]>
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" |
e02119d5 CM |
30 | |
31 | /* magic values for the inode_only field in btrfs_log_inode: | |
32 | * | |
33 | * LOG_INODE_ALL means to log everything | |
34 | * LOG_INODE_EXISTS means to log just enough to recreate the inode | |
35 | * during log replay | |
36 | */ | |
37 | #define LOG_INODE_ALL 0 | |
38 | #define LOG_INODE_EXISTS 1 | |
39 | ||
12fcfd22 CM |
40 | /* |
41 | * directory trouble cases | |
42 | * | |
43 | * 1) on rename or unlink, if the inode being unlinked isn't in the fsync | |
44 | * log, we must force a full commit before doing an fsync of the directory | |
45 | * where the unlink was done. | |
46 | * ---> record transid of last unlink/rename per directory | |
47 | * | |
48 | * mkdir foo/some_dir | |
49 | * normal commit | |
50 | * rename foo/some_dir foo2/some_dir | |
51 | * mkdir foo/some_dir | |
52 | * fsync foo/some_dir/some_file | |
53 | * | |
54 | * The fsync above will unlink the original some_dir without recording | |
55 | * it in its new location (foo2). After a crash, some_dir will be gone | |
56 | * unless the fsync of some_file forces a full commit | |
57 | * | |
58 | * 2) we must log any new names for any file or dir that is in the fsync | |
59 | * log. ---> check inode while renaming/linking. | |
60 | * | |
61 | * 2a) we must log any new names for any file or dir during rename | |
62 | * when the directory they are being removed from was logged. | |
63 | * ---> check inode and old parent dir during rename | |
64 | * | |
65 | * 2a is actually the more important variant. With the extra logging | |
66 | * a crash might unlink the old name without recreating the new one | |
67 | * | |
68 | * 3) after a crash, we must go through any directories with a link count | |
69 | * of zero and redo the rm -rf | |
70 | * | |
71 | * mkdir f1/foo | |
72 | * normal commit | |
73 | * rm -rf f1/foo | |
74 | * fsync(f1) | |
75 | * | |
76 | * The directory f1 was fully removed from the FS, but fsync was never | |
77 | * called on f1, only its parent dir. After a crash the rm -rf must | |
78 | * be replayed. This must be able to recurse down the entire | |
79 | * directory tree. The inode link count fixup code takes care of the | |
80 | * ugly details. | |
81 | */ | |
82 | ||
e02119d5 CM |
83 | /* |
84 | * stages for the tree walking. The first | |
85 | * stage (0) is to only pin down the blocks we find | |
86 | * the second stage (1) is to make sure that all the inodes | |
87 | * we find in the log are created in the subvolume. | |
88 | * | |
89 | * The last stage is to deal with directories and links and extents | |
90 | * and all the other fun semantics | |
91 | */ | |
92 | #define LOG_WALK_PIN_ONLY 0 | |
93 | #define LOG_WALK_REPLAY_INODES 1 | |
dd8e7217 JB |
94 | #define LOG_WALK_REPLAY_DIR_INDEX 2 |
95 | #define LOG_WALK_REPLAY_ALL 3 | |
e02119d5 | 96 | |
12fcfd22 | 97 | static int btrfs_log_inode(struct btrfs_trans_handle *trans, |
49dae1bc FM |
98 | struct btrfs_root *root, struct inode *inode, |
99 | int inode_only, | |
100 | const loff_t start, | |
8407f553 FM |
101 | const loff_t end, |
102 | struct btrfs_log_ctx *ctx); | |
ec051c0f YZ |
103 | static int link_to_fixup_dir(struct btrfs_trans_handle *trans, |
104 | struct btrfs_root *root, | |
105 | struct btrfs_path *path, u64 objectid); | |
12fcfd22 CM |
106 | static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans, |
107 | struct btrfs_root *root, | |
108 | struct btrfs_root *log, | |
109 | struct btrfs_path *path, | |
110 | u64 dirid, int del_all); | |
e02119d5 CM |
111 | |
112 | /* | |
113 | * tree logging is a special write ahead log used to make sure that | |
114 | * fsyncs and O_SYNCs can happen without doing full tree commits. | |
115 | * | |
116 | * Full tree commits are expensive because they require commonly | |
117 | * modified blocks to be recowed, creating many dirty pages in the | |
118 | * extent tree an 4x-6x higher write load than ext3. | |
119 | * | |
120 | * Instead of doing a tree commit on every fsync, we use the | |
121 | * key ranges and transaction ids to find items for a given file or directory | |
122 | * that have changed in this transaction. Those items are copied into | |
123 | * a special tree (one per subvolume root), that tree is written to disk | |
124 | * and then the fsync is considered complete. | |
125 | * | |
126 | * After a crash, items are copied out of the log-tree back into the | |
127 | * subvolume tree. Any file data extents found are recorded in the extent | |
128 | * allocation tree, and the log-tree freed. | |
129 | * | |
130 | * The log tree is read three times, once to pin down all the extents it is | |
131 | * using in ram and once, once to create all the inodes logged in the tree | |
132 | * and once to do all the other items. | |
133 | */ | |
134 | ||
e02119d5 CM |
135 | /* |
136 | * start a sub transaction and setup the log tree | |
137 | * this increments the log tree writer count to make the people | |
138 | * syncing the tree wait for us to finish | |
139 | */ | |
140 | static int start_log_trans(struct btrfs_trans_handle *trans, | |
8b050d35 MX |
141 | struct btrfs_root *root, |
142 | struct btrfs_log_ctx *ctx) | |
e02119d5 | 143 | { |
34eb2a52 | 144 | int ret = 0; |
7237f183 YZ |
145 | |
146 | mutex_lock(&root->log_mutex); | |
34eb2a52 | 147 | |
7237f183 | 148 | if (root->log_root) { |
995946dd | 149 | if (btrfs_need_log_full_commit(root->fs_info, trans)) { |
50471a38 MX |
150 | ret = -EAGAIN; |
151 | goto out; | |
152 | } | |
34eb2a52 | 153 | |
ff782e0a | 154 | if (!root->log_start_pid) { |
27cdeb70 | 155 | clear_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state); |
34eb2a52 | 156 | root->log_start_pid = current->pid; |
ff782e0a | 157 | } else if (root->log_start_pid != current->pid) { |
27cdeb70 | 158 | set_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state); |
ff782e0a | 159 | } |
34eb2a52 Z |
160 | } else { |
161 | mutex_lock(&root->fs_info->tree_log_mutex); | |
162 | if (!root->fs_info->log_root_tree) | |
163 | ret = btrfs_init_log_root_tree(trans, root->fs_info); | |
164 | mutex_unlock(&root->fs_info->tree_log_mutex); | |
165 | if (ret) | |
166 | goto out; | |
ff782e0a | 167 | |
e02119d5 | 168 | ret = btrfs_add_log_tree(trans, root); |
4a500fd1 | 169 | if (ret) |
e87ac136 | 170 | goto out; |
34eb2a52 Z |
171 | |
172 | clear_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state); | |
173 | root->log_start_pid = current->pid; | |
e02119d5 | 174 | } |
34eb2a52 | 175 | |
2ecb7923 | 176 | atomic_inc(&root->log_batch); |
7237f183 | 177 | atomic_inc(&root->log_writers); |
8b050d35 | 178 | if (ctx) { |
34eb2a52 | 179 | int index = root->log_transid % 2; |
8b050d35 | 180 | list_add_tail(&ctx->list, &root->log_ctxs[index]); |
d1433deb | 181 | ctx->log_transid = root->log_transid; |
8b050d35 | 182 | } |
34eb2a52 | 183 | |
e87ac136 | 184 | out: |
7237f183 | 185 | mutex_unlock(&root->log_mutex); |
e87ac136 | 186 | return ret; |
e02119d5 CM |
187 | } |
188 | ||
189 | /* | |
190 | * returns 0 if there was a log transaction running and we were able | |
191 | * to join, or returns -ENOENT if there were not transactions | |
192 | * in progress | |
193 | */ | |
194 | static int join_running_log_trans(struct btrfs_root *root) | |
195 | { | |
196 | int ret = -ENOENT; | |
197 | ||
198 | smp_mb(); | |
199 | if (!root->log_root) | |
200 | return -ENOENT; | |
201 | ||
7237f183 | 202 | mutex_lock(&root->log_mutex); |
e02119d5 CM |
203 | if (root->log_root) { |
204 | ret = 0; | |
7237f183 | 205 | atomic_inc(&root->log_writers); |
e02119d5 | 206 | } |
7237f183 | 207 | mutex_unlock(&root->log_mutex); |
e02119d5 CM |
208 | return ret; |
209 | } | |
210 | ||
12fcfd22 CM |
211 | /* |
212 | * This either makes the current running log transaction wait | |
213 | * until you call btrfs_end_log_trans() or it makes any future | |
214 | * log transactions wait until you call btrfs_end_log_trans() | |
215 | */ | |
216 | int btrfs_pin_log_trans(struct btrfs_root *root) | |
217 | { | |
218 | int ret = -ENOENT; | |
219 | ||
220 | mutex_lock(&root->log_mutex); | |
221 | atomic_inc(&root->log_writers); | |
222 | mutex_unlock(&root->log_mutex); | |
223 | return ret; | |
224 | } | |
225 | ||
e02119d5 CM |
226 | /* |
227 | * indicate we're done making changes to the log tree | |
228 | * and wake up anyone waiting to do a sync | |
229 | */ | |
143bede5 | 230 | void btrfs_end_log_trans(struct btrfs_root *root) |
e02119d5 | 231 | { |
7237f183 | 232 | if (atomic_dec_and_test(&root->log_writers)) { |
779adf0f DS |
233 | /* |
234 | * Implicit memory barrier after atomic_dec_and_test | |
235 | */ | |
7237f183 YZ |
236 | if (waitqueue_active(&root->log_writer_wait)) |
237 | wake_up(&root->log_writer_wait); | |
238 | } | |
e02119d5 CM |
239 | } |
240 | ||
241 | ||
242 | /* | |
243 | * the walk control struct is used to pass state down the chain when | |
244 | * processing the log tree. The stage field tells us which part | |
245 | * of the log tree processing we are currently doing. The others | |
246 | * are state fields used for that specific part | |
247 | */ | |
248 | struct walk_control { | |
249 | /* should we free the extent on disk when done? This is used | |
250 | * at transaction commit time while freeing a log tree | |
251 | */ | |
252 | int free; | |
253 | ||
254 | /* should we write out the extent buffer? This is used | |
255 | * while flushing the log tree to disk during a sync | |
256 | */ | |
257 | int write; | |
258 | ||
259 | /* should we wait for the extent buffer io to finish? Also used | |
260 | * while flushing the log tree to disk for a sync | |
261 | */ | |
262 | int wait; | |
263 | ||
264 | /* pin only walk, we record which extents on disk belong to the | |
265 | * log trees | |
266 | */ | |
267 | int pin; | |
268 | ||
269 | /* what stage of the replay code we're currently in */ | |
270 | int stage; | |
271 | ||
272 | /* the root we are currently replaying */ | |
273 | struct btrfs_root *replay_dest; | |
274 | ||
275 | /* the trans handle for the current replay */ | |
276 | struct btrfs_trans_handle *trans; | |
277 | ||
278 | /* the function that gets used to process blocks we find in the | |
279 | * tree. Note the extent_buffer might not be up to date when it is | |
280 | * passed in, and it must be checked or read if you need the data | |
281 | * inside it | |
282 | */ | |
283 | int (*process_func)(struct btrfs_root *log, struct extent_buffer *eb, | |
284 | struct walk_control *wc, u64 gen); | |
285 | }; | |
286 | ||
287 | /* | |
288 | * process_func used to pin down extents, write them or wait on them | |
289 | */ | |
290 | static int process_one_buffer(struct btrfs_root *log, | |
291 | struct extent_buffer *eb, | |
292 | struct walk_control *wc, u64 gen) | |
293 | { | |
b50c6e25 JB |
294 | int ret = 0; |
295 | ||
8c2a1a30 JB |
296 | /* |
297 | * If this fs is mixed then we need to be able to process the leaves to | |
298 | * pin down any logged extents, so we have to read the block. | |
299 | */ | |
300 | if (btrfs_fs_incompat(log->fs_info, MIXED_GROUPS)) { | |
301 | ret = btrfs_read_buffer(eb, gen); | |
302 | if (ret) | |
303 | return ret; | |
304 | } | |
305 | ||
04018de5 | 306 | if (wc->pin) |
b50c6e25 JB |
307 | ret = btrfs_pin_extent_for_log_replay(log->fs_info->extent_root, |
308 | eb->start, eb->len); | |
e02119d5 | 309 | |
b50c6e25 | 310 | if (!ret && btrfs_buffer_uptodate(eb, gen, 0)) { |
8c2a1a30 JB |
311 | if (wc->pin && btrfs_header_level(eb) == 0) |
312 | ret = btrfs_exclude_logged_extents(log, eb); | |
e02119d5 CM |
313 | if (wc->write) |
314 | btrfs_write_tree_block(eb); | |
315 | if (wc->wait) | |
316 | btrfs_wait_tree_block_writeback(eb); | |
317 | } | |
b50c6e25 | 318 | return ret; |
e02119d5 CM |
319 | } |
320 | ||
321 | /* | |
322 | * Item overwrite used by replay and tree logging. eb, slot and key all refer | |
323 | * to the src data we are copying out. | |
324 | * | |
325 | * root is the tree we are copying into, and path is a scratch | |
326 | * path for use in this function (it should be released on entry and | |
327 | * will be released on exit). | |
328 | * | |
329 | * If the key is already in the destination tree the existing item is | |
330 | * overwritten. If the existing item isn't big enough, it is extended. | |
331 | * If it is too large, it is truncated. | |
332 | * | |
333 | * If the key isn't in the destination yet, a new item is inserted. | |
334 | */ | |
335 | static noinline int overwrite_item(struct btrfs_trans_handle *trans, | |
336 | struct btrfs_root *root, | |
337 | struct btrfs_path *path, | |
338 | struct extent_buffer *eb, int slot, | |
339 | struct btrfs_key *key) | |
340 | { | |
341 | int ret; | |
342 | u32 item_size; | |
343 | u64 saved_i_size = 0; | |
344 | int save_old_i_size = 0; | |
345 | unsigned long src_ptr; | |
346 | unsigned long dst_ptr; | |
347 | int overwrite_root = 0; | |
4bc4bee4 | 348 | bool inode_item = key->type == BTRFS_INODE_ITEM_KEY; |
e02119d5 CM |
349 | |
350 | if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) | |
351 | overwrite_root = 1; | |
352 | ||
353 | item_size = btrfs_item_size_nr(eb, slot); | |
354 | src_ptr = btrfs_item_ptr_offset(eb, slot); | |
355 | ||
356 | /* look for the key in the destination tree */ | |
357 | ret = btrfs_search_slot(NULL, root, key, path, 0, 0); | |
4bc4bee4 JB |
358 | if (ret < 0) |
359 | return ret; | |
360 | ||
e02119d5 CM |
361 | if (ret == 0) { |
362 | char *src_copy; | |
363 | char *dst_copy; | |
364 | u32 dst_size = btrfs_item_size_nr(path->nodes[0], | |
365 | path->slots[0]); | |
366 | if (dst_size != item_size) | |
367 | goto insert; | |
368 | ||
369 | if (item_size == 0) { | |
b3b4aa74 | 370 | btrfs_release_path(path); |
e02119d5 CM |
371 | return 0; |
372 | } | |
373 | dst_copy = kmalloc(item_size, GFP_NOFS); | |
374 | src_copy = kmalloc(item_size, GFP_NOFS); | |
2a29edc6 | 375 | if (!dst_copy || !src_copy) { |
b3b4aa74 | 376 | btrfs_release_path(path); |
2a29edc6 | 377 | kfree(dst_copy); |
378 | kfree(src_copy); | |
379 | return -ENOMEM; | |
380 | } | |
e02119d5 CM |
381 | |
382 | read_extent_buffer(eb, src_copy, src_ptr, item_size); | |
383 | ||
384 | dst_ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]); | |
385 | read_extent_buffer(path->nodes[0], dst_copy, dst_ptr, | |
386 | item_size); | |
387 | ret = memcmp(dst_copy, src_copy, item_size); | |
388 | ||
389 | kfree(dst_copy); | |
390 | kfree(src_copy); | |
391 | /* | |
392 | * they have the same contents, just return, this saves | |
393 | * us from cowing blocks in the destination tree and doing | |
394 | * extra writes that may not have been done by a previous | |
395 | * sync | |
396 | */ | |
397 | if (ret == 0) { | |
b3b4aa74 | 398 | btrfs_release_path(path); |
e02119d5 CM |
399 | return 0; |
400 | } | |
401 | ||
4bc4bee4 JB |
402 | /* |
403 | * We need to load the old nbytes into the inode so when we | |
404 | * replay the extents we've logged we get the right nbytes. | |
405 | */ | |
406 | if (inode_item) { | |
407 | struct btrfs_inode_item *item; | |
408 | u64 nbytes; | |
d555438b | 409 | u32 mode; |
4bc4bee4 JB |
410 | |
411 | item = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
412 | struct btrfs_inode_item); | |
413 | nbytes = btrfs_inode_nbytes(path->nodes[0], item); | |
414 | item = btrfs_item_ptr(eb, slot, | |
415 | struct btrfs_inode_item); | |
416 | btrfs_set_inode_nbytes(eb, item, nbytes); | |
d555438b JB |
417 | |
418 | /* | |
419 | * If this is a directory we need to reset the i_size to | |
420 | * 0 so that we can set it up properly when replaying | |
421 | * the rest of the items in this log. | |
422 | */ | |
423 | mode = btrfs_inode_mode(eb, item); | |
424 | if (S_ISDIR(mode)) | |
425 | btrfs_set_inode_size(eb, item, 0); | |
4bc4bee4 JB |
426 | } |
427 | } else if (inode_item) { | |
428 | struct btrfs_inode_item *item; | |
d555438b | 429 | u32 mode; |
4bc4bee4 JB |
430 | |
431 | /* | |
432 | * New inode, set nbytes to 0 so that the nbytes comes out | |
433 | * properly when we replay the extents. | |
434 | */ | |
435 | item = btrfs_item_ptr(eb, slot, struct btrfs_inode_item); | |
436 | btrfs_set_inode_nbytes(eb, item, 0); | |
d555438b JB |
437 | |
438 | /* | |
439 | * If this is a directory we need to reset the i_size to 0 so | |
440 | * that we can set it up properly when replaying the rest of | |
441 | * the items in this log. | |
442 | */ | |
443 | mode = btrfs_inode_mode(eb, item); | |
444 | if (S_ISDIR(mode)) | |
445 | btrfs_set_inode_size(eb, item, 0); | |
e02119d5 CM |
446 | } |
447 | insert: | |
b3b4aa74 | 448 | btrfs_release_path(path); |
e02119d5 | 449 | /* try to insert the key into the destination tree */ |
df8d116f | 450 | path->skip_release_on_error = 1; |
e02119d5 CM |
451 | ret = btrfs_insert_empty_item(trans, root, path, |
452 | key, item_size); | |
df8d116f | 453 | path->skip_release_on_error = 0; |
e02119d5 CM |
454 | |
455 | /* make sure any existing item is the correct size */ | |
df8d116f | 456 | if (ret == -EEXIST || ret == -EOVERFLOW) { |
e02119d5 CM |
457 | u32 found_size; |
458 | found_size = btrfs_item_size_nr(path->nodes[0], | |
459 | path->slots[0]); | |
143bede5 | 460 | if (found_size > item_size) |
afe5fea7 | 461 | btrfs_truncate_item(root, path, item_size, 1); |
143bede5 | 462 | else if (found_size < item_size) |
4b90c680 | 463 | btrfs_extend_item(root, path, |
143bede5 | 464 | item_size - found_size); |
e02119d5 | 465 | } else if (ret) { |
4a500fd1 | 466 | return ret; |
e02119d5 CM |
467 | } |
468 | dst_ptr = btrfs_item_ptr_offset(path->nodes[0], | |
469 | path->slots[0]); | |
470 | ||
471 | /* don't overwrite an existing inode if the generation number | |
472 | * was logged as zero. This is done when the tree logging code | |
473 | * is just logging an inode to make sure it exists after recovery. | |
474 | * | |
475 | * Also, don't overwrite i_size on directories during replay. | |
476 | * log replay inserts and removes directory items based on the | |
477 | * state of the tree found in the subvolume, and i_size is modified | |
478 | * as it goes | |
479 | */ | |
480 | if (key->type == BTRFS_INODE_ITEM_KEY && ret == -EEXIST) { | |
481 | struct btrfs_inode_item *src_item; | |
482 | struct btrfs_inode_item *dst_item; | |
483 | ||
484 | src_item = (struct btrfs_inode_item *)src_ptr; | |
485 | dst_item = (struct btrfs_inode_item *)dst_ptr; | |
486 | ||
1a4bcf47 FM |
487 | if (btrfs_inode_generation(eb, src_item) == 0) { |
488 | struct extent_buffer *dst_eb = path->nodes[0]; | |
2f2ff0ee | 489 | const u64 ino_size = btrfs_inode_size(eb, src_item); |
1a4bcf47 | 490 | |
2f2ff0ee FM |
491 | /* |
492 | * For regular files an ino_size == 0 is used only when | |
493 | * logging that an inode exists, as part of a directory | |
494 | * fsync, and the inode wasn't fsynced before. In this | |
495 | * case don't set the size of the inode in the fs/subvol | |
496 | * tree, otherwise we would be throwing valid data away. | |
497 | */ | |
1a4bcf47 | 498 | if (S_ISREG(btrfs_inode_mode(eb, src_item)) && |
2f2ff0ee FM |
499 | S_ISREG(btrfs_inode_mode(dst_eb, dst_item)) && |
500 | ino_size != 0) { | |
1a4bcf47 | 501 | struct btrfs_map_token token; |
1a4bcf47 FM |
502 | |
503 | btrfs_init_map_token(&token); | |
504 | btrfs_set_token_inode_size(dst_eb, dst_item, | |
505 | ino_size, &token); | |
506 | } | |
e02119d5 | 507 | goto no_copy; |
1a4bcf47 | 508 | } |
e02119d5 CM |
509 | |
510 | if (overwrite_root && | |
511 | S_ISDIR(btrfs_inode_mode(eb, src_item)) && | |
512 | S_ISDIR(btrfs_inode_mode(path->nodes[0], dst_item))) { | |
513 | save_old_i_size = 1; | |
514 | saved_i_size = btrfs_inode_size(path->nodes[0], | |
515 | dst_item); | |
516 | } | |
517 | } | |
518 | ||
519 | copy_extent_buffer(path->nodes[0], eb, dst_ptr, | |
520 | src_ptr, item_size); | |
521 | ||
522 | if (save_old_i_size) { | |
523 | struct btrfs_inode_item *dst_item; | |
524 | dst_item = (struct btrfs_inode_item *)dst_ptr; | |
525 | btrfs_set_inode_size(path->nodes[0], dst_item, saved_i_size); | |
526 | } | |
527 | ||
528 | /* make sure the generation is filled in */ | |
529 | if (key->type == BTRFS_INODE_ITEM_KEY) { | |
530 | struct btrfs_inode_item *dst_item; | |
531 | dst_item = (struct btrfs_inode_item *)dst_ptr; | |
532 | if (btrfs_inode_generation(path->nodes[0], dst_item) == 0) { | |
533 | btrfs_set_inode_generation(path->nodes[0], dst_item, | |
534 | trans->transid); | |
535 | } | |
536 | } | |
537 | no_copy: | |
538 | btrfs_mark_buffer_dirty(path->nodes[0]); | |
b3b4aa74 | 539 | btrfs_release_path(path); |
e02119d5 CM |
540 | return 0; |
541 | } | |
542 | ||
543 | /* | |
544 | * simple helper to read an inode off the disk from a given root | |
545 | * This can only be called for subvolume roots and not for the log | |
546 | */ | |
547 | static noinline struct inode *read_one_inode(struct btrfs_root *root, | |
548 | u64 objectid) | |
549 | { | |
5d4f98a2 | 550 | struct btrfs_key key; |
e02119d5 | 551 | struct inode *inode; |
e02119d5 | 552 | |
5d4f98a2 YZ |
553 | key.objectid = objectid; |
554 | key.type = BTRFS_INODE_ITEM_KEY; | |
555 | key.offset = 0; | |
73f73415 | 556 | inode = btrfs_iget(root->fs_info->sb, &key, root, NULL); |
5d4f98a2 YZ |
557 | if (IS_ERR(inode)) { |
558 | inode = NULL; | |
559 | } else if (is_bad_inode(inode)) { | |
e02119d5 CM |
560 | iput(inode); |
561 | inode = NULL; | |
562 | } | |
563 | return inode; | |
564 | } | |
565 | ||
566 | /* replays a single extent in 'eb' at 'slot' with 'key' into the | |
567 | * subvolume 'root'. path is released on entry and should be released | |
568 | * on exit. | |
569 | * | |
570 | * extents in the log tree have not been allocated out of the extent | |
571 | * tree yet. So, this completes the allocation, taking a reference | |
572 | * as required if the extent already exists or creating a new extent | |
573 | * if it isn't in the extent allocation tree yet. | |
574 | * | |
575 | * The extent is inserted into the file, dropping any existing extents | |
576 | * from the file that overlap the new one. | |
577 | */ | |
578 | static noinline int replay_one_extent(struct btrfs_trans_handle *trans, | |
579 | struct btrfs_root *root, | |
580 | struct btrfs_path *path, | |
581 | struct extent_buffer *eb, int slot, | |
582 | struct btrfs_key *key) | |
583 | { | |
584 | int found_type; | |
e02119d5 | 585 | u64 extent_end; |
e02119d5 | 586 | u64 start = key->offset; |
4bc4bee4 | 587 | u64 nbytes = 0; |
e02119d5 CM |
588 | struct btrfs_file_extent_item *item; |
589 | struct inode *inode = NULL; | |
590 | unsigned long size; | |
591 | int ret = 0; | |
592 | ||
593 | item = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item); | |
594 | found_type = btrfs_file_extent_type(eb, item); | |
595 | ||
d899e052 | 596 | if (found_type == BTRFS_FILE_EXTENT_REG || |
4bc4bee4 JB |
597 | found_type == BTRFS_FILE_EXTENT_PREALLOC) { |
598 | nbytes = btrfs_file_extent_num_bytes(eb, item); | |
599 | extent_end = start + nbytes; | |
600 | ||
601 | /* | |
602 | * We don't add to the inodes nbytes if we are prealloc or a | |
603 | * hole. | |
604 | */ | |
605 | if (btrfs_file_extent_disk_bytenr(eb, item) == 0) | |
606 | nbytes = 0; | |
607 | } else if (found_type == BTRFS_FILE_EXTENT_INLINE) { | |
514ac8ad | 608 | size = btrfs_file_extent_inline_len(eb, slot, item); |
4bc4bee4 | 609 | nbytes = btrfs_file_extent_ram_bytes(eb, item); |
fda2832f | 610 | extent_end = ALIGN(start + size, root->sectorsize); |
e02119d5 CM |
611 | } else { |
612 | ret = 0; | |
613 | goto out; | |
614 | } | |
615 | ||
616 | inode = read_one_inode(root, key->objectid); | |
617 | if (!inode) { | |
618 | ret = -EIO; | |
619 | goto out; | |
620 | } | |
621 | ||
622 | /* | |
623 | * first check to see if we already have this extent in the | |
624 | * file. This must be done before the btrfs_drop_extents run | |
625 | * so we don't try to drop this extent. | |
626 | */ | |
33345d01 | 627 | ret = btrfs_lookup_file_extent(trans, root, path, btrfs_ino(inode), |
e02119d5 CM |
628 | start, 0); |
629 | ||
d899e052 YZ |
630 | if (ret == 0 && |
631 | (found_type == BTRFS_FILE_EXTENT_REG || | |
632 | found_type == BTRFS_FILE_EXTENT_PREALLOC)) { | |
e02119d5 CM |
633 | struct btrfs_file_extent_item cmp1; |
634 | struct btrfs_file_extent_item cmp2; | |
635 | struct btrfs_file_extent_item *existing; | |
636 | struct extent_buffer *leaf; | |
637 | ||
638 | leaf = path->nodes[0]; | |
639 | existing = btrfs_item_ptr(leaf, path->slots[0], | |
640 | struct btrfs_file_extent_item); | |
641 | ||
642 | read_extent_buffer(eb, &cmp1, (unsigned long)item, | |
643 | sizeof(cmp1)); | |
644 | read_extent_buffer(leaf, &cmp2, (unsigned long)existing, | |
645 | sizeof(cmp2)); | |
646 | ||
647 | /* | |
648 | * we already have a pointer to this exact extent, | |
649 | * we don't have to do anything | |
650 | */ | |
651 | if (memcmp(&cmp1, &cmp2, sizeof(cmp1)) == 0) { | |
b3b4aa74 | 652 | btrfs_release_path(path); |
e02119d5 CM |
653 | goto out; |
654 | } | |
655 | } | |
b3b4aa74 | 656 | btrfs_release_path(path); |
e02119d5 CM |
657 | |
658 | /* drop any overlapping extents */ | |
2671485d | 659 | ret = btrfs_drop_extents(trans, root, inode, start, extent_end, 1); |
3650860b JB |
660 | if (ret) |
661 | goto out; | |
e02119d5 | 662 | |
07d400a6 YZ |
663 | if (found_type == BTRFS_FILE_EXTENT_REG || |
664 | found_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
5d4f98a2 | 665 | u64 offset; |
07d400a6 YZ |
666 | unsigned long dest_offset; |
667 | struct btrfs_key ins; | |
668 | ||
669 | ret = btrfs_insert_empty_item(trans, root, path, key, | |
670 | sizeof(*item)); | |
3650860b JB |
671 | if (ret) |
672 | goto out; | |
07d400a6 YZ |
673 | dest_offset = btrfs_item_ptr_offset(path->nodes[0], |
674 | path->slots[0]); | |
675 | copy_extent_buffer(path->nodes[0], eb, dest_offset, | |
676 | (unsigned long)item, sizeof(*item)); | |
677 | ||
678 | ins.objectid = btrfs_file_extent_disk_bytenr(eb, item); | |
679 | ins.offset = btrfs_file_extent_disk_num_bytes(eb, item); | |
680 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
5d4f98a2 | 681 | offset = key->offset - btrfs_file_extent_offset(eb, item); |
07d400a6 YZ |
682 | |
683 | if (ins.objectid > 0) { | |
684 | u64 csum_start; | |
685 | u64 csum_end; | |
686 | LIST_HEAD(ordered_sums); | |
687 | /* | |
688 | * is this extent already allocated in the extent | |
689 | * allocation tree? If so, just add a reference | |
690 | */ | |
1a4ed8fd | 691 | ret = btrfs_lookup_data_extent(root, ins.objectid, |
07d400a6 YZ |
692 | ins.offset); |
693 | if (ret == 0) { | |
694 | ret = btrfs_inc_extent_ref(trans, root, | |
695 | ins.objectid, ins.offset, | |
5d4f98a2 | 696 | 0, root->root_key.objectid, |
b06c4bf5 | 697 | key->objectid, offset); |
b50c6e25 JB |
698 | if (ret) |
699 | goto out; | |
07d400a6 YZ |
700 | } else { |
701 | /* | |
702 | * insert the extent pointer in the extent | |
