Commit | Line | Data |
---|---|---|
c1d7c514 | 1 | // SPDX-License-Identifier: GPL-2.0 |
e02119d5 CM |
2 | /* |
3 | * Copyright (C) 2008 Oracle. All rights reserved. | |
e02119d5 CM |
4 | */ |
5 | ||
6 | #include <linux/sched.h> | |
5a0e3ad6 | 7 | #include <linux/slab.h> |
c6adc9cc | 8 | #include <linux/blkdev.h> |
5dc562c5 | 9 | #include <linux/list_sort.h> |
c7f88c4e | 10 | #include <linux/iversion.h> |
9678c543 | 11 | #include "ctree.h" |
995946dd | 12 | #include "tree-log.h" |
e02119d5 CM |
13 | #include "disk-io.h" |
14 | #include "locking.h" | |
15 | #include "print-tree.h" | |
f186373f | 16 | #include "backref.h" |
ebb8765b | 17 | #include "compression.h" |
df2c95f3 | 18 | #include "qgroup.h" |
900c9981 | 19 | #include "inode-map.h" |
e02119d5 CM |
20 | |
21 | /* magic values for the inode_only field in btrfs_log_inode: | |
22 | * | |
23 | * LOG_INODE_ALL means to log everything | |
24 | * LOG_INODE_EXISTS means to log just enough to recreate the inode | |
25 | * during log replay | |
26 | */ | |
27 | #define LOG_INODE_ALL 0 | |
28 | #define LOG_INODE_EXISTS 1 | |
781feef7 | 29 | #define LOG_OTHER_INODE 2 |
e02119d5 | 30 | |
12fcfd22 CM |
31 | /* |
32 | * directory trouble cases | |
33 | * | |
34 | * 1) on rename or unlink, if the inode being unlinked isn't in the fsync | |
35 | * log, we must force a full commit before doing an fsync of the directory | |
36 | * where the unlink was done. | |
37 | * ---> record transid of last unlink/rename per directory | |
38 | * | |
39 | * mkdir foo/some_dir | |
40 | * normal commit | |
41 | * rename foo/some_dir foo2/some_dir | |
42 | * mkdir foo/some_dir | |
43 | * fsync foo/some_dir/some_file | |
44 | * | |
45 | * The fsync above will unlink the original some_dir without recording | |
46 | * it in its new location (foo2). After a crash, some_dir will be gone | |
47 | * unless the fsync of some_file forces a full commit | |
48 | * | |
49 | * 2) we must log any new names for any file or dir that is in the fsync | |
50 | * log. ---> check inode while renaming/linking. | |
51 | * | |
52 | * 2a) we must log any new names for any file or dir during rename | |
53 | * when the directory they are being removed from was logged. | |
54 | * ---> check inode and old parent dir during rename | |
55 | * | |
56 | * 2a is actually the more important variant. With the extra logging | |
57 | * a crash might unlink the old name without recreating the new one | |
58 | * | |
59 | * 3) after a crash, we must go through any directories with a link count | |
60 | * of zero and redo the rm -rf | |
61 | * | |
62 | * mkdir f1/foo | |
63 | * normal commit | |
64 | * rm -rf f1/foo | |
65 | * fsync(f1) | |
66 | * | |
67 | * The directory f1 was fully removed from the FS, but fsync was never | |
68 | * called on f1, only its parent dir. After a crash the rm -rf must | |
69 | * be replayed. This must be able to recurse down the entire | |
70 | * directory tree. The inode link count fixup code takes care of the | |
71 | * ugly details. | |
72 | */ | |
73 | ||
e02119d5 CM |
74 | /* |
75 | * stages for the tree walking. The first | |
76 | * stage (0) is to only pin down the blocks we find | |
77 | * the second stage (1) is to make sure that all the inodes | |
78 | * we find in the log are created in the subvolume. | |
79 | * | |
80 | * The last stage is to deal with directories and links and extents | |
81 | * and all the other fun semantics | |
82 | */ | |
83 | #define LOG_WALK_PIN_ONLY 0 | |
84 | #define LOG_WALK_REPLAY_INODES 1 | |
dd8e7217 JB |
85 | #define LOG_WALK_REPLAY_DIR_INDEX 2 |
86 | #define LOG_WALK_REPLAY_ALL 3 | |
e02119d5 | 87 | |
12fcfd22 | 88 | static int btrfs_log_inode(struct btrfs_trans_handle *trans, |
a59108a7 | 89 | struct btrfs_root *root, struct btrfs_inode *inode, |
49dae1bc FM |
90 | int inode_only, |
91 | const loff_t start, | |
8407f553 FM |
92 | const loff_t end, |
93 | struct btrfs_log_ctx *ctx); | |
ec051c0f YZ |
94 | static int link_to_fixup_dir(struct btrfs_trans_handle *trans, |
95 | struct btrfs_root *root, | |
96 | struct btrfs_path *path, u64 objectid); | |
12fcfd22 CM |
97 | static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans, |
98 | struct btrfs_root *root, | |
99 | struct btrfs_root *log, | |
100 | struct btrfs_path *path, | |
101 | u64 dirid, int del_all); | |
e02119d5 CM |
102 | |
103 | /* | |
104 | * tree logging is a special write ahead log used to make sure that | |
105 | * fsyncs and O_SYNCs can happen without doing full tree commits. | |
106 | * | |
107 | * Full tree commits are expensive because they require commonly | |
108 | * modified blocks to be recowed, creating many dirty pages in the | |
109 | * extent tree an 4x-6x higher write load than ext3. | |
110 | * | |
111 | * Instead of doing a tree commit on every fsync, we use the | |
112 | * key ranges and transaction ids to find items for a given file or directory | |
113 | * that have changed in this transaction. Those items are copied into | |
114 | * a special tree (one per subvolume root), that tree is written to disk | |
115 | * and then the fsync is considered complete. | |
116 | * | |
117 | * After a crash, items are copied out of the log-tree back into the | |
118 | * subvolume tree. Any file data extents found are recorded in the extent | |
119 | * allocation tree, and the log-tree freed. | |
120 | * | |
121 | * The log tree is read three times, once to pin down all the extents it is | |
122 | * using in ram and once, once to create all the inodes logged in the tree | |
123 | * and once to do all the other items. | |
124 | */ | |
125 | ||
e02119d5 CM |
126 | /* |
127 | * start a sub transaction and setup the log tree | |
128 | * this increments the log tree writer count to make the people | |
129 | * syncing the tree wait for us to finish | |
130 | */ | |
131 | static int start_log_trans(struct btrfs_trans_handle *trans, | |
8b050d35 MX |
132 | struct btrfs_root *root, |
133 | struct btrfs_log_ctx *ctx) | |
e02119d5 | 134 | { |
0b246afa | 135 | struct btrfs_fs_info *fs_info = root->fs_info; |
34eb2a52 | 136 | int ret = 0; |
7237f183 YZ |
137 | |
138 | mutex_lock(&root->log_mutex); | |
34eb2a52 | 139 | |
7237f183 | 140 | if (root->log_root) { |
0b246afa | 141 | if (btrfs_need_log_full_commit(fs_info, trans)) { |
50471a38 MX |
142 | ret = -EAGAIN; |
143 | goto out; | |
144 | } | |
34eb2a52 | 145 | |
ff782e0a | 146 | if (!root->log_start_pid) { |
27cdeb70 | 147 | clear_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state); |
34eb2a52 | 148 | root->log_start_pid = current->pid; |
ff782e0a | 149 | } else if (root->log_start_pid != current->pid) { |
27cdeb70 | 150 | set_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state); |
ff782e0a | 151 | } |
34eb2a52 | 152 | } else { |
0b246afa JM |
153 | mutex_lock(&fs_info->tree_log_mutex); |
154 | if (!fs_info->log_root_tree) | |
155 | ret = btrfs_init_log_root_tree(trans, fs_info); | |
156 | mutex_unlock(&fs_info->tree_log_mutex); | |
34eb2a52 Z |
157 | if (ret) |
158 | goto out; | |
ff782e0a | 159 | |
e02119d5 | 160 | ret = btrfs_add_log_tree(trans, root); |
4a500fd1 | 161 | if (ret) |
e87ac136 | 162 | goto out; |
34eb2a52 Z |
163 | |
164 | clear_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state); | |
165 | root->log_start_pid = current->pid; | |
e02119d5 | 166 | } |
34eb2a52 | 167 | |
2ecb7923 | 168 | atomic_inc(&root->log_batch); |
7237f183 | 169 | atomic_inc(&root->log_writers); |
8b050d35 | 170 | if (ctx) { |
34eb2a52 | 171 | int index = root->log_transid % 2; |
8b050d35 | 172 | list_add_tail(&ctx->list, &root->log_ctxs[index]); |
d1433deb | 173 | ctx->log_transid = root->log_transid; |
8b050d35 | 174 | } |
34eb2a52 | 175 | |
e87ac136 | 176 | out: |
7237f183 | 177 | mutex_unlock(&root->log_mutex); |
e87ac136 | 178 | return ret; |
e02119d5 CM |
179 | } |
180 | ||
181 | /* | |
182 | * returns 0 if there was a log transaction running and we were able | |
183 | * to join, or returns -ENOENT if there were not transactions | |
184 | * in progress | |
185 | */ | |
186 | static int join_running_log_trans(struct btrfs_root *root) | |
187 | { | |
188 | int ret = -ENOENT; | |
189 | ||
190 | smp_mb(); | |
191 | if (!root->log_root) | |
192 | return -ENOENT; | |
193 | ||
7237f183 | 194 | mutex_lock(&root->log_mutex); |
e02119d5 CM |
195 | if (root->log_root) { |
196 | ret = 0; | |
7237f183 | 197 | atomic_inc(&root->log_writers); |
e02119d5 | 198 | } |
7237f183 | 199 | mutex_unlock(&root->log_mutex); |
e02119d5 CM |
200 | return ret; |
201 | } | |
202 | ||
12fcfd22 CM |
203 | /* |
204 | * This either makes the current running log transaction wait | |
205 | * until you call btrfs_end_log_trans() or it makes any future | |
206 | * log transactions wait until you call btrfs_end_log_trans() | |
207 | */ | |
208 | int btrfs_pin_log_trans(struct btrfs_root *root) | |
209 | { | |
210 | int ret = -ENOENT; | |
211 | ||
212 | mutex_lock(&root->log_mutex); | |
213 | atomic_inc(&root->log_writers); | |
214 | mutex_unlock(&root->log_mutex); | |
215 | return ret; | |
216 | } | |
217 | ||
e02119d5 CM |
218 | /* |
219 | * indicate we're done making changes to the log tree | |
220 | * and wake up anyone waiting to do a sync | |
221 | */ | |
143bede5 | 222 | void btrfs_end_log_trans(struct btrfs_root *root) |
e02119d5 | 223 | { |
7237f183 | 224 | if (atomic_dec_and_test(&root->log_writers)) { |
779adf0f DS |
225 | /* |
226 | * Implicit memory barrier after atomic_dec_and_test | |
227 | */ | |
7237f183 YZ |
228 | if (waitqueue_active(&root->log_writer_wait)) |
229 | wake_up(&root->log_writer_wait); | |
230 | } | |
e02119d5 CM |
231 | } |
232 | ||
233 | ||
234 | /* | |
235 | * the walk control struct is used to pass state down the chain when | |
236 | * processing the log tree. The stage field tells us which part | |
237 | * of the log tree processing we are currently doing. The others | |
238 | * are state fields used for that specific part | |
239 | */ | |
240 | struct walk_control { | |
241 | /* should we free the extent on disk when done? This is used | |
242 | * at transaction commit time while freeing a log tree | |
243 | */ | |
244 | int free; | |
245 | ||
246 | /* should we write out the extent buffer? This is used | |
247 | * while flushing the log tree to disk during a sync | |
248 | */ | |
249 | int write; | |
250 | ||
251 | /* should we wait for the extent buffer io to finish? Also used | |
252 | * while flushing the log tree to disk for a sync | |
253 | */ | |
254 | int wait; | |
255 | ||
256 | /* pin only walk, we record which extents on disk belong to the | |
257 | * log trees | |
258 | */ | |
259 | int pin; | |
260 | ||
261 | /* what stage of the replay code we're currently in */ | |
262 | int stage; | |
263 | ||
264 | /* the root we are currently replaying */ | |
265 | struct btrfs_root *replay_dest; | |
266 | ||
267 | /* the trans handle for the current replay */ | |
268 | struct btrfs_trans_handle *trans; | |
269 | ||
270 | /* the function that gets used to process blocks we find in the | |
271 | * tree. Note the extent_buffer might not be up to date when it is | |
272 | * passed in, and it must be checked or read if you need the data | |
273 | * inside it | |
274 | */ | |
275 | int (*process_func)(struct btrfs_root *log, struct extent_buffer *eb, | |
581c1760 | 276 | struct walk_control *wc, u64 gen, int level); |
e02119d5 CM |
277 | }; |
278 | ||
279 | /* | |
280 | * process_func used to pin down extents, write them or wait on them | |
281 | */ | |
282 | static int process_one_buffer(struct btrfs_root *log, | |
283 | struct extent_buffer *eb, | |
581c1760 | 284 | struct walk_control *wc, u64 gen, int level) |
e02119d5 | 285 | { |
0b246afa | 286 | struct btrfs_fs_info *fs_info = log->fs_info; |
b50c6e25 JB |
287 | int ret = 0; |
288 | ||
8c2a1a30 JB |
289 | /* |
290 | * If this fs is mixed then we need to be able to process the leaves to | |
291 | * pin down any logged extents, so we have to read the block. | |
292 | */ | |
0b246afa | 293 | if (btrfs_fs_incompat(fs_info, MIXED_GROUPS)) { |
581c1760 | 294 | ret = btrfs_read_buffer(eb, gen, level, NULL); |
8c2a1a30 JB |
295 | if (ret) |
296 | return ret; | |
297 | } | |
298 | ||
04018de5 | 299 | if (wc->pin) |
2ff7e61e JM |
300 | ret = btrfs_pin_extent_for_log_replay(fs_info, eb->start, |
301 | eb->len); | |
e02119d5 | 302 | |
b50c6e25 | 303 | if (!ret && btrfs_buffer_uptodate(eb, gen, 0)) { |
8c2a1a30 | 304 | if (wc->pin && btrfs_header_level(eb) == 0) |
2ff7e61e | 305 | ret = btrfs_exclude_logged_extents(fs_info, eb); |
e02119d5 CM |
306 | if (wc->write) |
307 | btrfs_write_tree_block(eb); | |
308 | if (wc->wait) | |
309 | btrfs_wait_tree_block_writeback(eb); | |
310 | } | |
b50c6e25 | 311 | return ret; |
e02119d5 CM |
312 | } |
313 | ||
314 | /* | |
315 | * Item overwrite used by replay and tree logging. eb, slot and key all refer | |
316 | * to the src data we are copying out. | |
317 | * | |
318 | * root is the tree we are copying into, and path is a scratch | |
319 | * path for use in this function (it should be released on entry and | |
320 | * will be released on exit). | |
321 | * | |
322 | * If the key is already in the destination tree the existing item is | |
323 | * overwritten. If the existing item isn't big enough, it is extended. | |
324 | * If it is too large, it is truncated. | |
325 | * | |
326 | * If the key isn't in the destination yet, a new item is inserted. | |
327 | */ | |
328 | static noinline int overwrite_item(struct btrfs_trans_handle *trans, | |
329 | struct btrfs_root *root, | |
330 | struct btrfs_path *path, | |
331 | struct extent_buffer *eb, int slot, | |
332 | struct btrfs_key *key) | |
333 | { | |
2ff7e61e | 334 | struct btrfs_fs_info *fs_info = root->fs_info; |
e02119d5 CM |
335 | int ret; |
336 | u32 item_size; | |
337 | u64 saved_i_size = 0; | |
338 | int save_old_i_size = 0; | |
339 | unsigned long src_ptr; | |
340 | unsigned long dst_ptr; | |
341 | int overwrite_root = 0; | |
4bc4bee4 | 342 | bool inode_item = key->type == BTRFS_INODE_ITEM_KEY; |
e02119d5 CM |
343 | |
344 | if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) | |
345 | overwrite_root = 1; | |
346 | ||
347 | item_size = btrfs_item_size_nr(eb, slot); | |
348 | src_ptr = btrfs_item_ptr_offset(eb, slot); | |
349 | ||
350 | /* look for the key in the destination tree */ | |
351 | ret = btrfs_search_slot(NULL, root, key, path, 0, 0); | |
4bc4bee4 JB |
352 | if (ret < 0) |
353 | return ret; | |
354 | ||
e02119d5 CM |
355 | if (ret == 0) { |
356 | char *src_copy; | |
357 | char *dst_copy; | |
358 | u32 dst_size = btrfs_item_size_nr(path->nodes[0], | |
359 | path->slots[0]); | |
360 | if (dst_size != item_size) | |
361 | goto insert; | |
362 | ||
363 | if (item_size == 0) { | |
b3b4aa74 | 364 | btrfs_release_path(path); |
e02119d5 CM |
365 | return 0; |
366 | } | |
367 | dst_copy = kmalloc(item_size, GFP_NOFS); | |
368 | src_copy = kmalloc(item_size, GFP_NOFS); | |
2a29edc6 | 369 | if (!dst_copy || !src_copy) { |
b3b4aa74 | 370 | btrfs_release_path(path); |
2a29edc6 | 371 | kfree(dst_copy); |
372 | kfree(src_copy); | |
373 | return -ENOMEM; | |
374 | } | |
e02119d5 CM |
375 | |
376 | read_extent_buffer(eb, src_copy, src_ptr, item_size); | |
377 | ||
378 | dst_ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]); | |
379 | read_extent_buffer(path->nodes[0], dst_copy, dst_ptr, | |
380 | item_size); | |
381 | ret = memcmp(dst_copy, src_copy, item_size); | |
382 | ||
383 | kfree(dst_copy); | |
384 | kfree(src_copy); | |
385 | /* | |
386 | * they have the same contents, just return, this saves | |
387 | * us from cowing blocks in the destination tree and doing | |
388 | * extra writes that may not have been done by a previous | |
389 | * sync | |
390 | */ | |
391 | if (ret == 0) { | |
b3b4aa74 | 392 | btrfs_release_path(path); |
e02119d5 CM |
393 | return 0; |
394 | } | |
395 | ||
4bc4bee4 JB |
396 | /* |
397 | * We need to load the old nbytes into the inode so when we | |
398 | * replay the extents we've logged we get the right nbytes. | |
399 | */ | |
400 | if (inode_item) { | |
401 | struct btrfs_inode_item *item; | |
402 | u64 nbytes; | |
d555438b | 403 | u32 mode; |
4bc4bee4 JB |
404 | |
405 | item = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
406 | struct btrfs_inode_item); | |
407 | nbytes = btrfs_inode_nbytes(path->nodes[0], item); | |
408 | item = btrfs_item_ptr(eb, slot, | |
409 | struct btrfs_inode_item); | |
410 | btrfs_set_inode_nbytes(eb, item, nbytes); | |
d555438b JB |
411 | |
412 | /* | |
413 | * If this is a directory we need to reset the i_size to | |
414 | * 0 so that we can set it up properly when replaying | |
415 | * the rest of the items in this log. | |
416 | */ | |
417 | mode = btrfs_inode_mode(eb, item); | |
418 | if (S_ISDIR(mode)) | |
419 | btrfs_set_inode_size(eb, item, 0); | |
4bc4bee4 JB |
420 | } |
421 | } else if (inode_item) { | |
422 | struct btrfs_inode_item *item; | |
d555438b | 423 | u32 mode; |
4bc4bee4 JB |
424 | |
425 | /* | |
426 | * New inode, set nbytes to 0 so that the nbytes comes out | |
427 | * properly when we replay the extents. | |
428 | */ | |
429 | item = btrfs_item_ptr(eb, slot, struct btrfs_inode_item); | |
430 | btrfs_set_inode_nbytes(eb, item, 0); | |
d555438b JB |
431 | |
432 | /* | |
433 | * If this is a directory we need to reset the i_size to 0 so | |
434 | * that we can set it up properly when replaying the rest of | |
435 | * the items in this log. | |
436 | */ | |
437 | mode = btrfs_inode_mode(eb, item); | |
438 | if (S_ISDIR(mode)) | |
439 | btrfs_set_inode_size(eb, item, 0); | |
e02119d5 CM |
440 | } |
441 | insert: | |
b3b4aa74 | 442 | btrfs_release_path(path); |
e02119d5 | 443 | /* try to insert the key into the destination tree */ |
df8d116f | 444 | path->skip_release_on_error = 1; |
e02119d5 CM |
445 | ret = btrfs_insert_empty_item(trans, root, path, |
446 | key, item_size); | |
df8d116f | 447 | path->skip_release_on_error = 0; |
e02119d5 CM |
448 | |
449 | /* make sure any existing item is the correct size */ | |
df8d116f | 450 | if (ret == -EEXIST || ret == -EOVERFLOW) { |
e02119d5 CM |
451 | u32 found_size; |
452 | found_size = btrfs_item_size_nr(path->nodes[0], | |
453 | path->slots[0]); | |
143bede5 | 454 | if (found_size > item_size) |
2ff7e61e | 455 | btrfs_truncate_item(fs_info, path, item_size, 1); |
143bede5 | 456 | else if (found_size < item_size) |
2ff7e61e | 457 | btrfs_extend_item(fs_info, path, |
143bede5 | 458 | item_size - found_size); |
e02119d5 | 459 | } else if (ret) { |
4a500fd1 | 460 | return ret; |
e02119d5 CM |
461 | } |
462 | dst_ptr = btrfs_item_ptr_offset(path->nodes[0], | |
463 | path->slots[0]); | |
464 | ||
465 | /* don't overwrite an existing inode if the generation number | |
466 | * was logged as zero. This is done when the tree logging code | |
467 | * is just logging an inode to make sure it exists after recovery. | |
468 | * | |
469 | * Also, don't overwrite i_size on directories during replay. | |
470 | * log replay inserts and removes directory items based on the | |
471 | * state of the tree found in the subvolume, and i_size is modified | |
472 | * as it goes | |
473 | */ | |
474 | if (key->type == BTRFS_INODE_ITEM_KEY && ret == -EEXIST) { | |
475 | struct btrfs_inode_item *src_item; | |
476 | struct btrfs_inode_item *dst_item; | |
477 | ||
478 | src_item = (struct btrfs_inode_item *)src_ptr; | |
479 | dst_item = (struct btrfs_inode_item *)dst_ptr; | |
480 | ||
1a4bcf47 FM |
481 | if (btrfs_inode_generation(eb, src_item) == 0) { |
482 | struct extent_buffer *dst_eb = path->nodes[0]; | |
2f2ff0ee | 483 | const u64 ino_size = btrfs_inode_size(eb, src_item); |
1a4bcf47 | 484 | |
2f2ff0ee FM |
485 | /* |
486 | * For regular files an ino_size == 0 is used only when | |
487 | * logging that an inode exists, as part of a directory | |
488 | * fsync, and the inode wasn't fsynced before. In this | |
489 | * case don't set the size of the inode in the fs/subvol | |
490 | * tree, otherwise we would be throwing valid data away. | |
491 | */ | |
1a4bcf47 | 492 | if (S_ISREG(btrfs_inode_mode(eb, src_item)) && |
2f2ff0ee FM |
493 | S_ISREG(btrfs_inode_mode(dst_eb, dst_item)) && |
494 | ino_size != 0) { | |
1a4bcf47 | 495 | struct btrfs_map_token token; |
1a4bcf47 FM |
496 | |
497 | btrfs_init_map_token(&token); | |
498 | btrfs_set_token_inode_size(dst_eb, dst_item, | |
499 | ino_size, &token); | |
500 | } | |
e02119d5 | 501 | goto no_copy; |
1a4bcf47 | 502 | } |
e02119d5 CM |
503 | |
504 | if (overwrite_root && | |
505 | S_ISDIR(btrfs_inode_mode(eb, src_item)) && | |
506 | S_ISDIR(btrfs_inode_mode(path->nodes[0], dst_item))) { | |
507 | save_old_i_size = 1; | |
508 | saved_i_size = btrfs_inode_size(path->nodes[0], | |
509 | dst_item); | |
510 | } | |
511 | } | |
512 | ||
513 | copy_extent_buffer(path->nodes[0], eb, dst_ptr, | |
514 | src_ptr, item_size); | |
515 | ||
516 | if (save_old_i_size) { | |
517 | struct btrfs_inode_item *dst_item; | |
518 | dst_item = (struct btrfs_inode_item *)dst_ptr; | |
519 | btrfs_set_inode_size(path->nodes[0], dst_item, saved_i_size); | |
520 | } | |
521 | ||
522 | /* make sure the generation is filled in */ | |
523 | if (key->type == BTRFS_INODE_ITEM_KEY) { | |
524 | struct btrfs_inode_item *dst_item; | |
525 | dst_item = (struct btrfs_inode_item *)dst_ptr; | |
526 | if (btrfs_inode_generation(path->nodes[0], dst_item) == 0) { | |
527 | btrfs_set_inode_generation(path->nodes[0], dst_item, | |
528 | trans->transid); | |
529 | } | |
530 | } | |
531 | no_copy: | |
532 | btrfs_mark_buffer_dirty(path->nodes[0]); | |
b3b4aa74 | 533 | btrfs_release_path(path); |
e02119d5 CM |
534 | return 0; |
535 | } | |
536 | ||
537 | /* | |
538 | * simple helper to read an inode off the disk from a given root | |
539 | * This can only be called for subvolume roots and not for the log | |
540 | */ | |
541 | static noinline struct inode *read_one_inode(struct btrfs_root *root, | |
542 | u64 objectid) | |
543 | { | |
5d4f98a2 | 544 | struct btrfs_key key; |
e02119d5 | 545 | struct inode *inode; |
e02119d5 | 546 | |
5d4f98a2 YZ |
547 | key.objectid = objectid; |
548 | key.type = BTRFS_INODE_ITEM_KEY; | |
549 | key.offset = 0; | |
73f73415 | 550 | inode = btrfs_iget(root->fs_info->sb, &key, root, NULL); |
5d4f98a2 YZ |
551 | if (IS_ERR(inode)) { |
552 | inode = NULL; | |
553 | } else if (is_bad_inode(inode)) { | |
e02119d5 CM |
554 | iput(inode); |
555 | inode = NULL; | |
556 | } | |
557 | return inode; | |
558 | } | |
559 | ||
560 | /* replays a single extent in 'eb' at 'slot' with 'key' into the | |
561 | * subvolume 'root'. path is released on entry and should be released | |
562 | * on exit. | |
563 | * | |
564 | * extents in the log tree have not been allocated out of the extent | |
565 | * tree yet. So, this completes the allocation, taking a reference | |
566 | * as required if the extent already exists or creating a new extent | |
567 | * if it isn't in the extent allocation tree yet. | |
568 | * | |
569 | * The extent is inserted into the file, dropping any existing extents | |
570 | * from the file that overlap the new one. | |
571 | */ | |
572 | static noinline int replay_one_extent(struct btrfs_trans_handle *trans, | |
573 | struct btrfs_root *root, | |
574 | struct btrfs_path *path, | |
575 | struct extent_buffer *eb, int slot, | |
576 | struct btrfs_key *key) | |
577 | { | |
0b246afa | 578 | struct btrfs_fs_info *fs_info = root->fs_info; |
e02119d5 | 579 | int found_type; |
e02119d5 | 580 | u64 extent_end; |
e02119d5 | 581 | u64 start = key->offset; |
4bc4bee4 | 582 | u64 nbytes = 0; |
e02119d5 CM |
583 | struct btrfs_file_extent_item *item; |
584 | struct inode *inode = NULL; | |
585 | unsigned long size; | |
586 | int ret = 0; | |
587 | ||
588 | item = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item); | |
589 | found_type = btrfs_file_extent_type(eb, item); | |
590 | ||
d899e052 | 591 | if (found_type == BTRFS_FILE_EXTENT_REG || |
4bc4bee4 JB |
592 | found_type == BTRFS_FILE_EXTENT_PREALLOC) { |
593 | nbytes = btrfs_file_extent_num_bytes(eb, item); | |
594 | extent_end = start + nbytes; | |
595 | ||
596 | /* | |
597 | * We don't add to the inodes nbytes if we are prealloc or a | |
598 | * hole. | |
599 | */ | |
600 | if (btrfs_file_extent_disk_bytenr(eb, item) == 0) | |
601 | nbytes = 0; | |
602 | } else if (found_type == BTRFS_FILE_EXTENT_INLINE) { | |
514ac8ad | 603 | size = btrfs_file_extent_inline_len(eb, slot, item); |
4bc4bee4 | 604 | nbytes = btrfs_file_extent_ram_bytes(eb, item); |
da17066c | 605 | extent_end = ALIGN(start + size, |
0b246afa | 606 | fs_info->sectorsize); |
e02119d5 CM |
607 | } else { |
608 | ret = 0; | |
609 | goto out; | |
610 | } | |
611 | ||
612 | inode = read_one_inode(root, key->objectid); | |
613 | if (!inode) { | |
614 | ret = -EIO; | |
615 | goto out; | |
616 | } | |
617 | ||
618 | /* | |
619 | * first check to see if we already have this extent in the | |
620 | * file. This must be done before the btrfs_drop_extents run | |
621 | * so we don't try to drop this extent. | |
622 | */ | |
f85b7379 DS |
623 | ret = btrfs_lookup_file_extent(trans, root, path, |
624 | btrfs_ino(BTRFS_I(inode)), start, 0); | |
e02119d5 | 625 | |
d899e052 YZ |
626 | if (ret == 0 && |
627 | (found_type == BTRFS_FILE_EXTENT_REG || | |
628 | found_type == BTRFS_FILE_EXTENT_PREALLOC)) { | |
e02119d5 CM |
629 | struct btrfs_file_extent_item cmp1; |
630 | struct btrfs_file_extent_item cmp2; | |
631 | struct btrfs_file_extent_item *existing; | |
632 | struct extent_buffer *leaf; | |
633 | ||
634 | leaf = path->nodes[0]; | |
635 | existing = btrfs_item_ptr(leaf, path->slots[0], | |
636 | struct btrfs_file_extent_item); | |
637 | ||
638 | read_extent_buffer(eb, &cmp1, (unsigned long)item, | |
639 | sizeof(cmp1)); | |
640 | read_extent_buffer(leaf, &cmp2, (unsigned long)existing, | |
641 | sizeof(cmp2)); | |
642 | ||
643 | /* | |
644 | * we already have a pointer to this exact extent, | |
645 | * we don't have to do anything | |
646 | */ | |
647 | if (memcmp(&cmp1, &cmp2, sizeof(cmp1)) == 0) { | |
b3b4aa74 | 648 | btrfs_release_path(path); |
e02119d5 CM |
649 | goto out; |
650 | } | |
651 | } | |
b3b4aa74 | 652 | btrfs_release_path(path); |
e02119d5 CM |
653 | |
654 | /* drop any overlapping extents */ | |
2671485d | 655 | ret = btrfs_drop_extents(trans, root, inode, start, extent_end, 1); |
3650860b JB |
656 | if (ret) |
657 | goto out; | |
e02119d5 | 658 | |
07d400a6 YZ |
659 | if (found_type == BTRFS_FILE_EXTENT_REG || |
660 | found_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
5d4f98a2 | 661 | u64 offset; |
07d400a6 YZ |
662 | unsigned long dest_offset; |
663 | struct btrfs_key ins; | |
664 | ||
3168021c FM |
665 | if (btrfs_file_extent_disk_bytenr(eb, item) == 0 && |
666 | btrfs_fs_incompat(fs_info, NO_HOLES)) | |
667 | goto update_inode; | |
668 | ||
07d400a6 YZ |
669 | ret = btrfs_insert_empty_item(trans, root, path, key, |
670 | sizeof(*item)); | |
3650860b JB |
671 | if (ret) |
672 | goto out; | |
07d400a6 YZ |
673 | dest_offset = btrfs_item_ptr_offset(path->nodes[0], |
674 | path->slots[0]); | |
675 | copy_extent_buffer(path->nodes[0], eb, dest_offset, | |
676 | (unsigned long)item, sizeof(*item)); | |
677 | ||
678 | ins.objectid = btrfs_file_extent_disk_bytenr(eb, item); | |
679 | ins.offset = btrfs_file_extent_disk_num_bytes(eb, item); | |
680 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
5d4f98a2 | 681 | offset = key->offset - btrfs_file_extent_offset(eb, item); |
07d400a6 | 682 | |
df2c95f3 QW |
683 | /* |
684 | * Manually record dirty extent, as here we did a shallow | |
685 | * file extent item copy and skip normal backref update, | |
686 | * but modifying extent tree all by ourselves. | |
687 | * So need to manually record dirty extent for qgroup, | |
688 | * as the owner of the file extent changed from log tree | |
689 | * (doesn't affect qgroup) to fs/file tree(affects qgroup) | |
690 | */ | |
0b246afa | 691 | ret = btrfs_qgroup_trace_extent(trans, fs_info, |
df2c95f3 QW |
692 | btrfs_file_extent_disk_bytenr(eb, item), |
693 | btrfs_file_extent_disk_num_bytes(eb, item), | |
694 | GFP_NOFS); | |
695 | if (ret < 0) | |
696 | goto out; | |
697 | ||
07d400a6 YZ |
698 | if (ins.objectid > 0) { |
699 | u64 csum_start; | |
700 | u64 csum_end; | |
701 | LIST_HEAD(ordered_sums); | |
702 | /* | |
703 | * is this extent already allocated in the extent | |
704 | * allocation tree? If so, just add a reference | |
705 | */ | |
2ff7e61e | 706 | ret = btrfs_lookup_data_extent(fs_info, ins.objectid, |
07d400a6 YZ |
707 | ins.offset); |
708 | if (ret == 0) { | |
84f7d8e6 | 709 | ret = btrfs_inc_extent_ref(trans, root, |
07d400a6 | 710 | ins.objectid, ins.offset, |
5d4f98a2 | 711 | 0, root->root_key.objectid, |
b06c4bf5 | 712 | key->objectid, offset); |
b50c6e25 JB |
713 | if (ret) |
714 | goto out; | |
07d400a6 YZ |
715 | } else { |
716 | /* | |
717 | * insert the extent pointer in the extent | |
718 | * allocation tree | |
719 | */ | |
5d4f98a2 | 720 | ret = btrfs_alloc_logged_file_extent(trans, |
2ff7e61e JM |
721 | fs_info, |
722 | root->root_key.objectid, | |
5d4f98a2 | 723 | key->objectid, offset, &ins); |
b50c6e25 JB |
724 | if (ret) |
725 | goto out; | |
07d400a6 | 726 | } |
b3b4aa74 | 727 | btrfs_release_path(path); |
07d400a6 YZ |
728 | |
729 | if (btrfs_file_extent_compression(eb, item)) { | |
730 | csum_start = ins.objectid; | |
731 | csum_end = csum_start + ins.offset; | |
