SIZE - bytes); | |
70826b6b | 10274 | kunmap_local(kaddr); |
7c0c7269 OS |
10275 | } |
10276 | ||
10277 | for (;;) { | |
10278 | struct btrfs_ordered_extent *ordered; | |
10279 | ||
10280 | ret = btrfs_wait_ordered_range(&inode->vfs_inode, start, num_bytes); | |
10281 | if (ret) | |
10282 | goto out_pages; | |
10283 | ret = invalidate_inode_pages2_range(inode->vfs_inode.i_mapping, | |
10284 | start >> PAGE_SHIFT, | |
10285 | end >> PAGE_SHIFT); | |
10286 | if (ret) | |
10287 | goto out_pages; | |
570eb97b | 10288 | lock_extent(io_tree, start, end, &cached_state); |
7c0c7269 OS |
10289 | ordered = btrfs_lookup_ordered_range(inode, start, num_bytes); |
10290 | if (!ordered && | |
10291 | !filemap_range_has_page(inode->vfs_inode.i_mapping, start, end)) | |
10292 | break; | |
10293 | if (ordered) | |
10294 | btrfs_put_ordered_extent(ordered); | |
570eb97b | 10295 | unlock_extent(io_tree, start, end, &cached_state); |
7c0c7269 OS |
10296 | cond_resched(); |
10297 | } | |
10298 | ||
10299 | /* | |
10300 | * We don't use the higher-level delalloc space functions because our | |
10301 | * num_bytes and disk_num_bytes are different. | |
10302 | */ | |
10303 | ret = btrfs_alloc_data_chunk_ondemand(inode, disk_num_bytes); | |
10304 | if (ret) | |
10305 | goto out_unlock; | |
10306 | ret = btrfs_qgroup_reserve_data(inode, &data_reserved, start, num_bytes); | |
10307 | if (ret) | |
10308 | goto out_free_data_space; | |
d4135134 FM |
10309 | ret = btrfs_delalloc_reserve_metadata(inode, num_bytes, disk_num_bytes, |
10310 | false); | |
7c0c7269 OS |
10311 | if (ret) |
10312 | goto out_qgroup_free_data; | |
10313 | ||
10314 | /* Try an inline extent first. */ | |
10315 | if (start == 0 && encoded->unencoded_len == encoded->len && | |
10316 | encoded->unencoded_offset == 0) { | |
10317 | ret = cow_file_range_inline(inode, encoded->len, orig_count, | |
10318 | compression, pages, true); | |
10319 | if (ret <= 0) { | |
10320 | if (ret == 0) | |
10321 | ret = orig_count; | |
10322 | goto out_delalloc_release; | |
10323 | } | |
10324 | } | |
10325 | ||
10326 | ret = btrfs_reserve_extent(root, disk_num_bytes, disk_num_bytes, | |
10327 | disk_num_bytes, 0, 0, &ins, 1, 1); | |
10328 | if (ret) | |
10329 | goto out_delalloc_release; | |
10330 | extent_reserved = true; | |
10331 | ||
10332 | em = create_io_em(inode, start, num_bytes, | |
10333 | start - encoded->unencoded_offset, ins.objectid, | |
10334 | ins.offset, ins.offset, ram_bytes, compression, | |
10335 | BTRFS_ORDERED_COMPRESSED); | |
10336 | if (IS_ERR(em)) { | |
10337 | ret = PTR_ERR(em); | |
10338 | goto out_free_reserved; | |
10339 | } | |
10340 | free_extent_map(em); | |
10341 | ||
d611935b | 10342 | ordered = btrfs_alloc_ordered_extent(inode, start, num_bytes, ram_bytes, |
7c0c7269 OS |
10343 | ins.objectid, ins.offset, |
10344 | encoded->unencoded_offset, | |
10345 | (1 << BTRFS_ORDERED_ENCODED) | | |
10346 | (1 << BTRFS_ORDERED_COMPRESSED), | |
10347 | compression); | |
d611935b | 10348 | if (IS_ERR(ordered)) { |
4c0c8cfc | 10349 | btrfs_drop_extent_map_range(inode, start, end, false); |
d611935b | 10350 | ret = PTR_ERR(ordered); |
7c0c7269 OS |
10351 | goto out_free_reserved; |
10352 | } | |
10353 | btrfs_dec_block_group_reservations(fs_info, ins.objectid); | |
10354 | ||
10355 | if (start + encoded->len > inode->vfs_inode.i_size) | |
10356 | i_size_write(&inode->vfs_inode, start + encoded->len); | |
10357 | ||
