]>
Commit | Line | Data |
---|---|---|
c1d7c514 | 1 | // SPDX-License-Identifier: GPL-2.0 |
e02119d5 CM |
2 | /* |
3 | * Copyright (C) 2008 Oracle. All rights reserved. | |
e02119d5 CM |
4 | */ |
5 | ||
6 | #include <linux/sched.h> | |
5a0e3ad6 | 7 | #include <linux/slab.h> |
c6adc9cc | 8 | #include <linux/blkdev.h> |
5dc562c5 | 9 | #include <linux/list_sort.h> |
c7f88c4e | 10 | #include <linux/iversion.h> |
9678c543 | 11 | #include "ctree.h" |
995946dd | 12 | #include "tree-log.h" |
e02119d5 CM |
13 | #include "disk-io.h" |
14 | #include "locking.h" | |
15 | #include "print-tree.h" | |
f186373f | 16 | #include "backref.h" |
ebb8765b | 17 | #include "compression.h" |
df2c95f3 | 18 | #include "qgroup.h" |
900c9981 | 19 | #include "inode-map.h" |
e02119d5 CM |
20 | |
21 | /* magic values for the inode_only field in btrfs_log_inode: | |
22 | * | |
23 | * LOG_INODE_ALL means to log everything | |
24 | * LOG_INODE_EXISTS means to log just enough to recreate the inode | |
25 | * during log replay | |
26 | */ | |
27 | #define LOG_INODE_ALL 0 | |
28 | #define LOG_INODE_EXISTS 1 | |
781feef7 | 29 | #define LOG_OTHER_INODE 2 |
e02119d5 | 30 | |
12fcfd22 CM |
31 | /* |
32 | * directory trouble cases | |
33 | * | |
34 | * 1) on rename or unlink, if the inode being unlinked isn't in the fsync | |
35 | * log, we must force a full commit before doing an fsync of the directory | |
36 | * where the unlink was done. | |
37 | * ---> record transid of last unlink/rename per directory | |
38 | * | |
39 | * mkdir foo/some_dir | |
40 | * normal commit | |
41 | * rename foo/some_dir foo2/some_dir | |
42 | * mkdir foo/some_dir | |
43 | * fsync foo/some_dir/some_file | |
44 | * | |
45 | * The fsync above will unlink the original some_dir without recording | |
46 | * it in its new location (foo2). After a crash, some_dir will be gone | |
47 | * unless the fsync of some_file forces a full commit | |
48 | * | |
49 | * 2) we must log any new names for any file or dir that is in the fsync | |
50 | * log. ---> check inode while renaming/linking. | |
51 | * | |
52 | * 2a) we must log any new names for any file or dir during rename | |
53 | * when the directory they are being removed from was logged. | |
54 | * ---> check inode and old parent dir during rename | |
55 | * | |
56 | * 2a is actually the more important variant. With the extra logging | |
57 | * a crash might unlink the old name without recreating the new one | |
58 | * | |
59 | * 3) after a crash, we must go through any directories with a link count | |
60 | * of zero and redo the rm -rf | |
61 | * | |
62 | * mkdir f1/foo | |
63 | * normal commit | |
64 | * rm -rf f1/foo | |
65 | * fsync(f1) | |
66 | * | |
67 | * The directory f1 was fully removed from the FS, but fsync was never | |
68 | * called on f1, only its parent dir. After a crash the rm -rf must | |
69 | * be replayed. This must be able to recurse down the entire | |
70 | * directory tree. The inode link count fixup code takes care of the | |
71 | * ugly details. | |
72 | */ | |
73 | ||
e02119d5 CM |
74 | /* |
75 | * stages for the tree walking. The first | |
76 | * stage (0) is to only pin down the blocks we find | |
77 | * the second stage (1) is to make sure that all the inodes | |
78 | * we find in the log are created in the subvolume. | |
79 | * | |
80 | * The last stage is to deal with directories and links and extents | |
81 | * and all the other fun semantics | |
82 | */ | |
83 | #define LOG_WALK_PIN_ONLY 0 | |
84 | #define LOG_WALK_REPLAY_INODES 1 | |
dd8e7217 JB |
85 | #define LOG_WALK_REPLAY_DIR_INDEX 2 |
86 | #define LOG_WALK_REPLAY_ALL 3 | |
e02119d5 | 87 | |
12fcfd22 | 88 | static int btrfs_log_inode(struct btrfs_trans_handle *trans, |
a59108a7 | 89 | struct btrfs_root *root, struct btrfs_inode *inode, |
49dae1bc FM |
90 | int inode_only, |
91 | const loff_t start, | |
8407f553 FM |
92 | const loff_t end, |
93 | struct btrfs_log_ctx *ctx); | |
ec051c0f YZ |
94 | static int link_to_fixup_dir(struct btrfs_trans_handle *trans, |
95 | struct btrfs_root *root, | |
96 | struct btrfs_path *path, u64 objectid); | |
12fcfd22 CM |
97 | static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans, |
98 | struct btrfs_root *root, | |
99 | struct btrfs_root *log, | |
100 | struct btrfs_path *path, | |
101 | u64 dirid, int del_all); | |
e02119d5 CM |
102 | |
103 | /* | |
104 | * tree logging is a special write ahead log used to make sure that | |
105 | * fsyncs and O_SYNCs can happen without doing full tree commits. | |
106 | * | |
107 | * Full tree commits are expensive because they require commonly | |
108 | * modified blocks to be recowed, creating many dirty pages in the | |
109 | * extent tree an 4x-6x higher write load than ext3. | |
110 | * | |
111 | * Instead of doing a tree commit on every fsync, we use the | |
112 | * key ranges and transaction ids to find items for a given file or directory | |
113 | * that have changed in this transaction. Those items are copied into | |
114 | * a special tree (one per subvolume root), that tree is written to disk | |
115 | * and then the fsync is considered complete. | |
116 | * | |
117 | * After a crash, items are copied out of the log-tree back into the | |
118 | * subvolume tree. Any file data extents found are recorded in the extent | |
119 | * allocation tree, and the log-tree freed. | |
120 | * | |
121 | * The log tree is read three times, once to pin down all the extents it is | |
122 | * using in ram and once, once to create all the inodes logged in the tree | |
123 | * and once to do all the other items. | |
124 | */ | |
125 | ||
e02119d5 CM |
126 | /* |
127 | * start a sub transaction and setup the log tree | |
128 | * this increments the log tree writer count to make the people | |
129 | * syncing the tree wait for us to finish | |
130 | */ | |
131 | static int start_log_trans(struct btrfs_trans_handle *trans, | |
8b050d35 MX |
132 | struct btrfs_root *root, |
133 | struct btrfs_log_ctx *ctx) | |
e02119d5 | 134 | { |
0b246afa | 135 | struct btrfs_fs_info *fs_info = root->fs_info; |
34eb2a52 | 136 | int ret = 0; |
7237f183 YZ |
137 | |
138 | mutex_lock(&root->log_mutex); | |
34eb2a52 | 139 | |
7237f183 | 140 | if (root->log_root) { |
0b246afa | 141 | if (btrfs_need_log_full_commit(fs_info, trans)) { |
50471a38 MX |
142 | ret = -EAGAIN; |
143 | goto out; | |
144 | } | |
34eb2a52 | 145 | |
ff782e0a | 146 | if (!root->log_start_pid) { |
27cdeb70 | 147 | clear_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state); |
34eb2a52 | 148 | root->log_start_pid = current->pid; |
ff782e0a | 149 | } else if (root->log_start_pid != current->pid) { |
27cdeb70 | 150 | set_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state); |
ff782e0a | 151 | } |
34eb2a52 | 152 | } else { |
0b246afa JM |
153 | mutex_lock(&fs_info->tree_log_mutex); |
154 | if (!fs_info->log_root_tree) | |
155 | ret = btrfs_init_log_root_tree(trans, fs_info); | |
156 | mutex_unlock(&fs_info->tree_log_mutex); | |
34eb2a52 Z |
157 | if (ret) |
158 | goto out; | |
ff782e0a | 159 | |
e02119d5 | 160 | ret = btrfs_add_log_tree(trans, root); |
4a500fd1 | 161 | if (ret) |
e87ac136 | 162 | goto out; |
34eb2a52 Z |
163 | |
164 | clear_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state); | |
165 | root->log_start_pid = current->pid; | |
e02119d5 | 166 | } |
34eb2a52 | 167 | |
2ecb7923 | 168 | atomic_inc(&root->log_batch); |
7237f183 | 169 | atomic_inc(&root->log_writers); |
8b050d35 | 170 | if (ctx) { |
34eb2a52 | 171 | int index = root->log_transid % 2; |
8b050d35 | 172 | list_add_tail(&ctx->list, &root->log_ctxs[index]); |
d1433deb | 173 | ctx->log_transid = root->log_transid; |
8b050d35 | 174 | } |
34eb2a52 | 175 | |
e87ac136 | 176 | out: |
7237f183 | 177 | mutex_unlock(&root->log_mutex); |
e87ac136 | 178 | return ret; |
e02119d5 CM |
179 | } |
180 | ||
181 | /* | |
182 | * returns 0 if there was a log transaction running and we were able | |
183 | * to join, or returns -ENOENT if there were not transactions | |
184 | * in progress | |
185 | */ | |
186 | static int join_running_log_trans(struct btrfs_root *root) | |
187 | { | |
188 | int ret = -ENOENT; | |
189 | ||
190 | smp_mb(); | |
191 | if (!root->log_root) | |
192 | return -ENOENT; | |
193 | ||
7237f183 | 194 | mutex_lock(&root->log_mutex); |
e02119d5 CM |
195 | if (root->log_root) { |
196 | ret = 0; | |
7237f183 | 197 | atomic_inc(&root->log_writers); |
e02119d5 | 198 | } |
7237f183 | 199 | mutex_unlock(&root->log_mutex); |
e02119d5 CM |
200 | return ret; |
201 | } | |
202 | ||
12fcfd22 CM |
203 | /* |
204 | * This either makes the current running log transaction wait | |
205 | * until you call btrfs_end_log_trans() or it makes any future | |
206 | * log transactions wait until you call btrfs_end_log_trans() | |
207 | */ | |
45128b08 | 208 | void btrfs_pin_log_trans(struct btrfs_root *root) |
12fcfd22 | 209 | { |
12fcfd22 CM |
210 | mutex_lock(&root->log_mutex); |
211 | atomic_inc(&root->log_writers); | |
212 | mutex_unlock(&root->log_mutex); | |
12fcfd22 CM |
213 | } |
214 | ||
e02119d5 CM |
215 | /* |
216 | * indicate we're done making changes to the log tree | |
217 | * and wake up anyone waiting to do a sync | |
218 | */ | |
143bede5 | 219 | void btrfs_end_log_trans(struct btrfs_root *root) |
e02119d5 | 220 | { |
7237f183 | 221 | if (atomic_dec_and_test(&root->log_writers)) { |
093258e6 DS |
222 | /* atomic_dec_and_test implies a barrier */ |
223 | cond_wake_up_nomb(&root->log_writer_wait); | |
7237f183 | 224 | } |
e02119d5 CM |
225 | } |
226 | ||
227 | ||
228 | /* | |
229 | * the walk control struct is used to pass state down the chain when | |
230 | * processing the log tree. The stage field tells us which part | |
231 | * of the log tree processing we are currently doing. The others | |
232 | * are state fields used for that specific part | |
233 | */ | |
234 | struct walk_control { | |
235 | /* should we free the extent on disk when done? This is used | |
236 | * at transaction commit time while freeing a log tree | |
237 | */ | |
238 | int free; | |
239 | ||
240 | /* should we write out the extent buffer? This is used | |
241 | * while flushing the log tree to disk during a sync | |
242 | */ | |
243 | int write; | |
244 | ||
245 | /* should we wait for the extent buffer io to finish? Also used | |
246 | * while flushing the log tree to disk for a sync | |
247 | */ | |
248 | int wait; | |
249 | ||
250 | /* pin only walk, we record which extents on disk belong to the | |
251 | * log trees | |
252 | */ | |
253 | int pin; | |
254 | ||
255 | /* what stage of the replay code we're currently in */ | |
256 | int stage; | |
257 | ||
258 | /* the root we are currently replaying */ | |
259 | struct btrfs_root *replay_dest; | |
260 | ||
261 | /* the trans handle for the current replay */ | |
262 | struct btrfs_trans_handle *trans; | |
263 | ||
264 | /* the function that gets used to process blocks we find in the | |
265 | * tree. Note the extent_buffer might not be up to date when it is | |
266 | * passed in, and it must be checked or read if you need the data | |
267 | * inside it | |
268 | */ | |
269 | int (*process_func)(struct btrfs_root *log, struct extent_buffer *eb, | |
581c1760 | 270 | struct walk_control *wc, u64 gen, int level); |
e02119d5 CM |
271 | }; |
272 | ||
273 | /* | |
274 | * process_func used to pin down extents, write them or wait on them | |
275 | */ | |
276 | static int process_one_buffer(struct btrfs_root *log, | |
277 | struct extent_buffer *eb, | |
581c1760 | 278 | struct walk_control *wc, u64 gen, int level) |
e02119d5 | 279 | { |
0b246afa | 280 | struct btrfs_fs_info *fs_info = log->fs_info; |
b50c6e25 JB |
281 | int ret = 0; |
282 | ||
8c2a1a30 JB |
283 | /* |
284 | * If this fs is mixed then we need to be able to process the leaves to | |
285 | * pin down any logged extents, so we have to read the block. | |
286 | */ | |
0b246afa | 287 | if (btrfs_fs_incompat(fs_info, MIXED_GROUPS)) { |
581c1760 | 288 | ret = btrfs_read_buffer(eb, gen, level, NULL); |
8c2a1a30 JB |
289 | if (ret) |
290 | return ret; | |
291 | } | |
292 | ||
04018de5 | 293 | if (wc->pin) |
2ff7e61e JM |
294 | ret = btrfs_pin_extent_for_log_replay(fs_info, eb->start, |
295 | eb->len); | |
e02119d5 | 296 | |
b50c6e25 | 297 | if (!ret && btrfs_buffer_uptodate(eb, gen, 0)) { |
8c2a1a30 | 298 | if (wc->pin && btrfs_header_level(eb) == 0) |
2ff7e61e | 299 | ret = btrfs_exclude_logged_extents(fs_info, eb); |
e02119d5 CM |
300 | if (wc->write) |
301 | btrfs_write_tree_block(eb); | |
302 | if (wc->wait) | |
303 | btrfs_wait_tree_block_writeback(eb); | |
304 | } | |
b50c6e25 | 305 | return ret; |
e02119d5 CM |
306 | } |
307 | ||
308 | /* | |
309 | * Item overwrite used by replay and tree logging. eb, slot and key all refer | |
310 | * to the src data we are copying out. | |
311 | * | |
312 | * root is the tree we are copying into, and path is a scratch | |
313 | * path for use in this function (it should be released on entry and | |
314 | * will be released on exit). | |
315 | * | |
316 | * If the key is already in the destination tree the existing item is | |
317 | * overwritten. If the existing item isn't big enough, it is extended. | |
318 | * If it is too large, it is truncated. | |
319 | * | |
320 | * If the key isn't in the destination yet, a new item is inserted. | |
321 | */ | |
322 | static noinline int overwrite_item(struct btrfs_trans_handle *trans, | |
323 | struct btrfs_root *root, | |
324 | struct btrfs_path *path, | |
325 | struct extent_buffer *eb, int slot, | |
326 | struct btrfs_key *key) | |
327 | { | |
2ff7e61e | 328 | struct btrfs_fs_info *fs_info = root->fs_info; |
e02119d5 CM |
329 | int ret; |
330 | u32 item_size; | |
331 | u64 saved_i_size = 0; | |
332 | int save_old_i_size = 0; | |
333 | unsigned long src_ptr; | |
334 | unsigned long dst_ptr; | |
335 | int overwrite_root = 0; | |
4bc4bee4 | 336 | bool inode_item = key->type == BTRFS_INODE_ITEM_KEY; |
e02119d5 CM |
337 | |
338 | if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) | |
339 | overwrite_root = 1; | |
340 | ||
341 | item_size = btrfs_item_size_nr(eb, slot); | |
342 | src_ptr = btrfs_item_ptr_offset(eb, slot); | |
343 | ||
344 | /* look for the key in the destination tree */ | |
345 | ret = btrfs_search_slot(NULL, root, key, path, 0, 0); | |
4bc4bee4 JB |
346 | if (ret < 0) |
347 | return ret; | |
348 | ||
e02119d5 CM |
349 | if (ret == 0) { |
350 | char *src_copy; | |
351 | char *dst_copy; | |
352 | u32 dst_size = btrfs_item_size_nr(path->nodes[0], | |
353 | path->slots[0]); | |
354 | if (dst_size != item_size) | |
355 | goto insert; | |
356 | ||
357 | if (item_size == 0) { | |
b3b4aa74 | 358 | btrfs_release_path(path); |
e02119d5 CM |
359 | return 0; |
360 | } | |
361 | dst_copy = kmalloc(item_size, GFP_NOFS); | |
362 | src_copy = kmalloc(item_size, GFP_NOFS); | |
2a29edc6 | 363 | if (!dst_copy || !src_copy) { |
b3b4aa74 | 364 | btrfs_release_path(path); |
2a29edc6 | 365 | kfree(dst_copy); |
366 | kfree(src_copy); | |
367 | return -ENOMEM; | |
368 | } | |
e02119d5 CM |
369 | |
370 | read_extent_buffer(eb, src_copy, src_ptr, item_size); | |
371 | ||
372 | dst_ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]); | |
373 | read_extent_buffer(path->nodes[0], dst_copy, dst_ptr, | |
374 | item_size); | |
375 | ret = memcmp(dst_copy, src_copy, item_size); | |
376 | ||
377 | kfree(dst_copy); | |
378 | kfree(src_copy); | |
379 | /* | |
380 | * they have the same contents, just return, this saves | |
381 | * us from cowing blocks in the destination tree and doing | |
382 | * extra writes that may not have been done by a previous | |
383 | * sync | |
384 | */ | |
385 | if (ret == 0) { | |
b3b4aa74 | 386 | btrfs_release_path(path); |
e02119d5 CM |
387 | return 0; |
388 | } | |
389 | ||
4bc4bee4 JB |
390 | /* |
391 | * We need to load the old nbytes into the inode so when we | |
392 | * replay the extents we've logged we get the right nbytes. | |
393 | */ | |
394 | if (inode_item) { | |
395 | struct btrfs_inode_item *item; | |
396 | u64 nbytes; | |
d555438b | 397 | u32 mode; |
4bc4bee4 JB |
398 | |
399 | item = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
400 | struct btrfs_inode_item); | |
401 | nbytes = btrfs_inode_nbytes(path->nodes[0], item); | |
402 | item = btrfs_item_ptr(eb, slot, | |
403 | struct btrfs_inode_item); | |
404 | btrfs_set_inode_nbytes(eb, item, nbytes); | |
d555438b JB |
405 | |
406 | /* | |
407 | * If this is a directory we need to reset the i_size to | |
408 | * 0 so that we can set it up properly when replaying | |
409 | * the rest of the items in this log. | |
410 | */ | |
411 | mode = btrfs_inode_mode(eb, item); | |
412 | if (S_ISDIR(mode)) | |
413 | btrfs_set_inode_size(eb, item, 0); | |
4bc4bee4 JB |
414 | } |
415 | } else if (inode_item) { | |
416 | struct btrfs_inode_item *item; | |
d555438b | 417 | u32 mode; |
4bc4bee4 JB |
418 | |
419 | /* | |
420 | * New inode, set nbytes to 0 so that the nbytes comes out | |
421 | * properly when we replay the extents. | |
422 | */ | |
423 | item = btrfs_item_ptr(eb, slot, struct btrfs_inode_item); | |
424 | btrfs_set_inode_nbytes(eb, item, 0); | |
d555438b JB |
425 | |
426 | /* | |
427 | * If this is a directory we need to reset the i_size to 0 so | |
428 | * that we can set it up properly when replaying the rest of | |
429 | * the items in this log. | |
430 | */ | |
431 | mode = btrfs_inode_mode(eb, item); | |
432 | if (S_ISDIR(mode)) | |
433 | btrfs_set_inode_size(eb, item, 0); | |
e02119d5 CM |
434 | } |
435 | insert: | |
b3b4aa74 | 436 | btrfs_release_path(path); |
e02119d5 | 437 | /* try to insert the key into the destination tree */ |
df8d116f | 438 | path->skip_release_on_error = 1; |
e02119d5 CM |
439 | ret = btrfs_insert_empty_item(trans, root, path, |
440 | key, item_size); | |
df8d116f | 441 | path->skip_release_on_error = 0; |
e02119d5 CM |
442 | |
443 | /* make sure any existing item is the correct size */ | |
df8d116f | 444 | if (ret == -EEXIST || ret == -EOVERFLOW) { |
e02119d5 CM |
445 | u32 found_size; |
446 | found_size = btrfs_item_size_nr(path->nodes[0], | |
447 | path->slots[0]); | |
143bede5 | 448 | if (found_size > item_size) |
2ff7e61e | 449 | btrfs_truncate_item(fs_info, path, item_size, 1); |
143bede5 | 450 | else if (found_size < item_size) |
2ff7e61e | 451 | btrfs_extend_item(fs_info, path, |
143bede5 | 452 | item_size - found_size); |
e02119d5 | 453 | } else if (ret) { |
4a500fd1 | 454 | return ret; |
e02119d5 CM |
455 | } |
456 | dst_ptr = btrfs_item_ptr_offset(path->nodes[0], | |
457 | path->slots[0]); | |
458 | ||
459 | /* don't overwrite an existing inode if the generation number | |
460 | * was logged as zero. This is done when the tree logging code | |
461 | * is just logging an inode to make sure it exists after recovery. | |
462 | * | |
463 | * Also, don't overwrite i_size on directories during replay. | |
464 | * log replay inserts and removes directory items based on the | |
465 | * state of the tree found in the subvolume, and i_size is modified | |
466 | * as it goes | |
467 | */ | |
468 | if (key->type == BTRFS_INODE_ITEM_KEY && ret == -EEXIST) { | |
469 | struct btrfs_inode_item *src_item; | |
470 | struct btrfs_inode_item *dst_item; | |
471 | ||
472 | src_item = (struct btrfs_inode_item *)src_ptr; | |
473 | dst_item = (struct btrfs_inode_item *)dst_ptr; | |
474 | ||
1a4bcf47 FM |
475 | if (btrfs_inode_generation(eb, src_item) == 0) { |
476 | struct extent_buffer *dst_eb = path->nodes[0]; | |
2f2ff0ee | 477 | const u64 ino_size = btrfs_inode_size(eb, src_item); |
1a4bcf47 | 478 | |
2f2ff0ee FM |
479 | /* |
480 | * For regular files an ino_size == 0 is used only when | |
481 | * logging that an inode exists, as part of a directory | |
482 | * fsync, and the inode wasn't fsynced before. In this | |
483 | * case don't set the size of the inode in the fs/subvol | |
484 | * tree, otherwise we would be throwing valid data away. | |
485 | */ | |
1a4bcf47 | 486 | if (S_ISREG(btrfs_inode_mode(eb, src_item)) && |
2f2ff0ee FM |
487 | S_ISREG(btrfs_inode_mode(dst_eb, dst_item)) && |
488 | ino_size != 0) { | |
1a4bcf47 | 489 | struct btrfs_map_token token; |
1a4bcf47 FM |
490 | |
491 | btrfs_init_map_token(&token); | |
492 | btrfs_set_token_inode_size(dst_eb, dst_item, | |
493 | ino_size, &token); | |
494 | } | |
e02119d5 | 495 | goto no_copy; |
1a4bcf47 | 496 | } |
e02119d5 CM |
497 | |
498 | if (overwrite_root && | |
499 | S_ISDIR(btrfs_inode_mode(eb, src_item)) && | |
500 | S_ISDIR(btrfs_inode_mode(path->nodes[0], dst_item))) { | |
501 | save_old_i_size = 1; | |
502 | saved_i_size = btrfs_inode_size(path->nodes[0], | |
503 | dst_item); | |
504 | } | |
505 | } | |
506 | ||
507 | copy_extent_buffer(path->nodes[0], eb, dst_ptr, | |
508 | src_ptr, item_size); | |
509 | ||
510 | if (save_old_i_size) { | |
511 | struct btrfs_inode_item *dst_item; | |
512 | dst_item = (struct btrfs_inode_item *)dst_ptr; | |
513 | btrfs_set_inode_size(path->nodes[0], dst_item, saved_i_size); | |
514 | } | |
515 | ||
516 | /* make sure the generation is filled in */ | |
517 | if (key->type == BTRFS_INODE_ITEM_KEY) { | |
518 | struct btrfs_inode_item *dst_item; | |
519 | dst_item = (struct btrfs_inode_item *)dst_ptr; | |
520 | if (btrfs_inode_generation(path->nodes[0], dst_item) == 0) { | |
521 | btrfs_set_inode_generation(path->nodes[0], dst_item, | |
522 | trans->transid); | |
523 | } | |
524 | } | |
525 | no_copy: | |
526 | btrfs_mark_buffer_dirty(path->nodes[0]); | |
b3b4aa74 | 527 | btrfs_release_path(path); |
e02119d5 CM |
528 | return 0; |
529 | } | |
530 | ||
531 | /* | |
532 | * simple helper to read an inode off the disk from a given root | |
533 | * This can only be called for subvolume roots and not for the log | |
534 | */ | |
535 | static noinline struct inode *read_one_inode(struct btrfs_root *root, | |
536 | u64 objectid) | |
537 | { | |
5d4f98a2 | 538 | struct btrfs_key key; |
e02119d5 | 539 | struct inode *inode; |
e02119d5 | 540 | |
5d4f98a2 YZ |
541 | key.objectid = objectid; |
542 | key.type = BTRFS_INODE_ITEM_KEY; | |
543 | key.offset = 0; | |
73f73415 | 544 | inode = btrfs_iget(root->fs_info->sb, &key, root, NULL); |
2e19f1f9 | 545 | if (IS_ERR(inode)) |
5d4f98a2 | 546 | inode = NULL; |
e02119d5 CM |
547 | return inode; |
548 | } | |
549 | ||
550 | /* replays a single extent in 'eb' at 'slot' with 'key' into the | |
551 | * subvolume 'root'. path is released on entry and should be released | |
552 | * on exit. | |
553 | * | |
554 | * extents in the log tree have not been allocated out of the extent | |
555 | * tree yet. So, this completes the allocation, taking a reference | |
556 | * as required if the extent already exists or creating a new extent | |
557 | * if it isn't in the extent allocation tree yet. | |
558 | * | |
559 | * The extent is inserted into the file, dropping any existing extents | |
560 | * from the file that overlap the new one. | |
561 | */ | |
562 | static noinline int replay_one_extent(struct btrfs_trans_handle *trans, | |
563 | struct btrfs_root *root, | |
564 | struct btrfs_path *path, | |
565 | struct extent_buffer *eb, int slot, | |
566 | struct btrfs_key *key) | |
567 | { | |
0b246afa | 568 | struct btrfs_fs_info *fs_info = root->fs_info; |
e02119d5 | 569 | int found_type; |
e02119d5 | 570 | u64 extent_end; |
e02119d5 | 571 | u64 start = key->offset; |
4bc4bee4 | 572 | u64 nbytes = 0; |
e02119d5 CM |
573 | struct btrfs_file_extent_item *item; |
574 | struct inode *inode = NULL; | |
575 | unsigned long size; | |
576 | int ret = 0; | |
577 | ||
578 | item = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item); | |
579 | found_type = btrfs_file_extent_type(eb, item); | |
580 | ||
d899e052 | 581 | if (found_type == BTRFS_FILE_EXTENT_REG || |
4bc4bee4 JB |
582 | found_type == BTRFS_FILE_EXTENT_PREALLOC) { |
583 | nbytes = btrfs_file_extent_num_bytes(eb, item); | |
584 | extent_end = start + nbytes; | |
585 | ||
586 | /* | |
587 | * We don't add to the inodes nbytes if we are prealloc or a | |
588 | * hole. | |
589 | */ | |
590 | if (btrfs_file_extent_disk_bytenr(eb, item) == 0) | |
591 | nbytes = 0; | |
592 | } else if (found_type == BTRFS_FILE_EXTENT_INLINE) { | |
e41ca589 | 593 | size = btrfs_file_extent_ram_bytes(eb, item); |
4bc4bee4 | 594 | nbytes = btrfs_file_extent_ram_bytes(eb, item); |
da17066c | 595 | extent_end = ALIGN(start + size, |
0b246afa | 596 | fs_info->sectorsize); |
e02119d5 CM |
597 | } else { |
598 | ret = 0; | |
599 | goto out; | |
600 | } | |
601 | ||
602 | inode = read_one_inode(root, key->objectid); | |
603 | if (!inode) { | |
604 | ret = -EIO; | |
605 | goto out; | |
606 | } | |
607 | ||
608 | /* | |
609 | * first check to see if we already have this extent in the | |
610 | * file. This must be done before the btrfs_drop_extents run | |
611 | * so we don't try to drop this extent. | |
612 | */ | |
f85b7379 DS |
613 | ret = btrfs_lookup_file_extent(trans, root, path, |
614 | btrfs_ino(BTRFS_I(inode)), start, 0); | |
e02119d5 | 615 | |
d899e052 YZ |
616 | if (ret == 0 && |
617 | (found_type == BTRFS_FILE_EXTENT_REG || | |
618 | found_type == BTRFS_FILE_EXTENT_PREALLOC)) { | |
e02119d5 CM |
619 | struct btrfs_file_extent_item cmp1; |
620 | struct btrfs_file_extent_item cmp2; | |
621 | struct btrfs_file_extent_item *existing; | |
622 | struct extent_buffer *leaf; | |
623 | ||
624 | leaf = path->nodes[0]; | |
625 | existing = btrfs_item_ptr(leaf, path->slots[0], | |
626 | struct btrfs_file_extent_item); | |
627 | ||
628 | read_extent_buffer(eb, &cmp1, (unsigned long)item, | |
629 | sizeof(cmp1)); | |
630 | read_extent_buffer(leaf, &cmp2, (unsigned long)existing, | |
631 | sizeof(cmp2)); | |
632 | ||
633 | /* | |
634 | * we already have a pointer to this exact extent, | |
635 | * we don't have to do anything | |
636 | */ | |
637 | if (memcmp(&cmp1, &cmp2, sizeof(cmp1)) == 0) { | |
b3b4aa74 | 638 | btrfs_release_path(path); |
e02119d5 CM |
639 | goto out; |
640 | } | |
641 | } | |
b3b4aa74 | 642 | btrfs_release_path(path); |
e02119d5 CM |
643 | |
644 | /* drop any overlapping extents */ | |
2671485d | 645 | ret = btrfs_drop_extents(trans, root, inode, start, extent_end, 1); |
3650860b JB |
646 | if (ret) |
647 | goto out; | |
e02119d5 | 648 | |
07d400a6 YZ |
649 | if (found_type == BTRFS_FILE_EXTENT_REG || |
650 | found_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
5d4f98a2 | 651 | u64 offset; |
07d400a6 YZ |
652 | unsigned long dest_offset; |
653 | struct btrfs_key ins; | |
654 | ||
3168021c FM |
655 | if (btrfs_file_extent_disk_bytenr(eb, item) == 0 && |
656 | btrfs_fs_incompat(fs_info, NO_HOLES)) | |
657 | goto update_inode; | |
658 | ||
07d400a6 YZ |
659 | ret = btrfs_insert_empty_item(trans, root, path, key, |
660 | sizeof(*item)); | |
3650860b JB |
661 | if (ret) |
662 | goto out; | |
07d400a6 YZ |
663 | dest_offset = btrfs_item_ptr_offset(path->nodes[0], |
664 | path->slots[0]); | |
665 | copy_extent_buffer(path->nodes[0], eb, dest_offset, | |
666 | (unsigned long)item, sizeof(*item)); | |
667 | ||
668 | ins.objectid = btrfs_file_extent_disk_bytenr(eb, item); | |
669 | ins.offset = btrfs_file_extent_disk_num_bytes(eb, item); | |
670 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
5d4f98a2 | 671 | offset = key->offset - btrfs_file_extent_offset(eb, item); |
07d400a6 | 672 | |
df2c95f3 QW |
673 | /* |
674 | * Manually record dirty extent, as here we did a shallow | |
675 | * file extent item copy and skip normal backref update, | |
676 | * but modifying extent tree all by ourselves. | |
677 | * So need to manually record dirty extent for qgroup, | |
678 | * as the owner of the file extent changed from log tree | |
679 | * (doesn't affect qgroup) to fs/file tree(affects qgroup) | |
680 | */ | |
a95f3aaf | 681 | ret = btrfs_qgroup_trace_extent(trans, |
df2c95f3 QW |
682 | btrfs_file_extent_disk_bytenr(eb, item), |
683 | btrfs_file_extent_disk_num_bytes(eb, item), | |
684 | GFP_NOFS); | |
685 | if (ret < 0) | |
686 | goto out; | |
687 | ||
07d400a6 YZ |
688 | if (ins.objectid > 0) { |
689 | u64 csum_start; | |
690 | u64 csum_end; | |
691 | LIST_HEAD(ordered_sums); | |
692 | /* | |
693 | * is this extent already allocated in the extent | |
694 | * allocation tree? If so, just add a reference | |
695 | */ | |
2ff7e61e | 696 | ret = btrfs_lookup_data_extent(fs_info, ins.objectid, |
07d400a6 YZ |
697 | ins.offset); |
698 | if (ret == 0) { | |
84f7d8e6 | 699 | ret = btrfs_inc_extent_ref(trans, root, |
07d400a6 | 700 | ins.objectid, ins.offset, |
5d4f98a2 | 701 | 0, root->root_key.objectid, |
b06c4bf5 | 702 | key->objectid, offset); |
b50c6e25 JB |
703 | if (ret) |
704 | goto out; | |
07d400a6 YZ |
705 | } else { |
706 | /* | |
707 | * insert the extent pointer in the extent | |
708 | * allocation tree | |
709 | */ | |
5d4f98a2 | 710 | ret = btrfs_alloc_logged_file_extent(trans, |
2ff7e61e | 711 | root->root_key.objectid, |
5d4f98a2 | 712 | key->objectid, offset, &ins); |
b50c6e25 JB |
713 | if (ret) |
714 | goto out; | |
07d400a6 | 715 | } |
b3b4aa74 | 716 | btrfs_release_path(path); |
07d400a6 YZ |
717 | |
718 | if (btrfs_file_extent_compression(eb, item)) { | |
719 | csum_start = ins.objectid; | |
720 | csum_end = csum_start + ins.offset; | |
721 | } else { | |
722 | csum_start = ins.objectid + | |
723 | btrfs_file_extent_offset(eb, item); | |
724 | csum_end = csum_start + | |
725 | btrfs_file_extent_num_bytes(eb, item); | |
726 | } | |
727 | ||
728 | ret = btrfs_lookup_csums_range(root->log_root, | |
729 | csum_start, csum_end - 1, | |
