]>
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 | |
6e26c442 | 128 | static noinline int cow_file_range(struct btrfs_inode *inode, |
771ed689 | 129 | struct page *locked_page, |
74e9194a | 130 | u64 start, u64 end, int *page_started, |
ba9145ad | 131 | unsigned long *nr_written, u64 *done_offset, |
53ffb30a | 132 | bool keep_locked, bool no_inline); |
4b67c11d NB |
133 | static struct extent_map *create_io_em(struct btrfs_inode *inode, u64 start, |
134 | u64 len, u64 orig_start, u64 block_start, | |
6f9994db LB |
135 | u64 block_len, u64 orig_block_len, |
136 | u64 ram_bytes, int compress_type, | |
137 | int type); | |
7b128766 | 138 | |
b9a9a850 QW |
139 | static int data_reloc_print_warning_inode(u64 inum, u64 offset, u64 num_bytes, |
140 | u64 root, void *warn_ctx) | |
141 | { | |
142 | struct data_reloc_warn *warn = warn_ctx; | |
143 | struct btrfs_fs_info *fs_info = warn->fs_info; | |
144 | struct extent_buffer *eb; | |
145 | struct btrfs_inode_item *inode_item; | |
146 | struct inode_fs_paths *ipath = NULL; | |
147 | struct btrfs_root *local_root; | |
148 | struct btrfs_key key; | |
149 | unsigned int nofs_flag; | |
150 | u32 nlink; | |
151 | int ret; | |
152 | ||
153 | local_root = btrfs_get_fs_root(fs_info, root, true); | |
154 | if (IS_ERR(local_root)) { | |
155 | ret = PTR_ERR(local_root); | |
156 | goto err; | |
157 | } | |
158 | ||
159 | /* This makes the path point to (inum INODE_ITEM ioff). */ | |
160 | key.objectid = inum; | |
161 | key.type = BTRFS_INODE_ITEM_KEY; | |
162 | key.offset = 0; | |
163 | ||
164 | ret = btrfs_search_slot(NULL, local_root, &key, &warn->path, 0, 0); | |
165 | if (ret) { | |
166 | btrfs_put_root(local_root); | |
167 | btrfs_release_path(&warn->path); | |
168 | goto err; | |
169 | } | |
170 | ||
171 | eb = warn->path.nodes[0]; | |
172 | inode_item = btrfs_item_ptr(eb, warn->path.slots[0], struct btrfs_inode_item); | |
173 | nlink = btrfs_inode_nlink(eb, inode_item); | |
174 | btrfs_release_path(&warn->path); | |
175 | ||
176 | nofs_flag = memalloc_nofs_save(); | |
177 | ipath = init_ipath(4096, local_root, &warn->path); | |
178 | memalloc_nofs_restore(nofs_flag); | |
179 | if (IS_ERR(ipath)) { | |
180 | btrfs_put_root(local_root); | |
181 | ret = PTR_ERR(ipath); | |
182 | ipath = NULL; | |
183 | /* | |
184 | * -ENOMEM, not a critical error, just output an generic error | |
185 | * without filename. | |
186 | */ | |
187 | btrfs_warn(fs_info, | |
188 | "checksum error at logical %llu mirror %u root %llu, inode %llu offset %llu", | |
189 | warn->logical, warn->mirror_num, root, inum, offset); | |
190 | return ret; | |
191 | } | |
192 | ret = paths_from_inode(inum, ipath); | |
193 | if (ret < 0) | |
194 | goto err; | |
195 | ||
196 | /* | |
197 | * We deliberately ignore the bit ipath might have been too small to | |
198 | * hold all of the paths here | |
199 | */ | |
200 | for (int i = 0; i < ipath->fspath->elem_cnt; i++) { | |
201 | btrfs_warn(fs_info, | |
202 | "checksum error at logical %llu mirror %u root %llu inode %llu offset %llu length %u links %u (path: %s)", | |
203 | warn->logical, warn->mirror_num, root, inum, offset, | |
204 | fs_info->sectorsize, nlink, | |
205 | (char *)(unsigned long)ipath->fspath->val[i]); | |
206 | } | |
207 | ||
208 | btrfs_put_root(local_root); | |
209 | free_ipath(ipath); | |
210 | return 0; | |
211 | ||
212 | err: | |
213 | btrfs_warn(fs_info, | |
214 | "checksum error at logical %llu mirror %u root %llu inode %llu offset %llu, path resolving failed with ret=%d", | |
215 | warn->logical, warn->mirror_num, root, inum, offset, ret); | |
216 | ||
217 | free_ipath(ipath); | |
218 | return ret; | |
219 | } | |
220 | ||
221 | /* | |
222 | * Do extra user-friendly error output (e.g. lookup all the affected files). | |
223 | * | |
224 | * Return true if we succeeded doing the backref lookup. | |
225 | * Return false if such lookup failed, and has to fallback to the old error message. | |
226 | */ | |
227 | static void print_data_reloc_error(const struct btrfs_inode *inode, u64 file_off, | |
228 | const u8 *csum, const u8 *csum_expected, | |
229 | int mirror_num) | |
230 | { | |
231 | struct btrfs_fs_info *fs_info = inode->root->fs_info; | |
232 | struct btrfs_path path = { 0 }; | |
233 | struct btrfs_key found_key = { 0 }; | |
234 | struct extent_buffer *eb; | |
235 | struct btrfs_extent_item *ei; | |
236 | const u32 csum_size = fs_info->csum_size; | |
237 | u64 logical; | |
238 | u64 flags; | |
239 | u32 item_size; | |
240 | int ret; | |
241 | ||
242 | mutex_lock(&fs_info->reloc_mutex); | |
243 | logical = btrfs_get_reloc_bg_bytenr(fs_info); | |
244 | mutex_unlock(&fs_info->reloc_mutex); | |
245 | ||
246 | if (logical == U64_MAX) { | |
247 | btrfs_warn_rl(fs_info, "has data reloc tree but no running relocation"); | |
248 | btrfs_warn_rl(fs_info, | |
249 | "csum failed root %lld ino %llu off %llu csum " CSUM_FMT " expected csum " CSUM_FMT " mirror %d", | |
250 | inode->root->root_key.objectid, btrfs_ino(inode), file_off, | |
251 | CSUM_FMT_VALUE(csum_size, csum), | |
252 | CSUM_FMT_VALUE(csum_size, csum_expected), | |
253 | mirror_num); | |
254 | return; | |
255 | } | |
256 | ||
257 | logical += file_off; | |
258 | btrfs_warn_rl(fs_info, | |
259 | "csum failed root %lld ino %llu off %llu logical %llu csum " CSUM_FMT " expected csum " CSUM_FMT " mirror %d", | |
260 | inode->root->root_key.objectid, | |
261 | btrfs_ino(inode), file_off, logical, | |
262 | CSUM_FMT_VALUE(csum_size, csum), | |
263 | CSUM_FMT_VALUE(csum_size, csum_expected), | |
264 | mirror_num); | |
265 | ||
266 | ret = extent_from_logical(fs_info, logical, &path, &found_key, &flags); | |
267 | if (ret < 0) { | |
268 | btrfs_err_rl(fs_info, "failed to lookup extent item for logical %llu: %d", | |
269 | logical, ret); | |
270 | return; | |
271 | } | |
272 | eb = path.nodes[0]; | |
273 | ei = btrfs_item_ptr(eb, path.slots[0], struct btrfs_extent_item); | |
274 | item_size = btrfs_item_size(eb, path.slots[0]); | |
275 | if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) { | |
276 | unsigned long ptr = 0; | |
277 | u64 ref_root; | |
278 | u8 ref_level; | |
279 | ||
b7f9945a | 280 | while (true) { |
b9a9a850 QW |
281 | ret = tree_backref_for_extent(&ptr, eb, &found_key, ei, |
282 | item_size, &ref_root, | |
283 | &ref_level); | |
b7f9945a QW |
284 | if (ret < 0) { |
285 | btrfs_warn_rl(fs_info, | |
286 | "failed to resolve tree backref for logical %llu: %d", | |
287 | logical, ret); | |
288 | break; | |
289 | } | |
290 | if (ret > 0) | |
291 | break; | |
292 | ||
b9a9a850 QW |
293 | btrfs_warn_rl(fs_info, |
294 | "csum error at logical %llu mirror %u: metadata %s (level %d) in tree %llu", | |
295 | logical, mirror_num, | |
296 | (ref_level ? "node" : "leaf"), | |
b7f9945a QW |
297 | ref_level, ref_root); |
298 | } | |
b9a9a850 QW |
299 | btrfs_release_path(&path); |
300 | } else { | |
301 | struct btrfs_backref_walk_ctx ctx = { 0 }; | |
302 | struct data_reloc_warn reloc_warn = { 0 }; | |
303 | ||
304 | btrfs_release_path(&path); | |
305 | ||
306 | ctx.bytenr = found_key.objectid; | |
307 | ctx.extent_item_pos = logical - found_key.objectid; | |
308 | ctx.fs_info = fs_info; | |
309 | ||
310 | reloc_warn.logical = logical; | |
311 | reloc_warn.extent_item_size = found_key.offset; | |
312 | reloc_warn.mirror_num = mirror_num; | |
313 | reloc_warn.fs_info = fs_info; | |
314 | ||
315 | iterate_extent_inodes(&ctx, true, | |
316 | data_reloc_print_warning_inode, &reloc_warn); | |
317 | } | |
318 | } | |
319 | ||
f60acad3 JB |
320 | static void __cold btrfs_print_data_csum_error(struct btrfs_inode *inode, |
321 | u64 logical_start, u8 *csum, u8 *csum_expected, int mirror_num) | |
322 | { | |
323 | struct btrfs_root *root = inode->root; | |
324 | const u32 csum_size = root->fs_info->csum_size; | |
325 | ||
b9a9a850 QW |
326 | /* For data reloc tree, it's better to do a backref lookup instead. */ |
327 | if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID) | |
328 | return print_data_reloc_error(inode, logical_start, csum, | |
329 | csum_expected, mirror_num); | |
330 | ||
f60acad3 JB |
331 | /* Output without objectid, which is more meaningful */ |
332 | if (root->root_key.objectid >= BTRFS_LAST_FREE_OBJECTID) { | |
333 | btrfs_warn_rl(root->fs_info, | |
334 | "csum failed root %lld ino %lld off %llu csum " CSUM_FMT " expected csum " CSUM_FMT " mirror %d", | |
335 | root->root_key.objectid, btrfs_ino(inode), | |
336 | logical_start, | |
337 | CSUM_FMT_VALUE(csum_size, csum), | |
338 | CSUM_FMT_VALUE(csum_size, csum_expected), | |
339 | mirror_num); | |
340 | } else { | |
341 | btrfs_warn_rl(root->fs_info, | |
342 | "csum failed root %llu ino %llu off %llu csum " CSUM_FMT " expected csum " CSUM_FMT " mirror %d", | |
343 | root->root_key.objectid, btrfs_ino(inode), | |
344 | logical_start, | |
345 | CSUM_FMT_VALUE(csum_size, csum), | |
346 | CSUM_FMT_VALUE(csum_size, csum_expected), | |
347 | mirror_num); | |
348 | } | |
349 | } | |
350 | ||
a14b78ad GR |
351 | /* |
352 | * btrfs_inode_lock - lock inode i_rwsem based on arguments passed | |
353 | * | |
354 | * ilock_flags can have the following bit set: | |
355 | * | |
356 | * BTRFS_ILOCK_SHARED - acquire a shared lock on the inode | |
357 | * BTRFS_ILOCK_TRY - try to acquire the lock, if fails on first attempt | |
358 | * return -EAGAIN | |
8318ba79 | 359 | * BTRFS_ILOCK_MMAP - acquire a write lock on the i_mmap_lock |
a14b78ad | 360 | */ |
29b6352b | 361 | int btrfs_inode_lock(struct btrfs_inode *inode, unsigned int ilock_flags) |
a14b78ad GR |
362 | { |
363 | if (ilock_flags & BTRFS_ILOCK_SHARED) { | |
364 | if (ilock_flags & BTRFS_ILOCK_TRY) { | |
29b6352b | 365 | if (!inode_trylock_shared(&inode->vfs_inode)) |
a14b78ad GR |
366 | return -EAGAIN; |
367 | else | |
368 | return 0; | |
369 | } | |
29b6352b | 370 | inode_lock_shared(&inode->vfs_inode); |
a14b78ad GR |
371 | } else { |
372 | if (ilock_flags & BTRFS_ILOCK_TRY) { | |
29b6352b | 373 | if (!inode_trylock(&inode->vfs_inode)) |
a14b78ad GR |
374 | return -EAGAIN; |
375 | else | |
376 | return 0; | |
377 | } | |
29b6352b | 378 | inode_lock(&inode->vfs_inode); |
a14b78ad | 379 | } |
8318ba79 | 380 | if (ilock_flags & BTRFS_ILOCK_MMAP) |
29b6352b | 381 | down_write(&inode->i_mmap_lock); |
a14b78ad GR |
382 | return 0; |
383 | } | |
384 | ||
385 | /* | |
386 | * btrfs_inode_unlock - unock inode i_rwsem | |
387 | * | |
388 | * ilock_flags should contain the same bits set as passed to btrfs_inode_lock() | |
389 | * to decide whether the lock acquired is shared or exclusive. | |
390 | */ | |
e5d4d75b | 391 | void btrfs_inode_unlock(struct btrfs_inode *inode, unsigned int ilock_flags) |
a14b78ad | 392 | { |
8318ba79 | 393 | if (ilock_flags & BTRFS_ILOCK_MMAP) |
e5d4d75b | 394 | up_write(&inode->i_mmap_lock); |
a14b78ad | 395 | if (ilock_flags & BTRFS_ILOCK_SHARED) |
e5d4d75b | 396 | inode_unlock_shared(&inode->vfs_inode); |
a14b78ad | 397 | else |
e5d4d75b | 398 | inode_unlock(&inode->vfs_inode); |
a14b78ad GR |
399 | } |
400 | ||
52427260 QW |
401 | /* |
402 | * Cleanup all submitted ordered extents in specified range to handle errors | |
52042d8e | 403 | * from the btrfs_run_delalloc_range() callback. |
52427260 QW |
404 | * |
405 | * NOTE: caller must ensure that when an error happens, it can not call | |
406 | * extent_clear_unlock_delalloc() to clear both the bits EXTENT_DO_ACCOUNTING | |
407 | * and EXTENT_DELALLOC simultaneously, because that causes the reserved metadata | |
408 | * to be released, which we want to happen only when finishing the ordered | |
d1051d6e | 409 | * extent (btrfs_finish_ordered_io()). |
52427260 | 410 | */ |
64e1db56 | 411 | static inline void btrfs_cleanup_ordered_extents(struct btrfs_inode *inode, |
d1051d6e NB |
412 | struct page *locked_page, |
413 | u64 offset, u64 bytes) | |
52427260 | 414 | { |
63d71450 NA |
415 | unsigned long index = offset >> PAGE_SHIFT; |
416 | unsigned long end_index = (offset + bytes - 1) >> PAGE_SHIFT; | |
0e47b25c | 417 | u64 page_start = 0, page_end = 0; |
63d71450 NA |
418 | struct page *page; |
419 | ||
99826e4c NA |
420 | if (locked_page) { |
421 | page_start = page_offset(locked_page); | |
422 | page_end = page_start + PAGE_SIZE - 1; | |
423 | } | |
424 | ||
63d71450 | 425 | while (index <= end_index) { |
968f2566 QW |
426 | /* |
427 | * For locked page, we will call end_extent_writepage() on it | |
428 | * in run_delalloc_range() for the error handling. That | |
429 | * end_extent_writepage() function will call | |
430 | * btrfs_mark_ordered_io_finished() to clear page Ordered and | |
431 | * run the ordered extent accounting. | |
432 | * | |
433 | * Here we can't just clear the Ordered bit, or | |
434 | * btrfs_mark_ordered_io_finished() would skip the accounting | |
435 | * for the page range, and the ordered extent will never finish. | |
436 | */ | |
99826e4c | 437 | if (locked_page && index == (page_start >> PAGE_SHIFT)) { |
968f2566 QW |
438 | index++; |
439 | continue; | |
440 | } | |
64e1db56 | 441 | page = find_get_page(inode->vfs_inode.i_mapping, index); |
63d71450 NA |
442 | index++; |
443 | if (!page) | |
444 | continue; | |
968f2566 QW |
445 | |
446 | /* | |
447 | * Here we just clear all Ordered bits for every page in the | |
711f447b | 448 | * range, then btrfs_mark_ordered_io_finished() will handle |
968f2566 QW |
449 | * the ordered extent accounting for the range. |
450 | */ | |
b945a463 QW |
451 | btrfs_page_clamp_clear_ordered(inode->root->fs_info, page, |
452 | offset, bytes); | |
63d71450 NA |
453 | put_page(page); |
454 | } | |
d1051d6e | 455 | |
99826e4c NA |
456 | if (locked_page) { |
457 | /* The locked page covers the full range, nothing needs to be done */ | |
458 | if (bytes + offset <= page_start + PAGE_SIZE) | |
459 | return; | |
460 | /* | |
461 | * In case this page belongs to the delalloc range being | |
462 | * instantiated then skip it, since the first page of a range is | |
463 | * going to be properly cleaned up by the caller of | |
464 | * run_delalloc_range | |
465 | */ | |
466 | if (page_start >= offset && page_end <= (offset + bytes - 1)) { | |
467 | bytes = offset + bytes - page_offset(locked_page) - PAGE_SIZE; | |
468 | offset = page_offset(locked_page) + PAGE_SIZE; | |
469 | } | |
d1051d6e NB |
470 | } |
471 | ||
711f447b | 472 | return btrfs_mark_ordered_io_finished(inode, NULL, offset, bytes, false); |
52427260 QW |
473 | } |
474 | ||
7152b425 | 475 | static int btrfs_dirty_inode(struct btrfs_inode *inode); |
7b128766 | 476 | |
f34f57a3 | 477 | static int btrfs_init_inode_security(struct btrfs_trans_handle *trans, |
3538d68d | 478 | struct btrfs_new_inode_args *args) |
0279b4cd JO |
479 | { |
480 | int err; | |
481 | ||
3538d68d OS |
482 | if (args->default_acl) { |
483 | err = __btrfs_set_acl(trans, args->inode, args->default_acl, | |
484 | ACL_TYPE_DEFAULT); | |
485 | if (err) | |
486 | return err; | |
487 | } | |
488 | if (args->acl) { | |
489 | err = __btrfs_set_acl(trans, args->inode, args->acl, ACL_TYPE_ACCESS); | |
490 | if (err) | |
491 | return err; | |
492 | } | |
493 | if (!args->default_acl && !args->acl) | |
494 | cache_no_acl(args->inode); | |
495 | return btrfs_xattr_security_init(trans, args->inode, args->dir, | |
496 | &args->dentry->d_name); | |
0279b4cd JO |
497 | } |
498 | ||
c8b97818 CM |
499 | /* |
500 | * this does all the hard work for inserting an inline extent into | |
501 | * the btree. The caller should have done a btrfs_drop_extents so that | |
502 | * no overlapping inline items exist in the btree | |
503 | */ | |
40f76580 | 504 | static int insert_inline_extent(struct btrfs_trans_handle *trans, |
8dd9872d OS |
505 | struct btrfs_path *path, |
506 | struct btrfs_inode *inode, bool extent_inserted, | |
507 | size_t size, size_t compressed_size, | |
fe3f566c | 508 | int compress_type, |
d9496e8a OS |
509 | struct page **compressed_pages, |
510 | bool update_i_size) | |
c8b97818 | 511 | { |
8dd9872d | 512 | struct btrfs_root *root = inode->root; |
c8b97818 CM |
513 | struct extent_buffer *leaf; |
514 | struct page *page = NULL; | |
515 | char *kaddr; | |
516 | unsigned long ptr; | |
517 | struct btrfs_file_extent_item *ei; | |
c8b97818 CM |
518 | int ret; |
519 | size_t cur_size = size; | |
d9496e8a | 520 | u64 i_size; |
c8b97818 | 521 | |
982f1f5d JJB |
522 | ASSERT((compressed_size > 0 && compressed_pages) || |
523 | (compressed_size == 0 && !compressed_pages)); | |
524 | ||
fe3f566c | 525 | if (compressed_size && compressed_pages) |
c8b97818 | 526 | cur_size = compressed_size; |
c8b97818 | 527 | |
1acae57b FDBM |
528 | if (!extent_inserted) { |
529 | struct btrfs_key key; | |
530 | size_t datasize; | |
c8b97818 | 531 | |
8dd9872d OS |
532 | key.objectid = btrfs_ino(inode); |
533 | key.offset = 0; | |
962a298f | 534 | key.type = BTRFS_EXTENT_DATA_KEY; |
c8b97818 | 535 | |
1acae57b | 536 | datasize = btrfs_file_extent_calc_inline_size(cur_size); |
1acae57b FDBM |
537 | ret = btrfs_insert_empty_item(trans, root, path, &key, |
538 | datasize); | |
79b4f4c6 | 539 | if (ret) |
1acae57b | 540 | goto fail; |
c8b97818 CM |
541 | } |
542 | leaf = path->nodes[0]; | |
543 | ei = btrfs_item_ptr(leaf, path->slots[0], | |
544 | struct btrfs_file_extent_item); | |
545 | btrfs_set_file_extent_generation(leaf, ei, trans->transid); | |
546 | btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE); | |
547 | btrfs_set_file_extent_encryption(leaf, ei, 0); | |
548 | btrfs_set_file_extent_other_encoding(leaf, ei, 0); | |
549 | btrfs_set_file_extent_ram_bytes(leaf, ei, size); | |
550 | ptr = btrfs_file_extent_inline_start(ei); | |
551 | ||
261507a0 | 552 | if (compress_type != BTRFS_COMPRESS_NONE) { |
c8b97818 CM |
553 | struct page *cpage; |
554 | int i = 0; | |
d397712b | 555 | while (compressed_size > 0) { |
c8b97818 | 556 | cpage = compressed_pages[i]; |
5b050f04 | 557 | cur_size = min_t(unsigned long, compressed_size, |
09cbfeaf | 558 | PAGE_SIZE); |
c8b97818 | 559 | |
4cb2e5e8 | 560 | kaddr = kmap_local_page(cpage); |
c8b97818 | 561 | write_extent_buffer(leaf, kaddr, ptr, cur_size); |
4cb2e5e8 | 562 | kunmap_local(kaddr); |
c8b97818 CM |
563 | |
564 | i++; | |
565 | ptr += cur_size; | |
566 | compressed_size -= cur_size; | |
567 | } | |
568 | btrfs_set_file_extent_compression(leaf, ei, | |
261507a0 | 569 | compress_type); |
c8b97818 | 570 | } else { |
8dd9872d | 571 | page = find_get_page(inode->vfs_inode.i_mapping, 0); |
c8b97818 | 572 | btrfs_set_file_extent_compression(leaf, ei, 0); |
4cb2e5e8 | 573 | kaddr = kmap_local_page(page); |
8dd9872d | 574 | write_extent_buffer(leaf, kaddr, ptr, size); |
4cb2e5e8 | 575 | kunmap_local(kaddr); |
09cbfeaf | 576 | put_page(page); |
c8b97818 CM |
577 | } |
578 | btrfs_mark_buffer_dirty(leaf); | |
1acae57b | 579 | btrfs_release_path(path); |
c8b97818 | 580 | |
9ddc959e JB |
581 | /* |
582 | * We align size to sectorsize for inline extents just for simplicity | |
583 | * sake. | |
584 | */ | |
8dd9872d OS |
585 | ret = btrfs_inode_set_file_extent_range(inode, 0, |
586 | ALIGN(size, root->fs_info->sectorsize)); | |
9ddc959e JB |
587 | if (ret) |
588 | goto fail; | |
589 | ||
c2167754 | 590 | /* |
d9496e8a OS |
591 | * We're an inline extent, so nobody can extend the file past i_size |
592 | * without locking a page we already have locked. | |
c2167754 | 593 | * |
d9496e8a OS |
594 | * We must do any i_size and inode updates before we unlock the pages. |
595 | * Otherwise we could end up racing with unlink. | |
c2167754 | 596 | */ |
d9496e8a OS |
597 | i_size = i_size_read(&inode->vfs_inode); |
598 | if (update_i_size && size > i_size) { | |
599 | i_size_write(&inode->vfs_inode, size); | |
600 | i_size = size; | |
601 | } | |
602 | inode->disk_i_size = i_size; | |
8dd9872d | 603 | |
c8b97818 | 604 | fail: |
79b4f4c6 | 605 | return ret; |
c8b97818 CM |
606 | } |
607 | ||
608 | ||
609 | /* | |
610 | * conditionally insert an inline extent into the file. This | |
611 | * does the checks required to make sure the data is small enough | |
612 | * to fit as an inline extent. | |
613 | */ | |
8dd9872d OS |
614 | static noinline int cow_file_range_inline(struct btrfs_inode *inode, u64 size, |
615 | size_t compressed_size, | |
00361589 | 616 | int compress_type, |
d9496e8a OS |
617 | struct page **compressed_pages, |
618 | bool update_i_size) | |
c8b97818 | 619 | { |
5893dfb9 | 620 | struct btrfs_drop_extents_args drop_args = { 0 }; |
a0349401 | 621 | struct btrfs_root *root = inode->root; |
0b246afa | 622 | struct btrfs_fs_info *fs_info = root->fs_info; |
00361589 | 623 | struct btrfs_trans_handle *trans; |
8dd9872d | 624 | u64 data_len = (compressed_size ?: size); |
c8b97818 | 625 | int ret; |
1acae57b | 626 | struct btrfs_path *path; |
c8b97818 | 627 | |
8dd9872d OS |
628 | /* |
629 | * We can create an inline extent if it ends at or beyond the current | |
630 | * i_size, is no larger than a sector (decompressed), and the (possibly | |
631 | * compressed) data fits in a leaf and the configured maximum inline | |
632 | * size. | |
633 | */ | |
634 | if (size < i_size_read(&inode->vfs_inode) || | |
635 | size > fs_info->sectorsize || | |
0b246afa | 636 | data_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info) || |
8dd9872d | 637 | data_len > fs_info->max_inline) |
c8b97818 | 638 | return 1; |
c8b97818 | 639 | |
1acae57b FDBM |
640 | path = btrfs_alloc_path(); |
641 | if (!path) | |
642 | return -ENOMEM; | |
643 | ||
00361589 | 644 | trans = btrfs_join_transaction(root); |
1acae57b FDBM |
645 | if (IS_ERR(trans)) { |
646 | btrfs_free_path(path); | |
00361589 | 647 | return PTR_ERR(trans); |
1acae57b | 648 | } |
a0349401 | 649 | trans->block_rsv = &inode->block_rsv; |
00361589 | 650 | |
5893dfb9 | 651 | drop_args.path = path; |
8dd9872d OS |
652 | drop_args.start = 0; |
653 | drop_args.end = fs_info->sectorsize; | |
5893dfb9 FM |
654 | drop_args.drop_cache = true; |
655 | drop_args.replace_extent = true; | |
8dd9872d | 656 | drop_args.extent_item_size = btrfs_file_extent_calc_inline_size(data_len); |
5893dfb9 | 657 | ret = btrfs_drop_extents(trans, root, inode, &drop_args); |
00361589 | 658 | if (ret) { |
66642832 | 659 | btrfs_abort_transaction(trans, ret); |
00361589 JB |
660 | goto out; |
661 | } | |
c8b97818 | 662 | |
8dd9872d OS |
663 | ret = insert_inline_extent(trans, path, inode, drop_args.extent_inserted, |
664 | size, compressed_size, compress_type, | |
d9496e8a | 665 | compressed_pages, update_i_size); |
2adcac1a | 666 | if (ret && ret != -ENOSPC) { |
66642832 | 667 | btrfs_abort_transaction(trans, ret); |
00361589 | 668 | goto out; |
2adcac1a | 669 | } else if (ret == -ENOSPC) { |
00361589 JB |
670 | ret = 1; |
671 | goto out; | |
79787eaa | 672 | } |
2adcac1a | 673 | |
8dd9872d | 674 | btrfs_update_inode_bytes(inode, size, drop_args.bytes_found); |
9a56fcd1 | 675 | ret = btrfs_update_inode(trans, root, inode); |
2766ff61 FM |
676 | if (ret && ret != -ENOSPC) { |
677 | btrfs_abort_transaction(trans, ret); | |
678 | goto out; | |
679 | } else if (ret == -ENOSPC) { | |
680 | ret = 1; | |
681 | goto out; | |
682 | } | |
683 | ||
23e3337f | 684 | btrfs_set_inode_full_sync(inode); |
00361589 | 685 | out: |
94ed938a QW |
686 | /* |
687 | * Don't forget to free the reserved space, as for inlined extent | |
688 | * it won't count as data extent, free them directly here. | |
689 | * And at reserve time, it's always aligned to page size, so | |
690 | * just free one page here. | |
691 | */ | |
a0349401 | 692 | btrfs_qgroup_free_data(inode, NULL, 0, PAGE_SIZE); |
1acae57b | 693 | btrfs_free_path(path); |
3a45bb20 | 694 | btrfs_end_transaction(trans); |
00361589 | 695 | return ret; |
c8b97818 CM |
696 | } |
697 | ||
771ed689 CM |
698 | struct async_extent { |
699 | u64 start; | |
700 | u64 ram_size; | |
701 | u64 compressed_size; | |
702 | struct page **pages; | |
703 | unsigned long nr_pages; | |
261507a0 | 704 | int compress_type; |
771ed689 CM |
705 | struct list_head list; |
706 | }; | |
707 | ||
97db1204 | 708 | struct async_chunk { |
99a81a44 | 709 | struct btrfs_inode *inode; |
771ed689 CM |
710 | struct page *locked_page; |
711 | u64 start; | |
712 | u64 end; | |
bf9486d6 | 713 | blk_opf_t write_flags; |
771ed689 | 714 | struct list_head extents; |
ec39f769 | 715 | struct cgroup_subsys_state *blkcg_css; |
771ed689 | 716 | struct btrfs_work work; |
9e895a8f | 717 | struct async_cow *async_cow; |
771ed689 CM |
718 | }; |
719 | ||
97db1204 | 720 | struct async_cow { |
97db1204 NB |
721 | atomic_t num_chunks; |
722 | struct async_chunk chunks[]; | |
771ed689 CM |
723 | }; |
724 | ||
97db1204 | 725 | static noinline int add_async_extent(struct async_chunk *cow, |
771ed689 CM |
726 | u64 start, u64 ram_size, |
727 | u64 compressed_size, | |
728 | struct page **pages, | |
261507a0 LZ |
729 | unsigned long nr_pages, |
730 | int compress_type) | |
771ed689 CM |
731 | { |
732 | struct async_extent *async_extent; | |
733 | ||
734 | async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS); | |
79787eaa | 735 | BUG_ON(!async_extent); /* -ENOMEM */ |
771ed689 CM |
736 | async_extent->start = start; |
737 | async_extent->ram_size = ram_size; | |
738 | async_extent->compressed_size = compressed_size; | |
739 | async_extent->pages = pages; | |
740 | async_extent->nr_pages = nr_pages; | |
261507a0 | 741 | async_extent->compress_type = compress_type; |
771ed689 CM |
742 | list_add_tail(&async_extent->list, &cow->extents); |
743 | return 0; | |
744 | } | |
745 | ||
42c16da6 QW |
746 | /* |
747 | * Check if the inode needs to be submitted to compression, based on mount | |
748 | * options, defragmentation, properties or heuristics. | |
749 | */ | |
808a1292 NB |
750 | static inline int inode_need_compress(struct btrfs_inode *inode, u64 start, |
751 | u64 end) | |
f79707b0 | 752 | { |
808a1292 | 753 | struct btrfs_fs_info *fs_info = inode->root->fs_info; |
f79707b0 | 754 | |
e6f9d696 | 755 | if (!btrfs_inode_can_compress(inode)) { |
42c16da6 QW |
756 | WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG), |
757 | KERN_ERR "BTRFS: unexpected compression for ino %llu\n", | |
808a1292 | 758 | btrfs_ino(inode)); |
42c16da6 QW |
759 | return 0; |
760 | } | |
0cf9b244 QW |
761 | /* |
762 | * Special check for subpage. | |
763 | * | |
764 | * We lock the full page then run each delalloc range in the page, thus | |
765 | * for the following case, we will hit some subpage specific corner case: | |
766 | * | |
767 | * 0 32K 64K | |
768 | * | |///////| |///////| | |
769 | * \- A \- B | |
770 | * | |
771 | * In above case, both range A and range B will try to unlock the full | |
772 | * page [0, 64K), causing the one finished later will have page | |
773 | * unlocked already, triggering various page lock requirement BUG_ON()s. | |
774 | * | |
775 | * So here we add an artificial limit that subpage compression can only | |
776 | * if the range is fully page aligned. | |
777 | * | |
778 | * In theory we only need to ensure the first page is fully covered, but | |
779 | * the tailing partial page will be locked until the full compression | |
780 | * finishes, delaying the write of other range. | |
781 | * | |
782 | * TODO: Make btrfs_run_delalloc_range() to lock all delalloc range | |
783 | * first to prevent any submitted async extent to unlock the full page. | |
784 | * By this, we can ensure for subpage case that only the last async_cow | |
785 | * will unlock the full page. | |
786 | */ | |
787 | if (fs_info->sectorsize < PAGE_SIZE) { | |
1280d2d1 FK |
788 | if (!PAGE_ALIGNED(start) || |
789 | !PAGE_ALIGNED(end + 1)) | |
0cf9b244 QW |
790 | return 0; |
791 | } | |
792 | ||
f79707b0 | 793 | /* force compress */ |
0b246afa | 794 | if (btrfs_test_opt(fs_info, FORCE_COMPRESS)) |
f79707b0 | 795 | return 1; |
eec63c65 | 796 | /* defrag ioctl */ |
808a1292 | 797 | if (inode->defrag_compress) |
eec63c65 | 798 | return 1; |
f79707b0 | 799 | /* bad compression ratios */ |
808a1292 | 800 | if (inode->flags & BTRFS_INODE_NOCOMPRESS) |
f79707b0 | 801 | return 0; |
0b246afa | 802 | if (btrfs_test_opt(fs_info, COMPRESS) || |
808a1292 NB |
803 | inode->flags & BTRFS_INODE_COMPRESS || |
804 | inode->prop_compress) | |
805 | return btrfs_compress_heuristic(&inode->vfs_inode, start, end); | |
f79707b0 WS |
806 | return 0; |
807 | } | |
808 | ||
6158e1ce | 809 | static inline void inode_should_defrag(struct btrfs_inode *inode, |
558732df | 810 | u64 start, u64 end, u64 num_bytes, u32 small_write) |
26d30f85 AJ |
811 | { |
812 | /* If this is a small write inside eof, kick off a defrag */ | |
813 | if (num_bytes < small_write && | |
6158e1ce | 814 | (start > 0 || end + 1 < inode->disk_i_size)) |
558732df | 815 | btrfs_add_inode_defrag(NULL, inode, small_write); |
26d30f85 AJ |
816 | } |
817 | ||
d352ac68 | 818 | /* |
771ed689 CM |
819 | * we create compressed extents in two phases. The first |
820 | * phase compresses a range of pages that have already been | |
821 | * locked (both pages and state bits are locked). | |
c8b97818 | 822 | * |
771ed689 CM |
823 | * This is done inside an ordered work queue, and the compression |
824 | * is spread across many cpus. The actual IO submission is step | |
825 | * two, and the ordered work queue takes care of making sure that | |
826 | * happens in the same order things were put onto the queue by | |
827 | * writepages and friends. | |
c8b97818 | 828 | * |
771ed689 CM |
829 | * If this code finds it can't get good compression, it puts an |
830 | * entry onto the work queue to write the uncompressed bytes. This | |
831 | * makes sure that both compressed inodes and uncompressed inodes | |
b2570314 AB |
832 | * are written in the same order that the flusher thread sent them |
833 | * down. | |
d352ac68 | 834 | */ |
ac3e9933 | 835 | static noinline int compress_file_range(struct async_chunk *async_chunk) |
b888db2b | 836 | { |
99a01bd6 DS |
837 | struct btrfs_inode *inode = async_chunk->inode; |
838 | struct btrfs_fs_info *fs_info = inode->root->fs_info; | |
a994310a | 839 | struct address_space *mapping = inode->vfs_inode.i_mapping; |
0b246afa | 840 | u64 blocksize = fs_info->sectorsize; |
1368c6da NB |
841 | u64 start = async_chunk->start; |
842 | u64 end = async_chunk->end; | |
c8b97818 | 843 | u64 actual_end; |
d98da499 | 844 | u64 i_size; |
e6dcd2dc | 845 | int ret = 0; |
c8b97818 CM |
846 | struct page **pages = NULL; |
847 | unsigned long nr_pages; | |
c8b97818 CM |
848 | unsigned long total_compressed = 0; |
849 | unsigned long total_in = 0; | |
c8b97818 CM |
850 | int i; |
851 | int will_compress; | |
0b246afa | 852 | int compress_type = fs_info->compress_type; |
ac3e9933 | 853 | int compressed_extents = 0; |
4adaa611 | 854 | int redirty = 0; |
b888db2b | 855 | |
99a01bd6 | 856 | inode_should_defrag(inode, start, end, end - start + 1, SZ_16K); |
4cb5300b | 857 | |
d98da499 JB |
858 | /* |
859 | * We need to save i_size before now because it could change in between | |
860 | * us evaluating the size and assigning it. This is because we lock and | |
861 | * unlock the page in truncate and fallocate, and then modify the i_size | |
862 | * later on. | |
863 | * | |
864 | * The barriers are to emulate READ_ONCE, remove that once i_size_read | |
865 | * does that for us. | |
866 | */ | |
867 | barrier(); | |
99a01bd6 | 868 | i_size = i_size_read(&inode->vfs_inode); |
d98da499 JB |
869 | barrier(); |
870 | actual_end = min_t(u64, i_size, end + 1); | |
c8b97818 CM |
871 | again: |
872 | will_compress = 0; | |
09cbfeaf | 873 | nr_pages = (end >> PAGE_SHIFT) - (start >> PAGE_SHIFT) + 1; |
544fe4a9 | 874 | nr_pages = min_t(unsigned long, nr_pages, BTRFS_MAX_COMPRESSED_PAGES); |
be20aa9d | 875 | |
f03d9301 CM |
876 | /* |
877 | * we don't want to send crud past the end of i_size through | |
878 | * compression, that's just a waste of CPU time. So, if the | |
879 | * end of the file is before the start of our current | |
880 | * requested range of bytes, we bail out to the uncompressed | |
881 | * cleanup code that can deal with all of this. | |
882 | * | |
883 | * It isn't really the fastest way to fix things, but this is a | |
884 | * very uncommon corner. | |
885 | */ | |
886 | if (actual_end <= start) | |
887 | goto cleanup_and_bail_uncompressed; | |
888 | ||
c8b97818 CM |
889 | total_compressed = actual_end - start; |
890 | ||
4bcbb332 | 891 | /* |
0cf9b244 | 892 | * Skip compression for a small file range(<=blocksize) that |
01327610 | 893 | * isn't an inline extent, since it doesn't save disk space at all. |
4bcbb332 SW |
894 | */ |
895 | if (total_compressed <= blocksize && | |
99a01bd6 | 896 | (start > 0 || end + 1 < inode->disk_i_size)) |
4bcbb332 SW |
897 | goto cleanup_and_bail_uncompressed; |
898 | ||
0cf9b244 QW |
899 | /* |
900 | * For subpage case, we require full page alignment for the sector | |
901 | * aligned range. | |
902 | * Thus we must also check against @actual_end, not just @end. | |
903 | */ | |
904 | if (blocksize < PAGE_SIZE) { | |
1280d2d1 FK |
905 | if (!PAGE_ALIGNED(start) || |
906 | !PAGE_ALIGNED(round_up(actual_end, blocksize))) | |
0cf9b244 QW |
907 | goto cleanup_and_bail_uncompressed; |
908 | } | |
909 | ||
069eac78 DS |
910 | total_compressed = min_t(unsigned long, total_compressed, |
911 | BTRFS_MAX_UNCOMPRESSED); | |
c8b97818 CM |
912 | total_in = 0; |
913 | ret = 0; | |
db94535d | 914 | |
771ed689 CM |
915 | /* |
916 | * we do compression for mount -o compress and when the | |
917 | * inode has not been flagged as nocompress. This flag can | |
918 | * change at any time if we discover bad compression ratios. | |
c8b97818 | 919 | */ |
99a01bd6 | 920 | if (inode_need_compress(inode, start, end)) { |
c8b97818 | 921 | WARN_ON(pages); |
31e818fe | 922 | pages = kcalloc(nr_pages, sizeof(struct page *), GFP_NOFS); |
560f7d75 LZ |
923 | if (!pages) { |
924 | /* just bail out to the uncompressed code */ | |
3527a018 | 925 | nr_pages = 0; |
560f7d75 LZ |
926 | goto cont; |
927 | } | |
c8b97818 | 928 | |
99a01bd6 DS |
929 | if (inode->defrag_compress) |
930 | compress_type = inode->defrag_compress; | |
931 | else if (inode->prop_compress) | |
932 | compress_type = inode->prop_compress; | |
261507a0 | 933 | |
4adaa611 CM |
934 | /* |
935 | * we need to call clear_page_dirty_for_io on each | |
936 | * page in the range. Otherwise applications with the file | |
937 | * mmap'd can wander in and change the page contents while | |
938 | * we are compressing them. | |
939 | * | |
940 | * If the compression fails for any reason, we set the pages | |
941 | * dirty again later on. | |
e9679de3 TT |
942 | * |
943 | * Note that the remaining part is redirtied, the start pointer | |
944 | * has moved, the end is the original one. | |
4adaa611 | 945 | */ |
e9679de3 | 946 | if (!redirty) { |
99a01bd6 | 947 | extent_range_clear_dirty_for_io(&inode->vfs_inode, start, end); |
e9679de3 TT |
948 | redirty = 1; |
949 | } | |
f51d2b59 DS |
950 | |
951 | /* Compression level is applied here and only here */ | |
952 | ret = btrfs_compress_pages( | |
953 | compress_type | (fs_info->compress_level << 4), | |
a994310a | 954 | mapping, start, |
38c31464 | 955 | pages, |
4d3a800e | 956 | &nr_pages, |
261507a0 | 957 | &total_in, |
e5d74902 | 958 | &total_compressed); |
c8b97818 CM |
959 | |
960 | if (!ret) { | |
7073017a | 961 | unsigned long offset = offset_in_page(total_compressed); |
4d3a800e | 962 | struct page *page = pages[nr_pages - 1]; |
c8b97818 CM |
963 | |
964 | /* zero the tail end of the last page, we might be | |
965 | * sending it down to disk | |
966 | */ | |
d048b9c2 IW |
967 | if (offset) |
968 | memzero_page(page, offset, PAGE_SIZE - offset); | |
c8b97818 CM |
969 | will_compress = 1; |
970 | } | |
971 | } | |
560f7d75 | 972 | cont: |
7367253a QW |
973 | /* |
974 | * Check cow_file_range() for why we don't even try to create inline | |
975 | * extent for subpage case. | |
976 | */ | |
977 | if (start == 0 && fs_info->sectorsize == PAGE_SIZE) { | |
c8b97818 | 978 | /* lets try to make an inline extent */ |
6018ba0a | 979 | if (ret || total_in < actual_end) { |
c8b97818 | 980 | /* we didn't compress the entire range, try |
771ed689 | 981 | * to make an uncompressed inline extent. |
c8b97818 | 982 | */ |
99a01bd6 | 983 | ret = cow_file_range_inline(inode, actual_end, |
a0349401 | 984 | 0, BTRFS_COMPRESS_NONE, |
d9496e8a | 985 | NULL, false); |
c8b97818 | 986 | } else { |
771ed689 | 987 | /* try making a compressed inline extent */ |
99a01bd6 | 988 | ret = cow_file_range_inline(inode, actual_end, |
fe3f566c | 989 | total_compressed, |
d9496e8a OS |
990 | compress_type, pages, |
991 | false); | |
c8b97818 | 992 | } |
79787eaa | 993 | if (ret <= 0) { |
151a41bc | 994 | unsigned long clear_flags = EXTENT_DELALLOC | |
8b62f87b JB |
995 | EXTENT_DELALLOC_NEW | EXTENT_DEFRAG | |
996 | EXTENT_DO_ACCOUNTING; | |
e6eb4314 | 997 | |
a994310a CH |
998 | if (ret < 0) |
999 | mapping_set_error(mapping, -EIO); | |
151a41bc | 1000 | |
771ed689 | 1001 | /* |
79787eaa JM |
1002 | * inline extent creation worked or returned error, |
1003 | * we don't need to create any more async work items. | |
1004 | * Unlock and free up our temp pages. | |
8b62f87b JB |
1005 | * |
1006 | * We use DO_ACCOUNTING here because we need the | |
1007 | * delalloc_release_metadata to be done _after_ we drop | |
1008 | * our outstanding extent for clearing delalloc for this | |
1009 | * range. | |
771ed689 | 1010 | */ |
99a01bd6 | 1011 | extent_clear_unlock_delalloc(inode, start, end, |
ad7ff17b | 1012 | NULL, |
74e9194a | 1013 | clear_flags, |
ba8b04c1 | 1014 | PAGE_UNLOCK | |
6869b0a8 | 1015 | PAGE_START_WRITEBACK | |
c2790a2e | 1016 | PAGE_END_WRITEBACK); |
cecc8d90 | 1017 | |
1e6e238c QW |
1018 | /* |
1019 | * Ensure we only free the compressed pages if we have | |
1020 | * them allocated, as we can still reach here with | |
1021 | * inode_need_compress() == false. | |
1022 | */ | |
1023 | if (pages) { | |
1024 | for (i = 0; i < nr_pages; i++) { | |
1025 | WARN_ON(pages[i]->mapping); | |
1026 | put_page(pages[i]); | |
1027 | } | |
1028 | kfree(pages); | |
cecc8d90 | 1029 | } |
cecc8d90 | 1030 | return 0; |
c8b97818 CM |
1031 | } |
1032 | } | |
1033 | ||
1034 | if (will_compress) { | |
1035 | /* | |
1036 | * we aren't doing an inline extent round the compressed size | |
1037 | * up to a block size boundary so the allocator does sane | |
1038 | * things | |
1039 | */ | |
fda2832f | 1040 | total_compressed = ALIGN(total_compressed, blocksize); |
c8b97818 CM |
1041 | |
1042 | /* | |
1043 | * one last check to make sure the compression is really a | |
170607eb TT |
1044 | * win, compare the page count read with the blocks on disk, |
1045 | * compression must free at least one sector size | |
c8b97818 | 1046 | */ |
4c162778 | 1047 | total_in = round_up(total_in, fs_info->sectorsize); |
170607eb | 1048 | if (total_compressed + blocksize <= total_in) { |
ac3e9933 | 1049 | compressed_extents++; |
c8bb0c8b AS |
1050 | |
1051 | /* | |
1052 | * The async work queues will take care of doing actual | |
1053 | * allocation on disk for these compressed pages, and | |
1054 | * will submit them to the elevator. | |
1055 | */ | |
b5326271 | 1056 | add_async_extent(async_chunk, start, total_in, |
4d3a800e | 1057 | total_compressed, pages, nr_pages, |
c8bb0c8b AS |
1058 | compress_type); |
1059 | ||
1170862d TT |
1060 | if (start + total_in < end) { |
1061 | start += total_in; | |
c8bb0c8b AS |
1062 | pages = NULL; |
1063 | cond_resched(); | |
1064 | goto again; | |
1065 | } | |
ac3e9933 | 1066 | return compressed_extents; |
c8b97818 CM |
1067 | } |
1068 | } | |
c8bb0c8b | 1069 | if (pages) { |
c8b97818 CM |
1070 | /* |
1071 | * the compression code ran but failed to make things smaller, | |
1072 | * free any pages it allocated and our page pointer array | |
1073 | */ | |
4d3a800e | 1074 | for (i = 0; i < nr_pages; i++) { |
70b99e69 | 1075 | WARN_ON(pages[i]->mapping); |
09cbfeaf | 1076 | put_page(pages[i]); |
c8b97818 CM |
1077 | } |
1078 | kfree(pages); | |
1079 | pages = NULL; | |
1080 | total_compressed = 0; | |
4d3a800e | 1081 | nr_pages = 0; |
c8b97818 CM |
1082 | |
1083 | /* flag the file so we don't compress in the future */ | |
0b246afa | 1084 | if (!btrfs_test_opt(fs_info, FORCE_COMPRESS) && |
99a01bd6 DS |
1085 | !(inode->prop_compress)) { |
1086 | inode->flags |= BTRFS_INODE_NOCOMPRESS; | |
1e701a32 | 1087 | } |
c8b97818 | 1088 | } |
f03d9301 | 1089 | cleanup_and_bail_uncompressed: |
c8bb0c8b AS |
1090 | /* |
1091 | * No compression, but we still need to write the pages in the file | |
1092 | * we've been given so far. redirty the locked page if it corresponds | |
1093 | * to our extent and set things up for the async work queue to run | |
1094 | * cow_file_range to do the normal delalloc dance. | |
1095 | */ | |
1d53c9e6 CM |
1096 | if (async_chunk->locked_page && |
1097 | (page_offset(async_chunk->locked_page) >= start && | |
1098 | page_offset(async_chunk->locked_page)) <= end) { | |
1368c6da | 1099 | __set_page_dirty_nobuffers(async_chunk->locked_page); |
c8bb0c8b | 1100 | /* unlocked later on in the async handlers */ |
1d53c9e6 | 1101 | } |
c8bb0c8b AS |
1102 | |
1103 | if (redirty) | |
99a01bd6 | 1104 | extent_range_redirty_for_io(&inode->vfs_inode, start, end); |
b5326271 | 1105 | add_async_extent(async_chunk, start, end - start + 1, 0, NULL, 0, |
c8bb0c8b | 1106 | BTRFS_COMPRESS_NONE); |
ac3e9933 | 1107 | compressed_extents++; |
3b951516 | 1108 | |
ac3e9933 | 1109 | return compressed_extents; |
771ed689 | 1110 | } |
771ed689 | 1111 | |
40ae837b FM |
1112 | static void free_async_extent_pages(struct async_extent *async_extent) |
1113 | { | |
1114 | int i; | |
1115 | ||
1116 | if (!async_extent->pages) | |
1117 | return; | |
1118 | ||
1119 | for (i = 0; i < async_extent->nr_pages; i++) { | |
1120 | WARN_ON(async_extent->pages[i]->mapping); | |
09cbfeaf | 1121 | put_page(async_extent->pages[i]); |
40ae837b FM |
1122 | } |
1123 | kfree(async_extent->pages); | |
1124 | async_extent->nr_pages = 0; | |
1125 | async_extent->pages = NULL; | |
771ed689 CM |
1126 | } |
1127 | ||
2b83a0ee QW |
1128 | static int submit_uncompressed_range(struct btrfs_inode *inode, |
1129 | struct async_extent *async_extent, | |
1130 | struct page *locked_page) | |
771ed689 | 1131 | { |
2b83a0ee QW |
1132 | u64 start = async_extent->start; |
1133 | u64 end = async_extent->start + async_extent->ram_size - 1; | |
1134 | unsigned long nr_written = 0; | |
1135 | int page_started = 0; | |
1136 | int ret; | |
7027f871 CH |
1137 | struct writeback_control wbc = { |
1138 | .sync_mode = WB_SYNC_ALL, | |
1139 | .range_start = start, | |
1140 | .range_end = end, | |
1141 | .no_cgroup_owner = 1, | |
1142 | }; | |
771ed689 | 1143 | |
2b83a0ee QW |
1144 | /* |
1145 | * Call cow_file_range() to run the delalloc range directly, since we | |
1146 | * won't go to NOCOW or async path again. | |
1147 | * | |
1148 | * Also we call cow_file_range() with @unlock_page == 0, so that we | |
1149 | * can directly submit them without interruption. | |
1150 | */ | |
1151 | ret = cow_file_range(inode, locked_page, start, end, &page_started, | |
53ffb30a | 1152 | &nr_written, NULL, true, false); |
2b83a0ee | 1153 | /* Inline extent inserted, page gets unlocked and everything is done */ |
e43a6210 CH |
1154 | if (page_started) |
1155 | return 0; | |
1156 | ||
2b83a0ee | 1157 | if (ret < 0) { |
71aa147b NA |
1158 | btrfs_cleanup_ordered_extents(inode, locked_page, start, end - start + 1); |
1159 | if (locked_page) { | |
1160 | const u64 page_start = page_offset(locked_page); | |
1161 | const u64 page_end = page_start + PAGE_SIZE - 1; | |
1162 | ||
71aa147b NA |
1163 | set_page_writeback(locked_page); |
1164 | end_page_writeback(locked_page); | |
1165 | end_extent_writepage(locked_page, ret, page_start, page_end); | |
2b83a0ee | 1166 | unlock_page(locked_page); |
71aa147b | 1167 | } |
e43a6210 | 1168 | return ret; |
2b83a0ee | 1169 | } |
771ed689 | 1170 | |
2b83a0ee | 1171 | /* All pages will be unlocked, including @locked_page */ |
7027f871 CH |
1172 | wbc_attach_fdatawrite_inode(&wbc, &inode->vfs_inode); |
1173 | ret = extent_write_locked_range(&inode->vfs_inode, start, end, &wbc); | |
1174 | wbc_detach_inode(&wbc); | |
1175 | return ret; | |
2b83a0ee | 1176 | } |
79787eaa | 1177 | |
b4ccace8 QW |
1178 | static int submit_one_async_extent(struct btrfs_inode *inode, |
1179 | struct async_chunk *async_chunk, | |
1180 | struct async_extent *async_extent, | |
1181 | u64 *alloc_hint) | |
771ed689 | 1182 | { |
b4ccace8 QW |
1183 | struct extent_io_tree *io_tree = &inode->io_tree; |
1184 | struct btrfs_root *root = inode->root; | |
1185 | struct btrfs_fs_info *fs_info = root->fs_info; | |
d611935b | 1186 | struct btrfs_ordered_extent *ordered; |
771ed689 | 1187 | struct btrfs_key ins; |
2b83a0ee | 1188 | struct page *locked_page = NULL; |
771ed689 | 1189 | struct extent_map *em; |
f5a84ee3 | 1190 | int ret = 0; |
b4ccace8 QW |
1191 | u64 start = async_extent->start; |
1192 | u64 end = async_extent->start + async_extent->ram_size - 1; | |
771ed689 | 1193 | |
896d7c1a CH |
1194 | if (async_chunk->blkcg_css) |
1195 | kthread_associate_blkcg(async_chunk->blkcg_css); | |
1196 | ||
2b83a0ee QW |
1197 | /* |
1198 | * If async_chunk->locked_page is in the async_extent range, we need to | |
1199 | * handle it. | |
1200 | */ | |
1201 | if (async_chunk->locked_page) { | |
1202 | u64 locked_page_start = page_offset(async_chunk->locked_page); | |
1203 | u64 locked_page_end = locked_page_start + PAGE_SIZE - 1; | |
3e04e7f1 | 1204 | |
2b83a0ee QW |
1205 | if (!(start >= locked_page_end || end <= locked_page_start)) |
1206 | locked_page = async_chunk->locked_page; | |
b4ccace8 | 1207 | } |
570eb97b | 1208 | lock_extent(io_tree, start, end, NULL); |
ce62003f | 1209 | |
2b83a0ee | 1210 | /* We have fall back to uncompressed write */ |
e43a6210 CH |
1211 | if (!async_extent->pages) { |
1212 | ret = submit_uncompressed_range(inode, async_extent, locked_page); | |
1213 | goto done; | |
1214 | } | |
ce62003f | 1215 | |
b4ccace8 QW |
1216 | ret = btrfs_reserve_extent(root, async_extent->ram_size, |
1217 | async_extent->compressed_size, | |
1218 | async_extent->compressed_size, | |
1219 | 0, *alloc_hint, &ins, 1, 1); | |
1220 | if (ret) { | |
c2167754 | 1221 | /* |
b4ccace8 QW |
1222 | * Here we used to try again by going back to non-compressed |
1223 | * path for ENOSPC. But we can't reserve space even for | |
1224 | * compressed size, how could it work for uncompressed size | |
1225 | * which requires larger size? So here we directly go error | |
1226 | * path. | |
c2167754 | 1227 | */ |
b4ccace8 QW |
1228 | goto out_free; |
1229 | } | |
1230 | ||
1231 | /* Here we're doing allocation and writeback of the compressed pages */ | |
1232 | em = create_io_em(inode, start, | |
1233 | async_extent->ram_size, /* len */ | |
1234 | start, /* orig_start */ | |
1235 | ins.objectid, /* block_start */ | |
1236 | ins.offset, /* block_len */ | |
1237 | ins.offset, /* orig_block_len */ | |
1238 | async_extent->ram_size, /* ram_bytes */ | |
1239 | async_extent->compress_type, | |
1240 | BTRFS_ORDERED_COMPRESSED); | |
1241 | if (IS_ERR(em)) { | |
1242 | ret = PTR_ERR(em); | |
1243 | goto out_free_reserve; | |
1244 | } | |
1245 | free_extent_map(em); | |
771ed689 | 1246 | |
d611935b | 1247 | ordered = btrfs_alloc_ordered_extent(inode, start, /* file_offset */ |
cb36a9bb OS |
1248 | async_extent->ram_size, /* num_bytes */ |
1249 | async_extent->ram_size, /* ram_bytes */ | |
1250 | ins.objectid, /* disk_bytenr */ | |
1251 | ins.offset, /* disk_num_bytes */ | |
1252 | 0, /* offset */ | |
1253 | 1 << BTRFS_ORDERED_COMPRESSED, | |
1254 | async_extent->compress_type); | |
d611935b | 1255 | if (IS_ERR(ordered)) { |
4c0c8cfc | 1256 | btrfs_drop_extent_map_range(inode, start, end, false); |
d611935b | 1257 | ret = PTR_ERR(ordered); |
b4ccace8 | 1258 | goto out_free_reserve; |
771ed689 | 1259 | } |
b4ccace8 QW |
1260 | btrfs_dec_block_group_reservations(fs_info, ins.objectid); |
1261 | ||
1262 | /* Clear dirty, set writeback and unlock the pages. */ | |
1263 | extent_clear_unlock_delalloc(inode, start, end, | |
1264 | NULL, EXTENT_LOCKED | EXTENT_DELALLOC, | |
1265 | PAGE_UNLOCK | PAGE_START_WRITEBACK); | |
d611935b | 1266 | btrfs_submit_compressed_write(ordered, |
b4ccace8 QW |
1267 | async_extent->pages, /* compressed_pages */ |
1268 | async_extent->nr_pages, | |
05d06a5c | 1269 | async_chunk->write_flags, true); |
b4ccace8 | 1270 | *alloc_hint = ins.objectid + ins.offset; |
e43a6210 | 1271 | done: |
896d7c1a CH |
1272 | if (async_chunk->blkcg_css) |
1273 | kthread_associate_blkcg(NULL); | |
b4ccace8 QW |
1274 | kfree(async_extent); |
1275 | return ret; | |
1276 | ||
3e04e7f1 | 1277 | out_free_reserve: |
0b246afa | 1278 | btrfs_dec_block_group_reservations(fs_info, ins.objectid); |
2ff7e61e | 1279 | btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1); |
79787eaa | 1280 | out_free: |
a994310a | 1281 | mapping_set_error(inode->vfs_inode.i_mapping, -EIO); |
b4ccace8 | 1282 | extent_clear_unlock_delalloc(inode, start, end, |
c2790a2e | 1283 | NULL, EXTENT_LOCKED | EXTENT_DELALLOC | |
a7e3b975 | 1284 | EXTENT_DELALLOC_NEW | |
151a41bc | 1285 | EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING, |
6869b0a8 | 1286 | PAGE_UNLOCK | PAGE_START_WRITEBACK | |
a994310a | 1287 | PAGE_END_WRITEBACK); |
40ae837b | 1288 | free_async_extent_pages(async_extent); |
e43a6210 | 1289 | goto done; |
b4ccace8 QW |
1290 | } |
1291 | ||
1292 | /* | |
1293 | * Phase two of compressed writeback. This is the ordered portion of the code, | |
1294 | * which only gets called in the order the work was queued. We walk all the | |
1295 | * async extents created by compress_file_range and send them down to the disk. | |
1296 | */ | |
1297 | static noinline void submit_compressed_extents(struct async_chunk *async_chunk) | |
1298 | { | |
99a81a44 | 1299 | struct btrfs_inode *inode = async_chunk->inode; |
b4ccace8 QW |
1300 | struct btrfs_fs_info *fs_info = inode->root->fs_info; |
1301 | struct async_extent *async_extent; | |
1302 | u64 alloc_hint = 0; | |
1303 | int ret = 0; | |
1304 | ||
1305 | while (!list_empty(&async_chunk->extents)) { | |
1306 | u64 extent_start; | |
1307 | u64 ram_size; | |
1308 | ||
1309 | async_extent = list_entry(async_chunk->extents.next, | |
1310 | struct async_extent, list); | |
1311 | list_del(&async_extent->list); | |
1312 | extent_start = async_extent->start; | |
1313 | ram_size = async_extent->ram_size; | |
1314 | ||
1315 | ret = submit_one_async_extent(inode, async_chunk, async_extent, | |
1316 | &alloc_hint); | |
1317 | btrfs_debug(fs_info, | |
1318 | "async extent submission failed root=%lld inode=%llu start=%llu len=%llu ret=%d", | |
1319 | inode->root->root_key.objectid, | |
1320 | btrfs_ino(inode), extent_start, ram_size, ret); | |
1321 | } | |
771ed689 CM |
1322 | } |
1323 | ||
43c69849 | 1324 | static u64 get_extent_allocation_hint(struct btrfs_inode *inode, u64 start, |
4b46fce2 JB |
1325 | u64 num_bytes) |
1326 | { | |
43c69849 | 1327 | struct extent_map_tree *em_tree = &inode->extent_tree; |
4b46fce2 JB |
1328 | struct extent_map *em; |
1329 | u64 alloc_hint = 0; | |
1330 | ||
1331 | read_lock(&em_tree->lock); | |
1332 | em = search_extent_mapping(em_tree, start, num_bytes); | |
1333 | if (em) { | |
1334 | /* | |
1335 | * if block start isn't an actual block number then find the | |
1336 | * first block in this inode and use that as a hint. If that | |
1337 | * block is also bogus then just don't worry about it. | |
1338 | */ | |
1339 | if (em->block_start >= EXTENT_MAP_LAST_BYTE) { | |
1340 | free_extent_map(em); | |
1341 | em = search_extent_mapping(em_tree, 0, 0); | |
1342 | if (em && em->block_start < EXTENT_MAP_LAST_BYTE) | |
1343 | alloc_hint = em->block_start; | |
1344 | if (em) | |
1345 | free_extent_map(em); | |
1346 | } else { | |
1347 | alloc_hint = em->block_start; | |
1348 | free_extent_map(em); | |
1349 | } | |
1350 | } | |
1351 | read_unlock(&em_tree->lock); | |
1352 | ||
1353 | return alloc_hint; | |
1354 | } | |
1355 | ||
771ed689 CM |
1356 | /* |
1357 | * when extent_io.c finds a delayed allocation range in the file, | |
1358 | * the call backs end up in this code. The basic idea is to | |
1359 | * allocate extents on disk for the range, and create ordered data structs | |
1360 | * in ram to track those extents. | |
1361 | * | |
1362 | * locked_page is the page that writepage had locked already. We use | |
1363 | * it to make sure we don't do extra locks or unlocks. | |
1364 | * | |
ba9145ad | 1365 | * When this function fails, it unlocks all pages except @locked_page. |
9ce7466f | 1366 | * |
ba9145ad CH |
1367 | * When this function successfully creates an inline extent, it sets page_started |
1368 | * to 1 and unlocks all pages including locked_page and starts I/O on them. | |
1369 | * (In reality inline extents are limited to a single page, so locked_page is | |
1370 | * the only page handled anyway). | |
9ce7466f | 1371 | * |
ba9145ad CH |
1372 | * When this function succeed and creates a normal extent, the page locking |
1373 | * status depends on the passed in flags: | |
9ce7466f | 1374 | * |
ba9145ad CH |
1375 | * - If @keep_locked is set, all pages are kept locked. |
1376 | * - Else all pages except for @locked_page are unlocked. | |
9ce7466f NA |
1377 | * |
1378 | * When a failure happens in the second or later iteration of the | |
1379 | * while-loop, the ordered extents created in previous iterations are kept | |
1380 | * intact. So, the caller must clean them up by calling | |
1381 | * btrfs_cleanup_ordered_extents(). See btrfs_run_delalloc_range() for | |
1382 | * example. | |
771ed689 | 1383 | */ |
6e26c442 | 1384 | static noinline int cow_file_range(struct btrfs_inode *inode, |
00361589 | 1385 | struct page *locked_page, |
74e9194a | 1386 | u64 start, u64 end, int *page_started, |
ba9145ad | 1387 | unsigned long *nr_written, u64 *done_offset, |
53ffb30a | 1388 | bool keep_locked, bool no_inline) |
771ed689 | 1389 | { |
6e26c442 NB |
1390 | struct btrfs_root *root = inode->root; |
1391 | struct btrfs_fs_info *fs_info = root->fs_info; | |
771ed689 | 1392 | u64 alloc_hint = 0; |
9ce7466f | 1393 | u64 orig_start = start; |
771ed689 CM |
1394 | u64 num_bytes; |
1395 | unsigned long ram_size; | |
a315e68f | 1396 | u64 cur_alloc_size = 0; |
432cd2a1 | 1397 | u64 min_alloc_size; |
0b246afa | 1398 | u64 blocksize = fs_info->sectorsize; |
771ed689 CM |
1399 | struct btrfs_key ins; |
1400 | struct extent_map *em; | |
a315e68f FM |
1401 | unsigned clear_bits; |
1402 | unsigned long page_ops; | |
1403 | bool extent_reserved = false; | |
771ed689 CM |
1404 | int ret = 0; |
1405 | ||
6e26c442 | 1406 | if (btrfs_is_free_space_inode(inode)) { |
29bce2f3 JB |
1407 | ret = -EINVAL; |
1408 | goto out_unlock; | |
02ecd2c2 | 1409 | } |
771ed689 | 1410 | |
fda2832f | 1411 | num_bytes = ALIGN(end - start + 1, blocksize); |
771ed689 | 1412 | num_bytes = max(blocksize, num_bytes); |
566b1760 | 1413 | ASSERT(num_bytes <= btrfs_super_total_bytes(fs_info->super_copy)); |
771ed689 | 1414 | |
6e26c442 | 1415 | inode_should_defrag(inode, start, end, num_bytes, SZ_64K); |
4cb5300b | 1416 | |
7367253a QW |
1417 | /* |
1418 | * Due to the page size limit, for subpage we can only trigger the | |
1419 | * writeback for the dirty sectors of page, that means data writeback | |
1420 | * is doing more writeback than what we want. | |
1421 | * | |
1422 | * This is especially unexpected for some call sites like fallocate, | |
1423 | * where we only increase i_size after everything is done. | |
1424 | * This means we can trigger inline extent even if we didn't want to. | |
1425 | * So here we skip inline extent creation completely. | |
1426 | */ | |
53ffb30a | 1427 | if (start == 0 && fs_info->sectorsize == PAGE_SIZE && !no_inline) { |
8dd9872d OS |
1428 | u64 actual_end = min_t(u64, i_size_read(&inode->vfs_inode), |
1429 | end + 1); | |
1430 | ||
771ed689 | 1431 | /* lets try to make an inline extent */ |
8dd9872d | 1432 | ret = cow_file_range_inline(inode, actual_end, 0, |
d9496e8a | 1433 | BTRFS_COMPRESS_NONE, NULL, false); |
771ed689 | 1434 | if (ret == 0) { |
8b62f87b JB |
1435 | /* |
1436 | * We use DO_ACCOUNTING here because we need the | |
1437 | * delalloc_release_metadata to be run _after_ we drop | |
1438 | * our outstanding extent for clearing delalloc for this | |
1439 | * range. | |
1440 | */ | |
4750af3b QW |
1441 | extent_clear_unlock_delalloc(inode, start, end, |
1442 | locked_page, | |
c2790a2e | 1443 | EXTENT_LOCKED | EXTENT_DELALLOC | |
8b62f87b JB |
1444 | EXTENT_DELALLOC_NEW | EXTENT_DEFRAG | |
1445 | EXTENT_DO_ACCOUNTING, PAGE_UNLOCK | | |
6869b0a8 | 1446 | PAGE_START_WRITEBACK | PAGE_END_WRITEBACK); |
771ed689 | 1447 | *nr_written = *nr_written + |
09cbfeaf | 1448 | (end - start + PAGE_SIZE) / PAGE_SIZE; |
771ed689 | 1449 | *page_started = 1; |
4750af3b QW |
1450 | /* |
1451 | * locked_page is locked by the caller of | |
1452 | * writepage_delalloc(), not locked by | |
1453 | * __process_pages_contig(). | |
1454 | * | |
1455 | * We can't let __process_pages_contig() to unlock it, | |
1456 | * as it doesn't have any subpage::writers recorded. | |
1457 | * | |
1458 | * Here we manually unlock the page, since the caller | |
1459 | * can't use page_started to determine if it's an | |
1460 | * inline extent or a compressed extent. | |
1461 | */ | |
1462 | unlock_page(locked_page); | |
771ed689 | 1463 | goto out; |
79787eaa | 1464 | } else if (ret < 0) { |
79787eaa | 1465 | goto out_unlock; |
771ed689 CM |
1466 | } |
1467 | } | |
1468 | ||
6e26c442 | 1469 | alloc_hint = get_extent_allocation_hint(inode, start, num_bytes); |
771ed689 | 1470 | |
432cd2a1 FM |
1471 | /* |
1472 | * Relocation relies on the relocated extents to have exactly the same | |
1473 | * size as the original extents. Normally writeback for relocation data | |
1474 | * extents follows a NOCOW path because relocation preallocates the | |
1475 | * extents. However, due to an operation such as scrub turning a block | |
1476 | * group to RO mode, it may fallback to COW mode, so we must make sure | |
1477 | * an extent allocated during COW has exactly the requested size and can | |
1478 | * not be split into smaller extents, otherwise relocation breaks and | |
1479 | * fails during the stage where it updates the bytenr of file extent | |
1480 | * items. | |
1481 | */ | |
37f00a6d | 1482 | if (btrfs_is_data_reloc_root(root)) |
432cd2a1 FM |
1483 | min_alloc_size = num_bytes; |
1484 | else | |
1485 | min_alloc_size = fs_info->sectorsize; | |
1486 | ||
3752d22f | 1487 | while (num_bytes > 0) { |
34bfaf15 CH |
1488 | struct btrfs_ordered_extent *ordered; |
1489 | ||
3752d22f | 1490 | cur_alloc_size = num_bytes; |
18513091 | 1491 | ret = btrfs_reserve_extent(root, cur_alloc_size, cur_alloc_size, |
432cd2a1 | 1492 | min_alloc_size, 0, alloc_hint, |
