]>
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 |
a3baaf0d | 30 | #define LOG_OTHER_INODE_ALL 3 |
e02119d5 | 31 | |
12fcfd22 CM |
32 | /* |
33 | * directory trouble cases | |
34 | * | |
35 | * 1) on rename or unlink, if the inode being unlinked isn't in the fsync | |
36 | * log, we must force a full commit before doing an fsync of the directory | |
37 | * where the unlink was done. | |
38 | * ---> record transid of last unlink/rename per directory | |
39 | * | |
40 | * mkdir foo/some_dir | |
41 | * normal commit | |
42 | * rename foo/some_dir foo2/some_dir | |
43 | * mkdir foo/some_dir | |
44 | * fsync foo/some_dir/some_file | |
45 | * | |
46 | * The fsync above will unlink the original some_dir without recording | |
47 | * it in its new location (foo2). After a crash, some_dir will be gone | |
48 | * unless the fsync of some_file forces a full commit | |
49 | * | |
50 | * 2) we must log any new names for any file or dir that is in the fsync | |
51 | * log. ---> check inode while renaming/linking. | |
52 | * | |
53 | * 2a) we must log any new names for any file or dir during rename | |
54 | * when the directory they are being removed from was logged. | |
55 | * ---> check inode and old parent dir during rename | |
56 | * | |
57 | * 2a is actually the more important variant. With the extra logging | |
58 | * a crash might unlink the old name without recreating the new one | |
59 | * | |
60 | * 3) after a crash, we must go through any directories with a link count | |
61 | * of zero and redo the rm -rf | |
62 | * | |
63 | * mkdir f1/foo | |
64 | * normal commit | |
65 | * rm -rf f1/foo | |
66 | * fsync(f1) | |
67 | * | |
68 | * The directory f1 was fully removed from the FS, but fsync was never | |
69 | * called on f1, only its parent dir. After a crash the rm -rf must | |
70 | * be replayed. This must be able to recurse down the entire | |
71 | * directory tree. The inode link count fixup code takes care of the | |
72 | * ugly details. | |
73 | */ | |
74 | ||
e02119d5 CM |
75 | /* |
76 | * stages for the tree walking. The first | |
77 | * stage (0) is to only pin down the blocks we find | |
78 | * the second stage (1) is to make sure that all the inodes | |
79 | * we find in the log are created in the subvolume. | |
80 | * | |
81 | * The last stage is to deal with directories and links and extents | |
82 | * and all the other fun semantics | |
83 | */ | |
84 | #define LOG_WALK_PIN_ONLY 0 | |
85 | #define LOG_WALK_REPLAY_INODES 1 | |
dd8e7217 JB |
86 | #define LOG_WALK_REPLAY_DIR_INDEX 2 |
87 | #define LOG_WALK_REPLAY_ALL 3 | |
e02119d5 | 88 | |
12fcfd22 | 89 | static int btrfs_log_inode(struct btrfs_trans_handle *trans, |
a59108a7 | 90 | struct btrfs_root *root, struct btrfs_inode *inode, |
49dae1bc FM |
91 | int inode_only, |
92 | const loff_t start, | |
8407f553 FM |
93 | const loff_t end, |
94 | struct btrfs_log_ctx *ctx); | |
ec051c0f YZ |
95 | static int link_to_fixup_dir(struct btrfs_trans_handle *trans, |
96 | struct btrfs_root *root, | |
97 | struct btrfs_path *path, u64 objectid); | |
12fcfd22 CM |
98 | static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans, |
99 | struct btrfs_root *root, | |
100 | struct btrfs_root *log, | |
101 | struct btrfs_path *path, | |
102 | u64 dirid, int del_all); | |
e02119d5 CM |
103 | |
104 | /* | |
105 | * tree logging is a special write ahead log used to make sure that | |
106 | * fsyncs and O_SYNCs can happen without doing full tree commits. | |
107 | * | |
108 | * Full tree commits are expensive because they require commonly | |
109 | * modified blocks to be recowed, creating many dirty pages in the | |
110 | * extent tree an 4x-6x higher write load than ext3. | |
111 | * | |
112 | * Instead of doing a tree commit on every fsync, we use the | |
113 | * key ranges and transaction ids to find items for a given file or directory | |
114 | * that have changed in this transaction. Those items are copied into | |
115 | * a special tree (one per subvolume root), that tree is written to disk | |
116 | * and then the fsync is considered complete. | |
117 | * | |
118 | * After a crash, items are copied out of the log-tree back into the | |
119 | * subvolume tree. Any file data extents found are recorded in the extent | |
120 | * allocation tree, and the log-tree freed. | |
121 | * | |
122 | * The log tree is read three times, once to pin down all the extents it is | |
123 | * using in ram and once, once to create all the inodes logged in the tree | |
124 | * and once to do all the other items. | |
125 | */ | |
126 | ||
e02119d5 CM |
127 | /* |
128 | * start a sub transaction and setup the log tree | |
129 | * this increments the log tree writer count to make the people | |
130 | * syncing the tree wait for us to finish | |
131 | */ | |
132 | static int start_log_trans(struct btrfs_trans_handle *trans, | |
8b050d35 MX |
133 | struct btrfs_root *root, |
134 | struct btrfs_log_ctx *ctx) | |
e02119d5 | 135 | { |
0b246afa | 136 | struct btrfs_fs_info *fs_info = root->fs_info; |
34eb2a52 | 137 | int ret = 0; |
7237f183 YZ |
138 | |
139 | mutex_lock(&root->log_mutex); | |
34eb2a52 | 140 | |
7237f183 | 141 | if (root->log_root) { |
0b246afa | 142 | if (btrfs_need_log_full_commit(fs_info, trans)) { |
50471a38 MX |
143 | ret = -EAGAIN; |
144 | goto out; | |
145 | } | |
34eb2a52 | 146 | |
ff782e0a | 147 | if (!root->log_start_pid) { |
27cdeb70 | 148 | clear_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state); |
34eb2a52 | 149 | root->log_start_pid = current->pid; |
ff782e0a | 150 | } else if (root->log_start_pid != current->pid) { |
27cdeb70 | 151 | set_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state); |
ff782e0a | 152 | } |
34eb2a52 | 153 | } else { |
0b246afa JM |
154 | mutex_lock(&fs_info->tree_log_mutex); |
155 | if (!fs_info->log_root_tree) | |
156 | ret = btrfs_init_log_root_tree(trans, fs_info); | |
157 | mutex_unlock(&fs_info->tree_log_mutex); | |
34eb2a52 Z |
158 | if (ret) |
159 | goto out; | |
ff782e0a | 160 | |
e02119d5 | 161 | ret = btrfs_add_log_tree(trans, root); |
4a500fd1 | 162 | if (ret) |
e87ac136 | 163 | goto out; |
34eb2a52 Z |
164 | |
165 | clear_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state); | |
166 | root->log_start_pid = current->pid; | |
e02119d5 | 167 | } |
34eb2a52 | 168 | |
2ecb7923 | 169 | atomic_inc(&root->log_batch); |
7237f183 | 170 | atomic_inc(&root->log_writers); |
8b050d35 | 171 | if (ctx) { |
34eb2a52 | 172 | int index = root->log_transid % 2; |
8b050d35 | 173 | list_add_tail(&ctx->list, &root->log_ctxs[index]); |
d1433deb | 174 | ctx->log_transid = root->log_transid; |
8b050d35 | 175 | } |
34eb2a52 | 176 | |
e87ac136 | 177 | out: |
7237f183 | 178 | mutex_unlock(&root->log_mutex); |
e87ac136 | 179 | return ret; |
e02119d5 CM |
180 | } |
181 | ||
182 | /* | |
183 | * returns 0 if there was a log transaction running and we were able | |
184 | * to join, or returns -ENOENT if there were not transactions | |
185 | * in progress | |
186 | */ | |
187 | static int join_running_log_trans(struct btrfs_root *root) | |
188 | { | |
189 | int ret = -ENOENT; | |
190 | ||
191 | smp_mb(); | |
192 | if (!root->log_root) | |
193 | return -ENOENT; | |
194 | ||
7237f183 | 195 | mutex_lock(&root->log_mutex); |
e02119d5 CM |
196 | if (root->log_root) { |
197 | ret = 0; | |
7237f183 | 198 | atomic_inc(&root->log_writers); |
e02119d5 | 199 | } |
7237f183 | 200 | mutex_unlock(&root->log_mutex); |
e02119d5 CM |
201 | return ret; |
202 | } | |
203 | ||
12fcfd22 CM |
204 | /* |
205 | * This either makes the current running log transaction wait | |
206 | * until you call btrfs_end_log_trans() or it makes any future | |
207 | * log transactions wait until you call btrfs_end_log_trans() | |
208 | */ | |
45128b08 | 209 | void btrfs_pin_log_trans(struct btrfs_root *root) |
12fcfd22 | 210 | { |
12fcfd22 CM |
211 | mutex_lock(&root->log_mutex); |
212 | atomic_inc(&root->log_writers); | |
213 | mutex_unlock(&root->log_mutex); | |
12fcfd22 CM |
214 | } |
215 | ||
e02119d5 CM |
216 | /* |
217 | * indicate we're done making changes to the log tree | |
218 | * and wake up anyone waiting to do a sync | |
219 | */ | |
143bede5 | 220 | void btrfs_end_log_trans(struct btrfs_root *root) |
e02119d5 | 221 | { |
7237f183 | 222 | if (atomic_dec_and_test(&root->log_writers)) { |
093258e6 DS |
223 | /* atomic_dec_and_test implies a barrier */ |
224 | cond_wake_up_nomb(&root->log_writer_wait); | |
7237f183 | 225 | } |
e02119d5 CM |
226 | } |
227 | ||
247462a5 DS |
228 | static int btrfs_write_tree_block(struct extent_buffer *buf) |
229 | { | |
230 | return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start, | |
231 | buf->start + buf->len - 1); | |
232 | } | |
233 | ||
234 | static void btrfs_wait_tree_block_writeback(struct extent_buffer *buf) | |
235 | { | |
236 | filemap_fdatawait_range(buf->pages[0]->mapping, | |
237 | buf->start, buf->start + buf->len - 1); | |
238 | } | |
e02119d5 CM |
239 | |
240 | /* | |
241 | * the walk control struct is used to pass state down the chain when | |
242 | * processing the log tree. The stage field tells us which part | |
243 | * of the log tree processing we are currently doing. The others | |
244 | * are state fields used for that specific part | |
245 | */ | |
246 | struct walk_control { | |
247 | /* should we free the extent on disk when done? This is used | |
248 | * at transaction commit time while freeing a log tree | |
249 | */ | |
250 | int free; | |
251 | ||
252 | /* should we write out the extent buffer? This is used | |
253 | * while flushing the log tree to disk during a sync | |
254 | */ | |
255 | int write; | |
256 | ||
257 | /* should we wait for the extent buffer io to finish? Also used | |
258 | * while flushing the log tree to disk for a sync | |
259 | */ | |
260 | int wait; | |
261 | ||
262 | /* pin only walk, we record which extents on disk belong to the | |
263 | * log trees | |
264 | */ | |
265 | int pin; | |
266 | ||
267 | /* what stage of the replay code we're currently in */ | |
268 | int stage; | |
269 | ||
f2d72f42 FM |
270 | /* |
271 | * Ignore any items from the inode currently being processed. Needs | |
272 | * to be set every time we find a BTRFS_INODE_ITEM_KEY and we are in | |
273 | * the LOG_WALK_REPLAY_INODES stage. | |
274 | */ | |
275 | bool ignore_cur_inode; | |
276 | ||
e02119d5 CM |
277 | /* the root we are currently replaying */ |
278 | struct btrfs_root *replay_dest; | |
279 | ||
280 | /* the trans handle for the current replay */ | |
281 | struct btrfs_trans_handle *trans; | |
282 | ||
283 | /* the function that gets used to process blocks we find in the | |
284 | * tree. Note the extent_buffer might not be up to date when it is | |
285 | * passed in, and it must be checked or read if you need the data | |
286 | * inside it | |
287 | */ | |
288 | int (*process_func)(struct btrfs_root *log, struct extent_buffer *eb, | |
581c1760 | 289 | struct walk_control *wc, u64 gen, int level); |
e02119d5 CM |
290 | }; |
291 | ||
292 | /* | |
293 | * process_func used to pin down extents, write them or wait on them | |
294 | */ | |
295 | static int process_one_buffer(struct btrfs_root *log, | |
296 | struct extent_buffer *eb, | |
581c1760 | 297 | struct walk_control *wc, u64 gen, int level) |
e02119d5 | 298 | { |
0b246afa | 299 | struct btrfs_fs_info *fs_info = log->fs_info; |
b50c6e25 JB |
300 | int ret = 0; |
301 | ||
8c2a1a30 JB |
302 | /* |
303 | * If this fs is mixed then we need to be able to process the leaves to | |
304 | * pin down any logged extents, so we have to read the block. | |
305 | */ | |
0b246afa | 306 | if (btrfs_fs_incompat(fs_info, MIXED_GROUPS)) { |
581c1760 | 307 | ret = btrfs_read_buffer(eb, gen, level, NULL); |
8c2a1a30 JB |
308 | if (ret) |
309 | return ret; | |
310 | } | |
311 | ||
04018de5 | 312 | if (wc->pin) |
2ff7e61e JM |
313 | ret = btrfs_pin_extent_for_log_replay(fs_info, eb->start, |
314 | eb->len); | |
e02119d5 | 315 | |
b50c6e25 | 316 | if (!ret && btrfs_buffer_uptodate(eb, gen, 0)) { |
8c2a1a30 | 317 | if (wc->pin && btrfs_header_level(eb) == 0) |
bcdc428c | 318 | ret = btrfs_exclude_logged_extents(eb); |
e02119d5 CM |
319 | if (wc->write) |
320 | btrfs_write_tree_block(eb); | |
321 | if (wc->wait) | |
322 | btrfs_wait_tree_block_writeback(eb); | |
323 | } | |
b50c6e25 | 324 | return ret; |
e02119d5 CM |
325 | } |
326 | ||
327 | /* | |
328 | * Item overwrite used by replay and tree logging. eb, slot and key all refer | |
329 | * to the src data we are copying out. | |
330 | * | |
331 | * root is the tree we are copying into, and path is a scratch | |
332 | * path for use in this function (it should be released on entry and | |
333 | * will be released on exit). | |
334 | * | |
335 | * If the key is already in the destination tree the existing item is | |
336 | * overwritten. If the existing item isn't big enough, it is extended. | |
337 | * If it is too large, it is truncated. | |
338 | * | |
339 | * If the key isn't in the destination yet, a new item is inserted. | |
340 | */ | |
341 | static noinline int overwrite_item(struct btrfs_trans_handle *trans, | |
342 | struct btrfs_root *root, | |
343 | struct btrfs_path *path, | |
344 | struct extent_buffer *eb, int slot, | |
345 | struct btrfs_key *key) | |
346 | { | |
2ff7e61e | 347 | struct btrfs_fs_info *fs_info = root->fs_info; |
e02119d5 CM |
348 | int ret; |
349 | u32 item_size; | |
350 | u64 saved_i_size = 0; | |
351 | int save_old_i_size = 0; | |
352 | unsigned long src_ptr; | |
353 | unsigned long dst_ptr; | |
354 | int overwrite_root = 0; | |
4bc4bee4 | 355 | bool inode_item = key->type == BTRFS_INODE_ITEM_KEY; |
e02119d5 CM |
356 | |
357 | if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) | |
358 | overwrite_root = 1; | |
359 | ||
360 | item_size = btrfs_item_size_nr(eb, slot); | |
361 | src_ptr = btrfs_item_ptr_offset(eb, slot); | |
362 | ||
363 | /* look for the key in the destination tree */ | |
364 | ret = btrfs_search_slot(NULL, root, key, path, 0, 0); | |
4bc4bee4 JB |
365 | if (ret < 0) |
366 | return ret; | |
367 | ||
e02119d5 CM |
368 | if (ret == 0) { |
369 | char *src_copy; | |
370 | char *dst_copy; | |
371 | u32 dst_size = btrfs_item_size_nr(path->nodes[0], | |
372 | path->slots[0]); | |
373 | if (dst_size != item_size) | |
374 | goto insert; | |
375 | ||
376 | if (item_size == 0) { | |
b3b4aa74 | 377 | btrfs_release_path(path); |
e02119d5 CM |
378 | return 0; |
379 | } | |
380 | dst_copy = kmalloc(item_size, GFP_NOFS); | |
381 | src_copy = kmalloc(item_size, GFP_NOFS); | |
2a29edc6 | 382 | if (!dst_copy || !src_copy) { |
b3b4aa74 | 383 | btrfs_release_path(path); |
2a29edc6 | 384 | kfree(dst_copy); |
385 | kfree(src_copy); | |
386 | return -ENOMEM; | |
387 | } | |
e02119d5 CM |
388 | |
389 | read_extent_buffer(eb, src_copy, src_ptr, item_size); | |
390 | ||
391 | dst_ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]); | |
392 | read_extent_buffer(path->nodes[0], dst_copy, dst_ptr, | |
393 | item_size); | |
394 | ret = memcmp(dst_copy, src_copy, item_size); | |
395 | ||
396 | kfree(dst_copy); | |
397 | kfree(src_copy); | |
398 | /* | |
399 | * they have the same contents, just return, this saves | |
400 | * us from cowing blocks in the destination tree and doing | |
401 | * extra writes that may not have been done by a previous | |
402 | * sync | |
403 | */ | |
404 | if (ret == 0) { | |
b3b4aa74 | 405 | btrfs_release_path(path); |
e02119d5 CM |
406 | return 0; |
407 | } | |
408 | ||
4bc4bee4 JB |
409 | /* |
410 | * We need to load the old nbytes into the inode so when we | |
411 | * replay the extents we've logged we get the right nbytes. | |
412 | */ | |
413 | if (inode_item) { | |
414 | struct btrfs_inode_item *item; | |
415 | u64 nbytes; | |
d555438b | 416 | u32 mode; |
4bc4bee4 JB |
417 | |
418 | item = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
419 | struct btrfs_inode_item); | |
420 | nbytes = btrfs_inode_nbytes(path->nodes[0], item); | |
421 | item = btrfs_item_ptr(eb, slot, | |
422 | struct btrfs_inode_item); | |
423 | btrfs_set_inode_nbytes(eb, item, nbytes); | |
d555438b JB |
424 | |
425 | /* | |
426 | * If this is a directory we need to reset the i_size to | |
427 | * 0 so that we can set it up properly when replaying | |
428 | * the rest of the items in this log. | |
429 | */ | |
430 | mode = btrfs_inode_mode(eb, item); | |
431 | if (S_ISDIR(mode)) | |
432 | btrfs_set_inode_size(eb, item, 0); | |
4bc4bee4 JB |
433 | } |
434 | } else if (inode_item) { | |
435 | struct btrfs_inode_item *item; | |
d555438b | 436 | u32 mode; |
4bc4bee4 JB |
437 | |
438 | /* | |
439 | * New inode, set nbytes to 0 so that the nbytes comes out | |
440 | * properly when we replay the extents. | |
441 | */ | |
442 | item = btrfs_item_ptr(eb, slot, struct btrfs_inode_item); | |
443 | btrfs_set_inode_nbytes(eb, item, 0); | |
d555438b JB |
444 | |
445 | /* | |
446 | * If this is a directory we need to reset the i_size to 0 so | |
447 | * that we can set it up properly when replaying the rest of | |
448 | * the items in this log. | |
449 | */ | |
450 | mode = btrfs_inode_mode(eb, item); | |
451 | if (S_ISDIR(mode)) | |
452 | btrfs_set_inode_size(eb, item, 0); | |
e02119d5 CM |
453 | } |
454 | insert: | |
b3b4aa74 | 455 | btrfs_release_path(path); |
e02119d5 | 456 | /* try to insert the key into the destination tree */ |
df8d116f | 457 | path->skip_release_on_error = 1; |
e02119d5 CM |
458 | ret = btrfs_insert_empty_item(trans, root, path, |
459 | key, item_size); | |
df8d116f | 460 | path->skip_release_on_error = 0; |
e02119d5 CM |
461 | |
462 | /* make sure any existing item is the correct size */ | |
df8d116f | 463 | if (ret == -EEXIST || ret == -EOVERFLOW) { |
e02119d5 CM |
464 | u32 found_size; |
465 | found_size = btrfs_item_size_nr(path->nodes[0], | |
466 | path->slots[0]); | |
143bede5 | 467 | if (found_size > item_size) |
2ff7e61e | 468 | btrfs_truncate_item(fs_info, path, item_size, 1); |
143bede5 | 469 | else if (found_size < item_size) |
2ff7e61e | 470 | btrfs_extend_item(fs_info, path, |
143bede5 | 471 | item_size - found_size); |
e02119d5 | 472 | } else if (ret) { |
4a500fd1 | 473 | return ret; |
e02119d5 CM |
474 | } |
475 | dst_ptr = btrfs_item_ptr_offset(path->nodes[0], | |
476 | path->slots[0]); | |
477 | ||
478 | /* don't overwrite an existing inode if the generation number | |
479 | * was logged as zero. This is done when the tree logging code | |
480 | * is just logging an inode to make sure it exists after recovery. | |
481 | * | |
482 | * Also, don't overwrite i_size on directories during replay. | |
483 | * log replay inserts and removes directory items based on the | |
484 | * state of the tree found in the subvolume, and i_size is modified | |
485 | * as it goes | |
486 | */ | |
487 | if (key->type == BTRFS_INODE_ITEM_KEY && ret == -EEXIST) { | |
488 | struct btrfs_inode_item *src_item; | |
489 | struct btrfs_inode_item *dst_item; | |
490 | ||
491 | src_item = (struct btrfs_inode_item *)src_ptr; | |
492 | dst_item = (struct btrfs_inode_item *)dst_ptr; | |
493 | ||
1a4bcf47 FM |
494 | if (btrfs_inode_generation(eb, src_item) == 0) { |
495 | struct extent_buffer *dst_eb = path->nodes[0]; | |
2f2ff0ee | 496 | const u64 ino_size = btrfs_inode_size(eb, src_item); |
1a4bcf47 | 497 | |
2f2ff0ee FM |
498 | /* |
499 | * For regular files an ino_size == 0 is used only when | |
500 | * logging that an inode exists, as part of a directory | |
501 | * fsync, and the inode wasn't fsynced before. In this | |
502 | * case don't set the size of the inode in the fs/subvol | |
503 | * tree, otherwise we would be throwing valid data away. | |
504 | */ | |
1a4bcf47 | 505 | if (S_ISREG(btrfs_inode_mode(eb, src_item)) && |
2f2ff0ee FM |
506 | S_ISREG(btrfs_inode_mode(dst_eb, dst_item)) && |
507 | ino_size != 0) { | |
1a4bcf47 | 508 | struct btrfs_map_token token; |
1a4bcf47 FM |
509 | |
510 | btrfs_init_map_token(&token); | |
511 | btrfs_set_token_inode_size(dst_eb, dst_item, | |
512 | ino_size, &token); | |
513 | } | |
e02119d5 | 514 | goto no_copy; |
1a4bcf47 | 515 | } |
e02119d5 CM |
516 | |
517 | if (overwrite_root && | |
518 | S_ISDIR(btrfs_inode_mode(eb, src_item)) && | |
519 | S_ISDIR(btrfs_inode_mode(path->nodes[0], dst_item))) { | |
520 | save_old_i_size = 1; | |
521 | saved_i_size = btrfs_inode_size(path->nodes[0], | |
522 | dst_item); | |
523 | } | |
524 | } | |
525 | ||
526 | copy_extent_buffer(path->nodes[0], eb, dst_ptr, | |
527 | src_ptr, item_size); | |
528 | ||
529 | if (save_old_i_size) { | |
530 | struct btrfs_inode_item *dst_item; | |
531 | dst_item = (struct btrfs_inode_item *)dst_ptr; | |
532 | btrfs_set_inode_size(path->nodes[0], dst_item, saved_i_size); | |
533 | } | |
534 | ||
535 | /* make sure the generation is filled in */ | |
536 | if (key->type == BTRFS_INODE_ITEM_KEY) { | |
537 | struct btrfs_inode_item *dst_item; | |
538 | dst_item = (struct btrfs_inode_item *)dst_ptr; | |
539 | if (btrfs_inode_generation(path->nodes[0], dst_item) == 0) { | |
540 | btrfs_set_inode_generation(path->nodes[0], dst_item, | |
541 | trans->transid); | |
542 | } | |
543 | } | |
544 | no_copy: | |
545 | btrfs_mark_buffer_dirty(path->nodes[0]); | |
b3b4aa74 | 546 | btrfs_release_path(path); |
e02119d5 CM |
547 | return 0; |
548 | } | |
549 | ||
550 | /* | |
551 | * simple helper to read an inode off the disk from a given root | |
552 | * This can only be called for subvolume roots and not for the log | |
553 | */ | |
554 | static noinline struct inode *read_one_inode(struct btrfs_root *root, | |
555 | u64 objectid) | |
556 | { | |
5d4f98a2 | 557 | struct btrfs_key key; |
e02119d5 | 558 | struct inode *inode; |
e02119d5 | 559 | |
5d4f98a2 YZ |
560 | key.objectid = objectid; |
561 | key.type = BTRFS_INODE_ITEM_KEY; | |
562 | key.offset = 0; | |
73f73415 | 563 | inode = btrfs_iget(root->fs_info->sb, &key, root, NULL); |
2e19f1f9 | 564 | if (IS_ERR(inode)) |
5d4f98a2 | 565 | inode = NULL; |
e02119d5 CM |
566 | return inode; |
567 | } | |
568 | ||
569 | /* replays a single extent in 'eb' at 'slot' with 'key' into the | |
570 | * subvolume 'root'. path is released on entry and should be released | |
571 | * on exit. | |
572 | * | |
573 | * extents in the log tree have not been allocated out of the extent | |
574 | * tree yet. So, this completes the allocation, taking a reference | |
575 | * as required if the extent already exists or creating a new extent | |
576 | * if it isn't in the extent allocation tree yet. | |
577 | * | |
578 | * The extent is inserted into the file, dropping any existing extents | |
579 | * from the file that overlap the new one. | |
580 | */ | |
581 | static noinline int replay_one_extent(struct btrfs_trans_handle *trans, | |
582 | struct btrfs_root *root, | |
583 | struct btrfs_path *path, | |
584 | struct extent_buffer *eb, int slot, | |
585 | struct btrfs_key *key) | |
586 | { | |
0b246afa | 587 | struct btrfs_fs_info *fs_info = root->fs_info; |
e02119d5 | 588 | int found_type; |
e02119d5 | 589 | u64 extent_end; |
e02119d5 | 590 | u64 start = key->offset; |
4bc4bee4 | 591 | u64 nbytes = 0; |
e02119d5 CM |
592 | struct btrfs_file_extent_item *item; |
593 | struct inode *inode = NULL; | |
594 | unsigned long size; | |
595 | int ret = 0; | |
596 | ||
597 | item = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item); | |
598 | found_type = btrfs_file_extent_type(eb, item); | |
599 | ||
d899e052 | 600 | if (found_type == BTRFS_FILE_EXTENT_REG || |
4bc4bee4 JB |
601 | found_type == BTRFS_FILE_EXTENT_PREALLOC) { |
602 | nbytes = btrfs_file_extent_num_bytes(eb, item); | |
603 | extent_end = start + nbytes; | |
604 | ||
605 | /* | |
606 | * We don't add to the inodes nbytes if we are prealloc or a | |
607 | * hole. | |
608 | */ | |
609 | if (btrfs_file_extent_disk_bytenr(eb, item) == 0) | |
610 | nbytes = 0; | |
611 | } else if (found_type == BTRFS_FILE_EXTENT_INLINE) { | |
e41ca589 | 612 | size = btrfs_file_extent_ram_bytes(eb, item); |
4bc4bee4 | 613 | nbytes = btrfs_file_extent_ram_bytes(eb, item); |
da17066c | 614 | extent_end = ALIGN(start + size, |
0b246afa | 615 | fs_info->sectorsize); |
e02119d5 CM |
616 | } else { |
617 | ret = 0; | |
618 | goto out; | |
619 | } | |
620 | ||
621 | inode = read_one_inode(root, key->objectid); | |
622 | if (!inode) { | |
623 | ret = -EIO; | |
624 | goto out; | |
625 | } | |
626 | ||
627 | /* | |
628 | * first check to see if we already have this extent in the | |
629 | * file. This must be done before the btrfs_drop_extents run | |
630 | * so we don't try to drop this extent. | |
631 | */ | |
f85b7379 DS |
632 | ret = btrfs_lookup_file_extent(trans, root, path, |
633 | btrfs_ino(BTRFS_I(inode)), start, 0); | |
e02119d5 | 634 | |
d899e052 YZ |
635 | if (ret == 0 && |
636 | (found_type == BTRFS_FILE_EXTENT_REG || | |
637 | found_type == BTRFS_FILE_EXTENT_PREALLOC)) { | |
e02119d5 CM |
638 | struct btrfs_file_extent_item cmp1; |
639 | struct btrfs_file_extent_item cmp2; | |
640 | struct btrfs_file_extent_item *existing; | |
641 | struct extent_buffer *leaf; | |
642 | ||
643 | leaf = path->nodes[0]; | |
644 | existing = btrfs_item_ptr(leaf, path->slots[0], | |
645 | struct btrfs_file_extent_item); | |
646 | ||
647 | read_extent_buffer(eb, &cmp1, (unsigned long)item, | |
648 | sizeof(cmp1)); | |
649 | read_extent_buffer(leaf, &cmp2, (unsigned long)existing, | |
650 | sizeof(cmp2)); | |
651 | ||
652 | /* | |
653 | * we already have a pointer to this exact extent, | |
654 | * we don't have to do anything | |
655 | */ | |
656 | if (memcmp(&cmp1, &cmp2, sizeof(cmp1)) == 0) { | |
b3b4aa74 | 657 | btrfs_release_path(path); |
e02119d5 CM |
658 | goto out; |
659 | } | |
660 | } | |
b3b4aa74 | 661 | btrfs_release_path(path); |
e02119d5 CM |
662 | |
663 | /* drop any overlapping extents */ | |
2671485d | 664 | ret = btrfs_drop_extents(trans, root, inode, start, extent_end, 1); |
3650860b JB |
665 | if (ret) |
666 | goto out; | |
e02119d5 | 667 | |
07d400a6 YZ |
668 | if (found_type == BTRFS_FILE_EXTENT_REG || |
669 | found_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
5d4f98a2 | 670 | u64 offset; |
07d400a6 YZ |
671 | unsigned long dest_offset; |
672 | struct btrfs_key ins; | |
673 | ||
3168021c FM |
674 | if (btrfs_file_extent_disk_bytenr(eb, item) == 0 && |
675 | btrfs_fs_incompat(fs_info, NO_HOLES)) | |
676 | goto update_inode; | |
677 | ||
07d400a6 YZ |
678 | ret = btrfs_insert_empty_item(trans, root, path, key, |
679 | sizeof(*item)); | |
3650860b JB |
680 | if (ret) |
681 | goto out; | |
07d400a6 YZ |
682 | dest_offset = btrfs_item_ptr_offset(path->nodes[0], |
683 | path->slots[0]); | |
684 | copy_extent_buffer(path->nodes[0], eb, dest_offset, | |
685 | (unsigned long)item, sizeof(*item)); | |
686 | ||
687 | ins.objectid = btrfs_file_extent_disk_bytenr(eb, item); | |
688 | ins.offset = btrfs_file_extent_disk_num_bytes(eb, item); | |
689 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
5d4f98a2 | 690 | offset = key->offset - btrfs_file_extent_offset(eb, item); |
07d400a6 | 691 | |
df2c95f3 QW |
692 | /* |
693 | * Manually record dirty extent, as here we did a shallow | |
694 | * file extent item copy and skip normal backref update, | |
695 | * but modifying extent tree all by ourselves. | |
696 | * So need to manually record dirty extent for qgroup, | |
697 | * as the owner of the file extent changed from log tree | |
698 | * (doesn't affect qgroup) to fs/file tree(affects qgroup) | |
699 | */ | |
a95f3aaf | 700 | ret = btrfs_qgroup_trace_extent(trans, |
df2c95f3 QW |
701 | btrfs_file_extent_disk_bytenr(eb, item), |
702 | btrfs_file_extent_disk_num_bytes(eb, item), | |
703 | GFP_NOFS); | |
704 | if (ret < 0) | |
705 | goto out; | |
706 | ||
07d400a6 YZ |
707 | if (ins.objectid > 0) { |
708 | u64 csum_start; | |
709 | u64 csum_end; | |
710 | LIST_HEAD(ordered_sums); | |
711 | /* | |
712 | * is this extent already allocated in the extent | |
713 | * allocation tree? If so, just add a reference | |
714 | */ | |
2ff7e61e | 715 | ret = btrfs_lookup_data_extent(fs_info, ins.objectid, |
07d400a6 YZ |
716 | ins.offset); |
717 | if (ret == 0) { | |
84f7d8e6 | 718 | ret = btrfs_inc_extent_ref(trans, root, |
07d400a6 | 719 | ins.objectid, ins.offset, |
5d4f98a2 | 720 | 0, root->root_key.objectid, |
b06c4bf5 | 721 | key->objectid, offset); |
b50c6e25 JB |
722 | if (ret) |
723 | goto out; | |
07d400a6 YZ |
724 | } else { |
725 | /* | |
726 | * insert the extent pointer in the extent | |
727 | * allocation tree | |
728 | */ | |
5d4f98a2 | 729 | ret = btrfs_alloc_logged_file_extent(trans, |
2ff7e61e | 730 | root->root_key.objectid, |
5d4f98a2 | 731 | key->objectid, offset, &ins); |
b50c6e25 JB |
732 | if (ret) |
733 | goto out; | |
07d400a6 | 734 | } |
b3b4aa74 | 735 | btrfs_release_path(path); |
07d400a6 YZ |
736 | |
737 | if (btrfs_file_extent_compression(eb, item)) { | |
738 | csum_start = ins.objectid; | |
739 | csum_end = csum_start + ins.offset; | |
740 | } else { | |
741 | csum_start = ins.objectid + | |
742 | btrfs_file_extent_offset(eb, item); | |
743 | csum_end = csum_start + | |
744 | btrfs_file_extent_num_bytes(eb, item); | |
745 | } | |
746 | ||
747 | ret = btrfs_lookup_csums_range(root->log_root, | |
748 | csum_start, csum_end - 1, | |
a2de733c | 749 | &ordered_sums, 0); |
3650860b JB |
750 | if (ret) |
751 | goto out; | |
b84b8390 FM |
752 | /* |
753 | * Now delete all existing cums in the csum root that | |
754 | * cover our range. We do this because we can have an | |
755 | * extent that is completely referenced by one file | |
756 | * extent item and partially referenced by another | |
757 | * file extent item (like after using the clone or | |
758 | * extent_same ioctls). In this case if we end up doing | |
759 | * the replay of the one that partially references the | |
760 | * extent first, and we do not do the csum deletion | |
761 | * below, we can get 2 csum items in the csum tree that | |
762 | * overlap each other. For example, imagine our log has | |
763 | * the two following file extent items: | |
764 | * | |
765 | * key (257 EXTENT_DATA 409600) | |
766 | * extent data disk byte 12845056 nr 102400 | |
767 | * extent data offset 20480 nr 20480 ram 102400 | |
768 | * | |
769 | * key (257 EXTENT_DATA 819200) | |
770 | * extent data disk byte 12845056 nr 102400 | |
771 | * extent data offset 0 nr 102400 ram 102400 | |
772 | * | |
773 | * Where the second one fully references the 100K extent | |
774 | * that starts at disk byte 12845056, and the log tree | |
775 | * has a single csum item that covers the entire range | |
776 | * of the extent: | |
777 | * | |
778 | * key (EXTENT_CSUM EXTENT_CSUM 12845056) itemsize 100 | |
779 | * | |
780 | * After the first file extent item is replayed, the | |