732 | } else { | |
733 | csum_start = ins.objectid + | |
734 | btrfs_file_extent_offset(eb, item); | |
735 | csum_end = csum_start + | |
736 | btrfs_file_extent_num_bytes(eb, item); | |
737 | } | |
738 | ||
739 | ret = btrfs_lookup_csums_range(root->log_root, | |
740 | csum_start, csum_end - 1, | |
a2de733c | 741 | &ordered_sums, 0); |
3650860b JB |
742 | if (ret) |
743 | goto out; | |
b84b8390 FM |
744 | /* |
745 | * Now delete all existing cums in the csum root that | |
746 | * cover our range. We do this because we can have an | |
747 | * extent that is completely referenced by one file | |
748 | * extent item and partially referenced by another | |
749 | * file extent item (like after using the clone or | |
750 | * extent_same ioctls). In this case if we end up doing | |
751 | * the replay of the one that partially references the | |
752 | * extent first, and we do not do the csum deletion | |
753 | * below, we can get 2 csum items in the csum tree that | |
754 | * overlap each other. For example, imagine our log has | |
755 | * the two following file extent items: | |
756 | * | |
757 | * key (257 EXTENT_DATA 409600) | |
758 | * extent data disk byte 12845056 nr 102400 | |
759 | * extent data offset 20480 nr 20480 ram 102400 | |
760 | * | |
761 | * key (257 EXTENT_DATA 819200) | |
762 | * extent data disk byte 12845056 nr 102400 | |
763 | * extent data offset 0 nr 102400 ram 102400 | |
764 | * | |
765 | * Where the second one fully references the 100K extent | |
766 | * that starts at disk byte 12845056, and the log tree | |
767 | * has a single csum item that covers the entire range | |
768 | * of the extent: | |
769 | * | |
770 | * key (EXTENT_CSUM EXTENT_CSUM 12845056) itemsize 100 | |
771 | * | |
772 | * After the first file extent item is replayed, the | |
773 | * csum tree gets the following csum item: | |
774 | * | |
775 | * key (EXTENT_CSUM EXTENT_CSUM 12865536) itemsize 20 | |
776 | * | |
777 | * Which covers the 20K sub-range starting at offset 20K | |
778 | * of our extent. Now when we replay the second file | |
779 | * extent item, if we do not delete existing csum items | |
780 | * that cover any of its blocks, we end up getting two | |
781 | * csum items in our csum tree that overlap each other: | |
782 | * | |
783 | * key (EXTENT_CSUM EXTENT_CSUM 12845056) itemsize 100 | |
784 | * key (EXTENT_CSUM EXTENT_CSUM 12865536) itemsize 20 | |
785 | * | |
786 | * Which is a problem, because after this anyone trying | |
787 | * to lookup up for the checksum of any block of our | |
788 | * extent starting at an offset of 40K or higher, will | |
789 | * end up looking at the second csum item only, which | |
790 | * does not contain the checksum for any block starting | |
791 | * at offset 40K or higher of our extent. | |
792 | */ | |
07d400a6 YZ |
793 | while (!list_empty(&ordered_sums)) { |
794 | struct btrfs_ordered_sum *sums; | |
795 | sums = list_entry(ordered_sums.next, | |
796 | struct btrfs_ordered_sum, | |
797 | list); | |
b84b8390 | 798 | if (!ret) |
0b246afa | 799 | ret = btrfs_del_csums(trans, fs_info, |
5b4aacef JM |
800 | sums->bytenr, |
801 | sums->len); | |
3650860b JB |
802 | if (!ret) |
803 | ret = btrfs_csum_file_blocks(trans, | |
0b246afa | 804 | fs_info->csum_root, sums); |
07d400a6 YZ |
805 | list_del(&sums->list); |
806 | kfree(sums); | |
807 | } | |
3650860b JB |
808 | if (ret) |
809 | goto out; | |
07d400a6 | 810 | } else { |
b3b4aa74 | 811 | btrfs_release_path(path); |
07d400a6 YZ |
812 | } |
813 | } else if (found_type == BTRFS_FILE_EXTENT_INLINE) { | |
814 | /* inline extents are easy, we just overwrite them */ | |
815 | ret = overwrite_item(trans, root, path, eb, slot, key); | |
3650860b JB |
816 | if (ret) |
817 | goto out; | |
07d400a6 | 818 | } |
e02119d5 | 819 | |
4bc4bee4 | 820 | inode_add_bytes(inode, nbytes); |
3168021c | 821 | update_inode: |
b9959295 | 822 | ret = btrfs_update_inode(trans, root, inode); |
e02119d5 CM |
823 | out: |
824 | if (inode) | |
825 | iput(inode); | |
826 | return ret; | |
827 | } | |
828 | ||
829 | /* | |
830 | * when cleaning up conflicts between the directory names in the | |
831 | * subvolume, directory names in the log and directory names in the | |
832 | * inode back references, we may have to unlink inodes from directories. | |
833 | * | |
834 | * This is a helper function to do the unlink of a specific directory | |
835 | * item | |
836 | */ | |
837 | static noinline int drop_one_dir_item(struct btrfs_trans_handle *trans, | |
838 | struct btrfs_root *root, | |
839 | struct btrfs_path *path, | |
207e7d92 | 840 | struct btrfs_inode *dir, |
e02119d5 CM |
841 | struct btrfs_dir_item *di) |
842 | { | |
843 | struct inode *inode; | |
844 | char *name; | |
845 | int name_len; | |
846 | struct extent_buffer *leaf; | |
847 | struct btrfs_key location; | |
848 | int ret; | |
849 | ||
850 | leaf = path->nodes[0]; | |
851 | ||
852 | btrfs_dir_item_key_to_cpu(leaf, di, &location); | |
853 | name_len = btrfs_dir_name_len(leaf, di); | |
854 | name = kmalloc(name_len, GFP_NOFS); | |
2a29edc6 | 855 | if (!name) |
856 | return -ENOMEM; | |
857 | ||
e02119d5 | 858 | read_extent_buffer(leaf, name, (unsigned long)(di + 1), name_len); |
b3b4aa74 | 859 | btrfs_release_path(path); |
e02119d5 CM |
860 | |
861 | inode = read_one_inode(root, location.objectid); | |
c00e9493 | 862 | if (!inode) { |
3650860b JB |
863 | ret = -EIO; |
864 | goto out; | |
c00e9493 | 865 | } |
e02119d5 | 866 | |
ec051c0f | 867 | ret = link_to_fixup_dir(trans, root, path, location.objectid); |
3650860b JB |
868 | if (ret) |
869 | goto out; | |
12fcfd22 | 870 | |
207e7d92 NB |
871 | ret = btrfs_unlink_inode(trans, root, dir, BTRFS_I(inode), name, |
872 | name_len); | |
3650860b JB |
873 | if (ret) |
874 | goto out; | |
ada9af21 | 875 | else |
e5c304e6 | 876 | ret = btrfs_run_delayed_items(trans); |
3650860b | 877 | out: |
e02119d5 | 878 | kfree(name); |
e02119d5 CM |
879 | iput(inode); |
880 | return ret; | |
881 | } | |
882 | ||
883 | /* | |
884 | * helper function to see if a given name and sequence number found | |
885 | * in an inode back reference are already in a directory and correctly | |
886 | * point to this inode | |
887 | */ | |
888 | static noinline int inode_in_dir(struct btrfs_root *root, | |
889 | struct btrfs_path *path, | |
890 | u64 dirid, u64 objectid, u64 index, | |
891 | const char *name, int name_len) | |
892 | { | |
893 | struct btrfs_dir_item *di; | |
894 | struct btrfs_key location; | |
895 | int match = 0; | |
896 | ||
897 | di = btrfs_lookup_dir_index_item(NULL, root, path, dirid, | |
898 | index, name, name_len, 0); | |
899 | if (di && !IS_ERR(di)) { | |
900 | btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location); | |
901 | if (location.objectid != objectid) | |
902 | goto out; | |
903 | } else | |
904 | goto out; | |
b3b4aa74 | 905 | btrfs_release_path(path); |
e02119d5 CM |
906 | |
907 | di = btrfs_lookup_dir_item(NULL, root, path, dirid, name, name_len, 0); | |
908 | if (di && !IS_ERR(di)) { | |
909 | btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location); | |
910 | if (location.objectid != objectid) | |
911 | goto out; | |
912 | } else | |
913 | goto out; | |
914 | match = 1; | |
915 | out: | |
b3b4aa74 | 916 | btrfs_release_path(path); |
e02119d5 CM |
917 | return match; |
918 | } | |
919 | ||
920 | /* | |
921 | * helper function to check a log tree for a named back reference in | |
922 | * an inode. This is used to decide if a back reference that is | |
923 | * found in the subvolume conflicts with what we find in the log. | |
924 | * | |
925 | * inode backreferences may have multiple refs in a single item, | |
926 | * during replay we process one reference at a time, and we don't | |
927 | * want to delete valid links to a file from the subvolume if that | |
928 | * link is also in the log. | |
929 | */ | |
930 | static noinline int backref_in_log(struct btrfs_root *log, | |
931 | struct btrfs_key *key, | |
f186373f | 932 | u64 ref_objectid, |
df8d116f | 933 | const char *name, int namelen) |
e02119d5 CM |
934 | { |
935 | struct btrfs_path *path; | |
936 | struct btrfs_inode_ref *ref; | |
937 | unsigned long ptr; | |
938 | unsigned long ptr_end; | |
939 | unsigned long name_ptr; | |
940 | int found_name_len; | |
941 | int item_size; | |
942 | int ret; | |
943 | int match = 0; | |
944 | ||
945 | path = btrfs_alloc_path(); | |
2a29edc6 | 946 | if (!path) |
947 | return -ENOMEM; | |
948 | ||
e02119d5 CM |
949 | ret = btrfs_search_slot(NULL, log, key, path, 0, 0); |
950 | if (ret != 0) | |
951 | goto out; | |
952 | ||
e02119d5 | 953 | ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]); |
f186373f MF |
954 | |
955 | if (key->type == BTRFS_INODE_EXTREF_KEY) { | |
1f250e92 FM |
956 | if (btrfs_find_name_in_ext_backref(path->nodes[0], |
957 | path->slots[0], | |
958 | ref_objectid, | |
f186373f MF |
959 | name, namelen, NULL)) |
960 | match = 1; | |
961 | ||
962 | goto out; | |
963 | } | |
964 | ||
965 | item_size = btrfs_item_size_nr(path->nodes[0], path->slots[0]); | |
e02119d5 CM |
966 | ptr_end = ptr + item_size; |
967 | while (ptr < ptr_end) { | |
968 | ref = (struct btrfs_inode_ref *)ptr; | |
969 | found_name_len = btrfs_inode_ref_name_len(path->nodes[0], ref); | |
970 | if (found_name_len == namelen) { | |
971 | name_ptr = (unsigned long)(ref + 1); | |
972 | ret = memcmp_extent_buffer(path->nodes[0], name, | |
973 | name_ptr, namelen); | |
974 | if (ret == 0) { | |
975 | match = 1; | |
976 | goto out; | |
977 | } | |
978 | } | |
979 | ptr = (unsigned long)(ref + 1) + found_name_len; | |
980 | } | |
981 | out: | |
982 | btrfs_free_path(path); | |
983 | return match; | |
984 | } | |
985 | ||
5a1d7843 | 986 | static inline int __add_inode_ref(struct btrfs_trans_handle *trans, |
e02119d5 | 987 | struct btrfs_root *root, |
e02119d5 | 988 | struct btrfs_path *path, |
5a1d7843 | 989 | struct btrfs_root *log_root, |
94c91a1f NB |
990 | struct btrfs_inode *dir, |
991 | struct btrfs_inode *inode, | |
f186373f MF |
992 | u64 inode_objectid, u64 parent_objectid, |
993 | u64 ref_index, char *name, int namelen, | |
994 | int *search_done) | |
e02119d5 | 995 | { |
34f3e4f2 | 996 | int ret; |
f186373f MF |
997 | char *victim_name; |
998 | int victim_name_len; | |
999 | struct extent_buffer *leaf; | |
5a1d7843 | 1000 | struct btrfs_dir_item *di; |
f186373f MF |
1001 | struct btrfs_key search_key; |
1002 | struct btrfs_inode_extref *extref; | |
c622ae60 | 1003 | |
f186373f MF |
1004 | again: |
1005 | /* Search old style refs */ | |
1006 | search_key.objectid = inode_objectid; | |
1007 | search_key.type = BTRFS_INODE_REF_KEY; | |
1008 | search_key.offset = parent_objectid; | |
1009 | ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0); | |
e02119d5 | 1010 | if (ret == 0) { |
e02119d5 CM |
1011 | struct btrfs_inode_ref *victim_ref; |
1012 | unsigned long ptr; | |
1013 | unsigned long ptr_end; | |
f186373f MF |
1014 | |
1015 | leaf = path->nodes[0]; | |
e02119d5 CM |
1016 | |
1017 | /* are we trying to overwrite a back ref for the root directory | |
1018 | * if so, just jump out, we're done | |
1019 | */ | |
f186373f | 1020 | if (search_key.objectid == search_key.offset) |
5a1d7843 | 1021 | return 1; |
e02119d5 CM |
1022 | |
1023 | /* check all the names in this back reference to see | |
1024 | * if they are in the log. if so, we allow them to stay | |
1025 | * otherwise they must be unlinked as a conflict | |
1026 | */ | |
1027 | ptr = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
1028 | ptr_end = ptr + btrfs_item_size_nr(leaf, path->slots[0]); | |
d397712b | 1029 | while (ptr < ptr_end) { |
e02119d5 CM |
1030 | victim_ref = (struct btrfs_inode_ref *)ptr; |
1031 | victim_name_len = btrfs_inode_ref_name_len(leaf, | |
1032 | victim_ref); | |
1033 | victim_name = kmalloc(victim_name_len, GFP_NOFS); | |
3650860b JB |
1034 | if (!victim_name) |
1035 | return -ENOMEM; | |
e02119d5 CM |
1036 | |
1037 | read_extent_buffer(leaf, victim_name, | |
1038 | (unsigned long)(victim_ref + 1), | |
1039 | victim_name_len); | |
1040 | ||
f186373f MF |
1041 | if (!backref_in_log(log_root, &search_key, |
1042 | parent_objectid, | |
1043 | victim_name, | |
e02119d5 | 1044 | victim_name_len)) { |
94c91a1f | 1045 | inc_nlink(&inode->vfs_inode); |
b3b4aa74 | 1046 | btrfs_release_path(path); |
12fcfd22 | 1047 | |
94c91a1f | 1048 | ret = btrfs_unlink_inode(trans, root, dir, inode, |
4ec5934e | 1049 | victim_name, victim_name_len); |
f186373f | 1050 | kfree(victim_name); |
3650860b JB |
1051 | if (ret) |
1052 | return ret; | |
e5c304e6 | 1053 | ret = btrfs_run_delayed_items(trans); |
ada9af21 FDBM |
1054 | if (ret) |
1055 | return ret; | |
f186373f MF |
1056 | *search_done = 1; |
1057 | goto again; | |
e02119d5 CM |
1058 | } |
1059 | kfree(victim_name); | |
f186373f | 1060 | |
e02119d5 CM |
1061 | ptr = (unsigned long)(victim_ref + 1) + victim_name_len; |
1062 | } | |
e02119d5 | 1063 | |
c622ae60 | 1064 | /* |
1065 | * NOTE: we have searched root tree and checked the | |
bb7ab3b9 | 1066 | * corresponding ref, it does not need to check again. |
c622ae60 | 1067 | */ |
5a1d7843 | 1068 | *search_done = 1; |
e02119d5 | 1069 | } |
b3b4aa74 | 1070 | btrfs_release_path(path); |
e02119d5 | 1071 | |
f186373f MF |
1072 | /* Same search but for extended refs */ |
1073 | extref = btrfs_lookup_inode_extref(NULL, root, path, name, namelen, | |
1074 | inode_objectid, parent_objectid, 0, | |
1075 | 0); | |
1076 | if (!IS_ERR_OR_NULL(extref)) { | |
1077 | u32 item_size; | |
1078 | u32 cur_offset = 0; | |
1079 | unsigned long base; | |
1080 | struct inode *victim_parent; | |
1081 | ||
1082 | leaf = path->nodes[0]; | |
1083 | ||
1084 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1085 | base = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
1086 | ||
1087 | while (cur_offset < item_size) { | |
dd9ef135 | 1088 | extref = (struct btrfs_inode_extref *)(base + cur_offset); |
f186373f MF |
1089 | |
1090 | victim_name_len = btrfs_inode_extref_name_len(leaf, extref); | |
1091 | ||
1092 | if (btrfs_inode_extref_parent(leaf, extref) != parent_objectid) | |
1093 | goto next; | |
1094 | ||
1095 | victim_name = kmalloc(victim_name_len, GFP_NOFS); | |
3650860b JB |
1096 | if (!victim_name) |
1097 | return -ENOMEM; | |
f186373f MF |
1098 | read_extent_buffer(leaf, victim_name, (unsigned long)&extref->name, |
1099 | victim_name_len); | |
1100 | ||
1101 | search_key.objectid = inode_objectid; | |
1102 | search_key.type = BTRFS_INODE_EXTREF_KEY; | |
1103 | search_key.offset = btrfs_extref_hash(parent_objectid, | |
1104 | victim_name, | |
1105 | victim_name_len); | |
1106 | ret = 0; | |
1107 | if (!backref_in_log(log_root, &search_key, | |
1108 | parent_objectid, victim_name, | |
1109 | victim_name_len)) { | |
1110 | ret = -ENOENT; | |
1111 | victim_parent = read_one_inode(root, | |
94c91a1f | 1112 | parent_objectid); |
f186373f | 1113 | if (victim_parent) { |
94c91a1f | 1114 | inc_nlink(&inode->vfs_inode); |
f186373f MF |
1115 | btrfs_release_path(path); |
1116 | ||
1117 | ret = btrfs_unlink_inode(trans, root, | |
4ec5934e | 1118 | BTRFS_I(victim_parent), |
94c91a1f | 1119 | inode, |
4ec5934e NB |
1120 | victim_name, |
1121 | victim_name_len); | |
ada9af21 FDBM |
1122 | if (!ret) |
1123 | ret = btrfs_run_delayed_items( | |
e5c304e6 | 1124 | trans); |
f186373f | 1125 | } |
f186373f MF |
1126 | iput(victim_parent); |
1127 | kfree(victim_name); | |
3650860b JB |
1128 | if (ret) |
1129 | return ret; | |
f186373f MF |
1130 | *search_done = 1; |
1131 | goto again; | |
1132 | } | |
1133 | kfree(victim_name); | |
f186373f MF |
1134 | next: |
1135 | cur_offset += victim_name_len + sizeof(*extref); | |
1136 | } | |
1137 | *search_done = 1; | |
1138 | } | |
1139 | btrfs_release_path(path); | |
1140 | ||
34f3e4f2 | 1141 | /* look for a conflicting sequence number */ |
94c91a1f | 1142 | di = btrfs_lookup_dir_index_item(trans, root, path, btrfs_ino(dir), |
f186373f | 1143 | ref_index, name, namelen, 0); |
34f3e4f2 | 1144 | if (di && !IS_ERR(di)) { |
94c91a1f | 1145 | ret = drop_one_dir_item(trans, root, path, dir, di); |
3650860b JB |
1146 | if (ret) |
1147 | return ret; | |
34f3e4f2 | 1148 | } |
1149 | btrfs_release_path(path); | |
1150 | ||
1151 | /* look for a conflicing name */ | |
94c91a1f | 1152 | di = btrfs_lookup_dir_item(trans, root, path, btrfs_ino(dir), |
34f3e4f2 | 1153 | name, namelen, 0); |
1154 | if (di && !IS_ERR(di)) { | |
94c91a1f | 1155 | ret = drop_one_dir_item(trans, root, path, dir, di); |
3650860b JB |
1156 | if (ret) |
1157 | return ret; | |
34f3e4f2 | 1158 | } |
1159 | btrfs_release_path(path); | |
1160 | ||
5a1d7843 JS |
1161 | return 0; |
1162 | } | |
e02119d5 | 1163 | |
bae15d95 QW |
1164 | static int extref_get_fields(struct extent_buffer *eb, unsigned long ref_ptr, |
1165 | u32 *namelen, char **name, u64 *index, | |
1166 | u64 *parent_objectid) | |
f186373f MF |
1167 | { |
1168 | struct btrfs_inode_extref *extref; | |
1169 | ||
1170 | extref = (struct btrfs_inode_extref *)ref_ptr; | |
1171 | ||
1172 | *namelen = btrfs_inode_extref_name_len(eb, extref); | |
1173 | *name = kmalloc(*namelen, GFP_NOFS); | |
1174 | if (*name == NULL) | |
1175 | return -ENOMEM; | |
1176 | ||
1177 | read_extent_buffer(eb, *name, (unsigned long)&extref->name, | |
1178 | *namelen); | |
1179 | ||
1f250e92 FM |
1180 | if (index) |
1181 | *index = btrfs_inode_extref_index(eb, extref); | |
f186373f MF |
1182 | if (parent_objectid) |
1183 | *parent_objectid = btrfs_inode_extref_parent(eb, extref); | |
1184 | ||
1185 | return 0; | |
1186 | } | |
1187 | ||
bae15d95 QW |
1188 | static int ref_get_fields(struct extent_buffer *eb, unsigned long ref_ptr, |
1189 | u32 *namelen, char **name, u64 *index) | |
f186373f MF |
1190 | { |
1191 | struct btrfs_inode_ref *ref; | |
1192 | ||
1193 | ref = (struct btrfs_inode_ref *)ref_ptr; | |
1194 | ||
1195 | *namelen = btrfs_inode_ref_name_len(eb, ref); | |
1196 | *name = kmalloc(*namelen, GFP_NOFS); | |
1197 | if (*name == NULL) | |
1198 | return -ENOMEM; | |
1199 | ||
1200 | read_extent_buffer(eb, *name, (unsigned long)(ref + 1), *namelen); | |
1201 | ||
1f250e92 FM |
1202 | if (index) |
1203 | *index = btrfs_inode_ref_index(eb, ref); | |
f186373f MF |
1204 | |
1205 | return 0; | |
1206 | } | |
1207 | ||
1f250e92 FM |
1208 | /* |
1209 | * Take an inode reference item from the log tree and iterate all names from the | |
1210 | * inode reference item in the subvolume tree with the same key (if it exists). | |
1211 | * For any name that is not in the inode reference item from the log tree, do a | |
1212 | * proper unlink of that name (that is, remove its entry from the inode | |
1213 | * reference item and both dir index keys). | |
1214 | */ | |
1215 | static int unlink_old_inode_refs(struct btrfs_trans_handle *trans, | |
1216 | struct btrfs_root *root, | |
1217 | struct btrfs_path *path, | |
1218 | struct btrfs_inode *inode, | |
1219 | struct extent_buffer *log_eb, | |
1220 | int log_slot, | |
1221 | struct btrfs_key *key) | |
1222 | { | |
1223 | int ret; | |
1224 | unsigned long ref_ptr; | |
1225 | unsigned long ref_end; | |
1226 | struct extent_buffer *eb; | |
1227 | ||
1228 | again: | |
1229 | btrfs_release_path(path); | |
1230 | ret = btrfs_search_slot(NULL, root, key, path, 0, 0); | |
1231 | if (ret > 0) { | |
1232 | ret = 0; | |
1233 | goto out; | |
1234 | } | |
1235 | if (ret < 0) | |
1236 | goto out; | |
1237 | ||
1238 | eb = path->nodes[0]; | |
1239 | ref_ptr = btrfs_item_ptr_offset(eb, path->slots[0]); | |
1240 | ref_end = ref_ptr + btrfs_item_size_nr(eb, path->slots[0]); | |
1241 | while (ref_ptr < ref_end) { | |
1242 | char *name = NULL; | |
1243 | int namelen; | |
1244 | u64 parent_id; | |
1245 | ||
1246 | if (key->type == BTRFS_INODE_EXTREF_KEY) { | |
1247 | ret = extref_get_fields(eb, ref_ptr, &namelen, &name, | |
1248 | NULL, &parent_id); | |
1249 | } else { | |
1250 | parent_id = key->offset; | |
1251 | ret = ref_get_fields(eb, ref_ptr, &namelen, &name, | |
1252 | NULL); | |
1253 | } | |
1254 | if (ret) | |
1255 | goto out; | |
1256 | ||
1257 | if (key->type == BTRFS_INODE_EXTREF_KEY) | |
1258 | ret = btrfs_find_name_in_ext_backref(log_eb, log_slot, | |
1259 | parent_id, name, | |
1260 | namelen, NULL); | |
1261 | else | |
1262 | ret = btrfs_find_name_in_backref(log_eb, log_slot, name, | |
1263 | namelen, NULL); | |
1264 | ||
1265 | if (!ret) { | |
1266 | struct inode *dir; | |
1267 | ||
1268 | btrfs_release_path(path); | |
1269 | dir = read_one_inode(root, parent_id); | |
1270 | if (!dir) { | |
1271 | ret = -ENOENT; | |
1272 | kfree(name); | |
1273 | goto out; | |
1274 | } | |
1275 | ret = btrfs_unlink_inode(trans, root, BTRFS_I(dir), | |
1276 | inode, name, namelen); | |
1277 | kfree(name); | |
1278 | iput(dir); | |
1279 | if (ret) | |
1280 | goto out; | |
1281 | goto again; | |
1282 | } | |
1283 | ||
1284 | kfree(name); | |
1285 | ref_ptr += namelen; | |
1286 | if (key->type == BTRFS_INODE_EXTREF_KEY) | |
1287 | ref_ptr += sizeof(struct btrfs_inode_extref); | |
1288 | else | |
1289 | ref_ptr += sizeof(struct btrfs_inode_ref); | |
1290 | } | |
1291 | ret = 0; | |
1292 | out: | |
1293 | btrfs_release_path(path); | |
1294 | return ret; | |
1295 | } | |
1296 | ||
5a1d7843 JS |
1297 | /* |
1298 | * replay one inode back reference item found in the log tree. | |
1299 | * eb, slot and key refer to the buffer and key found in the log tree. | |
1300 | * root is the destination we are replaying into, and path is for temp | |
1301 | * use by this function. (it should be released on return). | |
1302 | */ | |
1303 | static noinline int add_inode_ref(struct btrfs_trans_handle *trans, | |
1304 | struct btrfs_root *root, | |
1305 | struct btrfs_root *log, | |
1306 | struct btrfs_path *path, | |
1307 | struct extent_buffer *eb, int slot, | |
1308 | struct btrfs_key *key) | |
1309 | { | |
03b2f08b GB |
1310 | struct inode *dir = NULL; |
1311 | struct inode *inode = NULL; | |
5a1d7843 JS |
1312 | unsigned long ref_ptr; |
1313 | unsigned long ref_end; | |
03b2f08b | 1314 | char *name = NULL; |
5a1d7843 JS |
1315 | int namelen; |
1316 | int ret; | |
1317 | int search_done = 0; | |
f186373f MF |
1318 | int log_ref_ver = 0; |
1319 | u64 parent_objectid; | |
1320 | u64 inode_objectid; | |
f46dbe3d | 1321 | u64 ref_index = 0; |
f186373f MF |
1322 | int ref_struct_size; |
1323 | ||
1324 | ref_ptr = btrfs_item_ptr_offset(eb, slot); | |
1325 | ref_end = ref_ptr + btrfs_item_size_nr(eb, slot); | |
1326 | ||
1327 | if (key->type == BTRFS_INODE_EXTREF_KEY) { | |
1328 | struct btrfs_inode_extref *r; | |
1329 | ||
1330 | ref_struct_size = sizeof(struct btrfs_inode_extref); | |
1331 | log_ref_ver = 1; | |
1332 | r = (struct btrfs_inode_extref *)ref_ptr; | |
1333 | parent_objectid = btrfs_inode_extref_parent(eb, r); | |
1334 | } else { | |
1335 | ref_struct_size = sizeof(struct btrfs_inode_ref); | |
1336 | parent_objectid = key->offset; | |
1337 | } | |
1338 | inode_objectid = key->objectid; | |
e02119d5 | 1339 | |
5a1d7843 JS |
1340 | /* |
1341 | * it is possible that we didn't log all the parent directories | |
1342 | * for a given inode. If we don't find the dir, just don't | |
1343 | * copy the back ref in. The link count fixup code will take | |
1344 | * care of the rest | |
1345 | */ | |
f186373f | 1346 | dir = read_one_inode(root, parent_objectid); |
03b2f08b GB |
1347 | if (!dir) { |
1348 | ret = -ENOENT; | |
1349 | goto out; | |
1350 | } | |
5a1d7843 | 1351 | |
f186373f | 1352 | inode = read_one_inode(root, inode_objectid); |
5a1d7843 | 1353 | if (!inode) { |
03b2f08b GB |
1354 | ret = -EIO; |
1355 | goto out; | |
5a1d7843 JS |
1356 | } |
1357 | ||
5a1d7843 | 1358 | while (ref_ptr < ref_end) { |
f186373f | 1359 | if (log_ref_ver) { |
bae15d95 QW |
1360 | ret = extref_get_fields(eb, ref_ptr, &namelen, &name, |
1361 | &ref_index, &parent_objectid); | |
f186373f MF |
1362 | /* |
1363 | * parent object can change from one array | |
1364 | * item to another. | |
1365 | */ | |
1366 | if (!dir) | |
1367 | dir = read_one_inode(root, parent_objectid); | |
03b2f08b GB |
1368 | if (!dir) { |
1369 | ret = -ENOENT; | |
1370 | goto out; | |
1371 | } | |
f186373f | 1372 | } else { |
bae15d95 QW |
1373 | ret = ref_get_fields(eb, ref_ptr, &namelen, &name, |
1374 | &ref_index); | |
f186373f MF |
1375 | } |
1376 | if (ret) | |
03b2f08b | 1377 | goto out; |
5a1d7843 JS |
1378 | |
1379 | /* if we already have a perfect match, we're done */ | |
f85b7379 DS |
1380 | if (!inode_in_dir(root, path, btrfs_ino(BTRFS_I(dir)), |
1381 | btrfs_ino(BTRFS_I(inode)), ref_index, | |
1382 | name, namelen)) { | |
5a1d7843 JS |
1383 | /* |
1384 | * look for a conflicting back reference in the | |
1385 | * metadata. if we find one we have to unlink that name | |
1386 | * of the file before we add our new link. Later on, we | |
1387 | * overwrite any existing back reference, and we don't | |
1388 | * want to create dangling pointers in the directory. | |
1389 | */ | |
1390 | ||
1391 | if (!search_done) { | |
1392 | ret = __add_inode_ref(trans, root, path, log, | |
94c91a1f | 1393 | BTRFS_I(dir), |
d75eefdf | 1394 | BTRFS_I(inode), |
f186373f MF |
1395 | inode_objectid, |
1396 | parent_objectid, | |
1397 | ref_index, name, namelen, | |
5a1d7843 | 1398 | &search_done); |
03b2f08b GB |
1399 | if (ret) { |
1400 | if (ret == 1) | |
1401 | ret = 0; | |
3650860b JB |
1402 | goto out; |
1403 | } | |
5a1d7843 JS |
1404 | } |
1405 | ||
1406 | /* insert our name */ | |
db0a669f NB |
1407 | ret = btrfs_add_link(trans, BTRFS_I(dir), |
1408 | BTRFS_I(inode), | |
1409 | name, namelen, 0, ref_index); | |
3650860b JB |
1410 | if (ret) |
1411 | goto out; | |
5a1d7843 JS |
1412 | |
1413 | btrfs_update_inode(trans, root, inode); | |
1414 | } | |
1415 | ||
f186373f | 1416 | ref_ptr = (unsigned long)(ref_ptr + ref_struct_size) + namelen; |
5a1d7843 | 1417 | kfree(name); |
03b2f08b | 1418 | name = NULL; |
f186373f MF |
1419 | if (log_ref_ver) { |
1420 | iput(dir); | |
1421 | dir = NULL; | |
1422 | } | |
5a1d7843 | 1423 | } |
e02119d5 | 1424 | |
1f250e92 FM |
1425 | /* |
1426 | * Before we overwrite the inode reference item in the subvolume tree | |
1427 | * with the item from the log tree, we must unlink all names from the | |
1428 | * parent directory that are in the subvolume's tree inode reference | |
1429 | * item, otherwise we end up with an inconsistent subvolume tree where | |
1430 | * dir index entries exist for a name but there is no inode reference | |
1431 | * item with the same name. | |
1432 | */ | |
1433 | ret = unlink_old_inode_refs(trans, root, path, BTRFS_I(inode), eb, slot, | |
1434 | key); | |
1435 | if (ret) | |
1436 | goto out; | |
1437 | ||
e02119d5 CM |
1438 | /* finally write the back reference in the inode */ |
1439 | ret = overwrite_item(trans, root, path, eb, slot, key); | |
5a1d7843 | 1440 | out: |
b3b4aa74 | 1441 | btrfs_release_path(path); |
03b2f08b | 1442 | kfree(name); |
e02119d5 CM |
1443 | iput(dir); |
1444 | iput(inode); | |
3650860b | 1445 | return ret; |
e02119d5 CM |
1446 | } |
1447 | ||
c71bf099 | 1448 | static int insert_orphan_item(struct btrfs_trans_handle *trans, |
9c4f61f0 | 1449 | struct btrfs_root *root, u64 ino) |
c71bf099 YZ |
1450 | { |
1451 | int ret; | |
381cf658 | 1452 | |
9c4f61f0 DS |
1453 | ret = btrfs_insert_orphan_item(trans, root, ino); |
1454 | if (ret == -EEXIST) | |
1455 | ret = 0; | |
381cf658 | 1456 | |
c71bf099 YZ |
1457 | return ret; |
1458 | } | |
1459 | ||
f186373f | 1460 | static int count_inode_extrefs(struct btrfs_root *root, |
36283658 | 1461 | struct btrfs_inode *inode, struct btrfs_path *path) |
f186373f MF |
1462 | { |
1463 | int ret = 0; | |
1464 | int name_len; | |
1465 | unsigned int nlink = 0; | |
1466 | u32 item_size; | |
1467 | u32 cur_offset = 0; | |
36283658 | 1468 | u64 inode_objectid = btrfs_ino(inode); |
f186373f MF |
1469 | u64 offset = 0; |
1470 | unsigned long ptr; | |
1471 | struct btrfs_inode_extref *extref; | |
1472 | struct extent_buffer *leaf; | |
1473 | ||
1474 | while (1) { | |
1475 | ret = btrfs_find_one_extref(root, inode_objectid, offset, path, | |
1476 | &extref, &offset); | |
1477 | if (ret) | |
1478 | break; | |
c71bf099 | 1479 | |
f186373f MF |
1480 | leaf = path->nodes[0]; |
1481 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1482 | ptr = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
2c2c452b | 1483 | cur_offset = 0; |
f186373f MF |
1484 | |
1485 | while (cur_offset < item_size) { | |
1486 | extref = (struct btrfs_inode_extref *) (ptr + cur_offset); | |
1487 | name_len = btrfs_inode_extref_name_len(leaf, extref); | |
1488 | ||
1489 | nlink++; | |
1490 | ||
1491 | cur_offset += name_len + sizeof(*extref); | |
1492 | } | |
1493 | ||
1494 | offset++; | |
1495 | btrfs_release_path(path); | |
1496 | } | |
1497 | btrfs_release_path(path); | |
1498 | ||
2c2c452b | 1499 | if (ret < 0 && ret != -ENOENT) |
f186373f MF |
1500 | return ret; |
1501 | return nlink; | |
1502 | } | |
1503 | ||
1504 | static int count_inode_refs(struct btrfs_root *root, | |
f329e319 | 1505 | struct btrfs_inode *inode, struct btrfs_path *path) |
e02119d5 | 1506 | { |
e02119d5 CM |
1507 | int ret; |
1508 | struct btrfs_key key; | |
f186373f | 1509 | unsigned int nlink = 0; |
e02119d5 CM |
1510 | unsigned long ptr; |
1511 | unsigned long ptr_end; | |
1512 | int name_len; | |
f329e319 | 1513 | u64 ino = btrfs_ino(inode); |
e02119d5 | 1514 | |
33345d01 | 1515 | key.objectid = ino; |
e02119d5 CM |