570eb97b | 10358 | unlock_extent(io_tree, start, end, &cached_state); |
7c0c7269 OS |
10359 | |
10360 | btrfs_delalloc_release_extents(inode, num_bytes); | |
10361 | ||
d611935b | 10362 | btrfs_submit_compressed_write(ordered, pages, nr_pages, 0, false); |
7c0c7269 OS |
10363 | ret = orig_count; |
10364 | goto out; | |
10365 | ||
10366 | out_free_reserved: | |
10367 | btrfs_dec_block_group_reservations(fs_info, ins.objectid); | |
10368 | btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1); | |
10369 | out_delalloc_release: | |
10370 | btrfs_delalloc_release_extents(inode, num_bytes); | |
10371 | btrfs_delalloc_release_metadata(inode, disk_num_bytes, ret < 0); | |
10372 | out_qgroup_free_data: | |
10373 | if (ret < 0) | |
10374 | btrfs_qgroup_free_data(inode, data_reserved, start, num_bytes); | |
10375 | out_free_data_space: | |
10376 | /* | |
10377 | * If btrfs_reserve_extent() succeeded, then we already decremented | |
10378 | * bytes_may_use. | |
10379 | */ | |
10380 | if (!extent_reserved) | |
10381 | btrfs_free_reserved_data_space_noquota(fs_info, disk_num_bytes); | |
10382 | out_unlock: | |
570eb97b | 10383 | unlock_extent(io_tree, start, end, &cached_state); |
7c0c7269 OS |
10384 | out_pages: |
10385 | for (i = 0; i < nr_pages; i++) { | |
10386 | if (pages[i]) | |
10387 | __free_page(pages[i]); | |
10388 | } | |
10389 | kvfree(pages); | |
10390 | out: | |
10391 | if (ret >= 0) | |
10392 | iocb->ki_pos += encoded->len; | |
10393 | return ret; | |
10394 | } | |
10395 | ||
ed46ff3d OS |
10396 | #ifdef CONFIG_SWAP |
10397 | /* | |
10398 | * Add an entry indicating a block group or device which is pinned by a | |
10399 | * swapfile. Returns 0 on success, 1 if there is already an entry for it, or a | |
10400 | * negative errno on failure. | |
10401 | */ | |
10402 | static int btrfs_add_swapfile_pin(struct inode *inode, void *ptr, | |
10403 | bool is_block_group) | |
10404 | { | |
10405 | struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info; | |
10406 | struct btrfs_swapfile_pin *sp, *entry; | |
10407 | struct rb_node **p; | |
10408 | struct rb_node *parent = NULL; | |
10409 | ||
10410 | sp = kmalloc(sizeof(*sp), GFP_NOFS); | |
10411 | if (!sp) | |
10412 | return -ENOMEM; | |
10413 | sp->ptr = ptr; | |
10414 | sp->inode = inode; | |
10415 | sp->is_block_group = is_block_group; | |
195a49ea | 10416 | sp->bg_extent_count = 1; |
ed46ff3d OS |
10417 | |
10418 | spin_lock(&fs_info->swapfile_pins_lock); | |
10419 | p = &fs_info->swapfile_pins.rb_node; | |
10420 | while (*p) { | |
10421 | parent = *p; | |
10422 | entry = rb_entry(parent, struct btrfs_swapfile_pin, node); | |
10423 | if (sp->ptr < entry->ptr || | |
10424 | (sp->ptr == entry->ptr && sp->inode < entry->inode)) { | |
10425 | p = &(*p)->rb_left; | |
10426 | } else if (sp->ptr > entry->ptr || | |
10427 | (sp->ptr == entry->ptr && sp->inode > entry->inode)) { | |
10428 | p = &(*p)->rb_right; | |
10429 | } else { | |
195a49ea FM |
10430 | if (is_block_group) |
10431 | entry->bg_extent_count++; | |
ed46ff3d OS |
10432 | spin_unlock(&fs_info->swapfile_pins_lock); |
10433 | kfree(sp); | |
10434 | return 1; | |
10435 | } | |
10436 | } | |
10437 | rb_link_node(&sp->node, parent, p); | |
10438 | rb_insert_color(&sp->node, &fs_info->swapfile_pins); | |
10439 | spin_unlock(&fs_info->swapfile_pins_lock); | |
10440 | return 0; | |
10441 | } | |
10442 | ||
10443 | /* Free all of the entries pinned by this swapfile. */ | |
