Commit | Line | Data |
---|---|---|
c1d7c514 | 1 | // SPDX-License-Identifier: GPL-2.0 |
e02119d5 CM |
2 | /* |
3 | * Copyright (C) 2008 Oracle. All rights reserved. | |
e02119d5 CM |
4 | */ |
5 | ||
6 | #include <linux/sched.h> | |
5a0e3ad6 | 7 | #include <linux/slab.h> |
c6adc9cc | 8 | #include <linux/blkdev.h> |
5dc562c5 | 9 | #include <linux/list_sort.h> |
c7f88c4e | 10 | #include <linux/iversion.h> |
602cbe91 | 11 | #include "misc.h" |
9678c543 | 12 | #include "ctree.h" |
995946dd | 13 | #include "tree-log.h" |
e02119d5 CM |
14 | #include "disk-io.h" |
15 | #include "locking.h" | |
16 | #include "print-tree.h" | |
f186373f | 17 | #include "backref.h" |
ebb8765b | 18 | #include "compression.h" |
df2c95f3 | 19 | #include "qgroup.h" |
900c9981 | 20 | #include "inode-map.h" |
6787bb9f NB |
21 | #include "block-group.h" |
22 | #include "space-info.h" | |
e02119d5 CM |
23 | |
24 | /* magic values for the inode_only field in btrfs_log_inode: | |
25 | * | |
26 | * LOG_INODE_ALL means to log everything | |
27 | * LOG_INODE_EXISTS means to log just enough to recreate the inode | |
28 | * during log replay | |
29 | */ | |
e13976cf DS |
30 | enum { |
31 | LOG_INODE_ALL, | |
32 | LOG_INODE_EXISTS, | |
33 | LOG_OTHER_INODE, | |
34 | LOG_OTHER_INODE_ALL, | |
35 | }; | |
e02119d5 | 36 | |
12fcfd22 CM |
37 | /* |
38 | * directory trouble cases | |
39 | * | |
40 | * 1) on rename or unlink, if the inode being unlinked isn't in the fsync | |
41 | * log, we must force a full commit before doing an fsync of the directory | |
42 | * where the unlink was done. | |
43 | * ---> record transid of last unlink/rename per directory | |
44 | * | |
45 | * mkdir foo/some_dir | |
46 | * normal commit | |
47 | * rename foo/some_dir foo2/some_dir | |
48 | * mkdir foo/some_dir | |
49 | * fsync foo/some_dir/some_file | |
50 | * | |
51 | * The fsync above will unlink the original some_dir without recording | |
52 | * it in its new location (foo2). After a crash, some_dir will be gone | |
53 | * unless the fsync of some_file forces a full commit | |
54 | * | |
55 | * 2) we must log any new names for any file or dir that is in the fsync | |
56 | * log. ---> check inode while renaming/linking. | |
57 | * | |
58 | * 2a) we must log any new names for any file or dir during rename | |
59 | * when the directory they are being removed from was logged. | |
60 | * ---> check inode and old parent dir during rename | |
61 | * | |
62 | * 2a is actually the more important variant. With the extra logging | |
63 | * a crash might unlink the old name without recreating the new one | |
64 | * | |
65 | * 3) after a crash, we must go through any directories with a link count | |
66 | * of zero and redo the rm -rf | |
67 | * | |
68 | * mkdir f1/foo | |
69 | * normal commit | |
70 | * rm -rf f1/foo | |
71 | * fsync(f1) | |
72 | * | |
73 | * The directory f1 was fully removed from the FS, but fsync was never | |
74 | * called on f1, only its parent dir. After a crash the rm -rf must | |
75 | * be replayed. This must be able to recurse down the entire | |
76 | * directory tree. The inode link count fixup code takes care of the | |
77 | * ugly details. | |
78 | */ | |
79 | ||
e02119d5 CM |
80 | /* |
81 | * stages for the tree walking. The first | |
82 | * stage (0) is to only pin down the blocks we find | |
83 | * the second stage (1) is to make sure that all the inodes | |
84 | * we find in the log are created in the subvolume. | |
85 | * | |
86 | * The last stage is to deal with directories and links and extents | |
87 | * and all the other fun semantics | |
88 | */ | |
e13976cf DS |
89 | enum { |
90 | LOG_WALK_PIN_ONLY, | |
91 | LOG_WALK_REPLAY_INODES, | |
92 | LOG_WALK_REPLAY_DIR_INDEX, | |
93 | LOG_WALK_REPLAY_ALL, | |
94 | }; | |
e02119d5 | 95 | |
12fcfd22 | 96 | static int btrfs_log_inode(struct btrfs_trans_handle *trans, |
a59108a7 | 97 | struct btrfs_root *root, struct btrfs_inode *inode, |
49dae1bc | 98 | int inode_only, |
7af59743 FM |
99 | const loff_t start, |
100 | const loff_t end, | |
8407f553 | 101 | struct btrfs_log_ctx *ctx); |
ec051c0f YZ |
102 | static int link_to_fixup_dir(struct btrfs_trans_handle *trans, |
103 | struct btrfs_root *root, | |
104 | struct btrfs_path *path, u64 objectid); | |
12fcfd22 CM |
105 | static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans, |
106 | struct btrfs_root *root, | |
107 | struct btrfs_root *log, | |
108 | struct btrfs_path *path, | |
109 | u64 dirid, int del_all); | |
e02119d5 CM |
110 | |
111 | /* | |
112 | * tree logging is a special write ahead log used to make sure that | |
113 | * fsyncs and O_SYNCs can happen without doing full tree commits. | |
114 | * | |
115 | * Full tree commits are expensive because they require commonly | |
116 | * modified blocks to be recowed, creating many dirty pages in the | |
117 | * extent tree an 4x-6x higher write load than ext3. | |
118 | * | |
119 | * Instead of doing a tree commit on every fsync, we use the | |
120 | * key ranges and transaction ids to find items for a given file or directory | |
121 | * that have changed in this transaction. Those items are copied into | |
122 | * a special tree (one per subvolume root), that tree is written to disk | |
123 | * and then the fsync is considered complete. | |
124 | * | |
125 | * After a crash, items are copied out of the log-tree back into the | |
126 | * subvolume tree. Any file data extents found are recorded in the extent | |
127 | * allocation tree, and the log-tree freed. | |
128 | * | |
129 | * The log tree is read three times, once to pin down all the extents it is | |
130 | * using in ram and once, once to create all the inodes logged in the tree | |
131 | * and once to do all the other items. | |
132 | */ | |
133 | ||
e02119d5 CM |
134 | /* |
135 | * start a sub transaction and setup the log tree | |
136 | * this increments the log tree writer count to make the people | |
137 | * syncing the tree wait for us to finish | |
138 | */ | |
139 | static int start_log_trans(struct btrfs_trans_handle *trans, | |
8b050d35 MX |
140 | struct btrfs_root *root, |
141 | struct btrfs_log_ctx *ctx) | |
e02119d5 | 142 | { |
0b246afa | 143 | struct btrfs_fs_info *fs_info = root->fs_info; |
34eb2a52 | 144 | int ret = 0; |
7237f183 YZ |
145 | |
146 | mutex_lock(&root->log_mutex); | |
34eb2a52 | 147 | |
7237f183 | 148 | if (root->log_root) { |
4884b8e8 | 149 | if (btrfs_need_log_full_commit(trans)) { |
50471a38 MX |
150 | ret = -EAGAIN; |
151 | goto out; | |
152 | } | |
34eb2a52 | 153 | |
ff782e0a | 154 | if (!root->log_start_pid) { |
27cdeb70 | 155 | clear_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state); |
34eb2a52 | 156 | root->log_start_pid = current->pid; |
ff782e0a | 157 | } else if (root->log_start_pid != current->pid) { |
27cdeb70 | 158 | set_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state); |
ff782e0a | 159 | } |
34eb2a52 | 160 | } else { |
0b246afa JM |
161 | mutex_lock(&fs_info->tree_log_mutex); |
162 | if (!fs_info->log_root_tree) | |
163 | ret = btrfs_init_log_root_tree(trans, fs_info); | |
164 | mutex_unlock(&fs_info->tree_log_mutex); | |
34eb2a52 Z |
165 | if (ret) |
166 | goto out; | |
ff782e0a | 167 | |
e02119d5 | 168 | ret = btrfs_add_log_tree(trans, root); |
4a500fd1 | 169 | if (ret) |
e87ac136 | 170 | goto out; |
34eb2a52 | 171 | |
e7a79811 | 172 | set_bit(BTRFS_ROOT_HAS_LOG_TREE, &root->state); |
34eb2a52 Z |
173 | clear_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state); |
174 | root->log_start_pid = current->pid; | |
e02119d5 | 175 | } |
34eb2a52 | 176 | |
2ecb7923 | 177 | atomic_inc(&root->log_batch); |
7237f183 | 178 | atomic_inc(&root->log_writers); |
8b050d35 | 179 | if (ctx) { |
34eb2a52 | 180 | int index = root->log_transid % 2; |
8b050d35 | 181 | list_add_tail(&ctx->list, &root->log_ctxs[index]); |
d1433deb | 182 | ctx->log_transid = root->log_transid; |
8b050d35 | 183 | } |
34eb2a52 | 184 | |
e87ac136 | 185 | out: |
7237f183 | 186 | mutex_unlock(&root->log_mutex); |
e87ac136 | 187 | return ret; |
e02119d5 CM |
188 | } |
189 | ||
190 | /* | |
191 | * returns 0 if there was a log transaction running and we were able | |
192 | * to join, or returns -ENOENT if there were not transactions | |
193 | * in progress | |
194 | */ | |
195 | static int join_running_log_trans(struct btrfs_root *root) | |
196 | { | |
197 | int ret = -ENOENT; | |
198 | ||
e7a79811 FM |
199 | if (!test_bit(BTRFS_ROOT_HAS_LOG_TREE, &root->state)) |
200 | return ret; | |
201 | ||
7237f183 | 202 | mutex_lock(&root->log_mutex); |
e02119d5 CM |
203 | if (root->log_root) { |
204 | ret = 0; | |
7237f183 | 205 | atomic_inc(&root->log_writers); |
e02119d5 | 206 | } |
7237f183 | 207 | mutex_unlock(&root->log_mutex); |
e02119d5 CM |
208 | return ret; |
209 | } | |
210 | ||
12fcfd22 CM |
211 | /* |
212 | * This either makes the current running log transaction wait | |
213 | * until you call btrfs_end_log_trans() or it makes any future | |
214 | * log transactions wait until you call btrfs_end_log_trans() | |
215 | */ | |
45128b08 | 216 | void btrfs_pin_log_trans(struct btrfs_root *root) |
12fcfd22 | 217 | { |
12fcfd22 CM |
218 | mutex_lock(&root->log_mutex); |
219 | atomic_inc(&root->log_writers); | |
220 | mutex_unlock(&root->log_mutex); | |
12fcfd22 CM |
221 | } |
222 | ||
e02119d5 CM |
223 | /* |
224 | * indicate we're done making changes to the log tree | |
225 | * and wake up anyone waiting to do a sync | |
226 | */ | |
143bede5 | 227 | void btrfs_end_log_trans(struct btrfs_root *root) |
e02119d5 | 228 | { |
7237f183 | 229 | if (atomic_dec_and_test(&root->log_writers)) { |
093258e6 DS |
230 | /* atomic_dec_and_test implies a barrier */ |
231 | cond_wake_up_nomb(&root->log_writer_wait); | |
7237f183 | 232 | } |
e02119d5 CM |
233 | } |
234 | ||
247462a5 DS |
235 | static int btrfs_write_tree_block(struct extent_buffer *buf) |
236 | { | |
237 | return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start, | |
238 | buf->start + buf->len - 1); | |
239 | } | |
240 | ||
241 | static void btrfs_wait_tree_block_writeback(struct extent_buffer *buf) | |
242 | { | |
243 | filemap_fdatawait_range(buf->pages[0]->mapping, | |
244 | buf->start, buf->start + buf->len - 1); | |
245 | } | |
e02119d5 CM |
246 | |
247 | /* | |
248 | * the walk control struct is used to pass state down the chain when | |
249 | * processing the log tree. The stage field tells us which part | |
250 | * of the log tree processing we are currently doing. The others | |
251 | * are state fields used for that specific part | |
252 | */ | |
253 | struct walk_control { | |
254 | /* should we free the extent on disk when done? This is used | |
255 | * at transaction commit time while freeing a log tree | |
256 | */ | |
257 | int free; | |
258 | ||
259 | /* should we write out the extent buffer? This is used | |
260 | * while flushing the log tree to disk during a sync | |
261 | */ | |
262 | int write; | |
263 | ||
264 | /* should we wait for the extent buffer io to finish? Also used | |
265 | * while flushing the log tree to disk for a sync | |
266 | */ | |
267 | int wait; | |
268 | ||
269 | /* pin only walk, we record which extents on disk belong to the | |
270 | * log trees | |
271 | */ | |
272 | int pin; | |
273 | ||
274 | /* what stage of the replay code we're currently in */ | |
275 | int stage; | |
276 | ||
f2d72f42 FM |
277 | /* |
278 | * Ignore any items from the inode currently being processed. Needs | |
279 | * to be set every time we find a BTRFS_INODE_ITEM_KEY and we are in | |
280 | * the LOG_WALK_REPLAY_INODES stage. | |
281 | */ | |
282 | bool ignore_cur_inode; | |
283 | ||
e02119d5 CM |
284 | /* the root we are currently replaying */ |
285 | struct btrfs_root *replay_dest; | |
286 | ||
287 | /* the trans handle for the current replay */ | |
288 | struct btrfs_trans_handle *trans; | |
289 | ||
290 | /* the function that gets used to process blocks we find in the | |
291 | * tree. Note the extent_buffer might not be up to date when it is | |
292 | * passed in, and it must be checked or read if you need the data | |
293 | * inside it | |
294 | */ | |
295 | int (*process_func)(struct btrfs_root *log, struct extent_buffer *eb, | |
581c1760 | 296 | struct walk_control *wc, u64 gen, int level); |
e02119d5 CM |
297 | }; |
298 | ||
299 | /* | |
300 | * process_func used to pin down extents, write them or wait on them | |
301 | */ | |
302 | static int process_one_buffer(struct btrfs_root *log, | |
303 | struct extent_buffer *eb, | |
581c1760 | 304 | struct walk_control *wc, u64 gen, int level) |
e02119d5 | 305 | { |
0b246afa | 306 | struct btrfs_fs_info *fs_info = log->fs_info; |
b50c6e25 JB |
307 | int ret = 0; |
308 | ||
8c2a1a30 JB |
309 | /* |
310 | * If this fs is mixed then we need to be able to process the leaves to | |
311 | * pin down any logged extents, so we have to read the block. | |
312 | */ | |
0b246afa | 313 | if (btrfs_fs_incompat(fs_info, MIXED_GROUPS)) { |
581c1760 | 314 | ret = btrfs_read_buffer(eb, gen, level, NULL); |
8c2a1a30 JB |
315 | if (ret) |
316 | return ret; | |
317 | } | |
318 | ||
04018de5 | 319 | if (wc->pin) |
9fce5704 | 320 | ret = btrfs_pin_extent_for_log_replay(wc->trans, eb->start, |
2ff7e61e | 321 | eb->len); |
e02119d5 | 322 | |
b50c6e25 | 323 | if (!ret && btrfs_buffer_uptodate(eb, gen, 0)) { |
8c2a1a30 | 324 | if (wc->pin && btrfs_header_level(eb) == 0) |
bcdc428c | 325 | ret = btrfs_exclude_logged_extents(eb); |
e02119d5 CM |
326 | if (wc->write) |
327 | btrfs_write_tree_block(eb); | |
328 | if (wc->wait) | |
329 | btrfs_wait_tree_block_writeback(eb); | |
330 | } | |
b50c6e25 | 331 | return ret; |
e02119d5 CM |
332 | } |
333 | ||
334 | /* | |
335 | * Item overwrite used by replay and tree logging. eb, slot and key all refer | |
336 | * to the src data we are copying out. | |
337 | * | |
338 | * root is the tree we are copying into, and path is a scratch | |
339 | * path for use in this function (it should be released on entry and | |
340 | * will be released on exit). | |
341 | * | |
342 | * If the key is already in the destination tree the existing item is | |
343 | * overwritten. If the existing item isn't big enough, it is extended. | |
344 | * If it is too large, it is truncated. | |
345 | * | |
346 | * If the key isn't in the destination yet, a new item is inserted. | |
347 | */ | |
348 | static noinline int overwrite_item(struct btrfs_trans_handle *trans, | |
349 | struct btrfs_root *root, | |
350 | struct btrfs_path *path, | |
351 | struct extent_buffer *eb, int slot, | |
352 | struct btrfs_key *key) | |
353 | { | |
354 | int ret; | |
355 | u32 item_size; | |
356 | u64 saved_i_size = 0; | |
357 | int save_old_i_size = 0; | |
358 | unsigned long src_ptr; | |
359 | unsigned long dst_ptr; | |
360 | int overwrite_root = 0; | |
4bc4bee4 | 361 | bool inode_item = key->type == BTRFS_INODE_ITEM_KEY; |
e02119d5 CM |
362 | |
363 | if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) | |
364 | overwrite_root = 1; | |
365 | ||
366 | item_size = btrfs_item_size_nr(eb, slot); | |
367 | src_ptr = btrfs_item_ptr_offset(eb, slot); | |
368 | ||
369 | /* look for the key in the destination tree */ | |
370 | ret = btrfs_search_slot(NULL, root, key, path, 0, 0); | |
4bc4bee4 JB |
371 | if (ret < 0) |
372 | return ret; | |
373 | ||
e02119d5 CM |
374 | if (ret == 0) { |
375 | char *src_copy; | |
376 | char *dst_copy; | |
377 | u32 dst_size = btrfs_item_size_nr(path->nodes[0], | |
378 | path->slots[0]); | |
379 | if (dst_size != item_size) | |
380 | goto insert; | |
381 | ||
382 | if (item_size == 0) { | |
b3b4aa74 | 383 | btrfs_release_path(path); |
e02119d5 CM |
384 | return 0; |
385 | } | |
386 | dst_copy = kmalloc(item_size, GFP_NOFS); | |
387 | src_copy = kmalloc(item_size, GFP_NOFS); | |
2a29edc6 | 388 | if (!dst_copy || !src_copy) { |
b3b4aa74 | 389 | btrfs_release_path(path); |
2a29edc6 | 390 | kfree(dst_copy); |
391 | kfree(src_copy); | |
392 | return -ENOMEM; | |
393 | } | |
e02119d5 CM |
394 | |
395 | read_extent_buffer(eb, src_copy, src_ptr, item_size); | |
396 | ||
397 | dst_ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]); | |
398 | read_extent_buffer(path->nodes[0], dst_copy, dst_ptr, | |
399 | item_size); | |
400 | ret = memcmp(dst_copy, src_copy, item_size); | |
401 | ||
402 | kfree(dst_copy); | |
403 | kfree(src_copy); | |
404 | /* | |
405 | * they have the same contents, just return, this saves | |
406 | * us from cowing blocks in the destination tree and doing | |
407 | * extra writes that may not have been done by a previous | |
408 | * sync | |
409 | */ | |
410 | if (ret == 0) { | |
b3b4aa74 | 411 | btrfs_release_path(path); |
e02119d5 CM |
412 | return 0; |
413 | } | |
414 | ||
4bc4bee4 JB |
415 | /* |
416 | * We need to load the old nbytes into the inode so when we | |
417 | * replay the extents we've logged we get the right nbytes. | |
418 | */ | |
419 | if (inode_item) { | |
420 | struct btrfs_inode_item *item; | |
421 | u64 nbytes; | |
d555438b | 422 | u32 mode; |
4bc4bee4 JB |
423 | |
424 | item = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
425 | struct btrfs_inode_item); | |
426 | nbytes = btrfs_inode_nbytes(path->nodes[0], item); | |
427 | item = btrfs_item_ptr(eb, slot, | |
428 | struct btrfs_inode_item); | |
429 | btrfs_set_inode_nbytes(eb, item, nbytes); | |
d555438b JB |
430 | |
431 | /* | |
432 | * If this is a directory we need to reset the i_size to | |
433 | * 0 so that we can set it up properly when replaying | |
434 | * the rest of the items in this log. | |
435 | */ | |
436 | mode = btrfs_inode_mode(eb, item); | |
437 | if (S_ISDIR(mode)) | |
438 | btrfs_set_inode_size(eb, item, 0); | |
4bc4bee4 JB |
439 | } |
440 | } else if (inode_item) { | |
441 | struct btrfs_inode_item *item; | |
d555438b | 442 | u32 mode; |
4bc4bee4 JB |
443 | |
444 | /* | |
445 | * New inode, set nbytes to 0 so that the nbytes comes out | |
446 | * properly when we replay the extents. | |
447 | */ | |
448 | item = btrfs_item_ptr(eb, slot, struct btrfs_inode_item); | |
449 | btrfs_set_inode_nbytes(eb, item, 0); | |
d555438b JB |
450 | |
451 | /* | |
452 | * If this is a directory we need to reset the i_size to 0 so | |
453 | * that we can set it up properly when replaying the rest of | |
454 | * the items in this log. | |
455 | */ | |
456 | mode = btrfs_inode_mode(eb, item); | |
457 | if (S_ISDIR(mode)) | |
458 | btrfs_set_inode_size(eb, item, 0); | |
e02119d5 CM |
459 | } |
460 | insert: | |
b3b4aa74 | 461 | btrfs_release_path(path); |
e02119d5 | 462 | /* try to insert the key into the destination tree */ |
df8d116f | 463 | path->skip_release_on_error = 1; |
e02119d5 CM |
464 | ret = btrfs_insert_empty_item(trans, root, path, |
465 | key, item_size); | |
df8d116f | 466 | path->skip_release_on_error = 0; |
e02119d5 CM |
467 | |
468 | /* make sure any existing item is the correct size */ | |
df8d116f | 469 | if (ret == -EEXIST || ret == -EOVERFLOW) { |
e02119d5 CM |
470 | u32 found_size; |
471 | found_size = btrfs_item_size_nr(path->nodes[0], | |
472 | path->slots[0]); | |
143bede5 | 473 | if (found_size > item_size) |
78ac4f9e | 474 | btrfs_truncate_item(path, item_size, 1); |
143bede5 | 475 | else if (found_size < item_size) |
c71dd880 | 476 | btrfs_extend_item(path, item_size - found_size); |
e02119d5 | 477 | } else if (ret) { |
4a500fd1 | 478 | return ret; |
e02119d5 CM |
479 | } |
480 | dst_ptr = btrfs_item_ptr_offset(path->nodes[0], | |
481 | path->slots[0]); | |
482 | ||
483 | /* don't overwrite an existing inode if the generation number | |
484 | * was logged as zero. This is done when the tree logging code | |
485 | * is just logging an inode to make sure it exists after recovery. | |
486 | * | |
487 | * Also, don't overwrite i_size on directories during replay. | |
488 | * log replay inserts and removes directory items based on the | |
489 | * state of the tree found in the subvolume, and i_size is modified | |
490 | * as it goes | |
491 | */ | |
492 | if (key->type == BTRFS_INODE_ITEM_KEY && ret == -EEXIST) { | |
493 | struct btrfs_inode_item *src_item; | |
494 | struct btrfs_inode_item *dst_item; | |
495 | ||
496 | src_item = (struct btrfs_inode_item *)src_ptr; | |
497 | dst_item = (struct btrfs_inode_item *)dst_ptr; | |
498 | ||
1a4bcf47 FM |
499 | if (btrfs_inode_generation(eb, src_item) == 0) { |
500 | struct extent_buffer *dst_eb = path->nodes[0]; | |
2f2ff0ee | 501 | const u64 ino_size = btrfs_inode_size(eb, src_item); |
1a4bcf47 | 502 | |
2f2ff0ee FM |
503 | /* |
504 | * For regular files an ino_size == 0 is used only when | |
505 | * logging that an inode exists, as part of a directory | |
506 | * fsync, and the inode wasn't fsynced before. In this | |
507 | * case don't set the size of the inode in the fs/subvol | |
508 | * tree, otherwise we would be throwing valid data away. | |
509 | */ | |
1a4bcf47 | 510 | if (S_ISREG(btrfs_inode_mode(eb, src_item)) && |
2f2ff0ee | 511 | S_ISREG(btrfs_inode_mode(dst_eb, dst_item)) && |
60d48e2e DS |
512 | ino_size != 0) |
513 | btrfs_set_inode_size(dst_eb, dst_item, ino_size); | |
e02119d5 | 514 | goto no_copy; |
1a4bcf47 | 515 | } |
e02119d5 CM |
516 | |
517 | if (overwrite_root && | |
518 | S_ISDIR(btrfs_inode_mode(eb, src_item)) && | |
519 | S_ISDIR(btrfs_inode_mode(path->nodes[0], dst_item))) { | |
520 | save_old_i_size = 1; | |
521 | saved_i_size = btrfs_inode_size(path->nodes[0], | |
522 | dst_item); | |
523 | } | |
524 | } | |
525 | ||
526 | copy_extent_buffer(path->nodes[0], eb, dst_ptr, | |
527 | src_ptr, item_size); | |
528 | ||
529 | if (save_old_i_size) { | |
530 | struct btrfs_inode_item *dst_item; | |
531 | dst_item = (struct btrfs_inode_item *)dst_ptr; | |
532 | btrfs_set_inode_size(path->nodes[0], dst_item, saved_i_size); | |
533 | } | |
534 | ||
535 | /* make sure the generation is filled in */ | |
536 | if (key->type == BTRFS_INODE_ITEM_KEY) { | |
537 | struct btrfs_inode_item *dst_item; | |
538 | dst_item = (struct btrfs_inode_item *)dst_ptr; | |
539 | if (btrfs_inode_generation(path->nodes[0], dst_item) == 0) { | |
540 | btrfs_set_inode_generation(path->nodes[0], dst_item, | |
541 | trans->transid); | |
542 | } | |
543 | } | |
544 | no_copy: | |
545 | btrfs_mark_buffer_dirty(path->nodes[0]); | |
b3b4aa74 | 546 | btrfs_release_path(path); |
e02119d5 CM |
547 | return 0; |
548 | } | |
549 | ||
550 | /* | |
551 | * simple helper to read an inode off the disk from a given root | |
552 | * This can only be called for subvolume roots and not for the log | |
553 | */ | |
554 | static noinline struct inode *read_one_inode(struct btrfs_root *root, | |
555 | u64 objectid) | |
556 | { | |
557 | struct inode *inode; | |
e02119d5 | 558 | |
0202e83f | 559 | inode = btrfs_iget(root->fs_info->sb, objectid, root); |
2e19f1f9 | 560 | if (IS_ERR(inode)) |
5d4f98a2 | 561 | inode = NULL; |
e02119d5 CM |
562 | return inode; |
563 | } | |
564 | ||
565 | /* replays a single extent in 'eb' at 'slot' with 'key' into the | |
566 | * subvolume 'root'. path is released on entry and should be released | |
567 | * on exit. | |
568 | * | |
569 | * extents in the log tree have not been allocated out of the extent | |
570 | * tree yet. So, this completes the allocation, taking a reference | |
571 | * as required if the extent already exists or creating a new extent | |
572 | * if it isn't in the extent allocation tree yet. | |
573 | * | |
574 | * The extent is inserted into the file, dropping any existing extents | |
575 | * from the file that overlap the new one. | |
576 | */ | |
577 | static noinline int replay_one_extent(struct btrfs_trans_handle *trans, | |
578 | struct btrfs_root *root, | |
579 | struct btrfs_path *path, | |
580 | struct extent_buffer *eb, int slot, | |
581 | struct btrfs_key *key) | |
582 | { | |
0b246afa | 583 | struct btrfs_fs_info *fs_info = root->fs_info; |
e02119d5 | 584 | int found_type; |
e02119d5 | 585 | u64 extent_end; |
e02119d5 | 586 | u64 start = key->offset; |
4bc4bee4 | 587 | u64 nbytes = 0; |
e02119d5 CM |
588 | struct btrfs_file_extent_item *item; |
589 | struct inode *inode = NULL; | |
590 | unsigned long size; | |
591 | int ret = 0; | |
592 | ||
593 | item = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item); | |
594 | found_type = btrfs_file_extent_type(eb, item); | |
595 | ||
d899e052 | 596 | if (found_type == BTRFS_FILE_EXTENT_REG || |
4bc4bee4 JB |
597 | found_type == BTRFS_FILE_EXTENT_PREALLOC) { |
598 | nbytes = btrfs_file_extent_num_bytes(eb, item); | |
599 | extent_end = start + nbytes; | |
600 | ||
601 | /* | |
602 | * We don't add to the inodes nbytes if we are prealloc or a | |
603 | * hole. | |
604 | */ | |
605 | if (btrfs_file_extent_disk_bytenr(eb, item) == 0) | |
606 | nbytes = 0; | |
607 | } else if (found_type == BTRFS_FILE_EXTENT_INLINE) { | |
e41ca589 | 608 | size = btrfs_file_extent_ram_bytes(eb, item); |
4bc4bee4 | 609 | nbytes = btrfs_file_extent_ram_bytes(eb, item); |
da17066c | 610 | extent_end = ALIGN(start + size, |
0b246afa | 611 | fs_info->sectorsize); |
e02119d5 CM |
612 | } else { |
613 | ret = 0; | |
614 | goto out; | |
615 | } | |
616 | ||
617 | inode = read_one_inode(root, key->objectid); | |
618 | if (!inode) { | |
619 | ret = -EIO; | |
620 | goto out; | |
621 | } | |
622 | ||
623 | /* | |
624 | * first check to see if we already have this extent in the | |
625 | * file. This must be done before the btrfs_drop_extents run | |
626 | * so we don't try to drop this extent. | |
627 | */ | |
f85b7379 DS |
628 | ret = btrfs_lookup_file_extent(trans, root, path, |
629 | btrfs_ino(BTRFS_I(inode)), start, 0); | |
e02119d5 | 630 | |
d899e052 YZ |
631 | if (ret == 0 && |
632 | (found_type == BTRFS_FILE_EXTENT_REG || | |
633 | found_type == BTRFS_FILE_EXTENT_PREALLOC)) { | |
e02119d5 CM |
634 | struct btrfs_file_extent_item cmp1; |
635 | struct btrfs_file_extent_item cmp2; | |
636 | struct btrfs_file_extent_item *existing; | |
637 | struct extent_buffer *leaf; | |
638 | ||
639 | leaf = path->nodes[0]; | |
640 | existing = btrfs_item_ptr(leaf, path->slots[0], | |
641 | struct btrfs_file_extent_item); | |
642 | ||
643 | read_extent_buffer(eb, &cmp1, (unsigned long)item, | |
644 | sizeof(cmp1)); | |
645 | read_extent_buffer(leaf, &cmp2, (unsigned long)existing, | |
646 | sizeof(cmp2)); | |
647 | ||
648 | /* | |
649 | * we already have a pointer to this exact extent, | |
650 | * we don't have to do anything | |
651 | */ | |
652 | if (memcmp(&cmp1, &cmp2, sizeof(cmp1)) == 0) { | |
b3b4aa74 | 653 | btrfs_release_path(path); |
e02119d5 CM |
654 | goto out; |
655 | } | |
656 | } | |
b3b4aa74 | 657 | btrfs_release_path(path); |
e02119d5 CM |
658 | |
659 | /* drop any overlapping extents */ | |
2671485d | 660 | ret = btrfs_drop_extents(trans, root, inode, start, extent_end, 1); |
3650860b JB |
661 | if (ret) |
662 | goto out; | |
e02119d5 | 663 | |
07d400a6 YZ |
664 | if (found_type == BTRFS_FILE_EXTENT_REG || |
665 | found_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
5d4f98a2 | 666 | u64 offset; |
07d400a6 YZ |
667 | unsigned long dest_offset; |
668 | struct btrfs_key ins; | |
669 | ||
3168021c FM |
670 | if (btrfs_file_extent_disk_bytenr(eb, item) == 0 && |
671 | btrfs_fs_incompat(fs_info, NO_HOLES)) | |
672 | goto update_inode; | |
673 | ||
07d400a6 YZ |
674 | ret = btrfs_insert_empty_item(trans, root, path, key, |
675 | sizeof(*item)); | |
3650860b JB |
676 | if (ret) |
677 | goto out; | |
07d400a6 YZ |
678 | dest_offset = btrfs_item_ptr_offset(path->nodes[0], |
679 | path->slots[0]); | |
680 | copy_extent_buffer(path->nodes[0], eb, dest_offset, | |
681 | (unsigned long)item, sizeof(*item)); | |
682 | ||
683 | ins.objectid = btrfs_file_extent_disk_bytenr(eb, item); | |
684 | ins.offset = btrfs_file_extent_disk_num_bytes(eb, item); | |
685 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
5d4f98a2 | 686 | offset = key->offset - btrfs_file_extent_offset(eb, item); |
07d400a6 | 687 | |
df2c95f3 QW |
688 | /* |
689 | * Manually record dirty extent, as here we did a shallow | |
690 | * file extent item copy and skip normal backref update, | |
691 | * but modifying extent tree all by ourselves. | |
692 | * So need to manually record dirty extent for qgroup, | |
693 | * as the owner of the file extent changed from log tree | |
694 | * (doesn't affect qgroup) to fs/file tree(affects qgroup) | |
695 | */ | |
a95f3aaf | 696 | ret = btrfs_qgroup_trace_extent(trans, |
df2c95f3 QW |
697 | btrfs_file_extent_disk_bytenr(eb, item), |
698 | btrfs_file_extent_disk_num_bytes(eb, item), | |
699 | GFP_NOFS); | |
700 | if (ret < 0) | |
701 | goto out; | |
702 | ||
07d400a6 | 703 | if (ins.objectid > 0) { |
82fa113f | 704 | struct btrfs_ref ref = { 0 }; |
07d400a6 YZ |
705 | u64 csum_start; |
706 | u64 csum_end; | |
707 | LIST_HEAD(ordered_sums); | |
82fa113f | 708 | |
07d400a6 YZ |
709 | /* |
710 | * is this extent already allocated in the extent | |
711 | * allocation tree? If so, just add a reference | |
712 | */ | |
2ff7e61e | 713 | ret = btrfs_lookup_data_extent(fs_info, ins.objectid, |
07d400a6 YZ |
714 | ins.offset); |
715 | if (ret == 0) { | |
82fa113f QW |
716 | btrfs_init_generic_ref(&ref, |
717 | BTRFS_ADD_DELAYED_REF, | |
718 | ins.objectid, ins.offset, 0); | |
719 | btrfs_init_data_ref(&ref, | |
720 | root->root_key.objectid, | |
b06c4bf5 | 721 | key->objectid, offset); |
82fa113f | 722 | ret = btrfs_inc_extent_ref(trans, &ref); |
b50c6e25 JB |
723 | if (ret) |
724 | goto out; | |
07d400a6 YZ |
725 | } else { |
726 | /* | |
727 | * insert the extent pointer in the extent | |
728 | * allocation tree | |
729 | */ | |
5d4f98a2 | 730 | ret = btrfs_alloc_logged_file_extent(trans, |
2ff7e61e | 731 | root->root_key.objectid, |
5d4f98a2 | 732 | key->objectid, offset, &ins); |
b50c6e25 JB |
733 | if (ret) |
734 | goto out; | |
07d400a6 | 735 | } |
b3b4aa74 | 736 | btrfs_release_path(path); |
07d400a6 YZ |
737 | |
738 | if (btrfs_file_extent_compression(eb, item)) { | |
739 | csum_start = ins.objectid; | |
740 | csum_end = csum_start + ins.offset; | |
741 | } else { | |
742 | csum_start = ins.objectid + | |
743 | btrfs_file_extent_offset(eb, item); | |
744 | csum_end = csum_start + | |
745 | btrfs_file_extent_num_bytes(eb, item); | |
746 | } | |
747 | ||
748 | ret = btrfs_lookup_csums_range(root->log_root, | |
749 | csum_start, csum_end - 1, | |
a2de733c | 750 | &ordered_sums, 0); |
3650860b JB |
751 | if (ret) |
752 | goto out; | |
b84b8390 FM |
753 | /* |
754 | * Now delete all existing cums in the csum root that | |
755 | * cover our range. We do this because we can have an | |
756 | * extent that is completely referenced by one file | |
757 | * extent item and partially referenced by another | |
758 | * file extent item (like after using the clone or | |
759 | * extent_same ioctls). In this case if we end up doing | |
760 | * the replay of the one that partially references the | |
761 | * extent first, and we do not do the csum deletion | |
762 | * below, we can get 2 csum items in the csum tree that | |
763 | * overlap each other. For example, imagine our log has | |
764 | * the two following file extent items: | |
765 | * | |
766 | * key (257 EXTENT_DATA 409600) | |
767 | * extent data disk byte 12845056 nr 102400 | |
768 | * extent data offset 20480 nr 20480 ram 102400 | |
769 | * | |
770 | * key (257 EXTENT_DATA 819200) | |
771 | * extent data disk byte 12845056 nr 102400 | |
772 | * extent data offset 0 nr 102400 ram 102400 | |
