]>
Commit | Line | Data |
---|---|---|
e02119d5 CM |
1 | /* |
2 | * Copyright (C) 2008 Oracle. All rights reserved. | |
3 | * | |
4 | * This program is free software; you can redistribute it and/or | |
5 | * modify it under the terms of the GNU General Public | |
6 | * License v2 as published by the Free Software Foundation. | |
7 | * | |
8 | * This program is distributed in the hope that it will be useful, | |
9 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
10 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
11 | * General Public License for more details. | |
12 | * | |
13 | * You should have received a copy of the GNU General Public | |
14 | * License along with this program; if not, write to the | |
15 | * Free Software Foundation, Inc., 59 Temple Place - Suite 330, | |
16 | * Boston, MA 021110-1307, USA. | |
17 | */ | |
18 | ||
19 | #include <linux/sched.h> | |
5a0e3ad6 | 20 | #include <linux/slab.h> |
c6adc9cc | 21 | #include <linux/blkdev.h> |
5dc562c5 | 22 | #include <linux/list_sort.h> |
995946dd | 23 | #include "tree-log.h" |
e02119d5 CM |
24 | #include "disk-io.h" |
25 | #include "locking.h" | |
26 | #include "print-tree.h" | |
f186373f | 27 | #include "backref.h" |
f186373f | 28 | #include "hash.h" |
ebb8765b | 29 | #include "compression.h" |
df2c95f3 | 30 | #include "qgroup.h" |
e02119d5 CM |
31 | |
32 | /* magic values for the inode_only field in btrfs_log_inode: | |
33 | * | |
34 | * LOG_INODE_ALL means to log everything | |
35 | * LOG_INODE_EXISTS means to log just enough to recreate the inode | |
36 | * during log replay | |
37 | */ | |
38 | #define LOG_INODE_ALL 0 | |
39 | #define LOG_INODE_EXISTS 1 | |
781feef7 | 40 | #define LOG_OTHER_INODE 2 |
e02119d5 | 41 | |
12fcfd22 CM |
42 | /* |
43 | * directory trouble cases | |
44 | * | |
45 | * 1) on rename or unlink, if the inode being unlinked isn't in the fsync | |
46 | * log, we must force a full commit before doing an fsync of the directory | |
47 | * where the unlink was done. | |
48 | * ---> record transid of last unlink/rename per directory | |
49 | * | |
50 | * mkdir foo/some_dir | |
51 | * normal commit | |
52 | * rename foo/some_dir foo2/some_dir | |
53 | * mkdir foo/some_dir | |
54 | * fsync foo/some_dir/some_file | |
55 | * | |
56 | * The fsync above will unlink the original some_dir without recording | |
57 | * it in its new location (foo2). After a crash, some_dir will be gone | |
58 | * unless the fsync of some_file forces a full commit | |
59 | * | |
60 | * 2) we must log any new names for any file or dir that is in the fsync | |
61 | * log. ---> check inode while renaming/linking. | |
62 | * | |
63 | * 2a) we must log any new names for any file or dir during rename | |
64 | * when the directory they are being removed from was logged. | |
65 | * ---> check inode and old parent dir during rename | |
66 | * | |
67 | * 2a is actually the more important variant. With the extra logging | |
68 | * a crash might unlink the old name without recreating the new one | |
69 | * | |
70 | * 3) after a crash, we must go through any directories with a link count | |
71 | * of zero and redo the rm -rf | |
72 | * | |
73 | * mkdir f1/foo | |
74 | * normal commit | |
75 | * rm -rf f1/foo | |
76 | * fsync(f1) | |
77 | * | |
78 | * The directory f1 was fully removed from the FS, but fsync was never | |
79 | * called on f1, only its parent dir. After a crash the rm -rf must | |
80 | * be replayed. This must be able to recurse down the entire | |
81 | * directory tree. The inode link count fixup code takes care of the | |
82 | * ugly details. | |
83 | */ | |
84 | ||
e02119d5 CM |
85 | /* |
86 | * stages for the tree walking. The first | |
87 | * stage (0) is to only pin down the blocks we find | |
88 | * the second stage (1) is to make sure that all the inodes | |
89 | * we find in the log are created in the subvolume. | |
90 | * | |
91 | * The last stage is to deal with directories and links and extents | |
92 | * and all the other fun semantics | |
93 | */ | |
94 | #define LOG_WALK_PIN_ONLY 0 | |
95 | #define LOG_WALK_REPLAY_INODES 1 | |
dd8e7217 JB |
96 | #define LOG_WALK_REPLAY_DIR_INDEX 2 |
97 | #define LOG_WALK_REPLAY_ALL 3 | |
e02119d5 | 98 | |
12fcfd22 | 99 | static int btrfs_log_inode(struct btrfs_trans_handle *trans, |
a59108a7 | 100 | struct btrfs_root *root, struct btrfs_inode *inode, |
49dae1bc FM |
101 | int inode_only, |
102 | const loff_t start, | |
8407f553 FM |
103 | const loff_t end, |
104 | struct btrfs_log_ctx *ctx); | |
ec051c0f YZ |
105 | static int link_to_fixup_dir(struct btrfs_trans_handle *trans, |
106 | struct btrfs_root *root, | |
107 | struct btrfs_path *path, u64 objectid); | |
12fcfd22 CM |
108 | static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans, |
109 | struct btrfs_root *root, | |
110 | struct btrfs_root *log, | |
111 | struct btrfs_path *path, | |
112 | u64 dirid, int del_all); | |
e02119d5 CM |
113 | |
114 | /* | |
115 | * tree logging is a special write ahead log used to make sure that | |
116 | * fsyncs and O_SYNCs can happen without doing full tree commits. | |
117 | * | |
118 | * Full tree commits are expensive because they require commonly | |
119 | * modified blocks to be recowed, creating many dirty pages in the | |
120 | * extent tree an 4x-6x higher write load than ext3. | |
121 | * | |
122 | * Instead of doing a tree commit on every fsync, we use the | |
123 | * key ranges and transaction ids to find items for a given file or directory | |
124 | * that have changed in this transaction. Those items are copied into | |
125 | * a special tree (one per subvolume root), that tree is written to disk | |
126 | * and then the fsync is considered complete. | |
127 | * | |
128 | * After a crash, items are copied out of the log-tree back into the | |
129 | * subvolume tree. Any file data extents found are recorded in the extent | |
130 | * allocation tree, and the log-tree freed. | |
131 | * | |
132 | * The log tree is read three times, once to pin down all the extents it is | |
133 | * using in ram and once, once to create all the inodes logged in the tree | |
134 | * and once to do all the other items. | |
135 | */ | |
136 | ||
e02119d5 CM |
137 | /* |
138 | * start a sub transaction and setup the log tree | |
139 | * this increments the log tree writer count to make the people | |
140 | * syncing the tree wait for us to finish | |
141 | */ | |
142 | static int start_log_trans(struct btrfs_trans_handle *trans, | |
8b050d35 MX |
143 | struct btrfs_root *root, |
144 | struct btrfs_log_ctx *ctx) | |
e02119d5 | 145 | { |
0b246afa | 146 | struct btrfs_fs_info *fs_info = root->fs_info; |
34eb2a52 | 147 | int ret = 0; |
7237f183 YZ |
148 | |
149 | mutex_lock(&root->log_mutex); | |
34eb2a52 | 150 | |
7237f183 | 151 | if (root->log_root) { |
0b246afa | 152 | if (btrfs_need_log_full_commit(fs_info, trans)) { |
50471a38 MX |
153 | ret = -EAGAIN; |
154 | goto out; | |
155 | } | |
34eb2a52 | 156 | |
ff782e0a | 157 | if (!root->log_start_pid) { |
27cdeb70 | 158 | clear_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state); |
34eb2a52 | 159 | root->log_start_pid = current->pid; |
ff782e0a | 160 | } else if (root->log_start_pid != current->pid) { |
27cdeb70 | 161 | set_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state); |
ff782e0a | 162 | } |
34eb2a52 | 163 | } else { |
0b246afa JM |
164 | mutex_lock(&fs_info->tree_log_mutex); |
165 | if (!fs_info->log_root_tree) | |
166 | ret = btrfs_init_log_root_tree(trans, fs_info); | |
167 | mutex_unlock(&fs_info->tree_log_mutex); | |
34eb2a52 Z |
168 | if (ret) |
169 | goto out; | |
ff782e0a | 170 | |
e02119d5 | 171 | ret = btrfs_add_log_tree(trans, root); |
4a500fd1 | 172 | if (ret) |
e87ac136 | 173 | goto out; |
34eb2a52 Z |
174 | |
175 | clear_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state); | |
176 | root->log_start_pid = current->pid; | |
e02119d5 | 177 | } |
34eb2a52 | 178 | |
2ecb7923 | 179 | atomic_inc(&root->log_batch); |
7237f183 | 180 | atomic_inc(&root->log_writers); |
8b050d35 | 181 | if (ctx) { |
34eb2a52 | 182 | int index = root->log_transid % 2; |
8b050d35 | 183 | list_add_tail(&ctx->list, &root->log_ctxs[index]); |
d1433deb | 184 | ctx->log_transid = root->log_transid; |
8b050d35 | 185 | } |
34eb2a52 | 186 | |
e87ac136 | 187 | out: |
7237f183 | 188 | mutex_unlock(&root->log_mutex); |
e87ac136 | 189 | return ret; |
e02119d5 CM |
190 | } |
191 | ||
192 | /* | |
193 | * returns 0 if there was a log transaction running and we were able | |
194 | * to join, or returns -ENOENT if there were not transactions | |
195 | * in progress | |
196 | */ | |
197 | static int join_running_log_trans(struct btrfs_root *root) | |
198 | { | |
199 | int ret = -ENOENT; | |
200 | ||
201 | smp_mb(); | |
202 | if (!root->log_root) | |
203 | return -ENOENT; | |
204 | ||
7237f183 | 205 | mutex_lock(&root->log_mutex); |
e02119d5 CM |
206 | if (root->log_root) { |
207 | ret = 0; | |
7237f183 | 208 | atomic_inc(&root->log_writers); |
e02119d5 | 209 | } |
7237f183 | 210 | mutex_unlock(&root->log_mutex); |
e02119d5 CM |
211 | return ret; |
212 | } | |
213 | ||
12fcfd22 CM |
214 | /* |
215 | * This either makes the current running log transaction wait | |
216 | * until you call btrfs_end_log_trans() or it makes any future | |
217 | * log transactions wait until you call btrfs_end_log_trans() | |
218 | */ | |
219 | int btrfs_pin_log_trans(struct btrfs_root *root) | |
220 | { | |
221 | int ret = -ENOENT; | |
222 | ||
223 | mutex_lock(&root->log_mutex); | |
224 | atomic_inc(&root->log_writers); | |
225 | mutex_unlock(&root->log_mutex); | |
226 | return ret; | |
227 | } | |
228 | ||
e02119d5 CM |
229 | /* |
230 | * indicate we're done making changes to the log tree | |
231 | * and wake up anyone waiting to do a sync | |
232 | */ | |
143bede5 | 233 | void btrfs_end_log_trans(struct btrfs_root *root) |
e02119d5 | 234 | { |
7237f183 | 235 | if (atomic_dec_and_test(&root->log_writers)) { |
779adf0f DS |
236 | /* |
237 | * Implicit memory barrier after atomic_dec_and_test | |
238 | */ | |
7237f183 YZ |
239 | if (waitqueue_active(&root->log_writer_wait)) |
240 | wake_up(&root->log_writer_wait); | |
241 | } | |
e02119d5 CM |
242 | } |
243 | ||
244 | ||
245 | /* | |
246 | * the walk control struct is used to pass state down the chain when | |
247 | * processing the log tree. The stage field tells us which part | |
248 | * of the log tree processing we are currently doing. The others | |
249 | * are state fields used for that specific part | |
250 | */ | |
251 | struct walk_control { | |
252 | /* should we free the extent on disk when done? This is used | |
253 | * at transaction commit time while freeing a log tree | |
254 | */ | |
255 | int free; | |
256 | ||
257 | /* should we write out the extent buffer? This is used | |
258 | * while flushing the log tree to disk during a sync | |
259 | */ | |
260 | int write; | |
261 | ||
262 | /* should we wait for the extent buffer io to finish? Also used | |
263 | * while flushing the log tree to disk for a sync | |
264 | */ | |
265 | int wait; | |
266 | ||
267 | /* pin only walk, we record which extents on disk belong to the | |
268 | * log trees | |
269 | */ | |
270 | int pin; | |
271 | ||
272 | /* what stage of the replay code we're currently in */ | |
273 | int stage; | |
274 | ||
275 | /* the root we are currently replaying */ | |
276 | struct btrfs_root *replay_dest; | |
277 | ||
278 | /* the trans handle for the current replay */ | |
279 | struct btrfs_trans_handle *trans; | |
280 | ||
281 | /* the function that gets used to process blocks we find in the | |
282 | * tree. Note the extent_buffer might not be up to date when it is | |
283 | * passed in, and it must be checked or read if you need the data | |
284 | * inside it | |
285 | */ | |
286 | int (*process_func)(struct btrfs_root *log, struct extent_buffer *eb, | |
287 | struct walk_control *wc, u64 gen); | |
288 | }; | |
289 | ||
290 | /* | |
291 | * process_func used to pin down extents, write them or wait on them | |
292 | */ | |
293 | static int process_one_buffer(struct btrfs_root *log, | |
294 | struct extent_buffer *eb, | |
295 | struct walk_control *wc, u64 gen) | |
296 | { | |
0b246afa | 297 | struct btrfs_fs_info *fs_info = log->fs_info; |
b50c6e25 JB |
298 | int ret = 0; |
299 | ||
8c2a1a30 JB |
300 | /* |
301 | * If this fs is mixed then we need to be able to process the leaves to | |
302 | * pin down any logged extents, so we have to read the block. | |
303 | */ | |
0b246afa | 304 | if (btrfs_fs_incompat(fs_info, MIXED_GROUPS)) { |
8c2a1a30 JB |
305 | ret = btrfs_read_buffer(eb, gen); |
306 | if (ret) | |
307 | return ret; | |
308 | } | |
309 | ||
04018de5 | 310 | if (wc->pin) |
2ff7e61e JM |
311 | ret = btrfs_pin_extent_for_log_replay(fs_info, eb->start, |
312 | eb->len); | |
e02119d5 | 313 | |
b50c6e25 | 314 | if (!ret && btrfs_buffer_uptodate(eb, gen, 0)) { |
8c2a1a30 | 315 | if (wc->pin && btrfs_header_level(eb) == 0) |
2ff7e61e | 316 | ret = btrfs_exclude_logged_extents(fs_info, eb); |
e02119d5 CM |
317 | if (wc->write) |
318 | btrfs_write_tree_block(eb); | |
319 | if (wc->wait) | |
320 | btrfs_wait_tree_block_writeback(eb); | |
321 | } | |
b50c6e25 | 322 | return ret; |
e02119d5 CM |
323 | } |
324 | ||
325 | /* | |
326 | * Item overwrite used by replay and tree logging. eb, slot and key all refer | |
327 | * to the src data we are copying out. | |
328 | * | |
329 | * root is the tree we are copying into, and path is a scratch | |
330 | * path for use in this function (it should be released on entry and | |
331 | * will be released on exit). | |
332 | * | |
333 | * If the key is already in the destination tree the existing item is | |
334 | * overwritten. If the existing item isn't big enough, it is extended. | |
335 | * If it is too large, it is truncated. | |
336 | * | |
337 | * If the key isn't in the destination yet, a new item is inserted. | |
338 | */ | |
339 | static noinline int overwrite_item(struct btrfs_trans_handle *trans, | |
340 | struct btrfs_root *root, | |
341 | struct btrfs_path *path, | |
342 | struct extent_buffer *eb, int slot, | |
343 | struct btrfs_key *key) | |
344 | { | |
2ff7e61e | 345 | struct btrfs_fs_info *fs_info = root->fs_info; |
e02119d5 CM |
346 | int ret; |
347 | u32 item_size; | |
348 | u64 saved_i_size = 0; | |
349 | int save_old_i_size = 0; | |
350 | unsigned long src_ptr; | |
351 | unsigned long dst_ptr; | |
352 | int overwrite_root = 0; | |
4bc4bee4 | 353 | bool inode_item = key->type == BTRFS_INODE_ITEM_KEY; |
e02119d5 CM |
354 | |
355 | if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) | |
356 | overwrite_root = 1; | |
357 | ||
358 | item_size = btrfs_item_size_nr(eb, slot); | |
359 | src_ptr = btrfs_item_ptr_offset(eb, slot); | |
360 | ||
361 | /* look for the key in the destination tree */ | |
362 | ret = btrfs_search_slot(NULL, root, key, path, 0, 0); | |
4bc4bee4 JB |
363 | if (ret < 0) |
364 | return ret; | |
365 | ||
e02119d5 CM |
366 | if (ret == 0) { |
367 | char *src_copy; | |
368 | char *dst_copy; | |
369 | u32 dst_size = btrfs_item_size_nr(path->nodes[0], | |
370 | path->slots[0]); | |
371 | if (dst_size != item_size) | |
372 | goto insert; | |
373 | ||
374 | if (item_size == 0) { | |
b3b4aa74 | 375 | btrfs_release_path(path); |
e02119d5 CM |
376 | return 0; |
377 | } | |
378 | dst_copy = kmalloc(item_size, GFP_NOFS); | |
379 | src_copy = kmalloc(item_size, GFP_NOFS); | |
2a29edc6 | 380 | if (!dst_copy || !src_copy) { |
b3b4aa74 | 381 | btrfs_release_path(path); |
2a29edc6 | 382 | kfree(dst_copy); |
383 | kfree(src_copy); | |
384 | return -ENOMEM; | |
385 | } | |
e02119d5 CM |
386 | |
387 | read_extent_buffer(eb, src_copy, src_ptr, item_size); | |
388 | ||
389 | dst_ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]); | |
390 | read_extent_buffer(path->nodes[0], dst_copy, dst_ptr, | |
391 | item_size); | |
392 | ret = memcmp(dst_copy, src_copy, item_size); | |
393 | ||
394 | kfree(dst_copy); | |
395 | kfree(src_copy); | |
396 | /* | |
397 | * they have the same contents, just return, this saves | |
398 | * us from cowing blocks in the destination tree and doing | |
399 | * extra writes that may not have been done by a previous | |
400 | * sync | |
401 | */ | |
402 | if (ret == 0) { | |
b3b4aa74 | 403 | btrfs_release_path(path); |
e02119d5 CM |
404 | return 0; |
405 | } | |
406 | ||
4bc4bee4 JB |
407 | /* |
408 | * We need to load the old nbytes into the inode so when we | |
409 | * replay the extents we've logged we get the right nbytes. | |
410 | */ | |
411 | if (inode_item) { | |
412 | struct btrfs_inode_item *item; | |
413 | u64 nbytes; | |
d555438b | 414 | u32 mode; |
4bc4bee4 JB |
415 | |
416 | item = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
417 | struct btrfs_inode_item); | |
418 | nbytes = btrfs_inode_nbytes(path->nodes[0], item); | |
419 | item = btrfs_item_ptr(eb, slot, | |
420 | struct btrfs_inode_item); | |
421 | btrfs_set_inode_nbytes(eb, item, nbytes); | |
d555438b JB |
422 | |
423 | /* | |
424 | * If this is a directory we need to reset the i_size to | |
425 | * 0 so that we can set it up properly when replaying | |
426 | * the rest of the items in this log. | |
427 | */ | |
428 | mode = btrfs_inode_mode(eb, item); | |
429 | if (S_ISDIR(mode)) | |
430 | btrfs_set_inode_size(eb, item, 0); | |
4bc4bee4 JB |
431 | } |
432 | } else if (inode_item) { | |
433 | struct btrfs_inode_item *item; | |
d555438b | 434 | u32 mode; |
4bc4bee4 JB |
435 | |
436 | /* | |
437 | * New inode, set nbytes to 0 so that the nbytes comes out | |
438 | * properly when we replay the extents. | |
439 | */ | |
440 | item = btrfs_item_ptr(eb, slot, struct btrfs_inode_item); | |
441 | btrfs_set_inode_nbytes(eb, item, 0); | |
d555438b JB |
442 | |
443 | /* | |
444 | * If this is a directory we need to reset the i_size to 0 so | |
445 | * that we can set it up properly when replaying the rest of | |
446 | * the items in this log. | |
447 | */ | |
448 | mode = btrfs_inode_mode(eb, item); | |
449 | if (S_ISDIR(mode)) | |
450 | btrfs_set_inode_size(eb, item, 0); | |
e02119d5 CM |
451 | } |
452 | insert: | |
b3b4aa74 | 453 | btrfs_release_path(path); |
e02119d5 | 454 | /* try to insert the key into the destination tree */ |
df8d116f | 455 | path->skip_release_on_error = 1; |
e02119d5 CM |
456 | ret = btrfs_insert_empty_item(trans, root, path, |
457 | key, item_size); | |
df8d116f | 458 | path->skip_release_on_error = 0; |
e02119d5 CM |
459 | |
460 | /* make sure any existing item is the correct size */ | |
df8d116f | 461 | if (ret == -EEXIST || ret == -EOVERFLOW) { |
e02119d5 CM |
462 | u32 found_size; |
463 | found_size = btrfs_item_size_nr(path->nodes[0], | |
464 | path->slots[0]); | |
143bede5 | 465 | if (found_size > item_size) |
2ff7e61e | 466 | btrfs_truncate_item(fs_info, path, item_size, 1); |
143bede5 | 467 | else if (found_size < item_size) |
2ff7e61e | 468 | btrfs_extend_item(fs_info, path, |
143bede5 | 469 | item_size - found_size); |
e02119d5 | 470 | } else if (ret) { |
4a500fd1 | 471 | return ret; |
e02119d5 CM |
472 | } |
473 | dst_ptr = btrfs_item_ptr_offset(path->nodes[0], | |
474 | path->slots[0]); | |
475 | ||
476 | /* don't overwrite an existing inode if the generation number | |
477 | * was logged as zero. This is done when the tree logging code | |
478 | * is just logging an inode to make sure it exists after recovery. | |
479 | * | |
480 | * Also, don't overwrite i_size on directories during replay. | |
481 | * log replay inserts and removes directory items based on the | |
482 | * state of the tree found in the subvolume, and i_size is modified | |
483 | * as it goes | |
484 | */ | |
485 | if (key->type == BTRFS_INODE_ITEM_KEY && ret == -EEXIST) { | |
486 | struct btrfs_inode_item *src_item; | |
487 | struct btrfs_inode_item *dst_item; | |
488 | ||
489 | src_item = (struct btrfs_inode_item *)src_ptr; | |
490 | dst_item = (struct btrfs_inode_item *)dst_ptr; | |
491 | ||
1a4bcf47 FM |
492 | if (btrfs_inode_generation(eb, src_item) == 0) { |
493 | struct extent_buffer *dst_eb = path->nodes[0]; | |
2f2ff0ee | 494 | const u64 ino_size = btrfs_inode_size(eb, src_item); |
1a4bcf47 | 495 | |
2f2ff0ee FM |
496 | /* |
497 | * For regular files an ino_size == 0 is used only when | |
498 | * logging that an inode exists, as part of a directory | |
499 | * fsync, and the inode wasn't fsynced before. In this | |
500 | * case don't set the size of the inode in the fs/subvol | |
501 | * tree, otherwise we would be throwing valid data away. | |
502 | */ | |
1a4bcf47 | 503 | if (S_ISREG(btrfs_inode_mode(eb, src_item)) && |
2f2ff0ee FM |
504 | S_ISREG(btrfs_inode_mode(dst_eb, dst_item)) && |
505 | ino_size != 0) { | |
1a4bcf47 | 506 | struct btrfs_map_token token; |
1a4bcf47 FM |
507 | |
508 | btrfs_init_map_token(&token); | |
509 | btrfs_set_token_inode_size(dst_eb, dst_item, | |
510 | ino_size, &token); | |
511 | } | |
e02119d5 | 512 | goto no_copy; |
1a4bcf47 | 513 | } |
e02119d5 CM |
514 | |
515 | if (overwrite_root && | |
516 | S_ISDIR(btrfs_inode_mode(eb, src_item)) && | |
517 | S_ISDIR(btrfs_inode_mode(path->nodes[0], dst_item))) { | |
518 | save_old_i_size = 1; | |
519 | saved_i_size = btrfs_inode_size(path->nodes[0], | |
520 | dst_item); | |
521 | } | |
522 | } | |
523 | ||
524 | copy_extent_buffer(path->nodes[0], eb, dst_ptr, | |
525 | src_ptr, item_size); | |
526 | ||
527 | if (save_old_i_size) { | |
528 | struct btrfs_inode_item *dst_item; | |
529 | dst_item = (struct btrfs_inode_item *)dst_ptr; | |
530 | btrfs_set_inode_size(path->nodes[0], dst_item, saved_i_size); | |
531 | } | |
532 | ||
533 | /* make sure the generation is filled in */ | |
534 | if (key->type == BTRFS_INODE_ITEM_KEY) { | |
535 | struct btrfs_inode_item *dst_item; | |
536 | dst_item = (struct btrfs_inode_item *)dst_ptr; | |
537 | if (btrfs_inode_generation(path->nodes[0], dst_item) == 0) { | |
538 | btrfs_set_inode_generation(path->nodes[0], dst_item, | |
539 | trans->transid); | |
540 | } | |
541 | } | |
542 | no_copy: | |
543 | btrfs_mark_buffer_dirty(path->nodes[0]); | |
b3b4aa74 | 544 | btrfs_release_path(path); |
e02119d5 CM |
545 | return 0; |
546 | } | |
547 | ||
548 | /* | |
549 | * simple helper to read an inode off the disk from a given root | |
550 | * This can only be called for subvolume roots and not for the log | |
551 | */ | |
552 | static noinline struct inode *read_one_inode(struct btrfs_root *root, | |
553 | u64 objectid) | |
554 | { | |
5d4f98a2 | 555 | struct btrfs_key key; |
e02119d5 | 556 | struct inode *inode; |
e02119d5 | 557 | |
5d4f98a2 YZ |
558 | key.objectid = objectid; |
559 | key.type = BTRFS_INODE_ITEM_KEY; | |
560 | key.offset = 0; | |
73f73415 | 561 | inode = btrfs_iget(root->fs_info->sb, &key, root, NULL); |
5d4f98a2 YZ |
562 | if (IS_ERR(inode)) { |
563 | inode = NULL; | |
564 | } else if (is_bad_inode(inode)) { | |
e02119d5 CM |
565 | iput(inode); |
566 | inode = NULL; | |
567 | } | |
568 | return inode; | |
569 | } | |
570 | ||
571 | /* replays a single extent in 'eb' at 'slot' with 'key' into the | |
572 | * subvolume 'root'. path is released on entry and should be released | |
573 | * on exit. | |
574 | * | |
575 | * extents in the log tree have not been allocated out of the extent | |
576 | * tree yet. So, this completes the allocation, taking a reference | |
577 | * as required if the extent already exists or creating a new extent | |
578 | * if it isn't in the extent allocation tree yet. | |
579 | * | |
580 | * The extent is inserted into the file, dropping any existing extents | |
581 | * from the file that overlap the new one. | |
582 | */ | |
583 | static noinline int replay_one_extent(struct btrfs_trans_handle *trans, | |
584 | struct btrfs_root *root, | |
585 | struct btrfs_path *path, | |
586 | struct extent_buffer *eb, int slot, | |
587 | struct btrfs_key *key) | |
588 | { | |
0b246afa | 589 | struct btrfs_fs_info *fs_info = root->fs_info; |
e02119d5 | 590 | int found_type; |
e02119d5 | 591 | u64 extent_end; |
e02119d5 | 592 | u64 start = key->offset; |
4bc4bee4 | 593 | u64 nbytes = 0; |
e02119d5 CM |
594 | struct btrfs_file_extent_item *item; |
595 | struct inode *inode = NULL; | |
596 | unsigned long size; | |
597 | int ret = 0; | |
598 | ||
599 | item = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item); | |
600 | found_type = btrfs_file_extent_type(eb, item); | |
601 | ||
d899e052 | 602 | if (found_type == BTRFS_FILE_EXTENT_REG || |
4bc4bee4 JB |
603 | found_type == BTRFS_FILE_EXTENT_PREALLOC) { |
604 | nbytes = btrfs_file_extent_num_bytes(eb, item); | |
605 | extent_end = start + nbytes; | |
606 | ||
607 | /* | |
608 | * We don't add to the inodes nbytes if we are prealloc or a | |
609 | * hole. | |
610 | */ | |
611 | if (btrfs_file_extent_disk_bytenr(eb, item) == 0) | |
612 | nbytes = 0; | |
613 | } else if (found_type == BTRFS_FILE_EXTENT_INLINE) { | |
514ac8ad | 614 | size = btrfs_file_extent_inline_len(eb, slot, item); |
4bc4bee4 | 615 | nbytes = btrfs_file_extent_ram_bytes(eb, item); |
da17066c | 616 | extent_end = ALIGN(start + size, |
0b246afa | 617 | fs_info->sectorsize); |
e02119d5 CM |
618 | } else { |
619 | ret = 0; | |
620 | goto out; | |
621 | } | |
622 | ||
623 | inode = read_one_inode(root, key->objectid); | |
624 | if (!inode) { | |
625 | ret = -EIO; | |
626 | goto out; | |
627 | } | |
628 | ||
629 | /* | |
630 | * first check to see if we already have this extent in the | |
631 | * file. This must be done before the btrfs_drop_extents run | |
632 | * so we don't try to drop this extent. | |
633 | */ | |
f85b7379 DS |
634 | ret = btrfs_lookup_file_extent(trans, root, path, |
635 | btrfs_ino(BTRFS_I(inode)), start, 0); | |
e02119d5 | 636 | |
d899e052 YZ |
637 | if (ret == 0 && |
638 | (found_type == BTRFS_FILE_EXTENT_REG || | |
639 | found_type == BTRFS_FILE_EXTENT_PREALLOC)) { | |
e02119d5 CM |
640 | struct btrfs_file_extent_item cmp1; |
641 | struct btrfs_file_extent_item cmp2; | |
642 | struct btrfs_file_extent_item *existing; | |
643 | struct extent_buffer *leaf; | |
644 | ||
645 | leaf = path->nodes[0]; | |
646 | existing = btrfs_item_ptr(leaf, path->slots[0], | |
647 | struct btrfs_file_extent_item); | |
648 | ||
649 | read_extent_buffer(eb, &cmp1, (unsigned long)item, | |
650 | sizeof(cmp1)); | |
651 | read_extent_buffer(leaf, &cmp2, (unsigned long)existing, | |
652 | sizeof(cmp2)); | |
653 | ||
654 | /* | |
655 | * we already have a pointer to this exact extent, | |
656 | * we don't have to do anything | |
657 | */ | |
658 | if (memcmp(&cmp1, &cmp2, sizeof(cmp1)) == 0) { | |
b3b4aa74 | 659 | btrfs_release_path(path); |
e02119d5 CM |
660 | goto out; |
661 | } | |
662 | } | |
b3b4aa74 | 663 | btrfs_release_path(path); |
e02119d5 CM |
664 | |
665 | /* drop any overlapping extents */ | |
2671485d | 666 | ret = btrfs_drop_extents(trans, root, inode, start, extent_end, 1); |
3650860b JB |
667 | if (ret) |
668 | goto out; | |
e02119d5 | 669 | |
07d400a6 YZ |
670 | if (found_type == BTRFS_FILE_EXTENT_REG || |
671 | found_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
5d4f98a2 | 672 | u64 offset; |
07d400a6 YZ |
673 | unsigned long dest_offset; |
674 | struct btrfs_key ins; | |
675 | ||
3168021c FM |
676 | if (btrfs_file_extent_disk_bytenr(eb, item) == 0 && |
677 | btrfs_fs_incompat(fs_info, NO_HOLES)) | |
678 | goto update_inode; | |
679 | ||
07d400a6 YZ |
680 | ret = btrfs_insert_empty_item(trans, root, path, key, |
681 | sizeof(*item)); | |
3650860b JB |
682 | if (ret) |
683 | goto out; | |
07d400a6 YZ |
684 | dest_offset = btrfs_item_ptr_offset(path->nodes[0], |
685 | path->slots[0]); | |
686 | copy_extent_buffer(path->nodes[0], eb, dest_offset, | |
687 | (unsigned long)item, sizeof(*item)); | |
688 | ||
689 | ins.objectid = btrfs_file_extent_disk_bytenr(eb, item); | |
690 | ins.offset = btrfs_file_extent_disk_num_bytes(eb, item); | |
691 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
5d4f98a2 | 692 | offset = key->offset - btrfs_file_extent_offset(eb, item); |
07d400a6 | 693 | |
df2c95f3 QW |
694 | /* |
695 | * Manually record dirty extent, as here we did a shallow | |
696 | * file extent item copy and skip normal backref update, | |
697 | * but modifying extent tree all by ourselves. | |
698 | * So need to manually record dirty extent for qgroup, | |
699 | * as the owner of the file extent changed from log tree | |
700 | * (doesn't affect qgroup) to fs/file tree(affects qgroup) | |
701 | */ | |
0b246afa | 702 | ret = btrfs_qgroup_trace_extent(trans, fs_info, |
df2c95f3 QW |
703 | btrfs_file_extent_disk_bytenr(eb, item), |