10444 | static void btrfs_free_swapfile_pins(struct inode *inode) | |
10445 | { | |
10446 | struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info; | |
10447 | struct btrfs_swapfile_pin *sp; | |
10448 | struct rb_node *node, *next; | |
10449 | ||
10450 | spin_lock(&fs_info->swapfile_pins_lock); | |
10451 | node = rb_first(&fs_info->swapfile_pins); | |
10452 | while (node) { | |
10453 | next = rb_next(node); | |
10454 | sp = rb_entry(node, struct btrfs_swapfile_pin, node); | |
10455 | if (sp->inode == inode) { | |
10456 | rb_erase(&sp->node, &fs_info->swapfile_pins); | |
195a49ea FM |
10457 | if (sp->is_block_group) { |
10458 | btrfs_dec_block_group_swap_extents(sp->ptr, | |
10459 | sp->bg_extent_count); | |
ed46ff3d | 10460 | btrfs_put_block_group(sp->ptr); |
195a49ea | 10461 | } |
ed46ff3d OS |
10462 | kfree(sp); |
10463 | } | |
10464 | node = next; | |
10465 | } | |
10466 | spin_unlock(&fs_info->swapfile_pins_lock); | |
10467 | } | |
10468 | ||
10469 | struct btrfs_swap_info { | |
10470 | u64 start; | |
10471 | u64 block_start; | |
10472 | u64 block_len; | |
10473 | u64 lowest_ppage; | |
10474 | u64 highest_ppage; | |
10475 | unsigned long nr_pages; | |
10476 | int nr_extents; | |
10477 | }; | |
10478 | ||
10479 | static int btrfs_add_swap_extent(struct swap_info_struct *sis, | |
10480 | struct btrfs_swap_info *bsi) | |
10481 | { | |
10482 | unsigned long nr_pages; | |
c2f82263 | 10483 | unsigned long max_pages; |
ed46ff3d OS |
10484 | u64 first_ppage, first_ppage_reported, next_ppage; |
10485 | int ret; | |
10486 | ||
c2f82263 FM |
10487 | /* |
10488 | * Our swapfile may have had its size extended after the swap header was | |
10489 | * written. In that case activating the swapfile should not go beyond | |
10490 | * the max size set in the swap header. | |
10491 | */ | |
10492 | if (bsi->nr_pages >= sis->max) | |
10493 | return 0; | |
10494 | ||
10495 | max_pages = sis->max - bsi->nr_pages; | |
ce394a7f YZ |
10496 | first_ppage = PAGE_ALIGN(bsi->block_start) >> PAGE_SHIFT; |
10497 | next_ppage = PAGE_ALIGN_DOWN(bsi->block_start + bsi->block_len) >> PAGE_SHIFT; | |
ed46ff3d OS |
10498 | |
10499 | if (first_ppage >= next_ppage) | |
10500 | return 0; | |
10501 | nr_pages = next_ppage - first_ppage; | |
c2f82263 | 10502 | nr_pages = min(nr_pages, max_pages); |
ed46ff3d OS |
10503 | |
10504 | first_ppage_reported = first_ppage; | |
10505 | if (bsi->start == 0) | |
10506 | first_ppage_reported++; | |
10507 | if (bsi->lowest_ppage > first_ppage_reported) | |
10508 | bsi->lowest_ppage = first_ppage_reported; | |
10509 | if (bsi->highest_ppage < (next_ppage - 1)) | |
10510 | bsi->highest_ppage = next_ppage - 1; | |
10511 | ||
10512 | ret = add_swap_extent(sis, bsi->nr_pages, nr_pages, first_ppage); | |
10513 | if (ret < 0) | |
10514 | return ret; | |
10515 | bsi->nr_extents += ret; | |
10516 | bsi->nr_pages += nr_pages; | |
10517 | return 0; | |
10518 | } | |
10519 | ||
10520 | static void btrfs_swap_deactivate(struct file *file) | |
10521 | { | |
10522 | struct inode *inode = file_inode(file); | |
10523 | ||
10524 | btrfs_free_swapfile_pins(inode); | |
10525 | atomic_dec(&BTRFS_I(inode)->root->nr_swapfiles); | |
10526 | } | |
10527 | ||
10528 | static int btrfs_swap_activate(struct swap_info_struct *sis, struct file *file, | |
10529 | sector_t *span) | |
10530 | { | |
10531 | struct inode *inode = file_inode(file); | |
dd0734f2 FM |
10532 | struct btrfs_root *root = BTRFS_I(inode)->root; |
