]>
Commit | Line | Data |
---|---|---|
e02119d5 CM |
1 | /* |
2 | * Copyright (C) 2008 Oracle. All rights reserved. | |
3 | * | |
4 | * This program is free software; you can redistribute it and/or | |
5 | * modify it under the terms of the GNU General Public | |
6 | * License v2 as published by the Free Software Foundation. | |
7 | * | |
8 | * This program is distributed in the hope that it will be useful, | |
9 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
10 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
11 | * General Public License for more details. | |
12 | * | |
13 | * You should have received a copy of the GNU General Public | |
14 | * License along with this program; if not, write to the | |
15 | * Free Software Foundation, Inc., 59 Temple Place - Suite 330, | |
16 | * Boston, MA 021110-1307, USA. | |
17 | */ | |
18 | ||
19 | #include <linux/sched.h> | |
5a0e3ad6 | 20 | #include <linux/slab.h> |
c6adc9cc | 21 | #include <linux/blkdev.h> |
5dc562c5 | 22 | #include <linux/list_sort.h> |
995946dd | 23 | #include "tree-log.h" |
e02119d5 CM |
24 | #include "disk-io.h" |
25 | #include "locking.h" | |
26 | #include "print-tree.h" | |
f186373f | 27 | #include "backref.h" |
f186373f | 28 | #include "hash.h" |
e02119d5 CM |
29 | |
30 | /* magic values for the inode_only field in btrfs_log_inode: | |
31 | * | |
32 | * LOG_INODE_ALL means to log everything | |
33 | * LOG_INODE_EXISTS means to log just enough to recreate the inode | |
34 | * during log replay | |
35 | */ | |
36 | #define LOG_INODE_ALL 0 | |
37 | #define LOG_INODE_EXISTS 1 | |
38 | ||
12fcfd22 CM |
39 | /* |
40 | * directory trouble cases | |
41 | * | |
42 | * 1) on rename or unlink, if the inode being unlinked isn't in the fsync | |
43 | * log, we must force a full commit before doing an fsync of the directory | |
44 | * where the unlink was done. | |
45 | * ---> record transid of last unlink/rename per directory | |
46 | * | |
47 | * mkdir foo/some_dir | |
48 | * normal commit | |
49 | * rename foo/some_dir foo2/some_dir | |
50 | * mkdir foo/some_dir | |
51 | * fsync foo/some_dir/some_file | |
52 | * | |
53 | * The fsync above will unlink the original some_dir without recording | |
54 | * it in its new location (foo2). After a crash, some_dir will be gone | |
55 | * unless the fsync of some_file forces a full commit | |
56 | * | |
57 | * 2) we must log any new names for any file or dir that is in the fsync | |
58 | * log. ---> check inode while renaming/linking. | |
59 | * | |
60 | * 2a) we must log any new names for any file or dir during rename | |
61 | * when the directory they are being removed from was logged. | |
62 | * ---> check inode and old parent dir during rename | |
63 | * | |
64 | * 2a is actually the more important variant. With the extra logging | |
65 | * a crash might unlink the old name without recreating the new one | |
66 | * | |
67 | * 3) after a crash, we must go through any directories with a link count | |
68 | * of zero and redo the rm -rf | |
69 | * | |
70 | * mkdir f1/foo | |
71 | * normal commit | |
72 | * rm -rf f1/foo | |
73 | * fsync(f1) | |
74 | * | |
75 | * The directory f1 was fully removed from the FS, but fsync was never | |
76 | * called on f1, only its parent dir. After a crash the rm -rf must | |
77 | * be replayed. This must be able to recurse down the entire | |
78 | * directory tree. The inode link count fixup code takes care of the | |
79 | * ugly details. | |
80 | */ | |
81 | ||
e02119d5 CM |
82 | /* |
83 | * stages for the tree walking. The first | |
84 | * stage (0) is to only pin down the blocks we find | |
85 | * the second stage (1) is to make sure that all the inodes | |
86 | * we find in the log are created in the subvolume. | |
87 | * | |
88 | * The last stage is to deal with directories and links and extents | |
89 | * and all the other fun semantics | |
90 | */ | |
91 | #define LOG_WALK_PIN_ONLY 0 | |
92 | #define LOG_WALK_REPLAY_INODES 1 | |
dd8e7217 JB |
93 | #define LOG_WALK_REPLAY_DIR_INDEX 2 |
94 | #define LOG_WALK_REPLAY_ALL 3 | |
e02119d5 | 95 | |
12fcfd22 | 96 | static int btrfs_log_inode(struct btrfs_trans_handle *trans, |
49dae1bc FM |
97 | struct btrfs_root *root, struct inode *inode, |
98 | int inode_only, | |
99 | const loff_t start, | |
8407f553 FM |
100 | const loff_t end, |
101 | struct btrfs_log_ctx *ctx); | |
ec051c0f YZ |
102 | static int link_to_fixup_dir(struct btrfs_trans_handle *trans, |
103 | struct btrfs_root *root, | |
104 | struct btrfs_path *path, u64 objectid); | |
12fcfd22 CM |
105 | static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans, |
106 | struct btrfs_root *root, | |
107 | struct btrfs_root *log, | |
108 | struct btrfs_path *path, | |
109 | u64 dirid, int del_all); | |
e02119d5 CM |
110 | |
111 | /* | |
112 | * tree logging is a special write ahead log used to make sure that | |
113 | * fsyncs and O_SYNCs can happen without doing full tree commits. | |
114 | * | |
115 | * Full tree commits are expensive because they require commonly | |
116 | * modified blocks to be recowed, creating many dirty pages in the | |
117 | * extent tree an 4x-6x higher write load than ext3. | |
118 | * | |
119 | * Instead of doing a tree commit on every fsync, we use the | |
120 | * key ranges and transaction ids to find items for a given file or directory | |
121 | * that have changed in this transaction. Those items are copied into | |
122 | * a special tree (one per subvolume root), that tree is written to disk | |
123 | * and then the fsync is considered complete. | |
124 | * | |
125 | * After a crash, items are copied out of the log-tree back into the | |
126 | * subvolume tree. Any file data extents found are recorded in the extent | |
127 | * allocation tree, and the log-tree freed. | |
128 | * | |
129 | * The log tree is read three times, once to pin down all the extents it is | |
130 | * using in ram and once, once to create all the inodes logged in the tree | |
131 | * and once to do all the other items. | |
132 | */ | |
133 | ||
e02119d5 CM |
134 | /* |
135 | * start a sub transaction and setup the log tree | |
136 | * this increments the log tree writer count to make the people | |
137 | * syncing the tree wait for us to finish | |
138 | */ | |
139 | static int start_log_trans(struct btrfs_trans_handle *trans, | |
8b050d35 MX |
140 | struct btrfs_root *root, |
141 | struct btrfs_log_ctx *ctx) | |
e02119d5 | 142 | { |
8b050d35 | 143 | int index; |
e02119d5 | 144 | int ret; |
7237f183 YZ |
145 | |
146 | mutex_lock(&root->log_mutex); | |
147 | if (root->log_root) { | |
995946dd | 148 | if (btrfs_need_log_full_commit(root->fs_info, trans)) { |
50471a38 MX |
149 | ret = -EAGAIN; |
150 | goto out; | |
151 | } | |
ff782e0a JB |
152 | if (!root->log_start_pid) { |
153 | root->log_start_pid = current->pid; | |
27cdeb70 | 154 | clear_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state); |
ff782e0a | 155 | } else if (root->log_start_pid != current->pid) { |
27cdeb70 | 156 | set_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state); |
ff782e0a JB |
157 | } |
158 | ||
2ecb7923 | 159 | atomic_inc(&root->log_batch); |
7237f183 | 160 | atomic_inc(&root->log_writers); |
8b050d35 MX |
161 | if (ctx) { |
162 | index = root->log_transid % 2; | |
163 | list_add_tail(&ctx->list, &root->log_ctxs[index]); | |
d1433deb | 164 | ctx->log_transid = root->log_transid; |
8b050d35 | 165 | } |
7237f183 YZ |
166 | mutex_unlock(&root->log_mutex); |
167 | return 0; | |
168 | } | |
e87ac136 MX |
169 | |
170 | ret = 0; | |
e02119d5 | 171 | mutex_lock(&root->fs_info->tree_log_mutex); |
e87ac136 | 172 | if (!root->fs_info->log_root_tree) |
e02119d5 | 173 | ret = btrfs_init_log_root_tree(trans, root->fs_info); |
e87ac136 MX |
174 | mutex_unlock(&root->fs_info->tree_log_mutex); |
175 | if (ret) | |
176 | goto out; | |
177 | ||
178 | if (!root->log_root) { | |
e02119d5 | 179 | ret = btrfs_add_log_tree(trans, root); |
4a500fd1 | 180 | if (ret) |
e87ac136 | 181 | goto out; |
e02119d5 | 182 | } |
27cdeb70 | 183 | clear_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state); |
e87ac136 | 184 | root->log_start_pid = current->pid; |
2ecb7923 | 185 | atomic_inc(&root->log_batch); |
7237f183 | 186 | atomic_inc(&root->log_writers); |
8b050d35 MX |
187 | if (ctx) { |
188 | index = root->log_transid % 2; | |
189 | list_add_tail(&ctx->list, &root->log_ctxs[index]); | |
d1433deb | 190 | ctx->log_transid = root->log_transid; |
8b050d35 | 191 | } |
e87ac136 | 192 | out: |
7237f183 | 193 | mutex_unlock(&root->log_mutex); |
e87ac136 | 194 | return ret; |
e02119d5 CM |
195 | } |
196 | ||
197 | /* | |
198 | * returns 0 if there was a log transaction running and we were able | |
199 | * to join, or returns -ENOENT if there were not transactions | |
200 | * in progress | |
201 | */ | |
202 | static int join_running_log_trans(struct btrfs_root *root) | |
203 | { | |
204 | int ret = -ENOENT; | |
205 | ||
206 | smp_mb(); | |
207 | if (!root->log_root) | |
208 | return -ENOENT; | |
209 | ||
7237f183 | 210 | mutex_lock(&root->log_mutex); |
e02119d5 CM |
211 | if (root->log_root) { |
212 | ret = 0; | |
7237f183 | 213 | atomic_inc(&root->log_writers); |
e02119d5 | 214 | } |
7237f183 | 215 | mutex_unlock(&root->log_mutex); |
e02119d5 CM |
216 | return ret; |
217 | } | |
218 | ||
12fcfd22 CM |
219 | /* |
220 | * This either makes the current running log transaction wait | |
221 | * until you call btrfs_end_log_trans() or it makes any future | |
222 | * log transactions wait until you call btrfs_end_log_trans() | |
223 | */ | |
224 | int btrfs_pin_log_trans(struct btrfs_root *root) | |
225 | { | |
226 | int ret = -ENOENT; | |
227 | ||
228 | mutex_lock(&root->log_mutex); | |
229 | atomic_inc(&root->log_writers); | |
230 | mutex_unlock(&root->log_mutex); | |
231 | return ret; | |
232 | } | |
233 | ||
e02119d5 CM |
234 | /* |
235 | * indicate we're done making changes to the log tree | |
236 | * and wake up anyone waiting to do a sync | |
237 | */ | |
143bede5 | 238 | void btrfs_end_log_trans(struct btrfs_root *root) |
e02119d5 | 239 | { |
7237f183 YZ |
240 | if (atomic_dec_and_test(&root->log_writers)) { |
241 | smp_mb(); | |
242 | if (waitqueue_active(&root->log_writer_wait)) | |
243 | wake_up(&root->log_writer_wait); | |
244 | } | |
e02119d5 CM |
245 | } |
246 | ||
247 | ||
248 | /* | |
249 | * the walk control struct is used to pass state down the chain when | |
250 | * processing the log tree. The stage field tells us which part | |
251 | * of the log tree processing we are currently doing. The others | |
252 | * are state fields used for that specific part | |
253 | */ | |
254 | struct walk_control { | |
255 | /* should we free the extent on disk when done? This is used | |
256 | * at transaction commit time while freeing a log tree | |
257 | */ | |
258 | int free; | |
259 | ||
260 | /* should we write out the extent buffer? This is used | |
261 | * while flushing the log tree to disk during a sync | |
262 | */ | |
263 | int write; | |
264 | ||
265 | /* should we wait for the extent buffer io to finish? Also used | |
266 | * while flushing the log tree to disk for a sync | |
267 | */ | |
268 | int wait; | |
269 | ||
270 | /* pin only walk, we record which extents on disk belong to the | |
271 | * log trees | |
272 | */ | |
273 | int pin; | |
274 | ||
275 | /* what stage of the replay code we're currently in */ | |
276 | int stage; | |
277 | ||
278 | /* the root we are currently replaying */ | |
279 | struct btrfs_root *replay_dest; | |
280 | ||
281 | /* the trans handle for the current replay */ | |
282 | struct btrfs_trans_handle *trans; | |
283 | ||
284 | /* the function that gets used to process blocks we find in the | |
285 | * tree. Note the extent_buffer might not be up to date when it is | |
286 | * passed in, and it must be checked or read if you need the data | |
287 | * inside it | |
288 | */ | |
289 | int (*process_func)(struct btrfs_root *log, struct extent_buffer *eb, | |
290 | struct walk_control *wc, u64 gen); | |
291 | }; | |
292 | ||
293 | /* | |
294 | * process_func used to pin down extents, write them or wait on them | |
295 | */ | |
296 | static int process_one_buffer(struct btrfs_root *log, | |
297 | struct extent_buffer *eb, | |
298 | struct walk_control *wc, u64 gen) | |
299 | { | |
b50c6e25 JB |
300 | int ret = 0; |
301 | ||
8c2a1a30 JB |
302 | /* |
303 | * If this fs is mixed then we need to be able to process the leaves to | |
304 | * pin down any logged extents, so we have to read the block. | |
305 | */ | |
306 | if (btrfs_fs_incompat(log->fs_info, MIXED_GROUPS)) { | |
307 | ret = btrfs_read_buffer(eb, gen); | |
308 | if (ret) | |
309 | return ret; | |
310 | } | |
311 | ||
04018de5 | 312 | if (wc->pin) |
b50c6e25 JB |
313 | ret = btrfs_pin_extent_for_log_replay(log->fs_info->extent_root, |
314 | eb->start, eb->len); | |
e02119d5 | 315 | |
b50c6e25 | 316 | if (!ret && btrfs_buffer_uptodate(eb, gen, 0)) { |
8c2a1a30 JB |
317 | if (wc->pin && btrfs_header_level(eb) == 0) |
318 | ret = btrfs_exclude_logged_extents(log, eb); | |
e02119d5 CM |
319 | if (wc->write) |
320 | btrfs_write_tree_block(eb); | |
321 | if (wc->wait) | |
322 | btrfs_wait_tree_block_writeback(eb); | |
323 | } | |
b50c6e25 | 324 | return ret; |
e02119d5 CM |
325 | } |
326 | ||
327 | /* | |
328 | * Item overwrite used by replay and tree logging. eb, slot and key all refer | |
329 | * to the src data we are copying out. | |
330 | * | |
331 | * root is the tree we are copying into, and path is a scratch | |
332 | * path for use in this function (it should be released on entry and | |
333 | * will be released on exit). | |
334 | * | |
335 | * If the key is already in the destination tree the existing item is | |
336 | * overwritten. If the existing item isn't big enough, it is extended. | |
337 | * If it is too large, it is truncated. | |
338 | * | |
339 | * If the key isn't in the destination yet, a new item is inserted. | |
340 | */ | |
341 | static noinline int overwrite_item(struct btrfs_trans_handle *trans, | |
342 | struct btrfs_root *root, | |
343 | struct btrfs_path *path, | |
344 | struct extent_buffer *eb, int slot, | |
345 | struct btrfs_key *key) | |
346 | { | |
347 | int ret; | |
348 | u32 item_size; | |
349 | u64 saved_i_size = 0; | |
350 | int save_old_i_size = 0; | |
351 | unsigned long src_ptr; | |
352 | unsigned long dst_ptr; | |
353 | int overwrite_root = 0; | |
4bc4bee4 | 354 | bool inode_item = key->type == BTRFS_INODE_ITEM_KEY; |
e02119d5 CM |
355 | |
356 | if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) | |
357 | overwrite_root = 1; | |
358 | ||
359 | item_size = btrfs_item_size_nr(eb, slot); | |
360 | src_ptr = btrfs_item_ptr_offset(eb, slot); | |
361 | ||
362 | /* look for the key in the destination tree */ | |
363 | ret = btrfs_search_slot(NULL, root, key, path, 0, 0); | |
4bc4bee4 JB |
364 | if (ret < 0) |
365 | return ret; | |
366 | ||
e02119d5 CM |
367 | if (ret == 0) { |
368 | char *src_copy; | |
369 | char *dst_copy; | |
370 | u32 dst_size = btrfs_item_size_nr(path->nodes[0], | |
371 | path->slots[0]); | |
372 | if (dst_size != item_size) | |
373 | goto insert; | |
374 | ||
375 | if (item_size == 0) { | |
b3b4aa74 | 376 | btrfs_release_path(path); |
e02119d5 CM |
377 | return 0; |
378 | } | |
379 | dst_copy = kmalloc(item_size, GFP_NOFS); | |
380 | src_copy = kmalloc(item_size, GFP_NOFS); | |
2a29edc6 | 381 | if (!dst_copy || !src_copy) { |
b3b4aa74 | 382 | btrfs_release_path(path); |
2a29edc6 | 383 | kfree(dst_copy); |
384 | kfree(src_copy); | |
385 | return -ENOMEM; | |
386 | } | |
e02119d5 CM |
387 | |
388 | read_extent_buffer(eb, src_copy, src_ptr, item_size); | |
389 | ||
390 | dst_ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]); | |
391 | read_extent_buffer(path->nodes[0], dst_copy, dst_ptr, | |
392 | item_size); | |
393 | ret = memcmp(dst_copy, src_copy, item_size); | |
394 | ||
395 | kfree(dst_copy); | |
396 | kfree(src_copy); | |
397 | /* | |
398 | * they have the same contents, just return, this saves | |
399 | * us from cowing blocks in the destination tree and doing | |
400 | * extra writes that may not have been done by a previous | |
401 | * sync | |
402 | */ | |
403 | if (ret == 0) { | |
b3b4aa74 | 404 | btrfs_release_path(path); |
e02119d5 CM |
405 | return 0; |
406 | } | |
407 | ||
4bc4bee4 JB |
408 | /* |
409 | * We need to load the old nbytes into the inode so when we | |
410 | * replay the extents we've logged we get the right nbytes. | |
411 | */ | |
412 | if (inode_item) { | |
413 | struct btrfs_inode_item *item; | |
414 | u64 nbytes; | |
d555438b | 415 | u32 mode; |
4bc4bee4 JB |
416 | |
417 | item = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
418 | struct btrfs_inode_item); | |
419 | nbytes = btrfs_inode_nbytes(path->nodes[0], item); | |
420 | item = btrfs_item_ptr(eb, slot, | |
421 | struct btrfs_inode_item); | |
422 | btrfs_set_inode_nbytes(eb, item, nbytes); | |
d555438b JB |
423 | |
424 | /* | |
425 | * If this is a directory we need to reset the i_size to | |
426 | * 0 so that we can set it up properly when replaying | |
427 | * the rest of the items in this log. | |
428 | */ | |
429 | mode = btrfs_inode_mode(eb, item); | |
430 | if (S_ISDIR(mode)) | |
431 | btrfs_set_inode_size(eb, item, 0); | |
4bc4bee4 JB |
432 | } |
433 | } else if (inode_item) { | |
434 | struct btrfs_inode_item *item; | |
d555438b | 435 | u32 mode; |
4bc4bee4 JB |
436 | |
437 | /* | |
438 | * New inode, set nbytes to 0 so that the nbytes comes out | |
439 | * properly when we replay the extents. | |
440 | */ | |
441 | item = btrfs_item_ptr(eb, slot, struct btrfs_inode_item); | |
442 | btrfs_set_inode_nbytes(eb, item, 0); | |
d555438b JB |
443 | |
444 | /* | |
445 | * If this is a directory we need to reset the i_size to 0 so | |
446 | * that we can set it up properly when replaying the rest of | |
447 | * the items in this log. | |
448 | */ | |
449 | mode = btrfs_inode_mode(eb, item); | |
450 | if (S_ISDIR(mode)) | |
451 | btrfs_set_inode_size(eb, item, 0); | |
e02119d5 CM |
452 | } |
453 | insert: | |
b3b4aa74 | 454 | btrfs_release_path(path); |
e02119d5 CM |
455 | /* try to insert the key into the destination tree */ |
456 | ret = btrfs_insert_empty_item(trans, root, path, | |
457 | key, item_size); | |
458 | ||
459 | /* make sure any existing item is the correct size */ | |
460 | if (ret == -EEXIST) { | |
461 | u32 found_size; | |
462 | found_size = btrfs_item_size_nr(path->nodes[0], | |
463 | path->slots[0]); | |
143bede5 | 464 | if (found_size > item_size) |
afe5fea7 | 465 | btrfs_truncate_item(root, path, item_size, 1); |
143bede5 | 466 | else if (found_size < item_size) |
4b90c680 | 467 | btrfs_extend_item(root, path, |
143bede5 | 468 | item_size - found_size); |