1516 | key.type = BTRFS_INODE_REF_KEY; |
1517 | key.offset = (u64)-1; | |
1518 | ||
d397712b | 1519 | while (1) { |
e02119d5 CM |
1520 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
1521 | if (ret < 0) | |
1522 | break; | |
1523 | if (ret > 0) { | |
1524 | if (path->slots[0] == 0) | |
1525 | break; | |
1526 | path->slots[0]--; | |
1527 | } | |
e93ae26f | 1528 | process_slot: |
e02119d5 CM |
1529 | btrfs_item_key_to_cpu(path->nodes[0], &key, |
1530 | path->slots[0]); | |
33345d01 | 1531 | if (key.objectid != ino || |
e02119d5 CM |
1532 | key.type != BTRFS_INODE_REF_KEY) |
1533 | break; | |
1534 | ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]); | |
1535 | ptr_end = ptr + btrfs_item_size_nr(path->nodes[0], | |
1536 | path->slots[0]); | |
d397712b | 1537 | while (ptr < ptr_end) { |
e02119d5 CM |
1538 | struct btrfs_inode_ref *ref; |
1539 | ||
1540 | ref = (struct btrfs_inode_ref *)ptr; | |
1541 | name_len = btrfs_inode_ref_name_len(path->nodes[0], | |
1542 | ref); | |
1543 | ptr = (unsigned long)(ref + 1) + name_len; | |
1544 | nlink++; | |
1545 | } | |
1546 | ||
1547 | if (key.offset == 0) | |
1548 | break; | |
e93ae26f FDBM |
1549 | if (path->slots[0] > 0) { |
1550 | path->slots[0]--; | |
1551 | goto process_slot; | |
1552 | } | |
e02119d5 | 1553 | key.offset--; |
b3b4aa74 | 1554 | btrfs_release_path(path); |
e02119d5 | 1555 | } |
b3b4aa74 | 1556 | btrfs_release_path(path); |
f186373f MF |
1557 | |
1558 | return nlink; | |
1559 | } | |
1560 | ||
1561 | /* | |
1562 | * There are a few corners where the link count of the file can't | |
1563 | * be properly maintained during replay. So, instead of adding | |
1564 | * lots of complexity to the log code, we just scan the backrefs | |
1565 | * for any file that has been through replay. | |
1566 | * | |
1567 | * The scan will update the link count on the inode to reflect the | |
1568 | * number of back refs found. If it goes down to zero, the iput | |
1569 | * will free the inode. | |
1570 | */ | |
1571 | static noinline int fixup_inode_link_count(struct btrfs_trans_handle *trans, | |
1572 | struct btrfs_root *root, | |
1573 | struct inode *inode) | |
1574 | { | |
1575 | struct btrfs_path *path; | |
1576 | int ret; | |
1577 | u64 nlink = 0; | |
4a0cc7ca | 1578 | u64 ino = btrfs_ino(BTRFS_I(inode)); |
f186373f MF |
1579 | |
1580 | path = btrfs_alloc_path(); | |
1581 | if (!path) | |
1582 | return -ENOMEM; | |
1583 | ||
f329e319 | 1584 | ret = count_inode_refs(root, BTRFS_I(inode), path); |
f186373f MF |
1585 | if (ret < 0) |
1586 | goto out; | |
1587 | ||
1588 | nlink = ret; | |
1589 | ||
36283658 | 1590 | ret = count_inode_extrefs(root, BTRFS_I(inode), path); |
f186373f MF |
1591 | if (ret < 0) |
1592 | goto out; | |
1593 | ||
1594 | nlink += ret; | |
1595 | ||
1596 | ret = 0; | |
1597 | ||
e02119d5 | 1598 | if (nlink != inode->i_nlink) { |
bfe86848 | 1599 | set_nlink(inode, nlink); |
e02119d5 CM |
1600 | btrfs_update_inode(trans, root, inode); |
1601 | } | |
8d5bf1cb | 1602 | BTRFS_I(inode)->index_cnt = (u64)-1; |
e02119d5 | 1603 | |
c71bf099 YZ |
1604 | if (inode->i_nlink == 0) { |
1605 | if (S_ISDIR(inode->i_mode)) { | |
1606 | ret = replay_dir_deletes(trans, root, NULL, path, | |
33345d01 | 1607 | ino, 1); |
3650860b JB |
1608 | if (ret) |
1609 | goto out; | |
c71bf099 | 1610 | } |
33345d01 | 1611 | ret = insert_orphan_item(trans, root, ino); |
12fcfd22 | 1612 | } |
12fcfd22 | 1613 | |
f186373f MF |
1614 | out: |
1615 | btrfs_free_path(path); | |
1616 | return ret; | |
e02119d5 CM |
1617 | } |
1618 | ||
1619 | static noinline int fixup_inode_link_counts(struct btrfs_trans_handle *trans, | |
1620 | struct btrfs_root *root, | |
1621 | struct btrfs_path *path) | |
1622 | { | |
1623 | int ret; | |
1624 | struct btrfs_key key; | |
1625 | struct inode *inode; | |
1626 | ||
1627 | key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID; | |
1628 | key.type = BTRFS_ORPHAN_ITEM_KEY; | |
1629 | key.offset = (u64)-1; | |
d397712b | 1630 | while (1) { |
e02119d5 CM |
1631 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
1632 | if (ret < 0) | |
1633 | break; | |
1634 | ||
1635 | if (ret == 1) { | |
1636 | if (path->slots[0] == 0) | |
1637 | break; | |
1638 | path->slots[0]--; | |
1639 | } | |
1640 | ||
1641 | btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]); | |
1642 | if (key.objectid != BTRFS_TREE_LOG_FIXUP_OBJECTID || | |
1643 | key.type != BTRFS_ORPHAN_ITEM_KEY) | |
1644 | break; | |
1645 | ||
1646 | ret = btrfs_del_item(trans, root, path); | |
65a246c5 TI |
1647 | if (ret) |
1648 | goto out; | |
e02119d5 | 1649 | |
b3b4aa74 | 1650 | btrfs_release_path(path); |
e02119d5 | 1651 | inode = read_one_inode(root, key.offset); |
c00e9493 TI |
1652 | if (!inode) |
1653 | return -EIO; | |
e02119d5 CM |
1654 | |
1655 | ret = fixup_inode_link_count(trans, root, inode); | |
e02119d5 | 1656 | iput(inode); |
3650860b JB |
1657 | if (ret) |
1658 | goto out; | |
e02119d5 | 1659 | |
12fcfd22 CM |
1660 | /* |
1661 | * fixup on a directory may create new entries, | |
1662 | * make sure we always look for the highset possible | |
1663 | * offset | |
1664 | */ | |
1665 | key.offset = (u64)-1; | |
e02119d5 | 1666 | } |
65a246c5 TI |
1667 | ret = 0; |
1668 | out: | |
b3b4aa74 | 1669 | btrfs_release_path(path); |
65a246c5 | 1670 | return ret; |
e02119d5 CM |
1671 | } |
1672 | ||
1673 | ||
1674 | /* | |
1675 | * record a given inode in the fixup dir so we can check its link | |
1676 | * count when replay is done. The link count is incremented here | |
1677 | * so the inode won't go away until we check it | |
1678 | */ | |
1679 | static noinline int link_to_fixup_dir(struct btrfs_trans_handle *trans, | |
1680 | struct btrfs_root *root, | |
1681 | struct btrfs_path *path, | |
1682 | u64 objectid) | |
1683 | { | |
1684 | struct btrfs_key key; | |
1685 | int ret = 0; | |
1686 | struct inode *inode; | |
1687 | ||
1688 | inode = read_one_inode(root, objectid); | |
c00e9493 TI |
1689 | if (!inode) |
1690 | return -EIO; | |
e02119d5 CM |
1691 | |
1692 | key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID; | |
962a298f | 1693 | key.type = BTRFS_ORPHAN_ITEM_KEY; |
e02119d5 CM |
1694 | key.offset = objectid; |
1695 | ||
1696 | ret = btrfs_insert_empty_item(trans, root, path, &key, 0); | |
1697 | ||
b3b4aa74 | 1698 | btrfs_release_path(path); |
e02119d5 | 1699 | if (ret == 0) { |
9bf7a489 JB |
1700 | if (!inode->i_nlink) |
1701 | set_nlink(inode, 1); | |
1702 | else | |
8b558c5f | 1703 | inc_nlink(inode); |
b9959295 | 1704 | ret = btrfs_update_inode(trans, root, inode); |
e02119d5 CM |
1705 | } else if (ret == -EEXIST) { |
1706 | ret = 0; | |
1707 | } else { | |
3650860b | 1708 | BUG(); /* Logic Error */ |
e02119d5 CM |
1709 | } |
1710 | iput(inode); | |
1711 | ||
1712 | return ret; | |
1713 | } | |
1714 | ||
1715 | /* | |
1716 | * when replaying the log for a directory, we only insert names | |
1717 | * for inodes that actually exist. This means an fsync on a directory | |
1718 | * does not implicitly fsync all the new files in it | |
1719 | */ | |
1720 | static noinline int insert_one_name(struct btrfs_trans_handle *trans, | |
1721 | struct btrfs_root *root, | |
e02119d5 | 1722 | u64 dirid, u64 index, |
60d53eb3 | 1723 | char *name, int name_len, |
e02119d5 CM |
1724 | struct btrfs_key *location) |
1725 | { | |
1726 | struct inode *inode; | |
1727 | struct inode *dir; | |
1728 | int ret; | |
1729 | ||
1730 | inode = read_one_inode(root, location->objectid); | |
1731 | if (!inode) | |
1732 | return -ENOENT; | |
1733 | ||
1734 | dir = read_one_inode(root, dirid); | |
1735 | if (!dir) { | |
1736 | iput(inode); | |
1737 | return -EIO; | |
1738 | } | |
d555438b | 1739 | |
db0a669f NB |
1740 | ret = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode), name, |
1741 | name_len, 1, index); | |
e02119d5 CM |
1742 | |
1743 | /* FIXME, put inode into FIXUP list */ | |
1744 | ||
1745 | iput(inode); | |
1746 | iput(dir); | |
1747 | return ret; | |
1748 | } | |
1749 | ||
df8d116f FM |
1750 | /* |
1751 | * Return true if an inode reference exists in the log for the given name, | |
1752 | * inode and parent inode. | |
1753 | */ | |
1754 | static bool name_in_log_ref(struct btrfs_root *log_root, | |
1755 | const char *name, const int name_len, | |
1756 | const u64 dirid, const u64 ino) | |
1757 | { | |
1758 | struct btrfs_key search_key; | |
1759 | ||
1760 | search_key.objectid = ino; | |
1761 | search_key.type = BTRFS_INODE_REF_KEY; | |
1762 | search_key.offset = dirid; | |
1763 | if (backref_in_log(log_root, &search_key, dirid, name, name_len)) | |
1764 | return true; | |
1765 | ||
1766 | search_key.type = BTRFS_INODE_EXTREF_KEY; | |
1767 | search_key.offset = btrfs_extref_hash(dirid, name, name_len); | |
1768 | if (backref_in_log(log_root, &search_key, dirid, name, name_len)) | |
1769 | return true; | |
1770 | ||
1771 | return false; | |
1772 | } | |
1773 | ||
e02119d5 CM |
1774 | /* |
1775 | * take a single entry in a log directory item and replay it into | |
1776 | * the subvolume. | |
1777 | * | |
1778 | * if a conflicting item exists in the subdirectory already, | |
1779 | * the inode it points to is unlinked and put into the link count | |
1780 | * fix up tree. | |
1781 | * | |
1782 | * If a name from the log points to a file or directory that does | |
1783 | * not exist in the FS, it is skipped. fsyncs on directories | |
1784 | * do not force down inodes inside that directory, just changes to the | |
1785 | * names or unlinks in a directory. | |
bb53eda9 FM |
1786 | * |
1787 | * Returns < 0 on error, 0 if the name wasn't replayed (dentry points to a | |
1788 | * non-existing inode) and 1 if the name was replayed. | |
e02119d5 CM |
1789 | */ |
1790 | static noinline int replay_one_name(struct btrfs_trans_handle *trans, | |
1791 | struct btrfs_root *root, | |
1792 | struct btrfs_path *path, | |
1793 | struct extent_buffer *eb, | |
1794 | struct btrfs_dir_item *di, | |
1795 | struct btrfs_key *key) | |
1796 | { | |
1797 | char *name; | |
1798 | int name_len; | |
1799 | struct btrfs_dir_item *dst_di; | |
1800 | struct btrfs_key found_key; | |
1801 | struct btrfs_key log_key; | |
1802 | struct inode *dir; | |
e02119d5 | 1803 | u8 log_type; |
4bef0848 | 1804 | int exists; |
3650860b | 1805 | int ret = 0; |
d555438b | 1806 | bool update_size = (key->type == BTRFS_DIR_INDEX_KEY); |
bb53eda9 | 1807 | bool name_added = false; |
e02119d5 CM |
1808 | |
1809 | dir = read_one_inode(root, key->objectid); | |
c00e9493 TI |
1810 | if (!dir) |
1811 | return -EIO; | |
e02119d5 CM |
1812 | |
1813 | name_len = btrfs_dir_name_len(eb, di); | |
1814 | name = kmalloc(name_len, GFP_NOFS); | |
2bac325e FDBM |
1815 | if (!name) { |
1816 | ret = -ENOMEM; | |
1817 | goto out; | |
1818 | } | |
2a29edc6 | 1819 | |
e02119d5 CM |
1820 | log_type = btrfs_dir_type(eb, di); |
1821 | read_extent_buffer(eb, name, (unsigned long)(di + 1), | |
1822 | name_len); | |
1823 | ||
1824 | btrfs_dir_item_key_to_cpu(eb, di, &log_key); | |
4bef0848 CM |
1825 | exists = btrfs_lookup_inode(trans, root, path, &log_key, 0); |
1826 | if (exists == 0) | |
1827 | exists = 1; | |
1828 | else | |
1829 | exists = 0; | |
b3b4aa74 | 1830 | btrfs_release_path(path); |
4bef0848 | 1831 | |
e02119d5 CM |
1832 | if (key->type == BTRFS_DIR_ITEM_KEY) { |
1833 | dst_di = btrfs_lookup_dir_item(trans, root, path, key->objectid, | |
1834 | name, name_len, 1); | |
d397712b | 1835 | } else if (key->type == BTRFS_DIR_INDEX_KEY) { |
e02119d5 CM |
1836 | dst_di = btrfs_lookup_dir_index_item(trans, root, path, |
1837 | key->objectid, | |
1838 | key->offset, name, | |
1839 | name_len, 1); | |
1840 | } else { | |
3650860b JB |
1841 | /* Corruption */ |
1842 | ret = -EINVAL; | |
1843 | goto out; | |
e02119d5 | 1844 | } |
c704005d | 1845 | if (IS_ERR_OR_NULL(dst_di)) { |
e02119d5 CM |
1846 | /* we need a sequence number to insert, so we only |
1847 | * do inserts for the BTRFS_DIR_INDEX_KEY types | |
1848 | */ | |
1849 | if (key->type != BTRFS_DIR_INDEX_KEY) | |
1850 | goto out; | |
1851 | goto insert; | |
1852 | } | |
1853 | ||
1854 | btrfs_dir_item_key_to_cpu(path->nodes[0], dst_di, &found_key); | |
1855 | /* the existing item matches the logged item */ | |
1856 | if (found_key.objectid == log_key.objectid && | |
1857 | found_key.type == log_key.type && | |
1858 | found_key.offset == log_key.offset && | |
1859 | btrfs_dir_type(path->nodes[0], dst_di) == log_type) { | |
a2cc11db | 1860 | update_size = false; |
e02119d5 CM |
1861 | goto out; |
1862 | } | |
1863 | ||
1864 | /* | |
1865 | * don't drop the conflicting directory entry if the inode | |
1866 | * for the new entry doesn't exist | |
1867 | */ | |
4bef0848 | 1868 | if (!exists) |
e02119d5 CM |
1869 | goto out; |
1870 | ||
207e7d92 | 1871 | ret = drop_one_dir_item(trans, root, path, BTRFS_I(dir), dst_di); |
3650860b JB |
1872 | if (ret) |
1873 | goto out; | |
e02119d5 CM |
1874 | |
1875 | if (key->type == BTRFS_DIR_INDEX_KEY) | |
1876 | goto insert; | |
1877 | out: | |
b3b4aa74 | 1878 | btrfs_release_path(path); |
d555438b | 1879 | if (!ret && update_size) { |
6ef06d27 | 1880 | btrfs_i_size_write(BTRFS_I(dir), dir->i_size + name_len * 2); |
d555438b JB |
1881 | ret = btrfs_update_inode(trans, root, dir); |
1882 | } | |
e02119d5 CM |
1883 | kfree(name); |
1884 | iput(dir); | |
bb53eda9 FM |
1885 | if (!ret && name_added) |
1886 | ret = 1; | |
3650860b | 1887 | return ret; |
e02119d5 CM |
1888 | |
1889 | insert: | |
df8d116f FM |
1890 | if (name_in_log_ref(root->log_root, name, name_len, |
1891 | key->objectid, log_key.objectid)) { | |
1892 | /* The dentry will be added later. */ | |
1893 | ret = 0; | |
1894 | update_size = false; | |
1895 | goto out; | |
1896 | } | |
b3b4aa74 | 1897 | btrfs_release_path(path); |
60d53eb3 Z |
1898 | ret = insert_one_name(trans, root, key->objectid, key->offset, |
1899 | name, name_len, &log_key); | |
df8d116f | 1900 | if (ret && ret != -ENOENT && ret != -EEXIST) |
3650860b | 1901 | goto out; |
bb53eda9 FM |
1902 | if (!ret) |
1903 | name_added = true; | |
d555438b | 1904 | update_size = false; |
3650860b | 1905 | ret = 0; |
e02119d5 CM |
1906 | goto out; |
1907 | } | |
1908 | ||
1909 | /* | |
1910 | * find all the names in a directory item and reconcile them into | |
1911 | * the subvolume. Only BTRFS_DIR_ITEM_KEY types will have more than | |
1912 | * one name in a directory item, but the same code gets used for | |
1913 | * both directory index types | |
1914 | */ | |
1915 | static noinline int replay_one_dir_item(struct btrfs_trans_handle *trans, | |
1916 | struct btrfs_root *root, | |
1917 | struct btrfs_path *path, | |
1918 | struct extent_buffer *eb, int slot, | |
1919 | struct btrfs_key *key) | |
1920 | { | |
bb53eda9 | 1921 | int ret = 0; |
e02119d5 CM |
1922 | u32 item_size = btrfs_item_size_nr(eb, slot); |
1923 | struct btrfs_dir_item *di; | |
1924 | int name_len; | |
1925 | unsigned long ptr; | |
1926 | unsigned long ptr_end; | |
bb53eda9 | 1927 | struct btrfs_path *fixup_path = NULL; |
e02119d5 CM |
1928 | |
1929 | ptr = btrfs_item_ptr_offset(eb, slot); | |
1930 | ptr_end = ptr + item_size; | |
d397712b | 1931 | while (ptr < ptr_end) { |
e02119d5 CM |
1932 | di = (struct btrfs_dir_item *)ptr; |
1933 | name_len = btrfs_dir_name_len(eb, di); | |
1934 | ret = replay_one_name(trans, root, path, eb, di, key); | |
bb53eda9 FM |
1935 | if (ret < 0) |
1936 | break; | |
e02119d5 CM |
1937 | ptr = (unsigned long)(di + 1); |
1938 | ptr += name_len; | |
bb53eda9 FM |
1939 | |
1940 | /* | |
1941 | * If this entry refers to a non-directory (directories can not | |
1942 | * have a link count > 1) and it was added in the transaction | |
1943 | * that was not committed, make sure we fixup the link count of | |
1944 | * the inode it the entry points to. Otherwise something like | |
1945 | * the following would result in a directory pointing to an | |
1946 | * inode with a wrong link that does not account for this dir | |
1947 | * entry: | |
1948 | * | |
1949 | * mkdir testdir | |
1950 | * touch testdir/foo | |
1951 | * touch testdir/bar | |
1952 | * sync | |
1953 | * | |
1954 | * ln testdir/bar testdir/bar_link | |
1955 | * ln testdir/foo testdir/foo_link | |
1956 | * xfs_io -c "fsync" testdir/bar | |
1957 | * | |
1958 | * <power failure> | |
1959 | * | |
1960 | * mount fs, log replay happens | |
1961 | * | |
1962 | * File foo would remain with a link count of 1 when it has two | |
1963 | * entries pointing to it in the directory testdir. This would | |
1964 | * make it impossible to ever delete the parent directory has | |
1965 | * it would result in stale dentries that can never be deleted. | |
1966 | */ | |
1967 | if (ret == 1 && btrfs_dir_type(eb, di) != BTRFS_FT_DIR) { | |
1968 | struct btrfs_key di_key; | |
1969 | ||
1970 | if (!fixup_path) { | |
1971 | fixup_path = btrfs_alloc_path(); | |
1972 | if (!fixup_path) { | |
1973 | ret = -ENOMEM; | |
1974 | break; | |
1975 | } | |
1976 | } | |
1977 | ||
1978 | btrfs_dir_item_key_to_cpu(eb, di, &di_key); | |
1979 | ret = link_to_fixup_dir(trans, root, fixup_path, | |
1980 | di_key.objectid); | |
1981 | if (ret) | |
1982 | break; | |
1983 | } | |
1984 | ret = 0; | |
e02119d5 | 1985 | } |
bb53eda9 FM |
1986 | btrfs_free_path(fixup_path); |
1987 | return ret; | |
e02119d5 CM |
1988 | } |
1989 | ||
1990 | /* | |
1991 | * directory replay has two parts. There are the standard directory | |
1992 | * items in the log copied from the subvolume, and range items | |
1993 | * created in the log while the subvolume was logged. | |
1994 | * | |
1995 | * The range items tell us which parts of the key space the log | |
1996 | * is authoritative for. During replay, if a key in the subvolume | |
1997 | * directory is in a logged range item, but not actually in the log | |
1998 | * that means it was deleted from the directory before the fsync | |
1999 | * and should be removed. | |
2000 | */ | |
2001 | static noinline int find_dir_range(struct btrfs_root *root, | |
2002 | struct btrfs_path *path, | |
2003 | u64 dirid, int key_type, | |
2004 | u64 *start_ret, u64 *end_ret) | |
2005 | { | |
2006 | struct btrfs_key key; | |
2007 | u64 found_end; | |
2008 | struct btrfs_dir_log_item *item; | |
2009 | int ret; | |
2010 | int nritems; | |
2011 | ||
2012 | if (*start_ret == (u64)-1) | |
2013 | return 1; | |
2014 | ||
2015 | key.objectid = dirid; | |
2016 | key.type = key_type; | |
2017 | key.offset = *start_ret; | |
2018 | ||
2019 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
2020 | if (ret < 0) | |
2021 | goto out; | |
2022 | if (ret > 0) { | |
2023 | if (path->slots[0] == 0) | |
2024 | goto out; | |
2025 | path->slots[0]--; | |
2026 | } | |
2027 | if (ret != 0) | |
2028 | btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]); | |
2029 | ||
2030 | if (key.type != key_type || key.objectid != dirid) { | |
2031 | ret = 1; | |
2032 | goto next; | |
2033 | } | |
2034 | item = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
2035 | struct btrfs_dir_log_item); | |
2036 | found_end = btrfs_dir_log_end(path->nodes[0], item); | |
2037 | ||
2038 | if (*start_ret >= key.offset && *start_ret <= found_end) { | |
2039 | ret = 0; | |
2040 | *start_ret = key.offset; | |
2041 | *end_ret = found_end; | |
2042 | goto out; | |
2043 | } | |
2044 | ret = 1; | |
2045 | next: | |
2046 | /* check the next slot in the tree to see if it is a valid item */ | |
2047 | nritems = btrfs_header_nritems(path->nodes[0]); | |
2a7bf53f | 2048 | path->slots[0]++; |
e02119d5 CM |
2049 | if (path->slots[0] >= nritems) { |
2050 | ret = btrfs_next_leaf(root, path); | |
2051 | if (ret) | |
2052 | goto out; | |
e02119d5 CM |
2053 | } |
2054 | ||
2055 | btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]); | |
2056 | ||
2057 | if (key.type != key_type || key.objectid != dirid) { | |
2058 | ret = 1; | |
2059 | goto out; | |
2060 | } | |
2061 | item = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
2062 | struct btrfs_dir_log_item); | |
2063 | found_end = btrfs_dir_log_end(path->nodes[0], item); | |
2064 | *start_ret = key.offset; | |
2065 | *end_ret = found_end; | |
2066 | ret = 0; | |
2067 | out: | |
b3b4aa74 | 2068 | btrfs_release_path(path); |
e02119d5 CM |
2069 | return ret; |
2070 | } | |
2071 | ||
2072 | /* | |
2073 | * this looks for a given directory item in the log. If the directory | |
2074 | * item is not in the log, the item is removed and the inode it points | |
2075 | * to is unlinked | |
2076 | */ | |
2077 | static noinline int check_item_in_log(struct btrfs_trans_handle *trans, | |
2078 | struct btrfs_root *root, | |
2079 | struct btrfs_root *log, | |
2080 | struct btrfs_path *path, | |
2081 | struct btrfs_path *log_path, | |
2082 | struct inode *dir, | |
2083 | struct btrfs_key *dir_key) | |
2084 | { | |
2085 | int ret; | |
2086 | struct extent_buffer *eb; | |
2087 | int slot; | |
2088 | u32 item_size; | |
2089 | struct btrfs_dir_item *di; | |
2090 | struct btrfs_dir_item *log_di; | |
2091 | int name_len; | |
2092 | unsigned long ptr; | |
2093 | unsigned long ptr_end; | |
2094 | char *name; | |
2095 | struct inode *inode; | |
2096 | struct btrfs_key location; | |
2097 | ||
2098 | again: | |
2099 | eb = path->nodes[0]; | |
2100 | slot = path->slots[0]; | |
2101 | item_size = btrfs_item_size_nr(eb, slot); | |
2102 | ptr = btrfs_item_ptr_offset(eb, slot); | |
2103 | ptr_end = ptr + item_size; | |
d397712b | 2104 | while (ptr < ptr_end) { |
e02119d5 CM |
2105 | di = (struct btrfs_dir_item *)ptr; |
2106 | name_len = btrfs_dir_name_len(eb, di); | |
2107 | name = kmalloc(name_len, GFP_NOFS); | |
2108 | if (!name) { | |
2109 | ret = -ENOMEM; | |
2110 | goto out; | |
2111 | } | |
2112 | read_extent_buffer(eb, name, (unsigned long)(di + 1), | |
2113 | name_len); | |
2114 | log_di = NULL; | |
12fcfd22 | 2115 | if (log && dir_key->type == BTRFS_DIR_ITEM_KEY) { |
e02119d5 CM |
2116 | log_di = btrfs_lookup_dir_item(trans, log, log_path, |
2117 | dir_key->objectid, | |
2118 | name, name_len, 0); | |
12fcfd22 | 2119 | } else if (log && dir_key->type == BTRFS_DIR_INDEX_KEY) { |
e02119d5 CM |
2120 | log_di = btrfs_lookup_dir_index_item(trans, log, |
2121 | log_path, | |
2122 | dir_key->objectid, | |
2123 | dir_key->offset, | |
2124 | name, name_len, 0); | |
2125 | } | |
269d040f | 2126 | if (!log_di || (IS_ERR(log_di) && PTR_ERR(log_di) == -ENOENT)) { |
e02119d5 | 2127 | btrfs_dir_item_key_to_cpu(eb, di, &location); |
b3b4aa74 DS |
2128 | btrfs_release_path(path); |
2129 | btrfs_release_path(log_path); | |
e02119d5 | 2130 | inode = read_one_inode(root, location.objectid); |
c00e9493 TI |
2131 | if (!inode) { |
2132 | kfree(name); | |
2133 | return -EIO; | |
2134 | } | |
e02119d5 CM |
2135 | |
2136 | ret = link_to_fixup_dir(trans, root, | |
2137 | path, location.objectid); | |
3650860b JB |
2138 | if (ret) { |
2139 | kfree(name); | |
2140 | iput(inode); | |
2141 | goto out; | |
2142 | } | |
2143 | ||
8b558c5f | 2144 | inc_nlink(inode); |
4ec5934e NB |
2145 | ret = btrfs_unlink_inode(trans, root, BTRFS_I(dir), |
2146 | BTRFS_I(inode), name, name_len); | |
3650860b | 2147 | if (!ret) |
e5c304e6 | 2148 | ret = btrfs_run_delayed_items(trans); |
e02119d5 CM |
2149 | kfree(name); |
2150 | iput(inode); | |
3650860b JB |
2151 | if (ret) |
2152 | goto out; | |
e02119d5 CM |
2153 | |
2154 | /* there might still be more names under this key | |
2155 | * check and repeat if required | |
2156 | */ | |
2157 | ret = btrfs_search_slot(NULL, root, dir_key, path, | |
2158 | 0, 0); | |
2159 | if (ret == 0) | |
2160 | goto again; | |
2161 | ret = 0; | |
2162 | goto out; | |
269d040f FDBM |
2163 | } else if (IS_ERR(log_di)) { |
2164 | kfree(name); | |
2165 | return PTR_ERR(log_di); | |
e02119d5 | 2166 | } |
b3b4aa74 | 2167 | btrfs_release_path(log_path); |
e02119d5 CM |
2168 | kfree(name); |
2169 | ||
2170 | ptr = (unsigned long)(di + 1); | |
2171 | ptr += name_len; | |
2172 | } | |
2173 | ret = 0; | |
2174 | out: | |
b3b4aa74 DS |
2175 | btrfs_release_path(path); |
2176 | btrfs_release_path(log_path); | |
e02119d5 CM |
2177 | return ret; |
2178 | } | |
2179 | ||
4f764e51 FM |
2180 | static int replay_xattr_deletes(struct btrfs_trans_handle *trans, |
2181 | struct btrfs_root *root, | |
2182 | struct btrfs_root *log, | |
2183 | struct btrfs_path *path, | |
2184 | const u64 ino) | |
2185 | { | |
2186 | struct btrfs_key search_key; | |
2187 | struct btrfs_path *log_path; | |
2188 | int i; | |
2189 | int nritems; | |
2190 | int ret; | |
2191 | ||
2192 | log_path = btrfs_alloc_path(); | |
2193 | if (!log_path) | |
2194 | return -ENOMEM; | |
2195 | ||
2196 | search_key.objectid = ino; | |
2197 | search_key.type = BTRFS_XATTR_ITEM_KEY; | |
2198 | search_key.offset = 0; | |
2199 | again: | |
2200 | ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0); | |
2201 | if (ret < 0) | |
2202 | goto out; | |
2203 | process_leaf: | |
2204 | nritems = btrfs_header_nritems(path->nodes[0]); | |
2205 | for (i = path->slots[0]; i < nritems; i++) { | |
2206 | struct btrfs_key key; | |
2207 | struct btrfs_dir_item *di; | |
2208 | struct btrfs_dir_item *log_di; | |
2209 | u32 total_size; | |
2210 | u32 cur; | |
2211 | ||
2212 | btrfs_item_key_to_cpu(path->nodes[0], &key, i); | |
2213 | if (key.objectid != ino || key.type != BTRFS_XATTR_ITEM_KEY) { | |
2214 | ret = 0; | |
2215 | goto out; | |
2216 | } | |
2217 | ||
2218 | di = btrfs_item_ptr(path->nodes[0], i, struct btrfs_dir_item); | |
2219 | total_size = btrfs_item_size_nr(path->nodes[0], i); | |
2220 | cur = 0; | |
2221 | while (cur < total_size) { | |
2222 | u16 name_len = btrfs_dir_name_len(path->nodes[0], di); | |
2223 | u16 data_len = btrfs_dir_data_len(path->nodes[0], di); | |
2224 | u32 this_len = sizeof(*di) + name_len + data_len; | |
2225 | char *name; | |
2226 | ||
2227 | name = kmalloc(name_len, GFP_NOFS); | |
2228 | if (!name) { | |
2229 | ret = -ENOMEM; | |
2230 | goto out; | |
2231 | } | |
2232 | read_extent_buffer(path->nodes[0], name, | |
2233 | (unsigned long)(di + 1), name_len); | |
2234 | ||
2235 | log_di = btrfs_lookup_xattr(NULL, log, log_path, ino, | |
2236 | name, name_len, 0); | |
2237 | btrfs_release_path(log_path); | |
2238 | if (!log_di) { | |
2239 | /* Doesn't exist in log tree, so delete it. */ | |
2240 | btrfs_release_path(path); | |
2241 | di = btrfs_lookup_xattr(trans, root, path, ino, | |
2242 | name, name_len, -1); | |
2243 | kfree(name); | |
2244 | if (IS_ERR(di)) { | |
2245 | ret = PTR_ERR(di); | |
2246 | goto out; | |
2247 | } | |
2248 | ASSERT(di); | |
2249 | ret = btrfs_delete_one_dir_name(trans, root, | |
2250 | path, di); | |
2251 | if (ret) | |
2252 | goto out; | |
2253 | btrfs_release_path(path); | |
2254 | search_key = key; | |
2255 | goto again; | |
2256 | } | |
2257 | kfree(name); | |
2258 | if (IS_ERR(log_di)) { | |
2259 | ret = PTR_ERR(log_di); | |
2260 | goto out; | |
2261 | } | |
2262 | cur += this_len; | |
2263 | di = (struct btrfs_dir_item *)((char *)di + this_len); | |
2264 | } | |
2265 | } | |
2266 | ret = btrfs_next_leaf(root, path); | |
2267 | if (ret > 0) | |
2268 | ret = 0; | |
2269 | else if (ret == 0) | |
2270 | goto process_leaf; | |
2271 | out: | |
2272 | btrfs_free_path(log_path); | |
2273 | btrfs_release_path(path); | |
2274 | return ret; | |
2275 | } | |
2276 | ||
2277 | ||
e02119d5 CM |
2278 | /* |
2279 | * deletion replay happens before we copy any new directory items | |
2280 | * out of the log or out of backreferences from inodes. It | |
2281 | * scans the log to find ranges of keys that log is authoritative for, | |
2282 | * and then scans the directory to find items in those ranges that are | |
2283 | * not present in the log. | |
2284 | * | |
2285 | * Anything we don't find in the log is unlinked and removed from the | |
2286 | * directory. | |
2287 | */ | |
2288 | static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans, | |
2289 | struct btrfs_root *root, | |
2290 | struct btrfs_root *log, | |
2291 | struct btrfs_path *path, | |
12fcfd22 | 2292 | u64 dirid, int del_all) |
e02119d5 CM |
2293 | { |
2294 | u64 range_start; | |
2295 | u64 range_end; | |
2296 | int key_type = BTRFS_DIR_LOG_ITEM_KEY; | |
2297 | int ret = 0; | |
2298 | struct btrfs_key dir_key; | |
2299 | struct btrfs_key found_key; | |
2300 | struct btrfs_path *log_path; | |
2301 | struct inode *dir; | |
2302 | ||
2303 | dir_key.objectid = dirid; | |
2304 | dir_key.type = BTRFS_DIR_ITEM_KEY; | |
2305 | log_path = btrfs_alloc_path(); | |
2306 | if (!log_path) | |
2307 | return -ENOMEM; | |
2308 | ||
2309 | dir = read_one_inode(root, dirid); | |
2310 | /* it isn't an error if the inode isn't there, that can happen | |
2311 | * because we replay the deletes before we copy in the inode item | |
2312 | * from the log | |
2313 | */ | |
2314 | if (!dir) { | |
2315 | btrfs_free_path(log_path); | |
2316 | return 0; | |
2317 | } | |
2318 | again: | |
2319 | range_start = 0; | |
2320 | range_end = 0; | |
d397712b | 2321 | while (1) { |
12fcfd22 CM |
2322 | if (del_all) |
2323 | range_end = (u64)-1; | |
2324 | else { | |
2325 | ret = find_dir_range(log, path, dirid, key_type, | |
2326 | &range_start, &range_end); | |
2327 | if (ret != 0) | |
2328 | break; | |
2329 | } | |
e02119d5 CM |
2330 | |
2331 | dir_key.offset = range_start; | |
d397712b | 2332 | while (1) { |
e02119d5 CM |
2333 | int nritems; |
2334 | ret = btrfs_search_slot(NULL, root, &dir_key, path, | |
2335 | 0, 0); | |
2336 | if (ret < 0) | |
2337 | goto out; | |
2338 | ||
2339 | nritems = btrfs_header_nritems(path->nodes[0]); | |
