]>
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> |
602cbe91 | 11 | #include "misc.h" |
9678c543 | 12 | #include "ctree.h" |
995946dd | 13 | #include "tree-log.h" |
e02119d5 CM |
14 | #include "disk-io.h" |
15 | #include "locking.h" | |
16 | #include "print-tree.h" | |
f186373f | 17 | #include "backref.h" |
ebb8765b | 18 | #include "compression.h" |
df2c95f3 | 19 | #include "qgroup.h" |
6787bb9f NB |
20 | #include "block-group.h" |
21 | #include "space-info.h" | |
d3575156 | 22 | #include "zoned.h" |
e02119d5 CM |
23 | |
24 | /* magic values for the inode_only field in btrfs_log_inode: | |
25 | * | |
26 | * LOG_INODE_ALL means to log everything | |
27 | * LOG_INODE_EXISTS means to log just enough to recreate the inode | |
28 | * during log replay | |
29 | */ | |
e13976cf DS |
30 | enum { |
31 | LOG_INODE_ALL, | |
32 | LOG_INODE_EXISTS, | |
33 | LOG_OTHER_INODE, | |
34 | LOG_OTHER_INODE_ALL, | |
35 | }; | |
e02119d5 | 36 | |
12fcfd22 CM |
37 | /* |
38 | * directory trouble cases | |
39 | * | |
40 | * 1) on rename or unlink, if the inode being unlinked isn't in the fsync | |
41 | * log, we must force a full commit before doing an fsync of the directory | |
42 | * where the unlink was done. | |
43 | * ---> record transid of last unlink/rename per directory | |
44 | * | |
45 | * mkdir foo/some_dir | |
46 | * normal commit | |
47 | * rename foo/some_dir foo2/some_dir | |
48 | * mkdir foo/some_dir | |
49 | * fsync foo/some_dir/some_file | |
50 | * | |
51 | * The fsync above will unlink the original some_dir without recording | |
52 | * it in its new location (foo2). After a crash, some_dir will be gone | |
53 | * unless the fsync of some_file forces a full commit | |
54 | * | |
55 | * 2) we must log any new names for any file or dir that is in the fsync | |
56 | * log. ---> check inode while renaming/linking. | |
57 | * | |
58 | * 2a) we must log any new names for any file or dir during rename | |
59 | * when the directory they are being removed from was logged. | |
60 | * ---> check inode and old parent dir during rename | |
61 | * | |
62 | * 2a is actually the more important variant. With the extra logging | |
63 | * a crash might unlink the old name without recreating the new one | |
64 | * | |
65 | * 3) after a crash, we must go through any directories with a link count | |
66 | * of zero and redo the rm -rf | |
67 | * | |
68 | * mkdir f1/foo | |
69 | * normal commit | |
70 | * rm -rf f1/foo | |
71 | * fsync(f1) | |
72 | * | |
73 | * The directory f1 was fully removed from the FS, but fsync was never | |
74 | * called on f1, only its parent dir. After a crash the rm -rf must | |
75 | * be replayed. This must be able to recurse down the entire | |
76 | * directory tree. The inode link count fixup code takes care of the | |
77 | * ugly details. | |
78 | */ | |
79 | ||
e02119d5 CM |
80 | /* |
81 | * stages for the tree walking. The first | |
82 | * stage (0) is to only pin down the blocks we find | |
83 | * the second stage (1) is to make sure that all the inodes | |
84 | * we find in the log are created in the subvolume. | |
85 | * | |
86 | * The last stage is to deal with directories and links and extents | |
87 | * and all the other fun semantics | |
88 | */ | |
e13976cf DS |
89 | enum { |
90 | LOG_WALK_PIN_ONLY, | |
91 | LOG_WALK_REPLAY_INODES, | |
92 | LOG_WALK_REPLAY_DIR_INDEX, | |
93 | LOG_WALK_REPLAY_ALL, | |
94 | }; | |
e02119d5 | 95 | |
12fcfd22 | 96 | static int btrfs_log_inode(struct btrfs_trans_handle *trans, |
a59108a7 | 97 | struct btrfs_root *root, struct btrfs_inode *inode, |
49dae1bc | 98 | int inode_only, |
8407f553 | 99 | struct btrfs_log_ctx *ctx); |
ec051c0f YZ |
100 | static int link_to_fixup_dir(struct btrfs_trans_handle *trans, |
101 | struct btrfs_root *root, | |
102 | struct btrfs_path *path, u64 objectid); | |
12fcfd22 CM |
103 | static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans, |
104 | struct btrfs_root *root, | |
105 | struct btrfs_root *log, | |
106 | struct btrfs_path *path, | |
107 | u64 dirid, int del_all); | |
fa1a0f42 | 108 | static void wait_log_commit(struct btrfs_root *root, int transid); |
e02119d5 CM |
109 | |
110 | /* | |
111 | * tree logging is a special write ahead log used to make sure that | |
112 | * fsyncs and O_SYNCs can happen without doing full tree commits. | |
113 | * | |
114 | * Full tree commits are expensive because they require commonly | |
115 | * modified blocks to be recowed, creating many dirty pages in the | |
116 | * extent tree an 4x-6x higher write load than ext3. | |
117 | * | |
118 | * Instead of doing a tree commit on every fsync, we use the | |
119 | * key ranges and transaction ids to find items for a given file or directory | |
120 | * that have changed in this transaction. Those items are copied into | |
121 | * a special tree (one per subvolume root), that tree is written to disk | |
122 | * and then the fsync is considered complete. | |
123 | * | |
124 | * After a crash, items are copied out of the log-tree back into the | |
125 | * subvolume tree. Any file data extents found are recorded in the extent | |
126 | * allocation tree, and the log-tree freed. | |
127 | * | |
128 | * The log tree is read three times, once to pin down all the extents it is | |
129 | * using in ram and once, once to create all the inodes logged in the tree | |
130 | * and once to do all the other items. | |
131 | */ | |
132 | ||
e02119d5 CM |
133 | /* |
134 | * start a sub transaction and setup the log tree | |
135 | * this increments the log tree writer count to make the people | |
136 | * syncing the tree wait for us to finish | |
137 | */ | |
138 | static int start_log_trans(struct btrfs_trans_handle *trans, | |
8b050d35 MX |
139 | struct btrfs_root *root, |
140 | struct btrfs_log_ctx *ctx) | |
e02119d5 | 141 | { |
0b246afa | 142 | struct btrfs_fs_info *fs_info = root->fs_info; |
47876f7c | 143 | struct btrfs_root *tree_root = fs_info->tree_root; |
fa1a0f42 | 144 | const bool zoned = btrfs_is_zoned(fs_info); |
34eb2a52 | 145 | int ret = 0; |
fa1a0f42 | 146 | bool created = false; |
7237f183 | 147 | |
47876f7c FM |
148 | /* |
149 | * First check if the log root tree was already created. If not, create | |
150 | * it before locking the root's log_mutex, just to keep lockdep happy. | |
151 | */ | |
152 | if (!test_bit(BTRFS_ROOT_HAS_LOG_TREE, &tree_root->state)) { | |
153 | mutex_lock(&tree_root->log_mutex); | |
154 | if (!fs_info->log_root_tree) { | |
155 | ret = btrfs_init_log_root_tree(trans, fs_info); | |
fa1a0f42 | 156 | if (!ret) { |
47876f7c | 157 | set_bit(BTRFS_ROOT_HAS_LOG_TREE, &tree_root->state); |
fa1a0f42 NA |
158 | created = true; |
159 | } | |
47876f7c FM |
160 | } |
161 | mutex_unlock(&tree_root->log_mutex); | |
162 | if (ret) | |
163 | return ret; | |
164 | } | |
165 | ||
7237f183 | 166 | mutex_lock(&root->log_mutex); |
34eb2a52 | 167 | |
fa1a0f42 | 168 | again: |
7237f183 | 169 | if (root->log_root) { |
fa1a0f42 NA |
170 | int index = (root->log_transid + 1) % 2; |
171 | ||
4884b8e8 | 172 | if (btrfs_need_log_full_commit(trans)) { |
50471a38 MX |
173 | ret = -EAGAIN; |
174 | goto out; | |
175 | } | |
34eb2a52 | 176 | |
fa1a0f42 NA |
177 | if (zoned && atomic_read(&root->log_commit[index])) { |
178 | wait_log_commit(root, root->log_transid - 1); | |
179 | goto again; | |
180 | } | |
181 | ||
ff782e0a | 182 | if (!root->log_start_pid) { |
27cdeb70 | 183 | clear_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state); |
34eb2a52 | 184 | root->log_start_pid = current->pid; |
ff782e0a | 185 | } else if (root->log_start_pid != current->pid) { |
27cdeb70 | 186 | set_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state); |
ff782e0a | 187 | } |
34eb2a52 | 188 | } else { |
fa1a0f42 NA |
189 | /* |
190 | * This means fs_info->log_root_tree was already created | |
191 | * for some other FS trees. Do the full commit not to mix | |
192 | * nodes from multiple log transactions to do sequential | |
193 | * writing. | |
194 | */ | |
195 | if (zoned && !created) { | |
196 | ret = -EAGAIN; | |
197 | goto out; | |
198 | } | |
199 | ||
e02119d5 | 200 | ret = btrfs_add_log_tree(trans, root); |
4a500fd1 | 201 | if (ret) |
e87ac136 | 202 | goto out; |
34eb2a52 | 203 | |
e7a79811 | 204 | set_bit(BTRFS_ROOT_HAS_LOG_TREE, &root->state); |
34eb2a52 Z |
205 | clear_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state); |
206 | root->log_start_pid = current->pid; | |
e02119d5 | 207 | } |
34eb2a52 | 208 | |
7237f183 | 209 | atomic_inc(&root->log_writers); |
75b463d2 | 210 | if (ctx && !ctx->logging_new_name) { |
34eb2a52 | 211 | int index = root->log_transid % 2; |
8b050d35 | 212 | list_add_tail(&ctx->list, &root->log_ctxs[index]); |
d1433deb | 213 | ctx->log_transid = root->log_transid; |
8b050d35 | 214 | } |
34eb2a52 | 215 | |
e87ac136 | 216 | out: |
7237f183 | 217 | mutex_unlock(&root->log_mutex); |
e87ac136 | 218 | return ret; |
e02119d5 CM |
219 | } |
220 | ||
221 | /* | |
222 | * returns 0 if there was a log transaction running and we were able | |
223 | * to join, or returns -ENOENT if there were not transactions | |
224 | * in progress | |
225 | */ | |
226 | static int join_running_log_trans(struct btrfs_root *root) | |
227 | { | |
fa1a0f42 | 228 | const bool zoned = btrfs_is_zoned(root->fs_info); |
e02119d5 CM |
229 | int ret = -ENOENT; |
230 | ||
e7a79811 FM |
231 | if (!test_bit(BTRFS_ROOT_HAS_LOG_TREE, &root->state)) |
232 | return ret; | |
233 | ||
7237f183 | 234 | mutex_lock(&root->log_mutex); |
fa1a0f42 | 235 | again: |
e02119d5 | 236 | if (root->log_root) { |
fa1a0f42 NA |
237 | int index = (root->log_transid + 1) % 2; |
238 | ||
e02119d5 | 239 | ret = 0; |
fa1a0f42 NA |
240 | if (zoned && atomic_read(&root->log_commit[index])) { |
241 | wait_log_commit(root, root->log_transid - 1); | |
242 | goto again; | |
243 | } | |
7237f183 | 244 | atomic_inc(&root->log_writers); |
e02119d5 | 245 | } |
7237f183 | 246 | mutex_unlock(&root->log_mutex); |
e02119d5 CM |
247 | return ret; |
248 | } | |
249 | ||
12fcfd22 CM |
250 | /* |
251 | * This either makes the current running log transaction wait | |
252 | * until you call btrfs_end_log_trans() or it makes any future | |
253 | * log transactions wait until you call btrfs_end_log_trans() | |
254 | */ | |
45128b08 | 255 | void btrfs_pin_log_trans(struct btrfs_root *root) |
12fcfd22 | 256 | { |
12fcfd22 | 257 | atomic_inc(&root->log_writers); |
12fcfd22 CM |
258 | } |
259 | ||
e02119d5 CM |
260 | /* |
261 | * indicate we're done making changes to the log tree | |
262 | * and wake up anyone waiting to do a sync | |
263 | */ | |
143bede5 | 264 | void btrfs_end_log_trans(struct btrfs_root *root) |
e02119d5 | 265 | { |
7237f183 | 266 | if (atomic_dec_and_test(&root->log_writers)) { |
093258e6 DS |
267 | /* atomic_dec_and_test implies a barrier */ |
268 | cond_wake_up_nomb(&root->log_writer_wait); | |
7237f183 | 269 | } |
e02119d5 CM |
270 | } |
271 | ||
247462a5 DS |
272 | static int btrfs_write_tree_block(struct extent_buffer *buf) |
273 | { | |
274 | return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start, | |
275 | buf->start + buf->len - 1); | |
276 | } | |
277 | ||
278 | static void btrfs_wait_tree_block_writeback(struct extent_buffer *buf) | |
279 | { | |
280 | filemap_fdatawait_range(buf->pages[0]->mapping, | |
281 | buf->start, buf->start + buf->len - 1); | |
282 | } | |
e02119d5 CM |
283 | |
284 | /* | |
285 | * the walk control struct is used to pass state down the chain when | |
286 | * processing the log tree. The stage field tells us which part | |
287 | * of the log tree processing we are currently doing. The others | |
288 | * are state fields used for that specific part | |
289 | */ | |
290 | struct walk_control { | |
291 | /* should we free the extent on disk when done? This is used | |
292 | * at transaction commit time while freeing a log tree | |
293 | */ | |
294 | int free; | |
295 | ||
296 | /* should we write out the extent buffer? This is used | |
297 | * while flushing the log tree to disk during a sync | |
298 | */ | |
299 | int write; | |
300 | ||
301 | /* should we wait for the extent buffer io to finish? Also used | |
302 | * while flushing the log tree to disk for a sync | |
303 | */ | |
304 | int wait; | |
305 | ||
306 | /* pin only walk, we record which extents on disk belong to the | |
307 | * log trees | |
308 | */ | |
309 | int pin; | |
310 | ||
311 | /* what stage of the replay code we're currently in */ | |
312 | int stage; | |
313 | ||
f2d72f42 FM |
314 | /* |
315 | * Ignore any items from the inode currently being processed. Needs | |
316 | * to be set every time we find a BTRFS_INODE_ITEM_KEY and we are in | |
317 | * the LOG_WALK_REPLAY_INODES stage. | |
318 | */ | |
319 | bool ignore_cur_inode; | |
320 | ||
e02119d5 CM |
321 | /* the root we are currently replaying */ |
322 | struct btrfs_root *replay_dest; | |
323 | ||
324 | /* the trans handle for the current replay */ | |
325 | struct btrfs_trans_handle *trans; | |
326 | ||
327 | /* the function that gets used to process blocks we find in the | |
328 | * tree. Note the extent_buffer might not be up to date when it is | |
329 | * passed in, and it must be checked or read if you need the data | |
330 | * inside it | |
331 | */ | |
332 | int (*process_func)(struct btrfs_root *log, struct extent_buffer *eb, | |
581c1760 | 333 | struct walk_control *wc, u64 gen, int level); |
e02119d5 CM |
334 | }; |
335 | ||
336 | /* | |
337 | * process_func used to pin down extents, write them or wait on them | |
338 | */ | |
339 | static int process_one_buffer(struct btrfs_root *log, | |
340 | struct extent_buffer *eb, | |
581c1760 | 341 | struct walk_control *wc, u64 gen, int level) |
e02119d5 | 342 | { |
0b246afa | 343 | struct btrfs_fs_info *fs_info = log->fs_info; |
b50c6e25 JB |
344 | int ret = 0; |
345 | ||
8c2a1a30 JB |
346 | /* |
347 | * If this fs is mixed then we need to be able to process the leaves to | |
348 | * pin down any logged extents, so we have to read the block. | |
349 | */ | |
0b246afa | 350 | if (btrfs_fs_incompat(fs_info, MIXED_GROUPS)) { |
581c1760 | 351 | ret = btrfs_read_buffer(eb, gen, level, NULL); |
8c2a1a30 JB |
352 | if (ret) |
353 | return ret; | |
354 | } | |
355 | ||
04018de5 | 356 | if (wc->pin) |
9fce5704 | 357 | ret = btrfs_pin_extent_for_log_replay(wc->trans, eb->start, |
2ff7e61e | 358 | eb->len); |
e02119d5 | 359 | |
b50c6e25 | 360 | if (!ret && btrfs_buffer_uptodate(eb, gen, 0)) { |
8c2a1a30 | 361 | if (wc->pin && btrfs_header_level(eb) == 0) |
bcdc428c | 362 | ret = btrfs_exclude_logged_extents(eb); |
e02119d5 CM |
363 | if (wc->write) |
364 | btrfs_write_tree_block(eb); | |
365 | if (wc->wait) | |
366 | btrfs_wait_tree_block_writeback(eb); | |
367 | } | |
b50c6e25 | 368 | return ret; |
e02119d5 CM |
369 | } |
370 | ||
371 | /* | |
372 | * Item overwrite used by replay and tree logging. eb, slot and key all refer | |
373 | * to the src data we are copying out. | |
374 | * | |
375 | * root is the tree we are copying into, and path is a scratch | |
376 | * path for use in this function (it should be released on entry and | |
377 | * will be released on exit). | |
378 | * | |
379 | * If the key is already in the destination tree the existing item is | |
380 | * overwritten. If the existing item isn't big enough, it is extended. | |
381 | * If it is too large, it is truncated. | |
382 | * | |
383 | * If the key isn't in the destination yet, a new item is inserted. | |
384 | */ | |
385 | static noinline int overwrite_item(struct btrfs_trans_handle *trans, | |
386 | struct btrfs_root *root, | |
387 | struct btrfs_path *path, | |
388 | struct extent_buffer *eb, int slot, | |
389 | struct btrfs_key *key) | |
390 | { | |
391 | int ret; | |
392 | u32 item_size; | |
393 | u64 saved_i_size = 0; | |
394 | int save_old_i_size = 0; | |
395 | unsigned long src_ptr; | |
396 | unsigned long dst_ptr; | |
397 | int overwrite_root = 0; | |
4bc4bee4 | 398 | bool inode_item = key->type == BTRFS_INODE_ITEM_KEY; |
e02119d5 CM |
399 | |
400 | if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) | |
401 | overwrite_root = 1; | |
402 | ||
403 | item_size = btrfs_item_size_nr(eb, slot); | |
404 | src_ptr = btrfs_item_ptr_offset(eb, slot); | |
405 | ||
406 | /* look for the key in the destination tree */ | |
407 | ret = btrfs_search_slot(NULL, root, key, path, 0, 0); | |
4bc4bee4 JB |
408 | if (ret < 0) |
409 | return ret; | |
410 | ||
e02119d5 CM |
411 | if (ret == 0) { |
412 | char *src_copy; | |
413 | char *dst_copy; | |
414 | u32 dst_size = btrfs_item_size_nr(path->nodes[0], | |
415 | path->slots[0]); | |
416 | if (dst_size != item_size) | |
417 | goto insert; | |
418 | ||
419 | if (item_size == 0) { | |
b3b4aa74 | 420 | btrfs_release_path(path); |
e02119d5 CM |
421 | return 0; |
422 | } | |
423 | dst_copy = kmalloc(item_size, GFP_NOFS); | |
424 | src_copy = kmalloc(item_size, GFP_NOFS); | |
2a29edc6 | 425 | if (!dst_copy || !src_copy) { |
b3b4aa74 | 426 | btrfs_release_path(path); |
2a29edc6 | 427 | kfree(dst_copy); |
428 | kfree(src_copy); | |
429 | return -ENOMEM; | |
430 | } | |
e02119d5 CM |
431 | |
432 | read_extent_buffer(eb, src_copy, src_ptr, item_size); | |
433 | ||
434 | dst_ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]); | |
435 | read_extent_buffer(path->nodes[0], dst_copy, dst_ptr, | |
436 | item_size); | |
437 | ret = memcmp(dst_copy, src_copy, item_size); | |
438 | ||
439 | kfree(dst_copy); | |
440 | kfree(src_copy); | |
441 | /* | |
442 | * they have the same contents, just return, this saves | |
443 | * us from cowing blocks in the destination tree and doing | |
444 | * extra writes that may not have been done by a previous | |
445 | * sync | |
446 | */ | |
447 | if (ret == 0) { | |
b3b4aa74 | 448 | btrfs_release_path(path); |
e02119d5 CM |
449 | return 0; |
450 | } | |
451 | ||
4bc4bee4 JB |
452 | /* |
453 | * We need to load the old nbytes into the inode so when we | |
454 | * replay the extents we've logged we get the right nbytes. | |
455 | */ | |
456 | if (inode_item) { | |
457 | struct btrfs_inode_item *item; | |
458 | u64 nbytes; | |
d555438b | 459 | u32 mode; |
4bc4bee4 JB |
460 | |
461 | item = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
462 | struct btrfs_inode_item); | |
463 | nbytes = btrfs_inode_nbytes(path->nodes[0], item); | |
464 | item = btrfs_item_ptr(eb, slot, | |
465 | struct btrfs_inode_item); | |
466 | btrfs_set_inode_nbytes(eb, item, nbytes); | |
d555438b JB |
467 | |
468 | /* | |
469 | * If this is a directory we need to reset the i_size to | |
470 | * 0 so that we can set it up properly when replaying | |
471 | * the rest of the items in this log. | |
472 | */ | |
473 | mode = btrfs_inode_mode(eb, item); | |
474 | if (S_ISDIR(mode)) | |
475 | btrfs_set_inode_size(eb, item, 0); | |
4bc4bee4 JB |
476 | } |
477 | } else if (inode_item) { | |
478 | struct btrfs_inode_item *item; | |
d555438b | 479 | u32 mode; |
4bc4bee4 JB |
480 | |
481 | /* | |
482 | * New inode, set nbytes to 0 so that the nbytes comes out | |
483 | * properly when we replay the extents. | |
484 | */ | |
485 | item = btrfs_item_ptr(eb, slot, struct btrfs_inode_item); | |
486 | btrfs_set_inode_nbytes(eb, item, 0); | |
d555438b JB |
487 | |
488 | /* | |
489 | * If this is a directory we need to reset the i_size to 0 so | |
490 | * that we can set it up properly when replaying the rest of | |
491 | * the items in this log. | |
492 | */ | |
493 | mode = btrfs_inode_mode(eb, item); | |
494 | if (S_ISDIR(mode)) | |
495 | btrfs_set_inode_size(eb, item, 0); | |
e02119d5 CM |
496 | } |
497 | insert: | |
b3b4aa74 | 498 | btrfs_release_path(path); |
e02119d5 | 499 | /* try to insert the key into the destination tree */ |
df8d116f | 500 | path->skip_release_on_error = 1; |
e02119d5 CM |
501 | ret = btrfs_insert_empty_item(trans, root, path, |
502 | key, item_size); | |
df8d116f | 503 | path->skip_release_on_error = 0; |
e02119d5 CM |
504 | |
505 | /* make sure any existing item is the correct size */ | |
df8d116f | 506 | if (ret == -EEXIST || ret == -EOVERFLOW) { |
e02119d5 CM |
507 | u32 found_size; |
508 | found_size = btrfs_item_size_nr(path->nodes[0], | |
509 | path->slots[0]); | |
143bede5 | 510 | if (found_size > item_size) |
78ac4f9e | 511 | btrfs_truncate_item(path, item_size, 1); |
143bede5 | 512 | else if (found_size < item_size) |
c71dd880 | 513 | btrfs_extend_item(path, item_size - found_size); |
e02119d5 | 514 | } else if (ret) { |
4a500fd1 | 515 | return ret; |
e02119d5 CM |
516 | } |
517 | dst_ptr = btrfs_item_ptr_offset(path->nodes[0], | |
518 | path->slots[0]); | |
519 | ||
520 | /* don't overwrite an existing inode if the generation number | |
521 | * was logged as zero. This is done when the tree logging code | |
522 | * is just logging an inode to make sure it exists after recovery. | |
523 | * | |
524 | * Also, don't overwrite i_size on directories during replay. | |
525 | * log replay inserts and removes directory items based on the | |
526 | * state of the tree found in the subvolume, and i_size is modified | |
527 | * as it goes | |
528 | */ | |
529 | if (key->type == BTRFS_INODE_ITEM_KEY && ret == -EEXIST) { | |
530 | struct btrfs_inode_item *src_item; | |
531 | struct btrfs_inode_item *dst_item; | |
532 | ||
533 | src_item = (struct btrfs_inode_item *)src_ptr; | |
534 | dst_item = (struct btrfs_inode_item *)dst_ptr; | |
535 | ||
1a4bcf47 FM |
536 | if (btrfs_inode_generation(eb, src_item) == 0) { |
537 | struct extent_buffer *dst_eb = path->nodes[0]; | |
2f2ff0ee | 538 | const u64 ino_size = btrfs_inode_size(eb, src_item); |
1a4bcf47 | 539 | |
2f2ff0ee FM |
540 | /* |
541 | * For regular files an ino_size == 0 is used only when | |
542 | * logging that an inode exists, as part of a directory | |
543 | * fsync, and the inode wasn't fsynced before. In this | |
544 | * case don't set the size of the inode in the fs/subvol | |
545 | * tree, otherwise we would be throwing valid data away. | |
546 | */ | |
1a4bcf47 | 547 | if (S_ISREG(btrfs_inode_mode(eb, src_item)) && |
2f2ff0ee | 548 | S_ISREG(btrfs_inode_mode(dst_eb, dst_item)) && |
60d48e2e DS |
549 | ino_size != 0) |
550 | btrfs_set_inode_size(dst_eb, dst_item, ino_size); | |
e02119d5 | 551 | goto no_copy; |
1a4bcf47 | 552 | } |
e02119d5 CM |
553 | |
554 | if (overwrite_root && | |
555 | S_ISDIR(btrfs_inode_mode(eb, src_item)) && | |
556 | S_ISDIR(btrfs_inode_mode(path->nodes[0], dst_item))) { | |
557 | save_old_i_size = 1; | |
558 | saved_i_size = btrfs_inode_size(path->nodes[0], | |
559 | dst_item); | |
560 | } | |
561 | } | |
562 | ||
563 | copy_extent_buffer(path->nodes[0], eb, dst_ptr, | |
564 | src_ptr, item_size); | |
565 | ||
566 | if (save_old_i_size) { | |
567 | struct btrfs_inode_item *dst_item; | |
568 | dst_item = (struct btrfs_inode_item *)dst_ptr; | |
569 | btrfs_set_inode_size(path->nodes[0], dst_item, saved_i_size); | |
570 | } | |
571 | ||
572 | /* make sure the generation is filled in */ | |
573 | if (key->type == BTRFS_INODE_ITEM_KEY) { | |
574 | struct btrfs_inode_item *dst_item; | |
575 | dst_item = (struct btrfs_inode_item *)dst_ptr; | |
576 | if (btrfs_inode_generation(path->nodes[0], dst_item) == 0) { | |
577 | btrfs_set_inode_generation(path->nodes[0], dst_item, | |
578 | trans->transid); | |
579 | } | |
580 | } | |
581 | no_copy: | |
582 | btrfs_mark_buffer_dirty(path->nodes[0]); | |
b3b4aa74 | 583 | btrfs_release_path(path); |
e02119d5 CM |
584 | return 0; |
585 | } | |
586 | ||
587 | /* | |
588 | * simple helper to read an inode off the disk from a given root | |
589 | * This can only be called for subvolume roots and not for the log | |
590 | */ | |
591 | static noinline struct inode *read_one_inode(struct btrfs_root *root, | |
592 | u64 objectid) | |
593 | { | |
594 | struct inode *inode; | |
e02119d5 | 595 | |
0202e83f | 596 | inode = btrfs_iget(root->fs_info->sb, objectid, root); |
2e19f1f9 | 597 | if (IS_ERR(inode)) |
5d4f98a2 | 598 | inode = NULL; |
e02119d5 CM |
599 | return inode; |
600 | } | |
601 | ||
602 | /* replays a single extent in 'eb' at 'slot' with 'key' into the | |
603 | * subvolume 'root'. path is released on entry and should be released | |
604 | * on exit. | |
605 | * | |
606 | * extents in the log tree have not been allocated out of the extent | |
607 | * tree yet. So, this completes the allocation, taking a reference | |
608 | * as required if the extent already exists or creating a new extent | |
609 | * if it isn't in the extent allocation tree yet. | |
610 | * | |
611 | * The extent is inserted into the file, dropping any existing extents | |
612 | * from the file that overlap the new one. | |
613 | */ | |
614 | static noinline int replay_one_extent(struct btrfs_trans_handle *trans, | |
615 | struct btrfs_root *root, | |
616 | struct btrfs_path *path, | |
617 | struct extent_buffer *eb, int slot, | |
618 | struct btrfs_key *key) | |
619 | { | |
5893dfb9 | 620 | struct btrfs_drop_extents_args drop_args = { 0 }; |
0b246afa | 621 | struct btrfs_fs_info *fs_info = root->fs_info; |
e02119d5 | 622 | int found_type; |
e02119d5 | 623 | u64 extent_end; |
e02119d5 | 624 | u64 start = key->offset; |
4bc4bee4 | 625 | u64 nbytes = 0; |
e02119d5 CM |
626 | struct btrfs_file_extent_item *item; |
627 | struct inode *inode = NULL; | |
628 | unsigned long size; | |
629 | int ret = 0; | |
630 | ||
631 | item = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item); | |
632 | found_type = btrfs_file_extent_type(eb, item); | |
633 | ||
d899e052 | 634 | if (found_type == BTRFS_FILE_EXTENT_REG || |
4bc4bee4 JB |
635 | found_type == BTRFS_FILE_EXTENT_PREALLOC) { |
636 | nbytes = btrfs_file_extent_num_bytes(eb, item); | |
637 | extent_end = start + nbytes; | |
638 | ||
639 | /* | |
640 | * We don't add to the inodes nbytes if we are prealloc or a | |
641 | * hole. | |
642 | */ | |
643 | if (btrfs_file_extent_disk_bytenr(eb, item) == 0) | |
644 | nbytes = 0; | |
645 | } else if (found_type == BTRFS_FILE_EXTENT_INLINE) { | |
e41ca589 | 646 | size = btrfs_file_extent_ram_bytes(eb, item); |
4bc4bee4 | 647 | nbytes = btrfs_file_extent_ram_bytes(eb, item); |
da17066c | 648 | extent_end = ALIGN(start + size, |
0b246afa | 649 | fs_info->sectorsize); |
e02119d5 CM |
650 | } else { |
651 | ret = 0; | |
652 | goto out; | |
653 | } | |
654 | ||
655 | inode = read_one_inode(root, key->objectid); | |
656 | if (!inode) { | |
657 | ret = -EIO; | |
658 | goto out; | |
659 | } | |
660 | ||
661 | /* | |
662 | * first check to see if we already have this extent in the | |
663 | * file. This must be done before the btrfs_drop_extents run | |
664 | * so we don't try to drop this extent. | |
665 | */ | |
f85b7379 DS |
666 | ret = btrfs_lookup_file_extent(trans, root, path, |
667 | btrfs_ino(BTRFS_I(inode)), start, 0); | |
e02119d5 | 668 | |
d899e052 YZ |
669 | if (ret == 0 && |
670 | (found_type == BTRFS_FILE_EXTENT_REG || | |
671 | found_type == BTRFS_FILE_EXTENT_PREALLOC)) { | |
e02119d5 CM |
672 | struct btrfs_file_extent_item cmp1; |
673 | struct btrfs_file_extent_item cmp2; | |
674 | struct btrfs_file_extent_item *existing; | |
675 | struct extent_buffer *leaf; | |
676 | ||
677 | leaf = path->nodes[0]; | |
678 | existing = btrfs_item_ptr(leaf, path->slots[0], | |
679 | struct btrfs_file_extent_item); | |
680 | ||
681 | read_extent_buffer(eb, &cmp1, (unsigned long)item, | |
682 | sizeof(cmp1)); | |
683 | read_extent_buffer(leaf, &cmp2, (unsigned long)existing, | |
684 | sizeof(cmp2)); | |
685 | ||
686 | /* | |
687 | * we already have a pointer to this exact extent, | |
688 | * we don't have to do anything | |
689 | */ | |
690 | if (memcmp(&cmp1, &cmp2, sizeof(cmp1)) == 0) { | |
b3b4aa74 | 691 | btrfs_release_path(path); |
e02119d5 CM |
692 | goto out; |
693 | } | |
694 | } | |
b3b4aa74 | 695 | btrfs_release_path(path); |
e02119d5 CM |
696 | |
697 | /* drop any overlapping extents */ | |
5893dfb9 FM |
698 | drop_args.start = start; |
699 | drop_args.end = extent_end; | |
700 | drop_args.drop_cache = true; | |
701 | ret = btrfs_drop_extents(trans, root, BTRFS_I(inode), &drop_args); | |
3650860b JB |
702 | if (ret) |
703 | goto out; | |
e02119d5 | 704 | |
07d400a6 YZ |
705 | if (found_type == BTRFS_FILE_EXTENT_REG || |
706 | found_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
5d4f98a2 | 707 | u64 offset; |
07d400a6 YZ |
708 | unsigned long dest_offset; |
709 | struct btrfs_key ins; | |
710 | ||
3168021c FM |
711 | if (btrfs_file_extent_disk_bytenr(eb, item) == 0 && |
712 | btrfs_fs_incompat(fs_info, NO_HOLES)) | |
713 | goto update_inode; | |
714 | ||
07d400a6 YZ |
715 | ret = btrfs_insert_empty_item(trans, root, path, key, |
716 | sizeof(*item)); | |
3650860b JB |
717 | if (ret) |
718 | goto out; | |
07d400a6 YZ |
719 | dest_offset = btrfs_item_ptr_offset(path->nodes[0], |
720 | path->slots[0]); | |
721 | copy_extent_buffer(path->nodes[0], eb, dest_offset, | |
722 | (unsigned long)item, sizeof(*item)); | |
723 | ||
724 | ins.objectid = btrfs_file_extent_disk_bytenr(eb, item); | |
725 | ins.offset = btrfs_file_extent_disk_num_bytes(eb, item); | |
726 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
5d4f98a2 | 727 | offset = key->offset - btrfs_file_extent_offset(eb, item); |
07d400a6 | 728 | |
df2c95f3 QW |
729 | /* |
730 | * Manually record dirty extent, as here we did a shallow | |
731 | * file extent item copy and skip normal backref update, | |
732 | * but modifying extent tree all by ourselves. | |
733 | * So need to manually record dirty extent for qgroup, | |
734 | * as the owner of the file extent changed from log tree | |
735 | * (doesn't affect qgroup) to fs/file tree(affects qgroup) | |
736 | */ | |
a95f3aaf | 737 | ret = btrfs_qgroup_trace_extent(trans, |
df2c95f3 QW |
738 | btrfs_file_extent_disk_bytenr(eb, item), |
739 | btrfs_file_extent_disk_num_bytes(eb, item), | |
740 | GFP_NOFS); | |
741 | if (ret < 0) | |
742 | goto out; | |
743 | ||
07d400a6 | 744 | if (ins.objectid > 0) { |
82fa113f | 745 | struct btrfs_ref ref = { 0 }; |
07d400a6 YZ |
746 | u64 csum_start; |
747 | u64 csum_end; | |
748 | LIST_HEAD(ordered_sums); | |
82fa113f | 749 | |
07d400a6 YZ |
750 | /* |
751 | * is this extent already allocated in the extent | |
752 | * allocation tree? If so, just add a reference | |
753 | */ | |
2ff7e61e | 754 | ret = btrfs_lookup_data_extent(fs_info, ins.objectid, |
07d400a6 YZ |
755 | ins.offset); |
756 | if (ret == 0) { | |
82fa113f QW |
757 | btrfs_init_generic_ref(&ref, |
758 | BTRFS_ADD_DELAYED_REF, | |
759 | ins.objectid, ins.offset, 0); | |
760 | btrfs_init_data_ref(&ref, | |
761 | root->root_key.objectid, | |
b06c4bf5 | 762 | key->objectid, offset); |
82fa113f | 763 | ret = btrfs_inc_extent_ref(trans, &ref); |
b50c6e25 JB |
764 | if (ret) |
765 | goto out; | |
07d400a6 YZ |
766 | } else { |
767 | /* | |
768 | * insert the extent pointer in the extent | |
769 | * allocation tree | |
770 | */ | |
5d4f98a2 | 771 | ret = btrfs_alloc_logged_file_extent(trans, |
2ff7e61e | 772 | root->root_key.objectid, |
5d4f98a2 | 773 | key->objectid, offset, &ins); |
b50c6e25 JB |
774 | if (ret) |
775 | goto out; | |
07d400a6 | 776 | } |
b3b4aa74 | 777 | btrfs_release_path(path); |
07d400a6 YZ |
778 | |
779 | if (btrfs_file_extent_compression(eb, item)) { | |
780 | csum_start = ins.objectid; | |
781 | csum_end = csum_start + ins.offset; | |
782 | } else { | |
783 | csum_start = ins.objectid + | |
784 | btrfs_file_extent_offset(eb, item); | |
785 | csum_end = csum_start + | |
786 | btrfs_file_extent_num_bytes(eb, item); | |
787 | } | |
788 | ||
789 | ret = btrfs_lookup_csums_range(root->log_root, | |
790 | csum_start, csum_end - 1, | |
a2de733c | 791 | &ordered_sums, 0); |
3650860b JB |
792 | if (ret) |
793 | goto out; | |
b84b8390 FM |
794 | /* |
795 | * Now delete all existing cums in the csum root that | |