e02119d5 | 469 | } else if (ret) { |
4a500fd1 | 470 | return ret; |
e02119d5 CM |
471 | } |
472 | dst_ptr = btrfs_item_ptr_offset(path->nodes[0], | |
473 | path->slots[0]); | |
474 | ||
475 | /* don't overwrite an existing inode if the generation number | |
476 | * was logged as zero. This is done when the tree logging code | |
477 | * is just logging an inode to make sure it exists after recovery. | |
478 | * | |
479 | * Also, don't overwrite i_size on directories during replay. | |
480 | * log replay inserts and removes directory items based on the | |
481 | * state of the tree found in the subvolume, and i_size is modified | |
482 | * as it goes | |
483 | */ | |
484 | if (key->type == BTRFS_INODE_ITEM_KEY && ret == -EEXIST) { | |
485 | struct btrfs_inode_item *src_item; | |
486 | struct btrfs_inode_item *dst_item; | |
487 | ||
488 | src_item = (struct btrfs_inode_item *)src_ptr; | |
489 | dst_item = (struct btrfs_inode_item *)dst_ptr; | |
490 | ||
491 | if (btrfs_inode_generation(eb, src_item) == 0) | |
492 | goto no_copy; | |
493 | ||
494 | if (overwrite_root && | |
495 | S_ISDIR(btrfs_inode_mode(eb, src_item)) && | |
496 | S_ISDIR(btrfs_inode_mode(path->nodes[0], dst_item))) { | |
497 | save_old_i_size = 1; | |
498 | saved_i_size = btrfs_inode_size(path->nodes[0], | |
499 | dst_item); | |
500 | } | |
501 | } | |
502 | ||
503 | copy_extent_buffer(path->nodes[0], eb, dst_ptr, | |
504 | src_ptr, item_size); | |
505 | ||
506 | if (save_old_i_size) { | |
507 | struct btrfs_inode_item *dst_item; | |
508 | dst_item = (struct btrfs_inode_item *)dst_ptr; | |
509 | btrfs_set_inode_size(path->nodes[0], dst_item, saved_i_size); | |
510 | } | |
511 | ||
512 | /* make sure the generation is filled in */ | |
513 | if (key->type == BTRFS_INODE_ITEM_KEY) { | |
514 | struct btrfs_inode_item *dst_item; | |
515 | dst_item = (struct btrfs_inode_item *)dst_ptr; | |
516 | if (btrfs_inode_generation(path->nodes[0], dst_item) == 0) { | |
517 | btrfs_set_inode_generation(path->nodes[0], dst_item, | |
518 | trans->transid); | |
519 | } | |
520 | } | |
521 | no_copy: | |
522 | btrfs_mark_buffer_dirty(path->nodes[0]); | |
b3b4aa74 | 523 | btrfs_release_path(path); |
e02119d5 CM |
524 | return 0; |
525 | } | |
526 | ||
527 | /* | |
528 | * simple helper to read an inode off the disk from a given root | |
529 | * This can only be called for subvolume roots and not for the log | |
530 | */ | |
531 | static noinline struct inode *read_one_inode(struct btrfs_root *root, | |
532 | u64 objectid) | |
533 | { | |
5d4f98a2 | 534 | struct btrfs_key key; |
e02119d5 | 535 | struct inode *inode; |
e02119d5 | 536 | |
5d4f98a2 YZ |
537 | key.objectid = objectid; |
538 | key.type = BTRFS_INODE_ITEM_KEY; | |
539 | key.offset = 0; | |
73f73415 | 540 | inode = btrfs_iget(root->fs_info->sb, &key, root, NULL); |
5d4f98a2 YZ |
541 | if (IS_ERR(inode)) { |
542 | inode = NULL; | |
543 | } else if (is_bad_inode(inode)) { | |
e02119d5 CM |
544 | iput(inode); |
545 | inode = NULL; | |
546 | } | |
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 | { | |
568 | int found_type; | |
e02119d5 | 569 | u64 extent_end; |
e02119d5 | 570 | u64 start = key->offset; |
4bc4bee4 | 571 | u64 nbytes = 0; |
e02119d5 CM |
572 | struct btrfs_file_extent_item *item; |
573 | struct inode *inode = NULL; | |
574 | unsigned long size; | |
575 | int ret = 0; | |
576 | ||
577 | item = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item); | |
578 | found_type = btrfs_file_extent_type(eb, item); | |
579 | ||
d899e052 | 580 | if (found_type == BTRFS_FILE_EXTENT_REG || |
4bc4bee4 JB |
581 | found_type == BTRFS_FILE_EXTENT_PREALLOC) { |
582 | nbytes = btrfs_file_extent_num_bytes(eb, item); | |
583 | extent_end = start + nbytes; | |
584 | ||
585 | /* | |
586 | * We don't add to the inodes nbytes if we are prealloc or a | |
587 | * hole. | |
588 | */ | |
589 | if (btrfs_file_extent_disk_bytenr(eb, item) == 0) | |
590 | nbytes = 0; | |
591 | } else if (found_type == BTRFS_FILE_EXTENT_INLINE) { | |
514ac8ad | 592 | size = btrfs_file_extent_inline_len(eb, slot, item); |
4bc4bee4 | 593 | nbytes = btrfs_file_extent_ram_bytes(eb, item); |
fda2832f | 594 | extent_end = ALIGN(start + size, root->sectorsize); |
e02119d5 CM |
595 | } else { |
596 | ret = 0; | |
597 | goto out; | |
598 | } | |
599 | ||
600 | inode = read_one_inode(root, key->objectid); | |
601 | if (!inode) { | |
602 | ret = -EIO; | |
603 | goto out; | |
604 | } | |
605 | ||
606 | /* | |
607 | * first check to see if we already have this extent in the | |
608 | * file. This must be done before the btrfs_drop_extents run | |
609 | * so we don't try to drop this extent. | |
610 | */ | |
33345d01 | 611 | ret = btrfs_lookup_file_extent(trans, root, path, btrfs_ino(inode), |
e02119d5 CM |
612 | start, 0); |
613 | ||
d899e052 YZ |
614 | if (ret == 0 && |
615 | (found_type == BTRFS_FILE_EXTENT_REG || | |
616 | found_type == BTRFS_FILE_EXTENT_PREALLOC)) { | |
e02119d5 CM |
617 | struct btrfs_file_extent_item cmp1; |
618 | struct btrfs_file_extent_item cmp2; | |
619 | struct btrfs_file_extent_item *existing; | |
620 | struct extent_buffer *leaf; | |
621 | ||
622 | leaf = path->nodes[0]; | |
623 | existing = btrfs_item_ptr(leaf, path->slots[0], | |
624 | struct btrfs_file_extent_item); | |
625 | ||
626 | read_extent_buffer(eb, &cmp1, (unsigned long)item, | |
627 | sizeof(cmp1)); | |
628 | read_extent_buffer(leaf, &cmp2, (unsigned long)existing, | |
629 | sizeof(cmp2)); | |
630 | ||
631 | /* | |
632 | * we already have a pointer to this exact extent, | |
633 | * we don't have to do anything | |
634 | */ | |
635 | if (memcmp(&cmp1, &cmp2, sizeof(cmp1)) == 0) { | |
b3b4aa74 | 636 | btrfs_release_path(path); |
e02119d5 CM |
637 | goto out; |
638 | } | |
639 | } | |
b3b4aa74 | 640 | btrfs_release_path(path); |
e02119d5 CM |
641 | |
642 | /* drop any overlapping extents */ | |
2671485d | 643 | ret = btrfs_drop_extents(trans, root, inode, start, extent_end, 1); |
3650860b JB |
644 | if (ret) |
645 | goto out; | |
e02119d5 | 646 | |
07d400a6 YZ |
647 | if (found_type == BTRFS_FILE_EXTENT_REG || |
648 | found_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
5d4f98a2 | 649 | u64 offset; |
07d400a6 YZ |
650 | unsigned long dest_offset; |
651 | struct btrfs_key ins; | |
652 | ||
653 | ret = btrfs_insert_empty_item(trans, root, path, key, | |
654 | sizeof(*item)); | |
3650860b JB |
655 | if (ret) |
656 | goto out; | |
07d400a6 YZ |
657 | dest_offset = btrfs_item_ptr_offset(path->nodes[0], |
658 | path->slots[0]); | |
659 | copy_extent_buffer(path->nodes[0], eb, dest_offset, | |
660 | (unsigned long)item, sizeof(*item)); | |
661 | ||
662 | ins.objectid = btrfs_file_extent_disk_bytenr(eb, item); | |
663 | ins.offset = btrfs_file_extent_disk_num_bytes(eb, item); | |
664 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
5d4f98a2 | 665 | offset = key->offset - btrfs_file_extent_offset(eb, item); |
07d400a6 YZ |
666 | |
667 | if (ins.objectid > 0) { | |
668 | u64 csum_start; | |
669 | u64 csum_end; | |
670 | LIST_HEAD(ordered_sums); | |
671 | /* | |
672 | * is this extent already allocated in the extent | |
673 | * allocation tree? If so, just add a reference | |
674 | */ | |
1a4ed8fd | 675 | ret = btrfs_lookup_data_extent(root, ins.objectid, |
07d400a6 YZ |
676 | ins.offset); |
677 | if (ret == 0) { | |
678 | ret = btrfs_inc_extent_ref(trans, root, | |
679 | ins.objectid, ins.offset, | |
5d4f98a2 | 680 | 0, root->root_key.objectid, |
66d7e7f0 | 681 | key->objectid, offset, 0); |
b50c6e25 JB |
682 | if (ret) |
683 | goto out; | |
07d400a6 YZ |
684 | } else { |
685 | /* | |
686 | * insert the extent pointer in the extent | |
687 | * allocation tree | |
688 | */ | |
5d4f98a2 YZ |
689 | ret = btrfs_alloc_logged_file_extent(trans, |
690 | root, root->root_key.objectid, | |
691 | key->objectid, offset, &ins); | |
b50c6e25 JB |
692 | if (ret) |
693 | goto out; | |
07d400a6 | 694 | } |
b3b4aa74 | 695 | btrfs_release_path(path); |
07d400a6 YZ |
696 | |
697 | if (btrfs_file_extent_compression(eb, item)) { | |
698 | csum_start = ins.objectid; | |
699 | csum_end = csum_start + ins.offset; | |
700 | } else { | |
701 | csum_start = ins.objectid + | |
702 | btrfs_file_extent_offset(eb, item); | |
703 | csum_end = csum_start + | |
704 | btrfs_file_extent_num_bytes(eb, item); | |
705 | } | |
706 | ||
707 | ret = btrfs_lookup_csums_range(root->log_root, | |
708 | csum_start, csum_end - 1, | |
a2de733c | 709 | &ordered_sums, 0); |
3650860b JB |
710 | if (ret) |
711 | goto out; | |
07d400a6 YZ |
712 | while (!list_empty(&ordered_sums)) { |
713 | struct btrfs_ordered_sum *sums; | |
714 | sums = list_entry(ordered_sums.next, | |
715 | struct btrfs_ordered_sum, | |
716 | list); | |
3650860b JB |
717 | if (!ret) |
718 | ret = btrfs_csum_file_blocks(trans, | |
07d400a6 YZ |
719 | root->fs_info->csum_root, |
720 | sums); | |
07d400a6 YZ |
721 | list_del(&sums->list); |
722 | kfree(sums); | |
723 | } | |
3650860b JB |
724 | if (ret) |
725 | goto out; | |
07d400a6 | 726 | } else { |
b3b4aa74 | 727 | btrfs_release_path(path); |
07d400a6 YZ |
728 | } |
729 | } else if (found_type == BTRFS_FILE_EXTENT_INLINE) { | |
730 | /* inline extents are easy, we just overwrite them */ | |
731 | ret = overwrite_item(trans, root, path, eb, slot, key); | |
3650860b JB |
732 | if (ret) |
733 | goto out; | |
07d400a6 | 734 | } |
e02119d5 | 735 | |
4bc4bee4 | 736 | inode_add_bytes(inode, nbytes); |
b9959295 | 737 | ret = btrfs_update_inode(trans, root, inode); |
e02119d5 CM |
738 | out: |
739 | if (inode) | |
740 | iput(inode); | |
741 | return ret; | |
742 | } | |
743 | ||
744 | /* | |
745 | * when cleaning up conflicts between the directory names in the | |
746 | * subvolume, directory names in the log and directory names in the | |
747 | * inode back references, we may have to unlink inodes from directories. | |
748 | * | |
749 | * This is a helper function to do the unlink of a specific directory | |
750 | * item | |
751 | */ | |
752 | static noinline int drop_one_dir_item(struct btrfs_trans_handle *trans, | |
753 | struct btrfs_root *root, | |
754 | struct btrfs_path *path, | |
755 | struct inode *dir, | |
756 | struct btrfs_dir_item *di) | |
757 | { | |
758 | struct inode *inode; | |
759 | char *name; | |
760 | int name_len; | |
761 | struct extent_buffer *leaf; | |
762 | struct btrfs_key location; | |
763 | int ret; | |
764 | ||
765 | leaf = path->nodes[0]; | |
766 | ||
767 | btrfs_dir_item_key_to_cpu(leaf, di, &location); | |
768 | name_len = btrfs_dir_name_len(leaf, di); | |
769 | name = kmalloc(name_len, GFP_NOFS); | |
2a29edc6 | 770 | if (!name) |
771 | return -ENOMEM; | |
772 | ||
e02119d5 | 773 | read_extent_buffer(leaf, name, (unsigned long)(di + 1), name_len); |
b3b4aa74 | 774 | btrfs_release_path(path); |
e02119d5 CM |
775 | |
776 | inode = read_one_inode(root, location.objectid); | |
c00e9493 | 777 | if (!inode) { |
3650860b JB |
778 | ret = -EIO; |
779 | goto out; | |
c00e9493 | 780 | } |
e02119d5 | 781 | |
ec051c0f | 782 | ret = link_to_fixup_dir(trans, root, path, location.objectid); |
3650860b JB |
783 | if (ret) |
784 | goto out; | |
12fcfd22 | 785 | |
e02119d5 | 786 | ret = btrfs_unlink_inode(trans, root, dir, inode, name, name_len); |
3650860b JB |
787 | if (ret) |
788 | goto out; | |
ada9af21 FDBM |
789 | else |
790 | ret = btrfs_run_delayed_items(trans, root); | |
3650860b | 791 | out: |
e02119d5 | 792 | kfree(name); |
e02119d5 CM |
793 | iput(inode); |
794 | return ret; | |
795 | } | |
796 | ||
797 | /* | |
798 | * helper function to see if a given name and sequence number found | |
799 | * in an inode back reference are already in a directory and correctly | |
800 | * point to this inode | |
801 | */ | |
802 | static noinline int inode_in_dir(struct btrfs_root *root, | |
803 | struct btrfs_path *path, | |
804 | u64 dirid, u64 objectid, u64 index, | |
805 | const char *name, int name_len) | |
806 | { | |
807 | struct btrfs_dir_item *di; | |
808 | struct btrfs_key location; | |
809 | int match = 0; | |
810 | ||
811 | di = btrfs_lookup_dir_index_item(NULL, root, path, dirid, | |
812 | index, name, name_len, 0); | |
813 | if (di && !IS_ERR(di)) { | |
814 | btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location); | |
815 | if (location.objectid != objectid) | |
816 | goto out; | |
817 | } else | |
818 | goto out; | |
b3b4aa74 | 819 | btrfs_release_path(path); |
e02119d5 CM |
820 | |
821 | di = btrfs_lookup_dir_item(NULL, root, path, dirid, name, name_len, 0); | |
822 | if (di && !IS_ERR(di)) { | |
823 | btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location); | |
824 | if (location.objectid != objectid) | |
825 | goto out; | |
826 | } else | |
827 | goto out; | |
828 | match = 1; | |
829 | out: | |
b3b4aa74 | 830 | btrfs_release_path(path); |
e02119d5 CM |
831 | return match; |
832 | } | |
833 | ||
834 | /* | |
835 | * helper function to check a log tree for a named back reference in | |
836 | * an inode. This is used to decide if a back reference that is | |
837 | * found in the subvolume conflicts with what we find in the log. | |
838 | * | |
839 | * inode backreferences may have multiple refs in a single item, | |
840 | * during replay we process one reference at a time, and we don't | |
841 | * want to delete valid links to a file from the subvolume if that | |
842 | * link is also in the log. | |
843 | */ | |
844 | static noinline int backref_in_log(struct btrfs_root *log, | |
845 | struct btrfs_key *key, | |
f186373f | 846 | u64 ref_objectid, |
e02119d5 CM |
847 | char *name, int namelen) |
848 | { | |
849 | struct btrfs_path *path; | |
850 | struct btrfs_inode_ref *ref; | |
851 | unsigned long ptr; | |
852 | unsigned long ptr_end; | |
853 | unsigned long name_ptr; | |
854 | int found_name_len; | |
855 | int item_size; | |
856 | int ret; | |
857 | int match = 0; | |
858 | ||
859 | path = btrfs_alloc_path(); | |
2a29edc6 | 860 | if (!path) |
861 | return -ENOMEM; | |
862 | ||
e02119d5 CM |
863 | ret = btrfs_search_slot(NULL, log, key, path, 0, 0); |
864 | if (ret != 0) | |
865 | goto out; | |
866 | ||
e02119d5 | 867 | ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]); |
f186373f MF |
868 | |
869 | if (key->type == BTRFS_INODE_EXTREF_KEY) { | |
870 | if (btrfs_find_name_in_ext_backref(path, ref_objectid, | |
871 | name, namelen, NULL)) | |
872 | match = 1; | |
873 | ||
874 | goto out; | |
875 | } | |
876 | ||
877 | item_size = btrfs_item_size_nr(path->nodes[0], path->slots[0]); | |
e02119d5 CM |
878 | ptr_end = ptr + item_size; |
879 | while (ptr < ptr_end) { | |
880 | ref = (struct btrfs_inode_ref *)ptr; | |
881 | found_name_len = btrfs_inode_ref_name_len(path->nodes[0], ref); | |
882 | if (found_name_len == namelen) { | |
883 | name_ptr = (unsigned long)(ref + 1); | |
884 | ret = memcmp_extent_buffer(path->nodes[0], name, | |
885 | name_ptr, namelen); | |
886 | if (ret == 0) { | |
887 | match = 1; | |
888 | goto out; | |
889 | } | |
890 | } | |
891 | ptr = (unsigned long)(ref + 1) + found_name_len; | |
892 | } | |
893 | out: | |
894 | btrfs_free_path(path); | |
895 | return match; | |
896 | } | |
897 | ||
5a1d7843 | 898 | static inline int __add_inode_ref(struct btrfs_trans_handle *trans, |
e02119d5 | 899 | struct btrfs_root *root, |
e02119d5 | 900 | struct btrfs_path *path, |
5a1d7843 JS |
901 | struct btrfs_root *log_root, |
902 | struct inode *dir, struct inode *inode, | |
5a1d7843 | 903 | struct extent_buffer *eb, |
f186373f MF |
904 | u64 inode_objectid, u64 parent_objectid, |
905 | u64 ref_index, char *name, int namelen, | |
906 | int *search_done) | |
e02119d5 | 907 | { |
34f3e4f2 | 908 | int ret; |
f186373f MF |
909 | char *victim_name; |
910 | int victim_name_len; | |
911 | struct extent_buffer *leaf; | |
5a1d7843 | 912 | struct btrfs_dir_item *di; |
f186373f MF |
913 | struct btrfs_key search_key; |
914 | struct btrfs_inode_extref *extref; | |
c622ae60 | 915 | |
f186373f MF |
916 | again: |
917 | /* Search old style refs */ | |
918 | search_key.objectid = inode_objectid; | |
919 | search_key.type = BTRFS_INODE_REF_KEY; | |
920 | search_key.offset = parent_objectid; | |
921 | ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0); | |
e02119d5 | 922 | if (ret == 0) { |
e02119d5 CM |
923 | struct btrfs_inode_ref *victim_ref; |
924 | unsigned long ptr; | |
925 | unsigned long ptr_end; | |
f186373f MF |
926 | |
927 | leaf = path->nodes[0]; | |
e02119d5 CM |
928 | |
929 | /* are we trying to overwrite a back ref for the root directory | |
930 | * if so, just jump out, we're done | |
931 | */ | |
f186373f | 932 | if (search_key.objectid == search_key.offset) |
5a1d7843 | 933 | return 1; |
e02119d5 CM |
934 | |
935 | /* check all the names in this back reference to see | |
936 | * if they are in the log. if so, we allow them to stay | |
937 | * otherwise they must be unlinked as a conflict | |
938 | */ | |
939 | ptr = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
940 | ptr_end = ptr + btrfs_item_size_nr(leaf, path->slots[0]); | |
d397712b | 941 | while (ptr < ptr_end) { |
e02119d5 CM |
942 | victim_ref = (struct btrfs_inode_ref *)ptr; |
943 | victim_name_len = btrfs_inode_ref_name_len(leaf, | |
944 | victim_ref); | |
945 | victim_name = kmalloc(victim_name_len, GFP_NOFS); | |
3650860b JB |
946 | if (!victim_name) |
947 | return -ENOMEM; | |
e02119d5 CM |
948 | |
949 | read_extent_buffer(leaf, victim_name, | |
950 | (unsigned long)(victim_ref + 1), | |
951 | victim_name_len); | |
952 | ||
f186373f MF |
953 | if (!backref_in_log(log_root, &search_key, |
954 | parent_objectid, | |
955 | victim_name, | |
e02119d5 | 956 | victim_name_len)) { |
8b558c5f | 957 | inc_nlink(inode); |
b3b4aa74 | 958 | btrfs_release_path(path); |
12fcfd22 | 959 | |
e02119d5 CM |
960 | ret = btrfs_unlink_inode(trans, root, dir, |
961 | inode, victim_name, | |
962 | victim_name_len); | |
f186373f | 963 | kfree(victim_name); |
3650860b JB |
964 | if (ret) |
965 | return ret; | |
ada9af21 FDBM |
966 | ret = btrfs_run_delayed_items(trans, root); |
967 | if (ret) | |
968 | return ret; | |
f186373f MF |
969 | *search_done = 1; |
970 | goto again; | |
e02119d5 CM |
971 | } |
972 | kfree(victim_name); | |
f186373f | 973 | |
e02119d5 CM |
974 | ptr = (unsigned long)(victim_ref + 1) + victim_name_len; |
975 | } | |
e02119d5 | 976 | |
c622ae60 | 977 | /* |
978 | * NOTE: we have searched root tree and checked the | |
979 | * coresponding ref, it does not need to check again. | |
980 | */ | |
5a1d7843 | 981 | *search_done = 1; |
e02119d5 | 982 | } |
b3b4aa74 | 983 | btrfs_release_path(path); |
e02119d5 | 984 | |
f186373f MF |
985 | /* Same search but for extended refs */ |
986 | extref = btrfs_lookup_inode_extref(NULL, root, path, name, namelen, | |
987 | inode_objectid, parent_objectid, 0, | |