a2de733c | 730 | &ordered_sums, 0); |
3650860b JB |
731 | if (ret) |
732 | goto out; | |
b84b8390 FM |
733 | /* |
734 | * Now delete all existing cums in the csum root that | |
735 | * cover our range. We do this because we can have an | |
736 | * extent that is completely referenced by one file | |
737 | * extent item and partially referenced by another | |
738 | * file extent item (like after using the clone or | |
739 | * extent_same ioctls). In this case if we end up doing | |
740 | * the replay of the one that partially references the | |
741 | * extent first, and we do not do the csum deletion | |
742 | * below, we can get 2 csum items in the csum tree that | |
743 | * overlap each other. For example, imagine our log has | |
744 | * the two following file extent items: | |
745 | * | |
746 | * key (257 EXTENT_DATA 409600) | |
747 | * extent data disk byte 12845056 nr 102400 | |
748 | * extent data offset 20480 nr 20480 ram 102400 | |
749 | * | |
750 | * key (257 EXTENT_DATA 819200) | |
751 | * extent data disk byte 12845056 nr 102400 | |
752 | * extent data offset 0 nr 102400 ram 102400 | |
753 | * | |
754 | * Where the second one fully references the 100K extent | |
755 | * that starts at disk byte 12845056, and the log tree | |
756 | * has a single csum item that covers the entire range | |
757 | * of the extent: | |
758 | * | |
759 | * key (EXTENT_CSUM EXTENT_CSUM 12845056) itemsize 100 | |
760 | * | |
761 | * After the first file extent item is replayed, the | |
762 | * csum tree gets the following csum item: | |
763 | * | |
764 | * key (EXTENT_CSUM EXTENT_CSUM 12865536) itemsize 20 | |
765 | * | |
766 | * Which covers the 20K sub-range starting at offset 20K | |
767 | * of our extent. Now when we replay the second file | |
768 | * extent item, if we do not delete existing csum items | |
769 | * that cover any of its blocks, we end up getting two | |
770 | * csum items in our csum tree that overlap each other: | |
771 | * | |
772 | * key (EXTENT_CSUM EXTENT_CSUM 12845056) itemsize 100 | |
773 | * key (EXTENT_CSUM EXTENT_CSUM 12865536) itemsize 20 | |
774 | * | |
775 | * Which is a problem, because after this anyone trying | |
776 | * to lookup up for the checksum of any block of our | |
777 | * extent starting at an offset of 40K or higher, will | |
778 | * end up looking at the second csum item only, which | |
779 | * does not contain the checksum for any block starting | |
780 | * at offset 40K or higher of our extent. | |
781 | */ | |
07d400a6 YZ |
782 | while (!list_empty(&ordered_sums)) { |
783 | struct btrfs_ordered_sum *sums; | |
784 | sums = list_entry(ordered_sums.next, | |
785 | struct btrfs_ordered_sum, | |
786 | list); | |
b84b8390 | 787 | if (!ret) |
0b246afa | 788 | ret = btrfs_del_csums(trans, fs_info, |
5b4aacef JM |
789 | sums->bytenr, |
790 | sums->len); | |
3650860b JB |
791 | if (!ret) |
792 | ret = btrfs_csum_file_blocks(trans, | |
0b246afa | 793 | fs_info->csum_root, sums); |
07d400a6 YZ |
794 | list_del(&sums->list); |
795 | kfree(sums); | |
796 | } | |
3650860b JB |
797 | if (ret) |
798 | goto out; | |
07d400a6 | 799 | } else { |
b3b4aa74 | 800 | btrfs_release_path(path); |
07d400a6 YZ |
801 | } |
802 | } else if (found_type == BTRFS_FILE_EXTENT_INLINE) { | |
803 | /* inline extents are easy, we just overwrite them */ | |
804 | ret = overwrite_item(trans, root, path, eb, slot, key); | |
3650860b JB |
805 | if (ret) |
806 | goto out; | |
07d400a6 | 807 | } |
e02119d5 | 808 | |
4bc4bee4 | 809 | inode_add_bytes(inode, nbytes); |
3168021c | 810 | update_inode: |
b9959295 | 811 | ret = btrfs_update_inode(trans, root, inode); |
e02119d5 CM |
812 | out: |
813 | if (inode) | |
814 | iput(inode); | |
815 | return ret; | |
816 | } | |
817 | ||
818 | /* | |
819 | * when cleaning up conflicts between the directory names in the | |
820 | * subvolume, directory names in the log and directory names in the | |
821 | * inode back references, we may have to unlink inodes from directories. | |
822 | * | |
823 | * This is a helper function to do the unlink of a specific directory | |
824 | * item | |
825 | */ | |
826 | static noinline int drop_one_dir_item(struct btrfs_trans_handle *trans, | |
827 | struct btrfs_root *root, | |
828 | struct btrfs_path *path, | |
207e7d92 | 829 | struct btrfs_inode *dir, |
e02119d5 CM |
830 | struct btrfs_dir_item *di) |
831 | { | |
832 | struct inode *inode; | |
833 | char *name; | |
834 | int name_len; | |
835 | struct extent_buffer *leaf; | |
836 | struct btrfs_key location; | |
837 | int ret; | |
838 | ||
839 | leaf = path->nodes[0]; | |
840 | ||
841 | btrfs_dir_item_key_to_cpu(leaf, di, &location); | |
842 | name_len = btrfs_dir_name_len(leaf, di); | |
843 | name = kmalloc(name_len, GFP_NOFS); | |
2a29edc6 | 844 | if (!name) |
845 | return -ENOMEM; | |
846 | ||
e02119d5 | 847 | read_extent_buffer(leaf, name, (unsigned long)(di + 1), name_len); |
b3b4aa74 | 848 | btrfs_release_path(path); |
e02119d5 CM |
849 | |
850 | inode = read_one_inode(root, location.objectid); | |
c00e9493 | 851 | if (!inode) { |
3650860b JB |
852 | ret = -EIO; |
853 | goto out; | |
c00e9493 | 854 | } |
e02119d5 | 855 | |
ec051c0f | 856 | ret = link_to_fixup_dir(trans, root, path, location.objectid); |
3650860b JB |
857 | if (ret) |
858 | goto out; | |
12fcfd22 | 859 | |
207e7d92 NB |
860 | ret = btrfs_unlink_inode(trans, root, dir, BTRFS_I(inode), name, |
861 | name_len); | |
3650860b JB |
862 | if (ret) |
863 | goto out; | |
ada9af21 | 864 | else |
e5c304e6 | 865 | ret = btrfs_run_delayed_items(trans); |
3650860b | 866 | out: |
e02119d5 | 867 | kfree(name); |
e02119d5 CM |
868 | iput(inode); |
869 | return ret; | |
870 | } | |
871 | ||
872 | /* | |
873 | * helper function to see if a given name and sequence number found | |
874 | * in an inode back reference are already in a directory and correctly | |
875 | * point to this inode | |
876 | */ | |
877 | static noinline int inode_in_dir(struct btrfs_root *root, | |
878 | struct btrfs_path *path, | |
879 | u64 dirid, u64 objectid, u64 index, | |
880 | const char *name, int name_len) | |
881 | { | |
882 | struct btrfs_dir_item *di; | |
883 | struct btrfs_key location; | |
884 | int match = 0; | |
885 | ||
886 | di = btrfs_lookup_dir_index_item(NULL, root, path, dirid, | |
887 | index, name, name_len, 0); | |
888 | if (di && !IS_ERR(di)) { | |
889 | btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location); | |
890 | if (location.objectid != objectid) | |
891 | goto out; | |
892 | } else | |
893 | goto out; | |
b3b4aa74 | 894 | btrfs_release_path(path); |
e02119d5 CM |
895 | |
896 | di = btrfs_lookup_dir_item(NULL, root, path, dirid, name, name_len, 0); | |
897 | if (di && !IS_ERR(di)) { | |
898 | btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location); | |
899 | if (location.objectid != objectid) | |
900 | goto out; | |
901 | } else | |
902 | goto out; | |
903 | match = 1; | |
904 | out: | |
b3b4aa74 | 905 | btrfs_release_path(path); |
e02119d5 CM |
906 | return match; |
907 | } | |
908 | ||
909 | /* | |
910 | * helper function to check a log tree for a named back reference in | |
911 | * an inode. This is used to decide if a back reference that is | |
912 | * found in the subvolume conflicts with what we find in the log. | |
913 | * | |
914 | * inode backreferences may have multiple refs in a single item, | |
915 | * during replay we process one reference at a time, and we don't | |
916 | * want to delete valid links to a file from the subvolume if that | |
917 | * link is also in the log. | |
918 | */ | |
919 | static noinline int backref_in_log(struct btrfs_root *log, | |
920 | struct btrfs_key *key, | |
f186373f | 921 | u64 ref_objectid, |
df8d116f | 922 | const char *name, int namelen) |
e02119d5 CM |
923 | { |
924 | struct btrfs_path *path; | |
925 | struct btrfs_inode_ref *ref; | |
926 | unsigned long ptr; | |
927 | unsigned long ptr_end; | |
928 | unsigned long name_ptr; | |
929 | int found_name_len; | |
930 | int item_size; | |
931 | int ret; | |
932 | int match = 0; | |
933 | ||
934 | path = btrfs_alloc_path(); | |
2a29edc6 | 935 | if (!path) |
936 | return -ENOMEM; | |
937 | ||
e02119d5 CM |
938 | ret = btrfs_search_slot(NULL, log, key, path, 0, 0); |
939 | if (ret != 0) | |
940 | goto out; | |
941 | ||
e02119d5 | 942 | ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]); |
f186373f MF |
943 | |
944 | if (key->type == BTRFS_INODE_EXTREF_KEY) { | |
1f250e92 FM |
945 | if (btrfs_find_name_in_ext_backref(path->nodes[0], |
946 | path->slots[0], | |
947 | ref_objectid, | |
f186373f MF |
948 | name, namelen, NULL)) |
949 | match = 1; | |
950 | ||
951 | goto out; | |
952 | } | |
953 | ||
954 | item_size = btrfs_item_size_nr(path->nodes[0], path->slots[0]); | |
e02119d5 CM |
955 | ptr_end = ptr + item_size; |
956 | while (ptr < ptr_end) { | |
957 | ref = (struct btrfs_inode_ref *)ptr; | |
958 | found_name_len = btrfs_inode_ref_name_len(path->nodes[0], ref); | |
959 | if (found_name_len == namelen) { | |
960 | name_ptr = (unsigned long)(ref + 1); | |
961 | ret = memcmp_extent_buffer(path->nodes[0], name, | |
962 | name_ptr, namelen); | |
963 | if (ret == 0) { | |
964 | match = 1; | |
965 | goto out; | |
966 | } | |
967 | } | |
968 | ptr = (unsigned long)(ref + 1) + found_name_len; | |
969 | } | |
970 | out: | |
971 | btrfs_free_path(path); | |
972 | return match; | |
973 | } | |
974 | ||
5a1d7843 | 975 | static inline int __add_inode_ref(struct btrfs_trans_handle *trans, |
e02119d5 | 976 | struct btrfs_root *root, |
e02119d5 | 977 | struct btrfs_path *path, |
5a1d7843 | 978 | struct btrfs_root *log_root, |
94c91a1f NB |
979 | struct btrfs_inode *dir, |
980 | struct btrfs_inode *inode, | |
f186373f MF |
981 | u64 inode_objectid, u64 parent_objectid, |
982 | u64 ref_index, char *name, int namelen, | |
983 | int *search_done) | |
e02119d5 | 984 | { |
34f3e4f2 | 985 | int ret; |
f186373f MF |
986 | char *victim_name; |
987 | int victim_name_len; | |
988 | struct extent_buffer *leaf; | |
5a1d7843 | 989 | struct btrfs_dir_item *di; |
f186373f MF |
990 | struct btrfs_key search_key; |
991 | struct btrfs_inode_extref *extref; | |
c622ae60 | 992 | |
f186373f MF |
993 | again: |
994 | /* Search old style refs */ | |
995 | search_key.objectid = inode_objectid; | |
996 | search_key.type = BTRFS_INODE_REF_KEY; | |
997 | search_key.offset = parent_objectid; | |
998 | ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0); | |
e02119d5 | 999 | if (ret == 0) { |
e02119d5 CM |
1000 | struct btrfs_inode_ref *victim_ref; |
1001 | unsigned long ptr; | |
1002 | unsigned long ptr_end; | |
f186373f MF |
1003 | |
1004 | leaf = path->nodes[0]; | |
e02119d5 CM |
1005 | |
1006 | /* are we trying to overwrite a back ref for the root directory | |
1007 | * if so, just jump out, we're done | |
1008 | */ | |
f186373f | 1009 | if (search_key.objectid == search_key.offset) |
5a1d7843 | 1010 | return 1; |
e02119d5 CM |
1011 | |
1012 | /* check all the names in this back reference to see | |
1013 | * if they are in the log. if so, we allow them to stay | |
1014 | * otherwise they must be unlinked as a conflict | |
1015 | */ | |
1016 | ptr = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
1017 | ptr_end = ptr + btrfs_item_size_nr(leaf, path->slots[0]); | |
d397712b | 1018 | while (ptr < ptr_end) { |
e02119d5 CM |
1019 | victim_ref = (struct btrfs_inode_ref *)ptr; |
1020 | victim_name_len = btrfs_inode_ref_name_len(leaf, | |
1021 | victim_ref); | |
1022 | victim_name = kmalloc(victim_name_len, GFP_NOFS); | |
3650860b JB |
1023 | if (!victim_name) |
1024 | return -ENOMEM; | |
e02119d5 CM |
1025 | |
1026 | read_extent_buffer(leaf, victim_name, | |
1027 | (unsigned long)(victim_ref + 1), | |
1028 | victim_name_len); | |
1029 | ||
f186373f MF |
1030 | if (!backref_in_log(log_root, &search_key, |
1031 | parent_objectid, | |
1032 | victim_name, | |
e02119d5 | 1033 | victim_name_len)) { |
94c91a1f | 1034 | inc_nlink(&inode->vfs_inode); |
b3b4aa74 | 1035 | btrfs_release_path(path); |
12fcfd22 | 1036 | |
94c91a1f | 1037 | ret = btrfs_unlink_inode(trans, root, dir, inode, |
4ec5934e | 1038 | victim_name, victim_name_len); |
f186373f | 1039 | kfree(victim_name); |
3650860b JB |
1040 | if (ret) |
1041 | return ret; | |
e5c304e6 | 1042 | ret = btrfs_run_delayed_items(trans); |
ada9af21 FDBM |
1043 | if (ret) |
1044 | return ret; | |
f186373f MF |
1045 | *search_done = 1; |
1046 | goto again; | |
e02119d5 CM |
1047 | } |
1048 | kfree(victim_name); | |
f186373f | 1049 | |
e02119d5 CM |
1050 | ptr = (unsigned long)(victim_ref + 1) + victim_name_len; |
1051 | } | |
e02119d5 | 1052 | |
c622ae60 | 1053 | /* |
1054 | * NOTE: we have searched root tree and checked the | |
bb7ab3b9 | 1055 | * corresponding ref, it does not need to check again. |
c622ae60 | 1056 | */ |
5a1d7843 | 1057 | *search_done = 1; |
e02119d5 | 1058 | } |
b3b4aa74 | 1059 | btrfs_release_path(path); |
e02119d5 | 1060 | |
f186373f MF |
1061 | /* Same search but for extended refs */ |
1062 | extref = btrfs_lookup_inode_extref(NULL, root, path, name, namelen, | |
1063 | inode_objectid, parent_objectid, 0, | |
1064 | 0); | |
1065 | if (!IS_ERR_OR_NULL(extref)) { | |
1066 | u32 item_size; | |
1067 | u32 cur_offset = 0; | |
1068 | unsigned long base; | |
1069 | struct inode *victim_parent; | |
1070 | ||
1071 | leaf = path->nodes[0]; | |
1072 | ||
1073 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1074 | base = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
1075 | ||
1076 | while (cur_offset < item_size) { | |
dd9ef135 | 1077 | extref = (struct btrfs_inode_extref *)(base + cur_offset); |
f186373f MF |
1078 | |
1079 | victim_name_len = btrfs_inode_extref_name_len(leaf, extref); | |
1080 | ||
1081 | if (btrfs_inode_extref_parent(leaf, extref) != parent_objectid) | |
1082 | goto next; | |
1083 | ||
1084 | victim_name = kmalloc(victim_name_len, GFP_NOFS); | |
3650860b JB |
1085 | if (!victim_name) |
1086 | return -ENOMEM; | |
f186373f MF |
1087 | read_extent_buffer(leaf, victim_name, (unsigned long)&extref->name, |
1088 | victim_name_len); | |
1089 | ||
1090 | search_key.objectid = inode_objectid; | |
1091 | search_key.type = BTRFS_INODE_EXTREF_KEY; | |
1092 | search_key.offset = btrfs_extref_hash(parent_objectid, | |
1093 | victim_name, | |
1094 | victim_name_len); | |
1095 | ret = 0; | |
1096 | if (!backref_in_log(log_root, &search_key, | |
1097 | parent_objectid, victim_name, | |
1098 | victim_name_len)) { | |
1099 | ret = -ENOENT; | |
1100 | victim_parent = read_one_inode(root, | |
94c91a1f | 1101 | parent_objectid); |
f186373f | 1102 | if (victim_parent) { |
94c91a1f | 1103 | inc_nlink(&inode->vfs_inode); |
f186373f MF |
1104 | btrfs_release_path(path); |
1105 | ||
1106 | ret = btrfs_unlink_inode(trans, root, | |
4ec5934e | 1107 | BTRFS_I(victim_parent), |
94c91a1f | 1108 | inode, |
4ec5934e NB |
1109 | victim_name, |
1110 | victim_name_len); | |
ada9af21 FDBM |
1111 | if (!ret) |
1112 | ret = btrfs_run_delayed_items( | |
e5c304e6 | 1113 | trans); |
f186373f | 1114 | } |
f186373f MF |
1115 | iput(victim_parent); |
1116 | kfree(victim_name); | |
3650860b JB |
1117 | if (ret) |
1118 | return ret; | |
f186373f MF |
1119 | *search_done = 1; |
1120 | goto again; | |
1121 | } | |
1122 | kfree(victim_name); | |
f186373f MF |
1123 | next: |
1124 | cur_offset += victim_name_len + sizeof(*extref); | |
1125 | } | |
1126 | *search_done = 1; | |
1127 | } | |
1128 | btrfs_release_path(path); | |
1129 | ||
34f3e4f2 | 1130 | /* look for a conflicting sequence number */ |
94c91a1f | 1131 | di = btrfs_lookup_dir_index_item(trans, root, path, btrfs_ino(dir), |
f186373f | 1132 | ref_index, name, namelen, 0); |
34f3e4f2 | 1133 | if (di && !IS_ERR(di)) { |
94c91a1f | 1134 | ret = drop_one_dir_item(trans, root, path, dir, di); |
3650860b JB |
1135 | if (ret) |
1136 | return ret; | |
34f3e4f2 | 1137 | } |
1138 | btrfs_release_path(path); | |
1139 | ||
1140 | /* look for a conflicing name */ | |
94c91a1f | 1141 | di = btrfs_lookup_dir_item(trans, root, path, btrfs_ino(dir), |
34f3e4f2 | 1142 | name, namelen, 0); |
1143 | if (di && !IS_ERR(di)) { | |
94c91a1f | 1144 | ret = drop_one_dir_item(trans, root, path, dir, di); |
3650860b JB |
1145 | if (ret) |
1146 | return ret; | |
34f3e4f2 | 1147 | } |
1148 | btrfs_release_path(path); | |
1149 | ||
5a1d7843 JS |
1150 | return 0; |
1151 | } | |
e02119d5 | 1152 | |
bae15d95 QW |
1153 | static int extref_get_fields(struct extent_buffer *eb, unsigned long ref_ptr, |
1154 | u32 *namelen, char **name, u64 *index, | |
1155 | u64 *parent_objectid) | |
f186373f MF |
1156 | { |
1157 | struct btrfs_inode_extref *extref; | |
1158 | ||
1159 | extref = (struct btrfs_inode_extref *)ref_ptr; | |
1160 | ||
1161 | *namelen = btrfs_inode_extref_name_len(eb, extref); | |
1162 | *name = kmalloc(*namelen, GFP_NOFS); | |
1163 | if (*name == NULL) | |
1164 | return -ENOMEM; | |
1165 | ||
1166 | read_extent_buffer(eb, *name, (unsigned long)&extref->name, | |
1167 | *namelen); | |
1168 | ||
1f250e92 FM |
1169 | if (index) |
1170 | *index = btrfs_inode_extref_index(eb, extref); | |
f186373f MF |
1171 | if (parent_objectid) |
1172 | *parent_objectid = btrfs_inode_extref_parent(eb, extref); | |
1173 | ||
1174 | return 0; | |
1175 | } | |
1176 | ||
bae15d95 QW |
1177 | static int ref_get_fields(struct extent_buffer *eb, unsigned long ref_ptr, |
1178 | u32 *namelen, char **name, u64 *index) | |
f186373f MF |
1179 | { |
1180 | struct btrfs_inode_ref *ref; | |
1181 | ||
1182 | ref = (struct btrfs_inode_ref *)ref_ptr; | |
1183 | ||
1184 | *namelen = btrfs_inode_ref_name_len(eb, ref); | |
1185 | *name = kmalloc(*namelen, GFP_NOFS); | |
1186 | if (*name == NULL) | |
1187 | return -ENOMEM; | |
1188 | ||
1189 | read_extent_buffer(eb, *name, (unsigned long)(ref + 1), *namelen); | |
1190 | ||
1f250e92 FM |
1191 | if (index) |
1192 | *index = btrfs_inode_ref_index(eb, ref); | |
f186373f MF |
1193 | |
1194 | return 0; | |
1195 | } | |
1196 | ||
1f250e92 FM |
1197 | /* |
1198 | * Take an inode reference item from the log tree and iterate all names from the | |
1199 | * inode reference item in the subvolume tree with the same key (if it exists). | |
1200 | * For any name that is not in the inode reference item from the log tree, do a | |
1201 | * proper unlink of that name (that is, remove its entry from the inode | |
1202 | * reference item and both dir index keys). | |
1203 | */ | |
1204 | static int unlink_old_inode_refs(struct btrfs_trans_handle *trans, | |
1205 | struct btrfs_root *root, | |
1206 | struct btrfs_path *path, | |
1207 | struct btrfs_inode *inode, | |
1208 | struct extent_buffer *log_eb, | |
1209 | int log_slot, | |
1210 | struct btrfs_key *key) | |
1211 | { | |
1212 | int ret; | |
1213 | unsigned long ref_ptr; | |
1214 | unsigned long ref_end; | |
1215 | struct extent_buffer *eb; | |
1216 | ||
1217 | again: | |
1218 | btrfs_release_path(path); | |
1219 | ret = btrfs_search_slot(NULL, root, key, path, 0, 0); | |
1220 | if (ret > 0) { | |
1221 | ret = 0; | |
1222 | goto out; | |
1223 | } | |
1224 | if (ret < 0) | |
1225 | goto out; | |
1226 | ||
1227 | eb = path->nodes[0]; | |
1228 | ref_ptr = btrfs_item_ptr_offset(eb, path->slots[0]); | |
1229 | ref_end = ref_ptr + btrfs_item_size_nr(eb, path->slots[0]); | |
1230 | while (ref_ptr < ref_end) { | |
1231 | char *name = NULL; | |
1232 | int namelen; | |
1233 | u64 parent_id; | |
1234 | ||
1235 | if (key->type == BTRFS_INODE_EXTREF_KEY) { | |
1236 | ret = extref_get_fields(eb, ref_ptr, &namelen, &name, | |
1237 | NULL, &parent_id); | |
1238 | } else { | |
1239 | parent_id = key->offset; | |
1240 | ret = ref_get_fields(eb, ref_ptr, &namelen, &name, | |
1241 | NULL); | |
1242 | } | |
1243 | if (ret) | |
1244 | goto out; | |
1245 | ||
1246 | if (key->type == BTRFS_INODE_EXTREF_KEY) | |
1247 | ret = btrfs_find_name_in_ext_backref(log_eb, log_slot, | |
1248 | parent_id, name, | |
1249 | namelen, NULL); | |
1250 | else | |
1251 | ret = btrfs_find_name_in_backref(log_eb, log_slot, name, | |
1252 | namelen, NULL); | |
1253 | ||
1254 | if (!ret) { | |
1255 | struct inode *dir; | |
1256 | ||
1257 | btrfs_release_path(path); | |
1258 | dir = read_one_inode(root, parent_id); | |
1259 | if (!dir) { | |
1260 | ret = -ENOENT; | |
1261 | kfree(name); | |
1262 | goto out; | |
1263 | } | |
1264 | ret = btrfs_unlink_inode(trans, root, BTRFS_I(dir), | |
1265 | inode, name, namelen); | |
1266 | kfree(name); | |
1267 | iput(dir); | |
1268 | if (ret) | |
1269 | goto out; | |
1270 | goto again; | |
1271 | } | |
1272 | ||
1273 | kfree(name); | |
1274 | ref_ptr += namelen; | |
1275 | if (key->type == BTRFS_INODE_EXTREF_KEY) | |
1276 | ref_ptr += sizeof(struct btrfs_inode_extref); | |
1277 | else | |
1278 | ref_ptr += sizeof(struct btrfs_inode_ref); | |
1279 | } | |
1280 | ret = 0; | |
1281 | out: | |
1282 | btrfs_release_path(path); | |
1283 | return ret; | |
1284 | } | |
1285 | ||
0d836392 FM |
1286 | static int btrfs_inode_ref_exists(struct inode *inode, struct inode *dir, |
1287 | const u8 ref_type, const char *name, | |
1288 | const int namelen) | |
1289 | { | |
1290 | struct btrfs_key key; | |
1291 | struct btrfs_path *path; | |
1292 | const u64 parent_id = btrfs_ino(BTRFS_I(dir)); | |
1293 | int ret; | |
1294 | ||
1295 | path = btrfs_alloc_path(); | |
1296 | if (!path) | |
1297 | return -ENOMEM; | |
1298 | ||
1299 | key.objectid = btrfs_ino(BTRFS_I(inode)); | |
1300 | key.type = ref_type; | |
1301 | if (key.type == BTRFS_INODE_REF_KEY) | |
1302 | key.offset = parent_id; | |
1303 | else | |
1304 | key.offset = btrfs_extref_hash(parent_id, name, namelen); | |
1305 | ||
1306 | ret = btrfs_search_slot(NULL, BTRFS_I(inode)->root, &key, path, 0, 0); | |
1307 | if (ret < 0) | |
1308 | goto out; | |
1309 | if (ret > 0) { | |
1310 | ret = 0; | |
1311 | goto out; | |
1312 | } | |
1313 | if (key.type == BTRFS_INODE_EXTREF_KEY) | |
1314 | ret = btrfs_find_name_in_ext_backref(path->nodes[0], | |
1315 | path->slots[0], parent_id, | |
1316 | name, namelen, NULL); | |
1317 | else | |
1318 | ret = btrfs_find_name_in_backref(path->nodes[0], path->slots[0], | |
1319 | name, namelen, NULL); | |
1320 | ||
1321 | out: | |
1322 | btrfs_free_path(path); | |
1323 | return ret; | |
1324 | } | |
1325 | ||
5a1d7843 JS |
1326 | /* |
1327 | * replay one inode back reference item found in the log tree. | |
1328 | * eb, slot and key refer to the buffer and key found in the log tree. | |
1329 | * root is the destination we are replaying into, and path is for temp | |
1330 | * use by this function. (it should be released on return). | |
1331 | */ | |
1332 | static noinline int add_inode_ref(struct btrfs_trans_handle *trans, | |
1333 | struct btrfs_root *root, | |
1334 | struct btrfs_root *log, | |
1335 | struct btrfs_path *path, | |
1336 | struct extent_buffer *eb, int slot, | |
1337 | struct btrfs_key *key) | |
1338 | { | |
03b2f08b GB |
1339 | struct inode *dir = NULL; |
1340 | struct inode *inode = NULL; | |
5a1d7843 JS |
1341 | unsigned long ref_ptr; |
1342 | unsigned long ref_end; | |
03b2f08b | 1343 | char *name = NULL; |
5a1d7843 JS |
1344 | int namelen; |
1345 | int ret; | |
1346 | int search_done = 0; | |
f186373f MF |
1347 | int log_ref_ver = 0; |
1348 | u64 parent_objectid; | |
1349 | u64 inode_objectid; | |
f46dbe3d | 1350 | u64 ref_index = 0; |
f186373f MF |
1351 | int ref_struct_size; |
1352 | ||
1353 | ref_ptr = btrfs_item_ptr_offset(eb, slot); | |
1354 | ref_end = ref_ptr + btrfs_item_size_nr(eb, slot); | |
1355 | ||
1356 | if (key->type == BTRFS_INODE_EXTREF_KEY) { | |
1357 | struct btrfs_inode_extref *r; | |
1358 | ||
1359 | ref_struct_size = sizeof(struct btrfs_inode_extref); | |
1360 | log_ref_ver = 1; | |
1361 | r = (struct btrfs_inode_extref *)ref_ptr; | |
1362 | parent_objectid = btrfs_inode_extref_parent(eb, r); | |
1363 | } else { | |
1364 | ref_struct_size = sizeof(struct btrfs_inode_ref); | |
1365 | parent_objectid = key->offset; | |
1366 | } | |
1367 | inode_objectid = key->objectid; | |
e02119d5 | 1368 | |
5a1d7843 JS |
1369 | /* |
1370 | * it is possible that we didn't log all the parent directories | |
1371 | * for a given inode. If we don't find the dir, just don't | |
1372 | * copy the back ref in. The link count fixup code will take | |
1373 | * care of the rest | |
1374 | */ | |
f186373f | 1375 | dir = read_one_inode(root, parent_objectid); |
03b2f08b GB |
1376 | if (!dir) { |
1377 | ret = -ENOENT; | |
1378 | goto out; | |
1379 | } | |
5a1d7843 | 1380 | |
f186373f | 1381 | inode = read_one_inode(root, inode_objectid); |
5a1d7843 | 1382 | if (!inode) { |
03b2f08b GB |
1383 | ret = -EIO; |
1384 | goto out; | |
5a1d7843 JS |
1385 | } |
1386 | ||
5a1d7843 | 1387 | while (ref_ptr < ref_end) { |
f186373f | 1388 | if (log_ref_ver) { |
bae15d95 QW |
1389 | ret = extref_get_fields(eb, ref_ptr, &namelen, &name, |
1390 | &ref_index, &parent_objectid); | |
f186373f MF |
1391 | /* |
1392 | * parent object can change from one array | |
1393 | * item to another. | |
1394 | */ | |
1395 | if (!dir) | |
1396 | dir = read_one_inode(root, parent_objectid); | |
03b2f08b GB |
1397 | if (!dir) { |
1398 | ret = -ENOENT; | |
1399 | goto out; | |
1400 | } | |
f186373f | 1401 | } else { |
bae15d95 QW |
1402 | ret = ref_get_fields(eb, ref_ptr, &namelen, &name, |
1403 | &ref_index); | |
f186373f MF |
1404 | } |
1405 | if (ret) | |
03b2f08b | 1406 | goto out; |
5a1d7843 JS |
1407 | |
1408 | /* if we already have a perfect match, we're done */ | |
f85b7379 DS |
1409 | if (!inode_in_dir(root, path, btrfs_ino(BTRFS_I(dir)), |
1410 | btrfs_ino(BTRFS_I(inode)), ref_index, | |
1411 | name, namelen)) { | |
5a1d7843 JS |
1412 | /* |
1413 | * look for a conflicting back reference in the | |
1414 | * metadata. if we find one we have to unlink that name | |
1415 | * of the file before we add our new link. Later on, we | |
1416 | * overwrite any existing back reference, and we don't | |
1417 | * want to create dangling pointers in the directory. | |
1418 | */ | |
1419 | ||
1420 | if (!search_done) { | |
1421 | ret = __add_inode_ref(trans, root, path, log, | |
94c91a1f | 1422 | BTRFS_I(dir), |
d75eefdf | 1423 | BTRFS_I(inode), |
f186373f MF |
1424 | inode_objectid, |
1425 | parent_objectid, | |
1426 | ref_index, name, namelen, | |
5a1d7843 | 1427 | &search_done); |
03b2f08b GB |
1428 | if (ret) { |
1429 | if (ret == 1) | |
1430 | ret = 0; | |
3650860b JB |
1431 | goto out; |
1432 | } | |
5a1d7843 JS |
1433 | } |
1434 | ||
0d836392 FM |
1435 | /* |
1436 | * If a reference item already exists for this inode | |
1437 | * with the same parent and name, but different index, | |
1438 | * drop it and the corresponding directory index entries | |
1439 | * from the parent before adding the new reference item | |
1440 | * and dir index entries, otherwise we would fail with | |
1441 | * -EEXIST returned from btrfs_add_link() below. | |
1442 | */ | |
1443 | ret = btrfs_inode_ref_exists(inode, dir, key->type, | |
1444 | name, namelen); | |
1445 | if (ret > 0) { | |
1446 | ret = btrfs_unlink_inode(trans, root, | |
1447 | BTRFS_I(dir), | |
1448 | BTRFS_I(inode), | |
1449 | name, namelen); | |
1450 | /* | |
1451 | * If we dropped the link count to 0, bump it so | |
1452 | * that later the iput() on the inode will not | |
1453 | * free it. We will fixup the link count later. | |
1454 | */ | |
1455 | if (!ret && inode->i_nlink == 0) | |
1456 | inc_nlink(inode); | |
1457 | } | |
1458 | if (ret < 0) | |
1459 | goto out; | |
1460 | ||
5a1d7843 | 1461 | /* insert our name */ |
db0a669f NB |
1462 | ret = btrfs_add_link(trans, BTRFS_I(dir), |
1463 | BTRFS_I(inode), | |
1464 | name, namelen, 0, ref_index); | |
3650860b JB |
1465 | if (ret) |
1466 | goto out; | |
5a1d7843 JS |
1467 | |
1468 | btrfs_update_inode(trans, root, inode); | |
1469 | } | |
1470 | ||
f186373f | 1471 | ref_ptr = (unsigned long)(ref_ptr + ref_struct_size) + namelen; |
5a1d7843 | 1472 | kfree(name); |
03b2f08b | 1473 | name = NULL; |
f186373f MF |
1474 | if (log_ref_ver) { |
1475 | iput(dir); | |
1476 | dir = NULL; | |
1477 | } | |
5a1d7843 | 1478 | } |
e02119d5 | 1479 | |
1f250e92 FM |
1480 | /* |
1481 | * Before we overwrite the inode reference item in the subvolume tree | |
1482 | * with the item from the log tree, we must unlink all names from the | |
1483 | * parent directory that are in the subvolume's tree inode reference | |
1484 | * item, otherwise we end up with an inconsistent subvolume tree where | |
1485 | * dir index entries exist for a name but there is no inode reference | |
1486 | * item with the same name. | |
1487 | */ | |
1488 | ret = unlink_old_inode_refs(trans, root, path, BTRFS_I(inode), eb, slot, | |
1489 | key); | |
1490 | if (ret) | |
1491 | goto out; | |
1492 | ||
e02119d5 CM |
1493 | /* finally write the back reference in the inode */ |
1494 | ret = overwrite_item(trans, root, path, eb, slot, key); | |
5a1d7843 | 1495 | out: |
b3b4aa74 | 1496 | btrfs_release_path(path); |
03b2f08b | 1497 | kfree(name); |
e02119d5 CM |
1498 | iput(dir); |
1499 | iput(inode); | |
3650860b | 1500 | return ret; |
e02119d5 CM |
1501 | } |
1502 | ||
c71bf099 | 1503 | static int insert_orphan_item(struct btrfs_trans_handle *trans, |
9c4f61f0 | 1504 | struct btrfs_root *root, u64 ino) |
c71bf099 YZ |
1505 | { |
1506 | int ret; | |
381cf658 | 1507 | |
9c4f61f0 DS |
1508 | ret = btrfs_insert_orphan_item(trans, root, ino); |
1509 | if (ret == -EEXIST) | |
1510 | ret = 0; | |
381cf658 | 1511 | |
c71bf099 YZ |
1512 | return ret; |
1513 | } | |
1514 | ||
f186373f | 1515 | static int count_inode_extrefs(struct btrfs_root *root, |