781 | * csum tree gets the following csum item: | |
782 | * | |
783 | * key (EXTENT_CSUM EXTENT_CSUM 12865536) itemsize 20 | |
784 | * | |
785 | * Which covers the 20K sub-range starting at offset 20K | |
786 | * of our extent. Now when we replay the second file | |
787 | * extent item, if we do not delete existing csum items | |
788 | * that cover any of its blocks, we end up getting two | |
789 | * csum items in our csum tree that overlap each other: | |
790 | * | |
791 | * key (EXTENT_CSUM EXTENT_CSUM 12845056) itemsize 100 | |
792 | * key (EXTENT_CSUM EXTENT_CSUM 12865536) itemsize 20 | |
793 | * | |
794 | * Which is a problem, because after this anyone trying | |
795 | * to lookup up for the checksum of any block of our | |
796 | * extent starting at an offset of 40K or higher, will | |
797 | * end up looking at the second csum item only, which | |
798 | * does not contain the checksum for any block starting | |
799 | * at offset 40K or higher of our extent. | |
800 | */ | |
07d400a6 YZ |
801 | while (!list_empty(&ordered_sums)) { |
802 | struct btrfs_ordered_sum *sums; | |
803 | sums = list_entry(ordered_sums.next, | |
804 | struct btrfs_ordered_sum, | |
805 | list); | |
b84b8390 | 806 | if (!ret) |
0b246afa | 807 | ret = btrfs_del_csums(trans, fs_info, |
5b4aacef JM |
808 | sums->bytenr, |
809 | sums->len); | |
3650860b JB |
810 | if (!ret) |
811 | ret = btrfs_csum_file_blocks(trans, | |
0b246afa | 812 | fs_info->csum_root, sums); |
07d400a6 YZ |
813 | list_del(&sums->list); |
814 | kfree(sums); | |
815 | } | |
3650860b JB |
816 | if (ret) |
817 | goto out; | |
07d400a6 | 818 | } else { |
b3b4aa74 | 819 | btrfs_release_path(path); |
07d400a6 YZ |
820 | } |
821 | } else if (found_type == BTRFS_FILE_EXTENT_INLINE) { | |
822 | /* inline extents are easy, we just overwrite them */ | |
823 | ret = overwrite_item(trans, root, path, eb, slot, key); | |
3650860b JB |
824 | if (ret) |
825 | goto out; | |
07d400a6 | 826 | } |
e02119d5 | 827 | |
4bc4bee4 | 828 | inode_add_bytes(inode, nbytes); |
3168021c | 829 | update_inode: |
b9959295 | 830 | ret = btrfs_update_inode(trans, root, inode); |
e02119d5 CM |
831 | out: |
832 | if (inode) | |
833 | iput(inode); | |
834 | return ret; | |
835 | } | |
836 | ||
837 | /* | |
838 | * when cleaning up conflicts between the directory names in the | |
839 | * subvolume, directory names in the log and directory names in the | |
840 | * inode back references, we may have to unlink inodes from directories. | |
841 | * | |
842 | * This is a helper function to do the unlink of a specific directory | |
843 | * item | |
844 | */ | |
845 | static noinline int drop_one_dir_item(struct btrfs_trans_handle *trans, | |
846 | struct btrfs_root *root, | |
847 | struct btrfs_path *path, | |
207e7d92 | 848 | struct btrfs_inode *dir, |
e02119d5 CM |
849 | struct btrfs_dir_item *di) |
850 | { | |
851 | struct inode *inode; | |
852 | char *name; | |
853 | int name_len; | |
854 | struct extent_buffer *leaf; | |
855 | struct btrfs_key location; | |
856 | int ret; | |
857 | ||
858 | leaf = path->nodes[0]; | |
859 | ||
860 | btrfs_dir_item_key_to_cpu(leaf, di, &location); | |
861 | name_len = btrfs_dir_name_len(leaf, di); | |
862 | name = kmalloc(name_len, GFP_NOFS); | |
2a29edc6 | 863 | if (!name) |
864 | return -ENOMEM; | |
865 | ||
e02119d5 | 866 | read_extent_buffer(leaf, name, (unsigned long)(di + 1), name_len); |
b3b4aa74 | 867 | btrfs_release_path(path); |
e02119d5 CM |
868 | |
869 | inode = read_one_inode(root, location.objectid); | |
c00e9493 | 870 | if (!inode) { |
3650860b JB |
871 | ret = -EIO; |
872 | goto out; | |
c00e9493 | 873 | } |
e02119d5 | 874 | |
ec051c0f | 875 | ret = link_to_fixup_dir(trans, root, path, location.objectid); |
3650860b JB |
876 | if (ret) |
877 | goto out; | |
12fcfd22 | 878 | |
207e7d92 NB |
879 | ret = btrfs_unlink_inode(trans, root, dir, BTRFS_I(inode), name, |
880 | name_len); | |
3650860b JB |
881 | if (ret) |
882 | goto out; | |
ada9af21 | 883 | else |
e5c304e6 | 884 | ret = btrfs_run_delayed_items(trans); |
3650860b | 885 | out: |
e02119d5 | 886 | kfree(name); |
e02119d5 CM |
887 | iput(inode); |
888 | return ret; | |
889 | } | |
890 | ||
891 | /* | |
892 | * helper function to see if a given name and sequence number found | |
893 | * in an inode back reference are already in a directory and correctly | |
894 | * point to this inode | |
895 | */ | |
896 | static noinline int inode_in_dir(struct btrfs_root *root, | |
897 | struct btrfs_path *path, | |
898 | u64 dirid, u64 objectid, u64 index, | |
899 | const char *name, int name_len) | |
900 | { | |
901 | struct btrfs_dir_item *di; | |
902 | struct btrfs_key location; | |
903 | int match = 0; | |
904 | ||
905 | di = btrfs_lookup_dir_index_item(NULL, root, path, dirid, | |
906 | index, name, name_len, 0); | |
907 | if (di && !IS_ERR(di)) { | |
908 | btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location); | |
909 | if (location.objectid != objectid) | |
910 | goto out; | |
911 | } else | |
912 | goto out; | |
b3b4aa74 | 913 | btrfs_release_path(path); |
e02119d5 CM |
914 | |
915 | di = btrfs_lookup_dir_item(NULL, root, path, dirid, name, name_len, 0); | |
916 | if (di && !IS_ERR(di)) { | |
917 | btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location); | |
918 | if (location.objectid != objectid) | |
919 | goto out; | |
920 | } else | |
921 | goto out; | |
922 | match = 1; | |
923 | out: | |
b3b4aa74 | 924 | btrfs_release_path(path); |
e02119d5 CM |
925 | return match; |
926 | } | |
927 | ||
928 | /* | |
929 | * helper function to check a log tree for a named back reference in | |
930 | * an inode. This is used to decide if a back reference that is | |
931 | * found in the subvolume conflicts with what we find in the log. | |
932 | * | |
933 | * inode backreferences may have multiple refs in a single item, | |
934 | * during replay we process one reference at a time, and we don't | |
935 | * want to delete valid links to a file from the subvolume if that | |
936 | * link is also in the log. | |
937 | */ | |
938 | static noinline int backref_in_log(struct btrfs_root *log, | |
939 | struct btrfs_key *key, | |
f186373f | 940 | u64 ref_objectid, |
df8d116f | 941 | const char *name, int namelen) |
e02119d5 CM |
942 | { |
943 | struct btrfs_path *path; | |
944 | struct btrfs_inode_ref *ref; | |
945 | unsigned long ptr; | |
946 | unsigned long ptr_end; | |
947 | unsigned long name_ptr; | |
948 | int found_name_len; | |
949 | int item_size; | |
950 | int ret; | |
951 | int match = 0; | |
952 | ||
953 | path = btrfs_alloc_path(); | |
2a29edc6 | 954 | if (!path) |
955 | return -ENOMEM; | |
956 | ||
e02119d5 CM |
957 | ret = btrfs_search_slot(NULL, log, key, path, 0, 0); |
958 | if (ret != 0) | |
959 | goto out; | |
960 | ||
e02119d5 | 961 | ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]); |
f186373f MF |
962 | |
963 | if (key->type == BTRFS_INODE_EXTREF_KEY) { | |
1f250e92 FM |
964 | if (btrfs_find_name_in_ext_backref(path->nodes[0], |
965 | path->slots[0], | |
966 | ref_objectid, | |
f186373f MF |
967 | name, namelen, NULL)) |
968 | match = 1; | |
969 | ||
970 | goto out; | |
971 | } | |
972 | ||
973 | item_size = btrfs_item_size_nr(path->nodes[0], path->slots[0]); | |
e02119d5 CM |
974 | ptr_end = ptr + item_size; |
975 | while (ptr < ptr_end) { | |
976 | ref = (struct btrfs_inode_ref *)ptr; | |
977 | found_name_len = btrfs_inode_ref_name_len(path->nodes[0], ref); | |
978 | if (found_name_len == namelen) { | |
979 | name_ptr = (unsigned long)(ref + 1); | |
980 | ret = memcmp_extent_buffer(path->nodes[0], name, | |
981 | name_ptr, namelen); | |
982 | if (ret == 0) { | |
983 | match = 1; | |
984 | goto out; | |
985 | } | |
986 | } | |
987 | ptr = (unsigned long)(ref + 1) + found_name_len; | |
988 | } | |
989 | out: | |
990 | btrfs_free_path(path); | |
991 | return match; | |
992 | } | |
993 | ||
5a1d7843 | 994 | static inline int __add_inode_ref(struct btrfs_trans_handle *trans, |
e02119d5 | 995 | struct btrfs_root *root, |
e02119d5 | 996 | struct btrfs_path *path, |
5a1d7843 | 997 | struct btrfs_root *log_root, |
94c91a1f NB |
998 | struct btrfs_inode *dir, |
999 | struct btrfs_inode *inode, | |
f186373f MF |
1000 | u64 inode_objectid, u64 parent_objectid, |
1001 | u64 ref_index, char *name, int namelen, | |
1002 | int *search_done) | |
e02119d5 | 1003 | { |
34f3e4f2 | 1004 | int ret; |
f186373f MF |
1005 | char *victim_name; |
1006 | int victim_name_len; | |
1007 | struct extent_buffer *leaf; | |
5a1d7843 | 1008 | struct btrfs_dir_item *di; |
f186373f MF |
1009 | struct btrfs_key search_key; |
1010 | struct btrfs_inode_extref *extref; | |
c622ae60 | 1011 | |
f186373f MF |
1012 | again: |
1013 | /* Search old style refs */ | |
1014 | search_key.objectid = inode_objectid; | |
1015 | search_key.type = BTRFS_INODE_REF_KEY; | |
1016 | search_key.offset = parent_objectid; | |
1017 | ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0); | |
e02119d5 | 1018 | if (ret == 0) { |
e02119d5 CM |
1019 | struct btrfs_inode_ref *victim_ref; |
1020 | unsigned long ptr; | |
1021 | unsigned long ptr_end; | |
f186373f MF |
1022 | |
1023 | leaf = path->nodes[0]; | |
e02119d5 CM |
1024 | |
1025 | /* are we trying to overwrite a back ref for the root directory | |
1026 | * if so, just jump out, we're done | |
1027 | */ | |
f186373f | 1028 | if (search_key.objectid == search_key.offset) |
5a1d7843 | 1029 | return 1; |
e02119d5 CM |
1030 | |
1031 | /* check all the names in this back reference to see | |
1032 | * if they are in the log. if so, we allow them to stay | |
1033 | * otherwise they must be unlinked as a conflict | |
1034 | */ | |
1035 | ptr = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
1036 | ptr_end = ptr + btrfs_item_size_nr(leaf, path->slots[0]); | |
d397712b | 1037 | while (ptr < ptr_end) { |
e02119d5 CM |
1038 | victim_ref = (struct btrfs_inode_ref *)ptr; |
1039 | victim_name_len = btrfs_inode_ref_name_len(leaf, | |
1040 | victim_ref); | |
1041 | victim_name = kmalloc(victim_name_len, GFP_NOFS); | |
3650860b JB |
1042 | if (!victim_name) |
1043 | return -ENOMEM; | |
e02119d5 CM |
1044 | |
1045 | read_extent_buffer(leaf, victim_name, | |
1046 | (unsigned long)(victim_ref + 1), | |
1047 | victim_name_len); | |
1048 | ||
f186373f MF |
1049 | if (!backref_in_log(log_root, &search_key, |
1050 | parent_objectid, | |
1051 | victim_name, | |
e02119d5 | 1052 | victim_name_len)) { |
94c91a1f | 1053 | inc_nlink(&inode->vfs_inode); |
b3b4aa74 | 1054 | btrfs_release_path(path); |
12fcfd22 | 1055 | |
94c91a1f | 1056 | ret = btrfs_unlink_inode(trans, root, dir, inode, |
4ec5934e | 1057 | victim_name, victim_name_len); |
f186373f | 1058 | kfree(victim_name); |
3650860b JB |
1059 | if (ret) |
1060 | return ret; | |
e5c304e6 | 1061 | ret = btrfs_run_delayed_items(trans); |
ada9af21 FDBM |
1062 | if (ret) |
1063 | return ret; | |
f186373f MF |
1064 | *search_done = 1; |
1065 | goto again; | |
e02119d5 CM |
1066 | } |
1067 | kfree(victim_name); | |
f186373f | 1068 | |
e02119d5 CM |
1069 | ptr = (unsigned long)(victim_ref + 1) + victim_name_len; |
1070 | } | |
e02119d5 | 1071 | |
c622ae60 | 1072 | /* |
1073 | * NOTE: we have searched root tree and checked the | |
bb7ab3b9 | 1074 | * corresponding ref, it does not need to check again. |
c622ae60 | 1075 | */ |
5a1d7843 | 1076 | *search_done = 1; |
e02119d5 | 1077 | } |
b3b4aa74 | 1078 | btrfs_release_path(path); |
e02119d5 | 1079 | |
f186373f MF |
1080 | /* Same search but for extended refs */ |
1081 | extref = btrfs_lookup_inode_extref(NULL, root, path, name, namelen, | |
1082 | inode_objectid, parent_objectid, 0, | |
1083 | 0); | |
1084 | if (!IS_ERR_OR_NULL(extref)) { | |
1085 | u32 item_size; | |
1086 | u32 cur_offset = 0; | |
1087 | unsigned long base; | |
1088 | struct inode *victim_parent; | |
1089 | ||
1090 | leaf = path->nodes[0]; | |
1091 | ||
1092 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1093 | base = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
1094 | ||
1095 | while (cur_offset < item_size) { | |
dd9ef135 | 1096 | extref = (struct btrfs_inode_extref *)(base + cur_offset); |
f186373f MF |
1097 | |
1098 | victim_name_len = btrfs_inode_extref_name_len(leaf, extref); | |
1099 | ||
1100 | if (btrfs_inode_extref_parent(leaf, extref) != parent_objectid) | |
1101 | goto next; | |
1102 | ||
1103 | victim_name = kmalloc(victim_name_len, GFP_NOFS); | |
3650860b JB |
1104 | if (!victim_name) |
1105 | return -ENOMEM; | |
f186373f MF |
1106 | read_extent_buffer(leaf, victim_name, (unsigned long)&extref->name, |
1107 | victim_name_len); | |
1108 | ||
1109 | search_key.objectid = inode_objectid; | |
1110 | search_key.type = BTRFS_INODE_EXTREF_KEY; | |
1111 | search_key.offset = btrfs_extref_hash(parent_objectid, | |
1112 | victim_name, | |
1113 | victim_name_len); | |
1114 | ret = 0; | |
1115 | if (!backref_in_log(log_root, &search_key, | |
1116 | parent_objectid, victim_name, | |
1117 | victim_name_len)) { | |
1118 | ret = -ENOENT; | |
1119 | victim_parent = read_one_inode(root, | |
94c91a1f | 1120 | parent_objectid); |
f186373f | 1121 | if (victim_parent) { |
94c91a1f | 1122 | inc_nlink(&inode->vfs_inode); |
f186373f MF |
1123 | btrfs_release_path(path); |
1124 | ||
1125 | ret = btrfs_unlink_inode(trans, root, | |
4ec5934e | 1126 | BTRFS_I(victim_parent), |
94c91a1f | 1127 | inode, |
4ec5934e NB |
1128 | victim_name, |
1129 | victim_name_len); | |
ada9af21 FDBM |
1130 | if (!ret) |
1131 | ret = btrfs_run_delayed_items( | |
e5c304e6 | 1132 | trans); |
f186373f | 1133 | } |
f186373f MF |
1134 | iput(victim_parent); |
1135 | kfree(victim_name); | |
3650860b JB |
1136 | if (ret) |
1137 | return ret; | |
f186373f MF |
1138 | *search_done = 1; |
1139 | goto again; | |
1140 | } | |
1141 | kfree(victim_name); | |
f186373f MF |
1142 | next: |
1143 | cur_offset += victim_name_len + sizeof(*extref); | |
1144 | } | |
1145 | *search_done = 1; | |
1146 | } | |
1147 | btrfs_release_path(path); | |
1148 | ||
34f3e4f2 | 1149 | /* look for a conflicting sequence number */ |
94c91a1f | 1150 | di = btrfs_lookup_dir_index_item(trans, root, path, btrfs_ino(dir), |
f186373f | 1151 | ref_index, name, namelen, 0); |
34f3e4f2 | 1152 | if (di && !IS_ERR(di)) { |
94c91a1f | 1153 | ret = drop_one_dir_item(trans, root, path, dir, di); |
3650860b JB |
1154 | if (ret) |
1155 | return ret; | |
34f3e4f2 | 1156 | } |
1157 | btrfs_release_path(path); | |
1158 | ||
52042d8e | 1159 | /* look for a conflicting name */ |
94c91a1f | 1160 | di = btrfs_lookup_dir_item(trans, root, path, btrfs_ino(dir), |
34f3e4f2 | 1161 | name, namelen, 0); |
1162 | if (di && !IS_ERR(di)) { | |
94c91a1f | 1163 | ret = drop_one_dir_item(trans, root, path, dir, di); |
3650860b JB |
1164 | if (ret) |
1165 | return ret; | |
34f3e4f2 | 1166 | } |
1167 | btrfs_release_path(path); | |
1168 | ||
5a1d7843 JS |
1169 | return 0; |
1170 | } | |
e02119d5 | 1171 | |
bae15d95 QW |
1172 | static int extref_get_fields(struct extent_buffer *eb, unsigned long ref_ptr, |
1173 | u32 *namelen, char **name, u64 *index, | |
1174 | u64 *parent_objectid) | |
f186373f MF |
1175 | { |
1176 | struct btrfs_inode_extref *extref; | |
1177 | ||
1178 | extref = (struct btrfs_inode_extref *)ref_ptr; | |
1179 | ||
1180 | *namelen = btrfs_inode_extref_name_len(eb, extref); | |
1181 | *name = kmalloc(*namelen, GFP_NOFS); | |
1182 | if (*name == NULL) | |
1183 | return -ENOMEM; | |
1184 | ||
1185 | read_extent_buffer(eb, *name, (unsigned long)&extref->name, | |
1186 | *namelen); | |
1187 | ||
1f250e92 FM |
1188 | if (index) |
1189 | *index = btrfs_inode_extref_index(eb, extref); | |
f186373f MF |
1190 | if (parent_objectid) |
1191 | *parent_objectid = btrfs_inode_extref_parent(eb, extref); | |
1192 | ||
1193 | return 0; | |
1194 | } | |
1195 | ||
bae15d95 QW |
1196 | static int ref_get_fields(struct extent_buffer *eb, unsigned long ref_ptr, |
1197 | u32 *namelen, char **name, u64 *index) | |
f186373f MF |
1198 | { |
1199 | struct btrfs_inode_ref *ref; | |
1200 | ||
1201 | ref = (struct btrfs_inode_ref *)ref_ptr; | |
1202 | ||
1203 | *namelen = btrfs_inode_ref_name_len(eb, ref); | |
1204 | *name = kmalloc(*namelen, GFP_NOFS); | |
1205 | if (*name == NULL) | |
1206 | return -ENOMEM; | |
1207 | ||
1208 | read_extent_buffer(eb, *name, (unsigned long)(ref + 1), *namelen); | |
1209 | ||
1f250e92 FM |
1210 | if (index) |
1211 | *index = btrfs_inode_ref_index(eb, ref); | |
f186373f MF |
1212 | |
1213 | return 0; | |
1214 | } | |
1215 | ||
1f250e92 FM |
1216 | /* |
1217 | * Take an inode reference item from the log tree and iterate all names from the | |
1218 | * inode reference item in the subvolume tree with the same key (if it exists). | |
1219 | * For any name that is not in the inode reference item from the log tree, do a | |
1220 | * proper unlink of that name (that is, remove its entry from the inode | |
1221 | * reference item and both dir index keys). | |
1222 | */ | |
1223 | static int unlink_old_inode_refs(struct btrfs_trans_handle *trans, | |
1224 | struct btrfs_root *root, | |
1225 | struct btrfs_path *path, | |
1226 | struct btrfs_inode *inode, | |
1227 | struct extent_buffer *log_eb, | |
1228 | int log_slot, | |
1229 | struct btrfs_key *key) | |
1230 | { | |
1231 | int ret; | |
1232 | unsigned long ref_ptr; | |
1233 | unsigned long ref_end; | |
1234 | struct extent_buffer *eb; | |
1235 | ||
1236 | again: | |
1237 | btrfs_release_path(path); | |
1238 | ret = btrfs_search_slot(NULL, root, key, path, 0, 0); | |
1239 | if (ret > 0) { | |
1240 | ret = 0; | |
1241 | goto out; | |
1242 | } | |
1243 | if (ret < 0) | |
1244 | goto out; | |
1245 | ||
1246 | eb = path->nodes[0]; | |
1247 | ref_ptr = btrfs_item_ptr_offset(eb, path->slots[0]); | |
1248 | ref_end = ref_ptr + btrfs_item_size_nr(eb, path->slots[0]); | |
1249 | while (ref_ptr < ref_end) { | |
1250 | char *name = NULL; | |
1251 | int namelen; | |
1252 | u64 parent_id; | |
1253 | ||
1254 | if (key->type == BTRFS_INODE_EXTREF_KEY) { | |
1255 | ret = extref_get_fields(eb, ref_ptr, &namelen, &name, | |
1256 | NULL, &parent_id); | |
1257 | } else { | |
1258 | parent_id = key->offset; | |
1259 | ret = ref_get_fields(eb, ref_ptr, &namelen, &name, | |
1260 | NULL); | |
1261 | } | |
1262 | if (ret) | |
1263 | goto out; | |
1264 | ||
1265 | if (key->type == BTRFS_INODE_EXTREF_KEY) | |
1266 | ret = btrfs_find_name_in_ext_backref(log_eb, log_slot, | |
1267 | parent_id, name, | |
1268 | namelen, NULL); | |
1269 | else | |
1270 | ret = btrfs_find_name_in_backref(log_eb, log_slot, name, | |
1271 | namelen, NULL); | |
1272 | ||
1273 | if (!ret) { | |
1274 | struct inode *dir; | |
1275 | ||
1276 | btrfs_release_path(path); | |
1277 | dir = read_one_inode(root, parent_id); | |
1278 | if (!dir) { | |
1279 | ret = -ENOENT; | |
1280 | kfree(name); | |
1281 | goto out; | |
1282 | } | |
1283 | ret = btrfs_unlink_inode(trans, root, BTRFS_I(dir), | |
1284 | inode, name, namelen); | |
1285 | kfree(name); | |
1286 | iput(dir); | |
1287 | if (ret) | |
1288 | goto out; | |
1289 | goto again; | |
1290 | } | |
1291 | ||
1292 | kfree(name); | |
1293 | ref_ptr += namelen; | |
1294 | if (key->type == BTRFS_INODE_EXTREF_KEY) | |
1295 | ref_ptr += sizeof(struct btrfs_inode_extref); | |
1296 | else | |
1297 | ref_ptr += sizeof(struct btrfs_inode_ref); | |
1298 | } | |
1299 | ret = 0; | |
1300 | out: | |
1301 | btrfs_release_path(path); | |
1302 | return ret; | |
1303 | } | |
1304 | ||
0d836392 FM |
1305 | static int btrfs_inode_ref_exists(struct inode *inode, struct inode *dir, |
1306 | const u8 ref_type, const char *name, | |
1307 | const int namelen) | |
1308 | { | |
1309 | struct btrfs_key key; | |
1310 | struct btrfs_path *path; | |
1311 | const u64 parent_id = btrfs_ino(BTRFS_I(dir)); | |
1312 | int ret; | |
1313 | ||
1314 | path = btrfs_alloc_path(); | |
1315 | if (!path) | |
1316 | return -ENOMEM; | |
1317 | ||
1318 | key.objectid = btrfs_ino(BTRFS_I(inode)); | |
1319 | key.type = ref_type; | |
1320 | if (key.type == BTRFS_INODE_REF_KEY) | |
1321 | key.offset = parent_id; | |
1322 | else | |
1323 | key.offset = btrfs_extref_hash(parent_id, name, namelen); | |
1324 | ||
1325 | ret = btrfs_search_slot(NULL, BTRFS_I(inode)->root, &key, path, 0, 0); | |
1326 | if (ret < 0) | |
1327 | goto out; | |
1328 | if (ret > 0) { | |
1329 | ret = 0; | |
1330 | goto out; | |
1331 | } | |
1332 | if (key.type == BTRFS_INODE_EXTREF_KEY) | |
1333 | ret = btrfs_find_name_in_ext_backref(path->nodes[0], | |
1334 | path->slots[0], parent_id, | |
1335 | name, namelen, NULL); | |
1336 | else | |
1337 | ret = btrfs_find_name_in_backref(path->nodes[0], path->slots[0], | |
1338 | name, namelen, NULL); | |
1339 | ||
1340 | out: | |
1341 | btrfs_free_path(path); | |
1342 | return ret; | |
1343 | } | |
1344 | ||
6b5fc433 FM |
1345 | static int add_link(struct btrfs_trans_handle *trans, struct btrfs_root *root, |
1346 | struct inode *dir, struct inode *inode, const char *name, | |
1347 | int namelen, u64 ref_index) | |
1348 | { | |
1349 | struct btrfs_dir_item *dir_item; | |
1350 | struct btrfs_key key; | |
1351 | struct btrfs_path *path; | |
1352 | struct inode *other_inode = NULL; | |
1353 | int ret; | |
1354 | ||
1355 | path = btrfs_alloc_path(); | |
1356 | if (!path) | |
1357 | return -ENOMEM; | |
1358 | ||
1359 | dir_item = btrfs_lookup_dir_item(NULL, root, path, | |
1360 | btrfs_ino(BTRFS_I(dir)), | |
1361 | name, namelen, 0); | |
1362 | if (!dir_item) { | |
1363 | btrfs_release_path(path); | |
1364 | goto add_link; | |
1365 | } else if (IS_ERR(dir_item)) { | |
1366 | ret = PTR_ERR(dir_item); | |
1367 | goto out; | |
1368 | } | |
1369 | ||
1370 | /* | |
1371 | * Our inode's dentry collides with the dentry of another inode which is | |
1372 | * in the log but not yet processed since it has a higher inode number. | |
1373 | * So delete that other dentry. | |
1374 | */ | |
1375 | btrfs_dir_item_key_to_cpu(path->nodes[0], dir_item, &key); | |
1376 | btrfs_release_path(path); | |
1377 | other_inode = read_one_inode(root, key.objectid); | |
1378 | if (!other_inode) { | |
1379 | ret = -ENOENT; | |
1380 | goto out; | |
1381 | } | |
1382 | ret = btrfs_unlink_inode(trans, root, BTRFS_I(dir), BTRFS_I(other_inode), | |
1383 | name, namelen); | |
1384 | if (ret) | |
1385 | goto out; | |
1386 | /* | |
1387 | * If we dropped the link count to 0, bump it so that later the iput() | |
1388 | * on the inode will not free it. We will fixup the link count later. | |
1389 | */ | |
1390 | if (other_inode->i_nlink == 0) | |
1391 | inc_nlink(other_inode); | |
1392 | ||
1393 | ret = btrfs_run_delayed_items(trans); | |
1394 | if (ret) | |
1395 | goto out; | |
1396 | add_link: | |
1397 | ret = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode), | |
1398 | name, namelen, 0, ref_index); | |
1399 | out: | |
1400 | iput(other_inode); | |
1401 | btrfs_free_path(path); | |
1402 | ||
1403 | return ret; | |
1404 | } | |
1405 | ||
5a1d7843 JS |
1406 | /* |
1407 | * replay one inode back reference item found in the log tree. | |
1408 | * eb, slot and key refer to the buffer and key found in the log tree. | |
1409 | * root is the destination we are replaying into, and path is for temp | |
1410 | * use by this function. (it should be released on return). | |
1411 | */ | |
1412 | static noinline int add_inode_ref(struct btrfs_trans_handle *trans, | |
1413 | struct btrfs_root *root, | |
1414 | struct btrfs_root *log, | |
1415 | struct btrfs_path *path, | |
1416 | struct extent_buffer *eb, int slot, | |
1417 | struct btrfs_key *key) | |
1418 | { | |
03b2f08b GB |
1419 | struct inode *dir = NULL; |
1420 | struct inode *inode = NULL; | |
5a1d7843 JS |
1421 | unsigned long ref_ptr; |
1422 | unsigned long ref_end; | |
03b2f08b | 1423 | char *name = NULL; |
5a1d7843 JS |
1424 | int namelen; |
1425 | int ret; | |
1426 | int search_done = 0; | |
f186373f MF |
1427 | int log_ref_ver = 0; |
1428 | u64 parent_objectid; | |
1429 | u64 inode_objectid; | |
f46dbe3d | 1430 | u64 ref_index = 0; |
f186373f MF |
1431 | int ref_struct_size; |
1432 | ||
1433 | ref_ptr = btrfs_item_ptr_offset(eb, slot); | |
1434 | ref_end = ref_ptr + btrfs_item_size_nr(eb, slot); | |
1435 | ||
1436 | if (key->type == BTRFS_INODE_EXTREF_KEY) { | |
1437 | struct btrfs_inode_extref *r; | |
1438 | ||
1439 | ref_struct_size = sizeof(struct btrfs_inode_extref); | |
1440 | log_ref_ver = 1; | |
1441 | r = (struct btrfs_inode_extref *)ref_ptr; | |
1442 | parent_objectid = btrfs_inode_extref_parent(eb, r); | |
1443 | } else { | |
1444 | ref_struct_size = sizeof(struct btrfs_inode_ref); | |
1445 | parent_objectid = key->offset; | |
1446 | } | |
1447 | inode_objectid = key->objectid; | |
e02119d5 | 1448 | |
5a1d7843 JS |
1449 | /* |
1450 | * it is possible that we didn't log all the parent directories | |
1451 | * for a given inode. If we don't find the dir, just don't | |
1452 | * copy the back ref in. The link count fixup code will take | |
1453 | * care of the rest | |
1454 | */ | |
f186373f | 1455 | dir = read_one_inode(root, parent_objectid); |
03b2f08b GB |
1456 | if (!dir) { |
1457 | ret = -ENOENT; | |
1458 | goto out; | |
1459 | } | |
5a1d7843 | 1460 | |
f186373f | 1461 | inode = read_one_inode(root, inode_objectid); |
5a1d7843 | 1462 | if (!inode) { |
03b2f08b GB |
1463 | ret = -EIO; |
1464 | goto out; | |
5a1d7843 JS |
1465 | } |
1466 | ||
5a1d7843 | 1467 | while (ref_ptr < ref_end) { |
f186373f | 1468 | if (log_ref_ver) { |
bae15d95 QW |
1469 | ret = extref_get_fields(eb, ref_ptr, &namelen, &name, |
1470 | &ref_index, &parent_objectid); | |
f186373f MF |
1471 | /* |
1472 | * parent object can change from one array | |
1473 | * item to another. | |
1474 | */ | |
1475 | if (!dir) | |
1476 | dir = read_one_inode(root, parent_objectid); | |
03b2f08b GB |
1477 | if (!dir) { |
1478 | ret = -ENOENT; | |
1479 | goto out; | |
1480 | } | |
f186373f | 1481 | } else { |
bae15d95 QW |
1482 | ret = ref_get_fields(eb, ref_ptr, &namelen, &name, |
1483 | &ref_index); | |
f186373f MF |
1484 | } |
1485 | if (ret) | |
03b2f08b | 1486 | goto out; |
5a1d7843 JS |
1487 | |
1488 | /* if we already have a perfect match, we're done */ | |
f85b7379 DS |
1489 | if (!inode_in_dir(root, path, btrfs_ino(BTRFS_I(dir)), |
1490 | btrfs_ino(BTRFS_I(inode)), ref_index, | |
1491 | name, namelen)) { | |
5a1d7843 JS |
1492 | /* |
1493 | * look for a conflicting back reference in the | |
1494 | * metadata. if we find one we have to unlink that name | |
1495 | * of the file before we add our new link. Later on, we | |
1496 | * overwrite any existing back reference, and we don't | |
1497 | * want to create dangling pointers in the directory. | |
1498 | */ | |
1499 | ||
1500 | if (!search_done) { | |
1501 | ret = __add_inode_ref(trans, root, path, log, | |
94c91a1f | 1502 | BTRFS_I(dir), |
d75eefdf | 1503 | BTRFS_I(inode), |
f186373f MF |
1504 | inode_objectid, |
1505 | parent_objectid, | |
1506 | ref_index, name, namelen, | |
5a1d7843 | 1507 | &search_done); |
03b2f08b GB |
1508 | if (ret) { |
1509 | if (ret == 1) | |
1510 | ret = 0; | |
3650860b JB |
1511 | goto out; |
1512 | } | |
5a1d7843 JS |
1513 | } |
1514 | ||
0d836392 FM |
1515 | /* |
1516 | * If a reference item already exists for this inode | |
1517 | * with the same parent and name, but different index, | |
1518 | * drop it and the corresponding directory index entries | |
1519 | * from the parent before adding the new reference item | |
1520 | * and dir index entries, otherwise we would fail with | |
1521 | * -EEXIST returned from btrfs_add_link() below. | |
1522 | */ | |
1523 | ret = btrfs_inode_ref_exists(inode, dir, key->type, | |
1524 | name, namelen); | |
1525 | if (ret > 0) { | |
1526 | ret = btrfs_unlink_inode(trans, root, | |
1527 | BTRFS_I(dir), | |
1528 | BTRFS_I(inode), | |
1529 | name, namelen); | |
1530 | /* | |
1531 | * If we dropped the link count to 0, bump it so | |
1532 | * that later the iput() on the inode will not | |
1533 | * free it. We will fixup the link count later. | |
1534 | */ | |
1535 | if (!ret && inode->i_nlink == 0) | |
1536 | inc_nlink(inode); | |
1537 | } | |
1538 | if (ret < 0) | |
1539 | goto out; | |
1540 | ||
5a1d7843 | 1541 | /* insert our name */ |
6b5fc433 FM |
1542 | ret = add_link(trans, root, dir, inode, name, namelen, |
1543 | ref_index); | |
3650860b JB |
1544 | if (ret) |
1545 | goto out; | |
5a1d7843 JS |
1546 | |
1547 | btrfs_update_inode(trans, root, inode); | |
1548 | } | |
1549 | ||
f186373f | 1550 | ref_ptr = (unsigned long)(ref_ptr + ref_struct_size) + namelen; |
5a1d7843 | 1551 | kfree(name); |
03b2f08b | 1552 | name = NULL; |
f186373f MF |
1553 | if (log_ref_ver) { |
1554 | iput(dir); | |
1555 | dir = NULL; | |
1556 | } | |
5a1d7843 | 1557 | } |
e02119d5 | 1558 | |
1f250e92 FM |
1559 | /* |
1560 | * Before we overwrite the inode reference item in the subvolume tree | |
1561 | * with the item from the log tree, we must unlink all names from the | |
1562 | * parent directory that are in the subvolume's tree inode reference | |
1563 | * item, otherwise we end up with an inconsistent subvolume tree where | |
1564 | * dir index entries exist for a name but there is no inode reference | |
1565 | * item with the same name. | |
1566 | */ | |
1567 | ret = unlink_old_inode_refs(trans, root, path, BTRFS_I(inode), eb, slot, | |
1568 | key); | |
1569 | if (ret) | |
1570 | goto out; | |
1571 | ||
e02119d5 CM |
1572 | /* finally write the back reference in the inode */ |
1573 | ret = overwrite_item(trans, root, path, eb, slot, key); | |
5a1d7843 | 1574 | out: |
b3b4aa74 | 1575 | btrfs_release_path(path); |
03b2f08b | 1576 | kfree(name); |
e02119d5 CM |
1577 | iput(dir); |
1578 | iput(inode); | |
3650860b | 1579 | return ret; |
e02119d5 CM |
1580 | } |
1581 | ||
c71bf099 | 1582 | static int insert_orphan_item(struct btrfs_trans_handle *trans, |
9c4f61f0 | 1583 | struct btrfs_root *root, u64 ino) |
c71bf099 YZ |
1584 | { |
1585 | int ret; | |
381cf658 | 1586 | |
9c4f61f0 DS |
1587 | ret = btrfs_insert_orphan_item(trans, root, ino); |
1588 | if (ret == -EEXIST) | |
1589 | ret = 0; | |
381cf658 | 1590 | |
c71bf099 YZ |
1591 | return ret; |
1592 | } | |
1593 | ||
f186373f | 1594 | static int count_inode_extrefs(struct btrfs_root *root, |
36283658 | 1595 | struct btrfs_inode *inode, struct btrfs_path *path) |
f186373f MF |
1596 | { |
1597 | int ret = 0; | |
1598 | int name_len; | |
1599 | unsigned int nlink = 0; | |
1600 | u32 item_size; | |
1601 | u32 cur_offset = 0; | |
36283658 | 1602 | u64 inode_objectid = btrfs_ino(inode); |
f186373f MF |
1603 | u64 offset = 0; |
1604 | unsigned long ptr; | |
1605 | struct btrfs_inode_extref *extref; | |
1606 | struct extent_buffer *leaf; | |
1607 | ||
1608 | while (1) { | |
