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" |
e02119d5 CM |
29 | |
30 | /* magic values for the inode_only field in btrfs_log_inode: | |
31 | * | |
32 | * LOG_INODE_ALL means to log everything | |
33 | * LOG_INODE_EXISTS means to log just enough to recreate the inode | |
34 | * during log replay | |
35 | */ | |
36 | #define LOG_INODE_ALL 0 | |
37 | #define LOG_INODE_EXISTS 1 | |
38 | ||
12fcfd22 CM |
39 | /* |
40 | * directory trouble cases | |
41 | * | |
42 | * 1) on rename or unlink, if the inode being unlinked isn't in the fsync | |
43 | * log, we must force a full commit before doing an fsync of the directory | |
44 | * where the unlink was done. | |
45 | * ---> record transid of last unlink/rename per directory | |
46 | * | |
47 | * mkdir foo/some_dir | |
48 | * normal commit | |
49 | * rename foo/some_dir foo2/some_dir | |
50 | * mkdir foo/some_dir | |
51 | * fsync foo/some_dir/some_file | |
52 | * | |
53 | * The fsync above will unlink the original some_dir without recording | |
54 | * it in its new location (foo2). After a crash, some_dir will be gone | |
55 | * unless the fsync of some_file forces a full commit | |
56 | * | |
57 | * 2) we must log any new names for any file or dir that is in the fsync | |
58 | * log. ---> check inode while renaming/linking. | |
59 | * | |
60 | * 2a) we must log any new names for any file or dir during rename | |
61 | * when the directory they are being removed from was logged. | |
62 | * ---> check inode and old parent dir during rename | |
63 | * | |
64 | * 2a is actually the more important variant. With the extra logging | |
65 | * a crash might unlink the old name without recreating the new one | |
66 | * | |
67 | * 3) after a crash, we must go through any directories with a link count | |
68 | * of zero and redo the rm -rf | |
69 | * | |
70 | * mkdir f1/foo | |
71 | * normal commit | |
72 | * rm -rf f1/foo | |
73 | * fsync(f1) | |
74 | * | |
75 | * The directory f1 was fully removed from the FS, but fsync was never | |
76 | * called on f1, only its parent dir. After a crash the rm -rf must | |
77 | * be replayed. This must be able to recurse down the entire | |
78 | * directory tree. The inode link count fixup code takes care of the | |
79 | * ugly details. | |
80 | */ | |
81 | ||
e02119d5 CM |
82 | /* |
83 | * stages for the tree walking. The first | |
84 | * stage (0) is to only pin down the blocks we find | |
85 | * the second stage (1) is to make sure that all the inodes | |
86 | * we find in the log are created in the subvolume. | |
87 | * | |
88 | * The last stage is to deal with directories and links and extents | |
89 | * and all the other fun semantics | |
90 | */ | |
91 | #define LOG_WALK_PIN_ONLY 0 | |
92 | #define LOG_WALK_REPLAY_INODES 1 | |
dd8e7217 JB |
93 | #define LOG_WALK_REPLAY_DIR_INDEX 2 |
94 | #define LOG_WALK_REPLAY_ALL 3 | |
e02119d5 | 95 | |
12fcfd22 | 96 | static int btrfs_log_inode(struct btrfs_trans_handle *trans, |
49dae1bc FM |
97 | struct btrfs_root *root, struct inode *inode, |
98 | int inode_only, | |
99 | const loff_t start, | |
8407f553 FM |
100 | const loff_t end, |
101 | struct btrfs_log_ctx *ctx); | |
ec051c0f YZ |
102 | static int link_to_fixup_dir(struct btrfs_trans_handle *trans, |
103 | struct btrfs_root *root, | |
104 | struct btrfs_path *path, u64 objectid); | |
12fcfd22 CM |
105 | static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans, |
106 | struct btrfs_root *root, | |
107 | struct btrfs_root *log, | |
108 | struct btrfs_path *path, | |
109 | u64 dirid, int del_all); | |
e02119d5 CM |
110 | |
111 | /* | |
112 | * tree logging is a special write ahead log used to make sure that | |
113 | * fsyncs and O_SYNCs can happen without doing full tree commits. | |
114 | * | |
115 | * Full tree commits are expensive because they require commonly | |
116 | * modified blocks to be recowed, creating many dirty pages in the | |
117 | * extent tree an 4x-6x higher write load than ext3. | |
118 | * | |
119 | * Instead of doing a tree commit on every fsync, we use the | |
120 | * key ranges and transaction ids to find items for a given file or directory | |
121 | * that have changed in this transaction. Those items are copied into | |
122 | * a special tree (one per subvolume root), that tree is written to disk | |
123 | * and then the fsync is considered complete. | |
124 | * | |
125 | * After a crash, items are copied out of the log-tree back into the | |
126 | * subvolume tree. Any file data extents found are recorded in the extent | |
127 | * allocation tree, and the log-tree freed. | |
128 | * | |
129 | * The log tree is read three times, once to pin down all the extents it is | |
130 | * using in ram and once, once to create all the inodes logged in the tree | |
131 | * and once to do all the other items. | |
132 | */ | |
133 | ||
e02119d5 CM |
134 | /* |
135 | * start a sub transaction and setup the log tree | |
136 | * this increments the log tree writer count to make the people | |
137 | * syncing the tree wait for us to finish | |
138 | */ | |
139 | static int start_log_trans(struct btrfs_trans_handle *trans, | |
8b050d35 MX |
140 | struct btrfs_root *root, |
141 | struct btrfs_log_ctx *ctx) | |
e02119d5 | 142 | { |
34eb2a52 | 143 | int ret = 0; |
7237f183 YZ |
144 | |
145 | mutex_lock(&root->log_mutex); | |
34eb2a52 | 146 | |
7237f183 | 147 | if (root->log_root) { |
995946dd | 148 | if (btrfs_need_log_full_commit(root->fs_info, trans)) { |
50471a38 MX |
149 | ret = -EAGAIN; |
150 | goto out; | |
151 | } | |
34eb2a52 | 152 | |
ff782e0a | 153 | if (!root->log_start_pid) { |
27cdeb70 | 154 | clear_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state); |
34eb2a52 | 155 | root->log_start_pid = current->pid; |
ff782e0a | 156 | } else if (root->log_start_pid != current->pid) { |
27cdeb70 | 157 | set_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state); |
ff782e0a | 158 | } |
34eb2a52 Z |
159 | } else { |
160 | mutex_lock(&root->fs_info->tree_log_mutex); | |
161 | if (!root->fs_info->log_root_tree) | |
162 | ret = btrfs_init_log_root_tree(trans, root->fs_info); | |
163 | mutex_unlock(&root->fs_info->tree_log_mutex); | |
164 | if (ret) | |
165 | goto out; | |
ff782e0a | 166 | |
e02119d5 | 167 | ret = btrfs_add_log_tree(trans, root); |
4a500fd1 | 168 | if (ret) |
e87ac136 | 169 | goto out; |
34eb2a52 Z |
170 | |
171 | clear_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state); | |
172 | root->log_start_pid = current->pid; | |
e02119d5 | 173 | } |
34eb2a52 | 174 | |
2ecb7923 | 175 | atomic_inc(&root->log_batch); |
7237f183 | 176 | atomic_inc(&root->log_writers); |
8b050d35 | 177 | if (ctx) { |
34eb2a52 | 178 | int index = root->log_transid % 2; |
8b050d35 | 179 | list_add_tail(&ctx->list, &root->log_ctxs[index]); |
d1433deb | 180 | ctx->log_transid = root->log_transid; |
8b050d35 | 181 | } |
34eb2a52 | 182 | |
e87ac136 | 183 | out: |
7237f183 | 184 | mutex_unlock(&root->log_mutex); |
e87ac136 | 185 | return ret; |
e02119d5 CM |
186 | } |
187 | ||
188 | /* | |
189 | * returns 0 if there was a log transaction running and we were able | |
190 | * to join, or returns -ENOENT if there were not transactions | |
191 | * in progress | |
192 | */ | |
193 | static int join_running_log_trans(struct btrfs_root *root) | |
194 | { | |
195 | int ret = -ENOENT; | |
196 | ||
197 | smp_mb(); | |
198 | if (!root->log_root) | |
199 | return -ENOENT; | |
200 | ||
7237f183 | 201 | mutex_lock(&root->log_mutex); |
e02119d5 CM |
202 | if (root->log_root) { |
203 | ret = 0; | |
7237f183 | 204 | atomic_inc(&root->log_writers); |
e02119d5 | 205 | } |
7237f183 | 206 | mutex_unlock(&root->log_mutex); |
e02119d5 CM |
207 | return ret; |
208 | } | |
209 | ||
12fcfd22 CM |
210 | /* |
211 | * This either makes the current running log transaction wait | |
212 | * until you call btrfs_end_log_trans() or it makes any future | |
213 | * log transactions wait until you call btrfs_end_log_trans() | |
214 | */ | |
215 | int btrfs_pin_log_trans(struct btrfs_root *root) | |
216 | { | |
217 | int ret = -ENOENT; | |
218 | ||
219 | mutex_lock(&root->log_mutex); | |
220 | atomic_inc(&root->log_writers); | |
221 | mutex_unlock(&root->log_mutex); | |
222 | return ret; | |
223 | } | |
224 | ||
e02119d5 CM |
225 | /* |
226 | * indicate we're done making changes to the log tree | |
227 | * and wake up anyone waiting to do a sync | |
228 | */ | |
143bede5 | 229 | void btrfs_end_log_trans(struct btrfs_root *root) |
e02119d5 | 230 | { |
7237f183 YZ |
231 | if (atomic_dec_and_test(&root->log_writers)) { |
232 | smp_mb(); | |
233 | if (waitqueue_active(&root->log_writer_wait)) | |
234 | wake_up(&root->log_writer_wait); | |
235 | } | |
e02119d5 CM |
236 | } |
237 | ||
238 | ||
239 | /* | |
240 | * the walk control struct is used to pass state down the chain when | |
241 | * processing the log tree. The stage field tells us which part | |
242 | * of the log tree processing we are currently doing. The others | |
243 | * are state fields used for that specific part | |
244 | */ | |
245 | struct walk_control { | |
246 | /* should we free the extent on disk when done? This is used | |
247 | * at transaction commit time while freeing a log tree | |
248 | */ | |
249 | int free; | |
250 | ||
251 | /* should we write out the extent buffer? This is used | |
252 | * while flushing the log tree to disk during a sync | |
253 | */ | |
254 | int write; | |
255 | ||
256 | /* should we wait for the extent buffer io to finish? Also used | |
257 | * while flushing the log tree to disk for a sync | |
258 | */ | |
259 | int wait; | |
260 | ||
261 | /* pin only walk, we record which extents on disk belong to the | |
262 | * log trees | |
263 | */ | |
264 | int pin; | |
265 | ||
266 | /* what stage of the replay code we're currently in */ | |
267 | int stage; | |
268 | ||
269 | /* the root we are currently replaying */ | |
270 | struct btrfs_root *replay_dest; | |
271 | ||
272 | /* the trans handle for the current replay */ | |
273 | struct btrfs_trans_handle *trans; | |
274 | ||
275 | /* the function that gets used to process blocks we find in the | |
276 | * tree. Note the extent_buffer might not be up to date when it is | |
277 | * passed in, and it must be checked or read if you need the data | |
278 | * inside it | |
279 | */ | |
280 | int (*process_func)(struct btrfs_root *log, struct extent_buffer *eb, | |
281 | struct walk_control *wc, u64 gen); | |
282 | }; | |
283 | ||
284 | /* | |
285 | * process_func used to pin down extents, write them or wait on them | |
286 | */ | |
287 | static int process_one_buffer(struct btrfs_root *log, | |
288 | struct extent_buffer *eb, | |
289 | struct walk_control *wc, u64 gen) | |
290 | { | |
b50c6e25 JB |
291 | int ret = 0; |
292 | ||
8c2a1a30 JB |
293 | /* |
294 | * If this fs is mixed then we need to be able to process the leaves to | |
295 | * pin down any logged extents, so we have to read the block. | |
296 | */ | |
297 | if (btrfs_fs_incompat(log->fs_info, MIXED_GROUPS)) { | |
298 | ret = btrfs_read_buffer(eb, gen); | |
299 | if (ret) | |
300 | return ret; | |
301 | } | |
302 | ||
04018de5 | 303 | if (wc->pin) |
b50c6e25 JB |
304 | ret = btrfs_pin_extent_for_log_replay(log->fs_info->extent_root, |
305 | eb->start, eb->len); | |
e02119d5 | 306 | |
b50c6e25 | 307 | if (!ret && btrfs_buffer_uptodate(eb, gen, 0)) { |
8c2a1a30 JB |
308 | if (wc->pin && btrfs_header_level(eb) == 0) |
309 | ret = btrfs_exclude_logged_extents(log, eb); | |
e02119d5 CM |
310 | if (wc->write) |
311 | btrfs_write_tree_block(eb); | |
312 | if (wc->wait) | |
313 | btrfs_wait_tree_block_writeback(eb); | |
314 | } | |
b50c6e25 | 315 | return ret; |
e02119d5 CM |
316 | } |
317 | ||
318 | /* | |
319 | * Item overwrite used by replay and tree logging. eb, slot and key all refer | |
320 | * to the src data we are copying out. | |
321 | * | |
322 | * root is the tree we are copying into, and path is a scratch | |
323 | * path for use in this function (it should be released on entry and | |
324 | * will be released on exit). | |
325 | * | |
326 | * If the key is already in the destination tree the existing item is | |
327 | * overwritten. If the existing item isn't big enough, it is extended. | |
328 | * If it is too large, it is truncated. | |
329 | * | |
330 | * If the key isn't in the destination yet, a new item is inserted. | |
331 | */ | |
332 | static noinline int overwrite_item(struct btrfs_trans_handle *trans, | |
333 | struct btrfs_root *root, | |
334 | struct btrfs_path *path, | |
335 | struct extent_buffer *eb, int slot, | |
336 | struct btrfs_key *key) | |
337 | { | |
338 | int ret; | |
339 | u32 item_size; | |
340 | u64 saved_i_size = 0; | |
341 | int save_old_i_size = 0; | |
342 | unsigned long src_ptr; | |
343 | unsigned long dst_ptr; | |
344 | int overwrite_root = 0; | |
4bc4bee4 | 345 | bool inode_item = key->type == BTRFS_INODE_ITEM_KEY; |
e02119d5 CM |
346 | |
347 | if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) | |
348 | overwrite_root = 1; | |
349 | ||
350 | item_size = btrfs_item_size_nr(eb, slot); | |
351 | src_ptr = btrfs_item_ptr_offset(eb, slot); | |
352 | ||
353 | /* look for the key in the destination tree */ | |
354 | ret = btrfs_search_slot(NULL, root, key, path, 0, 0); | |
4bc4bee4 JB |
355 | if (ret < 0) |
356 | return ret; | |
357 | ||
e02119d5 CM |
358 | if (ret == 0) { |
359 | char *src_copy; | |
360 | char *dst_copy; | |
361 | u32 dst_size = btrfs_item_size_nr(path->nodes[0], | |
362 | path->slots[0]); | |
363 | if (dst_size != item_size) | |
364 | goto insert; | |
365 | ||
366 | if (item_size == 0) { | |
b3b4aa74 | 367 | btrfs_release_path(path); |
e02119d5 CM |
368 | return 0; |
369 | } | |
370 | dst_copy = kmalloc(item_size, GFP_NOFS); | |
371 | src_copy = kmalloc(item_size, GFP_NOFS); | |
2a29edc6 | 372 | if (!dst_copy || !src_copy) { |
b3b4aa74 | 373 | btrfs_release_path(path); |
2a29edc6 | 374 | kfree(dst_copy); |
375 | kfree(src_copy); | |
376 | return -ENOMEM; | |
377 | } | |
e02119d5 CM |
378 | |
379 | read_extent_buffer(eb, src_copy, src_ptr, item_size); | |
380 | ||
381 | dst_ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]); | |
382 | read_extent_buffer(path->nodes[0], dst_copy, dst_ptr, | |
383 | item_size); | |
384 | ret = memcmp(dst_copy, src_copy, item_size); | |
385 | ||
386 | kfree(dst_copy); | |
387 | kfree(src_copy); | |
388 | /* | |
389 | * they have the same contents, just return, this saves | |
390 | * us from cowing blocks in the destination tree and doing | |
391 | * extra writes that may not have been done by a previous | |
392 | * sync | |
393 | */ | |
394 | if (ret == 0) { | |
b3b4aa74 | 395 | btrfs_release_path(path); |
e02119d5 CM |
396 | return 0; |
397 | } | |
398 | ||
4bc4bee4 JB |
399 | /* |
400 | * We need to load the old nbytes into the inode so when we | |
401 | * replay the extents we've logged we get the right nbytes. | |
402 | */ | |
403 | if (inode_item) { | |
404 | struct btrfs_inode_item *item; | |
405 | u64 nbytes; | |
d555438b | 406 | u32 mode; |
4bc4bee4 JB |
407 | |
408 | item = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
409 | struct btrfs_inode_item); | |
410 | nbytes = btrfs_inode_nbytes(path->nodes[0], item); | |
411 | item = btrfs_item_ptr(eb, slot, | |
412 | struct btrfs_inode_item); | |
413 | btrfs_set_inode_nbytes(eb, item, nbytes); | |
d555438b JB |
414 | |
415 | /* | |
416 | * If this is a directory we need to reset the i_size to | |
417 | * 0 so that we can set it up properly when replaying | |
418 | * the rest of the items in this log. | |
419 | */ | |
420 | mode = btrfs_inode_mode(eb, item); | |
421 | if (S_ISDIR(mode)) | |
422 | btrfs_set_inode_size(eb, item, 0); | |
4bc4bee4 JB |
423 | } |
424 | } else if (inode_item) { | |
425 | struct btrfs_inode_item *item; | |
d555438b | 426 | u32 mode; |
4bc4bee4 JB |
427 | |
428 | /* | |
429 | * New inode, set nbytes to 0 so that the nbytes comes out | |
430 | * properly when we replay the extents. | |
431 | */ | |
432 | item = btrfs_item_ptr(eb, slot, struct btrfs_inode_item); | |
433 | btrfs_set_inode_nbytes(eb, item, 0); | |
d555438b JB |
434 | |
435 | /* | |
436 | * If this is a directory we need to reset the i_size to 0 so | |
437 | * that we can set it up properly when replaying the rest of | |
438 | * the items in this log. | |
439 | */ | |
440 | mode = btrfs_inode_mode(eb, item); | |
441 | if (S_ISDIR(mode)) | |
442 | btrfs_set_inode_size(eb, item, 0); | |
e02119d5 CM |
443 | } |
444 | insert: | |
b3b4aa74 | 445 | btrfs_release_path(path); |
e02119d5 | 446 | /* try to insert the key into the destination tree */ |
df8d116f | 447 | path->skip_release_on_error = 1; |
e02119d5 CM |
448 | ret = btrfs_insert_empty_item(trans, root, path, |
449 | key, item_size); | |
df8d116f | 450 | path->skip_release_on_error = 0; |
e02119d5 CM |
451 | |
452 | /* make sure any existing item is the correct size */ | |
df8d116f | 453 | if (ret == -EEXIST || ret == -EOVERFLOW) { |
e02119d5 CM |
454 | u32 found_size; |
455 | found_size = btrfs_item_size_nr(path->nodes[0], | |
456 | path->slots[0]); | |
143bede5 | 457 | if (found_size > item_size) |
afe5fea7 | 458 | btrfs_truncate_item(root, path, item_size, 1); |
143bede5 | 459 | else if (found_size < item_size) |
4b90c680 | 460 | btrfs_extend_item(root, path, |
143bede5 | 461 | item_size - found_size); |
e02119d5 | 462 | } else if (ret) { |
4a500fd1 | 463 | return ret; |
e02119d5 CM |
464 | } |
465 | dst_ptr = btrfs_item_ptr_offset(path->nodes[0], | |
466 | path->slots[0]); | |
467 | ||
468 | /* don't overwrite an existing inode if the generation number | |
469 | * was logged as zero. This is done when the tree logging code | |
470 | * is just logging an inode to make sure it exists after recovery. | |
471 | * | |
472 | * Also, don't overwrite i_size on directories during replay. | |
473 | * log replay inserts and removes directory items based on the | |
474 | * state of the tree found in the subvolume, and i_size is modified | |
475 | * as it goes | |
476 | */ | |
477 | if (key->type == BTRFS_INODE_ITEM_KEY && ret == -EEXIST) { | |
478 | struct btrfs_inode_item *src_item; | |
479 | struct btrfs_inode_item *dst_item; | |
480 | ||
481 | src_item = (struct btrfs_inode_item *)src_ptr; | |
482 | dst_item = (struct btrfs_inode_item *)dst_ptr; | |
483 | ||
1a4bcf47 FM |
484 | if (btrfs_inode_generation(eb, src_item) == 0) { |
485 | struct extent_buffer *dst_eb = path->nodes[0]; | |
2f2ff0ee | 486 | const u64 ino_size = btrfs_inode_size(eb, src_item); |
1a4bcf47 | 487 | |
2f2ff0ee FM |
488 | /* |
489 | * For regular files an ino_size == 0 is used only when | |
490 | * logging that an inode exists, as part of a directory | |
491 | * fsync, and the inode wasn't fsynced before. In this | |
492 | * case don't set the size of the inode in the fs/subvol | |
493 | * tree, otherwise we would be throwing valid data away. | |
494 | */ | |
1a4bcf47 | 495 | if (S_ISREG(btrfs_inode_mode(eb, src_item)) && |
2f2ff0ee FM |
496 | S_ISREG(btrfs_inode_mode(dst_eb, dst_item)) && |
497 | ino_size != 0) { | |
1a4bcf47 | 498 | struct btrfs_map_token token; |
1a4bcf47 FM |
499 | |
500 | btrfs_init_map_token(&token); | |
501 | btrfs_set_token_inode_size(dst_eb, dst_item, | |
502 | ino_size, &token); | |
503 | } | |
e02119d5 | 504 | goto no_copy; |
1a4bcf47 | 505 | } |
e02119d5 CM |
506 | |
507 | if (overwrite_root && | |
508 | S_ISDIR(btrfs_inode_mode(eb, src_item)) && | |
509 | S_ISDIR(btrfs_inode_mode(path->nodes[0], dst_item))) { | |
510 | save_old_i_size = 1; | |
511 | saved_i_size = btrfs_inode_size(path->nodes[0], | |
512 | dst_item); | |
513 | } | |
514 | } | |
515 | ||
516 | copy_extent_buffer(path->nodes[0], eb, dst_ptr, | |
517 | src_ptr, item_size); | |
518 | ||
519 | if (save_old_i_size) { | |
520 | struct btrfs_inode_item *dst_item; | |
521 | dst_item = (struct btrfs_inode_item *)dst_ptr; | |
522 | btrfs_set_inode_size(path->nodes[0], dst_item, saved_i_size); | |
523 | } | |
524 | ||
525 | /* make sure the generation is filled in */ | |
526 | if (key->type == BTRFS_INODE_ITEM_KEY) { | |
527 | struct btrfs_inode_item *dst_item; | |
528 | dst_item = (struct btrfs_inode_item *)dst_ptr; | |
529 | if (btrfs_inode_generation(path->nodes[0], dst_item) == 0) { | |
530 | btrfs_set_inode_generation(path->nodes[0], dst_item, | |
531 | trans->transid); | |
532 | } | |
533 | } | |
534 | no_copy: | |
535 | btrfs_mark_buffer_dirty(path->nodes[0]); | |
b3b4aa74 | 536 | btrfs_release_path(path); |
e02119d5 CM |
537 | return 0; |
538 | } | |
539 | ||
540 | /* | |
541 | * simple helper to read an inode off the disk from a given root | |
542 | * This can only be called for subvolume roots and not for the log | |
543 | */ | |
544 | static noinline struct inode *read_one_inode(struct btrfs_root *root, | |
545 | u64 objectid) | |
546 | { | |
5d4f98a2 | 547 | struct btrfs_key key; |
e02119d5 | 548 | struct inode *inode; |
e02119d5 | 549 | |
5d4f98a2 YZ |
550 | key.objectid = objectid; |
551 | key.type = BTRFS_INODE_ITEM_KEY; | |
552 | key.offset = 0; | |
73f73415 | 553 | inode = btrfs_iget(root->fs_info->sb, &key, root, NULL); |
5d4f98a2 YZ |
554 | if (IS_ERR(inode)) { |
555 | inode = NULL; | |
556 | } else if (is_bad_inode(inode)) { | |
e02119d5 CM |
557 | iput(inode); |
558 | inode = NULL; | |
559 | } | |
560 | return inode; | |
561 | } | |
562 | ||
563 | /* replays a single extent in 'eb' at 'slot' with 'key' into the | |
564 | * subvolume 'root'. path is released on entry and should be released | |
565 | * on exit. | |
566 | * | |
567 | * extents in the log tree have not been allocated out of the extent | |
568 | * tree yet. So, this completes the allocation, taking a reference | |
569 | * as required if the extent already exists or creating a new extent | |
570 | * if it isn't in the extent allocation tree yet. | |
571 | * | |
572 | * The extent is inserted into the file, dropping any existing extents | |
573 | * from the file that overlap the new one. | |
574 | */ | |
575 | static noinline int replay_one_extent(struct btrfs_trans_handle *trans, | |
576 | struct btrfs_root *root, | |
577 | struct btrfs_path *path, | |
578 | struct extent_buffer *eb, int slot, | |
579 | struct btrfs_key *key) | |
580 | { | |
581 | int found_type; | |
e02119d5 | 582 | u64 extent_end; |
e02119d5 | 583 | u64 start = key->offset; |
4bc4bee4 | 584 | u64 nbytes = 0; |
e02119d5 CM |
585 | struct btrfs_file_extent_item *item; |
586 | struct inode *inode = NULL; | |
587 | unsigned long size; | |
588 | int ret = 0; | |
589 | ||
590 | item = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item); | |
591 | found_type = btrfs_file_extent_type(eb, item); | |
592 | ||
d899e052 | 593 | if (found_type == BTRFS_FILE_EXTENT_REG || |
4bc4bee4 JB |
594 | found_type == BTRFS_FILE_EXTENT_PREALLOC) { |
595 | nbytes = btrfs_file_extent_num_bytes(eb, item); | |
596 | extent_end = start + nbytes; | |
597 | ||
598 | /* | |
599 | * We don't add to the inodes nbytes if we are prealloc or a | |
600 | * hole. | |
601 | */ | |
602 | if (btrfs_file_extent_disk_bytenr(eb, item) == 0) | |
603 | nbytes = 0; | |
604 | } else if (found_type == BTRFS_FILE_EXTENT_INLINE) { | |
514ac8ad | 605 | size = btrfs_file_extent_inline_len(eb, slot, item); |
4bc4bee4 | 606 | nbytes = btrfs_file_extent_ram_bytes(eb, item); |
fda2832f | 607 | extent_end = ALIGN(start + size, root->sectorsize); |
e02119d5 CM |
608 | } else { |
609 | ret = 0; | |
610 | goto out; | |
611 | } | |
612 | ||
613 | inode = read_one_inode(root, key->objectid); | |
614 | if (!inode) { | |
615 | ret = -EIO; | |
616 | goto out; | |
617 | } | |
618 | ||
619 | /* | |
620 | * first check to see if we already have this extent in the | |
621 | * file. This must be done before the btrfs_drop_extents run | |
622 | * so we don't try to drop this extent. | |
623 | */ | |
33345d01 | 624 | ret = btrfs_lookup_file_extent(trans, root, path, btrfs_ino(inode), |
e02119d5 CM |
625 | start, 0); |
626 | ||
d899e052 YZ |
627 | if (ret == 0 && |
628 | (found_type == BTRFS_FILE_EXTENT_REG || | |
629 | found_type == BTRFS_FILE_EXTENT_PREALLOC)) { | |
e02119d5 CM |
630 | struct btrfs_file_extent_item cmp1; |
631 | struct btrfs_file_extent_item cmp2; | |
632 | struct btrfs_file_extent_item *existing; | |
633 | struct extent_buffer *leaf; | |
634 | ||
635 | leaf = path->nodes[0]; | |
636 | existing = btrfs_item_ptr(leaf, path->slots[0], | |
637 | struct btrfs_file_extent_item); | |
638 | ||
639 | read_extent_buffer(eb, &cmp1, (unsigned long)item, | |
640 | sizeof(cmp1)); | |
641 | read_extent_buffer(leaf, &cmp2, (unsigned long)existing, | |
642 | sizeof(cmp2)); | |
643 | ||
644 | /* | |
645 | * we already have a pointer to this exact extent, | |
646 | * we don't have to do anything | |
647 | */ | |
648 | if (memcmp(&cmp1, &cmp2, sizeof(cmp1)) == 0) { | |
b3b4aa74 | 649 | btrfs_release_path(path); |
e02119d5 CM |
650 | goto out; |
651 | } | |
652 | } | |
b3b4aa74 | 653 | btrfs_release_path(path); |
e02119d5 CM |
654 | |
655 | /* drop any overlapping extents */ | |
2671485d | 656 | ret = btrfs_drop_extents(trans, root, inode, start, extent_end, 1); |
3650860b JB |
657 | if (ret) |
658 | goto out; | |
e02119d5 | 659 | |
07d400a6 YZ |
660 | if (found_type == BTRFS_FILE_EXTENT_REG || |
661 | found_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
5d4f98a2 | 662 | u64 offset; |
07d400a6 YZ |
663 | unsigned long dest_offset; |
664 | struct btrfs_key ins; | |
665 | ||
666 | ret = btrfs_insert_empty_item(trans, root, path, key, | |
667 | sizeof(*item)); | |
3650860b JB |
668 | if (ret) |
669 | goto out; | |
07d400a6 YZ |
670 | dest_offset = btrfs_item_ptr_offset(path->nodes[0], |
671 | path->slots[0]); | |
672 | copy_extent_buffer(path->nodes[0], eb, dest_offset, | |
673 | (unsigned long)item, sizeof(*item)); | |
674 | ||
675 | ins.objectid = btrfs_file_extent_disk_bytenr(eb, item); | |