988 | 0); | |
989 | if (!IS_ERR_OR_NULL(extref)) { | |
990 | u32 item_size; | |
991 | u32 cur_offset = 0; | |
992 | unsigned long base; | |
993 | struct inode *victim_parent; | |
994 | ||
995 | leaf = path->nodes[0]; | |
996 | ||
997 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
998 | base = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
999 | ||
1000 | while (cur_offset < item_size) { | |
1001 | extref = (struct btrfs_inode_extref *)base + cur_offset; | |
1002 | ||
1003 | victim_name_len = btrfs_inode_extref_name_len(leaf, extref); | |
1004 | ||
1005 | if (btrfs_inode_extref_parent(leaf, extref) != parent_objectid) | |
1006 | goto next; | |
1007 | ||
1008 | victim_name = kmalloc(victim_name_len, GFP_NOFS); | |
3650860b JB |
1009 | if (!victim_name) |
1010 | return -ENOMEM; | |
f186373f MF |
1011 | read_extent_buffer(leaf, victim_name, (unsigned long)&extref->name, |
1012 | victim_name_len); | |
1013 | ||
1014 | search_key.objectid = inode_objectid; | |
1015 | search_key.type = BTRFS_INODE_EXTREF_KEY; | |
1016 | search_key.offset = btrfs_extref_hash(parent_objectid, | |
1017 | victim_name, | |
1018 | victim_name_len); | |
1019 | ret = 0; | |
1020 | if (!backref_in_log(log_root, &search_key, | |
1021 | parent_objectid, victim_name, | |
1022 | victim_name_len)) { | |
1023 | ret = -ENOENT; | |
1024 | victim_parent = read_one_inode(root, | |
1025 | parent_objectid); | |
1026 | if (victim_parent) { | |
8b558c5f | 1027 | inc_nlink(inode); |
f186373f MF |
1028 | btrfs_release_path(path); |
1029 | ||
1030 | ret = btrfs_unlink_inode(trans, root, | |
1031 | victim_parent, | |
1032 | inode, | |
1033 | victim_name, | |
1034 | victim_name_len); | |
ada9af21 FDBM |
1035 | if (!ret) |
1036 | ret = btrfs_run_delayed_items( | |
1037 | trans, root); | |
f186373f | 1038 | } |
f186373f MF |
1039 | iput(victim_parent); |
1040 | kfree(victim_name); | |
3650860b JB |
1041 | if (ret) |
1042 | return ret; | |
f186373f MF |
1043 | *search_done = 1; |
1044 | goto again; | |
1045 | } | |
1046 | kfree(victim_name); | |
3650860b JB |
1047 | if (ret) |
1048 | return ret; | |
f186373f MF |
1049 | next: |
1050 | cur_offset += victim_name_len + sizeof(*extref); | |
1051 | } | |
1052 | *search_done = 1; | |
1053 | } | |
1054 | btrfs_release_path(path); | |
1055 | ||
34f3e4f2 | 1056 | /* look for a conflicting sequence number */ |
1057 | di = btrfs_lookup_dir_index_item(trans, root, path, btrfs_ino(dir), | |
f186373f | 1058 | ref_index, name, namelen, 0); |
34f3e4f2 | 1059 | if (di && !IS_ERR(di)) { |
1060 | ret = drop_one_dir_item(trans, root, path, dir, di); | |
3650860b JB |
1061 | if (ret) |
1062 | return ret; | |
34f3e4f2 | 1063 | } |
1064 | btrfs_release_path(path); | |
1065 | ||
1066 | /* look for a conflicing name */ | |
1067 | di = btrfs_lookup_dir_item(trans, root, path, btrfs_ino(dir), | |
1068 | name, namelen, 0); | |
1069 | if (di && !IS_ERR(di)) { | |
1070 | ret = drop_one_dir_item(trans, root, path, dir, di); | |
3650860b JB |
1071 | if (ret) |
1072 | return ret; | |
34f3e4f2 | 1073 | } |
1074 | btrfs_release_path(path); | |
1075 | ||
5a1d7843 JS |
1076 | return 0; |
1077 | } | |
e02119d5 | 1078 | |
f186373f MF |
1079 | static int extref_get_fields(struct extent_buffer *eb, unsigned long ref_ptr, |
1080 | u32 *namelen, char **name, u64 *index, | |
1081 | u64 *parent_objectid) | |
1082 | { | |
1083 | struct btrfs_inode_extref *extref; | |
1084 | ||
1085 | extref = (struct btrfs_inode_extref *)ref_ptr; | |
1086 | ||
1087 | *namelen = btrfs_inode_extref_name_len(eb, extref); | |
1088 | *name = kmalloc(*namelen, GFP_NOFS); | |
1089 | if (*name == NULL) | |
1090 | return -ENOMEM; | |
1091 | ||
1092 | read_extent_buffer(eb, *name, (unsigned long)&extref->name, | |
1093 | *namelen); | |
1094 | ||
1095 | *index = btrfs_inode_extref_index(eb, extref); | |
1096 | if (parent_objectid) | |
1097 | *parent_objectid = btrfs_inode_extref_parent(eb, extref); | |
1098 | ||
1099 | return 0; | |
1100 | } | |
1101 | ||
1102 | static int ref_get_fields(struct extent_buffer *eb, unsigned long ref_ptr, | |
1103 | u32 *namelen, char **name, u64 *index) | |
1104 | { | |
1105 | struct btrfs_inode_ref *ref; | |
1106 | ||
1107 | ref = (struct btrfs_inode_ref *)ref_ptr; | |
1108 | ||
1109 | *namelen = btrfs_inode_ref_name_len(eb, ref); | |
1110 | *name = kmalloc(*namelen, GFP_NOFS); | |
1111 | if (*name == NULL) | |
1112 | return -ENOMEM; | |
1113 | ||
1114 | read_extent_buffer(eb, *name, (unsigned long)(ref + 1), *namelen); | |
1115 | ||
1116 | *index = btrfs_inode_ref_index(eb, ref); | |
1117 | ||
1118 | return 0; | |
1119 | } | |
1120 | ||
5a1d7843 JS |
1121 | /* |
1122 | * replay one inode back reference item found in the log tree. | |
1123 | * eb, slot and key refer to the buffer and key found in the log tree. | |
1124 | * root is the destination we are replaying into, and path is for temp | |
1125 | * use by this function. (it should be released on return). | |
1126 | */ | |
1127 | static noinline int add_inode_ref(struct btrfs_trans_handle *trans, | |
1128 | struct btrfs_root *root, | |
1129 | struct btrfs_root *log, | |
1130 | struct btrfs_path *path, | |
1131 | struct extent_buffer *eb, int slot, | |
1132 | struct btrfs_key *key) | |
1133 | { | |
03b2f08b GB |
1134 | struct inode *dir = NULL; |
1135 | struct inode *inode = NULL; | |
5a1d7843 JS |
1136 | unsigned long ref_ptr; |
1137 | unsigned long ref_end; | |
03b2f08b | 1138 | char *name = NULL; |
5a1d7843 JS |
1139 | int namelen; |
1140 | int ret; | |
1141 | int search_done = 0; | |
f186373f MF |
1142 | int log_ref_ver = 0; |
1143 | u64 parent_objectid; | |
1144 | u64 inode_objectid; | |
f46dbe3d | 1145 | u64 ref_index = 0; |
f186373f MF |
1146 | int ref_struct_size; |
1147 | ||
1148 | ref_ptr = btrfs_item_ptr_offset(eb, slot); | |
1149 | ref_end = ref_ptr + btrfs_item_size_nr(eb, slot); | |
1150 | ||
1151 | if (key->type == BTRFS_INODE_EXTREF_KEY) { | |
1152 | struct btrfs_inode_extref *r; | |
1153 | ||
1154 | ref_struct_size = sizeof(struct btrfs_inode_extref); | |
1155 | log_ref_ver = 1; | |
1156 | r = (struct btrfs_inode_extref *)ref_ptr; | |
1157 | parent_objectid = btrfs_inode_extref_parent(eb, r); | |
1158 | } else { | |
1159 | ref_struct_size = sizeof(struct btrfs_inode_ref); | |
1160 | parent_objectid = key->offset; | |
1161 | } | |
1162 | inode_objectid = key->objectid; | |
e02119d5 | 1163 | |
5a1d7843 JS |
1164 | /* |
1165 | * it is possible that we didn't log all the parent directories | |
1166 | * for a given inode. If we don't find the dir, just don't | |
1167 | * copy the back ref in. The link count fixup code will take | |
1168 | * care of the rest | |
1169 | */ | |
f186373f | 1170 | dir = read_one_inode(root, parent_objectid); |
03b2f08b GB |
1171 | if (!dir) { |
1172 | ret = -ENOENT; | |
1173 | goto out; | |
1174 | } | |
5a1d7843 | 1175 | |
f186373f | 1176 | inode = read_one_inode(root, inode_objectid); |
5a1d7843 | 1177 | if (!inode) { |
03b2f08b GB |
1178 | ret = -EIO; |
1179 | goto out; | |
5a1d7843 JS |
1180 | } |
1181 | ||
5a1d7843 | 1182 | while (ref_ptr < ref_end) { |
f186373f MF |
1183 | if (log_ref_ver) { |
1184 | ret = extref_get_fields(eb, ref_ptr, &namelen, &name, | |
1185 | &ref_index, &parent_objectid); | |
1186 | /* | |
1187 | * parent object can change from one array | |
1188 | * item to another. | |
1189 | */ | |
1190 | if (!dir) | |
1191 | dir = read_one_inode(root, parent_objectid); | |
03b2f08b GB |
1192 | if (!dir) { |
1193 | ret = -ENOENT; | |
1194 | goto out; | |
1195 | } | |
f186373f MF |
1196 | } else { |
1197 | ret = ref_get_fields(eb, ref_ptr, &namelen, &name, | |
1198 | &ref_index); | |
1199 | } | |
1200 | if (ret) | |
03b2f08b | 1201 | goto out; |
5a1d7843 JS |
1202 | |
1203 | /* if we already have a perfect match, we're done */ | |
1204 | if (!inode_in_dir(root, path, btrfs_ino(dir), btrfs_ino(inode), | |
f186373f | 1205 | ref_index, name, namelen)) { |
5a1d7843 JS |
1206 | /* |
1207 | * look for a conflicting back reference in the | |
1208 | * metadata. if we find one we have to unlink that name | |
1209 | * of the file before we add our new link. Later on, we | |
1210 | * overwrite any existing back reference, and we don't | |
1211 | * want to create dangling pointers in the directory. | |
1212 | */ | |
1213 | ||
1214 | if (!search_done) { | |
1215 | ret = __add_inode_ref(trans, root, path, log, | |
f186373f MF |
1216 | dir, inode, eb, |
1217 | inode_objectid, | |
1218 | parent_objectid, | |
1219 | ref_index, name, namelen, | |
5a1d7843 | 1220 | &search_done); |
03b2f08b GB |
1221 | if (ret) { |
1222 | if (ret == 1) | |
1223 | ret = 0; | |
3650860b JB |
1224 | goto out; |
1225 | } | |
5a1d7843 JS |
1226 | } |
1227 | ||
1228 | /* insert our name */ | |
1229 | ret = btrfs_add_link(trans, dir, inode, name, namelen, | |
f186373f | 1230 | 0, ref_index); |
3650860b JB |
1231 | if (ret) |
1232 | goto out; | |
5a1d7843 JS |
1233 | |
1234 | btrfs_update_inode(trans, root, inode); | |
1235 | } | |
1236 | ||
f186373f | 1237 | ref_ptr = (unsigned long)(ref_ptr + ref_struct_size) + namelen; |
5a1d7843 | 1238 | kfree(name); |
03b2f08b | 1239 | name = NULL; |
f186373f MF |
1240 | if (log_ref_ver) { |
1241 | iput(dir); | |
1242 | dir = NULL; | |
1243 | } | |
5a1d7843 | 1244 | } |
e02119d5 CM |
1245 | |
1246 | /* finally write the back reference in the inode */ | |
1247 | ret = overwrite_item(trans, root, path, eb, slot, key); | |
5a1d7843 | 1248 | out: |
b3b4aa74 | 1249 | btrfs_release_path(path); |
03b2f08b | 1250 | kfree(name); |
e02119d5 CM |
1251 | iput(dir); |
1252 | iput(inode); | |
3650860b | 1253 | return ret; |
e02119d5 CM |
1254 | } |
1255 | ||
c71bf099 YZ |
1256 | static int insert_orphan_item(struct btrfs_trans_handle *trans, |
1257 | struct btrfs_root *root, u64 offset) | |
1258 | { | |
1259 | int ret; | |
3f870c28 KN |
1260 | ret = btrfs_find_item(root, NULL, BTRFS_ORPHAN_OBJECTID, |
1261 | offset, BTRFS_ORPHAN_ITEM_KEY, NULL); | |
c71bf099 YZ |
1262 | if (ret > 0) |
1263 | ret = btrfs_insert_orphan_item(trans, root, offset); | |
1264 | return ret; | |
1265 | } | |
1266 | ||
f186373f MF |
1267 | static int count_inode_extrefs(struct btrfs_root *root, |
1268 | struct inode *inode, struct btrfs_path *path) | |
1269 | { | |
1270 | int ret = 0; | |
1271 | int name_len; | |
1272 | unsigned int nlink = 0; | |
1273 | u32 item_size; | |
1274 | u32 cur_offset = 0; | |
1275 | u64 inode_objectid = btrfs_ino(inode); | |
1276 | u64 offset = 0; | |
1277 | unsigned long ptr; | |
1278 | struct btrfs_inode_extref *extref; | |
1279 | struct extent_buffer *leaf; | |
1280 | ||
1281 | while (1) { | |
1282 | ret = btrfs_find_one_extref(root, inode_objectid, offset, path, | |
1283 | &extref, &offset); | |
1284 | if (ret) | |
1285 | break; | |
c71bf099 | 1286 | |
f186373f MF |
1287 | leaf = path->nodes[0]; |
1288 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1289 | ptr = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
1290 | ||
1291 | while (cur_offset < item_size) { | |
1292 | extref = (struct btrfs_inode_extref *) (ptr + cur_offset); | |
1293 | name_len = btrfs_inode_extref_name_len(leaf, extref); | |
1294 | ||
1295 | nlink++; | |
1296 | ||
1297 | cur_offset += name_len + sizeof(*extref); | |
1298 | } | |
1299 | ||
1300 | offset++; | |
1301 | btrfs_release_path(path); | |
1302 | } | |
1303 | btrfs_release_path(path); | |
1304 | ||
1305 | if (ret < 0) | |
1306 | return ret; | |
1307 | return nlink; | |
1308 | } | |
1309 | ||
1310 | static int count_inode_refs(struct btrfs_root *root, | |
1311 | struct inode *inode, struct btrfs_path *path) | |
e02119d5 | 1312 | { |
e02119d5 CM |
1313 | int ret; |
1314 | struct btrfs_key key; | |
f186373f | 1315 | unsigned int nlink = 0; |
e02119d5 CM |
1316 | unsigned long ptr; |
1317 | unsigned long ptr_end; | |
1318 | int name_len; | |
33345d01 | 1319 | u64 ino = btrfs_ino(inode); |
e02119d5 | 1320 | |
33345d01 | 1321 | key.objectid = ino; |
e02119d5 CM |
1322 | key.type = BTRFS_INODE_REF_KEY; |
1323 | key.offset = (u64)-1; | |
1324 | ||
d397712b | 1325 | while (1) { |
e02119d5 CM |
1326 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
1327 | if (ret < 0) | |
1328 | break; | |
1329 | if (ret > 0) { | |
1330 | if (path->slots[0] == 0) | |
1331 | break; | |
1332 | path->slots[0]--; | |
1333 | } | |
e93ae26f | 1334 | process_slot: |
e02119d5 CM |
1335 | btrfs_item_key_to_cpu(path->nodes[0], &key, |
1336 | path->slots[0]); | |
33345d01 | 1337 | if (key.objectid != ino || |
e02119d5 CM |
1338 | key.type != BTRFS_INODE_REF_KEY) |
1339 | break; | |
1340 | ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]); | |
1341 | ptr_end = ptr + btrfs_item_size_nr(path->nodes[0], | |
1342 | path->slots[0]); | |
d397712b | 1343 | while (ptr < ptr_end) { |
e02119d5 CM |
1344 | struct btrfs_inode_ref *ref; |
1345 | ||
1346 | ref = (struct btrfs_inode_ref *)ptr; | |
1347 | name_len = btrfs_inode_ref_name_len(path->nodes[0], | |
1348 | ref); | |
1349 | ptr = (unsigned long)(ref + 1) + name_len; | |
1350 | nlink++; | |
1351 | } | |
1352 | ||
1353 | if (key.offset == 0) | |
1354 | break; | |
e93ae26f FDBM |
1355 | if (path->slots[0] > 0) { |
1356 | path->slots[0]--; | |
1357 | goto process_slot; | |
1358 | } | |
e02119d5 | 1359 | key.offset--; |
b3b4aa74 | 1360 | btrfs_release_path(path); |
e02119d5 | 1361 | } |
b3b4aa74 | 1362 | btrfs_release_path(path); |
f186373f MF |
1363 | |
1364 | return nlink; | |
1365 | } | |
1366 | ||
1367 | /* | |
1368 | * There are a few corners where the link count of the file can't | |
1369 | * be properly maintained during replay. So, instead of adding | |
1370 | * lots of complexity to the log code, we just scan the backrefs | |
1371 | * for any file that has been through replay. | |
1372 | * | |
1373 | * The scan will update the link count on the inode to reflect the | |
1374 | * number of back refs found. If it goes down to zero, the iput | |
1375 | * will free the inode. | |
1376 | */ | |
1377 | static noinline int fixup_inode_link_count(struct btrfs_trans_handle *trans, | |
1378 | struct btrfs_root *root, | |
1379 | struct inode *inode) | |
1380 | { | |
1381 | struct btrfs_path *path; | |
1382 | int ret; | |
1383 | u64 nlink = 0; | |
1384 | u64 ino = btrfs_ino(inode); | |
1385 | ||
1386 | path = btrfs_alloc_path(); | |
1387 | if (!path) | |
1388 | return -ENOMEM; | |
1389 | ||
1390 | ret = count_inode_refs(root, inode, path); | |
1391 | if (ret < 0) | |
1392 | goto out; | |
1393 | ||
1394 | nlink = ret; | |
1395 | ||
1396 | ret = count_inode_extrefs(root, inode, path); | |
1397 | if (ret == -ENOENT) | |
1398 | ret = 0; | |
1399 | ||
1400 | if (ret < 0) | |
1401 | goto out; | |
1402 | ||
1403 | nlink += ret; | |
1404 | ||
1405 | ret = 0; | |
1406 | ||
e02119d5 | 1407 | if (nlink != inode->i_nlink) { |
bfe86848 | 1408 | set_nlink(inode, nlink); |
e02119d5 CM |
1409 | btrfs_update_inode(trans, root, inode); |
1410 | } | |
8d5bf1cb | 1411 | BTRFS_I(inode)->index_cnt = (u64)-1; |
e02119d5 | 1412 | |
c71bf099 YZ |
1413 | if (inode->i_nlink == 0) { |
1414 | if (S_ISDIR(inode->i_mode)) { | |
1415 | ret = replay_dir_deletes(trans, root, NULL, path, | |
33345d01 | 1416 | ino, 1); |
3650860b JB |
1417 | if (ret) |
1418 | goto out; | |
c71bf099 | 1419 | } |
33345d01 | 1420 | ret = insert_orphan_item(trans, root, ino); |
12fcfd22 | 1421 | } |
12fcfd22 | 1422 | |
f186373f MF |
1423 | out: |
1424 | btrfs_free_path(path); | |
1425 | return ret; | |
e02119d5 CM |
1426 | } |
1427 | ||
1428 | static noinline int fixup_inode_link_counts(struct btrfs_trans_handle *trans, | |
1429 | struct btrfs_root *root, | |
1430 | struct btrfs_path *path) | |
1431 | { | |
1432 | int ret; | |
1433 | struct btrfs_key key; | |
1434 | struct inode *inode; | |
1435 | ||
1436 | key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID; | |
1437 | key.type = BTRFS_ORPHAN_ITEM_KEY; | |
1438 | key.offset = (u64)-1; | |
d397712b | 1439 | while (1) { |
e02119d5 CM |
1440 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
1441 | if (ret < 0) | |
1442 | break; | |
1443 | ||
1444 | if (ret == 1) { | |
1445 | if (path->slots[0] == 0) | |
1446 | break; | |
1447 | path->slots[0]--; | |
1448 | } | |
1449 | ||
1450 | btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]); | |
1451 | if (key.objectid != BTRFS_TREE_LOG_FIXUP_OBJECTID || | |
1452 | key.type != BTRFS_ORPHAN_ITEM_KEY) | |
1453 | break; | |
1454 | ||
1455 | ret = btrfs_del_item(trans, root, path); | |
65a246c5 TI |
1456 | if (ret) |
1457 | goto out; | |
e02119d5 | 1458 | |
b3b4aa74 | 1459 | btrfs_release_path(path); |
e02119d5 | 1460 | inode = read_one_inode(root, key.offset); |
c00e9493 TI |
1461 | if (!inode) |
1462 | return -EIO; | |
e02119d5 CM |
1463 | |
1464 | ret = fixup_inode_link_count(trans, root, inode); | |
e02119d5 | 1465 | iput(inode); |
3650860b JB |
1466 | if (ret) |
1467 | goto out; | |
e02119d5 | 1468 | |
12fcfd22 CM |
1469 | /* |
1470 | * fixup on a directory may create new entries, | |
1471 | * make sure we always look for the highset possible | |
1472 | * offset | |
1473 | */ | |
1474 | key.offset = (u64)-1; | |
e02119d5 | 1475 | } |
65a246c5 TI |
1476 | ret = 0; |
1477 | out: | |
b3b4aa74 | 1478 | btrfs_release_path(path); |
65a246c5 | 1479 | return ret; |
e02119d5 CM |
1480 | } |
1481 | ||
1482 | ||
1483 | /* | |
1484 | * record a given inode in the fixup dir so we can check its link | |
1485 | * count when replay is done. The link count is incremented here | |
1486 | * so the inode won't go away until we check it | |
1487 | */ | |
1488 | static noinline int link_to_fixup_dir(struct btrfs_trans_handle *trans, | |
1489 | struct btrfs_root *root, | |
1490 | struct btrfs_path *path, | |
1491 | u64 objectid) | |
1492 | { | |
1493 | struct btrfs_key key; | |
1494 | int ret = 0; | |
1495 | struct inode *inode; | |
1496 | ||
1497 | inode = read_one_inode(root, objectid); | |
c00e9493 TI |
1498 | if (!inode) |
1499 | return -EIO; | |
e02119d5 CM |
1500 | |
1501 | key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID; | |
962a298f | 1502 | key.type = BTRFS_ORPHAN_ITEM_KEY; |
e02119d5 CM |
1503 | key.offset = objectid; |
1504 | ||
1505 | ret = btrfs_insert_empty_item(trans, root, path, &key, 0); | |
1506 | ||
b3b4aa74 | 1507 | btrfs_release_path(path); |
e02119d5 | 1508 | if (ret == 0) { |
9bf7a489 JB |
1509 | if (!inode->i_nlink) |
1510 | set_nlink(inode, 1); | |
1511 | else | |
8b558c5f | 1512 | inc_nlink(inode); |
b9959295 | 1513 | ret = btrfs_update_inode(trans, root, inode); |
e02119d5 CM |
1514 | } else if (ret == -EEXIST) { |
1515 | ret = 0; | |
1516 | } else { | |
3650860b | 1517 | BUG(); /* Logic Error */ |
e02119d5 CM |
1518 | } |
1519 | iput(inode); | |
1520 | ||
1521 | return ret; | |
1522 | } | |
1523 | ||
1524 | /* | |