704 | btrfs_file_extent_disk_num_bytes(eb, item), | |
705 | GFP_NOFS); | |
706 | if (ret < 0) | |
707 | goto out; | |
708 | ||
07d400a6 YZ |
709 | if (ins.objectid > 0) { |
710 | u64 csum_start; | |
711 | u64 csum_end; | |
712 | LIST_HEAD(ordered_sums); | |
713 | /* | |
714 | * is this extent already allocated in the extent | |
715 | * allocation tree? If so, just add a reference | |
716 | */ | |
2ff7e61e | 717 | ret = btrfs_lookup_data_extent(fs_info, ins.objectid, |
07d400a6 YZ |
718 | ins.offset); |
719 | if (ret == 0) { | |
2ff7e61e | 720 | ret = btrfs_inc_extent_ref(trans, fs_info, |
07d400a6 | 721 | ins.objectid, ins.offset, |
5d4f98a2 | 722 | 0, root->root_key.objectid, |
b06c4bf5 | 723 | key->objectid, offset); |
b50c6e25 JB |
724 | if (ret) |
725 | goto out; | |
07d400a6 YZ |
726 | } else { |
727 | /* | |
728 | * insert the extent pointer in the extent | |
729 | * allocation tree | |
730 | */ | |
5d4f98a2 | 731 | ret = btrfs_alloc_logged_file_extent(trans, |
2ff7e61e JM |
732 | fs_info, |
733 | root->root_key.objectid, | |
5d4f98a2 | 734 | key->objectid, offset, &ins); |
b50c6e25 JB |
735 | if (ret) |
736 | goto out; | |
07d400a6 | 737 | } |
b3b4aa74 | 738 | btrfs_release_path(path); |
07d400a6 YZ |
739 | |
740 | if (btrfs_file_extent_compression(eb, item)) { | |
741 | csum_start = ins.objectid; | |
742 | csum_end = csum_start + ins.offset; | |
743 | } else { | |
744 | csum_start = ins.objectid + | |
745 | btrfs_file_extent_offset(eb, item); | |
746 | csum_end = csum_start + | |
747 | btrfs_file_extent_num_bytes(eb, item); | |
748 | } | |
749 | ||
750 | ret = btrfs_lookup_csums_range(root->log_root, | |
751 | csum_start, csum_end - 1, | |
a2de733c | 752 | &ordered_sums, 0); |
3650860b JB |
753 | if (ret) |
754 | goto out; | |
b84b8390 FM |
755 | /* |
756 | * Now delete all existing cums in the csum root that | |
757 | * cover our range. We do this because we can have an | |
758 | * extent that is completely referenced by one file | |
759 | * extent item and partially referenced by another | |
760 | * file extent item (like after using the clone or | |
761 | * extent_same ioctls). In this case if we end up doing | |
762 | * the replay of the one that partially references the | |
763 | * extent first, and we do not do the csum deletion | |
764 | * below, we can get 2 csum items in the csum tree that | |
765 | * overlap each other. For example, imagine our log has | |
766 | * the two following file extent items: | |
767 | * | |
768 | * key (257 EXTENT_DATA 409600) | |
769 | * extent data disk byte 12845056 nr 102400 | |
770 | * extent data offset 20480 nr 20480 ram 102400 | |
771 | * | |
772 | * key (257 EXTENT_DATA 819200) | |
773 | * extent data disk byte 12845056 nr 102400 | |
774 | * extent data offset 0 nr 102400 ram 102400 | |
775 | * | |
776 | * Where the second one fully references the 100K extent | |
777 | * that starts at disk byte 12845056, and the log tree | |
778 | * has a single csum item that covers the entire range | |
779 | * of the extent: | |
780 | * | |
781 | * key (EXTENT_CSUM EXTENT_CSUM 12845056) itemsize 100 | |
782 | * | |
783 | * After the first file extent item is replayed, the | |
784 | * csum tree gets the following csum item: | |
785 | * | |
786 | * key (EXTENT_CSUM EXTENT_CSUM 12865536) itemsize 20 | |
787 | * | |
788 | * Which covers the 20K sub-range starting at offset 20K | |
789 | * of our extent. Now when we replay the second file | |
790 | * extent item, if we do not delete existing csum items | |
791 | * that cover any of its blocks, we end up getting two | |
792 | * csum items in our csum tree that overlap each other: | |
793 | * | |
794 | * key (EXTENT_CSUM EXTENT_CSUM 12845056) itemsize 100 | |
795 | * key (EXTENT_CSUM EXTENT_CSUM 12865536) itemsize 20 | |
796 | * | |
797 | * Which is a problem, because after this anyone trying | |
798 | * to lookup up for the checksum of any block of our | |
799 | * extent starting at an offset of 40K or higher, will | |
800 | * end up looking at the second csum item only, which | |
801 | * does not contain the checksum for any block starting | |
802 | * at offset 40K or higher of our extent. | |
803 | */ | |
07d400a6 YZ |
804 | while (!list_empty(&ordered_sums)) { |
805 | struct btrfs_ordered_sum *sums; | |
806 | sums = list_entry(ordered_sums.next, | |
807 | struct btrfs_ordered_sum, | |
808 | list); | |
b84b8390 | 809 | if (!ret) |
0b246afa | 810 | ret = btrfs_del_csums(trans, fs_info, |
5b4aacef JM |
811 | sums->bytenr, |
812 | sums->len); | |
3650860b JB |
813 | if (!ret) |
814 | ret = btrfs_csum_file_blocks(trans, | |
0b246afa | 815 | fs_info->csum_root, sums); |
07d400a6 YZ |
816 | list_del(&sums->list); |
817 | kfree(sums); | |
818 | } | |
3650860b JB |
819 | if (ret) |
820 | goto out; | |
07d400a6 | 821 | } else { |
b3b4aa74 | 822 | btrfs_release_path(path); |
07d400a6 YZ |
823 | } |
824 | } else if (found_type == BTRFS_FILE_EXTENT_INLINE) { | |
825 | /* inline extents are easy, we just overwrite them */ | |
826 | ret = overwrite_item(trans, root, path, eb, slot, key); | |
3650860b JB |
827 | if (ret) |
828 | goto out; | |
07d400a6 | 829 | } |
e02119d5 | 830 | |
4bc4bee4 | 831 | inode_add_bytes(inode, nbytes); |
3168021c | 832 | update_inode: |
b9959295 | 833 | ret = btrfs_update_inode(trans, root, inode); |
e02119d5 CM |
834 | out: |
835 | if (inode) | |
836 | iput(inode); | |
837 | return ret; | |
838 | } | |
839 | ||
840 | /* | |
841 | * when cleaning up conflicts between the directory names in the | |
842 | * subvolume, directory names in the log and directory names in the | |
843 | * inode back references, we may have to unlink inodes from directories. | |
844 | * | |
845 | * This is a helper function to do the unlink of a specific directory | |
846 | * item | |
847 | */ | |
848 | static noinline int drop_one_dir_item(struct btrfs_trans_handle *trans, | |
849 | struct btrfs_root *root, | |
850 | struct btrfs_path *path, | |
207e7d92 | 851 | struct btrfs_inode *dir, |
e02119d5 CM |
852 | struct btrfs_dir_item *di) |
853 | { | |
2ff7e61e | 854 | struct btrfs_fs_info *fs_info = root->fs_info; |
e02119d5 CM |
855 | struct inode *inode; |
856 | char *name; | |
857 | int name_len; | |
858 | struct extent_buffer *leaf; | |
859 | struct btrfs_key location; | |
860 | int ret; | |
861 | ||
862 | leaf = path->nodes[0]; | |
863 | ||
864 | btrfs_dir_item_key_to_cpu(leaf, di, &location); | |
865 | name_len = btrfs_dir_name_len(leaf, di); | |
866 | name = kmalloc(name_len, GFP_NOFS); | |
2a29edc6 | 867 | if (!name) |
868 | return -ENOMEM; | |
869 | ||
e02119d5 | 870 | read_extent_buffer(leaf, name, (unsigned long)(di + 1), name_len); |
b3b4aa74 | 871 | btrfs_release_path(path); |
e02119d5 CM |
872 | |
873 | inode = read_one_inode(root, location.objectid); | |
c00e9493 | 874 | if (!inode) { |
3650860b JB |
875 | ret = -EIO; |
876 | goto out; | |
c00e9493 | 877 | } |
e02119d5 | 878 | |
ec051c0f | 879 | ret = link_to_fixup_dir(trans, root, path, location.objectid); |
3650860b JB |
880 | if (ret) |
881 | goto out; | |
12fcfd22 | 882 | |
207e7d92 NB |
883 | ret = btrfs_unlink_inode(trans, root, dir, BTRFS_I(inode), name, |
884 | name_len); | |
3650860b JB |
885 | if (ret) |
886 | goto out; | |
ada9af21 | 887 | else |
2ff7e61e | 888 | ret = btrfs_run_delayed_items(trans, fs_info); |
3650860b | 889 | out: |
e02119d5 | 890 | kfree(name); |
e02119d5 CM |
891 | iput(inode); |
892 | return ret; | |
893 | } | |
894 | ||
895 | /* | |
896 | * helper function to see if a given name and sequence number found | |
897 | * in an inode back reference are already in a directory and correctly | |
898 | * point to this inode | |
899 | */ | |
900 | static noinline int inode_in_dir(struct btrfs_root *root, | |
901 | struct btrfs_path *path, | |
902 | u64 dirid, u64 objectid, u64 index, | |
903 | const char *name, int name_len) | |
904 | { | |
905 | struct btrfs_dir_item *di; | |
906 | struct btrfs_key location; | |
907 | int match = 0; | |
908 | ||
909 | di = btrfs_lookup_dir_index_item(NULL, root, path, dirid, | |
910 | index, name, name_len, 0); | |
911 | if (di && !IS_ERR(di)) { | |
912 | btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location); | |
913 | if (location.objectid != objectid) | |
914 | goto out; | |
915 | } else | |
916 | goto out; | |
b3b4aa74 | 917 | btrfs_release_path(path); |
e02119d5 CM |
918 | |
919 | di = btrfs_lookup_dir_item(NULL, root, path, dirid, name, name_len, 0); | |
920 | if (di && !IS_ERR(di)) { | |
921 | btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location); | |
922 | if (location.objectid != objectid) | |
923 | goto out; | |
924 | } else | |
925 | goto out; | |
926 | match = 1; | |
927 | out: | |
b3b4aa74 | 928 | btrfs_release_path(path); |
e02119d5 CM |
929 | return match; |
930 | } | |
931 | ||
932 | /* | |
933 | * helper function to check a log tree for a named back reference in | |
934 | * an inode. This is used to decide if a back reference that is | |
935 | * found in the subvolume conflicts with what we find in the log. | |
936 | * | |
937 | * inode backreferences may have multiple refs in a single item, | |
938 | * during replay we process one reference at a time, and we don't | |
939 | * want to delete valid links to a file from the subvolume if that | |
940 | * link is also in the log. | |
941 | */ | |
942 | static noinline int backref_in_log(struct btrfs_root *log, | |
943 | struct btrfs_key *key, | |
f186373f | 944 | u64 ref_objectid, |
df8d116f | 945 | const char *name, int namelen) |
e02119d5 CM |
946 | { |
947 | struct btrfs_path *path; | |
948 | struct btrfs_inode_ref *ref; | |
949 | unsigned long ptr; | |
950 | unsigned long ptr_end; | |
951 | unsigned long name_ptr; | |
952 | int found_name_len; | |
953 | int item_size; | |
954 | int ret; | |
955 | int match = 0; | |
956 | ||
957 | path = btrfs_alloc_path(); | |
2a29edc6 | 958 | if (!path) |
959 | return -ENOMEM; | |
960 | ||
e02119d5 CM |
961 | ret = btrfs_search_slot(NULL, log, key, path, 0, 0); |
962 | if (ret != 0) | |
963 | goto out; | |
964 | ||
e02119d5 | 965 | ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]); |
f186373f MF |
966 | |
967 | if (key->type == BTRFS_INODE_EXTREF_KEY) { | |
968 | if (btrfs_find_name_in_ext_backref(path, ref_objectid, | |
969 | name, namelen, NULL)) | |
970 | match = 1; | |
971 | ||
972 | goto out; | |
973 | } | |
974 | ||
975 | item_size = btrfs_item_size_nr(path->nodes[0], path->slots[0]); | |
e02119d5 CM |
976 | ptr_end = ptr + item_size; |
977 | while (ptr < ptr_end) { | |
978 | ref = (struct btrfs_inode_ref *)ptr; | |
979 | found_name_len = btrfs_inode_ref_name_len(path->nodes[0], ref); | |
980 | if (found_name_len == namelen) { | |
981 | name_ptr = (unsigned long)(ref + 1); | |
982 | ret = memcmp_extent_buffer(path->nodes[0], name, | |
983 | name_ptr, namelen); | |
984 | if (ret == 0) { | |
985 | match = 1; | |
986 | goto out; | |
987 | } | |
988 | } | |
989 | ptr = (unsigned long)(ref + 1) + found_name_len; | |
990 | } | |
991 | out: | |
992 | btrfs_free_path(path); | |
993 | return match; | |
994 | } | |
995 | ||
5a1d7843 | 996 | static inline int __add_inode_ref(struct btrfs_trans_handle *trans, |
e02119d5 | 997 | struct btrfs_root *root, |
e02119d5 | 998 | struct btrfs_path *path, |
5a1d7843 | 999 | struct btrfs_root *log_root, |
94c91a1f NB |
1000 | struct btrfs_inode *dir, |
1001 | struct btrfs_inode *inode, | |
f186373f MF |
1002 | u64 inode_objectid, u64 parent_objectid, |
1003 | u64 ref_index, char *name, int namelen, | |
1004 | int *search_done) | |
e02119d5 | 1005 | { |
2ff7e61e | 1006 | struct btrfs_fs_info *fs_info = root->fs_info; |
34f3e4f2 | 1007 | int ret; |
f186373f MF |
1008 | char *victim_name; |
1009 | int victim_name_len; | |
1010 | struct extent_buffer *leaf; | |
5a1d7843 | 1011 | struct btrfs_dir_item *di; |
f186373f MF |
1012 | struct btrfs_key search_key; |
1013 | struct btrfs_inode_extref *extref; | |
c622ae60 | 1014 | |
f186373f MF |
1015 | again: |
1016 | /* Search old style refs */ | |
1017 | search_key.objectid = inode_objectid; | |
1018 | search_key.type = BTRFS_INODE_REF_KEY; | |
1019 | search_key.offset = parent_objectid; | |
1020 | ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0); | |
e02119d5 | 1021 | if (ret == 0) { |
e02119d5 CM |
1022 | struct btrfs_inode_ref *victim_ref; |
1023 | unsigned long ptr; | |
1024 | unsigned long ptr_end; | |
f186373f MF |
1025 | |
1026 | leaf = path->nodes[0]; | |
e02119d5 CM |
1027 | |
1028 | /* are we trying to overwrite a back ref for the root directory | |
1029 | * if so, just jump out, we're done | |
1030 | */ | |
f186373f | 1031 | if (search_key.objectid == search_key.offset) |
5a1d7843 | 1032 | return 1; |
e02119d5 CM |
1033 | |
1034 | /* check all the names in this back reference to see | |
1035 | * if they are in the log. if so, we allow them to stay | |
1036 | * otherwise they must be unlinked as a conflict | |
1037 | */ | |
1038 | ptr = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
1039 | ptr_end = ptr + btrfs_item_size_nr(leaf, path->slots[0]); | |
d397712b | 1040 | while (ptr < ptr_end) { |
e02119d5 CM |
1041 | victim_ref = (struct btrfs_inode_ref *)ptr; |
1042 | victim_name_len = btrfs_inode_ref_name_len(leaf, | |
1043 | victim_ref); | |
1044 | victim_name = kmalloc(victim_name_len, GFP_NOFS); | |
3650860b JB |
1045 | if (!victim_name) |
1046 | return -ENOMEM; | |
e02119d5 CM |
1047 | |
1048 | read_extent_buffer(leaf, victim_name, | |
1049 | (unsigned long)(victim_ref + 1), | |
1050 | victim_name_len); | |
1051 | ||
f186373f MF |
1052 | if (!backref_in_log(log_root, &search_key, |
1053 | parent_objectid, | |
1054 | victim_name, | |
e02119d5 | 1055 | victim_name_len)) { |
94c91a1f | 1056 | inc_nlink(&inode->vfs_inode); |
b3b4aa74 | 1057 | btrfs_release_path(path); |
12fcfd22 | 1058 | |
94c91a1f | 1059 | ret = btrfs_unlink_inode(trans, root, dir, inode, |
4ec5934e | 1060 | victim_name, victim_name_len); |
f186373f | 1061 | kfree(victim_name); |
3650860b JB |
1062 | if (ret) |
1063 | return ret; | |
2ff7e61e | 1064 | ret = btrfs_run_delayed_items(trans, fs_info); |
ada9af21 FDBM |
1065 | if (ret) |
1066 | return ret; | |
f186373f MF |
1067 | *search_done = 1; |
1068 | goto again; | |
e02119d5 CM |
1069 | } |
1070 | kfree(victim_name); | |
f186373f | 1071 | |
e02119d5 CM |
1072 | ptr = (unsigned long)(victim_ref + 1) + victim_name_len; |
1073 | } | |
e02119d5 | 1074 | |
c622ae60 | 1075 | /* |
1076 | * NOTE: we have searched root tree and checked the | |
bb7ab3b9 | 1077 | * corresponding ref, it does not need to check again. |
c622ae60 | 1078 | */ |
5a1d7843 | 1079 | *search_done = 1; |
e02119d5 | 1080 | } |
b3b4aa74 | 1081 | btrfs_release_path(path); |
e02119d5 | 1082 | |
f186373f MF |
1083 | /* Same search but for extended refs */ |
1084 | extref = btrfs_lookup_inode_extref(NULL, root, path, name, namelen, | |
1085 | inode_objectid, parent_objectid, 0, | |
1086 | 0); | |
1087 | if (!IS_ERR_OR_NULL(extref)) { | |
1088 | u32 item_size; | |
1089 | u32 cur_offset = 0; | |
1090 | unsigned long base; | |
1091 | struct inode *victim_parent; | |
1092 | ||
1093 | leaf = path->nodes[0]; | |
1094 | ||
1095 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1096 | base = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
1097 | ||
1098 | while (cur_offset < item_size) { | |
dd9ef135 | 1099 | extref = (struct btrfs_inode_extref *)(base + cur_offset); |
f186373f MF |
1100 | |
1101 | victim_name_len = btrfs_inode_extref_name_len(leaf, extref); | |
1102 | ||
1103 | if (btrfs_inode_extref_parent(leaf, extref) != parent_objectid) | |
1104 | goto next; | |
1105 | ||
1106 | victim_name = kmalloc(victim_name_len, GFP_NOFS); | |
3650860b JB |
1107 | if (!victim_name) |
1108 | return -ENOMEM; | |
f186373f MF |
1109 | read_extent_buffer(leaf, victim_name, (unsigned long)&extref->name, |
1110 | victim_name_len); | |
1111 | ||
1112 | search_key.objectid = inode_objectid; | |
1113 | search_key.type = BTRFS_INODE_EXTREF_KEY; | |
1114 | search_key.offset = btrfs_extref_hash(parent_objectid, | |
1115 | victim_name, | |
1116 | victim_name_len); | |
1117 | ret = 0; | |
1118 | if (!backref_in_log(log_root, &search_key, | |
1119 | parent_objectid, victim_name, | |
1120 | victim_name_len)) { | |
1121 | ret = -ENOENT; | |
1122 | victim_parent = read_one_inode(root, | |
94c91a1f | 1123 | parent_objectid); |
f186373f | 1124 | if (victim_parent) { |
94c91a1f | 1125 | inc_nlink(&inode->vfs_inode); |
f186373f MF |
1126 | btrfs_release_path(path); |
1127 | ||
1128 | ret = btrfs_unlink_inode(trans, root, | |
4ec5934e | 1129 | BTRFS_I(victim_parent), |
94c91a1f | 1130 | inode, |
4ec5934e NB |
1131 | victim_name, |
1132 | victim_name_len); | |
ada9af21 FDBM |
1133 | if (!ret) |
1134 | ret = btrfs_run_delayed_items( | |
2ff7e61e JM |
1135 | trans, |
1136 | fs_info); | |
f186373f | 1137 | } |
f186373f MF |
1138 | iput(victim_parent); |
1139 | kfree(victim_name); | |
3650860b JB |
1140 | if (ret) |
1141 | return ret; | |
f186373f MF |
1142 | *search_done = 1; |
1143 | goto again; | |
1144 | } | |
1145 | kfree(victim_name); | |
3650860b JB |
1146 | if (ret) |
1147 | return ret; | |
f186373f MF |
1148 | next: |
1149 | cur_offset += victim_name_len + sizeof(*extref); | |
1150 | } | |
1151 | *search_done = 1; | |
1152 | } | |
1153 | btrfs_release_path(path); | |
1154 | ||
34f3e4f2 | 1155 | /* look for a conflicting sequence number */ |
94c91a1f | 1156 | di = btrfs_lookup_dir_index_item(trans, root, path, btrfs_ino(dir), |
f186373f | 1157 | ref_index, name, namelen, 0); |
34f3e4f2 | 1158 | if (di && !IS_ERR(di)) { |
94c91a1f | 1159 | ret = drop_one_dir_item(trans, root, path, dir, di); |
3650860b JB |
1160 | if (ret) |
1161 | return ret; | |
34f3e4f2 | 1162 | } |
1163 | btrfs_release_path(path); | |
1164 | ||
1165 | /* look for a conflicing name */ | |
94c91a1f | 1166 | di = btrfs_lookup_dir_item(trans, root, path, btrfs_ino(dir), |
34f3e4f2 | 1167 | name, namelen, 0); |
1168 | if (di && !IS_ERR(di)) { | |
94c91a1f | 1169 | ret = drop_one_dir_item(trans, root, path, dir, di); |
3650860b JB |
1170 | if (ret) |
1171 | return ret; | |
34f3e4f2 | 1172 | } |
1173 | btrfs_release_path(path); | |
1174 | ||
5a1d7843 JS |
1175 | return 0; |
1176 | } | |
e02119d5 | 1177 | |
26a836ce SY |
1178 | static int extref_get_fields(struct extent_buffer *eb, int slot, |
1179 | unsigned long ref_ptr, u32 *namelen, char **name, | |
1180 | u64 *index, u64 *parent_objectid) | |
f186373f MF |
1181 | { |
1182 | struct btrfs_inode_extref *extref; | |
1183 | ||
1184 | extref = (struct btrfs_inode_extref *)ref_ptr; | |
1185 | ||
1186 | *namelen = btrfs_inode_extref_name_len(eb, extref); | |
26a836ce SY |
1187 | if (!btrfs_is_name_len_valid(eb, slot, (unsigned long)&extref->name, |
1188 | *namelen)) | |
1189 | return -EIO; | |
1190 | ||
f186373f MF |
1191 | *name = kmalloc(*namelen, GFP_NOFS); |
1192 | if (*name == NULL) | |
1193 | return -ENOMEM; | |
1194 | ||
1195 | read_extent_buffer(eb, *name, (unsigned long)&extref->name, | |
1196 | *namelen); | |
1197 | ||
1198 | *index = btrfs_inode_extref_index(eb, extref); | |
1199 | if (parent_objectid) | |
1200 | *parent_objectid = btrfs_inode_extref_parent(eb, extref); | |
1201 | ||
1202 | return 0; | |
1203 | } | |
1204 | ||
26a836ce SY |
1205 | static int ref_get_fields(struct extent_buffer *eb, int slot, |
1206 | unsigned long ref_ptr, u32 *namelen, char **name, | |
1207 | u64 *index) | |
f186373f MF |
1208 | { |
1209 | struct btrfs_inode_ref *ref; | |
1210 | ||
1211 | ref = (struct btrfs_inode_ref *)ref_ptr; | |
1212 | ||
1213 | *namelen = btrfs_inode_ref_name_len(eb, ref); | |
26a836ce SY |
1214 | if (!btrfs_is_name_len_valid(eb, slot, (unsigned long)(ref + 1), |
1215 | *namelen)) | |
1216 | return -EIO; | |
1217 | ||
f186373f MF |
1218 | *name = kmalloc(*namelen, GFP_NOFS); |
1219 | if (*name == NULL) | |
1220 | return -ENOMEM; | |
1221 | ||
1222 | read_extent_buffer(eb, *name, (unsigned long)(ref + 1), *namelen); | |
1223 | ||
1224 | *index = btrfs_inode_ref_index(eb, ref); | |
1225 | ||
1226 | return 0; | |
1227 | } | |
1228 | ||
5a1d7843 JS |
1229 | /* |
1230 | * replay one inode back reference item found in the log tree. | |
1231 | * eb, slot and key refer to the buffer and key found in the log tree. | |
1232 | * root is the destination we are replaying into, and path is for temp | |
1233 | * use by this function. (it should be released on return). | |
1234 | */ | |
1235 | static noinline int add_inode_ref(struct btrfs_trans_handle *trans, | |
1236 | struct btrfs_root *root, | |
1237 | struct btrfs_root *log, | |
1238 | struct btrfs_path *path, | |
1239 | struct extent_buffer *eb, int slot, | |
1240 | struct btrfs_key *key) | |
1241 | { | |
03b2f08b GB |
1242 | struct inode *dir = NULL; |
1243 | struct inode *inode = NULL; | |
5a1d7843 JS |
1244 | unsigned long ref_ptr; |
1245 | unsigned long ref_end; | |
03b2f08b | 1246 | char *name = NULL; |
5a1d7843 JS |
1247 | int namelen; |
1248 | int ret; | |
1249 | int search_done = 0; | |
f186373f MF |
1250 | int log_ref_ver = 0; |
1251 | u64 parent_objectid; | |
1252 | u64 inode_objectid; | |
f46dbe3d | 1253 | u64 ref_index = 0; |
f186373f MF |
1254 | int ref_struct_size; |
1255 | ||
1256 | ref_ptr = btrfs_item_ptr_offset(eb, slot); | |
1257 | ref_end = ref_ptr + btrfs_item_size_nr(eb, slot); | |
1258 | ||
1259 | if (key->type == BTRFS_INODE_EXTREF_KEY) { | |
1260 | struct btrfs_inode_extref *r; | |
1261 | ||
1262 | ref_struct_size = sizeof(struct btrfs_inode_extref); | |
1263 | log_ref_ver = 1; | |
1264 | r = (struct btrfs_inode_extref *)ref_ptr; | |
1265 | parent_objectid = btrfs_inode_extref_parent(eb, r); | |
1266 | } else { | |
1267 | ref_struct_size = sizeof(struct btrfs_inode_ref); | |
1268 | parent_objectid = key->offset; | |
1269 | } | |
1270 | inode_objectid = key->objectid; | |
e02119d5 | 1271 | |
5a1d7843 JS |
1272 | /* |
1273 | * it is possible that we didn't log all the parent directories | |
1274 | * for a given inode. If we don't find the dir, just don't | |
1275 | * copy the back ref in. The link count fixup code will take | |
1276 | * care of the rest | |
1277 | */ | |
f186373f | 1278 | dir = read_one_inode(root, parent_objectid); |
03b2f08b GB |
1279 | if (!dir) { |
1280 | ret = -ENOENT; | |
1281 | goto out; | |
1282 | } | |
5a1d7843 | 1283 | |
f186373f | 1284 | inode = read_one_inode(root, inode_objectid); |
5a1d7843 | 1285 | if (!inode) { |
03b2f08b GB |
1286 | ret = -EIO; |
1287 | goto out; | |
5a1d7843 JS |
1288 | } |
1289 | ||
5a1d7843 | 1290 | while (ref_ptr < ref_end) { |
f186373f | 1291 | if (log_ref_ver) { |
26a836ce SY |
1292 | ret = extref_get_fields(eb, slot, ref_ptr, &namelen, |
1293 | &name, &ref_index, &parent_objectid); | |
f186373f MF |
1294 | /* |
1295 | * parent object can change from one array | |
1296 | * item to another. | |
1297 | */ | |
1298 | if (!dir) | |
1299 | dir = read_one_inode(root, parent_objectid); | |
03b2f08b GB |
1300 | if (!dir) { |
1301 | ret = -ENOENT; | |
1302 | goto out; | |
1303 | } | |
f186373f | 1304 | } else { |
26a836ce SY |
1305 | ret = ref_get_fields(eb, slot, ref_ptr, &namelen, |
1306 | &name, &ref_index); | |
f186373f MF |
1307 | } |
1308 | if (ret) | |
03b2f08b | 1309 | goto out; |
5a1d7843 JS |
1310 | |
1311 | /* if we already have a perfect match, we're done */ | |
f85b7379 DS |
1312 | if (!inode_in_dir(root, path, btrfs_ino(BTRFS_I(dir)), |
1313 | btrfs_ino(BTRFS_I(inode)), ref_index, | |
1314 | name, namelen)) { | |
5a1d7843 JS |
1315 | /* |
1316 | * look for a conflicting back reference in the | |
1317 | * metadata. if we find one we have to unlink that name | |
1318 | * of the file before we add our new link. Later on, we | |
1319 | * overwrite any existing back reference, and we don't | |
1320 | * want to create dangling pointers in the directory. | |
1321 | */ | |
1322 | ||
1323 | if (!search_done) { | |
1324 | ret = __add_inode_ref(trans, root, path, log, | |
94c91a1f | 1325 | BTRFS_I(dir), |
d75eefdf | 1326 | BTRFS_I(inode), |
f186373f MF |
1327 | inode_objectid, |
1328 | parent_objectid, | |
1329 | ref_index, name, namelen, | |
5a1d7843 | 1330 | &search_done); |
03b2f08b GB |
1331 | if (ret) { |
1332 | if (ret == 1) | |
1333 | ret = 0; | |
3650860b JB |
1334 | goto out; |
1335 | } | |
5a1d7843 JS |
1336 | } |
1337 | ||
1338 | /* insert our name */ | |
db0a669f NB |
1339 | ret = btrfs_add_link(trans, BTRFS_I(dir), |
1340 | BTRFS_I(inode), | |
1341 | name, namelen, 0, ref_index); | |
3650860b JB |
1342 | if (ret) |
1343 | goto out; | |
5a1d7843 JS |
1344 | |
1345 | btrfs_update_inode(trans, root, inode); | |
1346 | } | |
1347 | ||
f186373f | 1348 | ref_ptr = (unsigned long)(ref_ptr + ref_struct_size) + namelen; |
5a1d7843 | 1349 | kfree(name); |
03b2f08b | 1350 | name = NULL; |
f186373f MF |
1351 | if (log_ref_ver) { |
1352 | iput(dir); | |
1353 | dir = NULL; | |
1354 | } | |
5a1d7843 | 1355 | } |
e02119d5 CM |
1356 | |
1357 | /* finally write the back reference in the inode */ | |
1358 | ret = overwrite_item(trans, root, path, eb, slot, key); | |
5a1d7843 | 1359 | out: |
b3b4aa74 | 1360 | btrfs_release_path(path); |
03b2f08b | 1361 | kfree(name); |
e02119d5 CM |
1362 | iput(dir); |
1363 | iput(inode); | |
3650860b | 1364 | return ret; |
e02119d5 CM |
1365 | } |
1366 | ||
c71bf099 | 1367 | static int insert_orphan_item(struct btrfs_trans_handle *trans, |
9c4f61f0 | 1368 | struct btrfs_root *root, u64 ino) |
c71bf099 YZ |
1369 | { |
1370 | int ret; | |
381cf658 | 1371 | |
9c4f61f0 DS |
1372 | ret = btrfs_insert_orphan_item(trans, root, ino); |
1373 | if (ret == -EEXIST) | |
1374 | ret = 0; | |
381cf658 | 1375 | |
c71bf099 YZ |
1376 | return ret; |
1377 | } | |
1378 | ||
f186373f | 1379 | static int count_inode_extrefs(struct btrfs_root *root, |
36283658 | 1380 | struct btrfs_inode *inode, struct btrfs_path *path) |
f186373f MF |
1381 | { |
1382 | int ret = 0; | |
1383 | int name_len; | |
1384 | unsigned int nlink = 0; | |
1385 | u32 item_size; | |
1386 | u32 cur_offset = 0; | |
36283658 | 1387 | u64 inode_objectid = btrfs_ino(inode); |
f186373f MF |
1388 | u64 offset = 0; |
1389 | unsigned long ptr; | |
1390 | struct btrfs_inode_extref *extref; | |
1391 | struct extent_buffer *leaf; | |
1392 | ||
1393 | while (1) { | |
1394 | ret = btrfs_find_one_extref(root, inode_objectid, offset, path, | |
1395 | &extref, &offset); | |
1396 | if (ret) | |
1397 | break; | |
c71bf099 | 1398 | |
f186373f MF |
1399 | leaf = path->nodes[0]; |
1400 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1401 | ptr = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
2c2c452b | 1402 | cur_offset = 0; |
f186373f MF |
1403 | |
1404 | while (cur_offset < item_size) { | |
1405 | extref = (struct btrfs_inode_extref *) (ptr + cur_offset); | |
1406 | name_len = btrfs_inode_extref_name_len(leaf, extref); | |
1407 | ||
1408 | nlink++; | |
1409 | ||
1410 | cur_offset += name_len + sizeof(*extref); | |
1411 | } | |
1412 | ||
1413 | offset++; | |
1414 | btrfs_release_path(path); | |
1415 | } | |
1416 | btrfs_release_path(path); | |
1417 | ||
2c2c452b | 1418 | if (ret < 0 && ret != -ENOENT) |
f186373f MF |
1419 | return ret; |
1420 | return nlink; | |
1421 | } | |
1422 | ||
1423 | static int count_inode_refs(struct btrfs_root *root, | |
f329e319 | 1424 | struct btrfs_inode *inode, struct btrfs_path *path) |
e02119d5 | 1425 | { |
e02119d5 CM |
1426 | int ret; |
1427 | struct btrfs_key key; | |
f186373f | 1428 | unsigned int nlink = 0; |
e02119d5 CM |
1429 | unsigned long ptr; |
1430 | unsigned long ptr_end; | |
1431 | int name_len; | |
f329e319 | 1432 | u64 ino = btrfs_ino(inode); |
e02119d5 | 1433 | |
33345d01 | 1434 | key.objectid = ino; |
e02119d5 CM |
1435 | key.type = BTRFS_INODE_REF_KEY; |
1436 | key.offset = (u64)-1; | |
1437 | ||
d397712b | 1438 | while (1) { |
e02119d5 CM |
1439 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
1440 | if (ret < 0) | |
1441 | break; | |
1442 | if (ret > 0) { | |
1443 | if (path->slots[0] == 0) | |
1444 | break; | |
1445 | path->slots[0]--; | |
1446 | } | |
e93ae26f | 1447 | process_slot: |
e02119d5 CM |
1448 | btrfs_item_key_to_cpu(path->nodes[0], &key, |
1449 | path->slots[0]); | |
33345d01 | 1450 | if (key.objectid != ino || |
e02119d5 CM |
1451 | key.type != BTRFS_INODE_REF_KEY) |
1452 | break; | |
1453 | ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]); | |
1454 | ptr_end = ptr + btrfs_item_size_nr(path->nodes[0], | |
1455 | path->slots[0]); | |
d397712b | 1456 | while (ptr < ptr_end) { |
e02119d5 CM |
1457 | struct btrfs_inode_ref *ref; |
1458 | ||
1459 | ref = (struct btrfs_inode_ref *)ptr; | |