703 | * allocation tree | |
704 | */ | |
5d4f98a2 YZ |
705 | ret = btrfs_alloc_logged_file_extent(trans, |
706 | root, root->root_key.objectid, | |
707 | key->objectid, offset, &ins); | |
b50c6e25 JB |
708 | if (ret) |
709 | goto out; | |
07d400a6 | 710 | } |
b3b4aa74 | 711 | btrfs_release_path(path); |
07d400a6 YZ |
712 | |
713 | if (btrfs_file_extent_compression(eb, item)) { | |
714 | csum_start = ins.objectid; | |
715 | csum_end = csum_start + ins.offset; | |
716 | } else { | |
717 | csum_start = ins.objectid + | |
718 | btrfs_file_extent_offset(eb, item); | |
719 | csum_end = csum_start + | |
720 | btrfs_file_extent_num_bytes(eb, item); | |
721 | } | |
722 | ||
723 | ret = btrfs_lookup_csums_range(root->log_root, | |
724 | csum_start, csum_end - 1, | |
a2de733c | 725 | &ordered_sums, 0); |
3650860b JB |
726 | if (ret) |
727 | goto out; | |
b84b8390 FM |
728 | /* |
729 | * Now delete all existing cums in the csum root that | |
730 | * cover our range. We do this because we can have an | |
731 | * extent that is completely referenced by one file | |
732 | * extent item and partially referenced by another | |
733 | * file extent item (like after using the clone or | |
734 | * extent_same ioctls). In this case if we end up doing | |
735 | * the replay of the one that partially references the | |
736 | * extent first, and we do not do the csum deletion | |
737 | * below, we can get 2 csum items in the csum tree that | |
738 | * overlap each other. For example, imagine our log has | |
739 | * the two following file extent items: | |
740 | * | |
741 | * key (257 EXTENT_DATA 409600) | |
742 | * extent data disk byte 12845056 nr 102400 | |
743 | * extent data offset 20480 nr 20480 ram 102400 | |
744 | * | |
745 | * key (257 EXTENT_DATA 819200) | |
746 | * extent data disk byte 12845056 nr 102400 | |
747 | * extent data offset 0 nr 102400 ram 102400 | |
748 | * | |
749 | * Where the second one fully references the 100K extent | |
750 | * that starts at disk byte 12845056, and the log tree | |
751 | * has a single csum item that covers the entire range | |
752 | * of the extent: | |
753 | * | |
754 | * key (EXTENT_CSUM EXTENT_CSUM 12845056) itemsize 100 | |
755 | * | |
756 | * After the first file extent item is replayed, the | |
757 | * csum tree gets the following csum item: | |
758 | * | |
759 | * key (EXTENT_CSUM EXTENT_CSUM 12865536) itemsize 20 | |
760 | * | |
761 | * Which covers the 20K sub-range starting at offset 20K | |
762 | * of our extent. Now when we replay the second file | |
763 | * extent item, if we do not delete existing csum items | |
764 | * that cover any of its blocks, we end up getting two | |
765 | * csum items in our csum tree that overlap each other: | |
766 | * | |
767 | * key (EXTENT_CSUM EXTENT_CSUM 12845056) itemsize 100 | |
768 | * key (EXTENT_CSUM EXTENT_CSUM 12865536) itemsize 20 | |
769 | * | |
770 | * Which is a problem, because after this anyone trying | |
771 | * to lookup up for the checksum of any block of our | |
772 | * extent starting at an offset of 40K or higher, will | |
773 | * end up looking at the second csum item only, which | |
774 | * does not contain the checksum for any block starting | |
775 | * at offset 40K or higher of our extent. | |
776 | */ | |
07d400a6 YZ |
777 | while (!list_empty(&ordered_sums)) { |
778 | struct btrfs_ordered_sum *sums; | |
779 | sums = list_entry(ordered_sums.next, | |
780 | struct btrfs_ordered_sum, | |
781 | list); | |
b84b8390 FM |
782 | if (!ret) |
783 | ret = btrfs_del_csums(trans, | |
784 | root->fs_info->csum_root, | |
785 | sums->bytenr, | |
786 | sums->len); | |
3650860b JB |
787 | if (!ret) |
788 | ret = btrfs_csum_file_blocks(trans, | |
07d400a6 YZ |
789 | root->fs_info->csum_root, |
790 | sums); | |
07d400a6 YZ |
791 | list_del(&sums->list); |
792 | kfree(sums); | |
793 | } | |
3650860b JB |
794 | if (ret) |
795 | goto out; | |
07d400a6 | 796 | } else { |
b3b4aa74 | 797 | btrfs_release_path(path); |
07d400a6 YZ |
798 | } |
799 | } else if (found_type == BTRFS_FILE_EXTENT_INLINE) { | |
800 | /* inline extents are easy, we just overwrite them */ | |
801 | ret = overwrite_item(trans, root, path, eb, slot, key); | |
3650860b JB |
802 | if (ret) |
803 | goto out; | |
07d400a6 | 804 | } |
e02119d5 | 805 | |
4bc4bee4 | 806 | inode_add_bytes(inode, nbytes); |
b9959295 | 807 | ret = btrfs_update_inode(trans, root, inode); |
e02119d5 CM |
808 | out: |
809 | if (inode) | |
810 | iput(inode); | |
811 | return ret; | |
812 | } | |
813 | ||
814 | /* | |
815 | * when cleaning up conflicts between the directory names in the | |
816 | * subvolume, directory names in the log and directory names in the | |
817 | * inode back references, we may have to unlink inodes from directories. | |
818 | * | |
819 | * This is a helper function to do the unlink of a specific directory | |
820 | * item | |
821 | */ | |
822 | static noinline int drop_one_dir_item(struct btrfs_trans_handle *trans, | |
823 | struct btrfs_root *root, | |
824 | struct btrfs_path *path, | |
825 | struct inode *dir, | |
826 | struct btrfs_dir_item *di) | |
827 | { | |
828 | struct inode *inode; | |
829 | char *name; | |
830 | int name_len; | |
831 | struct extent_buffer *leaf; | |
832 | struct btrfs_key location; | |
833 | int ret; | |
834 | ||
835 | leaf = path->nodes[0]; | |
836 | ||
837 | btrfs_dir_item_key_to_cpu(leaf, di, &location); | |
838 | name_len = btrfs_dir_name_len(leaf, di); | |
839 | name = kmalloc(name_len, GFP_NOFS); | |
2a29edc6 | 840 | if (!name) |
841 | return -ENOMEM; | |
842 | ||
e02119d5 | 843 | read_extent_buffer(leaf, name, (unsigned long)(di + 1), name_len); |
b3b4aa74 | 844 | btrfs_release_path(path); |
e02119d5 CM |
845 | |
846 | inode = read_one_inode(root, location.objectid); | |
c00e9493 | 847 | if (!inode) { |
3650860b JB |
848 | ret = -EIO; |
849 | goto out; | |
c00e9493 | 850 | } |
e02119d5 | 851 | |
ec051c0f | 852 | ret = link_to_fixup_dir(trans, root, path, location.objectid); |
3650860b JB |
853 | if (ret) |
854 | goto out; | |
12fcfd22 | 855 | |
e02119d5 | 856 | ret = btrfs_unlink_inode(trans, root, dir, inode, name, name_len); |
3650860b JB |
857 | if (ret) |
858 | goto out; | |
ada9af21 FDBM |
859 | else |
860 | ret = btrfs_run_delayed_items(trans, root); | |
3650860b | 861 | out: |
e02119d5 | 862 | kfree(name); |
e02119d5 CM |
863 | iput(inode); |
864 | return ret; | |
865 | } | |
866 | ||
867 | /* | |
868 | * helper function to see if a given name and sequence number found | |
869 | * in an inode back reference are already in a directory and correctly | |
870 | * point to this inode | |
871 | */ | |
872 | static noinline int inode_in_dir(struct btrfs_root *root, | |
873 | struct btrfs_path *path, | |
874 | u64 dirid, u64 objectid, u64 index, | |
875 | const char *name, int name_len) | |
876 | { | |
877 | struct btrfs_dir_item *di; | |
878 | struct btrfs_key location; | |
879 | int match = 0; | |
880 | ||
881 | di = btrfs_lookup_dir_index_item(NULL, root, path, dirid, | |
882 | index, name, name_len, 0); | |
883 | if (di && !IS_ERR(di)) { | |
884 | btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location); | |
885 | if (location.objectid != objectid) | |
886 | goto out; | |
887 | } else | |
888 | goto out; | |
b3b4aa74 | 889 | btrfs_release_path(path); |
e02119d5 CM |
890 | |
891 | di = btrfs_lookup_dir_item(NULL, root, path, dirid, name, name_len, 0); | |
892 | if (di && !IS_ERR(di)) { | |
893 | btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location); | |
894 | if (location.objectid != objectid) | |
895 | goto out; | |
896 | } else | |
897 | goto out; | |
898 | match = 1; | |
899 | out: | |
b3b4aa74 | 900 | btrfs_release_path(path); |
e02119d5 CM |
901 | return match; |
902 | } | |
903 | ||
904 | /* | |
905 | * helper function to check a log tree for a named back reference in | |
906 | * an inode. This is used to decide if a back reference that is | |
907 | * found in the subvolume conflicts with what we find in the log. | |
908 | * | |
909 | * inode backreferences may have multiple refs in a single item, | |
910 | * during replay we process one reference at a time, and we don't | |
911 | * want to delete valid links to a file from the subvolume if that | |
912 | * link is also in the log. | |
913 | */ | |
914 | static noinline int backref_in_log(struct btrfs_root *log, | |
915 | struct btrfs_key *key, | |
f186373f | 916 | u64 ref_objectid, |
df8d116f | 917 | const char *name, int namelen) |
e02119d5 CM |
918 | { |
919 | struct btrfs_path *path; | |
920 | struct btrfs_inode_ref *ref; | |
921 | unsigned long ptr; | |
922 | unsigned long ptr_end; | |
923 | unsigned long name_ptr; | |
924 | int found_name_len; | |
925 | int item_size; | |
926 | int ret; | |
927 | int match = 0; | |
928 | ||
929 | path = btrfs_alloc_path(); | |
2a29edc6 | 930 | if (!path) |
931 | return -ENOMEM; | |
932 | ||
e02119d5 CM |
933 | ret = btrfs_search_slot(NULL, log, key, path, 0, 0); |
934 | if (ret != 0) | |
935 | goto out; | |
936 | ||
e02119d5 | 937 | ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]); |
f186373f MF |
938 | |
939 | if (key->type == BTRFS_INODE_EXTREF_KEY) { | |
940 | if (btrfs_find_name_in_ext_backref(path, ref_objectid, | |
941 | name, namelen, NULL)) | |
942 | match = 1; | |
943 | ||
944 | goto out; | |
945 | } | |
946 | ||
947 | item_size = btrfs_item_size_nr(path->nodes[0], path->slots[0]); | |
e02119d5 CM |
948 | ptr_end = ptr + item_size; |
949 | while (ptr < ptr_end) { | |
950 | ref = (struct btrfs_inode_ref *)ptr; | |
951 | found_name_len = btrfs_inode_ref_name_len(path->nodes[0], ref); | |
952 | if (found_name_len == namelen) { | |
953 | name_ptr = (unsigned long)(ref + 1); | |
954 | ret = memcmp_extent_buffer(path->nodes[0], name, | |
955 | name_ptr, namelen); | |
956 | if (ret == 0) { | |
957 | match = 1; | |
958 | goto out; | |
959 | } | |
960 | } | |
961 | ptr = (unsigned long)(ref + 1) + found_name_len; | |
962 | } | |
963 | out: | |
964 | btrfs_free_path(path); | |
965 | return match; | |
966 | } | |
967 | ||
5a1d7843 | 968 | static inline int __add_inode_ref(struct btrfs_trans_handle *trans, |
e02119d5 | 969 | struct btrfs_root *root, |
e02119d5 | 970 | struct btrfs_path *path, |
5a1d7843 JS |
971 | struct btrfs_root *log_root, |
972 | struct inode *dir, struct inode *inode, | |
5a1d7843 | 973 | struct extent_buffer *eb, |
f186373f MF |
974 | u64 inode_objectid, u64 parent_objectid, |
975 | u64 ref_index, char *name, int namelen, | |
976 | int *search_done) | |
e02119d5 | 977 | { |
34f3e4f2 | 978 | int ret; |
f186373f MF |
979 | char *victim_name; |
980 | int victim_name_len; | |
981 | struct extent_buffer *leaf; | |
5a1d7843 | 982 | struct btrfs_dir_item *di; |
f186373f MF |
983 | struct btrfs_key search_key; |
984 | struct btrfs_inode_extref *extref; | |
c622ae60 | 985 | |
f186373f MF |
986 | again: |
987 | /* Search old style refs */ | |
988 | search_key.objectid = inode_objectid; | |
989 | search_key.type = BTRFS_INODE_REF_KEY; | |
990 | search_key.offset = parent_objectid; | |
991 | ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0); | |
e02119d5 | 992 | if (ret == 0) { |
e02119d5 CM |
993 | struct btrfs_inode_ref *victim_ref; |
994 | unsigned long ptr; | |
995 | unsigned long ptr_end; | |
f186373f MF |
996 | |
997 | leaf = path->nodes[0]; | |
e02119d5 CM |
998 | |
999 | /* are we trying to overwrite a back ref for the root directory | |
1000 | * if so, just jump out, we're done | |
1001 | */ | |
f186373f | 1002 | if (search_key.objectid == search_key.offset) |
5a1d7843 | 1003 | return 1; |
e02119d5 CM |
1004 | |
1005 | /* check all the names in this back reference to see | |
1006 | * if they are in the log. if so, we allow them to stay | |
1007 | * otherwise they must be unlinked as a conflict | |
1008 | */ | |
1009 | ptr = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
1010 | ptr_end = ptr + btrfs_item_size_nr(leaf, path->slots[0]); | |
d397712b | 1011 | while (ptr < ptr_end) { |
e02119d5 CM |
1012 | victim_ref = (struct btrfs_inode_ref *)ptr; |
1013 | victim_name_len = btrfs_inode_ref_name_len(leaf, | |
1014 | victim_ref); | |
1015 | victim_name = kmalloc(victim_name_len, GFP_NOFS); | |
3650860b JB |
1016 | if (!victim_name) |
1017 | return -ENOMEM; | |
e02119d5 CM |
1018 | |
1019 | read_extent_buffer(leaf, victim_name, | |
1020 | (unsigned long)(victim_ref + 1), | |
1021 | victim_name_len); | |
1022 | ||
f186373f MF |
1023 | if (!backref_in_log(log_root, &search_key, |
1024 | parent_objectid, | |
1025 | victim_name, | |
e02119d5 | 1026 | victim_name_len)) { |
8b558c5f | 1027 | inc_nlink(inode); |
b3b4aa74 | 1028 | btrfs_release_path(path); |
12fcfd22 | 1029 | |
e02119d5 CM |
1030 | ret = btrfs_unlink_inode(trans, root, dir, |
1031 | inode, victim_name, | |
1032 | victim_name_len); | |
f186373f | 1033 | kfree(victim_name); |
3650860b JB |
1034 | if (ret) |
1035 | return ret; | |
ada9af21 FDBM |
1036 | ret = btrfs_run_delayed_items(trans, root); |
1037 | if (ret) | |
1038 | return ret; | |
f186373f MF |
1039 | *search_done = 1; |
1040 | goto again; | |
e02119d5 CM |
1041 | } |
1042 | kfree(victim_name); | |
f186373f | 1043 | |
e02119d5 CM |
1044 | ptr = (unsigned long)(victim_ref + 1) + victim_name_len; |
1045 | } | |
e02119d5 | 1046 | |
c622ae60 | 1047 | /* |
1048 | * NOTE: we have searched root tree and checked the | |
1049 | * coresponding ref, it does not need to check again. | |
1050 | */ | |
5a1d7843 | 1051 | *search_done = 1; |
e02119d5 | 1052 | } |
b3b4aa74 | 1053 | btrfs_release_path(path); |
e02119d5 | 1054 | |
f186373f MF |
1055 | /* Same search but for extended refs */ |
1056 | extref = btrfs_lookup_inode_extref(NULL, root, path, name, namelen, | |
1057 | inode_objectid, parent_objectid, 0, | |
1058 | 0); | |
1059 | if (!IS_ERR_OR_NULL(extref)) { | |
1060 | u32 item_size; | |
1061 | u32 cur_offset = 0; | |
1062 | unsigned long base; | |
1063 | struct inode *victim_parent; | |
1064 | ||
1065 | leaf = path->nodes[0]; | |
1066 | ||
1067 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1068 | base = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
1069 | ||
1070 | while (cur_offset < item_size) { | |
dd9ef135 | 1071 | extref = (struct btrfs_inode_extref *)(base + cur_offset); |
f186373f MF |
1072 | |
1073 | victim_name_len = btrfs_inode_extref_name_len(leaf, extref); | |
1074 | ||
1075 | if (btrfs_inode_extref_parent(leaf, extref) != parent_objectid) | |
1076 | goto next; | |
1077 | ||
1078 | victim_name = kmalloc(victim_name_len, GFP_NOFS); | |
3650860b JB |
1079 | if (!victim_name) |
1080 | return -ENOMEM; | |
f186373f MF |
1081 | read_extent_buffer(leaf, victim_name, (unsigned long)&extref->name, |
1082 | victim_name_len); | |
1083 | ||
1084 | search_key.objectid = inode_objectid; | |
1085 | search_key.type = BTRFS_INODE_EXTREF_KEY; | |
1086 | search_key.offset = btrfs_extref_hash(parent_objectid, | |
1087 | victim_name, | |
1088 | victim_name_len); | |
1089 | ret = 0; | |
1090 | if (!backref_in_log(log_root, &search_key, | |
1091 | parent_objectid, victim_name, | |
1092 | victim_name_len)) { | |
1093 | ret = -ENOENT; | |
1094 | victim_parent = read_one_inode(root, | |
1095 | parent_objectid); | |
1096 | if (victim_parent) { | |
8b558c5f | 1097 | inc_nlink(inode); |
f186373f MF |
1098 | btrfs_release_path(path); |
1099 | ||
1100 | ret = btrfs_unlink_inode(trans, root, | |
1101 | victim_parent, | |
1102 | inode, | |
1103 | victim_name, | |
1104 | victim_name_len); | |
ada9af21 FDBM |
1105 | if (!ret) |
1106 | ret = btrfs_run_delayed_items( | |
1107 | trans, root); | |
f186373f | 1108 | } |
f186373f MF |
1109 | iput(victim_parent); |
1110 | kfree(victim_name); | |
3650860b JB |
1111 | if (ret) |
1112 | return ret; | |
f186373f MF |
1113 | *search_done = 1; |
1114 | goto again; | |
1115 | } | |
1116 | kfree(victim_name); | |
3650860b JB |
1117 | if (ret) |
1118 | return ret; | |
f186373f MF |
1119 | next: |
1120 | cur_offset += victim_name_len + sizeof(*extref); | |
1121 | } | |
1122 | *search_done = 1; | |
1123 | } | |
1124 | btrfs_release_path(path); | |
1125 | ||
34f3e4f2 | 1126 | /* look for a conflicting sequence number */ |
1127 | di = btrfs_lookup_dir_index_item(trans, root, path, btrfs_ino(dir), | |
f186373f | 1128 | ref_index, name, namelen, 0); |
34f3e4f2 | 1129 | if (di && !IS_ERR(di)) { |
1130 | ret = drop_one_dir_item(trans, root, path, dir, di); | |
3650860b JB |
1131 | if (ret) |
1132 | return ret; | |
34f3e4f2 | 1133 | } |
1134 | btrfs_release_path(path); | |
1135 | ||
1136 | /* look for a conflicing name */ | |
1137 | di = btrfs_lookup_dir_item(trans, root, path, btrfs_ino(dir), | |
1138 | name, namelen, 0); | |
1139 | if (di && !IS_ERR(di)) { | |
1140 | ret = drop_one_dir_item(trans, root, path, dir, di); | |
3650860b JB |
1141 | if (ret) |
1142 | return ret; | |
34f3e4f2 | 1143 | } |
1144 | btrfs_release_path(path); | |
1145 | ||
5a1d7843 JS |
1146 | return 0; |
1147 | } | |
e02119d5 | 1148 | |
f186373f MF |
1149 | static int extref_get_fields(struct extent_buffer *eb, unsigned long ref_ptr, |
1150 | u32 *namelen, char **name, u64 *index, | |
1151 | u64 *parent_objectid) | |
1152 | { | |
1153 | struct btrfs_inode_extref *extref; | |
1154 | ||
1155 | extref = (struct btrfs_inode_extref *)ref_ptr; | |
1156 | ||
1157 | *namelen = btrfs_inode_extref_name_len(eb, extref); | |
1158 | *name = kmalloc(*namelen, GFP_NOFS); | |
1159 | if (*name == NULL) | |
1160 | return -ENOMEM; | |
1161 | ||
1162 | read_extent_buffer(eb, *name, (unsigned long)&extref->name, | |
1163 | *namelen); | |
1164 | ||
1165 | *index = btrfs_inode_extref_index(eb, extref); | |
1166 | if (parent_objectid) | |
1167 | *parent_objectid = btrfs_inode_extref_parent(eb, extref); | |
1168 | ||
1169 | return 0; | |
1170 | } | |
1171 | ||
1172 | static int ref_get_fields(struct extent_buffer *eb, unsigned long ref_ptr, | |
1173 | u32 *namelen, char **name, u64 *index) | |
1174 | { | |
1175 | struct btrfs_inode_ref *ref; | |
1176 | ||
1177 | ref = (struct btrfs_inode_ref *)ref_ptr; | |
1178 | ||
1179 | *namelen = btrfs_inode_ref_name_len(eb, ref); | |
1180 | *name = kmalloc(*namelen, GFP_NOFS); | |
1181 | if (*name == NULL) | |
1182 | return -ENOMEM; | |
1183 | ||
1184 | read_extent_buffer(eb, *name, (unsigned long)(ref + 1), *namelen); | |
1185 | ||
1186 | *index = btrfs_inode_ref_index(eb, ref); | |
1187 | ||
1188 | return 0; | |
1189 | } | |
1190 | ||
5a1d7843 JS |
1191 | /* |
1192 | * replay one inode back reference item found in the log tree. | |
1193 | * eb, slot and key refer to the buffer and key found in the log tree. | |
1194 | * root is the destination we are replaying into, and path is for temp | |
1195 | * use by this function. (it should be released on return). | |
1196 | */ | |
1197 | static noinline int add_inode_ref(struct btrfs_trans_handle *trans, | |
1198 | struct btrfs_root *root, | |
1199 | struct btrfs_root *log, | |
1200 | struct btrfs_path *path, | |
1201 | struct extent_buffer *eb, int slot, | |
1202 | struct btrfs_key *key) | |
1203 | { | |
03b2f08b GB |
1204 | struct inode *dir = NULL; |
1205 | struct inode *inode = NULL; | |
5a1d7843 JS |
1206 | unsigned long ref_ptr; |
1207 | unsigned long ref_end; | |
03b2f08b | 1208 | char *name = NULL; |
5a1d7843 JS |
1209 | int namelen; |
1210 | int ret; | |
1211 | int search_done = 0; | |
f186373f MF |
1212 | int log_ref_ver = 0; |
1213 | u64 parent_objectid; | |
1214 | u64 inode_objectid; | |
f46dbe3d | 1215 | u64 ref_index = 0; |
f186373f MF |
1216 | int ref_struct_size; |
1217 | ||
1218 | ref_ptr = btrfs_item_ptr_offset(eb, slot); | |
1219 | ref_end = ref_ptr + btrfs_item_size_nr(eb, slot); | |
1220 | ||
1221 | if (key->type == BTRFS_INODE_EXTREF_KEY) { | |
1222 | struct btrfs_inode_extref *r; | |
1223 | ||
1224 | ref_struct_size = sizeof(struct btrfs_inode_extref); | |
1225 | log_ref_ver = 1; | |
1226 | r = (struct btrfs_inode_extref *)ref_ptr; | |
1227 | parent_objectid = btrfs_inode_extref_parent(eb, r); | |
1228 | } else { | |
1229 | ref_struct_size = sizeof(struct btrfs_inode_ref); | |
1230 | parent_objectid = key->offset; | |
1231 | } | |
1232 | inode_objectid = key->objectid; | |
e02119d5 | 1233 | |
5a1d7843 JS |
1234 | /* |
1235 | * it is possible that we didn't log all the parent directories | |
1236 | * for a given inode. If we don't find the dir, just don't | |
1237 | * copy the back ref in. The link count fixup code will take | |
1238 | * care of the rest | |
1239 | */ | |
f186373f | 1240 | dir = read_one_inode(root, parent_objectid); |
03b2f08b GB |
1241 | if (!dir) { |
1242 | ret = -ENOENT; | |
1243 | goto out; | |
1244 | } | |
5a1d7843 | 1245 | |
f186373f | 1246 | inode = read_one_inode(root, inode_objectid); |
5a1d7843 | 1247 | if (!inode) { |
03b2f08b GB |
1248 | ret = -EIO; |
1249 | goto out; | |
5a1d7843 JS |
1250 | } |
1251 | ||
5a1d7843 | 1252 | while (ref_ptr < ref_end) { |
f186373f MF |
1253 | if (log_ref_ver) { |
1254 | ret = extref_get_fields(eb, ref_ptr, &namelen, &name, | |
1255 | &ref_index, &parent_objectid); | |
1256 | /* | |
1257 | * parent object can change from one array | |
1258 | * item to another. | |
1259 | */ | |
1260 | if (!dir) | |
1261 | dir = read_one_inode(root, parent_objectid); | |
03b2f08b GB |
1262 | if (!dir) { |
1263 | ret = -ENOENT; | |
1264 | goto out; | |
1265 | } | |
f186373f MF |
1266 | } else { |
1267 | ret = ref_get_fields(eb, ref_ptr, &namelen, &name, | |
1268 | &ref_index); | |
1269 | } | |
1270 | if (ret) | |
03b2f08b | 1271 | goto out; |
5a1d7843 JS |
1272 | |
1273 | /* if we already have a perfect match, we're done */ | |
1274 | if (!inode_in_dir(root, path, btrfs_ino(dir), btrfs_ino(inode), | |
f186373f | 1275 | ref_index, name, namelen)) { |
5a1d7843 JS |
1276 | /* |
1277 | * look for a conflicting back reference in the | |
1278 | * metadata. if we find one we have to unlink that name | |
1279 | * of the file before we add our new link. Later on, we | |
1280 | * overwrite any existing back reference, and we don't | |
1281 | * want to create dangling pointers in the directory. | |
1282 | */ | |
1283 | ||
1284 | if (!search_done) { | |
1285 | ret = __add_inode_ref(trans, root, path, log, | |
f186373f MF |
1286 | dir, inode, eb, |
1287 | inode_objectid, | |
1288 | parent_objectid, | |
1289 | ref_index, name, namelen, | |
5a1d7843 | 1290 | &search_done); |
03b2f08b GB |
1291 | if (ret) { |
1292 | if (ret == 1) | |
1293 | ret = 0; | |
3650860b JB |
1294 | goto out; |
1295 | } | |
5a1d7843 JS |
1296 | } |
1297 | ||
1298 | /* insert our name */ | |
1299 | ret = btrfs_add_link(trans, dir, inode, name, namelen, | |
f186373f | 1300 | 0, ref_index); |
3650860b JB |
1301 | if (ret) |
1302 | goto out; | |
5a1d7843 JS |
1303 | |
1304 | btrfs_update_inode(trans, root, inode); | |
1305 | } | |
1306 | ||
f186373f | 1307 | ref_ptr = (unsigned long)(ref_ptr + ref_struct_size) + namelen; |
5a1d7843 | 1308 | kfree(name); |
03b2f08b | 1309 | name = NULL; |
f186373f MF |
1310 | if (log_ref_ver) { |
1311 | iput(dir); | |
1312 | dir = NULL; | |
1313 | } | |
5a1d7843 | 1314 | } |
e02119d5 CM |
1315 | |
1316 | /* finally write the back reference in the inode */ | |
1317 | ret = overwrite_item(trans, root, path, eb, slot, key); | |
5a1d7843 | 1318 | out: |
b3b4aa74 | 1319 | btrfs_release_path(path); |
03b2f08b | 1320 | kfree(name); |
e02119d5 CM |
1321 | iput(dir); |
1322 | iput(inode); | |
3650860b | 1323 | return ret; |
e02119d5 CM |
1324 | } |
1325 | ||
c71bf099 | 1326 | static int insert_orphan_item(struct btrfs_trans_handle *trans, |
9c4f61f0 | 1327 | struct btrfs_root *root, u64 ino) |
c71bf099 YZ |
1328 | { |
1329 | int ret; | |
381cf658 | 1330 | |
9c4f61f0 DS |
1331 | ret = btrfs_insert_orphan_item(trans, root, ino); |
1332 | if (ret == -EEXIST) | |
1333 | ret = 0; | |
381cf658 | 1334 | |
c71bf099 YZ |
1335 | return ret; |
1336 | } | |
1337 | ||
f186373f MF |
1338 | static int count_inode_extrefs(struct btrfs_root *root, |
1339 | struct inode *inode, struct btrfs_path *path) | |
1340 | { | |
1341 | int ret = 0; | |
1342 | int name_len; | |
1343 | unsigned int nlink = 0; | |
1344 | u32 item_size; | |
1345 | u32 cur_offset = 0; | |
1346 | u64 inode_objectid = btrfs_ino(inode); | |
1347 | u64 offset = 0; | |
1348 | unsigned long ptr; | |
1349 | struct btrfs_inode_extref *extref; | |
1350 | struct extent_buffer *leaf; | |
1351 | ||
1352 | while (1) { | |
1353 | ret = btrfs_find_one_extref(root, inode_objectid, offset, path, | |
1354 | &extref, &offset); | |
1355 | if (ret) | |
1356 | break; | |
c71bf099 | 1357 | |
f186373f MF |
1358 | leaf = path->nodes[0]; |
1359 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1360 | ptr = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
2c2c452b | 1361 | cur_offset = 0; |
f186373f MF |
1362 | |
1363 | while (cur_offset < item_size) { | |
1364 | extref = (struct btrfs_inode_extref *) (ptr + cur_offset); | |
1365 | name_len = btrfs_inode_extref_name_len(leaf, extref); | |
1366 | ||
1367 | nlink++; | |
1368 | ||
1369 | cur_offset += name_len + sizeof(*extref); | |
1370 | } | |
1371 | ||
1372 | offset++; | |
1373 | btrfs_release_path(path); | |
1374 | } | |
1375 | btrfs_release_path(path); | |
1376 | ||
2c2c452b | 1377 | if (ret < 0 && ret != -ENOENT) |
f186373f MF |
1378 | return ret; |
1379 | return nlink; | |
1380 | } | |
1381 | ||
1382 | static int count_inode_refs(struct btrfs_root *root, | |
1383 | struct inode *inode, struct btrfs_path *path) | |
e02119d5 | 1384 | { |
e02119d5 CM |
1385 | int ret; |
1386 | struct btrfs_key key; | |
f186373f | 1387 | unsigned int nlink = 0; |
e02119d5 CM |
1388 | unsigned long ptr; |
1389 | unsigned long ptr_end; | |
1390 | int name_len; | |
33345d01 | 1391 | u64 ino = btrfs_ino(inode); |
e02119d5 | 1392 | |
33345d01 | 1393 | key.objectid = ino; |
e02119d5 CM |
1394 | key.type = BTRFS_INODE_REF_KEY; |
1395 | key.offset = (u64)-1; | |
1396 | ||
d397712b | 1397 | while (1) { |
e02119d5 CM |
1398 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
1399 | if (ret < 0) | |
1400 | break; | |
1401 | if (ret > 0) { | |
1402 | if (path->slots[0] == 0) | |
1403 | break; | |
1404 | path->slots[0]--; | |
1405 | } | |
e93ae26f | 1406 | process_slot: |
e02119d5 CM |
1407 | btrfs_item_key_to_cpu(path->nodes[0], &key, |
1408 | path->slots[0]); | |
33345d01 | 1409 | if (key.objectid != ino || |
e02119d5 CM |
1410 | key.type != BTRFS_INODE_REF_KEY) |
1411 | break; | |
1412 | ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]); | |
1413 | ptr_end = ptr + btrfs_item_size_nr(path->nodes[0], | |
1414 | path->slots[0]); | |
d397712b | 1415 | while (ptr < ptr_end) { |
e02119d5 CM |
1416 | struct btrfs_inode_ref *ref; |
1417 | ||
1418 | ref = (struct btrfs_inode_ref *)ptr; | |
1419 | name_len = btrfs_inode_ref_name_len(path->nodes[0], | |
1420 | ref); | |
1421 | ptr = (unsigned long)(ref + 1) + name_len; | |
1422 | nlink++; | |
1423 | } | |
1424 | ||