676 | ins.offset = btrfs_file_extent_disk_num_bytes(eb, item); | |
677 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
5d4f98a2 | 678 | offset = key->offset - btrfs_file_extent_offset(eb, item); |
07d400a6 YZ |
679 | |
680 | if (ins.objectid > 0) { | |
681 | u64 csum_start; | |
682 | u64 csum_end; | |
683 | LIST_HEAD(ordered_sums); | |
684 | /* | |
685 | * is this extent already allocated in the extent | |
686 | * allocation tree? If so, just add a reference | |
687 | */ | |
1a4ed8fd | 688 | ret = btrfs_lookup_data_extent(root, ins.objectid, |
07d400a6 YZ |
689 | ins.offset); |
690 | if (ret == 0) { | |
691 | ret = btrfs_inc_extent_ref(trans, root, | |
692 | ins.objectid, ins.offset, | |
5d4f98a2 | 693 | 0, root->root_key.objectid, |
66d7e7f0 | 694 | key->objectid, offset, 0); |
b50c6e25 JB |
695 | if (ret) |
696 | goto out; | |
07d400a6 YZ |
697 | } else { |
698 | /* | |
699 | * insert the extent pointer in the extent | |
700 | * allocation tree | |
701 | */ | |
5d4f98a2 YZ |
702 | ret = btrfs_alloc_logged_file_extent(trans, |
703 | root, root->root_key.objectid, | |
704 | key->objectid, offset, &ins); | |
b50c6e25 JB |
705 | if (ret) |
706 | goto out; | |
07d400a6 | 707 | } |
b3b4aa74 | 708 | btrfs_release_path(path); |
07d400a6 YZ |
709 | |
710 | if (btrfs_file_extent_compression(eb, item)) { | |
711 | csum_start = ins.objectid; | |
712 | csum_end = csum_start + ins.offset; | |
713 | } else { | |
714 | csum_start = ins.objectid + | |
715 | btrfs_file_extent_offset(eb, item); | |
716 | csum_end = csum_start + | |
717 | btrfs_file_extent_num_bytes(eb, item); | |
718 | } | |
719 | ||
720 | ret = btrfs_lookup_csums_range(root->log_root, | |
721 | csum_start, csum_end - 1, | |
a2de733c | 722 | &ordered_sums, 0); |
3650860b JB |
723 | if (ret) |
724 | goto out; | |
07d400a6 YZ |
725 | while (!list_empty(&ordered_sums)) { |
726 | struct btrfs_ordered_sum *sums; | |
727 | sums = list_entry(ordered_sums.next, | |
728 | struct btrfs_ordered_sum, | |
729 | list); | |
3650860b JB |
730 | if (!ret) |
731 | ret = btrfs_csum_file_blocks(trans, | |
07d400a6 YZ |
732 | root->fs_info->csum_root, |
733 | sums); | |
07d400a6 YZ |
734 | list_del(&sums->list); |
735 | kfree(sums); | |
736 | } | |
3650860b JB |
737 | if (ret) |
738 | goto out; | |
07d400a6 | 739 | } else { |
b3b4aa74 | 740 | btrfs_release_path(path); |
07d400a6 YZ |
741 | } |
742 | } else if (found_type == BTRFS_FILE_EXTENT_INLINE) { | |
743 | /* inline extents are easy, we just overwrite them */ | |
744 | ret = overwrite_item(trans, root, path, eb, slot, key); | |
3650860b JB |
745 | if (ret) |
746 | goto out; | |
07d400a6 | 747 | } |
e02119d5 | 748 | |
4bc4bee4 | 749 | inode_add_bytes(inode, nbytes); |
b9959295 | 750 | ret = btrfs_update_inode(trans, root, inode); |
e02119d5 CM |
751 | out: |
752 | if (inode) | |
753 | iput(inode); | |
754 | return ret; | |
755 | } | |
756 | ||
757 | /* | |
758 | * when cleaning up conflicts between the directory names in the | |
759 | * subvolume, directory names in the log and directory names in the | |
760 | * inode back references, we may have to unlink inodes from directories. | |
761 | * | |
762 | * This is a helper function to do the unlink of a specific directory | |
763 | * item | |
764 | */ | |
765 | static noinline int drop_one_dir_item(struct btrfs_trans_handle *trans, | |
766 | struct btrfs_root *root, | |
767 | struct btrfs_path *path, | |
768 | struct inode *dir, | |
769 | struct btrfs_dir_item *di) | |
770 | { | |
771 | struct inode *inode; | |
772 | char *name; | |
773 | int name_len; | |
774 | struct extent_buffer *leaf; | |
775 | struct btrfs_key location; | |
776 | int ret; | |
777 | ||
778 | leaf = path->nodes[0]; | |
779 | ||
780 | btrfs_dir_item_key_to_cpu(leaf, di, &location); | |
781 | name_len = btrfs_dir_name_len(leaf, di); | |
782 | name = kmalloc(name_len, GFP_NOFS); | |
2a29edc6 | 783 | if (!name) |
784 | return -ENOMEM; | |
785 | ||
e02119d5 | 786 | read_extent_buffer(leaf, name, (unsigned long)(di + 1), name_len); |
b3b4aa74 | 787 | btrfs_release_path(path); |
e02119d5 CM |
788 | |
789 | inode = read_one_inode(root, location.objectid); | |
c00e9493 | 790 | if (!inode) { |
3650860b JB |
791 | ret = -EIO; |
792 | goto out; | |
c00e9493 | 793 | } |
e02119d5 | 794 | |
ec051c0f | 795 | ret = link_to_fixup_dir(trans, root, path, location.objectid); |
3650860b JB |
796 | if (ret) |
797 | goto out; | |
12fcfd22 | 798 | |
e02119d5 | 799 | ret = btrfs_unlink_inode(trans, root, dir, inode, name, name_len); |
3650860b JB |
800 | if (ret) |
801 | goto out; | |
ada9af21 FDBM |
802 | else |
803 | ret = btrfs_run_delayed_items(trans, root); | |
3650860b | 804 | out: |
e02119d5 | 805 | kfree(name); |
e02119d5 CM |
806 | iput(inode); |
807 | return ret; | |
808 | } | |
809 | ||
810 | /* | |
811 | * helper function to see if a given name and sequence number found | |
812 | * in an inode back reference are already in a directory and correctly | |
813 | * point to this inode | |
814 | */ | |
815 | static noinline int inode_in_dir(struct btrfs_root *root, | |
816 | struct btrfs_path *path, | |
817 | u64 dirid, u64 objectid, u64 index, | |
818 | const char *name, int name_len) | |
819 | { | |
820 | struct btrfs_dir_item *di; | |
821 | struct btrfs_key location; | |
822 | int match = 0; | |
823 | ||
824 | di = btrfs_lookup_dir_index_item(NULL, root, path, dirid, | |
825 | index, name, name_len, 0); | |
826 | if (di && !IS_ERR(di)) { | |
827 | btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location); | |
828 | if (location.objectid != objectid) | |
829 | goto out; | |
830 | } else | |
831 | goto out; | |
b3b4aa74 | 832 | btrfs_release_path(path); |
e02119d5 CM |
833 | |
834 | di = btrfs_lookup_dir_item(NULL, root, path, dirid, name, name_len, 0); | |
835 | if (di && !IS_ERR(di)) { | |
836 | btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location); | |
837 | if (location.objectid != objectid) | |
838 | goto out; | |
839 | } else | |
840 | goto out; | |
841 | match = 1; | |
842 | out: | |
b3b4aa74 | 843 | btrfs_release_path(path); |
e02119d5 CM |
844 | return match; |
845 | } | |
846 | ||
847 | /* | |
848 | * helper function to check a log tree for a named back reference in | |
849 | * an inode. This is used to decide if a back reference that is | |
850 | * found in the subvolume conflicts with what we find in the log. | |
851 | * | |
852 | * inode backreferences may have multiple refs in a single item, | |
853 | * during replay we process one reference at a time, and we don't | |
854 | * want to delete valid links to a file from the subvolume if that | |
855 | * link is also in the log. | |
856 | */ | |
857 | static noinline int backref_in_log(struct btrfs_root *log, | |
858 | struct btrfs_key *key, | |
f186373f | 859 | u64 ref_objectid, |
df8d116f | 860 | const char *name, int namelen) |
e02119d5 CM |
861 | { |
862 | struct btrfs_path *path; | |
863 | struct btrfs_inode_ref *ref; | |
864 | unsigned long ptr; | |
865 | unsigned long ptr_end; | |
866 | unsigned long name_ptr; | |
867 | int found_name_len; | |
868 | int item_size; | |
869 | int ret; | |
870 | int match = 0; | |
871 | ||
872 | path = btrfs_alloc_path(); | |
2a29edc6 | 873 | if (!path) |
874 | return -ENOMEM; | |
875 | ||
e02119d5 CM |
876 | ret = btrfs_search_slot(NULL, log, key, path, 0, 0); |
877 | if (ret != 0) | |
878 | goto out; | |
879 | ||
e02119d5 | 880 | ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]); |
f186373f MF |
881 | |
882 | if (key->type == BTRFS_INODE_EXTREF_KEY) { | |
883 | if (btrfs_find_name_in_ext_backref(path, ref_objectid, | |
884 | name, namelen, NULL)) | |
885 | match = 1; | |
886 | ||
887 | goto out; | |
888 | } | |
889 | ||
890 | item_size = btrfs_item_size_nr(path->nodes[0], path->slots[0]); | |
e02119d5 CM |
891 | ptr_end = ptr + item_size; |
892 | while (ptr < ptr_end) { | |
893 | ref = (struct btrfs_inode_ref *)ptr; | |
894 | found_name_len = btrfs_inode_ref_name_len(path->nodes[0], ref); | |
895 | if (found_name_len == namelen) { | |
896 | name_ptr = (unsigned long)(ref + 1); | |
897 | ret = memcmp_extent_buffer(path->nodes[0], name, | |
898 | name_ptr, namelen); | |
899 | if (ret == 0) { | |
900 | match = 1; | |
901 | goto out; | |
902 | } | |
903 | } | |
904 | ptr = (unsigned long)(ref + 1) + found_name_len; | |
905 | } | |
906 | out: | |
907 | btrfs_free_path(path); | |
908 | return match; | |
909 | } | |
910 | ||
5a1d7843 | 911 | static inline int __add_inode_ref(struct btrfs_trans_handle *trans, |
e02119d5 | 912 | struct btrfs_root *root, |
e02119d5 | 913 | struct btrfs_path *path, |
5a1d7843 JS |
914 | struct btrfs_root *log_root, |
915 | struct inode *dir, struct inode *inode, | |
5a1d7843 | 916 | struct extent_buffer *eb, |
f186373f MF |
917 | u64 inode_objectid, u64 parent_objectid, |
918 | u64 ref_index, char *name, int namelen, | |
919 | int *search_done) | |
e02119d5 | 920 | { |
34f3e4f2 | 921 | int ret; |
f186373f MF |
922 | char *victim_name; |
923 | int victim_name_len; | |
924 | struct extent_buffer *leaf; | |
5a1d7843 | 925 | struct btrfs_dir_item *di; |
f186373f MF |
926 | struct btrfs_key search_key; |
927 | struct btrfs_inode_extref *extref; | |
c622ae60 | 928 | |
f186373f MF |
929 | again: |
930 | /* Search old style refs */ | |
931 | search_key.objectid = inode_objectid; | |
932 | search_key.type = BTRFS_INODE_REF_KEY; | |
933 | search_key.offset = parent_objectid; | |
934 | ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0); | |
e02119d5 | 935 | if (ret == 0) { |
e02119d5 CM |
936 | struct btrfs_inode_ref *victim_ref; |
937 | unsigned long ptr; | |
938 | unsigned long ptr_end; | |
f186373f MF |
939 | |
940 | leaf = path->nodes[0]; | |
e02119d5 CM |
941 | |
942 | /* are we trying to overwrite a back ref for the root directory | |
943 | * if so, just jump out, we're done | |
944 | */ | |
f186373f | 945 | if (search_key.objectid == search_key.offset) |
5a1d7843 | 946 | return 1; |
e02119d5 CM |
947 | |
948 | /* check all the names in this back reference to see | |
949 | * if they are in the log. if so, we allow them to stay | |
950 | * otherwise they must be unlinked as a conflict | |
951 | */ | |
952 | ptr = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
953 | ptr_end = ptr + btrfs_item_size_nr(leaf, path->slots[0]); | |
d397712b | 954 | while (ptr < ptr_end) { |
e02119d5 CM |
955 | victim_ref = (struct btrfs_inode_ref *)ptr; |
956 | victim_name_len = btrfs_inode_ref_name_len(leaf, | |
957 | victim_ref); | |
958 | victim_name = kmalloc(victim_name_len, GFP_NOFS); | |
3650860b JB |
959 | if (!victim_name) |
960 | return -ENOMEM; | |
e02119d5 CM |
961 | |
962 | read_extent_buffer(leaf, victim_name, | |
963 | (unsigned long)(victim_ref + 1), | |
964 | victim_name_len); | |
965 | ||
f186373f MF |
966 | if (!backref_in_log(log_root, &search_key, |
967 | parent_objectid, | |
968 | victim_name, | |
e02119d5 | 969 | victim_name_len)) { |
8b558c5f | 970 | inc_nlink(inode); |
b3b4aa74 | 971 | btrfs_release_path(path); |
12fcfd22 | 972 | |
e02119d5 CM |
973 | ret = btrfs_unlink_inode(trans, root, dir, |
974 | inode, victim_name, | |
975 | victim_name_len); | |
f186373f | 976 | kfree(victim_name); |
3650860b JB |
977 | if (ret) |
978 | return ret; | |
ada9af21 FDBM |
979 | ret = btrfs_run_delayed_items(trans, root); |
980 | if (ret) | |
981 | return ret; | |
f186373f MF |
982 | *search_done = 1; |
983 | goto again; | |
e02119d5 CM |
984 | } |
985 | kfree(victim_name); | |
f186373f | 986 | |
e02119d5 CM |
987 | ptr = (unsigned long)(victim_ref + 1) + victim_name_len; |
988 | } | |
e02119d5 | 989 | |
c622ae60 | 990 | /* |
991 | * NOTE: we have searched root tree and checked the | |
992 | * coresponding ref, it does not need to check again. | |
993 | */ | |
5a1d7843 | 994 | *search_done = 1; |
e02119d5 | 995 | } |
b3b4aa74 | 996 | btrfs_release_path(path); |
e02119d5 | 997 | |
f186373f MF |
998 | /* Same search but for extended refs */ |
999 | extref = btrfs_lookup_inode_extref(NULL, root, path, name, namelen, | |
1000 | inode_objectid, parent_objectid, 0, | |
1001 | 0); | |
1002 | if (!IS_ERR_OR_NULL(extref)) { | |
1003 | u32 item_size; | |
1004 | u32 cur_offset = 0; | |
1005 | unsigned long base; | |
1006 | struct inode *victim_parent; | |
1007 | ||
1008 | leaf = path->nodes[0]; | |
1009 | ||
1010 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1011 | base = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
1012 | ||
1013 | while (cur_offset < item_size) { | |
dd9ef135 | 1014 | extref = (struct btrfs_inode_extref *)(base + cur_offset); |
f186373f MF |
1015 | |
1016 | victim_name_len = btrfs_inode_extref_name_len(leaf, extref); | |
1017 | ||
1018 | if (btrfs_inode_extref_parent(leaf, extref) != parent_objectid) | |
1019 | goto next; | |
1020 | ||
1021 | victim_name = kmalloc(victim_name_len, GFP_NOFS); | |
3650860b JB |
1022 | if (!victim_name) |
1023 | return -ENOMEM; | |
f186373f MF |
1024 | read_extent_buffer(leaf, victim_name, (unsigned long)&extref->name, |
1025 | victim_name_len); | |
1026 | ||
1027 | search_key.objectid = inode_objectid; | |
1028 | search_key.type = BTRFS_INODE_EXTREF_KEY; | |
1029 | search_key.offset = btrfs_extref_hash(parent_objectid, | |
1030 | victim_name, | |
1031 | victim_name_len); | |
1032 | ret = 0; | |
1033 | if (!backref_in_log(log_root, &search_key, | |
1034 | parent_objectid, victim_name, | |
1035 | victim_name_len)) { | |
1036 | ret = -ENOENT; | |
1037 | victim_parent = read_one_inode(root, | |
1038 | parent_objectid); | |
1039 | if (victim_parent) { | |
8b558c5f | 1040 | inc_nlink(inode); |
f186373f MF |
1041 | btrfs_release_path(path); |
1042 | ||
1043 | ret = btrfs_unlink_inode(trans, root, | |
1044 | victim_parent, | |
1045 | inode, | |
1046 | victim_name, | |
1047 | victim_name_len); | |
ada9af21 FDBM |
1048 | if (!ret) |
1049 | ret = btrfs_run_delayed_items( | |
1050 | trans, root); | |
f186373f | 1051 | } |
f186373f MF |
1052 | iput(victim_parent); |
1053 | kfree(victim_name); | |
3650860b JB |
1054 | if (ret) |
1055 | return ret; | |
f186373f MF |
1056 | *search_done = 1; |
1057 | goto again; | |
1058 | } | |
1059 | kfree(victim_name); | |
3650860b JB |
1060 | if (ret) |
1061 | return ret; | |
f186373f MF |
1062 | next: |
1063 | cur_offset += victim_name_len + sizeof(*extref); | |
1064 | } | |
1065 | *search_done = 1; | |
1066 | } | |
1067 | btrfs_release_path(path); | |
1068 | ||
34f3e4f2 | 1069 | /* look for a conflicting sequence number */ |
1070 | di = btrfs_lookup_dir_index_item(trans, root, path, btrfs_ino(dir), | |
f186373f | 1071 | ref_index, name, namelen, 0); |
34f3e4f2 | 1072 | if (di && !IS_ERR(di)) { |
1073 | ret = drop_one_dir_item(trans, root, path, dir, di); | |
3650860b JB |
1074 | if (ret) |
1075 | return ret; | |
34f3e4f2 | 1076 | } |
1077 | btrfs_release_path(path); | |
1078 | ||
1079 | /* look for a conflicing name */ | |
1080 | di = btrfs_lookup_dir_item(trans, root, path, btrfs_ino(dir), | |
1081 | name, namelen, 0); | |
1082 | if (di && !IS_ERR(di)) { | |
1083 | ret = drop_one_dir_item(trans, root, path, dir, di); | |
3650860b JB |
1084 | if (ret) |
1085 | return ret; | |
34f3e4f2 | 1086 | } |
1087 | btrfs_release_path(path); | |
1088 | ||
5a1d7843 JS |
1089 | return 0; |
1090 | } | |
e02119d5 | 1091 | |
f186373f MF |
1092 | static int extref_get_fields(struct extent_buffer *eb, unsigned long ref_ptr, |
1093 | u32 *namelen, char **name, u64 *index, | |
1094 | u64 *parent_objectid) | |
1095 | { | |
1096 | struct btrfs_inode_extref *extref; | |
1097 | ||
1098 | extref = (struct btrfs_inode_extref *)ref_ptr; | |
1099 | ||
1100 | *namelen = btrfs_inode_extref_name_len(eb, extref); | |
1101 | *name = kmalloc(*namelen, GFP_NOFS); | |
1102 | if (*name == NULL) | |
1103 | return -ENOMEM; | |
1104 | ||
1105 | read_extent_buffer(eb, *name, (unsigned long)&extref->name, | |
1106 | *namelen); | |
1107 | ||
1108 | *index = btrfs_inode_extref_index(eb, extref); | |
1109 | if (parent_objectid) | |
1110 | *parent_objectid = btrfs_inode_extref_parent(eb, extref); | |
1111 | ||
1112 | return 0; | |
1113 | } | |
1114 | ||
1115 | static int ref_get_fields(struct extent_buffer *eb, unsigned long ref_ptr, | |
1116 | u32 *namelen, char **name, u64 *index) | |
1117 | { | |
1118 | struct btrfs_inode_ref *ref; | |
1119 | ||
1120 | ref = (struct btrfs_inode_ref *)ref_ptr; | |
1121 | ||
1122 | *namelen = btrfs_inode_ref_name_len(eb, ref); | |
1123 | *name = kmalloc(*namelen, GFP_NOFS); | |
1124 | if (*name == NULL) | |
1125 | return -ENOMEM; | |
1126 | ||
1127 | read_extent_buffer(eb, *name, (unsigned long)(ref + 1), *namelen); | |
1128 | ||
1129 | *index = btrfs_inode_ref_index(eb, ref); | |
1130 | ||
1131 | return 0; | |
1132 | } | |
1133 | ||
5a1d7843 JS |
1134 | /* |
1135 | * replay one inode back reference item found in the log tree. | |
1136 | * eb, slot and key refer to the buffer and key found in the log tree. | |
1137 | * root is the destination we are replaying into, and path is for temp | |
1138 | * use by this function. (it should be released on return). | |
1139 | */ | |
1140 | static noinline int add_inode_ref(struct btrfs_trans_handle *trans, | |
1141 | struct btrfs_root *root, | |
1142 | struct btrfs_root *log, | |
1143 | struct btrfs_path *path, | |
1144 | struct extent_buffer *eb, int slot, | |
1145 | struct btrfs_key *key) | |
1146 | { | |
03b2f08b GB |
1147 | struct inode *dir = NULL; |
1148 | struct inode *inode = NULL; | |
5a1d7843 JS |
1149 | unsigned long ref_ptr; |
1150 | unsigned long ref_end; | |
03b2f08b | 1151 | char *name = NULL; |
5a1d7843 JS |
1152 | int namelen; |
1153 | int ret; | |
1154 | int search_done = 0; | |
f186373f MF |
1155 | int log_ref_ver = 0; |
1156 | u64 parent_objectid; | |
1157 | u64 inode_objectid; | |
f46dbe3d | 1158 | u64 ref_index = 0; |
f186373f MF |
1159 | int ref_struct_size; |
1160 | ||
1161 | ref_ptr = btrfs_item_ptr_offset(eb, slot); | |
1162 | ref_end = ref_ptr + btrfs_item_size_nr(eb, slot); | |
1163 | ||
1164 | if (key->type == BTRFS_INODE_EXTREF_KEY) { | |
1165 | struct btrfs_inode_extref *r; | |
1166 | ||
1167 | ref_struct_size = sizeof(struct btrfs_inode_extref); | |
1168 | log_ref_ver = 1; | |
1169 | r = (struct btrfs_inode_extref *)ref_ptr; | |
1170 | parent_objectid = btrfs_inode_extref_parent(eb, r); | |
1171 | } else { | |
1172 | ref_struct_size = sizeof(struct btrfs_inode_ref); | |
1173 | parent_objectid = key->offset; | |
1174 | } | |
1175 | inode_objectid = key->objectid; | |
e02119d5 | 1176 | |
5a1d7843 JS |
1177 | /* |
1178 | * it is possible that we didn't log all the parent directories | |
1179 | * for a given inode. If we don't find the dir, just don't | |
1180 | * copy the back ref in. The link count fixup code will take | |
1181 | * care of the rest | |
1182 | */ | |
f186373f | 1183 | dir = read_one_inode(root, parent_objectid); |
03b2f08b GB |
1184 | if (!dir) { |
1185 | ret = -ENOENT; | |
1186 | goto out; | |
1187 | } | |
5a1d7843 | 1188 | |
f186373f | 1189 | inode = read_one_inode(root, inode_objectid); |
5a1d7843 | 1190 | if (!inode) { |
03b2f08b GB |
1191 | ret = -EIO; |
1192 | goto out; | |
5a1d7843 JS |
1193 | } |
1194 | ||
5a1d7843 | 1195 | while (ref_ptr < ref_end) { |
f186373f MF |
1196 | if (log_ref_ver) { |
1197 | ret = extref_get_fields(eb, ref_ptr, &namelen, &name, | |
1198 | &ref_index, &parent_objectid); | |
1199 | /* | |
1200 | * parent object can change from one array | |
1201 | * item to another. | |
1202 | */ | |
1203 | if (!dir) | |
1204 | dir = read_one_inode(root, parent_objectid); | |
03b2f08b GB |
1205 | if (!dir) { |
1206 | ret = -ENOENT; | |
1207 | goto out; | |
1208 | } | |
f186373f MF |
1209 | } else { |
1210 | ret = ref_get_fields(eb, ref_ptr, &namelen, &name, | |
1211 | &ref_index); | |
1212 | } | |
1213 | if (ret) | |
03b2f08b | 1214 | goto out; |
5a1d7843 JS |
1215 | |
1216 | /* if we already have a perfect match, we're done */ | |
1217 | if (!inode_in_dir(root, path, btrfs_ino(dir), btrfs_ino(inode), | |
f186373f | 1218 | ref_index, name, namelen)) { |
5a1d7843 JS |
1219 | /* |
1220 | * look for a conflicting back reference in the | |
1221 | * metadata. if we find one we have to unlink that name | |
1222 | * of the file before we add our new link. Later on, we | |
1223 | * overwrite any existing back reference, and we don't | |
1224 | * want to create dangling pointers in the directory. | |
1225 | */ | |
1226 | ||
1227 | if (!search_done) { | |
1228 | ret = __add_inode_ref(trans, root, path, log, | |
f186373f MF |
1229 | dir, inode, eb, |
1230 | inode_objectid, | |
1231 | parent_objectid, | |
1232 | ref_index, name, namelen, | |
5a1d7843 | 1233 | &search_done); |
03b2f08b GB |
1234 | if (ret) { |
1235 | if (ret == 1) | |
1236 | ret = 0; | |
3650860b JB |
1237 | goto out; |
1238 | } | |
5a1d7843 JS |
1239 | } |
1240 | ||
1241 | /* insert our name */ | |
1242 | ret = btrfs_add_link(trans, dir, inode, name, namelen, | |
f186373f | 1243 | 0, ref_index); |
3650860b JB |
1244 | if (ret) |
1245 | goto out; | |
5a1d7843 JS |
1246 | |
1247 | btrfs_update_inode(trans, root, inode); | |
1248 | } | |
1249 | ||
f186373f | 1250 | ref_ptr = (unsigned long)(ref_ptr + ref_struct_size) + namelen; |
5a1d7843 | 1251 | kfree(name); |
03b2f08b | 1252 | name = NULL; |
f186373f MF |
1253 | if (log_ref_ver) { |
1254 | iput(dir); | |
1255 | dir = NULL; | |
1256 | } | |
5a1d7843 | 1257 | } |
e02119d5 CM |
1258 | |
1259 | /* finally write the back reference in the inode */ | |
1260 | ret = overwrite_item(trans, root, path, eb, slot, key); | |
5a1d7843 | 1261 | out: |
b3b4aa74 | 1262 | btrfs_release_path(path); |
03b2f08b | 1263 | kfree(name); |
e02119d5 CM |
1264 | iput(dir); |
1265 | iput(inode); | |
3650860b | 1266 | return ret; |
e02119d5 CM |
1267 | } |
1268 | ||
c71bf099 | 1269 | static int insert_orphan_item(struct btrfs_trans_handle *trans, |
9c4f61f0 | 1270 | struct btrfs_root *root, u64 ino) |
c71bf099 YZ |
1271 | { |
1272 | int ret; | |
381cf658 | 1273 | |
9c4f61f0 DS |
1274 | ret = btrfs_insert_orphan_item(trans, root, ino); |
1275 | if (ret == -EEXIST) | |
1276 | ret = 0; | |
381cf658 | 1277 | |
c71bf099 YZ |
1278 | return ret; |
1279 | } | |
1280 | ||
f186373f MF |
1281 | static int count_inode_extrefs(struct btrfs_root *root, |
1282 | struct inode *inode, struct btrfs_path *path) | |
1283 | { | |
1284 | int ret = 0; | |
1285 | int name_len; | |
1286 | unsigned int nlink = 0; | |
1287 | u32 item_size; | |
1288 | u32 cur_offset = 0; | |
1289 | u64 inode_objectid = btrfs_ino(inode); | |
1290 | u64 offset = 0; | |
1291 | unsigned long ptr; | |
1292 | struct btrfs_inode_extref *extref; | |
1293 | struct extent_buffer *leaf; | |
1294 | ||
1295 | while (1) { | |
1296 | ret = btrfs_find_one_extref(root, inode_objectid, offset, path, | |
1297 | &extref, &offset); | |
1298 | if (ret) | |
1299 | break; | |
c71bf099 | 1300 | |
f186373f MF |
1301 | leaf = path->nodes[0]; |
1302 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1303 | ptr = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
2c2c452b | 1304 | cur_offset = 0; |
f186373f MF |
1305 | |
1306 | while (cur_offset < item_size) { | |
1307 | extref = (struct btrfs_inode_extref *) (ptr + cur_offset); | |
1308 | name_len = btrfs_inode_extref_name_len(leaf, extref); | |
1309 | ||
1310 | nlink++; | |
1311 | ||
1312 | cur_offset += name_len + sizeof(*extref); | |
1313 | } | |
1314 | ||
1315 | offset++; | |
1316 | btrfs_release_path(path); | |
1317 | } | |
1318 | btrfs_release_path(path); | |
1319 | ||
2c2c452b | 1320 | if (ret < 0 && ret != -ENOENT) |
f186373f MF |
1321 | return ret; |
1322 | return nlink; | |
1323 | } | |
1324 | ||
1325 | static int count_inode_refs(struct btrfs_root *root, | |
1326 | struct inode *inode, struct btrfs_path *path) | |
e02119d5 | 1327 | { |
e02119d5 CM |
1328 | int ret; |
1329 | struct btrfs_key key; | |
f186373f | 1330 | unsigned int nlink = 0; |
e02119d5 CM |
1331 | unsigned long ptr; |
1332 | unsigned long ptr_end; | |
1333 | int name_len; | |
33345d01 | 1334 | u64 ino = btrfs_ino(inode); |
e02119d5 | 1335 | |
33345d01 | 1336 | key.objectid = ino; |
e02119d5 CM |
1337 | key.type = BTRFS_INODE_REF_KEY; |
1338 | key.offset = (u64)-1; | |
1339 | ||
d397712b | 1340 | while (1) { |
e02119d5 CM |
1341 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
1342 | if (ret < 0) | |
1343 | break; | |
1344 | if (ret > 0) { | |
1345 | if (path->slots[0] == 0) | |
1346 | break; | |
1347 | path->slots[0]--; | |
1348 | } | |
e93ae26f | 1349 | process_slot: |
e02119d5 CM |
1350 | btrfs_item_key_to_cpu(path->nodes[0], &key, |
1351 | path->slots[0]); | |
33345d01 | 1352 | if (key.objectid != ino || |
e02119d5 CM |
1353 | key.type != BTRFS_INODE_REF_KEY) |
1354 | break; | |
1355 | ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]); | |
1356 | ptr_end = ptr + btrfs_item_size_nr(path->nodes[0], | |
1357 | path->slots[0]); | |
d397712b | 1358 | while (ptr < ptr_end) { |
e02119d5 CM |
1359 | struct btrfs_inode_ref *ref; |
1360 | ||
1361 | ref = (struct btrfs_inode_ref *)ptr; | |
1362 | name_len = btrfs_inode_ref_name_len(path->nodes[0], | |
1363 | ref); | |
1364 | ptr = (unsigned long)(ref + 1) + name_len; | |
1365 | nlink++; | |
1366 | } | |
1367 | ||
1368 | if (key.offset == 0) | |
1369 | break; | |
e93ae26f FDBM |
1370 | if (path->slots[0] > 0) { |
1371 | path->slots[0]--; | |
1372 | goto process_slot; | |
1373 | } | |
e02119d5 | 1374 | key.offset--; |
b3b4aa74 | 1375 | btrfs_release_path(path); |
e02119d5 | 1376 | } |
b3b4aa74 | 1377 | btrfs_release_path(path); |
f186373f MF |
1378 | |
1379 | return nlink; | |
1380 | } | |
1381 | ||
1382 | /* | |
1383 | * There are a few corners where the link count of the file can't | |
1384 | * be properly maintained during replay. So, instead of adding | |
1385 | * lots of complexity to the log code, we just scan the backrefs | |
1386 | * for any file that has been through replay. | |
1387 | * | |