1460 | name_len = btrfs_inode_ref_name_len(path->nodes[0], | |
1461 | ref); | |
1462 | ptr = (unsigned long)(ref + 1) + name_len; | |
1463 | nlink++; | |
1464 | } | |
1465 | ||
1466 | if (key.offset == 0) | |
1467 | break; | |
e93ae26f FDBM |
1468 | if (path->slots[0] > 0) { |
1469 | path->slots[0]--; | |
1470 | goto process_slot; | |
1471 | } | |
e02119d5 | 1472 | key.offset--; |
b3b4aa74 | 1473 | btrfs_release_path(path); |
e02119d5 | 1474 | } |
b3b4aa74 | 1475 | btrfs_release_path(path); |
f186373f MF |
1476 | |
1477 | return nlink; | |
1478 | } | |
1479 | ||
1480 | /* | |
1481 | * There are a few corners where the link count of the file can't | |
1482 | * be properly maintained during replay. So, instead of adding | |
1483 | * lots of complexity to the log code, we just scan the backrefs | |
1484 | * for any file that has been through replay. | |
1485 | * | |
1486 | * The scan will update the link count on the inode to reflect the | |
1487 | * number of back refs found. If it goes down to zero, the iput | |
1488 | * will free the inode. | |
1489 | */ | |
1490 | static noinline int fixup_inode_link_count(struct btrfs_trans_handle *trans, | |
1491 | struct btrfs_root *root, | |
1492 | struct inode *inode) | |
1493 | { | |
1494 | struct btrfs_path *path; | |
1495 | int ret; | |
1496 | u64 nlink = 0; | |
4a0cc7ca | 1497 | u64 ino = btrfs_ino(BTRFS_I(inode)); |
f186373f MF |
1498 | |
1499 | path = btrfs_alloc_path(); | |
1500 | if (!path) | |
1501 | return -ENOMEM; | |
1502 | ||
f329e319 | 1503 | ret = count_inode_refs(root, BTRFS_I(inode), path); |
f186373f MF |
1504 | if (ret < 0) |
1505 | goto out; | |
1506 | ||
1507 | nlink = ret; | |
1508 | ||
36283658 | 1509 | ret = count_inode_extrefs(root, BTRFS_I(inode), path); |
f186373f MF |
1510 | if (ret < 0) |
1511 | goto out; | |
1512 | ||
1513 | nlink += ret; | |
1514 | ||
1515 | ret = 0; | |
1516 | ||
e02119d5 | 1517 | if (nlink != inode->i_nlink) { |
bfe86848 | 1518 | set_nlink(inode, nlink); |
e02119d5 CM |
1519 | btrfs_update_inode(trans, root, inode); |
1520 | } | |
8d5bf1cb | 1521 | BTRFS_I(inode)->index_cnt = (u64)-1; |
e02119d5 | 1522 | |
c71bf099 YZ |
1523 | if (inode->i_nlink == 0) { |
1524 | if (S_ISDIR(inode->i_mode)) { | |
1525 | ret = replay_dir_deletes(trans, root, NULL, path, | |
33345d01 | 1526 | ino, 1); |
3650860b JB |
1527 | if (ret) |
1528 | goto out; | |
c71bf099 | 1529 | } |
33345d01 | 1530 | ret = insert_orphan_item(trans, root, ino); |
12fcfd22 | 1531 | } |
12fcfd22 | 1532 | |
f186373f MF |
1533 | out: |
1534 | btrfs_free_path(path); | |
1535 | return ret; | |
e02119d5 CM |
1536 | } |
1537 | ||
1538 | static noinline int fixup_inode_link_counts(struct btrfs_trans_handle *trans, | |
1539 | struct btrfs_root *root, | |
1540 | struct btrfs_path *path) | |
1541 | { | |
1542 | int ret; | |
1543 | struct btrfs_key key; | |
1544 | struct inode *inode; | |
1545 | ||
1546 | key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID; | |
1547 | key.type = BTRFS_ORPHAN_ITEM_KEY; | |
1548 | key.offset = (u64)-1; | |
d397712b | 1549 | while (1) { |
e02119d5 CM |
1550 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
1551 | if (ret < 0) | |
1552 | break; | |
1553 | ||
1554 | if (ret == 1) { | |
1555 | if (path->slots[0] == 0) | |
1556 | break; | |
1557 | path->slots[0]--; | |
1558 | } | |
1559 | ||
1560 | btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]); | |
1561 | if (key.objectid != BTRFS_TREE_LOG_FIXUP_OBJECTID || | |
1562 | key.type != BTRFS_ORPHAN_ITEM_KEY) | |
1563 | break; | |
1564 | ||
1565 | ret = btrfs_del_item(trans, root, path); | |
65a246c5 TI |
1566 | if (ret) |
1567 | goto out; | |
e02119d5 | 1568 | |
b3b4aa74 | 1569 | btrfs_release_path(path); |
e02119d5 | 1570 | inode = read_one_inode(root, key.offset); |
c00e9493 TI |
1571 | if (!inode) |
1572 | return -EIO; | |
e02119d5 CM |
1573 | |
1574 | ret = fixup_inode_link_count(trans, root, inode); | |
e02119d5 | 1575 | iput(inode); |
3650860b JB |
1576 | if (ret) |
1577 | goto out; | |
e02119d5 | 1578 | |
12fcfd22 CM |
1579 | /* |
1580 | * fixup on a directory may create new entries, | |
1581 | * make sure we always look for the highset possible | |
1582 | * offset | |
1583 | */ | |
1584 | key.offset = (u64)-1; | |
e02119d5 | 1585 | } |
65a246c5 TI |
1586 | ret = 0; |
1587 | out: | |
b3b4aa74 | 1588 | btrfs_release_path(path); |
65a246c5 | 1589 | return ret; |
e02119d5 CM |
1590 | } |
1591 | ||
1592 | ||
1593 | /* | |
1594 | * record a given inode in the fixup dir so we can check its link | |
1595 | * count when replay is done. The link count is incremented here | |
1596 | * so the inode won't go away until we check it | |
1597 | */ | |
1598 | static noinline int link_to_fixup_dir(struct btrfs_trans_handle *trans, | |
1599 | struct btrfs_root *root, | |
1600 | struct btrfs_path *path, | |
1601 | u64 objectid) | |
1602 | { | |
1603 | struct btrfs_key key; | |
1604 | int ret = 0; | |
1605 | struct inode *inode; | |
1606 | ||
1607 | inode = read_one_inode(root, objectid); | |
c00e9493 TI |
1608 | if (!inode) |
1609 | return -EIO; | |
e02119d5 CM |
1610 | |
1611 | key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID; | |
962a298f | 1612 | key.type = BTRFS_ORPHAN_ITEM_KEY; |
e02119d5 CM |
1613 | key.offset = objectid; |
1614 | ||
1615 | ret = btrfs_insert_empty_item(trans, root, path, &key, 0); | |
1616 | ||
b3b4aa74 | 1617 | btrfs_release_path(path); |
e02119d5 | 1618 | if (ret == 0) { |
9bf7a489 JB |
1619 | if (!inode->i_nlink) |
1620 | set_nlink(inode, 1); | |
1621 | else | |
8b558c5f | 1622 | inc_nlink(inode); |
b9959295 | 1623 | ret = btrfs_update_inode(trans, root, inode); |
e02119d5 CM |
1624 | } else if (ret == -EEXIST) { |
1625 | ret = 0; | |
1626 | } else { | |
3650860b | 1627 | BUG(); /* Logic Error */ |
e02119d5 CM |
1628 | } |
1629 | iput(inode); | |
1630 | ||
1631 | return ret; | |
1632 | } | |
1633 | ||
1634 | /* | |
1635 | * when replaying the log for a directory, we only insert names | |
1636 | * for inodes that actually exist. This means an fsync on a directory | |
1637 | * does not implicitly fsync all the new files in it | |
1638 | */ | |
1639 | static noinline int insert_one_name(struct btrfs_trans_handle *trans, | |
1640 | struct btrfs_root *root, | |
e02119d5 | 1641 | u64 dirid, u64 index, |
60d53eb3 | 1642 | char *name, int name_len, |
e02119d5 CM |
1643 | struct btrfs_key *location) |
1644 | { | |
1645 | struct inode *inode; | |
1646 | struct inode *dir; | |
1647 | int ret; | |
1648 | ||
1649 | inode = read_one_inode(root, location->objectid); | |
1650 | if (!inode) | |
1651 | return -ENOENT; | |
1652 | ||
1653 | dir = read_one_inode(root, dirid); | |
1654 | if (!dir) { | |
1655 | iput(inode); | |
1656 | return -EIO; | |
1657 | } | |
d555438b | 1658 | |
db0a669f NB |
1659 | ret = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode), name, |
1660 | name_len, 1, index); | |
e02119d5 CM |
1661 | |
1662 | /* FIXME, put inode into FIXUP list */ | |
1663 | ||
1664 | iput(inode); | |
1665 | iput(dir); | |
1666 | return ret; | |
1667 | } | |
1668 | ||
df8d116f FM |
1669 | /* |
1670 | * Return true if an inode reference exists in the log for the given name, | |
1671 | * inode and parent inode. | |
1672 | */ | |
1673 | static bool name_in_log_ref(struct btrfs_root *log_root, | |
1674 | const char *name, const int name_len, | |
1675 | const u64 dirid, const u64 ino) | |
1676 | { | |
1677 | struct btrfs_key search_key; | |
1678 | ||
1679 | search_key.objectid = ino; | |
1680 | search_key.type = BTRFS_INODE_REF_KEY; | |
1681 | search_key.offset = dirid; | |
1682 | if (backref_in_log(log_root, &search_key, dirid, name, name_len)) | |
1683 | return true; | |
1684 | ||
1685 | search_key.type = BTRFS_INODE_EXTREF_KEY; | |
1686 | search_key.offset = btrfs_extref_hash(dirid, name, name_len); | |
1687 | if (backref_in_log(log_root, &search_key, dirid, name, name_len)) | |
1688 | return true; | |
1689 | ||
1690 | return false; | |
1691 | } | |
1692 | ||
e02119d5 CM |
1693 | /* |
1694 | * take a single entry in a log directory item and replay it into | |
1695 | * the subvolume. | |
1696 | * | |
1697 | * if a conflicting item exists in the subdirectory already, | |
1698 | * the inode it points to is unlinked and put into the link count | |
1699 | * fix up tree. | |
1700 | * | |
1701 | * If a name from the log points to a file or directory that does | |
1702 | * not exist in the FS, it is skipped. fsyncs on directories | |
1703 | * do not force down inodes inside that directory, just changes to the | |
1704 | * names or unlinks in a directory. | |
bb53eda9 FM |
1705 | * |
1706 | * Returns < 0 on error, 0 if the name wasn't replayed (dentry points to a | |
1707 | * non-existing inode) and 1 if the name was replayed. | |
e02119d5 CM |
1708 | */ |
1709 | static noinline int replay_one_name(struct btrfs_trans_handle *trans, | |
1710 | struct btrfs_root *root, | |
1711 | struct btrfs_path *path, | |
1712 | struct extent_buffer *eb, | |
1713 | struct btrfs_dir_item *di, | |
1714 | struct btrfs_key *key) | |
1715 | { | |
1716 | char *name; | |
1717 | int name_len; | |
1718 | struct btrfs_dir_item *dst_di; | |
1719 | struct btrfs_key found_key; | |
1720 | struct btrfs_key log_key; | |
1721 | struct inode *dir; | |
e02119d5 | 1722 | u8 log_type; |
4bef0848 | 1723 | int exists; |
3650860b | 1724 | int ret = 0; |
d555438b | 1725 | bool update_size = (key->type == BTRFS_DIR_INDEX_KEY); |
bb53eda9 | 1726 | bool name_added = false; |
e02119d5 CM |
1727 | |
1728 | dir = read_one_inode(root, key->objectid); | |
c00e9493 TI |
1729 | if (!dir) |
1730 | return -EIO; | |
e02119d5 CM |
1731 | |
1732 | name_len = btrfs_dir_name_len(eb, di); | |
1733 | name = kmalloc(name_len, GFP_NOFS); | |
2bac325e FDBM |
1734 | if (!name) { |
1735 | ret = -ENOMEM; | |
1736 | goto out; | |
1737 | } | |
2a29edc6 | 1738 | |
e02119d5 CM |
1739 | log_type = btrfs_dir_type(eb, di); |
1740 | read_extent_buffer(eb, name, (unsigned long)(di + 1), | |
1741 | name_len); | |
1742 | ||
1743 | btrfs_dir_item_key_to_cpu(eb, di, &log_key); | |
4bef0848 CM |
1744 | exists = btrfs_lookup_inode(trans, root, path, &log_key, 0); |
1745 | if (exists == 0) | |
1746 | exists = 1; | |
1747 | else | |
1748 | exists = 0; | |
b3b4aa74 | 1749 | btrfs_release_path(path); |
4bef0848 | 1750 | |
e02119d5 CM |
1751 | if (key->type == BTRFS_DIR_ITEM_KEY) { |
1752 | dst_di = btrfs_lookup_dir_item(trans, root, path, key->objectid, | |
1753 | name, name_len, 1); | |
d397712b | 1754 | } else if (key->type == BTRFS_DIR_INDEX_KEY) { |
e02119d5 CM |
1755 | dst_di = btrfs_lookup_dir_index_item(trans, root, path, |
1756 | key->objectid, | |
1757 | key->offset, name, | |
1758 | name_len, 1); | |
1759 | } else { | |
3650860b JB |
1760 | /* Corruption */ |
1761 | ret = -EINVAL; | |
1762 | goto out; | |
e02119d5 | 1763 | } |
c704005d | 1764 | if (IS_ERR_OR_NULL(dst_di)) { |
e02119d5 CM |
1765 | /* we need a sequence number to insert, so we only |
1766 | * do inserts for the BTRFS_DIR_INDEX_KEY types | |
1767 | */ | |
1768 | if (key->type != BTRFS_DIR_INDEX_KEY) | |
1769 | goto out; | |
1770 | goto insert; | |
1771 | } | |
1772 | ||
1773 | btrfs_dir_item_key_to_cpu(path->nodes[0], dst_di, &found_key); | |
1774 | /* the existing item matches the logged item */ | |
1775 | if (found_key.objectid == log_key.objectid && | |
1776 | found_key.type == log_key.type && | |
1777 | found_key.offset == log_key.offset && | |
1778 | btrfs_dir_type(path->nodes[0], dst_di) == log_type) { | |
a2cc11db | 1779 | update_size = false; |
e02119d5 CM |
1780 | goto out; |
1781 | } | |
1782 | ||
1783 | /* | |
1784 | * don't drop the conflicting directory entry if the inode | |
1785 | * for the new entry doesn't exist | |
1786 | */ | |
4bef0848 | 1787 | if (!exists) |
e02119d5 CM |
1788 | goto out; |
1789 | ||
207e7d92 | 1790 | ret = drop_one_dir_item(trans, root, path, BTRFS_I(dir), dst_di); |
3650860b JB |
1791 | if (ret) |
1792 | goto out; | |
e02119d5 CM |
1793 | |
1794 | if (key->type == BTRFS_DIR_INDEX_KEY) | |
1795 | goto insert; | |
1796 | out: | |
b3b4aa74 | 1797 | btrfs_release_path(path); |
d555438b | 1798 | if (!ret && update_size) { |
6ef06d27 | 1799 | btrfs_i_size_write(BTRFS_I(dir), dir->i_size + name_len * 2); |
d555438b JB |
1800 | ret = btrfs_update_inode(trans, root, dir); |
1801 | } | |
e02119d5 CM |
1802 | kfree(name); |
1803 | iput(dir); | |
bb53eda9 FM |
1804 | if (!ret && name_added) |
1805 | ret = 1; | |
3650860b | 1806 | return ret; |
e02119d5 CM |
1807 | |
1808 | insert: | |
df8d116f FM |
1809 | if (name_in_log_ref(root->log_root, name, name_len, |
1810 | key->objectid, log_key.objectid)) { | |
1811 | /* The dentry will be added later. */ | |
1812 | ret = 0; | |
1813 | update_size = false; | |
1814 | goto out; | |
1815 | } | |
b3b4aa74 | 1816 | btrfs_release_path(path); |
60d53eb3 Z |
1817 | ret = insert_one_name(trans, root, key->objectid, key->offset, |
1818 | name, name_len, &log_key); | |
df8d116f | 1819 | if (ret && ret != -ENOENT && ret != -EEXIST) |
3650860b | 1820 | goto out; |
bb53eda9 FM |
1821 | if (!ret) |
1822 | name_added = true; | |
d555438b | 1823 | update_size = false; |
3650860b | 1824 | ret = 0; |
e02119d5 CM |
1825 | goto out; |
1826 | } | |
1827 | ||
1828 | /* | |
1829 | * find all the names in a directory item and reconcile them into | |
1830 | * the subvolume. Only BTRFS_DIR_ITEM_KEY types will have more than | |
1831 | * one name in a directory item, but the same code gets used for | |
1832 | * both directory index types | |
1833 | */ | |
1834 | static noinline int replay_one_dir_item(struct btrfs_trans_handle *trans, | |
1835 | struct btrfs_root *root, | |
1836 | struct btrfs_path *path, | |
1837 | struct extent_buffer *eb, int slot, | |
1838 | struct btrfs_key *key) | |
1839 | { | |
2ff7e61e | 1840 | struct btrfs_fs_info *fs_info = root->fs_info; |
bb53eda9 | 1841 | int ret = 0; |
e02119d5 CM |
1842 | u32 item_size = btrfs_item_size_nr(eb, slot); |
1843 | struct btrfs_dir_item *di; | |
1844 | int name_len; | |
1845 | unsigned long ptr; | |
1846 | unsigned long ptr_end; | |
bb53eda9 | 1847 | struct btrfs_path *fixup_path = NULL; |
e02119d5 CM |
1848 | |
1849 | ptr = btrfs_item_ptr_offset(eb, slot); | |
1850 | ptr_end = ptr + item_size; | |
d397712b | 1851 | while (ptr < ptr_end) { |
e02119d5 | 1852 | di = (struct btrfs_dir_item *)ptr; |
e79a3327 | 1853 | if (verify_dir_item(fs_info, eb, slot, di)) |
22a94d44 | 1854 | return -EIO; |
e02119d5 CM |
1855 | name_len = btrfs_dir_name_len(eb, di); |
1856 | ret = replay_one_name(trans, root, path, eb, di, key); | |
bb53eda9 FM |
1857 | if (ret < 0) |
1858 | break; | |
e02119d5 CM |
1859 | ptr = (unsigned long)(di + 1); |
1860 | ptr += name_len; | |
bb53eda9 FM |
1861 | |
1862 | /* | |
1863 | * If this entry refers to a non-directory (directories can not | |
1864 | * have a link count > 1) and it was added in the transaction | |
1865 | * that was not committed, make sure we fixup the link count of | |
1866 | * the inode it the entry points to. Otherwise something like | |
1867 | * the following would result in a directory pointing to an | |
1868 | * inode with a wrong link that does not account for this dir | |
1869 | * entry: | |
1870 | * | |
1871 | * mkdir testdir | |
1872 | * touch testdir/foo | |
1873 | * touch testdir/bar | |
1874 | * sync | |
1875 | * | |
1876 | * ln testdir/bar testdir/bar_link | |
1877 | * ln testdir/foo testdir/foo_link | |
1878 | * xfs_io -c "fsync" testdir/bar | |
1879 | * | |
1880 | * <power failure> | |
1881 | * | |
1882 | * mount fs, log replay happens | |
1883 | * | |
1884 | * File foo would remain with a link count of 1 when it has two | |
1885 | * entries pointing to it in the directory testdir. This would | |
1886 | * make it impossible to ever delete the parent directory has | |
1887 | * it would result in stale dentries that can never be deleted. | |
1888 | */ | |
1889 | if (ret == 1 && btrfs_dir_type(eb, di) != BTRFS_FT_DIR) { | |
1890 | struct btrfs_key di_key; | |
1891 | ||
1892 | if (!fixup_path) { | |
1893 | fixup_path = btrfs_alloc_path(); | |
1894 | if (!fixup_path) { | |
1895 | ret = -ENOMEM; | |
1896 | break; | |
1897 | } | |
1898 | } | |
1899 | ||
1900 | btrfs_dir_item_key_to_cpu(eb, di, &di_key); | |
1901 | ret = link_to_fixup_dir(trans, root, fixup_path, | |
1902 | di_key.objectid); | |
1903 | if (ret) | |
1904 | break; | |
1905 | } | |
1906 | ret = 0; | |
e02119d5 | 1907 | } |
bb53eda9 FM |
1908 | btrfs_free_path(fixup_path); |
1909 | return ret; | |
e02119d5 CM |
1910 | } |
1911 | ||
1912 | /* | |
1913 | * directory replay has two parts. There are the standard directory | |
1914 | * items in the log copied from the subvolume, and range items | |
1915 | * created in the log while the subvolume was logged. | |
1916 | * | |
1917 | * The range items tell us which parts of the key space the log | |
1918 | * is authoritative for. During replay, if a key in the subvolume | |
1919 | * directory is in a logged range item, but not actually in the log | |
1920 | * that means it was deleted from the directory before the fsync | |
1921 | * and should be removed. | |
1922 | */ | |
1923 | static noinline int find_dir_range(struct btrfs_root *root, | |
1924 | struct btrfs_path *path, | |
1925 | u64 dirid, int key_type, | |
1926 | u64 *start_ret, u64 *end_ret) | |
1927 | { | |
1928 | struct btrfs_key key; | |
1929 | u64 found_end; | |
1930 | struct btrfs_dir_log_item *item; | |
1931 | int ret; | |
1932 | int nritems; | |
1933 | ||
1934 | if (*start_ret == (u64)-1) | |
1935 | return 1; | |
1936 | ||
1937 | key.objectid = dirid; | |
1938 | key.type = key_type; | |
1939 | key.offset = *start_ret; | |
1940 | ||
1941 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
1942 | if (ret < 0) | |
1943 | goto out; | |
1944 | if (ret > 0) { | |
1945 | if (path->slots[0] == 0) | |
1946 | goto out; | |
1947 | path->slots[0]--; | |
1948 | } | |
1949 | if (ret != 0) | |
1950 | btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]); | |
1951 | ||
1952 | if (key.type != key_type || key.objectid != dirid) { | |
1953 | ret = 1; | |
1954 | goto next; | |
1955 | } | |
1956 | item = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
1957 | struct btrfs_dir_log_item); | |
1958 | found_end = btrfs_dir_log_end(path->nodes[0], item); | |
1959 | ||
1960 | if (*start_ret >= key.offset && *start_ret <= found_end) { | |
1961 | ret = 0; | |
1962 | *start_ret = key.offset; | |
1963 | *end_ret = found_end; | |
1964 | goto out; | |
1965 | } | |
1966 | ret = 1; | |
1967 | next: | |
1968 | /* check the next slot in the tree to see if it is a valid item */ | |
1969 | nritems = btrfs_header_nritems(path->nodes[0]); | |
2a7bf53f | 1970 | path->slots[0]++; |
e02119d5 CM |
1971 | if (path->slots[0] >= nritems) { |
1972 | ret = btrfs_next_leaf(root, path); | |
1973 | if (ret) | |
1974 | goto out; | |
e02119d5 CM |
1975 | } |
1976 | ||
1977 | btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]); | |
1978 | ||
1979 | if (key.type != key_type || key.objectid != dirid) { | |
1980 | ret = 1; | |
1981 | goto out; | |
1982 | } | |
1983 | item = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
1984 | struct btrfs_dir_log_item); | |
1985 | found_end = btrfs_dir_log_end(path->nodes[0], item); | |
1986 | *start_ret = key.offset; | |
1987 | *end_ret = found_end; | |
1988 | ret = 0; | |
1989 | out: | |
b3b4aa74 | 1990 | btrfs_release_path(path); |
e02119d5 CM |
1991 | return ret; |
1992 | } | |
1993 | ||
1994 | /* | |
1995 | * this looks for a given directory item in the log. If the directory | |
1996 | * item is not in the log, the item is removed and the inode it points | |
1997 | * to is unlinked | |
1998 | */ | |
1999 | static noinline int check_item_in_log(struct btrfs_trans_handle *trans, | |
2000 | struct btrfs_root *root, | |
2001 | struct btrfs_root *log, | |
2002 | struct btrfs_path *path, | |
2003 | struct btrfs_path *log_path, | |
2004 | struct inode *dir, | |
2005 | struct btrfs_key *dir_key) | |
2006 | { | |
2ff7e61e | 2007 | struct btrfs_fs_info *fs_info = root->fs_info; |
e02119d5 CM |
2008 | int ret; |
2009 | struct extent_buffer *eb; | |
2010 | int slot; | |
2011 | u32 item_size; | |
2012 | struct btrfs_dir_item *di; | |
2013 | struct btrfs_dir_item *log_di; | |
2014 | int name_len; | |
2015 | unsigned long ptr; | |
2016 | unsigned long ptr_end; | |
2017 | char *name; | |
2018 | struct inode *inode; | |
2019 | struct btrfs_key location; | |
2020 | ||
2021 | again: | |
2022 | eb = path->nodes[0]; | |
2023 | slot = path->slots[0]; | |
2024 | item_size = btrfs_item_size_nr(eb, slot); | |
2025 | ptr = btrfs_item_ptr_offset(eb, slot); | |
2026 | ptr_end = ptr + item_size; | |
d397712b | 2027 | while (ptr < ptr_end) { |
e02119d5 | 2028 | di = (struct btrfs_dir_item *)ptr; |
e79a3327 | 2029 | if (verify_dir_item(fs_info, eb, slot, di)) { |
22a94d44 JB |
2030 | ret = -EIO; |
2031 | goto out; | |
2032 | } | |
2033 | ||
e02119d5 CM |
2034 | name_len = btrfs_dir_name_len(eb, di); |
2035 | name = kmalloc(name_len, GFP_NOFS); | |
2036 | if (!name) { | |
2037 | ret = -ENOMEM; | |
2038 | goto out; | |
2039 | } | |
2040 | read_extent_buffer(eb, name, (unsigned long)(di + 1), | |
2041 | name_len); | |
2042 | log_di = NULL; | |
12fcfd22 | 2043 | if (log && dir_key->type == BTRFS_DIR_ITEM_KEY) { |
e02119d5 CM |
2044 | log_di = btrfs_lookup_dir_item(trans, log, log_path, |
2045 | dir_key->objectid, | |
2046 | name, name_len, 0); | |
12fcfd22 | 2047 | } else if (log && dir_key->type == BTRFS_DIR_INDEX_KEY) { |
e02119d5 CM |
2048 | log_di = btrfs_lookup_dir_index_item(trans, log, |
2049 | log_path, | |
2050 | dir_key->objectid, | |
2051 | dir_key->offset, | |
2052 | name, name_len, 0); | |
2053 | } | |
269d040f | 2054 | if (!log_di || (IS_ERR(log_di) && PTR_ERR(log_di) == -ENOENT)) { |
e02119d5 | 2055 | btrfs_dir_item_key_to_cpu(eb, di, &location); |
b3b4aa74 DS |
2056 | btrfs_release_path(path); |
2057 | btrfs_release_path(log_path); | |
e02119d5 | 2058 | inode = read_one_inode(root, location.objectid); |
c00e9493 TI |
2059 | if (!inode) { |
2060 | kfree(name); | |
2061 | return -EIO; | |
2062 | } | |
e02119d5 CM |
2063 | |
2064 | ret = link_to_fixup_dir(trans, root, | |
2065 | path, location.objectid); | |
3650860b JB |
2066 | if (ret) { |
2067 | kfree(name); | |
2068 | iput(inode); | |
2069 | goto out; | |
2070 | } | |
2071 | ||
8b558c5f | 2072 | inc_nlink(inode); |
4ec5934e NB |
2073 | ret = btrfs_unlink_inode(trans, root, BTRFS_I(dir), |
2074 | BTRFS_I(inode), name, name_len); | |
3650860b | 2075 | if (!ret) |
2ff7e61e | 2076 | ret = btrfs_run_delayed_items(trans, fs_info); |
e02119d5 CM |
2077 | kfree(name); |
2078 | iput(inode); | |
3650860b JB |
2079 | if (ret) |
2080 | goto out; | |
e02119d5 CM |
2081 | |
2082 | /* there might still be more names under this key | |
2083 | * check and repeat if required | |
2084 | */ | |
2085 | ret = btrfs_search_slot(NULL, root, dir_key, path, | |
2086 | 0, 0); | |
2087 | if (ret == 0) | |
2088 | goto again; | |
2089 | ret = 0; | |
2090 | goto out; | |
269d040f FDBM |
2091 | } else if (IS_ERR(log_di)) { |
2092 | kfree(name); | |
2093 | return PTR_ERR(log_di); | |
e02119d5 | 2094 | } |
b3b4aa74 | 2095 | btrfs_release_path(log_path); |
e02119d5 CM |
2096 | kfree(name); |
2097 | ||
2098 | ptr = (unsigned long)(di + 1); | |
2099 | ptr += name_len; | |
2100 | } | |
2101 | ret = 0; | |
2102 | out: | |
b3b4aa74 DS |
2103 | btrfs_release_path(path); |
2104 | btrfs_release_path(log_path); | |
e02119d5 CM |
2105 | return ret; |
2106 | } | |
2107 | ||
4f764e51 FM |
2108 | static int replay_xattr_deletes(struct btrfs_trans_handle *trans, |
2109 | struct btrfs_root *root, | |
2110 | struct btrfs_root *log, | |
2111 | struct btrfs_path *path, | |
2112 | const u64 ino) | |
2113 | { | |
8ee8c2d6 | 2114 | struct btrfs_fs_info *fs_info = root->fs_info; |
4f764e51 FM |
2115 | struct btrfs_key search_key; |
2116 | struct btrfs_path *log_path; | |
2117 | int i; | |
2118 | int nritems; | |
2119 | int ret; | |
2120 | ||
2121 | log_path = btrfs_alloc_path(); | |
2122 | if (!log_path) | |
2123 | return -ENOMEM; | |
2124 | ||
2125 | search_key.objectid = ino; | |
2126 | search_key.type = BTRFS_XATTR_ITEM_KEY; | |
2127 | search_key.offset = 0; | |
2128 | again: | |
2129 | ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0); | |
2130 | if (ret < 0) | |
2131 | goto out; | |
2132 | process_leaf: | |
2133 | nritems = btrfs_header_nritems(path->nodes[0]); | |
2134 | for (i = path->slots[0]; i < nritems; i++) { | |
2135 | struct btrfs_key key; | |
2136 | struct btrfs_dir_item *di; | |
2137 | struct btrfs_dir_item *log_di; | |
2138 | u32 total_size; | |
2139 | u32 cur; | |
2140 | ||
2141 | btrfs_item_key_to_cpu(path->nodes[0], &key, i); | |
2142 | if (key.objectid != ino || key.type != BTRFS_XATTR_ITEM_KEY) { | |
2143 | ret = 0; | |
2144 | goto out; | |
2145 | } | |
2146 | ||
2147 | di = btrfs_item_ptr(path->nodes[0], i, struct btrfs_dir_item); | |
2148 | total_size = btrfs_item_size_nr(path->nodes[0], i); | |
2149 | cur = 0; | |
2150 | while (cur < total_size) { | |
2151 | u16 name_len = btrfs_dir_name_len(path->nodes[0], di); | |
2152 | u16 data_len = btrfs_dir_data_len(path->nodes[0], di); | |
2153 | u32 this_len = sizeof(*di) + name_len + data_len; | |
2154 | char *name; | |
2155 | ||
e33bf723 | 2156 | ret = verify_dir_item(fs_info, path->nodes[0], i, di); |
8ee8c2d6 SY |
2157 | if (ret) { |
2158 | ret = -EIO; | |
2159 | goto out; | |
2160 | } | |
4f764e51 FM |
2161 | name = kmalloc(name_len, GFP_NOFS); |
2162 | if (!name) { | |
2163 | ret = -ENOMEM; | |
2164 | goto out; | |
2165 | } | |
2166 | read_extent_buffer(path->nodes[0], name, | |
2167 | (unsigned long)(di + 1), name_len); | |
2168 | ||
2169 | log_di = btrfs_lookup_xattr(NULL, log, log_path, ino, | |
2170 | name, name_len, 0); | |
2171 | btrfs_release_path(log_path); | |
2172 | if (!log_di) { | |
2173 | /* Doesn't exist in log tree, so delete it. */ | |
2174 | btrfs_release_path(path); | |
2175 | di = btrfs_lookup_xattr(trans, root, path, ino, | |
2176 | name, name_len, -1); | |
2177 | kfree(name); | |
2178 | if (IS_ERR(di)) { | |
2179 | ret = PTR_ERR(di); | |
2180 | goto out; | |
2181 | } | |
2182 | ASSERT(di); | |
2183 | ret = btrfs_delete_one_dir_name(trans, root, | |
2184 | path, di); | |
2185 | if (ret) | |
2186 | goto out; | |
2187 | btrfs_release_path(path); | |
2188 | search_key = key; | |
2189 | goto again; | |
2190 | } | |
2191 | kfree(name); | |
2192 | if (IS_ERR(log_di)) { | |
2193 | ret = PTR_ERR(log_di); | |
2194 | goto out; | |
2195 | } | |
2196 | cur += this_len; | |
2197 | di = (struct btrfs_dir_item *)((char *)di + this_len); | |
2198 | } | |
2199 | } | |
2200 | ret = btrfs_next_leaf(root, path); | |
2201 | if (ret > 0) | |
2202 | ret = 0; | |
2203 | else if (ret == 0) | |
2204 | goto process_leaf; | |
2205 | out: | |
2206 | btrfs_free_path(log_path); | |
2207 | btrfs_release_path(path); | |
2208 | return ret; | |
2209 | } | |
2210 | ||
2211 | ||
e02119d5 CM |
2212 | /* |
2213 | * deletion replay happens before we copy any new directory items | |
2214 | * out of the log or out of backreferences from inodes. It | |
2215 | * scans the log to find ranges of keys that log is authoritative for, | |
2216 | * and then scans the directory to find items in those ranges that are | |
2217 | * not present in the log. | |
2218 | * | |
2219 | * Anything we don't find in the log is unlinked and removed from the | |
2220 | * directory. | |
2221 | */ | |
2222 | static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans, | |
2223 | struct btrfs_root *root, | |
2224 | struct btrfs_root *log, | |
2225 | struct btrfs_path *path, | |
12fcfd22 | 2226 | u64 dirid, int del_all) |
e02119d5 CM |
2227 | { |
2228 | u64 range_start; | |
2229 | u64 range_end; | |
2230 | int key_type = BTRFS_DIR_LOG_ITEM_KEY; | |
2231 | int ret = 0; | |
2232 | struct btrfs_key dir_key; | |
2233 | struct btrfs_key found_key; | |
2234 | struct btrfs_path *log_path; | |
2235 | struct inode *dir; | |
2236 | ||
2237 | dir_key.objectid = dirid; | |
2238 | dir_key.type = BTRFS_DIR_ITEM_KEY; | |
2239 | log_path = btrfs_alloc_path(); | |
2240 | if (!log_path) | |
2241 | return -ENOMEM; | |
2242 | ||
2243 | dir = read_one_inode(root, dirid); | |
2244 | /* it isn't an error if the inode isn't there, that can happen | |
2245 | * because we replay the deletes before we copy in the inode item | |
2246 | * from the log | |
2247 | */ | |
2248 | if (!dir) { | |
2249 | btrfs_free_path(log_path); | |
2250 | return 0; | |
2251 | } | |
2252 | again: | |
2253 | range_start = 0; | |
2254 | range_end = 0; | |