773 | * | |
774 | * Where the second one fully references the 100K extent | |
775 | * that starts at disk byte 12845056, and the log tree | |
776 | * has a single csum item that covers the entire range | |
777 | * of the extent: | |
778 | * | |
779 | * key (EXTENT_CSUM EXTENT_CSUM 12845056) itemsize 100 | |
780 | * | |
781 | * After the first file extent item is replayed, the | |
782 | * csum tree gets the following csum item: | |
783 | * | |
784 | * key (EXTENT_CSUM EXTENT_CSUM 12865536) itemsize 20 | |
785 | * | |
786 | * Which covers the 20K sub-range starting at offset 20K | |
787 | * of our extent. Now when we replay the second file | |
788 | * extent item, if we do not delete existing csum items | |
789 | * that cover any of its blocks, we end up getting two | |
790 | * csum items in our csum tree that overlap each other: | |
791 | * | |
792 | * key (EXTENT_CSUM EXTENT_CSUM 12845056) itemsize 100 | |
793 | * key (EXTENT_CSUM EXTENT_CSUM 12865536) itemsize 20 | |
794 | * | |
795 | * Which is a problem, because after this anyone trying | |
796 | * to lookup up for the checksum of any block of our | |
797 | * extent starting at an offset of 40K or higher, will | |
798 | * end up looking at the second csum item only, which | |
799 | * does not contain the checksum for any block starting | |
800 | * at offset 40K or higher of our extent. | |
801 | */ | |
07d400a6 YZ |
802 | while (!list_empty(&ordered_sums)) { |
803 | struct btrfs_ordered_sum *sums; | |
804 | sums = list_entry(ordered_sums.next, | |
805 | struct btrfs_ordered_sum, | |
806 | list); | |
b84b8390 | 807 | if (!ret) |
40e046ac FM |
808 | ret = btrfs_del_csums(trans, |
809 | fs_info->csum_root, | |
5b4aacef JM |
810 | sums->bytenr, |
811 | sums->len); | |
3650860b JB |
812 | if (!ret) |
813 | ret = btrfs_csum_file_blocks(trans, | |
0b246afa | 814 | fs_info->csum_root, sums); |
07d400a6 YZ |
815 | list_del(&sums->list); |
816 | kfree(sums); | |
817 | } | |
3650860b JB |
818 | if (ret) |
819 | goto out; | |
07d400a6 | 820 | } else { |
b3b4aa74 | 821 | btrfs_release_path(path); |
07d400a6 YZ |
822 | } |
823 | } else if (found_type == BTRFS_FILE_EXTENT_INLINE) { | |
824 | /* inline extents are easy, we just overwrite them */ | |
825 | ret = overwrite_item(trans, root, path, eb, slot, key); | |
3650860b JB |
826 | if (ret) |
827 | goto out; | |
07d400a6 | 828 | } |
e02119d5 | 829 | |
9ddc959e JB |
830 | ret = btrfs_inode_set_file_extent_range(BTRFS_I(inode), start, |
831 | extent_end - start); | |
832 | if (ret) | |
833 | goto out; | |
834 | ||
4bc4bee4 | 835 | inode_add_bytes(inode, nbytes); |
3168021c | 836 | update_inode: |
b9959295 | 837 | ret = btrfs_update_inode(trans, root, inode); |
e02119d5 CM |
838 | out: |
839 | if (inode) | |
840 | iput(inode); | |
841 | return ret; | |
842 | } | |
843 | ||
844 | /* | |
845 | * when cleaning up conflicts between the directory names in the | |
846 | * subvolume, directory names in the log and directory names in the | |
847 | * inode back references, we may have to unlink inodes from directories. | |
848 | * | |
849 | * This is a helper function to do the unlink of a specific directory | |
850 | * item | |
851 | */ | |
852 | static noinline int drop_one_dir_item(struct btrfs_trans_handle *trans, | |
853 | struct btrfs_root *root, | |
854 | struct btrfs_path *path, | |
207e7d92 | 855 | struct btrfs_inode *dir, |
e02119d5 CM |
856 | struct btrfs_dir_item *di) |
857 | { | |
858 | struct inode *inode; | |
859 | char *name; | |
860 | int name_len; | |
861 | struct extent_buffer *leaf; | |
862 | struct btrfs_key location; | |
863 | int ret; | |
864 | ||
865 | leaf = path->nodes[0]; | |
866 | ||
867 | btrfs_dir_item_key_to_cpu(leaf, di, &location); | |
868 | name_len = btrfs_dir_name_len(leaf, di); | |
869 | name = kmalloc(name_len, GFP_NOFS); | |
2a29edc6 | 870 | if (!name) |
871 | return -ENOMEM; | |
872 | ||
e02119d5 | 873 | read_extent_buffer(leaf, name, (unsigned long)(di + 1), name_len); |
b3b4aa74 | 874 | btrfs_release_path(path); |
e02119d5 CM |
875 | |
876 | inode = read_one_inode(root, location.objectid); | |
c00e9493 | 877 | if (!inode) { |
3650860b JB |
878 | ret = -EIO; |
879 | goto out; | |
c00e9493 | 880 | } |
e02119d5 | 881 | |
ec051c0f | 882 | ret = link_to_fixup_dir(trans, root, path, location.objectid); |
3650860b JB |
883 | if (ret) |
884 | goto out; | |
12fcfd22 | 885 | |
207e7d92 NB |
886 | ret = btrfs_unlink_inode(trans, root, dir, BTRFS_I(inode), name, |
887 | name_len); | |
3650860b JB |
888 | if (ret) |
889 | goto out; | |
ada9af21 | 890 | else |
e5c304e6 | 891 | ret = btrfs_run_delayed_items(trans); |
3650860b | 892 | out: |
e02119d5 | 893 | kfree(name); |
e02119d5 CM |
894 | iput(inode); |
895 | return ret; | |
896 | } | |
897 | ||
898 | /* | |
899 | * helper function to see if a given name and sequence number found | |
900 | * in an inode back reference are already in a directory and correctly | |
901 | * point to this inode | |
902 | */ | |
903 | static noinline int inode_in_dir(struct btrfs_root *root, | |
904 | struct btrfs_path *path, | |
905 | u64 dirid, u64 objectid, u64 index, | |
906 | const char *name, int name_len) | |
907 | { | |
908 | struct btrfs_dir_item *di; | |
909 | struct btrfs_key location; | |
910 | int match = 0; | |
911 | ||
912 | di = btrfs_lookup_dir_index_item(NULL, root, path, dirid, | |
913 | index, name, name_len, 0); | |
914 | if (di && !IS_ERR(di)) { | |
915 | btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location); | |
916 | if (location.objectid != objectid) | |
917 | goto out; | |
918 | } else | |
919 | goto out; | |
b3b4aa74 | 920 | btrfs_release_path(path); |
e02119d5 CM |
921 | |
922 | di = btrfs_lookup_dir_item(NULL, root, path, dirid, name, name_len, 0); | |
923 | if (di && !IS_ERR(di)) { | |
924 | btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location); | |
925 | if (location.objectid != objectid) | |
926 | goto out; | |
927 | } else | |
928 | goto out; | |
929 | match = 1; | |
930 | out: | |
b3b4aa74 | 931 | btrfs_release_path(path); |
e02119d5 CM |
932 | return match; |
933 | } | |
934 | ||
935 | /* | |
936 | * helper function to check a log tree for a named back reference in | |
937 | * an inode. This is used to decide if a back reference that is | |
938 | * found in the subvolume conflicts with what we find in the log. | |
939 | * | |
940 | * inode backreferences may have multiple refs in a single item, | |
941 | * during replay we process one reference at a time, and we don't | |
942 | * want to delete valid links to a file from the subvolume if that | |
943 | * link is also in the log. | |
944 | */ | |
945 | static noinline int backref_in_log(struct btrfs_root *log, | |
946 | struct btrfs_key *key, | |
f186373f | 947 | u64 ref_objectid, |
df8d116f | 948 | const char *name, int namelen) |
e02119d5 CM |
949 | { |
950 | struct btrfs_path *path; | |
e02119d5 | 951 | int ret; |
e02119d5 CM |
952 | |
953 | path = btrfs_alloc_path(); | |
2a29edc6 | 954 | if (!path) |
955 | return -ENOMEM; | |
956 | ||
e02119d5 | 957 | ret = btrfs_search_slot(NULL, log, key, path, 0, 0); |
d3316c82 NB |
958 | if (ret < 0) { |
959 | goto out; | |
960 | } else if (ret == 1) { | |
89cbf5f6 | 961 | ret = 0; |
f186373f MF |
962 | goto out; |
963 | } | |
964 | ||
89cbf5f6 NB |
965 | if (key->type == BTRFS_INODE_EXTREF_KEY) |
966 | ret = !!btrfs_find_name_in_ext_backref(path->nodes[0], | |
967 | path->slots[0], | |
968 | ref_objectid, | |
969 | name, namelen); | |
970 | else | |
971 | ret = !!btrfs_find_name_in_backref(path->nodes[0], | |
972 | path->slots[0], | |
973 | name, namelen); | |
e02119d5 CM |
974 | out: |
975 | btrfs_free_path(path); | |
89cbf5f6 | 976 | return ret; |
e02119d5 CM |
977 | } |
978 | ||
5a1d7843 | 979 | static inline int __add_inode_ref(struct btrfs_trans_handle *trans, |
e02119d5 | 980 | struct btrfs_root *root, |
e02119d5 | 981 | struct btrfs_path *path, |
5a1d7843 | 982 | struct btrfs_root *log_root, |
94c91a1f NB |
983 | struct btrfs_inode *dir, |
984 | struct btrfs_inode *inode, | |
f186373f MF |
985 | u64 inode_objectid, u64 parent_objectid, |
986 | u64 ref_index, char *name, int namelen, | |
987 | int *search_done) | |
e02119d5 | 988 | { |
34f3e4f2 | 989 | int ret; |
f186373f MF |
990 | char *victim_name; |
991 | int victim_name_len; | |
992 | struct extent_buffer *leaf; | |
5a1d7843 | 993 | struct btrfs_dir_item *di; |
f186373f MF |
994 | struct btrfs_key search_key; |
995 | struct btrfs_inode_extref *extref; | |
c622ae60 | 996 | |
f186373f MF |
997 | again: |
998 | /* Search old style refs */ | |
999 | search_key.objectid = inode_objectid; | |
1000 | search_key.type = BTRFS_INODE_REF_KEY; | |
1001 | search_key.offset = parent_objectid; | |
1002 | ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0); | |
e02119d5 | 1003 | if (ret == 0) { |
e02119d5 CM |
1004 | struct btrfs_inode_ref *victim_ref; |
1005 | unsigned long ptr; | |
1006 | unsigned long ptr_end; | |
f186373f MF |
1007 | |
1008 | leaf = path->nodes[0]; | |
e02119d5 CM |
1009 | |
1010 | /* are we trying to overwrite a back ref for the root directory | |
1011 | * if so, just jump out, we're done | |
1012 | */ | |
f186373f | 1013 | if (search_key.objectid == search_key.offset) |
5a1d7843 | 1014 | return 1; |
e02119d5 CM |
1015 | |
1016 | /* check all the names in this back reference to see | |
1017 | * if they are in the log. if so, we allow them to stay | |
1018 | * otherwise they must be unlinked as a conflict | |
1019 | */ | |
1020 | ptr = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
1021 | ptr_end = ptr + btrfs_item_size_nr(leaf, path->slots[0]); | |
d397712b | 1022 | while (ptr < ptr_end) { |
e02119d5 CM |
1023 | victim_ref = (struct btrfs_inode_ref *)ptr; |
1024 | victim_name_len = btrfs_inode_ref_name_len(leaf, | |
1025 | victim_ref); | |
1026 | victim_name = kmalloc(victim_name_len, GFP_NOFS); | |
3650860b JB |
1027 | if (!victim_name) |
1028 | return -ENOMEM; | |
e02119d5 CM |
1029 | |
1030 | read_extent_buffer(leaf, victim_name, | |
1031 | (unsigned long)(victim_ref + 1), | |
1032 | victim_name_len); | |
1033 | ||
d3316c82 NB |
1034 | ret = backref_in_log(log_root, &search_key, |
1035 | parent_objectid, victim_name, | |
1036 | victim_name_len); | |
1037 | if (ret < 0) { | |
1038 | kfree(victim_name); | |
1039 | return ret; | |
1040 | } else if (!ret) { | |
94c91a1f | 1041 | inc_nlink(&inode->vfs_inode); |
b3b4aa74 | 1042 | btrfs_release_path(path); |
12fcfd22 | 1043 | |
94c91a1f | 1044 | ret = btrfs_unlink_inode(trans, root, dir, inode, |
4ec5934e | 1045 | victim_name, victim_name_len); |
f186373f | 1046 | kfree(victim_name); |
3650860b JB |
1047 | if (ret) |
1048 | return ret; | |
e5c304e6 | 1049 | ret = btrfs_run_delayed_items(trans); |
ada9af21 FDBM |
1050 | if (ret) |
1051 | return ret; | |
f186373f MF |
1052 | *search_done = 1; |
1053 | goto again; | |
e02119d5 CM |
1054 | } |
1055 | kfree(victim_name); | |
f186373f | 1056 | |
e02119d5 CM |
1057 | ptr = (unsigned long)(victim_ref + 1) + victim_name_len; |
1058 | } | |
e02119d5 | 1059 | |
c622ae60 | 1060 | /* |
1061 | * NOTE: we have searched root tree and checked the | |
bb7ab3b9 | 1062 | * corresponding ref, it does not need to check again. |
c622ae60 | 1063 | */ |
5a1d7843 | 1064 | *search_done = 1; |
e02119d5 | 1065 | } |
b3b4aa74 | 1066 | btrfs_release_path(path); |
e02119d5 | 1067 | |
f186373f MF |
1068 | /* Same search but for extended refs */ |
1069 | extref = btrfs_lookup_inode_extref(NULL, root, path, name, namelen, | |
1070 | inode_objectid, parent_objectid, 0, | |
1071 | 0); | |
1072 | if (!IS_ERR_OR_NULL(extref)) { | |
1073 | u32 item_size; | |
1074 | u32 cur_offset = 0; | |
1075 | unsigned long base; | |
1076 | struct inode *victim_parent; | |
1077 | ||
1078 | leaf = path->nodes[0]; | |
1079 | ||
1080 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1081 | base = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
1082 | ||
1083 | while (cur_offset < item_size) { | |
dd9ef135 | 1084 | extref = (struct btrfs_inode_extref *)(base + cur_offset); |
f186373f MF |
1085 | |
1086 | victim_name_len = btrfs_inode_extref_name_len(leaf, extref); | |
1087 | ||
1088 | if (btrfs_inode_extref_parent(leaf, extref) != parent_objectid) | |
1089 | goto next; | |
1090 | ||
1091 | victim_name = kmalloc(victim_name_len, GFP_NOFS); | |
3650860b JB |
1092 | if (!victim_name) |
1093 | return -ENOMEM; | |
f186373f MF |
1094 | read_extent_buffer(leaf, victim_name, (unsigned long)&extref->name, |
1095 | victim_name_len); | |
1096 | ||
1097 | search_key.objectid = inode_objectid; | |
1098 | search_key.type = BTRFS_INODE_EXTREF_KEY; | |
1099 | search_key.offset = btrfs_extref_hash(parent_objectid, | |
1100 | victim_name, | |
1101 | victim_name_len); | |
d3316c82 NB |
1102 | ret = backref_in_log(log_root, &search_key, |
1103 | parent_objectid, victim_name, | |
1104 | victim_name_len); | |
1105 | if (ret < 0) { | |
1106 | return ret; | |
1107 | } else if (!ret) { | |
f186373f MF |
1108 | ret = -ENOENT; |
1109 | victim_parent = read_one_inode(root, | |
94c91a1f | 1110 | parent_objectid); |
f186373f | 1111 | if (victim_parent) { |
94c91a1f | 1112 | inc_nlink(&inode->vfs_inode); |
f186373f MF |
1113 | btrfs_release_path(path); |
1114 | ||
1115 | ret = btrfs_unlink_inode(trans, root, | |
4ec5934e | 1116 | BTRFS_I(victim_parent), |
94c91a1f | 1117 | inode, |
4ec5934e NB |
1118 | victim_name, |
1119 | victim_name_len); | |
ada9af21 FDBM |
1120 | if (!ret) |
1121 | ret = btrfs_run_delayed_items( | |
e5c304e6 | 1122 | trans); |
f186373f | 1123 | } |
f186373f MF |
1124 | iput(victim_parent); |
1125 | kfree(victim_name); | |
3650860b JB |
1126 | if (ret) |
1127 | return ret; | |
f186373f MF |
1128 | *search_done = 1; |
1129 | goto again; | |
1130 | } | |
1131 | kfree(victim_name); | |
f186373f MF |
1132 | next: |
1133 | cur_offset += victim_name_len + sizeof(*extref); | |
1134 | } | |
1135 | *search_done = 1; | |
1136 | } | |
1137 | btrfs_release_path(path); | |
1138 | ||
34f3e4f2 | 1139 | /* look for a conflicting sequence number */ |
94c91a1f | 1140 | di = btrfs_lookup_dir_index_item(trans, root, path, btrfs_ino(dir), |
f186373f | 1141 | ref_index, name, namelen, 0); |
34f3e4f2 | 1142 | if (di && !IS_ERR(di)) { |
94c91a1f | 1143 | ret = drop_one_dir_item(trans, root, path, dir, di); |
3650860b JB |
1144 | if (ret) |
1145 | return ret; | |
34f3e4f2 | 1146 | } |
1147 | btrfs_release_path(path); | |
1148 | ||
52042d8e | 1149 | /* look for a conflicting name */ |
94c91a1f | 1150 | di = btrfs_lookup_dir_item(trans, root, path, btrfs_ino(dir), |
34f3e4f2 | 1151 | name, namelen, 0); |
1152 | if (di && !IS_ERR(di)) { | |
94c91a1f | 1153 | ret = drop_one_dir_item(trans, root, path, dir, di); |
3650860b JB |
1154 | if (ret) |
1155 | return ret; | |
34f3e4f2 | 1156 | } |
1157 | btrfs_release_path(path); | |
1158 | ||
5a1d7843 JS |
1159 | return 0; |
1160 | } | |
e02119d5 | 1161 | |
bae15d95 QW |
1162 | static int extref_get_fields(struct extent_buffer *eb, unsigned long ref_ptr, |
1163 | u32 *namelen, char **name, u64 *index, | |
1164 | u64 *parent_objectid) | |
f186373f MF |
1165 | { |
1166 | struct btrfs_inode_extref *extref; | |
1167 | ||
1168 | extref = (struct btrfs_inode_extref *)ref_ptr; | |
1169 | ||
1170 | *namelen = btrfs_inode_extref_name_len(eb, extref); | |
1171 | *name = kmalloc(*namelen, GFP_NOFS); | |
1172 | if (*name == NULL) | |
1173 | return -ENOMEM; | |
1174 | ||
1175 | read_extent_buffer(eb, *name, (unsigned long)&extref->name, | |
1176 | *namelen); | |
1177 | ||
1f250e92 FM |
1178 | if (index) |
1179 | *index = btrfs_inode_extref_index(eb, extref); | |
f186373f MF |
1180 | if (parent_objectid) |
1181 | *parent_objectid = btrfs_inode_extref_parent(eb, extref); | |
1182 | ||
1183 | return 0; | |
1184 | } | |
1185 | ||
bae15d95 QW |
1186 | static int ref_get_fields(struct extent_buffer *eb, unsigned long ref_ptr, |
1187 | u32 *namelen, char **name, u64 *index) | |
f186373f MF |
1188 | { |
1189 | struct btrfs_inode_ref *ref; | |
1190 | ||
1191 | ref = (struct btrfs_inode_ref *)ref_ptr; | |
1192 | ||
1193 | *namelen = btrfs_inode_ref_name_len(eb, ref); | |
1194 | *name = kmalloc(*namelen, GFP_NOFS); | |
1195 | if (*name == NULL) | |
1196 | return -ENOMEM; | |
1197 | ||
1198 | read_extent_buffer(eb, *name, (unsigned long)(ref + 1), *namelen); | |
1199 | ||
1f250e92 FM |
1200 | if (index) |
1201 | *index = btrfs_inode_ref_index(eb, ref); | |
f186373f MF |
1202 | |
1203 | return 0; | |
1204 | } | |
1205 | ||
1f250e92 FM |
1206 | /* |
1207 | * Take an inode reference item from the log tree and iterate all names from the | |
1208 | * inode reference item in the subvolume tree with the same key (if it exists). | |
1209 | * For any name that is not in the inode reference item from the log tree, do a | |
1210 | * proper unlink of that name (that is, remove its entry from the inode | |
1211 | * reference item and both dir index keys). | |
1212 | */ | |
1213 | static int unlink_old_inode_refs(struct btrfs_trans_handle *trans, | |
1214 | struct btrfs_root *root, | |
1215 | struct btrfs_path *path, | |
1216 | struct btrfs_inode *inode, | |
1217 | struct extent_buffer *log_eb, | |
1218 | int log_slot, | |
1219 | struct btrfs_key *key) | |
1220 | { | |
1221 | int ret; | |
1222 | unsigned long ref_ptr; | |
1223 | unsigned long ref_end; | |
1224 | struct extent_buffer *eb; | |
1225 | ||
1226 | again: | |
1227 | btrfs_release_path(path); | |
1228 | ret = btrfs_search_slot(NULL, root, key, path, 0, 0); | |
1229 | if (ret > 0) { | |
1230 | ret = 0; | |
1231 | goto out; | |
1232 | } | |
1233 | if (ret < 0) | |
1234 | goto out; | |
1235 | ||
1236 | eb = path->nodes[0]; | |
1237 | ref_ptr = btrfs_item_ptr_offset(eb, path->slots[0]); | |
1238 | ref_end = ref_ptr + btrfs_item_size_nr(eb, path->slots[0]); | |
1239 | while (ref_ptr < ref_end) { | |
1240 | char *name = NULL; | |
1241 | int namelen; | |
1242 | u64 parent_id; | |
1243 | ||
1244 | if (key->type == BTRFS_INODE_EXTREF_KEY) { | |
1245 | ret = extref_get_fields(eb, ref_ptr, &namelen, &name, | |
1246 | NULL, &parent_id); | |
1247 | } else { | |
1248 | parent_id = key->offset; | |
1249 | ret = ref_get_fields(eb, ref_ptr, &namelen, &name, | |
1250 | NULL); | |
1251 | } | |
1252 | if (ret) | |
1253 | goto out; | |
1254 | ||
1255 | if (key->type == BTRFS_INODE_EXTREF_KEY) | |
6ff49c6a NB |
1256 | ret = !!btrfs_find_name_in_ext_backref(log_eb, log_slot, |
1257 | parent_id, name, | |
1258 | namelen); | |
1f250e92 | 1259 | else |
9bb8407f NB |
1260 | ret = !!btrfs_find_name_in_backref(log_eb, log_slot, |
1261 | name, namelen); | |
1f250e92 FM |
1262 | |
1263 | if (!ret) { | |
1264 | struct inode *dir; | |
1265 | ||
1266 | btrfs_release_path(path); | |
1267 | dir = read_one_inode(root, parent_id); | |
1268 | if (!dir) { | |
1269 | ret = -ENOENT; | |
1270 | kfree(name); | |
1271 | goto out; | |
1272 | } | |
1273 | ret = btrfs_unlink_inode(trans, root, BTRFS_I(dir), | |
1274 | inode, name, namelen); | |
1275 | kfree(name); | |
1276 | iput(dir); | |
1277 | if (ret) | |
1278 | goto out; | |
1279 | goto again; | |
1280 | } | |
1281 | ||
1282 | kfree(name); | |
1283 | ref_ptr += namelen; | |
1284 | if (key->type == BTRFS_INODE_EXTREF_KEY) | |
1285 | ref_ptr += sizeof(struct btrfs_inode_extref); | |
1286 | else | |
1287 | ref_ptr += sizeof(struct btrfs_inode_ref); | |
1288 | } | |
1289 | ret = 0; | |
1290 | out: | |
1291 | btrfs_release_path(path); | |
1292 | return ret; | |
1293 | } | |
1294 | ||
0d836392 FM |
1295 | static int btrfs_inode_ref_exists(struct inode *inode, struct inode *dir, |
1296 | const u8 ref_type, const char *name, | |
1297 | const int namelen) | |
1298 | { | |
1299 | struct btrfs_key key; | |
1300 | struct btrfs_path *path; | |
1301 | const u64 parent_id = btrfs_ino(BTRFS_I(dir)); | |
1302 | int ret; | |
1303 | ||
1304 | path = btrfs_alloc_path(); | |
1305 | if (!path) | |
1306 | return -ENOMEM; | |
1307 | ||
1308 | key.objectid = btrfs_ino(BTRFS_I(inode)); | |
1309 | key.type = ref_type; | |
1310 | if (key.type == BTRFS_INODE_REF_KEY) | |
1311 | key.offset = parent_id; | |
1312 | else | |
1313 | key.offset = btrfs_extref_hash(parent_id, name, namelen); | |
1314 | ||
1315 | ret = btrfs_search_slot(NULL, BTRFS_I(inode)->root, &key, path, 0, 0); | |
1316 | if (ret < 0) | |
1317 | goto out; | |
1318 | if (ret > 0) { | |
1319 | ret = 0; | |
1320 | goto out; | |
1321 | } | |
1322 | if (key.type == BTRFS_INODE_EXTREF_KEY) | |
6ff49c6a NB |
1323 | ret = !!btrfs_find_name_in_ext_backref(path->nodes[0], |
1324 | path->slots[0], parent_id, name, namelen); | |
0d836392 | 1325 | else |
9bb8407f NB |
1326 | ret = !!btrfs_find_name_in_backref(path->nodes[0], path->slots[0], |
1327 | name, namelen); | |
0d836392 FM |
1328 | |
1329 | out: | |
1330 | btrfs_free_path(path); | |
1331 | return ret; | |
1332 | } | |
1333 | ||
6b5fc433 FM |
1334 | static int add_link(struct btrfs_trans_handle *trans, struct btrfs_root *root, |
1335 | struct inode *dir, struct inode *inode, const char *name, | |
1336 | int namelen, u64 ref_index) | |
1337 | { | |
1338 | struct btrfs_dir_item *dir_item; | |
1339 | struct btrfs_key key; | |
1340 | struct btrfs_path *path; | |
1341 | struct inode *other_inode = NULL; | |
1342 | int ret; | |
1343 | ||
1344 | path = btrfs_alloc_path(); | |
1345 | if (!path) | |
1346 | return -ENOMEM; | |
1347 | ||
1348 | dir_item = btrfs_lookup_dir_item(NULL, root, path, | |
1349 | btrfs_ino(BTRFS_I(dir)), | |
1350 | name, namelen, 0); | |
1351 | if (!dir_item) { | |
1352 | btrfs_release_path(path); | |
1353 | goto add_link; | |
1354 | } else if (IS_ERR(dir_item)) { | |
1355 | ret = PTR_ERR(dir_item); | |
1356 | goto out; | |
1357 | } | |
1358 | ||
1359 | /* | |
1360 | * Our inode's dentry collides with the dentry of another inode which is | |
1361 | * in the log but not yet processed since it has a higher inode number. | |
1362 | * So delete that other dentry. | |
1363 | */ | |
1364 | btrfs_dir_item_key_to_cpu(path->nodes[0], dir_item, &key); | |
1365 | btrfs_release_path(path); | |
1366 | other_inode = read_one_inode(root, key.objectid); | |
1367 | if (!other_inode) { | |
1368 | ret = -ENOENT; | |
1369 | goto out; | |
1370 | } | |
1371 | ret = btrfs_unlink_inode(trans, root, BTRFS_I(dir), BTRFS_I(other_inode), | |
1372 | name, namelen); | |
1373 | if (ret) | |
1374 | goto out; | |
1375 | /* | |
1376 | * If we dropped the link count to 0, bump it so that later the iput() | |
1377 | * on the inode will not free it. We will fixup the link count later. | |
1378 | */ | |
1379 | if (other_inode->i_nlink == 0) | |
1380 | inc_nlink(other_inode); | |
1381 | ||
1382 | ret = btrfs_run_delayed_items(trans); | |
1383 | if (ret) | |
1384 | goto out; | |
1385 | add_link: | |
1386 | ret = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode), | |
1387 | name, namelen, 0, ref_index); | |
1388 | out: | |
1389 | iput(other_inode); | |
1390 | btrfs_free_path(path); | |
1391 | ||
1392 | return ret; | |
1393 | } | |
1394 | ||
5a1d7843 JS |
1395 | /* |
1396 | * replay one inode back reference item found in the log tree. | |
1397 | * eb, slot and key refer to the buffer and key found in the log tree. | |
1398 | * root is the destination we are replaying into, and path is for temp | |
1399 | * use by this function. (it should be released on return). | |
1400 | */ | |
1401 | static noinline int add_inode_ref(struct btrfs_trans_handle *trans, | |
1402 | struct btrfs_root *root, | |
1403 | struct btrfs_root *log, | |
1404 | struct btrfs_path *path, | |
1405 | struct extent_buffer *eb, int slot, | |
1406 | struct btrfs_key *key) | |
1407 | { | |
03b2f08b GB |
1408 | struct inode *dir = NULL; |
1409 | struct inode *inode = NULL; | |
5a1d7843 JS |
1410 | unsigned long ref_ptr; |
1411 | unsigned long ref_end; | |
03b2f08b | 1412 | char *name = NULL; |
5a1d7843 JS |
1413 | int namelen; |
1414 | int ret; | |
1415 | int search_done = 0; | |
f186373f MF |
1416 | int log_ref_ver = 0; |
1417 | u64 parent_objectid; | |
1418 | u64 inode_objectid; | |
f46dbe3d | 1419 | u64 ref_index = 0; |
f186373f MF |
1420 | int ref_struct_size; |
1421 | ||
1422 | ref_ptr = btrfs_item_ptr_offset(eb, slot); | |
1423 | ref_end = ref_ptr + btrfs_item_size_nr(eb, slot); | |
1424 | ||
1425 | if (key->type == BTRFS_INODE_EXTREF_KEY) { | |
1426 | struct btrfs_inode_extref *r; | |
1427 | ||
1428 | ref_struct_size = sizeof(struct btrfs_inode_extref); | |
1429 | log_ref_ver = 1; | |
1430 | r = (struct btrfs_inode_extref *)ref_ptr; | |
1431 | parent_objectid = btrfs_inode_extref_parent(eb, r); | |
1432 | } else { | |
1433 | ref_struct_size = sizeof(struct btrfs_inode_ref); | |
1434 | parent_objectid = key->offset; | |
1435 | } | |
1436 | inode_objectid = key->objectid; | |
e02119d5 | 1437 | |
5a1d7843 JS |
1438 | /* |
1439 | * it is possible that we didn't log all the parent directories | |
1440 | * for a given inode. If we don't find the dir, just don't | |
1441 | * copy the back ref in. The link count fixup code will take | |
1442 | * care of the rest | |
1443 | */ | |
f186373f | 1444 | dir = read_one_inode(root, parent_objectid); |
03b2f08b GB |
1445 | if (!dir) { |
1446 | ret = -ENOENT; | |
1447 | goto out; | |
1448 | } | |
5a1d7843 | 1449 | |
f186373f | 1450 | inode = read_one_inode(root, inode_objectid); |
5a1d7843 | 1451 | if (!inode) { |
03b2f08b GB |
1452 | ret = -EIO; |
1453 | goto out; | |
5a1d7843 JS |
1454 | } |
1455 | ||
5a1d7843 | 1456 | while (ref_ptr < ref_end) { |
f186373f | 1457 | if (log_ref_ver) { |
bae15d95 QW |
1458 | ret = extref_get_fields(eb, ref_ptr, &namelen, &name, |
1459 | &ref_index, &parent_objectid); | |
f186373f MF |
1460 | /* |
1461 | * parent object can change from one array | |
1462 | * item to another. | |
1463 | */ | |
1464 | if (!dir) | |
1465 | dir = read_one_inode(root, parent_objectid); | |
03b2f08b GB |
1466 | if (!dir) { |
1467 | ret = -ENOENT; | |
1468 | goto out; | |
1469 | } | |
f186373f | 1470 | } else { |
bae15d95 QW |
1471 | ret = ref_get_fields(eb, ref_ptr, &namelen, &name, |
1472 | &ref_index); | |
f186373f MF |
1473 | } |
1474 | if (ret) | |
03b2f08b | 1475 | goto out; |
5a1d7843 JS |
1476 | |
1477 | /* if we already have a perfect match, we're done */ | |
f85b7379 DS |
1478 | if (!inode_in_dir(root, path, btrfs_ino(BTRFS_I(dir)), |
1479 | btrfs_ino(BTRFS_I(inode)), ref_index, | |
1480 | name, namelen)) { | |
5a1d7843 JS |
1481 | /* |
1482 | * look for a conflicting back reference in the | |
1483 | * metadata. if we find one we have to unlink that name | |
1484 | * of the file before we add our new link. Later on, we | |
1485 | * overwrite any existing back reference, and we don't | |
1486 | * want to create dangling pointers in the directory. | |
1487 | */ | |
1488 | ||
1489 | if (!search_done) { | |
1490 | ret = __add_inode_ref(trans, root, path, log, | |
94c91a1f | 1491 | BTRFS_I(dir), |
d75eefdf | 1492 | BTRFS_I(inode), |
f186373f MF |
1493 | inode_objectid, |
1494 | parent_objectid, | |
1495 | ref_index, name, namelen, | |
5a1d7843 | 1496 | &search_done); |
03b2f08b GB |
1497 | if (ret) { |
1498 | if (ret == 1) | |
1499 | ret = 0; | |
3650860b JB |
1500 | goto out; |
1501 | } | |
5a1d7843 JS |
1502 | } |
1503 | ||
0d836392 FM |
1504 | /* |
1505 | * If a reference item already exists for this inode | |
1506 | * with the same parent and name, but different index, | |
1507 | * drop it and the corresponding directory index entries | |
1508 | * from the parent before adding the new reference item | |
1509 | * and dir index entries, otherwise we would fail with | |
1510 | * -EEXIST returned from btrfs_add_link() below. | |
1511 | */ | |
1512 | ret = btrfs_inode_ref_exists(inode, dir, key->type, | |
1513 | name, namelen); | |
1514 | if (ret > 0) { | |
1515 | ret = btrfs_unlink_inode(trans, root, | |
1516 | BTRFS_I(dir), | |
1517 | BTRFS_I(inode), | |
1518 | name, namelen); | |
1519 | /* | |
1520 | * If we dropped the link count to 0, bump it so | |
1521 | * that later the iput() on the inode will not | |
1522 | * free it. We will fixup the link count later. | |
1523 | */ | |
1524 | if (!ret && inode->i_nlink == 0) | |
1525 | inc_nlink(inode); | |
1526 | } | |
1527 | if (ret < 0) | |
1528 | goto out; | |
1529 | ||
5a1d7843 | 1530 | /* insert our name */ |
6b5fc433 FM |
1531 | ret = add_link(trans, root, dir, inode, name, namelen, |
1532 | ref_index); | |
3650860b JB |
1533 | if (ret) |
1534 | goto out; | |
5a1d7843 JS |
1535 | |
1536 | btrfs_update_inode(trans, root, inode); | |
1537 | } | |
1538 | ||
f186373f | 1539 | ref_ptr = (unsigned long)(ref_ptr + ref_struct_size) + namelen; |
5a1d7843 | 1540 | kfree(name); |
03b2f08b | 1541 | name = NULL; |
f186373f MF |
1542 | if (log_ref_ver) { |
1543 | iput(dir); | |
1544 | dir = NULL; | |
1545 | } | |
5a1d7843 | 1546 | } |
e02119d5 | 1547 | |
1f250e92 FM |
1548 | /* |
1549 | * Before we overwrite the inode reference item in the subvolume tree | |
1550 | * with the item from the log tree, we must unlink all names from the | |
1551 | * parent directory that are in the subvolume's tree inode reference | |
1552 | * item, otherwise we end up with an inconsistent subvolume tree where | |
1553 | * dir index entries exist for a name but there is no inode reference | |
1554 | * item with the same name. | |
1555 | */ | |
1556 | ret = unlink_old_inode_refs(trans, root, path, BTRFS_I(inode), eb, slot, | |
1557 | key); | |
1558 | if (ret) | |
1559 | goto out; | |
1560 | ||
e02119d5 CM |
1561 | /* finally write the back reference in the inode */ |
1562 | ret = overwrite_item(trans, root, path, eb, slot, key); | |
5a1d7843 | 1563 | out: |
b3b4aa74 | 1564 | btrfs_release_path(path); |
03b2f08b | 1565 | kfree(name); |
e02119d5 CM |
1566 | iput(dir); |
1567 | iput(inode); | |
3650860b | 1568 | return ret; |
e02119d5 CM |
1569 | } |
1570 | ||
c71bf099 | 1571 | static int insert_orphan_item(struct btrfs_trans_handle *trans, |
9c4f61f0 | 1572 | struct btrfs_root *root, u64 ino) |
c71bf099 YZ |
1573 | { |
1574 | int ret; | |
381cf658 | 1575 | |
9c4f61f0 DS |
1576 | ret = btrfs_insert_orphan_item(trans, root, ino); |
1577 | if (ret == -EEXIST) | |
1578 | ret = 0; | |
381cf658 | 1579 | |
c71bf099 YZ |
1580 | return ret; |
1581 | } | |
1582 | ||
f186373f | 1583 | static int count_inode_extrefs(struct btrfs_root *root, |
36283658 | 1584 | struct btrfs_inode *inode, struct btrfs_path *path) |
f186373f MF |
1585 | { |
1586 | int ret = 0; | |
1587 | int name_len; | |
1588 | unsigned int nlink = 0; | |
1589 | u32 item_size; | |
1590 | u32 cur_offset = 0; | |
36283658 | 1591 | u64 inode_objectid = btrfs_ino(inode); |
f186373f MF |
1592 | u64 offset = 0; |
1593 | unsigned long ptr; | |
1594 | struct btrfs_inode_extref *extref; | |
1595 | struct extent_buffer *leaf; | |
1596 | ||
1597 | while (1) { | |
1598 | ret = btrfs_find_one_extref(root, inode_objectid, offset, path, | |
1599 | &extref, &offset); | |
1600 | if (ret) | |
1601 | break; | |
c71bf099 | 1602 | |
f186373f MF |
1603 | leaf = path->nodes[0]; |
1604 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1605 | ptr = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
2c2c452b | 1606 | cur_offset = 0; |
f186373f MF |
1607 | |
1608 | while (cur_offset < item_size) { | |
1609 | extref = (struct btrfs_inode_extref *) (ptr + cur_offset); | |
1610 | name_len = btrfs_inode_extref_name_len(leaf, extref); | |
1611 | ||
1612 | nlink++; | |
1613 | ||