36283658 | 1516 | struct btrfs_inode *inode, struct btrfs_path *path) |
f186373f MF |
1517 | { |
1518 | int ret = 0; | |
1519 | int name_len; | |
1520 | unsigned int nlink = 0; | |
1521 | u32 item_size; | |
1522 | u32 cur_offset = 0; | |
36283658 | 1523 | u64 inode_objectid = btrfs_ino(inode); |
f186373f MF |
1524 | u64 offset = 0; |
1525 | unsigned long ptr; | |
1526 | struct btrfs_inode_extref *extref; | |
1527 | struct extent_buffer *leaf; | |
1528 | ||
1529 | while (1) { | |
1530 | ret = btrfs_find_one_extref(root, inode_objectid, offset, path, | |
1531 | &extref, &offset); | |
1532 | if (ret) | |
1533 | break; | |
c71bf099 | 1534 | |
f186373f MF |
1535 | leaf = path->nodes[0]; |
1536 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1537 | ptr = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
2c2c452b | 1538 | cur_offset = 0; |
f186373f MF |
1539 | |
1540 | while (cur_offset < item_size) { | |
1541 | extref = (struct btrfs_inode_extref *) (ptr + cur_offset); | |
1542 | name_len = btrfs_inode_extref_name_len(leaf, extref); | |
1543 | ||
1544 | nlink++; | |
1545 | ||
1546 | cur_offset += name_len + sizeof(*extref); | |
1547 | } | |
1548 | ||
1549 | offset++; | |
1550 | btrfs_release_path(path); | |
1551 | } | |
1552 | btrfs_release_path(path); | |
1553 | ||
2c2c452b | 1554 | if (ret < 0 && ret != -ENOENT) |
f186373f MF |
1555 | return ret; |
1556 | return nlink; | |
1557 | } | |
1558 | ||
1559 | static int count_inode_refs(struct btrfs_root *root, | |
f329e319 | 1560 | struct btrfs_inode *inode, struct btrfs_path *path) |
e02119d5 | 1561 | { |
e02119d5 CM |
1562 | int ret; |
1563 | struct btrfs_key key; | |
f186373f | 1564 | unsigned int nlink = 0; |
e02119d5 CM |
1565 | unsigned long ptr; |
1566 | unsigned long ptr_end; | |
1567 | int name_len; | |
f329e319 | 1568 | u64 ino = btrfs_ino(inode); |
e02119d5 | 1569 | |
33345d01 | 1570 | key.objectid = ino; |
e02119d5 CM |
1571 | key.type = BTRFS_INODE_REF_KEY; |
1572 | key.offset = (u64)-1; | |
1573 | ||
d397712b | 1574 | while (1) { |
e02119d5 CM |
1575 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
1576 | if (ret < 0) | |
1577 | break; | |
1578 | if (ret > 0) { | |
1579 | if (path->slots[0] == 0) | |
1580 | break; | |
1581 | path->slots[0]--; | |
1582 | } | |
e93ae26f | 1583 | process_slot: |
e02119d5 CM |
1584 | btrfs_item_key_to_cpu(path->nodes[0], &key, |
1585 | path->slots[0]); | |
33345d01 | 1586 | if (key.objectid != ino || |
e02119d5 CM |
1587 | key.type != BTRFS_INODE_REF_KEY) |
1588 | break; | |
1589 | ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]); | |
1590 | ptr_end = ptr + btrfs_item_size_nr(path->nodes[0], | |
1591 | path->slots[0]); | |
d397712b | 1592 | while (ptr < ptr_end) { |
e02119d5 CM |
1593 | struct btrfs_inode_ref *ref; |
1594 | ||
1595 | ref = (struct btrfs_inode_ref *)ptr; | |
1596 | name_len = btrfs_inode_ref_name_len(path->nodes[0], | |
1597 | ref); | |
1598 | ptr = (unsigned long)(ref + 1) + name_len; | |
1599 | nlink++; | |
1600 | } | |
1601 | ||
1602 | if (key.offset == 0) | |
1603 | break; | |
e93ae26f FDBM |
1604 | if (path->slots[0] > 0) { |
1605 | path->slots[0]--; | |
1606 | goto process_slot; | |
1607 | } | |
e02119d5 | 1608 | key.offset--; |
b3b4aa74 | 1609 | btrfs_release_path(path); |
e02119d5 | 1610 | } |
b3b4aa74 | 1611 | btrfs_release_path(path); |
f186373f MF |
1612 | |
1613 | return nlink; | |
1614 | } | |
1615 | ||
1616 | /* | |
1617 | * There are a few corners where the link count of the file can't | |
1618 | * be properly maintained during replay. So, instead of adding | |
1619 | * lots of complexity to the log code, we just scan the backrefs | |
1620 | * for any file that has been through replay. | |
1621 | * | |
1622 | * The scan will update the link count on the inode to reflect the | |
1623 | * number of back refs found. If it goes down to zero, the iput | |
1624 | * will free the inode. | |
1625 | */ | |
1626 | static noinline int fixup_inode_link_count(struct btrfs_trans_handle *trans, | |
1627 | struct btrfs_root *root, | |
1628 | struct inode *inode) | |
1629 | { | |
1630 | struct btrfs_path *path; | |
1631 | int ret; | |
1632 | u64 nlink = 0; | |
4a0cc7ca | 1633 | u64 ino = btrfs_ino(BTRFS_I(inode)); |
f186373f MF |
1634 | |
1635 | path = btrfs_alloc_path(); | |
1636 | if (!path) | |
1637 | return -ENOMEM; | |
1638 | ||
f329e319 | 1639 | ret = count_inode_refs(root, BTRFS_I(inode), path); |
f186373f MF |
1640 | if (ret < 0) |
1641 | goto out; | |
1642 | ||
1643 | nlink = ret; | |
1644 | ||
36283658 | 1645 | ret = count_inode_extrefs(root, BTRFS_I(inode), path); |
f186373f MF |
1646 | if (ret < 0) |
1647 | goto out; | |
1648 | ||
1649 | nlink += ret; | |
1650 | ||
1651 | ret = 0; | |
1652 | ||
e02119d5 | 1653 | if (nlink != inode->i_nlink) { |
bfe86848 | 1654 | set_nlink(inode, nlink); |
e02119d5 CM |
1655 | btrfs_update_inode(trans, root, inode); |
1656 | } | |
8d5bf1cb | 1657 | BTRFS_I(inode)->index_cnt = (u64)-1; |
e02119d5 | 1658 | |
c71bf099 YZ |
1659 | if (inode->i_nlink == 0) { |
1660 | if (S_ISDIR(inode->i_mode)) { | |
1661 | ret = replay_dir_deletes(trans, root, NULL, path, | |
33345d01 | 1662 | ino, 1); |
3650860b JB |
1663 | if (ret) |
1664 | goto out; | |
c71bf099 | 1665 | } |
33345d01 | 1666 | ret = insert_orphan_item(trans, root, ino); |
12fcfd22 | 1667 | } |
12fcfd22 | 1668 | |
f186373f MF |
1669 | out: |
1670 | btrfs_free_path(path); | |
1671 | return ret; | |
e02119d5 CM |
1672 | } |
1673 | ||
1674 | static noinline int fixup_inode_link_counts(struct btrfs_trans_handle *trans, | |
1675 | struct btrfs_root *root, | |
1676 | struct btrfs_path *path) | |
1677 | { | |
1678 | int ret; | |
1679 | struct btrfs_key key; | |
1680 | struct inode *inode; | |
1681 | ||
1682 | key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID; | |
1683 | key.type = BTRFS_ORPHAN_ITEM_KEY; | |
1684 | key.offset = (u64)-1; | |
d397712b | 1685 | while (1) { |
e02119d5 CM |
1686 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
1687 | if (ret < 0) | |
1688 | break; | |
1689 | ||
1690 | if (ret == 1) { | |
1691 | if (path->slots[0] == 0) | |
1692 | break; | |
1693 | path->slots[0]--; | |
1694 | } | |
1695 | ||
1696 | btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]); | |
1697 | if (key.objectid != BTRFS_TREE_LOG_FIXUP_OBJECTID || | |
1698 | key.type != BTRFS_ORPHAN_ITEM_KEY) | |
1699 | break; | |
1700 | ||
1701 | ret = btrfs_del_item(trans, root, path); | |
65a246c5 TI |
1702 | if (ret) |
1703 | goto out; | |
e02119d5 | 1704 | |
b3b4aa74 | 1705 | btrfs_release_path(path); |
e02119d5 | 1706 | inode = read_one_inode(root, key.offset); |
c00e9493 TI |
1707 | if (!inode) |
1708 | return -EIO; | |
e02119d5 CM |
1709 | |
1710 | ret = fixup_inode_link_count(trans, root, inode); | |
e02119d5 | 1711 | iput(inode); |
3650860b JB |
1712 | if (ret) |
1713 | goto out; | |
e02119d5 | 1714 | |
12fcfd22 CM |
1715 | /* |
1716 | * fixup on a directory may create new entries, | |
1717 | * make sure we always look for the highset possible | |
1718 | * offset | |
1719 | */ | |
1720 | key.offset = (u64)-1; | |
e02119d5 | 1721 | } |
65a246c5 TI |
1722 | ret = 0; |
1723 | out: | |
b3b4aa74 | 1724 | btrfs_release_path(path); |
65a246c5 | 1725 | return ret; |
e02119d5 CM |
1726 | } |
1727 | ||
1728 | ||
1729 | /* | |
1730 | * record a given inode in the fixup dir so we can check its link | |
1731 | * count when replay is done. The link count is incremented here | |
1732 | * so the inode won't go away until we check it | |
1733 | */ | |
1734 | static noinline int link_to_fixup_dir(struct btrfs_trans_handle *trans, | |
1735 | struct btrfs_root *root, | |
1736 | struct btrfs_path *path, | |
1737 | u64 objectid) | |
1738 | { | |
1739 | struct btrfs_key key; | |
1740 | int ret = 0; | |
1741 | struct inode *inode; | |
1742 | ||
1743 | inode = read_one_inode(root, objectid); | |
c00e9493 TI |
1744 | if (!inode) |
1745 | return -EIO; | |
e02119d5 CM |
1746 | |
1747 | key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID; | |
962a298f | 1748 | key.type = BTRFS_ORPHAN_ITEM_KEY; |
e02119d5 CM |
1749 | key.offset = objectid; |
1750 | ||
1751 | ret = btrfs_insert_empty_item(trans, root, path, &key, 0); | |
1752 | ||
b3b4aa74 | 1753 | btrfs_release_path(path); |
e02119d5 | 1754 | if (ret == 0) { |
9bf7a489 JB |
1755 | if (!inode->i_nlink) |
1756 | set_nlink(inode, 1); | |
1757 | else | |
8b558c5f | 1758 | inc_nlink(inode); |
b9959295 | 1759 | ret = btrfs_update_inode(trans, root, inode); |
e02119d5 CM |
1760 | } else if (ret == -EEXIST) { |
1761 | ret = 0; | |
1762 | } else { | |
3650860b | 1763 | BUG(); /* Logic Error */ |
e02119d5 CM |
1764 | } |
1765 | iput(inode); | |
1766 | ||
1767 | return ret; | |
1768 | } | |
1769 | ||
1770 | /* | |
1771 | * when replaying the log for a directory, we only insert names | |
1772 | * for inodes that actually exist. This means an fsync on a directory | |
1773 | * does not implicitly fsync all the new files in it | |
1774 | */ | |
1775 | static noinline int insert_one_name(struct btrfs_trans_handle *trans, | |
1776 | struct btrfs_root *root, | |
e02119d5 | 1777 | u64 dirid, u64 index, |
60d53eb3 | 1778 | char *name, int name_len, |
e02119d5 CM |
1779 | struct btrfs_key *location) |
1780 | { | |
1781 | struct inode *inode; | |
1782 | struct inode *dir; | |
1783 | int ret; | |
1784 | ||
1785 | inode = read_one_inode(root, location->objectid); | |
1786 | if (!inode) | |
1787 | return -ENOENT; | |
1788 | ||
1789 | dir = read_one_inode(root, dirid); | |
1790 | if (!dir) { | |
1791 | iput(inode); | |
1792 | return -EIO; | |
1793 | } | |
d555438b | 1794 | |
db0a669f NB |
1795 | ret = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode), name, |
1796 | name_len, 1, index); | |
e02119d5 CM |
1797 | |
1798 | /* FIXME, put inode into FIXUP list */ | |
1799 | ||
1800 | iput(inode); | |
1801 | iput(dir); | |
1802 | return ret; | |
1803 | } | |
1804 | ||
df8d116f FM |
1805 | /* |
1806 | * Return true if an inode reference exists in the log for the given name, | |
1807 | * inode and parent inode. | |
1808 | */ | |
1809 | static bool name_in_log_ref(struct btrfs_root *log_root, | |
1810 | const char *name, const int name_len, | |
1811 | const u64 dirid, const u64 ino) | |
1812 | { | |
1813 | struct btrfs_key search_key; | |
1814 | ||
1815 | search_key.objectid = ino; | |
1816 | search_key.type = BTRFS_INODE_REF_KEY; | |
1817 | search_key.offset = dirid; | |
1818 | if (backref_in_log(log_root, &search_key, dirid, name, name_len)) | |
1819 | return true; | |
1820 | ||
1821 | search_key.type = BTRFS_INODE_EXTREF_KEY; | |
1822 | search_key.offset = btrfs_extref_hash(dirid, name, name_len); | |
1823 | if (backref_in_log(log_root, &search_key, dirid, name, name_len)) | |
1824 | return true; | |
1825 | ||
1826 | return false; | |
1827 | } | |
1828 | ||
e02119d5 CM |
1829 | /* |
1830 | * take a single entry in a log directory item and replay it into | |
1831 | * the subvolume. | |
1832 | * | |
1833 | * if a conflicting item exists in the subdirectory already, | |
1834 | * the inode it points to is unlinked and put into the link count | |
1835 | * fix up tree. | |
1836 | * | |
1837 | * If a name from the log points to a file or directory that does | |
1838 | * not exist in the FS, it is skipped. fsyncs on directories | |
1839 | * do not force down inodes inside that directory, just changes to the | |
1840 | * names or unlinks in a directory. | |
bb53eda9 FM |
1841 | * |
1842 | * Returns < 0 on error, 0 if the name wasn't replayed (dentry points to a | |
1843 | * non-existing inode) and 1 if the name was replayed. | |
e02119d5 CM |
1844 | */ |
1845 | static noinline int replay_one_name(struct btrfs_trans_handle *trans, | |
1846 | struct btrfs_root *root, | |
1847 | struct btrfs_path *path, | |
1848 | struct extent_buffer *eb, | |
1849 | struct btrfs_dir_item *di, | |
1850 | struct btrfs_key *key) | |
1851 | { | |
1852 | char *name; | |
1853 | int name_len; | |
1854 | struct btrfs_dir_item *dst_di; | |
1855 | struct btrfs_key found_key; | |
1856 | struct btrfs_key log_key; | |
1857 | struct inode *dir; | |
e02119d5 | 1858 | u8 log_type; |
4bef0848 | 1859 | int exists; |
3650860b | 1860 | int ret = 0; |
d555438b | 1861 | bool update_size = (key->type == BTRFS_DIR_INDEX_KEY); |
bb53eda9 | 1862 | bool name_added = false; |
e02119d5 CM |
1863 | |
1864 | dir = read_one_inode(root, key->objectid); | |
c00e9493 TI |
1865 | if (!dir) |
1866 | return -EIO; | |
e02119d5 CM |
1867 | |
1868 | name_len = btrfs_dir_name_len(eb, di); | |
1869 | name = kmalloc(name_len, GFP_NOFS); | |
2bac325e FDBM |
1870 | if (!name) { |
1871 | ret = -ENOMEM; | |
1872 | goto out; | |
1873 | } | |
2a29edc6 | 1874 | |
e02119d5 CM |
1875 | log_type = btrfs_dir_type(eb, di); |
1876 | read_extent_buffer(eb, name, (unsigned long)(di + 1), | |
1877 | name_len); | |
1878 | ||
1879 | btrfs_dir_item_key_to_cpu(eb, di, &log_key); | |
4bef0848 CM |
1880 | exists = btrfs_lookup_inode(trans, root, path, &log_key, 0); |
1881 | if (exists == 0) | |
1882 | exists = 1; | |
1883 | else | |
1884 | exists = 0; | |
b3b4aa74 | 1885 | btrfs_release_path(path); |
4bef0848 | 1886 | |
e02119d5 CM |
1887 | if (key->type == BTRFS_DIR_ITEM_KEY) { |
1888 | dst_di = btrfs_lookup_dir_item(trans, root, path, key->objectid, | |
1889 | name, name_len, 1); | |
d397712b | 1890 | } else if (key->type == BTRFS_DIR_INDEX_KEY) { |
e02119d5 CM |
1891 | dst_di = btrfs_lookup_dir_index_item(trans, root, path, |
1892 | key->objectid, | |
1893 | key->offset, name, | |
1894 | name_len, 1); | |
1895 | } else { | |
3650860b JB |
1896 | /* Corruption */ |
1897 | ret = -EINVAL; | |
1898 | goto out; | |
e02119d5 | 1899 | } |
c704005d | 1900 | if (IS_ERR_OR_NULL(dst_di)) { |
e02119d5 CM |
1901 | /* we need a sequence number to insert, so we only |
1902 | * do inserts for the BTRFS_DIR_INDEX_KEY types | |
1903 | */ | |
1904 | if (key->type != BTRFS_DIR_INDEX_KEY) | |
1905 | goto out; | |
1906 | goto insert; | |
1907 | } | |
1908 | ||
1909 | btrfs_dir_item_key_to_cpu(path->nodes[0], dst_di, &found_key); | |
1910 | /* the existing item matches the logged item */ | |
1911 | if (found_key.objectid == log_key.objectid && | |
1912 | found_key.type == log_key.type && | |
1913 | found_key.offset == log_key.offset && | |
1914 | btrfs_dir_type(path->nodes[0], dst_di) == log_type) { | |
a2cc11db | 1915 | update_size = false; |
e02119d5 CM |
1916 | goto out; |
1917 | } | |
1918 | ||
1919 | /* | |
1920 | * don't drop the conflicting directory entry if the inode | |
1921 | * for the new entry doesn't exist | |
1922 | */ | |
4bef0848 | 1923 | if (!exists) |
e02119d5 CM |
1924 | goto out; |
1925 | ||
207e7d92 | 1926 | ret = drop_one_dir_item(trans, root, path, BTRFS_I(dir), dst_di); |
3650860b JB |
1927 | if (ret) |
1928 | goto out; | |
e02119d5 CM |
1929 | |
1930 | if (key->type == BTRFS_DIR_INDEX_KEY) | |
1931 | goto insert; | |
1932 | out: | |
b3b4aa74 | 1933 | btrfs_release_path(path); |
d555438b | 1934 | if (!ret && update_size) { |
6ef06d27 | 1935 | btrfs_i_size_write(BTRFS_I(dir), dir->i_size + name_len * 2); |
d555438b JB |
1936 | ret = btrfs_update_inode(trans, root, dir); |
1937 | } | |
e02119d5 CM |
1938 | kfree(name); |
1939 | iput(dir); | |
bb53eda9 FM |
1940 | if (!ret && name_added) |
1941 | ret = 1; | |
3650860b | 1942 | return ret; |
e02119d5 CM |
1943 | |
1944 | insert: | |
df8d116f FM |
1945 | if (name_in_log_ref(root->log_root, name, name_len, |
1946 | key->objectid, log_key.objectid)) { | |
1947 | /* The dentry will be added later. */ | |
1948 | ret = 0; | |
1949 | update_size = false; | |
1950 | goto out; | |
1951 | } | |
b3b4aa74 | 1952 | btrfs_release_path(path); |
60d53eb3 Z |
1953 | ret = insert_one_name(trans, root, key->objectid, key->offset, |
1954 | name, name_len, &log_key); | |
df8d116f | 1955 | if (ret && ret != -ENOENT && ret != -EEXIST) |
3650860b | 1956 | goto out; |
bb53eda9 FM |
1957 | if (!ret) |
1958 | name_added = true; | |
d555438b | 1959 | update_size = false; |
3650860b | 1960 | ret = 0; |
e02119d5 CM |
1961 | goto out; |
1962 | } | |
1963 | ||
1964 | /* | |
1965 | * find all the names in a directory item and reconcile them into | |
1966 | * the subvolume. Only BTRFS_DIR_ITEM_KEY types will have more than | |
1967 | * one name in a directory item, but the same code gets used for | |
1968 | * both directory index types | |
1969 | */ | |
1970 | static noinline int replay_one_dir_item(struct btrfs_trans_handle *trans, | |
1971 | struct btrfs_root *root, | |
1972 | struct btrfs_path *path, | |
1973 | struct extent_buffer *eb, int slot, | |
1974 | struct btrfs_key *key) | |
1975 | { | |
bb53eda9 | 1976 | int ret = 0; |
e02119d5 CM |
1977 | u32 item_size = btrfs_item_size_nr(eb, slot); |
1978 | struct btrfs_dir_item *di; | |
1979 | int name_len; | |
1980 | unsigned long ptr; | |
1981 | unsigned long ptr_end; | |
bb53eda9 | 1982 | struct btrfs_path *fixup_path = NULL; |
e02119d5 CM |
1983 | |
1984 | ptr = btrfs_item_ptr_offset(eb, slot); | |
1985 | ptr_end = ptr + item_size; | |
d397712b | 1986 | while (ptr < ptr_end) { |
e02119d5 CM |
1987 | di = (struct btrfs_dir_item *)ptr; |
1988 | name_len = btrfs_dir_name_len(eb, di); | |
1989 | ret = replay_one_name(trans, root, path, eb, di, key); | |
bb53eda9 FM |
1990 | if (ret < 0) |
1991 | break; | |
e02119d5 CM |
1992 | ptr = (unsigned long)(di + 1); |
1993 | ptr += name_len; | |
bb53eda9 FM |
1994 | |
1995 | /* | |
1996 | * If this entry refers to a non-directory (directories can not | |
1997 | * have a link count > 1) and it was added in the transaction | |
1998 | * that was not committed, make sure we fixup the link count of | |
1999 | * the inode it the entry points to. Otherwise something like | |
2000 | * the following would result in a directory pointing to an | |
2001 | * inode with a wrong link that does not account for this dir | |
2002 | * entry: | |
2003 | * | |
2004 | * mkdir testdir | |
2005 | * touch testdir/foo | |
2006 | * touch testdir/bar | |
2007 | * sync | |
2008 | * | |
2009 | * ln testdir/bar testdir/bar_link | |
2010 | * ln testdir/foo testdir/foo_link | |
2011 | * xfs_io -c "fsync" testdir/bar | |
2012 | * | |
2013 | * <power failure> | |
2014 | * | |
2015 | * mount fs, log replay happens | |
2016 | * | |
2017 | * File foo would remain with a link count of 1 when it has two | |
2018 | * entries pointing to it in the directory testdir. This would | |
2019 | * make it impossible to ever delete the parent directory has | |
2020 | * it would result in stale dentries that can never be deleted. | |
2021 | */ | |
2022 | if (ret == 1 && btrfs_dir_type(eb, di) != BTRFS_FT_DIR) { | |
2023 | struct btrfs_key di_key; | |
2024 | ||
2025 | if (!fixup_path) { | |
2026 | fixup_path = btrfs_alloc_path(); | |
2027 | if (!fixup_path) { | |
2028 | ret = -ENOMEM; | |
2029 | break; | |
2030 | } | |
2031 | } | |
2032 | ||
2033 | btrfs_dir_item_key_to_cpu(eb, di, &di_key); | |
2034 | ret = link_to_fixup_dir(trans, root, fixup_path, | |
2035 | di_key.objectid); | |
2036 | if (ret) | |
2037 | break; | |
2038 | } | |
2039 | ret = 0; | |
e02119d5 | 2040 | } |
bb53eda9 FM |
2041 | btrfs_free_path(fixup_path); |
2042 | return ret; | |
e02119d5 CM |
2043 | } |
2044 | ||
2045 | /* | |
2046 | * directory replay has two parts. There are the standard directory | |
2047 | * items in the log copied from the subvolume, and range items | |
2048 | * created in the log while the subvolume was logged. | |
2049 | * | |
2050 | * The range items tell us which parts of the key space the log | |
2051 | * is authoritative for. During replay, if a key in the subvolume | |
2052 | * directory is in a logged range item, but not actually in the log | |
2053 | * that means it was deleted from the directory before the fsync | |
2054 | * and should be removed. | |
2055 | */ | |
2056 | static noinline int find_dir_range(struct btrfs_root *root, | |
2057 | struct btrfs_path *path, | |
2058 | u64 dirid, int key_type, | |
2059 | u64 *start_ret, u64 *end_ret) | |
2060 | { | |
2061 | struct btrfs_key key; | |
2062 | u64 found_end; | |
2063 | struct btrfs_dir_log_item *item; | |
2064 | int ret; | |
2065 | int nritems; | |
2066 | ||
2067 | if (*start_ret == (u64)-1) | |
2068 | return 1; | |
2069 | ||
2070 | key.objectid = dirid; | |
2071 | key.type = key_type; | |
2072 | key.offset = *start_ret; | |
2073 | ||
2074 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
2075 | if (ret < 0) | |
2076 | goto out; | |
2077 | if (ret > 0) { | |
2078 | if (path->slots[0] == 0) | |
2079 | goto out; | |
2080 | path->slots[0]--; | |
2081 | } | |
2082 | if (ret != 0) | |
2083 | btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]); | |
2084 | ||
2085 | if (key.type != key_type || key.objectid != dirid) { | |
2086 | ret = 1; | |
2087 | goto next; | |
2088 | } | |
2089 | item = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
2090 | struct btrfs_dir_log_item); | |
2091 | found_end = btrfs_dir_log_end(path->nodes[0], item); | |
2092 | ||
2093 | if (*start_ret >= key.offset && *start_ret <= found_end) { | |
2094 | ret = 0; | |
2095 | *start_ret = key.offset; | |
2096 | *end_ret = found_end; | |
2097 | goto out; | |
2098 | } | |
2099 | ret = 1; | |
2100 | next: | |
2101 | /* check the next slot in the tree to see if it is a valid item */ | |
2102 | nritems = btrfs_header_nritems(path->nodes[0]); | |
2a7bf53f | 2103 | path->slots[0]++; |
e02119d5 CM |
2104 | if (path->slots[0] >= nritems) { |
2105 | ret = btrfs_next_leaf(root, path); | |
2106 | if (ret) | |
2107 | goto out; | |
e02119d5 CM |
2108 | } |
2109 | ||
2110 | btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]); | |
2111 | ||
2112 | if (key.type != key_type || key.objectid != dirid) { | |
2113 | ret = 1; | |
2114 | goto out; | |
2115 | } | |
2116 | item = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
2117 | struct btrfs_dir_log_item); | |
2118 | found_end = btrfs_dir_log_end(path->nodes[0], item); | |
2119 | *start_ret = key.offset; | |
2120 | *end_ret = found_end; | |
2121 | ret = 0; | |
2122 | out: | |
b3b4aa74 | 2123 | btrfs_release_path(path); |
e02119d5 CM |
2124 | return ret; |
2125 | } | |
2126 | ||
2127 | /* | |
2128 | * this looks for a given directory item in the log. If the directory | |
2129 | * item is not in the log, the item is removed and the inode it points | |
2130 | * to is unlinked | |
2131 | */ | |
2132 | static noinline int check_item_in_log(struct btrfs_trans_handle *trans, | |
2133 | struct btrfs_root *root, | |
2134 | struct btrfs_root *log, | |
2135 | struct btrfs_path *path, | |
2136 | struct btrfs_path *log_path, | |
2137 | struct inode *dir, | |
2138 | struct btrfs_key *dir_key) | |
2139 | { | |
2140 | int ret; | |
2141 | struct extent_buffer *eb; | |
2142 | int slot; | |
2143 | u32 item_size; | |
2144 | struct btrfs_dir_item *di; | |
2145 | struct btrfs_dir_item *log_di; | |
2146 | int name_len; | |
2147 | unsigned long ptr; | |
2148 | unsigned long ptr_end; | |
2149 | char *name; | |
2150 | struct inode *inode; | |
2151 | struct btrfs_key location; | |
2152 | ||
2153 | again: | |
2154 | eb = path->nodes[0]; | |
2155 | slot = path->slots[0]; | |
2156 | item_size = btrfs_item_size_nr(eb, slot); | |
2157 | ptr = btrfs_item_ptr_offset(eb, slot); | |
2158 | ptr_end = ptr + item_size; | |
d397712b | 2159 | while (ptr < ptr_end) { |
e02119d5 CM |
2160 | di = (struct btrfs_dir_item *)ptr; |
2161 | name_len = btrfs_dir_name_len(eb, di); | |
2162 | name = kmalloc(name_len, GFP_NOFS); | |
2163 | if (!name) { | |
2164 | ret = -ENOMEM; | |
2165 | goto out; | |
2166 | } | |
2167 | read_extent_buffer(eb, name, (unsigned long)(di + 1), | |
2168 | name_len); | |
2169 | log_di = NULL; | |
12fcfd22 | 2170 | if (log && dir_key->type == BTRFS_DIR_ITEM_KEY) { |
e02119d5 CM |
2171 | log_di = btrfs_lookup_dir_item(trans, log, log_path, |
2172 | dir_key->objectid, | |
2173 | name, name_len, 0); | |
12fcfd22 | 2174 | } else if (log && dir_key->type == BTRFS_DIR_INDEX_KEY) { |
e02119d5 CM |
2175 | log_di = btrfs_lookup_dir_index_item(trans, log, |
2176 | log_path, | |
2177 | dir_key->objectid, | |
2178 | dir_key->offset, | |
2179 | name, name_len, 0); | |
2180 | } | |
8d9e220c | 2181 | if (!log_di || log_di == ERR_PTR(-ENOENT)) { |
e02119d5 | 2182 | btrfs_dir_item_key_to_cpu(eb, di, &location); |
b3b4aa74 DS |
2183 | btrfs_release_path(path); |
2184 | btrfs_release_path(log_path); | |
e02119d5 | 2185 | inode = read_one_inode(root, location.objectid); |
c00e9493 TI |
2186 | if (!inode) { |
2187 | kfree(name); | |
2188 | return -EIO; | |
2189 | } | |
e02119d5 CM |
2190 | |
2191 | ret = link_to_fixup_dir(trans, root, | |
2192 | path, location.objectid); | |
3650860b JB |
2193 | if (ret) { |
2194 | kfree(name); | |
2195 | iput(inode); | |
2196 | goto out; | |
2197 | } | |
2198 | ||
8b558c5f | 2199 | inc_nlink(inode); |
4ec5934e NB |
2200 | ret = btrfs_unlink_inode(trans, root, BTRFS_I(dir), |
2201 | BTRFS_I(inode), name, name_len); | |
3650860b | 2202 | if (!ret) |
e5c304e6 | 2203 | ret = btrfs_run_delayed_items(trans); |
e02119d5 CM |
2204 | kfree(name); |
2205 | iput(inode); | |
3650860b JB |
2206 | if (ret) |
2207 | goto out; | |
e02119d5 CM |
2208 | |
2209 | /* there might still be more names under this key | |
2210 | * check and repeat if required | |
2211 | */ | |
2212 | ret = btrfs_search_slot(NULL, root, dir_key, path, | |
2213 | 0, 0); | |
2214 | if (ret == 0) | |
2215 | goto again; | |
2216 | ret = 0; | |
2217 | goto out; | |
269d040f FDBM |
2218 | } else if (IS_ERR(log_di)) { |
2219 | kfree(name); | |
2220 | return PTR_ERR(log_di); | |
e02119d5 | 2221 | } |
b3b4aa74 | 2222 | btrfs_release_path(log_path); |
e02119d5 CM |
2223 | kfree(name); |
2224 | ||
2225 | ptr = (unsigned long)(di + 1); | |
2226 | ptr += name_len; | |
2227 | } | |
2228 | ret = 0; | |
2229 | out: | |
b3b4aa74 DS |
2230 | btrfs_release_path(path); |
2231 | btrfs_release_path(log_path); | |
e02119d5 CM |
2232 | return ret; |
2233 | } | |
2234 | ||
4f764e51 FM |
2235 | static int replay_xattr_deletes(struct btrfs_trans_handle *trans, |
2236 | struct btrfs_root *root, | |
2237 | struct btrfs_root *log, | |
2238 | struct btrfs_path *path, | |
2239 | const u64 ino) | |
2240 | { | |
2241 | struct btrfs_key search_key; | |
2242 | struct btrfs_path *log_path; | |
2243 | int i; | |
2244 | int nritems; | |
2245 | int ret; | |
2246 | ||
2247 | log_path = btrfs_alloc_path(); | |
2248 | if (!log_path) | |
2249 | return -ENOMEM; | |
2250 | ||
2251 | search_key.objectid = ino; | |
2252 | search_key.type = BTRFS_XATTR_ITEM_KEY; | |
2253 | search_key.offset = 0; | |
2254 | again: | |
2255 | ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0); | |
2256 | if (ret < 0) | |
2257 | goto out; | |
2258 | process_leaf: | |
2259 | nritems = btrfs_header_nritems(path->nodes[0]); | |
2260 | for (i = path->slots[0]; i < nritems; i++) { | |
2261 | struct btrfs_key key; | |
2262 | struct btrfs_dir_item *di; | |
2263 | struct btrfs_dir_item *log_di; | |
2264 | u32 total_size; | |
2265 | u32 cur; | |
2266 | ||
2267 | btrfs_item_key_to_cpu(path->nodes[0], &key, i); | |
2268 | if (key.objectid != ino || key.type != BTRFS_XATTR_ITEM_KEY) { | |
2269 | ret = 0; | |
2270 | goto out; | |
2271 | } | |
2272 | ||
2273 | di = btrfs_item_ptr(path->nodes[0], i, struct btrfs_dir_item); | |
2274 | total_size = btrfs_item_size_nr(path->nodes[0], i); | |
2275 | cur = 0; | |
2276 | while (cur < total_size) { | |
2277 | u16 name_len = btrfs_dir_name_len(path->nodes[0], di); | |
2278 | u16 data_len = btrfs_dir_data_len(path->nodes[0], di); | |
2279 | u32 this_len = sizeof(*di) + name_len + data_len; | |
2280 | char *name; | |
2281 | ||
2282 | name = kmalloc(name_len, GFP_NOFS); | |
2283 | if (!name) { | |
2284 | ret = -ENOMEM; | |
2285 | goto out; | |
2286 | } | |
2287 | read_extent_buffer(path->nodes[0], name, | |
2288 | (unsigned long)(di + 1), name_len); | |
2289 | ||
2290 | log_di = btrfs_lookup_xattr(NULL, log, log_path, ino, | |
2291 | name, name_len, 0); | |
2292 | btrfs_release_path(log_path); | |
2293 | if (!log_di) { | |
2294 | /* Doesn't exist in log tree, so delete it. */ | |
2295 | btrfs_release_path(path); | |
2296 | di = btrfs_lookup_xattr(trans, root, path, ino, | |
2297 | name, name_len, -1); | |
2298 | kfree(name); | |
2299 | if (IS_ERR(di)) { | |
2300 | ret = PTR_ERR(di); | |
2301 | goto out; | |
2302 | } | |
2303 | ASSERT(di); | |
2304 | ret = btrfs_delete_one_dir_name(trans, root, | |
2305 | path, di); | |
2306 | if (ret) | |
2307 | goto out; | |
2308 | btrfs_release_path(path); | |
2309 | search_key = key; | |
2310 | goto again; | |
2311 | } | |
2312 | kfree(name); | |
2313 | if (IS_ERR(log_di)) { | |
2314 | ret = PTR_ERR(log_di); | |
2315 | goto out; | |
2316 | } | |
2317 | cur += this_len; | |
2318 | di = (struct btrfs_dir_item *)((char *)di + this_len); | |
2319 | } | |
2320 | } | |
2321 | ret = btrfs_next_leaf(root, path); | |
2322 | if (ret > 0) | |
2323 | ret = 0; | |
2324 | else if (ret == 0) | |
2325 | goto process_leaf; | |
2326 | out: | |
2327 | btrfs_free_path(log_path); | |
2328 | btrfs_release_path(path); | |
2329 | return ret; | |
2330 | } | |
2331 | ||
2332 | ||
e02119d5 CM |
2333 | /* |
2334 | * deletion replay happens before we copy any new directory items | |
2335 | * out of the log or out of backreferences from inodes. It | |
2336 | * scans the log to find ranges of keys that log is authoritative for, | |
2337 | * and then scans the directory to find items in those ranges that are | |
2338 | * not present in the log. | |
2339 | * | |