e570fd27 | 1493 | &ins, 1, 1); |
00361589 | 1494 | if (ret < 0) |
79787eaa | 1495 | goto out_unlock; |
a315e68f FM |
1496 | cur_alloc_size = ins.offset; |
1497 | extent_reserved = true; | |
d397712b | 1498 | |
771ed689 | 1499 | ram_size = ins.offset; |
6e26c442 | 1500 | em = create_io_em(inode, start, ins.offset, /* len */ |
6f9994db LB |
1501 | start, /* orig_start */ |
1502 | ins.objectid, /* block_start */ | |
1503 | ins.offset, /* block_len */ | |
1504 | ins.offset, /* orig_block_len */ | |
1505 | ram_size, /* ram_bytes */ | |
1506 | BTRFS_COMPRESS_NONE, /* compress_type */ | |
1af4a0aa | 1507 | BTRFS_ORDERED_REGULAR /* type */); |
090a127a SY |
1508 | if (IS_ERR(em)) { |
1509 | ret = PTR_ERR(em); | |
ace68bac | 1510 | goto out_reserve; |
090a127a | 1511 | } |
6f9994db | 1512 | free_extent_map(em); |
e6dcd2dc | 1513 | |
34bfaf15 CH |
1514 | ordered = btrfs_alloc_ordered_extent(inode, start, ram_size, |
1515 | ram_size, ins.objectid, cur_alloc_size, | |
1516 | 0, 1 << BTRFS_ORDERED_REGULAR, | |
1517 | BTRFS_COMPRESS_NONE); | |
1518 | if (IS_ERR(ordered)) { | |
1519 | ret = PTR_ERR(ordered); | |
d9f85963 | 1520 | goto out_drop_extent_cache; |
34bfaf15 | 1521 | } |
c8b97818 | 1522 | |
37f00a6d | 1523 | if (btrfs_is_data_reloc_root(root)) { |
34bfaf15 CH |
1524 | ret = btrfs_reloc_clone_csums(ordered); |
1525 | ||
4dbd80fb QW |
1526 | /* |
1527 | * Only drop cache here, and process as normal. | |
1528 | * | |
1529 | * We must not allow extent_clear_unlock_delalloc() | |
1530 | * at out_unlock label to free meta of this ordered | |
1531 | * extent, as its meta should be freed by | |
1532 | * btrfs_finish_ordered_io(). | |
1533 | * | |
1534 | * So we must continue until @start is increased to | |
1535 | * skip current ordered extent. | |
1536 | */ | |
00361589 | 1537 | if (ret) |
4c0c8cfc FM |
1538 | btrfs_drop_extent_map_range(inode, start, |
1539 | start + ram_size - 1, | |
1540 | false); | |
17d217fe | 1541 | } |
34bfaf15 | 1542 | btrfs_put_ordered_extent(ordered); |
17d217fe | 1543 | |
0b246afa | 1544 | btrfs_dec_block_group_reservations(fs_info, ins.objectid); |
9cfa3e34 | 1545 | |
f57ad937 QW |
1546 | /* |
1547 | * We're not doing compressed IO, don't unlock the first page | |
1548 | * (which the caller expects to stay locked), don't clear any | |
1549 | * dirty bits and don't set any writeback bits | |
8b62b72b | 1550 | * |
f57ad937 QW |
1551 | * Do set the Ordered (Private2) bit so we know this page was |
1552 | * properly setup for writepage. | |
c8b97818 | 1553 | */ |
ba9145ad | 1554 | page_ops = (keep_locked ? 0 : PAGE_UNLOCK); |
f57ad937 | 1555 | page_ops |= PAGE_SET_ORDERED; |
a791e35e | 1556 | |
6e26c442 | 1557 | extent_clear_unlock_delalloc(inode, start, start + ram_size - 1, |
74e9194a | 1558 | locked_page, |
c2790a2e | 1559 | EXTENT_LOCKED | EXTENT_DELALLOC, |
a315e68f | 1560 | page_ops); |
3752d22f AJ |
1561 | if (num_bytes < cur_alloc_size) |
1562 | num_bytes = 0; | |
4dbd80fb | 1563 | else |
3752d22f | 1564 | num_bytes -= cur_alloc_size; |
c59f8951 CM |
1565 | alloc_hint = ins.objectid + ins.offset; |
1566 | start += cur_alloc_size; | |
a315e68f | 1567 | extent_reserved = false; |
4dbd80fb QW |
1568 | |
1569 | /* | |
1570 | * btrfs_reloc_clone_csums() error, since start is increased | |
1571 | * extent_clear_unlock_delalloc() at out_unlock label won't | |
1572 | * free metadata of current ordered extent, we're OK to exit. | |
1573 | */ | |
1574 | if (ret) | |
1575 | goto out_unlock; | |
b888db2b | 1576 | } |
79787eaa | 1577 | out: |
be20aa9d | 1578 | return ret; |
b7d5b0a8 | 1579 | |
d9f85963 | 1580 | out_drop_extent_cache: |
4c0c8cfc | 1581 | btrfs_drop_extent_map_range(inode, start, start + ram_size - 1, false); |
ace68bac | 1582 | out_reserve: |
0b246afa | 1583 | btrfs_dec_block_group_reservations(fs_info, ins.objectid); |
2ff7e61e | 1584 | btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1); |
79787eaa | 1585 | out_unlock: |
898793d9 NA |
1586 | /* |
1587 | * If done_offset is non-NULL and ret == -EAGAIN, we expect the | |
1588 | * caller to write out the successfully allocated region and retry. | |
1589 | */ | |
1590 | if (done_offset && ret == -EAGAIN) { | |
1591 | if (orig_start < start) | |
1592 | *done_offset = start - 1; | |
1593 | else | |
1594 | *done_offset = start; | |
1595 | return ret; | |
1596 | } else if (ret == -EAGAIN) { | |
1597 | /* Convert to -ENOSPC since the caller cannot retry. */ | |
1598 | ret = -ENOSPC; | |
1599 | } | |
1600 | ||
9ce7466f NA |
1601 | /* |
1602 | * Now, we have three regions to clean up: | |
1603 | * | |
1604 | * |-------(1)----|---(2)---|-------------(3)----------| | |
1605 | * `- orig_start `- start `- start + cur_alloc_size `- end | |
1606 | * | |
1607 | * We process each region below. | |
1608 | */ | |
1609 | ||
a7e3b975 FM |
1610 | clear_bits = EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DELALLOC_NEW | |
1611 | EXTENT_DEFRAG | EXTENT_CLEAR_META_RESV; | |
6869b0a8 | 1612 | page_ops = PAGE_UNLOCK | PAGE_START_WRITEBACK | PAGE_END_WRITEBACK; |
9ce7466f | 1613 | |
a315e68f | 1614 | /* |
9ce7466f NA |
1615 | * For the range (1). We have already instantiated the ordered extents |
1616 | * for this region. They are cleaned up by | |
1617 | * btrfs_cleanup_ordered_extents() in e.g, | |
1618 | * btrfs_run_delalloc_range(). EXTENT_LOCKED | EXTENT_DELALLOC are | |
1619 | * already cleared in the above loop. And, EXTENT_DELALLOC_NEW | | |
1620 | * EXTENT_DEFRAG | EXTENT_CLEAR_META_RESV are handled by the cleanup | |
1621 | * function. | |
1622 | * | |
ba9145ad | 1623 | * However, in case of @keep_locked, we still need to unlock the pages |
9ce7466f NA |
1624 | * (except @locked_page) to ensure all the pages are unlocked. |
1625 | */ | |
ba9145ad | 1626 | if (keep_locked && orig_start < start) { |
71aa147b NA |
1627 | if (!locked_page) |
1628 | mapping_set_error(inode->vfs_inode.i_mapping, ret); | |
9ce7466f NA |
1629 | extent_clear_unlock_delalloc(inode, orig_start, start - 1, |
1630 | locked_page, 0, page_ops); | |
71aa147b | 1631 | } |
9ce7466f | 1632 | |
a315e68f | 1633 | /* |
9ce7466f NA |
1634 | * For the range (2). If we reserved an extent for our delalloc range |
1635 | * (or a subrange) and failed to create the respective ordered extent, | |
1636 | * then it means that when we reserved the extent we decremented the | |
1637 | * extent's size from the data space_info's bytes_may_use counter and | |
1638 | * incremented the space_info's bytes_reserved counter by the same | |
1639 | * amount. We must make sure extent_clear_unlock_delalloc() does not try | |
1640 | * to decrement again the data space_info's bytes_may_use counter, | |
1641 | * therefore we do not pass it the flag EXTENT_CLEAR_DATA_RESV. | |
a315e68f FM |
1642 | */ |
1643 | if (extent_reserved) { | |
6e26c442 | 1644 | extent_clear_unlock_delalloc(inode, start, |
e2c8e92d | 1645 | start + cur_alloc_size - 1, |
a315e68f FM |
1646 | locked_page, |
1647 | clear_bits, | |
1648 | page_ops); | |
1649 | start += cur_alloc_size; | |
a315e68f | 1650 | } |
9ce7466f NA |
1651 | |
1652 | /* | |
1653 | * For the range (3). We never touched the region. In addition to the | |
1654 | * clear_bits above, we add EXTENT_CLEAR_DATA_RESV to release the data | |
1655 | * space_info's bytes_may_use counter, reserved in | |
1656 | * btrfs_check_data_free_space(). | |
1657 | */ | |
12b2d64e CH |
1658 | if (start < end) { |
1659 | clear_bits |= EXTENT_CLEAR_DATA_RESV; | |
1660 | extent_clear_unlock_delalloc(inode, start, end, locked_page, | |
1661 | clear_bits, page_ops); | |
1662 | } | |
aaafa1eb | 1663 | return ret; |
771ed689 | 1664 | } |
c8b97818 | 1665 | |
771ed689 CM |
1666 | /* |
1667 | * work queue call back to started compression on a file and pages | |
1668 | */ | |
1669 | static noinline void async_cow_start(struct btrfs_work *work) | |
1670 | { | |
b5326271 | 1671 | struct async_chunk *async_chunk; |
ac3e9933 | 1672 | int compressed_extents; |
771ed689 | 1673 | |
b5326271 | 1674 | async_chunk = container_of(work, struct async_chunk, work); |
771ed689 | 1675 | |
ac3e9933 NB |
1676 | compressed_extents = compress_file_range(async_chunk); |
1677 | if (compressed_extents == 0) { | |
b5326271 NB |
1678 | btrfs_add_delayed_iput(async_chunk->inode); |
1679 | async_chunk->inode = NULL; | |
8180ef88 | 1680 | } |
771ed689 CM |
1681 | } |
1682 | ||
1683 | /* | |
1684 | * work queue call back to submit previously compressed pages | |
1685 | */ | |
1686 | static noinline void async_cow_submit(struct btrfs_work *work) | |
1687 | { | |
c5a68aec NB |
1688 | struct async_chunk *async_chunk = container_of(work, struct async_chunk, |
1689 | work); | |
1690 | struct btrfs_fs_info *fs_info = btrfs_work_owner(work); | |
771ed689 CM |
1691 | unsigned long nr_pages; |
1692 | ||
b5326271 | 1693 | nr_pages = (async_chunk->end - async_chunk->start + PAGE_SIZE) >> |
09cbfeaf | 1694 | PAGE_SHIFT; |
771ed689 | 1695 | |
4546d178 | 1696 | /* |
b5326271 | 1697 | * ->inode could be NULL if async_chunk_start has failed to compress, |
4546d178 NB |
1698 | * in which case we don't have anything to submit, yet we need to |
1699 | * always adjust ->async_delalloc_pages as its paired with the init | |
bb7b05fe | 1700 | * happening in run_delalloc_compressed |
4546d178 | 1701 | */ |
b5326271 NB |
1702 | if (async_chunk->inode) |
1703 | submit_compressed_extents(async_chunk); | |
ac98141d JB |
1704 | |
1705 | /* atomic_sub_return implies a barrier */ | |
1706 | if (atomic_sub_return(nr_pages, &fs_info->async_delalloc_pages) < | |
1707 | 5 * SZ_1M) | |
1708 | cond_wake_up_nomb(&fs_info->async_submit_wait); | |
771ed689 | 1709 | } |
c8b97818 | 1710 | |
771ed689 CM |
1711 | static noinline void async_cow_free(struct btrfs_work *work) |
1712 | { | |
b5326271 | 1713 | struct async_chunk *async_chunk; |
9e895a8f | 1714 | struct async_cow *async_cow; |
97db1204 | 1715 | |
b5326271 NB |
1716 | async_chunk = container_of(work, struct async_chunk, work); |
1717 | if (async_chunk->inode) | |
1718 | btrfs_add_delayed_iput(async_chunk->inode); | |
ec39f769 CM |
1719 | if (async_chunk->blkcg_css) |
1720 | css_put(async_chunk->blkcg_css); | |
9e895a8f QW |
1721 | |
1722 | async_cow = async_chunk->async_cow; | |
1723 | if (atomic_dec_and_test(&async_cow->num_chunks)) | |
1724 | kvfree(async_cow); | |
771ed689 CM |
1725 | } |
1726 | ||
bb7b05fe CH |
1727 | static bool run_delalloc_compressed(struct btrfs_inode *inode, |
1728 | struct writeback_control *wbc, | |
1729 | struct page *locked_page, | |
1730 | u64 start, u64 end, int *page_started, | |
1731 | unsigned long *nr_written) | |
771ed689 | 1732 | { |
751b6431 | 1733 | struct btrfs_fs_info *fs_info = inode->root->fs_info; |
ec39f769 | 1734 | struct cgroup_subsys_state *blkcg_css = wbc_blkcg_css(wbc); |
97db1204 NB |
1735 | struct async_cow *ctx; |
1736 | struct async_chunk *async_chunk; | |
771ed689 | 1737 | unsigned long nr_pages; |
97db1204 NB |
1738 | u64 num_chunks = DIV_ROUND_UP(end - start, SZ_512K); |
1739 | int i; | |
b1c16ac9 | 1740 | unsigned nofs_flag; |
bf9486d6 | 1741 | const blk_opf_t write_flags = wbc_to_write_flags(wbc); |
771ed689 | 1742 | |
b1c16ac9 NB |
1743 | nofs_flag = memalloc_nofs_save(); |
1744 | ctx = kvmalloc(struct_size(ctx, chunks, num_chunks), GFP_KERNEL); | |
1745 | memalloc_nofs_restore(nofs_flag); | |
973fb26e CH |
1746 | if (!ctx) |
1747 | return false; | |
b1c16ac9 | 1748 | |
973fb26e CH |
1749 | unlock_extent(&inode->io_tree, start, end, NULL); |
1750 | set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT, &inode->runtime_flags); | |
97db1204 NB |
1751 | |
1752 | async_chunk = ctx->chunks; | |
1753 | atomic_set(&ctx->num_chunks, num_chunks); | |
1754 | ||
1755 | for (i = 0; i < num_chunks; i++) { | |
973fb26e | 1756 | u64 cur_end = min(end, start + SZ_512K - 1); |
771ed689 | 1757 | |
bd4691a0 NB |
1758 | /* |
1759 | * igrab is called higher up in the call chain, take only the | |
1760 | * lightweight reference for the callback lifetime | |
1761 | */ | |
751b6431 | 1762 | ihold(&inode->vfs_inode); |
9e895a8f | 1763 | async_chunk[i].async_cow = ctx; |
99a81a44 | 1764 | async_chunk[i].inode = inode; |
97db1204 NB |
1765 | async_chunk[i].start = start; |
1766 | async_chunk[i].end = cur_end; | |
97db1204 NB |
1767 | async_chunk[i].write_flags = write_flags; |
1768 | INIT_LIST_HEAD(&async_chunk[i].extents); | |
1769 | ||
1d53c9e6 CM |
1770 | /* |
1771 | * The locked_page comes all the way from writepage and its | |
1772 | * the original page we were actually given. As we spread | |
1773 | * this large delalloc region across multiple async_chunk | |
1774 | * structs, only the first struct needs a pointer to locked_page | |
1775 | * | |
1776 | * This way we don't need racey decisions about who is supposed | |
1777 | * to unlock it. | |
1778 | */ | |
1779 | if (locked_page) { | |
ec39f769 CM |
1780 | /* |
1781 | * Depending on the compressibility, the pages might or | |
1782 | * might not go through async. We want all of them to | |
1783 | * be accounted against wbc once. Let's do it here | |
1784 | * before the paths diverge. wbc accounting is used | |
1785 | * only for foreign writeback detection and doesn't | |
1786 | * need full accuracy. Just account the whole thing | |
1787 | * against the first page. | |
1788 | */ | |
1789 | wbc_account_cgroup_owner(wbc, locked_page, | |
1790 | cur_end - start); | |
1d53c9e6 CM |
1791 | async_chunk[i].locked_page = locked_page; |
1792 | locked_page = NULL; | |
1793 | } else { | |
1794 | async_chunk[i].locked_page = NULL; | |
1795 | } | |
1796 | ||
ec39f769 CM |
1797 | if (blkcg_css != blkcg_root_css) { |
1798 | css_get(blkcg_css); | |
1799 | async_chunk[i].blkcg_css = blkcg_css; | |
3480373e | 1800 | async_chunk[i].write_flags |= REQ_BTRFS_CGROUP_PUNT; |
ec39f769 CM |
1801 | } else { |
1802 | async_chunk[i].blkcg_css = NULL; | |
1803 | } | |
1804 | ||
a0cac0ec OS |
1805 | btrfs_init_work(&async_chunk[i].work, async_cow_start, |
1806 | async_cow_submit, async_cow_free); | |
771ed689 | 1807 | |
97db1204 | 1808 | nr_pages = DIV_ROUND_UP(cur_end - start, PAGE_SIZE); |
0b246afa | 1809 | atomic_add(nr_pages, &fs_info->async_delalloc_pages); |
771ed689 | 1810 | |
97db1204 | 1811 | btrfs_queue_work(fs_info->delalloc_workers, &async_chunk[i].work); |
771ed689 | 1812 | |
771ed689 CM |
1813 | *nr_written += nr_pages; |
1814 | start = cur_end + 1; | |
1815 | } | |
1816 | *page_started = 1; | |
973fb26e | 1817 | return true; |
be20aa9d CM |
1818 | } |
1819 | ||
42c01100 NA |
1820 | static noinline int run_delalloc_zoned(struct btrfs_inode *inode, |
1821 | struct page *locked_page, u64 start, | |
1822 | u64 end, int *page_started, | |
7027f871 CH |
1823 | unsigned long *nr_written, |
1824 | struct writeback_control *wbc) | |
42c01100 | 1825 | { |
898793d9 | 1826 | u64 done_offset = end; |
42c01100 | 1827 | int ret; |
898793d9 | 1828 | bool locked_page_done = false; |
42c01100 | 1829 | |
898793d9 NA |
1830 | while (start <= end) { |
1831 | ret = cow_file_range(inode, locked_page, start, end, page_started, | |
53ffb30a | 1832 | nr_written, &done_offset, true, false); |
898793d9 NA |
1833 | if (ret && ret != -EAGAIN) |
1834 | return ret; | |
42c01100 | 1835 | |
898793d9 NA |
1836 | if (*page_started) { |
1837 | ASSERT(ret == 0); | |
1838 | return 0; | |
1839 | } | |
1840 | ||
1841 | if (ret == 0) | |
1842 | done_offset = end; | |
1843 | ||
2ce543f4 | 1844 | if (done_offset == start) { |
d5b81ced NA |
1845 | wait_on_bit_io(&inode->root->fs_info->flags, |
1846 | BTRFS_FS_NEED_ZONE_FINISH, | |
1847 | TASK_UNINTERRUPTIBLE); | |
2ce543f4 NA |
1848 | continue; |
1849 | } | |
898793d9 NA |
1850 | |
1851 | if (!locked_page_done) { | |
1852 | __set_page_dirty_nobuffers(locked_page); | |
1853 | account_page_redirty(locked_page); | |
1854 | } | |
1855 | locked_page_done = true; | |
7027f871 CH |
1856 | extent_write_locked_range(&inode->vfs_inode, start, done_offset, |
1857 | wbc); | |
898793d9 NA |
1858 | start = done_offset + 1; |
1859 | } | |
42c01100 | 1860 | |
42c01100 NA |
1861 | *page_started = 1; |
1862 | ||
1863 | return 0; | |
1864 | } | |
1865 | ||
2ff7e61e | 1866 | static noinline int csum_exist_in_range(struct btrfs_fs_info *fs_info, |
26ce9114 | 1867 | u64 bytenr, u64 num_bytes, bool nowait) |
17d217fe | 1868 | { |
fc28b25e | 1869 | struct btrfs_root *csum_root = btrfs_csum_root(fs_info, bytenr); |
17d217fe | 1870 | struct btrfs_ordered_sum *sums; |
fc28b25e | 1871 | int ret; |
17d217fe YZ |
1872 | LIST_HEAD(list); |
1873 | ||
97e38239 QW |
1874 | ret = btrfs_lookup_csums_list(csum_root, bytenr, bytenr + num_bytes - 1, |
1875 | &list, 0, nowait); | |
17d217fe YZ |
1876 | if (ret == 0 && list_empty(&list)) |
1877 | return 0; | |
1878 | ||
1879 | while (!list_empty(&list)) { | |
1880 | sums = list_entry(list.next, struct btrfs_ordered_sum, list); | |
1881 | list_del(&sums->list); | |
1882 | kfree(sums); | |
1883 | } | |
58113753 LB |
1884 | if (ret < 0) |
1885 | return ret; | |
17d217fe YZ |
1886 | return 1; |
1887 | } | |
1888 | ||
8ba96f3d | 1889 | static int fallback_to_cow(struct btrfs_inode *inode, struct page *locked_page, |
53ffb30a | 1890 | const u64 start, const u64 end) |
467dc47e | 1891 | { |
8ba96f3d | 1892 | const bool is_space_ino = btrfs_is_free_space_inode(inode); |
37f00a6d | 1893 | const bool is_reloc_ino = btrfs_is_data_reloc_root(inode->root); |
2166e5ed | 1894 | const u64 range_bytes = end + 1 - start; |
8ba96f3d | 1895 | struct extent_io_tree *io_tree = &inode->io_tree; |
53ffb30a CH |
1896 | int page_started = 0; |
1897 | unsigned long nr_written; | |
467dc47e FM |
1898 | u64 range_start = start; |
1899 | u64 count; | |
53ffb30a | 1900 | int ret; |
467dc47e FM |
1901 | |
1902 | /* | |
1903 | * If EXTENT_NORESERVE is set it means that when the buffered write was | |
1904 | * made we had not enough available data space and therefore we did not | |
1905 | * reserve data space for it, since we though we could do NOCOW for the | |
1906 | * respective file range (either there is prealloc extent or the inode | |
1907 | * has the NOCOW bit set). | |
1908 | * | |
1909 | * However when we need to fallback to COW mode (because for example the | |
1910 | * block group for the corresponding extent was turned to RO mode by a | |
1911 | * scrub or relocation) we need to do the following: | |
1912 | * | |
1913 | * 1) We increment the bytes_may_use counter of the data space info. | |
1914 | * If COW succeeds, it allocates a new data extent and after doing | |
1915 | * that it decrements the space info's bytes_may_use counter and | |
1916 | * increments its bytes_reserved counter by the same amount (we do | |
1917 | * this at btrfs_add_reserved_bytes()). So we need to increment the | |
1918 | * bytes_may_use counter to compensate (when space is reserved at | |
1919 | * buffered write time, the bytes_may_use counter is incremented); | |
1920 | * | |
1921 | * 2) We clear the EXTENT_NORESERVE bit from the range. We do this so | |
1922 | * that if the COW path fails for any reason, it decrements (through | |
1923 | * extent_clear_unlock_delalloc()) the bytes_may_use counter of the | |
1924 | * data space info, which we incremented in the step above. | |
2166e5ed FM |
1925 | * |
1926 | * If we need to fallback to cow and the inode corresponds to a free | |
6bd335b4 FM |
1927 | * space cache inode or an inode of the data relocation tree, we must |
1928 | * also increment bytes_may_use of the data space_info for the same | |
1929 | * reason. Space caches and relocated data extents always get a prealloc | |
2166e5ed | 1930 | * extent for them, however scrub or balance may have set the block |
6bd335b4 FM |
1931 | * group that contains that extent to RO mode and therefore force COW |
1932 | * when starting writeback. | |
467dc47e | 1933 | */ |
2166e5ed | 1934 | count = count_range_bits(io_tree, &range_start, end, range_bytes, |
8c6e53a7 | 1935 | EXTENT_NORESERVE, 0, NULL); |
6bd335b4 FM |
1936 | if (count > 0 || is_space_ino || is_reloc_ino) { |
1937 | u64 bytes = count; | |
8ba96f3d | 1938 | struct btrfs_fs_info *fs_info = inode->root->fs_info; |
467dc47e FM |
1939 | struct btrfs_space_info *sinfo = fs_info->data_sinfo; |
1940 | ||
6bd335b4 FM |
1941 | if (is_space_ino || is_reloc_ino) |
1942 | bytes = range_bytes; | |
1943 | ||
467dc47e | 1944 | spin_lock(&sinfo->lock); |
2166e5ed | 1945 | btrfs_space_info_update_bytes_may_use(fs_info, sinfo, bytes); |
467dc47e FM |
1946 | spin_unlock(&sinfo->lock); |
1947 | ||
2166e5ed FM |
1948 | if (count > 0) |
1949 | clear_extent_bit(io_tree, start, end, EXTENT_NORESERVE, | |
bd015294 | 1950 | NULL); |
467dc47e FM |
1951 | } |
1952 | ||
53ffb30a CH |
1953 | /* |
1954 | * Don't try to create inline extents, as a mix of inline extent that | |
1955 | * is written out and unlocked directly and a normal NOCOW extent | |
1956 | * doesn't work. | |
1957 | */ | |
1958 | ret = cow_file_range(inode, locked_page, start, end, &page_started, | |
1959 | &nr_written, NULL, false, true); | |
1960 | ASSERT(!page_started); | |
1961 | return ret; | |
467dc47e FM |
1962 | } |
1963 | ||
619104ba FM |
1964 | struct can_nocow_file_extent_args { |
1965 | /* Input fields. */ | |
1966 | ||
1967 | /* Start file offset of the range we want to NOCOW. */ | |
1968 | u64 start; | |
1969 | /* End file offset (inclusive) of the range we want to NOCOW. */ | |
1970 | u64 end; | |
1971 | bool writeback_path; | |
1972 | bool strict; | |
1973 | /* | |
1974 | * Free the path passed to can_nocow_file_extent() once it's not needed | |
1975 | * anymore. | |
1976 | */ | |
1977 | bool free_path; | |
1978 | ||
1979 | /* Output fields. Only set when can_nocow_file_extent() returns 1. */ | |
1980 | ||
1981 | u64 disk_bytenr; | |
1982 | u64 disk_num_bytes; | |
1983 | u64 extent_offset; | |
1984 | /* Number of bytes that can be written to in NOCOW mode. */ | |
1985 | u64 num_bytes; | |
1986 | }; | |
1987 | ||
1988 | /* | |
1989 | * Check if we can NOCOW the file extent that the path points to. | |
1990 | * This function may return with the path released, so the caller should check | |
1991 | * if path->nodes[0] is NULL or not if it needs to use the path afterwards. | |
1992 | * | |
1993 | * Returns: < 0 on error | |
1994 | * 0 if we can not NOCOW | |
1995 | * 1 if we can NOCOW | |
1996 | */ | |
1997 | static int can_nocow_file_extent(struct btrfs_path *path, | |
1998 | struct btrfs_key *key, | |
1999 | struct btrfs_inode *inode, | |
2000 | struct can_nocow_file_extent_args *args) | |
2001 | { | |
2002 | const bool is_freespace_inode = btrfs_is_free_space_inode(inode); | |
2003 | struct extent_buffer *leaf = path->nodes[0]; | |
2004 | struct btrfs_root *root = inode->root; | |
2005 | struct btrfs_file_extent_item *fi; | |
2006 | u64 extent_end; | |
2007 | u8 extent_type; | |
2008 | int can_nocow = 0; | |
2009 | int ret = 0; | |
26ce9114 | 2010 | bool nowait = path->nowait; |
619104ba FM |
2011 | |
2012 | fi = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_file_extent_item); | |
2013 | extent_type = btrfs_file_extent_type(leaf, fi); | |
2014 | ||
2015 | if (extent_type == BTRFS_FILE_EXTENT_INLINE) | |
2016 | goto out; | |
2017 | ||
2018 | /* Can't access these fields unless we know it's not an inline extent. */ | |
2019 | args->disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); | |
2020 | args->disk_num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi); | |
2021 | args->extent_offset = btrfs_file_extent_offset(leaf, fi); | |
2022 | ||
2023 | if (!(inode->flags & BTRFS_INODE_NODATACOW) && | |
2024 | extent_type == BTRFS_FILE_EXTENT_REG) | |
2025 | goto out; | |
2026 | ||
2027 | /* | |
2028 | * If the extent was created before the generation where the last snapshot | |
2029 | * for its subvolume was created, then this implies the extent is shared, | |
2030 | * hence we must COW. | |
2031 | */ | |
a7bb6bd4 | 2032 | if (!args->strict && |
619104ba FM |
2033 | btrfs_file_extent_generation(leaf, fi) <= |
2034 | btrfs_root_last_snapshot(&root->root_item)) | |
2035 | goto out; | |
2036 | ||
2037 | /* An explicit hole, must COW. */ | |
2038 | if (args->disk_bytenr == 0) | |
2039 | goto out; | |
2040 | ||
2041 | /* Compressed/encrypted/encoded extents must be COWed. */ | |
2042 | if (btrfs_file_extent_compression(leaf, fi) || | |
2043 | btrfs_file_extent_encryption(leaf, fi) || | |
2044 | btrfs_file_extent_other_encoding(leaf, fi)) | |
2045 | goto out; | |
2046 | ||
2047 | extent_end = btrfs_file_extent_end(path); | |
2048 | ||
2049 | /* | |
2050 | * The following checks can be expensive, as they need to take other | |
2051 | * locks and do btree or rbtree searches, so release the path to avoid | |
2052 | * blocking other tasks for too long. | |
2053 | */ | |
2054 | btrfs_release_path(path); | |
2055 | ||
2056 | ret = btrfs_cross_ref_exist(root, btrfs_ino(inode), | |
2057 | key->offset - args->extent_offset, | |
deccae40 | 2058 | args->disk_bytenr, args->strict, path); |
619104ba FM |
2059 | WARN_ON_ONCE(ret > 0 && is_freespace_inode); |
2060 | if (ret != 0) | |
2061 | goto out; | |
2062 | ||
2063 | if (args->free_path) { | |
2064 | /* | |
2065 | * We don't need the path anymore, plus through the | |
2066 | * csum_exist_in_range() call below we will end up allocating | |
2067 | * another path. So free the path to avoid unnecessary extra | |
2068 | * memory usage. | |
2069 | */ | |
2070 | btrfs_free_path(path); | |
2071 | path = NULL; | |
2072 | } | |
2073 | ||
2074 | /* If there are pending snapshots for this root, we must COW. */ | |
2075 | if (args->writeback_path && !is_freespace_inode && | |
2076 | atomic_read(&root->snapshot_force_cow)) | |
2077 | goto out; | |
2078 | ||
2079 | args->disk_bytenr += args->extent_offset; | |
2080 | args->disk_bytenr += args->start - key->offset; | |
2081 | args->num_bytes = min(args->end + 1, extent_end) - args->start; | |
2082 | ||
2083 | /* | |
2084 | * Force COW if csums exist in the range. This ensures that csums for a | |
2085 | * given extent are either valid or do not exist. | |
2086 | */ | |
26ce9114 JB |
2087 | ret = csum_exist_in_range(root->fs_info, args->disk_bytenr, args->num_bytes, |
2088 | nowait); | |
619104ba FM |
2089 | WARN_ON_ONCE(ret > 0 && is_freespace_inode); |
2090 | if (ret != 0) | |
2091 | goto out; | |
2092 | ||
2093 | can_nocow = 1; | |
2094 | out: | |
2095 | if (args->free_path && path) | |
2096 | btrfs_free_path(path); | |
2097 | ||
2098 | return ret < 0 ? ret : can_nocow; | |
2099 | } | |
2100 | ||
d352ac68 CM |
2101 | /* |
2102 | * when nowcow writeback call back. This checks for snapshots or COW copies | |
2103 | * of the extents that exist in the file, and COWs the file as required. | |
2104 | * | |
2105 | * If no cow copies or snapshots exist, we write directly to the existing | |
2106 | * blocks on disk | |
2107 | */ | |
968322c8 | 2108 | static noinline int run_delalloc_nocow(struct btrfs_inode *inode, |
7f366cfe | 2109 | struct page *locked_page, |
53ffb30a | 2110 | const u64 start, const u64 end) |
be20aa9d | 2111 | { |
968322c8 NB |
2112 | struct btrfs_fs_info *fs_info = inode->root->fs_info; |
2113 | struct btrfs_root *root = inode->root; | |
be20aa9d | 2114 | struct btrfs_path *path; |
3e024846 NB |
2115 | u64 cow_start = (u64)-1; |
2116 | u64 cur_offset = start; | |
8ecebf4d | 2117 | int ret; |
3e024846 | 2118 | bool check_prev = true; |
968322c8 | 2119 | u64 ino = btrfs_ino(inode); |
2306e83e | 2120 | struct btrfs_block_group *bg; |
762bf098 | 2121 | bool nocow = false; |
619104ba | 2122 | struct can_nocow_file_extent_args nocow_args = { 0 }; |
be20aa9d CM |
2123 | |
2124 | path = btrfs_alloc_path(); | |
17ca04af | 2125 | if (!path) { |
968322c8 | 2126 | extent_clear_unlock_delalloc(inode, start, end, locked_page, |
c2790a2e | 2127 | EXTENT_LOCKED | EXTENT_DELALLOC | |
151a41bc JB |
2128 | EXTENT_DO_ACCOUNTING | |
2129 | EXTENT_DEFRAG, PAGE_UNLOCK | | |
6869b0a8 | 2130 | PAGE_START_WRITEBACK | |
c2790a2e | 2131 | PAGE_END_WRITEBACK); |
d8926bb3 | 2132 | return -ENOMEM; |
17ca04af | 2133 | } |
82d5902d | 2134 | |
619104ba FM |
2135 | nocow_args.end = end; |
2136 | nocow_args.writeback_path = true; | |
2137 | ||
80ff3856 | 2138 | while (1) { |
34bfaf15 | 2139 | struct btrfs_ordered_extent *ordered; |
3e024846 NB |
2140 | struct btrfs_key found_key; |
2141 | struct btrfs_file_extent_item *fi; | |
2142 | struct extent_buffer *leaf; | |
2143 | u64 extent_end; | |
3e024846 | 2144 | u64 ram_bytes; |
619104ba | 2145 | u64 nocow_end; |
3e024846 | 2146 | int extent_type; |
3daea5fd | 2147 | bool is_prealloc; |
762bf098 NB |
2148 | |
2149 | nocow = false; | |
3e024846 | 2150 | |
e4c3b2dc | 2151 | ret = btrfs_lookup_file_extent(NULL, root, path, ino, |
80ff3856 | 2152 | cur_offset, 0); |
d788a349 | 2153 | if (ret < 0) |
79787eaa | 2154 | goto error; |
a6bd9cd1 NB |
2155 | |
2156 | /* | |
2157 | * If there is no extent for our range when doing the initial | |
2158 | * search, then go back to the previous slot as it will be the | |
2159 | * one containing the search offset | |
2160 | */ | |
80ff3856 YZ |
2161 | if (ret > 0 && path->slots[0] > 0 && check_prev) { |
2162 | leaf = path->nodes[0]; | |
2163 | btrfs_item_key_to_cpu(leaf, &found_key, | |
2164 | path->slots[0] - 1); | |
33345d01 | 2165 | if (found_key.objectid == ino && |
80ff3856 YZ |
2166 | found_key.type == BTRFS_EXTENT_DATA_KEY) |
2167 | path->slots[0]--; | |
2168 | } | |
3e024846 | 2169 | check_prev = false; |
80ff3856 | 2170 | next_slot: |
a6bd9cd1 | 2171 | /* Go to next leaf if we have exhausted the current one */ |
80ff3856 YZ |
2172 | leaf = path->nodes[0]; |
2173 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
2174 | ret = btrfs_next_leaf(root, path); | |
e8916699 LB |
2175 | if (ret < 0) { |
2176 | if (cow_start != (u64)-1) | |
2177 | cur_offset = cow_start; | |
79787eaa | 2178 | goto error; |
e8916699 | 2179 | } |
80ff3856 YZ |
2180 | if (ret > 0) |
2181 | break; | |
2182 | leaf = path->nodes[0]; | |
2183 | } | |
be20aa9d | 2184 | |
80ff3856 YZ |
2185 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
2186 | ||
a6bd9cd1 | 2187 | /* Didn't find anything for our INO */ |
1d512cb7 FM |
2188 | if (found_key.objectid > ino) |
2189 | break; | |
a6bd9cd1 NB |
2190 | /* |
2191 | * Keep searching until we find an EXTENT_ITEM or there are no | |
2192 | * more extents for this inode | |
2193 | */ | |
1d512cb7 FM |
2194 | if (WARN_ON_ONCE(found_key.objectid < ino) || |
2195 | found_key.type < BTRFS_EXTENT_DATA_KEY) { | |
2196 | path->slots[0]++; | |
2197 | goto next_slot; | |
2198 | } | |
a6bd9cd1 NB |
2199 | |
2200 | /* Found key is not EXTENT_DATA_KEY or starts after req range */ | |
1d512cb7 | 2201 | if (found_key.type > BTRFS_EXTENT_DATA_KEY || |
80ff3856 YZ |
2202 | found_key.offset > end) |
2203 | break; | |
2204 | ||
a6bd9cd1 NB |
2205 | /* |
2206 | * If the found extent starts after requested offset, then | |
2207 | * adjust extent_end to be right before this extent begins | |
2208 | */ | |
80ff3856 YZ |
2209 | if (found_key.offset > cur_offset) { |
2210 | extent_end = found_key.offset; | |
e9061e21 | 2211 | extent_type = 0; |
80ff3856 YZ |
2212 | goto out_check; |
2213 | } | |
2214 | ||
a6bd9cd1 NB |
2215 | /* |
2216 | * Found extent which begins before our range and potentially | |
2217 | * intersect it | |
2218 | */ | |
80ff3856 YZ |
2219 | fi = btrfs_item_ptr(leaf, path->slots[0], |
2220 | struct btrfs_file_extent_item); | |
2221 | extent_type = btrfs_file_extent_type(leaf, fi); | |
619104ba FM |
2222 | /* If this is triggered then we have a memory corruption. */ |
2223 | ASSERT(extent_type < BTRFS_NR_FILE_EXTENT_TYPES); | |
2224 | if (WARN_ON(extent_type >= BTRFS_NR_FILE_EXTENT_TYPES)) { | |
2225 | ret = -EUCLEAN; | |
2226 | goto error; | |
2227 | } | |
cc95bef6 | 2228 | ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi); |
619104ba | 2229 | extent_end = btrfs_file_extent_end(path); |
c65ca98f | 2230 | |
619104ba FM |
2231 | /* |
2232 | * If the extent we got ends before our current offset, skip to | |
2233 | * the next extent. | |
2234 | */ | |
2235 | if (extent_end <= cur_offset) { | |
2236 | path->slots[0]++; | |
2237 | goto next_slot; | |
2238 | } | |
c65ca98f | 2239 | |
619104ba FM |
2240 | nocow_args.start = cur_offset; |
2241 | ret = can_nocow_file_extent(path, &found_key, inode, &nocow_args); | |
2242 | if (ret < 0) { | |
2243 | if (cow_start != (u64)-1) | |
2244 | cur_offset = cow_start; | |
2245 | goto error; | |
2246 | } else if (ret == 0) { | |
2247 | goto out_check; | |
2248 | } | |
58113753 | 2249 | |
619104ba | 2250 | ret = 0; |
2306e83e FM |
2251 | bg = btrfs_inc_nocow_writers(fs_info, nocow_args.disk_bytenr); |
2252 | if (bg) | |
3e024846 | 2253 | nocow = true; |
80ff3856 | 2254 | out_check: |
a6bd9cd1 NB |
2255 | /* |
2256 | * If nocow is false then record the beginning of the range | |
2257 | * that needs to be COWed | |
2258 | */ | |
80ff3856 YZ |
2259 | if (!nocow) { |
2260 | if (cow_start == (u64)-1) | |
2261 | cow_start = cur_offset; | |
2262 | cur_offset = extent_end; | |
2263 | if (cur_offset > end) | |
2264 | break; | |
c65ca98f FM |
2265 | if (!path->nodes[0]) |
2266 | continue; | |
80ff3856 YZ |
2267 | path->slots[0]++; |
2268 | goto next_slot; | |
7ea394f1 YZ |
2269 | } |
2270 | ||
a6bd9cd1 NB |
2271 | /* |
2272 | * COW range from cow_start to found_key.offset - 1. As the key | |
2273 | * will contain the beginning of the first extent that can be | |
2274 | * NOCOW, following one which needs to be COW'ed | |
2275 | */ | |
80ff3856 | 2276 | if (cow_start != (u64)-1) { |
968322c8 | 2277 | ret = fallback_to_cow(inode, locked_page, |
53ffb30a | 2278 | cow_start, found_key.offset - 1); |
230ed397 | 2279 | if (ret) |
79787eaa | 2280 | goto error; |
80ff3856 | 2281 | cow_start = (u64)-1; |
7ea394f1 | 2282 | } |
80ff3856 | 2283 | |
619104ba | 2284 | nocow_end = cur_offset + nocow_args.num_bytes - 1; |
3daea5fd CH |
2285 | is_prealloc = extent_type == BTRFS_FILE_EXTENT_PREALLOC; |
2286 | if (is_prealloc) { | |
619104ba | 2287 | u64 orig_start = found_key.offset - nocow_args.extent_offset; |
3e024846 | 2288 | struct extent_map *em; |
6f9994db | 2289 | |
619104ba | 2290 | em = create_io_em(inode, cur_offset, nocow_args.num_bytes, |
6f9994db | 2291 | orig_start, |
619104ba FM |
2292 | nocow_args.disk_bytenr, /* block_start */ |
2293 | nocow_args.num_bytes, /* block_len */ | |
2294 | nocow_args.disk_num_bytes, /* orig_block_len */ | |
6f9994db LB |
2295 | ram_bytes, BTRFS_COMPRESS_NONE, |
2296 | BTRFS_ORDERED_PREALLOC); | |
2297 | if (IS_ERR(em)) { | |
6f9994db LB |
2298 | ret = PTR_ERR(em); |
2299 | goto error; | |
d899e052 | 2300 | } |
6f9994db | 2301 | free_extent_map(em); |
3daea5fd CH |
2302 | } |
2303 | ||
34bfaf15 | 2304 | ordered = btrfs_alloc_ordered_extent(inode, cur_offset, |
3daea5fd CH |
2305 | nocow_args.num_bytes, nocow_args.num_bytes, |
2306 | nocow_args.disk_bytenr, nocow_args.num_bytes, 0, | |
2307 | is_prealloc | |
2308 | ? (1 << BTRFS_ORDERED_PREALLOC) | |
2309 | : (1 << BTRFS_ORDERED_NOCOW), | |
2310 | BTRFS_COMPRESS_NONE); | |
34bfaf15 | 2311 | if (IS_ERR(ordered)) { |
3daea5fd | 2312 | if (is_prealloc) { |
4c0c8cfc FM |
2313 | btrfs_drop_extent_map_range(inode, cur_offset, |
2314 | nocow_end, false); | |
762bf098 | 2315 | } |
34bfaf15 | 2316 | ret = PTR_ERR(ordered); |
3daea5fd | 2317 | goto error; |
d899e052 | 2318 | } |
80ff3856 | 2319 | |
2306e83e FM |
2320 | if (nocow) { |
2321 | btrfs_dec_nocow_writers(bg); | |
2322 | nocow = false; | |
2323 | } | |
771ed689 | 2324 | |
37f00a6d | 2325 | if (btrfs_is_data_reloc_root(root)) |
4dbd80fb QW |
2326 | /* |
2327 | * Error handled later, as we must prevent | |
2328 | * extent_clear_unlock_delalloc() in error handler | |
2329 | * from freeing metadata of created ordered extent. | |
2330 | */ | |
34bfaf15 CH |
2331 | ret = btrfs_reloc_clone_csums(ordered); |
2332 | btrfs_put_ordered_extent(ordered); | |
efa56464 | 2333 | |
619104ba | 2334 | extent_clear_unlock_delalloc(inode, cur_offset, nocow_end, |
c2790a2e | 2335 | locked_page, EXTENT_LOCKED | |
18513091 WX |
2336 | EXTENT_DELALLOC | |
2337 | EXTENT_CLEAR_DATA_RESV, | |
f57ad937 | 2338 | PAGE_UNLOCK | PAGE_SET_ORDERED); |
18513091 | 2339 | |
80ff3856 | 2340 | cur_offset = extent_end; |
4dbd80fb QW |
2341 | |
2342 | /* | |
2343 | * btrfs_reloc_clone_csums() error, now we're OK to call error | |
2344 | * handler, as metadata for created ordered extent will only | |
2345 | * be freed by btrfs_finish_ordered_io(). | |
2346 | */ | |
2347 | if (ret) | |
2348 | goto error; | |
80ff3856 YZ |
2349 | if (cur_offset > end) |
2350 | break; | |
be20aa9d | 2351 | } |
b3b4aa74 | 2352 | btrfs_release_path(path); |
80ff3856 | 2353 | |
506481b2 | 2354 | if (cur_offset <= end && cow_start == (u64)-1) |
80ff3856 | 2355 | cow_start = cur_offset; |
17ca04af | 2356 | |
80ff3856 | 2357 | if (cow_start != (u64)-1) { |
506481b2 | 2358 | cur_offset = end; |
53ffb30a | 2359 | ret = fallback_to_cow(inode, locked_page, cow_start, end); |
d788a349 | 2360 | if (ret) |
79787eaa | 2361 | goto error; |
80ff3856 YZ |
2362 | } |
2363 | ||
79787eaa | 2364 | error: |
762bf098 | 2365 | if (nocow) |
2306e83e | 2366 | btrfs_dec_nocow_writers(bg); |
762bf098 | 2367 | |
17ca04af | 2368 | if (ret && cur_offset < end) |
968322c8 | 2369 | extent_clear_unlock_delalloc(inode, cur_offset, end, |
c2790a2e | 2370 | locked_page, EXTENT_LOCKED | |
151a41bc JB |
2371 | EXTENT_DELALLOC | EXTENT_DEFRAG | |
2372 | EXTENT_DO_ACCOUNTING, PAGE_UNLOCK | | |
6869b0a8 | 2373 | PAGE_START_WRITEBACK | |
c2790a2e | 2374 | PAGE_END_WRITEBACK); |
7ea394f1 | 2375 | btrfs_free_path(path); |
79787eaa | 2376 | return ret; |
be20aa9d CM |
2377 | } |
2378 | ||
6e65ae76 | 2379 | static bool should_nocow(struct btrfs_inode *inode, u64 start, u64 end) |
47059d93 | 2380 | { |
6e65ae76 GR |
2381 | if (inode->flags & (BTRFS_INODE_NODATACOW | BTRFS_INODE_PREALLOC)) { |
2382 | if (inode->defrag_bytes && | |
2383 | test_range_bit(&inode->io_tree, start, end, EXTENT_DEFRAG, | |
2384 | 0, NULL)) | |
2385 | return false; | |
2386 | return true; | |
2387 | } | |
2388 | return false; | |
47059d93 WS |
2389 | } |
2390 | ||
d352ac68 | 2391 | /* |
5eaad97a NB |
2392 | * Function to process delayed allocation (create CoW) for ranges which are |
2393 | * being touched for the first time. | |
d352ac68 | 2394 | */ |
98456b9c | 2395 | int btrfs_run_delalloc_range(struct btrfs_inode *inode, struct page *locked_page, |
5eaad97a NB |
2396 | u64 start, u64 end, int *page_started, unsigned long *nr_written, |
2397 | struct writeback_control *wbc) | |
be20aa9d | 2398 | { |
973fb26e | 2399 | int ret = 0; |
42c01100 | 2400 | const bool zoned = btrfs_is_zoned(inode->root->fs_info); |
a2135011 | 2401 | |
2749f7ef QW |
2402 | /* |
2403 | * The range must cover part of the @locked_page, or the returned | |
2404 | * @page_started can confuse the caller. | |
2405 | */ | |
2406 | ASSERT(!(end <= page_offset(locked_page) || | |
2407 | start >= page_offset(locked_page) + PAGE_SIZE)); | |
2408 | ||
6e65ae76 | 2409 | if (should_nocow(inode, start, end)) { |
2adada88 JT |
2410 | /* |
2411 | * Normally on a zoned device we're only doing COW writes, but | |
2412 | * in case of relocation on a zoned filesystem we have taken | |
2413 | * precaution, that we're only writing sequentially. It's safe | |
2414 | * to use run_delalloc_nocow() here, like for regular | |
2415 | * preallocated inodes. | |
2416 | */ | |
9435be73 | 2417 | ASSERT(!zoned || btrfs_is_data_reloc_root(inode->root)); |
53ffb30a | 2418 | ret = run_delalloc_nocow(inode, locked_page, start, end); |
973fb26e | 2419 | goto out; |
7ddf5a42 | 2420 | } |
973fb26e CH |
2421 | |
2422 | if (btrfs_inode_can_compress(inode) && | |
2423 | inode_need_compress(inode, start, end) && | |
bb7b05fe CH |
2424 | run_delalloc_compressed(inode, wbc, locked_page, start, |
2425 | end, page_started, nr_written)) | |
973fb26e CH |
2426 | goto out; |
2427 | ||
2428 | if (zoned) | |
2429 | ret = run_delalloc_zoned(inode, locked_page, start, end, | |
7027f871 | 2430 | page_started, nr_written, wbc); |
973fb26e CH |
2431 | else |
2432 | ret = cow_file_range(inode, locked_page, start, end, | |
53ffb30a | 2433 | page_started, nr_written, NULL, false, false); |
973fb26e CH |
2434 | |
2435 | out: | |
7361b4ae | 2436 | ASSERT(ret <= 0); |
52427260 | 2437 | if (ret) |
98456b9c | 2438 | btrfs_cleanup_ordered_extents(inode, locked_page, start, |
d1051d6e | 2439 | end - start + 1); |
b888db2b CM |
2440 | return ret; |
2441 | } | |
2442 | ||
62798a49 | 2443 | void btrfs_split_delalloc_extent(struct btrfs_inode *inode, |
abbb55f4 | 2444 | struct extent_state *orig, u64 split) |
9ed74f2d | 2445 | { |
62798a49 | 2446 | struct btrfs_fs_info *fs_info = inode->root->fs_info; |
dcab6a3b JB |
2447 | u64 size; |
2448 | ||
0ca1f7ce | 2449 | /* not delalloc, ignore it */ |
9ed74f2d | 2450 | if (!(orig->state & EXTENT_DELALLOC)) |
1bf85046 | 2451 | return; |
9ed74f2d | 2452 | |
dcab6a3b | 2453 | size = orig->end - orig->start + 1; |
f7b12a62 | 2454 | if (size > fs_info->max_extent_size) { |
823bb20a | 2455 | u32 num_extents; |
dcab6a3b JB |
2456 | u64 new_size; |
2457 | ||
2458 | /* | |
5c848198 | 2459 | * See the explanation in btrfs_merge_delalloc_extent, the same |
ba117213 | 2460 | * applies here, just in reverse. |
dcab6a3b JB |
2461 | */ |
2462 | new_size = orig->end - split + 1; | |
7d7672bc | 2463 | num_extents = count_max_extents(fs_info, new_size); |
ba117213 | 2464 | new_size = split - orig->start; |
7d7672bc NA |
2465 | num_extents += count_max_extents(fs_info, new_size); |
2466 | if (count_max_extents(fs_info, size) >= num_extents) | |
dcab6a3b JB |
2467 | return; |
2468 | } | |
2469 | ||
62798a49 DS |
2470 | spin_lock(&inode->lock); |
2471 | btrfs_mod_outstanding_extents(inode, 1); | |
2472 | spin_unlock(&inode->lock); | |
9ed74f2d JB |
2473 | } |
2474 | ||
2475 | /* | |
5c848198 NB |
2476 | * Handle merged delayed allocation extents so we can keep track of new extents |
2477 | * that are just merged onto old extents, such as when we are doing sequential | |
2478 | * writes, so we can properly account for the metadata space we'll need. | |
9ed74f2d | 2479 | */ |
2454151c | 2480 | void btrfs_merge_delalloc_extent(struct btrfs_inode *inode, struct extent_state *new, |
5c848198 | 2481 | struct extent_state *other) |
9ed74f2d | 2482 | { |
2454151c | 2483 | struct btrfs_fs_info *fs_info = inode->root->fs_info; |
dcab6a3b | 2484 | u64 new_size, old_size; |
823bb20a | 2485 | u32 num_extents; |
dcab6a3b | 2486 | |
9ed74f2d JB |
2487 | /* not delalloc, ignore it */ |
2488 | if (!(other->state & EXTENT_DELALLOC)) | |
1bf85046 | 2489 | return; |
9ed74f2d | 2490 | |
8461a3de JB |
2491 | if (new->start > other->start) |
2492 | new_size = new->end - other->start + 1; | |
2493 | else | |
2494 | new_size = other->end - new->start + 1; | |
dcab6a3b JB |
2495 | |
2496 | /* we're not bigger than the max, unreserve the space and go */ | |
f7b12a62 | 2497 | if (new_size <= fs_info->max_extent_size) { |
2454151c DS |
2498 | spin_lock(&inode->lock); |
2499 | btrfs_mod_outstanding_extents(inode, -1); | |
2500 | spin_unlock(&inode->lock); | |
dcab6a3b JB |
2501 | return; |
2502 | } | |
2503 | ||
2504 | /* | |
ba117213 JB |
2505 | * We have to add up either side to figure out how many extents were |
2506 | * accounted for before we merged into one big extent. If the number of | |
2507 | * extents we accounted for is <= the amount we need for the new range | |
2508 | * then we can return, otherwise drop. Think of it like this | |
2509 | * | |
2510 | * [ 4k][MAX_SIZE] | |
2511 | * | |
2512 | * So we've grown the extent by a MAX_SIZE extent, this would mean we | |
2513 | * need 2 outstanding extents, on one side we have 1 and the other side | |
2514 | * we have 1 so they are == and we can return. But in this case | |
2515 | * | |
2516 | * [MAX_SIZE+4k][MAX_SIZE+4k] | |
2517 | * | |
2518 | * Each range on their own accounts for 2 extents, but merged together | |
2519 | * they are only 3 extents worth of accounting, so we need to drop in | |
2520 | * this case. | |
dcab6a3b | 2521 | */ |
ba117213 | 2522 | old_size = other->end - other->start + 1; |
7d7672bc | 2523 | num_extents = count_max_extents(fs_info, old_size); |
ba117213 | 2524 | old_size = new->end - new->start + 1; |
7d7672bc NA |
2525 | num_extents += count_max_extents(fs_info, old_size); |
2526 | if (count_max_extents(fs_info, new_size) >= num_extents) | |
dcab6a3b JB |
2527 | return; |
2528 | ||
2454151c DS |
2529 | spin_lock(&inode->lock); |
2530 | btrfs_mod_outstanding_extents(inode, -1); | |
2531 | spin_unlock(&inode->lock); | |
9ed74f2d JB |
2532 | } |
2533 | ||
eb73c1b7 | 2534 | static void btrfs_add_delalloc_inodes(struct btrfs_root *root, |
82ca5a04 | 2535 | struct btrfs_inode *inode) |
eb73c1b7 | 2536 | { |
82ca5a04 | 2537 | struct btrfs_fs_info *fs_info = inode->root->fs_info; |
0b246afa | 2538 | |
eb73c1b7 | 2539 | spin_lock(&root->delalloc_lock); |
82ca5a04 DS |
2540 | if (list_empty(&inode->delalloc_inodes)) { |
2541 | list_add_tail(&inode->delalloc_inodes, &root->delalloc_inodes); | |
2542 | set_bit(BTRFS_INODE_IN_DELALLOC_LIST, &inode->runtime_flags); | |
eb73c1b7 MX |
2543 | root->nr_delalloc_inodes++; |
2544 | if (root->nr_delalloc_inodes == 1) { | |
0b246afa | 2545 | spin_lock(&fs_info->delalloc_root_lock); |
eb73c1b7 MX |
2546 | BUG_ON(!list_empty(&root->delalloc_root)); |
2547 | list_add_tail(&root->delalloc_root, | |
0b246afa JM |
2548 | &fs_info->delalloc_roots); |
2549 | spin_unlock(&fs_info->delalloc_root_lock); | |
eb73c1b7 MX |
2550 | } |
2551 | } | |
2552 | spin_unlock(&root->delalloc_lock); | |
2553 | } | |
2554 | ||
2b877331 NB |
2555 | void __btrfs_del_delalloc_inode(struct btrfs_root *root, |
2556 | struct btrfs_inode *inode) | |
eb73c1b7 | 2557 | { |
3ffbd68c | 2558 | struct btrfs_fs_info *fs_info = root->fs_info; |
0b246afa | 2559 | |
9e3e97f4 NB |
2560 | if (!list_empty(&inode->delalloc_inodes)) { |
2561 | list_del_init(&inode->delalloc_inodes); | |
eb73c1b7 | 2562 | clear_bit(BTRFS_INODE_IN_DELALLOC_LIST, |
9e3e97f4 | 2563 | &inode->runtime_flags); |
eb73c1b7 MX |
2564 | root->nr_delalloc_inodes--; |
2565 | if (!root->nr_delalloc_inodes) { | |
7c8a0d36 | 2566 | ASSERT(list_empty(&root->delalloc_inodes)); |
0b246afa | 2567 | spin_lock(&fs_info->delalloc_root_lock); |
eb73c1b7 MX |
2568 | BUG_ON(list_empty(&root->delalloc_root)); |
2569 | list_del_init(&root->delalloc_root); | |
0b246afa | 2570 | spin_unlock(&fs_info->delalloc_root_lock); |
eb73c1b7 MX |
2571 | } |
2572 | } | |
2b877331 NB |
2573 | } |
2574 | ||
2575 | static void btrfs_del_delalloc_inode(struct btrfs_root *root, | |
2576 | struct btrfs_inode *inode) | |
2577 | { | |
2578 | spin_lock(&root->delalloc_lock); | |
2579 | __btrfs_del_delalloc_inode(root, inode); | |
eb73c1b7 MX |
2580 | spin_unlock(&root->delalloc_lock); |
2581 | } | |
2582 | ||
d352ac68 | 2583 | /* |
e06a1fc9 NB |
2584 | * Properly track delayed allocation bytes in the inode and to maintain the |
2585 | * list of inodes that have pending delalloc work to be done. | |
d352ac68 | 2586 | */ |
4c5d166f | 2587 | void btrfs_set_delalloc_extent(struct btrfs_inode *inode, struct extent_state *state, |
6d92b304 | 2588 | u32 bits) |
291d673e | 2589 | { |
4c5d166f | 2590 | struct btrfs_fs_info *fs_info = inode->root->fs_info; |
0b246afa | 2591 | |
6d92b304 | 2592 | if ((bits & EXTENT_DEFRAG) && !(bits & EXTENT_DELALLOC)) |
47059d93 | 2593 | WARN_ON(1); |
75eff68e CM |
2594 | /* |
2595 | * set_bit and clear bit hooks normally require _irqsave/restore | |
27160b6b | 2596 | * but in this case, we are only testing for the DELALLOC |
75eff68e CM |
2597 | * bit, which is only set or cleared with irqs on |
2598 | */ | |
6d92b304 | 2599 | if (!(state->state & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) { |
4c5d166f | 2600 | struct btrfs_root *root = inode->root; |
0ca1f7ce | 2601 | u64 len = state->end + 1 - state->start; |
7d7672bc | 2602 | u32 num_extents = count_max_extents(fs_info, len); |
4c5d166f | 2603 | bool do_list = !btrfs_is_free_space_inode(inode); |
9ed74f2d | 2604 | |
4c5d166f DS |
2605 | spin_lock(&inode->lock); |
2606 | btrfs_mod_outstanding_extents(inode, num_extents); | |
2607 | spin_unlock(&inode->lock); | |
287a0ab9 | 2608 | |
6a3891c5 | 2609 | /* For sanity tests */ |
0b246afa | 2610 | if (btrfs_is_testing(fs_info)) |
6a3891c5 JB |
2611 | return; |
2612 | ||
104b4e51 NB |
2613 | percpu_counter_add_batch(&fs_info->delalloc_bytes, len, |
2614 | fs_info->delalloc_batch); | |
4c5d166f DS |
2615 | spin_lock(&inode->lock); |
2616 | inode->delalloc_bytes += len; | |
6d92b304 | 2617 | if (bits & EXTENT_DEFRAG) |
4c5d166f | 2618 | inode->defrag_bytes += len; |
df0af1a5 | 2619 | if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST, |
4c5d166f | 2620 | &inode->runtime_flags)) |
eb73c1b7 | 2621 | btrfs_add_delalloc_inodes(root, inode); |
4c5d166f | 2622 | spin_unlock(&inode->lock); |
291d673e | 2623 | } |
a7e3b975 FM |
2624 | |
2625 | if (!(state->state & EXTENT_DELALLOC_NEW) && | |
6d92b304 | 2626 | (bits & EXTENT_DELALLOC_NEW)) { |
4c5d166f DS |
2627 | spin_lock(&inode->lock); |
2628 | inode->new_delalloc_bytes += state->end + 1 - state->start; | |
2629 | spin_unlock(&inode->lock); | |
a7e3b975 | 2630 | } |
291d673e CM |
2631 | } |
2632 | ||
d352ac68 | 2633 | /* |
a36bb5f9 NB |
2634 | * Once a range is no longer delalloc this function ensures that proper |
2635 | * accounting happens. | |
d352ac68 | 2636 | */ |
bd54766e | 2637 | void btrfs_clear_delalloc_extent(struct btrfs_inode *inode, |
6d92b304 | 2638 | struct extent_state *state, u32 bits) |
291d673e | 2639 | { |
bd54766e | 2640 | struct btrfs_fs_info *fs_info = inode->root->fs_info; |
47059d93 | 2641 | u64 len = state->end + 1 - state->start; |
7d7672bc | 2642 | u32 num_extents = count_max_extents(fs_info, len); |
47059d93 | 2643 | |
6d92b304 | 2644 | if ((state->state & EXTENT_DEFRAG) && (bits & EXTENT_DEFRAG)) { |
4a4b964f | 2645 | spin_lock(&inode->lock); |
6fc0ef68 | 2646 | inode->defrag_bytes -= len; |
4a4b964f FM |
2647 | spin_unlock(&inode->lock); |
2648 | } | |
47059d93 | 2649 | |
75eff68e CM |
2650 | /* |
2651 | * set_bit and clear bit hooks normally require _irqsave/restore | |
27160b6b | 2652 | * but in this case, we are only testing for the DELALLOC |
75eff68e CM |
2653 | * bit, which is only set or cleared with irqs on |
2654 | */ | |
6d92b304 | 2655 | if ((state->state & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) { |
6fc0ef68 | 2656 | struct btrfs_root *root = inode->root; |
83eea1f1 | 2657 | bool do_list = !btrfs_is_free_space_inode(inode); |
bcbfce8a | 2658 | |
8b62f87b JB |
2659 | spin_lock(&inode->lock); |
2660 | btrfs_mod_outstanding_extents(inode, -num_extents); | |
2661 | spin_unlock(&inode->lock); | |
0ca1f7ce | 2662 | |
b6d08f06 JB |
2663 | /* |
2664 | * We don't reserve metadata space for space cache inodes so we | |
52042d8e | 2665 | * don't need to call delalloc_release_metadata if there is an |
b6d08f06 JB |
2666 | * error. |
2667 | */ | |
6d92b304 | 2668 | if (bits & EXTENT_CLEAR_META_RESV && |
0b246afa | 2669 | root != fs_info->tree_root) |
43b18595 | 2670 | btrfs_delalloc_release_metadata(inode, len, false); |
0ca1f7ce | 2671 | |
6a3891c5 | 2672 | /* For sanity tests. */ |
0b246afa | 2673 | if (btrfs_is_testing(fs_info)) |
6a3891c5 JB |
2674 | return; |
2675 | ||
37f00a6d | 2676 | if (!btrfs_is_data_reloc_root(root) && |
a315e68f | 2677 | do_list && !(state->state & EXTENT_NORESERVE) && |
6d92b304 | 2678 | (bits & EXTENT_CLEAR_DATA_RESV)) |
9db5d510 | 2679 | btrfs_free_reserved_data_space_noquota(fs_info, len); |
9ed74f2d | 2680 | |
104b4e51 NB |
2681 | percpu_counter_add_batch(&fs_info->delalloc_bytes, -len, |
2682 | fs_info->delalloc_batch); | |
6fc0ef68 NB |
2683 | spin_lock(&inode->lock); |
2684 | inode->delalloc_bytes -= len; | |
2685 | if (do_list && inode->delalloc_bytes == 0 && | |
df0af1a5 | 2686 | test_bit(BTRFS_INODE_IN_DELALLOC_LIST, |
9e3e97f4 | 2687 | &inode->runtime_flags)) |
eb73c1b7 | 2688 | btrfs_del_delalloc_inode(root, inode); |
6fc0ef68 | 2689 | spin_unlock(&inode->lock); |
291d673e | 2690 | } |
a7e3b975 FM |
2691 | |
2692 | if ((state->state & EXTENT_DELALLOC_NEW) && | |
6d92b304 | 2693 | (bits & EXTENT_DELALLOC_NEW)) { |
a7e3b975 FM |
2694 | spin_lock(&inode->lock); |
2695 | ASSERT(inode->new_delalloc_bytes >= len); | |
2696 | inode->new_delalloc_bytes -= len; | |
6d92b304 | 2697 | if (bits & EXTENT_ADD_INODE_BYTES) |
2766ff61 | 2698 | inode_add_bytes(&inode->vfs_inode, len); |
a7e3b975 FM |
2699 | spin_unlock(&inode->lock); |
2700 | } | |
291d673e CM |
2701 | } |
2702 | ||
71df088c CH |
2703 | static int btrfs_extract_ordered_extent(struct btrfs_bio *bbio, |
2704 | struct btrfs_ordered_extent *ordered) | |
d22002fd | 2705 | { |
69ccf3f4 CH |
2706 | u64 start = (u64)bbio->bio.bi_iter.bi_sector << SECTOR_SHIFT; |
2707 | u64 len = bbio->bio.bi_iter.bi_size; | |
b0307e28 | 2708 | struct btrfs_ordered_extent *new; |
ebdb44a0 | 2709 | int ret; |
d22002fd | 2710 | |
11d33ab6 | 2711 | /* Must always be called for the beginning of an ordered extent. */ |
7edd339c CH |
2712 | if (WARN_ON_ONCE(start != ordered->disk_bytenr)) |
2713 | return -EINVAL; | |
d22002fd | 2714 | |
11d33ab6 | 2715 | /* No need to split if the ordered extent covers the entire bio. */ |
ec63b84d CH |
2716 | if (ordered->disk_num_bytes == len) { |
2717 | refcount_inc(&ordered->refs); | |
2718 | bbio->ordered = ordered; | |
7edd339c | 2719 | return 0; |
ec63b84d | 2720 | } |
d22002fd | 2721 | |
f0f5329a BB |
2722 | /* |
2723 | * Don't split the extent_map for NOCOW extents, as we're writing into | |
2724 | * a pre-existing one. | |
2725 | */ | |
ebdb44a0 CH |
2726 | if (!test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags)) { |
2727 | ret = split_extent_map(bbio->inode, bbio->file_offset, | |
f000bc6f CH |
2728 | ordered->num_bytes, len, |
2729 | ordered->disk_bytenr); | |
ebdb44a0 CH |
2730 | if (ret) |
2731 | return ret; | |
2732 | } | |
f0f5329a | 2733 | |
b0307e28 CH |
2734 | new = btrfs_split_ordered_extent(ordered, len); |
2735 | if (IS_ERR(new)) | |
2736 | return PTR_ERR(new); | |
ec63b84d | 2737 | bbio->ordered = new; |
b0307e28 | 2738 | return 0; |
d22002fd NA |
2739 | } |
2740 | ||
d352ac68 CM |
2741 | /* |
2742 | * given a list of ordered sums record them in the inode. This happens | |
2743 | * at IO completion time based on sums calculated at bio submission time. | |
2744 | */ | |
510f85ed NB |
2745 | static int add_pending_csums(struct btrfs_trans_handle *trans, |
2746 | struct list_head *list) | |
e6dcd2dc | 2747 | { |
e6dcd2dc | 2748 | struct btrfs_ordered_sum *sum; |
fc28b25e | 2749 | struct btrfs_root *csum_root = NULL; |
ac01f26a | 2750 | int ret; |
e6dcd2dc | 2751 | |
c6e30871 | 2752 | list_for_each_entry(sum, list, list) { |
7c2871a2 | 2753 | trans->adding_csums = true; |
fc28b25e JB |
2754 | if (!csum_root) |
2755 | csum_root = btrfs_csum_root(trans->fs_info, | |
5cfe76f8 | 2756 | sum->logical); |
fc28b25e | 2757 | ret = btrfs_csum_file_blocks(trans, csum_root, sum); |
7c2871a2 | 2758 | trans->adding_csums = false; |
ac01f26a NB |
2759 | if (ret) |
2760 | return ret; | |
e6dcd2dc CM |
2761 | } |
2762 | return 0; | |
2763 | } | |
2764 | ||
c3347309 FM |
2765 | static int btrfs_find_new_delalloc_bytes(struct btrfs_inode *inode, |
2766 | const u64 start, | |
2767 | const u64 len, | |
2768 | struct extent_state **cached_state) | |
2769 | { | |
2770 | u64 search_start = start; | |
2771 | const u64 end = start + len - 1; | |
2772 | ||
2773 | while (search_start < end) { | |
2774 | const u64 search_len = end - search_start + 1; | |
2775 | struct extent_map *em; | |
2776 | u64 em_len; | |
2777 | int ret = 0; | |
2778 | ||
2779 | em = btrfs_get_extent(inode, NULL, 0, search_start, search_len); | |
2780 | if (IS_ERR(em)) | |
2781 | return PTR_ERR(em); | |
2782 | ||
2783 | if (em->block_start != EXTENT_MAP_HOLE) | |
2784 | goto next; | |
2785 | ||
2786 | em_len = em->len; | |
2787 | if (em->start < search_start) | |
2788 | em_len -= search_start - em->start; | |
2789 | if (em_len > search_len) | |
2790 | em_len = search_len; | |
2791 | ||
2792 | ret = set_extent_bit(&inode->io_tree, search_start, | |
2793 | search_start + em_len - 1, | |
1d126800 | 2794 | EXTENT_DELALLOC_NEW, cached_state); |
c3347309 FM |
2795 | next: |
2796 | search_start = extent_map_end(em); | |
2797 | free_extent_map(em); | |
2798 | if (ret) | |
2799 | return ret; | |
2800 | } | |
2801 | return 0; | |
2802 | } | |
2803 | ||
c2566f22 | 2804 | int btrfs_set_extent_delalloc(struct btrfs_inode *inode, u64 start, u64 end, |
e3b8a485 | 2805 | unsigned int extra_bits, |
330a5827 | 2806 | struct extent_state **cached_state) |
ea8c2819 | 2807 | { |
fdb1e121 | 2808 | WARN_ON(PAGE_ALIGNED(end)); |
c3347309 FM |
2809 | |
2810 | if (start >= i_size_read(&inode->vfs_inode) && | |
2811 | !(inode->flags & BTRFS_INODE_PREALLOC)) { | |
2812 | /* | |
2813 | * There can't be any extents following eof in this case so just | |
2814 | * set the delalloc new bit for the range directly. | |
2815 | */ | |
2816 | extra_bits |= EXTENT_DELALLOC_NEW; | |
2817 | } else { | |
2818 | int ret; | |
2819 | ||
2820 | ret = btrfs_find_new_delalloc_bytes(inode, start, | |
2821 | end + 1 - start, | |
2822 | cached_state); | |
2823 | if (ret) | |
2824 | return ret; | |
2825 | } | |
2826 | ||
66240ab1 | 2827 | return set_extent_bit(&inode->io_tree, start, end, |
1d126800 | 2828 | EXTENT_DELALLOC | extra_bits, cached_state); |
ea8c2819 CM |
2829 | } |
2830 | ||
d352ac68 | 2831 | /* see btrfs_writepage_start_hook for details on why this is required */ |
247e743c CM |
2832 | struct btrfs_writepage_fixup { |
2833 | struct page *page; | |
36eeaef5 | 2834 | struct btrfs_inode *inode; |
247e743c CM |
2835 | struct btrfs_work work; |
2836 | }; | |
2837 | ||
b2950863 | 2838 | static void btrfs_writepage_fixup_worker(struct btrfs_work *work) |
247e743c CM |
2839 | { |
2840 | struct btrfs_writepage_fixup *fixup; | |
2841 | struct btrfs_ordered_extent *ordered; | |
2ac55d41 | 2842 | struct extent_state *cached_state = NULL; |
364ecf36 | 2843 | struct extent_changeset *data_reserved = NULL; |
247e743c | 2844 | struct page *page; |
65d87f79 | 2845 | struct btrfs_inode *inode; |
247e743c CM |
2846 | u64 page_start; |
2847 | u64 page_end; | |
25f3c502 | 2848 | int ret = 0; |
f4b1363c | 2849 | bool free_delalloc_space = true; |
247e743c CM |
2850 | |
2851 | fixup = container_of(work, struct btrfs_writepage_fixup, work); | |
2852 | page = fixup->page; | |
36eeaef5 | 2853 | inode = fixup->inode; |
f4b1363c JB |
2854 | page_start = page_offset(page); |
2855 | page_end = page_offset(page) + PAGE_SIZE - 1; | |
2856 | ||
2857 | /* | |
2858 | * This is similar to page_mkwrite, we need to reserve the space before | |
2859 | * we take the page lock. | |
2860 | */ | |
65d87f79 NB |
2861 | ret = btrfs_delalloc_reserve_space(inode, &data_reserved, page_start, |