796 | * cover our range. We do this because we can have an | |
797 | * extent that is completely referenced by one file | |
798 | * extent item and partially referenced by another | |
799 | * file extent item (like after using the clone or | |
800 | * extent_same ioctls). In this case if we end up doing | |
801 | * the replay of the one that partially references the | |
802 | * extent first, and we do not do the csum deletion | |
803 | * below, we can get 2 csum items in the csum tree that | |
804 | * overlap each other. For example, imagine our log has | |
805 | * the two following file extent items: | |
806 | * | |
807 | * key (257 EXTENT_DATA 409600) | |
808 | * extent data disk byte 12845056 nr 102400 | |
809 | * extent data offset 20480 nr 20480 ram 102400 | |
810 | * | |
811 | * key (257 EXTENT_DATA 819200) | |
812 | * extent data disk byte 12845056 nr 102400 | |
813 | * extent data offset 0 nr 102400 ram 102400 | |
814 | * | |
815 | * Where the second one fully references the 100K extent | |
816 | * that starts at disk byte 12845056, and the log tree | |
817 | * has a single csum item that covers the entire range | |
818 | * of the extent: | |
819 | * | |
820 | * key (EXTENT_CSUM EXTENT_CSUM 12845056) itemsize 100 | |
821 | * | |
822 | * After the first file extent item is replayed, the | |
823 | * csum tree gets the following csum item: | |
824 | * | |
825 | * key (EXTENT_CSUM EXTENT_CSUM 12865536) itemsize 20 | |
826 | * | |
827 | * Which covers the 20K sub-range starting at offset 20K | |
828 | * of our extent. Now when we replay the second file | |
829 | * extent item, if we do not delete existing csum items | |
830 | * that cover any of its blocks, we end up getting two | |
831 | * csum items in our csum tree that overlap each other: | |
832 | * | |
833 | * key (EXTENT_CSUM EXTENT_CSUM 12845056) itemsize 100 | |
834 | * key (EXTENT_CSUM EXTENT_CSUM 12865536) itemsize 20 | |
835 | * | |
836 | * Which is a problem, because after this anyone trying | |
837 | * to lookup up for the checksum of any block of our | |
838 | * extent starting at an offset of 40K or higher, will | |
839 | * end up looking at the second csum item only, which | |
840 | * does not contain the checksum for any block starting | |
841 | * at offset 40K or higher of our extent. | |
842 | */ | |
07d400a6 YZ |
843 | while (!list_empty(&ordered_sums)) { |
844 | struct btrfs_ordered_sum *sums; | |
845 | sums = list_entry(ordered_sums.next, | |
846 | struct btrfs_ordered_sum, | |
847 | list); | |
b84b8390 | 848 | if (!ret) |
40e046ac FM |
849 | ret = btrfs_del_csums(trans, |
850 | fs_info->csum_root, | |
5b4aacef JM |
851 | sums->bytenr, |
852 | sums->len); | |
3650860b JB |
853 | if (!ret) |
854 | ret = btrfs_csum_file_blocks(trans, | |
0b246afa | 855 | fs_info->csum_root, sums); |
07d400a6 YZ |
856 | list_del(&sums->list); |
857 | kfree(sums); | |
858 | } | |
3650860b JB |
859 | if (ret) |
860 | goto out; | |
07d400a6 | 861 | } else { |
b3b4aa74 | 862 | btrfs_release_path(path); |
07d400a6 YZ |
863 | } |
864 | } else if (found_type == BTRFS_FILE_EXTENT_INLINE) { | |
865 | /* inline extents are easy, we just overwrite them */ | |
866 | ret = overwrite_item(trans, root, path, eb, slot, key); | |
3650860b JB |
867 | if (ret) |
868 | goto out; | |
07d400a6 | 869 | } |
e02119d5 | 870 | |
9ddc959e JB |
871 | ret = btrfs_inode_set_file_extent_range(BTRFS_I(inode), start, |
872 | extent_end - start); | |
873 | if (ret) | |
874 | goto out; | |
875 | ||
3168021c | 876 | update_inode: |
2766ff61 | 877 | btrfs_update_inode_bytes(BTRFS_I(inode), nbytes, drop_args.bytes_found); |
9a56fcd1 | 878 | ret = btrfs_update_inode(trans, root, BTRFS_I(inode)); |
e02119d5 CM |
879 | out: |
880 | if (inode) | |
881 | iput(inode); | |
882 | return ret; | |
883 | } | |
884 | ||
885 | /* | |
886 | * when cleaning up conflicts between the directory names in the | |
887 | * subvolume, directory names in the log and directory names in the | |
888 | * inode back references, we may have to unlink inodes from directories. | |
889 | * | |
890 | * This is a helper function to do the unlink of a specific directory | |
891 | * item | |
892 | */ | |
893 | static noinline int drop_one_dir_item(struct btrfs_trans_handle *trans, | |
894 | struct btrfs_root *root, | |
895 | struct btrfs_path *path, | |
207e7d92 | 896 | struct btrfs_inode *dir, |
e02119d5 CM |
897 | struct btrfs_dir_item *di) |
898 | { | |
899 | struct inode *inode; | |
900 | char *name; | |
901 | int name_len; | |
902 | struct extent_buffer *leaf; | |
903 | struct btrfs_key location; | |
904 | int ret; | |
905 | ||
906 | leaf = path->nodes[0]; | |
907 | ||
908 | btrfs_dir_item_key_to_cpu(leaf, di, &location); | |
909 | name_len = btrfs_dir_name_len(leaf, di); | |
910 | name = kmalloc(name_len, GFP_NOFS); | |
2a29edc6 | 911 | if (!name) |
912 | return -ENOMEM; | |
913 | ||
e02119d5 | 914 | read_extent_buffer(leaf, name, (unsigned long)(di + 1), name_len); |
b3b4aa74 | 915 | btrfs_release_path(path); |
e02119d5 CM |
916 | |
917 | inode = read_one_inode(root, location.objectid); | |
c00e9493 | 918 | if (!inode) { |
3650860b JB |
919 | ret = -EIO; |
920 | goto out; | |
c00e9493 | 921 | } |
e02119d5 | 922 | |
ec051c0f | 923 | ret = link_to_fixup_dir(trans, root, path, location.objectid); |
3650860b JB |
924 | if (ret) |
925 | goto out; | |
12fcfd22 | 926 | |
207e7d92 NB |
927 | ret = btrfs_unlink_inode(trans, root, dir, BTRFS_I(inode), name, |
928 | name_len); | |
3650860b JB |
929 | if (ret) |
930 | goto out; | |
ada9af21 | 931 | else |
e5c304e6 | 932 | ret = btrfs_run_delayed_items(trans); |
3650860b | 933 | out: |
e02119d5 | 934 | kfree(name); |
e02119d5 CM |
935 | iput(inode); |
936 | return ret; | |
937 | } | |
938 | ||
939 | /* | |
940 | * helper function to see if a given name and sequence number found | |
941 | * in an inode back reference are already in a directory and correctly | |
942 | * point to this inode | |
943 | */ | |
944 | static noinline int inode_in_dir(struct btrfs_root *root, | |
945 | struct btrfs_path *path, | |
946 | u64 dirid, u64 objectid, u64 index, | |
947 | const char *name, int name_len) | |
948 | { | |
949 | struct btrfs_dir_item *di; | |
950 | struct btrfs_key location; | |
951 | int match = 0; | |
952 | ||
953 | di = btrfs_lookup_dir_index_item(NULL, root, path, dirid, | |
954 | index, name, name_len, 0); | |
955 | if (di && !IS_ERR(di)) { | |
956 | btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location); | |
957 | if (location.objectid != objectid) | |
958 | goto out; | |
959 | } else | |
960 | goto out; | |
b3b4aa74 | 961 | btrfs_release_path(path); |
e02119d5 CM |
962 | |
963 | di = btrfs_lookup_dir_item(NULL, root, path, dirid, name, name_len, 0); | |
964 | if (di && !IS_ERR(di)) { | |
965 | btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location); | |
966 | if (location.objectid != objectid) | |
967 | goto out; | |
968 | } else | |
969 | goto out; | |
970 | match = 1; | |
971 | out: | |
b3b4aa74 | 972 | btrfs_release_path(path); |
e02119d5 CM |
973 | return match; |
974 | } | |
975 | ||
976 | /* | |
977 | * helper function to check a log tree for a named back reference in | |
978 | * an inode. This is used to decide if a back reference that is | |
979 | * found in the subvolume conflicts with what we find in the log. | |
980 | * | |
981 | * inode backreferences may have multiple refs in a single item, | |
982 | * during replay we process one reference at a time, and we don't | |
983 | * want to delete valid links to a file from the subvolume if that | |
984 | * link is also in the log. | |
985 | */ | |
986 | static noinline int backref_in_log(struct btrfs_root *log, | |
987 | struct btrfs_key *key, | |
f186373f | 988 | u64 ref_objectid, |
df8d116f | 989 | const char *name, int namelen) |
e02119d5 CM |
990 | { |
991 | struct btrfs_path *path; | |
e02119d5 | 992 | int ret; |
e02119d5 CM |
993 | |
994 | path = btrfs_alloc_path(); | |
2a29edc6 | 995 | if (!path) |
996 | return -ENOMEM; | |
997 | ||
e02119d5 | 998 | ret = btrfs_search_slot(NULL, log, key, path, 0, 0); |
d3316c82 NB |
999 | if (ret < 0) { |
1000 | goto out; | |
1001 | } else if (ret == 1) { | |
89cbf5f6 | 1002 | ret = 0; |
f186373f MF |
1003 | goto out; |
1004 | } | |
1005 | ||
89cbf5f6 NB |
1006 | if (key->type == BTRFS_INODE_EXTREF_KEY) |
1007 | ret = !!btrfs_find_name_in_ext_backref(path->nodes[0], | |
1008 | path->slots[0], | |
1009 | ref_objectid, | |
1010 | name, namelen); | |
1011 | else | |
1012 | ret = !!btrfs_find_name_in_backref(path->nodes[0], | |
1013 | path->slots[0], | |
1014 | name, namelen); | |
e02119d5 CM |
1015 | out: |
1016 | btrfs_free_path(path); | |
89cbf5f6 | 1017 | return ret; |
e02119d5 CM |
1018 | } |
1019 | ||
5a1d7843 | 1020 | static inline int __add_inode_ref(struct btrfs_trans_handle *trans, |
e02119d5 | 1021 | struct btrfs_root *root, |
e02119d5 | 1022 | struct btrfs_path *path, |
5a1d7843 | 1023 | struct btrfs_root *log_root, |
94c91a1f NB |
1024 | struct btrfs_inode *dir, |
1025 | struct btrfs_inode *inode, | |
f186373f MF |
1026 | u64 inode_objectid, u64 parent_objectid, |
1027 | u64 ref_index, char *name, int namelen, | |
1028 | int *search_done) | |
e02119d5 | 1029 | { |
34f3e4f2 | 1030 | int ret; |
f186373f MF |
1031 | char *victim_name; |
1032 | int victim_name_len; | |
1033 | struct extent_buffer *leaf; | |
5a1d7843 | 1034 | struct btrfs_dir_item *di; |
f186373f MF |
1035 | struct btrfs_key search_key; |
1036 | struct btrfs_inode_extref *extref; | |
c622ae60 | 1037 | |
f186373f MF |
1038 | again: |
1039 | /* Search old style refs */ | |
1040 | search_key.objectid = inode_objectid; | |
1041 | search_key.type = BTRFS_INODE_REF_KEY; | |
1042 | search_key.offset = parent_objectid; | |
1043 | ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0); | |
e02119d5 | 1044 | if (ret == 0) { |
e02119d5 CM |
1045 | struct btrfs_inode_ref *victim_ref; |
1046 | unsigned long ptr; | |
1047 | unsigned long ptr_end; | |
f186373f MF |
1048 | |
1049 | leaf = path->nodes[0]; | |
e02119d5 CM |
1050 | |
1051 | /* are we trying to overwrite a back ref for the root directory | |
1052 | * if so, just jump out, we're done | |
1053 | */ | |
f186373f | 1054 | if (search_key.objectid == search_key.offset) |
5a1d7843 | 1055 | return 1; |
e02119d5 CM |
1056 | |
1057 | /* check all the names in this back reference to see | |
1058 | * if they are in the log. if so, we allow them to stay | |
1059 | * otherwise they must be unlinked as a conflict | |
1060 | */ | |
1061 | ptr = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
1062 | ptr_end = ptr + btrfs_item_size_nr(leaf, path->slots[0]); | |
d397712b | 1063 | while (ptr < ptr_end) { |
e02119d5 CM |
1064 | victim_ref = (struct btrfs_inode_ref *)ptr; |
1065 | victim_name_len = btrfs_inode_ref_name_len(leaf, | |
1066 | victim_ref); | |
1067 | victim_name = kmalloc(victim_name_len, GFP_NOFS); | |
3650860b JB |
1068 | if (!victim_name) |
1069 | return -ENOMEM; | |
e02119d5 CM |
1070 | |
1071 | read_extent_buffer(leaf, victim_name, | |
1072 | (unsigned long)(victim_ref + 1), | |
1073 | victim_name_len); | |
1074 | ||
d3316c82 NB |
1075 | ret = backref_in_log(log_root, &search_key, |
1076 | parent_objectid, victim_name, | |
1077 | victim_name_len); | |
1078 | if (ret < 0) { | |
1079 | kfree(victim_name); | |
1080 | return ret; | |
1081 | } else if (!ret) { | |
94c91a1f | 1082 | inc_nlink(&inode->vfs_inode); |
b3b4aa74 | 1083 | btrfs_release_path(path); |
12fcfd22 | 1084 | |
94c91a1f | 1085 | ret = btrfs_unlink_inode(trans, root, dir, inode, |
4ec5934e | 1086 | victim_name, victim_name_len); |
f186373f | 1087 | kfree(victim_name); |
3650860b JB |
1088 | if (ret) |
1089 | return ret; | |
e5c304e6 | 1090 | ret = btrfs_run_delayed_items(trans); |
ada9af21 FDBM |
1091 | if (ret) |
1092 | return ret; | |
f186373f MF |
1093 | *search_done = 1; |
1094 | goto again; | |
e02119d5 CM |
1095 | } |
1096 | kfree(victim_name); | |
f186373f | 1097 | |
e02119d5 CM |
1098 | ptr = (unsigned long)(victim_ref + 1) + victim_name_len; |
1099 | } | |
e02119d5 | 1100 | |
c622ae60 | 1101 | /* |
1102 | * NOTE: we have searched root tree and checked the | |
bb7ab3b9 | 1103 | * corresponding ref, it does not need to check again. |
c622ae60 | 1104 | */ |
5a1d7843 | 1105 | *search_done = 1; |
e02119d5 | 1106 | } |
b3b4aa74 | 1107 | btrfs_release_path(path); |
e02119d5 | 1108 | |
f186373f MF |
1109 | /* Same search but for extended refs */ |
1110 | extref = btrfs_lookup_inode_extref(NULL, root, path, name, namelen, | |
1111 | inode_objectid, parent_objectid, 0, | |
1112 | 0); | |
1113 | if (!IS_ERR_OR_NULL(extref)) { | |
1114 | u32 item_size; | |
1115 | u32 cur_offset = 0; | |
1116 | unsigned long base; | |
1117 | struct inode *victim_parent; | |
1118 | ||
1119 | leaf = path->nodes[0]; | |
1120 | ||
1121 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1122 | base = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
1123 | ||
1124 | while (cur_offset < item_size) { | |
dd9ef135 | 1125 | extref = (struct btrfs_inode_extref *)(base + cur_offset); |
f186373f MF |
1126 | |
1127 | victim_name_len = btrfs_inode_extref_name_len(leaf, extref); | |
1128 | ||
1129 | if (btrfs_inode_extref_parent(leaf, extref) != parent_objectid) | |
1130 | goto next; | |
1131 | ||
1132 | victim_name = kmalloc(victim_name_len, GFP_NOFS); | |
3650860b JB |
1133 | if (!victim_name) |
1134 | return -ENOMEM; | |
f186373f MF |
1135 | read_extent_buffer(leaf, victim_name, (unsigned long)&extref->name, |
1136 | victim_name_len); | |
1137 | ||
1138 | search_key.objectid = inode_objectid; | |
1139 | search_key.type = BTRFS_INODE_EXTREF_KEY; | |
1140 | search_key.offset = btrfs_extref_hash(parent_objectid, | |
1141 | victim_name, | |
1142 | victim_name_len); | |
d3316c82 NB |
1143 | ret = backref_in_log(log_root, &search_key, |
1144 | parent_objectid, victim_name, | |
1145 | victim_name_len); | |
1146 | if (ret < 0) { | |
1147 | return ret; | |
1148 | } else if (!ret) { | |
f186373f MF |
1149 | ret = -ENOENT; |
1150 | victim_parent = read_one_inode(root, | |
94c91a1f | 1151 | parent_objectid); |
f186373f | 1152 | if (victim_parent) { |
94c91a1f | 1153 | inc_nlink(&inode->vfs_inode); |
f186373f MF |
1154 | btrfs_release_path(path); |
1155 | ||
1156 | ret = btrfs_unlink_inode(trans, root, | |
4ec5934e | 1157 | BTRFS_I(victim_parent), |
94c91a1f | 1158 | inode, |
4ec5934e NB |
1159 | victim_name, |
1160 | victim_name_len); | |
ada9af21 FDBM |
1161 | if (!ret) |
1162 | ret = btrfs_run_delayed_items( | |
e5c304e6 | 1163 | trans); |
f186373f | 1164 | } |
f186373f MF |
1165 | iput(victim_parent); |
1166 | kfree(victim_name); | |
3650860b JB |
1167 | if (ret) |
1168 | return ret; | |
f186373f MF |
1169 | *search_done = 1; |
1170 | goto again; | |
1171 | } | |
1172 | kfree(victim_name); | |
f186373f MF |
1173 | next: |
1174 | cur_offset += victim_name_len + sizeof(*extref); | |
1175 | } | |
1176 | *search_done = 1; | |
1177 | } | |
1178 | btrfs_release_path(path); | |
1179 | ||
34f3e4f2 | 1180 | /* look for a conflicting sequence number */ |
94c91a1f | 1181 | di = btrfs_lookup_dir_index_item(trans, root, path, btrfs_ino(dir), |
f186373f | 1182 | ref_index, name, namelen, 0); |
34f3e4f2 | 1183 | if (di && !IS_ERR(di)) { |
94c91a1f | 1184 | ret = drop_one_dir_item(trans, root, path, dir, di); |
3650860b JB |
1185 | if (ret) |
1186 | return ret; | |
34f3e4f2 | 1187 | } |
1188 | btrfs_release_path(path); | |
1189 | ||
52042d8e | 1190 | /* look for a conflicting name */ |
94c91a1f | 1191 | di = btrfs_lookup_dir_item(trans, root, path, btrfs_ino(dir), |
34f3e4f2 | 1192 | name, namelen, 0); |
1193 | if (di && !IS_ERR(di)) { | |
94c91a1f | 1194 | ret = drop_one_dir_item(trans, root, path, dir, di); |
3650860b JB |
1195 | if (ret) |
1196 | return ret; | |
34f3e4f2 | 1197 | } |
1198 | btrfs_release_path(path); | |
1199 | ||
5a1d7843 JS |
1200 | return 0; |
1201 | } | |
e02119d5 | 1202 | |
bae15d95 QW |
1203 | static int extref_get_fields(struct extent_buffer *eb, unsigned long ref_ptr, |
1204 | u32 *namelen, char **name, u64 *index, | |
1205 | u64 *parent_objectid) | |
f186373f MF |
1206 | { |
1207 | struct btrfs_inode_extref *extref; | |
1208 | ||
1209 | extref = (struct btrfs_inode_extref *)ref_ptr; | |
1210 | ||
1211 | *namelen = btrfs_inode_extref_name_len(eb, extref); | |
1212 | *name = kmalloc(*namelen, GFP_NOFS); | |
1213 | if (*name == NULL) | |
1214 | return -ENOMEM; | |
1215 | ||
1216 | read_extent_buffer(eb, *name, (unsigned long)&extref->name, | |
1217 | *namelen); | |
1218 | ||
1f250e92 FM |
1219 | if (index) |
1220 | *index = btrfs_inode_extref_index(eb, extref); | |
f186373f MF |
1221 | if (parent_objectid) |
1222 | *parent_objectid = btrfs_inode_extref_parent(eb, extref); | |
1223 | ||
1224 | return 0; | |
1225 | } | |
1226 | ||
bae15d95 QW |
1227 | static int ref_get_fields(struct extent_buffer *eb, unsigned long ref_ptr, |
1228 | u32 *namelen, char **name, u64 *index) | |
f186373f MF |
1229 | { |
1230 | struct btrfs_inode_ref *ref; | |
1231 | ||
1232 | ref = (struct btrfs_inode_ref *)ref_ptr; | |
1233 | ||
1234 | *namelen = btrfs_inode_ref_name_len(eb, ref); | |
1235 | *name = kmalloc(*namelen, GFP_NOFS); | |
1236 | if (*name == NULL) | |
1237 | return -ENOMEM; | |
1238 | ||
1239 | read_extent_buffer(eb, *name, (unsigned long)(ref + 1), *namelen); | |
1240 | ||
1f250e92 FM |
1241 | if (index) |
1242 | *index = btrfs_inode_ref_index(eb, ref); | |
f186373f MF |
1243 | |
1244 | return 0; | |
1245 | } | |
1246 | ||
1f250e92 FM |
1247 | /* |
1248 | * Take an inode reference item from the log tree and iterate all names from the | |
1249 | * inode reference item in the subvolume tree with the same key (if it exists). | |
1250 | * For any name that is not in the inode reference item from the log tree, do a | |
1251 | * proper unlink of that name (that is, remove its entry from the inode | |
1252 | * reference item and both dir index keys). | |
1253 | */ | |
1254 | static int unlink_old_inode_refs(struct btrfs_trans_handle *trans, | |
1255 | struct btrfs_root *root, | |
1256 | struct btrfs_path *path, | |
1257 | struct btrfs_inode *inode, | |
1258 | struct extent_buffer *log_eb, | |
1259 | int log_slot, | |
1260 | struct btrfs_key *key) | |
1261 | { | |
1262 | int ret; | |
1263 | unsigned long ref_ptr; | |
1264 | unsigned long ref_end; | |
1265 | struct extent_buffer *eb; | |
1266 | ||
1267 | again: | |
1268 | btrfs_release_path(path); | |
1269 | ret = btrfs_search_slot(NULL, root, key, path, 0, 0); | |
1270 | if (ret > 0) { | |
1271 | ret = 0; | |
1272 | goto out; | |
1273 | } | |
1274 | if (ret < 0) | |
1275 | goto out; | |
1276 | ||
1277 | eb = path->nodes[0]; | |
1278 | ref_ptr = btrfs_item_ptr_offset(eb, path->slots[0]); | |
1279 | ref_end = ref_ptr + btrfs_item_size_nr(eb, path->slots[0]); | |
1280 | while (ref_ptr < ref_end) { | |
1281 | char *name = NULL; | |
1282 | int namelen; | |
1283 | u64 parent_id; | |
1284 | ||
1285 | if (key->type == BTRFS_INODE_EXTREF_KEY) { | |
1286 | ret = extref_get_fields(eb, ref_ptr, &namelen, &name, | |
1287 | NULL, &parent_id); | |
1288 | } else { | |
1289 | parent_id = key->offset; | |
1290 | ret = ref_get_fields(eb, ref_ptr, &namelen, &name, | |
1291 | NULL); | |
1292 | } | |
1293 | if (ret) | |
1294 | goto out; | |
1295 | ||
1296 | if (key->type == BTRFS_INODE_EXTREF_KEY) | |
6ff49c6a NB |
1297 | ret = !!btrfs_find_name_in_ext_backref(log_eb, log_slot, |
1298 | parent_id, name, | |
1299 | namelen); | |
1f250e92 | 1300 | else |
9bb8407f NB |
1301 | ret = !!btrfs_find_name_in_backref(log_eb, log_slot, |
1302 | name, namelen); | |
1f250e92 FM |
1303 | |
1304 | if (!ret) { | |
1305 | struct inode *dir; | |
1306 | ||
1307 | btrfs_release_path(path); | |
1308 | dir = read_one_inode(root, parent_id); | |
1309 | if (!dir) { | |
1310 | ret = -ENOENT; | |
1311 | kfree(name); | |
1312 | goto out; | |
1313 | } | |
1314 | ret = btrfs_unlink_inode(trans, root, BTRFS_I(dir), | |
1315 | inode, name, namelen); | |
1316 | kfree(name); | |
1317 | iput(dir); | |
1318 | if (ret) | |
1319 | goto out; | |
1320 | goto again; | |
1321 | } | |
1322 | ||
1323 | kfree(name); | |
1324 | ref_ptr += namelen; | |
1325 | if (key->type == BTRFS_INODE_EXTREF_KEY) | |
1326 | ref_ptr += sizeof(struct btrfs_inode_extref); | |
1327 | else | |
1328 | ref_ptr += sizeof(struct btrfs_inode_ref); | |
1329 | } | |
1330 | ret = 0; | |
1331 | out: | |
1332 | btrfs_release_path(path); | |
1333 | return ret; | |
1334 | } | |
1335 | ||
0d836392 FM |
1336 | static int btrfs_inode_ref_exists(struct inode *inode, struct inode *dir, |
1337 | const u8 ref_type, const char *name, | |
1338 | const int namelen) | |
1339 | { | |
1340 | struct btrfs_key key; | |
1341 | struct btrfs_path *path; | |
1342 | const u64 parent_id = btrfs_ino(BTRFS_I(dir)); | |
1343 | int ret; | |
1344 | ||
1345 | path = btrfs_alloc_path(); | |
1346 | if (!path) | |
1347 | return -ENOMEM; | |
1348 | ||
1349 | key.objectid = btrfs_ino(BTRFS_I(inode)); | |
1350 | key.type = ref_type; | |
1351 | if (key.type == BTRFS_INODE_REF_KEY) | |
1352 | key.offset = parent_id; | |
1353 | else | |
1354 | key.offset = btrfs_extref_hash(parent_id, name, namelen); | |
1355 | ||
1356 | ret = btrfs_search_slot(NULL, BTRFS_I(inode)->root, &key, path, 0, 0); | |
1357 | if (ret < 0) | |
1358 | goto out; | |
1359 | if (ret > 0) { | |
1360 | ret = 0; | |
1361 | goto out; | |
1362 | } | |
1363 | if (key.type == BTRFS_INODE_EXTREF_KEY) | |
6ff49c6a NB |
1364 | ret = !!btrfs_find_name_in_ext_backref(path->nodes[0], |
1365 | path->slots[0], parent_id, name, namelen); | |
0d836392 | 1366 | else |
9bb8407f NB |
1367 | ret = !!btrfs_find_name_in_backref(path->nodes[0], path->slots[0], |
1368 | name, namelen); | |
0d836392 FM |
1369 | |
1370 | out: | |
1371 | btrfs_free_path(path); | |
1372 | return ret; | |
1373 | } | |
1374 | ||
6b5fc433 FM |
1375 | static int add_link(struct btrfs_trans_handle *trans, struct btrfs_root *root, |
1376 | struct inode *dir, struct inode *inode, const char *name, | |
1377 | int namelen, u64 ref_index) | |
1378 | { | |
1379 | struct btrfs_dir_item *dir_item; | |
1380 | struct btrfs_key key; | |
1381 | struct btrfs_path *path; | |
1382 | struct inode *other_inode = NULL; | |
1383 | int ret; | |
1384 | ||
1385 | path = btrfs_alloc_path(); | |
1386 | if (!path) | |
1387 | return -ENOMEM; | |
1388 | ||
1389 | dir_item = btrfs_lookup_dir_item(NULL, root, path, | |
1390 | btrfs_ino(BTRFS_I(dir)), | |
1391 | name, namelen, 0); | |
1392 | if (!dir_item) { | |
1393 | btrfs_release_path(path); | |
1394 | goto add_link; | |
1395 | } else if (IS_ERR(dir_item)) { | |
1396 | ret = PTR_ERR(dir_item); | |
1397 | goto out; | |
1398 | } | |
1399 | ||
1400 | /* | |
1401 | * Our inode's dentry collides with the dentry of another inode which is | |
1402 | * in the log but not yet processed since it has a higher inode number. | |
1403 | * So delete that other dentry. | |
1404 | */ | |
1405 | btrfs_dir_item_key_to_cpu(path->nodes[0], dir_item, &key); | |
1406 | btrfs_release_path(path); | |
1407 | other_inode = read_one_inode(root, key.objectid); | |
1408 | if (!other_inode) { | |
1409 | ret = -ENOENT; | |
1410 | goto out; | |
1411 | } | |
1412 | ret = btrfs_unlink_inode(trans, root, BTRFS_I(dir), BTRFS_I(other_inode), | |
1413 | name, namelen); | |
1414 | if (ret) | |
1415 | goto out; | |
1416 | /* | |
1417 | * If we dropped the link count to 0, bump it so that later the iput() | |
1418 | * on the inode will not free it. We will fixup the link count later. | |
1419 | */ | |
1420 | if (other_inode->i_nlink == 0) | |
1421 | inc_nlink(other_inode); | |
1422 | ||
1423 | ret = btrfs_run_delayed_items(trans); | |
1424 | if (ret) | |
1425 | goto out; | |
1426 | add_link: | |
1427 | ret = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode), | |
1428 | name, namelen, 0, ref_index); | |
1429 | out: | |
1430 | iput(other_inode); | |
1431 | btrfs_free_path(path); | |
1432 | ||
1433 | return ret; | |
1434 | } | |
1435 | ||
5a1d7843 JS |
1436 | /* |
1437 | * replay one inode back reference item found in the log tree. | |
1438 | * eb, slot and key refer to the buffer and key found in the log tree. | |
1439 | * root is the destination we are replaying into, and path is for temp | |
1440 | * use by this function. (it should be released on return). | |
1441 | */ | |
1442 | static noinline int add_inode_ref(struct btrfs_trans_handle *trans, | |
1443 | struct btrfs_root *root, | |
1444 | struct btrfs_root *log, | |
1445 | struct btrfs_path *path, | |
1446 | struct extent_buffer *eb, int slot, | |
1447 | struct btrfs_key *key) | |
1448 | { | |
03b2f08b GB |
1449 | struct inode *dir = NULL; |
1450 | struct inode *inode = NULL; | |
5a1d7843 JS |
1451 | unsigned long ref_ptr; |
1452 | unsigned long ref_end; | |
03b2f08b | 1453 | char *name = NULL; |
5a1d7843 JS |
1454 | int namelen; |
1455 | int ret; | |
1456 | int search_done = 0; | |
f186373f MF |
1457 | int log_ref_ver = 0; |
1458 | u64 parent_objectid; | |
1459 | u64 inode_objectid; | |
f46dbe3d | 1460 | u64 ref_index = 0; |
f186373f MF |
1461 | int ref_struct_size; |
1462 | ||
1463 | ref_ptr = btrfs_item_ptr_offset(eb, slot); | |
1464 | ref_end = ref_ptr + btrfs_item_size_nr(eb, slot); | |
1465 | ||
1466 | if (key->type == BTRFS_INODE_EXTREF_KEY) { | |
1467 | struct btrfs_inode_extref *r; | |
1468 | ||
1469 | ref_struct_size = sizeof(struct btrfs_inode_extref); | |
1470 | log_ref_ver = 1; | |
1471 | r = (struct btrfs_inode_extref *)ref_ptr; | |
1472 | parent_objectid = btrfs_inode_extref_parent(eb, r); | |
1473 | } else { | |
1474 | ref_struct_size = sizeof(struct btrfs_inode_ref); | |
1475 | parent_objectid = key->offset; | |
1476 | } | |
1477 | inode_objectid = key->objectid; | |
e02119d5 | 1478 | |
5a1d7843 JS |
1479 | /* |
1480 | * it is possible that we didn't log all the parent directories | |
1481 | * for a given inode. If we don't find the dir, just don't | |
1482 | * copy the back ref in. The link count fixup code will take | |
1483 | * care of the rest | |
1484 | */ | |
f186373f | 1485 | dir = read_one_inode(root, parent_objectid); |
03b2f08b GB |
1486 | if (!dir) { |
1487 | ret = -ENOENT; | |
1488 | goto out; | |
1489 | } | |
5a1d7843 | 1490 | |
f186373f | 1491 | inode = read_one_inode(root, inode_objectid); |
5a1d7843 | 1492 | if (!inode) { |
03b2f08b GB |
1493 | ret = -EIO; |
1494 | goto out; | |
5a1d7843 JS |
1495 | } |
1496 | ||
5a1d7843 | 1497 | while (ref_ptr < ref_end) { |
f186373f | 1498 | if (log_ref_ver) { |
bae15d95 QW |
1499 | ret = extref_get_fields(eb, ref_ptr, &namelen, &name, |
1500 | &ref_index, &parent_objectid); | |
f186373f MF |
1501 | /* |
1502 | * parent object can change from one array | |
1503 | * item to another. | |
1504 | */ | |
1505 | if (!dir) | |
1506 | dir = read_one_inode(root, parent_objectid); | |
03b2f08b GB |
1507 | if (!dir) { |
1508 | ret = -ENOENT; | |
1509 | goto out; | |
1510 | } | |
f186373f | 1511 | } else { |
bae15d95 QW |
1512 | ret = ref_get_fields(eb, ref_ptr, &namelen, &name, |
1513 | &ref_index); | |
f186373f MF |
1514 | } |
1515 | if (ret) | |
03b2f08b | 1516 | goto out; |
5a1d7843 JS |
1517 | |
1518 | /* if we already have a perfect match, we're done */ | |
f85b7379 DS |
1519 | if (!inode_in_dir(root, path, btrfs_ino(BTRFS_I(dir)), |
1520 | btrfs_ino(BTRFS_I(inode)), ref_index, | |
1521 | name, namelen)) { | |
5a1d7843 JS |
1522 | /* |
1523 | * look for a conflicting back reference in the | |
1524 | * metadata. if we find one we have to unlink that name | |
1525 | * of the file before we add our new link. Later on, we | |
1526 | * overwrite any existing back reference, and we don't | |
1527 | * want to create dangling pointers in the directory. | |
1528 | */ | |
1529 | ||
1530 | if (!search_done) { | |
1531 | ret = __add_inode_ref(trans, root, path, log, | |
94c91a1f | 1532 | BTRFS_I(dir), |
d75eefdf | 1533 | BTRFS_I(inode), |
f186373f MF |
1534 | inode_objectid, |
1535 | parent_objectid, | |
1536 | ref_index, name, namelen, | |
5a1d7843 | 1537 | &search_done); |
03b2f08b GB |
1538 | if (ret) { |
1539 | if (ret == 1) | |
1540 | ret = 0; | |
3650860b JB |
1541 | goto out; |
1542 | } | |
5a1d7843 JS |
1543 | } |
1544 | ||
0d836392 FM |
1545 | /* |
1546 | * If a reference item already exists for this inode | |
1547 | * with the same parent and name, but different index, | |
1548 | * drop it and the corresponding directory index entries | |
1549 | * from the parent before adding the new reference item | |
1550 | * and dir index entries, otherwise we would fail with | |
1551 | * -EEXIST returned from btrfs_add_link() below. | |
1552 | */ | |
1553 | ret = btrfs_inode_ref_exists(inode, dir, key->type, | |
1554 | name, namelen); | |
1555 | if (ret > 0) { | |
1556 | ret = btrfs_unlink_inode(trans, root, | |
1557 | BTRFS_I(dir), | |
1558 | BTRFS_I(inode), | |
1559 | name, namelen); | |
1560 | /* | |
1561 | * If we dropped the link count to 0, bump it so | |
1562 | * that later the iput() on the inode will not | |
1563 | * free it. We will fixup the link count later. | |
1564 | */ | |
1565 | if (!ret && inode->i_nlink == 0) | |
1566 | inc_nlink(inode); | |
1567 | } | |
1568 | if (ret < 0) | |
1569 | goto out; | |
1570 | ||
5a1d7843 | 1571 | /* insert our name */ |
6b5fc433 FM |
1572 | ret = add_link(trans, root, dir, inode, name, namelen, |
1573 | ref_index); | |
3650860b JB |
1574 | if (ret) |
1575 | goto out; | |
5a1d7843 | 1576 | |
f96d4474 JB |
1577 | ret = btrfs_update_inode(trans, root, BTRFS_I(inode)); |
1578 | if (ret) | |
1579 | goto out; | |
5a1d7843 JS |
1580 | } |
1581 | ||
f186373f | 1582 | ref_ptr = (unsigned long)(ref_ptr + ref_struct_size) + namelen; |
5a1d7843 | 1583 | kfree(name); |
03b2f08b | 1584 | name = NULL; |
f186373f MF |
1585 | if (log_ref_ver) { |
1586 | iput(dir); | |
1587 | dir = NULL; | |
1588 | } | |
5a1d7843 | 1589 | } |
e02119d5 | 1590 | |
1f250e92 FM |
1591 | /* |
1592 | * Before we overwrite the inode reference item in the subvolume tree | |
1593 | * with the item from the log tree, we must unlink all names from the | |
1594 | * parent directory that are in the subvolume's tree inode reference | |
1595 | * item, otherwise we end up with an inconsistent subvolume tree where | |
1596 | * dir index entries exist for a name but there is no inode reference | |
1597 | * item with the same name. | |
1598 | */ | |
1599 | ret = unlink_old_inode_refs(trans, root, path, BTRFS_I(inode), eb, slot, | |
1600 | key); | |
1601 | if (ret) | |
1602 | goto out; | |
1603 | ||
e02119d5 CM |
1604 | /* finally write the back reference in the inode */ |
1605 | ret = overwrite_item(trans, root, path, eb, slot, key); | |
5a1d7843 | 1606 | out: |
b3b4aa74 | 1607 | btrfs_release_path(path); |
03b2f08b | 1608 | kfree(name); |
e02119d5 CM |
1609 | iput(dir); |
1610 | iput(inode); | |
3650860b | 1611 | return ret; |
e02119d5 CM |
1612 | } |
1613 | ||
f186373f | 1614 | static int count_inode_extrefs(struct btrfs_root *root, |
36283658 | 1615 | struct btrfs_inode *inode, struct btrfs_path *path) |
f186373f MF |
1616 | { |
1617 | int ret = 0; | |
1618 | int name_len; | |
1619 | unsigned int nlink = 0; | |
1620 | u32 item_size; | |
1621 | u32 cur_offset = 0; | |
36283658 | 1622 | u64 inode_objectid = btrfs_ino(inode); |
f186373f MF |
1623 | u64 offset = 0; |
1624 | unsigned long ptr; | |
1625 | struct btrfs_inode_extref *extref; | |
1626 | struct extent_buffer *leaf; | |
1627 | ||
1628 | while (1) { | |
1629 | ret = btrfs_find_one_extref(root, inode_objectid, offset, path, | |
1630 | &extref, &offset); | |
1631 | if (ret) | |
1632 | break; | |
c71bf099 | 1633 | |
f186373f MF |
1634 | leaf = path->nodes[0]; |
1635 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1636 | ptr = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