1614 | cur_offset += name_len + sizeof(*extref); | |
1615 | } | |
1616 | ||
1617 | offset++; | |
1618 | btrfs_release_path(path); | |
1619 | } | |
1620 | btrfs_release_path(path); | |
1621 | ||
2c2c452b | 1622 | if (ret < 0 && ret != -ENOENT) |
f186373f MF |
1623 | return ret; |
1624 | return nlink; | |
1625 | } | |
1626 | ||
1627 | static int count_inode_refs(struct btrfs_root *root, | |
f329e319 | 1628 | struct btrfs_inode *inode, struct btrfs_path *path) |
e02119d5 | 1629 | { |
e02119d5 CM |
1630 | int ret; |
1631 | struct btrfs_key key; | |
f186373f | 1632 | unsigned int nlink = 0; |
e02119d5 CM |
1633 | unsigned long ptr; |
1634 | unsigned long ptr_end; | |
1635 | int name_len; | |
f329e319 | 1636 | u64 ino = btrfs_ino(inode); |
e02119d5 | 1637 | |
33345d01 | 1638 | key.objectid = ino; |
e02119d5 CM |
1639 | key.type = BTRFS_INODE_REF_KEY; |
1640 | key.offset = (u64)-1; | |
1641 | ||
d397712b | 1642 | while (1) { |
e02119d5 CM |
1643 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
1644 | if (ret < 0) | |
1645 | break; | |
1646 | if (ret > 0) { | |
1647 | if (path->slots[0] == 0) | |
1648 | break; | |
1649 | path->slots[0]--; | |
1650 | } | |
e93ae26f | 1651 | process_slot: |
e02119d5 CM |
1652 | btrfs_item_key_to_cpu(path->nodes[0], &key, |
1653 | path->slots[0]); | |
33345d01 | 1654 | if (key.objectid != ino || |
e02119d5 CM |
1655 | key.type != BTRFS_INODE_REF_KEY) |
1656 | break; | |
1657 | ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]); | |
1658 | ptr_end = ptr + btrfs_item_size_nr(path->nodes[0], | |
1659 | path->slots[0]); | |
d397712b | 1660 | while (ptr < ptr_end) { |
e02119d5 CM |
1661 | struct btrfs_inode_ref *ref; |
1662 | ||
1663 | ref = (struct btrfs_inode_ref *)ptr; | |
1664 | name_len = btrfs_inode_ref_name_len(path->nodes[0], | |
1665 | ref); | |
1666 | ptr = (unsigned long)(ref + 1) + name_len; | |
1667 | nlink++; | |
1668 | } | |
1669 | ||
1670 | if (key.offset == 0) | |
1671 | break; | |
e93ae26f FDBM |
1672 | if (path->slots[0] > 0) { |
1673 | path->slots[0]--; | |
1674 | goto process_slot; | |
1675 | } | |
e02119d5 | 1676 | key.offset--; |
b3b4aa74 | 1677 | btrfs_release_path(path); |
e02119d5 | 1678 | } |
b3b4aa74 | 1679 | btrfs_release_path(path); |
f186373f MF |
1680 | |
1681 | return nlink; | |
1682 | } | |
1683 | ||
1684 | /* | |
1685 | * There are a few corners where the link count of the file can't | |
1686 | * be properly maintained during replay. So, instead of adding | |
1687 | * lots of complexity to the log code, we just scan the backrefs | |
1688 | * for any file that has been through replay. | |
1689 | * | |
1690 | * The scan will update the link count on the inode to reflect the | |
1691 | * number of back refs found. If it goes down to zero, the iput | |
1692 | * will free the inode. | |
1693 | */ | |
1694 | static noinline int fixup_inode_link_count(struct btrfs_trans_handle *trans, | |
1695 | struct btrfs_root *root, | |
1696 | struct inode *inode) | |
1697 | { | |
1698 | struct btrfs_path *path; | |
1699 | int ret; | |
1700 | u64 nlink = 0; | |
4a0cc7ca | 1701 | u64 ino = btrfs_ino(BTRFS_I(inode)); |
f186373f MF |
1702 | |
1703 | path = btrfs_alloc_path(); | |
1704 | if (!path) | |
1705 | return -ENOMEM; | |
1706 | ||
f329e319 | 1707 | ret = count_inode_refs(root, BTRFS_I(inode), path); |
f186373f MF |
1708 | if (ret < 0) |
1709 | goto out; | |
1710 | ||
1711 | nlink = ret; | |
1712 | ||
36283658 | 1713 | ret = count_inode_extrefs(root, BTRFS_I(inode), path); |
f186373f MF |
1714 | if (ret < 0) |
1715 | goto out; | |
1716 | ||
1717 | nlink += ret; | |
1718 | ||
1719 | ret = 0; | |
1720 | ||
e02119d5 | 1721 | if (nlink != inode->i_nlink) { |
bfe86848 | 1722 | set_nlink(inode, nlink); |
e02119d5 CM |
1723 | btrfs_update_inode(trans, root, inode); |
1724 | } | |
8d5bf1cb | 1725 | BTRFS_I(inode)->index_cnt = (u64)-1; |
e02119d5 | 1726 | |
c71bf099 YZ |
1727 | if (inode->i_nlink == 0) { |
1728 | if (S_ISDIR(inode->i_mode)) { | |
1729 | ret = replay_dir_deletes(trans, root, NULL, path, | |
33345d01 | 1730 | ino, 1); |
3650860b JB |
1731 | if (ret) |
1732 | goto out; | |
c71bf099 | 1733 | } |
33345d01 | 1734 | ret = insert_orphan_item(trans, root, ino); |
12fcfd22 | 1735 | } |
12fcfd22 | 1736 | |
f186373f MF |
1737 | out: |
1738 | btrfs_free_path(path); | |
1739 | return ret; | |
e02119d5 CM |
1740 | } |
1741 | ||
1742 | static noinline int fixup_inode_link_counts(struct btrfs_trans_handle *trans, | |
1743 | struct btrfs_root *root, | |
1744 | struct btrfs_path *path) | |
1745 | { | |
1746 | int ret; | |
1747 | struct btrfs_key key; | |
1748 | struct inode *inode; | |
1749 | ||
1750 | key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID; | |
1751 | key.type = BTRFS_ORPHAN_ITEM_KEY; | |
1752 | key.offset = (u64)-1; | |
d397712b | 1753 | while (1) { |
e02119d5 CM |
1754 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
1755 | if (ret < 0) | |
1756 | break; | |
1757 | ||
1758 | if (ret == 1) { | |
1759 | if (path->slots[0] == 0) | |
1760 | break; | |
1761 | path->slots[0]--; | |
1762 | } | |
1763 | ||
1764 | btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]); | |
1765 | if (key.objectid != BTRFS_TREE_LOG_FIXUP_OBJECTID || | |
1766 | key.type != BTRFS_ORPHAN_ITEM_KEY) | |
1767 | break; | |
1768 | ||
1769 | ret = btrfs_del_item(trans, root, path); | |
65a246c5 TI |
1770 | if (ret) |
1771 | goto out; | |
e02119d5 | 1772 | |
b3b4aa74 | 1773 | btrfs_release_path(path); |
e02119d5 | 1774 | inode = read_one_inode(root, key.offset); |
c00e9493 TI |
1775 | if (!inode) |
1776 | return -EIO; | |
e02119d5 CM |
1777 | |
1778 | ret = fixup_inode_link_count(trans, root, inode); | |
e02119d5 | 1779 | iput(inode); |
3650860b JB |
1780 | if (ret) |
1781 | goto out; | |
e02119d5 | 1782 | |
12fcfd22 CM |
1783 | /* |
1784 | * fixup on a directory may create new entries, | |
1785 | * make sure we always look for the highset possible | |
1786 | * offset | |
1787 | */ | |
1788 | key.offset = (u64)-1; | |
e02119d5 | 1789 | } |
65a246c5 TI |
1790 | ret = 0; |
1791 | out: | |
b3b4aa74 | 1792 | btrfs_release_path(path); |
65a246c5 | 1793 | return ret; |
e02119d5 CM |
1794 | } |
1795 | ||
1796 | ||
1797 | /* | |
1798 | * record a given inode in the fixup dir so we can check its link | |
1799 | * count when replay is done. The link count is incremented here | |
1800 | * so the inode won't go away until we check it | |
1801 | */ | |
1802 | static noinline int link_to_fixup_dir(struct btrfs_trans_handle *trans, | |
1803 | struct btrfs_root *root, | |
1804 | struct btrfs_path *path, | |
1805 | u64 objectid) | |
1806 | { | |
1807 | struct btrfs_key key; | |
1808 | int ret = 0; | |
1809 | struct inode *inode; | |
1810 | ||
1811 | inode = read_one_inode(root, objectid); | |
c00e9493 TI |
1812 | if (!inode) |
1813 | return -EIO; | |
e02119d5 CM |
1814 | |
1815 | key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID; | |
962a298f | 1816 | key.type = BTRFS_ORPHAN_ITEM_KEY; |
e02119d5 CM |
1817 | key.offset = objectid; |
1818 | ||
1819 | ret = btrfs_insert_empty_item(trans, root, path, &key, 0); | |
1820 | ||
b3b4aa74 | 1821 | btrfs_release_path(path); |
e02119d5 | 1822 | if (ret == 0) { |
9bf7a489 JB |
1823 | if (!inode->i_nlink) |
1824 | set_nlink(inode, 1); | |
1825 | else | |
8b558c5f | 1826 | inc_nlink(inode); |
b9959295 | 1827 | ret = btrfs_update_inode(trans, root, inode); |
e02119d5 CM |
1828 | } else if (ret == -EEXIST) { |
1829 | ret = 0; | |
1830 | } else { | |
3650860b | 1831 | BUG(); /* Logic Error */ |
e02119d5 CM |
1832 | } |
1833 | iput(inode); | |
1834 | ||
1835 | return ret; | |
1836 | } | |
1837 | ||
1838 | /* | |
1839 | * when replaying the log for a directory, we only insert names | |
1840 | * for inodes that actually exist. This means an fsync on a directory | |
1841 | * does not implicitly fsync all the new files in it | |
1842 | */ | |
1843 | static noinline int insert_one_name(struct btrfs_trans_handle *trans, | |
1844 | struct btrfs_root *root, | |
e02119d5 | 1845 | u64 dirid, u64 index, |
60d53eb3 | 1846 | char *name, int name_len, |
e02119d5 CM |
1847 | struct btrfs_key *location) |
1848 | { | |
1849 | struct inode *inode; | |
1850 | struct inode *dir; | |
1851 | int ret; | |
1852 | ||
1853 | inode = read_one_inode(root, location->objectid); | |
1854 | if (!inode) | |
1855 | return -ENOENT; | |
1856 | ||
1857 | dir = read_one_inode(root, dirid); | |
1858 | if (!dir) { | |
1859 | iput(inode); | |
1860 | return -EIO; | |
1861 | } | |
d555438b | 1862 | |
db0a669f NB |
1863 | ret = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode), name, |
1864 | name_len, 1, index); | |
e02119d5 CM |
1865 | |
1866 | /* FIXME, put inode into FIXUP list */ | |
1867 | ||
1868 | iput(inode); | |
1869 | iput(dir); | |
1870 | return ret; | |
1871 | } | |
1872 | ||
1873 | /* | |
1874 | * take a single entry in a log directory item and replay it into | |
1875 | * the subvolume. | |
1876 | * | |
1877 | * if a conflicting item exists in the subdirectory already, | |
1878 | * the inode it points to is unlinked and put into the link count | |
1879 | * fix up tree. | |
1880 | * | |
1881 | * If a name from the log points to a file or directory that does | |
1882 | * not exist in the FS, it is skipped. fsyncs on directories | |
1883 | * do not force down inodes inside that directory, just changes to the | |
1884 | * names or unlinks in a directory. | |
bb53eda9 FM |
1885 | * |
1886 | * Returns < 0 on error, 0 if the name wasn't replayed (dentry points to a | |
1887 | * non-existing inode) and 1 if the name was replayed. | |
e02119d5 CM |
1888 | */ |
1889 | static noinline int replay_one_name(struct btrfs_trans_handle *trans, | |
1890 | struct btrfs_root *root, | |
1891 | struct btrfs_path *path, | |
1892 | struct extent_buffer *eb, | |
1893 | struct btrfs_dir_item *di, | |
1894 | struct btrfs_key *key) | |
1895 | { | |
1896 | char *name; | |
1897 | int name_len; | |
1898 | struct btrfs_dir_item *dst_di; | |
1899 | struct btrfs_key found_key; | |
1900 | struct btrfs_key log_key; | |
1901 | struct inode *dir; | |
e02119d5 | 1902 | u8 log_type; |
4bef0848 | 1903 | int exists; |
3650860b | 1904 | int ret = 0; |
d555438b | 1905 | bool update_size = (key->type == BTRFS_DIR_INDEX_KEY); |
bb53eda9 | 1906 | bool name_added = false; |
e02119d5 CM |
1907 | |
1908 | dir = read_one_inode(root, key->objectid); | |
c00e9493 TI |
1909 | if (!dir) |
1910 | return -EIO; | |
e02119d5 CM |
1911 | |
1912 | name_len = btrfs_dir_name_len(eb, di); | |
1913 | name = kmalloc(name_len, GFP_NOFS); | |
2bac325e FDBM |
1914 | if (!name) { |
1915 | ret = -ENOMEM; | |
1916 | goto out; | |
1917 | } | |
2a29edc6 | 1918 | |
e02119d5 CM |
1919 | log_type = btrfs_dir_type(eb, di); |
1920 | read_extent_buffer(eb, name, (unsigned long)(di + 1), | |
1921 | name_len); | |
1922 | ||
1923 | btrfs_dir_item_key_to_cpu(eb, di, &log_key); | |
4bef0848 CM |
1924 | exists = btrfs_lookup_inode(trans, root, path, &log_key, 0); |
1925 | if (exists == 0) | |
1926 | exists = 1; | |
1927 | else | |
1928 | exists = 0; | |
b3b4aa74 | 1929 | btrfs_release_path(path); |
4bef0848 | 1930 | |
e02119d5 CM |
1931 | if (key->type == BTRFS_DIR_ITEM_KEY) { |
1932 | dst_di = btrfs_lookup_dir_item(trans, root, path, key->objectid, | |
1933 | name, name_len, 1); | |
d397712b | 1934 | } else if (key->type == BTRFS_DIR_INDEX_KEY) { |
e02119d5 CM |
1935 | dst_di = btrfs_lookup_dir_index_item(trans, root, path, |
1936 | key->objectid, | |
1937 | key->offset, name, | |
1938 | name_len, 1); | |
1939 | } else { | |
3650860b JB |
1940 | /* Corruption */ |
1941 | ret = -EINVAL; | |
1942 | goto out; | |
e02119d5 | 1943 | } |
c704005d | 1944 | if (IS_ERR_OR_NULL(dst_di)) { |
e02119d5 CM |
1945 | /* we need a sequence number to insert, so we only |
1946 | * do inserts for the BTRFS_DIR_INDEX_KEY types | |
1947 | */ | |
1948 | if (key->type != BTRFS_DIR_INDEX_KEY) | |
1949 | goto out; | |
1950 | goto insert; | |
1951 | } | |
1952 | ||
1953 | btrfs_dir_item_key_to_cpu(path->nodes[0], dst_di, &found_key); | |
1954 | /* the existing item matches the logged item */ | |
1955 | if (found_key.objectid == log_key.objectid && | |
1956 | found_key.type == log_key.type && | |
1957 | found_key.offset == log_key.offset && | |
1958 | btrfs_dir_type(path->nodes[0], dst_di) == log_type) { | |
a2cc11db | 1959 | update_size = false; |
e02119d5 CM |
1960 | goto out; |
1961 | } | |
1962 | ||
1963 | /* | |
1964 | * don't drop the conflicting directory entry if the inode | |
1965 | * for the new entry doesn't exist | |
1966 | */ | |
4bef0848 | 1967 | if (!exists) |
e02119d5 CM |
1968 | goto out; |
1969 | ||
207e7d92 | 1970 | ret = drop_one_dir_item(trans, root, path, BTRFS_I(dir), dst_di); |
3650860b JB |
1971 | if (ret) |
1972 | goto out; | |
e02119d5 CM |
1973 | |
1974 | if (key->type == BTRFS_DIR_INDEX_KEY) | |
1975 | goto insert; | |
1976 | out: | |
b3b4aa74 | 1977 | btrfs_release_path(path); |
d555438b | 1978 | if (!ret && update_size) { |
6ef06d27 | 1979 | btrfs_i_size_write(BTRFS_I(dir), dir->i_size + name_len * 2); |
d555438b JB |
1980 | ret = btrfs_update_inode(trans, root, dir); |
1981 | } | |
e02119d5 CM |
1982 | kfree(name); |
1983 | iput(dir); | |
bb53eda9 FM |
1984 | if (!ret && name_added) |
1985 | ret = 1; | |
3650860b | 1986 | return ret; |
e02119d5 CM |
1987 | |
1988 | insert: | |
725af92a NB |
1989 | /* |
1990 | * Check if the inode reference exists in the log for the given name, | |
1991 | * inode and parent inode | |
1992 | */ | |
1993 | found_key.objectid = log_key.objectid; | |
1994 | found_key.type = BTRFS_INODE_REF_KEY; | |
1995 | found_key.offset = key->objectid; | |
1996 | ret = backref_in_log(root->log_root, &found_key, 0, name, name_len); | |
1997 | if (ret < 0) { | |
1998 | goto out; | |
1999 | } else if (ret) { | |
2000 | /* The dentry will be added later. */ | |
2001 | ret = 0; | |
2002 | update_size = false; | |
2003 | goto out; | |
2004 | } | |
2005 | ||
2006 | found_key.objectid = log_key.objectid; | |
2007 | found_key.type = BTRFS_INODE_EXTREF_KEY; | |
2008 | found_key.offset = key->objectid; | |
2009 | ret = backref_in_log(root->log_root, &found_key, key->objectid, name, | |
2010 | name_len); | |
2011 | if (ret < 0) { | |
2012 | goto out; | |
2013 | } else if (ret) { | |
df8d116f FM |
2014 | /* The dentry will be added later. */ |
2015 | ret = 0; | |
2016 | update_size = false; | |
2017 | goto out; | |
2018 | } | |
b3b4aa74 | 2019 | btrfs_release_path(path); |
60d53eb3 Z |
2020 | ret = insert_one_name(trans, root, key->objectid, key->offset, |
2021 | name, name_len, &log_key); | |
df8d116f | 2022 | if (ret && ret != -ENOENT && ret != -EEXIST) |
3650860b | 2023 | goto out; |
bb53eda9 FM |
2024 | if (!ret) |
2025 | name_added = true; | |
d555438b | 2026 | update_size = false; |
3650860b | 2027 | ret = 0; |
e02119d5 CM |
2028 | goto out; |
2029 | } | |
2030 | ||
2031 | /* | |
2032 | * find all the names in a directory item and reconcile them into | |
2033 | * the subvolume. Only BTRFS_DIR_ITEM_KEY types will have more than | |
2034 | * one name in a directory item, but the same code gets used for | |
2035 | * both directory index types | |
2036 | */ | |
2037 | static noinline int replay_one_dir_item(struct btrfs_trans_handle *trans, | |
2038 | struct btrfs_root *root, | |
2039 | struct btrfs_path *path, | |
2040 | struct extent_buffer *eb, int slot, | |
2041 | struct btrfs_key *key) | |
2042 | { | |
bb53eda9 | 2043 | int ret = 0; |
e02119d5 CM |
2044 | u32 item_size = btrfs_item_size_nr(eb, slot); |
2045 | struct btrfs_dir_item *di; | |
2046 | int name_len; | |
2047 | unsigned long ptr; | |
2048 | unsigned long ptr_end; | |
bb53eda9 | 2049 | struct btrfs_path *fixup_path = NULL; |
e02119d5 CM |
2050 | |
2051 | ptr = btrfs_item_ptr_offset(eb, slot); | |
2052 | ptr_end = ptr + item_size; | |
d397712b | 2053 | while (ptr < ptr_end) { |
e02119d5 CM |
2054 | di = (struct btrfs_dir_item *)ptr; |
2055 | name_len = btrfs_dir_name_len(eb, di); | |
2056 | ret = replay_one_name(trans, root, path, eb, di, key); | |
bb53eda9 FM |
2057 | if (ret < 0) |
2058 | break; | |
e02119d5 CM |
2059 | ptr = (unsigned long)(di + 1); |
2060 | ptr += name_len; | |
bb53eda9 FM |
2061 | |
2062 | /* | |
2063 | * If this entry refers to a non-directory (directories can not | |
2064 | * have a link count > 1) and it was added in the transaction | |
2065 | * that was not committed, make sure we fixup the link count of | |
2066 | * the inode it the entry points to. Otherwise something like | |
2067 | * the following would result in a directory pointing to an | |
2068 | * inode with a wrong link that does not account for this dir | |
2069 | * entry: | |
2070 | * | |
2071 | * mkdir testdir | |
2072 | * touch testdir/foo | |
2073 | * touch testdir/bar | |
2074 | * sync | |
2075 | * | |
2076 | * ln testdir/bar testdir/bar_link | |
2077 | * ln testdir/foo testdir/foo_link | |
2078 | * xfs_io -c "fsync" testdir/bar | |
2079 | * | |
2080 | * <power failure> | |
2081 | * | |
2082 | * mount fs, log replay happens | |
2083 | * | |
2084 | * File foo would remain with a link count of 1 when it has two | |
2085 | * entries pointing to it in the directory testdir. This would | |
2086 | * make it impossible to ever delete the parent directory has | |
2087 | * it would result in stale dentries that can never be deleted. | |
2088 | */ | |
2089 | if (ret == 1 && btrfs_dir_type(eb, di) != BTRFS_FT_DIR) { | |
2090 | struct btrfs_key di_key; | |
2091 | ||
2092 | if (!fixup_path) { | |
2093 | fixup_path = btrfs_alloc_path(); | |
2094 | if (!fixup_path) { | |
2095 | ret = -ENOMEM; | |
2096 | break; | |
2097 | } | |
2098 | } | |
2099 | ||
2100 | btrfs_dir_item_key_to_cpu(eb, di, &di_key); | |
2101 | ret = link_to_fixup_dir(trans, root, fixup_path, | |
2102 | di_key.objectid); | |
2103 | if (ret) | |
2104 | break; | |
2105 | } | |
2106 | ret = 0; | |
e02119d5 | 2107 | } |
bb53eda9 FM |
2108 | btrfs_free_path(fixup_path); |
2109 | return ret; | |
e02119d5 CM |
2110 | } |
2111 | ||
2112 | /* | |
2113 | * directory replay has two parts. There are the standard directory | |
2114 | * items in the log copied from the subvolume, and range items | |
2115 | * created in the log while the subvolume was logged. | |
2116 | * | |
2117 | * The range items tell us which parts of the key space the log | |
2118 | * is authoritative for. During replay, if a key in the subvolume | |
2119 | * directory is in a logged range item, but not actually in the log | |
2120 | * that means it was deleted from the directory before the fsync | |
2121 | * and should be removed. | |
2122 | */ | |
2123 | static noinline int find_dir_range(struct btrfs_root *root, | |
2124 | struct btrfs_path *path, | |
2125 | u64 dirid, int key_type, | |
2126 | u64 *start_ret, u64 *end_ret) | |
2127 | { | |
2128 | struct btrfs_key key; | |
2129 | u64 found_end; | |
2130 | struct btrfs_dir_log_item *item; | |
2131 | int ret; | |
2132 | int nritems; | |
2133 | ||
2134 | if (*start_ret == (u64)-1) | |
2135 | return 1; | |
2136 | ||
2137 | key.objectid = dirid; | |
2138 | key.type = key_type; | |
2139 | key.offset = *start_ret; | |
2140 | ||
2141 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
2142 | if (ret < 0) | |
2143 | goto out; | |
2144 | if (ret > 0) { | |
2145 | if (path->slots[0] == 0) | |
2146 | goto out; | |
2147 | path->slots[0]--; | |
2148 | } | |
2149 | if (ret != 0) | |
2150 | btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]); | |
2151 | ||
2152 | if (key.type != key_type || key.objectid != dirid) { | |
2153 | ret = 1; | |
2154 | goto next; | |
2155 | } | |
2156 | item = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
2157 | struct btrfs_dir_log_item); | |
2158 | found_end = btrfs_dir_log_end(path->nodes[0], item); | |
2159 | ||
2160 | if (*start_ret >= key.offset && *start_ret <= found_end) { | |
2161 | ret = 0; | |
2162 | *start_ret = key.offset; | |
2163 | *end_ret = found_end; | |
2164 | goto out; | |
2165 | } | |
2166 | ret = 1; | |
2167 | next: | |
2168 | /* check the next slot in the tree to see if it is a valid item */ | |
2169 | nritems = btrfs_header_nritems(path->nodes[0]); | |
2a7bf53f | 2170 | path->slots[0]++; |
e02119d5 CM |
2171 | if (path->slots[0] >= nritems) { |
2172 | ret = btrfs_next_leaf(root, path); | |
2173 | if (ret) | |
2174 | goto out; | |
e02119d5 CM |
2175 | } |
2176 | ||
2177 | btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]); | |
2178 | ||
2179 | if (key.type != key_type || key.objectid != dirid) { | |
2180 | ret = 1; | |
2181 | goto out; | |
2182 | } | |
2183 | item = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
2184 | struct btrfs_dir_log_item); | |
2185 | found_end = btrfs_dir_log_end(path->nodes[0], item); | |
2186 | *start_ret = key.offset; | |
2187 | *end_ret = found_end; | |
2188 | ret = 0; | |
2189 | out: | |
b3b4aa74 | 2190 | btrfs_release_path(path); |
e02119d5 CM |
2191 | return ret; |
2192 | } | |
2193 | ||
2194 | /* | |
2195 | * this looks for a given directory item in the log. If the directory | |
2196 | * item is not in the log, the item is removed and the inode it points | |
2197 | * to is unlinked | |
2198 | */ | |
2199 | static noinline int check_item_in_log(struct btrfs_trans_handle *trans, | |
2200 | struct btrfs_root *root, | |
2201 | struct btrfs_root *log, | |
2202 | struct btrfs_path *path, | |
2203 | struct btrfs_path *log_path, | |
2204 | struct inode *dir, | |
2205 | struct btrfs_key *dir_key) | |
2206 | { | |
2207 | int ret; | |
2208 | struct extent_buffer *eb; | |
2209 | int slot; | |
2210 | u32 item_size; | |
2211 | struct btrfs_dir_item *di; | |
2212 | struct btrfs_dir_item *log_di; | |
2213 | int name_len; | |
2214 | unsigned long ptr; | |
2215 | unsigned long ptr_end; | |
2216 | char *name; | |
2217 | struct inode *inode; | |
2218 | struct btrfs_key location; | |
2219 | ||
2220 | again: | |
2221 | eb = path->nodes[0]; | |
2222 | slot = path->slots[0]; | |
2223 | item_size = btrfs_item_size_nr(eb, slot); | |
2224 | ptr = btrfs_item_ptr_offset(eb, slot); | |
2225 | ptr_end = ptr + item_size; | |
d397712b | 2226 | while (ptr < ptr_end) { |
e02119d5 CM |
2227 | di = (struct btrfs_dir_item *)ptr; |
2228 | name_len = btrfs_dir_name_len(eb, di); | |
2229 | name = kmalloc(name_len, GFP_NOFS); | |
2230 | if (!name) { | |
2231 | ret = -ENOMEM; | |
2232 | goto out; | |
2233 | } | |
2234 | read_extent_buffer(eb, name, (unsigned long)(di + 1), | |
2235 | name_len); | |
2236 | log_di = NULL; | |
12fcfd22 | 2237 | if (log && dir_key->type == BTRFS_DIR_ITEM_KEY) { |
e02119d5 CM |
2238 | log_di = btrfs_lookup_dir_item(trans, log, log_path, |
2239 | dir_key->objectid, | |
2240 | name, name_len, 0); | |
12fcfd22 | 2241 | } else if (log && dir_key->type == BTRFS_DIR_INDEX_KEY) { |
e02119d5 CM |
2242 | log_di = btrfs_lookup_dir_index_item(trans, log, |
2243 | log_path, | |
2244 | dir_key->objectid, | |
2245 | dir_key->offset, | |
2246 | name, name_len, 0); | |
2247 | } | |
8d9e220c | 2248 | if (!log_di || log_di == ERR_PTR(-ENOENT)) { |
e02119d5 | 2249 | btrfs_dir_item_key_to_cpu(eb, di, &location); |
b3b4aa74 DS |
2250 | btrfs_release_path(path); |
2251 | btrfs_release_path(log_path); | |
e02119d5 | 2252 | inode = read_one_inode(root, location.objectid); |
c00e9493 TI |
2253 | if (!inode) { |
2254 | kfree(name); | |
2255 | return -EIO; | |
2256 | } | |
e02119d5 CM |
2257 | |
2258 | ret = link_to_fixup_dir(trans, root, | |
2259 | path, location.objectid); | |
3650860b JB |
2260 | if (ret) { |
2261 | kfree(name); | |
2262 | iput(inode); | |
2263 | goto out; | |
2264 | } | |
2265 | ||
8b558c5f | 2266 | inc_nlink(inode); |
4ec5934e NB |
2267 | ret = btrfs_unlink_inode(trans, root, BTRFS_I(dir), |
2268 | BTRFS_I(inode), name, name_len); | |
3650860b | 2269 | if (!ret) |
e5c304e6 | 2270 | ret = btrfs_run_delayed_items(trans); |
e02119d5 CM |
2271 | kfree(name); |
2272 | iput(inode); | |
3650860b JB |
2273 | if (ret) |
2274 | goto out; | |
e02119d5 CM |
2275 | |
2276 | /* there might still be more names under this key | |
2277 | * check and repeat if required | |
2278 | */ | |
2279 | ret = btrfs_search_slot(NULL, root, dir_key, path, | |
2280 | 0, 0); | |
2281 | if (ret == 0) | |
2282 | goto again; | |
2283 | ret = 0; | |
2284 | goto out; | |
269d040f FDBM |
2285 | } else if (IS_ERR(log_di)) { |
2286 | kfree(name); | |
2287 | return PTR_ERR(log_di); | |
e02119d5 | 2288 | } |
b3b4aa74 | 2289 | btrfs_release_path(log_path); |
e02119d5 CM |
2290 | kfree(name); |
2291 | ||
2292 | ptr = (unsigned long)(di + 1); | |
2293 | ptr += name_len; | |
2294 | } | |
2295 | ret = 0; | |
2296 | out: | |
b3b4aa74 DS |
2297 | btrfs_release_path(path); |
2298 | btrfs_release_path(log_path); | |
e02119d5 CM |
2299 | return ret; |
2300 | } | |
2301 | ||
4f764e51 FM |
2302 | static int replay_xattr_deletes(struct btrfs_trans_handle *trans, |
2303 | struct btrfs_root *root, | |
2304 | struct btrfs_root *log, | |
2305 | struct btrfs_path *path, | |
2306 | const u64 ino) | |
2307 | { | |
2308 | struct btrfs_key search_key; | |
2309 | struct btrfs_path *log_path; | |
2310 | int i; | |
2311 | int nritems; | |
2312 | int ret; | |
2313 | ||
2314 | log_path = btrfs_alloc_path(); | |
2315 | if (!log_path) | |
2316 | return -ENOMEM; | |
2317 | ||
2318 | search_key.objectid = ino; | |
2319 | search_key.type = BTRFS_XATTR_ITEM_KEY; | |
2320 | search_key.offset = 0; | |
2321 | again: | |
2322 | ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0); | |
2323 | if (ret < 0) | |
2324 | goto out; | |
2325 | process_leaf: | |
2326 | nritems = btrfs_header_nritems(path->nodes[0]); | |
2327 | for (i = path->slots[0]; i < nritems; i++) { | |
2328 | struct btrfs_key key; | |
2329 | struct btrfs_dir_item *di; | |
2330 | struct btrfs_dir_item *log_di; | |
2331 | u32 total_size; | |
2332 | u32 cur; | |
2333 | ||
2334 | btrfs_item_key_to_cpu(path->nodes[0], &key, i); | |
2335 | if (key.objectid != ino || key.type != BTRFS_XATTR_ITEM_KEY) { | |
2336 | ret = 0; | |
2337 | goto out; | |
2338 | } | |
2339 | ||
2340 | di = btrfs_item_ptr(path->nodes[0], i, struct btrfs_dir_item); | |
2341 | total_size = btrfs_item_size_nr(path->nodes[0], i); | |
2342 | cur = 0; | |
2343 | while (cur < total_size) { | |
2344 | u16 name_len = btrfs_dir_name_len(path->nodes[0], di); | |
2345 | u16 data_len = btrfs_dir_data_len(path->nodes[0], di); | |
2346 | u32 this_len = sizeof(*di) + name_len + data_len; | |
2347 | char *name; | |
2348 | ||
2349 | name = kmalloc(name_len, GFP_NOFS); | |
2350 | if (!name) { | |
2351 | ret = -ENOMEM; | |
2352 | goto out; | |
2353 | } | |
2354 | read_extent_buffer(path->nodes[0], name, | |
2355 | (unsigned long)(di + 1), name_len); | |
2356 | ||
2357 | log_di = btrfs_lookup_xattr(NULL, log, log_path, ino, | |
2358 | name, name_len, 0); | |
2359 | btrfs_release_path(log_path); | |
2360 | if (!log_di) { | |
2361 | /* Doesn't exist in log tree, so delete it. */ | |
2362 | btrfs_release_path(path); | |
2363 | di = btrfs_lookup_xattr(trans, root, path, ino, | |
2364 | name, name_len, -1); | |
2365 | kfree(name); | |
2366 | if (IS_ERR(di)) { | |
2367 | ret = PTR_ERR(di); | |
2368 | goto out; | |
2369 | } | |
2370 | ASSERT(di); | |
2371 | ret = btrfs_delete_one_dir_name(trans, root, | |
2372 | path, di); | |
2373 | if (ret) | |
2374 | goto out; | |
2375 | btrfs_release_path(path); | |
2376 | search_key = key; | |
2377 | goto again; | |
2378 | } | |
2379 | kfree(name); | |
2380 | if (IS_ERR(log_di)) { | |
2381 | ret = PTR_ERR(log_di); | |
2382 | goto out; | |
2383 | } | |
2384 | cur += this_len; | |
2385 | di = (struct btrfs_dir_item *)((char *)di + this_len); | |
2386 | } | |
2387 | } | |
2388 | ret = btrfs_next_leaf(root, path); | |
2389 | if (ret > 0) | |
2390 | ret = 0; | |
2391 | else if (ret == 0) | |
2392 | goto process_leaf; | |
2393 | out: | |
2394 | btrfs_free_path(log_path); | |
2395 | btrfs_release_path(path); | |
2396 | return ret; | |
2397 | } | |
2398 | ||
2399 | ||
e02119d5 CM |
2400 | /* |
2401 | * deletion replay happens before we copy any new directory items | |
2402 | * out of the log or out of backreferences from inodes. It | |
2403 | * scans the log to find ranges of keys that log is authoritative for, | |
2404 | * and then scans the directory to find items in those ranges that are | |
2405 | * not present in the log. | |
2406 | * | |
2407 | * Anything we don't find in the log is unlinked and removed from the | |
2408 | * directory. | |
2409 | */ | |
2410 | static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans, | |
2411 | struct btrfs_root *root, | |
2412 | struct btrfs_root *log, | |
2413 | struct btrfs_path *path, | |
12fcfd22 | 2414 | u64 dirid, int del_all) |
e02119d5 CM |
2415 | { |
2416 | u64 range_start; | |
2417 | u64 range_end; | |
2418 | int key_type = BTRFS_DIR_LOG_ITEM_KEY; | |
2419 | int ret = 0; | |
2420 | struct btrfs_key dir_key; | |
2421 | struct btrfs_key found_key; | |
2422 | struct btrfs_path *log_path; | |
2423 | struct inode *dir; | |
2424 | ||
2425 | dir_key.objectid = dirid; | |
2426 | dir_key.type = BTRFS_DIR_ITEM_KEY; | |
2427 | log_path = btrfs_alloc_path(); | |
2428 | if (!log_path) | |
2429 | return -ENOMEM; | |
2430 | ||
2431 | dir = read_one_inode(root, dirid); | |
2432 | /* it isn't an error if the inode isn't there, that can happen | |
2433 | * because we replay the deletes before we copy in the inode item | |
2434 | * from the log | |
2435 | */ | |
2436 | if (!dir) { | |
2437 | btrfs_free_path(log_path); | |
2438 | return 0; | |
2439 | } | |
2440 | again: | |
2441 | range_start = 0; | |
2442 | range_end = 0; | |
d397712b | 2443 | while (1) { |
12fcfd22 CM |
2444 | if (del_all) |
2445 | range_end = (u64)-1; | |
2446 | else { | |
2447 | ret = find_dir_range(log, path, dirid, key_type, | |
2448 | &range_start, &range_end); | |
2449 | if (ret != 0) | |
2450 | break; | |
2451 | } | |
e02119d5 CM |
2452 | |
2453 | dir_key.offset = range_start; | |
d397712b | 2454 | while (1) { |
e02119d5 CM |
2455 | int nritems; |
2456 | ret = btrfs_search_slot(NULL, root, &dir_key, path, | |
2457 | 0, 0); | |
2458 | if (ret < 0) | |
2459 | goto out; | |
2460 | ||
2461 | nritems = btrfs_header_nritems(path->nodes[0]); | |
2462 | if (path->slots[0] >= nritems) { | |
2463 | ret = btrfs_next_leaf(root, path); | |
b98def7c | 2464 | if (ret == 1) |
e02119d5 | 2465 | break; |
b98def7c LB |
2466 | else if (ret < 0) |
2467 | goto out; | |
e02119d5 CM |
2468 | } |
2469 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, | |
2470 | path->slots[0]); | |
2471 | if (found_key.objectid != dirid || | |
2472 | found_key.type != dir_key.type) | |
2473 | goto next_type; | |
2474 | ||
2475 | if (found_key.offset > range_end) | |
2476 | break; | |
2477 | ||
2478 | ret = check_item_in_log(trans, root, log, path, | |
12fcfd22 CM |
2479 | log_path, dir, |
2480 | &found_key); | |
3650860b JB |
2481 | if (ret) |
2482 | goto out; | |
e02119d5 CM |
2483 | if (found_key.offset == (u64)-1) |
2484 | break; | |
2485 | dir_key.offset = found_key.offset + 1; | |
2486 | } | |
b3b4aa74 | 2487 | btrfs_release_path(path); |
e02119d5 CM |
2488 | if (range_end == (u64)-1) |
2489 | break; | |
2490 | range_start = range_end + 1; | |
2491 | } | |
2492 | ||
2493 | next_type: | |
2494 | ret = 0; | |
2495 | if (key_type == BTRFS_DIR_LOG_ITEM_KEY) { | |
2496 | key_type = BTRFS_DIR_LOG_INDEX_KEY; | |
2497 | dir_key.type = BTRFS_DIR_INDEX_KEY; | |
b3b4aa74 | 2498 | btrfs_release_path(path); |