1525 | * when replaying the log for a directory, we only insert names | |
1526 | * for inodes that actually exist. This means an fsync on a directory | |
1527 | * does not implicitly fsync all the new files in it | |
1528 | */ | |
1529 | static noinline int insert_one_name(struct btrfs_trans_handle *trans, | |
1530 | struct btrfs_root *root, | |
1531 | struct btrfs_path *path, | |
1532 | u64 dirid, u64 index, | |
1533 | char *name, int name_len, u8 type, | |
1534 | struct btrfs_key *location) | |
1535 | { | |
1536 | struct inode *inode; | |
1537 | struct inode *dir; | |
1538 | int ret; | |
1539 | ||
1540 | inode = read_one_inode(root, location->objectid); | |
1541 | if (!inode) | |
1542 | return -ENOENT; | |
1543 | ||
1544 | dir = read_one_inode(root, dirid); | |
1545 | if (!dir) { | |
1546 | iput(inode); | |
1547 | return -EIO; | |
1548 | } | |
d555438b | 1549 | |
e02119d5 CM |
1550 | ret = btrfs_add_link(trans, dir, inode, name, name_len, 1, index); |
1551 | ||
1552 | /* FIXME, put inode into FIXUP list */ | |
1553 | ||
1554 | iput(inode); | |
1555 | iput(dir); | |
1556 | return ret; | |
1557 | } | |
1558 | ||
1559 | /* | |
1560 | * take a single entry in a log directory item and replay it into | |
1561 | * the subvolume. | |
1562 | * | |
1563 | * if a conflicting item exists in the subdirectory already, | |
1564 | * the inode it points to is unlinked and put into the link count | |
1565 | * fix up tree. | |
1566 | * | |
1567 | * If a name from the log points to a file or directory that does | |
1568 | * not exist in the FS, it is skipped. fsyncs on directories | |
1569 | * do not force down inodes inside that directory, just changes to the | |
1570 | * names or unlinks in a directory. | |
1571 | */ | |
1572 | static noinline int replay_one_name(struct btrfs_trans_handle *trans, | |
1573 | struct btrfs_root *root, | |
1574 | struct btrfs_path *path, | |
1575 | struct extent_buffer *eb, | |
1576 | struct btrfs_dir_item *di, | |
1577 | struct btrfs_key *key) | |
1578 | { | |
1579 | char *name; | |
1580 | int name_len; | |
1581 | struct btrfs_dir_item *dst_di; | |
1582 | struct btrfs_key found_key; | |
1583 | struct btrfs_key log_key; | |
1584 | struct inode *dir; | |
e02119d5 | 1585 | u8 log_type; |
4bef0848 | 1586 | int exists; |
3650860b | 1587 | int ret = 0; |
d555438b | 1588 | bool update_size = (key->type == BTRFS_DIR_INDEX_KEY); |
e02119d5 CM |
1589 | |
1590 | dir = read_one_inode(root, key->objectid); | |
c00e9493 TI |
1591 | if (!dir) |
1592 | return -EIO; | |
e02119d5 CM |
1593 | |
1594 | name_len = btrfs_dir_name_len(eb, di); | |
1595 | name = kmalloc(name_len, GFP_NOFS); | |
2bac325e FDBM |
1596 | if (!name) { |
1597 | ret = -ENOMEM; | |
1598 | goto out; | |
1599 | } | |
2a29edc6 | 1600 | |
e02119d5 CM |
1601 | log_type = btrfs_dir_type(eb, di); |
1602 | read_extent_buffer(eb, name, (unsigned long)(di + 1), | |
1603 | name_len); | |
1604 | ||
1605 | btrfs_dir_item_key_to_cpu(eb, di, &log_key); | |
4bef0848 CM |
1606 | exists = btrfs_lookup_inode(trans, root, path, &log_key, 0); |
1607 | if (exists == 0) | |
1608 | exists = 1; | |
1609 | else | |
1610 | exists = 0; | |
b3b4aa74 | 1611 | btrfs_release_path(path); |
4bef0848 | 1612 | |
e02119d5 CM |
1613 | if (key->type == BTRFS_DIR_ITEM_KEY) { |
1614 | dst_di = btrfs_lookup_dir_item(trans, root, path, key->objectid, | |
1615 | name, name_len, 1); | |
d397712b | 1616 | } else if (key->type == BTRFS_DIR_INDEX_KEY) { |
e02119d5 CM |
1617 | dst_di = btrfs_lookup_dir_index_item(trans, root, path, |
1618 | key->objectid, | |
1619 | key->offset, name, | |
1620 | name_len, 1); | |
1621 | } else { | |
3650860b JB |
1622 | /* Corruption */ |
1623 | ret = -EINVAL; | |
1624 | goto out; | |
e02119d5 | 1625 | } |
c704005d | 1626 | if (IS_ERR_OR_NULL(dst_di)) { |
e02119d5 CM |
1627 | /* we need a sequence number to insert, so we only |
1628 | * do inserts for the BTRFS_DIR_INDEX_KEY types | |
1629 | */ | |
1630 | if (key->type != BTRFS_DIR_INDEX_KEY) | |
1631 | goto out; | |
1632 | goto insert; | |
1633 | } | |
1634 | ||
1635 | btrfs_dir_item_key_to_cpu(path->nodes[0], dst_di, &found_key); | |
1636 | /* the existing item matches the logged item */ | |
1637 | if (found_key.objectid == log_key.objectid && | |
1638 | found_key.type == log_key.type && | |
1639 | found_key.offset == log_key.offset && | |
1640 | btrfs_dir_type(path->nodes[0], dst_di) == log_type) { | |
a2cc11db | 1641 | update_size = false; |
e02119d5 CM |
1642 | goto out; |
1643 | } | |
1644 | ||
1645 | /* | |
1646 | * don't drop the conflicting directory entry if the inode | |
1647 | * for the new entry doesn't exist | |
1648 | */ | |
4bef0848 | 1649 | if (!exists) |
e02119d5 CM |
1650 | goto out; |
1651 | ||
e02119d5 | 1652 | ret = drop_one_dir_item(trans, root, path, dir, dst_di); |
3650860b JB |
1653 | if (ret) |
1654 | goto out; | |
e02119d5 CM |
1655 | |
1656 | if (key->type == BTRFS_DIR_INDEX_KEY) | |
1657 | goto insert; | |
1658 | out: | |
b3b4aa74 | 1659 | btrfs_release_path(path); |
d555438b JB |
1660 | if (!ret && update_size) { |
1661 | btrfs_i_size_write(dir, dir->i_size + name_len * 2); | |
1662 | ret = btrfs_update_inode(trans, root, dir); | |
1663 | } | |
e02119d5 CM |
1664 | kfree(name); |
1665 | iput(dir); | |
3650860b | 1666 | return ret; |
e02119d5 CM |
1667 | |
1668 | insert: | |
b3b4aa74 | 1669 | btrfs_release_path(path); |
e02119d5 CM |
1670 | ret = insert_one_name(trans, root, path, key->objectid, key->offset, |
1671 | name, name_len, log_type, &log_key); | |
3650860b JB |
1672 | if (ret && ret != -ENOENT) |
1673 | goto out; | |
d555438b | 1674 | update_size = false; |
3650860b | 1675 | ret = 0; |
e02119d5 CM |
1676 | goto out; |
1677 | } | |
1678 | ||
1679 | /* | |
1680 | * find all the names in a directory item and reconcile them into | |
1681 | * the subvolume. Only BTRFS_DIR_ITEM_KEY types will have more than | |
1682 | * one name in a directory item, but the same code gets used for | |
1683 | * both directory index types | |
1684 | */ | |
1685 | static noinline int replay_one_dir_item(struct btrfs_trans_handle *trans, | |
1686 | struct btrfs_root *root, | |
1687 | struct btrfs_path *path, | |
1688 | struct extent_buffer *eb, int slot, | |
1689 | struct btrfs_key *key) | |
1690 | { | |
1691 | int ret; | |
1692 | u32 item_size = btrfs_item_size_nr(eb, slot); | |
1693 | struct btrfs_dir_item *di; | |
1694 | int name_len; | |
1695 | unsigned long ptr; | |
1696 | unsigned long ptr_end; | |
1697 | ||
1698 | ptr = btrfs_item_ptr_offset(eb, slot); | |
1699 | ptr_end = ptr + item_size; | |
d397712b | 1700 | while (ptr < ptr_end) { |
e02119d5 | 1701 | di = (struct btrfs_dir_item *)ptr; |
22a94d44 JB |
1702 | if (verify_dir_item(root, eb, di)) |
1703 | return -EIO; | |
e02119d5 CM |
1704 | name_len = btrfs_dir_name_len(eb, di); |
1705 | ret = replay_one_name(trans, root, path, eb, di, key); | |
3650860b JB |
1706 | if (ret) |
1707 | return ret; | |
e02119d5 CM |
1708 | ptr = (unsigned long)(di + 1); |
1709 | ptr += name_len; | |
1710 | } | |
1711 | return 0; | |
1712 | } | |
1713 | ||
1714 | /* | |
1715 | * directory replay has two parts. There are the standard directory | |
1716 | * items in the log copied from the subvolume, and range items | |
1717 | * created in the log while the subvolume was logged. | |
1718 | * | |
1719 | * The range items tell us which parts of the key space the log | |
1720 | * is authoritative for. During replay, if a key in the subvolume | |
1721 | * directory is in a logged range item, but not actually in the log | |
1722 | * that means it was deleted from the directory before the fsync | |
1723 | * and should be removed. | |
1724 | */ | |
1725 | static noinline int find_dir_range(struct btrfs_root *root, | |
1726 | struct btrfs_path *path, | |
1727 | u64 dirid, int key_type, | |
1728 | u64 *start_ret, u64 *end_ret) | |
1729 | { | |
1730 | struct btrfs_key key; | |
1731 | u64 found_end; | |
1732 | struct btrfs_dir_log_item *item; | |
1733 | int ret; | |
1734 | int nritems; | |
1735 | ||
1736 | if (*start_ret == (u64)-1) | |
1737 | return 1; | |
1738 | ||
1739 | key.objectid = dirid; | |
1740 | key.type = key_type; | |
1741 | key.offset = *start_ret; | |
1742 | ||
1743 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
1744 | if (ret < 0) | |
1745 | goto out; | |
1746 | if (ret > 0) { | |
1747 | if (path->slots[0] == 0) | |
1748 | goto out; | |
1749 | path->slots[0]--; | |
1750 | } | |
1751 | if (ret != 0) | |
1752 | btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]); | |
1753 | ||
1754 | if (key.type != key_type || key.objectid != dirid) { | |
1755 | ret = 1; | |
1756 | goto next; | |
1757 | } | |
1758 | item = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
1759 | struct btrfs_dir_log_item); | |
1760 | found_end = btrfs_dir_log_end(path->nodes[0], item); | |
1761 | ||
1762 | if (*start_ret >= key.offset && *start_ret <= found_end) { | |
1763 | ret = 0; | |
1764 | *start_ret = key.offset; | |
1765 | *end_ret = found_end; | |
1766 | goto out; | |
1767 | } | |
1768 | ret = 1; | |
1769 | next: | |
1770 | /* check the next slot in the tree to see if it is a valid item */ | |
1771 | nritems = btrfs_header_nritems(path->nodes[0]); | |
1772 | if (path->slots[0] >= nritems) { | |
1773 | ret = btrfs_next_leaf(root, path); | |
1774 | if (ret) | |
1775 | goto out; | |
1776 | } else { | |
1777 | path->slots[0]++; | |
1778 | } | |
1779 | ||
1780 | btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]); | |
1781 | ||
1782 | if (key.type != key_type || key.objectid != dirid) { | |
1783 | ret = 1; | |
1784 | goto out; | |
1785 | } | |
1786 | item = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
1787 | struct btrfs_dir_log_item); | |
1788 | found_end = btrfs_dir_log_end(path->nodes[0], item); | |
1789 | *start_ret = key.offset; | |
1790 | *end_ret = found_end; | |
1791 | ret = 0; | |
1792 | out: | |
b3b4aa74 | 1793 | btrfs_release_path(path); |
e02119d5 CM |
1794 | return ret; |
1795 | } | |
1796 | ||
1797 | /* | |
1798 | * this looks for a given directory item in the log. If the directory | |
1799 | * item is not in the log, the item is removed and the inode it points | |
1800 | * to is unlinked | |
1801 | */ | |
1802 | static noinline int check_item_in_log(struct btrfs_trans_handle *trans, | |
1803 | struct btrfs_root *root, | |
1804 | struct btrfs_root *log, | |
1805 | struct btrfs_path *path, | |
1806 | struct btrfs_path *log_path, | |
1807 | struct inode *dir, | |
1808 | struct btrfs_key *dir_key) | |
1809 | { | |
1810 | int ret; | |
1811 | struct extent_buffer *eb; | |
1812 | int slot; | |
1813 | u32 item_size; | |
1814 | struct btrfs_dir_item *di; | |
1815 | struct btrfs_dir_item *log_di; | |
1816 | int name_len; | |
1817 | unsigned long ptr; | |
1818 | unsigned long ptr_end; | |
1819 | char *name; | |
1820 | struct inode *inode; | |
1821 | struct btrfs_key location; | |
1822 | ||
1823 | again: | |
1824 | eb = path->nodes[0]; | |
1825 | slot = path->slots[0]; | |
1826 | item_size = btrfs_item_size_nr(eb, slot); | |
1827 | ptr = btrfs_item_ptr_offset(eb, slot); | |
1828 | ptr_end = ptr + item_size; | |
d397712b | 1829 | while (ptr < ptr_end) { |
e02119d5 | 1830 | di = (struct btrfs_dir_item *)ptr; |
22a94d44 JB |
1831 | if (verify_dir_item(root, eb, di)) { |
1832 | ret = -EIO; | |
1833 | goto out; | |
1834 | } | |
1835 | ||
e02119d5 CM |
1836 | name_len = btrfs_dir_name_len(eb, di); |
1837 | name = kmalloc(name_len, GFP_NOFS); | |
1838 | if (!name) { | |
1839 | ret = -ENOMEM; | |
1840 | goto out; | |
1841 | } | |
1842 | read_extent_buffer(eb, name, (unsigned long)(di + 1), | |
1843 | name_len); | |
1844 | log_di = NULL; | |
12fcfd22 | 1845 | if (log && dir_key->type == BTRFS_DIR_ITEM_KEY) { |
e02119d5 CM |
1846 | log_di = btrfs_lookup_dir_item(trans, log, log_path, |
1847 | dir_key->objectid, | |
1848 | name, name_len, 0); | |
12fcfd22 | 1849 | } else if (log && dir_key->type == BTRFS_DIR_INDEX_KEY) { |
e02119d5 CM |
1850 | log_di = btrfs_lookup_dir_index_item(trans, log, |
1851 | log_path, | |
1852 | dir_key->objectid, | |
1853 | dir_key->offset, | |
1854 | name, name_len, 0); | |
1855 | } | |
269d040f | 1856 | if (!log_di || (IS_ERR(log_di) && PTR_ERR(log_di) == -ENOENT)) { |
e02119d5 | 1857 | btrfs_dir_item_key_to_cpu(eb, di, &location); |
b3b4aa74 DS |
1858 | btrfs_release_path(path); |
1859 | btrfs_release_path(log_path); | |
e02119d5 | 1860 | inode = read_one_inode(root, location.objectid); |
c00e9493 TI |
1861 | if (!inode) { |
1862 | kfree(name); | |
1863 | return -EIO; | |
1864 | } | |
e02119d5 CM |
1865 | |
1866 | ret = link_to_fixup_dir(trans, root, | |
1867 | path, location.objectid); | |
3650860b JB |
1868 | if (ret) { |
1869 | kfree(name); | |
1870 | iput(inode); | |
1871 | goto out; | |
1872 | } | |
1873 | ||
8b558c5f | 1874 | inc_nlink(inode); |
e02119d5 CM |
1875 | ret = btrfs_unlink_inode(trans, root, dir, inode, |
1876 | name, name_len); | |
3650860b | 1877 | if (!ret) |
ada9af21 | 1878 | ret = btrfs_run_delayed_items(trans, root); |
e02119d5 CM |
1879 | kfree(name); |
1880 | iput(inode); | |
3650860b JB |
1881 | if (ret) |
1882 | goto out; | |
e02119d5 CM |
1883 | |
1884 | /* there might still be more names under this key | |
1885 | * check and repeat if required | |
1886 | */ | |
1887 | ret = btrfs_search_slot(NULL, root, dir_key, path, | |
1888 | 0, 0); | |
1889 | if (ret == 0) | |
1890 | goto again; | |
1891 | ret = 0; | |
1892 | goto out; | |
269d040f FDBM |
1893 | } else if (IS_ERR(log_di)) { |
1894 | kfree(name); | |
1895 | return PTR_ERR(log_di); | |
e02119d5 | 1896 | } |
b3b4aa74 | 1897 | btrfs_release_path(log_path); |
e02119d5 CM |
1898 | kfree(name); |
1899 | ||
1900 | ptr = (unsigned long)(di + 1); | |
1901 | ptr += name_len; | |
1902 | } | |
1903 | ret = 0; | |
1904 | out: | |
b3b4aa74 DS |
1905 | btrfs_release_path(path); |
1906 | btrfs_release_path(log_path); | |
e02119d5 CM |
1907 | return ret; |
1908 | } | |
1909 | ||
1910 | /* | |
1911 | * deletion replay happens before we copy any new directory items | |
1912 | * out of the log or out of backreferences from inodes. It | |
1913 | * scans the log to find ranges of keys that log is authoritative for, | |
1914 | * and then scans the directory to find items in those ranges that are | |
1915 | * not present in the log. | |
1916 | * | |
1917 | * Anything we don't find in the log is unlinked and removed from the | |
1918 | * directory. | |
1919 | */ | |
1920 | static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans, | |
1921 | struct btrfs_root *root, | |
1922 | struct btrfs_root *log, | |
1923 | struct btrfs_path *path, | |
12fcfd22 | 1924 | u64 dirid, int del_all) |
e02119d5 CM |
1925 | { |
1926 | u64 range_start; | |
1927 | u64 range_end; | |
1928 | int key_type = BTRFS_DIR_LOG_ITEM_KEY; | |
1929 | int ret = 0; | |
1930 | struct btrfs_key dir_key; | |
1931 | struct btrfs_key found_key; | |
1932 | struct btrfs_path *log_path; | |
1933 | struct inode *dir; | |
1934 | ||
1935 | dir_key.objectid = dirid; | |
1936 | dir_key.type = BTRFS_DIR_ITEM_KEY; | |
1937 | log_path = btrfs_alloc_path(); | |
1938 | if (!log_path) | |
1939 | return -ENOMEM; | |
1940 | ||
1941 | dir = read_one_inode(root, dirid); | |
1942 | /* it isn't an error if the inode isn't there, that can happen | |
1943 | * because we replay the deletes before we copy in the inode item | |
1944 | * from the log | |
1945 | */ | |
1946 | if (!dir) { | |
1947 | btrfs_free_path(log_path); | |
1948 | return 0; | |
1949 | } | |
1950 | again: | |
1951 | range_start = 0; | |
1952 | range_end = 0; | |
d397712b | 1953 | while (1) { |
12fcfd22 CM |
1954 | if (del_all) |
1955 | range_end = (u64)-1; | |
1956 | else { | |
1957 | ret = find_dir_range(log, path, dirid, key_type, | |
1958 | &range_start, &range_end); | |
1959 | if (ret != 0) | |
1960 | break; | |
1961 | } | |
e02119d5 CM |
1962 | |
1963 | dir_key.offset = range_start; | |
d397712b | 1964 | while (1) { |
e02119d5 CM |
1965 | int nritems; |
1966 | ret = btrfs_search_slot(NULL, root, &dir_key, path, | |
1967 | 0, 0); | |
1968 | if (ret < 0) | |
1969 | goto out; | |
1970 | ||
1971 | nritems = btrfs_header_nritems(path->nodes[0]); | |
1972 | if (path->slots[0] >= nritems) { | |
1973 | ret = btrfs_next_leaf(root, path); | |
1974 | if (ret) | |
1975 | break; | |
1976 | } | |
1977 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, | |
1978 | path->slots[0]); | |
1979 | if (found_key.objectid != dirid || | |
1980 | found_key.type != dir_key.type) | |
1981 | goto next_type; | |
1982 | ||
1983 | if (found_key.offset > range_end) | |
1984 | break; | |
1985 | ||
1986 | ret = check_item_in_log(trans, root, log, path, | |
12fcfd22 CM |
1987 | log_path, dir, |
1988 | &found_key); | |
3650860b JB |
1989 | if (ret) |
1990 | goto out; | |
e02119d5 CM |
1991 | if (found_key.offset == (u64)-1) |
1992 | break; | |
1993 | dir_key.offset = found_key.offset + 1; | |
1994 | } | |
b3b4aa74 | 1995 | btrfs_release_path(path); |
e02119d5 CM |
1996 | if (range_end == (u64)-1) |
1997 | break; | |
1998 | range_start = range_end + 1; | |
1999 | } | |
2000 | ||
2001 | next_type: | |
2002 | ret = 0; | |
2003 | if (key_type == BTRFS_DIR_LOG_ITEM_KEY) { | |
2004 | key_type = BTRFS_DIR_LOG_INDEX_KEY; | |
2005 | dir_key.type = BTRFS_DIR_INDEX_KEY; | |
b3b4aa74 | 2006 | btrfs_release_path(path); |
e02119d5 CM |
2007 | goto again; |
2008 | } | |
2009 | out: | |
b3b4aa74 | 2010 | btrfs_release_path(path); |
e02119d5 CM |
2011 | btrfs_free_path(log_path); |
2012 | iput(dir); | |
2013 | return ret; | |
2014 | } | |
2015 | ||
2016 | /* | |
2017 | * the process_func used to replay items from the log tree. This | |
2018 | * gets called in two different stages. The first stage just looks | |
2019 | * for inodes and makes sure they are all copied into the subvolume. | |
2020 | * | |
2021 | * The second stage copies all the other item types from the log into | |
2022 | * the subvolume. The two stage approach is slower, but gets rid of | |
2023 | * lots of complexity around inodes referencing other inodes that exist | |
2024 | * only in the log (references come from either directory items or inode | |
2025 | * back refs). | |
2026 | */ | |
2027 | static int replay_one_buffer(struct btrfs_root *log, struct extent_buffer *eb, | |
2028 | struct walk_control *wc, u64 gen) | |
2029 | { | |
2030 | int nritems; | |
2031 | struct btrfs_path *path; | |
2032 | struct btrfs_root *root = wc->replay_dest; | |
2033 | struct btrfs_key key; | |
e02119d5 CM |
2034 | int level; |
2035 | int i; | |
2036 | int ret; | |
2037 | ||
018642a1 TI |
2038 | ret = btrfs_read_buffer(eb, gen); |
2039 | if (ret) | |
2040 | return ret; | |
e02119d5 CM |
2041 | |
2042 | level = btrfs_header_level(eb); | |
2043 | ||
2044 | if (level != 0) | |
2045 | return 0; | |
