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, |
49dae1bc FM |
100 | struct btrfs_root *root, struct inode *inode, |
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 | */ | |
4a0cc7ca | 634 | ret = btrfs_lookup_file_extent(trans, root, path, btrfs_ino(BTRFS_I(inode)), |
e02119d5 CM |
635 | start, 0); |
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 | ||
676 | ret = btrfs_insert_empty_item(trans, root, path, key, | |
677 | sizeof(*item)); | |
3650860b JB |
678 | if (ret) |
679 | goto out; | |
07d400a6 YZ |
680 | dest_offset = btrfs_item_ptr_offset(path->nodes[0], |
681 | path->slots[0]); | |
682 | copy_extent_buffer(path->nodes[0], eb, dest_offset, | |
683 | (unsigned long)item, sizeof(*item)); | |
684 | ||
685 | ins.objectid = btrfs_file_extent_disk_bytenr(eb, item); | |
686 | ins.offset = btrfs_file_extent_disk_num_bytes(eb, item); | |
687 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
5d4f98a2 | 688 | offset = key->offset - btrfs_file_extent_offset(eb, item); |
07d400a6 | 689 | |
df2c95f3 QW |
690 | /* |
691 | * Manually record dirty extent, as here we did a shallow | |
692 | * file extent item copy and skip normal backref update, | |
693 | * but modifying extent tree all by ourselves. | |
694 | * So need to manually record dirty extent for qgroup, | |
695 | * as the owner of the file extent changed from log tree | |
696 | * (doesn't affect qgroup) to fs/file tree(affects qgroup) | |
697 | */ | |
0b246afa | 698 | ret = btrfs_qgroup_trace_extent(trans, fs_info, |
df2c95f3 QW |
699 | btrfs_file_extent_disk_bytenr(eb, item), |
700 | btrfs_file_extent_disk_num_bytes(eb, item), | |
701 | GFP_NOFS); | |
702 | if (ret < 0) | |
703 | goto out; | |
704 | ||
07d400a6 YZ |
705 | if (ins.objectid > 0) { |
706 | u64 csum_start; | |
707 | u64 csum_end; | |
708 | LIST_HEAD(ordered_sums); | |
709 | /* | |
710 | * is this extent already allocated in the extent | |
711 | * allocation tree? If so, just add a reference | |
712 | */ | |
2ff7e61e | 713 | ret = btrfs_lookup_data_extent(fs_info, ins.objectid, |
07d400a6 YZ |
714 | ins.offset); |
715 | if (ret == 0) { | |
2ff7e61e | 716 | ret = btrfs_inc_extent_ref(trans, fs_info, |
07d400a6 | 717 | ins.objectid, ins.offset, |
5d4f98a2 | 718 | 0, root->root_key.objectid, |
b06c4bf5 | 719 | key->objectid, offset); |
b50c6e25 JB |
720 | if (ret) |
721 | goto out; | |
07d400a6 YZ |
722 | } else { |
723 | /* | |
724 | * insert the extent pointer in the extent | |
725 | * allocation tree | |
726 | */ | |
5d4f98a2 | 727 | ret = btrfs_alloc_logged_file_extent(trans, |
2ff7e61e JM |
728 | fs_info, |
729 | root->root_key.objectid, | |
5d4f98a2 | 730 | key->objectid, offset, &ins); |
b50c6e25 JB |
731 | if (ret) |
732 | goto out; | |
07d400a6 | 733 | } |
b3b4aa74 | 734 | btrfs_release_path(path); |
07d400a6 YZ |
735 | |
736 | if (btrfs_file_extent_compression(eb, item)) { | |
737 | csum_start = ins.objectid; | |
738 | csum_end = csum_start + ins.offset; | |
739 | } else { | |
740 | csum_start = ins.objectid + | |
741 | btrfs_file_extent_offset(eb, item); | |
742 | csum_end = csum_start + | |
743 | btrfs_file_extent_num_bytes(eb, item); | |
744 | } | |
745 | ||
746 | ret = btrfs_lookup_csums_range(root->log_root, | |
747 | csum_start, csum_end - 1, | |
a2de733c | 748 | &ordered_sums, 0); |
3650860b JB |
749 | if (ret) |
750 | goto out; | |
b84b8390 FM |
751 | /* |
752 | * Now delete all existing cums in the csum root that | |
753 | * cover our range. We do this because we can have an | |
754 | * extent that is completely referenced by one file | |
755 | * extent item and partially referenced by another | |
756 | * file extent item (like after using the clone or | |
757 | * extent_same ioctls). In this case if we end up doing | |
758 | * the replay of the one that partially references the | |
759 | * extent first, and we do not do the csum deletion | |
760 | * below, we can get 2 csum items in the csum tree that | |
761 | * overlap each other. For example, imagine our log has | |
762 | * the two following file extent items: | |
763 | * | |
764 | * key (257 EXTENT_DATA 409600) | |
765 | * extent data disk byte 12845056 nr 102400 | |
766 | * extent data offset 20480 nr 20480 ram 102400 | |
767 | * | |
768 | * key (257 EXTENT_DATA 819200) | |
769 | * extent data disk byte 12845056 nr 102400 | |
770 | * extent data offset 0 nr 102400 ram 102400 | |
771 | * | |
772 | * Where the second one fully references the 100K extent | |
773 | * that starts at disk byte 12845056, and the log tree | |
774 | * has a single csum item that covers the entire range | |
775 | * of the extent: | |
776 | * | |
777 | * key (EXTENT_CSUM EXTENT_CSUM 12845056) itemsize 100 | |
778 | * | |
779 | * After the first file extent item is replayed, the | |
780 | * csum tree gets the following csum item: | |
781 | * | |
782 | * key (EXTENT_CSUM EXTENT_CSUM 12865536) itemsize 20 | |
783 | * | |
784 | * Which covers the 20K sub-range starting at offset 20K | |
785 | * of our extent. Now when we replay the second file | |
786 | * extent item, if we do not delete existing csum items | |
787 | * that cover any of its blocks, we end up getting two | |
788 | * csum items in our csum tree that overlap each other: | |
789 | * | |
790 | * key (EXTENT_CSUM EXTENT_CSUM 12845056) itemsize 100 | |
791 | * key (EXTENT_CSUM EXTENT_CSUM 12865536) itemsize 20 | |
792 | * | |
793 | * Which is a problem, because after this anyone trying | |
794 | * to lookup up for the checksum of any block of our | |
795 | * extent starting at an offset of 40K or higher, will | |
796 | * end up looking at the second csum item only, which | |
797 | * does not contain the checksum for any block starting | |
798 | * at offset 40K or higher of our extent. | |
799 | */ | |
07d400a6 YZ |
800 | while (!list_empty(&ordered_sums)) { |
801 | struct btrfs_ordered_sum *sums; | |
802 | sums = list_entry(ordered_sums.next, | |
803 | struct btrfs_ordered_sum, | |
804 | list); | |
b84b8390 | 805 | if (!ret) |
0b246afa | 806 | ret = btrfs_del_csums(trans, fs_info, |
5b4aacef JM |
807 | sums->bytenr, |
808 | sums->len); | |
3650860b JB |
809 | if (!ret) |
810 | ret = btrfs_csum_file_blocks(trans, | |
0b246afa | 811 | fs_info->csum_root, sums); |
07d400a6 YZ |
812 | list_del(&sums->list); |
813 | kfree(sums); | |
814 | } | |
3650860b JB |
815 | if (ret) |
816 | goto out; | |
07d400a6 | 817 | } else { |
b3b4aa74 | 818 | btrfs_release_path(path); |
07d400a6 YZ |
819 | } |
820 | } else if (found_type == BTRFS_FILE_EXTENT_INLINE) { | |
821 | /* inline extents are easy, we just overwrite them */ | |
822 | ret = overwrite_item(trans, root, path, eb, slot, key); | |
3650860b JB |
823 | if (ret) |
824 | goto out; | |
07d400a6 | 825 | } |
e02119d5 | 826 | |
4bc4bee4 | 827 | inode_add_bytes(inode, nbytes); |
b9959295 | 828 | ret = btrfs_update_inode(trans, root, inode); |
e02119d5 CM |
829 | out: |
830 | if (inode) | |
831 | iput(inode); | |
832 | return ret; | |
833 | } | |
834 | ||
835 | /* | |
836 | * when cleaning up conflicts between the directory names in the | |
837 | * subvolume, directory names in the log and directory names in the | |
838 | * inode back references, we may have to unlink inodes from directories. | |
839 | * | |
840 | * This is a helper function to do the unlink of a specific directory | |
841 | * item | |
842 | */ | |
843 | static noinline int drop_one_dir_item(struct btrfs_trans_handle *trans, | |
844 | struct btrfs_root *root, | |
845 | struct btrfs_path *path, | |
846 | struct inode *dir, | |
847 | struct btrfs_dir_item *di) | |
848 | { | |
2ff7e61e | 849 | struct btrfs_fs_info *fs_info = root->fs_info; |
e02119d5 CM |
850 | struct inode *inode; |
851 | char *name; | |
852 | int name_len; | |
853 | struct extent_buffer *leaf; | |
854 | struct btrfs_key location; | |
855 | int ret; | |
856 | ||
857 | leaf = path->nodes[0]; | |
858 | ||
859 | btrfs_dir_item_key_to_cpu(leaf, di, &location); | |
860 | name_len = btrfs_dir_name_len(leaf, di); | |
861 | name = kmalloc(name_len, GFP_NOFS); | |
2a29edc6 | 862 | if (!name) |
863 | return -ENOMEM; | |
864 | ||
e02119d5 | 865 | read_extent_buffer(leaf, name, (unsigned long)(di + 1), name_len); |
b3b4aa74 | 866 | btrfs_release_path(path); |
e02119d5 CM |
867 | |
868 | inode = read_one_inode(root, location.objectid); | |
c00e9493 | 869 | if (!inode) { |
3650860b JB |
870 | ret = -EIO; |
871 | goto out; | |
c00e9493 | 872 | } |
e02119d5 | 873 | |
ec051c0f | 874 | ret = link_to_fixup_dir(trans, root, path, location.objectid); |
3650860b JB |
875 | if (ret) |
876 | goto out; | |
12fcfd22 | 877 | |
e02119d5 | 878 | ret = btrfs_unlink_inode(trans, root, dir, inode, name, name_len); |
3650860b JB |
879 | if (ret) |
880 | goto out; | |
ada9af21 | 881 | else |
2ff7e61e | 882 | ret = btrfs_run_delayed_items(trans, fs_info); |
3650860b | 883 | out: |
e02119d5 | 884 | kfree(name); |
e02119d5 CM |
885 | iput(inode); |
886 | return ret; | |
887 | } | |
888 | ||
889 | /* | |
890 | * helper function to see if a given name and sequence number found | |
891 | * in an inode back reference are already in a directory and correctly | |
892 | * point to this inode | |
893 | */ | |
894 | static noinline int inode_in_dir(struct btrfs_root *root, | |
895 | struct btrfs_path *path, | |
896 | u64 dirid, u64 objectid, u64 index, | |
897 | const char *name, int name_len) | |
898 | { | |
899 | struct btrfs_dir_item *di; | |
900 | struct btrfs_key location; | |
901 | int match = 0; | |
902 | ||
903 | di = btrfs_lookup_dir_index_item(NULL, root, path, dirid, | |
904 | index, name, name_len, 0); | |
905 | if (di && !IS_ERR(di)) { | |
906 | btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location); | |
907 | if (location.objectid != objectid) | |
908 | goto out; | |
909 | } else | |
910 | goto out; | |
b3b4aa74 | 911 | btrfs_release_path(path); |
e02119d5 CM |
912 | |
913 | di = btrfs_lookup_dir_item(NULL, root, path, dirid, name, name_len, 0); | |
914 | if (di && !IS_ERR(di)) { | |
915 | btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location); | |
916 | if (location.objectid != objectid) | |
917 | goto out; | |
918 | } else | |
919 | goto out; | |
920 | match = 1; | |
921 | out: | |
b3b4aa74 | 922 | btrfs_release_path(path); |
e02119d5 CM |
923 | return match; |
924 | } | |
925 | ||
926 | /* | |
927 | * helper function to check a log tree for a named back reference in | |
928 | * an inode. This is used to decide if a back reference that is | |
929 | * found in the subvolume conflicts with what we find in the log. | |
930 | * | |
931 | * inode backreferences may have multiple refs in a single item, | |
932 | * during replay we process one reference at a time, and we don't | |
933 | * want to delete valid links to a file from the subvolume if that | |
934 | * link is also in the log. | |
935 | */ | |
936 | static noinline int backref_in_log(struct btrfs_root *log, | |
937 | struct btrfs_key *key, | |
f186373f | 938 | u64 ref_objectid, |
df8d116f | 939 | const char *name, int namelen) |
e02119d5 CM |
940 | { |
941 | struct btrfs_path *path; | |
942 | struct btrfs_inode_ref *ref; | |
943 | unsigned long ptr; | |
944 | unsigned long ptr_end; | |
945 | unsigned long name_ptr; | |
946 | int found_name_len; | |
947 | int item_size; | |
948 | int ret; | |
949 | int match = 0; | |
950 | ||
951 | path = btrfs_alloc_path(); | |
2a29edc6 | 952 | if (!path) |
953 | return -ENOMEM; | |
954 | ||
e02119d5 CM |
955 | ret = btrfs_search_slot(NULL, log, key, path, 0, 0); |
956 | if (ret != 0) | |
957 | goto out; | |
958 | ||
e02119d5 | 959 | ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]); |
f186373f MF |
960 | |
961 | if (key->type == BTRFS_INODE_EXTREF_KEY) { | |
962 | if (btrfs_find_name_in_ext_backref(path, ref_objectid, | |
963 | name, namelen, NULL)) | |
964 | match = 1; | |
965 | ||
966 | goto out; | |
967 | } | |
968 | ||
969 | item_size = btrfs_item_size_nr(path->nodes[0], path->slots[0]); | |
e02119d5 CM |
970 | ptr_end = ptr + item_size; |
971 | while (ptr < ptr_end) { | |
972 | ref = (struct btrfs_inode_ref *)ptr; | |
973 | found_name_len = btrfs_inode_ref_name_len(path->nodes[0], ref); | |
974 | if (found_name_len == namelen) { | |
975 | name_ptr = (unsigned long)(ref + 1); | |
976 | ret = memcmp_extent_buffer(path->nodes[0], name, | |
977 | name_ptr, namelen); | |
978 | if (ret == 0) { | |
979 | match = 1; | |
980 | goto out; | |
981 | } | |
982 | } | |
983 | ptr = (unsigned long)(ref + 1) + found_name_len; | |
984 | } | |
985 | out: | |
986 | btrfs_free_path(path); | |
987 | return match; | |
988 | } | |
989 | ||
5a1d7843 | 990 | static inline int __add_inode_ref(struct btrfs_trans_handle *trans, |
e02119d5 | 991 | struct btrfs_root *root, |
e02119d5 | 992 | struct btrfs_path *path, |
5a1d7843 JS |
993 | struct btrfs_root *log_root, |
994 | struct inode *dir, struct inode *inode, | |
5a1d7843 | 995 | struct extent_buffer *eb, |
f186373f MF |
996 | u64 inode_objectid, u64 parent_objectid, |
997 | u64 ref_index, char *name, int namelen, | |
998 | int *search_done) | |
e02119d5 | 999 | { |
2ff7e61e | 1000 | struct btrfs_fs_info *fs_info = root->fs_info; |
34f3e4f2 | 1001 | int ret; |
f186373f MF |
1002 | char *victim_name; |
1003 | int victim_name_len; | |
1004 | struct extent_buffer *leaf; | |
5a1d7843 | 1005 | struct btrfs_dir_item *di; |
f186373f MF |
1006 | struct btrfs_key search_key; |
1007 | struct btrfs_inode_extref *extref; | |
c622ae60 | 1008 | |
f186373f MF |
1009 | again: |
1010 | /* Search old style refs */ | |
1011 | search_key.objectid = inode_objectid; | |
1012 | search_key.type = BTRFS_INODE_REF_KEY; | |
1013 | search_key.offset = parent_objectid; | |
1014 | ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0); | |
e02119d5 | 1015 | if (ret == 0) { |
e02119d5 CM |
1016 | struct btrfs_inode_ref *victim_ref; |
1017 | unsigned long ptr; | |
1018 | unsigned long ptr_end; | |
f186373f MF |
1019 | |
1020 | leaf = path->nodes[0]; | |
e02119d5 CM |
1021 | |
1022 | /* are we trying to overwrite a back ref for the root directory | |
1023 | * if so, just jump out, we're done | |
1024 | */ | |
f186373f | 1025 | if (search_key.objectid == search_key.offset) |
5a1d7843 | 1026 | return 1; |
e02119d5 CM |
1027 | |
1028 | /* check all the names in this back reference to see | |
1029 | * if they are in the log. if so, we allow them to stay | |
1030 | * otherwise they must be unlinked as a conflict | |
1031 | */ | |
1032 | ptr = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
1033 | ptr_end = ptr + btrfs_item_size_nr(leaf, path->slots[0]); | |
d397712b | 1034 | while (ptr < ptr_end) { |
e02119d5 CM |
1035 | victim_ref = (struct btrfs_inode_ref *)ptr; |
1036 | victim_name_len = btrfs_inode_ref_name_len(leaf, | |
1037 | victim_ref); | |
1038 | victim_name = kmalloc(victim_name_len, GFP_NOFS); | |
3650860b JB |
1039 | if (!victim_name) |
1040 | return -ENOMEM; | |
e02119d5 CM |
1041 | |
1042 | read_extent_buffer(leaf, victim_name, | |
1043 | (unsigned long)(victim_ref + 1), | |
1044 | victim_name_len); | |
1045 | ||
f186373f MF |
1046 | if (!backref_in_log(log_root, &search_key, |
1047 | parent_objectid, | |
1048 | victim_name, | |
e02119d5 | 1049 | victim_name_len)) { |
8b558c5f | 1050 | inc_nlink(inode); |
b3b4aa74 | 1051 | btrfs_release_path(path); |
12fcfd22 | 1052 | |
e02119d5 CM |
1053 | ret = btrfs_unlink_inode(trans, root, dir, |
1054 | inode, victim_name, | |
1055 | victim_name_len); | |
f186373f | 1056 | kfree(victim_name); |
3650860b JB |
1057 | if (ret) |
1058 | return ret; | |
2ff7e61e | 1059 | ret = btrfs_run_delayed_items(trans, fs_info); |
ada9af21 FDBM |
1060 | if (ret) |
1061 | return ret; | |
f186373f MF |
1062 | *search_done = 1; |
1063 | goto again; | |
e02119d5 CM |
1064 | } |
1065 | kfree(victim_name); | |
f186373f | 1066 | |
e02119d5 CM |
1067 | ptr = (unsigned long)(victim_ref + 1) + victim_name_len; |
1068 | } | |
e02119d5 | 1069 | |
c622ae60 | 1070 | /* |
1071 | * NOTE: we have searched root tree and checked the | |
bb7ab3b9 | 1072 | * corresponding ref, it does not need to check again. |
c622ae60 | 1073 | */ |
5a1d7843 | 1074 | *search_done = 1; |
e02119d5 | 1075 | } |
b3b4aa74 | 1076 | btrfs_release_path(path); |
e02119d5 | 1077 | |
f186373f MF |
1078 | /* Same search but for extended refs */ |
1079 | extref = btrfs_lookup_inode_extref(NULL, root, path, name, namelen, | |
1080 | inode_objectid, parent_objectid, 0, | |
1081 | 0); | |
1082 | if (!IS_ERR_OR_NULL(extref)) { | |
1083 | u32 item_size; | |
1084 | u32 cur_offset = 0; | |
1085 | unsigned long base; | |
1086 | struct inode *victim_parent; | |
1087 | ||
1088 | leaf = path->nodes[0]; | |
1089 | ||
1090 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1091 | base = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
1092 | ||
1093 | while (cur_offset < item_size) { | |
dd9ef135 | 1094 | extref = (struct btrfs_inode_extref *)(base + cur_offset); |
f186373f MF |
1095 | |
1096 | victim_name_len = btrfs_inode_extref_name_len(leaf, extref); | |
1097 | ||
1098 | if (btrfs_inode_extref_parent(leaf, extref) != parent_objectid) | |
1099 | goto next; | |
1100 | ||
1101 | victim_name = kmalloc(victim_name_len, GFP_NOFS); | |
3650860b JB |
1102 | if (!victim_name) |
1103 | return -ENOMEM; | |
f186373f MF |
1104 | read_extent_buffer(leaf, victim_name, (unsigned long)&extref->name, |
1105 | victim_name_len); | |
1106 | ||
1107 | search_key.objectid = inode_objectid; | |
1108 | search_key.type = BTRFS_INODE_EXTREF_KEY; | |
1109 | search_key.offset = btrfs_extref_hash(parent_objectid, | |
1110 | victim_name, | |
1111 | victim_name_len); | |
1112 | ret = 0; | |
1113 | if (!backref_in_log(log_root, &search_key, | |
1114 | parent_objectid, victim_name, | |
1115 | victim_name_len)) { | |
1116 | ret = -ENOENT; | |
1117 | victim_parent = read_one_inode(root, | |
1118 | parent_objectid); | |
1119 | if (victim_parent) { | |
8b558c5f | 1120 | inc_nlink(inode); |
f186373f MF |
1121 | btrfs_release_path(path); |
1122 | ||
1123 | ret = btrfs_unlink_inode(trans, root, | |
1124 | victim_parent, | |
1125 | inode, | |
1126 | victim_name, | |
1127 | victim_name_len); | |
ada9af21 FDBM |
1128 | if (!ret) |
1129 | ret = btrfs_run_delayed_items( | |
2ff7e61e JM |
1130 | trans, |
1131 | fs_info); | |
f186373f | 1132 | } |
f186373f MF |
1133 | iput(victim_parent); |
1134 | kfree(victim_name); | |
3650860b JB |
1135 | if (ret) |
1136 | return ret; | |
f186373f MF |
1137 | *search_done = 1; |
1138 | goto again; | |
1139 | } | |
1140 | kfree(victim_name); | |
3650860b JB |
1141 | if (ret) |
1142 | return ret; | |
f186373f MF |
1143 | next: |
1144 | cur_offset += victim_name_len + sizeof(*extref); | |
1145 | } | |
1146 | *search_done = 1; | |
1147 | } | |
1148 | btrfs_release_path(path); | |
1149 | ||
34f3e4f2 | 1150 | /* look for a conflicting sequence number */ |
4a0cc7ca | 1151 | di = btrfs_lookup_dir_index_item(trans, root, path, btrfs_ino(BTRFS_I(dir)), |
f186373f | 1152 | ref_index, name, namelen, 0); |
34f3e4f2 | 1153 | if (di && !IS_ERR(di)) { |
1154 | ret = drop_one_dir_item(trans, root, path, dir, di); | |
3650860b JB |
1155 | if (ret) |
1156 | return ret; | |
34f3e4f2 | 1157 | } |
1158 | btrfs_release_path(path); | |
1159 | ||
1160 | /* look for a conflicing name */ | |
4a0cc7ca | 1161 | di = btrfs_lookup_dir_item(trans, root, path, btrfs_ino(BTRFS_I(dir)), |
34f3e4f2 | 1162 | name, namelen, 0); |
1163 | if (di && !IS_ERR(di)) { | |
1164 | ret = drop_one_dir_item(trans, root, path, dir, di); | |
3650860b JB |
1165 | if (ret) |
1166 | return ret; | |
34f3e4f2 | 1167 | } |
1168 | btrfs_release_path(path); | |
1169 | ||
5a1d7843 JS |
1170 | return 0; |
1171 | } | |
e02119d5 | 1172 | |
f186373f MF |
1173 | static int extref_get_fields(struct extent_buffer *eb, unsigned long ref_ptr, |
1174 | u32 *namelen, char **name, u64 *index, | |
1175 | u64 *parent_objectid) | |
1176 | { | |
1177 | struct btrfs_inode_extref *extref; | |
1178 | ||
1179 | extref = (struct btrfs_inode_extref *)ref_ptr; | |
1180 | ||
1181 | *namelen = btrfs_inode_extref_name_len(eb, extref); | |
1182 | *name = kmalloc(*namelen, GFP_NOFS); | |
1183 | if (*name == NULL) | |
1184 | return -ENOMEM; | |
1185 | ||
1186 | read_extent_buffer(eb, *name, (unsigned long)&extref->name, | |
1187 | *namelen); | |
1188 | ||
1189 | *index = btrfs_inode_extref_index(eb, extref); | |
1190 | if (parent_objectid) | |
1191 | *parent_objectid = btrfs_inode_extref_parent(eb, extref); | |
1192 | ||
1193 | return 0; | |
1194 | } | |
1195 | ||
1196 | static int ref_get_fields(struct extent_buffer *eb, unsigned long ref_ptr, | |
1197 | u32 *namelen, char **name, u64 *index) | |
1198 | { | |
1199 | struct btrfs_inode_ref *ref; | |
1200 | ||
1201 | ref = (struct btrfs_inode_ref *)ref_ptr; | |
1202 | ||
1203 | *namelen = btrfs_inode_ref_name_len(eb, ref); | |
1204 | *name = kmalloc(*namelen, GFP_NOFS); | |
1205 | if (*name == NULL) | |
1206 | return -ENOMEM; | |
1207 | ||
1208 | read_extent_buffer(eb, *name, (unsigned long)(ref + 1), *namelen); | |
1209 | ||
1210 | *index = btrfs_inode_ref_index(eb, ref); | |
1211 | ||
1212 | return 0; | |
1213 | } | |
1214 | ||
5a1d7843 JS |
1215 | /* |
1216 | * replay one inode back reference item found in the log tree. | |
1217 | * eb, slot and key refer to the buffer and key found in the log tree. | |
1218 | * root is the destination we are replaying into, and path is for temp | |
1219 | * use by this function. (it should be released on return). | |
1220 | */ | |
1221 | static noinline int add_inode_ref(struct btrfs_trans_handle *trans, | |
1222 | struct btrfs_root *root, | |
1223 | struct btrfs_root *log, | |
1224 | struct btrfs_path *path, | |
1225 | struct extent_buffer *eb, int slot, | |
1226 | struct btrfs_key *key) | |
1227 | { | |
03b2f08b GB |
1228 | struct inode *dir = NULL; |
1229 | struct inode *inode = NULL; | |
5a1d7843 JS |
1230 | unsigned long ref_ptr; |
1231 | unsigned long ref_end; | |
03b2f08b | 1232 | char *name = NULL; |
5a1d7843 JS |
1233 | int namelen; |
1234 | int ret; | |
1235 | int search_done = 0; | |
f186373f MF |
1236 | int log_ref_ver = 0; |
1237 | u64 parent_objectid; | |
1238 | u64 inode_objectid; | |
f46dbe3d | 1239 | u64 ref_index = 0; |
f186373f MF |
1240 | int ref_struct_size; |
1241 | ||
1242 | ref_ptr = btrfs_item_ptr_offset(eb, slot); | |
1243 | ref_end = ref_ptr + btrfs_item_size_nr(eb, slot); | |
1244 | ||
1245 | if (key->type == BTRFS_INODE_EXTREF_KEY) { | |
1246 | struct btrfs_inode_extref *r; | |
1247 | ||
1248 | ref_struct_size = sizeof(struct btrfs_inode_extref); | |
1249 | log_ref_ver = 1; | |
1250 | r = (struct btrfs_inode_extref *)ref_ptr; | |
1251 | parent_objectid = btrfs_inode_extref_parent(eb, r); | |
1252 | } else { | |
1253 | ref_struct_size = sizeof(struct btrfs_inode_ref); | |
1254 | parent_objectid = key->offset; | |
1255 | } | |
1256 | inode_objectid = key->objectid; | |
e02119d5 | 1257 | |
5a1d7843 JS |
1258 | /* |
1259 | * it is possible that we didn't log all the parent directories | |
1260 | * for a given inode. If we don't find the dir, just don't | |
1261 | * copy the back ref in. The link count fixup code will take | |
1262 | * care of the rest | |
1263 | */ | |
f186373f | 1264 | dir = read_one_inode(root, parent_objectid); |
03b2f08b GB |
1265 | if (!dir) { |
1266 | ret = -ENOENT; | |
1267 | goto out; | |
1268 | } | |
5a1d7843 | 1269 | |
f186373f | 1270 | inode = read_one_inode(root, inode_objectid); |
5a1d7843 | 1271 | if (!inode) { |
03b2f08b GB |
1272 | ret = -EIO; |
1273 | goto out; | |
5a1d7843 JS |
1274 | } |
1275 | ||
5a1d7843 | 1276 | while (ref_ptr < ref_end) { |
f186373f MF |
1277 | if (log_ref_ver) { |
1278 | ret = extref_get_fields(eb, ref_ptr, &namelen, &name, | |
1279 | &ref_index, &parent_objectid); | |
1280 | /* | |
1281 | * parent object can change from one array | |
1282 | * item to another. | |
1283 | */ | |
1284 | if (!dir) | |
1285 | dir = read_one_inode(root, parent_objectid); | |
03b2f08b GB |
1286 | if (!dir) { |
1287 | ret = -ENOENT; | |
1288 | goto out; | |
1289 | } | |
f186373f MF |
1290 | } else { |
1291 | ret = ref_get_fields(eb, ref_ptr, &namelen, &name, | |
1292 | &ref_index); | |
1293 | } | |
1294 | if (ret) | |
03b2f08b | 1295 | goto out; |
5a1d7843 JS |
1296 | |
1297 | /* if we already have a perfect match, we're done */ | |
4a0cc7ca | 1298 | if (!inode_in_dir(root, path, btrfs_ino(BTRFS_I(dir)), btrfs_ino(BTRFS_I(inode)), |
f186373f | 1299 | ref_index, name, namelen)) { |
5a1d7843 JS |
1300 | /* |
1301 | * look for a conflicting back reference in the | |
1302 | * metadata. if we find one we have to unlink that name | |
1303 | * of the file before we add our new link. Later on, we | |
1304 | * overwrite any existing back reference, and we don't | |
1305 | * want to create dangling pointers in the directory. | |
1306 | */ | |
1307 | ||
1308 | if (!search_done) { | |
1309 | ret = __add_inode_ref(trans, root, path, log, | |
f186373f MF |
1310 | dir, inode, eb, |
1311 | inode_objectid, | |
1312 | parent_objectid, | |