1425 | if (key.offset == 0) | |
1426 | break; | |
e93ae26f FDBM |
1427 | if (path->slots[0] > 0) { |
1428 | path->slots[0]--; | |
1429 | goto process_slot; | |
1430 | } | |
e02119d5 | 1431 | key.offset--; |
b3b4aa74 | 1432 | btrfs_release_path(path); |
e02119d5 | 1433 | } |
b3b4aa74 | 1434 | btrfs_release_path(path); |
f186373f MF |
1435 | |
1436 | return nlink; | |
1437 | } | |
1438 | ||
1439 | /* | |
1440 | * There are a few corners where the link count of the file can't | |
1441 | * be properly maintained during replay. So, instead of adding | |
1442 | * lots of complexity to the log code, we just scan the backrefs | |
1443 | * for any file that has been through replay. | |
1444 | * | |
1445 | * The scan will update the link count on the inode to reflect the | |
1446 | * number of back refs found. If it goes down to zero, the iput | |
1447 | * will free the inode. | |
1448 | */ | |
1449 | static noinline int fixup_inode_link_count(struct btrfs_trans_handle *trans, | |
1450 | struct btrfs_root *root, | |
1451 | struct inode *inode) | |
1452 | { | |
1453 | struct btrfs_path *path; | |
1454 | int ret; | |
1455 | u64 nlink = 0; | |
1456 | u64 ino = btrfs_ino(inode); | |
1457 | ||
1458 | path = btrfs_alloc_path(); | |
1459 | if (!path) | |
1460 | return -ENOMEM; | |
1461 | ||
1462 | ret = count_inode_refs(root, inode, path); | |
1463 | if (ret < 0) | |
1464 | goto out; | |
1465 | ||
1466 | nlink = ret; | |
1467 | ||
1468 | ret = count_inode_extrefs(root, inode, path); | |
f186373f MF |
1469 | if (ret < 0) |
1470 | goto out; | |
1471 | ||
1472 | nlink += ret; | |
1473 | ||
1474 | ret = 0; | |
1475 | ||
e02119d5 | 1476 | if (nlink != inode->i_nlink) { |
bfe86848 | 1477 | set_nlink(inode, nlink); |
e02119d5 CM |
1478 | btrfs_update_inode(trans, root, inode); |
1479 | } | |
8d5bf1cb | 1480 | BTRFS_I(inode)->index_cnt = (u64)-1; |
e02119d5 | 1481 | |
c71bf099 YZ |
1482 | if (inode->i_nlink == 0) { |
1483 | if (S_ISDIR(inode->i_mode)) { | |
1484 | ret = replay_dir_deletes(trans, root, NULL, path, | |
33345d01 | 1485 | ino, 1); |
3650860b JB |
1486 | if (ret) |
1487 | goto out; | |
c71bf099 | 1488 | } |
33345d01 | 1489 | ret = insert_orphan_item(trans, root, ino); |
12fcfd22 | 1490 | } |
12fcfd22 | 1491 | |
f186373f MF |
1492 | out: |
1493 | btrfs_free_path(path); | |
1494 | return ret; | |
e02119d5 CM |
1495 | } |
1496 | ||
1497 | static noinline int fixup_inode_link_counts(struct btrfs_trans_handle *trans, | |
1498 | struct btrfs_root *root, | |
1499 | struct btrfs_path *path) | |
1500 | { | |
1501 | int ret; | |
1502 | struct btrfs_key key; | |
1503 | struct inode *inode; | |
1504 | ||
1505 | key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID; | |
1506 | key.type = BTRFS_ORPHAN_ITEM_KEY; | |
1507 | key.offset = (u64)-1; | |
d397712b | 1508 | while (1) { |
e02119d5 CM |
1509 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
1510 | if (ret < 0) | |
1511 | break; | |
1512 | ||
1513 | if (ret == 1) { | |
1514 | if (path->slots[0] == 0) | |
1515 | break; | |
1516 | path->slots[0]--; | |
1517 | } | |
1518 | ||
1519 | btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]); | |
1520 | if (key.objectid != BTRFS_TREE_LOG_FIXUP_OBJECTID || | |
1521 | key.type != BTRFS_ORPHAN_ITEM_KEY) | |
1522 | break; | |
1523 | ||
1524 | ret = btrfs_del_item(trans, root, path); | |
65a246c5 TI |
1525 | if (ret) |
1526 | goto out; | |
e02119d5 | 1527 | |
b3b4aa74 | 1528 | btrfs_release_path(path); |
e02119d5 | 1529 | inode = read_one_inode(root, key.offset); |
c00e9493 TI |
1530 | if (!inode) |
1531 | return -EIO; | |
e02119d5 CM |
1532 | |
1533 | ret = fixup_inode_link_count(trans, root, inode); | |
e02119d5 | 1534 | iput(inode); |
3650860b JB |
1535 | if (ret) |
1536 | goto out; | |
e02119d5 | 1537 | |
12fcfd22 CM |
1538 | /* |
1539 | * fixup on a directory may create new entries, | |
1540 | * make sure we always look for the highset possible | |
1541 | * offset | |
1542 | */ | |
1543 | key.offset = (u64)-1; | |
e02119d5 | 1544 | } |
65a246c5 TI |
1545 | ret = 0; |
1546 | out: | |
b3b4aa74 | 1547 | btrfs_release_path(path); |
65a246c5 | 1548 | return ret; |
e02119d5 CM |
1549 | } |
1550 | ||
1551 | ||
1552 | /* | |
1553 | * record a given inode in the fixup dir so we can check its link | |
1554 | * count when replay is done. The link count is incremented here | |
1555 | * so the inode won't go away until we check it | |
1556 | */ | |
1557 | static noinline int link_to_fixup_dir(struct btrfs_trans_handle *trans, | |
1558 | struct btrfs_root *root, | |
1559 | struct btrfs_path *path, | |
1560 | u64 objectid) | |
1561 | { | |
1562 | struct btrfs_key key; | |
1563 | int ret = 0; | |
1564 | struct inode *inode; | |
1565 | ||
1566 | inode = read_one_inode(root, objectid); | |
c00e9493 TI |
1567 | if (!inode) |
1568 | return -EIO; | |
e02119d5 CM |
1569 | |
1570 | key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID; | |
962a298f | 1571 | key.type = BTRFS_ORPHAN_ITEM_KEY; |
e02119d5 CM |
1572 | key.offset = objectid; |
1573 | ||
1574 | ret = btrfs_insert_empty_item(trans, root, path, &key, 0); | |
1575 | ||
b3b4aa74 | 1576 | btrfs_release_path(path); |
e02119d5 | 1577 | if (ret == 0) { |
9bf7a489 JB |
1578 | if (!inode->i_nlink) |
1579 | set_nlink(inode, 1); | |
1580 | else | |
8b558c5f | 1581 | inc_nlink(inode); |
b9959295 | 1582 | ret = btrfs_update_inode(trans, root, inode); |
e02119d5 CM |
1583 | } else if (ret == -EEXIST) { |
1584 | ret = 0; | |
1585 | } else { | |
3650860b | 1586 | BUG(); /* Logic Error */ |
e02119d5 CM |
1587 | } |
1588 | iput(inode); | |
1589 | ||
1590 | return ret; | |
1591 | } | |
1592 | ||
1593 | /* | |
1594 | * when replaying the log for a directory, we only insert names | |
1595 | * for inodes that actually exist. This means an fsync on a directory | |
1596 | * does not implicitly fsync all the new files in it | |
1597 | */ | |
1598 | static noinline int insert_one_name(struct btrfs_trans_handle *trans, | |
1599 | struct btrfs_root *root, | |
e02119d5 | 1600 | u64 dirid, u64 index, |
60d53eb3 | 1601 | char *name, int name_len, |
e02119d5 CM |
1602 | struct btrfs_key *location) |
1603 | { | |
1604 | struct inode *inode; | |
1605 | struct inode *dir; | |
1606 | int ret; | |
1607 | ||
1608 | inode = read_one_inode(root, location->objectid); | |
1609 | if (!inode) | |
1610 | return -ENOENT; | |
1611 | ||
1612 | dir = read_one_inode(root, dirid); | |
1613 | if (!dir) { | |
1614 | iput(inode); | |
1615 | return -EIO; | |
1616 | } | |
d555438b | 1617 | |
e02119d5 CM |
1618 | ret = btrfs_add_link(trans, dir, inode, name, name_len, 1, index); |
1619 | ||
1620 | /* FIXME, put inode into FIXUP list */ | |
1621 | ||
1622 | iput(inode); | |
1623 | iput(dir); | |
1624 | return ret; | |
1625 | } | |
1626 | ||
df8d116f FM |
1627 | /* |
1628 | * Return true if an inode reference exists in the log for the given name, | |
1629 | * inode and parent inode. | |
1630 | */ | |
1631 | static bool name_in_log_ref(struct btrfs_root *log_root, | |
1632 | const char *name, const int name_len, | |
1633 | const u64 dirid, const u64 ino) | |
1634 | { | |
1635 | struct btrfs_key search_key; | |
1636 | ||
1637 | search_key.objectid = ino; | |
1638 | search_key.type = BTRFS_INODE_REF_KEY; | |
1639 | search_key.offset = dirid; | |
1640 | if (backref_in_log(log_root, &search_key, dirid, name, name_len)) | |
1641 | return true; | |
1642 | ||
1643 | search_key.type = BTRFS_INODE_EXTREF_KEY; | |
1644 | search_key.offset = btrfs_extref_hash(dirid, name, name_len); | |
1645 | if (backref_in_log(log_root, &search_key, dirid, name, name_len)) | |
1646 | return true; | |
1647 | ||
1648 | return false; | |
1649 | } | |
1650 | ||
e02119d5 CM |
1651 | /* |
1652 | * take a single entry in a log directory item and replay it into | |
1653 | * the subvolume. | |
1654 | * | |
1655 | * if a conflicting item exists in the subdirectory already, | |
1656 | * the inode it points to is unlinked and put into the link count | |
1657 | * fix up tree. | |
1658 | * | |
1659 | * If a name from the log points to a file or directory that does | |
1660 | * not exist in the FS, it is skipped. fsyncs on directories | |
1661 | * do not force down inodes inside that directory, just changes to the | |
1662 | * names or unlinks in a directory. | |
bb53eda9 FM |
1663 | * |
1664 | * Returns < 0 on error, 0 if the name wasn't replayed (dentry points to a | |
1665 | * non-existing inode) and 1 if the name was replayed. | |
e02119d5 CM |
1666 | */ |
1667 | static noinline int replay_one_name(struct btrfs_trans_handle *trans, | |
1668 | struct btrfs_root *root, | |
1669 | struct btrfs_path *path, | |
1670 | struct extent_buffer *eb, | |
1671 | struct btrfs_dir_item *di, | |
1672 | struct btrfs_key *key) | |
1673 | { | |
1674 | char *name; | |
1675 | int name_len; | |
1676 | struct btrfs_dir_item *dst_di; | |
1677 | struct btrfs_key found_key; | |
1678 | struct btrfs_key log_key; | |
1679 | struct inode *dir; | |
e02119d5 | 1680 | u8 log_type; |
4bef0848 | 1681 | int exists; |
3650860b | 1682 | int ret = 0; |
d555438b | 1683 | bool update_size = (key->type == BTRFS_DIR_INDEX_KEY); |
bb53eda9 | 1684 | bool name_added = false; |
e02119d5 CM |
1685 | |
1686 | dir = read_one_inode(root, key->objectid); | |
c00e9493 TI |
1687 | if (!dir) |
1688 | return -EIO; | |
e02119d5 CM |
1689 | |
1690 | name_len = btrfs_dir_name_len(eb, di); | |
1691 | name = kmalloc(name_len, GFP_NOFS); | |
2bac325e FDBM |
1692 | if (!name) { |
1693 | ret = -ENOMEM; | |
1694 | goto out; | |
1695 | } | |
2a29edc6 | 1696 | |
e02119d5 CM |
1697 | log_type = btrfs_dir_type(eb, di); |
1698 | read_extent_buffer(eb, name, (unsigned long)(di + 1), | |
1699 | name_len); | |
1700 | ||
1701 | btrfs_dir_item_key_to_cpu(eb, di, &log_key); | |
4bef0848 CM |
1702 | exists = btrfs_lookup_inode(trans, root, path, &log_key, 0); |
1703 | if (exists == 0) | |
1704 | exists = 1; | |
1705 | else | |
1706 | exists = 0; | |
b3b4aa74 | 1707 | btrfs_release_path(path); |
4bef0848 | 1708 | |
e02119d5 CM |
1709 | if (key->type == BTRFS_DIR_ITEM_KEY) { |
1710 | dst_di = btrfs_lookup_dir_item(trans, root, path, key->objectid, | |
1711 | name, name_len, 1); | |
d397712b | 1712 | } else if (key->type == BTRFS_DIR_INDEX_KEY) { |
e02119d5 CM |
1713 | dst_di = btrfs_lookup_dir_index_item(trans, root, path, |
1714 | key->objectid, | |
1715 | key->offset, name, | |
1716 | name_len, 1); | |
1717 | } else { | |
3650860b JB |
1718 | /* Corruption */ |
1719 | ret = -EINVAL; | |
1720 | goto out; | |
e02119d5 | 1721 | } |
c704005d | 1722 | if (IS_ERR_OR_NULL(dst_di)) { |
e02119d5 CM |
1723 | /* we need a sequence number to insert, so we only |
1724 | * do inserts for the BTRFS_DIR_INDEX_KEY types | |
1725 | */ | |
1726 | if (key->type != BTRFS_DIR_INDEX_KEY) | |
1727 | goto out; | |
1728 | goto insert; | |
1729 | } | |
1730 | ||
1731 | btrfs_dir_item_key_to_cpu(path->nodes[0], dst_di, &found_key); | |
1732 | /* the existing item matches the logged item */ | |
1733 | if (found_key.objectid == log_key.objectid && | |
1734 | found_key.type == log_key.type && | |
1735 | found_key.offset == log_key.offset && | |
1736 | btrfs_dir_type(path->nodes[0], dst_di) == log_type) { | |
a2cc11db | 1737 | update_size = false; |
e02119d5 CM |
1738 | goto out; |
1739 | } | |
1740 | ||
1741 | /* | |
1742 | * don't drop the conflicting directory entry if the inode | |
1743 | * for the new entry doesn't exist | |
1744 | */ | |
4bef0848 | 1745 | if (!exists) |
e02119d5 CM |
1746 | goto out; |
1747 | ||
e02119d5 | 1748 | ret = drop_one_dir_item(trans, root, path, dir, dst_di); |
3650860b JB |
1749 | if (ret) |
1750 | goto out; | |
e02119d5 CM |
1751 | |
1752 | if (key->type == BTRFS_DIR_INDEX_KEY) | |
1753 | goto insert; | |
1754 | out: | |
b3b4aa74 | 1755 | btrfs_release_path(path); |
d555438b JB |
1756 | if (!ret && update_size) { |
1757 | btrfs_i_size_write(dir, dir->i_size + name_len * 2); | |
1758 | ret = btrfs_update_inode(trans, root, dir); | |
1759 | } | |
e02119d5 CM |
1760 | kfree(name); |
1761 | iput(dir); | |
bb53eda9 FM |
1762 | if (!ret && name_added) |
1763 | ret = 1; | |
3650860b | 1764 | return ret; |
e02119d5 CM |
1765 | |
1766 | insert: | |
df8d116f FM |
1767 | if (name_in_log_ref(root->log_root, name, name_len, |
1768 | key->objectid, log_key.objectid)) { | |
1769 | /* The dentry will be added later. */ | |
1770 | ret = 0; | |
1771 | update_size = false; | |
1772 | goto out; | |
1773 | } | |
b3b4aa74 | 1774 | btrfs_release_path(path); |
60d53eb3 Z |
1775 | ret = insert_one_name(trans, root, key->objectid, key->offset, |
1776 | name, name_len, &log_key); | |
df8d116f | 1777 | if (ret && ret != -ENOENT && ret != -EEXIST) |
3650860b | 1778 | goto out; |
bb53eda9 FM |
1779 | if (!ret) |
1780 | name_added = true; | |
d555438b | 1781 | update_size = false; |
3650860b | 1782 | ret = 0; |
e02119d5 CM |
1783 | goto out; |
1784 | } | |
1785 | ||
1786 | /* | |
1787 | * find all the names in a directory item and reconcile them into | |
1788 | * the subvolume. Only BTRFS_DIR_ITEM_KEY types will have more than | |
1789 | * one name in a directory item, but the same code gets used for | |
1790 | * both directory index types | |
1791 | */ | |
1792 | static noinline int replay_one_dir_item(struct btrfs_trans_handle *trans, | |
1793 | struct btrfs_root *root, | |
1794 | struct btrfs_path *path, | |
1795 | struct extent_buffer *eb, int slot, | |
1796 | struct btrfs_key *key) | |
1797 | { | |
bb53eda9 | 1798 | int ret = 0; |
e02119d5 CM |
1799 | u32 item_size = btrfs_item_size_nr(eb, slot); |
1800 | struct btrfs_dir_item *di; | |
1801 | int name_len; | |
1802 | unsigned long ptr; | |
1803 | unsigned long ptr_end; | |
bb53eda9 | 1804 | struct btrfs_path *fixup_path = NULL; |
e02119d5 CM |
1805 | |
1806 | ptr = btrfs_item_ptr_offset(eb, slot); | |
1807 | ptr_end = ptr + item_size; | |
d397712b | 1808 | while (ptr < ptr_end) { |
e02119d5 | 1809 | di = (struct btrfs_dir_item *)ptr; |
22a94d44 JB |
1810 | if (verify_dir_item(root, eb, di)) |
1811 | return -EIO; | |
e02119d5 CM |
1812 | name_len = btrfs_dir_name_len(eb, di); |
1813 | ret = replay_one_name(trans, root, path, eb, di, key); | |
bb53eda9 FM |
1814 | if (ret < 0) |
1815 | break; | |
e02119d5 CM |
1816 | ptr = (unsigned long)(di + 1); |
1817 | ptr += name_len; | |
bb53eda9 FM |
1818 | |
1819 | /* | |
1820 | * If this entry refers to a non-directory (directories can not | |
1821 | * have a link count > 1) and it was added in the transaction | |
1822 | * that was not committed, make sure we fixup the link count of | |
1823 | * the inode it the entry points to. Otherwise something like | |
1824 | * the following would result in a directory pointing to an | |
1825 | * inode with a wrong link that does not account for this dir | |
1826 | * entry: | |
1827 | * | |
1828 | * mkdir testdir | |
1829 | * touch testdir/foo | |
1830 | * touch testdir/bar | |
1831 | * sync | |
1832 | * | |
1833 | * ln testdir/bar testdir/bar_link | |
1834 | * ln testdir/foo testdir/foo_link | |
1835 | * xfs_io -c "fsync" testdir/bar | |
1836 | * | |
1837 | * <power failure> | |
1838 | * | |
1839 | * mount fs, log replay happens | |
1840 | * | |
1841 | * File foo would remain with a link count of 1 when it has two | |
1842 | * entries pointing to it in the directory testdir. This would | |
1843 | * make it impossible to ever delete the parent directory has | |
1844 | * it would result in stale dentries that can never be deleted. | |
1845 | */ | |
1846 | if (ret == 1 && btrfs_dir_type(eb, di) != BTRFS_FT_DIR) { | |
1847 | struct btrfs_key di_key; | |
1848 | ||
1849 | if (!fixup_path) { | |
1850 | fixup_path = btrfs_alloc_path(); | |
1851 | if (!fixup_path) { | |
1852 | ret = -ENOMEM; | |
1853 | break; | |
1854 | } | |
1855 | } | |
1856 | ||
1857 | btrfs_dir_item_key_to_cpu(eb, di, &di_key); | |
1858 | ret = link_to_fixup_dir(trans, root, fixup_path, | |
1859 | di_key.objectid); | |
1860 | if (ret) | |
1861 | break; | |
1862 | } | |
1863 | ret = 0; | |
e02119d5 | 1864 | } |
bb53eda9 FM |
1865 | btrfs_free_path(fixup_path); |
1866 | return ret; | |
e02119d5 CM |
1867 | } |
1868 | ||
1869 | /* | |
1870 | * directory replay has two parts. There are the standard directory | |
1871 | * items in the log copied from the subvolume, and range items | |
1872 | * created in the log while the subvolume was logged. | |
1873 | * | |
1874 | * The range items tell us which parts of the key space the log | |
1875 | * is authoritative for. During replay, if a key in the subvolume | |
1876 | * directory is in a logged range item, but not actually in the log | |
1877 | * that means it was deleted from the directory before the fsync | |
1878 | * and should be removed. | |
1879 | */ | |
1880 | static noinline int find_dir_range(struct btrfs_root *root, | |
1881 | struct btrfs_path *path, | |
1882 | u64 dirid, int key_type, | |
1883 | u64 *start_ret, u64 *end_ret) | |
1884 | { | |
1885 | struct btrfs_key key; | |
1886 | u64 found_end; | |
1887 | struct btrfs_dir_log_item *item; | |
1888 | int ret; | |
1889 | int nritems; | |
1890 | ||
1891 | if (*start_ret == (u64)-1) | |
1892 | return 1; | |
1893 | ||
1894 | key.objectid = dirid; | |
1895 | key.type = key_type; | |
1896 | key.offset = *start_ret; | |
1897 | ||
1898 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
1899 | if (ret < 0) | |
1900 | goto out; | |
1901 | if (ret > 0) { | |
1902 | if (path->slots[0] == 0) | |
1903 | goto out; | |
1904 | path->slots[0]--; | |
1905 | } | |
1906 | if (ret != 0) | |
1907 | btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]); | |
1908 | ||
1909 | if (key.type != key_type || key.objectid != dirid) { | |
1910 | ret = 1; | |
1911 | goto next; | |
1912 | } | |
1913 | item = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
1914 | struct btrfs_dir_log_item); | |
1915 | found_end = btrfs_dir_log_end(path->nodes[0], item); | |
1916 | ||
1917 | if (*start_ret >= key.offset && *start_ret <= found_end) { | |
1918 | ret = 0; | |
1919 | *start_ret = key.offset; | |
1920 | *end_ret = found_end; | |
1921 | goto out; | |
1922 | } | |
1923 | ret = 1; | |
1924 | next: | |
1925 | /* check the next slot in the tree to see if it is a valid item */ | |
1926 | nritems = btrfs_header_nritems(path->nodes[0]); | |
1927 | if (path->slots[0] >= nritems) { | |
1928 | ret = btrfs_next_leaf(root, path); | |
1929 | if (ret) | |
1930 | goto out; | |
1931 | } else { | |
1932 | path->slots[0]++; | |
1933 | } | |
1934 | ||
1935 | btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]); | |
1936 | ||
1937 | if (key.type != key_type || key.objectid != dirid) { | |
1938 | ret = 1; | |
1939 | goto out; | |
1940 | } | |
1941 | item = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
1942 | struct btrfs_dir_log_item); | |
1943 | found_end = btrfs_dir_log_end(path->nodes[0], item); | |
1944 | *start_ret = key.offset; | |
1945 | *end_ret = found_end; | |
1946 | ret = 0; | |
1947 | out: | |
b3b4aa74 | 1948 | btrfs_release_path(path); |
e02119d5 CM |
1949 | return ret; |
1950 | } | |
1951 | ||
1952 | /* | |
1953 | * this looks for a given directory item in the log. If the directory | |
1954 | * item is not in the log, the item is removed and the inode it points | |
1955 | * to is unlinked | |
1956 | */ | |
1957 | static noinline int check_item_in_log(struct btrfs_trans_handle *trans, | |
1958 | struct btrfs_root *root, | |
1959 | struct btrfs_root *log, | |
1960 | struct btrfs_path *path, | |
1961 | struct btrfs_path *log_path, | |
1962 | struct inode *dir, | |
1963 | struct btrfs_key *dir_key) | |
1964 | { | |
1965 | int ret; | |
1966 | struct extent_buffer *eb; | |
1967 | int slot; | |
1968 | u32 item_size; | |
1969 | struct btrfs_dir_item *di; | |
1970 | struct btrfs_dir_item *log_di; | |
1971 | int name_len; | |
1972 | unsigned long ptr; | |
1973 | unsigned long ptr_end; | |
1974 | char *name; | |
1975 | struct inode *inode; | |
1976 | struct btrfs_key location; | |
1977 | ||
1978 | again: | |
1979 | eb = path->nodes[0]; | |
1980 | slot = path->slots[0]; | |
1981 | item_size = btrfs_item_size_nr(eb, slot); | |
1982 | ptr = btrfs_item_ptr_offset(eb, slot); | |
1983 | ptr_end = ptr + item_size; | |
d397712b | 1984 | while (ptr < ptr_end) { |
e02119d5 | 1985 | di = (struct btrfs_dir_item *)ptr; |
22a94d44 JB |
1986 | if (verify_dir_item(root, eb, di)) { |
1987 | ret = -EIO; | |
1988 | goto out; | |
1989 | } | |
1990 | ||
e02119d5 CM |
1991 | name_len = btrfs_dir_name_len(eb, di); |
1992 | name = kmalloc(name_len, GFP_NOFS); | |
1993 | if (!name) { | |
1994 | ret = -ENOMEM; | |
1995 | goto out; | |
1996 | } | |
1997 | read_extent_buffer(eb, name, (unsigned long)(di + 1), | |
1998 | name_len); | |
1999 | log_di = NULL; | |
12fcfd22 | 2000 | if (log && dir_key->type == BTRFS_DIR_ITEM_KEY) { |
e02119d5 CM |
2001 | log_di = btrfs_lookup_dir_item(trans, log, log_path, |
2002 | dir_key->objectid, | |
2003 | name, name_len, 0); | |
12fcfd22 | 2004 | } else if (log && dir_key->type == BTRFS_DIR_INDEX_KEY) { |
e02119d5 CM |
2005 | log_di = btrfs_lookup_dir_index_item(trans, log, |
2006 | log_path, | |
2007 | dir_key->objectid, | |
2008 | dir_key->offset, | |
2009 | name, name_len, 0); | |
2010 | } | |
269d040f | 2011 | if (!log_di || (IS_ERR(log_di) && PTR_ERR(log_di) == -ENOENT)) { |
e02119d5 | 2012 | btrfs_dir_item_key_to_cpu(eb, di, &location); |
b3b4aa74 DS |
2013 | btrfs_release_path(path); |
2014 | btrfs_release_path(log_path); | |
e02119d5 | 2015 | inode = read_one_inode(root, location.objectid); |
c00e9493 TI |
2016 | if (!inode) { |
2017 | kfree(name); | |
2018 | return -EIO; | |
2019 | } | |
e02119d5 CM |
2020 | |
2021 | ret = link_to_fixup_dir(trans, root, | |
2022 | path, location.objectid); | |
3650860b JB |
2023 | if (ret) { |
2024 | kfree(name); | |
2025 | iput(inode); | |
2026 | goto out; | |
2027 | } | |
2028 | ||
8b558c5f | 2029 | inc_nlink(inode); |
e02119d5 CM |
2030 | ret = btrfs_unlink_inode(trans, root, dir, inode, |
2031 | name, name_len); | |
3650860b | 2032 | if (!ret) |
ada9af21 | 2033 | ret = btrfs_run_delayed_items(trans, root); |
e02119d5 CM |
2034 | kfree(name); |
2035 | iput(inode); | |
3650860b JB |
2036 | if (ret) |
2037 | goto out; | |
e02119d5 CM |
2038 | |
2039 | /* there might still be more names under this key | |
2040 | * check and repeat if required | |
2041 | */ | |
2042 | ret = btrfs_search_slot(NULL, root, dir_key, path, | |
2043 | 0, 0); | |
2044 | if (ret == 0) | |
2045 | goto again; | |
2046 | ret = 0; | |
2047 | goto out; | |
269d040f FDBM |
2048 | } else if (IS_ERR(log_di)) { |
2049 | kfree(name); | |
2050 | return PTR_ERR(log_di); | |
e02119d5 | 2051 | } |
b3b4aa74 | 2052 | btrfs_release_path(log_path); |
e02119d5 CM |
2053 | kfree(name); |
2054 | ||
2055 | ptr = (unsigned long)(di + 1); | |
2056 | ptr += name_len; | |
2057 | } | |
2058 | ret = 0; | |
2059 | out: | |
b3b4aa74 DS |
2060 | btrfs_release_path(path); |
2061 | btrfs_release_path(log_path); | |
e02119d5 CM |
2062 | return ret; |
2063 | } | |
2064 | ||
4f764e51 FM |
2065 | static int replay_xattr_deletes(struct btrfs_trans_handle *trans, |
2066 | struct btrfs_root *root, | |
2067 | struct btrfs_root *log, | |
2068 | struct btrfs_path *path, | |
2069 | const u64 ino) | |
2070 | { | |
2071 | struct btrfs_key search_key; | |
2072 | struct btrfs_path *log_path; | |
2073 | int i; | |
2074 | int nritems; | |
2075 | int ret; | |
2076 | ||
2077 | log_path = btrfs_alloc_path(); | |
2078 | if (!log_path) | |
2079 | return -ENOMEM; | |
2080 | ||
2081 | search_key.objectid = ino; | |
2082 | search_key.type = BTRFS_XATTR_ITEM_KEY; | |
2083 | search_key.offset = 0; | |
2084 | again: | |
2085 | ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0); | |
2086 | if (ret < 0) | |
2087 | goto out; | |
2088 | process_leaf: | |
2089 | nritems = btrfs_header_nritems(path->nodes[0]); | |
2090 | for (i = path->slots[0]; i < nritems; i++) { | |
2091 | struct btrfs_key key; | |
2092 | struct btrfs_dir_item *di; | |
2093 | struct btrfs_dir_item *log_di; | |
2094 | u32 total_size; | |
2095 | u32 cur; | |
2096 | ||
2097 | btrfs_item_key_to_cpu(path->nodes[0], &key, i); | |
2098 | if (key.objectid != ino || key.type != BTRFS_XATTR_ITEM_KEY) { | |
2099 | ret = 0; | |
2100 | goto out; | |
2101 | } | |
2102 | ||
2103 | di = btrfs_item_ptr(path->nodes[0], i, struct btrfs_dir_item); | |
2104 | total_size = btrfs_item_size_nr(path->nodes[0], i); | |
2105 | cur = 0; | |
2106 | while (cur < total_size) { | |
2107 | u16 name_len = btrfs_dir_name_len(path->nodes[0], di); | |
2108 | u16 data_len = btrfs_dir_data_len(path->nodes[0], di); | |
2109 | u32 this_len = sizeof(*di) + name_len + data_len; | |
2110 | char *name; | |
2111 | ||
2112 | name = kmalloc(name_len, GFP_NOFS); | |
2113 | if (!name) { | |
2114 | ret = -ENOMEM; | |
2115 | goto out; | |
2116 | } | |
2117 | read_extent_buffer(path->nodes[0], name, | |
2118 | (unsigned long)(di + 1), name_len); | |
2119 | ||
2120 | log_di = btrfs_lookup_xattr(NULL, log, log_path, ino, | |
2121 | name, name_len, 0); | |
2122 | btrfs_release_path(log_path); | |
2123 | if (!log_di) { | |
2124 | /* Doesn't exist in log tree, so delete it. */ | |
2125 | btrfs_release_path(path); | |
2126 | di = btrfs_lookup_xattr(trans, root, path, ino, | |
2127 | name, name_len, -1); | |
2128 | kfree(name); | |
2129 | if (IS_ERR(di)) { | |
2130 | ret = PTR_ERR(di); | |
2131 | goto out; | |
2132 | } | |
2133 | ASSERT(di); | |
2134 | ret = btrfs_delete_one_dir_name(trans, root, | |
2135 | path, di); | |
2136 | if (ret) | |
2137 | goto out; | |
2138 | btrfs_release_path(path); | |
2139 | search_key = key; | |
2140 | goto again; | |
2141 | } | |
2142 | kfree(name); | |
2143 | if (IS_ERR(log_di)) { | |
2144 | ret = PTR_ERR(log_di); | |
2145 | goto out; | |
2146 | } | |
2147 | cur += this_len; | |
2148 | di = (struct btrfs_dir_item *)((char *)di + this_len); | |
2149 | } | |
2150 | } | |
2151 | ret = btrfs_next_leaf(root, path); | |
2152 | if (ret > 0) | |
2153 | ret = 0; | |
2154 | else if (ret == 0) | |
2155 | goto process_leaf; | |
2156 | out: | |
2157 | btrfs_free_path(log_path); | |
2158 | btrfs_release_path(path); | |
2159 | return ret; | |
2160 | } | |
2161 | ||
2162 | ||
e02119d5 CM |
2163 | /* |
2164 | * deletion replay happens before we copy any new directory items | |
2165 | * out of the log or out of backreferences from inodes. It | |
2166 | * scans the log to find ranges of keys that log is authoritative for, | |
2167 | * and then scans the directory to find items in those ranges that are | |
2168 | * not present in the log. | |
2169 | * | |
2170 | * Anything we don't find in the log is unlinked and removed from the | |
2171 | * directory. | |
2172 | */ | |
2173 | static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans, | |
2174 | struct btrfs_root *root, | |
2175 | struct btrfs_root *log, | |
2176 | struct btrfs_path *path, | |
12fcfd22 | 2177 | u64 dirid, int del_all) |