2046 | ||
2047 | path = btrfs_alloc_path(); | |
1e5063d0 MF |
2048 | if (!path) |
2049 | return -ENOMEM; | |
e02119d5 CM |
2050 | |
2051 | nritems = btrfs_header_nritems(eb); | |
2052 | for (i = 0; i < nritems; i++) { | |
2053 | btrfs_item_key_to_cpu(eb, &key, i); | |
e02119d5 CM |
2054 | |
2055 | /* inode keys are done during the first stage */ | |
2056 | if (key.type == BTRFS_INODE_ITEM_KEY && | |
2057 | wc->stage == LOG_WALK_REPLAY_INODES) { | |
e02119d5 CM |
2058 | struct btrfs_inode_item *inode_item; |
2059 | u32 mode; | |
2060 | ||
2061 | inode_item = btrfs_item_ptr(eb, i, | |
2062 | struct btrfs_inode_item); | |
2063 | mode = btrfs_inode_mode(eb, inode_item); | |
2064 | if (S_ISDIR(mode)) { | |
2065 | ret = replay_dir_deletes(wc->trans, | |
12fcfd22 | 2066 | root, log, path, key.objectid, 0); |
b50c6e25 JB |
2067 | if (ret) |
2068 | break; | |
e02119d5 CM |
2069 | } |
2070 | ret = overwrite_item(wc->trans, root, path, | |
2071 | eb, i, &key); | |
b50c6e25 JB |
2072 | if (ret) |
2073 | break; | |
e02119d5 | 2074 | |
c71bf099 YZ |
2075 | /* for regular files, make sure corresponding |
2076 | * orhpan item exist. extents past the new EOF | |
2077 | * will be truncated later by orphan cleanup. | |
e02119d5 CM |
2078 | */ |
2079 | if (S_ISREG(mode)) { | |
c71bf099 YZ |
2080 | ret = insert_orphan_item(wc->trans, root, |
2081 | key.objectid); | |
b50c6e25 JB |
2082 | if (ret) |
2083 | break; | |
e02119d5 | 2084 | } |
c71bf099 | 2085 | |
e02119d5 CM |
2086 | ret = link_to_fixup_dir(wc->trans, root, |
2087 | path, key.objectid); | |
b50c6e25 JB |
2088 | if (ret) |
2089 | break; | |
e02119d5 | 2090 | } |
dd8e7217 JB |
2091 | |
2092 | if (key.type == BTRFS_DIR_INDEX_KEY && | |
2093 | wc->stage == LOG_WALK_REPLAY_DIR_INDEX) { | |
2094 | ret = replay_one_dir_item(wc->trans, root, path, | |
2095 | eb, i, &key); | |
2096 | if (ret) | |
2097 | break; | |
2098 | } | |
2099 | ||
e02119d5 CM |
2100 | if (wc->stage < LOG_WALK_REPLAY_ALL) |
2101 | continue; | |
2102 | ||
2103 | /* these keys are simply copied */ | |
2104 | if (key.type == BTRFS_XATTR_ITEM_KEY) { | |
2105 | ret = overwrite_item(wc->trans, root, path, | |
2106 | eb, i, &key); | |
b50c6e25 JB |
2107 | if (ret) |
2108 | break; | |
2da1c669 LB |
2109 | } else if (key.type == BTRFS_INODE_REF_KEY || |
2110 | key.type == BTRFS_INODE_EXTREF_KEY) { | |
f186373f MF |
2111 | ret = add_inode_ref(wc->trans, root, log, path, |
2112 | eb, i, &key); | |
b50c6e25 JB |
2113 | if (ret && ret != -ENOENT) |
2114 | break; | |
2115 | ret = 0; | |
e02119d5 CM |
2116 | } else if (key.type == BTRFS_EXTENT_DATA_KEY) { |
2117 | ret = replay_one_extent(wc->trans, root, path, | |
2118 | eb, i, &key); | |
b50c6e25 JB |
2119 | if (ret) |
2120 | break; | |
dd8e7217 | 2121 | } else if (key.type == BTRFS_DIR_ITEM_KEY) { |
e02119d5 CM |
2122 | ret = replay_one_dir_item(wc->trans, root, path, |
2123 | eb, i, &key); | |
b50c6e25 JB |
2124 | if (ret) |
2125 | break; | |
e02119d5 CM |
2126 | } |
2127 | } | |
2128 | btrfs_free_path(path); | |
b50c6e25 | 2129 | return ret; |
e02119d5 CM |
2130 | } |
2131 | ||
d397712b | 2132 | static noinline int walk_down_log_tree(struct btrfs_trans_handle *trans, |
e02119d5 CM |
2133 | struct btrfs_root *root, |
2134 | struct btrfs_path *path, int *level, | |
2135 | struct walk_control *wc) | |
2136 | { | |
2137 | u64 root_owner; | |
e02119d5 CM |
2138 | u64 bytenr; |
2139 | u64 ptr_gen; | |
2140 | struct extent_buffer *next; | |
2141 | struct extent_buffer *cur; | |
2142 | struct extent_buffer *parent; | |
2143 | u32 blocksize; | |
2144 | int ret = 0; | |
2145 | ||
2146 | WARN_ON(*level < 0); | |
2147 | WARN_ON(*level >= BTRFS_MAX_LEVEL); | |
2148 | ||
d397712b | 2149 | while (*level > 0) { |
e02119d5 CM |
2150 | WARN_ON(*level < 0); |
2151 | WARN_ON(*level >= BTRFS_MAX_LEVEL); | |
2152 | cur = path->nodes[*level]; | |
2153 | ||
fae7f21c | 2154 | WARN_ON(btrfs_header_level(cur) != *level); |
e02119d5 CM |
2155 | |
2156 | if (path->slots[*level] >= | |
2157 | btrfs_header_nritems(cur)) | |
2158 | break; | |
2159 | ||
2160 | bytenr = btrfs_node_blockptr(cur, path->slots[*level]); | |
2161 | ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]); | |
707e8a07 | 2162 | blocksize = root->nodesize; |
e02119d5 CM |
2163 | |
2164 | parent = path->nodes[*level]; | |
2165 | root_owner = btrfs_header_owner(parent); | |
e02119d5 CM |
2166 | |
2167 | next = btrfs_find_create_tree_block(root, bytenr, blocksize); | |
2a29edc6 | 2168 | if (!next) |
2169 | return -ENOMEM; | |
e02119d5 | 2170 | |
e02119d5 | 2171 | if (*level == 1) { |
1e5063d0 | 2172 | ret = wc->process_func(root, next, wc, ptr_gen); |
b50c6e25 JB |
2173 | if (ret) { |
2174 | free_extent_buffer(next); | |
1e5063d0 | 2175 | return ret; |
b50c6e25 | 2176 | } |
4a500fd1 | 2177 | |
e02119d5 CM |
2178 | path->slots[*level]++; |
2179 | if (wc->free) { | |
018642a1 TI |
2180 | ret = btrfs_read_buffer(next, ptr_gen); |
2181 | if (ret) { | |
2182 | free_extent_buffer(next); | |
2183 | return ret; | |
2184 | } | |
e02119d5 | 2185 | |
681ae509 JB |
2186 | if (trans) { |
2187 | btrfs_tree_lock(next); | |
2188 | btrfs_set_lock_blocking(next); | |
2189 | clean_tree_block(trans, root, next); | |
2190 | btrfs_wait_tree_block_writeback(next); | |
2191 | btrfs_tree_unlock(next); | |
2192 | } | |
e02119d5 | 2193 | |
e02119d5 CM |
2194 | WARN_ON(root_owner != |
2195 | BTRFS_TREE_LOG_OBJECTID); | |
e688b725 | 2196 | ret = btrfs_free_and_pin_reserved_extent(root, |
d00aff00 | 2197 | bytenr, blocksize); |
3650860b JB |
2198 | if (ret) { |
2199 | free_extent_buffer(next); | |
2200 | return ret; | |
2201 | } | |
e02119d5 CM |
2202 | } |
2203 | free_extent_buffer(next); | |
2204 | continue; | |
2205 | } | |
018642a1 TI |
2206 | ret = btrfs_read_buffer(next, ptr_gen); |
2207 | if (ret) { | |
2208 | free_extent_buffer(next); | |
2209 | return ret; | |
2210 | } | |
e02119d5 CM |
2211 | |
2212 | WARN_ON(*level <= 0); | |
2213 | if (path->nodes[*level-1]) | |
2214 | free_extent_buffer(path->nodes[*level-1]); | |
2215 | path->nodes[*level-1] = next; | |
2216 | *level = btrfs_header_level(next); | |
2217 | path->slots[*level] = 0; | |
2218 | cond_resched(); | |
2219 | } | |
2220 | WARN_ON(*level < 0); | |
2221 | WARN_ON(*level >= BTRFS_MAX_LEVEL); | |
2222 | ||
4a500fd1 | 2223 | path->slots[*level] = btrfs_header_nritems(path->nodes[*level]); |
e02119d5 CM |
2224 | |
2225 | cond_resched(); | |
2226 | return 0; | |
2227 | } | |
2228 | ||
d397712b | 2229 | static noinline int walk_up_log_tree(struct btrfs_trans_handle *trans, |
e02119d5 CM |
2230 | struct btrfs_root *root, |
2231 | struct btrfs_path *path, int *level, | |
2232 | struct walk_control *wc) | |
2233 | { | |
2234 | u64 root_owner; | |
e02119d5 CM |
2235 | int i; |
2236 | int slot; | |
2237 | int ret; | |
2238 | ||
d397712b | 2239 | for (i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) { |
e02119d5 | 2240 | slot = path->slots[i]; |
4a500fd1 | 2241 | if (slot + 1 < btrfs_header_nritems(path->nodes[i])) { |
e02119d5 CM |
2242 | path->slots[i]++; |
2243 | *level = i; | |
2244 | WARN_ON(*level == 0); | |
2245 | return 0; | |
2246 | } else { | |
31840ae1 ZY |
2247 | struct extent_buffer *parent; |
2248 | if (path->nodes[*level] == root->node) | |
2249 | parent = path->nodes[*level]; | |
2250 | else | |
2251 | parent = path->nodes[*level + 1]; | |
2252 | ||
2253 | root_owner = btrfs_header_owner(parent); | |
1e5063d0 | 2254 | ret = wc->process_func(root, path->nodes[*level], wc, |
e02119d5 | 2255 | btrfs_header_generation(path->nodes[*level])); |
1e5063d0 MF |
2256 | if (ret) |
2257 | return ret; | |
2258 | ||
e02119d5 CM |
2259 | if (wc->free) { |
2260 | struct extent_buffer *next; | |
2261 | ||
2262 | next = path->nodes[*level]; | |
2263 | ||
681ae509 JB |
2264 | if (trans) { |
2265 | btrfs_tree_lock(next); | |
2266 | btrfs_set_lock_blocking(next); | |
2267 | clean_tree_block(trans, root, next); | |
2268 | btrfs_wait_tree_block_writeback(next); | |
2269 | btrfs_tree_unlock(next); | |
2270 | } | |
e02119d5 | 2271 | |
e02119d5 | 2272 | WARN_ON(root_owner != BTRFS_TREE_LOG_OBJECTID); |
e688b725 | 2273 | ret = btrfs_free_and_pin_reserved_extent(root, |
e02119d5 | 2274 | path->nodes[*level]->start, |
d00aff00 | 2275 | path->nodes[*level]->len); |
3650860b JB |
2276 | if (ret) |
2277 | return ret; | |
e02119d5 CM |
2278 | } |
2279 | free_extent_buffer(path->nodes[*level]); | |
2280 | path->nodes[*level] = NULL; | |
2281 | *level = i + 1; | |
2282 | } | |
2283 | } | |
2284 | return 1; | |
2285 | } | |
2286 | ||
2287 | /* | |
2288 | * drop the reference count on the tree rooted at 'snap'. This traverses | |
2289 | * the tree freeing any blocks that have a ref count of zero after being | |
2290 | * decremented. | |
2291 | */ | |
2292 | static int walk_log_tree(struct btrfs_trans_handle *trans, | |
2293 | struct btrfs_root *log, struct walk_control *wc) | |
2294 | { | |
2295 | int ret = 0; | |
2296 | int wret; | |
2297 | int level; | |
2298 | struct btrfs_path *path; | |
e02119d5 CM |
2299 | int orig_level; |
2300 | ||
2301 | path = btrfs_alloc_path(); | |
db5b493a TI |
2302 | if (!path) |
2303 | return -ENOMEM; | |
e02119d5 CM |
2304 | |
2305 | level = btrfs_header_level(log->node); | |
2306 | orig_level = level; | |
2307 | path->nodes[level] = log->node; | |
2308 | extent_buffer_get(log->node); | |
2309 | path->slots[level] = 0; | |
2310 | ||
d397712b | 2311 | while (1) { |
e02119d5 CM |
2312 | wret = walk_down_log_tree(trans, log, path, &level, wc); |
2313 | if (wret > 0) | |
2314 | break; | |
79787eaa | 2315 | if (wret < 0) { |
e02119d5 | 2316 | ret = wret; |
79787eaa JM |
2317 | goto out; |
2318 | } | |
e02119d5 CM |
2319 | |
2320 | wret = walk_up_log_tree(trans, log, path, &level, wc); | |
2321 | if (wret > 0) | |
2322 | break; | |
79787eaa | 2323 | if (wret < 0) { |
e02119d5 | 2324 | ret = wret; |
79787eaa JM |
2325 | goto out; |
2326 | } | |
e02119d5 CM |
2327 | } |
2328 | ||
2329 | /* was the root node processed? if not, catch it here */ | |
2330 | if (path->nodes[orig_level]) { | |
79787eaa | 2331 | ret = wc->process_func(log, path->nodes[orig_level], wc, |
e02119d5 | 2332 | btrfs_header_generation(path->nodes[orig_level])); |
79787eaa JM |
2333 | if (ret) |
2334 | goto out; | |
e02119d5 CM |
2335 | if (wc->free) { |
2336 | struct extent_buffer *next; | |
2337 | ||
2338 | next = path->nodes[orig_level]; | |
2339 | ||
681ae509 JB |
2340 | if (trans) { |
2341 | btrfs_tree_lock(next); | |
2342 | btrfs_set_lock_blocking(next); | |
2343 | clean_tree_block(trans, log, next); | |
2344 | btrfs_wait_tree_block_writeback(next); | |
2345 | btrfs_tree_unlock(next); | |
2346 | } | |
e02119d5 | 2347 | |
e02119d5 CM |
2348 | WARN_ON(log->root_key.objectid != |
2349 | BTRFS_TREE_LOG_OBJECTID); | |
e688b725 | 2350 | ret = btrfs_free_and_pin_reserved_extent(log, next->start, |
d00aff00 | 2351 | next->len); |
3650860b JB |
2352 | if (ret) |
2353 | goto out; | |
e02119d5 CM |
2354 | } |
2355 | } | |
2356 | ||
79787eaa | 2357 | out: |
e02119d5 | 2358 | btrfs_free_path(path); |
e02119d5 CM |
2359 | return ret; |
2360 | } | |
2361 | ||
7237f183 YZ |
2362 | /* |
2363 | * helper function to update the item for a given subvolumes log root | |
2364 | * in the tree of log roots | |
2365 | */ | |
2366 | static int update_log_root(struct btrfs_trans_handle *trans, | |
2367 | struct btrfs_root *log) | |
2368 | { | |
2369 | int ret; | |
2370 | ||
2371 | if (log->log_transid == 1) { | |
2372 | /* insert root item on the first sync */ | |
2373 | ret = btrfs_insert_root(trans, log->fs_info->log_root_tree, | |
2374 | &log->root_key, &log->root_item); | |
2375 | } else { | |
2376 | ret = btrfs_update_root(trans, log->fs_info->log_root_tree, | |
2377 | &log->root_key, &log->root_item); | |
2378 | } | |
2379 | return ret; | |
2380 | } | |
2381 | ||
8b050d35 MX |
2382 | static void wait_log_commit(struct btrfs_trans_handle *trans, |
2383 | struct btrfs_root *root, int transid) | |
e02119d5 CM |
2384 | { |
2385 | DEFINE_WAIT(wait); | |
7237f183 | 2386 | int index = transid % 2; |
e02119d5 | 2387 | |
7237f183 YZ |
2388 | /* |
2389 | * we only allow two pending log transactions at a time, | |
2390 | * so we know that if ours is more than 2 older than the | |
2391 | * current transaction, we're done | |
2392 | */ | |
e02119d5 | 2393 | do { |
7237f183 YZ |
2394 | prepare_to_wait(&root->log_commit_wait[index], |
2395 | &wait, TASK_UNINTERRUPTIBLE); | |
2396 | mutex_unlock(&root->log_mutex); | |
12fcfd22 | 2397 | |
d1433deb | 2398 | if (root->log_transid_committed < transid && |
7237f183 YZ |
2399 | atomic_read(&root->log_commit[index])) |
2400 | schedule(); | |
12fcfd22 | 2401 | |
7237f183 YZ |
2402 | finish_wait(&root->log_commit_wait[index], &wait); |
2403 | mutex_lock(&root->log_mutex); | |
d1433deb | 2404 | } while (root->log_transid_committed < transid && |
7237f183 | 2405 | atomic_read(&root->log_commit[index])); |
7237f183 YZ |
2406 | } |
2407 | ||
143bede5 JM |
2408 | static void wait_for_writer(struct btrfs_trans_handle *trans, |
2409 | struct btrfs_root *root) | |
7237f183 YZ |
2410 | { |
2411 | DEFINE_WAIT(wait); | |
8b050d35 MX |
2412 | |
2413 | while (atomic_read(&root->log_writers)) { | |
7237f183 YZ |
2414 | prepare_to_wait(&root->log_writer_wait, |
2415 | &wait, TASK_UNINTERRUPTIBLE); | |
2416 | mutex_unlock(&root->log_mutex); | |
8b050d35 | 2417 | if (atomic_read(&root->log_writers)) |
e02119d5 | 2418 | schedule(); |
7237f183 YZ |
2419 | mutex_lock(&root->log_mutex); |
2420 | finish_wait(&root->log_writer_wait, &wait); | |
2421 | } | |
e02119d5 CM |
2422 | } |
2423 | ||
8b050d35 MX |
2424 | static inline void btrfs_remove_log_ctx(struct btrfs_root *root, |
2425 | struct btrfs_log_ctx *ctx) | |
2426 | { | |
2427 | if (!ctx) | |
2428 | return; | |
2429 | ||
2430 | mutex_lock(&root->log_mutex); | |
2431 | list_del_init(&ctx->list); | |
2432 | mutex_unlock(&root->log_mutex); | |
2433 | } | |
2434 | ||
2435 | /* | |
2436 | * Invoked in log mutex context, or be sure there is no other task which | |
2437 | * can access the list. | |
2438 | */ | |
2439 | static inline void btrfs_remove_all_log_ctxs(struct btrfs_root *root, | |
2440 | int index, int error) | |
2441 | { | |
2442 | struct btrfs_log_ctx *ctx; | |
2443 | ||
2444 | if (!error) { | |
2445 | INIT_LIST_HEAD(&root->log_ctxs[index]); | |
2446 | return; | |
2447 | } | |
2448 | ||
2449 | list_for_each_entry(ctx, &root->log_ctxs[index], list) | |
2450 | ctx->log_ret = error; | |
2451 | ||
2452 | INIT_LIST_HEAD(&root->log_ctxs[index]); | |
2453 | } | |
2454 | ||
e02119d5 CM |
2455 | /* |
2456 | * btrfs_sync_log does sends a given tree log down to the disk and | |
2457 | * updates the super blocks to record it. When this call is done, | |
12fcfd22 CM |
2458 | * you know that any inodes previously logged are safely on disk only |
2459 | * if it returns 0. | |
2460 | * | |
2461 | * Any other return value means you need to call btrfs_commit_transaction. | |
2462 | * Some of the edge cases for fsyncing directories that have had unlinks | |
2463 | * or renames done in the past mean that sometimes the only safe | |
2464 | * fsync is to commit the whole FS. When btrfs_sync_log returns -EAGAIN, | |
2465 | * that has happened. | |
e02119d5 CM |
2466 | */ |
2467 | int btrfs_sync_log(struct btrfs_trans_handle *trans, | |
8b050d35 | 2468 | struct btrfs_root *root, struct btrfs_log_ctx *ctx) |
e02119d5 | 2469 | { |
7237f183 YZ |
2470 | int index1; |
2471 | int index2; | |
8cef4e16 | 2472 | int mark; |
e02119d5 | 2473 | int ret; |
e02119d5 | 2474 | struct btrfs_root *log = root->log_root; |
7237f183 | 2475 | struct btrfs_root *log_root_tree = root->fs_info->log_root_tree; |
bb14a59b | 2476 | int log_transid = 0; |
8b050d35 | 2477 | struct btrfs_log_ctx root_log_ctx; |
c6adc9cc | 2478 | struct blk_plug plug; |
e02119d5 | 2479 | |
7237f183 | 2480 | mutex_lock(&root->log_mutex); |
d1433deb MX |
2481 | log_transid = ctx->log_transid; |
2482 | if (root->log_transid_committed >= log_transid) { | |
2483 | mutex_unlock(&root->log_mutex); | |
2484 | return ctx->log_ret; | |
2485 | } | |
2486 | ||
2487 | index1 = log_transid % 2; | |
7237f183 | 2488 | if (atomic_read(&root->log_commit[index1])) { |
d1433deb | 2489 | wait_log_commit(trans, root, log_transid); |
7237f183 | 2490 | mutex_unlock(&root->log_mutex); |
8b050d35 | 2491 | return ctx->log_ret; |
e02119d5 | 2492 | } |
d1433deb | 2493 | ASSERT(log_transid == root->log_transid); |
7237f183 YZ |
2494 | atomic_set(&root->log_commit[index1], 1); |
2495 | ||
2496 | /* wait for previous tree log sync to complete */ | |
2497 | if (atomic_read(&root->log_commit[(index1 + 1) % 2])) | |
d1433deb | 2498 | wait_log_commit(trans, root, log_transid - 1); |
48cab2e0 | 2499 | |
86df7eb9 | 2500 | while (1) { |
2ecb7923 | 2501 | int batch = atomic_read(&root->log_batch); |
cd354ad6 | 2502 | /* when we're on an ssd, just kick the log commit out */ |
27cdeb70 MX |
2503 | if (!btrfs_test_opt(root, SSD) && |
2504 | test_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state)) { | |
86df7eb9 YZ |
2505 | mutex_unlock(&root->log_mutex); |
2506 | schedule_timeout_uninterruptible(1); | |
2507 | mutex_lock(&root->log_mutex); | |
2508 | } | |
12fcfd22 | 2509 | wait_for_writer(trans, root); |
2ecb7923 | 2510 | if (batch == atomic_read(&root->log_batch)) |
e02119d5 CM |
2511 | break; |
2512 | } | |
e02119d5 | 2513 | |
12fcfd22 | 2514 | /* bail out if we need to do a full commit */ |
995946dd | 2515 | if (btrfs_need_log_full_commit(root->fs_info, trans)) { |
12fcfd22 | 2516 | ret = -EAGAIN; |
2ab28f32 | 2517 | btrfs_free_logged_extents(log, log_transid); |
12fcfd22 CM |
2518 | mutex_unlock(&root->log_mutex); |
2519 | goto out; | |
2520 | } | |
2521 | ||
8cef4e16 YZ |
2522 | if (log_transid % 2 == 0) |
2523 | mark = EXTENT_DIRTY; | |
2524 | else | |
2525 | mark = EXTENT_NEW; | |
2526 | ||
690587d1 CM |
2527 | /* we start IO on all the marked extents here, but we don't actually |
2528 | * wait for them until later. | |
2529 | */ | |
c6adc9cc | 2530 | blk_start_plug(&plug); |
8cef4e16 | 2531 | ret = btrfs_write_marked_extents(log, &log->dirty_log_pages, mark); |
79787eaa | 2532 | if (ret) { |
c6adc9cc | 2533 | blk_finish_plug(&plug); |
79787eaa | 2534 | btrfs_abort_transaction(trans, root, ret); |
2ab28f32 | 2535 | btrfs_free_logged_extents(log, log_transid); |
995946dd | 2536 | btrfs_set_log_full_commit(root->fs_info, trans); |
79787eaa JM |
2537 | mutex_unlock(&root->log_mutex); |
2538 | goto out; | |
2539 | } | |
7237f183 | 2540 | |
5d4f98a2 | 2541 | btrfs_set_root_node(&log->root_item, log->node); |
7237f183 | 2542 | |
7237f183 YZ |
2543 | root->log_transid++; |
2544 | log->log_transid = root->log_transid; | |
ff782e0a | 2545 | root->log_start_pid = 0; |
7237f183 | 2546 | /* |