2862 | PAGE_SIZE); | |
4a096752 | 2863 | again: |
247e743c | 2864 | lock_page(page); |
25f3c502 CM |
2865 | |
2866 | /* | |
2867 | * Before we queued this fixup, we took a reference on the page. | |
2868 | * page->mapping may go NULL, but it shouldn't be moved to a different | |
2869 | * address space. | |
2870 | */ | |
f4b1363c JB |
2871 | if (!page->mapping || !PageDirty(page) || !PageChecked(page)) { |
2872 | /* | |
2873 | * Unfortunately this is a little tricky, either | |
2874 | * | |
2875 | * 1) We got here and our page had already been dealt with and | |
2876 | * we reserved our space, thus ret == 0, so we need to just | |
2877 | * drop our space reservation and bail. This can happen the | |
2878 | * first time we come into the fixup worker, or could happen | |
2879 | * while waiting for the ordered extent. | |
2880 | * 2) Our page was already dealt with, but we happened to get an | |
2881 | * ENOSPC above from the btrfs_delalloc_reserve_space. In | |
2882 | * this case we obviously don't have anything to release, but | |
2883 | * because the page was already dealt with we don't want to | |
2884 | * mark the page with an error, so make sure we're resetting | |
2885 | * ret to 0. This is why we have this check _before_ the ret | |
2886 | * check, because we do not want to have a surprise ENOSPC | |
2887 | * when the page was already properly dealt with. | |
2888 | */ | |
2889 | if (!ret) { | |
65d87f79 NB |
2890 | btrfs_delalloc_release_extents(inode, PAGE_SIZE); |
2891 | btrfs_delalloc_release_space(inode, data_reserved, | |
f4b1363c JB |
2892 | page_start, PAGE_SIZE, |
2893 | true); | |
2894 | } | |
2895 | ret = 0; | |
247e743c | 2896 | goto out_page; |
f4b1363c | 2897 | } |
247e743c | 2898 | |
25f3c502 | 2899 | /* |
f4b1363c JB |
2900 | * We can't mess with the page state unless it is locked, so now that |
2901 | * it is locked bail if we failed to make our space reservation. | |
25f3c502 | 2902 | */ |
f4b1363c JB |
2903 | if (ret) |
2904 | goto out_page; | |
247e743c | 2905 | |
570eb97b | 2906 | lock_extent(&inode->io_tree, page_start, page_end, &cached_state); |
4a096752 CM |
2907 | |
2908 | /* already ordered? We're done */ | |
f57ad937 | 2909 | if (PageOrdered(page)) |
f4b1363c | 2910 | goto out_reserved; |
4a096752 | 2911 | |
65d87f79 | 2912 | ordered = btrfs_lookup_ordered_range(inode, page_start, PAGE_SIZE); |
4a096752 | 2913 | if (ordered) { |
570eb97b JB |
2914 | unlock_extent(&inode->io_tree, page_start, page_end, |
2915 | &cached_state); | |
4a096752 | 2916 | unlock_page(page); |
36d45567 | 2917 | btrfs_start_ordered_extent(ordered); |
87826df0 | 2918 | btrfs_put_ordered_extent(ordered); |
4a096752 CM |
2919 | goto again; |
2920 | } | |
247e743c | 2921 | |
65d87f79 | 2922 | ret = btrfs_set_extent_delalloc(inode, page_start, page_end, 0, |
330a5827 | 2923 | &cached_state); |
25f3c502 | 2924 | if (ret) |
53687007 | 2925 | goto out_reserved; |
f3038ee3 | 2926 | |
25f3c502 CM |
2927 | /* |
2928 | * Everything went as planned, we're now the owner of a dirty page with | |
2929 | * delayed allocation bits set and space reserved for our COW | |
2930 | * destination. | |
2931 | * | |
2932 | * The page was dirty when we started, nothing should have cleaned it. | |
2933 | */ | |
2934 | BUG_ON(!PageDirty(page)); | |
f4b1363c | 2935 | free_delalloc_space = false; |
53687007 | 2936 | out_reserved: |
65d87f79 | 2937 | btrfs_delalloc_release_extents(inode, PAGE_SIZE); |
f4b1363c | 2938 | if (free_delalloc_space) |
65d87f79 NB |
2939 | btrfs_delalloc_release_space(inode, data_reserved, page_start, |
2940 | PAGE_SIZE, true); | |
570eb97b | 2941 | unlock_extent(&inode->io_tree, page_start, page_end, &cached_state); |
247e743c | 2942 | out_page: |
25f3c502 CM |
2943 | if (ret) { |
2944 | /* | |
2945 | * We hit ENOSPC or other errors. Update the mapping and page | |
2946 | * to reflect the errors and clean the page. | |
2947 | */ | |
2948 | mapping_set_error(page->mapping, ret); | |
2949 | end_extent_writepage(page, ret, page_start, page_end); | |
2950 | clear_page_dirty_for_io(page); | |
25f3c502 | 2951 | } |
e4f94347 | 2952 | btrfs_page_clear_checked(inode->root->fs_info, page, page_start, PAGE_SIZE); |
247e743c | 2953 | unlock_page(page); |
09cbfeaf | 2954 | put_page(page); |
b897abec | 2955 | kfree(fixup); |
364ecf36 | 2956 | extent_changeset_free(data_reserved); |
f4b1363c JB |
2957 | /* |
2958 | * As a precaution, do a delayed iput in case it would be the last iput | |
2959 | * that could need flushing space. Recursing back to fixup worker would | |
2960 | * deadlock. | |
2961 | */ | |
e55cf7ca | 2962 | btrfs_add_delayed_iput(inode); |
247e743c CM |
2963 | } |
2964 | ||
2965 | /* | |
2966 | * There are a few paths in the higher layers of the kernel that directly | |
2967 | * set the page dirty bit without asking the filesystem if it is a | |
2968 | * good idea. This causes problems because we want to make sure COW | |
2969 | * properly happens and the data=ordered rules are followed. | |
2970 | * | |
c8b97818 | 2971 | * In our case any range that doesn't have the ORDERED bit set |
247e743c CM |
2972 | * hasn't been properly setup for IO. We kick off an async process |
2973 | * to fix it up. The async helper will wait for ordered extents, set | |
2974 | * the delalloc bit and make it safe to write the page. | |
2975 | */ | |
a129ffb8 | 2976 | int btrfs_writepage_cow_fixup(struct page *page) |
247e743c CM |
2977 | { |
2978 | struct inode *inode = page->mapping->host; | |
0b246afa | 2979 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
247e743c | 2980 | struct btrfs_writepage_fixup *fixup; |
247e743c | 2981 | |
f57ad937 QW |
2982 | /* This page has ordered extent covering it already */ |
2983 | if (PageOrdered(page)) | |
247e743c CM |
2984 | return 0; |
2985 | ||
25f3c502 CM |
2986 | /* |
2987 | * PageChecked is set below when we create a fixup worker for this page, | |
2988 | * don't try to create another one if we're already PageChecked() | |
2989 | * | |
2990 | * The extent_io writepage code will redirty the page if we send back | |
2991 | * EAGAIN. | |
2992 | */ | |
247e743c CM |
2993 | if (PageChecked(page)) |
2994 | return -EAGAIN; | |
2995 | ||
2996 | fixup = kzalloc(sizeof(*fixup), GFP_NOFS); | |
2997 | if (!fixup) | |
2998 | return -EAGAIN; | |
f421950f | 2999 | |
f4b1363c JB |
3000 | /* |
3001 | * We are already holding a reference to this inode from | |
3002 | * write_cache_pages. We need to hold it because the space reservation | |
3003 | * takes place outside of the page lock, and we can't trust | |
3004 | * page->mapping outside of the page lock. | |
3005 | */ | |
3006 | ihold(inode); | |
e4f94347 | 3007 | btrfs_page_set_checked(fs_info, page, page_offset(page), PAGE_SIZE); |
09cbfeaf | 3008 | get_page(page); |
a0cac0ec | 3009 | btrfs_init_work(&fixup->work, btrfs_writepage_fixup_worker, NULL, NULL); |
247e743c | 3010 | fixup->page = page; |
36eeaef5 | 3011 | fixup->inode = BTRFS_I(inode); |
0b246afa | 3012 | btrfs_queue_work(fs_info->fixup_workers, &fixup->work); |
25f3c502 CM |
3013 | |
3014 | return -EAGAIN; | |
247e743c CM |
3015 | } |
3016 | ||
d899e052 | 3017 | static int insert_reserved_file_extent(struct btrfs_trans_handle *trans, |
c553f94d | 3018 | struct btrfs_inode *inode, u64 file_pos, |
9729f10a | 3019 | struct btrfs_file_extent_item *stack_fi, |
2766ff61 | 3020 | const bool update_inode_bytes, |
9729f10a | 3021 | u64 qgroup_reserved) |
d899e052 | 3022 | { |
c553f94d | 3023 | struct btrfs_root *root = inode->root; |
2766ff61 | 3024 | const u64 sectorsize = root->fs_info->sectorsize; |
d899e052 YZ |
3025 | struct btrfs_path *path; |
3026 | struct extent_buffer *leaf; | |
3027 | struct btrfs_key ins; | |
203f44c5 QW |
3028 | u64 disk_num_bytes = btrfs_stack_file_extent_disk_num_bytes(stack_fi); |
3029 | u64 disk_bytenr = btrfs_stack_file_extent_disk_bytenr(stack_fi); | |
cb36a9bb | 3030 | u64 offset = btrfs_stack_file_extent_offset(stack_fi); |
203f44c5 QW |
3031 | u64 num_bytes = btrfs_stack_file_extent_num_bytes(stack_fi); |
3032 | u64 ram_bytes = btrfs_stack_file_extent_ram_bytes(stack_fi); | |
5893dfb9 | 3033 | struct btrfs_drop_extents_args drop_args = { 0 }; |
d899e052 YZ |
3034 | int ret; |
3035 | ||
3036 | path = btrfs_alloc_path(); | |
d8926bb3 MF |
3037 | if (!path) |
3038 | return -ENOMEM; | |
d899e052 | 3039 | |
a1ed835e CM |
3040 | /* |
3041 | * we may be replacing one extent in the tree with another. | |
3042 | * The new extent is pinned in the extent map, and we don't want | |
3043 | * to drop it from the cache until it is completely in the btree. | |
3044 | * | |
3045 | * So, tell btrfs_drop_extents to leave this extent in the cache. | |
3046 | * the caller is expected to unpin it and allow it to be merged | |
3047 | * with the others. | |
3048 | */ | |
5893dfb9 FM |
3049 | drop_args.path = path; |
3050 | drop_args.start = file_pos; | |
3051 | drop_args.end = file_pos + num_bytes; | |
3052 | drop_args.replace_extent = true; | |
3053 | drop_args.extent_item_size = sizeof(*stack_fi); | |
3054 | ret = btrfs_drop_extents(trans, root, inode, &drop_args); | |
79787eaa JM |
3055 | if (ret) |
3056 | goto out; | |
d899e052 | 3057 | |
5893dfb9 | 3058 | if (!drop_args.extent_inserted) { |
c553f94d | 3059 | ins.objectid = btrfs_ino(inode); |
1acae57b FDBM |
3060 | ins.offset = file_pos; |
3061 | ins.type = BTRFS_EXTENT_DATA_KEY; | |
3062 | ||
1acae57b | 3063 | ret = btrfs_insert_empty_item(trans, root, path, &ins, |
203f44c5 | 3064 | sizeof(*stack_fi)); |
1acae57b FDBM |
3065 | if (ret) |
3066 | goto out; | |
3067 | } | |
d899e052 | 3068 | leaf = path->nodes[0]; |
203f44c5 QW |
3069 | btrfs_set_stack_file_extent_generation(stack_fi, trans->transid); |
3070 | write_extent_buffer(leaf, stack_fi, | |
3071 | btrfs_item_ptr_offset(leaf, path->slots[0]), | |
3072 | sizeof(struct btrfs_file_extent_item)); | |
b9473439 | 3073 | |
d899e052 | 3074 | btrfs_mark_buffer_dirty(leaf); |
ce195332 | 3075 | btrfs_release_path(path); |
d899e052 | 3076 | |
2766ff61 FM |
3077 | /* |
3078 | * If we dropped an inline extent here, we know the range where it is | |
3079 | * was not marked with the EXTENT_DELALLOC_NEW bit, so we update the | |
1a9fd417 | 3080 | * number of bytes only for that range containing the inline extent. |
2766ff61 FM |
3081 | * The remaining of the range will be processed when clearning the |
3082 | * EXTENT_DELALLOC_BIT bit through the ordered extent completion. | |
3083 | */ | |
3084 | if (file_pos == 0 && !IS_ALIGNED(drop_args.bytes_found, sectorsize)) { | |
3085 | u64 inline_size = round_down(drop_args.bytes_found, sectorsize); | |
3086 | ||
3087 | inline_size = drop_args.bytes_found - inline_size; | |
3088 | btrfs_update_inode_bytes(inode, sectorsize, inline_size); | |
3089 | drop_args.bytes_found -= inline_size; | |
3090 | num_bytes -= sectorsize; | |
3091 | } | |
3092 | ||
3093 | if (update_inode_bytes) | |
3094 | btrfs_update_inode_bytes(inode, num_bytes, drop_args.bytes_found); | |
d899e052 YZ |
3095 | |
3096 | ins.objectid = disk_bytenr; | |
3097 | ins.offset = disk_num_bytes; | |
3098 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
a12b877b | 3099 | |
c553f94d | 3100 | ret = btrfs_inode_set_file_extent_range(inode, file_pos, ram_bytes); |
9ddc959e JB |
3101 | if (ret) |
3102 | goto out; | |
3103 | ||
c553f94d | 3104 | ret = btrfs_alloc_reserved_file_extent(trans, root, btrfs_ino(inode), |
cb36a9bb OS |
3105 | file_pos - offset, |
3106 | qgroup_reserved, &ins); | |
79787eaa | 3107 | out: |
d899e052 | 3108 | btrfs_free_path(path); |
b9473439 | 3109 | |
79787eaa | 3110 | return ret; |
d899e052 YZ |
3111 | } |
3112 | ||
2ff7e61e | 3113 | static void btrfs_release_delalloc_bytes(struct btrfs_fs_info *fs_info, |
e570fd27 MX |
3114 | u64 start, u64 len) |
3115 | { | |
32da5386 | 3116 | struct btrfs_block_group *cache; |
e570fd27 | 3117 | |
0b246afa | 3118 | cache = btrfs_lookup_block_group(fs_info, start); |
e570fd27 MX |
3119 | ASSERT(cache); |
3120 | ||
3121 | spin_lock(&cache->lock); | |
3122 | cache->delalloc_bytes -= len; | |
3123 | spin_unlock(&cache->lock); | |
3124 | ||
3125 | btrfs_put_block_group(cache); | |
3126 | } | |
3127 | ||
203f44c5 | 3128 | static int insert_ordered_extent_file_extent(struct btrfs_trans_handle *trans, |
203f44c5 QW |
3129 | struct btrfs_ordered_extent *oe) |
3130 | { | |
3131 | struct btrfs_file_extent_item stack_fi; | |
2766ff61 | 3132 | bool update_inode_bytes; |
cb36a9bb OS |
3133 | u64 num_bytes = oe->num_bytes; |
3134 | u64 ram_bytes = oe->ram_bytes; | |
203f44c5 QW |
3135 | |
3136 | memset(&stack_fi, 0, sizeof(stack_fi)); | |
3137 | btrfs_set_stack_file_extent_type(&stack_fi, BTRFS_FILE_EXTENT_REG); | |
3138 | btrfs_set_stack_file_extent_disk_bytenr(&stack_fi, oe->disk_bytenr); | |
3139 | btrfs_set_stack_file_extent_disk_num_bytes(&stack_fi, | |
3140 | oe->disk_num_bytes); | |
cb36a9bb | 3141 | btrfs_set_stack_file_extent_offset(&stack_fi, oe->offset); |
c1867eb3 DS |
3142 | if (test_bit(BTRFS_ORDERED_TRUNCATED, &oe->flags)) { |
3143 | num_bytes = oe->truncated_len; | |
3144 | ram_bytes = num_bytes; | |
3145 | } | |
cb36a9bb OS |
3146 | btrfs_set_stack_file_extent_num_bytes(&stack_fi, num_bytes); |
3147 | btrfs_set_stack_file_extent_ram_bytes(&stack_fi, ram_bytes); | |
203f44c5 QW |
3148 | btrfs_set_stack_file_extent_compression(&stack_fi, oe->compress_type); |
3149 | /* Encryption and other encoding is reserved and all 0 */ | |
3150 | ||
2766ff61 FM |
3151 | /* |
3152 | * For delalloc, when completing an ordered extent we update the inode's | |
3153 | * bytes when clearing the range in the inode's io tree, so pass false | |
3154 | * as the argument 'update_inode_bytes' to insert_reserved_file_extent(), | |
3155 | * except if the ordered extent was truncated. | |
3156 | */ | |
3157 | update_inode_bytes = test_bit(BTRFS_ORDERED_DIRECT, &oe->flags) || | |
7c0c7269 | 3158 | test_bit(BTRFS_ORDERED_ENCODED, &oe->flags) || |
2766ff61 FM |
3159 | test_bit(BTRFS_ORDERED_TRUNCATED, &oe->flags); |
3160 | ||
3c38c877 NB |
3161 | return insert_reserved_file_extent(trans, BTRFS_I(oe->inode), |
3162 | oe->file_offset, &stack_fi, | |
2766ff61 | 3163 | update_inode_bytes, oe->qgroup_rsv); |
203f44c5 QW |
3164 | } |
3165 | ||
3166 | /* | |
3167 | * As ordered data IO finishes, this gets called so we can finish | |
d352ac68 CM |
3168 | * an ordered extent if the range of bytes in the file it covers are |
3169 | * fully written. | |
3170 | */ | |
71df088c | 3171 | int btrfs_finish_one_ordered(struct btrfs_ordered_extent *ordered_extent) |
e6dcd2dc | 3172 | { |
72e7e6ed NB |
3173 | struct btrfs_inode *inode = BTRFS_I(ordered_extent->inode); |
3174 | struct btrfs_root *root = inode->root; | |
3175 | struct btrfs_fs_info *fs_info = root->fs_info; | |
0ca1f7ce | 3176 | struct btrfs_trans_handle *trans = NULL; |
72e7e6ed | 3177 | struct extent_io_tree *io_tree = &inode->io_tree; |
2ac55d41 | 3178 | struct extent_state *cached_state = NULL; |
bffe633e | 3179 | u64 start, end; |
261507a0 | 3180 | int compress_type = 0; |
77cef2ec | 3181 | int ret = 0; |
bffe633e | 3182 | u64 logical_len = ordered_extent->num_bytes; |
8d510121 | 3183 | bool freespace_inode; |
77cef2ec | 3184 | bool truncated = false; |
49940bdd | 3185 | bool clear_reserved_extent = true; |
2766ff61 | 3186 | unsigned int clear_bits = EXTENT_DEFRAG; |
a7e3b975 | 3187 | |
bffe633e OS |
3188 | start = ordered_extent->file_offset; |
3189 | end = start + ordered_extent->num_bytes - 1; | |
3190 | ||
a7e3b975 FM |
3191 | if (!test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) && |
3192 | !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags) && | |
7c0c7269 OS |
3193 | !test_bit(BTRFS_ORDERED_DIRECT, &ordered_extent->flags) && |
3194 | !test_bit(BTRFS_ORDERED_ENCODED, &ordered_extent->flags)) | |
2766ff61 | 3195 | clear_bits |= EXTENT_DELALLOC_NEW; |
e6dcd2dc | 3196 | |
72e7e6ed | 3197 | freespace_inode = btrfs_is_free_space_inode(inode); |
5f4403e1 IA |
3198 | if (!freespace_inode) |
3199 | btrfs_lockdep_acquire(fs_info, btrfs_ordered_extent); | |
0cb59c99 | 3200 | |
5fd02043 JB |
3201 | if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) { |
3202 | ret = -EIO; | |
3203 | goto out; | |
3204 | } | |
3205 | ||
71df088c | 3206 | if (btrfs_is_zoned(fs_info)) |
be1a1d7a NA |
3207 | btrfs_zone_finish_endio(fs_info, ordered_extent->disk_bytenr, |
3208 | ordered_extent->disk_num_bytes); | |
d8e3fb10 | 3209 | |
77cef2ec JB |
3210 | if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) { |
3211 | truncated = true; | |
3212 | logical_len = ordered_extent->truncated_len; | |
3213 | /* Truncated the entire extent, don't bother adding */ | |
3214 | if (!logical_len) | |
3215 | goto out; | |
3216 | } | |
3217 | ||
c2167754 | 3218 | if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) { |
79787eaa | 3219 | BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */ |
94ed938a | 3220 | |
72e7e6ed | 3221 | btrfs_inode_safe_disk_i_size_write(inode, 0); |
8d510121 NB |
3222 | if (freespace_inode) |
3223 | trans = btrfs_join_transaction_spacecache(root); | |
6c760c07 JB |
3224 | else |
3225 | trans = btrfs_join_transaction(root); | |
3226 | if (IS_ERR(trans)) { | |
3227 | ret = PTR_ERR(trans); | |
3228 | trans = NULL; | |
3229 | goto out; | |
c2167754 | 3230 | } |
72e7e6ed | 3231 | trans->block_rsv = &inode->block_rsv; |
729f7961 | 3232 | ret = btrfs_update_inode_fallback(trans, root, inode); |
6c760c07 | 3233 | if (ret) /* -ENOMEM or corruption */ |
66642832 | 3234 | btrfs_abort_transaction(trans, ret); |
c2167754 YZ |
3235 | goto out; |
3236 | } | |
e6dcd2dc | 3237 | |
2766ff61 | 3238 | clear_bits |= EXTENT_LOCKED; |
570eb97b | 3239 | lock_extent(io_tree, start, end, &cached_state); |
e6dcd2dc | 3240 | |
8d510121 NB |
3241 | if (freespace_inode) |
3242 | trans = btrfs_join_transaction_spacecache(root); | |
0cb59c99 | 3243 | else |
7a7eaa40 | 3244 | trans = btrfs_join_transaction(root); |
79787eaa JM |
3245 | if (IS_ERR(trans)) { |
3246 | ret = PTR_ERR(trans); | |
3247 | trans = NULL; | |
a7e3b975 | 3248 | goto out; |
79787eaa | 3249 | } |
a79b7d4b | 3250 | |
72e7e6ed | 3251 | trans->block_rsv = &inode->block_rsv; |
c2167754 | 3252 | |
c8b97818 | 3253 | if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags)) |
261507a0 | 3254 | compress_type = ordered_extent->compress_type; |
d899e052 | 3255 | if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) { |
261507a0 | 3256 | BUG_ON(compress_type); |
72e7e6ed | 3257 | ret = btrfs_mark_extent_written(trans, inode, |
d899e052 YZ |
3258 | ordered_extent->file_offset, |
3259 | ordered_extent->file_offset + | |
77cef2ec | 3260 | logical_len); |
343d8a30 NA |
3261 | btrfs_zoned_release_data_reloc_bg(fs_info, ordered_extent->disk_bytenr, |
3262 | ordered_extent->disk_num_bytes); | |
d899e052 | 3263 | } else { |
0b246afa | 3264 | BUG_ON(root == fs_info->tree_root); |
3c38c877 | 3265 | ret = insert_ordered_extent_file_extent(trans, ordered_extent); |
49940bdd JB |
3266 | if (!ret) { |
3267 | clear_reserved_extent = false; | |
2ff7e61e | 3268 | btrfs_release_delalloc_bytes(fs_info, |
bffe633e OS |
3269 | ordered_extent->disk_bytenr, |
3270 | ordered_extent->disk_num_bytes); | |
49940bdd | 3271 | } |
d899e052 | 3272 | } |
72e7e6ed | 3273 | unpin_extent_cache(&inode->extent_tree, ordered_extent->file_offset, |
bffe633e | 3274 | ordered_extent->num_bytes, trans->transid); |
79787eaa | 3275 | if (ret < 0) { |
66642832 | 3276 | btrfs_abort_transaction(trans, ret); |
a7e3b975 | 3277 | goto out; |
79787eaa | 3278 | } |
2ac55d41 | 3279 | |
510f85ed | 3280 | ret = add_pending_csums(trans, &ordered_extent->list); |
ac01f26a NB |
3281 | if (ret) { |
3282 | btrfs_abort_transaction(trans, ret); | |
3283 | goto out; | |
3284 | } | |
e6dcd2dc | 3285 | |
2766ff61 FM |
3286 | /* |
3287 | * If this is a new delalloc range, clear its new delalloc flag to | |
3288 | * update the inode's number of bytes. This needs to be done first | |
3289 | * before updating the inode item. | |
3290 | */ | |
3291 | if ((clear_bits & EXTENT_DELALLOC_NEW) && | |
3292 | !test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) | |
72e7e6ed | 3293 | clear_extent_bit(&inode->io_tree, start, end, |
2766ff61 | 3294 | EXTENT_DELALLOC_NEW | EXTENT_ADD_INODE_BYTES, |
bd015294 | 3295 | &cached_state); |
2766ff61 | 3296 | |
72e7e6ed | 3297 | btrfs_inode_safe_disk_i_size_write(inode, 0); |
729f7961 | 3298 | ret = btrfs_update_inode_fallback(trans, root, inode); |
6c760c07 | 3299 | if (ret) { /* -ENOMEM or corruption */ |
66642832 | 3300 | btrfs_abort_transaction(trans, ret); |
a7e3b975 | 3301 | goto out; |
1ef30be1 JB |
3302 | } |
3303 | ret = 0; | |
c2167754 | 3304 | out: |
bd015294 | 3305 | clear_extent_bit(&inode->io_tree, start, end, clear_bits, |
313facc5 | 3306 | &cached_state); |
a7e3b975 | 3307 | |
a698d075 | 3308 | if (trans) |
3a45bb20 | 3309 | btrfs_end_transaction(trans); |
0cb59c99 | 3310 | |
77cef2ec | 3311 | if (ret || truncated) { |
bffe633e | 3312 | u64 unwritten_start = start; |
77cef2ec | 3313 | |
d61bec08 JB |
3314 | /* |
3315 | * If we failed to finish this ordered extent for any reason we | |
3316 | * need to make sure BTRFS_ORDERED_IOERR is set on the ordered | |
3317 | * extent, and mark the inode with the error if it wasn't | |
3318 | * already set. Any error during writeback would have already | |
3319 | * set the mapping error, so we need to set it if we're the ones | |
3320 | * marking this ordered extent as failed. | |
3321 | */ | |
3322 | if (ret && !test_and_set_bit(BTRFS_ORDERED_IOERR, | |
3323 | &ordered_extent->flags)) | |
3324 | mapping_set_error(ordered_extent->inode->i_mapping, -EIO); | |
3325 | ||
77cef2ec | 3326 | if (truncated) |
bffe633e OS |
3327 | unwritten_start += logical_len; |
3328 | clear_extent_uptodate(io_tree, unwritten_start, end, NULL); | |
77cef2ec | 3329 | |
4c0c8cfc FM |
3330 | /* Drop extent maps for the part of the extent we didn't write. */ |
3331 | btrfs_drop_extent_map_range(inode, unwritten_start, end, false); | |
5fd02043 | 3332 | |
0bec9ef5 JB |
3333 | /* |
3334 | * If the ordered extent had an IOERR or something else went | |
3335 | * wrong we need to return the space for this ordered extent | |
77cef2ec JB |
3336 | * back to the allocator. We only free the extent in the |
3337 | * truncated case if we didn't write out the extent at all. | |
49940bdd JB |
3338 | * |
3339 | * If we made it past insert_reserved_file_extent before we | |
3340 | * errored out then we don't need to do this as the accounting | |
3341 | * has already been done. | |
0bec9ef5 | 3342 | */ |
77cef2ec | 3343 | if ((ret || !logical_len) && |
49940bdd | 3344 | clear_reserved_extent && |
77cef2ec | 3345 | !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) && |
4eaaec24 NB |
3346 | !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) { |
3347 | /* | |
3348 | * Discard the range before returning it back to the | |
3349 | * free space pool | |
3350 | */ | |
46b27f50 | 3351 | if (ret && btrfs_test_opt(fs_info, DISCARD_SYNC)) |
4eaaec24 | 3352 | btrfs_discard_extent(fs_info, |
bffe633e OS |
3353 | ordered_extent->disk_bytenr, |
3354 | ordered_extent->disk_num_bytes, | |
3355 | NULL); | |
2ff7e61e | 3356 | btrfs_free_reserved_extent(fs_info, |
bffe633e OS |
3357 | ordered_extent->disk_bytenr, |
3358 | ordered_extent->disk_num_bytes, 1); | |
e28b0211 BB |
3359 | /* |
3360 | * Actually free the qgroup rsv which was released when | |
3361 | * the ordered extent was created. | |
3362 | */ | |
3363 | btrfs_qgroup_free_refroot(fs_info, inode->root->root_key.objectid, | |
3364 | ordered_extent->qgroup_rsv, | |
3365 | BTRFS_QGROUP_RSV_DATA); | |
4eaaec24 | 3366 | } |
0bec9ef5 JB |
3367 | } |
3368 | ||
5fd02043 | 3369 | /* |
8bad3c02 LB |
3370 | * This needs to be done to make sure anybody waiting knows we are done |
3371 | * updating everything for this ordered extent. | |
5fd02043 | 3372 | */ |
72e7e6ed | 3373 | btrfs_remove_ordered_extent(inode, ordered_extent); |
5fd02043 | 3374 | |
e6dcd2dc CM |
3375 | /* once for us */ |
3376 | btrfs_put_ordered_extent(ordered_extent); | |
3377 | /* once for the tree */ | |
3378 | btrfs_put_ordered_extent(ordered_extent); | |
3379 | ||
5fd02043 JB |
3380 | return ret; |
3381 | } | |
3382 | ||
71df088c CH |
3383 | int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered) |
3384 | { | |
3385 | if (btrfs_is_zoned(btrfs_sb(ordered->inode->i_sb)) && | |
3386 | !test_bit(BTRFS_ORDERED_IOERR, &ordered->flags)) | |
3387 | btrfs_finish_ordered_zoned(ordered); | |
3388 | return btrfs_finish_one_ordered(ordered); | |
3389 | } | |
3390 | ||
ae643a74 QW |
3391 | /* |
3392 | * Verify the checksum for a single sector without any extra action that depend | |
3393 | * on the type of I/O. | |
3394 | */ | |
3395 | int btrfs_check_sector_csum(struct btrfs_fs_info *fs_info, struct page *page, | |
3396 | u32 pgoff, u8 *csum, const u8 * const csum_expected) | |
3397 | { | |
3398 | SHASH_DESC_ON_STACK(shash, fs_info->csum_shash); | |
3399 | char *kaddr; | |
3400 | ||
3401 | ASSERT(pgoff + fs_info->sectorsize <= PAGE_SIZE); | |
3402 | ||
3403 | shash->tfm = fs_info->csum_shash; | |
3404 | ||
3405 | kaddr = kmap_local_page(page) + pgoff; | |
3406 | crypto_shash_digest(shash, kaddr, fs_info->sectorsize, csum); | |
3407 | kunmap_local(kaddr); | |
3408 | ||
3409 | if (memcmp(csum, csum_expected, fs_info->csum_size)) | |
3410 | return -EIO; | |
3411 | return 0; | |
211f90e6 CM |
3412 | } |
3413 | ||
265d4ac0 | 3414 | /* |
e5219044 CH |
3415 | * Verify the checksum of a single data sector. |
3416 | * | |
3417 | * @bbio: btrfs_io_bio which contains the csum | |
3418 | * @dev: device the sector is on | |
7ffd27e3 | 3419 | * @bio_offset: offset to the beginning of the bio (in bytes) |
e5219044 | 3420 | * @bv: bio_vec to check |
265d4ac0 | 3421 | * |
e5219044 CH |
3422 | * Check if the checksum on a data block is valid. When a checksum mismatch is |
3423 | * detected, report the error and fill the corrupted range with zero. | |
ae643a74 | 3424 | * |
e5219044 | 3425 | * Return %true if the sector is ok or had no checksum to start with, else %false. |
265d4ac0 | 3426 | */ |
e5219044 CH |
3427 | bool btrfs_data_csum_ok(struct btrfs_bio *bbio, struct btrfs_device *dev, |
3428 | u32 bio_offset, struct bio_vec *bv) | |
dc380aea | 3429 | { |
e5219044 | 3430 | struct btrfs_inode *inode = bbio->inode; |
621af94a | 3431 | struct btrfs_fs_info *fs_info = inode->root->fs_info; |
e5219044 CH |
3432 | u64 file_offset = bbio->file_offset + bio_offset; |
3433 | u64 end = file_offset + bv->bv_len - 1; | |
d5178578 JT |
3434 | u8 *csum_expected; |
3435 | u8 csum[BTRFS_CSUM_SIZE]; | |
dc380aea | 3436 | |
3d49d0d3 | 3437 | ASSERT(bv->bv_len == fs_info->sectorsize); |
265d4ac0 | 3438 | |
e5219044 CH |
3439 | if (!bbio->csum) |
3440 | return true; | |
d5178578 | 3441 | |
e5219044 CH |
3442 | if (btrfs_is_data_reloc_root(inode->root) && |
3443 | test_range_bit(&inode->io_tree, file_offset, end, EXTENT_NODATASUM, | |
3444 | 1, NULL)) { | |
3445 | /* Skip the range without csum for data reloc inode */ | |
3446 | clear_extent_bits(&inode->io_tree, file_offset, end, | |
3447 | EXTENT_NODATASUM); | |
3448 | return true; | |
3449 | } | |
3450 | ||
fa13661c JT |
3451 | csum_expected = bbio->csum + (bio_offset >> fs_info->sectorsize_bits) * |
3452 | fs_info->csum_size; | |
3d49d0d3 CH |
3453 | if (btrfs_check_sector_csum(fs_info, bv->bv_page, bv->bv_offset, csum, |
3454 | csum_expected)) | |
dc380aea | 3455 | goto zeroit; |
e5219044 | 3456 | return true; |
ae643a74 | 3457 | |
dc380aea | 3458 | zeroit: |
3d49d0d3 CH |
3459 | btrfs_print_data_csum_error(inode, file_offset, csum, csum_expected, |
3460 | bbio->mirror_num); | |
3461 | if (dev) | |
3462 | btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS); | |
3463 | memzero_bvec(bv); | |
3464 | return false; | |
07157aac | 3465 | } |
b888db2b | 3466 | |
c1c3fac2 NB |
3467 | /* |
3468 | * btrfs_add_delayed_iput - perform a delayed iput on @inode | |
3469 | * | |
3470 | * @inode: The inode we want to perform iput on | |
3471 | * | |
3472 | * This function uses the generic vfs_inode::i_count to track whether we should | |
3473 | * just decrement it (in case it's > 1) or if this is the last iput then link | |
3474 | * the inode to the delayed iput machinery. Delayed iputs are processed at | |
3475 | * transaction commit time/superblock commit/cleaner kthread. | |
3476 | */ | |
e55cf7ca | 3477 | void btrfs_add_delayed_iput(struct btrfs_inode *inode) |
24bbcf04 | 3478 | { |
e55cf7ca | 3479 | struct btrfs_fs_info *fs_info = inode->root->fs_info; |
866e98a4 | 3480 | unsigned long flags; |
24bbcf04 | 3481 | |
e55cf7ca | 3482 | if (atomic_add_unless(&inode->vfs_inode.i_count, -1, 1)) |
24bbcf04 YZ |
3483 | return; |
3484 | ||
034f784d | 3485 | atomic_inc(&fs_info->nr_delayed_iputs); |
866e98a4 FM |
3486 | /* |
3487 | * Need to be irq safe here because we can be called from either an irq | |
3488 | * context (see bio.c and btrfs_put_ordered_extent()) or a non-irq | |
3489 | * context. | |
3490 | */ | |
3491 | spin_lock_irqsave(&fs_info->delayed_iput_lock, flags); | |
e55cf7ca DS |
3492 | ASSERT(list_empty(&inode->delayed_iput)); |
3493 | list_add_tail(&inode->delayed_iput, &fs_info->delayed_iputs); | |
866e98a4 | 3494 | spin_unlock_irqrestore(&fs_info->delayed_iput_lock, flags); |
fd340d0f JB |
3495 | if (!test_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags)) |
3496 | wake_up_process(fs_info->cleaner_kthread); | |
24bbcf04 YZ |
3497 | } |
3498 | ||
63611e73 JB |
3499 | static void run_delayed_iput_locked(struct btrfs_fs_info *fs_info, |
3500 | struct btrfs_inode *inode) | |
3501 | { | |
3502 | list_del_init(&inode->delayed_iput); | |
866e98a4 | 3503 | spin_unlock_irq(&fs_info->delayed_iput_lock); |
63611e73 JB |
3504 | iput(&inode->vfs_inode); |
3505 | if (atomic_dec_and_test(&fs_info->nr_delayed_iputs)) | |
3506 | wake_up(&fs_info->delayed_iputs_wait); | |
866e98a4 | 3507 | spin_lock_irq(&fs_info->delayed_iput_lock); |
63611e73 JB |
3508 | } |
3509 | ||
3510 | static void btrfs_run_delayed_iput(struct btrfs_fs_info *fs_info, | |
3511 | struct btrfs_inode *inode) | |
3512 | { | |
3513 | if (!list_empty(&inode->delayed_iput)) { | |
866e98a4 | 3514 | spin_lock_irq(&fs_info->delayed_iput_lock); |
63611e73 JB |
3515 | if (!list_empty(&inode->delayed_iput)) |
3516 | run_delayed_iput_locked(fs_info, inode); | |
866e98a4 | 3517 | spin_unlock_irq(&fs_info->delayed_iput_lock); |
63611e73 JB |
3518 | } |
3519 | } | |
3520 | ||
2ff7e61e | 3521 | void btrfs_run_delayed_iputs(struct btrfs_fs_info *fs_info) |
24bbcf04 | 3522 | { |
866e98a4 FM |
3523 | /* |
3524 | * btrfs_put_ordered_extent() can run in irq context (see bio.c), which | |
3525 | * calls btrfs_add_delayed_iput() and that needs to lock | |
3526 | * fs_info->delayed_iput_lock. So we need to disable irqs here to | |
3527 | * prevent a deadlock. | |
3528 | */ | |
3529 | spin_lock_irq(&fs_info->delayed_iput_lock); | |
8089fe62 DS |
3530 | while (!list_empty(&fs_info->delayed_iputs)) { |
3531 | struct btrfs_inode *inode; | |
3532 | ||
3533 | inode = list_first_entry(&fs_info->delayed_iputs, | |
3534 | struct btrfs_inode, delayed_iput); | |
63611e73 | 3535 | run_delayed_iput_locked(fs_info, inode); |
866e98a4 FM |
3536 | if (need_resched()) { |
3537 | spin_unlock_irq(&fs_info->delayed_iput_lock); | |
3538 | cond_resched(); | |
3539 | spin_lock_irq(&fs_info->delayed_iput_lock); | |
3540 | } | |
24bbcf04 | 3541 | } |
866e98a4 | 3542 | spin_unlock_irq(&fs_info->delayed_iput_lock); |
24bbcf04 YZ |
3543 | } |
3544 | ||
e43eec81 | 3545 | /* |
2639631d NB |
3546 | * Wait for flushing all delayed iputs |
3547 | * | |
3548 | * @fs_info: the filesystem | |
034f784d JB |
3549 | * |
3550 | * This will wait on any delayed iputs that are currently running with KILLABLE | |
3551 | * set. Once they are all done running we will return, unless we are killed in | |
3552 | * which case we return EINTR. This helps in user operations like fallocate etc | |
3553 | * that might get blocked on the iputs. | |
2639631d NB |
3554 | * |
3555 | * Return EINTR if we were killed, 0 if nothing's pending | |
034f784d JB |
3556 | */ |
3557 | int btrfs_wait_on_delayed_iputs(struct btrfs_fs_info *fs_info) | |
3558 | { | |
3559 | int ret = wait_event_killable(fs_info->delayed_iputs_wait, | |
3560 | atomic_read(&fs_info->nr_delayed_iputs) == 0); | |
3561 | if (ret) | |
3562 | return -EINTR; | |
3563 | return 0; | |
3564 | } | |
3565 | ||
7b128766 | 3566 | /* |
f7e9e8fc OS |
3567 | * This creates an orphan entry for the given inode in case something goes wrong |
3568 | * in the middle of an unlink. | |
7b128766 | 3569 | */ |
73f2e545 | 3570 | int btrfs_orphan_add(struct btrfs_trans_handle *trans, |
27919067 | 3571 | struct btrfs_inode *inode) |
7b128766 | 3572 | { |
d68fc57b | 3573 | int ret; |
7b128766 | 3574 | |
27919067 OS |
3575 | ret = btrfs_insert_orphan_item(trans, inode->root, btrfs_ino(inode)); |
3576 | if (ret && ret != -EEXIST) { | |
3577 | btrfs_abort_transaction(trans, ret); | |
3578 | return ret; | |
d68fc57b YZ |
3579 | } |
3580 | ||
d68fc57b | 3581 | return 0; |
7b128766 JB |
3582 | } |
3583 | ||
3584 | /* | |
f7e9e8fc OS |
3585 | * We have done the delete so we can go ahead and remove the orphan item for |
3586 | * this particular inode. | |
7b128766 | 3587 | */ |
48a3b636 | 3588 | static int btrfs_orphan_del(struct btrfs_trans_handle *trans, |
3d6ae7bb | 3589 | struct btrfs_inode *inode) |
7b128766 | 3590 | { |
27919067 | 3591 | return btrfs_del_orphan_item(trans, inode->root, btrfs_ino(inode)); |
7b128766 JB |
3592 | } |
3593 | ||
3594 | /* | |
3595 | * this cleans up any orphans that may be left on the list from the last use | |
3596 | * of this root. | |
3597 | */ | |
66b4ffd1 | 3598 | int btrfs_orphan_cleanup(struct btrfs_root *root) |
7b128766 | 3599 | { |
0b246afa | 3600 | struct btrfs_fs_info *fs_info = root->fs_info; |
7b128766 JB |
3601 | struct btrfs_path *path; |
3602 | struct extent_buffer *leaf; | |
7b128766 JB |
3603 | struct btrfs_key key, found_key; |
3604 | struct btrfs_trans_handle *trans; | |
3605 | struct inode *inode; | |
8f6d7f4f | 3606 | u64 last_objectid = 0; |
f7e9e8fc | 3607 | int ret = 0, nr_unlink = 0; |
7b128766 | 3608 | |
54230013 | 3609 | if (test_and_set_bit(BTRFS_ROOT_ORPHAN_CLEANUP, &root->state)) |
66b4ffd1 | 3610 | return 0; |
c71bf099 YZ |
3611 | |
3612 | path = btrfs_alloc_path(); | |
66b4ffd1 JB |
3613 | if (!path) { |
3614 | ret = -ENOMEM; | |
3615 | goto out; | |
3616 | } | |
e4058b54 | 3617 | path->reada = READA_BACK; |
7b128766 JB |
3618 | |
3619 | key.objectid = BTRFS_ORPHAN_OBJECTID; | |
962a298f | 3620 | key.type = BTRFS_ORPHAN_ITEM_KEY; |
7b128766 JB |
3621 | key.offset = (u64)-1; |
3622 | ||
7b128766 JB |
3623 | while (1) { |
3624 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
66b4ffd1 JB |
3625 | if (ret < 0) |
3626 | goto out; | |
7b128766 JB |
3627 | |
3628 | /* | |
3629 | * if ret == 0 means we found what we were searching for, which | |
25985edc | 3630 | * is weird, but possible, so only screw with path if we didn't |
7b128766 JB |
3631 | * find the key and see if we have stuff that matches |
3632 | */ | |
3633 | if (ret > 0) { | |
66b4ffd1 | 3634 | ret = 0; |
7b128766 JB |
3635 | if (path->slots[0] == 0) |
3636 | break; | |
3637 | path->slots[0]--; | |
3638 | } | |
3639 | ||
3640 | /* pull out the item */ | |
3641 | leaf = path->nodes[0]; | |
7b128766 JB |
3642 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
3643 | ||
3644 | /* make sure the item matches what we want */ | |
3645 | if (found_key.objectid != BTRFS_ORPHAN_OBJECTID) | |
3646 | break; | |
962a298f | 3647 | if (found_key.type != BTRFS_ORPHAN_ITEM_KEY) |
7b128766 JB |
3648 | break; |
3649 | ||
3650 | /* release the path since we're done with it */ | |
b3b4aa74 | 3651 | btrfs_release_path(path); |
7b128766 JB |
3652 | |
3653 | /* | |
3654 | * this is where we are basically btrfs_lookup, without the | |
3655 | * crossing root thing. we store the inode number in the | |
3656 | * offset of the orphan item. | |
3657 | */ | |
8f6d7f4f JB |
3658 | |
3659 | if (found_key.offset == last_objectid) { | |
0b246afa JM |
3660 | btrfs_err(fs_info, |
3661 | "Error removing orphan entry, stopping orphan cleanup"); | |
8f6d7f4f JB |
3662 | ret = -EINVAL; |
3663 | goto out; | |
3664 | } | |
3665 | ||
3666 | last_objectid = found_key.offset; | |
3667 | ||
5d4f98a2 YZ |
3668 | found_key.objectid = found_key.offset; |
3669 | found_key.type = BTRFS_INODE_ITEM_KEY; | |
3670 | found_key.offset = 0; | |
0202e83f | 3671 | inode = btrfs_iget(fs_info->sb, last_objectid, root); |
cbaee87f FM |
3672 | if (IS_ERR(inode)) { |
3673 | ret = PTR_ERR(inode); | |
3674 | inode = NULL; | |
3675 | if (ret != -ENOENT) | |
3676 | goto out; | |
3677 | } | |
7b128766 | 3678 | |
cbaee87f | 3679 | if (!inode && root == fs_info->tree_root) { |
f8e9e0b0 | 3680 | struct btrfs_root *dead_root; |
f8e9e0b0 AJ |
3681 | int is_dead_root = 0; |
3682 | ||
3683 | /* | |
0c0218e9 | 3684 | * This is an orphan in the tree root. Currently these |
f8e9e0b0 | 3685 | * could come from 2 sources: |
0c0218e9 | 3686 | * a) a root (snapshot/subvolume) deletion in progress |
f8e9e0b0 | 3687 | * b) a free space cache inode |
0c0218e9 FM |
3688 | * We need to distinguish those two, as the orphan item |
3689 | * for a root must not get deleted before the deletion | |
3690 | * of the snapshot/subvolume's tree completes. | |
3691 | * | |
3692 | * btrfs_find_orphan_roots() ran before us, which has | |
3693 | * found all deleted roots and loaded them into | |
fc7cbcd4 | 3694 | * fs_info->fs_roots_radix. So here we can find if an |
0c0218e9 | 3695 | * orphan item corresponds to a deleted root by looking |
fc7cbcd4 | 3696 | * up the root from that radix tree. |
f8e9e0b0 | 3697 | */ |
a619b3c7 | 3698 | |
fc7cbcd4 DS |
3699 | spin_lock(&fs_info->fs_roots_radix_lock); |
3700 | dead_root = radix_tree_lookup(&fs_info->fs_roots_radix, | |
3701 | (unsigned long)found_key.objectid); | |
a619b3c7 RK |
3702 | if (dead_root && btrfs_root_refs(&dead_root->root_item) == 0) |
3703 | is_dead_root = 1; | |
fc7cbcd4 | 3704 | spin_unlock(&fs_info->fs_roots_radix_lock); |
a619b3c7 | 3705 | |
f8e9e0b0 AJ |
3706 | if (is_dead_root) { |
3707 | /* prevent this orphan from being found again */ | |
3708 | key.offset = found_key.objectid - 1; | |
3709 | continue; | |
3710 | } | |
f7e9e8fc | 3711 | |
f8e9e0b0 | 3712 | } |
f7e9e8fc | 3713 | |
7b128766 | 3714 | /* |
f7e9e8fc | 3715 | * If we have an inode with links, there are a couple of |
70524253 BB |
3716 | * possibilities: |
3717 | * | |
3718 | * 1. We were halfway through creating fsverity metadata for the | |
3719 | * file. In that case, the orphan item represents incomplete | |
3720 | * fsverity metadata which must be cleaned up with | |
3721 | * btrfs_drop_verity_items and deleting the orphan item. | |
3722 | ||
3723 | * 2. Old kernels (before v3.12) used to create an | |
f7e9e8fc OS |
3724 | * orphan item for truncate indicating that there were possibly |
3725 | * extent items past i_size that needed to be deleted. In v3.12, | |
3726 | * truncate was changed to update i_size in sync with the extent | |
3727 | * items, but the (useless) orphan item was still created. Since | |
3728 | * v4.18, we don't create the orphan item for truncate at all. | |
3729 | * | |
3730 | * So, this item could mean that we need to do a truncate, but | |
3731 | * only if this filesystem was last used on a pre-v3.12 kernel | |
3732 | * and was not cleanly unmounted. The odds of that are quite | |
3733 | * slim, and it's a pain to do the truncate now, so just delete | |
3734 | * the orphan item. | |
3735 | * | |
3736 | * It's also possible that this orphan item was supposed to be | |
3737 | * deleted but wasn't. The inode number may have been reused, | |
3738 | * but either way, we can delete the orphan item. | |
7b128766 | 3739 | */ |
cbaee87f FM |
3740 | if (!inode || inode->i_nlink) { |
3741 | if (inode) { | |
70524253 | 3742 | ret = btrfs_drop_verity_items(BTRFS_I(inode)); |
f7e9e8fc | 3743 | iput(inode); |
b777d279 | 3744 | inode = NULL; |
70524253 BB |
3745 | if (ret) |
3746 | goto out; | |
3747 | } | |
a8c9e576 | 3748 | trans = btrfs_start_transaction(root, 1); |
66b4ffd1 JB |
3749 | if (IS_ERR(trans)) { |
3750 | ret = PTR_ERR(trans); | |
3751 | goto out; | |
3752 | } | |
0b246afa JM |
3753 | btrfs_debug(fs_info, "auto deleting %Lu", |
3754 | found_key.objectid); | |
a8c9e576 JB |
3755 | ret = btrfs_del_orphan_item(trans, root, |
3756 | found_key.objectid); | |
3a45bb20 | 3757 | btrfs_end_transaction(trans); |
cbaee87f | 3758 | if (ret) |
4ef31a45 | 3759 | goto out; |
7b128766 JB |
3760 | continue; |
3761 | } | |
3762 | ||
f7e9e8fc | 3763 | nr_unlink++; |
7b128766 JB |
3764 | |
3765 | /* this will do delete_inode and everything for us */ | |
3766 | iput(inode); | |
3767 | } | |
3254c876 MX |
3768 | /* release the path since we're done with it */ |
3769 | btrfs_release_path(path); | |
3770 | ||
a575ceeb | 3771 | if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state)) { |
7a7eaa40 | 3772 | trans = btrfs_join_transaction(root); |
66b4ffd1 | 3773 | if (!IS_ERR(trans)) |
3a45bb20 | 3774 | btrfs_end_transaction(trans); |
d68fc57b | 3775 | } |
7b128766 JB |
3776 | |
3777 | if (nr_unlink) | |
0b246afa | 3778 | btrfs_debug(fs_info, "unlinked %d orphans", nr_unlink); |
66b4ffd1 JB |
3779 | |
3780 | out: | |
3781 | if (ret) | |
0b246afa | 3782 | btrfs_err(fs_info, "could not do orphan cleanup %d", ret); |
66b4ffd1 JB |
3783 | btrfs_free_path(path); |
3784 | return ret; | |
7b128766 JB |
3785 | } |
3786 | ||
46a53cca CM |
3787 | /* |
3788 | * very simple check to peek ahead in the leaf looking for xattrs. If we | |
3789 | * don't find any xattrs, we know there can't be any acls. | |
3790 | * | |
3791 | * slot is the slot the inode is in, objectid is the objectid of the inode | |
3792 | */ | |
3793 | static noinline int acls_after_inode_item(struct extent_buffer *leaf, | |
63541927 FDBM |
3794 | int slot, u64 objectid, |
3795 | int *first_xattr_slot) | |
46a53cca CM |
3796 | { |
3797 | u32 nritems = btrfs_header_nritems(leaf); | |
3798 | struct btrfs_key found_key; | |
f23b5a59 JB |
3799 | static u64 xattr_access = 0; |
3800 | static u64 xattr_default = 0; | |
46a53cca CM |
3801 | int scanned = 0; |
3802 | ||
f23b5a59 | 3803 | if (!xattr_access) { |
97d79299 AG |
3804 | xattr_access = btrfs_name_hash(XATTR_NAME_POSIX_ACL_ACCESS, |
3805 | strlen(XATTR_NAME_POSIX_ACL_ACCESS)); | |
3806 | xattr_default = btrfs_name_hash(XATTR_NAME_POSIX_ACL_DEFAULT, | |
3807 | strlen(XATTR_NAME_POSIX_ACL_DEFAULT)); | |
f23b5a59 JB |
3808 | } |
3809 | ||
46a53cca | 3810 | slot++; |
63541927 | 3811 | *first_xattr_slot = -1; |
46a53cca CM |
3812 | while (slot < nritems) { |
3813 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
3814 | ||
3815 | /* we found a different objectid, there must not be acls */ | |
3816 | if (found_key.objectid != objectid) | |
3817 | return 0; | |
3818 | ||
3819 | /* we found an xattr, assume we've got an acl */ | |
f23b5a59 | 3820 | if (found_key.type == BTRFS_XATTR_ITEM_KEY) { |
63541927 FDBM |
3821 | if (*first_xattr_slot == -1) |
3822 | *first_xattr_slot = slot; | |
f23b5a59 JB |
3823 | if (found_key.offset == xattr_access || |
3824 | found_key.offset == xattr_default) | |
3825 | return 1; | |
3826 | } | |
46a53cca CM |
3827 | |
3828 | /* | |
3829 | * we found a key greater than an xattr key, there can't | |
3830 | * be any acls later on | |
3831 | */ | |
3832 | if (found_key.type > BTRFS_XATTR_ITEM_KEY) | |
3833 | return 0; | |
3834 | ||
3835 | slot++; | |
3836 | scanned++; | |
3837 | ||
3838 | /* | |
3839 | * it goes inode, inode backrefs, xattrs, extents, | |
3840 | * so if there are a ton of hard links to an inode there can | |
3841 | * be a lot of backrefs. Don't waste time searching too hard, | |
3842 | * this is just an optimization | |
3843 | */ | |
3844 | if (scanned >= 8) | |
3845 | break; | |
3846 | } | |
3847 | /* we hit the end of the leaf before we found an xattr or | |
3848 | * something larger than an xattr. We have to assume the inode | |
3849 | * has acls | |
3850 | */ | |
63541927 FDBM |
3851 | if (*first_xattr_slot == -1) |
3852 | *first_xattr_slot = slot; | |
46a53cca CM |
3853 | return 1; |
3854 | } | |
3855 | ||
d352ac68 CM |
3856 | /* |
3857 | * read an inode from the btree into the in-memory inode | |
3858 | */ | |
4222ea71 FM |
3859 | static int btrfs_read_locked_inode(struct inode *inode, |
3860 | struct btrfs_path *in_path) | |
39279cc3 | 3861 | { |
0b246afa | 3862 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
4222ea71 | 3863 | struct btrfs_path *path = in_path; |
5f39d397 | 3864 | struct extent_buffer *leaf; |
39279cc3 CM |
3865 | struct btrfs_inode_item *inode_item; |
3866 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
3867 | struct btrfs_key location; | |
67de1176 | 3868 | unsigned long ptr; |
46a53cca | 3869 | int maybe_acls; |
618e21d5 | 3870 | u32 rdev; |
39279cc3 | 3871 | int ret; |
2f7e33d4 | 3872 | bool filled = false; |
63541927 | 3873 | int first_xattr_slot; |
2f7e33d4 MX |
3874 | |
3875 | ret = btrfs_fill_inode(inode, &rdev); | |
3876 | if (!ret) | |
3877 | filled = true; | |
39279cc3 | 3878 | |
4222ea71 FM |
3879 | if (!path) { |
3880 | path = btrfs_alloc_path(); | |
3881 | if (!path) | |
3882 | return -ENOMEM; | |
3883 | } | |
1748f843 | 3884 | |
39279cc3 | 3885 | memcpy(&location, &BTRFS_I(inode)->location, sizeof(location)); |
dc17ff8f | 3886 | |
39279cc3 | 3887 | ret = btrfs_lookup_inode(NULL, root, path, &location, 0); |
67710892 | 3888 | if (ret) { |
4222ea71 FM |
3889 | if (path != in_path) |
3890 | btrfs_free_path(path); | |
f5b3a417 | 3891 | return ret; |
67710892 | 3892 | } |
39279cc3 | 3893 | |
5f39d397 | 3894 | leaf = path->nodes[0]; |
2f7e33d4 MX |
3895 | |
3896 | if (filled) | |
67de1176 | 3897 | goto cache_index; |
2f7e33d4 | 3898 | |
5f39d397 CM |
3899 | inode_item = btrfs_item_ptr(leaf, path->slots[0], |
3900 | struct btrfs_inode_item); | |
5f39d397 | 3901 | inode->i_mode = btrfs_inode_mode(leaf, inode_item); |
bfe86848 | 3902 | set_nlink(inode, btrfs_inode_nlink(leaf, inode_item)); |
2f2f43d3 EB |
3903 | i_uid_write(inode, btrfs_inode_uid(leaf, inode_item)); |
3904 | i_gid_write(inode, btrfs_inode_gid(leaf, inode_item)); | |
6ef06d27 | 3905 | btrfs_i_size_write(BTRFS_I(inode), btrfs_inode_size(leaf, inode_item)); |
41a2ee75 JB |
3906 | btrfs_inode_set_file_extent_range(BTRFS_I(inode), 0, |
3907 | round_up(i_size_read(inode), fs_info->sectorsize)); | |
5f39d397 | 3908 | |
a937b979 DS |
3909 | inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->atime); |
3910 | inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->atime); | |
5f39d397 | 3911 | |
a937b979 DS |
3912 | inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->mtime); |
3913 | inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->mtime); | |
5f39d397 | 3914 | |
a937b979 DS |
3915 | inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->ctime); |
3916 | inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->ctime); | |
5f39d397 | 3917 | |
9cc97d64 | 3918 | BTRFS_I(inode)->i_otime.tv_sec = |
3919 | btrfs_timespec_sec(leaf, &inode_item->otime); | |
3920 | BTRFS_I(inode)->i_otime.tv_nsec = | |
3921 | btrfs_timespec_nsec(leaf, &inode_item->otime); | |
5f39d397 | 3922 | |
a76a3cd4 | 3923 | inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item)); |
e02119d5 | 3924 | BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item); |
5dc562c5 JB |
3925 | BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item); |
3926 | ||
c7f88c4e JL |
3927 | inode_set_iversion_queried(inode, |
3928 | btrfs_inode_sequence(leaf, inode_item)); | |
6e17d30b YD |
3929 | inode->i_generation = BTRFS_I(inode)->generation; |
3930 | inode->i_rdev = 0; | |
3931 | rdev = btrfs_inode_rdev(leaf, inode_item); | |
3932 | ||
3933 | BTRFS_I(inode)->index_cnt = (u64)-1; | |
77eea05e BB |
3934 | btrfs_inode_split_flags(btrfs_inode_flags(leaf, inode_item), |
3935 | &BTRFS_I(inode)->flags, &BTRFS_I(inode)->ro_flags); | |
6e17d30b YD |
3936 | |
3937 | cache_index: | |
5dc562c5 JB |
3938 | /* |
3939 | * If we were modified in the current generation and evicted from memory | |
3940 | * and then re-read we need to do a full sync since we don't have any | |
3941 | * idea about which extents were modified before we were evicted from | |
3942 | * cache. | |
6e17d30b YD |
3943 | * |
3944 | * This is required for both inode re-read from disk and delayed inode | |
088aea3b | 3945 | * in delayed_nodes_tree. |
5dc562c5 | 3946 | */ |
0b246afa | 3947 | if (BTRFS_I(inode)->last_trans == fs_info->generation) |
5dc562c5 JB |
3948 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, |
3949 | &BTRFS_I(inode)->runtime_flags); | |
3950 | ||
bde6c242 FM |
3951 | /* |
3952 | * We don't persist the id of the transaction where an unlink operation | |
3953 | * against the inode was last made. So here we assume the inode might | |
3954 | * have been evicted, and therefore the exact value of last_unlink_trans | |
3955 | * lost, and set it to last_trans to avoid metadata inconsistencies | |
3956 | * between the inode and its parent if the inode is fsync'ed and the log | |
3957 | * replayed. For example, in the scenario: | |
3958 | * | |
3959 | * touch mydir/foo | |
3960 | * ln mydir/foo mydir/bar | |
3961 | * sync | |
3962 | * unlink mydir/bar | |
3963 | * echo 2 > /proc/sys/vm/drop_caches # evicts inode | |
3964 | * xfs_io -c fsync mydir/foo | |
3965 | * <power failure> | |
3966 | * mount fs, triggers fsync log replay | |
3967 | * | |
3968 | * We must make sure that when we fsync our inode foo we also log its | |
3969 | * parent inode, otherwise after log replay the parent still has the | |
3970 | * dentry with the "bar" name but our inode foo has a link count of 1 | |
3971 | * and doesn't have an inode ref with the name "bar" anymore. | |
3972 | * | |
3973 | * Setting last_unlink_trans to last_trans is a pessimistic approach, | |
01327610 | 3974 | * but it guarantees correctness at the expense of occasional full |
bde6c242 FM |
3975 | * transaction commits on fsync if our inode is a directory, or if our |
3976 | * inode is not a directory, logging its parent unnecessarily. | |
3977 | */ | |
3978 | BTRFS_I(inode)->last_unlink_trans = BTRFS_I(inode)->last_trans; | |
3979 | ||
3ebac17c FM |
3980 | /* |
3981 | * Same logic as for last_unlink_trans. We don't persist the generation | |
3982 | * of the last transaction where this inode was used for a reflink | |
3983 | * operation, so after eviction and reloading the inode we must be | |
3984 | * pessimistic and assume the last transaction that modified the inode. | |
3985 | */ | |
3986 | BTRFS_I(inode)->last_reflink_trans = BTRFS_I(inode)->last_trans; | |
3987 | ||
67de1176 MX |
3988 | path->slots[0]++; |
3989 | if (inode->i_nlink != 1 || | |
3990 | path->slots[0] >= btrfs_header_nritems(leaf)) | |
3991 | goto cache_acl; | |
3992 | ||
3993 | btrfs_item_key_to_cpu(leaf, &location, path->slots[0]); | |
4a0cc7ca | 3994 | if (location.objectid != btrfs_ino(BTRFS_I(inode))) |
67de1176 MX |
3995 | goto cache_acl; |
3996 | ||
3997 | ptr = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
3998 | if (location.type == BTRFS_INODE_REF_KEY) { | |
3999 | struct btrfs_inode_ref *ref; | |
4000 | ||
4001 | ref = (struct btrfs_inode_ref *)ptr; | |
4002 | BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref); | |
4003 | } else if (location.type == BTRFS_INODE_EXTREF_KEY) { | |
4004 | struct btrfs_inode_extref *extref; | |
4005 | ||
4006 | extref = (struct btrfs_inode_extref *)ptr; | |
4007 | BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf, | |
4008 | extref); | |
4009 | } | |
2f7e33d4 | 4010 | cache_acl: |
46a53cca CM |
4011 | /* |
4012 | * try to precache a NULL acl entry for files that don't have | |
4013 | * any xattrs or acls | |
4014 | */ | |
33345d01 | 4015 | maybe_acls = acls_after_inode_item(leaf, path->slots[0], |
f85b7379 | 4016 | btrfs_ino(BTRFS_I(inode)), &first_xattr_slot); |
63541927 FDBM |
4017 | if (first_xattr_slot != -1) { |
4018 | path->slots[0] = first_xattr_slot; | |
4019 | ret = btrfs_load_inode_props(inode, path); | |
4020 | if (ret) | |
0b246afa | 4021 | btrfs_err(fs_info, |
351fd353 | 4022 | "error loading props for ino %llu (root %llu): %d", |
4a0cc7ca | 4023 | btrfs_ino(BTRFS_I(inode)), |
63541927 FDBM |
4024 | root->root_key.objectid, ret); |
4025 | } | |
4222ea71 FM |
4026 | if (path != in_path) |
4027 | btrfs_free_path(path); | |
63541927 | 4028 | |
72c04902 AV |
4029 | if (!maybe_acls) |
4030 | cache_no_acl(inode); | |
46a53cca | 4031 | |
39279cc3 | 4032 | switch (inode->i_mode & S_IFMT) { |
39279cc3 CM |
4033 | case S_IFREG: |
4034 | inode->i_mapping->a_ops = &btrfs_aops; | |
4035 | inode->i_fop = &btrfs_file_operations; | |
4036 | inode->i_op = &btrfs_file_inode_operations; | |
4037 | break; | |
4038 | case S_IFDIR: | |
4039 | inode->i_fop = &btrfs_dir_file_operations; | |
67ade058 | 4040 | inode->i_op = &btrfs_dir_inode_operations; |
39279cc3 CM |
4041 | break; |
4042 | case S_IFLNK: | |
4043 | inode->i_op = &btrfs_symlink_inode_operations; | |
21fc61c7 | 4044 | inode_nohighmem(inode); |
4779cc04 | 4045 | inode->i_mapping->a_ops = &btrfs_aops; |
39279cc3 | 4046 | break; |
618e21d5 | 4047 | default: |
0279b4cd | 4048 | inode->i_op = &btrfs_special_inode_operations; |
618e21d5 JB |
4049 | init_special_inode(inode, inode->i_mode, rdev); |
4050 | break; | |
39279cc3 | 4051 | } |
6cbff00f | 4052 | |
7b6a221e | 4053 | btrfs_sync_inode_flags_to_i_flags(inode); |
67710892 | 4054 | return 0; |
39279cc3 CM |
4055 | } |
4056 | ||
d352ac68 CM |
4057 | /* |
4058 | * given a leaf and an inode, copy the inode fields into the leaf | |
4059 | */ | |
e02119d5 CM |
4060 | static void fill_inode_item(struct btrfs_trans_handle *trans, |
4061 | struct extent_buffer *leaf, | |
5f39d397 | 4062 | struct btrfs_inode_item *item, |
39279cc3 CM |
4063 | struct inode *inode) |
4064 | { | |
51fab693 | 4065 | struct btrfs_map_token token; |
77eea05e | 4066 | u64 flags; |
51fab693 | 4067 | |
c82f823c | 4068 | btrfs_init_map_token(&token, leaf); |
5f39d397 | 4069 | |
cc4c13d5 DS |
4070 | btrfs_set_token_inode_uid(&token, item, i_uid_read(inode)); |
4071 | btrfs_set_token_inode_gid(&token, item, i_gid_read(inode)); | |
4072 | btrfs_set_token_inode_size(&token, item, BTRFS_I(inode)->disk_i_size); | |
4073 | btrfs_set_token_inode_mode(&token, item, inode->i_mode); | |
4074 | btrfs_set_token_inode_nlink(&token, item, inode->i_nlink); | |
4075 | ||
4076 | btrfs_set_token_timespec_sec(&token, &item->atime, | |
4077 | inode->i_atime.tv_sec); | |
4078 | btrfs_set_token_timespec_nsec(&token, &item->atime, | |
4079 | inode->i_atime.tv_nsec); | |
4080 | ||
4081 | btrfs_set_token_timespec_sec(&token, &item->mtime, | |
4082 | inode->i_mtime.tv_sec); | |
4083 | btrfs_set_token_timespec_nsec(&token, &item->mtime, | |
4084 | inode->i_mtime.tv_nsec); | |
4085 | ||
4086 | btrfs_set_token_timespec_sec(&token, &item->ctime, | |
4087 | inode->i_ctime.tv_sec); | |
4088 | btrfs_set_token_timespec_nsec(&token, &item->ctime, | |
4089 | inode->i_ctime.tv_nsec); | |
4090 | ||
4091 | btrfs_set_token_timespec_sec(&token, &item->otime, | |
4092 | BTRFS_I(inode)->i_otime.tv_sec); | |
4093 | btrfs_set_token_timespec_nsec(&token, &item->otime, | |
4094 | BTRFS_I(inode)->i_otime.tv_nsec); | |
4095 | ||
4096 | btrfs_set_token_inode_nbytes(&token, item, inode_get_bytes(inode)); | |
4097 | btrfs_set_token_inode_generation(&token, item, | |
4098 | BTRFS_I(inode)->generation); | |
4099 | btrfs_set_token_inode_sequence(&token, item, inode_peek_iversion(inode)); | |
4100 | btrfs_set_token_inode_transid(&token, item, trans->transid); | |
4101 | btrfs_set_token_inode_rdev(&token, item, inode->i_rdev); | |
77eea05e BB |
4102 | flags = btrfs_inode_combine_flags(BTRFS_I(inode)->flags, |
4103 | BTRFS_I(inode)->ro_flags); | |
4104 | btrfs_set_token_inode_flags(&token, item, flags); | |
cc4c13d5 | 4105 | btrfs_set_token_inode_block_group(&token, item, 0); |
39279cc3 CM |
4106 | } |
4107 | ||
d352ac68 CM |
4108 | /* |
4109 | * copy everything in the in-memory inode into the btree. | |
4110 | */ | |
2115133f | 4111 | static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans, |
dfeb9e7c NB |
4112 | struct btrfs_root *root, |
4113 | struct btrfs_inode *inode) | |
39279cc3 CM |
4114 | { |
4115 | struct btrfs_inode_item *inode_item; | |
4116 | struct btrfs_path *path; | |
5f39d397 | 4117 | struct extent_buffer *leaf; |
39279cc3 CM |
4118 | int ret; |
4119 | ||
4120 | path = btrfs_alloc_path(); | |
16cdcec7 MX |
4121 | if (!path) |
4122 | return -ENOMEM; | |
4123 | ||
dfeb9e7c | 4124 | ret = btrfs_lookup_inode(trans, root, path, &inode->location, 1); |
39279cc3 CM |
4125 | if (ret) { |
4126 | if (ret > 0) | |
4127 | ret = -ENOENT; | |
4128 | goto failed; | |
4129 | } | |
4130 | ||
5f39d397 CM |
4131 | leaf = path->nodes[0]; |
4132 | inode_item = btrfs_item_ptr(leaf, path->slots[0], | |
16cdcec7 | 4133 | struct btrfs_inode_item); |
39279cc3 | 4134 | |
dfeb9e7c | 4135 | fill_inode_item(trans, leaf, inode_item, &inode->vfs_inode); |
5f39d397 | 4136 | btrfs_mark_buffer_dirty(leaf); |
dfeb9e7c | 4137 | btrfs_set_inode_last_trans(trans, inode); |
39279cc3 CM |
4138 | ret = 0; |
4139 | failed: | |
39279cc3 CM |
4140 | btrfs_free_path(path); |
4141 | return ret; | |
4142 | } | |
4143 | ||
2115133f CM |
4144 | /* |
4145 | * copy everything in the in-memory inode into the btree. | |
4146 | */ | |
4147 | noinline int btrfs_update_inode(struct btrfs_trans_handle *trans, | |
9a56fcd1 NB |
4148 | struct btrfs_root *root, |
4149 | struct btrfs_inode *inode) | |
2115133f | 4150 | { |
0b246afa | 4151 | struct btrfs_fs_info *fs_info = root->fs_info; |
2115133f CM |
4152 | int ret; |
4153 | ||
4154 | /* | |
4155 | * If the inode is a free space inode, we can deadlock during commit | |
4156 | * if we put it into the delayed code. | |
4157 | * | |
4158 | * The data relocation inode should also be directly updated | |
4159 | * without delay | |
4160 | */ | |
9a56fcd1 | 4161 | if (!btrfs_is_free_space_inode(inode) |
37f00a6d | 4162 | && !btrfs_is_data_reloc_root(root) |
0b246afa | 4163 | && !test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) { |
8ea05e3a AB |
4164 | btrfs_update_root_times(trans, root); |
4165 | ||
9a56fcd1 | 4166 | ret = btrfs_delayed_update_inode(trans, root, inode); |
2115133f | 4167 | if (!ret) |
9a56fcd1 | 4168 | btrfs_set_inode_last_trans(trans, inode); |
2115133f CM |
4169 | return ret; |
4170 | } | |
4171 | ||
9a56fcd1 | 4172 | return btrfs_update_inode_item(trans, root, inode); |
2115133f CM |
4173 | } |
4174 | ||
729f7961 NB |
4175 | int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans, |
4176 | struct btrfs_root *root, struct btrfs_inode *inode) | |