2340 | * Anything we don't find in the log is unlinked and removed from the | |
2341 | * directory. | |
2342 | */ | |
2343 | static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans, | |
2344 | struct btrfs_root *root, | |
2345 | struct btrfs_root *log, | |
2346 | struct btrfs_path *path, | |
12fcfd22 | 2347 | u64 dirid, int del_all) |
e02119d5 CM |
2348 | { |
2349 | u64 range_start; | |
2350 | u64 range_end; | |
2351 | int key_type = BTRFS_DIR_LOG_ITEM_KEY; | |
2352 | int ret = 0; | |
2353 | struct btrfs_key dir_key; | |
2354 | struct btrfs_key found_key; | |
2355 | struct btrfs_path *log_path; | |
2356 | struct inode *dir; | |
2357 | ||
2358 | dir_key.objectid = dirid; | |
2359 | dir_key.type = BTRFS_DIR_ITEM_KEY; | |
2360 | log_path = btrfs_alloc_path(); | |
2361 | if (!log_path) | |
2362 | return -ENOMEM; | |
2363 | ||
2364 | dir = read_one_inode(root, dirid); | |
2365 | /* it isn't an error if the inode isn't there, that can happen | |
2366 | * because we replay the deletes before we copy in the inode item | |
2367 | * from the log | |
2368 | */ | |
2369 | if (!dir) { | |
2370 | btrfs_free_path(log_path); | |
2371 | return 0; | |
2372 | } | |
2373 | again: | |
2374 | range_start = 0; | |
2375 | range_end = 0; | |
d397712b | 2376 | while (1) { |
12fcfd22 CM |
2377 | if (del_all) |
2378 | range_end = (u64)-1; | |
2379 | else { | |
2380 | ret = find_dir_range(log, path, dirid, key_type, | |
2381 | &range_start, &range_end); | |
2382 | if (ret != 0) | |
2383 | break; | |
2384 | } | |
e02119d5 CM |
2385 | |
2386 | dir_key.offset = range_start; | |
d397712b | 2387 | while (1) { |
e02119d5 CM |
2388 | int nritems; |
2389 | ret = btrfs_search_slot(NULL, root, &dir_key, path, | |
2390 | 0, 0); | |
2391 | if (ret < 0) | |
2392 | goto out; | |
2393 | ||
2394 | nritems = btrfs_header_nritems(path->nodes[0]); | |
2395 | if (path->slots[0] >= nritems) { | |
2396 | ret = btrfs_next_leaf(root, path); | |
b98def7c | 2397 | if (ret == 1) |
e02119d5 | 2398 | break; |
b98def7c LB |
2399 | else if (ret < 0) |
2400 | goto out; | |
e02119d5 CM |
2401 | } |
2402 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, | |
2403 | path->slots[0]); | |
2404 | if (found_key.objectid != dirid || | |
2405 | found_key.type != dir_key.type) | |
2406 | goto next_type; | |
2407 | ||
2408 | if (found_key.offset > range_end) | |
2409 | break; | |
2410 | ||
2411 | ret = check_item_in_log(trans, root, log, path, | |
12fcfd22 CM |
2412 | log_path, dir, |
2413 | &found_key); | |
3650860b JB |
2414 | if (ret) |
2415 | goto out; | |
e02119d5 CM |
2416 | if (found_key.offset == (u64)-1) |
2417 | break; | |
2418 | dir_key.offset = found_key.offset + 1; | |
2419 | } | |
b3b4aa74 | 2420 | btrfs_release_path(path); |
e02119d5 CM |
2421 | if (range_end == (u64)-1) |
2422 | break; | |
2423 | range_start = range_end + 1; | |
2424 | } | |
2425 | ||
2426 | next_type: | |
2427 | ret = 0; | |
2428 | if (key_type == BTRFS_DIR_LOG_ITEM_KEY) { | |
2429 | key_type = BTRFS_DIR_LOG_INDEX_KEY; | |
2430 | dir_key.type = BTRFS_DIR_INDEX_KEY; | |
b3b4aa74 | 2431 | btrfs_release_path(path); |
e02119d5 CM |
2432 | goto again; |
2433 | } | |
2434 | out: | |
b3b4aa74 | 2435 | btrfs_release_path(path); |
e02119d5 CM |
2436 | btrfs_free_path(log_path); |
2437 | iput(dir); | |
2438 | return ret; | |
2439 | } | |
2440 | ||
2441 | /* | |
2442 | * the process_func used to replay items from the log tree. This | |
2443 | * gets called in two different stages. The first stage just looks | |
2444 | * for inodes and makes sure they are all copied into the subvolume. | |
2445 | * | |
2446 | * The second stage copies all the other item types from the log into | |
2447 | * the subvolume. The two stage approach is slower, but gets rid of | |
2448 | * lots of complexity around inodes referencing other inodes that exist | |
2449 | * only in the log (references come from either directory items or inode | |
2450 | * back refs). | |
2451 | */ | |
2452 | static int replay_one_buffer(struct btrfs_root *log, struct extent_buffer *eb, | |
581c1760 | 2453 | struct walk_control *wc, u64 gen, int level) |
e02119d5 CM |
2454 | { |
2455 | int nritems; | |
2456 | struct btrfs_path *path; | |
2457 | struct btrfs_root *root = wc->replay_dest; | |
2458 | struct btrfs_key key; | |
e02119d5 CM |
2459 | int i; |
2460 | int ret; | |
2461 | ||
581c1760 | 2462 | ret = btrfs_read_buffer(eb, gen, level, NULL); |
018642a1 TI |
2463 | if (ret) |
2464 | return ret; | |
e02119d5 CM |
2465 | |
2466 | level = btrfs_header_level(eb); | |
2467 | ||
2468 | if (level != 0) | |
2469 | return 0; | |
2470 | ||
2471 | path = btrfs_alloc_path(); | |
1e5063d0 MF |
2472 | if (!path) |
2473 | return -ENOMEM; | |
e02119d5 CM |
2474 | |
2475 | nritems = btrfs_header_nritems(eb); | |
2476 | for (i = 0; i < nritems; i++) { | |
2477 | btrfs_item_key_to_cpu(eb, &key, i); | |
e02119d5 CM |
2478 | |
2479 | /* inode keys are done during the first stage */ | |
2480 | if (key.type == BTRFS_INODE_ITEM_KEY && | |
2481 | wc->stage == LOG_WALK_REPLAY_INODES) { | |
e02119d5 CM |
2482 | struct btrfs_inode_item *inode_item; |
2483 | u32 mode; | |
2484 | ||
2485 | inode_item = btrfs_item_ptr(eb, i, | |
2486 | struct btrfs_inode_item); | |
4f764e51 FM |
2487 | ret = replay_xattr_deletes(wc->trans, root, log, |
2488 | path, key.objectid); | |
2489 | if (ret) | |
2490 | break; | |
e02119d5 CM |
2491 | mode = btrfs_inode_mode(eb, inode_item); |
2492 | if (S_ISDIR(mode)) { | |
2493 | ret = replay_dir_deletes(wc->trans, | |
12fcfd22 | 2494 | root, log, path, key.objectid, 0); |
b50c6e25 JB |
2495 | if (ret) |
2496 | break; | |
e02119d5 CM |
2497 | } |
2498 | ret = overwrite_item(wc->trans, root, path, | |
2499 | eb, i, &key); | |
b50c6e25 JB |
2500 | if (ret) |
2501 | break; | |
e02119d5 | 2502 | |
471d557a FM |
2503 | /* |
2504 | * Before replaying extents, truncate the inode to its | |
2505 | * size. We need to do it now and not after log replay | |
2506 | * because before an fsync we can have prealloc extents | |
2507 | * added beyond the inode's i_size. If we did it after, | |
2508 | * through orphan cleanup for example, we would drop | |
2509 | * those prealloc extents just after replaying them. | |
e02119d5 CM |
2510 | */ |
2511 | if (S_ISREG(mode)) { | |
471d557a FM |
2512 | struct inode *inode; |
2513 | u64 from; | |
2514 | ||
2515 | inode = read_one_inode(root, key.objectid); | |
2516 | if (!inode) { | |
2517 | ret = -EIO; | |
2518 | break; | |
2519 | } | |
2520 | from = ALIGN(i_size_read(inode), | |
2521 | root->fs_info->sectorsize); | |
2522 | ret = btrfs_drop_extents(wc->trans, root, inode, | |
2523 | from, (u64)-1, 1); | |
2524 | /* | |
2525 | * If the nlink count is zero here, the iput | |
2526 | * will free the inode. We bump it to make | |
2527 | * sure it doesn't get freed until the link | |
2528 | * count fixup is done. | |
2529 | */ | |
2530 | if (!ret) { | |
2531 | if (inode->i_nlink == 0) | |
2532 | inc_nlink(inode); | |
2533 | /* Update link count and nbytes. */ | |
2534 | ret = btrfs_update_inode(wc->trans, | |
2535 | root, inode); | |
2536 | } | |
2537 | iput(inode); | |
b50c6e25 JB |
2538 | if (ret) |
2539 | break; | |
e02119d5 | 2540 | } |
c71bf099 | 2541 | |
e02119d5 CM |
2542 | ret = link_to_fixup_dir(wc->trans, root, |
2543 | path, key.objectid); | |
b50c6e25 JB |
2544 | if (ret) |
2545 | break; | |
e02119d5 | 2546 | } |
dd8e7217 JB |
2547 | |
2548 | if (key.type == BTRFS_DIR_INDEX_KEY && | |
2549 | wc->stage == LOG_WALK_REPLAY_DIR_INDEX) { | |
2550 | ret = replay_one_dir_item(wc->trans, root, path, | |
2551 | eb, i, &key); | |
2552 | if (ret) | |
2553 | break; | |
2554 | } | |
2555 | ||
e02119d5 CM |
2556 | if (wc->stage < LOG_WALK_REPLAY_ALL) |
2557 | continue; | |
2558 | ||
2559 | /* these keys are simply copied */ | |
2560 | if (key.type == BTRFS_XATTR_ITEM_KEY) { | |
2561 | ret = overwrite_item(wc->trans, root, path, | |
2562 | eb, i, &key); | |
b50c6e25 JB |
2563 | if (ret) |
2564 | break; | |
2da1c669 LB |
2565 | } else if (key.type == BTRFS_INODE_REF_KEY || |
2566 | key.type == BTRFS_INODE_EXTREF_KEY) { | |
f186373f MF |
2567 | ret = add_inode_ref(wc->trans, root, log, path, |
2568 | eb, i, &key); | |
b50c6e25 JB |
2569 | if (ret && ret != -ENOENT) |
2570 | break; | |
2571 | ret = 0; | |
e02119d5 CM |
2572 | } else if (key.type == BTRFS_EXTENT_DATA_KEY) { |
2573 | ret = replay_one_extent(wc->trans, root, path, | |
2574 | eb, i, &key); | |
b50c6e25 JB |
2575 | if (ret) |
2576 | break; | |
dd8e7217 | 2577 | } else if (key.type == BTRFS_DIR_ITEM_KEY) { |
e02119d5 CM |
2578 | ret = replay_one_dir_item(wc->trans, root, path, |
2579 | eb, i, &key); | |
b50c6e25 JB |
2580 | if (ret) |
2581 | break; | |
e02119d5 CM |
2582 | } |
2583 | } | |
2584 | btrfs_free_path(path); | |
b50c6e25 | 2585 | return ret; |
e02119d5 CM |
2586 | } |
2587 | ||
d397712b | 2588 | static noinline int walk_down_log_tree(struct btrfs_trans_handle *trans, |
e02119d5 CM |
2589 | struct btrfs_root *root, |
2590 | struct btrfs_path *path, int *level, | |
2591 | struct walk_control *wc) | |
2592 | { | |
0b246afa | 2593 | struct btrfs_fs_info *fs_info = root->fs_info; |
e02119d5 | 2594 | u64 root_owner; |
e02119d5 CM |
2595 | u64 bytenr; |
2596 | u64 ptr_gen; | |
2597 | struct extent_buffer *next; | |
2598 | struct extent_buffer *cur; | |
2599 | struct extent_buffer *parent; | |
2600 | u32 blocksize; | |
2601 | int ret = 0; | |
2602 | ||
2603 | WARN_ON(*level < 0); | |
2604 | WARN_ON(*level >= BTRFS_MAX_LEVEL); | |
2605 | ||
d397712b | 2606 | while (*level > 0) { |
581c1760 QW |
2607 | struct btrfs_key first_key; |
2608 | ||
e02119d5 CM |
2609 | WARN_ON(*level < 0); |
2610 | WARN_ON(*level >= BTRFS_MAX_LEVEL); | |
2611 | cur = path->nodes[*level]; | |
2612 | ||
fae7f21c | 2613 | WARN_ON(btrfs_header_level(cur) != *level); |
e02119d5 CM |
2614 | |
2615 | if (path->slots[*level] >= | |
2616 | btrfs_header_nritems(cur)) | |
2617 | break; | |
2618 | ||
2619 | bytenr = btrfs_node_blockptr(cur, path->slots[*level]); | |
2620 | ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]); | |
581c1760 | 2621 | btrfs_node_key_to_cpu(cur, &first_key, path->slots[*level]); |
0b246afa | 2622 | blocksize = fs_info->nodesize; |
e02119d5 CM |
2623 | |
2624 | parent = path->nodes[*level]; | |
2625 | root_owner = btrfs_header_owner(parent); | |
e02119d5 | 2626 | |
2ff7e61e | 2627 | next = btrfs_find_create_tree_block(fs_info, bytenr); |
c871b0f2 LB |
2628 | if (IS_ERR(next)) |
2629 | return PTR_ERR(next); | |
e02119d5 | 2630 | |
e02119d5 | 2631 | if (*level == 1) { |
581c1760 QW |
2632 | ret = wc->process_func(root, next, wc, ptr_gen, |
2633 | *level - 1); | |
b50c6e25 JB |
2634 | if (ret) { |
2635 | free_extent_buffer(next); | |
1e5063d0 | 2636 | return ret; |
b50c6e25 | 2637 | } |
4a500fd1 | 2638 | |
e02119d5 CM |
2639 | path->slots[*level]++; |
2640 | if (wc->free) { | |
581c1760 QW |
2641 | ret = btrfs_read_buffer(next, ptr_gen, |
2642 | *level - 1, &first_key); | |
018642a1 TI |
2643 | if (ret) { |
2644 | free_extent_buffer(next); | |
2645 | return ret; | |
2646 | } | |
e02119d5 | 2647 | |
681ae509 JB |
2648 | if (trans) { |
2649 | btrfs_tree_lock(next); | |
2650 | btrfs_set_lock_blocking(next); | |
7c302b49 | 2651 | clean_tree_block(fs_info, next); |
681ae509 JB |
2652 | btrfs_wait_tree_block_writeback(next); |
2653 | btrfs_tree_unlock(next); | |
1846430c LB |
2654 | } else { |
2655 | if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &next->bflags)) | |
2656 | clear_extent_buffer_dirty(next); | |
681ae509 | 2657 | } |
e02119d5 | 2658 | |
e02119d5 CM |
2659 | WARN_ON(root_owner != |
2660 | BTRFS_TREE_LOG_OBJECTID); | |
2ff7e61e JM |
2661 | ret = btrfs_free_and_pin_reserved_extent( |
2662 | fs_info, bytenr, | |
2663 | blocksize); | |
3650860b JB |
2664 | if (ret) { |
2665 | free_extent_buffer(next); | |
2666 | return ret; | |
2667 | } | |
e02119d5 CM |
2668 | } |
2669 | free_extent_buffer(next); | |
2670 | continue; | |
2671 | } | |
581c1760 | 2672 | ret = btrfs_read_buffer(next, ptr_gen, *level - 1, &first_key); |
018642a1 TI |
2673 | if (ret) { |
2674 | free_extent_buffer(next); | |
2675 | return ret; | |
2676 | } | |
e02119d5 CM |
2677 | |
2678 | WARN_ON(*level <= 0); | |
2679 | if (path->nodes[*level-1]) | |
2680 | free_extent_buffer(path->nodes[*level-1]); | |
2681 | path->nodes[*level-1] = next; | |
2682 | *level = btrfs_header_level(next); | |
2683 | path->slots[*level] = 0; | |
2684 | cond_resched(); | |
2685 | } | |
2686 | WARN_ON(*level < 0); | |
2687 | WARN_ON(*level >= BTRFS_MAX_LEVEL); | |
2688 | ||
4a500fd1 | 2689 | path->slots[*level] = btrfs_header_nritems(path->nodes[*level]); |
e02119d5 CM |
2690 | |
2691 | cond_resched(); | |
2692 | return 0; | |
2693 | } | |
2694 | ||
d397712b | 2695 | static noinline int walk_up_log_tree(struct btrfs_trans_handle *trans, |
e02119d5 CM |
2696 | struct btrfs_root *root, |
2697 | struct btrfs_path *path, int *level, | |
2698 | struct walk_control *wc) | |
2699 | { | |
0b246afa | 2700 | struct btrfs_fs_info *fs_info = root->fs_info; |
e02119d5 | 2701 | u64 root_owner; |
e02119d5 CM |
2702 | int i; |
2703 | int slot; | |
2704 | int ret; | |
2705 | ||
d397712b | 2706 | for (i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) { |
e02119d5 | 2707 | slot = path->slots[i]; |
4a500fd1 | 2708 | if (slot + 1 < btrfs_header_nritems(path->nodes[i])) { |
e02119d5 CM |
2709 | path->slots[i]++; |
2710 | *level = i; | |
2711 | WARN_ON(*level == 0); | |
2712 | return 0; | |
2713 | } else { | |
31840ae1 ZY |
2714 | struct extent_buffer *parent; |
2715 | if (path->nodes[*level] == root->node) | |
2716 | parent = path->nodes[*level]; | |
2717 | else | |
2718 | parent = path->nodes[*level + 1]; | |
2719 | ||
2720 | root_owner = btrfs_header_owner(parent); | |
1e5063d0 | 2721 | ret = wc->process_func(root, path->nodes[*level], wc, |
581c1760 QW |
2722 | btrfs_header_generation(path->nodes[*level]), |
2723 | *level); | |
1e5063d0 MF |
2724 | if (ret) |
2725 | return ret; | |
2726 | ||
e02119d5 CM |
2727 | if (wc->free) { |
2728 | struct extent_buffer *next; | |
2729 | ||
2730 | next = path->nodes[*level]; | |
2731 | ||
681ae509 JB |
2732 | if (trans) { |
2733 | btrfs_tree_lock(next); | |
2734 | btrfs_set_lock_blocking(next); | |
7c302b49 | 2735 | clean_tree_block(fs_info, next); |
681ae509 JB |
2736 | btrfs_wait_tree_block_writeback(next); |
2737 | btrfs_tree_unlock(next); | |
1846430c LB |
2738 | } else { |
2739 | if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &next->bflags)) | |
2740 | clear_extent_buffer_dirty(next); | |
681ae509 | 2741 | } |
e02119d5 | 2742 | |
e02119d5 | 2743 | WARN_ON(root_owner != BTRFS_TREE_LOG_OBJECTID); |
2ff7e61e JM |
2744 | ret = btrfs_free_and_pin_reserved_extent( |
2745 | fs_info, | |
e02119d5 | 2746 | path->nodes[*level]->start, |
d00aff00 | 2747 | path->nodes[*level]->len); |
3650860b JB |
2748 | if (ret) |
2749 | return ret; | |
e02119d5 CM |
2750 | } |
2751 | free_extent_buffer(path->nodes[*level]); | |
2752 | path->nodes[*level] = NULL; | |
2753 | *level = i + 1; | |
2754 | } | |
2755 | } | |
2756 | return 1; | |
2757 | } | |
2758 | ||
2759 | /* | |
2760 | * drop the reference count on the tree rooted at 'snap'. This traverses | |
2761 | * the tree freeing any blocks that have a ref count of zero after being | |
2762 | * decremented. | |
2763 | */ | |
2764 | static int walk_log_tree(struct btrfs_trans_handle *trans, | |
2765 | struct btrfs_root *log, struct walk_control *wc) | |
2766 | { | |
2ff7e61e | 2767 | struct btrfs_fs_info *fs_info = log->fs_info; |
e02119d5 CM |
2768 | int ret = 0; |
2769 | int wret; | |
2770 | int level; | |
2771 | struct btrfs_path *path; | |
e02119d5 CM |
2772 | int orig_level; |
2773 | ||
2774 | path = btrfs_alloc_path(); | |
db5b493a TI |
2775 | if (!path) |
2776 | return -ENOMEM; | |
e02119d5 CM |
2777 | |
2778 | level = btrfs_header_level(log->node); | |
2779 | orig_level = level; | |
2780 | path->nodes[level] = log->node; | |
2781 | extent_buffer_get(log->node); | |
2782 | path->slots[level] = 0; | |
2783 | ||
d397712b | 2784 | while (1) { |
e02119d5 CM |
2785 | wret = walk_down_log_tree(trans, log, path, &level, wc); |
2786 | if (wret > 0) | |
2787 | break; | |
79787eaa | 2788 | if (wret < 0) { |
e02119d5 | 2789 | ret = wret; |
79787eaa JM |
2790 | goto out; |
2791 | } | |
e02119d5 CM |
2792 | |
2793 | wret = walk_up_log_tree(trans, log, path, &level, wc); | |
2794 | if (wret > 0) | |
2795 | break; | |
79787eaa | 2796 | if (wret < 0) { |
e02119d5 | 2797 | ret = wret; |
79787eaa JM |
2798 | goto out; |
2799 | } | |
e02119d5 CM |
2800 | } |
2801 | ||
2802 | /* was the root node processed? if not, catch it here */ | |
2803 | if (path->nodes[orig_level]) { | |
79787eaa | 2804 | ret = wc->process_func(log, path->nodes[orig_level], wc, |
581c1760 QW |
2805 | btrfs_header_generation(path->nodes[orig_level]), |
2806 | orig_level); | |
79787eaa JM |
2807 | if (ret) |
2808 | goto out; | |
e02119d5 CM |
2809 | if (wc->free) { |
2810 | struct extent_buffer *next; | |
2811 | ||
2812 | next = path->nodes[orig_level]; | |
2813 | ||
681ae509 JB |
2814 | if (trans) { |
2815 | btrfs_tree_lock(next); | |
2816 | btrfs_set_lock_blocking(next); | |
7c302b49 | 2817 | clean_tree_block(fs_info, next); |
681ae509 JB |
2818 | btrfs_wait_tree_block_writeback(next); |
2819 | btrfs_tree_unlock(next); | |
1846430c LB |
2820 | } else { |
2821 | if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &next->bflags)) | |
2822 | clear_extent_buffer_dirty(next); | |
681ae509 | 2823 | } |
e02119d5 | 2824 | |
e02119d5 CM |
2825 | WARN_ON(log->root_key.objectid != |
2826 | BTRFS_TREE_LOG_OBJECTID); | |
2ff7e61e JM |
2827 | ret = btrfs_free_and_pin_reserved_extent(fs_info, |
2828 | next->start, next->len); | |
3650860b JB |
2829 | if (ret) |
2830 | goto out; | |
e02119d5 CM |
2831 | } |
2832 | } | |
2833 | ||
79787eaa | 2834 | out: |
e02119d5 | 2835 | btrfs_free_path(path); |
e02119d5 CM |
2836 | return ret; |
2837 | } | |
2838 | ||
7237f183 YZ |
2839 | /* |
2840 | * helper function to update the item for a given subvolumes log root | |
2841 | * in the tree of log roots | |
2842 | */ | |
2843 | static int update_log_root(struct btrfs_trans_handle *trans, | |
2844 | struct btrfs_root *log) | |
2845 | { | |
0b246afa | 2846 | struct btrfs_fs_info *fs_info = log->fs_info; |
7237f183 YZ |
2847 | int ret; |
2848 | ||
2849 | if (log->log_transid == 1) { | |
2850 | /* insert root item on the first sync */ | |
0b246afa | 2851 | ret = btrfs_insert_root(trans, fs_info->log_root_tree, |
7237f183 YZ |
2852 | &log->root_key, &log->root_item); |
2853 | } else { | |
0b246afa | 2854 | ret = btrfs_update_root(trans, fs_info->log_root_tree, |
7237f183 YZ |
2855 | &log->root_key, &log->root_item); |
2856 | } | |
2857 | return ret; | |
2858 | } | |
2859 | ||
60d53eb3 | 2860 | static void wait_log_commit(struct btrfs_root *root, int transid) |
e02119d5 CM |
2861 | { |
2862 | DEFINE_WAIT(wait); | |
7237f183 | 2863 | int index = transid % 2; |
e02119d5 | 2864 | |
7237f183 YZ |
2865 | /* |
2866 | * we only allow two pending log transactions at a time, | |
2867 | * so we know that if ours is more than 2 older than the | |
2868 | * current transaction, we're done | |
2869 | */ | |
49e83f57 | 2870 | for (;;) { |
7237f183 YZ |
2871 | prepare_to_wait(&root->log_commit_wait[index], |
2872 | &wait, TASK_UNINTERRUPTIBLE); | |
12fcfd22 | 2873 | |
49e83f57 LB |
2874 | if (!(root->log_transid_committed < transid && |
2875 | atomic_read(&root->log_commit[index]))) | |
2876 | break; | |
12fcfd22 | 2877 | |
49e83f57 LB |
2878 | mutex_unlock(&root->log_mutex); |
2879 | schedule(); | |
7237f183 | 2880 | mutex_lock(&root->log_mutex); |
49e83f57 LB |
2881 | } |
2882 | finish_wait(&root->log_commit_wait[index], &wait); | |
7237f183 YZ |
2883 | } |
2884 | ||
60d53eb3 | 2885 | static void wait_for_writer(struct btrfs_root *root) |
7237f183 YZ |
2886 | { |
2887 | DEFINE_WAIT(wait); | |
8b050d35 | 2888 | |
49e83f57 LB |
2889 | for (;;) { |
2890 | prepare_to_wait(&root->log_writer_wait, &wait, | |
2891 | TASK_UNINTERRUPTIBLE); | |
2892 | if (!atomic_read(&root->log_writers)) | |
2893 | break; | |
2894 | ||
7237f183 | 2895 | mutex_unlock(&root->log_mutex); |
49e83f57 | 2896 | schedule(); |
575849ec | 2897 | mutex_lock(&root->log_mutex); |
7237f183 | 2898 | } |
49e83f57 | 2899 | finish_wait(&root->log_writer_wait, &wait); |
e02119d5 CM |
2900 | } |
2901 | ||
8b050d35 MX |
2902 | static inline void btrfs_remove_log_ctx(struct btrfs_root *root, |
2903 | struct btrfs_log_ctx *ctx) | |
2904 | { | |
2905 | if (!ctx) | |
2906 | return; | |
2907 | ||
2908 | mutex_lock(&root->log_mutex); | |
2909 | list_del_init(&ctx->list); | |
2910 | mutex_unlock(&root->log_mutex); | |
2911 | } | |
2912 | ||
2913 | /* | |
2914 | * Invoked in log mutex context, or be sure there is no other task which | |
2915 | * can access the list. | |
2916 | */ | |
2917 | static inline void btrfs_remove_all_log_ctxs(struct btrfs_root *root, | |
2918 | int index, int error) | |
2919 | { | |
2920 | struct btrfs_log_ctx *ctx; | |
570dd450 | 2921 | struct btrfs_log_ctx *safe; |
8b050d35 | 2922 | |
570dd450 CM |
2923 | list_for_each_entry_safe(ctx, safe, &root->log_ctxs[index], list) { |
2924 | list_del_init(&ctx->list); | |
8b050d35 | 2925 | ctx->log_ret = error; |
570dd450 | 2926 | } |
8b050d35 MX |
2927 | |
2928 | INIT_LIST_HEAD(&root->log_ctxs[index]); | |
2929 | } | |
2930 | ||
e02119d5 CM |
2931 | /* |
2932 | * btrfs_sync_log does sends a given tree log down to the disk and | |
2933 | * updates the super blocks to record it. When this call is done, | |
12fcfd22 CM |
2934 | * you know that any inodes previously logged are safely on disk only |
2935 | * if it returns 0. | |
2936 | * | |
2937 | * Any other return value means you need to call btrfs_commit_transaction. | |
2938 | * Some of the edge cases for fsyncing directories that have had unlinks | |
2939 | * or renames done in the past mean that sometimes the only safe | |
2940 | * fsync is to commit the whole FS. When btrfs_sync_log returns -EAGAIN, | |
2941 | * that has happened. | |
e02119d5 CM |
2942 | */ |
2943 | int btrfs_sync_log(struct btrfs_trans_handle *trans, | |
8b050d35 | 2944 | struct btrfs_root *root, struct btrfs_log_ctx *ctx) |
e02119d5 | 2945 | { |
7237f183 YZ |
2946 | int index1; |
2947 | int index2; | |
8cef4e16 | 2948 | int mark; |
e02119d5 | 2949 | int ret; |
0b246afa | 2950 | struct btrfs_fs_info *fs_info = root->fs_info; |
e02119d5 | 2951 | struct btrfs_root *log = root->log_root; |
0b246afa | 2952 | struct btrfs_root *log_root_tree = fs_info->log_root_tree; |
bb14a59b | 2953 | int log_transid = 0; |
8b050d35 | 2954 | struct btrfs_log_ctx root_log_ctx; |
c6adc9cc | 2955 | struct blk_plug plug; |
e02119d5 | 2956 | |
7237f183 | 2957 | mutex_lock(&root->log_mutex); |
d1433deb MX |
2958 | log_transid = ctx->log_transid; |
2959 | if (root->log_transid_committed >= log_transid) { | |
2960 | mutex_unlock(&root->log_mutex); | |
2961 | return ctx->log_ret; | |
2962 | } | |
2963 | ||
2964 | index1 = log_transid % 2; | |
7237f183 | 2965 | if (atomic_read(&root->log_commit[index1])) { |
60d53eb3 | 2966 | wait_log_commit(root, log_transid); |
7237f183 | 2967 | mutex_unlock(&root->log_mutex); |
8b050d35 | 2968 | return ctx->log_ret; |
e02119d5 | 2969 | } |
d1433deb | 2970 | ASSERT(log_transid == root->log_transid); |
7237f183 YZ |
2971 | atomic_set(&root->log_commit[index1], 1); |
2972 | ||
2973 | /* wait for previous tree log sync to complete */ | |
2974 | if (atomic_read(&root->log_commit[(index1 + 1) % 2])) | |
60d53eb3 | 2975 | wait_log_commit(root, log_transid - 1); |
48cab2e0 | 2976 | |
86df7eb9 | 2977 | while (1) { |
2ecb7923 | 2978 | int batch = atomic_read(&root->log_batch); |
cd354ad6 | 2979 | /* when we're on an ssd, just kick the log commit out */ |
0b246afa | 2980 | if (!btrfs_test_opt(fs_info, SSD) && |
27cdeb70 | 2981 | test_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state)) { |
86df7eb9 YZ |
2982 | mutex_unlock(&root->log_mutex); |
2983 | schedule_timeout_uninterruptible(1); | |
2984 | mutex_lock(&root->log_mutex); | |
2985 | } | |
60d53eb3 | 2986 | wait_for_writer(root); |
2ecb7923 | 2987 | if (batch == atomic_read(&root->log_batch)) |
e02119d5 CM |
2988 | break; |
2989 | } | |
e02119d5 | 2990 | |
12fcfd22 | 2991 | /* bail out if we need to do a full commit */ |
0b246afa | 2992 | if (btrfs_need_log_full_commit(fs_info, trans)) { |
12fcfd22 CM |
2993 | ret = -EAGAIN; |
2994 | mutex_unlock(&root->log_mutex); | |
2995 | goto out; | |
2996 | } | |
2997 | ||
8cef4e16 YZ |
2998 | if (log_transid % 2 == 0) |
2999 | mark = EXTENT_DIRTY; | |
3000 | else | |
3001 | mark = EXTENT_NEW; | |
3002 | ||
690587d1 CM |
3003 | /* we start IO on all the marked extents here, but we don't actually |
3004 | * wait for them until later. | |
3005 | */ | |
c6adc9cc | 3006 | blk_start_plug(&plug); |
2ff7e61e | 3007 | ret = btrfs_write_marked_extents(fs_info, &log->dirty_log_pages, mark); |
79787eaa | 3008 | if (ret) { |
c6adc9cc | 3009 | blk_finish_plug(&plug); |
66642832 | 3010 | btrfs_abort_transaction(trans, ret); |
0b246afa | 3011 | btrfs_set_log_full_commit(fs_info, trans); |
79787eaa JM |
3012 | mutex_unlock(&root->log_mutex); |
3013 | goto out; | |
3014 | } | |
7237f183 | 3015 | |
5d4f98a2 | 3016 | btrfs_set_root_node(&log->root_item, log->node); |
7237f183 | 3017 | |
7237f183 YZ |
3018 | root->log_transid++; |
3019 | log->log_transid = root->log_transid; | |
ff782e0a | 3020 | root->log_start_pid = 0; |
7237f183 | 3021 | /* |
8cef4e16 YZ |
3022 | * IO has been started, blocks of the log tree have WRITTEN flag set |
3023 | * in their headers. new modifications of the log will be written to | |
3024 | * new positions. so it's safe to allow log writers to go in. | |
7237f183 YZ |
3025 | */ |
3026 | mutex_unlock(&root->log_mutex); | |
3027 | ||
28a23593 | 3028 | btrfs_init_log_ctx(&root_log_ctx, NULL); |
d1433deb | 3029 | |
7237f183 | 3030 | mutex_lock(&log_root_tree->log_mutex); |
2ecb7923 | 3031 | atomic_inc(&log_root_tree->log_batch); |
7237f183 | 3032 | atomic_inc(&log_root_tree->log_writers); |
d1433deb MX |
3033 | |
3034 | index2 = log_root_tree->log_transid % 2; | |
3035 | list_add_tail(&root_log_ctx.list, &log_root_tree->log_ctxs[index2]); | |
3036 | root_log_ctx.log_transid = log_root_tree->log_transid; | |
3037 | ||
7237f183 YZ |
3038 | mutex_unlock(&log_root_tree->log_mutex); |
3039 | ||
3040 | ret = update_log_root(trans, log); | |
7237f183 YZ |
3041 | |
3042 | mutex_lock(&log_root_tree->log_mutex); | |
3043 | if (atomic_dec_and_test(&log_root_tree->log_writers)) { | |
093258e6 DS |
3044 | /* atomic_dec_and_test implies a barrier */ |
3045 | cond_wake_up_nomb(&log_root_tree->log_writer_wait); | |
7237f183 YZ |
3046 | } |
3047 | ||
4a500fd1 | 3048 | if (ret) { |
d1433deb MX |
3049 | if (!list_empty(&root_log_ctx.list)) |
3050 | list_del_init(&root_log_ctx.list); | |
3051 | ||
c6adc9cc | 3052 | blk_finish_plug(&plug); |
0b246afa | 3053 | btrfs_set_log_full_commit(fs_info, trans); |
995946dd | 3054 | |
79787eaa | 3055 | if (ret != -ENOSPC) { |
66642832 | 3056 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
3057 | mutex_unlock(&log_root_tree->log_mutex); |
3058 | goto out; | |
3059 | } | |
bf89d38f | 3060 | btrfs_wait_tree_log_extents(log, mark); |
4a500fd1 YZ |
3061 | mutex_unlock(&log_root_tree->log_mutex); |
3062 | ret = -EAGAIN; | |
3063 | goto out; | |
3064 | } | |
3065 | ||
d1433deb | 3066 | if (log_root_tree->log_transid_committed >= root_log_ctx.log_transid) { |
3da5ab56 | 3067 | blk_finish_plug(&plug); |
cbd60aa7 | 3068 | list_del_init(&root_log_ctx.list); |
d1433deb MX |
3069 | mutex_unlock(&log_root_tree->log_mutex); |
3070 | ret = root_log_ctx.log_ret; | |
3071 | goto out; | |
3072 | } | |
8b050d35 | 3073 | |
d1433deb | 3074 | index2 = root_log_ctx.log_transid % 2; |
7237f183 | 3075 | if (atomic_read(&log_root_tree->log_commit[index2])) { |
c6adc9cc | 3076 | blk_finish_plug(&plug); |
bf89d38f | 3077 | ret = btrfs_wait_tree_log_extents(log, mark); |
60d53eb3 | 3078 | wait_log_commit(log_root_tree, |
d1433deb | 3079 | root_log_ctx.log_transid); |
7237f183 | 3080 | mutex_unlock(&log_root_tree->log_mutex); |
5ab5e44a FM |
3081 | if (!ret) |
3082 | ret = root_log_ctx.log_ret; | |
7237f183 YZ |
3083 | goto out; |
3084 | } | |
d1433deb | 3085 | ASSERT(root_log_ctx.log_transid == log_root_tree->log_transid); |
7237f183 YZ |
3086 | atomic_set(&log_root_tree->log_commit[index2], 1); |
3087 | ||
12fcfd22 | 3088 | if (atomic_read(&log_root_tree->log_commit[(index2 + 1) % 2])) { |
60d53eb3 | 3089 | wait_log_commit(log_root_tree, |
d1433deb | 3090 | root_log_ctx.log_transid - 1); |
12fcfd22 CM |
3091 | } |
3092 | ||
60d53eb3 | 3093 | wait_for_writer(log_root_tree); |
7237f183 | 3094 | |
12fcfd22 CM |
3095 | /* |
3096 | * now that we've moved on to the tree of log tree roots, | |
3097 | * check the full commit flag again | |
3098 | */ | |
0b246afa | 3099 | if (btrfs_need_log_full_commit(fs_info, trans)) { |
c6adc9cc | 3100 | blk_finish_plug(&plug); |
bf89d38f | 3101 | btrfs_wait_tree_log_extents(log, mark); |
12fcfd22 CM |
3102 | mutex_unlock(&log_root_tree->log_mutex); |
3103 | ret = -EAGAIN; | |
3104 | goto out_wake_log_root; | |
3105 | } | |
7237f183 | 3106 | |
2ff7e61e | 3107 | ret = btrfs_write_marked_extents(fs_info, |
c6adc9cc MX |
3108 | &log_root_tree->dirty_log_pages, |
3109 | EXTENT_DIRTY | EXTENT_NEW); | |
3110 | blk_finish_plug(&plug); | |
79787eaa | 3111 | if (ret) { |
0b246afa | 3112 | btrfs_set_log_full_commit(fs_info, trans); |