1609 | ret = btrfs_find_one_extref(root, inode_objectid, offset, path, | |
1610 | &extref, &offset); | |
1611 | if (ret) | |
1612 | break; | |
c71bf099 | 1613 | |
f186373f MF |
1614 | leaf = path->nodes[0]; |
1615 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1616 | ptr = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
2c2c452b | 1617 | cur_offset = 0; |
f186373f MF |
1618 | |
1619 | while (cur_offset < item_size) { | |
1620 | extref = (struct btrfs_inode_extref *) (ptr + cur_offset); | |
1621 | name_len = btrfs_inode_extref_name_len(leaf, extref); | |
1622 | ||
1623 | nlink++; | |
1624 | ||
1625 | cur_offset += name_len + sizeof(*extref); | |
1626 | } | |
1627 | ||
1628 | offset++; | |
1629 | btrfs_release_path(path); | |
1630 | } | |
1631 | btrfs_release_path(path); | |
1632 | ||
2c2c452b | 1633 | if (ret < 0 && ret != -ENOENT) |
f186373f MF |
1634 | return ret; |
1635 | return nlink; | |
1636 | } | |
1637 | ||
1638 | static int count_inode_refs(struct btrfs_root *root, | |
f329e319 | 1639 | struct btrfs_inode *inode, struct btrfs_path *path) |
e02119d5 | 1640 | { |
e02119d5 CM |
1641 | int ret; |
1642 | struct btrfs_key key; | |
f186373f | 1643 | unsigned int nlink = 0; |
e02119d5 CM |
1644 | unsigned long ptr; |
1645 | unsigned long ptr_end; | |
1646 | int name_len; | |
f329e319 | 1647 | u64 ino = btrfs_ino(inode); |
e02119d5 | 1648 | |
33345d01 | 1649 | key.objectid = ino; |
e02119d5 CM |
1650 | key.type = BTRFS_INODE_REF_KEY; |
1651 | key.offset = (u64)-1; | |
1652 | ||
d397712b | 1653 | while (1) { |
e02119d5 CM |
1654 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
1655 | if (ret < 0) | |
1656 | break; | |
1657 | if (ret > 0) { | |
1658 | if (path->slots[0] == 0) | |
1659 | break; | |
1660 | path->slots[0]--; | |
1661 | } | |
e93ae26f | 1662 | process_slot: |
e02119d5 CM |
1663 | btrfs_item_key_to_cpu(path->nodes[0], &key, |
1664 | path->slots[0]); | |
33345d01 | 1665 | if (key.objectid != ino || |
e02119d5 CM |
1666 | key.type != BTRFS_INODE_REF_KEY) |
1667 | break; | |
1668 | ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]); | |
1669 | ptr_end = ptr + btrfs_item_size_nr(path->nodes[0], | |
1670 | path->slots[0]); | |
d397712b | 1671 | while (ptr < ptr_end) { |
e02119d5 CM |
1672 | struct btrfs_inode_ref *ref; |
1673 | ||
1674 | ref = (struct btrfs_inode_ref *)ptr; | |
1675 | name_len = btrfs_inode_ref_name_len(path->nodes[0], | |
1676 | ref); | |
1677 | ptr = (unsigned long)(ref + 1) + name_len; | |
1678 | nlink++; | |
1679 | } | |
1680 | ||
1681 | if (key.offset == 0) | |
1682 | break; | |
e93ae26f FDBM |
1683 | if (path->slots[0] > 0) { |
1684 | path->slots[0]--; | |
1685 | goto process_slot; | |
1686 | } | |
e02119d5 | 1687 | key.offset--; |
b3b4aa74 | 1688 | btrfs_release_path(path); |
e02119d5 | 1689 | } |
b3b4aa74 | 1690 | btrfs_release_path(path); |
f186373f MF |
1691 | |
1692 | return nlink; | |
1693 | } | |
1694 | ||
1695 | /* | |
1696 | * There are a few corners where the link count of the file can't | |
1697 | * be properly maintained during replay. So, instead of adding | |
1698 | * lots of complexity to the log code, we just scan the backrefs | |
1699 | * for any file that has been through replay. | |
1700 | * | |
1701 | * The scan will update the link count on the inode to reflect the | |
1702 | * number of back refs found. If it goes down to zero, the iput | |
1703 | * will free the inode. | |
1704 | */ | |
1705 | static noinline int fixup_inode_link_count(struct btrfs_trans_handle *trans, | |
1706 | struct btrfs_root *root, | |
1707 | struct inode *inode) | |
1708 | { | |
1709 | struct btrfs_path *path; | |
1710 | int ret; | |
1711 | u64 nlink = 0; | |
4a0cc7ca | 1712 | u64 ino = btrfs_ino(BTRFS_I(inode)); |
f186373f MF |
1713 | |
1714 | path = btrfs_alloc_path(); | |
1715 | if (!path) | |
1716 | return -ENOMEM; | |
1717 | ||
f329e319 | 1718 | ret = count_inode_refs(root, BTRFS_I(inode), path); |
f186373f MF |
1719 | if (ret < 0) |
1720 | goto out; | |
1721 | ||
1722 | nlink = ret; | |
1723 | ||
36283658 | 1724 | ret = count_inode_extrefs(root, BTRFS_I(inode), path); |
f186373f MF |
1725 | if (ret < 0) |
1726 | goto out; | |
1727 | ||
1728 | nlink += ret; | |
1729 | ||
1730 | ret = 0; | |
1731 | ||
e02119d5 | 1732 | if (nlink != inode->i_nlink) { |
bfe86848 | 1733 | set_nlink(inode, nlink); |
e02119d5 CM |
1734 | btrfs_update_inode(trans, root, inode); |
1735 | } | |
8d5bf1cb | 1736 | BTRFS_I(inode)->index_cnt = (u64)-1; |
e02119d5 | 1737 | |
c71bf099 YZ |
1738 | if (inode->i_nlink == 0) { |
1739 | if (S_ISDIR(inode->i_mode)) { | |
1740 | ret = replay_dir_deletes(trans, root, NULL, path, | |
33345d01 | 1741 | ino, 1); |
3650860b JB |
1742 | if (ret) |
1743 | goto out; | |
c71bf099 | 1744 | } |
33345d01 | 1745 | ret = insert_orphan_item(trans, root, ino); |
12fcfd22 | 1746 | } |
12fcfd22 | 1747 | |
f186373f MF |
1748 | out: |
1749 | btrfs_free_path(path); | |
1750 | return ret; | |
e02119d5 CM |
1751 | } |
1752 | ||
1753 | static noinline int fixup_inode_link_counts(struct btrfs_trans_handle *trans, | |
1754 | struct btrfs_root *root, | |
1755 | struct btrfs_path *path) | |
1756 | { | |
1757 | int ret; | |
1758 | struct btrfs_key key; | |
1759 | struct inode *inode; | |
1760 | ||
1761 | key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID; | |
1762 | key.type = BTRFS_ORPHAN_ITEM_KEY; | |
1763 | key.offset = (u64)-1; | |
d397712b | 1764 | while (1) { |
e02119d5 CM |
1765 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
1766 | if (ret < 0) | |
1767 | break; | |
1768 | ||
1769 | if (ret == 1) { | |
1770 | if (path->slots[0] == 0) | |
1771 | break; | |
1772 | path->slots[0]--; | |
1773 | } | |
1774 | ||
1775 | btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]); | |
1776 | if (key.objectid != BTRFS_TREE_LOG_FIXUP_OBJECTID || | |
1777 | key.type != BTRFS_ORPHAN_ITEM_KEY) | |
1778 | break; | |
1779 | ||
1780 | ret = btrfs_del_item(trans, root, path); | |
65a246c5 TI |
1781 | if (ret) |
1782 | goto out; | |
e02119d5 | 1783 | |
b3b4aa74 | 1784 | btrfs_release_path(path); |
e02119d5 | 1785 | inode = read_one_inode(root, key.offset); |
c00e9493 TI |
1786 | if (!inode) |
1787 | return -EIO; | |
e02119d5 CM |
1788 | |
1789 | ret = fixup_inode_link_count(trans, root, inode); | |
e02119d5 | 1790 | iput(inode); |
3650860b JB |
1791 | if (ret) |
1792 | goto out; | |
e02119d5 | 1793 | |
12fcfd22 CM |
1794 | /* |
1795 | * fixup on a directory may create new entries, | |
1796 | * make sure we always look for the highset possible | |
1797 | * offset | |
1798 | */ | |
1799 | key.offset = (u64)-1; | |
e02119d5 | 1800 | } |
65a246c5 TI |
1801 | ret = 0; |
1802 | out: | |
b3b4aa74 | 1803 | btrfs_release_path(path); |
65a246c5 | 1804 | return ret; |
e02119d5 CM |
1805 | } |
1806 | ||
1807 | ||
1808 | /* | |
1809 | * record a given inode in the fixup dir so we can check its link | |
1810 | * count when replay is done. The link count is incremented here | |
1811 | * so the inode won't go away until we check it | |
1812 | */ | |
1813 | static noinline int link_to_fixup_dir(struct btrfs_trans_handle *trans, | |
1814 | struct btrfs_root *root, | |
1815 | struct btrfs_path *path, | |
1816 | u64 objectid) | |
1817 | { | |
1818 | struct btrfs_key key; | |
1819 | int ret = 0; | |
1820 | struct inode *inode; | |
1821 | ||
1822 | inode = read_one_inode(root, objectid); | |
c00e9493 TI |
1823 | if (!inode) |
1824 | return -EIO; | |
e02119d5 CM |
1825 | |
1826 | key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID; | |
962a298f | 1827 | key.type = BTRFS_ORPHAN_ITEM_KEY; |
e02119d5 CM |
1828 | key.offset = objectid; |
1829 | ||
1830 | ret = btrfs_insert_empty_item(trans, root, path, &key, 0); | |
1831 | ||
b3b4aa74 | 1832 | btrfs_release_path(path); |
e02119d5 | 1833 | if (ret == 0) { |
9bf7a489 JB |
1834 | if (!inode->i_nlink) |
1835 | set_nlink(inode, 1); | |
1836 | else | |
8b558c5f | 1837 | inc_nlink(inode); |
b9959295 | 1838 | ret = btrfs_update_inode(trans, root, inode); |
e02119d5 CM |
1839 | } else if (ret == -EEXIST) { |
1840 | ret = 0; | |
1841 | } else { | |
3650860b | 1842 | BUG(); /* Logic Error */ |
e02119d5 CM |
1843 | } |
1844 | iput(inode); | |
1845 | ||
1846 | return ret; | |
1847 | } | |
1848 | ||
1849 | /* | |
1850 | * when replaying the log for a directory, we only insert names | |
1851 | * for inodes that actually exist. This means an fsync on a directory | |
1852 | * does not implicitly fsync all the new files in it | |
1853 | */ | |
1854 | static noinline int insert_one_name(struct btrfs_trans_handle *trans, | |
1855 | struct btrfs_root *root, | |
e02119d5 | 1856 | u64 dirid, u64 index, |
60d53eb3 | 1857 | char *name, int name_len, |
e02119d5 CM |
1858 | struct btrfs_key *location) |
1859 | { | |
1860 | struct inode *inode; | |
1861 | struct inode *dir; | |
1862 | int ret; | |
1863 | ||
1864 | inode = read_one_inode(root, location->objectid); | |
1865 | if (!inode) | |
1866 | return -ENOENT; | |
1867 | ||
1868 | dir = read_one_inode(root, dirid); | |
1869 | if (!dir) { | |
1870 | iput(inode); | |
1871 | return -EIO; | |
1872 | } | |
d555438b | 1873 | |
db0a669f NB |
1874 | ret = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode), name, |
1875 | name_len, 1, index); | |
e02119d5 CM |
1876 | |
1877 | /* FIXME, put inode into FIXUP list */ | |
1878 | ||
1879 | iput(inode); | |
1880 | iput(dir); | |
1881 | return ret; | |
1882 | } | |
1883 | ||
df8d116f FM |
1884 | /* |
1885 | * Return true if an inode reference exists in the log for the given name, | |
1886 | * inode and parent inode. | |
1887 | */ | |
1888 | static bool name_in_log_ref(struct btrfs_root *log_root, | |
1889 | const char *name, const int name_len, | |
1890 | const u64 dirid, const u64 ino) | |
1891 | { | |
1892 | struct btrfs_key search_key; | |
1893 | ||
1894 | search_key.objectid = ino; | |
1895 | search_key.type = BTRFS_INODE_REF_KEY; | |
1896 | search_key.offset = dirid; | |
1897 | if (backref_in_log(log_root, &search_key, dirid, name, name_len)) | |
1898 | return true; | |
1899 | ||
1900 | search_key.type = BTRFS_INODE_EXTREF_KEY; | |
1901 | search_key.offset = btrfs_extref_hash(dirid, name, name_len); | |
1902 | if (backref_in_log(log_root, &search_key, dirid, name, name_len)) | |
1903 | return true; | |
1904 | ||
1905 | return false; | |
1906 | } | |
1907 | ||
e02119d5 CM |
1908 | /* |
1909 | * take a single entry in a log directory item and replay it into | |
1910 | * the subvolume. | |
1911 | * | |
1912 | * if a conflicting item exists in the subdirectory already, | |
1913 | * the inode it points to is unlinked and put into the link count | |
1914 | * fix up tree. | |
1915 | * | |
1916 | * If a name from the log points to a file or directory that does | |
1917 | * not exist in the FS, it is skipped. fsyncs on directories | |
1918 | * do not force down inodes inside that directory, just changes to the | |
1919 | * names or unlinks in a directory. | |
bb53eda9 FM |
1920 | * |
1921 | * Returns < 0 on error, 0 if the name wasn't replayed (dentry points to a | |
1922 | * non-existing inode) and 1 if the name was replayed. | |
e02119d5 CM |
1923 | */ |
1924 | static noinline int replay_one_name(struct btrfs_trans_handle *trans, | |
1925 | struct btrfs_root *root, | |
1926 | struct btrfs_path *path, | |
1927 | struct extent_buffer *eb, | |
1928 | struct btrfs_dir_item *di, | |
1929 | struct btrfs_key *key) | |
1930 | { | |
1931 | char *name; | |
1932 | int name_len; | |
1933 | struct btrfs_dir_item *dst_di; | |
1934 | struct btrfs_key found_key; | |
1935 | struct btrfs_key log_key; | |
1936 | struct inode *dir; | |
e02119d5 | 1937 | u8 log_type; |
4bef0848 | 1938 | int exists; |
3650860b | 1939 | int ret = 0; |
d555438b | 1940 | bool update_size = (key->type == BTRFS_DIR_INDEX_KEY); |
bb53eda9 | 1941 | bool name_added = false; |
e02119d5 CM |
1942 | |
1943 | dir = read_one_inode(root, key->objectid); | |
c00e9493 TI |
1944 | if (!dir) |
1945 | return -EIO; | |
e02119d5 CM |
1946 | |
1947 | name_len = btrfs_dir_name_len(eb, di); | |
1948 | name = kmalloc(name_len, GFP_NOFS); | |
2bac325e FDBM |
1949 | if (!name) { |
1950 | ret = -ENOMEM; | |
1951 | goto out; | |
1952 | } | |
2a29edc6 | 1953 | |
e02119d5 CM |
1954 | log_type = btrfs_dir_type(eb, di); |
1955 | read_extent_buffer(eb, name, (unsigned long)(di + 1), | |
1956 | name_len); | |
1957 | ||
1958 | btrfs_dir_item_key_to_cpu(eb, di, &log_key); | |
4bef0848 CM |
1959 | exists = btrfs_lookup_inode(trans, root, path, &log_key, 0); |
1960 | if (exists == 0) | |
1961 | exists = 1; | |
1962 | else | |
1963 | exists = 0; | |
b3b4aa74 | 1964 | btrfs_release_path(path); |
4bef0848 | 1965 | |
e02119d5 CM |
1966 | if (key->type == BTRFS_DIR_ITEM_KEY) { |
1967 | dst_di = btrfs_lookup_dir_item(trans, root, path, key->objectid, | |
1968 | name, name_len, 1); | |
d397712b | 1969 | } else if (key->type == BTRFS_DIR_INDEX_KEY) { |
e02119d5 CM |
1970 | dst_di = btrfs_lookup_dir_index_item(trans, root, path, |
1971 | key->objectid, | |
1972 | key->offset, name, | |
1973 | name_len, 1); | |
1974 | } else { | |
3650860b JB |
1975 | /* Corruption */ |
1976 | ret = -EINVAL; | |
1977 | goto out; | |
e02119d5 | 1978 | } |
c704005d | 1979 | if (IS_ERR_OR_NULL(dst_di)) { |
e02119d5 CM |
1980 | /* we need a sequence number to insert, so we only |
1981 | * do inserts for the BTRFS_DIR_INDEX_KEY types | |
1982 | */ | |
1983 | if (key->type != BTRFS_DIR_INDEX_KEY) | |
1984 | goto out; | |
1985 | goto insert; | |
1986 | } | |
1987 | ||
1988 | btrfs_dir_item_key_to_cpu(path->nodes[0], dst_di, &found_key); | |
1989 | /* the existing item matches the logged item */ | |
1990 | if (found_key.objectid == log_key.objectid && | |
1991 | found_key.type == log_key.type && | |
1992 | found_key.offset == log_key.offset && | |
1993 | btrfs_dir_type(path->nodes[0], dst_di) == log_type) { | |
a2cc11db | 1994 | update_size = false; |
e02119d5 CM |
1995 | goto out; |
1996 | } | |
1997 | ||
1998 | /* | |
1999 | * don't drop the conflicting directory entry if the inode | |
2000 | * for the new entry doesn't exist | |
2001 | */ | |
4bef0848 | 2002 | if (!exists) |
e02119d5 CM |
2003 | goto out; |
2004 | ||
207e7d92 | 2005 | ret = drop_one_dir_item(trans, root, path, BTRFS_I(dir), dst_di); |
3650860b JB |
2006 | if (ret) |
2007 | goto out; | |
e02119d5 CM |
2008 | |
2009 | if (key->type == BTRFS_DIR_INDEX_KEY) | |
2010 | goto insert; | |
2011 | out: | |
b3b4aa74 | 2012 | btrfs_release_path(path); |
d555438b | 2013 | if (!ret && update_size) { |
6ef06d27 | 2014 | btrfs_i_size_write(BTRFS_I(dir), dir->i_size + name_len * 2); |
d555438b JB |
2015 | ret = btrfs_update_inode(trans, root, dir); |
2016 | } | |
e02119d5 CM |
2017 | kfree(name); |
2018 | iput(dir); | |
bb53eda9 FM |
2019 | if (!ret && name_added) |
2020 | ret = 1; | |
3650860b | 2021 | return ret; |
e02119d5 CM |
2022 | |
2023 | insert: | |
df8d116f FM |
2024 | if (name_in_log_ref(root->log_root, name, name_len, |
2025 | key->objectid, log_key.objectid)) { | |
2026 | /* The dentry will be added later. */ | |
2027 | ret = 0; | |
2028 | update_size = false; | |
2029 | goto out; | |
2030 | } | |
b3b4aa74 | 2031 | btrfs_release_path(path); |
60d53eb3 Z |
2032 | ret = insert_one_name(trans, root, key->objectid, key->offset, |
2033 | name, name_len, &log_key); | |
df8d116f | 2034 | if (ret && ret != -ENOENT && ret != -EEXIST) |
3650860b | 2035 | goto out; |
bb53eda9 FM |
2036 | if (!ret) |
2037 | name_added = true; | |
d555438b | 2038 | update_size = false; |
3650860b | 2039 | ret = 0; |
e02119d5 CM |
2040 | goto out; |
2041 | } | |
2042 | ||
2043 | /* | |
2044 | * find all the names in a directory item and reconcile them into | |
2045 | * the subvolume. Only BTRFS_DIR_ITEM_KEY types will have more than | |
2046 | * one name in a directory item, but the same code gets used for | |
2047 | * both directory index types | |
2048 | */ | |
2049 | static noinline int replay_one_dir_item(struct btrfs_trans_handle *trans, | |
2050 | struct btrfs_root *root, | |
2051 | struct btrfs_path *path, | |
2052 | struct extent_buffer *eb, int slot, | |
2053 | struct btrfs_key *key) | |
2054 | { | |
bb53eda9 | 2055 | int ret = 0; |
e02119d5 CM |
2056 | u32 item_size = btrfs_item_size_nr(eb, slot); |
2057 | struct btrfs_dir_item *di; | |
2058 | int name_len; | |
2059 | unsigned long ptr; | |
2060 | unsigned long ptr_end; | |
bb53eda9 | 2061 | struct btrfs_path *fixup_path = NULL; |
e02119d5 CM |
2062 | |
2063 | ptr = btrfs_item_ptr_offset(eb, slot); | |
2064 | ptr_end = ptr + item_size; | |
d397712b | 2065 | while (ptr < ptr_end) { |
e02119d5 CM |
2066 | di = (struct btrfs_dir_item *)ptr; |
2067 | name_len = btrfs_dir_name_len(eb, di); | |
2068 | ret = replay_one_name(trans, root, path, eb, di, key); | |
bb53eda9 FM |
2069 | if (ret < 0) |
2070 | break; | |
e02119d5 CM |
2071 | ptr = (unsigned long)(di + 1); |
2072 | ptr += name_len; | |
bb53eda9 FM |
2073 | |
2074 | /* | |
2075 | * If this entry refers to a non-directory (directories can not | |
2076 | * have a link count > 1) and it was added in the transaction | |
2077 | * that was not committed, make sure we fixup the link count of | |
2078 | * the inode it the entry points to. Otherwise something like | |
2079 | * the following would result in a directory pointing to an | |
2080 | * inode with a wrong link that does not account for this dir | |
2081 | * entry: | |
2082 | * | |
2083 | * mkdir testdir | |
2084 | * touch testdir/foo | |
2085 | * touch testdir/bar | |
2086 | * sync | |
2087 | * | |
2088 | * ln testdir/bar testdir/bar_link | |
2089 | * ln testdir/foo testdir/foo_link | |
2090 | * xfs_io -c "fsync" testdir/bar | |
2091 | * | |
2092 | * <power failure> | |
2093 | * | |
2094 | * mount fs, log replay happens | |
2095 | * | |
2096 | * File foo would remain with a link count of 1 when it has two | |
2097 | * entries pointing to it in the directory testdir. This would | |
2098 | * make it impossible to ever delete the parent directory has | |
2099 | * it would result in stale dentries that can never be deleted. | |
2100 | */ | |
2101 | if (ret == 1 && btrfs_dir_type(eb, di) != BTRFS_FT_DIR) { | |
2102 | struct btrfs_key di_key; | |
2103 | ||
2104 | if (!fixup_path) { | |
2105 | fixup_path = btrfs_alloc_path(); | |
2106 | if (!fixup_path) { | |
2107 | ret = -ENOMEM; | |
2108 | break; | |
2109 | } | |
2110 | } | |
2111 | ||
2112 | btrfs_dir_item_key_to_cpu(eb, di, &di_key); | |
2113 | ret = link_to_fixup_dir(trans, root, fixup_path, | |
2114 | di_key.objectid); | |
2115 | if (ret) | |
2116 | break; | |
2117 | } | |
2118 | ret = 0; | |
e02119d5 | 2119 | } |
bb53eda9 FM |
2120 | btrfs_free_path(fixup_path); |
2121 | return ret; | |
e02119d5 CM |
2122 | } |
2123 | ||
2124 | /* | |
2125 | * directory replay has two parts. There are the standard directory | |
2126 | * items in the log copied from the subvolume, and range items | |
2127 | * created in the log while the subvolume was logged. | |
2128 | * | |
2129 | * The range items tell us which parts of the key space the log | |
2130 | * is authoritative for. During replay, if a key in the subvolume | |
2131 | * directory is in a logged range item, but not actually in the log | |
2132 | * that means it was deleted from the directory before the fsync | |
2133 | * and should be removed. | |
2134 | */ | |
2135 | static noinline int find_dir_range(struct btrfs_root *root, | |
2136 | struct btrfs_path *path, | |
2137 | u64 dirid, int key_type, | |
2138 | u64 *start_ret, u64 *end_ret) | |
2139 | { | |
2140 | struct btrfs_key key; | |
2141 | u64 found_end; | |
2142 | struct btrfs_dir_log_item *item; | |
2143 | int ret; | |
2144 | int nritems; | |
2145 | ||
2146 | if (*start_ret == (u64)-1) | |
2147 | return 1; | |
2148 | ||
2149 | key.objectid = dirid; | |
2150 | key.type = key_type; | |
2151 | key.offset = *start_ret; | |
2152 | ||
2153 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
2154 | if (ret < 0) | |
2155 | goto out; | |
2156 | if (ret > 0) { | |
2157 | if (path->slots[0] == 0) | |
2158 | goto out; | |
2159 | path->slots[0]--; | |
2160 | } | |
2161 | if (ret != 0) | |
2162 | btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]); | |
2163 | ||
2164 | if (key.type != key_type || key.objectid != dirid) { | |
2165 | ret = 1; | |
2166 | goto next; | |
2167 | } | |
2168 | item = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
2169 | struct btrfs_dir_log_item); | |
2170 | found_end = btrfs_dir_log_end(path->nodes[0], item); | |
2171 | ||
2172 | if (*start_ret >= key.offset && *start_ret <= found_end) { | |
2173 | ret = 0; | |
2174 | *start_ret = key.offset; | |
2175 | *end_ret = found_end; | |
2176 | goto out; | |
2177 | } | |
2178 | ret = 1; | |
2179 | next: | |
2180 | /* check the next slot in the tree to see if it is a valid item */ | |
2181 | nritems = btrfs_header_nritems(path->nodes[0]); | |
2a7bf53f | 2182 | path->slots[0]++; |
e02119d5 CM |
2183 | if (path->slots[0] >= nritems) { |
2184 | ret = btrfs_next_leaf(root, path); | |
2185 | if (ret) | |
2186 | goto out; | |
e02119d5 CM |
2187 | } |
2188 | ||
2189 | btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]); | |
2190 | ||
2191 | if (key.type != key_type || key.objectid != dirid) { | |
2192 | ret = 1; | |
2193 | goto out; | |
2194 | } | |
2195 | item = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
2196 | struct btrfs_dir_log_item); | |
2197 | found_end = btrfs_dir_log_end(path->nodes[0], item); | |
2198 | *start_ret = key.offset; | |
2199 | *end_ret = found_end; | |
2200 | ret = 0; | |
2201 | out: | |
b3b4aa74 | 2202 | btrfs_release_path(path); |
e02119d5 CM |
2203 | return ret; |
2204 | } | |
2205 | ||
2206 | /* | |
2207 | * this looks for a given directory item in the log. If the directory | |
2208 | * item is not in the log, the item is removed and the inode it points | |
2209 | * to is unlinked | |
2210 | */ | |
2211 | static noinline int check_item_in_log(struct btrfs_trans_handle *trans, | |
2212 | struct btrfs_root *root, | |
2213 | struct btrfs_root *log, | |
2214 | struct btrfs_path *path, | |
2215 | struct btrfs_path *log_path, | |
2216 | struct inode *dir, | |
2217 | struct btrfs_key *dir_key) | |
2218 | { | |
2219 | int ret; | |
2220 | struct extent_buffer *eb; | |
2221 | int slot; | |
2222 | u32 item_size; | |
2223 | struct btrfs_dir_item *di; | |
2224 | struct btrfs_dir_item *log_di; | |
2225 | int name_len; | |
2226 | unsigned long ptr; | |
2227 | unsigned long ptr_end; | |
2228 | char *name; | |
2229 | struct inode *inode; | |
2230 | struct btrfs_key location; | |
2231 | ||
2232 | again: | |
2233 | eb = path->nodes[0]; | |
2234 | slot = path->slots[0]; | |
2235 | item_size = btrfs_item_size_nr(eb, slot); | |
2236 | ptr = btrfs_item_ptr_offset(eb, slot); | |
2237 | ptr_end = ptr + item_size; | |
d397712b | 2238 | while (ptr < ptr_end) { |
e02119d5 CM |
2239 | di = (struct btrfs_dir_item *)ptr; |
2240 | name_len = btrfs_dir_name_len(eb, di); | |
2241 | name = kmalloc(name_len, GFP_NOFS); | |
2242 | if (!name) { | |
2243 | ret = -ENOMEM; | |
2244 | goto out; | |
2245 | } | |
2246 | read_extent_buffer(eb, name, (unsigned long)(di + 1), | |
2247 | name_len); | |
2248 | log_di = NULL; | |
12fcfd22 | 2249 | if (log && dir_key->type == BTRFS_DIR_ITEM_KEY) { |
e02119d5 CM |
2250 | log_di = btrfs_lookup_dir_item(trans, log, log_path, |
2251 | dir_key->objectid, | |
2252 | name, name_len, 0); | |
12fcfd22 | 2253 | } else if (log && dir_key->type == BTRFS_DIR_INDEX_KEY) { |
e02119d5 CM |
2254 | log_di = btrfs_lookup_dir_index_item(trans, log, |
2255 | log_path, | |
2256 | dir_key->objectid, | |
2257 | dir_key->offset, | |
2258 | name, name_len, 0); | |
2259 | } | |
8d9e220c | 2260 | if (!log_di || log_di == ERR_PTR(-ENOENT)) { |
e02119d5 | 2261 | btrfs_dir_item_key_to_cpu(eb, di, &location); |
b3b4aa74 DS |
2262 | btrfs_release_path(path); |
2263 | btrfs_release_path(log_path); | |
e02119d5 | 2264 | inode = read_one_inode(root, location.objectid); |
c00e9493 TI |
2265 | if (!inode) { |
2266 | kfree(name); | |
2267 | return -EIO; | |
2268 | } | |
e02119d5 CM |
2269 | |
2270 | ret = link_to_fixup_dir(trans, root, | |
2271 | path, location.objectid); | |
3650860b JB |
2272 | if (ret) { |
2273 | kfree(name); | |
2274 | iput(inode); | |
2275 | goto out; | |
2276 | } | |
2277 | ||
8b558c5f | 2278 | inc_nlink(inode); |
4ec5934e NB |
2279 | ret = btrfs_unlink_inode(trans, root, BTRFS_I(dir), |
2280 | BTRFS_I(inode), name, name_len); | |
3650860b | 2281 | if (!ret) |
e5c304e6 | 2282 | ret = btrfs_run_delayed_items(trans); |
e02119d5 CM |
2283 | kfree(name); |
2284 | iput(inode); | |
3650860b JB |
2285 | if (ret) |
2286 | goto out; | |
e02119d5 CM |
2287 | |
2288 | /* there might still be more names under this key | |
2289 | * check and repeat if required | |
2290 | */ | |
2291 | ret = btrfs_search_slot(NULL, root, dir_key, path, | |
2292 | 0, 0); | |
2293 | if (ret == 0) | |
2294 | goto again; | |
2295 | ret = 0; | |
2296 | goto out; | |
269d040f FDBM |
2297 | } else if (IS_ERR(log_di)) { |
2298 | kfree(name); | |
2299 | return PTR_ERR(log_di); | |
e02119d5 | 2300 | } |
b3b4aa74 | 2301 | btrfs_release_path(log_path); |
e02119d5 CM |
2302 | kfree(name); |
2303 | ||
2304 | ptr = (unsigned long)(di + 1); | |
2305 | ptr += name_len; | |
2306 | } | |
2307 | ret = 0; | |
2308 | out: | |
b3b4aa74 DS |
2309 | btrfs_release_path(path); |
2310 | btrfs_release_path(log_path); | |
e02119d5 CM |
2311 | return ret; |
2312 | } | |
2313 | ||
4f764e51 FM |
2314 | static int replay_xattr_deletes(struct btrfs_trans_handle *trans, |
2315 | struct btrfs_root *root, | |
2316 | struct btrfs_root *log, | |
2317 | struct btrfs_path *path, | |
2318 | const u64 ino) | |
2319 | { | |
2320 | struct btrfs_key search_key; | |
2321 | struct btrfs_path *log_path; | |
2322 | int i; | |
2323 | int nritems; | |
2324 | int ret; | |
2325 | ||
2326 | log_path = btrfs_alloc_path(); | |
2327 | if (!log_path) | |
2328 | return -ENOMEM; | |
2329 | ||
2330 | search_key.objectid = ino; | |
2331 | search_key.type = BTRFS_XATTR_ITEM_KEY; | |
2332 | search_key.offset = 0; | |
2333 | again: | |
2334 | ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0); | |
2335 | if (ret < 0) | |
2336 | goto out; | |
2337 | process_leaf: | |
2338 | nritems = btrfs_header_nritems(path->nodes[0]); | |
2339 | for (i = path->slots[0]; i < nritems; i++) { | |
2340 | struct btrfs_key key; | |
2341 | struct btrfs_dir_item *di; | |
2342 | struct btrfs_dir_item *log_di; | |
2343 | u32 total_size; | |
2344 | u32 cur; | |
2345 | ||
2346 | btrfs_item_key_to_cpu(path->nodes[0], &key, i); | |
2347 | if (key.objectid != ino || key.type != BTRFS_XATTR_ITEM_KEY) { | |
2348 | ret = 0; | |
2349 | goto out; | |
2350 | } | |
2351 | ||
2352 | di = btrfs_item_ptr(path->nodes[0], i, struct btrfs_dir_item); | |
2353 | total_size = btrfs_item_size_nr(path->nodes[0], i); | |
2354 | cur = 0; | |
2355 | while (cur < total_size) { | |
2356 | u16 name_len = btrfs_dir_name_len(path->nodes[0], di); | |
2357 | u16 data_len = btrfs_dir_data_len(path->nodes[0], di); | |
2358 | u32 this_len = sizeof(*di) + name_len + data_len; | |
2359 | char *name; | |
2360 | ||
2361 | name = kmalloc(name_len, GFP_NOFS); | |
2362 | if (!name) { | |
2363 | ret = -ENOMEM; | |
2364 | goto out; | |
2365 | } | |
2366 | read_extent_buffer(path->nodes[0], name, | |
2367 | (unsigned long)(di + 1), name_len); | |
2368 | ||
2369 | log_di = btrfs_lookup_xattr(NULL, log, log_path, ino, | |
2370 | name, name_len, 0); | |
2371 | btrfs_release_path(log_path); | |
2372 | if (!log_di) { | |
2373 | /* Doesn't exist in log tree, so delete it. */ | |
2374 | btrfs_release_path(path); | |
2375 | di = btrfs_lookup_xattr(trans, root, path, ino, | |
2376 | name, name_len, -1); | |
2377 | kfree(name); | |
2378 | if (IS_ERR(di)) { | |
2379 | ret = PTR_ERR(di); | |
2380 | goto out; | |
2381 | } | |
2382 | ASSERT(di); | |
2383 | ret = btrfs_delete_one_dir_name(trans, root, | |
2384 | path, di); | |
2385 | if (ret) | |
2386 | goto out; | |
2387 | btrfs_release_path(path); | |
2388 | search_key = key; | |
2389 | goto again; | |
2390 | } | |
2391 | kfree(name); | |
2392 | if (IS_ERR(log_di)) { | |
2393 | ret = PTR_ERR(log_di); | |
2394 | goto out; | |
2395 | } | |
2396 | cur += this_len; | |
2397 | di = (struct btrfs_dir_item *)((char *)di + this_len); | |
2398 | } | |
2399 | } | |
2400 | ret = btrfs_next_leaf(root, path); | |
2401 | if (ret > 0) | |
2402 | ret = 0; | |
2403 | else if (ret == 0) | |
2404 | goto process_leaf; | |
2405 | out: | |
2406 | btrfs_free_path(log_path); | |
2407 | btrfs_release_path(path); | |
2408 | return ret; | |
2409 | } | |
2410 | ||
2411 | ||
e02119d5 CM |
2412 | /* |
2413 | * deletion replay happens before we copy any new directory items | |
2414 | * out of the log or out of backreferences from inodes. It | |
2415 | * scans the log to find ranges of keys that log is authoritative for, | |
2416 | * and then scans the directory to find items in those ranges that are | |
2417 | * not present in the log. | |
2418 | * | |
2419 | * Anything we don't find in the log is unlinked and removed from the | |
2420 | * directory. | |
2421 | */ | |
2422 | static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans, | |
2423 | struct btrfs_root *root, | |
2424 | struct btrfs_root *log, | |
2425 | struct btrfs_path *path, | |
12fcfd22 | 2426 | u64 dirid, int del_all) |
e02119d5 CM |
2427 | { |
2428 | u64 range_start; | |
2429 | u64 range_end; | |
2430 | int key_type = BTRFS_DIR_LOG_ITEM_KEY; | |
2431 | int ret = 0; | |
2432 | struct btrfs_key dir_key; | |
2433 | struct btrfs_key found_key; | |
2434 | struct btrfs_path *log_path; | |
2435 | struct inode *dir; | |
2436 | ||
2437 | dir_key.objectid = dirid; | |
2438 | dir_key.type = BTRFS_DIR_ITEM_KEY; | |
2439 | log_path = btrfs_alloc_path(); | |
2440 | if (!log_path) | |
2441 | return -ENOMEM; | |
2442 | ||
2443 | dir = read_one_inode(root, dirid); | |
2444 | /* it isn't an error if the inode isn't there, that can happen | |
2445 | * because we replay the deletes before we copy in the inode item | |
2446 | * from the log | |
2447 | */ | |
2448 | if (!dir) { | |
2449 | btrfs_free_path(log_path); | |
2450 | return 0; | |
2451 | } | |
2452 | again: | |
2453 | range_start = 0; | |
2454 | range_end = 0; | |
d397712b | 2455 | while (1) { |
12fcfd22 CM |
2456 | if (del_all) |
2457 | range_end = (u64)-1; | |
2458 | else { | |
2459 | ret = find_dir_range(log, path, dirid, key_type, | |
2460 | &range_start, &range_end); | |
2461 | if (ret != 0) | |
2462 | break; | |
2463 | } | |
e02119d5 CM |
2464 | |
2465 | dir_key.offset = range_start; | |
d397712b | 2466 | while (1) { |
e02119d5 CM |
2467 | int nritems; |