e02119d5 CM |
2499 | goto again; |
2500 | } | |
2501 | out: | |
b3b4aa74 | 2502 | btrfs_release_path(path); |
e02119d5 CM |
2503 | btrfs_free_path(log_path); |
2504 | iput(dir); | |
2505 | return ret; | |
2506 | } | |
2507 | ||
2508 | /* | |
2509 | * the process_func used to replay items from the log tree. This | |
2510 | * gets called in two different stages. The first stage just looks | |
2511 | * for inodes and makes sure they are all copied into the subvolume. | |
2512 | * | |
2513 | * The second stage copies all the other item types from the log into | |
2514 | * the subvolume. The two stage approach is slower, but gets rid of | |
2515 | * lots of complexity around inodes referencing other inodes that exist | |
2516 | * only in the log (references come from either directory items or inode | |
2517 | * back refs). | |
2518 | */ | |
2519 | static int replay_one_buffer(struct btrfs_root *log, struct extent_buffer *eb, | |
581c1760 | 2520 | struct walk_control *wc, u64 gen, int level) |
e02119d5 CM |
2521 | { |
2522 | int nritems; | |
2523 | struct btrfs_path *path; | |
2524 | struct btrfs_root *root = wc->replay_dest; | |
2525 | struct btrfs_key key; | |
e02119d5 CM |
2526 | int i; |
2527 | int ret; | |
2528 | ||
581c1760 | 2529 | ret = btrfs_read_buffer(eb, gen, level, NULL); |
018642a1 TI |
2530 | if (ret) |
2531 | return ret; | |
e02119d5 CM |
2532 | |
2533 | level = btrfs_header_level(eb); | |
2534 | ||
2535 | if (level != 0) | |
2536 | return 0; | |
2537 | ||
2538 | path = btrfs_alloc_path(); | |
1e5063d0 MF |
2539 | if (!path) |
2540 | return -ENOMEM; | |
e02119d5 CM |
2541 | |
2542 | nritems = btrfs_header_nritems(eb); | |
2543 | for (i = 0; i < nritems; i++) { | |
2544 | btrfs_item_key_to_cpu(eb, &key, i); | |
e02119d5 CM |
2545 | |
2546 | /* inode keys are done during the first stage */ | |
2547 | if (key.type == BTRFS_INODE_ITEM_KEY && | |
2548 | wc->stage == LOG_WALK_REPLAY_INODES) { | |
e02119d5 CM |
2549 | struct btrfs_inode_item *inode_item; |
2550 | u32 mode; | |
2551 | ||
2552 | inode_item = btrfs_item_ptr(eb, i, | |
2553 | struct btrfs_inode_item); | |
f2d72f42 FM |
2554 | /* |
2555 | * If we have a tmpfile (O_TMPFILE) that got fsync'ed | |
2556 | * and never got linked before the fsync, skip it, as | |
2557 | * replaying it is pointless since it would be deleted | |
2558 | * later. We skip logging tmpfiles, but it's always | |
2559 | * possible we are replaying a log created with a kernel | |
2560 | * that used to log tmpfiles. | |
2561 | */ | |
2562 | if (btrfs_inode_nlink(eb, inode_item) == 0) { | |
2563 | wc->ignore_cur_inode = true; | |
2564 | continue; | |
2565 | } else { | |
2566 | wc->ignore_cur_inode = false; | |
2567 | } | |
4f764e51 FM |
2568 | ret = replay_xattr_deletes(wc->trans, root, log, |
2569 | path, key.objectid); | |
2570 | if (ret) | |
2571 | break; | |
e02119d5 CM |
2572 | mode = btrfs_inode_mode(eb, inode_item); |
2573 | if (S_ISDIR(mode)) { | |
2574 | ret = replay_dir_deletes(wc->trans, | |
12fcfd22 | 2575 | root, log, path, key.objectid, 0); |
b50c6e25 JB |
2576 | if (ret) |
2577 | break; | |
e02119d5 CM |
2578 | } |
2579 | ret = overwrite_item(wc->trans, root, path, | |
2580 | eb, i, &key); | |
b50c6e25 JB |
2581 | if (ret) |
2582 | break; | |
e02119d5 | 2583 | |
471d557a FM |
2584 | /* |
2585 | * Before replaying extents, truncate the inode to its | |
2586 | * size. We need to do it now and not after log replay | |
2587 | * because before an fsync we can have prealloc extents | |
2588 | * added beyond the inode's i_size. If we did it after, | |
2589 | * through orphan cleanup for example, we would drop | |
2590 | * those prealloc extents just after replaying them. | |
e02119d5 CM |
2591 | */ |
2592 | if (S_ISREG(mode)) { | |
471d557a FM |
2593 | struct inode *inode; |
2594 | u64 from; | |
2595 | ||
2596 | inode = read_one_inode(root, key.objectid); | |
2597 | if (!inode) { | |
2598 | ret = -EIO; | |
2599 | break; | |
2600 | } | |
2601 | from = ALIGN(i_size_read(inode), | |
2602 | root->fs_info->sectorsize); | |
2603 | ret = btrfs_drop_extents(wc->trans, root, inode, | |
2604 | from, (u64)-1, 1); | |
471d557a | 2605 | if (!ret) { |
f2d72f42 | 2606 | /* Update the inode's nbytes. */ |
471d557a FM |
2607 | ret = btrfs_update_inode(wc->trans, |
2608 | root, inode); | |
2609 | } | |
2610 | iput(inode); | |
b50c6e25 JB |
2611 | if (ret) |
2612 | break; | |
e02119d5 | 2613 | } |
c71bf099 | 2614 | |
e02119d5 CM |
2615 | ret = link_to_fixup_dir(wc->trans, root, |
2616 | path, key.objectid); | |
b50c6e25 JB |
2617 | if (ret) |
2618 | break; | |
e02119d5 | 2619 | } |
dd8e7217 | 2620 | |
f2d72f42 FM |
2621 | if (wc->ignore_cur_inode) |
2622 | continue; | |
2623 | ||
dd8e7217 JB |
2624 | if (key.type == BTRFS_DIR_INDEX_KEY && |
2625 | wc->stage == LOG_WALK_REPLAY_DIR_INDEX) { | |
2626 | ret = replay_one_dir_item(wc->trans, root, path, | |
2627 | eb, i, &key); | |
2628 | if (ret) | |
2629 | break; | |
2630 | } | |
2631 | ||
e02119d5 CM |
2632 | if (wc->stage < LOG_WALK_REPLAY_ALL) |
2633 | continue; | |
2634 | ||
2635 | /* these keys are simply copied */ | |
2636 | if (key.type == BTRFS_XATTR_ITEM_KEY) { | |
2637 | ret = overwrite_item(wc->trans, root, path, | |
2638 | eb, i, &key); | |
b50c6e25 JB |
2639 | if (ret) |
2640 | break; | |
2da1c669 LB |
2641 | } else if (key.type == BTRFS_INODE_REF_KEY || |
2642 | key.type == BTRFS_INODE_EXTREF_KEY) { | |
f186373f MF |
2643 | ret = add_inode_ref(wc->trans, root, log, path, |
2644 | eb, i, &key); | |
b50c6e25 JB |
2645 | if (ret && ret != -ENOENT) |
2646 | break; | |
2647 | ret = 0; | |
e02119d5 CM |
2648 | } else if (key.type == BTRFS_EXTENT_DATA_KEY) { |
2649 | ret = replay_one_extent(wc->trans, root, path, | |
2650 | eb, i, &key); | |
b50c6e25 JB |
2651 | if (ret) |
2652 | break; | |
dd8e7217 | 2653 | } else if (key.type == BTRFS_DIR_ITEM_KEY) { |
e02119d5 CM |
2654 | ret = replay_one_dir_item(wc->trans, root, path, |
2655 | eb, i, &key); | |
b50c6e25 JB |
2656 | if (ret) |
2657 | break; | |
e02119d5 CM |
2658 | } |
2659 | } | |
2660 | btrfs_free_path(path); | |
b50c6e25 | 2661 | return ret; |
e02119d5 CM |
2662 | } |
2663 | ||
6787bb9f NB |
2664 | /* |
2665 | * Correctly adjust the reserved bytes occupied by a log tree extent buffer | |
2666 | */ | |
2667 | static void unaccount_log_buffer(struct btrfs_fs_info *fs_info, u64 start) | |
2668 | { | |
2669 | struct btrfs_block_group *cache; | |
2670 | ||
2671 | cache = btrfs_lookup_block_group(fs_info, start); | |
2672 | if (!cache) { | |
2673 | btrfs_err(fs_info, "unable to find block group for %llu", start); | |
2674 | return; | |
2675 | } | |
2676 | ||
2677 | spin_lock(&cache->space_info->lock); | |
2678 | spin_lock(&cache->lock); | |
2679 | cache->reserved -= fs_info->nodesize; | |
2680 | cache->space_info->bytes_reserved -= fs_info->nodesize; | |
2681 | spin_unlock(&cache->lock); | |
2682 | spin_unlock(&cache->space_info->lock); | |
2683 | ||
2684 | btrfs_put_block_group(cache); | |
2685 | } | |
2686 | ||
d397712b | 2687 | static noinline int walk_down_log_tree(struct btrfs_trans_handle *trans, |
e02119d5 CM |
2688 | struct btrfs_root *root, |
2689 | struct btrfs_path *path, int *level, | |
2690 | struct walk_control *wc) | |
2691 | { | |
0b246afa | 2692 | struct btrfs_fs_info *fs_info = root->fs_info; |
e02119d5 CM |
2693 | u64 bytenr; |
2694 | u64 ptr_gen; | |
2695 | struct extent_buffer *next; | |
2696 | struct extent_buffer *cur; | |
e02119d5 CM |
2697 | u32 blocksize; |
2698 | int ret = 0; | |
2699 | ||
d397712b | 2700 | while (*level > 0) { |
581c1760 QW |
2701 | struct btrfs_key first_key; |
2702 | ||
e02119d5 CM |
2703 | cur = path->nodes[*level]; |
2704 | ||
fae7f21c | 2705 | WARN_ON(btrfs_header_level(cur) != *level); |
e02119d5 CM |
2706 | |
2707 | if (path->slots[*level] >= | |
2708 | btrfs_header_nritems(cur)) | |
2709 | break; | |
2710 | ||
2711 | bytenr = btrfs_node_blockptr(cur, path->slots[*level]); | |
2712 | ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]); | |
581c1760 | 2713 | btrfs_node_key_to_cpu(cur, &first_key, path->slots[*level]); |
0b246afa | 2714 | blocksize = fs_info->nodesize; |
e02119d5 | 2715 | |
2ff7e61e | 2716 | next = btrfs_find_create_tree_block(fs_info, bytenr); |
c871b0f2 LB |
2717 | if (IS_ERR(next)) |
2718 | return PTR_ERR(next); | |
e02119d5 | 2719 | |
e02119d5 | 2720 | if (*level == 1) { |
581c1760 QW |
2721 | ret = wc->process_func(root, next, wc, ptr_gen, |
2722 | *level - 1); | |
b50c6e25 JB |
2723 | if (ret) { |
2724 | free_extent_buffer(next); | |
1e5063d0 | 2725 | return ret; |
b50c6e25 | 2726 | } |
4a500fd1 | 2727 | |
e02119d5 CM |
2728 | path->slots[*level]++; |
2729 | if (wc->free) { | |
581c1760 QW |
2730 | ret = btrfs_read_buffer(next, ptr_gen, |
2731 | *level - 1, &first_key); | |
018642a1 TI |
2732 | if (ret) { |
2733 | free_extent_buffer(next); | |
2734 | return ret; | |
2735 | } | |
e02119d5 | 2736 | |
681ae509 JB |
2737 | if (trans) { |
2738 | btrfs_tree_lock(next); | |
8bead258 | 2739 | btrfs_set_lock_blocking_write(next); |
6a884d7d | 2740 | btrfs_clean_tree_block(next); |
681ae509 JB |
2741 | btrfs_wait_tree_block_writeback(next); |
2742 | btrfs_tree_unlock(next); | |
7bfc1007 | 2743 | ret = btrfs_pin_reserved_extent(trans, |
10e958d5 NB |
2744 | bytenr, blocksize); |
2745 | if (ret) { | |
2746 | free_extent_buffer(next); | |
2747 | return ret; | |
2748 | } | |
1846430c LB |
2749 | } else { |
2750 | if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &next->bflags)) | |
2751 | clear_extent_buffer_dirty(next); | |
10e958d5 | 2752 | unaccount_log_buffer(fs_info, bytenr); |
3650860b | 2753 | } |
e02119d5 CM |
2754 | } |
2755 | free_extent_buffer(next); | |
2756 | continue; | |
2757 | } | |
581c1760 | 2758 | ret = btrfs_read_buffer(next, ptr_gen, *level - 1, &first_key); |
018642a1 TI |
2759 | if (ret) { |
2760 | free_extent_buffer(next); | |
2761 | return ret; | |
2762 | } | |
e02119d5 | 2763 | |
e02119d5 CM |
2764 | if (path->nodes[*level-1]) |
2765 | free_extent_buffer(path->nodes[*level-1]); | |
2766 | path->nodes[*level-1] = next; | |
2767 | *level = btrfs_header_level(next); | |
2768 | path->slots[*level] = 0; | |
2769 | cond_resched(); | |
2770 | } | |
4a500fd1 | 2771 | path->slots[*level] = btrfs_header_nritems(path->nodes[*level]); |
e02119d5 CM |
2772 | |
2773 | cond_resched(); | |
2774 | return 0; | |
2775 | } | |
2776 | ||
d397712b | 2777 | static noinline int walk_up_log_tree(struct btrfs_trans_handle *trans, |
e02119d5 CM |
2778 | struct btrfs_root *root, |
2779 | struct btrfs_path *path, int *level, | |
2780 | struct walk_control *wc) | |
2781 | { | |
0b246afa | 2782 | struct btrfs_fs_info *fs_info = root->fs_info; |
e02119d5 CM |
2783 | int i; |
2784 | int slot; | |
2785 | int ret; | |
2786 | ||
d397712b | 2787 | for (i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) { |
e02119d5 | 2788 | slot = path->slots[i]; |
4a500fd1 | 2789 | if (slot + 1 < btrfs_header_nritems(path->nodes[i])) { |
e02119d5 CM |
2790 | path->slots[i]++; |
2791 | *level = i; | |
2792 | WARN_ON(*level == 0); | |
2793 | return 0; | |
2794 | } else { | |
1e5063d0 | 2795 | ret = wc->process_func(root, path->nodes[*level], wc, |
581c1760 QW |
2796 | btrfs_header_generation(path->nodes[*level]), |
2797 | *level); | |
1e5063d0 MF |
2798 | if (ret) |
2799 | return ret; | |
2800 | ||
e02119d5 CM |
2801 | if (wc->free) { |
2802 | struct extent_buffer *next; | |
2803 | ||
2804 | next = path->nodes[*level]; | |
2805 | ||
681ae509 JB |
2806 | if (trans) { |
2807 | btrfs_tree_lock(next); | |
8bead258 | 2808 | btrfs_set_lock_blocking_write(next); |
6a884d7d | 2809 | btrfs_clean_tree_block(next); |
681ae509 JB |
2810 | btrfs_wait_tree_block_writeback(next); |
2811 | btrfs_tree_unlock(next); | |
7bfc1007 | 2812 | ret = btrfs_pin_reserved_extent(trans, |
10e958d5 NB |
2813 | path->nodes[*level]->start, |
2814 | path->nodes[*level]->len); | |
2815 | if (ret) | |
2816 | return ret; | |
1846430c LB |
2817 | } else { |
2818 | if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &next->bflags)) | |
2819 | clear_extent_buffer_dirty(next); | |
e02119d5 | 2820 | |
10e958d5 NB |
2821 | unaccount_log_buffer(fs_info, |
2822 | path->nodes[*level]->start); | |
2823 | } | |
e02119d5 CM |
2824 | } |
2825 | free_extent_buffer(path->nodes[*level]); | |
2826 | path->nodes[*level] = NULL; | |
2827 | *level = i + 1; | |
2828 | } | |
2829 | } | |
2830 | return 1; | |
2831 | } | |
2832 | ||
2833 | /* | |
2834 | * drop the reference count on the tree rooted at 'snap'. This traverses | |
2835 | * the tree freeing any blocks that have a ref count of zero after being | |
2836 | * decremented. | |
2837 | */ | |
2838 | static int walk_log_tree(struct btrfs_trans_handle *trans, | |
2839 | struct btrfs_root *log, struct walk_control *wc) | |
2840 | { | |
2ff7e61e | 2841 | struct btrfs_fs_info *fs_info = log->fs_info; |
e02119d5 CM |
2842 | int ret = 0; |
2843 | int wret; | |
2844 | int level; | |
2845 | struct btrfs_path *path; | |
e02119d5 CM |
2846 | int orig_level; |
2847 | ||
2848 | path = btrfs_alloc_path(); | |
db5b493a TI |
2849 | if (!path) |
2850 | return -ENOMEM; | |
e02119d5 CM |
2851 | |
2852 | level = btrfs_header_level(log->node); | |
2853 | orig_level = level; | |
2854 | path->nodes[level] = log->node; | |
67439dad | 2855 | atomic_inc(&log->node->refs); |
e02119d5 CM |
2856 | path->slots[level] = 0; |
2857 | ||
d397712b | 2858 | while (1) { |
e02119d5 CM |
2859 | wret = walk_down_log_tree(trans, log, path, &level, wc); |
2860 | if (wret > 0) | |
2861 | break; | |
79787eaa | 2862 | if (wret < 0) { |
e02119d5 | 2863 | ret = wret; |
79787eaa JM |
2864 | goto out; |
2865 | } | |
e02119d5 CM |
2866 | |
2867 | wret = walk_up_log_tree(trans, log, path, &level, wc); | |
2868 | if (wret > 0) | |
2869 | break; | |
79787eaa | 2870 | if (wret < 0) { |
e02119d5 | 2871 | ret = wret; |
79787eaa JM |
2872 | goto out; |
2873 | } | |
e02119d5 CM |
2874 | } |
2875 | ||
2876 | /* was the root node processed? if not, catch it here */ | |
2877 | if (path->nodes[orig_level]) { | |
79787eaa | 2878 | ret = wc->process_func(log, path->nodes[orig_level], wc, |
581c1760 QW |
2879 | btrfs_header_generation(path->nodes[orig_level]), |
2880 | orig_level); | |
79787eaa JM |
2881 | if (ret) |
2882 | goto out; | |
e02119d5 CM |
2883 | if (wc->free) { |
2884 | struct extent_buffer *next; | |
2885 | ||
2886 | next = path->nodes[orig_level]; | |
2887 | ||
681ae509 JB |
2888 | if (trans) { |
2889 | btrfs_tree_lock(next); | |
8bead258 | 2890 | btrfs_set_lock_blocking_write(next); |
6a884d7d | 2891 | btrfs_clean_tree_block(next); |
681ae509 JB |
2892 | btrfs_wait_tree_block_writeback(next); |
2893 | btrfs_tree_unlock(next); | |
7bfc1007 | 2894 | ret = btrfs_pin_reserved_extent(trans, |
10e958d5 NB |
2895 | next->start, next->len); |
2896 | if (ret) | |
2897 | goto out; | |
1846430c LB |
2898 | } else { |
2899 | if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &next->bflags)) | |
2900 | clear_extent_buffer_dirty(next); | |
10e958d5 | 2901 | unaccount_log_buffer(fs_info, next->start); |
681ae509 | 2902 | } |
e02119d5 CM |
2903 | } |
2904 | } | |
2905 | ||
79787eaa | 2906 | out: |
e02119d5 | 2907 | btrfs_free_path(path); |
e02119d5 CM |
2908 | return ret; |
2909 | } | |
2910 | ||
7237f183 YZ |
2911 | /* |
2912 | * helper function to update the item for a given subvolumes log root | |
2913 | * in the tree of log roots | |
2914 | */ | |
2915 | static int update_log_root(struct btrfs_trans_handle *trans, | |
4203e968 JB |
2916 | struct btrfs_root *log, |
2917 | struct btrfs_root_item *root_item) | |
7237f183 | 2918 | { |
0b246afa | 2919 | struct btrfs_fs_info *fs_info = log->fs_info; |
7237f183 YZ |
2920 | int ret; |
2921 | ||
2922 | if (log->log_transid == 1) { | |
2923 | /* insert root item on the first sync */ | |
0b246afa | 2924 | ret = btrfs_insert_root(trans, fs_info->log_root_tree, |
4203e968 | 2925 | &log->root_key, root_item); |
7237f183 | 2926 | } else { |
0b246afa | 2927 | ret = btrfs_update_root(trans, fs_info->log_root_tree, |
4203e968 | 2928 | &log->root_key, root_item); |
7237f183 YZ |
2929 | } |
2930 | return ret; | |
2931 | } | |
2932 | ||
60d53eb3 | 2933 | static void wait_log_commit(struct btrfs_root *root, int transid) |
e02119d5 CM |
2934 | { |
2935 | DEFINE_WAIT(wait); | |
7237f183 | 2936 | int index = transid % 2; |
e02119d5 | 2937 | |
7237f183 YZ |
2938 | /* |
2939 | * we only allow two pending log transactions at a time, | |
2940 | * so we know that if ours is more than 2 older than the | |
2941 | * current transaction, we're done | |
2942 | */ | |
49e83f57 | 2943 | for (;;) { |
7237f183 YZ |
2944 | prepare_to_wait(&root->log_commit_wait[index], |
2945 | &wait, TASK_UNINTERRUPTIBLE); | |
12fcfd22 | 2946 | |
49e83f57 LB |
2947 | if (!(root->log_transid_committed < transid && |
2948 | atomic_read(&root->log_commit[index]))) | |
2949 | break; | |
12fcfd22 | 2950 | |
49e83f57 LB |
2951 | mutex_unlock(&root->log_mutex); |
2952 | schedule(); | |
7237f183 | 2953 | mutex_lock(&root->log_mutex); |
49e83f57 LB |
2954 | } |
2955 | finish_wait(&root->log_commit_wait[index], &wait); | |
7237f183 YZ |
2956 | } |
2957 | ||
60d53eb3 | 2958 | static void wait_for_writer(struct btrfs_root *root) |
7237f183 YZ |
2959 | { |
2960 | DEFINE_WAIT(wait); | |
8b050d35 | 2961 | |
49e83f57 LB |
2962 | for (;;) { |
2963 | prepare_to_wait(&root->log_writer_wait, &wait, | |
2964 | TASK_UNINTERRUPTIBLE); | |
2965 | if (!atomic_read(&root->log_writers)) | |
2966 | break; | |
2967 | ||
7237f183 | 2968 | mutex_unlock(&root->log_mutex); |
49e83f57 | 2969 | schedule(); |
575849ec | 2970 | mutex_lock(&root->log_mutex); |
7237f183 | 2971 | } |
49e83f57 | 2972 | finish_wait(&root->log_writer_wait, &wait); |
e02119d5 CM |
2973 | } |
2974 | ||
8b050d35 MX |
2975 | static inline void btrfs_remove_log_ctx(struct btrfs_root *root, |
2976 | struct btrfs_log_ctx *ctx) | |
2977 | { | |
2978 | if (!ctx) | |
2979 | return; | |
2980 | ||
2981 | mutex_lock(&root->log_mutex); | |
2982 | list_del_init(&ctx->list); | |
2983 | mutex_unlock(&root->log_mutex); | |
2984 | } | |
2985 | ||
2986 | /* | |
2987 | * Invoked in log mutex context, or be sure there is no other task which | |
2988 | * can access the list. | |
2989 | */ | |
2990 | static inline void btrfs_remove_all_log_ctxs(struct btrfs_root *root, | |
2991 | int index, int error) | |
2992 | { | |
2993 | struct btrfs_log_ctx *ctx; | |
570dd450 | 2994 | struct btrfs_log_ctx *safe; |
8b050d35 | 2995 | |
570dd450 CM |
2996 | list_for_each_entry_safe(ctx, safe, &root->log_ctxs[index], list) { |
2997 | list_del_init(&ctx->list); | |
8b050d35 | 2998 | ctx->log_ret = error; |
570dd450 | 2999 | } |
8b050d35 MX |
3000 | |
3001 | INIT_LIST_HEAD(&root->log_ctxs[index]); | |
3002 | } | |
3003 | ||
e02119d5 CM |
3004 | /* |
3005 | * btrfs_sync_log does sends a given tree log down to the disk and | |
3006 | * updates the super blocks to record it. When this call is done, | |
12fcfd22 CM |
3007 | * you know that any inodes previously logged are safely on disk only |
3008 | * if it returns 0. | |
3009 | * | |
3010 | * Any other return value means you need to call btrfs_commit_transaction. | |
3011 | * Some of the edge cases for fsyncing directories that have had unlinks | |
3012 | * or renames done in the past mean that sometimes the only safe | |
3013 | * fsync is to commit the whole FS. When btrfs_sync_log returns -EAGAIN, | |
3014 | * that has happened. | |
e02119d5 CM |
3015 | */ |
3016 | int btrfs_sync_log(struct btrfs_trans_handle *trans, | |
8b050d35 | 3017 | struct btrfs_root *root, struct btrfs_log_ctx *ctx) |
e02119d5 | 3018 | { |
7237f183 YZ |
3019 | int index1; |
3020 | int index2; | |
8cef4e16 | 3021 | int mark; |
e02119d5 | 3022 | int ret; |
0b246afa | 3023 | struct btrfs_fs_info *fs_info = root->fs_info; |
e02119d5 | 3024 | struct btrfs_root *log = root->log_root; |
0b246afa | 3025 | struct btrfs_root *log_root_tree = fs_info->log_root_tree; |
4203e968 | 3026 | struct btrfs_root_item new_root_item; |
bb14a59b | 3027 | int log_transid = 0; |
8b050d35 | 3028 | struct btrfs_log_ctx root_log_ctx; |
c6adc9cc | 3029 | struct blk_plug plug; |
e02119d5 | 3030 | |
7237f183 | 3031 | mutex_lock(&root->log_mutex); |
d1433deb MX |
3032 | log_transid = ctx->log_transid; |
3033 | if (root->log_transid_committed >= log_transid) { | |
3034 | mutex_unlock(&root->log_mutex); | |
3035 | return ctx->log_ret; | |
3036 | } | |
3037 | ||
3038 | index1 = log_transid % 2; | |
7237f183 | 3039 | if (atomic_read(&root->log_commit[index1])) { |
60d53eb3 | 3040 | wait_log_commit(root, log_transid); |
7237f183 | 3041 | mutex_unlock(&root->log_mutex); |
8b050d35 | 3042 | return ctx->log_ret; |
e02119d5 | 3043 | } |
d1433deb | 3044 | ASSERT(log_transid == root->log_transid); |
7237f183 YZ |
3045 | atomic_set(&root->log_commit[index1], 1); |
3046 | ||
3047 | /* wait for previous tree log sync to complete */ | |
3048 | if (atomic_read(&root->log_commit[(index1 + 1) % 2])) | |
60d53eb3 | 3049 | wait_log_commit(root, log_transid - 1); |
48cab2e0 | 3050 | |
86df7eb9 | 3051 | while (1) { |
2ecb7923 | 3052 | int batch = atomic_read(&root->log_batch); |
cd354ad6 | 3053 | /* when we're on an ssd, just kick the log commit out */ |
0b246afa | 3054 | if (!btrfs_test_opt(fs_info, SSD) && |
27cdeb70 | 3055 | test_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state)) { |
86df7eb9 YZ |
3056 | mutex_unlock(&root->log_mutex); |
3057 | schedule_timeout_uninterruptible(1); | |
3058 | mutex_lock(&root->log_mutex); | |
3059 | } | |
60d53eb3 | 3060 | wait_for_writer(root); |
2ecb7923 | 3061 | if (batch == atomic_read(&root->log_batch)) |
e02119d5 CM |
3062 | break; |
3063 | } | |
e02119d5 | 3064 | |
12fcfd22 | 3065 | /* bail out if we need to do a full commit */ |
4884b8e8 | 3066 | if (btrfs_need_log_full_commit(trans)) { |
12fcfd22 CM |
3067 | ret = -EAGAIN; |
3068 | mutex_unlock(&root->log_mutex); | |
3069 | goto out; | |
3070 | } | |
3071 | ||
8cef4e16 YZ |
3072 | if (log_transid % 2 == 0) |
3073 | mark = EXTENT_DIRTY; | |
3074 | else | |
3075 | mark = EXTENT_NEW; | |
3076 | ||
690587d1 CM |
3077 | /* we start IO on all the marked extents here, but we don't actually |
3078 | * wait for them until later. | |
3079 | */ | |
c6adc9cc | 3080 | blk_start_plug(&plug); |
2ff7e61e | 3081 | ret = btrfs_write_marked_extents(fs_info, &log->dirty_log_pages, mark); |
79787eaa | 3082 | if (ret) { |
c6adc9cc | 3083 | blk_finish_plug(&plug); |
66642832 | 3084 | btrfs_abort_transaction(trans, ret); |
90787766 | 3085 | btrfs_set_log_full_commit(trans); |
79787eaa JM |
3086 | mutex_unlock(&root->log_mutex); |
3087 | goto out; | |
3088 | } | |
7237f183 | 3089 | |
4203e968 JB |
3090 | /* |
3091 | * We _must_ update under the root->log_mutex in order to make sure we | |
3092 | * have a consistent view of the log root we are trying to commit at | |
3093 | * this moment. | |
3094 | * | |
3095 | * We _must_ copy this into a local copy, because we are not holding the | |
3096 | * log_root_tree->log_mutex yet. This is important because when we | |
3097 | * commit the log_root_tree we must have a consistent view of the | |
3098 | * log_root_tree when we update the super block to point at the | |
3099 | * log_root_tree bytenr. If we update the log_root_tree here we'll race | |
3100 | * with the commit and possibly point at the new block which we may not | |
3101 | * have written out. | |
3102 | */ | |
5d4f98a2 | 3103 | btrfs_set_root_node(&log->root_item, log->node); |
4203e968 | 3104 | memcpy(&new_root_item, &log->root_item, sizeof(new_root_item)); |
7237f183 | 3105 | |
7237f183 YZ |
3106 | root->log_transid++; |
3107 | log->log_transid = root->log_transid; | |
ff782e0a | 3108 | root->log_start_pid = 0; |
7237f183 | 3109 | /* |
8cef4e16 YZ |
3110 | * IO has been started, blocks of the log tree have WRITTEN flag set |
3111 | * in their headers. new modifications of the log will be written to | |
3112 | * new positions. so it's safe to allow log writers to go in. | |
7237f183 YZ |
3113 | */ |
3114 | mutex_unlock(&root->log_mutex); | |
3115 | ||
28a23593 | 3116 | btrfs_init_log_ctx(&root_log_ctx, NULL); |
d1433deb | 3117 | |
7237f183 | 3118 | mutex_lock(&log_root_tree->log_mutex); |
d1433deb MX |
3119 | |
3120 | index2 = log_root_tree->log_transid % 2; | |
3121 | list_add_tail(&root_log_ctx.list, &log_root_tree->log_ctxs[index2]); | |
3122 | root_log_ctx.log_transid = log_root_tree->log_transid; | |
3123 | ||
4203e968 JB |
3124 | /* |
3125 | * Now we are safe to update the log_root_tree because we're under the | |
3126 | * log_mutex, and we're a current writer so we're holding the commit | |
3127 | * open until we drop the log_mutex. | |
3128 | */ | |
3129 | ret = update_log_root(trans, log, &new_root_item); | |
4a500fd1 | 3130 | if (ret) { |
d1433deb MX |
3131 | if (!list_empty(&root_log_ctx.list)) |
3132 | list_del_init(&root_log_ctx.list); | |
3133 | ||
c6adc9cc | 3134 | blk_finish_plug(&plug); |
90787766 | 3135 | btrfs_set_log_full_commit(trans); |
995946dd | 3136 | |
79787eaa | 3137 | if (ret != -ENOSPC) { |
66642832 | 3138 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
3139 | mutex_unlock(&log_root_tree->log_mutex); |
3140 | goto out; | |
3141 | } | |
bf89d38f | 3142 | btrfs_wait_tree_log_extents(log, mark); |
4a500fd1 YZ |
3143 | mutex_unlock(&log_root_tree->log_mutex); |
3144 | ret = -EAGAIN; | |
3145 | goto out; | |
3146 | } | |
3147 | ||
d1433deb | 3148 | if (log_root_tree->log_transid_committed >= root_log_ctx.log_transid) { |
3da5ab56 | 3149 | blk_finish_plug(&plug); |
cbd60aa7 | 3150 | list_del_init(&root_log_ctx.list); |
d1433deb MX |
3151 | mutex_unlock(&log_root_tree->log_mutex); |
3152 | ret = root_log_ctx.log_ret; | |
3153 | goto out; | |
3154 | } | |
8b050d35 | 3155 | |
d1433deb | 3156 | index2 = root_log_ctx.log_transid % 2; |
7237f183 | 3157 | if (atomic_read(&log_root_tree->log_commit[index2])) { |
c6adc9cc | 3158 | blk_finish_plug(&plug); |
bf89d38f | 3159 | ret = btrfs_wait_tree_log_extents(log, mark); |
60d53eb3 | 3160 | wait_log_commit(log_root_tree, |
d1433deb | 3161 | root_log_ctx.log_transid); |
7237f183 | 3162 | mutex_unlock(&log_root_tree->log_mutex); |
5ab5e44a FM |
3163 | if (!ret) |
3164 | ret = root_log_ctx.log_ret; | |
7237f183 YZ |
3165 | goto out; |
3166 | } | |
d1433deb | 3167 | ASSERT(root_log_ctx.log_transid == log_root_tree->log_transid); |
7237f183 YZ |
3168 | atomic_set(&log_root_tree->log_commit[index2], 1); |
3169 | ||
12fcfd22 | 3170 | if (atomic_read(&log_root_tree->log_commit[(index2 + 1) % 2])) { |
60d53eb3 | 3171 | wait_log_commit(log_root_tree, |
d1433deb | 3172 | root_log_ctx.log_transid - 1); |
12fcfd22 CM |
3173 | } |
3174 | ||
12fcfd22 CM |
3175 | /* |
3176 | * now that we've moved on to the tree of log tree roots, | |
3177 | * check the full commit flag again | |
3178 | */ | |
4884b8e8 | 3179 | if (btrfs_need_log_full_commit(trans)) { |
c6adc9cc | 3180 | blk_finish_plug(&plug); |
bf89d38f | 3181 | btrfs_wait_tree_log_extents(log, mark); |
12fcfd22 CM |
3182 | mutex_unlock(&log_root_tree->log_mutex); |
3183 | ret = -EAGAIN; | |
3184 | goto out_wake_log_root; | |
3185 | } | |
7237f183 | 3186 | |
2ff7e61e | 3187 | ret = btrfs_write_marked_extents(fs_info, |
c6adc9cc MX |
3188 | &log_root_tree->dirty_log_pages, |
3189 | EXTENT_DIRTY | EXTENT_NEW); | |
3190 | blk_finish_plug(&plug); | |
79787eaa | 3191 | if (ret) { |
90787766 | 3192 | btrfs_set_log_full_commit(trans); |
66642832 | 3193 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
3194 | mutex_unlock(&log_root_tree->log_mutex); |
3195 | goto out_wake_log_root; | |
3196 | } | |
bf89d38f | 3197 | ret = btrfs_wait_tree_log_extents(log, mark); |
5ab5e44a | 3198 | if (!ret) |
bf89d38f JM |
3199 | ret = btrfs_wait_tree_log_extents(log_root_tree, |
3200 | EXTENT_NEW | EXTENT_DIRTY); | |
5ab5e44a | 3201 | if (ret) { |
90787766 | 3202 | btrfs_set_log_full_commit(trans); |
5ab5e44a FM |
3203 | mutex_unlock(&log_root_tree->log_mutex); |
3204 | goto out_wake_log_root; | |
3205 | } | |
e02119d5 | 3206 | |
0b246afa JM |
3207 | btrfs_set_super_log_root(fs_info->super_for_commit, |
3208 | log_root_tree->node->start); | |
3209 | btrfs_set_super_log_root_level(fs_info->super_for_commit, | |
3210 | btrfs_header_level(log_root_tree->node)); | |
e02119d5 | 3211 | |
7237f183 | 3212 | log_root_tree->log_transid++; |
7237f183 YZ |
3213 | mutex_unlock(&log_root_tree->log_mutex); |
3214 | ||
3215 | /* | |
52042d8e | 3216 | * Nobody else is going to jump in and write the ctree |
7237f183 YZ |
3217 | * super here because the log_commit atomic below is protecting |
3218 | * us. We must be called with a transaction handle pinning | |
3219 | * the running transaction open, so a full commit can't hop | |
3220 | * in and cause problems either. | |
3221 | */ | |
eece6a9c | 3222 | ret = write_all_supers(fs_info, 1); |
5af3e8cc | 3223 | if (ret) { |
90787766 | 3224 | btrfs_set_log_full_commit(trans); |
66642832 | 3225 | btrfs_abort_transaction(trans, ret); |
5af3e8cc SB |
3226 | goto out_wake_log_root; |
3227 | } | |
7237f183 | 3228 | |
257c62e1 CM |
3229 | mutex_lock(&root->log_mutex); |
3230 | if (root->last_log_commit < log_transid) | |
3231 | root->last_log_commit = log_transid; | |
3232 | mutex_unlock(&root->log_mutex); | |
3233 | ||
12fcfd22 | 3234 | out_wake_log_root: |
570dd450 | 3235 | mutex_lock(&log_root_tree->log_mutex); |
8b050d35 MX |
3236 | btrfs_remove_all_log_ctxs(log_root_tree, index2, ret); |
3237 | ||
d1433deb | 3238 | log_root_tree->log_transid_committed++; |
7237f183 | 3239 | atomic_set(&log_root_tree->log_commit[index2], 0); |
d1433deb MX |
3240 | mutex_unlock(&log_root_tree->log_mutex); |
3241 | ||
33a9eca7 | 3242 | /* |
093258e6 DS |
3243 | * The barrier before waitqueue_active (in cond_wake_up) is needed so |
3244 | * all the updates above are seen by the woken threads. It might not be | |
3245 | * necessary, but proving that seems to be hard. | |
33a9eca7 | 3246 | */ |
093258e6 | 3247 | cond_wake_up(&log_root_tree->log_commit_wait[index2]); |
e02119d5 | 3248 | out: |
d1433deb | 3249 | mutex_lock(&root->log_mutex); |
570dd450 | 3250 | btrfs_remove_all_log_ctxs(root, index1, ret); |
d1433deb | 3251 | root->log_transid_committed++; |
7237f183 | 3252 | atomic_set(&root->log_commit[index1], 0); |
d1433deb | 3253 | mutex_unlock(&root->log_mutex); |
8b050d35 | 3254 | |
33a9eca7 | 3255 | /* |
093258e6 DS |
3256 | * The barrier before waitqueue_active (in cond_wake_up) is needed so |
3257 | * all the updates above are seen by the woken threads. It might not be | |
3258 | * necessary, but proving that seems to be hard. | |
33a9eca7 | 3259 | */ |
093258e6 | 3260 | cond_wake_up(&root->log_commit_wait[index1]); |
b31eabd8 | 3261 | return ret; |
e02119d5 CM |
3262 | } |
3263 | ||
4a500fd1 YZ |
3264 | static void free_log_tree(struct btrfs_trans_handle *trans, |
3265 | struct btrfs_root *log) | |
e02119d5 CM |
3266 | { |
3267 | int ret; | |
e02119d5 CM |
3268 | struct walk_control wc = { |
3269 | .free = 1, | |
3270 | .process_func = process_one_buffer | |
3271 | }; | |
3272 | ||
681ae509 | 3273 | ret = walk_log_tree(trans, log, &wc); |
374b0e2d JM |
3274 | if (ret) { |
3275 | if (trans) | |
3276 | btrfs_abort_transaction(trans, ret); | |
3277 | else | |
3278 | btrfs_handle_fs_error(log->fs_info, ret, NULL); | |
3279 | } | |
e02119d5 | 3280 | |
59b0713a FM |