1313 | ref_index, name, namelen, | |
5a1d7843 | 1314 | &search_done); |
03b2f08b GB |
1315 | if (ret) { |
1316 | if (ret == 1) | |
1317 | ret = 0; | |
3650860b JB |
1318 | goto out; |
1319 | } | |
5a1d7843 JS |
1320 | } |
1321 | ||
1322 | /* insert our name */ | |
1323 | ret = btrfs_add_link(trans, dir, inode, name, namelen, | |
f186373f | 1324 | 0, ref_index); |
3650860b JB |
1325 | if (ret) |
1326 | goto out; | |
5a1d7843 JS |
1327 | |
1328 | btrfs_update_inode(trans, root, inode); | |
1329 | } | |
1330 | ||
f186373f | 1331 | ref_ptr = (unsigned long)(ref_ptr + ref_struct_size) + namelen; |
5a1d7843 | 1332 | kfree(name); |
03b2f08b | 1333 | name = NULL; |
f186373f MF |
1334 | if (log_ref_ver) { |
1335 | iput(dir); | |
1336 | dir = NULL; | |
1337 | } | |
5a1d7843 | 1338 | } |
e02119d5 CM |
1339 | |
1340 | /* finally write the back reference in the inode */ | |
1341 | ret = overwrite_item(trans, root, path, eb, slot, key); | |
5a1d7843 | 1342 | out: |
b3b4aa74 | 1343 | btrfs_release_path(path); |
03b2f08b | 1344 | kfree(name); |
e02119d5 CM |
1345 | iput(dir); |
1346 | iput(inode); | |
3650860b | 1347 | return ret; |
e02119d5 CM |
1348 | } |
1349 | ||
c71bf099 | 1350 | static int insert_orphan_item(struct btrfs_trans_handle *trans, |
9c4f61f0 | 1351 | struct btrfs_root *root, u64 ino) |
c71bf099 YZ |
1352 | { |
1353 | int ret; | |
381cf658 | 1354 | |
9c4f61f0 DS |
1355 | ret = btrfs_insert_orphan_item(trans, root, ino); |
1356 | if (ret == -EEXIST) | |
1357 | ret = 0; | |
381cf658 | 1358 | |
c71bf099 YZ |
1359 | return ret; |
1360 | } | |
1361 | ||
f186373f MF |
1362 | static int count_inode_extrefs(struct btrfs_root *root, |
1363 | struct inode *inode, struct btrfs_path *path) | |
1364 | { | |
1365 | int ret = 0; | |
1366 | int name_len; | |
1367 | unsigned int nlink = 0; | |
1368 | u32 item_size; | |
1369 | u32 cur_offset = 0; | |
4a0cc7ca | 1370 | u64 inode_objectid = btrfs_ino(BTRFS_I(inode)); |
f186373f MF |
1371 | u64 offset = 0; |
1372 | unsigned long ptr; | |
1373 | struct btrfs_inode_extref *extref; | |
1374 | struct extent_buffer *leaf; | |
1375 | ||
1376 | while (1) { | |
1377 | ret = btrfs_find_one_extref(root, inode_objectid, offset, path, | |
1378 | &extref, &offset); | |
1379 | if (ret) | |
1380 | break; | |
c71bf099 | 1381 | |
f186373f MF |
1382 | leaf = path->nodes[0]; |
1383 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1384 | ptr = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
2c2c452b | 1385 | cur_offset = 0; |
f186373f MF |
1386 | |
1387 | while (cur_offset < item_size) { | |
1388 | extref = (struct btrfs_inode_extref *) (ptr + cur_offset); | |
1389 | name_len = btrfs_inode_extref_name_len(leaf, extref); | |
1390 | ||
1391 | nlink++; | |
1392 | ||
1393 | cur_offset += name_len + sizeof(*extref); | |
1394 | } | |
1395 | ||
1396 | offset++; | |
1397 | btrfs_release_path(path); | |
1398 | } | |
1399 | btrfs_release_path(path); | |
1400 | ||
2c2c452b | 1401 | if (ret < 0 && ret != -ENOENT) |
f186373f MF |
1402 | return ret; |
1403 | return nlink; | |
1404 | } | |
1405 | ||
1406 | static int count_inode_refs(struct btrfs_root *root, | |
1407 | struct inode *inode, struct btrfs_path *path) | |
e02119d5 | 1408 | { |
e02119d5 CM |
1409 | int ret; |
1410 | struct btrfs_key key; | |
f186373f | 1411 | unsigned int nlink = 0; |
e02119d5 CM |
1412 | unsigned long ptr; |
1413 | unsigned long ptr_end; | |
1414 | int name_len; | |
4a0cc7ca | 1415 | u64 ino = btrfs_ino(BTRFS_I(inode)); |
e02119d5 | 1416 | |
33345d01 | 1417 | key.objectid = ino; |
e02119d5 CM |
1418 | key.type = BTRFS_INODE_REF_KEY; |
1419 | key.offset = (u64)-1; | |
1420 | ||
d397712b | 1421 | while (1) { |
e02119d5 CM |
1422 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
1423 | if (ret < 0) | |
1424 | break; | |
1425 | if (ret > 0) { | |
1426 | if (path->slots[0] == 0) | |
1427 | break; | |
1428 | path->slots[0]--; | |
1429 | } | |
e93ae26f | 1430 | process_slot: |
e02119d5 CM |
1431 | btrfs_item_key_to_cpu(path->nodes[0], &key, |
1432 | path->slots[0]); | |
33345d01 | 1433 | if (key.objectid != ino || |
e02119d5 CM |
1434 | key.type != BTRFS_INODE_REF_KEY) |
1435 | break; | |
1436 | ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]); | |
1437 | ptr_end = ptr + btrfs_item_size_nr(path->nodes[0], | |
1438 | path->slots[0]); | |
d397712b | 1439 | while (ptr < ptr_end) { |
e02119d5 CM |
1440 | struct btrfs_inode_ref *ref; |
1441 | ||
1442 | ref = (struct btrfs_inode_ref *)ptr; | |
1443 | name_len = btrfs_inode_ref_name_len(path->nodes[0], | |
1444 | ref); | |
1445 | ptr = (unsigned long)(ref + 1) + name_len; | |
1446 | nlink++; | |
1447 | } | |
1448 | ||
1449 | if (key.offset == 0) | |
1450 | break; | |
e93ae26f FDBM |
1451 | if (path->slots[0] > 0) { |
1452 | path->slots[0]--; | |
1453 | goto process_slot; | |
1454 | } | |
e02119d5 | 1455 | key.offset--; |
b3b4aa74 | 1456 | btrfs_release_path(path); |
e02119d5 | 1457 | } |
b3b4aa74 | 1458 | btrfs_release_path(path); |
f186373f MF |
1459 | |
1460 | return nlink; | |
1461 | } | |
1462 | ||
1463 | /* | |
1464 | * There are a few corners where the link count of the file can't | |
1465 | * be properly maintained during replay. So, instead of adding | |
1466 | * lots of complexity to the log code, we just scan the backrefs | |
1467 | * for any file that has been through replay. | |
1468 | * | |
1469 | * The scan will update the link count on the inode to reflect the | |
1470 | * number of back refs found. If it goes down to zero, the iput | |
1471 | * will free the inode. | |
1472 | */ | |
1473 | static noinline int fixup_inode_link_count(struct btrfs_trans_handle *trans, | |
1474 | struct btrfs_root *root, | |
1475 | struct inode *inode) | |
1476 | { | |
1477 | struct btrfs_path *path; | |
1478 | int ret; | |
1479 | u64 nlink = 0; | |
4a0cc7ca | 1480 | u64 ino = btrfs_ino(BTRFS_I(inode)); |
f186373f MF |
1481 | |
1482 | path = btrfs_alloc_path(); | |
1483 | if (!path) | |
1484 | return -ENOMEM; | |
1485 | ||
1486 | ret = count_inode_refs(root, inode, path); | |
1487 | if (ret < 0) | |
1488 | goto out; | |
1489 | ||
1490 | nlink = ret; | |
1491 | ||
1492 | ret = count_inode_extrefs(root, inode, path); | |
f186373f MF |
1493 | if (ret < 0) |
1494 | goto out; | |
1495 | ||
1496 | nlink += ret; | |
1497 | ||
1498 | ret = 0; | |
1499 | ||
e02119d5 | 1500 | if (nlink != inode->i_nlink) { |
bfe86848 | 1501 | set_nlink(inode, nlink); |
e02119d5 CM |
1502 | btrfs_update_inode(trans, root, inode); |
1503 | } | |
8d5bf1cb | 1504 | BTRFS_I(inode)->index_cnt = (u64)-1; |
e02119d5 | 1505 | |
c71bf099 YZ |
1506 | if (inode->i_nlink == 0) { |
1507 | if (S_ISDIR(inode->i_mode)) { | |
1508 | ret = replay_dir_deletes(trans, root, NULL, path, | |
33345d01 | 1509 | ino, 1); |
3650860b JB |
1510 | if (ret) |
1511 | goto out; | |
c71bf099 | 1512 | } |
33345d01 | 1513 | ret = insert_orphan_item(trans, root, ino); |
12fcfd22 | 1514 | } |
12fcfd22 | 1515 | |
f186373f MF |
1516 | out: |
1517 | btrfs_free_path(path); | |
1518 | return ret; | |
e02119d5 CM |
1519 | } |
1520 | ||
1521 | static noinline int fixup_inode_link_counts(struct btrfs_trans_handle *trans, | |
1522 | struct btrfs_root *root, | |
1523 | struct btrfs_path *path) | |
1524 | { | |
1525 | int ret; | |
1526 | struct btrfs_key key; | |
1527 | struct inode *inode; | |
1528 | ||
1529 | key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID; | |
1530 | key.type = BTRFS_ORPHAN_ITEM_KEY; | |
1531 | key.offset = (u64)-1; | |
d397712b | 1532 | while (1) { |
e02119d5 CM |
1533 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
1534 | if (ret < 0) | |
1535 | break; | |
1536 | ||
1537 | if (ret == 1) { | |
1538 | if (path->slots[0] == 0) | |
1539 | break; | |
1540 | path->slots[0]--; | |
1541 | } | |
1542 | ||
1543 | btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]); | |
1544 | if (key.objectid != BTRFS_TREE_LOG_FIXUP_OBJECTID || | |
1545 | key.type != BTRFS_ORPHAN_ITEM_KEY) | |
1546 | break; | |
1547 | ||
1548 | ret = btrfs_del_item(trans, root, path); | |
65a246c5 TI |
1549 | if (ret) |
1550 | goto out; | |
e02119d5 | 1551 | |
b3b4aa74 | 1552 | btrfs_release_path(path); |
e02119d5 | 1553 | inode = read_one_inode(root, key.offset); |
c00e9493 TI |
1554 | if (!inode) |
1555 | return -EIO; | |
e02119d5 CM |
1556 | |
1557 | ret = fixup_inode_link_count(trans, root, inode); | |
e02119d5 | 1558 | iput(inode); |
3650860b JB |
1559 | if (ret) |
1560 | goto out; | |
e02119d5 | 1561 | |
12fcfd22 CM |
1562 | /* |
1563 | * fixup on a directory may create new entries, | |
1564 | * make sure we always look for the highset possible | |
1565 | * offset | |
1566 | */ | |
1567 | key.offset = (u64)-1; | |
e02119d5 | 1568 | } |
65a246c5 TI |
1569 | ret = 0; |
1570 | out: | |
b3b4aa74 | 1571 | btrfs_release_path(path); |
65a246c5 | 1572 | return ret; |
e02119d5 CM |
1573 | } |
1574 | ||
1575 | ||
1576 | /* | |
1577 | * record a given inode in the fixup dir so we can check its link | |
1578 | * count when replay is done. The link count is incremented here | |
1579 | * so the inode won't go away until we check it | |
1580 | */ | |
1581 | static noinline int link_to_fixup_dir(struct btrfs_trans_handle *trans, | |
1582 | struct btrfs_root *root, | |
1583 | struct btrfs_path *path, | |
1584 | u64 objectid) | |
1585 | { | |
1586 | struct btrfs_key key; | |
1587 | int ret = 0; | |
1588 | struct inode *inode; | |
1589 | ||
1590 | inode = read_one_inode(root, objectid); | |
c00e9493 TI |
1591 | if (!inode) |
1592 | return -EIO; | |
e02119d5 CM |
1593 | |
1594 | key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID; | |
962a298f | 1595 | key.type = BTRFS_ORPHAN_ITEM_KEY; |
e02119d5 CM |
1596 | key.offset = objectid; |
1597 | ||
1598 | ret = btrfs_insert_empty_item(trans, root, path, &key, 0); | |
1599 | ||
b3b4aa74 | 1600 | btrfs_release_path(path); |
e02119d5 | 1601 | if (ret == 0) { |
9bf7a489 JB |
1602 | if (!inode->i_nlink) |
1603 | set_nlink(inode, 1); | |
1604 | else | |
8b558c5f | 1605 | inc_nlink(inode); |
b9959295 | 1606 | ret = btrfs_update_inode(trans, root, inode); |
e02119d5 CM |
1607 | } else if (ret == -EEXIST) { |
1608 | ret = 0; | |
1609 | } else { | |
3650860b | 1610 | BUG(); /* Logic Error */ |
e02119d5 CM |
1611 | } |
1612 | iput(inode); | |
1613 | ||
1614 | return ret; | |
1615 | } | |
1616 | ||
1617 | /* | |
1618 | * when replaying the log for a directory, we only insert names | |
1619 | * for inodes that actually exist. This means an fsync on a directory | |
1620 | * does not implicitly fsync all the new files in it | |
1621 | */ | |
1622 | static noinline int insert_one_name(struct btrfs_trans_handle *trans, | |
1623 | struct btrfs_root *root, | |
e02119d5 | 1624 | u64 dirid, u64 index, |
60d53eb3 | 1625 | char *name, int name_len, |
e02119d5 CM |
1626 | struct btrfs_key *location) |
1627 | { | |
1628 | struct inode *inode; | |
1629 | struct inode *dir; | |
1630 | int ret; | |
1631 | ||
1632 | inode = read_one_inode(root, location->objectid); | |
1633 | if (!inode) | |
1634 | return -ENOENT; | |
1635 | ||
1636 | dir = read_one_inode(root, dirid); | |
1637 | if (!dir) { | |
1638 | iput(inode); | |
1639 | return -EIO; | |
1640 | } | |
d555438b | 1641 | |
e02119d5 CM |
1642 | ret = btrfs_add_link(trans, dir, inode, name, name_len, 1, index); |
1643 | ||
1644 | /* FIXME, put inode into FIXUP list */ | |
1645 | ||
1646 | iput(inode); | |
1647 | iput(dir); | |
1648 | return ret; | |
1649 | } | |
1650 | ||
df8d116f FM |
1651 | /* |
1652 | * Return true if an inode reference exists in the log for the given name, | |
1653 | * inode and parent inode. | |
1654 | */ | |
1655 | static bool name_in_log_ref(struct btrfs_root *log_root, | |
1656 | const char *name, const int name_len, | |
1657 | const u64 dirid, const u64 ino) | |
1658 | { | |
1659 | struct btrfs_key search_key; | |
1660 | ||
1661 | search_key.objectid = ino; | |
1662 | search_key.type = BTRFS_INODE_REF_KEY; | |
1663 | search_key.offset = dirid; | |
1664 | if (backref_in_log(log_root, &search_key, dirid, name, name_len)) | |
1665 | return true; | |
1666 | ||
1667 | search_key.type = BTRFS_INODE_EXTREF_KEY; | |
1668 | search_key.offset = btrfs_extref_hash(dirid, name, name_len); | |
1669 | if (backref_in_log(log_root, &search_key, dirid, name, name_len)) | |
1670 | return true; | |
1671 | ||
1672 | return false; | |
1673 | } | |
1674 | ||
e02119d5 CM |
1675 | /* |
1676 | * take a single entry in a log directory item and replay it into | |
1677 | * the subvolume. | |
1678 | * | |
1679 | * if a conflicting item exists in the subdirectory already, | |
1680 | * the inode it points to is unlinked and put into the link count | |
1681 | * fix up tree. | |
1682 | * | |
1683 | * If a name from the log points to a file or directory that does | |
1684 | * not exist in the FS, it is skipped. fsyncs on directories | |
1685 | * do not force down inodes inside that directory, just changes to the | |
1686 | * names or unlinks in a directory. | |
bb53eda9 FM |
1687 | * |
1688 | * Returns < 0 on error, 0 if the name wasn't replayed (dentry points to a | |
1689 | * non-existing inode) and 1 if the name was replayed. | |
e02119d5 CM |
1690 | */ |
1691 | static noinline int replay_one_name(struct btrfs_trans_handle *trans, | |
1692 | struct btrfs_root *root, | |
1693 | struct btrfs_path *path, | |
1694 | struct extent_buffer *eb, | |
1695 | struct btrfs_dir_item *di, | |
1696 | struct btrfs_key *key) | |
1697 | { | |
1698 | char *name; | |
1699 | int name_len; | |
1700 | struct btrfs_dir_item *dst_di; | |
1701 | struct btrfs_key found_key; | |
1702 | struct btrfs_key log_key; | |
1703 | struct inode *dir; | |
e02119d5 | 1704 | u8 log_type; |
4bef0848 | 1705 | int exists; |
3650860b | 1706 | int ret = 0; |
d555438b | 1707 | bool update_size = (key->type == BTRFS_DIR_INDEX_KEY); |
bb53eda9 | 1708 | bool name_added = false; |
e02119d5 CM |
1709 | |
1710 | dir = read_one_inode(root, key->objectid); | |
c00e9493 TI |
1711 | if (!dir) |
1712 | return -EIO; | |
e02119d5 CM |
1713 | |
1714 | name_len = btrfs_dir_name_len(eb, di); | |
1715 | name = kmalloc(name_len, GFP_NOFS); | |
2bac325e FDBM |
1716 | if (!name) { |
1717 | ret = -ENOMEM; | |
1718 | goto out; | |
1719 | } | |
2a29edc6 | 1720 | |
e02119d5 CM |
1721 | log_type = btrfs_dir_type(eb, di); |
1722 | read_extent_buffer(eb, name, (unsigned long)(di + 1), | |
1723 | name_len); | |
1724 | ||
1725 | btrfs_dir_item_key_to_cpu(eb, di, &log_key); | |
4bef0848 CM |
1726 | exists = btrfs_lookup_inode(trans, root, path, &log_key, 0); |
1727 | if (exists == 0) | |
1728 | exists = 1; | |
1729 | else | |
1730 | exists = 0; | |
b3b4aa74 | 1731 | btrfs_release_path(path); |
4bef0848 | 1732 | |
e02119d5 CM |
1733 | if (key->type == BTRFS_DIR_ITEM_KEY) { |
1734 | dst_di = btrfs_lookup_dir_item(trans, root, path, key->objectid, | |
1735 | name, name_len, 1); | |
d397712b | 1736 | } else if (key->type == BTRFS_DIR_INDEX_KEY) { |
e02119d5 CM |
1737 | dst_di = btrfs_lookup_dir_index_item(trans, root, path, |
1738 | key->objectid, | |
1739 | key->offset, name, | |
1740 | name_len, 1); | |
1741 | } else { | |
3650860b JB |
1742 | /* Corruption */ |
1743 | ret = -EINVAL; | |
1744 | goto out; | |
e02119d5 | 1745 | } |
c704005d | 1746 | if (IS_ERR_OR_NULL(dst_di)) { |
e02119d5 CM |
1747 | /* we need a sequence number to insert, so we only |
1748 | * do inserts for the BTRFS_DIR_INDEX_KEY types | |
1749 | */ | |
1750 | if (key->type != BTRFS_DIR_INDEX_KEY) | |
1751 | goto out; | |
1752 | goto insert; | |
1753 | } | |
1754 | ||
1755 | btrfs_dir_item_key_to_cpu(path->nodes[0], dst_di, &found_key); | |
1756 | /* the existing item matches the logged item */ | |
1757 | if (found_key.objectid == log_key.objectid && | |
1758 | found_key.type == log_key.type && | |
1759 | found_key.offset == log_key.offset && | |
1760 | btrfs_dir_type(path->nodes[0], dst_di) == log_type) { | |
a2cc11db | 1761 | update_size = false; |
e02119d5 CM |
1762 | goto out; |
1763 | } | |
1764 | ||
1765 | /* | |
1766 | * don't drop the conflicting directory entry if the inode | |
1767 | * for the new entry doesn't exist | |
1768 | */ | |
4bef0848 | 1769 | if (!exists) |
e02119d5 CM |
1770 | goto out; |
1771 | ||
e02119d5 | 1772 | ret = drop_one_dir_item(trans, root, path, dir, dst_di); |
3650860b JB |
1773 | if (ret) |
1774 | goto out; | |
e02119d5 CM |
1775 | |
1776 | if (key->type == BTRFS_DIR_INDEX_KEY) | |
1777 | goto insert; | |
1778 | out: | |
b3b4aa74 | 1779 | btrfs_release_path(path); |
d555438b JB |
1780 | if (!ret && update_size) { |
1781 | btrfs_i_size_write(dir, dir->i_size + name_len * 2); | |
1782 | ret = btrfs_update_inode(trans, root, dir); | |
1783 | } | |
e02119d5 CM |
1784 | kfree(name); |
1785 | iput(dir); | |
bb53eda9 FM |
1786 | if (!ret && name_added) |
1787 | ret = 1; | |
3650860b | 1788 | return ret; |
e02119d5 CM |
1789 | |
1790 | insert: | |
df8d116f FM |
1791 | if (name_in_log_ref(root->log_root, name, name_len, |
1792 | key->objectid, log_key.objectid)) { | |
1793 | /* The dentry will be added later. */ | |
1794 | ret = 0; | |
1795 | update_size = false; | |
1796 | goto out; | |
1797 | } | |
b3b4aa74 | 1798 | btrfs_release_path(path); |
60d53eb3 Z |
1799 | ret = insert_one_name(trans, root, key->objectid, key->offset, |
1800 | name, name_len, &log_key); | |
df8d116f | 1801 | if (ret && ret != -ENOENT && ret != -EEXIST) |
3650860b | 1802 | goto out; |
bb53eda9 FM |
1803 | if (!ret) |
1804 | name_added = true; | |
d555438b | 1805 | update_size = false; |
3650860b | 1806 | ret = 0; |
e02119d5 CM |
1807 | goto out; |
1808 | } | |
1809 | ||
1810 | /* | |
1811 | * find all the names in a directory item and reconcile them into | |
1812 | * the subvolume. Only BTRFS_DIR_ITEM_KEY types will have more than | |
1813 | * one name in a directory item, but the same code gets used for | |
1814 | * both directory index types | |
1815 | */ | |
1816 | static noinline int replay_one_dir_item(struct btrfs_trans_handle *trans, | |
1817 | struct btrfs_root *root, | |
1818 | struct btrfs_path *path, | |
1819 | struct extent_buffer *eb, int slot, | |
1820 | struct btrfs_key *key) | |
1821 | { | |
2ff7e61e | 1822 | struct btrfs_fs_info *fs_info = root->fs_info; |
bb53eda9 | 1823 | int ret = 0; |
e02119d5 CM |
1824 | u32 item_size = btrfs_item_size_nr(eb, slot); |
1825 | struct btrfs_dir_item *di; | |
1826 | int name_len; | |
1827 | unsigned long ptr; | |
1828 | unsigned long ptr_end; | |
bb53eda9 | 1829 | struct btrfs_path *fixup_path = NULL; |
e02119d5 CM |
1830 | |
1831 | ptr = btrfs_item_ptr_offset(eb, slot); | |
1832 | ptr_end = ptr + item_size; | |
d397712b | 1833 | while (ptr < ptr_end) { |
e02119d5 | 1834 | di = (struct btrfs_dir_item *)ptr; |
2ff7e61e | 1835 | if (verify_dir_item(fs_info, eb, di)) |
22a94d44 | 1836 | return -EIO; |
e02119d5 CM |
1837 | name_len = btrfs_dir_name_len(eb, di); |
1838 | ret = replay_one_name(trans, root, path, eb, di, key); | |
bb53eda9 FM |
1839 | if (ret < 0) |
1840 | break; | |
e02119d5 CM |
1841 | ptr = (unsigned long)(di + 1); |
1842 | ptr += name_len; | |
bb53eda9 FM |
1843 | |
1844 | /* | |
1845 | * If this entry refers to a non-directory (directories can not | |
1846 | * have a link count > 1) and it was added in the transaction | |
1847 | * that was not committed, make sure we fixup the link count of | |
1848 | * the inode it the entry points to. Otherwise something like | |
1849 | * the following would result in a directory pointing to an | |
1850 | * inode with a wrong link that does not account for this dir | |
1851 | * entry: | |
1852 | * | |
1853 | * mkdir testdir | |
1854 | * touch testdir/foo | |
1855 | * touch testdir/bar | |
1856 | * sync | |
1857 | * | |
1858 | * ln testdir/bar testdir/bar_link | |
1859 | * ln testdir/foo testdir/foo_link | |
1860 | * xfs_io -c "fsync" testdir/bar | |
1861 | * | |
1862 | * <power failure> | |
1863 | * | |
1864 | * mount fs, log replay happens | |
1865 | * | |
1866 | * File foo would remain with a link count of 1 when it has two | |
1867 | * entries pointing to it in the directory testdir. This would | |
1868 | * make it impossible to ever delete the parent directory has | |
1869 | * it would result in stale dentries that can never be deleted. | |
1870 | */ | |
1871 | if (ret == 1 && btrfs_dir_type(eb, di) != BTRFS_FT_DIR) { | |
1872 | struct btrfs_key di_key; | |
1873 | ||
1874 | if (!fixup_path) { | |
1875 | fixup_path = btrfs_alloc_path(); | |
1876 | if (!fixup_path) { | |
1877 | ret = -ENOMEM; | |
1878 | break; | |
1879 | } | |
1880 | } | |
1881 | ||
1882 | btrfs_dir_item_key_to_cpu(eb, di, &di_key); | |
1883 | ret = link_to_fixup_dir(trans, root, fixup_path, | |
1884 | di_key.objectid); | |
1885 | if (ret) | |
1886 | break; | |
1887 | } | |
1888 | ret = 0; | |
e02119d5 | 1889 | } |
bb53eda9 FM |
1890 | btrfs_free_path(fixup_path); |
1891 | return ret; | |
e02119d5 CM |
1892 | } |
1893 | ||
1894 | /* | |
1895 | * directory replay has two parts. There are the standard directory | |
1896 | * items in the log copied from the subvolume, and range items | |
1897 | * created in the log while the subvolume was logged. | |
1898 | * | |
1899 | * The range items tell us which parts of the key space the log | |
1900 | * is authoritative for. During replay, if a key in the subvolume | |
1901 | * directory is in a logged range item, but not actually in the log | |
1902 | * that means it was deleted from the directory before the fsync | |
1903 | * and should be removed. | |
1904 | */ | |
1905 | static noinline int find_dir_range(struct btrfs_root *root, | |
1906 | struct btrfs_path *path, | |
1907 | u64 dirid, int key_type, | |
1908 | u64 *start_ret, u64 *end_ret) | |
1909 | { | |
1910 | struct btrfs_key key; | |
1911 | u64 found_end; | |
1912 | struct btrfs_dir_log_item *item; | |
1913 | int ret; | |
1914 | int nritems; | |
1915 | ||
1916 | if (*start_ret == (u64)-1) | |
1917 | return 1; | |
1918 | ||
1919 | key.objectid = dirid; | |
1920 | key.type = key_type; | |
1921 | key.offset = *start_ret; | |
1922 | ||
1923 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
1924 | if (ret < 0) | |
1925 | goto out; | |
1926 | if (ret > 0) { | |
1927 | if (path->slots[0] == 0) | |
1928 | goto out; | |
1929 | path->slots[0]--; | |
1930 | } | |
1931 | if (ret != 0) | |
1932 | btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]); | |
1933 | ||
1934 | if (key.type != key_type || key.objectid != dirid) { | |
1935 | ret = 1; | |
1936 | goto next; | |
1937 | } | |
1938 | item = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
1939 | struct btrfs_dir_log_item); | |
1940 | found_end = btrfs_dir_log_end(path->nodes[0], item); | |
1941 | ||
1942 | if (*start_ret >= key.offset && *start_ret <= found_end) { | |
1943 | ret = 0; | |
1944 | *start_ret = key.offset; | |
1945 | *end_ret = found_end; | |
1946 | goto out; | |
1947 | } | |
1948 | ret = 1; | |
1949 | next: | |
1950 | /* check the next slot in the tree to see if it is a valid item */ | |
1951 | nritems = btrfs_header_nritems(path->nodes[0]); | |
2a7bf53f | 1952 | path->slots[0]++; |
e02119d5 CM |
1953 | if (path->slots[0] >= nritems) { |
1954 | ret = btrfs_next_leaf(root, path); | |
1955 | if (ret) | |
1956 | goto out; | |
e02119d5 CM |
1957 | } |
1958 | ||
1959 | btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]); | |
1960 | ||
1961 | if (key.type != key_type || key.objectid != dirid) { | |
1962 | ret = 1; | |
1963 | goto out; | |
1964 | } | |
1965 | item = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
1966 | struct btrfs_dir_log_item); | |
1967 | found_end = btrfs_dir_log_end(path->nodes[0], item); | |
1968 | *start_ret = key.offset; | |
1969 | *end_ret = found_end; | |
1970 | ret = 0; | |
1971 | out: | |
b3b4aa74 | 1972 | btrfs_release_path(path); |
e02119d5 CM |
1973 | return ret; |
1974 | } | |
1975 | ||
1976 | /* | |
1977 | * this looks for a given directory item in the log. If the directory | |
1978 | * item is not in the log, the item is removed and the inode it points | |
1979 | * to is unlinked | |
1980 | */ | |
1981 | static noinline int check_item_in_log(struct btrfs_trans_handle *trans, | |
1982 | struct btrfs_root *root, | |
1983 | struct btrfs_root *log, | |
1984 | struct btrfs_path *path, | |
1985 | struct btrfs_path *log_path, | |
1986 | struct inode *dir, | |
1987 | struct btrfs_key *dir_key) | |
1988 | { | |
2ff7e61e | 1989 | struct btrfs_fs_info *fs_info = root->fs_info; |
e02119d5 CM |
1990 | int ret; |
1991 | struct extent_buffer *eb; | |
1992 | int slot; | |
1993 | u32 item_size; | |
1994 | struct btrfs_dir_item *di; | |
1995 | struct btrfs_dir_item *log_di; | |
1996 | int name_len; | |
1997 | unsigned long ptr; | |
1998 | unsigned long ptr_end; | |
1999 | char *name; | |
2000 | struct inode *inode; | |
2001 | struct btrfs_key location; | |
2002 | ||
2003 | again: | |
2004 | eb = path->nodes[0]; | |
2005 | slot = path->slots[0]; | |
2006 | item_size = btrfs_item_size_nr(eb, slot); | |
2007 | ptr = btrfs_item_ptr_offset(eb, slot); | |