2340 | if (path->slots[0] >= nritems) { | |
2341 | ret = btrfs_next_leaf(root, path); | |
b98def7c | 2342 | if (ret == 1) |
e02119d5 | 2343 | break; |
b98def7c LB |
2344 | else if (ret < 0) |
2345 | goto out; | |
e02119d5 CM |
2346 | } |
2347 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, | |
2348 | path->slots[0]); | |
2349 | if (found_key.objectid != dirid || | |
2350 | found_key.type != dir_key.type) | |
2351 | goto next_type; | |
2352 | ||
2353 | if (found_key.offset > range_end) | |
2354 | break; | |
2355 | ||
2356 | ret = check_item_in_log(trans, root, log, path, | |
12fcfd22 CM |
2357 | log_path, dir, |
2358 | &found_key); | |
3650860b JB |
2359 | if (ret) |
2360 | goto out; | |
e02119d5 CM |
2361 | if (found_key.offset == (u64)-1) |
2362 | break; | |
2363 | dir_key.offset = found_key.offset + 1; | |
2364 | } | |
b3b4aa74 | 2365 | btrfs_release_path(path); |
e02119d5 CM |
2366 | if (range_end == (u64)-1) |
2367 | break; | |
2368 | range_start = range_end + 1; | |
2369 | } | |
2370 | ||
2371 | next_type: | |
2372 | ret = 0; | |
2373 | if (key_type == BTRFS_DIR_LOG_ITEM_KEY) { | |
2374 | key_type = BTRFS_DIR_LOG_INDEX_KEY; | |
2375 | dir_key.type = BTRFS_DIR_INDEX_KEY; | |
b3b4aa74 | 2376 | btrfs_release_path(path); |
e02119d5 CM |
2377 | goto again; |
2378 | } | |
2379 | out: | |
b3b4aa74 | 2380 | btrfs_release_path(path); |
e02119d5 CM |
2381 | btrfs_free_path(log_path); |
2382 | iput(dir); | |
2383 | return ret; | |
2384 | } | |
2385 | ||
2386 | /* | |
2387 | * the process_func used to replay items from the log tree. This | |
2388 | * gets called in two different stages. The first stage just looks | |
2389 | * for inodes and makes sure they are all copied into the subvolume. | |
2390 | * | |
2391 | * The second stage copies all the other item types from the log into | |
2392 | * the subvolume. The two stage approach is slower, but gets rid of | |
2393 | * lots of complexity around inodes referencing other inodes that exist | |
2394 | * only in the log (references come from either directory items or inode | |
2395 | * back refs). | |
2396 | */ | |
2397 | static int replay_one_buffer(struct btrfs_root *log, struct extent_buffer *eb, | |
581c1760 | 2398 | struct walk_control *wc, u64 gen, int level) |
e02119d5 CM |
2399 | { |
2400 | int nritems; | |
2401 | struct btrfs_path *path; | |
2402 | struct btrfs_root *root = wc->replay_dest; | |
2403 | struct btrfs_key key; | |
e02119d5 CM |
2404 | int i; |
2405 | int ret; | |
2406 | ||
581c1760 | 2407 | ret = btrfs_read_buffer(eb, gen, level, NULL); |
018642a1 TI |
2408 | if (ret) |
2409 | return ret; | |
e02119d5 CM |
2410 | |
2411 | level = btrfs_header_level(eb); | |
2412 | ||
2413 | if (level != 0) | |
2414 | return 0; | |
2415 | ||
2416 | path = btrfs_alloc_path(); | |
1e5063d0 MF |
2417 | if (!path) |
2418 | return -ENOMEM; | |
e02119d5 CM |
2419 | |
2420 | nritems = btrfs_header_nritems(eb); | |
2421 | for (i = 0; i < nritems; i++) { | |
2422 | btrfs_item_key_to_cpu(eb, &key, i); | |
e02119d5 CM |
2423 | |
2424 | /* inode keys are done during the first stage */ | |
2425 | if (key.type == BTRFS_INODE_ITEM_KEY && | |
2426 | wc->stage == LOG_WALK_REPLAY_INODES) { | |
e02119d5 CM |
2427 | struct btrfs_inode_item *inode_item; |
2428 | u32 mode; | |
2429 | ||
2430 | inode_item = btrfs_item_ptr(eb, i, | |
2431 | struct btrfs_inode_item); | |
4f764e51 FM |
2432 | ret = replay_xattr_deletes(wc->trans, root, log, |
2433 | path, key.objectid); | |
2434 | if (ret) | |
2435 | break; | |
e02119d5 CM |
2436 | mode = btrfs_inode_mode(eb, inode_item); |
2437 | if (S_ISDIR(mode)) { | |
2438 | ret = replay_dir_deletes(wc->trans, | |
12fcfd22 | 2439 | root, log, path, key.objectid, 0); |
b50c6e25 JB |
2440 | if (ret) |
2441 | break; | |
e02119d5 CM |
2442 | } |
2443 | ret = overwrite_item(wc->trans, root, path, | |
2444 | eb, i, &key); | |
b50c6e25 JB |
2445 | if (ret) |
2446 | break; | |
e02119d5 | 2447 | |
471d557a FM |
2448 | /* |
2449 | * Before replaying extents, truncate the inode to its | |
2450 | * size. We need to do it now and not after log replay | |
2451 | * because before an fsync we can have prealloc extents | |
2452 | * added beyond the inode's i_size. If we did it after, | |
2453 | * through orphan cleanup for example, we would drop | |
2454 | * those prealloc extents just after replaying them. | |
e02119d5 CM |
2455 | */ |
2456 | if (S_ISREG(mode)) { | |
471d557a FM |
2457 | struct inode *inode; |
2458 | u64 from; | |
2459 | ||
2460 | inode = read_one_inode(root, key.objectid); | |
2461 | if (!inode) { | |
2462 | ret = -EIO; | |
2463 | break; | |
2464 | } | |
2465 | from = ALIGN(i_size_read(inode), | |
2466 | root->fs_info->sectorsize); | |
2467 | ret = btrfs_drop_extents(wc->trans, root, inode, | |
2468 | from, (u64)-1, 1); | |
2469 | /* | |
2470 | * If the nlink count is zero here, the iput | |
2471 | * will free the inode. We bump it to make | |
2472 | * sure it doesn't get freed until the link | |
2473 | * count fixup is done. | |
2474 | */ | |
2475 | if (!ret) { | |
2476 | if (inode->i_nlink == 0) | |
2477 | inc_nlink(inode); | |
2478 | /* Update link count and nbytes. */ | |
2479 | ret = btrfs_update_inode(wc->trans, | |
2480 | root, inode); | |
2481 | } | |
2482 | iput(inode); | |
b50c6e25 JB |
2483 | if (ret) |
2484 | break; | |
e02119d5 | 2485 | } |
c71bf099 | 2486 | |
e02119d5 CM |
2487 | ret = link_to_fixup_dir(wc->trans, root, |
2488 | path, key.objectid); | |
b50c6e25 JB |
2489 | if (ret) |
2490 | break; | |
e02119d5 | 2491 | } |
dd8e7217 JB |
2492 | |
2493 | if (key.type == BTRFS_DIR_INDEX_KEY && | |
2494 | wc->stage == LOG_WALK_REPLAY_DIR_INDEX) { | |
2495 | ret = replay_one_dir_item(wc->trans, root, path, | |
2496 | eb, i, &key); | |
2497 | if (ret) | |
2498 | break; | |
2499 | } | |
2500 | ||
e02119d5 CM |
2501 | if (wc->stage < LOG_WALK_REPLAY_ALL) |
2502 | continue; | |
2503 | ||
2504 | /* these keys are simply copied */ | |
2505 | if (key.type == BTRFS_XATTR_ITEM_KEY) { | |
2506 | ret = overwrite_item(wc->trans, root, path, | |
2507 | eb, i, &key); | |
b50c6e25 JB |
2508 | if (ret) |
2509 | break; | |
2da1c669 LB |
2510 | } else if (key.type == BTRFS_INODE_REF_KEY || |
2511 | key.type == BTRFS_INODE_EXTREF_KEY) { | |
f186373f MF |
2512 | ret = add_inode_ref(wc->trans, root, log, path, |
2513 | eb, i, &key); | |
b50c6e25 JB |
2514 | if (ret && ret != -ENOENT) |
2515 | break; | |
2516 | ret = 0; | |
e02119d5 CM |
2517 | } else if (key.type == BTRFS_EXTENT_DATA_KEY) { |
2518 | ret = replay_one_extent(wc->trans, root, path, | |
2519 | eb, i, &key); | |
b50c6e25 JB |
2520 | if (ret) |
2521 | break; | |
dd8e7217 | 2522 | } else if (key.type == BTRFS_DIR_ITEM_KEY) { |
e02119d5 CM |
2523 | ret = replay_one_dir_item(wc->trans, root, path, |
2524 | eb, i, &key); | |
b50c6e25 JB |
2525 | if (ret) |
2526 | break; | |
e02119d5 CM |
2527 | } |
2528 | } | |
2529 | btrfs_free_path(path); | |
b50c6e25 | 2530 | return ret; |
e02119d5 CM |
2531 | } |
2532 | ||
d397712b | 2533 | static noinline int walk_down_log_tree(struct btrfs_trans_handle *trans, |
e02119d5 CM |
2534 | struct btrfs_root *root, |
2535 | struct btrfs_path *path, int *level, | |
2536 | struct walk_control *wc) | |
2537 | { | |
0b246afa | 2538 | struct btrfs_fs_info *fs_info = root->fs_info; |
e02119d5 | 2539 | u64 root_owner; |
e02119d5 CM |
2540 | u64 bytenr; |
2541 | u64 ptr_gen; | |
2542 | struct extent_buffer *next; | |
2543 | struct extent_buffer *cur; | |
2544 | struct extent_buffer *parent; | |
2545 | u32 blocksize; | |
2546 | int ret = 0; | |
2547 | ||
2548 | WARN_ON(*level < 0); | |
2549 | WARN_ON(*level >= BTRFS_MAX_LEVEL); | |
2550 | ||
d397712b | 2551 | while (*level > 0) { |
581c1760 QW |
2552 | struct btrfs_key first_key; |
2553 | ||
e02119d5 CM |
2554 | WARN_ON(*level < 0); |
2555 | WARN_ON(*level >= BTRFS_MAX_LEVEL); | |
2556 | cur = path->nodes[*level]; | |
2557 | ||
fae7f21c | 2558 | WARN_ON(btrfs_header_level(cur) != *level); |
e02119d5 CM |
2559 | |
2560 | if (path->slots[*level] >= | |
2561 | btrfs_header_nritems(cur)) | |
2562 | break; | |
2563 | ||
2564 | bytenr = btrfs_node_blockptr(cur, path->slots[*level]); | |
2565 | ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]); | |
581c1760 | 2566 | btrfs_node_key_to_cpu(cur, &first_key, path->slots[*level]); |
0b246afa | 2567 | blocksize = fs_info->nodesize; |
e02119d5 CM |
2568 | |
2569 | parent = path->nodes[*level]; | |
2570 | root_owner = btrfs_header_owner(parent); | |
e02119d5 | 2571 | |
2ff7e61e | 2572 | next = btrfs_find_create_tree_block(fs_info, bytenr); |
c871b0f2 LB |
2573 | if (IS_ERR(next)) |
2574 | return PTR_ERR(next); | |
e02119d5 | 2575 | |
e02119d5 | 2576 | if (*level == 1) { |
581c1760 QW |
2577 | ret = wc->process_func(root, next, wc, ptr_gen, |
2578 | *level - 1); | |
b50c6e25 JB |
2579 | if (ret) { |
2580 | free_extent_buffer(next); | |
1e5063d0 | 2581 | return ret; |
b50c6e25 | 2582 | } |
4a500fd1 | 2583 | |
e02119d5 CM |
2584 | path->slots[*level]++; |
2585 | if (wc->free) { | |
581c1760 QW |
2586 | ret = btrfs_read_buffer(next, ptr_gen, |
2587 | *level - 1, &first_key); | |
018642a1 TI |
2588 | if (ret) { |
2589 | free_extent_buffer(next); | |
2590 | return ret; | |
2591 | } | |
e02119d5 | 2592 | |
681ae509 JB |
2593 | if (trans) { |
2594 | btrfs_tree_lock(next); | |
2595 | btrfs_set_lock_blocking(next); | |
7c302b49 | 2596 | clean_tree_block(fs_info, next); |
681ae509 JB |
2597 | btrfs_wait_tree_block_writeback(next); |
2598 | btrfs_tree_unlock(next); | |
1846430c LB |
2599 | } else { |
2600 | if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &next->bflags)) | |
2601 | clear_extent_buffer_dirty(next); | |
681ae509 | 2602 | } |
e02119d5 | 2603 | |
e02119d5 CM |
2604 | WARN_ON(root_owner != |
2605 | BTRFS_TREE_LOG_OBJECTID); | |
2ff7e61e JM |
2606 | ret = btrfs_free_and_pin_reserved_extent( |
2607 | fs_info, bytenr, | |
2608 | blocksize); | |
3650860b JB |
2609 | if (ret) { |
2610 | free_extent_buffer(next); | |
2611 | return ret; | |
2612 | } | |
e02119d5 CM |
2613 | } |
2614 | free_extent_buffer(next); | |
2615 | continue; | |
2616 | } | |
581c1760 | 2617 | ret = btrfs_read_buffer(next, ptr_gen, *level - 1, &first_key); |
018642a1 TI |
2618 | if (ret) { |
2619 | free_extent_buffer(next); | |
2620 | return ret; | |
2621 | } | |
e02119d5 CM |
2622 | |
2623 | WARN_ON(*level <= 0); | |
2624 | if (path->nodes[*level-1]) | |
2625 | free_extent_buffer(path->nodes[*level-1]); | |
2626 | path->nodes[*level-1] = next; | |
2627 | *level = btrfs_header_level(next); | |
2628 | path->slots[*level] = 0; | |
2629 | cond_resched(); | |
2630 | } | |
2631 | WARN_ON(*level < 0); | |
2632 | WARN_ON(*level >= BTRFS_MAX_LEVEL); | |
2633 | ||
4a500fd1 | 2634 | path->slots[*level] = btrfs_header_nritems(path->nodes[*level]); |
e02119d5 CM |
2635 | |
2636 | cond_resched(); | |
2637 | return 0; | |
2638 | } | |
2639 | ||
d397712b | 2640 | static noinline int walk_up_log_tree(struct btrfs_trans_handle *trans, |
e02119d5 CM |
2641 | struct btrfs_root *root, |
2642 | struct btrfs_path *path, int *level, | |
2643 | struct walk_control *wc) | |
2644 | { | |
0b246afa | 2645 | struct btrfs_fs_info *fs_info = root->fs_info; |
e02119d5 | 2646 | u64 root_owner; |
e02119d5 CM |
2647 | int i; |
2648 | int slot; | |
2649 | int ret; | |
2650 | ||
d397712b | 2651 | for (i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) { |
e02119d5 | 2652 | slot = path->slots[i]; |
4a500fd1 | 2653 | if (slot + 1 < btrfs_header_nritems(path->nodes[i])) { |
e02119d5 CM |
2654 | path->slots[i]++; |
2655 | *level = i; | |
2656 | WARN_ON(*level == 0); | |
2657 | return 0; | |
2658 | } else { | |
31840ae1 ZY |
2659 | struct extent_buffer *parent; |
2660 | if (path->nodes[*level] == root->node) | |
2661 | parent = path->nodes[*level]; | |
2662 | else | |
2663 | parent = path->nodes[*level + 1]; | |
2664 | ||
2665 | root_owner = btrfs_header_owner(parent); | |
1e5063d0 | 2666 | ret = wc->process_func(root, path->nodes[*level], wc, |
581c1760 QW |
2667 | btrfs_header_generation(path->nodes[*level]), |
2668 | *level); | |
1e5063d0 MF |
2669 | if (ret) |
2670 | return ret; | |
2671 | ||
e02119d5 CM |
2672 | if (wc->free) { |
2673 | struct extent_buffer *next; | |
2674 | ||
2675 | next = path->nodes[*level]; | |
2676 | ||
681ae509 JB |
2677 | if (trans) { |
2678 | btrfs_tree_lock(next); | |
2679 | btrfs_set_lock_blocking(next); | |
7c302b49 | 2680 | clean_tree_block(fs_info, next); |
681ae509 JB |
2681 | btrfs_wait_tree_block_writeback(next); |
2682 | btrfs_tree_unlock(next); | |
1846430c LB |
2683 | } else { |
2684 | if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &next->bflags)) | |
2685 | clear_extent_buffer_dirty(next); | |
681ae509 | 2686 | } |
e02119d5 | 2687 | |
e02119d5 | 2688 | WARN_ON(root_owner != BTRFS_TREE_LOG_OBJECTID); |
2ff7e61e JM |
2689 | ret = btrfs_free_and_pin_reserved_extent( |
2690 | fs_info, | |
e02119d5 | 2691 | path->nodes[*level]->start, |
d00aff00 | 2692 | path->nodes[*level]->len); |
3650860b JB |
2693 | if (ret) |
2694 | return ret; | |
e02119d5 CM |
2695 | } |
2696 | free_extent_buffer(path->nodes[*level]); | |
2697 | path->nodes[*level] = NULL; | |
2698 | *level = i + 1; | |
2699 | } | |
2700 | } | |
2701 | return 1; | |
2702 | } | |
2703 | ||
2704 | /* | |
2705 | * drop the reference count on the tree rooted at 'snap'. This traverses | |
2706 | * the tree freeing any blocks that have a ref count of zero after being | |
2707 | * decremented. | |
2708 | */ | |
2709 | static int walk_log_tree(struct btrfs_trans_handle *trans, | |
2710 | struct btrfs_root *log, struct walk_control *wc) | |
2711 | { | |
2ff7e61e | 2712 | struct btrfs_fs_info *fs_info = log->fs_info; |
e02119d5 CM |
2713 | int ret = 0; |
2714 | int wret; | |
2715 | int level; | |
2716 | struct btrfs_path *path; | |
e02119d5 CM |
2717 | int orig_level; |
2718 | ||
2719 | path = btrfs_alloc_path(); | |
db5b493a TI |
2720 | if (!path) |
2721 | return -ENOMEM; | |
e02119d5 CM |
2722 | |
2723 | level = btrfs_header_level(log->node); | |
2724 | orig_level = level; | |
2725 | path->nodes[level] = log->node; | |
2726 | extent_buffer_get(log->node); | |
2727 | path->slots[level] = 0; | |
2728 | ||
d397712b | 2729 | while (1) { |
e02119d5 CM |
2730 | wret = walk_down_log_tree(trans, log, path, &level, wc); |
2731 | if (wret > 0) | |
2732 | break; | |
79787eaa | 2733 | if (wret < 0) { |
e02119d5 | 2734 | ret = wret; |
79787eaa JM |
2735 | goto out; |
2736 | } | |
e02119d5 CM |
2737 | |
2738 | wret = walk_up_log_tree(trans, log, path, &level, wc); | |
2739 | if (wret > 0) | |
2740 | break; | |
79787eaa | 2741 | if (wret < 0) { |
e02119d5 | 2742 | ret = wret; |
79787eaa JM |
2743 | goto out; |
2744 | } | |
e02119d5 CM |
2745 | } |
2746 | ||
2747 | /* was the root node processed? if not, catch it here */ | |
2748 | if (path->nodes[orig_level]) { | |
79787eaa | 2749 | ret = wc->process_func(log, path->nodes[orig_level], wc, |
581c1760 QW |
2750 | btrfs_header_generation(path->nodes[orig_level]), |
2751 | orig_level); | |
79787eaa JM |
2752 | if (ret) |
2753 | goto out; | |
e02119d5 CM |
2754 | if (wc->free) { |
2755 | struct extent_buffer *next; | |
2756 | ||
2757 | next = path->nodes[orig_level]; | |
2758 | ||
681ae509 JB |
2759 | if (trans) { |
2760 | btrfs_tree_lock(next); | |
2761 | btrfs_set_lock_blocking(next); | |
7c302b49 | 2762 | clean_tree_block(fs_info, next); |
681ae509 JB |
2763 | btrfs_wait_tree_block_writeback(next); |
2764 | btrfs_tree_unlock(next); | |
1846430c LB |
2765 | } else { |
2766 | if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &next->bflags)) | |
2767 | clear_extent_buffer_dirty(next); | |
681ae509 | 2768 | } |
e02119d5 | 2769 | |
e02119d5 CM |
2770 | WARN_ON(log->root_key.objectid != |
2771 | BTRFS_TREE_LOG_OBJECTID); | |
2ff7e61e JM |
2772 | ret = btrfs_free_and_pin_reserved_extent(fs_info, |
2773 | next->start, next->len); | |
3650860b JB |
2774 | if (ret) |
2775 | goto out; | |
e02119d5 CM |
2776 | } |
2777 | } | |
2778 | ||
79787eaa | 2779 | out: |
e02119d5 | 2780 | btrfs_free_path(path); |
e02119d5 CM |
2781 | return ret; |
2782 | } | |
2783 | ||
7237f183 YZ |
2784 | /* |
2785 | * helper function to update the item for a given subvolumes log root | |
2786 | * in the tree of log roots | |
2787 | */ | |
2788 | static int update_log_root(struct btrfs_trans_handle *trans, | |
2789 | struct btrfs_root *log) | |
2790 | { | |
0b246afa | 2791 | struct btrfs_fs_info *fs_info = log->fs_info; |
7237f183 YZ |
2792 | int ret; |
2793 | ||
2794 | if (log->log_transid == 1) { | |
2795 | /* insert root item on the first sync */ | |
0b246afa | 2796 | ret = btrfs_insert_root(trans, fs_info->log_root_tree, |
7237f183 YZ |
2797 | &log->root_key, &log->root_item); |
2798 | } else { | |
0b246afa | 2799 | ret = btrfs_update_root(trans, fs_info->log_root_tree, |
7237f183 YZ |
2800 | &log->root_key, &log->root_item); |
2801 | } | |
2802 | return ret; | |
2803 | } | |
2804 | ||
60d53eb3 | 2805 | static void wait_log_commit(struct btrfs_root *root, int transid) |
e02119d5 CM |
2806 | { |
2807 | DEFINE_WAIT(wait); | |
7237f183 | 2808 | int index = transid % 2; |
e02119d5 | 2809 | |
7237f183 YZ |
2810 | /* |
2811 | * we only allow two pending log transactions at a time, | |
2812 | * so we know that if ours is more than 2 older than the | |
2813 | * current transaction, we're done | |
2814 | */ | |
49e83f57 | 2815 | for (;;) { |
7237f183 YZ |
2816 | prepare_to_wait(&root->log_commit_wait[index], |
2817 | &wait, TASK_UNINTERRUPTIBLE); | |
12fcfd22 | 2818 | |
49e83f57 LB |
2819 | if (!(root->log_transid_committed < transid && |
2820 | atomic_read(&root->log_commit[index]))) | |
2821 | break; | |
12fcfd22 | 2822 | |
49e83f57 LB |
2823 | mutex_unlock(&root->log_mutex); |
2824 | schedule(); | |
7237f183 | 2825 | mutex_lock(&root->log_mutex); |
49e83f57 LB |
2826 | } |
2827 | finish_wait(&root->log_commit_wait[index], &wait); | |
7237f183 YZ |
2828 | } |
2829 | ||
60d53eb3 | 2830 | static void wait_for_writer(struct btrfs_root *root) |
7237f183 YZ |
2831 | { |
2832 | DEFINE_WAIT(wait); | |
8b050d35 | 2833 | |
49e83f57 LB |
2834 | for (;;) { |
2835 | prepare_to_wait(&root->log_writer_wait, &wait, | |
2836 | TASK_UNINTERRUPTIBLE); | |
2837 | if (!atomic_read(&root->log_writers)) | |
2838 | break; | |
2839 | ||
7237f183 | 2840 | mutex_unlock(&root->log_mutex); |
49e83f57 | 2841 | schedule(); |
575849ec | 2842 | mutex_lock(&root->log_mutex); |
7237f183 | 2843 | } |
49e83f57 | 2844 | finish_wait(&root->log_writer_wait, &wait); |
e02119d5 CM |
2845 | } |
2846 | ||
8b050d35 MX |
2847 | static inline void btrfs_remove_log_ctx(struct btrfs_root *root, |
2848 | struct btrfs_log_ctx *ctx) | |
2849 | { | |
2850 | if (!ctx) | |
2851 | return; | |
2852 | ||
2853 | mutex_lock(&root->log_mutex); | |
2854 | list_del_init(&ctx->list); | |
2855 | mutex_unlock(&root->log_mutex); | |
2856 | } | |
2857 | ||
2858 | /* | |
2859 | * Invoked in log mutex context, or be sure there is no other task which | |
2860 | * can access the list. | |
2861 | */ | |
2862 | static inline void btrfs_remove_all_log_ctxs(struct btrfs_root *root, | |
2863 | int index, int error) | |
2864 | { | |
2865 | struct btrfs_log_ctx *ctx; | |
570dd450 | 2866 | struct btrfs_log_ctx *safe; |
8b050d35 | 2867 | |
570dd450 CM |
2868 | list_for_each_entry_safe(ctx, safe, &root->log_ctxs[index], list) { |
2869 | list_del_init(&ctx->list); | |
8b050d35 | 2870 | ctx->log_ret = error; |
570dd450 | 2871 | } |
8b050d35 MX |
2872 | |
2873 | INIT_LIST_HEAD(&root->log_ctxs[index]); | |
2874 | } | |
2875 | ||
e02119d5 CM |
2876 | /* |
2877 | * btrfs_sync_log does sends a given tree log down to the disk and | |
2878 | * updates the super blocks to record it. When this call is done, | |
12fcfd22 CM |
2879 | * you know that any inodes previously logged are safely on disk only |
2880 | * if it returns 0. | |
2881 | * | |
2882 | * Any other return value means you need to call btrfs_commit_transaction. | |
2883 | * Some of the edge cases for fsyncing directories that have had unlinks | |
2884 | * or renames done in the past mean that sometimes the only safe | |
2885 | * fsync is to commit the whole FS. When btrfs_sync_log returns -EAGAIN, | |
2886 | * that has happened. | |
e02119d5 CM |
2887 | */ |
2888 | int btrfs_sync_log(struct btrfs_trans_handle *trans, | |
8b050d35 | 2889 | struct btrfs_root *root, struct btrfs_log_ctx *ctx) |
e02119d5 | 2890 | { |
7237f183 YZ |
2891 | int index1; |
2892 | int index2; | |
8cef4e16 | 2893 | int mark; |
e02119d5 | 2894 | int ret; |
0b246afa | 2895 | struct btrfs_fs_info *fs_info = root->fs_info; |
e02119d5 | 2896 | struct btrfs_root *log = root->log_root; |
0b246afa | 2897 | struct btrfs_root *log_root_tree = fs_info->log_root_tree; |
bb14a59b | 2898 | int log_transid = 0; |
8b050d35 | 2899 | struct btrfs_log_ctx root_log_ctx; |
c6adc9cc | 2900 | struct blk_plug plug; |
e02119d5 | 2901 | |
7237f183 | 2902 | mutex_lock(&root->log_mutex); |
d1433deb MX |
2903 | log_transid = ctx->log_transid; |
2904 | if (root->log_transid_committed >= log_transid) { | |
2905 | mutex_unlock(&root->log_mutex); | |
2906 | return ctx->log_ret; | |
2907 | } | |
2908 | ||
2909 | index1 = log_transid % 2; | |
7237f183 | 2910 | if (atomic_read(&root->log_commit[index1])) { |
60d53eb3 | 2911 | wait_log_commit(root, log_transid); |
7237f183 | 2912 | mutex_unlock(&root->log_mutex); |
8b050d35 | 2913 | return ctx->log_ret; |
e02119d5 | 2914 | } |
d1433deb | 2915 | ASSERT(log_transid == root->log_transid); |
7237f183 YZ |
2916 | atomic_set(&root->log_commit[index1], 1); |
2917 | ||
2918 | /* wait for previous tree log sync to complete */ | |
2919 | if (atomic_read(&root->log_commit[(index1 + 1) % 2])) | |
60d53eb3 | 2920 | wait_log_commit(root, log_transid - 1); |
48cab2e0 | 2921 | |
86df7eb9 | 2922 | while (1) { |
2ecb7923 | 2923 | int batch = atomic_read(&root->log_batch); |
cd354ad6 | 2924 | /* when we're on an ssd, just kick the log commit out */ |
0b246afa | 2925 | if (!btrfs_test_opt(fs_info, SSD) && |
27cdeb70 | 2926 | test_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state)) { |
86df7eb9 YZ |
2927 | mutex_unlock(&root->log_mutex); |
2928 | schedule_timeout_uninterruptible(1); | |
2929 | mutex_lock(&root->log_mutex); | |
2930 | } | |
60d53eb3 | 2931 | wait_for_writer(root); |
2ecb7923 | 2932 | if (batch == atomic_read(&root->log_batch)) |
e02119d5 CM |
2933 | break; |
2934 | } | |
e02119d5 | 2935 | |
12fcfd22 | 2936 | /* bail out if we need to do a full commit */ |
0b246afa | 2937 | if (btrfs_need_log_full_commit(fs_info, trans)) { |
12fcfd22 | 2938 | ret = -EAGAIN; |
2ab28f32 | 2939 | btrfs_free_logged_extents(log, log_transid); |
12fcfd22 CM |
2940 | mutex_unlock(&root->log_mutex); |
2941 | goto out; | |
2942 | } | |
2943 | ||
8cef4e16 YZ |
2944 | if (log_transid % 2 == 0) |
2945 | mark = EXTENT_DIRTY; | |
2946 | else | |
2947 | mark = EXTENT_NEW; | |
2948 | ||
690587d1 CM |
2949 | /* we start IO on all the marked extents here, but we don't actually |
2950 | * wait for them until later. | |
2951 | */ | |
c6adc9cc | 2952 | blk_start_plug(&plug); |
2ff7e61e | 2953 | ret = btrfs_write_marked_extents(fs_info, &log->dirty_log_pages, mark); |
79787eaa | 2954 | if (ret) { |
c6adc9cc | 2955 | blk_finish_plug(&plug); |
66642832 | 2956 | btrfs_abort_transaction(trans, ret); |
2ab28f32 | 2957 | btrfs_free_logged_extents(log, log_transid); |
0b246afa | 2958 | btrfs_set_log_full_commit(fs_info, trans); |
79787eaa JM |
2959 | mutex_unlock(&root->log_mutex); |
2960 | goto out; | |
2961 | } | |
7237f183 | 2962 | |
5d4f98a2 | 2963 | btrfs_set_root_node(&log->root_item, log->node); |
7237f183 | 2964 | |
7237f183 YZ |
2965 | root->log_transid++; |
2966 | log->log_transid = root->log_transid; | |
ff782e0a | 2967 | root->log_start_pid = 0; |
7237f183 | 2968 | /* |
8cef4e16 YZ |
2969 | * IO has been started, blocks of the log tree have WRITTEN flag set |
2970 | * in their headers. new modifications of the log will be written to | |
2971 | * new positions. so it's safe to allow log writers to go in. | |
7237f183 YZ |
2972 | */ |
2973 | mutex_unlock(&root->log_mutex); | |
2974 | ||
28a23593 | 2975 | btrfs_init_log_ctx(&root_log_ctx, NULL); |
d1433deb | 2976 | |
7237f183 | 2977 | mutex_lock(&log_root_tree->log_mutex); |
2ecb7923 | 2978 | atomic_inc(&log_root_tree->log_batch); |
7237f183 | 2979 | atomic_inc(&log_root_tree->log_writers); |
d1433deb MX |
2980 | |
2981 | index2 = log_root_tree->log_transid % 2; | |
2982 | list_add_tail(&root_log_ctx.list, &log_root_tree->log_ctxs[index2]); | |
2983 | root_log_ctx.log_transid = log_root_tree->log_transid; | |
2984 | ||
7237f183 YZ |
2985 | mutex_unlock(&log_root_tree->log_mutex); |
2986 | ||
2987 | ret = update_log_root(trans, log); | |
7237f183 YZ |
2988 | |
2989 | mutex_lock(&log_root_tree->log_mutex); | |
2990 | if (atomic_dec_and_test(&log_root_tree->log_writers)) { | |
779adf0f DS |
2991 | /* |
2992 | * Implicit memory barrier after atomic_dec_and_test | |
2993 | */ | |
7237f183 YZ |
2994 | if (waitqueue_active(&log_root_tree->log_writer_wait)) |
2995 | wake_up(&log_root_tree->log_writer_wait); | |
2996 | } | |
2997 | ||
4a500fd1 | 2998 | if (ret) { |
d1433deb MX |
2999 | if (!list_empty(&root_log_ctx.list)) |
3000 | list_del_init(&root_log_ctx.list); | |
3001 | ||
c6adc9cc | 3002 | blk_finish_plug(&plug); |
0b246afa | 3003 | btrfs_set_log_full_commit(fs_info, trans); |
995946dd | 3004 | |
79787eaa | 3005 | if (ret != -ENOSPC) { |
66642832 | 3006 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
3007 | mutex_unlock(&log_root_tree->log_mutex); |
3008 | goto out; | |
3009 | } | |
bf89d38f | 3010 | btrfs_wait_tree_log_extents(log, mark); |
2ab28f32 | 3011 | btrfs_free_logged_extents(log, log_transid); |
4a500fd1 YZ |
3012 | mutex_unlock(&log_root_tree->log_mutex); |
3013 | ret = -EAGAIN; | |
3014 | goto out; | |
3015 | } | |
3016 | ||
d1433deb | 3017 | if (log_root_tree->log_transid_committed >= root_log_ctx.log_transid) { |
3da5ab56 | 3018 | blk_finish_plug(&plug); |
cbd60aa7 | 3019 | list_del_init(&root_log_ctx.list); |
d1433deb MX |
3020 | mutex_unlock(&log_root_tree->log_mutex); |
3021 | ret = root_log_ctx.log_ret; | |
3022 | goto out; | |
3023 | } | |
8b050d35 | 3024 | |
d1433deb | 3025 | index2 = root_log_ctx.log_transid % 2; |
7237f183 | 3026 | if (atomic_read(&log_root_tree->log_commit[index2])) { |
c6adc9cc | 3027 | blk_finish_plug(&plug); |
bf89d38f | 3028 | ret = btrfs_wait_tree_log_extents(log, mark); |
50d9aa99 | 3029 | btrfs_wait_logged_extents(trans, log, log_transid); |
60d53eb3 | 3030 | wait_log_commit(log_root_tree, |
d1433deb | 3031 | root_log_ctx.log_transid); |
7237f183 | 3032 | mutex_unlock(&log_root_tree->log_mutex); |
5ab5e44a FM |
3033 | if (!ret) |
3034 | ret = root_log_ctx.log_ret; | |
7237f183 YZ |
3035 | goto out; |
3036 | } | |
d1433deb | 3037 | ASSERT(root_log_ctx.log_transid == log_root_tree->log_transid); |
7237f183 YZ |
3038 | atomic_set(&log_root_tree->log_commit[index2], 1); |
3039 | ||
12fcfd22 | 3040 | if (atomic_read(&log_root_tree->log_commit[(index2 + 1) % 2])) { |
60d53eb3 | 3041 | wait_log_commit(log_root_tree, |
d1433deb | 3042 | root_log_ctx.log_transid - 1); |
12fcfd22 CM |
3043 | } |
3044 | ||
60d53eb3 | 3045 | wait_for_writer(log_root_tree); |
7237f183 | 3046 | |
12fcfd22 CM |
3047 | /* |
3048 | * now that we've moved on to the tree of log tree roots, | |
3049 | * check the full commit flag again | |
3050 | */ | |
0b246afa | 3051 | if (btrfs_need_log_full_commit(fs_info, trans)) { |
c6adc9cc | 3052 | blk_finish_plug(&plug); |
bf89d38f | 3053 | btrfs_wait_tree_log_extents(log, mark); |
2ab28f32 | 3054 | btrfs_free_logged_extents(log, log_transid); |
12fcfd22 CM |
3055 | mutex_unlock(&log_root_tree->log_mutex); |
3056 | ret = -EAGAIN; | |
3057 | goto out_wake_log_root; | |
3058 | } | |
7237f183 | 3059 | |
2ff7e61e | 3060 | ret = btrfs_write_marked_extents(fs_info, |
c6adc9cc MX |
3061 | &log_root_tree->dirty_log_pages, |
3062 | EXTENT_DIRTY | EXTENT_NEW); | |
3063 | blk_finish_plug(&plug); | |
79787eaa | 3064 | if (ret) { |
0b246afa | 3065 | btrfs_set_log_full_commit(fs_info, trans); |
66642832 | 3066 | btrfs_abort_transaction(trans, ret); |
2ab28f32 | 3067 | btrfs_free_logged_extents(log, log_transid); |
79787eaa JM |
3068 | mutex_unlock(&log_root_tree->log_mutex); |
3069 | goto out_wake_log_root; | |
3070 | } | |
bf89d38f | 3071 | ret = btrfs_wait_tree_log_extents(log, mark); |
5ab5e44a | 3072 | if (!ret) |
bf89d38f JM |
3073 | ret = btrfs_wait_tree_log_extents(log_root_tree, |
3074 | EXTENT_NEW | EXTENT_DIRTY); | |
5ab5e44a | 3075 | if (ret) { |
0b246afa | 3076 | btrfs_set_log_full_commit(fs_info, trans); |
5ab5e44a FM |
3077 | btrfs_free_logged_extents(log, log_transid); |
3078 | mutex_unlock(&log_root_tree->log_mutex); | |
3079 | goto out_wake_log_root; | |
3080 | } | |
50d9aa99 | 3081 | btrfs_wait_logged_extents(trans, log, log_transid); |
e02119d5 | 3082 | |
0b246afa JM |
3083 | btrfs_set_super_log_root(fs_info->super_for_commit, |
3084 | log_root_tree->node->start); | |
3085 | btrfs_set_super_log_root_level(fs_info->super_for_commit, | |
3086 | btrfs_header_level(log_root_tree->node)); | |
e02119d5 | 3087 | |
7237f183 | 3088 | log_root_tree->log_transid++; |
7237f183 YZ |
3089 | mutex_unlock(&log_root_tree->log_mutex); |
3090 | ||
3091 | /* | |
3092 | * nobody else is going to jump in and write the the ctree | |
3093 | * super here because the log_commit atomic below is protecting | |