2c2c452b | 1637 | cur_offset = 0; |
f186373f MF |
1638 | |
1639 | while (cur_offset < item_size) { | |
1640 | extref = (struct btrfs_inode_extref *) (ptr + cur_offset); | |
1641 | name_len = btrfs_inode_extref_name_len(leaf, extref); | |
1642 | ||
1643 | nlink++; | |
1644 | ||
1645 | cur_offset += name_len + sizeof(*extref); | |
1646 | } | |
1647 | ||
1648 | offset++; | |
1649 | btrfs_release_path(path); | |
1650 | } | |
1651 | btrfs_release_path(path); | |
1652 | ||
2c2c452b | 1653 | if (ret < 0 && ret != -ENOENT) |
f186373f MF |
1654 | return ret; |
1655 | return nlink; | |
1656 | } | |
1657 | ||
1658 | static int count_inode_refs(struct btrfs_root *root, | |
f329e319 | 1659 | struct btrfs_inode *inode, struct btrfs_path *path) |
e02119d5 | 1660 | { |
e02119d5 CM |
1661 | int ret; |
1662 | struct btrfs_key key; | |
f186373f | 1663 | unsigned int nlink = 0; |
e02119d5 CM |
1664 | unsigned long ptr; |
1665 | unsigned long ptr_end; | |
1666 | int name_len; | |
f329e319 | 1667 | u64 ino = btrfs_ino(inode); |
e02119d5 | 1668 | |
33345d01 | 1669 | key.objectid = ino; |
e02119d5 CM |
1670 | key.type = BTRFS_INODE_REF_KEY; |
1671 | key.offset = (u64)-1; | |
1672 | ||
d397712b | 1673 | while (1) { |
e02119d5 CM |
1674 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
1675 | if (ret < 0) | |
1676 | break; | |
1677 | if (ret > 0) { | |
1678 | if (path->slots[0] == 0) | |
1679 | break; | |
1680 | path->slots[0]--; | |
1681 | } | |
e93ae26f | 1682 | process_slot: |
e02119d5 CM |
1683 | btrfs_item_key_to_cpu(path->nodes[0], &key, |
1684 | path->slots[0]); | |
33345d01 | 1685 | if (key.objectid != ino || |
e02119d5 CM |
1686 | key.type != BTRFS_INODE_REF_KEY) |
1687 | break; | |
1688 | ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]); | |
1689 | ptr_end = ptr + btrfs_item_size_nr(path->nodes[0], | |
1690 | path->slots[0]); | |
d397712b | 1691 | while (ptr < ptr_end) { |
e02119d5 CM |
1692 | struct btrfs_inode_ref *ref; |
1693 | ||
1694 | ref = (struct btrfs_inode_ref *)ptr; | |
1695 | name_len = btrfs_inode_ref_name_len(path->nodes[0], | |
1696 | ref); | |
1697 | ptr = (unsigned long)(ref + 1) + name_len; | |
1698 | nlink++; | |
1699 | } | |
1700 | ||
1701 | if (key.offset == 0) | |
1702 | break; | |
e93ae26f FDBM |
1703 | if (path->slots[0] > 0) { |
1704 | path->slots[0]--; | |
1705 | goto process_slot; | |
1706 | } | |
e02119d5 | 1707 | key.offset--; |
b3b4aa74 | 1708 | btrfs_release_path(path); |
e02119d5 | 1709 | } |
b3b4aa74 | 1710 | btrfs_release_path(path); |
f186373f MF |
1711 | |
1712 | return nlink; | |
1713 | } | |
1714 | ||
1715 | /* | |
1716 | * There are a few corners where the link count of the file can't | |
1717 | * be properly maintained during replay. So, instead of adding | |
1718 | * lots of complexity to the log code, we just scan the backrefs | |
1719 | * for any file that has been through replay. | |
1720 | * | |
1721 | * The scan will update the link count on the inode to reflect the | |
1722 | * number of back refs found. If it goes down to zero, the iput | |
1723 | * will free the inode. | |
1724 | */ | |
1725 | static noinline int fixup_inode_link_count(struct btrfs_trans_handle *trans, | |
1726 | struct btrfs_root *root, | |
1727 | struct inode *inode) | |
1728 | { | |
1729 | struct btrfs_path *path; | |
1730 | int ret; | |
1731 | u64 nlink = 0; | |
4a0cc7ca | 1732 | u64 ino = btrfs_ino(BTRFS_I(inode)); |
f186373f MF |
1733 | |
1734 | path = btrfs_alloc_path(); | |
1735 | if (!path) | |
1736 | return -ENOMEM; | |
1737 | ||
f329e319 | 1738 | ret = count_inode_refs(root, BTRFS_I(inode), path); |
f186373f MF |
1739 | if (ret < 0) |
1740 | goto out; | |
1741 | ||
1742 | nlink = ret; | |
1743 | ||
36283658 | 1744 | ret = count_inode_extrefs(root, BTRFS_I(inode), path); |
f186373f MF |
1745 | if (ret < 0) |
1746 | goto out; | |
1747 | ||
1748 | nlink += ret; | |
1749 | ||
1750 | ret = 0; | |
1751 | ||
e02119d5 | 1752 | if (nlink != inode->i_nlink) { |
bfe86848 | 1753 | set_nlink(inode, nlink); |
f96d4474 JB |
1754 | ret = btrfs_update_inode(trans, root, BTRFS_I(inode)); |
1755 | if (ret) | |
1756 | goto out; | |
e02119d5 | 1757 | } |
8d5bf1cb | 1758 | BTRFS_I(inode)->index_cnt = (u64)-1; |
e02119d5 | 1759 | |
c71bf099 YZ |
1760 | if (inode->i_nlink == 0) { |
1761 | if (S_ISDIR(inode->i_mode)) { | |
1762 | ret = replay_dir_deletes(trans, root, NULL, path, | |
33345d01 | 1763 | ino, 1); |
3650860b JB |
1764 | if (ret) |
1765 | goto out; | |
c71bf099 | 1766 | } |
ecdcf3c2 NB |
1767 | ret = btrfs_insert_orphan_item(trans, root, ino); |
1768 | if (ret == -EEXIST) | |
1769 | ret = 0; | |
12fcfd22 | 1770 | } |
12fcfd22 | 1771 | |
f186373f MF |
1772 | out: |
1773 | btrfs_free_path(path); | |
1774 | return ret; | |
e02119d5 CM |
1775 | } |
1776 | ||
1777 | static noinline int fixup_inode_link_counts(struct btrfs_trans_handle *trans, | |
1778 | struct btrfs_root *root, | |
1779 | struct btrfs_path *path) | |
1780 | { | |
1781 | int ret; | |
1782 | struct btrfs_key key; | |
1783 | struct inode *inode; | |
1784 | ||
1785 | key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID; | |
1786 | key.type = BTRFS_ORPHAN_ITEM_KEY; | |
1787 | key.offset = (u64)-1; | |
d397712b | 1788 | while (1) { |
e02119d5 CM |
1789 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
1790 | if (ret < 0) | |
1791 | break; | |
1792 | ||
1793 | if (ret == 1) { | |
011b28ac | 1794 | ret = 0; |
e02119d5 CM |
1795 | if (path->slots[0] == 0) |
1796 | break; | |
1797 | path->slots[0]--; | |
1798 | } | |
1799 | ||
1800 | btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]); | |
1801 | if (key.objectid != BTRFS_TREE_LOG_FIXUP_OBJECTID || | |
1802 | key.type != BTRFS_ORPHAN_ITEM_KEY) | |
1803 | break; | |
1804 | ||
1805 | ret = btrfs_del_item(trans, root, path); | |
65a246c5 | 1806 | if (ret) |
011b28ac | 1807 | break; |
e02119d5 | 1808 | |
b3b4aa74 | 1809 | btrfs_release_path(path); |
e02119d5 | 1810 | inode = read_one_inode(root, key.offset); |
011b28ac JB |
1811 | if (!inode) { |
1812 | ret = -EIO; | |
1813 | break; | |
1814 | } | |
e02119d5 CM |
1815 | |
1816 | ret = fixup_inode_link_count(trans, root, inode); | |
e02119d5 | 1817 | iput(inode); |
3650860b | 1818 | if (ret) |
011b28ac | 1819 | break; |
e02119d5 | 1820 | |
12fcfd22 CM |
1821 | /* |
1822 | * fixup on a directory may create new entries, | |
1823 | * make sure we always look for the highset possible | |
1824 | * offset | |
1825 | */ | |
1826 | key.offset = (u64)-1; | |
e02119d5 | 1827 | } |
b3b4aa74 | 1828 | btrfs_release_path(path); |
65a246c5 | 1829 | return ret; |
e02119d5 CM |
1830 | } |
1831 | ||
1832 | ||
1833 | /* | |
1834 | * record a given inode in the fixup dir so we can check its link | |
1835 | * count when replay is done. The link count is incremented here | |
1836 | * so the inode won't go away until we check it | |
1837 | */ | |
1838 | static noinline int link_to_fixup_dir(struct btrfs_trans_handle *trans, | |
1839 | struct btrfs_root *root, | |
1840 | struct btrfs_path *path, | |
1841 | u64 objectid) | |
1842 | { | |
1843 | struct btrfs_key key; | |
1844 | int ret = 0; | |
1845 | struct inode *inode; | |
1846 | ||
1847 | inode = read_one_inode(root, objectid); | |
c00e9493 TI |
1848 | if (!inode) |
1849 | return -EIO; | |
e02119d5 CM |
1850 | |
1851 | key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID; | |
962a298f | 1852 | key.type = BTRFS_ORPHAN_ITEM_KEY; |
e02119d5 CM |
1853 | key.offset = objectid; |
1854 | ||
1855 | ret = btrfs_insert_empty_item(trans, root, path, &key, 0); | |
1856 | ||
b3b4aa74 | 1857 | btrfs_release_path(path); |
e02119d5 | 1858 | if (ret == 0) { |
9bf7a489 JB |
1859 | if (!inode->i_nlink) |
1860 | set_nlink(inode, 1); | |
1861 | else | |
8b558c5f | 1862 | inc_nlink(inode); |
9a56fcd1 | 1863 | ret = btrfs_update_inode(trans, root, BTRFS_I(inode)); |
e02119d5 CM |
1864 | } else if (ret == -EEXIST) { |
1865 | ret = 0; | |
e02119d5 CM |
1866 | } |
1867 | iput(inode); | |
1868 | ||
1869 | return ret; | |
1870 | } | |
1871 | ||
1872 | /* | |
1873 | * when replaying the log for a directory, we only insert names | |
1874 | * for inodes that actually exist. This means an fsync on a directory | |
1875 | * does not implicitly fsync all the new files in it | |
1876 | */ | |
1877 | static noinline int insert_one_name(struct btrfs_trans_handle *trans, | |
1878 | struct btrfs_root *root, | |
e02119d5 | 1879 | u64 dirid, u64 index, |
60d53eb3 | 1880 | char *name, int name_len, |
e02119d5 CM |
1881 | struct btrfs_key *location) |
1882 | { | |
1883 | struct inode *inode; | |
1884 | struct inode *dir; | |
1885 | int ret; | |
1886 | ||
1887 | inode = read_one_inode(root, location->objectid); | |
1888 | if (!inode) | |
1889 | return -ENOENT; | |
1890 | ||
1891 | dir = read_one_inode(root, dirid); | |
1892 | if (!dir) { | |
1893 | iput(inode); | |
1894 | return -EIO; | |
1895 | } | |
d555438b | 1896 | |
db0a669f NB |
1897 | ret = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode), name, |
1898 | name_len, 1, index); | |
e02119d5 CM |
1899 | |
1900 | /* FIXME, put inode into FIXUP list */ | |
1901 | ||
1902 | iput(inode); | |
1903 | iput(dir); | |
1904 | return ret; | |
1905 | } | |
1906 | ||
1907 | /* | |
1908 | * take a single entry in a log directory item and replay it into | |
1909 | * the subvolume. | |
1910 | * | |
1911 | * if a conflicting item exists in the subdirectory already, | |
1912 | * the inode it points to is unlinked and put into the link count | |
1913 | * fix up tree. | |
1914 | * | |
1915 | * If a name from the log points to a file or directory that does | |
1916 | * not exist in the FS, it is skipped. fsyncs on directories | |
1917 | * do not force down inodes inside that directory, just changes to the | |
1918 | * names or unlinks in a directory. | |
bb53eda9 FM |
1919 | * |
1920 | * Returns < 0 on error, 0 if the name wasn't replayed (dentry points to a | |
1921 | * non-existing inode) and 1 if the name was replayed. | |
e02119d5 CM |
1922 | */ |
1923 | static noinline int replay_one_name(struct btrfs_trans_handle *trans, | |
1924 | struct btrfs_root *root, | |
1925 | struct btrfs_path *path, | |
1926 | struct extent_buffer *eb, | |
1927 | struct btrfs_dir_item *di, | |
1928 | struct btrfs_key *key) | |
1929 | { | |
1930 | char *name; | |
1931 | int name_len; | |
1932 | struct btrfs_dir_item *dst_di; | |
1933 | struct btrfs_key found_key; | |
1934 | struct btrfs_key log_key; | |
1935 | struct inode *dir; | |
e02119d5 | 1936 | u8 log_type; |
4bef0848 | 1937 | int exists; |
3650860b | 1938 | int ret = 0; |
d555438b | 1939 | bool update_size = (key->type == BTRFS_DIR_INDEX_KEY); |
bb53eda9 | 1940 | bool name_added = false; |
e02119d5 CM |
1941 | |
1942 | dir = read_one_inode(root, key->objectid); | |
c00e9493 TI |
1943 | if (!dir) |
1944 | return -EIO; | |
e02119d5 CM |
1945 | |
1946 | name_len = btrfs_dir_name_len(eb, di); | |
1947 | name = kmalloc(name_len, GFP_NOFS); | |
2bac325e FDBM |
1948 | if (!name) { |
1949 | ret = -ENOMEM; | |
1950 | goto out; | |
1951 | } | |
2a29edc6 | 1952 | |
e02119d5 CM |
1953 | log_type = btrfs_dir_type(eb, di); |
1954 | read_extent_buffer(eb, name, (unsigned long)(di + 1), | |
1955 | name_len); | |
1956 | ||
1957 | btrfs_dir_item_key_to_cpu(eb, di, &log_key); | |
4bef0848 CM |
1958 | exists = btrfs_lookup_inode(trans, root, path, &log_key, 0); |
1959 | if (exists == 0) | |
1960 | exists = 1; | |
1961 | else | |
1962 | exists = 0; | |
b3b4aa74 | 1963 | btrfs_release_path(path); |
4bef0848 | 1964 | |
e02119d5 CM |
1965 | if (key->type == BTRFS_DIR_ITEM_KEY) { |
1966 | dst_di = btrfs_lookup_dir_item(trans, root, path, key->objectid, | |
1967 | name, name_len, 1); | |
d397712b | 1968 | } else if (key->type == BTRFS_DIR_INDEX_KEY) { |
e02119d5 CM |
1969 | dst_di = btrfs_lookup_dir_index_item(trans, root, path, |
1970 | key->objectid, | |
1971 | key->offset, name, | |
1972 | name_len, 1); | |
1973 | } else { | |
3650860b JB |
1974 | /* Corruption */ |
1975 | ret = -EINVAL; | |
1976 | goto out; | |
e02119d5 | 1977 | } |
c704005d | 1978 | if (IS_ERR_OR_NULL(dst_di)) { |
e02119d5 CM |
1979 | /* we need a sequence number to insert, so we only |
1980 | * do inserts for the BTRFS_DIR_INDEX_KEY types | |
1981 | */ | |
1982 | if (key->type != BTRFS_DIR_INDEX_KEY) | |
1983 | goto out; | |
1984 | goto insert; | |
1985 | } | |
1986 | ||
1987 | btrfs_dir_item_key_to_cpu(path->nodes[0], dst_di, &found_key); | |
1988 | /* the existing item matches the logged item */ | |
1989 | if (found_key.objectid == log_key.objectid && | |
1990 | found_key.type == log_key.type && | |
1991 | found_key.offset == log_key.offset && | |
1992 | btrfs_dir_type(path->nodes[0], dst_di) == log_type) { | |
a2cc11db | 1993 | update_size = false; |
e02119d5 CM |
1994 | goto out; |
1995 | } | |
1996 | ||
1997 | /* | |
1998 | * don't drop the conflicting directory entry if the inode | |
1999 | * for the new entry doesn't exist | |
2000 | */ | |
4bef0848 | 2001 | if (!exists) |
e02119d5 CM |
2002 | goto out; |
2003 | ||
207e7d92 | 2004 | ret = drop_one_dir_item(trans, root, path, BTRFS_I(dir), dst_di); |
3650860b JB |
2005 | if (ret) |
2006 | goto out; | |
e02119d5 CM |
2007 | |
2008 | if (key->type == BTRFS_DIR_INDEX_KEY) | |
2009 | goto insert; | |
2010 | out: | |
b3b4aa74 | 2011 | btrfs_release_path(path); |
d555438b | 2012 | if (!ret && update_size) { |
6ef06d27 | 2013 | btrfs_i_size_write(BTRFS_I(dir), dir->i_size + name_len * 2); |
9a56fcd1 | 2014 | ret = btrfs_update_inode(trans, root, BTRFS_I(dir)); |
d555438b | 2015 | } |
e02119d5 CM |
2016 | kfree(name); |
2017 | iput(dir); | |
bb53eda9 FM |
2018 | if (!ret && name_added) |
2019 | ret = 1; | |
3650860b | 2020 | return ret; |
e02119d5 CM |
2021 | |
2022 | insert: | |
725af92a NB |
2023 | /* |
2024 | * Check if the inode reference exists in the log for the given name, | |
2025 | * inode and parent inode | |
2026 | */ | |
2027 | found_key.objectid = log_key.objectid; | |
2028 | found_key.type = BTRFS_INODE_REF_KEY; | |
2029 | found_key.offset = key->objectid; | |
2030 | ret = backref_in_log(root->log_root, &found_key, 0, name, name_len); | |
2031 | if (ret < 0) { | |
2032 | goto out; | |
2033 | } else if (ret) { | |
2034 | /* The dentry will be added later. */ | |
2035 | ret = 0; | |
2036 | update_size = false; | |
2037 | goto out; | |
2038 | } | |
2039 | ||
2040 | found_key.objectid = log_key.objectid; | |
2041 | found_key.type = BTRFS_INODE_EXTREF_KEY; | |
2042 | found_key.offset = key->objectid; | |
2043 | ret = backref_in_log(root->log_root, &found_key, key->objectid, name, | |
2044 | name_len); | |
2045 | if (ret < 0) { | |
2046 | goto out; | |
2047 | } else if (ret) { | |
df8d116f FM |
2048 | /* The dentry will be added later. */ |
2049 | ret = 0; | |
2050 | update_size = false; | |
2051 | goto out; | |
2052 | } | |
b3b4aa74 | 2053 | btrfs_release_path(path); |
60d53eb3 Z |
2054 | ret = insert_one_name(trans, root, key->objectid, key->offset, |
2055 | name, name_len, &log_key); | |
df8d116f | 2056 | if (ret && ret != -ENOENT && ret != -EEXIST) |
3650860b | 2057 | goto out; |
bb53eda9 FM |
2058 | if (!ret) |
2059 | name_added = true; | |
d555438b | 2060 | update_size = false; |
3650860b | 2061 | ret = 0; |
e02119d5 CM |
2062 | goto out; |
2063 | } | |
2064 | ||
2065 | /* | |
2066 | * find all the names in a directory item and reconcile them into | |
2067 | * the subvolume. Only BTRFS_DIR_ITEM_KEY types will have more than | |
2068 | * one name in a directory item, but the same code gets used for | |
2069 | * both directory index types | |
2070 | */ | |
2071 | static noinline int replay_one_dir_item(struct btrfs_trans_handle *trans, | |
2072 | struct btrfs_root *root, | |
2073 | struct btrfs_path *path, | |
2074 | struct extent_buffer *eb, int slot, | |
2075 | struct btrfs_key *key) | |
2076 | { | |
bb53eda9 | 2077 | int ret = 0; |
e02119d5 CM |
2078 | u32 item_size = btrfs_item_size_nr(eb, slot); |
2079 | struct btrfs_dir_item *di; | |
2080 | int name_len; | |
2081 | unsigned long ptr; | |
2082 | unsigned long ptr_end; | |
bb53eda9 | 2083 | struct btrfs_path *fixup_path = NULL; |
e02119d5 CM |
2084 | |
2085 | ptr = btrfs_item_ptr_offset(eb, slot); | |
2086 | ptr_end = ptr + item_size; | |
d397712b | 2087 | while (ptr < ptr_end) { |
e02119d5 CM |
2088 | di = (struct btrfs_dir_item *)ptr; |
2089 | name_len = btrfs_dir_name_len(eb, di); | |
2090 | ret = replay_one_name(trans, root, path, eb, di, key); | |
bb53eda9 FM |
2091 | if (ret < 0) |
2092 | break; | |
e02119d5 CM |
2093 | ptr = (unsigned long)(di + 1); |
2094 | ptr += name_len; | |
bb53eda9 FM |
2095 | |
2096 | /* | |
2097 | * If this entry refers to a non-directory (directories can not | |
2098 | * have a link count > 1) and it was added in the transaction | |
2099 | * that was not committed, make sure we fixup the link count of | |
2100 | * the inode it the entry points to. Otherwise something like | |
2101 | * the following would result in a directory pointing to an | |
2102 | * inode with a wrong link that does not account for this dir | |
2103 | * entry: | |
2104 | * | |
2105 | * mkdir testdir | |
2106 | * touch testdir/foo | |
2107 | * touch testdir/bar | |
2108 | * sync | |
2109 | * | |
2110 | * ln testdir/bar testdir/bar_link | |
2111 | * ln testdir/foo testdir/foo_link | |
2112 | * xfs_io -c "fsync" testdir/bar | |
2113 | * | |
2114 | * <power failure> | |
2115 | * | |
2116 | * mount fs, log replay happens | |
2117 | * | |
2118 | * File foo would remain with a link count of 1 when it has two | |
2119 | * entries pointing to it in the directory testdir. This would | |
2120 | * make it impossible to ever delete the parent directory has | |
2121 | * it would result in stale dentries that can never be deleted. | |
2122 | */ | |
2123 | if (ret == 1 && btrfs_dir_type(eb, di) != BTRFS_FT_DIR) { | |
2124 | struct btrfs_key di_key; | |
2125 | ||
2126 | if (!fixup_path) { | |
2127 | fixup_path = btrfs_alloc_path(); | |
2128 | if (!fixup_path) { | |
2129 | ret = -ENOMEM; | |
2130 | break; | |
2131 | } | |
2132 | } | |
2133 | ||
2134 | btrfs_dir_item_key_to_cpu(eb, di, &di_key); | |
2135 | ret = link_to_fixup_dir(trans, root, fixup_path, | |
2136 | di_key.objectid); | |
2137 | if (ret) | |
2138 | break; | |
2139 | } | |
2140 | ret = 0; | |
e02119d5 | 2141 | } |
bb53eda9 FM |
2142 | btrfs_free_path(fixup_path); |
2143 | return ret; | |
e02119d5 CM |
2144 | } |
2145 | ||
2146 | /* | |
2147 | * directory replay has two parts. There are the standard directory | |
2148 | * items in the log copied from the subvolume, and range items | |
2149 | * created in the log while the subvolume was logged. | |
2150 | * | |
2151 | * The range items tell us which parts of the key space the log | |
2152 | * is authoritative for. During replay, if a key in the subvolume | |
2153 | * directory is in a logged range item, but not actually in the log | |
2154 | * that means it was deleted from the directory before the fsync | |
2155 | * and should be removed. | |
2156 | */ | |
2157 | static noinline int find_dir_range(struct btrfs_root *root, | |
2158 | struct btrfs_path *path, | |
2159 | u64 dirid, int key_type, | |
2160 | u64 *start_ret, u64 *end_ret) | |
2161 | { | |
2162 | struct btrfs_key key; | |
2163 | u64 found_end; | |
2164 | struct btrfs_dir_log_item *item; | |
2165 | int ret; | |
2166 | int nritems; | |
2167 | ||
2168 | if (*start_ret == (u64)-1) | |
2169 | return 1; | |
2170 | ||
2171 | key.objectid = dirid; | |
2172 | key.type = key_type; | |
2173 | key.offset = *start_ret; | |
2174 | ||
2175 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
2176 | if (ret < 0) | |
2177 | goto out; | |
2178 | if (ret > 0) { | |
2179 | if (path->slots[0] == 0) | |
2180 | goto out; | |
2181 | path->slots[0]--; | |
2182 | } | |
2183 | if (ret != 0) | |
2184 | btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]); | |
2185 | ||
2186 | if (key.type != key_type || key.objectid != dirid) { | |
2187 | ret = 1; | |
2188 | goto next; | |
2189 | } | |
2190 | item = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
2191 | struct btrfs_dir_log_item); | |
2192 | found_end = btrfs_dir_log_end(path->nodes[0], item); | |
2193 | ||
2194 | if (*start_ret >= key.offset && *start_ret <= found_end) { | |
2195 | ret = 0; | |
2196 | *start_ret = key.offset; | |
2197 | *end_ret = found_end; | |
2198 | goto out; | |
2199 | } | |
2200 | ret = 1; | |
2201 | next: | |
2202 | /* check the next slot in the tree to see if it is a valid item */ | |
2203 | nritems = btrfs_header_nritems(path->nodes[0]); | |
2a7bf53f | 2204 | path->slots[0]++; |
e02119d5 CM |
2205 | if (path->slots[0] >= nritems) { |
2206 | ret = btrfs_next_leaf(root, path); | |
2207 | if (ret) | |
2208 | goto out; | |
e02119d5 CM |
2209 | } |
2210 | ||
2211 | btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]); | |
2212 | ||
2213 | if (key.type != key_type || key.objectid != dirid) { | |
2214 | ret = 1; | |
2215 | goto out; | |
2216 | } | |
2217 | item = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
2218 | struct btrfs_dir_log_item); | |
2219 | found_end = btrfs_dir_log_end(path->nodes[0], item); | |
2220 | *start_ret = key.offset; | |
2221 | *end_ret = found_end; | |
2222 | ret = 0; | |
2223 | out: | |
b3b4aa74 | 2224 | btrfs_release_path(path); |
e02119d5 CM |
2225 | return ret; |
2226 | } | |
2227 | ||
2228 | /* | |
2229 | * this looks for a given directory item in the log. If the directory | |
2230 | * item is not in the log, the item is removed and the inode it points | |
2231 | * to is unlinked | |
2232 | */ | |
2233 | static noinline int check_item_in_log(struct btrfs_trans_handle *trans, | |
2234 | struct btrfs_root *root, | |
2235 | struct btrfs_root *log, | |
2236 | struct btrfs_path *path, | |
2237 | struct btrfs_path *log_path, | |
2238 | struct inode *dir, | |
2239 | struct btrfs_key *dir_key) | |
2240 | { | |
2241 | int ret; | |
2242 | struct extent_buffer *eb; | |
2243 | int slot; | |
2244 | u32 item_size; | |
2245 | struct btrfs_dir_item *di; | |
2246 | struct btrfs_dir_item *log_di; | |
2247 | int name_len; | |
2248 | unsigned long ptr; | |
2249 | unsigned long ptr_end; | |
2250 | char *name; | |
2251 | struct inode *inode; | |
2252 | struct btrfs_key location; | |
2253 | ||
2254 | again: | |
2255 | eb = path->nodes[0]; | |
2256 | slot = path->slots[0]; | |
2257 | item_size = btrfs_item_size_nr(eb, slot); | |
2258 | ptr = btrfs_item_ptr_offset(eb, slot); | |
2259 | ptr_end = ptr + item_size; | |
d397712b | 2260 | while (ptr < ptr_end) { |
e02119d5 CM |
2261 | di = (struct btrfs_dir_item *)ptr; |
2262 | name_len = btrfs_dir_name_len(eb, di); | |
2263 | name = kmalloc(name_len, GFP_NOFS); | |
2264 | if (!name) { | |
2265 | ret = -ENOMEM; | |
2266 | goto out; | |
2267 | } | |
2268 | read_extent_buffer(eb, name, (unsigned long)(di + 1), | |
2269 | name_len); | |
2270 | log_di = NULL; | |
12fcfd22 | 2271 | if (log && dir_key->type == BTRFS_DIR_ITEM_KEY) { |
e02119d5 CM |
2272 | log_di = btrfs_lookup_dir_item(trans, log, log_path, |
2273 | dir_key->objectid, | |
2274 | name, name_len, 0); | |
12fcfd22 | 2275 | } else if (log && dir_key->type == BTRFS_DIR_INDEX_KEY) { |
e02119d5 CM |
2276 | log_di = btrfs_lookup_dir_index_item(trans, log, |
2277 | log_path, | |
2278 | dir_key->objectid, | |
2279 | dir_key->offset, | |
2280 | name, name_len, 0); | |
2281 | } | |
8d9e220c | 2282 | if (!log_di || log_di == ERR_PTR(-ENOENT)) { |
e02119d5 | 2283 | btrfs_dir_item_key_to_cpu(eb, di, &location); |
b3b4aa74 DS |
2284 | btrfs_release_path(path); |
2285 | btrfs_release_path(log_path); | |
e02119d5 | 2286 | inode = read_one_inode(root, location.objectid); |
c00e9493 TI |
2287 | if (!inode) { |
2288 | kfree(name); | |
2289 | return -EIO; | |
2290 | } | |
e02119d5 CM |
2291 | |
2292 | ret = link_to_fixup_dir(trans, root, | |
2293 | path, location.objectid); | |
3650860b JB |
2294 | if (ret) { |
2295 | kfree(name); | |
2296 | iput(inode); | |
2297 | goto out; | |
2298 | } | |
2299 | ||
8b558c5f | 2300 | inc_nlink(inode); |
4ec5934e NB |
2301 | ret = btrfs_unlink_inode(trans, root, BTRFS_I(dir), |
2302 | BTRFS_I(inode), name, name_len); | |
3650860b | 2303 | if (!ret) |
e5c304e6 | 2304 | ret = btrfs_run_delayed_items(trans); |
e02119d5 CM |
2305 | kfree(name); |
2306 | iput(inode); | |
3650860b JB |
2307 | if (ret) |
2308 | goto out; | |
e02119d5 CM |
2309 | |
2310 | /* there might still be more names under this key | |
2311 | * check and repeat if required | |
2312 | */ | |
2313 | ret = btrfs_search_slot(NULL, root, dir_key, path, | |
2314 | 0, 0); | |
2315 | if (ret == 0) | |
2316 | goto again; | |
2317 | ret = 0; | |
2318 | goto out; | |
269d040f FDBM |
2319 | } else if (IS_ERR(log_di)) { |
2320 | kfree(name); | |
2321 | return PTR_ERR(log_di); | |
e02119d5 | 2322 | } |
b3b4aa74 | 2323 | btrfs_release_path(log_path); |
e02119d5 CM |
2324 | kfree(name); |
2325 | ||
2326 | ptr = (unsigned long)(di + 1); | |
2327 | ptr += name_len; | |
2328 | } | |
2329 | ret = 0; | |
2330 | out: | |
b3b4aa74 DS |
2331 | btrfs_release_path(path); |
2332 | btrfs_release_path(log_path); | |
e02119d5 CM |
2333 | return ret; |
2334 | } | |
2335 | ||
4f764e51 FM |
2336 | static int replay_xattr_deletes(struct btrfs_trans_handle *trans, |
2337 | struct btrfs_root *root, | |
2338 | struct btrfs_root *log, | |
2339 | struct btrfs_path *path, | |
2340 | const u64 ino) | |
2341 | { | |
2342 | struct btrfs_key search_key; | |
2343 | struct btrfs_path *log_path; | |
2344 | int i; | |
2345 | int nritems; | |
2346 | int ret; | |
2347 | ||
2348 | log_path = btrfs_alloc_path(); | |
2349 | if (!log_path) | |
2350 | return -ENOMEM; | |
2351 | ||
2352 | search_key.objectid = ino; | |
2353 | search_key.type = BTRFS_XATTR_ITEM_KEY; | |
2354 | search_key.offset = 0; | |
2355 | again: | |
2356 | ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0); | |
2357 | if (ret < 0) | |
2358 | goto out; | |
2359 | process_leaf: | |
2360 | nritems = btrfs_header_nritems(path->nodes[0]); | |
2361 | for (i = path->slots[0]; i < nritems; i++) { | |
2362 | struct btrfs_key key; | |
2363 | struct btrfs_dir_item *di; | |
2364 | struct btrfs_dir_item *log_di; | |
2365 | u32 total_size; | |
2366 | u32 cur; | |
2367 | ||
2368 | btrfs_item_key_to_cpu(path->nodes[0], &key, i); | |
2369 | if (key.objectid != ino || key.type != BTRFS_XATTR_ITEM_KEY) { | |
2370 | ret = 0; | |
2371 | goto out; | |
2372 | } | |
2373 | ||
2374 | di = btrfs_item_ptr(path->nodes[0], i, struct btrfs_dir_item); | |
2375 | total_size = btrfs_item_size_nr(path->nodes[0], i); | |
2376 | cur = 0; | |
2377 | while (cur < total_size) { | |
2378 | u16 name_len = btrfs_dir_name_len(path->nodes[0], di); | |
2379 | u16 data_len = btrfs_dir_data_len(path->nodes[0], di); | |
2380 | u32 this_len = sizeof(*di) + name_len + data_len; | |
2381 | char *name; | |
2382 | ||
2383 | name = kmalloc(name_len, GFP_NOFS); | |
2384 | if (!name) { | |
2385 | ret = -ENOMEM; | |
2386 | goto out; | |
2387 | } | |
2388 | read_extent_buffer(path->nodes[0], name, | |
2389 | (unsigned long)(di + 1), name_len); | |
2390 | ||
2391 | log_di = btrfs_lookup_xattr(NULL, log, log_path, ino, | |
2392 | name, name_len, 0); | |
2393 | btrfs_release_path(log_path); | |
2394 | if (!log_di) { | |
2395 | /* Doesn't exist in log tree, so delete it. */ | |
2396 | btrfs_release_path(path); | |
2397 | di = btrfs_lookup_xattr(trans, root, path, ino, | |
2398 | name, name_len, -1); | |
2399 | kfree(name); | |
2400 | if (IS_ERR(di)) { | |
2401 | ret = PTR_ERR(di); | |
2402 | goto out; | |
2403 | } | |
2404 | ASSERT(di); | |
2405 | ret = btrfs_delete_one_dir_name(trans, root, | |
2406 | path, di); | |
2407 | if (ret) | |
2408 | goto out; | |
2409 | btrfs_release_path(path); | |
2410 | search_key = key; | |
2411 | goto again; | |
2412 | } | |
2413 | kfree(name); | |
2414 | if (IS_ERR(log_di)) { | |
2415 | ret = PTR_ERR(log_di); | |
2416 | goto out; | |
2417 | } | |
2418 | cur += this_len; | |
2419 | di = (struct btrfs_dir_item *)((char *)di + this_len); | |
2420 | } | |
2421 | } | |
2422 | ret = btrfs_next_leaf(root, path); | |
2423 | if (ret > 0) | |
2424 | ret = 0; | |
2425 | else if (ret == 0) | |
2426 | goto process_leaf; | |
2427 | out: | |
2428 | btrfs_free_path(log_path); | |
2429 | btrfs_release_path(path); | |
2430 | return ret; | |
2431 | } | |
2432 | ||
2433 | ||
e02119d5 CM |
2434 | /* |
2435 | * deletion replay happens before we copy any new directory items | |
2436 | * out of the log or out of backreferences from inodes. It | |
2437 | * scans the log to find ranges of keys that log is authoritative for, | |
2438 | * and then scans the directory to find items in those ranges that are | |
2439 | * not present in the log. | |
2440 | * | |
2441 | * Anything we don't find in the log is unlinked and removed from the | |
2442 | * directory. | |
2443 | */ | |
2444 | static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans, | |
2445 | struct btrfs_root *root, | |
2446 | struct btrfs_root *log, | |
2447 | struct btrfs_path *path, | |
12fcfd22 | 2448 | u64 dirid, int del_all) |
e02119d5 CM |
2449 | { |
2450 | u64 range_start; | |
2451 | u64 range_end; | |
2452 | int key_type = BTRFS_DIR_LOG_ITEM_KEY; | |
2453 | int ret = 0; | |
2454 | struct btrfs_key dir_key; | |
2455 | struct btrfs_key found_key; | |
2456 | struct btrfs_path *log_path; | |
2457 | struct inode *dir; | |
2458 | ||
2459 | dir_key.objectid = dirid; | |
2460 | dir_key.type = BTRFS_DIR_ITEM_KEY; | |
2461 | log_path = btrfs_alloc_path(); | |
2462 | if (!log_path) | |
2463 | return -ENOMEM; | |
2464 | ||
2465 | dir = read_one_inode(root, dirid); | |
2466 | /* it isn't an error if the inode isn't there, that can happen | |
2467 | * because we replay the deletes before we copy in the inode item | |
2468 | * from the log | |
2469 | */ | |
2470 | if (!dir) { | |
2471 | btrfs_free_path(log_path); | |
2472 | return 0; | |
2473 | } | |
2474 | again: | |
2475 | range_start = 0; | |
2476 | range_end = 0; | |
d397712b | 2477 | while (1) { |
12fcfd22 CM |
2478 | if (del_all) |
2479 | range_end = (u64)-1; | |
2480 | else { | |
2481 | ret = find_dir_range(log, path, dirid, key_type, | |
2482 | &range_start, &range_end); | |
2483 | if (ret != 0) | |
2484 | break; | |
2485 | } | |
e02119d5 CM |
2486 | |
2487 | dir_key.offset = range_start; | |
d397712b | 2488 | while (1) { |
e02119d5 CM |
2489 | int nritems; |
2490 | ret = btrfs_search_slot(NULL, root, &dir_key, path, | |
2491 | 0, 0); | |
2492 | if (ret < 0) | |
2493 | goto out; | |
2494 | ||
2495 | nritems = btrfs_header_nritems(path->nodes[0]); | |
2496 | if (path->slots[0] >= nritems) { | |
2497 | ret = btrfs_next_leaf(root, path); | |
b98def7c | 2498 | if (ret == 1) |
e02119d5 | 2499 | break; |
b98def7c LB |
2500 | else if (ret < 0) |
2501 | goto out; | |
e02119d5 CM |
2502 | } |
2503 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, | |
2504 | path->slots[0]); | |
2505 | if (found_key.objectid != dirid || | |
2506 | found_key.type != dir_key.type) | |
2507 | goto next_type; | |
2508 | ||
2509 | if (found_key.offset > range_end) | |
2510 | break; | |
2511 | ||
2512 | ret = check_item_in_log(trans, root, log, path, | |
12fcfd22 CM |
2513 | log_path, dir, |
2514 | &found_key); | |
3650860b JB |
2515 | if (ret) |
2516 | goto out; | |
e02119d5 CM |
2517 | if (found_key.offset == (u64)-1) |
2518 | break; | |
2519 | dir_key.offset = found_key.offset + 1; | |
2520 | } | |
b3b4aa74 | 2521 | btrfs_release_path(path); |
e02119d5 CM |
2522 | if (range_end == (u64)-1) |
2523 | break; | |
2524 | range_start = range_end + 1; | |
2525 | } | |
2526 | ||
2527 | next_type: | |
2528 | ret = 0; | |
2529 | if (key_type == BTRFS_DIR_LOG_ITEM_KEY) { | |
2530 | key_type = BTRFS_DIR_LOG_INDEX_KEY; | |