d397712b | 2255 | while (1) { |
12fcfd22 CM |
2256 | if (del_all) |
2257 | range_end = (u64)-1; | |
2258 | else { | |
2259 | ret = find_dir_range(log, path, dirid, key_type, | |
2260 | &range_start, &range_end); | |
2261 | if (ret != 0) | |
2262 | break; | |
2263 | } | |
e02119d5 CM |
2264 | |
2265 | dir_key.offset = range_start; | |
d397712b | 2266 | while (1) { |
e02119d5 CM |
2267 | int nritems; |
2268 | ret = btrfs_search_slot(NULL, root, &dir_key, path, | |
2269 | 0, 0); | |
2270 | if (ret < 0) | |
2271 | goto out; | |
2272 | ||
2273 | nritems = btrfs_header_nritems(path->nodes[0]); | |
2274 | if (path->slots[0] >= nritems) { | |
2275 | ret = btrfs_next_leaf(root, path); | |
2276 | if (ret) | |
2277 | break; | |
2278 | } | |
2279 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, | |
2280 | path->slots[0]); | |
2281 | if (found_key.objectid != dirid || | |
2282 | found_key.type != dir_key.type) | |
2283 | goto next_type; | |
2284 | ||
2285 | if (found_key.offset > range_end) | |
2286 | break; | |
2287 | ||
2288 | ret = check_item_in_log(trans, root, log, path, | |
12fcfd22 CM |
2289 | log_path, dir, |
2290 | &found_key); | |
3650860b JB |
2291 | if (ret) |
2292 | goto out; | |
e02119d5 CM |
2293 | if (found_key.offset == (u64)-1) |
2294 | break; | |
2295 | dir_key.offset = found_key.offset + 1; | |
2296 | } | |
b3b4aa74 | 2297 | btrfs_release_path(path); |
e02119d5 CM |
2298 | if (range_end == (u64)-1) |
2299 | break; | |
2300 | range_start = range_end + 1; | |
2301 | } | |
2302 | ||
2303 | next_type: | |
2304 | ret = 0; | |
2305 | if (key_type == BTRFS_DIR_LOG_ITEM_KEY) { | |
2306 | key_type = BTRFS_DIR_LOG_INDEX_KEY; | |
2307 | dir_key.type = BTRFS_DIR_INDEX_KEY; | |
b3b4aa74 | 2308 | btrfs_release_path(path); |
e02119d5 CM |
2309 | goto again; |
2310 | } | |
2311 | out: | |
b3b4aa74 | 2312 | btrfs_release_path(path); |
e02119d5 CM |
2313 | btrfs_free_path(log_path); |
2314 | iput(dir); | |
2315 | return ret; | |
2316 | } | |
2317 | ||
2318 | /* | |
2319 | * the process_func used to replay items from the log tree. This | |
2320 | * gets called in two different stages. The first stage just looks | |
2321 | * for inodes and makes sure they are all copied into the subvolume. | |
2322 | * | |
2323 | * The second stage copies all the other item types from the log into | |
2324 | * the subvolume. The two stage approach is slower, but gets rid of | |
2325 | * lots of complexity around inodes referencing other inodes that exist | |
2326 | * only in the log (references come from either directory items or inode | |
2327 | * back refs). | |
2328 | */ | |
2329 | static int replay_one_buffer(struct btrfs_root *log, struct extent_buffer *eb, | |
2330 | struct walk_control *wc, u64 gen) | |
2331 | { | |
2332 | int nritems; | |
2333 | struct btrfs_path *path; | |
2334 | struct btrfs_root *root = wc->replay_dest; | |
2335 | struct btrfs_key key; | |
e02119d5 CM |
2336 | int level; |
2337 | int i; | |
2338 | int ret; | |
2339 | ||
018642a1 TI |
2340 | ret = btrfs_read_buffer(eb, gen); |
2341 | if (ret) | |
2342 | return ret; | |
e02119d5 CM |
2343 | |
2344 | level = btrfs_header_level(eb); | |
2345 | ||
2346 | if (level != 0) | |
2347 | return 0; | |
2348 | ||
2349 | path = btrfs_alloc_path(); | |
1e5063d0 MF |
2350 | if (!path) |
2351 | return -ENOMEM; | |
e02119d5 CM |
2352 | |
2353 | nritems = btrfs_header_nritems(eb); | |
2354 | for (i = 0; i < nritems; i++) { | |
2355 | btrfs_item_key_to_cpu(eb, &key, i); | |
e02119d5 CM |
2356 | |
2357 | /* inode keys are done during the first stage */ | |
2358 | if (key.type == BTRFS_INODE_ITEM_KEY && | |
2359 | wc->stage == LOG_WALK_REPLAY_INODES) { | |
e02119d5 CM |
2360 | struct btrfs_inode_item *inode_item; |
2361 | u32 mode; | |
2362 | ||
2363 | inode_item = btrfs_item_ptr(eb, i, | |
2364 | struct btrfs_inode_item); | |
4f764e51 FM |
2365 | ret = replay_xattr_deletes(wc->trans, root, log, |
2366 | path, key.objectid); | |
2367 | if (ret) | |
2368 | break; | |
e02119d5 CM |
2369 | mode = btrfs_inode_mode(eb, inode_item); |
2370 | if (S_ISDIR(mode)) { | |
2371 | ret = replay_dir_deletes(wc->trans, | |
12fcfd22 | 2372 | root, log, path, key.objectid, 0); |
b50c6e25 JB |
2373 | if (ret) |
2374 | break; | |
e02119d5 CM |
2375 | } |
2376 | ret = overwrite_item(wc->trans, root, path, | |
2377 | eb, i, &key); | |
b50c6e25 JB |
2378 | if (ret) |
2379 | break; | |
e02119d5 | 2380 | |
c71bf099 | 2381 | /* for regular files, make sure corresponding |
01327610 | 2382 | * orphan item exist. extents past the new EOF |
c71bf099 | 2383 | * will be truncated later by orphan cleanup. |
e02119d5 CM |
2384 | */ |
2385 | if (S_ISREG(mode)) { | |
c71bf099 YZ |
2386 | ret = insert_orphan_item(wc->trans, root, |
2387 | key.objectid); | |
b50c6e25 JB |
2388 | if (ret) |
2389 | break; | |
e02119d5 | 2390 | } |
c71bf099 | 2391 | |
e02119d5 CM |
2392 | ret = link_to_fixup_dir(wc->trans, root, |
2393 | path, key.objectid); | |
b50c6e25 JB |
2394 | if (ret) |
2395 | break; | |
e02119d5 | 2396 | } |
dd8e7217 JB |
2397 | |
2398 | if (key.type == BTRFS_DIR_INDEX_KEY && | |
2399 | wc->stage == LOG_WALK_REPLAY_DIR_INDEX) { | |
2400 | ret = replay_one_dir_item(wc->trans, root, path, | |
2401 | eb, i, &key); | |
2402 | if (ret) | |
2403 | break; | |
2404 | } | |
2405 | ||
e02119d5 CM |
2406 | if (wc->stage < LOG_WALK_REPLAY_ALL) |
2407 | continue; | |
2408 | ||
2409 | /* these keys are simply copied */ | |
2410 | if (key.type == BTRFS_XATTR_ITEM_KEY) { | |
2411 | ret = overwrite_item(wc->trans, root, path, | |
2412 | eb, i, &key); | |
b50c6e25 JB |
2413 | if (ret) |
2414 | break; | |
2da1c669 LB |
2415 | } else if (key.type == BTRFS_INODE_REF_KEY || |
2416 | key.type == BTRFS_INODE_EXTREF_KEY) { | |
f186373f MF |
2417 | ret = add_inode_ref(wc->trans, root, log, path, |
2418 | eb, i, &key); | |
b50c6e25 JB |
2419 | if (ret && ret != -ENOENT) |
2420 | break; | |
2421 | ret = 0; | |
e02119d5 CM |
2422 | } else if (key.type == BTRFS_EXTENT_DATA_KEY) { |
2423 | ret = replay_one_extent(wc->trans, root, path, | |
2424 | eb, i, &key); | |
b50c6e25 JB |
2425 | if (ret) |
2426 | break; | |
dd8e7217 | 2427 | } else if (key.type == BTRFS_DIR_ITEM_KEY) { |
e02119d5 CM |
2428 | ret = replay_one_dir_item(wc->trans, root, path, |
2429 | eb, i, &key); | |
b50c6e25 JB |
2430 | if (ret) |
2431 | break; | |
e02119d5 CM |
2432 | } |
2433 | } | |
2434 | btrfs_free_path(path); | |
b50c6e25 | 2435 | return ret; |
e02119d5 CM |
2436 | } |
2437 | ||
d397712b | 2438 | static noinline int walk_down_log_tree(struct btrfs_trans_handle *trans, |
e02119d5 CM |
2439 | struct btrfs_root *root, |
2440 | struct btrfs_path *path, int *level, | |
2441 | struct walk_control *wc) | |
2442 | { | |
0b246afa | 2443 | struct btrfs_fs_info *fs_info = root->fs_info; |
e02119d5 | 2444 | u64 root_owner; |
e02119d5 CM |
2445 | u64 bytenr; |
2446 | u64 ptr_gen; | |
2447 | struct extent_buffer *next; | |
2448 | struct extent_buffer *cur; | |
2449 | struct extent_buffer *parent; | |
2450 | u32 blocksize; | |
2451 | int ret = 0; | |
2452 | ||
2453 | WARN_ON(*level < 0); | |
2454 | WARN_ON(*level >= BTRFS_MAX_LEVEL); | |
2455 | ||
d397712b | 2456 | while (*level > 0) { |
e02119d5 CM |
2457 | WARN_ON(*level < 0); |
2458 | WARN_ON(*level >= BTRFS_MAX_LEVEL); | |
2459 | cur = path->nodes[*level]; | |
2460 | ||
fae7f21c | 2461 | WARN_ON(btrfs_header_level(cur) != *level); |
e02119d5 CM |
2462 | |
2463 | if (path->slots[*level] >= | |
2464 | btrfs_header_nritems(cur)) | |
2465 | break; | |
2466 | ||
2467 | bytenr = btrfs_node_blockptr(cur, path->slots[*level]); | |
2468 | ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]); | |
0b246afa | 2469 | blocksize = fs_info->nodesize; |
e02119d5 CM |
2470 | |
2471 | parent = path->nodes[*level]; | |
2472 | root_owner = btrfs_header_owner(parent); | |
e02119d5 | 2473 | |
2ff7e61e | 2474 | next = btrfs_find_create_tree_block(fs_info, bytenr); |
c871b0f2 LB |
2475 | if (IS_ERR(next)) |
2476 | return PTR_ERR(next); | |
e02119d5 | 2477 | |
e02119d5 | 2478 | if (*level == 1) { |
1e5063d0 | 2479 | ret = wc->process_func(root, next, wc, ptr_gen); |
b50c6e25 JB |
2480 | if (ret) { |
2481 | free_extent_buffer(next); | |
1e5063d0 | 2482 | return ret; |
b50c6e25 | 2483 | } |
4a500fd1 | 2484 | |
e02119d5 CM |
2485 | path->slots[*level]++; |
2486 | if (wc->free) { | |
018642a1 TI |
2487 | ret = btrfs_read_buffer(next, ptr_gen); |
2488 | if (ret) { | |
2489 | free_extent_buffer(next); | |
2490 | return ret; | |
2491 | } | |
e02119d5 | 2492 | |
681ae509 JB |
2493 | if (trans) { |
2494 | btrfs_tree_lock(next); | |
2495 | btrfs_set_lock_blocking(next); | |
7c302b49 | 2496 | clean_tree_block(fs_info, next); |
681ae509 JB |
2497 | btrfs_wait_tree_block_writeback(next); |
2498 | btrfs_tree_unlock(next); | |
2499 | } | |
e02119d5 | 2500 | |
e02119d5 CM |
2501 | WARN_ON(root_owner != |
2502 | BTRFS_TREE_LOG_OBJECTID); | |
2ff7e61e JM |
2503 | ret = btrfs_free_and_pin_reserved_extent( |
2504 | fs_info, bytenr, | |
2505 | blocksize); | |
3650860b JB |
2506 | if (ret) { |
2507 | free_extent_buffer(next); | |
2508 | return ret; | |
2509 | } | |
e02119d5 CM |
2510 | } |
2511 | free_extent_buffer(next); | |
2512 | continue; | |
2513 | } | |
018642a1 TI |
2514 | ret = btrfs_read_buffer(next, ptr_gen); |
2515 | if (ret) { | |
2516 | free_extent_buffer(next); | |
2517 | return ret; | |
2518 | } | |
e02119d5 CM |
2519 | |
2520 | WARN_ON(*level <= 0); | |
2521 | if (path->nodes[*level-1]) | |
2522 | free_extent_buffer(path->nodes[*level-1]); | |
2523 | path->nodes[*level-1] = next; | |
2524 | *level = btrfs_header_level(next); | |
2525 | path->slots[*level] = 0; | |
2526 | cond_resched(); | |
2527 | } | |
2528 | WARN_ON(*level < 0); | |
2529 | WARN_ON(*level >= BTRFS_MAX_LEVEL); | |
2530 | ||
4a500fd1 | 2531 | path->slots[*level] = btrfs_header_nritems(path->nodes[*level]); |
e02119d5 CM |
2532 | |
2533 | cond_resched(); | |
2534 | return 0; | |
2535 | } | |
2536 | ||
d397712b | 2537 | static noinline int walk_up_log_tree(struct btrfs_trans_handle *trans, |
e02119d5 CM |
2538 | struct btrfs_root *root, |
2539 | struct btrfs_path *path, int *level, | |
2540 | struct walk_control *wc) | |
2541 | { | |
0b246afa | 2542 | struct btrfs_fs_info *fs_info = root->fs_info; |
e02119d5 | 2543 | u64 root_owner; |
e02119d5 CM |
2544 | int i; |
2545 | int slot; | |
2546 | int ret; | |
2547 | ||
d397712b | 2548 | for (i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) { |
e02119d5 | 2549 | slot = path->slots[i]; |
4a500fd1 | 2550 | if (slot + 1 < btrfs_header_nritems(path->nodes[i])) { |
e02119d5 CM |
2551 | path->slots[i]++; |
2552 | *level = i; | |
2553 | WARN_ON(*level == 0); | |
2554 | return 0; | |
2555 | } else { | |
31840ae1 ZY |
2556 | struct extent_buffer *parent; |
2557 | if (path->nodes[*level] == root->node) | |
2558 | parent = path->nodes[*level]; | |
2559 | else | |
2560 | parent = path->nodes[*level + 1]; | |
2561 | ||
2562 | root_owner = btrfs_header_owner(parent); | |
1e5063d0 | 2563 | ret = wc->process_func(root, path->nodes[*level], wc, |
e02119d5 | 2564 | btrfs_header_generation(path->nodes[*level])); |
1e5063d0 MF |
2565 | if (ret) |
2566 | return ret; | |
2567 | ||
e02119d5 CM |
2568 | if (wc->free) { |
2569 | struct extent_buffer *next; | |
2570 | ||
2571 | next = path->nodes[*level]; | |
2572 | ||
681ae509 JB |
2573 | if (trans) { |
2574 | btrfs_tree_lock(next); | |
2575 | btrfs_set_lock_blocking(next); | |
7c302b49 | 2576 | clean_tree_block(fs_info, next); |
681ae509 JB |
2577 | btrfs_wait_tree_block_writeback(next); |
2578 | btrfs_tree_unlock(next); | |
2579 | } | |
e02119d5 | 2580 | |
e02119d5 | 2581 | WARN_ON(root_owner != BTRFS_TREE_LOG_OBJECTID); |
2ff7e61e JM |
2582 | ret = btrfs_free_and_pin_reserved_extent( |
2583 | fs_info, | |
e02119d5 | 2584 | path->nodes[*level]->start, |
d00aff00 | 2585 | path->nodes[*level]->len); |
3650860b JB |
2586 | if (ret) |
2587 | return ret; | |
e02119d5 CM |
2588 | } |
2589 | free_extent_buffer(path->nodes[*level]); | |
2590 | path->nodes[*level] = NULL; | |
2591 | *level = i + 1; | |
2592 | } | |
2593 | } | |
2594 | return 1; | |
2595 | } | |
2596 | ||
2597 | /* | |
2598 | * drop the reference count on the tree rooted at 'snap'. This traverses | |
2599 | * the tree freeing any blocks that have a ref count of zero after being | |
2600 | * decremented. | |
2601 | */ | |
2602 | static int walk_log_tree(struct btrfs_trans_handle *trans, | |
2603 | struct btrfs_root *log, struct walk_control *wc) | |
2604 | { | |
2ff7e61e | 2605 | struct btrfs_fs_info *fs_info = log->fs_info; |
e02119d5 CM |
2606 | int ret = 0; |
2607 | int wret; | |
2608 | int level; | |
2609 | struct btrfs_path *path; | |
e02119d5 CM |
2610 | int orig_level; |
2611 | ||
2612 | path = btrfs_alloc_path(); | |
db5b493a TI |
2613 | if (!path) |
2614 | return -ENOMEM; | |
e02119d5 CM |
2615 | |
2616 | level = btrfs_header_level(log->node); | |
2617 | orig_level = level; | |
2618 | path->nodes[level] = log->node; | |
2619 | extent_buffer_get(log->node); | |
2620 | path->slots[level] = 0; | |
2621 | ||
d397712b | 2622 | while (1) { |
e02119d5 CM |
2623 | wret = walk_down_log_tree(trans, log, path, &level, wc); |
2624 | if (wret > 0) | |
2625 | break; | |
79787eaa | 2626 | if (wret < 0) { |
e02119d5 | 2627 | ret = wret; |
79787eaa JM |
2628 | goto out; |
2629 | } | |
e02119d5 CM |
2630 | |
2631 | wret = walk_up_log_tree(trans, log, path, &level, wc); | |
2632 | if (wret > 0) | |
2633 | break; | |
79787eaa | 2634 | if (wret < 0) { |
e02119d5 | 2635 | ret = wret; |
79787eaa JM |
2636 | goto out; |
2637 | } | |
e02119d5 CM |
2638 | } |
2639 | ||
2640 | /* was the root node processed? if not, catch it here */ | |
2641 | if (path->nodes[orig_level]) { | |
79787eaa | 2642 | ret = wc->process_func(log, path->nodes[orig_level], wc, |
e02119d5 | 2643 | btrfs_header_generation(path->nodes[orig_level])); |
79787eaa JM |
2644 | if (ret) |
2645 | goto out; | |
e02119d5 CM |
2646 | if (wc->free) { |
2647 | struct extent_buffer *next; | |
2648 | ||
2649 | next = path->nodes[orig_level]; | |
2650 | ||
681ae509 JB |
2651 | if (trans) { |
2652 | btrfs_tree_lock(next); | |
2653 | btrfs_set_lock_blocking(next); | |
7c302b49 | 2654 | clean_tree_block(fs_info, next); |
681ae509 JB |
2655 | btrfs_wait_tree_block_writeback(next); |
2656 | btrfs_tree_unlock(next); | |
2657 | } | |
e02119d5 | 2658 | |
e02119d5 CM |
2659 | WARN_ON(log->root_key.objectid != |
2660 | BTRFS_TREE_LOG_OBJECTID); | |
2ff7e61e JM |
2661 | ret = btrfs_free_and_pin_reserved_extent(fs_info, |
2662 | next->start, next->len); | |
3650860b JB |
2663 | if (ret) |
2664 | goto out; | |
e02119d5 CM |
2665 | } |
2666 | } | |
2667 | ||
79787eaa | 2668 | out: |
e02119d5 | 2669 | btrfs_free_path(path); |
e02119d5 CM |
2670 | return ret; |
2671 | } | |
2672 | ||
7237f183 YZ |
2673 | /* |
2674 | * helper function to update the item for a given subvolumes log root | |
2675 | * in the tree of log roots | |
2676 | */ | |
2677 | static int update_log_root(struct btrfs_trans_handle *trans, | |
2678 | struct btrfs_root *log) | |
2679 | { | |
0b246afa | 2680 | struct btrfs_fs_info *fs_info = log->fs_info; |
7237f183 YZ |
2681 | int ret; |
2682 | ||
2683 | if (log->log_transid == 1) { | |
2684 | /* insert root item on the first sync */ | |
0b246afa | 2685 | ret = btrfs_insert_root(trans, fs_info->log_root_tree, |
7237f183 YZ |
2686 | &log->root_key, &log->root_item); |
2687 | } else { | |
0b246afa | 2688 | ret = btrfs_update_root(trans, fs_info->log_root_tree, |
7237f183 YZ |
2689 | &log->root_key, &log->root_item); |
2690 | } | |
2691 | return ret; | |
2692 | } | |
2693 | ||
60d53eb3 | 2694 | static void wait_log_commit(struct btrfs_root *root, int transid) |
e02119d5 CM |
2695 | { |
2696 | DEFINE_WAIT(wait); | |
7237f183 | 2697 | int index = transid % 2; |
e02119d5 | 2698 | |
7237f183 YZ |
2699 | /* |
2700 | * we only allow two pending log transactions at a time, | |
2701 | * so we know that if ours is more than 2 older than the | |
2702 | * current transaction, we're done | |
2703 | */ | |
e02119d5 | 2704 | do { |
7237f183 YZ |
2705 | prepare_to_wait(&root->log_commit_wait[index], |
2706 | &wait, TASK_UNINTERRUPTIBLE); | |
2707 | mutex_unlock(&root->log_mutex); | |
12fcfd22 | 2708 | |
d1433deb | 2709 | if (root->log_transid_committed < transid && |
7237f183 YZ |
2710 | atomic_read(&root->log_commit[index])) |
2711 | schedule(); | |
12fcfd22 | 2712 | |
7237f183 YZ |
2713 | finish_wait(&root->log_commit_wait[index], &wait); |
2714 | mutex_lock(&root->log_mutex); | |
d1433deb | 2715 | } while (root->log_transid_committed < transid && |
7237f183 | 2716 | atomic_read(&root->log_commit[index])); |
7237f183 YZ |
2717 | } |
2718 | ||
60d53eb3 | 2719 | static void wait_for_writer(struct btrfs_root *root) |
7237f183 YZ |
2720 | { |
2721 | DEFINE_WAIT(wait); | |
8b050d35 MX |
2722 | |
2723 | while (atomic_read(&root->log_writers)) { | |
7237f183 YZ |
2724 | prepare_to_wait(&root->log_writer_wait, |
2725 | &wait, TASK_UNINTERRUPTIBLE); | |
2726 | mutex_unlock(&root->log_mutex); | |
8b050d35 | 2727 | if (atomic_read(&root->log_writers)) |
e02119d5 | 2728 | schedule(); |
7237f183 | 2729 | finish_wait(&root->log_writer_wait, &wait); |
575849ec | 2730 | mutex_lock(&root->log_mutex); |
7237f183 | 2731 | } |
e02119d5 CM |
2732 | } |
2733 | ||
8b050d35 MX |
2734 | static inline void btrfs_remove_log_ctx(struct btrfs_root *root, |
2735 | struct btrfs_log_ctx *ctx) | |
2736 | { | |
2737 | if (!ctx) | |
2738 | return; | |
2739 | ||
2740 | mutex_lock(&root->log_mutex); | |
2741 | list_del_init(&ctx->list); | |
2742 | mutex_unlock(&root->log_mutex); | |
2743 | } | |
2744 | ||
2745 | /* | |
2746 | * Invoked in log mutex context, or be sure there is no other task which | |
2747 | * can access the list. | |
2748 | */ | |
2749 | static inline void btrfs_remove_all_log_ctxs(struct btrfs_root *root, | |
2750 | int index, int error) | |
2751 | { | |
2752 | struct btrfs_log_ctx *ctx; | |
570dd450 | 2753 | struct btrfs_log_ctx *safe; |
8b050d35 | 2754 | |
570dd450 CM |
2755 | list_for_each_entry_safe(ctx, safe, &root->log_ctxs[index], list) { |
2756 | list_del_init(&ctx->list); | |
8b050d35 | 2757 | ctx->log_ret = error; |
570dd450 | 2758 | } |
8b050d35 MX |
2759 | |
2760 | INIT_LIST_HEAD(&root->log_ctxs[index]); | |
2761 | } | |
2762 | ||
e02119d5 CM |
2763 | /* |
2764 | * btrfs_sync_log does sends a given tree log down to the disk and | |
2765 | * updates the super blocks to record it. When this call is done, | |
12fcfd22 CM |
2766 | * you know that any inodes previously logged are safely on disk only |
2767 | * if it returns 0. | |
2768 | * | |
2769 | * Any other return value means you need to call btrfs_commit_transaction. | |
2770 | * Some of the edge cases for fsyncing directories that have had unlinks | |
2771 | * or renames done in the past mean that sometimes the only safe | |
2772 | * fsync is to commit the whole FS. When btrfs_sync_log returns -EAGAIN, | |
2773 | * that has happened. | |
e02119d5 CM |
2774 | */ |
2775 | int btrfs_sync_log(struct btrfs_trans_handle *trans, | |
8b050d35 | 2776 | struct btrfs_root *root, struct btrfs_log_ctx *ctx) |
e02119d5 | 2777 | { |
7237f183 YZ |
2778 | int index1; |
2779 | int index2; | |
8cef4e16 | 2780 | int mark; |
e02119d5 | 2781 | int ret; |
0b246afa | 2782 | struct btrfs_fs_info *fs_info = root->fs_info; |
e02119d5 | 2783 | struct btrfs_root *log = root->log_root; |
0b246afa | 2784 | struct btrfs_root *log_root_tree = fs_info->log_root_tree; |
bb14a59b | 2785 | int log_transid = 0; |
8b050d35 | 2786 | struct btrfs_log_ctx root_log_ctx; |
c6adc9cc | 2787 | struct blk_plug plug; |
e02119d5 | 2788 | |
7237f183 | 2789 | mutex_lock(&root->log_mutex); |
d1433deb MX |
2790 | log_transid = ctx->log_transid; |
2791 | if (root->log_transid_committed >= log_transid) { | |
2792 | mutex_unlock(&root->log_mutex); | |
2793 | return ctx->log_ret; | |
2794 | } | |
2795 | ||
2796 | index1 = log_transid % 2; | |
7237f183 | 2797 | if (atomic_read(&root->log_commit[index1])) { |
60d53eb3 | 2798 | wait_log_commit(root, log_transid); |
7237f183 | 2799 | mutex_unlock(&root->log_mutex); |
8b050d35 | 2800 | return ctx->log_ret; |
e02119d5 | 2801 | } |
d1433deb | 2802 | ASSERT(log_transid == root->log_transid); |
7237f183 YZ |
2803 | atomic_set(&root->log_commit[index1], 1); |
2804 | ||
2805 | /* wait for previous tree log sync to complete */ | |
2806 | if (atomic_read(&root->log_commit[(index1 + 1) % 2])) | |
60d53eb3 | 2807 | wait_log_commit(root, log_transid - 1); |
48cab2e0 | 2808 | |
86df7eb9 | 2809 | while (1) { |
2ecb7923 | 2810 | int batch = atomic_read(&root->log_batch); |
cd354ad6 | 2811 | /* when we're on an ssd, just kick the log commit out */ |
0b246afa | 2812 | if (!btrfs_test_opt(fs_info, SSD) && |
27cdeb70 | 2813 | test_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state)) { |
86df7eb9 YZ |
2814 | mutex_unlock(&root->log_mutex); |
2815 | schedule_timeout_uninterruptible(1); | |
2816 | mutex_lock(&root->log_mutex); | |
2817 | } | |
60d53eb3 | 2818 | wait_for_writer(root); |
2ecb7923 | 2819 | if (batch == atomic_read(&root->log_batch)) |
e02119d5 CM |
2820 | break; |
2821 | } | |
e02119d5 | 2822 | |
12fcfd22 | 2823 | /* bail out if we need to do a full commit */ |
0b246afa | 2824 | if (btrfs_need_log_full_commit(fs_info, trans)) { |
12fcfd22 | 2825 | ret = -EAGAIN; |
2ab28f32 | 2826 | btrfs_free_logged_extents(log, log_transid); |
12fcfd22 CM |
2827 | mutex_unlock(&root->log_mutex); |
2828 | goto out; | |
2829 | } | |
2830 | ||
8cef4e16 YZ |
2831 | if (log_transid % 2 == 0) |
2832 | mark = EXTENT_DIRTY; | |
2833 | else | |
2834 | mark = EXTENT_NEW; | |
2835 | ||
690587d1 CM |
2836 | /* we start IO on all the marked extents here, but we don't actually |
2837 | * wait for them until later. | |
2838 | */ | |
c6adc9cc | 2839 | blk_start_plug(&plug); |
2ff7e61e | 2840 | ret = btrfs_write_marked_extents(fs_info, &log->dirty_log_pages, mark); |
79787eaa | 2841 | if (ret) { |
c6adc9cc | 2842 | blk_finish_plug(&plug); |
66642832 | 2843 | btrfs_abort_transaction(trans, ret); |
2ab28f32 | 2844 | btrfs_free_logged_extents(log, log_transid); |
0b246afa | 2845 | btrfs_set_log_full_commit(fs_info, trans); |
79787eaa JM |
2846 | mutex_unlock(&root->log_mutex); |
2847 | goto out; | |
2848 | } | |
7237f183 | 2849 | |
5d4f98a2 | 2850 | btrfs_set_root_node(&log->root_item, log->node); |
7237f183 | 2851 | |
7237f183 YZ |
2852 | root->log_transid++; |
2853 | log->log_transid = root->log_transid; | |
ff782e0a | 2854 | root->log_start_pid = 0; |
7237f183 | 2855 | /* |
8cef4e16 YZ |
2856 | * IO has been started, blocks of the log tree have WRITTEN flag set |
2857 | * in their headers. new modifications of the log will be written to | |
2858 | * new positions. so it's safe to allow log writers to go in. | |
7237f183 YZ |
2859 | */ |
2860 | mutex_unlock(&root->log_mutex); | |
2861 | ||
28a23593 | 2862 | btrfs_init_log_ctx(&root_log_ctx, NULL); |
d1433deb | 2863 | |
7237f183 | 2864 | mutex_lock(&log_root_tree->log_mutex); |
2ecb7923 | 2865 | atomic_inc(&log_root_tree->log_batch); |
7237f183 | 2866 | atomic_inc(&log_root_tree->log_writers); |
d1433deb MX |
2867 | |
2868 | index2 = log_root_tree->log_transid % 2; | |
2869 | list_add_tail(&root_log_ctx.list, &log_root_tree->log_ctxs[index2]); | |
2870 | root_log_ctx.log_transid = log_root_tree->log_transid; | |
2871 | ||
7237f183 YZ |
2872 | mutex_unlock(&log_root_tree->log_mutex); |
2873 | ||
2874 | ret = update_log_root(trans, log); | |
7237f183 YZ |
2875 | |
2876 | mutex_lock(&log_root_tree->log_mutex); | |
2877 | if (atomic_dec_and_test(&log_root_tree->log_writers)) { | |
779adf0f DS |
2878 | /* |
2879 | * Implicit memory barrier after atomic_dec_and_test | |
2880 | */ | |
7237f183 YZ |
2881 | if (waitqueue_active(&log_root_tree->log_writer_wait)) |
2882 | wake_up(&log_root_tree->log_writer_wait); | |
2883 | } | |
2884 | ||
4a500fd1 | 2885 | if (ret) { |
d1433deb MX |
2886 | if (!list_empty(&root_log_ctx.list)) |
2887 | list_del_init(&root_log_ctx.list); | |
2888 | ||
c6adc9cc | 2889 | blk_finish_plug(&plug); |
0b246afa | 2890 | btrfs_set_log_full_commit(fs_info, trans); |
995946dd | 2891 | |
79787eaa | 2892 | if (ret != -ENOSPC) { |
66642832 | 2893 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
2894 | mutex_unlock(&log_root_tree->log_mutex); |
2895 | goto out; | |
2896 | } | |
bf89d38f | 2897 | btrfs_wait_tree_log_extents(log, mark); |
2ab28f32 | 2898 | btrfs_free_logged_extents(log, log_transid); |
4a500fd1 YZ |
2899 | mutex_unlock(&log_root_tree->log_mutex); |
2900 | ret = -EAGAIN; | |
2901 | goto out; | |
2902 | } | |
2903 | ||
d1433deb | 2904 | if (log_root_tree->log_transid_committed >= root_log_ctx.log_transid) { |
3da5ab56 | 2905 | blk_finish_plug(&plug); |
cbd60aa7 | 2906 | list_del_init(&root_log_ctx.list); |
d1433deb MX |
2907 | mutex_unlock(&log_root_tree->log_mutex); |
2908 | ret = root_log_ctx.log_ret; | |
2909 | goto out; | |
2910 | } | |
8b050d35 | 2911 | |
d1433deb | 2912 | index2 = root_log_ctx.log_transid % 2; |
7237f183 | 2913 | if (atomic_read(&log_root_tree->log_commit[index2])) { |
c6adc9cc | 2914 | blk_finish_plug(&plug); |
bf89d38f | 2915 | ret = btrfs_wait_tree_log_extents(log, mark); |
50d9aa99 | 2916 | btrfs_wait_logged_extents(trans, log, log_transid); |
60d53eb3 | 2917 | wait_log_commit(log_root_tree, |
d1433deb | 2918 | root_log_ctx.log_transid); |
7237f183 | 2919 | mutex_unlock(&log_root_tree->log_mutex); |
5ab5e44a FM |
2920 | if (!ret) |
2921 | ret = root_log_ctx.log_ret; | |
7237f183 YZ |
2922 | goto out; |
2923 | } | |
d1433deb | 2924 | ASSERT(root_log_ctx.log_transid == log_root_tree->log_transid); |
7237f183 YZ |
2925 | atomic_set(&log_root_tree->log_commit[index2], 1); |
2926 | ||
12fcfd22 | 2927 | if (atomic_read(&log_root_tree->log_commit[(index2 + 1) % 2])) { |
60d53eb3 | 2928 | wait_log_commit(log_root_tree, |
d1433deb | 2929 | root_log_ctx.log_transid - 1); |
12fcfd22 CM |
2930 | } |
2931 | ||
60d53eb3 | 2932 | wait_for_writer(log_root_tree); |
7237f183 | 2933 | |
12fcfd22 CM |
2934 | /* |
2935 | * now that we've moved on to the tree of log tree roots, | |
2936 | * check the full commit flag again | |
2937 | */ | |
0b246afa | 2938 | if (btrfs_need_log_full_commit(fs_info, trans)) { |
c6adc9cc | 2939 | blk_finish_plug(&plug); |
bf89d38f | 2940 | btrfs_wait_tree_log_extents(log, mark); |
2ab28f32 | 2941 | btrfs_free_logged_extents(log, log_transid); |
12fcfd22 CM |
2942 | mutex_unlock(&log_root_tree->log_mutex); |
2943 | ret = -EAGAIN; | |
2944 | goto out_wake_log_root; | |
2945 | } | |
7237f183 | 2946 | |
2ff7e61e | 2947 | ret = btrfs_write_marked_extents(fs_info, |
c6adc9cc MX |
2948 | &log_root_tree->dirty_log_pages, |
2949 | EXTENT_DIRTY | EXTENT_NEW); | |
2950 | blk_finish_plug(&plug); | |
79787eaa | 2951 | if (ret) { |
0b246afa | 2952 | btrfs_set_log_full_commit(fs_info, trans); |
66642832 | 2953 | btrfs_abort_transaction(trans, ret); |
2ab28f32 | 2954 | btrfs_free_logged_extents(log, log_transid); |
79787eaa JM |
2955 | mutex_unlock(&log_root_tree->log_mutex); |
2956 | goto out_wake_log_root; | |
2957 | } | |
bf89d38f | 2958 | ret = btrfs_wait_tree_log_extents(log, mark); |
5ab5e44a | 2959 | if (!ret) |
bf89d38f JM |
2960 | ret = btrfs_wait_tree_log_extents(log_root_tree, |
2961 | EXTENT_NEW | EXTENT_DIRTY); | |
5ab5e44a | 2962 | if (ret) { |
0b246afa | 2963 | btrfs_set_log_full_commit(fs_info, trans); |
5ab5e44a FM |
2964 | btrfs_free_logged_extents(log, log_transid); |
2965 | mutex_unlock(&log_root_tree->log_mutex); | |
2966 | goto out_wake_log_root; | |
2967 | } | |
50d9aa99 | 2968 | btrfs_wait_logged_extents(trans, log, log_transid); |
e02119d5 | 2969 | |
0b246afa JM |
2970 | btrfs_set_super_log_root(fs_info->super_for_commit, |
2971 | log_root_tree->node->start); | |
2972 | btrfs_set_super_log_root_level(fs_info->super_for_commit, | |
2973 | btrfs_header_level(log_root_tree->node)); | |
e02119d5 | 2974 | |
7237f183 | 2975 | log_root_tree->log_transid++; |
7237f183 YZ |
2976 | mutex_unlock(&log_root_tree->log_mutex); |
2977 | ||
2978 | /* | |
2979 | * nobody else is going to jump in and write the the ctree | |
2980 | * super here because the log_commit atomic below is protecting | |
2981 | * us. We must be called with a transaction handle pinning | |
2982 | * the running transaction open, so a full commit can't hop | |
2983 | * in and cause problems either. | |
2984 | */ | |
eece6a9c | 2985 | ret = write_all_supers(fs_info, 1); |
5af3e8cc | 2986 | if (ret) { |
0b246afa | 2987 | btrfs_set_log_full_commit(fs_info, trans); |