2468 | ret = btrfs_search_slot(NULL, root, &dir_key, path, | |
2469 | 0, 0); | |
2470 | if (ret < 0) | |
2471 | goto out; | |
2472 | ||
2473 | nritems = btrfs_header_nritems(path->nodes[0]); | |
2474 | if (path->slots[0] >= nritems) { | |
2475 | ret = btrfs_next_leaf(root, path); | |
b98def7c | 2476 | if (ret == 1) |
e02119d5 | 2477 | break; |
b98def7c LB |
2478 | else if (ret < 0) |
2479 | goto out; | |
e02119d5 CM |
2480 | } |
2481 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, | |
2482 | path->slots[0]); | |
2483 | if (found_key.objectid != dirid || | |
2484 | found_key.type != dir_key.type) | |
2485 | goto next_type; | |
2486 | ||
2487 | if (found_key.offset > range_end) | |
2488 | break; | |
2489 | ||
2490 | ret = check_item_in_log(trans, root, log, path, | |
12fcfd22 CM |
2491 | log_path, dir, |
2492 | &found_key); | |
3650860b JB |
2493 | if (ret) |
2494 | goto out; | |
e02119d5 CM |
2495 | if (found_key.offset == (u64)-1) |
2496 | break; | |
2497 | dir_key.offset = found_key.offset + 1; | |
2498 | } | |
b3b4aa74 | 2499 | btrfs_release_path(path); |
e02119d5 CM |
2500 | if (range_end == (u64)-1) |
2501 | break; | |
2502 | range_start = range_end + 1; | |
2503 | } | |
2504 | ||
2505 | next_type: | |
2506 | ret = 0; | |
2507 | if (key_type == BTRFS_DIR_LOG_ITEM_KEY) { | |
2508 | key_type = BTRFS_DIR_LOG_INDEX_KEY; | |
2509 | dir_key.type = BTRFS_DIR_INDEX_KEY; | |
b3b4aa74 | 2510 | btrfs_release_path(path); |
e02119d5 CM |
2511 | goto again; |
2512 | } | |
2513 | out: | |
b3b4aa74 | 2514 | btrfs_release_path(path); |
e02119d5 CM |
2515 | btrfs_free_path(log_path); |
2516 | iput(dir); | |
2517 | return ret; | |
2518 | } | |
2519 | ||
2520 | /* | |
2521 | * the process_func used to replay items from the log tree. This | |
2522 | * gets called in two different stages. The first stage just looks | |
2523 | * for inodes and makes sure they are all copied into the subvolume. | |
2524 | * | |
2525 | * The second stage copies all the other item types from the log into | |
2526 | * the subvolume. The two stage approach is slower, but gets rid of | |
2527 | * lots of complexity around inodes referencing other inodes that exist | |
2528 | * only in the log (references come from either directory items or inode | |
2529 | * back refs). | |
2530 | */ | |
2531 | static int replay_one_buffer(struct btrfs_root *log, struct extent_buffer *eb, | |
581c1760 | 2532 | struct walk_control *wc, u64 gen, int level) |
e02119d5 CM |
2533 | { |
2534 | int nritems; | |
2535 | struct btrfs_path *path; | |
2536 | struct btrfs_root *root = wc->replay_dest; | |
2537 | struct btrfs_key key; | |
e02119d5 CM |
2538 | int i; |
2539 | int ret; | |
2540 | ||
581c1760 | 2541 | ret = btrfs_read_buffer(eb, gen, level, NULL); |
018642a1 TI |
2542 | if (ret) |
2543 | return ret; | |
e02119d5 CM |
2544 | |
2545 | level = btrfs_header_level(eb); | |
2546 | ||
2547 | if (level != 0) | |
2548 | return 0; | |
2549 | ||
2550 | path = btrfs_alloc_path(); | |
1e5063d0 MF |
2551 | if (!path) |
2552 | return -ENOMEM; | |
e02119d5 CM |
2553 | |
2554 | nritems = btrfs_header_nritems(eb); | |
2555 | for (i = 0; i < nritems; i++) { | |
2556 | btrfs_item_key_to_cpu(eb, &key, i); | |
e02119d5 CM |
2557 | |
2558 | /* inode keys are done during the first stage */ | |
2559 | if (key.type == BTRFS_INODE_ITEM_KEY && | |
2560 | wc->stage == LOG_WALK_REPLAY_INODES) { | |
e02119d5 CM |
2561 | struct btrfs_inode_item *inode_item; |
2562 | u32 mode; | |
2563 | ||
2564 | inode_item = btrfs_item_ptr(eb, i, | |
2565 | struct btrfs_inode_item); | |
f2d72f42 FM |
2566 | /* |
2567 | * If we have a tmpfile (O_TMPFILE) that got fsync'ed | |
2568 | * and never got linked before the fsync, skip it, as | |
2569 | * replaying it is pointless since it would be deleted | |
2570 | * later. We skip logging tmpfiles, but it's always | |
2571 | * possible we are replaying a log created with a kernel | |
2572 | * that used to log tmpfiles. | |
2573 | */ | |
2574 | if (btrfs_inode_nlink(eb, inode_item) == 0) { | |
2575 | wc->ignore_cur_inode = true; | |
2576 | continue; | |
2577 | } else { | |
2578 | wc->ignore_cur_inode = false; | |
2579 | } | |
4f764e51 FM |
2580 | ret = replay_xattr_deletes(wc->trans, root, log, |
2581 | path, key.objectid); | |
2582 | if (ret) | |
2583 | break; | |
e02119d5 CM |
2584 | mode = btrfs_inode_mode(eb, inode_item); |
2585 | if (S_ISDIR(mode)) { | |
2586 | ret = replay_dir_deletes(wc->trans, | |
12fcfd22 | 2587 | root, log, path, key.objectid, 0); |
b50c6e25 JB |
2588 | if (ret) |
2589 | break; | |
e02119d5 CM |
2590 | } |
2591 | ret = overwrite_item(wc->trans, root, path, | |
2592 | eb, i, &key); | |
b50c6e25 JB |
2593 | if (ret) |
2594 | break; | |
e02119d5 | 2595 | |
471d557a FM |
2596 | /* |
2597 | * Before replaying extents, truncate the inode to its | |
2598 | * size. We need to do it now and not after log replay | |
2599 | * because before an fsync we can have prealloc extents | |
2600 | * added beyond the inode's i_size. If we did it after, | |
2601 | * through orphan cleanup for example, we would drop | |
2602 | * those prealloc extents just after replaying them. | |
e02119d5 CM |
2603 | */ |
2604 | if (S_ISREG(mode)) { | |
471d557a FM |
2605 | struct inode *inode; |
2606 | u64 from; | |
2607 | ||
2608 | inode = read_one_inode(root, key.objectid); | |
2609 | if (!inode) { | |
2610 | ret = -EIO; | |
2611 | break; | |
2612 | } | |
2613 | from = ALIGN(i_size_read(inode), | |
2614 | root->fs_info->sectorsize); | |
2615 | ret = btrfs_drop_extents(wc->trans, root, inode, | |
2616 | from, (u64)-1, 1); | |
471d557a | 2617 | if (!ret) { |
f2d72f42 | 2618 | /* Update the inode's nbytes. */ |
471d557a FM |
2619 | ret = btrfs_update_inode(wc->trans, |
2620 | root, inode); | |
2621 | } | |
2622 | iput(inode); | |
b50c6e25 JB |
2623 | if (ret) |
2624 | break; | |
e02119d5 | 2625 | } |
c71bf099 | 2626 | |
e02119d5 CM |
2627 | ret = link_to_fixup_dir(wc->trans, root, |
2628 | path, key.objectid); | |
b50c6e25 JB |
2629 | if (ret) |
2630 | break; | |
e02119d5 | 2631 | } |
dd8e7217 | 2632 | |
f2d72f42 FM |
2633 | if (wc->ignore_cur_inode) |
2634 | continue; | |
2635 | ||
dd8e7217 JB |
2636 | if (key.type == BTRFS_DIR_INDEX_KEY && |
2637 | wc->stage == LOG_WALK_REPLAY_DIR_INDEX) { | |
2638 | ret = replay_one_dir_item(wc->trans, root, path, | |
2639 | eb, i, &key); | |
2640 | if (ret) | |
2641 | break; | |
2642 | } | |
2643 | ||
e02119d5 CM |
2644 | if (wc->stage < LOG_WALK_REPLAY_ALL) |
2645 | continue; | |
2646 | ||
2647 | /* these keys are simply copied */ | |
2648 | if (key.type == BTRFS_XATTR_ITEM_KEY) { | |
2649 | ret = overwrite_item(wc->trans, root, path, | |
2650 | eb, i, &key); | |
b50c6e25 JB |
2651 | if (ret) |
2652 | break; | |
2da1c669 LB |
2653 | } else if (key.type == BTRFS_INODE_REF_KEY || |
2654 | key.type == BTRFS_INODE_EXTREF_KEY) { | |
f186373f MF |
2655 | ret = add_inode_ref(wc->trans, root, log, path, |
2656 | eb, i, &key); | |
b50c6e25 JB |
2657 | if (ret && ret != -ENOENT) |
2658 | break; | |
2659 | ret = 0; | |
e02119d5 CM |
2660 | } else if (key.type == BTRFS_EXTENT_DATA_KEY) { |
2661 | ret = replay_one_extent(wc->trans, root, path, | |
2662 | eb, i, &key); | |
b50c6e25 JB |
2663 | if (ret) |
2664 | break; | |
dd8e7217 | 2665 | } else if (key.type == BTRFS_DIR_ITEM_KEY) { |
e02119d5 CM |
2666 | ret = replay_one_dir_item(wc->trans, root, path, |
2667 | eb, i, &key); | |
b50c6e25 JB |
2668 | if (ret) |
2669 | break; | |
e02119d5 CM |
2670 | } |
2671 | } | |
2672 | btrfs_free_path(path); | |
b50c6e25 | 2673 | return ret; |
e02119d5 CM |
2674 | } |
2675 | ||
d397712b | 2676 | static noinline int walk_down_log_tree(struct btrfs_trans_handle *trans, |
e02119d5 CM |
2677 | struct btrfs_root *root, |
2678 | struct btrfs_path *path, int *level, | |
2679 | struct walk_control *wc) | |
2680 | { | |
0b246afa | 2681 | struct btrfs_fs_info *fs_info = root->fs_info; |
e02119d5 | 2682 | u64 root_owner; |
e02119d5 CM |
2683 | u64 bytenr; |
2684 | u64 ptr_gen; | |
2685 | struct extent_buffer *next; | |
2686 | struct extent_buffer *cur; | |
2687 | struct extent_buffer *parent; | |
2688 | u32 blocksize; | |
2689 | int ret = 0; | |
2690 | ||
2691 | WARN_ON(*level < 0); | |
2692 | WARN_ON(*level >= BTRFS_MAX_LEVEL); | |
2693 | ||
d397712b | 2694 | while (*level > 0) { |
581c1760 QW |
2695 | struct btrfs_key first_key; |
2696 | ||
e02119d5 CM |
2697 | WARN_ON(*level < 0); |
2698 | WARN_ON(*level >= BTRFS_MAX_LEVEL); | |
2699 | cur = path->nodes[*level]; | |
2700 | ||
fae7f21c | 2701 | WARN_ON(btrfs_header_level(cur) != *level); |
e02119d5 CM |
2702 | |
2703 | if (path->slots[*level] >= | |
2704 | btrfs_header_nritems(cur)) | |
2705 | break; | |
2706 | ||
2707 | bytenr = btrfs_node_blockptr(cur, path->slots[*level]); | |
2708 | ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]); | |
581c1760 | 2709 | btrfs_node_key_to_cpu(cur, &first_key, path->slots[*level]); |
0b246afa | 2710 | blocksize = fs_info->nodesize; |
e02119d5 CM |
2711 | |
2712 | parent = path->nodes[*level]; | |
2713 | root_owner = btrfs_header_owner(parent); | |
e02119d5 | 2714 | |
2ff7e61e | 2715 | next = btrfs_find_create_tree_block(fs_info, bytenr); |
c871b0f2 LB |
2716 | if (IS_ERR(next)) |
2717 | return PTR_ERR(next); | |
e02119d5 | 2718 | |
e02119d5 | 2719 | if (*level == 1) { |
581c1760 QW |
2720 | ret = wc->process_func(root, next, wc, ptr_gen, |
2721 | *level - 1); | |
b50c6e25 JB |
2722 | if (ret) { |
2723 | free_extent_buffer(next); | |
1e5063d0 | 2724 | return ret; |
b50c6e25 | 2725 | } |
4a500fd1 | 2726 | |
e02119d5 CM |
2727 | path->slots[*level]++; |
2728 | if (wc->free) { | |
581c1760 QW |
2729 | ret = btrfs_read_buffer(next, ptr_gen, |
2730 | *level - 1, &first_key); | |
018642a1 TI |
2731 | if (ret) { |
2732 | free_extent_buffer(next); | |
2733 | return ret; | |
2734 | } | |
e02119d5 | 2735 | |
681ae509 JB |
2736 | if (trans) { |
2737 | btrfs_tree_lock(next); | |
8bead258 | 2738 | btrfs_set_lock_blocking_write(next); |
7c302b49 | 2739 | clean_tree_block(fs_info, next); |
681ae509 JB |
2740 | btrfs_wait_tree_block_writeback(next); |
2741 | btrfs_tree_unlock(next); | |
1846430c LB |
2742 | } else { |
2743 | if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &next->bflags)) | |
2744 | clear_extent_buffer_dirty(next); | |
681ae509 | 2745 | } |
e02119d5 | 2746 | |
e02119d5 CM |
2747 | WARN_ON(root_owner != |
2748 | BTRFS_TREE_LOG_OBJECTID); | |
2ff7e61e JM |
2749 | ret = btrfs_free_and_pin_reserved_extent( |
2750 | fs_info, bytenr, | |
2751 | blocksize); | |
3650860b JB |
2752 | if (ret) { |
2753 | free_extent_buffer(next); | |
2754 | return ret; | |
2755 | } | |
e02119d5 CM |
2756 | } |
2757 | free_extent_buffer(next); | |
2758 | continue; | |
2759 | } | |
581c1760 | 2760 | ret = btrfs_read_buffer(next, ptr_gen, *level - 1, &first_key); |
018642a1 TI |
2761 | if (ret) { |
2762 | free_extent_buffer(next); | |
2763 | return ret; | |
2764 | } | |
e02119d5 CM |
2765 | |
2766 | WARN_ON(*level <= 0); | |
2767 | if (path->nodes[*level-1]) | |
2768 | free_extent_buffer(path->nodes[*level-1]); | |
2769 | path->nodes[*level-1] = next; | |
2770 | *level = btrfs_header_level(next); | |
2771 | path->slots[*level] = 0; | |
2772 | cond_resched(); | |
2773 | } | |
2774 | WARN_ON(*level < 0); | |
2775 | WARN_ON(*level >= BTRFS_MAX_LEVEL); | |
2776 | ||
4a500fd1 | 2777 | path->slots[*level] = btrfs_header_nritems(path->nodes[*level]); |
e02119d5 CM |
2778 | |
2779 | cond_resched(); | |
2780 | return 0; | |
2781 | } | |
2782 | ||
d397712b | 2783 | static noinline int walk_up_log_tree(struct btrfs_trans_handle *trans, |
e02119d5 CM |
2784 | struct btrfs_root *root, |
2785 | struct btrfs_path *path, int *level, | |
2786 | struct walk_control *wc) | |
2787 | { | |
0b246afa | 2788 | struct btrfs_fs_info *fs_info = root->fs_info; |
e02119d5 | 2789 | u64 root_owner; |
e02119d5 CM |
2790 | int i; |
2791 | int slot; | |
2792 | int ret; | |
2793 | ||
d397712b | 2794 | for (i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) { |
e02119d5 | 2795 | slot = path->slots[i]; |
4a500fd1 | 2796 | if (slot + 1 < btrfs_header_nritems(path->nodes[i])) { |
e02119d5 CM |
2797 | path->slots[i]++; |
2798 | *level = i; | |
2799 | WARN_ON(*level == 0); | |
2800 | return 0; | |
2801 | } else { | |
31840ae1 ZY |
2802 | struct extent_buffer *parent; |
2803 | if (path->nodes[*level] == root->node) | |
2804 | parent = path->nodes[*level]; | |
2805 | else | |
2806 | parent = path->nodes[*level + 1]; | |
2807 | ||
2808 | root_owner = btrfs_header_owner(parent); | |
1e5063d0 | 2809 | ret = wc->process_func(root, path->nodes[*level], wc, |
581c1760 QW |
2810 | btrfs_header_generation(path->nodes[*level]), |
2811 | *level); | |
1e5063d0 MF |
2812 | if (ret) |
2813 | return ret; | |
2814 | ||
e02119d5 CM |
2815 | if (wc->free) { |
2816 | struct extent_buffer *next; | |
2817 | ||
2818 | next = path->nodes[*level]; | |
2819 | ||
681ae509 JB |
2820 | if (trans) { |
2821 | btrfs_tree_lock(next); | |
8bead258 | 2822 | btrfs_set_lock_blocking_write(next); |
7c302b49 | 2823 | clean_tree_block(fs_info, next); |
681ae509 JB |
2824 | btrfs_wait_tree_block_writeback(next); |
2825 | btrfs_tree_unlock(next); | |
1846430c LB |
2826 | } else { |
2827 | if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &next->bflags)) | |
2828 | clear_extent_buffer_dirty(next); | |
681ae509 | 2829 | } |
e02119d5 | 2830 | |
e02119d5 | 2831 | WARN_ON(root_owner != BTRFS_TREE_LOG_OBJECTID); |
2ff7e61e JM |
2832 | ret = btrfs_free_and_pin_reserved_extent( |
2833 | fs_info, | |
e02119d5 | 2834 | path->nodes[*level]->start, |
d00aff00 | 2835 | path->nodes[*level]->len); |
3650860b JB |
2836 | if (ret) |
2837 | return ret; | |
e02119d5 CM |
2838 | } |
2839 | free_extent_buffer(path->nodes[*level]); | |
2840 | path->nodes[*level] = NULL; | |
2841 | *level = i + 1; | |
2842 | } | |
2843 | } | |
2844 | return 1; | |
2845 | } | |
2846 | ||
2847 | /* | |
2848 | * drop the reference count on the tree rooted at 'snap'. This traverses | |
2849 | * the tree freeing any blocks that have a ref count of zero after being | |
2850 | * decremented. | |
2851 | */ | |
2852 | static int walk_log_tree(struct btrfs_trans_handle *trans, | |
2853 | struct btrfs_root *log, struct walk_control *wc) | |
2854 | { | |
2ff7e61e | 2855 | struct btrfs_fs_info *fs_info = log->fs_info; |
e02119d5 CM |
2856 | int ret = 0; |
2857 | int wret; | |
2858 | int level; | |
2859 | struct btrfs_path *path; | |
e02119d5 CM |
2860 | int orig_level; |
2861 | ||
2862 | path = btrfs_alloc_path(); | |
db5b493a TI |
2863 | if (!path) |
2864 | return -ENOMEM; | |
e02119d5 CM |
2865 | |
2866 | level = btrfs_header_level(log->node); | |
2867 | orig_level = level; | |
2868 | path->nodes[level] = log->node; | |
2869 | extent_buffer_get(log->node); | |
2870 | path->slots[level] = 0; | |
2871 | ||
d397712b | 2872 | while (1) { |
e02119d5 CM |
2873 | wret = walk_down_log_tree(trans, log, path, &level, wc); |
2874 | if (wret > 0) | |
2875 | break; | |
79787eaa | 2876 | if (wret < 0) { |
e02119d5 | 2877 | ret = wret; |
79787eaa JM |
2878 | goto out; |
2879 | } | |
e02119d5 CM |
2880 | |
2881 | wret = walk_up_log_tree(trans, log, path, &level, wc); | |
2882 | if (wret > 0) | |
2883 | break; | |
79787eaa | 2884 | if (wret < 0) { |
e02119d5 | 2885 | ret = wret; |
79787eaa JM |
2886 | goto out; |
2887 | } | |
e02119d5 CM |
2888 | } |
2889 | ||
2890 | /* was the root node processed? if not, catch it here */ | |
2891 | if (path->nodes[orig_level]) { | |
79787eaa | 2892 | ret = wc->process_func(log, path->nodes[orig_level], wc, |
581c1760 QW |
2893 | btrfs_header_generation(path->nodes[orig_level]), |
2894 | orig_level); | |
79787eaa JM |
2895 | if (ret) |
2896 | goto out; | |
e02119d5 CM |
2897 | if (wc->free) { |
2898 | struct extent_buffer *next; | |
2899 | ||
2900 | next = path->nodes[orig_level]; | |
2901 | ||
681ae509 JB |
2902 | if (trans) { |
2903 | btrfs_tree_lock(next); | |
8bead258 | 2904 | btrfs_set_lock_blocking_write(next); |
7c302b49 | 2905 | clean_tree_block(fs_info, next); |
681ae509 JB |
2906 | btrfs_wait_tree_block_writeback(next); |
2907 | btrfs_tree_unlock(next); | |
1846430c LB |
2908 | } else { |
2909 | if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &next->bflags)) | |
2910 | clear_extent_buffer_dirty(next); | |
681ae509 | 2911 | } |
e02119d5 | 2912 | |
e02119d5 CM |
2913 | WARN_ON(log->root_key.objectid != |
2914 | BTRFS_TREE_LOG_OBJECTID); | |
2ff7e61e JM |
2915 | ret = btrfs_free_and_pin_reserved_extent(fs_info, |
2916 | next->start, next->len); | |
3650860b JB |
2917 | if (ret) |
2918 | goto out; | |
e02119d5 CM |
2919 | } |
2920 | } | |
2921 | ||
79787eaa | 2922 | out: |
e02119d5 | 2923 | btrfs_free_path(path); |
e02119d5 CM |
2924 | return ret; |
2925 | } | |
2926 | ||
7237f183 YZ |
2927 | /* |
2928 | * helper function to update the item for a given subvolumes log root | |
2929 | * in the tree of log roots | |
2930 | */ | |
2931 | static int update_log_root(struct btrfs_trans_handle *trans, | |
2932 | struct btrfs_root *log) | |
2933 | { | |
0b246afa | 2934 | struct btrfs_fs_info *fs_info = log->fs_info; |
7237f183 YZ |
2935 | int ret; |
2936 | ||
2937 | if (log->log_transid == 1) { | |
2938 | /* insert root item on the first sync */ | |
0b246afa | 2939 | ret = btrfs_insert_root(trans, fs_info->log_root_tree, |
7237f183 YZ |
2940 | &log->root_key, &log->root_item); |
2941 | } else { | |
0b246afa | 2942 | ret = btrfs_update_root(trans, fs_info->log_root_tree, |
7237f183 YZ |
2943 | &log->root_key, &log->root_item); |
2944 | } | |
2945 | return ret; | |
2946 | } | |
2947 | ||
60d53eb3 | 2948 | static void wait_log_commit(struct btrfs_root *root, int transid) |
e02119d5 CM |
2949 | { |
2950 | DEFINE_WAIT(wait); | |
7237f183 | 2951 | int index = transid % 2; |
e02119d5 | 2952 | |
7237f183 YZ |
2953 | /* |
2954 | * we only allow two pending log transactions at a time, | |
2955 | * so we know that if ours is more than 2 older than the | |
2956 | * current transaction, we're done | |
2957 | */ | |
49e83f57 | 2958 | for (;;) { |
7237f183 YZ |
2959 | prepare_to_wait(&root->log_commit_wait[index], |
2960 | &wait, TASK_UNINTERRUPTIBLE); | |
12fcfd22 | 2961 | |
49e83f57 LB |
2962 | if (!(root->log_transid_committed < transid && |
2963 | atomic_read(&root->log_commit[index]))) | |
2964 | break; | |
12fcfd22 | 2965 | |
49e83f57 LB |
2966 | mutex_unlock(&root->log_mutex); |
2967 | schedule(); | |
7237f183 | 2968 | mutex_lock(&root->log_mutex); |
49e83f57 LB |
2969 | } |
2970 | finish_wait(&root->log_commit_wait[index], &wait); | |
7237f183 YZ |
2971 | } |
2972 | ||
60d53eb3 | 2973 | static void wait_for_writer(struct btrfs_root *root) |
7237f183 YZ |
2974 | { |
2975 | DEFINE_WAIT(wait); | |
8b050d35 | 2976 | |
49e83f57 LB |
2977 | for (;;) { |
2978 | prepare_to_wait(&root->log_writer_wait, &wait, | |
2979 | TASK_UNINTERRUPTIBLE); | |
2980 | if (!atomic_read(&root->log_writers)) | |
2981 | break; | |
2982 | ||
7237f183 | 2983 | mutex_unlock(&root->log_mutex); |
49e83f57 | 2984 | schedule(); |
575849ec | 2985 | mutex_lock(&root->log_mutex); |
7237f183 | 2986 | } |
49e83f57 | 2987 | finish_wait(&root->log_writer_wait, &wait); |
e02119d5 CM |
2988 | } |
2989 | ||
8b050d35 MX |
2990 | static inline void btrfs_remove_log_ctx(struct btrfs_root *root, |
2991 | struct btrfs_log_ctx *ctx) | |
2992 | { | |
2993 | if (!ctx) | |
2994 | return; | |
2995 | ||
2996 | mutex_lock(&root->log_mutex); | |
2997 | list_del_init(&ctx->list); | |
2998 | mutex_unlock(&root->log_mutex); | |
2999 | } | |
3000 | ||
3001 | /* | |
3002 | * Invoked in log mutex context, or be sure there is no other task which | |
3003 | * can access the list. | |
3004 | */ | |
3005 | static inline void btrfs_remove_all_log_ctxs(struct btrfs_root *root, | |
3006 | int index, int error) | |
3007 | { | |
3008 | struct btrfs_log_ctx *ctx; | |
570dd450 | 3009 | struct btrfs_log_ctx *safe; |
8b050d35 | 3010 | |
570dd450 CM |
3011 | list_for_each_entry_safe(ctx, safe, &root->log_ctxs[index], list) { |
3012 | list_del_init(&ctx->list); | |
8b050d35 | 3013 | ctx->log_ret = error; |
570dd450 | 3014 | } |
8b050d35 MX |
3015 | |
3016 | INIT_LIST_HEAD(&root->log_ctxs[index]); | |
3017 | } | |
3018 | ||
e02119d5 CM |
3019 | /* |
3020 | * btrfs_sync_log does sends a given tree log down to the disk and | |
3021 | * updates the super blocks to record it. When this call is done, | |
12fcfd22 CM |
3022 | * you know that any inodes previously logged are safely on disk only |
3023 | * if it returns 0. | |
3024 | * | |
3025 | * Any other return value means you need to call btrfs_commit_transaction. | |
3026 | * Some of the edge cases for fsyncing directories that have had unlinks | |
3027 | * or renames done in the past mean that sometimes the only safe | |
3028 | * fsync is to commit the whole FS. When btrfs_sync_log returns -EAGAIN, | |
3029 | * that has happened. | |
e02119d5 CM |
3030 | */ |
3031 | int btrfs_sync_log(struct btrfs_trans_handle *trans, | |
8b050d35 | 3032 | struct btrfs_root *root, struct btrfs_log_ctx *ctx) |
e02119d5 | 3033 | { |
7237f183 YZ |
3034 | int index1; |
3035 | int index2; | |
8cef4e16 | 3036 | int mark; |
e02119d5 | 3037 | int ret; |
0b246afa | 3038 | struct btrfs_fs_info *fs_info = root->fs_info; |
e02119d5 | 3039 | struct btrfs_root *log = root->log_root; |
0b246afa | 3040 | struct btrfs_root *log_root_tree = fs_info->log_root_tree; |
bb14a59b | 3041 | int log_transid = 0; |
8b050d35 | 3042 | struct btrfs_log_ctx root_log_ctx; |
c6adc9cc | 3043 | struct blk_plug plug; |
e02119d5 | 3044 | |
7237f183 | 3045 | mutex_lock(&root->log_mutex); |
d1433deb MX |
3046 | log_transid = ctx->log_transid; |
3047 | if (root->log_transid_committed >= log_transid) { | |
3048 | mutex_unlock(&root->log_mutex); | |
3049 | return ctx->log_ret; | |
3050 | } | |
3051 | ||
3052 | index1 = log_transid % 2; | |
7237f183 | 3053 | if (atomic_read(&root->log_commit[index1])) { |
60d53eb3 | 3054 | wait_log_commit(root, log_transid); |
7237f183 | 3055 | mutex_unlock(&root->log_mutex); |
8b050d35 | 3056 | return ctx->log_ret; |
e02119d5 | 3057 | } |
d1433deb | 3058 | ASSERT(log_transid == root->log_transid); |
7237f183 YZ |
3059 | atomic_set(&root->log_commit[index1], 1); |
3060 | ||
3061 | /* wait for previous tree log sync to complete */ | |
3062 | if (atomic_read(&root->log_commit[(index1 + 1) % 2])) | |
60d53eb3 | 3063 | wait_log_commit(root, log_transid - 1); |
48cab2e0 | 3064 | |
86df7eb9 | 3065 | while (1) { |
2ecb7923 | 3066 | int batch = atomic_read(&root->log_batch); |
cd354ad6 | 3067 | /* when we're on an ssd, just kick the log commit out */ |
0b246afa | 3068 | if (!btrfs_test_opt(fs_info, SSD) && |
27cdeb70 | 3069 | test_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state)) { |
86df7eb9 YZ |
3070 | mutex_unlock(&root->log_mutex); |
3071 | schedule_timeout_uninterruptible(1); | |
3072 | mutex_lock(&root->log_mutex); | |
3073 | } | |
60d53eb3 | 3074 | wait_for_writer(root); |
2ecb7923 | 3075 | if (batch == atomic_read(&root->log_batch)) |
e02119d5 CM |
3076 | break; |
3077 | } | |
e02119d5 | 3078 | |
12fcfd22 | 3079 | /* bail out if we need to do a full commit */ |
0b246afa | 3080 | if (btrfs_need_log_full_commit(fs_info, trans)) { |
12fcfd22 CM |
3081 | ret = -EAGAIN; |
3082 | mutex_unlock(&root->log_mutex); | |
3083 | goto out; | |
3084 | } | |
3085 | ||
8cef4e16 YZ |
3086 | if (log_transid % 2 == 0) |
3087 | mark = EXTENT_DIRTY; | |
3088 | else | |
3089 | mark = EXTENT_NEW; | |
3090 | ||
690587d1 CM |
3091 | /* we start IO on all the marked extents here, but we don't actually |
3092 | * wait for them until later. | |
3093 | */ | |
c6adc9cc | 3094 | blk_start_plug(&plug); |
2ff7e61e | 3095 | ret = btrfs_write_marked_extents(fs_info, &log->dirty_log_pages, mark); |
79787eaa | 3096 | if (ret) { |
c6adc9cc | 3097 | blk_finish_plug(&plug); |
66642832 | 3098 | btrfs_abort_transaction(trans, ret); |
0b246afa | 3099 | btrfs_set_log_full_commit(fs_info, trans); |
79787eaa JM |
3100 | mutex_unlock(&root->log_mutex); |
3101 | goto out; | |
3102 | } | |
7237f183 | 3103 | |
5d4f98a2 | 3104 | btrfs_set_root_node(&log->root_item, log->node); |
7237f183 | 3105 | |
7237f183 YZ |
3106 | root->log_transid++; |
3107 | log->log_transid = root->log_transid; | |
ff782e0a | 3108 | root->log_start_pid = 0; |
7237f183 | 3109 | /* |
8cef4e16 YZ |
3110 | * IO has been started, blocks of the log tree have WRITTEN flag set |
3111 | * in their headers. new modifications of the log will be written to | |
3112 | * new positions. so it's safe to allow log writers to go in. | |
7237f183 YZ |
3113 | */ |
3114 | mutex_unlock(&root->log_mutex); | |
3115 | ||
28a23593 | 3116 | btrfs_init_log_ctx(&root_log_ctx, NULL); |
d1433deb | 3117 | |
7237f183 | 3118 | mutex_lock(&log_root_tree->log_mutex); |
2ecb7923 | 3119 | atomic_inc(&log_root_tree->log_batch); |
7237f183 | 3120 | atomic_inc(&log_root_tree->log_writers); |
d1433deb MX |
3121 | |
3122 | index2 = log_root_tree->log_transid % 2; | |
3123 | list_add_tail(&root_log_ctx.list, &log_root_tree->log_ctxs[index2]); | |
3124 | root_log_ctx.log_transid = log_root_tree->log_transid; | |
3125 | ||
7237f183 YZ |
3126 | mutex_unlock(&log_root_tree->log_mutex); |
3127 | ||
3128 | ret = update_log_root(trans, log); | |
7237f183 YZ |
3129 | |
3130 | mutex_lock(&log_root_tree->log_mutex); | |
3131 | if (atomic_dec_and_test(&log_root_tree->log_writers)) { | |
093258e6 DS |
3132 | /* atomic_dec_and_test implies a barrier */ |
3133 | cond_wake_up_nomb(&log_root_tree->log_writer_wait); | |
7237f183 YZ |
3134 | } |
3135 | ||
4a500fd1 | 3136 | if (ret) { |
d1433deb MX |
3137 | if (!list_empty(&root_log_ctx.list)) |
3138 | list_del_init(&root_log_ctx.list); | |
3139 | ||
c6adc9cc | 3140 | blk_finish_plug(&plug); |
0b246afa | 3141 | btrfs_set_log_full_commit(fs_info, trans); |
995946dd | 3142 | |
79787eaa | 3143 | if (ret != -ENOSPC) { |
66642832 | 3144 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
3145 | mutex_unlock(&log_root_tree->log_mutex); |
3146 | goto out; | |
3147 | } | |
bf89d38f | 3148 | btrfs_wait_tree_log_extents(log, mark); |
4a500fd1 YZ |
3149 | mutex_unlock(&log_root_tree->log_mutex); |
3150 | ret = -EAGAIN; | |
3151 | goto out; | |
3152 | } | |
3153 | ||
d1433deb | 3154 | if (log_root_tree->log_transid_committed >= root_log_ctx.log_transid) { |
3da5ab56 | 3155 | blk_finish_plug(&plug); |
cbd60aa7 | 3156 | list_del_init(&root_log_ctx.list); |
d1433deb MX |
3157 | mutex_unlock(&log_root_tree->log_mutex); |
3158 | ret = root_log_ctx.log_ret; | |
3159 | goto out; | |
3160 | } | |
8b050d35 | 3161 | |
d1433deb | 3162 | index2 = root_log_ctx.log_transid % 2; |
7237f183 | 3163 | if (atomic_read(&log_root_tree->log_commit[index2])) { |
c6adc9cc | 3164 | blk_finish_plug(&plug); |
bf89d38f | 3165 | ret = btrfs_wait_tree_log_extents(log, mark); |
60d53eb3 | 3166 | wait_log_commit(log_root_tree, |
d1433deb | 3167 | root_log_ctx.log_transid); |
7237f183 | 3168 | mutex_unlock(&log_root_tree->log_mutex); |
5ab5e44a FM |
3169 | if (!ret) |
3170 | ret = root_log_ctx.log_ret; | |
7237f183 YZ |
3171 | goto out; |
3172 | } | |
d1433deb | 3173 | ASSERT(root_log_ctx.log_transid == log_root_tree->log_transid); |
7237f183 YZ |
3174 | atomic_set(&log_root_tree->log_commit[index2], 1); |
3175 | ||
12fcfd22 | 3176 | if (atomic_read(&log_root_tree->log_commit[(index2 + 1) % 2])) { |
60d53eb3 | 3177 | wait_log_commit(log_root_tree, |
d1433deb | 3178 | root_log_ctx.log_transid - 1); |
12fcfd22 CM |
3179 | } |
3180 | ||
60d53eb3 | 3181 | wait_for_writer(log_root_tree); |
7237f183 | 3182 | |
12fcfd22 CM |
3183 | /* |
3184 | * now that we've moved on to the tree of log tree roots, | |
3185 | * check the full commit flag again | |
3186 | */ | |
0b246afa | 3187 | if (btrfs_need_log_full_commit(fs_info, trans)) { |
c6adc9cc | 3188 | blk_finish_plug(&plug); |
bf89d38f | 3189 | btrfs_wait_tree_log_extents(log, mark); |
12fcfd22 CM |
3190 | mutex_unlock(&log_root_tree->log_mutex); |
3191 | ret = -EAGAIN; | |
3192 | goto out_wake_log_root; | |
3193 | } | |
7237f183 | 3194 | |
2ff7e61e | 3195 | ret = btrfs_write_marked_extents(fs_info, |
c6adc9cc MX |
3196 | &log_root_tree->dirty_log_pages, |
3197 | EXTENT_DIRTY | EXTENT_NEW); | |
3198 | blk_finish_plug(&plug); | |
79787eaa | 3199 | if (ret) { |
0b246afa | 3200 | btrfs_set_log_full_commit(fs_info, trans); |
66642832 | 3201 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
3202 | mutex_unlock(&log_root_tree->log_mutex); |
3203 | goto out_wake_log_root; | |
3204 | } | |
bf89d38f | 3205 | ret = btrfs_wait_tree_log_extents(log, mark); |
5ab5e44a | 3206 | if (!ret) |
bf89d38f JM |
3207 | ret = btrfs_wait_tree_log_extents(log_root_tree, |
3208 | EXTENT_NEW | EXTENT_DIRTY); | |
5ab5e44a | 3209 | if (ret) { |
0b246afa | 3210 | btrfs_set_log_full_commit(fs_info, trans); |
5ab5e44a FM |
3211 | mutex_unlock(&log_root_tree->log_mutex); |
3212 | goto out_wake_log_root; | |
3213 | } | |
e02119d5 | 3214 | |
0b246afa JM |
3215 | btrfs_set_super_log_root(fs_info->super_for_commit, |
3216 | log_root_tree->node->start); | |
3217 | btrfs_set_super_log_root_level(fs_info->super_for_commit, | |
3218 | btrfs_header_level(log_root_tree->node)); | |
e02119d5 | 3219 | |
7237f183 | 3220 | log_root_tree->log_transid++; |
7237f183 YZ |
3221 | mutex_unlock(&log_root_tree->log_mutex); |
3222 | ||
3223 | /* | |
52042d8e | 3224 | * Nobody else is going to jump in and write the ctree |
7237f183 YZ |
3225 | * super here because the log_commit atomic below is protecting |
3226 | * us. We must be called with a transaction handle pinning | |
3227 | * the running transaction open, so a full commit can't hop | |
3228 | * in and cause problems either. | |
3229 | */ | |
eece6a9c | 3230 | ret = write_all_supers(fs_info, 1); |
5af3e8cc | 3231 | if (ret) { |
0b246afa | 3232 | btrfs_set_log_full_commit(fs_info, trans); |
66642832 | 3233 | btrfs_abort_transaction(trans, ret); |
5af3e8cc SB |
3234 | goto out_wake_log_root; |
3235 | } | |
7237f183 | 3236 | |
257c62e1 CM |
3237 | mutex_lock(&root->log_mutex); |
3238 | if (root->last_log_commit < log_transid) | |
3239 | root->last_log_commit = log_transid; | |
3240 | mutex_unlock(&root->log_mutex); | |
3241 | ||
12fcfd22 | 3242 | out_wake_log_root: |
570dd450 | 3243 | mutex_lock(&log_root_tree->log_mutex); |
8b050d35 MX |
3244 | btrfs_remove_all_log_ctxs(log_root_tree, index2, ret); |
3245 | ||
d1433deb | 3246 | log_root_tree->log_transid_committed++; |
7237f183 | 3247 | atomic_set(&log_root_tree->log_commit[index2], 0); |
d1433deb MX |
3248 | mutex_unlock(&log_root_tree->log_mutex); |
3249 | ||
33a9eca7 | 3250 | /* |
093258e6 DS |
3251 | * The barrier before waitqueue_active (in cond_wake_up) is needed so |