1388 | * The scan will update the link count on the inode to reflect the | |
1389 | * number of back refs found. If it goes down to zero, the iput | |
1390 | * will free the inode. | |
1391 | */ | |
1392 | static noinline int fixup_inode_link_count(struct btrfs_trans_handle *trans, | |
1393 | struct btrfs_root *root, | |
1394 | struct inode *inode) | |
1395 | { | |
1396 | struct btrfs_path *path; | |
1397 | int ret; | |
1398 | u64 nlink = 0; | |
1399 | u64 ino = btrfs_ino(inode); | |
1400 | ||
1401 | path = btrfs_alloc_path(); | |
1402 | if (!path) | |
1403 | return -ENOMEM; | |
1404 | ||
1405 | ret = count_inode_refs(root, inode, path); | |
1406 | if (ret < 0) | |
1407 | goto out; | |
1408 | ||
1409 | nlink = ret; | |
1410 | ||
1411 | ret = count_inode_extrefs(root, inode, path); | |
f186373f MF |
1412 | if (ret < 0) |
1413 | goto out; | |
1414 | ||
1415 | nlink += ret; | |
1416 | ||
1417 | ret = 0; | |
1418 | ||
e02119d5 | 1419 | if (nlink != inode->i_nlink) { |
bfe86848 | 1420 | set_nlink(inode, nlink); |
e02119d5 CM |
1421 | btrfs_update_inode(trans, root, inode); |
1422 | } | |
8d5bf1cb | 1423 | BTRFS_I(inode)->index_cnt = (u64)-1; |
e02119d5 | 1424 | |
c71bf099 YZ |
1425 | if (inode->i_nlink == 0) { |
1426 | if (S_ISDIR(inode->i_mode)) { | |
1427 | ret = replay_dir_deletes(trans, root, NULL, path, | |
33345d01 | 1428 | ino, 1); |
3650860b JB |
1429 | if (ret) |
1430 | goto out; | |
c71bf099 | 1431 | } |
33345d01 | 1432 | ret = insert_orphan_item(trans, root, ino); |
12fcfd22 | 1433 | } |
12fcfd22 | 1434 | |
f186373f MF |
1435 | out: |
1436 | btrfs_free_path(path); | |
1437 | return ret; | |
e02119d5 CM |
1438 | } |
1439 | ||
1440 | static noinline int fixup_inode_link_counts(struct btrfs_trans_handle *trans, | |
1441 | struct btrfs_root *root, | |
1442 | struct btrfs_path *path) | |
1443 | { | |
1444 | int ret; | |
1445 | struct btrfs_key key; | |
1446 | struct inode *inode; | |
1447 | ||
1448 | key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID; | |
1449 | key.type = BTRFS_ORPHAN_ITEM_KEY; | |
1450 | key.offset = (u64)-1; | |
d397712b | 1451 | while (1) { |
e02119d5 CM |
1452 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
1453 | if (ret < 0) | |
1454 | break; | |
1455 | ||
1456 | if (ret == 1) { | |
1457 | if (path->slots[0] == 0) | |
1458 | break; | |
1459 | path->slots[0]--; | |
1460 | } | |
1461 | ||
1462 | btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]); | |
1463 | if (key.objectid != BTRFS_TREE_LOG_FIXUP_OBJECTID || | |
1464 | key.type != BTRFS_ORPHAN_ITEM_KEY) | |
1465 | break; | |
1466 | ||
1467 | ret = btrfs_del_item(trans, root, path); | |
65a246c5 TI |
1468 | if (ret) |
1469 | goto out; | |
e02119d5 | 1470 | |
b3b4aa74 | 1471 | btrfs_release_path(path); |
e02119d5 | 1472 | inode = read_one_inode(root, key.offset); |
c00e9493 TI |
1473 | if (!inode) |
1474 | return -EIO; | |
e02119d5 CM |
1475 | |
1476 | ret = fixup_inode_link_count(trans, root, inode); | |
e02119d5 | 1477 | iput(inode); |
3650860b JB |
1478 | if (ret) |
1479 | goto out; | |
e02119d5 | 1480 | |
12fcfd22 CM |
1481 | /* |
1482 | * fixup on a directory may create new entries, | |
1483 | * make sure we always look for the highset possible | |
1484 | * offset | |
1485 | */ | |
1486 | key.offset = (u64)-1; | |
e02119d5 | 1487 | } |
65a246c5 TI |
1488 | ret = 0; |
1489 | out: | |
b3b4aa74 | 1490 | btrfs_release_path(path); |
65a246c5 | 1491 | return ret; |
e02119d5 CM |
1492 | } |
1493 | ||
1494 | ||
1495 | /* | |
1496 | * record a given inode in the fixup dir so we can check its link | |
1497 | * count when replay is done. The link count is incremented here | |
1498 | * so the inode won't go away until we check it | |
1499 | */ | |
1500 | static noinline int link_to_fixup_dir(struct btrfs_trans_handle *trans, | |
1501 | struct btrfs_root *root, | |
1502 | struct btrfs_path *path, | |
1503 | u64 objectid) | |
1504 | { | |
1505 | struct btrfs_key key; | |
1506 | int ret = 0; | |
1507 | struct inode *inode; | |
1508 | ||
1509 | inode = read_one_inode(root, objectid); | |
c00e9493 TI |
1510 | if (!inode) |
1511 | return -EIO; | |
e02119d5 CM |
1512 | |
1513 | key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID; | |
962a298f | 1514 | key.type = BTRFS_ORPHAN_ITEM_KEY; |
e02119d5 CM |
1515 | key.offset = objectid; |
1516 | ||
1517 | ret = btrfs_insert_empty_item(trans, root, path, &key, 0); | |
1518 | ||
b3b4aa74 | 1519 | btrfs_release_path(path); |
e02119d5 | 1520 | if (ret == 0) { |
9bf7a489 JB |
1521 | if (!inode->i_nlink) |
1522 | set_nlink(inode, 1); | |
1523 | else | |
8b558c5f | 1524 | inc_nlink(inode); |
b9959295 | 1525 | ret = btrfs_update_inode(trans, root, inode); |
e02119d5 CM |
1526 | } else if (ret == -EEXIST) { |
1527 | ret = 0; | |
1528 | } else { | |
3650860b | 1529 | BUG(); /* Logic Error */ |
e02119d5 CM |
1530 | } |
1531 | iput(inode); | |
1532 | ||
1533 | return ret; | |
1534 | } | |
1535 | ||
1536 | /* | |
1537 | * when replaying the log for a directory, we only insert names | |
1538 | * for inodes that actually exist. This means an fsync on a directory | |
1539 | * does not implicitly fsync all the new files in it | |
1540 | */ | |
1541 | static noinline int insert_one_name(struct btrfs_trans_handle *trans, | |
1542 | struct btrfs_root *root, | |
1543 | struct btrfs_path *path, | |
1544 | u64 dirid, u64 index, | |
1545 | char *name, int name_len, u8 type, | |
1546 | struct btrfs_key *location) | |
1547 | { | |
1548 | struct inode *inode; | |
1549 | struct inode *dir; | |
1550 | int ret; | |
1551 | ||
1552 | inode = read_one_inode(root, location->objectid); | |
1553 | if (!inode) | |
1554 | return -ENOENT; | |
1555 | ||
1556 | dir = read_one_inode(root, dirid); | |
1557 | if (!dir) { | |
1558 | iput(inode); | |
1559 | return -EIO; | |
1560 | } | |
d555438b | 1561 | |
e02119d5 CM |
1562 | ret = btrfs_add_link(trans, dir, inode, name, name_len, 1, index); |
1563 | ||
1564 | /* FIXME, put inode into FIXUP list */ | |
1565 | ||
1566 | iput(inode); | |
1567 | iput(dir); | |
1568 | return ret; | |
1569 | } | |
1570 | ||
df8d116f FM |
1571 | /* |
1572 | * Return true if an inode reference exists in the log for the given name, | |
1573 | * inode and parent inode. | |
1574 | */ | |
1575 | static bool name_in_log_ref(struct btrfs_root *log_root, | |
1576 | const char *name, const int name_len, | |
1577 | const u64 dirid, const u64 ino) | |
1578 | { | |
1579 | struct btrfs_key search_key; | |
1580 | ||
1581 | search_key.objectid = ino; | |
1582 | search_key.type = BTRFS_INODE_REF_KEY; | |
1583 | search_key.offset = dirid; | |
1584 | if (backref_in_log(log_root, &search_key, dirid, name, name_len)) | |
1585 | return true; | |
1586 | ||
1587 | search_key.type = BTRFS_INODE_EXTREF_KEY; | |
1588 | search_key.offset = btrfs_extref_hash(dirid, name, name_len); | |
1589 | if (backref_in_log(log_root, &search_key, dirid, name, name_len)) | |
1590 | return true; | |
1591 | ||
1592 | return false; | |
1593 | } | |
1594 | ||
e02119d5 CM |
1595 | /* |
1596 | * take a single entry in a log directory item and replay it into | |
1597 | * the subvolume. | |
1598 | * | |
1599 | * if a conflicting item exists in the subdirectory already, | |
1600 | * the inode it points to is unlinked and put into the link count | |
1601 | * fix up tree. | |
1602 | * | |
1603 | * If a name from the log points to a file or directory that does | |
1604 | * not exist in the FS, it is skipped. fsyncs on directories | |
1605 | * do not force down inodes inside that directory, just changes to the | |
1606 | * names or unlinks in a directory. | |
bb53eda9 FM |
1607 | * |
1608 | * Returns < 0 on error, 0 if the name wasn't replayed (dentry points to a | |
1609 | * non-existing inode) and 1 if the name was replayed. | |
e02119d5 CM |
1610 | */ |
1611 | static noinline int replay_one_name(struct btrfs_trans_handle *trans, | |
1612 | struct btrfs_root *root, | |
1613 | struct btrfs_path *path, | |
1614 | struct extent_buffer *eb, | |
1615 | struct btrfs_dir_item *di, | |
1616 | struct btrfs_key *key) | |
1617 | { | |
1618 | char *name; | |
1619 | int name_len; | |
1620 | struct btrfs_dir_item *dst_di; | |
1621 | struct btrfs_key found_key; | |
1622 | struct btrfs_key log_key; | |
1623 | struct inode *dir; | |
e02119d5 | 1624 | u8 log_type; |
4bef0848 | 1625 | int exists; |
3650860b | 1626 | int ret = 0; |
d555438b | 1627 | bool update_size = (key->type == BTRFS_DIR_INDEX_KEY); |
bb53eda9 | 1628 | bool name_added = false; |
e02119d5 CM |
1629 | |
1630 | dir = read_one_inode(root, key->objectid); | |
c00e9493 TI |
1631 | if (!dir) |
1632 | return -EIO; | |
e02119d5 CM |
1633 | |
1634 | name_len = btrfs_dir_name_len(eb, di); | |
1635 | name = kmalloc(name_len, GFP_NOFS); | |
2bac325e FDBM |
1636 | if (!name) { |
1637 | ret = -ENOMEM; | |
1638 | goto out; | |
1639 | } | |
2a29edc6 | 1640 | |
e02119d5 CM |
1641 | log_type = btrfs_dir_type(eb, di); |
1642 | read_extent_buffer(eb, name, (unsigned long)(di + 1), | |
1643 | name_len); | |
1644 | ||
1645 | btrfs_dir_item_key_to_cpu(eb, di, &log_key); | |
4bef0848 CM |
1646 | exists = btrfs_lookup_inode(trans, root, path, &log_key, 0); |
1647 | if (exists == 0) | |
1648 | exists = 1; | |
1649 | else | |
1650 | exists = 0; | |
b3b4aa74 | 1651 | btrfs_release_path(path); |
4bef0848 | 1652 | |
e02119d5 CM |
1653 | if (key->type == BTRFS_DIR_ITEM_KEY) { |
1654 | dst_di = btrfs_lookup_dir_item(trans, root, path, key->objectid, | |
1655 | name, name_len, 1); | |
d397712b | 1656 | } else if (key->type == BTRFS_DIR_INDEX_KEY) { |
e02119d5 CM |
1657 | dst_di = btrfs_lookup_dir_index_item(trans, root, path, |
1658 | key->objectid, | |
1659 | key->offset, name, | |
1660 | name_len, 1); | |
1661 | } else { | |
3650860b JB |
1662 | /* Corruption */ |
1663 | ret = -EINVAL; | |
1664 | goto out; | |
e02119d5 | 1665 | } |
c704005d | 1666 | if (IS_ERR_OR_NULL(dst_di)) { |
e02119d5 CM |
1667 | /* we need a sequence number to insert, so we only |
1668 | * do inserts for the BTRFS_DIR_INDEX_KEY types | |
1669 | */ | |
1670 | if (key->type != BTRFS_DIR_INDEX_KEY) | |
1671 | goto out; | |
1672 | goto insert; | |
1673 | } | |
1674 | ||
1675 | btrfs_dir_item_key_to_cpu(path->nodes[0], dst_di, &found_key); | |
1676 | /* the existing item matches the logged item */ | |
1677 | if (found_key.objectid == log_key.objectid && | |
1678 | found_key.type == log_key.type && | |
1679 | found_key.offset == log_key.offset && | |
1680 | btrfs_dir_type(path->nodes[0], dst_di) == log_type) { | |
a2cc11db | 1681 | update_size = false; |
e02119d5 CM |
1682 | goto out; |
1683 | } | |
1684 | ||
1685 | /* | |
1686 | * don't drop the conflicting directory entry if the inode | |
1687 | * for the new entry doesn't exist | |
1688 | */ | |
4bef0848 | 1689 | if (!exists) |
e02119d5 CM |
1690 | goto out; |
1691 | ||
e02119d5 | 1692 | ret = drop_one_dir_item(trans, root, path, dir, dst_di); |
3650860b JB |
1693 | if (ret) |
1694 | goto out; | |
e02119d5 CM |
1695 | |
1696 | if (key->type == BTRFS_DIR_INDEX_KEY) | |
1697 | goto insert; | |
1698 | out: | |
b3b4aa74 | 1699 | btrfs_release_path(path); |
d555438b JB |
1700 | if (!ret && update_size) { |
1701 | btrfs_i_size_write(dir, dir->i_size + name_len * 2); | |
1702 | ret = btrfs_update_inode(trans, root, dir); | |
1703 | } | |
e02119d5 CM |
1704 | kfree(name); |
1705 | iput(dir); | |
bb53eda9 FM |
1706 | if (!ret && name_added) |
1707 | ret = 1; | |
3650860b | 1708 | return ret; |
e02119d5 CM |
1709 | |
1710 | insert: | |
df8d116f FM |
1711 | if (name_in_log_ref(root->log_root, name, name_len, |
1712 | key->objectid, log_key.objectid)) { | |
1713 | /* The dentry will be added later. */ | |
1714 | ret = 0; | |
1715 | update_size = false; | |
1716 | goto out; | |
1717 | } | |
b3b4aa74 | 1718 | btrfs_release_path(path); |
e02119d5 CM |
1719 | ret = insert_one_name(trans, root, path, key->objectid, key->offset, |
1720 | name, name_len, log_type, &log_key); | |
df8d116f | 1721 | if (ret && ret != -ENOENT && ret != -EEXIST) |
3650860b | 1722 | goto out; |
bb53eda9 FM |
1723 | if (!ret) |
1724 | name_added = true; | |
d555438b | 1725 | update_size = false; |
3650860b | 1726 | ret = 0; |
e02119d5 CM |
1727 | goto out; |
1728 | } | |
1729 | ||
1730 | /* | |
1731 | * find all the names in a directory item and reconcile them into | |
1732 | * the subvolume. Only BTRFS_DIR_ITEM_KEY types will have more than | |
1733 | * one name in a directory item, but the same code gets used for | |
1734 | * both directory index types | |
1735 | */ | |
1736 | static noinline int replay_one_dir_item(struct btrfs_trans_handle *trans, | |
1737 | struct btrfs_root *root, | |
1738 | struct btrfs_path *path, | |
1739 | struct extent_buffer *eb, int slot, | |
1740 | struct btrfs_key *key) | |
1741 | { | |
bb53eda9 | 1742 | int ret = 0; |
e02119d5 CM |
1743 | u32 item_size = btrfs_item_size_nr(eb, slot); |
1744 | struct btrfs_dir_item *di; | |
1745 | int name_len; | |
1746 | unsigned long ptr; | |
1747 | unsigned long ptr_end; | |
bb53eda9 | 1748 | struct btrfs_path *fixup_path = NULL; |
e02119d5 CM |
1749 | |
1750 | ptr = btrfs_item_ptr_offset(eb, slot); | |
1751 | ptr_end = ptr + item_size; | |
d397712b | 1752 | while (ptr < ptr_end) { |
e02119d5 | 1753 | di = (struct btrfs_dir_item *)ptr; |
22a94d44 JB |
1754 | if (verify_dir_item(root, eb, di)) |
1755 | return -EIO; | |
e02119d5 CM |
1756 | name_len = btrfs_dir_name_len(eb, di); |
1757 | ret = replay_one_name(trans, root, path, eb, di, key); | |
bb53eda9 FM |
1758 | if (ret < 0) |
1759 | break; | |
e02119d5 CM |
1760 | ptr = (unsigned long)(di + 1); |
1761 | ptr += name_len; | |
bb53eda9 FM |
1762 | |
1763 | /* | |
1764 | * If this entry refers to a non-directory (directories can not | |
1765 | * have a link count > 1) and it was added in the transaction | |
1766 | * that was not committed, make sure we fixup the link count of | |
1767 | * the inode it the entry points to. Otherwise something like | |
1768 | * the following would result in a directory pointing to an | |
1769 | * inode with a wrong link that does not account for this dir | |
1770 | * entry: | |
1771 | * | |
1772 | * mkdir testdir | |
1773 | * touch testdir/foo | |
1774 | * touch testdir/bar | |
1775 | * sync | |
1776 | * | |
1777 | * ln testdir/bar testdir/bar_link | |
1778 | * ln testdir/foo testdir/foo_link | |
1779 | * xfs_io -c "fsync" testdir/bar | |
1780 | * | |
1781 | * <power failure> | |
1782 | * | |
1783 | * mount fs, log replay happens | |
1784 | * | |
1785 | * File foo would remain with a link count of 1 when it has two | |
1786 | * entries pointing to it in the directory testdir. This would | |
1787 | * make it impossible to ever delete the parent directory has | |
1788 | * it would result in stale dentries that can never be deleted. | |
1789 | */ | |
1790 | if (ret == 1 && btrfs_dir_type(eb, di) != BTRFS_FT_DIR) { | |
1791 | struct btrfs_key di_key; | |
1792 | ||
1793 | if (!fixup_path) { | |
1794 | fixup_path = btrfs_alloc_path(); | |
1795 | if (!fixup_path) { | |
1796 | ret = -ENOMEM; | |
1797 | break; | |
1798 | } | |
1799 | } | |
1800 | ||
1801 | btrfs_dir_item_key_to_cpu(eb, di, &di_key); | |
1802 | ret = link_to_fixup_dir(trans, root, fixup_path, | |
1803 | di_key.objectid); | |
1804 | if (ret) | |
1805 | break; | |
1806 | } | |
1807 | ret = 0; | |
e02119d5 | 1808 | } |
bb53eda9 FM |
1809 | btrfs_free_path(fixup_path); |
1810 | return ret; | |
e02119d5 CM |
1811 | } |
1812 | ||
1813 | /* | |
1814 | * directory replay has two parts. There are the standard directory | |
1815 | * items in the log copied from the subvolume, and range items | |
1816 | * created in the log while the subvolume was logged. | |
1817 | * | |
1818 | * The range items tell us which parts of the key space the log | |
1819 | * is authoritative for. During replay, if a key in the subvolume | |
1820 | * directory is in a logged range item, but not actually in the log | |
1821 | * that means it was deleted from the directory before the fsync | |
1822 | * and should be removed. | |
1823 | */ | |
1824 | static noinline int find_dir_range(struct btrfs_root *root, | |
1825 | struct btrfs_path *path, | |
1826 | u64 dirid, int key_type, | |
1827 | u64 *start_ret, u64 *end_ret) | |
1828 | { | |
1829 | struct btrfs_key key; | |
1830 | u64 found_end; | |
1831 | struct btrfs_dir_log_item *item; | |
1832 | int ret; | |
1833 | int nritems; | |
1834 | ||
1835 | if (*start_ret == (u64)-1) | |
1836 | return 1; | |
1837 | ||
1838 | key.objectid = dirid; | |
1839 | key.type = key_type; | |
1840 | key.offset = *start_ret; | |
1841 | ||
1842 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
1843 | if (ret < 0) | |
1844 | goto out; | |
1845 | if (ret > 0) { | |
1846 | if (path->slots[0] == 0) | |
1847 | goto out; | |
1848 | path->slots[0]--; | |
1849 | } | |
1850 | if (ret != 0) | |
1851 | btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]); | |
1852 | ||
1853 | if (key.type != key_type || key.objectid != dirid) { | |
1854 | ret = 1; | |
1855 | goto next; | |
1856 | } | |
1857 | item = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
1858 | struct btrfs_dir_log_item); | |
1859 | found_end = btrfs_dir_log_end(path->nodes[0], item); | |
1860 | ||
1861 | if (*start_ret >= key.offset && *start_ret <= found_end) { | |
1862 | ret = 0; | |
1863 | *start_ret = key.offset; | |
1864 | *end_ret = found_end; | |
1865 | goto out; | |
1866 | } | |
1867 | ret = 1; | |
1868 | next: | |
1869 | /* check the next slot in the tree to see if it is a valid item */ | |
1870 | nritems = btrfs_header_nritems(path->nodes[0]); | |
1871 | if (path->slots[0] >= nritems) { | |
1872 | ret = btrfs_next_leaf(root, path); | |
1873 | if (ret) | |
1874 | goto out; | |
1875 | } else { | |
1876 | path->slots[0]++; | |
1877 | } | |
1878 | ||
1879 | btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]); | |
1880 | ||
1881 | if (key.type != key_type || key.objectid != dirid) { | |
1882 | ret = 1; | |
1883 | goto out; | |
1884 | } | |
1885 | item = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
1886 | struct btrfs_dir_log_item); | |
1887 | found_end = btrfs_dir_log_end(path->nodes[0], item); | |
1888 | *start_ret = key.offset; | |
1889 | *end_ret = found_end; | |
1890 | ret = 0; | |
1891 | out: | |
b3b4aa74 | 1892 | btrfs_release_path(path); |
e02119d5 CM |
1893 | return ret; |
1894 | } | |
1895 | ||
1896 | /* | |
1897 | * this looks for a given directory item in the log. If the directory | |
1898 | * item is not in the log, the item is removed and the inode it points | |
1899 | * to is unlinked | |
1900 | */ | |
1901 | static noinline int check_item_in_log(struct btrfs_trans_handle *trans, | |
1902 | struct btrfs_root *root, | |
1903 | struct btrfs_root *log, | |
1904 | struct btrfs_path *path, | |
1905 | struct btrfs_path *log_path, | |
1906 | struct inode *dir, | |
1907 | struct btrfs_key *dir_key) | |
1908 | { | |
1909 | int ret; | |
1910 | struct extent_buffer *eb; | |
1911 | int slot; | |
1912 | u32 item_size; | |
1913 | struct btrfs_dir_item *di; | |
1914 | struct btrfs_dir_item *log_di; | |
1915 | int name_len; | |
1916 | unsigned long ptr; | |
1917 | unsigned long ptr_end; | |
1918 | char *name; | |
1919 | struct inode *inode; | |
1920 | struct btrfs_key location; | |
1921 | ||
1922 | again: | |
1923 | eb = path->nodes[0]; | |
1924 | slot = path->slots[0]; | |
1925 | item_size = btrfs_item_size_nr(eb, slot); | |
1926 | ptr = btrfs_item_ptr_offset(eb, slot); | |
1927 | ptr_end = ptr + item_size; | |
d397712b | 1928 | while (ptr < ptr_end) { |
e02119d5 | 1929 | di = (struct btrfs_dir_item *)ptr; |
22a94d44 JB |
1930 | if (verify_dir_item(root, eb, di)) { |
1931 | ret = -EIO; | |
1932 | goto out; | |
1933 | } | |
1934 | ||
e02119d5 CM |
1935 | name_len = btrfs_dir_name_len(eb, di); |
1936 | name = kmalloc(name_len, GFP_NOFS); | |
1937 | if (!name) { | |
1938 | ret = -ENOMEM; | |
1939 | goto out; | |
1940 | } | |
1941 | read_extent_buffer(eb, name, (unsigned long)(di + 1), | |
1942 | name_len); | |
1943 | log_di = NULL; | |
12fcfd22 | 1944 | if (log && dir_key->type == BTRFS_DIR_ITEM_KEY) { |
e02119d5 CM |
1945 | log_di = btrfs_lookup_dir_item(trans, log, log_path, |
1946 | dir_key->objectid, | |
1947 | name, name_len, 0); | |
12fcfd22 | 1948 | } else if (log && dir_key->type == BTRFS_DIR_INDEX_KEY) { |
e02119d5 CM |
1949 | log_di = btrfs_lookup_dir_index_item(trans, log, |
1950 | log_path, | |
1951 | dir_key->objectid, | |
1952 | dir_key->offset, | |
1953 | name, name_len, 0); | |
1954 | } | |
269d040f | 1955 | if (!log_di || (IS_ERR(log_di) && PTR_ERR(log_di) == -ENOENT)) { |
e02119d5 | 1956 | btrfs_dir_item_key_to_cpu(eb, di, &location); |
b3b4aa74 DS |
1957 | btrfs_release_path(path); |
1958 | btrfs_release_path(log_path); | |
e02119d5 | 1959 | inode = read_one_inode(root, location.objectid); |
c00e9493 TI |
1960 | if (!inode) { |
1961 | kfree(name); | |
1962 | return -EIO; | |
1963 | } | |
e02119d5 CM |
1964 | |
1965 | ret = link_to_fixup_dir(trans, root, | |
1966 | path, location.objectid); | |
3650860b JB |
1967 | if (ret) { |
1968 | kfree(name); | |
1969 | iput(inode); | |
1970 | goto out; | |
1971 | } | |
1972 | ||
8b558c5f | 1973 | inc_nlink(inode); |
e02119d5 CM |
1974 | ret = btrfs_unlink_inode(trans, root, dir, inode, |
1975 | name, name_len); | |
3650860b | 1976 | if (!ret) |
ada9af21 | 1977 | ret = btrfs_run_delayed_items(trans, root); |
e02119d5 CM |
1978 | kfree(name); |
1979 | iput(inode); | |
3650860b JB |
1980 | if (ret) |
1981 | goto out; | |
e02119d5 CM |
1982 | |
1983 | /* there might still be more names under this key | |
1984 | * check and repeat if required | |
1985 | */ | |
1986 | ret = btrfs_search_slot(NULL, root, dir_key, path, | |
1987 | 0, 0); | |
1988 | if (ret == 0) | |
1989 | goto again; | |
1990 | ret = 0; | |
1991 | goto out; | |
269d040f FDBM |
1992 | } else if (IS_ERR(log_di)) { |
1993 | kfree(name); | |
1994 | return PTR_ERR(log_di); | |
e02119d5 | 1995 | } |
b3b4aa74 | 1996 | btrfs_release_path(log_path); |
e02119d5 CM |
1997 | kfree(name); |
1998 | ||
1999 | ptr = (unsigned long)(di + 1); | |
2000 | ptr += name_len; | |
2001 | } | |
2002 | ret = 0; | |
2003 | out: | |
b3b4aa74 DS |
2004 | btrfs_release_path(path); |
2005 | btrfs_release_path(log_path); | |
e02119d5 CM |
2006 | return ret; |
2007 | } | |
2008 | ||
4f764e51 FM |
2009 | static int replay_xattr_deletes(struct btrfs_trans_handle *trans, |
2010 | struct btrfs_root *root, | |
2011 | struct btrfs_root *log, | |
2012 | struct btrfs_path *path, | |
2013 | const u64 ino) | |
2014 | { | |
2015 | struct btrfs_key search_key; | |
2016 | struct btrfs_path *log_path; | |
2017 | int i; | |
2018 | int nritems; | |
2019 | int ret; | |
2020 | ||
2021 | log_path = btrfs_alloc_path(); | |
2022 | if (!log_path) | |
2023 | return -ENOMEM; | |
2024 | ||
2025 | search_key.objectid = ino; | |
2026 | search_key.type = BTRFS_XATTR_ITEM_KEY; | |
2027 | search_key.offset = 0; | |
2028 | again: | |
2029 | ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0); | |
2030 | if (ret < 0) | |
2031 | goto out; | |
2032 | process_leaf: | |
2033 | nritems = btrfs_header_nritems(path->nodes[0]); | |
2034 | for (i = path->slots[0]; i < nritems; i++) { | |
2035 | struct btrfs_key key; | |
2036 | struct btrfs_dir_item *di; | |
2037 | struct btrfs_dir_item *log_di; | |
2038 | u32 total_size; | |
2039 | u32 cur; | |
2040 | ||
2041 | btrfs_item_key_to_cpu(path->nodes[0], &key, i); | |
2042 | if (key.objectid != ino || key.type != BTRFS_XATTR_ITEM_KEY) { | |
2043 | ret = 0; | |
2044 | goto out; | |
2045 | } | |
2046 | ||
2047 | di = btrfs_item_ptr(path->nodes[0], i, struct btrfs_dir_item); | |
2048 | total_size = btrfs_item_size_nr(path->nodes[0], i); | |
2049 | cur = 0; | |
2050 | while (cur < total_size) { | |
2051 | u16 name_len = btrfs_dir_name_len(path->nodes[0], di); | |
2052 | u16 data_len = btrfs_dir_data_len(path->nodes[0], di); | |
2053 | u32 this_len = sizeof(*di) + name_len + data_len; | |
2054 | char *name; | |
2055 | ||
2056 | name = kmalloc(name_len, GFP_NOFS); | |
2057 | if (!name) { | |
2058 | ret = -ENOMEM; | |
2059 | goto out; | |
2060 | } | |
2061 | read_extent_buffer(path->nodes[0], name, | |
2062 | (unsigned long)(di + 1), name_len); | |
2063 | ||
2064 | log_di = btrfs_lookup_xattr(NULL, log, log_path, ino, | |
2065 | name, name_len, 0); | |
2066 | btrfs_release_path(log_path); | |
2067 | if (!log_di) { | |
2068 | /* Doesn't exist in log tree, so delete it. */ | |
2069 | btrfs_release_path(path); | |
2070 | di = btrfs_lookup_xattr(trans, root, path, ino, | |
2071 | name, name_len, -1); | |
2072 | kfree(name); | |
2073 | if (IS_ERR(di)) { | |
2074 | ret = PTR_ERR(di); | |
2075 | goto out; | |
2076 | } | |
2077 | ASSERT(di); | |
2078 | ret = btrfs_delete_one_dir_name(trans, root, | |
2079 | path, di); | |
2080 | if (ret) | |
2081 | goto out; | |
2082 | btrfs_release_path(path); | |
2083 | search_key = key; | |
2084 | goto again; | |
2085 | } | |
2086 | kfree(name); | |
2087 | if (IS_ERR(log_di)) { | |
2088 | ret = PTR_ERR(log_di); | |
2089 | goto out; | |
2090 | } | |
2091 | cur += this_len; | |
2092 | di = (struct btrfs_dir_item *)((char *)di + this_len); | |
2093 | } | |
2094 | } | |
2095 | ret = btrfs_next_leaf(root, path); | |
2096 | if (ret > 0) | |
2097 | ret = 0; | |
2098 | else if (ret == 0) | |
2099 | goto process_leaf; | |
2100 | out: | |
2101 | btrfs_free_path(log_path); | |
2102 | btrfs_release_path(path); | |
2103 | return ret; | |
2104 | } | |
2105 | ||
2106 | ||
e02119d5 CM |
2107 | /* |
2108 | * deletion replay happens before we copy any new directory items | |
2109 | * out of the log or out of backreferences from inodes. It | |
2110 | * scans the log to find ranges of keys that log is authoritative for, | |
2111 | * and then scans the directory to find items in those ranges that are | |
2112 | * not present in the log. | |
2113 | * | |