3094 | * us. We must be called with a transaction handle pinning | |
3095 | * the running transaction open, so a full commit can't hop | |
3096 | * in and cause problems either. | |
3097 | */ | |
eece6a9c | 3098 | ret = write_all_supers(fs_info, 1); |
5af3e8cc | 3099 | if (ret) { |
0b246afa | 3100 | btrfs_set_log_full_commit(fs_info, trans); |
66642832 | 3101 | btrfs_abort_transaction(trans, ret); |
5af3e8cc SB |
3102 | goto out_wake_log_root; |
3103 | } | |
7237f183 | 3104 | |
257c62e1 CM |
3105 | mutex_lock(&root->log_mutex); |
3106 | if (root->last_log_commit < log_transid) | |
3107 | root->last_log_commit = log_transid; | |
3108 | mutex_unlock(&root->log_mutex); | |
3109 | ||
12fcfd22 | 3110 | out_wake_log_root: |
570dd450 | 3111 | mutex_lock(&log_root_tree->log_mutex); |
8b050d35 MX |
3112 | btrfs_remove_all_log_ctxs(log_root_tree, index2, ret); |
3113 | ||
d1433deb | 3114 | log_root_tree->log_transid_committed++; |
7237f183 | 3115 | atomic_set(&log_root_tree->log_commit[index2], 0); |
d1433deb MX |
3116 | mutex_unlock(&log_root_tree->log_mutex); |
3117 | ||
33a9eca7 DS |
3118 | /* |
3119 | * The barrier before waitqueue_active is implied by mutex_unlock | |
3120 | */ | |
7237f183 YZ |
3121 | if (waitqueue_active(&log_root_tree->log_commit_wait[index2])) |
3122 | wake_up(&log_root_tree->log_commit_wait[index2]); | |
e02119d5 | 3123 | out: |
d1433deb | 3124 | mutex_lock(&root->log_mutex); |
570dd450 | 3125 | btrfs_remove_all_log_ctxs(root, index1, ret); |
d1433deb | 3126 | root->log_transid_committed++; |
7237f183 | 3127 | atomic_set(&root->log_commit[index1], 0); |
d1433deb | 3128 | mutex_unlock(&root->log_mutex); |
8b050d35 | 3129 | |
33a9eca7 DS |
3130 | /* |
3131 | * The barrier before waitqueue_active is implied by mutex_unlock | |
3132 | */ | |
7237f183 YZ |
3133 | if (waitqueue_active(&root->log_commit_wait[index1])) |
3134 | wake_up(&root->log_commit_wait[index1]); | |
b31eabd8 | 3135 | return ret; |
e02119d5 CM |
3136 | } |
3137 | ||
4a500fd1 YZ |
3138 | static void free_log_tree(struct btrfs_trans_handle *trans, |
3139 | struct btrfs_root *log) | |
e02119d5 CM |
3140 | { |
3141 | int ret; | |
d0c803c4 CM |
3142 | u64 start; |
3143 | u64 end; | |
e02119d5 CM |
3144 | struct walk_control wc = { |
3145 | .free = 1, | |
3146 | .process_func = process_one_buffer | |
3147 | }; | |
3148 | ||
681ae509 JB |
3149 | ret = walk_log_tree(trans, log, &wc); |
3150 | /* I don't think this can happen but just in case */ | |
3151 | if (ret) | |
66642832 | 3152 | btrfs_abort_transaction(trans, ret); |
e02119d5 | 3153 | |
d397712b | 3154 | while (1) { |
d0c803c4 | 3155 | ret = find_first_extent_bit(&log->dirty_log_pages, |
55237a5f LB |
3156 | 0, &start, &end, |
3157 | EXTENT_DIRTY | EXTENT_NEW | EXTENT_NEED_WAIT, | |
e6138876 | 3158 | NULL); |
d0c803c4 CM |
3159 | if (ret) |
3160 | break; | |
3161 | ||
8cef4e16 | 3162 | clear_extent_bits(&log->dirty_log_pages, start, end, |
55237a5f | 3163 | EXTENT_DIRTY | EXTENT_NEW | EXTENT_NEED_WAIT); |
d0c803c4 CM |
3164 | } |
3165 | ||
2ab28f32 JB |
3166 | /* |
3167 | * We may have short-circuited the log tree with the full commit logic | |
3168 | * and left ordered extents on our list, so clear these out to keep us | |
3169 | * from leaking inodes and memory. | |
3170 | */ | |
3171 | btrfs_free_logged_extents(log, 0); | |
3172 | btrfs_free_logged_extents(log, 1); | |
3173 | ||
7237f183 YZ |
3174 | free_extent_buffer(log->node); |
3175 | kfree(log); | |
4a500fd1 YZ |
3176 | } |
3177 | ||
3178 | /* | |
3179 | * free all the extents used by the tree log. This should be called | |
3180 | * at commit time of the full transaction | |
3181 | */ | |
3182 | int btrfs_free_log(struct btrfs_trans_handle *trans, struct btrfs_root *root) | |
3183 | { | |
3184 | if (root->log_root) { | |
3185 | free_log_tree(trans, root->log_root); | |
3186 | root->log_root = NULL; | |
3187 | } | |
3188 | return 0; | |
3189 | } | |
3190 | ||
3191 | int btrfs_free_log_root_tree(struct btrfs_trans_handle *trans, | |
3192 | struct btrfs_fs_info *fs_info) | |
3193 | { | |
3194 | if (fs_info->log_root_tree) { | |
3195 | free_log_tree(trans, fs_info->log_root_tree); | |
3196 | fs_info->log_root_tree = NULL; | |
3197 | } | |
e02119d5 CM |
3198 | return 0; |
3199 | } | |
3200 | ||
e02119d5 CM |
3201 | /* |
3202 | * If both a file and directory are logged, and unlinks or renames are | |
3203 | * mixed in, we have a few interesting corners: | |
3204 | * | |
3205 | * create file X in dir Y | |
3206 | * link file X to X.link in dir Y | |
3207 | * fsync file X | |
3208 | * unlink file X but leave X.link | |
3209 | * fsync dir Y | |
3210 | * | |
3211 | * After a crash we would expect only X.link to exist. But file X | |
3212 | * didn't get fsync'd again so the log has back refs for X and X.link. | |
3213 | * | |
3214 | * We solve this by removing directory entries and inode backrefs from the | |
3215 | * log when a file that was logged in the current transaction is | |
3216 | * unlinked. Any later fsync will include the updated log entries, and | |
3217 | * we'll be able to reconstruct the proper directory items from backrefs. | |
3218 | * | |
3219 | * This optimizations allows us to avoid relogging the entire inode | |
3220 | * or the entire directory. | |
3221 | */ | |
3222 | int btrfs_del_dir_entries_in_log(struct btrfs_trans_handle *trans, | |
3223 | struct btrfs_root *root, | |
3224 | const char *name, int name_len, | |
49f34d1f | 3225 | struct btrfs_inode *dir, u64 index) |
e02119d5 CM |
3226 | { |
3227 | struct btrfs_root *log; | |
3228 | struct btrfs_dir_item *di; | |
3229 | struct btrfs_path *path; | |
3230 | int ret; | |
4a500fd1 | 3231 | int err = 0; |
e02119d5 | 3232 | int bytes_del = 0; |
49f34d1f | 3233 | u64 dir_ino = btrfs_ino(dir); |
e02119d5 | 3234 | |
49f34d1f | 3235 | if (dir->logged_trans < trans->transid) |
3a5f1d45 CM |
3236 | return 0; |
3237 | ||
e02119d5 CM |
3238 | ret = join_running_log_trans(root); |
3239 | if (ret) | |
3240 | return 0; | |
3241 | ||
49f34d1f | 3242 | mutex_lock(&dir->log_mutex); |
e02119d5 CM |
3243 | |
3244 | log = root->log_root; | |
3245 | path = btrfs_alloc_path(); | |
a62f44a5 TI |
3246 | if (!path) { |
3247 | err = -ENOMEM; | |
3248 | goto out_unlock; | |
3249 | } | |
2a29edc6 | 3250 | |
33345d01 | 3251 | di = btrfs_lookup_dir_item(trans, log, path, dir_ino, |
e02119d5 | 3252 | name, name_len, -1); |
4a500fd1 YZ |
3253 | if (IS_ERR(di)) { |
3254 | err = PTR_ERR(di); | |
3255 | goto fail; | |
3256 | } | |
3257 | if (di) { | |
e02119d5 CM |
3258 | ret = btrfs_delete_one_dir_name(trans, log, path, di); |
3259 | bytes_del += name_len; | |
3650860b JB |
3260 | if (ret) { |
3261 | err = ret; | |
3262 | goto fail; | |
3263 | } | |
e02119d5 | 3264 | } |
b3b4aa74 | 3265 | btrfs_release_path(path); |
33345d01 | 3266 | di = btrfs_lookup_dir_index_item(trans, log, path, dir_ino, |
e02119d5 | 3267 | index, name, name_len, -1); |
4a500fd1 YZ |
3268 | if (IS_ERR(di)) { |
3269 | err = PTR_ERR(di); | |
3270 | goto fail; | |
3271 | } | |
3272 | if (di) { | |
e02119d5 CM |
3273 | ret = btrfs_delete_one_dir_name(trans, log, path, di); |
3274 | bytes_del += name_len; | |
3650860b JB |
3275 | if (ret) { |
3276 | err = ret; | |
3277 | goto fail; | |
3278 | } | |
e02119d5 CM |
3279 | } |
3280 | ||
3281 | /* update the directory size in the log to reflect the names | |
3282 | * we have removed | |
3283 | */ | |
3284 | if (bytes_del) { | |
3285 | struct btrfs_key key; | |
3286 | ||
33345d01 | 3287 | key.objectid = dir_ino; |
e02119d5 CM |
3288 | key.offset = 0; |
3289 | key.type = BTRFS_INODE_ITEM_KEY; | |
b3b4aa74 | 3290 | btrfs_release_path(path); |
e02119d5 CM |
3291 | |
3292 | ret = btrfs_search_slot(trans, log, &key, path, 0, 1); | |
4a500fd1 YZ |
3293 | if (ret < 0) { |
3294 | err = ret; | |
3295 | goto fail; | |
3296 | } | |
e02119d5 CM |
3297 | if (ret == 0) { |
3298 | struct btrfs_inode_item *item; | |
3299 | u64 i_size; | |
3300 | ||
3301 | item = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
3302 | struct btrfs_inode_item); | |
3303 | i_size = btrfs_inode_size(path->nodes[0], item); | |
3304 | if (i_size > bytes_del) | |
3305 | i_size -= bytes_del; | |
3306 | else | |
3307 | i_size = 0; | |
3308 | btrfs_set_inode_size(path->nodes[0], item, i_size); | |
3309 | btrfs_mark_buffer_dirty(path->nodes[0]); | |
3310 | } else | |
3311 | ret = 0; | |
b3b4aa74 | 3312 | btrfs_release_path(path); |
e02119d5 | 3313 | } |
4a500fd1 | 3314 | fail: |
e02119d5 | 3315 | btrfs_free_path(path); |
a62f44a5 | 3316 | out_unlock: |
49f34d1f | 3317 | mutex_unlock(&dir->log_mutex); |
4a500fd1 | 3318 | if (ret == -ENOSPC) { |
995946dd | 3319 | btrfs_set_log_full_commit(root->fs_info, trans); |
4a500fd1 | 3320 | ret = 0; |
79787eaa | 3321 | } else if (ret < 0) |
66642832 | 3322 | btrfs_abort_transaction(trans, ret); |
79787eaa | 3323 | |
12fcfd22 | 3324 | btrfs_end_log_trans(root); |
e02119d5 | 3325 | |
411fc6bc | 3326 | return err; |
e02119d5 CM |
3327 | } |
3328 | ||
3329 | /* see comments for btrfs_del_dir_entries_in_log */ | |
3330 | int btrfs_del_inode_ref_in_log(struct btrfs_trans_handle *trans, | |
3331 | struct btrfs_root *root, | |
3332 | const char *name, int name_len, | |
a491abb2 | 3333 | struct btrfs_inode *inode, u64 dirid) |
e02119d5 | 3334 | { |
0b246afa | 3335 | struct btrfs_fs_info *fs_info = root->fs_info; |
e02119d5 CM |
3336 | struct btrfs_root *log; |
3337 | u64 index; | |
3338 | int ret; | |
3339 | ||
a491abb2 | 3340 | if (inode->logged_trans < trans->transid) |
3a5f1d45 CM |
3341 | return 0; |
3342 | ||
e02119d5 CM |
3343 | ret = join_running_log_trans(root); |
3344 | if (ret) | |
3345 | return 0; | |
3346 | log = root->log_root; | |
a491abb2 | 3347 | mutex_lock(&inode->log_mutex); |
e02119d5 | 3348 | |
a491abb2 | 3349 | ret = btrfs_del_inode_ref(trans, log, name, name_len, btrfs_ino(inode), |
e02119d5 | 3350 | dirid, &index); |
a491abb2 | 3351 | mutex_unlock(&inode->log_mutex); |
4a500fd1 | 3352 | if (ret == -ENOSPC) { |
0b246afa | 3353 | btrfs_set_log_full_commit(fs_info, trans); |
4a500fd1 | 3354 | ret = 0; |
79787eaa | 3355 | } else if (ret < 0 && ret != -ENOENT) |
66642832 | 3356 | btrfs_abort_transaction(trans, ret); |
12fcfd22 | 3357 | btrfs_end_log_trans(root); |
e02119d5 | 3358 | |
e02119d5 CM |
3359 | return ret; |
3360 | } | |
3361 | ||
3362 | /* | |
3363 | * creates a range item in the log for 'dirid'. first_offset and | |
3364 | * last_offset tell us which parts of the key space the log should | |
3365 | * be considered authoritative for. | |
3366 | */ | |
3367 | static noinline int insert_dir_log_key(struct btrfs_trans_handle *trans, | |
3368 | struct btrfs_root *log, | |
3369 | struct btrfs_path *path, | |
3370 | int key_type, u64 dirid, | |
3371 | u64 first_offset, u64 last_offset) | |
3372 | { | |
3373 | int ret; | |
3374 | struct btrfs_key key; | |
3375 | struct btrfs_dir_log_item *item; | |
3376 | ||
3377 | key.objectid = dirid; | |
3378 | key.offset = first_offset; | |
3379 | if (key_type == BTRFS_DIR_ITEM_KEY) | |
3380 | key.type = BTRFS_DIR_LOG_ITEM_KEY; | |
3381 | else | |
3382 | key.type = BTRFS_DIR_LOG_INDEX_KEY; | |
3383 | ret = btrfs_insert_empty_item(trans, log, path, &key, sizeof(*item)); | |
4a500fd1 YZ |
3384 | if (ret) |
3385 | return ret; | |
e02119d5 CM |
3386 | |
3387 | item = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
3388 | struct btrfs_dir_log_item); | |
3389 | btrfs_set_dir_log_end(path->nodes[0], item, last_offset); | |
3390 | btrfs_mark_buffer_dirty(path->nodes[0]); | |
b3b4aa74 | 3391 | btrfs_release_path(path); |
e02119d5 CM |
3392 | return 0; |
3393 | } | |
3394 | ||
3395 | /* | |
3396 | * log all the items included in the current transaction for a given | |
3397 | * directory. This also creates the range items in the log tree required | |
3398 | * to replay anything deleted before the fsync | |
3399 | */ | |
3400 | static noinline int log_dir_items(struct btrfs_trans_handle *trans, | |
684a5773 | 3401 | struct btrfs_root *root, struct btrfs_inode *inode, |
e02119d5 CM |
3402 | struct btrfs_path *path, |
3403 | struct btrfs_path *dst_path, int key_type, | |
2f2ff0ee | 3404 | struct btrfs_log_ctx *ctx, |
e02119d5 CM |
3405 | u64 min_offset, u64 *last_offset_ret) |
3406 | { | |
3407 | struct btrfs_key min_key; | |
e02119d5 CM |
3408 | struct btrfs_root *log = root->log_root; |
3409 | struct extent_buffer *src; | |
4a500fd1 | 3410 | int err = 0; |
e02119d5 CM |
3411 | int ret; |
3412 | int i; | |
3413 | int nritems; | |
3414 | u64 first_offset = min_offset; | |
3415 | u64 last_offset = (u64)-1; | |
684a5773 | 3416 | u64 ino = btrfs_ino(inode); |
e02119d5 CM |
3417 | |
3418 | log = root->log_root; | |
e02119d5 | 3419 | |
33345d01 | 3420 | min_key.objectid = ino; |
e02119d5 CM |
3421 | min_key.type = key_type; |
3422 | min_key.offset = min_offset; | |
3423 | ||
6174d3cb | 3424 | ret = btrfs_search_forward(root, &min_key, path, trans->transid); |
e02119d5 CM |
3425 | |
3426 | /* | |
3427 | * we didn't find anything from this transaction, see if there | |
3428 | * is anything at all | |
3429 | */ | |
33345d01 LZ |
3430 | if (ret != 0 || min_key.objectid != ino || min_key.type != key_type) { |
3431 | min_key.objectid = ino; | |
e02119d5 CM |
3432 | min_key.type = key_type; |
3433 | min_key.offset = (u64)-1; | |
b3b4aa74 | 3434 | btrfs_release_path(path); |
e02119d5 CM |
3435 | ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0); |
3436 | if (ret < 0) { | |
b3b4aa74 | 3437 | btrfs_release_path(path); |
e02119d5 CM |
3438 | return ret; |
3439 | } | |
33345d01 | 3440 | ret = btrfs_previous_item(root, path, ino, key_type); |
e02119d5 CM |
3441 | |
3442 | /* if ret == 0 there are items for this type, | |
3443 | * create a range to tell us the last key of this type. | |
3444 | * otherwise, there are no items in this directory after | |
3445 | * *min_offset, and we create a range to indicate that. | |
3446 | */ | |
3447 | if (ret == 0) { | |
3448 | struct btrfs_key tmp; | |
3449 | btrfs_item_key_to_cpu(path->nodes[0], &tmp, | |
3450 | path->slots[0]); | |
d397712b | 3451 | if (key_type == tmp.type) |
e02119d5 | 3452 | first_offset = max(min_offset, tmp.offset) + 1; |
e02119d5 CM |
3453 | } |
3454 | goto done; | |
3455 | } | |
3456 | ||
3457 | /* go backward to find any previous key */ | |
33345d01 | 3458 | ret = btrfs_previous_item(root, path, ino, key_type); |
e02119d5 CM |
3459 | if (ret == 0) { |
3460 | struct btrfs_key tmp; | |
3461 | btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]); | |
3462 | if (key_type == tmp.type) { | |
3463 | first_offset = tmp.offset; | |
3464 | ret = overwrite_item(trans, log, dst_path, | |
3465 | path->nodes[0], path->slots[0], | |
3466 | &tmp); | |
4a500fd1 YZ |
3467 | if (ret) { |
3468 | err = ret; | |
3469 | goto done; | |
3470 | } | |
e02119d5 CM |
3471 | } |
3472 | } | |
b3b4aa74 | 3473 | btrfs_release_path(path); |
e02119d5 CM |
3474 | |
3475 | /* find the first key from this transaction again */ | |
3476 | ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0); | |
fae7f21c | 3477 | if (WARN_ON(ret != 0)) |
e02119d5 | 3478 | goto done; |
e02119d5 CM |
3479 | |
3480 | /* | |
3481 | * we have a block from this transaction, log every item in it | |
3482 | * from our directory | |
3483 | */ | |
d397712b | 3484 | while (1) { |
e02119d5 CM |
3485 | struct btrfs_key tmp; |
3486 | src = path->nodes[0]; | |
3487 | nritems = btrfs_header_nritems(src); | |
3488 | for (i = path->slots[0]; i < nritems; i++) { | |
2f2ff0ee FM |
3489 | struct btrfs_dir_item *di; |
3490 | ||
e02119d5 CM |
3491 | btrfs_item_key_to_cpu(src, &min_key, i); |
3492 | ||
33345d01 | 3493 | if (min_key.objectid != ino || min_key.type != key_type) |
e02119d5 CM |
3494 | goto done; |
3495 | ret = overwrite_item(trans, log, dst_path, src, i, | |
3496 | &min_key); | |
4a500fd1 YZ |
3497 | if (ret) { |
3498 | err = ret; | |
3499 | goto done; | |
3500 | } | |
2f2ff0ee FM |
3501 | |
3502 | /* | |
3503 | * We must make sure that when we log a directory entry, | |
3504 | * the corresponding inode, after log replay, has a | |
3505 | * matching link count. For example: | |
3506 | * | |
3507 | * touch foo | |
3508 | * mkdir mydir | |
3509 | * sync | |
3510 | * ln foo mydir/bar | |
3511 | * xfs_io -c "fsync" mydir | |
3512 | * <crash> | |
3513 | * <mount fs and log replay> | |
3514 | * | |
3515 | * Would result in a fsync log that when replayed, our | |
3516 | * file inode would have a link count of 1, but we get | |
3517 | * two directory entries pointing to the same inode. | |
3518 | * After removing one of the names, it would not be | |
3519 | * possible to remove the other name, which resulted | |
3520 | * always in stale file handle errors, and would not | |
3521 | * be possible to rmdir the parent directory, since | |
3522 | * its i_size could never decrement to the value | |
3523 | * BTRFS_EMPTY_DIR_SIZE, resulting in -ENOTEMPTY errors. | |
3524 | */ | |
3525 | di = btrfs_item_ptr(src, i, struct btrfs_dir_item); | |
3526 | btrfs_dir_item_key_to_cpu(src, di, &tmp); | |
3527 | if (ctx && | |
3528 | (btrfs_dir_transid(src, di) == trans->transid || | |
3529 | btrfs_dir_type(src, di) == BTRFS_FT_DIR) && | |
3530 | tmp.type != BTRFS_ROOT_ITEM_KEY) | |
3531 | ctx->log_new_dentries = true; | |
e02119d5 CM |
3532 | } |
3533 | path->slots[0] = nritems; | |
3534 | ||
3535 | /* | |
3536 | * look ahead to the next item and see if it is also | |
3537 | * from this directory and from this transaction | |
3538 | */ | |
3539 | ret = btrfs_next_leaf(root, path); | |
80c0b421 LB |
3540 | if (ret) { |
3541 | if (ret == 1) | |
3542 | last_offset = (u64)-1; | |
3543 | else | |
3544 | err = ret; | |
e02119d5 CM |
3545 | goto done; |
3546 | } | |
3547 | btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]); | |
33345d01 | 3548 | if (tmp.objectid != ino || tmp.type != key_type) { |
e02119d5 CM |
3549 | last_offset = (u64)-1; |
3550 | goto done; | |
3551 | } | |
3552 | if (btrfs_header_generation(path->nodes[0]) != trans->transid) { | |
3553 | ret = overwrite_item(trans, log, dst_path, | |
3554 | path->nodes[0], path->slots[0], | |
3555 | &tmp); | |
4a500fd1 YZ |
3556 | if (ret) |
3557 | err = ret; | |
3558 | else | |
3559 | last_offset = tmp.offset; | |
e02119d5 CM |
3560 | goto done; |
3561 | } | |
3562 | } | |
3563 | done: | |
b3b4aa74 DS |
3564 | btrfs_release_path(path); |
3565 | btrfs_release_path(dst_path); | |
e02119d5 | 3566 | |
4a500fd1 YZ |
3567 | if (err == 0) { |
3568 | *last_offset_ret = last_offset; | |
3569 | /* | |
3570 | * insert the log range keys to indicate where the log | |
3571 | * is valid | |
3572 | */ | |
3573 | ret = insert_dir_log_key(trans, log, path, key_type, | |
33345d01 | 3574 | ino, first_offset, last_offset); |
4a500fd1 YZ |
3575 | if (ret) |
3576 | err = ret; | |
3577 | } | |
3578 | return err; | |
e02119d5 CM |
3579 | } |
3580 | ||
3581 | /* | |
3582 | * logging directories is very similar to logging inodes, We find all the items | |
3583 | * from the current transaction and write them to the log. | |
3584 | * | |
3585 | * The recovery code scans the directory in the subvolume, and if it finds a | |
3586 | * key in the range logged that is not present in the log tree, then it means | |
3587 | * that dir entry was unlinked during the transaction. | |
3588 | * | |
3589 | * In order for that scan to work, we must include one key smaller than | |
3590 | * the smallest logged by this transaction and one key larger than the largest | |
3591 | * key logged by this transaction. | |
3592 | */ | |
3593 | static noinline int log_directory_changes(struct btrfs_trans_handle *trans, | |
dbf39ea4 | 3594 | struct btrfs_root *root, struct btrfs_inode *inode, |
e02119d5 | 3595 | struct btrfs_path *path, |
2f2ff0ee FM |
3596 | struct btrfs_path *dst_path, |
3597 | struct btrfs_log_ctx *ctx) | |
e02119d5 CM |
3598 | { |
3599 | u64 min_key; | |
3600 | u64 max_key; | |
3601 | int ret; | |
3602 | int key_type = BTRFS_DIR_ITEM_KEY; | |
3603 | ||
3604 | again: | |
3605 | min_key = 0; | |
3606 | max_key = 0; | |
d397712b | 3607 | while (1) { |
dbf39ea4 NB |
3608 | ret = log_dir_items(trans, root, inode, path, dst_path, key_type, |
3609 | ctx, min_key, &max_key); | |
4a500fd1 YZ |
3610 | if (ret) |
3611 | return ret; | |
e02119d5 CM |
3612 | if (max_key == (u64)-1) |
3613 | break; | |
3614 | min_key = max_key + 1; | |
3615 | } | |
3616 | ||
3617 | if (key_type == BTRFS_DIR_ITEM_KEY) { | |
3618 | key_type = BTRFS_DIR_INDEX_KEY; | |
3619 | goto again; | |
3620 | } | |
3621 | return 0; | |
3622 | } | |
3623 | ||
3624 | /* | |
3625 | * a helper function to drop items from the log before we relog an | |
3626 | * inode. max_key_type indicates the highest item type to remove. | |
3627 | * This cannot be run for file data extents because it does not | |
3628 | * free the extents they point to. | |
3629 | */ | |
3630 | static int drop_objectid_items(struct btrfs_trans_handle *trans, | |
3631 | struct btrfs_root *log, | |
3632 | struct btrfs_path *path, | |
3633 | u64 objectid, int max_key_type) | |
3634 | { | |
3635 | int ret; | |
3636 | struct btrfs_key key; | |
3637 | struct btrfs_key found_key; | |
18ec90d6 | 3638 | int start_slot; |
e02119d5 CM |
3639 | |
3640 | key.objectid = objectid; | |
3641 | key.type = max_key_type; | |
3642 | key.offset = (u64)-1; | |
3643 | ||
d397712b | 3644 | while (1) { |
e02119d5 | 3645 | ret = btrfs_search_slot(trans, log, &key, path, -1, 1); |
3650860b | 3646 | BUG_ON(ret == 0); /* Logic error */ |
4a500fd1 | 3647 | if (ret < 0) |
e02119d5 CM |
3648 | break; |
3649 | ||
3650 | if (path->slots[0] == 0) | |
3651 | break; | |
3652 | ||
3653 | path->slots[0]--; | |
3654 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, | |
3655 | path->slots[0]); | |
3656 | ||
3657 | if (found_key.objectid != objectid) | |
3658 | break; | |
3659 | ||
18ec90d6 JB |
3660 | found_key.offset = 0; |
3661 | found_key.type = 0; | |
3662 | ret = btrfs_bin_search(path->nodes[0], &found_key, 0, | |
3663 | &start_slot); | |
3664 | ||
3665 | ret = btrfs_del_items(trans, log, path, start_slot, | |
3666 | path->slots[0] - start_slot + 1); | |
3667 | /* | |
3668 | * If start slot isn't 0 then we don't need to re-search, we've | |
3669 | * found the last guy with the objectid in this tree. | |
3670 | */ | |
3671 | if (ret || start_slot != 0) | |
65a246c5 | 3672 | break; |
b3b4aa74 | 3673 | btrfs_release_path(path); |
e02119d5 | 3674 | } |
b3b4aa74 | 3675 | btrfs_release_path(path); |
5bdbeb21 JB |
3676 | if (ret > 0) |
3677 | ret = 0; | |
4a500fd1 | 3678 | return ret; |
e02119d5 CM |
3679 | } |
3680 | ||
94edf4ae JB |
3681 | static void fill_inode_item(struct btrfs_trans_handle *trans, |
3682 | struct extent_buffer *leaf, | |
3683 | struct btrfs_inode_item *item, | |
1a4bcf47 FM |
3684 | struct inode *inode, int log_inode_only, |
3685 | u64 logged_isize) | |
94edf4ae | 3686 | { |
0b1c6cca JB |
3687 | struct btrfs_map_token token; |
3688 | ||
3689 | btrfs_init_map_token(&token); | |
94edf4ae JB |
3690 | |
3691 | if (log_inode_only) { | |
3692 | /* set the generation to zero so the recover code | |
3693 | * can tell the difference between an logging | |
3694 | * just to say 'this inode exists' and a logging | |
3695 | * to say 'update this inode with these values' | |
3696 | */ | |
0b1c6cca | 3697 | btrfs_set_token_inode_generation(leaf, item, 0, &token); |
1a4bcf47 | 3698 | btrfs_set_token_inode_size(leaf, item, logged_isize, &token); |
94edf4ae | 3699 | } else { |
0b1c6cca JB |
3700 | btrfs_set_token_inode_generation(leaf, item, |
3701 | BTRFS_I(inode)->generation, | |
3702 | &token); | |
3703 | btrfs_set_token_inode_size(leaf, item, inode->i_size, &token); | |
3704 | } | |
3705 | ||
3706 | btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token); | |
3707 | btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token); | |
3708 | btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token); | |
3709 | btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token); | |
3710 | ||
a937b979 | 3711 | btrfs_set_token_timespec_sec(leaf, &item->atime, |
0b1c6cca | 3712 | inode->i_atime.tv_sec, &token); |
a937b979 | 3713 | btrfs_set_token_timespec_nsec(leaf, &item->atime, |
0b1c6cca JB |
3714 | inode->i_atime.tv_nsec, &token); |
3715 | ||
a937b979 | 3716 | btrfs_set_token_timespec_sec(leaf, &item->mtime, |
0b1c6cca | 3717 | inode->i_mtime.tv_sec, &token); |
a937b979 | 3718 | btrfs_set_token_timespec_nsec(leaf, &item->mtime, |
0b1c6cca JB |
3719 | inode->i_mtime.tv_nsec, &token); |
3720 | ||
a937b979 | 3721 | btrfs_set_token_timespec_sec(leaf, &item->ctime, |
0b1c6cca | 3722 | inode->i_ctime.tv_sec, &token); |
a937b979 | 3723 | btrfs_set_token_timespec_nsec(leaf, &item->ctime, |
0b1c6cca JB |
3724 | inode->i_ctime.tv_nsec, &token); |
3725 | ||
3726 | btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode), | |
3727 | &token); | |
3728 | ||
c7f88c4e JL |
3729 | btrfs_set_token_inode_sequence(leaf, item, |
3730 | inode_peek_iversion(inode), &token); | |
0b1c6cca JB |
3731 | btrfs_set_token_inode_transid(leaf, item, trans->transid, &token); |
3732 | btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token); | |
3733 | btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token); | |
3734 | btrfs_set_token_inode_block_group(leaf, item, 0, &token); | |
94edf4ae JB |
3735 | } |
3736 | ||
a95249b3 JB |
3737 | static int log_inode_item(struct btrfs_trans_handle *trans, |
3738 | struct btrfs_root *log, struct btrfs_path *path, | |
6d889a3b | 3739 | struct btrfs_inode *inode) |
a95249b3 JB |
3740 | { |
3741 | struct btrfs_inode_item *inode_item; | |
a95249b3 JB |
3742 | int ret; |
3743 | ||
efd0c405 | 3744 | ret = btrfs_insert_empty_item(trans, log, path, |
6d889a3b | 3745 | &inode->location, sizeof(*inode_item)); |
a95249b3 JB |
3746 | if (ret && ret != -EEXIST) |
3747 | return ret; | |
3748 | inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
3749 | struct btrfs_inode_item); | |
6d889a3b NB |
3750 | fill_inode_item(trans, path->nodes[0], inode_item, &inode->vfs_inode, |
3751 | 0, 0); | |
a95249b3 JB |
3752 | btrfs_release_path(path); |
3753 | return 0; | |
3754 | } | |
3755 | ||
31ff1cd2 | 3756 | static noinline int copy_items(struct btrfs_trans_handle *trans, |
44d70e19 | 3757 | struct btrfs_inode *inode, |
31ff1cd2 | 3758 | struct btrfs_path *dst_path, |
16e7549f | 3759 | struct btrfs_path *src_path, u64 *last_extent, |
1a4bcf47 FM |
3760 | int start_slot, int nr, int inode_only, |
3761 | u64 logged_isize) | |
31ff1cd2 | 3762 | { |
44d70e19 | 3763 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb); |
31ff1cd2 CM |
3764 | unsigned long src_offset; |
3765 | unsigned long dst_offset; | |
44d70e19 | 3766 | struct btrfs_root *log = inode->root->log_root; |
31ff1cd2 CM |
3767 | struct btrfs_file_extent_item *extent; |
3768 | struct btrfs_inode_item *inode_item; | |
16e7549f JB |
3769 | struct extent_buffer *src = src_path->nodes[0]; |
3770 | struct btrfs_key first_key, last_key, key; | |
31ff1cd2 CM |
3771 | int ret; |
3772 | struct btrfs_key *ins_keys; | |
3773 | u32 *ins_sizes; | |
3774 | char *ins_data; | |
3775 | int i; | |
d20f7043 | 3776 | struct list_head ordered_sums; |
44d70e19 | 3777 | int skip_csum = inode->flags & BTRFS_INODE_NODATASUM; |
16e7549f | 3778 | bool has_extents = false; |
74121f7c | 3779 | bool need_find_last_extent = true; |
16e7549f | 3780 | bool done = false; |
d20f7043 CM |
3781 | |
3782 | INIT_LIST_HEAD(&ordered_sums); | |
31ff1cd2 CM |
3783 | |
3784 | ins_data = kmalloc(nr * sizeof(struct btrfs_key) + | |
3785 | nr * sizeof(u32), GFP_NOFS); | |
2a29edc6 | 3786 | if (!ins_data) |
3787 | return -ENOMEM; | |
3788 | ||
16e7549f JB |
3789 | first_key.objectid = (u64)-1; |
3790 | ||
31ff1cd2 CM |
3791 | ins_sizes = (u32 *)ins_data; |
3792 | ins_keys = (struct btrfs_key *)(ins_data + nr * sizeof(u32)); | |
3793 | ||
3794 | for (i = 0; i < nr; i++) { | |
3795 | ins_sizes[i] = btrfs_item_size_nr(src, i + start_slot); | |
3796 | btrfs_item_key_to_cpu(src, ins_keys + i, i + start_slot); | |
3797 | } | |
3798 | ret = btrfs_insert_empty_items(trans, log, dst_path, | |
3799 | ins_keys, ins_sizes, nr); | |
4a500fd1 YZ |
3800 | if (ret) { |
3801 | kfree(ins_data); | |
3802 | return ret; | |
3803 | } | |
31ff1cd2 | 3804 | |
5d4f98a2 | 3805 | for (i = 0; i < nr; i++, dst_path->slots[0]++) { |
31ff1cd2 CM |
3806 | dst_offset = btrfs_item_ptr_offset(dst_path->nodes[0], |
3807 | dst_path->slots[0]); | |
3808 | ||
3809 | src_offset = btrfs_item_ptr_offset(src, start_slot + i); | |
3810 | ||
0dde10be | 3811 | if (i == nr - 1) |
16e7549f JB |
3812 | last_key = ins_keys[i]; |
3813 | ||
94edf4ae | 3814 | if (ins_keys[i].type == BTRFS_INODE_ITEM_KEY) { |
31ff1cd2 CM |
3815 | inode_item = btrfs_item_ptr(dst_path->nodes[0], |
3816 | dst_path->slots[0], | |
3817 | struct btrfs_inode_item); | |
94edf4ae | 3818 | fill_inode_item(trans, dst_path->nodes[0], inode_item, |
f85b7379 DS |
3819 | &inode->vfs_inode, |
3820 | inode_only == LOG_INODE_EXISTS, | |
1a4bcf47 | 3821 | logged_isize); |
94edf4ae JB |
3822 | } else { |
3823 | copy_extent_buffer(dst_path->nodes[0], src, dst_offset, | |
3824 | src_offset, ins_sizes[i]); | |
31ff1cd2 | 3825 | } |
94edf4ae | 3826 | |
16e7549f JB |
3827 | /* |
3828 | * We set need_find_last_extent here in case we know we were | |
3829 | * processing other items and then walk into the first extent in | |
3830 | * the inode. If we don't hit an extent then nothing changes, | |