2531 | dir_key.type = BTRFS_DIR_INDEX_KEY; | |
b3b4aa74 | 2532 | btrfs_release_path(path); |
e02119d5 CM |
2533 | goto again; |
2534 | } | |
2535 | out: | |
b3b4aa74 | 2536 | btrfs_release_path(path); |
e02119d5 CM |
2537 | btrfs_free_path(log_path); |
2538 | iput(dir); | |
2539 | return ret; | |
2540 | } | |
2541 | ||
2542 | /* | |
2543 | * the process_func used to replay items from the log tree. This | |
2544 | * gets called in two different stages. The first stage just looks | |
2545 | * for inodes and makes sure they are all copied into the subvolume. | |
2546 | * | |
2547 | * The second stage copies all the other item types from the log into | |
2548 | * the subvolume. The two stage approach is slower, but gets rid of | |
2549 | * lots of complexity around inodes referencing other inodes that exist | |
2550 | * only in the log (references come from either directory items or inode | |
2551 | * back refs). | |
2552 | */ | |
2553 | static int replay_one_buffer(struct btrfs_root *log, struct extent_buffer *eb, | |
581c1760 | 2554 | struct walk_control *wc, u64 gen, int level) |
e02119d5 CM |
2555 | { |
2556 | int nritems; | |
2557 | struct btrfs_path *path; | |
2558 | struct btrfs_root *root = wc->replay_dest; | |
2559 | struct btrfs_key key; | |
e02119d5 CM |
2560 | int i; |
2561 | int ret; | |
2562 | ||
581c1760 | 2563 | ret = btrfs_read_buffer(eb, gen, level, NULL); |
018642a1 TI |
2564 | if (ret) |
2565 | return ret; | |
e02119d5 CM |
2566 | |
2567 | level = btrfs_header_level(eb); | |
2568 | ||
2569 | if (level != 0) | |
2570 | return 0; | |
2571 | ||
2572 | path = btrfs_alloc_path(); | |
1e5063d0 MF |
2573 | if (!path) |
2574 | return -ENOMEM; | |
e02119d5 CM |
2575 | |
2576 | nritems = btrfs_header_nritems(eb); | |
2577 | for (i = 0; i < nritems; i++) { | |
2578 | btrfs_item_key_to_cpu(eb, &key, i); | |
e02119d5 CM |
2579 | |
2580 | /* inode keys are done during the first stage */ | |
2581 | if (key.type == BTRFS_INODE_ITEM_KEY && | |
2582 | wc->stage == LOG_WALK_REPLAY_INODES) { | |
e02119d5 CM |
2583 | struct btrfs_inode_item *inode_item; |
2584 | u32 mode; | |
2585 | ||
2586 | inode_item = btrfs_item_ptr(eb, i, | |
2587 | struct btrfs_inode_item); | |
f2d72f42 FM |
2588 | /* |
2589 | * If we have a tmpfile (O_TMPFILE) that got fsync'ed | |
2590 | * and never got linked before the fsync, skip it, as | |
2591 | * replaying it is pointless since it would be deleted | |
2592 | * later. We skip logging tmpfiles, but it's always | |
2593 | * possible we are replaying a log created with a kernel | |
2594 | * that used to log tmpfiles. | |
2595 | */ | |
2596 | if (btrfs_inode_nlink(eb, inode_item) == 0) { | |
2597 | wc->ignore_cur_inode = true; | |
2598 | continue; | |
2599 | } else { | |
2600 | wc->ignore_cur_inode = false; | |
2601 | } | |
4f764e51 FM |
2602 | ret = replay_xattr_deletes(wc->trans, root, log, |
2603 | path, key.objectid); | |
2604 | if (ret) | |
2605 | break; | |
e02119d5 CM |
2606 | mode = btrfs_inode_mode(eb, inode_item); |
2607 | if (S_ISDIR(mode)) { | |
2608 | ret = replay_dir_deletes(wc->trans, | |
12fcfd22 | 2609 | root, log, path, key.objectid, 0); |
b50c6e25 JB |
2610 | if (ret) |
2611 | break; | |
e02119d5 CM |
2612 | } |
2613 | ret = overwrite_item(wc->trans, root, path, | |
2614 | eb, i, &key); | |
b50c6e25 JB |
2615 | if (ret) |
2616 | break; | |
e02119d5 | 2617 | |
471d557a FM |
2618 | /* |
2619 | * Before replaying extents, truncate the inode to its | |
2620 | * size. We need to do it now and not after log replay | |
2621 | * because before an fsync we can have prealloc extents | |
2622 | * added beyond the inode's i_size. If we did it after, | |
2623 | * through orphan cleanup for example, we would drop | |
2624 | * those prealloc extents just after replaying them. | |
e02119d5 CM |
2625 | */ |
2626 | if (S_ISREG(mode)) { | |
5893dfb9 | 2627 | struct btrfs_drop_extents_args drop_args = { 0 }; |
471d557a FM |
2628 | struct inode *inode; |
2629 | u64 from; | |
2630 | ||
2631 | inode = read_one_inode(root, key.objectid); | |
2632 | if (!inode) { | |
2633 | ret = -EIO; | |
2634 | break; | |
2635 | } | |
2636 | from = ALIGN(i_size_read(inode), | |
2637 | root->fs_info->sectorsize); | |
5893dfb9 FM |
2638 | drop_args.start = from; |
2639 | drop_args.end = (u64)-1; | |
2640 | drop_args.drop_cache = true; | |
2641 | ret = btrfs_drop_extents(wc->trans, root, | |
2642 | BTRFS_I(inode), | |
2643 | &drop_args); | |
471d557a | 2644 | if (!ret) { |
2766ff61 FM |
2645 | inode_sub_bytes(inode, |
2646 | drop_args.bytes_found); | |
f2d72f42 | 2647 | /* Update the inode's nbytes. */ |
471d557a | 2648 | ret = btrfs_update_inode(wc->trans, |
9a56fcd1 | 2649 | root, BTRFS_I(inode)); |
471d557a FM |
2650 | } |
2651 | iput(inode); | |
b50c6e25 JB |
2652 | if (ret) |
2653 | break; | |
e02119d5 | 2654 | } |
c71bf099 | 2655 | |
e02119d5 CM |
2656 | ret = link_to_fixup_dir(wc->trans, root, |
2657 | path, key.objectid); | |
b50c6e25 JB |
2658 | if (ret) |
2659 | break; | |
e02119d5 | 2660 | } |
dd8e7217 | 2661 | |
f2d72f42 FM |
2662 | if (wc->ignore_cur_inode) |
2663 | continue; | |
2664 | ||
dd8e7217 JB |
2665 | if (key.type == BTRFS_DIR_INDEX_KEY && |
2666 | wc->stage == LOG_WALK_REPLAY_DIR_INDEX) { | |
2667 | ret = replay_one_dir_item(wc->trans, root, path, | |
2668 | eb, i, &key); | |
2669 | if (ret) | |
2670 | break; | |
2671 | } | |
2672 | ||
e02119d5 CM |
2673 | if (wc->stage < LOG_WALK_REPLAY_ALL) |
2674 | continue; | |
2675 | ||
2676 | /* these keys are simply copied */ | |
2677 | if (key.type == BTRFS_XATTR_ITEM_KEY) { | |
2678 | ret = overwrite_item(wc->trans, root, path, | |
2679 | eb, i, &key); | |
b50c6e25 JB |
2680 | if (ret) |
2681 | break; | |
2da1c669 LB |
2682 | } else if (key.type == BTRFS_INODE_REF_KEY || |
2683 | key.type == BTRFS_INODE_EXTREF_KEY) { | |
f186373f MF |
2684 | ret = add_inode_ref(wc->trans, root, log, path, |
2685 | eb, i, &key); | |
b50c6e25 JB |
2686 | if (ret && ret != -ENOENT) |
2687 | break; | |
2688 | ret = 0; | |
e02119d5 CM |
2689 | } else if (key.type == BTRFS_EXTENT_DATA_KEY) { |
2690 | ret = replay_one_extent(wc->trans, root, path, | |
2691 | eb, i, &key); | |
b50c6e25 JB |
2692 | if (ret) |
2693 | break; | |
dd8e7217 | 2694 | } else if (key.type == BTRFS_DIR_ITEM_KEY) { |
e02119d5 CM |
2695 | ret = replay_one_dir_item(wc->trans, root, path, |
2696 | eb, i, &key); | |
b50c6e25 JB |
2697 | if (ret) |
2698 | break; | |
e02119d5 CM |
2699 | } |
2700 | } | |
2701 | btrfs_free_path(path); | |
b50c6e25 | 2702 | return ret; |
e02119d5 CM |
2703 | } |
2704 | ||
6787bb9f NB |
2705 | /* |
2706 | * Correctly adjust the reserved bytes occupied by a log tree extent buffer | |
2707 | */ | |
2708 | static void unaccount_log_buffer(struct btrfs_fs_info *fs_info, u64 start) | |
2709 | { | |
2710 | struct btrfs_block_group *cache; | |
2711 | ||
2712 | cache = btrfs_lookup_block_group(fs_info, start); | |
2713 | if (!cache) { | |
2714 | btrfs_err(fs_info, "unable to find block group for %llu", start); | |
2715 | return; | |
2716 | } | |
2717 | ||
2718 | spin_lock(&cache->space_info->lock); | |
2719 | spin_lock(&cache->lock); | |
2720 | cache->reserved -= fs_info->nodesize; | |
2721 | cache->space_info->bytes_reserved -= fs_info->nodesize; | |
2722 | spin_unlock(&cache->lock); | |
2723 | spin_unlock(&cache->space_info->lock); | |
2724 | ||
2725 | btrfs_put_block_group(cache); | |
2726 | } | |
2727 | ||
d397712b | 2728 | static noinline int walk_down_log_tree(struct btrfs_trans_handle *trans, |
e02119d5 CM |
2729 | struct btrfs_root *root, |
2730 | struct btrfs_path *path, int *level, | |
2731 | struct walk_control *wc) | |
2732 | { | |
0b246afa | 2733 | struct btrfs_fs_info *fs_info = root->fs_info; |
e02119d5 CM |
2734 | u64 bytenr; |
2735 | u64 ptr_gen; | |
2736 | struct extent_buffer *next; | |
2737 | struct extent_buffer *cur; | |
e02119d5 CM |
2738 | u32 blocksize; |
2739 | int ret = 0; | |
2740 | ||
d397712b | 2741 | while (*level > 0) { |
581c1760 QW |
2742 | struct btrfs_key first_key; |
2743 | ||
e02119d5 CM |
2744 | cur = path->nodes[*level]; |
2745 | ||
fae7f21c | 2746 | WARN_ON(btrfs_header_level(cur) != *level); |
e02119d5 CM |
2747 | |
2748 | if (path->slots[*level] >= | |
2749 | btrfs_header_nritems(cur)) | |
2750 | break; | |
2751 | ||
2752 | bytenr = btrfs_node_blockptr(cur, path->slots[*level]); | |
2753 | ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]); | |
581c1760 | 2754 | btrfs_node_key_to_cpu(cur, &first_key, path->slots[*level]); |
0b246afa | 2755 | blocksize = fs_info->nodesize; |
e02119d5 | 2756 | |
3fbaf258 JB |
2757 | next = btrfs_find_create_tree_block(fs_info, bytenr, |
2758 | btrfs_header_owner(cur), | |
2759 | *level - 1); | |
c871b0f2 LB |
2760 | if (IS_ERR(next)) |
2761 | return PTR_ERR(next); | |
e02119d5 | 2762 | |
e02119d5 | 2763 | if (*level == 1) { |
581c1760 QW |
2764 | ret = wc->process_func(root, next, wc, ptr_gen, |
2765 | *level - 1); | |
b50c6e25 JB |
2766 | if (ret) { |
2767 | free_extent_buffer(next); | |
1e5063d0 | 2768 | return ret; |
b50c6e25 | 2769 | } |
4a500fd1 | 2770 | |
e02119d5 CM |
2771 | path->slots[*level]++; |
2772 | if (wc->free) { | |
581c1760 QW |
2773 | ret = btrfs_read_buffer(next, ptr_gen, |
2774 | *level - 1, &first_key); | |
018642a1 TI |
2775 | if (ret) { |
2776 | free_extent_buffer(next); | |
2777 | return ret; | |
2778 | } | |
e02119d5 | 2779 | |
681ae509 JB |
2780 | if (trans) { |
2781 | btrfs_tree_lock(next); | |
6a884d7d | 2782 | btrfs_clean_tree_block(next); |
681ae509 JB |
2783 | btrfs_wait_tree_block_writeback(next); |
2784 | btrfs_tree_unlock(next); | |
7bfc1007 | 2785 | ret = btrfs_pin_reserved_extent(trans, |
10e958d5 NB |
2786 | bytenr, blocksize); |
2787 | if (ret) { | |
2788 | free_extent_buffer(next); | |
2789 | return ret; | |
2790 | } | |
d3575156 NA |
2791 | btrfs_redirty_list_add( |
2792 | trans->transaction, next); | |
1846430c LB |
2793 | } else { |
2794 | if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &next->bflags)) | |
2795 | clear_extent_buffer_dirty(next); | |
10e958d5 | 2796 | unaccount_log_buffer(fs_info, bytenr); |
3650860b | 2797 | } |
e02119d5 CM |
2798 | } |
2799 | free_extent_buffer(next); | |
2800 | continue; | |
2801 | } | |
581c1760 | 2802 | ret = btrfs_read_buffer(next, ptr_gen, *level - 1, &first_key); |
018642a1 TI |
2803 | if (ret) { |
2804 | free_extent_buffer(next); | |
2805 | return ret; | |
2806 | } | |
e02119d5 | 2807 | |
e02119d5 CM |
2808 | if (path->nodes[*level-1]) |
2809 | free_extent_buffer(path->nodes[*level-1]); | |
2810 | path->nodes[*level-1] = next; | |
2811 | *level = btrfs_header_level(next); | |
2812 | path->slots[*level] = 0; | |
2813 | cond_resched(); | |
2814 | } | |
4a500fd1 | 2815 | path->slots[*level] = btrfs_header_nritems(path->nodes[*level]); |
e02119d5 CM |
2816 | |
2817 | cond_resched(); | |
2818 | return 0; | |
2819 | } | |
2820 | ||
d397712b | 2821 | static noinline int walk_up_log_tree(struct btrfs_trans_handle *trans, |
e02119d5 CM |
2822 | struct btrfs_root *root, |
2823 | struct btrfs_path *path, int *level, | |
2824 | struct walk_control *wc) | |
2825 | { | |
0b246afa | 2826 | struct btrfs_fs_info *fs_info = root->fs_info; |
e02119d5 CM |
2827 | int i; |
2828 | int slot; | |
2829 | int ret; | |
2830 | ||
d397712b | 2831 | for (i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) { |
e02119d5 | 2832 | slot = path->slots[i]; |
4a500fd1 | 2833 | if (slot + 1 < btrfs_header_nritems(path->nodes[i])) { |
e02119d5 CM |
2834 | path->slots[i]++; |
2835 | *level = i; | |
2836 | WARN_ON(*level == 0); | |
2837 | return 0; | |
2838 | } else { | |
1e5063d0 | 2839 | ret = wc->process_func(root, path->nodes[*level], wc, |
581c1760 QW |
2840 | btrfs_header_generation(path->nodes[*level]), |
2841 | *level); | |
1e5063d0 MF |
2842 | if (ret) |
2843 | return ret; | |
2844 | ||
e02119d5 CM |
2845 | if (wc->free) { |
2846 | struct extent_buffer *next; | |
2847 | ||
2848 | next = path->nodes[*level]; | |
2849 | ||
681ae509 JB |
2850 | if (trans) { |
2851 | btrfs_tree_lock(next); | |
6a884d7d | 2852 | btrfs_clean_tree_block(next); |
681ae509 JB |
2853 | btrfs_wait_tree_block_writeback(next); |
2854 | btrfs_tree_unlock(next); | |
7bfc1007 | 2855 | ret = btrfs_pin_reserved_extent(trans, |
10e958d5 NB |
2856 | path->nodes[*level]->start, |
2857 | path->nodes[*level]->len); | |
2858 | if (ret) | |
2859 | return ret; | |
1846430c LB |
2860 | } else { |
2861 | if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &next->bflags)) | |
2862 | clear_extent_buffer_dirty(next); | |
e02119d5 | 2863 | |
10e958d5 NB |
2864 | unaccount_log_buffer(fs_info, |
2865 | path->nodes[*level]->start); | |
2866 | } | |
e02119d5 CM |
2867 | } |
2868 | free_extent_buffer(path->nodes[*level]); | |
2869 | path->nodes[*level] = NULL; | |
2870 | *level = i + 1; | |
2871 | } | |
2872 | } | |
2873 | return 1; | |
2874 | } | |
2875 | ||
2876 | /* | |
2877 | * drop the reference count on the tree rooted at 'snap'. This traverses | |
2878 | * the tree freeing any blocks that have a ref count of zero after being | |
2879 | * decremented. | |
2880 | */ | |
2881 | static int walk_log_tree(struct btrfs_trans_handle *trans, | |
2882 | struct btrfs_root *log, struct walk_control *wc) | |
2883 | { | |
2ff7e61e | 2884 | struct btrfs_fs_info *fs_info = log->fs_info; |
e02119d5 CM |
2885 | int ret = 0; |
2886 | int wret; | |
2887 | int level; | |
2888 | struct btrfs_path *path; | |
e02119d5 CM |
2889 | int orig_level; |
2890 | ||
2891 | path = btrfs_alloc_path(); | |
db5b493a TI |
2892 | if (!path) |
2893 | return -ENOMEM; | |
e02119d5 CM |
2894 | |
2895 | level = btrfs_header_level(log->node); | |
2896 | orig_level = level; | |
2897 | path->nodes[level] = log->node; | |
67439dad | 2898 | atomic_inc(&log->node->refs); |
e02119d5 CM |
2899 | path->slots[level] = 0; |
2900 | ||
d397712b | 2901 | while (1) { |
e02119d5 CM |
2902 | wret = walk_down_log_tree(trans, log, path, &level, wc); |
2903 | if (wret > 0) | |
2904 | break; | |
79787eaa | 2905 | if (wret < 0) { |
e02119d5 | 2906 | ret = wret; |
79787eaa JM |
2907 | goto out; |
2908 | } | |
e02119d5 CM |
2909 | |
2910 | wret = walk_up_log_tree(trans, log, path, &level, wc); | |
2911 | if (wret > 0) | |
2912 | break; | |
79787eaa | 2913 | if (wret < 0) { |
e02119d5 | 2914 | ret = wret; |
79787eaa JM |
2915 | goto out; |
2916 | } | |
e02119d5 CM |
2917 | } |
2918 | ||
2919 | /* was the root node processed? if not, catch it here */ | |
2920 | if (path->nodes[orig_level]) { | |
79787eaa | 2921 | ret = wc->process_func(log, path->nodes[orig_level], wc, |
581c1760 QW |
2922 | btrfs_header_generation(path->nodes[orig_level]), |
2923 | orig_level); | |
79787eaa JM |
2924 | if (ret) |
2925 | goto out; | |
e02119d5 CM |
2926 | if (wc->free) { |
2927 | struct extent_buffer *next; | |
2928 | ||
2929 | next = path->nodes[orig_level]; | |
2930 | ||
681ae509 JB |
2931 | if (trans) { |
2932 | btrfs_tree_lock(next); | |
6a884d7d | 2933 | btrfs_clean_tree_block(next); |
681ae509 JB |
2934 | btrfs_wait_tree_block_writeback(next); |
2935 | btrfs_tree_unlock(next); | |
7bfc1007 | 2936 | ret = btrfs_pin_reserved_extent(trans, |
10e958d5 NB |
2937 | next->start, next->len); |
2938 | if (ret) | |
2939 | goto out; | |
1846430c LB |
2940 | } else { |
2941 | if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &next->bflags)) | |
2942 | clear_extent_buffer_dirty(next); | |
10e958d5 | 2943 | unaccount_log_buffer(fs_info, next->start); |
681ae509 | 2944 | } |
e02119d5 CM |
2945 | } |
2946 | } | |
2947 | ||
79787eaa | 2948 | out: |
e02119d5 | 2949 | btrfs_free_path(path); |
e02119d5 CM |
2950 | return ret; |
2951 | } | |
2952 | ||
7237f183 YZ |
2953 | /* |
2954 | * helper function to update the item for a given subvolumes log root | |
2955 | * in the tree of log roots | |
2956 | */ | |
2957 | static int update_log_root(struct btrfs_trans_handle *trans, | |
4203e968 JB |
2958 | struct btrfs_root *log, |
2959 | struct btrfs_root_item *root_item) | |
7237f183 | 2960 | { |
0b246afa | 2961 | struct btrfs_fs_info *fs_info = log->fs_info; |
7237f183 YZ |
2962 | int ret; |
2963 | ||
2964 | if (log->log_transid == 1) { | |
2965 | /* insert root item on the first sync */ | |
0b246afa | 2966 | ret = btrfs_insert_root(trans, fs_info->log_root_tree, |
4203e968 | 2967 | &log->root_key, root_item); |
7237f183 | 2968 | } else { |
0b246afa | 2969 | ret = btrfs_update_root(trans, fs_info->log_root_tree, |
4203e968 | 2970 | &log->root_key, root_item); |
7237f183 YZ |
2971 | } |
2972 | return ret; | |
2973 | } | |
2974 | ||
60d53eb3 | 2975 | static void wait_log_commit(struct btrfs_root *root, int transid) |
e02119d5 CM |
2976 | { |
2977 | DEFINE_WAIT(wait); | |
7237f183 | 2978 | int index = transid % 2; |
e02119d5 | 2979 | |
7237f183 YZ |
2980 | /* |
2981 | * we only allow two pending log transactions at a time, | |
2982 | * so we know that if ours is more than 2 older than the | |
2983 | * current transaction, we're done | |
2984 | */ | |
49e83f57 | 2985 | for (;;) { |
7237f183 YZ |
2986 | prepare_to_wait(&root->log_commit_wait[index], |
2987 | &wait, TASK_UNINTERRUPTIBLE); | |
12fcfd22 | 2988 | |
49e83f57 LB |
2989 | if (!(root->log_transid_committed < transid && |
2990 | atomic_read(&root->log_commit[index]))) | |
2991 | break; | |
12fcfd22 | 2992 | |
49e83f57 LB |
2993 | mutex_unlock(&root->log_mutex); |
2994 | schedule(); | |
7237f183 | 2995 | mutex_lock(&root->log_mutex); |
49e83f57 LB |
2996 | } |
2997 | finish_wait(&root->log_commit_wait[index], &wait); | |
7237f183 YZ |
2998 | } |
2999 | ||
60d53eb3 | 3000 | static void wait_for_writer(struct btrfs_root *root) |
7237f183 YZ |
3001 | { |
3002 | DEFINE_WAIT(wait); | |
8b050d35 | 3003 | |
49e83f57 LB |
3004 | for (;;) { |
3005 | prepare_to_wait(&root->log_writer_wait, &wait, | |
3006 | TASK_UNINTERRUPTIBLE); | |
3007 | if (!atomic_read(&root->log_writers)) | |
3008 | break; | |
3009 | ||
7237f183 | 3010 | mutex_unlock(&root->log_mutex); |
49e83f57 | 3011 | schedule(); |
575849ec | 3012 | mutex_lock(&root->log_mutex); |
7237f183 | 3013 | } |
49e83f57 | 3014 | finish_wait(&root->log_writer_wait, &wait); |
e02119d5 CM |
3015 | } |
3016 | ||
8b050d35 MX |
3017 | static inline void btrfs_remove_log_ctx(struct btrfs_root *root, |
3018 | struct btrfs_log_ctx *ctx) | |
3019 | { | |
3020 | if (!ctx) | |
3021 | return; | |
3022 | ||
3023 | mutex_lock(&root->log_mutex); | |
3024 | list_del_init(&ctx->list); | |
3025 | mutex_unlock(&root->log_mutex); | |
3026 | } | |
3027 | ||
3028 | /* | |
3029 | * Invoked in log mutex context, or be sure there is no other task which | |
3030 | * can access the list. | |
3031 | */ | |
3032 | static inline void btrfs_remove_all_log_ctxs(struct btrfs_root *root, | |
3033 | int index, int error) | |
3034 | { | |
3035 | struct btrfs_log_ctx *ctx; | |
570dd450 | 3036 | struct btrfs_log_ctx *safe; |
8b050d35 | 3037 | |
570dd450 CM |
3038 | list_for_each_entry_safe(ctx, safe, &root->log_ctxs[index], list) { |
3039 | list_del_init(&ctx->list); | |
8b050d35 | 3040 | ctx->log_ret = error; |
570dd450 | 3041 | } |
8b050d35 MX |
3042 | |
3043 | INIT_LIST_HEAD(&root->log_ctxs[index]); | |
3044 | } | |
3045 | ||
e02119d5 CM |
3046 | /* |
3047 | * btrfs_sync_log does sends a given tree log down to the disk and | |
3048 | * updates the super blocks to record it. When this call is done, | |
12fcfd22 CM |
3049 | * you know that any inodes previously logged are safely on disk only |
3050 | * if it returns 0. | |
3051 | * | |
3052 | * Any other return value means you need to call btrfs_commit_transaction. | |
3053 | * Some of the edge cases for fsyncing directories that have had unlinks | |
3054 | * or renames done in the past mean that sometimes the only safe | |
3055 | * fsync is to commit the whole FS. When btrfs_sync_log returns -EAGAIN, | |
3056 | * that has happened. | |
e02119d5 CM |
3057 | */ |
3058 | int btrfs_sync_log(struct btrfs_trans_handle *trans, | |
8b050d35 | 3059 | struct btrfs_root *root, struct btrfs_log_ctx *ctx) |
e02119d5 | 3060 | { |
7237f183 YZ |
3061 | int index1; |
3062 | int index2; | |
8cef4e16 | 3063 | int mark; |
e02119d5 | 3064 | int ret; |
0b246afa | 3065 | struct btrfs_fs_info *fs_info = root->fs_info; |
e02119d5 | 3066 | struct btrfs_root *log = root->log_root; |
0b246afa | 3067 | struct btrfs_root *log_root_tree = fs_info->log_root_tree; |
4203e968 | 3068 | struct btrfs_root_item new_root_item; |
bb14a59b | 3069 | int log_transid = 0; |
8b050d35 | 3070 | struct btrfs_log_ctx root_log_ctx; |
c6adc9cc | 3071 | struct blk_plug plug; |
47876f7c FM |
3072 | u64 log_root_start; |
3073 | u64 log_root_level; | |
e02119d5 | 3074 | |
7237f183 | 3075 | mutex_lock(&root->log_mutex); |
d1433deb MX |
3076 | log_transid = ctx->log_transid; |
3077 | if (root->log_transid_committed >= log_transid) { | |
3078 | mutex_unlock(&root->log_mutex); | |
3079 | return ctx->log_ret; | |
3080 | } | |
3081 | ||
3082 | index1 = log_transid % 2; | |
7237f183 | 3083 | if (atomic_read(&root->log_commit[index1])) { |
60d53eb3 | 3084 | wait_log_commit(root, log_transid); |
7237f183 | 3085 | mutex_unlock(&root->log_mutex); |
8b050d35 | 3086 | return ctx->log_ret; |
e02119d5 | 3087 | } |
d1433deb | 3088 | ASSERT(log_transid == root->log_transid); |
7237f183 YZ |
3089 | atomic_set(&root->log_commit[index1], 1); |
3090 | ||
3091 | /* wait for previous tree log sync to complete */ | |
3092 | if (atomic_read(&root->log_commit[(index1 + 1) % 2])) | |
60d53eb3 | 3093 | wait_log_commit(root, log_transid - 1); |
48cab2e0 | 3094 | |
86df7eb9 | 3095 | while (1) { |
2ecb7923 | 3096 | int batch = atomic_read(&root->log_batch); |
cd354ad6 | 3097 | /* when we're on an ssd, just kick the log commit out */ |
0b246afa | 3098 | if (!btrfs_test_opt(fs_info, SSD) && |
27cdeb70 | 3099 | test_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state)) { |
86df7eb9 YZ |
3100 | mutex_unlock(&root->log_mutex); |
3101 | schedule_timeout_uninterruptible(1); | |
3102 | mutex_lock(&root->log_mutex); | |
3103 | } | |
60d53eb3 | 3104 | wait_for_writer(root); |
2ecb7923 | 3105 | if (batch == atomic_read(&root->log_batch)) |
e02119d5 CM |
3106 | break; |
3107 | } | |
e02119d5 | 3108 | |
12fcfd22 | 3109 | /* bail out if we need to do a full commit */ |
4884b8e8 | 3110 | if (btrfs_need_log_full_commit(trans)) { |
12fcfd22 CM |
3111 | ret = -EAGAIN; |
3112 | mutex_unlock(&root->log_mutex); | |
3113 | goto out; | |
3114 | } | |
3115 | ||
8cef4e16 YZ |
3116 | if (log_transid % 2 == 0) |
3117 | mark = EXTENT_DIRTY; | |
3118 | else | |
3119 | mark = EXTENT_NEW; | |
3120 | ||
690587d1 CM |
3121 | /* we start IO on all the marked extents here, but we don't actually |
3122 | * wait for them until later. | |
3123 | */ | |
c6adc9cc | 3124 | blk_start_plug(&plug); |
2ff7e61e | 3125 | ret = btrfs_write_marked_extents(fs_info, &log->dirty_log_pages, mark); |
b528f467 NA |
3126 | /* |
3127 | * -EAGAIN happens when someone, e.g., a concurrent transaction | |
3128 | * commit, writes a dirty extent in this tree-log commit. This | |
3129 | * concurrent write will create a hole writing out the extents, | |
3130 | * and we cannot proceed on a zoned filesystem, requiring | |
3131 | * sequential writing. While we can bail out to a full commit | |
3132 | * here, but we can continue hoping the concurrent writing fills | |
3133 | * the hole. | |
3134 | */ | |
3135 | if (ret == -EAGAIN && btrfs_is_zoned(fs_info)) | |
3136 | ret = 0; | |
79787eaa | 3137 | if (ret) { |
c6adc9cc | 3138 | blk_finish_plug(&plug); |
66642832 | 3139 | btrfs_abort_transaction(trans, ret); |
90787766 | 3140 | btrfs_set_log_full_commit(trans); |
79787eaa JM |
3141 | mutex_unlock(&root->log_mutex); |
3142 | goto out; | |
3143 | } | |
7237f183 | 3144 | |
4203e968 JB |
3145 | /* |
3146 | * We _must_ update under the root->log_mutex in order to make sure we | |
3147 | * have a consistent view of the log root we are trying to commit at | |
3148 | * this moment. | |
3149 | * | |
3150 | * We _must_ copy this into a local copy, because we are not holding the | |
3151 | * log_root_tree->log_mutex yet. This is important because when we | |
3152 | * commit the log_root_tree we must have a consistent view of the | |
3153 | * log_root_tree when we update the super block to point at the | |
3154 | * log_root_tree bytenr. If we update the log_root_tree here we'll race | |
3155 | * with the commit and possibly point at the new block which we may not | |
3156 | * have written out. | |
3157 | */ | |
5d4f98a2 | 3158 | btrfs_set_root_node(&log->root_item, log->node); |
4203e968 | 3159 | memcpy(&new_root_item, &log->root_item, sizeof(new_root_item)); |
7237f183 | 3160 | |
7237f183 YZ |
3161 | root->log_transid++; |
3162 | log->log_transid = root->log_transid; | |
ff782e0a | 3163 | root->log_start_pid = 0; |
7237f183 | 3164 | /* |
8cef4e16 YZ |
3165 | * IO has been started, blocks of the log tree have WRITTEN flag set |
3166 | * in their headers. new modifications of the log will be written to | |
3167 | * new positions. so it's safe to allow log writers to go in. | |
7237f183 YZ |
3168 | */ |
3169 | mutex_unlock(&root->log_mutex); | |
3170 | ||
3ddebf27 | 3171 | if (btrfs_is_zoned(fs_info)) { |
e75f9fd1 | 3172 | mutex_lock(&fs_info->tree_root->log_mutex); |
3ddebf27 NA |
3173 | if (!log_root_tree->node) { |
3174 | ret = btrfs_alloc_log_tree_node(trans, log_root_tree); | |
3175 | if (ret) { | |
e75f9fd1 | 3176 | mutex_unlock(&fs_info->tree_log_mutex); |
3ddebf27 NA |
3177 | goto out; |
3178 | } | |
3179 | } | |
e75f9fd1 | 3180 | mutex_unlock(&fs_info->tree_root->log_mutex); |
3ddebf27 NA |
3181 | } |
3182 | ||
e75f9fd1 NA |
3183 | btrfs_init_log_ctx(&root_log_ctx, NULL); |
3184 | ||
3185 | mutex_lock(&log_root_tree->log_mutex); | |
3186 | ||
e3d3b415 FM |
3187 | index2 = log_root_tree->log_transid % 2; |
3188 | list_add_tail(&root_log_ctx.list, &log_root_tree->log_ctxs[index2]); | |
3189 | root_log_ctx.log_transid = log_root_tree->log_transid; | |
3190 | ||
4203e968 JB |
3191 | /* |
3192 | * Now we are safe to update the log_root_tree because we're under the | |
3193 | * log_mutex, and we're a current writer so we're holding the commit | |
3194 | * open until we drop the log_mutex. | |
3195 | */ | |
3196 | ret = update_log_root(trans, log, &new_root_item); | |
4a500fd1 | 3197 | if (ret) { |
d1433deb MX |
3198 | if (!list_empty(&root_log_ctx.list)) |
3199 | list_del_init(&root_log_ctx.list); | |
3200 | ||
c6adc9cc | 3201 | blk_finish_plug(&plug); |
90787766 | 3202 | btrfs_set_log_full_commit(trans); |
995946dd | 3203 | |
79787eaa | 3204 | if (ret != -ENOSPC) { |
66642832 | 3205 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
3206 | mutex_unlock(&log_root_tree->log_mutex); |
3207 | goto out; | |
3208 | } | |
bf89d38f | 3209 | btrfs_wait_tree_log_extents(log, mark); |
4a500fd1 YZ |
3210 | mutex_unlock(&log_root_tree->log_mutex); |
3211 | ret = -EAGAIN; | |
3212 | goto out; | |
3213 | } | |
3214 | ||
d1433deb | 3215 | if (log_root_tree->log_transid_committed >= root_log_ctx.log_transid) { |
3da5ab56 | 3216 | blk_finish_plug(&plug); |
cbd60aa7 | 3217 | list_del_init(&root_log_ctx.list); |
d1433deb MX |
3218 | mutex_unlock(&log_root_tree->log_mutex); |
3219 | ret = root_log_ctx.log_ret; | |
3220 | goto out; | |
3221 | } | |
8b050d35 | 3222 | |
d1433deb | 3223 | index2 = root_log_ctx.log_transid % 2; |
7237f183 | 3224 | if (atomic_read(&log_root_tree->log_commit[index2])) { |
c6adc9cc | 3225 | blk_finish_plug(&plug); |
bf89d38f | 3226 | ret = btrfs_wait_tree_log_extents(log, mark); |
60d53eb3 | 3227 | wait_log_commit(log_root_tree, |
d1433deb | 3228 | root_log_ctx.log_transid); |
7237f183 | 3229 | mutex_unlock(&log_root_tree->log_mutex); |
5ab5e44a FM |
3230 | if (!ret) |
3231 | ret = root_log_ctx.log_ret; | |
7237f183 YZ |
3232 | goto out; |
3233 | } | |
d1433deb | 3234 | ASSERT(root_log_ctx.log_transid == log_root_tree->log_transid); |
7237f183 YZ |
3235 | atomic_set(&log_root_tree->log_commit[index2], 1); |
3236 | ||
12fcfd22 | 3237 | if (atomic_read(&log_root_tree->log_commit[(index2 + 1) % 2])) { |
60d53eb3 | 3238 | wait_log_commit(log_root_tree, |
d1433deb | 3239 | root_log_ctx.log_transid - 1); |
12fcfd22 CM |
3240 | } |
3241 | ||
12fcfd22 CM |
3242 | /* |
3243 | * now that we've moved on to the tree of log tree roots, | |
3244 | * check the full commit flag again | |
3245 | */ | |
4884b8e8 | 3246 | if (btrfs_need_log_full_commit(trans)) { |
c6adc9cc | 3247 | blk_finish_plug(&plug); |
bf89d38f | 3248 | btrfs_wait_tree_log_extents(log, mark); |
12fcfd22 CM |
3249 | mutex_unlock(&log_root_tree->log_mutex); |
3250 | ret = -EAGAIN; | |
3251 | goto out_wake_log_root; | |
3252 | } | |
7237f183 | 3253 | |
2ff7e61e | 3254 | ret = btrfs_write_marked_extents(fs_info, |
c6adc9cc MX |
3255 | &log_root_tree->dirty_log_pages, |
3256 | EXTENT_DIRTY | EXTENT_NEW); | |
3257 | blk_finish_plug(&plug); | |
b528f467 NA |
3258 | /* |
3259 | * As described above, -EAGAIN indicates a hole in the extents. We | |
3260 | * cannot wait for these write outs since the waiting cause a | |
3261 | * deadlock. Bail out to the full commit instead. | |
3262 | */ | |
3263 | if (ret == -EAGAIN && btrfs_is_zoned(fs_info)) { | |
3264 | btrfs_set_log_full_commit(trans); | |
3265 | btrfs_wait_tree_log_extents(log, mark); | |
3266 | mutex_unlock(&log_root_tree->log_mutex); | |
3267 | goto out_wake_log_root; | |
3268 | } else if (ret) { | |
90787766 | 3269 | btrfs_set_log_full_commit(trans); |
66642832 | 3270 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
3271 | mutex_unlock(&log_root_tree->log_mutex); |
3272 | goto out_wake_log_root; | |
3273 | } | |
bf89d38f | 3274 | ret = btrfs_wait_tree_log_extents(log, mark); |
5ab5e44a | 3275 | if (!ret) |
bf89d38f JM |
3276 | ret = btrfs_wait_tree_log_extents(log_root_tree, |
3277 | EXTENT_NEW | EXTENT_DIRTY); | |
5ab5e44a | 3278 | if (ret) { |
90787766 | 3279 | btrfs_set_log_full_commit(trans); |
5ab5e44a FM |
3280 | mutex_unlock(&log_root_tree->log_mutex); |
3281 | goto out_wake_log_root; | |
3282 | } | |
e02119d5 | 3283 | |
47876f7c FM |
3284 | log_root_start = log_root_tree->node->start; |
3285 | log_root_level = btrfs_header_level(log_root_tree->node); | |
7237f183 | 3286 | log_root_tree->log_transid++; |
7237f183 YZ |
3287 | mutex_unlock(&log_root_tree->log_mutex); |
3288 | ||
3289 | /* | |
47876f7c FM |
3290 | * Here we are guaranteed that nobody is going to write the superblock |
3291 | * for the current transaction before us and that neither we do write | |
3292 | * our superblock before the previous transaction finishes its commit | |
3293 | * and writes its superblock, because: | |
3294 | * | |
3295 | * 1) We are holding a handle on the current transaction, so no body | |
3296 | * can commit it until we release the handle; | |
3297 | * | |
3298 | * 2) Before writing our superblock we acquire the tree_log_mutex, so | |
3299 | * if the previous transaction is still committing, and hasn't yet | |
3300 | * written its superblock, we wait for it to do it, because a | |
3301 | * transaction commit acquires the tree_log_mutex when the commit | |
3302 | * begins and releases it only after writing its superblock. | |
7237f183 | 3303 | */ |
47876f7c FM |
3304 | mutex_lock(&fs_info->tree_log_mutex); |
3305 | btrfs_set_super_log_root(fs_info->super_for_commit, log_root_start); | |
3306 | btrfs_set_super_log_root_level(fs_info->super_for_commit, log_root_level); | |
eece6a9c | 3307 | ret = write_all_supers(fs_info, 1); |
47876f7c | 3308 | mutex_unlock(&fs_info->tree_log_mutex); |
5af3e8cc | 3309 | if (ret) { |
90787766 | 3310 | btrfs_set_log_full_commit(trans); |
66642832 | 3311 | btrfs_abort_transaction(trans, ret); |
5af3e8cc SB |
3312 | goto out_wake_log_root; |
3313 | } | |
7237f183 | 3314 | |
257c62e1 CM |
3315 | mutex_lock(&root->log_mutex); |