66642832 | 3113 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
3114 | mutex_unlock(&log_root_tree->log_mutex); |
3115 | goto out_wake_log_root; | |
3116 | } | |
bf89d38f | 3117 | ret = btrfs_wait_tree_log_extents(log, mark); |
5ab5e44a | 3118 | if (!ret) |
bf89d38f JM |
3119 | ret = btrfs_wait_tree_log_extents(log_root_tree, |
3120 | EXTENT_NEW | EXTENT_DIRTY); | |
5ab5e44a | 3121 | if (ret) { |
0b246afa | 3122 | btrfs_set_log_full_commit(fs_info, trans); |
5ab5e44a FM |
3123 | mutex_unlock(&log_root_tree->log_mutex); |
3124 | goto out_wake_log_root; | |
3125 | } | |
e02119d5 | 3126 | |
0b246afa JM |
3127 | btrfs_set_super_log_root(fs_info->super_for_commit, |
3128 | log_root_tree->node->start); | |
3129 | btrfs_set_super_log_root_level(fs_info->super_for_commit, | |
3130 | btrfs_header_level(log_root_tree->node)); | |
e02119d5 | 3131 | |
7237f183 | 3132 | log_root_tree->log_transid++; |
7237f183 YZ |
3133 | mutex_unlock(&log_root_tree->log_mutex); |
3134 | ||
3135 | /* | |
3136 | * nobody else is going to jump in and write the the ctree | |
3137 | * super here because the log_commit atomic below is protecting | |
3138 | * us. We must be called with a transaction handle pinning | |
3139 | * the running transaction open, so a full commit can't hop | |
3140 | * in and cause problems either. | |
3141 | */ | |
eece6a9c | 3142 | ret = write_all_supers(fs_info, 1); |
5af3e8cc | 3143 | if (ret) { |
0b246afa | 3144 | btrfs_set_log_full_commit(fs_info, trans); |
66642832 | 3145 | btrfs_abort_transaction(trans, ret); |
5af3e8cc SB |
3146 | goto out_wake_log_root; |
3147 | } | |
7237f183 | 3148 | |
257c62e1 CM |
3149 | mutex_lock(&root->log_mutex); |
3150 | if (root->last_log_commit < log_transid) | |
3151 | root->last_log_commit = log_transid; | |
3152 | mutex_unlock(&root->log_mutex); | |
3153 | ||
12fcfd22 | 3154 | out_wake_log_root: |
570dd450 | 3155 | mutex_lock(&log_root_tree->log_mutex); |
8b050d35 MX |
3156 | btrfs_remove_all_log_ctxs(log_root_tree, index2, ret); |
3157 | ||
d1433deb | 3158 | log_root_tree->log_transid_committed++; |
7237f183 | 3159 | atomic_set(&log_root_tree->log_commit[index2], 0); |
d1433deb MX |
3160 | mutex_unlock(&log_root_tree->log_mutex); |
3161 | ||
33a9eca7 | 3162 | /* |
093258e6 DS |
3163 | * The barrier before waitqueue_active (in cond_wake_up) is needed so |
3164 | * all the updates above are seen by the woken threads. It might not be | |
3165 | * necessary, but proving that seems to be hard. | |
33a9eca7 | 3166 | */ |
093258e6 | 3167 | cond_wake_up(&log_root_tree->log_commit_wait[index2]); |
e02119d5 | 3168 | out: |
d1433deb | 3169 | mutex_lock(&root->log_mutex); |
570dd450 | 3170 | btrfs_remove_all_log_ctxs(root, index1, ret); |
d1433deb | 3171 | root->log_transid_committed++; |
7237f183 | 3172 | atomic_set(&root->log_commit[index1], 0); |
d1433deb | 3173 | mutex_unlock(&root->log_mutex); |
8b050d35 | 3174 | |
33a9eca7 | 3175 | /* |
093258e6 DS |
3176 | * The barrier before waitqueue_active (in cond_wake_up) is needed so |
3177 | * all the updates above are seen by the woken threads. It might not be | |
3178 | * necessary, but proving that seems to be hard. | |
33a9eca7 | 3179 | */ |
093258e6 | 3180 | cond_wake_up(&root->log_commit_wait[index1]); |
b31eabd8 | 3181 | return ret; |
e02119d5 CM |
3182 | } |
3183 | ||
4a500fd1 YZ |
3184 | static void free_log_tree(struct btrfs_trans_handle *trans, |
3185 | struct btrfs_root *log) | |
e02119d5 CM |
3186 | { |
3187 | int ret; | |
d0c803c4 CM |
3188 | u64 start; |
3189 | u64 end; | |
e02119d5 CM |
3190 | struct walk_control wc = { |
3191 | .free = 1, | |
3192 | .process_func = process_one_buffer | |
3193 | }; | |
3194 | ||
681ae509 JB |
3195 | ret = walk_log_tree(trans, log, &wc); |
3196 | /* I don't think this can happen but just in case */ | |
3197 | if (ret) | |
66642832 | 3198 | btrfs_abort_transaction(trans, ret); |
e02119d5 | 3199 | |
d397712b | 3200 | while (1) { |
d0c803c4 | 3201 | ret = find_first_extent_bit(&log->dirty_log_pages, |
55237a5f LB |
3202 | 0, &start, &end, |
3203 | EXTENT_DIRTY | EXTENT_NEW | EXTENT_NEED_WAIT, | |
e6138876 | 3204 | NULL); |
d0c803c4 CM |
3205 | if (ret) |
3206 | break; | |
3207 | ||
8cef4e16 | 3208 | clear_extent_bits(&log->dirty_log_pages, start, end, |
55237a5f | 3209 | EXTENT_DIRTY | EXTENT_NEW | EXTENT_NEED_WAIT); |
d0c803c4 CM |
3210 | } |
3211 | ||
7237f183 YZ |
3212 | free_extent_buffer(log->node); |
3213 | kfree(log); | |
4a500fd1 YZ |
3214 | } |
3215 | ||
3216 | /* | |
3217 | * free all the extents used by the tree log. This should be called | |
3218 | * at commit time of the full transaction | |
3219 | */ | |
3220 | int btrfs_free_log(struct btrfs_trans_handle *trans, struct btrfs_root *root) | |
3221 | { | |
3222 | if (root->log_root) { | |
3223 | free_log_tree(trans, root->log_root); | |
3224 | root->log_root = NULL; | |
3225 | } | |
3226 | return 0; | |
3227 | } | |
3228 | ||
3229 | int btrfs_free_log_root_tree(struct btrfs_trans_handle *trans, | |
3230 | struct btrfs_fs_info *fs_info) | |
3231 | { | |
3232 | if (fs_info->log_root_tree) { | |
3233 | free_log_tree(trans, fs_info->log_root_tree); | |
3234 | fs_info->log_root_tree = NULL; | |
3235 | } | |
e02119d5 CM |
3236 | return 0; |
3237 | } | |
3238 | ||
e02119d5 CM |
3239 | /* |
3240 | * If both a file and directory are logged, and unlinks or renames are | |
3241 | * mixed in, we have a few interesting corners: | |
3242 | * | |
3243 | * create file X in dir Y | |
3244 | * link file X to X.link in dir Y | |
3245 | * fsync file X | |
3246 | * unlink file X but leave X.link | |
3247 | * fsync dir Y | |
3248 | * | |
3249 | * After a crash we would expect only X.link to exist. But file X | |
3250 | * didn't get fsync'd again so the log has back refs for X and X.link. | |
3251 | * | |
3252 | * We solve this by removing directory entries and inode backrefs from the | |
3253 | * log when a file that was logged in the current transaction is | |
3254 | * unlinked. Any later fsync will include the updated log entries, and | |
3255 | * we'll be able to reconstruct the proper directory items from backrefs. | |
3256 | * | |
3257 | * This optimizations allows us to avoid relogging the entire inode | |
3258 | * or the entire directory. | |
3259 | */ | |
3260 | int btrfs_del_dir_entries_in_log(struct btrfs_trans_handle *trans, | |
3261 | struct btrfs_root *root, | |
3262 | const char *name, int name_len, | |
49f34d1f | 3263 | struct btrfs_inode *dir, u64 index) |
e02119d5 CM |
3264 | { |
3265 | struct btrfs_root *log; | |
3266 | struct btrfs_dir_item *di; | |
3267 | struct btrfs_path *path; | |
3268 | int ret; | |
4a500fd1 | 3269 | int err = 0; |
e02119d5 | 3270 | int bytes_del = 0; |
49f34d1f | 3271 | u64 dir_ino = btrfs_ino(dir); |
e02119d5 | 3272 | |
49f34d1f | 3273 | if (dir->logged_trans < trans->transid) |
3a5f1d45 CM |
3274 | return 0; |
3275 | ||
e02119d5 CM |
3276 | ret = join_running_log_trans(root); |
3277 | if (ret) | |
3278 | return 0; | |
3279 | ||
49f34d1f | 3280 | mutex_lock(&dir->log_mutex); |
e02119d5 CM |
3281 | |
3282 | log = root->log_root; | |
3283 | path = btrfs_alloc_path(); | |
a62f44a5 TI |
3284 | if (!path) { |
3285 | err = -ENOMEM; | |
3286 | goto out_unlock; | |
3287 | } | |
2a29edc6 | 3288 | |
33345d01 | 3289 | di = btrfs_lookup_dir_item(trans, log, path, dir_ino, |
e02119d5 | 3290 | name, name_len, -1); |
4a500fd1 YZ |
3291 | if (IS_ERR(di)) { |
3292 | err = PTR_ERR(di); | |
3293 | goto fail; | |
3294 | } | |
3295 | if (di) { | |
e02119d5 CM |
3296 | ret = btrfs_delete_one_dir_name(trans, log, path, di); |
3297 | bytes_del += name_len; | |
3650860b JB |
3298 | if (ret) { |
3299 | err = ret; | |
3300 | goto fail; | |
3301 | } | |
e02119d5 | 3302 | } |
b3b4aa74 | 3303 | btrfs_release_path(path); |
33345d01 | 3304 | di = btrfs_lookup_dir_index_item(trans, log, path, dir_ino, |
e02119d5 | 3305 | index, name, name_len, -1); |
4a500fd1 YZ |
3306 | if (IS_ERR(di)) { |
3307 | err = PTR_ERR(di); | |
3308 | goto fail; | |
3309 | } | |
3310 | if (di) { | |
e02119d5 CM |
3311 | ret = btrfs_delete_one_dir_name(trans, log, path, di); |
3312 | bytes_del += name_len; | |
3650860b JB |
3313 | if (ret) { |
3314 | err = ret; | |
3315 | goto fail; | |
3316 | } | |
e02119d5 CM |
3317 | } |
3318 | ||
3319 | /* update the directory size in the log to reflect the names | |
3320 | * we have removed | |
3321 | */ | |
3322 | if (bytes_del) { | |
3323 | struct btrfs_key key; | |
3324 | ||
33345d01 | 3325 | key.objectid = dir_ino; |
e02119d5 CM |
3326 | key.offset = 0; |
3327 | key.type = BTRFS_INODE_ITEM_KEY; | |
b3b4aa74 | 3328 | btrfs_release_path(path); |
e02119d5 CM |
3329 | |
3330 | ret = btrfs_search_slot(trans, log, &key, path, 0, 1); | |
4a500fd1 YZ |
3331 | if (ret < 0) { |
3332 | err = ret; | |
3333 | goto fail; | |
3334 | } | |
e02119d5 CM |
3335 | if (ret == 0) { |
3336 | struct btrfs_inode_item *item; | |
3337 | u64 i_size; | |
3338 | ||
3339 | item = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
3340 | struct btrfs_inode_item); | |
3341 | i_size = btrfs_inode_size(path->nodes[0], item); | |
3342 | if (i_size > bytes_del) | |
3343 | i_size -= bytes_del; | |
3344 | else | |
3345 | i_size = 0; | |
3346 | btrfs_set_inode_size(path->nodes[0], item, i_size); | |
3347 | btrfs_mark_buffer_dirty(path->nodes[0]); | |
3348 | } else | |
3349 | ret = 0; | |
b3b4aa74 | 3350 | btrfs_release_path(path); |
e02119d5 | 3351 | } |
4a500fd1 | 3352 | fail: |
e02119d5 | 3353 | btrfs_free_path(path); |
a62f44a5 | 3354 | out_unlock: |
49f34d1f | 3355 | mutex_unlock(&dir->log_mutex); |
4a500fd1 | 3356 | if (ret == -ENOSPC) { |
995946dd | 3357 | btrfs_set_log_full_commit(root->fs_info, trans); |
4a500fd1 | 3358 | ret = 0; |
79787eaa | 3359 | } else if (ret < 0) |
66642832 | 3360 | btrfs_abort_transaction(trans, ret); |
79787eaa | 3361 | |
12fcfd22 | 3362 | btrfs_end_log_trans(root); |
e02119d5 | 3363 | |
411fc6bc | 3364 | return err; |
e02119d5 CM |
3365 | } |
3366 | ||
3367 | /* see comments for btrfs_del_dir_entries_in_log */ | |
3368 | int btrfs_del_inode_ref_in_log(struct btrfs_trans_handle *trans, | |
3369 | struct btrfs_root *root, | |
3370 | const char *name, int name_len, | |
a491abb2 | 3371 | struct btrfs_inode *inode, u64 dirid) |
e02119d5 | 3372 | { |
0b246afa | 3373 | struct btrfs_fs_info *fs_info = root->fs_info; |
e02119d5 CM |
3374 | struct btrfs_root *log; |
3375 | u64 index; | |
3376 | int ret; | |
3377 | ||
a491abb2 | 3378 | if (inode->logged_trans < trans->transid) |
3a5f1d45 CM |
3379 | return 0; |
3380 | ||
e02119d5 CM |
3381 | ret = join_running_log_trans(root); |
3382 | if (ret) | |
3383 | return 0; | |
3384 | log = root->log_root; | |
a491abb2 | 3385 | mutex_lock(&inode->log_mutex); |
e02119d5 | 3386 | |
a491abb2 | 3387 | ret = btrfs_del_inode_ref(trans, log, name, name_len, btrfs_ino(inode), |
e02119d5 | 3388 | dirid, &index); |
a491abb2 | 3389 | mutex_unlock(&inode->log_mutex); |
4a500fd1 | 3390 | if (ret == -ENOSPC) { |
0b246afa | 3391 | btrfs_set_log_full_commit(fs_info, trans); |
4a500fd1 | 3392 | ret = 0; |
79787eaa | 3393 | } else if (ret < 0 && ret != -ENOENT) |
66642832 | 3394 | btrfs_abort_transaction(trans, ret); |
12fcfd22 | 3395 | btrfs_end_log_trans(root); |
e02119d5 | 3396 | |
e02119d5 CM |
3397 | return ret; |
3398 | } | |
3399 | ||
3400 | /* | |
3401 | * creates a range item in the log for 'dirid'. first_offset and | |
3402 | * last_offset tell us which parts of the key space the log should | |
3403 | * be considered authoritative for. | |
3404 | */ | |
3405 | static noinline int insert_dir_log_key(struct btrfs_trans_handle *trans, | |
3406 | struct btrfs_root *log, | |
3407 | struct btrfs_path *path, | |
3408 | int key_type, u64 dirid, | |
3409 | u64 first_offset, u64 last_offset) | |
3410 | { | |
3411 | int ret; | |
3412 | struct btrfs_key key; | |
3413 | struct btrfs_dir_log_item *item; | |
3414 | ||
3415 | key.objectid = dirid; | |
3416 | key.offset = first_offset; | |
3417 | if (key_type == BTRFS_DIR_ITEM_KEY) | |
3418 | key.type = BTRFS_DIR_LOG_ITEM_KEY; | |
3419 | else | |
3420 | key.type = BTRFS_DIR_LOG_INDEX_KEY; | |
3421 | ret = btrfs_insert_empty_item(trans, log, path, &key, sizeof(*item)); | |
4a500fd1 YZ |
3422 | if (ret) |
3423 | return ret; | |
e02119d5 CM |
3424 | |
3425 | item = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
3426 | struct btrfs_dir_log_item); | |
3427 | btrfs_set_dir_log_end(path->nodes[0], item, last_offset); | |
3428 | btrfs_mark_buffer_dirty(path->nodes[0]); | |
b3b4aa74 | 3429 | btrfs_release_path(path); |
e02119d5 CM |
3430 | return 0; |
3431 | } | |
3432 | ||
3433 | /* | |
3434 | * log all the items included in the current transaction for a given | |
3435 | * directory. This also creates the range items in the log tree required | |
3436 | * to replay anything deleted before the fsync | |
3437 | */ | |
3438 | static noinline int log_dir_items(struct btrfs_trans_handle *trans, | |
684a5773 | 3439 | struct btrfs_root *root, struct btrfs_inode *inode, |
e02119d5 CM |
3440 | struct btrfs_path *path, |
3441 | struct btrfs_path *dst_path, int key_type, | |
2f2ff0ee | 3442 | struct btrfs_log_ctx *ctx, |
e02119d5 CM |
3443 | u64 min_offset, u64 *last_offset_ret) |
3444 | { | |
3445 | struct btrfs_key min_key; | |
e02119d5 CM |
3446 | struct btrfs_root *log = root->log_root; |
3447 | struct extent_buffer *src; | |
4a500fd1 | 3448 | int err = 0; |
e02119d5 CM |
3449 | int ret; |
3450 | int i; | |
3451 | int nritems; | |
3452 | u64 first_offset = min_offset; | |
3453 | u64 last_offset = (u64)-1; | |
684a5773 | 3454 | u64 ino = btrfs_ino(inode); |
e02119d5 CM |
3455 | |
3456 | log = root->log_root; | |
e02119d5 | 3457 | |
33345d01 | 3458 | min_key.objectid = ino; |
e02119d5 CM |
3459 | min_key.type = key_type; |
3460 | min_key.offset = min_offset; | |
3461 | ||
6174d3cb | 3462 | ret = btrfs_search_forward(root, &min_key, path, trans->transid); |
e02119d5 CM |
3463 | |
3464 | /* | |
3465 | * we didn't find anything from this transaction, see if there | |
3466 | * is anything at all | |
3467 | */ | |
33345d01 LZ |
3468 | if (ret != 0 || min_key.objectid != ino || min_key.type != key_type) { |
3469 | min_key.objectid = ino; | |
e02119d5 CM |
3470 | min_key.type = key_type; |
3471 | min_key.offset = (u64)-1; | |
b3b4aa74 | 3472 | btrfs_release_path(path); |
e02119d5 CM |
3473 | ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0); |
3474 | if (ret < 0) { | |
b3b4aa74 | 3475 | btrfs_release_path(path); |
e02119d5 CM |
3476 | return ret; |
3477 | } | |
33345d01 | 3478 | ret = btrfs_previous_item(root, path, ino, key_type); |
e02119d5 CM |
3479 | |
3480 | /* if ret == 0 there are items for this type, | |
3481 | * create a range to tell us the last key of this type. | |
3482 | * otherwise, there are no items in this directory after | |
3483 | * *min_offset, and we create a range to indicate that. | |
3484 | */ | |
3485 | if (ret == 0) { | |
3486 | struct btrfs_key tmp; | |
3487 | btrfs_item_key_to_cpu(path->nodes[0], &tmp, | |
3488 | path->slots[0]); | |
d397712b | 3489 | if (key_type == tmp.type) |
e02119d5 | 3490 | first_offset = max(min_offset, tmp.offset) + 1; |
e02119d5 CM |
3491 | } |
3492 | goto done; | |
3493 | } | |
3494 | ||
3495 | /* go backward to find any previous key */ | |
33345d01 | 3496 | ret = btrfs_previous_item(root, path, ino, key_type); |
e02119d5 CM |
3497 | if (ret == 0) { |
3498 | struct btrfs_key tmp; | |
3499 | btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]); | |
3500 | if (key_type == tmp.type) { | |
3501 | first_offset = tmp.offset; | |
3502 | ret = overwrite_item(trans, log, dst_path, | |
3503 | path->nodes[0], path->slots[0], | |
3504 | &tmp); | |
4a500fd1 YZ |
3505 | if (ret) { |
3506 | err = ret; | |
3507 | goto done; | |
3508 | } | |
e02119d5 CM |
3509 | } |
3510 | } | |
b3b4aa74 | 3511 | btrfs_release_path(path); |
e02119d5 CM |
3512 | |
3513 | /* find the first key from this transaction again */ | |
3514 | ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0); | |
fae7f21c | 3515 | if (WARN_ON(ret != 0)) |
e02119d5 | 3516 | goto done; |
e02119d5 CM |
3517 | |
3518 | /* | |
3519 | * we have a block from this transaction, log every item in it | |
3520 | * from our directory | |
3521 | */ | |
d397712b | 3522 | while (1) { |
e02119d5 CM |
3523 | struct btrfs_key tmp; |
3524 | src = path->nodes[0]; | |
3525 | nritems = btrfs_header_nritems(src); | |
3526 | for (i = path->slots[0]; i < nritems; i++) { | |
2f2ff0ee FM |
3527 | struct btrfs_dir_item *di; |
3528 | ||
e02119d5 CM |
3529 | btrfs_item_key_to_cpu(src, &min_key, i); |
3530 | ||
33345d01 | 3531 | if (min_key.objectid != ino || min_key.type != key_type) |
e02119d5 CM |
3532 | goto done; |
3533 | ret = overwrite_item(trans, log, dst_path, src, i, | |
3534 | &min_key); | |
4a500fd1 YZ |
3535 | if (ret) { |
3536 | err = ret; | |
3537 | goto done; | |
3538 | } | |
2f2ff0ee FM |
3539 | |
3540 | /* | |
3541 | * We must make sure that when we log a directory entry, | |
3542 | * the corresponding inode, after log replay, has a | |
3543 | * matching link count. For example: | |
3544 | * | |
3545 | * touch foo | |
3546 | * mkdir mydir | |
3547 | * sync | |
3548 | * ln foo mydir/bar | |
3549 | * xfs_io -c "fsync" mydir | |
3550 | * <crash> | |
3551 | * <mount fs and log replay> | |
3552 | * | |
3553 | * Would result in a fsync log that when replayed, our | |
3554 | * file inode would have a link count of 1, but we get | |
3555 | * two directory entries pointing to the same inode. | |
3556 | * After removing one of the names, it would not be | |
3557 | * possible to remove the other name, which resulted | |
3558 | * always in stale file handle errors, and would not | |
3559 | * be possible to rmdir the parent directory, since | |
3560 | * its i_size could never decrement to the value | |
3561 | * BTRFS_EMPTY_DIR_SIZE, resulting in -ENOTEMPTY errors. | |
3562 | */ | |
3563 | di = btrfs_item_ptr(src, i, struct btrfs_dir_item); | |
3564 | btrfs_dir_item_key_to_cpu(src, di, &tmp); | |
3565 | if (ctx && | |
3566 | (btrfs_dir_transid(src, di) == trans->transid || | |
3567 | btrfs_dir_type(src, di) == BTRFS_FT_DIR) && | |
3568 | tmp.type != BTRFS_ROOT_ITEM_KEY) | |
3569 | ctx->log_new_dentries = true; | |
e02119d5 CM |
3570 | } |
3571 | path->slots[0] = nritems; | |
3572 | ||
3573 | /* | |
3574 | * look ahead to the next item and see if it is also | |
3575 | * from this directory and from this transaction | |
3576 | */ | |
3577 | ret = btrfs_next_leaf(root, path); | |
80c0b421 LB |
3578 | if (ret) { |
3579 | if (ret == 1) | |
3580 | last_offset = (u64)-1; | |
3581 | else | |
3582 | err = ret; | |
e02119d5 CM |
3583 | goto done; |
3584 | } | |
3585 | btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]); | |
33345d01 | 3586 | if (tmp.objectid != ino || tmp.type != key_type) { |
e02119d5 CM |
3587 | last_offset = (u64)-1; |
3588 | goto done; | |
3589 | } | |
3590 | if (btrfs_header_generation(path->nodes[0]) != trans->transid) { | |
3591 | ret = overwrite_item(trans, log, dst_path, | |
3592 | path->nodes[0], path->slots[0], | |
3593 | &tmp); | |
4a500fd1 YZ |
3594 | if (ret) |
3595 | err = ret; | |
3596 | else | |
3597 | last_offset = tmp.offset; | |
e02119d5 CM |
3598 | goto done; |
3599 | } | |
3600 | } | |
3601 | done: | |
b3b4aa74 DS |
3602 | btrfs_release_path(path); |
3603 | btrfs_release_path(dst_path); | |
e02119d5 | 3604 | |
4a500fd1 YZ |
3605 | if (err == 0) { |
3606 | *last_offset_ret = last_offset; | |
3607 | /* | |
3608 | * insert the log range keys to indicate where the log | |
3609 | * is valid | |
3610 | */ | |
3611 | ret = insert_dir_log_key(trans, log, path, key_type, | |
33345d01 | 3612 | ino, first_offset, last_offset); |
4a500fd1 YZ |
3613 | if (ret) |
3614 | err = ret; | |
3615 | } | |
3616 | return err; | |
e02119d5 CM |
3617 | } |
3618 | ||
3619 | /* | |
3620 | * logging directories is very similar to logging inodes, We find all the items | |
3621 | * from the current transaction and write them to the log. | |
3622 | * | |
3623 | * The recovery code scans the directory in the subvolume, and if it finds a | |
3624 | * key in the range logged that is not present in the log tree, then it means | |
3625 | * that dir entry was unlinked during the transaction. | |
3626 | * | |
3627 | * In order for that scan to work, we must include one key smaller than | |
3628 | * the smallest logged by this transaction and one key larger than the largest | |
3629 | * key logged by this transaction. | |
3630 | */ | |
3631 | static noinline int log_directory_changes(struct btrfs_trans_handle *trans, | |
dbf39ea4 | 3632 | struct btrfs_root *root, struct btrfs_inode *inode, |
e02119d5 | 3633 | struct btrfs_path *path, |
2f2ff0ee FM |
3634 | struct btrfs_path *dst_path, |
3635 | struct btrfs_log_ctx *ctx) | |
e02119d5 CM |
3636 | { |
3637 | u64 min_key; | |
3638 | u64 max_key; | |
3639 | int ret; | |
3640 | int key_type = BTRFS_DIR_ITEM_KEY; | |
3641 | ||
3642 | again: | |
3643 | min_key = 0; | |
3644 | max_key = 0; | |
d397712b | 3645 | while (1) { |
dbf39ea4 NB |
3646 | ret = log_dir_items(trans, root, inode, path, dst_path, key_type, |
3647 | ctx, min_key, &max_key); | |
4a500fd1 YZ |
3648 | if (ret) |
3649 | return ret; | |
e02119d5 CM |
3650 | if (max_key == (u64)-1) |
3651 | break; | |
3652 | min_key = max_key + 1; | |
3653 | } | |
3654 | ||
3655 | if (key_type == BTRFS_DIR_ITEM_KEY) { | |
3656 | key_type = BTRFS_DIR_INDEX_KEY; | |
3657 | goto again; | |
3658 | } | |
3659 | return 0; | |
3660 | } | |
3661 | ||
3662 | /* | |
3663 | * a helper function to drop items from the log before we relog an | |
3664 | * inode. max_key_type indicates the highest item type to remove. | |
3665 | * This cannot be run for file data extents because it does not | |
3666 | * free the extents they point to. | |
3667 | */ | |
3668 | static int drop_objectid_items(struct btrfs_trans_handle *trans, | |
3669 | struct btrfs_root *log, | |
3670 | struct btrfs_path *path, | |
3671 | u64 objectid, int max_key_type) | |
3672 | { | |
3673 | int ret; | |
3674 | struct btrfs_key key; | |
3675 | struct btrfs_key found_key; | |
18ec90d6 | 3676 | int start_slot; |
e02119d5 CM |
3677 | |
3678 | key.objectid = objectid; | |
3679 | key.type = max_key_type; | |
3680 | key.offset = (u64)-1; | |
3681 | ||
d397712b | 3682 | while (1) { |
e02119d5 | 3683 | ret = btrfs_search_slot(trans, log, &key, path, -1, 1); |
3650860b | 3684 | BUG_ON(ret == 0); /* Logic error */ |
4a500fd1 | 3685 | if (ret < 0) |
e02119d5 CM |
3686 | break; |
3687 | ||
3688 | if (path->slots[0] == 0) | |
3689 | break; | |
3690 | ||
3691 | path->slots[0]--; | |
3692 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, | |
3693 | path->slots[0]); | |
3694 | ||
3695 | if (found_key.objectid != objectid) | |
3696 | break; | |
3697 | ||
18ec90d6 JB |
3698 | found_key.offset = 0; |
3699 | found_key.type = 0; | |
3700 | ret = btrfs_bin_search(path->nodes[0], &found_key, 0, | |
3701 | &start_slot); | |
3702 | ||
3703 | ret = btrfs_del_items(trans, log, path, start_slot, | |
3704 | path->slots[0] - start_slot + 1); | |
3705 | /* | |
3706 | * If start slot isn't 0 then we don't need to re-search, we've | |
3707 | * found the last guy with the objectid in this tree. | |
3708 | */ | |
3709 | if (ret || start_slot != 0) | |
65a246c5 | 3710 | break; |
b3b4aa74 | 3711 | btrfs_release_path(path); |
e02119d5 | 3712 | } |
b3b4aa74 | 3713 | btrfs_release_path(path); |
5bdbeb21 JB |
3714 | if (ret > 0) |
3715 | ret = 0; | |
4a500fd1 | 3716 | return ret; |
e02119d5 CM |
3717 | } |
3718 | ||
94edf4ae JB |
3719 | static void fill_inode_item(struct btrfs_trans_handle *trans, |
3720 | struct extent_buffer *leaf, | |
3721 | struct btrfs_inode_item *item, | |
1a4bcf47 FM |
3722 | struct inode *inode, int log_inode_only, |
3723 | u64 logged_isize) | |
94edf4ae | 3724 | { |
0b1c6cca JB |
3725 | struct btrfs_map_token token; |
3726 | ||
3727 | btrfs_init_map_token(&token); | |
94edf4ae JB |
3728 | |
3729 | if (log_inode_only) { | |
3730 | /* set the generation to zero so the recover code | |
3731 | * can tell the difference between an logging | |
3732 | * just to say 'this inode exists' and a logging | |
3733 | * to say 'update this inode with these values' | |
3734 | */ | |
0b1c6cca | 3735 | btrfs_set_token_inode_generation(leaf, item, 0, &token); |
1a4bcf47 | 3736 | btrfs_set_token_inode_size(leaf, item, logged_isize, &token); |
94edf4ae | 3737 | } else { |
0b1c6cca JB |
3738 | btrfs_set_token_inode_generation(leaf, item, |
3739 | BTRFS_I(inode)->generation, | |
3740 | &token); | |
3741 | btrfs_set_token_inode_size(leaf, item, inode->i_size, &token); | |
3742 | } | |
3743 | ||
3744 | btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token); | |
3745 | btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token); | |
3746 | btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token); | |
3747 | btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token); | |
3748 | ||
a937b979 | 3749 | btrfs_set_token_timespec_sec(leaf, &item->atime, |
0b1c6cca | 3750 | inode->i_atime.tv_sec, &token); |
a937b979 | 3751 | btrfs_set_token_timespec_nsec(leaf, &item->atime, |
0b1c6cca JB |
3752 | inode->i_atime.tv_nsec, &token); |
3753 | ||
a937b979 | 3754 | btrfs_set_token_timespec_sec(leaf, &item->mtime, |
0b1c6cca | 3755 | inode->i_mtime.tv_sec, &token); |
a937b979 | 3756 | btrfs_set_token_timespec_nsec(leaf, &item->mtime, |
0b1c6cca JB |
3757 | inode->i_mtime.tv_nsec, &token); |
3758 | ||
a937b979 | 3759 | btrfs_set_token_timespec_sec(leaf, &item->ctime, |
0b1c6cca | 3760 | inode->i_ctime.tv_sec, &token); |
a937b979 | 3761 | btrfs_set_token_timespec_nsec(leaf, &item->ctime, |
0b1c6cca JB |
3762 | inode->i_ctime.tv_nsec, &token); |
3763 | ||
3764 | btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode), | |
3765 | &token); | |
3766 | ||
c7f88c4e JL |
3767 | btrfs_set_token_inode_sequence(leaf, item, |
3768 | inode_peek_iversion(inode), &token); | |
0b1c6cca JB |
3769 | btrfs_set_token_inode_transid(leaf, item, trans->transid, &token); |
3770 | btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token); | |
3771 | btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token); | |
3772 | btrfs_set_token_inode_block_group(leaf, item, 0, &token); | |
94edf4ae JB |
3773 | } |
3774 | ||
a95249b3 JB |
3775 | static int log_inode_item(struct btrfs_trans_handle *trans, |
3776 | struct btrfs_root *log, struct btrfs_path *path, | |
6d889a3b | 3777 | struct btrfs_inode *inode) |
a95249b3 JB |
3778 | { |
3779 | struct btrfs_inode_item *inode_item; | |
a95249b3 JB |
3780 | int ret; |
3781 | ||
efd0c405 | 3782 | ret = btrfs_insert_empty_item(trans, log, path, |
6d889a3b | 3783 | &inode->location, sizeof(*inode_item)); |
a95249b3 JB |
3784 | if (ret && ret != -EEXIST) |
3785 | return ret; | |
3786 | inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
3787 | struct btrfs_inode_item); | |
6d889a3b NB |
3788 | fill_inode_item(trans, path->nodes[0], inode_item, &inode->vfs_inode, |
3789 | 0, 0); | |
a95249b3 JB |
3790 | btrfs_release_path(path); |
3791 | return 0; | |
3792 | } | |
3793 | ||
31ff1cd2 | 3794 | static noinline int copy_items(struct btrfs_trans_handle *trans, |
44d70e19 | 3795 | struct btrfs_inode *inode, |
31ff1cd2 | 3796 | struct btrfs_path *dst_path, |
16e7549f | 3797 | struct btrfs_path *src_path, u64 *last_extent, |
1a4bcf47 FM |
3798 | int start_slot, int nr, int inode_only, |
3799 | u64 logged_isize) | |
31ff1cd2 | 3800 | { |
3ffbd68c | 3801 | struct btrfs_fs_info *fs_info = trans->fs_info; |
31ff1cd2 CM |
3802 | unsigned long src_offset; |
3803 | unsigned long dst_offset; | |
44d70e19 | 3804 | struct btrfs_root *log = inode->root->log_root; |
31ff1cd2 CM |
3805 | struct btrfs_file_extent_item *extent; |
3806 | struct btrfs_inode_item *inode_item; | |
16e7549f JB |
3807 | struct extent_buffer *src = src_path->nodes[0]; |
3808 | struct btrfs_key first_key, last_key, key; | |
31ff1cd2 CM |
3809 | int ret; |
3810 | struct btrfs_key *ins_keys; | |
3811 | u32 *ins_sizes; | |
3812 | char *ins_data; | |
3813 | int i; | |
d20f7043 | 3814 | struct list_head ordered_sums; |
44d70e19 | 3815 | int skip_csum = inode->flags & BTRFS_INODE_NODATASUM; |
16e7549f | 3816 | bool has_extents = false; |
74121f7c | 3817 | bool need_find_last_extent = true; |
16e7549f | 3818 | bool done = false; |
d20f7043 CM |
3819 | |
3820 | INIT_LIST_HEAD(&ordered_sums); | |
31ff1cd2 CM |
3821 | |
3822 | ins_data = kmalloc(nr * sizeof(struct btrfs_key) + | |
3823 | nr * sizeof(u32), GFP_NOFS); | |
2a29edc6 | 3824 | if (!ins_data) |
3825 | return -ENOMEM; | |
3826 | ||
16e7549f JB |
3827 | first_key.objectid = (u64)-1; |
3828 | ||
31ff1cd2 CM |
3829 | ins_sizes = (u32 *)ins_data; |
3830 | ins_keys = (struct btrfs_key *)(ins_data + nr * sizeof(u32)); | |
3831 | ||
3832 | for (i = 0; i < nr; i++) { | |
3833 | ins_sizes[i] = btrfs_item_size_nr(src, i + start_slot); | |
3834 | btrfs_item_key_to_cpu(src, ins_keys + i, i + start_slot); | |
3835 | } | |
3836 | ret = btrfs_insert_empty_items(trans, log, dst_path, | |
3837 | ins_keys, ins_sizes, nr); | |
4a500fd1 YZ |
3838 | if (ret) { |
3839 | kfree(ins_data); | |
3840 | return ret; | |
3841 | } | |
31ff1cd2 | 3842 | |
5d4f98a2 | 3843 | for (i = 0; i < nr; i++, dst_path->slots[0]++) { |
31ff1cd2 CM |
3844 | dst_offset = btrfs_item_ptr_offset(dst_path->nodes[0], |
3845 | dst_path->slots[0]); | |
3846 | ||
3847 | src_offset = btrfs_item_ptr_offset(src, start_slot + i); | |
3848 | ||
0dde10be | 3849 | if (i == nr - 1) |
16e7549f JB |