66642832 | 2988 | btrfs_abort_transaction(trans, ret); |
5af3e8cc SB |
2989 | goto out_wake_log_root; |
2990 | } | |
7237f183 | 2991 | |
257c62e1 CM |
2992 | mutex_lock(&root->log_mutex); |
2993 | if (root->last_log_commit < log_transid) | |
2994 | root->last_log_commit = log_transid; | |
2995 | mutex_unlock(&root->log_mutex); | |
2996 | ||
12fcfd22 | 2997 | out_wake_log_root: |
570dd450 | 2998 | mutex_lock(&log_root_tree->log_mutex); |
8b050d35 MX |
2999 | btrfs_remove_all_log_ctxs(log_root_tree, index2, ret); |
3000 | ||
d1433deb | 3001 | log_root_tree->log_transid_committed++; |
7237f183 | 3002 | atomic_set(&log_root_tree->log_commit[index2], 0); |
d1433deb MX |
3003 | mutex_unlock(&log_root_tree->log_mutex); |
3004 | ||
33a9eca7 DS |
3005 | /* |
3006 | * The barrier before waitqueue_active is implied by mutex_unlock | |
3007 | */ | |
7237f183 YZ |
3008 | if (waitqueue_active(&log_root_tree->log_commit_wait[index2])) |
3009 | wake_up(&log_root_tree->log_commit_wait[index2]); | |
e02119d5 | 3010 | out: |
d1433deb | 3011 | mutex_lock(&root->log_mutex); |
570dd450 | 3012 | btrfs_remove_all_log_ctxs(root, index1, ret); |
d1433deb | 3013 | root->log_transid_committed++; |
7237f183 | 3014 | atomic_set(&root->log_commit[index1], 0); |
d1433deb | 3015 | mutex_unlock(&root->log_mutex); |
8b050d35 | 3016 | |
33a9eca7 DS |
3017 | /* |
3018 | * The barrier before waitqueue_active is implied by mutex_unlock | |
3019 | */ | |
7237f183 YZ |
3020 | if (waitqueue_active(&root->log_commit_wait[index1])) |
3021 | wake_up(&root->log_commit_wait[index1]); | |
b31eabd8 | 3022 | return ret; |
e02119d5 CM |
3023 | } |
3024 | ||
4a500fd1 YZ |
3025 | static void free_log_tree(struct btrfs_trans_handle *trans, |
3026 | struct btrfs_root *log) | |
e02119d5 CM |
3027 | { |
3028 | int ret; | |
d0c803c4 CM |
3029 | u64 start; |
3030 | u64 end; | |
e02119d5 CM |
3031 | struct walk_control wc = { |
3032 | .free = 1, | |
3033 | .process_func = process_one_buffer | |
3034 | }; | |
3035 | ||
681ae509 JB |
3036 | ret = walk_log_tree(trans, log, &wc); |
3037 | /* I don't think this can happen but just in case */ | |
3038 | if (ret) | |
66642832 | 3039 | btrfs_abort_transaction(trans, ret); |
e02119d5 | 3040 | |
d397712b | 3041 | while (1) { |
d0c803c4 | 3042 | ret = find_first_extent_bit(&log->dirty_log_pages, |
e6138876 JB |
3043 | 0, &start, &end, EXTENT_DIRTY | EXTENT_NEW, |
3044 | NULL); | |
d0c803c4 CM |
3045 | if (ret) |
3046 | break; | |
3047 | ||
8cef4e16 | 3048 | clear_extent_bits(&log->dirty_log_pages, start, end, |
91166212 | 3049 | EXTENT_DIRTY | EXTENT_NEW); |
d0c803c4 CM |
3050 | } |
3051 | ||
2ab28f32 JB |
3052 | /* |
3053 | * We may have short-circuited the log tree with the full commit logic | |
3054 | * and left ordered extents on our list, so clear these out to keep us | |
3055 | * from leaking inodes and memory. | |
3056 | */ | |
3057 | btrfs_free_logged_extents(log, 0); | |
3058 | btrfs_free_logged_extents(log, 1); | |
3059 | ||
7237f183 YZ |
3060 | free_extent_buffer(log->node); |
3061 | kfree(log); | |
4a500fd1 YZ |
3062 | } |
3063 | ||
3064 | /* | |
3065 | * free all the extents used by the tree log. This should be called | |
3066 | * at commit time of the full transaction | |
3067 | */ | |
3068 | int btrfs_free_log(struct btrfs_trans_handle *trans, struct btrfs_root *root) | |
3069 | { | |
3070 | if (root->log_root) { | |
3071 | free_log_tree(trans, root->log_root); | |
3072 | root->log_root = NULL; | |
3073 | } | |
3074 | return 0; | |
3075 | } | |
3076 | ||
3077 | int btrfs_free_log_root_tree(struct btrfs_trans_handle *trans, | |
3078 | struct btrfs_fs_info *fs_info) | |
3079 | { | |
3080 | if (fs_info->log_root_tree) { | |
3081 | free_log_tree(trans, fs_info->log_root_tree); | |
3082 | fs_info->log_root_tree = NULL; | |
3083 | } | |
e02119d5 CM |
3084 | return 0; |
3085 | } | |
3086 | ||
e02119d5 CM |
3087 | /* |
3088 | * If both a file and directory are logged, and unlinks or renames are | |
3089 | * mixed in, we have a few interesting corners: | |
3090 | * | |
3091 | * create file X in dir Y | |
3092 | * link file X to X.link in dir Y | |
3093 | * fsync file X | |
3094 | * unlink file X but leave X.link | |
3095 | * fsync dir Y | |
3096 | * | |
3097 | * After a crash we would expect only X.link to exist. But file X | |
3098 | * didn't get fsync'd again so the log has back refs for X and X.link. | |
3099 | * | |
3100 | * We solve this by removing directory entries and inode backrefs from the | |
3101 | * log when a file that was logged in the current transaction is | |
3102 | * unlinked. Any later fsync will include the updated log entries, and | |
3103 | * we'll be able to reconstruct the proper directory items from backrefs. | |
3104 | * | |
3105 | * This optimizations allows us to avoid relogging the entire inode | |
3106 | * or the entire directory. | |
3107 | */ | |
3108 | int btrfs_del_dir_entries_in_log(struct btrfs_trans_handle *trans, | |
3109 | struct btrfs_root *root, | |
3110 | const char *name, int name_len, | |
49f34d1f | 3111 | struct btrfs_inode *dir, u64 index) |
e02119d5 CM |
3112 | { |
3113 | struct btrfs_root *log; | |
3114 | struct btrfs_dir_item *di; | |
3115 | struct btrfs_path *path; | |
3116 | int ret; | |
4a500fd1 | 3117 | int err = 0; |
e02119d5 | 3118 | int bytes_del = 0; |
49f34d1f | 3119 | u64 dir_ino = btrfs_ino(dir); |
e02119d5 | 3120 | |
49f34d1f | 3121 | if (dir->logged_trans < trans->transid) |
3a5f1d45 CM |
3122 | return 0; |
3123 | ||
e02119d5 CM |
3124 | ret = join_running_log_trans(root); |
3125 | if (ret) | |
3126 | return 0; | |
3127 | ||
49f34d1f | 3128 | mutex_lock(&dir->log_mutex); |
e02119d5 CM |
3129 | |
3130 | log = root->log_root; | |
3131 | path = btrfs_alloc_path(); | |
a62f44a5 TI |
3132 | if (!path) { |
3133 | err = -ENOMEM; | |
3134 | goto out_unlock; | |
3135 | } | |
2a29edc6 | 3136 | |
33345d01 | 3137 | di = btrfs_lookup_dir_item(trans, log, path, dir_ino, |
e02119d5 | 3138 | name, name_len, -1); |
4a500fd1 YZ |
3139 | if (IS_ERR(di)) { |
3140 | err = PTR_ERR(di); | |
3141 | goto fail; | |
3142 | } | |
3143 | if (di) { | |
e02119d5 CM |
3144 | ret = btrfs_delete_one_dir_name(trans, log, path, di); |
3145 | bytes_del += name_len; | |
3650860b JB |
3146 | if (ret) { |
3147 | err = ret; | |
3148 | goto fail; | |
3149 | } | |
e02119d5 | 3150 | } |
b3b4aa74 | 3151 | btrfs_release_path(path); |
33345d01 | 3152 | di = btrfs_lookup_dir_index_item(trans, log, path, dir_ino, |
e02119d5 | 3153 | index, name, name_len, -1); |
4a500fd1 YZ |
3154 | if (IS_ERR(di)) { |
3155 | err = PTR_ERR(di); | |
3156 | goto fail; | |
3157 | } | |
3158 | if (di) { | |
e02119d5 CM |
3159 | ret = btrfs_delete_one_dir_name(trans, log, path, di); |
3160 | bytes_del += name_len; | |
3650860b JB |
3161 | if (ret) { |
3162 | err = ret; | |
3163 | goto fail; | |
3164 | } | |
e02119d5 CM |
3165 | } |
3166 | ||
3167 | /* update the directory size in the log to reflect the names | |
3168 | * we have removed | |
3169 | */ | |
3170 | if (bytes_del) { | |
3171 | struct btrfs_key key; | |
3172 | ||
33345d01 | 3173 | key.objectid = dir_ino; |
e02119d5 CM |
3174 | key.offset = 0; |
3175 | key.type = BTRFS_INODE_ITEM_KEY; | |
b3b4aa74 | 3176 | btrfs_release_path(path); |
e02119d5 CM |
3177 | |
3178 | ret = btrfs_search_slot(trans, log, &key, path, 0, 1); | |
4a500fd1 YZ |
3179 | if (ret < 0) { |
3180 | err = ret; | |
3181 | goto fail; | |
3182 | } | |
e02119d5 CM |
3183 | if (ret == 0) { |
3184 | struct btrfs_inode_item *item; | |
3185 | u64 i_size; | |
3186 | ||
3187 | item = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
3188 | struct btrfs_inode_item); | |
3189 | i_size = btrfs_inode_size(path->nodes[0], item); | |
3190 | if (i_size > bytes_del) | |
3191 | i_size -= bytes_del; | |
3192 | else | |
3193 | i_size = 0; | |
3194 | btrfs_set_inode_size(path->nodes[0], item, i_size); | |
3195 | btrfs_mark_buffer_dirty(path->nodes[0]); | |
3196 | } else | |
3197 | ret = 0; | |
b3b4aa74 | 3198 | btrfs_release_path(path); |
e02119d5 | 3199 | } |
4a500fd1 | 3200 | fail: |
e02119d5 | 3201 | btrfs_free_path(path); |
a62f44a5 | 3202 | out_unlock: |
49f34d1f | 3203 | mutex_unlock(&dir->log_mutex); |
4a500fd1 | 3204 | if (ret == -ENOSPC) { |
995946dd | 3205 | btrfs_set_log_full_commit(root->fs_info, trans); |
4a500fd1 | 3206 | ret = 0; |
79787eaa | 3207 | } else if (ret < 0) |
66642832 | 3208 | btrfs_abort_transaction(trans, ret); |
79787eaa | 3209 | |
12fcfd22 | 3210 | btrfs_end_log_trans(root); |
e02119d5 | 3211 | |
411fc6bc | 3212 | return err; |
e02119d5 CM |
3213 | } |
3214 | ||
3215 | /* see comments for btrfs_del_dir_entries_in_log */ | |
3216 | int btrfs_del_inode_ref_in_log(struct btrfs_trans_handle *trans, | |
3217 | struct btrfs_root *root, | |
3218 | const char *name, int name_len, | |
a491abb2 | 3219 | struct btrfs_inode *inode, u64 dirid) |
e02119d5 | 3220 | { |
0b246afa | 3221 | struct btrfs_fs_info *fs_info = root->fs_info; |
e02119d5 CM |
3222 | struct btrfs_root *log; |
3223 | u64 index; | |
3224 | int ret; | |
3225 | ||
a491abb2 | 3226 | if (inode->logged_trans < trans->transid) |
3a5f1d45 CM |
3227 | return 0; |
3228 | ||
e02119d5 CM |
3229 | ret = join_running_log_trans(root); |
3230 | if (ret) | |
3231 | return 0; | |
3232 | log = root->log_root; | |
a491abb2 | 3233 | mutex_lock(&inode->log_mutex); |
e02119d5 | 3234 | |
a491abb2 | 3235 | ret = btrfs_del_inode_ref(trans, log, name, name_len, btrfs_ino(inode), |
e02119d5 | 3236 | dirid, &index); |
a491abb2 | 3237 | mutex_unlock(&inode->log_mutex); |
4a500fd1 | 3238 | if (ret == -ENOSPC) { |
0b246afa | 3239 | btrfs_set_log_full_commit(fs_info, trans); |
4a500fd1 | 3240 | ret = 0; |
79787eaa | 3241 | } else if (ret < 0 && ret != -ENOENT) |
66642832 | 3242 | btrfs_abort_transaction(trans, ret); |
12fcfd22 | 3243 | btrfs_end_log_trans(root); |
e02119d5 | 3244 | |
e02119d5 CM |
3245 | return ret; |
3246 | } | |
3247 | ||
3248 | /* | |
3249 | * creates a range item in the log for 'dirid'. first_offset and | |
3250 | * last_offset tell us which parts of the key space the log should | |
3251 | * be considered authoritative for. | |
3252 | */ | |
3253 | static noinline int insert_dir_log_key(struct btrfs_trans_handle *trans, | |
3254 | struct btrfs_root *log, | |
3255 | struct btrfs_path *path, | |
3256 | int key_type, u64 dirid, | |
3257 | u64 first_offset, u64 last_offset) | |
3258 | { | |
3259 | int ret; | |
3260 | struct btrfs_key key; | |
3261 | struct btrfs_dir_log_item *item; | |
3262 | ||
3263 | key.objectid = dirid; | |
3264 | key.offset = first_offset; | |
3265 | if (key_type == BTRFS_DIR_ITEM_KEY) | |
3266 | key.type = BTRFS_DIR_LOG_ITEM_KEY; | |
3267 | else | |
3268 | key.type = BTRFS_DIR_LOG_INDEX_KEY; | |
3269 | ret = btrfs_insert_empty_item(trans, log, path, &key, sizeof(*item)); | |
4a500fd1 YZ |
3270 | if (ret) |
3271 | return ret; | |
e02119d5 CM |
3272 | |
3273 | item = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
3274 | struct btrfs_dir_log_item); | |
3275 | btrfs_set_dir_log_end(path->nodes[0], item, last_offset); | |
3276 | btrfs_mark_buffer_dirty(path->nodes[0]); | |
b3b4aa74 | 3277 | btrfs_release_path(path); |
e02119d5 CM |
3278 | return 0; |
3279 | } | |
3280 | ||
3281 | /* | |
3282 | * log all the items included in the current transaction for a given | |
3283 | * directory. This also creates the range items in the log tree required | |
3284 | * to replay anything deleted before the fsync | |
3285 | */ | |
3286 | static noinline int log_dir_items(struct btrfs_trans_handle *trans, | |
684a5773 | 3287 | struct btrfs_root *root, struct btrfs_inode *inode, |
e02119d5 CM |
3288 | struct btrfs_path *path, |
3289 | struct btrfs_path *dst_path, int key_type, | |
2f2ff0ee | 3290 | struct btrfs_log_ctx *ctx, |
e02119d5 CM |
3291 | u64 min_offset, u64 *last_offset_ret) |
3292 | { | |
3293 | struct btrfs_key min_key; | |
e02119d5 CM |
3294 | struct btrfs_root *log = root->log_root; |
3295 | struct extent_buffer *src; | |
4a500fd1 | 3296 | int err = 0; |
e02119d5 CM |
3297 | int ret; |
3298 | int i; | |
3299 | int nritems; | |
3300 | u64 first_offset = min_offset; | |
3301 | u64 last_offset = (u64)-1; | |
684a5773 | 3302 | u64 ino = btrfs_ino(inode); |
e02119d5 CM |
3303 | |
3304 | log = root->log_root; | |
e02119d5 | 3305 | |
33345d01 | 3306 | min_key.objectid = ino; |
e02119d5 CM |
3307 | min_key.type = key_type; |
3308 | min_key.offset = min_offset; | |
3309 | ||
6174d3cb | 3310 | ret = btrfs_search_forward(root, &min_key, path, trans->transid); |
e02119d5 CM |
3311 | |
3312 | /* | |
3313 | * we didn't find anything from this transaction, see if there | |
3314 | * is anything at all | |
3315 | */ | |
33345d01 LZ |
3316 | if (ret != 0 || min_key.objectid != ino || min_key.type != key_type) { |
3317 | min_key.objectid = ino; | |
e02119d5 CM |
3318 | min_key.type = key_type; |
3319 | min_key.offset = (u64)-1; | |
b3b4aa74 | 3320 | btrfs_release_path(path); |
e02119d5 CM |
3321 | ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0); |
3322 | if (ret < 0) { | |
b3b4aa74 | 3323 | btrfs_release_path(path); |
e02119d5 CM |
3324 | return ret; |
3325 | } | |
33345d01 | 3326 | ret = btrfs_previous_item(root, path, ino, key_type); |
e02119d5 CM |
3327 | |
3328 | /* if ret == 0 there are items for this type, | |
3329 | * create a range to tell us the last key of this type. | |
3330 | * otherwise, there are no items in this directory after | |
3331 | * *min_offset, and we create a range to indicate that. | |
3332 | */ | |
3333 | if (ret == 0) { | |
3334 | struct btrfs_key tmp; | |
3335 | btrfs_item_key_to_cpu(path->nodes[0], &tmp, | |
3336 | path->slots[0]); | |
d397712b | 3337 | if (key_type == tmp.type) |
e02119d5 | 3338 | first_offset = max(min_offset, tmp.offset) + 1; |
e02119d5 CM |
3339 | } |
3340 | goto done; | |
3341 | } | |
3342 | ||
3343 | /* go backward to find any previous key */ | |
33345d01 | 3344 | ret = btrfs_previous_item(root, path, ino, key_type); |
e02119d5 CM |
3345 | if (ret == 0) { |
3346 | struct btrfs_key tmp; | |
3347 | btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]); | |
3348 | if (key_type == tmp.type) { | |
3349 | first_offset = tmp.offset; | |
3350 | ret = overwrite_item(trans, log, dst_path, | |
3351 | path->nodes[0], path->slots[0], | |
3352 | &tmp); | |
4a500fd1 YZ |
3353 | if (ret) { |
3354 | err = ret; | |
3355 | goto done; | |
3356 | } | |
e02119d5 CM |
3357 | } |
3358 | } | |
b3b4aa74 | 3359 | btrfs_release_path(path); |
e02119d5 CM |
3360 | |
3361 | /* find the first key from this transaction again */ | |
3362 | ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0); | |
fae7f21c | 3363 | if (WARN_ON(ret != 0)) |
e02119d5 | 3364 | goto done; |
e02119d5 CM |
3365 | |
3366 | /* | |
3367 | * we have a block from this transaction, log every item in it | |
3368 | * from our directory | |
3369 | */ | |
d397712b | 3370 | while (1) { |
e02119d5 CM |
3371 | struct btrfs_key tmp; |
3372 | src = path->nodes[0]; | |
3373 | nritems = btrfs_header_nritems(src); | |
3374 | for (i = path->slots[0]; i < nritems; i++) { | |
2f2ff0ee FM |
3375 | struct btrfs_dir_item *di; |
3376 | ||
e02119d5 CM |
3377 | btrfs_item_key_to_cpu(src, &min_key, i); |
3378 | ||
33345d01 | 3379 | if (min_key.objectid != ino || min_key.type != key_type) |
e02119d5 CM |
3380 | goto done; |
3381 | ret = overwrite_item(trans, log, dst_path, src, i, | |
3382 | &min_key); | |
4a500fd1 YZ |
3383 | if (ret) { |
3384 | err = ret; | |
3385 | goto done; | |
3386 | } | |
2f2ff0ee FM |
3387 | |
3388 | /* | |
3389 | * We must make sure that when we log a directory entry, | |
3390 | * the corresponding inode, after log replay, has a | |
3391 | * matching link count. For example: | |
3392 | * | |
3393 | * touch foo | |
3394 | * mkdir mydir | |
3395 | * sync | |
3396 | * ln foo mydir/bar | |
3397 | * xfs_io -c "fsync" mydir | |
3398 | * <crash> | |
3399 | * <mount fs and log replay> | |
3400 | * | |
3401 | * Would result in a fsync log that when replayed, our | |
3402 | * file inode would have a link count of 1, but we get | |
3403 | * two directory entries pointing to the same inode. | |
3404 | * After removing one of the names, it would not be | |
3405 | * possible to remove the other name, which resulted | |
3406 | * always in stale file handle errors, and would not | |
3407 | * be possible to rmdir the parent directory, since | |
3408 | * its i_size could never decrement to the value | |
3409 | * BTRFS_EMPTY_DIR_SIZE, resulting in -ENOTEMPTY errors. | |
3410 | */ | |
3411 | di = btrfs_item_ptr(src, i, struct btrfs_dir_item); | |
3412 | btrfs_dir_item_key_to_cpu(src, di, &tmp); | |
3413 | if (ctx && | |
3414 | (btrfs_dir_transid(src, di) == trans->transid || | |
3415 | btrfs_dir_type(src, di) == BTRFS_FT_DIR) && | |
3416 | tmp.type != BTRFS_ROOT_ITEM_KEY) | |
3417 | ctx->log_new_dentries = true; | |
e02119d5 CM |
3418 | } |
3419 | path->slots[0] = nritems; | |
3420 | ||
3421 | /* | |
3422 | * look ahead to the next item and see if it is also | |
3423 | * from this directory and from this transaction | |
3424 | */ | |
3425 | ret = btrfs_next_leaf(root, path); | |
3426 | if (ret == 1) { | |
3427 | last_offset = (u64)-1; | |
3428 | goto done; | |
3429 | } | |
3430 | btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]); | |
33345d01 | 3431 | if (tmp.objectid != ino || tmp.type != key_type) { |
e02119d5 CM |
3432 | last_offset = (u64)-1; |
3433 | goto done; | |
3434 | } | |
3435 | if (btrfs_header_generation(path->nodes[0]) != trans->transid) { | |
3436 | ret = overwrite_item(trans, log, dst_path, | |
3437 | path->nodes[0], path->slots[0], | |
3438 | &tmp); | |
4a500fd1 YZ |
3439 | if (ret) |
3440 | err = ret; | |
3441 | else | |
3442 | last_offset = tmp.offset; | |
e02119d5 CM |
3443 | goto done; |
3444 | } | |
3445 | } | |
3446 | done: | |
b3b4aa74 DS |
3447 | btrfs_release_path(path); |
3448 | btrfs_release_path(dst_path); | |
e02119d5 | 3449 | |
4a500fd1 YZ |
3450 | if (err == 0) { |
3451 | *last_offset_ret = last_offset; | |
3452 | /* | |
3453 | * insert the log range keys to indicate where the log | |
3454 | * is valid | |
3455 | */ | |
3456 | ret = insert_dir_log_key(trans, log, path, key_type, | |
33345d01 | 3457 | ino, first_offset, last_offset); |
4a500fd1 YZ |
3458 | if (ret) |
3459 | err = ret; | |
3460 | } | |
3461 | return err; | |
e02119d5 CM |
3462 | } |
3463 | ||
3464 | /* | |
3465 | * logging directories is very similar to logging inodes, We find all the items | |
3466 | * from the current transaction and write them to the log. | |
3467 | * | |
3468 | * The recovery code scans the directory in the subvolume, and if it finds a | |
3469 | * key in the range logged that is not present in the log tree, then it means | |
3470 | * that dir entry was unlinked during the transaction. | |
3471 | * | |
3472 | * In order for that scan to work, we must include one key smaller than | |
3473 | * the smallest logged by this transaction and one key larger than the largest | |
3474 | * key logged by this transaction. | |
3475 | */ | |
3476 | static noinline int log_directory_changes(struct btrfs_trans_handle *trans, | |
dbf39ea4 | 3477 | struct btrfs_root *root, struct btrfs_inode *inode, |
e02119d5 | 3478 | struct btrfs_path *path, |
2f2ff0ee FM |
3479 | struct btrfs_path *dst_path, |
3480 | struct btrfs_log_ctx *ctx) | |
e02119d5 CM |
3481 | { |
3482 | u64 min_key; | |
3483 | u64 max_key; | |
3484 | int ret; | |
3485 | int key_type = BTRFS_DIR_ITEM_KEY; | |
3486 | ||
3487 | again: | |
3488 | min_key = 0; | |
3489 | max_key = 0; | |
d397712b | 3490 | while (1) { |
dbf39ea4 NB |
3491 | ret = log_dir_items(trans, root, inode, path, dst_path, key_type, |
3492 | ctx, min_key, &max_key); | |
4a500fd1 YZ |
3493 | if (ret) |
3494 | return ret; | |
e02119d5 CM |
3495 | if (max_key == (u64)-1) |
3496 | break; | |
3497 | min_key = max_key + 1; | |
3498 | } | |
3499 | ||
3500 | if (key_type == BTRFS_DIR_ITEM_KEY) { | |
3501 | key_type = BTRFS_DIR_INDEX_KEY; | |
3502 | goto again; | |
3503 | } | |
3504 | return 0; | |
3505 | } | |
3506 | ||
3507 | /* | |
3508 | * a helper function to drop items from the log before we relog an | |
3509 | * inode. max_key_type indicates the highest item type to remove. | |
3510 | * This cannot be run for file data extents because it does not | |
3511 | * free the extents they point to. | |
3512 | */ | |
3513 | static int drop_objectid_items(struct btrfs_trans_handle *trans, | |
3514 | struct btrfs_root *log, | |
3515 | struct btrfs_path *path, | |
3516 | u64 objectid, int max_key_type) | |
3517 | { | |
3518 | int ret; | |
3519 | struct btrfs_key key; | |
3520 | struct btrfs_key found_key; | |
18ec90d6 | 3521 | int start_slot; |
e02119d5 CM |
3522 | |
3523 | key.objectid = objectid; | |
3524 | key.type = max_key_type; | |
3525 | key.offset = (u64)-1; | |
3526 | ||
d397712b | 3527 | while (1) { |
e02119d5 | 3528 | ret = btrfs_search_slot(trans, log, &key, path, -1, 1); |
3650860b | 3529 | BUG_ON(ret == 0); /* Logic error */ |
4a500fd1 | 3530 | if (ret < 0) |
e02119d5 CM |
3531 | break; |
3532 | ||
3533 | if (path->slots[0] == 0) | |
3534 | break; | |
3535 | ||
3536 | path->slots[0]--; | |
3537 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, | |
3538 | path->slots[0]); | |
3539 | ||
3540 | if (found_key.objectid != objectid) | |
3541 | break; | |
3542 | ||
18ec90d6 JB |
3543 | found_key.offset = 0; |
3544 | found_key.type = 0; | |
3545 | ret = btrfs_bin_search(path->nodes[0], &found_key, 0, | |
3546 | &start_slot); | |
3547 | ||
3548 | ret = btrfs_del_items(trans, log, path, start_slot, | |
3549 | path->slots[0] - start_slot + 1); | |
3550 | /* | |
3551 | * If start slot isn't 0 then we don't need to re-search, we've | |
3552 | * found the last guy with the objectid in this tree. | |
3553 | */ | |
3554 | if (ret || start_slot != 0) | |
65a246c5 | 3555 | break; |
b3b4aa74 | 3556 | btrfs_release_path(path); |
e02119d5 | 3557 | } |
b3b4aa74 | 3558 | btrfs_release_path(path); |
5bdbeb21 JB |
3559 | if (ret > 0) |
3560 | ret = 0; | |
4a500fd1 | 3561 | return ret; |
e02119d5 CM |
3562 | } |
3563 | ||
94edf4ae JB |
3564 | static void fill_inode_item(struct btrfs_trans_handle *trans, |
3565 | struct extent_buffer *leaf, | |
3566 | struct btrfs_inode_item *item, | |
1a4bcf47 FM |
3567 | struct inode *inode, int log_inode_only, |
3568 | u64 logged_isize) | |
94edf4ae | 3569 | { |
0b1c6cca JB |
3570 | struct btrfs_map_token token; |
3571 | ||
3572 | btrfs_init_map_token(&token); | |
94edf4ae JB |
3573 | |
3574 | if (log_inode_only) { | |
3575 | /* set the generation to zero so the recover code | |
3576 | * can tell the difference between an logging | |
3577 | * just to say 'this inode exists' and a logging | |
3578 | * to say 'update this inode with these values' | |
3579 | */ | |
0b1c6cca | 3580 | btrfs_set_token_inode_generation(leaf, item, 0, &token); |
1a4bcf47 | 3581 | btrfs_set_token_inode_size(leaf, item, logged_isize, &token); |
94edf4ae | 3582 | } else { |
0b1c6cca JB |
3583 | btrfs_set_token_inode_generation(leaf, item, |
3584 | BTRFS_I(inode)->generation, | |
3585 | &token); | |
3586 | btrfs_set_token_inode_size(leaf, item, inode->i_size, &token); | |
3587 | } | |
3588 | ||
3589 | btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token); | |
3590 | btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token); | |
3591 | btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token); | |
3592 | btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token); | |
3593 | ||
a937b979 | 3594 | btrfs_set_token_timespec_sec(leaf, &item->atime, |
0b1c6cca | 3595 | inode->i_atime.tv_sec, &token); |
a937b979 | 3596 | btrfs_set_token_timespec_nsec(leaf, &item->atime, |
0b1c6cca JB |
3597 | inode->i_atime.tv_nsec, &token); |
3598 | ||
a937b979 | 3599 | btrfs_set_token_timespec_sec(leaf, &item->mtime, |
0b1c6cca | 3600 | inode->i_mtime.tv_sec, &token); |
a937b979 | 3601 | btrfs_set_token_timespec_nsec(leaf, &item->mtime, |
0b1c6cca JB |
3602 | inode->i_mtime.tv_nsec, &token); |
3603 | ||
a937b979 | 3604 | btrfs_set_token_timespec_sec(leaf, &item->ctime, |
0b1c6cca | 3605 | inode->i_ctime.tv_sec, &token); |
a937b979 | 3606 | btrfs_set_token_timespec_nsec(leaf, &item->ctime, |
0b1c6cca JB |
3607 | inode->i_ctime.tv_nsec, &token); |
3608 | ||
3609 | btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode), | |
3610 | &token); | |
3611 | ||
3612 | btrfs_set_token_inode_sequence(leaf, item, inode->i_version, &token); | |
3613 | btrfs_set_token_inode_transid(leaf, item, trans->transid, &token); | |
3614 | btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token); | |
3615 | btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token); | |
3616 | btrfs_set_token_inode_block_group(leaf, item, 0, &token); | |
94edf4ae JB |
3617 | } |
3618 | ||
a95249b3 JB |
3619 | static int log_inode_item(struct btrfs_trans_handle *trans, |
3620 | struct btrfs_root *log, struct btrfs_path *path, | |
6d889a3b | 3621 | struct btrfs_inode *inode) |
a95249b3 JB |
3622 | { |
3623 | struct btrfs_inode_item *inode_item; | |
a95249b3 JB |
3624 | int ret; |
3625 | ||
efd0c405 | 3626 | ret = btrfs_insert_empty_item(trans, log, path, |
6d889a3b | 3627 | &inode->location, sizeof(*inode_item)); |
a95249b3 JB |
3628 | if (ret && ret != -EEXIST) |
3629 | return ret; | |
3630 | inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
3631 | struct btrfs_inode_item); | |
6d889a3b NB |
3632 | fill_inode_item(trans, path->nodes[0], inode_item, &inode->vfs_inode, |
3633 | 0, 0); | |
a95249b3 JB |
3634 | btrfs_release_path(path); |
3635 | return 0; | |
3636 | } | |
3637 | ||
31ff1cd2 | 3638 | static noinline int copy_items(struct btrfs_trans_handle *trans, |
44d70e19 | 3639 | struct btrfs_inode *inode, |
31ff1cd2 | 3640 | struct btrfs_path *dst_path, |
16e7549f | 3641 | struct btrfs_path *src_path, u64 *last_extent, |
1a4bcf47 FM |
3642 | int start_slot, int nr, int inode_only, |
3643 | u64 logged_isize) | |
31ff1cd2 | 3644 | { |
44d70e19 | 3645 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb); |
31ff1cd2 CM |
3646 | unsigned long src_offset; |
3647 | unsigned long dst_offset; | |
44d70e19 | 3648 | struct btrfs_root *log = inode->root->log_root; |
31ff1cd2 CM |
3649 | struct btrfs_file_extent_item *extent; |
3650 | struct btrfs_inode_item *inode_item; | |
16e7549f JB |
3651 | struct extent_buffer *src = src_path->nodes[0]; |
3652 | struct btrfs_key first_key, last_key, key; | |
31ff1cd2 CM |
3653 | int ret; |
3654 | struct btrfs_key *ins_keys; | |
3655 | u32 *ins_sizes; | |
3656 | char *ins_data; | |
3657 | int i; | |
d20f7043 | 3658 | struct list_head ordered_sums; |
44d70e19 | 3659 | int skip_csum = inode->flags & BTRFS_INODE_NODATASUM; |
16e7549f | 3660 | bool has_extents = false; |
74121f7c | 3661 | bool need_find_last_extent = true; |
16e7549f | 3662 | bool done = false; |
d20f7043 CM |
3663 | |
3664 | INIT_LIST_HEAD(&ordered_sums); | |
31ff1cd2 CM |
3665 | |
3666 | ins_data = kmalloc(nr * sizeof(struct btrfs_key) + | |
3667 | nr * sizeof(u32), GFP_NOFS); | |
2a29edc6 | 3668 | if (!ins_data) |
3669 | return -ENOMEM; | |
3670 | ||
16e7549f JB |
3671 | first_key.objectid = (u64)-1; |
3672 | ||
31ff1cd2 CM |
3673 | ins_sizes = (u32 *)ins_data; |
3674 | ins_keys = (struct btrfs_key *)(ins_data + nr * sizeof(u32)); | |
3675 | ||
3676 | for (i = 0; i < nr; i++) { | |
3677 | ins_sizes[i] = btrfs_item_size_nr(src, i + start_slot); | |
3678 | btrfs_item_key_to_cpu(src, ins_keys + i, i + start_slot); | |
3679 | } | |
3680 | ret = btrfs_insert_empty_items(trans, log, dst_path, | |
3681 | ins_keys, ins_sizes, nr); | |
4a500fd1 YZ |
3682 | if (ret) { |
3683 | kfree(ins_data); | |
3684 | return ret; | |
3685 | } | |
31ff1cd2 | 3686 | |
5d4f98a2 | 3687 | for (i = 0; i < nr; i++, dst_path->slots[0]++) { |
31ff1cd2 CM |
3688 | dst_offset = btrfs_item_ptr_offset(dst_path->nodes[0], |
3689 | dst_path->slots[0]); | |
3690 | ||
3691 | src_offset = btrfs_item_ptr_offset(src, start_slot + i); | |
3692 | ||
16e7549f JB |
3693 | if ((i == (nr - 1))) |
3694 | last_key = ins_keys[i]; | |
3695 | ||
94edf4ae | 3696 | if (ins_keys[i].type == BTRFS_INODE_ITEM_KEY) { |
31ff1cd2 CM |
3697 | inode_item = btrfs_item_ptr(dst_path->nodes[0], |
3698 | dst_path->slots[0], | |
3699 | struct btrfs_inode_item); | |
94edf4ae | 3700 | fill_inode_item(trans, dst_path->nodes[0], inode_item, |
f85b7379 DS |
3701 | &inode->vfs_inode, |
3702 | inode_only == LOG_INODE_EXISTS, | |
1a4bcf47 | 3703 | logged_isize); |
94edf4ae JB |
3704 | } else { |
3705 | copy_extent_buffer(dst_path->nodes[0], src, dst_offset, | |
3706 | src_offset, ins_sizes[i]); | |