2008 | ptr_end = ptr + item_size; | |
d397712b | 2009 | while (ptr < ptr_end) { |
e02119d5 | 2010 | di = (struct btrfs_dir_item *)ptr; |
2ff7e61e | 2011 | if (verify_dir_item(fs_info, eb, di)) { |
22a94d44 JB |
2012 | ret = -EIO; |
2013 | goto out; | |
2014 | } | |
2015 | ||
e02119d5 CM |
2016 | name_len = btrfs_dir_name_len(eb, di); |
2017 | name = kmalloc(name_len, GFP_NOFS); | |
2018 | if (!name) { | |
2019 | ret = -ENOMEM; | |
2020 | goto out; | |
2021 | } | |
2022 | read_extent_buffer(eb, name, (unsigned long)(di + 1), | |
2023 | name_len); | |
2024 | log_di = NULL; | |
12fcfd22 | 2025 | if (log && dir_key->type == BTRFS_DIR_ITEM_KEY) { |
e02119d5 CM |
2026 | log_di = btrfs_lookup_dir_item(trans, log, log_path, |
2027 | dir_key->objectid, | |
2028 | name, name_len, 0); | |
12fcfd22 | 2029 | } else if (log && dir_key->type == BTRFS_DIR_INDEX_KEY) { |
e02119d5 CM |
2030 | log_di = btrfs_lookup_dir_index_item(trans, log, |
2031 | log_path, | |
2032 | dir_key->objectid, | |
2033 | dir_key->offset, | |
2034 | name, name_len, 0); | |
2035 | } | |
269d040f | 2036 | if (!log_di || (IS_ERR(log_di) && PTR_ERR(log_di) == -ENOENT)) { |
e02119d5 | 2037 | btrfs_dir_item_key_to_cpu(eb, di, &location); |
b3b4aa74 DS |
2038 | btrfs_release_path(path); |
2039 | btrfs_release_path(log_path); | |
e02119d5 | 2040 | inode = read_one_inode(root, location.objectid); |
c00e9493 TI |
2041 | if (!inode) { |
2042 | kfree(name); | |
2043 | return -EIO; | |
2044 | } | |
e02119d5 CM |
2045 | |
2046 | ret = link_to_fixup_dir(trans, root, | |
2047 | path, location.objectid); | |
3650860b JB |
2048 | if (ret) { |
2049 | kfree(name); | |
2050 | iput(inode); | |
2051 | goto out; | |
2052 | } | |
2053 | ||
8b558c5f | 2054 | inc_nlink(inode); |
e02119d5 CM |
2055 | ret = btrfs_unlink_inode(trans, root, dir, inode, |
2056 | name, name_len); | |
3650860b | 2057 | if (!ret) |
2ff7e61e | 2058 | ret = btrfs_run_delayed_items(trans, fs_info); |
e02119d5 CM |
2059 | kfree(name); |
2060 | iput(inode); | |
3650860b JB |
2061 | if (ret) |
2062 | goto out; | |
e02119d5 CM |
2063 | |
2064 | /* there might still be more names under this key | |
2065 | * check and repeat if required | |
2066 | */ | |
2067 | ret = btrfs_search_slot(NULL, root, dir_key, path, | |
2068 | 0, 0); | |
2069 | if (ret == 0) | |
2070 | goto again; | |
2071 | ret = 0; | |
2072 | goto out; | |
269d040f FDBM |
2073 | } else if (IS_ERR(log_di)) { |
2074 | kfree(name); | |
2075 | return PTR_ERR(log_di); | |
e02119d5 | 2076 | } |
b3b4aa74 | 2077 | btrfs_release_path(log_path); |
e02119d5 CM |
2078 | kfree(name); |
2079 | ||
2080 | ptr = (unsigned long)(di + 1); | |
2081 | ptr += name_len; | |
2082 | } | |
2083 | ret = 0; | |
2084 | out: | |
b3b4aa74 DS |
2085 | btrfs_release_path(path); |
2086 | btrfs_release_path(log_path); | |
e02119d5 CM |
2087 | return ret; |
2088 | } | |
2089 | ||
4f764e51 FM |
2090 | static int replay_xattr_deletes(struct btrfs_trans_handle *trans, |
2091 | struct btrfs_root *root, | |
2092 | struct btrfs_root *log, | |
2093 | struct btrfs_path *path, | |
2094 | const u64 ino) | |
2095 | { | |
2096 | struct btrfs_key search_key; | |
2097 | struct btrfs_path *log_path; | |
2098 | int i; | |
2099 | int nritems; | |
2100 | int ret; | |
2101 | ||
2102 | log_path = btrfs_alloc_path(); | |
2103 | if (!log_path) | |
2104 | return -ENOMEM; | |
2105 | ||
2106 | search_key.objectid = ino; | |
2107 | search_key.type = BTRFS_XATTR_ITEM_KEY; | |
2108 | search_key.offset = 0; | |
2109 | again: | |
2110 | ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0); | |
2111 | if (ret < 0) | |
2112 | goto out; | |
2113 | process_leaf: | |
2114 | nritems = btrfs_header_nritems(path->nodes[0]); | |
2115 | for (i = path->slots[0]; i < nritems; i++) { | |
2116 | struct btrfs_key key; | |
2117 | struct btrfs_dir_item *di; | |
2118 | struct btrfs_dir_item *log_di; | |
2119 | u32 total_size; | |
2120 | u32 cur; | |
2121 | ||
2122 | btrfs_item_key_to_cpu(path->nodes[0], &key, i); | |
2123 | if (key.objectid != ino || key.type != BTRFS_XATTR_ITEM_KEY) { | |
2124 | ret = 0; | |
2125 | goto out; | |
2126 | } | |
2127 | ||
2128 | di = btrfs_item_ptr(path->nodes[0], i, struct btrfs_dir_item); | |
2129 | total_size = btrfs_item_size_nr(path->nodes[0], i); | |
2130 | cur = 0; | |
2131 | while (cur < total_size) { | |
2132 | u16 name_len = btrfs_dir_name_len(path->nodes[0], di); | |
2133 | u16 data_len = btrfs_dir_data_len(path->nodes[0], di); | |
2134 | u32 this_len = sizeof(*di) + name_len + data_len; | |
2135 | char *name; | |
2136 | ||
2137 | name = kmalloc(name_len, GFP_NOFS); | |
2138 | if (!name) { | |
2139 | ret = -ENOMEM; | |
2140 | goto out; | |
2141 | } | |
2142 | read_extent_buffer(path->nodes[0], name, | |
2143 | (unsigned long)(di + 1), name_len); | |
2144 | ||
2145 | log_di = btrfs_lookup_xattr(NULL, log, log_path, ino, | |
2146 | name, name_len, 0); | |
2147 | btrfs_release_path(log_path); | |
2148 | if (!log_di) { | |
2149 | /* Doesn't exist in log tree, so delete it. */ | |
2150 | btrfs_release_path(path); | |
2151 | di = btrfs_lookup_xattr(trans, root, path, ino, | |
2152 | name, name_len, -1); | |
2153 | kfree(name); | |
2154 | if (IS_ERR(di)) { | |
2155 | ret = PTR_ERR(di); | |
2156 | goto out; | |
2157 | } | |
2158 | ASSERT(di); | |
2159 | ret = btrfs_delete_one_dir_name(trans, root, | |
2160 | path, di); | |
2161 | if (ret) | |
2162 | goto out; | |
2163 | btrfs_release_path(path); | |
2164 | search_key = key; | |
2165 | goto again; | |
2166 | } | |
2167 | kfree(name); | |
2168 | if (IS_ERR(log_di)) { | |
2169 | ret = PTR_ERR(log_di); | |
2170 | goto out; | |
2171 | } | |
2172 | cur += this_len; | |
2173 | di = (struct btrfs_dir_item *)((char *)di + this_len); | |
2174 | } | |
2175 | } | |
2176 | ret = btrfs_next_leaf(root, path); | |
2177 | if (ret > 0) | |
2178 | ret = 0; | |
2179 | else if (ret == 0) | |
2180 | goto process_leaf; | |
2181 | out: | |
2182 | btrfs_free_path(log_path); | |
2183 | btrfs_release_path(path); | |
2184 | return ret; | |
2185 | } | |
2186 | ||
2187 | ||
e02119d5 CM |
2188 | /* |
2189 | * deletion replay happens before we copy any new directory items | |
2190 | * out of the log or out of backreferences from inodes. It | |
2191 | * scans the log to find ranges of keys that log is authoritative for, | |
2192 | * and then scans the directory to find items in those ranges that are | |
2193 | * not present in the log. | |
2194 | * | |
2195 | * Anything we don't find in the log is unlinked and removed from the | |
2196 | * directory. | |
2197 | */ | |
2198 | static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans, | |
2199 | struct btrfs_root *root, | |
2200 | struct btrfs_root *log, | |
2201 | struct btrfs_path *path, | |
12fcfd22 | 2202 | u64 dirid, int del_all) |
e02119d5 CM |
2203 | { |
2204 | u64 range_start; | |
2205 | u64 range_end; | |
2206 | int key_type = BTRFS_DIR_LOG_ITEM_KEY; | |
2207 | int ret = 0; | |
2208 | struct btrfs_key dir_key; | |
2209 | struct btrfs_key found_key; | |
2210 | struct btrfs_path *log_path; | |
2211 | struct inode *dir; | |
2212 | ||
2213 | dir_key.objectid = dirid; | |
2214 | dir_key.type = BTRFS_DIR_ITEM_KEY; | |
2215 | log_path = btrfs_alloc_path(); | |
2216 | if (!log_path) | |
2217 | return -ENOMEM; | |
2218 | ||
2219 | dir = read_one_inode(root, dirid); | |
2220 | /* it isn't an error if the inode isn't there, that can happen | |
2221 | * because we replay the deletes before we copy in the inode item | |
2222 | * from the log | |
2223 | */ | |
2224 | if (!dir) { | |
2225 | btrfs_free_path(log_path); | |
2226 | return 0; | |
2227 | } | |
2228 | again: | |
2229 | range_start = 0; | |
2230 | range_end = 0; | |
d397712b | 2231 | while (1) { |
12fcfd22 CM |
2232 | if (del_all) |
2233 | range_end = (u64)-1; | |
2234 | else { | |
2235 | ret = find_dir_range(log, path, dirid, key_type, | |
2236 | &range_start, &range_end); | |
2237 | if (ret != 0) | |
2238 | break; | |
2239 | } | |
e02119d5 CM |
2240 | |
2241 | dir_key.offset = range_start; | |
d397712b | 2242 | while (1) { |
e02119d5 CM |
2243 | int nritems; |
2244 | ret = btrfs_search_slot(NULL, root, &dir_key, path, | |
2245 | 0, 0); | |
2246 | if (ret < 0) | |
2247 | goto out; | |
2248 | ||
2249 | nritems = btrfs_header_nritems(path->nodes[0]); | |
2250 | if (path->slots[0] >= nritems) { | |
2251 | ret = btrfs_next_leaf(root, path); | |
2252 | if (ret) | |
2253 | break; | |
2254 | } | |
2255 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, | |
2256 | path->slots[0]); | |
2257 | if (found_key.objectid != dirid || | |
2258 | found_key.type != dir_key.type) | |
2259 | goto next_type; | |
2260 | ||
2261 | if (found_key.offset > range_end) | |
2262 | break; | |
2263 | ||
2264 | ret = check_item_in_log(trans, root, log, path, | |
12fcfd22 CM |
2265 | log_path, dir, |
2266 | &found_key); | |
3650860b JB |
2267 | if (ret) |
2268 | goto out; | |
e02119d5 CM |
2269 | if (found_key.offset == (u64)-1) |
2270 | break; | |
2271 | dir_key.offset = found_key.offset + 1; | |
2272 | } | |
b3b4aa74 | 2273 | btrfs_release_path(path); |
e02119d5 CM |
2274 | if (range_end == (u64)-1) |
2275 | break; | |
2276 | range_start = range_end + 1; | |
2277 | } | |
2278 | ||
2279 | next_type: | |
2280 | ret = 0; | |
2281 | if (key_type == BTRFS_DIR_LOG_ITEM_KEY) { | |
2282 | key_type = BTRFS_DIR_LOG_INDEX_KEY; | |
2283 | dir_key.type = BTRFS_DIR_INDEX_KEY; | |
b3b4aa74 | 2284 | btrfs_release_path(path); |
e02119d5 CM |
2285 | goto again; |
2286 | } | |
2287 | out: | |
b3b4aa74 | 2288 | btrfs_release_path(path); |
e02119d5 CM |
2289 | btrfs_free_path(log_path); |
2290 | iput(dir); | |
2291 | return ret; | |
2292 | } | |
2293 | ||
2294 | /* | |
2295 | * the process_func used to replay items from the log tree. This | |
2296 | * gets called in two different stages. The first stage just looks | |
2297 | * for inodes and makes sure they are all copied into the subvolume. | |
2298 | * | |
2299 | * The second stage copies all the other item types from the log into | |
2300 | * the subvolume. The two stage approach is slower, but gets rid of | |
2301 | * lots of complexity around inodes referencing other inodes that exist | |
2302 | * only in the log (references come from either directory items or inode | |
2303 | * back refs). | |
2304 | */ | |
2305 | static int replay_one_buffer(struct btrfs_root *log, struct extent_buffer *eb, | |
2306 | struct walk_control *wc, u64 gen) | |
2307 | { | |
2308 | int nritems; | |
2309 | struct btrfs_path *path; | |
2310 | struct btrfs_root *root = wc->replay_dest; | |
2311 | struct btrfs_key key; | |
e02119d5 CM |
2312 | int level; |
2313 | int i; | |
2314 | int ret; | |
2315 | ||
018642a1 TI |
2316 | ret = btrfs_read_buffer(eb, gen); |
2317 | if (ret) | |
2318 | return ret; | |
e02119d5 CM |
2319 | |
2320 | level = btrfs_header_level(eb); | |
2321 | ||
2322 | if (level != 0) | |
2323 | return 0; | |
2324 | ||
2325 | path = btrfs_alloc_path(); | |
1e5063d0 MF |
2326 | if (!path) |
2327 | return -ENOMEM; | |
e02119d5 CM |
2328 | |
2329 | nritems = btrfs_header_nritems(eb); | |
2330 | for (i = 0; i < nritems; i++) { | |
2331 | btrfs_item_key_to_cpu(eb, &key, i); | |
e02119d5 CM |
2332 | |
2333 | /* inode keys are done during the first stage */ | |
2334 | if (key.type == BTRFS_INODE_ITEM_KEY && | |
2335 | wc->stage == LOG_WALK_REPLAY_INODES) { | |
e02119d5 CM |
2336 | struct btrfs_inode_item *inode_item; |
2337 | u32 mode; | |
2338 | ||
2339 | inode_item = btrfs_item_ptr(eb, i, | |
2340 | struct btrfs_inode_item); | |
4f764e51 FM |
2341 | ret = replay_xattr_deletes(wc->trans, root, log, |
2342 | path, key.objectid); | |
2343 | if (ret) | |
2344 | break; | |
e02119d5 CM |
2345 | mode = btrfs_inode_mode(eb, inode_item); |
2346 | if (S_ISDIR(mode)) { | |
2347 | ret = replay_dir_deletes(wc->trans, | |
12fcfd22 | 2348 | root, log, path, key.objectid, 0); |
b50c6e25 JB |
2349 | if (ret) |
2350 | break; | |
e02119d5 CM |
2351 | } |
2352 | ret = overwrite_item(wc->trans, root, path, | |
2353 | eb, i, &key); | |
b50c6e25 JB |
2354 | if (ret) |
2355 | break; | |
e02119d5 | 2356 | |
c71bf099 | 2357 | /* for regular files, make sure corresponding |
01327610 | 2358 | * orphan item exist. extents past the new EOF |
c71bf099 | 2359 | * will be truncated later by orphan cleanup. |
e02119d5 CM |
2360 | */ |
2361 | if (S_ISREG(mode)) { | |
c71bf099 YZ |
2362 | ret = insert_orphan_item(wc->trans, root, |
2363 | key.objectid); | |
b50c6e25 JB |
2364 | if (ret) |
2365 | break; | |
e02119d5 | 2366 | } |
c71bf099 | 2367 | |
e02119d5 CM |
2368 | ret = link_to_fixup_dir(wc->trans, root, |
2369 | path, key.objectid); | |
b50c6e25 JB |
2370 | if (ret) |
2371 | break; | |
e02119d5 | 2372 | } |
dd8e7217 JB |
2373 | |
2374 | if (key.type == BTRFS_DIR_INDEX_KEY && | |
2375 | wc->stage == LOG_WALK_REPLAY_DIR_INDEX) { | |
2376 | ret = replay_one_dir_item(wc->trans, root, path, | |
2377 | eb, i, &key); | |
2378 | if (ret) | |
2379 | break; | |
2380 | } | |
2381 | ||
e02119d5 CM |
2382 | if (wc->stage < LOG_WALK_REPLAY_ALL) |
2383 | continue; | |
2384 | ||
2385 | /* these keys are simply copied */ | |
2386 | if (key.type == BTRFS_XATTR_ITEM_KEY) { | |
2387 | ret = overwrite_item(wc->trans, root, path, | |
2388 | eb, i, &key); | |
b50c6e25 JB |
2389 | if (ret) |
2390 | break; | |
2da1c669 LB |
2391 | } else if (key.type == BTRFS_INODE_REF_KEY || |
2392 | key.type == BTRFS_INODE_EXTREF_KEY) { | |
f186373f MF |
2393 | ret = add_inode_ref(wc->trans, root, log, path, |
2394 | eb, i, &key); | |
b50c6e25 JB |
2395 | if (ret && ret != -ENOENT) |
2396 | break; | |
2397 | ret = 0; | |
e02119d5 CM |
2398 | } else if (key.type == BTRFS_EXTENT_DATA_KEY) { |
2399 | ret = replay_one_extent(wc->trans, root, path, | |
2400 | eb, i, &key); | |
b50c6e25 JB |
2401 | if (ret) |
2402 | break; | |
dd8e7217 | 2403 | } else if (key.type == BTRFS_DIR_ITEM_KEY) { |
e02119d5 CM |
2404 | ret = replay_one_dir_item(wc->trans, root, path, |
2405 | eb, i, &key); | |
b50c6e25 JB |
2406 | if (ret) |
2407 | break; | |
e02119d5 CM |
2408 | } |
2409 | } | |
2410 | btrfs_free_path(path); | |
b50c6e25 | 2411 | return ret; |
e02119d5 CM |
2412 | } |
2413 | ||
d397712b | 2414 | static noinline int walk_down_log_tree(struct btrfs_trans_handle *trans, |
e02119d5 CM |
2415 | struct btrfs_root *root, |
2416 | struct btrfs_path *path, int *level, | |
2417 | struct walk_control *wc) | |
2418 | { | |
0b246afa | 2419 | struct btrfs_fs_info *fs_info = root->fs_info; |
e02119d5 | 2420 | u64 root_owner; |
e02119d5 CM |
2421 | u64 bytenr; |
2422 | u64 ptr_gen; | |
2423 | struct extent_buffer *next; | |
2424 | struct extent_buffer *cur; | |
2425 | struct extent_buffer *parent; | |
2426 | u32 blocksize; | |
2427 | int ret = 0; | |
2428 | ||
2429 | WARN_ON(*level < 0); | |
2430 | WARN_ON(*level >= BTRFS_MAX_LEVEL); | |
2431 | ||
d397712b | 2432 | while (*level > 0) { |
e02119d5 CM |
2433 | WARN_ON(*level < 0); |
2434 | WARN_ON(*level >= BTRFS_MAX_LEVEL); | |
2435 | cur = path->nodes[*level]; | |
2436 | ||
fae7f21c | 2437 | WARN_ON(btrfs_header_level(cur) != *level); |
e02119d5 CM |
2438 | |
2439 | if (path->slots[*level] >= | |
2440 | btrfs_header_nritems(cur)) | |
2441 | break; | |
2442 | ||
2443 | bytenr = btrfs_node_blockptr(cur, path->slots[*level]); | |
2444 | ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]); | |
0b246afa | 2445 | blocksize = fs_info->nodesize; |
e02119d5 CM |
2446 | |
2447 | parent = path->nodes[*level]; | |
2448 | root_owner = btrfs_header_owner(parent); | |
e02119d5 | 2449 | |
2ff7e61e | 2450 | next = btrfs_find_create_tree_block(fs_info, bytenr); |
c871b0f2 LB |
2451 | if (IS_ERR(next)) |
2452 | return PTR_ERR(next); | |
e02119d5 | 2453 | |
e02119d5 | 2454 | if (*level == 1) { |
1e5063d0 | 2455 | ret = wc->process_func(root, next, wc, ptr_gen); |
b50c6e25 JB |
2456 | if (ret) { |
2457 | free_extent_buffer(next); | |
1e5063d0 | 2458 | return ret; |
b50c6e25 | 2459 | } |
4a500fd1 | 2460 | |
e02119d5 CM |
2461 | path->slots[*level]++; |
2462 | if (wc->free) { | |
018642a1 TI |
2463 | ret = btrfs_read_buffer(next, ptr_gen); |
2464 | if (ret) { | |
2465 | free_extent_buffer(next); | |
2466 | return ret; | |
2467 | } | |
e02119d5 | 2468 | |
681ae509 JB |
2469 | if (trans) { |
2470 | btrfs_tree_lock(next); | |
2471 | btrfs_set_lock_blocking(next); | |
0b246afa | 2472 | clean_tree_block(trans, fs_info, next); |
681ae509 JB |
2473 | btrfs_wait_tree_block_writeback(next); |
2474 | btrfs_tree_unlock(next); | |
2475 | } | |
e02119d5 | 2476 | |
e02119d5 CM |
2477 | WARN_ON(root_owner != |
2478 | BTRFS_TREE_LOG_OBJECTID); | |
2ff7e61e JM |
2479 | ret = btrfs_free_and_pin_reserved_extent( |
2480 | fs_info, bytenr, | |
2481 | blocksize); | |
3650860b JB |
2482 | if (ret) { |
2483 | free_extent_buffer(next); | |
2484 | return ret; | |
2485 | } | |
e02119d5 CM |
2486 | } |
2487 | free_extent_buffer(next); | |
2488 | continue; | |
2489 | } | |
018642a1 TI |
2490 | ret = btrfs_read_buffer(next, ptr_gen); |
2491 | if (ret) { | |
2492 | free_extent_buffer(next); | |
2493 | return ret; | |
2494 | } | |
e02119d5 CM |
2495 | |
2496 | WARN_ON(*level <= 0); | |
2497 | if (path->nodes[*level-1]) | |
2498 | free_extent_buffer(path->nodes[*level-1]); | |
2499 | path->nodes[*level-1] = next; | |
2500 | *level = btrfs_header_level(next); | |
2501 | path->slots[*level] = 0; | |
2502 | cond_resched(); | |
2503 | } | |
2504 | WARN_ON(*level < 0); | |
2505 | WARN_ON(*level >= BTRFS_MAX_LEVEL); | |
2506 | ||
4a500fd1 | 2507 | path->slots[*level] = btrfs_header_nritems(path->nodes[*level]); |
e02119d5 CM |
2508 | |
2509 | cond_resched(); | |
2510 | return 0; | |
2511 | } | |
2512 | ||
d397712b | 2513 | static noinline int walk_up_log_tree(struct btrfs_trans_handle *trans, |
e02119d5 CM |
2514 | struct btrfs_root *root, |
2515 | struct btrfs_path *path, int *level, | |
2516 | struct walk_control *wc) | |
2517 | { | |
0b246afa | 2518 | struct btrfs_fs_info *fs_info = root->fs_info; |
e02119d5 | 2519 | u64 root_owner; |
e02119d5 CM |
2520 | int i; |
2521 | int slot; | |
2522 | int ret; | |
2523 | ||
d397712b | 2524 | for (i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) { |
e02119d5 | 2525 | slot = path->slots[i]; |
4a500fd1 | 2526 | if (slot + 1 < btrfs_header_nritems(path->nodes[i])) { |
e02119d5 CM |
2527 | path->slots[i]++; |
2528 | *level = i; | |
2529 | WARN_ON(*level == 0); | |
2530 | return 0; | |
2531 | } else { | |
31840ae1 ZY |
2532 | struct extent_buffer *parent; |
2533 | if (path->nodes[*level] == root->node) | |
2534 | parent = path->nodes[*level]; | |
2535 | else | |
2536 | parent = path->nodes[*level + 1]; | |
2537 | ||
2538 | root_owner = btrfs_header_owner(parent); | |
1e5063d0 | 2539 | ret = wc->process_func(root, path->nodes[*level], wc, |
e02119d5 | 2540 | btrfs_header_generation(path->nodes[*level])); |
1e5063d0 MF |
2541 | if (ret) |
2542 | return ret; | |
2543 | ||
e02119d5 CM |
2544 | if (wc->free) { |
2545 | struct extent_buffer *next; | |
2546 | ||
2547 | next = path->nodes[*level]; | |
2548 | ||
681ae509 JB |
2549 | if (trans) { |
2550 | btrfs_tree_lock(next); | |
2551 | btrfs_set_lock_blocking(next); | |
0b246afa | 2552 | clean_tree_block(trans, fs_info, next); |
681ae509 JB |
2553 | btrfs_wait_tree_block_writeback(next); |
2554 | btrfs_tree_unlock(next); | |
2555 | } | |
e02119d5 | 2556 | |
e02119d5 | 2557 | WARN_ON(root_owner != BTRFS_TREE_LOG_OBJECTID); |
2ff7e61e JM |
2558 | ret = btrfs_free_and_pin_reserved_extent( |
2559 | fs_info, | |
e02119d5 | 2560 | path->nodes[*level]->start, |
d00aff00 | 2561 | path->nodes[*level]->len); |
3650860b JB |
2562 | if (ret) |
2563 | return ret; | |
e02119d5 CM |
2564 | } |
2565 | free_extent_buffer(path->nodes[*level]); | |
2566 | path->nodes[*level] = NULL; | |
2567 | *level = i + 1; | |
2568 | } | |
2569 | } | |
2570 | return 1; | |
2571 | } | |
2572 | ||
2573 | /* | |
2574 | * drop the reference count on the tree rooted at 'snap'. This traverses | |
2575 | * the tree freeing any blocks that have a ref count of zero after being | |
2576 | * decremented. | |
2577 | */ | |
2578 | static int walk_log_tree(struct btrfs_trans_handle *trans, | |
2579 | struct btrfs_root *log, struct walk_control *wc) | |
2580 | { | |
2ff7e61e | 2581 | struct btrfs_fs_info *fs_info = log->fs_info; |
e02119d5 CM |
2582 | int ret = 0; |
2583 | int wret; | |
2584 | int level; | |
2585 | struct btrfs_path *path; | |
e02119d5 CM |
2586 | int orig_level; |
2587 | ||
2588 | path = btrfs_alloc_path(); | |
db5b493a TI |
2589 | if (!path) |
2590 | return -ENOMEM; | |
e02119d5 CM |
2591 | |
2592 | level = btrfs_header_level(log->node); | |
2593 | orig_level = level; | |
2594 | path->nodes[level] = log->node; | |
2595 | extent_buffer_get(log->node); | |
2596 | path->slots[level] = 0; | |
2597 | ||
d397712b | 2598 | while (1) { |
e02119d5 CM |
2599 | wret = walk_down_log_tree(trans, log, path, &level, wc); |
2600 | if (wret > 0) | |
2601 | break; | |
79787eaa | 2602 | if (wret < 0) { |
e02119d5 | 2603 | ret = wret; |
79787eaa JM |
2604 | goto out; |
2605 | } | |
e02119d5 CM |
2606 | |
2607 | wret = walk_up_log_tree(trans, log, path, &level, wc); | |
2608 | if (wret > 0) | |
2609 | break; | |
79787eaa | 2610 | if (wret < 0) { |
e02119d5 | 2611 | ret = wret; |
79787eaa JM |
2612 | goto out; |
2613 | } | |
e02119d5 CM |
2614 | } |
2615 | ||
2616 | /* was the root node processed? if not, catch it here */ | |
2617 | if (path->nodes[orig_level]) { | |
79787eaa | 2618 | ret = wc->process_func(log, path->nodes[orig_level], wc, |
e02119d5 | 2619 | btrfs_header_generation(path->nodes[orig_level])); |
79787eaa JM |
2620 | if (ret) |
2621 | goto out; | |
e02119d5 CM |
2622 | if (wc->free) { |
2623 | struct extent_buffer *next; | |
2624 | ||
2625 | next = path->nodes[orig_level]; | |
2626 | ||
681ae509 JB |
2627 | if (trans) { |
2628 | btrfs_tree_lock(next); | |
2629 | btrfs_set_lock_blocking(next); | |
2ff7e61e | 2630 | clean_tree_block(trans, fs_info, next); |
681ae509 JB |
2631 | btrfs_wait_tree_block_writeback(next); |
2632 | btrfs_tree_unlock(next); | |
2633 | } | |
e02119d5 | 2634 | |
e02119d5 CM |
2635 | WARN_ON(log->root_key.objectid != |
2636 | BTRFS_TREE_LOG_OBJECTID); | |
2ff7e61e JM |
2637 | ret = btrfs_free_and_pin_reserved_extent(fs_info, |
2638 | next->start, next->len); | |
3650860b JB |
2639 | if (ret) |
2640 | goto out; | |
e02119d5 CM |
2641 | } |
2642 | } | |
2643 | ||
79787eaa | 2644 | out: |
e02119d5 | 2645 | btrfs_free_path(path); |
e02119d5 CM |
2646 | return ret; |
2647 | } | |
2648 | ||
7237f183 YZ |
2649 | /* |
2650 | * helper function to update the item for a given subvolumes log root | |
2651 | * in the tree of log roots | |
2652 | */ | |
2653 | static int update_log_root(struct btrfs_trans_handle *trans, | |
2654 | struct btrfs_root *log) | |
2655 | { | |
0b246afa | 2656 | struct btrfs_fs_info *fs_info = log->fs_info; |
7237f183 YZ |
2657 | int ret; |
2658 | ||
2659 | if (log->log_transid == 1) { | |
2660 | /* insert root item on the first sync */ | |
0b246afa | 2661 | ret = btrfs_insert_root(trans, fs_info->log_root_tree, |
7237f183 YZ |
2662 | &log->root_key, &log->root_item); |
2663 | } else { | |
0b246afa | 2664 | ret = btrfs_update_root(trans, fs_info->log_root_tree, |
7237f183 YZ |
2665 | &log->root_key, &log->root_item); |
2666 | } | |
2667 | return ret; | |
2668 | } | |
2669 | ||
60d53eb3 | 2670 | static void wait_log_commit(struct btrfs_root *root, int transid) |
e02119d5 CM |
2671 | { |
2672 | DEFINE_WAIT(wait); | |
7237f183 | 2673 | int index = transid % 2; |
e02119d5 | 2674 | |
7237f183 YZ |
2675 | /* |
2676 | * we only allow two pending log transactions at a time, | |
2677 | * so we know that if ours is more than 2 older than the | |
2678 | * current transaction, we're done | |
2679 | */ | |
e02119d5 | 2680 | do { |
7237f183 YZ |
2681 | prepare_to_wait(&root->log_commit_wait[index], |
2682 | &wait, TASK_UNINTERRUPTIBLE); | |
2683 | mutex_unlock(&root->log_mutex); | |
12fcfd22 | 2684 | |
d1433deb | 2685 | if (root->log_transid_committed < transid && |
7237f183 YZ |
2686 | atomic_read(&root->log_commit[index])) |
2687 | schedule(); | |
12fcfd22 | 2688 | |
7237f183 YZ |
2689 | finish_wait(&root->log_commit_wait[index], &wait); |
2690 | mutex_lock(&root->log_mutex); | |
d1433deb | 2691 | } while (root->log_transid_committed < transid && |
7237f183 | 2692 | atomic_read(&root->log_commit[index])); |
7237f183 YZ |
2693 | } |
2694 | ||
60d53eb3 | 2695 | static void wait_for_writer(struct btrfs_root *root) |
7237f183 YZ |
2696 | { |
2697 | DEFINE_WAIT(wait); | |
8b050d35 MX |
2698 | |
2699 | while (atomic_read(&root->log_writers)) { | |
7237f183 YZ |
2700 | prepare_to_wait(&root->log_writer_wait, |
2701 | &wait, TASK_UNINTERRUPTIBLE); | |
2702 | mutex_unlock(&root->log_mutex); | |
8b050d35 | 2703 | if (atomic_read(&root->log_writers)) |
e02119d5 | 2704 | schedule(); |
7237f183 | 2705 | finish_wait(&root->log_writer_wait, &wait); |
575849ec | 2706 | mutex_lock(&root->log_mutex); |
7237f183 | 2707 | } |
e02119d5 CM |