3850 | last_key = ins_keys[i]; |
3851 | ||
94edf4ae | 3852 | if (ins_keys[i].type == BTRFS_INODE_ITEM_KEY) { |
31ff1cd2 CM |
3853 | inode_item = btrfs_item_ptr(dst_path->nodes[0], |
3854 | dst_path->slots[0], | |
3855 | struct btrfs_inode_item); | |
94edf4ae | 3856 | fill_inode_item(trans, dst_path->nodes[0], inode_item, |
f85b7379 DS |
3857 | &inode->vfs_inode, |
3858 | inode_only == LOG_INODE_EXISTS, | |
1a4bcf47 | 3859 | logged_isize); |
94edf4ae JB |
3860 | } else { |
3861 | copy_extent_buffer(dst_path->nodes[0], src, dst_offset, | |
3862 | src_offset, ins_sizes[i]); | |
31ff1cd2 | 3863 | } |
94edf4ae | 3864 | |
16e7549f JB |
3865 | /* |
3866 | * We set need_find_last_extent here in case we know we were | |
3867 | * processing other items and then walk into the first extent in | |
3868 | * the inode. If we don't hit an extent then nothing changes, | |
3869 | * we'll do the last search the next time around. | |
3870 | */ | |
3871 | if (ins_keys[i].type == BTRFS_EXTENT_DATA_KEY) { | |
3872 | has_extents = true; | |
74121f7c | 3873 | if (first_key.objectid == (u64)-1) |
16e7549f JB |
3874 | first_key = ins_keys[i]; |
3875 | } else { | |
3876 | need_find_last_extent = false; | |
3877 | } | |
3878 | ||
31ff1cd2 CM |
3879 | /* take a reference on file data extents so that truncates |
3880 | * or deletes of this inode don't have to relog the inode | |
3881 | * again | |
3882 | */ | |
962a298f | 3883 | if (ins_keys[i].type == BTRFS_EXTENT_DATA_KEY && |
d2794405 | 3884 | !skip_csum) { |
31ff1cd2 CM |
3885 | int found_type; |
3886 | extent = btrfs_item_ptr(src, start_slot + i, | |
3887 | struct btrfs_file_extent_item); | |
3888 | ||
8e531cdf | 3889 | if (btrfs_file_extent_generation(src, extent) < trans->transid) |
3890 | continue; | |
3891 | ||
31ff1cd2 | 3892 | found_type = btrfs_file_extent_type(src, extent); |
6f1fed77 | 3893 | if (found_type == BTRFS_FILE_EXTENT_REG) { |
5d4f98a2 YZ |
3894 | u64 ds, dl, cs, cl; |
3895 | ds = btrfs_file_extent_disk_bytenr(src, | |
3896 | extent); | |
3897 | /* ds == 0 is a hole */ | |
3898 | if (ds == 0) | |
3899 | continue; | |
3900 | ||
3901 | dl = btrfs_file_extent_disk_num_bytes(src, | |
3902 | extent); | |
3903 | cs = btrfs_file_extent_offset(src, extent); | |
3904 | cl = btrfs_file_extent_num_bytes(src, | |
a419aef8 | 3905 | extent); |
580afd76 CM |
3906 | if (btrfs_file_extent_compression(src, |
3907 | extent)) { | |
3908 | cs = 0; | |
3909 | cl = dl; | |
3910 | } | |
5d4f98a2 YZ |
3911 | |
3912 | ret = btrfs_lookup_csums_range( | |
0b246afa | 3913 | fs_info->csum_root, |
5d4f98a2 | 3914 | ds + cs, ds + cs + cl - 1, |
a2de733c | 3915 | &ordered_sums, 0); |
3650860b JB |
3916 | if (ret) { |
3917 | btrfs_release_path(dst_path); | |
3918 | kfree(ins_data); | |
3919 | return ret; | |
3920 | } | |
31ff1cd2 CM |
3921 | } |
3922 | } | |
31ff1cd2 CM |
3923 | } |
3924 | ||
3925 | btrfs_mark_buffer_dirty(dst_path->nodes[0]); | |
b3b4aa74 | 3926 | btrfs_release_path(dst_path); |
31ff1cd2 | 3927 | kfree(ins_data); |
d20f7043 CM |
3928 | |
3929 | /* | |
3930 | * we have to do this after the loop above to avoid changing the | |
3931 | * log tree while trying to change the log tree. | |
3932 | */ | |
4a500fd1 | 3933 | ret = 0; |
d397712b | 3934 | while (!list_empty(&ordered_sums)) { |
d20f7043 CM |
3935 | struct btrfs_ordered_sum *sums = list_entry(ordered_sums.next, |
3936 | struct btrfs_ordered_sum, | |
3937 | list); | |
4a500fd1 YZ |
3938 | if (!ret) |
3939 | ret = btrfs_csum_file_blocks(trans, log, sums); | |
d20f7043 CM |
3940 | list_del(&sums->list); |
3941 | kfree(sums); | |
3942 | } | |
16e7549f JB |
3943 | |
3944 | if (!has_extents) | |
3945 | return ret; | |
3946 | ||
74121f7c FM |
3947 | if (need_find_last_extent && *last_extent == first_key.offset) { |
3948 | /* | |
3949 | * We don't have any leafs between our current one and the one | |
3950 | * we processed before that can have file extent items for our | |
3951 | * inode (and have a generation number smaller than our current | |
3952 | * transaction id). | |
3953 | */ | |
3954 | need_find_last_extent = false; | |
3955 | } | |
3956 | ||
16e7549f JB |
3957 | /* |
3958 | * Because we use btrfs_search_forward we could skip leaves that were | |
3959 | * not modified and then assume *last_extent is valid when it really | |
3960 | * isn't. So back up to the previous leaf and read the end of the last | |
3961 | * extent before we go and fill in holes. | |
3962 | */ | |
3963 | if (need_find_last_extent) { | |
3964 | u64 len; | |
3965 | ||
44d70e19 | 3966 | ret = btrfs_prev_leaf(inode->root, src_path); |
16e7549f JB |
3967 | if (ret < 0) |
3968 | return ret; | |
3969 | if (ret) | |
3970 | goto fill_holes; | |
3971 | if (src_path->slots[0]) | |
3972 | src_path->slots[0]--; | |
3973 | src = src_path->nodes[0]; | |
3974 | btrfs_item_key_to_cpu(src, &key, src_path->slots[0]); | |
44d70e19 | 3975 | if (key.objectid != btrfs_ino(inode) || |
16e7549f JB |
3976 | key.type != BTRFS_EXTENT_DATA_KEY) |
3977 | goto fill_holes; | |
3978 | extent = btrfs_item_ptr(src, src_path->slots[0], | |
3979 | struct btrfs_file_extent_item); | |
3980 | if (btrfs_file_extent_type(src, extent) == | |
3981 | BTRFS_FILE_EXTENT_INLINE) { | |
e41ca589 | 3982 | len = btrfs_file_extent_ram_bytes(src, extent); |
16e7549f | 3983 | *last_extent = ALIGN(key.offset + len, |
0b246afa | 3984 | fs_info->sectorsize); |
16e7549f JB |
3985 | } else { |
3986 | len = btrfs_file_extent_num_bytes(src, extent); | |
3987 | *last_extent = key.offset + len; | |
3988 | } | |
3989 | } | |
3990 | fill_holes: | |
3991 | /* So we did prev_leaf, now we need to move to the next leaf, but a few | |
3992 | * things could have happened | |
3993 | * | |
3994 | * 1) A merge could have happened, so we could currently be on a leaf | |
3995 | * that holds what we were copying in the first place. | |
3996 | * 2) A split could have happened, and now not all of the items we want | |
3997 | * are on the same leaf. | |
3998 | * | |
3999 | * So we need to adjust how we search for holes, we need to drop the | |
4000 | * path and re-search for the first extent key we found, and then walk | |
4001 | * forward until we hit the last one we copied. | |
4002 | */ | |
4003 | if (need_find_last_extent) { | |
4004 | /* btrfs_prev_leaf could return 1 without releasing the path */ | |
4005 | btrfs_release_path(src_path); | |
f85b7379 DS |
4006 | ret = btrfs_search_slot(NULL, inode->root, &first_key, |
4007 | src_path, 0, 0); | |
16e7549f JB |
4008 | if (ret < 0) |
4009 | return ret; | |
4010 | ASSERT(ret == 0); | |
4011 | src = src_path->nodes[0]; | |
4012 | i = src_path->slots[0]; | |
4013 | } else { | |
4014 | i = start_slot; | |
4015 | } | |
4016 | ||
4017 | /* | |
4018 | * Ok so here we need to go through and fill in any holes we may have | |
4019 | * to make sure that holes are punched for those areas in case they had | |
4020 | * extents previously. | |
4021 | */ | |
4022 | while (!done) { | |
4023 | u64 offset, len; | |
4024 | u64 extent_end; | |
4025 | ||
4026 | if (i >= btrfs_header_nritems(src_path->nodes[0])) { | |
44d70e19 | 4027 | ret = btrfs_next_leaf(inode->root, src_path); |
16e7549f JB |
4028 | if (ret < 0) |
4029 | return ret; | |
4030 | ASSERT(ret == 0); | |
4031 | src = src_path->nodes[0]; | |
4032 | i = 0; | |
8434ec46 | 4033 | need_find_last_extent = true; |
16e7549f JB |
4034 | } |
4035 | ||
4036 | btrfs_item_key_to_cpu(src, &key, i); | |
4037 | if (!btrfs_comp_cpu_keys(&key, &last_key)) | |
4038 | done = true; | |
44d70e19 | 4039 | if (key.objectid != btrfs_ino(inode) || |
16e7549f JB |
4040 | key.type != BTRFS_EXTENT_DATA_KEY) { |
4041 | i++; | |
4042 | continue; | |
4043 | } | |
4044 | extent = btrfs_item_ptr(src, i, struct btrfs_file_extent_item); | |
4045 | if (btrfs_file_extent_type(src, extent) == | |
4046 | BTRFS_FILE_EXTENT_INLINE) { | |
e41ca589 | 4047 | len = btrfs_file_extent_ram_bytes(src, extent); |
da17066c | 4048 | extent_end = ALIGN(key.offset + len, |
0b246afa | 4049 | fs_info->sectorsize); |
16e7549f JB |
4050 | } else { |
4051 | len = btrfs_file_extent_num_bytes(src, extent); | |
4052 | extent_end = key.offset + len; | |
4053 | } | |
4054 | i++; | |
4055 | ||
4056 | if (*last_extent == key.offset) { | |
4057 | *last_extent = extent_end; | |
4058 | continue; | |
4059 | } | |
4060 | offset = *last_extent; | |
4061 | len = key.offset - *last_extent; | |
44d70e19 | 4062 | ret = btrfs_insert_file_extent(trans, log, btrfs_ino(inode), |
f85b7379 | 4063 | offset, 0, 0, len, 0, len, 0, 0, 0); |
16e7549f JB |
4064 | if (ret) |
4065 | break; | |
74121f7c | 4066 | *last_extent = extent_end; |
16e7549f | 4067 | } |
4ee3fad3 FM |
4068 | |
4069 | /* | |
4070 | * Check if there is a hole between the last extent found in our leaf | |
4071 | * and the first extent in the next leaf. If there is one, we need to | |
4072 | * log an explicit hole so that at replay time we can punch the hole. | |
4073 | */ | |
4074 | if (ret == 0 && | |
4075 | key.objectid == btrfs_ino(inode) && | |
4076 | key.type == BTRFS_EXTENT_DATA_KEY && | |
4077 | i == btrfs_header_nritems(src_path->nodes[0])) { | |
4078 | ret = btrfs_next_leaf(inode->root, src_path); | |
4079 | need_find_last_extent = true; | |
4080 | if (ret > 0) { | |
4081 | ret = 0; | |
4082 | } else if (ret == 0) { | |
4083 | btrfs_item_key_to_cpu(src_path->nodes[0], &key, | |
4084 | src_path->slots[0]); | |
4085 | if (key.objectid == btrfs_ino(inode) && | |
4086 | key.type == BTRFS_EXTENT_DATA_KEY && | |
4087 | *last_extent < key.offset) { | |
4088 | const u64 len = key.offset - *last_extent; | |
4089 | ||
4090 | ret = btrfs_insert_file_extent(trans, log, | |
4091 | btrfs_ino(inode), | |
4092 | *last_extent, 0, | |
4093 | 0, len, 0, len, | |
4094 | 0, 0, 0); | |
4095 | } | |
4096 | } | |
4097 | } | |
16e7549f JB |
4098 | /* |
4099 | * Need to let the callers know we dropped the path so they should | |
4100 | * re-search. | |
4101 | */ | |
4102 | if (!ret && need_find_last_extent) | |
4103 | ret = 1; | |
4a500fd1 | 4104 | return ret; |
31ff1cd2 CM |
4105 | } |
4106 | ||
5dc562c5 JB |
4107 | static int extent_cmp(void *priv, struct list_head *a, struct list_head *b) |
4108 | { | |
4109 | struct extent_map *em1, *em2; | |
4110 | ||
4111 | em1 = list_entry(a, struct extent_map, list); | |
4112 | em2 = list_entry(b, struct extent_map, list); | |
4113 | ||
4114 | if (em1->start < em2->start) | |
4115 | return -1; | |
4116 | else if (em1->start > em2->start) | |
4117 | return 1; | |
4118 | return 0; | |
4119 | } | |
4120 | ||
e7175a69 JB |
4121 | static int log_extent_csums(struct btrfs_trans_handle *trans, |
4122 | struct btrfs_inode *inode, | |
a9ecb653 | 4123 | struct btrfs_root *log_root, |
e7175a69 | 4124 | const struct extent_map *em) |
5dc562c5 | 4125 | { |
2ab28f32 JB |
4126 | u64 csum_offset; |
4127 | u64 csum_len; | |
8407f553 FM |
4128 | LIST_HEAD(ordered_sums); |
4129 | int ret = 0; | |
0aa4a17d | 4130 | |
e7175a69 JB |
4131 | if (inode->flags & BTRFS_INODE_NODATASUM || |
4132 | test_bit(EXTENT_FLAG_PREALLOC, &em->flags) || | |
8407f553 | 4133 | em->block_start == EXTENT_MAP_HOLE) |
70c8a91c | 4134 | return 0; |
5dc562c5 | 4135 | |
e7175a69 | 4136 | /* If we're compressed we have to save the entire range of csums. */ |
488111aa FDBM |
4137 | if (em->compress_type) { |
4138 | csum_offset = 0; | |
8407f553 | 4139 | csum_len = max(em->block_len, em->orig_block_len); |
488111aa | 4140 | } else { |
e7175a69 JB |
4141 | csum_offset = em->mod_start - em->start; |
4142 | csum_len = em->mod_len; | |
488111aa | 4143 | } |
2ab28f32 | 4144 | |
70c8a91c | 4145 | /* block start is already adjusted for the file extent offset. */ |
a9ecb653 | 4146 | ret = btrfs_lookup_csums_range(trans->fs_info->csum_root, |
70c8a91c JB |
4147 | em->block_start + csum_offset, |
4148 | em->block_start + csum_offset + | |
4149 | csum_len - 1, &ordered_sums, 0); | |
4150 | if (ret) | |
4151 | return ret; | |
5dc562c5 | 4152 | |
70c8a91c JB |
4153 | while (!list_empty(&ordered_sums)) { |
4154 | struct btrfs_ordered_sum *sums = list_entry(ordered_sums.next, | |
4155 | struct btrfs_ordered_sum, | |
4156 | list); | |
4157 | if (!ret) | |
a9ecb653 | 4158 | ret = btrfs_csum_file_blocks(trans, log_root, sums); |
70c8a91c JB |
4159 | list_del(&sums->list); |
4160 | kfree(sums); | |
5dc562c5 JB |
4161 | } |
4162 | ||
70c8a91c | 4163 | return ret; |
5dc562c5 JB |
4164 | } |
4165 | ||
8407f553 | 4166 | static int log_one_extent(struct btrfs_trans_handle *trans, |
9d122629 | 4167 | struct btrfs_inode *inode, struct btrfs_root *root, |
8407f553 FM |
4168 | const struct extent_map *em, |
4169 | struct btrfs_path *path, | |
8407f553 FM |
4170 | struct btrfs_log_ctx *ctx) |
4171 | { | |
4172 | struct btrfs_root *log = root->log_root; | |
4173 | struct btrfs_file_extent_item *fi; | |
4174 | struct extent_buffer *leaf; | |
4175 | struct btrfs_map_token token; | |
4176 | struct btrfs_key key; | |
4177 | u64 extent_offset = em->start - em->orig_start; | |
4178 | u64 block_len; | |
4179 | int ret; | |
4180 | int extent_inserted = 0; | |
8407f553 | 4181 | |
a9ecb653 | 4182 | ret = log_extent_csums(trans, inode, log, em); |
8407f553 FM |
4183 | if (ret) |
4184 | return ret; | |
4185 | ||
8407f553 FM |
4186 | btrfs_init_map_token(&token); |
4187 | ||
9d122629 | 4188 | ret = __btrfs_drop_extents(trans, log, &inode->vfs_inode, path, em->start, |
8407f553 FM |
4189 | em->start + em->len, NULL, 0, 1, |
4190 | sizeof(*fi), &extent_inserted); | |
4191 | if (ret) | |
4192 | return ret; | |
4193 | ||
4194 | if (!extent_inserted) { | |
9d122629 | 4195 | key.objectid = btrfs_ino(inode); |
8407f553 FM |
4196 | key.type = BTRFS_EXTENT_DATA_KEY; |
4197 | key.offset = em->start; | |
4198 | ||
4199 | ret = btrfs_insert_empty_item(trans, log, path, &key, | |
4200 | sizeof(*fi)); | |
4201 | if (ret) | |
4202 | return ret; | |
4203 | } | |
4204 | leaf = path->nodes[0]; | |
4205 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
4206 | struct btrfs_file_extent_item); | |
4207 | ||
50d9aa99 | 4208 | btrfs_set_token_file_extent_generation(leaf, fi, trans->transid, |
8407f553 FM |
4209 | &token); |
4210 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) | |
4211 | btrfs_set_token_file_extent_type(leaf, fi, | |
4212 | BTRFS_FILE_EXTENT_PREALLOC, | |
4213 | &token); | |
4214 | else | |
4215 | btrfs_set_token_file_extent_type(leaf, fi, | |
4216 | BTRFS_FILE_EXTENT_REG, | |
4217 | &token); | |
4218 | ||
4219 | block_len = max(em->block_len, em->orig_block_len); | |
4220 | if (em->compress_type != BTRFS_COMPRESS_NONE) { | |
4221 | btrfs_set_token_file_extent_disk_bytenr(leaf, fi, | |
4222 | em->block_start, | |
4223 | &token); | |
4224 | btrfs_set_token_file_extent_disk_num_bytes(leaf, fi, block_len, | |
4225 | &token); | |
4226 | } else if (em->block_start < EXTENT_MAP_LAST_BYTE) { | |
4227 | btrfs_set_token_file_extent_disk_bytenr(leaf, fi, | |
4228 | em->block_start - | |
4229 | extent_offset, &token); | |
4230 | btrfs_set_token_file_extent_disk_num_bytes(leaf, fi, block_len, | |
4231 | &token); | |
4232 | } else { | |
4233 | btrfs_set_token_file_extent_disk_bytenr(leaf, fi, 0, &token); | |
4234 | btrfs_set_token_file_extent_disk_num_bytes(leaf, fi, 0, | |
4235 | &token); | |
4236 | } | |
4237 | ||
4238 | btrfs_set_token_file_extent_offset(leaf, fi, extent_offset, &token); | |
4239 | btrfs_set_token_file_extent_num_bytes(leaf, fi, em->len, &token); | |
4240 | btrfs_set_token_file_extent_ram_bytes(leaf, fi, em->ram_bytes, &token); | |
4241 | btrfs_set_token_file_extent_compression(leaf, fi, em->compress_type, | |
4242 | &token); | |
4243 | btrfs_set_token_file_extent_encryption(leaf, fi, 0, &token); | |
4244 | btrfs_set_token_file_extent_other_encoding(leaf, fi, 0, &token); | |
4245 | btrfs_mark_buffer_dirty(leaf); | |
4246 | ||
4247 | btrfs_release_path(path); | |
4248 | ||
4249 | return ret; | |
4250 | } | |
4251 | ||
31d11b83 FM |
4252 | /* |
4253 | * Log all prealloc extents beyond the inode's i_size to make sure we do not | |
4254 | * lose them after doing a fast fsync and replaying the log. We scan the | |
4255 | * subvolume's root instead of iterating the inode's extent map tree because | |
4256 | * otherwise we can log incorrect extent items based on extent map conversion. | |
4257 | * That can happen due to the fact that extent maps are merged when they | |
4258 | * are not in the extent map tree's list of modified extents. | |
4259 | */ | |
4260 | static int btrfs_log_prealloc_extents(struct btrfs_trans_handle *trans, | |
4261 | struct btrfs_inode *inode, | |
4262 | struct btrfs_path *path) | |
4263 | { | |
4264 | struct btrfs_root *root = inode->root; | |
4265 | struct btrfs_key key; | |
4266 | const u64 i_size = i_size_read(&inode->vfs_inode); | |
4267 | const u64 ino = btrfs_ino(inode); | |
4268 | struct btrfs_path *dst_path = NULL; | |
4269 | u64 last_extent = (u64)-1; | |
4270 | int ins_nr = 0; | |
4271 | int start_slot; | |
4272 | int ret; | |
4273 | ||
4274 | if (!(inode->flags & BTRFS_INODE_PREALLOC)) | |
4275 | return 0; | |
4276 | ||
4277 | key.objectid = ino; | |
4278 | key.type = BTRFS_EXTENT_DATA_KEY; | |
4279 | key.offset = i_size; | |
4280 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
4281 | if (ret < 0) | |
4282 | goto out; | |
4283 | ||
4284 | while (true) { | |
4285 | struct extent_buffer *leaf = path->nodes[0]; | |
4286 | int slot = path->slots[0]; | |
4287 | ||
4288 | if (slot >= btrfs_header_nritems(leaf)) { | |
4289 | if (ins_nr > 0) { | |
4290 | ret = copy_items(trans, inode, dst_path, path, | |
4291 | &last_extent, start_slot, | |
4292 | ins_nr, 1, 0); | |
4293 | if (ret < 0) | |
4294 | goto out; | |
4295 | ins_nr = 0; | |
4296 | } | |
4297 | ret = btrfs_next_leaf(root, path); | |
4298 | if (ret < 0) | |
4299 | goto out; | |
4300 | if (ret > 0) { | |
4301 | ret = 0; | |
4302 | break; | |
4303 | } | |
4304 | continue; | |
4305 | } | |
4306 | ||
4307 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
4308 | if (key.objectid > ino) | |
4309 | break; | |
4310 | if (WARN_ON_ONCE(key.objectid < ino) || | |
4311 | key.type < BTRFS_EXTENT_DATA_KEY || | |
4312 | key.offset < i_size) { | |
4313 | path->slots[0]++; | |
4314 | continue; | |
4315 | } | |
4316 | if (last_extent == (u64)-1) { | |
4317 | last_extent = key.offset; | |
4318 | /* | |
4319 | * Avoid logging extent items logged in past fsync calls | |
4320 | * and leading to duplicate keys in the log tree. | |
4321 | */ | |
4322 | do { | |
4323 | ret = btrfs_truncate_inode_items(trans, | |
4324 | root->log_root, | |
4325 | &inode->vfs_inode, | |
4326 | i_size, | |
4327 | BTRFS_EXTENT_DATA_KEY); | |
4328 | } while (ret == -EAGAIN); | |
4329 | if (ret) | |
4330 | goto out; | |
4331 | } | |
4332 | if (ins_nr == 0) | |
4333 | start_slot = slot; | |
4334 | ins_nr++; | |
4335 | path->slots[0]++; | |
4336 | if (!dst_path) { | |
4337 | dst_path = btrfs_alloc_path(); | |
4338 | if (!dst_path) { | |
4339 | ret = -ENOMEM; | |
4340 | goto out; | |
4341 | } | |
4342 | } | |
4343 | } | |
4344 | if (ins_nr > 0) { | |
4345 | ret = copy_items(trans, inode, dst_path, path, &last_extent, | |
4346 | start_slot, ins_nr, 1, 0); | |
4347 | if (ret > 0) | |
4348 | ret = 0; | |
4349 | } | |
4350 | out: | |
4351 | btrfs_release_path(path); | |
4352 | btrfs_free_path(dst_path); | |
4353 | return ret; | |
4354 | } | |
4355 | ||
5dc562c5 JB |
4356 | static int btrfs_log_changed_extents(struct btrfs_trans_handle *trans, |
4357 | struct btrfs_root *root, | |
9d122629 | 4358 | struct btrfs_inode *inode, |
827463c4 | 4359 | struct btrfs_path *path, |
de0ee0ed FM |
4360 | struct btrfs_log_ctx *ctx, |
4361 | const u64 start, | |
4362 | const u64 end) | |
5dc562c5 | 4363 | { |
5dc562c5 JB |
4364 | struct extent_map *em, *n; |
4365 | struct list_head extents; | |
9d122629 | 4366 | struct extent_map_tree *tree = &inode->extent_tree; |
8c6c5928 | 4367 | u64 logged_start, logged_end; |
5dc562c5 JB |
4368 | u64 test_gen; |
4369 | int ret = 0; | |
2ab28f32 | 4370 | int num = 0; |
5dc562c5 JB |
4371 | |
4372 | INIT_LIST_HEAD(&extents); | |
4373 | ||
9d122629 | 4374 | down_write(&inode->dio_sem); |
5dc562c5 JB |
4375 | write_lock(&tree->lock); |
4376 | test_gen = root->fs_info->last_trans_committed; | |
8c6c5928 JB |
4377 | logged_start = start; |
4378 | logged_end = end; | |
5dc562c5 JB |
4379 | |
4380 | list_for_each_entry_safe(em, n, &tree->modified_extents, list) { | |
4381 | list_del_init(&em->list); | |
2ab28f32 JB |
4382 | /* |
4383 | * Just an arbitrary number, this can be really CPU intensive | |
4384 | * once we start getting a lot of extents, and really once we | |
4385 | * have a bunch of extents we just want to commit since it will | |
4386 | * be faster. | |
4387 | */ | |
4388 | if (++num > 32768) { | |
4389 | list_del_init(&tree->modified_extents); | |
4390 | ret = -EFBIG; | |
4391 | goto process; | |
4392 | } | |
4393 | ||
5dc562c5 JB |
4394 | if (em->generation <= test_gen) |
4395 | continue; | |
8c6c5928 | 4396 | |
31d11b83 FM |
4397 | /* We log prealloc extents beyond eof later. */ |
4398 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) && | |
4399 | em->start >= i_size_read(&inode->vfs_inode)) | |
4400 | continue; | |
4401 | ||
8c6c5928 JB |
4402 | if (em->start < logged_start) |
4403 | logged_start = em->start; | |
4404 | if ((em->start + em->len - 1) > logged_end) | |
4405 | logged_end = em->start + em->len - 1; | |
4406 | ||
ff44c6e3 | 4407 | /* Need a ref to keep it from getting evicted from cache */ |
490b54d6 | 4408 | refcount_inc(&em->refs); |
ff44c6e3 | 4409 | set_bit(EXTENT_FLAG_LOGGING, &em->flags); |
5dc562c5 | 4410 | list_add_tail(&em->list, &extents); |
2ab28f32 | 4411 | num++; |
5dc562c5 JB |
4412 | } |
4413 | ||
4414 | list_sort(NULL, &extents, extent_cmp); | |
2ab28f32 | 4415 | process: |
5dc562c5 JB |
4416 | while (!list_empty(&extents)) { |
4417 | em = list_entry(extents.next, struct extent_map, list); | |
4418 | ||
4419 | list_del_init(&em->list); | |
4420 | ||
4421 | /* | |
4422 | * If we had an error we just need to delete everybody from our | |
4423 | * private list. | |
4424 | */ | |
ff44c6e3 | 4425 | if (ret) { |
201a9038 | 4426 | clear_em_logging(tree, em); |
ff44c6e3 | 4427 | free_extent_map(em); |
5dc562c5 | 4428 | continue; |
ff44c6e3 JB |
4429 | } |
4430 | ||
4431 | write_unlock(&tree->lock); | |
5dc562c5 | 4432 | |
a2120a47 | 4433 | ret = log_one_extent(trans, inode, root, em, path, ctx); |
ff44c6e3 | 4434 | write_lock(&tree->lock); |
201a9038 JB |
4435 | clear_em_logging(tree, em); |
4436 | free_extent_map(em); | |
5dc562c5 | 4437 | } |
ff44c6e3 JB |
4438 | WARN_ON(!list_empty(&extents)); |
4439 | write_unlock(&tree->lock); | |
9d122629 | 4440 | up_write(&inode->dio_sem); |
5dc562c5 | 4441 | |
5dc562c5 | 4442 | btrfs_release_path(path); |
31d11b83 FM |
4443 | if (!ret) |
4444 | ret = btrfs_log_prealloc_extents(trans, inode, path); | |
4445 | ||
5dc562c5 JB |
4446 | return ret; |
4447 | } | |
4448 | ||
481b01c0 | 4449 | static int logged_inode_size(struct btrfs_root *log, struct btrfs_inode *inode, |
1a4bcf47 FM |
4450 | struct btrfs_path *path, u64 *size_ret) |
4451 | { | |
4452 | struct btrfs_key key; | |
4453 | int ret; | |
4454 | ||
481b01c0 | 4455 | key.objectid = btrfs_ino(inode); |
1a4bcf47 FM |
4456 | key.type = BTRFS_INODE_ITEM_KEY; |
4457 | key.offset = 0; | |
4458 | ||
4459 | ret = btrfs_search_slot(NULL, log, &key, path, 0, 0); | |
4460 | if (ret < 0) { | |
4461 | return ret; | |
4462 | } else if (ret > 0) { | |
2f2ff0ee | 4463 | *size_ret = 0; |
1a4bcf47 FM |
4464 | } else { |
4465 | struct btrfs_inode_item *item; | |
4466 | ||
4467 | item = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
4468 | struct btrfs_inode_item); | |
4469 | *size_ret = btrfs_inode_size(path->nodes[0], item); | |
4470 | } | |
4471 | ||
4472 | btrfs_release_path(path); | |
4473 | return 0; | |
4474 | } | |
4475 | ||
36283bf7 FM |
4476 | /* |
4477 | * At the moment we always log all xattrs. This is to figure out at log replay | |
4478 | * time which xattrs must have their deletion replayed. If a xattr is missing | |
4479 | * in the log tree and exists in the fs/subvol tree, we delete it. This is | |
4480 | * because if a xattr is deleted, the inode is fsynced and a power failure | |
4481 | * happens, causing the log to be replayed the next time the fs is mounted, | |
4482 | * we want the xattr to not exist anymore (same behaviour as other filesystems | |
4483 | * with a journal, ext3/4, xfs, f2fs, etc). | |
4484 | */ | |
4485 | static int btrfs_log_all_xattrs(struct btrfs_trans_handle *trans, | |
4486 | struct btrfs_root *root, | |
1a93c36a | 4487 | struct btrfs_inode *inode, |
36283bf7 FM |
4488 | struct btrfs_path *path, |
4489 | struct btrfs_path *dst_path) | |
4490 | { | |
4491 | int ret; | |
4492 | struct btrfs_key key; | |
1a93c36a | 4493 | const u64 ino = btrfs_ino(inode); |
36283bf7 FM |
4494 | int ins_nr = 0; |
4495 | int start_slot = 0; | |
4496 | ||
4497 | key.objectid = ino; | |
4498 | key.type = BTRFS_XATTR_ITEM_KEY; | |
4499 | key.offset = 0; | |
4500 | ||
4501 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
4502 | if (ret < 0) | |
4503 | return ret; | |
4504 | ||
4505 | while (true) { | |
4506 | int slot = path->slots[0]; | |
4507 | struct extent_buffer *leaf = path->nodes[0]; | |
4508 | int nritems = btrfs_header_nritems(leaf); | |
4509 | ||
4510 | if (slot >= nritems) { | |
4511 | if (ins_nr > 0) { | |
4512 | u64 last_extent = 0; | |
4513 | ||
1a93c36a | 4514 | ret = copy_items(trans, inode, dst_path, path, |
36283bf7 FM |
4515 | &last_extent, start_slot, |
4516 | ins_nr, 1, 0); | |
4517 | /* can't be 1, extent items aren't processed */ | |
4518 | ASSERT(ret <= 0); | |
4519 | if (ret < 0) | |
4520 | return ret; | |
4521 | ins_nr = 0; | |
4522 | } | |
4523 | ret = btrfs_next_leaf(root, path); | |
4524 | if (ret < 0) | |
4525 | return ret; | |
4526 | else if (ret > 0) | |
4527 | break; | |
4528 | continue; | |
4529 | } | |
4530 | ||
4531 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
4532 | if (key.objectid != ino || key.type != BTRFS_XATTR_ITEM_KEY) | |
4533 | break; | |
4534 | ||
4535 | if (ins_nr == 0) | |
4536 | start_slot = slot; | |
4537 | ins_nr++; | |
4538 | path->slots[0]++; | |
4539 | cond_resched(); | |
4540 | } | |
4541 | if (ins_nr > 0) { | |
4542 | u64 last_extent = 0; | |
4543 | ||
1a93c36a | 4544 | ret = copy_items(trans, inode, dst_path, path, |
36283bf7 FM |
4545 | &last_extent, start_slot, |
4546 | ins_nr, 1, 0); | |
4547 | /* can't be 1, extent items aren't processed */ | |
4548 | ASSERT(ret <= 0); | |
4549 | if (ret < 0) | |
4550 | return ret; | |
4551 | } | |
4552 | ||
4553 | return 0; | |
4554 | } | |
4555 | ||
a89ca6f2 FM |
4556 | /* |
4557 | * If the no holes feature is enabled we need to make sure any hole between the | |
4558 | * last extent and the i_size of our inode is explicitly marked in the log. This | |
4559 | * is to make sure that doing something like: | |
4560 | * | |
4561 | * 1) create file with 128Kb of data | |
4562 | * 2) truncate file to 64Kb | |
4563 | * 3) truncate file to 256Kb | |
4564 | * 4) fsync file | |
4565 | * 5) <crash/power failure> | |
4566 | * 6) mount fs and trigger log replay | |
4567 | * | |
4568 | * Will give us a file with a size of 256Kb, the first 64Kb of data match what | |
4569 | * the file had in its first 64Kb of data at step 1 and the last 192Kb of the | |
4570 | * file correspond to a hole. The presence of explicit holes in a log tree is | |
4571 | * what guarantees that log replay will remove/adjust file extent items in the | |
4572 | * fs/subvol tree. | |
4573 | * | |
4574 | * Here we do not need to care about holes between extents, that is already done | |
4575 | * by copy_items(). We also only need to do this in the full sync path, where we | |
4576 | * lookup for extents from the fs/subvol tree only. In the fast path case, we | |
4577 | * lookup the list of modified extent maps and if any represents a hole, we | |
4578 | * insert a corresponding extent representing a hole in the log tree. | |
4579 | */ | |
4580 | static int btrfs_log_trailing_hole(struct btrfs_trans_handle *trans, | |
4581 | struct btrfs_root *root, | |
a0308dd7 | 4582 | struct btrfs_inode *inode, |
a89ca6f2 FM |
4583 | struct btrfs_path *path) |
4584 | { | |
0b246afa | 4585 | struct btrfs_fs_info *fs_info = root->fs_info; |
a89ca6f2 FM |
4586 | int ret; |
4587 | struct btrfs_key key; | |
4588 | u64 hole_start; | |
4589 | u64 hole_size; | |
4590 | struct extent_buffer *leaf; | |
4591 | struct btrfs_root *log = root->log_root; | |
a0308dd7 NB |
4592 | const u64 ino = btrfs_ino(inode); |
4593 | const u64 i_size = i_size_read(&inode->vfs_inode); | |
a89ca6f2 | 4594 | |
0b246afa | 4595 | if (!btrfs_fs_incompat(fs_info, NO_HOLES)) |
a89ca6f2 FM |
4596 | return 0; |
4597 | ||
4598 | key.objectid = ino; | |
4599 | key.type = BTRFS_EXTENT_DATA_KEY; | |
4600 | key.offset = (u64)-1; | |
4601 | ||
4602 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
4603 | ASSERT(ret != 0); | |
4604 | if (ret < 0) | |
4605 | return ret; | |
4606 | ||
4607 | ASSERT(path->slots[0] > 0); | |
4608 | path->slots[0]--; | |
4609 | leaf = path->nodes[0]; | |
4610 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
4611 | ||
4612 | if (key.objectid != ino || key.type != BTRFS_EXTENT_DATA_KEY) { | |
4613 | /* inode does not have any extents */ | |
4614 | hole_start = 0; | |
4615 | hole_size = i_size; | |
4616 | } else { | |
4617 | struct btrfs_file_extent_item *extent; | |
4618 | u64 len; | |
4619 | ||
4620 | /* | |