3281 | clear_extent_bits(&log->dirty_log_pages, 0, (u64)-1, |
3282 | EXTENT_DIRTY | EXTENT_NEW | EXTENT_NEED_WAIT); | |
e289f03e | 3283 | extent_io_tree_release(&log->log_csum_range); |
00246528 | 3284 | btrfs_put_root(log); |
4a500fd1 YZ |
3285 | } |
3286 | ||
3287 | /* | |
3288 | * free all the extents used by the tree log. This should be called | |
3289 | * at commit time of the full transaction | |
3290 | */ | |
3291 | int btrfs_free_log(struct btrfs_trans_handle *trans, struct btrfs_root *root) | |
3292 | { | |
3293 | if (root->log_root) { | |
3294 | free_log_tree(trans, root->log_root); | |
3295 | root->log_root = NULL; | |
e7a79811 | 3296 | clear_bit(BTRFS_ROOT_HAS_LOG_TREE, &root->state); |
4a500fd1 YZ |
3297 | } |
3298 | return 0; | |
3299 | } | |
3300 | ||
3301 | int btrfs_free_log_root_tree(struct btrfs_trans_handle *trans, | |
3302 | struct btrfs_fs_info *fs_info) | |
3303 | { | |
3304 | if (fs_info->log_root_tree) { | |
3305 | free_log_tree(trans, fs_info->log_root_tree); | |
3306 | fs_info->log_root_tree = NULL; | |
3307 | } | |
e02119d5 CM |
3308 | return 0; |
3309 | } | |
3310 | ||
803f0f64 FM |
3311 | /* |
3312 | * Check if an inode was logged in the current transaction. We can't always rely | |
3313 | * on an inode's logged_trans value, because it's an in-memory only field and | |
3314 | * therefore not persisted. This means that its value is lost if the inode gets | |
3315 | * evicted and loaded again from disk (in which case it has a value of 0, and | |
3316 | * certainly it is smaller then any possible transaction ID), when that happens | |
3317 | * the full_sync flag is set in the inode's runtime flags, so on that case we | |
3318 | * assume eviction happened and ignore the logged_trans value, assuming the | |
3319 | * worst case, that the inode was logged before in the current transaction. | |
3320 | */ | |
3321 | static bool inode_logged(struct btrfs_trans_handle *trans, | |
3322 | struct btrfs_inode *inode) | |
3323 | { | |
3324 | if (inode->logged_trans == trans->transid) | |
3325 | return true; | |
3326 | ||
3327 | if (inode->last_trans == trans->transid && | |
3328 | test_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &inode->runtime_flags) && | |
3329 | !test_bit(BTRFS_FS_LOG_RECOVERING, &trans->fs_info->flags)) | |
3330 | return true; | |
3331 | ||
3332 | return false; | |
3333 | } | |
3334 | ||
e02119d5 CM |
3335 | /* |
3336 | * If both a file and directory are logged, and unlinks or renames are | |
3337 | * mixed in, we have a few interesting corners: | |
3338 | * | |
3339 | * create file X in dir Y | |
3340 | * link file X to X.link in dir Y | |
3341 | * fsync file X | |
3342 | * unlink file X but leave X.link | |
3343 | * fsync dir Y | |
3344 | * | |
3345 | * After a crash we would expect only X.link to exist. But file X | |
3346 | * didn't get fsync'd again so the log has back refs for X and X.link. | |
3347 | * | |
3348 | * We solve this by removing directory entries and inode backrefs from the | |
3349 | * log when a file that was logged in the current transaction is | |
3350 | * unlinked. Any later fsync will include the updated log entries, and | |
3351 | * we'll be able to reconstruct the proper directory items from backrefs. | |
3352 | * | |
3353 | * This optimizations allows us to avoid relogging the entire inode | |
3354 | * or the entire directory. | |
3355 | */ | |
3356 | int btrfs_del_dir_entries_in_log(struct btrfs_trans_handle *trans, | |
3357 | struct btrfs_root *root, | |
3358 | const char *name, int name_len, | |
49f34d1f | 3359 | struct btrfs_inode *dir, u64 index) |
e02119d5 CM |
3360 | { |
3361 | struct btrfs_root *log; | |
3362 | struct btrfs_dir_item *di; | |
3363 | struct btrfs_path *path; | |
3364 | int ret; | |
4a500fd1 | 3365 | int err = 0; |
e02119d5 | 3366 | int bytes_del = 0; |
49f34d1f | 3367 | u64 dir_ino = btrfs_ino(dir); |
e02119d5 | 3368 | |
803f0f64 | 3369 | if (!inode_logged(trans, dir)) |
3a5f1d45 CM |
3370 | return 0; |
3371 | ||
e02119d5 CM |
3372 | ret = join_running_log_trans(root); |
3373 | if (ret) | |
3374 | return 0; | |
3375 | ||
49f34d1f | 3376 | mutex_lock(&dir->log_mutex); |
e02119d5 CM |
3377 | |
3378 | log = root->log_root; | |
3379 | path = btrfs_alloc_path(); | |
a62f44a5 TI |
3380 | if (!path) { |
3381 | err = -ENOMEM; | |
3382 | goto out_unlock; | |
3383 | } | |
2a29edc6 | 3384 | |
33345d01 | 3385 | di = btrfs_lookup_dir_item(trans, log, path, dir_ino, |
e02119d5 | 3386 | name, name_len, -1); |
4a500fd1 YZ |
3387 | if (IS_ERR(di)) { |
3388 | err = PTR_ERR(di); | |
3389 | goto fail; | |
3390 | } | |
3391 | if (di) { | |
e02119d5 CM |
3392 | ret = btrfs_delete_one_dir_name(trans, log, path, di); |
3393 | bytes_del += name_len; | |
3650860b JB |
3394 | if (ret) { |
3395 | err = ret; | |
3396 | goto fail; | |
3397 | } | |
e02119d5 | 3398 | } |
b3b4aa74 | 3399 | btrfs_release_path(path); |
33345d01 | 3400 | di = btrfs_lookup_dir_index_item(trans, log, path, dir_ino, |
e02119d5 | 3401 | index, name, name_len, -1); |
4a500fd1 YZ |
3402 | if (IS_ERR(di)) { |
3403 | err = PTR_ERR(di); | |
3404 | goto fail; | |
3405 | } | |
3406 | if (di) { | |
e02119d5 CM |
3407 | ret = btrfs_delete_one_dir_name(trans, log, path, di); |
3408 | bytes_del += name_len; | |
3650860b JB |
3409 | if (ret) { |
3410 | err = ret; | |
3411 | goto fail; | |
3412 | } | |
e02119d5 CM |
3413 | } |
3414 | ||
3415 | /* update the directory size in the log to reflect the names | |
3416 | * we have removed | |
3417 | */ | |
3418 | if (bytes_del) { | |
3419 | struct btrfs_key key; | |
3420 | ||
33345d01 | 3421 | key.objectid = dir_ino; |
e02119d5 CM |
3422 | key.offset = 0; |
3423 | key.type = BTRFS_INODE_ITEM_KEY; | |
b3b4aa74 | 3424 | btrfs_release_path(path); |
e02119d5 CM |
3425 | |
3426 | ret = btrfs_search_slot(trans, log, &key, path, 0, 1); | |
4a500fd1 YZ |
3427 | if (ret < 0) { |
3428 | err = ret; | |
3429 | goto fail; | |
3430 | } | |
e02119d5 CM |
3431 | if (ret == 0) { |
3432 | struct btrfs_inode_item *item; | |
3433 | u64 i_size; | |
3434 | ||
3435 | item = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
3436 | struct btrfs_inode_item); | |
3437 | i_size = btrfs_inode_size(path->nodes[0], item); | |
3438 | if (i_size > bytes_del) | |
3439 | i_size -= bytes_del; | |
3440 | else | |
3441 | i_size = 0; | |
3442 | btrfs_set_inode_size(path->nodes[0], item, i_size); | |
3443 | btrfs_mark_buffer_dirty(path->nodes[0]); | |
3444 | } else | |
3445 | ret = 0; | |
b3b4aa74 | 3446 | btrfs_release_path(path); |
e02119d5 | 3447 | } |
4a500fd1 | 3448 | fail: |
e02119d5 | 3449 | btrfs_free_path(path); |
a62f44a5 | 3450 | out_unlock: |
49f34d1f | 3451 | mutex_unlock(&dir->log_mutex); |
4a500fd1 | 3452 | if (ret == -ENOSPC) { |
90787766 | 3453 | btrfs_set_log_full_commit(trans); |
4a500fd1 | 3454 | ret = 0; |
79787eaa | 3455 | } else if (ret < 0) |
66642832 | 3456 | btrfs_abort_transaction(trans, ret); |
79787eaa | 3457 | |
12fcfd22 | 3458 | btrfs_end_log_trans(root); |
e02119d5 | 3459 | |
411fc6bc | 3460 | return err; |
e02119d5 CM |
3461 | } |
3462 | ||
3463 | /* see comments for btrfs_del_dir_entries_in_log */ | |
3464 | int btrfs_del_inode_ref_in_log(struct btrfs_trans_handle *trans, | |
3465 | struct btrfs_root *root, | |
3466 | const char *name, int name_len, | |
a491abb2 | 3467 | struct btrfs_inode *inode, u64 dirid) |
e02119d5 CM |
3468 | { |
3469 | struct btrfs_root *log; | |
3470 | u64 index; | |
3471 | int ret; | |
3472 | ||
803f0f64 | 3473 | if (!inode_logged(trans, inode)) |
3a5f1d45 CM |
3474 | return 0; |
3475 | ||
e02119d5 CM |
3476 | ret = join_running_log_trans(root); |
3477 | if (ret) | |
3478 | return 0; | |
3479 | log = root->log_root; | |
a491abb2 | 3480 | mutex_lock(&inode->log_mutex); |
e02119d5 | 3481 | |
a491abb2 | 3482 | ret = btrfs_del_inode_ref(trans, log, name, name_len, btrfs_ino(inode), |
e02119d5 | 3483 | dirid, &index); |
a491abb2 | 3484 | mutex_unlock(&inode->log_mutex); |
4a500fd1 | 3485 | if (ret == -ENOSPC) { |
90787766 | 3486 | btrfs_set_log_full_commit(trans); |
4a500fd1 | 3487 | ret = 0; |
79787eaa | 3488 | } else if (ret < 0 && ret != -ENOENT) |
66642832 | 3489 | btrfs_abort_transaction(trans, ret); |
12fcfd22 | 3490 | btrfs_end_log_trans(root); |
e02119d5 | 3491 | |
e02119d5 CM |
3492 | return ret; |
3493 | } | |
3494 | ||
3495 | /* | |
3496 | * creates a range item in the log for 'dirid'. first_offset and | |
3497 | * last_offset tell us which parts of the key space the log should | |
3498 | * be considered authoritative for. | |
3499 | */ | |
3500 | static noinline int insert_dir_log_key(struct btrfs_trans_handle *trans, | |
3501 | struct btrfs_root *log, | |
3502 | struct btrfs_path *path, | |
3503 | int key_type, u64 dirid, | |
3504 | u64 first_offset, u64 last_offset) | |
3505 | { | |
3506 | int ret; | |
3507 | struct btrfs_key key; | |
3508 | struct btrfs_dir_log_item *item; | |
3509 | ||
3510 | key.objectid = dirid; | |
3511 | key.offset = first_offset; | |
3512 | if (key_type == BTRFS_DIR_ITEM_KEY) | |
3513 | key.type = BTRFS_DIR_LOG_ITEM_KEY; | |
3514 | else | |
3515 | key.type = BTRFS_DIR_LOG_INDEX_KEY; | |
3516 | ret = btrfs_insert_empty_item(trans, log, path, &key, sizeof(*item)); | |
4a500fd1 YZ |
3517 | if (ret) |
3518 | return ret; | |
e02119d5 CM |
3519 | |
3520 | item = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
3521 | struct btrfs_dir_log_item); | |
3522 | btrfs_set_dir_log_end(path->nodes[0], item, last_offset); | |
3523 | btrfs_mark_buffer_dirty(path->nodes[0]); | |
b3b4aa74 | 3524 | btrfs_release_path(path); |
e02119d5 CM |
3525 | return 0; |
3526 | } | |
3527 | ||
3528 | /* | |
3529 | * log all the items included in the current transaction for a given | |
3530 | * directory. This also creates the range items in the log tree required | |
3531 | * to replay anything deleted before the fsync | |
3532 | */ | |
3533 | static noinline int log_dir_items(struct btrfs_trans_handle *trans, | |
684a5773 | 3534 | struct btrfs_root *root, struct btrfs_inode *inode, |
e02119d5 CM |
3535 | struct btrfs_path *path, |
3536 | struct btrfs_path *dst_path, int key_type, | |
2f2ff0ee | 3537 | struct btrfs_log_ctx *ctx, |
e02119d5 CM |
3538 | u64 min_offset, u64 *last_offset_ret) |
3539 | { | |
3540 | struct btrfs_key min_key; | |
e02119d5 CM |
3541 | struct btrfs_root *log = root->log_root; |
3542 | struct extent_buffer *src; | |
4a500fd1 | 3543 | int err = 0; |
e02119d5 CM |
3544 | int ret; |
3545 | int i; | |
3546 | int nritems; | |
3547 | u64 first_offset = min_offset; | |
3548 | u64 last_offset = (u64)-1; | |
684a5773 | 3549 | u64 ino = btrfs_ino(inode); |
e02119d5 CM |
3550 | |
3551 | log = root->log_root; | |
e02119d5 | 3552 | |
33345d01 | 3553 | min_key.objectid = ino; |
e02119d5 CM |
3554 | min_key.type = key_type; |
3555 | min_key.offset = min_offset; | |
3556 | ||
6174d3cb | 3557 | ret = btrfs_search_forward(root, &min_key, path, trans->transid); |
e02119d5 CM |
3558 | |
3559 | /* | |
3560 | * we didn't find anything from this transaction, see if there | |
3561 | * is anything at all | |
3562 | */ | |
33345d01 LZ |
3563 | if (ret != 0 || min_key.objectid != ino || min_key.type != key_type) { |
3564 | min_key.objectid = ino; | |
e02119d5 CM |
3565 | min_key.type = key_type; |
3566 | min_key.offset = (u64)-1; | |
b3b4aa74 | 3567 | btrfs_release_path(path); |
e02119d5 CM |
3568 | ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0); |
3569 | if (ret < 0) { | |
b3b4aa74 | 3570 | btrfs_release_path(path); |
e02119d5 CM |
3571 | return ret; |
3572 | } | |
33345d01 | 3573 | ret = btrfs_previous_item(root, path, ino, key_type); |
e02119d5 CM |
3574 | |
3575 | /* if ret == 0 there are items for this type, | |
3576 | * create a range to tell us the last key of this type. | |
3577 | * otherwise, there are no items in this directory after | |
3578 | * *min_offset, and we create a range to indicate that. | |
3579 | */ | |
3580 | if (ret == 0) { | |
3581 | struct btrfs_key tmp; | |
3582 | btrfs_item_key_to_cpu(path->nodes[0], &tmp, | |
3583 | path->slots[0]); | |
d397712b | 3584 | if (key_type == tmp.type) |
e02119d5 | 3585 | first_offset = max(min_offset, tmp.offset) + 1; |
e02119d5 CM |
3586 | } |
3587 | goto done; | |
3588 | } | |
3589 | ||
3590 | /* go backward to find any previous key */ | |
33345d01 | 3591 | ret = btrfs_previous_item(root, path, ino, key_type); |
e02119d5 CM |
3592 | if (ret == 0) { |
3593 | struct btrfs_key tmp; | |
3594 | btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]); | |
3595 | if (key_type == tmp.type) { | |
3596 | first_offset = tmp.offset; | |
3597 | ret = overwrite_item(trans, log, dst_path, | |
3598 | path->nodes[0], path->slots[0], | |
3599 | &tmp); | |
4a500fd1 YZ |
3600 | if (ret) { |
3601 | err = ret; | |
3602 | goto done; | |
3603 | } | |
e02119d5 CM |
3604 | } |
3605 | } | |
b3b4aa74 | 3606 | btrfs_release_path(path); |
e02119d5 | 3607 | |
2cc83342 JB |
3608 | /* |
3609 | * Find the first key from this transaction again. See the note for | |
3610 | * log_new_dir_dentries, if we're logging a directory recursively we | |
3611 | * won't be holding its i_mutex, which means we can modify the directory | |
3612 | * while we're logging it. If we remove an entry between our first | |
3613 | * search and this search we'll not find the key again and can just | |
3614 | * bail. | |
3615 | */ | |
e02119d5 | 3616 | ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0); |
2cc83342 | 3617 | if (ret != 0) |
e02119d5 | 3618 | goto done; |
e02119d5 CM |
3619 | |
3620 | /* | |
3621 | * we have a block from this transaction, log every item in it | |
3622 | * from our directory | |
3623 | */ | |
d397712b | 3624 | while (1) { |
e02119d5 CM |
3625 | struct btrfs_key tmp; |
3626 | src = path->nodes[0]; | |
3627 | nritems = btrfs_header_nritems(src); | |
3628 | for (i = path->slots[0]; i < nritems; i++) { | |
2f2ff0ee FM |
3629 | struct btrfs_dir_item *di; |
3630 | ||
e02119d5 CM |
3631 | btrfs_item_key_to_cpu(src, &min_key, i); |
3632 | ||
33345d01 | 3633 | if (min_key.objectid != ino || min_key.type != key_type) |
e02119d5 CM |
3634 | goto done; |
3635 | ret = overwrite_item(trans, log, dst_path, src, i, | |
3636 | &min_key); | |
4a500fd1 YZ |
3637 | if (ret) { |
3638 | err = ret; | |
3639 | goto done; | |
3640 | } | |
2f2ff0ee FM |
3641 | |
3642 | /* | |
3643 | * We must make sure that when we log a directory entry, | |
3644 | * the corresponding inode, after log replay, has a | |
3645 | * matching link count. For example: | |
3646 | * | |
3647 | * touch foo | |
3648 | * mkdir mydir | |
3649 | * sync | |
3650 | * ln foo mydir/bar | |
3651 | * xfs_io -c "fsync" mydir | |
3652 | * <crash> | |
3653 | * <mount fs and log replay> | |
3654 | * | |
3655 | * Would result in a fsync log that when replayed, our | |
3656 | * file inode would have a link count of 1, but we get | |
3657 | * two directory entries pointing to the same inode. | |
3658 | * After removing one of the names, it would not be | |
3659 | * possible to remove the other name, which resulted | |
3660 | * always in stale file handle errors, and would not | |
3661 | * be possible to rmdir the parent directory, since | |
3662 | * its i_size could never decrement to the value | |
3663 | * BTRFS_EMPTY_DIR_SIZE, resulting in -ENOTEMPTY errors. | |
3664 | */ | |
3665 | di = btrfs_item_ptr(src, i, struct btrfs_dir_item); | |
3666 | btrfs_dir_item_key_to_cpu(src, di, &tmp); | |
3667 | if (ctx && | |
3668 | (btrfs_dir_transid(src, di) == trans->transid || | |
3669 | btrfs_dir_type(src, di) == BTRFS_FT_DIR) && | |
3670 | tmp.type != BTRFS_ROOT_ITEM_KEY) | |
3671 | ctx->log_new_dentries = true; | |
e02119d5 CM |
3672 | } |
3673 | path->slots[0] = nritems; | |
3674 | ||
3675 | /* | |
3676 | * look ahead to the next item and see if it is also | |
3677 | * from this directory and from this transaction | |
3678 | */ | |
3679 | ret = btrfs_next_leaf(root, path); | |
80c0b421 LB |
3680 | if (ret) { |
3681 | if (ret == 1) | |
3682 | last_offset = (u64)-1; | |
3683 | else | |
3684 | err = ret; | |
e02119d5 CM |
3685 | goto done; |
3686 | } | |
3687 | btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]); | |
33345d01 | 3688 | if (tmp.objectid != ino || tmp.type != key_type) { |
e02119d5 CM |
3689 | last_offset = (u64)-1; |
3690 | goto done; | |
3691 | } | |
3692 | if (btrfs_header_generation(path->nodes[0]) != trans->transid) { | |
3693 | ret = overwrite_item(trans, log, dst_path, | |
3694 | path->nodes[0], path->slots[0], | |
3695 | &tmp); | |
4a500fd1 YZ |
3696 | if (ret) |
3697 | err = ret; | |
3698 | else | |
3699 | last_offset = tmp.offset; | |
e02119d5 CM |
3700 | goto done; |
3701 | } | |
3702 | } | |
3703 | done: | |
b3b4aa74 DS |
3704 | btrfs_release_path(path); |
3705 | btrfs_release_path(dst_path); | |
e02119d5 | 3706 | |
4a500fd1 YZ |
3707 | if (err == 0) { |
3708 | *last_offset_ret = last_offset; | |
3709 | /* | |
3710 | * insert the log range keys to indicate where the log | |
3711 | * is valid | |
3712 | */ | |
3713 | ret = insert_dir_log_key(trans, log, path, key_type, | |
33345d01 | 3714 | ino, first_offset, last_offset); |
4a500fd1 YZ |
3715 | if (ret) |
3716 | err = ret; | |
3717 | } | |
3718 | return err; | |
e02119d5 CM |
3719 | } |
3720 | ||
3721 | /* | |
3722 | * logging directories is very similar to logging inodes, We find all the items | |
3723 | * from the current transaction and write them to the log. | |
3724 | * | |
3725 | * The recovery code scans the directory in the subvolume, and if it finds a | |
3726 | * key in the range logged that is not present in the log tree, then it means | |
3727 | * that dir entry was unlinked during the transaction. | |
3728 | * | |
3729 | * In order for that scan to work, we must include one key smaller than | |
3730 | * the smallest logged by this transaction and one key larger than the largest | |
3731 | * key logged by this transaction. | |
3732 | */ | |
3733 | static noinline int log_directory_changes(struct btrfs_trans_handle *trans, | |
dbf39ea4 | 3734 | struct btrfs_root *root, struct btrfs_inode *inode, |
e02119d5 | 3735 | struct btrfs_path *path, |
2f2ff0ee FM |
3736 | struct btrfs_path *dst_path, |
3737 | struct btrfs_log_ctx *ctx) | |
e02119d5 CM |
3738 | { |
3739 | u64 min_key; | |
3740 | u64 max_key; | |
3741 | int ret; | |
3742 | int key_type = BTRFS_DIR_ITEM_KEY; | |
3743 | ||
3744 | again: | |
3745 | min_key = 0; | |
3746 | max_key = 0; | |
d397712b | 3747 | while (1) { |
dbf39ea4 NB |
3748 | ret = log_dir_items(trans, root, inode, path, dst_path, key_type, |
3749 | ctx, min_key, &max_key); | |
4a500fd1 YZ |
3750 | if (ret) |
3751 | return ret; | |
e02119d5 CM |
3752 | if (max_key == (u64)-1) |
3753 | break; | |
3754 | min_key = max_key + 1; | |
3755 | } | |
3756 | ||
3757 | if (key_type == BTRFS_DIR_ITEM_KEY) { | |
3758 | key_type = BTRFS_DIR_INDEX_KEY; | |
3759 | goto again; | |
3760 | } | |
3761 | return 0; | |
3762 | } | |
3763 | ||
3764 | /* | |
3765 | * a helper function to drop items from the log before we relog an | |
3766 | * inode. max_key_type indicates the highest item type to remove. | |
3767 | * This cannot be run for file data extents because it does not | |
3768 | * free the extents they point to. | |
3769 | */ | |
3770 | static int drop_objectid_items(struct btrfs_trans_handle *trans, | |
3771 | struct btrfs_root *log, | |
3772 | struct btrfs_path *path, | |
3773 | u64 objectid, int max_key_type) | |
3774 | { | |
3775 | int ret; | |
3776 | struct btrfs_key key; | |
3777 | struct btrfs_key found_key; | |
18ec90d6 | 3778 | int start_slot; |
e02119d5 CM |
3779 | |
3780 | key.objectid = objectid; | |
3781 | key.type = max_key_type; | |
3782 | key.offset = (u64)-1; | |
3783 | ||
d397712b | 3784 | while (1) { |
e02119d5 | 3785 | ret = btrfs_search_slot(trans, log, &key, path, -1, 1); |
3650860b | 3786 | BUG_ON(ret == 0); /* Logic error */ |
4a500fd1 | 3787 | if (ret < 0) |
e02119d5 CM |
3788 | break; |
3789 | ||
3790 | if (path->slots[0] == 0) | |
3791 | break; | |
3792 | ||
3793 | path->slots[0]--; | |
3794 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, | |
3795 | path->slots[0]); | |
3796 | ||
3797 | if (found_key.objectid != objectid) | |
3798 | break; | |
3799 | ||
18ec90d6 JB |
3800 | found_key.offset = 0; |
3801 | found_key.type = 0; | |
e3b83361 | 3802 | ret = btrfs_bin_search(path->nodes[0], &found_key, &start_slot); |
cbca7d59 FM |
3803 | if (ret < 0) |
3804 | break; | |
18ec90d6 JB |
3805 | |
3806 | ret = btrfs_del_items(trans, log, path, start_slot, | |
3807 | path->slots[0] - start_slot + 1); | |
3808 | /* | |
3809 | * If start slot isn't 0 then we don't need to re-search, we've | |
3810 | * found the last guy with the objectid in this tree. | |
3811 | */ | |
3812 | if (ret || start_slot != 0) | |
65a246c5 | 3813 | break; |
b3b4aa74 | 3814 | btrfs_release_path(path); |
e02119d5 | 3815 | } |
b3b4aa74 | 3816 | btrfs_release_path(path); |
5bdbeb21 JB |
3817 | if (ret > 0) |
3818 | ret = 0; | |
4a500fd1 | 3819 | return ret; |
e02119d5 CM |
3820 | } |
3821 | ||
94edf4ae JB |
3822 | static void fill_inode_item(struct btrfs_trans_handle *trans, |
3823 | struct extent_buffer *leaf, | |
3824 | struct btrfs_inode_item *item, | |
1a4bcf47 FM |
3825 | struct inode *inode, int log_inode_only, |
3826 | u64 logged_isize) | |
94edf4ae | 3827 | { |
0b1c6cca JB |
3828 | struct btrfs_map_token token; |
3829 | ||
c82f823c | 3830 | btrfs_init_map_token(&token, leaf); |
94edf4ae JB |
3831 | |
3832 | if (log_inode_only) { | |
3833 | /* set the generation to zero so the recover code | |
3834 | * can tell the difference between an logging | |
3835 | * just to say 'this inode exists' and a logging | |
3836 | * to say 'update this inode with these values' | |
3837 | */ | |
cc4c13d5 DS |
3838 | btrfs_set_token_inode_generation(&token, item, 0); |
3839 | btrfs_set_token_inode_size(&token, item, logged_isize); | |
94edf4ae | 3840 | } else { |
cc4c13d5 DS |
3841 | btrfs_set_token_inode_generation(&token, item, |
3842 | BTRFS_I(inode)->generation); | |
3843 | btrfs_set_token_inode_size(&token, item, inode->i_size); | |
0b1c6cca JB |
3844 | } |
3845 | ||
cc4c13d5 DS |
3846 | btrfs_set_token_inode_uid(&token, item, i_uid_read(inode)); |
3847 | btrfs_set_token_inode_gid(&token, item, i_gid_read(inode)); | |
3848 | btrfs_set_token_inode_mode(&token, item, inode->i_mode); | |
3849 | btrfs_set_token_inode_nlink(&token, item, inode->i_nlink); | |
3850 | ||
3851 | btrfs_set_token_timespec_sec(&token, &item->atime, | |
3852 | inode->i_atime.tv_sec); | |
3853 | btrfs_set_token_timespec_nsec(&token, &item->atime, | |
3854 | inode->i_atime.tv_nsec); | |
3855 | ||
3856 | btrfs_set_token_timespec_sec(&token, &item->mtime, | |
3857 | inode->i_mtime.tv_sec); | |
3858 | btrfs_set_token_timespec_nsec(&token, &item->mtime, | |
3859 | inode->i_mtime.tv_nsec); | |
3860 | ||
3861 | btrfs_set_token_timespec_sec(&token, &item->ctime, | |
3862 | inode->i_ctime.tv_sec); | |
3863 | btrfs_set_token_timespec_nsec(&token, &item->ctime, | |
3864 | inode->i_ctime.tv_nsec); | |
3865 | ||
3866 | btrfs_set_token_inode_nbytes(&token, item, inode_get_bytes(inode)); | |
3867 | ||
3868 | btrfs_set_token_inode_sequence(&token, item, inode_peek_iversion(inode)); | |
3869 | btrfs_set_token_inode_transid(&token, item, trans->transid); | |
3870 | btrfs_set_token_inode_rdev(&token, item, inode->i_rdev); | |
3871 | btrfs_set_token_inode_flags(&token, item, BTRFS_I(inode)->flags); | |
3872 | btrfs_set_token_inode_block_group(&token, item, 0); | |
94edf4ae JB |
3873 | } |
3874 | ||
a95249b3 JB |
3875 | static int log_inode_item(struct btrfs_trans_handle *trans, |
3876 | struct btrfs_root *log, struct btrfs_path *path, | |
6d889a3b | 3877 | struct btrfs_inode *inode) |
a95249b3 JB |
3878 | { |
3879 | struct btrfs_inode_item *inode_item; | |
a95249b3 JB |
3880 | int ret; |
3881 | ||
efd0c405 | 3882 | ret = btrfs_insert_empty_item(trans, log, path, |
6d889a3b | 3883 | &inode->location, sizeof(*inode_item)); |
a95249b3 JB |
3884 | if (ret && ret != -EEXIST) |
3885 | return ret; | |
3886 | inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
3887 | struct btrfs_inode_item); | |
6d889a3b NB |
3888 | fill_inode_item(trans, path->nodes[0], inode_item, &inode->vfs_inode, |
3889 | 0, 0); | |
a95249b3 JB |
3890 | btrfs_release_path(path); |
3891 | return 0; | |
3892 | } | |
3893 | ||
40e046ac | 3894 | static int log_csums(struct btrfs_trans_handle *trans, |
3ebac17c | 3895 | struct btrfs_inode *inode, |
40e046ac FM |
3896 | struct btrfs_root *log_root, |
3897 | struct btrfs_ordered_sum *sums) | |
3898 | { | |
e289f03e FM |
3899 | const u64 lock_end = sums->bytenr + sums->len - 1; |
3900 | struct extent_state *cached_state = NULL; | |
40e046ac FM |
3901 | int ret; |
3902 | ||
3ebac17c FM |
3903 | /* |
3904 | * If this inode was not used for reflink operations in the current | |
3905 | * transaction with new extents, then do the fast path, no need to | |
3906 | * worry about logging checksum items with overlapping ranges. | |
3907 | */ | |
3908 | if (inode->last_reflink_trans < trans->transid) | |
3909 | return btrfs_csum_file_blocks(trans, log_root, sums); | |
3910 | ||
e289f03e FM |
3911 | /* |
3912 | * Serialize logging for checksums. This is to avoid racing with the | |
3913 | * same checksum being logged by another task that is logging another | |
3914 | * file which happens to refer to the same extent as well. Such races | |
3915 | * can leave checksum items in the log with overlapping ranges. | |
3916 | */ | |
3917 | ret = lock_extent_bits(&log_root->log_csum_range, sums->bytenr, | |
3918 | lock_end, &cached_state); | |
3919 | if (ret) | |
3920 | return ret; | |
40e046ac FM |
3921 | /* |
3922 | * Due to extent cloning, we might have logged a csum item that covers a | |
3923 | * subrange of a cloned extent, and later we can end up logging a csum | |
3924 | * item for a larger subrange of the same extent or the entire range. | |
3925 | * This would leave csum items in the log tree that cover the same range | |
3926 | * and break the searches for checksums in the log tree, resulting in | |
3927 | * some checksums missing in the fs/subvolume tree. So just delete (or | |
3928 | * trim and adjust) any existing csum items in the log for this range. | |
3929 | */ | |
3930 | ret = btrfs_del_csums(trans, log_root, sums->bytenr, sums->len); | |
e289f03e FM |
3931 | if (!ret) |
3932 | ret = btrfs_csum_file_blocks(trans, log_root, sums); | |
40e046ac | 3933 | |
e289f03e FM |
3934 | unlock_extent_cached(&log_root->log_csum_range, sums->bytenr, lock_end, |
3935 | &cached_state); | |
3936 | ||
3937 | return ret; | |
40e046ac FM |
3938 | } |
3939 | ||
31ff1cd2 | 3940 | static noinline int copy_items(struct btrfs_trans_handle *trans, |
44d70e19 | 3941 | struct btrfs_inode *inode, |
31ff1cd2 | 3942 | struct btrfs_path *dst_path, |
0e56315c | 3943 | struct btrfs_path *src_path, |
1a4bcf47 FM |
3944 | int start_slot, int nr, int inode_only, |
3945 | u64 logged_isize) | |
31ff1cd2 | 3946 | { |
3ffbd68c | 3947 | struct btrfs_fs_info *fs_info = trans->fs_info; |
31ff1cd2 CM |
3948 | unsigned long src_offset; |
3949 | unsigned long dst_offset; | |
44d70e19 | 3950 | struct btrfs_root *log = inode->root->log_root; |
31ff1cd2 CM |
3951 | struct btrfs_file_extent_item *extent; |
3952 | struct btrfs_inode_item *inode_item; | |
16e7549f | 3953 | struct extent_buffer *src = src_path->nodes[0]; |
31ff1cd2 CM |
3954 | int ret; |
3955 | struct btrfs_key *ins_keys; | |
3956 | u32 *ins_sizes; | |
3957 | char *ins_data; | |
3958 | int i; | |
d20f7043 | 3959 | struct list_head ordered_sums; |
44d70e19 | 3960 | int skip_csum = inode->flags & BTRFS_INODE_NODATASUM; |
d20f7043 CM |
3961 | |
3962 | INIT_LIST_HEAD(&ordered_sums); | |
31ff1cd2 CM |
3963 | |
3964 | ins_data = kmalloc(nr * sizeof(struct btrfs_key) + | |
3965 | nr * sizeof(u32), GFP_NOFS); | |
2a29edc6 | 3966 | if (!ins_data) |
3967 | return -ENOMEM; | |
3968 | ||
31ff1cd2 CM |
3969 | ins_sizes = (u32 *)ins_data; |
3970 | ins_keys = (struct btrfs_key *)(ins_data + nr * sizeof(u32)); | |
3971 | ||
3972 | for (i = 0; i < nr; i++) { | |
3973 | ins_sizes[i] = btrfs_item_size_nr(src, i + start_slot); | |
3974 | btrfs_item_key_to_cpu(src, ins_keys + i, i + start_slot); | |
3975 | } | |
3976 | ret = btrfs_insert_empty_items(trans, log, dst_path, | |
3977 | ins_keys, ins_sizes, nr); | |
4a500fd1 YZ |
3978 | if (ret) { |
3979 | kfree(ins_data); | |
3980 | return ret; | |
3981 | } | |
31ff1cd2 | 3982 | |
5d4f98a2 | 3983 | for (i = 0; i < nr; i++, dst_path->slots[0]++) { |
31ff1cd2 CM |
3984 | dst_offset = btrfs_item_ptr_offset(dst_path->nodes[0], |
3985 | dst_path->slots[0]); | |
3986 | ||
3987 | src_offset = btrfs_item_ptr_offset(src, start_slot + i); | |
3988 | ||
94edf4ae | 3989 | if (ins_keys[i].type == BTRFS_INODE_ITEM_KEY) { |
31ff1cd2 CM |
3990 | inode_item = btrfs_item_ptr(dst_path->nodes[0], |
3991 | dst_path->slots[0], | |
3992 | struct btrfs_inode_item); | |
94edf4ae | 3993 | fill_inode_item(trans, dst_path->nodes[0], inode_item, |
f85b7379 DS |
3994 | &inode->vfs_inode, |
3995 | inode_only == LOG_INODE_EXISTS, | |
1a4bcf47 | 3996 | logged_isize); |
94edf4ae JB |
3997 | } else { |
3998 | copy_extent_buffer(dst_path->nodes[0], src, dst_offset, | |
3999 | src_offset, ins_sizes[i]); | |
31ff1cd2 | 4000 | } |
94edf4ae | 4001 | |
31ff1cd2 CM |
4002 | /* take a reference on file data extents so that truncates |
4003 | * or deletes of this inode don't have to relog the inode | |
4004 | * again | |
4005 | */ | |
962a298f | 4006 | if (ins_keys[i].type == BTRFS_EXTENT_DATA_KEY && |
d2794405 | 4007 | !skip_csum) { |
31ff1cd2 CM |
4008 | int found_type; |
4009 | extent = btrfs_item_ptr(src, start_slot + i, | |
4010 | struct btrfs_file_extent_item); | |
4011 | ||
8e531cdf | 4012 | if (btrfs_file_extent_generation(src, extent) < trans->transid) |
4013 | continue; | |
4014 | ||
31ff1cd2 | 4015 | found_type = btrfs_file_extent_type(src, extent); |
6f1fed77 | 4016 | if (found_type == BTRFS_FILE_EXTENT_REG) { |
5d4f98a2 YZ |
4017 | u64 ds, dl, cs, cl; |
4018 | ds = btrfs_file_extent_disk_bytenr(src, | |
4019 | extent); | |
4020 | /* ds == 0 is a hole */ | |
4021 | if (ds == 0) | |
4022 | continue; | |
4023 | ||
4024 | dl = btrfs_file_extent_disk_num_bytes(src, | |
4025 | extent); | |
4026 | cs = btrfs_file_extent_offset(src, extent); | |
4027 | cl = btrfs_file_extent_num_bytes(src, | |
a419aef8 | 4028 | extent); |
580afd76 CM |
4029 | if (btrfs_file_extent_compression(src, |
4030 | extent)) { | |
4031 | cs = 0; | |
4032 | cl = dl; | |
4033 | } | |
5d4f98a2 YZ |
4034 | |
4035 | ret = btrfs_lookup_csums_range( | |
0b246afa | 4036 | fs_info->csum_root, |
5d4f98a2 | 4037 | ds + cs, ds + cs + cl - 1, |
a2de733c | 4038 | &ordered_sums, 0); |
3650860b JB |
4039 | if (ret) { |
4040 | btrfs_release_path(dst_path); | |
4041 | kfree(ins_data); | |
4042 | return ret; | |
4043 | } | |
31ff1cd2 CM |
4044 | } |
4045 | } | |
31ff1cd2 CM |
4046 | } |
4047 | ||
4048 | btrfs_mark_buffer_dirty(dst_path->nodes[0]); | |
b3b4aa74 | 4049 | btrfs_release_path(dst_path); |
31ff1cd2 | 4050 | kfree(ins_data); |
d20f7043 CM |
4051 | |
4052 | /* | |
4053 | * we have to do this after the loop above to avoid changing the | |
4054 | * log tree while trying to change the log tree. | |
4055 | */ | |
4a500fd1 | 4056 | ret = 0; |
d397712b | 4057 | while (!list_empty(&ordered_sums)) { |
d20f7043 CM |