3831 | * we'll do the last search the next time around. | |
3832 | */ | |
3833 | if (ins_keys[i].type == BTRFS_EXTENT_DATA_KEY) { | |
3834 | has_extents = true; | |
74121f7c | 3835 | if (first_key.objectid == (u64)-1) |
16e7549f JB |
3836 | first_key = ins_keys[i]; |
3837 | } else { | |
3838 | need_find_last_extent = false; | |
3839 | } | |
3840 | ||
31ff1cd2 CM |
3841 | /* take a reference on file data extents so that truncates |
3842 | * or deletes of this inode don't have to relog the inode | |
3843 | * again | |
3844 | */ | |
962a298f | 3845 | if (ins_keys[i].type == BTRFS_EXTENT_DATA_KEY && |
d2794405 | 3846 | !skip_csum) { |
31ff1cd2 CM |
3847 | int found_type; |
3848 | extent = btrfs_item_ptr(src, start_slot + i, | |
3849 | struct btrfs_file_extent_item); | |
3850 | ||
8e531cdf | 3851 | if (btrfs_file_extent_generation(src, extent) < trans->transid) |
3852 | continue; | |
3853 | ||
31ff1cd2 | 3854 | found_type = btrfs_file_extent_type(src, extent); |
6f1fed77 | 3855 | if (found_type == BTRFS_FILE_EXTENT_REG) { |
5d4f98a2 YZ |
3856 | u64 ds, dl, cs, cl; |
3857 | ds = btrfs_file_extent_disk_bytenr(src, | |
3858 | extent); | |
3859 | /* ds == 0 is a hole */ | |
3860 | if (ds == 0) | |
3861 | continue; | |
3862 | ||
3863 | dl = btrfs_file_extent_disk_num_bytes(src, | |
3864 | extent); | |
3865 | cs = btrfs_file_extent_offset(src, extent); | |
3866 | cl = btrfs_file_extent_num_bytes(src, | |
a419aef8 | 3867 | extent); |
580afd76 CM |
3868 | if (btrfs_file_extent_compression(src, |
3869 | extent)) { | |
3870 | cs = 0; | |
3871 | cl = dl; | |
3872 | } | |
5d4f98a2 YZ |
3873 | |
3874 | ret = btrfs_lookup_csums_range( | |
0b246afa | 3875 | fs_info->csum_root, |
5d4f98a2 | 3876 | ds + cs, ds + cs + cl - 1, |
a2de733c | 3877 | &ordered_sums, 0); |
3650860b JB |
3878 | if (ret) { |
3879 | btrfs_release_path(dst_path); | |
3880 | kfree(ins_data); | |
3881 | return ret; | |
3882 | } | |
31ff1cd2 CM |
3883 | } |
3884 | } | |
31ff1cd2 CM |
3885 | } |
3886 | ||
3887 | btrfs_mark_buffer_dirty(dst_path->nodes[0]); | |
b3b4aa74 | 3888 | btrfs_release_path(dst_path); |
31ff1cd2 | 3889 | kfree(ins_data); |
d20f7043 CM |
3890 | |
3891 | /* | |
3892 | * we have to do this after the loop above to avoid changing the | |
3893 | * log tree while trying to change the log tree. | |
3894 | */ | |
4a500fd1 | 3895 | ret = 0; |
d397712b | 3896 | while (!list_empty(&ordered_sums)) { |
d20f7043 CM |
3897 | struct btrfs_ordered_sum *sums = list_entry(ordered_sums.next, |
3898 | struct btrfs_ordered_sum, | |
3899 | list); | |
4a500fd1 YZ |
3900 | if (!ret) |
3901 | ret = btrfs_csum_file_blocks(trans, log, sums); | |
d20f7043 CM |
3902 | list_del(&sums->list); |
3903 | kfree(sums); | |
3904 | } | |
16e7549f JB |
3905 | |
3906 | if (!has_extents) | |
3907 | return ret; | |
3908 | ||
74121f7c FM |
3909 | if (need_find_last_extent && *last_extent == first_key.offset) { |
3910 | /* | |
3911 | * We don't have any leafs between our current one and the one | |
3912 | * we processed before that can have file extent items for our | |
3913 | * inode (and have a generation number smaller than our current | |
3914 | * transaction id). | |
3915 | */ | |
3916 | need_find_last_extent = false; | |
3917 | } | |
3918 | ||
16e7549f JB |
3919 | /* |
3920 | * Because we use btrfs_search_forward we could skip leaves that were | |
3921 | * not modified and then assume *last_extent is valid when it really | |
3922 | * isn't. So back up to the previous leaf and read the end of the last | |
3923 | * extent before we go and fill in holes. | |
3924 | */ | |
3925 | if (need_find_last_extent) { | |
3926 | u64 len; | |
3927 | ||
44d70e19 | 3928 | ret = btrfs_prev_leaf(inode->root, src_path); |
16e7549f JB |
3929 | if (ret < 0) |
3930 | return ret; | |
3931 | if (ret) | |
3932 | goto fill_holes; | |
3933 | if (src_path->slots[0]) | |
3934 | src_path->slots[0]--; | |
3935 | src = src_path->nodes[0]; | |
3936 | btrfs_item_key_to_cpu(src, &key, src_path->slots[0]); | |
44d70e19 | 3937 | if (key.objectid != btrfs_ino(inode) || |
16e7549f JB |
3938 | key.type != BTRFS_EXTENT_DATA_KEY) |
3939 | goto fill_holes; | |
3940 | extent = btrfs_item_ptr(src, src_path->slots[0], | |
3941 | struct btrfs_file_extent_item); | |
3942 | if (btrfs_file_extent_type(src, extent) == | |
3943 | BTRFS_FILE_EXTENT_INLINE) { | |
514ac8ad CM |
3944 | len = btrfs_file_extent_inline_len(src, |
3945 | src_path->slots[0], | |
3946 | extent); | |
16e7549f | 3947 | *last_extent = ALIGN(key.offset + len, |
0b246afa | 3948 | fs_info->sectorsize); |
16e7549f JB |
3949 | } else { |
3950 | len = btrfs_file_extent_num_bytes(src, extent); | |
3951 | *last_extent = key.offset + len; | |
3952 | } | |
3953 | } | |
3954 | fill_holes: | |
3955 | /* So we did prev_leaf, now we need to move to the next leaf, but a few | |
3956 | * things could have happened | |
3957 | * | |
3958 | * 1) A merge could have happened, so we could currently be on a leaf | |
3959 | * that holds what we were copying in the first place. | |
3960 | * 2) A split could have happened, and now not all of the items we want | |
3961 | * are on the same leaf. | |
3962 | * | |
3963 | * So we need to adjust how we search for holes, we need to drop the | |
3964 | * path and re-search for the first extent key we found, and then walk | |
3965 | * forward until we hit the last one we copied. | |
3966 | */ | |
3967 | if (need_find_last_extent) { | |
3968 | /* btrfs_prev_leaf could return 1 without releasing the path */ | |
3969 | btrfs_release_path(src_path); | |
f85b7379 DS |
3970 | ret = btrfs_search_slot(NULL, inode->root, &first_key, |
3971 | src_path, 0, 0); | |
16e7549f JB |
3972 | if (ret < 0) |
3973 | return ret; | |
3974 | ASSERT(ret == 0); | |
3975 | src = src_path->nodes[0]; | |
3976 | i = src_path->slots[0]; | |
3977 | } else { | |
3978 | i = start_slot; | |
3979 | } | |
3980 | ||
3981 | /* | |
3982 | * Ok so here we need to go through and fill in any holes we may have | |
3983 | * to make sure that holes are punched for those areas in case they had | |
3984 | * extents previously. | |
3985 | */ | |
3986 | while (!done) { | |
3987 | u64 offset, len; | |
3988 | u64 extent_end; | |
3989 | ||
3990 | if (i >= btrfs_header_nritems(src_path->nodes[0])) { | |
44d70e19 | 3991 | ret = btrfs_next_leaf(inode->root, src_path); |
16e7549f JB |
3992 | if (ret < 0) |
3993 | return ret; | |
3994 | ASSERT(ret == 0); | |
3995 | src = src_path->nodes[0]; | |
3996 | i = 0; | |
8434ec46 | 3997 | need_find_last_extent = true; |
16e7549f JB |
3998 | } |
3999 | ||
4000 | btrfs_item_key_to_cpu(src, &key, i); | |
4001 | if (!btrfs_comp_cpu_keys(&key, &last_key)) | |
4002 | done = true; | |
44d70e19 | 4003 | if (key.objectid != btrfs_ino(inode) || |
16e7549f JB |
4004 | key.type != BTRFS_EXTENT_DATA_KEY) { |
4005 | i++; | |
4006 | continue; | |
4007 | } | |
4008 | extent = btrfs_item_ptr(src, i, struct btrfs_file_extent_item); | |
4009 | if (btrfs_file_extent_type(src, extent) == | |
4010 | BTRFS_FILE_EXTENT_INLINE) { | |
514ac8ad | 4011 | len = btrfs_file_extent_inline_len(src, i, extent); |
da17066c | 4012 | extent_end = ALIGN(key.offset + len, |
0b246afa | 4013 | fs_info->sectorsize); |
16e7549f JB |
4014 | } else { |
4015 | len = btrfs_file_extent_num_bytes(src, extent); | |
4016 | extent_end = key.offset + len; | |
4017 | } | |
4018 | i++; | |
4019 | ||
4020 | if (*last_extent == key.offset) { | |
4021 | *last_extent = extent_end; | |
4022 | continue; | |
4023 | } | |
4024 | offset = *last_extent; | |
4025 | len = key.offset - *last_extent; | |
44d70e19 | 4026 | ret = btrfs_insert_file_extent(trans, log, btrfs_ino(inode), |
f85b7379 | 4027 | offset, 0, 0, len, 0, len, 0, 0, 0); |
16e7549f JB |
4028 | if (ret) |
4029 | break; | |
74121f7c | 4030 | *last_extent = extent_end; |
16e7549f | 4031 | } |
4ee3fad3 FM |
4032 | |
4033 | /* | |
4034 | * Check if there is a hole between the last extent found in our leaf | |
4035 | * and the first extent in the next leaf. If there is one, we need to | |
4036 | * log an explicit hole so that at replay time we can punch the hole. | |
4037 | */ | |
4038 | if (ret == 0 && | |
4039 | key.objectid == btrfs_ino(inode) && | |
4040 | key.type == BTRFS_EXTENT_DATA_KEY && | |
4041 | i == btrfs_header_nritems(src_path->nodes[0])) { | |
4042 | ret = btrfs_next_leaf(inode->root, src_path); | |
4043 | need_find_last_extent = true; | |
4044 | if (ret > 0) { | |
4045 | ret = 0; | |
4046 | } else if (ret == 0) { | |
4047 | btrfs_item_key_to_cpu(src_path->nodes[0], &key, | |
4048 | src_path->slots[0]); | |
4049 | if (key.objectid == btrfs_ino(inode) && | |
4050 | key.type == BTRFS_EXTENT_DATA_KEY && | |
4051 | *last_extent < key.offset) { | |
4052 | const u64 len = key.offset - *last_extent; | |
4053 | ||
4054 | ret = btrfs_insert_file_extent(trans, log, | |
4055 | btrfs_ino(inode), | |
4056 | *last_extent, 0, | |
4057 | 0, len, 0, len, | |
4058 | 0, 0, 0); | |
4059 | } | |
4060 | } | |
4061 | } | |
16e7549f JB |
4062 | /* |
4063 | * Need to let the callers know we dropped the path so they should | |
4064 | * re-search. | |
4065 | */ | |
4066 | if (!ret && need_find_last_extent) | |
4067 | ret = 1; | |
4a500fd1 | 4068 | return ret; |
31ff1cd2 CM |
4069 | } |
4070 | ||
5dc562c5 JB |
4071 | static int extent_cmp(void *priv, struct list_head *a, struct list_head *b) |
4072 | { | |
4073 | struct extent_map *em1, *em2; | |
4074 | ||
4075 | em1 = list_entry(a, struct extent_map, list); | |
4076 | em2 = list_entry(b, struct extent_map, list); | |
4077 | ||
4078 | if (em1->start < em2->start) | |
4079 | return -1; | |
4080 | else if (em1->start > em2->start) | |
4081 | return 1; | |
4082 | return 0; | |
4083 | } | |
4084 | ||
8407f553 FM |
4085 | static int wait_ordered_extents(struct btrfs_trans_handle *trans, |
4086 | struct inode *inode, | |
4087 | struct btrfs_root *root, | |
4088 | const struct extent_map *em, | |
4089 | const struct list_head *logged_list, | |
4090 | bool *ordered_io_error) | |
5dc562c5 | 4091 | { |
0b246afa | 4092 | struct btrfs_fs_info *fs_info = root->fs_info; |
2ab28f32 | 4093 | struct btrfs_ordered_extent *ordered; |
8407f553 | 4094 | struct btrfs_root *log = root->log_root; |
2ab28f32 JB |
4095 | u64 mod_start = em->mod_start; |
4096 | u64 mod_len = em->mod_len; | |
8407f553 | 4097 | const bool skip_csum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM; |
2ab28f32 JB |
4098 | u64 csum_offset; |
4099 | u64 csum_len; | |
8407f553 FM |
4100 | LIST_HEAD(ordered_sums); |
4101 | int ret = 0; | |
0aa4a17d | 4102 | |
8407f553 | 4103 | *ordered_io_error = false; |
0aa4a17d | 4104 | |
8407f553 FM |
4105 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) || |
4106 | em->block_start == EXTENT_MAP_HOLE) | |
70c8a91c | 4107 | return 0; |
5dc562c5 | 4108 | |
2ab28f32 | 4109 | /* |
8407f553 FM |
4110 | * Wait far any ordered extent that covers our extent map. If it |
4111 | * finishes without an error, first check and see if our csums are on | |
4112 | * our outstanding ordered extents. | |
2ab28f32 | 4113 | */ |
827463c4 | 4114 | list_for_each_entry(ordered, logged_list, log_list) { |
2ab28f32 JB |
4115 | struct btrfs_ordered_sum *sum; |
4116 | ||
4117 | if (!mod_len) | |
4118 | break; | |
4119 | ||
2ab28f32 JB |
4120 | if (ordered->file_offset + ordered->len <= mod_start || |
4121 | mod_start + mod_len <= ordered->file_offset) | |
4122 | continue; | |
4123 | ||
8407f553 FM |
4124 | if (!test_bit(BTRFS_ORDERED_IO_DONE, &ordered->flags) && |
4125 | !test_bit(BTRFS_ORDERED_IOERR, &ordered->flags) && | |
4126 | !test_bit(BTRFS_ORDERED_DIRECT, &ordered->flags)) { | |
4127 | const u64 start = ordered->file_offset; | |
4128 | const u64 end = ordered->file_offset + ordered->len - 1; | |
4129 | ||
4130 | WARN_ON(ordered->inode != inode); | |
4131 | filemap_fdatawrite_range(inode->i_mapping, start, end); | |
4132 | } | |
4133 | ||
4134 | wait_event(ordered->wait, | |
4135 | (test_bit(BTRFS_ORDERED_IO_DONE, &ordered->flags) || | |
4136 | test_bit(BTRFS_ORDERED_IOERR, &ordered->flags))); | |
4137 | ||
4138 | if (test_bit(BTRFS_ORDERED_IOERR, &ordered->flags)) { | |
b38ef71c FM |
4139 | /* |
4140 | * Clear the AS_EIO/AS_ENOSPC flags from the inode's | |
4141 | * i_mapping flags, so that the next fsync won't get | |
4142 | * an outdated io error too. | |
4143 | */ | |
f0312210 | 4144 | filemap_check_errors(inode->i_mapping); |
8407f553 FM |
4145 | *ordered_io_error = true; |
4146 | break; | |
4147 | } | |
2ab28f32 JB |
4148 | /* |
4149 | * We are going to copy all the csums on this ordered extent, so | |
4150 | * go ahead and adjust mod_start and mod_len in case this | |
4151 | * ordered extent has already been logged. | |
4152 | */ | |
4153 | if (ordered->file_offset > mod_start) { | |
4154 | if (ordered->file_offset + ordered->len >= | |
4155 | mod_start + mod_len) | |
4156 | mod_len = ordered->file_offset - mod_start; | |
4157 | /* | |
4158 | * If we have this case | |
4159 | * | |
4160 | * |--------- logged extent ---------| | |
4161 | * |----- ordered extent ----| | |
4162 | * | |
4163 | * Just don't mess with mod_start and mod_len, we'll | |
4164 | * just end up logging more csums than we need and it | |
4165 | * will be ok. | |
4166 | */ | |
4167 | } else { | |
4168 | if (ordered->file_offset + ordered->len < | |
4169 | mod_start + mod_len) { | |
4170 | mod_len = (mod_start + mod_len) - | |
4171 | (ordered->file_offset + ordered->len); | |
4172 | mod_start = ordered->file_offset + | |
4173 | ordered->len; | |
4174 | } else { | |
4175 | mod_len = 0; | |
4176 | } | |
4177 | } | |
4178 | ||
8407f553 FM |
4179 | if (skip_csum) |
4180 | continue; | |
4181 | ||
2ab28f32 JB |
4182 | /* |
4183 | * To keep us from looping for the above case of an ordered | |
4184 | * extent that falls inside of the logged extent. | |
4185 | */ | |
4186 | if (test_and_set_bit(BTRFS_ORDERED_LOGGED_CSUM, | |
4187 | &ordered->flags)) | |
4188 | continue; | |
2ab28f32 | 4189 | |
2ab28f32 JB |
4190 | list_for_each_entry(sum, &ordered->list, list) { |
4191 | ret = btrfs_csum_file_blocks(trans, log, sum); | |
827463c4 | 4192 | if (ret) |
8407f553 | 4193 | break; |
2ab28f32 | 4194 | } |
2ab28f32 | 4195 | } |
2ab28f32 | 4196 | |
8407f553 | 4197 | if (*ordered_io_error || !mod_len || ret || skip_csum) |
2ab28f32 JB |
4198 | return ret; |
4199 | ||
488111aa FDBM |
4200 | if (em->compress_type) { |
4201 | csum_offset = 0; | |
8407f553 | 4202 | csum_len = max(em->block_len, em->orig_block_len); |
488111aa FDBM |
4203 | } else { |
4204 | csum_offset = mod_start - em->start; | |
4205 | csum_len = mod_len; | |
4206 | } | |
2ab28f32 | 4207 | |
70c8a91c | 4208 | /* block start is already adjusted for the file extent offset. */ |
0b246afa | 4209 | ret = btrfs_lookup_csums_range(fs_info->csum_root, |
70c8a91c JB |
4210 | em->block_start + csum_offset, |
4211 | em->block_start + csum_offset + | |
4212 | csum_len - 1, &ordered_sums, 0); | |
4213 | if (ret) | |
4214 | return ret; | |
5dc562c5 | 4215 | |
70c8a91c JB |
4216 | while (!list_empty(&ordered_sums)) { |
4217 | struct btrfs_ordered_sum *sums = list_entry(ordered_sums.next, | |
4218 | struct btrfs_ordered_sum, | |
4219 | list); | |
4220 | if (!ret) | |
4221 | ret = btrfs_csum_file_blocks(trans, log, sums); | |
4222 | list_del(&sums->list); | |
4223 | kfree(sums); | |
5dc562c5 JB |
4224 | } |
4225 | ||
70c8a91c | 4226 | return ret; |
5dc562c5 JB |
4227 | } |
4228 | ||
8407f553 | 4229 | static int log_one_extent(struct btrfs_trans_handle *trans, |
9d122629 | 4230 | struct btrfs_inode *inode, struct btrfs_root *root, |
8407f553 FM |
4231 | const struct extent_map *em, |
4232 | struct btrfs_path *path, | |
4233 | const struct list_head *logged_list, | |
4234 | struct btrfs_log_ctx *ctx) | |
4235 | { | |
4236 | struct btrfs_root *log = root->log_root; | |
4237 | struct btrfs_file_extent_item *fi; | |
4238 | struct extent_buffer *leaf; | |
4239 | struct btrfs_map_token token; | |
4240 | struct btrfs_key key; | |
4241 | u64 extent_offset = em->start - em->orig_start; | |
4242 | u64 block_len; | |
4243 | int ret; | |
4244 | int extent_inserted = 0; | |
4245 | bool ordered_io_err = false; | |
4246 | ||
f85b7379 DS |
4247 | ret = wait_ordered_extents(trans, &inode->vfs_inode, root, em, |
4248 | logged_list, &ordered_io_err); | |
8407f553 FM |
4249 | if (ret) |
4250 | return ret; | |
4251 | ||
4252 | if (ordered_io_err) { | |
4253 | ctx->io_err = -EIO; | |
ebb70442 | 4254 | return ctx->io_err; |
8407f553 FM |
4255 | } |
4256 | ||
4257 | btrfs_init_map_token(&token); | |
4258 | ||
9d122629 | 4259 | ret = __btrfs_drop_extents(trans, log, &inode->vfs_inode, path, em->start, |
8407f553 FM |
4260 | em->start + em->len, NULL, 0, 1, |
4261 | sizeof(*fi), &extent_inserted); | |
4262 | if (ret) | |
4263 | return ret; | |
4264 | ||
4265 | if (!extent_inserted) { | |
9d122629 | 4266 | key.objectid = btrfs_ino(inode); |
8407f553 FM |
4267 | key.type = BTRFS_EXTENT_DATA_KEY; |
4268 | key.offset = em->start; | |
4269 | ||
4270 | ret = btrfs_insert_empty_item(trans, log, path, &key, | |
4271 | sizeof(*fi)); | |
4272 | if (ret) | |
4273 | return ret; | |
4274 | } | |
4275 | leaf = path->nodes[0]; | |
4276 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
4277 | struct btrfs_file_extent_item); | |
4278 | ||
50d9aa99 | 4279 | btrfs_set_token_file_extent_generation(leaf, fi, trans->transid, |
8407f553 FM |
4280 | &token); |
4281 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) | |
4282 | btrfs_set_token_file_extent_type(leaf, fi, | |
4283 | BTRFS_FILE_EXTENT_PREALLOC, | |
4284 | &token); | |
4285 | else | |
4286 | btrfs_set_token_file_extent_type(leaf, fi, | |
4287 | BTRFS_FILE_EXTENT_REG, | |
4288 | &token); | |
4289 | ||
4290 | block_len = max(em->block_len, em->orig_block_len); | |
4291 | if (em->compress_type != BTRFS_COMPRESS_NONE) { | |
4292 | btrfs_set_token_file_extent_disk_bytenr(leaf, fi, | |
4293 | em->block_start, | |
4294 | &token); | |
4295 | btrfs_set_token_file_extent_disk_num_bytes(leaf, fi, block_len, | |
4296 | &token); | |
4297 | } else if (em->block_start < EXTENT_MAP_LAST_BYTE) { | |
4298 | btrfs_set_token_file_extent_disk_bytenr(leaf, fi, | |
4299 | em->block_start - | |
4300 | extent_offset, &token); | |
4301 | btrfs_set_token_file_extent_disk_num_bytes(leaf, fi, block_len, | |
4302 | &token); | |
4303 | } else { | |
4304 | btrfs_set_token_file_extent_disk_bytenr(leaf, fi, 0, &token); | |
4305 | btrfs_set_token_file_extent_disk_num_bytes(leaf, fi, 0, | |
4306 | &token); | |
4307 | } | |
4308 | ||
4309 | btrfs_set_token_file_extent_offset(leaf, fi, extent_offset, &token); | |
4310 | btrfs_set_token_file_extent_num_bytes(leaf, fi, em->len, &token); | |
4311 | btrfs_set_token_file_extent_ram_bytes(leaf, fi, em->ram_bytes, &token); | |
4312 | btrfs_set_token_file_extent_compression(leaf, fi, em->compress_type, | |
4313 | &token); | |
4314 | btrfs_set_token_file_extent_encryption(leaf, fi, 0, &token); | |
4315 | btrfs_set_token_file_extent_other_encoding(leaf, fi, 0, &token); | |
4316 | btrfs_mark_buffer_dirty(leaf); | |
4317 | ||
4318 | btrfs_release_path(path); | |
4319 | ||
4320 | return ret; | |
4321 | } | |
4322 | ||
5dc562c5 JB |
4323 | static int btrfs_log_changed_extents(struct btrfs_trans_handle *trans, |
4324 | struct btrfs_root *root, | |
9d122629 | 4325 | struct btrfs_inode *inode, |
827463c4 | 4326 | struct btrfs_path *path, |
8407f553 | 4327 | struct list_head *logged_list, |
de0ee0ed FM |
4328 | struct btrfs_log_ctx *ctx, |
4329 | const u64 start, | |
4330 | const u64 end) | |
5dc562c5 | 4331 | { |
5dc562c5 JB |
4332 | struct extent_map *em, *n; |
4333 | struct list_head extents; | |
9d122629 | 4334 | struct extent_map_tree *tree = &inode->extent_tree; |
8c6c5928 | 4335 | u64 logged_start, logged_end; |
5dc562c5 JB |
4336 | u64 test_gen; |
4337 | int ret = 0; | |
2ab28f32 | 4338 | int num = 0; |
5dc562c5 JB |
4339 | |
4340 | INIT_LIST_HEAD(&extents); | |
4341 | ||
9d122629 | 4342 | down_write(&inode->dio_sem); |
5dc562c5 JB |
4343 | write_lock(&tree->lock); |
4344 | test_gen = root->fs_info->last_trans_committed; | |
8c6c5928 JB |
4345 | logged_start = start; |
4346 | logged_end = end; | |
5dc562c5 JB |
4347 | |
4348 | list_for_each_entry_safe(em, n, &tree->modified_extents, list) { | |
4349 | list_del_init(&em->list); | |
2ab28f32 JB |
4350 | /* |
4351 | * Just an arbitrary number, this can be really CPU intensive | |
4352 | * once we start getting a lot of extents, and really once we | |
4353 | * have a bunch of extents we just want to commit since it will | |
4354 | * be faster. | |
4355 | */ | |
4356 | if (++num > 32768) { | |
4357 | list_del_init(&tree->modified_extents); | |
4358 | ret = -EFBIG; | |
4359 | goto process; | |
4360 | } | |
4361 | ||
5dc562c5 JB |
4362 | if (em->generation <= test_gen) |
4363 | continue; | |
8c6c5928 JB |
4364 | |
4365 | if (em->start < logged_start) | |
4366 | logged_start = em->start; | |
4367 | if ((em->start + em->len - 1) > logged_end) | |
4368 | logged_end = em->start + em->len - 1; | |
4369 | ||
ff44c6e3 | 4370 | /* Need a ref to keep it from getting evicted from cache */ |
490b54d6 | 4371 | refcount_inc(&em->refs); |
ff44c6e3 | 4372 | set_bit(EXTENT_FLAG_LOGGING, &em->flags); |
5dc562c5 | 4373 | list_add_tail(&em->list, &extents); |
2ab28f32 | 4374 | num++; |
5dc562c5 JB |
4375 | } |
4376 | ||
471d557a FM |
4377 | /* |
4378 | * Add all prealloc extents beyond the inode's i_size to make sure we | |
4379 | * don't lose them after doing a fast fsync and replaying the log. | |
4380 | */ | |
4381 | if (inode->flags & BTRFS_INODE_PREALLOC) { | |
4382 | struct rb_node *node; | |
4383 | ||
4384 | for (node = rb_last(&tree->map); node; node = rb_prev(node)) { | |
4385 | em = rb_entry(node, struct extent_map, rb_node); | |
4386 | if (em->start < i_size_read(&inode->vfs_inode)) | |
4387 | break; | |
4388 | if (!list_empty(&em->list)) | |
4389 | continue; | |
4390 | /* Same as above loop. */ | |
4391 | if (++num > 32768) { | |
4392 | list_del_init(&tree->modified_extents); | |
4393 | ret = -EFBIG; | |
4394 | goto process; | |
4395 | } | |
4396 | refcount_inc(&em->refs); | |
4397 | set_bit(EXTENT_FLAG_LOGGING, &em->flags); | |
4398 | list_add_tail(&em->list, &extents); | |
4399 | } | |
4400 | } | |
4401 | ||
5dc562c5 | 4402 | list_sort(NULL, &extents, extent_cmp); |
8c6c5928 | 4403 | btrfs_get_logged_extents(inode, logged_list, logged_start, logged_end); |
de0ee0ed | 4404 | /* |
5f9a8a51 FM |
4405 | * Some ordered extents started by fsync might have completed |
4406 | * before we could collect them into the list logged_list, which | |
4407 | * means they're gone, not in our logged_list nor in the inode's | |
4408 | * ordered tree. We want the application/user space to know an | |
4409 | * error happened while attempting to persist file data so that | |
4410 | * it can take proper action. If such error happened, we leave | |
4411 | * without writing to the log tree and the fsync must report the | |
4412 | * file data write error and not commit the current transaction. | |
de0ee0ed | 4413 | */ |
9d122629 | 4414 | ret = filemap_check_errors(inode->vfs_inode.i_mapping); |
5f9a8a51 FM |
4415 | if (ret) |
4416 | ctx->io_err = ret; | |
2ab28f32 | 4417 | process: |
5dc562c5 JB |
4418 | while (!list_empty(&extents)) { |
4419 | em = list_entry(extents.next, struct extent_map, list); | |
4420 | ||
4421 | list_del_init(&em->list); | |
4422 | ||
4423 | /* | |
4424 | * If we had an error we just need to delete everybody from our | |
4425 | * private list. | |
4426 | */ | |
ff44c6e3 | 4427 | if (ret) { |
201a9038 | 4428 | clear_em_logging(tree, em); |
ff44c6e3 | 4429 | free_extent_map(em); |
5dc562c5 | 4430 | continue; |
ff44c6e3 JB |
4431 | } |
4432 | ||
4433 | write_unlock(&tree->lock); | |
5dc562c5 | 4434 | |
8407f553 FM |
4435 | ret = log_one_extent(trans, inode, root, em, path, logged_list, |
4436 | ctx); | |
ff44c6e3 | 4437 | write_lock(&tree->lock); |
201a9038 JB |
4438 | clear_em_logging(tree, em); |
4439 | free_extent_map(em); | |
5dc562c5 | 4440 | } |
ff44c6e3 JB |
4441 | WARN_ON(!list_empty(&extents)); |
4442 | write_unlock(&tree->lock); | |
9d122629 | 4443 | up_write(&inode->dio_sem); |
5dc562c5 | 4444 | |
5dc562c5 | 4445 | btrfs_release_path(path); |
5dc562c5 JB |
4446 | return ret; |
4447 | } | |
4448 | ||
481b01c0 | 4449 | static int logged_inode_size(struct btrfs_root *log, struct btrfs_inode *inode, |
1a4bcf47 FM |
4450 | struct btrfs_path *path, u64 *size_ret) |
4451 | { | |
4452 | struct btrfs_key key; | |
4453 | int ret; | |
4454 | ||
481b01c0 | 4455 | key.objectid = btrfs_ino(inode); |
1a4bcf47 FM |
4456 | key.type = BTRFS_INODE_ITEM_KEY; |
4457 | key.offset = 0; | |
4458 | ||
4459 | ret = btrfs_search_slot(NULL, log, &key, path, 0, 0); | |
4460 | if (ret < 0) { | |
4461 | return ret; | |
4462 | } else if (ret > 0) { | |
2f2ff0ee | 4463 | *size_ret = 0; |
1a4bcf47 FM |
4464 | } else { |
4465 | struct btrfs_inode_item *item; | |
4466 | ||
4467 | item = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
4468 | struct btrfs_inode_item); | |
4469 | *size_ret = btrfs_inode_size(path->nodes[0], item); | |
4470 | } | |
4471 | ||
4472 | btrfs_release_path(path); | |
4473 | return 0; | |
4474 | } | |
4475 | ||
36283bf7 FM |
4476 | /* |
4477 | * At the moment we always log all xattrs. This is to figure out at log replay | |
4478 | * time which xattrs must have their deletion replayed. If a xattr is missing | |
4479 | * in the log tree and exists in the fs/subvol tree, we delete it. This is | |
4480 | * because if a xattr is deleted, the inode is fsynced and a power failure | |
4481 | * happens, causing the log to be replayed the next time the fs is mounted, | |
4482 | * we want the xattr to not exist anymore (same behaviour as other filesystems | |
4483 | * with a journal, ext3/4, xfs, f2fs, etc). | |
4484 | */ | |
4485 | static int btrfs_log_all_xattrs(struct btrfs_trans_handle *trans, | |
4486 | struct btrfs_root *root, | |
1a93c36a | 4487 | struct btrfs_inode *inode, |
36283bf7 FM |
4488 | struct btrfs_path *path, |
4489 | struct btrfs_path *dst_path) | |
4490 | { | |
4491 | int ret; | |
4492 | struct btrfs_key key; | |
1a93c36a | 4493 | const u64 ino = btrfs_ino(inode); |
36283bf7 FM |
4494 | int ins_nr = 0; |
4495 | int start_slot = 0; | |
4496 | ||
4497 | key.objectid = ino; | |
4498 | key.type = BTRFS_XATTR_ITEM_KEY; | |
4499 | key.offset = 0; | |
4500 | ||
4501 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
4502 | if (ret < 0) | |
4503 | return ret; | |
4504 | ||
4505 | while (true) { | |
4506 | int slot = path->slots[0]; | |
4507 | struct extent_buffer *leaf = path->nodes[0]; | |
4508 | int nritems = btrfs_header_nritems(leaf); | |
4509 | ||
4510 | if (slot >= nritems) { | |
4511 | if (ins_nr > 0) { | |
4512 | u64 last_extent = 0; | |
4513 | ||
1a93c36a | 4514 | ret = copy_items(trans, inode, dst_path, path, |
36283bf7 FM |
4515 | &last_extent, start_slot, |
4516 | ins_nr, 1, 0); | |
4517 | /* can't be 1, extent items aren't processed */ | |
4518 | ASSERT(ret <= 0); | |
4519 | if (ret < 0) | |
4520 | return ret; | |
4521 | ins_nr = 0; | |
4522 | } | |
4523 | ret = btrfs_next_leaf(root, path); | |
4524 | if (ret < 0) | |
4525 | return ret; | |
4526 | else if (ret > 0) | |
4527 | break; | |
4528 | continue; | |
4529 | } | |
4530 | ||
4531 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
4532 | if (key.objectid != ino || key.type != BTRFS_XATTR_ITEM_KEY) | |
4533 | break; | |
4534 | ||
4535 | if (ins_nr == 0) | |
4536 | start_slot = slot; | |
4537 | ins_nr++; | |
4538 | path->slots[0]++; | |
4539 | cond_resched(); | |
4540 | } | |
4541 | if (ins_nr > 0) { | |
4542 | u64 last_extent = 0; | |
4543 | ||
1a93c36a | 4544 | ret = copy_items(trans, inode, dst_path, path, |
36283bf7 FM |
4545 | &last_extent, start_slot, |
4546 | ins_nr, 1, 0); | |
4547 | /* can't be 1, extent items aren't processed */ | |
4548 | ASSERT(ret <= 0); | |
4549 | if (ret < 0) | |
4550 | return ret; | |
4551 | } | |
4552 | ||
4553 | return 0; | |
4554 | } | |
4555 | ||
a89ca6f2 FM |
4556 | /* |
4557 | * If the no holes feature is enabled we need to make sure any hole between the | |
4558 | * last extent and the i_size of our inode is explicitly marked in the log. This | |
4559 | * is to make sure that doing something like: | |
4560 | * | |
4561 | * 1) create file with 128Kb of data | |
4562 | * 2) truncate file to 64Kb | |
4563 | * 3) truncate file to 256Kb | |
4564 | * 4) fsync file | |
4565 | * 5) <crash/power failure> | |
4566 | * 6) mount fs and trigger log replay | |
4567 | * | |
4568 | * Will give us a file with a size of 256Kb, the first 64Kb of data match what | |
4569 | * the file had in its first 64Kb of data at step 1 and the last 192Kb of the | |
4570 | * file correspond to a hole. The presence of explicit holes in a log tree is | |
4571 | * what guarantees that log replay will remove/adjust file extent items in the | |
4572 | * fs/subvol tree. | |
4573 | * | |
4574 | * Here we do not need to care about holes between extents, that is already done | |
4575 | * by copy_items(). We also only need to do this in the full sync path, where we | |
4576 | * lookup for extents from the fs/subvol tree only. In the fast path case, we | |
4577 | * lookup the list of modified extent maps and if any represents a hole, we | |
4578 | * insert a corresponding extent representing a hole in the log tree. | |
4579 | */ | |
4580 | static int btrfs_log_trailing_hole(struct btrfs_trans_handle *trans, | |