3252 | * all the updates above are seen by the woken threads. It might not be | |
3253 | * necessary, but proving that seems to be hard. | |
33a9eca7 | 3254 | */ |
093258e6 | 3255 | cond_wake_up(&log_root_tree->log_commit_wait[index2]); |
e02119d5 | 3256 | out: |
d1433deb | 3257 | mutex_lock(&root->log_mutex); |
570dd450 | 3258 | btrfs_remove_all_log_ctxs(root, index1, ret); |
d1433deb | 3259 | root->log_transid_committed++; |
7237f183 | 3260 | atomic_set(&root->log_commit[index1], 0); |
d1433deb | 3261 | mutex_unlock(&root->log_mutex); |
8b050d35 | 3262 | |
33a9eca7 | 3263 | /* |
093258e6 DS |
3264 | * The barrier before waitqueue_active (in cond_wake_up) is needed so |
3265 | * all the updates above are seen by the woken threads. It might not be | |
3266 | * necessary, but proving that seems to be hard. | |
33a9eca7 | 3267 | */ |
093258e6 | 3268 | cond_wake_up(&root->log_commit_wait[index1]); |
b31eabd8 | 3269 | return ret; |
e02119d5 CM |
3270 | } |
3271 | ||
4a500fd1 YZ |
3272 | static void free_log_tree(struct btrfs_trans_handle *trans, |
3273 | struct btrfs_root *log) | |
e02119d5 CM |
3274 | { |
3275 | int ret; | |
e02119d5 CM |
3276 | struct walk_control wc = { |
3277 | .free = 1, | |
3278 | .process_func = process_one_buffer | |
3279 | }; | |
3280 | ||
681ae509 | 3281 | ret = walk_log_tree(trans, log, &wc); |
374b0e2d JM |
3282 | if (ret) { |
3283 | if (trans) | |
3284 | btrfs_abort_transaction(trans, ret); | |
3285 | else | |
3286 | btrfs_handle_fs_error(log->fs_info, ret, NULL); | |
3287 | } | |
e02119d5 | 3288 | |
59b0713a FM |
3289 | clear_extent_bits(&log->dirty_log_pages, 0, (u64)-1, |
3290 | EXTENT_DIRTY | EXTENT_NEW | EXTENT_NEED_WAIT); | |
7237f183 YZ |
3291 | free_extent_buffer(log->node); |
3292 | kfree(log); | |
4a500fd1 YZ |
3293 | } |
3294 | ||
3295 | /* | |
3296 | * free all the extents used by the tree log. This should be called | |
3297 | * at commit time of the full transaction | |
3298 | */ | |
3299 | int btrfs_free_log(struct btrfs_trans_handle *trans, struct btrfs_root *root) | |
3300 | { | |
3301 | if (root->log_root) { | |
3302 | free_log_tree(trans, root->log_root); | |
3303 | root->log_root = NULL; | |
3304 | } | |
3305 | return 0; | |
3306 | } | |
3307 | ||
3308 | int btrfs_free_log_root_tree(struct btrfs_trans_handle *trans, | |
3309 | struct btrfs_fs_info *fs_info) | |
3310 | { | |
3311 | if (fs_info->log_root_tree) { | |
3312 | free_log_tree(trans, fs_info->log_root_tree); | |
3313 | fs_info->log_root_tree = NULL; | |
3314 | } | |
e02119d5 CM |
3315 | return 0; |
3316 | } | |
3317 | ||
e02119d5 CM |
3318 | /* |
3319 | * If both a file and directory are logged, and unlinks or renames are | |
3320 | * mixed in, we have a few interesting corners: | |
3321 | * | |
3322 | * create file X in dir Y | |
3323 | * link file X to X.link in dir Y | |
3324 | * fsync file X | |
3325 | * unlink file X but leave X.link | |
3326 | * fsync dir Y | |
3327 | * | |
3328 | * After a crash we would expect only X.link to exist. But file X | |
3329 | * didn't get fsync'd again so the log has back refs for X and X.link. | |
3330 | * | |
3331 | * We solve this by removing directory entries and inode backrefs from the | |
3332 | * log when a file that was logged in the current transaction is | |
3333 | * unlinked. Any later fsync will include the updated log entries, and | |
3334 | * we'll be able to reconstruct the proper directory items from backrefs. | |
3335 | * | |
3336 | * This optimizations allows us to avoid relogging the entire inode | |
3337 | * or the entire directory. | |
3338 | */ | |
3339 | int btrfs_del_dir_entries_in_log(struct btrfs_trans_handle *trans, | |
3340 | struct btrfs_root *root, | |
3341 | const char *name, int name_len, | |
49f34d1f | 3342 | struct btrfs_inode *dir, u64 index) |
e02119d5 CM |
3343 | { |
3344 | struct btrfs_root *log; | |
3345 | struct btrfs_dir_item *di; | |
3346 | struct btrfs_path *path; | |
3347 | int ret; | |
4a500fd1 | 3348 | int err = 0; |
e02119d5 | 3349 | int bytes_del = 0; |
49f34d1f | 3350 | u64 dir_ino = btrfs_ino(dir); |
e02119d5 | 3351 | |
49f34d1f | 3352 | if (dir->logged_trans < trans->transid) |
3a5f1d45 CM |
3353 | return 0; |
3354 | ||
e02119d5 CM |
3355 | ret = join_running_log_trans(root); |
3356 | if (ret) | |
3357 | return 0; | |
3358 | ||
49f34d1f | 3359 | mutex_lock(&dir->log_mutex); |
e02119d5 CM |
3360 | |
3361 | log = root->log_root; | |
3362 | path = btrfs_alloc_path(); | |
a62f44a5 TI |
3363 | if (!path) { |
3364 | err = -ENOMEM; | |
3365 | goto out_unlock; | |
3366 | } | |
2a29edc6 | 3367 | |
33345d01 | 3368 | di = btrfs_lookup_dir_item(trans, log, path, dir_ino, |
e02119d5 | 3369 | name, name_len, -1); |
4a500fd1 YZ |
3370 | if (IS_ERR(di)) { |
3371 | err = PTR_ERR(di); | |
3372 | goto fail; | |
3373 | } | |
3374 | if (di) { | |
e02119d5 CM |
3375 | ret = btrfs_delete_one_dir_name(trans, log, path, di); |
3376 | bytes_del += name_len; | |
3650860b JB |
3377 | if (ret) { |
3378 | err = ret; | |
3379 | goto fail; | |
3380 | } | |
e02119d5 | 3381 | } |
b3b4aa74 | 3382 | btrfs_release_path(path); |
33345d01 | 3383 | di = btrfs_lookup_dir_index_item(trans, log, path, dir_ino, |
e02119d5 | 3384 | index, name, name_len, -1); |
4a500fd1 YZ |
3385 | if (IS_ERR(di)) { |
3386 | err = PTR_ERR(di); | |
3387 | goto fail; | |
3388 | } | |
3389 | if (di) { | |
e02119d5 CM |
3390 | ret = btrfs_delete_one_dir_name(trans, log, path, di); |
3391 | bytes_del += name_len; | |
3650860b JB |
3392 | if (ret) { |
3393 | err = ret; | |
3394 | goto fail; | |
3395 | } | |
e02119d5 CM |
3396 | } |
3397 | ||
3398 | /* update the directory size in the log to reflect the names | |
3399 | * we have removed | |
3400 | */ | |
3401 | if (bytes_del) { | |
3402 | struct btrfs_key key; | |
3403 | ||
33345d01 | 3404 | key.objectid = dir_ino; |
e02119d5 CM |
3405 | key.offset = 0; |
3406 | key.type = BTRFS_INODE_ITEM_KEY; | |
b3b4aa74 | 3407 | btrfs_release_path(path); |
e02119d5 CM |
3408 | |
3409 | ret = btrfs_search_slot(trans, log, &key, path, 0, 1); | |
4a500fd1 YZ |
3410 | if (ret < 0) { |
3411 | err = ret; | |
3412 | goto fail; | |
3413 | } | |
e02119d5 CM |
3414 | if (ret == 0) { |
3415 | struct btrfs_inode_item *item; | |
3416 | u64 i_size; | |
3417 | ||
3418 | item = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
3419 | struct btrfs_inode_item); | |
3420 | i_size = btrfs_inode_size(path->nodes[0], item); | |
3421 | if (i_size > bytes_del) | |
3422 | i_size -= bytes_del; | |
3423 | else | |
3424 | i_size = 0; | |
3425 | btrfs_set_inode_size(path->nodes[0], item, i_size); | |
3426 | btrfs_mark_buffer_dirty(path->nodes[0]); | |
3427 | } else | |
3428 | ret = 0; | |
b3b4aa74 | 3429 | btrfs_release_path(path); |
e02119d5 | 3430 | } |
4a500fd1 | 3431 | fail: |
e02119d5 | 3432 | btrfs_free_path(path); |
a62f44a5 | 3433 | out_unlock: |
49f34d1f | 3434 | mutex_unlock(&dir->log_mutex); |
4a500fd1 | 3435 | if (ret == -ENOSPC) { |
995946dd | 3436 | btrfs_set_log_full_commit(root->fs_info, trans); |
4a500fd1 | 3437 | ret = 0; |
79787eaa | 3438 | } else if (ret < 0) |
66642832 | 3439 | btrfs_abort_transaction(trans, ret); |
79787eaa | 3440 | |
12fcfd22 | 3441 | btrfs_end_log_trans(root); |
e02119d5 | 3442 | |
411fc6bc | 3443 | return err; |
e02119d5 CM |
3444 | } |
3445 | ||
3446 | /* see comments for btrfs_del_dir_entries_in_log */ | |
3447 | int btrfs_del_inode_ref_in_log(struct btrfs_trans_handle *trans, | |
3448 | struct btrfs_root *root, | |
3449 | const char *name, int name_len, | |
a491abb2 | 3450 | struct btrfs_inode *inode, u64 dirid) |
e02119d5 | 3451 | { |
0b246afa | 3452 | struct btrfs_fs_info *fs_info = root->fs_info; |
e02119d5 CM |
3453 | struct btrfs_root *log; |
3454 | u64 index; | |
3455 | int ret; | |
3456 | ||
a491abb2 | 3457 | if (inode->logged_trans < trans->transid) |
3a5f1d45 CM |
3458 | return 0; |
3459 | ||
e02119d5 CM |
3460 | ret = join_running_log_trans(root); |
3461 | if (ret) | |
3462 | return 0; | |
3463 | log = root->log_root; | |
a491abb2 | 3464 | mutex_lock(&inode->log_mutex); |
e02119d5 | 3465 | |
a491abb2 | 3466 | ret = btrfs_del_inode_ref(trans, log, name, name_len, btrfs_ino(inode), |
e02119d5 | 3467 | dirid, &index); |
a491abb2 | 3468 | mutex_unlock(&inode->log_mutex); |
4a500fd1 | 3469 | if (ret == -ENOSPC) { |
0b246afa | 3470 | btrfs_set_log_full_commit(fs_info, trans); |
4a500fd1 | 3471 | ret = 0; |
79787eaa | 3472 | } else if (ret < 0 && ret != -ENOENT) |
66642832 | 3473 | btrfs_abort_transaction(trans, ret); |
12fcfd22 | 3474 | btrfs_end_log_trans(root); |
e02119d5 | 3475 | |
e02119d5 CM |
3476 | return ret; |
3477 | } | |
3478 | ||
3479 | /* | |
3480 | * creates a range item in the log for 'dirid'. first_offset and | |
3481 | * last_offset tell us which parts of the key space the log should | |
3482 | * be considered authoritative for. | |
3483 | */ | |
3484 | static noinline int insert_dir_log_key(struct btrfs_trans_handle *trans, | |
3485 | struct btrfs_root *log, | |
3486 | struct btrfs_path *path, | |
3487 | int key_type, u64 dirid, | |
3488 | u64 first_offset, u64 last_offset) | |
3489 | { | |
3490 | int ret; | |
3491 | struct btrfs_key key; | |
3492 | struct btrfs_dir_log_item *item; | |
3493 | ||
3494 | key.objectid = dirid; | |
3495 | key.offset = first_offset; | |
3496 | if (key_type == BTRFS_DIR_ITEM_KEY) | |
3497 | key.type = BTRFS_DIR_LOG_ITEM_KEY; | |
3498 | else | |
3499 | key.type = BTRFS_DIR_LOG_INDEX_KEY; | |
3500 | ret = btrfs_insert_empty_item(trans, log, path, &key, sizeof(*item)); | |
4a500fd1 YZ |
3501 | if (ret) |
3502 | return ret; | |
e02119d5 CM |
3503 | |
3504 | item = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
3505 | struct btrfs_dir_log_item); | |
3506 | btrfs_set_dir_log_end(path->nodes[0], item, last_offset); | |
3507 | btrfs_mark_buffer_dirty(path->nodes[0]); | |
b3b4aa74 | 3508 | btrfs_release_path(path); |
e02119d5 CM |
3509 | return 0; |
3510 | } | |
3511 | ||
3512 | /* | |
3513 | * log all the items included in the current transaction for a given | |
3514 | * directory. This also creates the range items in the log tree required | |
3515 | * to replay anything deleted before the fsync | |
3516 | */ | |
3517 | static noinline int log_dir_items(struct btrfs_trans_handle *trans, | |
684a5773 | 3518 | struct btrfs_root *root, struct btrfs_inode *inode, |
e02119d5 CM |
3519 | struct btrfs_path *path, |
3520 | struct btrfs_path *dst_path, int key_type, | |
2f2ff0ee | 3521 | struct btrfs_log_ctx *ctx, |
e02119d5 CM |
3522 | u64 min_offset, u64 *last_offset_ret) |
3523 | { | |
3524 | struct btrfs_key min_key; | |
e02119d5 CM |
3525 | struct btrfs_root *log = root->log_root; |
3526 | struct extent_buffer *src; | |
4a500fd1 | 3527 | int err = 0; |
e02119d5 CM |
3528 | int ret; |
3529 | int i; | |
3530 | int nritems; | |
3531 | u64 first_offset = min_offset; | |
3532 | u64 last_offset = (u64)-1; | |
684a5773 | 3533 | u64 ino = btrfs_ino(inode); |
e02119d5 CM |
3534 | |
3535 | log = root->log_root; | |
e02119d5 | 3536 | |
33345d01 | 3537 | min_key.objectid = ino; |
e02119d5 CM |
3538 | min_key.type = key_type; |
3539 | min_key.offset = min_offset; | |
3540 | ||
6174d3cb | 3541 | ret = btrfs_search_forward(root, &min_key, path, trans->transid); |
e02119d5 CM |
3542 | |
3543 | /* | |
3544 | * we didn't find anything from this transaction, see if there | |
3545 | * is anything at all | |
3546 | */ | |
33345d01 LZ |
3547 | if (ret != 0 || min_key.objectid != ino || min_key.type != key_type) { |
3548 | min_key.objectid = ino; | |
e02119d5 CM |
3549 | min_key.type = key_type; |
3550 | min_key.offset = (u64)-1; | |
b3b4aa74 | 3551 | btrfs_release_path(path); |
e02119d5 CM |
3552 | ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0); |
3553 | if (ret < 0) { | |
b3b4aa74 | 3554 | btrfs_release_path(path); |
e02119d5 CM |
3555 | return ret; |
3556 | } | |
33345d01 | 3557 | ret = btrfs_previous_item(root, path, ino, key_type); |
e02119d5 CM |
3558 | |
3559 | /* if ret == 0 there are items for this type, | |
3560 | * create a range to tell us the last key of this type. | |
3561 | * otherwise, there are no items in this directory after | |
3562 | * *min_offset, and we create a range to indicate that. | |
3563 | */ | |
3564 | if (ret == 0) { | |
3565 | struct btrfs_key tmp; | |
3566 | btrfs_item_key_to_cpu(path->nodes[0], &tmp, | |
3567 | path->slots[0]); | |
d397712b | 3568 | if (key_type == tmp.type) |
e02119d5 | 3569 | first_offset = max(min_offset, tmp.offset) + 1; |
e02119d5 CM |
3570 | } |
3571 | goto done; | |
3572 | } | |
3573 | ||
3574 | /* go backward to find any previous key */ | |
33345d01 | 3575 | ret = btrfs_previous_item(root, path, ino, key_type); |
e02119d5 CM |
3576 | if (ret == 0) { |
3577 | struct btrfs_key tmp; | |
3578 | btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]); | |
3579 | if (key_type == tmp.type) { | |
3580 | first_offset = tmp.offset; | |
3581 | ret = overwrite_item(trans, log, dst_path, | |
3582 | path->nodes[0], path->slots[0], | |
3583 | &tmp); | |
4a500fd1 YZ |
3584 | if (ret) { |
3585 | err = ret; | |
3586 | goto done; | |
3587 | } | |
e02119d5 CM |
3588 | } |
3589 | } | |
b3b4aa74 | 3590 | btrfs_release_path(path); |
e02119d5 | 3591 | |
2cc83342 JB |
3592 | /* |
3593 | * Find the first key from this transaction again. See the note for | |
3594 | * log_new_dir_dentries, if we're logging a directory recursively we | |
3595 | * won't be holding its i_mutex, which means we can modify the directory | |
3596 | * while we're logging it. If we remove an entry between our first | |
3597 | * search and this search we'll not find the key again and can just | |
3598 | * bail. | |
3599 | */ | |
e02119d5 | 3600 | ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0); |
2cc83342 | 3601 | if (ret != 0) |
e02119d5 | 3602 | goto done; |
e02119d5 CM |
3603 | |
3604 | /* | |
3605 | * we have a block from this transaction, log every item in it | |
3606 | * from our directory | |
3607 | */ | |
d397712b | 3608 | while (1) { |
e02119d5 CM |
3609 | struct btrfs_key tmp; |
3610 | src = path->nodes[0]; | |
3611 | nritems = btrfs_header_nritems(src); | |
3612 | for (i = path->slots[0]; i < nritems; i++) { | |
2f2ff0ee FM |
3613 | struct btrfs_dir_item *di; |
3614 | ||
e02119d5 CM |
3615 | btrfs_item_key_to_cpu(src, &min_key, i); |
3616 | ||
33345d01 | 3617 | if (min_key.objectid != ino || min_key.type != key_type) |
e02119d5 CM |
3618 | goto done; |
3619 | ret = overwrite_item(trans, log, dst_path, src, i, | |
3620 | &min_key); | |
4a500fd1 YZ |
3621 | if (ret) { |
3622 | err = ret; | |
3623 | goto done; | |
3624 | } | |
2f2ff0ee FM |
3625 | |
3626 | /* | |
3627 | * We must make sure that when we log a directory entry, | |
3628 | * the corresponding inode, after log replay, has a | |
3629 | * matching link count. For example: | |
3630 | * | |
3631 | * touch foo | |
3632 | * mkdir mydir | |
3633 | * sync | |
3634 | * ln foo mydir/bar | |
3635 | * xfs_io -c "fsync" mydir | |
3636 | * <crash> | |
3637 | * <mount fs and log replay> | |
3638 | * | |
3639 | * Would result in a fsync log that when replayed, our | |
3640 | * file inode would have a link count of 1, but we get | |
3641 | * two directory entries pointing to the same inode. | |
3642 | * After removing one of the names, it would not be | |
3643 | * possible to remove the other name, which resulted | |
3644 | * always in stale file handle errors, and would not | |
3645 | * be possible to rmdir the parent directory, since | |
3646 | * its i_size could never decrement to the value | |
3647 | * BTRFS_EMPTY_DIR_SIZE, resulting in -ENOTEMPTY errors. | |
3648 | */ | |
3649 | di = btrfs_item_ptr(src, i, struct btrfs_dir_item); | |
3650 | btrfs_dir_item_key_to_cpu(src, di, &tmp); | |
3651 | if (ctx && | |
3652 | (btrfs_dir_transid(src, di) == trans->transid || | |
3653 | btrfs_dir_type(src, di) == BTRFS_FT_DIR) && | |
3654 | tmp.type != BTRFS_ROOT_ITEM_KEY) | |
3655 | ctx->log_new_dentries = true; | |
e02119d5 CM |
3656 | } |
3657 | path->slots[0] = nritems; | |
3658 | ||
3659 | /* | |
3660 | * look ahead to the next item and see if it is also | |
3661 | * from this directory and from this transaction | |
3662 | */ | |
3663 | ret = btrfs_next_leaf(root, path); | |
80c0b421 LB |
3664 | if (ret) { |
3665 | if (ret == 1) | |
3666 | last_offset = (u64)-1; | |
3667 | else | |
3668 | err = ret; | |
e02119d5 CM |
3669 | goto done; |
3670 | } | |
3671 | btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]); | |
33345d01 | 3672 | if (tmp.objectid != ino || tmp.type != key_type) { |
e02119d5 CM |
3673 | last_offset = (u64)-1; |
3674 | goto done; | |
3675 | } | |
3676 | if (btrfs_header_generation(path->nodes[0]) != trans->transid) { | |
3677 | ret = overwrite_item(trans, log, dst_path, | |
3678 | path->nodes[0], path->slots[0], | |
3679 | &tmp); | |
4a500fd1 YZ |
3680 | if (ret) |
3681 | err = ret; | |
3682 | else | |
3683 | last_offset = tmp.offset; | |
e02119d5 CM |
3684 | goto done; |
3685 | } | |
3686 | } | |
3687 | done: | |
b3b4aa74 DS |
3688 | btrfs_release_path(path); |
3689 | btrfs_release_path(dst_path); | |
e02119d5 | 3690 | |
4a500fd1 YZ |
3691 | if (err == 0) { |
3692 | *last_offset_ret = last_offset; | |
3693 | /* | |
3694 | * insert the log range keys to indicate where the log | |
3695 | * is valid | |
3696 | */ | |
3697 | ret = insert_dir_log_key(trans, log, path, key_type, | |
33345d01 | 3698 | ino, first_offset, last_offset); |
4a500fd1 YZ |
3699 | if (ret) |
3700 | err = ret; | |
3701 | } | |
3702 | return err; | |
e02119d5 CM |
3703 | } |
3704 | ||
3705 | /* | |
3706 | * logging directories is very similar to logging inodes, We find all the items | |
3707 | * from the current transaction and write them to the log. | |
3708 | * | |
3709 | * The recovery code scans the directory in the subvolume, and if it finds a | |
3710 | * key in the range logged that is not present in the log tree, then it means | |
3711 | * that dir entry was unlinked during the transaction. | |
3712 | * | |
3713 | * In order for that scan to work, we must include one key smaller than | |
3714 | * the smallest logged by this transaction and one key larger than the largest | |
3715 | * key logged by this transaction. | |
3716 | */ | |
3717 | static noinline int log_directory_changes(struct btrfs_trans_handle *trans, | |
dbf39ea4 | 3718 | struct btrfs_root *root, struct btrfs_inode *inode, |
e02119d5 | 3719 | struct btrfs_path *path, |
2f2ff0ee FM |
3720 | struct btrfs_path *dst_path, |
3721 | struct btrfs_log_ctx *ctx) | |
e02119d5 CM |
3722 | { |
3723 | u64 min_key; | |
3724 | u64 max_key; | |
3725 | int ret; | |
3726 | int key_type = BTRFS_DIR_ITEM_KEY; | |
3727 | ||
3728 | again: | |
3729 | min_key = 0; | |
3730 | max_key = 0; | |
d397712b | 3731 | while (1) { |
dbf39ea4 NB |
3732 | ret = log_dir_items(trans, root, inode, path, dst_path, key_type, |
3733 | ctx, min_key, &max_key); | |
4a500fd1 YZ |
3734 | if (ret) |
3735 | return ret; | |
e02119d5 CM |
3736 | if (max_key == (u64)-1) |
3737 | break; | |
3738 | min_key = max_key + 1; | |
3739 | } | |
3740 | ||
3741 | if (key_type == BTRFS_DIR_ITEM_KEY) { | |
3742 | key_type = BTRFS_DIR_INDEX_KEY; | |
3743 | goto again; | |
3744 | } | |
3745 | return 0; | |
3746 | } | |
3747 | ||
3748 | /* | |
3749 | * a helper function to drop items from the log before we relog an | |
3750 | * inode. max_key_type indicates the highest item type to remove. | |
3751 | * This cannot be run for file data extents because it does not | |
3752 | * free the extents they point to. | |
3753 | */ | |
3754 | static int drop_objectid_items(struct btrfs_trans_handle *trans, | |
3755 | struct btrfs_root *log, | |
3756 | struct btrfs_path *path, | |
3757 | u64 objectid, int max_key_type) | |
3758 | { | |
3759 | int ret; | |
3760 | struct btrfs_key key; | |
3761 | struct btrfs_key found_key; | |
18ec90d6 | 3762 | int start_slot; |
e02119d5 CM |
3763 | |
3764 | key.objectid = objectid; | |
3765 | key.type = max_key_type; | |
3766 | key.offset = (u64)-1; | |
3767 | ||
d397712b | 3768 | while (1) { |
e02119d5 | 3769 | ret = btrfs_search_slot(trans, log, &key, path, -1, 1); |
3650860b | 3770 | BUG_ON(ret == 0); /* Logic error */ |
4a500fd1 | 3771 | if (ret < 0) |
e02119d5 CM |
3772 | break; |
3773 | ||
3774 | if (path->slots[0] == 0) | |
3775 | break; | |
3776 | ||
3777 | path->slots[0]--; | |
3778 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, | |
3779 | path->slots[0]); | |
3780 | ||
3781 | if (found_key.objectid != objectid) | |
3782 | break; | |
3783 | ||
18ec90d6 JB |
3784 | found_key.offset = 0; |
3785 | found_key.type = 0; | |
3786 | ret = btrfs_bin_search(path->nodes[0], &found_key, 0, | |
3787 | &start_slot); | |
cbca7d59 FM |
3788 | if (ret < 0) |
3789 | break; | |
18ec90d6 JB |
3790 | |
3791 | ret = btrfs_del_items(trans, log, path, start_slot, | |
3792 | path->slots[0] - start_slot + 1); | |
3793 | /* | |
3794 | * If start slot isn't 0 then we don't need to re-search, we've | |
3795 | * found the last guy with the objectid in this tree. | |
3796 | */ | |
3797 | if (ret || start_slot != 0) | |
65a246c5 | 3798 | break; |
b3b4aa74 | 3799 | btrfs_release_path(path); |
e02119d5 | 3800 | } |
b3b4aa74 | 3801 | btrfs_release_path(path); |
5bdbeb21 JB |
3802 | if (ret > 0) |
3803 | ret = 0; | |
4a500fd1 | 3804 | return ret; |
e02119d5 CM |
3805 | } |
3806 | ||
94edf4ae JB |
3807 | static void fill_inode_item(struct btrfs_trans_handle *trans, |
3808 | struct extent_buffer *leaf, | |
3809 | struct btrfs_inode_item *item, | |
1a4bcf47 FM |
3810 | struct inode *inode, int log_inode_only, |
3811 | u64 logged_isize) | |
94edf4ae | 3812 | { |
0b1c6cca JB |
3813 | struct btrfs_map_token token; |
3814 | ||
3815 | btrfs_init_map_token(&token); | |
94edf4ae JB |
3816 | |
3817 | if (log_inode_only) { | |
3818 | /* set the generation to zero so the recover code | |
3819 | * can tell the difference between an logging | |
3820 | * just to say 'this inode exists' and a logging | |
3821 | * to say 'update this inode with these values' | |
3822 | */ | |
0b1c6cca | 3823 | btrfs_set_token_inode_generation(leaf, item, 0, &token); |
1a4bcf47 | 3824 | btrfs_set_token_inode_size(leaf, item, logged_isize, &token); |
94edf4ae | 3825 | } else { |
0b1c6cca JB |
3826 | btrfs_set_token_inode_generation(leaf, item, |
3827 | BTRFS_I(inode)->generation, | |
3828 | &token); | |
3829 | btrfs_set_token_inode_size(leaf, item, inode->i_size, &token); | |
3830 | } | |
3831 | ||
3832 | btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token); | |
3833 | btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token); | |
3834 | btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token); | |
3835 | btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token); | |
3836 | ||
a937b979 | 3837 | btrfs_set_token_timespec_sec(leaf, &item->atime, |
0b1c6cca | 3838 | inode->i_atime.tv_sec, &token); |
a937b979 | 3839 | btrfs_set_token_timespec_nsec(leaf, &item->atime, |
0b1c6cca JB |
3840 | inode->i_atime.tv_nsec, &token); |
3841 | ||
a937b979 | 3842 | btrfs_set_token_timespec_sec(leaf, &item->mtime, |
0b1c6cca | 3843 | inode->i_mtime.tv_sec, &token); |
a937b979 | 3844 | btrfs_set_token_timespec_nsec(leaf, &item->mtime, |
0b1c6cca JB |
3845 | inode->i_mtime.tv_nsec, &token); |
3846 | ||
a937b979 | 3847 | btrfs_set_token_timespec_sec(leaf, &item->ctime, |
0b1c6cca | 3848 | inode->i_ctime.tv_sec, &token); |
a937b979 | 3849 | btrfs_set_token_timespec_nsec(leaf, &item->ctime, |
0b1c6cca JB |
3850 | inode->i_ctime.tv_nsec, &token); |
3851 | ||
3852 | btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode), | |
3853 | &token); | |
3854 | ||
c7f88c4e JL |
3855 | btrfs_set_token_inode_sequence(leaf, item, |
3856 | inode_peek_iversion(inode), &token); | |
0b1c6cca JB |
3857 | btrfs_set_token_inode_transid(leaf, item, trans->transid, &token); |
3858 | btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token); | |
3859 | btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token); | |
3860 | btrfs_set_token_inode_block_group(leaf, item, 0, &token); | |
94edf4ae JB |
3861 | } |
3862 | ||
a95249b3 JB |
3863 | static int log_inode_item(struct btrfs_trans_handle *trans, |
3864 | struct btrfs_root *log, struct btrfs_path *path, | |
6d889a3b | 3865 | struct btrfs_inode *inode) |
a95249b3 JB |
3866 | { |
3867 | struct btrfs_inode_item *inode_item; | |
a95249b3 JB |
3868 | int ret; |
3869 | ||
efd0c405 | 3870 | ret = btrfs_insert_empty_item(trans, log, path, |
6d889a3b | 3871 | &inode->location, sizeof(*inode_item)); |
a95249b3 JB |
3872 | if (ret && ret != -EEXIST) |
3873 | return ret; | |
3874 | inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
3875 | struct btrfs_inode_item); | |
6d889a3b NB |
3876 | fill_inode_item(trans, path->nodes[0], inode_item, &inode->vfs_inode, |
3877 | 0, 0); | |
a95249b3 JB |
3878 | btrfs_release_path(path); |
3879 | return 0; | |
3880 | } | |
3881 | ||
31ff1cd2 | 3882 | static noinline int copy_items(struct btrfs_trans_handle *trans, |
44d70e19 | 3883 | struct btrfs_inode *inode, |
31ff1cd2 | 3884 | struct btrfs_path *dst_path, |
16e7549f | 3885 | struct btrfs_path *src_path, u64 *last_extent, |
1a4bcf47 FM |
3886 | int start_slot, int nr, int inode_only, |
3887 | u64 logged_isize) | |
31ff1cd2 | 3888 | { |
3ffbd68c | 3889 | struct btrfs_fs_info *fs_info = trans->fs_info; |
31ff1cd2 CM |
3890 | unsigned long src_offset; |
3891 | unsigned long dst_offset; | |
44d70e19 | 3892 | struct btrfs_root *log = inode->root->log_root; |
31ff1cd2 CM |
3893 | struct btrfs_file_extent_item *extent; |
3894 | struct btrfs_inode_item *inode_item; | |
16e7549f JB |
3895 | struct extent_buffer *src = src_path->nodes[0]; |
3896 | struct btrfs_key first_key, last_key, key; | |
31ff1cd2 CM |
3897 | int ret; |
3898 | struct btrfs_key *ins_keys; | |
3899 | u32 *ins_sizes; | |
3900 | char *ins_data; | |
3901 | int i; | |
d20f7043 | 3902 | struct list_head ordered_sums; |
44d70e19 | 3903 | int skip_csum = inode->flags & BTRFS_INODE_NODATASUM; |
16e7549f | 3904 | bool has_extents = false; |
74121f7c | 3905 | bool need_find_last_extent = true; |
16e7549f | 3906 | bool done = false; |
d20f7043 CM |
3907 | |
3908 | INIT_LIST_HEAD(&ordered_sums); | |
31ff1cd2 CM |
3909 | |
3910 | ins_data = kmalloc(nr * sizeof(struct btrfs_key) + | |
3911 | nr * sizeof(u32), GFP_NOFS); | |
2a29edc6 | 3912 | if (!ins_data) |
3913 | return -ENOMEM; | |
3914 | ||
16e7549f JB |
3915 | first_key.objectid = (u64)-1; |
3916 | ||
31ff1cd2 CM |
3917 | ins_sizes = (u32 *)ins_data; |
3918 | ins_keys = (struct btrfs_key *)(ins_data + nr * sizeof(u32)); | |
3919 | ||
3920 | for (i = 0; i < nr; i++) { | |
3921 | ins_sizes[i] = btrfs_item_size_nr(src, i + start_slot); | |
3922 | btrfs_item_key_to_cpu(src, ins_keys + i, i + start_slot); | |
3923 | } | |
3924 | ret = btrfs_insert_empty_items(trans, log, dst_path, | |
3925 | ins_keys, ins_sizes, nr); | |
4a500fd1 YZ |
3926 | if (ret) { |
3927 | kfree(ins_data); | |
3928 | return ret; | |
3929 | } | |
31ff1cd2 | 3930 | |
5d4f98a2 | 3931 | for (i = 0; i < nr; i++, dst_path->slots[0]++) { |
31ff1cd2 CM |
3932 | dst_offset = btrfs_item_ptr_offset(dst_path->nodes[0], |
3933 | dst_path->slots[0]); | |
3934 | ||
3935 | src_offset = btrfs_item_ptr_offset(src, start_slot + i); | |
3936 | ||
0dde10be | 3937 | if (i == nr - 1) |
16e7549f JB |
3938 | last_key = ins_keys[i]; |
3939 | ||
94edf4ae | 3940 | if (ins_keys[i].type == BTRFS_INODE_ITEM_KEY) { |
31ff1cd2 CM |
3941 | inode_item = btrfs_item_ptr(dst_path->nodes[0], |
3942 | dst_path->slots[0], | |
3943 | struct btrfs_inode_item); | |
94edf4ae | 3944 | fill_inode_item(trans, dst_path->nodes[0], inode_item, |
f85b7379 DS |
3945 | &inode->vfs_inode, |
3946 | inode_only == LOG_INODE_EXISTS, | |
1a4bcf47 | 3947 | logged_isize); |
94edf4ae JB |
3948 | } else { |
3949 | copy_extent_buffer(dst_path->nodes[0], src, dst_offset, | |
3950 | src_offset, ins_sizes[i]); | |
31ff1cd2 | 3951 | } |
94edf4ae | 3952 | |
16e7549f JB |
3953 | /* |
3954 | * We set need_find_last_extent here in case we know we were | |
3955 | * processing other items and then walk into the first extent in | |
3956 | * the inode. If we don't hit an extent then nothing changes, | |
3957 | * we'll do the last search the next time around. | |
3958 | */ | |
3959 | if (ins_keys[i].type == BTRFS_EXTENT_DATA_KEY) { | |
3960 | has_extents = true; | |
74121f7c | 3961 | if (first_key.objectid == (u64)-1) |
16e7549f JB |
3962 | first_key = ins_keys[i]; |
3963 | } else { | |
3964 | need_find_last_extent = false; | |
3965 | } | |
3966 | ||
31ff1cd2 CM |
3967 | /* take a reference on file data extents so that truncates |
3968 | * or deletes of this inode don't have to relog the inode | |
3969 | * again | |
3970 | */ | |
962a298f | 3971 | if (ins_keys[i].type == BTRFS_EXTENT_DATA_KEY && |
d2794405 | 3972 | !skip_csum) { |
31ff1cd2 CM |
3973 | int found_type; |
3974 | extent = btrfs_item_ptr(src, start_slot + i, | |
3975 | struct btrfs_file_extent_item); | |
3976 | ||
8e531cdf | 3977 | if (btrfs_file_extent_generation(src, extent) < trans->transid) |
3978 | continue; | |
3979 | ||
31ff1cd2 | 3980 | found_type = btrfs_file_extent_type(src, extent); |
6f1fed77 | 3981 | if (found_type == BTRFS_FILE_EXTENT_REG) { |
5d4f98a2 YZ |
3982 | u64 ds, dl, cs, cl; |
3983 | ds = btrfs_file_extent_disk_bytenr(src, | |
3984 | extent); | |
3985 | /* ds == 0 is a hole */ | |
3986 | if (ds == 0) | |
3987 | continue; | |
3988 | ||
3989 | dl = btrfs_file_extent_disk_num_bytes(src, | |
3990 | extent); | |
3991 | cs = btrfs_file_extent_offset(src, extent); | |
3992 | cl = btrfs_file_extent_num_bytes(src, | |
a419aef8 | 3993 | extent); |
580afd76 CM |
3994 | if (btrfs_file_extent_compression(src, |
3995 | extent)) { | |
3996 | cs = 0; | |
3997 | cl = dl; | |
3998 | } | |
5d4f98a2 YZ |
3999 | |
4000 | ret = btrfs_lookup_csums_range( | |
0b246afa | 4001 | fs_info->csum_root, |
5d4f98a2 | 4002 | ds + cs, ds + cs + cl - 1, |
a2de733c | 4003 | &ordered_sums, 0); |
3650860b JB |
4004 | if (ret) { |
4005 | btrfs_release_path(dst_path); | |
4006 | kfree(ins_data); | |
4007 | return ret; | |
4008 | } | |
31ff1cd2 CM |
4009 | } |
4010 | } | |
31ff1cd2 CM |
4011 | } |
4012 | ||
4013 | btrfs_mark_buffer_dirty(dst_path->nodes[0]); | |
b3b4aa74 | 4014 | btrfs_release_path(dst_path); |
31ff1cd2 | 4015 | kfree(ins_data); |
d20f7043 CM |
4016 | |
4017 | /* | |
4018 | * we have to do this after the loop above to avoid changing the | |
4019 | * log tree while trying to change the log tree. | |
4020 | */ | |
4a500fd1 | 4021 | ret = 0; |
d397712b | 4022 | while (!list_empty(&ordered_sums)) { |
d20f7043 CM |
4023 | struct btrfs_ordered_sum *sums = list_entry(ordered_sums.next, |