31ff1cd2 | 3707 | } |
94edf4ae | 3708 | |
16e7549f JB |
3709 | /* |
3710 | * We set need_find_last_extent here in case we know we were | |
3711 | * processing other items and then walk into the first extent in | |
3712 | * the inode. If we don't hit an extent then nothing changes, | |
3713 | * we'll do the last search the next time around. | |
3714 | */ | |
3715 | if (ins_keys[i].type == BTRFS_EXTENT_DATA_KEY) { | |
3716 | has_extents = true; | |
74121f7c | 3717 | if (first_key.objectid == (u64)-1) |
16e7549f JB |
3718 | first_key = ins_keys[i]; |
3719 | } else { | |
3720 | need_find_last_extent = false; | |
3721 | } | |
3722 | ||
31ff1cd2 CM |
3723 | /* take a reference on file data extents so that truncates |
3724 | * or deletes of this inode don't have to relog the inode | |
3725 | * again | |
3726 | */ | |
962a298f | 3727 | if (ins_keys[i].type == BTRFS_EXTENT_DATA_KEY && |
d2794405 | 3728 | !skip_csum) { |
31ff1cd2 CM |
3729 | int found_type; |
3730 | extent = btrfs_item_ptr(src, start_slot + i, | |
3731 | struct btrfs_file_extent_item); | |
3732 | ||
8e531cdf | 3733 | if (btrfs_file_extent_generation(src, extent) < trans->transid) |
3734 | continue; | |
3735 | ||
31ff1cd2 | 3736 | found_type = btrfs_file_extent_type(src, extent); |
6f1fed77 | 3737 | if (found_type == BTRFS_FILE_EXTENT_REG) { |
5d4f98a2 YZ |
3738 | u64 ds, dl, cs, cl; |
3739 | ds = btrfs_file_extent_disk_bytenr(src, | |
3740 | extent); | |
3741 | /* ds == 0 is a hole */ | |
3742 | if (ds == 0) | |
3743 | continue; | |
3744 | ||
3745 | dl = btrfs_file_extent_disk_num_bytes(src, | |
3746 | extent); | |
3747 | cs = btrfs_file_extent_offset(src, extent); | |
3748 | cl = btrfs_file_extent_num_bytes(src, | |
a419aef8 | 3749 | extent); |
580afd76 CM |
3750 | if (btrfs_file_extent_compression(src, |
3751 | extent)) { | |
3752 | cs = 0; | |
3753 | cl = dl; | |
3754 | } | |
5d4f98a2 YZ |
3755 | |
3756 | ret = btrfs_lookup_csums_range( | |
0b246afa | 3757 | fs_info->csum_root, |
5d4f98a2 | 3758 | ds + cs, ds + cs + cl - 1, |
a2de733c | 3759 | &ordered_sums, 0); |
3650860b JB |
3760 | if (ret) { |
3761 | btrfs_release_path(dst_path); | |
3762 | kfree(ins_data); | |
3763 | return ret; | |
3764 | } | |
31ff1cd2 CM |
3765 | } |
3766 | } | |
31ff1cd2 CM |
3767 | } |
3768 | ||
3769 | btrfs_mark_buffer_dirty(dst_path->nodes[0]); | |
b3b4aa74 | 3770 | btrfs_release_path(dst_path); |
31ff1cd2 | 3771 | kfree(ins_data); |
d20f7043 CM |
3772 | |
3773 | /* | |
3774 | * we have to do this after the loop above to avoid changing the | |
3775 | * log tree while trying to change the log tree. | |
3776 | */ | |
4a500fd1 | 3777 | ret = 0; |
d397712b | 3778 | while (!list_empty(&ordered_sums)) { |
d20f7043 CM |
3779 | struct btrfs_ordered_sum *sums = list_entry(ordered_sums.next, |
3780 | struct btrfs_ordered_sum, | |
3781 | list); | |
4a500fd1 YZ |
3782 | if (!ret) |
3783 | ret = btrfs_csum_file_blocks(trans, log, sums); | |
d20f7043 CM |
3784 | list_del(&sums->list); |
3785 | kfree(sums); | |
3786 | } | |
16e7549f JB |
3787 | |
3788 | if (!has_extents) | |
3789 | return ret; | |
3790 | ||
74121f7c FM |
3791 | if (need_find_last_extent && *last_extent == first_key.offset) { |
3792 | /* | |
3793 | * We don't have any leafs between our current one and the one | |
3794 | * we processed before that can have file extent items for our | |
3795 | * inode (and have a generation number smaller than our current | |
3796 | * transaction id). | |
3797 | */ | |
3798 | need_find_last_extent = false; | |
3799 | } | |
3800 | ||
16e7549f JB |
3801 | /* |
3802 | * Because we use btrfs_search_forward we could skip leaves that were | |
3803 | * not modified and then assume *last_extent is valid when it really | |
3804 | * isn't. So back up to the previous leaf and read the end of the last | |
3805 | * extent before we go and fill in holes. | |
3806 | */ | |
3807 | if (need_find_last_extent) { | |
3808 | u64 len; | |
3809 | ||
44d70e19 | 3810 | ret = btrfs_prev_leaf(inode->root, src_path); |
16e7549f JB |
3811 | if (ret < 0) |
3812 | return ret; | |
3813 | if (ret) | |
3814 | goto fill_holes; | |
3815 | if (src_path->slots[0]) | |
3816 | src_path->slots[0]--; | |
3817 | src = src_path->nodes[0]; | |
3818 | btrfs_item_key_to_cpu(src, &key, src_path->slots[0]); | |
44d70e19 | 3819 | if (key.objectid != btrfs_ino(inode) || |
16e7549f JB |
3820 | key.type != BTRFS_EXTENT_DATA_KEY) |
3821 | goto fill_holes; | |
3822 | extent = btrfs_item_ptr(src, src_path->slots[0], | |
3823 | struct btrfs_file_extent_item); | |
3824 | if (btrfs_file_extent_type(src, extent) == | |
3825 | BTRFS_FILE_EXTENT_INLINE) { | |
514ac8ad CM |
3826 | len = btrfs_file_extent_inline_len(src, |
3827 | src_path->slots[0], | |
3828 | extent); | |
16e7549f | 3829 | *last_extent = ALIGN(key.offset + len, |
0b246afa | 3830 | fs_info->sectorsize); |
16e7549f JB |
3831 | } else { |
3832 | len = btrfs_file_extent_num_bytes(src, extent); | |
3833 | *last_extent = key.offset + len; | |
3834 | } | |
3835 | } | |
3836 | fill_holes: | |
3837 | /* So we did prev_leaf, now we need to move to the next leaf, but a few | |
3838 | * things could have happened | |
3839 | * | |
3840 | * 1) A merge could have happened, so we could currently be on a leaf | |
3841 | * that holds what we were copying in the first place. | |
3842 | * 2) A split could have happened, and now not all of the items we want | |
3843 | * are on the same leaf. | |
3844 | * | |
3845 | * So we need to adjust how we search for holes, we need to drop the | |
3846 | * path and re-search for the first extent key we found, and then walk | |
3847 | * forward until we hit the last one we copied. | |
3848 | */ | |
3849 | if (need_find_last_extent) { | |
3850 | /* btrfs_prev_leaf could return 1 without releasing the path */ | |
3851 | btrfs_release_path(src_path); | |
f85b7379 DS |
3852 | ret = btrfs_search_slot(NULL, inode->root, &first_key, |
3853 | src_path, 0, 0); | |
16e7549f JB |
3854 | if (ret < 0) |
3855 | return ret; | |
3856 | ASSERT(ret == 0); | |
3857 | src = src_path->nodes[0]; | |
3858 | i = src_path->slots[0]; | |
3859 | } else { | |
3860 | i = start_slot; | |
3861 | } | |
3862 | ||
3863 | /* | |
3864 | * Ok so here we need to go through and fill in any holes we may have | |
3865 | * to make sure that holes are punched for those areas in case they had | |
3866 | * extents previously. | |
3867 | */ | |
3868 | while (!done) { | |
3869 | u64 offset, len; | |
3870 | u64 extent_end; | |
3871 | ||
3872 | if (i >= btrfs_header_nritems(src_path->nodes[0])) { | |
44d70e19 | 3873 | ret = btrfs_next_leaf(inode->root, src_path); |
16e7549f JB |
3874 | if (ret < 0) |
3875 | return ret; | |
3876 | ASSERT(ret == 0); | |
3877 | src = src_path->nodes[0]; | |
3878 | i = 0; | |
3879 | } | |
3880 | ||
3881 | btrfs_item_key_to_cpu(src, &key, i); | |
3882 | if (!btrfs_comp_cpu_keys(&key, &last_key)) | |
3883 | done = true; | |
44d70e19 | 3884 | if (key.objectid != btrfs_ino(inode) || |
16e7549f JB |
3885 | key.type != BTRFS_EXTENT_DATA_KEY) { |
3886 | i++; | |
3887 | continue; | |
3888 | } | |
3889 | extent = btrfs_item_ptr(src, i, struct btrfs_file_extent_item); | |
3890 | if (btrfs_file_extent_type(src, extent) == | |
3891 | BTRFS_FILE_EXTENT_INLINE) { | |
514ac8ad | 3892 | len = btrfs_file_extent_inline_len(src, i, extent); |
da17066c | 3893 | extent_end = ALIGN(key.offset + len, |
0b246afa | 3894 | fs_info->sectorsize); |
16e7549f JB |
3895 | } else { |
3896 | len = btrfs_file_extent_num_bytes(src, extent); | |
3897 | extent_end = key.offset + len; | |
3898 | } | |
3899 | i++; | |
3900 | ||
3901 | if (*last_extent == key.offset) { | |
3902 | *last_extent = extent_end; | |
3903 | continue; | |
3904 | } | |
3905 | offset = *last_extent; | |
3906 | len = key.offset - *last_extent; | |
44d70e19 | 3907 | ret = btrfs_insert_file_extent(trans, log, btrfs_ino(inode), |
f85b7379 | 3908 | offset, 0, 0, len, 0, len, 0, 0, 0); |
16e7549f JB |
3909 | if (ret) |
3910 | break; | |
74121f7c | 3911 | *last_extent = extent_end; |
16e7549f JB |
3912 | } |
3913 | /* | |
3914 | * Need to let the callers know we dropped the path so they should | |
3915 | * re-search. | |
3916 | */ | |
3917 | if (!ret && need_find_last_extent) | |
3918 | ret = 1; | |
4a500fd1 | 3919 | return ret; |
31ff1cd2 CM |
3920 | } |
3921 | ||
5dc562c5 JB |
3922 | static int extent_cmp(void *priv, struct list_head *a, struct list_head *b) |
3923 | { | |
3924 | struct extent_map *em1, *em2; | |
3925 | ||
3926 | em1 = list_entry(a, struct extent_map, list); | |
3927 | em2 = list_entry(b, struct extent_map, list); | |
3928 | ||
3929 | if (em1->start < em2->start) | |
3930 | return -1; | |
3931 | else if (em1->start > em2->start) | |
3932 | return 1; | |
3933 | return 0; | |
3934 | } | |
3935 | ||
8407f553 FM |
3936 | static int wait_ordered_extents(struct btrfs_trans_handle *trans, |
3937 | struct inode *inode, | |
3938 | struct btrfs_root *root, | |
3939 | const struct extent_map *em, | |
3940 | const struct list_head *logged_list, | |
3941 | bool *ordered_io_error) | |
5dc562c5 | 3942 | { |
0b246afa | 3943 | struct btrfs_fs_info *fs_info = root->fs_info; |
2ab28f32 | 3944 | struct btrfs_ordered_extent *ordered; |
8407f553 | 3945 | struct btrfs_root *log = root->log_root; |
2ab28f32 JB |
3946 | u64 mod_start = em->mod_start; |
3947 | u64 mod_len = em->mod_len; | |
8407f553 | 3948 | const bool skip_csum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM; |
2ab28f32 JB |
3949 | u64 csum_offset; |
3950 | u64 csum_len; | |
8407f553 FM |
3951 | LIST_HEAD(ordered_sums); |
3952 | int ret = 0; | |
0aa4a17d | 3953 | |
8407f553 | 3954 | *ordered_io_error = false; |
0aa4a17d | 3955 | |
8407f553 FM |
3956 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) || |
3957 | em->block_start == EXTENT_MAP_HOLE) | |
70c8a91c | 3958 | return 0; |
5dc562c5 | 3959 | |
2ab28f32 | 3960 | /* |
8407f553 FM |
3961 | * Wait far any ordered extent that covers our extent map. If it |
3962 | * finishes without an error, first check and see if our csums are on | |
3963 | * our outstanding ordered extents. | |
2ab28f32 | 3964 | */ |
827463c4 | 3965 | list_for_each_entry(ordered, logged_list, log_list) { |
2ab28f32 JB |
3966 | struct btrfs_ordered_sum *sum; |
3967 | ||
3968 | if (!mod_len) | |
3969 | break; | |
3970 | ||
2ab28f32 JB |
3971 | if (ordered->file_offset + ordered->len <= mod_start || |
3972 | mod_start + mod_len <= ordered->file_offset) | |
3973 | continue; | |
3974 | ||
8407f553 FM |
3975 | if (!test_bit(BTRFS_ORDERED_IO_DONE, &ordered->flags) && |
3976 | !test_bit(BTRFS_ORDERED_IOERR, &ordered->flags) && | |
3977 | !test_bit(BTRFS_ORDERED_DIRECT, &ordered->flags)) { | |
3978 | const u64 start = ordered->file_offset; | |
3979 | const u64 end = ordered->file_offset + ordered->len - 1; | |
3980 | ||
3981 | WARN_ON(ordered->inode != inode); | |
3982 | filemap_fdatawrite_range(inode->i_mapping, start, end); | |
3983 | } | |
3984 | ||
3985 | wait_event(ordered->wait, | |
3986 | (test_bit(BTRFS_ORDERED_IO_DONE, &ordered->flags) || | |
3987 | test_bit(BTRFS_ORDERED_IOERR, &ordered->flags))); | |
3988 | ||
3989 | if (test_bit(BTRFS_ORDERED_IOERR, &ordered->flags)) { | |
b38ef71c FM |
3990 | /* |
3991 | * Clear the AS_EIO/AS_ENOSPC flags from the inode's | |
3992 | * i_mapping flags, so that the next fsync won't get | |
3993 | * an outdated io error too. | |
3994 | */ | |
f0312210 | 3995 | filemap_check_errors(inode->i_mapping); |
8407f553 FM |
3996 | *ordered_io_error = true; |
3997 | break; | |
3998 | } | |
2ab28f32 JB |
3999 | /* |
4000 | * We are going to copy all the csums on this ordered extent, so | |
4001 | * go ahead and adjust mod_start and mod_len in case this | |
4002 | * ordered extent has already been logged. | |
4003 | */ | |
4004 | if (ordered->file_offset > mod_start) { | |
4005 | if (ordered->file_offset + ordered->len >= | |
4006 | mod_start + mod_len) | |
4007 | mod_len = ordered->file_offset - mod_start; | |
4008 | /* | |
4009 | * If we have this case | |
4010 | * | |
4011 | * |--------- logged extent ---------| | |
4012 | * |----- ordered extent ----| | |
4013 | * | |
4014 | * Just don't mess with mod_start and mod_len, we'll | |
4015 | * just end up logging more csums than we need and it | |
4016 | * will be ok. | |
4017 | */ | |
4018 | } else { | |
4019 | if (ordered->file_offset + ordered->len < | |
4020 | mod_start + mod_len) { | |
4021 | mod_len = (mod_start + mod_len) - | |
4022 | (ordered->file_offset + ordered->len); | |
4023 | mod_start = ordered->file_offset + | |
4024 | ordered->len; | |
4025 | } else { | |
4026 | mod_len = 0; | |
4027 | } | |
4028 | } | |
4029 | ||
8407f553 FM |
4030 | if (skip_csum) |
4031 | continue; | |
4032 | ||
2ab28f32 JB |
4033 | /* |
4034 | * To keep us from looping for the above case of an ordered | |
4035 | * extent that falls inside of the logged extent. | |
4036 | */ | |
4037 | if (test_and_set_bit(BTRFS_ORDERED_LOGGED_CSUM, | |
4038 | &ordered->flags)) | |
4039 | continue; | |
2ab28f32 | 4040 | |
2ab28f32 JB |
4041 | list_for_each_entry(sum, &ordered->list, list) { |
4042 | ret = btrfs_csum_file_blocks(trans, log, sum); | |
827463c4 | 4043 | if (ret) |
8407f553 | 4044 | break; |
2ab28f32 | 4045 | } |
2ab28f32 | 4046 | } |
2ab28f32 | 4047 | |
8407f553 | 4048 | if (*ordered_io_error || !mod_len || ret || skip_csum) |
2ab28f32 JB |
4049 | return ret; |
4050 | ||
488111aa FDBM |
4051 | if (em->compress_type) { |
4052 | csum_offset = 0; | |
8407f553 | 4053 | csum_len = max(em->block_len, em->orig_block_len); |
488111aa FDBM |
4054 | } else { |
4055 | csum_offset = mod_start - em->start; | |
4056 | csum_len = mod_len; | |
4057 | } | |
2ab28f32 | 4058 | |
70c8a91c | 4059 | /* block start is already adjusted for the file extent offset. */ |
0b246afa | 4060 | ret = btrfs_lookup_csums_range(fs_info->csum_root, |
70c8a91c JB |
4061 | em->block_start + csum_offset, |
4062 | em->block_start + csum_offset + | |
4063 | csum_len - 1, &ordered_sums, 0); | |
4064 | if (ret) | |
4065 | return ret; | |
5dc562c5 | 4066 | |
70c8a91c JB |
4067 | while (!list_empty(&ordered_sums)) { |
4068 | struct btrfs_ordered_sum *sums = list_entry(ordered_sums.next, | |
4069 | struct btrfs_ordered_sum, | |
4070 | list); | |
4071 | if (!ret) | |
4072 | ret = btrfs_csum_file_blocks(trans, log, sums); | |
4073 | list_del(&sums->list); | |
4074 | kfree(sums); | |
5dc562c5 JB |
4075 | } |
4076 | ||
70c8a91c | 4077 | return ret; |
5dc562c5 JB |
4078 | } |
4079 | ||
8407f553 | 4080 | static int log_one_extent(struct btrfs_trans_handle *trans, |
9d122629 | 4081 | struct btrfs_inode *inode, struct btrfs_root *root, |
8407f553 FM |
4082 | const struct extent_map *em, |
4083 | struct btrfs_path *path, | |
4084 | const struct list_head *logged_list, | |
4085 | struct btrfs_log_ctx *ctx) | |
4086 | { | |
4087 | struct btrfs_root *log = root->log_root; | |
4088 | struct btrfs_file_extent_item *fi; | |
4089 | struct extent_buffer *leaf; | |
4090 | struct btrfs_map_token token; | |
4091 | struct btrfs_key key; | |
4092 | u64 extent_offset = em->start - em->orig_start; | |
4093 | u64 block_len; | |
4094 | int ret; | |
4095 | int extent_inserted = 0; | |
4096 | bool ordered_io_err = false; | |
4097 | ||
f85b7379 DS |
4098 | ret = wait_ordered_extents(trans, &inode->vfs_inode, root, em, |
4099 | logged_list, &ordered_io_err); | |
8407f553 FM |
4100 | if (ret) |
4101 | return ret; | |
4102 | ||
4103 | if (ordered_io_err) { | |
4104 | ctx->io_err = -EIO; | |
4105 | return 0; | |
4106 | } | |
4107 | ||
4108 | btrfs_init_map_token(&token); | |
4109 | ||
9d122629 | 4110 | ret = __btrfs_drop_extents(trans, log, &inode->vfs_inode, path, em->start, |
8407f553 FM |
4111 | em->start + em->len, NULL, 0, 1, |
4112 | sizeof(*fi), &extent_inserted); | |
4113 | if (ret) | |
4114 | return ret; | |
4115 | ||
4116 | if (!extent_inserted) { | |
9d122629 | 4117 | key.objectid = btrfs_ino(inode); |
8407f553 FM |
4118 | key.type = BTRFS_EXTENT_DATA_KEY; |
4119 | key.offset = em->start; | |
4120 | ||
4121 | ret = btrfs_insert_empty_item(trans, log, path, &key, | |
4122 | sizeof(*fi)); | |
4123 | if (ret) | |
4124 | return ret; | |
4125 | } | |
4126 | leaf = path->nodes[0]; | |
4127 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
4128 | struct btrfs_file_extent_item); | |
4129 | ||
50d9aa99 | 4130 | btrfs_set_token_file_extent_generation(leaf, fi, trans->transid, |
8407f553 FM |
4131 | &token); |
4132 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) | |
4133 | btrfs_set_token_file_extent_type(leaf, fi, | |
4134 | BTRFS_FILE_EXTENT_PREALLOC, | |
4135 | &token); | |
4136 | else | |
4137 | btrfs_set_token_file_extent_type(leaf, fi, | |
4138 | BTRFS_FILE_EXTENT_REG, | |
4139 | &token); | |
4140 | ||
4141 | block_len = max(em->block_len, em->orig_block_len); | |
4142 | if (em->compress_type != BTRFS_COMPRESS_NONE) { | |
4143 | btrfs_set_token_file_extent_disk_bytenr(leaf, fi, | |
4144 | em->block_start, | |
4145 | &token); | |
4146 | btrfs_set_token_file_extent_disk_num_bytes(leaf, fi, block_len, | |
4147 | &token); | |
4148 | } else if (em->block_start < EXTENT_MAP_LAST_BYTE) { | |
4149 | btrfs_set_token_file_extent_disk_bytenr(leaf, fi, | |
4150 | em->block_start - | |
4151 | extent_offset, &token); | |
4152 | btrfs_set_token_file_extent_disk_num_bytes(leaf, fi, block_len, | |
4153 | &token); | |
4154 | } else { | |
4155 | btrfs_set_token_file_extent_disk_bytenr(leaf, fi, 0, &token); | |
4156 | btrfs_set_token_file_extent_disk_num_bytes(leaf, fi, 0, | |
4157 | &token); | |
4158 | } | |
4159 | ||
4160 | btrfs_set_token_file_extent_offset(leaf, fi, extent_offset, &token); | |
4161 | btrfs_set_token_file_extent_num_bytes(leaf, fi, em->len, &token); | |
4162 | btrfs_set_token_file_extent_ram_bytes(leaf, fi, em->ram_bytes, &token); | |
4163 | btrfs_set_token_file_extent_compression(leaf, fi, em->compress_type, | |
4164 | &token); | |
4165 | btrfs_set_token_file_extent_encryption(leaf, fi, 0, &token); | |
4166 | btrfs_set_token_file_extent_other_encoding(leaf, fi, 0, &token); | |
4167 | btrfs_mark_buffer_dirty(leaf); | |
4168 | ||
4169 | btrfs_release_path(path); | |
4170 | ||
4171 | return ret; | |
4172 | } | |
4173 | ||
5dc562c5 JB |
4174 | static int btrfs_log_changed_extents(struct btrfs_trans_handle *trans, |
4175 | struct btrfs_root *root, | |
9d122629 | 4176 | struct btrfs_inode *inode, |
827463c4 | 4177 | struct btrfs_path *path, |
8407f553 | 4178 | struct list_head *logged_list, |
de0ee0ed FM |
4179 | struct btrfs_log_ctx *ctx, |
4180 | const u64 start, | |
4181 | const u64 end) | |
5dc562c5 | 4182 | { |
5dc562c5 JB |
4183 | struct extent_map *em, *n; |
4184 | struct list_head extents; | |
9d122629 | 4185 | struct extent_map_tree *tree = &inode->extent_tree; |
5dc562c5 JB |
4186 | u64 test_gen; |
4187 | int ret = 0; | |
2ab28f32 | 4188 | int num = 0; |
5dc562c5 JB |
4189 | |
4190 | INIT_LIST_HEAD(&extents); | |
4191 | ||
9d122629 | 4192 | down_write(&inode->dio_sem); |
5dc562c5 JB |
4193 | write_lock(&tree->lock); |
4194 | test_gen = root->fs_info->last_trans_committed; | |
4195 | ||
4196 | list_for_each_entry_safe(em, n, &tree->modified_extents, list) { | |
4197 | list_del_init(&em->list); | |
2ab28f32 JB |
4198 | |
4199 | /* | |
4200 | * Just an arbitrary number, this can be really CPU intensive | |
4201 | * once we start getting a lot of extents, and really once we | |
4202 | * have a bunch of extents we just want to commit since it will | |
4203 | * be faster. | |
4204 | */ | |
4205 | if (++num > 32768) { | |
4206 | list_del_init(&tree->modified_extents); | |
4207 | ret = -EFBIG; | |
4208 | goto process; | |
4209 | } | |
4210 | ||
5dc562c5 JB |
4211 | if (em->generation <= test_gen) |
4212 | continue; | |
ff44c6e3 | 4213 | /* Need a ref to keep it from getting evicted from cache */ |
490b54d6 | 4214 | refcount_inc(&em->refs); |
ff44c6e3 | 4215 | set_bit(EXTENT_FLAG_LOGGING, &em->flags); |
5dc562c5 | 4216 | list_add_tail(&em->list, &extents); |
2ab28f32 | 4217 | num++; |
5dc562c5 JB |
4218 | } |
4219 | ||
4220 | list_sort(NULL, &extents, extent_cmp); | |
9d122629 | 4221 | btrfs_get_logged_extents(inode, logged_list, start, end); |
de0ee0ed | 4222 | /* |
5f9a8a51 FM |
4223 | * Some ordered extents started by fsync might have completed |
4224 | * before we could collect them into the list logged_list, which | |
4225 | * means they're gone, not in our logged_list nor in the inode's | |
4226 | * ordered tree. We want the application/user space to know an | |
4227 | * error happened while attempting to persist file data so that | |
4228 | * it can take proper action. If such error happened, we leave | |
4229 | * without writing to the log tree and the fsync must report the | |
4230 | * file data write error and not commit the current transaction. | |
de0ee0ed | 4231 | */ |
9d122629 | 4232 | ret = filemap_check_errors(inode->vfs_inode.i_mapping); |
5f9a8a51 FM |
4233 | if (ret) |
4234 | ctx->io_err = ret; | |
2ab28f32 | 4235 | process: |
5dc562c5 JB |
4236 | while (!list_empty(&extents)) { |
4237 | em = list_entry(extents.next, struct extent_map, list); | |
4238 | ||
4239 | list_del_init(&em->list); | |
4240 | ||
4241 | /* | |
4242 | * If we had an error we just need to delete everybody from our | |
4243 | * private list. | |
4244 | */ | |
ff44c6e3 | 4245 | if (ret) { |
201a9038 | 4246 | clear_em_logging(tree, em); |
ff44c6e3 | 4247 | free_extent_map(em); |
5dc562c5 | 4248 | continue; |
ff44c6e3 JB |
4249 | } |
4250 | ||
4251 | write_unlock(&tree->lock); | |
5dc562c5 | 4252 | |
8407f553 FM |
4253 | ret = log_one_extent(trans, inode, root, em, path, logged_list, |
4254 | ctx); | |
ff44c6e3 | 4255 | write_lock(&tree->lock); |
201a9038 JB |
4256 | clear_em_logging(tree, em); |
4257 | free_extent_map(em); | |
5dc562c5 | 4258 | } |
ff44c6e3 JB |
4259 | WARN_ON(!list_empty(&extents)); |
4260 | write_unlock(&tree->lock); | |
9d122629 | 4261 | up_write(&inode->dio_sem); |
5dc562c5 | 4262 | |
5dc562c5 | 4263 | btrfs_release_path(path); |
5dc562c5 JB |
4264 | return ret; |
4265 | } | |
4266 | ||
481b01c0 | 4267 | static int logged_inode_size(struct btrfs_root *log, struct btrfs_inode *inode, |
1a4bcf47 FM |
4268 | struct btrfs_path *path, u64 *size_ret) |
4269 | { | |
4270 | struct btrfs_key key; | |
4271 | int ret; | |
4272 | ||
481b01c0 | 4273 | key.objectid = btrfs_ino(inode); |
1a4bcf47 FM |
4274 | key.type = BTRFS_INODE_ITEM_KEY; |
4275 | key.offset = 0; | |
4276 | ||
4277 | ret = btrfs_search_slot(NULL, log, &key, path, 0, 0); | |
4278 | if (ret < 0) { | |
4279 | return ret; | |
4280 | } else if (ret > 0) { | |
2f2ff0ee | 4281 | *size_ret = 0; |
1a4bcf47 FM |
4282 | } else { |
4283 | struct btrfs_inode_item *item; | |
4284 | ||
4285 | item = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
4286 | struct btrfs_inode_item); | |
4287 | *size_ret = btrfs_inode_size(path->nodes[0], item); | |
4288 | } | |
4289 | ||
4290 | btrfs_release_path(path); | |
4291 | return 0; | |
4292 | } | |
4293 | ||
36283bf7 FM |
4294 | /* |
4295 | * At the moment we always log all xattrs. This is to figure out at log replay | |
4296 | * time which xattrs must have their deletion replayed. If a xattr is missing | |
4297 | * in the log tree and exists in the fs/subvol tree, we delete it. This is | |
4298 | * because if a xattr is deleted, the inode is fsynced and a power failure | |
4299 | * happens, causing the log to be replayed the next time the fs is mounted, | |
4300 | * we want the xattr to not exist anymore (same behaviour as other filesystems | |
4301 | * with a journal, ext3/4, xfs, f2fs, etc). | |
4302 | */ | |
4303 | static int btrfs_log_all_xattrs(struct btrfs_trans_handle *trans, | |
4304 | struct btrfs_root *root, | |
1a93c36a | 4305 | struct btrfs_inode *inode, |
36283bf7 FM |
4306 | struct btrfs_path *path, |
4307 | struct btrfs_path *dst_path) | |
4308 | { | |
4309 | int ret; | |
4310 | struct btrfs_key key; | |
1a93c36a | 4311 | const u64 ino = btrfs_ino(inode); |
36283bf7 FM |
4312 | int ins_nr = 0; |
4313 | int start_slot = 0; | |
4314 | ||
4315 | key.objectid = ino; | |
4316 | key.type = BTRFS_XATTR_ITEM_KEY; | |
4317 | key.offset = 0; | |
4318 | ||
4319 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
4320 | if (ret < 0) | |
4321 | return ret; | |
4322 | ||
4323 | while (true) { | |
4324 | int slot = path->slots[0]; | |
4325 | struct extent_buffer *leaf = path->nodes[0]; | |
4326 | int nritems = btrfs_header_nritems(leaf); | |
4327 | ||
4328 | if (slot >= nritems) { | |
4329 | if (ins_nr > 0) { | |
4330 | u64 last_extent = 0; | |
4331 | ||
1a93c36a | 4332 | ret = copy_items(trans, inode, dst_path, path, |
36283bf7 FM |
4333 | &last_extent, start_slot, |
4334 | ins_nr, 1, 0); | |
4335 | /* can't be 1, extent items aren't processed */ | |
4336 | ASSERT(ret <= 0); | |
4337 | if (ret < 0) | |
4338 | return ret; | |
4339 | ins_nr = 0; | |
4340 | } | |
4341 | ret = btrfs_next_leaf(root, path); | |
4342 | if (ret < 0) | |
4343 | return ret; | |
4344 | else if (ret > 0) | |
4345 | break; | |
4346 | continue; | |
4347 | } | |
4348 | ||
4349 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
4350 | if (key.objectid != ino || key.type != BTRFS_XATTR_ITEM_KEY) | |
4351 | break; | |
4352 | ||
4353 | if (ins_nr == 0) | |
4354 | start_slot = slot; | |
4355 | ins_nr++; | |
4356 | path->slots[0]++; | |
4357 | cond_resched(); | |
4358 | } | |
4359 | if (ins_nr > 0) { | |
4360 | u64 last_extent = 0; | |
4361 | ||
1a93c36a | 4362 | ret = copy_items(trans, inode, dst_path, path, |
36283bf7 FM |
4363 | &last_extent, start_slot, |
4364 | ins_nr, 1, 0); | |
4365 | /* can't be 1, extent items aren't processed */ | |
4366 | ASSERT(ret <= 0); | |
4367 | if (ret < 0) | |
4368 | return ret; | |
4369 | } | |
4370 | ||
4371 | return 0; | |
4372 | } | |
4373 | ||
a89ca6f2 FM |
4374 | /* |
4375 | * If the no holes feature is enabled we need to make sure any hole between the | |
4376 | * last extent and the i_size of our inode is explicitly marked in the log. This | |
4377 | * is to make sure that doing something like: | |
4378 | * | |
4379 | * 1) create file with 128Kb of data | |
4380 | * 2) truncate file to 64Kb | |
4381 | * 3) truncate file to 256Kb | |
4382 | * 4) fsync file | |
4383 | * 5) <crash/power failure> | |
4384 | * 6) mount fs and trigger log replay | |
4385 | * | |
4386 | * Will give us a file with a size of 256Kb, the first 64Kb of data match what | |
4387 | * the file had in its first 64Kb of data at step 1 and the last 192Kb of the | |
4388 | * file correspond to a hole. The presence of explicit holes in a log tree is | |
4389 | * what guarantees that log replay will remove/adjust file extent items in the | |
4390 | * fs/subvol tree. | |
4391 | * | |
4392 | * Here we do not need to care about holes between extents, that is already done | |
4393 | * by copy_items(). We also only need to do this in the full sync path, where we | |
4394 | * lookup for extents from the fs/subvol tree only. In the fast path case, we | |
4395 | * lookup the list of modified extent maps and if any represents a hole, we | |
4396 | * insert a corresponding extent representing a hole in the log tree. | |
4397 | */ | |
4398 | static int btrfs_log_trailing_hole(struct btrfs_trans_handle *trans, | |
4399 | struct btrfs_root *root, | |
a0308dd7 | 4400 | struct btrfs_inode *inode, |
a89ca6f2 FM |
4401 | struct btrfs_path *path) |
4402 | { | |
0b246afa | 4403 | struct btrfs_fs_info *fs_info = root->fs_info; |
a89ca6f2 FM |
4404 | int ret; |
4405 | struct btrfs_key key; | |
4406 | u64 hole_start; | |
4407 | u64 hole_size; | |
4408 | struct extent_buffer *leaf; | |
4409 | struct btrfs_root *log = root->log_root; | |
a0308dd7 NB |
4410 | const u64 ino = btrfs_ino(inode); |
4411 | const u64 i_size = i_size_read(&inode->vfs_inode); | |
a89ca6f2 | 4412 | |
0b246afa | 4413 | if (!btrfs_fs_incompat(fs_info, NO_HOLES)) |
a89ca6f2 FM |
4414 | return 0; |
4415 | ||
4416 | key.objectid = ino; | |
4417 | key.type = BTRFS_EXTENT_DATA_KEY; | |
4418 | key.offset = (u64)-1; | |
4419 | ||
4420 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
4421 | ASSERT(ret != 0); | |
4422 | if (ret < 0) | |
4423 | return ret; | |
4424 | ||
4425 | ASSERT(path->slots[0] > 0); | |
4426 | path->slots[0]--; | |
4427 | leaf = path->nodes[0]; | |
4428 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
4429 | ||
4430 | if (key.objectid != ino || key.type != BTRFS_EXTENT_DATA_KEY) { | |
4431 | /* inode does not have any extents */ | |
4432 | hole_start = 0; | |
4433 | hole_size = i_size; | |
4434 | } else { | |
4435 | struct btrfs_file_extent_item *extent; | |