e02119d5 CM |
2178 | { |
2179 | u64 range_start; | |
2180 | u64 range_end; | |
2181 | int key_type = BTRFS_DIR_LOG_ITEM_KEY; | |
2182 | int ret = 0; | |
2183 | struct btrfs_key dir_key; | |
2184 | struct btrfs_key found_key; | |
2185 | struct btrfs_path *log_path; | |
2186 | struct inode *dir; | |
2187 | ||
2188 | dir_key.objectid = dirid; | |
2189 | dir_key.type = BTRFS_DIR_ITEM_KEY; | |
2190 | log_path = btrfs_alloc_path(); | |
2191 | if (!log_path) | |
2192 | return -ENOMEM; | |
2193 | ||
2194 | dir = read_one_inode(root, dirid); | |
2195 | /* it isn't an error if the inode isn't there, that can happen | |
2196 | * because we replay the deletes before we copy in the inode item | |
2197 | * from the log | |
2198 | */ | |
2199 | if (!dir) { | |
2200 | btrfs_free_path(log_path); | |
2201 | return 0; | |
2202 | } | |
2203 | again: | |
2204 | range_start = 0; | |
2205 | range_end = 0; | |
d397712b | 2206 | while (1) { |
12fcfd22 CM |
2207 | if (del_all) |
2208 | range_end = (u64)-1; | |
2209 | else { | |
2210 | ret = find_dir_range(log, path, dirid, key_type, | |
2211 | &range_start, &range_end); | |
2212 | if (ret != 0) | |
2213 | break; | |
2214 | } | |
e02119d5 CM |
2215 | |
2216 | dir_key.offset = range_start; | |
d397712b | 2217 | while (1) { |
e02119d5 CM |
2218 | int nritems; |
2219 | ret = btrfs_search_slot(NULL, root, &dir_key, path, | |
2220 | 0, 0); | |
2221 | if (ret < 0) | |
2222 | goto out; | |
2223 | ||
2224 | nritems = btrfs_header_nritems(path->nodes[0]); | |
2225 | if (path->slots[0] >= nritems) { | |
2226 | ret = btrfs_next_leaf(root, path); | |
2227 | if (ret) | |
2228 | break; | |
2229 | } | |
2230 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, | |
2231 | path->slots[0]); | |
2232 | if (found_key.objectid != dirid || | |
2233 | found_key.type != dir_key.type) | |
2234 | goto next_type; | |
2235 | ||
2236 | if (found_key.offset > range_end) | |
2237 | break; | |
2238 | ||
2239 | ret = check_item_in_log(trans, root, log, path, | |
12fcfd22 CM |
2240 | log_path, dir, |
2241 | &found_key); | |
3650860b JB |
2242 | if (ret) |
2243 | goto out; | |
e02119d5 CM |
2244 | if (found_key.offset == (u64)-1) |
2245 | break; | |
2246 | dir_key.offset = found_key.offset + 1; | |
2247 | } | |
b3b4aa74 | 2248 | btrfs_release_path(path); |
e02119d5 CM |
2249 | if (range_end == (u64)-1) |
2250 | break; | |
2251 | range_start = range_end + 1; | |
2252 | } | |
2253 | ||
2254 | next_type: | |
2255 | ret = 0; | |
2256 | if (key_type == BTRFS_DIR_LOG_ITEM_KEY) { | |
2257 | key_type = BTRFS_DIR_LOG_INDEX_KEY; | |
2258 | dir_key.type = BTRFS_DIR_INDEX_KEY; | |
b3b4aa74 | 2259 | btrfs_release_path(path); |
e02119d5 CM |
2260 | goto again; |
2261 | } | |
2262 | out: | |
b3b4aa74 | 2263 | btrfs_release_path(path); |
e02119d5 CM |
2264 | btrfs_free_path(log_path); |
2265 | iput(dir); | |
2266 | return ret; | |
2267 | } | |
2268 | ||
2269 | /* | |
2270 | * the process_func used to replay items from the log tree. This | |
2271 | * gets called in two different stages. The first stage just looks | |
2272 | * for inodes and makes sure they are all copied into the subvolume. | |
2273 | * | |
2274 | * The second stage copies all the other item types from the log into | |
2275 | * the subvolume. The two stage approach is slower, but gets rid of | |
2276 | * lots of complexity around inodes referencing other inodes that exist | |
2277 | * only in the log (references come from either directory items or inode | |
2278 | * back refs). | |
2279 | */ | |
2280 | static int replay_one_buffer(struct btrfs_root *log, struct extent_buffer *eb, | |
2281 | struct walk_control *wc, u64 gen) | |
2282 | { | |
2283 | int nritems; | |
2284 | struct btrfs_path *path; | |
2285 | struct btrfs_root *root = wc->replay_dest; | |
2286 | struct btrfs_key key; | |
e02119d5 CM |
2287 | int level; |
2288 | int i; | |
2289 | int ret; | |
2290 | ||
018642a1 TI |
2291 | ret = btrfs_read_buffer(eb, gen); |
2292 | if (ret) | |
2293 | return ret; | |
e02119d5 CM |
2294 | |
2295 | level = btrfs_header_level(eb); | |
2296 | ||
2297 | if (level != 0) | |
2298 | return 0; | |
2299 | ||
2300 | path = btrfs_alloc_path(); | |
1e5063d0 MF |
2301 | if (!path) |
2302 | return -ENOMEM; | |
e02119d5 CM |
2303 | |
2304 | nritems = btrfs_header_nritems(eb); | |
2305 | for (i = 0; i < nritems; i++) { | |
2306 | btrfs_item_key_to_cpu(eb, &key, i); | |
e02119d5 CM |
2307 | |
2308 | /* inode keys are done during the first stage */ | |
2309 | if (key.type == BTRFS_INODE_ITEM_KEY && | |
2310 | wc->stage == LOG_WALK_REPLAY_INODES) { | |
e02119d5 CM |
2311 | struct btrfs_inode_item *inode_item; |
2312 | u32 mode; | |
2313 | ||
2314 | inode_item = btrfs_item_ptr(eb, i, | |
2315 | struct btrfs_inode_item); | |
4f764e51 FM |
2316 | ret = replay_xattr_deletes(wc->trans, root, log, |
2317 | path, key.objectid); | |
2318 | if (ret) | |
2319 | break; | |
e02119d5 CM |
2320 | mode = btrfs_inode_mode(eb, inode_item); |
2321 | if (S_ISDIR(mode)) { | |
2322 | ret = replay_dir_deletes(wc->trans, | |
12fcfd22 | 2323 | root, log, path, key.objectid, 0); |
b50c6e25 JB |
2324 | if (ret) |
2325 | break; | |
e02119d5 CM |
2326 | } |
2327 | ret = overwrite_item(wc->trans, root, path, | |
2328 | eb, i, &key); | |
b50c6e25 JB |
2329 | if (ret) |
2330 | break; | |
e02119d5 | 2331 | |
c71bf099 YZ |
2332 | /* for regular files, make sure corresponding |
2333 | * orhpan item exist. extents past the new EOF | |
2334 | * will be truncated later by orphan cleanup. | |
e02119d5 CM |
2335 | */ |
2336 | if (S_ISREG(mode)) { | |
c71bf099 YZ |
2337 | ret = insert_orphan_item(wc->trans, root, |
2338 | key.objectid); | |
b50c6e25 JB |
2339 | if (ret) |
2340 | break; | |
e02119d5 | 2341 | } |
c71bf099 | 2342 | |
e02119d5 CM |
2343 | ret = link_to_fixup_dir(wc->trans, root, |
2344 | path, key.objectid); | |
b50c6e25 JB |
2345 | if (ret) |
2346 | break; | |
e02119d5 | 2347 | } |
dd8e7217 JB |
2348 | |
2349 | if (key.type == BTRFS_DIR_INDEX_KEY && | |
2350 | wc->stage == LOG_WALK_REPLAY_DIR_INDEX) { | |
2351 | ret = replay_one_dir_item(wc->trans, root, path, | |
2352 | eb, i, &key); | |
2353 | if (ret) | |
2354 | break; | |
2355 | } | |
2356 | ||
e02119d5 CM |
2357 | if (wc->stage < LOG_WALK_REPLAY_ALL) |
2358 | continue; | |
2359 | ||
2360 | /* these keys are simply copied */ | |
2361 | if (key.type == BTRFS_XATTR_ITEM_KEY) { | |
2362 | ret = overwrite_item(wc->trans, root, path, | |
2363 | eb, i, &key); | |
b50c6e25 JB |
2364 | if (ret) |
2365 | break; | |
2da1c669 LB |
2366 | } else if (key.type == BTRFS_INODE_REF_KEY || |
2367 | key.type == BTRFS_INODE_EXTREF_KEY) { | |
f186373f MF |
2368 | ret = add_inode_ref(wc->trans, root, log, path, |
2369 | eb, i, &key); | |
b50c6e25 JB |
2370 | if (ret && ret != -ENOENT) |
2371 | break; | |
2372 | ret = 0; | |
e02119d5 CM |
2373 | } else if (key.type == BTRFS_EXTENT_DATA_KEY) { |
2374 | ret = replay_one_extent(wc->trans, root, path, | |
2375 | eb, i, &key); | |
b50c6e25 JB |
2376 | if (ret) |
2377 | break; | |
dd8e7217 | 2378 | } else if (key.type == BTRFS_DIR_ITEM_KEY) { |
e02119d5 CM |
2379 | ret = replay_one_dir_item(wc->trans, root, path, |
2380 | eb, i, &key); | |
b50c6e25 JB |
2381 | if (ret) |
2382 | break; | |
e02119d5 CM |
2383 | } |
2384 | } | |
2385 | btrfs_free_path(path); | |
b50c6e25 | 2386 | return ret; |
e02119d5 CM |
2387 | } |
2388 | ||
d397712b | 2389 | static noinline int walk_down_log_tree(struct btrfs_trans_handle *trans, |
e02119d5 CM |
2390 | struct btrfs_root *root, |
2391 | struct btrfs_path *path, int *level, | |
2392 | struct walk_control *wc) | |
2393 | { | |
2394 | u64 root_owner; | |
e02119d5 CM |
2395 | u64 bytenr; |
2396 | u64 ptr_gen; | |
2397 | struct extent_buffer *next; | |
2398 | struct extent_buffer *cur; | |
2399 | struct extent_buffer *parent; | |
2400 | u32 blocksize; | |
2401 | int ret = 0; | |
2402 | ||
2403 | WARN_ON(*level < 0); | |
2404 | WARN_ON(*level >= BTRFS_MAX_LEVEL); | |
2405 | ||
d397712b | 2406 | while (*level > 0) { |
e02119d5 CM |
2407 | WARN_ON(*level < 0); |
2408 | WARN_ON(*level >= BTRFS_MAX_LEVEL); | |
2409 | cur = path->nodes[*level]; | |
2410 | ||
fae7f21c | 2411 | WARN_ON(btrfs_header_level(cur) != *level); |
e02119d5 CM |
2412 | |
2413 | if (path->slots[*level] >= | |
2414 | btrfs_header_nritems(cur)) | |
2415 | break; | |
2416 | ||
2417 | bytenr = btrfs_node_blockptr(cur, path->slots[*level]); | |
2418 | ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]); | |
707e8a07 | 2419 | blocksize = root->nodesize; |
e02119d5 CM |
2420 | |
2421 | parent = path->nodes[*level]; | |
2422 | root_owner = btrfs_header_owner(parent); | |
e02119d5 | 2423 | |
a83fffb7 | 2424 | next = btrfs_find_create_tree_block(root, bytenr); |
2a29edc6 | 2425 | if (!next) |
2426 | return -ENOMEM; | |
e02119d5 | 2427 | |
e02119d5 | 2428 | if (*level == 1) { |
1e5063d0 | 2429 | ret = wc->process_func(root, next, wc, ptr_gen); |
b50c6e25 JB |
2430 | if (ret) { |
2431 | free_extent_buffer(next); | |
1e5063d0 | 2432 | return ret; |
b50c6e25 | 2433 | } |
4a500fd1 | 2434 | |
e02119d5 CM |
2435 | path->slots[*level]++; |
2436 | if (wc->free) { | |
018642a1 TI |
2437 | ret = btrfs_read_buffer(next, ptr_gen); |
2438 | if (ret) { | |
2439 | free_extent_buffer(next); | |
2440 | return ret; | |
2441 | } | |
e02119d5 | 2442 | |
681ae509 JB |
2443 | if (trans) { |
2444 | btrfs_tree_lock(next); | |
2445 | btrfs_set_lock_blocking(next); | |
01d58472 DD |
2446 | clean_tree_block(trans, root->fs_info, |
2447 | next); | |
681ae509 JB |
2448 | btrfs_wait_tree_block_writeback(next); |
2449 | btrfs_tree_unlock(next); | |
2450 | } | |
e02119d5 | 2451 | |
e02119d5 CM |
2452 | WARN_ON(root_owner != |
2453 | BTRFS_TREE_LOG_OBJECTID); | |
e688b725 | 2454 | ret = btrfs_free_and_pin_reserved_extent(root, |
d00aff00 | 2455 | bytenr, blocksize); |
3650860b JB |
2456 | if (ret) { |
2457 | free_extent_buffer(next); | |
2458 | return ret; | |
2459 | } | |
e02119d5 CM |
2460 | } |
2461 | free_extent_buffer(next); | |
2462 | continue; | |
2463 | } | |
018642a1 TI |
2464 | ret = btrfs_read_buffer(next, ptr_gen); |
2465 | if (ret) { | |
2466 | free_extent_buffer(next); | |
2467 | return ret; | |
2468 | } | |
e02119d5 CM |
2469 | |
2470 | WARN_ON(*level <= 0); | |
2471 | if (path->nodes[*level-1]) | |
2472 | free_extent_buffer(path->nodes[*level-1]); | |
2473 | path->nodes[*level-1] = next; | |
2474 | *level = btrfs_header_level(next); | |
2475 | path->slots[*level] = 0; | |
2476 | cond_resched(); | |
2477 | } | |
2478 | WARN_ON(*level < 0); | |
2479 | WARN_ON(*level >= BTRFS_MAX_LEVEL); | |
2480 | ||
4a500fd1 | 2481 | path->slots[*level] = btrfs_header_nritems(path->nodes[*level]); |
e02119d5 CM |
2482 | |
2483 | cond_resched(); | |
2484 | return 0; | |
2485 | } | |
2486 | ||
d397712b | 2487 | static noinline int walk_up_log_tree(struct btrfs_trans_handle *trans, |
e02119d5 CM |
2488 | struct btrfs_root *root, |
2489 | struct btrfs_path *path, int *level, | |
2490 | struct walk_control *wc) | |
2491 | { | |
2492 | u64 root_owner; | |
e02119d5 CM |
2493 | int i; |
2494 | int slot; | |
2495 | int ret; | |
2496 | ||
d397712b | 2497 | for (i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) { |
e02119d5 | 2498 | slot = path->slots[i]; |
4a500fd1 | 2499 | if (slot + 1 < btrfs_header_nritems(path->nodes[i])) { |
e02119d5 CM |
2500 | path->slots[i]++; |
2501 | *level = i; | |
2502 | WARN_ON(*level == 0); | |
2503 | return 0; | |
2504 | } else { | |
31840ae1 ZY |
2505 | struct extent_buffer *parent; |
2506 | if (path->nodes[*level] == root->node) | |
2507 | parent = path->nodes[*level]; | |
2508 | else | |
2509 | parent = path->nodes[*level + 1]; | |
2510 | ||
2511 | root_owner = btrfs_header_owner(parent); | |
1e5063d0 | 2512 | ret = wc->process_func(root, path->nodes[*level], wc, |
e02119d5 | 2513 | btrfs_header_generation(path->nodes[*level])); |
1e5063d0 MF |
2514 | if (ret) |
2515 | return ret; | |
2516 | ||
e02119d5 CM |
2517 | if (wc->free) { |
2518 | struct extent_buffer *next; | |
2519 | ||
2520 | next = path->nodes[*level]; | |
2521 | ||
681ae509 JB |
2522 | if (trans) { |
2523 | btrfs_tree_lock(next); | |
2524 | btrfs_set_lock_blocking(next); | |
01d58472 DD |
2525 | clean_tree_block(trans, root->fs_info, |
2526 | next); | |
681ae509 JB |
2527 | btrfs_wait_tree_block_writeback(next); |
2528 | btrfs_tree_unlock(next); | |
2529 | } | |
e02119d5 | 2530 | |
e02119d5 | 2531 | WARN_ON(root_owner != BTRFS_TREE_LOG_OBJECTID); |
e688b725 | 2532 | ret = btrfs_free_and_pin_reserved_extent(root, |
e02119d5 | 2533 | path->nodes[*level]->start, |
d00aff00 | 2534 | path->nodes[*level]->len); |
3650860b JB |
2535 | if (ret) |
2536 | return ret; | |
e02119d5 CM |
2537 | } |
2538 | free_extent_buffer(path->nodes[*level]); | |
2539 | path->nodes[*level] = NULL; | |
2540 | *level = i + 1; | |
2541 | } | |
2542 | } | |
2543 | return 1; | |
2544 | } | |
2545 | ||
2546 | /* | |
2547 | * drop the reference count on the tree rooted at 'snap'. This traverses | |
2548 | * the tree freeing any blocks that have a ref count of zero after being | |
2549 | * decremented. | |
2550 | */ | |
2551 | static int walk_log_tree(struct btrfs_trans_handle *trans, | |
2552 | struct btrfs_root *log, struct walk_control *wc) | |
2553 | { | |
2554 | int ret = 0; | |
2555 | int wret; | |
2556 | int level; | |
2557 | struct btrfs_path *path; | |
e02119d5 CM |
2558 | int orig_level; |
2559 | ||
2560 | path = btrfs_alloc_path(); | |
db5b493a TI |
2561 | if (!path) |
2562 | return -ENOMEM; | |
e02119d5 CM |
2563 | |
2564 | level = btrfs_header_level(log->node); | |
2565 | orig_level = level; | |
2566 | path->nodes[level] = log->node; | |
2567 | extent_buffer_get(log->node); | |
2568 | path->slots[level] = 0; | |
2569 | ||
d397712b | 2570 | while (1) { |
e02119d5 CM |
2571 | wret = walk_down_log_tree(trans, log, path, &level, wc); |
2572 | if (wret > 0) | |
2573 | break; | |
79787eaa | 2574 | if (wret < 0) { |
e02119d5 | 2575 | ret = wret; |
79787eaa JM |
2576 | goto out; |
2577 | } | |
e02119d5 CM |
2578 | |
2579 | wret = walk_up_log_tree(trans, log, path, &level, wc); | |
2580 | if (wret > 0) | |
2581 | break; | |
79787eaa | 2582 | if (wret < 0) { |
e02119d5 | 2583 | ret = wret; |
79787eaa JM |
2584 | goto out; |
2585 | } | |
e02119d5 CM |
2586 | } |
2587 | ||
2588 | /* was the root node processed? if not, catch it here */ | |
2589 | if (path->nodes[orig_level]) { | |
79787eaa | 2590 | ret = wc->process_func(log, path->nodes[orig_level], wc, |
e02119d5 | 2591 | btrfs_header_generation(path->nodes[orig_level])); |
79787eaa JM |
2592 | if (ret) |
2593 | goto out; | |
e02119d5 CM |
2594 | if (wc->free) { |
2595 | struct extent_buffer *next; | |
2596 | ||
2597 | next = path->nodes[orig_level]; | |
2598 | ||
681ae509 JB |
2599 | if (trans) { |
2600 | btrfs_tree_lock(next); | |
2601 | btrfs_set_lock_blocking(next); | |
01d58472 | 2602 | clean_tree_block(trans, log->fs_info, next); |
681ae509 JB |
2603 | btrfs_wait_tree_block_writeback(next); |
2604 | btrfs_tree_unlock(next); | |
2605 | } | |
e02119d5 | 2606 | |
e02119d5 CM |
2607 | WARN_ON(log->root_key.objectid != |
2608 | BTRFS_TREE_LOG_OBJECTID); | |
e688b725 | 2609 | ret = btrfs_free_and_pin_reserved_extent(log, next->start, |
d00aff00 | 2610 | next->len); |
3650860b JB |
2611 | if (ret) |
2612 | goto out; | |
e02119d5 CM |
2613 | } |
2614 | } | |
2615 | ||
79787eaa | 2616 | out: |
e02119d5 | 2617 | btrfs_free_path(path); |
e02119d5 CM |
2618 | return ret; |
2619 | } | |
2620 | ||
7237f183 YZ |
2621 | /* |
2622 | * helper function to update the item for a given subvolumes log root | |
2623 | * in the tree of log roots | |
2624 | */ | |
2625 | static int update_log_root(struct btrfs_trans_handle *trans, | |
2626 | struct btrfs_root *log) | |
2627 | { | |
2628 | int ret; | |
2629 | ||
2630 | if (log->log_transid == 1) { | |
2631 | /* insert root item on the first sync */ | |
2632 | ret = btrfs_insert_root(trans, log->fs_info->log_root_tree, | |
2633 | &log->root_key, &log->root_item); | |
2634 | } else { | |
2635 | ret = btrfs_update_root(trans, log->fs_info->log_root_tree, | |
2636 | &log->root_key, &log->root_item); | |
2637 | } | |
2638 | return ret; | |
2639 | } | |
2640 | ||
60d53eb3 | 2641 | static void wait_log_commit(struct btrfs_root *root, int transid) |
e02119d5 CM |
2642 | { |
2643 | DEFINE_WAIT(wait); | |
7237f183 | 2644 | int index = transid % 2; |
e02119d5 | 2645 | |
7237f183 YZ |
2646 | /* |
2647 | * we only allow two pending log transactions at a time, | |
2648 | * so we know that if ours is more than 2 older than the | |
2649 | * current transaction, we're done | |
2650 | */ | |
e02119d5 | 2651 | do { |
7237f183 YZ |
2652 | prepare_to_wait(&root->log_commit_wait[index], |
2653 | &wait, TASK_UNINTERRUPTIBLE); | |
2654 | mutex_unlock(&root->log_mutex); | |
12fcfd22 | 2655 | |
d1433deb | 2656 | if (root->log_transid_committed < transid && |
7237f183 YZ |
2657 | atomic_read(&root->log_commit[index])) |
2658 | schedule(); | |
12fcfd22 | 2659 | |
7237f183 YZ |
2660 | finish_wait(&root->log_commit_wait[index], &wait); |
2661 | mutex_lock(&root->log_mutex); | |
d1433deb | 2662 | } while (root->log_transid_committed < transid && |
7237f183 | 2663 | atomic_read(&root->log_commit[index])); |
7237f183 YZ |
2664 | } |
2665 | ||
60d53eb3 | 2666 | static void wait_for_writer(struct btrfs_root *root) |
7237f183 YZ |
2667 | { |
2668 | DEFINE_WAIT(wait); | |
8b050d35 MX |
2669 | |
2670 | while (atomic_read(&root->log_writers)) { | |
7237f183 YZ |
2671 | prepare_to_wait(&root->log_writer_wait, |
2672 | &wait, TASK_UNINTERRUPTIBLE); | |
2673 | mutex_unlock(&root->log_mutex); | |
8b050d35 | 2674 | if (atomic_read(&root->log_writers)) |
e02119d5 | 2675 | schedule(); |
7237f183 | 2676 | finish_wait(&root->log_writer_wait, &wait); |
575849ec | 2677 | mutex_lock(&root->log_mutex); |
7237f183 | 2678 | } |
e02119d5 CM |
2679 | } |
2680 | ||
8b050d35 MX |
2681 | static inline void btrfs_remove_log_ctx(struct btrfs_root *root, |
2682 | struct btrfs_log_ctx *ctx) | |
2683 | { | |
2684 | if (!ctx) | |
2685 | return; | |
2686 | ||
2687 | mutex_lock(&root->log_mutex); | |
2688 | list_del_init(&ctx->list); | |
2689 | mutex_unlock(&root->log_mutex); | |
2690 | } | |
2691 | ||
2692 | /* | |
2693 | * Invoked in log mutex context, or be sure there is no other task which | |
2694 | * can access the list. | |
2695 | */ | |
2696 | static inline void btrfs_remove_all_log_ctxs(struct btrfs_root *root, | |
2697 | int index, int error) | |
2698 | { | |
2699 | struct btrfs_log_ctx *ctx; | |
2700 | ||
2701 | if (!error) { | |
2702 | INIT_LIST_HEAD(&root->log_ctxs[index]); | |
2703 | return; | |
2704 | } | |
2705 | ||
2706 | list_for_each_entry(ctx, &root->log_ctxs[index], list) | |
2707 | ctx->log_ret = error; | |
2708 | ||
2709 | INIT_LIST_HEAD(&root->log_ctxs[index]); | |
2710 | } | |
2711 | ||
e02119d5 CM |
2712 | /* |
2713 | * btrfs_sync_log does sends a given tree log down to the disk and | |
2714 | * updates the super blocks to record it. When this call is done, | |
12fcfd22 CM |
2715 | * you know that any inodes previously logged are safely on disk only |
2716 | * if it returns 0. | |
2717 | * | |
2718 | * Any other return value means you need to call btrfs_commit_transaction. | |
2719 | * Some of the edge cases for fsyncing directories that have had unlinks | |
2720 | * or renames done in the past mean that sometimes the only safe | |
2721 | * fsync is to commit the whole FS. When btrfs_sync_log returns -EAGAIN, | |
2722 | * that has happened. | |
e02119d5 CM |
2723 | */ |
2724 | int btrfs_sync_log(struct btrfs_trans_handle *trans, | |
8b050d35 | 2725 | struct btrfs_root *root, struct btrfs_log_ctx *ctx) |
e02119d5 | 2726 | { |
7237f183 YZ |
2727 | int index1; |
2728 | int index2; | |
8cef4e16 | 2729 | int mark; |
e02119d5 | 2730 | int ret; |
e02119d5 | 2731 | struct btrfs_root *log = root->log_root; |
7237f183 | 2732 | struct btrfs_root *log_root_tree = root->fs_info->log_root_tree; |
bb14a59b | 2733 | int log_transid = 0; |
8b050d35 | 2734 | struct btrfs_log_ctx root_log_ctx; |
c6adc9cc | 2735 | struct blk_plug plug; |
e02119d5 | 2736 | |
7237f183 | 2737 | mutex_lock(&root->log_mutex); |
d1433deb MX |
2738 | log_transid = ctx->log_transid; |
2739 | if (root->log_transid_committed >= log_transid) { | |
2740 | mutex_unlock(&root->log_mutex); | |
2741 | return ctx->log_ret; | |
2742 | } | |
2743 | ||
2744 | index1 = log_transid % 2; | |
7237f183 | 2745 | if (atomic_read(&root->log_commit[index1])) { |
60d53eb3 | 2746 | wait_log_commit(root, log_transid); |
7237f183 | 2747 | mutex_unlock(&root->log_mutex); |
8b050d35 | 2748 | return ctx->log_ret; |
e02119d5 | 2749 | } |
d1433deb | 2750 | ASSERT(log_transid == root->log_transid); |
7237f183 YZ |
2751 | atomic_set(&root->log_commit[index1], 1); |
2752 | ||
2753 | /* wait for previous tree log sync to complete */ | |
2754 | if (atomic_read(&root->log_commit[(index1 + 1) % 2])) | |
60d53eb3 | 2755 | wait_log_commit(root, log_transid - 1); |
48cab2e0 | 2756 | |
86df7eb9 | 2757 | while (1) { |
2ecb7923 | 2758 | int batch = atomic_read(&root->log_batch); |
cd354ad6 | 2759 | /* when we're on an ssd, just kick the log commit out */ |
27cdeb70 MX |
2760 | if (!btrfs_test_opt(root, SSD) && |
2761 | test_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state)) { | |
86df7eb9 YZ |
2762 | mutex_unlock(&root->log_mutex); |
2763 | schedule_timeout_uninterruptible(1); | |
2764 | mutex_lock(&root->log_mutex); | |
2765 | } | |
60d53eb3 | 2766 | wait_for_writer(root); |
2ecb7923 | 2767 | if (batch == atomic_read(&root->log_batch)) |
e02119d5 CM |
2768 | break; |
2769 | } | |
e02119d5 | 2770 | |
12fcfd22 | 2771 | /* bail out if we need to do a full commit */ |
995946dd | 2772 | if (btrfs_need_log_full_commit(root->fs_info, trans)) { |
12fcfd22 | 2773 | ret = -EAGAIN; |
2ab28f32 | 2774 | btrfs_free_logged_extents(log, log_transid); |
12fcfd22 CM |
2775 | mutex_unlock(&root->log_mutex); |
2776 | goto out; | |
2777 | } | |
2778 | ||
8cef4e16 YZ |
2779 | if (log_transid % 2 == 0) |
2780 | mark = EXTENT_DIRTY; | |
2781 | else | |
2782 | mark = EXTENT_NEW; | |
2783 | ||
690587d1 CM |
2784 | /* we start IO on all the marked extents here, but we don't actually |
2785 | * wait for them until later. | |
2786 | */ | |
c6adc9cc | 2787 | blk_start_plug(&plug); |
8cef4e16 | 2788 | ret = btrfs_write_marked_extents(log, &log->dirty_log_pages, mark); |
79787eaa | 2789 | if (ret) { |
c6adc9cc | 2790 | blk_finish_plug(&plug); |
79787eaa | 2791 | btrfs_abort_transaction(trans, root, ret); |
2ab28f32 | 2792 | btrfs_free_logged_extents(log, log_transid); |
995946dd | 2793 | btrfs_set_log_full_commit(root->fs_info, trans); |
79787eaa JM |
2794 | mutex_unlock(&root->log_mutex); |
2795 | goto out; | |
2796 | } | |
7237f183 | 2797 | |
5d4f98a2 | 2798 | btrfs_set_root_node(&log->root_item, log->node); |
7237f183 | 2799 | |
7237f183 YZ |
2800 | root->log_transid++; |
2801 | log->log_transid = root->log_transid; | |
ff782e0a | 2802 | root->log_start_pid = 0; |
7237f183 | 2803 | /* |
8cef4e16 YZ |
2804 | * IO has been started, blocks of the log tree have WRITTEN flag set |
2805 | * in their headers. new modifications of the log will be written to | |
2806 | * new positions. so it's safe to allow log writers to go in. | |
7237f183 YZ |
2807 | */ |
2808 | mutex_unlock(&root->log_mutex); | |
2809 | ||
d1433deb MX |
2810 | btrfs_init_log_ctx(&root_log_ctx); |
2811 | ||
7237f183 | 2812 | mutex_lock(&log_root_tree->log_mutex); |
2ecb7923 | 2813 | atomic_inc(&log_root_tree->log_batch); |
7237f183 | 2814 | atomic_inc(&log_root_tree->log_writers); |
d1433deb MX |
2815 | |
2816 | index2 = log_root_tree->log_transid % 2; | |
2817 | list_add_tail(&root_log_ctx.list, &log_root_tree->log_ctxs[index2]); | |
2818 | root_log_ctx.log_transid = log_root_tree->log_transid; | |
2819 | ||
7237f183 YZ |
2820 | mutex_unlock(&log_root_tree->log_mutex); |
2821 | ||
2822 | ret = update_log_root(trans, log); | |
7237f183 YZ |
2823 | |
2824 | mutex_lock(&log_root_tree->log_mutex); | |
2825 | if (atomic_dec_and_test(&log_root_tree->log_writers)) { | |
779adf0f DS |
2826 | /* |
2827 | * Implicit memory barrier after atomic_dec_and_test | |
2828 | */ | |
7237f183 YZ |
2829 | if (waitqueue_active(&log_root_tree->log_writer_wait)) |
2830 | wake_up(&log_root_tree->log_writer_wait); | |
2831 | } | |
2832 | ||
4a500fd1 | 2833 | if (ret) { |
d1433deb MX |
2834 | if (!list_empty(&root_log_ctx.list)) |
2835 | list_del_init(&root_log_ctx.list); | |
2836 | ||
c6adc9cc | 2837 | blk_finish_plug(&plug); |
995946dd MX |
2838 | btrfs_set_log_full_commit(root->fs_info, trans); |
2839 | ||
79787eaa JM |
2840 | if (ret != -ENOSPC) { |
2841 | btrfs_abort_transaction(trans, root, ret); | |
2842 | mutex_unlock(&log_root_tree->log_mutex); | |
2843 | goto out; | |
2844 | } | |
4a500fd1 | 2845 | btrfs_wait_marked_extents(log, &log->dirty_log_pages, mark); |
2ab28f32 | 2846 | btrfs_free_logged_extents(log, log_transid); |
4a500fd1 YZ |
2847 | mutex_unlock(&log_root_tree->log_mutex); |
2848 | ret = -EAGAIN; | |
2849 | goto out; | |
2850 | } | |
2851 | ||
d1433deb | 2852 | if (log_root_tree->log_transid_committed >= root_log_ctx.log_transid) { |
3da5ab56 | 2853 | blk_finish_plug(&plug); |
d1433deb MX |
2854 | mutex_unlock(&log_root_tree->log_mutex); |
2855 | ret = root_log_ctx.log_ret; | |
2856 | goto out; | |
2857 | } | |
8b050d35 | 2858 | |
d1433deb | 2859 | index2 = root_log_ctx.log_transid % 2; |
7237f183 | 2860 | if (atomic_read(&log_root_tree->log_commit[index2])) { |
c6adc9cc | 2861 | blk_finish_plug(&plug); |
5ab5e44a FM |
2862 | ret = btrfs_wait_marked_extents(log, &log->dirty_log_pages, |
2863 | mark); | |
50d9aa99 | 2864 | btrfs_wait_logged_extents(trans, log, log_transid); |
60d53eb3 | 2865 | wait_log_commit(log_root_tree, |
d1433deb | 2866 | root_log_ctx.log_transid); |
7237f183 | 2867 | mutex_unlock(&log_root_tree->log_mutex); |
5ab5e44a FM |
2868 | if (!ret) |
2869 | ret = root_log_ctx.log_ret; | |
7237f183 YZ |
2870 | goto out; |
2871 | } | |
d1433deb | 2872 | ASSERT(root_log_ctx.log_transid == log_root_tree->log_transid); |
7237f183 YZ |
2873 | atomic_set(&log_root_tree->log_commit[index2], 1); |
2874 | ||
12fcfd22 | 2875 | if (atomic_read(&log_root_tree->log_commit[(index2 + 1) % 2])) { |
60d53eb3 | 2876 | wait_log_commit(log_root_tree, |
d1433deb | 2877 | root_log_ctx.log_transid - 1); |
12fcfd22 CM |
2878 | } |
2879 | ||
60d53eb3 | 2880 | wait_for_writer(log_root_tree); |
7237f183 | 2881 | |
12fcfd22 CM |
2882 | /* |
2883 | * now that we've moved on to the tree of log tree roots, | |
2884 | * check the full commit flag again | |
2885 | */ | |
995946dd | 2886 | if (btrfs_need_log_full_commit(root->fs_info, trans)) { |
c6adc9cc | 2887 | blk_finish_plug(&plug); |
8cef4e16 | 2888 | btrfs_wait_marked_extents(log, &log->dirty_log_pages, mark); |
2ab28f32 | 2889 | btrfs_free_logged_extents(log, log_transid); |