10533 | struct btrfs_fs_info *fs_info = root->fs_info; | |
ed46ff3d OS |
10534 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
10535 | struct extent_state *cached_state = NULL; | |
10536 | struct extent_map *em = NULL; | |
10537 | struct btrfs_device *device = NULL; | |
10538 | struct btrfs_swap_info bsi = { | |
10539 | .lowest_ppage = (sector_t)-1ULL, | |
10540 | }; | |
10541 | int ret = 0; | |
10542 | u64 isize; | |
10543 | u64 start; | |
10544 | ||
10545 | /* | |
10546 | * If the swap file was just created, make sure delalloc is done. If the | |
10547 | * file changes again after this, the user is doing something stupid and | |
10548 | * we don't really care. | |
10549 | */ | |
10550 | ret = btrfs_wait_ordered_range(inode, 0, (u64)-1); | |
10551 | if (ret) | |
10552 | return ret; | |
10553 | ||
10554 | /* | |
10555 | * The inode is locked, so these flags won't change after we check them. | |
10556 | */ | |
10557 | if (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS) { | |
10558 | btrfs_warn(fs_info, "swapfile must not be compressed"); | |
10559 | return -EINVAL; | |
10560 | } | |
10561 | if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW)) { | |
10562 | btrfs_warn(fs_info, "swapfile must not be copy-on-write"); | |
10563 | return -EINVAL; | |
10564 | } | |
10565 | if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) { | |
10566 | btrfs_warn(fs_info, "swapfile must not be checksummed"); | |
10567 | return -EINVAL; | |
10568 | } | |
10569 | ||
10570 | /* | |
10571 | * Balance or device remove/replace/resize can move stuff around from | |
c3e1f96c GR |
10572 | * under us. The exclop protection makes sure they aren't running/won't |
10573 | * run concurrently while we are mapping the swap extents, and | |
10574 | * fs_info->swapfile_pins prevents them from running while the swap | |
10575 | * file is active and moving the extents. Note that this also prevents | |
10576 | * a concurrent device add which isn't actually necessary, but it's not | |
ed46ff3d OS |
10577 | * really worth the trouble to allow it. |
10578 | */ | |
c3e1f96c | 10579 | if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_SWAP_ACTIVATE)) { |
ed46ff3d OS |
10580 | btrfs_warn(fs_info, |
10581 | "cannot activate swapfile while exclusive operation is running"); | |
10582 | return -EBUSY; | |
10583 | } | |
dd0734f2 FM |
10584 | |
10585 | /* | |
10586 | * Prevent snapshot creation while we are activating the swap file. | |
10587 | * We do not want to race with snapshot creation. If snapshot creation | |
10588 | * already started before we bumped nr_swapfiles from 0 to 1 and | |
10589 | * completes before the first write into the swap file after it is | |
10590 | * activated, than that write would fallback to COW. | |
10591 | */ | |
10592 | if (!btrfs_drew_try_write_lock(&root->snapshot_lock)) { | |
10593 | btrfs_exclop_finish(fs_info); | |
10594 | btrfs_warn(fs_info, | |
10595 | "cannot activate swapfile because snapshot creation is in progress"); | |
10596 | return -EINVAL; | |
10597 | } | |
ed46ff3d OS |
10598 | /* |
10599 | * Snapshots can create extents which require COW even if NODATACOW is | |
10600 | * set. We use this counter to prevent snapshots. We must increment it | |
10601 | * before walking the extents because we don't want a concurrent | |
10602 | * snapshot to run after we've already checked the extents. | |
60021bd7 KH |
10603 | * |