3316 | if (root->last_log_commit < log_transid) | |
3317 | root->last_log_commit = log_transid; | |
3318 | mutex_unlock(&root->log_mutex); | |
3319 | ||
12fcfd22 | 3320 | out_wake_log_root: |
570dd450 | 3321 | mutex_lock(&log_root_tree->log_mutex); |
8b050d35 MX |
3322 | btrfs_remove_all_log_ctxs(log_root_tree, index2, ret); |
3323 | ||
d1433deb | 3324 | log_root_tree->log_transid_committed++; |
7237f183 | 3325 | atomic_set(&log_root_tree->log_commit[index2], 0); |
d1433deb MX |
3326 | mutex_unlock(&log_root_tree->log_mutex); |
3327 | ||
33a9eca7 | 3328 | /* |
093258e6 DS |
3329 | * The barrier before waitqueue_active (in cond_wake_up) is needed so |
3330 | * all the updates above are seen by the woken threads. It might not be | |
3331 | * necessary, but proving that seems to be hard. | |
33a9eca7 | 3332 | */ |
093258e6 | 3333 | cond_wake_up(&log_root_tree->log_commit_wait[index2]); |
e02119d5 | 3334 | out: |
d1433deb | 3335 | mutex_lock(&root->log_mutex); |
570dd450 | 3336 | btrfs_remove_all_log_ctxs(root, index1, ret); |
d1433deb | 3337 | root->log_transid_committed++; |
7237f183 | 3338 | atomic_set(&root->log_commit[index1], 0); |
d1433deb | 3339 | mutex_unlock(&root->log_mutex); |
8b050d35 | 3340 | |
33a9eca7 | 3341 | /* |
093258e6 DS |
3342 | * The barrier before waitqueue_active (in cond_wake_up) is needed so |
3343 | * all the updates above are seen by the woken threads. It might not be | |
3344 | * necessary, but proving that seems to be hard. | |
33a9eca7 | 3345 | */ |
093258e6 | 3346 | cond_wake_up(&root->log_commit_wait[index1]); |
b31eabd8 | 3347 | return ret; |
e02119d5 CM |
3348 | } |
3349 | ||
4a500fd1 YZ |
3350 | static void free_log_tree(struct btrfs_trans_handle *trans, |
3351 | struct btrfs_root *log) | |
e02119d5 CM |
3352 | { |
3353 | int ret; | |
e02119d5 CM |
3354 | struct walk_control wc = { |
3355 | .free = 1, | |
3356 | .process_func = process_one_buffer | |
3357 | }; | |
3358 | ||
3ddebf27 NA |
3359 | if (log->node) { |
3360 | ret = walk_log_tree(trans, log, &wc); | |
3361 | if (ret) { | |
3362 | if (trans) | |
3363 | btrfs_abort_transaction(trans, ret); | |
3364 | else | |
3365 | btrfs_handle_fs_error(log->fs_info, ret, NULL); | |
3366 | } | |
374b0e2d | 3367 | } |
e02119d5 | 3368 | |
59b0713a FM |
3369 | clear_extent_bits(&log->dirty_log_pages, 0, (u64)-1, |
3370 | EXTENT_DIRTY | EXTENT_NEW | EXTENT_NEED_WAIT); | |
e289f03e | 3371 | extent_io_tree_release(&log->log_csum_range); |
d3575156 NA |
3372 | |
3373 | if (trans && log->node) | |
3374 | btrfs_redirty_list_add(trans->transaction, log->node); | |
00246528 | 3375 | btrfs_put_root(log); |
4a500fd1 YZ |
3376 | } |
3377 | ||
3378 | /* | |
3379 | * free all the extents used by the tree log. This should be called | |
3380 | * at commit time of the full transaction | |
3381 | */ | |
3382 | int btrfs_free_log(struct btrfs_trans_handle *trans, struct btrfs_root *root) | |
3383 | { | |
3384 | if (root->log_root) { | |
3385 | free_log_tree(trans, root->log_root); | |
3386 | root->log_root = NULL; | |
e7a79811 | 3387 | clear_bit(BTRFS_ROOT_HAS_LOG_TREE, &root->state); |
4a500fd1 YZ |
3388 | } |
3389 | return 0; | |
3390 | } | |
3391 | ||
3392 | int btrfs_free_log_root_tree(struct btrfs_trans_handle *trans, | |
3393 | struct btrfs_fs_info *fs_info) | |
3394 | { | |
3395 | if (fs_info->log_root_tree) { | |
3396 | free_log_tree(trans, fs_info->log_root_tree); | |
3397 | fs_info->log_root_tree = NULL; | |
47876f7c | 3398 | clear_bit(BTRFS_ROOT_HAS_LOG_TREE, &fs_info->tree_root->state); |
4a500fd1 | 3399 | } |
e02119d5 CM |
3400 | return 0; |
3401 | } | |
3402 | ||
803f0f64 FM |
3403 | /* |
3404 | * Check if an inode was logged in the current transaction. We can't always rely | |
3405 | * on an inode's logged_trans value, because it's an in-memory only field and | |
3406 | * therefore not persisted. This means that its value is lost if the inode gets | |
3407 | * evicted and loaded again from disk (in which case it has a value of 0, and | |
3408 | * certainly it is smaller then any possible transaction ID), when that happens | |
3409 | * the full_sync flag is set in the inode's runtime flags, so on that case we | |
3410 | * assume eviction happened and ignore the logged_trans value, assuming the | |
3411 | * worst case, that the inode was logged before in the current transaction. | |
3412 | */ | |
3413 | static bool inode_logged(struct btrfs_trans_handle *trans, | |
3414 | struct btrfs_inode *inode) | |
3415 | { | |
3416 | if (inode->logged_trans == trans->transid) | |
3417 | return true; | |
3418 | ||
3419 | if (inode->last_trans == trans->transid && | |
3420 | test_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &inode->runtime_flags) && | |
3421 | !test_bit(BTRFS_FS_LOG_RECOVERING, &trans->fs_info->flags)) | |
3422 | return true; | |
3423 | ||
3424 | return false; | |
3425 | } | |
3426 | ||
e02119d5 CM |
3427 | /* |
3428 | * If both a file and directory are logged, and unlinks or renames are | |
3429 | * mixed in, we have a few interesting corners: | |
3430 | * | |
3431 | * create file X in dir Y | |
3432 | * link file X to X.link in dir Y | |
3433 | * fsync file X | |
3434 | * unlink file X but leave X.link | |
3435 | * fsync dir Y | |
3436 | * | |
3437 | * After a crash we would expect only X.link to exist. But file X | |
3438 | * didn't get fsync'd again so the log has back refs for X and X.link. | |
3439 | * | |
3440 | * We solve this by removing directory entries and inode backrefs from the | |
3441 | * log when a file that was logged in the current transaction is | |
3442 | * unlinked. Any later fsync will include the updated log entries, and | |
3443 | * we'll be able to reconstruct the proper directory items from backrefs. | |
3444 | * | |
3445 | * This optimizations allows us to avoid relogging the entire inode | |
3446 | * or the entire directory. | |
3447 | */ | |
3448 | int btrfs_del_dir_entries_in_log(struct btrfs_trans_handle *trans, | |
3449 | struct btrfs_root *root, | |
3450 | const char *name, int name_len, | |
49f34d1f | 3451 | struct btrfs_inode *dir, u64 index) |
e02119d5 CM |
3452 | { |
3453 | struct btrfs_root *log; | |
3454 | struct btrfs_dir_item *di; | |
3455 | struct btrfs_path *path; | |
3456 | int ret; | |
4a500fd1 | 3457 | int err = 0; |
49f34d1f | 3458 | u64 dir_ino = btrfs_ino(dir); |
e02119d5 | 3459 | |
803f0f64 | 3460 | if (!inode_logged(trans, dir)) |
3a5f1d45 CM |
3461 | return 0; |
3462 | ||
e02119d5 CM |
3463 | ret = join_running_log_trans(root); |
3464 | if (ret) | |
3465 | return 0; | |
3466 | ||
49f34d1f | 3467 | mutex_lock(&dir->log_mutex); |
e02119d5 CM |
3468 | |
3469 | log = root->log_root; | |
3470 | path = btrfs_alloc_path(); | |
a62f44a5 TI |
3471 | if (!path) { |
3472 | err = -ENOMEM; | |
3473 | goto out_unlock; | |
3474 | } | |
2a29edc6 | 3475 | |
33345d01 | 3476 | di = btrfs_lookup_dir_item(trans, log, path, dir_ino, |
e02119d5 | 3477 | name, name_len, -1); |
4a500fd1 YZ |
3478 | if (IS_ERR(di)) { |
3479 | err = PTR_ERR(di); | |
3480 | goto fail; | |
3481 | } | |
3482 | if (di) { | |
e02119d5 | 3483 | ret = btrfs_delete_one_dir_name(trans, log, path, di); |
3650860b JB |
3484 | if (ret) { |
3485 | err = ret; | |
3486 | goto fail; | |
3487 | } | |
e02119d5 | 3488 | } |
b3b4aa74 | 3489 | btrfs_release_path(path); |
33345d01 | 3490 | di = btrfs_lookup_dir_index_item(trans, log, path, dir_ino, |
e02119d5 | 3491 | index, name, name_len, -1); |
4a500fd1 YZ |
3492 | if (IS_ERR(di)) { |
3493 | err = PTR_ERR(di); | |
3494 | goto fail; | |
3495 | } | |
3496 | if (di) { | |
e02119d5 | 3497 | ret = btrfs_delete_one_dir_name(trans, log, path, di); |
3650860b JB |
3498 | if (ret) { |
3499 | err = ret; | |
3500 | goto fail; | |
3501 | } | |
e02119d5 CM |
3502 | } |
3503 | ||
ddffcf6f FM |
3504 | /* |
3505 | * We do not need to update the size field of the directory's inode item | |
3506 | * because on log replay we update the field to reflect all existing | |
3507 | * entries in the directory (see overwrite_item()). | |
e02119d5 | 3508 | */ |
4a500fd1 | 3509 | fail: |
e02119d5 | 3510 | btrfs_free_path(path); |
a62f44a5 | 3511 | out_unlock: |
49f34d1f | 3512 | mutex_unlock(&dir->log_mutex); |
fb2fecba | 3513 | if (err == -ENOSPC) { |
90787766 | 3514 | btrfs_set_log_full_commit(trans); |
fb2fecba JB |
3515 | err = 0; |
3516 | } else if (err < 0 && err != -ENOENT) { | |
3517 | /* ENOENT can be returned if the entry hasn't been fsynced yet */ | |
3518 | btrfs_abort_transaction(trans, err); | |
3519 | } | |
79787eaa | 3520 | |
12fcfd22 | 3521 | btrfs_end_log_trans(root); |
e02119d5 | 3522 | |
411fc6bc | 3523 | return err; |
e02119d5 CM |
3524 | } |
3525 | ||
3526 | /* see comments for btrfs_del_dir_entries_in_log */ | |
3527 | int btrfs_del_inode_ref_in_log(struct btrfs_trans_handle *trans, | |
3528 | struct btrfs_root *root, | |
3529 | const char *name, int name_len, | |
a491abb2 | 3530 | struct btrfs_inode *inode, u64 dirid) |
e02119d5 CM |
3531 | { |
3532 | struct btrfs_root *log; | |
3533 | u64 index; | |
3534 | int ret; | |
3535 | ||
803f0f64 | 3536 | if (!inode_logged(trans, inode)) |
3a5f1d45 CM |
3537 | return 0; |
3538 | ||
e02119d5 CM |
3539 | ret = join_running_log_trans(root); |
3540 | if (ret) | |
3541 | return 0; | |
3542 | log = root->log_root; | |
a491abb2 | 3543 | mutex_lock(&inode->log_mutex); |
e02119d5 | 3544 | |
a491abb2 | 3545 | ret = btrfs_del_inode_ref(trans, log, name, name_len, btrfs_ino(inode), |
e02119d5 | 3546 | dirid, &index); |
a491abb2 | 3547 | mutex_unlock(&inode->log_mutex); |
4a500fd1 | 3548 | if (ret == -ENOSPC) { |
90787766 | 3549 | btrfs_set_log_full_commit(trans); |
4a500fd1 | 3550 | ret = 0; |
79787eaa | 3551 | } else if (ret < 0 && ret != -ENOENT) |
66642832 | 3552 | btrfs_abort_transaction(trans, ret); |
12fcfd22 | 3553 | btrfs_end_log_trans(root); |
e02119d5 | 3554 | |
e02119d5 CM |
3555 | return ret; |
3556 | } | |
3557 | ||
3558 | /* | |
3559 | * creates a range item in the log for 'dirid'. first_offset and | |
3560 | * last_offset tell us which parts of the key space the log should | |
3561 | * be considered authoritative for. | |
3562 | */ | |
3563 | static noinline int insert_dir_log_key(struct btrfs_trans_handle *trans, | |
3564 | struct btrfs_root *log, | |
3565 | struct btrfs_path *path, | |
3566 | int key_type, u64 dirid, | |
3567 | u64 first_offset, u64 last_offset) | |
3568 | { | |
3569 | int ret; | |
3570 | struct btrfs_key key; | |
3571 | struct btrfs_dir_log_item *item; | |
3572 | ||
3573 | key.objectid = dirid; | |
3574 | key.offset = first_offset; | |
3575 | if (key_type == BTRFS_DIR_ITEM_KEY) | |
3576 | key.type = BTRFS_DIR_LOG_ITEM_KEY; | |
3577 | else | |
3578 | key.type = BTRFS_DIR_LOG_INDEX_KEY; | |
3579 | ret = btrfs_insert_empty_item(trans, log, path, &key, sizeof(*item)); | |
4a500fd1 YZ |
3580 | if (ret) |
3581 | return ret; | |
e02119d5 CM |
3582 | |
3583 | item = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
3584 | struct btrfs_dir_log_item); | |
3585 | btrfs_set_dir_log_end(path->nodes[0], item, last_offset); | |
3586 | btrfs_mark_buffer_dirty(path->nodes[0]); | |
b3b4aa74 | 3587 | btrfs_release_path(path); |
e02119d5 CM |
3588 | return 0; |
3589 | } | |
3590 | ||
3591 | /* | |
3592 | * log all the items included in the current transaction for a given | |
3593 | * directory. This also creates the range items in the log tree required | |
3594 | * to replay anything deleted before the fsync | |
3595 | */ | |
3596 | static noinline int log_dir_items(struct btrfs_trans_handle *trans, | |
684a5773 | 3597 | struct btrfs_root *root, struct btrfs_inode *inode, |
e02119d5 CM |
3598 | struct btrfs_path *path, |
3599 | struct btrfs_path *dst_path, int key_type, | |
2f2ff0ee | 3600 | struct btrfs_log_ctx *ctx, |
e02119d5 CM |
3601 | u64 min_offset, u64 *last_offset_ret) |
3602 | { | |
3603 | struct btrfs_key min_key; | |
e02119d5 CM |
3604 | struct btrfs_root *log = root->log_root; |
3605 | struct extent_buffer *src; | |
4a500fd1 | 3606 | int err = 0; |
e02119d5 CM |
3607 | int ret; |
3608 | int i; | |
3609 | int nritems; | |
3610 | u64 first_offset = min_offset; | |
3611 | u64 last_offset = (u64)-1; | |
684a5773 | 3612 | u64 ino = btrfs_ino(inode); |
e02119d5 CM |
3613 | |
3614 | log = root->log_root; | |
e02119d5 | 3615 | |
33345d01 | 3616 | min_key.objectid = ino; |
e02119d5 CM |
3617 | min_key.type = key_type; |
3618 | min_key.offset = min_offset; | |
3619 | ||
6174d3cb | 3620 | ret = btrfs_search_forward(root, &min_key, path, trans->transid); |
e02119d5 CM |
3621 | |
3622 | /* | |
3623 | * we didn't find anything from this transaction, see if there | |
3624 | * is anything at all | |
3625 | */ | |
33345d01 LZ |
3626 | if (ret != 0 || min_key.objectid != ino || min_key.type != key_type) { |
3627 | min_key.objectid = ino; | |
e02119d5 CM |
3628 | min_key.type = key_type; |
3629 | min_key.offset = (u64)-1; | |
b3b4aa74 | 3630 | btrfs_release_path(path); |
e02119d5 CM |
3631 | ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0); |
3632 | if (ret < 0) { | |
b3b4aa74 | 3633 | btrfs_release_path(path); |
e02119d5 CM |
3634 | return ret; |
3635 | } | |
33345d01 | 3636 | ret = btrfs_previous_item(root, path, ino, key_type); |
e02119d5 CM |
3637 | |
3638 | /* if ret == 0 there are items for this type, | |
3639 | * create a range to tell us the last key of this type. | |
3640 | * otherwise, there are no items in this directory after | |
3641 | * *min_offset, and we create a range to indicate that. | |
3642 | */ | |
3643 | if (ret == 0) { | |
3644 | struct btrfs_key tmp; | |
3645 | btrfs_item_key_to_cpu(path->nodes[0], &tmp, | |
3646 | path->slots[0]); | |
d397712b | 3647 | if (key_type == tmp.type) |
e02119d5 | 3648 | first_offset = max(min_offset, tmp.offset) + 1; |
e02119d5 CM |
3649 | } |
3650 | goto done; | |
3651 | } | |
3652 | ||
3653 | /* go backward to find any previous key */ | |
33345d01 | 3654 | ret = btrfs_previous_item(root, path, ino, key_type); |
e02119d5 CM |
3655 | if (ret == 0) { |
3656 | struct btrfs_key tmp; | |
3657 | btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]); | |
3658 | if (key_type == tmp.type) { | |
3659 | first_offset = tmp.offset; | |
3660 | ret = overwrite_item(trans, log, dst_path, | |
3661 | path->nodes[0], path->slots[0], | |
3662 | &tmp); | |
4a500fd1 YZ |
3663 | if (ret) { |
3664 | err = ret; | |
3665 | goto done; | |
3666 | } | |
e02119d5 CM |
3667 | } |
3668 | } | |
b3b4aa74 | 3669 | btrfs_release_path(path); |
e02119d5 | 3670 | |
2cc83342 JB |
3671 | /* |
3672 | * Find the first key from this transaction again. See the note for | |
3673 | * log_new_dir_dentries, if we're logging a directory recursively we | |
3674 | * won't be holding its i_mutex, which means we can modify the directory | |
3675 | * while we're logging it. If we remove an entry between our first | |
3676 | * search and this search we'll not find the key again and can just | |
3677 | * bail. | |
3678 | */ | |
bb56f02f | 3679 | search: |
e02119d5 | 3680 | ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0); |
2cc83342 | 3681 | if (ret != 0) |
e02119d5 | 3682 | goto done; |
e02119d5 CM |
3683 | |
3684 | /* | |
3685 | * we have a block from this transaction, log every item in it | |
3686 | * from our directory | |
3687 | */ | |
d397712b | 3688 | while (1) { |
e02119d5 CM |
3689 | struct btrfs_key tmp; |
3690 | src = path->nodes[0]; | |
3691 | nritems = btrfs_header_nritems(src); | |
3692 | for (i = path->slots[0]; i < nritems; i++) { | |
2f2ff0ee FM |
3693 | struct btrfs_dir_item *di; |
3694 | ||
e02119d5 CM |
3695 | btrfs_item_key_to_cpu(src, &min_key, i); |
3696 | ||
33345d01 | 3697 | if (min_key.objectid != ino || min_key.type != key_type) |
e02119d5 | 3698 | goto done; |
bb56f02f FM |
3699 | |
3700 | if (need_resched()) { | |
3701 | btrfs_release_path(path); | |
3702 | cond_resched(); | |
3703 | goto search; | |
3704 | } | |
3705 | ||
e02119d5 CM |
3706 | ret = overwrite_item(trans, log, dst_path, src, i, |
3707 | &min_key); | |
4a500fd1 YZ |
3708 | if (ret) { |
3709 | err = ret; | |
3710 | goto done; | |
3711 | } | |
2f2ff0ee FM |
3712 | |
3713 | /* | |
3714 | * We must make sure that when we log a directory entry, | |
3715 | * the corresponding inode, after log replay, has a | |
3716 | * matching link count. For example: | |
3717 | * | |
3718 | * touch foo | |
3719 | * mkdir mydir | |
3720 | * sync | |
3721 | * ln foo mydir/bar | |
3722 | * xfs_io -c "fsync" mydir | |
3723 | * <crash> | |
3724 | * <mount fs and log replay> | |
3725 | * | |
3726 | * Would result in a fsync log that when replayed, our | |
3727 | * file inode would have a link count of 1, but we get | |
3728 | * two directory entries pointing to the same inode. | |
3729 | * After removing one of the names, it would not be | |
3730 | * possible to remove the other name, which resulted | |
3731 | * always in stale file handle errors, and would not | |
3732 | * be possible to rmdir the parent directory, since | |
3733 | * its i_size could never decrement to the value | |
3734 | * BTRFS_EMPTY_DIR_SIZE, resulting in -ENOTEMPTY errors. | |
3735 | */ | |
3736 | di = btrfs_item_ptr(src, i, struct btrfs_dir_item); | |
3737 | btrfs_dir_item_key_to_cpu(src, di, &tmp); | |
3738 | if (ctx && | |
3739 | (btrfs_dir_transid(src, di) == trans->transid || | |
3740 | btrfs_dir_type(src, di) == BTRFS_FT_DIR) && | |
3741 | tmp.type != BTRFS_ROOT_ITEM_KEY) | |
3742 | ctx->log_new_dentries = true; | |
e02119d5 CM |
3743 | } |
3744 | path->slots[0] = nritems; | |
3745 | ||
3746 | /* | |
3747 | * look ahead to the next item and see if it is also | |
3748 | * from this directory and from this transaction | |
3749 | */ | |
3750 | ret = btrfs_next_leaf(root, path); | |
80c0b421 LB |
3751 | if (ret) { |
3752 | if (ret == 1) | |
3753 | last_offset = (u64)-1; | |
3754 | else | |
3755 | err = ret; | |
e02119d5 CM |
3756 | goto done; |
3757 | } | |
3758 | btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]); | |
33345d01 | 3759 | if (tmp.objectid != ino || tmp.type != key_type) { |
e02119d5 CM |
3760 | last_offset = (u64)-1; |
3761 | goto done; | |
3762 | } | |
3763 | if (btrfs_header_generation(path->nodes[0]) != trans->transid) { | |
3764 | ret = overwrite_item(trans, log, dst_path, | |
3765 | path->nodes[0], path->slots[0], | |
3766 | &tmp); | |
4a500fd1 YZ |
3767 | if (ret) |
3768 | err = ret; | |
3769 | else | |
3770 | last_offset = tmp.offset; | |
e02119d5 CM |
3771 | goto done; |
3772 | } | |
3773 | } | |
3774 | done: | |
b3b4aa74 DS |
3775 | btrfs_release_path(path); |
3776 | btrfs_release_path(dst_path); | |
e02119d5 | 3777 | |
4a500fd1 YZ |
3778 | if (err == 0) { |
3779 | *last_offset_ret = last_offset; | |
3780 | /* | |
3781 | * insert the log range keys to indicate where the log | |
3782 | * is valid | |
3783 | */ | |
3784 | ret = insert_dir_log_key(trans, log, path, key_type, | |
33345d01 | 3785 | ino, first_offset, last_offset); |
4a500fd1 YZ |
3786 | if (ret) |
3787 | err = ret; | |
3788 | } | |
3789 | return err; | |
e02119d5 CM |
3790 | } |
3791 | ||
3792 | /* | |
3793 | * logging directories is very similar to logging inodes, We find all the items | |
3794 | * from the current transaction and write them to the log. | |
3795 | * | |
3796 | * The recovery code scans the directory in the subvolume, and if it finds a | |
3797 | * key in the range logged that is not present in the log tree, then it means | |
3798 | * that dir entry was unlinked during the transaction. | |
3799 | * | |
3800 | * In order for that scan to work, we must include one key smaller than | |
3801 | * the smallest logged by this transaction and one key larger than the largest | |
3802 | * key logged by this transaction. | |
3803 | */ | |
3804 | static noinline int log_directory_changes(struct btrfs_trans_handle *trans, | |
dbf39ea4 | 3805 | struct btrfs_root *root, struct btrfs_inode *inode, |
e02119d5 | 3806 | struct btrfs_path *path, |
2f2ff0ee FM |
3807 | struct btrfs_path *dst_path, |
3808 | struct btrfs_log_ctx *ctx) | |
e02119d5 CM |
3809 | { |
3810 | u64 min_key; | |
3811 | u64 max_key; | |
3812 | int ret; | |
3813 | int key_type = BTRFS_DIR_ITEM_KEY; | |
3814 | ||
3815 | again: | |
3816 | min_key = 0; | |
3817 | max_key = 0; | |
d397712b | 3818 | while (1) { |
dbf39ea4 NB |
3819 | ret = log_dir_items(trans, root, inode, path, dst_path, key_type, |
3820 | ctx, min_key, &max_key); | |
4a500fd1 YZ |
3821 | if (ret) |
3822 | return ret; | |
e02119d5 CM |
3823 | if (max_key == (u64)-1) |
3824 | break; | |
3825 | min_key = max_key + 1; | |
3826 | } | |
3827 | ||
3828 | if (key_type == BTRFS_DIR_ITEM_KEY) { | |
3829 | key_type = BTRFS_DIR_INDEX_KEY; | |
3830 | goto again; | |
3831 | } | |
3832 | return 0; | |
3833 | } | |
3834 | ||
3835 | /* | |
3836 | * a helper function to drop items from the log before we relog an | |
3837 | * inode. max_key_type indicates the highest item type to remove. | |
3838 | * This cannot be run for file data extents because it does not | |
3839 | * free the extents they point to. | |
3840 | */ | |
3841 | static int drop_objectid_items(struct btrfs_trans_handle *trans, | |
3842 | struct btrfs_root *log, | |
3843 | struct btrfs_path *path, | |
3844 | u64 objectid, int max_key_type) | |
3845 | { | |
3846 | int ret; | |
3847 | struct btrfs_key key; | |
3848 | struct btrfs_key found_key; | |
18ec90d6 | 3849 | int start_slot; |
e02119d5 CM |
3850 | |
3851 | key.objectid = objectid; | |
3852 | key.type = max_key_type; | |
3853 | key.offset = (u64)-1; | |
3854 | ||
d397712b | 3855 | while (1) { |
e02119d5 | 3856 | ret = btrfs_search_slot(trans, log, &key, path, -1, 1); |
3650860b | 3857 | BUG_ON(ret == 0); /* Logic error */ |
4a500fd1 | 3858 | if (ret < 0) |
e02119d5 CM |
3859 | break; |
3860 | ||
3861 | if (path->slots[0] == 0) | |
3862 | break; | |
3863 | ||
3864 | path->slots[0]--; | |
3865 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, | |
3866 | path->slots[0]); | |
3867 | ||
3868 | if (found_key.objectid != objectid) | |
3869 | break; | |
3870 | ||
18ec90d6 JB |
3871 | found_key.offset = 0; |
3872 | found_key.type = 0; | |
e3b83361 | 3873 | ret = btrfs_bin_search(path->nodes[0], &found_key, &start_slot); |
cbca7d59 FM |
3874 | if (ret < 0) |
3875 | break; | |
18ec90d6 JB |
3876 | |
3877 | ret = btrfs_del_items(trans, log, path, start_slot, | |
3878 | path->slots[0] - start_slot + 1); | |
3879 | /* | |
3880 | * If start slot isn't 0 then we don't need to re-search, we've | |
3881 | * found the last guy with the objectid in this tree. | |
3882 | */ | |
3883 | if (ret || start_slot != 0) | |
65a246c5 | 3884 | break; |
b3b4aa74 | 3885 | btrfs_release_path(path); |
e02119d5 | 3886 | } |
b3b4aa74 | 3887 | btrfs_release_path(path); |
5bdbeb21 JB |
3888 | if (ret > 0) |
3889 | ret = 0; | |
4a500fd1 | 3890 | return ret; |
e02119d5 CM |
3891 | } |
3892 | ||
94edf4ae JB |
3893 | static void fill_inode_item(struct btrfs_trans_handle *trans, |
3894 | struct extent_buffer *leaf, | |
3895 | struct btrfs_inode_item *item, | |
1a4bcf47 FM |
3896 | struct inode *inode, int log_inode_only, |
3897 | u64 logged_isize) | |
94edf4ae | 3898 | { |
0b1c6cca JB |
3899 | struct btrfs_map_token token; |
3900 | ||
c82f823c | 3901 | btrfs_init_map_token(&token, leaf); |
94edf4ae JB |
3902 | |
3903 | if (log_inode_only) { | |
3904 | /* set the generation to zero so the recover code | |
3905 | * can tell the difference between an logging | |
3906 | * just to say 'this inode exists' and a logging | |
3907 | * to say 'update this inode with these values' | |
3908 | */ | |
cc4c13d5 DS |
3909 | btrfs_set_token_inode_generation(&token, item, 0); |
3910 | btrfs_set_token_inode_size(&token, item, logged_isize); | |
94edf4ae | 3911 | } else { |
cc4c13d5 DS |
3912 | btrfs_set_token_inode_generation(&token, item, |
3913 | BTRFS_I(inode)->generation); | |
3914 | btrfs_set_token_inode_size(&token, item, inode->i_size); | |
0b1c6cca JB |
3915 | } |
3916 | ||
cc4c13d5 DS |
3917 | btrfs_set_token_inode_uid(&token, item, i_uid_read(inode)); |
3918 | btrfs_set_token_inode_gid(&token, item, i_gid_read(inode)); | |
3919 | btrfs_set_token_inode_mode(&token, item, inode->i_mode); | |
3920 | btrfs_set_token_inode_nlink(&token, item, inode->i_nlink); | |
3921 | ||
3922 | btrfs_set_token_timespec_sec(&token, &item->atime, | |
3923 | inode->i_atime.tv_sec); | |
3924 | btrfs_set_token_timespec_nsec(&token, &item->atime, | |
3925 | inode->i_atime.tv_nsec); | |
3926 | ||
3927 | btrfs_set_token_timespec_sec(&token, &item->mtime, | |
3928 | inode->i_mtime.tv_sec); | |
3929 | btrfs_set_token_timespec_nsec(&token, &item->mtime, | |
3930 | inode->i_mtime.tv_nsec); | |
3931 | ||
3932 | btrfs_set_token_timespec_sec(&token, &item->ctime, | |
3933 | inode->i_ctime.tv_sec); | |
3934 | btrfs_set_token_timespec_nsec(&token, &item->ctime, | |
3935 | inode->i_ctime.tv_nsec); | |
3936 | ||
e593e54e FM |
3937 | /* |
3938 | * We do not need to set the nbytes field, in fact during a fast fsync | |
3939 | * its value may not even be correct, since a fast fsync does not wait | |
3940 | * for ordered extent completion, which is where we update nbytes, it | |
3941 | * only waits for writeback to complete. During log replay as we find | |
3942 | * file extent items and replay them, we adjust the nbytes field of the | |
3943 | * inode item in subvolume tree as needed (see overwrite_item()). | |
3944 | */ | |
cc4c13d5 DS |
3945 | |
3946 | btrfs_set_token_inode_sequence(&token, item, inode_peek_iversion(inode)); | |
3947 | btrfs_set_token_inode_transid(&token, item, trans->transid); | |
3948 | btrfs_set_token_inode_rdev(&token, item, inode->i_rdev); | |
3949 | btrfs_set_token_inode_flags(&token, item, BTRFS_I(inode)->flags); | |
3950 | btrfs_set_token_inode_block_group(&token, item, 0); | |
94edf4ae JB |
3951 | } |
3952 | ||
a95249b3 JB |
3953 | static int log_inode_item(struct btrfs_trans_handle *trans, |
3954 | struct btrfs_root *log, struct btrfs_path *path, | |
6d889a3b | 3955 | struct btrfs_inode *inode) |
a95249b3 JB |
3956 | { |
3957 | struct btrfs_inode_item *inode_item; | |
a95249b3 JB |
3958 | int ret; |
3959 | ||
efd0c405 | 3960 | ret = btrfs_insert_empty_item(trans, log, path, |
6d889a3b | 3961 | &inode->location, sizeof(*inode_item)); |
a95249b3 JB |
3962 | if (ret && ret != -EEXIST) |
3963 | return ret; | |
3964 | inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
3965 | struct btrfs_inode_item); | |
6d889a3b NB |
3966 | fill_inode_item(trans, path->nodes[0], inode_item, &inode->vfs_inode, |
3967 | 0, 0); | |
a95249b3 JB |
3968 | btrfs_release_path(path); |
3969 | return 0; | |
3970 | } | |
3971 | ||
40e046ac | 3972 | static int log_csums(struct btrfs_trans_handle *trans, |
3ebac17c | 3973 | struct btrfs_inode *inode, |
40e046ac FM |
3974 | struct btrfs_root *log_root, |
3975 | struct btrfs_ordered_sum *sums) | |
3976 | { | |
e289f03e FM |
3977 | const u64 lock_end = sums->bytenr + sums->len - 1; |
3978 | struct extent_state *cached_state = NULL; | |
40e046ac FM |
3979 | int ret; |
3980 | ||
3ebac17c FM |
3981 | /* |
3982 | * If this inode was not used for reflink operations in the current | |
3983 | * transaction with new extents, then do the fast path, no need to | |
3984 | * worry about logging checksum items with overlapping ranges. | |
3985 | */ | |
3986 | if (inode->last_reflink_trans < trans->transid) | |
3987 | return btrfs_csum_file_blocks(trans, log_root, sums); | |
3988 | ||
e289f03e FM |
3989 | /* |
3990 | * Serialize logging for checksums. This is to avoid racing with the | |
3991 | * same checksum being logged by another task that is logging another | |
3992 | * file which happens to refer to the same extent as well. Such races | |
3993 | * can leave checksum items in the log with overlapping ranges. | |
3994 | */ | |
3995 | ret = lock_extent_bits(&log_root->log_csum_range, sums->bytenr, | |
3996 | lock_end, &cached_state); | |
3997 | if (ret) | |
3998 | return ret; | |
40e046ac FM |
3999 | /* |
4000 | * Due to extent cloning, we might have logged a csum item that covers a | |
4001 | * subrange of a cloned extent, and later we can end up logging a csum | |
4002 | * item for a larger subrange of the same extent or the entire range. | |
4003 | * This would leave csum items in the log tree that cover the same range | |
4004 | * and break the searches for checksums in the log tree, resulting in | |
4005 | * some checksums missing in the fs/subvolume tree. So just delete (or | |
4006 | * trim and adjust) any existing csum items in the log for this range. | |
4007 | */ | |
4008 | ret = btrfs_del_csums(trans, log_root, sums->bytenr, sums->len); | |
e289f03e FM |
4009 | if (!ret) |
4010 | ret = btrfs_csum_file_blocks(trans, log_root, sums); | |
40e046ac | 4011 | |
e289f03e FM |
4012 | unlock_extent_cached(&log_root->log_csum_range, sums->bytenr, lock_end, |
4013 | &cached_state); | |
4014 | ||
4015 | return ret; | |
40e046ac FM |
4016 | } |
4017 | ||
31ff1cd2 | 4018 | static noinline int copy_items(struct btrfs_trans_handle *trans, |
44d70e19 | 4019 | struct btrfs_inode *inode, |
31ff1cd2 | 4020 | struct btrfs_path *dst_path, |
0e56315c | 4021 | struct btrfs_path *src_path, |
1a4bcf47 FM |
4022 | int start_slot, int nr, int inode_only, |
4023 | u64 logged_isize) | |
31ff1cd2 | 4024 | { |
3ffbd68c | 4025 | struct btrfs_fs_info *fs_info = trans->fs_info; |
31ff1cd2 CM |
4026 | unsigned long src_offset; |
4027 | unsigned long dst_offset; | |
44d70e19 | 4028 | struct btrfs_root *log = inode->root->log_root; |
31ff1cd2 CM |
4029 | struct btrfs_file_extent_item *extent; |
4030 | struct btrfs_inode_item *inode_item; | |
16e7549f | 4031 | struct extent_buffer *src = src_path->nodes[0]; |
31ff1cd2 CM |
4032 | int ret; |
4033 | struct btrfs_key *ins_keys; | |
4034 | u32 *ins_sizes; | |
4035 | char *ins_data; | |
4036 | int i; | |
d20f7043 | 4037 | struct list_head ordered_sums; |
44d70e19 | 4038 | int skip_csum = inode->flags & BTRFS_INODE_NODATASUM; |
d20f7043 CM |
4039 | |
4040 | INIT_LIST_HEAD(&ordered_sums); | |
31ff1cd2 CM |
4041 | |
4042 | ins_data = kmalloc(nr * sizeof(struct btrfs_key) + | |
4043 | nr * sizeof(u32), GFP_NOFS); | |
2a29edc6 | 4044 | if (!ins_data) |
4045 | return -ENOMEM; | |
4046 | ||
31ff1cd2 CM |
4047 | ins_sizes = (u32 *)ins_data; |
4048 | ins_keys = (struct btrfs_key *)(ins_data + nr * sizeof(u32)); | |
4049 | ||
4050 | for (i = 0; i < nr; i++) { | |
4051 | ins_sizes[i] = btrfs_item_size_nr(src, i + start_slot); | |
4052 | btrfs_item_key_to_cpu(src, ins_keys + i, i + start_slot); | |
4053 | } | |
4054 | ret = btrfs_insert_empty_items(trans, log, dst_path, | |
4055 | ins_keys, ins_sizes, nr); | |
4a500fd1 YZ |
4056 | if (ret) { |
4057 | kfree(ins_data); | |
4058 | return ret; | |
4059 | } | |
31ff1cd2 | 4060 | |
5d4f98a2 | 4061 | for (i = 0; i < nr; i++, dst_path->slots[0]++) { |
31ff1cd2 CM |
4062 | dst_offset = btrfs_item_ptr_offset(dst_path->nodes[0], |
4063 | dst_path->slots[0]); | |
4064 | ||
4065 | src_offset = btrfs_item_ptr_offset(src, start_slot + i); | |
4066 | ||
94edf4ae | 4067 | if (ins_keys[i].type == BTRFS_INODE_ITEM_KEY) { |
31ff1cd2 CM |
4068 | inode_item = btrfs_item_ptr(dst_path->nodes[0], |
4069 | dst_path->slots[0], | |
4070 | struct btrfs_inode_item); | |
94edf4ae | 4071 | fill_inode_item(trans, dst_path->nodes[0], inode_item, |
f85b7379 DS |
4072 | &inode->vfs_inode, |
4073 | inode_only == LOG_INODE_EXISTS, | |
1a4bcf47 | 4074 | logged_isize); |
94edf4ae JB |
4075 | } else { |
4076 | copy_extent_buffer(dst_path->nodes[0], src, dst_offset, | |
4077 | src_offset, ins_sizes[i]); | |
31ff1cd2 | 4078 | } |
94edf4ae | 4079 | |
31ff1cd2 CM |
4080 | /* take a reference on file data extents so that truncates |
4081 | * or deletes of this inode don't have to relog the inode | |
4082 | * again | |
4083 | */ | |