12fcfd22 CM |
2890 | mutex_unlock(&log_root_tree->log_mutex); |
2891 | ret = -EAGAIN; | |
2892 | goto out_wake_log_root; | |
2893 | } | |
7237f183 | 2894 | |
c6adc9cc MX |
2895 | ret = btrfs_write_marked_extents(log_root_tree, |
2896 | &log_root_tree->dirty_log_pages, | |
2897 | EXTENT_DIRTY | EXTENT_NEW); | |
2898 | blk_finish_plug(&plug); | |
79787eaa | 2899 | if (ret) { |
995946dd | 2900 | btrfs_set_log_full_commit(root->fs_info, trans); |
79787eaa | 2901 | btrfs_abort_transaction(trans, root, ret); |
2ab28f32 | 2902 | btrfs_free_logged_extents(log, log_transid); |
79787eaa JM |
2903 | mutex_unlock(&log_root_tree->log_mutex); |
2904 | goto out_wake_log_root; | |
2905 | } | |
5ab5e44a FM |
2906 | ret = btrfs_wait_marked_extents(log, &log->dirty_log_pages, mark); |
2907 | if (!ret) | |
2908 | ret = btrfs_wait_marked_extents(log_root_tree, | |
2909 | &log_root_tree->dirty_log_pages, | |
2910 | EXTENT_NEW | EXTENT_DIRTY); | |
2911 | if (ret) { | |
2912 | btrfs_set_log_full_commit(root->fs_info, trans); | |
2913 | btrfs_free_logged_extents(log, log_transid); | |
2914 | mutex_unlock(&log_root_tree->log_mutex); | |
2915 | goto out_wake_log_root; | |
2916 | } | |
50d9aa99 | 2917 | btrfs_wait_logged_extents(trans, log, log_transid); |
e02119d5 | 2918 | |
6c41761f | 2919 | btrfs_set_super_log_root(root->fs_info->super_for_commit, |
7237f183 | 2920 | log_root_tree->node->start); |
6c41761f | 2921 | btrfs_set_super_log_root_level(root->fs_info->super_for_commit, |
7237f183 | 2922 | btrfs_header_level(log_root_tree->node)); |
e02119d5 | 2923 | |
7237f183 | 2924 | log_root_tree->log_transid++; |
7237f183 YZ |
2925 | mutex_unlock(&log_root_tree->log_mutex); |
2926 | ||
2927 | /* | |
2928 | * nobody else is going to jump in and write the the ctree | |
2929 | * super here because the log_commit atomic below is protecting | |
2930 | * us. We must be called with a transaction handle pinning | |
2931 | * the running transaction open, so a full commit can't hop | |
2932 | * in and cause problems either. | |
2933 | */ | |
5af3e8cc | 2934 | ret = write_ctree_super(trans, root->fs_info->tree_root, 1); |
5af3e8cc | 2935 | if (ret) { |
995946dd | 2936 | btrfs_set_log_full_commit(root->fs_info, trans); |
5af3e8cc SB |
2937 | btrfs_abort_transaction(trans, root, ret); |
2938 | goto out_wake_log_root; | |
2939 | } | |
7237f183 | 2940 | |
257c62e1 CM |
2941 | mutex_lock(&root->log_mutex); |
2942 | if (root->last_log_commit < log_transid) | |
2943 | root->last_log_commit = log_transid; | |
2944 | mutex_unlock(&root->log_mutex); | |
2945 | ||
12fcfd22 | 2946 | out_wake_log_root: |
8b050d35 MX |
2947 | /* |
2948 | * We needn't get log_mutex here because we are sure all | |
2949 | * the other tasks are blocked. | |
2950 | */ | |
2951 | btrfs_remove_all_log_ctxs(log_root_tree, index2, ret); | |
2952 | ||
d1433deb MX |
2953 | mutex_lock(&log_root_tree->log_mutex); |
2954 | log_root_tree->log_transid_committed++; | |
7237f183 | 2955 | atomic_set(&log_root_tree->log_commit[index2], 0); |
d1433deb MX |
2956 | mutex_unlock(&log_root_tree->log_mutex); |
2957 | ||
33a9eca7 DS |
2958 | /* |
2959 | * The barrier before waitqueue_active is implied by mutex_unlock | |
2960 | */ | |
7237f183 YZ |
2961 | if (waitqueue_active(&log_root_tree->log_commit_wait[index2])) |
2962 | wake_up(&log_root_tree->log_commit_wait[index2]); | |
e02119d5 | 2963 | out: |
8b050d35 MX |
2964 | /* See above. */ |
2965 | btrfs_remove_all_log_ctxs(root, index1, ret); | |
2966 | ||
d1433deb MX |
2967 | mutex_lock(&root->log_mutex); |
2968 | root->log_transid_committed++; | |
7237f183 | 2969 | atomic_set(&root->log_commit[index1], 0); |
d1433deb | 2970 | mutex_unlock(&root->log_mutex); |
8b050d35 | 2971 | |
33a9eca7 DS |
2972 | /* |
2973 | * The barrier before waitqueue_active is implied by mutex_unlock | |
2974 | */ | |
7237f183 YZ |
2975 | if (waitqueue_active(&root->log_commit_wait[index1])) |
2976 | wake_up(&root->log_commit_wait[index1]); | |
b31eabd8 | 2977 | return ret; |
e02119d5 CM |
2978 | } |
2979 | ||
4a500fd1 YZ |
2980 | static void free_log_tree(struct btrfs_trans_handle *trans, |
2981 | struct btrfs_root *log) | |
e02119d5 CM |
2982 | { |
2983 | int ret; | |
d0c803c4 CM |
2984 | u64 start; |
2985 | u64 end; | |
e02119d5 CM |
2986 | struct walk_control wc = { |
2987 | .free = 1, | |
2988 | .process_func = process_one_buffer | |
2989 | }; | |
2990 | ||
681ae509 JB |
2991 | ret = walk_log_tree(trans, log, &wc); |
2992 | /* I don't think this can happen but just in case */ | |
2993 | if (ret) | |
2994 | btrfs_abort_transaction(trans, log, ret); | |
e02119d5 | 2995 | |
d397712b | 2996 | while (1) { |
d0c803c4 | 2997 | ret = find_first_extent_bit(&log->dirty_log_pages, |
e6138876 JB |
2998 | 0, &start, &end, EXTENT_DIRTY | EXTENT_NEW, |
2999 | NULL); | |
d0c803c4 CM |
3000 | if (ret) |
3001 | break; | |
3002 | ||
8cef4e16 YZ |
3003 | clear_extent_bits(&log->dirty_log_pages, start, end, |
3004 | EXTENT_DIRTY | EXTENT_NEW, GFP_NOFS); | |
d0c803c4 CM |
3005 | } |
3006 | ||
2ab28f32 JB |
3007 | /* |
3008 | * We may have short-circuited the log tree with the full commit logic | |
3009 | * and left ordered extents on our list, so clear these out to keep us | |
3010 | * from leaking inodes and memory. | |
3011 | */ | |
3012 | btrfs_free_logged_extents(log, 0); | |
3013 | btrfs_free_logged_extents(log, 1); | |
3014 | ||
7237f183 YZ |
3015 | free_extent_buffer(log->node); |
3016 | kfree(log); | |
4a500fd1 YZ |
3017 | } |
3018 | ||
3019 | /* | |
3020 | * free all the extents used by the tree log. This should be called | |
3021 | * at commit time of the full transaction | |
3022 | */ | |
3023 | int btrfs_free_log(struct btrfs_trans_handle *trans, struct btrfs_root *root) | |
3024 | { | |
3025 | if (root->log_root) { | |
3026 | free_log_tree(trans, root->log_root); | |
3027 | root->log_root = NULL; | |
3028 | } | |
3029 | return 0; | |
3030 | } | |
3031 | ||
3032 | int btrfs_free_log_root_tree(struct btrfs_trans_handle *trans, | |
3033 | struct btrfs_fs_info *fs_info) | |
3034 | { | |
3035 | if (fs_info->log_root_tree) { | |
3036 | free_log_tree(trans, fs_info->log_root_tree); | |
3037 | fs_info->log_root_tree = NULL; | |
3038 | } | |
e02119d5 CM |
3039 | return 0; |
3040 | } | |
3041 | ||
e02119d5 CM |
3042 | /* |
3043 | * If both a file and directory are logged, and unlinks or renames are | |
3044 | * mixed in, we have a few interesting corners: | |
3045 | * | |
3046 | * create file X in dir Y | |
3047 | * link file X to X.link in dir Y | |
3048 | * fsync file X | |
3049 | * unlink file X but leave X.link | |
3050 | * fsync dir Y | |
3051 | * | |
3052 | * After a crash we would expect only X.link to exist. But file X | |
3053 | * didn't get fsync'd again so the log has back refs for X and X.link. | |
3054 | * | |
3055 | * We solve this by removing directory entries and inode backrefs from the | |
3056 | * log when a file that was logged in the current transaction is | |
3057 | * unlinked. Any later fsync will include the updated log entries, and | |
3058 | * we'll be able to reconstruct the proper directory items from backrefs. | |
3059 | * | |
3060 | * This optimizations allows us to avoid relogging the entire inode | |
3061 | * or the entire directory. | |
3062 | */ | |
3063 | int btrfs_del_dir_entries_in_log(struct btrfs_trans_handle *trans, | |
3064 | struct btrfs_root *root, | |
3065 | const char *name, int name_len, | |
3066 | struct inode *dir, u64 index) | |
3067 | { | |
3068 | struct btrfs_root *log; | |
3069 | struct btrfs_dir_item *di; | |
3070 | struct btrfs_path *path; | |
3071 | int ret; | |
4a500fd1 | 3072 | int err = 0; |
e02119d5 | 3073 | int bytes_del = 0; |
33345d01 | 3074 | u64 dir_ino = btrfs_ino(dir); |
e02119d5 | 3075 | |
3a5f1d45 CM |
3076 | if (BTRFS_I(dir)->logged_trans < trans->transid) |
3077 | return 0; | |
3078 | ||
e02119d5 CM |
3079 | ret = join_running_log_trans(root); |
3080 | if (ret) | |
3081 | return 0; | |
3082 | ||
3083 | mutex_lock(&BTRFS_I(dir)->log_mutex); | |
3084 | ||
3085 | log = root->log_root; | |
3086 | path = btrfs_alloc_path(); | |
a62f44a5 TI |
3087 | if (!path) { |
3088 | err = -ENOMEM; | |
3089 | goto out_unlock; | |
3090 | } | |
2a29edc6 | 3091 | |
33345d01 | 3092 | di = btrfs_lookup_dir_item(trans, log, path, dir_ino, |
e02119d5 | 3093 | name, name_len, -1); |
4a500fd1 YZ |
3094 | if (IS_ERR(di)) { |
3095 | err = PTR_ERR(di); | |
3096 | goto fail; | |
3097 | } | |
3098 | if (di) { | |
e02119d5 CM |
3099 | ret = btrfs_delete_one_dir_name(trans, log, path, di); |
3100 | bytes_del += name_len; | |
3650860b JB |
3101 | if (ret) { |
3102 | err = ret; | |
3103 | goto fail; | |
3104 | } | |
e02119d5 | 3105 | } |
b3b4aa74 | 3106 | btrfs_release_path(path); |
33345d01 | 3107 | di = btrfs_lookup_dir_index_item(trans, log, path, dir_ino, |
e02119d5 | 3108 | index, name, name_len, -1); |
4a500fd1 YZ |
3109 | if (IS_ERR(di)) { |
3110 | err = PTR_ERR(di); | |
3111 | goto fail; | |
3112 | } | |
3113 | if (di) { | |
e02119d5 CM |
3114 | ret = btrfs_delete_one_dir_name(trans, log, path, di); |
3115 | bytes_del += name_len; | |
3650860b JB |
3116 | if (ret) { |
3117 | err = ret; | |
3118 | goto fail; | |
3119 | } | |
e02119d5 CM |
3120 | } |
3121 | ||
3122 | /* update the directory size in the log to reflect the names | |
3123 | * we have removed | |
3124 | */ | |
3125 | if (bytes_del) { | |
3126 | struct btrfs_key key; | |
3127 | ||
33345d01 | 3128 | key.objectid = dir_ino; |
e02119d5 CM |
3129 | key.offset = 0; |
3130 | key.type = BTRFS_INODE_ITEM_KEY; | |
b3b4aa74 | 3131 | btrfs_release_path(path); |
e02119d5 CM |
3132 | |
3133 | ret = btrfs_search_slot(trans, log, &key, path, 0, 1); | |
4a500fd1 YZ |
3134 | if (ret < 0) { |
3135 | err = ret; | |
3136 | goto fail; | |
3137 | } | |
e02119d5 CM |
3138 | if (ret == 0) { |
3139 | struct btrfs_inode_item *item; | |
3140 | u64 i_size; | |
3141 | ||
3142 | item = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
3143 | struct btrfs_inode_item); | |
3144 | i_size = btrfs_inode_size(path->nodes[0], item); | |
3145 | if (i_size > bytes_del) | |
3146 | i_size -= bytes_del; | |
3147 | else | |
3148 | i_size = 0; | |
3149 | btrfs_set_inode_size(path->nodes[0], item, i_size); | |
3150 | btrfs_mark_buffer_dirty(path->nodes[0]); | |
3151 | } else | |
3152 | ret = 0; | |
b3b4aa74 | 3153 | btrfs_release_path(path); |
e02119d5 | 3154 | } |
4a500fd1 | 3155 | fail: |
e02119d5 | 3156 | btrfs_free_path(path); |
a62f44a5 | 3157 | out_unlock: |
e02119d5 | 3158 | mutex_unlock(&BTRFS_I(dir)->log_mutex); |
4a500fd1 | 3159 | if (ret == -ENOSPC) { |
995946dd | 3160 | btrfs_set_log_full_commit(root->fs_info, trans); |
4a500fd1 | 3161 | ret = 0; |
79787eaa JM |
3162 | } else if (ret < 0) |
3163 | btrfs_abort_transaction(trans, root, ret); | |
3164 | ||
12fcfd22 | 3165 | btrfs_end_log_trans(root); |
e02119d5 | 3166 | |
411fc6bc | 3167 | return err; |
e02119d5 CM |
3168 | } |
3169 | ||
3170 | /* see comments for btrfs_del_dir_entries_in_log */ | |
3171 | int btrfs_del_inode_ref_in_log(struct btrfs_trans_handle *trans, | |
3172 | struct btrfs_root *root, | |
3173 | const char *name, int name_len, | |
3174 | struct inode *inode, u64 dirid) | |
3175 | { | |
3176 | struct btrfs_root *log; | |
3177 | u64 index; | |
3178 | int ret; | |
3179 | ||
3a5f1d45 CM |
3180 | if (BTRFS_I(inode)->logged_trans < trans->transid) |
3181 | return 0; | |
3182 | ||
e02119d5 CM |
3183 | ret = join_running_log_trans(root); |
3184 | if (ret) | |
3185 | return 0; | |
3186 | log = root->log_root; | |
3187 | mutex_lock(&BTRFS_I(inode)->log_mutex); | |
3188 | ||
33345d01 | 3189 | ret = btrfs_del_inode_ref(trans, log, name, name_len, btrfs_ino(inode), |
e02119d5 CM |
3190 | dirid, &index); |
3191 | mutex_unlock(&BTRFS_I(inode)->log_mutex); | |
4a500fd1 | 3192 | if (ret == -ENOSPC) { |
995946dd | 3193 | btrfs_set_log_full_commit(root->fs_info, trans); |
4a500fd1 | 3194 | ret = 0; |
79787eaa JM |
3195 | } else if (ret < 0 && ret != -ENOENT) |
3196 | btrfs_abort_transaction(trans, root, ret); | |
12fcfd22 | 3197 | btrfs_end_log_trans(root); |
e02119d5 | 3198 | |
e02119d5 CM |
3199 | return ret; |
3200 | } | |
3201 | ||
3202 | /* | |
3203 | * creates a range item in the log for 'dirid'. first_offset and | |
3204 | * last_offset tell us which parts of the key space the log should | |
3205 | * be considered authoritative for. | |
3206 | */ | |
3207 | static noinline int insert_dir_log_key(struct btrfs_trans_handle *trans, | |
3208 | struct btrfs_root *log, | |
3209 | struct btrfs_path *path, | |
3210 | int key_type, u64 dirid, | |
3211 | u64 first_offset, u64 last_offset) | |
3212 | { | |
3213 | int ret; | |
3214 | struct btrfs_key key; | |
3215 | struct btrfs_dir_log_item *item; | |
3216 | ||
3217 | key.objectid = dirid; | |
3218 | key.offset = first_offset; | |
3219 | if (key_type == BTRFS_DIR_ITEM_KEY) | |
3220 | key.type = BTRFS_DIR_LOG_ITEM_KEY; | |
3221 | else | |
3222 | key.type = BTRFS_DIR_LOG_INDEX_KEY; | |
3223 | ret = btrfs_insert_empty_item(trans, log, path, &key, sizeof(*item)); | |
4a500fd1 YZ |
3224 | if (ret) |
3225 | return ret; | |
e02119d5 CM |
3226 | |
3227 | item = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
3228 | struct btrfs_dir_log_item); | |
3229 | btrfs_set_dir_log_end(path->nodes[0], item, last_offset); | |
3230 | btrfs_mark_buffer_dirty(path->nodes[0]); | |
b3b4aa74 | 3231 | btrfs_release_path(path); |
e02119d5 CM |
3232 | return 0; |
3233 | } | |
3234 | ||
3235 | /* | |
3236 | * log all the items included in the current transaction for a given | |
3237 | * directory. This also creates the range items in the log tree required | |
3238 | * to replay anything deleted before the fsync | |
3239 | */ | |
3240 | static noinline int log_dir_items(struct btrfs_trans_handle *trans, | |
3241 | struct btrfs_root *root, struct inode *inode, | |
3242 | struct btrfs_path *path, | |
3243 | struct btrfs_path *dst_path, int key_type, | |
2f2ff0ee | 3244 | struct btrfs_log_ctx *ctx, |
e02119d5 CM |
3245 | u64 min_offset, u64 *last_offset_ret) |
3246 | { | |
3247 | struct btrfs_key min_key; | |
e02119d5 CM |
3248 | struct btrfs_root *log = root->log_root; |
3249 | struct extent_buffer *src; | |
4a500fd1 | 3250 | int err = 0; |
e02119d5 CM |
3251 | int ret; |
3252 | int i; | |
3253 | int nritems; | |
3254 | u64 first_offset = min_offset; | |
3255 | u64 last_offset = (u64)-1; | |
33345d01 | 3256 | u64 ino = btrfs_ino(inode); |
e02119d5 CM |
3257 | |
3258 | log = root->log_root; | |
e02119d5 | 3259 | |
33345d01 | 3260 | min_key.objectid = ino; |
e02119d5 CM |
3261 | min_key.type = key_type; |
3262 | min_key.offset = min_offset; | |
3263 | ||
6174d3cb | 3264 | ret = btrfs_search_forward(root, &min_key, path, trans->transid); |
e02119d5 CM |
3265 | |
3266 | /* | |
3267 | * we didn't find anything from this transaction, see if there | |
3268 | * is anything at all | |
3269 | */ | |
33345d01 LZ |
3270 | if (ret != 0 || min_key.objectid != ino || min_key.type != key_type) { |
3271 | min_key.objectid = ino; | |
e02119d5 CM |
3272 | min_key.type = key_type; |
3273 | min_key.offset = (u64)-1; | |
b3b4aa74 | 3274 | btrfs_release_path(path); |
e02119d5 CM |
3275 | ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0); |
3276 | if (ret < 0) { | |
b3b4aa74 | 3277 | btrfs_release_path(path); |
e02119d5 CM |
3278 | return ret; |
3279 | } | |
33345d01 | 3280 | ret = btrfs_previous_item(root, path, ino, key_type); |
e02119d5 CM |
3281 | |
3282 | /* if ret == 0 there are items for this type, | |
3283 | * create a range to tell us the last key of this type. | |
3284 | * otherwise, there are no items in this directory after | |
3285 | * *min_offset, and we create a range to indicate that. | |
3286 | */ | |
3287 | if (ret == 0) { | |
3288 | struct btrfs_key tmp; | |
3289 | btrfs_item_key_to_cpu(path->nodes[0], &tmp, | |
3290 | path->slots[0]); | |
d397712b | 3291 | if (key_type == tmp.type) |
e02119d5 | 3292 | first_offset = max(min_offset, tmp.offset) + 1; |
e02119d5 CM |
3293 | } |
3294 | goto done; | |
3295 | } | |
3296 | ||
3297 | /* go backward to find any previous key */ | |
33345d01 | 3298 | ret = btrfs_previous_item(root, path, ino, key_type); |
e02119d5 CM |
3299 | if (ret == 0) { |
3300 | struct btrfs_key tmp; | |
3301 | btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]); | |
3302 | if (key_type == tmp.type) { | |
3303 | first_offset = tmp.offset; | |
3304 | ret = overwrite_item(trans, log, dst_path, | |
3305 | path->nodes[0], path->slots[0], | |
3306 | &tmp); | |
4a500fd1 YZ |
3307 | if (ret) { |
3308 | err = ret; | |
3309 | goto done; | |
3310 | } | |
e02119d5 CM |
3311 | } |
3312 | } | |
b3b4aa74 | 3313 | btrfs_release_path(path); |
e02119d5 CM |
3314 | |
3315 | /* find the first key from this transaction again */ | |
3316 | ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0); | |
fae7f21c | 3317 | if (WARN_ON(ret != 0)) |
e02119d5 | 3318 | goto done; |
e02119d5 CM |
3319 | |
3320 | /* | |
3321 | * we have a block from this transaction, log every item in it | |
3322 | * from our directory | |
3323 | */ | |
d397712b | 3324 | while (1) { |
e02119d5 CM |
3325 | struct btrfs_key tmp; |
3326 | src = path->nodes[0]; | |
3327 | nritems = btrfs_header_nritems(src); | |
3328 | for (i = path->slots[0]; i < nritems; i++) { | |
2f2ff0ee FM |
3329 | struct btrfs_dir_item *di; |
3330 | ||
e02119d5 CM |
3331 | btrfs_item_key_to_cpu(src, &min_key, i); |
3332 | ||
33345d01 | 3333 | if (min_key.objectid != ino || min_key.type != key_type) |
e02119d5 CM |
3334 | goto done; |
3335 | ret = overwrite_item(trans, log, dst_path, src, i, | |
3336 | &min_key); | |
4a500fd1 YZ |
3337 | if (ret) { |
3338 | err = ret; | |
3339 | goto done; | |
3340 | } | |
2f2ff0ee FM |
3341 | |
3342 | /* | |
3343 | * We must make sure that when we log a directory entry, | |
3344 | * the corresponding inode, after log replay, has a | |
3345 | * matching link count. For example: | |
3346 | * | |
3347 | * touch foo | |
3348 | * mkdir mydir | |
3349 | * sync | |
3350 | * ln foo mydir/bar | |
3351 | * xfs_io -c "fsync" mydir | |
3352 | * <crash> | |
3353 | * <mount fs and log replay> | |
3354 | * | |
3355 | * Would result in a fsync log that when replayed, our | |
3356 | * file inode would have a link count of 1, but we get | |
3357 | * two directory entries pointing to the same inode. | |
3358 | * After removing one of the names, it would not be | |
3359 | * possible to remove the other name, which resulted | |
3360 | * always in stale file handle errors, and would not | |
3361 | * be possible to rmdir the parent directory, since | |
3362 | * its i_size could never decrement to the value | |
3363 | * BTRFS_EMPTY_DIR_SIZE, resulting in -ENOTEMPTY errors. | |
3364 | */ | |
3365 | di = btrfs_item_ptr(src, i, struct btrfs_dir_item); | |
3366 | btrfs_dir_item_key_to_cpu(src, di, &tmp); | |
3367 | if (ctx && | |
3368 | (btrfs_dir_transid(src, di) == trans->transid || | |
3369 | btrfs_dir_type(src, di) == BTRFS_FT_DIR) && | |
3370 | tmp.type != BTRFS_ROOT_ITEM_KEY) | |
3371 | ctx->log_new_dentries = true; | |
e02119d5 CM |
3372 | } |
3373 | path->slots[0] = nritems; | |
3374 | ||
3375 | /* | |
3376 | * look ahead to the next item and see if it is also | |
3377 | * from this directory and from this transaction | |
3378 | */ | |
3379 | ret = btrfs_next_leaf(root, path); | |
3380 | if (ret == 1) { | |
3381 | last_offset = (u64)-1; | |
3382 | goto done; | |
3383 | } | |
3384 | btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]); | |
33345d01 | 3385 | if (tmp.objectid != ino || tmp.type != key_type) { |
e02119d5 CM |
3386 | last_offset = (u64)-1; |
3387 | goto done; | |
3388 | } | |
3389 | if (btrfs_header_generation(path->nodes[0]) != trans->transid) { | |
3390 | ret = overwrite_item(trans, log, dst_path, | |
3391 | path->nodes[0], path->slots[0], | |
3392 | &tmp); | |
4a500fd1 YZ |
3393 | if (ret) |
3394 | err = ret; | |
3395 | else | |
3396 | last_offset = tmp.offset; | |
e02119d5 CM |
3397 | goto done; |
3398 | } | |
3399 | } | |
3400 | done: | |
b3b4aa74 DS |
3401 | btrfs_release_path(path); |
3402 | btrfs_release_path(dst_path); | |
e02119d5 | 3403 | |
4a500fd1 YZ |
3404 | if (err == 0) { |
3405 | *last_offset_ret = last_offset; | |
3406 | /* | |
3407 | * insert the log range keys to indicate where the log | |
3408 | * is valid | |
3409 | */ | |
3410 | ret = insert_dir_log_key(trans, log, path, key_type, | |
33345d01 | 3411 | ino, first_offset, last_offset); |
4a500fd1 YZ |
3412 | if (ret) |
3413 | err = ret; | |
3414 | } | |
3415 | return err; | |
e02119d5 CM |
3416 | } |
3417 | ||
3418 | /* | |
3419 | * logging directories is very similar to logging inodes, We find all the items | |
3420 | * from the current transaction and write them to the log. | |
3421 | * | |
3422 | * The recovery code scans the directory in the subvolume, and if it finds a | |
3423 | * key in the range logged that is not present in the log tree, then it means | |
3424 | * that dir entry was unlinked during the transaction. | |
3425 | * | |
3426 | * In order for that scan to work, we must include one key smaller than | |
3427 | * the smallest logged by this transaction and one key larger than the largest | |
3428 | * key logged by this transaction. | |
3429 | */ | |
3430 | static noinline int log_directory_changes(struct btrfs_trans_handle *trans, | |
3431 | struct btrfs_root *root, struct inode *inode, | |
3432 | struct btrfs_path *path, | |
2f2ff0ee FM |
3433 | struct btrfs_path *dst_path, |
3434 | struct btrfs_log_ctx *ctx) | |
e02119d5 CM |
3435 | { |
3436 | u64 min_key; | |
3437 | u64 max_key; | |
3438 | int ret; | |
3439 | int key_type = BTRFS_DIR_ITEM_KEY; | |
3440 | ||
3441 | again: | |
3442 | min_key = 0; | |
3443 | max_key = 0; | |
d397712b | 3444 | while (1) { |
e02119d5 | 3445 | ret = log_dir_items(trans, root, inode, path, |
2f2ff0ee | 3446 | dst_path, key_type, ctx, min_key, |
e02119d5 | 3447 | &max_key); |
4a500fd1 YZ |
3448 | if (ret) |
3449 | return ret; | |
e02119d5 CM |
3450 | if (max_key == (u64)-1) |
3451 | break; | |
3452 | min_key = max_key + 1; | |
3453 | } | |
3454 | ||
3455 | if (key_type == BTRFS_DIR_ITEM_KEY) { | |
3456 | key_type = BTRFS_DIR_INDEX_KEY; | |
3457 | goto again; | |
3458 | } | |
3459 | return 0; | |
3460 | } | |
3461 | ||
3462 | /* | |
3463 | * a helper function to drop items from the log before we relog an | |
3464 | * inode. max_key_type indicates the highest item type to remove. | |
3465 | * This cannot be run for file data extents because it does not | |
3466 | * free the extents they point to. | |
3467 | */ | |
3468 | static int drop_objectid_items(struct btrfs_trans_handle *trans, | |
3469 | struct btrfs_root *log, | |
3470 | struct btrfs_path *path, | |
3471 | u64 objectid, int max_key_type) | |
3472 | { | |
3473 | int ret; | |
3474 | struct btrfs_key key; | |
3475 | struct btrfs_key found_key; | |
18ec90d6 | 3476 | int start_slot; |
e02119d5 CM |
3477 | |
3478 | key.objectid = objectid; | |
3479 | key.type = max_key_type; | |
3480 | key.offset = (u64)-1; | |
3481 | ||
d397712b | 3482 | while (1) { |
e02119d5 | 3483 | ret = btrfs_search_slot(trans, log, &key, path, -1, 1); |
3650860b | 3484 | BUG_ON(ret == 0); /* Logic error */ |
4a500fd1 | 3485 | if (ret < 0) |
e02119d5 CM |
3486 | break; |
3487 | ||
3488 | if (path->slots[0] == 0) | |
3489 | break; | |
3490 | ||
3491 | path->slots[0]--; | |
3492 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, | |
3493 | path->slots[0]); | |
3494 | ||
3495 | if (found_key.objectid != objectid) | |
3496 | break; | |
3497 | ||
18ec90d6 JB |
3498 | found_key.offset = 0; |
3499 | found_key.type = 0; | |
3500 | ret = btrfs_bin_search(path->nodes[0], &found_key, 0, | |
3501 | &start_slot); | |
3502 | ||
3503 | ret = btrfs_del_items(trans, log, path, start_slot, | |
3504 | path->slots[0] - start_slot + 1); | |
3505 | /* | |
3506 | * If start slot isn't 0 then we don't need to re-search, we've | |
3507 | * found the last guy with the objectid in this tree. | |
3508 | */ | |
3509 | if (ret || start_slot != 0) | |
65a246c5 | 3510 | break; |
b3b4aa74 | 3511 | btrfs_release_path(path); |
e02119d5 | 3512 | } |
b3b4aa74 | 3513 | btrfs_release_path(path); |
5bdbeb21 JB |
3514 | if (ret > 0) |
3515 | ret = 0; | |
4a500fd1 | 3516 | return ret; |
e02119d5 CM |
3517 | } |
3518 | ||
94edf4ae JB |
3519 | static void fill_inode_item(struct btrfs_trans_handle *trans, |
3520 | struct extent_buffer *leaf, | |
3521 | struct btrfs_inode_item *item, | |
1a4bcf47 FM |
3522 | struct inode *inode, int log_inode_only, |
3523 | u64 logged_isize) | |
94edf4ae | 3524 | { |
0b1c6cca JB |
3525 | struct btrfs_map_token token; |
3526 | ||
3527 | btrfs_init_map_token(&token); | |
94edf4ae JB |
3528 | |
3529 | if (log_inode_only) { | |
3530 | /* set the generation to zero so the recover code | |
3531 | * can tell the difference between an logging | |
3532 | * just to say 'this inode exists' and a logging | |
3533 | * to say 'update this inode with these values' | |
3534 | */ | |
0b1c6cca | 3535 | btrfs_set_token_inode_generation(leaf, item, 0, &token); |
1a4bcf47 | 3536 | btrfs_set_token_inode_size(leaf, item, logged_isize, &token); |
94edf4ae | 3537 | } else { |
0b1c6cca JB |
3538 | btrfs_set_token_inode_generation(leaf, item, |
3539 | BTRFS_I(inode)->generation, | |
3540 | &token); | |
3541 | btrfs_set_token_inode_size(leaf, item, inode->i_size, &token); | |
3542 | } | |
3543 | ||
3544 | btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token); | |
3545 | btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token); | |
3546 | btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token); | |
3547 | btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token); | |
3548 | ||
a937b979 | 3549 | btrfs_set_token_timespec_sec(leaf, &item->atime, |
0b1c6cca | 3550 | inode->i_atime.tv_sec, &token); |
a937b979 | 3551 | btrfs_set_token_timespec_nsec(leaf, &item->atime, |
0b1c6cca JB |
3552 | inode->i_atime.tv_nsec, &token); |
3553 | ||
a937b979 | 3554 | btrfs_set_token_timespec_sec(leaf, &item->mtime, |
0b1c6cca | 3555 | inode->i_mtime.tv_sec, &token); |
a937b979 | 3556 | btrfs_set_token_timespec_nsec(leaf, &item->mtime, |
0b1c6cca JB |
3557 | inode->i_mtime.tv_nsec, &token); |
3558 | ||
a937b979 | 3559 | btrfs_set_token_timespec_sec(leaf, &item->ctime, |
0b1c6cca | 3560 | inode->i_ctime.tv_sec, &token); |
a937b979 | 3561 | btrfs_set_token_timespec_nsec(leaf, &item->ctime, |
0b1c6cca JB |
3562 | inode->i_ctime.tv_nsec, &token); |
3563 | ||
3564 | btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode), | |
3565 | &token); | |
3566 | ||
3567 | btrfs_set_token_inode_sequence(leaf, item, inode->i_version, &token); | |
3568 | btrfs_set_token_inode_transid(leaf, item, trans->transid, &token); | |
3569 | btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token); | |