2115133f CM |
4177 | { |
4178 | int ret; | |
4179 | ||
729f7961 | 4180 | ret = btrfs_update_inode(trans, root, inode); |
2115133f | 4181 | if (ret == -ENOSPC) |
729f7961 | 4182 | return btrfs_update_inode_item(trans, root, inode); |
2115133f CM |
4183 | return ret; |
4184 | } | |
4185 | ||
d352ac68 CM |
4186 | /* |
4187 | * unlink helper that gets used here in inode.c and in the tree logging | |
4188 | * recovery code. It remove a link in a directory with a given name, and | |
4189 | * also drops the back refs in the inode to the directory | |
4190 | */ | |
92986796 | 4191 | static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans, |
4ec5934e NB |
4192 | struct btrfs_inode *dir, |
4193 | struct btrfs_inode *inode, | |
6db75318 | 4194 | const struct fscrypt_str *name, |
88d2beec | 4195 | struct btrfs_rename_ctx *rename_ctx) |
39279cc3 | 4196 | { |
4467af88 | 4197 | struct btrfs_root *root = dir->root; |
0b246afa | 4198 | struct btrfs_fs_info *fs_info = root->fs_info; |
39279cc3 | 4199 | struct btrfs_path *path; |
39279cc3 | 4200 | int ret = 0; |
39279cc3 | 4201 | struct btrfs_dir_item *di; |
aec7477b | 4202 | u64 index; |
33345d01 LZ |
4203 | u64 ino = btrfs_ino(inode); |
4204 | u64 dir_ino = btrfs_ino(dir); | |
39279cc3 CM |
4205 | |
4206 | path = btrfs_alloc_path(); | |
54aa1f4d CM |
4207 | if (!path) { |
4208 | ret = -ENOMEM; | |
554233a6 | 4209 | goto out; |
54aa1f4d CM |
4210 | } |
4211 | ||
e43eec81 | 4212 | di = btrfs_lookup_dir_item(trans, root, path, dir_ino, name, -1); |
3cf5068f LB |
4213 | if (IS_ERR_OR_NULL(di)) { |
4214 | ret = di ? PTR_ERR(di) : -ENOENT; | |
39279cc3 CM |
4215 | goto err; |
4216 | } | |
39279cc3 | 4217 | ret = btrfs_delete_one_dir_name(trans, root, path, di); |
54aa1f4d CM |
4218 | if (ret) |
4219 | goto err; | |
b3b4aa74 | 4220 | btrfs_release_path(path); |
39279cc3 | 4221 | |
67de1176 MX |
4222 | /* |
4223 | * If we don't have dir index, we have to get it by looking up | |
4224 | * the inode ref, since we get the inode ref, remove it directly, | |
4225 | * it is unnecessary to do delayed deletion. | |
4226 | * | |
4227 | * But if we have dir index, needn't search inode ref to get it. | |
4228 | * Since the inode ref is close to the inode item, it is better | |
4229 | * that we delay to delete it, and just do this deletion when | |
4230 | * we update the inode item. | |
4231 | */ | |
4ec5934e | 4232 | if (inode->dir_index) { |
67de1176 MX |
4233 | ret = btrfs_delayed_delete_inode_ref(inode); |
4234 | if (!ret) { | |
4ec5934e | 4235 | index = inode->dir_index; |
67de1176 MX |
4236 | goto skip_backref; |
4237 | } | |
4238 | } | |
4239 | ||
e43eec81 | 4240 | ret = btrfs_del_inode_ref(trans, root, name, ino, dir_ino, &index); |
aec7477b | 4241 | if (ret) { |
0b246afa | 4242 | btrfs_info(fs_info, |
c2cf52eb | 4243 | "failed to delete reference to %.*s, inode %llu parent %llu", |
e43eec81 | 4244 | name->len, name->name, ino, dir_ino); |
66642832 | 4245 | btrfs_abort_transaction(trans, ret); |
aec7477b JB |
4246 | goto err; |
4247 | } | |
67de1176 | 4248 | skip_backref: |
88d2beec FM |
4249 | if (rename_ctx) |
4250 | rename_ctx->index = index; | |
4251 | ||
9add2945 | 4252 | ret = btrfs_delete_delayed_dir_index(trans, dir, index); |
79787eaa | 4253 | if (ret) { |
66642832 | 4254 | btrfs_abort_transaction(trans, ret); |
39279cc3 | 4255 | goto err; |
79787eaa | 4256 | } |
39279cc3 | 4257 | |
259c4b96 FM |
4258 | /* |
4259 | * If we are in a rename context, we don't need to update anything in the | |
4260 | * log. That will be done later during the rename by btrfs_log_new_name(). | |
143823cf | 4261 | * Besides that, doing it here would only cause extra unnecessary btree |
259c4b96 FM |
4262 | * operations on the log tree, increasing latency for applications. |
4263 | */ | |
4264 | if (!rename_ctx) { | |
e43eec81 STD |
4265 | btrfs_del_inode_ref_in_log(trans, root, name, inode, dir_ino); |
4266 | btrfs_del_dir_entries_in_log(trans, root, name, dir, index); | |
259c4b96 | 4267 | } |
63611e73 JB |
4268 | |
4269 | /* | |
4270 | * If we have a pending delayed iput we could end up with the final iput | |
4271 | * being run in btrfs-cleaner context. If we have enough of these built | |
4272 | * up we can end up burning a lot of time in btrfs-cleaner without any | |
4273 | * way to throttle the unlinks. Since we're currently holding a ref on | |
4274 | * the inode we can run the delayed iput here without any issues as the | |
4275 | * final iput won't be done until after we drop the ref we're currently | |
4276 | * holding. | |
4277 | */ | |
4278 | btrfs_run_delayed_iput(fs_info, inode); | |
39279cc3 CM |
4279 | err: |
4280 | btrfs_free_path(path); | |
e02119d5 CM |
4281 | if (ret) |
4282 | goto out; | |
4283 | ||
e43eec81 | 4284 | btrfs_i_size_write(dir, dir->vfs_inode.i_size - name->len * 2); |
4ec5934e NB |
4285 | inode_inc_iversion(&inode->vfs_inode); |
4286 | inode_inc_iversion(&dir->vfs_inode); | |
c1867eb3 DS |
4287 | inode->vfs_inode.i_ctime = current_time(&inode->vfs_inode); |
4288 | dir->vfs_inode.i_mtime = inode->vfs_inode.i_ctime; | |
4289 | dir->vfs_inode.i_ctime = inode->vfs_inode.i_ctime; | |
9a56fcd1 | 4290 | ret = btrfs_update_inode(trans, root, dir); |
e02119d5 | 4291 | out: |
39279cc3 CM |
4292 | return ret; |
4293 | } | |
4294 | ||
92986796 | 4295 | int btrfs_unlink_inode(struct btrfs_trans_handle *trans, |
4ec5934e | 4296 | struct btrfs_inode *dir, struct btrfs_inode *inode, |
6db75318 | 4297 | const struct fscrypt_str *name) |
92986796 AV |
4298 | { |
4299 | int ret; | |
e43eec81 STD |
4300 | |
4301 | ret = __btrfs_unlink_inode(trans, dir, inode, name, NULL); | |
92986796 | 4302 | if (!ret) { |
4ec5934e | 4303 | drop_nlink(&inode->vfs_inode); |
4467af88 | 4304 | ret = btrfs_update_inode(trans, inode->root, inode); |
92986796 AV |
4305 | } |
4306 | return ret; | |
4307 | } | |
39279cc3 | 4308 | |
a22285a6 YZ |
4309 | /* |
4310 | * helper to start transaction for unlink and rmdir. | |
4311 | * | |
d52be818 JB |
4312 | * unlink and rmdir are special in btrfs, they do not always free space, so |
4313 | * if we cannot make our reservations the normal way try and see if there is | |
4314 | * plenty of slack room in the global reserve to migrate, otherwise we cannot | |
4315 | * allow the unlink to occur. | |
a22285a6 | 4316 | */ |
e569b1d5 | 4317 | static struct btrfs_trans_handle *__unlink_start_trans(struct btrfs_inode *dir) |
4df27c4d | 4318 | { |
e569b1d5 | 4319 | struct btrfs_root *root = dir->root; |
4df27c4d | 4320 | |
5630e2bc FM |
4321 | return btrfs_start_transaction_fallback_global_rsv(root, |
4322 | BTRFS_UNLINK_METADATA_UNITS); | |
a22285a6 YZ |
4323 | } |
4324 | ||
4325 | static int btrfs_unlink(struct inode *dir, struct dentry *dentry) | |
4326 | { | |
a22285a6 | 4327 | struct btrfs_trans_handle *trans; |
2b0143b5 | 4328 | struct inode *inode = d_inode(dentry); |
a22285a6 | 4329 | int ret; |
ab3c5c18 | 4330 | struct fscrypt_name fname; |
a22285a6 | 4331 | |
ab3c5c18 STD |
4332 | ret = fscrypt_setup_filename(dir, &dentry->d_name, 1, &fname); |
4333 | if (ret) | |
4334 | return ret; | |
ab3c5c18 STD |
4335 | |
4336 | /* This needs to handle no-key deletions later on */ | |
a22285a6 | 4337 | |
e569b1d5 | 4338 | trans = __unlink_start_trans(BTRFS_I(dir)); |
ab3c5c18 STD |
4339 | if (IS_ERR(trans)) { |
4340 | ret = PTR_ERR(trans); | |
4341 | goto fscrypt_free; | |
4342 | } | |
5f39d397 | 4343 | |
4ec5934e | 4344 | btrfs_record_unlink_dir(trans, BTRFS_I(dir), BTRFS_I(d_inode(dentry)), |
59fcf388 | 4345 | false); |
12fcfd22 | 4346 | |
e43eec81 | 4347 | ret = btrfs_unlink_inode(trans, BTRFS_I(dir), BTRFS_I(d_inode(dentry)), |
6db75318 | 4348 | &fname.disk_name); |
b532402e | 4349 | if (ret) |
ab3c5c18 | 4350 | goto end_trans; |
7b128766 | 4351 | |
a22285a6 | 4352 | if (inode->i_nlink == 0) { |
73f2e545 | 4353 | ret = btrfs_orphan_add(trans, BTRFS_I(inode)); |
b532402e | 4354 | if (ret) |
ab3c5c18 | 4355 | goto end_trans; |
a22285a6 | 4356 | } |
7b128766 | 4357 | |
ab3c5c18 | 4358 | end_trans: |
3a45bb20 | 4359 | btrfs_end_transaction(trans); |
4467af88 | 4360 | btrfs_btree_balance_dirty(BTRFS_I(dir)->root->fs_info); |
ab3c5c18 STD |
4361 | fscrypt_free: |
4362 | fscrypt_free_filename(&fname); | |
39279cc3 CM |
4363 | return ret; |
4364 | } | |
4365 | ||
f60a2364 | 4366 | static int btrfs_unlink_subvol(struct btrfs_trans_handle *trans, |
5b7544cb | 4367 | struct btrfs_inode *dir, struct dentry *dentry) |
4df27c4d | 4368 | { |
5b7544cb | 4369 | struct btrfs_root *root = dir->root; |
045d3967 | 4370 | struct btrfs_inode *inode = BTRFS_I(d_inode(dentry)); |
4df27c4d YZ |
4371 | struct btrfs_path *path; |
4372 | struct extent_buffer *leaf; | |
4373 | struct btrfs_dir_item *di; | |
4374 | struct btrfs_key key; | |
4375 | u64 index; | |
4376 | int ret; | |
045d3967 | 4377 | u64 objectid; |
5b7544cb | 4378 | u64 dir_ino = btrfs_ino(dir); |
ab3c5c18 STD |
4379 | struct fscrypt_name fname; |
4380 | ||
5b7544cb | 4381 | ret = fscrypt_setup_filename(&dir->vfs_inode, &dentry->d_name, 1, &fname); |
ab3c5c18 STD |
4382 | if (ret) |
4383 | return ret; | |
ab3c5c18 STD |
4384 | |
4385 | /* This needs to handle no-key deletions later on */ | |
4df27c4d | 4386 | |
045d3967 JB |
4387 | if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID) { |
4388 | objectid = inode->root->root_key.objectid; | |
4389 | } else if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) { | |
4390 | objectid = inode->location.objectid; | |
4391 | } else { | |
4392 | WARN_ON(1); | |
ab3c5c18 | 4393 | fscrypt_free_filename(&fname); |
045d3967 JB |
4394 | return -EINVAL; |
4395 | } | |
4396 | ||
4df27c4d | 4397 | path = btrfs_alloc_path(); |
ab3c5c18 STD |
4398 | if (!path) { |
4399 | ret = -ENOMEM; | |
4400 | goto out; | |
4401 | } | |
4df27c4d | 4402 | |
33345d01 | 4403 | di = btrfs_lookup_dir_item(trans, root, path, dir_ino, |
6db75318 | 4404 | &fname.disk_name, -1); |
79787eaa | 4405 | if (IS_ERR_OR_NULL(di)) { |
3cf5068f | 4406 | ret = di ? PTR_ERR(di) : -ENOENT; |
79787eaa JM |
4407 | goto out; |
4408 | } | |
4df27c4d YZ |
4409 | |
4410 | leaf = path->nodes[0]; | |
4411 | btrfs_dir_item_key_to_cpu(leaf, di, &key); | |
4412 | WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid); | |
4413 | ret = btrfs_delete_one_dir_name(trans, root, path, di); | |
79787eaa | 4414 | if (ret) { |
66642832 | 4415 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
4416 | goto out; |
4417 | } | |
b3b4aa74 | 4418 | btrfs_release_path(path); |
4df27c4d | 4419 | |
d49d3287 JB |
4420 | /* |
4421 | * This is a placeholder inode for a subvolume we didn't have a | |
4422 | * reference to at the time of the snapshot creation. In the meantime | |
4423 | * we could have renamed the real subvol link into our snapshot, so | |
1a9fd417 | 4424 | * depending on btrfs_del_root_ref to return -ENOENT here is incorrect. |
d49d3287 JB |
4425 | * Instead simply lookup the dir_index_item for this entry so we can |
4426 | * remove it. Otherwise we know we have a ref to the root and we can | |
4427 | * call btrfs_del_root_ref, and it _shouldn't_ fail. | |
4428 | */ | |
4429 | if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) { | |
6db75318 | 4430 | di = btrfs_search_dir_index_item(root, path, dir_ino, &fname.disk_name); |
79787eaa JM |
4431 | if (IS_ERR_OR_NULL(di)) { |
4432 | if (!di) | |
4433 | ret = -ENOENT; | |
4434 | else | |
4435 | ret = PTR_ERR(di); | |
66642832 | 4436 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
4437 | goto out; |
4438 | } | |
4df27c4d YZ |
4439 | |
4440 | leaf = path->nodes[0]; | |
4441 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
4df27c4d | 4442 | index = key.offset; |
d49d3287 JB |
4443 | btrfs_release_path(path); |
4444 | } else { | |
4445 | ret = btrfs_del_root_ref(trans, objectid, | |
4446 | root->root_key.objectid, dir_ino, | |
6db75318 | 4447 | &index, &fname.disk_name); |
d49d3287 JB |
4448 | if (ret) { |
4449 | btrfs_abort_transaction(trans, ret); | |
4450 | goto out; | |
4451 | } | |
4df27c4d YZ |
4452 | } |
4453 | ||
5b7544cb | 4454 | ret = btrfs_delete_delayed_dir_index(trans, dir, index); |
79787eaa | 4455 | if (ret) { |
66642832 | 4456 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
4457 | goto out; |
4458 | } | |
4df27c4d | 4459 | |
5b7544cb DS |
4460 | btrfs_i_size_write(dir, dir->vfs_inode.i_size - fname.disk_name.len * 2); |
4461 | inode_inc_iversion(&dir->vfs_inode); | |
4462 | dir->vfs_inode.i_mtime = current_time(&dir->vfs_inode); | |
4463 | dir->vfs_inode.i_ctime = dir->vfs_inode.i_mtime; | |
4464 | ret = btrfs_update_inode_fallback(trans, root, dir); | |
79787eaa | 4465 | if (ret) |
66642832 | 4466 | btrfs_abort_transaction(trans, ret); |
79787eaa | 4467 | out: |
71d7aed0 | 4468 | btrfs_free_path(path); |
ab3c5c18 | 4469 | fscrypt_free_filename(&fname); |
79787eaa | 4470 | return ret; |
4df27c4d YZ |
4471 | } |
4472 | ||
ec42f167 MT |
4473 | /* |
4474 | * Helper to check if the subvolume references other subvolumes or if it's | |
4475 | * default. | |
4476 | */ | |
f60a2364 | 4477 | static noinline int may_destroy_subvol(struct btrfs_root *root) |
ec42f167 MT |
4478 | { |
4479 | struct btrfs_fs_info *fs_info = root->fs_info; | |
4480 | struct btrfs_path *path; | |
4481 | struct btrfs_dir_item *di; | |
4482 | struct btrfs_key key; | |
6db75318 | 4483 | struct fscrypt_str name = FSTR_INIT("default", 7); |
ec42f167 MT |
4484 | u64 dir_id; |
4485 | int ret; | |
4486 | ||
4487 | path = btrfs_alloc_path(); | |
4488 | if (!path) | |
4489 | return -ENOMEM; | |
4490 | ||
4491 | /* Make sure this root isn't set as the default subvol */ | |
4492 | dir_id = btrfs_super_root_dir(fs_info->super_copy); | |
4493 | di = btrfs_lookup_dir_item(NULL, fs_info->tree_root, path, | |
e43eec81 | 4494 | dir_id, &name, 0); |
ec42f167 MT |
4495 | if (di && !IS_ERR(di)) { |
4496 | btrfs_dir_item_key_to_cpu(path->nodes[0], di, &key); | |
4497 | if (key.objectid == root->root_key.objectid) { | |
4498 | ret = -EPERM; | |
4499 | btrfs_err(fs_info, | |
4500 | "deleting default subvolume %llu is not allowed", | |
4501 | key.objectid); | |
4502 | goto out; | |
4503 | } | |
4504 | btrfs_release_path(path); | |
4505 | } | |
4506 | ||
4507 | key.objectid = root->root_key.objectid; | |
4508 | key.type = BTRFS_ROOT_REF_KEY; | |
4509 | key.offset = (u64)-1; | |
4510 | ||
4511 | ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0); | |
4512 | if (ret < 0) | |
4513 | goto out; | |
4514 | BUG_ON(ret == 0); | |
4515 | ||
4516 | ret = 0; | |
4517 | if (path->slots[0] > 0) { | |
4518 | path->slots[0]--; | |
4519 | btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]); | |
4520 | if (key.objectid == root->root_key.objectid && | |
4521 | key.type == BTRFS_ROOT_REF_KEY) | |
4522 | ret = -ENOTEMPTY; | |
4523 | } | |
4524 | out: | |
4525 | btrfs_free_path(path); | |
4526 | return ret; | |
4527 | } | |
4528 | ||
20a68004 NB |
4529 | /* Delete all dentries for inodes belonging to the root */ |
4530 | static void btrfs_prune_dentries(struct btrfs_root *root) | |
4531 | { | |
4532 | struct btrfs_fs_info *fs_info = root->fs_info; | |
4533 | struct rb_node *node; | |
4534 | struct rb_node *prev; | |
4535 | struct btrfs_inode *entry; | |
4536 | struct inode *inode; | |
4537 | u64 objectid = 0; | |
4538 | ||
84961539 | 4539 | if (!BTRFS_FS_ERROR(fs_info)) |
20a68004 NB |
4540 | WARN_ON(btrfs_root_refs(&root->root_item) != 0); |
4541 | ||
4542 | spin_lock(&root->inode_lock); | |
4543 | again: | |
4544 | node = root->inode_tree.rb_node; | |
4545 | prev = NULL; | |
4546 | while (node) { | |
4547 | prev = node; | |
4548 | entry = rb_entry(node, struct btrfs_inode, rb_node); | |
4549 | ||
37508515 | 4550 | if (objectid < btrfs_ino(entry)) |
20a68004 | 4551 | node = node->rb_left; |
37508515 | 4552 | else if (objectid > btrfs_ino(entry)) |
20a68004 NB |
4553 | node = node->rb_right; |
4554 | else | |
4555 | break; | |
4556 | } | |
4557 | if (!node) { | |
4558 | while (prev) { | |
4559 | entry = rb_entry(prev, struct btrfs_inode, rb_node); | |
37508515 | 4560 | if (objectid <= btrfs_ino(entry)) { |
20a68004 NB |
4561 | node = prev; |
4562 | break; | |
4563 | } | |
4564 | prev = rb_next(prev); | |
4565 | } | |
4566 | } | |
4567 | while (node) { | |
4568 | entry = rb_entry(node, struct btrfs_inode, rb_node); | |
37508515 | 4569 | objectid = btrfs_ino(entry) + 1; |
20a68004 NB |
4570 | inode = igrab(&entry->vfs_inode); |
4571 | if (inode) { | |
4572 | spin_unlock(&root->inode_lock); | |
4573 | if (atomic_read(&inode->i_count) > 1) | |
4574 | d_prune_aliases(inode); | |
4575 | /* | |
4576 | * btrfs_drop_inode will have it removed from the inode | |
4577 | * cache when its usage count hits zero. | |
4578 | */ | |
4579 | iput(inode); | |
4580 | cond_resched(); | |
4581 | spin_lock(&root->inode_lock); | |
4582 | goto again; | |
4583 | } | |
4584 | ||
4585 | if (cond_resched_lock(&root->inode_lock)) | |
4586 | goto again; | |
4587 | ||
4588 | node = rb_next(node); | |
4589 | } | |
4590 | spin_unlock(&root->inode_lock); | |
4591 | } | |
4592 | ||
3c4f91e2 | 4593 | int btrfs_delete_subvolume(struct btrfs_inode *dir, struct dentry *dentry) |
f60a2364 MT |
4594 | { |
4595 | struct btrfs_fs_info *fs_info = btrfs_sb(dentry->d_sb); | |
3c4f91e2 | 4596 | struct btrfs_root *root = dir->root; |
f60a2364 MT |
4597 | struct inode *inode = d_inode(dentry); |
4598 | struct btrfs_root *dest = BTRFS_I(inode)->root; | |
4599 | struct btrfs_trans_handle *trans; | |
4600 | struct btrfs_block_rsv block_rsv; | |
4601 | u64 root_flags; | |
f60a2364 | 4602 | int ret; |
f60a2364 MT |
4603 | |
4604 | /* | |
4605 | * Don't allow to delete a subvolume with send in progress. This is | |
4606 | * inside the inode lock so the error handling that has to drop the bit | |
4607 | * again is not run concurrently. | |
4608 | */ | |
4609 | spin_lock(&dest->root_item_lock); | |
a7176f74 | 4610 | if (dest->send_in_progress) { |
f60a2364 MT |
4611 | spin_unlock(&dest->root_item_lock); |
4612 | btrfs_warn(fs_info, | |
4613 | "attempt to delete subvolume %llu during send", | |
4614 | dest->root_key.objectid); | |
4615 | return -EPERM; | |
4616 | } | |
60021bd7 KH |
4617 | if (atomic_read(&dest->nr_swapfiles)) { |
4618 | spin_unlock(&dest->root_item_lock); | |
4619 | btrfs_warn(fs_info, | |
4620 | "attempt to delete subvolume %llu with active swapfile", | |
4621 | root->root_key.objectid); | |
4622 | return -EPERM; | |
4623 | } | |
a7176f74 LF |
4624 | root_flags = btrfs_root_flags(&dest->root_item); |
4625 | btrfs_set_root_flags(&dest->root_item, | |
4626 | root_flags | BTRFS_ROOT_SUBVOL_DEAD); | |
4627 | spin_unlock(&dest->root_item_lock); | |
f60a2364 MT |
4628 | |
4629 | down_write(&fs_info->subvol_sem); | |
4630 | ||
ee0d904f NB |
4631 | ret = may_destroy_subvol(dest); |
4632 | if (ret) | |
f60a2364 MT |
4633 | goto out_up_write; |
4634 | ||
4635 | btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP); | |
4636 | /* | |
4637 | * One for dir inode, | |
4638 | * two for dir entries, | |
4639 | * two for root ref/backref. | |
4640 | */ | |
ee0d904f NB |
4641 | ret = btrfs_subvolume_reserve_metadata(root, &block_rsv, 5, true); |
4642 | if (ret) | |
f60a2364 MT |
4643 | goto out_up_write; |
4644 | ||
4645 | trans = btrfs_start_transaction(root, 0); | |
4646 | if (IS_ERR(trans)) { | |
ee0d904f | 4647 | ret = PTR_ERR(trans); |
f60a2364 MT |
4648 | goto out_release; |
4649 | } | |
4650 | trans->block_rsv = &block_rsv; | |
4651 | trans->bytes_reserved = block_rsv.size; | |
4652 | ||
3c4f91e2 | 4653 | btrfs_record_snapshot_destroy(trans, dir); |
f60a2364 | 4654 | |
045d3967 | 4655 | ret = btrfs_unlink_subvol(trans, dir, dentry); |
f60a2364 | 4656 | if (ret) { |
f60a2364 MT |
4657 | btrfs_abort_transaction(trans, ret); |
4658 | goto out_end_trans; | |
4659 | } | |
4660 | ||
2731f518 JB |
4661 | ret = btrfs_record_root_in_trans(trans, dest); |
4662 | if (ret) { | |
4663 | btrfs_abort_transaction(trans, ret); | |
4664 | goto out_end_trans; | |
4665 | } | |
f60a2364 MT |
4666 | |
4667 | memset(&dest->root_item.drop_progress, 0, | |
4668 | sizeof(dest->root_item.drop_progress)); | |
c8422684 | 4669 | btrfs_set_root_drop_level(&dest->root_item, 0); |
f60a2364 MT |
4670 | btrfs_set_root_refs(&dest->root_item, 0); |
4671 | ||
4672 | if (!test_and_set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &dest->state)) { | |
4673 | ret = btrfs_insert_orphan_item(trans, | |
4674 | fs_info->tree_root, | |
4675 | dest->root_key.objectid); | |
4676 | if (ret) { | |
4677 | btrfs_abort_transaction(trans, ret); | |
f60a2364 MT |
4678 | goto out_end_trans; |
4679 | } | |
4680 | } | |
4681 | ||
d1957791 | 4682 | ret = btrfs_uuid_tree_remove(trans, dest->root_item.uuid, |
f60a2364 MT |
4683 | BTRFS_UUID_KEY_SUBVOL, |
4684 | dest->root_key.objectid); | |
4685 | if (ret && ret != -ENOENT) { | |
4686 | btrfs_abort_transaction(trans, ret); | |
f60a2364 MT |
4687 | goto out_end_trans; |
4688 | } | |
4689 | if (!btrfs_is_empty_uuid(dest->root_item.received_uuid)) { | |
d1957791 | 4690 | ret = btrfs_uuid_tree_remove(trans, |
f60a2364 MT |
4691 | dest->root_item.received_uuid, |
4692 | BTRFS_UUID_KEY_RECEIVED_SUBVOL, | |
4693 | dest->root_key.objectid); | |
4694 | if (ret && ret != -ENOENT) { | |
4695 | btrfs_abort_transaction(trans, ret); | |
f60a2364 MT |
4696 | goto out_end_trans; |
4697 | } | |
4698 | } | |
4699 | ||
082b6c97 QW |
4700 | free_anon_bdev(dest->anon_dev); |
4701 | dest->anon_dev = 0; | |
f60a2364 MT |
4702 | out_end_trans: |
4703 | trans->block_rsv = NULL; | |
4704 | trans->bytes_reserved = 0; | |
4705 | ret = btrfs_end_transaction(trans); | |
f60a2364 MT |
4706 | inode->i_flags |= S_DEAD; |
4707 | out_release: | |
e85fde51 | 4708 | btrfs_subvolume_release_metadata(root, &block_rsv); |
f60a2364 MT |
4709 | out_up_write: |
4710 | up_write(&fs_info->subvol_sem); | |
ee0d904f | 4711 | if (ret) { |
f60a2364 MT |
4712 | spin_lock(&dest->root_item_lock); |
4713 | root_flags = btrfs_root_flags(&dest->root_item); | |
4714 | btrfs_set_root_flags(&dest->root_item, | |
4715 | root_flags & ~BTRFS_ROOT_SUBVOL_DEAD); | |
4716 | spin_unlock(&dest->root_item_lock); | |
4717 | } else { | |
4718 | d_invalidate(dentry); | |
20a68004 | 4719 | btrfs_prune_dentries(dest); |
f60a2364 | 4720 | ASSERT(dest->send_in_progress == 0); |
f60a2364 MT |
4721 | } |
4722 | ||
ee0d904f | 4723 | return ret; |
f60a2364 MT |
4724 | } |
4725 | ||
39279cc3 CM |
4726 | static int btrfs_rmdir(struct inode *dir, struct dentry *dentry) |
4727 | { | |
2b0143b5 | 4728 | struct inode *inode = d_inode(dentry); |
813febdb | 4729 | struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info; |
1832a6d5 | 4730 | int err = 0; |
39279cc3 | 4731 | struct btrfs_trans_handle *trans; |
44f714da | 4732 | u64 last_unlink_trans; |
ab3c5c18 | 4733 | struct fscrypt_name fname; |
39279cc3 | 4734 | |
b3ae244e | 4735 | if (inode->i_size > BTRFS_EMPTY_DIR_SIZE) |
134d4512 | 4736 | return -ENOTEMPTY; |
813febdb JB |
4737 | if (btrfs_ino(BTRFS_I(inode)) == BTRFS_FIRST_FREE_OBJECTID) { |
4738 | if (unlikely(btrfs_fs_incompat(fs_info, EXTENT_TREE_V2))) { | |
4739 | btrfs_err(fs_info, | |
4740 | "extent tree v2 doesn't support snapshot deletion yet"); | |
4741 | return -EOPNOTSUPP; | |
4742 | } | |
3c4f91e2 | 4743 | return btrfs_delete_subvolume(BTRFS_I(dir), dentry); |
813febdb | 4744 | } |
134d4512 | 4745 | |
ab3c5c18 STD |
4746 | err = fscrypt_setup_filename(dir, &dentry->d_name, 1, &fname); |
4747 | if (err) | |
4748 | return err; | |
ab3c5c18 STD |
4749 | |
4750 | /* This needs to handle no-key deletions later on */ | |
4751 | ||
e569b1d5 | 4752 | trans = __unlink_start_trans(BTRFS_I(dir)); |
ab3c5c18 STD |
4753 | if (IS_ERR(trans)) { |
4754 | err = PTR_ERR(trans); | |
4755 | goto out_notrans; | |
4756 | } | |
5df6a9f6 | 4757 | |
4a0cc7ca | 4758 | if (unlikely(btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) { |
5b7544cb | 4759 | err = btrfs_unlink_subvol(trans, BTRFS_I(dir), dentry); |
4df27c4d YZ |
4760 | goto out; |
4761 | } | |
4762 | ||
73f2e545 | 4763 | err = btrfs_orphan_add(trans, BTRFS_I(inode)); |
7b128766 | 4764 | if (err) |
4df27c4d | 4765 | goto out; |
7b128766 | 4766 | |
44f714da FM |
4767 | last_unlink_trans = BTRFS_I(inode)->last_unlink_trans; |
4768 | ||
39279cc3 | 4769 | /* now the directory is empty */ |
e43eec81 | 4770 | err = btrfs_unlink_inode(trans, BTRFS_I(dir), BTRFS_I(d_inode(dentry)), |
6db75318 | 4771 | &fname.disk_name); |
44f714da | 4772 | if (!err) { |
6ef06d27 | 4773 | btrfs_i_size_write(BTRFS_I(inode), 0); |
44f714da FM |
4774 | /* |
4775 | * Propagate the last_unlink_trans value of the deleted dir to | |
4776 | * its parent directory. This is to prevent an unrecoverable | |
4777 | * log tree in the case we do something like this: | |
4778 | * 1) create dir foo | |
4779 | * 2) create snapshot under dir foo | |
4780 | * 3) delete the snapshot | |
4781 | * 4) rmdir foo | |
4782 | * 5) mkdir foo | |
4783 | * 6) fsync foo or some file inside foo | |
4784 | */ | |
4785 | if (last_unlink_trans >= trans->transid) | |
4786 | BTRFS_I(dir)->last_unlink_trans = last_unlink_trans; | |
4787 | } | |
4df27c4d | 4788 | out: |
3a45bb20 | 4789 | btrfs_end_transaction(trans); |
ab3c5c18 | 4790 | out_notrans: |
813febdb | 4791 | btrfs_btree_balance_dirty(fs_info); |
ab3c5c18 | 4792 | fscrypt_free_filename(&fname); |
3954401f | 4793 | |
39279cc3 CM |
4794 | return err; |
4795 | } | |
4796 | ||
39279cc3 | 4797 | /* |
9703fefe | 4798 | * btrfs_truncate_block - read, zero a chunk and write a block |
2aaa6655 JB |
4799 | * @inode - inode that we're zeroing |
4800 | * @from - the offset to start zeroing | |
4801 | * @len - the length to zero, 0 to zero the entire range respective to the | |
4802 | * offset | |
4803 | * @front - zero up to the offset instead of from the offset on | |
4804 | * | |
9703fefe | 4805 | * This will find the block for the "from" offset and cow the block and zero the |
2aaa6655 | 4806 | * part we want to zero. This is used with truncate and hole punching. |
39279cc3 | 4807 | */ |
217f42eb NB |
4808 | int btrfs_truncate_block(struct btrfs_inode *inode, loff_t from, loff_t len, |
4809 | int front) | |
39279cc3 | 4810 | { |
217f42eb NB |
4811 | struct btrfs_fs_info *fs_info = inode->root->fs_info; |
4812 | struct address_space *mapping = inode->vfs_inode.i_mapping; | |
4813 | struct extent_io_tree *io_tree = &inode->io_tree; | |
e6dcd2dc | 4814 | struct btrfs_ordered_extent *ordered; |
2ac55d41 | 4815 | struct extent_state *cached_state = NULL; |
364ecf36 | 4816 | struct extent_changeset *data_reserved = NULL; |
6d4572a9 | 4817 | bool only_release_metadata = false; |
0b246afa | 4818 | u32 blocksize = fs_info->sectorsize; |
09cbfeaf | 4819 | pgoff_t index = from >> PAGE_SHIFT; |
9703fefe | 4820 | unsigned offset = from & (blocksize - 1); |
39279cc3 | 4821 | struct page *page; |
3b16a4e3 | 4822 | gfp_t mask = btrfs_alloc_write_mask(mapping); |
6d4572a9 | 4823 | size_t write_bytes = blocksize; |
39279cc3 | 4824 | int ret = 0; |
9703fefe CR |
4825 | u64 block_start; |
4826 | u64 block_end; | |
39279cc3 | 4827 | |
b03ebd99 NB |
4828 | if (IS_ALIGNED(offset, blocksize) && |
4829 | (!len || IS_ALIGNED(len, blocksize))) | |
39279cc3 | 4830 | goto out; |
9703fefe | 4831 | |
8b62f87b JB |
4832 | block_start = round_down(from, blocksize); |
4833 | block_end = block_start + blocksize - 1; | |
4834 | ||
217f42eb | 4835 | ret = btrfs_check_data_free_space(inode, &data_reserved, block_start, |
1daedb1d | 4836 | blocksize, false); |
6d4572a9 | 4837 | if (ret < 0) { |
80f9d241 | 4838 | if (btrfs_check_nocow_lock(inode, block_start, &write_bytes, false) > 0) { |
6d4572a9 QW |
4839 | /* For nocow case, no need to reserve data space */ |
4840 | only_release_metadata = true; | |
4841 | } else { | |
4842 | goto out; | |
4843 | } | |
4844 | } | |
d4135134 | 4845 | ret = btrfs_delalloc_reserve_metadata(inode, blocksize, blocksize, false); |
6d4572a9 QW |
4846 | if (ret < 0) { |
4847 | if (!only_release_metadata) | |
217f42eb NB |
4848 | btrfs_free_reserved_data_space(inode, data_reserved, |
4849 | block_start, blocksize); | |
6d4572a9 QW |
4850 | goto out; |
4851 | } | |
211c17f5 | 4852 | again: |
3b16a4e3 | 4853 | page = find_or_create_page(mapping, index, mask); |
5d5e103a | 4854 | if (!page) { |
217f42eb NB |
4855 | btrfs_delalloc_release_space(inode, data_reserved, block_start, |
4856 | blocksize, true); | |
4857 | btrfs_delalloc_release_extents(inode, blocksize); | |
ac6a2b36 | 4858 | ret = -ENOMEM; |
39279cc3 | 4859 | goto out; |
5d5e103a | 4860 | } |
e6dcd2dc | 4861 | |
39279cc3 | 4862 | if (!PageUptodate(page)) { |
fb12489b | 4863 | ret = btrfs_read_folio(NULL, page_folio(page)); |
39279cc3 | 4864 | lock_page(page); |
211c17f5 CM |
4865 | if (page->mapping != mapping) { |
4866 | unlock_page(page); | |
09cbfeaf | 4867 | put_page(page); |
211c17f5 CM |
4868 | goto again; |
4869 | } | |
39279cc3 CM |
4870 | if (!PageUptodate(page)) { |
4871 | ret = -EIO; | |
89642229 | 4872 | goto out_unlock; |
39279cc3 CM |
4873 | } |
4874 | } | |
17b17fcd JB |
4875 | |
4876 | /* | |
4877 | * We unlock the page after the io is completed and then re-lock it | |
4878 | * above. release_folio() could have come in between that and cleared | |
4879 | * PagePrivate(), but left the page in the mapping. Set the page mapped | |
4880 | * here to make sure it's properly set for the subpage stuff. | |
4881 | */ | |
4882 | ret = set_page_extent_mapped(page); | |
4883 | if (ret < 0) | |
4884 | goto out_unlock; | |
4885 | ||
211c17f5 | 4886 | wait_on_page_writeback(page); |
e6dcd2dc | 4887 | |
570eb97b | 4888 | lock_extent(io_tree, block_start, block_end, &cached_state); |
e6dcd2dc | 4889 | |
217f42eb | 4890 | ordered = btrfs_lookup_ordered_extent(inode, block_start); |
e6dcd2dc | 4891 | if (ordered) { |
570eb97b | 4892 | unlock_extent(io_tree, block_start, block_end, &cached_state); |
e6dcd2dc | 4893 | unlock_page(page); |
09cbfeaf | 4894 | put_page(page); |
36d45567 | 4895 | btrfs_start_ordered_extent(ordered); |
e6dcd2dc CM |
4896 | btrfs_put_ordered_extent(ordered); |
4897 | goto again; | |
4898 | } | |
4899 | ||
217f42eb | 4900 | clear_extent_bit(&inode->io_tree, block_start, block_end, |
e182163d | 4901 | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, |
bd015294 | 4902 | &cached_state); |
5d5e103a | 4903 | |
217f42eb | 4904 | ret = btrfs_set_extent_delalloc(inode, block_start, block_end, 0, |
330a5827 | 4905 | &cached_state); |
9ed74f2d | 4906 | if (ret) { |
570eb97b | 4907 | unlock_extent(io_tree, block_start, block_end, &cached_state); |
9ed74f2d JB |
4908 | goto out_unlock; |
4909 | } | |
4910 | ||
9703fefe | 4911 | if (offset != blocksize) { |
2aaa6655 | 4912 | if (!len) |
9703fefe | 4913 | len = blocksize - offset; |
2aaa6655 | 4914 | if (front) |
d048b9c2 IW |
4915 | memzero_page(page, (block_start - page_offset(page)), |
4916 | offset); | |
2aaa6655 | 4917 | else |
d048b9c2 IW |
4918 | memzero_page(page, (block_start - page_offset(page)) + offset, |
4919 | len); | |
e6dcd2dc | 4920 | } |
e4f94347 QW |
4921 | btrfs_page_clear_checked(fs_info, page, block_start, |
4922 | block_end + 1 - block_start); | |
6c9ac8be | 4923 | btrfs_page_set_dirty(fs_info, page, block_start, block_end + 1 - block_start); |
570eb97b | 4924 | unlock_extent(io_tree, block_start, block_end, &cached_state); |
39279cc3 | 4925 | |
6d4572a9 | 4926 | if (only_release_metadata) |
217f42eb | 4927 | set_extent_bit(&inode->io_tree, block_start, block_end, |
1d126800 | 4928 | EXTENT_NORESERVE, NULL); |
6d4572a9 | 4929 | |
89642229 | 4930 | out_unlock: |
6d4572a9 QW |
4931 | if (ret) { |
4932 | if (only_release_metadata) | |
217f42eb | 4933 | btrfs_delalloc_release_metadata(inode, blocksize, true); |
6d4572a9 | 4934 | else |
217f42eb | 4935 | btrfs_delalloc_release_space(inode, data_reserved, |
6d4572a9 QW |
4936 | block_start, blocksize, true); |
4937 | } | |
217f42eb | 4938 | btrfs_delalloc_release_extents(inode, blocksize); |
39279cc3 | 4939 | unlock_page(page); |
09cbfeaf | 4940 | put_page(page); |
39279cc3 | 4941 | out: |
6d4572a9 | 4942 | if (only_release_metadata) |
217f42eb | 4943 | btrfs_check_nocow_unlock(inode); |
364ecf36 | 4944 | extent_changeset_free(data_reserved); |
39279cc3 CM |
4945 | return ret; |
4946 | } | |
4947 | ||
a4ba6cc0 | 4948 | static int maybe_insert_hole(struct btrfs_root *root, struct btrfs_inode *inode, |
16e7549f JB |
4949 | u64 offset, u64 len) |
4950 | { | |
a4ba6cc0 | 4951 | struct btrfs_fs_info *fs_info = root->fs_info; |
16e7549f | 4952 | struct btrfs_trans_handle *trans; |
5893dfb9 | 4953 | struct btrfs_drop_extents_args drop_args = { 0 }; |
16e7549f JB |
4954 | int ret; |
4955 | ||
4956 | /* | |
cceaa89f FM |
4957 | * If NO_HOLES is enabled, we don't need to do anything. |
4958 | * Later, up in the call chain, either btrfs_set_inode_last_sub_trans() | |
4959 | * or btrfs_update_inode() will be called, which guarantee that the next | |
4960 | * fsync will know this inode was changed and needs to be logged. | |
16e7549f | 4961 | */ |
cceaa89f | 4962 | if (btrfs_fs_incompat(fs_info, NO_HOLES)) |
16e7549f | 4963 | return 0; |
16e7549f JB |
4964 | |
4965 | /* | |
4966 | * 1 - for the one we're dropping | |
4967 | * 1 - for the one we're adding | |
4968 | * 1 - for updating the inode. | |
4969 | */ | |
4970 | trans = btrfs_start_transaction(root, 3); | |
4971 | if (IS_ERR(trans)) | |
4972 | return PTR_ERR(trans); | |
4973 | ||
5893dfb9 FM |
4974 | drop_args.start = offset; |
4975 | drop_args.end = offset + len; | |
4976 | drop_args.drop_cache = true; | |
4977 | ||
a4ba6cc0 | 4978 | ret = btrfs_drop_extents(trans, root, inode, &drop_args); |
16e7549f | 4979 | if (ret) { |
66642832 | 4980 | btrfs_abort_transaction(trans, ret); |
3a45bb20 | 4981 | btrfs_end_transaction(trans); |
16e7549f JB |
4982 | return ret; |
4983 | } | |
4984 | ||
d1f68ba0 | 4985 | ret = btrfs_insert_hole_extent(trans, root, btrfs_ino(inode), offset, len); |
2766ff61 | 4986 | if (ret) { |
66642832 | 4987 | btrfs_abort_transaction(trans, ret); |
2766ff61 | 4988 | } else { |
a4ba6cc0 NB |
4989 | btrfs_update_inode_bytes(inode, 0, drop_args.bytes_found); |
4990 | btrfs_update_inode(trans, root, inode); | |
2766ff61 | 4991 | } |
3a45bb20 | 4992 | btrfs_end_transaction(trans); |
16e7549f JB |
4993 | return ret; |
4994 | } | |
4995 | ||
695a0d0d JB |
4996 | /* |
4997 | * This function puts in dummy file extents for the area we're creating a hole | |
4998 | * for. So if we are truncating this file to a larger size we need to insert | |
4999 | * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for | |
5000 | * the range between oldsize and size | |
5001 | */ | |
b06359a3 | 5002 | int btrfs_cont_expand(struct btrfs_inode *inode, loff_t oldsize, loff_t size) |
39279cc3 | 5003 | { |
b06359a3 NB |
5004 | struct btrfs_root *root = inode->root; |
5005 | struct btrfs_fs_info *fs_info = root->fs_info; | |
5006 | struct extent_io_tree *io_tree = &inode->io_tree; | |
a22285a6 | 5007 | struct extent_map *em = NULL; |
2ac55d41 | 5008 | struct extent_state *cached_state = NULL; |
0b246afa JM |
5009 | u64 hole_start = ALIGN(oldsize, fs_info->sectorsize); |
5010 | u64 block_end = ALIGN(size, fs_info->sectorsize); | |
9036c102 YZ |
5011 | u64 last_byte; |
5012 | u64 cur_offset; | |
5013 | u64 hole_size; | |
9ed74f2d | 5014 | int err = 0; |
39279cc3 | 5015 | |
a71754fc | 5016 | /* |
9703fefe CR |
5017 | * If our size started in the middle of a block we need to zero out the |
5018 | * rest of the block before we expand the i_size, otherwise we could | |
a71754fc JB |
5019 | * expose stale data. |
5020 | */ | |
b06359a3 | 5021 | err = btrfs_truncate_block(inode, oldsize, 0, 0); |
a71754fc JB |
5022 | if (err) |
5023 | return err; | |
5024 | ||
9036c102 YZ |
5025 | if (size <= hole_start) |
5026 | return 0; | |
5027 | ||
b06359a3 NB |
5028 | btrfs_lock_and_flush_ordered_range(inode, hole_start, block_end - 1, |
5029 | &cached_state); | |
9036c102 YZ |
5030 | cur_offset = hole_start; |
5031 | while (1) { | |
b06359a3 | 5032 | em = btrfs_get_extent(inode, NULL, 0, cur_offset, |
39b07b5d | 5033 | block_end - cur_offset); |
79787eaa JM |
5034 | if (IS_ERR(em)) { |
5035 | err = PTR_ERR(em); | |
f2767956 | 5036 | em = NULL; |
79787eaa JM |
5037 | break; |
5038 | } | |
9036c102 | 5039 | last_byte = min(extent_map_end(em), block_end); |
0b246afa | 5040 | last_byte = ALIGN(last_byte, fs_info->sectorsize); |
9ddc959e JB |
5041 | hole_size = last_byte - cur_offset; |
5042 | ||
8082510e | 5043 | if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) { |
5dc562c5 | 5044 | struct extent_map *hole_em; |
9ed74f2d | 5045 | |
b06359a3 NB |
5046 | err = maybe_insert_hole(root, inode, cur_offset, |
5047 | hole_size); | |
16e7549f | 5048 | if (err) |
3893e33b | 5049 | break; |
9ddc959e | 5050 | |
b06359a3 | 5051 | err = btrfs_inode_set_file_extent_range(inode, |
9ddc959e JB |
5052 | cur_offset, hole_size); |
5053 | if (err) | |
5054 | break; | |
5055 | ||
5dc562c5 JB |
5056 | hole_em = alloc_extent_map(); |
5057 | if (!hole_em) { | |
a1ba4c08 FM |
5058 | btrfs_drop_extent_map_range(inode, cur_offset, |
5059 | cur_offset + hole_size - 1, | |
5060 | false); | |
23e3337f | 5061 | btrfs_set_inode_full_sync(inode); |
5dc562c5 JB |
5062 | goto next; |
5063 | } | |
5064 | hole_em->start = cur_offset; | |
5065 | hole_em->len = hole_size; | |
5066 | hole_em->orig_start = cur_offset; | |
8082510e | 5067 | |
5dc562c5 JB |
5068 | hole_em->block_start = EXTENT_MAP_HOLE; |
5069 | hole_em->block_len = 0; | |
b4939680 | 5070 | hole_em->orig_block_len = 0; |
cc95bef6 | 5071 | hole_em->ram_bytes = hole_size; |
5dc562c5 | 5072 | hole_em->compress_type = BTRFS_COMPRESS_NONE; |
0b246afa | 5073 | hole_em->generation = fs_info->generation; |
8082510e | 5074 | |
a1ba4c08 | 5075 | err = btrfs_replace_extent_map_range(inode, hole_em, true); |
5dc562c5 | 5076 | free_extent_map(hole_em); |
9ddc959e | 5077 | } else { |
b06359a3 | 5078 | err = btrfs_inode_set_file_extent_range(inode, |
9ddc959e JB |
5079 | cur_offset, hole_size); |
5080 | if (err) | |
5081 | break; | |
9036c102 | 5082 | } |
16e7549f | 5083 | next: |
9036c102 | 5084 | free_extent_map(em); |
a22285a6 | 5085 | em = NULL; |
9036c102 | 5086 | cur_offset = last_byte; |
8082510e | 5087 | if (cur_offset >= block_end) |
9036c102 YZ |
5088 | break; |
5089 | } | |
a22285a6 | 5090 | free_extent_map(em); |
570eb97b | 5091 | unlock_extent(io_tree, hole_start, block_end - 1, &cached_state); |
9036c102 YZ |
5092 | return err; |
5093 | } | |
39279cc3 | 5094 | |
3972f260 | 5095 | static int btrfs_setsize(struct inode *inode, struct iattr *attr) |
8082510e | 5096 | { |
f4a2f4c5 MX |
5097 | struct btrfs_root *root = BTRFS_I(inode)->root; |
5098 | struct btrfs_trans_handle *trans; | |
a41ad394 | 5099 | loff_t oldsize = i_size_read(inode); |
3972f260 ES |
5100 | loff_t newsize = attr->ia_size; |
5101 | int mask = attr->ia_valid; | |
8082510e YZ |
5102 | int ret; |
5103 | ||
3972f260 ES |
5104 | /* |
5105 | * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a | |
5106 | * special case where we need to update the times despite not having | |
5107 | * these flags set. For all other operations the VFS set these flags | |
5108 | * explicitly if it wants a timestamp update. | |
5109 | */ | |
dff6efc3 CH |
5110 | if (newsize != oldsize) { |
5111 | inode_inc_iversion(inode); | |
c1867eb3 DS |
5112 | if (!(mask & (ATTR_CTIME | ATTR_MTIME))) { |
5113 | inode->i_mtime = current_time(inode); | |
5114 | inode->i_ctime = inode->i_mtime; | |
5115 | } | |
dff6efc3 | 5116 | } |
3972f260 | 5117 | |
a41ad394 | 5118 | if (newsize > oldsize) { |
9ea24bbe | 5119 | /* |
ea14b57f | 5120 | * Don't do an expanding truncate while snapshotting is ongoing. |
9ea24bbe FM |
5121 | * This is to ensure the snapshot captures a fully consistent |
5122 | * state of this file - if the snapshot captures this expanding | |
5123 | * truncation, it must capture all writes that happened before | |
5124 | * this truncation. | |
5125 | */ | |
dcc3eb96 | 5126 | btrfs_drew_write_lock(&root->snapshot_lock); |
b06359a3 | 5127 | ret = btrfs_cont_expand(BTRFS_I(inode), oldsize, newsize); |
9ea24bbe | 5128 | if (ret) { |
dcc3eb96 | 5129 | btrfs_drew_write_unlock(&root->snapshot_lock); |
8082510e | 5130 | return ret; |
9ea24bbe | 5131 | } |
8082510e | 5132 | |
f4a2f4c5 | 5133 | trans = btrfs_start_transaction(root, 1); |
9ea24bbe | 5134 | if (IS_ERR(trans)) { |
dcc3eb96 | 5135 | btrfs_drew_write_unlock(&root->snapshot_lock); |
f4a2f4c5 | 5136 | return PTR_ERR(trans); |
9ea24bbe | 5137 | } |
f4a2f4c5 MX |
5138 | |
5139 | i_size_write(inode, newsize); | |
76aea537 | 5140 | btrfs_inode_safe_disk_i_size_write(BTRFS_I(inode), 0); |
27772b68 | 5141 | pagecache_isize_extended(inode, oldsize, newsize); |
9a56fcd1 | 5142 | ret = btrfs_update_inode(trans, root, BTRFS_I(inode)); |
dcc3eb96 | 5143 | btrfs_drew_write_unlock(&root->snapshot_lock); |
3a45bb20 | 5144 | btrfs_end_transaction(trans); |
a41ad394 | 5145 | } else { |
24c0a722 NA |
5146 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
5147 | ||
5148 | if (btrfs_is_zoned(fs_info)) { | |
5149 | ret = btrfs_wait_ordered_range(inode, | |
5150 | ALIGN(newsize, fs_info->sectorsize), | |
5151 | (u64)-1); | |
5152 | if (ret) | |
5153 | return ret; | |
5154 | } | |
8082510e | 5155 | |
a41ad394 JB |
5156 | /* |
5157 | * We're truncating a file that used to have good data down to | |
1fd4033d NB |
5158 | * zero. Make sure any new writes to the file get on disk |
5159 | * on close. | |
a41ad394 JB |
5160 | */ |
5161 | if (newsize == 0) | |
1fd4033d | 5162 | set_bit(BTRFS_INODE_FLUSH_ON_CLOSE, |
72ac3c0d | 5163 | &BTRFS_I(inode)->runtime_flags); |
8082510e | 5164 | |
a41ad394 | 5165 | truncate_setsize(inode, newsize); |
2e60a51e | 5166 | |
2e60a51e | 5167 | inode_dio_wait(inode); |
2e60a51e | 5168 | |
d9dcae67 | 5169 | ret = btrfs_truncate(BTRFS_I(inode), newsize == oldsize); |
7f4f6e0a JB |
5170 | if (ret && inode->i_nlink) { |
5171 | int err; | |
5172 | ||
5173 | /* | |
f7e9e8fc OS |
5174 | * Truncate failed, so fix up the in-memory size. We |
5175 | * adjusted disk_i_size down as we removed extents, so | |
5176 | * wait for disk_i_size to be stable and then update the | |
5177 | * in-memory size to match. | |
7f4f6e0a | 5178 | */ |
f7e9e8fc | 5179 | err = btrfs_wait_ordered_range(inode, 0, (u64)-1); |
7f4f6e0a | 5180 | if (err) |
f7e9e8fc OS |
5181 | return err; |
5182 | i_size_write(inode, BTRFS_I(inode)->disk_i_size); | |
7f4f6e0a | 5183 | } |
8082510e YZ |
5184 | } |
5185 | ||
a41ad394 | 5186 | return ret; |
8082510e YZ |
5187 | } |
5188 | ||
c1632a0f | 5189 | static int btrfs_setattr(struct mnt_idmap *idmap, struct dentry *dentry, |
549c7297 | 5190 | struct iattr *attr) |
9036c102 | 5191 | { |
2b0143b5 | 5192 | struct inode *inode = d_inode(dentry); |
b83cc969 | 5193 | struct btrfs_root *root = BTRFS_I(inode)->root; |
9036c102 | 5194 | int err; |
39279cc3 | 5195 | |
b83cc969 LZ |
5196 | if (btrfs_root_readonly(root)) |
5197 | return -EROFS; | |
5198 | ||
c1632a0f | 5199 | err = setattr_prepare(idmap, dentry, attr); |
9036c102 YZ |
5200 | if (err) |
5201 | return err; | |
2bf5a725 | 5202 | |
5a3f23d5 | 5203 | if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) { |
3972f260 | 5204 | err = btrfs_setsize(inode, attr); |
8082510e YZ |
5205 | if (err) |
5206 | return err; | |
39279cc3 | 5207 | } |
9036c102 | 5208 | |
1025774c | 5209 | if (attr->ia_valid) { |
c1632a0f | 5210 | setattr_copy(idmap, inode, attr); |
0c4d2d95 | 5211 | inode_inc_iversion(inode); |
7152b425 | 5212 | err = btrfs_dirty_inode(BTRFS_I(inode)); |
1025774c | 5213 | |
22c44fe6 | 5214 | if (!err && attr->ia_valid & ATTR_MODE) |
13e83a49 | 5215 | err = posix_acl_chmod(idmap, dentry, inode->i_mode); |
1025774c | 5216 | } |
33268eaf | 5217 | |
39279cc3 CM |
5218 | return err; |
5219 | } | |
61295eb8 | 5220 | |
131e404a | 5221 | /* |
895586eb MWO |
5222 | * While truncating the inode pages during eviction, we get the VFS |
5223 | * calling btrfs_invalidate_folio() against each folio of the inode. This | |
5224 | * is slow because the calls to btrfs_invalidate_folio() result in a | |
570eb97b | 5225 | * huge amount of calls to lock_extent() and clear_extent_bit(), |
895586eb MWO |
5226 | * which keep merging and splitting extent_state structures over and over, |
5227 | * wasting lots of time. | |
131e404a | 5228 | * |
895586eb MWO |
5229 | * Therefore if the inode is being evicted, let btrfs_invalidate_folio() |
5230 | * skip all those expensive operations on a per folio basis and do only | |
5231 | * the ordered io finishing, while we release here the extent_map and | |
5232 | * extent_state structures, without the excessive merging and splitting. | |
131e404a FDBM |
5233 | */ |
5234 | static void evict_inode_truncate_pages(struct inode *inode) | |
5235 | { | |
5236 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; | |
131e404a FDBM |
5237 | struct rb_node *node; |
5238 | ||
5239 | ASSERT(inode->i_state & I_FREEING); | |
91b0abe3 | 5240 | truncate_inode_pages_final(&inode->i_data); |
131e404a | 5241 | |
9c9d1b4f | 5242 | btrfs_drop_extent_map_range(BTRFS_I(inode), 0, (u64)-1, false); |
131e404a | 5243 | |
6ca07097 FM |
5244 | /* |
5245 | * Keep looping until we have no more ranges in the io tree. | |
ba206a02 MWO |
5246 | * We can have ongoing bios started by readahead that have |
5247 | * their endio callback (extent_io.c:end_bio_extent_readpage) | |
9c6429d9 FM |
5248 | * still in progress (unlocked the pages in the bio but did not yet |
5249 | * unlocked the ranges in the io tree). Therefore this means some | |
6ca07097 FM |
5250 | * ranges can still be locked and eviction started because before |
5251 | * submitting those bios, which are executed by a separate task (work | |
5252 | * queue kthread), inode references (inode->i_count) were not taken | |
5253 | * (which would be dropped in the end io callback of each bio). | |
5254 | * Therefore here we effectively end up waiting for those bios and | |
5255 | * anyone else holding locked ranges without having bumped the inode's | |
5256 | * reference count - if we don't do it, when they access the inode's | |
5257 | * io_tree to unlock a range it may be too late, leading to an | |
5258 | * use-after-free issue. | |
5259 | */ | |
131e404a FDBM |
5260 | spin_lock(&io_tree->lock); |
5261 | while (!RB_EMPTY_ROOT(&io_tree->state)) { | |
5262 | struct extent_state *state; | |
5263 | struct extent_state *cached_state = NULL; | |
6ca07097 FM |
5264 | u64 start; |
5265 | u64 end; | |
421f0922 | 5266 | unsigned state_flags; |
131e404a FDBM |
5267 | |
5268 | node = rb_first(&io_tree->state); | |
5269 | state = rb_entry(node, struct extent_state, rb_node); | |
6ca07097 FM |
5270 | start = state->start; |
5271 | end = state->end; | |
421f0922 | 5272 | state_flags = state->state; |
131e404a FDBM |
5273 | spin_unlock(&io_tree->lock); |
5274 | ||
570eb97b | 5275 | lock_extent(io_tree, start, end, &cached_state); |
b9d0b389 QW |
5276 | |
5277 | /* | |
5278 | * If still has DELALLOC flag, the extent didn't reach disk, | |
5279 | * and its reserved space won't be freed by delayed_ref. | |
5280 | * So we need to free its reserved space here. | |
895586eb | 5281 | * (Refer to comment in btrfs_invalidate_folio, case 2) |
b9d0b389 QW |
5282 | * |
5283 | * Note, end is the bytenr of last byte, so we need + 1 here. | |
5284 | */ | |
421f0922 | 5285 | if (state_flags & EXTENT_DELALLOC) |
8b8a979f NB |
5286 | btrfs_qgroup_free_data(BTRFS_I(inode), NULL, start, |
5287 | end - start + 1); | |
b9d0b389 | 5288 | |
6ca07097 | 5289 | clear_extent_bit(io_tree, start, end, |
bd015294 | 5290 | EXTENT_CLEAR_ALL_BITS | EXTENT_DO_ACCOUNTING, |
e182163d | 5291 | &cached_state); |
131e404a | 5292 | |
7064dd5c | 5293 | cond_resched(); |
131e404a FDBM |
5294 | spin_lock(&io_tree->lock); |
5295 | } | |
5296 | spin_unlock(&io_tree->lock); | |
5297 | } | |
5298 | ||
4b9d7b59 | 5299 | static struct btrfs_trans_handle *evict_refill_and_join(struct btrfs_root *root, |
ad80cf50 | 5300 | struct btrfs_block_rsv *rsv) |
4b9d7b59 OS |
5301 | { |
5302 | struct btrfs_fs_info *fs_info = root->fs_info; | |
d3984c90 | 5303 | struct btrfs_trans_handle *trans; |
b13d57db | 5304 | u64 delayed_refs_extra = btrfs_calc_delayed_ref_bytes(fs_info, 1); |
d3984c90 | 5305 | int ret; |
4b9d7b59 | 5306 | |
d3984c90 JB |
5307 | /* |
5308 | * Eviction should be taking place at some place safe because of our | |
5309 | * delayed iputs. However the normal flushing code will run delayed | |
5310 | * iputs, so we cannot use FLUSH_ALL otherwise we'll deadlock. | |
5311 | * | |
5312 | * We reserve the delayed_refs_extra here again because we can't use | |
5313 | * btrfs_start_transaction(root, 0) for the same deadlocky reason as | |
5314 | * above. We reserve our extra bit here because we generate a ton of | |
5315 | * delayed refs activity by truncating. | |
5316 | * | |
ee6adbfd JB |
5317 | * BTRFS_RESERVE_FLUSH_EVICT will steal from the global_rsv if it can, |
5318 | * if we fail to make this reservation we can re-try without the | |
5319 | * delayed_refs_extra so we can make some forward progress. | |
d3984c90 | 5320 | */ |
9270501c | 5321 | ret = btrfs_block_rsv_refill(fs_info, rsv, rsv->size + delayed_refs_extra, |
d3984c90 JB |
5322 | BTRFS_RESERVE_FLUSH_EVICT); |
5323 | if (ret) { | |
9270501c | 5324 | ret = btrfs_block_rsv_refill(fs_info, rsv, rsv->size, |
ee6adbfd JB |
5325 | BTRFS_RESERVE_FLUSH_EVICT); |
5326 | if (ret) { | |
d3984c90 JB |
5327 | btrfs_warn(fs_info, |
5328 | "could not allocate space for delete; will truncate on mount"); | |
5329 | return ERR_PTR(-ENOSPC); | |
5330 | } | |
5331 | delayed_refs_extra = 0; | |
5332 | } | |
4b9d7b59 | 5333 | |
d3984c90 JB |
5334 | trans = btrfs_join_transaction(root); |
5335 | if (IS_ERR(trans)) | |
5336 | return trans; | |
5337 | ||
5338 | if (delayed_refs_extra) { | |
5339 | trans->block_rsv = &fs_info->trans_block_rsv; | |
5340 | trans->bytes_reserved = delayed_refs_extra; | |
5341 | btrfs_block_rsv_migrate(rsv, trans->block_rsv, | |
4e0527de | 5342 | delayed_refs_extra, true); |
4b9d7b59 | 5343 | } |
d3984c90 | 5344 | return trans; |
4b9d7b59 OS |
5345 | } |
5346 | ||
bd555975 | 5347 | void btrfs_evict_inode(struct inode *inode) |
39279cc3 | 5348 | { |
0b246afa | 5349 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
39279cc3 CM |
5350 | struct btrfs_trans_handle *trans; |
5351 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
b7b1167c | 5352 | struct btrfs_block_rsv *rsv = NULL; |
39279cc3 CM |
5353 | int ret; |
5354 | ||
1abe9b8a | 5355 | trace_btrfs_inode_evict(inode); |
5356 | ||
3d48d981 | 5357 | if (!root) { |
14605409 | 5358 | fsverity_cleanup_inode(inode); |
e8f1bc14 | 5359 | clear_inode(inode); |
3d48d981 NB |
5360 | return; |
5361 | } | |
5362 | ||
131e404a FDBM |
5363 | evict_inode_truncate_pages(inode); |
5364 | ||
69e9c6c6 SB |
5365 | if (inode->i_nlink && |
5366 | ((btrfs_root_refs(&root->root_item) != 0 && | |
5367 | root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) || | |
70ddc553 | 5368 | btrfs_is_free_space_inode(BTRFS_I(inode)))) |
b7b1167c | 5369 | goto out; |
bd555975 | 5370 | |
27919067 | 5371 | if (is_bad_inode(inode)) |
b7b1167c | 5372 | goto out; |
5f39d397 | 5373 | |
7b40b695 | 5374 | if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) |
b7b1167c | 5375 | goto out; |
c71bf099 | 5376 | |
76dda93c | 5377 | if (inode->i_nlink > 0) { |
69e9c6c6 SB |
5378 | BUG_ON(btrfs_root_refs(&root->root_item) != 0 && |
5379 | root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID); | |
b7b1167c | 5380 | goto out; |
76dda93c YZ |
5381 | } |
5382 | ||
2adc75d6 JB |
5383 | /* |
5384 | * This makes sure the inode item in tree is uptodate and the space for | |
5385 | * the inode update is released. | |
5386 | */ | |
aa79021f | 5387 | ret = btrfs_commit_inode_delayed_inode(BTRFS_I(inode)); |
27919067 | 5388 | if (ret) |
b7b1167c | 5389 | goto out; |
0e8c36a9 | 5390 | |
2adc75d6 JB |
5391 | /* |
5392 | * This drops any pending insert or delete operations we have for this | |
5393 | * inode. We could have a delayed dir index deletion queued up, but | |
5394 | * we're removing the inode completely so that'll be taken care of in | |
5395 | * the truncate. | |
5396 | */ | |
5397 | btrfs_kill_delayed_inode_items(BTRFS_I(inode)); | |
5398 | ||
2ff7e61e | 5399 | rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP); |
27919067 | 5400 | if (!rsv) |
b7b1167c | 5401 | goto out; |
2bd36e7b | 5402 | rsv->size = btrfs_calc_metadata_size(fs_info, 1); |
710d5921 | 5403 | rsv->failfast = true; |
4289a667 | 5404 | |
6ef06d27 | 5405 | btrfs_i_size_write(BTRFS_I(inode), 0); |
5f39d397 | 5406 | |
8082510e | 5407 | while (1) { |
d9ac19c3 | 5408 | struct btrfs_truncate_control control = { |
71d18b53 | 5409 | .inode = BTRFS_I(inode), |
487e81d2 | 5410 | .ino = btrfs_ino(BTRFS_I(inode)), |
d9ac19c3 JB |
5411 | .new_size = 0, |
5412 | .min_type = 0, | |
5413 | }; | |
5414 | ||
ad80cf50 | 5415 | trans = evict_refill_and_join(root, rsv); |
27919067 | 5416 | if (IS_ERR(trans)) |
b7b1167c | 5417 | goto out; |
7b128766 | 5418 | |
4289a667 JB |
5419 | trans->block_rsv = rsv; |
5420 | ||
71d18b53 | 5421 | ret = btrfs_truncate_inode_items(trans, root, &control); |
27919067 OS |
5422 | trans->block_rsv = &fs_info->trans_block_rsv; |
5423 | btrfs_end_transaction(trans); | |
afa4b0af FM |
5424 | /* |
5425 | * We have not added new delayed items for our inode after we | |
5426 | * have flushed its delayed items, so no need to throttle on | |
5427 | * delayed items. However we have modified extent buffers. | |
5428 | */ | |
5429 | btrfs_btree_balance_dirty_nodelay(fs_info); | |
27919067 | 5430 | if (ret && ret != -ENOSPC && ret != -EAGAIN) |
b7b1167c | 5431 | goto out; |
27919067 | 5432 | else if (!ret) |
8082510e | 5433 | break; |
8082510e | 5434 | } |
5f39d397 | 5435 | |
4ef31a45 | 5436 | /* |
27919067 OS |
5437 | * Errors here aren't a big deal, it just means we leave orphan items in |
5438 | * the tree. They will be cleaned up on the next mount. If the inode | |
5439 | * number gets reused, cleanup deletes the orphan item without doing | |
5440 | * anything, and unlink reuses the existing orphan item. | |
5441 | * | |
5442 | * If it turns out that we are dropping too many of these, we might want | |
5443 | * to add a mechanism for retrying these after a commit. | |
4ef31a45 | 5444 | */ |
ad80cf50 | 5445 | trans = evict_refill_and_join(root, rsv); |
27919067 OS |
5446 | if (!IS_ERR(trans)) { |
5447 | trans->block_rsv = rsv; | |
5448 | btrfs_orphan_del(trans, BTRFS_I(inode)); | |
5449 | trans->block_rsv = &fs_info->trans_block_rsv; | |
5450 | btrfs_end_transaction(trans); | |
5451 | } | |
54aa1f4d | 5452 | |
b7b1167c | 5453 | out: |
27919067 | 5454 | btrfs_free_block_rsv(fs_info, rsv); |
27919067 OS |
5455 | /* |
5456 | * If we didn't successfully delete, the orphan item will still be in | |
5457 | * the tree and we'll retry on the next mount. Again, we might also want | |
5458 | * to retry these periodically in the future. | |
5459 | */ | |
f48d1cf5 | 5460 | btrfs_remove_delayed_node(BTRFS_I(inode)); |
14605409 | 5461 | fsverity_cleanup_inode(inode); |
dbd5768f | 5462 | clear_inode(inode); |
39279cc3 CM |
5463 | } |
5464 | ||
5465 | /* | |
6bf9e4bd QW |
5466 | * Return the key found in the dir entry in the location pointer, fill @type |
5467 | * with BTRFS_FT_*, and return 0. | |
5468 | * | |
005d6712 SY |
5469 | * If no dir entries were found, returns -ENOENT. |
5470 | * If found a corrupted location in dir entry, returns -EUCLEAN. | |
39279cc3 | 5471 | */ |
d1de429b | 5472 | static int btrfs_inode_by_name(struct btrfs_inode *dir, struct dentry *dentry, |
6bf9e4bd | 5473 | struct btrfs_key *location, u8 *type) |
39279cc3 | 5474 | { |
39279cc3 CM |
5475 | struct btrfs_dir_item *di; |
5476 | struct btrfs_path *path; | |
d1de429b | 5477 | struct btrfs_root *root = dir->root; |
0d9f7f3e | 5478 | int ret = 0; |
ab3c5c18 | 5479 | struct fscrypt_name fname; |
39279cc3 CM |
5480 | |
5481 | path = btrfs_alloc_path(); | |
d8926bb3 MF |
5482 | if (!path) |
5483 | return -ENOMEM; | |
3954401f | 5484 | |
d1de429b | 5485 | ret = fscrypt_setup_filename(&dir->vfs_inode, &dentry->d_name, 1, &fname); |
10a8857a | 5486 | if (ret < 0) |
ab3c5c18 | 5487 | goto out; |
10a8857a STD |
5488 | /* |
5489 | * fscrypt_setup_filename() should never return a positive value, but | |
5490 | * gcc on sparc/parisc thinks it can, so assert that doesn't happen. | |
5491 | */ | |
5492 | ASSERT(ret == 0); | |
ab3c5c18 | 5493 | |
ab3c5c18 STD |
5494 | /* This needs to handle no-key deletions later on */ |
5495 | ||
d1de429b | 5496 | di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), |
6db75318 | 5497 | &fname.disk_name, 0); |
3cf5068f LB |
5498 | if (IS_ERR_OR_NULL(di)) { |
5499 | ret = di ? PTR_ERR(di) : -ENOENT; | |
005d6712 SY |
5500 | goto out; |
5501 | } | |
d397712b | 5502 | |
5f39d397 | 5503 | btrfs_dir_item_key_to_cpu(path->nodes[0], di, location); |
56a0e706 LB |
5504 | if (location->type != BTRFS_INODE_ITEM_KEY && |
5505 | location->type != BTRFS_ROOT_ITEM_KEY) { | |
005d6712 | 5506 | ret = -EUCLEAN; |
56a0e706 LB |
5507 | btrfs_warn(root->fs_info, |
5508 | "%s gets something invalid in DIR_ITEM (name %s, directory ino %llu, location(%llu %u %llu))", | |
d1de429b | 5509 | __func__, fname.disk_name.name, btrfs_ino(dir), |
56a0e706 | 5510 | location->objectid, location->type, location->offset); |
56a0e706 | 5511 | } |
6bf9e4bd | 5512 | if (!ret) |
94a48aef | 5513 | *type = btrfs_dir_ftype(path->nodes[0], di); |
39279cc3 | 5514 | out: |
ab3c5c18 | 5515 | fscrypt_free_filename(&fname); |
39279cc3 CM |
5516 | btrfs_free_path(path); |
5517 | return ret; | |
5518 | } | |
5519 | ||
5520 | /* | |
5521 | * when we hit a tree root in a directory, the btrfs part of the inode | |
5522 | * needs to be changed to reflect the root directory of the tree root. This | |
5523 | * is kind of like crossing a mount point. | |
5524 | */ | |
2ff7e61e | 5525 | static int fixup_tree_root_location(struct btrfs_fs_info *fs_info, |
3c1b1c4c | 5526 | struct btrfs_inode *dir, |
4df27c4d YZ |
5527 | struct dentry *dentry, |
5528 | struct btrfs_key *location, | |
5529 | struct btrfs_root **sub_root) | |
39279cc3 | 5530 | { |
4df27c4d YZ |
5531 | struct btrfs_path *path; |
5532 | struct btrfs_root *new_root; | |
5533 | struct btrfs_root_ref *ref; | |
5534 | struct extent_buffer *leaf; | |
1d4c08e0 | 5535 | struct btrfs_key key; |
4df27c4d YZ |
5536 | int ret; |
5537 | int err = 0; | |
ab3c5c18 | 5538 | struct fscrypt_name fname; |
ab3c5c18 | 5539 | |
3c1b1c4c | 5540 | ret = fscrypt_setup_filename(&dir->vfs_inode, &dentry->d_name, 0, &fname); |
ab3c5c18 STD |
5541 | if (ret) |
5542 | return ret; | |
39279cc3 | 5543 | |
4df27c4d YZ |
5544 | path = btrfs_alloc_path(); |
5545 | if (!path) { | |
5546 | err = -ENOMEM; | |