2708 | } |
2709 | ||
8b050d35 MX |
2710 | static inline void btrfs_remove_log_ctx(struct btrfs_root *root, |
2711 | struct btrfs_log_ctx *ctx) | |
2712 | { | |
2713 | if (!ctx) | |
2714 | return; | |
2715 | ||
2716 | mutex_lock(&root->log_mutex); | |
2717 | list_del_init(&ctx->list); | |
2718 | mutex_unlock(&root->log_mutex); | |
2719 | } | |
2720 | ||
2721 | /* | |
2722 | * Invoked in log mutex context, or be sure there is no other task which | |
2723 | * can access the list. | |
2724 | */ | |
2725 | static inline void btrfs_remove_all_log_ctxs(struct btrfs_root *root, | |
2726 | int index, int error) | |
2727 | { | |
2728 | struct btrfs_log_ctx *ctx; | |
570dd450 | 2729 | struct btrfs_log_ctx *safe; |
8b050d35 | 2730 | |
570dd450 CM |
2731 | list_for_each_entry_safe(ctx, safe, &root->log_ctxs[index], list) { |
2732 | list_del_init(&ctx->list); | |
8b050d35 | 2733 | ctx->log_ret = error; |
570dd450 | 2734 | } |
8b050d35 MX |
2735 | |
2736 | INIT_LIST_HEAD(&root->log_ctxs[index]); | |
2737 | } | |
2738 | ||
e02119d5 CM |
2739 | /* |
2740 | * btrfs_sync_log does sends a given tree log down to the disk and | |
2741 | * updates the super blocks to record it. When this call is done, | |
12fcfd22 CM |
2742 | * you know that any inodes previously logged are safely on disk only |
2743 | * if it returns 0. | |
2744 | * | |
2745 | * Any other return value means you need to call btrfs_commit_transaction. | |
2746 | * Some of the edge cases for fsyncing directories that have had unlinks | |
2747 | * or renames done in the past mean that sometimes the only safe | |
2748 | * fsync is to commit the whole FS. When btrfs_sync_log returns -EAGAIN, | |
2749 | * that has happened. | |
e02119d5 CM |
2750 | */ |
2751 | int btrfs_sync_log(struct btrfs_trans_handle *trans, | |
8b050d35 | 2752 | struct btrfs_root *root, struct btrfs_log_ctx *ctx) |
e02119d5 | 2753 | { |
7237f183 YZ |
2754 | int index1; |
2755 | int index2; | |
8cef4e16 | 2756 | int mark; |
e02119d5 | 2757 | int ret; |
0b246afa | 2758 | struct btrfs_fs_info *fs_info = root->fs_info; |
e02119d5 | 2759 | struct btrfs_root *log = root->log_root; |
0b246afa | 2760 | struct btrfs_root *log_root_tree = fs_info->log_root_tree; |
bb14a59b | 2761 | int log_transid = 0; |
8b050d35 | 2762 | struct btrfs_log_ctx root_log_ctx; |
c6adc9cc | 2763 | struct blk_plug plug; |
e02119d5 | 2764 | |
7237f183 | 2765 | mutex_lock(&root->log_mutex); |
d1433deb MX |
2766 | log_transid = ctx->log_transid; |
2767 | if (root->log_transid_committed >= log_transid) { | |
2768 | mutex_unlock(&root->log_mutex); | |
2769 | return ctx->log_ret; | |
2770 | } | |
2771 | ||
2772 | index1 = log_transid % 2; | |
7237f183 | 2773 | if (atomic_read(&root->log_commit[index1])) { |
60d53eb3 | 2774 | wait_log_commit(root, log_transid); |
7237f183 | 2775 | mutex_unlock(&root->log_mutex); |
8b050d35 | 2776 | return ctx->log_ret; |
e02119d5 | 2777 | } |
d1433deb | 2778 | ASSERT(log_transid == root->log_transid); |
7237f183 YZ |
2779 | atomic_set(&root->log_commit[index1], 1); |
2780 | ||
2781 | /* wait for previous tree log sync to complete */ | |
2782 | if (atomic_read(&root->log_commit[(index1 + 1) % 2])) | |
60d53eb3 | 2783 | wait_log_commit(root, log_transid - 1); |
48cab2e0 | 2784 | |
86df7eb9 | 2785 | while (1) { |
2ecb7923 | 2786 | int batch = atomic_read(&root->log_batch); |
cd354ad6 | 2787 | /* when we're on an ssd, just kick the log commit out */ |
0b246afa | 2788 | if (!btrfs_test_opt(fs_info, SSD) && |
27cdeb70 | 2789 | test_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state)) { |
86df7eb9 YZ |
2790 | mutex_unlock(&root->log_mutex); |
2791 | schedule_timeout_uninterruptible(1); | |
2792 | mutex_lock(&root->log_mutex); | |
2793 | } | |
60d53eb3 | 2794 | wait_for_writer(root); |
2ecb7923 | 2795 | if (batch == atomic_read(&root->log_batch)) |
e02119d5 CM |
2796 | break; |
2797 | } | |
e02119d5 | 2798 | |
12fcfd22 | 2799 | /* bail out if we need to do a full commit */ |
0b246afa | 2800 | if (btrfs_need_log_full_commit(fs_info, trans)) { |
12fcfd22 | 2801 | ret = -EAGAIN; |
2ab28f32 | 2802 | btrfs_free_logged_extents(log, log_transid); |
12fcfd22 CM |
2803 | mutex_unlock(&root->log_mutex); |
2804 | goto out; | |
2805 | } | |
2806 | ||
8cef4e16 YZ |
2807 | if (log_transid % 2 == 0) |
2808 | mark = EXTENT_DIRTY; | |
2809 | else | |
2810 | mark = EXTENT_NEW; | |
2811 | ||
690587d1 CM |
2812 | /* we start IO on all the marked extents here, but we don't actually |
2813 | * wait for them until later. | |
2814 | */ | |
c6adc9cc | 2815 | blk_start_plug(&plug); |
2ff7e61e | 2816 | ret = btrfs_write_marked_extents(fs_info, &log->dirty_log_pages, mark); |
79787eaa | 2817 | if (ret) { |
c6adc9cc | 2818 | blk_finish_plug(&plug); |
66642832 | 2819 | btrfs_abort_transaction(trans, ret); |
2ab28f32 | 2820 | btrfs_free_logged_extents(log, log_transid); |
0b246afa | 2821 | btrfs_set_log_full_commit(fs_info, trans); |
79787eaa JM |
2822 | mutex_unlock(&root->log_mutex); |
2823 | goto out; | |
2824 | } | |
7237f183 | 2825 | |
5d4f98a2 | 2826 | btrfs_set_root_node(&log->root_item, log->node); |
7237f183 | 2827 | |
7237f183 YZ |
2828 | root->log_transid++; |
2829 | log->log_transid = root->log_transid; | |
ff782e0a | 2830 | root->log_start_pid = 0; |
7237f183 | 2831 | /* |
8cef4e16 YZ |
2832 | * IO has been started, blocks of the log tree have WRITTEN flag set |
2833 | * in their headers. new modifications of the log will be written to | |
2834 | * new positions. so it's safe to allow log writers to go in. | |
7237f183 YZ |
2835 | */ |
2836 | mutex_unlock(&root->log_mutex); | |
2837 | ||
28a23593 | 2838 | btrfs_init_log_ctx(&root_log_ctx, NULL); |
d1433deb | 2839 | |
7237f183 | 2840 | mutex_lock(&log_root_tree->log_mutex); |
2ecb7923 | 2841 | atomic_inc(&log_root_tree->log_batch); |
7237f183 | 2842 | atomic_inc(&log_root_tree->log_writers); |
d1433deb MX |
2843 | |
2844 | index2 = log_root_tree->log_transid % 2; | |
2845 | list_add_tail(&root_log_ctx.list, &log_root_tree->log_ctxs[index2]); | |
2846 | root_log_ctx.log_transid = log_root_tree->log_transid; | |
2847 | ||
7237f183 YZ |
2848 | mutex_unlock(&log_root_tree->log_mutex); |
2849 | ||
2850 | ret = update_log_root(trans, log); | |
7237f183 YZ |
2851 | |
2852 | mutex_lock(&log_root_tree->log_mutex); | |
2853 | if (atomic_dec_and_test(&log_root_tree->log_writers)) { | |
779adf0f DS |
2854 | /* |
2855 | * Implicit memory barrier after atomic_dec_and_test | |
2856 | */ | |
7237f183 YZ |
2857 | if (waitqueue_active(&log_root_tree->log_writer_wait)) |
2858 | wake_up(&log_root_tree->log_writer_wait); | |
2859 | } | |
2860 | ||
4a500fd1 | 2861 | if (ret) { |
d1433deb MX |
2862 | if (!list_empty(&root_log_ctx.list)) |
2863 | list_del_init(&root_log_ctx.list); | |
2864 | ||
c6adc9cc | 2865 | blk_finish_plug(&plug); |
0b246afa | 2866 | btrfs_set_log_full_commit(fs_info, trans); |
995946dd | 2867 | |
79787eaa | 2868 | if (ret != -ENOSPC) { |
66642832 | 2869 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
2870 | mutex_unlock(&log_root_tree->log_mutex); |
2871 | goto out; | |
2872 | } | |
bf89d38f | 2873 | btrfs_wait_tree_log_extents(log, mark); |
2ab28f32 | 2874 | btrfs_free_logged_extents(log, log_transid); |
4a500fd1 YZ |
2875 | mutex_unlock(&log_root_tree->log_mutex); |
2876 | ret = -EAGAIN; | |
2877 | goto out; | |
2878 | } | |
2879 | ||
d1433deb | 2880 | if (log_root_tree->log_transid_committed >= root_log_ctx.log_transid) { |
3da5ab56 | 2881 | blk_finish_plug(&plug); |
cbd60aa7 | 2882 | list_del_init(&root_log_ctx.list); |
d1433deb MX |
2883 | mutex_unlock(&log_root_tree->log_mutex); |
2884 | ret = root_log_ctx.log_ret; | |
2885 | goto out; | |
2886 | } | |
8b050d35 | 2887 | |
d1433deb | 2888 | index2 = root_log_ctx.log_transid % 2; |
7237f183 | 2889 | if (atomic_read(&log_root_tree->log_commit[index2])) { |
c6adc9cc | 2890 | blk_finish_plug(&plug); |
bf89d38f | 2891 | ret = btrfs_wait_tree_log_extents(log, mark); |
50d9aa99 | 2892 | btrfs_wait_logged_extents(trans, log, log_transid); |
60d53eb3 | 2893 | wait_log_commit(log_root_tree, |
d1433deb | 2894 | root_log_ctx.log_transid); |
7237f183 | 2895 | mutex_unlock(&log_root_tree->log_mutex); |
5ab5e44a FM |
2896 | if (!ret) |
2897 | ret = root_log_ctx.log_ret; | |
7237f183 YZ |
2898 | goto out; |
2899 | } | |
d1433deb | 2900 | ASSERT(root_log_ctx.log_transid == log_root_tree->log_transid); |
7237f183 YZ |
2901 | atomic_set(&log_root_tree->log_commit[index2], 1); |
2902 | ||
12fcfd22 | 2903 | if (atomic_read(&log_root_tree->log_commit[(index2 + 1) % 2])) { |
60d53eb3 | 2904 | wait_log_commit(log_root_tree, |
d1433deb | 2905 | root_log_ctx.log_transid - 1); |
12fcfd22 CM |
2906 | } |
2907 | ||
60d53eb3 | 2908 | wait_for_writer(log_root_tree); |
7237f183 | 2909 | |
12fcfd22 CM |
2910 | /* |
2911 | * now that we've moved on to the tree of log tree roots, | |
2912 | * check the full commit flag again | |
2913 | */ | |
0b246afa | 2914 | if (btrfs_need_log_full_commit(fs_info, trans)) { |
c6adc9cc | 2915 | blk_finish_plug(&plug); |
bf89d38f | 2916 | btrfs_wait_tree_log_extents(log, mark); |
2ab28f32 | 2917 | btrfs_free_logged_extents(log, log_transid); |
12fcfd22 CM |
2918 | mutex_unlock(&log_root_tree->log_mutex); |
2919 | ret = -EAGAIN; | |
2920 | goto out_wake_log_root; | |
2921 | } | |
7237f183 | 2922 | |
2ff7e61e | 2923 | ret = btrfs_write_marked_extents(fs_info, |
c6adc9cc MX |
2924 | &log_root_tree->dirty_log_pages, |
2925 | EXTENT_DIRTY | EXTENT_NEW); | |
2926 | blk_finish_plug(&plug); | |
79787eaa | 2927 | if (ret) { |
0b246afa | 2928 | btrfs_set_log_full_commit(fs_info, trans); |
66642832 | 2929 | btrfs_abort_transaction(trans, ret); |
2ab28f32 | 2930 | btrfs_free_logged_extents(log, log_transid); |
79787eaa JM |
2931 | mutex_unlock(&log_root_tree->log_mutex); |
2932 | goto out_wake_log_root; | |
2933 | } | |
bf89d38f | 2934 | ret = btrfs_wait_tree_log_extents(log, mark); |
5ab5e44a | 2935 | if (!ret) |
bf89d38f JM |
2936 | ret = btrfs_wait_tree_log_extents(log_root_tree, |
2937 | EXTENT_NEW | EXTENT_DIRTY); | |
5ab5e44a | 2938 | if (ret) { |
0b246afa | 2939 | btrfs_set_log_full_commit(fs_info, trans); |
5ab5e44a FM |
2940 | btrfs_free_logged_extents(log, log_transid); |
2941 | mutex_unlock(&log_root_tree->log_mutex); | |
2942 | goto out_wake_log_root; | |
2943 | } | |
50d9aa99 | 2944 | btrfs_wait_logged_extents(trans, log, log_transid); |
e02119d5 | 2945 | |
0b246afa JM |
2946 | btrfs_set_super_log_root(fs_info->super_for_commit, |
2947 | log_root_tree->node->start); | |
2948 | btrfs_set_super_log_root_level(fs_info->super_for_commit, | |
2949 | btrfs_header_level(log_root_tree->node)); | |
e02119d5 | 2950 | |
7237f183 | 2951 | log_root_tree->log_transid++; |
7237f183 YZ |
2952 | mutex_unlock(&log_root_tree->log_mutex); |
2953 | ||
2954 | /* | |
2955 | * nobody else is going to jump in and write the the ctree | |
2956 | * super here because the log_commit atomic below is protecting | |
2957 | * us. We must be called with a transaction handle pinning | |
2958 | * the running transaction open, so a full commit can't hop | |
2959 | * in and cause problems either. | |
2960 | */ | |
2ff7e61e | 2961 | ret = write_ctree_super(trans, fs_info, 1); |
5af3e8cc | 2962 | if (ret) { |
0b246afa | 2963 | btrfs_set_log_full_commit(fs_info, trans); |
66642832 | 2964 | btrfs_abort_transaction(trans, ret); |
5af3e8cc SB |
2965 | goto out_wake_log_root; |
2966 | } | |
7237f183 | 2967 | |
257c62e1 CM |
2968 | mutex_lock(&root->log_mutex); |
2969 | if (root->last_log_commit < log_transid) | |
2970 | root->last_log_commit = log_transid; | |
2971 | mutex_unlock(&root->log_mutex); | |
2972 | ||
12fcfd22 | 2973 | out_wake_log_root: |
570dd450 | 2974 | mutex_lock(&log_root_tree->log_mutex); |
8b050d35 MX |
2975 | btrfs_remove_all_log_ctxs(log_root_tree, index2, ret); |
2976 | ||
d1433deb | 2977 | log_root_tree->log_transid_committed++; |
7237f183 | 2978 | atomic_set(&log_root_tree->log_commit[index2], 0); |
d1433deb MX |
2979 | mutex_unlock(&log_root_tree->log_mutex); |
2980 | ||
33a9eca7 DS |
2981 | /* |
2982 | * The barrier before waitqueue_active is implied by mutex_unlock | |
2983 | */ | |
7237f183 YZ |
2984 | if (waitqueue_active(&log_root_tree->log_commit_wait[index2])) |
2985 | wake_up(&log_root_tree->log_commit_wait[index2]); | |
e02119d5 | 2986 | out: |
d1433deb | 2987 | mutex_lock(&root->log_mutex); |
570dd450 | 2988 | btrfs_remove_all_log_ctxs(root, index1, ret); |
d1433deb | 2989 | root->log_transid_committed++; |
7237f183 | 2990 | atomic_set(&root->log_commit[index1], 0); |
d1433deb | 2991 | mutex_unlock(&root->log_mutex); |
8b050d35 | 2992 | |
33a9eca7 DS |
2993 | /* |
2994 | * The barrier before waitqueue_active is implied by mutex_unlock | |
2995 | */ | |
7237f183 YZ |
2996 | if (waitqueue_active(&root->log_commit_wait[index1])) |
2997 | wake_up(&root->log_commit_wait[index1]); | |
b31eabd8 | 2998 | return ret; |
e02119d5 CM |
2999 | } |
3000 | ||
4a500fd1 YZ |
3001 | static void free_log_tree(struct btrfs_trans_handle *trans, |
3002 | struct btrfs_root *log) | |
e02119d5 CM |
3003 | { |
3004 | int ret; | |
d0c803c4 CM |
3005 | u64 start; |
3006 | u64 end; | |
e02119d5 CM |
3007 | struct walk_control wc = { |
3008 | .free = 1, | |
3009 | .process_func = process_one_buffer | |
3010 | }; | |
3011 | ||
681ae509 JB |
3012 | ret = walk_log_tree(trans, log, &wc); |
3013 | /* I don't think this can happen but just in case */ | |
3014 | if (ret) | |
66642832 | 3015 | btrfs_abort_transaction(trans, ret); |
e02119d5 | 3016 | |
d397712b | 3017 | while (1) { |
d0c803c4 | 3018 | ret = find_first_extent_bit(&log->dirty_log_pages, |
e6138876 JB |
3019 | 0, &start, &end, EXTENT_DIRTY | EXTENT_NEW, |
3020 | NULL); | |
d0c803c4 CM |
3021 | if (ret) |
3022 | break; | |
3023 | ||
8cef4e16 | 3024 | clear_extent_bits(&log->dirty_log_pages, start, end, |
91166212 | 3025 | EXTENT_DIRTY | EXTENT_NEW); |
d0c803c4 CM |
3026 | } |
3027 | ||
2ab28f32 JB |
3028 | /* |
3029 | * We may have short-circuited the log tree with the full commit logic | |
3030 | * and left ordered extents on our list, so clear these out to keep us | |
3031 | * from leaking inodes and memory. | |
3032 | */ | |
3033 | btrfs_free_logged_extents(log, 0); | |
3034 | btrfs_free_logged_extents(log, 1); | |
3035 | ||
7237f183 YZ |
3036 | free_extent_buffer(log->node); |
3037 | kfree(log); | |
4a500fd1 YZ |
3038 | } |
3039 | ||
3040 | /* | |
3041 | * free all the extents used by the tree log. This should be called | |
3042 | * at commit time of the full transaction | |
3043 | */ | |
3044 | int btrfs_free_log(struct btrfs_trans_handle *trans, struct btrfs_root *root) | |
3045 | { | |
3046 | if (root->log_root) { | |
3047 | free_log_tree(trans, root->log_root); | |
3048 | root->log_root = NULL; | |
3049 | } | |
3050 | return 0; | |
3051 | } | |
3052 | ||
3053 | int btrfs_free_log_root_tree(struct btrfs_trans_handle *trans, | |
3054 | struct btrfs_fs_info *fs_info) | |
3055 | { | |
3056 | if (fs_info->log_root_tree) { | |
3057 | free_log_tree(trans, fs_info->log_root_tree); | |
3058 | fs_info->log_root_tree = NULL; | |
3059 | } | |
e02119d5 CM |
3060 | return 0; |
3061 | } | |
3062 | ||
e02119d5 CM |
3063 | /* |
3064 | * If both a file and directory are logged, and unlinks or renames are | |
3065 | * mixed in, we have a few interesting corners: | |
3066 | * | |
3067 | * create file X in dir Y | |
3068 | * link file X to X.link in dir Y | |
3069 | * fsync file X | |
3070 | * unlink file X but leave X.link | |
3071 | * fsync dir Y | |
3072 | * | |
3073 | * After a crash we would expect only X.link to exist. But file X | |
3074 | * didn't get fsync'd again so the log has back refs for X and X.link. | |
3075 | * | |
3076 | * We solve this by removing directory entries and inode backrefs from the | |
3077 | * log when a file that was logged in the current transaction is | |
3078 | * unlinked. Any later fsync will include the updated log entries, and | |
3079 | * we'll be able to reconstruct the proper directory items from backrefs. | |
3080 | * | |
3081 | * This optimizations allows us to avoid relogging the entire inode | |
3082 | * or the entire directory. | |
3083 | */ | |
3084 | int btrfs_del_dir_entries_in_log(struct btrfs_trans_handle *trans, | |
3085 | struct btrfs_root *root, | |
3086 | const char *name, int name_len, | |
49f34d1f | 3087 | struct btrfs_inode *dir, u64 index) |
e02119d5 CM |
3088 | { |
3089 | struct btrfs_root *log; | |
3090 | struct btrfs_dir_item *di; | |
3091 | struct btrfs_path *path; | |
3092 | int ret; | |
4a500fd1 | 3093 | int err = 0; |
e02119d5 | 3094 | int bytes_del = 0; |
49f34d1f | 3095 | u64 dir_ino = btrfs_ino(dir); |
e02119d5 | 3096 | |
49f34d1f | 3097 | if (dir->logged_trans < trans->transid) |
3a5f1d45 CM |
3098 | return 0; |
3099 | ||
e02119d5 CM |
3100 | ret = join_running_log_trans(root); |
3101 | if (ret) | |
3102 | return 0; | |
3103 | ||
49f34d1f | 3104 | mutex_lock(&dir->log_mutex); |
e02119d5 CM |
3105 | |
3106 | log = root->log_root; | |
3107 | path = btrfs_alloc_path(); | |
a62f44a5 TI |
3108 | if (!path) { |
3109 | err = -ENOMEM; | |
3110 | goto out_unlock; | |
3111 | } | |
2a29edc6 | 3112 | |
33345d01 | 3113 | di = btrfs_lookup_dir_item(trans, log, path, dir_ino, |
e02119d5 | 3114 | name, name_len, -1); |
4a500fd1 YZ |
3115 | if (IS_ERR(di)) { |
3116 | err = PTR_ERR(di); | |
3117 | goto fail; | |
3118 | } | |
3119 | if (di) { | |
e02119d5 CM |
3120 | ret = btrfs_delete_one_dir_name(trans, log, path, di); |
3121 | bytes_del += name_len; | |
3650860b JB |
3122 | if (ret) { |
3123 | err = ret; | |
3124 | goto fail; | |
3125 | } | |
e02119d5 | 3126 | } |
b3b4aa74 | 3127 | btrfs_release_path(path); |
33345d01 | 3128 | di = btrfs_lookup_dir_index_item(trans, log, path, dir_ino, |
e02119d5 | 3129 | index, name, name_len, -1); |
4a500fd1 YZ |
3130 | if (IS_ERR(di)) { |
3131 | err = PTR_ERR(di); | |
3132 | goto fail; | |
3133 | } | |
3134 | if (di) { | |
e02119d5 CM |
3135 | ret = btrfs_delete_one_dir_name(trans, log, path, di); |
3136 | bytes_del += name_len; | |
3650860b JB |
3137 | if (ret) { |
3138 | err = ret; | |
3139 | goto fail; | |
3140 | } | |
e02119d5 CM |
3141 | } |
3142 | ||
3143 | /* update the directory size in the log to reflect the names | |
3144 | * we have removed | |
3145 | */ | |
3146 | if (bytes_del) { | |
3147 | struct btrfs_key key; | |
3148 | ||
33345d01 | 3149 | key.objectid = dir_ino; |
e02119d5 CM |
3150 | key.offset = 0; |
3151 | key.type = BTRFS_INODE_ITEM_KEY; | |
b3b4aa74 | 3152 | btrfs_release_path(path); |
e02119d5 CM |
3153 | |
3154 | ret = btrfs_search_slot(trans, log, &key, path, 0, 1); | |
4a500fd1 YZ |
3155 | if (ret < 0) { |
3156 | err = ret; | |
3157 | goto fail; | |
3158 | } | |
e02119d5 CM |
3159 | if (ret == 0) { |
3160 | struct btrfs_inode_item *item; | |
3161 | u64 i_size; | |
3162 | ||
3163 | item = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
3164 | struct btrfs_inode_item); | |
3165 | i_size = btrfs_inode_size(path->nodes[0], item); | |
3166 | if (i_size > bytes_del) | |
3167 | i_size -= bytes_del; | |
3168 | else | |
3169 | i_size = 0; | |
3170 | btrfs_set_inode_size(path->nodes[0], item, i_size); | |
3171 | btrfs_mark_buffer_dirty(path->nodes[0]); | |
3172 | } else | |
3173 | ret = 0; | |
b3b4aa74 | 3174 | btrfs_release_path(path); |
e02119d5 | 3175 | } |
4a500fd1 | 3176 | fail: |
e02119d5 | 3177 | btrfs_free_path(path); |
a62f44a5 | 3178 | out_unlock: |
49f34d1f | 3179 | mutex_unlock(&dir->log_mutex); |
4a500fd1 | 3180 | if (ret == -ENOSPC) { |
995946dd | 3181 | btrfs_set_log_full_commit(root->fs_info, trans); |
4a500fd1 | 3182 | ret = 0; |
79787eaa | 3183 | } else if (ret < 0) |
66642832 | 3184 | btrfs_abort_transaction(trans, ret); |
79787eaa | 3185 | |
12fcfd22 | 3186 | btrfs_end_log_trans(root); |
e02119d5 | 3187 | |
411fc6bc | 3188 | return err; |
e02119d5 CM |
3189 | } |
3190 | ||
3191 | /* see comments for btrfs_del_dir_entries_in_log */ | |
3192 | int btrfs_del_inode_ref_in_log(struct btrfs_trans_handle *trans, | |
3193 | struct btrfs_root *root, | |
3194 | const char *name, int name_len, | |
a491abb2 | 3195 | struct btrfs_inode *inode, u64 dirid) |
e02119d5 | 3196 | { |
0b246afa | 3197 | struct btrfs_fs_info *fs_info = root->fs_info; |
e02119d5 CM |
3198 | struct btrfs_root *log; |
3199 | u64 index; | |
3200 | int ret; | |
3201 | ||
a491abb2 | 3202 | if (inode->logged_trans < trans->transid) |
3a5f1d45 CM |
3203 | return 0; |
3204 | ||
e02119d5 CM |
3205 | ret = join_running_log_trans(root); |
3206 | if (ret) | |
3207 | return 0; | |
3208 | log = root->log_root; | |
a491abb2 | 3209 | mutex_lock(&inode->log_mutex); |
e02119d5 | 3210 | |
a491abb2 | 3211 | ret = btrfs_del_inode_ref(trans, log, name, name_len, btrfs_ino(inode), |
e02119d5 | 3212 | dirid, &index); |
a491abb2 | 3213 | mutex_unlock(&inode->log_mutex); |
4a500fd1 | 3214 | if (ret == -ENOSPC) { |
0b246afa | 3215 | btrfs_set_log_full_commit(fs_info, trans); |
4a500fd1 | 3216 | ret = 0; |
79787eaa | 3217 | } else if (ret < 0 && ret != -ENOENT) |
66642832 | 3218 | btrfs_abort_transaction(trans, ret); |
12fcfd22 | 3219 | btrfs_end_log_trans(root); |
e02119d5 | 3220 | |
e02119d5 CM |
3221 | return ret; |
3222 | } | |
3223 | ||
3224 | /* | |
3225 | * creates a range item in the log for 'dirid'. first_offset and | |
3226 | * last_offset tell us which parts of the key space the log should | |
3227 | * be considered authoritative for. | |
3228 | */ | |
3229 | static noinline int insert_dir_log_key(struct btrfs_trans_handle *trans, | |
3230 | struct btrfs_root *log, | |
3231 | struct btrfs_path *path, | |
3232 | int key_type, u64 dirid, | |
3233 | u64 first_offset, u64 last_offset) | |
3234 | { | |
3235 | int ret; | |
3236 | struct btrfs_key key; | |
3237 | struct btrfs_dir_log_item *item; | |
3238 | ||
3239 | key.objectid = dirid; | |
3240 | key.offset = first_offset; | |
3241 | if (key_type == BTRFS_DIR_ITEM_KEY) | |
3242 | key.type = BTRFS_DIR_LOG_ITEM_KEY; | |
3243 | else | |
3244 | key.type = BTRFS_DIR_LOG_INDEX_KEY; | |
3245 | ret = btrfs_insert_empty_item(trans, log, path, &key, sizeof(*item)); | |
4a500fd1 YZ |
3246 | if (ret) |
3247 | return ret; | |
e02119d5 CM |
3248 | |
3249 | item = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
3250 | struct btrfs_dir_log_item); | |
3251 | btrfs_set_dir_log_end(path->nodes[0], item, last_offset); | |
3252 | btrfs_mark_buffer_dirty(path->nodes[0]); | |
b3b4aa74 | 3253 | btrfs_release_path(path); |
e02119d5 CM |
3254 | return 0; |
3255 | } | |
3256 | ||
3257 | /* | |
3258 | * log all the items included in the current transaction for a given | |
3259 | * directory. This also creates the range items in the log tree required | |
3260 | * to replay anything deleted before the fsync | |
3261 | */ | |
3262 | static noinline int log_dir_items(struct btrfs_trans_handle *trans, | |
3263 | struct btrfs_root *root, struct inode *inode, | |
3264 | struct btrfs_path *path, | |
3265 | struct btrfs_path *dst_path, int key_type, | |
2f2ff0ee | 3266 | struct btrfs_log_ctx *ctx, |
e02119d5 CM |
3267 | u64 min_offset, u64 *last_offset_ret) |
3268 | { | |
3269 | struct btrfs_key min_key; | |
e02119d5 CM |
3270 | struct btrfs_root *log = root->log_root; |
3271 | struct extent_buffer *src; | |
4a500fd1 | 3272 | int err = 0; |
e02119d5 CM |
3273 | int ret; |
3274 | int i; | |
3275 | int nritems; | |
3276 | u64 first_offset = min_offset; | |
3277 | u64 last_offset = (u64)-1; | |
4a0cc7ca | 3278 | u64 ino = btrfs_ino(BTRFS_I(inode)); |
e02119d5 CM |
3279 | |
3280 | log = root->log_root; | |
e02119d5 | 3281 | |
33345d01 | 3282 | min_key.objectid = ino; |
e02119d5 CM |
3283 | min_key.type = key_type; |
3284 | min_key.offset = min_offset; | |
3285 | ||
6174d3cb | 3286 | ret = btrfs_search_forward(root, &min_key, path, trans->transid); |
e02119d5 CM |
3287 | |
3288 | /* | |
3289 | * we didn't find anything from this transaction, see if there | |
3290 | * is anything at all | |
3291 | */ | |
33345d01 LZ |
3292 | if (ret != 0 || min_key.objectid != ino || min_key.type != key_type) { |
3293 | min_key.objectid = ino; | |
e02119d5 CM |
3294 | min_key.type = key_type; |
3295 | min_key.offset = (u64)-1; | |
b3b4aa74 | 3296 | btrfs_release_path(path); |
e02119d5 CM |
3297 | ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0); |
3298 | if (ret < 0) { | |
b3b4aa74 | 3299 | btrfs_release_path(path); |
e02119d5 CM |
3300 | return ret; |
3301 | } | |
33345d01 | 3302 | ret = btrfs_previous_item(root, path, ino, key_type); |
e02119d5 CM |
3303 | |
3304 | /* if ret == 0 there are items for this type, | |
3305 | * create a range to tell us the last key of this type. | |
3306 | * otherwise, there are no items in this directory after | |
3307 | * *min_offset, and we create a range to indicate that. | |
3308 | */ | |
3309 | if (ret == 0) { | |
3310 | struct btrfs_key tmp; | |
3311 | btrfs_item_key_to_cpu(path->nodes[0], &tmp, | |
3312 | path->slots[0]); | |
d397712b | 3313 | if (key_type == tmp.type) |
e02119d5 | 3314 | first_offset = max(min_offset, tmp.offset) + 1; |