4621 | * If there's an extent beyond i_size, an explicit hole was | |
4622 | * already inserted by copy_items(). | |
4623 | */ | |
4624 | if (key.offset >= i_size) | |
4625 | return 0; | |
4626 | ||
4627 | extent = btrfs_item_ptr(leaf, path->slots[0], | |
4628 | struct btrfs_file_extent_item); | |
4629 | ||
4630 | if (btrfs_file_extent_type(leaf, extent) == | |
4631 | BTRFS_FILE_EXTENT_INLINE) { | |
e41ca589 | 4632 | len = btrfs_file_extent_ram_bytes(leaf, extent); |
6399fb5a FM |
4633 | ASSERT(len == i_size || |
4634 | (len == fs_info->sectorsize && | |
4635 | btrfs_file_extent_compression(leaf, extent) != | |
4636 | BTRFS_COMPRESS_NONE)); | |
a89ca6f2 FM |
4637 | return 0; |
4638 | } | |
4639 | ||
4640 | len = btrfs_file_extent_num_bytes(leaf, extent); | |
4641 | /* Last extent goes beyond i_size, no need to log a hole. */ | |
4642 | if (key.offset + len > i_size) | |
4643 | return 0; | |
4644 | hole_start = key.offset + len; | |
4645 | hole_size = i_size - hole_start; | |
4646 | } | |
4647 | btrfs_release_path(path); | |
4648 | ||
4649 | /* Last extent ends at i_size. */ | |
4650 | if (hole_size == 0) | |
4651 | return 0; | |
4652 | ||
0b246afa | 4653 | hole_size = ALIGN(hole_size, fs_info->sectorsize); |
a89ca6f2 FM |
4654 | ret = btrfs_insert_file_extent(trans, log, ino, hole_start, 0, 0, |
4655 | hole_size, 0, hole_size, 0, 0, 0); | |
4656 | return ret; | |
4657 | } | |
4658 | ||
56f23fdb FM |
4659 | /* |
4660 | * When we are logging a new inode X, check if it doesn't have a reference that | |
4661 | * matches the reference from some other inode Y created in a past transaction | |
4662 | * and that was renamed in the current transaction. If we don't do this, then at | |
4663 | * log replay time we can lose inode Y (and all its files if it's a directory): | |
4664 | * | |
4665 | * mkdir /mnt/x | |
4666 | * echo "hello world" > /mnt/x/foobar | |
4667 | * sync | |
4668 | * mv /mnt/x /mnt/y | |
4669 | * mkdir /mnt/x # or touch /mnt/x | |
4670 | * xfs_io -c fsync /mnt/x | |
4671 | * <power fail> | |
4672 | * mount fs, trigger log replay | |
4673 | * | |
4674 | * After the log replay procedure, we would lose the first directory and all its | |
4675 | * files (file foobar). | |
4676 | * For the case where inode Y is not a directory we simply end up losing it: | |
4677 | * | |
4678 | * echo "123" > /mnt/foo | |
4679 | * sync | |
4680 | * mv /mnt/foo /mnt/bar | |
4681 | * echo "abc" > /mnt/foo | |
4682 | * xfs_io -c fsync /mnt/foo | |
4683 | * <power fail> | |
4684 | * | |
4685 | * We also need this for cases where a snapshot entry is replaced by some other | |
4686 | * entry (file or directory) otherwise we end up with an unreplayable log due to | |
4687 | * attempts to delete the snapshot entry (entry of type BTRFS_ROOT_ITEM_KEY) as | |
4688 | * if it were a regular entry: | |
4689 | * | |
4690 | * mkdir /mnt/x | |
4691 | * btrfs subvolume snapshot /mnt /mnt/x/snap | |
4692 | * btrfs subvolume delete /mnt/x/snap | |
4693 | * rmdir /mnt/x | |
4694 | * mkdir /mnt/x | |
4695 | * fsync /mnt/x or fsync some new file inside it | |
4696 | * <power fail> | |
4697 | * | |
4698 | * The snapshot delete, rmdir of x, mkdir of a new x and the fsync all happen in | |
4699 | * the same transaction. | |
4700 | */ | |
4701 | static int btrfs_check_ref_name_override(struct extent_buffer *eb, | |
4702 | const int slot, | |
4703 | const struct btrfs_key *key, | |
4791c8f1 | 4704 | struct btrfs_inode *inode, |
44f714da | 4705 | u64 *other_ino) |
56f23fdb FM |
4706 | { |
4707 | int ret; | |
4708 | struct btrfs_path *search_path; | |
4709 | char *name = NULL; | |
4710 | u32 name_len = 0; | |
4711 | u32 item_size = btrfs_item_size_nr(eb, slot); | |
4712 | u32 cur_offset = 0; | |
4713 | unsigned long ptr = btrfs_item_ptr_offset(eb, slot); | |
4714 | ||
4715 | search_path = btrfs_alloc_path(); | |
4716 | if (!search_path) | |
4717 | return -ENOMEM; | |
4718 | search_path->search_commit_root = 1; | |
4719 | search_path->skip_locking = 1; | |
4720 | ||
4721 | while (cur_offset < item_size) { | |
4722 | u64 parent; | |
4723 | u32 this_name_len; | |
4724 | u32 this_len; | |
4725 | unsigned long name_ptr; | |
4726 | struct btrfs_dir_item *di; | |
4727 | ||
4728 | if (key->type == BTRFS_INODE_REF_KEY) { | |
4729 | struct btrfs_inode_ref *iref; | |
4730 | ||
4731 | iref = (struct btrfs_inode_ref *)(ptr + cur_offset); | |
4732 | parent = key->offset; | |
4733 | this_name_len = btrfs_inode_ref_name_len(eb, iref); | |
4734 | name_ptr = (unsigned long)(iref + 1); | |
4735 | this_len = sizeof(*iref) + this_name_len; | |
4736 | } else { | |
4737 | struct btrfs_inode_extref *extref; | |
4738 | ||
4739 | extref = (struct btrfs_inode_extref *)(ptr + | |
4740 | cur_offset); | |
4741 | parent = btrfs_inode_extref_parent(eb, extref); | |
4742 | this_name_len = btrfs_inode_extref_name_len(eb, extref); | |
4743 | name_ptr = (unsigned long)&extref->name; | |
4744 | this_len = sizeof(*extref) + this_name_len; | |
4745 | } | |
4746 | ||
4747 | if (this_name_len > name_len) { | |
4748 | char *new_name; | |
4749 | ||
4750 | new_name = krealloc(name, this_name_len, GFP_NOFS); | |
4751 | if (!new_name) { | |
4752 | ret = -ENOMEM; | |
4753 | goto out; | |
4754 | } | |
4755 | name_len = this_name_len; | |
4756 | name = new_name; | |
4757 | } | |
4758 | ||
4759 | read_extent_buffer(eb, name, name_ptr, this_name_len); | |
4791c8f1 NB |
4760 | di = btrfs_lookup_dir_item(NULL, inode->root, search_path, |
4761 | parent, name, this_name_len, 0); | |
56f23fdb | 4762 | if (di && !IS_ERR(di)) { |
44f714da FM |
4763 | struct btrfs_key di_key; |
4764 | ||
4765 | btrfs_dir_item_key_to_cpu(search_path->nodes[0], | |
4766 | di, &di_key); | |
4767 | if (di_key.type == BTRFS_INODE_ITEM_KEY) { | |
4768 | ret = 1; | |
4769 | *other_ino = di_key.objectid; | |
4770 | } else { | |
4771 | ret = -EAGAIN; | |
4772 | } | |
56f23fdb FM |
4773 | goto out; |
4774 | } else if (IS_ERR(di)) { | |
4775 | ret = PTR_ERR(di); | |
4776 | goto out; | |
4777 | } | |
4778 | btrfs_release_path(search_path); | |
4779 | ||
4780 | cur_offset += this_len; | |
4781 | } | |
4782 | ret = 0; | |
4783 | out: | |
4784 | btrfs_free_path(search_path); | |
4785 | kfree(name); | |
4786 | return ret; | |
4787 | } | |
4788 | ||
e02119d5 CM |
4789 | /* log a single inode in the tree log. |
4790 | * At least one parent directory for this inode must exist in the tree | |
4791 | * or be logged already. | |
4792 | * | |
4793 | * Any items from this inode changed by the current transaction are copied | |
4794 | * to the log tree. An extra reference is taken on any extents in this | |
4795 | * file, allowing us to avoid a whole pile of corner cases around logging | |
4796 | * blocks that have been removed from the tree. | |
4797 | * | |
4798 | * See LOG_INODE_ALL and related defines for a description of what inode_only | |
4799 | * does. | |
4800 | * | |
4801 | * This handles both files and directories. | |
4802 | */ | |
12fcfd22 | 4803 | static int btrfs_log_inode(struct btrfs_trans_handle *trans, |
a59108a7 | 4804 | struct btrfs_root *root, struct btrfs_inode *inode, |
49dae1bc FM |
4805 | int inode_only, |
4806 | const loff_t start, | |
8407f553 FM |
4807 | const loff_t end, |
4808 | struct btrfs_log_ctx *ctx) | |
e02119d5 | 4809 | { |
0b246afa | 4810 | struct btrfs_fs_info *fs_info = root->fs_info; |
e02119d5 CM |
4811 | struct btrfs_path *path; |
4812 | struct btrfs_path *dst_path; | |
4813 | struct btrfs_key min_key; | |
4814 | struct btrfs_key max_key; | |
4815 | struct btrfs_root *log = root->log_root; | |
16e7549f | 4816 | u64 last_extent = 0; |
4a500fd1 | 4817 | int err = 0; |
e02119d5 | 4818 | int ret; |
3a5f1d45 | 4819 | int nritems; |
31ff1cd2 CM |
4820 | int ins_start_slot = 0; |
4821 | int ins_nr; | |
5dc562c5 | 4822 | bool fast_search = false; |
a59108a7 NB |
4823 | u64 ino = btrfs_ino(inode); |
4824 | struct extent_map_tree *em_tree = &inode->extent_tree; | |
1a4bcf47 | 4825 | u64 logged_isize = 0; |
e4545de5 | 4826 | bool need_log_inode_item = true; |
9a8fca62 | 4827 | bool xattrs_logged = false; |
e02119d5 | 4828 | |
e02119d5 | 4829 | path = btrfs_alloc_path(); |
5df67083 TI |
4830 | if (!path) |
4831 | return -ENOMEM; | |
e02119d5 | 4832 | dst_path = btrfs_alloc_path(); |
5df67083 TI |
4833 | if (!dst_path) { |
4834 | btrfs_free_path(path); | |
4835 | return -ENOMEM; | |
4836 | } | |
e02119d5 | 4837 | |
33345d01 | 4838 | min_key.objectid = ino; |
e02119d5 CM |
4839 | min_key.type = BTRFS_INODE_ITEM_KEY; |
4840 | min_key.offset = 0; | |
4841 | ||
33345d01 | 4842 | max_key.objectid = ino; |
12fcfd22 | 4843 | |
12fcfd22 | 4844 | |
5dc562c5 | 4845 | /* today the code can only do partial logging of directories */ |
a59108a7 | 4846 | if (S_ISDIR(inode->vfs_inode.i_mode) || |
5269b67e | 4847 | (!test_bit(BTRFS_INODE_NEEDS_FULL_SYNC, |
a59108a7 | 4848 | &inode->runtime_flags) && |
781feef7 | 4849 | inode_only >= LOG_INODE_EXISTS)) |
e02119d5 CM |
4850 | max_key.type = BTRFS_XATTR_ITEM_KEY; |
4851 | else | |
4852 | max_key.type = (u8)-1; | |
4853 | max_key.offset = (u64)-1; | |
4854 | ||
2c2c452b FM |
4855 | /* |
4856 | * Only run delayed items if we are a dir or a new file. | |
4857 | * Otherwise commit the delayed inode only, which is needed in | |
4858 | * order for the log replay code to mark inodes for link count | |
4859 | * fixup (create temporary BTRFS_TREE_LOG_FIXUP_OBJECTID items). | |
4860 | */ | |
a59108a7 NB |
4861 | if (S_ISDIR(inode->vfs_inode.i_mode) || |
4862 | inode->generation > fs_info->last_trans_committed) | |
4863 | ret = btrfs_commit_inode_delayed_items(trans, inode); | |
2c2c452b | 4864 | else |
a59108a7 | 4865 | ret = btrfs_commit_inode_delayed_inode(inode); |
2c2c452b FM |
4866 | |
4867 | if (ret) { | |
4868 | btrfs_free_path(path); | |
4869 | btrfs_free_path(dst_path); | |
4870 | return ret; | |
16cdcec7 MX |
4871 | } |
4872 | ||
781feef7 LB |
4873 | if (inode_only == LOG_OTHER_INODE) { |
4874 | inode_only = LOG_INODE_EXISTS; | |
a59108a7 | 4875 | mutex_lock_nested(&inode->log_mutex, SINGLE_DEPTH_NESTING); |
781feef7 | 4876 | } else { |
a59108a7 | 4877 | mutex_lock(&inode->log_mutex); |
781feef7 | 4878 | } |
e02119d5 CM |
4879 | |
4880 | /* | |
4881 | * a brute force approach to making sure we get the most uptodate | |
4882 | * copies of everything. | |
4883 | */ | |
a59108a7 | 4884 | if (S_ISDIR(inode->vfs_inode.i_mode)) { |
e02119d5 CM |
4885 | int max_key_type = BTRFS_DIR_LOG_INDEX_KEY; |
4886 | ||
4f764e51 FM |
4887 | if (inode_only == LOG_INODE_EXISTS) |
4888 | max_key_type = BTRFS_XATTR_ITEM_KEY; | |
33345d01 | 4889 | ret = drop_objectid_items(trans, log, path, ino, max_key_type); |
e02119d5 | 4890 | } else { |
1a4bcf47 FM |
4891 | if (inode_only == LOG_INODE_EXISTS) { |
4892 | /* | |
4893 | * Make sure the new inode item we write to the log has | |
4894 | * the same isize as the current one (if it exists). | |
4895 | * This is necessary to prevent data loss after log | |
4896 | * replay, and also to prevent doing a wrong expanding | |
4897 | * truncate - for e.g. create file, write 4K into offset | |
4898 | * 0, fsync, write 4K into offset 4096, add hard link, | |
4899 | * fsync some other file (to sync log), power fail - if | |
4900 | * we use the inode's current i_size, after log replay | |
4901 | * we get a 8Kb file, with the last 4Kb extent as a hole | |
4902 | * (zeroes), as if an expanding truncate happened, | |
4903 | * instead of getting a file of 4Kb only. | |
4904 | */ | |
a59108a7 | 4905 | err = logged_inode_size(log, inode, path, &logged_isize); |
1a4bcf47 FM |
4906 | if (err) |
4907 | goto out_unlock; | |
4908 | } | |
a742994a | 4909 | if (test_bit(BTRFS_INODE_NEEDS_FULL_SYNC, |
a59108a7 | 4910 | &inode->runtime_flags)) { |
a742994a | 4911 | if (inode_only == LOG_INODE_EXISTS) { |
4f764e51 | 4912 | max_key.type = BTRFS_XATTR_ITEM_KEY; |
a742994a FM |
4913 | ret = drop_objectid_items(trans, log, path, ino, |
4914 | max_key.type); | |
4915 | } else { | |
4916 | clear_bit(BTRFS_INODE_NEEDS_FULL_SYNC, | |
a59108a7 | 4917 | &inode->runtime_flags); |
a742994a | 4918 | clear_bit(BTRFS_INODE_COPY_EVERYTHING, |
a59108a7 | 4919 | &inode->runtime_flags); |
28ed1345 CM |
4920 | while(1) { |
4921 | ret = btrfs_truncate_inode_items(trans, | |
a59108a7 | 4922 | log, &inode->vfs_inode, 0, 0); |
28ed1345 CM |
4923 | if (ret != -EAGAIN) |
4924 | break; | |
4925 | } | |
a742994a | 4926 | } |
4f764e51 | 4927 | } else if (test_and_clear_bit(BTRFS_INODE_COPY_EVERYTHING, |
a59108a7 | 4928 | &inode->runtime_flags) || |
6cfab851 | 4929 | inode_only == LOG_INODE_EXISTS) { |
4f764e51 | 4930 | if (inode_only == LOG_INODE_ALL) |
183f37fa | 4931 | fast_search = true; |
4f764e51 | 4932 | max_key.type = BTRFS_XATTR_ITEM_KEY; |
5dc562c5 | 4933 | ret = drop_objectid_items(trans, log, path, ino, |
e9976151 | 4934 | max_key.type); |
a95249b3 JB |
4935 | } else { |
4936 | if (inode_only == LOG_INODE_ALL) | |
4937 | fast_search = true; | |
a95249b3 | 4938 | goto log_extents; |
5dc562c5 | 4939 | } |
a95249b3 | 4940 | |
e02119d5 | 4941 | } |
4a500fd1 YZ |
4942 | if (ret) { |
4943 | err = ret; | |
4944 | goto out_unlock; | |
4945 | } | |
e02119d5 | 4946 | |
d397712b | 4947 | while (1) { |
31ff1cd2 | 4948 | ins_nr = 0; |
6174d3cb | 4949 | ret = btrfs_search_forward(root, &min_key, |
de78b51a | 4950 | path, trans->transid); |
fb770ae4 LB |
4951 | if (ret < 0) { |
4952 | err = ret; | |
4953 | goto out_unlock; | |
4954 | } | |
e02119d5 CM |
4955 | if (ret != 0) |
4956 | break; | |
3a5f1d45 | 4957 | again: |
31ff1cd2 | 4958 | /* note, ins_nr might be > 0 here, cleanup outside the loop */ |
33345d01 | 4959 | if (min_key.objectid != ino) |
e02119d5 CM |
4960 | break; |
4961 | if (min_key.type > max_key.type) | |
4962 | break; | |
31ff1cd2 | 4963 | |
e4545de5 FM |
4964 | if (min_key.type == BTRFS_INODE_ITEM_KEY) |
4965 | need_log_inode_item = false; | |
4966 | ||
56f23fdb FM |
4967 | if ((min_key.type == BTRFS_INODE_REF_KEY || |
4968 | min_key.type == BTRFS_INODE_EXTREF_KEY) && | |
a59108a7 | 4969 | inode->generation == trans->transid) { |
44f714da FM |
4970 | u64 other_ino = 0; |
4971 | ||
56f23fdb | 4972 | ret = btrfs_check_ref_name_override(path->nodes[0], |
a59108a7 NB |
4973 | path->slots[0], &min_key, inode, |
4974 | &other_ino); | |
56f23fdb FM |
4975 | if (ret < 0) { |
4976 | err = ret; | |
4977 | goto out_unlock; | |
28a23593 | 4978 | } else if (ret > 0 && ctx && |
4a0cc7ca | 4979 | other_ino != btrfs_ino(BTRFS_I(ctx->inode))) { |
44f714da FM |
4980 | struct btrfs_key inode_key; |
4981 | struct inode *other_inode; | |
4982 | ||
4983 | if (ins_nr > 0) { | |
4984 | ins_nr++; | |
4985 | } else { | |
4986 | ins_nr = 1; | |
4987 | ins_start_slot = path->slots[0]; | |
4988 | } | |
a59108a7 | 4989 | ret = copy_items(trans, inode, dst_path, path, |
44f714da FM |
4990 | &last_extent, ins_start_slot, |
4991 | ins_nr, inode_only, | |
4992 | logged_isize); | |
4993 | if (ret < 0) { | |
4994 | err = ret; | |
4995 | goto out_unlock; | |
4996 | } | |
4997 | ins_nr = 0; | |
4998 | btrfs_release_path(path); | |
4999 | inode_key.objectid = other_ino; | |
5000 | inode_key.type = BTRFS_INODE_ITEM_KEY; | |
5001 | inode_key.offset = 0; | |
0b246afa | 5002 | other_inode = btrfs_iget(fs_info->sb, |
44f714da FM |
5003 | &inode_key, root, |
5004 | NULL); | |
5005 | /* | |
5006 | * If the other inode that had a conflicting dir | |
5007 | * entry was deleted in the current transaction, | |
5008 | * we don't need to do more work nor fallback to | |
5009 | * a transaction commit. | |
5010 | */ | |
8d9e220c | 5011 | if (other_inode == ERR_PTR(-ENOENT)) { |
44f714da FM |
5012 | goto next_key; |
5013 | } else if (IS_ERR(other_inode)) { | |
5014 | err = PTR_ERR(other_inode); | |
5015 | goto out_unlock; | |
5016 | } | |
5017 | /* | |
5018 | * We are safe logging the other inode without | |
5019 | * acquiring its i_mutex as long as we log with | |
5020 | * the LOG_INODE_EXISTS mode. We're safe against | |
5021 | * concurrent renames of the other inode as well | |
5022 | * because during a rename we pin the log and | |
5023 | * update the log with the new name before we | |
5024 | * unpin it. | |
5025 | */ | |
a59108a7 NB |
5026 | err = btrfs_log_inode(trans, root, |
5027 | BTRFS_I(other_inode), | |
5028 | LOG_OTHER_INODE, 0, LLONG_MAX, | |
5029 | ctx); | |
44f714da FM |
5030 | iput(other_inode); |
5031 | if (err) | |
5032 | goto out_unlock; | |
5033 | else | |
5034 | goto next_key; | |
56f23fdb FM |
5035 | } |
5036 | } | |
5037 | ||
36283bf7 FM |
5038 | /* Skip xattrs, we log them later with btrfs_log_all_xattrs() */ |
5039 | if (min_key.type == BTRFS_XATTR_ITEM_KEY) { | |
5040 | if (ins_nr == 0) | |
5041 | goto next_slot; | |
a59108a7 | 5042 | ret = copy_items(trans, inode, dst_path, path, |
36283bf7 FM |
5043 | &last_extent, ins_start_slot, |
5044 | ins_nr, inode_only, logged_isize); | |
5045 | if (ret < 0) { | |
5046 | err = ret; | |
5047 | goto out_unlock; | |
5048 | } | |
5049 | ins_nr = 0; | |
5050 | if (ret) { | |
5051 | btrfs_release_path(path); | |
5052 | continue; | |
5053 | } | |
5054 | goto next_slot; | |
5055 | } | |
5056 | ||
31ff1cd2 CM |
5057 | if (ins_nr && ins_start_slot + ins_nr == path->slots[0]) { |
5058 | ins_nr++; | |
5059 | goto next_slot; | |
5060 | } else if (!ins_nr) { | |
5061 | ins_start_slot = path->slots[0]; | |
5062 | ins_nr = 1; | |
5063 | goto next_slot; | |
e02119d5 CM |
5064 | } |
5065 | ||
a59108a7 | 5066 | ret = copy_items(trans, inode, dst_path, path, &last_extent, |
1a4bcf47 FM |
5067 | ins_start_slot, ins_nr, inode_only, |
5068 | logged_isize); | |
16e7549f | 5069 | if (ret < 0) { |
4a500fd1 YZ |
5070 | err = ret; |
5071 | goto out_unlock; | |
a71db86e RV |
5072 | } |
5073 | if (ret) { | |
16e7549f JB |
5074 | ins_nr = 0; |
5075 | btrfs_release_path(path); | |
5076 | continue; | |
4a500fd1 | 5077 | } |
31ff1cd2 CM |
5078 | ins_nr = 1; |
5079 | ins_start_slot = path->slots[0]; | |
5080 | next_slot: | |
e02119d5 | 5081 | |
3a5f1d45 CM |
5082 | nritems = btrfs_header_nritems(path->nodes[0]); |
5083 | path->slots[0]++; | |
5084 | if (path->slots[0] < nritems) { | |
5085 | btrfs_item_key_to_cpu(path->nodes[0], &min_key, | |
5086 | path->slots[0]); | |
5087 | goto again; | |
5088 | } | |
31ff1cd2 | 5089 | if (ins_nr) { |
a59108a7 | 5090 | ret = copy_items(trans, inode, dst_path, path, |
16e7549f | 5091 | &last_extent, ins_start_slot, |
1a4bcf47 | 5092 | ins_nr, inode_only, logged_isize); |
16e7549f | 5093 | if (ret < 0) { |
4a500fd1 YZ |
5094 | err = ret; |
5095 | goto out_unlock; | |
5096 | } | |
16e7549f | 5097 | ret = 0; |
31ff1cd2 CM |
5098 | ins_nr = 0; |
5099 | } | |
b3b4aa74 | 5100 | btrfs_release_path(path); |
44f714da | 5101 | next_key: |
3d41d702 | 5102 | if (min_key.offset < (u64)-1) { |
e02119d5 | 5103 | min_key.offset++; |
3d41d702 | 5104 | } else if (min_key.type < max_key.type) { |
e02119d5 | 5105 | min_key.type++; |
3d41d702 FDBM |
5106 | min_key.offset = 0; |
5107 | } else { | |
e02119d5 | 5108 | break; |
3d41d702 | 5109 | } |
e02119d5 | 5110 | } |
31ff1cd2 | 5111 | if (ins_nr) { |
a59108a7 | 5112 | ret = copy_items(trans, inode, dst_path, path, &last_extent, |
1a4bcf47 FM |
5113 | ins_start_slot, ins_nr, inode_only, |
5114 | logged_isize); | |
16e7549f | 5115 | if (ret < 0) { |
4a500fd1 YZ |
5116 | err = ret; |
5117 | goto out_unlock; | |
5118 | } | |
16e7549f | 5119 | ret = 0; |
31ff1cd2 CM |
5120 | ins_nr = 0; |
5121 | } | |
5dc562c5 | 5122 | |
36283bf7 FM |
5123 | btrfs_release_path(path); |
5124 | btrfs_release_path(dst_path); | |
a59108a7 | 5125 | err = btrfs_log_all_xattrs(trans, root, inode, path, dst_path); |
36283bf7 FM |
5126 | if (err) |
5127 | goto out_unlock; | |
9a8fca62 | 5128 | xattrs_logged = true; |
a89ca6f2 FM |
5129 | if (max_key.type >= BTRFS_EXTENT_DATA_KEY && !fast_search) { |
5130 | btrfs_release_path(path); | |
5131 | btrfs_release_path(dst_path); | |
a59108a7 | 5132 | err = btrfs_log_trailing_hole(trans, root, inode, path); |
a89ca6f2 FM |
5133 | if (err) |
5134 | goto out_unlock; | |
5135 | } | |
a95249b3 | 5136 | log_extents: |
f3b15ccd JB |
5137 | btrfs_release_path(path); |
5138 | btrfs_release_path(dst_path); | |
e4545de5 | 5139 | if (need_log_inode_item) { |
a59108a7 | 5140 | err = log_inode_item(trans, log, dst_path, inode); |
9a8fca62 FM |
5141 | if (!err && !xattrs_logged) { |
5142 | err = btrfs_log_all_xattrs(trans, root, inode, path, | |
5143 | dst_path); | |
5144 | btrfs_release_path(path); | |
5145 | } | |
e4545de5 FM |
5146 | if (err) |
5147 | goto out_unlock; | |
5148 | } | |
5dc562c5 | 5149 | if (fast_search) { |
a59108a7 | 5150 | ret = btrfs_log_changed_extents(trans, root, inode, dst_path, |
a2120a47 | 5151 | ctx, start, end); |
5dc562c5 JB |
5152 | if (ret) { |
5153 | err = ret; | |
5154 | goto out_unlock; | |
5155 | } | |
d006a048 | 5156 | } else if (inode_only == LOG_INODE_ALL) { |
06d3d22b LB |
5157 | struct extent_map *em, *n; |
5158 | ||
49dae1bc FM |
5159 | write_lock(&em_tree->lock); |
5160 | /* | |
5161 | * We can't just remove every em if we're called for a ranged | |
5162 | * fsync - that is, one that doesn't cover the whole possible | |
5163 | * file range (0 to LLONG_MAX). This is because we can have | |
5164 | * em's that fall outside the range we're logging and therefore | |
5165 | * their ordered operations haven't completed yet | |
5166 | * (btrfs_finish_ordered_io() not invoked yet). This means we | |
5167 | * didn't get their respective file extent item in the fs/subvol | |
5168 | * tree yet, and need to let the next fast fsync (one which | |
5169 | * consults the list of modified extent maps) find the em so | |
5170 | * that it logs a matching file extent item and waits for the | |
5171 | * respective ordered operation to complete (if it's still | |
5172 | * running). | |
5173 | * | |
5174 | * Removing every em outside the range we're logging would make | |
5175 | * the next fast fsync not log their matching file extent items, | |
5176 | * therefore making us lose data after a log replay. | |
5177 | */ | |
5178 | list_for_each_entry_safe(em, n, &em_tree->modified_extents, | |
5179 | list) { | |
5180 | const u64 mod_end = em->mod_start + em->mod_len - 1; | |
5181 | ||
5182 | if (em->mod_start >= start && mod_end <= end) | |
5183 | list_del_init(&em->list); | |
5184 | } | |
5185 | write_unlock(&em_tree->lock); | |
5dc562c5 JB |
5186 | } |
5187 | ||
a59108a7 NB |
5188 | if (inode_only == LOG_INODE_ALL && S_ISDIR(inode->vfs_inode.i_mode)) { |
5189 | ret = log_directory_changes(trans, root, inode, path, dst_path, | |
5190 | ctx); | |
4a500fd1 YZ |
5191 | if (ret) { |
5192 | err = ret; | |
5193 | goto out_unlock; | |
5194 | } | |
e02119d5 | 5195 | } |
49dae1bc | 5196 | |
a59108a7 NB |
5197 | spin_lock(&inode->lock); |
5198 | inode->logged_trans = trans->transid; | |
5199 | inode->last_log_commit = inode->last_sub_trans; | |
5200 | spin_unlock(&inode->lock); | |
4a500fd1 | 5201 | out_unlock: |
a59108a7 | 5202 | mutex_unlock(&inode->log_mutex); |
e02119d5 CM |
5203 | |
5204 | btrfs_free_path(path); | |
5205 | btrfs_free_path(dst_path); | |
4a500fd1 | 5206 | return err; |
e02119d5 CM |
5207 | } |
5208 | ||
2be63d5c FM |
5209 | /* |
5210 | * Check if we must fallback to a transaction commit when logging an inode. | |
5211 | * This must be called after logging the inode and is used only in the context | |
5212 | * when fsyncing an inode requires the need to log some other inode - in which | |
5213 | * case we can't lock the i_mutex of each other inode we need to log as that | |
5214 | * can lead to deadlocks with concurrent fsync against other inodes (as we can | |
5215 | * log inodes up or down in the hierarchy) or rename operations for example. So | |
5216 | * we take the log_mutex of the inode after we have logged it and then check for | |
5217 | * its last_unlink_trans value - this is safe because any task setting | |
5218 | * last_unlink_trans must take the log_mutex and it must do this before it does | |
5219 | * the actual unlink operation, so if we do this check before a concurrent task | |
5220 | * sets last_unlink_trans it means we've logged a consistent version/state of | |
5221 | * all the inode items, otherwise we are not sure and must do a transaction | |
01327610 | 5222 | * commit (the concurrent task might have only updated last_unlink_trans before |
2be63d5c FM |
5223 | * we logged the inode or it might have also done the unlink). |
5224 | */ | |
5225 | static bool btrfs_must_commit_transaction(struct btrfs_trans_handle *trans, | |
ab1717b2 | 5226 | struct btrfs_inode *inode) |
2be63d5c | 5227 | { |
ab1717b2 | 5228 | struct btrfs_fs_info *fs_info = inode->root->fs_info; |
2be63d5c FM |
5229 | bool ret = false; |
5230 | ||
ab1717b2 NB |
5231 | mutex_lock(&inode->log_mutex); |
5232 | if (inode->last_unlink_trans > fs_info->last_trans_committed) { | |
2be63d5c FM |
5233 | /* |
5234 | * Make sure any commits to the log are forced to be full | |
5235 | * commits. | |
5236 | */ | |
5237 | btrfs_set_log_full_commit(fs_info, trans); | |
5238 | ret = true; | |
5239 | } | |
ab1717b2 | 5240 | mutex_unlock(&inode->log_mutex); |
2be63d5c FM |
5241 | |
5242 | return ret; | |
5243 | } | |
5244 | ||
12fcfd22 CM |
5245 | /* |
5246 | * follow the dentry parent pointers up the chain and see if any | |
5247 | * of the directories in it require a full commit before they can | |
5248 | * be logged. Returns zero if nothing special needs to be done or 1 if | |
5249 | * a full commit is required. | |
5250 | */ | |
5251 | static noinline int check_parent_dirs_for_sync(struct btrfs_trans_handle *trans, | |
aefa6115 | 5252 | struct btrfs_inode *inode, |
12fcfd22 CM |
5253 | struct dentry *parent, |
5254 | struct super_block *sb, | |
5255 | u64 last_committed) | |
e02119d5 | 5256 | { |
12fcfd22 | 5257 | int ret = 0; |
6a912213 | 5258 | struct dentry *old_parent = NULL; |
aefa6115 | 5259 | struct btrfs_inode *orig_inode = inode; |
e02119d5 | 5260 | |
af4176b4 CM |
5261 | /* |
5262 | * for regular files, if its inode is already on disk, we don't | |
5263 | * have to worry about the parents at all. This is because | |
5264 | * we can use the last_unlink_trans field to record renames | |
5265 | * and other fun in this file. | |
5266 | */ | |
aefa6115 NB |
5267 | if (S_ISREG(inode->vfs_inode.i_mode) && |
5268 | inode->generation <= last_committed && | |
5269 | inode->last_unlink_trans <= last_committed) | |
5270 | goto out; | |
af4176b4 | 5271 | |
aefa6115 | 5272 | if (!S_ISDIR(inode->vfs_inode.i_mode)) { |
fc64005c | 5273 | if (!parent || d_really_is_negative(parent) || sb != parent->d_sb) |
12fcfd22 | 5274 | goto out; |
aefa6115 | 5275 | inode = BTRFS_I(d_inode(parent)); |
12fcfd22 CM |
5276 | } |
5277 | ||
5278 | while (1) { | |
de2b530b JB |
5279 | /* |
5280 | * If we are logging a directory then we start with our inode, | |
01327610 | 5281 | * not our parent's inode, so we need to skip setting the |
de2b530b JB |
5282 | * logged_trans so that further down in the log code we don't |
5283 | * think this inode has already been logged. | |
5284 | */ | |
5285 | if (inode != orig_inode) | |
aefa6115 | 5286 | inode->logged_trans = trans->transid; |
12fcfd22 CM |
5287 | smp_mb(); |
5288 | ||
aefa6115 | 5289 | if (btrfs_must_commit_transaction(trans, inode)) { |
12fcfd22 CM |
5290 | ret = 1; |
5291 | break; | |
5292 | } | |
5293 | ||
fc64005c | 5294 | if (!parent || d_really_is_negative(parent) || sb != parent->d_sb) |
12fcfd22 CM |
5295 | break; |
5296 | ||
44f714da | 5297 | if (IS_ROOT(parent)) { |
aefa6115 NB |
5298 | inode = BTRFS_I(d_inode(parent)); |
5299 | if (btrfs_must_commit_transaction(trans, inode)) | |
44f714da | 5300 | ret = 1; |
12fcfd22 | 5301 | break; |
44f714da | 5302 | } |
12fcfd22 | 5303 | |
6a912213 JB |
5304 | parent = dget_parent(parent); |
5305 | dput(old_parent); | |
5306 | old_parent = parent; | |
aefa6115 | 5307 | inode = BTRFS_I(d_inode(parent)); |
12fcfd22 CM |
5308 | |
5309 | } | |
6a912213 | 5310 | dput(old_parent); |
12fcfd22 | 5311 | out: |
e02119d5 CM |
5312 | return ret; |
5313 | } | |
5314 | ||
2f2ff0ee FM |
5315 | struct btrfs_dir_list { |
5316 | u64 ino; | |
5317 | struct list_head list; | |
5318 | }; | |
5319 | ||
5320 | /* | |
5321 | * Log the inodes of the new dentries of a directory. See log_dir_items() for | |
5322 | * details about the why it is needed. | |
5323 | * This is a recursive operation - if an existing dentry corresponds to a | |
5324 | * directory, that directory's new entries are logged too (same behaviour as | |
5325 | * ext3/4, xfs, f2fs, reiserfs, nilfs2). Note that when logging the inodes | |
5326 | * the dentries point to we do not lock their i_mutex, otherwise lockdep | |
5327 | * complains about the following circular lock dependency / possible deadlock: | |
5328 | * | |
5329 | * CPU0 CPU1 | |
5330 | * ---- ---- | |
5331 | * lock(&type->i_mutex_dir_key#3/2); | |
5332 | * lock(sb_internal#2); | |
5333 | * lock(&type->i_mutex_dir_key#3/2); | |
5334 | * lock(&sb->s_type->i_mutex_key#14); | |
5335 | * | |
5336 | * Where sb_internal is the lock (a counter that works as a lock) acquired by | |
5337 | * sb_start_intwrite() in btrfs_start_transaction(). | |
5338 | * Not locking i_mutex of the inodes is still safe because: | |
5339 | * | |
5340 | * 1) For regular files we log with a mode of LOG_INODE_EXISTS. It's possible | |
5341 | * that while logging the inode new references (names) are added or removed | |
5342 | * from the inode, leaving the logged inode item with a link count that does | |
5343 | * not match the number of logged inode reference items. This is fine because | |