2114 | * Anything we don't find in the log is unlinked and removed from the | |
2115 | * directory. | |
2116 | */ | |
2117 | static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans, | |
2118 | struct btrfs_root *root, | |
2119 | struct btrfs_root *log, | |
2120 | struct btrfs_path *path, | |
12fcfd22 | 2121 | u64 dirid, int del_all) |
e02119d5 CM |
2122 | { |
2123 | u64 range_start; | |
2124 | u64 range_end; | |
2125 | int key_type = BTRFS_DIR_LOG_ITEM_KEY; | |
2126 | int ret = 0; | |
2127 | struct btrfs_key dir_key; | |
2128 | struct btrfs_key found_key; | |
2129 | struct btrfs_path *log_path; | |
2130 | struct inode *dir; | |
2131 | ||
2132 | dir_key.objectid = dirid; | |
2133 | dir_key.type = BTRFS_DIR_ITEM_KEY; | |
2134 | log_path = btrfs_alloc_path(); | |
2135 | if (!log_path) | |
2136 | return -ENOMEM; | |
2137 | ||
2138 | dir = read_one_inode(root, dirid); | |
2139 | /* it isn't an error if the inode isn't there, that can happen | |
2140 | * because we replay the deletes before we copy in the inode item | |
2141 | * from the log | |
2142 | */ | |
2143 | if (!dir) { | |
2144 | btrfs_free_path(log_path); | |
2145 | return 0; | |
2146 | } | |
2147 | again: | |
2148 | range_start = 0; | |
2149 | range_end = 0; | |
d397712b | 2150 | while (1) { |
12fcfd22 CM |
2151 | if (del_all) |
2152 | range_end = (u64)-1; | |
2153 | else { | |
2154 | ret = find_dir_range(log, path, dirid, key_type, | |
2155 | &range_start, &range_end); | |
2156 | if (ret != 0) | |
2157 | break; | |
2158 | } | |
e02119d5 CM |
2159 | |
2160 | dir_key.offset = range_start; | |
d397712b | 2161 | while (1) { |
e02119d5 CM |
2162 | int nritems; |
2163 | ret = btrfs_search_slot(NULL, root, &dir_key, path, | |
2164 | 0, 0); | |
2165 | if (ret < 0) | |
2166 | goto out; | |
2167 | ||
2168 | nritems = btrfs_header_nritems(path->nodes[0]); | |
2169 | if (path->slots[0] >= nritems) { | |
2170 | ret = btrfs_next_leaf(root, path); | |
2171 | if (ret) | |
2172 | break; | |
2173 | } | |
2174 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, | |
2175 | path->slots[0]); | |
2176 | if (found_key.objectid != dirid || | |
2177 | found_key.type != dir_key.type) | |
2178 | goto next_type; | |
2179 | ||
2180 | if (found_key.offset > range_end) | |
2181 | break; | |
2182 | ||
2183 | ret = check_item_in_log(trans, root, log, path, | |
12fcfd22 CM |
2184 | log_path, dir, |
2185 | &found_key); | |
3650860b JB |
2186 | if (ret) |
2187 | goto out; | |
e02119d5 CM |
2188 | if (found_key.offset == (u64)-1) |
2189 | break; | |
2190 | dir_key.offset = found_key.offset + 1; | |
2191 | } | |
b3b4aa74 | 2192 | btrfs_release_path(path); |
e02119d5 CM |
2193 | if (range_end == (u64)-1) |
2194 | break; | |
2195 | range_start = range_end + 1; | |
2196 | } | |
2197 | ||
2198 | next_type: | |
2199 | ret = 0; | |
2200 | if (key_type == BTRFS_DIR_LOG_ITEM_KEY) { | |
2201 | key_type = BTRFS_DIR_LOG_INDEX_KEY; | |
2202 | dir_key.type = BTRFS_DIR_INDEX_KEY; | |
b3b4aa74 | 2203 | btrfs_release_path(path); |
e02119d5 CM |
2204 | goto again; |
2205 | } | |
2206 | out: | |
b3b4aa74 | 2207 | btrfs_release_path(path); |
e02119d5 CM |
2208 | btrfs_free_path(log_path); |
2209 | iput(dir); | |
2210 | return ret; | |
2211 | } | |
2212 | ||
2213 | /* | |
2214 | * the process_func used to replay items from the log tree. This | |
2215 | * gets called in two different stages. The first stage just looks | |
2216 | * for inodes and makes sure they are all copied into the subvolume. | |
2217 | * | |
2218 | * The second stage copies all the other item types from the log into | |
2219 | * the subvolume. The two stage approach is slower, but gets rid of | |
2220 | * lots of complexity around inodes referencing other inodes that exist | |
2221 | * only in the log (references come from either directory items or inode | |
2222 | * back refs). | |
2223 | */ | |
2224 | static int replay_one_buffer(struct btrfs_root *log, struct extent_buffer *eb, | |
2225 | struct walk_control *wc, u64 gen) | |
2226 | { | |
2227 | int nritems; | |
2228 | struct btrfs_path *path; | |
2229 | struct btrfs_root *root = wc->replay_dest; | |
2230 | struct btrfs_key key; | |
e02119d5 CM |
2231 | int level; |
2232 | int i; | |
2233 | int ret; | |
2234 | ||
018642a1 TI |
2235 | ret = btrfs_read_buffer(eb, gen); |
2236 | if (ret) | |
2237 | return ret; | |
e02119d5 CM |
2238 | |
2239 | level = btrfs_header_level(eb); | |
2240 | ||
2241 | if (level != 0) | |
2242 | return 0; | |
2243 | ||
2244 | path = btrfs_alloc_path(); | |
1e5063d0 MF |
2245 | if (!path) |
2246 | return -ENOMEM; | |
e02119d5 CM |
2247 | |
2248 | nritems = btrfs_header_nritems(eb); | |
2249 | for (i = 0; i < nritems; i++) { | |
2250 | btrfs_item_key_to_cpu(eb, &key, i); | |
e02119d5 CM |
2251 | |
2252 | /* inode keys are done during the first stage */ | |
2253 | if (key.type == BTRFS_INODE_ITEM_KEY && | |
2254 | wc->stage == LOG_WALK_REPLAY_INODES) { | |
e02119d5 CM |
2255 | struct btrfs_inode_item *inode_item; |
2256 | u32 mode; | |
2257 | ||
2258 | inode_item = btrfs_item_ptr(eb, i, | |
2259 | struct btrfs_inode_item); | |
4f764e51 FM |
2260 | ret = replay_xattr_deletes(wc->trans, root, log, |
2261 | path, key.objectid); | |
2262 | if (ret) | |
2263 | break; | |
e02119d5 CM |
2264 | mode = btrfs_inode_mode(eb, inode_item); |
2265 | if (S_ISDIR(mode)) { | |
2266 | ret = replay_dir_deletes(wc->trans, | |
12fcfd22 | 2267 | root, log, path, key.objectid, 0); |
b50c6e25 JB |
2268 | if (ret) |
2269 | break; | |
e02119d5 CM |
2270 | } |
2271 | ret = overwrite_item(wc->trans, root, path, | |
2272 | eb, i, &key); | |
b50c6e25 JB |
2273 | if (ret) |
2274 | break; | |
e02119d5 | 2275 | |
c71bf099 YZ |
2276 | /* for regular files, make sure corresponding |
2277 | * orhpan item exist. extents past the new EOF | |
2278 | * will be truncated later by orphan cleanup. | |
e02119d5 CM |
2279 | */ |
2280 | if (S_ISREG(mode)) { | |
c71bf099 YZ |
2281 | ret = insert_orphan_item(wc->trans, root, |
2282 | key.objectid); | |
b50c6e25 JB |
2283 | if (ret) |
2284 | break; | |
e02119d5 | 2285 | } |
c71bf099 | 2286 | |
e02119d5 CM |
2287 | ret = link_to_fixup_dir(wc->trans, root, |
2288 | path, key.objectid); | |
b50c6e25 JB |
2289 | if (ret) |
2290 | break; | |
e02119d5 | 2291 | } |
dd8e7217 JB |
2292 | |
2293 | if (key.type == BTRFS_DIR_INDEX_KEY && | |
2294 | wc->stage == LOG_WALK_REPLAY_DIR_INDEX) { | |
2295 | ret = replay_one_dir_item(wc->trans, root, path, | |
2296 | eb, i, &key); | |
2297 | if (ret) | |
2298 | break; | |
2299 | } | |
2300 | ||
e02119d5 CM |
2301 | if (wc->stage < LOG_WALK_REPLAY_ALL) |
2302 | continue; | |
2303 | ||
2304 | /* these keys are simply copied */ | |
2305 | if (key.type == BTRFS_XATTR_ITEM_KEY) { | |
2306 | ret = overwrite_item(wc->trans, root, path, | |
2307 | eb, i, &key); | |
b50c6e25 JB |
2308 | if (ret) |
2309 | break; | |
2da1c669 LB |
2310 | } else if (key.type == BTRFS_INODE_REF_KEY || |
2311 | key.type == BTRFS_INODE_EXTREF_KEY) { | |
f186373f MF |
2312 | ret = add_inode_ref(wc->trans, root, log, path, |
2313 | eb, i, &key); | |
b50c6e25 JB |
2314 | if (ret && ret != -ENOENT) |
2315 | break; | |
2316 | ret = 0; | |
e02119d5 CM |
2317 | } else if (key.type == BTRFS_EXTENT_DATA_KEY) { |
2318 | ret = replay_one_extent(wc->trans, root, path, | |
2319 | eb, i, &key); | |
b50c6e25 JB |
2320 | if (ret) |
2321 | break; | |
dd8e7217 | 2322 | } else if (key.type == BTRFS_DIR_ITEM_KEY) { |
e02119d5 CM |
2323 | ret = replay_one_dir_item(wc->trans, root, path, |
2324 | eb, i, &key); | |
b50c6e25 JB |
2325 | if (ret) |
2326 | break; | |
e02119d5 CM |
2327 | } |
2328 | } | |
2329 | btrfs_free_path(path); | |
b50c6e25 | 2330 | return ret; |
e02119d5 CM |
2331 | } |
2332 | ||
d397712b | 2333 | static noinline int walk_down_log_tree(struct btrfs_trans_handle *trans, |
e02119d5 CM |
2334 | struct btrfs_root *root, |
2335 | struct btrfs_path *path, int *level, | |
2336 | struct walk_control *wc) | |
2337 | { | |
2338 | u64 root_owner; | |
e02119d5 CM |
2339 | u64 bytenr; |
2340 | u64 ptr_gen; | |
2341 | struct extent_buffer *next; | |
2342 | struct extent_buffer *cur; | |
2343 | struct extent_buffer *parent; | |
2344 | u32 blocksize; | |
2345 | int ret = 0; | |
2346 | ||
2347 | WARN_ON(*level < 0); | |
2348 | WARN_ON(*level >= BTRFS_MAX_LEVEL); | |
2349 | ||
d397712b | 2350 | while (*level > 0) { |
e02119d5 CM |
2351 | WARN_ON(*level < 0); |
2352 | WARN_ON(*level >= BTRFS_MAX_LEVEL); | |
2353 | cur = path->nodes[*level]; | |
2354 | ||
fae7f21c | 2355 | WARN_ON(btrfs_header_level(cur) != *level); |
e02119d5 CM |
2356 | |
2357 | if (path->slots[*level] >= | |
2358 | btrfs_header_nritems(cur)) | |
2359 | break; | |
2360 | ||
2361 | bytenr = btrfs_node_blockptr(cur, path->slots[*level]); | |
2362 | ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]); | |
707e8a07 | 2363 | blocksize = root->nodesize; |
e02119d5 CM |
2364 | |
2365 | parent = path->nodes[*level]; | |
2366 | root_owner = btrfs_header_owner(parent); | |
e02119d5 | 2367 | |
a83fffb7 | 2368 | next = btrfs_find_create_tree_block(root, bytenr); |
2a29edc6 | 2369 | if (!next) |
2370 | return -ENOMEM; | |
e02119d5 | 2371 | |
e02119d5 | 2372 | if (*level == 1) { |
1e5063d0 | 2373 | ret = wc->process_func(root, next, wc, ptr_gen); |
b50c6e25 JB |
2374 | if (ret) { |
2375 | free_extent_buffer(next); | |
1e5063d0 | 2376 | return ret; |
b50c6e25 | 2377 | } |
4a500fd1 | 2378 | |
e02119d5 CM |
2379 | path->slots[*level]++; |
2380 | if (wc->free) { | |
018642a1 TI |
2381 | ret = btrfs_read_buffer(next, ptr_gen); |
2382 | if (ret) { | |
2383 | free_extent_buffer(next); | |
2384 | return ret; | |
2385 | } | |
e02119d5 | 2386 | |
681ae509 JB |
2387 | if (trans) { |
2388 | btrfs_tree_lock(next); | |
2389 | btrfs_set_lock_blocking(next); | |
01d58472 DD |
2390 | clean_tree_block(trans, root->fs_info, |
2391 | next); | |
681ae509 JB |
2392 | btrfs_wait_tree_block_writeback(next); |
2393 | btrfs_tree_unlock(next); | |
2394 | } | |
e02119d5 | 2395 | |
e02119d5 CM |
2396 | WARN_ON(root_owner != |
2397 | BTRFS_TREE_LOG_OBJECTID); | |
e688b725 | 2398 | ret = btrfs_free_and_pin_reserved_extent(root, |
d00aff00 | 2399 | bytenr, blocksize); |
3650860b JB |
2400 | if (ret) { |
2401 | free_extent_buffer(next); | |
2402 | return ret; | |
2403 | } | |
e02119d5 CM |
2404 | } |
2405 | free_extent_buffer(next); | |
2406 | continue; | |
2407 | } | |
018642a1 TI |
2408 | ret = btrfs_read_buffer(next, ptr_gen); |
2409 | if (ret) { | |
2410 | free_extent_buffer(next); | |
2411 | return ret; | |
2412 | } | |
e02119d5 CM |
2413 | |
2414 | WARN_ON(*level <= 0); | |
2415 | if (path->nodes[*level-1]) | |
2416 | free_extent_buffer(path->nodes[*level-1]); | |
2417 | path->nodes[*level-1] = next; | |
2418 | *level = btrfs_header_level(next); | |
2419 | path->slots[*level] = 0; | |
2420 | cond_resched(); | |
2421 | } | |
2422 | WARN_ON(*level < 0); | |
2423 | WARN_ON(*level >= BTRFS_MAX_LEVEL); | |
2424 | ||
4a500fd1 | 2425 | path->slots[*level] = btrfs_header_nritems(path->nodes[*level]); |
e02119d5 CM |
2426 | |
2427 | cond_resched(); | |
2428 | return 0; | |
2429 | } | |
2430 | ||
d397712b | 2431 | static noinline int walk_up_log_tree(struct btrfs_trans_handle *trans, |
e02119d5 CM |
2432 | struct btrfs_root *root, |
2433 | struct btrfs_path *path, int *level, | |
2434 | struct walk_control *wc) | |
2435 | { | |
2436 | u64 root_owner; | |
e02119d5 CM |
2437 | int i; |
2438 | int slot; | |
2439 | int ret; | |
2440 | ||
d397712b | 2441 | for (i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) { |
e02119d5 | 2442 | slot = path->slots[i]; |
4a500fd1 | 2443 | if (slot + 1 < btrfs_header_nritems(path->nodes[i])) { |
e02119d5 CM |
2444 | path->slots[i]++; |
2445 | *level = i; | |
2446 | WARN_ON(*level == 0); | |
2447 | return 0; | |
2448 | } else { | |
31840ae1 ZY |
2449 | struct extent_buffer *parent; |
2450 | if (path->nodes[*level] == root->node) | |
2451 | parent = path->nodes[*level]; | |
2452 | else | |
2453 | parent = path->nodes[*level + 1]; | |
2454 | ||
2455 | root_owner = btrfs_header_owner(parent); | |
1e5063d0 | 2456 | ret = wc->process_func(root, path->nodes[*level], wc, |
e02119d5 | 2457 | btrfs_header_generation(path->nodes[*level])); |
1e5063d0 MF |
2458 | if (ret) |
2459 | return ret; | |
2460 | ||
e02119d5 CM |
2461 | if (wc->free) { |
2462 | struct extent_buffer *next; | |
2463 | ||
2464 | next = path->nodes[*level]; | |
2465 | ||
681ae509 JB |
2466 | if (trans) { |
2467 | btrfs_tree_lock(next); | |
2468 | btrfs_set_lock_blocking(next); | |
01d58472 DD |
2469 | clean_tree_block(trans, root->fs_info, |
2470 | next); | |
681ae509 JB |
2471 | btrfs_wait_tree_block_writeback(next); |
2472 | btrfs_tree_unlock(next); | |
2473 | } | |
e02119d5 | 2474 | |
e02119d5 | 2475 | WARN_ON(root_owner != BTRFS_TREE_LOG_OBJECTID); |
e688b725 | 2476 | ret = btrfs_free_and_pin_reserved_extent(root, |
e02119d5 | 2477 | path->nodes[*level]->start, |
d00aff00 | 2478 | path->nodes[*level]->len); |
3650860b JB |
2479 | if (ret) |
2480 | return ret; | |
e02119d5 CM |
2481 | } |
2482 | free_extent_buffer(path->nodes[*level]); | |
2483 | path->nodes[*level] = NULL; | |
2484 | *level = i + 1; | |
2485 | } | |
2486 | } | |
2487 | return 1; | |
2488 | } | |
2489 | ||
2490 | /* | |
2491 | * drop the reference count on the tree rooted at 'snap'. This traverses | |
2492 | * the tree freeing any blocks that have a ref count of zero after being | |
2493 | * decremented. | |
2494 | */ | |
2495 | static int walk_log_tree(struct btrfs_trans_handle *trans, | |
2496 | struct btrfs_root *log, struct walk_control *wc) | |
2497 | { | |
2498 | int ret = 0; | |
2499 | int wret; | |
2500 | int level; | |
2501 | struct btrfs_path *path; | |
e02119d5 CM |
2502 | int orig_level; |
2503 | ||
2504 | path = btrfs_alloc_path(); | |
db5b493a TI |
2505 | if (!path) |
2506 | return -ENOMEM; | |
e02119d5 CM |
2507 | |
2508 | level = btrfs_header_level(log->node); | |
2509 | orig_level = level; | |
2510 | path->nodes[level] = log->node; | |
2511 | extent_buffer_get(log->node); | |
2512 | path->slots[level] = 0; | |
2513 | ||
d397712b | 2514 | while (1) { |
e02119d5 CM |
2515 | wret = walk_down_log_tree(trans, log, path, &level, wc); |
2516 | if (wret > 0) | |
2517 | break; | |
79787eaa | 2518 | if (wret < 0) { |
e02119d5 | 2519 | ret = wret; |
79787eaa JM |
2520 | goto out; |
2521 | } | |
e02119d5 CM |
2522 | |
2523 | wret = walk_up_log_tree(trans, log, path, &level, wc); | |
2524 | if (wret > 0) | |
2525 | break; | |
79787eaa | 2526 | if (wret < 0) { |
e02119d5 | 2527 | ret = wret; |
79787eaa JM |
2528 | goto out; |
2529 | } | |
e02119d5 CM |
2530 | } |
2531 | ||
2532 | /* was the root node processed? if not, catch it here */ | |
2533 | if (path->nodes[orig_level]) { | |
79787eaa | 2534 | ret = wc->process_func(log, path->nodes[orig_level], wc, |
e02119d5 | 2535 | btrfs_header_generation(path->nodes[orig_level])); |
79787eaa JM |
2536 | if (ret) |
2537 | goto out; | |
e02119d5 CM |
2538 | if (wc->free) { |
2539 | struct extent_buffer *next; | |
2540 | ||
2541 | next = path->nodes[orig_level]; | |
2542 | ||
681ae509 JB |
2543 | if (trans) { |
2544 | btrfs_tree_lock(next); | |
2545 | btrfs_set_lock_blocking(next); | |
01d58472 | 2546 | clean_tree_block(trans, log->fs_info, next); |
681ae509 JB |
2547 | btrfs_wait_tree_block_writeback(next); |
2548 | btrfs_tree_unlock(next); | |
2549 | } | |
e02119d5 | 2550 | |
e02119d5 CM |
2551 | WARN_ON(log->root_key.objectid != |
2552 | BTRFS_TREE_LOG_OBJECTID); | |
e688b725 | 2553 | ret = btrfs_free_and_pin_reserved_extent(log, next->start, |
d00aff00 | 2554 | next->len); |
3650860b JB |
2555 | if (ret) |
2556 | goto out; | |
e02119d5 CM |
2557 | } |
2558 | } | |
2559 | ||
79787eaa | 2560 | out: |
e02119d5 | 2561 | btrfs_free_path(path); |
e02119d5 CM |
2562 | return ret; |
2563 | } | |
2564 | ||
7237f183 YZ |
2565 | /* |
2566 | * helper function to update the item for a given subvolumes log root | |
2567 | * in the tree of log roots | |
2568 | */ | |
2569 | static int update_log_root(struct btrfs_trans_handle *trans, | |
2570 | struct btrfs_root *log) | |
2571 | { | |
2572 | int ret; | |
2573 | ||
2574 | if (log->log_transid == 1) { | |
2575 | /* insert root item on the first sync */ | |
2576 | ret = btrfs_insert_root(trans, log->fs_info->log_root_tree, | |
2577 | &log->root_key, &log->root_item); | |
2578 | } else { | |
2579 | ret = btrfs_update_root(trans, log->fs_info->log_root_tree, | |
2580 | &log->root_key, &log->root_item); | |
2581 | } | |
2582 | return ret; | |
2583 | } | |
2584 | ||
8b050d35 MX |
2585 | static void wait_log_commit(struct btrfs_trans_handle *trans, |
2586 | struct btrfs_root *root, int transid) | |
e02119d5 CM |
2587 | { |
2588 | DEFINE_WAIT(wait); | |
7237f183 | 2589 | int index = transid % 2; |
e02119d5 | 2590 | |
7237f183 YZ |
2591 | /* |
2592 | * we only allow two pending log transactions at a time, | |
2593 | * so we know that if ours is more than 2 older than the | |
2594 | * current transaction, we're done | |
2595 | */ | |
e02119d5 | 2596 | do { |
7237f183 YZ |
2597 | prepare_to_wait(&root->log_commit_wait[index], |
2598 | &wait, TASK_UNINTERRUPTIBLE); | |
2599 | mutex_unlock(&root->log_mutex); | |
12fcfd22 | 2600 | |
d1433deb | 2601 | if (root->log_transid_committed < transid && |
7237f183 YZ |
2602 | atomic_read(&root->log_commit[index])) |
2603 | schedule(); | |
12fcfd22 | 2604 | |
7237f183 YZ |
2605 | finish_wait(&root->log_commit_wait[index], &wait); |
2606 | mutex_lock(&root->log_mutex); | |
d1433deb | 2607 | } while (root->log_transid_committed < transid && |
7237f183 | 2608 | atomic_read(&root->log_commit[index])); |
7237f183 YZ |
2609 | } |
2610 | ||
143bede5 JM |
2611 | static void wait_for_writer(struct btrfs_trans_handle *trans, |
2612 | struct btrfs_root *root) | |
7237f183 YZ |
2613 | { |
2614 | DEFINE_WAIT(wait); | |
8b050d35 MX |
2615 | |
2616 | while (atomic_read(&root->log_writers)) { | |
7237f183 YZ |
2617 | prepare_to_wait(&root->log_writer_wait, |
2618 | &wait, TASK_UNINTERRUPTIBLE); | |
2619 | mutex_unlock(&root->log_mutex); | |
8b050d35 | 2620 | if (atomic_read(&root->log_writers)) |
e02119d5 | 2621 | schedule(); |
7237f183 | 2622 | finish_wait(&root->log_writer_wait, &wait); |
575849ec | 2623 | mutex_lock(&root->log_mutex); |
7237f183 | 2624 | } |
e02119d5 CM |
2625 | } |
2626 | ||
8b050d35 MX |
2627 | static inline void btrfs_remove_log_ctx(struct btrfs_root *root, |
2628 | struct btrfs_log_ctx *ctx) | |
2629 | { | |
2630 | if (!ctx) | |
2631 | return; | |
2632 | ||
2633 | mutex_lock(&root->log_mutex); | |
2634 | list_del_init(&ctx->list); | |
2635 | mutex_unlock(&root->log_mutex); | |
2636 | } | |
2637 | ||
2638 | /* | |
2639 | * Invoked in log mutex context, or be sure there is no other task which | |
2640 | * can access the list. | |
2641 | */ | |
2642 | static inline void btrfs_remove_all_log_ctxs(struct btrfs_root *root, | |
2643 | int index, int error) | |
2644 | { | |
2645 | struct btrfs_log_ctx *ctx; | |
2646 | ||
2647 | if (!error) { | |
2648 | INIT_LIST_HEAD(&root->log_ctxs[index]); | |
2649 | return; | |
2650 | } | |
2651 | ||
2652 | list_for_each_entry(ctx, &root->log_ctxs[index], list) | |
2653 | ctx->log_ret = error; | |
2654 | ||
2655 | INIT_LIST_HEAD(&root->log_ctxs[index]); | |
2656 | } | |
2657 | ||
e02119d5 CM |
2658 | /* |
2659 | * btrfs_sync_log does sends a given tree log down to the disk and | |
2660 | * updates the super blocks to record it. When this call is done, | |
12fcfd22 CM |
2661 | * you know that any inodes previously logged are safely on disk only |
2662 | * if it returns 0. | |
2663 | * | |
2664 | * Any other return value means you need to call btrfs_commit_transaction. | |
2665 | * Some of the edge cases for fsyncing directories that have had unlinks | |
2666 | * or renames done in the past mean that sometimes the only safe | |
2667 | * fsync is to commit the whole FS. When btrfs_sync_log returns -EAGAIN, | |
2668 | * that has happened. | |
e02119d5 CM |
2669 | */ |
2670 | int btrfs_sync_log(struct btrfs_trans_handle *trans, | |
8b050d35 | 2671 | struct btrfs_root *root, struct btrfs_log_ctx *ctx) |
e02119d5 | 2672 | { |
7237f183 YZ |
2673 | int index1; |
2674 | int index2; | |
8cef4e16 | 2675 | int mark; |
e02119d5 | 2676 | int ret; |
e02119d5 | 2677 | struct btrfs_root *log = root->log_root; |
7237f183 | 2678 | struct btrfs_root *log_root_tree = root->fs_info->log_root_tree; |
bb14a59b | 2679 | int log_transid = 0; |
8b050d35 | 2680 | struct btrfs_log_ctx root_log_ctx; |
c6adc9cc | 2681 | struct blk_plug plug; |
e02119d5 | 2682 | |
7237f183 | 2683 | mutex_lock(&root->log_mutex); |
d1433deb MX |
2684 | log_transid = ctx->log_transid; |
2685 | if (root->log_transid_committed >= log_transid) { | |
2686 | mutex_unlock(&root->log_mutex); | |
2687 | return ctx->log_ret; | |
2688 | } | |
2689 | ||
2690 | index1 = log_transid % 2; | |
7237f183 | 2691 | if (atomic_read(&root->log_commit[index1])) { |
d1433deb | 2692 | wait_log_commit(trans, root, log_transid); |
7237f183 | 2693 | mutex_unlock(&root->log_mutex); |
8b050d35 | 2694 | return ctx->log_ret; |
e02119d5 | 2695 | } |
d1433deb | 2696 | ASSERT(log_transid == root->log_transid); |
7237f183 YZ |
2697 | atomic_set(&root->log_commit[index1], 1); |
2698 | ||
2699 | /* wait for previous tree log sync to complete */ | |
2700 | if (atomic_read(&root->log_commit[(index1 + 1) % 2])) | |
d1433deb | 2701 | wait_log_commit(trans, root, log_transid - 1); |
48cab2e0 | 2702 | |
86df7eb9 | 2703 | while (1) { |
2ecb7923 | 2704 | int batch = atomic_read(&root->log_batch); |
cd354ad6 | 2705 | /* when we're on an ssd, just kick the log commit out */ |
27cdeb70 MX |
2706 | if (!btrfs_test_opt(root, SSD) && |
2707 | test_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state)) { | |
86df7eb9 YZ |
2708 | mutex_unlock(&root->log_mutex); |
2709 | schedule_timeout_uninterruptible(1); | |
2710 | mutex_lock(&root->log_mutex); | |
2711 | } | |
12fcfd22 | 2712 | wait_for_writer(trans, root); |
2ecb7923 | 2713 | if (batch == atomic_read(&root->log_batch)) |
e02119d5 CM |
2714 | break; |
2715 | } | |
e02119d5 | 2716 | |
12fcfd22 | 2717 | /* bail out if we need to do a full commit */ |
995946dd | 2718 | if (btrfs_need_log_full_commit(root->fs_info, trans)) { |
12fcfd22 | 2719 | ret = -EAGAIN; |
2ab28f32 | 2720 | btrfs_free_logged_extents(log, log_transid); |
12fcfd22 CM |
2721 | mutex_unlock(&root->log_mutex); |
2722 | goto out; | |
2723 | } | |
2724 | ||
8cef4e16 YZ |
2725 | if (log_transid % 2 == 0) |
2726 | mark = EXTENT_DIRTY; | |
2727 | else | |
2728 | mark = EXTENT_NEW; | |
2729 | ||
690587d1 CM |
2730 | /* we start IO on all the marked extents here, but we don't actually |
2731 | * wait for them until later. | |
2732 | */ | |
c6adc9cc | 2733 | blk_start_plug(&plug); |
8cef4e16 | 2734 | ret = btrfs_write_marked_extents(log, &log->dirty_log_pages, mark); |
79787eaa | 2735 | if (ret) { |
c6adc9cc | 2736 | blk_finish_plug(&plug); |
79787eaa | 2737 | btrfs_abort_transaction(trans, root, ret); |
2ab28f32 | 2738 | btrfs_free_logged_extents(log, log_transid); |
995946dd | 2739 | btrfs_set_log_full_commit(root->fs_info, trans); |
79787eaa JM |
2740 | mutex_unlock(&root->log_mutex); |
2741 | goto out; | |
2742 | } | |
7237f183 | 2743 | |
5d4f98a2 | 2744 | btrfs_set_root_node(&log->root_item, log->node); |
7237f183 | 2745 | |
7237f183 YZ |
2746 | root->log_transid++; |
2747 | log->log_transid = root->log_transid; | |
ff782e0a | 2748 | root->log_start_pid = 0; |
7237f183 | 2749 | /* |
8cef4e16 YZ |
2750 | * IO has been started, blocks of the log tree have WRITTEN flag set |
2751 | * in their headers. new modifications of the log will be written to | |
2752 | * new positions. so it's safe to allow log writers to go in. | |
7237f183 YZ |
2753 | */ |
2754 | mutex_unlock(&root->log_mutex); | |
2755 | ||
d1433deb MX |
2756 | btrfs_init_log_ctx(&root_log_ctx); |
2757 | ||
7237f183 | 2758 | mutex_lock(&log_root_tree->log_mutex); |
2ecb7923 | 2759 | atomic_inc(&log_root_tree->log_batch); |
7237f183 | 2760 | atomic_inc(&log_root_tree->log_writers); |
d1433deb MX |
2761 | |
2762 | index2 = log_root_tree->log_transid % 2; | |
2763 | list_add_tail(&root_log_ctx.list, &log_root_tree->log_ctxs[index2]); | |
2764 | root_log_ctx.log_transid = log_root_tree->log_transid; | |
2765 | ||
7237f183 YZ |
2766 | mutex_unlock(&log_root_tree->log_mutex); |
2767 | ||
2768 | ret = update_log_root(trans, log); | |
7237f183 YZ |
2769 | |
2770 | mutex_lock(&log_root_tree->log_mutex); | |
2771 | if (atomic_dec_and_test(&log_root_tree->log_writers)) { | |
2772 | smp_mb(); | |
2773 | if (waitqueue_active(&log_root_tree->log_writer_wait)) | |
2774 | wake_up(&log_root_tree->log_writer_wait); | |
2775 | } | |
2776 | ||
4a500fd1 | 2777 | if (ret) { |
d1433deb MX |
2778 | if (!list_empty(&root_log_ctx.list)) |
2779 | list_del_init(&root_log_ctx.list); | |
2780 | ||
c6adc9cc | 2781 | blk_finish_plug(&plug); |
995946dd MX |
2782 | btrfs_set_log_full_commit(root->fs_info, trans); |
2783 | ||
79787eaa JM |
2784 | if (ret != -ENOSPC) { |
2785 | btrfs_abort_transaction(trans, root, ret); | |
2786 | mutex_unlock(&log_root_tree->log_mutex); | |
2787 | goto out; | |
2788 | } | |
4a500fd1 | 2789 | btrfs_wait_marked_extents(log, &log->dirty_log_pages, mark); |
2ab28f32 | 2790 | btrfs_free_logged_extents(log, log_transid); |
4a500fd1 YZ |
2791 | mutex_unlock(&log_root_tree->log_mutex); |
2792 | ret = -EAGAIN; | |
2793 | goto out; | |
2794 | } | |
2795 | ||
d1433deb | 2796 | if (log_root_tree->log_transid_committed >= root_log_ctx.log_transid) { |
3da5ab56 | 2797 | blk_finish_plug(&plug); |
d1433deb MX |
2798 | mutex_unlock(&log_root_tree->log_mutex); |
2799 | ret = root_log_ctx.log_ret; | |
2800 | goto out; | |
2801 | } | |
8b050d35 | 2802 | |
d1433deb | 2803 | index2 = root_log_ctx.log_transid % 2; |
7237f183 | 2804 | if (atomic_read(&log_root_tree->log_commit[index2])) { |