5547 | goto out; | |
5548 | } | |
39279cc3 | 5549 | |
4df27c4d | 5550 | err = -ENOENT; |
3c1b1c4c | 5551 | key.objectid = dir->root->root_key.objectid; |
1d4c08e0 DS |
5552 | key.type = BTRFS_ROOT_REF_KEY; |
5553 | key.offset = location->objectid; | |
5554 | ||
0b246afa | 5555 | ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0); |
4df27c4d YZ |
5556 | if (ret) { |
5557 | if (ret < 0) | |
5558 | err = ret; | |
5559 | goto out; | |
5560 | } | |
39279cc3 | 5561 | |
4df27c4d YZ |
5562 | leaf = path->nodes[0]; |
5563 | ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref); | |
3c1b1c4c | 5564 | if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) || |
6db75318 | 5565 | btrfs_root_ref_name_len(leaf, ref) != fname.disk_name.len) |
4df27c4d | 5566 | goto out; |
39279cc3 | 5567 | |
6db75318 STD |
5568 | ret = memcmp_extent_buffer(leaf, fname.disk_name.name, |
5569 | (unsigned long)(ref + 1), fname.disk_name.len); | |
4df27c4d YZ |
5570 | if (ret) |
5571 | goto out; | |
5572 | ||
b3b4aa74 | 5573 | btrfs_release_path(path); |
4df27c4d | 5574 | |
56e9357a | 5575 | new_root = btrfs_get_fs_root(fs_info, location->objectid, true); |
4df27c4d YZ |
5576 | if (IS_ERR(new_root)) { |
5577 | err = PTR_ERR(new_root); | |
5578 | goto out; | |
5579 | } | |
5580 | ||
4df27c4d YZ |
5581 | *sub_root = new_root; |
5582 | location->objectid = btrfs_root_dirid(&new_root->root_item); | |
5583 | location->type = BTRFS_INODE_ITEM_KEY; | |
5584 | location->offset = 0; | |
5585 | err = 0; | |
5586 | out: | |
5587 | btrfs_free_path(path); | |
ab3c5c18 | 5588 | fscrypt_free_filename(&fname); |
4df27c4d | 5589 | return err; |
39279cc3 CM |
5590 | } |
5591 | ||
4c45a4f4 | 5592 | static void inode_tree_add(struct btrfs_inode *inode) |
5d4f98a2 | 5593 | { |
4c45a4f4 | 5594 | struct btrfs_root *root = inode->root; |
5d4f98a2 | 5595 | struct btrfs_inode *entry; |
03e860bd NP |
5596 | struct rb_node **p; |
5597 | struct rb_node *parent; | |
4c45a4f4 DS |
5598 | struct rb_node *new = &inode->rb_node; |
5599 | u64 ino = btrfs_ino(inode); | |
5d4f98a2 | 5600 | |
4c45a4f4 | 5601 | if (inode_unhashed(&inode->vfs_inode)) |
76dda93c | 5602 | return; |
e1409cef | 5603 | parent = NULL; |
5d4f98a2 | 5604 | spin_lock(&root->inode_lock); |
e1409cef | 5605 | p = &root->inode_tree.rb_node; |
5d4f98a2 YZ |
5606 | while (*p) { |
5607 | parent = *p; | |
5608 | entry = rb_entry(parent, struct btrfs_inode, rb_node); | |
5609 | ||
37508515 | 5610 | if (ino < btrfs_ino(entry)) |
03e860bd | 5611 | p = &parent->rb_left; |
37508515 | 5612 | else if (ino > btrfs_ino(entry)) |
03e860bd | 5613 | p = &parent->rb_right; |
5d4f98a2 YZ |
5614 | else { |
5615 | WARN_ON(!(entry->vfs_inode.i_state & | |
a4ffdde6 | 5616 | (I_WILL_FREE | I_FREEING))); |
cef21937 | 5617 | rb_replace_node(parent, new, &root->inode_tree); |
03e860bd NP |
5618 | RB_CLEAR_NODE(parent); |
5619 | spin_unlock(&root->inode_lock); | |
cef21937 | 5620 | return; |
5d4f98a2 YZ |
5621 | } |
5622 | } | |
cef21937 FDBM |
5623 | rb_link_node(new, parent, p); |
5624 | rb_insert_color(new, &root->inode_tree); | |
5d4f98a2 YZ |
5625 | spin_unlock(&root->inode_lock); |
5626 | } | |
5627 | ||
b79b7249 | 5628 | static void inode_tree_del(struct btrfs_inode *inode) |
5d4f98a2 | 5629 | { |
b79b7249 | 5630 | struct btrfs_root *root = inode->root; |
76dda93c | 5631 | int empty = 0; |
5d4f98a2 | 5632 | |
03e860bd | 5633 | spin_lock(&root->inode_lock); |
b79b7249 NB |
5634 | if (!RB_EMPTY_NODE(&inode->rb_node)) { |
5635 | rb_erase(&inode->rb_node, &root->inode_tree); | |
5636 | RB_CLEAR_NODE(&inode->rb_node); | |
76dda93c | 5637 | empty = RB_EMPTY_ROOT(&root->inode_tree); |
5d4f98a2 | 5638 | } |
03e860bd | 5639 | spin_unlock(&root->inode_lock); |
76dda93c | 5640 | |
69e9c6c6 | 5641 | if (empty && btrfs_root_refs(&root->root_item) == 0) { |
76dda93c YZ |
5642 | spin_lock(&root->inode_lock); |
5643 | empty = RB_EMPTY_ROOT(&root->inode_tree); | |
5644 | spin_unlock(&root->inode_lock); | |
5645 | if (empty) | |
5646 | btrfs_add_dead_root(root); | |
5647 | } | |
5648 | } | |
5649 | ||
5d4f98a2 | 5650 | |
e02119d5 CM |
5651 | static int btrfs_init_locked_inode(struct inode *inode, void *p) |
5652 | { | |
5653 | struct btrfs_iget_args *args = p; | |
0202e83f DS |
5654 | |
5655 | inode->i_ino = args->ino; | |
5656 | BTRFS_I(inode)->location.objectid = args->ino; | |
5657 | BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY; | |
5658 | BTRFS_I(inode)->location.offset = 0; | |
5c8fd99f JB |
5659 | BTRFS_I(inode)->root = btrfs_grab_root(args->root); |
5660 | BUG_ON(args->root && !BTRFS_I(inode)->root); | |
9b9b8854 JB |
5661 | |
5662 | if (args->root && args->root == args->root->fs_info->tree_root && | |
5663 | args->ino != BTRFS_BTREE_INODE_OBJECTID) | |
5664 | set_bit(BTRFS_INODE_FREE_SPACE_INODE, | |
5665 | &BTRFS_I(inode)->runtime_flags); | |
39279cc3 CM |
5666 | return 0; |
5667 | } | |
5668 | ||
5669 | static int btrfs_find_actor(struct inode *inode, void *opaque) | |
5670 | { | |
5671 | struct btrfs_iget_args *args = opaque; | |
0202e83f DS |
5672 | |
5673 | return args->ino == BTRFS_I(inode)->location.objectid && | |
d397712b | 5674 | args->root == BTRFS_I(inode)->root; |
39279cc3 CM |
5675 | } |
5676 | ||
0202e83f | 5677 | static struct inode *btrfs_iget_locked(struct super_block *s, u64 ino, |
5d4f98a2 | 5678 | struct btrfs_root *root) |
39279cc3 CM |
5679 | { |
5680 | struct inode *inode; | |
5681 | struct btrfs_iget_args args; | |
0202e83f | 5682 | unsigned long hashval = btrfs_inode_hash(ino, root); |
778ba82b | 5683 | |
0202e83f | 5684 | args.ino = ino; |
39279cc3 CM |
5685 | args.root = root; |
5686 | ||
778ba82b | 5687 | inode = iget5_locked(s, hashval, btrfs_find_actor, |
39279cc3 CM |
5688 | btrfs_init_locked_inode, |
5689 | (void *)&args); | |
5690 | return inode; | |
5691 | } | |
5692 | ||
4c66e0d4 | 5693 | /* |
0202e83f | 5694 | * Get an inode object given its inode number and corresponding root. |
4c66e0d4 DS |
5695 | * Path can be preallocated to prevent recursing back to iget through |
5696 | * allocator. NULL is also valid but may require an additional allocation | |
5697 | * later. | |
1a54ef8c | 5698 | */ |
0202e83f | 5699 | struct inode *btrfs_iget_path(struct super_block *s, u64 ino, |
4c66e0d4 | 5700 | struct btrfs_root *root, struct btrfs_path *path) |
1a54ef8c BR |
5701 | { |
5702 | struct inode *inode; | |
5703 | ||
0202e83f | 5704 | inode = btrfs_iget_locked(s, ino, root); |
1a54ef8c | 5705 | if (!inode) |
5d4f98a2 | 5706 | return ERR_PTR(-ENOMEM); |
1a54ef8c BR |
5707 | |
5708 | if (inode->i_state & I_NEW) { | |
67710892 FM |
5709 | int ret; |
5710 | ||
4222ea71 | 5711 | ret = btrfs_read_locked_inode(inode, path); |
9bc2ceff | 5712 | if (!ret) { |
4c45a4f4 | 5713 | inode_tree_add(BTRFS_I(inode)); |
1748f843 | 5714 | unlock_new_inode(inode); |
1748f843 | 5715 | } else { |
f5b3a417 AV |
5716 | iget_failed(inode); |
5717 | /* | |
5718 | * ret > 0 can come from btrfs_search_slot called by | |
5719 | * btrfs_read_locked_inode, this means the inode item | |
5720 | * was not found. | |
5721 | */ | |
5722 | if (ret > 0) | |
5723 | ret = -ENOENT; | |
5724 | inode = ERR_PTR(ret); | |
1748f843 MF |
5725 | } |
5726 | } | |
5727 | ||
1a54ef8c BR |
5728 | return inode; |
5729 | } | |
5730 | ||
0202e83f | 5731 | struct inode *btrfs_iget(struct super_block *s, u64 ino, struct btrfs_root *root) |
4222ea71 | 5732 | { |
0202e83f | 5733 | return btrfs_iget_path(s, ino, root, NULL); |
4222ea71 FM |
5734 | } |
5735 | ||
4df27c4d YZ |
5736 | static struct inode *new_simple_dir(struct super_block *s, |
5737 | struct btrfs_key *key, | |
5738 | struct btrfs_root *root) | |
5739 | { | |
5740 | struct inode *inode = new_inode(s); | |
5741 | ||
5742 | if (!inode) | |
5743 | return ERR_PTR(-ENOMEM); | |
5744 | ||
5c8fd99f | 5745 | BTRFS_I(inode)->root = btrfs_grab_root(root); |
4df27c4d | 5746 | memcpy(&BTRFS_I(inode)->location, key, sizeof(*key)); |
72ac3c0d | 5747 | set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags); |
4df27c4d YZ |
5748 | |
5749 | inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID; | |
6bb6b514 OS |
5750 | /* |
5751 | * We only need lookup, the rest is read-only and there's no inode | |
5752 | * associated with the dentry | |
5753 | */ | |
5754 | inode->i_op = &simple_dir_inode_operations; | |
1fdf4194 | 5755 | inode->i_opflags &= ~IOP_XATTR; |
4df27c4d YZ |
5756 | inode->i_fop = &simple_dir_operations; |
5757 | inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO; | |
c2050a45 | 5758 | inode->i_mtime = current_time(inode); |
9cc97d64 | 5759 | inode->i_atime = inode->i_mtime; |
5760 | inode->i_ctime = inode->i_mtime; | |
d3c6be6f | 5761 | BTRFS_I(inode)->i_otime = inode->i_mtime; |
4df27c4d YZ |
5762 | |
5763 | return inode; | |
5764 | } | |
5765 | ||
a55e65b8 DS |
5766 | static_assert(BTRFS_FT_UNKNOWN == FT_UNKNOWN); |
5767 | static_assert(BTRFS_FT_REG_FILE == FT_REG_FILE); | |
5768 | static_assert(BTRFS_FT_DIR == FT_DIR); | |
5769 | static_assert(BTRFS_FT_CHRDEV == FT_CHRDEV); | |
5770 | static_assert(BTRFS_FT_BLKDEV == FT_BLKDEV); | |
5771 | static_assert(BTRFS_FT_FIFO == FT_FIFO); | |
5772 | static_assert(BTRFS_FT_SOCK == FT_SOCK); | |
5773 | static_assert(BTRFS_FT_SYMLINK == FT_SYMLINK); | |
5774 | ||
6bf9e4bd QW |
5775 | static inline u8 btrfs_inode_type(struct inode *inode) |
5776 | { | |
6bf9e4bd QW |
5777 | return fs_umode_to_ftype(inode->i_mode); |
5778 | } | |
5779 | ||
3de4586c | 5780 | struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry) |
39279cc3 | 5781 | { |
0b246afa | 5782 | struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb); |
d397712b | 5783 | struct inode *inode; |
4df27c4d | 5784 | struct btrfs_root *root = BTRFS_I(dir)->root; |
39279cc3 CM |
5785 | struct btrfs_root *sub_root = root; |
5786 | struct btrfs_key location; | |
6bf9e4bd | 5787 | u8 di_type = 0; |
b4aff1f8 | 5788 | int ret = 0; |
39279cc3 CM |
5789 | |
5790 | if (dentry->d_name.len > BTRFS_NAME_LEN) | |
5791 | return ERR_PTR(-ENAMETOOLONG); | |
5f39d397 | 5792 | |
d1de429b | 5793 | ret = btrfs_inode_by_name(BTRFS_I(dir), dentry, &location, &di_type); |
39279cc3 CM |
5794 | if (ret < 0) |
5795 | return ERR_PTR(ret); | |
5f39d397 | 5796 | |
4df27c4d | 5797 | if (location.type == BTRFS_INODE_ITEM_KEY) { |
0202e83f | 5798 | inode = btrfs_iget(dir->i_sb, location.objectid, root); |
6bf9e4bd QW |
5799 | if (IS_ERR(inode)) |
5800 | return inode; | |
5801 | ||
5802 | /* Do extra check against inode mode with di_type */ | |
5803 | if (btrfs_inode_type(inode) != di_type) { | |
5804 | btrfs_crit(fs_info, | |
5805 | "inode mode mismatch with dir: inode mode=0%o btrfs type=%u dir type=%u", | |
5806 | inode->i_mode, btrfs_inode_type(inode), | |
5807 | di_type); | |
5808 | iput(inode); | |
5809 | return ERR_PTR(-EUCLEAN); | |
5810 | } | |
4df27c4d YZ |
5811 | return inode; |
5812 | } | |
5813 | ||
3c1b1c4c | 5814 | ret = fixup_tree_root_location(fs_info, BTRFS_I(dir), dentry, |
4df27c4d YZ |
5815 | &location, &sub_root); |
5816 | if (ret < 0) { | |
5817 | if (ret != -ENOENT) | |
5818 | inode = ERR_PTR(ret); | |
5819 | else | |
fc8b235f | 5820 | inode = new_simple_dir(dir->i_sb, &location, root); |
4df27c4d | 5821 | } else { |
0202e83f | 5822 | inode = btrfs_iget(dir->i_sb, location.objectid, sub_root); |
00246528 | 5823 | btrfs_put_root(sub_root); |
76dda93c | 5824 | |
fc8b235f NB |
5825 | if (IS_ERR(inode)) |
5826 | return inode; | |
5827 | ||
0b246afa | 5828 | down_read(&fs_info->cleanup_work_sem); |
bc98a42c | 5829 | if (!sb_rdonly(inode->i_sb)) |
66b4ffd1 | 5830 | ret = btrfs_orphan_cleanup(sub_root); |
0b246afa | 5831 | up_read(&fs_info->cleanup_work_sem); |
01cd3367 JB |
5832 | if (ret) { |
5833 | iput(inode); | |
66b4ffd1 | 5834 | inode = ERR_PTR(ret); |
01cd3367 | 5835 | } |
c71bf099 YZ |
5836 | } |
5837 | ||
3de4586c CM |
5838 | return inode; |
5839 | } | |
5840 | ||
fe15ce44 | 5841 | static int btrfs_dentry_delete(const struct dentry *dentry) |
76dda93c YZ |
5842 | { |
5843 | struct btrfs_root *root; | |
2b0143b5 | 5844 | struct inode *inode = d_inode(dentry); |
76dda93c | 5845 | |
848cce0d | 5846 | if (!inode && !IS_ROOT(dentry)) |
2b0143b5 | 5847 | inode = d_inode(dentry->d_parent); |
76dda93c | 5848 | |
848cce0d LZ |
5849 | if (inode) { |
5850 | root = BTRFS_I(inode)->root; | |
efefb143 YZ |
5851 | if (btrfs_root_refs(&root->root_item) == 0) |
5852 | return 1; | |
848cce0d | 5853 | |
4a0cc7ca | 5854 | if (btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) |
848cce0d | 5855 | return 1; |
efefb143 | 5856 | } |
76dda93c YZ |
5857 | return 0; |
5858 | } | |
5859 | ||
3de4586c | 5860 | static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry, |
00cd8dd3 | 5861 | unsigned int flags) |
3de4586c | 5862 | { |
3837d208 | 5863 | struct inode *inode = btrfs_lookup_dentry(dir, dentry); |
5662344b | 5864 | |
3837d208 AV |
5865 | if (inode == ERR_PTR(-ENOENT)) |
5866 | inode = NULL; | |
41d28bca | 5867 | return d_splice_alias(inode, dentry); |
39279cc3 CM |
5868 | } |
5869 | ||
9b378f6a FM |
5870 | /* |
5871 | * Find the highest existing sequence number in a directory and then set the | |
5872 | * in-memory index_cnt variable to the first free sequence number. | |
5873 | */ | |
5874 | static int btrfs_set_inode_index_count(struct btrfs_inode *inode) | |
5875 | { | |
5876 | struct btrfs_root *root = inode->root; | |
5877 | struct btrfs_key key, found_key; | |
5878 | struct btrfs_path *path; | |
5879 | struct extent_buffer *leaf; | |
5880 | int ret; | |
5881 | ||
5882 | key.objectid = btrfs_ino(inode); | |
5883 | key.type = BTRFS_DIR_INDEX_KEY; | |
5884 | key.offset = (u64)-1; | |
5885 | ||
5886 | path = btrfs_alloc_path(); | |
5887 | if (!path) | |
5888 | return -ENOMEM; | |
5889 | ||
5890 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
5891 | if (ret < 0) | |
5892 | goto out; | |
5893 | /* FIXME: we should be able to handle this */ | |
5894 | if (ret == 0) | |
5895 | goto out; | |
5896 | ret = 0; | |
5897 | ||
5898 | if (path->slots[0] == 0) { | |
5899 | inode->index_cnt = BTRFS_DIR_START_INDEX; | |
5900 | goto out; | |
5901 | } | |
5902 | ||
5903 | path->slots[0]--; | |
5904 | ||
5905 | leaf = path->nodes[0]; | |
5906 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
5907 | ||
5908 | if (found_key.objectid != btrfs_ino(inode) || | |
5909 | found_key.type != BTRFS_DIR_INDEX_KEY) { | |
5910 | inode->index_cnt = BTRFS_DIR_START_INDEX; | |
5911 | goto out; | |
5912 | } | |
5913 | ||
5914 | inode->index_cnt = found_key.offset + 1; | |
5915 | out: | |
5916 | btrfs_free_path(path); | |
5917 | return ret; | |
5918 | } | |
5919 | ||
5920 | static int btrfs_get_dir_last_index(struct btrfs_inode *dir, u64 *index) | |
5921 | { | |
5922 | if (dir->index_cnt == (u64)-1) { | |
5923 | int ret; | |
5924 | ||
5925 | ret = btrfs_inode_delayed_dir_index_count(dir); | |
5926 | if (ret) { | |
5927 | ret = btrfs_set_inode_index_count(dir); | |
5928 | if (ret) | |
5929 | return ret; | |
5930 | } | |
5931 | } | |
5932 | ||
5933 | *index = dir->index_cnt; | |
5934 | ||
5935 | return 0; | |
5936 | } | |
5937 | ||
23b5ec74 JB |
5938 | /* |
5939 | * All this infrastructure exists because dir_emit can fault, and we are holding | |
5940 | * the tree lock when doing readdir. For now just allocate a buffer and copy | |
5941 | * our information into that, and then dir_emit from the buffer. This is | |
5942 | * similar to what NFS does, only we don't keep the buffer around in pagecache | |
5943 | * because I'm afraid I'll mess that up. Long term we need to make filldir do | |
5944 | * copy_to_user_inatomic so we don't have to worry about page faulting under the | |
5945 | * tree lock. | |
5946 | */ | |
5947 | static int btrfs_opendir(struct inode *inode, struct file *file) | |
5948 | { | |
5949 | struct btrfs_file_private *private; | |
9b378f6a FM |
5950 | u64 last_index; |
5951 | int ret; | |
5952 | ||
5953 | ret = btrfs_get_dir_last_index(BTRFS_I(inode), &last_index); | |
5954 | if (ret) | |
5955 | return ret; | |
23b5ec74 JB |
5956 | |
5957 | private = kzalloc(sizeof(struct btrfs_file_private), GFP_KERNEL); | |
5958 | if (!private) | |
5959 | return -ENOMEM; | |
9b378f6a | 5960 | private->last_index = last_index; |
23b5ec74 JB |
5961 | private->filldir_buf = kzalloc(PAGE_SIZE, GFP_KERNEL); |
5962 | if (!private->filldir_buf) { | |
5963 | kfree(private); | |
5964 | return -ENOMEM; | |
5965 | } | |
5966 | file->private_data = private; | |
5967 | return 0; | |
5968 | } | |
5969 | ||
5970 | struct dir_entry { | |
5971 | u64 ino; | |
5972 | u64 offset; | |
5973 | unsigned type; | |
5974 | int name_len; | |
5975 | }; | |
5976 | ||
5977 | static int btrfs_filldir(void *addr, int entries, struct dir_context *ctx) | |
5978 | { | |
5979 | while (entries--) { | |
5980 | struct dir_entry *entry = addr; | |
5981 | char *name = (char *)(entry + 1); | |
5982 | ||
92d32170 DS |
5983 | ctx->pos = get_unaligned(&entry->offset); |
5984 | if (!dir_emit(ctx, name, get_unaligned(&entry->name_len), | |
5985 | get_unaligned(&entry->ino), | |
5986 | get_unaligned(&entry->type))) | |
23b5ec74 | 5987 | return 1; |
92d32170 DS |
5988 | addr += sizeof(struct dir_entry) + |
5989 | get_unaligned(&entry->name_len); | |
23b5ec74 JB |
5990 | ctx->pos++; |
5991 | } | |
5992 | return 0; | |
5993 | } | |
5994 | ||
9cdda8d3 | 5995 | static int btrfs_real_readdir(struct file *file, struct dir_context *ctx) |
39279cc3 | 5996 | { |
9cdda8d3 | 5997 | struct inode *inode = file_inode(file); |
39279cc3 | 5998 | struct btrfs_root *root = BTRFS_I(inode)->root; |
23b5ec74 | 5999 | struct btrfs_file_private *private = file->private_data; |
39279cc3 CM |
6000 | struct btrfs_dir_item *di; |
6001 | struct btrfs_key key; | |
5f39d397 | 6002 | struct btrfs_key found_key; |
39279cc3 | 6003 | struct btrfs_path *path; |
23b5ec74 | 6004 | void *addr; |
16cdcec7 MX |
6005 | struct list_head ins_list; |
6006 | struct list_head del_list; | |
39279cc3 | 6007 | int ret; |
5f39d397 CM |
6008 | char *name_ptr; |
6009 | int name_len; | |
23b5ec74 JB |
6010 | int entries = 0; |
6011 | int total_len = 0; | |
02dbfc99 | 6012 | bool put = false; |
c2951f32 | 6013 | struct btrfs_key location; |
5f39d397 | 6014 | |
9cdda8d3 AV |
6015 | if (!dir_emit_dots(file, ctx)) |
6016 | return 0; | |
6017 | ||
49593bfa | 6018 | path = btrfs_alloc_path(); |
16cdcec7 MX |
6019 | if (!path) |
6020 | return -ENOMEM; | |
ff5714cc | 6021 | |
23b5ec74 | 6022 | addr = private->filldir_buf; |
e4058b54 | 6023 | path->reada = READA_FORWARD; |
49593bfa | 6024 | |
c2951f32 JM |
6025 | INIT_LIST_HEAD(&ins_list); |
6026 | INIT_LIST_HEAD(&del_list); | |
9b378f6a FM |
6027 | put = btrfs_readdir_get_delayed_items(inode, private->last_index, |
6028 | &ins_list, &del_list); | |
16cdcec7 | 6029 | |
23b5ec74 | 6030 | again: |
c2951f32 | 6031 | key.type = BTRFS_DIR_INDEX_KEY; |
9cdda8d3 | 6032 | key.offset = ctx->pos; |
4a0cc7ca | 6033 | key.objectid = btrfs_ino(BTRFS_I(inode)); |
5f39d397 | 6034 | |
a8ce68fd | 6035 | btrfs_for_each_slot(root, &key, &found_key, path, ret) { |
23b5ec74 | 6036 | struct dir_entry *entry; |
a8ce68fd | 6037 | struct extent_buffer *leaf = path->nodes[0]; |
94a48aef | 6038 | u8 ftype; |
5f39d397 CM |
6039 | |
6040 | if (found_key.objectid != key.objectid) | |
39279cc3 | 6041 | break; |
c2951f32 | 6042 | if (found_key.type != BTRFS_DIR_INDEX_KEY) |
39279cc3 | 6043 | break; |
9cdda8d3 | 6044 | if (found_key.offset < ctx->pos) |
a8ce68fd | 6045 | continue; |
9b378f6a FM |
6046 | if (found_key.offset > private->last_index) |
6047 | break; | |
c2951f32 | 6048 | if (btrfs_should_delete_dir_index(&del_list, found_key.offset)) |
a8ce68fd GN |
6049 | continue; |
6050 | di = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dir_item); | |
c2951f32 | 6051 | name_len = btrfs_dir_name_len(leaf, di); |
23b5ec74 JB |
6052 | if ((total_len + sizeof(struct dir_entry) + name_len) >= |
6053 | PAGE_SIZE) { | |
6054 | btrfs_release_path(path); | |
6055 | ret = btrfs_filldir(private->filldir_buf, entries, ctx); | |
6056 | if (ret) | |
6057 | goto nopos; | |
6058 | addr = private->filldir_buf; | |
6059 | entries = 0; | |
6060 | total_len = 0; | |
6061 | goto again; | |
c2951f32 | 6062 | } |
23b5ec74 | 6063 | |
94a48aef | 6064 | ftype = btrfs_dir_flags_to_ftype(btrfs_dir_flags(leaf, di)); |
23b5ec74 | 6065 | entry = addr; |
23b5ec74 | 6066 | name_ptr = (char *)(entry + 1); |
94a48aef OS |
6067 | read_extent_buffer(leaf, name_ptr, |
6068 | (unsigned long)(di + 1), name_len); | |
6069 | put_unaligned(name_len, &entry->name_len); | |
6070 | put_unaligned(fs_ftype_to_dtype(ftype), &entry->type); | |
c2951f32 | 6071 | btrfs_dir_item_key_to_cpu(leaf, di, &location); |
92d32170 DS |
6072 | put_unaligned(location.objectid, &entry->ino); |
6073 | put_unaligned(found_key.offset, &entry->offset); | |
23b5ec74 JB |
6074 | entries++; |
6075 | addr += sizeof(struct dir_entry) + name_len; | |
6076 | total_len += sizeof(struct dir_entry) + name_len; | |
39279cc3 | 6077 | } |
a8ce68fd GN |
6078 | /* Catch error encountered during iteration */ |
6079 | if (ret < 0) | |
6080 | goto err; | |
6081 | ||
23b5ec74 JB |
6082 | btrfs_release_path(path); |
6083 | ||
6084 | ret = btrfs_filldir(private->filldir_buf, entries, ctx); | |
6085 | if (ret) | |
6086 | goto nopos; | |
49593bfa | 6087 | |
d2fbb2b5 | 6088 | ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list); |
c2951f32 | 6089 | if (ret) |
bc4ef759 DS |
6090 | goto nopos; |
6091 | ||
db62efbb ZB |
6092 | /* |
6093 | * Stop new entries from being returned after we return the last | |
6094 | * entry. | |
6095 | * | |
6096 | * New directory entries are assigned a strictly increasing | |
6097 | * offset. This means that new entries created during readdir | |
6098 | * are *guaranteed* to be seen in the future by that readdir. | |
6099 | * This has broken buggy programs which operate on names as | |
6100 | * they're returned by readdir. Until we re-use freed offsets | |
6101 | * we have this hack to stop new entries from being returned | |
6102 | * under the assumption that they'll never reach this huge | |
6103 | * offset. | |
6104 | * | |
6105 | * This is being careful not to overflow 32bit loff_t unless the | |
6106 | * last entry requires it because doing so has broken 32bit apps | |
6107 | * in the past. | |
6108 | */ | |
c2951f32 JM |
6109 | if (ctx->pos >= INT_MAX) |
6110 | ctx->pos = LLONG_MAX; | |
6111 | else | |
6112 | ctx->pos = INT_MAX; | |
39279cc3 CM |
6113 | nopos: |
6114 | ret = 0; | |
6115 | err: | |
02dbfc99 OS |
6116 | if (put) |
6117 | btrfs_readdir_put_delayed_items(inode, &ins_list, &del_list); | |
39279cc3 | 6118 | btrfs_free_path(path); |
39279cc3 CM |
6119 | return ret; |
6120 | } | |
6121 | ||
39279cc3 | 6122 | /* |
54aa1f4d | 6123 | * This is somewhat expensive, updating the tree every time the |
39279cc3 CM |
6124 | * inode changes. But, it is most likely to find the inode in cache. |
6125 | * FIXME, needs more benchmarking...there are no reasons other than performance | |
6126 | * to keep or drop this code. | |
6127 | */ | |
7152b425 | 6128 | static int btrfs_dirty_inode(struct btrfs_inode *inode) |
39279cc3 | 6129 | { |
7152b425 DS |
6130 | struct btrfs_root *root = inode->root; |
6131 | struct btrfs_fs_info *fs_info = root->fs_info; | |
39279cc3 | 6132 | struct btrfs_trans_handle *trans; |
8929ecfa YZ |
6133 | int ret; |
6134 | ||
7152b425 | 6135 | if (test_bit(BTRFS_INODE_DUMMY, &inode->runtime_flags)) |
22c44fe6 | 6136 | return 0; |
39279cc3 | 6137 | |
7a7eaa40 | 6138 | trans = btrfs_join_transaction(root); |
22c44fe6 JB |
6139 | if (IS_ERR(trans)) |
6140 | return PTR_ERR(trans); | |
8929ecfa | 6141 | |
7152b425 | 6142 | ret = btrfs_update_inode(trans, root, inode); |
4d14c5cd | 6143 | if (ret && (ret == -ENOSPC || ret == -EDQUOT)) { |
94b60442 | 6144 | /* whoops, lets try again with the full transaction */ |
3a45bb20 | 6145 | btrfs_end_transaction(trans); |
94b60442 | 6146 | trans = btrfs_start_transaction(root, 1); |
22c44fe6 JB |
6147 | if (IS_ERR(trans)) |
6148 | return PTR_ERR(trans); | |
8929ecfa | 6149 | |
7152b425 | 6150 | ret = btrfs_update_inode(trans, root, inode); |
94b60442 | 6151 | } |
3a45bb20 | 6152 | btrfs_end_transaction(trans); |
7152b425 | 6153 | if (inode->delayed_node) |
2ff7e61e | 6154 | btrfs_balance_delayed_items(fs_info); |
22c44fe6 JB |
6155 | |
6156 | return ret; | |
6157 | } | |
6158 | ||
6159 | /* | |
6160 | * This is a copy of file_update_time. We need this so we can return error on | |
6161 | * ENOSPC for updating the inode in the case of file write and mmap writes. | |
6162 | */ | |
95582b00 | 6163 | static int btrfs_update_time(struct inode *inode, struct timespec64 *now, |
e41f941a | 6164 | int flags) |
22c44fe6 | 6165 | { |
2bc55652 | 6166 | struct btrfs_root *root = BTRFS_I(inode)->root; |
3a8c7231 | 6167 | bool dirty = flags & ~S_VERSION; |
2bc55652 AB |
6168 | |
6169 | if (btrfs_root_readonly(root)) | |
6170 | return -EROFS; | |
6171 | ||
e41f941a | 6172 | if (flags & S_VERSION) |
3a8c7231 | 6173 | dirty |= inode_maybe_inc_iversion(inode, dirty); |
e41f941a JB |
6174 | if (flags & S_CTIME) |
6175 | inode->i_ctime = *now; | |
6176 | if (flags & S_MTIME) | |
6177 | inode->i_mtime = *now; | |
6178 | if (flags & S_ATIME) | |
6179 | inode->i_atime = *now; | |
7152b425 | 6180 | return dirty ? btrfs_dirty_inode(BTRFS_I(inode)) : 0; |
39279cc3 CM |
6181 | } |
6182 | ||
d352ac68 CM |
6183 | /* |
6184 | * helper to find a free sequence number in a given directory. This current | |
6185 | * code is very simple, later versions will do smarter things in the btree | |
6186 | */ | |
877574e2 | 6187 | int btrfs_set_inode_index(struct btrfs_inode *dir, u64 *index) |
aec7477b JB |
6188 | { |
6189 | int ret = 0; | |
6190 | ||
877574e2 NB |
6191 | if (dir->index_cnt == (u64)-1) { |
6192 | ret = btrfs_inode_delayed_dir_index_count(dir); | |
16cdcec7 MX |
6193 | if (ret) { |
6194 | ret = btrfs_set_inode_index_count(dir); | |
6195 | if (ret) | |
6196 | return ret; | |
6197 | } | |
aec7477b JB |
6198 | } |
6199 | ||
877574e2 NB |
6200 | *index = dir->index_cnt; |
6201 | dir->index_cnt++; | |
aec7477b JB |
6202 | |
6203 | return ret; | |
6204 | } | |
6205 | ||
b0d5d10f CM |
6206 | static int btrfs_insert_inode_locked(struct inode *inode) |
6207 | { | |
6208 | struct btrfs_iget_args args; | |
0202e83f DS |
6209 | |
6210 | args.ino = BTRFS_I(inode)->location.objectid; | |
b0d5d10f CM |
6211 | args.root = BTRFS_I(inode)->root; |
6212 | ||
6213 | return insert_inode_locked4(inode, | |
6214 | btrfs_inode_hash(inode->i_ino, BTRFS_I(inode)->root), | |
6215 | btrfs_find_actor, &args); | |
6216 | } | |
6217 | ||
3538d68d OS |
6218 | int btrfs_new_inode_prepare(struct btrfs_new_inode_args *args, |
6219 | unsigned int *trans_num_items) | |
6220 | { | |
6221 | struct inode *dir = args->dir; | |
6222 | struct inode *inode = args->inode; | |
6223 | int ret; | |
6224 | ||
ab3c5c18 STD |
6225 | if (!args->orphan) { |
6226 | ret = fscrypt_setup_filename(dir, &args->dentry->d_name, 0, | |
6227 | &args->fname); | |
6228 | if (ret) | |
6229 | return ret; | |
ab3c5c18 STD |
6230 | } |
6231 | ||
3538d68d | 6232 | ret = posix_acl_create(dir, &inode->i_mode, &args->default_acl, &args->acl); |
ab3c5c18 STD |
6233 | if (ret) { |
6234 | fscrypt_free_filename(&args->fname); | |
3538d68d | 6235 | return ret; |
ab3c5c18 | 6236 | } |
3538d68d OS |
6237 | |
6238 | /* 1 to add inode item */ | |
6239 | *trans_num_items = 1; | |
6240 | /* 1 to add compression property */ | |
6241 | if (BTRFS_I(dir)->prop_compress) | |
6242 | (*trans_num_items)++; | |
6243 | /* 1 to add default ACL xattr */ | |
6244 | if (args->default_acl) | |
6245 | (*trans_num_items)++; | |
6246 | /* 1 to add access ACL xattr */ | |
6247 | if (args->acl) | |
6248 | (*trans_num_items)++; | |
6249 | #ifdef CONFIG_SECURITY | |
6250 | /* 1 to add LSM xattr */ | |
6251 | if (dir->i_security) | |
6252 | (*trans_num_items)++; | |
6253 | #endif | |
6254 | if (args->orphan) { | |
6255 | /* 1 to add orphan item */ | |
6256 | (*trans_num_items)++; | |
6257 | } else { | |
6258 | /* | |
3538d68d OS |
6259 | * 1 to add dir item |
6260 | * 1 to add dir index | |
6261 | * 1 to update parent inode item | |
97bdf1a9 FM |
6262 | * |
6263 | * No need for 1 unit for the inode ref item because it is | |
6264 | * inserted in a batch together with the inode item at | |
6265 | * btrfs_create_new_inode(). | |
3538d68d | 6266 | */ |
97bdf1a9 | 6267 | *trans_num_items += 3; |
3538d68d OS |
6268 | } |
6269 | return 0; | |
6270 | } | |
6271 | ||
6272 | void btrfs_new_inode_args_destroy(struct btrfs_new_inode_args *args) | |
6273 | { | |
6274 | posix_acl_release(args->acl); | |
6275 | posix_acl_release(args->default_acl); | |
ab3c5c18 | 6276 | fscrypt_free_filename(&args->fname); |
3538d68d OS |
6277 | } |
6278 | ||
19aee8de AJ |
6279 | /* |
6280 | * Inherit flags from the parent inode. | |
6281 | * | |
6282 | * Currently only the compression flags and the cow flags are inherited. | |
6283 | */ | |
7a0443f0 | 6284 | static void btrfs_inherit_iflags(struct btrfs_inode *inode, struct btrfs_inode *dir) |
19aee8de AJ |
6285 | { |
6286 | unsigned int flags; | |
6287 | ||
7a0443f0 | 6288 | flags = dir->flags; |
19aee8de AJ |
6289 | |
6290 | if (flags & BTRFS_INODE_NOCOMPRESS) { | |
7a0443f0 DS |
6291 | inode->flags &= ~BTRFS_INODE_COMPRESS; |
6292 | inode->flags |= BTRFS_INODE_NOCOMPRESS; | |
19aee8de | 6293 | } else if (flags & BTRFS_INODE_COMPRESS) { |
7a0443f0 DS |
6294 | inode->flags &= ~BTRFS_INODE_NOCOMPRESS; |
6295 | inode->flags |= BTRFS_INODE_COMPRESS; | |
19aee8de AJ |
6296 | } |
6297 | ||
6298 | if (flags & BTRFS_INODE_NODATACOW) { | |
7a0443f0 DS |
6299 | inode->flags |= BTRFS_INODE_NODATACOW; |
6300 | if (S_ISREG(inode->vfs_inode.i_mode)) | |
6301 | inode->flags |= BTRFS_INODE_NODATASUM; | |
19aee8de AJ |
6302 | } |
6303 | ||
7a0443f0 | 6304 | btrfs_sync_inode_flags_to_i_flags(&inode->vfs_inode); |
19aee8de AJ |
6305 | } |
6306 | ||
3538d68d | 6307 | int btrfs_create_new_inode(struct btrfs_trans_handle *trans, |
caae78e0 | 6308 | struct btrfs_new_inode_args *args) |
39279cc3 | 6309 | { |
caae78e0 | 6310 | struct inode *dir = args->dir; |
3538d68d | 6311 | struct inode *inode = args->inode; |
6db75318 | 6312 | const struct fscrypt_str *name = args->orphan ? NULL : &args->fname.disk_name; |
caae78e0 | 6313 | struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb); |
3538d68d | 6314 | struct btrfs_root *root; |
5f39d397 | 6315 | struct btrfs_inode_item *inode_item; |
39279cc3 | 6316 | struct btrfs_key *location; |
5f39d397 | 6317 | struct btrfs_path *path; |
6437d458 | 6318 | u64 objectid; |
9c58309d CM |
6319 | struct btrfs_inode_ref *ref; |
6320 | struct btrfs_key key[2]; | |
6321 | u32 sizes[2]; | |
b7ef5f3a | 6322 | struct btrfs_item_batch batch; |
9c58309d | 6323 | unsigned long ptr; |
39279cc3 | 6324 | int ret; |
39279cc3 | 6325 | |
5f39d397 | 6326 | path = btrfs_alloc_path(); |
d8926bb3 | 6327 | if (!path) |
a1fd0c35 | 6328 | return -ENOMEM; |
39279cc3 | 6329 | |
3538d68d OS |
6330 | if (!args->subvol) |
6331 | BTRFS_I(inode)->root = btrfs_grab_root(BTRFS_I(dir)->root); | |
6332 | root = BTRFS_I(inode)->root; | |
6333 | ||
6437d458 | 6334 | ret = btrfs_get_free_objectid(root, &objectid); |
caae78e0 OS |
6335 | if (ret) |
6336 | goto out; | |
581bb050 LZ |
6337 | inode->i_ino = objectid; |
6338 | ||
caae78e0 OS |
6339 | if (args->orphan) { |
6340 | /* | |
6341 | * O_TMPFILE, set link count to 0, so that after this point, we | |
6342 | * fill in an inode item with the correct link count. | |
6343 | */ | |
6344 | set_nlink(inode, 0); | |
6345 | } else { | |
1abe9b8a | 6346 | trace_btrfs_inode_request(dir); |
6347 | ||
caae78e0 OS |
6348 | ret = btrfs_set_inode_index(BTRFS_I(dir), &BTRFS_I(inode)->dir_index); |
6349 | if (ret) | |
6350 | goto out; | |
aec7477b | 6351 | } |
49024388 FM |
6352 | /* index_cnt is ignored for everything but a dir. */ |
6353 | BTRFS_I(inode)->index_cnt = BTRFS_DIR_START_INDEX; | |
e02119d5 | 6354 | BTRFS_I(inode)->generation = trans->transid; |
76195853 | 6355 | inode->i_generation = BTRFS_I(inode)->generation; |
b888db2b | 6356 | |
caae78e0 OS |
6357 | /* |
6358 | * Subvolumes don't inherit flags from their parent directory. | |
6359 | * Originally this was probably by accident, but we probably can't | |
6360 | * change it now without compatibility issues. | |
6361 | */ | |
6362 | if (!args->subvol) | |
7a0443f0 | 6363 | btrfs_inherit_iflags(BTRFS_I(inode), BTRFS_I(dir)); |
305eaac0 | 6364 | |
a1fd0c35 | 6365 | if (S_ISREG(inode->i_mode)) { |
305eaac0 OS |
6366 | if (btrfs_test_opt(fs_info, NODATASUM)) |
6367 | BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM; | |
6368 | if (btrfs_test_opt(fs_info, NODATACOW)) | |
6369 | BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW | | |
6370 | BTRFS_INODE_NODATASUM; | |
6371 | } | |
6372 | ||
caae78e0 OS |
6373 | location = &BTRFS_I(inode)->location; |
6374 | location->objectid = objectid; | |
6375 | location->offset = 0; | |
6376 | location->type = BTRFS_INODE_ITEM_KEY; | |
6377 | ||
6378 | ret = btrfs_insert_inode_locked(inode); | |
6379 | if (ret < 0) { | |
6380 | if (!args->orphan) | |
6381 | BTRFS_I(dir)->index_cnt--; | |
6382 | goto out; | |
6383 | } | |
6384 | ||
5dc562c5 JB |
6385 | /* |
6386 | * We could have gotten an inode number from somebody who was fsynced | |
6387 | * and then removed in this same transaction, so let's just set full | |
6388 | * sync since it will be a full sync anyway and this will blow away the | |
6389 | * old info in the log. | |
6390 | */ | |
23e3337f | 6391 | btrfs_set_inode_full_sync(BTRFS_I(inode)); |
5dc562c5 | 6392 | |
9c58309d | 6393 | key[0].objectid = objectid; |
962a298f | 6394 | key[0].type = BTRFS_INODE_ITEM_KEY; |
9c58309d CM |
6395 | key[0].offset = 0; |
6396 | ||
9c58309d | 6397 | sizes[0] = sizeof(struct btrfs_inode_item); |
ef3b9af5 | 6398 | |
caae78e0 | 6399 | if (!args->orphan) { |
ef3b9af5 FM |
6400 | /* |
6401 | * Start new inodes with an inode_ref. This is slightly more | |
6402 | * efficient for small numbers of hard links since they will | |
6403 | * be packed into one item. Extended refs will kick in if we | |
6404 | * add more hard links than can fit in the ref item. | |
6405 | */ | |
6406 | key[1].objectid = objectid; | |
962a298f | 6407 | key[1].type = BTRFS_INODE_REF_KEY; |
caae78e0 | 6408 | if (args->subvol) { |
23c24ef8 | 6409 | key[1].offset = objectid; |
caae78e0 OS |
6410 | sizes[1] = 2 + sizeof(*ref); |
6411 | } else { | |
6412 | key[1].offset = btrfs_ino(BTRFS_I(dir)); | |
e43eec81 | 6413 | sizes[1] = name->len + sizeof(*ref); |
caae78e0 | 6414 | } |
ef3b9af5 | 6415 | } |
9c58309d | 6416 | |
b7ef5f3a FM |
6417 | batch.keys = &key[0]; |
6418 | batch.data_sizes = &sizes[0]; | |
caae78e0 OS |
6419 | batch.total_data_size = sizes[0] + (args->orphan ? 0 : sizes[1]); |
6420 | batch.nr = args->orphan ? 1 : 2; | |
b7ef5f3a | 6421 | ret = btrfs_insert_empty_items(trans, root, path, &batch); |
caae78e0 OS |
6422 | if (ret != 0) { |
6423 | btrfs_abort_transaction(trans, ret); | |
6424 | goto discard; | |
6425 | } | |
5f39d397 | 6426 | |
c2050a45 | 6427 | inode->i_mtime = current_time(inode); |
9cc97d64 | 6428 | inode->i_atime = inode->i_mtime; |
6429 | inode->i_ctime = inode->i_mtime; | |
d3c6be6f | 6430 | BTRFS_I(inode)->i_otime = inode->i_mtime; |
9cc97d64 | 6431 | |
caae78e0 OS |
6432 | /* |
6433 | * We're going to fill the inode item now, so at this point the inode | |
6434 | * must be fully initialized. | |
6435 | */ | |
6436 | ||
5f39d397 CM |
6437 | inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0], |
6438 | struct btrfs_inode_item); | |
b159fa28 | 6439 | memzero_extent_buffer(path->nodes[0], (unsigned long)inode_item, |
293f7e07 | 6440 | sizeof(*inode_item)); |
e02119d5 | 6441 | fill_inode_item(trans, path->nodes[0], inode_item, inode); |
9c58309d | 6442 | |
caae78e0 | 6443 | if (!args->orphan) { |
ef3b9af5 FM |
6444 | ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1, |
6445 | struct btrfs_inode_ref); | |
ef3b9af5 | 6446 | ptr = (unsigned long)(ref + 1); |
caae78e0 OS |
6447 | if (args->subvol) { |
6448 | btrfs_set_inode_ref_name_len(path->nodes[0], ref, 2); | |
6449 | btrfs_set_inode_ref_index(path->nodes[0], ref, 0); | |
6450 | write_extent_buffer(path->nodes[0], "..", ptr, 2); | |
6451 | } else { | |
e43eec81 STD |
6452 | btrfs_set_inode_ref_name_len(path->nodes[0], ref, |
6453 | name->len); | |
caae78e0 OS |
6454 | btrfs_set_inode_ref_index(path->nodes[0], ref, |
6455 | BTRFS_I(inode)->dir_index); | |
e43eec81 STD |
6456 | write_extent_buffer(path->nodes[0], name->name, ptr, |
6457 | name->len); | |
caae78e0 | 6458 | } |
ef3b9af5 | 6459 | } |
9c58309d | 6460 | |
5f39d397 | 6461 | btrfs_mark_buffer_dirty(path->nodes[0]); |
814e7718 FM |
6462 | /* |
6463 | * We don't need the path anymore, plus inheriting properties, adding | |
6464 | * ACLs, security xattrs, orphan item or adding the link, will result in | |
6465 | * allocating yet another path. So just free our path. | |
6466 | */ | |
6467 | btrfs_free_path(path); | |
6468 | path = NULL; | |
5f39d397 | 6469 | |
6c3636eb STD |
6470 | if (args->subvol) { |
6471 | struct inode *parent; | |
6472 | ||
6473 | /* | |
6474 | * Subvolumes inherit properties from their parent subvolume, | |
6475 | * not the directory they were created in. | |
6476 | */ | |
6477 | parent = btrfs_iget(fs_info->sb, BTRFS_FIRST_FREE_OBJECTID, | |
6478 | BTRFS_I(dir)->root); | |
6479 | if (IS_ERR(parent)) { | |
6480 | ret = PTR_ERR(parent); | |
6481 | } else { | |
6482 | ret = btrfs_inode_inherit_props(trans, inode, parent); | |
6483 | iput(parent); | |
6484 | } | |
6485 | } else { | |
6486 | ret = btrfs_inode_inherit_props(trans, inode, dir); | |
6487 | } | |
6488 | if (ret) { | |
6489 | btrfs_err(fs_info, | |
6490 | "error inheriting props for ino %llu (root %llu): %d", | |
6491 | btrfs_ino(BTRFS_I(inode)), root->root_key.objectid, | |
6492 | ret); | |
6493 | } | |
6494 | ||
6495 | /* | |
6496 | * Subvolumes don't inherit ACLs or get passed to the LSM. This is | |
6497 | * probably a bug. | |
6498 | */ | |
6499 | if (!args->subvol) { | |
6500 | ret = btrfs_init_inode_security(trans, args); | |
6501 | if (ret) { | |
6502 | btrfs_abort_transaction(trans, ret); | |
6503 | goto discard; | |
6504 | } | |
6505 | } | |
6506 | ||
4c45a4f4 | 6507 | inode_tree_add(BTRFS_I(inode)); |
1abe9b8a | 6508 | |
6509 | trace_btrfs_inode_new(inode); | |
d9094414 | 6510 | btrfs_set_inode_last_trans(trans, BTRFS_I(inode)); |
1abe9b8a | 6511 | |
8ea05e3a AB |
6512 | btrfs_update_root_times(trans, root); |
6513 | ||
caae78e0 OS |
6514 | if (args->orphan) { |
6515 | ret = btrfs_orphan_add(trans, BTRFS_I(inode)); | |
6516 | } else { | |
6517 | ret = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode), name, | |
e43eec81 | 6518 | 0, BTRFS_I(inode)->dir_index); |
caae78e0 OS |
6519 | } |
6520 | if (ret) { | |
6521 | btrfs_abort_transaction(trans, ret); | |
6522 | goto discard; | |
6523 | } | |
63541927 | 6524 | |
814e7718 | 6525 | return 0; |
b0d5d10f | 6526 | |
caae78e0 | 6527 | discard: |
a1fd0c35 OS |
6528 | /* |
6529 | * discard_new_inode() calls iput(), but the caller owns the reference | |
6530 | * to the inode. | |
6531 | */ | |
6532 | ihold(inode); | |
32955c54 | 6533 | discard_new_inode(inode); |
caae78e0 | 6534 | out: |
5f39d397 | 6535 | btrfs_free_path(path); |
a1fd0c35 | 6536 | return ret; |
39279cc3 CM |
6537 | } |
6538 | ||
d352ac68 CM |
6539 | /* |
6540 | * utility function to add 'inode' into 'parent_inode' with | |
6541 | * a give name and a given sequence number. | |
6542 | * if 'add_backref' is true, also insert a backref from the | |
6543 | * inode to the parent directory. | |
6544 | */ | |
e02119d5 | 6545 | int btrfs_add_link(struct btrfs_trans_handle *trans, |
db0a669f | 6546 | struct btrfs_inode *parent_inode, struct btrfs_inode *inode, |
6db75318 | 6547 | const struct fscrypt_str *name, int add_backref, u64 index) |
39279cc3 | 6548 | { |
4df27c4d | 6549 | int ret = 0; |
39279cc3 | 6550 | struct btrfs_key key; |
db0a669f NB |
6551 | struct btrfs_root *root = parent_inode->root; |
6552 | u64 ino = btrfs_ino(inode); | |
6553 | u64 parent_ino = btrfs_ino(parent_inode); | |
5f39d397 | 6554 | |
33345d01 | 6555 | if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) { |
db0a669f | 6556 | memcpy(&key, &inode->root->root_key, sizeof(key)); |
4df27c4d | 6557 | } else { |
33345d01 | 6558 | key.objectid = ino; |
962a298f | 6559 | key.type = BTRFS_INODE_ITEM_KEY; |
4df27c4d YZ |
6560 | key.offset = 0; |
6561 | } | |
6562 | ||
33345d01 | 6563 | if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) { |
6025c19f | 6564 | ret = btrfs_add_root_ref(trans, key.objectid, |
0b246afa | 6565 | root->root_key.objectid, parent_ino, |
e43eec81 | 6566 | index, name); |
4df27c4d | 6567 | } else if (add_backref) { |
e43eec81 STD |
6568 | ret = btrfs_insert_inode_ref(trans, root, name, |
6569 | ino, parent_ino, index); | |
4df27c4d | 6570 | } |
39279cc3 | 6571 | |
79787eaa JM |
6572 | /* Nothing to clean up yet */ |
6573 | if (ret) | |
6574 | return ret; | |
4df27c4d | 6575 | |
e43eec81 | 6576 | ret = btrfs_insert_dir_item(trans, name, parent_inode, &key, |
db0a669f | 6577 | btrfs_inode_type(&inode->vfs_inode), index); |
9c52057c | 6578 | if (ret == -EEXIST || ret == -EOVERFLOW) |
79787eaa JM |
6579 | goto fail_dir_item; |
6580 | else if (ret) { | |
66642832 | 6581 | btrfs_abort_transaction(trans, ret); |
79787eaa | 6582 | return ret; |
39279cc3 | 6583 | } |
79787eaa | 6584 | |
db0a669f | 6585 | btrfs_i_size_write(parent_inode, parent_inode->vfs_inode.i_size + |
e43eec81 | 6586 | name->len * 2); |
db0a669f | 6587 | inode_inc_iversion(&parent_inode->vfs_inode); |
5338e43a FM |
6588 | /* |
6589 | * If we are replaying a log tree, we do not want to update the mtime | |
6590 | * and ctime of the parent directory with the current time, since the | |
6591 | * log replay procedure is responsible for setting them to their correct | |
6592 | * values (the ones it had when the fsync was done). | |
6593 | */ | |
6594 | if (!test_bit(BTRFS_FS_LOG_RECOVERING, &root->fs_info->flags)) { | |
6595 | struct timespec64 now = current_time(&parent_inode->vfs_inode); | |
6596 | ||
6597 | parent_inode->vfs_inode.i_mtime = now; | |
6598 | parent_inode->vfs_inode.i_ctime = now; | |
6599 | } | |
9a56fcd1 | 6600 | ret = btrfs_update_inode(trans, root, parent_inode); |
79787eaa | 6601 | if (ret) |
66642832 | 6602 | btrfs_abort_transaction(trans, ret); |
39279cc3 | 6603 | return ret; |
fe66a05a CM |
6604 | |
6605 | fail_dir_item: | |
6606 | if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) { | |
6607 | u64 local_index; | |
6608 | int err; | |
3ee1c553 | 6609 | err = btrfs_del_root_ref(trans, key.objectid, |
0b246afa | 6610 | root->root_key.objectid, parent_ino, |
e43eec81 | 6611 | &local_index, name); |
1690dd41 JT |
6612 | if (err) |
6613 | btrfs_abort_transaction(trans, err); | |
fe66a05a CM |
6614 | } else if (add_backref) { |
6615 | u64 local_index; | |
6616 | int err; | |
6617 | ||
e43eec81 STD |
6618 | err = btrfs_del_inode_ref(trans, root, name, ino, parent_ino, |
6619 | &local_index); | |
1690dd41 JT |
6620 | if (err) |
6621 | btrfs_abort_transaction(trans, err); | |
fe66a05a | 6622 | } |
1690dd41 JT |
6623 | |
6624 | /* Return the original error code */ | |
fe66a05a | 6625 | return ret; |
39279cc3 CM |
6626 | } |
6627 | ||
5f465bf1 OS |
6628 | static int btrfs_create_common(struct inode *dir, struct dentry *dentry, |
6629 | struct inode *inode) | |
618e21d5 | 6630 | { |
2ff7e61e | 6631 | struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb); |
618e21d5 | 6632 | struct btrfs_root *root = BTRFS_I(dir)->root; |
3538d68d OS |
6633 | struct btrfs_new_inode_args new_inode_args = { |
6634 | .dir = dir, | |
6635 | .dentry = dentry, | |
6636 | .inode = inode, | |
6637 | }; | |
6638 | unsigned int trans_num_items; | |
5f465bf1 | 6639 | struct btrfs_trans_handle *trans; |
618e21d5 | 6640 | int err; |
618e21d5 | 6641 | |
3538d68d | 6642 | err = btrfs_new_inode_prepare(&new_inode_args, &trans_num_items); |
caae78e0 OS |
6643 | if (err) |
6644 | goto out_inode; | |
3538d68d OS |
6645 | |
6646 | trans = btrfs_start_transaction(root, trans_num_items); | |
a1fd0c35 | 6647 | if (IS_ERR(trans)) { |
3538d68d OS |
6648 | err = PTR_ERR(trans); |
6649 | goto out_new_inode_args; | |
a1fd0c35 | 6650 | } |
1832a6d5 | 6651 | |
caae78e0 OS |
6652 | err = btrfs_create_new_inode(trans, &new_inode_args); |
6653 | if (!err) | |
6654 | d_instantiate_new(dentry, inode); | |
b0d5d10f | 6655 | |
3a45bb20 | 6656 | btrfs_end_transaction(trans); |
5f465bf1 | 6657 | btrfs_btree_balance_dirty(fs_info); |
3538d68d OS |
6658 | out_new_inode_args: |
6659 | btrfs_new_inode_args_destroy(&new_inode_args); | |
caae78e0 OS |
6660 | out_inode: |
6661 | if (err) | |
6662 | iput(inode); | |
618e21d5 JB |
6663 | return err; |
6664 | } | |
6665 | ||
5ebb29be | 6666 | static int btrfs_mknod(struct mnt_idmap *idmap, struct inode *dir, |
5f465bf1 OS |
6667 | struct dentry *dentry, umode_t mode, dev_t rdev) |
6668 | { | |
6669 | struct inode *inode; | |
6670 | ||
6671 | inode = new_inode(dir->i_sb); | |
6672 | if (!inode) | |
6673 | return -ENOMEM; | |
f2d40141 | 6674 | inode_init_owner(idmap, inode, dir, mode); |
5f465bf1 OS |
6675 | inode->i_op = &btrfs_special_inode_operations; |
6676 | init_special_inode(inode, inode->i_mode, rdev); | |
6677 | return btrfs_create_common(dir, dentry, inode); | |
6678 | } | |
6679 | ||
6c960e68 | 6680 | static int btrfs_create(struct mnt_idmap *idmap, struct inode *dir, |
549c7297 | 6681 | struct dentry *dentry, umode_t mode, bool excl) |
39279cc3 | 6682 | { |
a1fd0c35 | 6683 | struct inode *inode; |
39279cc3 | 6684 | |
a1fd0c35 OS |
6685 | inode = new_inode(dir->i_sb); |
6686 | if (!inode) | |
6687 | return -ENOMEM; | |
f2d40141 | 6688 | inode_init_owner(idmap, inode, dir, mode); |
a1fd0c35 OS |
6689 | inode->i_fop = &btrfs_file_operations; |
6690 | inode->i_op = &btrfs_file_inode_operations; | |
6691 | inode->i_mapping->a_ops = &btrfs_aops; | |
5f465bf1 | 6692 | return btrfs_create_common(dir, dentry, inode); |
39279cc3 CM |
6693 | } |
6694 | ||
6695 | static int btrfs_link(struct dentry *old_dentry, struct inode *dir, | |
6696 | struct dentry *dentry) | |
6697 | { | |
271dba45 | 6698 | struct btrfs_trans_handle *trans = NULL; |
39279cc3 | 6699 | struct btrfs_root *root = BTRFS_I(dir)->root; |
2b0143b5 | 6700 | struct inode *inode = d_inode(old_dentry); |
2ff7e61e | 6701 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
ab3c5c18 | 6702 | struct fscrypt_name fname; |
00e4e6b3 | 6703 | u64 index; |
39279cc3 CM |
6704 | int err; |
6705 | int drop_inode = 0; | |
6706 | ||
4a8be425 | 6707 | /* do not allow sys_link's with other subvols of the same device */ |
4fd786e6 | 6708 | if (root->root_key.objectid != BTRFS_I(inode)->root->root_key.objectid) |
3ab3564f | 6709 | return -EXDEV; |
4a8be425 | 6710 | |
f186373f | 6711 | if (inode->i_nlink >= BTRFS_LINK_MAX) |
c055e99e | 6712 | return -EMLINK; |
4a8be425 | 6713 | |
ab3c5c18 STD |
6714 | err = fscrypt_setup_filename(dir, &dentry->d_name, 0, &fname); |
6715 | if (err) | |
6716 | goto fail; | |
6717 | ||
877574e2 | 6718 | err = btrfs_set_inode_index(BTRFS_I(dir), &index); |
aec7477b JB |
6719 | if (err) |
6720 | goto fail; | |
6721 | ||
a22285a6 | 6722 | /* |
7e6b6465 | 6723 | * 2 items for inode and inode ref |
a22285a6 | 6724 | * 2 items for dir items |
7e6b6465 | 6725 | * 1 item for parent inode |
399b0bbf | 6726 | * 1 item for orphan item deletion if O_TMPFILE |
a22285a6 | 6727 | */ |
399b0bbf | 6728 | trans = btrfs_start_transaction(root, inode->i_nlink ? 5 : 6); |
a22285a6 YZ |
6729 | if (IS_ERR(trans)) { |
6730 | err = PTR_ERR(trans); | |
271dba45 | 6731 | trans = NULL; |
a22285a6 YZ |
6732 | goto fail; |
6733 | } | |
5f39d397 | 6734 | |
67de1176 MX |
6735 | /* There are several dir indexes for this inode, clear the cache. */ |
6736 | BTRFS_I(inode)->dir_index = 0ULL; | |
8b558c5f | 6737 | inc_nlink(inode); |
0c4d2d95 | 6738 | inode_inc_iversion(inode); |
c2050a45 | 6739 | inode->i_ctime = current_time(inode); |
7de9c6ee | 6740 | ihold(inode); |
e9976151 | 6741 | set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags); |
aec7477b | 6742 | |
81512e89 | 6743 | err = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode), |
6db75318 | 6744 | &fname.disk_name, 1, index); |
5f39d397 | 6745 | |
a5719521 | 6746 | if (err) { |
54aa1f4d | 6747 | drop_inode = 1; |
a5719521 | 6748 | } else { |
10d9f309 | 6749 | struct dentry *parent = dentry->d_parent; |
d4682ba0 | 6750 | |
9a56fcd1 | 6751 | err = btrfs_update_inode(trans, root, BTRFS_I(inode)); |
79787eaa JM |
6752 | if (err) |
6753 | goto fail; | |
ef3b9af5 FM |
6754 | if (inode->i_nlink == 1) { |
6755 | /* | |
6756 | * If new hard link count is 1, it's a file created | |
6757 | * with open(2) O_TMPFILE flag. | |
6758 | */ | |
3d6ae7bb | 6759 | err = btrfs_orphan_del(trans, BTRFS_I(inode)); |
ef3b9af5 FM |
6760 | if (err) |
6761 | goto fail; | |
6762 | } | |
08c422c2 | 6763 | d_instantiate(dentry, inode); |
88d2beec | 6764 | btrfs_log_new_name(trans, old_dentry, NULL, 0, parent); |
a5719521 | 6765 | } |
39279cc3 | 6766 | |
1832a6d5 | 6767 | fail: |
ab3c5c18 | 6768 | fscrypt_free_filename(&fname); |
271dba45 | 6769 | if (trans) |
3a45bb20 | 6770 | btrfs_end_transaction(trans); |
39279cc3 CM |
6771 | if (drop_inode) { |
6772 | inode_dec_link_count(inode); | |
6773 | iput(inode); | |
6774 | } | |
2ff7e61e | 6775 | btrfs_btree_balance_dirty(fs_info); |
39279cc3 CM |
6776 | return err; |
6777 | } | |
6778 | ||
c54bd91e | 6779 | static int btrfs_mkdir(struct mnt_idmap *idmap, struct inode *dir, |
549c7297 | 6780 | struct dentry *dentry, umode_t mode) |
39279cc3 | 6781 | { |
a1fd0c35 | 6782 | struct inode *inode; |
39279cc3 | 6783 | |
a1fd0c35 OS |
6784 | inode = new_inode(dir->i_sb); |
6785 | if (!inode) | |
6786 | return -ENOMEM; | |
f2d40141 | 6787 | inode_init_owner(idmap, inode, dir, S_IFDIR | mode); |
a1fd0c35 OS |
6788 | inode->i_op = &btrfs_dir_inode_operations; |
6789 | inode->i_fop = &btrfs_dir_file_operations; | |
5f465bf1 | 6790 | return btrfs_create_common(dir, dentry, inode); |
39279cc3 CM |
6791 | } |
6792 | ||
c8b97818 | 6793 | static noinline int uncompress_inline(struct btrfs_path *path, |
e40da0e5 | 6794 | struct page *page, |
c8b97818 CM |
6795 | struct btrfs_file_extent_item *item) |
6796 | { | |
6797 | int ret; | |
6798 | struct extent_buffer *leaf = path->nodes[0]; | |
6799 | char *tmp; | |
6800 | size_t max_size; | |
6801 | unsigned long inline_size; | |
6802 | unsigned long ptr; | |
261507a0 | 6803 | int compress_type; |
c8b97818 | 6804 | |
261507a0 | 6805 | compress_type = btrfs_file_extent_compression(leaf, item); |
c8b97818 | 6806 | max_size = btrfs_file_extent_ram_bytes(leaf, item); |
437bd07e | 6807 | inline_size = btrfs_file_extent_inline_item_len(leaf, path->slots[0]); |
c8b97818 | 6808 | tmp = kmalloc(inline_size, GFP_NOFS); |
8d413713 TI |
6809 | if (!tmp) |
6810 | return -ENOMEM; | |
c8b97818 CM |
6811 | ptr = btrfs_file_extent_inline_start(item); |
6812 | ||
6813 | read_extent_buffer(leaf, tmp, ptr, inline_size); | |
6814 | ||
09cbfeaf | 6815 | max_size = min_t(unsigned long, PAGE_SIZE, max_size); |
a982fc82 | 6816 | ret = btrfs_decompress(compress_type, tmp, page, 0, inline_size, max_size); |
e1699d2d ZB |
6817 | |
6818 | /* | |
6819 | * decompression code contains a memset to fill in any space between the end | |
6820 | * of the uncompressed data and the end of max_size in case the decompressed | |
6821 | * data ends up shorter than ram_bytes. That doesn't cover the hole between | |
6822 | * the end of an inline extent and the beginning of the next block, so we | |
6823 | * cover that region here. | |
6824 | */ | |
6825 | ||
a982fc82 QW |
6826 | if (max_size < PAGE_SIZE) |
6827 | memzero_page(page, max_size, PAGE_SIZE - max_size); | |
c8b97818 | 6828 | kfree(tmp); |
166ae5a4 | 6829 | return ret; |
c8b97818 CM |
6830 | } |
6831 | ||
a982fc82 QW |
6832 | static int read_inline_extent(struct btrfs_inode *inode, struct btrfs_path *path, |
6833 | struct page *page) | |
6834 | { | |
6835 | struct btrfs_file_extent_item *fi; | |
6836 | void *kaddr; | |
6837 | size_t copy_size; | |
6838 | ||
6839 | if (!page || PageUptodate(page)) | |
6840 | return 0; | |
6841 | ||
6842 | ASSERT(page_offset(page) == 0); | |
6843 | ||
6844 | fi = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
6845 | struct btrfs_file_extent_item); | |
6846 | if (btrfs_file_extent_compression(path->nodes[0], fi) != BTRFS_COMPRESS_NONE) | |
6847 | return uncompress_inline(path, page, fi); | |
6848 | ||
6849 | copy_size = min_t(u64, PAGE_SIZE, | |
6850 | btrfs_file_extent_ram_bytes(path->nodes[0], fi)); | |
6851 | kaddr = kmap_local_page(page); | |
6852 | read_extent_buffer(path->nodes[0], kaddr, | |
6853 | btrfs_file_extent_inline_start(fi), copy_size); | |
6854 | kunmap_local(kaddr); | |
6855 | if (copy_size < PAGE_SIZE) | |
6856 | memzero_page(page, copy_size, PAGE_SIZE - copy_size); | |
6857 | return 0; | |
6858 | } | |
6859 | ||
43dd529a DS |
6860 | /* |
6861 | * Lookup the first extent overlapping a range in a file. | |
6862 | * | |
39b07b5d OS |
6863 | * @inode: file to search in |
6864 | * @page: page to read extent data into if the extent is inline | |
6865 | * @pg_offset: offset into @page to copy to | |
6866 | * @start: file offset | |
6867 | * @len: length of range starting at @start | |
6868 | * | |
43dd529a DS |
6869 | * Return the first &struct extent_map which overlaps the given range, reading |
6870 | * it from the B-tree and caching it if necessary. Note that there may be more | |
6871 | * extents which overlap the given range after the returned extent_map. | |
d352ac68 | 6872 | * |
39b07b5d OS |
6873 | * If @page is not NULL and the extent is inline, this also reads the extent |
6874 | * data directly into the page and marks the extent up to date in the io_tree. | |
6875 | * | |
6876 | * Return: ERR_PTR on error, non-NULL extent_map on success. | |
d352ac68 | 6877 | */ |
fc4f21b1 | 6878 | struct extent_map *btrfs_get_extent(struct btrfs_inode *inode, |
39b07b5d OS |
6879 | struct page *page, size_t pg_offset, |
6880 | u64 start, u64 len) | |
a52d9a80 | 6881 | { |
3ffbd68c | 6882 | struct btrfs_fs_info *fs_info = inode->root->fs_info; |
1028d1c4 | 6883 | int ret = 0; |
a52d9a80 CM |
6884 | u64 extent_start = 0; |
6885 | u64 extent_end = 0; | |
fc4f21b1 | 6886 | u64 objectid = btrfs_ino(inode); |
7e74e235 | 6887 | int extent_type = -1; |
f421950f | 6888 | struct btrfs_path *path = NULL; |
fc4f21b1 | 6889 | struct btrfs_root *root = inode->root; |
a52d9a80 | 6890 | struct btrfs_file_extent_item *item; |
5f39d397 CM |
6891 | struct extent_buffer *leaf; |
6892 | struct btrfs_key found_key; | |
a52d9a80 | 6893 | struct extent_map *em = NULL; |
fc4f21b1 | 6894 | struct extent_map_tree *em_tree = &inode->extent_tree; |
a52d9a80 | 6895 | |
890871be | 6896 | read_lock(&em_tree->lock); |
d1310b2e | 6897 | em = lookup_extent_mapping(em_tree, start, len); |
890871be | 6898 | read_unlock(&em_tree->lock); |
d1310b2e | 6899 | |
a52d9a80 | 6900 | if (em) { |
e1c4b745 CM |
6901 | if (em->start > start || em->start + em->len <= start) |
6902 | free_extent_map(em); | |
6903 | else if (em->block_start == EXTENT_MAP_INLINE && page) | |
70dec807 CM |
6904 | free_extent_map(em); |
6905 | else | |
6906 | goto out; | |
a52d9a80 | 6907 | } |
172ddd60 | 6908 | em = alloc_extent_map(); |
a52d9a80 | 6909 | if (!em) { |
1028d1c4 | 6910 | ret = -ENOMEM; |
d1310b2e | 6911 | goto out; |
a52d9a80 | 6912 | } |
d1310b2e | 6913 | em->start = EXTENT_MAP_HOLE; |
445a6944 | 6914 | em->orig_start = EXTENT_MAP_HOLE; |
d1310b2e | 6915 | em->len = (u64)-1; |
c8b97818 | 6916 | em->block_len = (u64)-1; |
f421950f | 6917 | |
bee6ec82 | 6918 | path = btrfs_alloc_path(); |
f421950f | 6919 | if (!path) { |
1028d1c4 | 6920 | ret = -ENOMEM; |
bee6ec82 | 6921 | goto out; |
f421950f CM |
6922 | } |
6923 | ||
bee6ec82 LB |
6924 | /* Chances are we'll be called again, so go ahead and do readahead */ |
6925 | path->reada = READA_FORWARD; | |
4d7240f0 JB |
6926 | |
6927 | /* | |
6928 | * The same explanation in load_free_space_cache applies here as well, | |
6929 | * we only read when we're loading the free space cache, and at that | |
6930 | * point the commit_root has everything we need. | |
6931 | */ | |
6932 | if (btrfs_is_free_space_inode(inode)) { | |
6933 | path->search_commit_root = 1; | |
6934 | path->skip_locking = 1; | |
6935 | } | |
51899412 | 6936 | |
5c9a702e | 6937 | ret = btrfs_lookup_file_extent(NULL, root, path, objectid, start, 0); |
a52d9a80 | 6938 | if (ret < 0) { |
a52d9a80 | 6939 | goto out; |
b8eeab7f | 6940 | } else if (ret > 0) { |
a52d9a80 CM |
6941 | if (path->slots[0] == 0) |
6942 | goto not_found; | |
6943 | path->slots[0]--; | |
1028d1c4 | 6944 | ret = 0; |
a52d9a80 CM |
6945 | } |
6946 | ||
5f39d397 CM |
6947 | leaf = path->nodes[0]; |
6948 | item = btrfs_item_ptr(leaf, path->slots[0], | |
a52d9a80 | 6949 | struct btrfs_file_extent_item); |
5f39d397 | 6950 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
5f39d397 | 6951 | if (found_key.objectid != objectid || |
694c12ed | 6952 | found_key.type != BTRFS_EXTENT_DATA_KEY) { |
25a50341 JB |
6953 | /* |
6954 | * If we backup past the first extent we want to move forward | |
6955 | * and see if there is an extent in front of us, otherwise we'll | |
6956 | * say there is a hole for our whole search range which can | |
6957 | * cause problems. | |
6958 | */ | |
6959 | extent_end = start; | |
6960 | goto next; | |
a52d9a80 CM |
6961 | } |
6962 | ||
694c12ed | 6963 | extent_type = btrfs_file_extent_type(leaf, item); |
5f39d397 | 6964 | extent_start = found_key.offset; |
a5eeb3d1 | 6965 | extent_end = btrfs_file_extent_end(path); |
694c12ed NB |
6966 | if (extent_type == BTRFS_FILE_EXTENT_REG || |
6967 | extent_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
6bf9e4bd QW |
6968 | /* Only regular file could have regular/prealloc extent */ |
6969 | if (!S_ISREG(inode->vfs_inode.i_mode)) { | |
1028d1c4 | 6970 | ret = -EUCLEAN; |
6bf9e4bd QW |
6971 | btrfs_crit(fs_info, |
6972 | "regular/prealloc extent found for non-regular inode %llu", | |
6973 | btrfs_ino(inode)); | |
6974 | goto out; | |
6975 | } | |
09ed2f16 LB |
6976 | trace_btrfs_get_extent_show_fi_regular(inode, leaf, item, |
6977 | extent_start); | |
694c12ed | 6978 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
09ed2f16 LB |
6979 | trace_btrfs_get_extent_show_fi_inline(inode, leaf, item, |
6980 | path->slots[0], | |
6981 | extent_start); | |
9036c102 | 6982 | } |
25a50341 | 6983 | next: |
9036c102 YZ |
6984 | if (start >= extent_end) { |
6985 | path->slots[0]++; | |
6986 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
6987 | ret = btrfs_next_leaf(root, path); | |