4058 | struct btrfs_ordered_sum *sums = list_entry(ordered_sums.next, |
4059 | struct btrfs_ordered_sum, | |
4060 | list); | |
4a500fd1 | 4061 | if (!ret) |
3ebac17c | 4062 | ret = log_csums(trans, inode, log, sums); |
d20f7043 CM |
4063 | list_del(&sums->list); |
4064 | kfree(sums); | |
4065 | } | |
16e7549f | 4066 | |
4a500fd1 | 4067 | return ret; |
31ff1cd2 CM |
4068 | } |
4069 | ||
5dc562c5 JB |
4070 | static int extent_cmp(void *priv, struct list_head *a, struct list_head *b) |
4071 | { | |
4072 | struct extent_map *em1, *em2; | |
4073 | ||
4074 | em1 = list_entry(a, struct extent_map, list); | |
4075 | em2 = list_entry(b, struct extent_map, list); | |
4076 | ||
4077 | if (em1->start < em2->start) | |
4078 | return -1; | |
4079 | else if (em1->start > em2->start) | |
4080 | return 1; | |
4081 | return 0; | |
4082 | } | |
4083 | ||
e7175a69 JB |
4084 | static int log_extent_csums(struct btrfs_trans_handle *trans, |
4085 | struct btrfs_inode *inode, | |
a9ecb653 | 4086 | struct btrfs_root *log_root, |
e7175a69 | 4087 | const struct extent_map *em) |
5dc562c5 | 4088 | { |
2ab28f32 JB |
4089 | u64 csum_offset; |
4090 | u64 csum_len; | |
8407f553 FM |
4091 | LIST_HEAD(ordered_sums); |
4092 | int ret = 0; | |
0aa4a17d | 4093 | |
e7175a69 JB |
4094 | if (inode->flags & BTRFS_INODE_NODATASUM || |
4095 | test_bit(EXTENT_FLAG_PREALLOC, &em->flags) || | |
8407f553 | 4096 | em->block_start == EXTENT_MAP_HOLE) |
70c8a91c | 4097 | return 0; |
5dc562c5 | 4098 | |
e7175a69 | 4099 | /* If we're compressed we have to save the entire range of csums. */ |
488111aa FDBM |
4100 | if (em->compress_type) { |
4101 | csum_offset = 0; | |
8407f553 | 4102 | csum_len = max(em->block_len, em->orig_block_len); |
488111aa | 4103 | } else { |
e7175a69 JB |
4104 | csum_offset = em->mod_start - em->start; |
4105 | csum_len = em->mod_len; | |
488111aa | 4106 | } |
2ab28f32 | 4107 | |
70c8a91c | 4108 | /* block start is already adjusted for the file extent offset. */ |
a9ecb653 | 4109 | ret = btrfs_lookup_csums_range(trans->fs_info->csum_root, |
70c8a91c JB |
4110 | em->block_start + csum_offset, |
4111 | em->block_start + csum_offset + | |
4112 | csum_len - 1, &ordered_sums, 0); | |
4113 | if (ret) | |
4114 | return ret; | |
5dc562c5 | 4115 | |
70c8a91c JB |
4116 | while (!list_empty(&ordered_sums)) { |
4117 | struct btrfs_ordered_sum *sums = list_entry(ordered_sums.next, | |
4118 | struct btrfs_ordered_sum, | |
4119 | list); | |
4120 | if (!ret) | |
3ebac17c | 4121 | ret = log_csums(trans, inode, log_root, sums); |
70c8a91c JB |
4122 | list_del(&sums->list); |
4123 | kfree(sums); | |
5dc562c5 JB |
4124 | } |
4125 | ||
70c8a91c | 4126 | return ret; |
5dc562c5 JB |
4127 | } |
4128 | ||
8407f553 | 4129 | static int log_one_extent(struct btrfs_trans_handle *trans, |
9d122629 | 4130 | struct btrfs_inode *inode, struct btrfs_root *root, |
8407f553 FM |
4131 | const struct extent_map *em, |
4132 | struct btrfs_path *path, | |
8407f553 FM |
4133 | struct btrfs_log_ctx *ctx) |
4134 | { | |
4135 | struct btrfs_root *log = root->log_root; | |
4136 | struct btrfs_file_extent_item *fi; | |
4137 | struct extent_buffer *leaf; | |
4138 | struct btrfs_map_token token; | |
4139 | struct btrfs_key key; | |
4140 | u64 extent_offset = em->start - em->orig_start; | |
4141 | u64 block_len; | |
4142 | int ret; | |
4143 | int extent_inserted = 0; | |
8407f553 | 4144 | |
a9ecb653 | 4145 | ret = log_extent_csums(trans, inode, log, em); |
8407f553 FM |
4146 | if (ret) |
4147 | return ret; | |
4148 | ||
906c448c | 4149 | ret = __btrfs_drop_extents(trans, log, inode, path, em->start, |
8407f553 FM |
4150 | em->start + em->len, NULL, 0, 1, |
4151 | sizeof(*fi), &extent_inserted); | |
4152 | if (ret) | |
4153 | return ret; | |
4154 | ||
4155 | if (!extent_inserted) { | |
9d122629 | 4156 | key.objectid = btrfs_ino(inode); |
8407f553 FM |
4157 | key.type = BTRFS_EXTENT_DATA_KEY; |
4158 | key.offset = em->start; | |
4159 | ||
4160 | ret = btrfs_insert_empty_item(trans, log, path, &key, | |
4161 | sizeof(*fi)); | |
4162 | if (ret) | |
4163 | return ret; | |
4164 | } | |
4165 | leaf = path->nodes[0]; | |
c82f823c | 4166 | btrfs_init_map_token(&token, leaf); |
8407f553 FM |
4167 | fi = btrfs_item_ptr(leaf, path->slots[0], |
4168 | struct btrfs_file_extent_item); | |
4169 | ||
cc4c13d5 | 4170 | btrfs_set_token_file_extent_generation(&token, fi, trans->transid); |
8407f553 | 4171 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) |
cc4c13d5 DS |
4172 | btrfs_set_token_file_extent_type(&token, fi, |
4173 | BTRFS_FILE_EXTENT_PREALLOC); | |
8407f553 | 4174 | else |
cc4c13d5 DS |
4175 | btrfs_set_token_file_extent_type(&token, fi, |
4176 | BTRFS_FILE_EXTENT_REG); | |
8407f553 FM |
4177 | |
4178 | block_len = max(em->block_len, em->orig_block_len); | |
4179 | if (em->compress_type != BTRFS_COMPRESS_NONE) { | |
cc4c13d5 DS |
4180 | btrfs_set_token_file_extent_disk_bytenr(&token, fi, |
4181 | em->block_start); | |
4182 | btrfs_set_token_file_extent_disk_num_bytes(&token, fi, block_len); | |
8407f553 | 4183 | } else if (em->block_start < EXTENT_MAP_LAST_BYTE) { |
cc4c13d5 | 4184 | btrfs_set_token_file_extent_disk_bytenr(&token, fi, |
8407f553 | 4185 | em->block_start - |
cc4c13d5 DS |
4186 | extent_offset); |
4187 | btrfs_set_token_file_extent_disk_num_bytes(&token, fi, block_len); | |
8407f553 | 4188 | } else { |
cc4c13d5 DS |
4189 | btrfs_set_token_file_extent_disk_bytenr(&token, fi, 0); |
4190 | btrfs_set_token_file_extent_disk_num_bytes(&token, fi, 0); | |
8407f553 FM |
4191 | } |
4192 | ||
cc4c13d5 DS |
4193 | btrfs_set_token_file_extent_offset(&token, fi, extent_offset); |
4194 | btrfs_set_token_file_extent_num_bytes(&token, fi, em->len); | |
4195 | btrfs_set_token_file_extent_ram_bytes(&token, fi, em->ram_bytes); | |
4196 | btrfs_set_token_file_extent_compression(&token, fi, em->compress_type); | |
4197 | btrfs_set_token_file_extent_encryption(&token, fi, 0); | |
4198 | btrfs_set_token_file_extent_other_encoding(&token, fi, 0); | |
8407f553 FM |
4199 | btrfs_mark_buffer_dirty(leaf); |
4200 | ||
4201 | btrfs_release_path(path); | |
4202 | ||
4203 | return ret; | |
4204 | } | |
4205 | ||
31d11b83 FM |
4206 | /* |
4207 | * Log all prealloc extents beyond the inode's i_size to make sure we do not | |
4208 | * lose them after doing a fast fsync and replaying the log. We scan the | |
4209 | * subvolume's root instead of iterating the inode's extent map tree because | |
4210 | * otherwise we can log incorrect extent items based on extent map conversion. | |
4211 | * That can happen due to the fact that extent maps are merged when they | |
4212 | * are not in the extent map tree's list of modified extents. | |
4213 | */ | |
4214 | static int btrfs_log_prealloc_extents(struct btrfs_trans_handle *trans, | |
4215 | struct btrfs_inode *inode, | |
4216 | struct btrfs_path *path) | |
4217 | { | |
4218 | struct btrfs_root *root = inode->root; | |
4219 | struct btrfs_key key; | |
4220 | const u64 i_size = i_size_read(&inode->vfs_inode); | |
4221 | const u64 ino = btrfs_ino(inode); | |
4222 | struct btrfs_path *dst_path = NULL; | |
0e56315c | 4223 | bool dropped_extents = false; |
f135cea3 FM |
4224 | u64 truncate_offset = i_size; |
4225 | struct extent_buffer *leaf; | |
4226 | int slot; | |
31d11b83 FM |
4227 | int ins_nr = 0; |
4228 | int start_slot; | |
4229 | int ret; | |
4230 | ||
4231 | if (!(inode->flags & BTRFS_INODE_PREALLOC)) | |
4232 | return 0; | |
4233 | ||
4234 | key.objectid = ino; | |
4235 | key.type = BTRFS_EXTENT_DATA_KEY; | |
4236 | key.offset = i_size; | |
4237 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
4238 | if (ret < 0) | |
4239 | goto out; | |
4240 | ||
f135cea3 FM |
4241 | /* |
4242 | * We must check if there is a prealloc extent that starts before the | |
4243 | * i_size and crosses the i_size boundary. This is to ensure later we | |
4244 | * truncate down to the end of that extent and not to the i_size, as | |
4245 | * otherwise we end up losing part of the prealloc extent after a log | |
4246 | * replay and with an implicit hole if there is another prealloc extent | |
4247 | * that starts at an offset beyond i_size. | |
4248 | */ | |
4249 | ret = btrfs_previous_item(root, path, ino, BTRFS_EXTENT_DATA_KEY); | |
4250 | if (ret < 0) | |
4251 | goto out; | |
4252 | ||
4253 | if (ret == 0) { | |
4254 | struct btrfs_file_extent_item *ei; | |
4255 | ||
4256 | leaf = path->nodes[0]; | |
4257 | slot = path->slots[0]; | |
4258 | ei = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item); | |
4259 | ||
4260 | if (btrfs_file_extent_type(leaf, ei) == | |
4261 | BTRFS_FILE_EXTENT_PREALLOC) { | |
4262 | u64 extent_end; | |
4263 | ||
4264 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
4265 | extent_end = key.offset + | |
4266 | btrfs_file_extent_num_bytes(leaf, ei); | |
4267 | ||
4268 | if (extent_end > i_size) | |
4269 | truncate_offset = extent_end; | |
4270 | } | |
4271 | } else { | |
4272 | ret = 0; | |
4273 | } | |
4274 | ||
31d11b83 | 4275 | while (true) { |
f135cea3 FM |
4276 | leaf = path->nodes[0]; |
4277 | slot = path->slots[0]; | |
31d11b83 FM |
4278 | |
4279 | if (slot >= btrfs_header_nritems(leaf)) { | |
4280 | if (ins_nr > 0) { | |
4281 | ret = copy_items(trans, inode, dst_path, path, | |
0e56315c | 4282 | start_slot, ins_nr, 1, 0); |
31d11b83 FM |
4283 | if (ret < 0) |
4284 | goto out; | |
4285 | ins_nr = 0; | |
4286 | } | |
4287 | ret = btrfs_next_leaf(root, path); | |
4288 | if (ret < 0) | |
4289 | goto out; | |
4290 | if (ret > 0) { | |
4291 | ret = 0; | |
4292 | break; | |
4293 | } | |
4294 | continue; | |
4295 | } | |
4296 | ||
4297 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
4298 | if (key.objectid > ino) | |
4299 | break; | |
4300 | if (WARN_ON_ONCE(key.objectid < ino) || | |
4301 | key.type < BTRFS_EXTENT_DATA_KEY || | |
4302 | key.offset < i_size) { | |
4303 | path->slots[0]++; | |
4304 | continue; | |
4305 | } | |
0e56315c | 4306 | if (!dropped_extents) { |
31d11b83 FM |
4307 | /* |
4308 | * Avoid logging extent items logged in past fsync calls | |
4309 | * and leading to duplicate keys in the log tree. | |
4310 | */ | |
4311 | do { | |
4312 | ret = btrfs_truncate_inode_items(trans, | |
4313 | root->log_root, | |
4314 | &inode->vfs_inode, | |
f135cea3 | 4315 | truncate_offset, |
31d11b83 FM |
4316 | BTRFS_EXTENT_DATA_KEY); |
4317 | } while (ret == -EAGAIN); | |
4318 | if (ret) | |
4319 | goto out; | |
0e56315c | 4320 | dropped_extents = true; |
31d11b83 FM |
4321 | } |
4322 | if (ins_nr == 0) | |
4323 | start_slot = slot; | |
4324 | ins_nr++; | |
4325 | path->slots[0]++; | |
4326 | if (!dst_path) { | |
4327 | dst_path = btrfs_alloc_path(); | |
4328 | if (!dst_path) { | |
4329 | ret = -ENOMEM; | |
4330 | goto out; | |
4331 | } | |
4332 | } | |
4333 | } | |
0bc2d3c0 | 4334 | if (ins_nr > 0) |
0e56315c | 4335 | ret = copy_items(trans, inode, dst_path, path, |
31d11b83 | 4336 | start_slot, ins_nr, 1, 0); |
31d11b83 FM |
4337 | out: |
4338 | btrfs_release_path(path); | |
4339 | btrfs_free_path(dst_path); | |
4340 | return ret; | |
4341 | } | |
4342 | ||
5dc562c5 JB |
4343 | static int btrfs_log_changed_extents(struct btrfs_trans_handle *trans, |
4344 | struct btrfs_root *root, | |
9d122629 | 4345 | struct btrfs_inode *inode, |
827463c4 | 4346 | struct btrfs_path *path, |
de0ee0ed FM |
4347 | struct btrfs_log_ctx *ctx, |
4348 | const u64 start, | |
4349 | const u64 end) | |
5dc562c5 | 4350 | { |
5dc562c5 JB |
4351 | struct extent_map *em, *n; |
4352 | struct list_head extents; | |
9d122629 | 4353 | struct extent_map_tree *tree = &inode->extent_tree; |
5dc562c5 JB |
4354 | u64 test_gen; |
4355 | int ret = 0; | |
2ab28f32 | 4356 | int num = 0; |
5dc562c5 JB |
4357 | |
4358 | INIT_LIST_HEAD(&extents); | |
4359 | ||
5dc562c5 JB |
4360 | write_lock(&tree->lock); |
4361 | test_gen = root->fs_info->last_trans_committed; | |
4362 | ||
4363 | list_for_each_entry_safe(em, n, &tree->modified_extents, list) { | |
008c6753 FM |
4364 | /* |
4365 | * Skip extents outside our logging range. It's important to do | |
4366 | * it for correctness because if we don't ignore them, we may | |
4367 | * log them before their ordered extent completes, and therefore | |
4368 | * we could log them without logging their respective checksums | |
4369 | * (the checksum items are added to the csum tree at the very | |
4370 | * end of btrfs_finish_ordered_io()). Also leave such extents | |
4371 | * outside of our range in the list, since we may have another | |
4372 | * ranged fsync in the near future that needs them. If an extent | |
4373 | * outside our range corresponds to a hole, log it to avoid | |
4374 | * leaving gaps between extents (fsck will complain when we are | |
4375 | * not using the NO_HOLES feature). | |
4376 | */ | |
4377 | if ((em->start > end || em->start + em->len <= start) && | |
4378 | em->block_start != EXTENT_MAP_HOLE) | |
4379 | continue; | |
4380 | ||
5dc562c5 | 4381 | list_del_init(&em->list); |
2ab28f32 JB |
4382 | /* |
4383 | * Just an arbitrary number, this can be really CPU intensive | |
4384 | * once we start getting a lot of extents, and really once we | |
4385 | * have a bunch of extents we just want to commit since it will | |
4386 | * be faster. | |
4387 | */ | |
4388 | if (++num > 32768) { | |
4389 | list_del_init(&tree->modified_extents); | |
4390 | ret = -EFBIG; | |
4391 | goto process; | |
4392 | } | |
4393 | ||
5dc562c5 JB |
4394 | if (em->generation <= test_gen) |
4395 | continue; | |
8c6c5928 | 4396 | |
31d11b83 FM |
4397 | /* We log prealloc extents beyond eof later. */ |
4398 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) && | |
4399 | em->start >= i_size_read(&inode->vfs_inode)) | |
4400 | continue; | |
4401 | ||
ff44c6e3 | 4402 | /* Need a ref to keep it from getting evicted from cache */ |
490b54d6 | 4403 | refcount_inc(&em->refs); |
ff44c6e3 | 4404 | set_bit(EXTENT_FLAG_LOGGING, &em->flags); |
5dc562c5 | 4405 | list_add_tail(&em->list, &extents); |
2ab28f32 | 4406 | num++; |
5dc562c5 JB |
4407 | } |
4408 | ||
4409 | list_sort(NULL, &extents, extent_cmp); | |
2ab28f32 | 4410 | process: |
5dc562c5 JB |
4411 | while (!list_empty(&extents)) { |
4412 | em = list_entry(extents.next, struct extent_map, list); | |
4413 | ||
4414 | list_del_init(&em->list); | |
4415 | ||
4416 | /* | |
4417 | * If we had an error we just need to delete everybody from our | |
4418 | * private list. | |
4419 | */ | |
ff44c6e3 | 4420 | if (ret) { |
201a9038 | 4421 | clear_em_logging(tree, em); |
ff44c6e3 | 4422 | free_extent_map(em); |
5dc562c5 | 4423 | continue; |
ff44c6e3 JB |
4424 | } |
4425 | ||
4426 | write_unlock(&tree->lock); | |
5dc562c5 | 4427 | |
a2120a47 | 4428 | ret = log_one_extent(trans, inode, root, em, path, ctx); |
ff44c6e3 | 4429 | write_lock(&tree->lock); |
201a9038 JB |
4430 | clear_em_logging(tree, em); |
4431 | free_extent_map(em); | |
5dc562c5 | 4432 | } |
ff44c6e3 JB |
4433 | WARN_ON(!list_empty(&extents)); |
4434 | write_unlock(&tree->lock); | |
5dc562c5 | 4435 | |
5dc562c5 | 4436 | btrfs_release_path(path); |
31d11b83 FM |
4437 | if (!ret) |
4438 | ret = btrfs_log_prealloc_extents(trans, inode, path); | |
4439 | ||
5dc562c5 JB |
4440 | return ret; |
4441 | } | |
4442 | ||
481b01c0 | 4443 | static int logged_inode_size(struct btrfs_root *log, struct btrfs_inode *inode, |
1a4bcf47 FM |
4444 | struct btrfs_path *path, u64 *size_ret) |
4445 | { | |
4446 | struct btrfs_key key; | |
4447 | int ret; | |
4448 | ||
481b01c0 | 4449 | key.objectid = btrfs_ino(inode); |
1a4bcf47 FM |
4450 | key.type = BTRFS_INODE_ITEM_KEY; |
4451 | key.offset = 0; | |
4452 | ||
4453 | ret = btrfs_search_slot(NULL, log, &key, path, 0, 0); | |
4454 | if (ret < 0) { | |
4455 | return ret; | |
4456 | } else if (ret > 0) { | |
2f2ff0ee | 4457 | *size_ret = 0; |
1a4bcf47 FM |
4458 | } else { |
4459 | struct btrfs_inode_item *item; | |
4460 | ||
4461 | item = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
4462 | struct btrfs_inode_item); | |
4463 | *size_ret = btrfs_inode_size(path->nodes[0], item); | |
bf504110 FM |
4464 | /* |
4465 | * If the in-memory inode's i_size is smaller then the inode | |
4466 | * size stored in the btree, return the inode's i_size, so | |
4467 | * that we get a correct inode size after replaying the log | |
4468 | * when before a power failure we had a shrinking truncate | |
4469 | * followed by addition of a new name (rename / new hard link). | |
4470 | * Otherwise return the inode size from the btree, to avoid | |
4471 | * data loss when replaying a log due to previously doing a | |
4472 | * write that expands the inode's size and logging a new name | |
4473 | * immediately after. | |
4474 | */ | |
4475 | if (*size_ret > inode->vfs_inode.i_size) | |
4476 | *size_ret = inode->vfs_inode.i_size; | |
1a4bcf47 FM |
4477 | } |
4478 | ||
4479 | btrfs_release_path(path); | |
4480 | return 0; | |
4481 | } | |
4482 | ||
36283bf7 FM |
4483 | /* |
4484 | * At the moment we always log all xattrs. This is to figure out at log replay | |
4485 | * time which xattrs must have their deletion replayed. If a xattr is missing | |
4486 | * in the log tree and exists in the fs/subvol tree, we delete it. This is | |
4487 | * because if a xattr is deleted, the inode is fsynced and a power failure | |
4488 | * happens, causing the log to be replayed the next time the fs is mounted, | |
4489 | * we want the xattr to not exist anymore (same behaviour as other filesystems | |
4490 | * with a journal, ext3/4, xfs, f2fs, etc). | |
4491 | */ | |
4492 | static int btrfs_log_all_xattrs(struct btrfs_trans_handle *trans, | |
4493 | struct btrfs_root *root, | |
1a93c36a | 4494 | struct btrfs_inode *inode, |
36283bf7 FM |
4495 | struct btrfs_path *path, |
4496 | struct btrfs_path *dst_path) | |
4497 | { | |
4498 | int ret; | |
4499 | struct btrfs_key key; | |
1a93c36a | 4500 | const u64 ino = btrfs_ino(inode); |
36283bf7 FM |
4501 | int ins_nr = 0; |
4502 | int start_slot = 0; | |
4503 | ||
4504 | key.objectid = ino; | |
4505 | key.type = BTRFS_XATTR_ITEM_KEY; | |
4506 | key.offset = 0; | |
4507 | ||
4508 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
4509 | if (ret < 0) | |
4510 | return ret; | |
4511 | ||
4512 | while (true) { | |
4513 | int slot = path->slots[0]; | |
4514 | struct extent_buffer *leaf = path->nodes[0]; | |
4515 | int nritems = btrfs_header_nritems(leaf); | |
4516 | ||
4517 | if (slot >= nritems) { | |
4518 | if (ins_nr > 0) { | |
1a93c36a | 4519 | ret = copy_items(trans, inode, dst_path, path, |
0e56315c | 4520 | start_slot, ins_nr, 1, 0); |
36283bf7 FM |
4521 | if (ret < 0) |
4522 | return ret; | |
4523 | ins_nr = 0; | |
4524 | } | |
4525 | ret = btrfs_next_leaf(root, path); | |
4526 | if (ret < 0) | |
4527 | return ret; | |
4528 | else if (ret > 0) | |
4529 | break; | |
4530 | continue; | |
4531 | } | |
4532 | ||
4533 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
4534 | if (key.objectid != ino || key.type != BTRFS_XATTR_ITEM_KEY) | |
4535 | break; | |
4536 | ||
4537 | if (ins_nr == 0) | |
4538 | start_slot = slot; | |
4539 | ins_nr++; | |
4540 | path->slots[0]++; | |
4541 | cond_resched(); | |
4542 | } | |
4543 | if (ins_nr > 0) { | |
1a93c36a | 4544 | ret = copy_items(trans, inode, dst_path, path, |
0e56315c | 4545 | start_slot, ins_nr, 1, 0); |
36283bf7 FM |
4546 | if (ret < 0) |
4547 | return ret; | |
4548 | } | |
4549 | ||
4550 | return 0; | |
4551 | } | |
4552 | ||
a89ca6f2 | 4553 | /* |
0e56315c FM |
4554 | * When using the NO_HOLES feature if we punched a hole that causes the |
4555 | * deletion of entire leafs or all the extent items of the first leaf (the one | |
4556 | * that contains the inode item and references) we may end up not processing | |
4557 | * any extents, because there are no leafs with a generation matching the | |
4558 | * current transaction that have extent items for our inode. So we need to find | |
4559 | * if any holes exist and then log them. We also need to log holes after any | |
4560 | * truncate operation that changes the inode's size. | |
a89ca6f2 | 4561 | */ |
0e56315c FM |
4562 | static int btrfs_log_holes(struct btrfs_trans_handle *trans, |
4563 | struct btrfs_root *root, | |
4564 | struct btrfs_inode *inode, | |
7af59743 | 4565 | struct btrfs_path *path) |
a89ca6f2 | 4566 | { |
0b246afa | 4567 | struct btrfs_fs_info *fs_info = root->fs_info; |
a89ca6f2 | 4568 | struct btrfs_key key; |
a0308dd7 NB |
4569 | const u64 ino = btrfs_ino(inode); |
4570 | const u64 i_size = i_size_read(&inode->vfs_inode); | |
7af59743 | 4571 | u64 prev_extent_end = 0; |
0e56315c | 4572 | int ret; |
a89ca6f2 | 4573 | |
0e56315c | 4574 | if (!btrfs_fs_incompat(fs_info, NO_HOLES) || i_size == 0) |
a89ca6f2 FM |
4575 | return 0; |
4576 | ||
4577 | key.objectid = ino; | |
4578 | key.type = BTRFS_EXTENT_DATA_KEY; | |
7af59743 | 4579 | key.offset = 0; |
a89ca6f2 FM |
4580 | |
4581 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
a89ca6f2 FM |
4582 | if (ret < 0) |
4583 | return ret; | |
4584 | ||
0e56315c | 4585 | while (true) { |
0e56315c | 4586 | struct extent_buffer *leaf = path->nodes[0]; |
a89ca6f2 | 4587 | |
0e56315c FM |
4588 | if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) { |
4589 | ret = btrfs_next_leaf(root, path); | |
4590 | if (ret < 0) | |
4591 | return ret; | |
4592 | if (ret > 0) { | |
4593 | ret = 0; | |
4594 | break; | |
4595 | } | |
4596 | leaf = path->nodes[0]; | |
4597 | } | |
4598 | ||
4599 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
4600 | if (key.objectid != ino || key.type != BTRFS_EXTENT_DATA_KEY) | |
4601 | break; | |
4602 | ||
4603 | /* We have a hole, log it. */ | |
4604 | if (prev_extent_end < key.offset) { | |
7af59743 | 4605 | const u64 hole_len = key.offset - prev_extent_end; |
0e56315c FM |
4606 | |
4607 | /* | |
4608 | * Release the path to avoid deadlocks with other code | |
4609 | * paths that search the root while holding locks on | |
4610 | * leafs from the log root. | |
4611 | */ | |
4612 | btrfs_release_path(path); | |
4613 | ret = btrfs_insert_file_extent(trans, root->log_root, | |
4614 | ino, prev_extent_end, 0, | |
4615 | 0, hole_len, 0, hole_len, | |
4616 | 0, 0, 0); | |
4617 | if (ret < 0) | |
4618 | return ret; | |
4619 | ||
4620 | /* | |
4621 | * Search for the same key again in the root. Since it's | |
4622 | * an extent item and we are holding the inode lock, the | |
4623 | * key must still exist. If it doesn't just emit warning | |
4624 | * and return an error to fall back to a transaction | |
4625 | * commit. | |
4626 | */ | |
4627 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
4628 | if (ret < 0) | |
4629 | return ret; | |
4630 | if (WARN_ON(ret > 0)) | |
4631 | return -ENOENT; | |
4632 | leaf = path->nodes[0]; | |
4633 | } | |
a89ca6f2 | 4634 | |
7af59743 | 4635 | prev_extent_end = btrfs_file_extent_end(path); |
0e56315c FM |
4636 | path->slots[0]++; |
4637 | cond_resched(); | |
a89ca6f2 | 4638 | } |
a89ca6f2 | 4639 | |
7af59743 | 4640 | if (prev_extent_end < i_size) { |
0e56315c | 4641 | u64 hole_len; |
a89ca6f2 | 4642 | |
0e56315c | 4643 | btrfs_release_path(path); |
7af59743 | 4644 | hole_len = ALIGN(i_size - prev_extent_end, fs_info->sectorsize); |
0e56315c FM |
4645 | ret = btrfs_insert_file_extent(trans, root->log_root, |
4646 | ino, prev_extent_end, 0, 0, | |
4647 | hole_len, 0, hole_len, | |
4648 | 0, 0, 0); | |
4649 | if (ret < 0) | |
4650 | return ret; | |
4651 | } | |
4652 | ||
4653 | return 0; | |
a89ca6f2 FM |
4654 | } |
4655 | ||
56f23fdb FM |
4656 | /* |
4657 | * When we are logging a new inode X, check if it doesn't have a reference that | |
4658 | * matches the reference from some other inode Y created in a past transaction | |
4659 | * and that was renamed in the current transaction. If we don't do this, then at | |
4660 | * log replay time we can lose inode Y (and all its files if it's a directory): | |
4661 | * | |
4662 | * mkdir /mnt/x | |
4663 | * echo "hello world" > /mnt/x/foobar | |
4664 | * sync | |
4665 | * mv /mnt/x /mnt/y | |
4666 | * mkdir /mnt/x # or touch /mnt/x | |
4667 | * xfs_io -c fsync /mnt/x | |
4668 | * <power fail> | |
4669 | * mount fs, trigger log replay | |
4670 | * | |
4671 | * After the log replay procedure, we would lose the first directory and all its | |
4672 | * files (file foobar). | |
4673 | * For the case where inode Y is not a directory we simply end up losing it: | |
4674 | * | |
4675 | * echo "123" > /mnt/foo | |
4676 | * sync | |
4677 | * mv /mnt/foo /mnt/bar | |
4678 | * echo "abc" > /mnt/foo | |
4679 | * xfs_io -c fsync /mnt/foo | |
4680 | * <power fail> | |
4681 | * | |
4682 | * We also need this for cases where a snapshot entry is replaced by some other | |
4683 | * entry (file or directory) otherwise we end up with an unreplayable log due to | |
4684 | * attempts to delete the snapshot entry (entry of type BTRFS_ROOT_ITEM_KEY) as | |
4685 | * if it were a regular entry: | |
4686 | * | |
4687 | * mkdir /mnt/x | |
4688 | * btrfs subvolume snapshot /mnt /mnt/x/snap | |
4689 | * btrfs subvolume delete /mnt/x/snap | |
4690 | * rmdir /mnt/x | |
4691 | * mkdir /mnt/x | |
4692 | * fsync /mnt/x or fsync some new file inside it | |
4693 | * <power fail> | |
4694 | * | |
4695 | * The snapshot delete, rmdir of x, mkdir of a new x and the fsync all happen in | |
4696 | * the same transaction. | |
4697 | */ | |
4698 | static int btrfs_check_ref_name_override(struct extent_buffer *eb, | |
4699 | const int slot, | |
4700 | const struct btrfs_key *key, | |
4791c8f1 | 4701 | struct btrfs_inode *inode, |
a3baaf0d | 4702 | u64 *other_ino, u64 *other_parent) |
56f23fdb FM |
4703 | { |
4704 | int ret; | |
4705 | struct btrfs_path *search_path; | |
4706 | char *name = NULL; | |
4707 | u32 name_len = 0; | |
4708 | u32 item_size = btrfs_item_size_nr(eb, slot); | |
4709 | u32 cur_offset = 0; | |
4710 | unsigned long ptr = btrfs_item_ptr_offset(eb, slot); | |
4711 | ||
4712 | search_path = btrfs_alloc_path(); | |
4713 | if (!search_path) | |
4714 | return -ENOMEM; | |
4715 | search_path->search_commit_root = 1; | |
4716 | search_path->skip_locking = 1; | |
4717 | ||
4718 | while (cur_offset < item_size) { | |
4719 | u64 parent; | |
4720 | u32 this_name_len; | |
4721 | u32 this_len; | |
4722 | unsigned long name_ptr; | |
4723 | struct btrfs_dir_item *di; | |
4724 | ||
4725 | if (key->type == BTRFS_INODE_REF_KEY) { | |
4726 | struct btrfs_inode_ref *iref; | |
4727 | ||
4728 | iref = (struct btrfs_inode_ref *)(ptr + cur_offset); | |
4729 | parent = key->offset; | |
4730 | this_name_len = btrfs_inode_ref_name_len(eb, iref); | |
4731 | name_ptr = (unsigned long)(iref + 1); | |
4732 | this_len = sizeof(*iref) + this_name_len; | |
4733 | } else { | |
4734 | struct btrfs_inode_extref *extref; | |
4735 | ||
4736 | extref = (struct btrfs_inode_extref *)(ptr + | |
4737 | cur_offset); | |
4738 | parent = btrfs_inode_extref_parent(eb, extref); | |
4739 | this_name_len = btrfs_inode_extref_name_len(eb, extref); | |
4740 | name_ptr = (unsigned long)&extref->name; | |
4741 | this_len = sizeof(*extref) + this_name_len; | |
4742 | } | |
4743 | ||
4744 | if (this_name_len > name_len) { | |
4745 | char *new_name; | |
4746 | ||
4747 | new_name = krealloc(name, this_name_len, GFP_NOFS); | |
4748 | if (!new_name) { | |
4749 | ret = -ENOMEM; | |
4750 | goto out; | |
4751 | } | |
4752 | name_len = this_name_len; | |
4753 | name = new_name; | |
4754 | } | |
4755 | ||
4756 | read_extent_buffer(eb, name, name_ptr, this_name_len); | |
4791c8f1 NB |
4757 | di = btrfs_lookup_dir_item(NULL, inode->root, search_path, |
4758 | parent, name, this_name_len, 0); | |
56f23fdb | 4759 | if (di && !IS_ERR(di)) { |
44f714da FM |
4760 | struct btrfs_key di_key; |
4761 | ||
4762 | btrfs_dir_item_key_to_cpu(search_path->nodes[0], | |
4763 | di, &di_key); | |
4764 | if (di_key.type == BTRFS_INODE_ITEM_KEY) { | |
6b5fc433 FM |
4765 | if (di_key.objectid != key->objectid) { |
4766 | ret = 1; | |
4767 | *other_ino = di_key.objectid; | |
a3baaf0d | 4768 | *other_parent = parent; |
6b5fc433 FM |
4769 | } else { |
4770 | ret = 0; | |
4771 | } | |
44f714da FM |
4772 | } else { |
4773 | ret = -EAGAIN; | |
4774 | } | |
56f23fdb FM |
4775 | goto out; |
4776 | } else if (IS_ERR(di)) { | |
4777 | ret = PTR_ERR(di); | |
4778 | goto out; | |
4779 | } | |
4780 | btrfs_release_path(search_path); | |
4781 | ||
4782 | cur_offset += this_len; | |
4783 | } | |
4784 | ret = 0; | |
4785 | out: | |
4786 | btrfs_free_path(search_path); | |
4787 | kfree(name); | |
4788 | return ret; | |
4789 | } | |
4790 | ||
6b5fc433 FM |
4791 | struct btrfs_ino_list { |
4792 | u64 ino; | |
a3baaf0d | 4793 | u64 parent; |
6b5fc433 FM |
4794 | struct list_head list; |
4795 | }; | |
4796 | ||
4797 | static int log_conflicting_inodes(struct btrfs_trans_handle *trans, | |
4798 | struct btrfs_root *root, | |
4799 | struct btrfs_path *path, | |
4800 | struct btrfs_log_ctx *ctx, | |
a3baaf0d | 4801 | u64 ino, u64 parent) |
6b5fc433 FM |
4802 | { |
4803 | struct btrfs_ino_list *ino_elem; | |
4804 | LIST_HEAD(inode_list); | |
4805 | int ret = 0; | |
4806 | ||
4807 | ino_elem = kmalloc(sizeof(*ino_elem), GFP_NOFS); | |
4808 | if (!ino_elem) | |
4809 | return -ENOMEM; | |
4810 | ino_elem->ino = ino; | |
a3baaf0d | 4811 | ino_elem->parent = parent; |
6b5fc433 FM |
4812 | list_add_tail(&ino_elem->list, &inode_list); |
4813 | ||
4814 | while (!list_empty(&inode_list)) { | |
4815 | struct btrfs_fs_info *fs_info = root->fs_info; | |
4816 | struct btrfs_key key; | |
4817 | struct inode *inode; | |
4818 | ||
4819 | ino_elem = list_first_entry(&inode_list, struct btrfs_ino_list, | |
4820 | list); | |
4821 | ino = ino_elem->ino; | |
a3baaf0d | 4822 | parent = ino_elem->parent; |
6b5fc433 FM |
4823 | list_del(&ino_elem->list); |
4824 | kfree(ino_elem); | |
4825 | if (ret) | |
4826 | continue; | |
4827 | ||
4828 | btrfs_release_path(path); | |
4829 | ||
0202e83f | 4830 | inode = btrfs_iget(fs_info->sb, ino, root); |
6b5fc433 FM |
4831 | /* |
4832 | * If the other inode that had a conflicting dir entry was | |
a3baaf0d FM |
4833 | * deleted in the current transaction, we need to log its parent |
4834 | * directory. | |
6b5fc433 FM |
4835 | */ |
4836 | if (IS_ERR(inode)) { | |
4837 | ret = PTR_ERR(inode); | |
a3baaf0d | 4838 | if (ret == -ENOENT) { |
0202e83f | 4839 | inode = btrfs_iget(fs_info->sb, parent, root); |
a3baaf0d FM |
4840 | if (IS_ERR(inode)) { |
4841 | ret = PTR_ERR(inode); | |
4842 | } else { | |
4843 | ret = btrfs_log_inode(trans, root, | |
4844 | BTRFS_I(inode), | |
4845 | LOG_OTHER_INODE_ALL, | |
4846 | 0, LLONG_MAX, ctx); | |
410f954c | 4847 | btrfs_add_delayed_iput(inode); |
a3baaf0d FM |
4848 | } |
4849 | } | |
6b5fc433 FM |
4850 | continue; |
4851 | } | |
b5e4ff9d FM |
4852 | /* |
4853 | * If the inode was already logged skip it - otherwise we can | |
4854 | * hit an infinite loop. Example: | |
4855 | * | |
4856 | * From the commit root (previous transaction) we have the | |
4857 | * following inodes: | |
4858 | * | |
4859 | * inode 257 a directory | |
4860 | * inode 258 with references "zz" and "zz_link" on inode 257 | |
4861 | * inode 259 with reference "a" on inode 257 | |
4862 | * | |
4863 | * And in the current (uncommitted) transaction we have: | |
4864 | * | |
4865 | * inode 257 a directory, unchanged | |
4866 | * inode 258 with references "a" and "a2" on inode 257 | |
4867 | * inode 259 with reference "zz_link" on inode 257 | |