4024 | struct btrfs_ordered_sum, | |
4025 | list); | |
4a500fd1 YZ |
4026 | if (!ret) |
4027 | ret = btrfs_csum_file_blocks(trans, log, sums); | |
d20f7043 CM |
4028 | list_del(&sums->list); |
4029 | kfree(sums); | |
4030 | } | |
16e7549f JB |
4031 | |
4032 | if (!has_extents) | |
4033 | return ret; | |
4034 | ||
74121f7c FM |
4035 | if (need_find_last_extent && *last_extent == first_key.offset) { |
4036 | /* | |
4037 | * We don't have any leafs between our current one and the one | |
4038 | * we processed before that can have file extent items for our | |
4039 | * inode (and have a generation number smaller than our current | |
4040 | * transaction id). | |
4041 | */ | |
4042 | need_find_last_extent = false; | |
4043 | } | |
4044 | ||
16e7549f JB |
4045 | /* |
4046 | * Because we use btrfs_search_forward we could skip leaves that were | |
4047 | * not modified and then assume *last_extent is valid when it really | |
4048 | * isn't. So back up to the previous leaf and read the end of the last | |
4049 | * extent before we go and fill in holes. | |
4050 | */ | |
4051 | if (need_find_last_extent) { | |
4052 | u64 len; | |
4053 | ||
44d70e19 | 4054 | ret = btrfs_prev_leaf(inode->root, src_path); |
16e7549f JB |
4055 | if (ret < 0) |
4056 | return ret; | |
4057 | if (ret) | |
4058 | goto fill_holes; | |
4059 | if (src_path->slots[0]) | |
4060 | src_path->slots[0]--; | |
4061 | src = src_path->nodes[0]; | |
4062 | btrfs_item_key_to_cpu(src, &key, src_path->slots[0]); | |
44d70e19 | 4063 | if (key.objectid != btrfs_ino(inode) || |
16e7549f JB |
4064 | key.type != BTRFS_EXTENT_DATA_KEY) |
4065 | goto fill_holes; | |
4066 | extent = btrfs_item_ptr(src, src_path->slots[0], | |
4067 | struct btrfs_file_extent_item); | |
4068 | if (btrfs_file_extent_type(src, extent) == | |
4069 | BTRFS_FILE_EXTENT_INLINE) { | |
e41ca589 | 4070 | len = btrfs_file_extent_ram_bytes(src, extent); |
16e7549f | 4071 | *last_extent = ALIGN(key.offset + len, |
0b246afa | 4072 | fs_info->sectorsize); |
16e7549f JB |
4073 | } else { |
4074 | len = btrfs_file_extent_num_bytes(src, extent); | |
4075 | *last_extent = key.offset + len; | |
4076 | } | |
4077 | } | |
4078 | fill_holes: | |
4079 | /* So we did prev_leaf, now we need to move to the next leaf, but a few | |
4080 | * things could have happened | |
4081 | * | |
4082 | * 1) A merge could have happened, so we could currently be on a leaf | |
4083 | * that holds what we were copying in the first place. | |
4084 | * 2) A split could have happened, and now not all of the items we want | |
4085 | * are on the same leaf. | |
4086 | * | |
4087 | * So we need to adjust how we search for holes, we need to drop the | |
4088 | * path and re-search for the first extent key we found, and then walk | |
4089 | * forward until we hit the last one we copied. | |
4090 | */ | |
4091 | if (need_find_last_extent) { | |
4092 | /* btrfs_prev_leaf could return 1 without releasing the path */ | |
4093 | btrfs_release_path(src_path); | |
f85b7379 DS |
4094 | ret = btrfs_search_slot(NULL, inode->root, &first_key, |
4095 | src_path, 0, 0); | |
16e7549f JB |
4096 | if (ret < 0) |
4097 | return ret; | |
4098 | ASSERT(ret == 0); | |
4099 | src = src_path->nodes[0]; | |
4100 | i = src_path->slots[0]; | |
4101 | } else { | |
4102 | i = start_slot; | |
4103 | } | |
4104 | ||
4105 | /* | |
4106 | * Ok so here we need to go through and fill in any holes we may have | |
4107 | * to make sure that holes are punched for those areas in case they had | |
4108 | * extents previously. | |
4109 | */ | |
4110 | while (!done) { | |
4111 | u64 offset, len; | |
4112 | u64 extent_end; | |
4113 | ||
4114 | if (i >= btrfs_header_nritems(src_path->nodes[0])) { | |
44d70e19 | 4115 | ret = btrfs_next_leaf(inode->root, src_path); |
16e7549f JB |
4116 | if (ret < 0) |
4117 | return ret; | |
4118 | ASSERT(ret == 0); | |
4119 | src = src_path->nodes[0]; | |
4120 | i = 0; | |
8434ec46 | 4121 | need_find_last_extent = true; |
16e7549f JB |
4122 | } |
4123 | ||
4124 | btrfs_item_key_to_cpu(src, &key, i); | |
4125 | if (!btrfs_comp_cpu_keys(&key, &last_key)) | |
4126 | done = true; | |
44d70e19 | 4127 | if (key.objectid != btrfs_ino(inode) || |
16e7549f JB |
4128 | key.type != BTRFS_EXTENT_DATA_KEY) { |
4129 | i++; | |
4130 | continue; | |
4131 | } | |
4132 | extent = btrfs_item_ptr(src, i, struct btrfs_file_extent_item); | |
4133 | if (btrfs_file_extent_type(src, extent) == | |
4134 | BTRFS_FILE_EXTENT_INLINE) { | |
e41ca589 | 4135 | len = btrfs_file_extent_ram_bytes(src, extent); |
da17066c | 4136 | extent_end = ALIGN(key.offset + len, |
0b246afa | 4137 | fs_info->sectorsize); |
16e7549f JB |
4138 | } else { |
4139 | len = btrfs_file_extent_num_bytes(src, extent); | |
4140 | extent_end = key.offset + len; | |
4141 | } | |
4142 | i++; | |
4143 | ||
4144 | if (*last_extent == key.offset) { | |
4145 | *last_extent = extent_end; | |
4146 | continue; | |
4147 | } | |
4148 | offset = *last_extent; | |
4149 | len = key.offset - *last_extent; | |
44d70e19 | 4150 | ret = btrfs_insert_file_extent(trans, log, btrfs_ino(inode), |
f85b7379 | 4151 | offset, 0, 0, len, 0, len, 0, 0, 0); |
16e7549f JB |
4152 | if (ret) |
4153 | break; | |
74121f7c | 4154 | *last_extent = extent_end; |
16e7549f | 4155 | } |
4ee3fad3 FM |
4156 | |
4157 | /* | |
4158 | * Check if there is a hole between the last extent found in our leaf | |
4159 | * and the first extent in the next leaf. If there is one, we need to | |
4160 | * log an explicit hole so that at replay time we can punch the hole. | |
4161 | */ | |
4162 | if (ret == 0 && | |
4163 | key.objectid == btrfs_ino(inode) && | |
4164 | key.type == BTRFS_EXTENT_DATA_KEY && | |
4165 | i == btrfs_header_nritems(src_path->nodes[0])) { | |
4166 | ret = btrfs_next_leaf(inode->root, src_path); | |
4167 | need_find_last_extent = true; | |
4168 | if (ret > 0) { | |
4169 | ret = 0; | |
4170 | } else if (ret == 0) { | |
4171 | btrfs_item_key_to_cpu(src_path->nodes[0], &key, | |
4172 | src_path->slots[0]); | |
4173 | if (key.objectid == btrfs_ino(inode) && | |
4174 | key.type == BTRFS_EXTENT_DATA_KEY && | |
4175 | *last_extent < key.offset) { | |
4176 | const u64 len = key.offset - *last_extent; | |
4177 | ||
4178 | ret = btrfs_insert_file_extent(trans, log, | |
4179 | btrfs_ino(inode), | |
4180 | *last_extent, 0, | |
4181 | 0, len, 0, len, | |
4182 | 0, 0, 0); | |
4183 | } | |
4184 | } | |
4185 | } | |
16e7549f JB |
4186 | /* |
4187 | * Need to let the callers know we dropped the path so they should | |
4188 | * re-search. | |
4189 | */ | |
4190 | if (!ret && need_find_last_extent) | |
4191 | ret = 1; | |
4a500fd1 | 4192 | return ret; |
31ff1cd2 CM |
4193 | } |
4194 | ||
5dc562c5 JB |
4195 | static int extent_cmp(void *priv, struct list_head *a, struct list_head *b) |
4196 | { | |
4197 | struct extent_map *em1, *em2; | |
4198 | ||
4199 | em1 = list_entry(a, struct extent_map, list); | |
4200 | em2 = list_entry(b, struct extent_map, list); | |
4201 | ||
4202 | if (em1->start < em2->start) | |
4203 | return -1; | |
4204 | else if (em1->start > em2->start) | |
4205 | return 1; | |
4206 | return 0; | |
4207 | } | |
4208 | ||
e7175a69 JB |
4209 | static int log_extent_csums(struct btrfs_trans_handle *trans, |
4210 | struct btrfs_inode *inode, | |
a9ecb653 | 4211 | struct btrfs_root *log_root, |
e7175a69 | 4212 | const struct extent_map *em) |
5dc562c5 | 4213 | { |
2ab28f32 JB |
4214 | u64 csum_offset; |
4215 | u64 csum_len; | |
8407f553 FM |
4216 | LIST_HEAD(ordered_sums); |
4217 | int ret = 0; | |
0aa4a17d | 4218 | |
e7175a69 JB |
4219 | if (inode->flags & BTRFS_INODE_NODATASUM || |
4220 | test_bit(EXTENT_FLAG_PREALLOC, &em->flags) || | |
8407f553 | 4221 | em->block_start == EXTENT_MAP_HOLE) |
70c8a91c | 4222 | return 0; |
5dc562c5 | 4223 | |
e7175a69 | 4224 | /* If we're compressed we have to save the entire range of csums. */ |
488111aa FDBM |
4225 | if (em->compress_type) { |
4226 | csum_offset = 0; | |
8407f553 | 4227 | csum_len = max(em->block_len, em->orig_block_len); |
488111aa | 4228 | } else { |
e7175a69 JB |
4229 | csum_offset = em->mod_start - em->start; |
4230 | csum_len = em->mod_len; | |
488111aa | 4231 | } |
2ab28f32 | 4232 | |
70c8a91c | 4233 | /* block start is already adjusted for the file extent offset. */ |
a9ecb653 | 4234 | ret = btrfs_lookup_csums_range(trans->fs_info->csum_root, |
70c8a91c JB |
4235 | em->block_start + csum_offset, |
4236 | em->block_start + csum_offset + | |
4237 | csum_len - 1, &ordered_sums, 0); | |
4238 | if (ret) | |
4239 | return ret; | |
5dc562c5 | 4240 | |
70c8a91c JB |
4241 | while (!list_empty(&ordered_sums)) { |
4242 | struct btrfs_ordered_sum *sums = list_entry(ordered_sums.next, | |
4243 | struct btrfs_ordered_sum, | |
4244 | list); | |
4245 | if (!ret) | |
a9ecb653 | 4246 | ret = btrfs_csum_file_blocks(trans, log_root, sums); |
70c8a91c JB |
4247 | list_del(&sums->list); |
4248 | kfree(sums); | |
5dc562c5 JB |
4249 | } |
4250 | ||
70c8a91c | 4251 | return ret; |
5dc562c5 JB |
4252 | } |
4253 | ||
8407f553 | 4254 | static int log_one_extent(struct btrfs_trans_handle *trans, |
9d122629 | 4255 | struct btrfs_inode *inode, struct btrfs_root *root, |
8407f553 FM |
4256 | const struct extent_map *em, |
4257 | struct btrfs_path *path, | |
8407f553 FM |
4258 | struct btrfs_log_ctx *ctx) |
4259 | { | |
4260 | struct btrfs_root *log = root->log_root; | |
4261 | struct btrfs_file_extent_item *fi; | |
4262 | struct extent_buffer *leaf; | |
4263 | struct btrfs_map_token token; | |
4264 | struct btrfs_key key; | |
4265 | u64 extent_offset = em->start - em->orig_start; | |
4266 | u64 block_len; | |
4267 | int ret; | |
4268 | int extent_inserted = 0; | |
8407f553 | 4269 | |
a9ecb653 | 4270 | ret = log_extent_csums(trans, inode, log, em); |
8407f553 FM |
4271 | if (ret) |
4272 | return ret; | |
4273 | ||
8407f553 FM |
4274 | btrfs_init_map_token(&token); |
4275 | ||
9d122629 | 4276 | ret = __btrfs_drop_extents(trans, log, &inode->vfs_inode, path, em->start, |
8407f553 FM |
4277 | em->start + em->len, NULL, 0, 1, |
4278 | sizeof(*fi), &extent_inserted); | |
4279 | if (ret) | |
4280 | return ret; | |
4281 | ||
4282 | if (!extent_inserted) { | |
9d122629 | 4283 | key.objectid = btrfs_ino(inode); |
8407f553 FM |
4284 | key.type = BTRFS_EXTENT_DATA_KEY; |
4285 | key.offset = em->start; | |
4286 | ||
4287 | ret = btrfs_insert_empty_item(trans, log, path, &key, | |
4288 | sizeof(*fi)); | |
4289 | if (ret) | |
4290 | return ret; | |
4291 | } | |
4292 | leaf = path->nodes[0]; | |
4293 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
4294 | struct btrfs_file_extent_item); | |
4295 | ||
50d9aa99 | 4296 | btrfs_set_token_file_extent_generation(leaf, fi, trans->transid, |
8407f553 FM |
4297 | &token); |
4298 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) | |
4299 | btrfs_set_token_file_extent_type(leaf, fi, | |
4300 | BTRFS_FILE_EXTENT_PREALLOC, | |
4301 | &token); | |
4302 | else | |
4303 | btrfs_set_token_file_extent_type(leaf, fi, | |
4304 | BTRFS_FILE_EXTENT_REG, | |
4305 | &token); | |
4306 | ||
4307 | block_len = max(em->block_len, em->orig_block_len); | |
4308 | if (em->compress_type != BTRFS_COMPRESS_NONE) { | |
4309 | btrfs_set_token_file_extent_disk_bytenr(leaf, fi, | |
4310 | em->block_start, | |
4311 | &token); | |
4312 | btrfs_set_token_file_extent_disk_num_bytes(leaf, fi, block_len, | |
4313 | &token); | |
4314 | } else if (em->block_start < EXTENT_MAP_LAST_BYTE) { | |
4315 | btrfs_set_token_file_extent_disk_bytenr(leaf, fi, | |
4316 | em->block_start - | |
4317 | extent_offset, &token); | |
4318 | btrfs_set_token_file_extent_disk_num_bytes(leaf, fi, block_len, | |
4319 | &token); | |
4320 | } else { | |
4321 | btrfs_set_token_file_extent_disk_bytenr(leaf, fi, 0, &token); | |
4322 | btrfs_set_token_file_extent_disk_num_bytes(leaf, fi, 0, | |
4323 | &token); | |
4324 | } | |
4325 | ||
4326 | btrfs_set_token_file_extent_offset(leaf, fi, extent_offset, &token); | |
4327 | btrfs_set_token_file_extent_num_bytes(leaf, fi, em->len, &token); | |
4328 | btrfs_set_token_file_extent_ram_bytes(leaf, fi, em->ram_bytes, &token); | |
4329 | btrfs_set_token_file_extent_compression(leaf, fi, em->compress_type, | |
4330 | &token); | |
4331 | btrfs_set_token_file_extent_encryption(leaf, fi, 0, &token); | |
4332 | btrfs_set_token_file_extent_other_encoding(leaf, fi, 0, &token); | |
4333 | btrfs_mark_buffer_dirty(leaf); | |
4334 | ||
4335 | btrfs_release_path(path); | |
4336 | ||
4337 | return ret; | |
4338 | } | |
4339 | ||
31d11b83 FM |
4340 | /* |
4341 | * Log all prealloc extents beyond the inode's i_size to make sure we do not | |
4342 | * lose them after doing a fast fsync and replaying the log. We scan the | |
4343 | * subvolume's root instead of iterating the inode's extent map tree because | |
4344 | * otherwise we can log incorrect extent items based on extent map conversion. | |
4345 | * That can happen due to the fact that extent maps are merged when they | |
4346 | * are not in the extent map tree's list of modified extents. | |
4347 | */ | |
4348 | static int btrfs_log_prealloc_extents(struct btrfs_trans_handle *trans, | |
4349 | struct btrfs_inode *inode, | |
4350 | struct btrfs_path *path) | |
4351 | { | |
4352 | struct btrfs_root *root = inode->root; | |
4353 | struct btrfs_key key; | |
4354 | const u64 i_size = i_size_read(&inode->vfs_inode); | |
4355 | const u64 ino = btrfs_ino(inode); | |
4356 | struct btrfs_path *dst_path = NULL; | |
4357 | u64 last_extent = (u64)-1; | |
4358 | int ins_nr = 0; | |
4359 | int start_slot; | |
4360 | int ret; | |
4361 | ||
4362 | if (!(inode->flags & BTRFS_INODE_PREALLOC)) | |
4363 | return 0; | |
4364 | ||
4365 | key.objectid = ino; | |
4366 | key.type = BTRFS_EXTENT_DATA_KEY; | |
4367 | key.offset = i_size; | |
4368 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
4369 | if (ret < 0) | |
4370 | goto out; | |
4371 | ||
4372 | while (true) { | |
4373 | struct extent_buffer *leaf = path->nodes[0]; | |
4374 | int slot = path->slots[0]; | |
4375 | ||
4376 | if (slot >= btrfs_header_nritems(leaf)) { | |
4377 | if (ins_nr > 0) { | |
4378 | ret = copy_items(trans, inode, dst_path, path, | |
4379 | &last_extent, start_slot, | |
4380 | ins_nr, 1, 0); | |
4381 | if (ret < 0) | |
4382 | goto out; | |
4383 | ins_nr = 0; | |
4384 | } | |
4385 | ret = btrfs_next_leaf(root, path); | |
4386 | if (ret < 0) | |
4387 | goto out; | |
4388 | if (ret > 0) { | |
4389 | ret = 0; | |
4390 | break; | |
4391 | } | |
4392 | continue; | |
4393 | } | |
4394 | ||
4395 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
4396 | if (key.objectid > ino) | |
4397 | break; | |
4398 | if (WARN_ON_ONCE(key.objectid < ino) || | |
4399 | key.type < BTRFS_EXTENT_DATA_KEY || | |
4400 | key.offset < i_size) { | |
4401 | path->slots[0]++; | |
4402 | continue; | |
4403 | } | |
4404 | if (last_extent == (u64)-1) { | |
4405 | last_extent = key.offset; | |
4406 | /* | |
4407 | * Avoid logging extent items logged in past fsync calls | |
4408 | * and leading to duplicate keys in the log tree. | |
4409 | */ | |
4410 | do { | |
4411 | ret = btrfs_truncate_inode_items(trans, | |
4412 | root->log_root, | |
4413 | &inode->vfs_inode, | |
4414 | i_size, | |
4415 | BTRFS_EXTENT_DATA_KEY); | |
4416 | } while (ret == -EAGAIN); | |
4417 | if (ret) | |
4418 | goto out; | |
4419 | } | |
4420 | if (ins_nr == 0) | |
4421 | start_slot = slot; | |
4422 | ins_nr++; | |
4423 | path->slots[0]++; | |
4424 | if (!dst_path) { | |
4425 | dst_path = btrfs_alloc_path(); | |
4426 | if (!dst_path) { | |
4427 | ret = -ENOMEM; | |
4428 | goto out; | |
4429 | } | |
4430 | } | |
4431 | } | |
4432 | if (ins_nr > 0) { | |
4433 | ret = copy_items(trans, inode, dst_path, path, &last_extent, | |
4434 | start_slot, ins_nr, 1, 0); | |
4435 | if (ret > 0) | |
4436 | ret = 0; | |
4437 | } | |
4438 | out: | |
4439 | btrfs_release_path(path); | |
4440 | btrfs_free_path(dst_path); | |
4441 | return ret; | |
4442 | } | |
4443 | ||
5dc562c5 JB |
4444 | static int btrfs_log_changed_extents(struct btrfs_trans_handle *trans, |
4445 | struct btrfs_root *root, | |
9d122629 | 4446 | struct btrfs_inode *inode, |
827463c4 | 4447 | struct btrfs_path *path, |
de0ee0ed FM |
4448 | struct btrfs_log_ctx *ctx, |
4449 | const u64 start, | |
4450 | const u64 end) | |
5dc562c5 | 4451 | { |
5dc562c5 JB |
4452 | struct extent_map *em, *n; |
4453 | struct list_head extents; | |
9d122629 | 4454 | struct extent_map_tree *tree = &inode->extent_tree; |
5dc562c5 JB |
4455 | u64 test_gen; |
4456 | int ret = 0; | |
2ab28f32 | 4457 | int num = 0; |
5dc562c5 JB |
4458 | |
4459 | INIT_LIST_HEAD(&extents); | |
4460 | ||
5dc562c5 JB |
4461 | write_lock(&tree->lock); |
4462 | test_gen = root->fs_info->last_trans_committed; | |
4463 | ||
4464 | list_for_each_entry_safe(em, n, &tree->modified_extents, list) { | |
008c6753 FM |
4465 | /* |
4466 | * Skip extents outside our logging range. It's important to do | |
4467 | * it for correctness because if we don't ignore them, we may | |
4468 | * log them before their ordered extent completes, and therefore | |
4469 | * we could log them without logging their respective checksums | |
4470 | * (the checksum items are added to the csum tree at the very | |
4471 | * end of btrfs_finish_ordered_io()). Also leave such extents | |
4472 | * outside of our range in the list, since we may have another | |
4473 | * ranged fsync in the near future that needs them. If an extent | |
4474 | * outside our range corresponds to a hole, log it to avoid | |
4475 | * leaving gaps between extents (fsck will complain when we are | |
4476 | * not using the NO_HOLES feature). | |
4477 | */ | |
4478 | if ((em->start > end || em->start + em->len <= start) && | |
4479 | em->block_start != EXTENT_MAP_HOLE) | |
4480 | continue; | |
4481 | ||
5dc562c5 | 4482 | list_del_init(&em->list); |
2ab28f32 JB |
4483 | /* |
4484 | * Just an arbitrary number, this can be really CPU intensive | |
4485 | * once we start getting a lot of extents, and really once we | |
4486 | * have a bunch of extents we just want to commit since it will | |
4487 | * be faster. | |
4488 | */ | |
4489 | if (++num > 32768) { | |
4490 | list_del_init(&tree->modified_extents); | |
4491 | ret = -EFBIG; | |
4492 | goto process; | |
4493 | } | |
4494 | ||
5dc562c5 JB |
4495 | if (em->generation <= test_gen) |
4496 | continue; | |
8c6c5928 | 4497 | |
31d11b83 FM |
4498 | /* We log prealloc extents beyond eof later. */ |
4499 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) && | |
4500 | em->start >= i_size_read(&inode->vfs_inode)) | |
4501 | continue; | |
4502 | ||
ff44c6e3 | 4503 | /* Need a ref to keep it from getting evicted from cache */ |
490b54d6 | 4504 | refcount_inc(&em->refs); |
ff44c6e3 | 4505 | set_bit(EXTENT_FLAG_LOGGING, &em->flags); |
5dc562c5 | 4506 | list_add_tail(&em->list, &extents); |
2ab28f32 | 4507 | num++; |
5dc562c5 JB |
4508 | } |
4509 | ||
4510 | list_sort(NULL, &extents, extent_cmp); | |
2ab28f32 | 4511 | process: |
5dc562c5 JB |
4512 | while (!list_empty(&extents)) { |
4513 | em = list_entry(extents.next, struct extent_map, list); | |
4514 | ||
4515 | list_del_init(&em->list); | |
4516 | ||
4517 | /* | |
4518 | * If we had an error we just need to delete everybody from our | |
4519 | * private list. | |
4520 | */ | |
ff44c6e3 | 4521 | if (ret) { |
201a9038 | 4522 | clear_em_logging(tree, em); |
ff44c6e3 | 4523 | free_extent_map(em); |
5dc562c5 | 4524 | continue; |
ff44c6e3 JB |
4525 | } |
4526 | ||
4527 | write_unlock(&tree->lock); | |
5dc562c5 | 4528 | |
a2120a47 | 4529 | ret = log_one_extent(trans, inode, root, em, path, ctx); |
ff44c6e3 | 4530 | write_lock(&tree->lock); |
201a9038 JB |
4531 | clear_em_logging(tree, em); |
4532 | free_extent_map(em); | |
5dc562c5 | 4533 | } |
ff44c6e3 JB |
4534 | WARN_ON(!list_empty(&extents)); |
4535 | write_unlock(&tree->lock); | |
5dc562c5 | 4536 | |
5dc562c5 | 4537 | btrfs_release_path(path); |
31d11b83 FM |
4538 | if (!ret) |
4539 | ret = btrfs_log_prealloc_extents(trans, inode, path); | |
4540 | ||
5dc562c5 JB |
4541 | return ret; |
4542 | } | |
4543 | ||
481b01c0 | 4544 | static int logged_inode_size(struct btrfs_root *log, struct btrfs_inode *inode, |
1a4bcf47 FM |
4545 | struct btrfs_path *path, u64 *size_ret) |
4546 | { | |
4547 | struct btrfs_key key; | |
4548 | int ret; | |
4549 | ||
481b01c0 | 4550 | key.objectid = btrfs_ino(inode); |
1a4bcf47 FM |
4551 | key.type = BTRFS_INODE_ITEM_KEY; |
4552 | key.offset = 0; | |
4553 | ||
4554 | ret = btrfs_search_slot(NULL, log, &key, path, 0, 0); | |
4555 | if (ret < 0) { | |
4556 | return ret; | |
4557 | } else if (ret > 0) { | |
2f2ff0ee | 4558 | *size_ret = 0; |
1a4bcf47 FM |
4559 | } else { |
4560 | struct btrfs_inode_item *item; | |
4561 | ||
4562 | item = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
4563 | struct btrfs_inode_item); | |
4564 | *size_ret = btrfs_inode_size(path->nodes[0], item); | |
bf504110 FM |
4565 | /* |
4566 | * If the in-memory inode's i_size is smaller then the inode | |
4567 | * size stored in the btree, return the inode's i_size, so | |
4568 | * that we get a correct inode size after replaying the log | |
4569 | * when before a power failure we had a shrinking truncate | |
4570 | * followed by addition of a new name (rename / new hard link). | |
4571 | * Otherwise return the inode size from the btree, to avoid | |
4572 | * data loss when replaying a log due to previously doing a | |
4573 | * write that expands the inode's size and logging a new name | |
4574 | * immediately after. | |
4575 | */ | |
4576 | if (*size_ret > inode->vfs_inode.i_size) | |
4577 | *size_ret = inode->vfs_inode.i_size; | |
1a4bcf47 FM |
4578 | } |
4579 | ||
4580 | btrfs_release_path(path); | |
4581 | return 0; | |
4582 | } | |
4583 | ||
36283bf7 FM |
4584 | /* |
4585 | * At the moment we always log all xattrs. This is to figure out at log replay | |
4586 | * time which xattrs must have their deletion replayed. If a xattr is missing | |
4587 | * in the log tree and exists in the fs/subvol tree, we delete it. This is | |
4588 | * because if a xattr is deleted, the inode is fsynced and a power failure | |
4589 | * happens, causing the log to be replayed the next time the fs is mounted, | |
4590 | * we want the xattr to not exist anymore (same behaviour as other filesystems | |
4591 | * with a journal, ext3/4, xfs, f2fs, etc). | |
4592 | */ | |
4593 | static int btrfs_log_all_xattrs(struct btrfs_trans_handle *trans, | |
4594 | struct btrfs_root *root, | |
1a93c36a | 4595 | struct btrfs_inode *inode, |
36283bf7 FM |
4596 | struct btrfs_path *path, |
4597 | struct btrfs_path *dst_path) | |
4598 | { | |
4599 | int ret; | |
4600 | struct btrfs_key key; | |
1a93c36a | 4601 | const u64 ino = btrfs_ino(inode); |
36283bf7 FM |
4602 | int ins_nr = 0; |
4603 | int start_slot = 0; | |
4604 | ||
4605 | key.objectid = ino; | |
4606 | key.type = BTRFS_XATTR_ITEM_KEY; | |
4607 | key.offset = 0; | |
4608 | ||
4609 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
4610 | if (ret < 0) | |
4611 | return ret; | |
4612 | ||
4613 | while (true) { | |
4614 | int slot = path->slots[0]; | |
4615 | struct extent_buffer *leaf = path->nodes[0]; | |
4616 | int nritems = btrfs_header_nritems(leaf); | |
4617 | ||
4618 | if (slot >= nritems) { | |
4619 | if (ins_nr > 0) { | |
4620 | u64 last_extent = 0; | |
4621 | ||
1a93c36a | 4622 | ret = copy_items(trans, inode, dst_path, path, |
36283bf7 FM |
4623 | &last_extent, start_slot, |
4624 | ins_nr, 1, 0); | |
4625 | /* can't be 1, extent items aren't processed */ | |
4626 | ASSERT(ret <= 0); | |
4627 | if (ret < 0) | |
4628 | return ret; | |
4629 | ins_nr = 0; | |
4630 | } | |
4631 | ret = btrfs_next_leaf(root, path); | |
4632 | if (ret < 0) | |
4633 | return ret; | |
4634 | else if (ret > 0) | |
4635 | break; | |
4636 | continue; | |
4637 | } | |
4638 | ||
4639 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
4640 | if (key.objectid != ino || key.type != BTRFS_XATTR_ITEM_KEY) | |
4641 | break; | |
4642 | ||
4643 | if (ins_nr == 0) | |
4644 | start_slot = slot; | |
4645 | ins_nr++; | |
4646 | path->slots[0]++; | |
4647 | cond_resched(); | |
4648 | } | |
4649 | if (ins_nr > 0) { | |
4650 | u64 last_extent = 0; | |
4651 | ||
1a93c36a | 4652 | ret = copy_items(trans, inode, dst_path, path, |
36283bf7 FM |
4653 | &last_extent, start_slot, |
4654 | ins_nr, 1, 0); | |
4655 | /* can't be 1, extent items aren't processed */ | |
4656 | ASSERT(ret <= 0); | |
4657 | if (ret < 0) | |
4658 | return ret; | |
4659 | } | |
4660 | ||
4661 | return 0; | |
4662 | } | |
4663 | ||
a89ca6f2 FM |
4664 | /* |
4665 | * If the no holes feature is enabled we need to make sure any hole between the | |
4666 | * last extent and the i_size of our inode is explicitly marked in the log. This | |
4667 | * is to make sure that doing something like: | |
4668 | * | |
4669 | * 1) create file with 128Kb of data | |
4670 | * 2) truncate file to 64Kb | |
4671 | * 3) truncate file to 256Kb | |
4672 | * 4) fsync file | |
4673 | * 5) <crash/power failure> | |
4674 | * 6) mount fs and trigger log replay | |
4675 | * | |
4676 | * Will give us a file with a size of 256Kb, the first 64Kb of data match what | |
4677 | * the file had in its first 64Kb of data at step 1 and the last 192Kb of the | |
4678 | * file correspond to a hole. The presence of explicit holes in a log tree is | |
4679 | * what guarantees that log replay will remove/adjust file extent items in the | |
4680 | * fs/subvol tree. | |
4681 | * | |
4682 | * Here we do not need to care about holes between extents, that is already done | |
4683 | * by copy_items(). We also only need to do this in the full sync path, where we | |
4684 | * lookup for extents from the fs/subvol tree only. In the fast path case, we | |
4685 | * lookup the list of modified extent maps and if any represents a hole, we | |
4686 | * insert a corresponding extent representing a hole in the log tree. | |
4687 | */ | |
4688 | static int btrfs_log_trailing_hole(struct btrfs_trans_handle *trans, | |
4689 | struct btrfs_root *root, | |
a0308dd7 | 4690 | struct btrfs_inode *inode, |
a89ca6f2 FM |
4691 | struct btrfs_path *path) |
4692 | { | |
0b246afa | 4693 | struct btrfs_fs_info *fs_info = root->fs_info; |
a89ca6f2 FM |
4694 | int ret; |
4695 | struct btrfs_key key; | |
4696 | u64 hole_start; | |
4697 | u64 hole_size; | |
4698 | struct extent_buffer *leaf; | |
4699 | struct btrfs_root *log = root->log_root; | |
a0308dd7 NB |
4700 | const u64 ino = btrfs_ino(inode); |
4701 | const u64 i_size = i_size_read(&inode->vfs_inode); | |
a89ca6f2 | 4702 | |
0b246afa | 4703 | if (!btrfs_fs_incompat(fs_info, NO_HOLES)) |
a89ca6f2 FM |
4704 | return 0; |
4705 | ||
4706 | key.objectid = ino; | |
4707 | key.type = BTRFS_EXTENT_DATA_KEY; | |
4708 | key.offset = (u64)-1; | |
4709 | ||
4710 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
4711 | ASSERT(ret != 0); | |
4712 | if (ret < 0) | |
4713 | return ret; | |
4714 | ||
4715 | ASSERT(path->slots[0] > 0); | |
4716 | path->slots[0]--; | |
4717 | leaf = path->nodes[0]; | |
4718 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
4719 | ||
4720 | if (key.objectid != ino || key.type != BTRFS_EXTENT_DATA_KEY) { | |
4721 | /* inode does not have any extents */ | |
4722 | hole_start = 0; | |
4723 | hole_size = i_size; | |
4724 | } else { | |
4725 | struct btrfs_file_extent_item *extent; | |
4726 | u64 len; | |
4727 | ||
4728 | /* | |
4729 | * If there's an extent beyond i_size, an explicit hole was | |
4730 | * already inserted by copy_items(). | |
4731 | */ | |
4732 | if (key.offset >= i_size) | |
4733 | return 0; | |
4734 | ||
4735 | extent = btrfs_item_ptr(leaf, path->slots[0], | |
4736 | struct btrfs_file_extent_item); | |
4737 | ||
4738 | if (btrfs_file_extent_type(leaf, extent) == | |
0ccc3876 | 4739 | BTRFS_FILE_EXTENT_INLINE) |
a89ca6f2 | 4740 | return 0; |
a89ca6f2 FM |
4741 | |
4742 | len = btrfs_file_extent_num_bytes(leaf, extent); | |
4743 | /* Last extent goes beyond i_size, no need to log a hole. */ | |
4744 | if (key.offset + len > i_size) | |
4745 | return 0; | |
4746 | hole_start = key.offset + len; | |
4747 | hole_size = i_size - hole_start; | |
4748 | } | |
4749 | btrfs_release_path(path); | |
4750 | ||
4751 | /* Last extent ends at i_size. */ | |
4752 | if (hole_size == 0) | |
4753 | return 0; | |
4754 | ||
0b246afa | 4755 | hole_size = ALIGN(hole_size, fs_info->sectorsize); |
a89ca6f2 FM |
4756 | ret = btrfs_insert_file_extent(trans, log, ino, hole_start, 0, 0, |
4757 | hole_size, 0, hole_size, 0, 0, 0); | |
4758 | return ret; | |
4759 | } | |
4760 | ||
56f23fdb FM |
4761 | /* |
4762 | * When we are logging a new inode X, check if it doesn't have a reference that | |
4763 | * matches the reference from some other inode Y created in a past transaction | |
4764 | * and that was renamed in the current transaction. If we don't do this, then at | |
4765 | * log replay time we can lose inode Y (and all its files if it's a directory): | |
4766 | * | |
4767 | * mkdir /mnt/x | |
4768 | * echo "hello world" > /mnt/x/foobar | |
4769 | * sync | |
4770 | * mv /mnt/x /mnt/y | |
4771 | * mkdir /mnt/x # or touch /mnt/x | |
4772 | * xfs_io -c fsync /mnt/x | |
4773 | * <power fail> | |
4774 | * mount fs, trigger log replay | |
4775 | * | |
4776 | * After the log replay procedure, we would lose the first directory and all its | |
4777 | * files (file foobar). | |
4778 | * For the case where inode Y is not a directory we simply end up losing it: | |
4779 | * | |
4780 | * echo "123" > /mnt/foo | |
4781 | * sync | |
4782 | * mv /mnt/foo /mnt/bar | |
4783 | * echo "abc" > /mnt/foo | |
4784 | * xfs_io -c fsync /mnt/foo | |
4785 | * <power fail> | |
4786 | * | |
4787 | * We also need this for cases where a snapshot entry is replaced by some other | |
4788 | * entry (file or directory) otherwise we end up with an unreplayable log due to | |
4789 | * attempts to delete the snapshot entry (entry of type BTRFS_ROOT_ITEM_KEY) as | |
4790 | * if it were a regular entry: | |
4791 | * | |
4792 | * mkdir /mnt/x | |
4793 | * btrfs subvolume snapshot /mnt /mnt/x/snap | |
4794 | * btrfs subvolume delete /mnt/x/snap | |
4795 | * rmdir /mnt/x | |
4796 | * mkdir /mnt/x | |
4797 | * fsync /mnt/x or fsync some new file inside it | |
4798 | * <power fail> | |
4799 | * | |
4800 | * The snapshot delete, rmdir of x, mkdir of a new x and the fsync all happen in | |
4801 | * the same transaction. | |
4802 | */ | |
4803 | static int btrfs_check_ref_name_override(struct extent_buffer *eb, | |
4804 | const int slot, | |
4805 | const struct btrfs_key *key, | |
4791c8f1 | 4806 | struct btrfs_inode *inode, |
a3baaf0d | 4807 | u64 *other_ino, u64 *other_parent) |