4581 | struct btrfs_root *root, | |
a0308dd7 | 4582 | struct btrfs_inode *inode, |
a89ca6f2 FM |
4583 | struct btrfs_path *path) |
4584 | { | |
0b246afa | 4585 | struct btrfs_fs_info *fs_info = root->fs_info; |
a89ca6f2 FM |
4586 | int ret; |
4587 | struct btrfs_key key; | |
4588 | u64 hole_start; | |
4589 | u64 hole_size; | |
4590 | struct extent_buffer *leaf; | |
4591 | struct btrfs_root *log = root->log_root; | |
a0308dd7 NB |
4592 | const u64 ino = btrfs_ino(inode); |
4593 | const u64 i_size = i_size_read(&inode->vfs_inode); | |
a89ca6f2 | 4594 | |
0b246afa | 4595 | if (!btrfs_fs_incompat(fs_info, NO_HOLES)) |
a89ca6f2 FM |
4596 | return 0; |
4597 | ||
4598 | key.objectid = ino; | |
4599 | key.type = BTRFS_EXTENT_DATA_KEY; | |
4600 | key.offset = (u64)-1; | |
4601 | ||
4602 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
4603 | ASSERT(ret != 0); | |
4604 | if (ret < 0) | |
4605 | return ret; | |
4606 | ||
4607 | ASSERT(path->slots[0] > 0); | |
4608 | path->slots[0]--; | |
4609 | leaf = path->nodes[0]; | |
4610 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
4611 | ||
4612 | if (key.objectid != ino || key.type != BTRFS_EXTENT_DATA_KEY) { | |
4613 | /* inode does not have any extents */ | |
4614 | hole_start = 0; | |
4615 | hole_size = i_size; | |
4616 | } else { | |
4617 | struct btrfs_file_extent_item *extent; | |
4618 | u64 len; | |
4619 | ||
4620 | /* | |
4621 | * If there's an extent beyond i_size, an explicit hole was | |
4622 | * already inserted by copy_items(). | |
4623 | */ | |
4624 | if (key.offset >= i_size) | |
4625 | return 0; | |
4626 | ||
4627 | extent = btrfs_item_ptr(leaf, path->slots[0], | |
4628 | struct btrfs_file_extent_item); | |
4629 | ||
4630 | if (btrfs_file_extent_type(leaf, extent) == | |
4631 | BTRFS_FILE_EXTENT_INLINE) { | |
4632 | len = btrfs_file_extent_inline_len(leaf, | |
4633 | path->slots[0], | |
4634 | extent); | |
6399fb5a FM |
4635 | ASSERT(len == i_size || |
4636 | (len == fs_info->sectorsize && | |
4637 | btrfs_file_extent_compression(leaf, extent) != | |
4638 | BTRFS_COMPRESS_NONE)); | |
a89ca6f2 FM |
4639 | return 0; |
4640 | } | |
4641 | ||
4642 | len = btrfs_file_extent_num_bytes(leaf, extent); | |
4643 | /* Last extent goes beyond i_size, no need to log a hole. */ | |
4644 | if (key.offset + len > i_size) | |
4645 | return 0; | |
4646 | hole_start = key.offset + len; | |
4647 | hole_size = i_size - hole_start; | |
4648 | } | |
4649 | btrfs_release_path(path); | |
4650 | ||
4651 | /* Last extent ends at i_size. */ | |
4652 | if (hole_size == 0) | |
4653 | return 0; | |
4654 | ||
0b246afa | 4655 | hole_size = ALIGN(hole_size, fs_info->sectorsize); |
a89ca6f2 FM |
4656 | ret = btrfs_insert_file_extent(trans, log, ino, hole_start, 0, 0, |
4657 | hole_size, 0, hole_size, 0, 0, 0); | |
4658 | return ret; | |
4659 | } | |
4660 | ||
56f23fdb FM |
4661 | /* |
4662 | * When we are logging a new inode X, check if it doesn't have a reference that | |
4663 | * matches the reference from some other inode Y created in a past transaction | |
4664 | * and that was renamed in the current transaction. If we don't do this, then at | |
4665 | * log replay time we can lose inode Y (and all its files if it's a directory): | |
4666 | * | |
4667 | * mkdir /mnt/x | |
4668 | * echo "hello world" > /mnt/x/foobar | |
4669 | * sync | |
4670 | * mv /mnt/x /mnt/y | |
4671 | * mkdir /mnt/x # or touch /mnt/x | |
4672 | * xfs_io -c fsync /mnt/x | |
4673 | * <power fail> | |
4674 | * mount fs, trigger log replay | |
4675 | * | |
4676 | * After the log replay procedure, we would lose the first directory and all its | |
4677 | * files (file foobar). | |
4678 | * For the case where inode Y is not a directory we simply end up losing it: | |
4679 | * | |
4680 | * echo "123" > /mnt/foo | |
4681 | * sync | |
4682 | * mv /mnt/foo /mnt/bar | |
4683 | * echo "abc" > /mnt/foo | |
4684 | * xfs_io -c fsync /mnt/foo | |
4685 | * <power fail> | |
4686 | * | |
4687 | * We also need this for cases where a snapshot entry is replaced by some other | |
4688 | * entry (file or directory) otherwise we end up with an unreplayable log due to | |
4689 | * attempts to delete the snapshot entry (entry of type BTRFS_ROOT_ITEM_KEY) as | |
4690 | * if it were a regular entry: | |
4691 | * | |
4692 | * mkdir /mnt/x | |
4693 | * btrfs subvolume snapshot /mnt /mnt/x/snap | |
4694 | * btrfs subvolume delete /mnt/x/snap | |
4695 | * rmdir /mnt/x | |
4696 | * mkdir /mnt/x | |
4697 | * fsync /mnt/x or fsync some new file inside it | |
4698 | * <power fail> | |
4699 | * | |
4700 | * The snapshot delete, rmdir of x, mkdir of a new x and the fsync all happen in | |
4701 | * the same transaction. | |
4702 | */ | |
4703 | static int btrfs_check_ref_name_override(struct extent_buffer *eb, | |
4704 | const int slot, | |
4705 | const struct btrfs_key *key, | |
4791c8f1 | 4706 | struct btrfs_inode *inode, |
44f714da | 4707 | u64 *other_ino) |
56f23fdb FM |
4708 | { |
4709 | int ret; | |
4710 | struct btrfs_path *search_path; | |
4711 | char *name = NULL; | |
4712 | u32 name_len = 0; | |
4713 | u32 item_size = btrfs_item_size_nr(eb, slot); | |
4714 | u32 cur_offset = 0; | |
4715 | unsigned long ptr = btrfs_item_ptr_offset(eb, slot); | |
4716 | ||
4717 | search_path = btrfs_alloc_path(); | |
4718 | if (!search_path) | |
4719 | return -ENOMEM; | |
4720 | search_path->search_commit_root = 1; | |
4721 | search_path->skip_locking = 1; | |
4722 | ||
4723 | while (cur_offset < item_size) { | |
4724 | u64 parent; | |
4725 | u32 this_name_len; | |
4726 | u32 this_len; | |
4727 | unsigned long name_ptr; | |
4728 | struct btrfs_dir_item *di; | |
4729 | ||
4730 | if (key->type == BTRFS_INODE_REF_KEY) { | |
4731 | struct btrfs_inode_ref *iref; | |
4732 | ||
4733 | iref = (struct btrfs_inode_ref *)(ptr + cur_offset); | |
4734 | parent = key->offset; | |
4735 | this_name_len = btrfs_inode_ref_name_len(eb, iref); | |
4736 | name_ptr = (unsigned long)(iref + 1); | |
4737 | this_len = sizeof(*iref) + this_name_len; | |
4738 | } else { | |
4739 | struct btrfs_inode_extref *extref; | |
4740 | ||
4741 | extref = (struct btrfs_inode_extref *)(ptr + | |
4742 | cur_offset); | |
4743 | parent = btrfs_inode_extref_parent(eb, extref); | |
4744 | this_name_len = btrfs_inode_extref_name_len(eb, extref); | |
4745 | name_ptr = (unsigned long)&extref->name; | |
4746 | this_len = sizeof(*extref) + this_name_len; | |
4747 | } | |
4748 | ||
4749 | if (this_name_len > name_len) { | |
4750 | char *new_name; | |
4751 | ||
4752 | new_name = krealloc(name, this_name_len, GFP_NOFS); | |
4753 | if (!new_name) { | |
4754 | ret = -ENOMEM; | |
4755 | goto out; | |
4756 | } | |
4757 | name_len = this_name_len; | |
4758 | name = new_name; | |
4759 | } | |
4760 | ||
4761 | read_extent_buffer(eb, name, name_ptr, this_name_len); | |
4791c8f1 NB |
4762 | di = btrfs_lookup_dir_item(NULL, inode->root, search_path, |
4763 | parent, name, this_name_len, 0); | |
56f23fdb | 4764 | if (di && !IS_ERR(di)) { |
44f714da FM |
4765 | struct btrfs_key di_key; |
4766 | ||
4767 | btrfs_dir_item_key_to_cpu(search_path->nodes[0], | |
4768 | di, &di_key); | |
4769 | if (di_key.type == BTRFS_INODE_ITEM_KEY) { | |
4770 | ret = 1; | |
4771 | *other_ino = di_key.objectid; | |
4772 | } else { | |
4773 | ret = -EAGAIN; | |
4774 | } | |
56f23fdb FM |
4775 | goto out; |
4776 | } else if (IS_ERR(di)) { | |
4777 | ret = PTR_ERR(di); | |
4778 | goto out; | |
4779 | } | |
4780 | btrfs_release_path(search_path); | |
4781 | ||
4782 | cur_offset += this_len; | |
4783 | } | |
4784 | ret = 0; | |
4785 | out: | |
4786 | btrfs_free_path(search_path); | |
4787 | kfree(name); | |
4788 | return ret; | |
4789 | } | |
4790 | ||
e02119d5 CM |
4791 | /* log a single inode in the tree log. |
4792 | * At least one parent directory for this inode must exist in the tree | |
4793 | * or be logged already. | |
4794 | * | |
4795 | * Any items from this inode changed by the current transaction are copied | |
4796 | * to the log tree. An extra reference is taken on any extents in this | |
4797 | * file, allowing us to avoid a whole pile of corner cases around logging | |
4798 | * blocks that have been removed from the tree. | |
4799 | * | |
4800 | * See LOG_INODE_ALL and related defines for a description of what inode_only | |
4801 | * does. | |
4802 | * | |
4803 | * This handles both files and directories. | |
4804 | */ | |
12fcfd22 | 4805 | static int btrfs_log_inode(struct btrfs_trans_handle *trans, |
a59108a7 | 4806 | struct btrfs_root *root, struct btrfs_inode *inode, |
49dae1bc FM |
4807 | int inode_only, |
4808 | const loff_t start, | |
8407f553 FM |
4809 | const loff_t end, |
4810 | struct btrfs_log_ctx *ctx) | |
e02119d5 | 4811 | { |
0b246afa | 4812 | struct btrfs_fs_info *fs_info = root->fs_info; |
e02119d5 CM |
4813 | struct btrfs_path *path; |
4814 | struct btrfs_path *dst_path; | |
4815 | struct btrfs_key min_key; | |
4816 | struct btrfs_key max_key; | |
4817 | struct btrfs_root *log = root->log_root; | |
827463c4 | 4818 | LIST_HEAD(logged_list); |
16e7549f | 4819 | u64 last_extent = 0; |
4a500fd1 | 4820 | int err = 0; |
e02119d5 | 4821 | int ret; |
3a5f1d45 | 4822 | int nritems; |
31ff1cd2 CM |
4823 | int ins_start_slot = 0; |
4824 | int ins_nr; | |
5dc562c5 | 4825 | bool fast_search = false; |
a59108a7 NB |
4826 | u64 ino = btrfs_ino(inode); |
4827 | struct extent_map_tree *em_tree = &inode->extent_tree; | |
1a4bcf47 | 4828 | u64 logged_isize = 0; |
e4545de5 | 4829 | bool need_log_inode_item = true; |
e02119d5 | 4830 | |
e02119d5 | 4831 | path = btrfs_alloc_path(); |
5df67083 TI |
4832 | if (!path) |
4833 | return -ENOMEM; | |
e02119d5 | 4834 | dst_path = btrfs_alloc_path(); |
5df67083 TI |
4835 | if (!dst_path) { |
4836 | btrfs_free_path(path); | |
4837 | return -ENOMEM; | |
4838 | } | |
e02119d5 | 4839 | |
33345d01 | 4840 | min_key.objectid = ino; |
e02119d5 CM |
4841 | min_key.type = BTRFS_INODE_ITEM_KEY; |
4842 | min_key.offset = 0; | |
4843 | ||
33345d01 | 4844 | max_key.objectid = ino; |
12fcfd22 | 4845 | |
12fcfd22 | 4846 | |
5dc562c5 | 4847 | /* today the code can only do partial logging of directories */ |
a59108a7 | 4848 | if (S_ISDIR(inode->vfs_inode.i_mode) || |
5269b67e | 4849 | (!test_bit(BTRFS_INODE_NEEDS_FULL_SYNC, |
a59108a7 | 4850 | &inode->runtime_flags) && |
781feef7 | 4851 | inode_only >= LOG_INODE_EXISTS)) |
e02119d5 CM |
4852 | max_key.type = BTRFS_XATTR_ITEM_KEY; |
4853 | else | |
4854 | max_key.type = (u8)-1; | |
4855 | max_key.offset = (u64)-1; | |
4856 | ||
2c2c452b FM |
4857 | /* |
4858 | * Only run delayed items if we are a dir or a new file. | |
4859 | * Otherwise commit the delayed inode only, which is needed in | |
4860 | * order for the log replay code to mark inodes for link count | |
4861 | * fixup (create temporary BTRFS_TREE_LOG_FIXUP_OBJECTID items). | |
4862 | */ | |
a59108a7 NB |
4863 | if (S_ISDIR(inode->vfs_inode.i_mode) || |
4864 | inode->generation > fs_info->last_trans_committed) | |
4865 | ret = btrfs_commit_inode_delayed_items(trans, inode); | |
2c2c452b | 4866 | else |
a59108a7 | 4867 | ret = btrfs_commit_inode_delayed_inode(inode); |
2c2c452b FM |
4868 | |
4869 | if (ret) { | |
4870 | btrfs_free_path(path); | |
4871 | btrfs_free_path(dst_path); | |
4872 | return ret; | |
16cdcec7 MX |
4873 | } |
4874 | ||
781feef7 LB |
4875 | if (inode_only == LOG_OTHER_INODE) { |
4876 | inode_only = LOG_INODE_EXISTS; | |
a59108a7 | 4877 | mutex_lock_nested(&inode->log_mutex, SINGLE_DEPTH_NESTING); |
781feef7 | 4878 | } else { |
a59108a7 | 4879 | mutex_lock(&inode->log_mutex); |
781feef7 | 4880 | } |
e02119d5 CM |
4881 | |
4882 | /* | |
4883 | * a brute force approach to making sure we get the most uptodate | |
4884 | * copies of everything. | |
4885 | */ | |
a59108a7 | 4886 | if (S_ISDIR(inode->vfs_inode.i_mode)) { |
e02119d5 CM |
4887 | int max_key_type = BTRFS_DIR_LOG_INDEX_KEY; |
4888 | ||
4f764e51 FM |
4889 | if (inode_only == LOG_INODE_EXISTS) |
4890 | max_key_type = BTRFS_XATTR_ITEM_KEY; | |
33345d01 | 4891 | ret = drop_objectid_items(trans, log, path, ino, max_key_type); |
e02119d5 | 4892 | } else { |
1a4bcf47 FM |
4893 | if (inode_only == LOG_INODE_EXISTS) { |
4894 | /* | |
4895 | * Make sure the new inode item we write to the log has | |
4896 | * the same isize as the current one (if it exists). | |
4897 | * This is necessary to prevent data loss after log | |
4898 | * replay, and also to prevent doing a wrong expanding | |
4899 | * truncate - for e.g. create file, write 4K into offset | |
4900 | * 0, fsync, write 4K into offset 4096, add hard link, | |
4901 | * fsync some other file (to sync log), power fail - if | |
4902 | * we use the inode's current i_size, after log replay | |
4903 | * we get a 8Kb file, with the last 4Kb extent as a hole | |
4904 | * (zeroes), as if an expanding truncate happened, | |
4905 | * instead of getting a file of 4Kb only. | |
4906 | */ | |
a59108a7 | 4907 | err = logged_inode_size(log, inode, path, &logged_isize); |
1a4bcf47 FM |
4908 | if (err) |
4909 | goto out_unlock; | |
4910 | } | |
a742994a | 4911 | if (test_bit(BTRFS_INODE_NEEDS_FULL_SYNC, |
a59108a7 | 4912 | &inode->runtime_flags)) { |
a742994a | 4913 | if (inode_only == LOG_INODE_EXISTS) { |
4f764e51 | 4914 | max_key.type = BTRFS_XATTR_ITEM_KEY; |
a742994a FM |
4915 | ret = drop_objectid_items(trans, log, path, ino, |
4916 | max_key.type); | |
4917 | } else { | |
4918 | clear_bit(BTRFS_INODE_NEEDS_FULL_SYNC, | |
a59108a7 | 4919 | &inode->runtime_flags); |
a742994a | 4920 | clear_bit(BTRFS_INODE_COPY_EVERYTHING, |
a59108a7 | 4921 | &inode->runtime_flags); |
28ed1345 CM |
4922 | while(1) { |
4923 | ret = btrfs_truncate_inode_items(trans, | |
a59108a7 | 4924 | log, &inode->vfs_inode, 0, 0); |
28ed1345 CM |
4925 | if (ret != -EAGAIN) |
4926 | break; | |
4927 | } | |
a742994a | 4928 | } |
4f764e51 | 4929 | } else if (test_and_clear_bit(BTRFS_INODE_COPY_EVERYTHING, |
a59108a7 | 4930 | &inode->runtime_flags) || |
6cfab851 | 4931 | inode_only == LOG_INODE_EXISTS) { |
4f764e51 | 4932 | if (inode_only == LOG_INODE_ALL) |
183f37fa | 4933 | fast_search = true; |
4f764e51 | 4934 | max_key.type = BTRFS_XATTR_ITEM_KEY; |
5dc562c5 | 4935 | ret = drop_objectid_items(trans, log, path, ino, |
e9976151 | 4936 | max_key.type); |
a95249b3 JB |
4937 | } else { |
4938 | if (inode_only == LOG_INODE_ALL) | |
4939 | fast_search = true; | |
a95249b3 | 4940 | goto log_extents; |
5dc562c5 | 4941 | } |
a95249b3 | 4942 | |
e02119d5 | 4943 | } |
4a500fd1 YZ |
4944 | if (ret) { |
4945 | err = ret; | |
4946 | goto out_unlock; | |
4947 | } | |
e02119d5 | 4948 | |
d397712b | 4949 | while (1) { |
31ff1cd2 | 4950 | ins_nr = 0; |
6174d3cb | 4951 | ret = btrfs_search_forward(root, &min_key, |
de78b51a | 4952 | path, trans->transid); |
fb770ae4 LB |
4953 | if (ret < 0) { |
4954 | err = ret; | |
4955 | goto out_unlock; | |
4956 | } | |
e02119d5 CM |
4957 | if (ret != 0) |
4958 | break; | |
3a5f1d45 | 4959 | again: |
31ff1cd2 | 4960 | /* note, ins_nr might be > 0 here, cleanup outside the loop */ |
33345d01 | 4961 | if (min_key.objectid != ino) |
e02119d5 CM |
4962 | break; |
4963 | if (min_key.type > max_key.type) | |
4964 | break; | |
31ff1cd2 | 4965 | |
e4545de5 FM |
4966 | if (min_key.type == BTRFS_INODE_ITEM_KEY) |
4967 | need_log_inode_item = false; | |
4968 | ||
56f23fdb FM |
4969 | if ((min_key.type == BTRFS_INODE_REF_KEY || |
4970 | min_key.type == BTRFS_INODE_EXTREF_KEY) && | |
a59108a7 | 4971 | inode->generation == trans->transid) { |
44f714da FM |
4972 | u64 other_ino = 0; |
4973 | ||
56f23fdb | 4974 | ret = btrfs_check_ref_name_override(path->nodes[0], |
a59108a7 NB |
4975 | path->slots[0], &min_key, inode, |
4976 | &other_ino); | |
56f23fdb FM |
4977 | if (ret < 0) { |
4978 | err = ret; | |
4979 | goto out_unlock; | |
28a23593 | 4980 | } else if (ret > 0 && ctx && |
4a0cc7ca | 4981 | other_ino != btrfs_ino(BTRFS_I(ctx->inode))) { |
44f714da FM |
4982 | struct btrfs_key inode_key; |
4983 | struct inode *other_inode; | |
4984 | ||
4985 | if (ins_nr > 0) { | |
4986 | ins_nr++; | |
4987 | } else { | |
4988 | ins_nr = 1; | |
4989 | ins_start_slot = path->slots[0]; | |
4990 | } | |
a59108a7 | 4991 | ret = copy_items(trans, inode, dst_path, path, |
44f714da FM |
4992 | &last_extent, ins_start_slot, |
4993 | ins_nr, inode_only, | |
4994 | logged_isize); | |
4995 | if (ret < 0) { | |
4996 | err = ret; | |
4997 | goto out_unlock; | |
4998 | } | |
4999 | ins_nr = 0; | |
5000 | btrfs_release_path(path); | |
5001 | inode_key.objectid = other_ino; | |
5002 | inode_key.type = BTRFS_INODE_ITEM_KEY; | |
5003 | inode_key.offset = 0; | |
0b246afa | 5004 | other_inode = btrfs_iget(fs_info->sb, |
44f714da FM |
5005 | &inode_key, root, |
5006 | NULL); | |
5007 | /* | |
5008 | * If the other inode that had a conflicting dir | |
5009 | * entry was deleted in the current transaction, | |
5010 | * we don't need to do more work nor fallback to | |
5011 | * a transaction commit. | |
5012 | */ | |
5013 | if (IS_ERR(other_inode) && | |
5014 | PTR_ERR(other_inode) == -ENOENT) { | |
5015 | goto next_key; | |
5016 | } else if (IS_ERR(other_inode)) { | |
5017 | err = PTR_ERR(other_inode); | |
5018 | goto out_unlock; | |
5019 | } | |
5020 | /* | |
5021 | * We are safe logging the other inode without | |
5022 | * acquiring its i_mutex as long as we log with | |
5023 | * the LOG_INODE_EXISTS mode. We're safe against | |
5024 | * concurrent renames of the other inode as well | |
5025 | * because during a rename we pin the log and | |
5026 | * update the log with the new name before we | |
5027 | * unpin it. | |
5028 | */ | |
a59108a7 NB |
5029 | err = btrfs_log_inode(trans, root, |
5030 | BTRFS_I(other_inode), | |
5031 | LOG_OTHER_INODE, 0, LLONG_MAX, | |
5032 | ctx); | |
44f714da FM |
5033 | iput(other_inode); |
5034 | if (err) | |
5035 | goto out_unlock; | |
5036 | else | |
5037 | goto next_key; | |
56f23fdb FM |
5038 | } |
5039 | } | |
5040 | ||
36283bf7 FM |
5041 | /* Skip xattrs, we log them later with btrfs_log_all_xattrs() */ |
5042 | if (min_key.type == BTRFS_XATTR_ITEM_KEY) { | |
5043 | if (ins_nr == 0) | |
5044 | goto next_slot; | |
a59108a7 | 5045 | ret = copy_items(trans, inode, dst_path, path, |
36283bf7 FM |
5046 | &last_extent, ins_start_slot, |
5047 | ins_nr, inode_only, logged_isize); | |
5048 | if (ret < 0) { | |
5049 | err = ret; | |
5050 | goto out_unlock; | |
5051 | } | |
5052 | ins_nr = 0; | |
5053 | if (ret) { | |
5054 | btrfs_release_path(path); | |
5055 | continue; | |
5056 | } | |
5057 | goto next_slot; | |
5058 | } | |
5059 | ||
31ff1cd2 CM |
5060 | if (ins_nr && ins_start_slot + ins_nr == path->slots[0]) { |
5061 | ins_nr++; | |
5062 | goto next_slot; | |
5063 | } else if (!ins_nr) { | |
5064 | ins_start_slot = path->slots[0]; | |
5065 | ins_nr = 1; | |
5066 | goto next_slot; | |
e02119d5 CM |
5067 | } |
5068 | ||
a59108a7 | 5069 | ret = copy_items(trans, inode, dst_path, path, &last_extent, |
1a4bcf47 FM |
5070 | ins_start_slot, ins_nr, inode_only, |
5071 | logged_isize); | |
16e7549f | 5072 | if (ret < 0) { |
4a500fd1 YZ |
5073 | err = ret; |
5074 | goto out_unlock; | |
a71db86e RV |
5075 | } |
5076 | if (ret) { | |
16e7549f JB |
5077 | ins_nr = 0; |
5078 | btrfs_release_path(path); | |
5079 | continue; | |
4a500fd1 | 5080 | } |
31ff1cd2 CM |
5081 | ins_nr = 1; |
5082 | ins_start_slot = path->slots[0]; | |
5083 | next_slot: | |
e02119d5 | 5084 | |
3a5f1d45 CM |
5085 | nritems = btrfs_header_nritems(path->nodes[0]); |
5086 | path->slots[0]++; | |
5087 | if (path->slots[0] < nritems) { | |
5088 | btrfs_item_key_to_cpu(path->nodes[0], &min_key, | |
5089 | path->slots[0]); | |
5090 | goto again; | |
5091 | } | |
31ff1cd2 | 5092 | if (ins_nr) { |
a59108a7 | 5093 | ret = copy_items(trans, inode, dst_path, path, |
16e7549f | 5094 | &last_extent, ins_start_slot, |
1a4bcf47 | 5095 | ins_nr, inode_only, logged_isize); |
16e7549f | 5096 | if (ret < 0) { |
4a500fd1 YZ |
5097 | err = ret; |
5098 | goto out_unlock; | |
5099 | } | |
16e7549f | 5100 | ret = 0; |
31ff1cd2 CM |
5101 | ins_nr = 0; |
5102 | } | |
b3b4aa74 | 5103 | btrfs_release_path(path); |
44f714da | 5104 | next_key: |
3d41d702 | 5105 | if (min_key.offset < (u64)-1) { |
e02119d5 | 5106 | min_key.offset++; |
3d41d702 | 5107 | } else if (min_key.type < max_key.type) { |
e02119d5 | 5108 | min_key.type++; |
3d41d702 FDBM |
5109 | min_key.offset = 0; |
5110 | } else { | |
e02119d5 | 5111 | break; |
3d41d702 | 5112 | } |
e02119d5 | 5113 | } |
31ff1cd2 | 5114 | if (ins_nr) { |
a59108a7 | 5115 | ret = copy_items(trans, inode, dst_path, path, &last_extent, |
1a4bcf47 FM |
5116 | ins_start_slot, ins_nr, inode_only, |
5117 | logged_isize); | |
16e7549f | 5118 | if (ret < 0) { |
4a500fd1 YZ |
5119 | err = ret; |
5120 | goto out_unlock; | |
5121 | } | |
16e7549f | 5122 | ret = 0; |
31ff1cd2 CM |
5123 | ins_nr = 0; |
5124 | } | |
5dc562c5 | 5125 | |
36283bf7 FM |
5126 | btrfs_release_path(path); |
5127 | btrfs_release_path(dst_path); | |
a59108a7 | 5128 | err = btrfs_log_all_xattrs(trans, root, inode, path, dst_path); |
36283bf7 FM |
5129 | if (err) |
5130 | goto out_unlock; | |
a89ca6f2 FM |
5131 | if (max_key.type >= BTRFS_EXTENT_DATA_KEY && !fast_search) { |
5132 | btrfs_release_path(path); | |
5133 | btrfs_release_path(dst_path); | |
a59108a7 | 5134 | err = btrfs_log_trailing_hole(trans, root, inode, path); |
a89ca6f2 FM |
5135 | if (err) |
5136 | goto out_unlock; | |
5137 | } | |
a95249b3 | 5138 | log_extents: |
f3b15ccd JB |
5139 | btrfs_release_path(path); |
5140 | btrfs_release_path(dst_path); | |
e4545de5 | 5141 | if (need_log_inode_item) { |
a59108a7 | 5142 | err = log_inode_item(trans, log, dst_path, inode); |
e4545de5 FM |
5143 | if (err) |
5144 | goto out_unlock; | |
5145 | } | |
5dc562c5 | 5146 | if (fast_search) { |
a59108a7 | 5147 | ret = btrfs_log_changed_extents(trans, root, inode, dst_path, |
de0ee0ed | 5148 | &logged_list, ctx, start, end); |
5dc562c5 JB |
5149 | if (ret) { |
5150 | err = ret; | |
5151 | goto out_unlock; | |
5152 | } | |
d006a048 | 5153 | } else if (inode_only == LOG_INODE_ALL) { |
06d3d22b LB |
5154 | struct extent_map *em, *n; |
5155 | ||
49dae1bc FM |
5156 | write_lock(&em_tree->lock); |
5157 | /* | |
5158 | * We can't just remove every em if we're called for a ranged | |
5159 | * fsync - that is, one that doesn't cover the whole possible | |
5160 | * file range (0 to LLONG_MAX). This is because we can have | |
5161 | * em's that fall outside the range we're logging and therefore | |
5162 | * their ordered operations haven't completed yet | |
5163 | * (btrfs_finish_ordered_io() not invoked yet). This means we | |
5164 | * didn't get their respective file extent item in the fs/subvol | |
5165 | * tree yet, and need to let the next fast fsync (one which | |
5166 | * consults the list of modified extent maps) find the em so | |
5167 | * that it logs a matching file extent item and waits for the | |
5168 | * respective ordered operation to complete (if it's still | |
5169 | * running). | |
5170 | * | |
5171 | * Removing every em outside the range we're logging would make | |
5172 | * the next fast fsync not log their matching file extent items, | |
5173 | * therefore making us lose data after a log replay. | |
5174 | */ | |
5175 | list_for_each_entry_safe(em, n, &em_tree->modified_extents, | |
5176 | list) { | |
5177 | const u64 mod_end = em->mod_start + em->mod_len - 1; | |
5178 | ||
5179 | if (em->mod_start >= start && mod_end <= end) | |
5180 | list_del_init(&em->list); | |
5181 | } | |
5182 | write_unlock(&em_tree->lock); | |
5dc562c5 JB |
5183 | } |
5184 | ||
a59108a7 NB |
5185 | if (inode_only == LOG_INODE_ALL && S_ISDIR(inode->vfs_inode.i_mode)) { |
5186 | ret = log_directory_changes(trans, root, inode, path, dst_path, | |
5187 | ctx); | |
4a500fd1 YZ |
5188 | if (ret) { |
5189 | err = ret; | |
5190 | goto out_unlock; | |
5191 | } | |
e02119d5 | 5192 | } |
49dae1bc | 5193 | |
a59108a7 NB |
5194 | spin_lock(&inode->lock); |
5195 | inode->logged_trans = trans->transid; | |
5196 | inode->last_log_commit = inode->last_sub_trans; | |
5197 | spin_unlock(&inode->lock); | |
4a500fd1 | 5198 | out_unlock: |
827463c4 MX |
5199 | if (unlikely(err)) |
5200 | btrfs_put_logged_extents(&logged_list); | |
5201 | else | |
5202 | btrfs_submit_logged_extents(&logged_list, log); | |
a59108a7 | 5203 | mutex_unlock(&inode->log_mutex); |
e02119d5 CM |
5204 | |
5205 | btrfs_free_path(path); | |
5206 | btrfs_free_path(dst_path); | |
4a500fd1 | 5207 | return err; |
e02119d5 CM |
5208 | } |
5209 | ||
2be63d5c FM |
5210 | /* |
5211 | * Check if we must fallback to a transaction commit when logging an inode. | |
5212 | * This must be called after logging the inode and is used only in the context | |
5213 | * when fsyncing an inode requires the need to log some other inode - in which | |
5214 | * case we can't lock the i_mutex of each other inode we need to log as that | |
5215 | * can lead to deadlocks with concurrent fsync against other inodes (as we can | |
5216 | * log inodes up or down in the hierarchy) or rename operations for example. So | |
5217 | * we take the log_mutex of the inode after we have logged it and then check for | |
5218 | * its last_unlink_trans value - this is safe because any task setting | |
5219 | * last_unlink_trans must take the log_mutex and it must do this before it does | |
5220 | * the actual unlink operation, so if we do this check before a concurrent task | |
5221 | * sets last_unlink_trans it means we've logged a consistent version/state of | |
5222 | * all the inode items, otherwise we are not sure and must do a transaction | |
01327610 | 5223 | * commit (the concurrent task might have only updated last_unlink_trans before |
2be63d5c FM |
5224 | * we logged the inode or it might have also done the unlink). |
5225 | */ | |
5226 | static bool btrfs_must_commit_transaction(struct btrfs_trans_handle *trans, | |
ab1717b2 | 5227 | struct btrfs_inode *inode) |
2be63d5c | 5228 | { |
ab1717b2 | 5229 | struct btrfs_fs_info *fs_info = inode->root->fs_info; |
2be63d5c FM |
5230 | bool ret = false; |
5231 | ||
ab1717b2 NB |
5232 | mutex_lock(&inode->log_mutex); |
5233 | if (inode->last_unlink_trans > fs_info->last_trans_committed) { | |
2be63d5c FM |
5234 | /* |
5235 | * Make sure any commits to the log are forced to be full | |
5236 | * commits. | |
5237 | */ | |
5238 | btrfs_set_log_full_commit(fs_info, trans); | |
5239 | ret = true; | |
5240 | } | |
ab1717b2 | 5241 | mutex_unlock(&inode->log_mutex); |
2be63d5c FM |
5242 | |
5243 | return ret; | |
5244 | } | |
5245 | ||
12fcfd22 CM |
5246 | /* |
5247 | * follow the dentry parent pointers up the chain and see if any | |
5248 | * of the directories in it require a full commit before they can | |
5249 | * be logged. Returns zero if nothing special needs to be done or 1 if | |
5250 | * a full commit is required. | |
5251 | */ | |
5252 | static noinline int check_parent_dirs_for_sync(struct btrfs_trans_handle *trans, | |
aefa6115 | 5253 | struct btrfs_inode *inode, |
12fcfd22 CM |
5254 | struct dentry *parent, |
5255 | struct super_block *sb, | |
5256 | u64 last_committed) | |
e02119d5 | 5257 | { |
12fcfd22 | 5258 | int ret = 0; |
6a912213 | 5259 | struct dentry *old_parent = NULL; |
aefa6115 | 5260 | struct btrfs_inode *orig_inode = inode; |
e02119d5 | 5261 | |
af4176b4 CM |
5262 | /* |
5263 | * for regular files, if its inode is already on disk, we don't | |
5264 | * have to worry about the parents at all. This is because | |
5265 | * we can use the last_unlink_trans field to record renames | |
5266 | * and other fun in this file. | |
5267 | */ | |
aefa6115 NB |
5268 | if (S_ISREG(inode->vfs_inode.i_mode) && |
5269 | inode->generation <= last_committed && | |
5270 | inode->last_unlink_trans <= last_committed) | |
5271 | goto out; | |
af4176b4 | 5272 | |
aefa6115 | 5273 | if (!S_ISDIR(inode->vfs_inode.i_mode)) { |
fc64005c | 5274 | if (!parent || d_really_is_negative(parent) || sb != parent->d_sb) |
12fcfd22 | 5275 | goto out; |
aefa6115 | 5276 | inode = BTRFS_I(d_inode(parent)); |
12fcfd22 CM |
5277 | } |
5278 | ||
5279 | while (1) { | |
de2b530b JB |
5280 | /* |
5281 | * If we are logging a directory then we start with our inode, | |
01327610 | 5282 | * not our parent's inode, so we need to skip setting the |
de2b530b JB |
5283 | * logged_trans so that further down in the log code we don't |
5284 | * think this inode has already been logged. | |
5285 | */ | |
5286 | if (inode != orig_inode) | |
aefa6115 | 5287 | inode->logged_trans = trans->transid; |
12fcfd22 CM |
5288 | smp_mb(); |
5289 | ||
aefa6115 | 5290 | if (btrfs_must_commit_transaction(trans, inode)) { |
12fcfd22 CM |
5291 | ret = 1; |
5292 | break; | |
5293 | } | |
5294 | ||
fc64005c | 5295 | if (!parent || d_really_is_negative(parent) || sb != parent->d_sb) |
12fcfd22 CM |
5296 | break; |
5297 | ||
44f714da | 5298 | if (IS_ROOT(parent)) { |
aefa6115 NB |
5299 | inode = BTRFS_I(d_inode(parent)); |
5300 | if (btrfs_must_commit_transaction(trans, inode)) | |
44f714da | 5301 | ret = 1; |
12fcfd22 | 5302 | break; |
44f714da | 5303 | } |
12fcfd22 | 5304 | |
6a912213 JB |
5305 | parent = dget_parent(parent); |
5306 | dput(old_parent); | |
5307 | old_parent = parent; | |
aefa6115 | 5308 | inode = BTRFS_I(d_inode(parent)); |
12fcfd22 CM |
5309 | |
5310 | } | |
6a912213 | 5311 | dput(old_parent); |
12fcfd22 | 5312 | out: |
e02119d5 CM |
5313 | return ret; |
5314 | } | |
5315 | ||