4868 | * inode 261 with reference "zz" on inode 257 | |
4869 | * | |
4870 | * When logging inode 261 the following infinite loop could | |
4871 | * happen if we don't skip already logged inodes: | |
4872 | * | |
4873 | * - we detect inode 258 as a conflicting inode, with inode 261 | |
4874 | * on reference "zz", and log it; | |
4875 | * | |
4876 | * - we detect inode 259 as a conflicting inode, with inode 258 | |
4877 | * on reference "a", and log it; | |
4878 | * | |
4879 | * - we detect inode 258 as a conflicting inode, with inode 259 | |
4880 | * on reference "zz_link", and log it - again! After this we | |
4881 | * repeat the above steps forever. | |
4882 | */ | |
4883 | spin_lock(&BTRFS_I(inode)->lock); | |
4884 | /* | |
4885 | * Check the inode's logged_trans only instead of | |
4886 | * btrfs_inode_in_log(). This is because the last_log_commit of | |
4887 | * the inode is not updated when we only log that it exists and | |
4888 | * and it has the full sync bit set (see btrfs_log_inode()). | |
4889 | */ | |
4890 | if (BTRFS_I(inode)->logged_trans == trans->transid) { | |
4891 | spin_unlock(&BTRFS_I(inode)->lock); | |
4892 | btrfs_add_delayed_iput(inode); | |
4893 | continue; | |
4894 | } | |
4895 | spin_unlock(&BTRFS_I(inode)->lock); | |
6b5fc433 FM |
4896 | /* |
4897 | * We are safe logging the other inode without acquiring its | |
4898 | * lock as long as we log with the LOG_INODE_EXISTS mode. We | |
4899 | * are safe against concurrent renames of the other inode as | |
4900 | * well because during a rename we pin the log and update the | |
4901 | * log with the new name before we unpin it. | |
4902 | */ | |
4903 | ret = btrfs_log_inode(trans, root, BTRFS_I(inode), | |
4904 | LOG_OTHER_INODE, 0, LLONG_MAX, ctx); | |
4905 | if (ret) { | |
410f954c | 4906 | btrfs_add_delayed_iput(inode); |
6b5fc433 FM |
4907 | continue; |
4908 | } | |
4909 | ||
4910 | key.objectid = ino; | |
4911 | key.type = BTRFS_INODE_REF_KEY; | |
4912 | key.offset = 0; | |
4913 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
4914 | if (ret < 0) { | |
410f954c | 4915 | btrfs_add_delayed_iput(inode); |
6b5fc433 FM |
4916 | continue; |
4917 | } | |
4918 | ||
4919 | while (true) { | |
4920 | struct extent_buffer *leaf = path->nodes[0]; | |
4921 | int slot = path->slots[0]; | |
4922 | u64 other_ino = 0; | |
a3baaf0d | 4923 | u64 other_parent = 0; |
6b5fc433 FM |
4924 | |
4925 | if (slot >= btrfs_header_nritems(leaf)) { | |
4926 | ret = btrfs_next_leaf(root, path); | |
4927 | if (ret < 0) { | |
4928 | break; | |
4929 | } else if (ret > 0) { | |
4930 | ret = 0; | |
4931 | break; | |
4932 | } | |
4933 | continue; | |
4934 | } | |
4935 | ||
4936 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
4937 | if (key.objectid != ino || | |
4938 | (key.type != BTRFS_INODE_REF_KEY && | |
4939 | key.type != BTRFS_INODE_EXTREF_KEY)) { | |
4940 | ret = 0; | |
4941 | break; | |
4942 | } | |
4943 | ||
4944 | ret = btrfs_check_ref_name_override(leaf, slot, &key, | |
a3baaf0d FM |
4945 | BTRFS_I(inode), &other_ino, |
4946 | &other_parent); | |
6b5fc433 FM |
4947 | if (ret < 0) |
4948 | break; | |
4949 | if (ret > 0) { | |
4950 | ino_elem = kmalloc(sizeof(*ino_elem), GFP_NOFS); | |
4951 | if (!ino_elem) { | |
4952 | ret = -ENOMEM; | |
4953 | break; | |
4954 | } | |
4955 | ino_elem->ino = other_ino; | |
a3baaf0d | 4956 | ino_elem->parent = other_parent; |
6b5fc433 FM |
4957 | list_add_tail(&ino_elem->list, &inode_list); |
4958 | ret = 0; | |
4959 | } | |
4960 | path->slots[0]++; | |
4961 | } | |
410f954c | 4962 | btrfs_add_delayed_iput(inode); |
6b5fc433 FM |
4963 | } |
4964 | ||
4965 | return ret; | |
4966 | } | |
4967 | ||
da447009 FM |
4968 | static int copy_inode_items_to_log(struct btrfs_trans_handle *trans, |
4969 | struct btrfs_inode *inode, | |
4970 | struct btrfs_key *min_key, | |
4971 | const struct btrfs_key *max_key, | |
4972 | struct btrfs_path *path, | |
4973 | struct btrfs_path *dst_path, | |
4974 | const u64 logged_isize, | |
4975 | const bool recursive_logging, | |
4976 | const int inode_only, | |
4977 | struct btrfs_log_ctx *ctx, | |
4978 | bool *need_log_inode_item) | |
4979 | { | |
4980 | struct btrfs_root *root = inode->root; | |
4981 | int ins_start_slot = 0; | |
4982 | int ins_nr = 0; | |
4983 | int ret; | |
4984 | ||
4985 | while (1) { | |
4986 | ret = btrfs_search_forward(root, min_key, path, trans->transid); | |
4987 | if (ret < 0) | |
4988 | return ret; | |
4989 | if (ret > 0) { | |
4990 | ret = 0; | |
4991 | break; | |
4992 | } | |
4993 | again: | |
4994 | /* Note, ins_nr might be > 0 here, cleanup outside the loop */ | |
4995 | if (min_key->objectid != max_key->objectid) | |
4996 | break; | |
4997 | if (min_key->type > max_key->type) | |
4998 | break; | |
4999 | ||
5000 | if (min_key->type == BTRFS_INODE_ITEM_KEY) | |
5001 | *need_log_inode_item = false; | |
5002 | ||
5003 | if ((min_key->type == BTRFS_INODE_REF_KEY || | |
5004 | min_key->type == BTRFS_INODE_EXTREF_KEY) && | |
5005 | inode->generation == trans->transid && | |
5006 | !recursive_logging) { | |
5007 | u64 other_ino = 0; | |
5008 | u64 other_parent = 0; | |
5009 | ||
5010 | ret = btrfs_check_ref_name_override(path->nodes[0], | |
5011 | path->slots[0], min_key, inode, | |
5012 | &other_ino, &other_parent); | |
5013 | if (ret < 0) { | |
5014 | return ret; | |
5015 | } else if (ret > 0 && ctx && | |
5016 | other_ino != btrfs_ino(BTRFS_I(ctx->inode))) { | |
5017 | if (ins_nr > 0) { | |
5018 | ins_nr++; | |
5019 | } else { | |
5020 | ins_nr = 1; | |
5021 | ins_start_slot = path->slots[0]; | |
5022 | } | |
5023 | ret = copy_items(trans, inode, dst_path, path, | |
5024 | ins_start_slot, ins_nr, | |
5025 | inode_only, logged_isize); | |
5026 | if (ret < 0) | |
5027 | return ret; | |
5028 | ins_nr = 0; | |
5029 | ||
5030 | ret = log_conflicting_inodes(trans, root, path, | |
5031 | ctx, other_ino, other_parent); | |
5032 | if (ret) | |
5033 | return ret; | |
5034 | btrfs_release_path(path); | |
5035 | goto next_key; | |
5036 | } | |
5037 | } | |
5038 | ||
5039 | /* Skip xattrs, we log them later with btrfs_log_all_xattrs() */ | |
5040 | if (min_key->type == BTRFS_XATTR_ITEM_KEY) { | |
5041 | if (ins_nr == 0) | |
5042 | goto next_slot; | |
5043 | ret = copy_items(trans, inode, dst_path, path, | |
5044 | ins_start_slot, | |
5045 | ins_nr, inode_only, logged_isize); | |
5046 | if (ret < 0) | |
5047 | return ret; | |
5048 | ins_nr = 0; | |
5049 | goto next_slot; | |
5050 | } | |
5051 | ||
5052 | if (ins_nr && ins_start_slot + ins_nr == path->slots[0]) { | |
5053 | ins_nr++; | |
5054 | goto next_slot; | |
5055 | } else if (!ins_nr) { | |
5056 | ins_start_slot = path->slots[0]; | |
5057 | ins_nr = 1; | |
5058 | goto next_slot; | |
5059 | } | |
5060 | ||
5061 | ret = copy_items(trans, inode, dst_path, path, ins_start_slot, | |
5062 | ins_nr, inode_only, logged_isize); | |
5063 | if (ret < 0) | |
5064 | return ret; | |
5065 | ins_nr = 1; | |
5066 | ins_start_slot = path->slots[0]; | |
5067 | next_slot: | |
5068 | path->slots[0]++; | |
5069 | if (path->slots[0] < btrfs_header_nritems(path->nodes[0])) { | |
5070 | btrfs_item_key_to_cpu(path->nodes[0], min_key, | |
5071 | path->slots[0]); | |
5072 | goto again; | |
5073 | } | |
5074 | if (ins_nr) { | |
5075 | ret = copy_items(trans, inode, dst_path, path, | |
5076 | ins_start_slot, ins_nr, inode_only, | |
5077 | logged_isize); | |
5078 | if (ret < 0) | |
5079 | return ret; | |
5080 | ins_nr = 0; | |
5081 | } | |
5082 | btrfs_release_path(path); | |
5083 | next_key: | |
5084 | if (min_key->offset < (u64)-1) { | |
5085 | min_key->offset++; | |
5086 | } else if (min_key->type < max_key->type) { | |
5087 | min_key->type++; | |
5088 | min_key->offset = 0; | |
5089 | } else { | |
5090 | break; | |
5091 | } | |
5092 | } | |
5093 | if (ins_nr) | |
5094 | ret = copy_items(trans, inode, dst_path, path, ins_start_slot, | |
5095 | ins_nr, inode_only, logged_isize); | |
5096 | ||
5097 | return ret; | |
5098 | } | |
5099 | ||
e02119d5 CM |
5100 | /* log a single inode in the tree log. |
5101 | * At least one parent directory for this inode must exist in the tree | |
5102 | * or be logged already. | |
5103 | * | |
5104 | * Any items from this inode changed by the current transaction are copied | |
5105 | * to the log tree. An extra reference is taken on any extents in this | |
5106 | * file, allowing us to avoid a whole pile of corner cases around logging | |
5107 | * blocks that have been removed from the tree. | |
5108 | * | |
5109 | * See LOG_INODE_ALL and related defines for a description of what inode_only | |
5110 | * does. | |
5111 | * | |
5112 | * This handles both files and directories. | |
5113 | */ | |
12fcfd22 | 5114 | static int btrfs_log_inode(struct btrfs_trans_handle *trans, |
a59108a7 | 5115 | struct btrfs_root *root, struct btrfs_inode *inode, |
49dae1bc | 5116 | int inode_only, |
7af59743 FM |
5117 | const loff_t start, |
5118 | const loff_t end, | |
8407f553 | 5119 | struct btrfs_log_ctx *ctx) |
e02119d5 CM |
5120 | { |
5121 | struct btrfs_path *path; | |
5122 | struct btrfs_path *dst_path; | |
5123 | struct btrfs_key min_key; | |
5124 | struct btrfs_key max_key; | |
5125 | struct btrfs_root *log = root->log_root; | |
4a500fd1 | 5126 | int err = 0; |
8c8648dd | 5127 | int ret = 0; |
5dc562c5 | 5128 | bool fast_search = false; |
a59108a7 NB |
5129 | u64 ino = btrfs_ino(inode); |
5130 | struct extent_map_tree *em_tree = &inode->extent_tree; | |
1a4bcf47 | 5131 | u64 logged_isize = 0; |
e4545de5 | 5132 | bool need_log_inode_item = true; |
9a8fca62 | 5133 | bool xattrs_logged = false; |
a3baaf0d | 5134 | bool recursive_logging = false; |
e02119d5 | 5135 | |
e02119d5 | 5136 | path = btrfs_alloc_path(); |
5df67083 TI |
5137 | if (!path) |
5138 | return -ENOMEM; | |
e02119d5 | 5139 | dst_path = btrfs_alloc_path(); |
5df67083 TI |
5140 | if (!dst_path) { |
5141 | btrfs_free_path(path); | |
5142 | return -ENOMEM; | |
5143 | } | |
e02119d5 | 5144 | |
33345d01 | 5145 | min_key.objectid = ino; |
e02119d5 CM |
5146 | min_key.type = BTRFS_INODE_ITEM_KEY; |
5147 | min_key.offset = 0; | |
5148 | ||
33345d01 | 5149 | max_key.objectid = ino; |
12fcfd22 | 5150 | |
12fcfd22 | 5151 | |
5dc562c5 | 5152 | /* today the code can only do partial logging of directories */ |
a59108a7 | 5153 | if (S_ISDIR(inode->vfs_inode.i_mode) || |
5269b67e | 5154 | (!test_bit(BTRFS_INODE_NEEDS_FULL_SYNC, |
a59108a7 | 5155 | &inode->runtime_flags) && |
781feef7 | 5156 | inode_only >= LOG_INODE_EXISTS)) |
e02119d5 CM |
5157 | max_key.type = BTRFS_XATTR_ITEM_KEY; |
5158 | else | |
5159 | max_key.type = (u8)-1; | |
5160 | max_key.offset = (u64)-1; | |
5161 | ||
2c2c452b | 5162 | /* |
5aa7d1a7 FM |
5163 | * Only run delayed items if we are a directory. We want to make sure |
5164 | * all directory indexes hit the fs/subvolume tree so we can find them | |
5165 | * and figure out which index ranges have to be logged. | |
5166 | * | |
8c8648dd FM |
5167 | * Otherwise commit the delayed inode only if the full sync flag is set, |
5168 | * as we want to make sure an up to date version is in the subvolume | |
5169 | * tree so copy_inode_items_to_log() / copy_items() can find it and copy | |
5170 | * it to the log tree. For a non full sync, we always log the inode item | |
5171 | * based on the in-memory struct btrfs_inode which is always up to date. | |
2c2c452b | 5172 | */ |
5aa7d1a7 | 5173 | if (S_ISDIR(inode->vfs_inode.i_mode)) |
a59108a7 | 5174 | ret = btrfs_commit_inode_delayed_items(trans, inode); |
8c8648dd | 5175 | else if (test_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &inode->runtime_flags)) |
a59108a7 | 5176 | ret = btrfs_commit_inode_delayed_inode(inode); |
2c2c452b FM |
5177 | |
5178 | if (ret) { | |
5179 | btrfs_free_path(path); | |
5180 | btrfs_free_path(dst_path); | |
5181 | return ret; | |
16cdcec7 MX |
5182 | } |
5183 | ||
a3baaf0d FM |
5184 | if (inode_only == LOG_OTHER_INODE || inode_only == LOG_OTHER_INODE_ALL) { |
5185 | recursive_logging = true; | |
5186 | if (inode_only == LOG_OTHER_INODE) | |
5187 | inode_only = LOG_INODE_EXISTS; | |
5188 | else | |
5189 | inode_only = LOG_INODE_ALL; | |
a59108a7 | 5190 | mutex_lock_nested(&inode->log_mutex, SINGLE_DEPTH_NESTING); |
781feef7 | 5191 | } else { |
a59108a7 | 5192 | mutex_lock(&inode->log_mutex); |
781feef7 | 5193 | } |
e02119d5 CM |
5194 | |
5195 | /* | |
5196 | * a brute force approach to making sure we get the most uptodate | |
5197 | * copies of everything. | |
5198 | */ | |
a59108a7 | 5199 | if (S_ISDIR(inode->vfs_inode.i_mode)) { |
e02119d5 CM |
5200 | int max_key_type = BTRFS_DIR_LOG_INDEX_KEY; |
5201 | ||
4f764e51 FM |
5202 | if (inode_only == LOG_INODE_EXISTS) |
5203 | max_key_type = BTRFS_XATTR_ITEM_KEY; | |
33345d01 | 5204 | ret = drop_objectid_items(trans, log, path, ino, max_key_type); |
e02119d5 | 5205 | } else { |
1a4bcf47 FM |
5206 | if (inode_only == LOG_INODE_EXISTS) { |
5207 | /* | |
5208 | * Make sure the new inode item we write to the log has | |
5209 | * the same isize as the current one (if it exists). | |
5210 | * This is necessary to prevent data loss after log | |
5211 | * replay, and also to prevent doing a wrong expanding | |
5212 | * truncate - for e.g. create file, write 4K into offset | |
5213 | * 0, fsync, write 4K into offset 4096, add hard link, | |
5214 | * fsync some other file (to sync log), power fail - if | |
5215 | * we use the inode's current i_size, after log replay | |
5216 | * we get a 8Kb file, with the last 4Kb extent as a hole | |
5217 | * (zeroes), as if an expanding truncate happened, | |
5218 | * instead of getting a file of 4Kb only. | |
5219 | */ | |
a59108a7 | 5220 | err = logged_inode_size(log, inode, path, &logged_isize); |
1a4bcf47 FM |
5221 | if (err) |
5222 | goto out_unlock; | |
5223 | } | |
a742994a | 5224 | if (test_bit(BTRFS_INODE_NEEDS_FULL_SYNC, |
a59108a7 | 5225 | &inode->runtime_flags)) { |
a742994a | 5226 | if (inode_only == LOG_INODE_EXISTS) { |
4f764e51 | 5227 | max_key.type = BTRFS_XATTR_ITEM_KEY; |
a742994a FM |
5228 | ret = drop_objectid_items(trans, log, path, ino, |
5229 | max_key.type); | |
5230 | } else { | |
5231 | clear_bit(BTRFS_INODE_NEEDS_FULL_SYNC, | |
a59108a7 | 5232 | &inode->runtime_flags); |
a742994a | 5233 | clear_bit(BTRFS_INODE_COPY_EVERYTHING, |
a59108a7 | 5234 | &inode->runtime_flags); |
28ed1345 CM |
5235 | while(1) { |
5236 | ret = btrfs_truncate_inode_items(trans, | |
a59108a7 | 5237 | log, &inode->vfs_inode, 0, 0); |
28ed1345 CM |
5238 | if (ret != -EAGAIN) |
5239 | break; | |
5240 | } | |
a742994a | 5241 | } |
4f764e51 | 5242 | } else if (test_and_clear_bit(BTRFS_INODE_COPY_EVERYTHING, |
a59108a7 | 5243 | &inode->runtime_flags) || |
6cfab851 | 5244 | inode_only == LOG_INODE_EXISTS) { |
4f764e51 | 5245 | if (inode_only == LOG_INODE_ALL) |
183f37fa | 5246 | fast_search = true; |
4f764e51 | 5247 | max_key.type = BTRFS_XATTR_ITEM_KEY; |
5dc562c5 | 5248 | ret = drop_objectid_items(trans, log, path, ino, |
e9976151 | 5249 | max_key.type); |
a95249b3 JB |
5250 | } else { |
5251 | if (inode_only == LOG_INODE_ALL) | |
5252 | fast_search = true; | |
a95249b3 | 5253 | goto log_extents; |
5dc562c5 | 5254 | } |
a95249b3 | 5255 | |
e02119d5 | 5256 | } |
4a500fd1 YZ |
5257 | if (ret) { |
5258 | err = ret; | |
5259 | goto out_unlock; | |
5260 | } | |
e02119d5 | 5261 | |
da447009 FM |
5262 | err = copy_inode_items_to_log(trans, inode, &min_key, &max_key, |
5263 | path, dst_path, logged_isize, | |
7af59743 FM |
5264 | recursive_logging, inode_only, ctx, |
5265 | &need_log_inode_item); | |
da447009 FM |
5266 | if (err) |
5267 | goto out_unlock; | |
5dc562c5 | 5268 | |
36283bf7 FM |
5269 | btrfs_release_path(path); |
5270 | btrfs_release_path(dst_path); | |
a59108a7 | 5271 | err = btrfs_log_all_xattrs(trans, root, inode, path, dst_path); |
36283bf7 FM |
5272 | if (err) |
5273 | goto out_unlock; | |
9a8fca62 | 5274 | xattrs_logged = true; |
a89ca6f2 FM |
5275 | if (max_key.type >= BTRFS_EXTENT_DATA_KEY && !fast_search) { |
5276 | btrfs_release_path(path); | |
5277 | btrfs_release_path(dst_path); | |
7af59743 | 5278 | err = btrfs_log_holes(trans, root, inode, path); |
a89ca6f2 FM |
5279 | if (err) |
5280 | goto out_unlock; | |
5281 | } | |
a95249b3 | 5282 | log_extents: |
f3b15ccd JB |
5283 | btrfs_release_path(path); |
5284 | btrfs_release_path(dst_path); | |
e4545de5 | 5285 | if (need_log_inode_item) { |
a59108a7 | 5286 | err = log_inode_item(trans, log, dst_path, inode); |
9a8fca62 FM |
5287 | if (!err && !xattrs_logged) { |
5288 | err = btrfs_log_all_xattrs(trans, root, inode, path, | |
5289 | dst_path); | |
5290 | btrfs_release_path(path); | |
5291 | } | |
e4545de5 FM |
5292 | if (err) |
5293 | goto out_unlock; | |
5294 | } | |
5dc562c5 | 5295 | if (fast_search) { |
a59108a7 | 5296 | ret = btrfs_log_changed_extents(trans, root, inode, dst_path, |
a2120a47 | 5297 | ctx, start, end); |
5dc562c5 JB |
5298 | if (ret) { |
5299 | err = ret; | |
5300 | goto out_unlock; | |
5301 | } | |
d006a048 | 5302 | } else if (inode_only == LOG_INODE_ALL) { |
06d3d22b LB |
5303 | struct extent_map *em, *n; |
5304 | ||
49dae1bc FM |
5305 | write_lock(&em_tree->lock); |
5306 | /* | |
5307 | * We can't just remove every em if we're called for a ranged | |
5308 | * fsync - that is, one that doesn't cover the whole possible | |
5309 | * file range (0 to LLONG_MAX). This is because we can have | |
5310 | * em's that fall outside the range we're logging and therefore | |
5311 | * their ordered operations haven't completed yet | |
5312 | * (btrfs_finish_ordered_io() not invoked yet). This means we | |
5313 | * didn't get their respective file extent item in the fs/subvol | |
5314 | * tree yet, and need to let the next fast fsync (one which | |
5315 | * consults the list of modified extent maps) find the em so | |
5316 | * that it logs a matching file extent item and waits for the | |
5317 | * respective ordered operation to complete (if it's still | |
5318 | * running). | |
5319 | * | |
5320 | * Removing every em outside the range we're logging would make | |
5321 | * the next fast fsync not log their matching file extent items, | |
5322 | * therefore making us lose data after a log replay. | |
5323 | */ | |
5324 | list_for_each_entry_safe(em, n, &em_tree->modified_extents, | |
5325 | list) { | |
5326 | const u64 mod_end = em->mod_start + em->mod_len - 1; | |
5327 | ||
5328 | if (em->mod_start >= start && mod_end <= end) | |
5329 | list_del_init(&em->list); | |
5330 | } | |
5331 | write_unlock(&em_tree->lock); | |
5dc562c5 JB |
5332 | } |
5333 | ||
a59108a7 NB |
5334 | if (inode_only == LOG_INODE_ALL && S_ISDIR(inode->vfs_inode.i_mode)) { |
5335 | ret = log_directory_changes(trans, root, inode, path, dst_path, | |
5336 | ctx); | |
4a500fd1 YZ |
5337 | if (ret) { |
5338 | err = ret; | |
5339 | goto out_unlock; | |
5340 | } | |
e02119d5 | 5341 | } |
49dae1bc | 5342 | |
d1d832a0 FM |
5343 | /* |
5344 | * Don't update last_log_commit if we logged that an inode exists after | |
5345 | * it was loaded to memory (full_sync bit set). | |
5346 | * This is to prevent data loss when we do a write to the inode, then | |
5347 | * the inode gets evicted after all delalloc was flushed, then we log | |
5348 | * it exists (due to a rename for example) and then fsync it. This last | |
5349 | * fsync would do nothing (not logging the extents previously written). | |
5350 | */ | |
a59108a7 NB |
5351 | spin_lock(&inode->lock); |
5352 | inode->logged_trans = trans->transid; | |
d1d832a0 FM |
5353 | if (inode_only != LOG_INODE_EXISTS || |
5354 | !test_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &inode->runtime_flags)) | |
5355 | inode->last_log_commit = inode->last_sub_trans; | |
a59108a7 | 5356 | spin_unlock(&inode->lock); |
4a500fd1 | 5357 | out_unlock: |
a59108a7 | 5358 | mutex_unlock(&inode->log_mutex); |
e02119d5 CM |
5359 | |
5360 | btrfs_free_path(path); | |
5361 | btrfs_free_path(dst_path); | |
4a500fd1 | 5362 | return err; |
e02119d5 CM |
5363 | } |
5364 | ||
2be63d5c FM |
5365 | /* |
5366 | * Check if we must fallback to a transaction commit when logging an inode. | |
5367 | * This must be called after logging the inode and is used only in the context | |
5368 | * when fsyncing an inode requires the need to log some other inode - in which | |
5369 | * case we can't lock the i_mutex of each other inode we need to log as that | |
5370 | * can lead to deadlocks with concurrent fsync against other inodes (as we can | |
5371 | * log inodes up or down in the hierarchy) or rename operations for example. So | |
5372 | * we take the log_mutex of the inode after we have logged it and then check for | |
5373 | * its last_unlink_trans value - this is safe because any task setting | |
5374 | * last_unlink_trans must take the log_mutex and it must do this before it does | |
5375 | * the actual unlink operation, so if we do this check before a concurrent task | |
5376 | * sets last_unlink_trans it means we've logged a consistent version/state of | |
5377 | * all the inode items, otherwise we are not sure and must do a transaction | |
01327610 | 5378 | * commit (the concurrent task might have only updated last_unlink_trans before |
2be63d5c FM |
5379 | * we logged the inode or it might have also done the unlink). |
5380 | */ | |
5381 | static bool btrfs_must_commit_transaction(struct btrfs_trans_handle *trans, | |
ab1717b2 | 5382 | struct btrfs_inode *inode) |
2be63d5c | 5383 | { |
ab1717b2 | 5384 | struct btrfs_fs_info *fs_info = inode->root->fs_info; |
2be63d5c FM |
5385 | bool ret = false; |
5386 | ||
ab1717b2 NB |
5387 | mutex_lock(&inode->log_mutex); |
5388 | if (inode->last_unlink_trans > fs_info->last_trans_committed) { | |
2be63d5c FM |
5389 | /* |
5390 | * Make sure any commits to the log are forced to be full | |
5391 | * commits. | |
5392 | */ | |
90787766 | 5393 | btrfs_set_log_full_commit(trans); |
2be63d5c FM |
5394 | ret = true; |
5395 | } | |
ab1717b2 | 5396 | mutex_unlock(&inode->log_mutex); |
2be63d5c FM |
5397 | |
5398 | return ret; | |
5399 | } | |
5400 | ||
12fcfd22 CM |
5401 | /* |
5402 | * follow the dentry parent pointers up the chain and see if any | |
5403 | * of the directories in it require a full commit before they can | |
5404 | * be logged. Returns zero if nothing special needs to be done or 1 if | |
5405 | * a full commit is required. | |
5406 | */ | |
5407 | static noinline int check_parent_dirs_for_sync(struct btrfs_trans_handle *trans, | |
aefa6115 | 5408 | struct btrfs_inode *inode, |
12fcfd22 CM |
5409 | struct dentry *parent, |
5410 | struct super_block *sb, | |
5411 | u64 last_committed) | |
e02119d5 | 5412 | { |
12fcfd22 | 5413 | int ret = 0; |
6a912213 | 5414 | struct dentry *old_parent = NULL; |
e02119d5 | 5415 | |
af4176b4 CM |
5416 | /* |
5417 | * for regular files, if its inode is already on disk, we don't | |
5418 | * have to worry about the parents at all. This is because | |
5419 | * we can use the last_unlink_trans field to record renames | |
5420 | * and other fun in this file. | |
5421 | */ | |
aefa6115 NB |
5422 | if (S_ISREG(inode->vfs_inode.i_mode) && |
5423 | inode->generation <= last_committed && | |
5424 | inode->last_unlink_trans <= last_committed) | |
5425 | goto out; | |
af4176b4 | 5426 | |
aefa6115 | 5427 | if (!S_ISDIR(inode->vfs_inode.i_mode)) { |
fc64005c | 5428 | if (!parent || d_really_is_negative(parent) || sb != parent->d_sb) |
12fcfd22 | 5429 | goto out; |
aefa6115 | 5430 | inode = BTRFS_I(d_inode(parent)); |
12fcfd22 CM |
5431 | } |
5432 | ||
5433 | while (1) { | |
aefa6115 | 5434 | if (btrfs_must_commit_transaction(trans, inode)) { |
12fcfd22 CM |
5435 | ret = 1; |
5436 | break; | |
5437 | } | |
5438 | ||
fc64005c | 5439 | if (!parent || d_really_is_negative(parent) || sb != parent->d_sb) |
12fcfd22 CM |
5440 | break; |
5441 | ||
44f714da | 5442 | if (IS_ROOT(parent)) { |
aefa6115 NB |
5443 | inode = BTRFS_I(d_inode(parent)); |
5444 | if (btrfs_must_commit_transaction(trans, inode)) | |
44f714da | 5445 | ret = 1; |
12fcfd22 | 5446 | break; |
44f714da | 5447 | } |
12fcfd22 | 5448 | |
6a912213 JB |
5449 | parent = dget_parent(parent); |
5450 | dput(old_parent); | |
5451 | old_parent = parent; | |
aefa6115 | 5452 | inode = BTRFS_I(d_inode(parent)); |
12fcfd22 CM |
5453 | |
5454 | } | |
6a912213 | 5455 | dput(old_parent); |
12fcfd22 | 5456 | out: |
e02119d5 CM |
5457 | return ret; |
5458 | } | |
5459 | ||
2f2ff0ee FM |
5460 | struct btrfs_dir_list { |
5461 | u64 ino; | |
5462 | struct list_head list; | |
5463 | }; | |
5464 | ||
5465 | /* | |
5466 | * Log the inodes of the new dentries of a directory. See log_dir_items() for | |
5467 | * details about the why it is needed. | |
5468 | * This is a recursive operation - if an existing dentry corresponds to a | |
5469 | * directory, that directory's new entries are logged too (same behaviour as | |
5470 | * ext3/4, xfs, f2fs, reiserfs, nilfs2). Note that when logging the inodes | |
5471 | * the dentries point to we do not lock their i_mutex, otherwise lockdep | |
5472 | * complains about the following circular lock dependency / possible deadlock: | |
5473 | * | |
5474 | * CPU0 CPU1 | |
5475 | * ---- ---- | |
5476 | * lock(&type->i_mutex_dir_key#3/2); | |
5477 | * lock(sb_internal#2); | |
5478 | * lock(&type->i_mutex_dir_key#3/2); | |
5479 | * lock(&sb->s_type->i_mutex_key#14); | |
5480 | * | |
5481 | * Where sb_internal is the lock (a counter that works as a lock) acquired by | |
5482 | * sb_start_intwrite() in btrfs_start_transaction(). | |
5483 | * Not locking i_mutex of the inodes is still safe because: | |
5484 | * | |
5485 | * 1) For regular files we log with a mode of LOG_INODE_EXISTS. It's possible | |
5486 | * that while logging the inode new references (names) are added or removed | |
5487 | * from the inode, leaving the logged inode item with a link count that does | |
5488 | * not match the number of logged inode reference items. This is fine because | |
5489 | * at log replay time we compute the real number of links and correct the | |
5490 | * link count in the inode item (see replay_one_buffer() and | |
5491 | * link_to_fixup_dir()); | |
5492 | * | |
5493 | * 2) For directories we log with a mode of LOG_INODE_ALL. It's possible that | |
5494 | * while logging the inode's items new items with keys BTRFS_DIR_ITEM_KEY and | |
5495 | * BTRFS_DIR_INDEX_KEY are added to fs/subvol tree and the logged inode item | |
5496 | * has a size that doesn't match the sum of the lengths of all the logged | |
5497 | * names. This does not result in a problem because if a dir_item key is | |
5498 | * logged but its matching dir_index key is not logged, at log replay time we | |
5499 | * don't use it to replay the respective name (see replay_one_name()). On the | |
5500 | * other hand if only the dir_index key ends up being logged, the respective | |
5501 | * name is added to the fs/subvol tree with both the dir_item and dir_index | |
5502 | * keys created (see replay_one_name()). | |
5503 | * The directory's inode item with a wrong i_size is not a problem as well, | |
5504 | * since we don't use it at log replay time to set the i_size in the inode | |
5505 | * item of the fs/subvol tree (see overwrite_item()). | |
5506 | */ | |
5507 | static int log_new_dir_dentries(struct btrfs_trans_handle *trans, | |
5508 | struct btrfs_root *root, | |
51cc0d32 | 5509 | struct btrfs_inode *start_inode, |
2f2ff0ee FM |
5510 | struct btrfs_log_ctx *ctx) |
5511 | { | |
0b246afa | 5512 | struct btrfs_fs_info *fs_info = root->fs_info; |
2f2ff0ee FM |
5513 | struct btrfs_root *log = root->log_root; |
5514 | struct btrfs_path *path; | |
5515 | LIST_HEAD(dir_list); | |
5516 | struct btrfs_dir_list *dir_elem; | |
5517 | int ret = 0; | |
5518 | ||
5519 | path = btrfs_alloc_path(); | |
5520 | if (!path) | |
5521 | return -ENOMEM; | |
5522 | ||
5523 | dir_elem = kmalloc(sizeof(*dir_elem), GFP_NOFS); | |
5524 | if (!dir_elem) { | |
5525 | btrfs_free_path(path); | |
5526 | return -ENOMEM; | |
5527 | } | |
51cc0d32 | 5528 | dir_elem->ino = btrfs_ino(start_inode); |
2f2ff0ee FM |
5529 | list_add_tail(&dir_elem->list, &dir_list); |
5530 | ||
5531 | while (!list_empty(&dir_list)) { | |
5532 | struct extent_buffer *leaf; | |
5533 | struct btrfs_key min_key; | |
5534 | int nritems; | |
5535 | int i; | |
5536 | ||
5537 | dir_elem = list_first_entry(&dir_list, struct btrfs_dir_list, | |
5538 | list); | |
5539 | if (ret) | |
5540 | goto next_dir_inode; | |
5541 | ||
5542 | min_key.objectid = dir_elem->ino; | |
5543 | min_key.type = BTRFS_DIR_ITEM_KEY; | |
5544 | min_key.offset = 0; | |
5545 | again: | |
5546 | btrfs_release_path(path); | |
5547 | ret = btrfs_search_forward(log, &min_key, path, trans->transid); | |
5548 | if (ret < 0) { | |
5549 | goto next_dir_inode; | |
5550 | } else if (ret > 0) { | |
5551 | ret = 0; | |
5552 | goto next_dir_inode; | |
5553 | } | |
5554 | ||
5555 | process_leaf: | |
5556 | leaf = path->nodes[0]; | |
5557 | nritems = btrfs_header_nritems(leaf); | |
5558 | for (i = path->slots[0]; i < nritems; i++) { | |
5559 | struct btrfs_dir_item *di; | |
5560 | struct btrfs_key di_key; | |
5561 | struct inode *di_inode; | |
5562 | struct btrfs_dir_list *new_dir_elem; | |
5563 | int log_mode = LOG_INODE_EXISTS; | |
5564 | int type; | |
5565 | ||
5566 | btrfs_item_key_to_cpu(leaf, &min_key, i); | |
5567 | if (min_key.objectid != dir_elem->ino || | |
5568 | min_key.type != BTRFS_DIR_ITEM_KEY) | |
5569 | goto next_dir_inode; | |
5570 | ||
5571 | di = btrfs_item_ptr(leaf, i, struct btrfs_dir_item); | |
5572 | type = btrfs_dir_type(leaf, di); | |
5573 | if (btrfs_dir_transid(leaf, di) < trans->transid && | |
5574 | type != BTRFS_FT_DIR) | |
5575 | continue; | |
5576 | btrfs_dir_item_key_to_cpu(leaf, di, &di_key); | |
5577 | if (di_key.type == BTRFS_ROOT_ITEM_KEY) | |
5578 | continue; | |
5579 | ||
ec125cfb | 5580 | btrfs_release_path(path); |
0202e83f | 5581 | di_inode = btrfs_iget(fs_info->sb, di_key.objectid, root); |
2f2ff0ee FM |
5582 | if (IS_ERR(di_inode)) { |
5583 | ret = PTR_ERR(di_inode); | |
5584 | goto next_dir_inode; | |
5585 | } | |
5586 | ||
0f8939b8 | 5587 | if (btrfs_inode_in_log(BTRFS_I(di_inode), trans->transid)) { |
410f954c | 5588 | btrfs_add_delayed_iput(di_inode); |
ec125cfb | 5589 | break; |
2f2ff0ee FM |
5590 | } |
5591 | ||
5592 | ctx->log_new_dentries = false; | |
3f9749f6 | 5593 | if (type == BTRFS_FT_DIR || type == BTRFS_FT_SYMLINK) |
2f2ff0ee | 5594 | log_mode = LOG_INODE_ALL; |
a59108a7 | 5595 | ret = btrfs_log_inode(trans, root, BTRFS_I(di_inode), |
2f2ff0ee | 5596 | log_mode, 0, LLONG_MAX, ctx); |
2be63d5c | 5597 | if (!ret && |
ab1717b2 | 5598 | btrfs_must_commit_transaction(trans, BTRFS_I(di_inode))) |
2be63d5c | 5599 | ret = 1; |
410f954c | 5600 | btrfs_add_delayed_iput(di_inode); |
2f2ff0ee FM |
5601 | if (ret) |
5602 | goto next_dir_inode; | |
5603 | if (ctx->log_new_dentries) { | |
5604 | new_dir_elem = kmalloc(sizeof(*new_dir_elem), | |
5605 | GFP_NOFS); | |
5606 | if (!new_dir_elem) { | |
5607 | ret = -ENOMEM; | |
5608 | goto next_dir_inode; | |
5609 | } | |
5610 | new_dir_elem->ino = di_key.objectid; | |
5611 | list_add_tail(&new_dir_elem->list, &dir_list); | |
5612 | } | |
5613 | break; | |
5614 | } | |
5615 | if (i == nritems) { | |
5616 | ret = btrfs_next_leaf(log, path); | |
5617 | if (ret < 0) { | |
5618 | goto next_dir_inode; | |
5619 | } else if (ret > 0) { | |