1028d1c4 | 6988 | if (ret < 0) |
9036c102 | 6989 | goto out; |
1028d1c4 | 6990 | else if (ret > 0) |
9036c102 | 6991 | goto not_found; |
1028d1c4 | 6992 | |
9036c102 | 6993 | leaf = path->nodes[0]; |
a52d9a80 | 6994 | } |
9036c102 YZ |
6995 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
6996 | if (found_key.objectid != objectid || | |
6997 | found_key.type != BTRFS_EXTENT_DATA_KEY) | |
6998 | goto not_found; | |
6999 | if (start + len <= found_key.offset) | |
7000 | goto not_found; | |
e2eca69d WS |
7001 | if (start > found_key.offset) |
7002 | goto next; | |
02a033df NB |
7003 | |
7004 | /* New extent overlaps with existing one */ | |
9036c102 | 7005 | em->start = start; |
70c8a91c | 7006 | em->orig_start = start; |
9036c102 | 7007 | em->len = found_key.offset - start; |
02a033df NB |
7008 | em->block_start = EXTENT_MAP_HOLE; |
7009 | goto insert; | |
9036c102 YZ |
7010 | } |
7011 | ||
280f15cb | 7012 | btrfs_extent_item_to_extent_map(inode, path, item, em); |
7ffbb598 | 7013 | |
694c12ed NB |
7014 | if (extent_type == BTRFS_FILE_EXTENT_REG || |
7015 | extent_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
a52d9a80 | 7016 | goto insert; |
694c12ed | 7017 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
affc5424 QW |
7018 | /* |
7019 | * Inline extent can only exist at file offset 0. This is | |
7020 | * ensured by tree-checker and inline extent creation path. | |
7021 | * Thus all members representing file offsets should be zero. | |
7022 | */ | |
affc5424 QW |
7023 | ASSERT(pg_offset == 0); |
7024 | ASSERT(extent_start == 0); | |
7025 | ASSERT(em->start == 0); | |
5f39d397 | 7026 | |
a196a894 QW |
7027 | /* |
7028 | * btrfs_extent_item_to_extent_map() should have properly | |
7029 | * initialized em members already. | |
7030 | * | |
7031 | * Other members are not utilized for inline extents. | |
7032 | */ | |
7033 | ASSERT(em->block_start == EXTENT_MAP_INLINE); | |
946c2923 | 7034 | ASSERT(em->len == fs_info->sectorsize); |
e49aabd9 | 7035 | |
a982fc82 QW |
7036 | ret = read_inline_extent(inode, path, page); |
7037 | if (ret < 0) | |
7038 | goto out; | |
a52d9a80 | 7039 | goto insert; |
a52d9a80 CM |
7040 | } |
7041 | not_found: | |
7042 | em->start = start; | |
70c8a91c | 7043 | em->orig_start = start; |
d1310b2e | 7044 | em->len = len; |
5f39d397 | 7045 | em->block_start = EXTENT_MAP_HOLE; |
a52d9a80 | 7046 | insert: |
1028d1c4 | 7047 | ret = 0; |
b3b4aa74 | 7048 | btrfs_release_path(path); |
d1310b2e | 7049 | if (em->start > start || extent_map_end(em) <= start) { |
0b246afa | 7050 | btrfs_err(fs_info, |
5d163e0e JM |
7051 | "bad extent! em: [%llu %llu] passed [%llu %llu]", |
7052 | em->start, em->len, start, len); | |
1028d1c4 | 7053 | ret = -EIO; |
a52d9a80 CM |
7054 | goto out; |
7055 | } | |
d1310b2e | 7056 | |
890871be | 7057 | write_lock(&em_tree->lock); |
1028d1c4 | 7058 | ret = btrfs_add_extent_mapping(fs_info, em_tree, &em, start, len); |
890871be | 7059 | write_unlock(&em_tree->lock); |
a52d9a80 | 7060 | out: |
c6414280 | 7061 | btrfs_free_path(path); |
1abe9b8a | 7062 | |
fc4f21b1 | 7063 | trace_btrfs_get_extent(root, inode, em); |
1abe9b8a | 7064 | |
1028d1c4 | 7065 | if (ret) { |
a52d9a80 | 7066 | free_extent_map(em); |
1028d1c4 | 7067 | return ERR_PTR(ret); |
a52d9a80 CM |
7068 | } |
7069 | return em; | |
7070 | } | |
7071 | ||
64f54188 | 7072 | static struct extent_map *btrfs_create_dio_extent(struct btrfs_inode *inode, |
53f2c206 | 7073 | struct btrfs_dio_data *dio_data, |
5f9a8a51 FM |
7074 | const u64 start, |
7075 | const u64 len, | |
7076 | const u64 orig_start, | |
7077 | const u64 block_start, | |
7078 | const u64 block_len, | |
7079 | const u64 orig_block_len, | |
7080 | const u64 ram_bytes, | |
7081 | const int type) | |
7082 | { | |
7083 | struct extent_map *em = NULL; | |
53f2c206 | 7084 | struct btrfs_ordered_extent *ordered; |
5f9a8a51 | 7085 | |
5f9a8a51 | 7086 | if (type != BTRFS_ORDERED_NOCOW) { |
64f54188 NB |
7087 | em = create_io_em(inode, start, len, orig_start, block_start, |
7088 | block_len, orig_block_len, ram_bytes, | |
6f9994db LB |
7089 | BTRFS_COMPRESS_NONE, /* compress_type */ |
7090 | type); | |
5f9a8a51 FM |
7091 | if (IS_ERR(em)) |
7092 | goto out; | |
7093 | } | |
53f2c206 BB |
7094 | ordered = btrfs_alloc_ordered_extent(inode, start, len, len, |
7095 | block_start, block_len, 0, | |
7096 | (1 << type) | | |
7097 | (1 << BTRFS_ORDERED_DIRECT), | |
7098 | BTRFS_COMPRESS_NONE); | |
7099 | if (IS_ERR(ordered)) { | |
5f9a8a51 FM |
7100 | if (em) { |
7101 | free_extent_map(em); | |
4c0c8cfc FM |
7102 | btrfs_drop_extent_map_range(inode, start, |
7103 | start + len - 1, false); | |
5f9a8a51 | 7104 | } |
53f2c206 BB |
7105 | em = ERR_CAST(ordered); |
7106 | } else { | |
7107 | ASSERT(!dio_data->ordered); | |
7108 | dio_data->ordered = ordered; | |
5f9a8a51 FM |
7109 | } |
7110 | out: | |
5f9a8a51 FM |
7111 | |
7112 | return em; | |
7113 | } | |
7114 | ||
9fc6f911 | 7115 | static struct extent_map *btrfs_new_extent_direct(struct btrfs_inode *inode, |
53f2c206 | 7116 | struct btrfs_dio_data *dio_data, |
4b46fce2 JB |
7117 | u64 start, u64 len) |
7118 | { | |
9fc6f911 NB |
7119 | struct btrfs_root *root = inode->root; |
7120 | struct btrfs_fs_info *fs_info = root->fs_info; | |
70c8a91c | 7121 | struct extent_map *em; |
4b46fce2 JB |
7122 | struct btrfs_key ins; |
7123 | u64 alloc_hint; | |
7124 | int ret; | |
4b46fce2 | 7125 | |
9fc6f911 | 7126 | alloc_hint = get_extent_allocation_hint(inode, start, len); |
0b246afa | 7127 | ret = btrfs_reserve_extent(root, len, len, fs_info->sectorsize, |
da17066c | 7128 | 0, alloc_hint, &ins, 1, 1); |
00361589 JB |
7129 | if (ret) |
7130 | return ERR_PTR(ret); | |
4b46fce2 | 7131 | |
53f2c206 | 7132 | em = btrfs_create_dio_extent(inode, dio_data, start, ins.offset, start, |
5f9a8a51 | 7133 | ins.objectid, ins.offset, ins.offset, |
6288d6ea | 7134 | ins.offset, BTRFS_ORDERED_REGULAR); |
0b246afa | 7135 | btrfs_dec_block_group_reservations(fs_info, ins.objectid); |
5f9a8a51 | 7136 | if (IS_ERR(em)) |
9fc6f911 NB |
7137 | btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, |
7138 | 1); | |
de0ee0ed | 7139 | |
4b46fce2 JB |
7140 | return em; |
7141 | } | |
7142 | ||
f4639636 | 7143 | static bool btrfs_extent_readonly(struct btrfs_fs_info *fs_info, u64 bytenr) |
05947ae1 AJ |
7144 | { |
7145 | struct btrfs_block_group *block_group; | |
f4639636 | 7146 | bool readonly = false; |
05947ae1 AJ |
7147 | |
7148 | block_group = btrfs_lookup_block_group(fs_info, bytenr); | |
7149 | if (!block_group || block_group->ro) | |
f4639636 | 7150 | readonly = true; |
05947ae1 AJ |
7151 | if (block_group) |
7152 | btrfs_put_block_group(block_group); | |
7153 | return readonly; | |
7154 | } | |
7155 | ||
46bfbb5c | 7156 | /* |
e4ecaf90 QW |
7157 | * Check if we can do nocow write into the range [@offset, @offset + @len) |
7158 | * | |
7159 | * @offset: File offset | |
7160 | * @len: The length to write, will be updated to the nocow writeable | |
7161 | * range | |
7162 | * @orig_start: (optional) Return the original file offset of the file extent | |
7163 | * @orig_len: (optional) Return the original on-disk length of the file extent | |
7164 | * @ram_bytes: (optional) Return the ram_bytes of the file extent | |
a84d5d42 BB |
7165 | * @strict: if true, omit optimizations that might force us into unnecessary |
7166 | * cow. e.g., don't trust generation number. | |
e4ecaf90 | 7167 | * |
e4ecaf90 QW |
7168 | * Return: |
7169 | * >0 and update @len if we can do nocow write | |
7170 | * 0 if we can't do nocow write | |
7171 | * <0 if error happened | |
7172 | * | |
7173 | * NOTE: This only checks the file extents, caller is responsible to wait for | |
7174 | * any ordered extents. | |
46bfbb5c | 7175 | */ |
00361589 | 7176 | noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len, |
7ee9e440 | 7177 | u64 *orig_start, u64 *orig_block_len, |
26ce9114 | 7178 | u64 *ram_bytes, bool nowait, bool strict) |
46bfbb5c | 7179 | { |
2ff7e61e | 7180 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
619104ba | 7181 | struct can_nocow_file_extent_args nocow_args = { 0 }; |
46bfbb5c CM |
7182 | struct btrfs_path *path; |
7183 | int ret; | |
7184 | struct extent_buffer *leaf; | |
7185 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
7b2b7085 | 7186 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
46bfbb5c CM |
7187 | struct btrfs_file_extent_item *fi; |
7188 | struct btrfs_key key; | |
46bfbb5c | 7189 | int found_type; |
e77751aa | 7190 | |
46bfbb5c CM |
7191 | path = btrfs_alloc_path(); |
7192 | if (!path) | |
7193 | return -ENOMEM; | |
26ce9114 | 7194 | path->nowait = nowait; |
46bfbb5c | 7195 | |
f85b7379 DS |
7196 | ret = btrfs_lookup_file_extent(NULL, root, path, |
7197 | btrfs_ino(BTRFS_I(inode)), offset, 0); | |
46bfbb5c CM |
7198 | if (ret < 0) |
7199 | goto out; | |
7200 | ||
46bfbb5c | 7201 | if (ret == 1) { |
619104ba | 7202 | if (path->slots[0] == 0) { |
46bfbb5c CM |
7203 | /* can't find the item, must cow */ |
7204 | ret = 0; | |
7205 | goto out; | |
7206 | } | |
619104ba | 7207 | path->slots[0]--; |
46bfbb5c CM |
7208 | } |
7209 | ret = 0; | |
7210 | leaf = path->nodes[0]; | |
619104ba | 7211 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
4a0cc7ca | 7212 | if (key.objectid != btrfs_ino(BTRFS_I(inode)) || |
46bfbb5c CM |
7213 | key.type != BTRFS_EXTENT_DATA_KEY) { |
7214 | /* not our file or wrong item type, must cow */ | |
7215 | goto out; | |
7216 | } | |
7217 | ||
7218 | if (key.offset > offset) { | |
7219 | /* Wrong offset, must cow */ | |
7220 | goto out; | |
7221 | } | |
7222 | ||
619104ba | 7223 | if (btrfs_file_extent_end(path) <= offset) |
7ee9e440 JB |
7224 | goto out; |
7225 | ||
619104ba FM |
7226 | fi = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_file_extent_item); |
7227 | found_type = btrfs_file_extent_type(leaf, fi); | |
7228 | if (ram_bytes) | |
7229 | *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi); | |
e77751aa | 7230 | |
619104ba FM |
7231 | nocow_args.start = offset; |
7232 | nocow_args.end = offset + *len - 1; | |
7233 | nocow_args.strict = strict; | |
7234 | nocow_args.free_path = true; | |
7ee9e440 | 7235 | |
619104ba FM |
7236 | ret = can_nocow_file_extent(path, &key, BTRFS_I(inode), &nocow_args); |
7237 | /* can_nocow_file_extent() has freed the path. */ | |
7238 | path = NULL; | |
7ee9e440 | 7239 | |
619104ba FM |
7240 | if (ret != 1) { |
7241 | /* Treat errors as not being able to NOCOW. */ | |
7242 | ret = 0; | |
78d4295b | 7243 | goto out; |
7ee9e440 | 7244 | } |
eb384b55 | 7245 | |
619104ba FM |
7246 | ret = 0; |
7247 | if (btrfs_extent_readonly(fs_info, nocow_args.disk_bytenr)) | |
46bfbb5c | 7248 | goto out; |
7b2b7085 | 7249 | |
619104ba FM |
7250 | if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) && |
7251 | found_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
7b2b7085 MX |
7252 | u64 range_end; |
7253 | ||
619104ba | 7254 | range_end = round_up(offset + nocow_args.num_bytes, |
da17066c | 7255 | root->fs_info->sectorsize) - 1; |
7b2b7085 MX |
7256 | ret = test_range_bit(io_tree, offset, range_end, |
7257 | EXTENT_DELALLOC, 0, NULL); | |
7258 | if (ret) { | |
7259 | ret = -EAGAIN; | |
7260 | goto out; | |
7261 | } | |
7262 | } | |
7263 | ||
619104ba FM |
7264 | if (orig_start) |
7265 | *orig_start = key.offset - nocow_args.extent_offset; | |
7266 | if (orig_block_len) | |
7267 | *orig_block_len = nocow_args.disk_num_bytes; | |
00361589 | 7268 | |
619104ba | 7269 | *len = nocow_args.num_bytes; |
46bfbb5c CM |
7270 | ret = 1; |
7271 | out: | |
7272 | btrfs_free_path(path); | |
7273 | return ret; | |
7274 | } | |
7275 | ||
eb838e73 | 7276 | static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend, |
59094403 FM |
7277 | struct extent_state **cached_state, |
7278 | unsigned int iomap_flags) | |
eb838e73 | 7279 | { |
59094403 FM |
7280 | const bool writing = (iomap_flags & IOMAP_WRITE); |
7281 | const bool nowait = (iomap_flags & IOMAP_NOWAIT); | |
7282 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; | |
eb838e73 JB |
7283 | struct btrfs_ordered_extent *ordered; |
7284 | int ret = 0; | |
7285 | ||
7286 | while (1) { | |
59094403 | 7287 | if (nowait) { |
83ae4133 JB |
7288 | if (!try_lock_extent(io_tree, lockstart, lockend, |
7289 | cached_state)) | |
59094403 FM |
7290 | return -EAGAIN; |
7291 | } else { | |
570eb97b | 7292 | lock_extent(io_tree, lockstart, lockend, cached_state); |
59094403 | 7293 | } |
eb838e73 JB |
7294 | /* |
7295 | * We're concerned with the entire range that we're going to be | |
01327610 | 7296 | * doing DIO to, so we need to make sure there's no ordered |
eb838e73 JB |
7297 | * extents in this range. |
7298 | */ | |
a776c6fa | 7299 | ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), lockstart, |
eb838e73 JB |
7300 | lockend - lockstart + 1); |
7301 | ||
7302 | /* | |
7303 | * We need to make sure there are no buffered pages in this | |
7304 | * range either, we could have raced between the invalidate in | |
7305 | * generic_file_direct_write and locking the extent. The | |
7306 | * invalidate needs to happen so that reads after a write do not | |
7307 | * get stale data. | |
7308 | */ | |
fc4adbff | 7309 | if (!ordered && |
051c98eb DS |
7310 | (!writing || !filemap_range_has_page(inode->i_mapping, |
7311 | lockstart, lockend))) | |
eb838e73 JB |
7312 | break; |
7313 | ||
570eb97b | 7314 | unlock_extent(io_tree, lockstart, lockend, cached_state); |
eb838e73 JB |
7315 | |
7316 | if (ordered) { | |
59094403 FM |
7317 | if (nowait) { |
7318 | btrfs_put_ordered_extent(ordered); | |
7319 | ret = -EAGAIN; | |
7320 | break; | |
7321 | } | |
ade77029 FM |
7322 | /* |
7323 | * If we are doing a DIO read and the ordered extent we | |
7324 | * found is for a buffered write, we can not wait for it | |
7325 | * to complete and retry, because if we do so we can | |
7326 | * deadlock with concurrent buffered writes on page | |
7327 | * locks. This happens only if our DIO read covers more | |
7328 | * than one extent map, if at this point has already | |
7329 | * created an ordered extent for a previous extent map | |
7330 | * and locked its range in the inode's io tree, and a | |
7331 | * concurrent write against that previous extent map's | |
7332 | * range and this range started (we unlock the ranges | |
7333 | * in the io tree only when the bios complete and | |
7334 | * buffered writes always lock pages before attempting | |
7335 | * to lock range in the io tree). | |
7336 | */ | |
7337 | if (writing || | |
7338 | test_bit(BTRFS_ORDERED_DIRECT, &ordered->flags)) | |
36d45567 | 7339 | btrfs_start_ordered_extent(ordered); |
ade77029 | 7340 | else |
59094403 | 7341 | ret = nowait ? -EAGAIN : -ENOTBLK; |
eb838e73 JB |
7342 | btrfs_put_ordered_extent(ordered); |
7343 | } else { | |
eb838e73 | 7344 | /* |
b850ae14 FM |
7345 | * We could trigger writeback for this range (and wait |
7346 | * for it to complete) and then invalidate the pages for | |
7347 | * this range (through invalidate_inode_pages2_range()), | |
7348 | * but that can lead us to a deadlock with a concurrent | |
ba206a02 | 7349 | * call to readahead (a buffered read or a defrag call |
b850ae14 FM |
7350 | * triggered a readahead) on a page lock due to an |
7351 | * ordered dio extent we created before but did not have | |
7352 | * yet a corresponding bio submitted (whence it can not | |
ba206a02 | 7353 | * complete), which makes readahead wait for that |
b850ae14 FM |
7354 | * ordered extent to complete while holding a lock on |
7355 | * that page. | |
eb838e73 | 7356 | */ |
59094403 | 7357 | ret = nowait ? -EAGAIN : -ENOTBLK; |
eb838e73 JB |
7358 | } |
7359 | ||
ade77029 FM |
7360 | if (ret) |
7361 | break; | |
7362 | ||
eb838e73 JB |
7363 | cond_resched(); |
7364 | } | |
7365 | ||
7366 | return ret; | |
7367 | } | |
7368 | ||
6f9994db | 7369 | /* The callers of this must take lock_extent() */ |
4b67c11d NB |
7370 | static struct extent_map *create_io_em(struct btrfs_inode *inode, u64 start, |
7371 | u64 len, u64 orig_start, u64 block_start, | |
6f9994db LB |
7372 | u64 block_len, u64 orig_block_len, |
7373 | u64 ram_bytes, int compress_type, | |
7374 | int type) | |
69ffb543 | 7375 | { |
69ffb543 | 7376 | struct extent_map *em; |
69ffb543 JB |
7377 | int ret; |
7378 | ||
6f9994db LB |
7379 | ASSERT(type == BTRFS_ORDERED_PREALLOC || |
7380 | type == BTRFS_ORDERED_COMPRESSED || | |
7381 | type == BTRFS_ORDERED_NOCOW || | |
1af4a0aa | 7382 | type == BTRFS_ORDERED_REGULAR); |
6f9994db | 7383 | |
69ffb543 JB |
7384 | em = alloc_extent_map(); |
7385 | if (!em) | |
7386 | return ERR_PTR(-ENOMEM); | |
7387 | ||
7388 | em->start = start; | |
7389 | em->orig_start = orig_start; | |
7390 | em->len = len; | |
7391 | em->block_len = block_len; | |
7392 | em->block_start = block_start; | |
b4939680 | 7393 | em->orig_block_len = orig_block_len; |
cc95bef6 | 7394 | em->ram_bytes = ram_bytes; |
70c8a91c | 7395 | em->generation = -1; |
69ffb543 | 7396 | set_bit(EXTENT_FLAG_PINNED, &em->flags); |
1af4a0aa | 7397 | if (type == BTRFS_ORDERED_PREALLOC) { |
b11e234d | 7398 | set_bit(EXTENT_FLAG_FILLING, &em->flags); |
1af4a0aa | 7399 | } else if (type == BTRFS_ORDERED_COMPRESSED) { |
6f9994db LB |
7400 | set_bit(EXTENT_FLAG_COMPRESSED, &em->flags); |
7401 | em->compress_type = compress_type; | |
7402 | } | |
69ffb543 | 7403 | |
a1ba4c08 | 7404 | ret = btrfs_replace_extent_map_range(inode, em, true); |
69ffb543 JB |
7405 | if (ret) { |
7406 | free_extent_map(em); | |
7407 | return ERR_PTR(ret); | |
7408 | } | |
7409 | ||
6f9994db | 7410 | /* em got 2 refs now, callers needs to do free_extent_map once. */ |
69ffb543 JB |
7411 | return em; |
7412 | } | |
7413 | ||
1c8d0175 | 7414 | |
c5794e51 | 7415 | static int btrfs_get_blocks_direct_write(struct extent_map **map, |
c5794e51 NB |
7416 | struct inode *inode, |
7417 | struct btrfs_dio_data *dio_data, | |
7833b865 | 7418 | u64 start, u64 *lenp, |
d7a8ab4e | 7419 | unsigned int iomap_flags) |
c5794e51 | 7420 | { |
d4135134 | 7421 | const bool nowait = (iomap_flags & IOMAP_NOWAIT); |
c5794e51 NB |
7422 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
7423 | struct extent_map *em = *map; | |
f0bfa76a FM |
7424 | int type; |
7425 | u64 block_start, orig_start, orig_block_len, ram_bytes; | |
2306e83e | 7426 | struct btrfs_block_group *bg; |
f0bfa76a FM |
7427 | bool can_nocow = false; |
7428 | bool space_reserved = false; | |
7833b865 | 7429 | u64 len = *lenp; |
6d82ad13 | 7430 | u64 prev_len; |
c5794e51 NB |
7431 | int ret = 0; |
7432 | ||
7433 | /* | |
7434 | * We don't allocate a new extent in the following cases | |
7435 | * | |
7436 | * 1) The inode is marked as NODATACOW. In this case we'll just use the | |
7437 | * existing extent. | |
7438 | * 2) The extent is marked as PREALLOC. We're good to go here and can | |
7439 | * just use the extent. | |
7440 | * | |
7441 | */ | |
7442 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) || | |
7443 | ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) && | |
7444 | em->block_start != EXTENT_MAP_HOLE)) { | |
c5794e51 NB |
7445 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) |
7446 | type = BTRFS_ORDERED_PREALLOC; | |
7447 | else | |
7448 | type = BTRFS_ORDERED_NOCOW; | |
7449 | len = min(len, em->len - (start - em->start)); | |
7450 | block_start = em->block_start + (start - em->start); | |
7451 | ||
7452 | if (can_nocow_extent(inode, start, &len, &orig_start, | |
26ce9114 | 7453 | &orig_block_len, &ram_bytes, false, false) == 1) { |
2306e83e FM |
7454 | bg = btrfs_inc_nocow_writers(fs_info, block_start); |
7455 | if (bg) | |
7456 | can_nocow = true; | |
7457 | } | |
f0bfa76a | 7458 | } |
c5794e51 | 7459 | |
6d82ad13 | 7460 | prev_len = len; |
f0bfa76a FM |
7461 | if (can_nocow) { |
7462 | struct extent_map *em2; | |
7463 | ||
7464 | /* We can NOCOW, so only need to reserve metadata space. */ | |
d4135134 FM |
7465 | ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode), len, len, |
7466 | nowait); | |
f0bfa76a FM |
7467 | if (ret < 0) { |
7468 | /* Our caller expects us to free the input extent map. */ | |
7469 | free_extent_map(em); | |
7470 | *map = NULL; | |
2306e83e | 7471 | btrfs_dec_nocow_writers(bg); |
d4135134 FM |
7472 | if (nowait && (ret == -ENOSPC || ret == -EDQUOT)) |
7473 | ret = -EAGAIN; | |
f0bfa76a FM |
7474 | goto out; |
7475 | } | |
7476 | space_reserved = true; | |
7477 | ||
53f2c206 | 7478 | em2 = btrfs_create_dio_extent(BTRFS_I(inode), dio_data, start, len, |
f0bfa76a FM |
7479 | orig_start, block_start, |
7480 | len, orig_block_len, | |
7481 | ram_bytes, type); | |
2306e83e | 7482 | btrfs_dec_nocow_writers(bg); |
f0bfa76a FM |
7483 | if (type == BTRFS_ORDERED_PREALLOC) { |
7484 | free_extent_map(em); | |
c1867eb3 DS |
7485 | *map = em2; |
7486 | em = em2; | |
f0bfa76a | 7487 | } |
c5794e51 | 7488 | |
f0bfa76a FM |
7489 | if (IS_ERR(em2)) { |
7490 | ret = PTR_ERR(em2); | |
7491 | goto out; | |
c5794e51 | 7492 | } |
f5585f4f FM |
7493 | |
7494 | dio_data->nocow_done = true; | |
f0bfa76a | 7495 | } else { |
f0bfa76a FM |
7496 | /* Our caller expects us to free the input extent map. */ |
7497 | free_extent_map(em); | |
7498 | *map = NULL; | |
7499 | ||
7833b865 CH |
7500 | if (nowait) { |
7501 | ret = -EAGAIN; | |
7502 | goto out; | |
7503 | } | |
d7a8ab4e | 7504 | |
f5585f4f FM |
7505 | /* |
7506 | * If we could not allocate data space before locking the file | |
7507 | * range and we can't do a NOCOW write, then we have to fail. | |
7508 | */ | |
7833b865 CH |
7509 | if (!dio_data->data_space_reserved) { |
7510 | ret = -ENOSPC; | |
7511 | goto out; | |
7512 | } | |
f5585f4f FM |
7513 | |
7514 | /* | |
7515 | * We have to COW and we have already reserved data space before, | |
7516 | * so now we reserve only metadata. | |
7517 | */ | |
7518 | ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode), len, len, | |
7519 | false); | |
f0bfa76a FM |
7520 | if (ret < 0) |
7521 | goto out; | |
7522 | space_reserved = true; | |
7523 | ||
53f2c206 | 7524 | em = btrfs_new_extent_direct(BTRFS_I(inode), dio_data, start, len); |
f0bfa76a FM |
7525 | if (IS_ERR(em)) { |
7526 | ret = PTR_ERR(em); | |
7527 | goto out; | |
7528 | } | |
7529 | *map = em; | |
7530 | len = min(len, em->len - (start - em->start)); | |
7531 | if (len < prev_len) | |
f5585f4f FM |
7532 | btrfs_delalloc_release_metadata(BTRFS_I(inode), |
7533 | prev_len - len, true); | |
c5794e51 NB |
7534 | } |
7535 | ||
f0bfa76a FM |
7536 | /* |
7537 | * We have created our ordered extent, so we can now release our reservation | |
7538 | * for an outstanding extent. | |
7539 | */ | |
6d82ad13 | 7540 | btrfs_delalloc_release_extents(BTRFS_I(inode), prev_len); |
c5794e51 | 7541 | |
c5794e51 NB |
7542 | /* |
7543 | * Need to update the i_size under the extent lock so buffered | |
7544 | * readers will get the updated i_size when we unlock. | |
7545 | */ | |
f85781fb | 7546 | if (start + len > i_size_read(inode)) |
c5794e51 | 7547 | i_size_write(inode, start + len); |
c5794e51 | 7548 | out: |
f0bfa76a FM |
7549 | if (ret && space_reserved) { |
7550 | btrfs_delalloc_release_extents(BTRFS_I(inode), len); | |
f5585f4f | 7551 | btrfs_delalloc_release_metadata(BTRFS_I(inode), len, true); |
f0bfa76a | 7552 | } |
7833b865 | 7553 | *lenp = len; |
c5794e51 NB |
7554 | return ret; |
7555 | } | |
7556 | ||
f85781fb GR |
7557 | static int btrfs_dio_iomap_begin(struct inode *inode, loff_t start, |
7558 | loff_t length, unsigned int flags, struct iomap *iomap, | |
7559 | struct iomap *srcmap) | |
4b46fce2 | 7560 | { |
491a6d01 | 7561 | struct iomap_iter *iter = container_of(iomap, struct iomap_iter, iomap); |
0b246afa | 7562 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
4b46fce2 | 7563 | struct extent_map *em; |
eb838e73 | 7564 | struct extent_state *cached_state = NULL; |
491a6d01 | 7565 | struct btrfs_dio_data *dio_data = iter->private; |
eb838e73 | 7566 | u64 lockstart, lockend; |
f85781fb | 7567 | const bool write = !!(flags & IOMAP_WRITE); |
0934856d | 7568 | int ret = 0; |
f85781fb | 7569 | u64 len = length; |
f5585f4f | 7570 | const u64 data_alloc_len = length; |
f85781fb | 7571 | bool unlock_extents = false; |
eb838e73 | 7572 | |
79d3d1d1 JB |
7573 | /* |
7574 | * We could potentially fault if we have a buffer > PAGE_SIZE, and if | |
7575 | * we're NOWAIT we may submit a bio for a partial range and return | |
7576 | * EIOCBQUEUED, which would result in an errant short read. | |
7577 | * | |
7578 | * The best way to handle this would be to allow for partial completions | |
7579 | * of iocb's, so we could submit the partial bio, return and fault in | |
7580 | * the rest of the pages, and then submit the io for the rest of the | |
7581 | * range. However we don't have that currently, so simply return | |
7582 | * -EAGAIN at this point so that the normal path is used. | |
7583 | */ | |
7584 | if (!write && (flags & IOMAP_NOWAIT) && length > PAGE_SIZE) | |
7585 | return -EAGAIN; | |
7586 | ||
ee5b46a3 CH |
7587 | /* |
7588 | * Cap the size of reads to that usually seen in buffered I/O as we need | |
7589 | * to allocate a contiguous array for the checksums. | |
7590 | */ | |
f85781fb | 7591 | if (!write) |
ee5b46a3 | 7592 | len = min_t(u64, len, fs_info->sectorsize * BTRFS_MAX_BIO_SECTORS); |
eb838e73 | 7593 | |
c329861d JB |
7594 | lockstart = start; |
7595 | lockend = start + len - 1; | |
7596 | ||
f85781fb | 7597 | /* |
b023e675 FM |
7598 | * iomap_dio_rw() only does filemap_write_and_wait_range(), which isn't |
7599 | * enough if we've written compressed pages to this area, so we need to | |
7600 | * flush the dirty pages again to make absolutely sure that any | |
7601 | * outstanding dirty pages are on disk - the first flush only starts | |
7602 | * compression on the data, while keeping the pages locked, so by the | |
7603 | * time the second flush returns we know bios for the compressed pages | |
7604 | * were submitted and finished, and the pages no longer under writeback. | |
7605 | * | |
7606 | * If we have a NOWAIT request and we have any pages in the range that | |
7607 | * are locked, likely due to compression still in progress, we don't want | |
7608 | * to block on page locks. We also don't want to block on pages marked as | |
7609 | * dirty or under writeback (same as for the non-compression case). | |
7610 | * iomap_dio_rw() did the same check, but after that and before we got | |
7611 | * here, mmap'ed writes may have happened or buffered reads started | |
7612 | * (readpage() and readahead(), which lock pages), as we haven't locked | |
7613 | * the file range yet. | |
f85781fb GR |
7614 | */ |
7615 | if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT, | |
7616 | &BTRFS_I(inode)->runtime_flags)) { | |
b023e675 FM |
7617 | if (flags & IOMAP_NOWAIT) { |
7618 | if (filemap_range_needs_writeback(inode->i_mapping, | |
7619 | lockstart, lockend)) | |
7620 | return -EAGAIN; | |
7621 | } else { | |
7622 | ret = filemap_fdatawrite_range(inode->i_mapping, start, | |
7623 | start + length - 1); | |
7624 | if (ret) | |
7625 | return ret; | |
7626 | } | |
f85781fb GR |
7627 | } |
7628 | ||
491a6d01 | 7629 | memset(dio_data, 0, sizeof(*dio_data)); |
f85781fb | 7630 | |
f5585f4f FM |
7631 | /* |
7632 | * We always try to allocate data space and must do it before locking | |
7633 | * the file range, to avoid deadlocks with concurrent writes to the same | |
7634 | * range if the range has several extents and the writes don't expand the | |
7635 | * current i_size (the inode lock is taken in shared mode). If we fail to | |
7636 | * allocate data space here we continue and later, after locking the | |
7637 | * file range, we fail with ENOSPC only if we figure out we can not do a | |
7638 | * NOCOW write. | |
7639 | */ | |
7640 | if (write && !(flags & IOMAP_NOWAIT)) { | |
7641 | ret = btrfs_check_data_free_space(BTRFS_I(inode), | |
7642 | &dio_data->data_reserved, | |
1daedb1d | 7643 | start, data_alloc_len, false); |
f5585f4f FM |
7644 | if (!ret) |
7645 | dio_data->data_space_reserved = true; | |
7646 | else if (ret && !(BTRFS_I(inode)->flags & | |
7647 | (BTRFS_INODE_NODATACOW | BTRFS_INODE_PREALLOC))) | |
7648 | goto err; | |
7649 | } | |
e1cbbfa5 | 7650 | |
eb838e73 JB |
7651 | /* |
7652 | * If this errors out it's because we couldn't invalidate pagecache for | |
59094403 FM |
7653 | * this range and we need to fallback to buffered IO, or we are doing a |
7654 | * NOWAIT read/write and we need to block. | |
eb838e73 | 7655 | */ |
59094403 FM |
7656 | ret = lock_extent_direct(inode, lockstart, lockend, &cached_state, flags); |
7657 | if (ret < 0) | |
9c9464cc | 7658 | goto err; |
eb838e73 | 7659 | |
39b07b5d | 7660 | em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len); |
eb838e73 JB |
7661 | if (IS_ERR(em)) { |
7662 | ret = PTR_ERR(em); | |
7663 | goto unlock_err; | |
7664 | } | |
4b46fce2 JB |
7665 | |
7666 | /* | |
7667 | * Ok for INLINE and COMPRESSED extents we need to fallback on buffered | |
7668 | * io. INLINE is special, and we could probably kludge it in here, but | |
7669 | * it's still buffered so for safety lets just fall back to the generic | |
7670 | * buffered path. | |
7671 | * | |
7672 | * For COMPRESSED we _have_ to read the entire extent in so we can | |
7673 | * decompress it, so there will be buffering required no matter what we | |
7674 | * do, so go ahead and fallback to buffered. | |
7675 | * | |
01327610 | 7676 | * We return -ENOTBLK because that's what makes DIO go ahead and go back |
4b46fce2 JB |
7677 | * to buffered IO. Don't blame me, this is the price we pay for using |
7678 | * the generic code. | |
7679 | */ | |
7680 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) || | |
7681 | em->block_start == EXTENT_MAP_INLINE) { | |
7682 | free_extent_map(em); | |
a4527e18 FM |
7683 | /* |
7684 | * If we are in a NOWAIT context, return -EAGAIN in order to | |
7685 | * fallback to buffered IO. This is not only because we can | |
7686 | * block with buffered IO (no support for NOWAIT semantics at | |
7687 | * the moment) but also to avoid returning short reads to user | |
7688 | * space - this happens if we were able to read some data from | |
7689 | * previous non-compressed extents and then when we fallback to | |
7690 | * buffered IO, at btrfs_file_read_iter() by calling | |
7691 | * filemap_read(), we fail to fault in pages for the read buffer, | |
7692 | * in which case filemap_read() returns a short read (the number | |
7693 | * of bytes previously read is > 0, so it does not return -EFAULT). | |
7694 | */ | |
7695 | ret = (flags & IOMAP_NOWAIT) ? -EAGAIN : -ENOTBLK; | |
eb838e73 | 7696 | goto unlock_err; |
4b46fce2 JB |
7697 | } |
7698 | ||
f85781fb | 7699 | len = min(len, em->len - (start - em->start)); |
ca93e44b FM |
7700 | |
7701 | /* | |
7702 | * If we have a NOWAIT request and the range contains multiple extents | |
7703 | * (or a mix of extents and holes), then we return -EAGAIN to make the | |
7704 | * caller fallback to a context where it can do a blocking (without | |
7705 | * NOWAIT) request. This way we avoid doing partial IO and returning | |
7706 | * success to the caller, which is not optimal for writes and for reads | |
7707 | * it can result in unexpected behaviour for an application. | |
7708 | * | |
7709 | * When doing a read, because we use IOMAP_DIO_PARTIAL when calling | |
7710 | * iomap_dio_rw(), we can end up returning less data then what the caller | |
7711 | * asked for, resulting in an unexpected, and incorrect, short read. | |
7712 | * That is, the caller asked to read N bytes and we return less than that, | |
7713 | * which is wrong unless we are crossing EOF. This happens if we get a | |
7714 | * page fault error when trying to fault in pages for the buffer that is | |
7715 | * associated to the struct iov_iter passed to iomap_dio_rw(), and we | |
7716 | * have previously submitted bios for other extents in the range, in | |
7717 | * which case iomap_dio_rw() may return us EIOCBQUEUED if not all of | |
7718 | * those bios have completed by the time we get the page fault error, | |
7719 | * which we return back to our caller - we should only return EIOCBQUEUED | |
7720 | * after we have submitted bios for all the extents in the range. | |
7721 | */ | |
7722 | if ((flags & IOMAP_NOWAIT) && len < length) { | |
7723 | free_extent_map(em); | |
7724 | ret = -EAGAIN; | |
7725 | goto unlock_err; | |
7726 | } | |
7727 | ||
f85781fb GR |
7728 | if (write) { |
7729 | ret = btrfs_get_blocks_direct_write(&em, inode, dio_data, | |
7833b865 | 7730 | start, &len, flags); |
c5794e51 NB |
7731 | if (ret < 0) |
7732 | goto unlock_err; | |
f85781fb GR |
7733 | unlock_extents = true; |
7734 | /* Recalc len in case the new em is smaller than requested */ | |
7735 | len = min(len, em->len - (start - em->start)); | |
f5585f4f FM |
7736 | if (dio_data->data_space_reserved) { |
7737 | u64 release_offset; | |
7738 | u64 release_len = 0; | |
7739 | ||
7740 | if (dio_data->nocow_done) { | |
7741 | release_offset = start; | |
7742 | release_len = data_alloc_len; | |
7743 | } else if (len < data_alloc_len) { | |
7744 | release_offset = start + len; | |
7745 | release_len = data_alloc_len - len; | |
7746 | } | |
7747 | ||
7748 | if (release_len > 0) | |
7749 | btrfs_free_reserved_data_space(BTRFS_I(inode), | |
7750 | dio_data->data_reserved, | |
7751 | release_offset, | |
7752 | release_len); | |
7753 | } | |
c5794e51 | 7754 | } else { |
1c8d0175 NB |
7755 | /* |
7756 | * We need to unlock only the end area that we aren't using. | |
7757 | * The rest is going to be unlocked by the endio routine. | |
7758 | */ | |
f85781fb GR |
7759 | lockstart = start + len; |
7760 | if (lockstart < lockend) | |
7761 | unlock_extents = true; | |
7762 | } | |
7763 | ||
7764 | if (unlock_extents) | |
570eb97b JB |
7765 | unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend, |
7766 | &cached_state); | |
f85781fb GR |
7767 | else |
7768 | free_extent_state(cached_state); | |
7769 | ||
7770 | /* | |
7771 | * Translate extent map information to iomap. | |
7772 | * We trim the extents (and move the addr) even though iomap code does | |
7773 | * that, since we have locked only the parts we are performing I/O in. | |
7774 | */ | |
7775 | if ((em->block_start == EXTENT_MAP_HOLE) || | |
7776 | (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) && !write)) { | |
7777 | iomap->addr = IOMAP_NULL_ADDR; | |
7778 | iomap->type = IOMAP_HOLE; | |
7779 | } else { | |
7780 | iomap->addr = em->block_start + (start - em->start); | |
7781 | iomap->type = IOMAP_MAPPED; | |
a43a67a2 | 7782 | } |
f85781fb | 7783 | iomap->offset = start; |
d24fa5c1 | 7784 | iomap->bdev = fs_info->fs_devices->latest_dev->bdev; |
f85781fb | 7785 | iomap->length = len; |
4b46fce2 JB |
7786 | free_extent_map(em); |
7787 | ||
7788 | return 0; | |
eb838e73 JB |
7789 | |
7790 | unlock_err: | |
570eb97b JB |
7791 | unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend, |
7792 | &cached_state); | |
9c9464cc | 7793 | err: |
f5585f4f FM |
7794 | if (dio_data->data_space_reserved) { |
7795 | btrfs_free_reserved_data_space(BTRFS_I(inode), | |
7796 | dio_data->data_reserved, | |
7797 | start, data_alloc_len); | |
7798 | extent_changeset_free(dio_data->data_reserved); | |
7799 | } | |
7800 | ||
f85781fb GR |
7801 | return ret; |
7802 | } | |
7803 | ||
7804 | static int btrfs_dio_iomap_end(struct inode *inode, loff_t pos, loff_t length, | |
7805 | ssize_t written, unsigned int flags, struct iomap *iomap) | |
7806 | { | |
491a6d01 CH |
7807 | struct iomap_iter *iter = container_of(iomap, struct iomap_iter, iomap); |
7808 | struct btrfs_dio_data *dio_data = iter->private; | |
f85781fb GR |
7809 | size_t submitted = dio_data->submitted; |
7810 | const bool write = !!(flags & IOMAP_WRITE); | |
491a6d01 | 7811 | int ret = 0; |
f85781fb GR |
7812 | |
7813 | if (!write && (iomap->type == IOMAP_HOLE)) { | |
7814 | /* If reading from a hole, unlock and return */ | |
570eb97b JB |
7815 | unlock_extent(&BTRFS_I(inode)->io_tree, pos, pos + length - 1, |
7816 | NULL); | |
491a6d01 | 7817 | return 0; |
f85781fb GR |
7818 | } |
7819 | ||
7820 | if (submitted < length) { | |
7821 | pos += submitted; | |
7822 | length -= submitted; | |
7823 | if (write) | |
b41b6f69 CH |
7824 | btrfs_finish_ordered_extent(dio_data->ordered, NULL, |
7825 | pos, length, false); | |
f85781fb GR |
7826 | else |
7827 | unlock_extent(&BTRFS_I(inode)->io_tree, pos, | |
570eb97b | 7828 | pos + length - 1, NULL); |
f85781fb GR |
7829 | ret = -ENOTBLK; |
7830 | } | |
53f2c206 BB |
7831 | if (write) { |
7832 | btrfs_put_ordered_extent(dio_data->ordered); | |
7833 | dio_data->ordered = NULL; | |
7834 | } | |
f85781fb | 7835 | |
f0bfa76a | 7836 | if (write) |
f85781fb | 7837 | extent_changeset_free(dio_data->data_reserved); |
8b110e39 MX |
7838 | return ret; |
7839 | } | |
7840 | ||
67d66982 | 7841 | static void btrfs_dio_end_io(struct btrfs_bio *bbio) |
8b110e39 | 7842 | { |
67d66982 CH |
7843 | struct btrfs_dio_private *dip = |
7844 | container_of(bbio, struct btrfs_dio_private, bbio); | |
7845 | struct btrfs_inode *inode = bbio->inode; | |
917f32a2 | 7846 | struct bio *bio = &bbio->bio; |
e65e1535 | 7847 | |
67d66982 CH |
7848 | if (bio->bi_status) { |
7849 | btrfs_warn(inode->root->fs_info, | |
7850 | "direct IO failed ino %llu op 0x%0x offset %#llx len %u err no %d", | |
7851 | btrfs_ino(inode), bio->bi_opf, | |
7852 | dip->file_offset, dip->bytes, bio->bi_status); | |
7609afac | 7853 | } |
1ae39938 | 7854 | |
b41b6f69 CH |
7855 | if (btrfs_op(bio) == BTRFS_MAP_WRITE) { |
7856 | btrfs_finish_ordered_extent(bbio->ordered, NULL, | |
7857 | dip->file_offset, dip->bytes, | |
7858 | !bio->bi_status); | |
7859 | } else { | |
67d66982 CH |
7860 | unlock_extent(&inode->io_tree, dip->file_offset, |
7861 | dip->file_offset + dip->bytes - 1, NULL); | |
b41b6f69 | 7862 | } |
ea1f0ced | 7863 | |
67d66982 CH |
7864 | bbio->bio.bi_private = bbio->private; |
7865 | iomap_dio_bio_end_io(bio); | |
e65e1535 MX |
7866 | } |
7867 | ||
67d66982 CH |
7868 | static void btrfs_dio_submit_io(const struct iomap_iter *iter, struct bio *bio, |
7869 | loff_t file_offset) | |
c36cac28 | 7870 | { |
67d66982 | 7871 | struct btrfs_bio *bbio = btrfs_bio(bio); |
642c5d34 | 7872 | struct btrfs_dio_private *dip = |
67d66982 | 7873 | container_of(bbio, struct btrfs_dio_private, bbio); |
491a6d01 | 7874 | struct btrfs_dio_data *dio_data = iter->private; |
544d24f9 | 7875 | |
4317ff00 QW |
7876 | btrfs_bio_init(bbio, BTRFS_I(iter->inode)->root->fs_info, |
7877 | btrfs_dio_end_io, bio->bi_private); | |
7878 | bbio->inode = BTRFS_I(iter->inode); | |
67d66982 | 7879 | bbio->file_offset = file_offset; |
e65e1535 | 7880 | |
67d66982 CH |
7881 | dip->file_offset = file_offset; |
7882 | dip->bytes = bio->bi_iter.bi_size; | |
e65e1535 | 7883 | |
67d66982 | 7884 | dio_data->submitted += bio->bi_iter.bi_size; |
b73a6fd1 BB |
7885 | |
7886 | /* | |
7887 | * Check if we are doing a partial write. If we are, we need to split | |
7888 | * the ordered extent to match the submitted bio. Hang on to the | |
7889 | * remaining unfinishable ordered_extent in dio_data so that it can be | |
7890 | * cancelled in iomap_end to avoid a deadlock wherein faulting the | |
7891 | * remaining pages is blocked on the outstanding ordered extent. | |
7892 | */ | |
7893 | if (iter->flags & IOMAP_WRITE) { | |
7894 | int ret; | |
7895 | ||
7896 | ret = btrfs_extract_ordered_extent(bbio, dio_data->ordered); | |
7897 | if (ret) { | |
7cad645e CH |
7898 | btrfs_finish_ordered_extent(dio_data->ordered, NULL, |
7899 | file_offset, dip->bytes, | |
7900 | !ret); | |
7901 | bio->bi_status = errno_to_blk_status(ret); | |
7902 | iomap_dio_bio_end_io(bio); | |
b73a6fd1 BB |
7903 | return; |
7904 | } | |
7905 | } | |
7906 | ||
ae42a154 | 7907 | btrfs_submit_bio(bbio, 0); |
4b46fce2 JB |
7908 | } |
7909 | ||
36e8c622 | 7910 | static const struct iomap_ops btrfs_dio_iomap_ops = { |
f85781fb GR |
7911 | .iomap_begin = btrfs_dio_iomap_begin, |
7912 | .iomap_end = btrfs_dio_iomap_end, | |
7913 | }; | |
7914 | ||
36e8c622 | 7915 | static const struct iomap_dio_ops btrfs_dio_ops = { |
67d66982 | 7916 | .submit_io = btrfs_dio_submit_io, |
642c5d34 | 7917 | .bio_set = &btrfs_dio_bioset, |
f85781fb GR |
7918 | }; |
7919 | ||
8184620a | 7920 | ssize_t btrfs_dio_read(struct kiocb *iocb, struct iov_iter *iter, size_t done_before) |
36e8c622 | 7921 | { |
53f2c206 | 7922 | struct btrfs_dio_data data = { 0 }; |
491a6d01 | 7923 | |
36e8c622 | 7924 | return iomap_dio_rw(iocb, iter, &btrfs_dio_iomap_ops, &btrfs_dio_ops, |
8184620a FM |
7925 | IOMAP_DIO_PARTIAL, &data, done_before); |
7926 | } | |
7927 | ||
7928 | struct iomap_dio *btrfs_dio_write(struct kiocb *iocb, struct iov_iter *iter, | |
7929 | size_t done_before) | |
7930 | { | |
53f2c206 | 7931 | struct btrfs_dio_data data = { 0 }; |
8184620a FM |
7932 | |
7933 | return __iomap_dio_rw(iocb, iter, &btrfs_dio_iomap_ops, &btrfs_dio_ops, | |
7934 | IOMAP_DIO_PARTIAL, &data, done_before); | |
36e8c622 CH |
7935 | } |
7936 | ||
1506fcc8 | 7937 | static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, |
bab16e21 | 7938 | u64 start, u64 len) |
1506fcc8 | 7939 | { |
05dadc09 TI |
7940 | int ret; |
7941 | ||
45dd052e | 7942 | ret = fiemap_prep(inode, fieinfo, start, &len, 0); |
05dadc09 TI |
7943 | if (ret) |
7944 | return ret; | |
7945 | ||
33a86cfa FM |
7946 | /* |
7947 | * fiemap_prep() called filemap_write_and_wait() for the whole possible | |
7948 | * file range (0 to LLONG_MAX), but that is not enough if we have | |
7949 | * compression enabled. The first filemap_fdatawrite_range() only kicks | |
7950 | * in the compression of data (in an async thread) and will return | |
7951 | * before the compression is done and writeback is started. A second | |
7952 | * filemap_fdatawrite_range() is needed to wait for the compression to | |
ac3c0d36 FM |
7953 | * complete and writeback to start. We also need to wait for ordered |
7954 | * extents to complete, because our fiemap implementation uses mainly | |
7955 | * file extent items to list the extents, searching for extent maps | |
7956 | * only for file ranges with holes or prealloc extents to figure out | |
7957 | * if we have delalloc in those ranges. | |
33a86cfa FM |
7958 | */ |
7959 | if (fieinfo->fi_flags & FIEMAP_FLAG_SYNC) { | |
ac3c0d36 | 7960 | ret = btrfs_wait_ordered_range(inode, 0, LLONG_MAX); |
33a86cfa FM |
7961 | if (ret) |
7962 | return ret; | |
7963 | } | |
7964 | ||
facee0a0 | 7965 | return extent_fiemap(BTRFS_I(inode), fieinfo, start, len); |
1506fcc8 YS |
7966 | } |
7967 | ||
48a3b636 ES |
7968 | static int btrfs_writepages(struct address_space *mapping, |
7969 | struct writeback_control *wbc) | |
b293f02e | 7970 | { |
8ae225a8 | 7971 | return extent_writepages(mapping, wbc); |
b293f02e CM |
7972 | } |
7973 | ||
ba206a02 | 7974 | static void btrfs_readahead(struct readahead_control *rac) |
3ab2fb5a | 7975 | { |
ba206a02 | 7976 | extent_readahead(rac); |
3ab2fb5a | 7977 | } |
2a3ff0ad | 7978 | |
7c11d0ae | 7979 | /* |
f913cff3 | 7980 | * For release_folio() and invalidate_folio() we have a race window where |
895586eb | 7981 | * folio_end_writeback() is called but the subpage spinlock is not yet released. |
7c11d0ae QW |
7982 | * If we continue to release/invalidate the page, we could cause use-after-free |
7983 | * for subpage spinlock. So this function is to spin and wait for subpage | |
7984 | * spinlock. | |
7985 | */ | |
7986 | static void wait_subpage_spinlock(struct page *page) | |
7987 | { | |
7988 | struct btrfs_fs_info *fs_info = btrfs_sb(page->mapping->host->i_sb); | |
7989 | struct btrfs_subpage *subpage; | |
7990 | ||
fbca46eb | 7991 | if (!btrfs_is_subpage(fs_info, page)) |
7c11d0ae QW |
7992 | return; |
7993 | ||
7994 | ASSERT(PagePrivate(page) && page->private); | |
7995 | subpage = (struct btrfs_subpage *)page->private; | |
7996 | ||
7997 | /* | |
7998 | * This may look insane as we just acquire the spinlock and release it, | |
7999 | * without doing anything. But we just want to make sure no one is | |
8000 | * still holding the subpage spinlock. | |
8001 | * And since the page is not dirty nor writeback, and we have page | |
8002 | * locked, the only possible way to hold a spinlock is from the endio | |
8003 | * function to clear page writeback. | |
8004 | * | |
8005 | * Here we just acquire the spinlock so that all existing callers | |
8006 | * should exit and we're safe to release/invalidate the page. | |
8007 | */ | |
8008 | spin_lock_irq(&subpage->lock); | |
8009 | spin_unlock_irq(&subpage->lock); | |
8010 | } | |
8011 | ||
f913cff3 | 8012 | static bool __btrfs_release_folio(struct folio *folio, gfp_t gfp_flags) |
9ebefb18 | 8013 | { |
f913cff3 | 8014 | int ret = try_release_extent_mapping(&folio->page, gfp_flags); |
7c11d0ae QW |
8015 | |
8016 | if (ret == 1) { | |
f913cff3 MWO |
8017 | wait_subpage_spinlock(&folio->page); |
8018 | clear_page_extent_mapped(&folio->page); | |
7c11d0ae | 8019 | } |
a52d9a80 | 8020 | return ret; |
39279cc3 CM |
8021 | } |
8022 | ||
f913cff3 | 8023 | static bool btrfs_release_folio(struct folio *folio, gfp_t gfp_flags) |
e6dcd2dc | 8024 | { |
f913cff3 MWO |
8025 | if (folio_test_writeback(folio) || folio_test_dirty(folio)) |
8026 | return false; | |
8027 | return __btrfs_release_folio(folio, gfp_flags); | |
e6dcd2dc CM |
8028 | } |
8029 | ||
f8e66081 | 8030 | #ifdef CONFIG_MIGRATION |
e7a60a17 MWO |
8031 | static int btrfs_migrate_folio(struct address_space *mapping, |
8032 | struct folio *dst, struct folio *src, | |
f8e66081 RG |
8033 | enum migrate_mode mode) |
8034 | { | |
e7a60a17 | 8035 | int ret = filemap_migrate_folio(mapping, dst, src, mode); |
f8e66081 | 8036 | |
f8e66081 RG |
8037 | if (ret != MIGRATEPAGE_SUCCESS) |
8038 | return ret; | |
8039 | ||
e7a60a17 MWO |
8040 | if (folio_test_ordered(src)) { |
8041 | folio_clear_ordered(src); | |
8042 | folio_set_ordered(dst); | |
f8e66081 RG |
8043 | } |
8044 | ||
f8e66081 RG |
8045 | return MIGRATEPAGE_SUCCESS; |
8046 | } | |
e7a60a17 MWO |
8047 | #else |
8048 | #define btrfs_migrate_folio NULL | |
f8e66081 RG |
8049 | #endif |
8050 | ||
895586eb MWO |
8051 | static void btrfs_invalidate_folio(struct folio *folio, size_t offset, |
8052 | size_t length) | |
39279cc3 | 8053 | { |
895586eb | 8054 | struct btrfs_inode *inode = BTRFS_I(folio->mapping->host); |
b945a463 | 8055 | struct btrfs_fs_info *fs_info = inode->root->fs_info; |
53ac7ead | 8056 | struct extent_io_tree *tree = &inode->io_tree; |
2ac55d41 | 8057 | struct extent_state *cached_state = NULL; |
895586eb MWO |
8058 | u64 page_start = folio_pos(folio); |
8059 | u64 page_end = page_start + folio_size(folio) - 1; | |
3b835840 | 8060 | u64 cur; |
53ac7ead | 8061 | int inode_evicting = inode->vfs_inode.i_state & I_FREEING; |
39279cc3 | 8062 | |
8b62b72b | 8063 | /* |
895586eb MWO |
8064 | * We have folio locked so no new ordered extent can be created on this |
8065 | * page, nor bio can be submitted for this folio. | |
8b62b72b | 8066 | * |
895586eb MWO |
8067 | * But already submitted bio can still be finished on this folio. |
8068 | * Furthermore, endio function won't skip folio which has Ordered | |
f57ad937 | 8069 | * (Private2) already cleared, so it's possible for endio and |
895586eb MWO |
8070 | * invalidate_folio to do the same ordered extent accounting twice |
8071 | * on one folio. | |
266a2586 QW |
8072 | * |
8073 | * So here we wait for any submitted bios to finish, so that we won't | |
895586eb | 8074 | * do double ordered extent accounting on the same folio. |
8b62b72b | 8075 | */ |
895586eb MWO |
8076 | folio_wait_writeback(folio); |
8077 | wait_subpage_spinlock(&folio->page); | |
8b62b72b | 8078 | |
bcd77455 QW |
8079 | /* |
8080 | * For subpage case, we have call sites like | |
8081 | * btrfs_punch_hole_lock_range() which passes range not aligned to | |
8082 | * sectorsize. | |
895586eb MWO |
8083 | * If the range doesn't cover the full folio, we don't need to and |
8084 | * shouldn't clear page extent mapped, as folio->private can still | |
bcd77455 QW |
8085 | * record subpage dirty bits for other part of the range. |
8086 | * | |
895586eb MWO |
8087 | * For cases that invalidate the full folio even the range doesn't |
8088 | * cover the full folio, like invalidating the last folio, we're | |
bcd77455 QW |
8089 | * still safe to wait for ordered extent to finish. |
8090 | */ | |
5a60542c | 8091 | if (!(offset == 0 && length == folio_size(folio))) { |
f913cff3 | 8092 | btrfs_release_folio(folio, GFP_NOFS); |
e6dcd2dc CM |
8093 | return; |
8094 | } | |
131e404a FDBM |
8095 | |
8096 | if (!inode_evicting) | |
570eb97b | 8097 | lock_extent(tree, page_start, page_end, &cached_state); |
951c80f8 | 8098 | |
3b835840 QW |
8099 | cur = page_start; |
8100 | while (cur < page_end) { | |
8101 | struct btrfs_ordered_extent *ordered; | |
3b835840 | 8102 | u64 range_end; |
b945a463 | 8103 | u32 range_len; |
bd015294 | 8104 | u32 extra_flags = 0; |
3b835840 QW |
8105 | |
8106 | ordered = btrfs_lookup_first_ordered_range(inode, cur, | |
8107 | page_end + 1 - cur); | |
8108 | if (!ordered) { | |
8109 | range_end = page_end; | |
8110 | /* | |
8111 | * No ordered extent covering this range, we are safe | |
8112 | * to delete all extent states in the range. | |
8113 | */ | |
bd015294 | 8114 | extra_flags = EXTENT_CLEAR_ALL_BITS; |
3b835840 QW |
8115 | goto next; |
8116 | } | |
8117 | if (ordered->file_offset > cur) { | |
8118 | /* | |
8119 | * There is a range between [cur, oe->file_offset) not | |
8120 | * covered by any ordered extent. | |
8121 | * We are safe to delete all extent states, and handle | |
8122 | * the ordered extent in the next iteration. | |
8123 | */ | |
8124 | range_end = ordered->file_offset - 1; | |
bd015294 | 8125 | extra_flags = EXTENT_CLEAR_ALL_BITS; |
3b835840 QW |
8126 | goto next; |
8127 | } | |
8128 | ||
8129 | range_end = min(ordered->file_offset + ordered->num_bytes - 1, | |
8130 | page_end); | |
b945a463 QW |
8131 | ASSERT(range_end + 1 - cur < U32_MAX); |
8132 | range_len = range_end + 1 - cur; | |
895586eb | 8133 | if (!btrfs_page_test_ordered(fs_info, &folio->page, cur, range_len)) { |
3b835840 | 8134 | /* |
f57ad937 QW |
8135 | * If Ordered (Private2) is cleared, it means endio has |
8136 | * already been executed for the range. | |
3b835840 QW |
8137 | * We can't delete the extent states as |
8138 | * btrfs_finish_ordered_io() may still use some of them. | |
8139 | */ | |
3b835840 QW |
8140 | goto next; |
8141 | } | |
895586eb | 8142 | btrfs_page_clear_ordered(fs_info, &folio->page, cur, range_len); |
3b835840 | 8143 | |
eb84ae03 | 8144 | /* |
2766ff61 FM |
8145 | * IO on this page will never be started, so we need to account |
8146 | * for any ordered extents now. Don't clear EXTENT_DELALLOC_NEW | |
8147 | * here, must leave that up for the ordered extent completion. | |
3b835840 QW |
8148 | * |
8149 | * This will also unlock the range for incoming | |
8150 | * btrfs_finish_ordered_io(). | |
eb84ae03 | 8151 | */ |
131e404a | 8152 | if (!inode_evicting) |
3b835840 | 8153 | clear_extent_bit(tree, cur, range_end, |
2766ff61 | 8154 | EXTENT_DELALLOC | |
131e404a | 8155 | EXTENT_LOCKED | EXTENT_DO_ACCOUNTING | |
bd015294 | 8156 | EXTENT_DEFRAG, &cached_state); |
3b835840 QW |
8157 | |
8158 | spin_lock_irq(&inode->ordered_tree.lock); | |
8159 | set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags); | |
8160 | ordered->truncated_len = min(ordered->truncated_len, | |
8161 | cur - ordered->file_offset); | |
8162 | spin_unlock_irq(&inode->ordered_tree.lock); | |
8163 | ||
bd015294 JB |
8164 | /* |
8165 | * If the ordered extent has finished, we're safe to delete all | |
8166 | * the extent states of the range, otherwise | |
8167 | * btrfs_finish_ordered_io() will get executed by endio for | |
8168 | * other pages, so we can't delete extent states. | |
8169 | */ | |
3b835840 | 8170 | if (btrfs_dec_test_ordered_pending(inode, &ordered, |
f41b6ba9 | 8171 | cur, range_end + 1 - cur)) { |
3b835840 QW |
8172 | btrfs_finish_ordered_io(ordered); |
8173 | /* | |
8174 | * The ordered extent has finished, now we're again | |
8175 | * safe to delete all extent states of the range. | |
8176 | */ | |
bd015294 | 8177 | extra_flags = EXTENT_CLEAR_ALL_BITS; |
3b835840 QW |
8178 | } |
8179 | next: | |
8180 | if (ordered) | |
8181 | btrfs_put_ordered_extent(ordered); | |
8b62b72b | 8182 | /* |
3b835840 QW |
8183 | * Qgroup reserved space handler |
8184 | * Sector(s) here will be either: | |
266a2586 | 8185 | * |
3b835840 QW |
8186 | * 1) Already written to disk or bio already finished |
8187 | * Then its QGROUP_RESERVED bit in io_tree is already cleared. | |
8188 | * Qgroup will be handled by its qgroup_record then. | |
8189 | * btrfs_qgroup_free_data() call will do nothing here. | |
8190 | * | |
8191 | * 2) Not written to disk yet | |
8192 | * Then btrfs_qgroup_free_data() call will clear the | |
8193 | * QGROUP_RESERVED bit of its io_tree, and free the qgroup | |
8194 | * reserved data space. | |
8195 | * Since the IO will never happen for this page. | |
8b62b72b | 8196 | */ |
3b835840 | 8197 | btrfs_qgroup_free_data(inode, NULL, cur, range_end + 1 - cur); |
131e404a | 8198 | if (!inode_evicting) { |
3b835840 QW |
8199 | clear_extent_bit(tree, cur, range_end, EXTENT_LOCKED | |
8200 | EXTENT_DELALLOC | EXTENT_UPTODATE | | |
bd015294 JB |
8201 | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG | |
8202 | extra_flags, &cached_state); | |
131e404a | 8203 | } |
3b835840 | 8204 | cur = range_end + 1; |
131e404a | 8205 | } |
b9d0b389 | 8206 | /* |
3b835840 | 8207 | * We have iterated through all ordered extents of the page, the page |
f57ad937 QW |
8208 | * should not have Ordered (Private2) anymore, or the above iteration |
8209 | * did something wrong. | |
b9d0b389 | 8210 | */ |
895586eb MWO |
8211 | ASSERT(!folio_test_ordered(folio)); |
8212 | btrfs_page_clear_checked(fs_info, &folio->page, folio_pos(folio), folio_size(folio)); | |
3b835840 | 8213 | if (!inode_evicting) |
f913cff3 | 8214 | __btrfs_release_folio(folio, GFP_NOFS); |
895586eb | 8215 | clear_page_extent_mapped(&folio->page); |
39279cc3 CM |
8216 | } |
8217 | ||
9ebefb18 CM |
8218 | /* |
8219 | * btrfs_page_mkwrite() is not allowed to change the file size as it gets | |
8220 | * called from a page fault handler when a page is first dirtied. Hence we must | |
8221 | * be careful to check for EOF conditions here. We set the page up correctly | |
8222 | * for a written page which means we get ENOSPC checking when writing into | |
8223 | * holes and correct delalloc and unwritten extent mapping on filesystems that | |
8224 | * support these features. | |
8225 | * | |
8226 | * We are not allowed to take the i_mutex here so we have to play games to | |
8227 | * protect against truncate races as the page could now be beyond EOF. Because | |
d1342aad OS |
8228 | * truncate_setsize() writes the inode size before removing pages, once we have |
8229 | * the page lock we can determine safely if the page is beyond EOF. If it is not | |
9ebefb18 CM |
8230 | * beyond EOF, then the page is guaranteed safe against truncation until we |
8231 | * unlock the page. | |
8232 | */ | |
a528a241 | 8233 | vm_fault_t btrfs_page_mkwrite(struct vm_fault *vmf) |
9ebefb18 | 8234 | { |
c2ec175c | 8235 | struct page *page = vmf->page; |
11bac800 | 8236 | struct inode *inode = file_inode(vmf->vma->vm_file); |
0b246afa | 8237 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
e6dcd2dc CM |
8238 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
8239 | struct btrfs_ordered_extent *ordered; | |
2ac55d41 | 8240 | struct extent_state *cached_state = NULL; |
364ecf36 | 8241 | struct extent_changeset *data_reserved = NULL; |
e6dcd2dc | 8242 | unsigned long zero_start; |
9ebefb18 | 8243 | loff_t size; |
a528a241 SJ |
8244 | vm_fault_t ret; |
8245 | int ret2; | |
9998eb70 | 8246 | int reserved = 0; |
d0b7da88 | 8247 | u64 reserved_space; |
a52d9a80 | 8248 | u64 page_start; |
e6dcd2dc | 8249 | u64 page_end; |
d0b7da88 CR |
8250 | u64 end; |
8251 | ||
09cbfeaf | 8252 | reserved_space = PAGE_SIZE; |
9ebefb18 | 8253 | |
b2b5ef5c | 8254 | sb_start_pagefault(inode->i_sb); |
df480633 | 8255 | page_start = page_offset(page); |
09cbfeaf | 8256 | page_end = page_start + PAGE_SIZE - 1; |
d0b7da88 | 8257 | end = page_end; |
df480633 | 8258 | |
d0b7da88 CR |
8259 | /* |
8260 | * Reserving delalloc space after obtaining the page lock can lead to | |
8261 | * deadlock. For example, if a dirty page is locked by this function | |
8262 | * and the call to btrfs_delalloc_reserve_space() ends up triggering | |
f3e90c1c | 8263 | * dirty page write out, then the btrfs_writepages() function could |
d0b7da88 CR |
8264 | * end up waiting indefinitely to get a lock on the page currently |
8265 | * being processed by btrfs_page_mkwrite() function. | |
8266 | */ | |
e5b7231e NB |
8267 | ret2 = btrfs_delalloc_reserve_space(BTRFS_I(inode), &data_reserved, |
8268 | page_start, reserved_space); | |
a528a241 SJ |
8269 | if (!ret2) { |
8270 | ret2 = file_update_time(vmf->vma->vm_file); | |
9998eb70 CM |
8271 | reserved = 1; |
8272 | } | |
a528a241 SJ |
8273 | if (ret2) { |
8274 | ret = vmf_error(ret2); | |
9998eb70 CM |
8275 | if (reserved) |
8276 | goto out; | |
8277 | goto out_noreserve; | |
56a76f82 | 8278 | } |
1832a6d5 | 8279 | |
56a76f82 | 8280 | ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */ |
e6dcd2dc | 8281 | again: |
8318ba79 | 8282 | down_read(&BTRFS_I(inode)->i_mmap_lock); |
9ebefb18 | 8283 | lock_page(page); |
9ebefb18 | 8284 | size = i_size_read(inode); |
a52d9a80 | 8285 | |
9ebefb18 | 8286 | if ((page->mapping != inode->i_mapping) || |
e6dcd2dc | 8287 | (page_start >= size)) { |
9ebefb18 CM |
8288 | /* page got truncated out from underneath us */ |
8289 | goto out_unlock; | |
8290 | } | |
e6dcd2dc CM |
8291 | wait_on_page_writeback(page); |
8292 | ||
570eb97b | 8293 | lock_extent(io_tree, page_start, page_end, &cached_state); |
32443de3 QW |
8294 | ret2 = set_page_extent_mapped(page); |
8295 | if (ret2 < 0) { | |
8296 | ret = vmf_error(ret2); | |
570eb97b | 8297 | unlock_extent(io_tree, page_start, page_end, &cached_state); |
32443de3 QW |
8298 | goto out_unlock; |
8299 | } | |
e6dcd2dc | 8300 | |
eb84ae03 CM |
8301 | /* |
8302 | * we can't set the delalloc bits if there are pending ordered | |
8303 | * extents. Drop our locks and wait for them to finish | |
8304 | */ | |
a776c6fa NB |
8305 | ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), page_start, |
8306 | PAGE_SIZE); | |
e6dcd2dc | 8307 | if (ordered) { |
570eb97b | 8308 | unlock_extent(io_tree, page_start, page_end, &cached_state); |
e6dcd2dc | 8309 | unlock_page(page); |
8318ba79 | 8310 | up_read(&BTRFS_I(inode)->i_mmap_lock); |
36d45567 | 8311 | btrfs_start_ordered_extent(ordered); |
e6dcd2dc CM |