10604 | * It is possible that subvolume is marked for deletion but still not | |
10605 | * removed yet. To prevent this race, we check the root status before | |
10606 | * activating the swapfile. | |
ed46ff3d | 10607 | */ |
60021bd7 KH |
10608 | spin_lock(&root->root_item_lock); |
10609 | if (btrfs_root_dead(root)) { | |
10610 | spin_unlock(&root->root_item_lock); | |
10611 | ||
10612 | btrfs_exclop_finish(fs_info); | |
10613 | btrfs_warn(fs_info, | |
10614 | "cannot activate swapfile because subvolume %llu is being deleted", | |
10615 | root->root_key.objectid); | |
10616 | return -EPERM; | |
10617 | } | |
dd0734f2 | 10618 | atomic_inc(&root->nr_swapfiles); |
60021bd7 | 10619 | spin_unlock(&root->root_item_lock); |
ed46ff3d OS |
10620 | |
10621 | isize = ALIGN_DOWN(inode->i_size, fs_info->sectorsize); | |
10622 | ||
570eb97b | 10623 | lock_extent(io_tree, 0, isize - 1, &cached_state); |
ed46ff3d OS |
10624 | start = 0; |
10625 | while (start < isize) { | |
10626 | u64 logical_block_start, physical_block_start; | |
32da5386 | 10627 | struct btrfs_block_group *bg; |
ed46ff3d OS |
10628 | u64 len = isize - start; |
10629 | ||
39b07b5d | 10630 | em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len); |
ed46ff3d OS |
10631 | if (IS_ERR(em)) { |
10632 | ret = PTR_ERR(em); | |
10633 | goto out; | |
10634 | } | |
10635 | ||
10636 | if (em->block_start == EXTENT_MAP_HOLE) { | |
10637 | btrfs_warn(fs_info, "swapfile must not have holes"); | |
10638 | ret = -EINVAL; | |
10639 | goto out; | |
10640 | } | |
10641 | if (em->block_start == EXTENT_MAP_INLINE) { | |
10642 | /* | |
10643 | * It's unlikely we'll ever actually find ourselves | |
10644 | * here, as a file small enough to fit inline won't be | |
10645 | * big enough to store more than the swap header, but in | |
10646 | * case something changes in the future, let's catch it | |
10647 | * here rather than later. | |
10648 | */ | |
10649 | btrfs_warn(fs_info, "swapfile must not be inline"); | |
10650 | ret = -EINVAL; | |
10651 | goto out; | |
10652 | } | |
10653 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) { | |
10654 | btrfs_warn(fs_info, "swapfile must not be compressed"); | |
10655 | ret = -EINVAL; | |
10656 | goto out; | |
10657 | } | |
10658 | ||
10659 | logical_block_start = em->block_start + (start - em->start); | |
10660 | len = min(len, em->len - (start - em->start)); | |
10661 | free_extent_map(em); | |
10662 | em = NULL; | |
10663 | ||
26ce9114 | 10664 | ret = can_nocow_extent(inode, start, &len, NULL, NULL, NULL, false, true); |
ed46ff3d OS |
10665 | if (ret < 0) { |
10666 | goto out; | |
10667 | } else if (ret) { | |
10668 | ret = 0; | |
10669 | } else { | |
10670 | btrfs_warn(fs_info, | |
10671 | "swapfile must not be copy-on-write"); | |
10672 | ret = -EINVAL; | |
10673 | goto out; | |
10674 | } | |
10675 | ||
10676 | em = btrfs_get_chunk_map(fs_info, logical_block_start, len); | |
10677 | if (IS_ERR(em)) { | |
10678 | ret = PTR_ERR(em); | |
10679 | goto out; | |
10680 | } | |
10681 | ||
10682 | if (em->map_lookup->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) { | |
10683 | btrfs_warn(fs_info, | |
10684 | "swapfile must have single data profile"); | |
10685 | ret = -EINVAL; | |
10686 | goto out; | |
10687 | } | |
10688 | ||
10689 | if (device == NULL) { | |
10690 | device = em->map_lookup->stripes[0].dev; | |
10691 | ret = btrfs_add_swapfile_pin(inode, device, false); | |
10692 | if (ret == 1) | |