e02119d5 CM |
3315 | } |
3316 | goto done; | |
3317 | } | |
3318 | ||
3319 | /* go backward to find any previous key */ | |
33345d01 | 3320 | ret = btrfs_previous_item(root, path, ino, key_type); |
e02119d5 CM |
3321 | if (ret == 0) { |
3322 | struct btrfs_key tmp; | |
3323 | btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]); | |
3324 | if (key_type == tmp.type) { | |
3325 | first_offset = tmp.offset; | |
3326 | ret = overwrite_item(trans, log, dst_path, | |
3327 | path->nodes[0], path->slots[0], | |
3328 | &tmp); | |
4a500fd1 YZ |
3329 | if (ret) { |
3330 | err = ret; | |
3331 | goto done; | |
3332 | } | |
e02119d5 CM |
3333 | } |
3334 | } | |
b3b4aa74 | 3335 | btrfs_release_path(path); |
e02119d5 CM |
3336 | |
3337 | /* find the first key from this transaction again */ | |
3338 | ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0); | |
fae7f21c | 3339 | if (WARN_ON(ret != 0)) |
e02119d5 | 3340 | goto done; |
e02119d5 CM |
3341 | |
3342 | /* | |
3343 | * we have a block from this transaction, log every item in it | |
3344 | * from our directory | |
3345 | */ | |
d397712b | 3346 | while (1) { |
e02119d5 CM |
3347 | struct btrfs_key tmp; |
3348 | src = path->nodes[0]; | |
3349 | nritems = btrfs_header_nritems(src); | |
3350 | for (i = path->slots[0]; i < nritems; i++) { | |
2f2ff0ee FM |
3351 | struct btrfs_dir_item *di; |
3352 | ||
e02119d5 CM |
3353 | btrfs_item_key_to_cpu(src, &min_key, i); |
3354 | ||
33345d01 | 3355 | if (min_key.objectid != ino || min_key.type != key_type) |
e02119d5 CM |
3356 | goto done; |
3357 | ret = overwrite_item(trans, log, dst_path, src, i, | |
3358 | &min_key); | |
4a500fd1 YZ |
3359 | if (ret) { |
3360 | err = ret; | |
3361 | goto done; | |
3362 | } | |
2f2ff0ee FM |
3363 | |
3364 | /* | |
3365 | * We must make sure that when we log a directory entry, | |
3366 | * the corresponding inode, after log replay, has a | |
3367 | * matching link count. For example: | |
3368 | * | |
3369 | * touch foo | |
3370 | * mkdir mydir | |
3371 | * sync | |
3372 | * ln foo mydir/bar | |
3373 | * xfs_io -c "fsync" mydir | |
3374 | * <crash> | |
3375 | * <mount fs and log replay> | |
3376 | * | |
3377 | * Would result in a fsync log that when replayed, our | |
3378 | * file inode would have a link count of 1, but we get | |
3379 | * two directory entries pointing to the same inode. | |
3380 | * After removing one of the names, it would not be | |
3381 | * possible to remove the other name, which resulted | |
3382 | * always in stale file handle errors, and would not | |
3383 | * be possible to rmdir the parent directory, since | |
3384 | * its i_size could never decrement to the value | |
3385 | * BTRFS_EMPTY_DIR_SIZE, resulting in -ENOTEMPTY errors. | |
3386 | */ | |
3387 | di = btrfs_item_ptr(src, i, struct btrfs_dir_item); | |
3388 | btrfs_dir_item_key_to_cpu(src, di, &tmp); | |
3389 | if (ctx && | |
3390 | (btrfs_dir_transid(src, di) == trans->transid || | |
3391 | btrfs_dir_type(src, di) == BTRFS_FT_DIR) && | |
3392 | tmp.type != BTRFS_ROOT_ITEM_KEY) | |
3393 | ctx->log_new_dentries = true; | |
e02119d5 CM |
3394 | } |
3395 | path->slots[0] = nritems; | |
3396 | ||
3397 | /* | |
3398 | * look ahead to the next item and see if it is also | |
3399 | * from this directory and from this transaction | |
3400 | */ | |
3401 | ret = btrfs_next_leaf(root, path); | |
3402 | if (ret == 1) { | |
3403 | last_offset = (u64)-1; | |
3404 | goto done; | |
3405 | } | |
3406 | btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]); | |
33345d01 | 3407 | if (tmp.objectid != ino || tmp.type != key_type) { |
e02119d5 CM |
3408 | last_offset = (u64)-1; |
3409 | goto done; | |
3410 | } | |
3411 | if (btrfs_header_generation(path->nodes[0]) != trans->transid) { | |
3412 | ret = overwrite_item(trans, log, dst_path, | |
3413 | path->nodes[0], path->slots[0], | |
3414 | &tmp); | |
4a500fd1 YZ |
3415 | if (ret) |
3416 | err = ret; | |
3417 | else | |
3418 | last_offset = tmp.offset; | |
e02119d5 CM |
3419 | goto done; |
3420 | } | |
3421 | } | |
3422 | done: | |
b3b4aa74 DS |
3423 | btrfs_release_path(path); |
3424 | btrfs_release_path(dst_path); | |
e02119d5 | 3425 | |
4a500fd1 YZ |
3426 | if (err == 0) { |
3427 | *last_offset_ret = last_offset; | |
3428 | /* | |
3429 | * insert the log range keys to indicate where the log | |
3430 | * is valid | |
3431 | */ | |
3432 | ret = insert_dir_log_key(trans, log, path, key_type, | |
33345d01 | 3433 | ino, first_offset, last_offset); |
4a500fd1 YZ |
3434 | if (ret) |
3435 | err = ret; | |
3436 | } | |
3437 | return err; | |
e02119d5 CM |
3438 | } |
3439 | ||
3440 | /* | |
3441 | * logging directories is very similar to logging inodes, We find all the items | |
3442 | * from the current transaction and write them to the log. | |
3443 | * | |
3444 | * The recovery code scans the directory in the subvolume, and if it finds a | |
3445 | * key in the range logged that is not present in the log tree, then it means | |
3446 | * that dir entry was unlinked during the transaction. | |
3447 | * | |
3448 | * In order for that scan to work, we must include one key smaller than | |
3449 | * the smallest logged by this transaction and one key larger than the largest | |
3450 | * key logged by this transaction. | |
3451 | */ | |
3452 | static noinline int log_directory_changes(struct btrfs_trans_handle *trans, | |
3453 | struct btrfs_root *root, struct inode *inode, | |
3454 | struct btrfs_path *path, | |
2f2ff0ee FM |
3455 | struct btrfs_path *dst_path, |
3456 | struct btrfs_log_ctx *ctx) | |
e02119d5 CM |
3457 | { |
3458 | u64 min_key; | |
3459 | u64 max_key; | |
3460 | int ret; | |
3461 | int key_type = BTRFS_DIR_ITEM_KEY; | |
3462 | ||
3463 | again: | |
3464 | min_key = 0; | |
3465 | max_key = 0; | |
d397712b | 3466 | while (1) { |
e02119d5 | 3467 | ret = log_dir_items(trans, root, inode, path, |
2f2ff0ee | 3468 | dst_path, key_type, ctx, min_key, |
e02119d5 | 3469 | &max_key); |
4a500fd1 YZ |
3470 | if (ret) |
3471 | return ret; | |
e02119d5 CM |
3472 | if (max_key == (u64)-1) |
3473 | break; | |
3474 | min_key = max_key + 1; | |
3475 | } | |
3476 | ||
3477 | if (key_type == BTRFS_DIR_ITEM_KEY) { | |
3478 | key_type = BTRFS_DIR_INDEX_KEY; | |
3479 | goto again; | |
3480 | } | |
3481 | return 0; | |
3482 | } | |
3483 | ||
3484 | /* | |
3485 | * a helper function to drop items from the log before we relog an | |
3486 | * inode. max_key_type indicates the highest item type to remove. | |
3487 | * This cannot be run for file data extents because it does not | |
3488 | * free the extents they point to. | |
3489 | */ | |
3490 | static int drop_objectid_items(struct btrfs_trans_handle *trans, | |
3491 | struct btrfs_root *log, | |
3492 | struct btrfs_path *path, | |
3493 | u64 objectid, int max_key_type) | |
3494 | { | |
3495 | int ret; | |
3496 | struct btrfs_key key; | |
3497 | struct btrfs_key found_key; | |
18ec90d6 | 3498 | int start_slot; |
e02119d5 CM |
3499 | |
3500 | key.objectid = objectid; | |
3501 | key.type = max_key_type; | |
3502 | key.offset = (u64)-1; | |
3503 | ||
d397712b | 3504 | while (1) { |
e02119d5 | 3505 | ret = btrfs_search_slot(trans, log, &key, path, -1, 1); |
3650860b | 3506 | BUG_ON(ret == 0); /* Logic error */ |
4a500fd1 | 3507 | if (ret < 0) |
e02119d5 CM |
3508 | break; |
3509 | ||
3510 | if (path->slots[0] == 0) | |
3511 | break; | |
3512 | ||
3513 | path->slots[0]--; | |
3514 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, | |
3515 | path->slots[0]); | |
3516 | ||
3517 | if (found_key.objectid != objectid) | |
3518 | break; | |
3519 | ||
18ec90d6 JB |
3520 | found_key.offset = 0; |
3521 | found_key.type = 0; | |
3522 | ret = btrfs_bin_search(path->nodes[0], &found_key, 0, | |
3523 | &start_slot); | |
3524 | ||
3525 | ret = btrfs_del_items(trans, log, path, start_slot, | |
3526 | path->slots[0] - start_slot + 1); | |
3527 | /* | |
3528 | * If start slot isn't 0 then we don't need to re-search, we've | |
3529 | * found the last guy with the objectid in this tree. | |
3530 | */ | |
3531 | if (ret || start_slot != 0) | |
65a246c5 | 3532 | break; |
b3b4aa74 | 3533 | btrfs_release_path(path); |
e02119d5 | 3534 | } |
b3b4aa74 | 3535 | btrfs_release_path(path); |
5bdbeb21 JB |
3536 | if (ret > 0) |
3537 | ret = 0; | |
4a500fd1 | 3538 | return ret; |
e02119d5 CM |
3539 | } |
3540 | ||
94edf4ae JB |
3541 | static void fill_inode_item(struct btrfs_trans_handle *trans, |
3542 | struct extent_buffer *leaf, | |
3543 | struct btrfs_inode_item *item, | |
1a4bcf47 FM |
3544 | struct inode *inode, int log_inode_only, |
3545 | u64 logged_isize) | |
94edf4ae | 3546 | { |
0b1c6cca JB |
3547 | struct btrfs_map_token token; |
3548 | ||
3549 | btrfs_init_map_token(&token); | |
94edf4ae JB |
3550 | |
3551 | if (log_inode_only) { | |
3552 | /* set the generation to zero so the recover code | |
3553 | * can tell the difference between an logging | |
3554 | * just to say 'this inode exists' and a logging | |
3555 | * to say 'update this inode with these values' | |
3556 | */ | |
0b1c6cca | 3557 | btrfs_set_token_inode_generation(leaf, item, 0, &token); |
1a4bcf47 | 3558 | btrfs_set_token_inode_size(leaf, item, logged_isize, &token); |
94edf4ae | 3559 | } else { |
0b1c6cca JB |
3560 | btrfs_set_token_inode_generation(leaf, item, |
3561 | BTRFS_I(inode)->generation, | |
3562 | &token); | |
3563 | btrfs_set_token_inode_size(leaf, item, inode->i_size, &token); | |
3564 | } | |
3565 | ||
3566 | btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token); | |
3567 | btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token); | |
3568 | btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token); | |
3569 | btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token); | |
3570 | ||
a937b979 | 3571 | btrfs_set_token_timespec_sec(leaf, &item->atime, |
0b1c6cca | 3572 | inode->i_atime.tv_sec, &token); |
a937b979 | 3573 | btrfs_set_token_timespec_nsec(leaf, &item->atime, |
0b1c6cca JB |
3574 | inode->i_atime.tv_nsec, &token); |
3575 | ||
a937b979 | 3576 | btrfs_set_token_timespec_sec(leaf, &item->mtime, |
0b1c6cca | 3577 | inode->i_mtime.tv_sec, &token); |
a937b979 | 3578 | btrfs_set_token_timespec_nsec(leaf, &item->mtime, |
0b1c6cca JB |
3579 | inode->i_mtime.tv_nsec, &token); |
3580 | ||
a937b979 | 3581 | btrfs_set_token_timespec_sec(leaf, &item->ctime, |
0b1c6cca | 3582 | inode->i_ctime.tv_sec, &token); |
a937b979 | 3583 | btrfs_set_token_timespec_nsec(leaf, &item->ctime, |
0b1c6cca JB |
3584 | inode->i_ctime.tv_nsec, &token); |
3585 | ||
3586 | btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode), | |
3587 | &token); | |
3588 | ||
3589 | btrfs_set_token_inode_sequence(leaf, item, inode->i_version, &token); | |
3590 | btrfs_set_token_inode_transid(leaf, item, trans->transid, &token); | |
3591 | btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token); | |
3592 | btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token); | |
3593 | btrfs_set_token_inode_block_group(leaf, item, 0, &token); | |
94edf4ae JB |
3594 | } |
3595 | ||
a95249b3 JB |
3596 | static int log_inode_item(struct btrfs_trans_handle *trans, |
3597 | struct btrfs_root *log, struct btrfs_path *path, | |
3598 | struct inode *inode) | |
3599 | { | |
3600 | struct btrfs_inode_item *inode_item; | |
a95249b3 JB |
3601 | int ret; |
3602 | ||
efd0c405 FDBM |
3603 | ret = btrfs_insert_empty_item(trans, log, path, |
3604 | &BTRFS_I(inode)->location, | |
a95249b3 JB |
3605 | sizeof(*inode_item)); |
3606 | if (ret && ret != -EEXIST) | |
3607 | return ret; | |
3608 | inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
3609 | struct btrfs_inode_item); | |
1a4bcf47 | 3610 | fill_inode_item(trans, path->nodes[0], inode_item, inode, 0, 0); |
a95249b3 JB |
3611 | btrfs_release_path(path); |
3612 | return 0; | |
3613 | } | |
3614 | ||
31ff1cd2 | 3615 | static noinline int copy_items(struct btrfs_trans_handle *trans, |
44d70e19 | 3616 | struct btrfs_inode *inode, |
31ff1cd2 | 3617 | struct btrfs_path *dst_path, |
16e7549f | 3618 | struct btrfs_path *src_path, u64 *last_extent, |
1a4bcf47 FM |
3619 | int start_slot, int nr, int inode_only, |
3620 | u64 logged_isize) | |
31ff1cd2 | 3621 | { |
44d70e19 | 3622 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb); |
31ff1cd2 CM |
3623 | unsigned long src_offset; |
3624 | unsigned long dst_offset; | |
44d70e19 | 3625 | struct btrfs_root *log = inode->root->log_root; |
31ff1cd2 CM |
3626 | struct btrfs_file_extent_item *extent; |
3627 | struct btrfs_inode_item *inode_item; | |
16e7549f JB |
3628 | struct extent_buffer *src = src_path->nodes[0]; |
3629 | struct btrfs_key first_key, last_key, key; | |
31ff1cd2 CM |
3630 | int ret; |
3631 | struct btrfs_key *ins_keys; | |
3632 | u32 *ins_sizes; | |
3633 | char *ins_data; | |
3634 | int i; | |
d20f7043 | 3635 | struct list_head ordered_sums; |
44d70e19 | 3636 | int skip_csum = inode->flags & BTRFS_INODE_NODATASUM; |
16e7549f | 3637 | bool has_extents = false; |
74121f7c | 3638 | bool need_find_last_extent = true; |
16e7549f | 3639 | bool done = false; |
d20f7043 CM |
3640 | |
3641 | INIT_LIST_HEAD(&ordered_sums); | |
31ff1cd2 CM |
3642 | |
3643 | ins_data = kmalloc(nr * sizeof(struct btrfs_key) + | |
3644 | nr * sizeof(u32), GFP_NOFS); | |
2a29edc6 | 3645 | if (!ins_data) |
3646 | return -ENOMEM; | |
3647 | ||
16e7549f JB |
3648 | first_key.objectid = (u64)-1; |
3649 | ||
31ff1cd2 CM |
3650 | ins_sizes = (u32 *)ins_data; |
3651 | ins_keys = (struct btrfs_key *)(ins_data + nr * sizeof(u32)); | |
3652 | ||
3653 | for (i = 0; i < nr; i++) { | |
3654 | ins_sizes[i] = btrfs_item_size_nr(src, i + start_slot); | |
3655 | btrfs_item_key_to_cpu(src, ins_keys + i, i + start_slot); | |
3656 | } | |
3657 | ret = btrfs_insert_empty_items(trans, log, dst_path, | |
3658 | ins_keys, ins_sizes, nr); | |
4a500fd1 YZ |
3659 | if (ret) { |
3660 | kfree(ins_data); | |
3661 | return ret; | |
3662 | } | |
31ff1cd2 | 3663 | |
5d4f98a2 | 3664 | for (i = 0; i < nr; i++, dst_path->slots[0]++) { |
31ff1cd2 CM |
3665 | dst_offset = btrfs_item_ptr_offset(dst_path->nodes[0], |
3666 | dst_path->slots[0]); | |
3667 | ||
3668 | src_offset = btrfs_item_ptr_offset(src, start_slot + i); | |
3669 | ||
16e7549f JB |
3670 | if ((i == (nr - 1))) |
3671 | last_key = ins_keys[i]; | |
3672 | ||
94edf4ae | 3673 | if (ins_keys[i].type == BTRFS_INODE_ITEM_KEY) { |
31ff1cd2 CM |
3674 | inode_item = btrfs_item_ptr(dst_path->nodes[0], |
3675 | dst_path->slots[0], | |
3676 | struct btrfs_inode_item); | |
94edf4ae | 3677 | fill_inode_item(trans, dst_path->nodes[0], inode_item, |
44d70e19 | 3678 | &inode->vfs_inode, inode_only == LOG_INODE_EXISTS, |
1a4bcf47 | 3679 | logged_isize); |
94edf4ae JB |
3680 | } else { |
3681 | copy_extent_buffer(dst_path->nodes[0], src, dst_offset, | |
3682 | src_offset, ins_sizes[i]); | |
31ff1cd2 | 3683 | } |
94edf4ae | 3684 | |
16e7549f JB |
3685 | /* |
3686 | * We set need_find_last_extent here in case we know we were | |
3687 | * processing other items and then walk into the first extent in | |
3688 | * the inode. If we don't hit an extent then nothing changes, | |
3689 | * we'll do the last search the next time around. | |
3690 | */ | |
3691 | if (ins_keys[i].type == BTRFS_EXTENT_DATA_KEY) { | |
3692 | has_extents = true; | |
74121f7c | 3693 | if (first_key.objectid == (u64)-1) |
16e7549f JB |
3694 | first_key = ins_keys[i]; |
3695 | } else { | |
3696 | need_find_last_extent = false; | |
3697 | } | |
3698 | ||
31ff1cd2 CM |
3699 | /* take a reference on file data extents so that truncates |
3700 | * or deletes of this inode don't have to relog the inode | |
3701 | * again | |
3702 | */ | |
962a298f | 3703 | if (ins_keys[i].type == BTRFS_EXTENT_DATA_KEY && |
d2794405 | 3704 | !skip_csum) { |
31ff1cd2 CM |
3705 | int found_type; |
3706 | extent = btrfs_item_ptr(src, start_slot + i, | |
3707 | struct btrfs_file_extent_item); | |
3708 | ||
8e531cdf | 3709 | if (btrfs_file_extent_generation(src, extent) < trans->transid) |
3710 | continue; | |
3711 | ||
31ff1cd2 | 3712 | found_type = btrfs_file_extent_type(src, extent); |
6f1fed77 | 3713 | if (found_type == BTRFS_FILE_EXTENT_REG) { |
5d4f98a2 YZ |
3714 | u64 ds, dl, cs, cl; |
3715 | ds = btrfs_file_extent_disk_bytenr(src, | |
3716 | extent); | |
3717 | /* ds == 0 is a hole */ | |
3718 | if (ds == 0) | |
3719 | continue; | |
3720 | ||
3721 | dl = btrfs_file_extent_disk_num_bytes(src, | |
3722 | extent); | |
3723 | cs = btrfs_file_extent_offset(src, extent); | |
3724 | cl = btrfs_file_extent_num_bytes(src, | |
a419aef8 | 3725 | extent); |
580afd76 CM |
3726 | if (btrfs_file_extent_compression(src, |
3727 | extent)) { | |
3728 | cs = 0; | |
3729 | cl = dl; | |
3730 | } | |
5d4f98a2 YZ |
3731 | |
3732 | ret = btrfs_lookup_csums_range( | |
0b246afa | 3733 | fs_info->csum_root, |
5d4f98a2 | 3734 | ds + cs, ds + cs + cl - 1, |
a2de733c | 3735 | &ordered_sums, 0); |
3650860b JB |
3736 | if (ret) { |
3737 | btrfs_release_path(dst_path); | |
3738 | kfree(ins_data); | |
3739 | return ret; | |
3740 | } | |
31ff1cd2 CM |
3741 | } |
3742 | } | |
31ff1cd2 CM |
3743 | } |
3744 | ||
3745 | btrfs_mark_buffer_dirty(dst_path->nodes[0]); | |
b3b4aa74 | 3746 | btrfs_release_path(dst_path); |
31ff1cd2 | 3747 | kfree(ins_data); |
d20f7043 CM |
3748 | |
3749 | /* | |
3750 | * we have to do this after the loop above to avoid changing the | |
3751 | * log tree while trying to change the log tree. | |
3752 | */ | |
4a500fd1 | 3753 | ret = 0; |
d397712b | 3754 | while (!list_empty(&ordered_sums)) { |
d20f7043 CM |
3755 | struct btrfs_ordered_sum *sums = list_entry(ordered_sums.next, |
3756 | struct btrfs_ordered_sum, | |
3757 | list); | |
4a500fd1 YZ |
3758 | if (!ret) |
3759 | ret = btrfs_csum_file_blocks(trans, log, sums); | |
d20f7043 CM |
3760 | list_del(&sums->list); |
3761 | kfree(sums); | |
3762 | } | |
16e7549f JB |
3763 | |
3764 | if (!has_extents) | |
3765 | return ret; | |
3766 | ||
74121f7c FM |
3767 | if (need_find_last_extent && *last_extent == first_key.offset) { |
3768 | /* | |
3769 | * We don't have any leafs between our current one and the one | |
3770 | * we processed before that can have file extent items for our | |
3771 | * inode (and have a generation number smaller than our current | |
3772 | * transaction id). | |
3773 | */ | |
3774 | need_find_last_extent = false; | |
3775 | } | |
3776 | ||
16e7549f JB |
3777 | /* |
3778 | * Because we use btrfs_search_forward we could skip leaves that were | |
3779 | * not modified and then assume *last_extent is valid when it really | |
3780 | * isn't. So back up to the previous leaf and read the end of the last | |
3781 | * extent before we go and fill in holes. | |
3782 | */ | |
3783 | if (need_find_last_extent) { | |
3784 | u64 len; | |
3785 | ||
44d70e19 | 3786 | ret = btrfs_prev_leaf(inode->root, src_path); |
16e7549f JB |
3787 | if (ret < 0) |
3788 | return ret; | |
3789 | if (ret) | |
3790 | goto fill_holes; | |
3791 | if (src_path->slots[0]) | |
3792 | src_path->slots[0]--; | |
3793 | src = src_path->nodes[0]; | |
3794 | btrfs_item_key_to_cpu(src, &key, src_path->slots[0]); | |
44d70e19 | 3795 | if (key.objectid != btrfs_ino(inode) || |
16e7549f JB |
3796 | key.type != BTRFS_EXTENT_DATA_KEY) |
3797 | goto fill_holes; | |
3798 | extent = btrfs_item_ptr(src, src_path->slots[0], | |
3799 | struct btrfs_file_extent_item); | |
3800 | if (btrfs_file_extent_type(src, extent) == | |
3801 | BTRFS_FILE_EXTENT_INLINE) { | |
514ac8ad CM |
3802 | len = btrfs_file_extent_inline_len(src, |
3803 | src_path->slots[0], | |
3804 | extent); | |
16e7549f | 3805 | *last_extent = ALIGN(key.offset + len, |
0b246afa | 3806 | fs_info->sectorsize); |
16e7549f JB |
3807 | } else { |
3808 | len = btrfs_file_extent_num_bytes(src, extent); | |
3809 | *last_extent = key.offset + len; | |
3810 | } | |
3811 | } | |
3812 | fill_holes: | |
3813 | /* So we did prev_leaf, now we need to move to the next leaf, but a few | |
3814 | * things could have happened | |
3815 | * | |
3816 | * 1) A merge could have happened, so we could currently be on a leaf | |
3817 | * that holds what we were copying in the first place. | |
3818 | * 2) A split could have happened, and now not all of the items we want | |
3819 | * are on the same leaf. | |
3820 | * | |
3821 | * So we need to adjust how we search for holes, we need to drop the | |
3822 | * path and re-search for the first extent key we found, and then walk | |
3823 | * forward until we hit the last one we copied. | |
3824 | */ | |
3825 | if (need_find_last_extent) { | |
3826 | /* btrfs_prev_leaf could return 1 without releasing the path */ | |
3827 | btrfs_release_path(src_path); | |
44d70e19 | 3828 | ret = btrfs_search_slot(NULL, inode->root, &first_key, src_path, 0, 0); |
16e7549f JB |
3829 | if (ret < 0) |
3830 | return ret; | |
3831 | ASSERT(ret == 0); | |
3832 | src = src_path->nodes[0]; | |
3833 | i = src_path->slots[0]; | |
3834 | } else { | |
3835 | i = start_slot; | |
3836 | } | |
3837 | ||
3838 | /* | |
3839 | * Ok so here we need to go through and fill in any holes we may have | |
3840 | * to make sure that holes are punched for those areas in case they had | |
3841 | * extents previously. | |
3842 | */ | |
3843 | while (!done) { | |
3844 | u64 offset, len; | |
3845 | u64 extent_end; | |
3846 | ||
3847 | if (i >= btrfs_header_nritems(src_path->nodes[0])) { | |
44d70e19 | 3848 | ret = btrfs_next_leaf(inode->root, src_path); |
16e7549f JB |
3849 | if (ret < 0) |
3850 | return ret; | |
3851 | ASSERT(ret == 0); | |
3852 | src = src_path->nodes[0]; | |
3853 | i = 0; | |
3854 | } | |
3855 | ||
3856 | btrfs_item_key_to_cpu(src, &key, i); | |
3857 | if (!btrfs_comp_cpu_keys(&key, &last_key)) | |
3858 | done = true; | |
44d70e19 | 3859 | if (key.objectid != btrfs_ino(inode) || |
16e7549f JB |
3860 | key.type != BTRFS_EXTENT_DATA_KEY) { |
3861 | i++; | |
3862 | continue; | |
3863 | } | |
3864 | extent = btrfs_item_ptr(src, i, struct btrfs_file_extent_item); | |
3865 | if (btrfs_file_extent_type(src, extent) == | |
3866 | BTRFS_FILE_EXTENT_INLINE) { | |
514ac8ad | 3867 | len = btrfs_file_extent_inline_len(src, i, extent); |
da17066c | 3868 | extent_end = ALIGN(key.offset + len, |
0b246afa | 3869 | fs_info->sectorsize); |
16e7549f JB |
3870 | } else { |
3871 | len = btrfs_file_extent_num_bytes(src, extent); | |
3872 | extent_end = key.offset + len; | |
3873 | } | |
3874 | i++; | |
3875 | ||
3876 | if (*last_extent == key.offset) { | |
3877 | *last_extent = extent_end; | |
3878 | continue; | |
3879 | } | |
3880 | offset = *last_extent; | |
3881 | len = key.offset - *last_extent; | |
44d70e19 NB |
3882 | ret = btrfs_insert_file_extent(trans, log, btrfs_ino(inode), |
3883 | offset, 0, 0, len, 0, len, 0, 0, 0); | |
16e7549f JB |
3884 | if (ret) |
3885 | break; | |
74121f7c | 3886 | *last_extent = extent_end; |
16e7549f JB |
3887 | } |
3888 | /* | |
3889 | * Need to let the callers know we dropped the path so they should | |
3890 | * re-search. | |
3891 | */ | |
3892 | if (!ret && need_find_last_extent) | |
3893 | ret = 1; | |
4a500fd1 | 3894 | return ret; |
31ff1cd2 CM |
3895 | } |
3896 | ||
5dc562c5 JB |
3897 | static int extent_cmp(void *priv, struct list_head *a, struct list_head *b) |
3898 | { | |
3899 | struct extent_map *em1, *em2; | |
3900 | ||
3901 | em1 = list_entry(a, struct extent_map, list); | |
3902 | em2 = list_entry(b, struct extent_map, list); | |
3903 | ||
3904 | if (em1->start < em2->start) | |
3905 | return -1; | |
3906 | else if (em1->start > em2->start) | |
3907 | return 1; | |
3908 | return 0; | |
3909 | } | |
3910 | ||
8407f553 FM |
3911 | static int wait_ordered_extents(struct btrfs_trans_handle *trans, |
3912 | struct inode *inode, | |
3913 | struct btrfs_root *root, | |
3914 | const struct extent_map *em, | |
3915 | const struct list_head *logged_list, | |
3916 | bool *ordered_io_error) | |
5dc562c5 | 3917 | { |
0b246afa | 3918 | struct btrfs_fs_info *fs_info = root->fs_info; |
2ab28f32 | 3919 | struct btrfs_ordered_extent *ordered; |
8407f553 | 3920 | struct btrfs_root *log = root->log_root; |
2ab28f32 JB |
3921 | u64 mod_start = em->mod_start; |
3922 | u64 mod_len = em->mod_len; | |
8407f553 | 3923 | const bool skip_csum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM; |
2ab28f32 JB |
3924 | u64 csum_offset; |
3925 | u64 csum_len; | |
8407f553 FM |
3926 | LIST_HEAD(ordered_sums); |
3927 | int ret = 0; | |
0aa4a17d | 3928 | |
8407f553 | 3929 | *ordered_io_error = false; |
0aa4a17d | 3930 | |
8407f553 FM |
3931 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) || |
3932 | em->block_start == EXTENT_MAP_HOLE) | |
70c8a91c | 3933 | return 0; |
5dc562c5 | 3934 | |
2ab28f32 | 3935 | /* |
8407f553 FM |
3936 | * Wait far any ordered extent that covers our extent map. If it |
3937 | * finishes without an error, first check and see if our csums are on | |
3938 | * our outstanding ordered extents. | |
2ab28f32 | 3939 | */ |
827463c4 | 3940 | list_for_each_entry(ordered, logged_list, log_list) { |
2ab28f32 JB |