4436 | u64 len; | |
4437 | ||
4438 | /* | |
4439 | * If there's an extent beyond i_size, an explicit hole was | |
4440 | * already inserted by copy_items(). | |
4441 | */ | |
4442 | if (key.offset >= i_size) | |
4443 | return 0; | |
4444 | ||
4445 | extent = btrfs_item_ptr(leaf, path->slots[0], | |
4446 | struct btrfs_file_extent_item); | |
4447 | ||
4448 | if (btrfs_file_extent_type(leaf, extent) == | |
4449 | BTRFS_FILE_EXTENT_INLINE) { | |
4450 | len = btrfs_file_extent_inline_len(leaf, | |
4451 | path->slots[0], | |
4452 | extent); | |
4453 | ASSERT(len == i_size); | |
4454 | return 0; | |
4455 | } | |
4456 | ||
4457 | len = btrfs_file_extent_num_bytes(leaf, extent); | |
4458 | /* Last extent goes beyond i_size, no need to log a hole. */ | |
4459 | if (key.offset + len > i_size) | |
4460 | return 0; | |
4461 | hole_start = key.offset + len; | |
4462 | hole_size = i_size - hole_start; | |
4463 | } | |
4464 | btrfs_release_path(path); | |
4465 | ||
4466 | /* Last extent ends at i_size. */ | |
4467 | if (hole_size == 0) | |
4468 | return 0; | |
4469 | ||
0b246afa | 4470 | hole_size = ALIGN(hole_size, fs_info->sectorsize); |
a89ca6f2 FM |
4471 | ret = btrfs_insert_file_extent(trans, log, ino, hole_start, 0, 0, |
4472 | hole_size, 0, hole_size, 0, 0, 0); | |
4473 | return ret; | |
4474 | } | |
4475 | ||
56f23fdb FM |
4476 | /* |
4477 | * When we are logging a new inode X, check if it doesn't have a reference that | |
4478 | * matches the reference from some other inode Y created in a past transaction | |
4479 | * and that was renamed in the current transaction. If we don't do this, then at | |
4480 | * log replay time we can lose inode Y (and all its files if it's a directory): | |
4481 | * | |
4482 | * mkdir /mnt/x | |
4483 | * echo "hello world" > /mnt/x/foobar | |
4484 | * sync | |
4485 | * mv /mnt/x /mnt/y | |
4486 | * mkdir /mnt/x # or touch /mnt/x | |
4487 | * xfs_io -c fsync /mnt/x | |
4488 | * <power fail> | |
4489 | * mount fs, trigger log replay | |
4490 | * | |
4491 | * After the log replay procedure, we would lose the first directory and all its | |
4492 | * files (file foobar). | |
4493 | * For the case where inode Y is not a directory we simply end up losing it: | |
4494 | * | |
4495 | * echo "123" > /mnt/foo | |
4496 | * sync | |
4497 | * mv /mnt/foo /mnt/bar | |
4498 | * echo "abc" > /mnt/foo | |
4499 | * xfs_io -c fsync /mnt/foo | |
4500 | * <power fail> | |
4501 | * | |
4502 | * We also need this for cases where a snapshot entry is replaced by some other | |
4503 | * entry (file or directory) otherwise we end up with an unreplayable log due to | |
4504 | * attempts to delete the snapshot entry (entry of type BTRFS_ROOT_ITEM_KEY) as | |
4505 | * if it were a regular entry: | |
4506 | * | |
4507 | * mkdir /mnt/x | |
4508 | * btrfs subvolume snapshot /mnt /mnt/x/snap | |
4509 | * btrfs subvolume delete /mnt/x/snap | |
4510 | * rmdir /mnt/x | |
4511 | * mkdir /mnt/x | |
4512 | * fsync /mnt/x or fsync some new file inside it | |
4513 | * <power fail> | |
4514 | * | |
4515 | * The snapshot delete, rmdir of x, mkdir of a new x and the fsync all happen in | |
4516 | * the same transaction. | |
4517 | */ | |
4518 | static int btrfs_check_ref_name_override(struct extent_buffer *eb, | |
4519 | const int slot, | |
4520 | const struct btrfs_key *key, | |
4791c8f1 | 4521 | struct btrfs_inode *inode, |
44f714da | 4522 | u64 *other_ino) |
56f23fdb FM |
4523 | { |
4524 | int ret; | |
4525 | struct btrfs_path *search_path; | |
4526 | char *name = NULL; | |
4527 | u32 name_len = 0; | |
4528 | u32 item_size = btrfs_item_size_nr(eb, slot); | |
4529 | u32 cur_offset = 0; | |
4530 | unsigned long ptr = btrfs_item_ptr_offset(eb, slot); | |
4531 | ||
4532 | search_path = btrfs_alloc_path(); | |
4533 | if (!search_path) | |
4534 | return -ENOMEM; | |
4535 | search_path->search_commit_root = 1; | |
4536 | search_path->skip_locking = 1; | |
4537 | ||
4538 | while (cur_offset < item_size) { | |
4539 | u64 parent; | |
4540 | u32 this_name_len; | |
4541 | u32 this_len; | |
4542 | unsigned long name_ptr; | |
4543 | struct btrfs_dir_item *di; | |
4544 | ||
4545 | if (key->type == BTRFS_INODE_REF_KEY) { | |
4546 | struct btrfs_inode_ref *iref; | |
4547 | ||
4548 | iref = (struct btrfs_inode_ref *)(ptr + cur_offset); | |
4549 | parent = key->offset; | |
4550 | this_name_len = btrfs_inode_ref_name_len(eb, iref); | |
4551 | name_ptr = (unsigned long)(iref + 1); | |
4552 | this_len = sizeof(*iref) + this_name_len; | |
4553 | } else { | |
4554 | struct btrfs_inode_extref *extref; | |
4555 | ||
4556 | extref = (struct btrfs_inode_extref *)(ptr + | |
4557 | cur_offset); | |
4558 | parent = btrfs_inode_extref_parent(eb, extref); | |
4559 | this_name_len = btrfs_inode_extref_name_len(eb, extref); | |
4560 | name_ptr = (unsigned long)&extref->name; | |
4561 | this_len = sizeof(*extref) + this_name_len; | |
4562 | } | |
4563 | ||
3c1d4184 SY |
4564 | ret = btrfs_is_name_len_valid(eb, slot, name_ptr, |
4565 | this_name_len); | |
4566 | if (!ret) { | |
4567 | ret = -EIO; | |
4568 | goto out; | |
4569 | } | |
56f23fdb FM |
4570 | if (this_name_len > name_len) { |
4571 | char *new_name; | |
4572 | ||
4573 | new_name = krealloc(name, this_name_len, GFP_NOFS); | |
4574 | if (!new_name) { | |
4575 | ret = -ENOMEM; | |
4576 | goto out; | |
4577 | } | |
4578 | name_len = this_name_len; | |
4579 | name = new_name; | |
4580 | } | |
4581 | ||
4582 | read_extent_buffer(eb, name, name_ptr, this_name_len); | |
4791c8f1 NB |
4583 | di = btrfs_lookup_dir_item(NULL, inode->root, search_path, |
4584 | parent, name, this_name_len, 0); | |
56f23fdb | 4585 | if (di && !IS_ERR(di)) { |
44f714da FM |
4586 | struct btrfs_key di_key; |
4587 | ||
4588 | btrfs_dir_item_key_to_cpu(search_path->nodes[0], | |
4589 | di, &di_key); | |
4590 | if (di_key.type == BTRFS_INODE_ITEM_KEY) { | |
4591 | ret = 1; | |
4592 | *other_ino = di_key.objectid; | |
4593 | } else { | |
4594 | ret = -EAGAIN; | |
4595 | } | |
56f23fdb FM |
4596 | goto out; |
4597 | } else if (IS_ERR(di)) { | |
4598 | ret = PTR_ERR(di); | |
4599 | goto out; | |
4600 | } | |
4601 | btrfs_release_path(search_path); | |
4602 | ||
4603 | cur_offset += this_len; | |
4604 | } | |
4605 | ret = 0; | |
4606 | out: | |
4607 | btrfs_free_path(search_path); | |
4608 | kfree(name); | |
4609 | return ret; | |
4610 | } | |
4611 | ||
e02119d5 CM |
4612 | /* log a single inode in the tree log. |
4613 | * At least one parent directory for this inode must exist in the tree | |
4614 | * or be logged already. | |
4615 | * | |
4616 | * Any items from this inode changed by the current transaction are copied | |
4617 | * to the log tree. An extra reference is taken on any extents in this | |
4618 | * file, allowing us to avoid a whole pile of corner cases around logging | |
4619 | * blocks that have been removed from the tree. | |
4620 | * | |
4621 | * See LOG_INODE_ALL and related defines for a description of what inode_only | |
4622 | * does. | |
4623 | * | |
4624 | * This handles both files and directories. | |
4625 | */ | |
12fcfd22 | 4626 | static int btrfs_log_inode(struct btrfs_trans_handle *trans, |
a59108a7 | 4627 | struct btrfs_root *root, struct btrfs_inode *inode, |
49dae1bc FM |
4628 | int inode_only, |
4629 | const loff_t start, | |
8407f553 FM |
4630 | const loff_t end, |
4631 | struct btrfs_log_ctx *ctx) | |
e02119d5 | 4632 | { |
0b246afa | 4633 | struct btrfs_fs_info *fs_info = root->fs_info; |
e02119d5 CM |
4634 | struct btrfs_path *path; |
4635 | struct btrfs_path *dst_path; | |
4636 | struct btrfs_key min_key; | |
4637 | struct btrfs_key max_key; | |
4638 | struct btrfs_root *log = root->log_root; | |
31ff1cd2 | 4639 | struct extent_buffer *src = NULL; |
827463c4 | 4640 | LIST_HEAD(logged_list); |
16e7549f | 4641 | u64 last_extent = 0; |
4a500fd1 | 4642 | int err = 0; |
e02119d5 | 4643 | int ret; |
3a5f1d45 | 4644 | int nritems; |
31ff1cd2 CM |
4645 | int ins_start_slot = 0; |
4646 | int ins_nr; | |
5dc562c5 | 4647 | bool fast_search = false; |
a59108a7 NB |
4648 | u64 ino = btrfs_ino(inode); |
4649 | struct extent_map_tree *em_tree = &inode->extent_tree; | |
1a4bcf47 | 4650 | u64 logged_isize = 0; |
e4545de5 | 4651 | bool need_log_inode_item = true; |
e02119d5 | 4652 | |
e02119d5 | 4653 | path = btrfs_alloc_path(); |
5df67083 TI |
4654 | if (!path) |
4655 | return -ENOMEM; | |
e02119d5 | 4656 | dst_path = btrfs_alloc_path(); |
5df67083 TI |
4657 | if (!dst_path) { |
4658 | btrfs_free_path(path); | |
4659 | return -ENOMEM; | |
4660 | } | |
e02119d5 | 4661 | |
33345d01 | 4662 | min_key.objectid = ino; |
e02119d5 CM |
4663 | min_key.type = BTRFS_INODE_ITEM_KEY; |
4664 | min_key.offset = 0; | |
4665 | ||
33345d01 | 4666 | max_key.objectid = ino; |
12fcfd22 | 4667 | |
12fcfd22 | 4668 | |
5dc562c5 | 4669 | /* today the code can only do partial logging of directories */ |
a59108a7 | 4670 | if (S_ISDIR(inode->vfs_inode.i_mode) || |
5269b67e | 4671 | (!test_bit(BTRFS_INODE_NEEDS_FULL_SYNC, |
a59108a7 | 4672 | &inode->runtime_flags) && |
781feef7 | 4673 | inode_only >= LOG_INODE_EXISTS)) |
e02119d5 CM |
4674 | max_key.type = BTRFS_XATTR_ITEM_KEY; |
4675 | else | |
4676 | max_key.type = (u8)-1; | |
4677 | max_key.offset = (u64)-1; | |
4678 | ||
2c2c452b FM |
4679 | /* |
4680 | * Only run delayed items if we are a dir or a new file. | |
4681 | * Otherwise commit the delayed inode only, which is needed in | |
4682 | * order for the log replay code to mark inodes for link count | |
4683 | * fixup (create temporary BTRFS_TREE_LOG_FIXUP_OBJECTID items). | |
4684 | */ | |
a59108a7 NB |
4685 | if (S_ISDIR(inode->vfs_inode.i_mode) || |
4686 | inode->generation > fs_info->last_trans_committed) | |
4687 | ret = btrfs_commit_inode_delayed_items(trans, inode); | |
2c2c452b | 4688 | else |
a59108a7 | 4689 | ret = btrfs_commit_inode_delayed_inode(inode); |
2c2c452b FM |
4690 | |
4691 | if (ret) { | |
4692 | btrfs_free_path(path); | |
4693 | btrfs_free_path(dst_path); | |
4694 | return ret; | |
16cdcec7 MX |
4695 | } |
4696 | ||
781feef7 LB |
4697 | if (inode_only == LOG_OTHER_INODE) { |
4698 | inode_only = LOG_INODE_EXISTS; | |
a59108a7 | 4699 | mutex_lock_nested(&inode->log_mutex, SINGLE_DEPTH_NESTING); |
781feef7 | 4700 | } else { |
a59108a7 | 4701 | mutex_lock(&inode->log_mutex); |
781feef7 | 4702 | } |
e02119d5 CM |
4703 | |
4704 | /* | |
4705 | * a brute force approach to making sure we get the most uptodate | |
4706 | * copies of everything. | |
4707 | */ | |
a59108a7 | 4708 | if (S_ISDIR(inode->vfs_inode.i_mode)) { |
e02119d5 CM |
4709 | int max_key_type = BTRFS_DIR_LOG_INDEX_KEY; |
4710 | ||
4f764e51 FM |
4711 | if (inode_only == LOG_INODE_EXISTS) |
4712 | max_key_type = BTRFS_XATTR_ITEM_KEY; | |
33345d01 | 4713 | ret = drop_objectid_items(trans, log, path, ino, max_key_type); |
e02119d5 | 4714 | } else { |
1a4bcf47 FM |
4715 | if (inode_only == LOG_INODE_EXISTS) { |
4716 | /* | |
4717 | * Make sure the new inode item we write to the log has | |
4718 | * the same isize as the current one (if it exists). | |
4719 | * This is necessary to prevent data loss after log | |
4720 | * replay, and also to prevent doing a wrong expanding | |
4721 | * truncate - for e.g. create file, write 4K into offset | |
4722 | * 0, fsync, write 4K into offset 4096, add hard link, | |
4723 | * fsync some other file (to sync log), power fail - if | |
4724 | * we use the inode's current i_size, after log replay | |
4725 | * we get a 8Kb file, with the last 4Kb extent as a hole | |
4726 | * (zeroes), as if an expanding truncate happened, | |
4727 | * instead of getting a file of 4Kb only. | |
4728 | */ | |
a59108a7 | 4729 | err = logged_inode_size(log, inode, path, &logged_isize); |
1a4bcf47 FM |
4730 | if (err) |
4731 | goto out_unlock; | |
4732 | } | |
a742994a | 4733 | if (test_bit(BTRFS_INODE_NEEDS_FULL_SYNC, |
a59108a7 | 4734 | &inode->runtime_flags)) { |
a742994a | 4735 | if (inode_only == LOG_INODE_EXISTS) { |
4f764e51 | 4736 | max_key.type = BTRFS_XATTR_ITEM_KEY; |
a742994a FM |
4737 | ret = drop_objectid_items(trans, log, path, ino, |
4738 | max_key.type); | |
4739 | } else { | |
4740 | clear_bit(BTRFS_INODE_NEEDS_FULL_SYNC, | |
a59108a7 | 4741 | &inode->runtime_flags); |
a742994a | 4742 | clear_bit(BTRFS_INODE_COPY_EVERYTHING, |
a59108a7 | 4743 | &inode->runtime_flags); |
28ed1345 CM |
4744 | while(1) { |
4745 | ret = btrfs_truncate_inode_items(trans, | |
a59108a7 | 4746 | log, &inode->vfs_inode, 0, 0); |
28ed1345 CM |
4747 | if (ret != -EAGAIN) |
4748 | break; | |
4749 | } | |
a742994a | 4750 | } |
4f764e51 | 4751 | } else if (test_and_clear_bit(BTRFS_INODE_COPY_EVERYTHING, |
a59108a7 | 4752 | &inode->runtime_flags) || |
6cfab851 | 4753 | inode_only == LOG_INODE_EXISTS) { |
4f764e51 | 4754 | if (inode_only == LOG_INODE_ALL) |
183f37fa | 4755 | fast_search = true; |
4f764e51 | 4756 | max_key.type = BTRFS_XATTR_ITEM_KEY; |
5dc562c5 | 4757 | ret = drop_objectid_items(trans, log, path, ino, |
e9976151 | 4758 | max_key.type); |
a95249b3 JB |
4759 | } else { |
4760 | if (inode_only == LOG_INODE_ALL) | |
4761 | fast_search = true; | |
a95249b3 | 4762 | goto log_extents; |
5dc562c5 | 4763 | } |
a95249b3 | 4764 | |
e02119d5 | 4765 | } |
4a500fd1 YZ |
4766 | if (ret) { |
4767 | err = ret; | |
4768 | goto out_unlock; | |
4769 | } | |
e02119d5 | 4770 | |
d397712b | 4771 | while (1) { |
31ff1cd2 | 4772 | ins_nr = 0; |
6174d3cb | 4773 | ret = btrfs_search_forward(root, &min_key, |
de78b51a | 4774 | path, trans->transid); |
fb770ae4 LB |
4775 | if (ret < 0) { |
4776 | err = ret; | |
4777 | goto out_unlock; | |
4778 | } | |
e02119d5 CM |
4779 | if (ret != 0) |
4780 | break; | |
3a5f1d45 | 4781 | again: |
31ff1cd2 | 4782 | /* note, ins_nr might be > 0 here, cleanup outside the loop */ |
33345d01 | 4783 | if (min_key.objectid != ino) |
e02119d5 CM |
4784 | break; |
4785 | if (min_key.type > max_key.type) | |
4786 | break; | |
31ff1cd2 | 4787 | |
e4545de5 FM |
4788 | if (min_key.type == BTRFS_INODE_ITEM_KEY) |
4789 | need_log_inode_item = false; | |
4790 | ||
56f23fdb FM |
4791 | if ((min_key.type == BTRFS_INODE_REF_KEY || |
4792 | min_key.type == BTRFS_INODE_EXTREF_KEY) && | |
a59108a7 | 4793 | inode->generation == trans->transid) { |
44f714da FM |
4794 | u64 other_ino = 0; |
4795 | ||
56f23fdb | 4796 | ret = btrfs_check_ref_name_override(path->nodes[0], |
a59108a7 NB |
4797 | path->slots[0], &min_key, inode, |
4798 | &other_ino); | |
56f23fdb FM |
4799 | if (ret < 0) { |
4800 | err = ret; | |
4801 | goto out_unlock; | |
28a23593 | 4802 | } else if (ret > 0 && ctx && |
4a0cc7ca | 4803 | other_ino != btrfs_ino(BTRFS_I(ctx->inode))) { |
44f714da FM |
4804 | struct btrfs_key inode_key; |
4805 | struct inode *other_inode; | |
4806 | ||
4807 | if (ins_nr > 0) { | |
4808 | ins_nr++; | |
4809 | } else { | |
4810 | ins_nr = 1; | |
4811 | ins_start_slot = path->slots[0]; | |
4812 | } | |
a59108a7 | 4813 | ret = copy_items(trans, inode, dst_path, path, |
44f714da FM |
4814 | &last_extent, ins_start_slot, |
4815 | ins_nr, inode_only, | |
4816 | logged_isize); | |
4817 | if (ret < 0) { | |
4818 | err = ret; | |
4819 | goto out_unlock; | |
4820 | } | |
4821 | ins_nr = 0; | |
4822 | btrfs_release_path(path); | |
4823 | inode_key.objectid = other_ino; | |
4824 | inode_key.type = BTRFS_INODE_ITEM_KEY; | |
4825 | inode_key.offset = 0; | |
0b246afa | 4826 | other_inode = btrfs_iget(fs_info->sb, |
44f714da FM |
4827 | &inode_key, root, |
4828 | NULL); | |
4829 | /* | |
4830 | * If the other inode that had a conflicting dir | |
4831 | * entry was deleted in the current transaction, | |
4832 | * we don't need to do more work nor fallback to | |
4833 | * a transaction commit. | |
4834 | */ | |
4835 | if (IS_ERR(other_inode) && | |
4836 | PTR_ERR(other_inode) == -ENOENT) { | |
4837 | goto next_key; | |
4838 | } else if (IS_ERR(other_inode)) { | |
4839 | err = PTR_ERR(other_inode); | |
4840 | goto out_unlock; | |
4841 | } | |
4842 | /* | |
4843 | * We are safe logging the other inode without | |
4844 | * acquiring its i_mutex as long as we log with | |
4845 | * the LOG_INODE_EXISTS mode. We're safe against | |
4846 | * concurrent renames of the other inode as well | |
4847 | * because during a rename we pin the log and | |
4848 | * update the log with the new name before we | |
4849 | * unpin it. | |
4850 | */ | |
a59108a7 NB |
4851 | err = btrfs_log_inode(trans, root, |
4852 | BTRFS_I(other_inode), | |
4853 | LOG_OTHER_INODE, 0, LLONG_MAX, | |
4854 | ctx); | |
44f714da FM |
4855 | iput(other_inode); |
4856 | if (err) | |
4857 | goto out_unlock; | |
4858 | else | |
4859 | goto next_key; | |
56f23fdb FM |
4860 | } |
4861 | } | |
4862 | ||
36283bf7 FM |
4863 | /* Skip xattrs, we log them later with btrfs_log_all_xattrs() */ |
4864 | if (min_key.type == BTRFS_XATTR_ITEM_KEY) { | |
4865 | if (ins_nr == 0) | |
4866 | goto next_slot; | |
a59108a7 | 4867 | ret = copy_items(trans, inode, dst_path, path, |
36283bf7 FM |
4868 | &last_extent, ins_start_slot, |
4869 | ins_nr, inode_only, logged_isize); | |
4870 | if (ret < 0) { | |
4871 | err = ret; | |
4872 | goto out_unlock; | |
4873 | } | |
4874 | ins_nr = 0; | |
4875 | if (ret) { | |
4876 | btrfs_release_path(path); | |
4877 | continue; | |
4878 | } | |
4879 | goto next_slot; | |
4880 | } | |
4881 | ||
e02119d5 | 4882 | src = path->nodes[0]; |
31ff1cd2 CM |
4883 | if (ins_nr && ins_start_slot + ins_nr == path->slots[0]) { |
4884 | ins_nr++; | |
4885 | goto next_slot; | |
4886 | } else if (!ins_nr) { | |
4887 | ins_start_slot = path->slots[0]; | |
4888 | ins_nr = 1; | |
4889 | goto next_slot; | |
e02119d5 CM |
4890 | } |
4891 | ||
a59108a7 | 4892 | ret = copy_items(trans, inode, dst_path, path, &last_extent, |
1a4bcf47 FM |
4893 | ins_start_slot, ins_nr, inode_only, |
4894 | logged_isize); | |
16e7549f | 4895 | if (ret < 0) { |
4a500fd1 YZ |
4896 | err = ret; |
4897 | goto out_unlock; | |
a71db86e RV |
4898 | } |
4899 | if (ret) { | |
16e7549f JB |
4900 | ins_nr = 0; |
4901 | btrfs_release_path(path); | |
4902 | continue; | |
4a500fd1 | 4903 | } |
31ff1cd2 CM |
4904 | ins_nr = 1; |
4905 | ins_start_slot = path->slots[0]; | |
4906 | next_slot: | |
e02119d5 | 4907 | |
3a5f1d45 CM |
4908 | nritems = btrfs_header_nritems(path->nodes[0]); |
4909 | path->slots[0]++; | |
4910 | if (path->slots[0] < nritems) { | |
4911 | btrfs_item_key_to_cpu(path->nodes[0], &min_key, | |
4912 | path->slots[0]); | |
4913 | goto again; | |
4914 | } | |
31ff1cd2 | 4915 | if (ins_nr) { |
a59108a7 | 4916 | ret = copy_items(trans, inode, dst_path, path, |
16e7549f | 4917 | &last_extent, ins_start_slot, |
1a4bcf47 | 4918 | ins_nr, inode_only, logged_isize); |
16e7549f | 4919 | if (ret < 0) { |
4a500fd1 YZ |
4920 | err = ret; |
4921 | goto out_unlock; | |
4922 | } | |
16e7549f | 4923 | ret = 0; |
31ff1cd2 CM |
4924 | ins_nr = 0; |
4925 | } | |
b3b4aa74 | 4926 | btrfs_release_path(path); |
44f714da | 4927 | next_key: |
3d41d702 | 4928 | if (min_key.offset < (u64)-1) { |
e02119d5 | 4929 | min_key.offset++; |
3d41d702 | 4930 | } else if (min_key.type < max_key.type) { |
e02119d5 | 4931 | min_key.type++; |
3d41d702 FDBM |
4932 | min_key.offset = 0; |
4933 | } else { | |
e02119d5 | 4934 | break; |
3d41d702 | 4935 | } |
e02119d5 | 4936 | } |
31ff1cd2 | 4937 | if (ins_nr) { |
a59108a7 | 4938 | ret = copy_items(trans, inode, dst_path, path, &last_extent, |
1a4bcf47 FM |
4939 | ins_start_slot, ins_nr, inode_only, |
4940 | logged_isize); | |
16e7549f | 4941 | if (ret < 0) { |
4a500fd1 YZ |
4942 | err = ret; |
4943 | goto out_unlock; | |
4944 | } | |
16e7549f | 4945 | ret = 0; |
31ff1cd2 CM |
4946 | ins_nr = 0; |
4947 | } | |
5dc562c5 | 4948 | |
36283bf7 FM |
4949 | btrfs_release_path(path); |
4950 | btrfs_release_path(dst_path); | |
a59108a7 | 4951 | err = btrfs_log_all_xattrs(trans, root, inode, path, dst_path); |
36283bf7 FM |
4952 | if (err) |
4953 | goto out_unlock; | |
a89ca6f2 FM |
4954 | if (max_key.type >= BTRFS_EXTENT_DATA_KEY && !fast_search) { |
4955 | btrfs_release_path(path); | |
4956 | btrfs_release_path(dst_path); | |
a59108a7 | 4957 | err = btrfs_log_trailing_hole(trans, root, inode, path); |
a89ca6f2 FM |
4958 | if (err) |
4959 | goto out_unlock; | |
4960 | } | |
a95249b3 | 4961 | log_extents: |
f3b15ccd JB |
4962 | btrfs_release_path(path); |
4963 | btrfs_release_path(dst_path); | |
e4545de5 | 4964 | if (need_log_inode_item) { |
a59108a7 | 4965 | err = log_inode_item(trans, log, dst_path, inode); |
e4545de5 FM |
4966 | if (err) |
4967 | goto out_unlock; | |
4968 | } | |
5dc562c5 | 4969 | if (fast_search) { |
a59108a7 | 4970 | ret = btrfs_log_changed_extents(trans, root, inode, dst_path, |
de0ee0ed | 4971 | &logged_list, ctx, start, end); |
5dc562c5 JB |
4972 | if (ret) { |
4973 | err = ret; | |
4974 | goto out_unlock; | |
4975 | } | |
d006a048 | 4976 | } else if (inode_only == LOG_INODE_ALL) { |
06d3d22b LB |
4977 | struct extent_map *em, *n; |
4978 | ||
49dae1bc FM |
4979 | write_lock(&em_tree->lock); |
4980 | /* | |
4981 | * We can't just remove every em if we're called for a ranged | |
4982 | * fsync - that is, one that doesn't cover the whole possible | |
4983 | * file range (0 to LLONG_MAX). This is because we can have | |
4984 | * em's that fall outside the range we're logging and therefore | |
4985 | * their ordered operations haven't completed yet | |
4986 | * (btrfs_finish_ordered_io() not invoked yet). This means we | |
4987 | * didn't get their respective file extent item in the fs/subvol | |
4988 | * tree yet, and need to let the next fast fsync (one which | |
4989 | * consults the list of modified extent maps) find the em so | |
4990 | * that it logs a matching file extent item and waits for the | |
4991 | * respective ordered operation to complete (if it's still | |
4992 | * running). | |
4993 | * | |
4994 | * Removing every em outside the range we're logging would make | |
4995 | * the next fast fsync not log their matching file extent items, | |
4996 | * therefore making us lose data after a log replay. | |
4997 | */ | |
4998 | list_for_each_entry_safe(em, n, &em_tree->modified_extents, | |
4999 | list) { | |
5000 | const u64 mod_end = em->mod_start + em->mod_len - 1; | |
5001 | ||
5002 | if (em->mod_start >= start && mod_end <= end) | |
5003 | list_del_init(&em->list); | |
5004 | } | |
5005 | write_unlock(&em_tree->lock); | |
5dc562c5 JB |
5006 | } |
5007 | ||
a59108a7 NB |
5008 | if (inode_only == LOG_INODE_ALL && S_ISDIR(inode->vfs_inode.i_mode)) { |
5009 | ret = log_directory_changes(trans, root, inode, path, dst_path, | |
5010 | ctx); | |
4a500fd1 YZ |
5011 | if (ret) { |
5012 | err = ret; | |
5013 | goto out_unlock; | |
5014 | } | |
e02119d5 | 5015 | } |
49dae1bc | 5016 | |
a59108a7 NB |
5017 | spin_lock(&inode->lock); |
5018 | inode->logged_trans = trans->transid; | |
5019 | inode->last_log_commit = inode->last_sub_trans; | |
5020 | spin_unlock(&inode->lock); | |
4a500fd1 | 5021 | out_unlock: |
827463c4 MX |
5022 | if (unlikely(err)) |
5023 | btrfs_put_logged_extents(&logged_list); | |
5024 | else | |
5025 | btrfs_submit_logged_extents(&logged_list, log); | |
a59108a7 | 5026 | mutex_unlock(&inode->log_mutex); |
e02119d5 CM |
5027 | |
5028 | btrfs_free_path(path); | |
5029 | btrfs_free_path(dst_path); | |
4a500fd1 | 5030 | return err; |
e02119d5 CM |
5031 | } |
5032 | ||
2be63d5c FM |
5033 | /* |
5034 | * Check if we must fallback to a transaction commit when logging an inode. | |
5035 | * This must be called after logging the inode and is used only in the context | |
5036 | * when fsyncing an inode requires the need to log some other inode - in which | |
5037 | * case we can't lock the i_mutex of each other inode we need to log as that | |
5038 | * can lead to deadlocks with concurrent fsync against other inodes (as we can | |
5039 | * log inodes up or down in the hierarchy) or rename operations for example. So | |
5040 | * we take the log_mutex of the inode after we have logged it and then check for | |
5041 | * its last_unlink_trans value - this is safe because any task setting | |
5042 | * last_unlink_trans must take the log_mutex and it must do this before it does | |
5043 | * the actual unlink operation, so if we do this check before a concurrent task | |
5044 | * sets last_unlink_trans it means we've logged a consistent version/state of | |
5045 | * all the inode items, otherwise we are not sure and must do a transaction | |
01327610 | 5046 | * commit (the concurrent task might have only updated last_unlink_trans before |
2be63d5c FM |
5047 | * we logged the inode or it might have also done the unlink). |
5048 | */ | |
5049 | static bool btrfs_must_commit_transaction(struct btrfs_trans_handle *trans, | |
ab1717b2 | 5050 | struct btrfs_inode *inode) |
2be63d5c | 5051 | { |
ab1717b2 | 5052 | struct btrfs_fs_info *fs_info = inode->root->fs_info; |
2be63d5c FM |
5053 | bool ret = false; |
5054 | ||
ab1717b2 NB |
5055 | mutex_lock(&inode->log_mutex); |
5056 | if (inode->last_unlink_trans > fs_info->last_trans_committed) { | |
2be63d5c FM |
5057 | /* |
5058 | * Make sure any commits to the log are forced to be full | |
5059 | * commits. | |
5060 | */ | |
5061 | btrfs_set_log_full_commit(fs_info, trans); | |
5062 | ret = true; | |
5063 | } | |
ab1717b2 | 5064 | mutex_unlock(&inode->log_mutex); |
2be63d5c FM |
5065 | |
5066 | return ret; | |
5067 | } | |
5068 | ||
12fcfd22 CM |
5069 | /* |
5070 | * follow the dentry parent pointers up the chain and see if any | |
5071 | * of the directories in it require a full commit before they can | |
5072 | * be logged. Returns zero if nothing special needs to be done or 1 if | |
5073 | * a full commit is required. | |
5074 | */ | |
5075 | static noinline int check_parent_dirs_for_sync(struct btrfs_trans_handle *trans, | |
aefa6115 | 5076 | struct btrfs_inode *inode, |
12fcfd22 CM |
5077 | struct dentry *parent, |
5078 | struct super_block *sb, | |
5079 | u64 last_committed) | |
e02119d5 | 5080 | { |
12fcfd22 | 5081 | int ret = 0; |
6a912213 | 5082 | struct dentry *old_parent = NULL; |
aefa6115 | 5083 | struct btrfs_inode *orig_inode = inode; |
e02119d5 | 5084 | |
af4176b4 CM |
5085 | /* |
5086 | * for regular files, if its inode is already on disk, we don't | |
5087 | * have to worry about the parents at all. This is because | |
5088 | * we can use the last_unlink_trans field to record renames | |
5089 | * and other fun in this file. | |
5090 | */ | |
aefa6115 NB |
5091 | if (S_ISREG(inode->vfs_inode.i_mode) && |
5092 | inode->generation <= last_committed && | |
5093 | inode->last_unlink_trans <= last_committed) | |
5094 | goto out; | |
af4176b4 | 5095 | |
aefa6115 | 5096 | if (!S_ISDIR(inode->vfs_inode.i_mode)) { |
fc64005c | 5097 | if (!parent || d_really_is_negative(parent) || sb != parent->d_sb) |
12fcfd22 | 5098 | goto out; |
aefa6115 | 5099 | inode = BTRFS_I(d_inode(parent)); |
12fcfd22 CM |
5100 | } |
5101 | ||
5102 | while (1) { | |
de2b530b JB |
5103 | /* |
5104 | * If we are logging a directory then we start with our inode, | |
01327610 | 5105 | * not our parent's inode, so we need to skip setting the |
de2b530b JB |
5106 | * logged_trans so that further down in the log code we don't |
5107 | * think this inode has already been logged. | |
5108 | */ | |
5109 | if (inode != orig_inode) | |
aefa6115 | 5110 | inode->logged_trans = trans->transid; |
12fcfd22 CM |
5111 | smp_mb(); |
5112 | ||
aefa6115 | 5113 | if (btrfs_must_commit_transaction(trans, inode)) { |
12fcfd22 CM |
5114 | ret = 1; |
5115 | break; | |
5116 | } | |
5117 | ||
fc64005c | 5118 | if (!parent || d_really_is_negative(parent) || sb != parent->d_sb) |
12fcfd22 CM |
5119 | break; |
5120 | ||
44f714da | 5121 | if (IS_ROOT(parent)) { |
aefa6115 NB |
5122 | inode = BTRFS_I(d_inode(parent)); |
5123 | if (btrfs_must_commit_transaction(trans, inode)) | |
44f714da | 5124 | ret = 1; |
12fcfd22 | 5125 | break; |
44f714da | 5126 | } |
12fcfd22 | 5127 | |
6a912213 JB |
5128 | parent = dget_parent(parent); |
5129 | dput(old_parent); | |
5130 | old_parent = parent; | |
aefa6115 | 5131 | inode = BTRFS_I(d_inode(parent)); |
12fcfd22 CM |
5132 | |
5133 | } | |
6a912213 | 5134 | dput(old_parent); |
12fcfd22 | 5135 | out: |
e02119d5 CM |
5136 | return ret; |
5137 | } | |
5138 | ||
2f2ff0ee FM |
5139 | struct btrfs_dir_list { |
5140 | u64 ino; | |
5141 | struct list_head list; | |
5142 | }; | |
5143 | ||
5144 | /* | |
5145 | * Log the inodes of the new dentries of a directory. See log_dir_items() for | |
5146 | * details about the why it is needed. | |
5147 | * This is a recursive operation - if an existing dentry corresponds to a | |