c6adc9cc | 2805 | blk_finish_plug(&plug); |
5ab5e44a FM |
2806 | ret = btrfs_wait_marked_extents(log, &log->dirty_log_pages, |
2807 | mark); | |
50d9aa99 | 2808 | btrfs_wait_logged_extents(trans, log, log_transid); |
8b050d35 | 2809 | wait_log_commit(trans, log_root_tree, |
d1433deb | 2810 | root_log_ctx.log_transid); |
7237f183 | 2811 | mutex_unlock(&log_root_tree->log_mutex); |
5ab5e44a FM |
2812 | if (!ret) |
2813 | ret = root_log_ctx.log_ret; | |
7237f183 YZ |
2814 | goto out; |
2815 | } | |
d1433deb | 2816 | ASSERT(root_log_ctx.log_transid == log_root_tree->log_transid); |
7237f183 YZ |
2817 | atomic_set(&log_root_tree->log_commit[index2], 1); |
2818 | ||
12fcfd22 CM |
2819 | if (atomic_read(&log_root_tree->log_commit[(index2 + 1) % 2])) { |
2820 | wait_log_commit(trans, log_root_tree, | |
d1433deb | 2821 | root_log_ctx.log_transid - 1); |
12fcfd22 CM |
2822 | } |
2823 | ||
2824 | wait_for_writer(trans, log_root_tree); | |
7237f183 | 2825 | |
12fcfd22 CM |
2826 | /* |
2827 | * now that we've moved on to the tree of log tree roots, | |
2828 | * check the full commit flag again | |
2829 | */ | |
995946dd | 2830 | if (btrfs_need_log_full_commit(root->fs_info, trans)) { |
c6adc9cc | 2831 | blk_finish_plug(&plug); |
8cef4e16 | 2832 | btrfs_wait_marked_extents(log, &log->dirty_log_pages, mark); |
2ab28f32 | 2833 | btrfs_free_logged_extents(log, log_transid); |
12fcfd22 CM |
2834 | mutex_unlock(&log_root_tree->log_mutex); |
2835 | ret = -EAGAIN; | |
2836 | goto out_wake_log_root; | |
2837 | } | |
7237f183 | 2838 | |
c6adc9cc MX |
2839 | ret = btrfs_write_marked_extents(log_root_tree, |
2840 | &log_root_tree->dirty_log_pages, | |
2841 | EXTENT_DIRTY | EXTENT_NEW); | |
2842 | blk_finish_plug(&plug); | |
79787eaa | 2843 | if (ret) { |
995946dd | 2844 | btrfs_set_log_full_commit(root->fs_info, trans); |
79787eaa | 2845 | btrfs_abort_transaction(trans, root, ret); |
2ab28f32 | 2846 | btrfs_free_logged_extents(log, log_transid); |
79787eaa JM |
2847 | mutex_unlock(&log_root_tree->log_mutex); |
2848 | goto out_wake_log_root; | |
2849 | } | |
5ab5e44a FM |
2850 | ret = btrfs_wait_marked_extents(log, &log->dirty_log_pages, mark); |
2851 | if (!ret) | |
2852 | ret = btrfs_wait_marked_extents(log_root_tree, | |
2853 | &log_root_tree->dirty_log_pages, | |
2854 | EXTENT_NEW | EXTENT_DIRTY); | |
2855 | if (ret) { | |
2856 | btrfs_set_log_full_commit(root->fs_info, trans); | |
2857 | btrfs_free_logged_extents(log, log_transid); | |
2858 | mutex_unlock(&log_root_tree->log_mutex); | |
2859 | goto out_wake_log_root; | |
2860 | } | |
50d9aa99 | 2861 | btrfs_wait_logged_extents(trans, log, log_transid); |
e02119d5 | 2862 | |
6c41761f | 2863 | btrfs_set_super_log_root(root->fs_info->super_for_commit, |
7237f183 | 2864 | log_root_tree->node->start); |
6c41761f | 2865 | btrfs_set_super_log_root_level(root->fs_info->super_for_commit, |
7237f183 | 2866 | btrfs_header_level(log_root_tree->node)); |
e02119d5 | 2867 | |
7237f183 | 2868 | log_root_tree->log_transid++; |
7237f183 YZ |
2869 | mutex_unlock(&log_root_tree->log_mutex); |
2870 | ||
2871 | /* | |
2872 | * nobody else is going to jump in and write the the ctree | |
2873 | * super here because the log_commit atomic below is protecting | |
2874 | * us. We must be called with a transaction handle pinning | |
2875 | * the running transaction open, so a full commit can't hop | |
2876 | * in and cause problems either. | |
2877 | */ | |
5af3e8cc | 2878 | ret = write_ctree_super(trans, root->fs_info->tree_root, 1); |
5af3e8cc | 2879 | if (ret) { |
995946dd | 2880 | btrfs_set_log_full_commit(root->fs_info, trans); |
5af3e8cc SB |
2881 | btrfs_abort_transaction(trans, root, ret); |
2882 | goto out_wake_log_root; | |
2883 | } | |
7237f183 | 2884 | |
257c62e1 CM |
2885 | mutex_lock(&root->log_mutex); |
2886 | if (root->last_log_commit < log_transid) | |
2887 | root->last_log_commit = log_transid; | |
2888 | mutex_unlock(&root->log_mutex); | |
2889 | ||
12fcfd22 | 2890 | out_wake_log_root: |
8b050d35 MX |
2891 | /* |
2892 | * We needn't get log_mutex here because we are sure all | |
2893 | * the other tasks are blocked. | |
2894 | */ | |
2895 | btrfs_remove_all_log_ctxs(log_root_tree, index2, ret); | |
2896 | ||
d1433deb MX |
2897 | mutex_lock(&log_root_tree->log_mutex); |
2898 | log_root_tree->log_transid_committed++; | |
7237f183 | 2899 | atomic_set(&log_root_tree->log_commit[index2], 0); |
d1433deb MX |
2900 | mutex_unlock(&log_root_tree->log_mutex); |
2901 | ||
7237f183 YZ |
2902 | if (waitqueue_active(&log_root_tree->log_commit_wait[index2])) |
2903 | wake_up(&log_root_tree->log_commit_wait[index2]); | |
e02119d5 | 2904 | out: |
8b050d35 MX |
2905 | /* See above. */ |
2906 | btrfs_remove_all_log_ctxs(root, index1, ret); | |
2907 | ||
d1433deb MX |
2908 | mutex_lock(&root->log_mutex); |
2909 | root->log_transid_committed++; | |
7237f183 | 2910 | atomic_set(&root->log_commit[index1], 0); |
d1433deb | 2911 | mutex_unlock(&root->log_mutex); |
8b050d35 | 2912 | |
7237f183 YZ |
2913 | if (waitqueue_active(&root->log_commit_wait[index1])) |
2914 | wake_up(&root->log_commit_wait[index1]); | |
b31eabd8 | 2915 | return ret; |
e02119d5 CM |
2916 | } |
2917 | ||
4a500fd1 YZ |
2918 | static void free_log_tree(struct btrfs_trans_handle *trans, |
2919 | struct btrfs_root *log) | |
e02119d5 CM |
2920 | { |
2921 | int ret; | |
d0c803c4 CM |
2922 | u64 start; |
2923 | u64 end; | |
e02119d5 CM |
2924 | struct walk_control wc = { |
2925 | .free = 1, | |
2926 | .process_func = process_one_buffer | |
2927 | }; | |
2928 | ||
681ae509 JB |
2929 | ret = walk_log_tree(trans, log, &wc); |
2930 | /* I don't think this can happen but just in case */ | |
2931 | if (ret) | |
2932 | btrfs_abort_transaction(trans, log, ret); | |
e02119d5 | 2933 | |
d397712b | 2934 | while (1) { |
d0c803c4 | 2935 | ret = find_first_extent_bit(&log->dirty_log_pages, |
e6138876 JB |
2936 | 0, &start, &end, EXTENT_DIRTY | EXTENT_NEW, |
2937 | NULL); | |
d0c803c4 CM |
2938 | if (ret) |
2939 | break; | |
2940 | ||
8cef4e16 YZ |
2941 | clear_extent_bits(&log->dirty_log_pages, start, end, |
2942 | EXTENT_DIRTY | EXTENT_NEW, GFP_NOFS); | |
d0c803c4 CM |
2943 | } |
2944 | ||
2ab28f32 JB |
2945 | /* |
2946 | * We may have short-circuited the log tree with the full commit logic | |
2947 | * and left ordered extents on our list, so clear these out to keep us | |
2948 | * from leaking inodes and memory. | |
2949 | */ | |
2950 | btrfs_free_logged_extents(log, 0); | |
2951 | btrfs_free_logged_extents(log, 1); | |
2952 | ||
7237f183 YZ |
2953 | free_extent_buffer(log->node); |
2954 | kfree(log); | |
4a500fd1 YZ |
2955 | } |
2956 | ||
2957 | /* | |
2958 | * free all the extents used by the tree log. This should be called | |
2959 | * at commit time of the full transaction | |
2960 | */ | |
2961 | int btrfs_free_log(struct btrfs_trans_handle *trans, struct btrfs_root *root) | |
2962 | { | |
2963 | if (root->log_root) { | |
2964 | free_log_tree(trans, root->log_root); | |
2965 | root->log_root = NULL; | |
2966 | } | |
2967 | return 0; | |
2968 | } | |
2969 | ||
2970 | int btrfs_free_log_root_tree(struct btrfs_trans_handle *trans, | |
2971 | struct btrfs_fs_info *fs_info) | |
2972 | { | |
2973 | if (fs_info->log_root_tree) { | |
2974 | free_log_tree(trans, fs_info->log_root_tree); | |
2975 | fs_info->log_root_tree = NULL; | |
2976 | } | |
e02119d5 CM |
2977 | return 0; |
2978 | } | |
2979 | ||
e02119d5 CM |
2980 | /* |
2981 | * If both a file and directory are logged, and unlinks or renames are | |
2982 | * mixed in, we have a few interesting corners: | |
2983 | * | |
2984 | * create file X in dir Y | |
2985 | * link file X to X.link in dir Y | |
2986 | * fsync file X | |
2987 | * unlink file X but leave X.link | |
2988 | * fsync dir Y | |
2989 | * | |
2990 | * After a crash we would expect only X.link to exist. But file X | |
2991 | * didn't get fsync'd again so the log has back refs for X and X.link. | |
2992 | * | |
2993 | * We solve this by removing directory entries and inode backrefs from the | |
2994 | * log when a file that was logged in the current transaction is | |
2995 | * unlinked. Any later fsync will include the updated log entries, and | |
2996 | * we'll be able to reconstruct the proper directory items from backrefs. | |
2997 | * | |
2998 | * This optimizations allows us to avoid relogging the entire inode | |
2999 | * or the entire directory. | |
3000 | */ | |
3001 | int btrfs_del_dir_entries_in_log(struct btrfs_trans_handle *trans, | |
3002 | struct btrfs_root *root, | |
3003 | const char *name, int name_len, | |
3004 | struct inode *dir, u64 index) | |
3005 | { | |
3006 | struct btrfs_root *log; | |
3007 | struct btrfs_dir_item *di; | |
3008 | struct btrfs_path *path; | |
3009 | int ret; | |
4a500fd1 | 3010 | int err = 0; |
e02119d5 | 3011 | int bytes_del = 0; |
33345d01 | 3012 | u64 dir_ino = btrfs_ino(dir); |
e02119d5 | 3013 | |
3a5f1d45 CM |
3014 | if (BTRFS_I(dir)->logged_trans < trans->transid) |
3015 | return 0; | |
3016 | ||
e02119d5 CM |
3017 | ret = join_running_log_trans(root); |
3018 | if (ret) | |
3019 | return 0; | |
3020 | ||
3021 | mutex_lock(&BTRFS_I(dir)->log_mutex); | |
3022 | ||
3023 | log = root->log_root; | |
3024 | path = btrfs_alloc_path(); | |
a62f44a5 TI |
3025 | if (!path) { |
3026 | err = -ENOMEM; | |
3027 | goto out_unlock; | |
3028 | } | |
2a29edc6 | 3029 | |
33345d01 | 3030 | di = btrfs_lookup_dir_item(trans, log, path, dir_ino, |
e02119d5 | 3031 | name, name_len, -1); |
4a500fd1 YZ |
3032 | if (IS_ERR(di)) { |
3033 | err = PTR_ERR(di); | |
3034 | goto fail; | |
3035 | } | |
3036 | if (di) { | |
e02119d5 CM |
3037 | ret = btrfs_delete_one_dir_name(trans, log, path, di); |
3038 | bytes_del += name_len; | |
3650860b JB |
3039 | if (ret) { |
3040 | err = ret; | |
3041 | goto fail; | |
3042 | } | |
e02119d5 | 3043 | } |
b3b4aa74 | 3044 | btrfs_release_path(path); |
33345d01 | 3045 | di = btrfs_lookup_dir_index_item(trans, log, path, dir_ino, |
e02119d5 | 3046 | index, name, name_len, -1); |
4a500fd1 YZ |
3047 | if (IS_ERR(di)) { |
3048 | err = PTR_ERR(di); | |
3049 | goto fail; | |
3050 | } | |
3051 | if (di) { | |
e02119d5 CM |
3052 | ret = btrfs_delete_one_dir_name(trans, log, path, di); |
3053 | bytes_del += name_len; | |
3650860b JB |
3054 | if (ret) { |
3055 | err = ret; | |
3056 | goto fail; | |
3057 | } | |
e02119d5 CM |
3058 | } |
3059 | ||
3060 | /* update the directory size in the log to reflect the names | |
3061 | * we have removed | |
3062 | */ | |
3063 | if (bytes_del) { | |
3064 | struct btrfs_key key; | |
3065 | ||
33345d01 | 3066 | key.objectid = dir_ino; |
e02119d5 CM |
3067 | key.offset = 0; |
3068 | key.type = BTRFS_INODE_ITEM_KEY; | |
b3b4aa74 | 3069 | btrfs_release_path(path); |
e02119d5 CM |
3070 | |
3071 | ret = btrfs_search_slot(trans, log, &key, path, 0, 1); | |
4a500fd1 YZ |
3072 | if (ret < 0) { |
3073 | err = ret; | |
3074 | goto fail; | |
3075 | } | |
e02119d5 CM |
3076 | if (ret == 0) { |
3077 | struct btrfs_inode_item *item; | |
3078 | u64 i_size; | |
3079 | ||
3080 | item = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
3081 | struct btrfs_inode_item); | |
3082 | i_size = btrfs_inode_size(path->nodes[0], item); | |
3083 | if (i_size > bytes_del) | |
3084 | i_size -= bytes_del; | |
3085 | else | |
3086 | i_size = 0; | |
3087 | btrfs_set_inode_size(path->nodes[0], item, i_size); | |
3088 | btrfs_mark_buffer_dirty(path->nodes[0]); | |
3089 | } else | |
3090 | ret = 0; | |
b3b4aa74 | 3091 | btrfs_release_path(path); |
e02119d5 | 3092 | } |
4a500fd1 | 3093 | fail: |
e02119d5 | 3094 | btrfs_free_path(path); |
a62f44a5 | 3095 | out_unlock: |
e02119d5 | 3096 | mutex_unlock(&BTRFS_I(dir)->log_mutex); |
4a500fd1 | 3097 | if (ret == -ENOSPC) { |
995946dd | 3098 | btrfs_set_log_full_commit(root->fs_info, trans); |
4a500fd1 | 3099 | ret = 0; |
79787eaa JM |
3100 | } else if (ret < 0) |
3101 | btrfs_abort_transaction(trans, root, ret); | |
3102 | ||
12fcfd22 | 3103 | btrfs_end_log_trans(root); |
e02119d5 | 3104 | |
411fc6bc | 3105 | return err; |
e02119d5 CM |
3106 | } |
3107 | ||
3108 | /* see comments for btrfs_del_dir_entries_in_log */ | |
3109 | int btrfs_del_inode_ref_in_log(struct btrfs_trans_handle *trans, | |
3110 | struct btrfs_root *root, | |
3111 | const char *name, int name_len, | |
3112 | struct inode *inode, u64 dirid) | |
3113 | { | |
3114 | struct btrfs_root *log; | |
3115 | u64 index; | |
3116 | int ret; | |
3117 | ||
3a5f1d45 CM |
3118 | if (BTRFS_I(inode)->logged_trans < trans->transid) |
3119 | return 0; | |
3120 | ||
e02119d5 CM |
3121 | ret = join_running_log_trans(root); |
3122 | if (ret) | |
3123 | return 0; | |
3124 | log = root->log_root; | |
3125 | mutex_lock(&BTRFS_I(inode)->log_mutex); | |
3126 | ||
33345d01 | 3127 | ret = btrfs_del_inode_ref(trans, log, name, name_len, btrfs_ino(inode), |
e02119d5 CM |
3128 | dirid, &index); |
3129 | mutex_unlock(&BTRFS_I(inode)->log_mutex); | |
4a500fd1 | 3130 | if (ret == -ENOSPC) { |
995946dd | 3131 | btrfs_set_log_full_commit(root->fs_info, trans); |
4a500fd1 | 3132 | ret = 0; |
79787eaa JM |
3133 | } else if (ret < 0 && ret != -ENOENT) |
3134 | btrfs_abort_transaction(trans, root, ret); | |
12fcfd22 | 3135 | btrfs_end_log_trans(root); |
e02119d5 | 3136 | |
e02119d5 CM |
3137 | return ret; |
3138 | } | |
3139 | ||
3140 | /* | |
3141 | * creates a range item in the log for 'dirid'. first_offset and | |
3142 | * last_offset tell us which parts of the key space the log should | |
3143 | * be considered authoritative for. | |
3144 | */ | |
3145 | static noinline int insert_dir_log_key(struct btrfs_trans_handle *trans, | |
3146 | struct btrfs_root *log, | |
3147 | struct btrfs_path *path, | |
3148 | int key_type, u64 dirid, | |
3149 | u64 first_offset, u64 last_offset) | |
3150 | { | |
3151 | int ret; | |
3152 | struct btrfs_key key; | |
3153 | struct btrfs_dir_log_item *item; | |
3154 | ||
3155 | key.objectid = dirid; | |
3156 | key.offset = first_offset; | |
3157 | if (key_type == BTRFS_DIR_ITEM_KEY) | |
3158 | key.type = BTRFS_DIR_LOG_ITEM_KEY; | |
3159 | else | |
3160 | key.type = BTRFS_DIR_LOG_INDEX_KEY; | |
3161 | ret = btrfs_insert_empty_item(trans, log, path, &key, sizeof(*item)); | |
4a500fd1 YZ |
3162 | if (ret) |
3163 | return ret; | |
e02119d5 CM |
3164 | |
3165 | item = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
3166 | struct btrfs_dir_log_item); | |
3167 | btrfs_set_dir_log_end(path->nodes[0], item, last_offset); | |
3168 | btrfs_mark_buffer_dirty(path->nodes[0]); | |
b3b4aa74 | 3169 | btrfs_release_path(path); |
e02119d5 CM |
3170 | return 0; |
3171 | } | |
3172 | ||
3173 | /* | |
3174 | * log all the items included in the current transaction for a given | |
3175 | * directory. This also creates the range items in the log tree required | |
3176 | * to replay anything deleted before the fsync | |
3177 | */ | |
3178 | static noinline int log_dir_items(struct btrfs_trans_handle *trans, | |
3179 | struct btrfs_root *root, struct inode *inode, | |
3180 | struct btrfs_path *path, | |
3181 | struct btrfs_path *dst_path, int key_type, | |
2f2ff0ee | 3182 | struct btrfs_log_ctx *ctx, |
e02119d5 CM |
3183 | u64 min_offset, u64 *last_offset_ret) |
3184 | { | |
3185 | struct btrfs_key min_key; | |
e02119d5 CM |
3186 | struct btrfs_root *log = root->log_root; |
3187 | struct extent_buffer *src; | |
4a500fd1 | 3188 | int err = 0; |
e02119d5 CM |
3189 | int ret; |
3190 | int i; | |
3191 | int nritems; | |
3192 | u64 first_offset = min_offset; | |
3193 | u64 last_offset = (u64)-1; | |
33345d01 | 3194 | u64 ino = btrfs_ino(inode); |
e02119d5 CM |
3195 | |
3196 | log = root->log_root; | |
e02119d5 | 3197 | |
33345d01 | 3198 | min_key.objectid = ino; |
e02119d5 CM |
3199 | min_key.type = key_type; |
3200 | min_key.offset = min_offset; | |
3201 | ||
6174d3cb | 3202 | ret = btrfs_search_forward(root, &min_key, path, trans->transid); |
e02119d5 CM |
3203 | |
3204 | /* | |
3205 | * we didn't find anything from this transaction, see if there | |
3206 | * is anything at all | |
3207 | */ | |
33345d01 LZ |
3208 | if (ret != 0 || min_key.objectid != ino || min_key.type != key_type) { |
3209 | min_key.objectid = ino; | |
e02119d5 CM |
3210 | min_key.type = key_type; |
3211 | min_key.offset = (u64)-1; | |
b3b4aa74 | 3212 | btrfs_release_path(path); |
e02119d5 CM |
3213 | ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0); |
3214 | if (ret < 0) { | |
b3b4aa74 | 3215 | btrfs_release_path(path); |
e02119d5 CM |
3216 | return ret; |
3217 | } | |
33345d01 | 3218 | ret = btrfs_previous_item(root, path, ino, key_type); |
e02119d5 CM |
3219 | |
3220 | /* if ret == 0 there are items for this type, | |
3221 | * create a range to tell us the last key of this type. | |
3222 | * otherwise, there are no items in this directory after | |
3223 | * *min_offset, and we create a range to indicate that. | |
3224 | */ | |
3225 | if (ret == 0) { | |
3226 | struct btrfs_key tmp; | |
3227 | btrfs_item_key_to_cpu(path->nodes[0], &tmp, | |
3228 | path->slots[0]); | |
d397712b | 3229 | if (key_type == tmp.type) |
e02119d5 | 3230 | first_offset = max(min_offset, tmp.offset) + 1; |
e02119d5 CM |
3231 | } |
3232 | goto done; | |
3233 | } | |
3234 | ||
3235 | /* go backward to find any previous key */ | |
33345d01 | 3236 | ret = btrfs_previous_item(root, path, ino, key_type); |
e02119d5 CM |
3237 | if (ret == 0) { |
3238 | struct btrfs_key tmp; | |
3239 | btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]); | |
3240 | if (key_type == tmp.type) { | |
3241 | first_offset = tmp.offset; | |
3242 | ret = overwrite_item(trans, log, dst_path, | |
3243 | path->nodes[0], path->slots[0], | |
3244 | &tmp); | |
4a500fd1 YZ |
3245 | if (ret) { |
3246 | err = ret; | |
3247 | goto done; | |
3248 | } | |
e02119d5 CM |
3249 | } |
3250 | } | |
b3b4aa74 | 3251 | btrfs_release_path(path); |
e02119d5 CM |
3252 | |
3253 | /* find the first key from this transaction again */ | |
3254 | ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0); | |
fae7f21c | 3255 | if (WARN_ON(ret != 0)) |
e02119d5 | 3256 | goto done; |
e02119d5 CM |
3257 | |
3258 | /* | |
3259 | * we have a block from this transaction, log every item in it | |
3260 | * from our directory | |
3261 | */ | |
d397712b | 3262 | while (1) { |
e02119d5 CM |
3263 | struct btrfs_key tmp; |
3264 | src = path->nodes[0]; | |
3265 | nritems = btrfs_header_nritems(src); | |
3266 | for (i = path->slots[0]; i < nritems; i++) { | |
2f2ff0ee FM |
3267 | struct btrfs_dir_item *di; |
3268 | ||
e02119d5 CM |
3269 | btrfs_item_key_to_cpu(src, &min_key, i); |
3270 | ||
33345d01 | 3271 | if (min_key.objectid != ino || min_key.type != key_type) |
e02119d5 CM |
3272 | goto done; |
3273 | ret = overwrite_item(trans, log, dst_path, src, i, | |
3274 | &min_key); | |
4a500fd1 YZ |
3275 | if (ret) { |
3276 | err = ret; | |
3277 | goto done; | |
3278 | } | |
2f2ff0ee FM |
3279 | |
3280 | /* | |
3281 | * We must make sure that when we log a directory entry, | |
3282 | * the corresponding inode, after log replay, has a | |
3283 | * matching link count. For example: | |
3284 | * | |
3285 | * touch foo | |
3286 | * mkdir mydir | |
3287 | * sync | |
3288 | * ln foo mydir/bar | |
3289 | * xfs_io -c "fsync" mydir | |
3290 | * <crash> | |
3291 | * <mount fs and log replay> | |
3292 | * | |
3293 | * Would result in a fsync log that when replayed, our | |
3294 | * file inode would have a link count of 1, but we get | |
3295 | * two directory entries pointing to the same inode. | |
3296 | * After removing one of the names, it would not be | |
3297 | * possible to remove the other name, which resulted | |
3298 | * always in stale file handle errors, and would not | |
3299 | * be possible to rmdir the parent directory, since | |
3300 | * its i_size could never decrement to the value | |
3301 | * BTRFS_EMPTY_DIR_SIZE, resulting in -ENOTEMPTY errors. | |
3302 | */ | |
3303 | di = btrfs_item_ptr(src, i, struct btrfs_dir_item); | |
3304 | btrfs_dir_item_key_to_cpu(src, di, &tmp); | |
3305 | if (ctx && | |
3306 | (btrfs_dir_transid(src, di) == trans->transid || | |
3307 | btrfs_dir_type(src, di) == BTRFS_FT_DIR) && | |
3308 | tmp.type != BTRFS_ROOT_ITEM_KEY) | |
3309 | ctx->log_new_dentries = true; | |
e02119d5 CM |
3310 | } |
3311 | path->slots[0] = nritems; | |
3312 | ||
3313 | /* | |
3314 | * look ahead to the next item and see if it is also | |
3315 | * from this directory and from this transaction | |
3316 | */ | |
3317 | ret = btrfs_next_leaf(root, path); | |
3318 | if (ret == 1) { | |
3319 | last_offset = (u64)-1; | |
3320 | goto done; | |
3321 | } | |
3322 | btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]); | |
33345d01 | 3323 | if (tmp.objectid != ino || tmp.type != key_type) { |
e02119d5 CM |
3324 | last_offset = (u64)-1; |
3325 | goto done; | |
3326 | } | |
3327 | if (btrfs_header_generation(path->nodes[0]) != trans->transid) { | |
3328 | ret = overwrite_item(trans, log, dst_path, | |
3329 | path->nodes[0], path->slots[0], | |
3330 | &tmp); | |
4a500fd1 YZ |
3331 | if (ret) |
3332 | err = ret; | |
3333 | else | |
3334 | last_offset = tmp.offset; | |
e02119d5 CM |
3335 | goto done; |
3336 | } | |
3337 | } | |
3338 | done: | |
b3b4aa74 DS |
3339 | btrfs_release_path(path); |
3340 | btrfs_release_path(dst_path); | |
e02119d5 | 3341 | |
4a500fd1 YZ |
3342 | if (err == 0) { |
3343 | *last_offset_ret = last_offset; | |
3344 | /* | |
3345 | * insert the log range keys to indicate where the log | |
3346 | * is valid | |
3347 | */ | |
3348 | ret = insert_dir_log_key(trans, log, path, key_type, | |
33345d01 | 3349 | ino, first_offset, last_offset); |
4a500fd1 YZ |
3350 | if (ret) |
3351 | err = ret; | |
3352 | } | |
3353 | return err; | |
e02119d5 CM |
3354 | } |
3355 | ||
3356 | /* | |
3357 | * logging directories is very similar to logging inodes, We find all the items | |
3358 | * from the current transaction and write them to the log. | |
3359 | * | |
3360 | * The recovery code scans the directory in the subvolume, and if it finds a | |
3361 | * key in the range logged that is not present in the log tree, then it means | |
3362 | * that dir entry was unlinked during the transaction. | |
3363 | * | |
3364 | * In order for that scan to work, we must include one key smaller than | |
3365 | * the smallest logged by this transaction and one key larger than the largest | |
3366 | * key logged by this transaction. | |
3367 | */ | |
3368 | static noinline int log_directory_changes(struct btrfs_trans_handle *trans, | |
3369 | struct btrfs_root *root, struct inode *inode, | |
3370 | struct btrfs_path *path, | |
2f2ff0ee FM |
3371 | struct btrfs_path *dst_path, |
3372 | struct btrfs_log_ctx *ctx) | |
e02119d5 CM |
3373 | { |
3374 | u64 min_key; | |
3375 | u64 max_key; | |
3376 | int ret; | |
3377 | int key_type = BTRFS_DIR_ITEM_KEY; | |
3378 | ||
3379 | again: | |
3380 | min_key = 0; | |
3381 | max_key = 0; | |
d397712b | 3382 | while (1) { |
e02119d5 | 3383 | ret = log_dir_items(trans, root, inode, path, |
2f2ff0ee | 3384 | dst_path, key_type, ctx, min_key, |
e02119d5 | 3385 | &max_key); |
4a500fd1 YZ |
3386 | if (ret) |
3387 | return ret; | |
e02119d5 CM |
3388 | if (max_key == (u64)-1) |
3389 | break; | |
3390 | min_key = max_key + 1; | |
3391 | } | |
3392 | ||
3393 | if (key_type == BTRFS_DIR_ITEM_KEY) { | |
3394 | key_type = BTRFS_DIR_INDEX_KEY; | |
3395 | goto again; | |
3396 | } | |
3397 | return 0; | |
3398 | } | |
3399 | ||
3400 | /* | |
3401 | * a helper function to drop items from the log before we relog an | |
3402 | * inode. max_key_type indicates the highest item type to remove. | |
3403 | * This cannot be run for file data extents because it does not | |
3404 | * free the extents they point to. | |
3405 | */ | |
3406 | static int drop_objectid_items(struct btrfs_trans_handle *trans, | |
3407 | struct btrfs_root *log, | |
3408 | struct btrfs_path *path, | |
3409 | u64 objectid, int max_key_type) | |
3410 | { | |
3411 | int ret; | |
3412 | struct btrfs_key key; | |
3413 | struct btrfs_key found_key; | |
18ec90d6 | 3414 | int start_slot; |
e02119d5 CM |
3415 | |
3416 | key.objectid = objectid; | |
3417 | key.type = max_key_type; | |
3418 | key.offset = (u64)-1; | |
3419 | ||
d397712b | 3420 | while (1) { |
e02119d5 | 3421 | ret = btrfs_search_slot(trans, log, &key, path, -1, 1); |
3650860b | 3422 | BUG_ON(ret == 0); /* Logic error */ |
4a500fd1 | 3423 | if (ret < 0) |
e02119d5 CM |
3424 | break; |
3425 | ||
3426 | if (path->slots[0] == 0) | |
3427 | break; | |
3428 | ||
3429 | path->slots[0]--; | |
3430 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, | |
3431 | path->slots[0]); | |
3432 | ||
3433 | if (found_key.objectid != objectid) | |
3434 | break; | |
3435 | ||
18ec90d6 JB |
3436 | found_key.offset = 0; |
3437 | found_key.type = 0; | |
3438 | ret = btrfs_bin_search(path->nodes[0], &found_key, 0, | |
3439 | &start_slot); | |
3440 | ||
3441 | ret = btrfs_del_items(trans, log, path, start_slot, | |
3442 | path->slots[0] - start_slot + 1); | |
3443 | /* | |
3444 | * If start slot isn't 0 then we don't need to re-search, we've | |
3445 | * found the last guy with the objectid in this tree. | |
3446 | */ | |
3447 | if (ret || start_slot != 0) | |
65a246c5 | 3448 | break; |
b3b4aa74 | 3449 | btrfs_release_path(path); |
e02119d5 | 3450 | } |
b3b4aa74 | 3451 | btrfs_release_path(path); |
5bdbeb21 JB |
3452 | if (ret > 0) |
3453 | ret = 0; | |
4a500fd1 | 3454 | return ret; |
e02119d5 CM |
3455 | } |
3456 | ||
94edf4ae JB |
3457 | static void fill_inode_item(struct btrfs_trans_handle *trans, |
3458 | struct extent_buffer *leaf, | |
3459 | struct btrfs_inode_item *item, | |
1a4bcf47 FM |
3460 | struct inode *inode, int log_inode_only, |
3461 | u64 logged_isize) | |
94edf4ae | 3462 | { |
0b1c6cca JB |
3463 | struct btrfs_map_token token; |
3464 | ||
3465 | btrfs_init_map_token(&token); | |
94edf4ae JB |
3466 | |
3467 | if (log_inode_only) { | |