962a298f | 4084 | if (ins_keys[i].type == BTRFS_EXTENT_DATA_KEY && |
d2794405 | 4085 | !skip_csum) { |
31ff1cd2 CM |
4086 | int found_type; |
4087 | extent = btrfs_item_ptr(src, start_slot + i, | |
4088 | struct btrfs_file_extent_item); | |
4089 | ||
8e531cdf | 4090 | if (btrfs_file_extent_generation(src, extent) < trans->transid) |
4091 | continue; | |
4092 | ||
31ff1cd2 | 4093 | found_type = btrfs_file_extent_type(src, extent); |
6f1fed77 | 4094 | if (found_type == BTRFS_FILE_EXTENT_REG) { |
5d4f98a2 YZ |
4095 | u64 ds, dl, cs, cl; |
4096 | ds = btrfs_file_extent_disk_bytenr(src, | |
4097 | extent); | |
4098 | /* ds == 0 is a hole */ | |
4099 | if (ds == 0) | |
4100 | continue; | |
4101 | ||
4102 | dl = btrfs_file_extent_disk_num_bytes(src, | |
4103 | extent); | |
4104 | cs = btrfs_file_extent_offset(src, extent); | |
4105 | cl = btrfs_file_extent_num_bytes(src, | |
a419aef8 | 4106 | extent); |
580afd76 CM |
4107 | if (btrfs_file_extent_compression(src, |
4108 | extent)) { | |
4109 | cs = 0; | |
4110 | cl = dl; | |
4111 | } | |
5d4f98a2 YZ |
4112 | |
4113 | ret = btrfs_lookup_csums_range( | |
0b246afa | 4114 | fs_info->csum_root, |
5d4f98a2 | 4115 | ds + cs, ds + cs + cl - 1, |
a2de733c | 4116 | &ordered_sums, 0); |
4f26433e FM |
4117 | if (ret) |
4118 | break; | |
31ff1cd2 CM |
4119 | } |
4120 | } | |
31ff1cd2 CM |
4121 | } |
4122 | ||
4123 | btrfs_mark_buffer_dirty(dst_path->nodes[0]); | |
b3b4aa74 | 4124 | btrfs_release_path(dst_path); |
31ff1cd2 | 4125 | kfree(ins_data); |
d20f7043 CM |
4126 | |
4127 | /* | |
4128 | * we have to do this after the loop above to avoid changing the | |
4129 | * log tree while trying to change the log tree. | |
4130 | */ | |
d397712b | 4131 | while (!list_empty(&ordered_sums)) { |
d20f7043 CM |
4132 | struct btrfs_ordered_sum *sums = list_entry(ordered_sums.next, |
4133 | struct btrfs_ordered_sum, | |
4134 | list); | |
4a500fd1 | 4135 | if (!ret) |
3ebac17c | 4136 | ret = log_csums(trans, inode, log, sums); |
d20f7043 CM |
4137 | list_del(&sums->list); |
4138 | kfree(sums); | |
4139 | } | |
16e7549f | 4140 | |
4a500fd1 | 4141 | return ret; |
31ff1cd2 CM |
4142 | } |
4143 | ||
5dc562c5 JB |
4144 | static int extent_cmp(void *priv, struct list_head *a, struct list_head *b) |
4145 | { | |
4146 | struct extent_map *em1, *em2; | |
4147 | ||
4148 | em1 = list_entry(a, struct extent_map, list); | |
4149 | em2 = list_entry(b, struct extent_map, list); | |
4150 | ||
4151 | if (em1->start < em2->start) | |
4152 | return -1; | |
4153 | else if (em1->start > em2->start) | |
4154 | return 1; | |
4155 | return 0; | |
4156 | } | |
4157 | ||
e7175a69 JB |
4158 | static int log_extent_csums(struct btrfs_trans_handle *trans, |
4159 | struct btrfs_inode *inode, | |
a9ecb653 | 4160 | struct btrfs_root *log_root, |
48778179 FM |
4161 | const struct extent_map *em, |
4162 | struct btrfs_log_ctx *ctx) | |
5dc562c5 | 4163 | { |
48778179 | 4164 | struct btrfs_ordered_extent *ordered; |
2ab28f32 JB |
4165 | u64 csum_offset; |
4166 | u64 csum_len; | |
48778179 FM |
4167 | u64 mod_start = em->mod_start; |
4168 | u64 mod_len = em->mod_len; | |
8407f553 FM |
4169 | LIST_HEAD(ordered_sums); |
4170 | int ret = 0; | |
0aa4a17d | 4171 | |
e7175a69 JB |
4172 | if (inode->flags & BTRFS_INODE_NODATASUM || |
4173 | test_bit(EXTENT_FLAG_PREALLOC, &em->flags) || | |
8407f553 | 4174 | em->block_start == EXTENT_MAP_HOLE) |
70c8a91c | 4175 | return 0; |
5dc562c5 | 4176 | |
48778179 FM |
4177 | list_for_each_entry(ordered, &ctx->ordered_extents, log_list) { |
4178 | const u64 ordered_end = ordered->file_offset + ordered->num_bytes; | |
4179 | const u64 mod_end = mod_start + mod_len; | |
4180 | struct btrfs_ordered_sum *sums; | |
4181 | ||
4182 | if (mod_len == 0) | |
4183 | break; | |
4184 | ||
4185 | if (ordered_end <= mod_start) | |
4186 | continue; | |
4187 | if (mod_end <= ordered->file_offset) | |
4188 | break; | |
4189 | ||
4190 | /* | |
4191 | * We are going to copy all the csums on this ordered extent, so | |
4192 | * go ahead and adjust mod_start and mod_len in case this ordered | |
4193 | * extent has already been logged. | |
4194 | */ | |
4195 | if (ordered->file_offset > mod_start) { | |
4196 | if (ordered_end >= mod_end) | |
4197 | mod_len = ordered->file_offset - mod_start; | |
4198 | /* | |
4199 | * If we have this case | |
4200 | * | |
4201 | * |--------- logged extent ---------| | |
4202 | * |----- ordered extent ----| | |
4203 | * | |
4204 | * Just don't mess with mod_start and mod_len, we'll | |
4205 | * just end up logging more csums than we need and it | |
4206 | * will be ok. | |
4207 | */ | |
4208 | } else { | |
4209 | if (ordered_end < mod_end) { | |
4210 | mod_len = mod_end - ordered_end; | |
4211 | mod_start = ordered_end; | |
4212 | } else { | |
4213 | mod_len = 0; | |
4214 | } | |
4215 | } | |
4216 | ||
4217 | /* | |
4218 | * To keep us from looping for the above case of an ordered | |
4219 | * extent that falls inside of the logged extent. | |
4220 | */ | |
4221 | if (test_and_set_bit(BTRFS_ORDERED_LOGGED_CSUM, &ordered->flags)) | |
4222 | continue; | |
4223 | ||
4224 | list_for_each_entry(sums, &ordered->list, list) { | |
4225 | ret = log_csums(trans, inode, log_root, sums); | |
4226 | if (ret) | |
4227 | return ret; | |
4228 | } | |
4229 | } | |
4230 | ||
4231 | /* We're done, found all csums in the ordered extents. */ | |
4232 | if (mod_len == 0) | |
4233 | return 0; | |
4234 | ||
e7175a69 | 4235 | /* If we're compressed we have to save the entire range of csums. */ |
488111aa FDBM |
4236 | if (em->compress_type) { |
4237 | csum_offset = 0; | |
8407f553 | 4238 | csum_len = max(em->block_len, em->orig_block_len); |
488111aa | 4239 | } else { |
48778179 FM |
4240 | csum_offset = mod_start - em->start; |
4241 | csum_len = mod_len; | |
488111aa | 4242 | } |
2ab28f32 | 4243 | |
70c8a91c | 4244 | /* block start is already adjusted for the file extent offset. */ |
a9ecb653 | 4245 | ret = btrfs_lookup_csums_range(trans->fs_info->csum_root, |
70c8a91c JB |
4246 | em->block_start + csum_offset, |
4247 | em->block_start + csum_offset + | |
4248 | csum_len - 1, &ordered_sums, 0); | |
4249 | if (ret) | |
4250 | return ret; | |
5dc562c5 | 4251 | |
70c8a91c JB |
4252 | while (!list_empty(&ordered_sums)) { |
4253 | struct btrfs_ordered_sum *sums = list_entry(ordered_sums.next, | |
4254 | struct btrfs_ordered_sum, | |
4255 | list); | |
4256 | if (!ret) | |
3ebac17c | 4257 | ret = log_csums(trans, inode, log_root, sums); |
70c8a91c JB |
4258 | list_del(&sums->list); |
4259 | kfree(sums); | |
5dc562c5 JB |
4260 | } |
4261 | ||
70c8a91c | 4262 | return ret; |
5dc562c5 JB |
4263 | } |
4264 | ||
8407f553 | 4265 | static int log_one_extent(struct btrfs_trans_handle *trans, |
9d122629 | 4266 | struct btrfs_inode *inode, struct btrfs_root *root, |
8407f553 FM |
4267 | const struct extent_map *em, |
4268 | struct btrfs_path *path, | |
8407f553 FM |
4269 | struct btrfs_log_ctx *ctx) |
4270 | { | |
5893dfb9 | 4271 | struct btrfs_drop_extents_args drop_args = { 0 }; |
8407f553 FM |
4272 | struct btrfs_root *log = root->log_root; |
4273 | struct btrfs_file_extent_item *fi; | |
4274 | struct extent_buffer *leaf; | |
4275 | struct btrfs_map_token token; | |
4276 | struct btrfs_key key; | |
4277 | u64 extent_offset = em->start - em->orig_start; | |
4278 | u64 block_len; | |
4279 | int ret; | |
8407f553 | 4280 | |
48778179 | 4281 | ret = log_extent_csums(trans, inode, log, em, ctx); |
8407f553 FM |
4282 | if (ret) |
4283 | return ret; | |
4284 | ||
5893dfb9 FM |
4285 | drop_args.path = path; |
4286 | drop_args.start = em->start; | |
4287 | drop_args.end = em->start + em->len; | |
4288 | drop_args.replace_extent = true; | |
4289 | drop_args.extent_item_size = sizeof(*fi); | |
4290 | ret = btrfs_drop_extents(trans, log, inode, &drop_args); | |
8407f553 FM |
4291 | if (ret) |
4292 | return ret; | |
4293 | ||
5893dfb9 | 4294 | if (!drop_args.extent_inserted) { |
9d122629 | 4295 | key.objectid = btrfs_ino(inode); |
8407f553 FM |
4296 | key.type = BTRFS_EXTENT_DATA_KEY; |
4297 | key.offset = em->start; | |
4298 | ||
4299 | ret = btrfs_insert_empty_item(trans, log, path, &key, | |
4300 | sizeof(*fi)); | |
4301 | if (ret) | |
4302 | return ret; | |
4303 | } | |
4304 | leaf = path->nodes[0]; | |
c82f823c | 4305 | btrfs_init_map_token(&token, leaf); |
8407f553 FM |
4306 | fi = btrfs_item_ptr(leaf, path->slots[0], |
4307 | struct btrfs_file_extent_item); | |
4308 | ||
cc4c13d5 | 4309 | btrfs_set_token_file_extent_generation(&token, fi, trans->transid); |
8407f553 | 4310 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) |
cc4c13d5 DS |
4311 | btrfs_set_token_file_extent_type(&token, fi, |
4312 | BTRFS_FILE_EXTENT_PREALLOC); | |
8407f553 | 4313 | else |
cc4c13d5 DS |
4314 | btrfs_set_token_file_extent_type(&token, fi, |
4315 | BTRFS_FILE_EXTENT_REG); | |
8407f553 FM |
4316 | |
4317 | block_len = max(em->block_len, em->orig_block_len); | |
4318 | if (em->compress_type != BTRFS_COMPRESS_NONE) { | |
cc4c13d5 DS |
4319 | btrfs_set_token_file_extent_disk_bytenr(&token, fi, |
4320 | em->block_start); | |
4321 | btrfs_set_token_file_extent_disk_num_bytes(&token, fi, block_len); | |
8407f553 | 4322 | } else if (em->block_start < EXTENT_MAP_LAST_BYTE) { |
cc4c13d5 | 4323 | btrfs_set_token_file_extent_disk_bytenr(&token, fi, |
8407f553 | 4324 | em->block_start - |
cc4c13d5 DS |
4325 | extent_offset); |
4326 | btrfs_set_token_file_extent_disk_num_bytes(&token, fi, block_len); | |
8407f553 | 4327 | } else { |
cc4c13d5 DS |
4328 | btrfs_set_token_file_extent_disk_bytenr(&token, fi, 0); |
4329 | btrfs_set_token_file_extent_disk_num_bytes(&token, fi, 0); | |
8407f553 FM |
4330 | } |
4331 | ||
cc4c13d5 DS |
4332 | btrfs_set_token_file_extent_offset(&token, fi, extent_offset); |
4333 | btrfs_set_token_file_extent_num_bytes(&token, fi, em->len); | |
4334 | btrfs_set_token_file_extent_ram_bytes(&token, fi, em->ram_bytes); | |
4335 | btrfs_set_token_file_extent_compression(&token, fi, em->compress_type); | |
4336 | btrfs_set_token_file_extent_encryption(&token, fi, 0); | |
4337 | btrfs_set_token_file_extent_other_encoding(&token, fi, 0); | |
8407f553 FM |
4338 | btrfs_mark_buffer_dirty(leaf); |
4339 | ||
4340 | btrfs_release_path(path); | |
4341 | ||
4342 | return ret; | |
4343 | } | |
4344 | ||
31d11b83 FM |
4345 | /* |
4346 | * Log all prealloc extents beyond the inode's i_size to make sure we do not | |
4347 | * lose them after doing a fast fsync and replaying the log. We scan the | |
4348 | * subvolume's root instead of iterating the inode's extent map tree because | |
4349 | * otherwise we can log incorrect extent items based on extent map conversion. | |
4350 | * That can happen due to the fact that extent maps are merged when they | |
4351 | * are not in the extent map tree's list of modified extents. | |
4352 | */ | |
4353 | static int btrfs_log_prealloc_extents(struct btrfs_trans_handle *trans, | |
4354 | struct btrfs_inode *inode, | |
4355 | struct btrfs_path *path) | |
4356 | { | |
4357 | struct btrfs_root *root = inode->root; | |
4358 | struct btrfs_key key; | |
4359 | const u64 i_size = i_size_read(&inode->vfs_inode); | |
4360 | const u64 ino = btrfs_ino(inode); | |
4361 | struct btrfs_path *dst_path = NULL; | |
0e56315c | 4362 | bool dropped_extents = false; |
f135cea3 FM |
4363 | u64 truncate_offset = i_size; |
4364 | struct extent_buffer *leaf; | |
4365 | int slot; | |
31d11b83 FM |
4366 | int ins_nr = 0; |
4367 | int start_slot; | |
4368 | int ret; | |
4369 | ||
4370 | if (!(inode->flags & BTRFS_INODE_PREALLOC)) | |
4371 | return 0; | |
4372 | ||
4373 | key.objectid = ino; | |
4374 | key.type = BTRFS_EXTENT_DATA_KEY; | |
4375 | key.offset = i_size; | |
4376 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
4377 | if (ret < 0) | |
4378 | goto out; | |
4379 | ||
f135cea3 FM |
4380 | /* |
4381 | * We must check if there is a prealloc extent that starts before the | |
4382 | * i_size and crosses the i_size boundary. This is to ensure later we | |
4383 | * truncate down to the end of that extent and not to the i_size, as | |
4384 | * otherwise we end up losing part of the prealloc extent after a log | |
4385 | * replay and with an implicit hole if there is another prealloc extent | |
4386 | * that starts at an offset beyond i_size. | |
4387 | */ | |
4388 | ret = btrfs_previous_item(root, path, ino, BTRFS_EXTENT_DATA_KEY); | |
4389 | if (ret < 0) | |
4390 | goto out; | |
4391 | ||
4392 | if (ret == 0) { | |
4393 | struct btrfs_file_extent_item *ei; | |
4394 | ||
4395 | leaf = path->nodes[0]; | |
4396 | slot = path->slots[0]; | |
4397 | ei = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item); | |
4398 | ||
4399 | if (btrfs_file_extent_type(leaf, ei) == | |
4400 | BTRFS_FILE_EXTENT_PREALLOC) { | |
4401 | u64 extent_end; | |
4402 | ||
4403 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
4404 | extent_end = key.offset + | |
4405 | btrfs_file_extent_num_bytes(leaf, ei); | |
4406 | ||
4407 | if (extent_end > i_size) | |
4408 | truncate_offset = extent_end; | |
4409 | } | |
4410 | } else { | |
4411 | ret = 0; | |
4412 | } | |
4413 | ||
31d11b83 | 4414 | while (true) { |
f135cea3 FM |
4415 | leaf = path->nodes[0]; |
4416 | slot = path->slots[0]; | |
31d11b83 FM |
4417 | |
4418 | if (slot >= btrfs_header_nritems(leaf)) { | |
4419 | if (ins_nr > 0) { | |
4420 | ret = copy_items(trans, inode, dst_path, path, | |
0e56315c | 4421 | start_slot, ins_nr, 1, 0); |
31d11b83 FM |
4422 | if (ret < 0) |
4423 | goto out; | |
4424 | ins_nr = 0; | |
4425 | } | |
4426 | ret = btrfs_next_leaf(root, path); | |
4427 | if (ret < 0) | |
4428 | goto out; | |
4429 | if (ret > 0) { | |
4430 | ret = 0; | |
4431 | break; | |
4432 | } | |
4433 | continue; | |
4434 | } | |
4435 | ||
4436 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
4437 | if (key.objectid > ino) | |
4438 | break; | |
4439 | if (WARN_ON_ONCE(key.objectid < ino) || | |
4440 | key.type < BTRFS_EXTENT_DATA_KEY || | |
4441 | key.offset < i_size) { | |
4442 | path->slots[0]++; | |
4443 | continue; | |
4444 | } | |
0e56315c | 4445 | if (!dropped_extents) { |
31d11b83 FM |
4446 | /* |
4447 | * Avoid logging extent items logged in past fsync calls | |
4448 | * and leading to duplicate keys in the log tree. | |
4449 | */ | |
4450 | do { | |
4451 | ret = btrfs_truncate_inode_items(trans, | |
4452 | root->log_root, | |
50743398 | 4453 | inode, truncate_offset, |
31d11b83 FM |
4454 | BTRFS_EXTENT_DATA_KEY); |
4455 | } while (ret == -EAGAIN); | |
4456 | if (ret) | |
4457 | goto out; | |
0e56315c | 4458 | dropped_extents = true; |
31d11b83 FM |
4459 | } |
4460 | if (ins_nr == 0) | |
4461 | start_slot = slot; | |
4462 | ins_nr++; | |
4463 | path->slots[0]++; | |
4464 | if (!dst_path) { | |
4465 | dst_path = btrfs_alloc_path(); | |
4466 | if (!dst_path) { | |
4467 | ret = -ENOMEM; | |
4468 | goto out; | |
4469 | } | |
4470 | } | |
4471 | } | |
0bc2d3c0 | 4472 | if (ins_nr > 0) |
0e56315c | 4473 | ret = copy_items(trans, inode, dst_path, path, |
31d11b83 | 4474 | start_slot, ins_nr, 1, 0); |
31d11b83 FM |
4475 | out: |
4476 | btrfs_release_path(path); | |
4477 | btrfs_free_path(dst_path); | |
4478 | return ret; | |
4479 | } | |
4480 | ||
5dc562c5 JB |
4481 | static int btrfs_log_changed_extents(struct btrfs_trans_handle *trans, |
4482 | struct btrfs_root *root, | |
9d122629 | 4483 | struct btrfs_inode *inode, |
827463c4 | 4484 | struct btrfs_path *path, |
48778179 | 4485 | struct btrfs_log_ctx *ctx) |
5dc562c5 | 4486 | { |
48778179 FM |
4487 | struct btrfs_ordered_extent *ordered; |
4488 | struct btrfs_ordered_extent *tmp; | |
5dc562c5 JB |
4489 | struct extent_map *em, *n; |
4490 | struct list_head extents; | |
9d122629 | 4491 | struct extent_map_tree *tree = &inode->extent_tree; |
5dc562c5 | 4492 | int ret = 0; |
2ab28f32 | 4493 | int num = 0; |
5dc562c5 JB |
4494 | |
4495 | INIT_LIST_HEAD(&extents); | |
4496 | ||
5dc562c5 | 4497 | write_lock(&tree->lock); |
5dc562c5 JB |
4498 | |
4499 | list_for_each_entry_safe(em, n, &tree->modified_extents, list) { | |
4500 | list_del_init(&em->list); | |
2ab28f32 JB |
4501 | /* |
4502 | * Just an arbitrary number, this can be really CPU intensive | |
4503 | * once we start getting a lot of extents, and really once we | |
4504 | * have a bunch of extents we just want to commit since it will | |
4505 | * be faster. | |
4506 | */ | |
4507 | if (++num > 32768) { | |
4508 | list_del_init(&tree->modified_extents); | |
4509 | ret = -EFBIG; | |
4510 | goto process; | |
4511 | } | |
4512 | ||
5f96bfb7 | 4513 | if (em->generation < trans->transid) |
5dc562c5 | 4514 | continue; |
8c6c5928 | 4515 | |
31d11b83 FM |
4516 | /* We log prealloc extents beyond eof later. */ |
4517 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) && | |
4518 | em->start >= i_size_read(&inode->vfs_inode)) | |
4519 | continue; | |
4520 | ||
ff44c6e3 | 4521 | /* Need a ref to keep it from getting evicted from cache */ |
490b54d6 | 4522 | refcount_inc(&em->refs); |
ff44c6e3 | 4523 | set_bit(EXTENT_FLAG_LOGGING, &em->flags); |
5dc562c5 | 4524 | list_add_tail(&em->list, &extents); |
2ab28f32 | 4525 | num++; |
5dc562c5 JB |
4526 | } |
4527 | ||
4528 | list_sort(NULL, &extents, extent_cmp); | |
2ab28f32 | 4529 | process: |
5dc562c5 JB |
4530 | while (!list_empty(&extents)) { |
4531 | em = list_entry(extents.next, struct extent_map, list); | |
4532 | ||
4533 | list_del_init(&em->list); | |
4534 | ||
4535 | /* | |
4536 | * If we had an error we just need to delete everybody from our | |
4537 | * private list. | |
4538 | */ | |
ff44c6e3 | 4539 | if (ret) { |
201a9038 | 4540 | clear_em_logging(tree, em); |
ff44c6e3 | 4541 | free_extent_map(em); |
5dc562c5 | 4542 | continue; |
ff44c6e3 JB |
4543 | } |
4544 | ||
4545 | write_unlock(&tree->lock); | |
5dc562c5 | 4546 | |
a2120a47 | 4547 | ret = log_one_extent(trans, inode, root, em, path, ctx); |
ff44c6e3 | 4548 | write_lock(&tree->lock); |
201a9038 JB |
4549 | clear_em_logging(tree, em); |
4550 | free_extent_map(em); | |
5dc562c5 | 4551 | } |
ff44c6e3 JB |
4552 | WARN_ON(!list_empty(&extents)); |
4553 | write_unlock(&tree->lock); | |
5dc562c5 | 4554 | |
5dc562c5 | 4555 | btrfs_release_path(path); |
31d11b83 FM |
4556 | if (!ret) |
4557 | ret = btrfs_log_prealloc_extents(trans, inode, path); | |
48778179 FM |
4558 | if (ret) |
4559 | return ret; | |
31d11b83 | 4560 | |
48778179 FM |
4561 | /* |
4562 | * We have logged all extents successfully, now make sure the commit of | |
4563 | * the current transaction waits for the ordered extents to complete | |
4564 | * before it commits and wipes out the log trees, otherwise we would | |
4565 | * lose data if an ordered extents completes after the transaction | |
4566 | * commits and a power failure happens after the transaction commit. | |
4567 | */ | |
4568 | list_for_each_entry_safe(ordered, tmp, &ctx->ordered_extents, log_list) { | |
4569 | list_del_init(&ordered->log_list); | |
4570 | set_bit(BTRFS_ORDERED_LOGGED, &ordered->flags); | |
4571 | ||
4572 | if (!test_bit(BTRFS_ORDERED_COMPLETE, &ordered->flags)) { | |
4573 | spin_lock_irq(&inode->ordered_tree.lock); | |
4574 | if (!test_bit(BTRFS_ORDERED_COMPLETE, &ordered->flags)) { | |
4575 | set_bit(BTRFS_ORDERED_PENDING, &ordered->flags); | |
4576 | atomic_inc(&trans->transaction->pending_ordered); | |
4577 | } | |
4578 | spin_unlock_irq(&inode->ordered_tree.lock); | |
4579 | } | |
4580 | btrfs_put_ordered_extent(ordered); | |
4581 | } | |
4582 | ||
4583 | return 0; | |
5dc562c5 JB |
4584 | } |
4585 | ||
481b01c0 | 4586 | static int logged_inode_size(struct btrfs_root *log, struct btrfs_inode *inode, |
1a4bcf47 FM |
4587 | struct btrfs_path *path, u64 *size_ret) |
4588 | { | |
4589 | struct btrfs_key key; | |
4590 | int ret; | |
4591 | ||
481b01c0 | 4592 | key.objectid = btrfs_ino(inode); |
1a4bcf47 FM |
4593 | key.type = BTRFS_INODE_ITEM_KEY; |
4594 | key.offset = 0; | |
4595 | ||
4596 | ret = btrfs_search_slot(NULL, log, &key, path, 0, 0); | |
4597 | if (ret < 0) { | |
4598 | return ret; | |
4599 | } else if (ret > 0) { | |
2f2ff0ee | 4600 | *size_ret = 0; |
1a4bcf47 FM |
4601 | } else { |
4602 | struct btrfs_inode_item *item; | |
4603 | ||
4604 | item = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
4605 | struct btrfs_inode_item); | |
4606 | *size_ret = btrfs_inode_size(path->nodes[0], item); | |
bf504110 FM |
4607 | /* |
4608 | * If the in-memory inode's i_size is smaller then the inode | |
4609 | * size stored in the btree, return the inode's i_size, so | |
4610 | * that we get a correct inode size after replaying the log | |
4611 | * when before a power failure we had a shrinking truncate | |
4612 | * followed by addition of a new name (rename / new hard link). | |
4613 | * Otherwise return the inode size from the btree, to avoid | |
4614 | * data loss when replaying a log due to previously doing a | |
4615 | * write that expands the inode's size and logging a new name | |
4616 | * immediately after. | |
4617 | */ | |
4618 | if (*size_ret > inode->vfs_inode.i_size) | |
4619 | *size_ret = inode->vfs_inode.i_size; | |
1a4bcf47 FM |
4620 | } |
4621 | ||
4622 | btrfs_release_path(path); | |
4623 | return 0; | |
4624 | } | |
4625 | ||
36283bf7 FM |
4626 | /* |
4627 | * At the moment we always log all xattrs. This is to figure out at log replay | |
4628 | * time which xattrs must have their deletion replayed. If a xattr is missing | |
4629 | * in the log tree and exists in the fs/subvol tree, we delete it. This is | |
4630 | * because if a xattr is deleted, the inode is fsynced and a power failure | |
4631 | * happens, causing the log to be replayed the next time the fs is mounted, | |
4632 | * we want the xattr to not exist anymore (same behaviour as other filesystems | |
4633 | * with a journal, ext3/4, xfs, f2fs, etc). | |
4634 | */ | |
4635 | static int btrfs_log_all_xattrs(struct btrfs_trans_handle *trans, | |
4636 | struct btrfs_root *root, | |
1a93c36a | 4637 | struct btrfs_inode *inode, |
36283bf7 FM |
4638 | struct btrfs_path *path, |
4639 | struct btrfs_path *dst_path) | |
4640 | { | |
4641 | int ret; | |
4642 | struct btrfs_key key; | |
1a93c36a | 4643 | const u64 ino = btrfs_ino(inode); |
36283bf7 FM |
4644 | int ins_nr = 0; |
4645 | int start_slot = 0; | |
f2f121ab FM |
4646 | bool found_xattrs = false; |
4647 | ||
4648 | if (test_bit(BTRFS_INODE_NO_XATTRS, &inode->runtime_flags)) | |
4649 | return 0; | |
36283bf7 FM |
4650 | |
4651 | key.objectid = ino; | |
4652 | key.type = BTRFS_XATTR_ITEM_KEY; | |
4653 | key.offset = 0; | |
4654 | ||
4655 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
4656 | if (ret < 0) | |
4657 | return ret; | |
4658 | ||
4659 | while (true) { | |
4660 | int slot = path->slots[0]; | |
4661 | struct extent_buffer *leaf = path->nodes[0]; | |
4662 | int nritems = btrfs_header_nritems(leaf); | |
4663 | ||
4664 | if (slot >= nritems) { | |
4665 | if (ins_nr > 0) { | |
1a93c36a | 4666 | ret = copy_items(trans, inode, dst_path, path, |
0e56315c | 4667 | start_slot, ins_nr, 1, 0); |
36283bf7 FM |
4668 | if (ret < 0) |
4669 | return ret; | |
4670 | ins_nr = 0; | |
4671 | } | |
4672 | ret = btrfs_next_leaf(root, path); | |
4673 | if (ret < 0) | |
4674 | return ret; | |
4675 | else if (ret > 0) | |
4676 | break; | |
4677 | continue; | |
4678 | } | |
4679 | ||
4680 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
4681 | if (key.objectid != ino || key.type != BTRFS_XATTR_ITEM_KEY) | |
4682 | break; | |
4683 | ||
4684 | if (ins_nr == 0) | |
4685 | start_slot = slot; | |
4686 | ins_nr++; | |
4687 | path->slots[0]++; | |
f2f121ab | 4688 | found_xattrs = true; |
36283bf7 FM |
4689 | cond_resched(); |
4690 | } | |
4691 | if (ins_nr > 0) { | |
1a93c36a | 4692 | ret = copy_items(trans, inode, dst_path, path, |
0e56315c | 4693 | start_slot, ins_nr, 1, 0); |
36283bf7 FM |
4694 | if (ret < 0) |
4695 | return ret; | |
4696 | } | |
4697 | ||
f2f121ab FM |
4698 | if (!found_xattrs) |
4699 | set_bit(BTRFS_INODE_NO_XATTRS, &inode->runtime_flags); | |
4700 | ||
36283bf7 FM |
4701 | return 0; |
4702 | } | |
4703 | ||
a89ca6f2 | 4704 | /* |
0e56315c FM |
4705 | * When using the NO_HOLES feature if we punched a hole that causes the |
4706 | * deletion of entire leafs or all the extent items of the first leaf (the one | |
4707 | * that contains the inode item and references) we may end up not processing | |
4708 | * any extents, because there are no leafs with a generation matching the | |
4709 | * current transaction that have extent items for our inode. So we need to find | |
4710 | * if any holes exist and then log them. We also need to log holes after any | |
4711 | * truncate operation that changes the inode's size. | |
a89ca6f2 | 4712 | */ |
0e56315c FM |
4713 | static int btrfs_log_holes(struct btrfs_trans_handle *trans, |
4714 | struct btrfs_root *root, | |
4715 | struct btrfs_inode *inode, | |
7af59743 | 4716 | struct btrfs_path *path) |
a89ca6f2 | 4717 | { |
0b246afa | 4718 | struct btrfs_fs_info *fs_info = root->fs_info; |
a89ca6f2 | 4719 | struct btrfs_key key; |
a0308dd7 NB |
4720 | const u64 ino = btrfs_ino(inode); |
4721 | const u64 i_size = i_size_read(&inode->vfs_inode); | |
7af59743 | 4722 | u64 prev_extent_end = 0; |
0e56315c | 4723 | int ret; |
a89ca6f2 | 4724 | |
0e56315c | 4725 | if (!btrfs_fs_incompat(fs_info, NO_HOLES) || i_size == 0) |
a89ca6f2 FM |
4726 | return 0; |
4727 | ||
4728 | key.objectid = ino; | |
4729 | key.type = BTRFS_EXTENT_DATA_KEY; | |
7af59743 | 4730 | key.offset = 0; |
a89ca6f2 FM |
4731 | |
4732 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
a89ca6f2 FM |
4733 | if (ret < 0) |
4734 | return ret; | |
4735 | ||
0e56315c | 4736 | while (true) { |
0e56315c | 4737 | struct extent_buffer *leaf = path->nodes[0]; |
a89ca6f2 | 4738 | |
0e56315c FM |
4739 | if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) { |
4740 | ret = btrfs_next_leaf(root, path); | |
4741 | if (ret < 0) | |
4742 | return ret; | |
4743 | if (ret > 0) { | |
4744 | ret = 0; | |
4745 | break; | |
4746 | } | |
4747 | leaf = path->nodes[0]; | |
4748 | } | |
4749 | ||
4750 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
4751 | if (key.objectid != ino || key.type != BTRFS_EXTENT_DATA_KEY) | |
4752 | break; | |
4753 | ||
4754 | /* We have a hole, log it. */ | |
4755 | if (prev_extent_end < key.offset) { | |
7af59743 | 4756 | const u64 hole_len = key.offset - prev_extent_end; |
0e56315c FM |
4757 | |
4758 | /* | |
4759 | * Release the path to avoid deadlocks with other code | |
4760 | * paths that search the root while holding locks on | |
4761 | * leafs from the log root. | |
4762 | */ | |
4763 | btrfs_release_path(path); | |
4764 | ret = btrfs_insert_file_extent(trans, root->log_root, | |
4765 | ino, prev_extent_end, 0, | |
4766 | 0, hole_len, 0, hole_len, | |
4767 | 0, 0, 0); | |
4768 | if (ret < 0) | |
4769 | return ret; | |
4770 | ||
4771 | /* | |
4772 | * Search for the same key again in the root. Since it's | |
4773 | * an extent item and we are holding the inode lock, the | |
4774 | * key must still exist. If it doesn't just emit warning | |
4775 | * and return an error to fall back to a transaction | |
4776 | * commit. | |
4777 | */ | |
4778 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
4779 | if (ret < 0) | |
4780 | return ret; | |
4781 | if (WARN_ON(ret > 0)) | |
4782 | return -ENOENT; | |
4783 | leaf = path->nodes[0]; | |
4784 | } | |
a89ca6f2 | 4785 | |
7af59743 | 4786 | prev_extent_end = btrfs_file_extent_end(path); |
0e56315c FM |
4787 | path->slots[0]++; |
4788 | cond_resched(); | |
a89ca6f2 | 4789 | } |
a89ca6f2 | 4790 | |
7af59743 | 4791 | if (prev_extent_end < i_size) { |
0e56315c | 4792 | u64 hole_len; |
a89ca6f2 | 4793 | |
0e56315c | 4794 | btrfs_release_path(path); |
7af59743 | 4795 | hole_len = ALIGN(i_size - prev_extent_end, fs_info->sectorsize); |
0e56315c FM |
4796 | ret = btrfs_insert_file_extent(trans, root->log_root, |
4797 | ino, prev_extent_end, 0, 0, | |
4798 | hole_len, 0, hole_len, | |
4799 | 0, 0, 0); | |
4800 | if (ret < 0) | |
4801 | return ret; | |
4802 | } | |
4803 | ||
4804 | return 0; | |
a89ca6f2 FM |
4805 | } |
4806 | ||
56f23fdb FM |
4807 | /* |
4808 | * When we are logging a new inode X, check if it doesn't have a reference that | |
4809 | * matches the reference from some other inode Y created in a past transaction | |
4810 | * and that was renamed in the current transaction. If we don't do this, then at | |
4811 | * log replay time we can lose inode Y (and all its files if it's a directory): | |
4812 | * | |
4813 | * mkdir /mnt/x | |
4814 | * echo "hello world" > /mnt/x/foobar | |
4815 | * sync | |
4816 | * mv /mnt/x /mnt/y | |
4817 | * mkdir /mnt/x # or touch /mnt/x | |
4818 | * xfs_io -c fsync /mnt/x | |
4819 | * <power fail> | |
4820 | * mount fs, trigger log replay | |
4821 | * | |
4822 | * After the log replay procedure, we would lose the first directory and all its | |
4823 | * files (file foobar). | |
4824 | * For the case where inode Y is not a directory we simply end up losing it: | |
4825 | * | |
4826 | * echo "123" > /mnt/foo | |
4827 | * sync | |
4828 | * mv /mnt/foo /mnt/bar | |
4829 | * echo "abc" > /mnt/foo | |
4830 | * xfs_io -c fsync /mnt/foo | |
4831 | * <power fail> | |
4832 | * | |
4833 | * We also need this for cases where a snapshot entry is replaced by some other | |
4834 | * entry (file or directory) otherwise we end up with an unreplayable log due to | |
4835 | * attempts to delete the snapshot entry (entry of type BTRFS_ROOT_ITEM_KEY) as | |
4836 | * if it were a regular entry: | |
4837 | * | |
4838 | * mkdir /mnt/x | |
4839 | * btrfs subvolume snapshot /mnt /mnt/x/snap | |
4840 | * btrfs subvolume delete /mnt/x/snap | |
4841 | * rmdir /mnt/x | |
4842 | * mkdir /mnt/x | |
4843 | * fsync /mnt/x or fsync some new file inside it | |
4844 | * <power fail> | |
4845 | * | |
4846 | * The snapshot delete, rmdir of x, mkdir of a new x and the fsync all happen in | |
4847 | * the same transaction. | |
4848 | */ | |
4849 | static int btrfs_check_ref_name_override(struct extent_buffer *eb, | |
4850 | const int slot, | |
4851 | const struct btrfs_key *key, | |
4791c8f1 | 4852 | struct btrfs_inode *inode, |
a3baaf0d | 4853 | u64 *other_ino, u64 *other_parent) |
56f23fdb FM |
4854 | { |
4855 | int ret; | |
4856 | struct btrfs_path *search_path; | |
4857 | char *name = NULL; | |
4858 | u32 name_len = 0; | |
4859 | u32 item_size = btrfs_item_size_nr(eb, slot); | |
4860 | u32 cur_offset = 0; | |
4861 | unsigned long ptr = btrfs_item_ptr_offset(eb, slot); | |
4862 | ||
4863 | search_path = btrfs_alloc_path(); | |
4864 | if (!search_path) | |
4865 | return -ENOMEM; | |
4866 | search_path->search_commit_root = 1; | |
4867 | search_path->skip_locking = 1; | |
4868 | ||
4869 | while (cur_offset < item_size) { | |
4870 | u64 parent; | |
4871 | u32 this_name_len; | |
4872 | u32 this_len; | |
4873 | unsigned long name_ptr; | |
4874 | struct btrfs_dir_item *di; | |
4875 | ||
4876 | if (key->type == BTRFS_INODE_REF_KEY) { | |
4877 | struct btrfs_inode_ref *iref; | |
4878 | ||
4879 | iref = (struct btrfs_inode_ref *)(ptr + cur_offset); | |
4880 | parent = key->offset; | |
4881 | this_name_len = btrfs_inode_ref_name_len(eb, iref); | |
4882 | name_ptr = (unsigned long)(iref + 1); | |
4883 | this_len = sizeof(*iref) + this_name_len; | |
4884 | } else { | |
4885 | struct btrfs_inode_extref *extref; | |
4886 | ||
4887 | extref = (struct btrfs_inode_extref *)(ptr + | |
4888 | cur_offset); | |
4889 | parent = btrfs_inode_extref_parent(eb, extref); | |
4890 | this_name_len = btrfs_inode_extref_name_len(eb, extref); | |
4891 | name_ptr = (unsigned long)&extref->name; | |
4892 | this_len = sizeof(*extref) + this_name_len; | |
4893 | } | |
4894 | ||
4895 | if (this_name_len > name_len) { | |
4896 | char *new_name; | |