56f23fdb FM |
4808 | { |
4809 | int ret; | |
4810 | struct btrfs_path *search_path; | |
4811 | char *name = NULL; | |
4812 | u32 name_len = 0; | |
4813 | u32 item_size = btrfs_item_size_nr(eb, slot); | |
4814 | u32 cur_offset = 0; | |
4815 | unsigned long ptr = btrfs_item_ptr_offset(eb, slot); | |
4816 | ||
4817 | search_path = btrfs_alloc_path(); | |
4818 | if (!search_path) | |
4819 | return -ENOMEM; | |
4820 | search_path->search_commit_root = 1; | |
4821 | search_path->skip_locking = 1; | |
4822 | ||
4823 | while (cur_offset < item_size) { | |
4824 | u64 parent; | |
4825 | u32 this_name_len; | |
4826 | u32 this_len; | |
4827 | unsigned long name_ptr; | |
4828 | struct btrfs_dir_item *di; | |
4829 | ||
4830 | if (key->type == BTRFS_INODE_REF_KEY) { | |
4831 | struct btrfs_inode_ref *iref; | |
4832 | ||
4833 | iref = (struct btrfs_inode_ref *)(ptr + cur_offset); | |
4834 | parent = key->offset; | |
4835 | this_name_len = btrfs_inode_ref_name_len(eb, iref); | |
4836 | name_ptr = (unsigned long)(iref + 1); | |
4837 | this_len = sizeof(*iref) + this_name_len; | |
4838 | } else { | |
4839 | struct btrfs_inode_extref *extref; | |
4840 | ||
4841 | extref = (struct btrfs_inode_extref *)(ptr + | |
4842 | cur_offset); | |
4843 | parent = btrfs_inode_extref_parent(eb, extref); | |
4844 | this_name_len = btrfs_inode_extref_name_len(eb, extref); | |
4845 | name_ptr = (unsigned long)&extref->name; | |
4846 | this_len = sizeof(*extref) + this_name_len; | |
4847 | } | |
4848 | ||
4849 | if (this_name_len > name_len) { | |
4850 | char *new_name; | |
4851 | ||
4852 | new_name = krealloc(name, this_name_len, GFP_NOFS); | |
4853 | if (!new_name) { | |
4854 | ret = -ENOMEM; | |
4855 | goto out; | |
4856 | } | |
4857 | name_len = this_name_len; | |
4858 | name = new_name; | |
4859 | } | |
4860 | ||
4861 | read_extent_buffer(eb, name, name_ptr, this_name_len); | |
4791c8f1 NB |
4862 | di = btrfs_lookup_dir_item(NULL, inode->root, search_path, |
4863 | parent, name, this_name_len, 0); | |
56f23fdb | 4864 | if (di && !IS_ERR(di)) { |
44f714da FM |
4865 | struct btrfs_key di_key; |
4866 | ||
4867 | btrfs_dir_item_key_to_cpu(search_path->nodes[0], | |
4868 | di, &di_key); | |
4869 | if (di_key.type == BTRFS_INODE_ITEM_KEY) { | |
6b5fc433 FM |
4870 | if (di_key.objectid != key->objectid) { |
4871 | ret = 1; | |
4872 | *other_ino = di_key.objectid; | |
a3baaf0d | 4873 | *other_parent = parent; |
6b5fc433 FM |
4874 | } else { |
4875 | ret = 0; | |
4876 | } | |
44f714da FM |
4877 | } else { |
4878 | ret = -EAGAIN; | |
4879 | } | |
56f23fdb FM |
4880 | goto out; |
4881 | } else if (IS_ERR(di)) { | |
4882 | ret = PTR_ERR(di); | |
4883 | goto out; | |
4884 | } | |
4885 | btrfs_release_path(search_path); | |
4886 | ||
4887 | cur_offset += this_len; | |
4888 | } | |
4889 | ret = 0; | |
4890 | out: | |
4891 | btrfs_free_path(search_path); | |
4892 | kfree(name); | |
4893 | return ret; | |
4894 | } | |
4895 | ||
6b5fc433 FM |
4896 | struct btrfs_ino_list { |
4897 | u64 ino; | |
a3baaf0d | 4898 | u64 parent; |
6b5fc433 FM |
4899 | struct list_head list; |
4900 | }; | |
4901 | ||
4902 | static int log_conflicting_inodes(struct btrfs_trans_handle *trans, | |
4903 | struct btrfs_root *root, | |
4904 | struct btrfs_path *path, | |
4905 | struct btrfs_log_ctx *ctx, | |
a3baaf0d | 4906 | u64 ino, u64 parent) |
6b5fc433 FM |
4907 | { |
4908 | struct btrfs_ino_list *ino_elem; | |
4909 | LIST_HEAD(inode_list); | |
4910 | int ret = 0; | |
4911 | ||
4912 | ino_elem = kmalloc(sizeof(*ino_elem), GFP_NOFS); | |
4913 | if (!ino_elem) | |
4914 | return -ENOMEM; | |
4915 | ino_elem->ino = ino; | |
a3baaf0d | 4916 | ino_elem->parent = parent; |
6b5fc433 FM |
4917 | list_add_tail(&ino_elem->list, &inode_list); |
4918 | ||
4919 | while (!list_empty(&inode_list)) { | |
4920 | struct btrfs_fs_info *fs_info = root->fs_info; | |
4921 | struct btrfs_key key; | |
4922 | struct inode *inode; | |
4923 | ||
4924 | ino_elem = list_first_entry(&inode_list, struct btrfs_ino_list, | |
4925 | list); | |
4926 | ino = ino_elem->ino; | |
a3baaf0d | 4927 | parent = ino_elem->parent; |
6b5fc433 FM |
4928 | list_del(&ino_elem->list); |
4929 | kfree(ino_elem); | |
4930 | if (ret) | |
4931 | continue; | |
4932 | ||
4933 | btrfs_release_path(path); | |
4934 | ||
4935 | key.objectid = ino; | |
4936 | key.type = BTRFS_INODE_ITEM_KEY; | |
4937 | key.offset = 0; | |
4938 | inode = btrfs_iget(fs_info->sb, &key, root, NULL); | |
4939 | /* | |
4940 | * If the other inode that had a conflicting dir entry was | |
a3baaf0d FM |
4941 | * deleted in the current transaction, we need to log its parent |
4942 | * directory. | |
6b5fc433 FM |
4943 | */ |
4944 | if (IS_ERR(inode)) { | |
4945 | ret = PTR_ERR(inode); | |
a3baaf0d FM |
4946 | if (ret == -ENOENT) { |
4947 | key.objectid = parent; | |
4948 | inode = btrfs_iget(fs_info->sb, &key, root, | |
4949 | NULL); | |
4950 | if (IS_ERR(inode)) { | |
4951 | ret = PTR_ERR(inode); | |
4952 | } else { | |
4953 | ret = btrfs_log_inode(trans, root, | |
4954 | BTRFS_I(inode), | |
4955 | LOG_OTHER_INODE_ALL, | |
4956 | 0, LLONG_MAX, ctx); | |
4957 | iput(inode); | |
4958 | } | |
4959 | } | |
6b5fc433 FM |
4960 | continue; |
4961 | } | |
4962 | /* | |
4963 | * We are safe logging the other inode without acquiring its | |
4964 | * lock as long as we log with the LOG_INODE_EXISTS mode. We | |
4965 | * are safe against concurrent renames of the other inode as | |
4966 | * well because during a rename we pin the log and update the | |
4967 | * log with the new name before we unpin it. | |
4968 | */ | |
4969 | ret = btrfs_log_inode(trans, root, BTRFS_I(inode), | |
4970 | LOG_OTHER_INODE, 0, LLONG_MAX, ctx); | |
4971 | if (ret) { | |
4972 | iput(inode); | |
4973 | continue; | |
4974 | } | |
4975 | ||
4976 | key.objectid = ino; | |
4977 | key.type = BTRFS_INODE_REF_KEY; | |
4978 | key.offset = 0; | |
4979 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
4980 | if (ret < 0) { | |
4981 | iput(inode); | |
4982 | continue; | |
4983 | } | |
4984 | ||
4985 | while (true) { | |
4986 | struct extent_buffer *leaf = path->nodes[0]; | |
4987 | int slot = path->slots[0]; | |
4988 | u64 other_ino = 0; | |
a3baaf0d | 4989 | u64 other_parent = 0; |
6b5fc433 FM |
4990 | |
4991 | if (slot >= btrfs_header_nritems(leaf)) { | |
4992 | ret = btrfs_next_leaf(root, path); | |
4993 | if (ret < 0) { | |
4994 | break; | |
4995 | } else if (ret > 0) { | |
4996 | ret = 0; | |
4997 | break; | |
4998 | } | |
4999 | continue; | |
5000 | } | |
5001 | ||
5002 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
5003 | if (key.objectid != ino || | |
5004 | (key.type != BTRFS_INODE_REF_KEY && | |
5005 | key.type != BTRFS_INODE_EXTREF_KEY)) { | |
5006 | ret = 0; | |
5007 | break; | |
5008 | } | |
5009 | ||
5010 | ret = btrfs_check_ref_name_override(leaf, slot, &key, | |
a3baaf0d FM |
5011 | BTRFS_I(inode), &other_ino, |
5012 | &other_parent); | |
6b5fc433 FM |
5013 | if (ret < 0) |
5014 | break; | |
5015 | if (ret > 0) { | |
5016 | ino_elem = kmalloc(sizeof(*ino_elem), GFP_NOFS); | |
5017 | if (!ino_elem) { | |
5018 | ret = -ENOMEM; | |
5019 | break; | |
5020 | } | |
5021 | ino_elem->ino = other_ino; | |
a3baaf0d | 5022 | ino_elem->parent = other_parent; |
6b5fc433 FM |
5023 | list_add_tail(&ino_elem->list, &inode_list); |
5024 | ret = 0; | |
5025 | } | |
5026 | path->slots[0]++; | |
5027 | } | |
5028 | iput(inode); | |
5029 | } | |
5030 | ||
5031 | return ret; | |
5032 | } | |
5033 | ||
e02119d5 CM |
5034 | /* log a single inode in the tree log. |
5035 | * At least one parent directory for this inode must exist in the tree | |
5036 | * or be logged already. | |
5037 | * | |
5038 | * Any items from this inode changed by the current transaction are copied | |
5039 | * to the log tree. An extra reference is taken on any extents in this | |
5040 | * file, allowing us to avoid a whole pile of corner cases around logging | |
5041 | * blocks that have been removed from the tree. | |
5042 | * | |
5043 | * See LOG_INODE_ALL and related defines for a description of what inode_only | |
5044 | * does. | |
5045 | * | |
5046 | * This handles both files and directories. | |
5047 | */ | |
12fcfd22 | 5048 | static int btrfs_log_inode(struct btrfs_trans_handle *trans, |
a59108a7 | 5049 | struct btrfs_root *root, struct btrfs_inode *inode, |
49dae1bc FM |
5050 | int inode_only, |
5051 | const loff_t start, | |
8407f553 FM |
5052 | const loff_t end, |
5053 | struct btrfs_log_ctx *ctx) | |
e02119d5 | 5054 | { |
0b246afa | 5055 | struct btrfs_fs_info *fs_info = root->fs_info; |
e02119d5 CM |
5056 | struct btrfs_path *path; |
5057 | struct btrfs_path *dst_path; | |
5058 | struct btrfs_key min_key; | |
5059 | struct btrfs_key max_key; | |
5060 | struct btrfs_root *log = root->log_root; | |
16e7549f | 5061 | u64 last_extent = 0; |
4a500fd1 | 5062 | int err = 0; |
e02119d5 | 5063 | int ret; |
3a5f1d45 | 5064 | int nritems; |
31ff1cd2 CM |
5065 | int ins_start_slot = 0; |
5066 | int ins_nr; | |
5dc562c5 | 5067 | bool fast_search = false; |
a59108a7 NB |
5068 | u64 ino = btrfs_ino(inode); |
5069 | struct extent_map_tree *em_tree = &inode->extent_tree; | |
1a4bcf47 | 5070 | u64 logged_isize = 0; |
e4545de5 | 5071 | bool need_log_inode_item = true; |
9a8fca62 | 5072 | bool xattrs_logged = false; |
a3baaf0d | 5073 | bool recursive_logging = false; |
e02119d5 | 5074 | |
e02119d5 | 5075 | path = btrfs_alloc_path(); |
5df67083 TI |
5076 | if (!path) |
5077 | return -ENOMEM; | |
e02119d5 | 5078 | dst_path = btrfs_alloc_path(); |
5df67083 TI |
5079 | if (!dst_path) { |
5080 | btrfs_free_path(path); | |
5081 | return -ENOMEM; | |
5082 | } | |
e02119d5 | 5083 | |
33345d01 | 5084 | min_key.objectid = ino; |
e02119d5 CM |
5085 | min_key.type = BTRFS_INODE_ITEM_KEY; |
5086 | min_key.offset = 0; | |
5087 | ||
33345d01 | 5088 | max_key.objectid = ino; |
12fcfd22 | 5089 | |
12fcfd22 | 5090 | |
5dc562c5 | 5091 | /* today the code can only do partial logging of directories */ |
a59108a7 | 5092 | if (S_ISDIR(inode->vfs_inode.i_mode) || |
5269b67e | 5093 | (!test_bit(BTRFS_INODE_NEEDS_FULL_SYNC, |
a59108a7 | 5094 | &inode->runtime_flags) && |
781feef7 | 5095 | inode_only >= LOG_INODE_EXISTS)) |
e02119d5 CM |
5096 | max_key.type = BTRFS_XATTR_ITEM_KEY; |
5097 | else | |
5098 | max_key.type = (u8)-1; | |
5099 | max_key.offset = (u64)-1; | |
5100 | ||
2c2c452b FM |
5101 | /* |
5102 | * Only run delayed items if we are a dir or a new file. | |
5103 | * Otherwise commit the delayed inode only, which is needed in | |
5104 | * order for the log replay code to mark inodes for link count | |
5105 | * fixup (create temporary BTRFS_TREE_LOG_FIXUP_OBJECTID items). | |
5106 | */ | |
a59108a7 NB |
5107 | if (S_ISDIR(inode->vfs_inode.i_mode) || |
5108 | inode->generation > fs_info->last_trans_committed) | |
5109 | ret = btrfs_commit_inode_delayed_items(trans, inode); | |
2c2c452b | 5110 | else |
a59108a7 | 5111 | ret = btrfs_commit_inode_delayed_inode(inode); |
2c2c452b FM |
5112 | |
5113 | if (ret) { | |
5114 | btrfs_free_path(path); | |
5115 | btrfs_free_path(dst_path); | |
5116 | return ret; | |
16cdcec7 MX |
5117 | } |
5118 | ||
a3baaf0d FM |
5119 | if (inode_only == LOG_OTHER_INODE || inode_only == LOG_OTHER_INODE_ALL) { |
5120 | recursive_logging = true; | |
5121 | if (inode_only == LOG_OTHER_INODE) | |
5122 | inode_only = LOG_INODE_EXISTS; | |
5123 | else | |
5124 | inode_only = LOG_INODE_ALL; | |
a59108a7 | 5125 | mutex_lock_nested(&inode->log_mutex, SINGLE_DEPTH_NESTING); |
781feef7 | 5126 | } else { |
a59108a7 | 5127 | mutex_lock(&inode->log_mutex); |
781feef7 | 5128 | } |
e02119d5 CM |
5129 | |
5130 | /* | |
5131 | * a brute force approach to making sure we get the most uptodate | |
5132 | * copies of everything. | |
5133 | */ | |
a59108a7 | 5134 | if (S_ISDIR(inode->vfs_inode.i_mode)) { |
e02119d5 CM |
5135 | int max_key_type = BTRFS_DIR_LOG_INDEX_KEY; |
5136 | ||
4f764e51 FM |
5137 | if (inode_only == LOG_INODE_EXISTS) |
5138 | max_key_type = BTRFS_XATTR_ITEM_KEY; | |
33345d01 | 5139 | ret = drop_objectid_items(trans, log, path, ino, max_key_type); |
e02119d5 | 5140 | } else { |
1a4bcf47 FM |
5141 | if (inode_only == LOG_INODE_EXISTS) { |
5142 | /* | |
5143 | * Make sure the new inode item we write to the log has | |
5144 | * the same isize as the current one (if it exists). | |
5145 | * This is necessary to prevent data loss after log | |
5146 | * replay, and also to prevent doing a wrong expanding | |
5147 | * truncate - for e.g. create file, write 4K into offset | |
5148 | * 0, fsync, write 4K into offset 4096, add hard link, | |
5149 | * fsync some other file (to sync log), power fail - if | |
5150 | * we use the inode's current i_size, after log replay | |
5151 | * we get a 8Kb file, with the last 4Kb extent as a hole | |
5152 | * (zeroes), as if an expanding truncate happened, | |
5153 | * instead of getting a file of 4Kb only. | |
5154 | */ | |
a59108a7 | 5155 | err = logged_inode_size(log, inode, path, &logged_isize); |
1a4bcf47 FM |
5156 | if (err) |
5157 | goto out_unlock; | |
5158 | } | |
a742994a | 5159 | if (test_bit(BTRFS_INODE_NEEDS_FULL_SYNC, |
a59108a7 | 5160 | &inode->runtime_flags)) { |
a742994a | 5161 | if (inode_only == LOG_INODE_EXISTS) { |
4f764e51 | 5162 | max_key.type = BTRFS_XATTR_ITEM_KEY; |
a742994a FM |
5163 | ret = drop_objectid_items(trans, log, path, ino, |
5164 | max_key.type); | |
5165 | } else { | |
5166 | clear_bit(BTRFS_INODE_NEEDS_FULL_SYNC, | |
a59108a7 | 5167 | &inode->runtime_flags); |
a742994a | 5168 | clear_bit(BTRFS_INODE_COPY_EVERYTHING, |
a59108a7 | 5169 | &inode->runtime_flags); |
28ed1345 CM |
5170 | while(1) { |
5171 | ret = btrfs_truncate_inode_items(trans, | |
a59108a7 | 5172 | log, &inode->vfs_inode, 0, 0); |
28ed1345 CM |
5173 | if (ret != -EAGAIN) |
5174 | break; | |
5175 | } | |
a742994a | 5176 | } |
4f764e51 | 5177 | } else if (test_and_clear_bit(BTRFS_INODE_COPY_EVERYTHING, |
a59108a7 | 5178 | &inode->runtime_flags) || |
6cfab851 | 5179 | inode_only == LOG_INODE_EXISTS) { |
4f764e51 | 5180 | if (inode_only == LOG_INODE_ALL) |
183f37fa | 5181 | fast_search = true; |
4f764e51 | 5182 | max_key.type = BTRFS_XATTR_ITEM_KEY; |
5dc562c5 | 5183 | ret = drop_objectid_items(trans, log, path, ino, |
e9976151 | 5184 | max_key.type); |
a95249b3 JB |
5185 | } else { |
5186 | if (inode_only == LOG_INODE_ALL) | |
5187 | fast_search = true; | |
a95249b3 | 5188 | goto log_extents; |
5dc562c5 | 5189 | } |
a95249b3 | 5190 | |
e02119d5 | 5191 | } |
4a500fd1 YZ |
5192 | if (ret) { |
5193 | err = ret; | |
5194 | goto out_unlock; | |
5195 | } | |
e02119d5 | 5196 | |
d397712b | 5197 | while (1) { |
31ff1cd2 | 5198 | ins_nr = 0; |
6174d3cb | 5199 | ret = btrfs_search_forward(root, &min_key, |
de78b51a | 5200 | path, trans->transid); |
fb770ae4 LB |
5201 | if (ret < 0) { |
5202 | err = ret; | |
5203 | goto out_unlock; | |
5204 | } | |
e02119d5 CM |
5205 | if (ret != 0) |
5206 | break; | |
3a5f1d45 | 5207 | again: |
31ff1cd2 | 5208 | /* note, ins_nr might be > 0 here, cleanup outside the loop */ |
33345d01 | 5209 | if (min_key.objectid != ino) |
e02119d5 CM |
5210 | break; |
5211 | if (min_key.type > max_key.type) | |
5212 | break; | |
31ff1cd2 | 5213 | |
e4545de5 FM |
5214 | if (min_key.type == BTRFS_INODE_ITEM_KEY) |
5215 | need_log_inode_item = false; | |
5216 | ||
56f23fdb FM |
5217 | if ((min_key.type == BTRFS_INODE_REF_KEY || |
5218 | min_key.type == BTRFS_INODE_EXTREF_KEY) && | |
6b5fc433 FM |
5219 | inode->generation == trans->transid && |
5220 | !recursive_logging) { | |
44f714da | 5221 | u64 other_ino = 0; |
a3baaf0d | 5222 | u64 other_parent = 0; |
44f714da | 5223 | |
56f23fdb | 5224 | ret = btrfs_check_ref_name_override(path->nodes[0], |
a59108a7 | 5225 | path->slots[0], &min_key, inode, |
a3baaf0d | 5226 | &other_ino, &other_parent); |
56f23fdb FM |
5227 | if (ret < 0) { |
5228 | err = ret; | |
5229 | goto out_unlock; | |
28a23593 | 5230 | } else if (ret > 0 && ctx && |
4a0cc7ca | 5231 | other_ino != btrfs_ino(BTRFS_I(ctx->inode))) { |
44f714da FM |
5232 | if (ins_nr > 0) { |
5233 | ins_nr++; | |
5234 | } else { | |
5235 | ins_nr = 1; | |
5236 | ins_start_slot = path->slots[0]; | |
5237 | } | |
a59108a7 | 5238 | ret = copy_items(trans, inode, dst_path, path, |
44f714da FM |
5239 | &last_extent, ins_start_slot, |
5240 | ins_nr, inode_only, | |
5241 | logged_isize); | |
5242 | if (ret < 0) { | |
5243 | err = ret; | |
5244 | goto out_unlock; | |
5245 | } | |
5246 | ins_nr = 0; | |
6b5fc433 FM |
5247 | |
5248 | err = log_conflicting_inodes(trans, root, path, | |
a3baaf0d | 5249 | ctx, other_ino, other_parent); |
44f714da FM |
5250 | if (err) |
5251 | goto out_unlock; | |
6b5fc433 FM |
5252 | btrfs_release_path(path); |
5253 | goto next_key; | |
56f23fdb FM |
5254 | } |
5255 | } | |
5256 | ||
36283bf7 FM |
5257 | /* Skip xattrs, we log them later with btrfs_log_all_xattrs() */ |
5258 | if (min_key.type == BTRFS_XATTR_ITEM_KEY) { | |
5259 | if (ins_nr == 0) | |
5260 | goto next_slot; | |
a59108a7 | 5261 | ret = copy_items(trans, inode, dst_path, path, |
36283bf7 FM |
5262 | &last_extent, ins_start_slot, |
5263 | ins_nr, inode_only, logged_isize); | |
5264 | if (ret < 0) { | |
5265 | err = ret; | |
5266 | goto out_unlock; | |
5267 | } | |
5268 | ins_nr = 0; | |
5269 | if (ret) { | |
5270 | btrfs_release_path(path); | |
5271 | continue; | |
5272 | } | |
5273 | goto next_slot; | |
5274 | } | |
5275 | ||
31ff1cd2 CM |
5276 | if (ins_nr && ins_start_slot + ins_nr == path->slots[0]) { |
5277 | ins_nr++; | |
5278 | goto next_slot; | |
5279 | } else if (!ins_nr) { | |
5280 | ins_start_slot = path->slots[0]; | |
5281 | ins_nr = 1; | |
5282 | goto next_slot; | |
e02119d5 CM |
5283 | } |
5284 | ||
a59108a7 | 5285 | ret = copy_items(trans, inode, dst_path, path, &last_extent, |
1a4bcf47 FM |
5286 | ins_start_slot, ins_nr, inode_only, |
5287 | logged_isize); | |
16e7549f | 5288 | if (ret < 0) { |
4a500fd1 YZ |
5289 | err = ret; |
5290 | goto out_unlock; | |
a71db86e RV |
5291 | } |
5292 | if (ret) { | |
16e7549f JB |
5293 | ins_nr = 0; |
5294 | btrfs_release_path(path); | |
5295 | continue; | |
4a500fd1 | 5296 | } |
31ff1cd2 CM |
5297 | ins_nr = 1; |
5298 | ins_start_slot = path->slots[0]; | |
5299 | next_slot: | |
e02119d5 | 5300 | |
3a5f1d45 CM |
5301 | nritems = btrfs_header_nritems(path->nodes[0]); |
5302 | path->slots[0]++; | |
5303 | if (path->slots[0] < nritems) { | |
5304 | btrfs_item_key_to_cpu(path->nodes[0], &min_key, | |
5305 | path->slots[0]); | |
5306 | goto again; | |
5307 | } | |
31ff1cd2 | 5308 | if (ins_nr) { |
a59108a7 | 5309 | ret = copy_items(trans, inode, dst_path, path, |
16e7549f | 5310 | &last_extent, ins_start_slot, |
1a4bcf47 | 5311 | ins_nr, inode_only, logged_isize); |
16e7549f | 5312 | if (ret < 0) { |
4a500fd1 YZ |
5313 | err = ret; |
5314 | goto out_unlock; | |
5315 | } | |
16e7549f | 5316 | ret = 0; |
31ff1cd2 CM |
5317 | ins_nr = 0; |
5318 | } | |
b3b4aa74 | 5319 | btrfs_release_path(path); |
44f714da | 5320 | next_key: |
3d41d702 | 5321 | if (min_key.offset < (u64)-1) { |
e02119d5 | 5322 | min_key.offset++; |
3d41d702 | 5323 | } else if (min_key.type < max_key.type) { |
e02119d5 | 5324 | min_key.type++; |
3d41d702 FDBM |
5325 | min_key.offset = 0; |
5326 | } else { | |
e02119d5 | 5327 | break; |
3d41d702 | 5328 | } |
e02119d5 | 5329 | } |
31ff1cd2 | 5330 | if (ins_nr) { |
a59108a7 | 5331 | ret = copy_items(trans, inode, dst_path, path, &last_extent, |
1a4bcf47 FM |
5332 | ins_start_slot, ins_nr, inode_only, |
5333 | logged_isize); | |
16e7549f | 5334 | if (ret < 0) { |
4a500fd1 YZ |
5335 | err = ret; |
5336 | goto out_unlock; | |
5337 | } | |
16e7549f | 5338 | ret = 0; |
31ff1cd2 CM |
5339 | ins_nr = 0; |
5340 | } | |
5dc562c5 | 5341 | |
36283bf7 FM |
5342 | btrfs_release_path(path); |
5343 | btrfs_release_path(dst_path); | |
a59108a7 | 5344 | err = btrfs_log_all_xattrs(trans, root, inode, path, dst_path); |
36283bf7 FM |
5345 | if (err) |
5346 | goto out_unlock; | |
9a8fca62 | 5347 | xattrs_logged = true; |
a89ca6f2 FM |
5348 | if (max_key.type >= BTRFS_EXTENT_DATA_KEY && !fast_search) { |
5349 | btrfs_release_path(path); | |
5350 | btrfs_release_path(dst_path); | |
a59108a7 | 5351 | err = btrfs_log_trailing_hole(trans, root, inode, path); |
a89ca6f2 FM |
5352 | if (err) |
5353 | goto out_unlock; | |
5354 | } | |
a95249b3 | 5355 | log_extents: |
f3b15ccd JB |
5356 | btrfs_release_path(path); |
5357 | btrfs_release_path(dst_path); | |
e4545de5 | 5358 | if (need_log_inode_item) { |
a59108a7 | 5359 | err = log_inode_item(trans, log, dst_path, inode); |
9a8fca62 FM |
5360 | if (!err && !xattrs_logged) { |
5361 | err = btrfs_log_all_xattrs(trans, root, inode, path, | |
5362 | dst_path); | |
5363 | btrfs_release_path(path); | |
5364 | } | |
e4545de5 FM |
5365 | if (err) |
5366 | goto out_unlock; | |
5367 | } | |
5dc562c5 | 5368 | if (fast_search) { |
a59108a7 | 5369 | ret = btrfs_log_changed_extents(trans, root, inode, dst_path, |
a2120a47 | 5370 | ctx, start, end); |
5dc562c5 JB |
5371 | if (ret) { |
5372 | err = ret; | |
5373 | goto out_unlock; | |
5374 | } | |
d006a048 | 5375 | } else if (inode_only == LOG_INODE_ALL) { |
06d3d22b LB |
5376 | struct extent_map *em, *n; |
5377 | ||
49dae1bc FM |
5378 | write_lock(&em_tree->lock); |
5379 | /* | |
5380 | * We can't just remove every em if we're called for a ranged | |
5381 | * fsync - that is, one that doesn't cover the whole possible | |
5382 | * file range (0 to LLONG_MAX). This is because we can have | |
5383 | * em's that fall outside the range we're logging and therefore | |
5384 | * their ordered operations haven't completed yet | |
5385 | * (btrfs_finish_ordered_io() not invoked yet). This means we | |
5386 | * didn't get their respective file extent item in the fs/subvol | |
5387 | * tree yet, and need to let the next fast fsync (one which | |
5388 | * consults the list of modified extent maps) find the em so | |
5389 | * that it logs a matching file extent item and waits for the | |
5390 | * respective ordered operation to complete (if it's still | |
5391 | * running). | |
5392 | * | |
5393 | * Removing every em outside the range we're logging would make | |
5394 | * the next fast fsync not log their matching file extent items, | |
5395 | * therefore making us lose data after a log replay. | |
5396 | */ | |
5397 | list_for_each_entry_safe(em, n, &em_tree->modified_extents, | |
5398 | list) { | |
5399 | const u64 mod_end = em->mod_start + em->mod_len - 1; | |
5400 | ||
5401 | if (em->mod_start >= start && mod_end <= end) | |
5402 | list_del_init(&em->list); | |
5403 | } | |
5404 | write_unlock(&em_tree->lock); | |
5dc562c5 JB |
5405 | } |
5406 | ||
a59108a7 NB |
5407 | if (inode_only == LOG_INODE_ALL && S_ISDIR(inode->vfs_inode.i_mode)) { |
5408 | ret = log_directory_changes(trans, root, inode, path, dst_path, | |
5409 | ctx); | |
4a500fd1 YZ |
5410 | if (ret) { |
5411 | err = ret; | |
5412 | goto out_unlock; | |
5413 | } | |
e02119d5 | 5414 | } |
49dae1bc | 5415 | |
a59108a7 NB |
5416 | spin_lock(&inode->lock); |
5417 | inode->logged_trans = trans->transid; | |
5418 | inode->last_log_commit = inode->last_sub_trans; | |
5419 | spin_unlock(&inode->lock); | |
4a500fd1 | 5420 | out_unlock: |
a59108a7 | 5421 | mutex_unlock(&inode->log_mutex); |
e02119d5 CM |
5422 | |
5423 | btrfs_free_path(path); | |
5424 | btrfs_free_path(dst_path); | |
4a500fd1 | 5425 | return err; |
e02119d5 CM |
5426 | } |
5427 | ||
2be63d5c FM |
5428 | /* |
5429 | * Check if we must fallback to a transaction commit when logging an inode. | |
5430 | * This must be called after logging the inode and is used only in the context | |
5431 | * when fsyncing an inode requires the need to log some other inode - in which | |
5432 | * case we can't lock the i_mutex of each other inode we need to log as that | |
5433 | * can lead to deadlocks with concurrent fsync against other inodes (as we can | |
5434 | * log inodes up or down in the hierarchy) or rename operations for example. So | |
5435 | * we take the log_mutex of the inode after we have logged it and then check for | |
5436 | * its last_unlink_trans value - this is safe because any task setting | |
5437 | * last_unlink_trans must take the log_mutex and it must do this before it does | |
5438 | * the actual unlink operation, so if we do this check before a concurrent task | |
5439 | * sets last_unlink_trans it means we've logged a consistent version/state of | |
5440 | * all the inode items, otherwise we are not sure and must do a transaction | |
01327610 | 5441 | * commit (the concurrent task might have only updated last_unlink_trans before |
2be63d5c FM |
5442 | * we logged the inode or it might have also done the unlink). |
5443 | */ | |
5444 | static bool btrfs_must_commit_transaction(struct btrfs_trans_handle *trans, | |
ab1717b2 | 5445 | struct btrfs_inode *inode) |
2be63d5c | 5446 | { |
ab1717b2 | 5447 | struct btrfs_fs_info *fs_info = inode->root->fs_info; |
2be63d5c FM |
5448 | bool ret = false; |
5449 | ||
ab1717b2 NB |
5450 | mutex_lock(&inode->log_mutex); |
5451 | if (inode->last_unlink_trans > fs_info->last_trans_committed) { | |
2be63d5c FM |
5452 | /* |
5453 | * Make sure any commits to the log are forced to be full | |
5454 | * commits. | |
5455 | */ | |
5456 | btrfs_set_log_full_commit(fs_info, trans); | |
5457 | ret = true; | |
5458 | } | |
ab1717b2 | 5459 | mutex_unlock(&inode->log_mutex); |
2be63d5c FM |
5460 | |
5461 | return ret; | |
5462 | } | |
5463 | ||
12fcfd22 CM |
5464 | /* |
5465 | * follow the dentry parent pointers up the chain and see if any | |
5466 | * of the directories in it require a full commit before they can | |
5467 | * be logged. Returns zero if nothing special needs to be done or 1 if | |
5468 | * a full commit is required. | |
5469 | */ | |
5470 | static noinline int check_parent_dirs_for_sync(struct btrfs_trans_handle *trans, | |
aefa6115 | 5471 | struct btrfs_inode *inode, |
12fcfd22 CM |
5472 | struct dentry *parent, |
5473 | struct super_block *sb, | |
5474 | u64 last_committed) | |
e02119d5 | 5475 | { |
12fcfd22 | 5476 | int ret = 0; |
6a912213 | 5477 | struct dentry *old_parent = NULL; |
aefa6115 | 5478 | struct btrfs_inode *orig_inode = inode; |
e02119d5 | 5479 | |
af4176b4 CM |
5480 | /* |
5481 | * for regular files, if its inode is already on disk, we don't | |
5482 | * have to worry about the parents at all. This is because | |
5483 | * we can use the last_unlink_trans field to record renames | |
5484 | * and other fun in this file. | |
5485 | */ | |
aefa6115 NB |
5486 | if (S_ISREG(inode->vfs_inode.i_mode) && |
5487 | inode->generation <= last_committed && | |
5488 | inode->last_unlink_trans <= last_committed) | |
5489 | goto out; | |
af4176b4 | 5490 | |
aefa6115 | 5491 | if (!S_ISDIR(inode->vfs_inode.i_mode)) { |
fc64005c | 5492 | if (!parent || d_really_is_negative(parent) || sb != parent->d_sb) |
12fcfd22 | 5493 | goto out; |
aefa6115 | 5494 | inode = BTRFS_I(d_inode(parent)); |
12fcfd22 CM |
5495 | } |
5496 | ||
5497 | while (1) { | |
de2b530b JB |
5498 | /* |
5499 | * If we are logging a directory then we start with our inode, | |
01327610 | 5500 | * not our parent's inode, so we need to skip setting the |
de2b530b JB |
5501 | * logged_trans so that further down in the log code we don't |
5502 | * think this inode has already been logged. | |
5503 | */ | |
5504 | if (inode != orig_inode) | |
aefa6115 | 5505 | inode->logged_trans = trans->transid; |
12fcfd22 CM |
5506 | smp_mb(); |
5507 | ||
aefa6115 | 5508 | if (btrfs_must_commit_transaction(trans, inode)) { |
12fcfd22 CM |
5509 | ret = 1; |
5510 | break; | |
5511 | } | |
5512 | ||
fc64005c | 5513 | if (!parent || d_really_is_negative(parent) || sb != parent->d_sb) |
12fcfd22 CM |
5514 | break; |
5515 | ||
44f714da | 5516 | if (IS_ROOT(parent)) { |
aefa6115 NB |
5517 | inode = BTRFS_I(d_inode(parent)); |
5518 | if (btrfs_must_commit_transaction(trans, inode)) | |
44f714da | 5519 | ret = 1; |
12fcfd22 | 5520 | break; |
44f714da | 5521 | } |
12fcfd22 | 5522 | |
6a912213 JB |
5523 | parent = dget_parent(parent); |
5524 | dput(old_parent); | |
5525 | old_parent = parent; | |
aefa6115 | 5526 | inode = BTRFS_I(d_inode(parent)); |
12fcfd22 CM |
5527 | |
5528 | } | |
6a912213 | 5529 | dput(old_parent); |
12fcfd22 | 5530 | out: |
e02119d5 CM |
5531 | return ret; |
5532 | } | |
5533 | ||
2f2ff0ee FM |
5534 | struct btrfs_dir_list { |
5535 | u64 ino; | |
5536 | struct list_head list; | |
5537 | }; | |
5538 | ||
5539 | /* | |
5540 | * Log the inodes of the new dentries of a directory. See log_dir_items() for | |
5541 | * details about the why it is needed. | |
5542 | * This is a recursive operation - if an existing dentry corresponds to a | |
5543 | * directory, that directory's new entries are logged too (same behaviour as | |
5544 | * ext3/4, xfs, f2fs, reiserfs, nilfs2). Note that when logging the inodes | |
5545 | * the dentries point to we do not lock their i_mutex, otherwise lockdep | |
5546 | * complains about the following circular lock dependency / possible deadlock: | |
5547 | * | |
5548 | * CPU0 CPU1 | |
5549 | * ---- ---- | |
5550 | * lock(&type->i_mutex_dir_key#3/2); | |
5551 | * lock(sb_internal#2); | |
5552 | * lock(&type->i_mutex_dir_key#3/2); | |
5553 | * lock(&sb->s_type->i_mutex_key#14); | |
5554 | * | |
5555 | * Where sb_internal is the lock (a counter that works as a lock) acquired by | |
5556 | * sb_start_intwrite() in btrfs_start_transaction(). | |
5557 | * Not locking i_mutex of the inodes is still safe because: | |
5558 | * | |
5559 | * 1) For regular files we log with a mode of LOG_INODE_EXISTS. It's possible | |
5560 | * that while logging the inode new references (names) are added or removed | |
5561 | * from the inode, leaving the logged inode item with a link count that does | |
5562 | * not match the number of logged inode reference items. This is fine because | |
5563 | * at log replay time we compute the real number of links and correct the | |
5564 | * link count in the inode item (see replay_one_buffer() and | |
5565 | * link_to_fixup_dir()); | |
5566 | * | |
5567 | * 2) For directories we log with a mode of LOG_INODE_ALL. It's possible that | |
5568 | * while logging the inode's items new items with keys BTRFS_DIR_ITEM_KEY and | |
5569 | * BTRFS_DIR_INDEX_KEY are added to fs/subvol tree and the logged inode item | |
5570 | * has a size that doesn't match the sum of the lengths of all the logged | |
5571 | * names. This does not result in a problem because if a dir_item key is | |
5572 | * logged but its matching dir_index key is not logged, at log replay time we | |
5573 | * don't use it to replay the respective name (see replay_one_name()). On the | |