3941 | struct btrfs_ordered_sum *sum; |
3942 | ||
3943 | if (!mod_len) | |
3944 | break; | |
3945 | ||
2ab28f32 JB |
3946 | if (ordered->file_offset + ordered->len <= mod_start || |
3947 | mod_start + mod_len <= ordered->file_offset) | |
3948 | continue; | |
3949 | ||
8407f553 FM |
3950 | if (!test_bit(BTRFS_ORDERED_IO_DONE, &ordered->flags) && |
3951 | !test_bit(BTRFS_ORDERED_IOERR, &ordered->flags) && | |
3952 | !test_bit(BTRFS_ORDERED_DIRECT, &ordered->flags)) { | |
3953 | const u64 start = ordered->file_offset; | |
3954 | const u64 end = ordered->file_offset + ordered->len - 1; | |
3955 | ||
3956 | WARN_ON(ordered->inode != inode); | |
3957 | filemap_fdatawrite_range(inode->i_mapping, start, end); | |
3958 | } | |
3959 | ||
3960 | wait_event(ordered->wait, | |
3961 | (test_bit(BTRFS_ORDERED_IO_DONE, &ordered->flags) || | |
3962 | test_bit(BTRFS_ORDERED_IOERR, &ordered->flags))); | |
3963 | ||
3964 | if (test_bit(BTRFS_ORDERED_IOERR, &ordered->flags)) { | |
b38ef71c FM |
3965 | /* |
3966 | * Clear the AS_EIO/AS_ENOSPC flags from the inode's | |
3967 | * i_mapping flags, so that the next fsync won't get | |
3968 | * an outdated io error too. | |
3969 | */ | |
f0312210 | 3970 | filemap_check_errors(inode->i_mapping); |
8407f553 FM |
3971 | *ordered_io_error = true; |
3972 | break; | |
3973 | } | |
2ab28f32 JB |
3974 | /* |
3975 | * We are going to copy all the csums on this ordered extent, so | |
3976 | * go ahead and adjust mod_start and mod_len in case this | |
3977 | * ordered extent has already been logged. | |
3978 | */ | |
3979 | if (ordered->file_offset > mod_start) { | |
3980 | if (ordered->file_offset + ordered->len >= | |
3981 | mod_start + mod_len) | |
3982 | mod_len = ordered->file_offset - mod_start; | |
3983 | /* | |
3984 | * If we have this case | |
3985 | * | |
3986 | * |--------- logged extent ---------| | |
3987 | * |----- ordered extent ----| | |
3988 | * | |
3989 | * Just don't mess with mod_start and mod_len, we'll | |
3990 | * just end up logging more csums than we need and it | |
3991 | * will be ok. | |
3992 | */ | |
3993 | } else { | |
3994 | if (ordered->file_offset + ordered->len < | |
3995 | mod_start + mod_len) { | |
3996 | mod_len = (mod_start + mod_len) - | |
3997 | (ordered->file_offset + ordered->len); | |
3998 | mod_start = ordered->file_offset + | |
3999 | ordered->len; | |
4000 | } else { | |
4001 | mod_len = 0; | |
4002 | } | |
4003 | } | |
4004 | ||
8407f553 FM |
4005 | if (skip_csum) |
4006 | continue; | |
4007 | ||
2ab28f32 JB |
4008 | /* |
4009 | * To keep us from looping for the above case of an ordered | |
4010 | * extent that falls inside of the logged extent. | |
4011 | */ | |
4012 | if (test_and_set_bit(BTRFS_ORDERED_LOGGED_CSUM, | |
4013 | &ordered->flags)) | |
4014 | continue; | |
2ab28f32 | 4015 | |
2ab28f32 JB |
4016 | list_for_each_entry(sum, &ordered->list, list) { |
4017 | ret = btrfs_csum_file_blocks(trans, log, sum); | |
827463c4 | 4018 | if (ret) |
8407f553 | 4019 | break; |
2ab28f32 | 4020 | } |
2ab28f32 | 4021 | } |
2ab28f32 | 4022 | |
8407f553 | 4023 | if (*ordered_io_error || !mod_len || ret || skip_csum) |
2ab28f32 JB |
4024 | return ret; |
4025 | ||
488111aa FDBM |
4026 | if (em->compress_type) { |
4027 | csum_offset = 0; | |
8407f553 | 4028 | csum_len = max(em->block_len, em->orig_block_len); |
488111aa FDBM |
4029 | } else { |
4030 | csum_offset = mod_start - em->start; | |
4031 | csum_len = mod_len; | |
4032 | } | |
2ab28f32 | 4033 | |
70c8a91c | 4034 | /* block start is already adjusted for the file extent offset. */ |
0b246afa | 4035 | ret = btrfs_lookup_csums_range(fs_info->csum_root, |
70c8a91c JB |
4036 | em->block_start + csum_offset, |
4037 | em->block_start + csum_offset + | |
4038 | csum_len - 1, &ordered_sums, 0); | |
4039 | if (ret) | |
4040 | return ret; | |
5dc562c5 | 4041 | |
70c8a91c JB |
4042 | while (!list_empty(&ordered_sums)) { |
4043 | struct btrfs_ordered_sum *sums = list_entry(ordered_sums.next, | |
4044 | struct btrfs_ordered_sum, | |
4045 | list); | |
4046 | if (!ret) | |
4047 | ret = btrfs_csum_file_blocks(trans, log, sums); | |
4048 | list_del(&sums->list); | |
4049 | kfree(sums); | |
5dc562c5 JB |
4050 | } |
4051 | ||
70c8a91c | 4052 | return ret; |
5dc562c5 JB |
4053 | } |
4054 | ||
8407f553 FM |
4055 | static int log_one_extent(struct btrfs_trans_handle *trans, |
4056 | struct inode *inode, struct btrfs_root *root, | |
4057 | const struct extent_map *em, | |
4058 | struct btrfs_path *path, | |
4059 | const struct list_head *logged_list, | |
4060 | struct btrfs_log_ctx *ctx) | |
4061 | { | |
4062 | struct btrfs_root *log = root->log_root; | |
4063 | struct btrfs_file_extent_item *fi; | |
4064 | struct extent_buffer *leaf; | |
4065 | struct btrfs_map_token token; | |
4066 | struct btrfs_key key; | |
4067 | u64 extent_offset = em->start - em->orig_start; | |
4068 | u64 block_len; | |
4069 | int ret; | |
4070 | int extent_inserted = 0; | |
4071 | bool ordered_io_err = false; | |
4072 | ||
4073 | ret = wait_ordered_extents(trans, inode, root, em, logged_list, | |
4074 | &ordered_io_err); | |
4075 | if (ret) | |
4076 | return ret; | |
4077 | ||
4078 | if (ordered_io_err) { | |
4079 | ctx->io_err = -EIO; | |
4080 | return 0; | |
4081 | } | |
4082 | ||
4083 | btrfs_init_map_token(&token); | |
4084 | ||
4085 | ret = __btrfs_drop_extents(trans, log, inode, path, em->start, | |
4086 | em->start + em->len, NULL, 0, 1, | |
4087 | sizeof(*fi), &extent_inserted); | |
4088 | if (ret) | |
4089 | return ret; | |
4090 | ||
4091 | if (!extent_inserted) { | |
4a0cc7ca | 4092 | key.objectid = btrfs_ino(BTRFS_I(inode)); |
8407f553 FM |
4093 | key.type = BTRFS_EXTENT_DATA_KEY; |
4094 | key.offset = em->start; | |
4095 | ||
4096 | ret = btrfs_insert_empty_item(trans, log, path, &key, | |
4097 | sizeof(*fi)); | |
4098 | if (ret) | |
4099 | return ret; | |
4100 | } | |
4101 | leaf = path->nodes[0]; | |
4102 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
4103 | struct btrfs_file_extent_item); | |
4104 | ||
50d9aa99 | 4105 | btrfs_set_token_file_extent_generation(leaf, fi, trans->transid, |
8407f553 FM |
4106 | &token); |
4107 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) | |
4108 | btrfs_set_token_file_extent_type(leaf, fi, | |
4109 | BTRFS_FILE_EXTENT_PREALLOC, | |
4110 | &token); | |
4111 | else | |
4112 | btrfs_set_token_file_extent_type(leaf, fi, | |
4113 | BTRFS_FILE_EXTENT_REG, | |
4114 | &token); | |
4115 | ||
4116 | block_len = max(em->block_len, em->orig_block_len); | |
4117 | if (em->compress_type != BTRFS_COMPRESS_NONE) { | |
4118 | btrfs_set_token_file_extent_disk_bytenr(leaf, fi, | |
4119 | em->block_start, | |
4120 | &token); | |
4121 | btrfs_set_token_file_extent_disk_num_bytes(leaf, fi, block_len, | |
4122 | &token); | |
4123 | } else if (em->block_start < EXTENT_MAP_LAST_BYTE) { | |
4124 | btrfs_set_token_file_extent_disk_bytenr(leaf, fi, | |
4125 | em->block_start - | |
4126 | extent_offset, &token); | |
4127 | btrfs_set_token_file_extent_disk_num_bytes(leaf, fi, block_len, | |
4128 | &token); | |
4129 | } else { | |
4130 | btrfs_set_token_file_extent_disk_bytenr(leaf, fi, 0, &token); | |
4131 | btrfs_set_token_file_extent_disk_num_bytes(leaf, fi, 0, | |
4132 | &token); | |
4133 | } | |
4134 | ||
4135 | btrfs_set_token_file_extent_offset(leaf, fi, extent_offset, &token); | |
4136 | btrfs_set_token_file_extent_num_bytes(leaf, fi, em->len, &token); | |
4137 | btrfs_set_token_file_extent_ram_bytes(leaf, fi, em->ram_bytes, &token); | |
4138 | btrfs_set_token_file_extent_compression(leaf, fi, em->compress_type, | |
4139 | &token); | |
4140 | btrfs_set_token_file_extent_encryption(leaf, fi, 0, &token); | |
4141 | btrfs_set_token_file_extent_other_encoding(leaf, fi, 0, &token); | |
4142 | btrfs_mark_buffer_dirty(leaf); | |
4143 | ||
4144 | btrfs_release_path(path); | |
4145 | ||
4146 | return ret; | |
4147 | } | |
4148 | ||
5dc562c5 JB |
4149 | static int btrfs_log_changed_extents(struct btrfs_trans_handle *trans, |
4150 | struct btrfs_root *root, | |
4151 | struct inode *inode, | |
827463c4 | 4152 | struct btrfs_path *path, |
8407f553 | 4153 | struct list_head *logged_list, |
de0ee0ed FM |
4154 | struct btrfs_log_ctx *ctx, |
4155 | const u64 start, | |
4156 | const u64 end) | |
5dc562c5 | 4157 | { |
5dc562c5 JB |
4158 | struct extent_map *em, *n; |
4159 | struct list_head extents; | |
4160 | struct extent_map_tree *tree = &BTRFS_I(inode)->extent_tree; | |
4161 | u64 test_gen; | |
4162 | int ret = 0; | |
2ab28f32 | 4163 | int num = 0; |
5dc562c5 JB |
4164 | |
4165 | INIT_LIST_HEAD(&extents); | |
4166 | ||
5f9a8a51 | 4167 | down_write(&BTRFS_I(inode)->dio_sem); |
5dc562c5 JB |
4168 | write_lock(&tree->lock); |
4169 | test_gen = root->fs_info->last_trans_committed; | |
4170 | ||
4171 | list_for_each_entry_safe(em, n, &tree->modified_extents, list) { | |
4172 | list_del_init(&em->list); | |
2ab28f32 JB |
4173 | |
4174 | /* | |
4175 | * Just an arbitrary number, this can be really CPU intensive | |
4176 | * once we start getting a lot of extents, and really once we | |
4177 | * have a bunch of extents we just want to commit since it will | |
4178 | * be faster. | |
4179 | */ | |
4180 | if (++num > 32768) { | |
4181 | list_del_init(&tree->modified_extents); | |
4182 | ret = -EFBIG; | |
4183 | goto process; | |
4184 | } | |
4185 | ||
5dc562c5 JB |
4186 | if (em->generation <= test_gen) |
4187 | continue; | |
ff44c6e3 JB |
4188 | /* Need a ref to keep it from getting evicted from cache */ |
4189 | atomic_inc(&em->refs); | |
4190 | set_bit(EXTENT_FLAG_LOGGING, &em->flags); | |
5dc562c5 | 4191 | list_add_tail(&em->list, &extents); |
2ab28f32 | 4192 | num++; |
5dc562c5 JB |
4193 | } |
4194 | ||
4195 | list_sort(NULL, &extents, extent_cmp); | |
5f9a8a51 | 4196 | btrfs_get_logged_extents(inode, logged_list, start, end); |
de0ee0ed | 4197 | /* |
5f9a8a51 FM |
4198 | * Some ordered extents started by fsync might have completed |
4199 | * before we could collect them into the list logged_list, which | |
4200 | * means they're gone, not in our logged_list nor in the inode's | |
4201 | * ordered tree. We want the application/user space to know an | |
4202 | * error happened while attempting to persist file data so that | |
4203 | * it can take proper action. If such error happened, we leave | |
4204 | * without writing to the log tree and the fsync must report the | |
4205 | * file data write error and not commit the current transaction. | |
de0ee0ed | 4206 | */ |
f0312210 | 4207 | ret = filemap_check_errors(inode->i_mapping); |
5f9a8a51 FM |
4208 | if (ret) |
4209 | ctx->io_err = ret; | |
2ab28f32 | 4210 | process: |
5dc562c5 JB |
4211 | while (!list_empty(&extents)) { |
4212 | em = list_entry(extents.next, struct extent_map, list); | |
4213 | ||
4214 | list_del_init(&em->list); | |
4215 | ||
4216 | /* | |
4217 | * If we had an error we just need to delete everybody from our | |
4218 | * private list. | |
4219 | */ | |
ff44c6e3 | 4220 | if (ret) { |
201a9038 | 4221 | clear_em_logging(tree, em); |
ff44c6e3 | 4222 | free_extent_map(em); |
5dc562c5 | 4223 | continue; |
ff44c6e3 JB |
4224 | } |
4225 | ||
4226 | write_unlock(&tree->lock); | |
5dc562c5 | 4227 | |
8407f553 FM |
4228 | ret = log_one_extent(trans, inode, root, em, path, logged_list, |
4229 | ctx); | |
ff44c6e3 | 4230 | write_lock(&tree->lock); |
201a9038 JB |
4231 | clear_em_logging(tree, em); |
4232 | free_extent_map(em); | |
5dc562c5 | 4233 | } |
ff44c6e3 JB |
4234 | WARN_ON(!list_empty(&extents)); |
4235 | write_unlock(&tree->lock); | |
5f9a8a51 | 4236 | up_write(&BTRFS_I(inode)->dio_sem); |
5dc562c5 | 4237 | |
5dc562c5 | 4238 | btrfs_release_path(path); |
5dc562c5 JB |
4239 | return ret; |
4240 | } | |
4241 | ||
481b01c0 | 4242 | static int logged_inode_size(struct btrfs_root *log, struct btrfs_inode *inode, |
1a4bcf47 FM |
4243 | struct btrfs_path *path, u64 *size_ret) |
4244 | { | |
4245 | struct btrfs_key key; | |
4246 | int ret; | |
4247 | ||
481b01c0 | 4248 | key.objectid = btrfs_ino(inode); |
1a4bcf47 FM |
4249 | key.type = BTRFS_INODE_ITEM_KEY; |
4250 | key.offset = 0; | |
4251 | ||
4252 | ret = btrfs_search_slot(NULL, log, &key, path, 0, 0); | |
4253 | if (ret < 0) { | |
4254 | return ret; | |
4255 | } else if (ret > 0) { | |
2f2ff0ee | 4256 | *size_ret = 0; |
1a4bcf47 FM |
4257 | } else { |
4258 | struct btrfs_inode_item *item; | |
4259 | ||
4260 | item = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
4261 | struct btrfs_inode_item); | |
4262 | *size_ret = btrfs_inode_size(path->nodes[0], item); | |
4263 | } | |
4264 | ||
4265 | btrfs_release_path(path); | |
4266 | return 0; | |
4267 | } | |
4268 | ||
36283bf7 FM |
4269 | /* |
4270 | * At the moment we always log all xattrs. This is to figure out at log replay | |
4271 | * time which xattrs must have their deletion replayed. If a xattr is missing | |
4272 | * in the log tree and exists in the fs/subvol tree, we delete it. This is | |
4273 | * because if a xattr is deleted, the inode is fsynced and a power failure | |
4274 | * happens, causing the log to be replayed the next time the fs is mounted, | |
4275 | * we want the xattr to not exist anymore (same behaviour as other filesystems | |
4276 | * with a journal, ext3/4, xfs, f2fs, etc). | |
4277 | */ | |
4278 | static int btrfs_log_all_xattrs(struct btrfs_trans_handle *trans, | |
4279 | struct btrfs_root *root, | |
4280 | struct inode *inode, | |
4281 | struct btrfs_path *path, | |
4282 | struct btrfs_path *dst_path) | |
4283 | { | |
4284 | int ret; | |
4285 | struct btrfs_key key; | |
4a0cc7ca | 4286 | const u64 ino = btrfs_ino(BTRFS_I(inode)); |
36283bf7 FM |
4287 | int ins_nr = 0; |
4288 | int start_slot = 0; | |
4289 | ||
4290 | key.objectid = ino; | |
4291 | key.type = BTRFS_XATTR_ITEM_KEY; | |
4292 | key.offset = 0; | |
4293 | ||
4294 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
4295 | if (ret < 0) | |
4296 | return ret; | |
4297 | ||
4298 | while (true) { | |
4299 | int slot = path->slots[0]; | |
4300 | struct extent_buffer *leaf = path->nodes[0]; | |
4301 | int nritems = btrfs_header_nritems(leaf); | |
4302 | ||
4303 | if (slot >= nritems) { | |
4304 | if (ins_nr > 0) { | |
4305 | u64 last_extent = 0; | |
4306 | ||
44d70e19 | 4307 | ret = copy_items(trans, BTRFS_I(inode), dst_path, path, |
36283bf7 FM |
4308 | &last_extent, start_slot, |
4309 | ins_nr, 1, 0); | |
4310 | /* can't be 1, extent items aren't processed */ | |
4311 | ASSERT(ret <= 0); | |
4312 | if (ret < 0) | |
4313 | return ret; | |
4314 | ins_nr = 0; | |
4315 | } | |
4316 | ret = btrfs_next_leaf(root, path); | |
4317 | if (ret < 0) | |
4318 | return ret; | |
4319 | else if (ret > 0) | |
4320 | break; | |
4321 | continue; | |
4322 | } | |
4323 | ||
4324 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
4325 | if (key.objectid != ino || key.type != BTRFS_XATTR_ITEM_KEY) | |
4326 | break; | |
4327 | ||
4328 | if (ins_nr == 0) | |
4329 | start_slot = slot; | |
4330 | ins_nr++; | |
4331 | path->slots[0]++; | |
4332 | cond_resched(); | |
4333 | } | |
4334 | if (ins_nr > 0) { | |
4335 | u64 last_extent = 0; | |
4336 | ||
44d70e19 | 4337 | ret = copy_items(trans, BTRFS_I(inode), dst_path, path, |
36283bf7 FM |
4338 | &last_extent, start_slot, |
4339 | ins_nr, 1, 0); | |
4340 | /* can't be 1, extent items aren't processed */ | |
4341 | ASSERT(ret <= 0); | |
4342 | if (ret < 0) | |
4343 | return ret; | |
4344 | } | |
4345 | ||
4346 | return 0; | |
4347 | } | |
4348 | ||
a89ca6f2 FM |
4349 | /* |
4350 | * If the no holes feature is enabled we need to make sure any hole between the | |
4351 | * last extent and the i_size of our inode is explicitly marked in the log. This | |
4352 | * is to make sure that doing something like: | |
4353 | * | |
4354 | * 1) create file with 128Kb of data | |
4355 | * 2) truncate file to 64Kb | |
4356 | * 3) truncate file to 256Kb | |
4357 | * 4) fsync file | |
4358 | * 5) <crash/power failure> | |
4359 | * 6) mount fs and trigger log replay | |
4360 | * | |
4361 | * Will give us a file with a size of 256Kb, the first 64Kb of data match what | |
4362 | * the file had in its first 64Kb of data at step 1 and the last 192Kb of the | |
4363 | * file correspond to a hole. The presence of explicit holes in a log tree is | |
4364 | * what guarantees that log replay will remove/adjust file extent items in the | |
4365 | * fs/subvol tree. | |
4366 | * | |
4367 | * Here we do not need to care about holes between extents, that is already done | |
4368 | * by copy_items(). We also only need to do this in the full sync path, where we | |
4369 | * lookup for extents from the fs/subvol tree only. In the fast path case, we | |
4370 | * lookup the list of modified extent maps and if any represents a hole, we | |
4371 | * insert a corresponding extent representing a hole in the log tree. | |
4372 | */ | |
4373 | static int btrfs_log_trailing_hole(struct btrfs_trans_handle *trans, | |
4374 | struct btrfs_root *root, | |
4375 | struct inode *inode, | |
4376 | struct btrfs_path *path) | |
4377 | { | |
0b246afa | 4378 | struct btrfs_fs_info *fs_info = root->fs_info; |
a89ca6f2 FM |
4379 | int ret; |
4380 | struct btrfs_key key; | |
4381 | u64 hole_start; | |
4382 | u64 hole_size; | |
4383 | struct extent_buffer *leaf; | |
4384 | struct btrfs_root *log = root->log_root; | |
4a0cc7ca | 4385 | const u64 ino = btrfs_ino(BTRFS_I(inode)); |
a89ca6f2 FM |
4386 | const u64 i_size = i_size_read(inode); |
4387 | ||
0b246afa | 4388 | if (!btrfs_fs_incompat(fs_info, NO_HOLES)) |
a89ca6f2 FM |
4389 | return 0; |
4390 | ||
4391 | key.objectid = ino; | |
4392 | key.type = BTRFS_EXTENT_DATA_KEY; | |
4393 | key.offset = (u64)-1; | |
4394 | ||
4395 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
4396 | ASSERT(ret != 0); | |
4397 | if (ret < 0) | |
4398 | return ret; | |
4399 | ||
4400 | ASSERT(path->slots[0] > 0); | |
4401 | path->slots[0]--; | |
4402 | leaf = path->nodes[0]; | |
4403 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
4404 | ||
4405 | if (key.objectid != ino || key.type != BTRFS_EXTENT_DATA_KEY) { | |
4406 | /* inode does not have any extents */ | |
4407 | hole_start = 0; | |
4408 | hole_size = i_size; | |
4409 | } else { | |
4410 | struct btrfs_file_extent_item *extent; | |
4411 | u64 len; | |
4412 | ||
4413 | /* | |
4414 | * If there's an extent beyond i_size, an explicit hole was | |
4415 | * already inserted by copy_items(). | |
4416 | */ | |
4417 | if (key.offset >= i_size) | |
4418 | return 0; | |
4419 | ||
4420 | extent = btrfs_item_ptr(leaf, path->slots[0], | |
4421 | struct btrfs_file_extent_item); | |
4422 | ||
4423 | if (btrfs_file_extent_type(leaf, extent) == | |
4424 | BTRFS_FILE_EXTENT_INLINE) { | |
4425 | len = btrfs_file_extent_inline_len(leaf, | |
4426 | path->slots[0], | |
4427 | extent); | |
4428 | ASSERT(len == i_size); | |
4429 | return 0; | |
4430 | } | |
4431 | ||
4432 | len = btrfs_file_extent_num_bytes(leaf, extent); | |
4433 | /* Last extent goes beyond i_size, no need to log a hole. */ | |
4434 | if (key.offset + len > i_size) | |
4435 | return 0; | |
4436 | hole_start = key.offset + len; | |
4437 | hole_size = i_size - hole_start; | |
4438 | } | |
4439 | btrfs_release_path(path); | |
4440 | ||
4441 | /* Last extent ends at i_size. */ | |
4442 | if (hole_size == 0) | |
4443 | return 0; | |
4444 | ||
0b246afa | 4445 | hole_size = ALIGN(hole_size, fs_info->sectorsize); |
a89ca6f2 FM |
4446 | ret = btrfs_insert_file_extent(trans, log, ino, hole_start, 0, 0, |
4447 | hole_size, 0, hole_size, 0, 0, 0); | |
4448 | return ret; | |
4449 | } | |
4450 | ||
56f23fdb FM |
4451 | /* |
4452 | * When we are logging a new inode X, check if it doesn't have a reference that | |
4453 | * matches the reference from some other inode Y created in a past transaction | |
4454 | * and that was renamed in the current transaction. If we don't do this, then at | |
4455 | * log replay time we can lose inode Y (and all its files if it's a directory): | |
4456 | * | |
4457 | * mkdir /mnt/x | |
4458 | * echo "hello world" > /mnt/x/foobar | |
4459 | * sync | |
4460 | * mv /mnt/x /mnt/y | |
4461 | * mkdir /mnt/x # or touch /mnt/x | |
4462 | * xfs_io -c fsync /mnt/x | |
4463 | * <power fail> | |
4464 | * mount fs, trigger log replay | |
4465 | * | |
4466 | * After the log replay procedure, we would lose the first directory and all its | |
4467 | * files (file foobar). | |
4468 | * For the case where inode Y is not a directory we simply end up losing it: | |
4469 | * | |
4470 | * echo "123" > /mnt/foo | |
4471 | * sync | |
4472 | * mv /mnt/foo /mnt/bar | |
4473 | * echo "abc" > /mnt/foo | |
4474 | * xfs_io -c fsync /mnt/foo | |
4475 | * <power fail> | |
4476 | * | |
4477 | * We also need this for cases where a snapshot entry is replaced by some other | |
4478 | * entry (file or directory) otherwise we end up with an unreplayable log due to | |
4479 | * attempts to delete the snapshot entry (entry of type BTRFS_ROOT_ITEM_KEY) as | |
4480 | * if it were a regular entry: | |
4481 | * | |
4482 | * mkdir /mnt/x | |
4483 | * btrfs subvolume snapshot /mnt /mnt/x/snap | |
4484 | * btrfs subvolume delete /mnt/x/snap | |
4485 | * rmdir /mnt/x | |
4486 | * mkdir /mnt/x | |
4487 | * fsync /mnt/x or fsync some new file inside it | |
4488 | * <power fail> | |
4489 | * | |
4490 | * The snapshot delete, rmdir of x, mkdir of a new x and the fsync all happen in | |
4491 | * the same transaction. | |
4492 | */ | |
4493 | static int btrfs_check_ref_name_override(struct extent_buffer *eb, | |
4494 | const int slot, | |
4495 | const struct btrfs_key *key, | |
4791c8f1 | 4496 | struct btrfs_inode *inode, |
44f714da | 4497 | u64 *other_ino) |
56f23fdb FM |
4498 | { |
4499 | int ret; | |
4500 | struct btrfs_path *search_path; | |
4501 | char *name = NULL; | |
4502 | u32 name_len = 0; | |
4503 | u32 item_size = btrfs_item_size_nr(eb, slot); | |
4504 | u32 cur_offset = 0; | |
4505 | unsigned long ptr = btrfs_item_ptr_offset(eb, slot); | |
4506 | ||
4507 | search_path = btrfs_alloc_path(); | |
4508 | if (!search_path) | |
4509 | return -ENOMEM; | |
4510 | search_path->search_commit_root = 1; | |
4511 | search_path->skip_locking = 1; | |
4512 | ||
4513 | while (cur_offset < item_size) { | |
4514 | u64 parent; | |
4515 | u32 this_name_len; | |
4516 | u32 this_len; | |
4517 | unsigned long name_ptr; | |
4518 | struct btrfs_dir_item *di; | |
4519 | ||
4520 | if (key->type == BTRFS_INODE_REF_KEY) { | |
4521 | struct btrfs_inode_ref *iref; | |
4522 | ||
4523 | iref = (struct btrfs_inode_ref *)(ptr + cur_offset); | |
4524 | parent = key->offset; | |
4525 | this_name_len = btrfs_inode_ref_name_len(eb, iref); | |
4526 | name_ptr = (unsigned long)(iref + 1); | |
4527 | this_len = sizeof(*iref) + this_name_len; | |
4528 | } else { | |
4529 | struct btrfs_inode_extref *extref; | |
4530 | ||
4531 | extref = (struct btrfs_inode_extref *)(ptr + | |
4532 | cur_offset); | |
4533 | parent = btrfs_inode_extref_parent(eb, extref); | |
4534 | this_name_len = btrfs_inode_extref_name_len(eb, extref); | |
4535 | name_ptr = (unsigned long)&extref->name; | |
4536 | this_len = sizeof(*extref) + this_name_len; | |
4537 | } | |
4538 | ||
4539 | if (this_name_len > name_len) { | |
4540 | char *new_name; | |
4541 | ||
4542 | new_name = krealloc(name, this_name_len, GFP_NOFS); | |
4543 | if (!new_name) { | |
4544 | ret = -ENOMEM; | |
4545 | goto out; | |
4546 | } | |
4547 | name_len = this_name_len; | |
4548 | name = new_name; | |
4549 | } | |
4550 | ||
4551 | read_extent_buffer(eb, name, name_ptr, this_name_len); | |
4791c8f1 NB |
4552 | di = btrfs_lookup_dir_item(NULL, inode->root, search_path, |
4553 | parent, name, this_name_len, 0); | |
56f23fdb | 4554 | if (di && !IS_ERR(di)) { |
44f714da FM |
4555 | struct btrfs_key di_key; |
4556 | ||
4557 | btrfs_dir_item_key_to_cpu(search_path->nodes[0], | |
4558 | di, &di_key); | |
4559 | if (di_key.type == BTRFS_INODE_ITEM_KEY) { | |
4560 | ret = 1; | |
4561 | *other_ino = di_key.objectid; | |
4562 | } else { | |
4563 | ret = -EAGAIN; | |
4564 | } | |
56f23fdb FM |
4565 | goto out; |
4566 | } else if (IS_ERR(di)) { | |
4567 | ret = PTR_ERR(di); | |
4568 | goto out; | |
4569 | } | |
4570 | btrfs_release_path(search_path); | |
4571 | ||
4572 | cur_offset += this_len; | |
4573 | } | |
4574 | ret = 0; | |
4575 | out: | |
4576 | btrfs_free_path(search_path); | |
4577 | kfree(name); | |
4578 | return ret; | |
4579 | } | |
4580 | ||
e02119d5 CM |
4581 | /* log a single inode in the tree log. |
4582 | * At least one parent directory for this inode must exist in the tree | |
4583 | * or be logged already. | |
4584 | * | |
4585 | * Any items from this inode changed by the current transaction are copied | |
4586 | * to the log tree. An extra reference is taken on any extents in this | |
4587 | * file, allowing us to avoid a whole pile of corner cases around logging | |
4588 | * blocks that have been removed from the tree. | |
4589 | * | |
4590 | * See LOG_INODE_ALL and related defines for a description of what inode_only | |