3570 | btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token); | |
3571 | btrfs_set_token_inode_block_group(leaf, item, 0, &token); | |
94edf4ae JB |
3572 | } |
3573 | ||
a95249b3 JB |
3574 | static int log_inode_item(struct btrfs_trans_handle *trans, |
3575 | struct btrfs_root *log, struct btrfs_path *path, | |
3576 | struct inode *inode) | |
3577 | { | |
3578 | struct btrfs_inode_item *inode_item; | |
a95249b3 JB |
3579 | int ret; |
3580 | ||
efd0c405 FDBM |
3581 | ret = btrfs_insert_empty_item(trans, log, path, |
3582 | &BTRFS_I(inode)->location, | |
a95249b3 JB |
3583 | sizeof(*inode_item)); |
3584 | if (ret && ret != -EEXIST) | |
3585 | return ret; | |
3586 | inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
3587 | struct btrfs_inode_item); | |
1a4bcf47 | 3588 | fill_inode_item(trans, path->nodes[0], inode_item, inode, 0, 0); |
a95249b3 JB |
3589 | btrfs_release_path(path); |
3590 | return 0; | |
3591 | } | |
3592 | ||
31ff1cd2 | 3593 | static noinline int copy_items(struct btrfs_trans_handle *trans, |
d2794405 | 3594 | struct inode *inode, |
31ff1cd2 | 3595 | struct btrfs_path *dst_path, |
16e7549f | 3596 | struct btrfs_path *src_path, u64 *last_extent, |
1a4bcf47 FM |
3597 | int start_slot, int nr, int inode_only, |
3598 | u64 logged_isize) | |
31ff1cd2 CM |
3599 | { |
3600 | unsigned long src_offset; | |
3601 | unsigned long dst_offset; | |
d2794405 | 3602 | struct btrfs_root *log = BTRFS_I(inode)->root->log_root; |
31ff1cd2 CM |
3603 | struct btrfs_file_extent_item *extent; |
3604 | struct btrfs_inode_item *inode_item; | |
16e7549f JB |
3605 | struct extent_buffer *src = src_path->nodes[0]; |
3606 | struct btrfs_key first_key, last_key, key; | |
31ff1cd2 CM |
3607 | int ret; |
3608 | struct btrfs_key *ins_keys; | |
3609 | u32 *ins_sizes; | |
3610 | char *ins_data; | |
3611 | int i; | |
d20f7043 | 3612 | struct list_head ordered_sums; |
d2794405 | 3613 | int skip_csum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM; |
16e7549f | 3614 | bool has_extents = false; |
74121f7c | 3615 | bool need_find_last_extent = true; |
16e7549f | 3616 | bool done = false; |
d20f7043 CM |
3617 | |
3618 | INIT_LIST_HEAD(&ordered_sums); | |
31ff1cd2 CM |
3619 | |
3620 | ins_data = kmalloc(nr * sizeof(struct btrfs_key) + | |
3621 | nr * sizeof(u32), GFP_NOFS); | |
2a29edc6 | 3622 | if (!ins_data) |
3623 | return -ENOMEM; | |
3624 | ||
16e7549f JB |
3625 | first_key.objectid = (u64)-1; |
3626 | ||
31ff1cd2 CM |
3627 | ins_sizes = (u32 *)ins_data; |
3628 | ins_keys = (struct btrfs_key *)(ins_data + nr * sizeof(u32)); | |
3629 | ||
3630 | for (i = 0; i < nr; i++) { | |
3631 | ins_sizes[i] = btrfs_item_size_nr(src, i + start_slot); | |
3632 | btrfs_item_key_to_cpu(src, ins_keys + i, i + start_slot); | |
3633 | } | |
3634 | ret = btrfs_insert_empty_items(trans, log, dst_path, | |
3635 | ins_keys, ins_sizes, nr); | |
4a500fd1 YZ |
3636 | if (ret) { |
3637 | kfree(ins_data); | |
3638 | return ret; | |
3639 | } | |
31ff1cd2 | 3640 | |
5d4f98a2 | 3641 | for (i = 0; i < nr; i++, dst_path->slots[0]++) { |
31ff1cd2 CM |
3642 | dst_offset = btrfs_item_ptr_offset(dst_path->nodes[0], |
3643 | dst_path->slots[0]); | |
3644 | ||
3645 | src_offset = btrfs_item_ptr_offset(src, start_slot + i); | |
3646 | ||
16e7549f JB |
3647 | if ((i == (nr - 1))) |
3648 | last_key = ins_keys[i]; | |
3649 | ||
94edf4ae | 3650 | if (ins_keys[i].type == BTRFS_INODE_ITEM_KEY) { |
31ff1cd2 CM |
3651 | inode_item = btrfs_item_ptr(dst_path->nodes[0], |
3652 | dst_path->slots[0], | |
3653 | struct btrfs_inode_item); | |
94edf4ae | 3654 | fill_inode_item(trans, dst_path->nodes[0], inode_item, |
1a4bcf47 FM |
3655 | inode, inode_only == LOG_INODE_EXISTS, |
3656 | logged_isize); | |
94edf4ae JB |
3657 | } else { |
3658 | copy_extent_buffer(dst_path->nodes[0], src, dst_offset, | |
3659 | src_offset, ins_sizes[i]); | |
31ff1cd2 | 3660 | } |
94edf4ae | 3661 | |
16e7549f JB |
3662 | /* |
3663 | * We set need_find_last_extent here in case we know we were | |
3664 | * processing other items and then walk into the first extent in | |
3665 | * the inode. If we don't hit an extent then nothing changes, | |
3666 | * we'll do the last search the next time around. | |
3667 | */ | |
3668 | if (ins_keys[i].type == BTRFS_EXTENT_DATA_KEY) { | |
3669 | has_extents = true; | |
74121f7c | 3670 | if (first_key.objectid == (u64)-1) |
16e7549f JB |
3671 | first_key = ins_keys[i]; |
3672 | } else { | |
3673 | need_find_last_extent = false; | |
3674 | } | |
3675 | ||
31ff1cd2 CM |
3676 | /* take a reference on file data extents so that truncates |
3677 | * or deletes of this inode don't have to relog the inode | |
3678 | * again | |
3679 | */ | |
962a298f | 3680 | if (ins_keys[i].type == BTRFS_EXTENT_DATA_KEY && |
d2794405 | 3681 | !skip_csum) { |
31ff1cd2 CM |
3682 | int found_type; |
3683 | extent = btrfs_item_ptr(src, start_slot + i, | |
3684 | struct btrfs_file_extent_item); | |
3685 | ||
8e531cdf | 3686 | if (btrfs_file_extent_generation(src, extent) < trans->transid) |
3687 | continue; | |
3688 | ||
31ff1cd2 | 3689 | found_type = btrfs_file_extent_type(src, extent); |
6f1fed77 | 3690 | if (found_type == BTRFS_FILE_EXTENT_REG) { |
5d4f98a2 YZ |
3691 | u64 ds, dl, cs, cl; |
3692 | ds = btrfs_file_extent_disk_bytenr(src, | |
3693 | extent); | |
3694 | /* ds == 0 is a hole */ | |
3695 | if (ds == 0) | |
3696 | continue; | |
3697 | ||
3698 | dl = btrfs_file_extent_disk_num_bytes(src, | |
3699 | extent); | |
3700 | cs = btrfs_file_extent_offset(src, extent); | |
3701 | cl = btrfs_file_extent_num_bytes(src, | |
a419aef8 | 3702 | extent); |
580afd76 CM |
3703 | if (btrfs_file_extent_compression(src, |
3704 | extent)) { | |
3705 | cs = 0; | |
3706 | cl = dl; | |
3707 | } | |
5d4f98a2 YZ |
3708 | |
3709 | ret = btrfs_lookup_csums_range( | |
3710 | log->fs_info->csum_root, | |
3711 | ds + cs, ds + cs + cl - 1, | |
a2de733c | 3712 | &ordered_sums, 0); |
3650860b JB |
3713 | if (ret) { |
3714 | btrfs_release_path(dst_path); | |
3715 | kfree(ins_data); | |
3716 | return ret; | |
3717 | } | |
31ff1cd2 CM |
3718 | } |
3719 | } | |
31ff1cd2 CM |
3720 | } |
3721 | ||
3722 | btrfs_mark_buffer_dirty(dst_path->nodes[0]); | |
b3b4aa74 | 3723 | btrfs_release_path(dst_path); |
31ff1cd2 | 3724 | kfree(ins_data); |
d20f7043 CM |
3725 | |
3726 | /* | |
3727 | * we have to do this after the loop above to avoid changing the | |
3728 | * log tree while trying to change the log tree. | |
3729 | */ | |
4a500fd1 | 3730 | ret = 0; |
d397712b | 3731 | while (!list_empty(&ordered_sums)) { |
d20f7043 CM |
3732 | struct btrfs_ordered_sum *sums = list_entry(ordered_sums.next, |
3733 | struct btrfs_ordered_sum, | |
3734 | list); | |
4a500fd1 YZ |
3735 | if (!ret) |
3736 | ret = btrfs_csum_file_blocks(trans, log, sums); | |
d20f7043 CM |
3737 | list_del(&sums->list); |
3738 | kfree(sums); | |
3739 | } | |
16e7549f JB |
3740 | |
3741 | if (!has_extents) | |
3742 | return ret; | |
3743 | ||
74121f7c FM |
3744 | if (need_find_last_extent && *last_extent == first_key.offset) { |
3745 | /* | |
3746 | * We don't have any leafs between our current one and the one | |
3747 | * we processed before that can have file extent items for our | |
3748 | * inode (and have a generation number smaller than our current | |
3749 | * transaction id). | |
3750 | */ | |
3751 | need_find_last_extent = false; | |
3752 | } | |
3753 | ||
16e7549f JB |
3754 | /* |
3755 | * Because we use btrfs_search_forward we could skip leaves that were | |
3756 | * not modified and then assume *last_extent is valid when it really | |
3757 | * isn't. So back up to the previous leaf and read the end of the last | |
3758 | * extent before we go and fill in holes. | |
3759 | */ | |
3760 | if (need_find_last_extent) { | |
3761 | u64 len; | |
3762 | ||
3763 | ret = btrfs_prev_leaf(BTRFS_I(inode)->root, src_path); | |
3764 | if (ret < 0) | |
3765 | return ret; | |
3766 | if (ret) | |
3767 | goto fill_holes; | |
3768 | if (src_path->slots[0]) | |
3769 | src_path->slots[0]--; | |
3770 | src = src_path->nodes[0]; | |
3771 | btrfs_item_key_to_cpu(src, &key, src_path->slots[0]); | |
3772 | if (key.objectid != btrfs_ino(inode) || | |
3773 | key.type != BTRFS_EXTENT_DATA_KEY) | |
3774 | goto fill_holes; | |
3775 | extent = btrfs_item_ptr(src, src_path->slots[0], | |
3776 | struct btrfs_file_extent_item); | |
3777 | if (btrfs_file_extent_type(src, extent) == | |
3778 | BTRFS_FILE_EXTENT_INLINE) { | |
514ac8ad CM |
3779 | len = btrfs_file_extent_inline_len(src, |
3780 | src_path->slots[0], | |
3781 | extent); | |
16e7549f JB |
3782 | *last_extent = ALIGN(key.offset + len, |
3783 | log->sectorsize); | |
3784 | } else { | |
3785 | len = btrfs_file_extent_num_bytes(src, extent); | |
3786 | *last_extent = key.offset + len; | |
3787 | } | |
3788 | } | |
3789 | fill_holes: | |
3790 | /* So we did prev_leaf, now we need to move to the next leaf, but a few | |
3791 | * things could have happened | |
3792 | * | |
3793 | * 1) A merge could have happened, so we could currently be on a leaf | |
3794 | * that holds what we were copying in the first place. | |
3795 | * 2) A split could have happened, and now not all of the items we want | |
3796 | * are on the same leaf. | |
3797 | * | |
3798 | * So we need to adjust how we search for holes, we need to drop the | |
3799 | * path and re-search for the first extent key we found, and then walk | |
3800 | * forward until we hit the last one we copied. | |
3801 | */ | |
3802 | if (need_find_last_extent) { | |
3803 | /* btrfs_prev_leaf could return 1 without releasing the path */ | |
3804 | btrfs_release_path(src_path); | |
3805 | ret = btrfs_search_slot(NULL, BTRFS_I(inode)->root, &first_key, | |
3806 | src_path, 0, 0); | |
3807 | if (ret < 0) | |
3808 | return ret; | |
3809 | ASSERT(ret == 0); | |
3810 | src = src_path->nodes[0]; | |
3811 | i = src_path->slots[0]; | |
3812 | } else { | |
3813 | i = start_slot; | |
3814 | } | |
3815 | ||
3816 | /* | |
3817 | * Ok so here we need to go through and fill in any holes we may have | |
3818 | * to make sure that holes are punched for those areas in case they had | |
3819 | * extents previously. | |
3820 | */ | |
3821 | while (!done) { | |
3822 | u64 offset, len; | |
3823 | u64 extent_end; | |
3824 | ||
3825 | if (i >= btrfs_header_nritems(src_path->nodes[0])) { | |
3826 | ret = btrfs_next_leaf(BTRFS_I(inode)->root, src_path); | |
3827 | if (ret < 0) | |
3828 | return ret; | |
3829 | ASSERT(ret == 0); | |
3830 | src = src_path->nodes[0]; | |
3831 | i = 0; | |
3832 | } | |
3833 | ||
3834 | btrfs_item_key_to_cpu(src, &key, i); | |
3835 | if (!btrfs_comp_cpu_keys(&key, &last_key)) | |
3836 | done = true; | |
3837 | if (key.objectid != btrfs_ino(inode) || | |
3838 | key.type != BTRFS_EXTENT_DATA_KEY) { | |
3839 | i++; | |
3840 | continue; | |
3841 | } | |
3842 | extent = btrfs_item_ptr(src, i, struct btrfs_file_extent_item); | |
3843 | if (btrfs_file_extent_type(src, extent) == | |
3844 | BTRFS_FILE_EXTENT_INLINE) { | |
514ac8ad | 3845 | len = btrfs_file_extent_inline_len(src, i, extent); |
16e7549f JB |
3846 | extent_end = ALIGN(key.offset + len, log->sectorsize); |
3847 | } else { | |
3848 | len = btrfs_file_extent_num_bytes(src, extent); | |
3849 | extent_end = key.offset + len; | |
3850 | } | |
3851 | i++; | |
3852 | ||
3853 | if (*last_extent == key.offset) { | |
3854 | *last_extent = extent_end; | |
3855 | continue; | |
3856 | } | |
3857 | offset = *last_extent; | |
3858 | len = key.offset - *last_extent; | |
3859 | ret = btrfs_insert_file_extent(trans, log, btrfs_ino(inode), | |
3860 | offset, 0, 0, len, 0, len, 0, | |
3861 | 0, 0); | |
3862 | if (ret) | |
3863 | break; | |
74121f7c | 3864 | *last_extent = extent_end; |
16e7549f JB |
3865 | } |
3866 | /* | |
3867 | * Need to let the callers know we dropped the path so they should | |
3868 | * re-search. | |
3869 | */ | |
3870 | if (!ret && need_find_last_extent) | |
3871 | ret = 1; | |
4a500fd1 | 3872 | return ret; |
31ff1cd2 CM |
3873 | } |
3874 | ||
5dc562c5 JB |
3875 | static int extent_cmp(void *priv, struct list_head *a, struct list_head *b) |
3876 | { | |
3877 | struct extent_map *em1, *em2; | |
3878 | ||
3879 | em1 = list_entry(a, struct extent_map, list); | |
3880 | em2 = list_entry(b, struct extent_map, list); | |
3881 | ||
3882 | if (em1->start < em2->start) | |
3883 | return -1; | |
3884 | else if (em1->start > em2->start) | |
3885 | return 1; | |
3886 | return 0; | |
3887 | } | |
3888 | ||
8407f553 FM |
3889 | static int wait_ordered_extents(struct btrfs_trans_handle *trans, |
3890 | struct inode *inode, | |
3891 | struct btrfs_root *root, | |
3892 | const struct extent_map *em, | |
3893 | const struct list_head *logged_list, | |
3894 | bool *ordered_io_error) | |
5dc562c5 | 3895 | { |
2ab28f32 | 3896 | struct btrfs_ordered_extent *ordered; |
8407f553 | 3897 | struct btrfs_root *log = root->log_root; |
2ab28f32 JB |
3898 | u64 mod_start = em->mod_start; |
3899 | u64 mod_len = em->mod_len; | |
8407f553 | 3900 | const bool skip_csum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM; |
2ab28f32 JB |
3901 | u64 csum_offset; |
3902 | u64 csum_len; | |
8407f553 FM |
3903 | LIST_HEAD(ordered_sums); |
3904 | int ret = 0; | |
0aa4a17d | 3905 | |
8407f553 | 3906 | *ordered_io_error = false; |
0aa4a17d | 3907 | |
8407f553 FM |
3908 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) || |
3909 | em->block_start == EXTENT_MAP_HOLE) | |
70c8a91c | 3910 | return 0; |
5dc562c5 | 3911 | |
2ab28f32 | 3912 | /* |
8407f553 FM |
3913 | * Wait far any ordered extent that covers our extent map. If it |
3914 | * finishes without an error, first check and see if our csums are on | |
3915 | * our outstanding ordered extents. | |
2ab28f32 | 3916 | */ |
827463c4 | 3917 | list_for_each_entry(ordered, logged_list, log_list) { |
2ab28f32 JB |
3918 | struct btrfs_ordered_sum *sum; |
3919 | ||
3920 | if (!mod_len) | |
3921 | break; | |
3922 | ||
2ab28f32 JB |
3923 | if (ordered->file_offset + ordered->len <= mod_start || |
3924 | mod_start + mod_len <= ordered->file_offset) | |
3925 | continue; | |
3926 | ||
8407f553 FM |
3927 | if (!test_bit(BTRFS_ORDERED_IO_DONE, &ordered->flags) && |
3928 | !test_bit(BTRFS_ORDERED_IOERR, &ordered->flags) && | |
3929 | !test_bit(BTRFS_ORDERED_DIRECT, &ordered->flags)) { | |
3930 | const u64 start = ordered->file_offset; | |
3931 | const u64 end = ordered->file_offset + ordered->len - 1; | |
3932 | ||
3933 | WARN_ON(ordered->inode != inode); | |
3934 | filemap_fdatawrite_range(inode->i_mapping, start, end); | |
3935 | } | |
3936 | ||
3937 | wait_event(ordered->wait, | |
3938 | (test_bit(BTRFS_ORDERED_IO_DONE, &ordered->flags) || | |
3939 | test_bit(BTRFS_ORDERED_IOERR, &ordered->flags))); | |
3940 | ||
3941 | if (test_bit(BTRFS_ORDERED_IOERR, &ordered->flags)) { | |
b38ef71c FM |
3942 | /* |
3943 | * Clear the AS_EIO/AS_ENOSPC flags from the inode's | |
3944 | * i_mapping flags, so that the next fsync won't get | |
3945 | * an outdated io error too. | |
3946 | */ | |
3947 | btrfs_inode_check_errors(inode); | |
8407f553 FM |
3948 | *ordered_io_error = true; |
3949 | break; | |
3950 | } | |
2ab28f32 JB |
3951 | /* |
3952 | * We are going to copy all the csums on this ordered extent, so | |
3953 | * go ahead and adjust mod_start and mod_len in case this | |
3954 | * ordered extent has already been logged. | |
3955 | */ | |
3956 | if (ordered->file_offset > mod_start) { | |
3957 | if (ordered->file_offset + ordered->len >= | |
3958 | mod_start + mod_len) | |
3959 | mod_len = ordered->file_offset - mod_start; | |
3960 | /* | |
3961 | * If we have this case | |
3962 | * | |
3963 | * |--------- logged extent ---------| | |
3964 | * |----- ordered extent ----| | |
3965 | * | |
3966 | * Just don't mess with mod_start and mod_len, we'll | |
3967 | * just end up logging more csums than we need and it | |
3968 | * will be ok. | |
3969 | */ | |
3970 | } else { | |
3971 | if (ordered->file_offset + ordered->len < | |
3972 | mod_start + mod_len) { | |
3973 | mod_len = (mod_start + mod_len) - | |
3974 | (ordered->file_offset + ordered->len); | |
3975 | mod_start = ordered->file_offset + | |
3976 | ordered->len; | |
3977 | } else { | |
3978 | mod_len = 0; | |
3979 | } | |
3980 | } | |
3981 | ||
8407f553 FM |
3982 | if (skip_csum) |
3983 | continue; | |
3984 | ||
2ab28f32 JB |
3985 | /* |
3986 | * To keep us from looping for the above case of an ordered | |
3987 | * extent that falls inside of the logged extent. | |
3988 | */ | |
3989 | if (test_and_set_bit(BTRFS_ORDERED_LOGGED_CSUM, | |
3990 | &ordered->flags)) | |
3991 | continue; | |
2ab28f32 | 3992 | |
2ab28f32 JB |
3993 | list_for_each_entry(sum, &ordered->list, list) { |
3994 | ret = btrfs_csum_file_blocks(trans, log, sum); | |
827463c4 | 3995 | if (ret) |
8407f553 | 3996 | break; |
2ab28f32 | 3997 | } |
2ab28f32 | 3998 | } |
2ab28f32 | 3999 | |
8407f553 | 4000 | if (*ordered_io_error || !mod_len || ret || skip_csum) |
2ab28f32 JB |
4001 | return ret; |
4002 | ||
488111aa FDBM |
4003 | if (em->compress_type) { |
4004 | csum_offset = 0; | |
8407f553 | 4005 | csum_len = max(em->block_len, em->orig_block_len); |
488111aa FDBM |
4006 | } else { |
4007 | csum_offset = mod_start - em->start; | |
4008 | csum_len = mod_len; | |
4009 | } | |
2ab28f32 | 4010 | |
70c8a91c JB |
4011 | /* block start is already adjusted for the file extent offset. */ |
4012 | ret = btrfs_lookup_csums_range(log->fs_info->csum_root, | |
4013 | em->block_start + csum_offset, | |
4014 | em->block_start + csum_offset + | |
4015 | csum_len - 1, &ordered_sums, 0); | |
4016 | if (ret) | |
4017 | return ret; | |
5dc562c5 | 4018 | |
70c8a91c JB |
4019 | while (!list_empty(&ordered_sums)) { |
4020 | struct btrfs_ordered_sum *sums = list_entry(ordered_sums.next, | |
4021 | struct btrfs_ordered_sum, | |
4022 | list); | |
4023 | if (!ret) | |
4024 | ret = btrfs_csum_file_blocks(trans, log, sums); | |
4025 | list_del(&sums->list); | |
4026 | kfree(sums); | |
5dc562c5 JB |
4027 | } |
4028 | ||
70c8a91c | 4029 | return ret; |
5dc562c5 JB |
4030 | } |
4031 | ||
8407f553 FM |
4032 | static int log_one_extent(struct btrfs_trans_handle *trans, |
4033 | struct inode *inode, struct btrfs_root *root, | |
4034 | const struct extent_map *em, | |
4035 | struct btrfs_path *path, | |
4036 | const struct list_head *logged_list, | |
4037 | struct btrfs_log_ctx *ctx) | |
4038 | { | |
4039 | struct btrfs_root *log = root->log_root; | |
4040 | struct btrfs_file_extent_item *fi; | |
4041 | struct extent_buffer *leaf; | |
4042 | struct btrfs_map_token token; | |
4043 | struct btrfs_key key; | |
4044 | u64 extent_offset = em->start - em->orig_start; | |
4045 | u64 block_len; | |
4046 | int ret; | |
4047 | int extent_inserted = 0; | |
4048 | bool ordered_io_err = false; | |
4049 | ||
4050 | ret = wait_ordered_extents(trans, inode, root, em, logged_list, | |
4051 | &ordered_io_err); | |
4052 | if (ret) | |
4053 | return ret; | |
4054 | ||
4055 | if (ordered_io_err) { | |
4056 | ctx->io_err = -EIO; | |
4057 | return 0; | |
4058 | } | |
4059 | ||
4060 | btrfs_init_map_token(&token); | |
4061 | ||
4062 | ret = __btrfs_drop_extents(trans, log, inode, path, em->start, | |
4063 | em->start + em->len, NULL, 0, 1, | |
4064 | sizeof(*fi), &extent_inserted); | |
4065 | if (ret) | |
4066 | return ret; | |
4067 | ||
4068 | if (!extent_inserted) { | |
4069 | key.objectid = btrfs_ino(inode); | |
4070 | key.type = BTRFS_EXTENT_DATA_KEY; | |
4071 | key.offset = em->start; | |
4072 | ||
4073 | ret = btrfs_insert_empty_item(trans, log, path, &key, | |
4074 | sizeof(*fi)); | |
4075 | if (ret) | |
4076 | return ret; | |
4077 | } | |
4078 | leaf = path->nodes[0]; | |
4079 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
4080 | struct btrfs_file_extent_item); | |
4081 | ||
50d9aa99 | 4082 | btrfs_set_token_file_extent_generation(leaf, fi, trans->transid, |
8407f553 FM |
4083 | &token); |
4084 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) | |
4085 | btrfs_set_token_file_extent_type(leaf, fi, | |
4086 | BTRFS_FILE_EXTENT_PREALLOC, | |
4087 | &token); | |
4088 | else | |
4089 | btrfs_set_token_file_extent_type(leaf, fi, | |
4090 | BTRFS_FILE_EXTENT_REG, | |
4091 | &token); | |
4092 | ||
4093 | block_len = max(em->block_len, em->orig_block_len); | |
4094 | if (em->compress_type != BTRFS_COMPRESS_NONE) { | |
4095 | btrfs_set_token_file_extent_disk_bytenr(leaf, fi, | |
4096 | em->block_start, | |
4097 | &token); | |
4098 | btrfs_set_token_file_extent_disk_num_bytes(leaf, fi, block_len, | |
4099 | &token); | |
4100 | } else if (em->block_start < EXTENT_MAP_LAST_BYTE) { | |
4101 | btrfs_set_token_file_extent_disk_bytenr(leaf, fi, | |
4102 | em->block_start - | |
4103 | extent_offset, &token); | |
4104 | btrfs_set_token_file_extent_disk_num_bytes(leaf, fi, block_len, | |
4105 | &token); | |
4106 | } else { | |
4107 | btrfs_set_token_file_extent_disk_bytenr(leaf, fi, 0, &token); | |
4108 | btrfs_set_token_file_extent_disk_num_bytes(leaf, fi, 0, | |
4109 | &token); | |
4110 | } | |
4111 | ||
4112 | btrfs_set_token_file_extent_offset(leaf, fi, extent_offset, &token); | |
4113 | btrfs_set_token_file_extent_num_bytes(leaf, fi, em->len, &token); | |
4114 | btrfs_set_token_file_extent_ram_bytes(leaf, fi, em->ram_bytes, &token); | |
4115 | btrfs_set_token_file_extent_compression(leaf, fi, em->compress_type, | |
4116 | &token); | |
4117 | btrfs_set_token_file_extent_encryption(leaf, fi, 0, &token); | |
4118 | btrfs_set_token_file_extent_other_encoding(leaf, fi, 0, &token); | |
4119 | btrfs_mark_buffer_dirty(leaf); | |
4120 | ||
4121 | btrfs_release_path(path); | |
4122 | ||
4123 | return ret; | |
4124 | } | |
4125 | ||
5dc562c5 JB |
4126 | static int btrfs_log_changed_extents(struct btrfs_trans_handle *trans, |
4127 | struct btrfs_root *root, | |
4128 | struct inode *inode, | |
827463c4 | 4129 | struct btrfs_path *path, |
8407f553 | 4130 | struct list_head *logged_list, |
de0ee0ed FM |
4131 | struct btrfs_log_ctx *ctx, |
4132 | const u64 start, | |
4133 | const u64 end) | |
5dc562c5 | 4134 | { |
5dc562c5 JB |
4135 | struct extent_map *em, *n; |
4136 | struct list_head extents; | |
4137 | struct extent_map_tree *tree = &BTRFS_I(inode)->extent_tree; | |
4138 | u64 test_gen; | |
4139 | int ret = 0; | |
2ab28f32 | 4140 | int num = 0; |
5dc562c5 JB |
4141 | |
4142 | INIT_LIST_HEAD(&extents); | |
4143 | ||
5dc562c5 JB |
4144 | write_lock(&tree->lock); |
4145 | test_gen = root->fs_info->last_trans_committed; | |
4146 | ||
4147 | list_for_each_entry_safe(em, n, &tree->modified_extents, list) { | |
4148 | list_del_init(&em->list); | |
2ab28f32 JB |
4149 | |
4150 | /* | |
4151 | * Just an arbitrary number, this can be really CPU intensive | |
4152 | * once we start getting a lot of extents, and really once we | |
4153 | * have a bunch of extents we just want to commit since it will | |
4154 | * be faster. | |
4155 | */ | |
4156 | if (++num > 32768) { | |
4157 | list_del_init(&tree->modified_extents); | |
4158 | ret = -EFBIG; | |
4159 | goto process; | |
4160 | } | |
4161 | ||
5dc562c5 JB |
4162 | if (em->generation <= test_gen) |
4163 | continue; | |
ff44c6e3 JB |
4164 | /* Need a ref to keep it from getting evicted from cache */ |
4165 | atomic_inc(&em->refs); | |
4166 | set_bit(EXTENT_FLAG_LOGGING, &em->flags); | |
5dc562c5 | 4167 | list_add_tail(&em->list, &extents); |
2ab28f32 | 4168 | num++; |
5dc562c5 JB |
4169 | } |
4170 | ||
4171 | list_sort(NULL, &extents, extent_cmp); | |
de0ee0ed FM |
4172 | /* |
4173 | * Collect any new ordered extents within the range. This is to | |
4174 | * prevent logging file extent items without waiting for the disk | |
4175 | * location they point to being written. We do this only to deal | |
4176 | * with races against concurrent lockless direct IO writes. | |
4177 | */ | |
4178 | btrfs_get_logged_extents(inode, logged_list, start, end); | |
2ab28f32 | 4179 | process: |
5dc562c5 JB |
4180 | while (!list_empty(&extents)) { |
4181 | em = list_entry(extents.next, struct extent_map, list); | |
4182 | ||
4183 | list_del_init(&em->list); | |
4184 | ||
4185 | /* | |
4186 | * If we had an error we just need to delete everybody from our | |
4187 | * private list. | |
4188 | */ | |
ff44c6e3 | 4189 | if (ret) { |
201a9038 | 4190 | clear_em_logging(tree, em); |
ff44c6e3 | 4191 | free_extent_map(em); |
5dc562c5 | 4192 | continue; |
ff44c6e3 JB |
4193 | } |
4194 | ||
4195 | write_unlock(&tree->lock); | |
5dc562c5 | 4196 | |
8407f553 FM |
4197 | ret = log_one_extent(trans, inode, root, em, path, logged_list, |
4198 | ctx); | |
ff44c6e3 | 4199 | write_lock(&tree->lock); |
201a9038 JB |
4200 | clear_em_logging(tree, em); |
4201 | free_extent_map(em); | |
5dc562c5 | 4202 | } |
ff44c6e3 JB |
4203 | WARN_ON(!list_empty(&extents)); |
4204 | write_unlock(&tree->lock); | |
5dc562c5 | 4205 | |
5dc562c5 | 4206 | btrfs_release_path(path); |
5dc562c5 JB |
4207 | return ret; |
4208 | } | |
4209 | ||
1a4bcf47 FM |
4210 | static int logged_inode_size(struct btrfs_root *log, struct inode *inode, |
4211 | struct btrfs_path *path, u64 *size_ret) | |
4212 | { | |
4213 | struct btrfs_key key; | |
4214 | int ret; | |
4215 | ||
4216 | key.objectid = btrfs_ino(inode); | |
4217 | key.type = BTRFS_INODE_ITEM_KEY; | |
4218 | key.offset = 0; | |
4219 | ||
4220 | ret = btrfs_search_slot(NULL, log, &key, path, 0, 0); | |
4221 | if (ret < 0) { | |
4222 | return ret; | |
4223 | } else if (ret > 0) { | |
2f2ff0ee | 4224 | *size_ret = 0; |
1a4bcf47 FM |
4225 | } else { |
4226 | struct btrfs_inode_item *item; | |
4227 | ||
4228 | item = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
4229 | struct btrfs_inode_item); | |
4230 | *size_ret = btrfs_inode_size(path->nodes[0], item); | |
4231 | } | |
4232 | ||
4233 | btrfs_release_path(path); | |
4234 | return 0; | |
4235 | } | |
4236 | ||
36283bf7 FM |
4237 | /* |
4238 | * At the moment we always log all xattrs. This is to figure out at log replay | |
4239 | * time which xattrs must have their deletion replayed. If a xattr is missing | |
4240 | * in the log tree and exists in the fs/subvol tree, we delete it. This is | |
4241 | * because if a xattr is deleted, the inode is fsynced and a power failure | |
4242 | * happens, causing the log to be replayed the next time the fs is mounted, | |
4243 | * we want the xattr to not exist anymore (same behaviour as other filesystems | |
4244 | * with a journal, ext3/4, xfs, f2fs, etc). | |
4245 | */ | |
4246 | static int btrfs_log_all_xattrs(struct btrfs_trans_handle *trans, | |
4247 | struct btrfs_root *root, | |
4248 | struct inode *inode, | |
4249 | struct btrfs_path *path, | |
4250 | struct btrfs_path *dst_path) | |
4251 | { | |
4252 | int ret; | |
4253 | struct btrfs_key key; | |
4254 | const u64 ino = btrfs_ino(inode); | |
4255 | int ins_nr = 0; | |
4256 | int start_slot = 0; | |
4257 | ||
4258 | key.objectid = ino; | |
4259 | key.type = BTRFS_XATTR_ITEM_KEY; | |