3468 | /* set the generation to zero so the recover code | |
3469 | * can tell the difference between an logging | |
3470 | * just to say 'this inode exists' and a logging | |
3471 | * to say 'update this inode with these values' | |
3472 | */ | |
0b1c6cca | 3473 | btrfs_set_token_inode_generation(leaf, item, 0, &token); |
1a4bcf47 | 3474 | btrfs_set_token_inode_size(leaf, item, logged_isize, &token); |
94edf4ae | 3475 | } else { |
0b1c6cca JB |
3476 | btrfs_set_token_inode_generation(leaf, item, |
3477 | BTRFS_I(inode)->generation, | |
3478 | &token); | |
3479 | btrfs_set_token_inode_size(leaf, item, inode->i_size, &token); | |
3480 | } | |
3481 | ||
3482 | btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token); | |
3483 | btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token); | |
3484 | btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token); | |
3485 | btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token); | |
3486 | ||
a937b979 | 3487 | btrfs_set_token_timespec_sec(leaf, &item->atime, |
0b1c6cca | 3488 | inode->i_atime.tv_sec, &token); |
a937b979 | 3489 | btrfs_set_token_timespec_nsec(leaf, &item->atime, |
0b1c6cca JB |
3490 | inode->i_atime.tv_nsec, &token); |
3491 | ||
a937b979 | 3492 | btrfs_set_token_timespec_sec(leaf, &item->mtime, |
0b1c6cca | 3493 | inode->i_mtime.tv_sec, &token); |
a937b979 | 3494 | btrfs_set_token_timespec_nsec(leaf, &item->mtime, |
0b1c6cca JB |
3495 | inode->i_mtime.tv_nsec, &token); |
3496 | ||
a937b979 | 3497 | btrfs_set_token_timespec_sec(leaf, &item->ctime, |
0b1c6cca | 3498 | inode->i_ctime.tv_sec, &token); |
a937b979 | 3499 | btrfs_set_token_timespec_nsec(leaf, &item->ctime, |
0b1c6cca JB |
3500 | inode->i_ctime.tv_nsec, &token); |
3501 | ||
3502 | btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode), | |
3503 | &token); | |
3504 | ||
3505 | btrfs_set_token_inode_sequence(leaf, item, inode->i_version, &token); | |
3506 | btrfs_set_token_inode_transid(leaf, item, trans->transid, &token); | |
3507 | btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token); | |
3508 | btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token); | |
3509 | btrfs_set_token_inode_block_group(leaf, item, 0, &token); | |
94edf4ae JB |
3510 | } |
3511 | ||
a95249b3 JB |
3512 | static int log_inode_item(struct btrfs_trans_handle *trans, |
3513 | struct btrfs_root *log, struct btrfs_path *path, | |
3514 | struct inode *inode) | |
3515 | { | |
3516 | struct btrfs_inode_item *inode_item; | |
a95249b3 JB |
3517 | int ret; |
3518 | ||
efd0c405 FDBM |
3519 | ret = btrfs_insert_empty_item(trans, log, path, |
3520 | &BTRFS_I(inode)->location, | |
a95249b3 JB |
3521 | sizeof(*inode_item)); |
3522 | if (ret && ret != -EEXIST) | |
3523 | return ret; | |
3524 | inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
3525 | struct btrfs_inode_item); | |
1a4bcf47 | 3526 | fill_inode_item(trans, path->nodes[0], inode_item, inode, 0, 0); |
a95249b3 JB |
3527 | btrfs_release_path(path); |
3528 | return 0; | |
3529 | } | |
3530 | ||
31ff1cd2 | 3531 | static noinline int copy_items(struct btrfs_trans_handle *trans, |
d2794405 | 3532 | struct inode *inode, |
31ff1cd2 | 3533 | struct btrfs_path *dst_path, |
16e7549f | 3534 | struct btrfs_path *src_path, u64 *last_extent, |
1a4bcf47 FM |
3535 | int start_slot, int nr, int inode_only, |
3536 | u64 logged_isize) | |
31ff1cd2 CM |
3537 | { |
3538 | unsigned long src_offset; | |
3539 | unsigned long dst_offset; | |
d2794405 | 3540 | struct btrfs_root *log = BTRFS_I(inode)->root->log_root; |
31ff1cd2 CM |
3541 | struct btrfs_file_extent_item *extent; |
3542 | struct btrfs_inode_item *inode_item; | |
16e7549f JB |
3543 | struct extent_buffer *src = src_path->nodes[0]; |
3544 | struct btrfs_key first_key, last_key, key; | |
31ff1cd2 CM |
3545 | int ret; |
3546 | struct btrfs_key *ins_keys; | |
3547 | u32 *ins_sizes; | |
3548 | char *ins_data; | |
3549 | int i; | |
d20f7043 | 3550 | struct list_head ordered_sums; |
d2794405 | 3551 | int skip_csum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM; |
16e7549f | 3552 | bool has_extents = false; |
74121f7c | 3553 | bool need_find_last_extent = true; |
16e7549f | 3554 | bool done = false; |
d20f7043 CM |
3555 | |
3556 | INIT_LIST_HEAD(&ordered_sums); | |
31ff1cd2 CM |
3557 | |
3558 | ins_data = kmalloc(nr * sizeof(struct btrfs_key) + | |
3559 | nr * sizeof(u32), GFP_NOFS); | |
2a29edc6 | 3560 | if (!ins_data) |
3561 | return -ENOMEM; | |
3562 | ||
16e7549f JB |
3563 | first_key.objectid = (u64)-1; |
3564 | ||
31ff1cd2 CM |
3565 | ins_sizes = (u32 *)ins_data; |
3566 | ins_keys = (struct btrfs_key *)(ins_data + nr * sizeof(u32)); | |
3567 | ||
3568 | for (i = 0; i < nr; i++) { | |
3569 | ins_sizes[i] = btrfs_item_size_nr(src, i + start_slot); | |
3570 | btrfs_item_key_to_cpu(src, ins_keys + i, i + start_slot); | |
3571 | } | |
3572 | ret = btrfs_insert_empty_items(trans, log, dst_path, | |
3573 | ins_keys, ins_sizes, nr); | |
4a500fd1 YZ |
3574 | if (ret) { |
3575 | kfree(ins_data); | |
3576 | return ret; | |
3577 | } | |
31ff1cd2 | 3578 | |
5d4f98a2 | 3579 | for (i = 0; i < nr; i++, dst_path->slots[0]++) { |
31ff1cd2 CM |
3580 | dst_offset = btrfs_item_ptr_offset(dst_path->nodes[0], |
3581 | dst_path->slots[0]); | |
3582 | ||
3583 | src_offset = btrfs_item_ptr_offset(src, start_slot + i); | |
3584 | ||
16e7549f JB |
3585 | if ((i == (nr - 1))) |
3586 | last_key = ins_keys[i]; | |
3587 | ||
94edf4ae | 3588 | if (ins_keys[i].type == BTRFS_INODE_ITEM_KEY) { |
31ff1cd2 CM |
3589 | inode_item = btrfs_item_ptr(dst_path->nodes[0], |
3590 | dst_path->slots[0], | |
3591 | struct btrfs_inode_item); | |
94edf4ae | 3592 | fill_inode_item(trans, dst_path->nodes[0], inode_item, |
1a4bcf47 FM |
3593 | inode, inode_only == LOG_INODE_EXISTS, |
3594 | logged_isize); | |
94edf4ae JB |
3595 | } else { |
3596 | copy_extent_buffer(dst_path->nodes[0], src, dst_offset, | |
3597 | src_offset, ins_sizes[i]); | |
31ff1cd2 | 3598 | } |
94edf4ae | 3599 | |
16e7549f JB |
3600 | /* |
3601 | * We set need_find_last_extent here in case we know we were | |
3602 | * processing other items and then walk into the first extent in | |
3603 | * the inode. If we don't hit an extent then nothing changes, | |
3604 | * we'll do the last search the next time around. | |
3605 | */ | |
3606 | if (ins_keys[i].type == BTRFS_EXTENT_DATA_KEY) { | |
3607 | has_extents = true; | |
74121f7c | 3608 | if (first_key.objectid == (u64)-1) |
16e7549f JB |
3609 | first_key = ins_keys[i]; |
3610 | } else { | |
3611 | need_find_last_extent = false; | |
3612 | } | |
3613 | ||
31ff1cd2 CM |
3614 | /* take a reference on file data extents so that truncates |
3615 | * or deletes of this inode don't have to relog the inode | |
3616 | * again | |
3617 | */ | |
962a298f | 3618 | if (ins_keys[i].type == BTRFS_EXTENT_DATA_KEY && |
d2794405 | 3619 | !skip_csum) { |
31ff1cd2 CM |
3620 | int found_type; |
3621 | extent = btrfs_item_ptr(src, start_slot + i, | |
3622 | struct btrfs_file_extent_item); | |
3623 | ||
8e531cdf | 3624 | if (btrfs_file_extent_generation(src, extent) < trans->transid) |
3625 | continue; | |
3626 | ||
31ff1cd2 | 3627 | found_type = btrfs_file_extent_type(src, extent); |
6f1fed77 | 3628 | if (found_type == BTRFS_FILE_EXTENT_REG) { |
5d4f98a2 YZ |
3629 | u64 ds, dl, cs, cl; |
3630 | ds = btrfs_file_extent_disk_bytenr(src, | |
3631 | extent); | |
3632 | /* ds == 0 is a hole */ | |
3633 | if (ds == 0) | |
3634 | continue; | |
3635 | ||
3636 | dl = btrfs_file_extent_disk_num_bytes(src, | |
3637 | extent); | |
3638 | cs = btrfs_file_extent_offset(src, extent); | |
3639 | cl = btrfs_file_extent_num_bytes(src, | |
a419aef8 | 3640 | extent); |
580afd76 CM |
3641 | if (btrfs_file_extent_compression(src, |
3642 | extent)) { | |
3643 | cs = 0; | |
3644 | cl = dl; | |
3645 | } | |
5d4f98a2 YZ |
3646 | |
3647 | ret = btrfs_lookup_csums_range( | |
3648 | log->fs_info->csum_root, | |
3649 | ds + cs, ds + cs + cl - 1, | |
a2de733c | 3650 | &ordered_sums, 0); |
3650860b JB |
3651 | if (ret) { |
3652 | btrfs_release_path(dst_path); | |
3653 | kfree(ins_data); | |
3654 | return ret; | |
3655 | } | |
31ff1cd2 CM |
3656 | } |
3657 | } | |
31ff1cd2 CM |
3658 | } |
3659 | ||
3660 | btrfs_mark_buffer_dirty(dst_path->nodes[0]); | |
b3b4aa74 | 3661 | btrfs_release_path(dst_path); |
31ff1cd2 | 3662 | kfree(ins_data); |
d20f7043 CM |
3663 | |
3664 | /* | |
3665 | * we have to do this after the loop above to avoid changing the | |
3666 | * log tree while trying to change the log tree. | |
3667 | */ | |
4a500fd1 | 3668 | ret = 0; |
d397712b | 3669 | while (!list_empty(&ordered_sums)) { |
d20f7043 CM |
3670 | struct btrfs_ordered_sum *sums = list_entry(ordered_sums.next, |
3671 | struct btrfs_ordered_sum, | |
3672 | list); | |
4a500fd1 YZ |
3673 | if (!ret) |
3674 | ret = btrfs_csum_file_blocks(trans, log, sums); | |
d20f7043 CM |
3675 | list_del(&sums->list); |
3676 | kfree(sums); | |
3677 | } | |
16e7549f JB |
3678 | |
3679 | if (!has_extents) | |
3680 | return ret; | |
3681 | ||
74121f7c FM |
3682 | if (need_find_last_extent && *last_extent == first_key.offset) { |
3683 | /* | |
3684 | * We don't have any leafs between our current one and the one | |
3685 | * we processed before that can have file extent items for our | |
3686 | * inode (and have a generation number smaller than our current | |
3687 | * transaction id). | |
3688 | */ | |
3689 | need_find_last_extent = false; | |
3690 | } | |
3691 | ||
16e7549f JB |
3692 | /* |
3693 | * Because we use btrfs_search_forward we could skip leaves that were | |
3694 | * not modified and then assume *last_extent is valid when it really | |
3695 | * isn't. So back up to the previous leaf and read the end of the last | |
3696 | * extent before we go and fill in holes. | |
3697 | */ | |
3698 | if (need_find_last_extent) { | |
3699 | u64 len; | |
3700 | ||
3701 | ret = btrfs_prev_leaf(BTRFS_I(inode)->root, src_path); | |
3702 | if (ret < 0) | |
3703 | return ret; | |
3704 | if (ret) | |
3705 | goto fill_holes; | |
3706 | if (src_path->slots[0]) | |
3707 | src_path->slots[0]--; | |
3708 | src = src_path->nodes[0]; | |
3709 | btrfs_item_key_to_cpu(src, &key, src_path->slots[0]); | |
3710 | if (key.objectid != btrfs_ino(inode) || | |
3711 | key.type != BTRFS_EXTENT_DATA_KEY) | |
3712 | goto fill_holes; | |
3713 | extent = btrfs_item_ptr(src, src_path->slots[0], | |
3714 | struct btrfs_file_extent_item); | |
3715 | if (btrfs_file_extent_type(src, extent) == | |
3716 | BTRFS_FILE_EXTENT_INLINE) { | |
514ac8ad CM |
3717 | len = btrfs_file_extent_inline_len(src, |
3718 | src_path->slots[0], | |
3719 | extent); | |
16e7549f JB |
3720 | *last_extent = ALIGN(key.offset + len, |
3721 | log->sectorsize); | |
3722 | } else { | |
3723 | len = btrfs_file_extent_num_bytes(src, extent); | |
3724 | *last_extent = key.offset + len; | |
3725 | } | |
3726 | } | |
3727 | fill_holes: | |
3728 | /* So we did prev_leaf, now we need to move to the next leaf, but a few | |
3729 | * things could have happened | |
3730 | * | |
3731 | * 1) A merge could have happened, so we could currently be on a leaf | |
3732 | * that holds what we were copying in the first place. | |
3733 | * 2) A split could have happened, and now not all of the items we want | |
3734 | * are on the same leaf. | |
3735 | * | |
3736 | * So we need to adjust how we search for holes, we need to drop the | |
3737 | * path and re-search for the first extent key we found, and then walk | |
3738 | * forward until we hit the last one we copied. | |
3739 | */ | |
3740 | if (need_find_last_extent) { | |
3741 | /* btrfs_prev_leaf could return 1 without releasing the path */ | |
3742 | btrfs_release_path(src_path); | |
3743 | ret = btrfs_search_slot(NULL, BTRFS_I(inode)->root, &first_key, | |
3744 | src_path, 0, 0); | |
3745 | if (ret < 0) | |
3746 | return ret; | |
3747 | ASSERT(ret == 0); | |
3748 | src = src_path->nodes[0]; | |
3749 | i = src_path->slots[0]; | |
3750 | } else { | |
3751 | i = start_slot; | |
3752 | } | |
3753 | ||
3754 | /* | |
3755 | * Ok so here we need to go through and fill in any holes we may have | |
3756 | * to make sure that holes are punched for those areas in case they had | |
3757 | * extents previously. | |
3758 | */ | |
3759 | while (!done) { | |
3760 | u64 offset, len; | |
3761 | u64 extent_end; | |
3762 | ||
3763 | if (i >= btrfs_header_nritems(src_path->nodes[0])) { | |
3764 | ret = btrfs_next_leaf(BTRFS_I(inode)->root, src_path); | |
3765 | if (ret < 0) | |
3766 | return ret; | |
3767 | ASSERT(ret == 0); | |
3768 | src = src_path->nodes[0]; | |
3769 | i = 0; | |
3770 | } | |
3771 | ||
3772 | btrfs_item_key_to_cpu(src, &key, i); | |
3773 | if (!btrfs_comp_cpu_keys(&key, &last_key)) | |
3774 | done = true; | |
3775 | if (key.objectid != btrfs_ino(inode) || | |
3776 | key.type != BTRFS_EXTENT_DATA_KEY) { | |
3777 | i++; | |
3778 | continue; | |
3779 | } | |
3780 | extent = btrfs_item_ptr(src, i, struct btrfs_file_extent_item); | |
3781 | if (btrfs_file_extent_type(src, extent) == | |
3782 | BTRFS_FILE_EXTENT_INLINE) { | |
514ac8ad | 3783 | len = btrfs_file_extent_inline_len(src, i, extent); |
16e7549f JB |
3784 | extent_end = ALIGN(key.offset + len, log->sectorsize); |
3785 | } else { | |
3786 | len = btrfs_file_extent_num_bytes(src, extent); | |
3787 | extent_end = key.offset + len; | |
3788 | } | |
3789 | i++; | |
3790 | ||
3791 | if (*last_extent == key.offset) { | |
3792 | *last_extent = extent_end; | |
3793 | continue; | |
3794 | } | |
3795 | offset = *last_extent; | |
3796 | len = key.offset - *last_extent; | |
3797 | ret = btrfs_insert_file_extent(trans, log, btrfs_ino(inode), | |
3798 | offset, 0, 0, len, 0, len, 0, | |
3799 | 0, 0); | |
3800 | if (ret) | |
3801 | break; | |
74121f7c | 3802 | *last_extent = extent_end; |
16e7549f JB |
3803 | } |
3804 | /* | |
3805 | * Need to let the callers know we dropped the path so they should | |
3806 | * re-search. | |
3807 | */ | |
3808 | if (!ret && need_find_last_extent) | |
3809 | ret = 1; | |
4a500fd1 | 3810 | return ret; |
31ff1cd2 CM |
3811 | } |
3812 | ||
5dc562c5 JB |
3813 | static int extent_cmp(void *priv, struct list_head *a, struct list_head *b) |
3814 | { | |
3815 | struct extent_map *em1, *em2; | |
3816 | ||
3817 | em1 = list_entry(a, struct extent_map, list); | |
3818 | em2 = list_entry(b, struct extent_map, list); | |
3819 | ||
3820 | if (em1->start < em2->start) | |
3821 | return -1; | |
3822 | else if (em1->start > em2->start) | |
3823 | return 1; | |
3824 | return 0; | |
3825 | } | |
3826 | ||
8407f553 FM |
3827 | static int wait_ordered_extents(struct btrfs_trans_handle *trans, |
3828 | struct inode *inode, | |
3829 | struct btrfs_root *root, | |
3830 | const struct extent_map *em, | |
3831 | const struct list_head *logged_list, | |
3832 | bool *ordered_io_error) | |
5dc562c5 | 3833 | { |
2ab28f32 | 3834 | struct btrfs_ordered_extent *ordered; |
8407f553 | 3835 | struct btrfs_root *log = root->log_root; |
2ab28f32 JB |
3836 | u64 mod_start = em->mod_start; |
3837 | u64 mod_len = em->mod_len; | |
8407f553 | 3838 | const bool skip_csum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM; |
2ab28f32 JB |
3839 | u64 csum_offset; |
3840 | u64 csum_len; | |
8407f553 FM |
3841 | LIST_HEAD(ordered_sums); |
3842 | int ret = 0; | |
0aa4a17d | 3843 | |
8407f553 | 3844 | *ordered_io_error = false; |
0aa4a17d | 3845 | |
8407f553 FM |
3846 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) || |
3847 | em->block_start == EXTENT_MAP_HOLE) | |
70c8a91c | 3848 | return 0; |
5dc562c5 | 3849 | |
2ab28f32 | 3850 | /* |
8407f553 FM |
3851 | * Wait far any ordered extent that covers our extent map. If it |
3852 | * finishes without an error, first check and see if our csums are on | |
3853 | * our outstanding ordered extents. | |
2ab28f32 | 3854 | */ |
827463c4 | 3855 | list_for_each_entry(ordered, logged_list, log_list) { |
2ab28f32 JB |
3856 | struct btrfs_ordered_sum *sum; |
3857 | ||
3858 | if (!mod_len) | |
3859 | break; | |
3860 | ||
2ab28f32 JB |
3861 | if (ordered->file_offset + ordered->len <= mod_start || |
3862 | mod_start + mod_len <= ordered->file_offset) | |
3863 | continue; | |
3864 | ||
8407f553 FM |
3865 | if (!test_bit(BTRFS_ORDERED_IO_DONE, &ordered->flags) && |
3866 | !test_bit(BTRFS_ORDERED_IOERR, &ordered->flags) && | |
3867 | !test_bit(BTRFS_ORDERED_DIRECT, &ordered->flags)) { | |
3868 | const u64 start = ordered->file_offset; | |
3869 | const u64 end = ordered->file_offset + ordered->len - 1; | |
3870 | ||
3871 | WARN_ON(ordered->inode != inode); | |
3872 | filemap_fdatawrite_range(inode->i_mapping, start, end); | |
3873 | } | |
3874 | ||
3875 | wait_event(ordered->wait, | |
3876 | (test_bit(BTRFS_ORDERED_IO_DONE, &ordered->flags) || | |
3877 | test_bit(BTRFS_ORDERED_IOERR, &ordered->flags))); | |
3878 | ||
3879 | if (test_bit(BTRFS_ORDERED_IOERR, &ordered->flags)) { | |
b38ef71c FM |
3880 | /* |
3881 | * Clear the AS_EIO/AS_ENOSPC flags from the inode's | |
3882 | * i_mapping flags, so that the next fsync won't get | |
3883 | * an outdated io error too. | |
3884 | */ | |
3885 | btrfs_inode_check_errors(inode); | |
8407f553 FM |
3886 | *ordered_io_error = true; |
3887 | break; | |
3888 | } | |
2ab28f32 JB |
3889 | /* |
3890 | * We are going to copy all the csums on this ordered extent, so | |
3891 | * go ahead and adjust mod_start and mod_len in case this | |
3892 | * ordered extent has already been logged. | |
3893 | */ | |
3894 | if (ordered->file_offset > mod_start) { | |
3895 | if (ordered->file_offset + ordered->len >= | |
3896 | mod_start + mod_len) | |
3897 | mod_len = ordered->file_offset - mod_start; | |
3898 | /* | |
3899 | * If we have this case | |
3900 | * | |
3901 | * |--------- logged extent ---------| | |
3902 | * |----- ordered extent ----| | |
3903 | * | |
3904 | * Just don't mess with mod_start and mod_len, we'll | |
3905 | * just end up logging more csums than we need and it | |
3906 | * will be ok. | |
3907 | */ | |
3908 | } else { | |
3909 | if (ordered->file_offset + ordered->len < | |
3910 | mod_start + mod_len) { | |
3911 | mod_len = (mod_start + mod_len) - | |
3912 | (ordered->file_offset + ordered->len); | |
3913 | mod_start = ordered->file_offset + | |
3914 | ordered->len; | |
3915 | } else { | |
3916 | mod_len = 0; | |
3917 | } | |
3918 | } | |
3919 | ||
8407f553 FM |
3920 | if (skip_csum) |
3921 | continue; | |
3922 | ||
2ab28f32 JB |
3923 | /* |
3924 | * To keep us from looping for the above case of an ordered | |
3925 | * extent that falls inside of the logged extent. | |
3926 | */ | |
3927 | if (test_and_set_bit(BTRFS_ORDERED_LOGGED_CSUM, | |
3928 | &ordered->flags)) | |
3929 | continue; | |
2ab28f32 | 3930 | |
2ab28f32 JB |
3931 | list_for_each_entry(sum, &ordered->list, list) { |
3932 | ret = btrfs_csum_file_blocks(trans, log, sum); | |
827463c4 | 3933 | if (ret) |
8407f553 | 3934 | break; |
2ab28f32 | 3935 | } |
2ab28f32 | 3936 | } |
2ab28f32 | 3937 | |
8407f553 | 3938 | if (*ordered_io_error || !mod_len || ret || skip_csum) |
2ab28f32 JB |
3939 | return ret; |
3940 | ||
488111aa FDBM |
3941 | if (em->compress_type) { |
3942 | csum_offset = 0; | |
8407f553 | 3943 | csum_len = max(em->block_len, em->orig_block_len); |
488111aa FDBM |
3944 | } else { |
3945 | csum_offset = mod_start - em->start; | |
3946 | csum_len = mod_len; | |
3947 | } | |
2ab28f32 | 3948 | |
70c8a91c JB |
3949 | /* block start is already adjusted for the file extent offset. */ |
3950 | ret = btrfs_lookup_csums_range(log->fs_info->csum_root, | |
3951 | em->block_start + csum_offset, | |
3952 | em->block_start + csum_offset + | |
3953 | csum_len - 1, &ordered_sums, 0); | |
3954 | if (ret) | |
3955 | return ret; | |
5dc562c5 | 3956 | |
70c8a91c JB |
3957 | while (!list_empty(&ordered_sums)) { |
3958 | struct btrfs_ordered_sum *sums = list_entry(ordered_sums.next, | |
3959 | struct btrfs_ordered_sum, | |
3960 | list); | |
3961 | if (!ret) | |
3962 | ret = btrfs_csum_file_blocks(trans, log, sums); | |
3963 | list_del(&sums->list); | |
3964 | kfree(sums); | |
5dc562c5 JB |
3965 | } |
3966 | ||
70c8a91c | 3967 | return ret; |
5dc562c5 JB |
3968 | } |
3969 | ||
8407f553 FM |
3970 | static int log_one_extent(struct btrfs_trans_handle *trans, |
3971 | struct inode *inode, struct btrfs_root *root, | |
3972 | const struct extent_map *em, | |
3973 | struct btrfs_path *path, | |
3974 | const struct list_head *logged_list, | |
3975 | struct btrfs_log_ctx *ctx) | |
3976 | { | |
3977 | struct btrfs_root *log = root->log_root; | |
3978 | struct btrfs_file_extent_item *fi; | |
3979 | struct extent_buffer *leaf; | |
3980 | struct btrfs_map_token token; | |
3981 | struct btrfs_key key; | |
3982 | u64 extent_offset = em->start - em->orig_start; | |
3983 | u64 block_len; | |
3984 | int ret; | |
3985 | int extent_inserted = 0; | |
3986 | bool ordered_io_err = false; | |
3987 | ||
3988 | ret = wait_ordered_extents(trans, inode, root, em, logged_list, | |
3989 | &ordered_io_err); | |
3990 | if (ret) | |
3991 | return ret; | |
3992 | ||
3993 | if (ordered_io_err) { | |
3994 | ctx->io_err = -EIO; | |
3995 | return 0; | |
3996 | } | |
3997 | ||
3998 | btrfs_init_map_token(&token); | |
3999 | ||
4000 | ret = __btrfs_drop_extents(trans, log, inode, path, em->start, | |
4001 | em->start + em->len, NULL, 0, 1, | |
4002 | sizeof(*fi), &extent_inserted); | |
4003 | if (ret) | |
4004 | return ret; | |
4005 | ||
4006 | if (!extent_inserted) { | |
4007 | key.objectid = btrfs_ino(inode); | |
4008 | key.type = BTRFS_EXTENT_DATA_KEY; | |
4009 | key.offset = em->start; | |
4010 | ||
4011 | ret = btrfs_insert_empty_item(trans, log, path, &key, | |
4012 | sizeof(*fi)); | |
4013 | if (ret) | |
4014 | return ret; | |
4015 | } | |
4016 | leaf = path->nodes[0]; | |
4017 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
4018 | struct btrfs_file_extent_item); | |
4019 | ||
50d9aa99 | 4020 | btrfs_set_token_file_extent_generation(leaf, fi, trans->transid, |
8407f553 FM |
4021 | &token); |
4022 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) | |
4023 | btrfs_set_token_file_extent_type(leaf, fi, | |
4024 | BTRFS_FILE_EXTENT_PREALLOC, | |
4025 | &token); | |
4026 | else | |
4027 | btrfs_set_token_file_extent_type(leaf, fi, | |
4028 | BTRFS_FILE_EXTENT_REG, | |
4029 | &token); | |
4030 | ||
4031 | block_len = max(em->block_len, em->orig_block_len); | |
4032 | if (em->compress_type != BTRFS_COMPRESS_NONE) { | |
4033 | btrfs_set_token_file_extent_disk_bytenr(leaf, fi, | |
4034 | em->block_start, | |
4035 | &token); | |
4036 | btrfs_set_token_file_extent_disk_num_bytes(leaf, fi, block_len, | |
4037 | &token); | |
4038 | } else if (em->block_start < EXTENT_MAP_LAST_BYTE) { | |
4039 | btrfs_set_token_file_extent_disk_bytenr(leaf, fi, | |
4040 | em->block_start - | |
4041 | extent_offset, &token); | |
4042 | btrfs_set_token_file_extent_disk_num_bytes(leaf, fi, block_len, | |
4043 | &token); | |
4044 | } else { | |
4045 | btrfs_set_token_file_extent_disk_bytenr(leaf, fi, 0, &token); | |
4046 | btrfs_set_token_file_extent_disk_num_bytes(leaf, fi, 0, | |
4047 | &token); | |
4048 | } | |
4049 | ||
4050 | btrfs_set_token_file_extent_offset(leaf, fi, extent_offset, &token); | |
4051 | btrfs_set_token_file_extent_num_bytes(leaf, fi, em->len, &token); | |
4052 | btrfs_set_token_file_extent_ram_bytes(leaf, fi, em->ram_bytes, &token); | |
4053 | btrfs_set_token_file_extent_compression(leaf, fi, em->compress_type, | |
4054 | &token); | |
4055 | btrfs_set_token_file_extent_encryption(leaf, fi, 0, &token); | |
4056 | btrfs_set_token_file_extent_other_encoding(leaf, fi, 0, &token); | |
4057 | btrfs_mark_buffer_dirty(leaf); | |
4058 | ||
4059 | btrfs_release_path(path); | |
4060 | ||
4061 | return ret; | |
4062 | } | |
4063 | ||
5dc562c5 JB |
4064 | static int btrfs_log_changed_extents(struct btrfs_trans_handle *trans, |
4065 | struct btrfs_root *root, | |
4066 | struct inode *inode, | |
827463c4 | 4067 | struct btrfs_path *path, |
8407f553 FM |
4068 | struct list_head *logged_list, |
4069 | struct btrfs_log_ctx *ctx) | |
5dc562c5 | 4070 | { |
5dc562c5 JB |
4071 | struct extent_map *em, *n; |
4072 | struct list_head extents; | |
4073 | struct extent_map_tree *tree = &BTRFS_I(inode)->extent_tree; | |
4074 | u64 test_gen; | |
4075 | int ret = 0; | |
2ab28f32 | 4076 | int num = 0; |
5dc562c5 JB |
4077 | |
4078 | INIT_LIST_HEAD(&extents); | |
4079 | ||
5dc562c5 JB |
4080 | write_lock(&tree->lock); |
4081 | test_gen = root->fs_info->last_trans_committed; | |
4082 | ||
4083 | list_for_each_entry_safe(em, n, &tree->modified_extents, list) { | |
4084 | list_del_init(&em->list); | |
2ab28f32 JB |
4085 | |
4086 | /* | |
4087 | * Just an arbitrary number, this can be really CPU intensive | |
4088 | * once we start getting a lot of extents, and really once we | |
4089 | * have a bunch of extents we just want to commit since it will | |
4090 | * be faster. | |
4091 | */ | |
4092 | if (++num > 32768) { | |
4093 | list_del_init(&tree->modified_extents); | |
4094 | ret = -EFBIG; | |
4095 | goto process; | |
4096 | } | |
4097 | ||
5dc562c5 JB |
4098 | if (em->generation <= test_gen) |
4099 | continue; | |
ff44c6e3 JB |
4100 | /* Need a ref to keep it from getting evicted from cache */ |
4101 | atomic_inc(&em->refs); | |
4102 | set_bit(EXTENT_FLAG_LOGGING, &em->flags); | |
5dc562c5 | 4103 | list_add_tail(&em->list, &extents); |
2ab28f32 | 4104 | num++; |
5dc562c5 JB |
4105 | } |
4106 | ||
4107 | list_sort(NULL, &extents, extent_cmp); | |
4108 | ||
2ab28f32 | 4109 | process: |
5dc562c5 JB |
4110 | while (!list_empty(&extents)) { |
4111 | em = list_entry(extents.next, struct extent_map, list); | |
4112 | ||
4113 | list_del_init(&em->list); | |
4114 | ||
4115 | /* | |
4116 | * If we had an error we just need to delete everybody from our | |
4117 | * private list. | |
4118 | */ | |
ff44c6e3 | 4119 | if (ret) { |