10693 | ret = 0; | |
10694 | else if (ret) | |
10695 | goto out; | |
10696 | } else if (device != em->map_lookup->stripes[0].dev) { | |
10697 | btrfs_warn(fs_info, "swapfile must be on one device"); | |
10698 | ret = -EINVAL; | |
10699 | goto out; | |
10700 | } | |
10701 | ||
10702 | physical_block_start = (em->map_lookup->stripes[0].physical + | |
10703 | (logical_block_start - em->start)); | |
10704 | len = min(len, em->len - (logical_block_start - em->start)); | |
10705 | free_extent_map(em); | |
10706 | em = NULL; | |
10707 | ||
10708 | bg = btrfs_lookup_block_group(fs_info, logical_block_start); | |
10709 | if (!bg) { | |
10710 | btrfs_warn(fs_info, | |
10711 | "could not find block group containing swapfile"); | |
10712 | ret = -EINVAL; | |
10713 | goto out; | |
10714 | } | |
10715 | ||
195a49ea FM |
10716 | if (!btrfs_inc_block_group_swap_extents(bg)) { |
10717 | btrfs_warn(fs_info, | |
10718 | "block group for swapfile at %llu is read-only%s", | |
10719 | bg->start, | |
10720 | atomic_read(&fs_info->scrubs_running) ? | |
10721 | " (scrub running)" : ""); | |
10722 | btrfs_put_block_group(bg); | |
10723 | ret = -EINVAL; | |
10724 | goto out; | |
10725 | } | |
10726 | ||
ed46ff3d OS |
10727 | ret = btrfs_add_swapfile_pin(inode, bg, true); |
10728 | if (ret) { | |
10729 | btrfs_put_block_group(bg); | |
10730 | if (ret == 1) | |
10731 | ret = 0; | |
10732 | else | |
10733 | goto out; | |
10734 | } | |
10735 | ||
10736 | if (bsi.block_len && | |
10737 | bsi.block_start + bsi.block_len == physical_block_start) { | |
10738 | bsi.block_len += len; | |
10739 | } else { | |
10740 | if (bsi.block_len) { | |
10741 | ret = btrfs_add_swap_extent(sis, &bsi); | |
10742 | if (ret) | |
10743 | goto out; | |
10744 | } | |
10745 | bsi.start = start; | |
10746 | bsi.block_start = physical_block_start; | |
10747 | bsi.block_len = len; | |
10748 | } | |
10749 | ||
10750 | start += len; | |
10751 | } | |
10752 | ||
10753 | if (bsi.block_len) | |
10754 | ret = btrfs_add_swap_extent(sis, &bsi); | |
10755 | ||
10756 | out: | |
10757 | if (!IS_ERR_OR_NULL(em)) | |
10758 | free_extent_map(em); | |
10759 | ||
570eb97b | 10760 | unlock_extent(io_tree, 0, isize - 1, &cached_state); |
ed46ff3d OS |
10761 | |
10762 | if (ret) | |
10763 | btrfs_swap_deactivate(file); | |
10764 | ||
dd0734f2 FM |
10765 | btrfs_drew_write_unlock(&root->snapshot_lock); |
10766 | ||
c3e1f96c | 10767 | btrfs_exclop_finish(fs_info); |
ed46ff3d OS |
10768 | |
10769 | if (ret) | |
10770 | return ret; | |
10771 | ||
10772 | if (device) | |
10773 | sis->bdev = device->bdev; | |
10774 | *span = bsi.highest_ppage - bsi.lowest_ppage + 1; | |
10775 | sis->max = bsi.nr_pages; | |
10776 | sis->pages = bsi.nr_pages - 1; | |
10777 | sis->highest_bit = bsi.nr_pages - 1; | |
10778 | return bsi.nr_extents; | |
10779 | } | |
10780 | #else | |
10781 | static void btrfs_swap_deactivate(struct file *file) | |
10782 | { | |
10783 | } | |
10784 | ||
10785 | static int btrfs_swap_activate(struct swap_info_struct *sis, struct file *file, | |
10786 | sector_t *span) | |
10787 | { | |
10788 | return -EOPNOTSUPP; | |
10789 | } | |
10790 | #endif | |
10791 | ||
2766ff61 FM |
10792 | /* |
10793 | * Update the number of bytes used in the VFS' inode. When we replace extents in | |
10794 | * a range (clone, dedupe, fallocate's zero range), we must update the number of | |
10795 | * bytes used by the inode in an atomic manner, so that concurrent stat(2) calls | |