8cef4e16 YZ |
2547 | * IO has been started, blocks of the log tree have WRITTEN flag set |
2548 | * in their headers. new modifications of the log will be written to | |
2549 | * new positions. so it's safe to allow log writers to go in. | |
7237f183 YZ |
2550 | */ |
2551 | mutex_unlock(&root->log_mutex); | |
2552 | ||
d1433deb MX |
2553 | btrfs_init_log_ctx(&root_log_ctx); |
2554 | ||
7237f183 | 2555 | mutex_lock(&log_root_tree->log_mutex); |
2ecb7923 | 2556 | atomic_inc(&log_root_tree->log_batch); |
7237f183 | 2557 | atomic_inc(&log_root_tree->log_writers); |
d1433deb MX |
2558 | |
2559 | index2 = log_root_tree->log_transid % 2; | |
2560 | list_add_tail(&root_log_ctx.list, &log_root_tree->log_ctxs[index2]); | |
2561 | root_log_ctx.log_transid = log_root_tree->log_transid; | |
2562 | ||
7237f183 YZ |
2563 | mutex_unlock(&log_root_tree->log_mutex); |
2564 | ||
2565 | ret = update_log_root(trans, log); | |
7237f183 YZ |
2566 | |
2567 | mutex_lock(&log_root_tree->log_mutex); | |
2568 | if (atomic_dec_and_test(&log_root_tree->log_writers)) { | |
2569 | smp_mb(); | |
2570 | if (waitqueue_active(&log_root_tree->log_writer_wait)) | |
2571 | wake_up(&log_root_tree->log_writer_wait); | |
2572 | } | |
2573 | ||
4a500fd1 | 2574 | if (ret) { |
d1433deb MX |
2575 | if (!list_empty(&root_log_ctx.list)) |
2576 | list_del_init(&root_log_ctx.list); | |
2577 | ||
c6adc9cc | 2578 | blk_finish_plug(&plug); |
995946dd MX |
2579 | btrfs_set_log_full_commit(root->fs_info, trans); |
2580 | ||
79787eaa JM |
2581 | if (ret != -ENOSPC) { |
2582 | btrfs_abort_transaction(trans, root, ret); | |
2583 | mutex_unlock(&log_root_tree->log_mutex); | |
2584 | goto out; | |
2585 | } | |
4a500fd1 | 2586 | btrfs_wait_marked_extents(log, &log->dirty_log_pages, mark); |
2ab28f32 | 2587 | btrfs_free_logged_extents(log, log_transid); |
4a500fd1 YZ |
2588 | mutex_unlock(&log_root_tree->log_mutex); |
2589 | ret = -EAGAIN; | |
2590 | goto out; | |
2591 | } | |
2592 | ||
d1433deb MX |
2593 | if (log_root_tree->log_transid_committed >= root_log_ctx.log_transid) { |
2594 | mutex_unlock(&log_root_tree->log_mutex); | |
2595 | ret = root_log_ctx.log_ret; | |
2596 | goto out; | |
2597 | } | |
8b050d35 | 2598 | |
d1433deb | 2599 | index2 = root_log_ctx.log_transid % 2; |
7237f183 | 2600 | if (atomic_read(&log_root_tree->log_commit[index2])) { |
c6adc9cc | 2601 | blk_finish_plug(&plug); |
5ab5e44a FM |
2602 | ret = btrfs_wait_marked_extents(log, &log->dirty_log_pages, |
2603 | mark); | |
50d9aa99 | 2604 | btrfs_wait_logged_extents(trans, log, log_transid); |
8b050d35 | 2605 | wait_log_commit(trans, log_root_tree, |
d1433deb | 2606 | root_log_ctx.log_transid); |
7237f183 | 2607 | mutex_unlock(&log_root_tree->log_mutex); |
5ab5e44a FM |
2608 | if (!ret) |
2609 | ret = root_log_ctx.log_ret; | |
7237f183 YZ |
2610 | goto out; |
2611 | } | |
d1433deb | 2612 | ASSERT(root_log_ctx.log_transid == log_root_tree->log_transid); |
7237f183 YZ |
2613 | atomic_set(&log_root_tree->log_commit[index2], 1); |
2614 | ||
12fcfd22 CM |
2615 | if (atomic_read(&log_root_tree->log_commit[(index2 + 1) % 2])) { |
2616 | wait_log_commit(trans, log_root_tree, | |
d1433deb | 2617 | root_log_ctx.log_transid - 1); |
12fcfd22 CM |
2618 | } |
2619 | ||
2620 | wait_for_writer(trans, log_root_tree); | |
7237f183 | 2621 | |
12fcfd22 CM |
2622 | /* |
2623 | * now that we've moved on to the tree of log tree roots, | |
2624 | * check the full commit flag again | |
2625 | */ | |
995946dd | 2626 | if (btrfs_need_log_full_commit(root->fs_info, trans)) { |
c6adc9cc | 2627 | blk_finish_plug(&plug); |
8cef4e16 | 2628 | btrfs_wait_marked_extents(log, &log->dirty_log_pages, mark); |
2ab28f32 | 2629 | btrfs_free_logged_extents(log, log_transid); |
12fcfd22 CM |
2630 | mutex_unlock(&log_root_tree->log_mutex); |
2631 | ret = -EAGAIN; | |
2632 | goto out_wake_log_root; | |
2633 | } | |
7237f183 | 2634 | |
c6adc9cc MX |
2635 | ret = btrfs_write_marked_extents(log_root_tree, |
2636 | &log_root_tree->dirty_log_pages, | |
2637 | EXTENT_DIRTY | EXTENT_NEW); | |
2638 | blk_finish_plug(&plug); | |
79787eaa | 2639 | if (ret) { |
995946dd | 2640 | btrfs_set_log_full_commit(root->fs_info, trans); |
79787eaa | 2641 | btrfs_abort_transaction(trans, root, ret); |
2ab28f32 | 2642 | btrfs_free_logged_extents(log, log_transid); |
79787eaa JM |
2643 | mutex_unlock(&log_root_tree->log_mutex); |
2644 | goto out_wake_log_root; | |
2645 | } | |
5ab5e44a FM |
2646 | ret = btrfs_wait_marked_extents(log, &log->dirty_log_pages, mark); |
2647 | if (!ret) | |
2648 | ret = btrfs_wait_marked_extents(log_root_tree, | |
2649 | &log_root_tree->dirty_log_pages, | |
2650 | EXTENT_NEW | EXTENT_DIRTY); | |
2651 | if (ret) { | |
2652 | btrfs_set_log_full_commit(root->fs_info, trans); | |
2653 | btrfs_free_logged_extents(log, log_transid); | |
2654 | mutex_unlock(&log_root_tree->log_mutex); | |
2655 | goto out_wake_log_root; | |
2656 | } | |
50d9aa99 | 2657 | btrfs_wait_logged_extents(trans, log, log_transid); |
e02119d5 | 2658 | |
6c41761f | 2659 | btrfs_set_super_log_root(root->fs_info->super_for_commit, |
7237f183 | 2660 | log_root_tree->node->start); |
6c41761f | 2661 | btrfs_set_super_log_root_level(root->fs_info->super_for_commit, |
7237f183 | 2662 | btrfs_header_level(log_root_tree->node)); |
e02119d5 | 2663 | |
7237f183 | 2664 | log_root_tree->log_transid++; |
7237f183 YZ |
2665 | mutex_unlock(&log_root_tree->log_mutex); |
2666 | ||
2667 | /* | |
2668 | * nobody else is going to jump in and write the the ctree | |
2669 | * super here because the log_commit atomic below is protecting | |
2670 | * us. We must be called with a transaction handle pinning | |
2671 | * the running transaction open, so a full commit can't hop | |
2672 | * in and cause problems either. | |
2673 | */ | |
5af3e8cc | 2674 | ret = write_ctree_super(trans, root->fs_info->tree_root, 1); |
5af3e8cc | 2675 | if (ret) { |
995946dd | 2676 | btrfs_set_log_full_commit(root->fs_info, trans); |
5af3e8cc SB |
2677 | btrfs_abort_transaction(trans, root, ret); |
2678 | goto out_wake_log_root; | |
2679 | } | |
7237f183 | 2680 | |
257c62e1 CM |
2681 | mutex_lock(&root->log_mutex); |
2682 | if (root->last_log_commit < log_transid) | |
2683 | root->last_log_commit = log_transid; | |
2684 | mutex_unlock(&root->log_mutex); | |
2685 | ||
12fcfd22 | 2686 | out_wake_log_root: |
8b050d35 MX |
2687 | /* |
2688 | * We needn't get log_mutex here because we are sure all | |
2689 | * the other tasks are blocked. | |
2690 | */ | |
2691 | btrfs_remove_all_log_ctxs(log_root_tree, index2, ret); | |
2692 | ||
d1433deb MX |
2693 | mutex_lock(&log_root_tree->log_mutex); |
2694 | log_root_tree->log_transid_committed++; | |
7237f183 | 2695 | atomic_set(&log_root_tree->log_commit[index2], 0); |
d1433deb MX |
2696 | mutex_unlock(&log_root_tree->log_mutex); |
2697 | ||
7237f183 YZ |
2698 | if (waitqueue_active(&log_root_tree->log_commit_wait[index2])) |
2699 | wake_up(&log_root_tree->log_commit_wait[index2]); | |
e02119d5 | 2700 | out: |
8b050d35 MX |
2701 | /* See above. */ |
2702 | btrfs_remove_all_log_ctxs(root, index1, ret); | |
2703 | ||
d1433deb MX |
2704 | mutex_lock(&root->log_mutex); |
2705 | root->log_transid_committed++; | |
7237f183 | 2706 | atomic_set(&root->log_commit[index1], 0); |
d1433deb | 2707 | mutex_unlock(&root->log_mutex); |
8b050d35 | 2708 | |
7237f183 YZ |
2709 | if (waitqueue_active(&root->log_commit_wait[index1])) |
2710 | wake_up(&root->log_commit_wait[index1]); | |
b31eabd8 | 2711 | return ret; |
e02119d5 CM |
2712 | } |
2713 | ||
4a500fd1 YZ |
2714 | static void free_log_tree(struct btrfs_trans_handle *trans, |
2715 | struct btrfs_root *log) | |
e02119d5 CM |
2716 | { |
2717 | int ret; | |
d0c803c4 CM |
2718 | u64 start; |
2719 | u64 end; | |
e02119d5 CM |
2720 | struct walk_control wc = { |
2721 | .free = 1, | |
2722 | .process_func = process_one_buffer | |
2723 | }; | |
2724 | ||
681ae509 JB |
2725 | ret = walk_log_tree(trans, log, &wc); |
2726 | /* I don't think this can happen but just in case */ | |
2727 | if (ret) | |
2728 | btrfs_abort_transaction(trans, log, ret); | |
e02119d5 | 2729 | |
d397712b | 2730 | while (1) { |
d0c803c4 | 2731 | ret = find_first_extent_bit(&log->dirty_log_pages, |
e6138876 JB |
2732 | 0, &start, &end, EXTENT_DIRTY | EXTENT_NEW, |
2733 | NULL); | |
d0c803c4 CM |
2734 | if (ret) |
2735 | break; | |
2736 | ||
8cef4e16 YZ |
2737 | clear_extent_bits(&log->dirty_log_pages, start, end, |
2738 | EXTENT_DIRTY | EXTENT_NEW, GFP_NOFS); | |
d0c803c4 CM |
2739 | } |
2740 | ||
2ab28f32 JB |
2741 | /* |
2742 | * We may have short-circuited the log tree with the full commit logic | |
2743 | * and left ordered extents on our list, so clear these out to keep us | |
2744 | * from leaking inodes and memory. | |
2745 | */ | |
2746 | btrfs_free_logged_extents(log, 0); | |
2747 | btrfs_free_logged_extents(log, 1); | |
2748 | ||
7237f183 YZ |
2749 | free_extent_buffer(log->node); |
2750 | kfree(log); | |
4a500fd1 YZ |
2751 | } |
2752 | ||
2753 | /* | |
2754 | * free all the extents used by the tree log. This should be called | |
2755 | * at commit time of the full transaction | |
2756 | */ | |
2757 | int btrfs_free_log(struct btrfs_trans_handle *trans, struct btrfs_root *root) | |
2758 | { | |
2759 | if (root->log_root) { | |
2760 | free_log_tree(trans, root->log_root); | |
2761 | root->log_root = NULL; | |
2762 | } | |
2763 | return 0; | |
2764 | } | |
2765 | ||
2766 | int btrfs_free_log_root_tree(struct btrfs_trans_handle *trans, | |
2767 | struct btrfs_fs_info *fs_info) | |
2768 | { | |
2769 | if (fs_info->log_root_tree) { | |
2770 | free_log_tree(trans, fs_info->log_root_tree); | |
2771 | fs_info->log_root_tree = NULL; | |
2772 | } | |
e02119d5 CM |
2773 | return 0; |
2774 | } | |
2775 | ||
e02119d5 CM |
2776 | /* |
2777 | * If both a file and directory are logged, and unlinks or renames are | |
2778 | * mixed in, we have a few interesting corners: | |
2779 | * | |
2780 | * create file X in dir Y | |
2781 | * link file X to X.link in dir Y | |
2782 | * fsync file X | |
2783 | * unlink file X but leave X.link | |
2784 | * fsync dir Y | |
2785 | * | |
2786 | * After a crash we would expect only X.link to exist. But file X | |
2787 | * didn't get fsync'd again so the log has back refs for X and X.link. | |
2788 | * | |
2789 | * We solve this by removing directory entries and inode backrefs from the | |
2790 | * log when a file that was logged in the current transaction is | |
2791 | * unlinked. Any later fsync will include the updated log entries, and | |
2792 | * we'll be able to reconstruct the proper directory items from backrefs. | |
2793 | * | |
2794 | * This optimizations allows us to avoid relogging the entire inode | |
2795 | * or the entire directory. | |
2796 | */ | |
2797 | int btrfs_del_dir_entries_in_log(struct btrfs_trans_handle *trans, | |
2798 | struct btrfs_root *root, | |
2799 | const char *name, int name_len, | |
2800 | struct inode *dir, u64 index) | |
2801 | { | |
2802 | struct btrfs_root *log; | |
2803 | struct btrfs_dir_item *di; | |
2804 | struct btrfs_path *path; | |
2805 | int ret; | |
4a500fd1 | 2806 | int err = 0; |
e02119d5 | 2807 | int bytes_del = 0; |
33345d01 | 2808 | u64 dir_ino = btrfs_ino(dir); |
e02119d5 | 2809 | |
3a5f1d45 CM |
2810 | if (BTRFS_I(dir)->logged_trans < trans->transid) |
2811 | return 0; | |
2812 | ||
e02119d5 CM |
2813 | ret = join_running_log_trans(root); |
2814 | if (ret) | |
2815 | return 0; | |
2816 | ||
2817 | mutex_lock(&BTRFS_I(dir)->log_mutex); | |
2818 | ||
2819 | log = root->log_root; | |
2820 | path = btrfs_alloc_path(); | |
a62f44a5 TI |
2821 | if (!path) { |
2822 | err = -ENOMEM; | |
2823 | goto out_unlock; | |
2824 | } | |
2a29edc6 | 2825 | |
33345d01 | 2826 | di = btrfs_lookup_dir_item(trans, log, path, dir_ino, |
e02119d5 | 2827 | name, name_len, -1); |
4a500fd1 YZ |
2828 | if (IS_ERR(di)) { |
2829 | err = PTR_ERR(di); | |
2830 | goto fail; | |
2831 | } | |
2832 | if (di) { | |
e02119d5 CM |
2833 | ret = btrfs_delete_one_dir_name(trans, log, path, di); |
2834 | bytes_del += name_len; | |
3650860b JB |
2835 | if (ret) { |
2836 | err = ret; | |
2837 | goto fail; | |
2838 | } | |
e02119d5 | 2839 | } |
b3b4aa74 | 2840 | btrfs_release_path(path); |
33345d01 | 2841 | di = btrfs_lookup_dir_index_item(trans, log, path, dir_ino, |
e02119d5 | 2842 | index, name, name_len, -1); |
4a500fd1 YZ |
2843 | if (IS_ERR(di)) { |
2844 | err = PTR_ERR(di); | |
2845 | goto fail; | |
2846 | } | |
2847 | if (di) { | |
e02119d5 CM |
2848 | ret = btrfs_delete_one_dir_name(trans, log, path, di); |
2849 | bytes_del += name_len; | |
3650860b JB |
2850 | if (ret) { |
2851 | err = ret; | |
2852 | goto fail; | |
2853 | } | |
e02119d5 CM |
2854 | } |
2855 | ||
2856 | /* update the directory size in the log to reflect the names | |
2857 | * we have removed | |
2858 | */ | |
2859 | if (bytes_del) { | |
2860 | struct btrfs_key key; | |
2861 | ||
33345d01 | 2862 | key.objectid = dir_ino; |
e02119d5 CM |
2863 | key.offset = 0; |
2864 | key.type = BTRFS_INODE_ITEM_KEY; | |
b3b4aa74 | 2865 | btrfs_release_path(path); |
e02119d5 CM |
2866 | |
2867 | ret = btrfs_search_slot(trans, log, &key, path, 0, 1); | |
4a500fd1 YZ |
2868 | if (ret < 0) { |
2869 | err = ret; | |
2870 | goto fail; | |
2871 | } | |
e02119d5 CM |
2872 | if (ret == 0) { |
2873 | struct btrfs_inode_item *item; | |
2874 | u64 i_size; | |
2875 | ||
2876 | item = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
2877 | struct btrfs_inode_item); | |
2878 | i_size = btrfs_inode_size(path->nodes[0], item); | |
2879 | if (i_size > bytes_del) | |
2880 | i_size -= bytes_del; | |
2881 | else | |
2882 | i_size = 0; | |
2883 | btrfs_set_inode_size(path->nodes[0], item, i_size); | |
2884 | btrfs_mark_buffer_dirty(path->nodes[0]); | |
2885 | } else | |
2886 | ret = 0; | |
b3b4aa74 | 2887 | btrfs_release_path(path); |
e02119d5 | 2888 | } |
4a500fd1 | 2889 | fail: |
e02119d5 | 2890 | btrfs_free_path(path); |
a62f44a5 | 2891 | out_unlock: |
e02119d5 | 2892 | mutex_unlock(&BTRFS_I(dir)->log_mutex); |
4a500fd1 | 2893 | if (ret == -ENOSPC) { |
995946dd | 2894 | btrfs_set_log_full_commit(root->fs_info, trans); |
4a500fd1 | 2895 | ret = 0; |
79787eaa JM |
2896 | } else if (ret < 0) |
2897 | btrfs_abort_transaction(trans, root, ret); | |
2898 | ||
12fcfd22 | 2899 | btrfs_end_log_trans(root); |
e02119d5 | 2900 | |
411fc6bc | 2901 | return err; |
e02119d5 CM |
2902 | } |
2903 | ||
2904 | /* see comments for btrfs_del_dir_entries_in_log */ | |
2905 | int btrfs_del_inode_ref_in_log(struct btrfs_trans_handle *trans, | |
2906 | struct btrfs_root *root, | |
2907 | const char *name, int name_len, | |
2908 | struct inode *inode, u64 dirid) | |
2909 | { | |
2910 | struct btrfs_root *log; | |
2911 | u64 index; | |
2912 | int ret; | |
2913 | ||
3a5f1d45 CM |
2914 | if (BTRFS_I(inode)->logged_trans < trans->transid) |
2915 | return 0; | |
2916 | ||
e02119d5 CM |
2917 | ret = join_running_log_trans(root); |
2918 | if (ret) | |
2919 | return 0; | |
2920 | log = root->log_root; | |
2921 | mutex_lock(&BTRFS_I(inode)->log_mutex); | |
2922 | ||
33345d01 | 2923 | ret = btrfs_del_inode_ref(trans, log, name, name_len, btrfs_ino(inode), |
e02119d5 CM |
2924 | dirid, &index); |
2925 | mutex_unlock(&BTRFS_I(inode)->log_mutex); | |
4a500fd1 | 2926 | if (ret == -ENOSPC) { |
995946dd | 2927 | btrfs_set_log_full_commit(root->fs_info, trans); |
4a500fd1 | 2928 | ret = 0; |
79787eaa JM |
2929 | } else if (ret < 0 && ret != -ENOENT) |
2930 | btrfs_abort_transaction(trans, root, ret); | |
12fcfd22 | 2931 | btrfs_end_log_trans(root); |
e02119d5 | 2932 | |
e02119d5 CM |
2933 | return ret; |
2934 | } | |
2935 | ||
2936 | /* | |
2937 | * creates a range item in the log for 'dirid'. first_offset and | |
2938 | * last_offset tell us which parts of the key space the log should | |
2939 | * be considered authoritative for. | |
2940 | */ | |
2941 | static noinline int insert_dir_log_key(struct btrfs_trans_handle *trans, | |
2942 | struct btrfs_root *log, | |
2943 | struct btrfs_path *path, | |
2944 | int key_type, u64 dirid, | |
2945 | u64 first_offset, u64 last_offset) | |
2946 | { | |
2947 | int ret; | |
2948 | struct btrfs_key key; | |
2949 | struct btrfs_dir_log_item *item; | |
2950 | ||
2951 | key.objectid = dirid; | |
2952 | key.offset = first_offset; | |
2953 | if (key_type == BTRFS_DIR_ITEM_KEY) | |
2954 | key.type = BTRFS_DIR_LOG_ITEM_KEY; | |
2955 | else | |
2956 | key.type = BTRFS_DIR_LOG_INDEX_KEY; | |
2957 | ret = btrfs_insert_empty_item(trans, log, path, &key, sizeof(*item)); | |
4a500fd1 YZ |
2958 | if (ret) |
2959 | return ret; | |
e02119d5 CM |
2960 | |
2961 | item = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
2962 | struct btrfs_dir_log_item); | |
2963 | btrfs_set_dir_log_end(path->nodes[0], item, last_offset); | |
2964 | btrfs_mark_buffer_dirty(path->nodes[0]); | |
b3b4aa74 | 2965 | btrfs_release_path(path); |
e02119d5 CM |
2966 | return 0; |
2967 | } | |
2968 | ||
2969 | /* | |
2970 | * log all the items included in the current transaction for a given | |
2971 | * directory. This also creates the range items in the log tree required | |
2972 | * to replay anything deleted before the fsync | |
2973 | */ | |
2974 | static noinline int log_dir_items(struct btrfs_trans_handle *trans, | |
2975 | struct btrfs_root *root, struct inode *inode, | |
2976 | struct btrfs_path *path, | |
2977 | struct btrfs_path *dst_path, int key_type, | |
2978 | u64 min_offset, u64 *last_offset_ret) | |
2979 | { | |
2980 | struct btrfs_key min_key; | |
e02119d5 CM |
2981 | struct btrfs_root *log = root->log_root; |
2982 | struct extent_buffer *src; | |
4a500fd1 | 2983 | int err = 0; |
e02119d5 CM |
2984 | int ret; |
2985 | int i; | |
2986 | int nritems; | |
2987 | u64 first_offset = min_offset; | |
2988 | u64 last_offset = (u64)-1; | |
33345d01 | 2989 | u64 ino = btrfs_ino(inode); |
e02119d5 CM |
2990 | |
2991 | log = root->log_root; | |
e02119d5 | 2992 | |
33345d01 | 2993 | min_key.objectid = ino; |
e02119d5 CM |
2994 | min_key.type = key_type; |
2995 | min_key.offset = min_offset; | |
2996 | ||
6174d3cb | 2997 | ret = btrfs_search_forward(root, &min_key, path, trans->transid); |
e02119d5 CM |
2998 | |
2999 | /* | |
3000 | * we didn't find anything from this transaction, see if there | |
3001 | * is anything at all | |
3002 | */ | |
33345d01 LZ |
3003 | if (ret != 0 || min_key.objectid != ino || min_key.type != key_type) { |
3004 | min_key.objectid = ino; | |
e02119d5 CM |
3005 | min_key.type = key_type; |
3006 | min_key.offset = (u64)-1; | |
b3b4aa74 | 3007 | btrfs_release_path(path); |
e02119d5 CM |
3008 | ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0); |
3009 | if (ret < 0) { | |
b3b4aa74 | 3010 | btrfs_release_path(path); |
e02119d5 CM |