8312 | btrfs_put_ordered_extent(ordered); |
8313 | goto again; | |
8314 | } | |
8315 | ||
09cbfeaf | 8316 | if (page->index == ((size - 1) >> PAGE_SHIFT)) { |
da17066c | 8317 | reserved_space = round_up(size - page_start, |
0b246afa | 8318 | fs_info->sectorsize); |
09cbfeaf | 8319 | if (reserved_space < PAGE_SIZE) { |
d0b7da88 | 8320 | end = page_start + reserved_space - 1; |
86d52921 NB |
8321 | btrfs_delalloc_release_space(BTRFS_I(inode), |
8322 | data_reserved, page_start, | |
8323 | PAGE_SIZE - reserved_space, true); | |
d0b7da88 CR |
8324 | } |
8325 | } | |
8326 | ||
fbf19087 | 8327 | /* |
5416034f LB |
8328 | * page_mkwrite gets called when the page is firstly dirtied after it's |
8329 | * faulted in, but write(2) could also dirty a page and set delalloc | |
8330 | * bits, thus in this case for space account reason, we still need to | |
8331 | * clear any delalloc bits within this page range since we have to | |
8332 | * reserve data&meta space before lock_page() (see above comments). | |
fbf19087 | 8333 | */ |
d0b7da88 | 8334 | clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, end, |
e182163d | 8335 | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING | |
bd015294 | 8336 | EXTENT_DEFRAG, &cached_state); |
fbf19087 | 8337 | |
c2566f22 | 8338 | ret2 = btrfs_set_extent_delalloc(BTRFS_I(inode), page_start, end, 0, |
330a5827 | 8339 | &cached_state); |
a528a241 | 8340 | if (ret2) { |
570eb97b | 8341 | unlock_extent(io_tree, page_start, page_end, &cached_state); |
9ed74f2d JB |
8342 | ret = VM_FAULT_SIGBUS; |
8343 | goto out_unlock; | |
8344 | } | |
9ebefb18 CM |
8345 | |
8346 | /* page is wholly or partially inside EOF */ | |
09cbfeaf | 8347 | if (page_start + PAGE_SIZE > size) |
7073017a | 8348 | zero_start = offset_in_page(size); |
9ebefb18 | 8349 | else |
09cbfeaf | 8350 | zero_start = PAGE_SIZE; |
9ebefb18 | 8351 | |
21a8935e | 8352 | if (zero_start != PAGE_SIZE) |
d048b9c2 | 8353 | memzero_page(page, zero_start, PAGE_SIZE - zero_start); |
21a8935e | 8354 | |
e4f94347 | 8355 | btrfs_page_clear_checked(fs_info, page, page_start, PAGE_SIZE); |
2d8ec40e QW |
8356 | btrfs_page_set_dirty(fs_info, page, page_start, end + 1 - page_start); |
8357 | btrfs_page_set_uptodate(fs_info, page, page_start, end + 1 - page_start); | |
5a3f23d5 | 8358 | |
bc0939fc | 8359 | btrfs_set_inode_last_sub_trans(BTRFS_I(inode)); |
257c62e1 | 8360 | |
570eb97b | 8361 | unlock_extent(io_tree, page_start, page_end, &cached_state); |
8318ba79 | 8362 | up_read(&BTRFS_I(inode)->i_mmap_lock); |
9ebefb18 | 8363 | |
76de60ed YY |
8364 | btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE); |
8365 | sb_end_pagefault(inode->i_sb); | |
8366 | extent_changeset_free(data_reserved); | |
8367 | return VM_FAULT_LOCKED; | |
717beb96 CM |
8368 | |
8369 | out_unlock: | |
9ebefb18 | 8370 | unlock_page(page); |
8318ba79 | 8371 | up_read(&BTRFS_I(inode)->i_mmap_lock); |
1832a6d5 | 8372 | out: |
8702ba93 | 8373 | btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE); |
86d52921 | 8374 | btrfs_delalloc_release_space(BTRFS_I(inode), data_reserved, page_start, |
43b18595 | 8375 | reserved_space, (ret != 0)); |
9998eb70 | 8376 | out_noreserve: |
b2b5ef5c | 8377 | sb_end_pagefault(inode->i_sb); |
364ecf36 | 8378 | extent_changeset_free(data_reserved); |
9ebefb18 CM |
8379 | return ret; |
8380 | } | |
8381 | ||
d9dcae67 | 8382 | static int btrfs_truncate(struct btrfs_inode *inode, bool skip_writeback) |
39279cc3 | 8383 | { |
d9ac19c3 | 8384 | struct btrfs_truncate_control control = { |
d9dcae67 DS |
8385 | .inode = inode, |
8386 | .ino = btrfs_ino(inode), | |
d9ac19c3 | 8387 | .min_type = BTRFS_EXTENT_DATA_KEY, |
655807b8 | 8388 | .clear_extent_range = true, |
d9ac19c3 | 8389 | }; |
d9dcae67 DS |
8390 | struct btrfs_root *root = inode->root; |
8391 | struct btrfs_fs_info *fs_info = root->fs_info; | |
fcb80c2a | 8392 | struct btrfs_block_rsv *rsv; |
ad7e1a74 | 8393 | int ret; |
39279cc3 | 8394 | struct btrfs_trans_handle *trans; |
0b246afa | 8395 | u64 mask = fs_info->sectorsize - 1; |
6822b3f7 | 8396 | const u64 min_size = btrfs_calc_metadata_size(fs_info, 1); |
39279cc3 | 8397 | |
213e8c55 | 8398 | if (!skip_writeback) { |
d9dcae67 DS |
8399 | ret = btrfs_wait_ordered_range(&inode->vfs_inode, |
8400 | inode->vfs_inode.i_size & (~mask), | |
213e8c55 FM |
8401 | (u64)-1); |
8402 | if (ret) | |
8403 | return ret; | |
8404 | } | |
39279cc3 | 8405 | |
fcb80c2a | 8406 | /* |
f7e9e8fc OS |
8407 | * Yes ladies and gentlemen, this is indeed ugly. We have a couple of |
8408 | * things going on here: | |
fcb80c2a | 8409 | * |
f7e9e8fc | 8410 | * 1) We need to reserve space to update our inode. |
fcb80c2a | 8411 | * |
f7e9e8fc | 8412 | * 2) We need to have something to cache all the space that is going to |
fcb80c2a JB |
8413 | * be free'd up by the truncate operation, but also have some slack |
8414 | * space reserved in case it uses space during the truncate (thank you | |
8415 | * very much snapshotting). | |
8416 | * | |
f7e9e8fc | 8417 | * And we need these to be separate. The fact is we can use a lot of |
fcb80c2a | 8418 | * space doing the truncate, and we have no earthly idea how much space |
01327610 | 8419 | * we will use, so we need the truncate reservation to be separate so it |
f7e9e8fc OS |
8420 | * doesn't end up using space reserved for updating the inode. We also |
8421 | * need to be able to stop the transaction and start a new one, which | |
8422 | * means we need to be able to update the inode several times, and we | |
8423 | * have no idea of knowing how many times that will be, so we can't just | |
8424 | * reserve 1 item for the entirety of the operation, so that has to be | |
8425 | * done separately as well. | |
fcb80c2a JB |
8426 | * |
8427 | * So that leaves us with | |
8428 | * | |
f7e9e8fc | 8429 | * 1) rsv - for the truncate reservation, which we will steal from the |
fcb80c2a | 8430 | * transaction reservation. |
f7e9e8fc | 8431 | * 2) fs_info->trans_block_rsv - this will have 1 items worth left for |
fcb80c2a JB |
8432 | * updating the inode. |
8433 | */ | |
2ff7e61e | 8434 | rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP); |
fcb80c2a JB |
8435 | if (!rsv) |
8436 | return -ENOMEM; | |
4a338542 | 8437 | rsv->size = min_size; |
710d5921 | 8438 | rsv->failfast = true; |
f0cd846e | 8439 | |
907cbceb | 8440 | /* |
07127184 | 8441 | * 1 for the truncate slack space |
907cbceb JB |
8442 | * 1 for updating the inode. |
8443 | */ | |
f3fe820c | 8444 | trans = btrfs_start_transaction(root, 2); |
fcb80c2a | 8445 | if (IS_ERR(trans)) { |
ad7e1a74 | 8446 | ret = PTR_ERR(trans); |
fcb80c2a JB |
8447 | goto out; |
8448 | } | |
f0cd846e | 8449 | |
907cbceb | 8450 | /* Migrate the slack space for the truncate to our reserve */ |
0b246afa | 8451 | ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv, rsv, |
3a584174 | 8452 | min_size, false); |
6822b3f7 FM |
8453 | /* |
8454 | * We have reserved 2 metadata units when we started the transaction and | |
8455 | * min_size matches 1 unit, so this should never fail, but if it does, | |
8456 | * it's not critical we just fail truncation. | |
8457 | */ | |
8458 | if (WARN_ON(ret)) { | |
8459 | btrfs_end_transaction(trans); | |
8460 | goto out; | |
8461 | } | |
f0cd846e | 8462 | |
ca7e70f5 | 8463 | trans->block_rsv = rsv; |
907cbceb | 8464 | |
8082510e | 8465 | while (1) { |
9a4a1429 | 8466 | struct extent_state *cached_state = NULL; |
d9dcae67 | 8467 | const u64 new_size = inode->vfs_inode.i_size; |
9a4a1429 JB |
8468 | const u64 lock_start = ALIGN_DOWN(new_size, fs_info->sectorsize); |
8469 | ||
d9ac19c3 | 8470 | control.new_size = new_size; |
d9dcae67 | 8471 | lock_extent(&inode->io_tree, lock_start, (u64)-1, &cached_state); |
9a4a1429 JB |
8472 | /* |
8473 | * We want to drop from the next block forward in case this new | |
8474 | * size is not block aligned since we will be keeping the last | |
8475 | * block of the extent just the way it is. | |
8476 | */ | |
d9dcae67 | 8477 | btrfs_drop_extent_map_range(inode, |
4c0c8cfc FM |
8478 | ALIGN(new_size, fs_info->sectorsize), |
8479 | (u64)-1, false); | |
9a4a1429 | 8480 | |
71d18b53 | 8481 | ret = btrfs_truncate_inode_items(trans, root, &control); |
c2ddb612 | 8482 | |
d9dcae67 DS |
8483 | inode_sub_bytes(&inode->vfs_inode, control.sub_bytes); |
8484 | btrfs_inode_safe_disk_i_size_write(inode, control.last_size); | |
c2ddb612 | 8485 | |
d9dcae67 | 8486 | unlock_extent(&inode->io_tree, lock_start, (u64)-1, &cached_state); |
9a4a1429 | 8487 | |
ddfae63c | 8488 | trans->block_rsv = &fs_info->trans_block_rsv; |
ad7e1a74 | 8489 | if (ret != -ENOSPC && ret != -EAGAIN) |
8082510e | 8490 | break; |
39279cc3 | 8491 | |
d9dcae67 | 8492 | ret = btrfs_update_inode(trans, root, inode); |
ad7e1a74 | 8493 | if (ret) |
3893e33b | 8494 | break; |
ca7e70f5 | 8495 | |
3a45bb20 | 8496 | btrfs_end_transaction(trans); |
2ff7e61e | 8497 | btrfs_btree_balance_dirty(fs_info); |
ca7e70f5 JB |
8498 | |
8499 | trans = btrfs_start_transaction(root, 2); | |
8500 | if (IS_ERR(trans)) { | |
ad7e1a74 | 8501 | ret = PTR_ERR(trans); |
ca7e70f5 JB |
8502 | trans = NULL; |
8503 | break; | |
8504 | } | |
8505 | ||
63f018be | 8506 | btrfs_block_rsv_release(fs_info, rsv, -1, NULL); |
0b246afa | 8507 | ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv, |
3a584174 | 8508 | rsv, min_size, false); |
6822b3f7 FM |
8509 | /* |
8510 | * We have reserved 2 metadata units when we started the | |
8511 | * transaction and min_size matches 1 unit, so this should never | |
8512 | * fail, but if it does, it's not critical we just fail truncation. | |
8513 | */ | |
8514 | if (WARN_ON(ret)) | |
8515 | break; | |
8516 | ||
ca7e70f5 | 8517 | trans->block_rsv = rsv; |
8082510e YZ |
8518 | } |
8519 | ||
ddfae63c JB |
8520 | /* |
8521 | * We can't call btrfs_truncate_block inside a trans handle as we could | |
54f03ab1 JB |
8522 | * deadlock with freeze, if we got BTRFS_NEED_TRUNCATE_BLOCK then we |
8523 | * know we've truncated everything except the last little bit, and can | |
8524 | * do btrfs_truncate_block and then update the disk_i_size. | |
ddfae63c | 8525 | */ |
54f03ab1 | 8526 | if (ret == BTRFS_NEED_TRUNCATE_BLOCK) { |
ddfae63c JB |
8527 | btrfs_end_transaction(trans); |
8528 | btrfs_btree_balance_dirty(fs_info); | |
8529 | ||
d9dcae67 | 8530 | ret = btrfs_truncate_block(inode, inode->vfs_inode.i_size, 0, 0); |
ddfae63c JB |
8531 | if (ret) |
8532 | goto out; | |
8533 | trans = btrfs_start_transaction(root, 1); | |
8534 | if (IS_ERR(trans)) { | |
8535 | ret = PTR_ERR(trans); | |
8536 | goto out; | |
8537 | } | |
d9dcae67 | 8538 | btrfs_inode_safe_disk_i_size_write(inode, 0); |
ddfae63c JB |
8539 | } |
8540 | ||
917c16b2 | 8541 | if (trans) { |
ad7e1a74 OS |
8542 | int ret2; |
8543 | ||
0b246afa | 8544 | trans->block_rsv = &fs_info->trans_block_rsv; |
d9dcae67 | 8545 | ret2 = btrfs_update_inode(trans, root, inode); |
ad7e1a74 OS |
8546 | if (ret2 && !ret) |
8547 | ret = ret2; | |
7b128766 | 8548 | |
ad7e1a74 OS |
8549 | ret2 = btrfs_end_transaction(trans); |
8550 | if (ret2 && !ret) | |
8551 | ret = ret2; | |
2ff7e61e | 8552 | btrfs_btree_balance_dirty(fs_info); |
917c16b2 | 8553 | } |
fcb80c2a | 8554 | out: |
2ff7e61e | 8555 | btrfs_free_block_rsv(fs_info, rsv); |
0d7d3165 FM |
8556 | /* |
8557 | * So if we truncate and then write and fsync we normally would just | |
8558 | * write the extents that changed, which is a problem if we need to | |
8559 | * first truncate that entire inode. So set this flag so we write out | |
8560 | * all of the extents in the inode to the sync log so we're completely | |
8561 | * safe. | |
8562 | * | |
8563 | * If no extents were dropped or trimmed we don't need to force the next | |
8564 | * fsync to truncate all the inode's items from the log and re-log them | |
8565 | * all. This means the truncate operation did not change the file size, | |
8566 | * or changed it to a smaller size but there was only an implicit hole | |
8567 | * between the old i_size and the new i_size, and there were no prealloc | |
8568 | * extents beyond i_size to drop. | |
8569 | */ | |
d9ac19c3 | 8570 | if (control.extents_found > 0) |
d9dcae67 | 8571 | btrfs_set_inode_full_sync(inode); |
fcb80c2a | 8572 | |
ad7e1a74 | 8573 | return ret; |
39279cc3 CM |
8574 | } |
8575 | ||
f2d40141 | 8576 | struct inode *btrfs_new_subvol_inode(struct mnt_idmap *idmap, |
a1fd0c35 OS |
8577 | struct inode *dir) |
8578 | { | |
8579 | struct inode *inode; | |
8580 | ||
8581 | inode = new_inode(dir->i_sb); | |
8582 | if (inode) { | |
8583 | /* | |
8584 | * Subvolumes don't inherit the sgid bit or the parent's gid if | |
8585 | * the parent's sgid bit is set. This is probably a bug. | |
8586 | */ | |
f2d40141 | 8587 | inode_init_owner(idmap, inode, NULL, |
a1fd0c35 OS |
8588 | S_IFDIR | (~current_umask() & S_IRWXUGO)); |
8589 | inode->i_op = &btrfs_dir_inode_operations; | |
8590 | inode->i_fop = &btrfs_dir_file_operations; | |
8591 | } | |
8592 | return inode; | |
8593 | } | |
8594 | ||
39279cc3 CM |
8595 | struct inode *btrfs_alloc_inode(struct super_block *sb) |
8596 | { | |
69fe2d75 | 8597 | struct btrfs_fs_info *fs_info = btrfs_sb(sb); |
39279cc3 | 8598 | struct btrfs_inode *ei; |
2ead6ae7 | 8599 | struct inode *inode; |
39279cc3 | 8600 | |
fd60b288 | 8601 | ei = alloc_inode_sb(sb, btrfs_inode_cachep, GFP_KERNEL); |
39279cc3 CM |
8602 | if (!ei) |
8603 | return NULL; | |
2ead6ae7 YZ |
8604 | |
8605 | ei->root = NULL; | |
2ead6ae7 | 8606 | ei->generation = 0; |
15ee9bc7 | 8607 | ei->last_trans = 0; |
257c62e1 | 8608 | ei->last_sub_trans = 0; |
e02119d5 | 8609 | ei->logged_trans = 0; |
2ead6ae7 | 8610 | ei->delalloc_bytes = 0; |
a7e3b975 | 8611 | ei->new_delalloc_bytes = 0; |
47059d93 | 8612 | ei->defrag_bytes = 0; |
2ead6ae7 YZ |
8613 | ei->disk_i_size = 0; |
8614 | ei->flags = 0; | |
77eea05e | 8615 | ei->ro_flags = 0; |
7709cde3 | 8616 | ei->csum_bytes = 0; |
2ead6ae7 | 8617 | ei->index_cnt = (u64)-1; |
67de1176 | 8618 | ei->dir_index = 0; |
2ead6ae7 | 8619 | ei->last_unlink_trans = 0; |
3ebac17c | 8620 | ei->last_reflink_trans = 0; |
46d8bc34 | 8621 | ei->last_log_commit = 0; |
2ead6ae7 | 8622 | |
9e0baf60 JB |
8623 | spin_lock_init(&ei->lock); |
8624 | ei->outstanding_extents = 0; | |
69fe2d75 JB |
8625 | if (sb->s_magic != BTRFS_TEST_MAGIC) |
8626 | btrfs_init_metadata_block_rsv(fs_info, &ei->block_rsv, | |
8627 | BTRFS_BLOCK_RSV_DELALLOC); | |
72ac3c0d | 8628 | ei->runtime_flags = 0; |
b52aa8c9 | 8629 | ei->prop_compress = BTRFS_COMPRESS_NONE; |
eec63c65 | 8630 | ei->defrag_compress = BTRFS_COMPRESS_NONE; |
2ead6ae7 | 8631 | |
16cdcec7 MX |
8632 | ei->delayed_node = NULL; |
8633 | ||
9cc97d64 | 8634 | ei->i_otime.tv_sec = 0; |
8635 | ei->i_otime.tv_nsec = 0; | |
8636 | ||
2ead6ae7 | 8637 | inode = &ei->vfs_inode; |
a8067e02 | 8638 | extent_map_tree_init(&ei->extent_tree); |
35da5a7e | 8639 | extent_io_tree_init(fs_info, &ei->io_tree, IO_TREE_INODE_IO); |
0988fc7b | 8640 | ei->io_tree.inode = ei; |
41a2ee75 | 8641 | extent_io_tree_init(fs_info, &ei->file_extent_tree, |
35da5a7e | 8642 | IO_TREE_INODE_FILE_EXTENT); |
2ead6ae7 | 8643 | mutex_init(&ei->log_mutex); |
e6dcd2dc | 8644 | btrfs_ordered_inode_tree_init(&ei->ordered_tree); |
2ead6ae7 | 8645 | INIT_LIST_HEAD(&ei->delalloc_inodes); |
8089fe62 | 8646 | INIT_LIST_HEAD(&ei->delayed_iput); |
2ead6ae7 | 8647 | RB_CLEAR_NODE(&ei->rb_node); |
8318ba79 | 8648 | init_rwsem(&ei->i_mmap_lock); |
2ead6ae7 YZ |
8649 | |
8650 | return inode; | |
39279cc3 CM |
8651 | } |
8652 | ||
aaedb55b JB |
8653 | #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS |
8654 | void btrfs_test_destroy_inode(struct inode *inode) | |
8655 | { | |
4c0c8cfc | 8656 | btrfs_drop_extent_map_range(BTRFS_I(inode), 0, (u64)-1, false); |
aaedb55b JB |
8657 | kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode)); |
8658 | } | |
8659 | #endif | |
8660 | ||
26602cab | 8661 | void btrfs_free_inode(struct inode *inode) |
fa0d7e3d | 8662 | { |
fa0d7e3d NP |
8663 | kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode)); |
8664 | } | |
8665 | ||
633cc816 | 8666 | void btrfs_destroy_inode(struct inode *vfs_inode) |
39279cc3 | 8667 | { |
e6dcd2dc | 8668 | struct btrfs_ordered_extent *ordered; |
633cc816 NB |
8669 | struct btrfs_inode *inode = BTRFS_I(vfs_inode); |
8670 | struct btrfs_root *root = inode->root; | |
5f4403e1 | 8671 | bool freespace_inode; |
5a3f23d5 | 8672 | |
633cc816 NB |
8673 | WARN_ON(!hlist_empty(&vfs_inode->i_dentry)); |
8674 | WARN_ON(vfs_inode->i_data.nrpages); | |
8675 | WARN_ON(inode->block_rsv.reserved); | |
8676 | WARN_ON(inode->block_rsv.size); | |
8677 | WARN_ON(inode->outstanding_extents); | |
dc287224 FM |
8678 | if (!S_ISDIR(vfs_inode->i_mode)) { |
8679 | WARN_ON(inode->delalloc_bytes); | |
8680 | WARN_ON(inode->new_delalloc_bytes); | |
8681 | } | |
633cc816 NB |
8682 | WARN_ON(inode->csum_bytes); |
8683 | WARN_ON(inode->defrag_bytes); | |
39279cc3 | 8684 | |
a6dbd429 JB |
8685 | /* |
8686 | * This can happen where we create an inode, but somebody else also | |
8687 | * created the same inode and we need to destroy the one we already | |
8688 | * created. | |
8689 | */ | |
8690 | if (!root) | |
26602cab | 8691 | return; |
a6dbd429 | 8692 | |
5f4403e1 IA |
8693 | /* |
8694 | * If this is a free space inode do not take the ordered extents lockdep | |
8695 | * map. | |
8696 | */ | |
8697 | freespace_inode = btrfs_is_free_space_inode(inode); | |
8698 | ||
d397712b | 8699 | while (1) { |
633cc816 | 8700 | ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1); |
e6dcd2dc CM |
8701 | if (!ordered) |
8702 | break; | |
8703 | else { | |
633cc816 | 8704 | btrfs_err(root->fs_info, |
5d163e0e | 8705 | "found ordered extent %llu %llu on inode cleanup", |
bffe633e | 8706 | ordered->file_offset, ordered->num_bytes); |
5f4403e1 IA |
8707 | |
8708 | if (!freespace_inode) | |
8709 | btrfs_lockdep_acquire(root->fs_info, btrfs_ordered_extent); | |
8710 | ||
71fe0a55 | 8711 | btrfs_remove_ordered_extent(inode, ordered); |
e6dcd2dc CM |
8712 | btrfs_put_ordered_extent(ordered); |
8713 | btrfs_put_ordered_extent(ordered); | |
8714 | } | |
8715 | } | |
633cc816 NB |
8716 | btrfs_qgroup_check_reserved_leak(inode); |
8717 | inode_tree_del(inode); | |
4c0c8cfc | 8718 | btrfs_drop_extent_map_range(inode, 0, (u64)-1, false); |
633cc816 NB |
8719 | btrfs_inode_clear_file_extent_range(inode, 0, (u64)-1); |
8720 | btrfs_put_root(inode->root); | |
39279cc3 CM |
8721 | } |
8722 | ||
45321ac5 | 8723 | int btrfs_drop_inode(struct inode *inode) |
76dda93c YZ |
8724 | { |
8725 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
45321ac5 | 8726 | |
6379ef9f NA |
8727 | if (root == NULL) |
8728 | return 1; | |
8729 | ||
fa6ac876 | 8730 | /* the snap/subvol tree is on deleting */ |
69e9c6c6 | 8731 | if (btrfs_root_refs(&root->root_item) == 0) |
45321ac5 | 8732 | return 1; |
76dda93c | 8733 | else |
45321ac5 | 8734 | return generic_drop_inode(inode); |
76dda93c YZ |
8735 | } |
8736 | ||
0ee0fda0 | 8737 | static void init_once(void *foo) |
39279cc3 | 8738 | { |
0d031dc4 | 8739 | struct btrfs_inode *ei = foo; |
39279cc3 CM |
8740 | |
8741 | inode_init_once(&ei->vfs_inode); | |
8742 | } | |
8743 | ||
e67c718b | 8744 | void __cold btrfs_destroy_cachep(void) |
39279cc3 | 8745 | { |
8c0a8537 KS |
8746 | /* |
8747 | * Make sure all delayed rcu free inodes are flushed before we | |
8748 | * destroy cache. | |
8749 | */ | |
8750 | rcu_barrier(); | |
642c5d34 | 8751 | bioset_exit(&btrfs_dio_bioset); |
5598e900 | 8752 | kmem_cache_destroy(btrfs_inode_cachep); |
39279cc3 CM |
8753 | } |
8754 | ||
f5c29bd9 | 8755 | int __init btrfs_init_cachep(void) |
39279cc3 | 8756 | { |
837e1972 | 8757 | btrfs_inode_cachep = kmem_cache_create("btrfs_inode", |
9601e3f6 | 8758 | sizeof(struct btrfs_inode), 0, |
5d097056 VD |
8759 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD | SLAB_ACCOUNT, |
8760 | init_once); | |
39279cc3 CM |
8761 | if (!btrfs_inode_cachep) |
8762 | goto fail; | |
9601e3f6 | 8763 | |
642c5d34 | 8764 | if (bioset_init(&btrfs_dio_bioset, BIO_POOL_SIZE, |
67d66982 | 8765 | offsetof(struct btrfs_dio_private, bbio.bio), |
642c5d34 CH |
8766 | BIOSET_NEED_BVECS)) |
8767 | goto fail; | |
8768 | ||
39279cc3 CM |
8769 | return 0; |
8770 | fail: | |
8771 | btrfs_destroy_cachep(); | |
8772 | return -ENOMEM; | |
8773 | } | |
8774 | ||
b74d24f7 | 8775 | static int btrfs_getattr(struct mnt_idmap *idmap, |
549c7297 | 8776 | const struct path *path, struct kstat *stat, |
a528d35e | 8777 | u32 request_mask, unsigned int flags) |
39279cc3 | 8778 | { |
df0af1a5 | 8779 | u64 delalloc_bytes; |
2766ff61 | 8780 | u64 inode_bytes; |
a528d35e | 8781 | struct inode *inode = d_inode(path->dentry); |
fadc0d8b | 8782 | u32 blocksize = inode->i_sb->s_blocksize; |
04a87e34 | 8783 | u32 bi_flags = BTRFS_I(inode)->flags; |
14605409 | 8784 | u32 bi_ro_flags = BTRFS_I(inode)->ro_flags; |
04a87e34 YS |
8785 | |
8786 | stat->result_mask |= STATX_BTIME; | |
8787 | stat->btime.tv_sec = BTRFS_I(inode)->i_otime.tv_sec; | |
8788 | stat->btime.tv_nsec = BTRFS_I(inode)->i_otime.tv_nsec; | |
8789 | if (bi_flags & BTRFS_INODE_APPEND) | |
8790 | stat->attributes |= STATX_ATTR_APPEND; | |
8791 | if (bi_flags & BTRFS_INODE_COMPRESS) | |
8792 | stat->attributes |= STATX_ATTR_COMPRESSED; | |
8793 | if (bi_flags & BTRFS_INODE_IMMUTABLE) | |
8794 | stat->attributes |= STATX_ATTR_IMMUTABLE; | |
8795 | if (bi_flags & BTRFS_INODE_NODUMP) | |
8796 | stat->attributes |= STATX_ATTR_NODUMP; | |
14605409 BB |
8797 | if (bi_ro_flags & BTRFS_INODE_RO_VERITY) |
8798 | stat->attributes |= STATX_ATTR_VERITY; | |
04a87e34 YS |
8799 | |
8800 | stat->attributes_mask |= (STATX_ATTR_APPEND | | |
8801 | STATX_ATTR_COMPRESSED | | |
8802 | STATX_ATTR_IMMUTABLE | | |
8803 | STATX_ATTR_NODUMP); | |
fadc0d8b | 8804 | |
b74d24f7 | 8805 | generic_fillattr(idmap, inode, stat); |
0ee5dc67 | 8806 | stat->dev = BTRFS_I(inode)->root->anon_dev; |
df0af1a5 MX |
8807 | |
8808 | spin_lock(&BTRFS_I(inode)->lock); | |
a7e3b975 | 8809 | delalloc_bytes = BTRFS_I(inode)->new_delalloc_bytes; |
2766ff61 | 8810 | inode_bytes = inode_get_bytes(inode); |
df0af1a5 | 8811 | spin_unlock(&BTRFS_I(inode)->lock); |
2766ff61 | 8812 | stat->blocks = (ALIGN(inode_bytes, blocksize) + |
29e70be2 | 8813 | ALIGN(delalloc_bytes, blocksize)) >> SECTOR_SHIFT; |
39279cc3 CM |
8814 | return 0; |
8815 | } | |
8816 | ||
cdd1fedf DF |
8817 | static int btrfs_rename_exchange(struct inode *old_dir, |
8818 | struct dentry *old_dentry, | |
8819 | struct inode *new_dir, | |
8820 | struct dentry *new_dentry) | |
8821 | { | |
0b246afa | 8822 | struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb); |
cdd1fedf | 8823 | struct btrfs_trans_handle *trans; |
c1621871 | 8824 | unsigned int trans_num_items; |
cdd1fedf DF |
8825 | struct btrfs_root *root = BTRFS_I(old_dir)->root; |
8826 | struct btrfs_root *dest = BTRFS_I(new_dir)->root; | |
8827 | struct inode *new_inode = new_dentry->d_inode; | |
8828 | struct inode *old_inode = old_dentry->d_inode; | |
95582b00 | 8829 | struct timespec64 ctime = current_time(old_inode); |
88d2beec FM |
8830 | struct btrfs_rename_ctx old_rename_ctx; |
8831 | struct btrfs_rename_ctx new_rename_ctx; | |
4a0cc7ca NB |
8832 | u64 old_ino = btrfs_ino(BTRFS_I(old_inode)); |
8833 | u64 new_ino = btrfs_ino(BTRFS_I(new_inode)); | |
cdd1fedf DF |
8834 | u64 old_idx = 0; |
8835 | u64 new_idx = 0; | |
cdd1fedf | 8836 | int ret; |
75b463d2 | 8837 | int ret2; |
dc09ef35 | 8838 | bool need_abort = false; |
ab3c5c18 | 8839 | struct fscrypt_name old_fname, new_fname; |
6db75318 | 8840 | struct fscrypt_str *old_name, *new_name; |
cdd1fedf | 8841 | |
3f79f6f6 N |
8842 | /* |
8843 | * For non-subvolumes allow exchange only within one subvolume, in the | |
8844 | * same inode namespace. Two subvolumes (represented as directory) can | |
8845 | * be exchanged as they're a logical link and have a fixed inode number. | |
8846 | */ | |
8847 | if (root != dest && | |
8848 | (old_ino != BTRFS_FIRST_FREE_OBJECTID || | |
8849 | new_ino != BTRFS_FIRST_FREE_OBJECTID)) | |
cdd1fedf DF |
8850 | return -EXDEV; |
8851 | ||
ab3c5c18 STD |
8852 | ret = fscrypt_setup_filename(old_dir, &old_dentry->d_name, 0, &old_fname); |
8853 | if (ret) | |
8854 | return ret; | |
8855 | ||
8856 | ret = fscrypt_setup_filename(new_dir, &new_dentry->d_name, 0, &new_fname); | |
8857 | if (ret) { | |
8858 | fscrypt_free_filename(&old_fname); | |
8859 | return ret; | |
8860 | } | |
8861 | ||
6db75318 STD |
8862 | old_name = &old_fname.disk_name; |
8863 | new_name = &new_fname.disk_name; | |
ab3c5c18 | 8864 | |
cdd1fedf | 8865 | /* close the race window with snapshot create/destroy ioctl */ |
943eb3bf JB |
8866 | if (old_ino == BTRFS_FIRST_FREE_OBJECTID || |
8867 | new_ino == BTRFS_FIRST_FREE_OBJECTID) | |
0b246afa | 8868 | down_read(&fs_info->subvol_sem); |
cdd1fedf DF |
8869 | |
8870 | /* | |
c1621871 OS |
8871 | * For each inode: |
8872 | * 1 to remove old dir item | |
8873 | * 1 to remove old dir index | |
8874 | * 1 to add new dir item | |
8875 | * 1 to add new dir index | |
8876 | * 1 to update parent inode | |
8877 | * | |
8878 | * If the parents are the same, we only need to account for one | |
cdd1fedf | 8879 | */ |
c1621871 OS |
8880 | trans_num_items = (old_dir == new_dir ? 9 : 10); |
8881 | if (old_ino == BTRFS_FIRST_FREE_OBJECTID) { | |
8882 | /* | |
8883 | * 1 to remove old root ref | |
8884 | * 1 to remove old root backref | |
8885 | * 1 to add new root ref | |
8886 | * 1 to add new root backref | |
8887 | */ | |
8888 | trans_num_items += 4; | |
8889 | } else { | |
8890 | /* | |
8891 | * 1 to update inode item | |
8892 | * 1 to remove old inode ref | |
8893 | * 1 to add new inode ref | |
8894 | */ | |
8895 | trans_num_items += 3; | |
8896 | } | |
8897 | if (new_ino == BTRFS_FIRST_FREE_OBJECTID) | |
8898 | trans_num_items += 4; | |
8899 | else | |
8900 | trans_num_items += 3; | |
8901 | trans = btrfs_start_transaction(root, trans_num_items); | |
cdd1fedf DF |
8902 | if (IS_ERR(trans)) { |
8903 | ret = PTR_ERR(trans); | |
8904 | goto out_notrans; | |
8905 | } | |
8906 | ||
00aa8e87 JB |
8907 | if (dest != root) { |
8908 | ret = btrfs_record_root_in_trans(trans, dest); | |
8909 | if (ret) | |
8910 | goto out_fail; | |
8911 | } | |
3e174099 | 8912 | |
cdd1fedf DF |
8913 | /* |
8914 | * We need to find a free sequence number both in the source and | |
8915 | * in the destination directory for the exchange. | |
8916 | */ | |
877574e2 | 8917 | ret = btrfs_set_inode_index(BTRFS_I(new_dir), &old_idx); |
cdd1fedf DF |
8918 | if (ret) |
8919 | goto out_fail; | |
877574e2 | 8920 | ret = btrfs_set_inode_index(BTRFS_I(old_dir), &new_idx); |
cdd1fedf DF |
8921 | if (ret) |
8922 | goto out_fail; | |
8923 | ||
8924 | BTRFS_I(old_inode)->dir_index = 0ULL; | |
8925 | BTRFS_I(new_inode)->dir_index = 0ULL; | |
8926 | ||
8927 | /* Reference for the source. */ | |
8928 | if (old_ino == BTRFS_FIRST_FREE_OBJECTID) { | |
8929 | /* force full log commit if subvolume involved. */ | |
90787766 | 8930 | btrfs_set_log_full_commit(trans); |
cdd1fedf | 8931 | } else { |
6db75318 | 8932 | ret = btrfs_insert_inode_ref(trans, dest, new_name, old_ino, |
f85b7379 DS |
8933 | btrfs_ino(BTRFS_I(new_dir)), |
8934 | old_idx); | |
cdd1fedf DF |
8935 | if (ret) |
8936 | goto out_fail; | |
dc09ef35 | 8937 | need_abort = true; |
cdd1fedf DF |
8938 | } |
8939 | ||
8940 | /* And now for the dest. */ | |
8941 | if (new_ino == BTRFS_FIRST_FREE_OBJECTID) { | |
8942 | /* force full log commit if subvolume involved. */ | |
90787766 | 8943 | btrfs_set_log_full_commit(trans); |
cdd1fedf | 8944 | } else { |
6db75318 | 8945 | ret = btrfs_insert_inode_ref(trans, root, old_name, new_ino, |
f85b7379 DS |
8946 | btrfs_ino(BTRFS_I(old_dir)), |
8947 | new_idx); | |
dc09ef35 JB |
8948 | if (ret) { |
8949 | if (need_abort) | |
8950 | btrfs_abort_transaction(trans, ret); | |
cdd1fedf | 8951 | goto out_fail; |
dc09ef35 | 8952 | } |
cdd1fedf DF |
8953 | } |
8954 | ||
8955 | /* Update inode version and ctime/mtime. */ | |
8956 | inode_inc_iversion(old_dir); | |
8957 | inode_inc_iversion(new_dir); | |
8958 | inode_inc_iversion(old_inode); | |
8959 | inode_inc_iversion(new_inode); | |
c1867eb3 DS |
8960 | old_dir->i_mtime = ctime; |
8961 | old_dir->i_ctime = ctime; | |
8962 | new_dir->i_mtime = ctime; | |
8963 | new_dir->i_ctime = ctime; | |
cdd1fedf DF |
8964 | old_inode->i_ctime = ctime; |
8965 | new_inode->i_ctime = ctime; | |
8966 | ||
8967 | if (old_dentry->d_parent != new_dentry->d_parent) { | |
f85b7379 | 8968 | btrfs_record_unlink_dir(trans, BTRFS_I(old_dir), |
59fcf388 | 8969 | BTRFS_I(old_inode), true); |
f85b7379 | 8970 | btrfs_record_unlink_dir(trans, BTRFS_I(new_dir), |
59fcf388 | 8971 | BTRFS_I(new_inode), true); |
cdd1fedf DF |
8972 | } |
8973 | ||
8974 | /* src is a subvolume */ | |
8975 | if (old_ino == BTRFS_FIRST_FREE_OBJECTID) { | |
5b7544cb | 8976 | ret = btrfs_unlink_subvol(trans, BTRFS_I(old_dir), old_dentry); |
cdd1fedf | 8977 | } else { /* src is an inode */ |
4467af88 | 8978 | ret = __btrfs_unlink_inode(trans, BTRFS_I(old_dir), |
4ec5934e | 8979 | BTRFS_I(old_dentry->d_inode), |
6db75318 | 8980 | old_name, &old_rename_ctx); |
cdd1fedf | 8981 | if (!ret) |
9a56fcd1 | 8982 | ret = btrfs_update_inode(trans, root, BTRFS_I(old_inode)); |
cdd1fedf DF |
8983 | } |
8984 | if (ret) { | |
66642832 | 8985 | btrfs_abort_transaction(trans, ret); |
cdd1fedf DF |
8986 | goto out_fail; |
8987 | } | |
8988 | ||
8989 | /* dest is a subvolume */ | |
8990 | if (new_ino == BTRFS_FIRST_FREE_OBJECTID) { | |
5b7544cb | 8991 | ret = btrfs_unlink_subvol(trans, BTRFS_I(new_dir), new_dentry); |
cdd1fedf | 8992 | } else { /* dest is an inode */ |
4467af88 | 8993 | ret = __btrfs_unlink_inode(trans, BTRFS_I(new_dir), |
4ec5934e | 8994 | BTRFS_I(new_dentry->d_inode), |
6db75318 | 8995 | new_name, &new_rename_ctx); |
cdd1fedf | 8996 | if (!ret) |
9a56fcd1 | 8997 | ret = btrfs_update_inode(trans, dest, BTRFS_I(new_inode)); |
cdd1fedf DF |
8998 | } |
8999 | if (ret) { | |
66642832 | 9000 | btrfs_abort_transaction(trans, ret); |
cdd1fedf DF |
9001 | goto out_fail; |
9002 | } | |
9003 | ||
db0a669f | 9004 | ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode), |
6db75318 | 9005 | new_name, 0, old_idx); |
cdd1fedf | 9006 | if (ret) { |
66642832 | 9007 | btrfs_abort_transaction(trans, ret); |
cdd1fedf DF |
9008 | goto out_fail; |
9009 | } | |
9010 | ||
db0a669f | 9011 | ret = btrfs_add_link(trans, BTRFS_I(old_dir), BTRFS_I(new_inode), |
6db75318 | 9012 | old_name, 0, new_idx); |
cdd1fedf | 9013 | if (ret) { |
66642832 | 9014 | btrfs_abort_transaction(trans, ret); |
cdd1fedf DF |
9015 | goto out_fail; |
9016 | } | |
9017 | ||
9018 | if (old_inode->i_nlink == 1) | |
9019 | BTRFS_I(old_inode)->dir_index = old_idx; | |
9020 | if (new_inode->i_nlink == 1) | |
9021 | BTRFS_I(new_inode)->dir_index = new_idx; | |
9022 | ||
259c4b96 FM |
9023 | /* |
9024 | * Now pin the logs of the roots. We do it to ensure that no other task | |
9025 | * can sync the logs while we are in progress with the rename, because | |
9026 | * that could result in an inconsistency in case any of the inodes that | |
9027 | * are part of this rename operation were logged before. | |
9028 | */ | |
9029 | if (old_ino != BTRFS_FIRST_FREE_OBJECTID) | |
9030 | btrfs_pin_log_trans(root); | |
9031 | if (new_ino != BTRFS_FIRST_FREE_OBJECTID) | |
9032 | btrfs_pin_log_trans(dest); | |
9033 | ||
9034 | /* Do the log updates for all inodes. */ | |
9035 | if (old_ino != BTRFS_FIRST_FREE_OBJECTID) | |
d5f5bd54 | 9036 | btrfs_log_new_name(trans, old_dentry, BTRFS_I(old_dir), |
88d2beec | 9037 | old_rename_ctx.index, new_dentry->d_parent); |
259c4b96 | 9038 | if (new_ino != BTRFS_FIRST_FREE_OBJECTID) |
d5f5bd54 | 9039 | btrfs_log_new_name(trans, new_dentry, BTRFS_I(new_dir), |
88d2beec | 9040 | new_rename_ctx.index, old_dentry->d_parent); |
259c4b96 FM |
9041 | |
9042 | /* Now unpin the logs. */ | |
9043 | if (old_ino != BTRFS_FIRST_FREE_OBJECTID) | |
9044 | btrfs_end_log_trans(root); | |
9045 | if (new_ino != BTRFS_FIRST_FREE_OBJECTID) | |
cdd1fedf | 9046 | btrfs_end_log_trans(dest); |
cdd1fedf | 9047 | out_fail: |
75b463d2 FM |
9048 | ret2 = btrfs_end_transaction(trans); |
9049 | ret = ret ? ret : ret2; | |
cdd1fedf | 9050 | out_notrans: |
943eb3bf JB |
9051 | if (new_ino == BTRFS_FIRST_FREE_OBJECTID || |
9052 | old_ino == BTRFS_FIRST_FREE_OBJECTID) | |
0b246afa | 9053 | up_read(&fs_info->subvol_sem); |
cdd1fedf | 9054 | |
ab3c5c18 STD |
9055 | fscrypt_free_filename(&new_fname); |
9056 | fscrypt_free_filename(&old_fname); | |
cdd1fedf DF |
9057 | return ret; |
9058 | } | |
9059 | ||
f2d40141 | 9060 | static struct inode *new_whiteout_inode(struct mnt_idmap *idmap, |
a1fd0c35 OS |
9061 | struct inode *dir) |
9062 | { | |
9063 | struct inode *inode; | |
9064 | ||
9065 | inode = new_inode(dir->i_sb); | |
9066 | if (inode) { | |
f2d40141 | 9067 | inode_init_owner(idmap, inode, dir, |
a1fd0c35 OS |
9068 | S_IFCHR | WHITEOUT_MODE); |
9069 | inode->i_op = &btrfs_special_inode_operations; | |
9070 | init_special_inode(inode, inode->i_mode, WHITEOUT_DEV); | |
9071 | } | |
9072 | return inode; | |
9073 | } | |
9074 | ||
f2d40141 | 9075 | static int btrfs_rename(struct mnt_idmap *idmap, |
ca07274c CB |
9076 | struct inode *old_dir, struct dentry *old_dentry, |
9077 | struct inode *new_dir, struct dentry *new_dentry, | |
9078 | unsigned int flags) | |
39279cc3 | 9079 | { |
0b246afa | 9080 | struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb); |
3538d68d OS |
9081 | struct btrfs_new_inode_args whiteout_args = { |
9082 | .dir = old_dir, | |
9083 | .dentry = old_dentry, | |
9084 | }; | |
39279cc3 | 9085 | struct btrfs_trans_handle *trans; |
5062af35 | 9086 | unsigned int trans_num_items; |
39279cc3 | 9087 | struct btrfs_root *root = BTRFS_I(old_dir)->root; |
4df27c4d | 9088 | struct btrfs_root *dest = BTRFS_I(new_dir)->root; |
2b0143b5 DH |
9089 | struct inode *new_inode = d_inode(new_dentry); |
9090 | struct inode *old_inode = d_inode(old_dentry); | |
88d2beec | 9091 | struct btrfs_rename_ctx rename_ctx; |
00e4e6b3 | 9092 | u64 index = 0; |
39279cc3 | 9093 | int ret; |
75b463d2 | 9094 | int ret2; |
4a0cc7ca | 9095 | u64 old_ino = btrfs_ino(BTRFS_I(old_inode)); |
ab3c5c18 | 9096 | struct fscrypt_name old_fname, new_fname; |
39279cc3 | 9097 | |
4a0cc7ca | 9098 | if (btrfs_ino(BTRFS_I(new_dir)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) |
f679a840 YZ |
9099 | return -EPERM; |
9100 | ||
4df27c4d | 9101 | /* we only allow rename subvolume link between subvolumes */ |
33345d01 | 9102 | if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest) |
3394e160 CM |
9103 | return -EXDEV; |
9104 | ||
33345d01 | 9105 | if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID || |
4a0cc7ca | 9106 | (new_inode && btrfs_ino(BTRFS_I(new_inode)) == BTRFS_FIRST_FREE_OBJECTID)) |
39279cc3 | 9107 | return -ENOTEMPTY; |
5f39d397 | 9108 | |
4df27c4d YZ |
9109 | if (S_ISDIR(old_inode->i_mode) && new_inode && |
9110 | new_inode->i_size > BTRFS_EMPTY_DIR_SIZE) | |
9111 | return -ENOTEMPTY; | |
9c52057c | 9112 | |
ab3c5c18 STD |
9113 | ret = fscrypt_setup_filename(old_dir, &old_dentry->d_name, 0, &old_fname); |
9114 | if (ret) | |
9115 | return ret; | |
9c52057c | 9116 | |
ab3c5c18 STD |
9117 | ret = fscrypt_setup_filename(new_dir, &new_dentry->d_name, 0, &new_fname); |
9118 | if (ret) { | |
9119 | fscrypt_free_filename(&old_fname); | |
9120 | return ret; | |
9121 | } | |
9c52057c | 9122 | |
9c52057c | 9123 | /* check for collisions, even if the name isn't there */ |
6db75318 | 9124 | ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino, &new_fname.disk_name); |
9c52057c CM |
9125 | if (ret) { |
9126 | if (ret == -EEXIST) { | |
9127 | /* we shouldn't get | |
9128 | * eexist without a new_inode */ | |
fae7f21c | 9129 | if (WARN_ON(!new_inode)) { |
ab3c5c18 | 9130 | goto out_fscrypt_names; |
9c52057c CM |
9131 | } |
9132 | } else { | |
9133 | /* maybe -EOVERFLOW */ | |
ab3c5c18 | 9134 | goto out_fscrypt_names; |
9c52057c CM |
9135 | } |
9136 | } | |
9137 | ret = 0; | |
9138 | ||
5a3f23d5 | 9139 | /* |
8d875f95 CM |
9140 | * we're using rename to replace one file with another. Start IO on it |
9141 | * now so we don't add too much work to the end of the transaction | |
5a3f23d5 | 9142 | */ |
8d875f95 | 9143 | if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size) |
5a3f23d5 CM |
9144 | filemap_flush(old_inode->i_mapping); |
9145 | ||
a1fd0c35 | 9146 | if (flags & RENAME_WHITEOUT) { |
f2d40141 | 9147 | whiteout_args.inode = new_whiteout_inode(idmap, old_dir); |
abe3bf74 CJ |
9148 | if (!whiteout_args.inode) { |
9149 | ret = -ENOMEM; | |
9150 | goto out_fscrypt_names; | |
9151 | } | |
3538d68d OS |
9152 | ret = btrfs_new_inode_prepare(&whiteout_args, &trans_num_items); |
9153 | if (ret) | |
9154 | goto out_whiteout_inode; | |
9155 | } else { | |
9156 | /* 1 to update the old parent inode. */ | |
9157 | trans_num_items = 1; | |
a1fd0c35 OS |
9158 | } |
9159 | ||
c1621871 OS |
9160 | if (old_ino == BTRFS_FIRST_FREE_OBJECTID) { |
9161 | /* Close the race window with snapshot create/destroy ioctl */ | |
0b246afa | 9162 | down_read(&fs_info->subvol_sem); |
c1621871 OS |
9163 | /* |
9164 | * 1 to remove old root ref | |
9165 | * 1 to remove old root backref | |
9166 | * 1 to add new root ref | |
9167 | * 1 to add new root backref | |
9168 | */ | |
3538d68d | 9169 | trans_num_items += 4; |
c1621871 OS |
9170 | } else { |
9171 | /* | |
9172 | * 1 to update inode | |
9173 | * 1 to remove old inode ref | |
9174 | * 1 to add new inode ref | |
9175 | */ | |
3538d68d | 9176 | trans_num_items += 3; |
c1621871 | 9177 | } |
a22285a6 | 9178 | /* |
c1621871 OS |
9179 | * 1 to remove old dir item |
9180 | * 1 to remove old dir index | |
c1621871 OS |
9181 | * 1 to add new dir item |
9182 | * 1 to add new dir index | |
a22285a6 | 9183 | */ |
3538d68d OS |
9184 | trans_num_items += 4; |
9185 | /* 1 to update new parent inode if it's not the same as the old parent */ | |
c1621871 OS |
9186 | if (new_dir != old_dir) |
9187 | trans_num_items++; | |
9188 | if (new_inode) { | |
9189 | /* | |
9190 | * 1 to update inode | |
9191 | * 1 to remove inode ref | |
9192 | * 1 to remove dir item | |
9193 | * 1 to remove dir index | |
9194 | * 1 to possibly add orphan item | |
9195 | */ | |
9196 | trans_num_items += 5; | |
9197 | } | |
5062af35 | 9198 | trans = btrfs_start_transaction(root, trans_num_items); |
b44c59a8 | 9199 | if (IS_ERR(trans)) { |
cdd1fedf DF |
9200 | ret = PTR_ERR(trans); |
9201 | goto out_notrans; | |
9202 | } | |
76dda93c | 9203 | |
b0fec6fd JB |
9204 | if (dest != root) { |
9205 | ret = btrfs_record_root_in_trans(trans, dest); | |
9206 | if (ret) | |
9207 | goto out_fail; | |
9208 | } | |
5f39d397 | 9209 | |
877574e2 | 9210 | ret = btrfs_set_inode_index(BTRFS_I(new_dir), &index); |
a5719521 YZ |
9211 | if (ret) |
9212 | goto out_fail; | |
5a3f23d5 | 9213 | |
67de1176 | 9214 | BTRFS_I(old_inode)->dir_index = 0ULL; |
33345d01 | 9215 | if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) { |
4df27c4d | 9216 | /* force full log commit if subvolume involved. */ |
90787766 | 9217 | btrfs_set_log_full_commit(trans); |
4df27c4d | 9218 | } else { |
6db75318 STD |
9219 | ret = btrfs_insert_inode_ref(trans, dest, &new_fname.disk_name, |
9220 | old_ino, btrfs_ino(BTRFS_I(new_dir)), | |
9221 | index); | |
a5719521 YZ |
9222 | if (ret) |
9223 | goto out_fail; | |
4df27c4d | 9224 | } |
5a3f23d5 | 9225 | |
0c4d2d95 JB |
9226 | inode_inc_iversion(old_dir); |
9227 | inode_inc_iversion(new_dir); | |
9228 | inode_inc_iversion(old_inode); | |
c1867eb3 DS |
9229 | old_dir->i_mtime = current_time(old_dir); |
9230 | old_dir->i_ctime = old_dir->i_mtime; | |
9231 | new_dir->i_mtime = old_dir->i_mtime; | |
9232 | new_dir->i_ctime = old_dir->i_mtime; | |
9233 | old_inode->i_ctime = old_dir->i_mtime; | |
5f39d397 | 9234 | |
12fcfd22 | 9235 | if (old_dentry->d_parent != new_dentry->d_parent) |
f85b7379 | 9236 | btrfs_record_unlink_dir(trans, BTRFS_I(old_dir), |
59fcf388 | 9237 | BTRFS_I(old_inode), true); |
12fcfd22 | 9238 | |
33345d01 | 9239 | if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) { |
5b7544cb | 9240 | ret = btrfs_unlink_subvol(trans, BTRFS_I(old_dir), old_dentry); |
4df27c4d | 9241 | } else { |
4467af88 | 9242 | ret = __btrfs_unlink_inode(trans, BTRFS_I(old_dir), |
e43eec81 | 9243 | BTRFS_I(d_inode(old_dentry)), |
6db75318 | 9244 | &old_fname.disk_name, &rename_ctx); |
92986796 | 9245 | if (!ret) |
9a56fcd1 | 9246 | ret = btrfs_update_inode(trans, root, BTRFS_I(old_inode)); |
4df27c4d | 9247 | } |
79787eaa | 9248 | if (ret) { |
66642832 | 9249 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
9250 | goto out_fail; |
9251 | } | |
39279cc3 CM |
9252 | |
9253 | if (new_inode) { | |
0c4d2d95 | 9254 | inode_inc_iversion(new_inode); |
c2050a45 | 9255 | new_inode->i_ctime = current_time(new_inode); |
4a0cc7ca | 9256 | if (unlikely(btrfs_ino(BTRFS_I(new_inode)) == |
4df27c4d | 9257 | BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) { |
5b7544cb | 9258 | ret = btrfs_unlink_subvol(trans, BTRFS_I(new_dir), new_dentry); |
4df27c4d YZ |
9259 | BUG_ON(new_inode->i_nlink == 0); |
9260 | } else { | |
4467af88 | 9261 | ret = btrfs_unlink_inode(trans, BTRFS_I(new_dir), |
4ec5934e | 9262 | BTRFS_I(d_inode(new_dentry)), |
6db75318 | 9263 | &new_fname.disk_name); |
4df27c4d | 9264 | } |
4ef31a45 | 9265 | if (!ret && new_inode->i_nlink == 0) |
73f2e545 NB |
9266 | ret = btrfs_orphan_add(trans, |
9267 | BTRFS_I(d_inode(new_dentry))); | |
79787eaa | 9268 | if (ret) { |
66642832 | 9269 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
9270 | goto out_fail; |
9271 | } | |
39279cc3 | 9272 | } |
aec7477b | 9273 | |
db0a669f | 9274 | ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode), |
6db75318 | 9275 | &new_fname.disk_name, 0, index); |
79787eaa | 9276 | if (ret) { |
66642832 | 9277 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
9278 | goto out_fail; |
9279 | } | |
39279cc3 | 9280 | |
67de1176 MX |
9281 | if (old_inode->i_nlink == 1) |
9282 | BTRFS_I(old_inode)->dir_index = index; | |
9283 | ||
259c4b96 | 9284 | if (old_ino != BTRFS_FIRST_FREE_OBJECTID) |
d5f5bd54 | 9285 | btrfs_log_new_name(trans, old_dentry, BTRFS_I(old_dir), |
88d2beec | 9286 | rename_ctx.index, new_dentry->d_parent); |
cdd1fedf DF |
9287 | |
9288 | if (flags & RENAME_WHITEOUT) { | |
caae78e0 | 9289 | ret = btrfs_create_new_inode(trans, &whiteout_args); |
cdd1fedf | 9290 | if (ret) { |
66642832 | 9291 | btrfs_abort_transaction(trans, ret); |
cdd1fedf | 9292 | goto out_fail; |
caae78e0 OS |
9293 | } else { |
9294 | unlock_new_inode(whiteout_args.inode); | |
9295 | iput(whiteout_args.inode); | |
9296 | whiteout_args.inode = NULL; | |
cdd1fedf | 9297 | } |
4df27c4d | 9298 | } |
39279cc3 | 9299 | out_fail: |
75b463d2 FM |
9300 | ret2 = btrfs_end_transaction(trans); |
9301 | ret = ret ? ret : ret2; | |
b44c59a8 | 9302 | out_notrans: |
33345d01 | 9303 | if (old_ino == BTRFS_FIRST_FREE_OBJECTID) |
0b246afa | 9304 | up_read(&fs_info->subvol_sem); |
a1fd0c35 | 9305 | if (flags & RENAME_WHITEOUT) |
3538d68d OS |
9306 | btrfs_new_inode_args_destroy(&whiteout_args); |
9307 | out_whiteout_inode: | |
9308 | if (flags & RENAME_WHITEOUT) | |
9309 | iput(whiteout_args.inode); | |
ab3c5c18 STD |
9310 | out_fscrypt_names: |
9311 | fscrypt_free_filename(&old_fname); | |
9312 | fscrypt_free_filename(&new_fname); | |
39279cc3 CM |
9313 | return ret; |
9314 | } | |
9315 | ||
e18275ae | 9316 | static int btrfs_rename2(struct mnt_idmap *idmap, struct inode *old_dir, |
549c7297 CB |
9317 | struct dentry *old_dentry, struct inode *new_dir, |
9318 | struct dentry *new_dentry, unsigned int flags) | |
80ace85c | 9319 | { |
ca6dee6b FM |
9320 | int ret; |
9321 | ||
cdd1fedf | 9322 | if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT)) |
80ace85c MS |
9323 | return -EINVAL; |
9324 | ||
cdd1fedf | 9325 | if (flags & RENAME_EXCHANGE) |
ca6dee6b FM |
9326 | ret = btrfs_rename_exchange(old_dir, old_dentry, new_dir, |
9327 | new_dentry); | |
9328 | else | |
f2d40141 | 9329 | ret = btrfs_rename(idmap, old_dir, old_dentry, new_dir, |
ca6dee6b | 9330 | new_dentry, flags); |
cdd1fedf | 9331 | |
ca6dee6b FM |
9332 | btrfs_btree_balance_dirty(BTRFS_I(new_dir)->root->fs_info); |
9333 | ||
9334 | return ret; | |
80ace85c MS |
9335 | } |
9336 | ||
3a2f8c07 NB |
9337 | struct btrfs_delalloc_work { |
9338 | struct inode *inode; | |
9339 | struct completion completion; | |
9340 | struct list_head list; | |
9341 | struct btrfs_work work; | |
9342 | }; | |
9343 | ||
8ccf6f19 MX |
9344 | static void btrfs_run_delalloc_work(struct btrfs_work *work) |
9345 | { | |
9346 | struct btrfs_delalloc_work *delalloc_work; | |
9f23e289 | 9347 | struct inode *inode; |
8ccf6f19 MX |
9348 | |
9349 | delalloc_work = container_of(work, struct btrfs_delalloc_work, | |
9350 | work); | |
9f23e289 | 9351 | inode = delalloc_work->inode; |
30424601 DS |
9352 | filemap_flush(inode->i_mapping); |
9353 | if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT, | |
9354 | &BTRFS_I(inode)->runtime_flags)) | |
9f23e289 | 9355 | filemap_flush(inode->i_mapping); |
8ccf6f19 | 9356 | |
076da91c | 9357 | iput(inode); |
8ccf6f19 MX |
9358 | complete(&delalloc_work->completion); |
9359 | } | |
9360 | ||
3a2f8c07 | 9361 | static struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode) |
8ccf6f19 MX |
9362 | { |
9363 | struct btrfs_delalloc_work *work; | |
9364 | ||
100d5702 | 9365 | work = kmalloc(sizeof(*work), GFP_NOFS); |
8ccf6f19 MX |
9366 | if (!work) |
9367 | return NULL; | |
9368 | ||
9369 | init_completion(&work->completion); | |
9370 | INIT_LIST_HEAD(&work->list); | |
9371 | work->inode = inode; | |
a0cac0ec | 9372 | btrfs_init_work(&work->work, btrfs_run_delalloc_work, NULL, NULL); |
8ccf6f19 MX |
9373 | |
9374 | return work; | |
9375 | } | |
9376 | ||
d352ac68 CM |
9377 | /* |
9378 | * some fairly slow code that needs optimization. This walks the list | |
9379 | * of all the inodes with pending delalloc and forces them to disk. | |
9380 | */ | |
e076ab2a JB |
9381 | static int start_delalloc_inodes(struct btrfs_root *root, |
9382 | struct writeback_control *wbc, bool snapshot, | |
3d45f221 | 9383 | bool in_reclaim_context) |
ea8c2819 | 9384 | { |
ea8c2819 | 9385 | struct btrfs_inode *binode; |
5b21f2ed | 9386 | struct inode *inode; |
8ccf6f19 MX |
9387 | struct btrfs_delalloc_work *work, *next; |
9388 | struct list_head works; | |
1eafa6c7 | 9389 | struct list_head splice; |
8ccf6f19 | 9390 | int ret = 0; |
e076ab2a | 9391 | bool full_flush = wbc->nr_to_write == LONG_MAX; |
ea8c2819 | 9392 | |
8ccf6f19 | 9393 | INIT_LIST_HEAD(&works); |
1eafa6c7 | 9394 | INIT_LIST_HEAD(&splice); |
63607cc8 | 9395 | |
573bfb72 | 9396 | mutex_lock(&root->delalloc_mutex); |
eb73c1b7 MX |
9397 | spin_lock(&root->delalloc_lock); |
9398 | list_splice_init(&root->delalloc_inodes, &splice); | |
1eafa6c7 MX |
9399 | while (!list_empty(&splice)) { |
9400 | binode = list_entry(splice.next, struct btrfs_inode, | |
ea8c2819 | 9401 | delalloc_inodes); |
1eafa6c7 | 9402 | |
eb73c1b7 MX |
9403 | list_move_tail(&binode->delalloc_inodes, |
9404 | &root->delalloc_inodes); | |
3d45f221 FM |
9405 | |
9406 | if (in_reclaim_context && | |
9407 | test_bit(BTRFS_INODE_NO_DELALLOC_FLUSH, &binode->runtime_flags)) | |
9408 | continue; | |
9409 | ||
5b21f2ed | 9410 | inode = igrab(&binode->vfs_inode); |
df0af1a5 | 9411 | if (!inode) { |
eb73c1b7 | 9412 | cond_resched_lock(&root->delalloc_lock); |
1eafa6c7 | 9413 | continue; |
df0af1a5 | 9414 | } |
eb73c1b7 | 9415 | spin_unlock(&root->delalloc_lock); |
1eafa6c7 | 9416 | |
3cd24c69 EL |
9417 | if (snapshot) |
9418 | set_bit(BTRFS_INODE_SNAPSHOT_FLUSH, | |
9419 | &binode->runtime_flags); | |
e076ab2a JB |
9420 | if (full_flush) { |
9421 | work = btrfs_alloc_delalloc_work(inode); | |
9422 | if (!work) { | |
9423 | iput(inode); | |
9424 | ret = -ENOMEM; | |
9425 | goto out; | |
9426 | } | |
9427 | list_add_tail(&work->list, &works); | |
9428 | btrfs_queue_work(root->fs_info->flush_workers, | |
9429 | &work->work); | |
9430 | } else { | |
b3776305 | 9431 | ret = filemap_fdatawrite_wbc(inode->i_mapping, wbc); |
e55cf7ca | 9432 | btrfs_add_delayed_iput(BTRFS_I(inode)); |
e076ab2a | 9433 | if (ret || wbc->nr_to_write <= 0) |
b4912139 JB |
9434 | goto out; |
9435 | } | |
5b21f2ed | 9436 | cond_resched(); |
eb73c1b7 | 9437 | spin_lock(&root->delalloc_lock); |
ea8c2819 | 9438 | } |
eb73c1b7 | 9439 | spin_unlock(&root->delalloc_lock); |
8c8bee1d | 9440 | |
a1ecaabb | 9441 | out: |
eb73c1b7 MX |
9442 | list_for_each_entry_safe(work, next, &works, list) { |
9443 | list_del_init(&work->list); | |
40012f96 NB |
9444 | wait_for_completion(&work->completion); |
9445 | kfree(work); | |
eb73c1b7 MX |
9446 | } |
9447 | ||
81f1d390 | 9448 | if (!list_empty(&splice)) { |
eb73c1b7 MX |
9449 | spin_lock(&root->delalloc_lock); |
9450 | list_splice_tail(&splice, &root->delalloc_inodes); | |
9451 | spin_unlock(&root->delalloc_lock); | |
9452 | } | |
573bfb72 | 9453 | mutex_unlock(&root->delalloc_mutex); |
eb73c1b7 MX |
9454 | return ret; |
9455 | } | |
1eafa6c7 | 9456 | |
f9baa501 | 9457 | int btrfs_start_delalloc_snapshot(struct btrfs_root *root, bool in_reclaim_context) |
eb73c1b7 | 9458 | { |
e076ab2a JB |
9459 | struct writeback_control wbc = { |
9460 | .nr_to_write = LONG_MAX, | |
9461 | .sync_mode = WB_SYNC_NONE, | |
9462 | .range_start = 0, | |
9463 | .range_end = LLONG_MAX, | |
9464 | }; | |
0b246afa | 9465 | struct btrfs_fs_info *fs_info = root->fs_info; |
1eafa6c7 | 9466 | |
84961539 | 9467 | if (BTRFS_FS_ERROR(fs_info)) |
eb73c1b7 MX |
9468 | return -EROFS; |
9469 | ||
f9baa501 | 9470 | return start_delalloc_inodes(root, &wbc, true, in_reclaim_context); |
eb73c1b7 MX |
9471 | } |
9472 | ||
9db4dc24 | 9473 | int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, long nr, |
3d45f221 | 9474 | bool in_reclaim_context) |
eb73c1b7 | 9475 | { |
e076ab2a | 9476 | struct writeback_control wbc = { |
9db4dc24 | 9477 | .nr_to_write = nr, |
e076ab2a JB |
9478 | .sync_mode = WB_SYNC_NONE, |
9479 | .range_start = 0, | |
9480 | .range_end = LLONG_MAX, | |
9481 | }; | |
eb73c1b7 MX |
9482 | struct btrfs_root *root; |
9483 | struct list_head splice; | |
9484 | int ret; | |
9485 | ||
84961539 | 9486 | if (BTRFS_FS_ERROR(fs_info)) |
eb73c1b7 MX |
9487 | return -EROFS; |
9488 | ||
9489 | INIT_LIST_HEAD(&splice); | |
9490 | ||
573bfb72 | 9491 | mutex_lock(&fs_info->delalloc_root_mutex); |
eb73c1b7 MX |
9492 | spin_lock(&fs_info->delalloc_root_lock); |
9493 | list_splice_init(&fs_info->delalloc_roots, &splice); | |
d7830b71 | 9494 | while (!list_empty(&splice)) { |
e076ab2a JB |
9495 | /* |
9496 | * Reset nr_to_write here so we know that we're doing a full | |
9497 | * flush. | |
9498 | */ | |
9db4dc24 | 9499 | if (nr == LONG_MAX) |
e076ab2a JB |
9500 | wbc.nr_to_write = LONG_MAX; |
9501 | ||
eb73c1b7 MX |
9502 | root = list_first_entry(&splice, struct btrfs_root, |
9503 | delalloc_root); | |
00246528 | 9504 | root = btrfs_grab_root(root); |
eb73c1b7 MX |
9505 | BUG_ON(!root); |
9506 | list_move_tail(&root->delalloc_root, | |
9507 | &fs_info->delalloc_roots); | |
9508 | spin_unlock(&fs_info->delalloc_root_lock); | |
9509 | ||
e076ab2a | 9510 | ret = start_delalloc_inodes(root, &wbc, false, in_reclaim_context); |
00246528 | 9511 | btrfs_put_root(root); |
e076ab2a | 9512 | if (ret < 0 || wbc.nr_to_write <= 0) |
eb73c1b7 | 9513 | goto out; |
eb73c1b7 | 9514 | spin_lock(&fs_info->delalloc_root_lock); |
8ccf6f19 | 9515 | } |
eb73c1b7 | 9516 | spin_unlock(&fs_info->delalloc_root_lock); |
1eafa6c7 | 9517 | |
6c255e67 | 9518 | ret = 0; |
eb73c1b7 | 9519 | out: |
81f1d390 | 9520 | if (!list_empty(&splice)) { |
eb73c1b7 MX |
9521 | spin_lock(&fs_info->delalloc_root_lock); |
9522 | list_splice_tail(&splice, &fs_info->delalloc_roots); | |
9523 | spin_unlock(&fs_info->delalloc_root_lock); | |
1eafa6c7 | 9524 | } |
573bfb72 | 9525 | mutex_unlock(&fs_info->delalloc_root_mutex); |
8ccf6f19 | 9526 | return ret; |
ea8c2819 CM |
9527 | } |
9528 | ||
7a77db95 | 9529 | static int btrfs_symlink(struct mnt_idmap *idmap, struct inode *dir, |
549c7297 | 9530 | struct dentry *dentry, const char *symname) |
39279cc3 | 9531 | { |
0b246afa | 9532 | struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb); |
39279cc3 CM |
9533 | struct btrfs_trans_handle *trans; |
9534 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
9535 | struct btrfs_path *path; | |
9536 | struct btrfs_key key; | |
a1fd0c35 | 9537 | struct inode *inode; |
3538d68d OS |
9538 | struct btrfs_new_inode_args new_inode_args = { |
9539 | .dir = dir, | |
9540 | .dentry = dentry, | |
9541 | }; | |
9542 | unsigned int trans_num_items; | |
39279cc3 | 9543 | int err; |
39279cc3 CM |
9544 | int name_len; |
9545 | int datasize; | |
5f39d397 | 9546 | unsigned long ptr; |
39279cc3 | 9547 | struct btrfs_file_extent_item *ei; |
5f39d397 | 9548 | struct extent_buffer *leaf; |
39279cc3 | 9549 | |
f06becc4 | 9550 | name_len = strlen(symname); |
0b246afa | 9551 | if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info)) |
39279cc3 | 9552 | return -ENAMETOOLONG; |
1832a6d5 | 9553 | |
a1fd0c35 OS |
9554 | inode = new_inode(dir->i_sb); |
9555 | if (!inode) | |
9556 | return -ENOMEM; | |
f2d40141 | 9557 | inode_init_owner(idmap, inode, dir, S_IFLNK | S_IRWXUGO); |
a1fd0c35 OS |
9558 | inode->i_op = &btrfs_symlink_inode_operations; |
9559 | inode_nohighmem(inode); | |
9560 | inode->i_mapping->a_ops = &btrfs_aops; | |
caae78e0 OS |
9561 | btrfs_i_size_write(BTRFS_I(inode), name_len); |
9562 | inode_set_bytes(inode, name_len); | |
a1fd0c35 | 9563 | |
3538d68d OS |
9564 | new_inode_args.inode = inode; |
9565 | err = btrfs_new_inode_prepare(&new_inode_args, &trans_num_items); | |
caae78e0 OS |
9566 | if (err) |
9567 | goto out_inode; | |
3538d68d OS |
9568 | /* 1 additional item for the inline extent */ |
9569 | trans_num_items++; | |
9570 | ||
9571 | trans = btrfs_start_transaction(root, trans_num_items); | |
a1fd0c35 | 9572 | if (IS_ERR(trans)) { |
3538d68d OS |
9573 | err = PTR_ERR(trans); |
9574 | goto out_new_inode_args; | |
a1fd0c35 | 9575 | } |
1832a6d5 | 9576 | |
caae78e0 | 9577 | err = btrfs_create_new_inode(trans, &new_inode_args); |
b0d5d10f | 9578 | if (err) |
caae78e0 | 9579 | goto out; |
ad19db71 | 9580 | |
39279cc3 | 9581 | path = btrfs_alloc_path(); |
d8926bb3 MF |
9582 | if (!path) { |
9583 | err = -ENOMEM; | |
caae78e0 OS |
9584 | btrfs_abort_transaction(trans, err); |
9585 | discard_new_inode(inode); | |
9586 | inode = NULL; | |
9587 | goto out; | |
d8926bb3 | 9588 | } |
4a0cc7ca | 9589 | key.objectid = btrfs_ino(BTRFS_I(inode)); |
39279cc3 | 9590 | key.offset = 0; |
962a298f | 9591 | key.type = BTRFS_EXTENT_DATA_KEY; |
39279cc3 CM |
9592 | datasize = btrfs_file_extent_calc_inline_size(name_len); |
9593 | err = btrfs_insert_empty_item(trans, root, path, &key, | |
9594 | datasize); | |
54aa1f4d | 9595 | if (err) { |
caae78e0 | 9596 | btrfs_abort_transaction(trans, err); |
b0839166 | 9597 | btrfs_free_path(path); |
caae78e0 OS |
9598 | discard_new_inode(inode); |
9599 | inode = NULL; | |
9600 | goto out; | |
54aa1f4d | 9601 | } |
5f39d397 CM |
9602 | leaf = path->nodes[0]; |
9603 | ei = btrfs_item_ptr(leaf, path->slots[0], | |
9604 | struct btrfs_file_extent_item); | |
9605 | btrfs_set_file_extent_generation(leaf, ei, trans->transid); | |
9606 | btrfs_set_file_extent_type(leaf, ei, | |
39279cc3 | 9607 | BTRFS_FILE_EXTENT_INLINE); |
c8b97818 CM |
9608 | btrfs_set_file_extent_encryption(leaf, ei, 0); |
9609 | btrfs_set_file_extent_compression(leaf, ei, 0); | |
9610 | btrfs_set_file_extent_other_encoding(leaf, ei, 0); | |
9611 | btrfs_set_file_extent_ram_bytes(leaf, ei, name_len); | |
9612 | ||
39279cc3 | 9613 | ptr = btrfs_file_extent_inline_start(ei); |
5f39d397 CM |
9614 | write_extent_buffer(leaf, symname, ptr, name_len); |
9615 | btrfs_mark_buffer_dirty(leaf); | |
39279cc3 | 9616 | btrfs_free_path(path); |
5f39d397 | 9617 | |
1e2e547a | 9618 | d_instantiate_new(dentry, inode); |
caae78e0 OS |
9619 | err = 0; |
9620 | out: | |
3a45bb20 | 9621 | btrfs_end_transaction(trans); |
2ff7e61e | 9622 | btrfs_btree_balance_dirty(fs_info); |
3538d68d OS |
9623 | out_new_inode_args: |
9624 | btrfs_new_inode_args_destroy(&new_inode_args); | |
caae78e0 OS |
9625 | out_inode: |
9626 | if (err) | |
9627 | iput(inode); | |
39279cc3 CM |
9628 | return err; |
9629 | } | |
16432985 | 9630 | |
8fccebfa FM |
9631 | static struct btrfs_trans_handle *insert_prealloc_file_extent( |
9632 | struct btrfs_trans_handle *trans_in, | |
90dffd0c NB |
9633 | struct btrfs_inode *inode, |
9634 | struct btrfs_key *ins, | |
203f44c5 QW |
9635 | u64 file_offset) |
9636 | { | |
9637 | struct btrfs_file_extent_item stack_fi; | |
bf385648 | 9638 | struct btrfs_replace_extent_info extent_info; |
8fccebfa FM |
9639 | struct btrfs_trans_handle *trans = trans_in; |
9640 | struct btrfs_path *path; | |
203f44c5 QW |
9641 | u64 start = ins->objectid; |
9642 | u64 len = ins->offset; | |
fbf48bb0 | 9643 | int qgroup_released; |
9729f10a | 9644 | int ret; |
203f44c5 QW |
9645 | |
9646 | memset(&stack_fi, 0, sizeof(stack_fi)); | |
9647 | ||
9648 | btrfs_set_stack_file_extent_type(&stack_fi, BTRFS_FILE_EXTENT_PREALLOC); | |
9649 | btrfs_set_stack_file_extent_disk_bytenr(&stack_fi, start); | |
9650 | btrfs_set_stack_file_extent_disk_num_bytes(&stack_fi, len); | |
9651 | btrfs_set_stack_file_extent_num_bytes(&stack_fi, len); | |
9652 | btrfs_set_stack_file_extent_ram_bytes(&stack_fi, len); | |
9653 | btrfs_set_stack_file_extent_compression(&stack_fi, BTRFS_COMPRESS_NONE); | |
9654 | /* Encryption and other encoding is reserved and all 0 */ | |
9655 | ||
fbf48bb0 QW |
9656 | qgroup_released = btrfs_qgroup_release_data(inode, file_offset, len); |
9657 | if (qgroup_released < 0) | |
9658 | return ERR_PTR(qgroup_released); | |
8fccebfa FM |
9659 | |
9660 | if (trans) { | |
90dffd0c | 9661 | ret = insert_reserved_file_extent(trans, inode, |
2766ff61 | 9662 | file_offset, &stack_fi, |
fbf48bb0 | 9663 | true, qgroup_released); |
8fccebfa | 9664 | if (ret) |
a3ee79bd | 9665 | goto free_qgroup; |
8fccebfa FM |
9666 | return trans; |
9667 | } | |
9668 | ||
9669 | extent_info.disk_offset = start; | |
9670 | extent_info.disk_len = len; | |
9671 | extent_info.data_offset = 0; | |
9672 | extent_info.data_len = len; | |
9673 | extent_info.file_offset = file_offset; | |
9674 | extent_info.extent_buf = (char *)&stack_fi; | |
8fccebfa | 9675 | extent_info.is_new_extent = true; |
983d8209 | 9676 | extent_info.update_times = true; |
fbf48bb0 | 9677 | extent_info.qgroup_reserved = qgroup_released; |
8fccebfa FM |
9678 | extent_info.insertions = 0; |
9679 | ||
9680 | path = btrfs_alloc_path(); | |
a3ee79bd QW |
9681 | if (!path) { |
9682 | ret = -ENOMEM; | |
9683 | goto free_qgroup; | |
9684 | } | |
8fccebfa | 9685 | |