2f2ff0ee FM |
5316 | struct btrfs_dir_list { |
5317 | u64 ino; | |
5318 | struct list_head list; | |
5319 | }; | |
5320 | ||
5321 | /* | |
5322 | * Log the inodes of the new dentries of a directory. See log_dir_items() for | |
5323 | * details about the why it is needed. | |
5324 | * This is a recursive operation - if an existing dentry corresponds to a | |
5325 | * directory, that directory's new entries are logged too (same behaviour as | |
5326 | * ext3/4, xfs, f2fs, reiserfs, nilfs2). Note that when logging the inodes | |
5327 | * the dentries point to we do not lock their i_mutex, otherwise lockdep | |
5328 | * complains about the following circular lock dependency / possible deadlock: | |
5329 | * | |
5330 | * CPU0 CPU1 | |
5331 | * ---- ---- | |
5332 | * lock(&type->i_mutex_dir_key#3/2); | |
5333 | * lock(sb_internal#2); | |
5334 | * lock(&type->i_mutex_dir_key#3/2); | |
5335 | * lock(&sb->s_type->i_mutex_key#14); | |
5336 | * | |
5337 | * Where sb_internal is the lock (a counter that works as a lock) acquired by | |
5338 | * sb_start_intwrite() in btrfs_start_transaction(). | |
5339 | * Not locking i_mutex of the inodes is still safe because: | |
5340 | * | |
5341 | * 1) For regular files we log with a mode of LOG_INODE_EXISTS. It's possible | |
5342 | * that while logging the inode new references (names) are added or removed | |
5343 | * from the inode, leaving the logged inode item with a link count that does | |
5344 | * not match the number of logged inode reference items. This is fine because | |
5345 | * at log replay time we compute the real number of links and correct the | |
5346 | * link count in the inode item (see replay_one_buffer() and | |
5347 | * link_to_fixup_dir()); | |
5348 | * | |
5349 | * 2) For directories we log with a mode of LOG_INODE_ALL. It's possible that | |
5350 | * while logging the inode's items new items with keys BTRFS_DIR_ITEM_KEY and | |
5351 | * BTRFS_DIR_INDEX_KEY are added to fs/subvol tree and the logged inode item | |
5352 | * has a size that doesn't match the sum of the lengths of all the logged | |
5353 | * names. This does not result in a problem because if a dir_item key is | |
5354 | * logged but its matching dir_index key is not logged, at log replay time we | |
5355 | * don't use it to replay the respective name (see replay_one_name()). On the | |
5356 | * other hand if only the dir_index key ends up being logged, the respective | |
5357 | * name is added to the fs/subvol tree with both the dir_item and dir_index | |
5358 | * keys created (see replay_one_name()). | |
5359 | * The directory's inode item with a wrong i_size is not a problem as well, | |
5360 | * since we don't use it at log replay time to set the i_size in the inode | |
5361 | * item of the fs/subvol tree (see overwrite_item()). | |
5362 | */ | |
5363 | static int log_new_dir_dentries(struct btrfs_trans_handle *trans, | |
5364 | struct btrfs_root *root, | |
51cc0d32 | 5365 | struct btrfs_inode *start_inode, |
2f2ff0ee FM |
5366 | struct btrfs_log_ctx *ctx) |
5367 | { | |
0b246afa | 5368 | struct btrfs_fs_info *fs_info = root->fs_info; |
2f2ff0ee FM |
5369 | struct btrfs_root *log = root->log_root; |
5370 | struct btrfs_path *path; | |
5371 | LIST_HEAD(dir_list); | |
5372 | struct btrfs_dir_list *dir_elem; | |
5373 | int ret = 0; | |
5374 | ||
5375 | path = btrfs_alloc_path(); | |
5376 | if (!path) | |
5377 | return -ENOMEM; | |
5378 | ||
5379 | dir_elem = kmalloc(sizeof(*dir_elem), GFP_NOFS); | |
5380 | if (!dir_elem) { | |
5381 | btrfs_free_path(path); | |
5382 | return -ENOMEM; | |
5383 | } | |
51cc0d32 | 5384 | dir_elem->ino = btrfs_ino(start_inode); |
2f2ff0ee FM |
5385 | list_add_tail(&dir_elem->list, &dir_list); |
5386 | ||
5387 | while (!list_empty(&dir_list)) { | |
5388 | struct extent_buffer *leaf; | |
5389 | struct btrfs_key min_key; | |
5390 | int nritems; | |
5391 | int i; | |
5392 | ||
5393 | dir_elem = list_first_entry(&dir_list, struct btrfs_dir_list, | |
5394 | list); | |
5395 | if (ret) | |
5396 | goto next_dir_inode; | |
5397 | ||
5398 | min_key.objectid = dir_elem->ino; | |
5399 | min_key.type = BTRFS_DIR_ITEM_KEY; | |
5400 | min_key.offset = 0; | |
5401 | again: | |
5402 | btrfs_release_path(path); | |
5403 | ret = btrfs_search_forward(log, &min_key, path, trans->transid); | |
5404 | if (ret < 0) { | |
5405 | goto next_dir_inode; | |
5406 | } else if (ret > 0) { | |
5407 | ret = 0; | |
5408 | goto next_dir_inode; | |
5409 | } | |
5410 | ||
5411 | process_leaf: | |
5412 | leaf = path->nodes[0]; | |
5413 | nritems = btrfs_header_nritems(leaf); | |
5414 | for (i = path->slots[0]; i < nritems; i++) { | |
5415 | struct btrfs_dir_item *di; | |
5416 | struct btrfs_key di_key; | |
5417 | struct inode *di_inode; | |
5418 | struct btrfs_dir_list *new_dir_elem; | |
5419 | int log_mode = LOG_INODE_EXISTS; | |
5420 | int type; | |
5421 | ||
5422 | btrfs_item_key_to_cpu(leaf, &min_key, i); | |
5423 | if (min_key.objectid != dir_elem->ino || | |
5424 | min_key.type != BTRFS_DIR_ITEM_KEY) | |
5425 | goto next_dir_inode; | |
5426 | ||
5427 | di = btrfs_item_ptr(leaf, i, struct btrfs_dir_item); | |
5428 | type = btrfs_dir_type(leaf, di); | |
5429 | if (btrfs_dir_transid(leaf, di) < trans->transid && | |
5430 | type != BTRFS_FT_DIR) | |
5431 | continue; | |
5432 | btrfs_dir_item_key_to_cpu(leaf, di, &di_key); | |
5433 | if (di_key.type == BTRFS_ROOT_ITEM_KEY) | |
5434 | continue; | |
5435 | ||
ec125cfb | 5436 | btrfs_release_path(path); |
0b246afa | 5437 | di_inode = btrfs_iget(fs_info->sb, &di_key, root, NULL); |
2f2ff0ee FM |
5438 | if (IS_ERR(di_inode)) { |
5439 | ret = PTR_ERR(di_inode); | |
5440 | goto next_dir_inode; | |
5441 | } | |
5442 | ||
0f8939b8 | 5443 | if (btrfs_inode_in_log(BTRFS_I(di_inode), trans->transid)) { |
2f2ff0ee | 5444 | iput(di_inode); |
ec125cfb | 5445 | break; |
2f2ff0ee FM |
5446 | } |
5447 | ||
5448 | ctx->log_new_dentries = false; | |
3f9749f6 | 5449 | if (type == BTRFS_FT_DIR || type == BTRFS_FT_SYMLINK) |
2f2ff0ee | 5450 | log_mode = LOG_INODE_ALL; |
a59108a7 | 5451 | ret = btrfs_log_inode(trans, root, BTRFS_I(di_inode), |
2f2ff0ee | 5452 | log_mode, 0, LLONG_MAX, ctx); |
2be63d5c | 5453 | if (!ret && |
ab1717b2 | 5454 | btrfs_must_commit_transaction(trans, BTRFS_I(di_inode))) |
2be63d5c | 5455 | ret = 1; |
2f2ff0ee FM |
5456 | iput(di_inode); |
5457 | if (ret) | |
5458 | goto next_dir_inode; | |
5459 | if (ctx->log_new_dentries) { | |
5460 | new_dir_elem = kmalloc(sizeof(*new_dir_elem), | |
5461 | GFP_NOFS); | |
5462 | if (!new_dir_elem) { | |
5463 | ret = -ENOMEM; | |
5464 | goto next_dir_inode; | |
5465 | } | |
5466 | new_dir_elem->ino = di_key.objectid; | |
5467 | list_add_tail(&new_dir_elem->list, &dir_list); | |
5468 | } | |
5469 | break; | |
5470 | } | |
5471 | if (i == nritems) { | |
5472 | ret = btrfs_next_leaf(log, path); | |
5473 | if (ret < 0) { | |
5474 | goto next_dir_inode; | |
5475 | } else if (ret > 0) { | |
5476 | ret = 0; | |
5477 | goto next_dir_inode; | |
5478 | } | |
5479 | goto process_leaf; | |
5480 | } | |
5481 | if (min_key.offset < (u64)-1) { | |
5482 | min_key.offset++; | |
5483 | goto again; | |
5484 | } | |
5485 | next_dir_inode: | |
5486 | list_del(&dir_elem->list); | |
5487 | kfree(dir_elem); | |
5488 | } | |
5489 | ||
5490 | btrfs_free_path(path); | |
5491 | return ret; | |
5492 | } | |
5493 | ||
18aa0922 | 5494 | static int btrfs_log_all_parents(struct btrfs_trans_handle *trans, |
d0a0b78d | 5495 | struct btrfs_inode *inode, |
18aa0922 FM |
5496 | struct btrfs_log_ctx *ctx) |
5497 | { | |
d0a0b78d | 5498 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb); |
18aa0922 FM |
5499 | int ret; |
5500 | struct btrfs_path *path; | |
5501 | struct btrfs_key key; | |
d0a0b78d NB |
5502 | struct btrfs_root *root = inode->root; |
5503 | const u64 ino = btrfs_ino(inode); | |
18aa0922 FM |
5504 | |
5505 | path = btrfs_alloc_path(); | |
5506 | if (!path) | |
5507 | return -ENOMEM; | |
5508 | path->skip_locking = 1; | |
5509 | path->search_commit_root = 1; | |
5510 | ||
5511 | key.objectid = ino; | |
5512 | key.type = BTRFS_INODE_REF_KEY; | |
5513 | key.offset = 0; | |
5514 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
5515 | if (ret < 0) | |
5516 | goto out; | |
5517 | ||
5518 | while (true) { | |
5519 | struct extent_buffer *leaf = path->nodes[0]; | |
5520 | int slot = path->slots[0]; | |
5521 | u32 cur_offset = 0; | |
5522 | u32 item_size; | |
5523 | unsigned long ptr; | |
5524 | ||
5525 | if (slot >= btrfs_header_nritems(leaf)) { | |
5526 | ret = btrfs_next_leaf(root, path); | |
5527 | if (ret < 0) | |
5528 | goto out; | |
5529 | else if (ret > 0) | |
5530 | break; | |
5531 | continue; | |
5532 | } | |
5533 | ||
5534 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
5535 | /* BTRFS_INODE_EXTREF_KEY is BTRFS_INODE_REF_KEY + 1 */ | |
5536 | if (key.objectid != ino || key.type > BTRFS_INODE_EXTREF_KEY) | |
5537 | break; | |
5538 | ||
5539 | item_size = btrfs_item_size_nr(leaf, slot); | |
5540 | ptr = btrfs_item_ptr_offset(leaf, slot); | |
5541 | while (cur_offset < item_size) { | |
5542 | struct btrfs_key inode_key; | |
5543 | struct inode *dir_inode; | |
5544 | ||
5545 | inode_key.type = BTRFS_INODE_ITEM_KEY; | |
5546 | inode_key.offset = 0; | |
5547 | ||
5548 | if (key.type == BTRFS_INODE_EXTREF_KEY) { | |
5549 | struct btrfs_inode_extref *extref; | |
5550 | ||
5551 | extref = (struct btrfs_inode_extref *) | |
5552 | (ptr + cur_offset); | |
5553 | inode_key.objectid = btrfs_inode_extref_parent( | |
5554 | leaf, extref); | |
5555 | cur_offset += sizeof(*extref); | |
5556 | cur_offset += btrfs_inode_extref_name_len(leaf, | |
5557 | extref); | |
5558 | } else { | |
5559 | inode_key.objectid = key.offset; | |
5560 | cur_offset = item_size; | |
5561 | } | |
5562 | ||
0b246afa | 5563 | dir_inode = btrfs_iget(fs_info->sb, &inode_key, |
18aa0922 FM |
5564 | root, NULL); |
5565 | /* If parent inode was deleted, skip it. */ | |
5566 | if (IS_ERR(dir_inode)) | |
5567 | continue; | |
5568 | ||
657ed1aa FM |
5569 | if (ctx) |
5570 | ctx->log_new_dentries = false; | |
a59108a7 | 5571 | ret = btrfs_log_inode(trans, root, BTRFS_I(dir_inode), |
18aa0922 | 5572 | LOG_INODE_ALL, 0, LLONG_MAX, ctx); |
2be63d5c | 5573 | if (!ret && |
ab1717b2 | 5574 | btrfs_must_commit_transaction(trans, BTRFS_I(dir_inode))) |
2be63d5c | 5575 | ret = 1; |
657ed1aa FM |
5576 | if (!ret && ctx && ctx->log_new_dentries) |
5577 | ret = log_new_dir_dentries(trans, root, | |
f85b7379 | 5578 | BTRFS_I(dir_inode), ctx); |
18aa0922 FM |
5579 | iput(dir_inode); |
5580 | if (ret) | |
5581 | goto out; | |
5582 | } | |
5583 | path->slots[0]++; | |
5584 | } | |
5585 | ret = 0; | |
5586 | out: | |
5587 | btrfs_free_path(path); | |
5588 | return ret; | |
5589 | } | |
5590 | ||
e02119d5 CM |
5591 | /* |
5592 | * helper function around btrfs_log_inode to make sure newly created | |
5593 | * parent directories also end up in the log. A minimal inode and backref | |
5594 | * only logging is done of any parent directories that are older than | |
5595 | * the last committed transaction | |
5596 | */ | |
48a3b636 | 5597 | static int btrfs_log_inode_parent(struct btrfs_trans_handle *trans, |
19df27a9 | 5598 | struct btrfs_inode *inode, |
49dae1bc FM |
5599 | struct dentry *parent, |
5600 | const loff_t start, | |
5601 | const loff_t end, | |
41a1eada | 5602 | int inode_only, |
8b050d35 | 5603 | struct btrfs_log_ctx *ctx) |
e02119d5 | 5604 | { |
f882274b | 5605 | struct btrfs_root *root = inode->root; |
0b246afa | 5606 | struct btrfs_fs_info *fs_info = root->fs_info; |
e02119d5 | 5607 | struct super_block *sb; |
6a912213 | 5608 | struct dentry *old_parent = NULL; |
12fcfd22 | 5609 | int ret = 0; |
0b246afa | 5610 | u64 last_committed = fs_info->last_trans_committed; |
2f2ff0ee | 5611 | bool log_dentries = false; |
19df27a9 | 5612 | struct btrfs_inode *orig_inode = inode; |
12fcfd22 | 5613 | |
19df27a9 | 5614 | sb = inode->vfs_inode.i_sb; |
12fcfd22 | 5615 | |
0b246afa | 5616 | if (btrfs_test_opt(fs_info, NOTREELOG)) { |
3a5e1404 SW |
5617 | ret = 1; |
5618 | goto end_no_trans; | |
5619 | } | |
5620 | ||
995946dd MX |
5621 | /* |
5622 | * The prev transaction commit doesn't complete, we need do | |
5623 | * full commit by ourselves. | |
5624 | */ | |
0b246afa JM |
5625 | if (fs_info->last_trans_log_full_commit > |
5626 | fs_info->last_trans_committed) { | |
12fcfd22 CM |
5627 | ret = 1; |
5628 | goto end_no_trans; | |
5629 | } | |
5630 | ||
f882274b | 5631 | if (btrfs_root_refs(&root->root_item) == 0) { |
76dda93c YZ |
5632 | ret = 1; |
5633 | goto end_no_trans; | |
5634 | } | |
5635 | ||
19df27a9 NB |
5636 | ret = check_parent_dirs_for_sync(trans, inode, parent, sb, |
5637 | last_committed); | |
12fcfd22 CM |
5638 | if (ret) |
5639 | goto end_no_trans; | |
e02119d5 | 5640 | |
19df27a9 | 5641 | if (btrfs_inode_in_log(inode, trans->transid)) { |
257c62e1 CM |
5642 | ret = BTRFS_NO_LOG_SYNC; |
5643 | goto end_no_trans; | |
5644 | } | |
5645 | ||
8b050d35 | 5646 | ret = start_log_trans(trans, root, ctx); |
4a500fd1 | 5647 | if (ret) |
e87ac136 | 5648 | goto end_no_trans; |
e02119d5 | 5649 | |
19df27a9 | 5650 | ret = btrfs_log_inode(trans, root, inode, inode_only, start, end, ctx); |
4a500fd1 YZ |
5651 | if (ret) |
5652 | goto end_trans; | |
12fcfd22 | 5653 | |
af4176b4 CM |
5654 | /* |
5655 | * for regular files, if its inode is already on disk, we don't | |
5656 | * have to worry about the parents at all. This is because | |
5657 | * we can use the last_unlink_trans field to record renames | |
5658 | * and other fun in this file. | |
5659 | */ | |
19df27a9 NB |
5660 | if (S_ISREG(inode->vfs_inode.i_mode) && |
5661 | inode->generation <= last_committed && | |
5662 | inode->last_unlink_trans <= last_committed) { | |
4a500fd1 YZ |
5663 | ret = 0; |
5664 | goto end_trans; | |
5665 | } | |
af4176b4 | 5666 | |
19df27a9 | 5667 | if (S_ISDIR(inode->vfs_inode.i_mode) && ctx && ctx->log_new_dentries) |
2f2ff0ee FM |
5668 | log_dentries = true; |
5669 | ||
18aa0922 | 5670 | /* |
01327610 | 5671 | * On unlink we must make sure all our current and old parent directory |
18aa0922 FM |
5672 | * inodes are fully logged. This is to prevent leaving dangling |
5673 | * directory index entries in directories that were our parents but are | |
5674 | * not anymore. Not doing this results in old parent directory being | |
5675 | * impossible to delete after log replay (rmdir will always fail with | |
5676 | * error -ENOTEMPTY). | |
5677 | * | |
5678 | * Example 1: | |
5679 | * | |
5680 | * mkdir testdir | |
5681 | * touch testdir/foo | |
5682 | * ln testdir/foo testdir/bar | |
5683 | * sync | |
5684 | * unlink testdir/bar | |
5685 | * xfs_io -c fsync testdir/foo | |
5686 | * <power failure> | |
5687 | * mount fs, triggers log replay | |
5688 | * | |
5689 | * If we don't log the parent directory (testdir), after log replay the | |
5690 | * directory still has an entry pointing to the file inode using the bar | |
5691 | * name, but a matching BTRFS_INODE_[REF|EXTREF]_KEY does not exist and | |
5692 | * the file inode has a link count of 1. | |
5693 | * | |
5694 | * Example 2: | |
5695 | * | |
5696 | * mkdir testdir | |
5697 | * touch foo | |
5698 | * ln foo testdir/foo2 | |
5699 | * ln foo testdir/foo3 | |
5700 | * sync | |
5701 | * unlink testdir/foo3 | |
5702 | * xfs_io -c fsync foo | |
5703 | * <power failure> | |
5704 | * mount fs, triggers log replay | |
5705 | * | |
5706 | * Similar as the first example, after log replay the parent directory | |
5707 | * testdir still has an entry pointing to the inode file with name foo3 | |
5708 | * but the file inode does not have a matching BTRFS_INODE_REF_KEY item | |
5709 | * and has a link count of 2. | |
5710 | */ | |
19df27a9 | 5711 | if (inode->last_unlink_trans > last_committed) { |
18aa0922 FM |
5712 | ret = btrfs_log_all_parents(trans, orig_inode, ctx); |
5713 | if (ret) | |
5714 | goto end_trans; | |
5715 | } | |
5716 | ||
12fcfd22 | 5717 | while (1) { |
fc64005c | 5718 | if (!parent || d_really_is_negative(parent) || sb != parent->d_sb) |
e02119d5 CM |
5719 | break; |
5720 | ||
19df27a9 NB |
5721 | inode = BTRFS_I(d_inode(parent)); |
5722 | if (root != inode->root) | |
76dda93c YZ |
5723 | break; |
5724 | ||
19df27a9 NB |
5725 | if (inode->generation > last_committed) { |
5726 | ret = btrfs_log_inode(trans, root, inode, | |
5727 | LOG_INODE_EXISTS, 0, LLONG_MAX, ctx); | |
4a500fd1 YZ |
5728 | if (ret) |
5729 | goto end_trans; | |
12fcfd22 | 5730 | } |
76dda93c | 5731 | if (IS_ROOT(parent)) |
e02119d5 | 5732 | break; |
12fcfd22 | 5733 | |
6a912213 JB |
5734 | parent = dget_parent(parent); |
5735 | dput(old_parent); | |
5736 | old_parent = parent; | |
e02119d5 | 5737 | } |
2f2ff0ee | 5738 | if (log_dentries) |
19df27a9 | 5739 | ret = log_new_dir_dentries(trans, root, orig_inode, ctx); |
2f2ff0ee FM |
5740 | else |
5741 | ret = 0; | |
4a500fd1 | 5742 | end_trans: |
6a912213 | 5743 | dput(old_parent); |
4a500fd1 | 5744 | if (ret < 0) { |
0b246afa | 5745 | btrfs_set_log_full_commit(fs_info, trans); |
4a500fd1 YZ |
5746 | ret = 1; |
5747 | } | |
8b050d35 MX |
5748 | |
5749 | if (ret) | |
5750 | btrfs_remove_log_ctx(root, ctx); | |
12fcfd22 CM |
5751 | btrfs_end_log_trans(root); |
5752 | end_no_trans: | |
5753 | return ret; | |
e02119d5 CM |
5754 | } |
5755 | ||
5756 | /* | |
5757 | * it is not safe to log dentry if the chunk root has added new | |
5758 | * chunks. This returns 0 if the dentry was logged, and 1 otherwise. | |
5759 | * If this returns 1, you must commit the transaction to safely get your | |
5760 | * data on disk. | |
5761 | */ | |
5762 | int btrfs_log_dentry_safe(struct btrfs_trans_handle *trans, | |
e5b84f7a | 5763 | struct dentry *dentry, |
49dae1bc FM |
5764 | const loff_t start, |
5765 | const loff_t end, | |
8b050d35 | 5766 | struct btrfs_log_ctx *ctx) |
e02119d5 | 5767 | { |
6a912213 JB |
5768 | struct dentry *parent = dget_parent(dentry); |
5769 | int ret; | |
5770 | ||
f882274b NB |
5771 | ret = btrfs_log_inode_parent(trans, BTRFS_I(d_inode(dentry)), parent, |
5772 | start, end, LOG_INODE_ALL, ctx); | |
6a912213 JB |
5773 | dput(parent); |
5774 | ||
5775 | return ret; | |
e02119d5 CM |
5776 | } |
5777 | ||
5778 | /* | |
5779 | * should be called during mount to recover any replay any log trees | |
5780 | * from the FS | |
5781 | */ | |
5782 | int btrfs_recover_log_trees(struct btrfs_root *log_root_tree) | |
5783 | { | |
5784 | int ret; | |
5785 | struct btrfs_path *path; | |
5786 | struct btrfs_trans_handle *trans; | |
5787 | struct btrfs_key key; | |
5788 | struct btrfs_key found_key; | |
5789 | struct btrfs_key tmp_key; | |
5790 | struct btrfs_root *log; | |
5791 | struct btrfs_fs_info *fs_info = log_root_tree->fs_info; | |
5792 | struct walk_control wc = { | |
5793 | .process_func = process_one_buffer, | |
5794 | .stage = 0, | |
5795 | }; | |
5796 | ||
e02119d5 | 5797 | path = btrfs_alloc_path(); |
db5b493a TI |
5798 | if (!path) |
5799 | return -ENOMEM; | |
5800 | ||
afcdd129 | 5801 | set_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags); |
e02119d5 | 5802 | |
4a500fd1 | 5803 | trans = btrfs_start_transaction(fs_info->tree_root, 0); |
79787eaa JM |
5804 | if (IS_ERR(trans)) { |
5805 | ret = PTR_ERR(trans); | |
5806 | goto error; | |
5807 | } | |
e02119d5 CM |
5808 | |
5809 | wc.trans = trans; | |
5810 | wc.pin = 1; | |
5811 | ||
db5b493a | 5812 | ret = walk_log_tree(trans, log_root_tree, &wc); |
79787eaa | 5813 | if (ret) { |
5d163e0e JM |
5814 | btrfs_handle_fs_error(fs_info, ret, |
5815 | "Failed to pin buffers while recovering log root tree."); | |
79787eaa JM |
5816 | goto error; |
5817 | } | |
e02119d5 CM |
5818 | |
5819 | again: | |
5820 | key.objectid = BTRFS_TREE_LOG_OBJECTID; | |
5821 | key.offset = (u64)-1; | |
962a298f | 5822 | key.type = BTRFS_ROOT_ITEM_KEY; |
e02119d5 | 5823 | |
d397712b | 5824 | while (1) { |
e02119d5 | 5825 | ret = btrfs_search_slot(NULL, log_root_tree, &key, path, 0, 0); |
79787eaa JM |
5826 | |
5827 | if (ret < 0) { | |
34d97007 | 5828 | btrfs_handle_fs_error(fs_info, ret, |
79787eaa JM |
5829 | "Couldn't find tree log root."); |
5830 | goto error; | |
5831 | } | |
e02119d5 CM |
5832 | if (ret > 0) { |
5833 | if (path->slots[0] == 0) | |
5834 | break; | |
5835 | path->slots[0]--; | |
5836 | } | |
5837 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, | |
5838 | path->slots[0]); | |
b3b4aa74 | 5839 | btrfs_release_path(path); |
e02119d5 CM |
5840 | if (found_key.objectid != BTRFS_TREE_LOG_OBJECTID) |
5841 | break; | |
5842 | ||
cb517eab | 5843 | log = btrfs_read_fs_root(log_root_tree, &found_key); |
79787eaa JM |
5844 | if (IS_ERR(log)) { |
5845 | ret = PTR_ERR(log); | |
34d97007 | 5846 | btrfs_handle_fs_error(fs_info, ret, |
79787eaa JM |
5847 | "Couldn't read tree log root."); |
5848 | goto error; | |
5849 | } | |
e02119d5 CM |
5850 | |
5851 | tmp_key.objectid = found_key.offset; | |
5852 | tmp_key.type = BTRFS_ROOT_ITEM_KEY; | |
5853 | tmp_key.offset = (u64)-1; | |
5854 | ||
5855 | wc.replay_dest = btrfs_read_fs_root_no_name(fs_info, &tmp_key); | |
79787eaa JM |
5856 | if (IS_ERR(wc.replay_dest)) { |
5857 | ret = PTR_ERR(wc.replay_dest); | |
b50c6e25 JB |
5858 | free_extent_buffer(log->node); |
5859 | free_extent_buffer(log->commit_root); | |
5860 | kfree(log); | |
5d163e0e JM |
5861 | btrfs_handle_fs_error(fs_info, ret, |
5862 | "Couldn't read target root for tree log recovery."); | |
79787eaa JM |
5863 | goto error; |
5864 | } | |
e02119d5 | 5865 | |
07d400a6 | 5866 | wc.replay_dest->log_root = log; |
5d4f98a2 | 5867 | btrfs_record_root_in_trans(trans, wc.replay_dest); |
e02119d5 | 5868 | ret = walk_log_tree(trans, log, &wc); |
e02119d5 | 5869 | |
b50c6e25 | 5870 | if (!ret && wc.stage == LOG_WALK_REPLAY_ALL) { |
e02119d5 CM |
5871 | ret = fixup_inode_link_counts(trans, wc.replay_dest, |
5872 | path); | |
e02119d5 CM |
5873 | } |
5874 | ||
900c9981 LB |
5875 | if (!ret && wc.stage == LOG_WALK_REPLAY_ALL) { |
5876 | struct btrfs_root *root = wc.replay_dest; | |
5877 | ||
5878 | btrfs_release_path(path); | |
5879 | ||
5880 | /* | |
5881 | * We have just replayed everything, and the highest | |
5882 | * objectid of fs roots probably has changed in case | |
5883 | * some inode_item's got replayed. | |
5884 | * | |
5885 | * root->objectid_mutex is not acquired as log replay | |
5886 | * could only happen during mount. | |
5887 | */ | |
5888 | ret = btrfs_find_highest_objectid(root, | |
5889 | &root->highest_objectid); | |
5890 | } | |
5891 | ||
e02119d5 | 5892 | key.offset = found_key.offset - 1; |
07d400a6 | 5893 | wc.replay_dest->log_root = NULL; |
e02119d5 | 5894 | free_extent_buffer(log->node); |
b263c2c8 | 5895 | free_extent_buffer(log->commit_root); |
e02119d5 CM |
5896 | kfree(log); |
5897 | ||
b50c6e25 JB |
5898 | if (ret) |
5899 | goto error; | |
5900 | ||
e02119d5 CM |
5901 | if (found_key.offset == 0) |
5902 | break; | |
5903 | } | |
b3b4aa74 | 5904 | btrfs_release_path(path); |
e02119d5 CM |
5905 | |
5906 | /* step one is to pin it all, step two is to replay just inodes */ | |
5907 | if (wc.pin) { | |
5908 | wc.pin = 0; | |
5909 | wc.process_func = replay_one_buffer; | |
5910 | wc.stage = LOG_WALK_REPLAY_INODES; | |
5911 | goto again; | |
5912 | } | |
5913 | /* step three is to replay everything */ | |
5914 | if (wc.stage < LOG_WALK_REPLAY_ALL) { | |
5915 | wc.stage++; | |
5916 | goto again; | |
5917 | } | |
5918 | ||
5919 | btrfs_free_path(path); | |
5920 | ||
abefa55a | 5921 | /* step 4: commit the transaction, which also unpins the blocks */ |
3a45bb20 | 5922 | ret = btrfs_commit_transaction(trans); |
abefa55a JB |
5923 | if (ret) |
5924 | return ret; | |
5925 | ||
e02119d5 CM |
5926 | free_extent_buffer(log_root_tree->node); |
5927 | log_root_tree->log_root = NULL; | |
afcdd129 | 5928 | clear_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags); |
e02119d5 | 5929 | kfree(log_root_tree); |
79787eaa | 5930 | |
abefa55a | 5931 | return 0; |
79787eaa | 5932 | error: |
b50c6e25 | 5933 | if (wc.trans) |
3a45bb20 | 5934 | btrfs_end_transaction(wc.trans); |
79787eaa JM |
5935 | btrfs_free_path(path); |
5936 | return ret; | |
e02119d5 | 5937 | } |
12fcfd22 CM |
5938 | |
5939 | /* | |
5940 | * there are some corner cases where we want to force a full | |
5941 | * commit instead of allowing a directory to be logged. | |
5942 | * | |
5943 | * They revolve around files there were unlinked from the directory, and | |
5944 | * this function updates the parent directory so that a full commit is | |
5945 | * properly done if it is fsync'd later after the unlinks are done. | |
2be63d5c FM |
5946 | * |
5947 | * Must be called before the unlink operations (updates to the subvolume tree, | |
5948 | * inodes, etc) are done. | |
12fcfd22 CM |
5949 | */ |
5950 | void btrfs_record_unlink_dir(struct btrfs_trans_handle *trans, | |
4176bdbf | 5951 | struct btrfs_inode *dir, struct btrfs_inode *inode, |
12fcfd22 CM |
5952 | int for_rename) |
5953 | { | |
af4176b4 CM |
5954 | /* |
5955 | * when we're logging a file, if it hasn't been renamed | |
5956 | * or unlinked, and its inode is fully committed on disk, | |
5957 | * we don't have to worry about walking up the directory chain | |
5958 | * to log its parents. | |
5959 | * | |
5960 | * So, we use the last_unlink_trans field to put this transid | |
5961 | * into the file. When the file is logged we check it and | |
5962 | * don't log the parents if the file is fully on disk. | |
5963 | */ | |
4176bdbf NB |
5964 | mutex_lock(&inode->log_mutex); |
5965 | inode->last_unlink_trans = trans->transid; | |
5966 | mutex_unlock(&inode->log_mutex); | |
af4176b4 | 5967 | |
12fcfd22 CM |
5968 | /* |
5969 | * if this directory was already logged any new | |
5970 | * names for this file/dir will get recorded | |
5971 | */ | |
5972 | smp_mb(); | |
4176bdbf | 5973 | if (dir->logged_trans == trans->transid) |
12fcfd22 CM |
5974 | return; |
5975 | ||
5976 | /* | |
5977 | * if the inode we're about to unlink was logged, | |
5978 | * the log will be properly updated for any new names | |
5979 | */ | |
4176bdbf | 5980 | if (inode->logged_trans == trans->transid) |
12fcfd22 CM |
5981 | return; |
5982 | ||
5983 | /* | |
5984 | * when renaming files across directories, if the directory | |
5985 | * there we're unlinking from gets fsync'd later on, there's | |
5986 | * no way to find the destination directory later and fsync it | |
5987 | * properly. So, we have to be conservative and force commits | |
5988 | * so the new name gets discovered. | |
5989 | */ | |
5990 | if (for_rename) | |
5991 | goto record; | |
5992 | ||
5993 | /* we can safely do the unlink without any special recording */ | |
5994 | return; | |
5995 | ||
5996 | record: | |
4176bdbf NB |
5997 | mutex_lock(&dir->log_mutex); |
5998 | dir->last_unlink_trans = trans->transid; | |
5999 | mutex_unlock(&dir->log_mutex); | |
1ec9a1ae FM |
6000 | } |
6001 | ||
6002 | /* | |
6003 | * Make sure that if someone attempts to fsync the parent directory of a deleted | |
6004 | * snapshot, it ends up triggering a transaction commit. This is to guarantee | |
6005 | * that after replaying the log tree of the parent directory's root we will not | |
6006 | * see the snapshot anymore and at log replay time we will not see any log tree | |
6007 | * corresponding to the deleted snapshot's root, which could lead to replaying | |
6008 | * it after replaying the log tree of the parent directory (which would replay | |
6009 | * the snapshot delete operation). | |
2be63d5c FM |
6010 | * |
6011 | * Must be called before the actual snapshot destroy operation (updates to the | |
6012 | * parent root and tree of tree roots trees, etc) are done. | |
1ec9a1ae FM |
6013 | */ |
6014 | void btrfs_record_snapshot_destroy(struct btrfs_trans_handle *trans, | |
43663557 | 6015 | struct btrfs_inode *dir) |
1ec9a1ae | 6016 | { |
43663557 NB |
6017 | mutex_lock(&dir->log_mutex); |
6018 | dir->last_unlink_trans = trans->transid; | |
6019 | mutex_unlock(&dir->log_mutex); | |
12fcfd22 CM |
6020 | } |
6021 | ||
6022 | /* | |
6023 | * Call this after adding a new name for a file and it will properly | |
6024 | * update the log to reflect the new name. | |
6025 | * | |
6026 | * It will return zero if all goes well, and it will return 1 if a | |
6027 | * full transaction commit is required. | |
6028 | */ | |
6029 | int btrfs_log_new_name(struct btrfs_trans_handle *trans, | |
9ca5fbfb | 6030 | struct btrfs_inode *inode, struct btrfs_inode *old_dir, |
12fcfd22 CM |
6031 | struct dentry *parent) |
6032 | { | |
9ca5fbfb | 6033 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb); |
12fcfd22 | 6034 | |
af4176b4 CM |
6035 | /* |
6036 | * this will force the logging code to walk the dentry chain | |
6037 | * up for the file | |
6038 | */ | |
9a6509c4 | 6039 | if (!S_ISDIR(inode->vfs_inode.i_mode)) |
9ca5fbfb | 6040 | inode->last_unlink_trans = trans->transid; |
af4176b4 | 6041 | |
12fcfd22 CM |
6042 | /* |
6043 | * if this inode hasn't been logged and directory we're renaming it | |
6044 | * from hasn't been logged, we don't need to log it | |
6045 | */ | |
9ca5fbfb NB |
6046 | if (inode->logged_trans <= fs_info->last_trans_committed && |
6047 | (!old_dir || old_dir->logged_trans <= fs_info->last_trans_committed)) | |
12fcfd22 CM |
6048 | return 0; |
6049 | ||
f882274b NB |
6050 | return btrfs_log_inode_parent(trans, inode, parent, 0, LLONG_MAX, |
6051 | LOG_INODE_EXISTS, NULL); | |
12fcfd22 CM |
6052 | } |
6053 |