4260 | key.offset = 0; | |
4261 | ||
4262 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
4263 | if (ret < 0) | |
4264 | return ret; | |
4265 | ||
4266 | while (true) { | |
4267 | int slot = path->slots[0]; | |
4268 | struct extent_buffer *leaf = path->nodes[0]; | |
4269 | int nritems = btrfs_header_nritems(leaf); | |
4270 | ||
4271 | if (slot >= nritems) { | |
4272 | if (ins_nr > 0) { | |
4273 | u64 last_extent = 0; | |
4274 | ||
4275 | ret = copy_items(trans, inode, dst_path, path, | |
4276 | &last_extent, start_slot, | |
4277 | ins_nr, 1, 0); | |
4278 | /* can't be 1, extent items aren't processed */ | |
4279 | ASSERT(ret <= 0); | |
4280 | if (ret < 0) | |
4281 | return ret; | |
4282 | ins_nr = 0; | |
4283 | } | |
4284 | ret = btrfs_next_leaf(root, path); | |
4285 | if (ret < 0) | |
4286 | return ret; | |
4287 | else if (ret > 0) | |
4288 | break; | |
4289 | continue; | |
4290 | } | |
4291 | ||
4292 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
4293 | if (key.objectid != ino || key.type != BTRFS_XATTR_ITEM_KEY) | |
4294 | break; | |
4295 | ||
4296 | if (ins_nr == 0) | |
4297 | start_slot = slot; | |
4298 | ins_nr++; | |
4299 | path->slots[0]++; | |
4300 | cond_resched(); | |
4301 | } | |
4302 | if (ins_nr > 0) { | |
4303 | u64 last_extent = 0; | |
4304 | ||
4305 | ret = copy_items(trans, inode, dst_path, path, | |
4306 | &last_extent, start_slot, | |
4307 | ins_nr, 1, 0); | |
4308 | /* can't be 1, extent items aren't processed */ | |
4309 | ASSERT(ret <= 0); | |
4310 | if (ret < 0) | |
4311 | return ret; | |
4312 | } | |
4313 | ||
4314 | return 0; | |
4315 | } | |
4316 | ||
a89ca6f2 FM |
4317 | /* |
4318 | * If the no holes feature is enabled we need to make sure any hole between the | |
4319 | * last extent and the i_size of our inode is explicitly marked in the log. This | |
4320 | * is to make sure that doing something like: | |
4321 | * | |
4322 | * 1) create file with 128Kb of data | |
4323 | * 2) truncate file to 64Kb | |
4324 | * 3) truncate file to 256Kb | |
4325 | * 4) fsync file | |
4326 | * 5) <crash/power failure> | |
4327 | * 6) mount fs and trigger log replay | |
4328 | * | |
4329 | * Will give us a file with a size of 256Kb, the first 64Kb of data match what | |
4330 | * the file had in its first 64Kb of data at step 1 and the last 192Kb of the | |
4331 | * file correspond to a hole. The presence of explicit holes in a log tree is | |
4332 | * what guarantees that log replay will remove/adjust file extent items in the | |
4333 | * fs/subvol tree. | |
4334 | * | |
4335 | * Here we do not need to care about holes between extents, that is already done | |
4336 | * by copy_items(). We also only need to do this in the full sync path, where we | |
4337 | * lookup for extents from the fs/subvol tree only. In the fast path case, we | |
4338 | * lookup the list of modified extent maps and if any represents a hole, we | |
4339 | * insert a corresponding extent representing a hole in the log tree. | |
4340 | */ | |
4341 | static int btrfs_log_trailing_hole(struct btrfs_trans_handle *trans, | |
4342 | struct btrfs_root *root, | |
4343 | struct inode *inode, | |
4344 | struct btrfs_path *path) | |
4345 | { | |
4346 | int ret; | |
4347 | struct btrfs_key key; | |
4348 | u64 hole_start; | |
4349 | u64 hole_size; | |
4350 | struct extent_buffer *leaf; | |
4351 | struct btrfs_root *log = root->log_root; | |
4352 | const u64 ino = btrfs_ino(inode); | |
4353 | const u64 i_size = i_size_read(inode); | |
4354 | ||
4355 | if (!btrfs_fs_incompat(root->fs_info, NO_HOLES)) | |
4356 | return 0; | |
4357 | ||
4358 | key.objectid = ino; | |
4359 | key.type = BTRFS_EXTENT_DATA_KEY; | |
4360 | key.offset = (u64)-1; | |
4361 | ||
4362 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
4363 | ASSERT(ret != 0); | |
4364 | if (ret < 0) | |
4365 | return ret; | |
4366 | ||
4367 | ASSERT(path->slots[0] > 0); | |
4368 | path->slots[0]--; | |
4369 | leaf = path->nodes[0]; | |
4370 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
4371 | ||
4372 | if (key.objectid != ino || key.type != BTRFS_EXTENT_DATA_KEY) { | |
4373 | /* inode does not have any extents */ | |
4374 | hole_start = 0; | |
4375 | hole_size = i_size; | |
4376 | } else { | |
4377 | struct btrfs_file_extent_item *extent; | |
4378 | u64 len; | |
4379 | ||
4380 | /* | |
4381 | * If there's an extent beyond i_size, an explicit hole was | |
4382 | * already inserted by copy_items(). | |
4383 | */ | |
4384 | if (key.offset >= i_size) | |
4385 | return 0; | |
4386 | ||
4387 | extent = btrfs_item_ptr(leaf, path->slots[0], | |
4388 | struct btrfs_file_extent_item); | |
4389 | ||
4390 | if (btrfs_file_extent_type(leaf, extent) == | |
4391 | BTRFS_FILE_EXTENT_INLINE) { | |
4392 | len = btrfs_file_extent_inline_len(leaf, | |
4393 | path->slots[0], | |
4394 | extent); | |
4395 | ASSERT(len == i_size); | |
4396 | return 0; | |
4397 | } | |
4398 | ||
4399 | len = btrfs_file_extent_num_bytes(leaf, extent); | |
4400 | /* Last extent goes beyond i_size, no need to log a hole. */ | |
4401 | if (key.offset + len > i_size) | |
4402 | return 0; | |
4403 | hole_start = key.offset + len; | |
4404 | hole_size = i_size - hole_start; | |
4405 | } | |
4406 | btrfs_release_path(path); | |
4407 | ||
4408 | /* Last extent ends at i_size. */ | |
4409 | if (hole_size == 0) | |
4410 | return 0; | |
4411 | ||
4412 | hole_size = ALIGN(hole_size, root->sectorsize); | |
4413 | ret = btrfs_insert_file_extent(trans, log, ino, hole_start, 0, 0, | |
4414 | hole_size, 0, hole_size, 0, 0, 0); | |
4415 | return ret; | |
4416 | } | |
4417 | ||
e02119d5 CM |
4418 | /* log a single inode in the tree log. |
4419 | * At least one parent directory for this inode must exist in the tree | |
4420 | * or be logged already. | |
4421 | * | |
4422 | * Any items from this inode changed by the current transaction are copied | |
4423 | * to the log tree. An extra reference is taken on any extents in this | |
4424 | * file, allowing us to avoid a whole pile of corner cases around logging | |
4425 | * blocks that have been removed from the tree. | |
4426 | * | |
4427 | * See LOG_INODE_ALL and related defines for a description of what inode_only | |
4428 | * does. | |
4429 | * | |
4430 | * This handles both files and directories. | |
4431 | */ | |
12fcfd22 | 4432 | static int btrfs_log_inode(struct btrfs_trans_handle *trans, |
49dae1bc FM |
4433 | struct btrfs_root *root, struct inode *inode, |
4434 | int inode_only, | |
4435 | const loff_t start, | |
8407f553 FM |
4436 | const loff_t end, |
4437 | struct btrfs_log_ctx *ctx) | |
e02119d5 CM |
4438 | { |
4439 | struct btrfs_path *path; | |
4440 | struct btrfs_path *dst_path; | |
4441 | struct btrfs_key min_key; | |
4442 | struct btrfs_key max_key; | |
4443 | struct btrfs_root *log = root->log_root; | |
31ff1cd2 | 4444 | struct extent_buffer *src = NULL; |
827463c4 | 4445 | LIST_HEAD(logged_list); |
16e7549f | 4446 | u64 last_extent = 0; |
4a500fd1 | 4447 | int err = 0; |
e02119d5 | 4448 | int ret; |
3a5f1d45 | 4449 | int nritems; |
31ff1cd2 CM |
4450 | int ins_start_slot = 0; |
4451 | int ins_nr; | |
5dc562c5 | 4452 | bool fast_search = false; |
33345d01 | 4453 | u64 ino = btrfs_ino(inode); |
49dae1bc | 4454 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; |
1a4bcf47 | 4455 | u64 logged_isize = 0; |
e4545de5 | 4456 | bool need_log_inode_item = true; |
e02119d5 | 4457 | |
e02119d5 | 4458 | path = btrfs_alloc_path(); |
5df67083 TI |
4459 | if (!path) |
4460 | return -ENOMEM; | |
e02119d5 | 4461 | dst_path = btrfs_alloc_path(); |
5df67083 TI |
4462 | if (!dst_path) { |
4463 | btrfs_free_path(path); | |
4464 | return -ENOMEM; | |
4465 | } | |
e02119d5 | 4466 | |
33345d01 | 4467 | min_key.objectid = ino; |
e02119d5 CM |
4468 | min_key.type = BTRFS_INODE_ITEM_KEY; |
4469 | min_key.offset = 0; | |
4470 | ||
33345d01 | 4471 | max_key.objectid = ino; |
12fcfd22 | 4472 | |
12fcfd22 | 4473 | |
5dc562c5 | 4474 | /* today the code can only do partial logging of directories */ |
5269b67e MX |
4475 | if (S_ISDIR(inode->i_mode) || |
4476 | (!test_bit(BTRFS_INODE_NEEDS_FULL_SYNC, | |
4477 | &BTRFS_I(inode)->runtime_flags) && | |
4478 | inode_only == LOG_INODE_EXISTS)) | |
e02119d5 CM |
4479 | max_key.type = BTRFS_XATTR_ITEM_KEY; |
4480 | else | |
4481 | max_key.type = (u8)-1; | |
4482 | max_key.offset = (u64)-1; | |
4483 | ||
2c2c452b FM |
4484 | /* |
4485 | * Only run delayed items if we are a dir or a new file. | |
4486 | * Otherwise commit the delayed inode only, which is needed in | |
4487 | * order for the log replay code to mark inodes for link count | |
4488 | * fixup (create temporary BTRFS_TREE_LOG_FIXUP_OBJECTID items). | |
4489 | */ | |
94edf4ae | 4490 | if (S_ISDIR(inode->i_mode) || |
2c2c452b | 4491 | BTRFS_I(inode)->generation > root->fs_info->last_trans_committed) |
94edf4ae | 4492 | ret = btrfs_commit_inode_delayed_items(trans, inode); |
2c2c452b FM |
4493 | else |
4494 | ret = btrfs_commit_inode_delayed_inode(inode); | |
4495 | ||
4496 | if (ret) { | |
4497 | btrfs_free_path(path); | |
4498 | btrfs_free_path(dst_path); | |
4499 | return ret; | |
16cdcec7 MX |
4500 | } |
4501 | ||
e02119d5 CM |
4502 | mutex_lock(&BTRFS_I(inode)->log_mutex); |
4503 | ||
5e33a2bd FM |
4504 | /* |
4505 | * Collect ordered extents only if we are logging data. This is to | |
4506 | * ensure a subsequent request to log this inode in LOG_INODE_ALL mode | |
4507 | * will process the ordered extents if they still exists at the time, | |
4508 | * because when we collect them we test and set for the flag | |
4509 | * BTRFS_ORDERED_LOGGED to prevent multiple log requests to process the | |
4510 | * same ordered extents. The consequence for the LOG_INODE_ALL log mode | |
4511 | * not processing the ordered extents is that we end up logging the | |
4512 | * corresponding file extent items, based on the extent maps in the | |
4513 | * inode's extent_map_tree's modified_list, without logging the | |
4514 | * respective checksums (since the may still be only attached to the | |
4515 | * ordered extents and have not been inserted in the csum tree by | |
4516 | * btrfs_finish_ordered_io() yet). | |
4517 | */ | |
4518 | if (inode_only == LOG_INODE_ALL) | |
4519 | btrfs_get_logged_extents(inode, &logged_list, start, end); | |
2ab28f32 | 4520 | |
e02119d5 CM |
4521 | /* |
4522 | * a brute force approach to making sure we get the most uptodate | |
4523 | * copies of everything. | |
4524 | */ | |
4525 | if (S_ISDIR(inode->i_mode)) { | |
4526 | int max_key_type = BTRFS_DIR_LOG_INDEX_KEY; | |
4527 | ||
4f764e51 FM |
4528 | if (inode_only == LOG_INODE_EXISTS) |
4529 | max_key_type = BTRFS_XATTR_ITEM_KEY; | |
33345d01 | 4530 | ret = drop_objectid_items(trans, log, path, ino, max_key_type); |
e02119d5 | 4531 | } else { |
1a4bcf47 FM |
4532 | if (inode_only == LOG_INODE_EXISTS) { |
4533 | /* | |
4534 | * Make sure the new inode item we write to the log has | |
4535 | * the same isize as the current one (if it exists). | |
4536 | * This is necessary to prevent data loss after log | |
4537 | * replay, and also to prevent doing a wrong expanding | |
4538 | * truncate - for e.g. create file, write 4K into offset | |
4539 | * 0, fsync, write 4K into offset 4096, add hard link, | |
4540 | * fsync some other file (to sync log), power fail - if | |
4541 | * we use the inode's current i_size, after log replay | |
4542 | * we get a 8Kb file, with the last 4Kb extent as a hole | |
4543 | * (zeroes), as if an expanding truncate happened, | |
4544 | * instead of getting a file of 4Kb only. | |
4545 | */ | |
4546 | err = logged_inode_size(log, inode, path, | |
4547 | &logged_isize); | |
4548 | if (err) | |
4549 | goto out_unlock; | |
4550 | } | |
a742994a FM |
4551 | if (test_bit(BTRFS_INODE_NEEDS_FULL_SYNC, |
4552 | &BTRFS_I(inode)->runtime_flags)) { | |
4553 | if (inode_only == LOG_INODE_EXISTS) { | |
4f764e51 | 4554 | max_key.type = BTRFS_XATTR_ITEM_KEY; |
a742994a FM |
4555 | ret = drop_objectid_items(trans, log, path, ino, |
4556 | max_key.type); | |
4557 | } else { | |
4558 | clear_bit(BTRFS_INODE_NEEDS_FULL_SYNC, | |
4559 | &BTRFS_I(inode)->runtime_flags); | |
4560 | clear_bit(BTRFS_INODE_COPY_EVERYTHING, | |
4561 | &BTRFS_I(inode)->runtime_flags); | |
28ed1345 CM |
4562 | while(1) { |
4563 | ret = btrfs_truncate_inode_items(trans, | |
4564 | log, inode, 0, 0); | |
4565 | if (ret != -EAGAIN) | |
4566 | break; | |
4567 | } | |
a742994a | 4568 | } |
4f764e51 FM |
4569 | } else if (test_and_clear_bit(BTRFS_INODE_COPY_EVERYTHING, |
4570 | &BTRFS_I(inode)->runtime_flags) || | |
6cfab851 | 4571 | inode_only == LOG_INODE_EXISTS) { |
4f764e51 | 4572 | if (inode_only == LOG_INODE_ALL) |
183f37fa | 4573 | fast_search = true; |
4f764e51 | 4574 | max_key.type = BTRFS_XATTR_ITEM_KEY; |
5dc562c5 | 4575 | ret = drop_objectid_items(trans, log, path, ino, |
e9976151 | 4576 | max_key.type); |
a95249b3 JB |
4577 | } else { |
4578 | if (inode_only == LOG_INODE_ALL) | |
4579 | fast_search = true; | |
a95249b3 | 4580 | goto log_extents; |
5dc562c5 | 4581 | } |
a95249b3 | 4582 | |
e02119d5 | 4583 | } |
4a500fd1 YZ |
4584 | if (ret) { |
4585 | err = ret; | |
4586 | goto out_unlock; | |
4587 | } | |
e02119d5 | 4588 | |
d397712b | 4589 | while (1) { |
31ff1cd2 | 4590 | ins_nr = 0; |
6174d3cb | 4591 | ret = btrfs_search_forward(root, &min_key, |
de78b51a | 4592 | path, trans->transid); |
e02119d5 CM |
4593 | if (ret != 0) |
4594 | break; | |
3a5f1d45 | 4595 | again: |
31ff1cd2 | 4596 | /* note, ins_nr might be > 0 here, cleanup outside the loop */ |
33345d01 | 4597 | if (min_key.objectid != ino) |
e02119d5 CM |
4598 | break; |
4599 | if (min_key.type > max_key.type) | |
4600 | break; | |
31ff1cd2 | 4601 | |
e4545de5 FM |
4602 | if (min_key.type == BTRFS_INODE_ITEM_KEY) |
4603 | need_log_inode_item = false; | |
4604 | ||
36283bf7 FM |
4605 | /* Skip xattrs, we log them later with btrfs_log_all_xattrs() */ |
4606 | if (min_key.type == BTRFS_XATTR_ITEM_KEY) { | |
4607 | if (ins_nr == 0) | |
4608 | goto next_slot; | |
4609 | ret = copy_items(trans, inode, dst_path, path, | |
4610 | &last_extent, ins_start_slot, | |
4611 | ins_nr, inode_only, logged_isize); | |
4612 | if (ret < 0) { | |
4613 | err = ret; | |
4614 | goto out_unlock; | |
4615 | } | |
4616 | ins_nr = 0; | |
4617 | if (ret) { | |
4618 | btrfs_release_path(path); | |
4619 | continue; | |
4620 | } | |
4621 | goto next_slot; | |
4622 | } | |
4623 | ||
e02119d5 | 4624 | src = path->nodes[0]; |
31ff1cd2 CM |
4625 | if (ins_nr && ins_start_slot + ins_nr == path->slots[0]) { |
4626 | ins_nr++; | |
4627 | goto next_slot; | |
4628 | } else if (!ins_nr) { | |
4629 | ins_start_slot = path->slots[0]; | |
4630 | ins_nr = 1; | |
4631 | goto next_slot; | |
e02119d5 CM |
4632 | } |
4633 | ||
16e7549f | 4634 | ret = copy_items(trans, inode, dst_path, path, &last_extent, |
1a4bcf47 FM |
4635 | ins_start_slot, ins_nr, inode_only, |
4636 | logged_isize); | |
16e7549f | 4637 | if (ret < 0) { |
4a500fd1 YZ |
4638 | err = ret; |
4639 | goto out_unlock; | |
a71db86e RV |
4640 | } |
4641 | if (ret) { | |
16e7549f JB |
4642 | ins_nr = 0; |
4643 | btrfs_release_path(path); | |
4644 | continue; | |
4a500fd1 | 4645 | } |
31ff1cd2 CM |
4646 | ins_nr = 1; |
4647 | ins_start_slot = path->slots[0]; | |
4648 | next_slot: | |
e02119d5 | 4649 | |
3a5f1d45 CM |
4650 | nritems = btrfs_header_nritems(path->nodes[0]); |
4651 | path->slots[0]++; | |
4652 | if (path->slots[0] < nritems) { | |
4653 | btrfs_item_key_to_cpu(path->nodes[0], &min_key, | |
4654 | path->slots[0]); | |
4655 | goto again; | |
4656 | } | |
31ff1cd2 | 4657 | if (ins_nr) { |
16e7549f JB |
4658 | ret = copy_items(trans, inode, dst_path, path, |
4659 | &last_extent, ins_start_slot, | |
1a4bcf47 | 4660 | ins_nr, inode_only, logged_isize); |
16e7549f | 4661 | if (ret < 0) { |
4a500fd1 YZ |
4662 | err = ret; |
4663 | goto out_unlock; | |
4664 | } | |
16e7549f | 4665 | ret = 0; |
31ff1cd2 CM |
4666 | ins_nr = 0; |
4667 | } | |
b3b4aa74 | 4668 | btrfs_release_path(path); |
3a5f1d45 | 4669 | |
3d41d702 | 4670 | if (min_key.offset < (u64)-1) { |
e02119d5 | 4671 | min_key.offset++; |
3d41d702 | 4672 | } else if (min_key.type < max_key.type) { |
e02119d5 | 4673 | min_key.type++; |
3d41d702 FDBM |
4674 | min_key.offset = 0; |
4675 | } else { | |
e02119d5 | 4676 | break; |
3d41d702 | 4677 | } |
e02119d5 | 4678 | } |
31ff1cd2 | 4679 | if (ins_nr) { |
16e7549f | 4680 | ret = copy_items(trans, inode, dst_path, path, &last_extent, |
1a4bcf47 FM |
4681 | ins_start_slot, ins_nr, inode_only, |
4682 | logged_isize); | |
16e7549f | 4683 | if (ret < 0) { |
4a500fd1 YZ |
4684 | err = ret; |
4685 | goto out_unlock; | |
4686 | } | |
16e7549f | 4687 | ret = 0; |
31ff1cd2 CM |
4688 | ins_nr = 0; |
4689 | } | |
5dc562c5 | 4690 | |
36283bf7 FM |
4691 | btrfs_release_path(path); |
4692 | btrfs_release_path(dst_path); | |
4693 | err = btrfs_log_all_xattrs(trans, root, inode, path, dst_path); | |
4694 | if (err) | |
4695 | goto out_unlock; | |
a89ca6f2 FM |
4696 | if (max_key.type >= BTRFS_EXTENT_DATA_KEY && !fast_search) { |
4697 | btrfs_release_path(path); | |
4698 | btrfs_release_path(dst_path); | |
4699 | err = btrfs_log_trailing_hole(trans, root, inode, path); | |
4700 | if (err) | |
4701 | goto out_unlock; | |
4702 | } | |
a95249b3 | 4703 | log_extents: |
f3b15ccd JB |
4704 | btrfs_release_path(path); |
4705 | btrfs_release_path(dst_path); | |
e4545de5 FM |
4706 | if (need_log_inode_item) { |
4707 | err = log_inode_item(trans, log, dst_path, inode); | |
4708 | if (err) | |
4709 | goto out_unlock; | |
4710 | } | |
5dc562c5 | 4711 | if (fast_search) { |
b38ef71c FM |
4712 | /* |
4713 | * Some ordered extents started by fsync might have completed | |
4714 | * before we collected the ordered extents in logged_list, which | |
4715 | * means they're gone, not in our logged_list nor in the inode's | |
4716 | * ordered tree. We want the application/user space to know an | |
4717 | * error happened while attempting to persist file data so that | |
4718 | * it can take proper action. If such error happened, we leave | |
4719 | * without writing to the log tree and the fsync must report the | |
4720 | * file data write error and not commit the current transaction. | |
4721 | */ | |
4722 | err = btrfs_inode_check_errors(inode); | |
4723 | if (err) { | |
4724 | ctx->io_err = err; | |
4725 | goto out_unlock; | |
4726 | } | |
827463c4 | 4727 | ret = btrfs_log_changed_extents(trans, root, inode, dst_path, |
de0ee0ed | 4728 | &logged_list, ctx, start, end); |
5dc562c5 JB |
4729 | if (ret) { |
4730 | err = ret; | |
4731 | goto out_unlock; | |
4732 | } | |
d006a048 | 4733 | } else if (inode_only == LOG_INODE_ALL) { |
06d3d22b LB |
4734 | struct extent_map *em, *n; |
4735 | ||
49dae1bc FM |
4736 | write_lock(&em_tree->lock); |
4737 | /* | |
4738 | * We can't just remove every em if we're called for a ranged | |
4739 | * fsync - that is, one that doesn't cover the whole possible | |
4740 | * file range (0 to LLONG_MAX). This is because we can have | |
4741 | * em's that fall outside the range we're logging and therefore | |
4742 | * their ordered operations haven't completed yet | |
4743 | * (btrfs_finish_ordered_io() not invoked yet). This means we | |
4744 | * didn't get their respective file extent item in the fs/subvol | |
4745 | * tree yet, and need to let the next fast fsync (one which | |
4746 | * consults the list of modified extent maps) find the em so | |
4747 | * that it logs a matching file extent item and waits for the | |
4748 | * respective ordered operation to complete (if it's still | |
4749 | * running). | |
4750 | * | |
4751 | * Removing every em outside the range we're logging would make | |
4752 | * the next fast fsync not log their matching file extent items, | |
4753 | * therefore making us lose data after a log replay. | |
4754 | */ | |
4755 | list_for_each_entry_safe(em, n, &em_tree->modified_extents, | |
4756 | list) { | |
4757 | const u64 mod_end = em->mod_start + em->mod_len - 1; | |
4758 | ||
4759 | if (em->mod_start >= start && mod_end <= end) | |
4760 | list_del_init(&em->list); | |
4761 | } | |
4762 | write_unlock(&em_tree->lock); | |
5dc562c5 JB |
4763 | } |
4764 | ||
9623f9a3 | 4765 | if (inode_only == LOG_INODE_ALL && S_ISDIR(inode->i_mode)) { |
2f2ff0ee FM |
4766 | ret = log_directory_changes(trans, root, inode, path, dst_path, |
4767 | ctx); | |
4a500fd1 YZ |
4768 | if (ret) { |
4769 | err = ret; | |
4770 | goto out_unlock; | |
4771 | } | |
e02119d5 | 4772 | } |
49dae1bc | 4773 | |
2f2ff0ee | 4774 | spin_lock(&BTRFS_I(inode)->lock); |
125c4cf9 FM |
4775 | BTRFS_I(inode)->logged_trans = trans->transid; |
4776 | BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->last_sub_trans; | |
2f2ff0ee | 4777 | spin_unlock(&BTRFS_I(inode)->lock); |
4a500fd1 | 4778 | out_unlock: |
827463c4 MX |
4779 | if (unlikely(err)) |
4780 | btrfs_put_logged_extents(&logged_list); | |
4781 | else | |
4782 | btrfs_submit_logged_extents(&logged_list, log); | |
e02119d5 CM |
4783 | mutex_unlock(&BTRFS_I(inode)->log_mutex); |
4784 | ||
4785 | btrfs_free_path(path); | |
4786 | btrfs_free_path(dst_path); | |
4a500fd1 | 4787 | return err; |
e02119d5 CM |
4788 | } |
4789 | ||
2be63d5c FM |
4790 | /* |
4791 | * Check if we must fallback to a transaction commit when logging an inode. | |
4792 | * This must be called after logging the inode and is used only in the context | |
4793 | * when fsyncing an inode requires the need to log some other inode - in which | |
4794 | * case we can't lock the i_mutex of each other inode we need to log as that | |
4795 | * can lead to deadlocks with concurrent fsync against other inodes (as we can | |
4796 | * log inodes up or down in the hierarchy) or rename operations for example. So | |
4797 | * we take the log_mutex of the inode after we have logged it and then check for | |
4798 | * its last_unlink_trans value - this is safe because any task setting | |
4799 | * last_unlink_trans must take the log_mutex and it must do this before it does | |
4800 | * the actual unlink operation, so if we do this check before a concurrent task | |
4801 | * sets last_unlink_trans it means we've logged a consistent version/state of | |
4802 | * all the inode items, otherwise we are not sure and must do a transaction | |
4803 | * commit (the concurrent task migth have only updated last_unlink_trans before | |
4804 | * we logged the inode or it might have also done the unlink). | |
4805 | */ | |
4806 | static bool btrfs_must_commit_transaction(struct btrfs_trans_handle *trans, | |
4807 | struct inode *inode) | |
4808 | { | |
4809 | struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info; | |
4810 | bool ret = false; | |
4811 | ||
4812 | mutex_lock(&BTRFS_I(inode)->log_mutex); | |
4813 | if (BTRFS_I(inode)->last_unlink_trans > fs_info->last_trans_committed) { | |
4814 | /* | |
4815 | * Make sure any commits to the log are forced to be full | |
4816 | * commits. | |
4817 | */ | |
4818 | btrfs_set_log_full_commit(fs_info, trans); | |
4819 | ret = true; | |
4820 | } | |
4821 | mutex_unlock(&BTRFS_I(inode)->log_mutex); | |
4822 | ||
4823 | return ret; | |
4824 | } | |
4825 | ||
12fcfd22 CM |
4826 | /* |
4827 | * follow the dentry parent pointers up the chain and see if any | |
4828 | * of the directories in it require a full commit before they can | |
4829 | * be logged. Returns zero if nothing special needs to be done or 1 if | |
4830 | * a full commit is required. | |
4831 | */ | |
4832 | static noinline int check_parent_dirs_for_sync(struct btrfs_trans_handle *trans, | |
4833 | struct inode *inode, | |
4834 | struct dentry *parent, | |
4835 | struct super_block *sb, | |
4836 | u64 last_committed) | |
e02119d5 | 4837 | { |
12fcfd22 | 4838 | int ret = 0; |
6a912213 | 4839 | struct dentry *old_parent = NULL; |
de2b530b | 4840 | struct inode *orig_inode = inode; |
e02119d5 | 4841 | |
af4176b4 CM |
4842 | /* |
4843 | * for regular files, if its inode is already on disk, we don't | |
4844 | * have to worry about the parents at all. This is because | |
4845 | * we can use the last_unlink_trans field to record renames | |
4846 | * and other fun in this file. | |
4847 | */ | |
4848 | if (S_ISREG(inode->i_mode) && | |
4849 | BTRFS_I(inode)->generation <= last_committed && | |
4850 | BTRFS_I(inode)->last_unlink_trans <= last_committed) | |
4851 | goto out; | |
4852 | ||
12fcfd22 | 4853 | if (!S_ISDIR(inode->i_mode)) { |
2b0143b5 | 4854 | if (!parent || d_really_is_negative(parent) || sb != d_inode(parent)->i_sb) |
12fcfd22 | 4855 | goto out; |
2b0143b5 | 4856 | inode = d_inode(parent); |
12fcfd22 CM |
4857 | } |
4858 | ||
4859 | while (1) { | |
de2b530b JB |
4860 | /* |
4861 | * If we are logging a directory then we start with our inode, | |
4862 | * not our parents inode, so we need to skipp setting the | |
4863 | * logged_trans so that further down in the log code we don't | |
4864 | * think this inode has already been logged. | |
4865 | */ | |
4866 | if (inode != orig_inode) | |
4867 | BTRFS_I(inode)->logged_trans = trans->transid; | |
12fcfd22 CM |
4868 | smp_mb(); |
4869 | ||
2be63d5c | 4870 | if (btrfs_must_commit_transaction(trans, inode)) { |
12fcfd22 CM |
4871 | ret = 1; |
4872 | break; | |
4873 | } | |
4874 | ||
2b0143b5 | 4875 | if (!parent || d_really_is_negative(parent) || sb != d_inode(parent)->i_sb) |
12fcfd22 CM |
4876 | break; |
4877 | ||
76dda93c | 4878 | if (IS_ROOT(parent)) |
12fcfd22 CM |
4879 | break; |
4880 | ||
6a912213 JB |
4881 | parent = dget_parent(parent); |
4882 | dput(old_parent); | |
4883 | old_parent = parent; | |
2b0143b5 | 4884 | inode = d_inode(parent); |
12fcfd22 CM |
4885 | |
4886 | } | |
6a912213 | 4887 | dput(old_parent); |
12fcfd22 | 4888 | out: |
e02119d5 CM |
4889 | return ret; |
4890 | } | |
4891 | ||
2f2ff0ee FM |
4892 | struct btrfs_dir_list { |
4893 | u64 ino; | |
4894 | struct list_head list; | |
4895 | }; | |
4896 | ||
4897 | /* | |
4898 | * Log the inodes of the new dentries of a directory. See log_dir_items() for | |
4899 | * details about the why it is needed. | |
4900 | * This is a recursive operation - if an existing dentry corresponds to a | |
4901 | * directory, that directory's new entries are logged too (same behaviour as | |
4902 | * ext3/4, xfs, f2fs, reiserfs, nilfs2). Note that when logging the inodes | |
4903 | * the dentries point to we do not lock their i_mutex, otherwise lockdep | |
4904 | * complains about the following circular lock dependency / possible deadlock: | |
4905 | * | |
4906 | * CPU0 CPU1 | |
4907 | * ---- ---- | |
4908 | * lock(&type->i_mutex_dir_key#3/2); | |
4909 | * lock(sb_internal#2); | |
4910 | * lock(&type->i_mutex_dir_key#3/2); | |
4911 | * lock(&sb->s_type->i_mutex_key#14); | |
4912 | * | |
4913 | * Where sb_internal is the lock (a counter that works as a lock) acquired by | |
4914 | * sb_start_intwrite() in btrfs_start_transaction(). | |