5620 | ret = 0; | |
5621 | goto next_dir_inode; | |
5622 | } | |
5623 | goto process_leaf; | |
5624 | } | |
5625 | if (min_key.offset < (u64)-1) { | |
5626 | min_key.offset++; | |
5627 | goto again; | |
5628 | } | |
5629 | next_dir_inode: | |
5630 | list_del(&dir_elem->list); | |
5631 | kfree(dir_elem); | |
5632 | } | |
5633 | ||
5634 | btrfs_free_path(path); | |
5635 | return ret; | |
5636 | } | |
5637 | ||
18aa0922 | 5638 | static int btrfs_log_all_parents(struct btrfs_trans_handle *trans, |
d0a0b78d | 5639 | struct btrfs_inode *inode, |
18aa0922 FM |
5640 | struct btrfs_log_ctx *ctx) |
5641 | { | |
3ffbd68c | 5642 | struct btrfs_fs_info *fs_info = trans->fs_info; |
18aa0922 FM |
5643 | int ret; |
5644 | struct btrfs_path *path; | |
5645 | struct btrfs_key key; | |
d0a0b78d NB |
5646 | struct btrfs_root *root = inode->root; |
5647 | const u64 ino = btrfs_ino(inode); | |
18aa0922 FM |
5648 | |
5649 | path = btrfs_alloc_path(); | |
5650 | if (!path) | |
5651 | return -ENOMEM; | |
5652 | path->skip_locking = 1; | |
5653 | path->search_commit_root = 1; | |
5654 | ||
5655 | key.objectid = ino; | |
5656 | key.type = BTRFS_INODE_REF_KEY; | |
5657 | key.offset = 0; | |
5658 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
5659 | if (ret < 0) | |
5660 | goto out; | |
5661 | ||
5662 | while (true) { | |
5663 | struct extent_buffer *leaf = path->nodes[0]; | |
5664 | int slot = path->slots[0]; | |
5665 | u32 cur_offset = 0; | |
5666 | u32 item_size; | |
5667 | unsigned long ptr; | |
5668 | ||
5669 | if (slot >= btrfs_header_nritems(leaf)) { | |
5670 | ret = btrfs_next_leaf(root, path); | |
5671 | if (ret < 0) | |
5672 | goto out; | |
5673 | else if (ret > 0) | |
5674 | break; | |
5675 | continue; | |
5676 | } | |
5677 | ||
5678 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
5679 | /* BTRFS_INODE_EXTREF_KEY is BTRFS_INODE_REF_KEY + 1 */ | |
5680 | if (key.objectid != ino || key.type > BTRFS_INODE_EXTREF_KEY) | |
5681 | break; | |
5682 | ||
5683 | item_size = btrfs_item_size_nr(leaf, slot); | |
5684 | ptr = btrfs_item_ptr_offset(leaf, slot); | |
5685 | while (cur_offset < item_size) { | |
5686 | struct btrfs_key inode_key; | |
5687 | struct inode *dir_inode; | |
5688 | ||
5689 | inode_key.type = BTRFS_INODE_ITEM_KEY; | |
5690 | inode_key.offset = 0; | |
5691 | ||
5692 | if (key.type == BTRFS_INODE_EXTREF_KEY) { | |
5693 | struct btrfs_inode_extref *extref; | |
5694 | ||
5695 | extref = (struct btrfs_inode_extref *) | |
5696 | (ptr + cur_offset); | |
5697 | inode_key.objectid = btrfs_inode_extref_parent( | |
5698 | leaf, extref); | |
5699 | cur_offset += sizeof(*extref); | |
5700 | cur_offset += btrfs_inode_extref_name_len(leaf, | |
5701 | extref); | |
5702 | } else { | |
5703 | inode_key.objectid = key.offset; | |
5704 | cur_offset = item_size; | |
5705 | } | |
5706 | ||
0202e83f DS |
5707 | dir_inode = btrfs_iget(fs_info->sb, inode_key.objectid, |
5708 | root); | |
0f375eed FM |
5709 | /* |
5710 | * If the parent inode was deleted, return an error to | |
5711 | * fallback to a transaction commit. This is to prevent | |
5712 | * getting an inode that was moved from one parent A to | |
5713 | * a parent B, got its former parent A deleted and then | |
5714 | * it got fsync'ed, from existing at both parents after | |
5715 | * a log replay (and the old parent still existing). | |
5716 | * Example: | |
5717 | * | |
5718 | * mkdir /mnt/A | |
5719 | * mkdir /mnt/B | |
5720 | * touch /mnt/B/bar | |
5721 | * sync | |
5722 | * mv /mnt/B/bar /mnt/A/bar | |
5723 | * mv -T /mnt/A /mnt/B | |
5724 | * fsync /mnt/B/bar | |
5725 | * <power fail> | |
5726 | * | |
5727 | * If we ignore the old parent B which got deleted, | |
5728 | * after a log replay we would have file bar linked | |
5729 | * at both parents and the old parent B would still | |
5730 | * exist. | |
5731 | */ | |
5732 | if (IS_ERR(dir_inode)) { | |
5733 | ret = PTR_ERR(dir_inode); | |
5734 | goto out; | |
5735 | } | |
18aa0922 | 5736 | |
657ed1aa FM |
5737 | if (ctx) |
5738 | ctx->log_new_dentries = false; | |
a59108a7 | 5739 | ret = btrfs_log_inode(trans, root, BTRFS_I(dir_inode), |
18aa0922 | 5740 | LOG_INODE_ALL, 0, LLONG_MAX, ctx); |
2be63d5c | 5741 | if (!ret && |
ab1717b2 | 5742 | btrfs_must_commit_transaction(trans, BTRFS_I(dir_inode))) |
2be63d5c | 5743 | ret = 1; |
657ed1aa FM |
5744 | if (!ret && ctx && ctx->log_new_dentries) |
5745 | ret = log_new_dir_dentries(trans, root, | |
f85b7379 | 5746 | BTRFS_I(dir_inode), ctx); |
410f954c | 5747 | btrfs_add_delayed_iput(dir_inode); |
18aa0922 FM |
5748 | if (ret) |
5749 | goto out; | |
5750 | } | |
5751 | path->slots[0]++; | |
5752 | } | |
5753 | ret = 0; | |
5754 | out: | |
5755 | btrfs_free_path(path); | |
5756 | return ret; | |
5757 | } | |
5758 | ||
b8aa330d FM |
5759 | static int log_new_ancestors(struct btrfs_trans_handle *trans, |
5760 | struct btrfs_root *root, | |
5761 | struct btrfs_path *path, | |
5762 | struct btrfs_log_ctx *ctx) | |
5763 | { | |
5764 | struct btrfs_key found_key; | |
5765 | ||
5766 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, path->slots[0]); | |
5767 | ||
5768 | while (true) { | |
5769 | struct btrfs_fs_info *fs_info = root->fs_info; | |
5770 | const u64 last_committed = fs_info->last_trans_committed; | |
5771 | struct extent_buffer *leaf = path->nodes[0]; | |
5772 | int slot = path->slots[0]; | |
5773 | struct btrfs_key search_key; | |
5774 | struct inode *inode; | |
0202e83f | 5775 | u64 ino; |
b8aa330d FM |
5776 | int ret = 0; |
5777 | ||
5778 | btrfs_release_path(path); | |
5779 | ||
0202e83f DS |
5780 | ino = found_key.offset; |
5781 | ||
b8aa330d FM |
5782 | search_key.objectid = found_key.offset; |
5783 | search_key.type = BTRFS_INODE_ITEM_KEY; | |
5784 | search_key.offset = 0; | |
0202e83f | 5785 | inode = btrfs_iget(fs_info->sb, ino, root); |
b8aa330d FM |
5786 | if (IS_ERR(inode)) |
5787 | return PTR_ERR(inode); | |
5788 | ||
5789 | if (BTRFS_I(inode)->generation > last_committed) | |
5790 | ret = btrfs_log_inode(trans, root, BTRFS_I(inode), | |
5791 | LOG_INODE_EXISTS, | |
5792 | 0, LLONG_MAX, ctx); | |
410f954c | 5793 | btrfs_add_delayed_iput(inode); |
b8aa330d FM |
5794 | if (ret) |
5795 | return ret; | |
5796 | ||
5797 | if (search_key.objectid == BTRFS_FIRST_FREE_OBJECTID) | |
5798 | break; | |
5799 | ||
5800 | search_key.type = BTRFS_INODE_REF_KEY; | |
5801 | ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0); | |
5802 | if (ret < 0) | |
5803 | return ret; | |
5804 | ||
5805 | leaf = path->nodes[0]; | |
5806 | slot = path->slots[0]; | |
5807 | if (slot >= btrfs_header_nritems(leaf)) { | |
5808 | ret = btrfs_next_leaf(root, path); | |
5809 | if (ret < 0) | |
5810 | return ret; | |
5811 | else if (ret > 0) | |
5812 | return -ENOENT; | |
5813 | leaf = path->nodes[0]; | |
5814 | slot = path->slots[0]; | |
5815 | } | |
5816 | ||
5817 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
5818 | if (found_key.objectid != search_key.objectid || | |
5819 | found_key.type != BTRFS_INODE_REF_KEY) | |
5820 | return -ENOENT; | |
5821 | } | |
5822 | return 0; | |
5823 | } | |
5824 | ||
5825 | static int log_new_ancestors_fast(struct btrfs_trans_handle *trans, | |
5826 | struct btrfs_inode *inode, | |
5827 | struct dentry *parent, | |
5828 | struct btrfs_log_ctx *ctx) | |
5829 | { | |
5830 | struct btrfs_root *root = inode->root; | |
5831 | struct btrfs_fs_info *fs_info = root->fs_info; | |
5832 | struct dentry *old_parent = NULL; | |
5833 | struct super_block *sb = inode->vfs_inode.i_sb; | |
5834 | int ret = 0; | |
5835 | ||
5836 | while (true) { | |
5837 | if (!parent || d_really_is_negative(parent) || | |
5838 | sb != parent->d_sb) | |
5839 | break; | |
5840 | ||
5841 | inode = BTRFS_I(d_inode(parent)); | |
5842 | if (root != inode->root) | |
5843 | break; | |
5844 | ||
5845 | if (inode->generation > fs_info->last_trans_committed) { | |
5846 | ret = btrfs_log_inode(trans, root, inode, | |
5847 | LOG_INODE_EXISTS, 0, LLONG_MAX, ctx); | |
5848 | if (ret) | |
5849 | break; | |
5850 | } | |
5851 | if (IS_ROOT(parent)) | |
5852 | break; | |
5853 | ||
5854 | parent = dget_parent(parent); | |
5855 | dput(old_parent); | |
5856 | old_parent = parent; | |
5857 | } | |
5858 | dput(old_parent); | |
5859 | ||
5860 | return ret; | |
5861 | } | |
5862 | ||
5863 | static int log_all_new_ancestors(struct btrfs_trans_handle *trans, | |
5864 | struct btrfs_inode *inode, | |
5865 | struct dentry *parent, | |
5866 | struct btrfs_log_ctx *ctx) | |
5867 | { | |
5868 | struct btrfs_root *root = inode->root; | |
5869 | const u64 ino = btrfs_ino(inode); | |
5870 | struct btrfs_path *path; | |
5871 | struct btrfs_key search_key; | |
5872 | int ret; | |
5873 | ||
5874 | /* | |
5875 | * For a single hard link case, go through a fast path that does not | |
5876 | * need to iterate the fs/subvolume tree. | |
5877 | */ | |
5878 | if (inode->vfs_inode.i_nlink < 2) | |
5879 | return log_new_ancestors_fast(trans, inode, parent, ctx); | |
5880 | ||
5881 | path = btrfs_alloc_path(); | |
5882 | if (!path) | |
5883 | return -ENOMEM; | |
5884 | ||
5885 | search_key.objectid = ino; | |
5886 | search_key.type = BTRFS_INODE_REF_KEY; | |
5887 | search_key.offset = 0; | |
5888 | again: | |
5889 | ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0); | |
5890 | if (ret < 0) | |
5891 | goto out; | |
5892 | if (ret == 0) | |
5893 | path->slots[0]++; | |
5894 | ||
5895 | while (true) { | |
5896 | struct extent_buffer *leaf = path->nodes[0]; | |
5897 | int slot = path->slots[0]; | |
5898 | struct btrfs_key found_key; | |
5899 | ||
5900 | if (slot >= btrfs_header_nritems(leaf)) { | |
5901 | ret = btrfs_next_leaf(root, path); | |
5902 | if (ret < 0) | |
5903 | goto out; | |
5904 | else if (ret > 0) | |
5905 | break; | |
5906 | continue; | |
5907 | } | |
5908 | ||
5909 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
5910 | if (found_key.objectid != ino || | |
5911 | found_key.type > BTRFS_INODE_EXTREF_KEY) | |
5912 | break; | |
5913 | ||
5914 | /* | |
5915 | * Don't deal with extended references because they are rare | |
5916 | * cases and too complex to deal with (we would need to keep | |
5917 | * track of which subitem we are processing for each item in | |
5918 | * this loop, etc). So just return some error to fallback to | |
5919 | * a transaction commit. | |
5920 | */ | |
5921 | if (found_key.type == BTRFS_INODE_EXTREF_KEY) { | |
5922 | ret = -EMLINK; | |
5923 | goto out; | |
5924 | } | |
5925 | ||
5926 | /* | |
5927 | * Logging ancestors needs to do more searches on the fs/subvol | |
5928 | * tree, so it releases the path as needed to avoid deadlocks. | |
5929 | * Keep track of the last inode ref key and resume from that key | |
5930 | * after logging all new ancestors for the current hard link. | |
5931 | */ | |
5932 | memcpy(&search_key, &found_key, sizeof(search_key)); | |
5933 | ||
5934 | ret = log_new_ancestors(trans, root, path, ctx); | |
5935 | if (ret) | |
5936 | goto out; | |
5937 | btrfs_release_path(path); | |
5938 | goto again; | |
5939 | } | |
5940 | ret = 0; | |
5941 | out: | |
5942 | btrfs_free_path(path); | |
5943 | return ret; | |
5944 | } | |
5945 | ||
e02119d5 CM |
5946 | /* |
5947 | * helper function around btrfs_log_inode to make sure newly created | |
5948 | * parent directories also end up in the log. A minimal inode and backref | |
5949 | * only logging is done of any parent directories that are older than | |
5950 | * the last committed transaction | |
5951 | */ | |
48a3b636 | 5952 | static int btrfs_log_inode_parent(struct btrfs_trans_handle *trans, |
19df27a9 | 5953 | struct btrfs_inode *inode, |
49dae1bc FM |
5954 | struct dentry *parent, |
5955 | const loff_t start, | |
5956 | const loff_t end, | |
41a1eada | 5957 | int inode_only, |
8b050d35 | 5958 | struct btrfs_log_ctx *ctx) |
e02119d5 | 5959 | { |
f882274b | 5960 | struct btrfs_root *root = inode->root; |
0b246afa | 5961 | struct btrfs_fs_info *fs_info = root->fs_info; |
e02119d5 | 5962 | struct super_block *sb; |
12fcfd22 | 5963 | int ret = 0; |
0b246afa | 5964 | u64 last_committed = fs_info->last_trans_committed; |
2f2ff0ee | 5965 | bool log_dentries = false; |
12fcfd22 | 5966 | |
19df27a9 | 5967 | sb = inode->vfs_inode.i_sb; |
12fcfd22 | 5968 | |
0b246afa | 5969 | if (btrfs_test_opt(fs_info, NOTREELOG)) { |
3a5e1404 SW |
5970 | ret = 1; |
5971 | goto end_no_trans; | |
5972 | } | |
5973 | ||
995946dd MX |
5974 | /* |
5975 | * The prev transaction commit doesn't complete, we need do | |
5976 | * full commit by ourselves. | |
5977 | */ | |
0b246afa JM |
5978 | if (fs_info->last_trans_log_full_commit > |
5979 | fs_info->last_trans_committed) { | |
12fcfd22 CM |
5980 | ret = 1; |
5981 | goto end_no_trans; | |
5982 | } | |
5983 | ||
f882274b | 5984 | if (btrfs_root_refs(&root->root_item) == 0) { |
76dda93c YZ |
5985 | ret = 1; |
5986 | goto end_no_trans; | |
5987 | } | |
5988 | ||
19df27a9 NB |
5989 | ret = check_parent_dirs_for_sync(trans, inode, parent, sb, |
5990 | last_committed); | |
12fcfd22 CM |
5991 | if (ret) |
5992 | goto end_no_trans; | |
e02119d5 | 5993 | |
f2d72f42 FM |
5994 | /* |
5995 | * Skip already logged inodes or inodes corresponding to tmpfiles | |
5996 | * (since logging them is pointless, a link count of 0 means they | |
5997 | * will never be accessible). | |
5998 | */ | |
5999 | if (btrfs_inode_in_log(inode, trans->transid) || | |
6000 | inode->vfs_inode.i_nlink == 0) { | |
257c62e1 CM |
6001 | ret = BTRFS_NO_LOG_SYNC; |
6002 | goto end_no_trans; | |
6003 | } | |
6004 | ||
8b050d35 | 6005 | ret = start_log_trans(trans, root, ctx); |
4a500fd1 | 6006 | if (ret) |
e87ac136 | 6007 | goto end_no_trans; |
e02119d5 | 6008 | |
19df27a9 | 6009 | ret = btrfs_log_inode(trans, root, inode, inode_only, start, end, ctx); |
4a500fd1 YZ |
6010 | if (ret) |
6011 | goto end_trans; | |
12fcfd22 | 6012 | |
af4176b4 CM |
6013 | /* |
6014 | * for regular files, if its inode is already on disk, we don't | |
6015 | * have to worry about the parents at all. This is because | |
6016 | * we can use the last_unlink_trans field to record renames | |
6017 | * and other fun in this file. | |
6018 | */ | |
19df27a9 NB |
6019 | if (S_ISREG(inode->vfs_inode.i_mode) && |
6020 | inode->generation <= last_committed && | |
6021 | inode->last_unlink_trans <= last_committed) { | |
4a500fd1 YZ |
6022 | ret = 0; |
6023 | goto end_trans; | |
6024 | } | |
af4176b4 | 6025 | |
19df27a9 | 6026 | if (S_ISDIR(inode->vfs_inode.i_mode) && ctx && ctx->log_new_dentries) |
2f2ff0ee FM |
6027 | log_dentries = true; |
6028 | ||
18aa0922 | 6029 | /* |
01327610 | 6030 | * On unlink we must make sure all our current and old parent directory |
18aa0922 FM |
6031 | * inodes are fully logged. This is to prevent leaving dangling |
6032 | * directory index entries in directories that were our parents but are | |
6033 | * not anymore. Not doing this results in old parent directory being | |
6034 | * impossible to delete after log replay (rmdir will always fail with | |
6035 | * error -ENOTEMPTY). | |
6036 | * | |
6037 | * Example 1: | |
6038 | * | |
6039 | * mkdir testdir | |
6040 | * touch testdir/foo | |
6041 | * ln testdir/foo testdir/bar | |
6042 | * sync | |
6043 | * unlink testdir/bar | |
6044 | * xfs_io -c fsync testdir/foo | |
6045 | * <power failure> | |
6046 | * mount fs, triggers log replay | |
6047 | * | |
6048 | * If we don't log the parent directory (testdir), after log replay the | |
6049 | * directory still has an entry pointing to the file inode using the bar | |
6050 | * name, but a matching BTRFS_INODE_[REF|EXTREF]_KEY does not exist and | |
6051 | * the file inode has a link count of 1. | |
6052 | * | |
6053 | * Example 2: | |
6054 | * | |
6055 | * mkdir testdir | |
6056 | * touch foo | |
6057 | * ln foo testdir/foo2 | |
6058 | * ln foo testdir/foo3 | |
6059 | * sync | |
6060 | * unlink testdir/foo3 | |
6061 | * xfs_io -c fsync foo | |
6062 | * <power failure> | |
6063 | * mount fs, triggers log replay | |
6064 | * | |
6065 | * Similar as the first example, after log replay the parent directory | |
6066 | * testdir still has an entry pointing to the inode file with name foo3 | |
6067 | * but the file inode does not have a matching BTRFS_INODE_REF_KEY item | |
6068 | * and has a link count of 2. | |
6069 | */ | |
19df27a9 | 6070 | if (inode->last_unlink_trans > last_committed) { |
b8aa330d | 6071 | ret = btrfs_log_all_parents(trans, inode, ctx); |
18aa0922 FM |
6072 | if (ret) |
6073 | goto end_trans; | |
6074 | } | |
6075 | ||
b8aa330d FM |
6076 | ret = log_all_new_ancestors(trans, inode, parent, ctx); |
6077 | if (ret) | |
41bd6067 | 6078 | goto end_trans; |
76dda93c | 6079 | |
2f2ff0ee | 6080 | if (log_dentries) |
b8aa330d | 6081 | ret = log_new_dir_dentries(trans, root, inode, ctx); |
2f2ff0ee FM |
6082 | else |
6083 | ret = 0; | |
4a500fd1 YZ |
6084 | end_trans: |
6085 | if (ret < 0) { | |
90787766 | 6086 | btrfs_set_log_full_commit(trans); |
4a500fd1 YZ |
6087 | ret = 1; |
6088 | } | |
8b050d35 MX |
6089 | |
6090 | if (ret) | |
6091 | btrfs_remove_log_ctx(root, ctx); | |
12fcfd22 CM |
6092 | btrfs_end_log_trans(root); |
6093 | end_no_trans: | |
6094 | return ret; | |
e02119d5 CM |
6095 | } |
6096 | ||
6097 | /* | |
6098 | * it is not safe to log dentry if the chunk root has added new | |
6099 | * chunks. This returns 0 if the dentry was logged, and 1 otherwise. | |
6100 | * If this returns 1, you must commit the transaction to safely get your | |
6101 | * data on disk. | |
6102 | */ | |
6103 | int btrfs_log_dentry_safe(struct btrfs_trans_handle *trans, | |
e5b84f7a | 6104 | struct dentry *dentry, |
49dae1bc FM |
6105 | const loff_t start, |
6106 | const loff_t end, | |
8b050d35 | 6107 | struct btrfs_log_ctx *ctx) |
e02119d5 | 6108 | { |
6a912213 JB |
6109 | struct dentry *parent = dget_parent(dentry); |
6110 | int ret; | |
6111 | ||
f882274b NB |
6112 | ret = btrfs_log_inode_parent(trans, BTRFS_I(d_inode(dentry)), parent, |
6113 | start, end, LOG_INODE_ALL, ctx); | |
6a912213 JB |
6114 | dput(parent); |
6115 | ||
6116 | return ret; | |
e02119d5 CM |
6117 | } |
6118 | ||
6119 | /* | |
6120 | * should be called during mount to recover any replay any log trees | |
6121 | * from the FS | |
6122 | */ | |
6123 | int btrfs_recover_log_trees(struct btrfs_root *log_root_tree) | |
6124 | { | |
6125 | int ret; | |
6126 | struct btrfs_path *path; | |
6127 | struct btrfs_trans_handle *trans; | |
6128 | struct btrfs_key key; | |
6129 | struct btrfs_key found_key; | |
e02119d5 CM |
6130 | struct btrfs_root *log; |
6131 | struct btrfs_fs_info *fs_info = log_root_tree->fs_info; | |
6132 | struct walk_control wc = { | |
6133 | .process_func = process_one_buffer, | |
430a6626 | 6134 | .stage = LOG_WALK_PIN_ONLY, |
e02119d5 CM |
6135 | }; |
6136 | ||
e02119d5 | 6137 | path = btrfs_alloc_path(); |
db5b493a TI |
6138 | if (!path) |
6139 | return -ENOMEM; | |
6140 | ||
afcdd129 | 6141 | set_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags); |
e02119d5 | 6142 | |
4a500fd1 | 6143 | trans = btrfs_start_transaction(fs_info->tree_root, 0); |
79787eaa JM |
6144 | if (IS_ERR(trans)) { |
6145 | ret = PTR_ERR(trans); | |
6146 | goto error; | |
6147 | } | |
e02119d5 CM |
6148 | |
6149 | wc.trans = trans; | |
6150 | wc.pin = 1; | |
6151 | ||
db5b493a | 6152 | ret = walk_log_tree(trans, log_root_tree, &wc); |
79787eaa | 6153 | if (ret) { |
5d163e0e JM |
6154 | btrfs_handle_fs_error(fs_info, ret, |
6155 | "Failed to pin buffers while recovering log root tree."); | |
79787eaa JM |
6156 | goto error; |
6157 | } | |
e02119d5 CM |
6158 | |
6159 | again: | |
6160 | key.objectid = BTRFS_TREE_LOG_OBJECTID; | |
6161 | key.offset = (u64)-1; | |
962a298f | 6162 | key.type = BTRFS_ROOT_ITEM_KEY; |
e02119d5 | 6163 | |
d397712b | 6164 | while (1) { |
e02119d5 | 6165 | ret = btrfs_search_slot(NULL, log_root_tree, &key, path, 0, 0); |
79787eaa JM |
6166 | |
6167 | if (ret < 0) { | |
34d97007 | 6168 | btrfs_handle_fs_error(fs_info, ret, |
79787eaa JM |
6169 | "Couldn't find tree log root."); |
6170 | goto error; | |
6171 | } | |
e02119d5 CM |
6172 | if (ret > 0) { |
6173 | if (path->slots[0] == 0) | |
6174 | break; | |
6175 | path->slots[0]--; | |
6176 | } | |
6177 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, | |
6178 | path->slots[0]); | |
b3b4aa74 | 6179 | btrfs_release_path(path); |
e02119d5 CM |
6180 | if (found_key.objectid != BTRFS_TREE_LOG_OBJECTID) |
6181 | break; | |
6182 | ||
62a2c73e | 6183 | log = btrfs_read_tree_root(log_root_tree, &found_key); |
79787eaa JM |
6184 | if (IS_ERR(log)) { |
6185 | ret = PTR_ERR(log); | |
34d97007 | 6186 | btrfs_handle_fs_error(fs_info, ret, |
79787eaa JM |
6187 | "Couldn't read tree log root."); |
6188 | goto error; | |
6189 | } | |
e02119d5 | 6190 | |
56e9357a DS |
6191 | wc.replay_dest = btrfs_get_fs_root(fs_info, found_key.offset, |
6192 | true); | |
79787eaa JM |
6193 | if (IS_ERR(wc.replay_dest)) { |
6194 | ret = PTR_ERR(wc.replay_dest); | |
9bc574de JB |
6195 | |
6196 | /* | |
6197 | * We didn't find the subvol, likely because it was | |
6198 | * deleted. This is ok, simply skip this log and go to | |
6199 | * the next one. | |
6200 | * | |
6201 | * We need to exclude the root because we can't have | |
6202 | * other log replays overwriting this log as we'll read | |
6203 | * it back in a few more times. This will keep our | |
6204 | * block from being modified, and we'll just bail for | |
6205 | * each subsequent pass. | |
6206 | */ | |
6207 | if (ret == -ENOENT) | |
9fce5704 | 6208 | ret = btrfs_pin_extent_for_log_replay(trans, |
9bc574de JB |
6209 | log->node->start, |
6210 | log->node->len); | |
00246528 | 6211 | btrfs_put_root(log); |
9bc574de JB |
6212 | |
6213 | if (!ret) | |
6214 | goto next; | |
5d163e0e JM |
6215 | btrfs_handle_fs_error(fs_info, ret, |
6216 | "Couldn't read target root for tree log recovery."); | |
79787eaa JM |
6217 | goto error; |
6218 | } | |
e02119d5 | 6219 | |
07d400a6 | 6220 | wc.replay_dest->log_root = log; |
5d4f98a2 | 6221 | btrfs_record_root_in_trans(trans, wc.replay_dest); |
e02119d5 | 6222 | ret = walk_log_tree(trans, log, &wc); |
e02119d5 | 6223 | |
b50c6e25 | 6224 | if (!ret && wc.stage == LOG_WALK_REPLAY_ALL) { |
e02119d5 CM |
6225 | ret = fixup_inode_link_counts(trans, wc.replay_dest, |
6226 | path); | |
e02119d5 CM |
6227 | } |
6228 | ||
900c9981 LB |
6229 | if (!ret && wc.stage == LOG_WALK_REPLAY_ALL) { |
6230 | struct btrfs_root *root = wc.replay_dest; | |
6231 | ||
6232 | btrfs_release_path(path); | |
6233 | ||
6234 | /* | |
6235 | * We have just replayed everything, and the highest | |
6236 | * objectid of fs roots probably has changed in case | |
6237 | * some inode_item's got replayed. | |
6238 | * | |
6239 | * root->objectid_mutex is not acquired as log replay | |
6240 | * could only happen during mount. | |
6241 | */ | |
6242 | ret = btrfs_find_highest_objectid(root, | |
6243 | &root->highest_objectid); | |
6244 | } | |
6245 | ||
07d400a6 | 6246 | wc.replay_dest->log_root = NULL; |
00246528 | 6247 | btrfs_put_root(wc.replay_dest); |
00246528 | 6248 | btrfs_put_root(log); |
e02119d5 | 6249 | |
b50c6e25 JB |
6250 | if (ret) |
6251 | goto error; | |
9bc574de | 6252 | next: |
e02119d5 CM |
6253 | if (found_key.offset == 0) |
6254 | break; | |
9bc574de | 6255 | key.offset = found_key.offset - 1; |
e02119d5 | 6256 | } |
b3b4aa74 | 6257 | btrfs_release_path(path); |
e02119d5 CM |
6258 | |
6259 | /* step one is to pin it all, step two is to replay just inodes */ | |
6260 | if (wc.pin) { | |
6261 | wc.pin = 0; | |
6262 | wc.process_func = replay_one_buffer; | |
6263 | wc.stage = LOG_WALK_REPLAY_INODES; | |
6264 | goto again; | |
6265 | } | |
6266 | /* step three is to replay everything */ | |
6267 | if (wc.stage < LOG_WALK_REPLAY_ALL) { | |
6268 | wc.stage++; | |
6269 | goto again; | |
6270 | } | |
6271 | ||
6272 | btrfs_free_path(path); | |
6273 | ||
abefa55a | 6274 | /* step 4: commit the transaction, which also unpins the blocks */ |
3a45bb20 | 6275 | ret = btrfs_commit_transaction(trans); |
abefa55a JB |
6276 | if (ret) |
6277 | return ret; | |
6278 | ||
e02119d5 | 6279 | log_root_tree->log_root = NULL; |
afcdd129 | 6280 | clear_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags); |
00246528 | 6281 | btrfs_put_root(log_root_tree); |
79787eaa | 6282 | |
abefa55a | 6283 | return 0; |
79787eaa | 6284 | error: |
b50c6e25 | 6285 | if (wc.trans) |
3a45bb20 | 6286 | btrfs_end_transaction(wc.trans); |
79787eaa JM |
6287 | btrfs_free_path(path); |
6288 | return ret; | |
e02119d5 | 6289 | } |
12fcfd22 CM |
6290 | |
6291 | /* | |
6292 | * there are some corner cases where we want to force a full | |
6293 | * commit instead of allowing a directory to be logged. | |
6294 | * | |
6295 | * They revolve around files there were unlinked from the directory, and | |
6296 | * this function updates the parent directory so that a full commit is | |
6297 | * properly done if it is fsync'd later after the unlinks are done. | |
2be63d5c FM |
6298 | * |
6299 | * Must be called before the unlink operations (updates to the subvolume tree, | |
6300 | * inodes, etc) are done. | |
12fcfd22 CM |
6301 | */ |
6302 | void btrfs_record_unlink_dir(struct btrfs_trans_handle *trans, | |
4176bdbf | 6303 | struct btrfs_inode *dir, struct btrfs_inode *inode, |
12fcfd22 CM |
6304 | int for_rename) |
6305 | { | |
af4176b4 CM |
6306 | /* |
6307 | * when we're logging a file, if it hasn't been renamed | |
6308 | * or unlinked, and its inode is fully committed on disk, | |
6309 | * we don't have to worry about walking up the directory chain | |
6310 | * to log its parents. | |
6311 | * | |
6312 | * So, we use the last_unlink_trans field to put this transid | |
6313 | * into the file. When the file is logged we check it and | |
6314 | * don't log the parents if the file is fully on disk. | |
6315 | */ | |
4176bdbf NB |
6316 | mutex_lock(&inode->log_mutex); |
6317 | inode->last_unlink_trans = trans->transid; | |
6318 | mutex_unlock(&inode->log_mutex); | |
af4176b4 | 6319 | |
12fcfd22 CM |
6320 | /* |
6321 | * if this directory was already logged any new | |
6322 | * names for this file/dir will get recorded | |
6323 | */ | |
4176bdbf | 6324 | if (dir->logged_trans == trans->transid) |
12fcfd22 CM |
6325 | return; |
6326 | ||
6327 | /* | |
6328 | * if the inode we're about to unlink was logged, | |
6329 | * the log will be properly updated for any new names | |
6330 | */ | |
4176bdbf | 6331 | if (inode->logged_trans == trans->transid) |
12fcfd22 CM |
6332 | return; |
6333 | ||
6334 | /* | |
6335 | * when renaming files across directories, if the directory | |
6336 | * there we're unlinking from gets fsync'd later on, there's | |
6337 | * no way to find the destination directory later and fsync it | |
6338 | * properly. So, we have to be conservative and force commits | |
6339 | * so the new name gets discovered. | |
6340 | */ | |
6341 | if (for_rename) | |
6342 | goto record; | |
6343 | ||
6344 | /* we can safely do the unlink without any special recording */ | |
6345 | return; | |
6346 | ||
6347 | record: | |
4176bdbf NB |
6348 | mutex_lock(&dir->log_mutex); |
6349 | dir->last_unlink_trans = trans->transid; | |
6350 | mutex_unlock(&dir->log_mutex); | |
1ec9a1ae FM |
6351 | } |
6352 | ||
6353 | /* | |
6354 | * Make sure that if someone attempts to fsync the parent directory of a deleted | |
6355 | * snapshot, it ends up triggering a transaction commit. This is to guarantee | |
6356 | * that after replaying the log tree of the parent directory's root we will not | |
6357 | * see the snapshot anymore and at log replay time we will not see any log tree | |
6358 | * corresponding to the deleted snapshot's root, which could lead to replaying | |
6359 | * it after replaying the log tree of the parent directory (which would replay | |
6360 | * the snapshot delete operation). | |
2be63d5c FM |
6361 | * |
6362 | * Must be called before the actual snapshot destroy operation (updates to the | |
6363 | * parent root and tree of tree roots trees, etc) are done. | |
1ec9a1ae FM |
6364 | */ |
6365 | void btrfs_record_snapshot_destroy(struct btrfs_trans_handle *trans, | |
43663557 | 6366 | struct btrfs_inode *dir) |
1ec9a1ae | 6367 | { |
43663557 NB |
6368 | mutex_lock(&dir->log_mutex); |
6369 | dir->last_unlink_trans = trans->transid; | |
6370 | mutex_unlock(&dir->log_mutex); | |
12fcfd22 CM |
6371 | } |
6372 | ||
6373 | /* | |
6374 | * Call this after adding a new name for a file and it will properly | |
6375 | * update the log to reflect the new name. | |
6376 | * | |
d4682ba0 FM |
6377 | * @ctx can not be NULL when @sync_log is false, and should be NULL when it's |
6378 | * true (because it's not used). | |
6379 | * | |
6380 | * Return value depends on whether @sync_log is true or false. | |
6381 | * When true: returns BTRFS_NEED_TRANS_COMMIT if the transaction needs to be | |
6382 | * committed by the caller, and BTRFS_DONT_NEED_TRANS_COMMIT | |
6383 | * otherwise. | |
6384 | * When false: returns BTRFS_DONT_NEED_LOG_SYNC if the caller does not need to | |
6385 | * to sync the log, BTRFS_NEED_LOG_SYNC if it needs to sync the log, | |
6386 | * or BTRFS_NEED_TRANS_COMMIT if the transaction needs to be | |
6387 | * committed (without attempting to sync the log). | |
12fcfd22 CM |
6388 | */ |
6389 | int btrfs_log_new_name(struct btrfs_trans_handle *trans, | |
9ca5fbfb | 6390 | struct btrfs_inode *inode, struct btrfs_inode *old_dir, |
d4682ba0 FM |
6391 | struct dentry *parent, |
6392 | bool sync_log, struct btrfs_log_ctx *ctx) | |
12fcfd22 | 6393 | { |
3ffbd68c | 6394 | struct btrfs_fs_info *fs_info = trans->fs_info; |
d4682ba0 | 6395 | int ret; |
12fcfd22 | 6396 | |
af4176b4 CM |
6397 | /* |
6398 | * this will force the logging code to walk the dentry chain | |
6399 | * up for the file | |
6400 | */ | |
9a6509c4 | 6401 | if (!S_ISDIR(inode->vfs_inode.i_mode)) |
9ca5fbfb | 6402 | inode->last_unlink_trans = trans->transid; |
af4176b4 | 6403 | |
12fcfd22 CM |
6404 | /* |
6405 | * if this inode hasn't been logged and directory we're renaming it | |
6406 | * from hasn't been logged, we don't need to log it | |
6407 | */ | |
9ca5fbfb NB |
6408 | if (inode->logged_trans <= fs_info->last_trans_committed && |
6409 | (!old_dir || old_dir->logged_trans <= fs_info->last_trans_committed)) | |
d4682ba0 FM |
6410 | return sync_log ? BTRFS_DONT_NEED_TRANS_COMMIT : |
6411 | BTRFS_DONT_NEED_LOG_SYNC; | |
6412 | ||
6413 | if (sync_log) { | |
6414 | struct btrfs_log_ctx ctx2; | |
6415 | ||
6416 | btrfs_init_log_ctx(&ctx2, &inode->vfs_inode); | |
6417 | ret = btrfs_log_inode_parent(trans, inode, parent, 0, LLONG_MAX, | |
6418 | LOG_INODE_EXISTS, &ctx2); | |
6419 | if (ret == BTRFS_NO_LOG_SYNC) | |
6420 | return BTRFS_DONT_NEED_TRANS_COMMIT; | |
6421 | else if (ret) | |
6422 | return BTRFS_NEED_TRANS_COMMIT; | |
6423 | ||
6424 | ret = btrfs_sync_log(trans, inode->root, &ctx2); | |
6425 | if (ret) | |
6426 | return BTRFS_NEED_TRANS_COMMIT; | |
6427 | return BTRFS_DONT_NEED_TRANS_COMMIT; | |
6428 | } | |
6429 | ||
6430 | ASSERT(ctx); | |
6431 | ret = btrfs_log_inode_parent(trans, inode, parent, 0, LLONG_MAX, | |
6432 | LOG_INODE_EXISTS, ctx); | |
6433 | if (ret == BTRFS_NO_LOG_SYNC) | |
6434 | return BTRFS_DONT_NEED_LOG_SYNC; | |
6435 | else if (ret) | |
6436 | return BTRFS_NEED_TRANS_COMMIT; | |
12fcfd22 | 6437 | |
d4682ba0 | 6438 | return BTRFS_NEED_LOG_SYNC; |
12fcfd22 CM |
6439 | } |
6440 |