10796 | * always get a correct value. | |
10797 | */ | |
10798 | void btrfs_update_inode_bytes(struct btrfs_inode *inode, | |
10799 | const u64 add_bytes, | |
10800 | const u64 del_bytes) | |
10801 | { | |
10802 | if (add_bytes == del_bytes) | |
10803 | return; | |
10804 | ||
10805 | spin_lock(&inode->lock); | |
10806 | if (del_bytes > 0) | |
10807 | inode_sub_bytes(&inode->vfs_inode, del_bytes); | |
10808 | if (add_bytes > 0) | |
10809 | inode_add_bytes(&inode->vfs_inode, add_bytes); | |
10810 | spin_unlock(&inode->lock); | |
10811 | } | |
10812 | ||
43dd529a | 10813 | /* |
63c34cb4 FM |
10814 | * Verify that there are no ordered extents for a given file range. |
10815 | * | |
10816 | * @inode: The target inode. | |
10817 | * @start: Start offset of the file range, should be sector size aligned. | |
10818 | * @end: End offset (inclusive) of the file range, its value +1 should be | |
10819 | * sector size aligned. | |
10820 | * | |
10821 | * This should typically be used for cases where we locked an inode's VFS lock in | |
10822 | * exclusive mode, we have also locked the inode's i_mmap_lock in exclusive mode, | |
10823 | * we have flushed all delalloc in the range, we have waited for all ordered | |
10824 | * extents in the range to complete and finally we have locked the file range in | |
10825 | * the inode's io_tree. | |
10826 | */ | |
10827 | void btrfs_assert_inode_range_clean(struct btrfs_inode *inode, u64 start, u64 end) | |
10828 | { | |
10829 | struct btrfs_root *root = inode->root; | |
10830 | struct btrfs_ordered_extent *ordered; | |
10831 | ||
10832 | if (!IS_ENABLED(CONFIG_BTRFS_ASSERT)) | |
10833 | return; | |
10834 | ||
10835 | ordered = btrfs_lookup_first_ordered_range(inode, start, end + 1 - start); | |
10836 | if (ordered) { | |
10837 | btrfs_err(root->fs_info, | |
10838 | "found unexpected ordered extent in file range [%llu, %llu] for inode %llu root %llu (ordered range [%llu, %llu])", | |
10839 | start, end, btrfs_ino(inode), root->root_key.objectid, | |
10840 | ordered->file_offset, | |
10841 | ordered->file_offset + ordered->num_bytes - 1); | |
10842 | btrfs_put_ordered_extent(ordered); | |
10843 | } | |
10844 | ||
10845 | ASSERT(ordered == NULL); | |
10846 | } | |
10847 | ||
6e1d5dcc | 10848 | static const struct inode_operations btrfs_dir_inode_operations = { |
3394e160 | 10849 | .getattr = btrfs_getattr, |
39279cc3 CM |
10850 | .lookup = btrfs_lookup, |
10851 | .create = btrfs_create, | |
10852 | .unlink = btrfs_unlink, | |
10853 | .link = btrfs_link, | |
10854 | .mkdir = btrfs_mkdir, | |
10855 | .rmdir = btrfs_rmdir, | |
2773bf00 | 10856 | .rename = btrfs_rename2, |
39279cc3 CM |
10857 | .symlink = btrfs_symlink, |
10858 | .setattr = btrfs_setattr, | |
618e21d5 | 10859 | .mknod = btrfs_mknod, |
5103e947 | 10860 | .listxattr = btrfs_listxattr, |
fdebe2bd | 10861 | .permission = btrfs_permission, |
cac2f8b8 | 10862 | .get_inode_acl = btrfs_get_acl, |
996a710d | 10863 | .set_acl = btrfs_set_acl, |
93fd63c2 | 10864 | .update_time = btrfs_update_time, |
ef3b9af5 | 10865 | .tmpfile = btrfs_tmpfile, |
97fc2977 MS |
10866 | .fileattr_get = btrfs_fileattr_get, |
10867 | .fileattr_set = btrfs_fileattr_set, | |
39279cc3 | 10868 | }; |
76dda93c | 10869 | |
828c0950 | 10870 | static const struct file_operations btrfs_dir_file_operations = { |
39279cc3 CM |
10871 | .llseek = generic_file_llseek, |
10872 | .read = generic_read_dir, | |