5344 | * at log replay time we compute the real number of links and correct the | |
5345 | * link count in the inode item (see replay_one_buffer() and | |
5346 | * link_to_fixup_dir()); | |
5347 | * | |
5348 | * 2) For directories we log with a mode of LOG_INODE_ALL. It's possible that | |
5349 | * while logging the inode's items new items with keys BTRFS_DIR_ITEM_KEY and | |
5350 | * BTRFS_DIR_INDEX_KEY are added to fs/subvol tree and the logged inode item | |
5351 | * has a size that doesn't match the sum of the lengths of all the logged | |
5352 | * names. This does not result in a problem because if a dir_item key is | |
5353 | * logged but its matching dir_index key is not logged, at log replay time we | |
5354 | * don't use it to replay the respective name (see replay_one_name()). On the | |
5355 | * other hand if only the dir_index key ends up being logged, the respective | |
5356 | * name is added to the fs/subvol tree with both the dir_item and dir_index | |
5357 | * keys created (see replay_one_name()). | |
5358 | * The directory's inode item with a wrong i_size is not a problem as well, | |
5359 | * since we don't use it at log replay time to set the i_size in the inode | |
5360 | * item of the fs/subvol tree (see overwrite_item()). | |
5361 | */ | |
5362 | static int log_new_dir_dentries(struct btrfs_trans_handle *trans, | |
5363 | struct btrfs_root *root, | |
51cc0d32 | 5364 | struct btrfs_inode *start_inode, |
2f2ff0ee FM |
5365 | struct btrfs_log_ctx *ctx) |
5366 | { | |
0b246afa | 5367 | struct btrfs_fs_info *fs_info = root->fs_info; |
2f2ff0ee FM |
5368 | struct btrfs_root *log = root->log_root; |
5369 | struct btrfs_path *path; | |
5370 | LIST_HEAD(dir_list); | |
5371 | struct btrfs_dir_list *dir_elem; | |
5372 | int ret = 0; | |
5373 | ||
5374 | path = btrfs_alloc_path(); | |
5375 | if (!path) | |
5376 | return -ENOMEM; | |
5377 | ||
5378 | dir_elem = kmalloc(sizeof(*dir_elem), GFP_NOFS); | |
5379 | if (!dir_elem) { | |
5380 | btrfs_free_path(path); | |
5381 | return -ENOMEM; | |
5382 | } | |
51cc0d32 | 5383 | dir_elem->ino = btrfs_ino(start_inode); |
2f2ff0ee FM |
5384 | list_add_tail(&dir_elem->list, &dir_list); |
5385 | ||
5386 | while (!list_empty(&dir_list)) { | |
5387 | struct extent_buffer *leaf; | |
5388 | struct btrfs_key min_key; | |
5389 | int nritems; | |
5390 | int i; | |
5391 | ||
5392 | dir_elem = list_first_entry(&dir_list, struct btrfs_dir_list, | |
5393 | list); | |
5394 | if (ret) | |
5395 | goto next_dir_inode; | |
5396 | ||
5397 | min_key.objectid = dir_elem->ino; | |
5398 | min_key.type = BTRFS_DIR_ITEM_KEY; | |
5399 | min_key.offset = 0; | |
5400 | again: | |
5401 | btrfs_release_path(path); | |
5402 | ret = btrfs_search_forward(log, &min_key, path, trans->transid); | |
5403 | if (ret < 0) { | |
5404 | goto next_dir_inode; | |
5405 | } else if (ret > 0) { | |
5406 | ret = 0; | |
5407 | goto next_dir_inode; | |
5408 | } | |
5409 | ||
5410 | process_leaf: | |
5411 | leaf = path->nodes[0]; | |
5412 | nritems = btrfs_header_nritems(leaf); | |
5413 | for (i = path->slots[0]; i < nritems; i++) { | |
5414 | struct btrfs_dir_item *di; | |
5415 | struct btrfs_key di_key; | |
5416 | struct inode *di_inode; | |
5417 | struct btrfs_dir_list *new_dir_elem; | |
5418 | int log_mode = LOG_INODE_EXISTS; | |
5419 | int type; | |
5420 | ||
5421 | btrfs_item_key_to_cpu(leaf, &min_key, i); | |
5422 | if (min_key.objectid != dir_elem->ino || | |
5423 | min_key.type != BTRFS_DIR_ITEM_KEY) | |
5424 | goto next_dir_inode; | |
5425 | ||
5426 | di = btrfs_item_ptr(leaf, i, struct btrfs_dir_item); | |
5427 | type = btrfs_dir_type(leaf, di); | |
5428 | if (btrfs_dir_transid(leaf, di) < trans->transid && | |
5429 | type != BTRFS_FT_DIR) | |
5430 | continue; | |
5431 | btrfs_dir_item_key_to_cpu(leaf, di, &di_key); | |
5432 | if (di_key.type == BTRFS_ROOT_ITEM_KEY) | |
5433 | continue; | |
5434 | ||
ec125cfb | 5435 | btrfs_release_path(path); |
0b246afa | 5436 | di_inode = btrfs_iget(fs_info->sb, &di_key, root, NULL); |
2f2ff0ee FM |
5437 | if (IS_ERR(di_inode)) { |
5438 | ret = PTR_ERR(di_inode); | |
5439 | goto next_dir_inode; | |
5440 | } | |
5441 | ||
0f8939b8 | 5442 | if (btrfs_inode_in_log(BTRFS_I(di_inode), trans->transid)) { |
2f2ff0ee | 5443 | iput(di_inode); |
ec125cfb | 5444 | break; |
2f2ff0ee FM |
5445 | } |
5446 | ||
5447 | ctx->log_new_dentries = false; | |
3f9749f6 | 5448 | if (type == BTRFS_FT_DIR || type == BTRFS_FT_SYMLINK) |
2f2ff0ee | 5449 | log_mode = LOG_INODE_ALL; |
a59108a7 | 5450 | ret = btrfs_log_inode(trans, root, BTRFS_I(di_inode), |
2f2ff0ee | 5451 | log_mode, 0, LLONG_MAX, ctx); |
2be63d5c | 5452 | if (!ret && |
ab1717b2 | 5453 | btrfs_must_commit_transaction(trans, BTRFS_I(di_inode))) |
2be63d5c | 5454 | ret = 1; |
2f2ff0ee FM |
5455 | iput(di_inode); |
5456 | if (ret) | |
5457 | goto next_dir_inode; | |
5458 | if (ctx->log_new_dentries) { | |
5459 | new_dir_elem = kmalloc(sizeof(*new_dir_elem), | |
5460 | GFP_NOFS); | |
5461 | if (!new_dir_elem) { | |
5462 | ret = -ENOMEM; | |
5463 | goto next_dir_inode; | |
5464 | } | |
5465 | new_dir_elem->ino = di_key.objectid; | |
5466 | list_add_tail(&new_dir_elem->list, &dir_list); | |
5467 | } | |
5468 | break; | |
5469 | } | |
5470 | if (i == nritems) { | |
5471 | ret = btrfs_next_leaf(log, path); | |
5472 | if (ret < 0) { | |
5473 | goto next_dir_inode; | |
5474 | } else if (ret > 0) { | |
5475 | ret = 0; | |
5476 | goto next_dir_inode; | |
5477 | } | |
5478 | goto process_leaf; | |
5479 | } | |
5480 | if (min_key.offset < (u64)-1) { | |
5481 | min_key.offset++; | |
5482 | goto again; | |
5483 | } | |
5484 | next_dir_inode: | |
5485 | list_del(&dir_elem->list); | |
5486 | kfree(dir_elem); | |
5487 | } | |
5488 | ||
5489 | btrfs_free_path(path); | |
5490 | return ret; | |
5491 | } | |
5492 | ||
18aa0922 | 5493 | static int btrfs_log_all_parents(struct btrfs_trans_handle *trans, |
d0a0b78d | 5494 | struct btrfs_inode *inode, |
18aa0922 FM |
5495 | struct btrfs_log_ctx *ctx) |
5496 | { | |
3ffbd68c | 5497 | struct btrfs_fs_info *fs_info = trans->fs_info; |
18aa0922 FM |
5498 | int ret; |
5499 | struct btrfs_path *path; | |
5500 | struct btrfs_key key; | |
d0a0b78d NB |
5501 | struct btrfs_root *root = inode->root; |
5502 | const u64 ino = btrfs_ino(inode); | |
18aa0922 FM |
5503 | |
5504 | path = btrfs_alloc_path(); | |
5505 | if (!path) | |
5506 | return -ENOMEM; | |
5507 | path->skip_locking = 1; | |
5508 | path->search_commit_root = 1; | |
5509 | ||
5510 | key.objectid = ino; | |
5511 | key.type = BTRFS_INODE_REF_KEY; | |
5512 | key.offset = 0; | |
5513 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
5514 | if (ret < 0) | |
5515 | goto out; | |
5516 | ||
5517 | while (true) { | |
5518 | struct extent_buffer *leaf = path->nodes[0]; | |
5519 | int slot = path->slots[0]; | |
5520 | u32 cur_offset = 0; | |
5521 | u32 item_size; | |
5522 | unsigned long ptr; | |
5523 | ||
5524 | if (slot >= btrfs_header_nritems(leaf)) { | |
5525 | ret = btrfs_next_leaf(root, path); | |
5526 | if (ret < 0) | |
5527 | goto out; | |
5528 | else if (ret > 0) | |
5529 | break; | |
5530 | continue; | |
5531 | } | |
5532 | ||
5533 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
5534 | /* BTRFS_INODE_EXTREF_KEY is BTRFS_INODE_REF_KEY + 1 */ | |
5535 | if (key.objectid != ino || key.type > BTRFS_INODE_EXTREF_KEY) | |
5536 | break; | |
5537 | ||
5538 | item_size = btrfs_item_size_nr(leaf, slot); | |
5539 | ptr = btrfs_item_ptr_offset(leaf, slot); | |
5540 | while (cur_offset < item_size) { | |
5541 | struct btrfs_key inode_key; | |
5542 | struct inode *dir_inode; | |
5543 | ||
5544 | inode_key.type = BTRFS_INODE_ITEM_KEY; | |
5545 | inode_key.offset = 0; | |
5546 | ||
5547 | if (key.type == BTRFS_INODE_EXTREF_KEY) { | |
5548 | struct btrfs_inode_extref *extref; | |
5549 | ||
5550 | extref = (struct btrfs_inode_extref *) | |
5551 | (ptr + cur_offset); | |
5552 | inode_key.objectid = btrfs_inode_extref_parent( | |
5553 | leaf, extref); | |
5554 | cur_offset += sizeof(*extref); | |
5555 | cur_offset += btrfs_inode_extref_name_len(leaf, | |
5556 | extref); | |
5557 | } else { | |
5558 | inode_key.objectid = key.offset; | |
5559 | cur_offset = item_size; | |
5560 | } | |
5561 | ||
0b246afa | 5562 | dir_inode = btrfs_iget(fs_info->sb, &inode_key, |
18aa0922 FM |
5563 | root, NULL); |
5564 | /* If parent inode was deleted, skip it. */ | |
5565 | if (IS_ERR(dir_inode)) | |
5566 | continue; | |
5567 | ||
657ed1aa FM |
5568 | if (ctx) |
5569 | ctx->log_new_dentries = false; | |
a59108a7 | 5570 | ret = btrfs_log_inode(trans, root, BTRFS_I(dir_inode), |
18aa0922 | 5571 | LOG_INODE_ALL, 0, LLONG_MAX, ctx); |
2be63d5c | 5572 | if (!ret && |
ab1717b2 | 5573 | btrfs_must_commit_transaction(trans, BTRFS_I(dir_inode))) |
2be63d5c | 5574 | ret = 1; |
657ed1aa FM |
5575 | if (!ret && ctx && ctx->log_new_dentries) |
5576 | ret = log_new_dir_dentries(trans, root, | |
f85b7379 | 5577 | BTRFS_I(dir_inode), ctx); |
18aa0922 FM |
5578 | iput(dir_inode); |
5579 | if (ret) | |
5580 | goto out; | |
5581 | } | |
5582 | path->slots[0]++; | |
5583 | } | |
5584 | ret = 0; | |
5585 | out: | |
5586 | btrfs_free_path(path); | |
5587 | return ret; | |
5588 | } | |
5589 | ||
e02119d5 CM |
5590 | /* |
5591 | * helper function around btrfs_log_inode to make sure newly created | |
5592 | * parent directories also end up in the log. A minimal inode and backref | |
5593 | * only logging is done of any parent directories that are older than | |
5594 | * the last committed transaction | |
5595 | */ | |
48a3b636 | 5596 | static int btrfs_log_inode_parent(struct btrfs_trans_handle *trans, |
19df27a9 | 5597 | struct btrfs_inode *inode, |
49dae1bc FM |
5598 | struct dentry *parent, |
5599 | const loff_t start, | |
5600 | const loff_t end, | |
41a1eada | 5601 | int inode_only, |
8b050d35 | 5602 | struct btrfs_log_ctx *ctx) |
e02119d5 | 5603 | { |
f882274b | 5604 | struct btrfs_root *root = inode->root; |
0b246afa | 5605 | struct btrfs_fs_info *fs_info = root->fs_info; |
e02119d5 | 5606 | struct super_block *sb; |
6a912213 | 5607 | struct dentry *old_parent = NULL; |
12fcfd22 | 5608 | int ret = 0; |
0b246afa | 5609 | u64 last_committed = fs_info->last_trans_committed; |
2f2ff0ee | 5610 | bool log_dentries = false; |
19df27a9 | 5611 | struct btrfs_inode *orig_inode = inode; |
12fcfd22 | 5612 | |
19df27a9 | 5613 | sb = inode->vfs_inode.i_sb; |
12fcfd22 | 5614 | |
0b246afa | 5615 | if (btrfs_test_opt(fs_info, NOTREELOG)) { |
3a5e1404 SW |
5616 | ret = 1; |
5617 | goto end_no_trans; | |
5618 | } | |
5619 | ||
995946dd MX |
5620 | /* |
5621 | * The prev transaction commit doesn't complete, we need do | |
5622 | * full commit by ourselves. | |
5623 | */ | |
0b246afa JM |
5624 | if (fs_info->last_trans_log_full_commit > |
5625 | fs_info->last_trans_committed) { | |
12fcfd22 CM |
5626 | ret = 1; |
5627 | goto end_no_trans; | |
5628 | } | |
5629 | ||
f882274b | 5630 | if (btrfs_root_refs(&root->root_item) == 0) { |
76dda93c YZ |
5631 | ret = 1; |
5632 | goto end_no_trans; | |
5633 | } | |
5634 | ||
19df27a9 NB |
5635 | ret = check_parent_dirs_for_sync(trans, inode, parent, sb, |
5636 | last_committed); | |
12fcfd22 CM |
5637 | if (ret) |
5638 | goto end_no_trans; | |
e02119d5 | 5639 | |
19df27a9 | 5640 | if (btrfs_inode_in_log(inode, trans->transid)) { |
257c62e1 CM |
5641 | ret = BTRFS_NO_LOG_SYNC; |
5642 | goto end_no_trans; | |
5643 | } | |
5644 | ||
8b050d35 | 5645 | ret = start_log_trans(trans, root, ctx); |
4a500fd1 | 5646 | if (ret) |
e87ac136 | 5647 | goto end_no_trans; |
e02119d5 | 5648 | |
19df27a9 | 5649 | ret = btrfs_log_inode(trans, root, inode, inode_only, start, end, ctx); |
4a500fd1 YZ |
5650 | if (ret) |
5651 | goto end_trans; | |
12fcfd22 | 5652 | |
af4176b4 CM |
5653 | /* |
5654 | * for regular files, if its inode is already on disk, we don't | |
5655 | * have to worry about the parents at all. This is because | |
5656 | * we can use the last_unlink_trans field to record renames | |
5657 | * and other fun in this file. | |
5658 | */ | |
19df27a9 NB |
5659 | if (S_ISREG(inode->vfs_inode.i_mode) && |
5660 | inode->generation <= last_committed && | |
5661 | inode->last_unlink_trans <= last_committed) { | |
4a500fd1 YZ |
5662 | ret = 0; |
5663 | goto end_trans; | |
5664 | } | |
af4176b4 | 5665 | |
19df27a9 | 5666 | if (S_ISDIR(inode->vfs_inode.i_mode) && ctx && ctx->log_new_dentries) |
2f2ff0ee FM |
5667 | log_dentries = true; |
5668 | ||
18aa0922 | 5669 | /* |
01327610 | 5670 | * On unlink we must make sure all our current and old parent directory |
18aa0922 FM |
5671 | * inodes are fully logged. This is to prevent leaving dangling |
5672 | * directory index entries in directories that were our parents but are | |
5673 | * not anymore. Not doing this results in old parent directory being | |
5674 | * impossible to delete after log replay (rmdir will always fail with | |
5675 | * error -ENOTEMPTY). | |
5676 | * | |
5677 | * Example 1: | |
5678 | * | |
5679 | * mkdir testdir | |
5680 | * touch testdir/foo | |
5681 | * ln testdir/foo testdir/bar | |
5682 | * sync | |
5683 | * unlink testdir/bar | |
5684 | * xfs_io -c fsync testdir/foo | |
5685 | * <power failure> | |
5686 | * mount fs, triggers log replay | |
5687 | * | |
5688 | * If we don't log the parent directory (testdir), after log replay the | |
5689 | * directory still has an entry pointing to the file inode using the bar | |
5690 | * name, but a matching BTRFS_INODE_[REF|EXTREF]_KEY does not exist and | |
5691 | * the file inode has a link count of 1. | |
5692 | * | |
5693 | * Example 2: | |
5694 | * | |
5695 | * mkdir testdir | |
5696 | * touch foo | |
5697 | * ln foo testdir/foo2 | |
5698 | * ln foo testdir/foo3 | |
5699 | * sync | |
5700 | * unlink testdir/foo3 | |
5701 | * xfs_io -c fsync foo | |
5702 | * <power failure> | |
5703 | * mount fs, triggers log replay | |
5704 | * | |
5705 | * Similar as the first example, after log replay the parent directory | |
5706 | * testdir still has an entry pointing to the inode file with name foo3 | |
5707 | * but the file inode does not have a matching BTRFS_INODE_REF_KEY item | |
5708 | * and has a link count of 2. | |
5709 | */ | |
19df27a9 | 5710 | if (inode->last_unlink_trans > last_committed) { |
18aa0922 FM |
5711 | ret = btrfs_log_all_parents(trans, orig_inode, ctx); |
5712 | if (ret) | |
5713 | goto end_trans; | |
5714 | } | |
5715 | ||
12fcfd22 | 5716 | while (1) { |
fc64005c | 5717 | if (!parent || d_really_is_negative(parent) || sb != parent->d_sb) |
e02119d5 CM |
5718 | break; |
5719 | ||
19df27a9 NB |
5720 | inode = BTRFS_I(d_inode(parent)); |
5721 | if (root != inode->root) | |
76dda93c YZ |
5722 | break; |
5723 | ||
19df27a9 NB |
5724 | if (inode->generation > last_committed) { |
5725 | ret = btrfs_log_inode(trans, root, inode, | |
5726 | LOG_INODE_EXISTS, 0, LLONG_MAX, ctx); | |
4a500fd1 YZ |
5727 | if (ret) |
5728 | goto end_trans; | |
12fcfd22 | 5729 | } |
76dda93c | 5730 | if (IS_ROOT(parent)) |
e02119d5 | 5731 | break; |
12fcfd22 | 5732 | |
6a912213 JB |
5733 | parent = dget_parent(parent); |
5734 | dput(old_parent); | |
5735 | old_parent = parent; | |
e02119d5 | 5736 | } |
2f2ff0ee | 5737 | if (log_dentries) |
19df27a9 | 5738 | ret = log_new_dir_dentries(trans, root, orig_inode, ctx); |
2f2ff0ee FM |
5739 | else |
5740 | ret = 0; | |
4a500fd1 | 5741 | end_trans: |
6a912213 | 5742 | dput(old_parent); |
4a500fd1 | 5743 | if (ret < 0) { |
0b246afa | 5744 | btrfs_set_log_full_commit(fs_info, trans); |
4a500fd1 YZ |
5745 | ret = 1; |
5746 | } | |
8b050d35 MX |
5747 | |
5748 | if (ret) | |
5749 | btrfs_remove_log_ctx(root, ctx); | |
12fcfd22 CM |
5750 | btrfs_end_log_trans(root); |
5751 | end_no_trans: | |
5752 | return ret; | |
e02119d5 CM |
5753 | } |
5754 | ||
5755 | /* | |
5756 | * it is not safe to log dentry if the chunk root has added new | |
5757 | * chunks. This returns 0 if the dentry was logged, and 1 otherwise. | |
5758 | * If this returns 1, you must commit the transaction to safely get your | |
5759 | * data on disk. | |
5760 | */ | |
5761 | int btrfs_log_dentry_safe(struct btrfs_trans_handle *trans, | |
e5b84f7a | 5762 | struct dentry *dentry, |
49dae1bc FM |
5763 | const loff_t start, |
5764 | const loff_t end, | |
8b050d35 | 5765 | struct btrfs_log_ctx *ctx) |
e02119d5 | 5766 | { |
6a912213 JB |
5767 | struct dentry *parent = dget_parent(dentry); |
5768 | int ret; | |
5769 | ||
f882274b NB |
5770 | ret = btrfs_log_inode_parent(trans, BTRFS_I(d_inode(dentry)), parent, |
5771 | start, end, LOG_INODE_ALL, ctx); | |
6a912213 JB |
5772 | dput(parent); |
5773 | ||
5774 | return ret; | |
e02119d5 CM |
5775 | } |
5776 | ||
5777 | /* | |
5778 | * should be called during mount to recover any replay any log trees | |
5779 | * from the FS | |
5780 | */ | |
5781 | int btrfs_recover_log_trees(struct btrfs_root *log_root_tree) | |
5782 | { | |
5783 | int ret; | |
5784 | struct btrfs_path *path; | |
5785 | struct btrfs_trans_handle *trans; | |
5786 | struct btrfs_key key; | |
5787 | struct btrfs_key found_key; | |
5788 | struct btrfs_key tmp_key; | |
5789 | struct btrfs_root *log; | |
5790 | struct btrfs_fs_info *fs_info = log_root_tree->fs_info; | |
5791 | struct walk_control wc = { | |
5792 | .process_func = process_one_buffer, | |
5793 | .stage = 0, | |
5794 | }; | |
5795 | ||
e02119d5 | 5796 | path = btrfs_alloc_path(); |
db5b493a TI |
5797 | if (!path) |
5798 | return -ENOMEM; | |
5799 | ||
afcdd129 | 5800 | set_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags); |
e02119d5 | 5801 | |
4a500fd1 | 5802 | trans = btrfs_start_transaction(fs_info->tree_root, 0); |
79787eaa JM |
5803 | if (IS_ERR(trans)) { |
5804 | ret = PTR_ERR(trans); | |
5805 | goto error; | |
5806 | } | |
e02119d5 CM |
5807 | |
5808 | wc.trans = trans; | |
5809 | wc.pin = 1; | |
5810 | ||
db5b493a | 5811 | ret = walk_log_tree(trans, log_root_tree, &wc); |
79787eaa | 5812 | if (ret) { |
5d163e0e JM |
5813 | btrfs_handle_fs_error(fs_info, ret, |
5814 | "Failed to pin buffers while recovering log root tree."); | |
79787eaa JM |
5815 | goto error; |
5816 | } | |
e02119d5 CM |
5817 | |
5818 | again: | |
5819 | key.objectid = BTRFS_TREE_LOG_OBJECTID; | |
5820 | key.offset = (u64)-1; | |
962a298f | 5821 | key.type = BTRFS_ROOT_ITEM_KEY; |
e02119d5 | 5822 | |
d397712b | 5823 | while (1) { |
e02119d5 | 5824 | ret = btrfs_search_slot(NULL, log_root_tree, &key, path, 0, 0); |
79787eaa JM |
5825 | |
5826 | if (ret < 0) { | |
34d97007 | 5827 | btrfs_handle_fs_error(fs_info, ret, |
79787eaa JM |
5828 | "Couldn't find tree log root."); |
5829 | goto error; | |
5830 | } | |
e02119d5 CM |
5831 | if (ret > 0) { |
5832 | if (path->slots[0] == 0) | |
5833 | break; | |
5834 | path->slots[0]--; | |
5835 | } | |
5836 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, | |
5837 | path->slots[0]); | |
b3b4aa74 | 5838 | btrfs_release_path(path); |
e02119d5 CM |
5839 | if (found_key.objectid != BTRFS_TREE_LOG_OBJECTID) |
5840 | break; | |
5841 | ||
cb517eab | 5842 | log = btrfs_read_fs_root(log_root_tree, &found_key); |
79787eaa JM |
5843 | if (IS_ERR(log)) { |
5844 | ret = PTR_ERR(log); | |
34d97007 | 5845 | btrfs_handle_fs_error(fs_info, ret, |
79787eaa JM |
5846 | "Couldn't read tree log root."); |
5847 | goto error; | |
5848 | } | |
e02119d5 CM |
5849 | |
5850 | tmp_key.objectid = found_key.offset; | |
5851 | tmp_key.type = BTRFS_ROOT_ITEM_KEY; | |
5852 | tmp_key.offset = (u64)-1; | |
5853 | ||
5854 | wc.replay_dest = btrfs_read_fs_root_no_name(fs_info, &tmp_key); | |
79787eaa JM |
5855 | if (IS_ERR(wc.replay_dest)) { |
5856 | ret = PTR_ERR(wc.replay_dest); | |
b50c6e25 JB |
5857 | free_extent_buffer(log->node); |
5858 | free_extent_buffer(log->commit_root); | |
5859 | kfree(log); | |
5d163e0e JM |
5860 | btrfs_handle_fs_error(fs_info, ret, |
5861 | "Couldn't read target root for tree log recovery."); | |
79787eaa JM |
5862 | goto error; |
5863 | } | |
e02119d5 | 5864 | |
07d400a6 | 5865 | wc.replay_dest->log_root = log; |
5d4f98a2 | 5866 | btrfs_record_root_in_trans(trans, wc.replay_dest); |
e02119d5 | 5867 | ret = walk_log_tree(trans, log, &wc); |
e02119d5 | 5868 | |
b50c6e25 | 5869 | if (!ret && wc.stage == LOG_WALK_REPLAY_ALL) { |
e02119d5 CM |
5870 | ret = fixup_inode_link_counts(trans, wc.replay_dest, |
5871 | path); | |
e02119d5 CM |
5872 | } |
5873 | ||
900c9981 LB |
5874 | if (!ret && wc.stage == LOG_WALK_REPLAY_ALL) { |
5875 | struct btrfs_root *root = wc.replay_dest; | |
5876 | ||
5877 | btrfs_release_path(path); | |
5878 | ||
5879 | /* | |
5880 | * We have just replayed everything, and the highest | |
5881 | * objectid of fs roots probably has changed in case | |
5882 | * some inode_item's got replayed. | |
5883 | * | |
5884 | * root->objectid_mutex is not acquired as log replay | |
5885 | * could only happen during mount. | |
5886 | */ | |
5887 | ret = btrfs_find_highest_objectid(root, | |
5888 | &root->highest_objectid); | |
5889 | } | |
5890 | ||
e02119d5 | 5891 | key.offset = found_key.offset - 1; |
07d400a6 | 5892 | wc.replay_dest->log_root = NULL; |
e02119d5 | 5893 | free_extent_buffer(log->node); |
b263c2c8 | 5894 | free_extent_buffer(log->commit_root); |
e02119d5 CM |
5895 | kfree(log); |
5896 | ||
b50c6e25 JB |
5897 | if (ret) |
5898 | goto error; | |
5899 | ||
e02119d5 CM |
5900 | if (found_key.offset == 0) |
5901 | break; | |
5902 | } | |
b3b4aa74 | 5903 | btrfs_release_path(path); |
e02119d5 CM |
5904 | |
5905 | /* step one is to pin it all, step two is to replay just inodes */ | |
5906 | if (wc.pin) { | |
5907 | wc.pin = 0; | |
5908 | wc.process_func = replay_one_buffer; | |
5909 | wc.stage = LOG_WALK_REPLAY_INODES; | |
5910 | goto again; | |
5911 | } | |
5912 | /* step three is to replay everything */ | |
5913 | if (wc.stage < LOG_WALK_REPLAY_ALL) { | |
5914 | wc.stage++; | |
5915 | goto again; | |
5916 | } | |
5917 | ||
5918 | btrfs_free_path(path); | |
5919 | ||
abefa55a | 5920 | /* step 4: commit the transaction, which also unpins the blocks */ |
3a45bb20 | 5921 | ret = btrfs_commit_transaction(trans); |
abefa55a JB |
5922 | if (ret) |
5923 | return ret; | |
5924 | ||
e02119d5 CM |
5925 | free_extent_buffer(log_root_tree->node); |
5926 | log_root_tree->log_root = NULL; | |
afcdd129 | 5927 | clear_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags); |
e02119d5 | 5928 | kfree(log_root_tree); |
79787eaa | 5929 | |
abefa55a | 5930 | return 0; |
79787eaa | 5931 | error: |
b50c6e25 | 5932 | if (wc.trans) |
3a45bb20 | 5933 | btrfs_end_transaction(wc.trans); |
79787eaa JM |
5934 | btrfs_free_path(path); |
5935 | return ret; | |
e02119d5 | 5936 | } |
12fcfd22 CM |
5937 | |
5938 | /* | |
5939 | * there are some corner cases where we want to force a full | |
5940 | * commit instead of allowing a directory to be logged. | |
5941 | * | |
5942 | * They revolve around files there were unlinked from the directory, and | |
5943 | * this function updates the parent directory so that a full commit is | |
5944 | * properly done if it is fsync'd later after the unlinks are done. | |
2be63d5c FM |
5945 | * |
5946 | * Must be called before the unlink operations (updates to the subvolume tree, | |
5947 | * inodes, etc) are done. | |
12fcfd22 CM |
5948 | */ |
5949 | void btrfs_record_unlink_dir(struct btrfs_trans_handle *trans, | |
4176bdbf | 5950 | struct btrfs_inode *dir, struct btrfs_inode *inode, |
12fcfd22 CM |
5951 | int for_rename) |
5952 | { | |
af4176b4 CM |
5953 | /* |
5954 | * when we're logging a file, if it hasn't been renamed | |
5955 | * or unlinked, and its inode is fully committed on disk, | |
5956 | * we don't have to worry about walking up the directory chain | |
5957 | * to log its parents. | |
5958 | * | |
5959 | * So, we use the last_unlink_trans field to put this transid | |
5960 | * into the file. When the file is logged we check it and | |
5961 | * don't log the parents if the file is fully on disk. | |
5962 | */ | |
4176bdbf NB |
5963 | mutex_lock(&inode->log_mutex); |
5964 | inode->last_unlink_trans = trans->transid; | |
5965 | mutex_unlock(&inode->log_mutex); | |
af4176b4 | 5966 | |
12fcfd22 CM |
5967 | /* |
5968 | * if this directory was already logged any new | |
5969 | * names for this file/dir will get recorded | |
5970 | */ | |
5971 | smp_mb(); | |
4176bdbf | 5972 | if (dir->logged_trans == trans->transid) |
12fcfd22 CM |
5973 | return; |
5974 | ||
5975 | /* | |
5976 | * if the inode we're about to unlink was logged, | |
5977 | * the log will be properly updated for any new names | |
5978 | */ | |
4176bdbf | 5979 | if (inode->logged_trans == trans->transid) |
12fcfd22 CM |
5980 | return; |
5981 | ||
5982 | /* | |
5983 | * when renaming files across directories, if the directory | |
5984 | * there we're unlinking from gets fsync'd later on, there's | |
5985 | * no way to find the destination directory later and fsync it | |
5986 | * properly. So, we have to be conservative and force commits | |
5987 | * so the new name gets discovered. | |
5988 | */ | |
5989 | if (for_rename) | |
5990 | goto record; | |
5991 | ||
5992 | /* we can safely do the unlink without any special recording */ | |
5993 | return; | |
5994 | ||
5995 | record: | |
4176bdbf NB |
5996 | mutex_lock(&dir->log_mutex); |
5997 | dir->last_unlink_trans = trans->transid; | |
5998 | mutex_unlock(&dir->log_mutex); | |
1ec9a1ae FM |
5999 | } |
6000 | ||
6001 | /* | |
6002 | * Make sure that if someone attempts to fsync the parent directory of a deleted | |
6003 | * snapshot, it ends up triggering a transaction commit. This is to guarantee | |
6004 | * that after replaying the log tree of the parent directory's root we will not | |
6005 | * see the snapshot anymore and at log replay time we will not see any log tree | |
6006 | * corresponding to the deleted snapshot's root, which could lead to replaying | |
6007 | * it after replaying the log tree of the parent directory (which would replay | |
6008 | * the snapshot delete operation). | |
2be63d5c FM |
6009 | * |
6010 | * Must be called before the actual snapshot destroy operation (updates to the | |
6011 | * parent root and tree of tree roots trees, etc) are done. | |
1ec9a1ae FM |
6012 | */ |
6013 | void btrfs_record_snapshot_destroy(struct btrfs_trans_handle *trans, | |
43663557 | 6014 | struct btrfs_inode *dir) |
1ec9a1ae | 6015 | { |
43663557 NB |
6016 | mutex_lock(&dir->log_mutex); |
6017 | dir->last_unlink_trans = trans->transid; | |
6018 | mutex_unlock(&dir->log_mutex); | |
12fcfd22 CM |
6019 | } |
6020 | ||
6021 | /* | |
6022 | * Call this after adding a new name for a file and it will properly | |
6023 | * update the log to reflect the new name. | |
6024 | * | |
d4682ba0 FM |
6025 | * @ctx can not be NULL when @sync_log is false, and should be NULL when it's |
6026 | * true (because it's not used). | |
6027 | * | |
6028 | * Return value depends on whether @sync_log is true or false. | |
6029 | * When true: returns BTRFS_NEED_TRANS_COMMIT if the transaction needs to be | |
6030 | * committed by the caller, and BTRFS_DONT_NEED_TRANS_COMMIT | |
6031 | * otherwise. | |
6032 | * When false: returns BTRFS_DONT_NEED_LOG_SYNC if the caller does not need to | |
6033 | * to sync the log, BTRFS_NEED_LOG_SYNC if it needs to sync the log, | |
6034 | * or BTRFS_NEED_TRANS_COMMIT if the transaction needs to be | |
6035 | * committed (without attempting to sync the log). | |
12fcfd22 CM |
6036 | */ |
6037 | int btrfs_log_new_name(struct btrfs_trans_handle *trans, | |
9ca5fbfb | 6038 | struct btrfs_inode *inode, struct btrfs_inode *old_dir, |
d4682ba0 FM |
6039 | struct dentry *parent, |
6040 | bool sync_log, struct btrfs_log_ctx *ctx) | |
12fcfd22 | 6041 | { |
3ffbd68c | 6042 | struct btrfs_fs_info *fs_info = trans->fs_info; |
d4682ba0 | 6043 | int ret; |
12fcfd22 | 6044 | |
af4176b4 CM |
6045 | /* |
6046 | * this will force the logging code to walk the dentry chain | |
6047 | * up for the file | |
6048 | */ | |
9a6509c4 | 6049 | if (!S_ISDIR(inode->vfs_inode.i_mode)) |
9ca5fbfb | 6050 | inode->last_unlink_trans = trans->transid; |
af4176b4 | 6051 | |
12fcfd22 CM |
6052 | /* |
6053 | * if this inode hasn't been logged and directory we're renaming it | |
6054 | * from hasn't been logged, we don't need to log it | |
6055 | */ | |
9ca5fbfb NB |
6056 | if (inode->logged_trans <= fs_info->last_trans_committed && |
6057 | (!old_dir || old_dir->logged_trans <= fs_info->last_trans_committed)) | |
d4682ba0 FM |
6058 | return sync_log ? BTRFS_DONT_NEED_TRANS_COMMIT : |
6059 | BTRFS_DONT_NEED_LOG_SYNC; | |
6060 | ||
6061 | if (sync_log) { | |
6062 | struct btrfs_log_ctx ctx2; | |
6063 | ||
6064 | btrfs_init_log_ctx(&ctx2, &inode->vfs_inode); | |
6065 | ret = btrfs_log_inode_parent(trans, inode, parent, 0, LLONG_MAX, | |
6066 | LOG_INODE_EXISTS, &ctx2); | |
6067 | if (ret == BTRFS_NO_LOG_SYNC) | |
6068 | return BTRFS_DONT_NEED_TRANS_COMMIT; | |
6069 | else if (ret) | |
6070 | return BTRFS_NEED_TRANS_COMMIT; | |
6071 | ||
6072 | ret = btrfs_sync_log(trans, inode->root, &ctx2); | |
6073 | if (ret) | |
6074 | return BTRFS_NEED_TRANS_COMMIT; | |
6075 | return BTRFS_DONT_NEED_TRANS_COMMIT; | |
6076 | } | |
6077 | ||
6078 | ASSERT(ctx); | |
6079 | ret = btrfs_log_inode_parent(trans, inode, parent, 0, LLONG_MAX, | |
6080 | LOG_INODE_EXISTS, ctx); | |
6081 | if (ret == BTRFS_NO_LOG_SYNC) | |
6082 | return BTRFS_DONT_NEED_LOG_SYNC; | |
6083 | else if (ret) | |
6084 | return BTRFS_NEED_TRANS_COMMIT; | |
12fcfd22 | 6085 | |
d4682ba0 | 6086 | return BTRFS_NEED_LOG_SYNC; |
12fcfd22 CM |
6087 | } |
6088 |