3011 | return ret; |
3012 | } | |
33345d01 | 3013 | ret = btrfs_previous_item(root, path, ino, key_type); |
e02119d5 CM |
3014 | |
3015 | /* if ret == 0 there are items for this type, | |
3016 | * create a range to tell us the last key of this type. | |
3017 | * otherwise, there are no items in this directory after | |
3018 | * *min_offset, and we create a range to indicate that. | |
3019 | */ | |
3020 | if (ret == 0) { | |
3021 | struct btrfs_key tmp; | |
3022 | btrfs_item_key_to_cpu(path->nodes[0], &tmp, | |
3023 | path->slots[0]); | |
d397712b | 3024 | if (key_type == tmp.type) |
e02119d5 | 3025 | first_offset = max(min_offset, tmp.offset) + 1; |
e02119d5 CM |
3026 | } |
3027 | goto done; | |
3028 | } | |
3029 | ||
3030 | /* go backward to find any previous key */ | |
33345d01 | 3031 | ret = btrfs_previous_item(root, path, ino, key_type); |
e02119d5 CM |
3032 | if (ret == 0) { |
3033 | struct btrfs_key tmp; | |
3034 | btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]); | |
3035 | if (key_type == tmp.type) { | |
3036 | first_offset = tmp.offset; | |
3037 | ret = overwrite_item(trans, log, dst_path, | |
3038 | path->nodes[0], path->slots[0], | |
3039 | &tmp); | |
4a500fd1 YZ |
3040 | if (ret) { |
3041 | err = ret; | |
3042 | goto done; | |
3043 | } | |
e02119d5 CM |
3044 | } |
3045 | } | |
b3b4aa74 | 3046 | btrfs_release_path(path); |
e02119d5 CM |
3047 | |
3048 | /* find the first key from this transaction again */ | |
3049 | ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0); | |
fae7f21c | 3050 | if (WARN_ON(ret != 0)) |
e02119d5 | 3051 | goto done; |
e02119d5 CM |
3052 | |
3053 | /* | |
3054 | * we have a block from this transaction, log every item in it | |
3055 | * from our directory | |
3056 | */ | |
d397712b | 3057 | while (1) { |
e02119d5 CM |
3058 | struct btrfs_key tmp; |
3059 | src = path->nodes[0]; | |
3060 | nritems = btrfs_header_nritems(src); | |
3061 | for (i = path->slots[0]; i < nritems; i++) { | |
3062 | btrfs_item_key_to_cpu(src, &min_key, i); | |
3063 | ||
33345d01 | 3064 | if (min_key.objectid != ino || min_key.type != key_type) |
e02119d5 CM |
3065 | goto done; |
3066 | ret = overwrite_item(trans, log, dst_path, src, i, | |
3067 | &min_key); | |
4a500fd1 YZ |
3068 | if (ret) { |
3069 | err = ret; | |
3070 | goto done; | |
3071 | } | |
e02119d5 CM |
3072 | } |
3073 | path->slots[0] = nritems; | |
3074 | ||
3075 | /* | |
3076 | * look ahead to the next item and see if it is also | |
3077 | * from this directory and from this transaction | |
3078 | */ | |
3079 | ret = btrfs_next_leaf(root, path); | |
3080 | if (ret == 1) { | |
3081 | last_offset = (u64)-1; | |
3082 | goto done; | |
3083 | } | |
3084 | btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]); | |
33345d01 | 3085 | if (tmp.objectid != ino || tmp.type != key_type) { |
e02119d5 CM |
3086 | last_offset = (u64)-1; |
3087 | goto done; | |
3088 | } | |
3089 | if (btrfs_header_generation(path->nodes[0]) != trans->transid) { | |
3090 | ret = overwrite_item(trans, log, dst_path, | |
3091 | path->nodes[0], path->slots[0], | |
3092 | &tmp); | |
4a500fd1 YZ |
3093 | if (ret) |
3094 | err = ret; | |
3095 | else | |
3096 | last_offset = tmp.offset; | |
e02119d5 CM |
3097 | goto done; |
3098 | } | |
3099 | } | |
3100 | done: | |
b3b4aa74 DS |
3101 | btrfs_release_path(path); |
3102 | btrfs_release_path(dst_path); | |
e02119d5 | 3103 | |
4a500fd1 YZ |
3104 | if (err == 0) { |
3105 | *last_offset_ret = last_offset; | |
3106 | /* | |
3107 | * insert the log range keys to indicate where the log | |
3108 | * is valid | |
3109 | */ | |
3110 | ret = insert_dir_log_key(trans, log, path, key_type, | |
33345d01 | 3111 | ino, first_offset, last_offset); |
4a500fd1 YZ |
3112 | if (ret) |
3113 | err = ret; | |
3114 | } | |
3115 | return err; | |
e02119d5 CM |
3116 | } |
3117 | ||
3118 | /* | |
3119 | * logging directories is very similar to logging inodes, We find all the items | |
3120 | * from the current transaction and write them to the log. | |
3121 | * | |
3122 | * The recovery code scans the directory in the subvolume, and if it finds a | |
3123 | * key in the range logged that is not present in the log tree, then it means | |
3124 | * that dir entry was unlinked during the transaction. | |
3125 | * | |
3126 | * In order for that scan to work, we must include one key smaller than | |
3127 | * the smallest logged by this transaction and one key larger than the largest | |
3128 | * key logged by this transaction. | |
3129 | */ | |
3130 | static noinline int log_directory_changes(struct btrfs_trans_handle *trans, | |
3131 | struct btrfs_root *root, struct inode *inode, | |
3132 | struct btrfs_path *path, | |
3133 | struct btrfs_path *dst_path) | |
3134 | { | |
3135 | u64 min_key; | |
3136 | u64 max_key; | |
3137 | int ret; | |
3138 | int key_type = BTRFS_DIR_ITEM_KEY; | |
3139 | ||
3140 | again: | |
3141 | min_key = 0; | |
3142 | max_key = 0; | |
d397712b | 3143 | while (1) { |
e02119d5 CM |
3144 | ret = log_dir_items(trans, root, inode, path, |
3145 | dst_path, key_type, min_key, | |
3146 | &max_key); | |
4a500fd1 YZ |
3147 | if (ret) |
3148 | return ret; | |
e02119d5 CM |
3149 | if (max_key == (u64)-1) |
3150 | break; | |
3151 | min_key = max_key + 1; | |
3152 | } | |
3153 | ||
3154 | if (key_type == BTRFS_DIR_ITEM_KEY) { | |
3155 | key_type = BTRFS_DIR_INDEX_KEY; | |
3156 | goto again; | |
3157 | } | |
3158 | return 0; | |
3159 | } | |
3160 | ||
3161 | /* | |
3162 | * a helper function to drop items from the log before we relog an | |
3163 | * inode. max_key_type indicates the highest item type to remove. | |
3164 | * This cannot be run for file data extents because it does not | |
3165 | * free the extents they point to. | |
3166 | */ | |
3167 | static int drop_objectid_items(struct btrfs_trans_handle *trans, | |
3168 | struct btrfs_root *log, | |
3169 | struct btrfs_path *path, | |
3170 | u64 objectid, int max_key_type) | |
3171 | { | |
3172 | int ret; | |
3173 | struct btrfs_key key; | |
3174 | struct btrfs_key found_key; | |
18ec90d6 | 3175 | int start_slot; |
e02119d5 CM |
3176 | |
3177 | key.objectid = objectid; | |
3178 | key.type = max_key_type; | |
3179 | key.offset = (u64)-1; | |
3180 | ||
d397712b | 3181 | while (1) { |
e02119d5 | 3182 | ret = btrfs_search_slot(trans, log, &key, path, -1, 1); |
3650860b | 3183 | BUG_ON(ret == 0); /* Logic error */ |
4a500fd1 | 3184 | if (ret < 0) |
e02119d5 CM |
3185 | break; |
3186 | ||
3187 | if (path->slots[0] == 0) | |
3188 | break; | |
3189 | ||
3190 | path->slots[0]--; | |
3191 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, | |
3192 | path->slots[0]); | |
3193 | ||
3194 | if (found_key.objectid != objectid) | |
3195 | break; | |
3196 | ||
18ec90d6 JB |
3197 | found_key.offset = 0; |
3198 | found_key.type = 0; | |
3199 | ret = btrfs_bin_search(path->nodes[0], &found_key, 0, | |
3200 | &start_slot); | |
3201 | ||
3202 | ret = btrfs_del_items(trans, log, path, start_slot, | |
3203 | path->slots[0] - start_slot + 1); | |
3204 | /* | |
3205 | * If start slot isn't 0 then we don't need to re-search, we've | |
3206 | * found the last guy with the objectid in this tree. | |
3207 | */ | |
3208 | if (ret || start_slot != 0) | |
65a246c5 | 3209 | break; |
b3b4aa74 | 3210 | btrfs_release_path(path); |
e02119d5 | 3211 | } |
b3b4aa74 | 3212 | btrfs_release_path(path); |
5bdbeb21 JB |
3213 | if (ret > 0) |
3214 | ret = 0; | |
4a500fd1 | 3215 | return ret; |
e02119d5 CM |
3216 | } |
3217 | ||
94edf4ae JB |
3218 | static void fill_inode_item(struct btrfs_trans_handle *trans, |
3219 | struct extent_buffer *leaf, | |
3220 | struct btrfs_inode_item *item, | |
3221 | struct inode *inode, int log_inode_only) | |
3222 | { | |
0b1c6cca JB |
3223 | struct btrfs_map_token token; |
3224 | ||
3225 | btrfs_init_map_token(&token); | |
94edf4ae JB |
3226 | |
3227 | if (log_inode_only) { | |
3228 | /* set the generation to zero so the recover code | |
3229 | * can tell the difference between an logging | |
3230 | * just to say 'this inode exists' and a logging | |
3231 | * to say 'update this inode with these values' | |
3232 | */ | |
0b1c6cca JB |
3233 | btrfs_set_token_inode_generation(leaf, item, 0, &token); |
3234 | btrfs_set_token_inode_size(leaf, item, 0, &token); | |
94edf4ae | 3235 | } else { |
0b1c6cca JB |
3236 | btrfs_set_token_inode_generation(leaf, item, |
3237 | BTRFS_I(inode)->generation, | |
3238 | &token); | |
3239 | btrfs_set_token_inode_size(leaf, item, inode->i_size, &token); | |
3240 | } | |
3241 | ||
3242 | btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token); | |
3243 | btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token); | |
3244 | btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token); | |
3245 | btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token); | |
3246 | ||
3247 | btrfs_set_token_timespec_sec(leaf, btrfs_inode_atime(item), | |
3248 | inode->i_atime.tv_sec, &token); | |
3249 | btrfs_set_token_timespec_nsec(leaf, btrfs_inode_atime(item), | |
3250 | inode->i_atime.tv_nsec, &token); | |
3251 | ||
3252 | btrfs_set_token_timespec_sec(leaf, btrfs_inode_mtime(item), | |
3253 | inode->i_mtime.tv_sec, &token); | |
3254 | btrfs_set_token_timespec_nsec(leaf, btrfs_inode_mtime(item), | |
3255 | inode->i_mtime.tv_nsec, &token); | |
3256 | ||
3257 | btrfs_set_token_timespec_sec(leaf, btrfs_inode_ctime(item), | |
3258 | inode->i_ctime.tv_sec, &token); | |
3259 | btrfs_set_token_timespec_nsec(leaf, btrfs_inode_ctime(item), | |
3260 | inode->i_ctime.tv_nsec, &token); | |
3261 | ||
3262 | btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode), | |
3263 | &token); | |
3264 | ||
3265 | btrfs_set_token_inode_sequence(leaf, item, inode->i_version, &token); | |
3266 | btrfs_set_token_inode_transid(leaf, item, trans->transid, &token); | |
3267 | btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token); | |
3268 | btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token); | |
3269 | btrfs_set_token_inode_block_group(leaf, item, 0, &token); | |
94edf4ae JB |
3270 | } |
3271 | ||
a95249b3 JB |
3272 | static int log_inode_item(struct btrfs_trans_handle *trans, |
3273 | struct btrfs_root *log, struct btrfs_path *path, | |
3274 | struct inode *inode) | |
3275 | { | |
3276 | struct btrfs_inode_item *inode_item; | |
a95249b3 JB |
3277 | int ret; |
3278 | ||
efd0c405 FDBM |
3279 | ret = btrfs_insert_empty_item(trans, log, path, |
3280 | &BTRFS_I(inode)->location, | |
a95249b3 JB |
3281 | sizeof(*inode_item)); |
3282 | if (ret && ret != -EEXIST) | |
3283 | return ret; | |
3284 | inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
3285 | struct btrfs_inode_item); | |
3286 | fill_inode_item(trans, path->nodes[0], inode_item, inode, 0); | |
3287 | btrfs_release_path(path); | |
3288 | return 0; | |
3289 | } | |
3290 | ||
31ff1cd2 | 3291 | static noinline int copy_items(struct btrfs_trans_handle *trans, |
d2794405 | 3292 | struct inode *inode, |
31ff1cd2 | 3293 | struct btrfs_path *dst_path, |
16e7549f | 3294 | struct btrfs_path *src_path, u64 *last_extent, |
31ff1cd2 CM |
3295 | int start_slot, int nr, int inode_only) |
3296 | { | |
3297 | unsigned long src_offset; | |
3298 | unsigned long dst_offset; | |
d2794405 | 3299 | struct btrfs_root *log = BTRFS_I(inode)->root->log_root; |
31ff1cd2 CM |
3300 | struct btrfs_file_extent_item *extent; |
3301 | struct btrfs_inode_item *inode_item; | |
16e7549f JB |
3302 | struct extent_buffer *src = src_path->nodes[0]; |
3303 | struct btrfs_key first_key, last_key, key; | |
31ff1cd2 CM |
3304 | int ret; |
3305 | struct btrfs_key *ins_keys; | |
3306 | u32 *ins_sizes; | |
3307 | char *ins_data; | |
3308 | int i; | |
d20f7043 | 3309 | struct list_head ordered_sums; |
d2794405 | 3310 | int skip_csum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM; |
16e7549f | 3311 | bool has_extents = false; |
74121f7c | 3312 | bool need_find_last_extent = true; |
16e7549f | 3313 | bool done = false; |
d20f7043 CM |
3314 | |
3315 | INIT_LIST_HEAD(&ordered_sums); | |
31ff1cd2 CM |
3316 | |
3317 | ins_data = kmalloc(nr * sizeof(struct btrfs_key) + | |
3318 | nr * sizeof(u32), GFP_NOFS); | |
2a29edc6 | 3319 | if (!ins_data) |
3320 | return -ENOMEM; | |
3321 | ||
16e7549f JB |
3322 | first_key.objectid = (u64)-1; |
3323 | ||
31ff1cd2 CM |
3324 | ins_sizes = (u32 *)ins_data; |
3325 | ins_keys = (struct btrfs_key *)(ins_data + nr * sizeof(u32)); | |
3326 | ||
3327 | for (i = 0; i < nr; i++) { | |
3328 | ins_sizes[i] = btrfs_item_size_nr(src, i + start_slot); | |
3329 | btrfs_item_key_to_cpu(src, ins_keys + i, i + start_slot); | |
3330 | } | |
3331 | ret = btrfs_insert_empty_items(trans, log, dst_path, | |
3332 | ins_keys, ins_sizes, nr); | |
4a500fd1 YZ |
3333 | if (ret) { |
3334 | kfree(ins_data); | |
3335 | return ret; | |
3336 | } | |
31ff1cd2 | 3337 | |
5d4f98a2 | 3338 | for (i = 0; i < nr; i++, dst_path->slots[0]++) { |
31ff1cd2 CM |
3339 | dst_offset = btrfs_item_ptr_offset(dst_path->nodes[0], |
3340 | dst_path->slots[0]); | |
3341 | ||
3342 | src_offset = btrfs_item_ptr_offset(src, start_slot + i); | |
3343 | ||
16e7549f JB |
3344 | if ((i == (nr - 1))) |
3345 | last_key = ins_keys[i]; | |
3346 | ||
94edf4ae | 3347 | if (ins_keys[i].type == BTRFS_INODE_ITEM_KEY) { |
31ff1cd2 CM |
3348 | inode_item = btrfs_item_ptr(dst_path->nodes[0], |
3349 | dst_path->slots[0], | |
3350 | struct btrfs_inode_item); | |
94edf4ae JB |
3351 | fill_inode_item(trans, dst_path->nodes[0], inode_item, |
3352 | inode, inode_only == LOG_INODE_EXISTS); | |
3353 | } else { | |
3354 | copy_extent_buffer(dst_path->nodes[0], src, dst_offset, | |
3355 | src_offset, ins_sizes[i]); | |
31ff1cd2 | 3356 | } |
94edf4ae | 3357 | |
16e7549f JB |
3358 | /* |
3359 | * We set need_find_last_extent here in case we know we were | |
3360 | * processing other items and then walk into the first extent in | |
3361 | * the inode. If we don't hit an extent then nothing changes, | |
3362 | * we'll do the last search the next time around. | |
3363 | */ | |
3364 | if (ins_keys[i].type == BTRFS_EXTENT_DATA_KEY) { | |
3365 | has_extents = true; | |
74121f7c | 3366 | if (first_key.objectid == (u64)-1) |
16e7549f JB |
3367 | first_key = ins_keys[i]; |
3368 | } else { | |
3369 | need_find_last_extent = false; | |
3370 | } | |
3371 | ||
31ff1cd2 CM |
3372 | /* take a reference on file data extents so that truncates |
3373 | * or deletes of this inode don't have to relog the inode | |
3374 | * again | |
3375 | */ | |
962a298f | 3376 | if (ins_keys[i].type == BTRFS_EXTENT_DATA_KEY && |
d2794405 | 3377 | !skip_csum) { |
31ff1cd2 CM |
3378 | int found_type; |
3379 | extent = btrfs_item_ptr(src, start_slot + i, | |
3380 | struct btrfs_file_extent_item); | |
3381 | ||
8e531cdf | 3382 | if (btrfs_file_extent_generation(src, extent) < trans->transid) |
3383 | continue; | |
3384 | ||
31ff1cd2 | 3385 | found_type = btrfs_file_extent_type(src, extent); |
6f1fed77 | 3386 | if (found_type == BTRFS_FILE_EXTENT_REG) { |
5d4f98a2 YZ |
3387 | u64 ds, dl, cs, cl; |
3388 | ds = btrfs_file_extent_disk_bytenr(src, | |
3389 | extent); | |
3390 | /* ds == 0 is a hole */ | |
3391 | if (ds == 0) | |
3392 | continue; | |
3393 | ||
3394 | dl = btrfs_file_extent_disk_num_bytes(src, | |
3395 | extent); | |
3396 | cs = btrfs_file_extent_offset(src, extent); | |
3397 | cl = btrfs_file_extent_num_bytes(src, | |
a419aef8 | 3398 | extent); |
580afd76 CM |
3399 | if (btrfs_file_extent_compression(src, |
3400 | extent)) { | |
3401 | cs = 0; | |
3402 | cl = dl; | |
3403 | } | |
5d4f98a2 YZ |
3404 | |
3405 | ret = btrfs_lookup_csums_range( | |
3406 | log->fs_info->csum_root, | |
3407 | ds + cs, ds + cs + cl - 1, | |
a2de733c | 3408 | &ordered_sums, 0); |
3650860b JB |
3409 | if (ret) { |
3410 | btrfs_release_path(dst_path); | |
3411 | kfree(ins_data); | |
3412 | return ret; | |
3413 | } | |
31ff1cd2 CM |
3414 | } |
3415 | } | |
31ff1cd2 CM |
3416 | } |
3417 | ||
3418 | btrfs_mark_buffer_dirty(dst_path->nodes[0]); | |
b3b4aa74 | 3419 | btrfs_release_path(dst_path); |
31ff1cd2 | 3420 | kfree(ins_data); |
d20f7043 CM |
3421 | |
3422 | /* | |
3423 | * we have to do this after the loop above to avoid changing the | |
3424 | * log tree while trying to change the log tree. | |
3425 | */ | |
4a500fd1 | 3426 | ret = 0; |
d397712b | 3427 | while (!list_empty(&ordered_sums)) { |
d20f7043 CM |
3428 | struct btrfs_ordered_sum *sums = list_entry(ordered_sums.next, |
3429 | struct btrfs_ordered_sum, | |
3430 | list); | |
4a500fd1 YZ |
3431 | if (!ret) |
3432 | ret = btrfs_csum_file_blocks(trans, log, sums); | |
d20f7043 CM |
3433 | list_del(&sums->list); |
3434 | kfree(sums); | |
3435 | } | |
16e7549f JB |
3436 | |
3437 | if (!has_extents) | |
3438 | return ret; | |
3439 | ||
74121f7c FM |
3440 | if (need_find_last_extent && *last_extent == first_key.offset) { |
3441 | /* | |
3442 | * We don't have any leafs between our current one and the one | |
3443 | * we processed before that can have file extent items for our | |
3444 | * inode (and have a generation number smaller than our current | |
3445 | * transaction id). | |
3446 | */ | |
3447 | need_find_last_extent = false; | |
3448 | } | |
3449 | ||
16e7549f JB |
3450 | /* |
3451 | * Because we use btrfs_search_forward we could skip leaves that were | |
3452 | * not modified and then assume *last_extent is valid when it really | |
3453 | * isn't. So back up to the previous leaf and read the end of the last | |
3454 | * extent before we go and fill in holes. | |
3455 | */ | |
3456 | if (need_find_last_extent) { | |
3457 | u64 len; | |
3458 | ||
3459 | ret = btrfs_prev_leaf(BTRFS_I(inode)->root, src_path); | |
3460 | if (ret < 0) | |
3461 | return ret; | |
3462 | if (ret) | |
3463 | goto fill_holes; | |
3464 | if (src_path->slots[0]) | |
3465 | src_path->slots[0]--; | |
3466 | src = src_path->nodes[0]; | |
3467 | btrfs_item_key_to_cpu(src, &key, src_path->slots[0]); | |
3468 | if (key.objectid != btrfs_ino(inode) || | |
3469 | key.type != BTRFS_EXTENT_DATA_KEY) | |
3470 | goto fill_holes; | |
3471 | extent = btrfs_item_ptr(src, src_path->slots[0], | |
3472 | struct btrfs_file_extent_item); | |
3473 | if (btrfs_file_extent_type(src, extent) == | |
3474 | BTRFS_FILE_EXTENT_INLINE) { | |
514ac8ad CM |
3475 | len = btrfs_file_extent_inline_len(src, |
3476 | src_path->slots[0], | |
3477 | extent); | |
16e7549f JB |
3478 | *last_extent = ALIGN(key.offset + len, |
3479 | log->sectorsize); | |
3480 | } else { | |
3481 | len = btrfs_file_extent_num_bytes(src, extent); | |
3482 | *last_extent = key.offset + len; | |
3483 | } | |
3484 | } | |
3485 | fill_holes: | |
3486 | /* So we did prev_leaf, now we need to move to the next leaf, but a few | |
3487 | * things could have happened | |
3488 | * | |