bfc78479 | 9686 | ret = btrfs_replace_file_extents(inode, path, file_offset, |
8fccebfa FM |
9687 | file_offset + len - 1, &extent_info, |
9688 | &trans); | |
9689 | btrfs_free_path(path); | |
9690 | if (ret) | |
a3ee79bd | 9691 | goto free_qgroup; |
8fccebfa | 9692 | return trans; |
a3ee79bd QW |
9693 | |
9694 | free_qgroup: | |
9695 | /* | |
9696 | * We have released qgroup data range at the beginning of the function, | |
9697 | * and normally qgroup_released bytes will be freed when committing | |
9698 | * transaction. | |
9699 | * But if we error out early, we have to free what we have released | |
9700 | * or we leak qgroup data reservation. | |
9701 | */ | |
9702 | btrfs_qgroup_free_refroot(inode->root->fs_info, | |
9703 | inode->root->root_key.objectid, qgroup_released, | |
9704 | BTRFS_QGROUP_RSV_DATA); | |
9705 | return ERR_PTR(ret); | |
203f44c5 | 9706 | } |
8fccebfa | 9707 | |
0af3d00b JB |
9708 | static int __btrfs_prealloc_file_range(struct inode *inode, int mode, |
9709 | u64 start, u64 num_bytes, u64 min_size, | |
9710 | loff_t actual_len, u64 *alloc_hint, | |
9711 | struct btrfs_trans_handle *trans) | |
d899e052 | 9712 | { |
0b246afa | 9713 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
5dc562c5 | 9714 | struct extent_map *em; |
d899e052 YZ |
9715 | struct btrfs_root *root = BTRFS_I(inode)->root; |
9716 | struct btrfs_key ins; | |
d899e052 | 9717 | u64 cur_offset = start; |
b778cf96 | 9718 | u64 clear_offset = start; |
55a61d1d | 9719 | u64 i_size; |
154ea289 | 9720 | u64 cur_bytes; |
0b670dc4 | 9721 | u64 last_alloc = (u64)-1; |
d899e052 | 9722 | int ret = 0; |
0af3d00b | 9723 | bool own_trans = true; |
18513091 | 9724 | u64 end = start + num_bytes - 1; |
d899e052 | 9725 | |
0af3d00b JB |
9726 | if (trans) |
9727 | own_trans = false; | |
d899e052 | 9728 | while (num_bytes > 0) { |
ee22184b | 9729 | cur_bytes = min_t(u64, num_bytes, SZ_256M); |
154ea289 | 9730 | cur_bytes = max(cur_bytes, min_size); |
0b670dc4 JB |
9731 | /* |
9732 | * If we are severely fragmented we could end up with really | |
9733 | * small allocations, so if the allocator is returning small | |
9734 | * chunks lets make its job easier by only searching for those | |
9735 | * sized chunks. | |
9736 | */ | |
9737 | cur_bytes = min(cur_bytes, last_alloc); | |
18513091 WX |
9738 | ret = btrfs_reserve_extent(root, cur_bytes, cur_bytes, |
9739 | min_size, 0, *alloc_hint, &ins, 1, 0); | |
8fccebfa | 9740 | if (ret) |
a22285a6 | 9741 | break; |
b778cf96 JB |
9742 | |
9743 | /* | |
9744 | * We've reserved this space, and thus converted it from | |
9745 | * ->bytes_may_use to ->bytes_reserved. Any error that happens | |
9746 | * from here on out we will only need to clear our reservation | |
9747 | * for the remaining unreserved area, so advance our | |
9748 | * clear_offset by our extent size. | |
9749 | */ | |
9750 | clear_offset += ins.offset; | |
5a303d5d | 9751 | |
0b670dc4 | 9752 | last_alloc = ins.offset; |
90dffd0c NB |
9753 | trans = insert_prealloc_file_extent(trans, BTRFS_I(inode), |
9754 | &ins, cur_offset); | |
1afc708d FM |
9755 | /* |
9756 | * Now that we inserted the prealloc extent we can finally | |
9757 | * decrement the number of reservations in the block group. | |
9758 | * If we did it before, we could race with relocation and have | |
9759 | * relocation miss the reserved extent, making it fail later. | |
9760 | */ | |
9761 | btrfs_dec_block_group_reservations(fs_info, ins.objectid); | |
8fccebfa FM |
9762 | if (IS_ERR(trans)) { |
9763 | ret = PTR_ERR(trans); | |
2ff7e61e | 9764 | btrfs_free_reserved_extent(fs_info, ins.objectid, |
e570fd27 | 9765 | ins.offset, 0); |
79787eaa JM |
9766 | break; |
9767 | } | |
31193213 | 9768 | |
5dc562c5 JB |
9769 | em = alloc_extent_map(); |
9770 | if (!em) { | |
a1ba4c08 FM |
9771 | btrfs_drop_extent_map_range(BTRFS_I(inode), cur_offset, |
9772 | cur_offset + ins.offset - 1, false); | |
23e3337f | 9773 | btrfs_set_inode_full_sync(BTRFS_I(inode)); |
5dc562c5 JB |
9774 | goto next; |
9775 | } | |
9776 | ||
9777 | em->start = cur_offset; | |
9778 | em->orig_start = cur_offset; | |
9779 | em->len = ins.offset; | |
9780 | em->block_start = ins.objectid; | |
9781 | em->block_len = ins.offset; | |
b4939680 | 9782 | em->orig_block_len = ins.offset; |
cc95bef6 | 9783 | em->ram_bytes = ins.offset; |
5dc562c5 JB |
9784 | set_bit(EXTENT_FLAG_PREALLOC, &em->flags); |
9785 | em->generation = trans->transid; | |
9786 | ||
a1ba4c08 | 9787 | ret = btrfs_replace_extent_map_range(BTRFS_I(inode), em, true); |
5dc562c5 JB |
9788 | free_extent_map(em); |
9789 | next: | |
d899e052 YZ |
9790 | num_bytes -= ins.offset; |
9791 | cur_offset += ins.offset; | |
efa56464 | 9792 | *alloc_hint = ins.objectid + ins.offset; |
5a303d5d | 9793 | |
0c4d2d95 | 9794 | inode_inc_iversion(inode); |
c2050a45 | 9795 | inode->i_ctime = current_time(inode); |
6cbff00f | 9796 | BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC; |
d899e052 | 9797 | if (!(mode & FALLOC_FL_KEEP_SIZE) && |
efa56464 YZ |
9798 | (actual_len > inode->i_size) && |
9799 | (cur_offset > inode->i_size)) { | |
d1ea6a61 | 9800 | if (cur_offset > actual_len) |
55a61d1d | 9801 | i_size = actual_len; |
d1ea6a61 | 9802 | else |
55a61d1d JB |
9803 | i_size = cur_offset; |
9804 | i_size_write(inode, i_size); | |
76aea537 | 9805 | btrfs_inode_safe_disk_i_size_write(BTRFS_I(inode), 0); |
5a303d5d YZ |
9806 | } |
9807 | ||
9a56fcd1 | 9808 | ret = btrfs_update_inode(trans, root, BTRFS_I(inode)); |
79787eaa JM |
9809 | |
9810 | if (ret) { | |
66642832 | 9811 | btrfs_abort_transaction(trans, ret); |
79787eaa | 9812 | if (own_trans) |
3a45bb20 | 9813 | btrfs_end_transaction(trans); |
79787eaa JM |
9814 | break; |
9815 | } | |
d899e052 | 9816 | |
8fccebfa | 9817 | if (own_trans) { |
3a45bb20 | 9818 | btrfs_end_transaction(trans); |
8fccebfa FM |
9819 | trans = NULL; |
9820 | } | |
5a303d5d | 9821 | } |
b778cf96 | 9822 | if (clear_offset < end) |
25ce28ca | 9823 | btrfs_free_reserved_data_space(BTRFS_I(inode), NULL, clear_offset, |
b778cf96 | 9824 | end - clear_offset + 1); |
d899e052 YZ |
9825 | return ret; |
9826 | } | |
9827 | ||
0af3d00b JB |
9828 | int btrfs_prealloc_file_range(struct inode *inode, int mode, |
9829 | u64 start, u64 num_bytes, u64 min_size, | |
9830 | loff_t actual_len, u64 *alloc_hint) | |
9831 | { | |
9832 | return __btrfs_prealloc_file_range(inode, mode, start, num_bytes, | |
9833 | min_size, actual_len, alloc_hint, | |
9834 | NULL); | |
9835 | } | |
9836 | ||
9837 | int btrfs_prealloc_file_range_trans(struct inode *inode, | |
9838 | struct btrfs_trans_handle *trans, int mode, | |
9839 | u64 start, u64 num_bytes, u64 min_size, | |
9840 | loff_t actual_len, u64 *alloc_hint) | |
9841 | { | |
9842 | return __btrfs_prealloc_file_range(inode, mode, start, num_bytes, | |
9843 | min_size, actual_len, alloc_hint, trans); | |
9844 | } | |
9845 | ||
4609e1f1 | 9846 | static int btrfs_permission(struct mnt_idmap *idmap, |
549c7297 | 9847 | struct inode *inode, int mask) |
fdebe2bd | 9848 | { |
b83cc969 | 9849 | struct btrfs_root *root = BTRFS_I(inode)->root; |
cb6db4e5 | 9850 | umode_t mode = inode->i_mode; |
b83cc969 | 9851 | |
cb6db4e5 JM |
9852 | if (mask & MAY_WRITE && |
9853 | (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) { | |
9854 | if (btrfs_root_readonly(root)) | |
9855 | return -EROFS; | |
9856 | if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY) | |
9857 | return -EACCES; | |
9858 | } | |
4609e1f1 | 9859 | return generic_permission(idmap, inode, mask); |
fdebe2bd | 9860 | } |
39279cc3 | 9861 | |
011e2b71 | 9862 | static int btrfs_tmpfile(struct mnt_idmap *idmap, struct inode *dir, |
863f144f | 9863 | struct file *file, umode_t mode) |
ef3b9af5 | 9864 | { |
2ff7e61e | 9865 | struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb); |
ef3b9af5 FM |
9866 | struct btrfs_trans_handle *trans; |
9867 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
a1fd0c35 | 9868 | struct inode *inode; |
3538d68d OS |
9869 | struct btrfs_new_inode_args new_inode_args = { |
9870 | .dir = dir, | |
863f144f | 9871 | .dentry = file->f_path.dentry, |
3538d68d OS |
9872 | .orphan = true, |
9873 | }; | |
9874 | unsigned int trans_num_items; | |
a1fd0c35 OS |
9875 | int ret; |
9876 | ||
9877 | inode = new_inode(dir->i_sb); | |
9878 | if (!inode) | |
9879 | return -ENOMEM; | |
f2d40141 | 9880 | inode_init_owner(idmap, inode, dir, mode); |
a1fd0c35 OS |
9881 | inode->i_fop = &btrfs_file_operations; |
9882 | inode->i_op = &btrfs_file_inode_operations; | |
9883 | inode->i_mapping->a_ops = &btrfs_aops; | |
ef3b9af5 | 9884 | |
3538d68d OS |
9885 | new_inode_args.inode = inode; |
9886 | ret = btrfs_new_inode_prepare(&new_inode_args, &trans_num_items); | |
caae78e0 OS |
9887 | if (ret) |
9888 | goto out_inode; | |
3538d68d OS |
9889 | |
9890 | trans = btrfs_start_transaction(root, trans_num_items); | |
a1fd0c35 | 9891 | if (IS_ERR(trans)) { |
3538d68d OS |
9892 | ret = PTR_ERR(trans); |
9893 | goto out_new_inode_args; | |
a1fd0c35 | 9894 | } |
ef3b9af5 | 9895 | |
caae78e0 | 9896 | ret = btrfs_create_new_inode(trans, &new_inode_args); |
ef3b9af5 | 9897 | |
5762b5c9 | 9898 | /* |
3538d68d OS |
9899 | * We set number of links to 0 in btrfs_create_new_inode(), and here we |
9900 | * set it to 1 because d_tmpfile() will issue a warning if the count is | |
9901 | * 0, through: | |
5762b5c9 FM |
9902 | * |
9903 | * d_tmpfile() -> inode_dec_link_count() -> drop_nlink() | |
9904 | */ | |
9905 | set_nlink(inode, 1); | |
caae78e0 OS |
9906 | |
9907 | if (!ret) { | |
863f144f | 9908 | d_tmpfile(file, inode); |
caae78e0 OS |
9909 | unlock_new_inode(inode); |
9910 | mark_inode_dirty(inode); | |
9911 | } | |
9912 | ||
3a45bb20 | 9913 | btrfs_end_transaction(trans); |
2ff7e61e | 9914 | btrfs_btree_balance_dirty(fs_info); |
3538d68d OS |
9915 | out_new_inode_args: |
9916 | btrfs_new_inode_args_destroy(&new_inode_args); | |
caae78e0 OS |
9917 | out_inode: |
9918 | if (ret) | |
9919 | iput(inode); | |
863f144f | 9920 | return finish_open_simple(file, ret); |
ef3b9af5 FM |
9921 | } |
9922 | ||
d2a91064 | 9923 | void btrfs_set_range_writeback(struct btrfs_inode *inode, u64 start, u64 end) |
c6100a4b | 9924 | { |
d2a91064 | 9925 | struct btrfs_fs_info *fs_info = inode->root->fs_info; |
c6100a4b JB |
9926 | unsigned long index = start >> PAGE_SHIFT; |
9927 | unsigned long end_index = end >> PAGE_SHIFT; | |
9928 | struct page *page; | |
d2a91064 | 9929 | u32 len; |
c6100a4b | 9930 | |
d2a91064 QW |
9931 | ASSERT(end + 1 - start <= U32_MAX); |
9932 | len = end + 1 - start; | |
c6100a4b | 9933 | while (index <= end_index) { |
d2a91064 | 9934 | page = find_get_page(inode->vfs_inode.i_mapping, index); |
c6100a4b | 9935 | ASSERT(page); /* Pages should be in the extent_io_tree */ |
d2a91064 QW |
9936 | |
9937 | btrfs_page_set_writeback(fs_info, page, start, len); | |
c6100a4b JB |
9938 | put_page(page); |
9939 | index++; | |
9940 | } | |
9941 | } | |
9942 | ||
3ea4dc5b OS |
9943 | int btrfs_encoded_io_compression_from_extent(struct btrfs_fs_info *fs_info, |
9944 | int compress_type) | |
1881fba8 OS |
9945 | { |
9946 | switch (compress_type) { | |
9947 | case BTRFS_COMPRESS_NONE: | |
9948 | return BTRFS_ENCODED_IO_COMPRESSION_NONE; | |
9949 | case BTRFS_COMPRESS_ZLIB: | |
9950 | return BTRFS_ENCODED_IO_COMPRESSION_ZLIB; | |
9951 | case BTRFS_COMPRESS_LZO: | |
9952 | /* | |
9953 | * The LZO format depends on the sector size. 64K is the maximum | |
9954 | * sector size that we support. | |
9955 | */ | |
9956 | if (fs_info->sectorsize < SZ_4K || fs_info->sectorsize > SZ_64K) | |
9957 | return -EINVAL; | |
9958 | return BTRFS_ENCODED_IO_COMPRESSION_LZO_4K + | |
9959 | (fs_info->sectorsize_bits - 12); | |
9960 | case BTRFS_COMPRESS_ZSTD: | |
9961 | return BTRFS_ENCODED_IO_COMPRESSION_ZSTD; | |
9962 | default: | |
9963 | return -EUCLEAN; | |
9964 | } | |
9965 | } | |
9966 | ||
9967 | static ssize_t btrfs_encoded_read_inline( | |
9968 | struct kiocb *iocb, | |
9969 | struct iov_iter *iter, u64 start, | |
9970 | u64 lockend, | |
9971 | struct extent_state **cached_state, | |
9972 | u64 extent_start, size_t count, | |
9973 | struct btrfs_ioctl_encoded_io_args *encoded, | |
9974 | bool *unlocked) | |
9975 | { | |
9976 | struct btrfs_inode *inode = BTRFS_I(file_inode(iocb->ki_filp)); | |
9977 | struct btrfs_root *root = inode->root; | |
9978 | struct btrfs_fs_info *fs_info = root->fs_info; | |
9979 | struct extent_io_tree *io_tree = &inode->io_tree; | |
9980 | struct btrfs_path *path; | |
9981 | struct extent_buffer *leaf; | |
9982 | struct btrfs_file_extent_item *item; | |
9983 | u64 ram_bytes; | |
9984 | unsigned long ptr; | |
9985 | void *tmp; | |
9986 | ssize_t ret; | |
9987 | ||
9988 | path = btrfs_alloc_path(); | |
9989 | if (!path) { | |
9990 | ret = -ENOMEM; | |
9991 | goto out; | |
9992 | } | |
9993 | ret = btrfs_lookup_file_extent(NULL, root, path, btrfs_ino(inode), | |
9994 | extent_start, 0); | |
9995 | if (ret) { | |
9996 | if (ret > 0) { | |
9997 | /* The extent item disappeared? */ | |
9998 | ret = -EIO; | |
9999 | } | |
10000 | goto out; | |
10001 | } | |
10002 | leaf = path->nodes[0]; | |
10003 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_file_extent_item); | |
10004 | ||
10005 | ram_bytes = btrfs_file_extent_ram_bytes(leaf, item); | |
10006 | ptr = btrfs_file_extent_inline_start(item); | |
10007 | ||
10008 | encoded->len = min_t(u64, extent_start + ram_bytes, | |
10009 | inode->vfs_inode.i_size) - iocb->ki_pos; | |
10010 | ret = btrfs_encoded_io_compression_from_extent(fs_info, | |
10011 | btrfs_file_extent_compression(leaf, item)); | |
10012 | if (ret < 0) | |
10013 | goto out; | |
10014 | encoded->compression = ret; | |
10015 | if (encoded->compression) { | |
10016 | size_t inline_size; | |
10017 | ||
10018 | inline_size = btrfs_file_extent_inline_item_len(leaf, | |
10019 | path->slots[0]); | |
10020 | if (inline_size > count) { | |
10021 | ret = -ENOBUFS; | |
10022 | goto out; | |
10023 | } | |
10024 | count = inline_size; | |
10025 | encoded->unencoded_len = ram_bytes; | |
10026 | encoded->unencoded_offset = iocb->ki_pos - extent_start; | |
10027 | } else { | |
10028 | count = min_t(u64, count, encoded->len); | |
10029 | encoded->len = count; | |
10030 | encoded->unencoded_len = count; | |
10031 | ptr += iocb->ki_pos - extent_start; | |
10032 | } | |
10033 | ||
10034 | tmp = kmalloc(count, GFP_NOFS); | |
10035 | if (!tmp) { | |
10036 | ret = -ENOMEM; | |
10037 | goto out; | |
10038 | } | |
10039 | read_extent_buffer(leaf, tmp, ptr, count); | |
10040 | btrfs_release_path(path); | |
570eb97b | 10041 | unlock_extent(io_tree, start, lockend, cached_state); |
e5d4d75b | 10042 | btrfs_inode_unlock(inode, BTRFS_ILOCK_SHARED); |
1881fba8 OS |
10043 | *unlocked = true; |
10044 | ||
10045 | ret = copy_to_iter(tmp, count, iter); | |
10046 | if (ret != count) | |
10047 | ret = -EFAULT; | |
10048 | kfree(tmp); | |
10049 | out: | |
10050 | btrfs_free_path(path); | |
10051 | return ret; | |
10052 | } | |
10053 | ||
10054 | struct btrfs_encoded_read_private { | |
1881fba8 OS |
10055 | wait_queue_head_t wait; |
10056 | atomic_t pending; | |
10057 | blk_status_t status; | |
1881fba8 OS |
10058 | }; |
10059 | ||
917f32a2 | 10060 | static void btrfs_encoded_read_endio(struct btrfs_bio *bbio) |
1881fba8 | 10061 | { |
917f32a2 | 10062 | struct btrfs_encoded_read_private *priv = bbio->private; |
1881fba8 | 10063 | |
7609afac | 10064 | if (bbio->bio.bi_status) { |
1881fba8 OS |
10065 | /* |
10066 | * The memory barrier implied by the atomic_dec_return() here | |
10067 | * pairs with the memory barrier implied by the | |
10068 | * atomic_dec_return() or io_wait_event() in | |
10069 | * btrfs_encoded_read_regular_fill_pages() to ensure that this | |
10070 | * write is observed before the load of status in | |
10071 | * btrfs_encoded_read_regular_fill_pages(). | |
10072 | */ | |
7609afac | 10073 | WRITE_ONCE(priv->status, bbio->bio.bi_status); |
1881fba8 OS |
10074 | } |
10075 | if (!atomic_dec_return(&priv->pending)) | |
10076 | wake_up(&priv->wait); | |
917f32a2 | 10077 | bio_put(&bbio->bio); |
1881fba8 OS |
10078 | } |
10079 | ||
3ea4dc5b OS |
10080 | int btrfs_encoded_read_regular_fill_pages(struct btrfs_inode *inode, |
10081 | u64 file_offset, u64 disk_bytenr, | |
10082 | u64 disk_io_size, struct page **pages) | |
1881fba8 | 10083 | { |
4317ff00 | 10084 | struct btrfs_fs_info *fs_info = inode->root->fs_info; |
1881fba8 | 10085 | struct btrfs_encoded_read_private priv = { |
1881fba8 | 10086 | .pending = ATOMIC_INIT(1), |
1881fba8 OS |
10087 | }; |
10088 | unsigned long i = 0; | |
b41bbd29 | 10089 | struct btrfs_bio *bbio; |
1881fba8 OS |
10090 | |
10091 | init_waitqueue_head(&priv.wait); | |
1881fba8 | 10092 | |
4317ff00 QW |
10093 | bbio = btrfs_bio_alloc(BIO_MAX_VECS, REQ_OP_READ, fs_info, |
10094 | btrfs_encoded_read_endio, &priv); | |
b41bbd29 | 10095 | bbio->bio.bi_iter.bi_sector = disk_bytenr >> SECTOR_SHIFT; |
4317ff00 | 10096 | bbio->inode = inode; |
1881fba8 | 10097 | |
34f888ce CH |
10098 | do { |
10099 | size_t bytes = min_t(u64, disk_io_size, PAGE_SIZE); | |
10100 | ||
b41bbd29 | 10101 | if (bio_add_page(&bbio->bio, pages[i], bytes, 0) < bytes) { |
34f888ce | 10102 | atomic_inc(&priv.pending); |
b41bbd29 | 10103 | btrfs_submit_bio(bbio, 0); |
34f888ce | 10104 | |
4317ff00 | 10105 | bbio = btrfs_bio_alloc(BIO_MAX_VECS, REQ_OP_READ, fs_info, |
b41bbd29 CH |
10106 | btrfs_encoded_read_endio, &priv); |
10107 | bbio->bio.bi_iter.bi_sector = disk_bytenr >> SECTOR_SHIFT; | |
4317ff00 | 10108 | bbio->inode = inode; |
34f888ce | 10109 | continue; |
1881fba8 | 10110 | } |
34f888ce CH |
10111 | |
10112 | i++; | |
10113 | disk_bytenr += bytes; | |
10114 | disk_io_size -= bytes; | |
10115 | } while (disk_io_size); | |
10116 | ||
10117 | atomic_inc(&priv.pending); | |
b41bbd29 | 10118 | btrfs_submit_bio(bbio, 0); |
1881fba8 | 10119 | |
1881fba8 OS |
10120 | if (atomic_dec_return(&priv.pending)) |
10121 | io_wait_event(priv.wait, !atomic_read(&priv.pending)); | |
10122 | /* See btrfs_encoded_read_endio() for ordering. */ | |
10123 | return blk_status_to_errno(READ_ONCE(priv.status)); | |
10124 | } | |
10125 | ||
10126 | static ssize_t btrfs_encoded_read_regular(struct kiocb *iocb, | |
10127 | struct iov_iter *iter, | |
10128 | u64 start, u64 lockend, | |
10129 | struct extent_state **cached_state, | |
10130 | u64 disk_bytenr, u64 disk_io_size, | |
10131 | size_t count, bool compressed, | |
10132 | bool *unlocked) | |
10133 | { | |
10134 | struct btrfs_inode *inode = BTRFS_I(file_inode(iocb->ki_filp)); | |
10135 | struct extent_io_tree *io_tree = &inode->io_tree; | |
10136 | struct page **pages; | |
10137 | unsigned long nr_pages, i; | |
10138 | u64 cur; | |
10139 | size_t page_offset; | |
10140 | ssize_t ret; | |
10141 | ||
10142 | nr_pages = DIV_ROUND_UP(disk_io_size, PAGE_SIZE); | |
10143 | pages = kcalloc(nr_pages, sizeof(struct page *), GFP_NOFS); | |
10144 | if (!pages) | |
10145 | return -ENOMEM; | |
dd137dd1 STD |
10146 | ret = btrfs_alloc_page_array(nr_pages, pages); |
10147 | if (ret) { | |
10148 | ret = -ENOMEM; | |
10149 | goto out; | |
1881fba8 | 10150 | } |
1881fba8 OS |
10151 | |
10152 | ret = btrfs_encoded_read_regular_fill_pages(inode, start, disk_bytenr, | |
10153 | disk_io_size, pages); | |
10154 | if (ret) | |
10155 | goto out; | |
10156 | ||
570eb97b | 10157 | unlock_extent(io_tree, start, lockend, cached_state); |
e5d4d75b | 10158 | btrfs_inode_unlock(inode, BTRFS_ILOCK_SHARED); |
1881fba8 OS |
10159 | *unlocked = true; |
10160 | ||
10161 | if (compressed) { | |
10162 | i = 0; | |
10163 | page_offset = 0; | |
10164 | } else { | |
10165 | i = (iocb->ki_pos - start) >> PAGE_SHIFT; | |
10166 | page_offset = (iocb->ki_pos - start) & (PAGE_SIZE - 1); | |
10167 | } | |
10168 | cur = 0; | |
10169 | while (cur < count) { | |
10170 | size_t bytes = min_t(size_t, count - cur, | |
10171 | PAGE_SIZE - page_offset); | |
10172 | ||
10173 | if (copy_page_to_iter(pages[i], page_offset, bytes, | |
10174 | iter) != bytes) { | |
10175 | ret = -EFAULT; | |
10176 | goto out; | |
10177 | } | |
10178 | i++; | |
10179 | cur += bytes; | |
10180 | page_offset = 0; | |
10181 | } | |
10182 | ret = count; | |
10183 | out: | |
10184 | for (i = 0; i < nr_pages; i++) { | |
10185 | if (pages[i]) | |
10186 | __free_page(pages[i]); | |
10187 | } | |
10188 | kfree(pages); | |
10189 | return ret; | |
10190 | } | |
10191 | ||
10192 | ssize_t btrfs_encoded_read(struct kiocb *iocb, struct iov_iter *iter, | |
10193 | struct btrfs_ioctl_encoded_io_args *encoded) | |
10194 | { | |
10195 | struct btrfs_inode *inode = BTRFS_I(file_inode(iocb->ki_filp)); | |
10196 | struct btrfs_fs_info *fs_info = inode->root->fs_info; | |
10197 | struct extent_io_tree *io_tree = &inode->io_tree; | |
10198 | ssize_t ret; | |
10199 | size_t count = iov_iter_count(iter); | |
10200 | u64 start, lockend, disk_bytenr, disk_io_size; | |
10201 | struct extent_state *cached_state = NULL; | |
10202 | struct extent_map *em; | |
10203 | bool unlocked = false; | |
10204 | ||
10205 | file_accessed(iocb->ki_filp); | |
10206 | ||
29b6352b | 10207 | btrfs_inode_lock(inode, BTRFS_ILOCK_SHARED); |
1881fba8 OS |
10208 | |
10209 | if (iocb->ki_pos >= inode->vfs_inode.i_size) { | |
e5d4d75b | 10210 | btrfs_inode_unlock(inode, BTRFS_ILOCK_SHARED); |
1881fba8 OS |
10211 | return 0; |
10212 | } | |
10213 | start = ALIGN_DOWN(iocb->ki_pos, fs_info->sectorsize); | |
10214 | /* | |
10215 | * We don't know how long the extent containing iocb->ki_pos is, but if | |
10216 | * it's compressed we know that it won't be longer than this. | |
10217 | */ | |
10218 | lockend = start + BTRFS_MAX_UNCOMPRESSED - 1; | |
10219 | ||
10220 | for (;;) { | |
10221 | struct btrfs_ordered_extent *ordered; | |
10222 | ||
10223 | ret = btrfs_wait_ordered_range(&inode->vfs_inode, start, | |
10224 | lockend - start + 1); | |
10225 | if (ret) | |
10226 | goto out_unlock_inode; | |
570eb97b | 10227 | lock_extent(io_tree, start, lockend, &cached_state); |
1881fba8 OS |
10228 | ordered = btrfs_lookup_ordered_range(inode, start, |
10229 | lockend - start + 1); | |
10230 | if (!ordered) | |
10231 | break; | |
10232 | btrfs_put_ordered_extent(ordered); | |
570eb97b | 10233 | unlock_extent(io_tree, start, lockend, &cached_state); |
1881fba8 OS |
10234 | cond_resched(); |
10235 | } | |
10236 | ||
10237 | em = btrfs_get_extent(inode, NULL, 0, start, lockend - start + 1); | |
10238 | if (IS_ERR(em)) { | |
10239 | ret = PTR_ERR(em); | |
10240 | goto out_unlock_extent; | |
10241 | } | |
10242 | ||
10243 | if (em->block_start == EXTENT_MAP_INLINE) { | |
10244 | u64 extent_start = em->start; | |
10245 | ||
10246 | /* | |
10247 | * For inline extents we get everything we need out of the | |
10248 | * extent item. | |
10249 | */ | |
10250 | free_extent_map(em); | |
10251 | em = NULL; | |
10252 | ret = btrfs_encoded_read_inline(iocb, iter, start, lockend, | |
10253 | &cached_state, extent_start, | |
10254 | count, encoded, &unlocked); | |
10255 | goto out; | |
10256 | } | |
10257 | ||
10258 | /* | |
10259 | * We only want to return up to EOF even if the extent extends beyond | |
10260 | * that. | |
10261 | */ | |
10262 | encoded->len = min_t(u64, extent_map_end(em), | |
10263 | inode->vfs_inode.i_size) - iocb->ki_pos; | |
10264 | if (em->block_start == EXTENT_MAP_HOLE || | |
10265 | test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) { | |
10266 | disk_bytenr = EXTENT_MAP_HOLE; | |
10267 | count = min_t(u64, count, encoded->len); | |
10268 | encoded->len = count; | |
10269 | encoded->unencoded_len = count; | |
10270 | } else if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) { | |
10271 | disk_bytenr = em->block_start; | |
10272 | /* | |
10273 | * Bail if the buffer isn't large enough to return the whole | |
10274 | * compressed extent. | |
10275 | */ | |
10276 | if (em->block_len > count) { | |
10277 | ret = -ENOBUFS; | |
10278 | goto out_em; | |
10279 | } | |
c1867eb3 DS |
10280 | disk_io_size = em->block_len; |
10281 | count = em->block_len; | |
1881fba8 OS |
10282 | encoded->unencoded_len = em->ram_bytes; |
10283 | encoded->unencoded_offset = iocb->ki_pos - em->orig_start; | |
10284 | ret = btrfs_encoded_io_compression_from_extent(fs_info, | |
10285 | em->compress_type); | |
10286 | if (ret < 0) | |
10287 | goto out_em; | |
10288 | encoded->compression = ret; | |
10289 | } else { | |
10290 | disk_bytenr = em->block_start + (start - em->start); | |
10291 | if (encoded->len > count) | |
10292 | encoded->len = count; | |
10293 | /* | |
10294 | * Don't read beyond what we locked. This also limits the page | |
10295 | * allocations that we'll do. | |
10296 | */ | |
10297 | disk_io_size = min(lockend + 1, iocb->ki_pos + encoded->len) - start; | |
10298 | count = start + disk_io_size - iocb->ki_pos; | |
10299 | encoded->len = count; | |
10300 | encoded->unencoded_len = count; | |
10301 | disk_io_size = ALIGN(disk_io_size, fs_info->sectorsize); | |
10302 | } | |
10303 | free_extent_map(em); | |
10304 | em = NULL; | |
10305 | ||
10306 | if (disk_bytenr == EXTENT_MAP_HOLE) { | |
570eb97b | 10307 | unlock_extent(io_tree, start, lockend, &cached_state); |
e5d4d75b | 10308 | btrfs_inode_unlock(inode, BTRFS_ILOCK_SHARED); |
1881fba8 OS |
10309 | unlocked = true; |
10310 | ret = iov_iter_zero(count, iter); | |
10311 | if (ret != count) | |
10312 | ret = -EFAULT; | |
10313 | } else { | |
10314 | ret = btrfs_encoded_read_regular(iocb, iter, start, lockend, | |
10315 | &cached_state, disk_bytenr, | |
10316 | disk_io_size, count, | |
10317 | encoded->compression, | |
10318 | &unlocked); | |
10319 | } | |
10320 | ||
10321 | out: | |
10322 | if (ret >= 0) | |
10323 | iocb->ki_pos += encoded->len; | |
10324 | out_em: | |
10325 | free_extent_map(em); | |
10326 | out_unlock_extent: | |
10327 | if (!unlocked) | |
570eb97b | 10328 | unlock_extent(io_tree, start, lockend, &cached_state); |
1881fba8 OS |
10329 | out_unlock_inode: |
10330 | if (!unlocked) | |
e5d4d75b | 10331 | btrfs_inode_unlock(inode, BTRFS_ILOCK_SHARED); |
1881fba8 OS |
10332 | return ret; |
10333 | } | |
10334 | ||
7c0c7269 OS |
10335 | ssize_t btrfs_do_encoded_write(struct kiocb *iocb, struct iov_iter *from, |
10336 | const struct btrfs_ioctl_encoded_io_args *encoded) | |
10337 | { | |
10338 | struct btrfs_inode *inode = BTRFS_I(file_inode(iocb->ki_filp)); | |
10339 | struct btrfs_root *root = inode->root; | |
10340 | struct btrfs_fs_info *fs_info = root->fs_info; | |
10341 | struct extent_io_tree *io_tree = &inode->io_tree; | |
10342 | struct extent_changeset *data_reserved = NULL; | |
10343 | struct extent_state *cached_state = NULL; | |
d611935b | 10344 | struct btrfs_ordered_extent *ordered; |
7c0c7269 OS |
10345 | int compression; |
10346 | size_t orig_count; | |
10347 | u64 start, end; | |
10348 | u64 num_bytes, ram_bytes, disk_num_bytes; | |
10349 | unsigned long nr_pages, i; | |
10350 | struct page **pages; | |
10351 | struct btrfs_key ins; | |
10352 | bool extent_reserved = false; | |
10353 | struct extent_map *em; | |
10354 | ssize_t ret; | |
10355 | ||
10356 | switch (encoded->compression) { | |
10357 | case BTRFS_ENCODED_IO_COMPRESSION_ZLIB: | |
10358 | compression = BTRFS_COMPRESS_ZLIB; | |
10359 | break; | |
10360 | case BTRFS_ENCODED_IO_COMPRESSION_ZSTD: | |
10361 | compression = BTRFS_COMPRESS_ZSTD; | |
10362 | break; | |
10363 | case BTRFS_ENCODED_IO_COMPRESSION_LZO_4K: | |
10364 | case BTRFS_ENCODED_IO_COMPRESSION_LZO_8K: | |
10365 | case BTRFS_ENCODED_IO_COMPRESSION_LZO_16K: | |
10366 | case BTRFS_ENCODED_IO_COMPRESSION_LZO_32K: | |
10367 | case BTRFS_ENCODED_IO_COMPRESSION_LZO_64K: | |
10368 | /* The sector size must match for LZO. */ | |
10369 | if (encoded->compression - | |
10370 | BTRFS_ENCODED_IO_COMPRESSION_LZO_4K + 12 != | |
10371 | fs_info->sectorsize_bits) | |
10372 | return -EINVAL; | |
10373 | compression = BTRFS_COMPRESS_LZO; | |
10374 | break; | |
10375 | default: | |
10376 | return -EINVAL; | |
10377 | } | |
10378 | if (encoded->encryption != BTRFS_ENCODED_IO_ENCRYPTION_NONE) | |
10379 | return -EINVAL; | |
10380 | ||
10381 | orig_count = iov_iter_count(from); | |
10382 | ||
10383 | /* The extent size must be sane. */ | |
10384 | if (encoded->unencoded_len > BTRFS_MAX_UNCOMPRESSED || | |
10385 | orig_count > BTRFS_MAX_COMPRESSED || orig_count == 0) | |
10386 | return -EINVAL; | |
10387 | ||
10388 | /* | |
10389 | * The compressed data must be smaller than the decompressed data. | |
10390 | * | |
10391 | * It's of course possible for data to compress to larger or the same | |
10392 | * size, but the buffered I/O path falls back to no compression for such | |
10393 | * data, and we don't want to break any assumptions by creating these | |
10394 | * extents. | |
10395 | * | |
10396 | * Note that this is less strict than the current check we have that the | |
10397 | * compressed data must be at least one sector smaller than the | |
10398 | * decompressed data. We only want to enforce the weaker requirement | |
10399 | * from old kernels that it is at least one byte smaller. | |
10400 | */ | |
10401 | if (orig_count >= encoded->unencoded_len) | |
10402 | return -EINVAL; | |
10403 | ||
10404 | /* The extent must start on a sector boundary. */ | |
10405 | start = iocb->ki_pos; | |
10406 | if (!IS_ALIGNED(start, fs_info->sectorsize)) | |
10407 | return -EINVAL; | |
10408 | ||
10409 | /* | |
10410 | * The extent must end on a sector boundary. However, we allow a write | |
10411 | * which ends at or extends i_size to have an unaligned length; we round | |
10412 | * up the extent size and set i_size to the unaligned end. | |
10413 | */ | |
10414 | if (start + encoded->len < inode->vfs_inode.i_size && | |
10415 | !IS_ALIGNED(start + encoded->len, fs_info->sectorsize)) | |
10416 | return -EINVAL; | |
10417 | ||
10418 | /* Finally, the offset in the unencoded data must be sector-aligned. */ | |
10419 | if (!IS_ALIGNED(encoded->unencoded_offset, fs_info->sectorsize)) | |
10420 | return -EINVAL; | |
10421 | ||
10422 | num_bytes = ALIGN(encoded->len, fs_info->sectorsize); | |
10423 | ram_bytes = ALIGN(encoded->unencoded_len, fs_info->sectorsize); | |
10424 | end = start + num_bytes - 1; | |
10425 | ||
10426 | /* | |
10427 | * If the extent cannot be inline, the compressed data on disk must be | |
10428 | * sector-aligned. For convenience, we extend it with zeroes if it | |
10429 | * isn't. | |
10430 | */ | |
10431 | disk_num_bytes = ALIGN(orig_count, fs_info->sectorsize); | |
10432 | nr_pages = DIV_ROUND_UP(disk_num_bytes, PAGE_SIZE); | |
10433 | pages = kvcalloc(nr_pages, sizeof(struct page *), GFP_KERNEL_ACCOUNT); | |
10434 | if (!pages) | |
10435 | return -ENOMEM; | |
10436 | for (i = 0; i < nr_pages; i++) { | |
10437 | size_t bytes = min_t(size_t, PAGE_SIZE, iov_iter_count(from)); | |
10438 | char *kaddr; | |
10439 | ||
10440 | pages[i] = alloc_page(GFP_KERNEL_ACCOUNT); | |
10441 | if (!pages[i]) { | |
10442 | ret = -ENOMEM; | |
10443 | goto out_pages; | |
10444 | } | |
70826b6b | 10445 | kaddr = kmap_local_page(pages[i]); |
7c0c7269 | 10446 | if (copy_from_iter(kaddr, bytes, from) != bytes) { |
70826b6b | 10447 | kunmap_local(kaddr); |
7c0c7269 OS |
10448 | ret = -EFAULT; |
10449 | goto out_pages; | |
10450 | } | |
10451 | if (bytes < PAGE_SIZE) | |
10452 | memset(kaddr + bytes, 0, PAGE_SIZE - bytes); | |
70826b6b | 10453 | kunmap_local(kaddr); |
7c0c7269 OS |
10454 | } |
10455 | ||
10456 | for (;;) { | |
10457 | struct btrfs_ordered_extent *ordered; | |
10458 | ||
10459 | ret = btrfs_wait_ordered_range(&inode->vfs_inode, start, num_bytes); | |
10460 | if (ret) | |
10461 | goto out_pages; | |
10462 | ret = invalidate_inode_pages2_range(inode->vfs_inode.i_mapping, | |
10463 | start >> PAGE_SHIFT, | |
10464 | end >> PAGE_SHIFT); | |
10465 | if (ret) | |
10466 | goto out_pages; | |
570eb97b | 10467 | lock_extent(io_tree, start, end, &cached_state); |
7c0c7269 OS |
10468 | ordered = btrfs_lookup_ordered_range(inode, start, num_bytes); |
10469 | if (!ordered && | |
10470 | !filemap_range_has_page(inode->vfs_inode.i_mapping, start, end)) | |
10471 | break; | |
10472 | if (ordered) | |
10473 | btrfs_put_ordered_extent(ordered); | |
570eb97b | 10474 | unlock_extent(io_tree, start, end, &cached_state); |
7c0c7269 OS |
10475 | cond_resched(); |
10476 | } | |
10477 | ||
10478 | /* | |
10479 | * We don't use the higher-level delalloc space functions because our | |
10480 | * num_bytes and disk_num_bytes are different. | |
10481 | */ | |
10482 | ret = btrfs_alloc_data_chunk_ondemand(inode, disk_num_bytes); | |
10483 | if (ret) | |
10484 | goto out_unlock; | |
10485 | ret = btrfs_qgroup_reserve_data(inode, &data_reserved, start, num_bytes); | |
10486 | if (ret) | |
10487 | goto out_free_data_space; | |
d4135134 FM |
10488 | ret = btrfs_delalloc_reserve_metadata(inode, num_bytes, disk_num_bytes, |
10489 | false); | |
7c0c7269 OS |
10490 | if (ret) |
10491 | goto out_qgroup_free_data; | |
10492 | ||
10493 | /* Try an inline extent first. */ | |
10494 | if (start == 0 && encoded->unencoded_len == encoded->len && | |
10495 | encoded->unencoded_offset == 0) { | |
10496 | ret = cow_file_range_inline(inode, encoded->len, orig_count, | |
10497 | compression, pages, true); | |
10498 | if (ret <= 0) { | |
10499 | if (ret == 0) | |
10500 | ret = orig_count; | |
10501 | goto out_delalloc_release; | |
10502 | } | |
10503 | } | |
10504 | ||
10505 | ret = btrfs_reserve_extent(root, disk_num_bytes, disk_num_bytes, | |
10506 | disk_num_bytes, 0, 0, &ins, 1, 1); | |
10507 | if (ret) | |
10508 | goto out_delalloc_release; | |
10509 | extent_reserved = true; | |
10510 | ||
10511 | em = create_io_em(inode, start, num_bytes, | |
10512 | start - encoded->unencoded_offset, ins.objectid, | |
10513 | ins.offset, ins.offset, ram_bytes, compression, | |
10514 | BTRFS_ORDERED_COMPRESSED); | |
10515 | if (IS_ERR(em)) { | |
10516 | ret = PTR_ERR(em); | |
10517 | goto out_free_reserved; | |
10518 | } | |
10519 | free_extent_map(em); | |
10520 | ||
d611935b | 10521 | ordered = btrfs_alloc_ordered_extent(inode, start, num_bytes, ram_bytes, |
7c0c7269 OS |
10522 | ins.objectid, ins.offset, |
10523 | encoded->unencoded_offset, | |
10524 | (1 << BTRFS_ORDERED_ENCODED) | | |
10525 | (1 << BTRFS_ORDERED_COMPRESSED), | |
10526 | compression); | |
d611935b | 10527 | if (IS_ERR(ordered)) { |
4c0c8cfc | 10528 | btrfs_drop_extent_map_range(inode, start, end, false); |
d611935b | 10529 | ret = PTR_ERR(ordered); |
7c0c7269 OS |
10530 | goto out_free_reserved; |
10531 | } | |
10532 | btrfs_dec_block_group_reservations(fs_info, ins.objectid); | |
10533 | ||
10534 | if (start + encoded->len > inode->vfs_inode.i_size) | |
10535 | i_size_write(&inode->vfs_inode, start + encoded->len); | |
10536 | ||
570eb97b | 10537 | unlock_extent(io_tree, start, end, &cached_state); |
7c0c7269 OS |
10538 | |
10539 | btrfs_delalloc_release_extents(inode, num_bytes); | |
10540 | ||
d611935b | 10541 | btrfs_submit_compressed_write(ordered, pages, nr_pages, 0, false); |
7c0c7269 OS |
10542 | ret = orig_count; |
10543 | goto out; | |
10544 | ||
10545 | out_free_reserved: | |
10546 | btrfs_dec_block_group_reservations(fs_info, ins.objectid); | |
10547 | btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1); | |
10548 | out_delalloc_release: | |
10549 | btrfs_delalloc_release_extents(inode, num_bytes); | |
10550 | btrfs_delalloc_release_metadata(inode, disk_num_bytes, ret < 0); | |
10551 | out_qgroup_free_data: | |
10552 | if (ret < 0) | |
10553 | btrfs_qgroup_free_data(inode, data_reserved, start, num_bytes); | |
10554 | out_free_data_space: | |
10555 | /* | |
10556 | * If btrfs_reserve_extent() succeeded, then we already decremented | |
10557 | * bytes_may_use. | |
10558 | */ | |
10559 | if (!extent_reserved) | |
10560 | btrfs_free_reserved_data_space_noquota(fs_info, disk_num_bytes); | |
10561 | out_unlock: | |
570eb97b | 10562 | unlock_extent(io_tree, start, end, &cached_state); |
7c0c7269 OS |
10563 | out_pages: |
10564 | for (i = 0; i < nr_pages; i++) { | |
10565 | if (pages[i]) | |
10566 | __free_page(pages[i]); | |
10567 | } | |
10568 | kvfree(pages); | |
10569 | out: | |
10570 | if (ret >= 0) | |
10571 | iocb->ki_pos += encoded->len; | |
10572 | return ret; | |
10573 | } | |
10574 | ||
ed46ff3d OS |
10575 | #ifdef CONFIG_SWAP |
10576 | /* | |
10577 | * Add an entry indicating a block group or device which is pinned by a | |
10578 | * swapfile. Returns 0 on success, 1 if there is already an entry for it, or a | |
10579 | * negative errno on failure. | |
10580 | */ | |
10581 | static int btrfs_add_swapfile_pin(struct inode *inode, void *ptr, | |
10582 | bool is_block_group) | |
10583 | { | |
10584 | struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info; | |
10585 | struct btrfs_swapfile_pin *sp, *entry; | |
10586 | struct rb_node **p; | |
10587 | struct rb_node *parent = NULL; | |
10588 | ||
10589 | sp = kmalloc(sizeof(*sp), GFP_NOFS); | |
10590 | if (!sp) | |
10591 | return -ENOMEM; | |
10592 | sp->ptr = ptr; | |
10593 | sp->inode = inode; | |
10594 | sp->is_block_group = is_block_group; | |
195a49ea | 10595 | sp->bg_extent_count = 1; |
ed46ff3d OS |
10596 | |
10597 | spin_lock(&fs_info->swapfile_pins_lock); | |
10598 | p = &fs_info->swapfile_pins.rb_node; | |
10599 | while (*p) { | |
10600 | parent = *p; | |
10601 | entry = rb_entry(parent, struct btrfs_swapfile_pin, node); | |
10602 | if (sp->ptr < entry->ptr || | |
10603 | (sp->ptr == entry->ptr && sp->inode < entry->inode)) { | |
10604 | p = &(*p)->rb_left; | |
10605 | } else if (sp->ptr > entry->ptr || | |
10606 | (sp->ptr == entry->ptr && sp->inode > entry->inode)) { | |
10607 | p = &(*p)->rb_right; | |
10608 | } else { | |
195a49ea FM |
10609 | if (is_block_group) |
10610 | entry->bg_extent_count++; | |
ed46ff3d OS |
10611 | spin_unlock(&fs_info->swapfile_pins_lock); |
10612 | kfree(sp); | |
10613 | return 1; | |
10614 | } | |
10615 | } | |
10616 | rb_link_node(&sp->node, parent, p); | |
10617 | rb_insert_color(&sp->node, &fs_info->swapfile_pins); | |
10618 | spin_unlock(&fs_info->swapfile_pins_lock); | |
10619 | return 0; | |
10620 | } | |
10621 | ||
10622 | /* Free all of the entries pinned by this swapfile. */ | |
10623 | static void btrfs_free_swapfile_pins(struct inode *inode) | |
10624 | { | |
10625 | struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info; | |
10626 | struct btrfs_swapfile_pin *sp; | |
10627 | struct rb_node *node, *next; | |
10628 | ||
10629 | spin_lock(&fs_info->swapfile_pins_lock); | |
10630 | node = rb_first(&fs_info->swapfile_pins); | |
10631 | while (node) { | |
10632 | next = rb_next(node); | |
10633 | sp = rb_entry(node, struct btrfs_swapfile_pin, node); | |
10634 | if (sp->inode == inode) { | |
10635 | rb_erase(&sp->node, &fs_info->swapfile_pins); | |
195a49ea FM |
10636 | if (sp->is_block_group) { |
10637 | btrfs_dec_block_group_swap_extents(sp->ptr, | |
10638 | sp->bg_extent_count); | |
ed46ff3d | 10639 | btrfs_put_block_group(sp->ptr); |
195a49ea | 10640 | } |
ed46ff3d OS |
10641 | kfree(sp); |
10642 | } | |
10643 | node = next; | |
10644 | } | |
10645 | spin_unlock(&fs_info->swapfile_pins_lock); | |
10646 | } | |
10647 | ||
10648 | struct btrfs_swap_info { | |
10649 | u64 start; | |
10650 | u64 block_start; | |
10651 | u64 block_len; | |
10652 | u64 lowest_ppage; | |
10653 | u64 highest_ppage; | |
10654 | unsigned long nr_pages; | |
10655 | int nr_extents; | |
10656 | }; | |
10657 | ||
10658 | static int btrfs_add_swap_extent(struct swap_info_struct *sis, | |
10659 | struct btrfs_swap_info *bsi) | |
10660 | { | |
10661 | unsigned long nr_pages; | |
c2f82263 | 10662 | unsigned long max_pages; |
ed46ff3d OS |
10663 | u64 first_ppage, first_ppage_reported, next_ppage; |
10664 | int ret; | |
10665 | ||
c2f82263 FM |
10666 | /* |
10667 | * Our swapfile may have had its size extended after the swap header was | |
10668 | * written. In that case activating the swapfile should not go beyond | |
10669 | * the max size set in the swap header. | |
10670 | */ | |
10671 | if (bsi->nr_pages >= sis->max) | |
10672 | return 0; | |
10673 | ||
10674 | max_pages = sis->max - bsi->nr_pages; | |
ce394a7f YZ |
10675 | first_ppage = PAGE_ALIGN(bsi->block_start) >> PAGE_SHIFT; |
10676 | next_ppage = PAGE_ALIGN_DOWN(bsi->block_start + bsi->block_len) >> PAGE_SHIFT; | |
ed46ff3d OS |
10677 | |
10678 | if (first_ppage >= next_ppage) | |
10679 | return 0; | |
10680 | nr_pages = next_ppage - first_ppage; | |
c2f82263 | 10681 | nr_pages = min(nr_pages, max_pages); |
ed46ff3d OS |
10682 | |
10683 | first_ppage_reported = first_ppage; | |
10684 | if (bsi->start == 0) | |
10685 | first_ppage_reported++; | |
10686 | if (bsi->lowest_ppage > first_ppage_reported) | |
10687 | bsi->lowest_ppage = first_ppage_reported; | |
10688 | if (bsi->highest_ppage < (next_ppage - 1)) | |
10689 | bsi->highest_ppage = next_ppage - 1; | |
10690 | ||
10691 | ret = add_swap_extent(sis, bsi->nr_pages, nr_pages, first_ppage); | |
10692 | if (ret < 0) | |
10693 | return ret; | |
10694 | bsi->nr_extents += ret; | |
10695 | bsi->nr_pages += nr_pages; | |
10696 | return 0; | |
10697 | } | |
10698 | ||
10699 | static void btrfs_swap_deactivate(struct file *file) | |
10700 | { | |
10701 | struct inode *inode = file_inode(file); | |
10702 | ||
10703 | btrfs_free_swapfile_pins(inode); | |
10704 | atomic_dec(&BTRFS_I(inode)->root->nr_swapfiles); | |
10705 | } | |
10706 | ||
10707 | static int btrfs_swap_activate(struct swap_info_struct *sis, struct file *file, | |
10708 | sector_t *span) | |
10709 | { | |
10710 | struct inode *inode = file_inode(file); | |
dd0734f2 FM |
10711 | struct btrfs_root *root = BTRFS_I(inode)->root; |
10712 | struct btrfs_fs_info *fs_info = root->fs_info; | |
ed46ff3d OS |
10713 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
10714 | struct extent_state *cached_state = NULL; | |
10715 | struct extent_map *em = NULL; | |
10716 | struct btrfs_device *device = NULL; | |
10717 | struct btrfs_swap_info bsi = { | |
10718 | .lowest_ppage = (sector_t)-1ULL, | |
10719 | }; | |
10720 | int ret = 0; | |
10721 | u64 isize; | |
10722 | u64 start; | |
10723 | ||
10724 | /* | |
10725 | * If the swap file was just created, make sure delalloc is done. If the | |
10726 | * file changes again after this, the user is doing something stupid and | |
10727 | * we don't really care. | |
10728 | */ | |
10729 | ret = btrfs_wait_ordered_range(inode, 0, (u64)-1); | |
10730 | if (ret) | |
10731 | return ret; | |
10732 | ||
10733 | /* | |
10734 | * The inode is locked, so these flags won't change after we check them. | |
10735 | */ | |
10736 | if (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS) { | |
10737 | btrfs_warn(fs_info, "swapfile must not be compressed"); | |
10738 | return -EINVAL; | |
10739 | } | |
10740 | if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW)) { | |
10741 | btrfs_warn(fs_info, "swapfile must not be copy-on-write"); | |
10742 | return -EINVAL; | |
10743 | } | |
10744 | if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) { | |
10745 | btrfs_warn(fs_info, "swapfile must not be checksummed"); | |
10746 | return -EINVAL; | |
10747 | } | |
10748 | ||
10749 | /* | |
10750 | * Balance or device remove/replace/resize can move stuff around from | |
c3e1f96c GR |
10751 | * under us. The exclop protection makes sure they aren't running/won't |
10752 | * run concurrently while we are mapping the swap extents, and | |
10753 | * fs_info->swapfile_pins prevents them from running while the swap | |
10754 | * file is active and moving the extents. Note that this also prevents | |
10755 | * a concurrent device add which isn't actually necessary, but it's not | |
ed46ff3d OS |
10756 | * really worth the trouble to allow it. |
10757 | */ | |
c3e1f96c | 10758 | if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_SWAP_ACTIVATE)) { |
ed46ff3d OS |
10759 | btrfs_warn(fs_info, |
10760 | "cannot activate swapfile while exclusive operation is running"); | |
10761 | return -EBUSY; | |
10762 | } | |
dd0734f2 FM |
10763 | |
10764 | /* | |
10765 | * Prevent snapshot creation while we are activating the swap file. | |
10766 | * We do not want to race with snapshot creation. If snapshot creation | |
10767 | * already started before we bumped nr_swapfiles from 0 to 1 and | |
10768 | * completes before the first write into the swap file after it is | |
10769 | * activated, than that write would fallback to COW. | |
10770 | */ | |
10771 | if (!btrfs_drew_try_write_lock(&root->snapshot_lock)) { | |
10772 | btrfs_exclop_finish(fs_info); | |
10773 | btrfs_warn(fs_info, | |
10774 | "cannot activate swapfile because snapshot creation is in progress"); | |
10775 | return -EINVAL; | |
10776 | } | |
ed46ff3d OS |
10777 | /* |
10778 | * Snapshots can create extents which require COW even if NODATACOW is | |
10779 | * set. We use this counter to prevent snapshots. We must increment it | |
10780 | * before walking the extents because we don't want a concurrent | |
10781 | * snapshot to run after we've already checked the extents. | |
60021bd7 KH |
10782 | * |
10783 | * It is possible that subvolume is marked for deletion but still not | |
10784 | * removed yet. To prevent this race, we check the root status before | |
10785 | * activating the swapfile. | |
ed46ff3d | 10786 | */ |
60021bd7 KH |
10787 | spin_lock(&root->root_item_lock); |
10788 | if (btrfs_root_dead(root)) { | |
10789 | spin_unlock(&root->root_item_lock); | |
10790 | ||
10791 | btrfs_exclop_finish(fs_info); | |
10792 | btrfs_warn(fs_info, | |
10793 | "cannot activate swapfile because subvolume %llu is being deleted", | |
10794 | root->root_key.objectid); | |
10795 | return -EPERM; | |
10796 | } | |
dd0734f2 | 10797 | atomic_inc(&root->nr_swapfiles); |
60021bd7 | 10798 | spin_unlock(&root->root_item_lock); |
ed46ff3d OS |
10799 | |
10800 | isize = ALIGN_DOWN(inode->i_size, fs_info->sectorsize); | |
10801 | ||
570eb97b | 10802 | lock_extent(io_tree, 0, isize - 1, &cached_state); |
ed46ff3d OS |
10803 | start = 0; |
10804 | while (start < isize) { | |
10805 | u64 logical_block_start, physical_block_start; | |
32da5386 | 10806 | struct btrfs_block_group *bg; |
ed46ff3d OS |
10807 | u64 len = isize - start; |
10808 | ||
39b07b5d | 10809 | em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len); |
ed46ff3d OS |
10810 | if (IS_ERR(em)) { |
10811 | ret = PTR_ERR(em); | |
10812 | goto out; | |
10813 | } | |
10814 | ||
10815 | if (em->block_start == EXTENT_MAP_HOLE) { | |
10816 | btrfs_warn(fs_info, "swapfile must not have holes"); | |
10817 | ret = -EINVAL; | |
10818 | goto out; | |
10819 | } | |
10820 | if (em->block_start == EXTENT_MAP_INLINE) { | |
10821 | /* | |
10822 | * It's unlikely we'll ever actually find ourselves | |
10823 | * here, as a file small enough to fit inline won't be | |
10824 | * big enough to store more than the swap header, but in | |
10825 | * case something changes in the future, let's catch it | |
10826 | * here rather than later. | |
10827 | */ | |
10828 | btrfs_warn(fs_info, "swapfile must not be inline"); | |
10829 | ret = -EINVAL; | |
10830 | goto out; | |
10831 | } | |
10832 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) { | |
10833 | btrfs_warn(fs_info, "swapfile must not be compressed"); | |
10834 | ret = -EINVAL; | |
10835 | goto out; | |
10836 | } | |
10837 | ||
10838 | logical_block_start = em->block_start + (start - em->start); | |
10839 | len = min(len, em->len - (start - em->start)); | |
10840 | free_extent_map(em); | |
10841 | em = NULL; | |
10842 | ||
26ce9114 | 10843 | ret = can_nocow_extent(inode, start, &len, NULL, NULL, NULL, false, true); |
ed46ff3d OS |
10844 | if (ret < 0) { |
10845 | goto out; | |
10846 | } else if (ret) { | |
10847 | ret = 0; | |
10848 | } else { | |
10849 | btrfs_warn(fs_info, | |
10850 | "swapfile must not be copy-on-write"); | |
10851 | ret = -EINVAL; | |
10852 | goto out; | |
10853 | } | |
10854 | ||
10855 | em = btrfs_get_chunk_map(fs_info, logical_block_start, len); | |
10856 | if (IS_ERR(em)) { | |
10857 | ret = PTR_ERR(em); | |
10858 | goto out; | |
10859 | } | |
10860 | ||
10861 | if (em->map_lookup->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) { | |
10862 | btrfs_warn(fs_info, | |
10863 | "swapfile must have single data profile"); | |
10864 | ret = -EINVAL; | |
10865 | goto out; | |
10866 | } | |
10867 | ||
10868 | if (device == NULL) { | |
10869 | device = em->map_lookup->stripes[0].dev; | |
10870 | ret = btrfs_add_swapfile_pin(inode, device, false); | |
10871 | if (ret == 1) | |
10872 | ret = 0; | |
10873 | else if (ret) | |
10874 | goto out; | |
10875 | } else if (device != em->map_lookup->stripes[0].dev) { | |
10876 | btrfs_warn(fs_info, "swapfile must be on one device"); | |
10877 | ret = -EINVAL; | |
10878 | goto out; | |
10879 | } | |
10880 | ||
10881 | physical_block_start = (em->map_lookup->stripes[0].physical + | |
10882 | (logical_block_start - em->start)); | |
10883 | len = min(len, em->len - (logical_block_start - em->start)); | |
10884 | free_extent_map(em); | |
10885 | em = NULL; | |
10886 | ||
10887 | bg = btrfs_lookup_block_group(fs_info, logical_block_start); | |
10888 | if (!bg) { | |
10889 | btrfs_warn(fs_info, | |
10890 | "could not find block group containing swapfile"); | |
10891 | ret = -EINVAL; | |
10892 | goto out; | |
10893 | } | |
10894 | ||
195a49ea FM |
10895 | if (!btrfs_inc_block_group_swap_extents(bg)) { |
10896 | btrfs_warn(fs_info, | |
10897 | "block group for swapfile at %llu is read-only%s", | |
10898 | bg->start, | |
10899 | atomic_read(&fs_info->scrubs_running) ? | |
10900 | " (scrub running)" : ""); | |
10901 | btrfs_put_block_group(bg); | |
10902 | ret = -EINVAL; | |
10903 | goto out; | |
10904 | } | |
10905 | ||
ed46ff3d OS |
10906 | ret = btrfs_add_swapfile_pin(inode, bg, true); |
10907 | if (ret) { | |
10908 | btrfs_put_block_group(bg); | |
10909 | if (ret == 1) | |
10910 | ret = 0; | |
10911 | else | |
10912 | goto out; | |
10913 | } | |
10914 | ||
10915 | if (bsi.block_len && | |
10916 | bsi.block_start + bsi.block_len == physical_block_start) { | |
10917 | bsi.block_len += len; | |
10918 | } else { | |
10919 | if (bsi.block_len) { | |
10920 | ret = btrfs_add_swap_extent(sis, &bsi); | |
10921 | if (ret) | |
10922 | goto out; | |
10923 | } | |
10924 | bsi.start = start; | |
10925 | bsi.block_start = physical_block_start; | |
10926 | bsi.block_len = len; | |
10927 | } | |
10928 | ||
10929 | start += len; | |
10930 | } | |
10931 | ||
10932 | if (bsi.block_len) | |
10933 | ret = btrfs_add_swap_extent(sis, &bsi); | |
10934 | ||
10935 | out: | |
10936 | if (!IS_ERR_OR_NULL(em)) | |
10937 | free_extent_map(em); | |
10938 | ||
570eb97b | 10939 | unlock_extent(io_tree, 0, isize - 1, &cached_state); |
ed46ff3d OS |
10940 | |
10941 | if (ret) | |
10942 | btrfs_swap_deactivate(file); | |
10943 | ||
dd0734f2 FM |
10944 | btrfs_drew_write_unlock(&root->snapshot_lock); |
10945 | ||
c3e1f96c | 10946 | btrfs_exclop_finish(fs_info); |
ed46ff3d OS |
10947 | |
10948 | if (ret) | |
10949 | return ret; | |
10950 | ||
10951 | if (device) | |
10952 | sis->bdev = device->bdev; | |
10953 | *span = bsi.highest_ppage - bsi.lowest_ppage + 1; | |
10954 | sis->max = bsi.nr_pages; | |
10955 | sis->pages = bsi.nr_pages - 1; | |
10956 | sis->highest_bit = bsi.nr_pages - 1; | |
10957 | return bsi.nr_extents; | |
10958 | } | |
10959 | #else | |
10960 | static void btrfs_swap_deactivate(struct file *file) | |
10961 | { | |
10962 | } | |
10963 | ||
10964 | static int btrfs_swap_activate(struct swap_info_struct *sis, struct file *file, | |
10965 | sector_t *span) | |
10966 | { | |
10967 | return -EOPNOTSUPP; | |
10968 | } | |
10969 | #endif | |
10970 | ||
2766ff61 FM |
10971 | /* |
10972 | * Update the number of bytes used in the VFS' inode. When we replace extents in | |
10973 | * a range (clone, dedupe, fallocate's zero range), we must update the number of | |
10974 | * bytes used by the inode in an atomic manner, so that concurrent stat(2) calls | |
10975 | * always get a correct value. | |
10976 | */ | |
10977 | void btrfs_update_inode_bytes(struct btrfs_inode *inode, | |
10978 | const u64 add_bytes, | |
10979 | const u64 del_bytes) | |
10980 | { | |
10981 | if (add_bytes == del_bytes) | |
10982 | return; | |
10983 | ||
10984 | spin_lock(&inode->lock); | |
10985 | if (del_bytes > 0) | |
10986 | inode_sub_bytes(&inode->vfs_inode, del_bytes); | |
10987 | if (add_bytes > 0) | |
10988 | inode_add_bytes(&inode->vfs_inode, add_bytes); | |
10989 | spin_unlock(&inode->lock); | |
10990 | } | |
10991 | ||
43dd529a | 10992 | /* |
63c34cb4 FM |
10993 | * Verify that there are no ordered extents for a given file range. |
10994 | * | |
10995 | * @inode: The target inode. | |
10996 | * @start: Start offset of the file range, should be sector size aligned. | |
10997 | * @end: End offset (inclusive) of the file range, its value +1 should be | |
10998 | * sector size aligned. | |
10999 | * | |
11000 | * This should typically be used for cases where we locked an inode's VFS lock in | |
11001 | * exclusive mode, we have also locked the inode's i_mmap_lock in exclusive mode, | |
11002 | * we have flushed all delalloc in the range, we have waited for all ordered | |
11003 | * extents in the range to complete and finally we have locked the file range in | |
11004 | * the inode's io_tree. | |
11005 | */ | |
11006 | void btrfs_assert_inode_range_clean(struct btrfs_inode *inode, u64 start, u64 end) | |
11007 | { | |
11008 | struct btrfs_root *root = inode->root; | |
11009 | struct btrfs_ordered_extent *ordered; | |
11010 | ||
11011 | if (!IS_ENABLED(CONFIG_BTRFS_ASSERT)) | |
11012 | return; | |
11013 | ||
11014 | ordered = btrfs_lookup_first_ordered_range(inode, start, end + 1 - start); | |
11015 | if (ordered) { | |
11016 | btrfs_err(root->fs_info, | |
11017 | "found unexpected ordered extent in file range [%llu, %llu] for inode %llu root %llu (ordered range [%llu, %llu])", | |
11018 | start, end, btrfs_ino(inode), root->root_key.objectid, | |
11019 | ordered->file_offset, | |
11020 | ordered->file_offset + ordered->num_bytes - 1); | |
11021 | btrfs_put_ordered_extent(ordered); | |
11022 | } | |
11023 | ||
11024 | ASSERT(ordered == NULL); | |
11025 | } | |
11026 | ||
6e1d5dcc | 11027 | static const struct inode_operations btrfs_dir_inode_operations = { |
3394e160 | 11028 | .getattr = btrfs_getattr, |
39279cc3 CM |
11029 | .lookup = btrfs_lookup, |
11030 | .create = btrfs_create, | |
11031 | .unlink = btrfs_unlink, | |
11032 | .link = btrfs_link, | |
11033 | .mkdir = btrfs_mkdir, | |
11034 | .rmdir = btrfs_rmdir, | |
2773bf00 | 11035 | .rename = btrfs_rename2, |
39279cc3 CM |
11036 | .symlink = btrfs_symlink, |
11037 | .setattr = btrfs_setattr, | |
618e21d5 | 11038 | .mknod = btrfs_mknod, |
5103e947 | 11039 | .listxattr = btrfs_listxattr, |
fdebe2bd | 11040 | .permission = btrfs_permission, |
cac2f8b8 | 11041 | .get_inode_acl = btrfs_get_acl, |
996a710d | 11042 | .set_acl = btrfs_set_acl, |
93fd63c2 | 11043 | .update_time = btrfs_update_time, |
ef3b9af5 | 11044 | .tmpfile = btrfs_tmpfile, |
97fc2977 MS |
11045 | .fileattr_get = btrfs_fileattr_get, |
11046 | .fileattr_set = btrfs_fileattr_set, | |
39279cc3 | 11047 | }; |
76dda93c | 11048 | |
828c0950 | 11049 | static const struct file_operations btrfs_dir_file_operations = { |
39279cc3 CM |
11050 | .llseek = generic_file_llseek, |
11051 | .read = generic_read_dir, | |
02dbfc99 | 11052 | .iterate_shared = btrfs_real_readdir, |
23b5ec74 | 11053 | .open = btrfs_opendir, |
34287aa3 | 11054 | .unlocked_ioctl = btrfs_ioctl, |
39279cc3 | 11055 | #ifdef CONFIG_COMPAT |
4c63c245 | 11056 | .compat_ioctl = btrfs_compat_ioctl, |
39279cc3 | 11057 | #endif |
6bf13c0c | 11058 | .release = btrfs_release_file, |
e02119d5 | 11059 | .fsync = btrfs_sync_file, |
39279cc3 CM |
11060 | }; |
11061 | ||
35054394 CM |
11062 | /* |
11063 | * btrfs doesn't support the bmap operation because swapfiles | |
11064 | * use bmap to make a mapping of extents in the file. They assume | |
11065 | * these extents won't change over the life of the file and they | |
11066 | * use the bmap result to do IO directly to the drive. | |
11067 | * | |
11068 | * the btrfs bmap call would return logical addresses that aren't | |
11069 | * suitable for IO and they also will change frequently as COW | |
11070 | * operations happen. So, swapfile + btrfs == corruption. | |
11071 | * | |
11072 | * For now we're avoiding this by dropping bmap. | |
11073 | */ | |
7f09410b | 11074 | static const struct address_space_operations btrfs_aops = { |
fb12489b | 11075 | .read_folio = btrfs_read_folio, |
b293f02e | 11076 | .writepages = btrfs_writepages, |
ba206a02 | 11077 | .readahead = btrfs_readahead, |
895586eb | 11078 | .invalidate_folio = btrfs_invalidate_folio, |
f913cff3 | 11079 | .release_folio = btrfs_release_folio, |
e7a60a17 | 11080 | .migrate_folio = btrfs_migrate_folio, |
187c82cb | 11081 | .dirty_folio = filemap_dirty_folio, |
465fdd97 | 11082 | .error_remove_page = generic_error_remove_page, |
ed46ff3d OS |
11083 | .swap_activate = btrfs_swap_activate, |
11084 | .swap_deactivate = btrfs_swap_deactivate, | |
39279cc3 CM |
11085 | }; |
11086 | ||
6e1d5dcc | 11087 | static const struct inode_operations btrfs_file_inode_operations = { |
39279cc3 CM |
11088 | .getattr = btrfs_getattr, |
11089 | .setattr = btrfs_setattr, | |
5103e947 | 11090 | .listxattr = btrfs_listxattr, |
fdebe2bd | 11091 | .permission = btrfs_permission, |
1506fcc8 | 11092 | .fiemap = btrfs_fiemap, |
cac2f8b8 | 11093 | .get_inode_acl = btrfs_get_acl, |
996a710d | 11094 | .set_acl = btrfs_set_acl, |
e41f941a | 11095 | .update_time = btrfs_update_time, |
97fc2977 MS |
11096 | .fileattr_get = btrfs_fileattr_get, |
11097 | .fileattr_set = btrfs_fileattr_set, | |
39279cc3 | 11098 | }; |
6e1d5dcc | 11099 | static const struct inode_operations btrfs_special_inode_operations = { |
618e21d5 JB |
11100 | .getattr = btrfs_getattr, |
11101 | .setattr = btrfs_setattr, | |
fdebe2bd | 11102 | .permission = btrfs_permission, |
33268eaf | 11103 | .listxattr = btrfs_listxattr, |
cac2f8b8 | 11104 | .get_inode_acl = btrfs_get_acl, |
996a710d | 11105 | .set_acl = btrfs_set_acl, |
e41f941a | 11106 | .update_time = btrfs_update_time, |
618e21d5 | 11107 | }; |
6e1d5dcc | 11108 | static const struct inode_operations btrfs_symlink_inode_operations = { |
6b255391 | 11109 | .get_link = page_get_link, |
f209561a | 11110 | .getattr = btrfs_getattr, |
22c44fe6 | 11111 | .setattr = btrfs_setattr, |
fdebe2bd | 11112 | .permission = btrfs_permission, |
0279b4cd | 11113 | .listxattr = btrfs_listxattr, |
e41f941a | 11114 | .update_time = btrfs_update_time, |
39279cc3 | 11115 | }; |
76dda93c | 11116 | |
82d339d9 | 11117 | const struct dentry_operations btrfs_dentry_operations = { |
76dda93c YZ |
11118 | .d_delete = btrfs_dentry_delete, |
11119 | }; |