4915 | * Not locking i_mutex of the inodes is still safe because: | |
4916 | * | |
4917 | * 1) For regular files we log with a mode of LOG_INODE_EXISTS. It's possible | |
4918 | * that while logging the inode new references (names) are added or removed | |
4919 | * from the inode, leaving the logged inode item with a link count that does | |
4920 | * not match the number of logged inode reference items. This is fine because | |
4921 | * at log replay time we compute the real number of links and correct the | |
4922 | * link count in the inode item (see replay_one_buffer() and | |
4923 | * link_to_fixup_dir()); | |
4924 | * | |
4925 | * 2) For directories we log with a mode of LOG_INODE_ALL. It's possible that | |
4926 | * while logging the inode's items new items with keys BTRFS_DIR_ITEM_KEY and | |
4927 | * BTRFS_DIR_INDEX_KEY are added to fs/subvol tree and the logged inode item | |
4928 | * has a size that doesn't match the sum of the lengths of all the logged | |
4929 | * names. This does not result in a problem because if a dir_item key is | |
4930 | * logged but its matching dir_index key is not logged, at log replay time we | |
4931 | * don't use it to replay the respective name (see replay_one_name()). On the | |
4932 | * other hand if only the dir_index key ends up being logged, the respective | |
4933 | * name is added to the fs/subvol tree with both the dir_item and dir_index | |
4934 | * keys created (see replay_one_name()). | |
4935 | * The directory's inode item with a wrong i_size is not a problem as well, | |
4936 | * since we don't use it at log replay time to set the i_size in the inode | |
4937 | * item of the fs/subvol tree (see overwrite_item()). | |
4938 | */ | |
4939 | static int log_new_dir_dentries(struct btrfs_trans_handle *trans, | |
4940 | struct btrfs_root *root, | |
4941 | struct inode *start_inode, | |
4942 | struct btrfs_log_ctx *ctx) | |
4943 | { | |
4944 | struct btrfs_root *log = root->log_root; | |
4945 | struct btrfs_path *path; | |
4946 | LIST_HEAD(dir_list); | |
4947 | struct btrfs_dir_list *dir_elem; | |
4948 | int ret = 0; | |
4949 | ||
4950 | path = btrfs_alloc_path(); | |
4951 | if (!path) | |
4952 | return -ENOMEM; | |
4953 | ||
4954 | dir_elem = kmalloc(sizeof(*dir_elem), GFP_NOFS); | |
4955 | if (!dir_elem) { | |
4956 | btrfs_free_path(path); | |
4957 | return -ENOMEM; | |
4958 | } | |
4959 | dir_elem->ino = btrfs_ino(start_inode); | |
4960 | list_add_tail(&dir_elem->list, &dir_list); | |
4961 | ||
4962 | while (!list_empty(&dir_list)) { | |
4963 | struct extent_buffer *leaf; | |
4964 | struct btrfs_key min_key; | |
4965 | int nritems; | |
4966 | int i; | |
4967 | ||
4968 | dir_elem = list_first_entry(&dir_list, struct btrfs_dir_list, | |
4969 | list); | |
4970 | if (ret) | |
4971 | goto next_dir_inode; | |
4972 | ||
4973 | min_key.objectid = dir_elem->ino; | |
4974 | min_key.type = BTRFS_DIR_ITEM_KEY; | |
4975 | min_key.offset = 0; | |
4976 | again: | |
4977 | btrfs_release_path(path); | |
4978 | ret = btrfs_search_forward(log, &min_key, path, trans->transid); | |
4979 | if (ret < 0) { | |
4980 | goto next_dir_inode; | |
4981 | } else if (ret > 0) { | |
4982 | ret = 0; | |
4983 | goto next_dir_inode; | |
4984 | } | |
4985 | ||
4986 | process_leaf: | |
4987 | leaf = path->nodes[0]; | |
4988 | nritems = btrfs_header_nritems(leaf); | |
4989 | for (i = path->slots[0]; i < nritems; i++) { | |
4990 | struct btrfs_dir_item *di; | |
4991 | struct btrfs_key di_key; | |
4992 | struct inode *di_inode; | |
4993 | struct btrfs_dir_list *new_dir_elem; | |
4994 | int log_mode = LOG_INODE_EXISTS; | |
4995 | int type; | |
4996 | ||
4997 | btrfs_item_key_to_cpu(leaf, &min_key, i); | |
4998 | if (min_key.objectid != dir_elem->ino || | |
4999 | min_key.type != BTRFS_DIR_ITEM_KEY) | |
5000 | goto next_dir_inode; | |
5001 | ||
5002 | di = btrfs_item_ptr(leaf, i, struct btrfs_dir_item); | |
5003 | type = btrfs_dir_type(leaf, di); | |
5004 | if (btrfs_dir_transid(leaf, di) < trans->transid && | |
5005 | type != BTRFS_FT_DIR) | |
5006 | continue; | |
5007 | btrfs_dir_item_key_to_cpu(leaf, di, &di_key); | |
5008 | if (di_key.type == BTRFS_ROOT_ITEM_KEY) | |
5009 | continue; | |
5010 | ||
5011 | di_inode = btrfs_iget(root->fs_info->sb, &di_key, | |
5012 | root, NULL); | |
5013 | if (IS_ERR(di_inode)) { | |
5014 | ret = PTR_ERR(di_inode); | |
5015 | goto next_dir_inode; | |
5016 | } | |
5017 | ||
5018 | if (btrfs_inode_in_log(di_inode, trans->transid)) { | |
5019 | iput(di_inode); | |
5020 | continue; | |
5021 | } | |
5022 | ||
5023 | ctx->log_new_dentries = false; | |
5024 | if (type == BTRFS_FT_DIR) | |
5025 | log_mode = LOG_INODE_ALL; | |
5026 | btrfs_release_path(path); | |
5027 | ret = btrfs_log_inode(trans, root, di_inode, | |
5028 | log_mode, 0, LLONG_MAX, ctx); | |
2be63d5c FM |
5029 | if (!ret && |
5030 | btrfs_must_commit_transaction(trans, di_inode)) | |
5031 | ret = 1; | |
2f2ff0ee FM |
5032 | iput(di_inode); |
5033 | if (ret) | |
5034 | goto next_dir_inode; | |
5035 | if (ctx->log_new_dentries) { | |
5036 | new_dir_elem = kmalloc(sizeof(*new_dir_elem), | |
5037 | GFP_NOFS); | |
5038 | if (!new_dir_elem) { | |
5039 | ret = -ENOMEM; | |
5040 | goto next_dir_inode; | |
5041 | } | |
5042 | new_dir_elem->ino = di_key.objectid; | |
5043 | list_add_tail(&new_dir_elem->list, &dir_list); | |
5044 | } | |
5045 | break; | |
5046 | } | |
5047 | if (i == nritems) { | |
5048 | ret = btrfs_next_leaf(log, path); | |
5049 | if (ret < 0) { | |
5050 | goto next_dir_inode; | |
5051 | } else if (ret > 0) { | |
5052 | ret = 0; | |
5053 | goto next_dir_inode; | |
5054 | } | |
5055 | goto process_leaf; | |
5056 | } | |
5057 | if (min_key.offset < (u64)-1) { | |
5058 | min_key.offset++; | |
5059 | goto again; | |
5060 | } | |
5061 | next_dir_inode: | |
5062 | list_del(&dir_elem->list); | |
5063 | kfree(dir_elem); | |
5064 | } | |
5065 | ||
5066 | btrfs_free_path(path); | |
5067 | return ret; | |
5068 | } | |
5069 | ||
18aa0922 FM |
5070 | static int btrfs_log_all_parents(struct btrfs_trans_handle *trans, |
5071 | struct inode *inode, | |
5072 | struct btrfs_log_ctx *ctx) | |
5073 | { | |
5074 | int ret; | |
5075 | struct btrfs_path *path; | |
5076 | struct btrfs_key key; | |
5077 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5078 | const u64 ino = btrfs_ino(inode); | |
5079 | ||
5080 | path = btrfs_alloc_path(); | |
5081 | if (!path) | |
5082 | return -ENOMEM; | |
5083 | path->skip_locking = 1; | |
5084 | path->search_commit_root = 1; | |
5085 | ||
5086 | key.objectid = ino; | |
5087 | key.type = BTRFS_INODE_REF_KEY; | |
5088 | key.offset = 0; | |
5089 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
5090 | if (ret < 0) | |
5091 | goto out; | |
5092 | ||
5093 | while (true) { | |
5094 | struct extent_buffer *leaf = path->nodes[0]; | |
5095 | int slot = path->slots[0]; | |
5096 | u32 cur_offset = 0; | |
5097 | u32 item_size; | |
5098 | unsigned long ptr; | |
5099 | ||
5100 | if (slot >= btrfs_header_nritems(leaf)) { | |
5101 | ret = btrfs_next_leaf(root, path); | |
5102 | if (ret < 0) | |
5103 | goto out; | |
5104 | else if (ret > 0) | |
5105 | break; | |
5106 | continue; | |
5107 | } | |
5108 | ||
5109 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
5110 | /* BTRFS_INODE_EXTREF_KEY is BTRFS_INODE_REF_KEY + 1 */ | |
5111 | if (key.objectid != ino || key.type > BTRFS_INODE_EXTREF_KEY) | |
5112 | break; | |
5113 | ||
5114 | item_size = btrfs_item_size_nr(leaf, slot); | |
5115 | ptr = btrfs_item_ptr_offset(leaf, slot); | |
5116 | while (cur_offset < item_size) { | |
5117 | struct btrfs_key inode_key; | |
5118 | struct inode *dir_inode; | |
5119 | ||
5120 | inode_key.type = BTRFS_INODE_ITEM_KEY; | |
5121 | inode_key.offset = 0; | |
5122 | ||
5123 | if (key.type == BTRFS_INODE_EXTREF_KEY) { | |
5124 | struct btrfs_inode_extref *extref; | |
5125 | ||
5126 | extref = (struct btrfs_inode_extref *) | |
5127 | (ptr + cur_offset); | |
5128 | inode_key.objectid = btrfs_inode_extref_parent( | |
5129 | leaf, extref); | |
5130 | cur_offset += sizeof(*extref); | |
5131 | cur_offset += btrfs_inode_extref_name_len(leaf, | |
5132 | extref); | |
5133 | } else { | |
5134 | inode_key.objectid = key.offset; | |
5135 | cur_offset = item_size; | |
5136 | } | |
5137 | ||
5138 | dir_inode = btrfs_iget(root->fs_info->sb, &inode_key, | |
5139 | root, NULL); | |
5140 | /* If parent inode was deleted, skip it. */ | |
5141 | if (IS_ERR(dir_inode)) | |
5142 | continue; | |
5143 | ||
5144 | ret = btrfs_log_inode(trans, root, dir_inode, | |
5145 | LOG_INODE_ALL, 0, LLONG_MAX, ctx); | |
2be63d5c FM |
5146 | if (!ret && |
5147 | btrfs_must_commit_transaction(trans, dir_inode)) | |
5148 | ret = 1; | |
18aa0922 FM |
5149 | iput(dir_inode); |
5150 | if (ret) | |
5151 | goto out; | |
5152 | } | |
5153 | path->slots[0]++; | |
5154 | } | |
5155 | ret = 0; | |
5156 | out: | |
5157 | btrfs_free_path(path); | |
5158 | return ret; | |
5159 | } | |
5160 | ||
e02119d5 CM |
5161 | /* |
5162 | * helper function around btrfs_log_inode to make sure newly created | |
5163 | * parent directories also end up in the log. A minimal inode and backref | |
5164 | * only logging is done of any parent directories that are older than | |
5165 | * the last committed transaction | |
5166 | */ | |
48a3b636 ES |
5167 | static int btrfs_log_inode_parent(struct btrfs_trans_handle *trans, |
5168 | struct btrfs_root *root, struct inode *inode, | |
49dae1bc FM |
5169 | struct dentry *parent, |
5170 | const loff_t start, | |
5171 | const loff_t end, | |
5172 | int exists_only, | |
8b050d35 | 5173 | struct btrfs_log_ctx *ctx) |
e02119d5 | 5174 | { |
12fcfd22 | 5175 | int inode_only = exists_only ? LOG_INODE_EXISTS : LOG_INODE_ALL; |
e02119d5 | 5176 | struct super_block *sb; |
6a912213 | 5177 | struct dentry *old_parent = NULL; |
12fcfd22 CM |
5178 | int ret = 0; |
5179 | u64 last_committed = root->fs_info->last_trans_committed; | |
2f2ff0ee FM |
5180 | bool log_dentries = false; |
5181 | struct inode *orig_inode = inode; | |
12fcfd22 CM |
5182 | |
5183 | sb = inode->i_sb; | |
5184 | ||
3a5e1404 SW |
5185 | if (btrfs_test_opt(root, NOTREELOG)) { |
5186 | ret = 1; | |
5187 | goto end_no_trans; | |
5188 | } | |
5189 | ||
995946dd MX |
5190 | /* |
5191 | * The prev transaction commit doesn't complete, we need do | |
5192 | * full commit by ourselves. | |
5193 | */ | |
12fcfd22 CM |
5194 | if (root->fs_info->last_trans_log_full_commit > |
5195 | root->fs_info->last_trans_committed) { | |
5196 | ret = 1; | |
5197 | goto end_no_trans; | |
5198 | } | |
5199 | ||
76dda93c YZ |
5200 | if (root != BTRFS_I(inode)->root || |
5201 | btrfs_root_refs(&root->root_item) == 0) { | |
5202 | ret = 1; | |
5203 | goto end_no_trans; | |
5204 | } | |
5205 | ||
12fcfd22 CM |
5206 | ret = check_parent_dirs_for_sync(trans, inode, parent, |
5207 | sb, last_committed); | |
5208 | if (ret) | |
5209 | goto end_no_trans; | |
e02119d5 | 5210 | |
22ee6985 | 5211 | if (btrfs_inode_in_log(inode, trans->transid)) { |
257c62e1 CM |
5212 | ret = BTRFS_NO_LOG_SYNC; |
5213 | goto end_no_trans; | |
5214 | } | |
5215 | ||
8b050d35 | 5216 | ret = start_log_trans(trans, root, ctx); |
4a500fd1 | 5217 | if (ret) |
e87ac136 | 5218 | goto end_no_trans; |
e02119d5 | 5219 | |
8407f553 | 5220 | ret = btrfs_log_inode(trans, root, inode, inode_only, start, end, ctx); |
4a500fd1 YZ |
5221 | if (ret) |
5222 | goto end_trans; | |
12fcfd22 | 5223 | |
af4176b4 CM |
5224 | /* |
5225 | * for regular files, if its inode is already on disk, we don't | |
5226 | * have to worry about the parents at all. This is because | |
5227 | * we can use the last_unlink_trans field to record renames | |
5228 | * and other fun in this file. | |
5229 | */ | |
5230 | if (S_ISREG(inode->i_mode) && | |
5231 | BTRFS_I(inode)->generation <= last_committed && | |
4a500fd1 YZ |
5232 | BTRFS_I(inode)->last_unlink_trans <= last_committed) { |
5233 | ret = 0; | |
5234 | goto end_trans; | |
5235 | } | |
af4176b4 | 5236 | |
2f2ff0ee FM |
5237 | if (S_ISDIR(inode->i_mode) && ctx && ctx->log_new_dentries) |
5238 | log_dentries = true; | |
5239 | ||
18aa0922 FM |
5240 | /* |
5241 | * On unlink we must make sure all our current and old parent directores | |
5242 | * inodes are fully logged. This is to prevent leaving dangling | |
5243 | * directory index entries in directories that were our parents but are | |
5244 | * not anymore. Not doing this results in old parent directory being | |
5245 | * impossible to delete after log replay (rmdir will always fail with | |
5246 | * error -ENOTEMPTY). | |
5247 | * | |
5248 | * Example 1: | |
5249 | * | |
5250 | * mkdir testdir | |
5251 | * touch testdir/foo | |
5252 | * ln testdir/foo testdir/bar | |
5253 | * sync | |
5254 | * unlink testdir/bar | |
5255 | * xfs_io -c fsync testdir/foo | |
5256 | * <power failure> | |
5257 | * mount fs, triggers log replay | |
5258 | * | |
5259 | * If we don't log the parent directory (testdir), after log replay the | |
5260 | * directory still has an entry pointing to the file inode using the bar | |
5261 | * name, but a matching BTRFS_INODE_[REF|EXTREF]_KEY does not exist and | |
5262 | * the file inode has a link count of 1. | |
5263 | * | |
5264 | * Example 2: | |
5265 | * | |
5266 | * mkdir testdir | |
5267 | * touch foo | |
5268 | * ln foo testdir/foo2 | |
5269 | * ln foo testdir/foo3 | |
5270 | * sync | |
5271 | * unlink testdir/foo3 | |
5272 | * xfs_io -c fsync foo | |
5273 | * <power failure> | |
5274 | * mount fs, triggers log replay | |
5275 | * | |
5276 | * Similar as the first example, after log replay the parent directory | |
5277 | * testdir still has an entry pointing to the inode file with name foo3 | |
5278 | * but the file inode does not have a matching BTRFS_INODE_REF_KEY item | |
5279 | * and has a link count of 2. | |
5280 | */ | |
5281 | if (BTRFS_I(inode)->last_unlink_trans > last_committed) { | |
5282 | ret = btrfs_log_all_parents(trans, orig_inode, ctx); | |
5283 | if (ret) | |
5284 | goto end_trans; | |
5285 | } | |
5286 | ||
12fcfd22 | 5287 | while (1) { |
2b0143b5 | 5288 | if (!parent || d_really_is_negative(parent) || sb != d_inode(parent)->i_sb) |
e02119d5 CM |
5289 | break; |
5290 | ||
2b0143b5 | 5291 | inode = d_inode(parent); |
76dda93c YZ |
5292 | if (root != BTRFS_I(inode)->root) |
5293 | break; | |
5294 | ||
18aa0922 FM |
5295 | if (BTRFS_I(inode)->generation > last_committed) { |
5296 | ret = btrfs_log_inode(trans, root, inode, | |
5297 | LOG_INODE_EXISTS, | |
8407f553 | 5298 | 0, LLONG_MAX, ctx); |
4a500fd1 YZ |
5299 | if (ret) |
5300 | goto end_trans; | |
12fcfd22 | 5301 | } |
76dda93c | 5302 | if (IS_ROOT(parent)) |
e02119d5 | 5303 | break; |
12fcfd22 | 5304 | |
6a912213 JB |
5305 | parent = dget_parent(parent); |
5306 | dput(old_parent); | |
5307 | old_parent = parent; | |
e02119d5 | 5308 | } |
2f2ff0ee FM |
5309 | if (log_dentries) |
5310 | ret = log_new_dir_dentries(trans, root, orig_inode, ctx); | |
5311 | else | |
5312 | ret = 0; | |
4a500fd1 | 5313 | end_trans: |
6a912213 | 5314 | dput(old_parent); |
4a500fd1 | 5315 | if (ret < 0) { |
995946dd | 5316 | btrfs_set_log_full_commit(root->fs_info, trans); |
4a500fd1 YZ |
5317 | ret = 1; |
5318 | } | |
8b050d35 MX |
5319 | |
5320 | if (ret) | |
5321 | btrfs_remove_log_ctx(root, ctx); | |
12fcfd22 CM |
5322 | btrfs_end_log_trans(root); |
5323 | end_no_trans: | |
5324 | return ret; | |
e02119d5 CM |
5325 | } |
5326 | ||
5327 | /* | |
5328 | * it is not safe to log dentry if the chunk root has added new | |
5329 | * chunks. This returns 0 if the dentry was logged, and 1 otherwise. | |
5330 | * If this returns 1, you must commit the transaction to safely get your | |
5331 | * data on disk. | |
5332 | */ | |
5333 | int btrfs_log_dentry_safe(struct btrfs_trans_handle *trans, | |
8b050d35 | 5334 | struct btrfs_root *root, struct dentry *dentry, |
49dae1bc FM |
5335 | const loff_t start, |
5336 | const loff_t end, | |
8b050d35 | 5337 | struct btrfs_log_ctx *ctx) |
e02119d5 | 5338 | { |
6a912213 JB |
5339 | struct dentry *parent = dget_parent(dentry); |
5340 | int ret; | |
5341 | ||
2b0143b5 | 5342 | ret = btrfs_log_inode_parent(trans, root, d_inode(dentry), parent, |
49dae1bc | 5343 | start, end, 0, ctx); |
6a912213 JB |
5344 | dput(parent); |
5345 | ||
5346 | return ret; | |
e02119d5 CM |
5347 | } |
5348 | ||
5349 | /* | |
5350 | * should be called during mount to recover any replay any log trees | |
5351 | * from the FS | |
5352 | */ | |
5353 | int btrfs_recover_log_trees(struct btrfs_root *log_root_tree) | |
5354 | { | |
5355 | int ret; | |
5356 | struct btrfs_path *path; | |
5357 | struct btrfs_trans_handle *trans; | |
5358 | struct btrfs_key key; | |
5359 | struct btrfs_key found_key; | |
5360 | struct btrfs_key tmp_key; | |
5361 | struct btrfs_root *log; | |
5362 | struct btrfs_fs_info *fs_info = log_root_tree->fs_info; | |
5363 | struct walk_control wc = { | |
5364 | .process_func = process_one_buffer, | |
5365 | .stage = 0, | |
5366 | }; | |
5367 | ||
e02119d5 | 5368 | path = btrfs_alloc_path(); |
db5b493a TI |
5369 | if (!path) |
5370 | return -ENOMEM; | |
5371 | ||
5372 | fs_info->log_root_recovering = 1; | |
e02119d5 | 5373 | |
4a500fd1 | 5374 | trans = btrfs_start_transaction(fs_info->tree_root, 0); |
79787eaa JM |
5375 | if (IS_ERR(trans)) { |
5376 | ret = PTR_ERR(trans); | |
5377 | goto error; | |
5378 | } | |
e02119d5 CM |
5379 | |
5380 | wc.trans = trans; | |
5381 | wc.pin = 1; | |
5382 | ||
db5b493a | 5383 | ret = walk_log_tree(trans, log_root_tree, &wc); |
79787eaa | 5384 | if (ret) { |
a4553fef | 5385 | btrfs_std_error(fs_info, ret, "Failed to pin buffers while " |
79787eaa JM |
5386 | "recovering log root tree."); |
5387 | goto error; | |
5388 | } | |
e02119d5 CM |
5389 | |
5390 | again: | |
5391 | key.objectid = BTRFS_TREE_LOG_OBJECTID; | |
5392 | key.offset = (u64)-1; | |
962a298f | 5393 | key.type = BTRFS_ROOT_ITEM_KEY; |
e02119d5 | 5394 | |
d397712b | 5395 | while (1) { |
e02119d5 | 5396 | ret = btrfs_search_slot(NULL, log_root_tree, &key, path, 0, 0); |
79787eaa JM |
5397 | |
5398 | if (ret < 0) { | |
a4553fef | 5399 | btrfs_std_error(fs_info, ret, |
79787eaa JM |
5400 | "Couldn't find tree log root."); |
5401 | goto error; | |
5402 | } | |
e02119d5 CM |
5403 | if (ret > 0) { |
5404 | if (path->slots[0] == 0) | |
5405 | break; | |
5406 | path->slots[0]--; | |
5407 | } | |
5408 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, | |
5409 | path->slots[0]); | |
b3b4aa74 | 5410 | btrfs_release_path(path); |
e02119d5 CM |
5411 | if (found_key.objectid != BTRFS_TREE_LOG_OBJECTID) |
5412 | break; | |
5413 | ||
cb517eab | 5414 | log = btrfs_read_fs_root(log_root_tree, &found_key); |
79787eaa JM |
5415 | if (IS_ERR(log)) { |
5416 | ret = PTR_ERR(log); | |
a4553fef | 5417 | btrfs_std_error(fs_info, ret, |
79787eaa JM |
5418 | "Couldn't read tree log root."); |
5419 | goto error; | |
5420 | } | |
e02119d5 CM |
5421 | |
5422 | tmp_key.objectid = found_key.offset; | |
5423 | tmp_key.type = BTRFS_ROOT_ITEM_KEY; | |
5424 | tmp_key.offset = (u64)-1; | |
5425 | ||
5426 | wc.replay_dest = btrfs_read_fs_root_no_name(fs_info, &tmp_key); | |
79787eaa JM |
5427 | if (IS_ERR(wc.replay_dest)) { |
5428 | ret = PTR_ERR(wc.replay_dest); | |
b50c6e25 JB |
5429 | free_extent_buffer(log->node); |
5430 | free_extent_buffer(log->commit_root); | |
5431 | kfree(log); | |
a4553fef | 5432 | btrfs_std_error(fs_info, ret, "Couldn't read target root " |
79787eaa JM |
5433 | "for tree log recovery."); |
5434 | goto error; | |
5435 | } | |
e02119d5 | 5436 | |
07d400a6 | 5437 | wc.replay_dest->log_root = log; |
5d4f98a2 | 5438 | btrfs_record_root_in_trans(trans, wc.replay_dest); |
e02119d5 | 5439 | ret = walk_log_tree(trans, log, &wc); |
e02119d5 | 5440 | |
b50c6e25 | 5441 | if (!ret && wc.stage == LOG_WALK_REPLAY_ALL) { |
e02119d5 CM |
5442 | ret = fixup_inode_link_counts(trans, wc.replay_dest, |
5443 | path); | |
e02119d5 CM |
5444 | } |
5445 | ||
5446 | key.offset = found_key.offset - 1; | |
07d400a6 | 5447 | wc.replay_dest->log_root = NULL; |
e02119d5 | 5448 | free_extent_buffer(log->node); |
b263c2c8 | 5449 | free_extent_buffer(log->commit_root); |
e02119d5 CM |
5450 | kfree(log); |
5451 | ||
b50c6e25 JB |
5452 | if (ret) |
5453 | goto error; | |
5454 | ||
e02119d5 CM |
5455 | if (found_key.offset == 0) |
5456 | break; | |
5457 | } | |
b3b4aa74 | 5458 | btrfs_release_path(path); |
e02119d5 CM |
5459 | |
5460 | /* step one is to pin it all, step two is to replay just inodes */ | |
5461 | if (wc.pin) { | |
5462 | wc.pin = 0; | |
5463 | wc.process_func = replay_one_buffer; | |
5464 | wc.stage = LOG_WALK_REPLAY_INODES; | |
5465 | goto again; | |
5466 | } | |
5467 | /* step three is to replay everything */ | |
5468 | if (wc.stage < LOG_WALK_REPLAY_ALL) { | |
5469 | wc.stage++; | |
5470 | goto again; | |
5471 | } | |
5472 | ||
5473 | btrfs_free_path(path); | |
5474 | ||
abefa55a JB |
5475 | /* step 4: commit the transaction, which also unpins the blocks */ |
5476 | ret = btrfs_commit_transaction(trans, fs_info->tree_root); | |
5477 | if (ret) | |
5478 | return ret; | |
5479 | ||
e02119d5 CM |
5480 | free_extent_buffer(log_root_tree->node); |
5481 | log_root_tree->log_root = NULL; | |
5482 | fs_info->log_root_recovering = 0; | |
e02119d5 | 5483 | kfree(log_root_tree); |
79787eaa | 5484 | |
abefa55a | 5485 | return 0; |
79787eaa | 5486 | error: |
b50c6e25 JB |
5487 | if (wc.trans) |
5488 | btrfs_end_transaction(wc.trans, fs_info->tree_root); | |
79787eaa JM |
5489 | btrfs_free_path(path); |
5490 | return ret; | |
e02119d5 | 5491 | } |
12fcfd22 CM |
5492 | |
5493 | /* | |
5494 | * there are some corner cases where we want to force a full | |
5495 | * commit instead of allowing a directory to be logged. | |
5496 | * | |
5497 | * They revolve around files there were unlinked from the directory, and | |
5498 | * this function updates the parent directory so that a full commit is | |
5499 | * properly done if it is fsync'd later after the unlinks are done. | |
2be63d5c FM |
5500 | * |
5501 | * Must be called before the unlink operations (updates to the subvolume tree, | |
5502 | * inodes, etc) are done. | |
12fcfd22 CM |
5503 | */ |
5504 | void btrfs_record_unlink_dir(struct btrfs_trans_handle *trans, | |
5505 | struct inode *dir, struct inode *inode, | |
5506 | int for_rename) | |
5507 | { | |
af4176b4 CM |
5508 | /* |
5509 | * when we're logging a file, if it hasn't been renamed | |
5510 | * or unlinked, and its inode is fully committed on disk, | |
5511 | * we don't have to worry about walking up the directory chain | |
5512 | * to log its parents. | |
5513 | * | |
5514 | * So, we use the last_unlink_trans field to put this transid | |
5515 | * into the file. When the file is logged we check it and | |
5516 | * don't log the parents if the file is fully on disk. | |
5517 | */ | |
2be63d5c FM |
5518 | if (S_ISREG(inode->i_mode)) { |
5519 | mutex_lock(&BTRFS_I(inode)->log_mutex); | |
af4176b4 | 5520 | BTRFS_I(inode)->last_unlink_trans = trans->transid; |
2be63d5c FM |
5521 | mutex_unlock(&BTRFS_I(inode)->log_mutex); |
5522 | } | |
af4176b4 | 5523 | |
12fcfd22 CM |
5524 | /* |
5525 | * if this directory was already logged any new | |
5526 | * names for this file/dir will get recorded | |
5527 | */ | |
5528 | smp_mb(); | |
5529 | if (BTRFS_I(dir)->logged_trans == trans->transid) | |
5530 | return; | |
5531 | ||
5532 | /* | |
5533 | * if the inode we're about to unlink was logged, | |
5534 | * the log will be properly updated for any new names | |
5535 | */ | |
5536 | if (BTRFS_I(inode)->logged_trans == trans->transid) | |
5537 | return; | |
5538 | ||
5539 | /* | |
5540 | * when renaming files across directories, if the directory | |
5541 | * there we're unlinking from gets fsync'd later on, there's | |
5542 | * no way to find the destination directory later and fsync it | |
5543 | * properly. So, we have to be conservative and force commits | |
5544 | * so the new name gets discovered. | |
5545 | */ | |
5546 | if (for_rename) | |
5547 | goto record; | |
5548 | ||
5549 | /* we can safely do the unlink without any special recording */ | |
5550 | return; | |
5551 | ||
5552 | record: | |
2be63d5c | 5553 | mutex_lock(&BTRFS_I(dir)->log_mutex); |
12fcfd22 | 5554 | BTRFS_I(dir)->last_unlink_trans = trans->transid; |
2be63d5c | 5555 | mutex_unlock(&BTRFS_I(dir)->log_mutex); |
1ec9a1ae FM |
5556 | } |
5557 | ||
5558 | /* | |
5559 | * Make sure that if someone attempts to fsync the parent directory of a deleted | |
5560 | * snapshot, it ends up triggering a transaction commit. This is to guarantee | |
5561 | * that after replaying the log tree of the parent directory's root we will not | |
5562 | * see the snapshot anymore and at log replay time we will not see any log tree | |
5563 | * corresponding to the deleted snapshot's root, which could lead to replaying | |
5564 | * it after replaying the log tree of the parent directory (which would replay | |
5565 | * the snapshot delete operation). | |
2be63d5c FM |
5566 | * |
5567 | * Must be called before the actual snapshot destroy operation (updates to the | |
5568 | * parent root and tree of tree roots trees, etc) are done. | |
1ec9a1ae FM |
5569 | */ |
5570 | void btrfs_record_snapshot_destroy(struct btrfs_trans_handle *trans, | |
5571 | struct inode *dir) | |
5572 | { | |
2be63d5c | 5573 | mutex_lock(&BTRFS_I(dir)->log_mutex); |
1ec9a1ae | 5574 | BTRFS_I(dir)->last_unlink_trans = trans->transid; |
2be63d5c | 5575 | mutex_unlock(&BTRFS_I(dir)->log_mutex); |
12fcfd22 CM |
5576 | } |
5577 | ||
5578 | /* | |
5579 | * Call this after adding a new name for a file and it will properly | |
5580 | * update the log to reflect the new name. | |
5581 | * | |
5582 | * It will return zero if all goes well, and it will return 1 if a | |
5583 | * full transaction commit is required. | |
5584 | */ | |
5585 | int btrfs_log_new_name(struct btrfs_trans_handle *trans, | |
5586 | struct inode *inode, struct inode *old_dir, | |
5587 | struct dentry *parent) | |
5588 | { | |
5589 | struct btrfs_root * root = BTRFS_I(inode)->root; | |
5590 | ||
af4176b4 CM |
5591 | /* |
5592 | * this will force the logging code to walk the dentry chain | |
5593 | * up for the file | |
5594 | */ | |
5595 | if (S_ISREG(inode->i_mode)) | |
5596 | BTRFS_I(inode)->last_unlink_trans = trans->transid; | |
5597 | ||
12fcfd22 CM |
5598 | /* |
5599 | * if this inode hasn't been logged and directory we're renaming it | |
5600 | * from hasn't been logged, we don't need to log it | |
5601 | */ | |
5602 | if (BTRFS_I(inode)->logged_trans <= | |
5603 | root->fs_info->last_trans_committed && | |
5604 | (!old_dir || BTRFS_I(old_dir)->logged_trans <= | |
5605 | root->fs_info->last_trans_committed)) | |
5606 | return 0; | |
5607 | ||
49dae1bc FM |
5608 | return btrfs_log_inode_parent(trans, root, inode, parent, 0, |
5609 | LLONG_MAX, 1, NULL); | |
12fcfd22 CM |
5610 | } |
5611 |