201a9038 | 4120 | clear_em_logging(tree, em); |
ff44c6e3 | 4121 | free_extent_map(em); |
5dc562c5 | 4122 | continue; |
ff44c6e3 JB |
4123 | } |
4124 | ||
4125 | write_unlock(&tree->lock); | |
5dc562c5 | 4126 | |
8407f553 FM |
4127 | ret = log_one_extent(trans, inode, root, em, path, logged_list, |
4128 | ctx); | |
ff44c6e3 | 4129 | write_lock(&tree->lock); |
201a9038 JB |
4130 | clear_em_logging(tree, em); |
4131 | free_extent_map(em); | |
5dc562c5 | 4132 | } |
ff44c6e3 JB |
4133 | WARN_ON(!list_empty(&extents)); |
4134 | write_unlock(&tree->lock); | |
5dc562c5 | 4135 | |
5dc562c5 | 4136 | btrfs_release_path(path); |
5dc562c5 JB |
4137 | return ret; |
4138 | } | |
4139 | ||
1a4bcf47 FM |
4140 | static int logged_inode_size(struct btrfs_root *log, struct inode *inode, |
4141 | struct btrfs_path *path, u64 *size_ret) | |
4142 | { | |
4143 | struct btrfs_key key; | |
4144 | int ret; | |
4145 | ||
4146 | key.objectid = btrfs_ino(inode); | |
4147 | key.type = BTRFS_INODE_ITEM_KEY; | |
4148 | key.offset = 0; | |
4149 | ||
4150 | ret = btrfs_search_slot(NULL, log, &key, path, 0, 0); | |
4151 | if (ret < 0) { | |
4152 | return ret; | |
4153 | } else if (ret > 0) { | |
2f2ff0ee | 4154 | *size_ret = 0; |
1a4bcf47 FM |
4155 | } else { |
4156 | struct btrfs_inode_item *item; | |
4157 | ||
4158 | item = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
4159 | struct btrfs_inode_item); | |
4160 | *size_ret = btrfs_inode_size(path->nodes[0], item); | |
4161 | } | |
4162 | ||
4163 | btrfs_release_path(path); | |
4164 | return 0; | |
4165 | } | |
4166 | ||
36283bf7 FM |
4167 | /* |
4168 | * At the moment we always log all xattrs. This is to figure out at log replay | |
4169 | * time which xattrs must have their deletion replayed. If a xattr is missing | |
4170 | * in the log tree and exists in the fs/subvol tree, we delete it. This is | |
4171 | * because if a xattr is deleted, the inode is fsynced and a power failure | |
4172 | * happens, causing the log to be replayed the next time the fs is mounted, | |
4173 | * we want the xattr to not exist anymore (same behaviour as other filesystems | |
4174 | * with a journal, ext3/4, xfs, f2fs, etc). | |
4175 | */ | |
4176 | static int btrfs_log_all_xattrs(struct btrfs_trans_handle *trans, | |
4177 | struct btrfs_root *root, | |
4178 | struct inode *inode, | |
4179 | struct btrfs_path *path, | |
4180 | struct btrfs_path *dst_path) | |
4181 | { | |
4182 | int ret; | |
4183 | struct btrfs_key key; | |
4184 | const u64 ino = btrfs_ino(inode); | |
4185 | int ins_nr = 0; | |
4186 | int start_slot = 0; | |
4187 | ||
4188 | key.objectid = ino; | |
4189 | key.type = BTRFS_XATTR_ITEM_KEY; | |
4190 | key.offset = 0; | |
4191 | ||
4192 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
4193 | if (ret < 0) | |
4194 | return ret; | |
4195 | ||
4196 | while (true) { | |
4197 | int slot = path->slots[0]; | |
4198 | struct extent_buffer *leaf = path->nodes[0]; | |
4199 | int nritems = btrfs_header_nritems(leaf); | |
4200 | ||
4201 | if (slot >= nritems) { | |
4202 | if (ins_nr > 0) { | |
4203 | u64 last_extent = 0; | |
4204 | ||
4205 | ret = copy_items(trans, inode, dst_path, path, | |
4206 | &last_extent, start_slot, | |
4207 | ins_nr, 1, 0); | |
4208 | /* can't be 1, extent items aren't processed */ | |
4209 | ASSERT(ret <= 0); | |
4210 | if (ret < 0) | |
4211 | return ret; | |
4212 | ins_nr = 0; | |
4213 | } | |
4214 | ret = btrfs_next_leaf(root, path); | |
4215 | if (ret < 0) | |
4216 | return ret; | |
4217 | else if (ret > 0) | |
4218 | break; | |
4219 | continue; | |
4220 | } | |
4221 | ||
4222 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
4223 | if (key.objectid != ino || key.type != BTRFS_XATTR_ITEM_KEY) | |
4224 | break; | |
4225 | ||
4226 | if (ins_nr == 0) | |
4227 | start_slot = slot; | |
4228 | ins_nr++; | |
4229 | path->slots[0]++; | |
4230 | cond_resched(); | |
4231 | } | |
4232 | if (ins_nr > 0) { | |
4233 | u64 last_extent = 0; | |
4234 | ||
4235 | ret = copy_items(trans, inode, dst_path, path, | |
4236 | &last_extent, start_slot, | |
4237 | ins_nr, 1, 0); | |
4238 | /* can't be 1, extent items aren't processed */ | |
4239 | ASSERT(ret <= 0); | |
4240 | if (ret < 0) | |
4241 | return ret; | |
4242 | } | |
4243 | ||
4244 | return 0; | |
4245 | } | |
4246 | ||
a89ca6f2 FM |
4247 | /* |
4248 | * If the no holes feature is enabled we need to make sure any hole between the | |
4249 | * last extent and the i_size of our inode is explicitly marked in the log. This | |
4250 | * is to make sure that doing something like: | |
4251 | * | |
4252 | * 1) create file with 128Kb of data | |
4253 | * 2) truncate file to 64Kb | |
4254 | * 3) truncate file to 256Kb | |
4255 | * 4) fsync file | |
4256 | * 5) <crash/power failure> | |
4257 | * 6) mount fs and trigger log replay | |
4258 | * | |
4259 | * Will give us a file with a size of 256Kb, the first 64Kb of data match what | |
4260 | * the file had in its first 64Kb of data at step 1 and the last 192Kb of the | |
4261 | * file correspond to a hole. The presence of explicit holes in a log tree is | |
4262 | * what guarantees that log replay will remove/adjust file extent items in the | |
4263 | * fs/subvol tree. | |
4264 | * | |
4265 | * Here we do not need to care about holes between extents, that is already done | |
4266 | * by copy_items(). We also only need to do this in the full sync path, where we | |
4267 | * lookup for extents from the fs/subvol tree only. In the fast path case, we | |
4268 | * lookup the list of modified extent maps and if any represents a hole, we | |
4269 | * insert a corresponding extent representing a hole in the log tree. | |
4270 | */ | |
4271 | static int btrfs_log_trailing_hole(struct btrfs_trans_handle *trans, | |
4272 | struct btrfs_root *root, | |
4273 | struct inode *inode, | |
4274 | struct btrfs_path *path) | |
4275 | { | |
4276 | int ret; | |
4277 | struct btrfs_key key; | |
4278 | u64 hole_start; | |
4279 | u64 hole_size; | |
4280 | struct extent_buffer *leaf; | |
4281 | struct btrfs_root *log = root->log_root; | |
4282 | const u64 ino = btrfs_ino(inode); | |
4283 | const u64 i_size = i_size_read(inode); | |
4284 | ||
4285 | if (!btrfs_fs_incompat(root->fs_info, NO_HOLES)) | |
4286 | return 0; | |
4287 | ||
4288 | key.objectid = ino; | |
4289 | key.type = BTRFS_EXTENT_DATA_KEY; | |
4290 | key.offset = (u64)-1; | |
4291 | ||
4292 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
4293 | ASSERT(ret != 0); | |
4294 | if (ret < 0) | |
4295 | return ret; | |
4296 | ||
4297 | ASSERT(path->slots[0] > 0); | |
4298 | path->slots[0]--; | |
4299 | leaf = path->nodes[0]; | |
4300 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
4301 | ||
4302 | if (key.objectid != ino || key.type != BTRFS_EXTENT_DATA_KEY) { | |
4303 | /* inode does not have any extents */ | |
4304 | hole_start = 0; | |
4305 | hole_size = i_size; | |
4306 | } else { | |
4307 | struct btrfs_file_extent_item *extent; | |
4308 | u64 len; | |
4309 | ||
4310 | /* | |
4311 | * If there's an extent beyond i_size, an explicit hole was | |
4312 | * already inserted by copy_items(). | |
4313 | */ | |
4314 | if (key.offset >= i_size) | |
4315 | return 0; | |
4316 | ||
4317 | extent = btrfs_item_ptr(leaf, path->slots[0], | |
4318 | struct btrfs_file_extent_item); | |
4319 | ||
4320 | if (btrfs_file_extent_type(leaf, extent) == | |
4321 | BTRFS_FILE_EXTENT_INLINE) { | |
4322 | len = btrfs_file_extent_inline_len(leaf, | |
4323 | path->slots[0], | |
4324 | extent); | |
4325 | ASSERT(len == i_size); | |
4326 | return 0; | |
4327 | } | |
4328 | ||
4329 | len = btrfs_file_extent_num_bytes(leaf, extent); | |
4330 | /* Last extent goes beyond i_size, no need to log a hole. */ | |
4331 | if (key.offset + len > i_size) | |
4332 | return 0; | |
4333 | hole_start = key.offset + len; | |
4334 | hole_size = i_size - hole_start; | |
4335 | } | |
4336 | btrfs_release_path(path); | |
4337 | ||
4338 | /* Last extent ends at i_size. */ | |
4339 | if (hole_size == 0) | |
4340 | return 0; | |
4341 | ||
4342 | hole_size = ALIGN(hole_size, root->sectorsize); | |
4343 | ret = btrfs_insert_file_extent(trans, log, ino, hole_start, 0, 0, | |
4344 | hole_size, 0, hole_size, 0, 0, 0); | |
4345 | return ret; | |
4346 | } | |
4347 | ||
e02119d5 CM |
4348 | /* log a single inode in the tree log. |
4349 | * At least one parent directory for this inode must exist in the tree | |
4350 | * or be logged already. | |
4351 | * | |
4352 | * Any items from this inode changed by the current transaction are copied | |
4353 | * to the log tree. An extra reference is taken on any extents in this | |
4354 | * file, allowing us to avoid a whole pile of corner cases around logging | |
4355 | * blocks that have been removed from the tree. | |
4356 | * | |
4357 | * See LOG_INODE_ALL and related defines for a description of what inode_only | |
4358 | * does. | |
4359 | * | |
4360 | * This handles both files and directories. | |
4361 | */ | |
12fcfd22 | 4362 | static int btrfs_log_inode(struct btrfs_trans_handle *trans, |
49dae1bc FM |
4363 | struct btrfs_root *root, struct inode *inode, |
4364 | int inode_only, | |
4365 | const loff_t start, | |
8407f553 FM |
4366 | const loff_t end, |
4367 | struct btrfs_log_ctx *ctx) | |
e02119d5 CM |
4368 | { |
4369 | struct btrfs_path *path; | |
4370 | struct btrfs_path *dst_path; | |
4371 | struct btrfs_key min_key; | |
4372 | struct btrfs_key max_key; | |
4373 | struct btrfs_root *log = root->log_root; | |
31ff1cd2 | 4374 | struct extent_buffer *src = NULL; |
827463c4 | 4375 | LIST_HEAD(logged_list); |
16e7549f | 4376 | u64 last_extent = 0; |
4a500fd1 | 4377 | int err = 0; |
e02119d5 | 4378 | int ret; |
3a5f1d45 | 4379 | int nritems; |
31ff1cd2 CM |
4380 | int ins_start_slot = 0; |
4381 | int ins_nr; | |
5dc562c5 | 4382 | bool fast_search = false; |
33345d01 | 4383 | u64 ino = btrfs_ino(inode); |
49dae1bc | 4384 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; |
1a4bcf47 | 4385 | u64 logged_isize = 0; |
e4545de5 | 4386 | bool need_log_inode_item = true; |
e02119d5 | 4387 | |
e02119d5 | 4388 | path = btrfs_alloc_path(); |
5df67083 TI |
4389 | if (!path) |
4390 | return -ENOMEM; | |
e02119d5 | 4391 | dst_path = btrfs_alloc_path(); |
5df67083 TI |
4392 | if (!dst_path) { |
4393 | btrfs_free_path(path); | |
4394 | return -ENOMEM; | |
4395 | } | |
e02119d5 | 4396 | |
33345d01 | 4397 | min_key.objectid = ino; |
e02119d5 CM |
4398 | min_key.type = BTRFS_INODE_ITEM_KEY; |
4399 | min_key.offset = 0; | |
4400 | ||
33345d01 | 4401 | max_key.objectid = ino; |
12fcfd22 | 4402 | |
12fcfd22 | 4403 | |
5dc562c5 | 4404 | /* today the code can only do partial logging of directories */ |
5269b67e MX |
4405 | if (S_ISDIR(inode->i_mode) || |
4406 | (!test_bit(BTRFS_INODE_NEEDS_FULL_SYNC, | |
4407 | &BTRFS_I(inode)->runtime_flags) && | |
4408 | inode_only == LOG_INODE_EXISTS)) | |
e02119d5 CM |
4409 | max_key.type = BTRFS_XATTR_ITEM_KEY; |
4410 | else | |
4411 | max_key.type = (u8)-1; | |
4412 | max_key.offset = (u64)-1; | |
4413 | ||
2c2c452b FM |
4414 | /* |
4415 | * Only run delayed items if we are a dir or a new file. | |
4416 | * Otherwise commit the delayed inode only, which is needed in | |
4417 | * order for the log replay code to mark inodes for link count | |
4418 | * fixup (create temporary BTRFS_TREE_LOG_FIXUP_OBJECTID items). | |
4419 | */ | |
94edf4ae | 4420 | if (S_ISDIR(inode->i_mode) || |
2c2c452b | 4421 | BTRFS_I(inode)->generation > root->fs_info->last_trans_committed) |
94edf4ae | 4422 | ret = btrfs_commit_inode_delayed_items(trans, inode); |
2c2c452b FM |
4423 | else |
4424 | ret = btrfs_commit_inode_delayed_inode(inode); | |
4425 | ||
4426 | if (ret) { | |
4427 | btrfs_free_path(path); | |
4428 | btrfs_free_path(dst_path); | |
4429 | return ret; | |
16cdcec7 MX |
4430 | } |
4431 | ||
e02119d5 CM |
4432 | mutex_lock(&BTRFS_I(inode)->log_mutex); |
4433 | ||
0870295b | 4434 | btrfs_get_logged_extents(inode, &logged_list, start, end); |
2ab28f32 | 4435 | |
e02119d5 CM |
4436 | /* |
4437 | * a brute force approach to making sure we get the most uptodate | |
4438 | * copies of everything. | |
4439 | */ | |
4440 | if (S_ISDIR(inode->i_mode)) { | |
4441 | int max_key_type = BTRFS_DIR_LOG_INDEX_KEY; | |
4442 | ||
4f764e51 FM |
4443 | if (inode_only == LOG_INODE_EXISTS) |
4444 | max_key_type = BTRFS_XATTR_ITEM_KEY; | |
33345d01 | 4445 | ret = drop_objectid_items(trans, log, path, ino, max_key_type); |
e02119d5 | 4446 | } else { |
1a4bcf47 FM |
4447 | if (inode_only == LOG_INODE_EXISTS) { |
4448 | /* | |
4449 | * Make sure the new inode item we write to the log has | |
4450 | * the same isize as the current one (if it exists). | |
4451 | * This is necessary to prevent data loss after log | |
4452 | * replay, and also to prevent doing a wrong expanding | |
4453 | * truncate - for e.g. create file, write 4K into offset | |
4454 | * 0, fsync, write 4K into offset 4096, add hard link, | |
4455 | * fsync some other file (to sync log), power fail - if | |
4456 | * we use the inode's current i_size, after log replay | |
4457 | * we get a 8Kb file, with the last 4Kb extent as a hole | |
4458 | * (zeroes), as if an expanding truncate happened, | |
4459 | * instead of getting a file of 4Kb only. | |
4460 | */ | |
4461 | err = logged_inode_size(log, inode, path, | |
4462 | &logged_isize); | |
4463 | if (err) | |
4464 | goto out_unlock; | |
4465 | } | |
a742994a FM |
4466 | if (test_bit(BTRFS_INODE_NEEDS_FULL_SYNC, |
4467 | &BTRFS_I(inode)->runtime_flags)) { | |
4468 | if (inode_only == LOG_INODE_EXISTS) { | |
4f764e51 | 4469 | max_key.type = BTRFS_XATTR_ITEM_KEY; |
a742994a FM |
4470 | ret = drop_objectid_items(trans, log, path, ino, |
4471 | max_key.type); | |
4472 | } else { | |
4473 | clear_bit(BTRFS_INODE_NEEDS_FULL_SYNC, | |
4474 | &BTRFS_I(inode)->runtime_flags); | |
4475 | clear_bit(BTRFS_INODE_COPY_EVERYTHING, | |
4476 | &BTRFS_I(inode)->runtime_flags); | |
28ed1345 CM |
4477 | while(1) { |
4478 | ret = btrfs_truncate_inode_items(trans, | |
4479 | log, inode, 0, 0); | |
4480 | if (ret != -EAGAIN) | |
4481 | break; | |
4482 | } | |
a742994a | 4483 | } |
4f764e51 FM |
4484 | } else if (test_and_clear_bit(BTRFS_INODE_COPY_EVERYTHING, |
4485 | &BTRFS_I(inode)->runtime_flags) || | |
6cfab851 | 4486 | inode_only == LOG_INODE_EXISTS) { |
4f764e51 | 4487 | if (inode_only == LOG_INODE_ALL) |
183f37fa | 4488 | fast_search = true; |
4f764e51 | 4489 | max_key.type = BTRFS_XATTR_ITEM_KEY; |
5dc562c5 | 4490 | ret = drop_objectid_items(trans, log, path, ino, |
e9976151 | 4491 | max_key.type); |
a95249b3 JB |
4492 | } else { |
4493 | if (inode_only == LOG_INODE_ALL) | |
4494 | fast_search = true; | |
a95249b3 | 4495 | goto log_extents; |
5dc562c5 | 4496 | } |
a95249b3 | 4497 | |
e02119d5 | 4498 | } |
4a500fd1 YZ |
4499 | if (ret) { |
4500 | err = ret; | |
4501 | goto out_unlock; | |
4502 | } | |
e02119d5 | 4503 | |
d397712b | 4504 | while (1) { |
31ff1cd2 | 4505 | ins_nr = 0; |
6174d3cb | 4506 | ret = btrfs_search_forward(root, &min_key, |
de78b51a | 4507 | path, trans->transid); |
e02119d5 CM |
4508 | if (ret != 0) |
4509 | break; | |
3a5f1d45 | 4510 | again: |
31ff1cd2 | 4511 | /* note, ins_nr might be > 0 here, cleanup outside the loop */ |
33345d01 | 4512 | if (min_key.objectid != ino) |
e02119d5 CM |
4513 | break; |
4514 | if (min_key.type > max_key.type) | |
4515 | break; | |
31ff1cd2 | 4516 | |
e4545de5 FM |
4517 | if (min_key.type == BTRFS_INODE_ITEM_KEY) |
4518 | need_log_inode_item = false; | |
4519 | ||
36283bf7 FM |
4520 | /* Skip xattrs, we log them later with btrfs_log_all_xattrs() */ |
4521 | if (min_key.type == BTRFS_XATTR_ITEM_KEY) { | |
4522 | if (ins_nr == 0) | |
4523 | goto next_slot; | |
4524 | ret = copy_items(trans, inode, dst_path, path, | |
4525 | &last_extent, ins_start_slot, | |
4526 | ins_nr, inode_only, logged_isize); | |
4527 | if (ret < 0) { | |
4528 | err = ret; | |
4529 | goto out_unlock; | |
4530 | } | |
4531 | ins_nr = 0; | |
4532 | if (ret) { | |
4533 | btrfs_release_path(path); | |
4534 | continue; | |
4535 | } | |
4536 | goto next_slot; | |
4537 | } | |
4538 | ||
e02119d5 | 4539 | src = path->nodes[0]; |
31ff1cd2 CM |
4540 | if (ins_nr && ins_start_slot + ins_nr == path->slots[0]) { |
4541 | ins_nr++; | |
4542 | goto next_slot; | |
4543 | } else if (!ins_nr) { | |
4544 | ins_start_slot = path->slots[0]; | |
4545 | ins_nr = 1; | |
4546 | goto next_slot; | |
e02119d5 CM |
4547 | } |
4548 | ||
16e7549f | 4549 | ret = copy_items(trans, inode, dst_path, path, &last_extent, |
1a4bcf47 FM |
4550 | ins_start_slot, ins_nr, inode_only, |
4551 | logged_isize); | |
16e7549f | 4552 | if (ret < 0) { |
4a500fd1 YZ |
4553 | err = ret; |
4554 | goto out_unlock; | |
a71db86e RV |
4555 | } |
4556 | if (ret) { | |
16e7549f JB |
4557 | ins_nr = 0; |
4558 | btrfs_release_path(path); | |
4559 | continue; | |
4a500fd1 | 4560 | } |
31ff1cd2 CM |
4561 | ins_nr = 1; |
4562 | ins_start_slot = path->slots[0]; | |
4563 | next_slot: | |
e02119d5 | 4564 | |
3a5f1d45 CM |
4565 | nritems = btrfs_header_nritems(path->nodes[0]); |
4566 | path->slots[0]++; | |
4567 | if (path->slots[0] < nritems) { | |
4568 | btrfs_item_key_to_cpu(path->nodes[0], &min_key, | |
4569 | path->slots[0]); | |
4570 | goto again; | |
4571 | } | |
31ff1cd2 | 4572 | if (ins_nr) { |
16e7549f JB |
4573 | ret = copy_items(trans, inode, dst_path, path, |
4574 | &last_extent, ins_start_slot, | |
1a4bcf47 | 4575 | ins_nr, inode_only, logged_isize); |
16e7549f | 4576 | if (ret < 0) { |
4a500fd1 YZ |
4577 | err = ret; |
4578 | goto out_unlock; | |
4579 | } | |
16e7549f | 4580 | ret = 0; |
31ff1cd2 CM |
4581 | ins_nr = 0; |
4582 | } | |
b3b4aa74 | 4583 | btrfs_release_path(path); |
3a5f1d45 | 4584 | |
3d41d702 | 4585 | if (min_key.offset < (u64)-1) { |
e02119d5 | 4586 | min_key.offset++; |
3d41d702 | 4587 | } else if (min_key.type < max_key.type) { |
e02119d5 | 4588 | min_key.type++; |
3d41d702 FDBM |
4589 | min_key.offset = 0; |
4590 | } else { | |
e02119d5 | 4591 | break; |
3d41d702 | 4592 | } |
e02119d5 | 4593 | } |
31ff1cd2 | 4594 | if (ins_nr) { |
16e7549f | 4595 | ret = copy_items(trans, inode, dst_path, path, &last_extent, |
1a4bcf47 FM |
4596 | ins_start_slot, ins_nr, inode_only, |
4597 | logged_isize); | |
16e7549f | 4598 | if (ret < 0) { |
4a500fd1 YZ |
4599 | err = ret; |
4600 | goto out_unlock; | |
4601 | } | |
16e7549f | 4602 | ret = 0; |
31ff1cd2 CM |
4603 | ins_nr = 0; |
4604 | } | |
5dc562c5 | 4605 | |
36283bf7 FM |
4606 | btrfs_release_path(path); |
4607 | btrfs_release_path(dst_path); | |
4608 | err = btrfs_log_all_xattrs(trans, root, inode, path, dst_path); | |
4609 | if (err) | |
4610 | goto out_unlock; | |
a89ca6f2 FM |
4611 | if (max_key.type >= BTRFS_EXTENT_DATA_KEY && !fast_search) { |
4612 | btrfs_release_path(path); | |
4613 | btrfs_release_path(dst_path); | |
4614 | err = btrfs_log_trailing_hole(trans, root, inode, path); | |
4615 | if (err) | |
4616 | goto out_unlock; | |
4617 | } | |
a95249b3 | 4618 | log_extents: |
f3b15ccd JB |
4619 | btrfs_release_path(path); |
4620 | btrfs_release_path(dst_path); | |
e4545de5 FM |
4621 | if (need_log_inode_item) { |
4622 | err = log_inode_item(trans, log, dst_path, inode); | |
4623 | if (err) | |
4624 | goto out_unlock; | |
4625 | } | |
5dc562c5 | 4626 | if (fast_search) { |
b38ef71c FM |
4627 | /* |
4628 | * Some ordered extents started by fsync might have completed | |
4629 | * before we collected the ordered extents in logged_list, which | |
4630 | * means they're gone, not in our logged_list nor in the inode's | |
4631 | * ordered tree. We want the application/user space to know an | |
4632 | * error happened while attempting to persist file data so that | |
4633 | * it can take proper action. If such error happened, we leave | |
4634 | * without writing to the log tree and the fsync must report the | |
4635 | * file data write error and not commit the current transaction. | |
4636 | */ | |
4637 | err = btrfs_inode_check_errors(inode); | |
4638 | if (err) { | |
4639 | ctx->io_err = err; | |
4640 | goto out_unlock; | |
4641 | } | |
827463c4 | 4642 | ret = btrfs_log_changed_extents(trans, root, inode, dst_path, |
8407f553 | 4643 | &logged_list, ctx); |
5dc562c5 JB |
4644 | if (ret) { |
4645 | err = ret; | |
4646 | goto out_unlock; | |
4647 | } | |
d006a048 | 4648 | } else if (inode_only == LOG_INODE_ALL) { |
06d3d22b LB |
4649 | struct extent_map *em, *n; |
4650 | ||
49dae1bc FM |
4651 | write_lock(&em_tree->lock); |
4652 | /* | |
4653 | * We can't just remove every em if we're called for a ranged | |
4654 | * fsync - that is, one that doesn't cover the whole possible | |
4655 | * file range (0 to LLONG_MAX). This is because we can have | |
4656 | * em's that fall outside the range we're logging and therefore | |
4657 | * their ordered operations haven't completed yet | |
4658 | * (btrfs_finish_ordered_io() not invoked yet). This means we | |
4659 | * didn't get their respective file extent item in the fs/subvol | |
4660 | * tree yet, and need to let the next fast fsync (one which | |
4661 | * consults the list of modified extent maps) find the em so | |
4662 | * that it logs a matching file extent item and waits for the | |
4663 | * respective ordered operation to complete (if it's still | |
4664 | * running). | |
4665 | * | |
4666 | * Removing every em outside the range we're logging would make | |
4667 | * the next fast fsync not log their matching file extent items, | |
4668 | * therefore making us lose data after a log replay. | |
4669 | */ | |
4670 | list_for_each_entry_safe(em, n, &em_tree->modified_extents, | |
4671 | list) { | |
4672 | const u64 mod_end = em->mod_start + em->mod_len - 1; | |
4673 | ||
4674 | if (em->mod_start >= start && mod_end <= end) | |
4675 | list_del_init(&em->list); | |
4676 | } | |
4677 | write_unlock(&em_tree->lock); | |
5dc562c5 JB |
4678 | } |
4679 | ||
9623f9a3 | 4680 | if (inode_only == LOG_INODE_ALL && S_ISDIR(inode->i_mode)) { |
2f2ff0ee FM |
4681 | ret = log_directory_changes(trans, root, inode, path, dst_path, |
4682 | ctx); | |
4a500fd1 YZ |
4683 | if (ret) { |
4684 | err = ret; | |
4685 | goto out_unlock; | |
4686 | } | |
e02119d5 | 4687 | } |
49dae1bc | 4688 | |
2f2ff0ee | 4689 | spin_lock(&BTRFS_I(inode)->lock); |
125c4cf9 FM |
4690 | BTRFS_I(inode)->logged_trans = trans->transid; |
4691 | BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->last_sub_trans; | |
2f2ff0ee | 4692 | spin_unlock(&BTRFS_I(inode)->lock); |
4a500fd1 | 4693 | out_unlock: |
827463c4 MX |
4694 | if (unlikely(err)) |
4695 | btrfs_put_logged_extents(&logged_list); | |
4696 | else | |
4697 | btrfs_submit_logged_extents(&logged_list, log); | |
e02119d5 CM |
4698 | mutex_unlock(&BTRFS_I(inode)->log_mutex); |
4699 | ||
4700 | btrfs_free_path(path); | |
4701 | btrfs_free_path(dst_path); | |
4a500fd1 | 4702 | return err; |
e02119d5 CM |
4703 | } |
4704 | ||
12fcfd22 CM |
4705 | /* |
4706 | * follow the dentry parent pointers up the chain and see if any | |
4707 | * of the directories in it require a full commit before they can | |
4708 | * be logged. Returns zero if nothing special needs to be done or 1 if | |
4709 | * a full commit is required. | |
4710 | */ | |
4711 | static noinline int check_parent_dirs_for_sync(struct btrfs_trans_handle *trans, | |
4712 | struct inode *inode, | |
4713 | struct dentry *parent, | |
4714 | struct super_block *sb, | |
4715 | u64 last_committed) | |
e02119d5 | 4716 | { |
12fcfd22 CM |
4717 | int ret = 0; |
4718 | struct btrfs_root *root; | |
6a912213 | 4719 | struct dentry *old_parent = NULL; |
de2b530b | 4720 | struct inode *orig_inode = inode; |
e02119d5 | 4721 | |
af4176b4 CM |
4722 | /* |
4723 | * for regular files, if its inode is already on disk, we don't | |
4724 | * have to worry about the parents at all. This is because | |
4725 | * we can use the last_unlink_trans field to record renames | |
4726 | * and other fun in this file. | |
4727 | */ | |
4728 | if (S_ISREG(inode->i_mode) && | |
4729 | BTRFS_I(inode)->generation <= last_committed && | |
4730 | BTRFS_I(inode)->last_unlink_trans <= last_committed) | |
4731 | goto out; | |
4732 | ||
12fcfd22 | 4733 | if (!S_ISDIR(inode->i_mode)) { |
2b0143b5 | 4734 | if (!parent || d_really_is_negative(parent) || sb != d_inode(parent)->i_sb) |
12fcfd22 | 4735 | goto out; |
2b0143b5 | 4736 | inode = d_inode(parent); |
12fcfd22 CM |
4737 | } |
4738 | ||
4739 | while (1) { | |
de2b530b JB |
4740 | /* |
4741 | * If we are logging a directory then we start with our inode, | |
4742 | * not our parents inode, so we need to skipp setting the | |
4743 | * logged_trans so that further down in the log code we don't | |
4744 | * think this inode has already been logged. | |
4745 | */ | |
4746 | if (inode != orig_inode) | |
4747 | BTRFS_I(inode)->logged_trans = trans->transid; | |
12fcfd22 CM |
4748 | smp_mb(); |
4749 | ||
4750 | if (BTRFS_I(inode)->last_unlink_trans > last_committed) { | |
4751 | root = BTRFS_I(inode)->root; | |
4752 | ||
4753 | /* | |
4754 | * make sure any commits to the log are forced | |
4755 | * to be full commits | |
4756 | */ | |
995946dd | 4757 | btrfs_set_log_full_commit(root->fs_info, trans); |
12fcfd22 CM |
4758 | ret = 1; |
4759 | break; | |
4760 | } | |
4761 | ||
2b0143b5 | 4762 | if (!parent || d_really_is_negative(parent) || sb != d_inode(parent)->i_sb) |
12fcfd22 CM |
4763 | break; |
4764 | ||
76dda93c | 4765 | if (IS_ROOT(parent)) |
12fcfd22 CM |
4766 | break; |
4767 | ||
6a912213 JB |
4768 | parent = dget_parent(parent); |
4769 | dput(old_parent); | |
4770 | old_parent = parent; | |
2b0143b5 | 4771 | inode = d_inode(parent); |
12fcfd22 CM |
4772 | |
4773 | } | |
6a912213 | 4774 | dput(old_parent); |
12fcfd22 | 4775 | out: |
e02119d5 CM |
4776 | return ret; |
4777 | } | |
4778 | ||
2f2ff0ee FM |
4779 | struct btrfs_dir_list { |
4780 | u64 ino; | |
4781 | struct list_head list; | |
4782 | }; | |
4783 | ||
4784 | /* | |