4897 | ||
4898 | new_name = krealloc(name, this_name_len, GFP_NOFS); | |
4899 | if (!new_name) { | |
4900 | ret = -ENOMEM; | |
4901 | goto out; | |
4902 | } | |
4903 | name_len = this_name_len; | |
4904 | name = new_name; | |
4905 | } | |
4906 | ||
4907 | read_extent_buffer(eb, name, name_ptr, this_name_len); | |
4791c8f1 NB |
4908 | di = btrfs_lookup_dir_item(NULL, inode->root, search_path, |
4909 | parent, name, this_name_len, 0); | |
56f23fdb | 4910 | if (di && !IS_ERR(di)) { |
44f714da FM |
4911 | struct btrfs_key di_key; |
4912 | ||
4913 | btrfs_dir_item_key_to_cpu(search_path->nodes[0], | |
4914 | di, &di_key); | |
4915 | if (di_key.type == BTRFS_INODE_ITEM_KEY) { | |
6b5fc433 FM |
4916 | if (di_key.objectid != key->objectid) { |
4917 | ret = 1; | |
4918 | *other_ino = di_key.objectid; | |
a3baaf0d | 4919 | *other_parent = parent; |
6b5fc433 FM |
4920 | } else { |
4921 | ret = 0; | |
4922 | } | |
44f714da FM |
4923 | } else { |
4924 | ret = -EAGAIN; | |
4925 | } | |
56f23fdb FM |
4926 | goto out; |
4927 | } else if (IS_ERR(di)) { | |
4928 | ret = PTR_ERR(di); | |
4929 | goto out; | |
4930 | } | |
4931 | btrfs_release_path(search_path); | |
4932 | ||
4933 | cur_offset += this_len; | |
4934 | } | |
4935 | ret = 0; | |
4936 | out: | |
4937 | btrfs_free_path(search_path); | |
4938 | kfree(name); | |
4939 | return ret; | |
4940 | } | |
4941 | ||
6b5fc433 FM |
4942 | struct btrfs_ino_list { |
4943 | u64 ino; | |
a3baaf0d | 4944 | u64 parent; |
6b5fc433 FM |
4945 | struct list_head list; |
4946 | }; | |
4947 | ||
4948 | static int log_conflicting_inodes(struct btrfs_trans_handle *trans, | |
4949 | struct btrfs_root *root, | |
4950 | struct btrfs_path *path, | |
4951 | struct btrfs_log_ctx *ctx, | |
a3baaf0d | 4952 | u64 ino, u64 parent) |
6b5fc433 FM |
4953 | { |
4954 | struct btrfs_ino_list *ino_elem; | |
4955 | LIST_HEAD(inode_list); | |
4956 | int ret = 0; | |
4957 | ||
4958 | ino_elem = kmalloc(sizeof(*ino_elem), GFP_NOFS); | |
4959 | if (!ino_elem) | |
4960 | return -ENOMEM; | |
4961 | ino_elem->ino = ino; | |
a3baaf0d | 4962 | ino_elem->parent = parent; |
6b5fc433 FM |
4963 | list_add_tail(&ino_elem->list, &inode_list); |
4964 | ||
4965 | while (!list_empty(&inode_list)) { | |
4966 | struct btrfs_fs_info *fs_info = root->fs_info; | |
4967 | struct btrfs_key key; | |
4968 | struct inode *inode; | |
4969 | ||
4970 | ino_elem = list_first_entry(&inode_list, struct btrfs_ino_list, | |
4971 | list); | |
4972 | ino = ino_elem->ino; | |
a3baaf0d | 4973 | parent = ino_elem->parent; |
6b5fc433 FM |
4974 | list_del(&ino_elem->list); |
4975 | kfree(ino_elem); | |
4976 | if (ret) | |
4977 | continue; | |
4978 | ||
4979 | btrfs_release_path(path); | |
4980 | ||
0202e83f | 4981 | inode = btrfs_iget(fs_info->sb, ino, root); |
6b5fc433 FM |
4982 | /* |
4983 | * If the other inode that had a conflicting dir entry was | |
a3baaf0d FM |
4984 | * deleted in the current transaction, we need to log its parent |
4985 | * directory. | |
6b5fc433 FM |
4986 | */ |
4987 | if (IS_ERR(inode)) { | |
4988 | ret = PTR_ERR(inode); | |
a3baaf0d | 4989 | if (ret == -ENOENT) { |
0202e83f | 4990 | inode = btrfs_iget(fs_info->sb, parent, root); |
a3baaf0d FM |
4991 | if (IS_ERR(inode)) { |
4992 | ret = PTR_ERR(inode); | |
4993 | } else { | |
4994 | ret = btrfs_log_inode(trans, root, | |
4995 | BTRFS_I(inode), | |
4996 | LOG_OTHER_INODE_ALL, | |
48778179 | 4997 | ctx); |
410f954c | 4998 | btrfs_add_delayed_iput(inode); |
a3baaf0d FM |
4999 | } |
5000 | } | |
6b5fc433 FM |
5001 | continue; |
5002 | } | |
b5e4ff9d FM |
5003 | /* |
5004 | * If the inode was already logged skip it - otherwise we can | |
5005 | * hit an infinite loop. Example: | |
5006 | * | |
5007 | * From the commit root (previous transaction) we have the | |
5008 | * following inodes: | |
5009 | * | |
5010 | * inode 257 a directory | |
5011 | * inode 258 with references "zz" and "zz_link" on inode 257 | |
5012 | * inode 259 with reference "a" on inode 257 | |
5013 | * | |
5014 | * And in the current (uncommitted) transaction we have: | |
5015 | * | |
5016 | * inode 257 a directory, unchanged | |
5017 | * inode 258 with references "a" and "a2" on inode 257 | |
5018 | * inode 259 with reference "zz_link" on inode 257 | |
5019 | * inode 261 with reference "zz" on inode 257 | |
5020 | * | |
5021 | * When logging inode 261 the following infinite loop could | |
5022 | * happen if we don't skip already logged inodes: | |
5023 | * | |
5024 | * - we detect inode 258 as a conflicting inode, with inode 261 | |
5025 | * on reference "zz", and log it; | |
5026 | * | |
5027 | * - we detect inode 259 as a conflicting inode, with inode 258 | |
5028 | * on reference "a", and log it; | |
5029 | * | |
5030 | * - we detect inode 258 as a conflicting inode, with inode 259 | |
5031 | * on reference "zz_link", and log it - again! After this we | |
5032 | * repeat the above steps forever. | |
5033 | */ | |
5034 | spin_lock(&BTRFS_I(inode)->lock); | |
5035 | /* | |
5036 | * Check the inode's logged_trans only instead of | |
5037 | * btrfs_inode_in_log(). This is because the last_log_commit of | |
5038 | * the inode is not updated when we only log that it exists and | |
260db43c | 5039 | * it has the full sync bit set (see btrfs_log_inode()). |
b5e4ff9d FM |
5040 | */ |
5041 | if (BTRFS_I(inode)->logged_trans == trans->transid) { | |
5042 | spin_unlock(&BTRFS_I(inode)->lock); | |
5043 | btrfs_add_delayed_iput(inode); | |
5044 | continue; | |
5045 | } | |
5046 | spin_unlock(&BTRFS_I(inode)->lock); | |
6b5fc433 FM |
5047 | /* |
5048 | * We are safe logging the other inode without acquiring its | |
5049 | * lock as long as we log with the LOG_INODE_EXISTS mode. We | |
5050 | * are safe against concurrent renames of the other inode as | |
5051 | * well because during a rename we pin the log and update the | |
5052 | * log with the new name before we unpin it. | |
5053 | */ | |
5054 | ret = btrfs_log_inode(trans, root, BTRFS_I(inode), | |
48778179 | 5055 | LOG_OTHER_INODE, ctx); |
6b5fc433 | 5056 | if (ret) { |
410f954c | 5057 | btrfs_add_delayed_iput(inode); |
6b5fc433 FM |
5058 | continue; |
5059 | } | |
5060 | ||
5061 | key.objectid = ino; | |
5062 | key.type = BTRFS_INODE_REF_KEY; | |
5063 | key.offset = 0; | |
5064 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
5065 | if (ret < 0) { | |
410f954c | 5066 | btrfs_add_delayed_iput(inode); |
6b5fc433 FM |
5067 | continue; |
5068 | } | |
5069 | ||
5070 | while (true) { | |
5071 | struct extent_buffer *leaf = path->nodes[0]; | |
5072 | int slot = path->slots[0]; | |
5073 | u64 other_ino = 0; | |
a3baaf0d | 5074 | u64 other_parent = 0; |
6b5fc433 FM |
5075 | |
5076 | if (slot >= btrfs_header_nritems(leaf)) { | |
5077 | ret = btrfs_next_leaf(root, path); | |
5078 | if (ret < 0) { | |
5079 | break; | |
5080 | } else if (ret > 0) { | |
5081 | ret = 0; | |
5082 | break; | |
5083 | } | |
5084 | continue; | |
5085 | } | |
5086 | ||
5087 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
5088 | if (key.objectid != ino || | |
5089 | (key.type != BTRFS_INODE_REF_KEY && | |
5090 | key.type != BTRFS_INODE_EXTREF_KEY)) { | |
5091 | ret = 0; | |
5092 | break; | |
5093 | } | |
5094 | ||
5095 | ret = btrfs_check_ref_name_override(leaf, slot, &key, | |
a3baaf0d FM |
5096 | BTRFS_I(inode), &other_ino, |
5097 | &other_parent); | |
6b5fc433 FM |
5098 | if (ret < 0) |
5099 | break; | |
5100 | if (ret > 0) { | |
5101 | ino_elem = kmalloc(sizeof(*ino_elem), GFP_NOFS); | |
5102 | if (!ino_elem) { | |
5103 | ret = -ENOMEM; | |
5104 | break; | |
5105 | } | |
5106 | ino_elem->ino = other_ino; | |
a3baaf0d | 5107 | ino_elem->parent = other_parent; |
6b5fc433 FM |
5108 | list_add_tail(&ino_elem->list, &inode_list); |
5109 | ret = 0; | |
5110 | } | |
5111 | path->slots[0]++; | |
5112 | } | |
410f954c | 5113 | btrfs_add_delayed_iput(inode); |
6b5fc433 FM |
5114 | } |
5115 | ||
5116 | return ret; | |
5117 | } | |
5118 | ||
da447009 FM |
5119 | static int copy_inode_items_to_log(struct btrfs_trans_handle *trans, |
5120 | struct btrfs_inode *inode, | |
5121 | struct btrfs_key *min_key, | |
5122 | const struct btrfs_key *max_key, | |
5123 | struct btrfs_path *path, | |
5124 | struct btrfs_path *dst_path, | |
5125 | const u64 logged_isize, | |
5126 | const bool recursive_logging, | |
5127 | const int inode_only, | |
5128 | struct btrfs_log_ctx *ctx, | |
5129 | bool *need_log_inode_item) | |
5130 | { | |
5131 | struct btrfs_root *root = inode->root; | |
5132 | int ins_start_slot = 0; | |
5133 | int ins_nr = 0; | |
5134 | int ret; | |
5135 | ||
5136 | while (1) { | |
5137 | ret = btrfs_search_forward(root, min_key, path, trans->transid); | |
5138 | if (ret < 0) | |
5139 | return ret; | |
5140 | if (ret > 0) { | |
5141 | ret = 0; | |
5142 | break; | |
5143 | } | |
5144 | again: | |
5145 | /* Note, ins_nr might be > 0 here, cleanup outside the loop */ | |
5146 | if (min_key->objectid != max_key->objectid) | |
5147 | break; | |
5148 | if (min_key->type > max_key->type) | |
5149 | break; | |
5150 | ||
5151 | if (min_key->type == BTRFS_INODE_ITEM_KEY) | |
5152 | *need_log_inode_item = false; | |
5153 | ||
5154 | if ((min_key->type == BTRFS_INODE_REF_KEY || | |
5155 | min_key->type == BTRFS_INODE_EXTREF_KEY) && | |
5156 | inode->generation == trans->transid && | |
5157 | !recursive_logging) { | |
5158 | u64 other_ino = 0; | |
5159 | u64 other_parent = 0; | |
5160 | ||
5161 | ret = btrfs_check_ref_name_override(path->nodes[0], | |
5162 | path->slots[0], min_key, inode, | |
5163 | &other_ino, &other_parent); | |
5164 | if (ret < 0) { | |
5165 | return ret; | |
5166 | } else if (ret > 0 && ctx && | |
5167 | other_ino != btrfs_ino(BTRFS_I(ctx->inode))) { | |
5168 | if (ins_nr > 0) { | |
5169 | ins_nr++; | |
5170 | } else { | |
5171 | ins_nr = 1; | |
5172 | ins_start_slot = path->slots[0]; | |
5173 | } | |
5174 | ret = copy_items(trans, inode, dst_path, path, | |
5175 | ins_start_slot, ins_nr, | |
5176 | inode_only, logged_isize); | |
5177 | if (ret < 0) | |
5178 | return ret; | |
5179 | ins_nr = 0; | |
5180 | ||
5181 | ret = log_conflicting_inodes(trans, root, path, | |
5182 | ctx, other_ino, other_parent); | |
5183 | if (ret) | |
5184 | return ret; | |
5185 | btrfs_release_path(path); | |
5186 | goto next_key; | |
5187 | } | |
5188 | } | |
5189 | ||
5190 | /* Skip xattrs, we log them later with btrfs_log_all_xattrs() */ | |
5191 | if (min_key->type == BTRFS_XATTR_ITEM_KEY) { | |
5192 | if (ins_nr == 0) | |
5193 | goto next_slot; | |
5194 | ret = copy_items(trans, inode, dst_path, path, | |
5195 | ins_start_slot, | |
5196 | ins_nr, inode_only, logged_isize); | |
5197 | if (ret < 0) | |
5198 | return ret; | |
5199 | ins_nr = 0; | |
5200 | goto next_slot; | |
5201 | } | |
5202 | ||
5203 | if (ins_nr && ins_start_slot + ins_nr == path->slots[0]) { | |
5204 | ins_nr++; | |
5205 | goto next_slot; | |
5206 | } else if (!ins_nr) { | |
5207 | ins_start_slot = path->slots[0]; | |
5208 | ins_nr = 1; | |
5209 | goto next_slot; | |
5210 | } | |
5211 | ||
5212 | ret = copy_items(trans, inode, dst_path, path, ins_start_slot, | |
5213 | ins_nr, inode_only, logged_isize); | |
5214 | if (ret < 0) | |
5215 | return ret; | |
5216 | ins_nr = 1; | |
5217 | ins_start_slot = path->slots[0]; | |
5218 | next_slot: | |
5219 | path->slots[0]++; | |
5220 | if (path->slots[0] < btrfs_header_nritems(path->nodes[0])) { | |
5221 | btrfs_item_key_to_cpu(path->nodes[0], min_key, | |
5222 | path->slots[0]); | |
5223 | goto again; | |
5224 | } | |
5225 | if (ins_nr) { | |
5226 | ret = copy_items(trans, inode, dst_path, path, | |
5227 | ins_start_slot, ins_nr, inode_only, | |
5228 | logged_isize); | |
5229 | if (ret < 0) | |
5230 | return ret; | |
5231 | ins_nr = 0; | |
5232 | } | |
5233 | btrfs_release_path(path); | |
5234 | next_key: | |
5235 | if (min_key->offset < (u64)-1) { | |
5236 | min_key->offset++; | |
5237 | } else if (min_key->type < max_key->type) { | |
5238 | min_key->type++; | |
5239 | min_key->offset = 0; | |
5240 | } else { | |
5241 | break; | |
5242 | } | |
5243 | } | |
5244 | if (ins_nr) | |
5245 | ret = copy_items(trans, inode, dst_path, path, ins_start_slot, | |
5246 | ins_nr, inode_only, logged_isize); | |
5247 | ||
5248 | return ret; | |
5249 | } | |
5250 | ||
e02119d5 CM |
5251 | /* log a single inode in the tree log. |
5252 | * At least one parent directory for this inode must exist in the tree | |
5253 | * or be logged already. | |
5254 | * | |
5255 | * Any items from this inode changed by the current transaction are copied | |
5256 | * to the log tree. An extra reference is taken on any extents in this | |
5257 | * file, allowing us to avoid a whole pile of corner cases around logging | |
5258 | * blocks that have been removed from the tree. | |
5259 | * | |
5260 | * See LOG_INODE_ALL and related defines for a description of what inode_only | |
5261 | * does. | |
5262 | * | |
5263 | * This handles both files and directories. | |
5264 | */ | |
12fcfd22 | 5265 | static int btrfs_log_inode(struct btrfs_trans_handle *trans, |
a59108a7 | 5266 | struct btrfs_root *root, struct btrfs_inode *inode, |
49dae1bc | 5267 | int inode_only, |
8407f553 | 5268 | struct btrfs_log_ctx *ctx) |
e02119d5 CM |
5269 | { |
5270 | struct btrfs_path *path; | |
5271 | struct btrfs_path *dst_path; | |
5272 | struct btrfs_key min_key; | |
5273 | struct btrfs_key max_key; | |
5274 | struct btrfs_root *log = root->log_root; | |
4a500fd1 | 5275 | int err = 0; |
8c8648dd | 5276 | int ret = 0; |
5dc562c5 | 5277 | bool fast_search = false; |
a59108a7 NB |
5278 | u64 ino = btrfs_ino(inode); |
5279 | struct extent_map_tree *em_tree = &inode->extent_tree; | |
1a4bcf47 | 5280 | u64 logged_isize = 0; |
e4545de5 | 5281 | bool need_log_inode_item = true; |
9a8fca62 | 5282 | bool xattrs_logged = false; |
a3baaf0d | 5283 | bool recursive_logging = false; |
e02119d5 | 5284 | |
e02119d5 | 5285 | path = btrfs_alloc_path(); |
5df67083 TI |
5286 | if (!path) |
5287 | return -ENOMEM; | |
e02119d5 | 5288 | dst_path = btrfs_alloc_path(); |
5df67083 TI |
5289 | if (!dst_path) { |
5290 | btrfs_free_path(path); | |
5291 | return -ENOMEM; | |
5292 | } | |
e02119d5 | 5293 | |
33345d01 | 5294 | min_key.objectid = ino; |
e02119d5 CM |
5295 | min_key.type = BTRFS_INODE_ITEM_KEY; |
5296 | min_key.offset = 0; | |
5297 | ||
33345d01 | 5298 | max_key.objectid = ino; |
12fcfd22 | 5299 | |
12fcfd22 | 5300 | |
5dc562c5 | 5301 | /* today the code can only do partial logging of directories */ |
a59108a7 | 5302 | if (S_ISDIR(inode->vfs_inode.i_mode) || |
5269b67e | 5303 | (!test_bit(BTRFS_INODE_NEEDS_FULL_SYNC, |
a59108a7 | 5304 | &inode->runtime_flags) && |
781feef7 | 5305 | inode_only >= LOG_INODE_EXISTS)) |
e02119d5 CM |
5306 | max_key.type = BTRFS_XATTR_ITEM_KEY; |
5307 | else | |
5308 | max_key.type = (u8)-1; | |
5309 | max_key.offset = (u64)-1; | |
5310 | ||
2c2c452b | 5311 | /* |
5aa7d1a7 FM |
5312 | * Only run delayed items if we are a directory. We want to make sure |
5313 | * all directory indexes hit the fs/subvolume tree so we can find them | |
5314 | * and figure out which index ranges have to be logged. | |
5315 | * | |
8c8648dd FM |
5316 | * Otherwise commit the delayed inode only if the full sync flag is set, |
5317 | * as we want to make sure an up to date version is in the subvolume | |
5318 | * tree so copy_inode_items_to_log() / copy_items() can find it and copy | |
5319 | * it to the log tree. For a non full sync, we always log the inode item | |
5320 | * based on the in-memory struct btrfs_inode which is always up to date. | |
2c2c452b | 5321 | */ |
5aa7d1a7 | 5322 | if (S_ISDIR(inode->vfs_inode.i_mode)) |
a59108a7 | 5323 | ret = btrfs_commit_inode_delayed_items(trans, inode); |
8c8648dd | 5324 | else if (test_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &inode->runtime_flags)) |
a59108a7 | 5325 | ret = btrfs_commit_inode_delayed_inode(inode); |
2c2c452b FM |
5326 | |
5327 | if (ret) { | |
5328 | btrfs_free_path(path); | |
5329 | btrfs_free_path(dst_path); | |
5330 | return ret; | |
16cdcec7 MX |
5331 | } |
5332 | ||
a3baaf0d FM |
5333 | if (inode_only == LOG_OTHER_INODE || inode_only == LOG_OTHER_INODE_ALL) { |
5334 | recursive_logging = true; | |
5335 | if (inode_only == LOG_OTHER_INODE) | |
5336 | inode_only = LOG_INODE_EXISTS; | |
5337 | else | |
5338 | inode_only = LOG_INODE_ALL; | |
a59108a7 | 5339 | mutex_lock_nested(&inode->log_mutex, SINGLE_DEPTH_NESTING); |
781feef7 | 5340 | } else { |
a59108a7 | 5341 | mutex_lock(&inode->log_mutex); |
781feef7 | 5342 | } |
e02119d5 | 5343 | |
64d6b281 FM |
5344 | /* |
5345 | * This is for cases where logging a directory could result in losing a | |
5346 | * a file after replaying the log. For example, if we move a file from a | |
5347 | * directory A to a directory B, then fsync directory A, we have no way | |
5348 | * to known the file was moved from A to B, so logging just A would | |
5349 | * result in losing the file after a log replay. | |
5350 | */ | |
5351 | if (S_ISDIR(inode->vfs_inode.i_mode) && | |
5352 | inode_only == LOG_INODE_ALL && | |
5353 | inode->last_unlink_trans >= trans->transid) { | |
5354 | btrfs_set_log_full_commit(trans); | |
5355 | err = 1; | |
5356 | goto out_unlock; | |
5357 | } | |
5358 | ||
e02119d5 CM |
5359 | /* |
5360 | * a brute force approach to making sure we get the most uptodate | |
5361 | * copies of everything. | |
5362 | */ | |
a59108a7 | 5363 | if (S_ISDIR(inode->vfs_inode.i_mode)) { |
e02119d5 CM |
5364 | int max_key_type = BTRFS_DIR_LOG_INDEX_KEY; |
5365 | ||
ab12313a | 5366 | clear_bit(BTRFS_INODE_COPY_EVERYTHING, &inode->runtime_flags); |
4f764e51 FM |
5367 | if (inode_only == LOG_INODE_EXISTS) |
5368 | max_key_type = BTRFS_XATTR_ITEM_KEY; | |
33345d01 | 5369 | ret = drop_objectid_items(trans, log, path, ino, max_key_type); |
e02119d5 | 5370 | } else { |
1a4bcf47 FM |
5371 | if (inode_only == LOG_INODE_EXISTS) { |
5372 | /* | |
5373 | * Make sure the new inode item we write to the log has | |
5374 | * the same isize as the current one (if it exists). | |
5375 | * This is necessary to prevent data loss after log | |
5376 | * replay, and also to prevent doing a wrong expanding | |
5377 | * truncate - for e.g. create file, write 4K into offset | |
5378 | * 0, fsync, write 4K into offset 4096, add hard link, | |
5379 | * fsync some other file (to sync log), power fail - if | |
5380 | * we use the inode's current i_size, after log replay | |
5381 | * we get a 8Kb file, with the last 4Kb extent as a hole | |
5382 | * (zeroes), as if an expanding truncate happened, | |
5383 | * instead of getting a file of 4Kb only. | |
5384 | */ | |
a59108a7 | 5385 | err = logged_inode_size(log, inode, path, &logged_isize); |
1a4bcf47 FM |
5386 | if (err) |
5387 | goto out_unlock; | |
5388 | } | |
a742994a | 5389 | if (test_bit(BTRFS_INODE_NEEDS_FULL_SYNC, |
a59108a7 | 5390 | &inode->runtime_flags)) { |
a742994a | 5391 | if (inode_only == LOG_INODE_EXISTS) { |
4f764e51 | 5392 | max_key.type = BTRFS_XATTR_ITEM_KEY; |
a742994a FM |
5393 | ret = drop_objectid_items(trans, log, path, ino, |
5394 | max_key.type); | |
5395 | } else { | |
5396 | clear_bit(BTRFS_INODE_NEEDS_FULL_SYNC, | |
a59108a7 | 5397 | &inode->runtime_flags); |
a742994a | 5398 | clear_bit(BTRFS_INODE_COPY_EVERYTHING, |
a59108a7 | 5399 | &inode->runtime_flags); |
28ed1345 CM |
5400 | while(1) { |
5401 | ret = btrfs_truncate_inode_items(trans, | |
50743398 | 5402 | log, inode, 0, 0); |
28ed1345 CM |
5403 | if (ret != -EAGAIN) |
5404 | break; | |
5405 | } | |
a742994a | 5406 | } |
4f764e51 | 5407 | } else if (test_and_clear_bit(BTRFS_INODE_COPY_EVERYTHING, |
a59108a7 | 5408 | &inode->runtime_flags) || |
6cfab851 | 5409 | inode_only == LOG_INODE_EXISTS) { |
4f764e51 | 5410 | if (inode_only == LOG_INODE_ALL) |
183f37fa | 5411 | fast_search = true; |
4f764e51 | 5412 | max_key.type = BTRFS_XATTR_ITEM_KEY; |
5dc562c5 | 5413 | ret = drop_objectid_items(trans, log, path, ino, |
e9976151 | 5414 | max_key.type); |
a95249b3 JB |
5415 | } else { |
5416 | if (inode_only == LOG_INODE_ALL) | |
5417 | fast_search = true; | |
a95249b3 | 5418 | goto log_extents; |
5dc562c5 | 5419 | } |
a95249b3 | 5420 | |
e02119d5 | 5421 | } |
4a500fd1 YZ |
5422 | if (ret) { |
5423 | err = ret; | |
5424 | goto out_unlock; | |
5425 | } | |
e02119d5 | 5426 | |
da447009 FM |
5427 | err = copy_inode_items_to_log(trans, inode, &min_key, &max_key, |
5428 | path, dst_path, logged_isize, | |
7af59743 FM |
5429 | recursive_logging, inode_only, ctx, |
5430 | &need_log_inode_item); | |
da447009 FM |
5431 | if (err) |
5432 | goto out_unlock; | |
5dc562c5 | 5433 | |
36283bf7 FM |
5434 | btrfs_release_path(path); |
5435 | btrfs_release_path(dst_path); | |
a59108a7 | 5436 | err = btrfs_log_all_xattrs(trans, root, inode, path, dst_path); |
36283bf7 FM |
5437 | if (err) |
5438 | goto out_unlock; | |
9a8fca62 | 5439 | xattrs_logged = true; |
a89ca6f2 FM |
5440 | if (max_key.type >= BTRFS_EXTENT_DATA_KEY && !fast_search) { |
5441 | btrfs_release_path(path); | |
5442 | btrfs_release_path(dst_path); | |
7af59743 | 5443 | err = btrfs_log_holes(trans, root, inode, path); |
a89ca6f2 FM |
5444 | if (err) |
5445 | goto out_unlock; | |
5446 | } | |
a95249b3 | 5447 | log_extents: |
f3b15ccd JB |
5448 | btrfs_release_path(path); |
5449 | btrfs_release_path(dst_path); | |
e4545de5 | 5450 | if (need_log_inode_item) { |
a59108a7 | 5451 | err = log_inode_item(trans, log, dst_path, inode); |
9a8fca62 FM |
5452 | if (!err && !xattrs_logged) { |
5453 | err = btrfs_log_all_xattrs(trans, root, inode, path, | |
5454 | dst_path); | |
5455 | btrfs_release_path(path); | |
5456 | } | |
e4545de5 FM |
5457 | if (err) |
5458 | goto out_unlock; | |
5459 | } | |
5dc562c5 | 5460 | if (fast_search) { |
a59108a7 | 5461 | ret = btrfs_log_changed_extents(trans, root, inode, dst_path, |
48778179 | 5462 | ctx); |
5dc562c5 JB |
5463 | if (ret) { |
5464 | err = ret; | |
5465 | goto out_unlock; | |
5466 | } | |
d006a048 | 5467 | } else if (inode_only == LOG_INODE_ALL) { |
06d3d22b LB |
5468 | struct extent_map *em, *n; |
5469 | ||
49dae1bc | 5470 | write_lock(&em_tree->lock); |
48778179 FM |
5471 | list_for_each_entry_safe(em, n, &em_tree->modified_extents, list) |
5472 | list_del_init(&em->list); | |
49dae1bc | 5473 | write_unlock(&em_tree->lock); |
5dc562c5 JB |
5474 | } |
5475 | ||
a59108a7 NB |
5476 | if (inode_only == LOG_INODE_ALL && S_ISDIR(inode->vfs_inode.i_mode)) { |
5477 | ret = log_directory_changes(trans, root, inode, path, dst_path, | |
5478 | ctx); | |
4a500fd1 YZ |
5479 | if (ret) { |
5480 | err = ret; | |
5481 | goto out_unlock; | |
5482 | } | |
e02119d5 | 5483 | } |
49dae1bc | 5484 | |
d1d832a0 | 5485 | /* |
75b463d2 FM |
5486 | * If we are logging that an ancestor inode exists as part of logging a |
5487 | * new name from a link or rename operation, don't mark the inode as | |
5488 | * logged - otherwise if an explicit fsync is made against an ancestor, | |
5489 | * the fsync considers the inode in the log and doesn't sync the log, | |
5490 | * resulting in the ancestor missing after a power failure unless the | |
5491 | * log was synced as part of an fsync against any other unrelated inode. | |
5492 | * So keep it simple for this case and just don't flag the ancestors as | |
5493 | * logged. | |
d1d832a0 | 5494 | */ |
75b463d2 FM |
5495 | if (!ctx || |
5496 | !(S_ISDIR(inode->vfs_inode.i_mode) && ctx->logging_new_name && | |
5497 | &inode->vfs_inode != ctx->inode)) { | |
5498 | spin_lock(&inode->lock); | |
5499 | inode->logged_trans = trans->transid; | |
5500 | /* | |
5501 | * Don't update last_log_commit if we logged that an inode exists | |
5502 | * after it was loaded to memory (full_sync bit set). | |
5503 | * This is to prevent data loss when we do a write to the inode, | |
5504 | * then the inode gets evicted after all delalloc was flushed, | |
5505 | * then we log it exists (due to a rename for example) and then | |
5506 | * fsync it. This last fsync would do nothing (not logging the | |
5507 | * extents previously written). | |
5508 | */ | |
5509 | if (inode_only != LOG_INODE_EXISTS || | |
5510 | !test_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &inode->runtime_flags)) | |
5511 | inode->last_log_commit = inode->last_sub_trans; | |
5512 | spin_unlock(&inode->lock); | |
5513 | } | |
4a500fd1 | 5514 | out_unlock: |
a59108a7 | 5515 | mutex_unlock(&inode->log_mutex); |
e02119d5 CM |
5516 | |
5517 | btrfs_free_path(path); | |
5518 | btrfs_free_path(dst_path); | |
4a500fd1 | 5519 | return err; |
e02119d5 CM |
5520 | } |
5521 | ||
ab12313a FM |
5522 | /* |
5523 | * Check if we need to log an inode. This is used in contexts where while | |
5524 | * logging an inode we need to log another inode (either that it exists or in | |
5525 | * full mode). This is used instead of btrfs_inode_in_log() because the later | |
5526 | * requires the inode to be in the log and have the log transaction committed, | |
5527 | * while here we do not care if the log transaction was already committed - our | |
5528 | * caller will commit the log later - and we want to avoid logging an inode | |
5529 | * multiple times when multiple tasks have joined the same log transaction. | |
5530 | */ | |
5531 | static bool need_log_inode(struct btrfs_trans_handle *trans, | |
5532 | struct btrfs_inode *inode) | |
5533 | { | |
5534 | /* | |
5535 | * If this inode does not have new/updated/deleted xattrs since the last | |
5536 | * time it was logged and is flagged as logged in the current transaction, | |
5537 | * we can skip logging it. As for new/deleted names, those are updated in | |
5538 | * the log by link/unlink/rename operations. | |
5539 | * In case the inode was logged and then evicted and reloaded, its | |
5540 | * logged_trans will be 0, in which case we have to fully log it since | |
5541 | * logged_trans is a transient field, not persisted. | |
5542 | */ | |
5543 | if (inode->logged_trans == trans->transid && | |
5544 | !test_bit(BTRFS_INODE_COPY_EVERYTHING, &inode->runtime_flags)) | |
5545 | return false; | |
5546 | ||
5547 | return true; | |
5548 | } | |
5549 | ||
2f2ff0ee FM |
5550 | struct btrfs_dir_list { |
5551 | u64 ino; | |
5552 | struct list_head list; | |
5553 | }; | |
5554 | ||
5555 | /* | |
5556 | * Log the inodes of the new dentries of a directory. See log_dir_items() for | |
5557 | * details about the why it is needed. | |
5558 | * This is a recursive operation - if an existing dentry corresponds to a | |
5559 | * directory, that directory's new entries are logged too (same behaviour as | |
5560 | * ext3/4, xfs, f2fs, reiserfs, nilfs2). Note that when logging the inodes | |
5561 | * the dentries point to we do not lock their i_mutex, otherwise lockdep | |
5562 | * complains about the following circular lock dependency / possible deadlock: | |
5563 | * | |
5564 | * CPU0 CPU1 | |
5565 | * ---- ---- | |
5566 | * lock(&type->i_mutex_dir_key#3/2); | |
5567 | * lock(sb_internal#2); | |
5568 | * lock(&type->i_mutex_dir_key#3/2); | |
5569 | * lock(&sb->s_type->i_mutex_key#14); | |
5570 | * | |
5571 | * Where sb_internal is the lock (a counter that works as a lock) acquired by | |
5572 | * sb_start_intwrite() in btrfs_start_transaction(). | |
5573 | * Not locking i_mutex of the inodes is still safe because: | |
5574 | * | |
5575 | * 1) For regular files we log with a mode of LOG_INODE_EXISTS. It's possible | |
5576 | * that while logging the inode new references (names) are added or removed | |
5577 | * from the inode, leaving the logged inode item with a link count that does | |
5578 | * not match the number of logged inode reference items. This is fine because | |
5579 | * at log replay time we compute the real number of links and correct the | |
5580 | * link count in the inode item (see replay_one_buffer() and | |
5581 | * link_to_fixup_dir()); | |
5582 | * | |
5583 | * 2) For directories we log with a mode of LOG_INODE_ALL. It's possible that | |
5584 | * while logging the inode's items new items with keys BTRFS_DIR_ITEM_KEY and | |
5585 | * BTRFS_DIR_INDEX_KEY are added to fs/subvol tree and the logged inode item | |
5586 | * has a size that doesn't match the sum of the lengths of all the logged | |
5587 | * names. This does not result in a problem because if a dir_item key is | |
5588 | * logged but its matching dir_index key is not logged, at log replay time we | |
5589 | * don't use it to replay the respective name (see replay_one_name()). On the | |
5590 | * other hand if only the dir_index key ends up being logged, the respective | |
5591 | * name is added to the fs/subvol tree with both the dir_item and dir_index | |
5592 | * keys created (see replay_one_name()). | |
5593 | * The directory's inode item with a wrong i_size is not a problem as well, | |
5594 | * since we don't use it at log replay time to set the i_size in the inode | |
5595 | * item of the fs/subvol tree (see overwrite_item()). | |
5596 | */ | |
5597 | static int log_new_dir_dentries(struct btrfs_trans_handle *trans, | |
5598 | struct btrfs_root *root, | |
51cc0d32 | 5599 | struct btrfs_inode *start_inode, |
2f2ff0ee FM |
5600 | struct btrfs_log_ctx *ctx) |
5601 | { | |
0b246afa | 5602 | struct btrfs_fs_info *fs_info = root->fs_info; |
2f2ff0ee FM |
5603 | struct btrfs_root *log = root->log_root; |
5604 | struct btrfs_path *path; | |
5605 | LIST_HEAD(dir_list); | |
5606 | struct btrfs_dir_list *dir_elem; | |
5607 | int ret = 0; | |
5608 | ||
5609 | path = btrfs_alloc_path(); | |
5610 | if (!path) | |
5611 | return -ENOMEM; | |
5612 | ||
5613 | dir_elem = kmalloc(sizeof(*dir_elem), GFP_NOFS); | |
5614 | if (!dir_elem) { | |
5615 | btrfs_free_path(path); | |
5616 | return -ENOMEM; | |
5617 | } | |
51cc0d32 | 5618 | dir_elem->ino = btrfs_ino(start_inode); |
2f2ff0ee FM |
5619 | list_add_tail(&dir_elem->list, &dir_list); |
5620 | ||
5621 | while (!list_empty(&dir_list)) { | |
5622 | struct extent_buffer *leaf; | |
5623 | struct btrfs_key min_key; | |
5624 | int nritems; | |
5625 | int i; | |
5626 | ||
5627 | dir_elem = list_first_entry(&dir_list, struct btrfs_dir_list, | |
5628 | list); | |
5629 | if (ret) | |
5630 | goto next_dir_inode; | |
5631 | ||
5632 | min_key.objectid = dir_elem->ino; | |
5633 | min_key.type = BTRFS_DIR_ITEM_KEY; | |
5634 | min_key.offset = 0; | |
5635 | again: | |
5636 | btrfs_release_path(path); | |
5637 | ret = btrfs_search_forward(log, &min_key, path, trans->transid); | |
5638 | if (ret < 0) { | |
5639 | goto next_dir_inode; | |
5640 | } else if (ret > 0) { | |
5641 | ret = 0; | |
5642 | goto next_dir_inode; | |
5643 | } | |
5644 | ||
5645 | process_leaf: | |
5646 | leaf = path->nodes[0]; | |
5647 | nritems = btrfs_header_nritems(leaf); | |
5648 | for (i = path->slots[0]; i < nritems; i++) { | |
5649 | struct btrfs_dir_item *di; | |
5650 | struct btrfs_key di_key; | |
5651 | struct inode *di_inode; | |
5652 | struct btrfs_dir_list *new_dir_elem; | |
5653 | int log_mode = LOG_INODE_EXISTS; | |
5654 | int type; | |
5655 | ||
5656 | btrfs_item_key_to_cpu(leaf, &min_key, i); | |
5657 | if (min_key.objectid != dir_elem->ino || | |
5658 | min_key.type != BTRFS_DIR_ITEM_KEY) | |
5659 | goto next_dir_inode; | |
5660 | ||
5661 | di = btrfs_item_ptr(leaf, i, struct btrfs_dir_item); | |
5662 | type = btrfs_dir_type(leaf, di); | |
5663 | if (btrfs_dir_transid(leaf, di) < trans->transid && | |
5664 | type != BTRFS_FT_DIR) | |
5665 | continue; | |
5666 | btrfs_dir_item_key_to_cpu(leaf, di, &di_key); | |
5667 | if (di_key.type == BTRFS_ROOT_ITEM_KEY) | |
5668 | continue; | |
5669 | ||
ec125cfb | 5670 | btrfs_release_path(path); |
0202e83f | 5671 | di_inode = btrfs_iget(fs_info->sb, di_key.objectid, root); |
2f2ff0ee FM |
5672 | if (IS_ERR(di_inode)) { |
5673 | ret = PTR_ERR(di_inode); | |
5674 | goto next_dir_inode; | |
5675 | } | |