5148 | * directory, that directory's new entries are logged too (same behaviour as | |
5149 | * ext3/4, xfs, f2fs, reiserfs, nilfs2). Note that when logging the inodes | |
5150 | * the dentries point to we do not lock their i_mutex, otherwise lockdep | |
5151 | * complains about the following circular lock dependency / possible deadlock: | |
5152 | * | |
5153 | * CPU0 CPU1 | |
5154 | * ---- ---- | |
5155 | * lock(&type->i_mutex_dir_key#3/2); | |
5156 | * lock(sb_internal#2); | |
5157 | * lock(&type->i_mutex_dir_key#3/2); | |
5158 | * lock(&sb->s_type->i_mutex_key#14); | |
5159 | * | |
5160 | * Where sb_internal is the lock (a counter that works as a lock) acquired by | |
5161 | * sb_start_intwrite() in btrfs_start_transaction(). | |
5162 | * Not locking i_mutex of the inodes is still safe because: | |
5163 | * | |
5164 | * 1) For regular files we log with a mode of LOG_INODE_EXISTS. It's possible | |
5165 | * that while logging the inode new references (names) are added or removed | |
5166 | * from the inode, leaving the logged inode item with a link count that does | |
5167 | * not match the number of logged inode reference items. This is fine because | |
5168 | * at log replay time we compute the real number of links and correct the | |
5169 | * link count in the inode item (see replay_one_buffer() and | |
5170 | * link_to_fixup_dir()); | |
5171 | * | |
5172 | * 2) For directories we log with a mode of LOG_INODE_ALL. It's possible that | |
5173 | * while logging the inode's items new items with keys BTRFS_DIR_ITEM_KEY and | |
5174 | * BTRFS_DIR_INDEX_KEY are added to fs/subvol tree and the logged inode item | |
5175 | * has a size that doesn't match the sum of the lengths of all the logged | |
5176 | * names. This does not result in a problem because if a dir_item key is | |
5177 | * logged but its matching dir_index key is not logged, at log replay time we | |
5178 | * don't use it to replay the respective name (see replay_one_name()). On the | |
5179 | * other hand if only the dir_index key ends up being logged, the respective | |
5180 | * name is added to the fs/subvol tree with both the dir_item and dir_index | |
5181 | * keys created (see replay_one_name()). | |
5182 | * The directory's inode item with a wrong i_size is not a problem as well, | |
5183 | * since we don't use it at log replay time to set the i_size in the inode | |
5184 | * item of the fs/subvol tree (see overwrite_item()). | |
5185 | */ | |
5186 | static int log_new_dir_dentries(struct btrfs_trans_handle *trans, | |
5187 | struct btrfs_root *root, | |
51cc0d32 | 5188 | struct btrfs_inode *start_inode, |
2f2ff0ee FM |
5189 | struct btrfs_log_ctx *ctx) |
5190 | { | |
0b246afa | 5191 | struct btrfs_fs_info *fs_info = root->fs_info; |
2f2ff0ee FM |
5192 | struct btrfs_root *log = root->log_root; |
5193 | struct btrfs_path *path; | |
5194 | LIST_HEAD(dir_list); | |
5195 | struct btrfs_dir_list *dir_elem; | |
5196 | int ret = 0; | |
5197 | ||
5198 | path = btrfs_alloc_path(); | |
5199 | if (!path) | |
5200 | return -ENOMEM; | |
5201 | ||
5202 | dir_elem = kmalloc(sizeof(*dir_elem), GFP_NOFS); | |
5203 | if (!dir_elem) { | |
5204 | btrfs_free_path(path); | |
5205 | return -ENOMEM; | |
5206 | } | |
51cc0d32 | 5207 | dir_elem->ino = btrfs_ino(start_inode); |
2f2ff0ee FM |
5208 | list_add_tail(&dir_elem->list, &dir_list); |
5209 | ||
5210 | while (!list_empty(&dir_list)) { | |
5211 | struct extent_buffer *leaf; | |
5212 | struct btrfs_key min_key; | |
5213 | int nritems; | |
5214 | int i; | |
5215 | ||
5216 | dir_elem = list_first_entry(&dir_list, struct btrfs_dir_list, | |
5217 | list); | |
5218 | if (ret) | |
5219 | goto next_dir_inode; | |
5220 | ||
5221 | min_key.objectid = dir_elem->ino; | |
5222 | min_key.type = BTRFS_DIR_ITEM_KEY; | |
5223 | min_key.offset = 0; | |
5224 | again: | |
5225 | btrfs_release_path(path); | |
5226 | ret = btrfs_search_forward(log, &min_key, path, trans->transid); | |
5227 | if (ret < 0) { | |
5228 | goto next_dir_inode; | |
5229 | } else if (ret > 0) { | |
5230 | ret = 0; | |
5231 | goto next_dir_inode; | |
5232 | } | |
5233 | ||
5234 | process_leaf: | |
5235 | leaf = path->nodes[0]; | |
5236 | nritems = btrfs_header_nritems(leaf); | |
5237 | for (i = path->slots[0]; i < nritems; i++) { | |
5238 | struct btrfs_dir_item *di; | |
5239 | struct btrfs_key di_key; | |
5240 | struct inode *di_inode; | |
5241 | struct btrfs_dir_list *new_dir_elem; | |
5242 | int log_mode = LOG_INODE_EXISTS; | |
5243 | int type; | |
5244 | ||
5245 | btrfs_item_key_to_cpu(leaf, &min_key, i); | |
5246 | if (min_key.objectid != dir_elem->ino || | |
5247 | min_key.type != BTRFS_DIR_ITEM_KEY) | |
5248 | goto next_dir_inode; | |
5249 | ||
5250 | di = btrfs_item_ptr(leaf, i, struct btrfs_dir_item); | |
5251 | type = btrfs_dir_type(leaf, di); | |
5252 | if (btrfs_dir_transid(leaf, di) < trans->transid && | |
5253 | type != BTRFS_FT_DIR) | |
5254 | continue; | |
5255 | btrfs_dir_item_key_to_cpu(leaf, di, &di_key); | |
5256 | if (di_key.type == BTRFS_ROOT_ITEM_KEY) | |
5257 | continue; | |
5258 | ||
ec125cfb | 5259 | btrfs_release_path(path); |
0b246afa | 5260 | di_inode = btrfs_iget(fs_info->sb, &di_key, root, NULL); |
2f2ff0ee FM |
5261 | if (IS_ERR(di_inode)) { |
5262 | ret = PTR_ERR(di_inode); | |
5263 | goto next_dir_inode; | |
5264 | } | |
5265 | ||
0f8939b8 | 5266 | if (btrfs_inode_in_log(BTRFS_I(di_inode), trans->transid)) { |
2f2ff0ee | 5267 | iput(di_inode); |
ec125cfb | 5268 | break; |
2f2ff0ee FM |
5269 | } |
5270 | ||
5271 | ctx->log_new_dentries = false; | |
3f9749f6 | 5272 | if (type == BTRFS_FT_DIR || type == BTRFS_FT_SYMLINK) |
2f2ff0ee | 5273 | log_mode = LOG_INODE_ALL; |
a59108a7 | 5274 | ret = btrfs_log_inode(trans, root, BTRFS_I(di_inode), |
2f2ff0ee | 5275 | log_mode, 0, LLONG_MAX, ctx); |
2be63d5c | 5276 | if (!ret && |
ab1717b2 | 5277 | btrfs_must_commit_transaction(trans, BTRFS_I(di_inode))) |
2be63d5c | 5278 | ret = 1; |
2f2ff0ee FM |
5279 | iput(di_inode); |
5280 | if (ret) | |
5281 | goto next_dir_inode; | |
5282 | if (ctx->log_new_dentries) { | |
5283 | new_dir_elem = kmalloc(sizeof(*new_dir_elem), | |
5284 | GFP_NOFS); | |
5285 | if (!new_dir_elem) { | |
5286 | ret = -ENOMEM; | |
5287 | goto next_dir_inode; | |
5288 | } | |
5289 | new_dir_elem->ino = di_key.objectid; | |
5290 | list_add_tail(&new_dir_elem->list, &dir_list); | |
5291 | } | |
5292 | break; | |
5293 | } | |
5294 | if (i == nritems) { | |
5295 | ret = btrfs_next_leaf(log, path); | |
5296 | if (ret < 0) { | |
5297 | goto next_dir_inode; | |
5298 | } else if (ret > 0) { | |
5299 | ret = 0; | |
5300 | goto next_dir_inode; | |
5301 | } | |
5302 | goto process_leaf; | |
5303 | } | |
5304 | if (min_key.offset < (u64)-1) { | |
5305 | min_key.offset++; | |
5306 | goto again; | |
5307 | } | |
5308 | next_dir_inode: | |
5309 | list_del(&dir_elem->list); | |
5310 | kfree(dir_elem); | |
5311 | } | |
5312 | ||
5313 | btrfs_free_path(path); | |
5314 | return ret; | |
5315 | } | |
5316 | ||
18aa0922 | 5317 | static int btrfs_log_all_parents(struct btrfs_trans_handle *trans, |
d0a0b78d | 5318 | struct btrfs_inode *inode, |
18aa0922 FM |
5319 | struct btrfs_log_ctx *ctx) |
5320 | { | |
d0a0b78d | 5321 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb); |
18aa0922 FM |
5322 | int ret; |
5323 | struct btrfs_path *path; | |
5324 | struct btrfs_key key; | |
d0a0b78d NB |
5325 | struct btrfs_root *root = inode->root; |
5326 | const u64 ino = btrfs_ino(inode); | |
18aa0922 FM |
5327 | |
5328 | path = btrfs_alloc_path(); | |
5329 | if (!path) | |
5330 | return -ENOMEM; | |
5331 | path->skip_locking = 1; | |
5332 | path->search_commit_root = 1; | |
5333 | ||
5334 | key.objectid = ino; | |
5335 | key.type = BTRFS_INODE_REF_KEY; | |
5336 | key.offset = 0; | |
5337 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
5338 | if (ret < 0) | |
5339 | goto out; | |
5340 | ||
5341 | while (true) { | |
5342 | struct extent_buffer *leaf = path->nodes[0]; | |
5343 | int slot = path->slots[0]; | |
5344 | u32 cur_offset = 0; | |
5345 | u32 item_size; | |
5346 | unsigned long ptr; | |
5347 | ||
5348 | if (slot >= btrfs_header_nritems(leaf)) { | |
5349 | ret = btrfs_next_leaf(root, path); | |
5350 | if (ret < 0) | |
5351 | goto out; | |
5352 | else if (ret > 0) | |
5353 | break; | |
5354 | continue; | |
5355 | } | |
5356 | ||
5357 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
5358 | /* BTRFS_INODE_EXTREF_KEY is BTRFS_INODE_REF_KEY + 1 */ | |
5359 | if (key.objectid != ino || key.type > BTRFS_INODE_EXTREF_KEY) | |
5360 | break; | |
5361 | ||
5362 | item_size = btrfs_item_size_nr(leaf, slot); | |
5363 | ptr = btrfs_item_ptr_offset(leaf, slot); | |
5364 | while (cur_offset < item_size) { | |
5365 | struct btrfs_key inode_key; | |
5366 | struct inode *dir_inode; | |
5367 | ||
5368 | inode_key.type = BTRFS_INODE_ITEM_KEY; | |
5369 | inode_key.offset = 0; | |
5370 | ||
5371 | if (key.type == BTRFS_INODE_EXTREF_KEY) { | |
5372 | struct btrfs_inode_extref *extref; | |
5373 | ||
5374 | extref = (struct btrfs_inode_extref *) | |
5375 | (ptr + cur_offset); | |
5376 | inode_key.objectid = btrfs_inode_extref_parent( | |
5377 | leaf, extref); | |
5378 | cur_offset += sizeof(*extref); | |
5379 | cur_offset += btrfs_inode_extref_name_len(leaf, | |
5380 | extref); | |
5381 | } else { | |
5382 | inode_key.objectid = key.offset; | |
5383 | cur_offset = item_size; | |
5384 | } | |
5385 | ||
0b246afa | 5386 | dir_inode = btrfs_iget(fs_info->sb, &inode_key, |
18aa0922 FM |
5387 | root, NULL); |
5388 | /* If parent inode was deleted, skip it. */ | |
5389 | if (IS_ERR(dir_inode)) | |
5390 | continue; | |
5391 | ||
657ed1aa FM |
5392 | if (ctx) |
5393 | ctx->log_new_dentries = false; | |
a59108a7 | 5394 | ret = btrfs_log_inode(trans, root, BTRFS_I(dir_inode), |
18aa0922 | 5395 | LOG_INODE_ALL, 0, LLONG_MAX, ctx); |
2be63d5c | 5396 | if (!ret && |
ab1717b2 | 5397 | btrfs_must_commit_transaction(trans, BTRFS_I(dir_inode))) |
2be63d5c | 5398 | ret = 1; |
657ed1aa FM |
5399 | if (!ret && ctx && ctx->log_new_dentries) |
5400 | ret = log_new_dir_dentries(trans, root, | |
f85b7379 | 5401 | BTRFS_I(dir_inode), ctx); |
18aa0922 FM |
5402 | iput(dir_inode); |
5403 | if (ret) | |
5404 | goto out; | |
5405 | } | |
5406 | path->slots[0]++; | |
5407 | } | |
5408 | ret = 0; | |
5409 | out: | |
5410 | btrfs_free_path(path); | |
5411 | return ret; | |
5412 | } | |
5413 | ||
e02119d5 CM |
5414 | /* |
5415 | * helper function around btrfs_log_inode to make sure newly created | |
5416 | * parent directories also end up in the log. A minimal inode and backref | |
5417 | * only logging is done of any parent directories that are older than | |
5418 | * the last committed transaction | |
5419 | */ | |
48a3b636 | 5420 | static int btrfs_log_inode_parent(struct btrfs_trans_handle *trans, |
19df27a9 NB |
5421 | struct btrfs_root *root, |
5422 | struct btrfs_inode *inode, | |
49dae1bc FM |
5423 | struct dentry *parent, |
5424 | const loff_t start, | |
5425 | const loff_t end, | |
5426 | int exists_only, | |
8b050d35 | 5427 | struct btrfs_log_ctx *ctx) |
e02119d5 | 5428 | { |
0b246afa | 5429 | struct btrfs_fs_info *fs_info = root->fs_info; |
12fcfd22 | 5430 | int inode_only = exists_only ? LOG_INODE_EXISTS : LOG_INODE_ALL; |
e02119d5 | 5431 | struct super_block *sb; |
6a912213 | 5432 | struct dentry *old_parent = NULL; |
12fcfd22 | 5433 | int ret = 0; |
0b246afa | 5434 | u64 last_committed = fs_info->last_trans_committed; |
2f2ff0ee | 5435 | bool log_dentries = false; |
19df27a9 | 5436 | struct btrfs_inode *orig_inode = inode; |
12fcfd22 | 5437 | |
19df27a9 | 5438 | sb = inode->vfs_inode.i_sb; |
12fcfd22 | 5439 | |
0b246afa | 5440 | if (btrfs_test_opt(fs_info, NOTREELOG)) { |
3a5e1404 SW |
5441 | ret = 1; |
5442 | goto end_no_trans; | |
5443 | } | |
5444 | ||
995946dd MX |
5445 | /* |
5446 | * The prev transaction commit doesn't complete, we need do | |
5447 | * full commit by ourselves. | |
5448 | */ | |
0b246afa JM |
5449 | if (fs_info->last_trans_log_full_commit > |
5450 | fs_info->last_trans_committed) { | |
12fcfd22 CM |
5451 | ret = 1; |
5452 | goto end_no_trans; | |
5453 | } | |
5454 | ||
19df27a9 | 5455 | if (root != inode->root || btrfs_root_refs(&root->root_item) == 0) { |
76dda93c YZ |
5456 | ret = 1; |
5457 | goto end_no_trans; | |
5458 | } | |
5459 | ||
19df27a9 NB |
5460 | ret = check_parent_dirs_for_sync(trans, inode, parent, sb, |
5461 | last_committed); | |
12fcfd22 CM |
5462 | if (ret) |
5463 | goto end_no_trans; | |
e02119d5 | 5464 | |
19df27a9 | 5465 | if (btrfs_inode_in_log(inode, trans->transid)) { |
257c62e1 CM |
5466 | ret = BTRFS_NO_LOG_SYNC; |
5467 | goto end_no_trans; | |
5468 | } | |
5469 | ||
8b050d35 | 5470 | ret = start_log_trans(trans, root, ctx); |
4a500fd1 | 5471 | if (ret) |
e87ac136 | 5472 | goto end_no_trans; |
e02119d5 | 5473 | |
19df27a9 | 5474 | ret = btrfs_log_inode(trans, root, inode, inode_only, start, end, ctx); |
4a500fd1 YZ |
5475 | if (ret) |
5476 | goto end_trans; | |
12fcfd22 | 5477 | |
af4176b4 CM |
5478 | /* |
5479 | * for regular files, if its inode is already on disk, we don't | |
5480 | * have to worry about the parents at all. This is because | |
5481 | * we can use the last_unlink_trans field to record renames | |
5482 | * and other fun in this file. | |
5483 | */ | |
19df27a9 NB |
5484 | if (S_ISREG(inode->vfs_inode.i_mode) && |
5485 | inode->generation <= last_committed && | |
5486 | inode->last_unlink_trans <= last_committed) { | |
4a500fd1 YZ |
5487 | ret = 0; |
5488 | goto end_trans; | |
5489 | } | |
af4176b4 | 5490 | |
19df27a9 | 5491 | if (S_ISDIR(inode->vfs_inode.i_mode) && ctx && ctx->log_new_dentries) |
2f2ff0ee FM |
5492 | log_dentries = true; |
5493 | ||
18aa0922 | 5494 | /* |
01327610 | 5495 | * On unlink we must make sure all our current and old parent directory |
18aa0922 FM |
5496 | * inodes are fully logged. This is to prevent leaving dangling |
5497 | * directory index entries in directories that were our parents but are | |
5498 | * not anymore. Not doing this results in old parent directory being | |
5499 | * impossible to delete after log replay (rmdir will always fail with | |
5500 | * error -ENOTEMPTY). | |
5501 | * | |
5502 | * Example 1: | |
5503 | * | |
5504 | * mkdir testdir | |
5505 | * touch testdir/foo | |
5506 | * ln testdir/foo testdir/bar | |
5507 | * sync | |
5508 | * unlink testdir/bar | |
5509 | * xfs_io -c fsync testdir/foo | |
5510 | * <power failure> | |
5511 | * mount fs, triggers log replay | |
5512 | * | |
5513 | * If we don't log the parent directory (testdir), after log replay the | |
5514 | * directory still has an entry pointing to the file inode using the bar | |
5515 | * name, but a matching BTRFS_INODE_[REF|EXTREF]_KEY does not exist and | |
5516 | * the file inode has a link count of 1. | |
5517 | * | |
5518 | * Example 2: | |
5519 | * | |
5520 | * mkdir testdir | |
5521 | * touch foo | |
5522 | * ln foo testdir/foo2 | |
5523 | * ln foo testdir/foo3 | |
5524 | * sync | |
5525 | * unlink testdir/foo3 | |
5526 | * xfs_io -c fsync foo | |
5527 | * <power failure> | |
5528 | * mount fs, triggers log replay | |
5529 | * | |
5530 | * Similar as the first example, after log replay the parent directory | |
5531 | * testdir still has an entry pointing to the inode file with name foo3 | |
5532 | * but the file inode does not have a matching BTRFS_INODE_REF_KEY item | |
5533 | * and has a link count of 2. | |
5534 | */ | |
19df27a9 | 5535 | if (inode->last_unlink_trans > last_committed) { |
18aa0922 FM |
5536 | ret = btrfs_log_all_parents(trans, orig_inode, ctx); |
5537 | if (ret) | |
5538 | goto end_trans; | |
5539 | } | |
5540 | ||
12fcfd22 | 5541 | while (1) { |
fc64005c | 5542 | if (!parent || d_really_is_negative(parent) || sb != parent->d_sb) |
e02119d5 CM |
5543 | break; |
5544 | ||
19df27a9 NB |
5545 | inode = BTRFS_I(d_inode(parent)); |
5546 | if (root != inode->root) | |
76dda93c YZ |
5547 | break; |
5548 | ||
19df27a9 NB |
5549 | if (inode->generation > last_committed) { |
5550 | ret = btrfs_log_inode(trans, root, inode, | |
5551 | LOG_INODE_EXISTS, 0, LLONG_MAX, ctx); | |
4a500fd1 YZ |
5552 | if (ret) |
5553 | goto end_trans; | |
12fcfd22 | 5554 | } |
76dda93c | 5555 | if (IS_ROOT(parent)) |
e02119d5 | 5556 | break; |
12fcfd22 | 5557 | |
6a912213 JB |
5558 | parent = dget_parent(parent); |
5559 | dput(old_parent); | |
5560 | old_parent = parent; | |
e02119d5 | 5561 | } |
2f2ff0ee | 5562 | if (log_dentries) |
19df27a9 | 5563 | ret = log_new_dir_dentries(trans, root, orig_inode, ctx); |
2f2ff0ee FM |
5564 | else |
5565 | ret = 0; | |
4a500fd1 | 5566 | end_trans: |
6a912213 | 5567 | dput(old_parent); |
4a500fd1 | 5568 | if (ret < 0) { |
0b246afa | 5569 | btrfs_set_log_full_commit(fs_info, trans); |
4a500fd1 YZ |
5570 | ret = 1; |
5571 | } | |
8b050d35 MX |
5572 | |
5573 | if (ret) | |
5574 | btrfs_remove_log_ctx(root, ctx); | |
12fcfd22 CM |
5575 | btrfs_end_log_trans(root); |
5576 | end_no_trans: | |
5577 | return ret; | |
e02119d5 CM |
5578 | } |
5579 | ||
5580 | /* | |
5581 | * it is not safe to log dentry if the chunk root has added new | |
5582 | * chunks. This returns 0 if the dentry was logged, and 1 otherwise. | |
5583 | * If this returns 1, you must commit the transaction to safely get your | |
5584 | * data on disk. | |
5585 | */ | |
5586 | int btrfs_log_dentry_safe(struct btrfs_trans_handle *trans, | |
8b050d35 | 5587 | struct btrfs_root *root, struct dentry *dentry, |
49dae1bc FM |
5588 | const loff_t start, |
5589 | const loff_t end, | |
8b050d35 | 5590 | struct btrfs_log_ctx *ctx) |
e02119d5 | 5591 | { |
6a912213 JB |
5592 | struct dentry *parent = dget_parent(dentry); |
5593 | int ret; | |
5594 | ||
19df27a9 NB |
5595 | ret = btrfs_log_inode_parent(trans, root, BTRFS_I(d_inode(dentry)), |
5596 | parent, start, end, 0, ctx); | |
6a912213 JB |
5597 | dput(parent); |
5598 | ||
5599 | return ret; | |
e02119d5 CM |
5600 | } |
5601 | ||
5602 | /* | |
5603 | * should be called during mount to recover any replay any log trees | |
5604 | * from the FS | |
5605 | */ | |
5606 | int btrfs_recover_log_trees(struct btrfs_root *log_root_tree) | |
5607 | { | |
5608 | int ret; | |
5609 | struct btrfs_path *path; | |
5610 | struct btrfs_trans_handle *trans; | |
5611 | struct btrfs_key key; | |
5612 | struct btrfs_key found_key; | |
5613 | struct btrfs_key tmp_key; | |
5614 | struct btrfs_root *log; | |
5615 | struct btrfs_fs_info *fs_info = log_root_tree->fs_info; | |
5616 | struct walk_control wc = { | |
5617 | .process_func = process_one_buffer, | |
5618 | .stage = 0, | |
5619 | }; | |
5620 | ||
e02119d5 | 5621 | path = btrfs_alloc_path(); |
db5b493a TI |
5622 | if (!path) |
5623 | return -ENOMEM; | |
5624 | ||
afcdd129 | 5625 | set_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags); |
e02119d5 | 5626 | |
4a500fd1 | 5627 | trans = btrfs_start_transaction(fs_info->tree_root, 0); |
79787eaa JM |
5628 | if (IS_ERR(trans)) { |
5629 | ret = PTR_ERR(trans); | |
5630 | goto error; | |
5631 | } | |
e02119d5 CM |
5632 | |
5633 | wc.trans = trans; | |
5634 | wc.pin = 1; | |
5635 | ||
db5b493a | 5636 | ret = walk_log_tree(trans, log_root_tree, &wc); |
79787eaa | 5637 | if (ret) { |
5d163e0e JM |
5638 | btrfs_handle_fs_error(fs_info, ret, |
5639 | "Failed to pin buffers while recovering log root tree."); | |
79787eaa JM |
5640 | goto error; |
5641 | } | |
e02119d5 CM |
5642 | |
5643 | again: | |
5644 | key.objectid = BTRFS_TREE_LOG_OBJECTID; | |
5645 | key.offset = (u64)-1; | |
962a298f | 5646 | key.type = BTRFS_ROOT_ITEM_KEY; |
e02119d5 | 5647 | |
d397712b | 5648 | while (1) { |
e02119d5 | 5649 | ret = btrfs_search_slot(NULL, log_root_tree, &key, path, 0, 0); |
79787eaa JM |
5650 | |
5651 | if (ret < 0) { | |
34d97007 | 5652 | btrfs_handle_fs_error(fs_info, ret, |
79787eaa JM |
5653 | "Couldn't find tree log root."); |
5654 | goto error; | |
5655 | } | |
e02119d5 CM |
5656 | if (ret > 0) { |
5657 | if (path->slots[0] == 0) | |
5658 | break; | |
5659 | path->slots[0]--; | |
5660 | } | |
5661 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, | |
5662 | path->slots[0]); | |
b3b4aa74 | 5663 | btrfs_release_path(path); |
e02119d5 CM |
5664 | if (found_key.objectid != BTRFS_TREE_LOG_OBJECTID) |
5665 | break; | |
5666 | ||
cb517eab | 5667 | log = btrfs_read_fs_root(log_root_tree, &found_key); |
79787eaa JM |
5668 | if (IS_ERR(log)) { |
5669 | ret = PTR_ERR(log); | |
34d97007 | 5670 | btrfs_handle_fs_error(fs_info, ret, |
79787eaa JM |
5671 | "Couldn't read tree log root."); |
5672 | goto error; | |
5673 | } | |
e02119d5 CM |
5674 | |
5675 | tmp_key.objectid = found_key.offset; | |
5676 | tmp_key.type = BTRFS_ROOT_ITEM_KEY; | |
5677 | tmp_key.offset = (u64)-1; | |
5678 | ||
5679 | wc.replay_dest = btrfs_read_fs_root_no_name(fs_info, &tmp_key); | |
79787eaa JM |
5680 | if (IS_ERR(wc.replay_dest)) { |
5681 | ret = PTR_ERR(wc.replay_dest); | |
b50c6e25 JB |
5682 | free_extent_buffer(log->node); |
5683 | free_extent_buffer(log->commit_root); | |
5684 | kfree(log); | |
5d163e0e JM |
5685 | btrfs_handle_fs_error(fs_info, ret, |
5686 | "Couldn't read target root for tree log recovery."); | |
79787eaa JM |
5687 | goto error; |
5688 | } | |
e02119d5 | 5689 | |
07d400a6 | 5690 | wc.replay_dest->log_root = log; |
5d4f98a2 | 5691 | btrfs_record_root_in_trans(trans, wc.replay_dest); |
e02119d5 | 5692 | ret = walk_log_tree(trans, log, &wc); |
e02119d5 | 5693 | |
b50c6e25 | 5694 | if (!ret && wc.stage == LOG_WALK_REPLAY_ALL) { |
e02119d5 CM |
5695 | ret = fixup_inode_link_counts(trans, wc.replay_dest, |
5696 | path); | |
e02119d5 CM |
5697 | } |
5698 | ||
5699 | key.offset = found_key.offset - 1; | |
07d400a6 | 5700 | wc.replay_dest->log_root = NULL; |
e02119d5 | 5701 | free_extent_buffer(log->node); |
b263c2c8 | 5702 | free_extent_buffer(log->commit_root); |
e02119d5 CM |
5703 | kfree(log); |
5704 | ||
b50c6e25 JB |
5705 | if (ret) |
5706 | goto error; | |
5707 | ||
e02119d5 CM |
5708 | if (found_key.offset == 0) |
5709 | break; | |
5710 | } | |
b3b4aa74 | 5711 | btrfs_release_path(path); |
e02119d5 CM |
5712 | |
5713 | /* step one is to pin it all, step two is to replay just inodes */ | |
5714 | if (wc.pin) { | |
5715 | wc.pin = 0; | |
5716 | wc.process_func = replay_one_buffer; | |
5717 | wc.stage = LOG_WALK_REPLAY_INODES; | |
5718 | goto again; | |
5719 | } | |
5720 | /* step three is to replay everything */ | |
5721 | if (wc.stage < LOG_WALK_REPLAY_ALL) { | |
5722 | wc.stage++; | |
5723 | goto again; | |
5724 | } | |
5725 | ||
5726 | btrfs_free_path(path); | |
5727 | ||
abefa55a | 5728 | /* step 4: commit the transaction, which also unpins the blocks */ |
3a45bb20 | 5729 | ret = btrfs_commit_transaction(trans); |
abefa55a JB |
5730 | if (ret) |
5731 | return ret; | |
5732 | ||
e02119d5 CM |
5733 | free_extent_buffer(log_root_tree->node); |
5734 | log_root_tree->log_root = NULL; | |
afcdd129 | 5735 | clear_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags); |
e02119d5 | 5736 | kfree(log_root_tree); |
79787eaa | 5737 | |
abefa55a | 5738 | return 0; |
79787eaa | 5739 | error: |
b50c6e25 | 5740 | if (wc.trans) |
3a45bb20 | 5741 | btrfs_end_transaction(wc.trans); |
79787eaa JM |
5742 | btrfs_free_path(path); |
5743 | return ret; | |
e02119d5 | 5744 | } |
12fcfd22 CM |
5745 | |
5746 | /* | |
5747 | * there are some corner cases where we want to force a full | |
5748 | * commit instead of allowing a directory to be logged. | |
5749 | * | |
5750 | * They revolve around files there were unlinked from the directory, and | |
5751 | * this function updates the parent directory so that a full commit is | |
5752 | * properly done if it is fsync'd later after the unlinks are done. | |
2be63d5c FM |
5753 | * |
5754 | * Must be called before the unlink operations (updates to the subvolume tree, | |
5755 | * inodes, etc) are done. | |
12fcfd22 CM |
5756 | */ |
5757 | void btrfs_record_unlink_dir(struct btrfs_trans_handle *trans, | |
4176bdbf | 5758 | struct btrfs_inode *dir, struct btrfs_inode *inode, |
12fcfd22 CM |
5759 | int for_rename) |
5760 | { | |
af4176b4 CM |
5761 | /* |
5762 | * when we're logging a file, if it hasn't been renamed | |
5763 | * or unlinked, and its inode is fully committed on disk, | |
5764 | * we don't have to worry about walking up the directory chain | |
5765 | * to log its parents. | |
5766 | * | |
5767 | * So, we use the last_unlink_trans field to put this transid | |
5768 | * into the file. When the file is logged we check it and | |
5769 | * don't log the parents if the file is fully on disk. | |
5770 | */ | |
4176bdbf NB |
5771 | mutex_lock(&inode->log_mutex); |
5772 | inode->last_unlink_trans = trans->transid; | |
5773 | mutex_unlock(&inode->log_mutex); | |
af4176b4 | 5774 | |
12fcfd22 CM |
5775 | /* |
5776 | * if this directory was already logged any new | |
5777 | * names for this file/dir will get recorded | |
5778 | */ | |
5779 | smp_mb(); | |
4176bdbf | 5780 | if (dir->logged_trans == trans->transid) |
12fcfd22 CM |
5781 | return; |
5782 | ||
5783 | /* | |
5784 | * if the inode we're about to unlink was logged, | |
5785 | * the log will be properly updated for any new names | |
5786 | */ | |
4176bdbf | 5787 | if (inode->logged_trans == trans->transid) |
12fcfd22 CM |
5788 | return; |
5789 | ||
5790 | /* | |
5791 | * when renaming files across directories, if the directory | |
5792 | * there we're unlinking from gets fsync'd later on, there's | |
5793 | * no way to find the destination directory later and fsync it | |
5794 | * properly. So, we have to be conservative and force commits | |
5795 | * so the new name gets discovered. | |
5796 | */ | |
5797 | if (for_rename) | |
5798 | goto record; | |
5799 | ||
5800 | /* we can safely do the unlink without any special recording */ | |
5801 | return; | |
5802 | ||
5803 | record: | |
4176bdbf NB |
5804 | mutex_lock(&dir->log_mutex); |
5805 | dir->last_unlink_trans = trans->transid; | |
5806 | mutex_unlock(&dir->log_mutex); | |
1ec9a1ae FM |
5807 | } |
5808 | ||
5809 | /* | |
5810 | * Make sure that if someone attempts to fsync the parent directory of a deleted | |
5811 | * snapshot, it ends up triggering a transaction commit. This is to guarantee | |
5812 | * that after replaying the log tree of the parent directory's root we will not | |
5813 | * see the snapshot anymore and at log replay time we will not see any log tree | |
5814 | * corresponding to the deleted snapshot's root, which could lead to replaying | |
5815 | * it after replaying the log tree of the parent directory (which would replay | |
5816 | * the snapshot delete operation). | |
2be63d5c FM |
5817 | * |
5818 | * Must be called before the actual snapshot destroy operation (updates to the | |
5819 | * parent root and tree of tree roots trees, etc) are done. | |
1ec9a1ae FM |
5820 | */ |
5821 | void btrfs_record_snapshot_destroy(struct btrfs_trans_handle *trans, | |
43663557 | 5822 | struct btrfs_inode *dir) |
1ec9a1ae | 5823 | { |
43663557 NB |
5824 | mutex_lock(&dir->log_mutex); |
5825 | dir->last_unlink_trans = trans->transid; | |
5826 | mutex_unlock(&dir->log_mutex); | |
12fcfd22 CM |
5827 | } |
5828 | ||
5829 | /* | |
5830 | * Call this after adding a new name for a file and it will properly | |
5831 | * update the log to reflect the new name. | |
5832 | * | |
5833 | * It will return zero if all goes well, and it will return 1 if a | |
5834 | * full transaction commit is required. | |
5835 | */ | |
5836 | int btrfs_log_new_name(struct btrfs_trans_handle *trans, | |
9ca5fbfb | 5837 | struct btrfs_inode *inode, struct btrfs_inode *old_dir, |
12fcfd22 CM |
5838 | struct dentry *parent) |
5839 | { | |
9ca5fbfb | 5840 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb); |
f85b7379 | 5841 | struct btrfs_root *root = inode->root; |
12fcfd22 | 5842 | |
af4176b4 CM |
5843 | /* |
5844 | * this will force the logging code to walk the dentry chain | |
5845 | * up for the file | |
5846 | */ | |
9ca5fbfb NB |
5847 | if (S_ISREG(inode->vfs_inode.i_mode)) |
5848 | inode->last_unlink_trans = trans->transid; | |
af4176b4 | 5849 | |
12fcfd22 CM |
5850 | /* |
5851 | * if this inode hasn't been logged and directory we're renaming it | |
5852 | * from hasn't been logged, we don't need to log it | |
5853 | */ | |
9ca5fbfb NB |
5854 | if (inode->logged_trans <= fs_info->last_trans_committed && |
5855 | (!old_dir || old_dir->logged_trans <= fs_info->last_trans_committed)) | |
12fcfd22 CM |
5856 | return 0; |
5857 | ||
19df27a9 | 5858 | return btrfs_log_inode_parent(trans, root, inode, parent, 0, |
49dae1bc | 5859 | LLONG_MAX, 1, NULL); |
12fcfd22 CM |
5860 | } |
5861 |