4785 | * Log the inodes of the new dentries of a directory. See log_dir_items() for | |
4786 | * details about the why it is needed. | |
4787 | * This is a recursive operation - if an existing dentry corresponds to a | |
4788 | * directory, that directory's new entries are logged too (same behaviour as | |
4789 | * ext3/4, xfs, f2fs, reiserfs, nilfs2). Note that when logging the inodes | |
4790 | * the dentries point to we do not lock their i_mutex, otherwise lockdep | |
4791 | * complains about the following circular lock dependency / possible deadlock: | |
4792 | * | |
4793 | * CPU0 CPU1 | |
4794 | * ---- ---- | |
4795 | * lock(&type->i_mutex_dir_key#3/2); | |
4796 | * lock(sb_internal#2); | |
4797 | * lock(&type->i_mutex_dir_key#3/2); | |
4798 | * lock(&sb->s_type->i_mutex_key#14); | |
4799 | * | |
4800 | * Where sb_internal is the lock (a counter that works as a lock) acquired by | |
4801 | * sb_start_intwrite() in btrfs_start_transaction(). | |
4802 | * Not locking i_mutex of the inodes is still safe because: | |
4803 | * | |
4804 | * 1) For regular files we log with a mode of LOG_INODE_EXISTS. It's possible | |
4805 | * that while logging the inode new references (names) are added or removed | |
4806 | * from the inode, leaving the logged inode item with a link count that does | |
4807 | * not match the number of logged inode reference items. This is fine because | |
4808 | * at log replay time we compute the real number of links and correct the | |
4809 | * link count in the inode item (see replay_one_buffer() and | |
4810 | * link_to_fixup_dir()); | |
4811 | * | |
4812 | * 2) For directories we log with a mode of LOG_INODE_ALL. It's possible that | |
4813 | * while logging the inode's items new items with keys BTRFS_DIR_ITEM_KEY and | |
4814 | * BTRFS_DIR_INDEX_KEY are added to fs/subvol tree and the logged inode item | |
4815 | * has a size that doesn't match the sum of the lengths of all the logged | |
4816 | * names. This does not result in a problem because if a dir_item key is | |
4817 | * logged but its matching dir_index key is not logged, at log replay time we | |
4818 | * don't use it to replay the respective name (see replay_one_name()). On the | |
4819 | * other hand if only the dir_index key ends up being logged, the respective | |
4820 | * name is added to the fs/subvol tree with both the dir_item and dir_index | |
4821 | * keys created (see replay_one_name()). | |
4822 | * The directory's inode item with a wrong i_size is not a problem as well, | |
4823 | * since we don't use it at log replay time to set the i_size in the inode | |
4824 | * item of the fs/subvol tree (see overwrite_item()). | |
4825 | */ | |
4826 | static int log_new_dir_dentries(struct btrfs_trans_handle *trans, | |
4827 | struct btrfs_root *root, | |
4828 | struct inode *start_inode, | |
4829 | struct btrfs_log_ctx *ctx) | |
4830 | { | |
4831 | struct btrfs_root *log = root->log_root; | |
4832 | struct btrfs_path *path; | |
4833 | LIST_HEAD(dir_list); | |
4834 | struct btrfs_dir_list *dir_elem; | |
4835 | int ret = 0; | |
4836 | ||
4837 | path = btrfs_alloc_path(); | |
4838 | if (!path) | |
4839 | return -ENOMEM; | |
4840 | ||
4841 | dir_elem = kmalloc(sizeof(*dir_elem), GFP_NOFS); | |
4842 | if (!dir_elem) { | |
4843 | btrfs_free_path(path); | |
4844 | return -ENOMEM; | |
4845 | } | |
4846 | dir_elem->ino = btrfs_ino(start_inode); | |
4847 | list_add_tail(&dir_elem->list, &dir_list); | |
4848 | ||
4849 | while (!list_empty(&dir_list)) { | |
4850 | struct extent_buffer *leaf; | |
4851 | struct btrfs_key min_key; | |
4852 | int nritems; | |
4853 | int i; | |
4854 | ||
4855 | dir_elem = list_first_entry(&dir_list, struct btrfs_dir_list, | |
4856 | list); | |
4857 | if (ret) | |
4858 | goto next_dir_inode; | |
4859 | ||
4860 | min_key.objectid = dir_elem->ino; | |
4861 | min_key.type = BTRFS_DIR_ITEM_KEY; | |
4862 | min_key.offset = 0; | |
4863 | again: | |
4864 | btrfs_release_path(path); | |
4865 | ret = btrfs_search_forward(log, &min_key, path, trans->transid); | |
4866 | if (ret < 0) { | |
4867 | goto next_dir_inode; | |
4868 | } else if (ret > 0) { | |
4869 | ret = 0; | |
4870 | goto next_dir_inode; | |
4871 | } | |
4872 | ||
4873 | process_leaf: | |
4874 | leaf = path->nodes[0]; | |
4875 | nritems = btrfs_header_nritems(leaf); | |
4876 | for (i = path->slots[0]; i < nritems; i++) { | |
4877 | struct btrfs_dir_item *di; | |
4878 | struct btrfs_key di_key; | |
4879 | struct inode *di_inode; | |
4880 | struct btrfs_dir_list *new_dir_elem; | |
4881 | int log_mode = LOG_INODE_EXISTS; | |
4882 | int type; | |
4883 | ||
4884 | btrfs_item_key_to_cpu(leaf, &min_key, i); | |
4885 | if (min_key.objectid != dir_elem->ino || | |
4886 | min_key.type != BTRFS_DIR_ITEM_KEY) | |
4887 | goto next_dir_inode; | |
4888 | ||
4889 | di = btrfs_item_ptr(leaf, i, struct btrfs_dir_item); | |
4890 | type = btrfs_dir_type(leaf, di); | |
4891 | if (btrfs_dir_transid(leaf, di) < trans->transid && | |
4892 | type != BTRFS_FT_DIR) | |
4893 | continue; | |
4894 | btrfs_dir_item_key_to_cpu(leaf, di, &di_key); | |
4895 | if (di_key.type == BTRFS_ROOT_ITEM_KEY) | |
4896 | continue; | |
4897 | ||
4898 | di_inode = btrfs_iget(root->fs_info->sb, &di_key, | |
4899 | root, NULL); | |
4900 | if (IS_ERR(di_inode)) { | |
4901 | ret = PTR_ERR(di_inode); | |
4902 | goto next_dir_inode; | |
4903 | } | |
4904 | ||
4905 | if (btrfs_inode_in_log(di_inode, trans->transid)) { | |
4906 | iput(di_inode); | |
4907 | continue; | |
4908 | } | |
4909 | ||
4910 | ctx->log_new_dentries = false; | |
4911 | if (type == BTRFS_FT_DIR) | |
4912 | log_mode = LOG_INODE_ALL; | |
4913 | btrfs_release_path(path); | |
4914 | ret = btrfs_log_inode(trans, root, di_inode, | |
4915 | log_mode, 0, LLONG_MAX, ctx); | |
4916 | iput(di_inode); | |
4917 | if (ret) | |
4918 | goto next_dir_inode; | |
4919 | if (ctx->log_new_dentries) { | |
4920 | new_dir_elem = kmalloc(sizeof(*new_dir_elem), | |
4921 | GFP_NOFS); | |
4922 | if (!new_dir_elem) { | |
4923 | ret = -ENOMEM; | |
4924 | goto next_dir_inode; | |
4925 | } | |
4926 | new_dir_elem->ino = di_key.objectid; | |
4927 | list_add_tail(&new_dir_elem->list, &dir_list); | |
4928 | } | |
4929 | break; | |
4930 | } | |
4931 | if (i == nritems) { | |
4932 | ret = btrfs_next_leaf(log, path); | |
4933 | if (ret < 0) { | |
4934 | goto next_dir_inode; | |
4935 | } else if (ret > 0) { | |
4936 | ret = 0; | |
4937 | goto next_dir_inode; | |
4938 | } | |
4939 | goto process_leaf; | |
4940 | } | |
4941 | if (min_key.offset < (u64)-1) { | |
4942 | min_key.offset++; | |
4943 | goto again; | |
4944 | } | |
4945 | next_dir_inode: | |
4946 | list_del(&dir_elem->list); | |
4947 | kfree(dir_elem); | |
4948 | } | |
4949 | ||
4950 | btrfs_free_path(path); | |
4951 | return ret; | |
4952 | } | |
4953 | ||
18aa0922 FM |
4954 | static int btrfs_log_all_parents(struct btrfs_trans_handle *trans, |
4955 | struct inode *inode, | |
4956 | struct btrfs_log_ctx *ctx) | |
4957 | { | |
4958 | int ret; | |
4959 | struct btrfs_path *path; | |
4960 | struct btrfs_key key; | |
4961 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
4962 | const u64 ino = btrfs_ino(inode); | |
4963 | ||
4964 | path = btrfs_alloc_path(); | |
4965 | if (!path) | |
4966 | return -ENOMEM; | |
4967 | path->skip_locking = 1; | |
4968 | path->search_commit_root = 1; | |
4969 | ||
4970 | key.objectid = ino; | |
4971 | key.type = BTRFS_INODE_REF_KEY; | |
4972 | key.offset = 0; | |
4973 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
4974 | if (ret < 0) | |
4975 | goto out; | |
4976 | ||
4977 | while (true) { | |
4978 | struct extent_buffer *leaf = path->nodes[0]; | |
4979 | int slot = path->slots[0]; | |
4980 | u32 cur_offset = 0; | |
4981 | u32 item_size; | |
4982 | unsigned long ptr; | |
4983 | ||
4984 | if (slot >= btrfs_header_nritems(leaf)) { | |
4985 | ret = btrfs_next_leaf(root, path); | |
4986 | if (ret < 0) | |
4987 | goto out; | |
4988 | else if (ret > 0) | |
4989 | break; | |
4990 | continue; | |
4991 | } | |
4992 | ||
4993 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
4994 | /* BTRFS_INODE_EXTREF_KEY is BTRFS_INODE_REF_KEY + 1 */ | |
4995 | if (key.objectid != ino || key.type > BTRFS_INODE_EXTREF_KEY) | |
4996 | break; | |
4997 | ||
4998 | item_size = btrfs_item_size_nr(leaf, slot); | |
4999 | ptr = btrfs_item_ptr_offset(leaf, slot); | |
5000 | while (cur_offset < item_size) { | |
5001 | struct btrfs_key inode_key; | |
5002 | struct inode *dir_inode; | |
5003 | ||
5004 | inode_key.type = BTRFS_INODE_ITEM_KEY; | |
5005 | inode_key.offset = 0; | |
5006 | ||
5007 | if (key.type == BTRFS_INODE_EXTREF_KEY) { | |
5008 | struct btrfs_inode_extref *extref; | |
5009 | ||
5010 | extref = (struct btrfs_inode_extref *) | |
5011 | (ptr + cur_offset); | |
5012 | inode_key.objectid = btrfs_inode_extref_parent( | |
5013 | leaf, extref); | |
5014 | cur_offset += sizeof(*extref); | |
5015 | cur_offset += btrfs_inode_extref_name_len(leaf, | |
5016 | extref); | |
5017 | } else { | |
5018 | inode_key.objectid = key.offset; | |
5019 | cur_offset = item_size; | |
5020 | } | |
5021 | ||
5022 | dir_inode = btrfs_iget(root->fs_info->sb, &inode_key, | |
5023 | root, NULL); | |
5024 | /* If parent inode was deleted, skip it. */ | |
5025 | if (IS_ERR(dir_inode)) | |
5026 | continue; | |
5027 | ||
5028 | ret = btrfs_log_inode(trans, root, dir_inode, | |
5029 | LOG_INODE_ALL, 0, LLONG_MAX, ctx); | |
5030 | iput(dir_inode); | |
5031 | if (ret) | |
5032 | goto out; | |
5033 | } | |
5034 | path->slots[0]++; | |
5035 | } | |
5036 | ret = 0; | |
5037 | out: | |
5038 | btrfs_free_path(path); | |
5039 | return ret; | |
5040 | } | |
5041 | ||
e02119d5 CM |
5042 | /* |
5043 | * helper function around btrfs_log_inode to make sure newly created | |
5044 | * parent directories also end up in the log. A minimal inode and backref | |
5045 | * only logging is done of any parent directories that are older than | |
5046 | * the last committed transaction | |
5047 | */ | |
48a3b636 ES |
5048 | static int btrfs_log_inode_parent(struct btrfs_trans_handle *trans, |
5049 | struct btrfs_root *root, struct inode *inode, | |
49dae1bc FM |
5050 | struct dentry *parent, |
5051 | const loff_t start, | |
5052 | const loff_t end, | |
5053 | int exists_only, | |
8b050d35 | 5054 | struct btrfs_log_ctx *ctx) |
e02119d5 | 5055 | { |
12fcfd22 | 5056 | int inode_only = exists_only ? LOG_INODE_EXISTS : LOG_INODE_ALL; |
e02119d5 | 5057 | struct super_block *sb; |
6a912213 | 5058 | struct dentry *old_parent = NULL; |
12fcfd22 CM |
5059 | int ret = 0; |
5060 | u64 last_committed = root->fs_info->last_trans_committed; | |
2f2ff0ee FM |
5061 | bool log_dentries = false; |
5062 | struct inode *orig_inode = inode; | |
12fcfd22 CM |
5063 | |
5064 | sb = inode->i_sb; | |
5065 | ||
3a5e1404 SW |
5066 | if (btrfs_test_opt(root, NOTREELOG)) { |
5067 | ret = 1; | |
5068 | goto end_no_trans; | |
5069 | } | |
5070 | ||
995946dd MX |
5071 | /* |
5072 | * The prev transaction commit doesn't complete, we need do | |
5073 | * full commit by ourselves. | |
5074 | */ | |
12fcfd22 CM |
5075 | if (root->fs_info->last_trans_log_full_commit > |
5076 | root->fs_info->last_trans_committed) { | |
5077 | ret = 1; | |
5078 | goto end_no_trans; | |
5079 | } | |
5080 | ||
76dda93c YZ |
5081 | if (root != BTRFS_I(inode)->root || |
5082 | btrfs_root_refs(&root->root_item) == 0) { | |
5083 | ret = 1; | |
5084 | goto end_no_trans; | |
5085 | } | |
5086 | ||
12fcfd22 CM |
5087 | ret = check_parent_dirs_for_sync(trans, inode, parent, |
5088 | sb, last_committed); | |
5089 | if (ret) | |
5090 | goto end_no_trans; | |
e02119d5 | 5091 | |
22ee6985 | 5092 | if (btrfs_inode_in_log(inode, trans->transid)) { |
257c62e1 CM |
5093 | ret = BTRFS_NO_LOG_SYNC; |
5094 | goto end_no_trans; | |
5095 | } | |
5096 | ||
8b050d35 | 5097 | ret = start_log_trans(trans, root, ctx); |
4a500fd1 | 5098 | if (ret) |
e87ac136 | 5099 | goto end_no_trans; |
e02119d5 | 5100 | |
8407f553 | 5101 | ret = btrfs_log_inode(trans, root, inode, inode_only, start, end, ctx); |
4a500fd1 YZ |
5102 | if (ret) |
5103 | goto end_trans; | |
12fcfd22 | 5104 | |
af4176b4 CM |
5105 | /* |
5106 | * for regular files, if its inode is already on disk, we don't | |
5107 | * have to worry about the parents at all. This is because | |
5108 | * we can use the last_unlink_trans field to record renames | |
5109 | * and other fun in this file. | |
5110 | */ | |
5111 | if (S_ISREG(inode->i_mode) && | |
5112 | BTRFS_I(inode)->generation <= last_committed && | |
4a500fd1 YZ |
5113 | BTRFS_I(inode)->last_unlink_trans <= last_committed) { |
5114 | ret = 0; | |
5115 | goto end_trans; | |
5116 | } | |
af4176b4 | 5117 | |
2f2ff0ee FM |
5118 | if (S_ISDIR(inode->i_mode) && ctx && ctx->log_new_dentries) |
5119 | log_dentries = true; | |
5120 | ||
18aa0922 FM |
5121 | /* |
5122 | * On unlink we must make sure all our current and old parent directores | |
5123 | * inodes are fully logged. This is to prevent leaving dangling | |
5124 | * directory index entries in directories that were our parents but are | |
5125 | * not anymore. Not doing this results in old parent directory being | |
5126 | * impossible to delete after log replay (rmdir will always fail with | |
5127 | * error -ENOTEMPTY). | |
5128 | * | |
5129 | * Example 1: | |
5130 | * | |
5131 | * mkdir testdir | |
5132 | * touch testdir/foo | |
5133 | * ln testdir/foo testdir/bar | |
5134 | * sync | |
5135 | * unlink testdir/bar | |
5136 | * xfs_io -c fsync testdir/foo | |
5137 | * <power failure> | |
5138 | * mount fs, triggers log replay | |
5139 | * | |
5140 | * If we don't log the parent directory (testdir), after log replay the | |
5141 | * directory still has an entry pointing to the file inode using the bar | |
5142 | * name, but a matching BTRFS_INODE_[REF|EXTREF]_KEY does not exist and | |
5143 | * the file inode has a link count of 1. | |
5144 | * | |
5145 | * Example 2: | |
5146 | * | |
5147 | * mkdir testdir | |
5148 | * touch foo | |
5149 | * ln foo testdir/foo2 | |
5150 | * ln foo testdir/foo3 | |
5151 | * sync | |
5152 | * unlink testdir/foo3 | |
5153 | * xfs_io -c fsync foo | |
5154 | * <power failure> | |
5155 | * mount fs, triggers log replay | |
5156 | * | |
5157 | * Similar as the first example, after log replay the parent directory | |
5158 | * testdir still has an entry pointing to the inode file with name foo3 | |
5159 | * but the file inode does not have a matching BTRFS_INODE_REF_KEY item | |
5160 | * and has a link count of 2. | |
5161 | */ | |
5162 | if (BTRFS_I(inode)->last_unlink_trans > last_committed) { | |
5163 | ret = btrfs_log_all_parents(trans, orig_inode, ctx); | |
5164 | if (ret) | |
5165 | goto end_trans; | |
5166 | } | |
5167 | ||
12fcfd22 | 5168 | while (1) { |
2b0143b5 | 5169 | if (!parent || d_really_is_negative(parent) || sb != d_inode(parent)->i_sb) |
e02119d5 CM |
5170 | break; |
5171 | ||
2b0143b5 | 5172 | inode = d_inode(parent); |
76dda93c YZ |
5173 | if (root != BTRFS_I(inode)->root) |
5174 | break; | |
5175 | ||
18aa0922 FM |
5176 | if (BTRFS_I(inode)->generation > last_committed) { |
5177 | ret = btrfs_log_inode(trans, root, inode, | |
5178 | LOG_INODE_EXISTS, | |
8407f553 | 5179 | 0, LLONG_MAX, ctx); |
4a500fd1 YZ |
5180 | if (ret) |
5181 | goto end_trans; | |
12fcfd22 | 5182 | } |
76dda93c | 5183 | if (IS_ROOT(parent)) |
e02119d5 | 5184 | break; |
12fcfd22 | 5185 | |
6a912213 JB |
5186 | parent = dget_parent(parent); |
5187 | dput(old_parent); | |
5188 | old_parent = parent; | |
e02119d5 | 5189 | } |
2f2ff0ee FM |
5190 | if (log_dentries) |
5191 | ret = log_new_dir_dentries(trans, root, orig_inode, ctx); | |
5192 | else | |
5193 | ret = 0; | |
4a500fd1 | 5194 | end_trans: |
6a912213 | 5195 | dput(old_parent); |
4a500fd1 | 5196 | if (ret < 0) { |
995946dd | 5197 | btrfs_set_log_full_commit(root->fs_info, trans); |
4a500fd1 YZ |
5198 | ret = 1; |
5199 | } | |
8b050d35 MX |
5200 | |
5201 | if (ret) | |
5202 | btrfs_remove_log_ctx(root, ctx); | |
12fcfd22 CM |
5203 | btrfs_end_log_trans(root); |
5204 | end_no_trans: | |
5205 | return ret; | |
e02119d5 CM |
5206 | } |
5207 | ||
5208 | /* | |
5209 | * it is not safe to log dentry if the chunk root has added new | |
5210 | * chunks. This returns 0 if the dentry was logged, and 1 otherwise. | |
5211 | * If this returns 1, you must commit the transaction to safely get your | |
5212 | * data on disk. | |
5213 | */ | |
5214 | int btrfs_log_dentry_safe(struct btrfs_trans_handle *trans, | |
8b050d35 | 5215 | struct btrfs_root *root, struct dentry *dentry, |
49dae1bc FM |
5216 | const loff_t start, |
5217 | const loff_t end, | |
8b050d35 | 5218 | struct btrfs_log_ctx *ctx) |
e02119d5 | 5219 | { |
6a912213 JB |
5220 | struct dentry *parent = dget_parent(dentry); |
5221 | int ret; | |
5222 | ||
2b0143b5 | 5223 | ret = btrfs_log_inode_parent(trans, root, d_inode(dentry), parent, |
49dae1bc | 5224 | start, end, 0, ctx); |
6a912213 JB |
5225 | dput(parent); |
5226 | ||
5227 | return ret; | |
e02119d5 CM |
5228 | } |
5229 | ||
5230 | /* | |
5231 | * should be called during mount to recover any replay any log trees | |
5232 | * from the FS | |
5233 | */ | |
5234 | int btrfs_recover_log_trees(struct btrfs_root *log_root_tree) | |
5235 | { | |
5236 | int ret; | |
5237 | struct btrfs_path *path; | |
5238 | struct btrfs_trans_handle *trans; | |
5239 | struct btrfs_key key; | |
5240 | struct btrfs_key found_key; | |
5241 | struct btrfs_key tmp_key; | |
5242 | struct btrfs_root *log; | |
5243 | struct btrfs_fs_info *fs_info = log_root_tree->fs_info; | |
5244 | struct walk_control wc = { | |
5245 | .process_func = process_one_buffer, | |
5246 | .stage = 0, | |
5247 | }; | |
5248 | ||
e02119d5 | 5249 | path = btrfs_alloc_path(); |
db5b493a TI |
5250 | if (!path) |
5251 | return -ENOMEM; | |
5252 | ||
5253 | fs_info->log_root_recovering = 1; | |
e02119d5 | 5254 | |
4a500fd1 | 5255 | trans = btrfs_start_transaction(fs_info->tree_root, 0); |
79787eaa JM |
5256 | if (IS_ERR(trans)) { |
5257 | ret = PTR_ERR(trans); | |
5258 | goto error; | |
5259 | } | |
e02119d5 CM |
5260 | |
5261 | wc.trans = trans; | |
5262 | wc.pin = 1; | |
5263 | ||
db5b493a | 5264 | ret = walk_log_tree(trans, log_root_tree, &wc); |
79787eaa JM |
5265 | if (ret) { |
5266 | btrfs_error(fs_info, ret, "Failed to pin buffers while " | |
5267 | "recovering log root tree."); | |
5268 | goto error; | |
5269 | } | |
e02119d5 CM |
5270 | |
5271 | again: | |
5272 | key.objectid = BTRFS_TREE_LOG_OBJECTID; | |
5273 | key.offset = (u64)-1; | |
962a298f | 5274 | key.type = BTRFS_ROOT_ITEM_KEY; |
e02119d5 | 5275 | |
d397712b | 5276 | while (1) { |
e02119d5 | 5277 | ret = btrfs_search_slot(NULL, log_root_tree, &key, path, 0, 0); |
79787eaa JM |
5278 | |
5279 | if (ret < 0) { | |
5280 | btrfs_error(fs_info, ret, | |
5281 | "Couldn't find tree log root."); | |
5282 | goto error; | |
5283 | } | |
e02119d5 CM |
5284 | if (ret > 0) { |
5285 | if (path->slots[0] == 0) | |
5286 | break; | |
5287 | path->slots[0]--; | |
5288 | } | |
5289 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, | |
5290 | path->slots[0]); | |
b3b4aa74 | 5291 | btrfs_release_path(path); |
e02119d5 CM |
5292 | if (found_key.objectid != BTRFS_TREE_LOG_OBJECTID) |
5293 | break; | |
5294 | ||
cb517eab | 5295 | log = btrfs_read_fs_root(log_root_tree, &found_key); |
79787eaa JM |
5296 | if (IS_ERR(log)) { |
5297 | ret = PTR_ERR(log); | |
5298 | btrfs_error(fs_info, ret, | |
5299 | "Couldn't read tree log root."); | |
5300 | goto error; | |
5301 | } | |
e02119d5 CM |
5302 | |
5303 | tmp_key.objectid = found_key.offset; | |
5304 | tmp_key.type = BTRFS_ROOT_ITEM_KEY; | |
5305 | tmp_key.offset = (u64)-1; | |
5306 | ||
5307 | wc.replay_dest = btrfs_read_fs_root_no_name(fs_info, &tmp_key); | |
79787eaa JM |
5308 | if (IS_ERR(wc.replay_dest)) { |
5309 | ret = PTR_ERR(wc.replay_dest); | |
b50c6e25 JB |
5310 | free_extent_buffer(log->node); |
5311 | free_extent_buffer(log->commit_root); | |
5312 | kfree(log); | |
79787eaa JM |
5313 | btrfs_error(fs_info, ret, "Couldn't read target root " |
5314 | "for tree log recovery."); | |
5315 | goto error; | |
5316 | } | |
e02119d5 | 5317 | |
07d400a6 | 5318 | wc.replay_dest->log_root = log; |
5d4f98a2 | 5319 | btrfs_record_root_in_trans(trans, wc.replay_dest); |
e02119d5 | 5320 | ret = walk_log_tree(trans, log, &wc); |
e02119d5 | 5321 | |
b50c6e25 | 5322 | if (!ret && wc.stage == LOG_WALK_REPLAY_ALL) { |
e02119d5 CM |
5323 | ret = fixup_inode_link_counts(trans, wc.replay_dest, |
5324 | path); | |
e02119d5 CM |
5325 | } |
5326 | ||
5327 | key.offset = found_key.offset - 1; | |
07d400a6 | 5328 | wc.replay_dest->log_root = NULL; |
e02119d5 | 5329 | free_extent_buffer(log->node); |
b263c2c8 | 5330 | free_extent_buffer(log->commit_root); |
e02119d5 CM |
5331 | kfree(log); |
5332 | ||
b50c6e25 JB |
5333 | if (ret) |
5334 | goto error; | |
5335 | ||
e02119d5 CM |
5336 | if (found_key.offset == 0) |
5337 | break; | |
5338 | } | |
b3b4aa74 | 5339 | btrfs_release_path(path); |
e02119d5 CM |
5340 | |
5341 | /* step one is to pin it all, step two is to replay just inodes */ | |
5342 | if (wc.pin) { | |
5343 | wc.pin = 0; | |
5344 | wc.process_func = replay_one_buffer; | |
5345 | wc.stage = LOG_WALK_REPLAY_INODES; | |
5346 | goto again; | |
5347 | } | |
5348 | /* step three is to replay everything */ | |
5349 | if (wc.stage < LOG_WALK_REPLAY_ALL) { | |
5350 | wc.stage++; | |
5351 | goto again; | |
5352 | } | |
5353 | ||
5354 | btrfs_free_path(path); | |
5355 | ||
abefa55a JB |
5356 | /* step 4: commit the transaction, which also unpins the blocks */ |
5357 | ret = btrfs_commit_transaction(trans, fs_info->tree_root); | |
5358 | if (ret) | |
5359 | return ret; | |
5360 | ||
e02119d5 CM |
5361 | free_extent_buffer(log_root_tree->node); |
5362 | log_root_tree->log_root = NULL; | |
5363 | fs_info->log_root_recovering = 0; | |
e02119d5 | 5364 | kfree(log_root_tree); |
79787eaa | 5365 | |
abefa55a | 5366 | return 0; |
79787eaa | 5367 | error: |
b50c6e25 JB |
5368 | if (wc.trans) |
5369 | btrfs_end_transaction(wc.trans, fs_info->tree_root); | |
79787eaa JM |
5370 | btrfs_free_path(path); |
5371 | return ret; | |
e02119d5 | 5372 | } |
12fcfd22 CM |
5373 | |
5374 | /* | |
5375 | * there are some corner cases where we want to force a full | |
5376 | * commit instead of allowing a directory to be logged. | |
5377 | * | |
5378 | * They revolve around files there were unlinked from the directory, and | |
5379 | * this function updates the parent directory so that a full commit is | |
5380 | * properly done if it is fsync'd later after the unlinks are done. | |
5381 | */ | |
5382 | void btrfs_record_unlink_dir(struct btrfs_trans_handle *trans, | |
5383 | struct inode *dir, struct inode *inode, | |
5384 | int for_rename) | |
5385 | { | |
af4176b4 CM |
5386 | /* |
5387 | * when we're logging a file, if it hasn't been renamed | |
5388 | * or unlinked, and its inode is fully committed on disk, | |
5389 | * we don't have to worry about walking up the directory chain | |
5390 | * to log its parents. | |
5391 | * | |
5392 | * So, we use the last_unlink_trans field to put this transid | |
5393 | * into the file. When the file is logged we check it and | |
5394 | * don't log the parents if the file is fully on disk. | |
5395 | */ | |
5396 | if (S_ISREG(inode->i_mode)) | |
5397 | BTRFS_I(inode)->last_unlink_trans = trans->transid; | |
5398 | ||
12fcfd22 CM |
5399 | /* |
5400 | * if this directory was already logged any new | |
5401 | * names for this file/dir will get recorded | |
5402 | */ | |
5403 | smp_mb(); | |
5404 | if (BTRFS_I(dir)->logged_trans == trans->transid) | |
5405 | return; | |
5406 | ||
5407 | /* | |
5408 | * if the inode we're about to unlink was logged, | |
5409 | * the log will be properly updated for any new names | |
5410 | */ | |
5411 | if (BTRFS_I(inode)->logged_trans == trans->transid) | |
5412 | return; | |
5413 | ||
5414 | /* | |
5415 | * when renaming files across directories, if the directory | |
5416 | * there we're unlinking from gets fsync'd later on, there's | |
5417 | * no way to find the destination directory later and fsync it | |
5418 | * properly. So, we have to be conservative and force commits | |
5419 | * so the new name gets discovered. | |
5420 | */ | |
5421 | if (for_rename) | |
5422 | goto record; | |
5423 | ||
5424 | /* we can safely do the unlink without any special recording */ | |
5425 | return; | |
5426 | ||
5427 | record: | |
5428 | BTRFS_I(dir)->last_unlink_trans = trans->transid; | |
5429 | } | |
5430 | ||
5431 | /* | |
5432 | * Call this after adding a new name for a file and it will properly | |
5433 | * update the log to reflect the new name. | |
5434 | * | |
5435 | * It will return zero if all goes well, and it will return 1 if a | |
5436 | * full transaction commit is required. | |
5437 | */ | |
5438 | int btrfs_log_new_name(struct btrfs_trans_handle *trans, | |
5439 | struct inode *inode, struct inode *old_dir, | |
5440 | struct dentry *parent) | |
5441 | { | |
5442 | struct btrfs_root * root = BTRFS_I(inode)->root; | |
5443 | ||
af4176b4 CM |
5444 | /* |
5445 | * this will force the logging code to walk the dentry chain | |
5446 | * up for the file | |
5447 | */ | |
5448 | if (S_ISREG(inode->i_mode)) | |
5449 | BTRFS_I(inode)->last_unlink_trans = trans->transid; | |
5450 | ||
12fcfd22 CM |
5451 | /* |
5452 | * if this inode hasn't been logged and directory we're renaming it | |
5453 | * from hasn't been logged, we don't need to log it | |
5454 | */ | |
5455 | if (BTRFS_I(inode)->logged_trans <= | |
5456 | root->fs_info->last_trans_committed && | |
5457 | (!old_dir || BTRFS_I(old_dir)->logged_trans <= | |
5458 | root->fs_info->last_trans_committed)) | |
5459 | return 0; | |
5460 | ||
49dae1bc FM |
5461 | return btrfs_log_inode_parent(trans, root, inode, parent, 0, |
5462 | LLONG_MAX, 1, NULL); | |
12fcfd22 CM |
5463 | } |
5464 |