4591 | * does. | |
4592 | * | |
4593 | * This handles both files and directories. | |
4594 | */ | |
12fcfd22 | 4595 | static int btrfs_log_inode(struct btrfs_trans_handle *trans, |
49dae1bc FM |
4596 | struct btrfs_root *root, struct inode *inode, |
4597 | int inode_only, | |
4598 | const loff_t start, | |
8407f553 FM |
4599 | const loff_t end, |
4600 | struct btrfs_log_ctx *ctx) | |
e02119d5 | 4601 | { |
0b246afa | 4602 | struct btrfs_fs_info *fs_info = root->fs_info; |
e02119d5 CM |
4603 | struct btrfs_path *path; |
4604 | struct btrfs_path *dst_path; | |
4605 | struct btrfs_key min_key; | |
4606 | struct btrfs_key max_key; | |
4607 | struct btrfs_root *log = root->log_root; | |
31ff1cd2 | 4608 | struct extent_buffer *src = NULL; |
827463c4 | 4609 | LIST_HEAD(logged_list); |
16e7549f | 4610 | u64 last_extent = 0; |
4a500fd1 | 4611 | int err = 0; |
e02119d5 | 4612 | int ret; |
3a5f1d45 | 4613 | int nritems; |
31ff1cd2 CM |
4614 | int ins_start_slot = 0; |
4615 | int ins_nr; | |
5dc562c5 | 4616 | bool fast_search = false; |
4a0cc7ca | 4617 | u64 ino = btrfs_ino(BTRFS_I(inode)); |
49dae1bc | 4618 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; |
1a4bcf47 | 4619 | u64 logged_isize = 0; |
e4545de5 | 4620 | bool need_log_inode_item = true; |
e02119d5 | 4621 | |
e02119d5 | 4622 | path = btrfs_alloc_path(); |
5df67083 TI |
4623 | if (!path) |
4624 | return -ENOMEM; | |
e02119d5 | 4625 | dst_path = btrfs_alloc_path(); |
5df67083 TI |
4626 | if (!dst_path) { |
4627 | btrfs_free_path(path); | |
4628 | return -ENOMEM; | |
4629 | } | |
e02119d5 | 4630 | |
33345d01 | 4631 | min_key.objectid = ino; |
e02119d5 CM |
4632 | min_key.type = BTRFS_INODE_ITEM_KEY; |
4633 | min_key.offset = 0; | |
4634 | ||
33345d01 | 4635 | max_key.objectid = ino; |
12fcfd22 | 4636 | |
12fcfd22 | 4637 | |
5dc562c5 | 4638 | /* today the code can only do partial logging of directories */ |
5269b67e MX |
4639 | if (S_ISDIR(inode->i_mode) || |
4640 | (!test_bit(BTRFS_INODE_NEEDS_FULL_SYNC, | |
4641 | &BTRFS_I(inode)->runtime_flags) && | |
781feef7 | 4642 | inode_only >= LOG_INODE_EXISTS)) |
e02119d5 CM |
4643 | max_key.type = BTRFS_XATTR_ITEM_KEY; |
4644 | else | |
4645 | max_key.type = (u8)-1; | |
4646 | max_key.offset = (u64)-1; | |
4647 | ||
2c2c452b FM |
4648 | /* |
4649 | * Only run delayed items if we are a dir or a new file. | |
4650 | * Otherwise commit the delayed inode only, which is needed in | |
4651 | * order for the log replay code to mark inodes for link count | |
4652 | * fixup (create temporary BTRFS_TREE_LOG_FIXUP_OBJECTID items). | |
4653 | */ | |
94edf4ae | 4654 | if (S_ISDIR(inode->i_mode) || |
0b246afa | 4655 | BTRFS_I(inode)->generation > fs_info->last_trans_committed) |
5f4b32e9 | 4656 | ret = btrfs_commit_inode_delayed_items(trans, BTRFS_I(inode)); |
2c2c452b | 4657 | else |
aa79021f | 4658 | ret = btrfs_commit_inode_delayed_inode(BTRFS_I(inode)); |
2c2c452b FM |
4659 | |
4660 | if (ret) { | |
4661 | btrfs_free_path(path); | |
4662 | btrfs_free_path(dst_path); | |
4663 | return ret; | |
16cdcec7 MX |
4664 | } |
4665 | ||
781feef7 LB |
4666 | if (inode_only == LOG_OTHER_INODE) { |
4667 | inode_only = LOG_INODE_EXISTS; | |
4668 | mutex_lock_nested(&BTRFS_I(inode)->log_mutex, | |
4669 | SINGLE_DEPTH_NESTING); | |
4670 | } else { | |
4671 | mutex_lock(&BTRFS_I(inode)->log_mutex); | |
4672 | } | |
e02119d5 CM |
4673 | |
4674 | /* | |
4675 | * a brute force approach to making sure we get the most uptodate | |
4676 | * copies of everything. | |
4677 | */ | |
4678 | if (S_ISDIR(inode->i_mode)) { | |
4679 | int max_key_type = BTRFS_DIR_LOG_INDEX_KEY; | |
4680 | ||
4f764e51 FM |
4681 | if (inode_only == LOG_INODE_EXISTS) |
4682 | max_key_type = BTRFS_XATTR_ITEM_KEY; | |
33345d01 | 4683 | ret = drop_objectid_items(trans, log, path, ino, max_key_type); |
e02119d5 | 4684 | } else { |
1a4bcf47 FM |
4685 | if (inode_only == LOG_INODE_EXISTS) { |
4686 | /* | |
4687 | * Make sure the new inode item we write to the log has | |
4688 | * the same isize as the current one (if it exists). | |
4689 | * This is necessary to prevent data loss after log | |
4690 | * replay, and also to prevent doing a wrong expanding | |
4691 | * truncate - for e.g. create file, write 4K into offset | |
4692 | * 0, fsync, write 4K into offset 4096, add hard link, | |
4693 | * fsync some other file (to sync log), power fail - if | |
4694 | * we use the inode's current i_size, after log replay | |
4695 | * we get a 8Kb file, with the last 4Kb extent as a hole | |
4696 | * (zeroes), as if an expanding truncate happened, | |
4697 | * instead of getting a file of 4Kb only. | |
4698 | */ | |
481b01c0 | 4699 | err = logged_inode_size(log, BTRFS_I(inode), path, |
1a4bcf47 FM |
4700 | &logged_isize); |
4701 | if (err) | |
4702 | goto out_unlock; | |
4703 | } | |
a742994a FM |
4704 | if (test_bit(BTRFS_INODE_NEEDS_FULL_SYNC, |
4705 | &BTRFS_I(inode)->runtime_flags)) { | |
4706 | if (inode_only == LOG_INODE_EXISTS) { | |
4f764e51 | 4707 | max_key.type = BTRFS_XATTR_ITEM_KEY; |
a742994a FM |
4708 | ret = drop_objectid_items(trans, log, path, ino, |
4709 | max_key.type); | |
4710 | } else { | |
4711 | clear_bit(BTRFS_INODE_NEEDS_FULL_SYNC, | |
4712 | &BTRFS_I(inode)->runtime_flags); | |
4713 | clear_bit(BTRFS_INODE_COPY_EVERYTHING, | |
4714 | &BTRFS_I(inode)->runtime_flags); | |
28ed1345 CM |
4715 | while(1) { |
4716 | ret = btrfs_truncate_inode_items(trans, | |
4717 | log, inode, 0, 0); | |
4718 | if (ret != -EAGAIN) | |
4719 | break; | |
4720 | } | |
a742994a | 4721 | } |
4f764e51 FM |
4722 | } else if (test_and_clear_bit(BTRFS_INODE_COPY_EVERYTHING, |
4723 | &BTRFS_I(inode)->runtime_flags) || | |
6cfab851 | 4724 | inode_only == LOG_INODE_EXISTS) { |
4f764e51 | 4725 | if (inode_only == LOG_INODE_ALL) |
183f37fa | 4726 | fast_search = true; |
4f764e51 | 4727 | max_key.type = BTRFS_XATTR_ITEM_KEY; |
5dc562c5 | 4728 | ret = drop_objectid_items(trans, log, path, ino, |
e9976151 | 4729 | max_key.type); |
a95249b3 JB |
4730 | } else { |
4731 | if (inode_only == LOG_INODE_ALL) | |
4732 | fast_search = true; | |
a95249b3 | 4733 | goto log_extents; |
5dc562c5 | 4734 | } |
a95249b3 | 4735 | |
e02119d5 | 4736 | } |
4a500fd1 YZ |
4737 | if (ret) { |
4738 | err = ret; | |
4739 | goto out_unlock; | |
4740 | } | |
e02119d5 | 4741 | |
d397712b | 4742 | while (1) { |
31ff1cd2 | 4743 | ins_nr = 0; |
6174d3cb | 4744 | ret = btrfs_search_forward(root, &min_key, |
de78b51a | 4745 | path, trans->transid); |
fb770ae4 LB |
4746 | if (ret < 0) { |
4747 | err = ret; | |
4748 | goto out_unlock; | |
4749 | } | |
e02119d5 CM |
4750 | if (ret != 0) |
4751 | break; | |
3a5f1d45 | 4752 | again: |
31ff1cd2 | 4753 | /* note, ins_nr might be > 0 here, cleanup outside the loop */ |
33345d01 | 4754 | if (min_key.objectid != ino) |
e02119d5 CM |
4755 | break; |
4756 | if (min_key.type > max_key.type) | |
4757 | break; | |
31ff1cd2 | 4758 | |
e4545de5 FM |
4759 | if (min_key.type == BTRFS_INODE_ITEM_KEY) |
4760 | need_log_inode_item = false; | |
4761 | ||
56f23fdb FM |
4762 | if ((min_key.type == BTRFS_INODE_REF_KEY || |
4763 | min_key.type == BTRFS_INODE_EXTREF_KEY) && | |
4764 | BTRFS_I(inode)->generation == trans->transid) { | |
44f714da FM |
4765 | u64 other_ino = 0; |
4766 | ||
56f23fdb FM |
4767 | ret = btrfs_check_ref_name_override(path->nodes[0], |
4768 | path->slots[0], | |
4791c8f1 | 4769 | &min_key, BTRFS_I(inode), |
44f714da | 4770 | &other_ino); |
56f23fdb FM |
4771 | if (ret < 0) { |
4772 | err = ret; | |
4773 | goto out_unlock; | |
28a23593 | 4774 | } else if (ret > 0 && ctx && |
4a0cc7ca | 4775 | other_ino != btrfs_ino(BTRFS_I(ctx->inode))) { |
44f714da FM |
4776 | struct btrfs_key inode_key; |
4777 | struct inode *other_inode; | |
4778 | ||
4779 | if (ins_nr > 0) { | |
4780 | ins_nr++; | |
4781 | } else { | |
4782 | ins_nr = 1; | |
4783 | ins_start_slot = path->slots[0]; | |
4784 | } | |
44d70e19 | 4785 | ret = copy_items(trans, BTRFS_I(inode), dst_path, path, |
44f714da FM |
4786 | &last_extent, ins_start_slot, |
4787 | ins_nr, inode_only, | |
4788 | logged_isize); | |
4789 | if (ret < 0) { | |
4790 | err = ret; | |
4791 | goto out_unlock; | |
4792 | } | |
4793 | ins_nr = 0; | |
4794 | btrfs_release_path(path); | |
4795 | inode_key.objectid = other_ino; | |
4796 | inode_key.type = BTRFS_INODE_ITEM_KEY; | |
4797 | inode_key.offset = 0; | |
0b246afa | 4798 | other_inode = btrfs_iget(fs_info->sb, |
44f714da FM |
4799 | &inode_key, root, |
4800 | NULL); | |
4801 | /* | |
4802 | * If the other inode that had a conflicting dir | |
4803 | * entry was deleted in the current transaction, | |
4804 | * we don't need to do more work nor fallback to | |
4805 | * a transaction commit. | |
4806 | */ | |
4807 | if (IS_ERR(other_inode) && | |
4808 | PTR_ERR(other_inode) == -ENOENT) { | |
4809 | goto next_key; | |
4810 | } else if (IS_ERR(other_inode)) { | |
4811 | err = PTR_ERR(other_inode); | |
4812 | goto out_unlock; | |
4813 | } | |
4814 | /* | |
4815 | * We are safe logging the other inode without | |
4816 | * acquiring its i_mutex as long as we log with | |
4817 | * the LOG_INODE_EXISTS mode. We're safe against | |
4818 | * concurrent renames of the other inode as well | |
4819 | * because during a rename we pin the log and | |
4820 | * update the log with the new name before we | |
4821 | * unpin it. | |
4822 | */ | |
4823 | err = btrfs_log_inode(trans, root, other_inode, | |
781feef7 | 4824 | LOG_OTHER_INODE, |
44f714da FM |
4825 | 0, LLONG_MAX, ctx); |
4826 | iput(other_inode); | |
4827 | if (err) | |
4828 | goto out_unlock; | |
4829 | else | |
4830 | goto next_key; | |
56f23fdb FM |
4831 | } |
4832 | } | |
4833 | ||
36283bf7 FM |
4834 | /* Skip xattrs, we log them later with btrfs_log_all_xattrs() */ |
4835 | if (min_key.type == BTRFS_XATTR_ITEM_KEY) { | |
4836 | if (ins_nr == 0) | |
4837 | goto next_slot; | |
44d70e19 | 4838 | ret = copy_items(trans, BTRFS_I(inode), dst_path, path, |
36283bf7 FM |
4839 | &last_extent, ins_start_slot, |
4840 | ins_nr, inode_only, logged_isize); | |
4841 | if (ret < 0) { | |
4842 | err = ret; | |
4843 | goto out_unlock; | |
4844 | } | |
4845 | ins_nr = 0; | |
4846 | if (ret) { | |
4847 | btrfs_release_path(path); | |
4848 | continue; | |
4849 | } | |
4850 | goto next_slot; | |
4851 | } | |
4852 | ||
e02119d5 | 4853 | src = path->nodes[0]; |
31ff1cd2 CM |
4854 | if (ins_nr && ins_start_slot + ins_nr == path->slots[0]) { |
4855 | ins_nr++; | |
4856 | goto next_slot; | |
4857 | } else if (!ins_nr) { | |
4858 | ins_start_slot = path->slots[0]; | |
4859 | ins_nr = 1; | |
4860 | goto next_slot; | |
e02119d5 CM |
4861 | } |
4862 | ||
44d70e19 | 4863 | ret = copy_items(trans, BTRFS_I(inode), dst_path, path, &last_extent, |
1a4bcf47 FM |
4864 | ins_start_slot, ins_nr, inode_only, |
4865 | logged_isize); | |
16e7549f | 4866 | if (ret < 0) { |
4a500fd1 YZ |
4867 | err = ret; |
4868 | goto out_unlock; | |
a71db86e RV |
4869 | } |
4870 | if (ret) { | |
16e7549f JB |
4871 | ins_nr = 0; |
4872 | btrfs_release_path(path); | |
4873 | continue; | |
4a500fd1 | 4874 | } |
31ff1cd2 CM |
4875 | ins_nr = 1; |
4876 | ins_start_slot = path->slots[0]; | |
4877 | next_slot: | |
e02119d5 | 4878 | |
3a5f1d45 CM |
4879 | nritems = btrfs_header_nritems(path->nodes[0]); |
4880 | path->slots[0]++; | |
4881 | if (path->slots[0] < nritems) { | |
4882 | btrfs_item_key_to_cpu(path->nodes[0], &min_key, | |
4883 | path->slots[0]); | |
4884 | goto again; | |
4885 | } | |
31ff1cd2 | 4886 | if (ins_nr) { |
44d70e19 | 4887 | ret = copy_items(trans, BTRFS_I(inode), dst_path, path, |
16e7549f | 4888 | &last_extent, ins_start_slot, |
1a4bcf47 | 4889 | ins_nr, inode_only, logged_isize); |
16e7549f | 4890 | if (ret < 0) { |
4a500fd1 YZ |
4891 | err = ret; |
4892 | goto out_unlock; | |
4893 | } | |
16e7549f | 4894 | ret = 0; |
31ff1cd2 CM |
4895 | ins_nr = 0; |
4896 | } | |
b3b4aa74 | 4897 | btrfs_release_path(path); |
44f714da | 4898 | next_key: |
3d41d702 | 4899 | if (min_key.offset < (u64)-1) { |
e02119d5 | 4900 | min_key.offset++; |
3d41d702 | 4901 | } else if (min_key.type < max_key.type) { |
e02119d5 | 4902 | min_key.type++; |
3d41d702 FDBM |
4903 | min_key.offset = 0; |
4904 | } else { | |
e02119d5 | 4905 | break; |
3d41d702 | 4906 | } |
e02119d5 | 4907 | } |
31ff1cd2 | 4908 | if (ins_nr) { |
44d70e19 | 4909 | ret = copy_items(trans, BTRFS_I(inode), dst_path, path, &last_extent, |
1a4bcf47 FM |
4910 | ins_start_slot, ins_nr, inode_only, |
4911 | logged_isize); | |
16e7549f | 4912 | if (ret < 0) { |
4a500fd1 YZ |
4913 | err = ret; |
4914 | goto out_unlock; | |
4915 | } | |
16e7549f | 4916 | ret = 0; |
31ff1cd2 CM |
4917 | ins_nr = 0; |
4918 | } | |
5dc562c5 | 4919 | |
36283bf7 FM |
4920 | btrfs_release_path(path); |
4921 | btrfs_release_path(dst_path); | |
4922 | err = btrfs_log_all_xattrs(trans, root, inode, path, dst_path); | |
4923 | if (err) | |
4924 | goto out_unlock; | |
a89ca6f2 FM |
4925 | if (max_key.type >= BTRFS_EXTENT_DATA_KEY && !fast_search) { |
4926 | btrfs_release_path(path); | |
4927 | btrfs_release_path(dst_path); | |
4928 | err = btrfs_log_trailing_hole(trans, root, inode, path); | |
4929 | if (err) | |
4930 | goto out_unlock; | |
4931 | } | |
a95249b3 | 4932 | log_extents: |
f3b15ccd JB |
4933 | btrfs_release_path(path); |
4934 | btrfs_release_path(dst_path); | |
e4545de5 FM |
4935 | if (need_log_inode_item) { |
4936 | err = log_inode_item(trans, log, dst_path, inode); | |
4937 | if (err) | |
4938 | goto out_unlock; | |
4939 | } | |
5dc562c5 | 4940 | if (fast_search) { |
827463c4 | 4941 | ret = btrfs_log_changed_extents(trans, root, inode, dst_path, |
de0ee0ed | 4942 | &logged_list, ctx, start, end); |
5dc562c5 JB |
4943 | if (ret) { |
4944 | err = ret; | |
4945 | goto out_unlock; | |
4946 | } | |
d006a048 | 4947 | } else if (inode_only == LOG_INODE_ALL) { |
06d3d22b LB |
4948 | struct extent_map *em, *n; |
4949 | ||
49dae1bc FM |
4950 | write_lock(&em_tree->lock); |
4951 | /* | |
4952 | * We can't just remove every em if we're called for a ranged | |
4953 | * fsync - that is, one that doesn't cover the whole possible | |
4954 | * file range (0 to LLONG_MAX). This is because we can have | |
4955 | * em's that fall outside the range we're logging and therefore | |
4956 | * their ordered operations haven't completed yet | |
4957 | * (btrfs_finish_ordered_io() not invoked yet). This means we | |
4958 | * didn't get their respective file extent item in the fs/subvol | |
4959 | * tree yet, and need to let the next fast fsync (one which | |
4960 | * consults the list of modified extent maps) find the em so | |
4961 | * that it logs a matching file extent item and waits for the | |
4962 | * respective ordered operation to complete (if it's still | |
4963 | * running). | |
4964 | * | |
4965 | * Removing every em outside the range we're logging would make | |
4966 | * the next fast fsync not log their matching file extent items, | |
4967 | * therefore making us lose data after a log replay. | |
4968 | */ | |
4969 | list_for_each_entry_safe(em, n, &em_tree->modified_extents, | |
4970 | list) { | |
4971 | const u64 mod_end = em->mod_start + em->mod_len - 1; | |
4972 | ||
4973 | if (em->mod_start >= start && mod_end <= end) | |
4974 | list_del_init(&em->list); | |
4975 | } | |
4976 | write_unlock(&em_tree->lock); | |
5dc562c5 JB |
4977 | } |
4978 | ||
9623f9a3 | 4979 | if (inode_only == LOG_INODE_ALL && S_ISDIR(inode->i_mode)) { |
2f2ff0ee FM |
4980 | ret = log_directory_changes(trans, root, inode, path, dst_path, |
4981 | ctx); | |
4a500fd1 YZ |
4982 | if (ret) { |
4983 | err = ret; | |
4984 | goto out_unlock; | |
4985 | } | |
e02119d5 | 4986 | } |
49dae1bc | 4987 | |
2f2ff0ee | 4988 | spin_lock(&BTRFS_I(inode)->lock); |
125c4cf9 FM |
4989 | BTRFS_I(inode)->logged_trans = trans->transid; |
4990 | BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->last_sub_trans; | |
2f2ff0ee | 4991 | spin_unlock(&BTRFS_I(inode)->lock); |
4a500fd1 | 4992 | out_unlock: |
827463c4 MX |
4993 | if (unlikely(err)) |
4994 | btrfs_put_logged_extents(&logged_list); | |
4995 | else | |
4996 | btrfs_submit_logged_extents(&logged_list, log); | |
e02119d5 CM |
4997 | mutex_unlock(&BTRFS_I(inode)->log_mutex); |
4998 | ||
4999 | btrfs_free_path(path); | |
5000 | btrfs_free_path(dst_path); | |
4a500fd1 | 5001 | return err; |
e02119d5 CM |
5002 | } |
5003 | ||
2be63d5c FM |
5004 | /* |
5005 | * Check if we must fallback to a transaction commit when logging an inode. | |
5006 | * This must be called after logging the inode and is used only in the context | |
5007 | * when fsyncing an inode requires the need to log some other inode - in which | |
5008 | * case we can't lock the i_mutex of each other inode we need to log as that | |
5009 | * can lead to deadlocks with concurrent fsync against other inodes (as we can | |
5010 | * log inodes up or down in the hierarchy) or rename operations for example. So | |
5011 | * we take the log_mutex of the inode after we have logged it and then check for | |
5012 | * its last_unlink_trans value - this is safe because any task setting | |
5013 | * last_unlink_trans must take the log_mutex and it must do this before it does | |
5014 | * the actual unlink operation, so if we do this check before a concurrent task | |
5015 | * sets last_unlink_trans it means we've logged a consistent version/state of | |
5016 | * all the inode items, otherwise we are not sure and must do a transaction | |
01327610 | 5017 | * commit (the concurrent task might have only updated last_unlink_trans before |
2be63d5c FM |
5018 | * we logged the inode or it might have also done the unlink). |
5019 | */ | |
5020 | static bool btrfs_must_commit_transaction(struct btrfs_trans_handle *trans, | |
ab1717b2 | 5021 | struct btrfs_inode *inode) |
2be63d5c | 5022 | { |
ab1717b2 | 5023 | struct btrfs_fs_info *fs_info = inode->root->fs_info; |
2be63d5c FM |
5024 | bool ret = false; |
5025 | ||
ab1717b2 NB |
5026 | mutex_lock(&inode->log_mutex); |
5027 | if (inode->last_unlink_trans > fs_info->last_trans_committed) { | |
2be63d5c FM |
5028 | /* |
5029 | * Make sure any commits to the log are forced to be full | |
5030 | * commits. | |
5031 | */ | |
5032 | btrfs_set_log_full_commit(fs_info, trans); | |
5033 | ret = true; | |
5034 | } | |
ab1717b2 | 5035 | mutex_unlock(&inode->log_mutex); |
2be63d5c FM |
5036 | |
5037 | return ret; | |
5038 | } | |
5039 | ||
12fcfd22 CM |
5040 | /* |
5041 | * follow the dentry parent pointers up the chain and see if any | |
5042 | * of the directories in it require a full commit before they can | |
5043 | * be logged. Returns zero if nothing special needs to be done or 1 if | |
5044 | * a full commit is required. | |
5045 | */ | |
5046 | static noinline int check_parent_dirs_for_sync(struct btrfs_trans_handle *trans, | |
5047 | struct inode *inode, | |
5048 | struct dentry *parent, | |
5049 | struct super_block *sb, | |
5050 | u64 last_committed) | |
e02119d5 | 5051 | { |
12fcfd22 | 5052 | int ret = 0; |
6a912213 | 5053 | struct dentry *old_parent = NULL; |
de2b530b | 5054 | struct inode *orig_inode = inode; |
e02119d5 | 5055 | |
af4176b4 CM |
5056 | /* |
5057 | * for regular files, if its inode is already on disk, we don't | |
5058 | * have to worry about the parents at all. This is because | |
5059 | * we can use the last_unlink_trans field to record renames | |
5060 | * and other fun in this file. | |
5061 | */ | |
5062 | if (S_ISREG(inode->i_mode) && | |
5063 | BTRFS_I(inode)->generation <= last_committed && | |
5064 | BTRFS_I(inode)->last_unlink_trans <= last_committed) | |
5065 | goto out; | |
5066 | ||
12fcfd22 | 5067 | if (!S_ISDIR(inode->i_mode)) { |
fc64005c | 5068 | if (!parent || d_really_is_negative(parent) || sb != parent->d_sb) |
12fcfd22 | 5069 | goto out; |
2b0143b5 | 5070 | inode = d_inode(parent); |
12fcfd22 CM |
5071 | } |
5072 | ||
5073 | while (1) { | |
de2b530b JB |
5074 | /* |
5075 | * If we are logging a directory then we start with our inode, | |
01327610 | 5076 | * not our parent's inode, so we need to skip setting the |
de2b530b JB |
5077 | * logged_trans so that further down in the log code we don't |
5078 | * think this inode has already been logged. | |
5079 | */ | |
5080 | if (inode != orig_inode) | |
5081 | BTRFS_I(inode)->logged_trans = trans->transid; | |
12fcfd22 CM |
5082 | smp_mb(); |
5083 | ||
ab1717b2 | 5084 | if (btrfs_must_commit_transaction(trans, BTRFS_I(inode))) { |
12fcfd22 CM |
5085 | ret = 1; |
5086 | break; | |
5087 | } | |
5088 | ||
fc64005c | 5089 | if (!parent || d_really_is_negative(parent) || sb != parent->d_sb) |
12fcfd22 CM |
5090 | break; |
5091 | ||
44f714da FM |
5092 | if (IS_ROOT(parent)) { |
5093 | inode = d_inode(parent); | |
ab1717b2 | 5094 | if (btrfs_must_commit_transaction(trans, BTRFS_I(inode))) |
44f714da | 5095 | ret = 1; |
12fcfd22 | 5096 | break; |
44f714da | 5097 | } |
12fcfd22 | 5098 | |
6a912213 JB |
5099 | parent = dget_parent(parent); |
5100 | dput(old_parent); | |
5101 | old_parent = parent; | |
2b0143b5 | 5102 | inode = d_inode(parent); |
12fcfd22 CM |
5103 | |
5104 | } | |
6a912213 | 5105 | dput(old_parent); |
12fcfd22 | 5106 | out: |
e02119d5 CM |
5107 | return ret; |
5108 | } | |
5109 | ||
2f2ff0ee FM |
5110 | struct btrfs_dir_list { |
5111 | u64 ino; | |
5112 | struct list_head list; | |
5113 | }; | |
5114 | ||
5115 | /* | |
5116 | * Log the inodes of the new dentries of a directory. See log_dir_items() for | |
5117 | * details about the why it is needed. | |
5118 | * This is a recursive operation - if an existing dentry corresponds to a | |
5119 | * directory, that directory's new entries are logged too (same behaviour as | |
5120 | * ext3/4, xfs, f2fs, reiserfs, nilfs2). Note that when logging the inodes | |
5121 | * the dentries point to we do not lock their i_mutex, otherwise lockdep | |
5122 | * complains about the following circular lock dependency / possible deadlock: | |
5123 | * | |
5124 | * CPU0 CPU1 | |
5125 | * ---- ---- | |
5126 | * lock(&type->i_mutex_dir_key#3/2); | |
5127 | * lock(sb_internal#2); | |
5128 | * lock(&type->i_mutex_dir_key#3/2); | |
5129 | * lock(&sb->s_type->i_mutex_key#14); | |
5130 | * | |
5131 | * Where sb_internal is the lock (a counter that works as a lock) acquired by | |
5132 | * sb_start_intwrite() in btrfs_start_transaction(). | |
5133 | * Not locking i_mutex of the inodes is still safe because: | |
5134 | * | |
5135 | * 1) For regular files we log with a mode of LOG_INODE_EXISTS. It's possible | |
5136 | * that while logging the inode new references (names) are added or removed | |
5137 | * from the inode, leaving the logged inode item with a link count that does | |
5138 | * not match the number of logged inode reference items. This is fine because | |
5139 | * at log replay time we compute the real number of links and correct the | |
5140 | * link count in the inode item (see replay_one_buffer() and | |
5141 | * link_to_fixup_dir()); | |
5142 | * | |
5143 | * 2) For directories we log with a mode of LOG_INODE_ALL. It's possible that | |
5144 | * while logging the inode's items new items with keys BTRFS_DIR_ITEM_KEY and | |
5145 | * BTRFS_DIR_INDEX_KEY are added to fs/subvol tree and the logged inode item | |
5146 | * has a size that doesn't match the sum of the lengths of all the logged | |
5147 | * names. This does not result in a problem because if a dir_item key is | |
5148 | * logged but its matching dir_index key is not logged, at log replay time we | |
5149 | * don't use it to replay the respective name (see replay_one_name()). On the | |
5150 | * other hand if only the dir_index key ends up being logged, the respective | |
5151 | * name is added to the fs/subvol tree with both the dir_item and dir_index | |
5152 | * keys created (see replay_one_name()). | |
5153 | * The directory's inode item with a wrong i_size is not a problem as well, | |
5154 | * since we don't use it at log replay time to set the i_size in the inode | |
5155 | * item of the fs/subvol tree (see overwrite_item()). | |
5156 | */ | |
5157 | static int log_new_dir_dentries(struct btrfs_trans_handle *trans, | |
5158 | struct btrfs_root *root, | |
5159 | struct inode *start_inode, | |
5160 | struct btrfs_log_ctx *ctx) | |
5161 | { | |
0b246afa | 5162 | struct btrfs_fs_info *fs_info = root->fs_info; |
2f2ff0ee FM |
5163 | struct btrfs_root *log = root->log_root; |
5164 | struct btrfs_path *path; | |
5165 | LIST_HEAD(dir_list); | |
5166 | struct btrfs_dir_list *dir_elem; | |
5167 | int ret = 0; | |
5168 | ||
5169 | path = btrfs_alloc_path(); | |
5170 | if (!path) | |
5171 | return -ENOMEM; | |
5172 | ||
5173 | dir_elem = kmalloc(sizeof(*dir_elem), GFP_NOFS); | |
5174 | if (!dir_elem) { | |
5175 | btrfs_free_path(path); | |
5176 | return -ENOMEM; | |
5177 | } | |
4a0cc7ca | 5178 | dir_elem->ino = btrfs_ino(BTRFS_I(start_inode)); |
2f2ff0ee FM |
5179 | list_add_tail(&dir_elem->list, &dir_list); |
5180 | ||
5181 | while (!list_empty(&dir_list)) { | |
5182 | struct extent_buffer *leaf; | |