5676 | ||
0e44cb3f | 5677 | if (!need_log_inode(trans, BTRFS_I(di_inode))) { |
410f954c | 5678 | btrfs_add_delayed_iput(di_inode); |
ec125cfb | 5679 | break; |
2f2ff0ee FM |
5680 | } |
5681 | ||
5682 | ctx->log_new_dentries = false; | |
3f9749f6 | 5683 | if (type == BTRFS_FT_DIR || type == BTRFS_FT_SYMLINK) |
2f2ff0ee | 5684 | log_mode = LOG_INODE_ALL; |
a59108a7 | 5685 | ret = btrfs_log_inode(trans, root, BTRFS_I(di_inode), |
48778179 | 5686 | log_mode, ctx); |
410f954c | 5687 | btrfs_add_delayed_iput(di_inode); |
2f2ff0ee FM |
5688 | if (ret) |
5689 | goto next_dir_inode; | |
5690 | if (ctx->log_new_dentries) { | |
5691 | new_dir_elem = kmalloc(sizeof(*new_dir_elem), | |
5692 | GFP_NOFS); | |
5693 | if (!new_dir_elem) { | |
5694 | ret = -ENOMEM; | |
5695 | goto next_dir_inode; | |
5696 | } | |
5697 | new_dir_elem->ino = di_key.objectid; | |
5698 | list_add_tail(&new_dir_elem->list, &dir_list); | |
5699 | } | |
5700 | break; | |
5701 | } | |
5702 | if (i == nritems) { | |
5703 | ret = btrfs_next_leaf(log, path); | |
5704 | if (ret < 0) { | |
5705 | goto next_dir_inode; | |
5706 | } else if (ret > 0) { | |
5707 | ret = 0; | |
5708 | goto next_dir_inode; | |
5709 | } | |
5710 | goto process_leaf; | |
5711 | } | |
5712 | if (min_key.offset < (u64)-1) { | |
5713 | min_key.offset++; | |
5714 | goto again; | |
5715 | } | |
5716 | next_dir_inode: | |
5717 | list_del(&dir_elem->list); | |
5718 | kfree(dir_elem); | |
5719 | } | |
5720 | ||
5721 | btrfs_free_path(path); | |
5722 | return ret; | |
5723 | } | |
5724 | ||
18aa0922 | 5725 | static int btrfs_log_all_parents(struct btrfs_trans_handle *trans, |
d0a0b78d | 5726 | struct btrfs_inode *inode, |
18aa0922 FM |
5727 | struct btrfs_log_ctx *ctx) |
5728 | { | |
3ffbd68c | 5729 | struct btrfs_fs_info *fs_info = trans->fs_info; |
18aa0922 FM |
5730 | int ret; |
5731 | struct btrfs_path *path; | |
5732 | struct btrfs_key key; | |
d0a0b78d NB |
5733 | struct btrfs_root *root = inode->root; |
5734 | const u64 ino = btrfs_ino(inode); | |
18aa0922 FM |
5735 | |
5736 | path = btrfs_alloc_path(); | |
5737 | if (!path) | |
5738 | return -ENOMEM; | |
5739 | path->skip_locking = 1; | |
5740 | path->search_commit_root = 1; | |
5741 | ||
5742 | key.objectid = ino; | |
5743 | key.type = BTRFS_INODE_REF_KEY; | |
5744 | key.offset = 0; | |
5745 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
5746 | if (ret < 0) | |
5747 | goto out; | |
5748 | ||
5749 | while (true) { | |
5750 | struct extent_buffer *leaf = path->nodes[0]; | |
5751 | int slot = path->slots[0]; | |
5752 | u32 cur_offset = 0; | |
5753 | u32 item_size; | |
5754 | unsigned long ptr; | |
5755 | ||
5756 | if (slot >= btrfs_header_nritems(leaf)) { | |
5757 | ret = btrfs_next_leaf(root, path); | |
5758 | if (ret < 0) | |
5759 | goto out; | |
5760 | else if (ret > 0) | |
5761 | break; | |
5762 | continue; | |
5763 | } | |
5764 | ||
5765 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
5766 | /* BTRFS_INODE_EXTREF_KEY is BTRFS_INODE_REF_KEY + 1 */ | |
5767 | if (key.objectid != ino || key.type > BTRFS_INODE_EXTREF_KEY) | |
5768 | break; | |
5769 | ||
5770 | item_size = btrfs_item_size_nr(leaf, slot); | |
5771 | ptr = btrfs_item_ptr_offset(leaf, slot); | |
5772 | while (cur_offset < item_size) { | |
5773 | struct btrfs_key inode_key; | |
5774 | struct inode *dir_inode; | |
5775 | ||
5776 | inode_key.type = BTRFS_INODE_ITEM_KEY; | |
5777 | inode_key.offset = 0; | |
5778 | ||
5779 | if (key.type == BTRFS_INODE_EXTREF_KEY) { | |
5780 | struct btrfs_inode_extref *extref; | |
5781 | ||
5782 | extref = (struct btrfs_inode_extref *) | |
5783 | (ptr + cur_offset); | |
5784 | inode_key.objectid = btrfs_inode_extref_parent( | |
5785 | leaf, extref); | |
5786 | cur_offset += sizeof(*extref); | |
5787 | cur_offset += btrfs_inode_extref_name_len(leaf, | |
5788 | extref); | |
5789 | } else { | |
5790 | inode_key.objectid = key.offset; | |
5791 | cur_offset = item_size; | |
5792 | } | |
5793 | ||
0202e83f DS |
5794 | dir_inode = btrfs_iget(fs_info->sb, inode_key.objectid, |
5795 | root); | |
0f375eed FM |
5796 | /* |
5797 | * If the parent inode was deleted, return an error to | |
5798 | * fallback to a transaction commit. This is to prevent | |
5799 | * getting an inode that was moved from one parent A to | |
5800 | * a parent B, got its former parent A deleted and then | |
5801 | * it got fsync'ed, from existing at both parents after | |
5802 | * a log replay (and the old parent still existing). | |
5803 | * Example: | |
5804 | * | |
5805 | * mkdir /mnt/A | |
5806 | * mkdir /mnt/B | |
5807 | * touch /mnt/B/bar | |
5808 | * sync | |
5809 | * mv /mnt/B/bar /mnt/A/bar | |
5810 | * mv -T /mnt/A /mnt/B | |
5811 | * fsync /mnt/B/bar | |
5812 | * <power fail> | |
5813 | * | |
5814 | * If we ignore the old parent B which got deleted, | |
5815 | * after a log replay we would have file bar linked | |
5816 | * at both parents and the old parent B would still | |
5817 | * exist. | |
5818 | */ | |
5819 | if (IS_ERR(dir_inode)) { | |
5820 | ret = PTR_ERR(dir_inode); | |
5821 | goto out; | |
5822 | } | |
18aa0922 | 5823 | |
3e6a86a1 FM |
5824 | if (!need_log_inode(trans, BTRFS_I(dir_inode))) { |
5825 | btrfs_add_delayed_iput(dir_inode); | |
5826 | continue; | |
5827 | } | |
5828 | ||
657ed1aa FM |
5829 | if (ctx) |
5830 | ctx->log_new_dentries = false; | |
a59108a7 | 5831 | ret = btrfs_log_inode(trans, root, BTRFS_I(dir_inode), |
48778179 | 5832 | LOG_INODE_ALL, ctx); |
657ed1aa FM |
5833 | if (!ret && ctx && ctx->log_new_dentries) |
5834 | ret = log_new_dir_dentries(trans, root, | |
f85b7379 | 5835 | BTRFS_I(dir_inode), ctx); |
410f954c | 5836 | btrfs_add_delayed_iput(dir_inode); |
18aa0922 FM |
5837 | if (ret) |
5838 | goto out; | |
5839 | } | |
5840 | path->slots[0]++; | |
5841 | } | |
5842 | ret = 0; | |
5843 | out: | |
5844 | btrfs_free_path(path); | |
5845 | return ret; | |
5846 | } | |
5847 | ||
b8aa330d FM |
5848 | static int log_new_ancestors(struct btrfs_trans_handle *trans, |
5849 | struct btrfs_root *root, | |
5850 | struct btrfs_path *path, | |
5851 | struct btrfs_log_ctx *ctx) | |
5852 | { | |
5853 | struct btrfs_key found_key; | |
5854 | ||
5855 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, path->slots[0]); | |
5856 | ||
5857 | while (true) { | |
5858 | struct btrfs_fs_info *fs_info = root->fs_info; | |
b8aa330d FM |
5859 | struct extent_buffer *leaf = path->nodes[0]; |
5860 | int slot = path->slots[0]; | |
5861 | struct btrfs_key search_key; | |
5862 | struct inode *inode; | |
0202e83f | 5863 | u64 ino; |
b8aa330d FM |
5864 | int ret = 0; |
5865 | ||
5866 | btrfs_release_path(path); | |
5867 | ||
0202e83f DS |
5868 | ino = found_key.offset; |
5869 | ||
b8aa330d FM |
5870 | search_key.objectid = found_key.offset; |
5871 | search_key.type = BTRFS_INODE_ITEM_KEY; | |
5872 | search_key.offset = 0; | |
0202e83f | 5873 | inode = btrfs_iget(fs_info->sb, ino, root); |
b8aa330d FM |
5874 | if (IS_ERR(inode)) |
5875 | return PTR_ERR(inode); | |
5876 | ||
ab12313a FM |
5877 | if (BTRFS_I(inode)->generation >= trans->transid && |
5878 | need_log_inode(trans, BTRFS_I(inode))) | |
b8aa330d | 5879 | ret = btrfs_log_inode(trans, root, BTRFS_I(inode), |
48778179 | 5880 | LOG_INODE_EXISTS, ctx); |
410f954c | 5881 | btrfs_add_delayed_iput(inode); |
b8aa330d FM |
5882 | if (ret) |
5883 | return ret; | |
5884 | ||
5885 | if (search_key.objectid == BTRFS_FIRST_FREE_OBJECTID) | |
5886 | break; | |
5887 | ||
5888 | search_key.type = BTRFS_INODE_REF_KEY; | |
5889 | ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0); | |
5890 | if (ret < 0) | |
5891 | return ret; | |
5892 | ||
5893 | leaf = path->nodes[0]; | |
5894 | slot = path->slots[0]; | |
5895 | if (slot >= btrfs_header_nritems(leaf)) { | |
5896 | ret = btrfs_next_leaf(root, path); | |
5897 | if (ret < 0) | |
5898 | return ret; | |
5899 | else if (ret > 0) | |
5900 | return -ENOENT; | |
5901 | leaf = path->nodes[0]; | |
5902 | slot = path->slots[0]; | |
5903 | } | |
5904 | ||
5905 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
5906 | if (found_key.objectid != search_key.objectid || | |
5907 | found_key.type != BTRFS_INODE_REF_KEY) | |
5908 | return -ENOENT; | |
5909 | } | |
5910 | return 0; | |
5911 | } | |
5912 | ||
5913 | static int log_new_ancestors_fast(struct btrfs_trans_handle *trans, | |
5914 | struct btrfs_inode *inode, | |
5915 | struct dentry *parent, | |
5916 | struct btrfs_log_ctx *ctx) | |
5917 | { | |
5918 | struct btrfs_root *root = inode->root; | |
b8aa330d FM |
5919 | struct dentry *old_parent = NULL; |
5920 | struct super_block *sb = inode->vfs_inode.i_sb; | |
5921 | int ret = 0; | |
5922 | ||
5923 | while (true) { | |
5924 | if (!parent || d_really_is_negative(parent) || | |
5925 | sb != parent->d_sb) | |
5926 | break; | |
5927 | ||
5928 | inode = BTRFS_I(d_inode(parent)); | |
5929 | if (root != inode->root) | |
5930 | break; | |
5931 | ||
ab12313a FM |
5932 | if (inode->generation >= trans->transid && |
5933 | need_log_inode(trans, inode)) { | |
b8aa330d | 5934 | ret = btrfs_log_inode(trans, root, inode, |
48778179 | 5935 | LOG_INODE_EXISTS, ctx); |
b8aa330d FM |
5936 | if (ret) |
5937 | break; | |
5938 | } | |
5939 | if (IS_ROOT(parent)) | |
5940 | break; | |
5941 | ||
5942 | parent = dget_parent(parent); | |
5943 | dput(old_parent); | |
5944 | old_parent = parent; | |
5945 | } | |
5946 | dput(old_parent); | |
5947 | ||
5948 | return ret; | |
5949 | } | |
5950 | ||
5951 | static int log_all_new_ancestors(struct btrfs_trans_handle *trans, | |
5952 | struct btrfs_inode *inode, | |
5953 | struct dentry *parent, | |
5954 | struct btrfs_log_ctx *ctx) | |
5955 | { | |
5956 | struct btrfs_root *root = inode->root; | |
5957 | const u64 ino = btrfs_ino(inode); | |
5958 | struct btrfs_path *path; | |
5959 | struct btrfs_key search_key; | |
5960 | int ret; | |
5961 | ||
5962 | /* | |
5963 | * For a single hard link case, go through a fast path that does not | |
5964 | * need to iterate the fs/subvolume tree. | |
5965 | */ | |
5966 | if (inode->vfs_inode.i_nlink < 2) | |
5967 | return log_new_ancestors_fast(trans, inode, parent, ctx); | |
5968 | ||
5969 | path = btrfs_alloc_path(); | |
5970 | if (!path) | |
5971 | return -ENOMEM; | |
5972 | ||
5973 | search_key.objectid = ino; | |
5974 | search_key.type = BTRFS_INODE_REF_KEY; | |
5975 | search_key.offset = 0; | |
5976 | again: | |
5977 | ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0); | |
5978 | if (ret < 0) | |
5979 | goto out; | |
5980 | if (ret == 0) | |
5981 | path->slots[0]++; | |
5982 | ||
5983 | while (true) { | |
5984 | struct extent_buffer *leaf = path->nodes[0]; | |
5985 | int slot = path->slots[0]; | |
5986 | struct btrfs_key found_key; | |
5987 | ||
5988 | if (slot >= btrfs_header_nritems(leaf)) { | |
5989 | ret = btrfs_next_leaf(root, path); | |
5990 | if (ret < 0) | |
5991 | goto out; | |
5992 | else if (ret > 0) | |
5993 | break; | |
5994 | continue; | |
5995 | } | |
5996 | ||
5997 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
5998 | if (found_key.objectid != ino || | |
5999 | found_key.type > BTRFS_INODE_EXTREF_KEY) | |
6000 | break; | |
6001 | ||
6002 | /* | |
6003 | * Don't deal with extended references because they are rare | |
6004 | * cases and too complex to deal with (we would need to keep | |
6005 | * track of which subitem we are processing for each item in | |
6006 | * this loop, etc). So just return some error to fallback to | |
6007 | * a transaction commit. | |
6008 | */ | |
6009 | if (found_key.type == BTRFS_INODE_EXTREF_KEY) { | |
6010 | ret = -EMLINK; | |
6011 | goto out; | |
6012 | } | |
6013 | ||
6014 | /* | |
6015 | * Logging ancestors needs to do more searches on the fs/subvol | |
6016 | * tree, so it releases the path as needed to avoid deadlocks. | |
6017 | * Keep track of the last inode ref key and resume from that key | |
6018 | * after logging all new ancestors for the current hard link. | |
6019 | */ | |
6020 | memcpy(&search_key, &found_key, sizeof(search_key)); | |
6021 | ||
6022 | ret = log_new_ancestors(trans, root, path, ctx); | |
6023 | if (ret) | |
6024 | goto out; | |
6025 | btrfs_release_path(path); | |
6026 | goto again; | |
6027 | } | |
6028 | ret = 0; | |
6029 | out: | |
6030 | btrfs_free_path(path); | |
6031 | return ret; | |
6032 | } | |
6033 | ||
e02119d5 CM |
6034 | /* |
6035 | * helper function around btrfs_log_inode to make sure newly created | |
6036 | * parent directories also end up in the log. A minimal inode and backref | |
6037 | * only logging is done of any parent directories that are older than | |
6038 | * the last committed transaction | |
6039 | */ | |
48a3b636 | 6040 | static int btrfs_log_inode_parent(struct btrfs_trans_handle *trans, |
19df27a9 | 6041 | struct btrfs_inode *inode, |
49dae1bc | 6042 | struct dentry *parent, |
41a1eada | 6043 | int inode_only, |
8b050d35 | 6044 | struct btrfs_log_ctx *ctx) |
e02119d5 | 6045 | { |
f882274b | 6046 | struct btrfs_root *root = inode->root; |
0b246afa | 6047 | struct btrfs_fs_info *fs_info = root->fs_info; |
12fcfd22 | 6048 | int ret = 0; |
2f2ff0ee | 6049 | bool log_dentries = false; |
12fcfd22 | 6050 | |
0b246afa | 6051 | if (btrfs_test_opt(fs_info, NOTREELOG)) { |
3a5e1404 SW |
6052 | ret = 1; |
6053 | goto end_no_trans; | |
6054 | } | |
6055 | ||
f882274b | 6056 | if (btrfs_root_refs(&root->root_item) == 0) { |
76dda93c YZ |
6057 | ret = 1; |
6058 | goto end_no_trans; | |
6059 | } | |
6060 | ||
f2d72f42 FM |
6061 | /* |
6062 | * Skip already logged inodes or inodes corresponding to tmpfiles | |
6063 | * (since logging them is pointless, a link count of 0 means they | |
6064 | * will never be accessible). | |
6065 | */ | |
626e9f41 FM |
6066 | if ((btrfs_inode_in_log(inode, trans->transid) && |
6067 | list_empty(&ctx->ordered_extents)) || | |
f2d72f42 | 6068 | inode->vfs_inode.i_nlink == 0) { |
257c62e1 CM |
6069 | ret = BTRFS_NO_LOG_SYNC; |
6070 | goto end_no_trans; | |
6071 | } | |
6072 | ||
8b050d35 | 6073 | ret = start_log_trans(trans, root, ctx); |
4a500fd1 | 6074 | if (ret) |
e87ac136 | 6075 | goto end_no_trans; |
e02119d5 | 6076 | |
48778179 | 6077 | ret = btrfs_log_inode(trans, root, inode, inode_only, ctx); |
4a500fd1 YZ |
6078 | if (ret) |
6079 | goto end_trans; | |
12fcfd22 | 6080 | |
af4176b4 CM |
6081 | /* |
6082 | * for regular files, if its inode is already on disk, we don't | |
6083 | * have to worry about the parents at all. This is because | |
6084 | * we can use the last_unlink_trans field to record renames | |
6085 | * and other fun in this file. | |
6086 | */ | |
19df27a9 | 6087 | if (S_ISREG(inode->vfs_inode.i_mode) && |
47d3db41 FM |
6088 | inode->generation < trans->transid && |
6089 | inode->last_unlink_trans < trans->transid) { | |
4a500fd1 YZ |
6090 | ret = 0; |
6091 | goto end_trans; | |
6092 | } | |
af4176b4 | 6093 | |
19df27a9 | 6094 | if (S_ISDIR(inode->vfs_inode.i_mode) && ctx && ctx->log_new_dentries) |
2f2ff0ee FM |
6095 | log_dentries = true; |
6096 | ||
18aa0922 | 6097 | /* |
01327610 | 6098 | * On unlink we must make sure all our current and old parent directory |
18aa0922 FM |
6099 | * inodes are fully logged. This is to prevent leaving dangling |
6100 | * directory index entries in directories that were our parents but are | |
6101 | * not anymore. Not doing this results in old parent directory being | |
6102 | * impossible to delete after log replay (rmdir will always fail with | |
6103 | * error -ENOTEMPTY). | |
6104 | * | |
6105 | * Example 1: | |
6106 | * | |
6107 | * mkdir testdir | |
6108 | * touch testdir/foo | |
6109 | * ln testdir/foo testdir/bar | |
6110 | * sync | |
6111 | * unlink testdir/bar | |
6112 | * xfs_io -c fsync testdir/foo | |
6113 | * <power failure> | |
6114 | * mount fs, triggers log replay | |
6115 | * | |
6116 | * If we don't log the parent directory (testdir), after log replay the | |
6117 | * directory still has an entry pointing to the file inode using the bar | |
6118 | * name, but a matching BTRFS_INODE_[REF|EXTREF]_KEY does not exist and | |
6119 | * the file inode has a link count of 1. | |
6120 | * | |
6121 | * Example 2: | |
6122 | * | |
6123 | * mkdir testdir | |
6124 | * touch foo | |
6125 | * ln foo testdir/foo2 | |
6126 | * ln foo testdir/foo3 | |
6127 | * sync | |
6128 | * unlink testdir/foo3 | |
6129 | * xfs_io -c fsync foo | |
6130 | * <power failure> | |
6131 | * mount fs, triggers log replay | |
6132 | * | |
6133 | * Similar as the first example, after log replay the parent directory | |
6134 | * testdir still has an entry pointing to the inode file with name foo3 | |
6135 | * but the file inode does not have a matching BTRFS_INODE_REF_KEY item | |
6136 | * and has a link count of 2. | |
6137 | */ | |
47d3db41 | 6138 | if (inode->last_unlink_trans >= trans->transid) { |
b8aa330d | 6139 | ret = btrfs_log_all_parents(trans, inode, ctx); |
18aa0922 FM |
6140 | if (ret) |
6141 | goto end_trans; | |
6142 | } | |
6143 | ||
b8aa330d FM |
6144 | ret = log_all_new_ancestors(trans, inode, parent, ctx); |
6145 | if (ret) | |
41bd6067 | 6146 | goto end_trans; |
76dda93c | 6147 | |
2f2ff0ee | 6148 | if (log_dentries) |
b8aa330d | 6149 | ret = log_new_dir_dentries(trans, root, inode, ctx); |
2f2ff0ee FM |
6150 | else |
6151 | ret = 0; | |
4a500fd1 YZ |
6152 | end_trans: |
6153 | if (ret < 0) { | |
90787766 | 6154 | btrfs_set_log_full_commit(trans); |
4a500fd1 YZ |
6155 | ret = 1; |
6156 | } | |
8b050d35 MX |
6157 | |
6158 | if (ret) | |
6159 | btrfs_remove_log_ctx(root, ctx); | |
12fcfd22 CM |
6160 | btrfs_end_log_trans(root); |
6161 | end_no_trans: | |
6162 | return ret; | |
e02119d5 CM |
6163 | } |
6164 | ||
6165 | /* | |
6166 | * it is not safe to log dentry if the chunk root has added new | |
6167 | * chunks. This returns 0 if the dentry was logged, and 1 otherwise. | |
6168 | * If this returns 1, you must commit the transaction to safely get your | |
6169 | * data on disk. | |
6170 | */ | |
6171 | int btrfs_log_dentry_safe(struct btrfs_trans_handle *trans, | |
e5b84f7a | 6172 | struct dentry *dentry, |
8b050d35 | 6173 | struct btrfs_log_ctx *ctx) |
e02119d5 | 6174 | { |
6a912213 JB |
6175 | struct dentry *parent = dget_parent(dentry); |
6176 | int ret; | |
6177 | ||
f882274b | 6178 | ret = btrfs_log_inode_parent(trans, BTRFS_I(d_inode(dentry)), parent, |
48778179 | 6179 | LOG_INODE_ALL, ctx); |
6a912213 JB |
6180 | dput(parent); |
6181 | ||
6182 | return ret; | |
e02119d5 CM |
6183 | } |
6184 | ||
6185 | /* | |
6186 | * should be called during mount to recover any replay any log trees | |
6187 | * from the FS | |
6188 | */ | |
6189 | int btrfs_recover_log_trees(struct btrfs_root *log_root_tree) | |
6190 | { | |
6191 | int ret; | |
6192 | struct btrfs_path *path; | |
6193 | struct btrfs_trans_handle *trans; | |
6194 | struct btrfs_key key; | |
6195 | struct btrfs_key found_key; | |
e02119d5 CM |
6196 | struct btrfs_root *log; |
6197 | struct btrfs_fs_info *fs_info = log_root_tree->fs_info; | |
6198 | struct walk_control wc = { | |
6199 | .process_func = process_one_buffer, | |
430a6626 | 6200 | .stage = LOG_WALK_PIN_ONLY, |
e02119d5 CM |
6201 | }; |
6202 | ||
e02119d5 | 6203 | path = btrfs_alloc_path(); |
db5b493a TI |
6204 | if (!path) |
6205 | return -ENOMEM; | |
6206 | ||
afcdd129 | 6207 | set_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags); |
e02119d5 | 6208 | |
4a500fd1 | 6209 | trans = btrfs_start_transaction(fs_info->tree_root, 0); |
79787eaa JM |
6210 | if (IS_ERR(trans)) { |
6211 | ret = PTR_ERR(trans); | |
6212 | goto error; | |
6213 | } | |
e02119d5 CM |
6214 | |
6215 | wc.trans = trans; | |
6216 | wc.pin = 1; | |
6217 | ||
db5b493a | 6218 | ret = walk_log_tree(trans, log_root_tree, &wc); |
79787eaa | 6219 | if (ret) { |
5d163e0e JM |
6220 | btrfs_handle_fs_error(fs_info, ret, |
6221 | "Failed to pin buffers while recovering log root tree."); | |
79787eaa JM |
6222 | goto error; |
6223 | } | |
e02119d5 CM |
6224 | |
6225 | again: | |
6226 | key.objectid = BTRFS_TREE_LOG_OBJECTID; | |
6227 | key.offset = (u64)-1; | |
962a298f | 6228 | key.type = BTRFS_ROOT_ITEM_KEY; |
e02119d5 | 6229 | |
d397712b | 6230 | while (1) { |
e02119d5 | 6231 | ret = btrfs_search_slot(NULL, log_root_tree, &key, path, 0, 0); |
79787eaa JM |
6232 | |
6233 | if (ret < 0) { | |
34d97007 | 6234 | btrfs_handle_fs_error(fs_info, ret, |
79787eaa JM |
6235 | "Couldn't find tree log root."); |
6236 | goto error; | |
6237 | } | |
e02119d5 CM |
6238 | if (ret > 0) { |
6239 | if (path->slots[0] == 0) | |
6240 | break; | |
6241 | path->slots[0]--; | |
6242 | } | |
6243 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, | |
6244 | path->slots[0]); | |
b3b4aa74 | 6245 | btrfs_release_path(path); |
e02119d5 CM |
6246 | if (found_key.objectid != BTRFS_TREE_LOG_OBJECTID) |
6247 | break; | |
6248 | ||
62a2c73e | 6249 | log = btrfs_read_tree_root(log_root_tree, &found_key); |
79787eaa JM |
6250 | if (IS_ERR(log)) { |
6251 | ret = PTR_ERR(log); | |
34d97007 | 6252 | btrfs_handle_fs_error(fs_info, ret, |
79787eaa JM |
6253 | "Couldn't read tree log root."); |
6254 | goto error; | |
6255 | } | |
e02119d5 | 6256 | |
56e9357a DS |
6257 | wc.replay_dest = btrfs_get_fs_root(fs_info, found_key.offset, |
6258 | true); | |
79787eaa JM |
6259 | if (IS_ERR(wc.replay_dest)) { |
6260 | ret = PTR_ERR(wc.replay_dest); | |
9bc574de JB |
6261 | |
6262 | /* | |
6263 | * We didn't find the subvol, likely because it was | |
6264 | * deleted. This is ok, simply skip this log and go to | |
6265 | * the next one. | |
6266 | * | |
6267 | * We need to exclude the root because we can't have | |
6268 | * other log replays overwriting this log as we'll read | |
6269 | * it back in a few more times. This will keep our | |
6270 | * block from being modified, and we'll just bail for | |
6271 | * each subsequent pass. | |
6272 | */ | |
6273 | if (ret == -ENOENT) | |
9fce5704 | 6274 | ret = btrfs_pin_extent_for_log_replay(trans, |
9bc574de JB |
6275 | log->node->start, |
6276 | log->node->len); | |
00246528 | 6277 | btrfs_put_root(log); |
9bc574de JB |
6278 | |
6279 | if (!ret) | |
6280 | goto next; | |
5d163e0e JM |
6281 | btrfs_handle_fs_error(fs_info, ret, |
6282 | "Couldn't read target root for tree log recovery."); | |
79787eaa JM |
6283 | goto error; |
6284 | } | |
e02119d5 | 6285 | |
07d400a6 | 6286 | wc.replay_dest->log_root = log; |
2002ae11 JB |
6287 | ret = btrfs_record_root_in_trans(trans, wc.replay_dest); |
6288 | if (ret) | |
6289 | /* The loop needs to continue due to the root refs */ | |
6290 | btrfs_handle_fs_error(fs_info, ret, | |
6291 | "failed to record the log root in transaction"); | |
6292 | else | |
6293 | ret = walk_log_tree(trans, log, &wc); | |
e02119d5 | 6294 | |
b50c6e25 | 6295 | if (!ret && wc.stage == LOG_WALK_REPLAY_ALL) { |
e02119d5 CM |
6296 | ret = fixup_inode_link_counts(trans, wc.replay_dest, |
6297 | path); | |
e02119d5 CM |
6298 | } |
6299 | ||
900c9981 LB |
6300 | if (!ret && wc.stage == LOG_WALK_REPLAY_ALL) { |
6301 | struct btrfs_root *root = wc.replay_dest; | |
6302 | ||
6303 | btrfs_release_path(path); | |
6304 | ||
6305 | /* | |
6306 | * We have just replayed everything, and the highest | |
6307 | * objectid of fs roots probably has changed in case | |
6308 | * some inode_item's got replayed. | |
6309 | * | |
6310 | * root->objectid_mutex is not acquired as log replay | |
6311 | * could only happen during mount. | |
6312 | */ | |
453e4873 | 6313 | ret = btrfs_init_root_free_objectid(root); |
900c9981 LB |
6314 | } |
6315 | ||
07d400a6 | 6316 | wc.replay_dest->log_root = NULL; |
00246528 | 6317 | btrfs_put_root(wc.replay_dest); |
00246528 | 6318 | btrfs_put_root(log); |
e02119d5 | 6319 | |
b50c6e25 JB |
6320 | if (ret) |
6321 | goto error; | |
9bc574de | 6322 | next: |
e02119d5 CM |
6323 | if (found_key.offset == 0) |
6324 | break; | |
9bc574de | 6325 | key.offset = found_key.offset - 1; |
e02119d5 | 6326 | } |
b3b4aa74 | 6327 | btrfs_release_path(path); |
e02119d5 CM |
6328 | |
6329 | /* step one is to pin it all, step two is to replay just inodes */ | |
6330 | if (wc.pin) { | |
6331 | wc.pin = 0; | |
6332 | wc.process_func = replay_one_buffer; | |
6333 | wc.stage = LOG_WALK_REPLAY_INODES; | |
6334 | goto again; | |
6335 | } | |
6336 | /* step three is to replay everything */ | |
6337 | if (wc.stage < LOG_WALK_REPLAY_ALL) { | |
6338 | wc.stage++; | |
6339 | goto again; | |
6340 | } | |
6341 | ||
6342 | btrfs_free_path(path); | |
6343 | ||
abefa55a | 6344 | /* step 4: commit the transaction, which also unpins the blocks */ |
3a45bb20 | 6345 | ret = btrfs_commit_transaction(trans); |
abefa55a JB |
6346 | if (ret) |
6347 | return ret; | |
6348 | ||
e02119d5 | 6349 | log_root_tree->log_root = NULL; |
afcdd129 | 6350 | clear_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags); |
00246528 | 6351 | btrfs_put_root(log_root_tree); |
79787eaa | 6352 | |
abefa55a | 6353 | return 0; |
79787eaa | 6354 | error: |
b50c6e25 | 6355 | if (wc.trans) |
3a45bb20 | 6356 | btrfs_end_transaction(wc.trans); |
79787eaa JM |
6357 | btrfs_free_path(path); |
6358 | return ret; | |
e02119d5 | 6359 | } |
12fcfd22 CM |
6360 | |
6361 | /* | |
6362 | * there are some corner cases where we want to force a full | |
6363 | * commit instead of allowing a directory to be logged. | |
6364 | * | |
6365 | * They revolve around files there were unlinked from the directory, and | |
6366 | * this function updates the parent directory so that a full commit is | |
6367 | * properly done if it is fsync'd later after the unlinks are done. | |
2be63d5c FM |
6368 | * |
6369 | * Must be called before the unlink operations (updates to the subvolume tree, | |
6370 | * inodes, etc) are done. | |
12fcfd22 CM |
6371 | */ |
6372 | void btrfs_record_unlink_dir(struct btrfs_trans_handle *trans, | |
4176bdbf | 6373 | struct btrfs_inode *dir, struct btrfs_inode *inode, |
12fcfd22 CM |
6374 | int for_rename) |
6375 | { | |
af4176b4 CM |
6376 | /* |
6377 | * when we're logging a file, if it hasn't been renamed | |
6378 | * or unlinked, and its inode is fully committed on disk, | |
6379 | * we don't have to worry about walking up the directory chain | |
6380 | * to log its parents. | |
6381 | * | |
6382 | * So, we use the last_unlink_trans field to put this transid | |
6383 | * into the file. When the file is logged we check it and | |
6384 | * don't log the parents if the file is fully on disk. | |
6385 | */ | |
4176bdbf NB |
6386 | mutex_lock(&inode->log_mutex); |
6387 | inode->last_unlink_trans = trans->transid; | |
6388 | mutex_unlock(&inode->log_mutex); | |
af4176b4 | 6389 | |
12fcfd22 CM |
6390 | /* |
6391 | * if this directory was already logged any new | |
6392 | * names for this file/dir will get recorded | |
6393 | */ | |
4176bdbf | 6394 | if (dir->logged_trans == trans->transid) |
12fcfd22 CM |
6395 | return; |
6396 | ||
6397 | /* | |
6398 | * if the inode we're about to unlink was logged, | |
6399 | * the log will be properly updated for any new names | |
6400 | */ | |
4176bdbf | 6401 | if (inode->logged_trans == trans->transid) |
12fcfd22 CM |
6402 | return; |
6403 | ||
6404 | /* | |
6405 | * when renaming files across directories, if the directory | |
6406 | * there we're unlinking from gets fsync'd later on, there's | |
6407 | * no way to find the destination directory later and fsync it | |
6408 | * properly. So, we have to be conservative and force commits | |
6409 | * so the new name gets discovered. | |
6410 | */ | |
6411 | if (for_rename) | |
6412 | goto record; | |
6413 | ||
6414 | /* we can safely do the unlink without any special recording */ | |
6415 | return; | |
6416 | ||
6417 | record: | |
4176bdbf NB |
6418 | mutex_lock(&dir->log_mutex); |
6419 | dir->last_unlink_trans = trans->transid; | |
6420 | mutex_unlock(&dir->log_mutex); | |
1ec9a1ae FM |
6421 | } |
6422 | ||
6423 | /* | |
6424 | * Make sure that if someone attempts to fsync the parent directory of a deleted | |
6425 | * snapshot, it ends up triggering a transaction commit. This is to guarantee | |
6426 | * that after replaying the log tree of the parent directory's root we will not | |
6427 | * see the snapshot anymore and at log replay time we will not see any log tree | |
6428 | * corresponding to the deleted snapshot's root, which could lead to replaying | |
6429 | * it after replaying the log tree of the parent directory (which would replay | |
6430 | * the snapshot delete operation). | |
2be63d5c FM |
6431 | * |
6432 | * Must be called before the actual snapshot destroy operation (updates to the | |
6433 | * parent root and tree of tree roots trees, etc) are done. | |
1ec9a1ae FM |
6434 | */ |
6435 | void btrfs_record_snapshot_destroy(struct btrfs_trans_handle *trans, | |
43663557 | 6436 | struct btrfs_inode *dir) |
1ec9a1ae | 6437 | { |
43663557 NB |
6438 | mutex_lock(&dir->log_mutex); |
6439 | dir->last_unlink_trans = trans->transid; | |
6440 | mutex_unlock(&dir->log_mutex); | |
12fcfd22 CM |
6441 | } |
6442 | ||
6443 | /* | |
6444 | * Call this after adding a new name for a file and it will properly | |
6445 | * update the log to reflect the new name. | |
12fcfd22 | 6446 | */ |
75b463d2 | 6447 | void btrfs_log_new_name(struct btrfs_trans_handle *trans, |
9ca5fbfb | 6448 | struct btrfs_inode *inode, struct btrfs_inode *old_dir, |
75b463d2 | 6449 | struct dentry *parent) |
12fcfd22 | 6450 | { |
75b463d2 | 6451 | struct btrfs_log_ctx ctx; |
12fcfd22 | 6452 | |
af4176b4 CM |
6453 | /* |
6454 | * this will force the logging code to walk the dentry chain | |
6455 | * up for the file | |
6456 | */ | |
9a6509c4 | 6457 | if (!S_ISDIR(inode->vfs_inode.i_mode)) |
9ca5fbfb | 6458 | inode->last_unlink_trans = trans->transid; |
af4176b4 | 6459 | |
12fcfd22 CM |
6460 | /* |
6461 | * if this inode hasn't been logged and directory we're renaming it | |
6462 | * from hasn't been logged, we don't need to log it | |
6463 | */ | |
de53d892 FM |
6464 | if (inode->logged_trans < trans->transid && |
6465 | (!old_dir || old_dir->logged_trans < trans->transid)) | |
75b463d2 | 6466 | return; |
12fcfd22 | 6467 | |
54a40fc3 FM |
6468 | /* |
6469 | * If we are doing a rename (old_dir is not NULL) from a directory that | |
6470 | * was previously logged, make sure the next log attempt on the directory | |
6471 | * is not skipped and logs the inode again. This is because the log may | |
6472 | * not currently be authoritative for a range including the old | |
6473 | * BTRFS_DIR_ITEM_KEY and BTRFS_DIR_INDEX_KEY keys, so we want to make | |
6474 | * sure after a log replay we do not end up with both the new and old | |
6475 | * dentries around (in case the inode is a directory we would have a | |
6476 | * directory with two hard links and 2 inode references for different | |
6477 | * parents). The next log attempt of old_dir will happen at | |
6478 | * btrfs_log_all_parents(), called through btrfs_log_inode_parent() | |
6479 | * below, because we have previously set inode->last_unlink_trans to the | |
6480 | * current transaction ID, either here or at btrfs_record_unlink_dir() in | |
6481 | * case inode is a directory. | |
6482 | */ | |
6483 | if (old_dir) | |
6484 | old_dir->logged_trans = 0; | |
6485 | ||
75b463d2 FM |
6486 | btrfs_init_log_ctx(&ctx, &inode->vfs_inode); |
6487 | ctx.logging_new_name = true; | |
6488 | /* | |
6489 | * We don't care about the return value. If we fail to log the new name | |
6490 | * then we know the next attempt to sync the log will fallback to a full | |
6491 | * transaction commit (due to a call to btrfs_set_log_full_commit()), so | |
6492 | * we don't need to worry about getting a log committed that has an | |
6493 | * inconsistent state after a rename operation. | |
6494 | */ | |
48778179 | 6495 | btrfs_log_inode_parent(trans, inode, parent, LOG_INODE_EXISTS, &ctx); |
12fcfd22 CM |
6496 | } |
6497 |