5574 | * other hand if only the dir_index key ends up being logged, the respective | |
5575 | * name is added to the fs/subvol tree with both the dir_item and dir_index | |
5576 | * keys created (see replay_one_name()). | |
5577 | * The directory's inode item with a wrong i_size is not a problem as well, | |
5578 | * since we don't use it at log replay time to set the i_size in the inode | |
5579 | * item of the fs/subvol tree (see overwrite_item()). | |
5580 | */ | |
5581 | static int log_new_dir_dentries(struct btrfs_trans_handle *trans, | |
5582 | struct btrfs_root *root, | |
51cc0d32 | 5583 | struct btrfs_inode *start_inode, |
2f2ff0ee FM |
5584 | struct btrfs_log_ctx *ctx) |
5585 | { | |
0b246afa | 5586 | struct btrfs_fs_info *fs_info = root->fs_info; |
2f2ff0ee FM |
5587 | struct btrfs_root *log = root->log_root; |
5588 | struct btrfs_path *path; | |
5589 | LIST_HEAD(dir_list); | |
5590 | struct btrfs_dir_list *dir_elem; | |
5591 | int ret = 0; | |
5592 | ||
5593 | path = btrfs_alloc_path(); | |
5594 | if (!path) | |
5595 | return -ENOMEM; | |
5596 | ||
5597 | dir_elem = kmalloc(sizeof(*dir_elem), GFP_NOFS); | |
5598 | if (!dir_elem) { | |
5599 | btrfs_free_path(path); | |
5600 | return -ENOMEM; | |
5601 | } | |
51cc0d32 | 5602 | dir_elem->ino = btrfs_ino(start_inode); |
2f2ff0ee FM |
5603 | list_add_tail(&dir_elem->list, &dir_list); |
5604 | ||
5605 | while (!list_empty(&dir_list)) { | |
5606 | struct extent_buffer *leaf; | |
5607 | struct btrfs_key min_key; | |
5608 | int nritems; | |
5609 | int i; | |
5610 | ||
5611 | dir_elem = list_first_entry(&dir_list, struct btrfs_dir_list, | |
5612 | list); | |
5613 | if (ret) | |
5614 | goto next_dir_inode; | |
5615 | ||
5616 | min_key.objectid = dir_elem->ino; | |
5617 | min_key.type = BTRFS_DIR_ITEM_KEY; | |
5618 | min_key.offset = 0; | |
5619 | again: | |
5620 | btrfs_release_path(path); | |
5621 | ret = btrfs_search_forward(log, &min_key, path, trans->transid); | |
5622 | if (ret < 0) { | |
5623 | goto next_dir_inode; | |
5624 | } else if (ret > 0) { | |
5625 | ret = 0; | |
5626 | goto next_dir_inode; | |
5627 | } | |
5628 | ||
5629 | process_leaf: | |
5630 | leaf = path->nodes[0]; | |
5631 | nritems = btrfs_header_nritems(leaf); | |
5632 | for (i = path->slots[0]; i < nritems; i++) { | |
5633 | struct btrfs_dir_item *di; | |
5634 | struct btrfs_key di_key; | |
5635 | struct inode *di_inode; | |
5636 | struct btrfs_dir_list *new_dir_elem; | |
5637 | int log_mode = LOG_INODE_EXISTS; | |
5638 | int type; | |
5639 | ||
5640 | btrfs_item_key_to_cpu(leaf, &min_key, i); | |
5641 | if (min_key.objectid != dir_elem->ino || | |
5642 | min_key.type != BTRFS_DIR_ITEM_KEY) | |
5643 | goto next_dir_inode; | |
5644 | ||
5645 | di = btrfs_item_ptr(leaf, i, struct btrfs_dir_item); | |
5646 | type = btrfs_dir_type(leaf, di); | |
5647 | if (btrfs_dir_transid(leaf, di) < trans->transid && | |
5648 | type != BTRFS_FT_DIR) | |
5649 | continue; | |
5650 | btrfs_dir_item_key_to_cpu(leaf, di, &di_key); | |
5651 | if (di_key.type == BTRFS_ROOT_ITEM_KEY) | |
5652 | continue; | |
5653 | ||
ec125cfb | 5654 | btrfs_release_path(path); |
0b246afa | 5655 | di_inode = btrfs_iget(fs_info->sb, &di_key, root, NULL); |
2f2ff0ee FM |
5656 | if (IS_ERR(di_inode)) { |
5657 | ret = PTR_ERR(di_inode); | |
5658 | goto next_dir_inode; | |
5659 | } | |
5660 | ||
0f8939b8 | 5661 | if (btrfs_inode_in_log(BTRFS_I(di_inode), trans->transid)) { |
2f2ff0ee | 5662 | iput(di_inode); |
ec125cfb | 5663 | break; |
2f2ff0ee FM |
5664 | } |
5665 | ||
5666 | ctx->log_new_dentries = false; | |
3f9749f6 | 5667 | if (type == BTRFS_FT_DIR || type == BTRFS_FT_SYMLINK) |
2f2ff0ee | 5668 | log_mode = LOG_INODE_ALL; |
a59108a7 | 5669 | ret = btrfs_log_inode(trans, root, BTRFS_I(di_inode), |
2f2ff0ee | 5670 | log_mode, 0, LLONG_MAX, ctx); |
2be63d5c | 5671 | if (!ret && |
ab1717b2 | 5672 | btrfs_must_commit_transaction(trans, BTRFS_I(di_inode))) |
2be63d5c | 5673 | ret = 1; |
2f2ff0ee FM |
5674 | iput(di_inode); |
5675 | if (ret) | |
5676 | goto next_dir_inode; | |
5677 | if (ctx->log_new_dentries) { | |
5678 | new_dir_elem = kmalloc(sizeof(*new_dir_elem), | |
5679 | GFP_NOFS); | |
5680 | if (!new_dir_elem) { | |
5681 | ret = -ENOMEM; | |
5682 | goto next_dir_inode; | |
5683 | } | |
5684 | new_dir_elem->ino = di_key.objectid; | |
5685 | list_add_tail(&new_dir_elem->list, &dir_list); | |
5686 | } | |
5687 | break; | |
5688 | } | |
5689 | if (i == nritems) { | |
5690 | ret = btrfs_next_leaf(log, path); | |
5691 | if (ret < 0) { | |
5692 | goto next_dir_inode; | |
5693 | } else if (ret > 0) { | |
5694 | ret = 0; | |
5695 | goto next_dir_inode; | |
5696 | } | |
5697 | goto process_leaf; | |
5698 | } | |
5699 | if (min_key.offset < (u64)-1) { | |
5700 | min_key.offset++; | |
5701 | goto again; | |
5702 | } | |
5703 | next_dir_inode: | |
5704 | list_del(&dir_elem->list); | |
5705 | kfree(dir_elem); | |
5706 | } | |
5707 | ||
5708 | btrfs_free_path(path); | |
5709 | return ret; | |
5710 | } | |
5711 | ||
18aa0922 | 5712 | static int btrfs_log_all_parents(struct btrfs_trans_handle *trans, |
d0a0b78d | 5713 | struct btrfs_inode *inode, |
18aa0922 FM |
5714 | struct btrfs_log_ctx *ctx) |
5715 | { | |
3ffbd68c | 5716 | struct btrfs_fs_info *fs_info = trans->fs_info; |
18aa0922 FM |
5717 | int ret; |
5718 | struct btrfs_path *path; | |
5719 | struct btrfs_key key; | |
d0a0b78d NB |
5720 | struct btrfs_root *root = inode->root; |
5721 | const u64 ino = btrfs_ino(inode); | |
18aa0922 FM |
5722 | |
5723 | path = btrfs_alloc_path(); | |
5724 | if (!path) | |
5725 | return -ENOMEM; | |
5726 | path->skip_locking = 1; | |
5727 | path->search_commit_root = 1; | |
5728 | ||
5729 | key.objectid = ino; | |
5730 | key.type = BTRFS_INODE_REF_KEY; | |
5731 | key.offset = 0; | |
5732 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
5733 | if (ret < 0) | |
5734 | goto out; | |
5735 | ||
5736 | while (true) { | |
5737 | struct extent_buffer *leaf = path->nodes[0]; | |
5738 | int slot = path->slots[0]; | |
5739 | u32 cur_offset = 0; | |
5740 | u32 item_size; | |
5741 | unsigned long ptr; | |
5742 | ||
5743 | if (slot >= btrfs_header_nritems(leaf)) { | |
5744 | ret = btrfs_next_leaf(root, path); | |
5745 | if (ret < 0) | |
5746 | goto out; | |
5747 | else if (ret > 0) | |
5748 | break; | |
5749 | continue; | |
5750 | } | |
5751 | ||
5752 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
5753 | /* BTRFS_INODE_EXTREF_KEY is BTRFS_INODE_REF_KEY + 1 */ | |
5754 | if (key.objectid != ino || key.type > BTRFS_INODE_EXTREF_KEY) | |
5755 | break; | |
5756 | ||
5757 | item_size = btrfs_item_size_nr(leaf, slot); | |
5758 | ptr = btrfs_item_ptr_offset(leaf, slot); | |
5759 | while (cur_offset < item_size) { | |
5760 | struct btrfs_key inode_key; | |
5761 | struct inode *dir_inode; | |
5762 | ||
5763 | inode_key.type = BTRFS_INODE_ITEM_KEY; | |
5764 | inode_key.offset = 0; | |
5765 | ||
5766 | if (key.type == BTRFS_INODE_EXTREF_KEY) { | |
5767 | struct btrfs_inode_extref *extref; | |
5768 | ||
5769 | extref = (struct btrfs_inode_extref *) | |
5770 | (ptr + cur_offset); | |
5771 | inode_key.objectid = btrfs_inode_extref_parent( | |
5772 | leaf, extref); | |
5773 | cur_offset += sizeof(*extref); | |
5774 | cur_offset += btrfs_inode_extref_name_len(leaf, | |
5775 | extref); | |
5776 | } else { | |
5777 | inode_key.objectid = key.offset; | |
5778 | cur_offset = item_size; | |
5779 | } | |
5780 | ||
0b246afa | 5781 | dir_inode = btrfs_iget(fs_info->sb, &inode_key, |
18aa0922 | 5782 | root, NULL); |
0f375eed FM |
5783 | /* |
5784 | * If the parent inode was deleted, return an error to | |
5785 | * fallback to a transaction commit. This is to prevent | |
5786 | * getting an inode that was moved from one parent A to | |
5787 | * a parent B, got its former parent A deleted and then | |
5788 | * it got fsync'ed, from existing at both parents after | |
5789 | * a log replay (and the old parent still existing). | |
5790 | * Example: | |
5791 | * | |
5792 | * mkdir /mnt/A | |
5793 | * mkdir /mnt/B | |
5794 | * touch /mnt/B/bar | |
5795 | * sync | |
5796 | * mv /mnt/B/bar /mnt/A/bar | |
5797 | * mv -T /mnt/A /mnt/B | |
5798 | * fsync /mnt/B/bar | |
5799 | * <power fail> | |
5800 | * | |
5801 | * If we ignore the old parent B which got deleted, | |
5802 | * after a log replay we would have file bar linked | |
5803 | * at both parents and the old parent B would still | |
5804 | * exist. | |
5805 | */ | |
5806 | if (IS_ERR(dir_inode)) { | |
5807 | ret = PTR_ERR(dir_inode); | |
5808 | goto out; | |
5809 | } | |
18aa0922 | 5810 | |
657ed1aa FM |
5811 | if (ctx) |
5812 | ctx->log_new_dentries = false; | |
a59108a7 | 5813 | ret = btrfs_log_inode(trans, root, BTRFS_I(dir_inode), |
18aa0922 | 5814 | LOG_INODE_ALL, 0, LLONG_MAX, ctx); |
2be63d5c | 5815 | if (!ret && |
ab1717b2 | 5816 | btrfs_must_commit_transaction(trans, BTRFS_I(dir_inode))) |
2be63d5c | 5817 | ret = 1; |
657ed1aa FM |
5818 | if (!ret && ctx && ctx->log_new_dentries) |
5819 | ret = log_new_dir_dentries(trans, root, | |
f85b7379 | 5820 | BTRFS_I(dir_inode), ctx); |
18aa0922 FM |
5821 | iput(dir_inode); |
5822 | if (ret) | |
5823 | goto out; | |
5824 | } | |
5825 | path->slots[0]++; | |
5826 | } | |
5827 | ret = 0; | |
5828 | out: | |
5829 | btrfs_free_path(path); | |
5830 | return ret; | |
5831 | } | |
5832 | ||
e02119d5 CM |
5833 | /* |
5834 | * helper function around btrfs_log_inode to make sure newly created | |
5835 | * parent directories also end up in the log. A minimal inode and backref | |
5836 | * only logging is done of any parent directories that are older than | |
5837 | * the last committed transaction | |
5838 | */ | |
48a3b636 | 5839 | static int btrfs_log_inode_parent(struct btrfs_trans_handle *trans, |
19df27a9 | 5840 | struct btrfs_inode *inode, |
49dae1bc FM |
5841 | struct dentry *parent, |
5842 | const loff_t start, | |
5843 | const loff_t end, | |
41a1eada | 5844 | int inode_only, |
8b050d35 | 5845 | struct btrfs_log_ctx *ctx) |
e02119d5 | 5846 | { |
f882274b | 5847 | struct btrfs_root *root = inode->root; |
0b246afa | 5848 | struct btrfs_fs_info *fs_info = root->fs_info; |
e02119d5 | 5849 | struct super_block *sb; |
6a912213 | 5850 | struct dentry *old_parent = NULL; |
12fcfd22 | 5851 | int ret = 0; |
0b246afa | 5852 | u64 last_committed = fs_info->last_trans_committed; |
2f2ff0ee | 5853 | bool log_dentries = false; |
19df27a9 | 5854 | struct btrfs_inode *orig_inode = inode; |
12fcfd22 | 5855 | |
19df27a9 | 5856 | sb = inode->vfs_inode.i_sb; |
12fcfd22 | 5857 | |
0b246afa | 5858 | if (btrfs_test_opt(fs_info, NOTREELOG)) { |
3a5e1404 SW |
5859 | ret = 1; |
5860 | goto end_no_trans; | |
5861 | } | |
5862 | ||
995946dd MX |
5863 | /* |
5864 | * The prev transaction commit doesn't complete, we need do | |
5865 | * full commit by ourselves. | |
5866 | */ | |
0b246afa JM |
5867 | if (fs_info->last_trans_log_full_commit > |
5868 | fs_info->last_trans_committed) { | |
12fcfd22 CM |
5869 | ret = 1; |
5870 | goto end_no_trans; | |
5871 | } | |
5872 | ||
f882274b | 5873 | if (btrfs_root_refs(&root->root_item) == 0) { |
76dda93c YZ |
5874 | ret = 1; |
5875 | goto end_no_trans; | |
5876 | } | |
5877 | ||
19df27a9 NB |
5878 | ret = check_parent_dirs_for_sync(trans, inode, parent, sb, |
5879 | last_committed); | |
12fcfd22 CM |
5880 | if (ret) |
5881 | goto end_no_trans; | |
e02119d5 | 5882 | |
f2d72f42 FM |
5883 | /* |
5884 | * Skip already logged inodes or inodes corresponding to tmpfiles | |
5885 | * (since logging them is pointless, a link count of 0 means they | |
5886 | * will never be accessible). | |
5887 | */ | |
5888 | if (btrfs_inode_in_log(inode, trans->transid) || | |
5889 | inode->vfs_inode.i_nlink == 0) { | |
257c62e1 CM |
5890 | ret = BTRFS_NO_LOG_SYNC; |
5891 | goto end_no_trans; | |
5892 | } | |
5893 | ||
8b050d35 | 5894 | ret = start_log_trans(trans, root, ctx); |
4a500fd1 | 5895 | if (ret) |
e87ac136 | 5896 | goto end_no_trans; |
e02119d5 | 5897 | |
19df27a9 | 5898 | ret = btrfs_log_inode(trans, root, inode, inode_only, start, end, ctx); |
4a500fd1 YZ |
5899 | if (ret) |
5900 | goto end_trans; | |
12fcfd22 | 5901 | |
af4176b4 CM |
5902 | /* |
5903 | * for regular files, if its inode is already on disk, we don't | |
5904 | * have to worry about the parents at all. This is because | |
5905 | * we can use the last_unlink_trans field to record renames | |
5906 | * and other fun in this file. | |
5907 | */ | |
19df27a9 NB |
5908 | if (S_ISREG(inode->vfs_inode.i_mode) && |
5909 | inode->generation <= last_committed && | |
5910 | inode->last_unlink_trans <= last_committed) { | |
4a500fd1 YZ |
5911 | ret = 0; |
5912 | goto end_trans; | |
5913 | } | |
af4176b4 | 5914 | |
19df27a9 | 5915 | if (S_ISDIR(inode->vfs_inode.i_mode) && ctx && ctx->log_new_dentries) |
2f2ff0ee FM |
5916 | log_dentries = true; |
5917 | ||
18aa0922 | 5918 | /* |
01327610 | 5919 | * On unlink we must make sure all our current and old parent directory |
18aa0922 FM |
5920 | * inodes are fully logged. This is to prevent leaving dangling |
5921 | * directory index entries in directories that were our parents but are | |
5922 | * not anymore. Not doing this results in old parent directory being | |
5923 | * impossible to delete after log replay (rmdir will always fail with | |
5924 | * error -ENOTEMPTY). | |
5925 | * | |
5926 | * Example 1: | |
5927 | * | |
5928 | * mkdir testdir | |
5929 | * touch testdir/foo | |
5930 | * ln testdir/foo testdir/bar | |
5931 | * sync | |
5932 | * unlink testdir/bar | |
5933 | * xfs_io -c fsync testdir/foo | |
5934 | * <power failure> | |
5935 | * mount fs, triggers log replay | |
5936 | * | |
5937 | * If we don't log the parent directory (testdir), after log replay the | |
5938 | * directory still has an entry pointing to the file inode using the bar | |
5939 | * name, but a matching BTRFS_INODE_[REF|EXTREF]_KEY does not exist and | |
5940 | * the file inode has a link count of 1. | |
5941 | * | |
5942 | * Example 2: | |
5943 | * | |
5944 | * mkdir testdir | |
5945 | * touch foo | |
5946 | * ln foo testdir/foo2 | |
5947 | * ln foo testdir/foo3 | |
5948 | * sync | |
5949 | * unlink testdir/foo3 | |
5950 | * xfs_io -c fsync foo | |
5951 | * <power failure> | |
5952 | * mount fs, triggers log replay | |
5953 | * | |
5954 | * Similar as the first example, after log replay the parent directory | |
5955 | * testdir still has an entry pointing to the inode file with name foo3 | |
5956 | * but the file inode does not have a matching BTRFS_INODE_REF_KEY item | |
5957 | * and has a link count of 2. | |
5958 | */ | |
19df27a9 | 5959 | if (inode->last_unlink_trans > last_committed) { |
18aa0922 FM |
5960 | ret = btrfs_log_all_parents(trans, orig_inode, ctx); |
5961 | if (ret) | |
5962 | goto end_trans; | |
5963 | } | |
5964 | ||
41bd6067 FM |
5965 | /* |
5966 | * If a new hard link was added to the inode in the current transaction | |
5967 | * and its link count is now greater than 1, we need to fallback to a | |
5968 | * transaction commit, otherwise we can end up not logging all its new | |
5969 | * parents for all the hard links. Here just from the dentry used to | |
5970 | * fsync, we can not visit the ancestor inodes for all the other hard | |
5971 | * links to figure out if any is new, so we fallback to a transaction | |
5972 | * commit (instead of adding a lot of complexity of scanning a btree, | |
5973 | * since this scenario is not a common use case). | |
5974 | */ | |
5975 | if (inode->vfs_inode.i_nlink > 1 && | |
5976 | inode->last_link_trans > last_committed) { | |
5977 | ret = -EMLINK; | |
5978 | goto end_trans; | |
5979 | } | |
5980 | ||
12fcfd22 | 5981 | while (1) { |
fc64005c | 5982 | if (!parent || d_really_is_negative(parent) || sb != parent->d_sb) |
e02119d5 CM |
5983 | break; |
5984 | ||
19df27a9 NB |
5985 | inode = BTRFS_I(d_inode(parent)); |
5986 | if (root != inode->root) | |
76dda93c YZ |
5987 | break; |
5988 | ||
19df27a9 NB |
5989 | if (inode->generation > last_committed) { |
5990 | ret = btrfs_log_inode(trans, root, inode, | |
5991 | LOG_INODE_EXISTS, 0, LLONG_MAX, ctx); | |
4a500fd1 YZ |
5992 | if (ret) |
5993 | goto end_trans; | |
12fcfd22 | 5994 | } |
76dda93c | 5995 | if (IS_ROOT(parent)) |
e02119d5 | 5996 | break; |
12fcfd22 | 5997 | |
6a912213 JB |
5998 | parent = dget_parent(parent); |
5999 | dput(old_parent); | |
6000 | old_parent = parent; | |
e02119d5 | 6001 | } |
2f2ff0ee | 6002 | if (log_dentries) |
19df27a9 | 6003 | ret = log_new_dir_dentries(trans, root, orig_inode, ctx); |
2f2ff0ee FM |
6004 | else |
6005 | ret = 0; | |
4a500fd1 | 6006 | end_trans: |
6a912213 | 6007 | dput(old_parent); |
4a500fd1 | 6008 | if (ret < 0) { |
0b246afa | 6009 | btrfs_set_log_full_commit(fs_info, trans); |
4a500fd1 YZ |
6010 | ret = 1; |
6011 | } | |
8b050d35 MX |
6012 | |
6013 | if (ret) | |
6014 | btrfs_remove_log_ctx(root, ctx); | |
12fcfd22 CM |
6015 | btrfs_end_log_trans(root); |
6016 | end_no_trans: | |
6017 | return ret; | |
e02119d5 CM |
6018 | } |
6019 | ||
6020 | /* | |
6021 | * it is not safe to log dentry if the chunk root has added new | |
6022 | * chunks. This returns 0 if the dentry was logged, and 1 otherwise. | |
6023 | * If this returns 1, you must commit the transaction to safely get your | |
6024 | * data on disk. | |
6025 | */ | |
6026 | int btrfs_log_dentry_safe(struct btrfs_trans_handle *trans, | |
e5b84f7a | 6027 | struct dentry *dentry, |
49dae1bc FM |
6028 | const loff_t start, |
6029 | const loff_t end, | |
8b050d35 | 6030 | struct btrfs_log_ctx *ctx) |
e02119d5 | 6031 | { |
6a912213 JB |
6032 | struct dentry *parent = dget_parent(dentry); |
6033 | int ret; | |
6034 | ||
f882274b NB |
6035 | ret = btrfs_log_inode_parent(trans, BTRFS_I(d_inode(dentry)), parent, |
6036 | start, end, LOG_INODE_ALL, ctx); | |
6a912213 JB |
6037 | dput(parent); |
6038 | ||
6039 | return ret; | |
e02119d5 CM |
6040 | } |
6041 | ||
6042 | /* | |
6043 | * should be called during mount to recover any replay any log trees | |
6044 | * from the FS | |
6045 | */ | |
6046 | int btrfs_recover_log_trees(struct btrfs_root *log_root_tree) | |
6047 | { | |
6048 | int ret; | |
6049 | struct btrfs_path *path; | |
6050 | struct btrfs_trans_handle *trans; | |
6051 | struct btrfs_key key; | |
6052 | struct btrfs_key found_key; | |
6053 | struct btrfs_key tmp_key; | |
6054 | struct btrfs_root *log; | |
6055 | struct btrfs_fs_info *fs_info = log_root_tree->fs_info; | |
6056 | struct walk_control wc = { | |
6057 | .process_func = process_one_buffer, | |
6058 | .stage = 0, | |
6059 | }; | |
6060 | ||
e02119d5 | 6061 | path = btrfs_alloc_path(); |
db5b493a TI |
6062 | if (!path) |
6063 | return -ENOMEM; | |
6064 | ||
afcdd129 | 6065 | set_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags); |
e02119d5 | 6066 | |
4a500fd1 | 6067 | trans = btrfs_start_transaction(fs_info->tree_root, 0); |
79787eaa JM |
6068 | if (IS_ERR(trans)) { |
6069 | ret = PTR_ERR(trans); | |
6070 | goto error; | |
6071 | } | |
e02119d5 CM |
6072 | |
6073 | wc.trans = trans; | |
6074 | wc.pin = 1; | |
6075 | ||
db5b493a | 6076 | ret = walk_log_tree(trans, log_root_tree, &wc); |
79787eaa | 6077 | if (ret) { |
5d163e0e JM |
6078 | btrfs_handle_fs_error(fs_info, ret, |
6079 | "Failed to pin buffers while recovering log root tree."); | |
79787eaa JM |
6080 | goto error; |
6081 | } | |
e02119d5 CM |
6082 | |
6083 | again: | |
6084 | key.objectid = BTRFS_TREE_LOG_OBJECTID; | |
6085 | key.offset = (u64)-1; | |
962a298f | 6086 | key.type = BTRFS_ROOT_ITEM_KEY; |
e02119d5 | 6087 | |
d397712b | 6088 | while (1) { |
e02119d5 | 6089 | ret = btrfs_search_slot(NULL, log_root_tree, &key, path, 0, 0); |
79787eaa JM |
6090 | |
6091 | if (ret < 0) { | |
34d97007 | 6092 | btrfs_handle_fs_error(fs_info, ret, |
79787eaa JM |
6093 | "Couldn't find tree log root."); |
6094 | goto error; | |
6095 | } | |
e02119d5 CM |
6096 | if (ret > 0) { |
6097 | if (path->slots[0] == 0) | |
6098 | break; | |
6099 | path->slots[0]--; | |
6100 | } | |
6101 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, | |
6102 | path->slots[0]); | |
b3b4aa74 | 6103 | btrfs_release_path(path); |
e02119d5 CM |
6104 | if (found_key.objectid != BTRFS_TREE_LOG_OBJECTID) |
6105 | break; | |
6106 | ||
cb517eab | 6107 | log = btrfs_read_fs_root(log_root_tree, &found_key); |
79787eaa JM |
6108 | if (IS_ERR(log)) { |
6109 | ret = PTR_ERR(log); | |
34d97007 | 6110 | btrfs_handle_fs_error(fs_info, ret, |
79787eaa JM |
6111 | "Couldn't read tree log root."); |
6112 | goto error; | |
6113 | } | |
e02119d5 CM |
6114 | |
6115 | tmp_key.objectid = found_key.offset; | |
6116 | tmp_key.type = BTRFS_ROOT_ITEM_KEY; | |
6117 | tmp_key.offset = (u64)-1; | |
6118 | ||
6119 | wc.replay_dest = btrfs_read_fs_root_no_name(fs_info, &tmp_key); | |
79787eaa JM |
6120 | if (IS_ERR(wc.replay_dest)) { |
6121 | ret = PTR_ERR(wc.replay_dest); | |
b50c6e25 JB |
6122 | free_extent_buffer(log->node); |
6123 | free_extent_buffer(log->commit_root); | |
6124 | kfree(log); | |
5d163e0e JM |
6125 | btrfs_handle_fs_error(fs_info, ret, |
6126 | "Couldn't read target root for tree log recovery."); | |
79787eaa JM |
6127 | goto error; |
6128 | } | |
e02119d5 | 6129 | |
07d400a6 | 6130 | wc.replay_dest->log_root = log; |
5d4f98a2 | 6131 | btrfs_record_root_in_trans(trans, wc.replay_dest); |
e02119d5 | 6132 | ret = walk_log_tree(trans, log, &wc); |
e02119d5 | 6133 | |
b50c6e25 | 6134 | if (!ret && wc.stage == LOG_WALK_REPLAY_ALL) { |
e02119d5 CM |
6135 | ret = fixup_inode_link_counts(trans, wc.replay_dest, |
6136 | path); | |
e02119d5 CM |
6137 | } |
6138 | ||
900c9981 LB |
6139 | if (!ret && wc.stage == LOG_WALK_REPLAY_ALL) { |
6140 | struct btrfs_root *root = wc.replay_dest; | |
6141 | ||
6142 | btrfs_release_path(path); | |
6143 | ||
6144 | /* | |
6145 | * We have just replayed everything, and the highest | |
6146 | * objectid of fs roots probably has changed in case | |
6147 | * some inode_item's got replayed. | |
6148 | * | |
6149 | * root->objectid_mutex is not acquired as log replay | |
6150 | * could only happen during mount. | |
6151 | */ | |
6152 | ret = btrfs_find_highest_objectid(root, | |
6153 | &root->highest_objectid); | |
6154 | } | |
6155 | ||
e02119d5 | 6156 | key.offset = found_key.offset - 1; |
07d400a6 | 6157 | wc.replay_dest->log_root = NULL; |
e02119d5 | 6158 | free_extent_buffer(log->node); |
b263c2c8 | 6159 | free_extent_buffer(log->commit_root); |
e02119d5 CM |
6160 | kfree(log); |
6161 | ||
b50c6e25 JB |
6162 | if (ret) |
6163 | goto error; | |
6164 | ||
e02119d5 CM |
6165 | if (found_key.offset == 0) |
6166 | break; | |
6167 | } | |
b3b4aa74 | 6168 | btrfs_release_path(path); |
e02119d5 CM |
6169 | |
6170 | /* step one is to pin it all, step two is to replay just inodes */ | |
6171 | if (wc.pin) { | |
6172 | wc.pin = 0; | |
6173 | wc.process_func = replay_one_buffer; | |
6174 | wc.stage = LOG_WALK_REPLAY_INODES; | |
6175 | goto again; | |
6176 | } | |
6177 | /* step three is to replay everything */ | |
6178 | if (wc.stage < LOG_WALK_REPLAY_ALL) { | |
6179 | wc.stage++; | |
6180 | goto again; | |
6181 | } | |
6182 | ||
6183 | btrfs_free_path(path); | |
6184 | ||
abefa55a | 6185 | /* step 4: commit the transaction, which also unpins the blocks */ |
3a45bb20 | 6186 | ret = btrfs_commit_transaction(trans); |
abefa55a JB |
6187 | if (ret) |
6188 | return ret; | |
6189 | ||
e02119d5 CM |
6190 | free_extent_buffer(log_root_tree->node); |
6191 | log_root_tree->log_root = NULL; | |
afcdd129 | 6192 | clear_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags); |
e02119d5 | 6193 | kfree(log_root_tree); |
79787eaa | 6194 | |
abefa55a | 6195 | return 0; |
79787eaa | 6196 | error: |
b50c6e25 | 6197 | if (wc.trans) |
3a45bb20 | 6198 | btrfs_end_transaction(wc.trans); |
79787eaa JM |
6199 | btrfs_free_path(path); |
6200 | return ret; | |
e02119d5 | 6201 | } |
12fcfd22 CM |
6202 | |
6203 | /* | |
6204 | * there are some corner cases where we want to force a full | |
6205 | * commit instead of allowing a directory to be logged. | |
6206 | * | |
6207 | * They revolve around files there were unlinked from the directory, and | |
6208 | * this function updates the parent directory so that a full commit is | |
6209 | * properly done if it is fsync'd later after the unlinks are done. | |
2be63d5c FM |
6210 | * |
6211 | * Must be called before the unlink operations (updates to the subvolume tree, | |
6212 | * inodes, etc) are done. | |
12fcfd22 CM |
6213 | */ |
6214 | void btrfs_record_unlink_dir(struct btrfs_trans_handle *trans, | |
4176bdbf | 6215 | struct btrfs_inode *dir, struct btrfs_inode *inode, |
12fcfd22 CM |
6216 | int for_rename) |
6217 | { | |
af4176b4 CM |
6218 | /* |
6219 | * when we're logging a file, if it hasn't been renamed | |
6220 | * or unlinked, and its inode is fully committed on disk, | |
6221 | * we don't have to worry about walking up the directory chain | |
6222 | * to log its parents. | |
6223 | * | |
6224 | * So, we use the last_unlink_trans field to put this transid | |
6225 | * into the file. When the file is logged we check it and | |
6226 | * don't log the parents if the file is fully on disk. | |
6227 | */ | |
4176bdbf NB |
6228 | mutex_lock(&inode->log_mutex); |
6229 | inode->last_unlink_trans = trans->transid; | |
6230 | mutex_unlock(&inode->log_mutex); | |
af4176b4 | 6231 | |
12fcfd22 CM |
6232 | /* |
6233 | * if this directory was already logged any new | |
6234 | * names for this file/dir will get recorded | |
6235 | */ | |
6236 | smp_mb(); | |
4176bdbf | 6237 | if (dir->logged_trans == trans->transid) |
12fcfd22 CM |
6238 | return; |
6239 | ||
6240 | /* | |
6241 | * if the inode we're about to unlink was logged, | |
6242 | * the log will be properly updated for any new names | |
6243 | */ | |
4176bdbf | 6244 | if (inode->logged_trans == trans->transid) |
12fcfd22 CM |
6245 | return; |
6246 | ||
6247 | /* | |
6248 | * when renaming files across directories, if the directory | |
6249 | * there we're unlinking from gets fsync'd later on, there's | |
6250 | * no way to find the destination directory later and fsync it | |
6251 | * properly. So, we have to be conservative and force commits | |
6252 | * so the new name gets discovered. | |
6253 | */ | |
6254 | if (for_rename) | |
6255 | goto record; | |
6256 | ||
6257 | /* we can safely do the unlink without any special recording */ | |
6258 | return; | |
6259 | ||
6260 | record: | |
4176bdbf NB |
6261 | mutex_lock(&dir->log_mutex); |
6262 | dir->last_unlink_trans = trans->transid; | |
6263 | mutex_unlock(&dir->log_mutex); | |
1ec9a1ae FM |
6264 | } |
6265 | ||
6266 | /* | |
6267 | * Make sure that if someone attempts to fsync the parent directory of a deleted | |
6268 | * snapshot, it ends up triggering a transaction commit. This is to guarantee | |
6269 | * that after replaying the log tree of the parent directory's root we will not | |
6270 | * see the snapshot anymore and at log replay time we will not see any log tree | |
6271 | * corresponding to the deleted snapshot's root, which could lead to replaying | |
6272 | * it after replaying the log tree of the parent directory (which would replay | |
6273 | * the snapshot delete operation). | |
2be63d5c FM |
6274 | * |
6275 | * Must be called before the actual snapshot destroy operation (updates to the | |
6276 | * parent root and tree of tree roots trees, etc) are done. | |
1ec9a1ae FM |
6277 | */ |
6278 | void btrfs_record_snapshot_destroy(struct btrfs_trans_handle *trans, | |
43663557 | 6279 | struct btrfs_inode *dir) |
1ec9a1ae | 6280 | { |
43663557 NB |
6281 | mutex_lock(&dir->log_mutex); |
6282 | dir->last_unlink_trans = trans->transid; | |
6283 | mutex_unlock(&dir->log_mutex); | |
12fcfd22 CM |
6284 | } |
6285 | ||
6286 | /* | |
6287 | * Call this after adding a new name for a file and it will properly | |
6288 | * update the log to reflect the new name. | |
6289 | * | |
d4682ba0 FM |
6290 | * @ctx can not be NULL when @sync_log is false, and should be NULL when it's |
6291 | * true (because it's not used). | |
6292 | * | |
6293 | * Return value depends on whether @sync_log is true or false. | |
6294 | * When true: returns BTRFS_NEED_TRANS_COMMIT if the transaction needs to be | |
6295 | * committed by the caller, and BTRFS_DONT_NEED_TRANS_COMMIT | |
6296 | * otherwise. | |
6297 | * When false: returns BTRFS_DONT_NEED_LOG_SYNC if the caller does not need to | |
6298 | * to sync the log, BTRFS_NEED_LOG_SYNC if it needs to sync the log, | |
6299 | * or BTRFS_NEED_TRANS_COMMIT if the transaction needs to be | |
6300 | * committed (without attempting to sync the log). | |
12fcfd22 CM |
6301 | */ |
6302 | int btrfs_log_new_name(struct btrfs_trans_handle *trans, | |
9ca5fbfb | 6303 | struct btrfs_inode *inode, struct btrfs_inode *old_dir, |
d4682ba0 FM |
6304 | struct dentry *parent, |
6305 | bool sync_log, struct btrfs_log_ctx *ctx) | |
12fcfd22 | 6306 | { |
3ffbd68c | 6307 | struct btrfs_fs_info *fs_info = trans->fs_info; |
d4682ba0 | 6308 | int ret; |
12fcfd22 | 6309 | |
af4176b4 CM |
6310 | /* |
6311 | * this will force the logging code to walk the dentry chain | |
6312 | * up for the file | |
6313 | */ | |
9a6509c4 | 6314 | if (!S_ISDIR(inode->vfs_inode.i_mode)) |
9ca5fbfb | 6315 | inode->last_unlink_trans = trans->transid; |
af4176b4 | 6316 | |
12fcfd22 CM |
6317 | /* |
6318 | * if this inode hasn't been logged and directory we're renaming it | |
6319 | * from hasn't been logged, we don't need to log it | |
6320 | */ | |
9ca5fbfb NB |
6321 | if (inode->logged_trans <= fs_info->last_trans_committed && |
6322 | (!old_dir || old_dir->logged_trans <= fs_info->last_trans_committed)) | |
d4682ba0 FM |
6323 | return sync_log ? BTRFS_DONT_NEED_TRANS_COMMIT : |
6324 | BTRFS_DONT_NEED_LOG_SYNC; | |
6325 | ||
6326 | if (sync_log) { | |
6327 | struct btrfs_log_ctx ctx2; | |
6328 | ||
6329 | btrfs_init_log_ctx(&ctx2, &inode->vfs_inode); | |
6330 | ret = btrfs_log_inode_parent(trans, inode, parent, 0, LLONG_MAX, | |
6331 | LOG_INODE_EXISTS, &ctx2); | |
6332 | if (ret == BTRFS_NO_LOG_SYNC) | |
6333 | return BTRFS_DONT_NEED_TRANS_COMMIT; | |
6334 | else if (ret) | |
6335 | return BTRFS_NEED_TRANS_COMMIT; | |
6336 | ||
6337 | ret = btrfs_sync_log(trans, inode->root, &ctx2); | |
6338 | if (ret) | |
6339 | return BTRFS_NEED_TRANS_COMMIT; | |
6340 | return BTRFS_DONT_NEED_TRANS_COMMIT; | |
6341 | } | |
6342 | ||
6343 | ASSERT(ctx); | |
6344 | ret = btrfs_log_inode_parent(trans, inode, parent, 0, LLONG_MAX, | |
6345 | LOG_INODE_EXISTS, ctx); | |
6346 | if (ret == BTRFS_NO_LOG_SYNC) | |
6347 | return BTRFS_DONT_NEED_LOG_SYNC; | |
6348 | else if (ret) | |
6349 | return BTRFS_NEED_TRANS_COMMIT; | |
12fcfd22 | 6350 | |
d4682ba0 | 6351 | return BTRFS_NEED_LOG_SYNC; |
12fcfd22 CM |
6352 | } |
6353 |