5183 | struct btrfs_key min_key; | |
5184 | int nritems; | |
5185 | int i; | |
5186 | ||
5187 | dir_elem = list_first_entry(&dir_list, struct btrfs_dir_list, | |
5188 | list); | |
5189 | if (ret) | |
5190 | goto next_dir_inode; | |
5191 | ||
5192 | min_key.objectid = dir_elem->ino; | |
5193 | min_key.type = BTRFS_DIR_ITEM_KEY; | |
5194 | min_key.offset = 0; | |
5195 | again: | |
5196 | btrfs_release_path(path); | |
5197 | ret = btrfs_search_forward(log, &min_key, path, trans->transid); | |
5198 | if (ret < 0) { | |
5199 | goto next_dir_inode; | |
5200 | } else if (ret > 0) { | |
5201 | ret = 0; | |
5202 | goto next_dir_inode; | |
5203 | } | |
5204 | ||
5205 | process_leaf: | |
5206 | leaf = path->nodes[0]; | |
5207 | nritems = btrfs_header_nritems(leaf); | |
5208 | for (i = path->slots[0]; i < nritems; i++) { | |
5209 | struct btrfs_dir_item *di; | |
5210 | struct btrfs_key di_key; | |
5211 | struct inode *di_inode; | |
5212 | struct btrfs_dir_list *new_dir_elem; | |
5213 | int log_mode = LOG_INODE_EXISTS; | |
5214 | int type; | |
5215 | ||
5216 | btrfs_item_key_to_cpu(leaf, &min_key, i); | |
5217 | if (min_key.objectid != dir_elem->ino || | |
5218 | min_key.type != BTRFS_DIR_ITEM_KEY) | |
5219 | goto next_dir_inode; | |
5220 | ||
5221 | di = btrfs_item_ptr(leaf, i, struct btrfs_dir_item); | |
5222 | type = btrfs_dir_type(leaf, di); | |
5223 | if (btrfs_dir_transid(leaf, di) < trans->transid && | |
5224 | type != BTRFS_FT_DIR) | |
5225 | continue; | |
5226 | btrfs_dir_item_key_to_cpu(leaf, di, &di_key); | |
5227 | if (di_key.type == BTRFS_ROOT_ITEM_KEY) | |
5228 | continue; | |
5229 | ||
ec125cfb | 5230 | btrfs_release_path(path); |
0b246afa | 5231 | di_inode = btrfs_iget(fs_info->sb, &di_key, root, NULL); |
2f2ff0ee FM |
5232 | if (IS_ERR(di_inode)) { |
5233 | ret = PTR_ERR(di_inode); | |
5234 | goto next_dir_inode; | |
5235 | } | |
5236 | ||
0f8939b8 | 5237 | if (btrfs_inode_in_log(BTRFS_I(di_inode), trans->transid)) { |
2f2ff0ee | 5238 | iput(di_inode); |
ec125cfb | 5239 | break; |
2f2ff0ee FM |
5240 | } |
5241 | ||
5242 | ctx->log_new_dentries = false; | |
3f9749f6 | 5243 | if (type == BTRFS_FT_DIR || type == BTRFS_FT_SYMLINK) |
2f2ff0ee | 5244 | log_mode = LOG_INODE_ALL; |
2f2ff0ee FM |
5245 | ret = btrfs_log_inode(trans, root, di_inode, |
5246 | log_mode, 0, LLONG_MAX, ctx); | |
2be63d5c | 5247 | if (!ret && |
ab1717b2 | 5248 | btrfs_must_commit_transaction(trans, BTRFS_I(di_inode))) |
2be63d5c | 5249 | ret = 1; |
2f2ff0ee FM |
5250 | iput(di_inode); |
5251 | if (ret) | |
5252 | goto next_dir_inode; | |
5253 | if (ctx->log_new_dentries) { | |
5254 | new_dir_elem = kmalloc(sizeof(*new_dir_elem), | |
5255 | GFP_NOFS); | |
5256 | if (!new_dir_elem) { | |
5257 | ret = -ENOMEM; | |
5258 | goto next_dir_inode; | |
5259 | } | |
5260 | new_dir_elem->ino = di_key.objectid; | |
5261 | list_add_tail(&new_dir_elem->list, &dir_list); | |
5262 | } | |
5263 | break; | |
5264 | } | |
5265 | if (i == nritems) { | |
5266 | ret = btrfs_next_leaf(log, path); | |
5267 | if (ret < 0) { | |
5268 | goto next_dir_inode; | |
5269 | } else if (ret > 0) { | |
5270 | ret = 0; | |
5271 | goto next_dir_inode; | |
5272 | } | |
5273 | goto process_leaf; | |
5274 | } | |
5275 | if (min_key.offset < (u64)-1) { | |
5276 | min_key.offset++; | |
5277 | goto again; | |
5278 | } | |
5279 | next_dir_inode: | |
5280 | list_del(&dir_elem->list); | |
5281 | kfree(dir_elem); | |
5282 | } | |
5283 | ||
5284 | btrfs_free_path(path); | |
5285 | return ret; | |
5286 | } | |
5287 | ||
18aa0922 FM |
5288 | static int btrfs_log_all_parents(struct btrfs_trans_handle *trans, |
5289 | struct inode *inode, | |
5290 | struct btrfs_log_ctx *ctx) | |
5291 | { | |
0b246afa | 5292 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
18aa0922 FM |
5293 | int ret; |
5294 | struct btrfs_path *path; | |
5295 | struct btrfs_key key; | |
5296 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
4a0cc7ca | 5297 | const u64 ino = btrfs_ino(BTRFS_I(inode)); |
18aa0922 FM |
5298 | |
5299 | path = btrfs_alloc_path(); | |
5300 | if (!path) | |
5301 | return -ENOMEM; | |
5302 | path->skip_locking = 1; | |
5303 | path->search_commit_root = 1; | |
5304 | ||
5305 | key.objectid = ino; | |
5306 | key.type = BTRFS_INODE_REF_KEY; | |
5307 | key.offset = 0; | |
5308 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
5309 | if (ret < 0) | |
5310 | goto out; | |
5311 | ||
5312 | while (true) { | |
5313 | struct extent_buffer *leaf = path->nodes[0]; | |
5314 | int slot = path->slots[0]; | |
5315 | u32 cur_offset = 0; | |
5316 | u32 item_size; | |
5317 | unsigned long ptr; | |
5318 | ||
5319 | if (slot >= btrfs_header_nritems(leaf)) { | |
5320 | ret = btrfs_next_leaf(root, path); | |
5321 | if (ret < 0) | |
5322 | goto out; | |
5323 | else if (ret > 0) | |
5324 | break; | |
5325 | continue; | |
5326 | } | |
5327 | ||
5328 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
5329 | /* BTRFS_INODE_EXTREF_KEY is BTRFS_INODE_REF_KEY + 1 */ | |
5330 | if (key.objectid != ino || key.type > BTRFS_INODE_EXTREF_KEY) | |
5331 | break; | |
5332 | ||
5333 | item_size = btrfs_item_size_nr(leaf, slot); | |
5334 | ptr = btrfs_item_ptr_offset(leaf, slot); | |
5335 | while (cur_offset < item_size) { | |
5336 | struct btrfs_key inode_key; | |
5337 | struct inode *dir_inode; | |
5338 | ||
5339 | inode_key.type = BTRFS_INODE_ITEM_KEY; | |
5340 | inode_key.offset = 0; | |
5341 | ||
5342 | if (key.type == BTRFS_INODE_EXTREF_KEY) { | |
5343 | struct btrfs_inode_extref *extref; | |
5344 | ||
5345 | extref = (struct btrfs_inode_extref *) | |
5346 | (ptr + cur_offset); | |
5347 | inode_key.objectid = btrfs_inode_extref_parent( | |
5348 | leaf, extref); | |
5349 | cur_offset += sizeof(*extref); | |
5350 | cur_offset += btrfs_inode_extref_name_len(leaf, | |
5351 | extref); | |
5352 | } else { | |
5353 | inode_key.objectid = key.offset; | |
5354 | cur_offset = item_size; | |
5355 | } | |
5356 | ||
0b246afa | 5357 | dir_inode = btrfs_iget(fs_info->sb, &inode_key, |
18aa0922 FM |
5358 | root, NULL); |
5359 | /* If parent inode was deleted, skip it. */ | |
5360 | if (IS_ERR(dir_inode)) | |
5361 | continue; | |
5362 | ||
657ed1aa FM |
5363 | if (ctx) |
5364 | ctx->log_new_dentries = false; | |
18aa0922 FM |
5365 | ret = btrfs_log_inode(trans, root, dir_inode, |
5366 | LOG_INODE_ALL, 0, LLONG_MAX, ctx); | |
2be63d5c | 5367 | if (!ret && |
ab1717b2 | 5368 | btrfs_must_commit_transaction(trans, BTRFS_I(dir_inode))) |
2be63d5c | 5369 | ret = 1; |
657ed1aa FM |
5370 | if (!ret && ctx && ctx->log_new_dentries) |
5371 | ret = log_new_dir_dentries(trans, root, | |
5372 | dir_inode, ctx); | |
18aa0922 FM |
5373 | iput(dir_inode); |
5374 | if (ret) | |
5375 | goto out; | |
5376 | } | |
5377 | path->slots[0]++; | |
5378 | } | |
5379 | ret = 0; | |
5380 | out: | |
5381 | btrfs_free_path(path); | |
5382 | return ret; | |
5383 | } | |
5384 | ||
e02119d5 CM |
5385 | /* |
5386 | * helper function around btrfs_log_inode to make sure newly created | |
5387 | * parent directories also end up in the log. A minimal inode and backref | |
5388 | * only logging is done of any parent directories that are older than | |
5389 | * the last committed transaction | |
5390 | */ | |
48a3b636 ES |
5391 | static int btrfs_log_inode_parent(struct btrfs_trans_handle *trans, |
5392 | struct btrfs_root *root, struct inode *inode, | |
49dae1bc FM |
5393 | struct dentry *parent, |
5394 | const loff_t start, | |
5395 | const loff_t end, | |
5396 | int exists_only, | |
8b050d35 | 5397 | struct btrfs_log_ctx *ctx) |
e02119d5 | 5398 | { |
0b246afa | 5399 | struct btrfs_fs_info *fs_info = root->fs_info; |
12fcfd22 | 5400 | int inode_only = exists_only ? LOG_INODE_EXISTS : LOG_INODE_ALL; |
e02119d5 | 5401 | struct super_block *sb; |
6a912213 | 5402 | struct dentry *old_parent = NULL; |
12fcfd22 | 5403 | int ret = 0; |
0b246afa | 5404 | u64 last_committed = fs_info->last_trans_committed; |
2f2ff0ee FM |
5405 | bool log_dentries = false; |
5406 | struct inode *orig_inode = inode; | |
12fcfd22 CM |
5407 | |
5408 | sb = inode->i_sb; | |
5409 | ||
0b246afa | 5410 | if (btrfs_test_opt(fs_info, NOTREELOG)) { |
3a5e1404 SW |
5411 | ret = 1; |
5412 | goto end_no_trans; | |
5413 | } | |
5414 | ||
995946dd MX |
5415 | /* |
5416 | * The prev transaction commit doesn't complete, we need do | |
5417 | * full commit by ourselves. | |
5418 | */ | |
0b246afa JM |
5419 | if (fs_info->last_trans_log_full_commit > |
5420 | fs_info->last_trans_committed) { | |
12fcfd22 CM |
5421 | ret = 1; |
5422 | goto end_no_trans; | |
5423 | } | |
5424 | ||
76dda93c YZ |
5425 | if (root != BTRFS_I(inode)->root || |
5426 | btrfs_root_refs(&root->root_item) == 0) { | |
5427 | ret = 1; | |
5428 | goto end_no_trans; | |
5429 | } | |
5430 | ||
12fcfd22 CM |
5431 | ret = check_parent_dirs_for_sync(trans, inode, parent, |
5432 | sb, last_committed); | |
5433 | if (ret) | |
5434 | goto end_no_trans; | |
e02119d5 | 5435 | |
0f8939b8 | 5436 | if (btrfs_inode_in_log(BTRFS_I(inode), trans->transid)) { |
257c62e1 CM |
5437 | ret = BTRFS_NO_LOG_SYNC; |
5438 | goto end_no_trans; | |
5439 | } | |
5440 | ||
8b050d35 | 5441 | ret = start_log_trans(trans, root, ctx); |
4a500fd1 | 5442 | if (ret) |
e87ac136 | 5443 | goto end_no_trans; |
e02119d5 | 5444 | |
8407f553 | 5445 | ret = btrfs_log_inode(trans, root, inode, inode_only, start, end, ctx); |
4a500fd1 YZ |
5446 | if (ret) |
5447 | goto end_trans; | |
12fcfd22 | 5448 | |
af4176b4 CM |
5449 | /* |
5450 | * for regular files, if its inode is already on disk, we don't | |
5451 | * have to worry about the parents at all. This is because | |
5452 | * we can use the last_unlink_trans field to record renames | |
5453 | * and other fun in this file. | |
5454 | */ | |
5455 | if (S_ISREG(inode->i_mode) && | |
5456 | BTRFS_I(inode)->generation <= last_committed && | |
4a500fd1 YZ |
5457 | BTRFS_I(inode)->last_unlink_trans <= last_committed) { |
5458 | ret = 0; | |
5459 | goto end_trans; | |
5460 | } | |
af4176b4 | 5461 | |
2f2ff0ee FM |
5462 | if (S_ISDIR(inode->i_mode) && ctx && ctx->log_new_dentries) |
5463 | log_dentries = true; | |
5464 | ||
18aa0922 | 5465 | /* |
01327610 | 5466 | * On unlink we must make sure all our current and old parent directory |
18aa0922 FM |
5467 | * inodes are fully logged. This is to prevent leaving dangling |
5468 | * directory index entries in directories that were our parents but are | |
5469 | * not anymore. Not doing this results in old parent directory being | |
5470 | * impossible to delete after log replay (rmdir will always fail with | |
5471 | * error -ENOTEMPTY). | |
5472 | * | |
5473 | * Example 1: | |
5474 | * | |
5475 | * mkdir testdir | |
5476 | * touch testdir/foo | |
5477 | * ln testdir/foo testdir/bar | |
5478 | * sync | |
5479 | * unlink testdir/bar | |
5480 | * xfs_io -c fsync testdir/foo | |
5481 | * <power failure> | |
5482 | * mount fs, triggers log replay | |
5483 | * | |
5484 | * If we don't log the parent directory (testdir), after log replay the | |
5485 | * directory still has an entry pointing to the file inode using the bar | |
5486 | * name, but a matching BTRFS_INODE_[REF|EXTREF]_KEY does not exist and | |
5487 | * the file inode has a link count of 1. | |
5488 | * | |
5489 | * Example 2: | |
5490 | * | |
5491 | * mkdir testdir | |
5492 | * touch foo | |
5493 | * ln foo testdir/foo2 | |
5494 | * ln foo testdir/foo3 | |
5495 | * sync | |
5496 | * unlink testdir/foo3 | |
5497 | * xfs_io -c fsync foo | |
5498 | * <power failure> | |
5499 | * mount fs, triggers log replay | |
5500 | * | |
5501 | * Similar as the first example, after log replay the parent directory | |
5502 | * testdir still has an entry pointing to the inode file with name foo3 | |
5503 | * but the file inode does not have a matching BTRFS_INODE_REF_KEY item | |
5504 | * and has a link count of 2. | |
5505 | */ | |
5506 | if (BTRFS_I(inode)->last_unlink_trans > last_committed) { | |
5507 | ret = btrfs_log_all_parents(trans, orig_inode, ctx); | |
5508 | if (ret) | |
5509 | goto end_trans; | |
5510 | } | |
5511 | ||
12fcfd22 | 5512 | while (1) { |
fc64005c | 5513 | if (!parent || d_really_is_negative(parent) || sb != parent->d_sb) |
e02119d5 CM |
5514 | break; |
5515 | ||
2b0143b5 | 5516 | inode = d_inode(parent); |
76dda93c YZ |
5517 | if (root != BTRFS_I(inode)->root) |
5518 | break; | |
5519 | ||
18aa0922 FM |
5520 | if (BTRFS_I(inode)->generation > last_committed) { |
5521 | ret = btrfs_log_inode(trans, root, inode, | |
5522 | LOG_INODE_EXISTS, | |
8407f553 | 5523 | 0, LLONG_MAX, ctx); |
4a500fd1 YZ |
5524 | if (ret) |
5525 | goto end_trans; | |
12fcfd22 | 5526 | } |
76dda93c | 5527 | if (IS_ROOT(parent)) |
e02119d5 | 5528 | break; |
12fcfd22 | 5529 | |
6a912213 JB |
5530 | parent = dget_parent(parent); |
5531 | dput(old_parent); | |
5532 | old_parent = parent; | |
e02119d5 | 5533 | } |
2f2ff0ee FM |
5534 | if (log_dentries) |
5535 | ret = log_new_dir_dentries(trans, root, orig_inode, ctx); | |
5536 | else | |
5537 | ret = 0; | |
4a500fd1 | 5538 | end_trans: |
6a912213 | 5539 | dput(old_parent); |
4a500fd1 | 5540 | if (ret < 0) { |
0b246afa | 5541 | btrfs_set_log_full_commit(fs_info, trans); |
4a500fd1 YZ |
5542 | ret = 1; |
5543 | } | |
8b050d35 MX |
5544 | |
5545 | if (ret) | |
5546 | btrfs_remove_log_ctx(root, ctx); | |
12fcfd22 CM |
5547 | btrfs_end_log_trans(root); |
5548 | end_no_trans: | |
5549 | return ret; | |
e02119d5 CM |
5550 | } |
5551 | ||
5552 | /* | |
5553 | * it is not safe to log dentry if the chunk root has added new | |
5554 | * chunks. This returns 0 if the dentry was logged, and 1 otherwise. | |
5555 | * If this returns 1, you must commit the transaction to safely get your | |
5556 | * data on disk. | |
5557 | */ | |
5558 | int btrfs_log_dentry_safe(struct btrfs_trans_handle *trans, | |
8b050d35 | 5559 | struct btrfs_root *root, struct dentry *dentry, |
49dae1bc FM |
5560 | const loff_t start, |
5561 | const loff_t end, | |
8b050d35 | 5562 | struct btrfs_log_ctx *ctx) |
e02119d5 | 5563 | { |
6a912213 JB |
5564 | struct dentry *parent = dget_parent(dentry); |
5565 | int ret; | |
5566 | ||
2b0143b5 | 5567 | ret = btrfs_log_inode_parent(trans, root, d_inode(dentry), parent, |
49dae1bc | 5568 | start, end, 0, ctx); |
6a912213 JB |
5569 | dput(parent); |
5570 | ||
5571 | return ret; | |
e02119d5 CM |
5572 | } |
5573 | ||
5574 | /* | |
5575 | * should be called during mount to recover any replay any log trees | |
5576 | * from the FS | |
5577 | */ | |
5578 | int btrfs_recover_log_trees(struct btrfs_root *log_root_tree) | |
5579 | { | |
5580 | int ret; | |
5581 | struct btrfs_path *path; | |
5582 | struct btrfs_trans_handle *trans; | |
5583 | struct btrfs_key key; | |
5584 | struct btrfs_key found_key; | |
5585 | struct btrfs_key tmp_key; | |
5586 | struct btrfs_root *log; | |
5587 | struct btrfs_fs_info *fs_info = log_root_tree->fs_info; | |
5588 | struct walk_control wc = { | |
5589 | .process_func = process_one_buffer, | |
5590 | .stage = 0, | |
5591 | }; | |
5592 | ||
e02119d5 | 5593 | path = btrfs_alloc_path(); |
db5b493a TI |
5594 | if (!path) |
5595 | return -ENOMEM; | |
5596 | ||
afcdd129 | 5597 | set_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags); |
e02119d5 | 5598 | |
4a500fd1 | 5599 | trans = btrfs_start_transaction(fs_info->tree_root, 0); |
79787eaa JM |
5600 | if (IS_ERR(trans)) { |
5601 | ret = PTR_ERR(trans); | |
5602 | goto error; | |
5603 | } | |
e02119d5 CM |
5604 | |
5605 | wc.trans = trans; | |
5606 | wc.pin = 1; | |
5607 | ||
db5b493a | 5608 | ret = walk_log_tree(trans, log_root_tree, &wc); |
79787eaa | 5609 | if (ret) { |
5d163e0e JM |
5610 | btrfs_handle_fs_error(fs_info, ret, |
5611 | "Failed to pin buffers while recovering log root tree."); | |
79787eaa JM |
5612 | goto error; |
5613 | } | |
e02119d5 CM |
5614 | |
5615 | again: | |
5616 | key.objectid = BTRFS_TREE_LOG_OBJECTID; | |
5617 | key.offset = (u64)-1; | |
962a298f | 5618 | key.type = BTRFS_ROOT_ITEM_KEY; |
e02119d5 | 5619 | |
d397712b | 5620 | while (1) { |
e02119d5 | 5621 | ret = btrfs_search_slot(NULL, log_root_tree, &key, path, 0, 0); |
79787eaa JM |
5622 | |
5623 | if (ret < 0) { | |
34d97007 | 5624 | btrfs_handle_fs_error(fs_info, ret, |
79787eaa JM |
5625 | "Couldn't find tree log root."); |
5626 | goto error; | |
5627 | } | |
e02119d5 CM |
5628 | if (ret > 0) { |
5629 | if (path->slots[0] == 0) | |
5630 | break; | |
5631 | path->slots[0]--; | |
5632 | } | |
5633 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, | |
5634 | path->slots[0]); | |
b3b4aa74 | 5635 | btrfs_release_path(path); |
e02119d5 CM |
5636 | if (found_key.objectid != BTRFS_TREE_LOG_OBJECTID) |
5637 | break; | |
5638 | ||
cb517eab | 5639 | log = btrfs_read_fs_root(log_root_tree, &found_key); |
79787eaa JM |
5640 | if (IS_ERR(log)) { |
5641 | ret = PTR_ERR(log); | |
34d97007 | 5642 | btrfs_handle_fs_error(fs_info, ret, |
79787eaa JM |
5643 | "Couldn't read tree log root."); |
5644 | goto error; | |
5645 | } | |
e02119d5 CM |
5646 | |
5647 | tmp_key.objectid = found_key.offset; | |
5648 | tmp_key.type = BTRFS_ROOT_ITEM_KEY; | |
5649 | tmp_key.offset = (u64)-1; | |
5650 | ||
5651 | wc.replay_dest = btrfs_read_fs_root_no_name(fs_info, &tmp_key); | |
79787eaa JM |
5652 | if (IS_ERR(wc.replay_dest)) { |
5653 | ret = PTR_ERR(wc.replay_dest); | |
b50c6e25 JB |
5654 | free_extent_buffer(log->node); |
5655 | free_extent_buffer(log->commit_root); | |
5656 | kfree(log); | |
5d163e0e JM |
5657 | btrfs_handle_fs_error(fs_info, ret, |
5658 | "Couldn't read target root for tree log recovery."); | |
79787eaa JM |
5659 | goto error; |
5660 | } | |
e02119d5 | 5661 | |
07d400a6 | 5662 | wc.replay_dest->log_root = log; |
5d4f98a2 | 5663 | btrfs_record_root_in_trans(trans, wc.replay_dest); |
e02119d5 | 5664 | ret = walk_log_tree(trans, log, &wc); |
e02119d5 | 5665 | |
b50c6e25 | 5666 | if (!ret && wc.stage == LOG_WALK_REPLAY_ALL) { |
e02119d5 CM |
5667 | ret = fixup_inode_link_counts(trans, wc.replay_dest, |
5668 | path); | |
e02119d5 CM |
5669 | } |
5670 | ||
5671 | key.offset = found_key.offset - 1; | |
07d400a6 | 5672 | wc.replay_dest->log_root = NULL; |
e02119d5 | 5673 | free_extent_buffer(log->node); |
b263c2c8 | 5674 | free_extent_buffer(log->commit_root); |
e02119d5 CM |
5675 | kfree(log); |
5676 | ||
b50c6e25 JB |
5677 | if (ret) |
5678 | goto error; | |
5679 | ||
e02119d5 CM |
5680 | if (found_key.offset == 0) |
5681 | break; | |
5682 | } | |
b3b4aa74 | 5683 | btrfs_release_path(path); |
e02119d5 CM |
5684 | |
5685 | /* step one is to pin it all, step two is to replay just inodes */ | |
5686 | if (wc.pin) { | |
5687 | wc.pin = 0; | |
5688 | wc.process_func = replay_one_buffer; | |
5689 | wc.stage = LOG_WALK_REPLAY_INODES; | |
5690 | goto again; | |
5691 | } | |
5692 | /* step three is to replay everything */ | |
5693 | if (wc.stage < LOG_WALK_REPLAY_ALL) { | |
5694 | wc.stage++; | |
5695 | goto again; | |
5696 | } | |
5697 | ||
5698 | btrfs_free_path(path); | |
5699 | ||
abefa55a | 5700 | /* step 4: commit the transaction, which also unpins the blocks */ |
3a45bb20 | 5701 | ret = btrfs_commit_transaction(trans); |
abefa55a JB |
5702 | if (ret) |
5703 | return ret; | |
5704 | ||
e02119d5 CM |
5705 | free_extent_buffer(log_root_tree->node); |
5706 | log_root_tree->log_root = NULL; | |
afcdd129 | 5707 | clear_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags); |
e02119d5 | 5708 | kfree(log_root_tree); |
79787eaa | 5709 | |
abefa55a | 5710 | return 0; |
79787eaa | 5711 | error: |
b50c6e25 | 5712 | if (wc.trans) |
3a45bb20 | 5713 | btrfs_end_transaction(wc.trans); |
79787eaa JM |
5714 | btrfs_free_path(path); |
5715 | return ret; | |
e02119d5 | 5716 | } |
12fcfd22 CM |
5717 | |
5718 | /* | |
5719 | * there are some corner cases where we want to force a full | |
5720 | * commit instead of allowing a directory to be logged. | |
5721 | * | |
5722 | * They revolve around files there were unlinked from the directory, and | |
5723 | * this function updates the parent directory so that a full commit is | |
5724 | * properly done if it is fsync'd later after the unlinks are done. | |
2be63d5c FM |
5725 | * |
5726 | * Must be called before the unlink operations (updates to the subvolume tree, | |
5727 | * inodes, etc) are done. | |
12fcfd22 CM |
5728 | */ |
5729 | void btrfs_record_unlink_dir(struct btrfs_trans_handle *trans, | |
4176bdbf | 5730 | struct btrfs_inode *dir, struct btrfs_inode *inode, |
12fcfd22 CM |
5731 | int for_rename) |
5732 | { | |
af4176b4 CM |
5733 | /* |
5734 | * when we're logging a file, if it hasn't been renamed | |
5735 | * or unlinked, and its inode is fully committed on disk, | |
5736 | * we don't have to worry about walking up the directory chain | |
5737 | * to log its parents. | |
5738 | * | |
5739 | * So, we use the last_unlink_trans field to put this transid | |
5740 | * into the file. When the file is logged we check it and | |
5741 | * don't log the parents if the file is fully on disk. | |
5742 | */ | |
4176bdbf NB |
5743 | mutex_lock(&inode->log_mutex); |
5744 | inode->last_unlink_trans = trans->transid; | |
5745 | mutex_unlock(&inode->log_mutex); | |
af4176b4 | 5746 | |
12fcfd22 CM |
5747 | /* |
5748 | * if this directory was already logged any new | |
5749 | * names for this file/dir will get recorded | |
5750 | */ | |
5751 | smp_mb(); | |
4176bdbf | 5752 | if (dir->logged_trans == trans->transid) |
12fcfd22 CM |
5753 | return; |
5754 | ||
5755 | /* | |
5756 | * if the inode we're about to unlink was logged, | |
5757 | * the log will be properly updated for any new names | |
5758 | */ | |
4176bdbf | 5759 | if (inode->logged_trans == trans->transid) |
12fcfd22 CM |
5760 | return; |
5761 | ||
5762 | /* | |
5763 | * when renaming files across directories, if the directory | |
5764 | * there we're unlinking from gets fsync'd later on, there's | |
5765 | * no way to find the destination directory later and fsync it | |
5766 | * properly. So, we have to be conservative and force commits | |
5767 | * so the new name gets discovered. | |
5768 | */ | |
5769 | if (for_rename) | |
5770 | goto record; | |
5771 | ||
5772 | /* we can safely do the unlink without any special recording */ | |
5773 | return; | |
5774 | ||
5775 | record: | |
4176bdbf NB |
5776 | mutex_lock(&dir->log_mutex); |
5777 | dir->last_unlink_trans = trans->transid; | |
5778 | mutex_unlock(&dir->log_mutex); | |
1ec9a1ae FM |
5779 | } |
5780 | ||
5781 | /* | |
5782 | * Make sure that if someone attempts to fsync the parent directory of a deleted | |
5783 | * snapshot, it ends up triggering a transaction commit. This is to guarantee | |
5784 | * that after replaying the log tree of the parent directory's root we will not | |
5785 | * see the snapshot anymore and at log replay time we will not see any log tree | |
5786 | * corresponding to the deleted snapshot's root, which could lead to replaying | |
5787 | * it after replaying the log tree of the parent directory (which would replay | |
5788 | * the snapshot delete operation). | |
2be63d5c FM |
5789 | * |
5790 | * Must be called before the actual snapshot destroy operation (updates to the | |
5791 | * parent root and tree of tree roots trees, etc) are done. | |
1ec9a1ae FM |
5792 | */ |
5793 | void btrfs_record_snapshot_destroy(struct btrfs_trans_handle *trans, | |
43663557 | 5794 | struct btrfs_inode *dir) |
1ec9a1ae | 5795 | { |
43663557 NB |
5796 | mutex_lock(&dir->log_mutex); |
5797 | dir->last_unlink_trans = trans->transid; | |
5798 | mutex_unlock(&dir->log_mutex); | |
12fcfd22 CM |
5799 | } |
5800 | ||
5801 | /* | |
5802 | * Call this after adding a new name for a file and it will properly | |
5803 | * update the log to reflect the new name. | |
5804 | * | |
5805 | * It will return zero if all goes well, and it will return 1 if a | |
5806 | * full transaction commit is required. | |
5807 | */ | |
5808 | int btrfs_log_new_name(struct btrfs_trans_handle *trans, | |
9ca5fbfb | 5809 | struct btrfs_inode *inode, struct btrfs_inode *old_dir, |
12fcfd22 CM |
5810 | struct dentry *parent) |
5811 | { | |
9ca5fbfb NB |
5812 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb); |
5813 | struct btrfs_root * root = inode->root; | |
12fcfd22 | 5814 | |
af4176b4 CM |
5815 | /* |
5816 | * this will force the logging code to walk the dentry chain | |
5817 | * up for the file | |
5818 | */ | |
9ca5fbfb NB |
5819 | if (S_ISREG(inode->vfs_inode.i_mode)) |
5820 | inode->last_unlink_trans = trans->transid; | |
af4176b4 | 5821 | |
12fcfd22 CM |
5822 | /* |
5823 | * if this inode hasn't been logged and directory we're renaming it | |
5824 | * from hasn't been logged, we don't need to log it | |
5825 | */ | |
9ca5fbfb NB |
5826 | if (inode->logged_trans <= fs_info->last_trans_committed && |
5827 | (!old_dir || old_dir->logged_trans <= fs_info->last_trans_committed)) | |
12fcfd22 CM |
5828 | return 0; |
5829 | ||
9ca5fbfb | 5830 | return btrfs_log_inode_parent(trans, root, &inode->vfs_inode, parent, 0, |
49dae1bc | 5831 | LLONG_MAX, 1, NULL); |
12fcfd22 CM |
5832 | } |
5833 |