02dbfc99 | 10873 | .iterate_shared = btrfs_real_readdir, |
23b5ec74 | 10874 | .open = btrfs_opendir, |
34287aa3 | 10875 | .unlocked_ioctl = btrfs_ioctl, |
39279cc3 | 10876 | #ifdef CONFIG_COMPAT |
4c63c245 | 10877 | .compat_ioctl = btrfs_compat_ioctl, |
39279cc3 | 10878 | #endif |
6bf13c0c | 10879 | .release = btrfs_release_file, |
e02119d5 | 10880 | .fsync = btrfs_sync_file, |
39279cc3 CM |
10881 | }; |
10882 | ||
35054394 CM |
10883 | /* |
10884 | * btrfs doesn't support the bmap operation because swapfiles | |
10885 | * use bmap to make a mapping of extents in the file. They assume | |
10886 | * these extents won't change over the life of the file and they | |
10887 | * use the bmap result to do IO directly to the drive. | |
10888 | * | |
10889 | * the btrfs bmap call would return logical addresses that aren't | |
10890 | * suitable for IO and they also will change frequently as COW | |
10891 | * operations happen. So, swapfile + btrfs == corruption. | |
10892 | * | |
10893 | * For now we're avoiding this by dropping bmap. | |
10894 | */ | |
7f09410b | 10895 | static const struct address_space_operations btrfs_aops = { |
fb12489b | 10896 | .read_folio = btrfs_read_folio, |
b293f02e | 10897 | .writepages = btrfs_writepages, |
ba206a02 | 10898 | .readahead = btrfs_readahead, |
895586eb | 10899 | .invalidate_folio = btrfs_invalidate_folio, |
f913cff3 | 10900 | .release_folio = btrfs_release_folio, |
e7a60a17 | 10901 | .migrate_folio = btrfs_migrate_folio, |
187c82cb | 10902 | .dirty_folio = filemap_dirty_folio, |
465fdd97 | 10903 | .error_remove_page = generic_error_remove_page, |
ed46ff3d OS |
10904 | .swap_activate = btrfs_swap_activate, |
10905 | .swap_deactivate = btrfs_swap_deactivate, | |
39279cc3 CM |
10906 | }; |
10907 | ||
6e1d5dcc | 10908 | static const struct inode_operations btrfs_file_inode_operations = { |
39279cc3 CM |
10909 | .getattr = btrfs_getattr, |
10910 | .setattr = btrfs_setattr, | |
5103e947 | 10911 | .listxattr = btrfs_listxattr, |
fdebe2bd | 10912 | .permission = btrfs_permission, |
1506fcc8 | 10913 | .fiemap = btrfs_fiemap, |
cac2f8b8 | 10914 | .get_inode_acl = btrfs_get_acl, |
996a710d | 10915 | .set_acl = btrfs_set_acl, |
e41f941a | 10916 | .update_time = btrfs_update_time, |
97fc2977 MS |
10917 | .fileattr_get = btrfs_fileattr_get, |
10918 | .fileattr_set = btrfs_fileattr_set, | |
39279cc3 | 10919 | }; |
6e1d5dcc | 10920 | static const struct inode_operations btrfs_special_inode_operations = { |
618e21d5 JB |
10921 | .getattr = btrfs_getattr, |
10922 | .setattr = btrfs_setattr, | |
fdebe2bd | 10923 | .permission = btrfs_permission, |
33268eaf | 10924 | .listxattr = btrfs_listxattr, |
cac2f8b8 | 10925 | .get_inode_acl = btrfs_get_acl, |
996a710d | 10926 | .set_acl = btrfs_set_acl, |
e41f941a | 10927 | .update_time = btrfs_update_time, |
618e21d5 | 10928 | }; |
6e1d5dcc | 10929 | static const struct inode_operations btrfs_symlink_inode_operations = { |
6b255391 | 10930 | .get_link = page_get_link, |
f209561a | 10931 | .getattr = btrfs_getattr, |
22c44fe6 | 10932 | .setattr = btrfs_setattr, |
fdebe2bd | 10933 | .permission = btrfs_permission, |
0279b4cd | 10934 | .listxattr = btrfs_listxattr, |
e41f941a | 10935 | .update_time = btrfs_update_time, |
39279cc3 | 10936 | }; |
76dda93c | 10937 | |
82d339d9 | 10938 | const struct dentry_operations btrfs_dentry_operations = { |
76dda93c YZ |
10939 | .d_delete = btrfs_dentry_delete, |
10940 | }; |