3489 | * 1) A merge could have happened, so we could currently be on a leaf | |
3490 | * that holds what we were copying in the first place. | |
3491 | * 2) A split could have happened, and now not all of the items we want | |
3492 | * are on the same leaf. | |
3493 | * | |
3494 | * So we need to adjust how we search for holes, we need to drop the | |
3495 | * path and re-search for the first extent key we found, and then walk | |
3496 | * forward until we hit the last one we copied. | |
3497 | */ | |
3498 | if (need_find_last_extent) { | |
3499 | /* btrfs_prev_leaf could return 1 without releasing the path */ | |
3500 | btrfs_release_path(src_path); | |
3501 | ret = btrfs_search_slot(NULL, BTRFS_I(inode)->root, &first_key, | |
3502 | src_path, 0, 0); | |
3503 | if (ret < 0) | |
3504 | return ret; | |
3505 | ASSERT(ret == 0); | |
3506 | src = src_path->nodes[0]; | |
3507 | i = src_path->slots[0]; | |
3508 | } else { | |
3509 | i = start_slot; | |
3510 | } | |
3511 | ||
3512 | /* | |
3513 | * Ok so here we need to go through and fill in any holes we may have | |
3514 | * to make sure that holes are punched for those areas in case they had | |
3515 | * extents previously. | |
3516 | */ | |
3517 | while (!done) { | |
3518 | u64 offset, len; | |
3519 | u64 extent_end; | |
3520 | ||
3521 | if (i >= btrfs_header_nritems(src_path->nodes[0])) { | |
3522 | ret = btrfs_next_leaf(BTRFS_I(inode)->root, src_path); | |
3523 | if (ret < 0) | |
3524 | return ret; | |
3525 | ASSERT(ret == 0); | |
3526 | src = src_path->nodes[0]; | |
3527 | i = 0; | |
3528 | } | |
3529 | ||
3530 | btrfs_item_key_to_cpu(src, &key, i); | |
3531 | if (!btrfs_comp_cpu_keys(&key, &last_key)) | |
3532 | done = true; | |
3533 | if (key.objectid != btrfs_ino(inode) || | |
3534 | key.type != BTRFS_EXTENT_DATA_KEY) { | |
3535 | i++; | |
3536 | continue; | |
3537 | } | |
3538 | extent = btrfs_item_ptr(src, i, struct btrfs_file_extent_item); | |
3539 | if (btrfs_file_extent_type(src, extent) == | |
3540 | BTRFS_FILE_EXTENT_INLINE) { | |
514ac8ad | 3541 | len = btrfs_file_extent_inline_len(src, i, extent); |
16e7549f JB |
3542 | extent_end = ALIGN(key.offset + len, log->sectorsize); |
3543 | } else { | |
3544 | len = btrfs_file_extent_num_bytes(src, extent); | |
3545 | extent_end = key.offset + len; | |
3546 | } | |
3547 | i++; | |
3548 | ||
3549 | if (*last_extent == key.offset) { | |
3550 | *last_extent = extent_end; | |
3551 | continue; | |
3552 | } | |
3553 | offset = *last_extent; | |
3554 | len = key.offset - *last_extent; | |
3555 | ret = btrfs_insert_file_extent(trans, log, btrfs_ino(inode), | |
3556 | offset, 0, 0, len, 0, len, 0, | |
3557 | 0, 0); | |
3558 | if (ret) | |
3559 | break; | |
74121f7c | 3560 | *last_extent = extent_end; |
16e7549f JB |
3561 | } |
3562 | /* | |
3563 | * Need to let the callers know we dropped the path so they should | |
3564 | * re-search. | |
3565 | */ | |
3566 | if (!ret && need_find_last_extent) | |
3567 | ret = 1; | |
4a500fd1 | 3568 | return ret; |
31ff1cd2 CM |
3569 | } |
3570 | ||
5dc562c5 JB |
3571 | static int extent_cmp(void *priv, struct list_head *a, struct list_head *b) |
3572 | { | |
3573 | struct extent_map *em1, *em2; | |
3574 | ||
3575 | em1 = list_entry(a, struct extent_map, list); | |
3576 | em2 = list_entry(b, struct extent_map, list); | |
3577 | ||
3578 | if (em1->start < em2->start) | |
3579 | return -1; | |
3580 | else if (em1->start > em2->start) | |
3581 | return 1; | |
3582 | return 0; | |
3583 | } | |
3584 | ||
8407f553 FM |
3585 | static int wait_ordered_extents(struct btrfs_trans_handle *trans, |
3586 | struct inode *inode, | |
3587 | struct btrfs_root *root, | |
3588 | const struct extent_map *em, | |
3589 | const struct list_head *logged_list, | |
3590 | bool *ordered_io_error) | |
5dc562c5 | 3591 | { |
2ab28f32 | 3592 | struct btrfs_ordered_extent *ordered; |
8407f553 | 3593 | struct btrfs_root *log = root->log_root; |
2ab28f32 JB |
3594 | u64 mod_start = em->mod_start; |
3595 | u64 mod_len = em->mod_len; | |
8407f553 | 3596 | const bool skip_csum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM; |
2ab28f32 JB |
3597 | u64 csum_offset; |
3598 | u64 csum_len; | |
8407f553 FM |
3599 | LIST_HEAD(ordered_sums); |
3600 | int ret = 0; | |
0aa4a17d | 3601 | |
8407f553 | 3602 | *ordered_io_error = false; |
0aa4a17d | 3603 | |
8407f553 FM |
3604 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) || |
3605 | em->block_start == EXTENT_MAP_HOLE) | |
70c8a91c | 3606 | return 0; |
5dc562c5 | 3607 | |
2ab28f32 | 3608 | /* |
8407f553 FM |
3609 | * Wait far any ordered extent that covers our extent map. If it |
3610 | * finishes without an error, first check and see if our csums are on | |
3611 | * our outstanding ordered extents. | |
2ab28f32 | 3612 | */ |
827463c4 | 3613 | list_for_each_entry(ordered, logged_list, log_list) { |
2ab28f32 JB |
3614 | struct btrfs_ordered_sum *sum; |
3615 | ||
3616 | if (!mod_len) | |
3617 | break; | |
3618 | ||
2ab28f32 JB |
3619 | if (ordered->file_offset + ordered->len <= mod_start || |
3620 | mod_start + mod_len <= ordered->file_offset) | |
3621 | continue; | |
3622 | ||
8407f553 FM |
3623 | if (!test_bit(BTRFS_ORDERED_IO_DONE, &ordered->flags) && |
3624 | !test_bit(BTRFS_ORDERED_IOERR, &ordered->flags) && | |
3625 | !test_bit(BTRFS_ORDERED_DIRECT, &ordered->flags)) { | |
3626 | const u64 start = ordered->file_offset; | |
3627 | const u64 end = ordered->file_offset + ordered->len - 1; | |
3628 | ||
3629 | WARN_ON(ordered->inode != inode); | |
3630 | filemap_fdatawrite_range(inode->i_mapping, start, end); | |
3631 | } | |
3632 | ||
3633 | wait_event(ordered->wait, | |
3634 | (test_bit(BTRFS_ORDERED_IO_DONE, &ordered->flags) || | |
3635 | test_bit(BTRFS_ORDERED_IOERR, &ordered->flags))); | |
3636 | ||
3637 | if (test_bit(BTRFS_ORDERED_IOERR, &ordered->flags)) { | |
b38ef71c FM |
3638 | /* |
3639 | * Clear the AS_EIO/AS_ENOSPC flags from the inode's | |
3640 | * i_mapping flags, so that the next fsync won't get | |
3641 | * an outdated io error too. | |
3642 | */ | |
3643 | btrfs_inode_check_errors(inode); | |
8407f553 FM |
3644 | *ordered_io_error = true; |
3645 | break; | |
3646 | } | |
2ab28f32 JB |
3647 | /* |
3648 | * We are going to copy all the csums on this ordered extent, so | |
3649 | * go ahead and adjust mod_start and mod_len in case this | |
3650 | * ordered extent has already been logged. | |
3651 | */ | |
3652 | if (ordered->file_offset > mod_start) { | |
3653 | if (ordered->file_offset + ordered->len >= | |
3654 | mod_start + mod_len) | |
3655 | mod_len = ordered->file_offset - mod_start; | |
3656 | /* | |
3657 | * If we have this case | |
3658 | * | |
3659 | * |--------- logged extent ---------| | |
3660 | * |----- ordered extent ----| | |
3661 | * | |
3662 | * Just don't mess with mod_start and mod_len, we'll | |
3663 | * just end up logging more csums than we need and it | |
3664 | * will be ok. | |
3665 | */ | |
3666 | } else { | |
3667 | if (ordered->file_offset + ordered->len < | |
3668 | mod_start + mod_len) { | |
3669 | mod_len = (mod_start + mod_len) - | |
3670 | (ordered->file_offset + ordered->len); | |
3671 | mod_start = ordered->file_offset + | |
3672 | ordered->len; | |
3673 | } else { | |
3674 | mod_len = 0; | |
3675 | } | |
3676 | } | |
3677 | ||
8407f553 FM |
3678 | if (skip_csum) |
3679 | continue; | |
3680 | ||
2ab28f32 JB |
3681 | /* |
3682 | * To keep us from looping for the above case of an ordered | |
3683 | * extent that falls inside of the logged extent. | |
3684 | */ | |
3685 | if (test_and_set_bit(BTRFS_ORDERED_LOGGED_CSUM, | |
3686 | &ordered->flags)) | |
3687 | continue; | |
2ab28f32 | 3688 | |
23c671a5 MX |
3689 | if (ordered->csum_bytes_left) { |
3690 | btrfs_start_ordered_extent(inode, ordered, 0); | |
3691 | wait_event(ordered->wait, | |
3692 | ordered->csum_bytes_left == 0); | |
3693 | } | |
2ab28f32 JB |
3694 | |
3695 | list_for_each_entry(sum, &ordered->list, list) { | |
3696 | ret = btrfs_csum_file_blocks(trans, log, sum); | |
827463c4 | 3697 | if (ret) |
8407f553 | 3698 | break; |
2ab28f32 | 3699 | } |
2ab28f32 | 3700 | } |
2ab28f32 | 3701 | |
8407f553 | 3702 | if (*ordered_io_error || !mod_len || ret || skip_csum) |
2ab28f32 JB |
3703 | return ret; |
3704 | ||
488111aa FDBM |
3705 | if (em->compress_type) { |
3706 | csum_offset = 0; | |
8407f553 | 3707 | csum_len = max(em->block_len, em->orig_block_len); |
488111aa FDBM |
3708 | } else { |
3709 | csum_offset = mod_start - em->start; | |
3710 | csum_len = mod_len; | |
3711 | } | |
2ab28f32 | 3712 | |
70c8a91c JB |
3713 | /* block start is already adjusted for the file extent offset. */ |
3714 | ret = btrfs_lookup_csums_range(log->fs_info->csum_root, | |
3715 | em->block_start + csum_offset, | |
3716 | em->block_start + csum_offset + | |
3717 | csum_len - 1, &ordered_sums, 0); | |
3718 | if (ret) | |
3719 | return ret; | |
5dc562c5 | 3720 | |
70c8a91c JB |
3721 | while (!list_empty(&ordered_sums)) { |
3722 | struct btrfs_ordered_sum *sums = list_entry(ordered_sums.next, | |
3723 | struct btrfs_ordered_sum, | |
3724 | list); | |
3725 | if (!ret) | |
3726 | ret = btrfs_csum_file_blocks(trans, log, sums); | |
3727 | list_del(&sums->list); | |
3728 | kfree(sums); | |
5dc562c5 JB |
3729 | } |
3730 | ||
70c8a91c | 3731 | return ret; |
5dc562c5 JB |
3732 | } |
3733 | ||
8407f553 FM |
3734 | static int log_one_extent(struct btrfs_trans_handle *trans, |
3735 | struct inode *inode, struct btrfs_root *root, | |
3736 | const struct extent_map *em, | |
3737 | struct btrfs_path *path, | |
3738 | const struct list_head *logged_list, | |
3739 | struct btrfs_log_ctx *ctx) | |
3740 | { | |
3741 | struct btrfs_root *log = root->log_root; | |
3742 | struct btrfs_file_extent_item *fi; | |
3743 | struct extent_buffer *leaf; | |
3744 | struct btrfs_map_token token; | |
3745 | struct btrfs_key key; | |
3746 | u64 extent_offset = em->start - em->orig_start; | |
3747 | u64 block_len; | |
3748 | int ret; | |
3749 | int extent_inserted = 0; | |
3750 | bool ordered_io_err = false; | |
3751 | ||
3752 | ret = wait_ordered_extents(trans, inode, root, em, logged_list, | |
3753 | &ordered_io_err); | |
3754 | if (ret) | |
3755 | return ret; | |
3756 | ||
3757 | if (ordered_io_err) { | |
3758 | ctx->io_err = -EIO; | |
3759 | return 0; | |
3760 | } | |
3761 | ||
3762 | btrfs_init_map_token(&token); | |
3763 | ||
3764 | ret = __btrfs_drop_extents(trans, log, inode, path, em->start, | |
3765 | em->start + em->len, NULL, 0, 1, | |
3766 | sizeof(*fi), &extent_inserted); | |
3767 | if (ret) | |
3768 | return ret; | |
3769 | ||
3770 | if (!extent_inserted) { | |
3771 | key.objectid = btrfs_ino(inode); | |
3772 | key.type = BTRFS_EXTENT_DATA_KEY; | |
3773 | key.offset = em->start; | |
3774 | ||
3775 | ret = btrfs_insert_empty_item(trans, log, path, &key, | |
3776 | sizeof(*fi)); | |
3777 | if (ret) | |
3778 | return ret; | |
3779 | } | |
3780 | leaf = path->nodes[0]; | |
3781 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
3782 | struct btrfs_file_extent_item); | |
3783 | ||
50d9aa99 | 3784 | btrfs_set_token_file_extent_generation(leaf, fi, trans->transid, |
8407f553 FM |
3785 | &token); |
3786 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) | |
3787 | btrfs_set_token_file_extent_type(leaf, fi, | |
3788 | BTRFS_FILE_EXTENT_PREALLOC, | |
3789 | &token); | |
3790 | else | |
3791 | btrfs_set_token_file_extent_type(leaf, fi, | |
3792 | BTRFS_FILE_EXTENT_REG, | |
3793 | &token); | |
3794 | ||
3795 | block_len = max(em->block_len, em->orig_block_len); | |
3796 | if (em->compress_type != BTRFS_COMPRESS_NONE) { | |
3797 | btrfs_set_token_file_extent_disk_bytenr(leaf, fi, | |
3798 | em->block_start, | |
3799 | &token); | |
3800 | btrfs_set_token_file_extent_disk_num_bytes(leaf, fi, block_len, | |
3801 | &token); | |
3802 | } else if (em->block_start < EXTENT_MAP_LAST_BYTE) { | |
3803 | btrfs_set_token_file_extent_disk_bytenr(leaf, fi, | |
3804 | em->block_start - | |
3805 | extent_offset, &token); | |
3806 | btrfs_set_token_file_extent_disk_num_bytes(leaf, fi, block_len, | |
3807 | &token); | |
3808 | } else { | |
3809 | btrfs_set_token_file_extent_disk_bytenr(leaf, fi, 0, &token); | |
3810 | btrfs_set_token_file_extent_disk_num_bytes(leaf, fi, 0, | |
3811 | &token); | |
3812 | } | |
3813 | ||
3814 | btrfs_set_token_file_extent_offset(leaf, fi, extent_offset, &token); | |
3815 | btrfs_set_token_file_extent_num_bytes(leaf, fi, em->len, &token); | |
3816 | btrfs_set_token_file_extent_ram_bytes(leaf, fi, em->ram_bytes, &token); | |
3817 | btrfs_set_token_file_extent_compression(leaf, fi, em->compress_type, | |
3818 | &token); | |
3819 | btrfs_set_token_file_extent_encryption(leaf, fi, 0, &token); | |
3820 | btrfs_set_token_file_extent_other_encoding(leaf, fi, 0, &token); | |
3821 | btrfs_mark_buffer_dirty(leaf); | |
3822 | ||
3823 | btrfs_release_path(path); | |
3824 | ||
3825 | return ret; | |
3826 | } | |
3827 | ||
5dc562c5 JB |
3828 | static int btrfs_log_changed_extents(struct btrfs_trans_handle *trans, |
3829 | struct btrfs_root *root, | |
3830 | struct inode *inode, | |
827463c4 | 3831 | struct btrfs_path *path, |
8407f553 FM |
3832 | struct list_head *logged_list, |
3833 | struct btrfs_log_ctx *ctx) | |
5dc562c5 | 3834 | { |
5dc562c5 JB |
3835 | struct extent_map *em, *n; |
3836 | struct list_head extents; | |
3837 | struct extent_map_tree *tree = &BTRFS_I(inode)->extent_tree; | |
3838 | u64 test_gen; | |
3839 | int ret = 0; | |
2ab28f32 | 3840 | int num = 0; |
5dc562c5 JB |
3841 | |
3842 | INIT_LIST_HEAD(&extents); | |
3843 | ||
5dc562c5 JB |
3844 | write_lock(&tree->lock); |
3845 | test_gen = root->fs_info->last_trans_committed; | |
3846 | ||
3847 | list_for_each_entry_safe(em, n, &tree->modified_extents, list) { | |
3848 | list_del_init(&em->list); | |
2ab28f32 JB |
3849 | |
3850 | /* | |
3851 | * Just an arbitrary number, this can be really CPU intensive | |
3852 | * once we start getting a lot of extents, and really once we | |
3853 | * have a bunch of extents we just want to commit since it will | |
3854 | * be faster. | |
3855 | */ | |
3856 | if (++num > 32768) { | |
3857 | list_del_init(&tree->modified_extents); | |
3858 | ret = -EFBIG; | |
3859 | goto process; | |
3860 | } | |
3861 | ||
5dc562c5 JB |
3862 | if (em->generation <= test_gen) |
3863 | continue; | |
ff44c6e3 JB |
3864 | /* Need a ref to keep it from getting evicted from cache */ |
3865 | atomic_inc(&em->refs); | |
3866 | set_bit(EXTENT_FLAG_LOGGING, &em->flags); | |
5dc562c5 | 3867 | list_add_tail(&em->list, &extents); |
2ab28f32 | 3868 | num++; |
5dc562c5 JB |
3869 | } |
3870 | ||
3871 | list_sort(NULL, &extents, extent_cmp); | |
3872 | ||
2ab28f32 | 3873 | process: |
5dc562c5 JB |
3874 | while (!list_empty(&extents)) { |
3875 | em = list_entry(extents.next, struct extent_map, list); | |
3876 | ||
3877 | list_del_init(&em->list); | |
3878 | ||
3879 | /* | |
3880 | * If we had an error we just need to delete everybody from our | |
3881 | * private list. | |
3882 | */ | |
ff44c6e3 | 3883 | if (ret) { |
201a9038 | 3884 | clear_em_logging(tree, em); |
ff44c6e3 | 3885 | free_extent_map(em); |
5dc562c5 | 3886 | continue; |
ff44c6e3 JB |
3887 | } |
3888 | ||
3889 | write_unlock(&tree->lock); | |
5dc562c5 | 3890 | |
8407f553 FM |
3891 | ret = log_one_extent(trans, inode, root, em, path, logged_list, |
3892 | ctx); | |
ff44c6e3 | 3893 | write_lock(&tree->lock); |
201a9038 JB |
3894 | clear_em_logging(tree, em); |
3895 | free_extent_map(em); | |
5dc562c5 | 3896 | } |
ff44c6e3 JB |
3897 | WARN_ON(!list_empty(&extents)); |
3898 | write_unlock(&tree->lock); | |
5dc562c5 | 3899 | |
5dc562c5 | 3900 | btrfs_release_path(path); |
5dc562c5 JB |
3901 | return ret; |
3902 | } | |
3903 | ||
e02119d5 CM |
3904 | /* log a single inode in the tree log. |
3905 | * At least one parent directory for this inode must exist in the tree | |
3906 | * or be logged already. | |
3907 | * | |
3908 | * Any items from this inode changed by the current transaction are copied | |
3909 | * to the log tree. An extra reference is taken on any extents in this | |
3910 | * file, allowing us to avoid a whole pile of corner cases around logging | |
3911 | * blocks that have been removed from the tree. | |
3912 | * | |
3913 | * See LOG_INODE_ALL and related defines for a description of what inode_only | |
3914 | * does. | |
3915 | * | |
3916 | * This handles both files and directories. | |
3917 | */ | |
12fcfd22 | 3918 | static int btrfs_log_inode(struct btrfs_trans_handle *trans, |
49dae1bc FM |
3919 | struct btrfs_root *root, struct inode *inode, |
3920 | int inode_only, | |
3921 | const loff_t start, | |
8407f553 FM |
3922 | const loff_t end, |
3923 | struct btrfs_log_ctx *ctx) | |
e02119d5 CM |
3924 | { |
3925 | struct btrfs_path *path; | |
3926 | struct btrfs_path *dst_path; | |
3927 | struct btrfs_key min_key; | |
3928 | struct btrfs_key max_key; | |
3929 | struct btrfs_root *log = root->log_root; | |
31ff1cd2 | 3930 | struct extent_buffer *src = NULL; |
827463c4 | 3931 | LIST_HEAD(logged_list); |
16e7549f | 3932 | u64 last_extent = 0; |
4a500fd1 | 3933 | int err = 0; |
e02119d5 | 3934 | int ret; |
3a5f1d45 | 3935 | int nritems; |
31ff1cd2 CM |
3936 | int ins_start_slot = 0; |
3937 | int ins_nr; | |
5dc562c5 | 3938 | bool fast_search = false; |
33345d01 | 3939 | u64 ino = btrfs_ino(inode); |
49dae1bc | 3940 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; |
e02119d5 | 3941 | |
e02119d5 | 3942 | path = btrfs_alloc_path(); |
5df67083 TI |
3943 | if (!path) |
3944 | return -ENOMEM; | |
e02119d5 | 3945 | dst_path = btrfs_alloc_path(); |
5df67083 TI |
3946 | if (!dst_path) { |
3947 | btrfs_free_path(path); | |
3948 | return -ENOMEM; | |
3949 | } | |
e02119d5 | 3950 | |
33345d01 | 3951 | min_key.objectid = ino; |
e02119d5 CM |
3952 | min_key.type = BTRFS_INODE_ITEM_KEY; |
3953 | min_key.offset = 0; | |
3954 | ||
33345d01 | 3955 | max_key.objectid = ino; |
12fcfd22 | 3956 | |
12fcfd22 | 3957 | |
5dc562c5 | 3958 | /* today the code can only do partial logging of directories */ |
5269b67e MX |
3959 | if (S_ISDIR(inode->i_mode) || |
3960 | (!test_bit(BTRFS_INODE_NEEDS_FULL_SYNC, | |
3961 | &BTRFS_I(inode)->runtime_flags) && | |
3962 | inode_only == LOG_INODE_EXISTS)) | |
e02119d5 CM |
3963 | max_key.type = BTRFS_XATTR_ITEM_KEY; |
3964 | else | |
3965 | max_key.type = (u8)-1; | |
3966 | max_key.offset = (u64)-1; | |
3967 | ||
94edf4ae JB |
3968 | /* Only run delayed items if we are a dir or a new file */ |
3969 | if (S_ISDIR(inode->i_mode) || | |
3970 | BTRFS_I(inode)->generation > root->fs_info->last_trans_committed) { | |
3971 | ret = btrfs_commit_inode_delayed_items(trans, inode); | |
3972 | if (ret) { | |
3973 | btrfs_free_path(path); | |
3974 | btrfs_free_path(dst_path); | |
3975 | return ret; | |
3976 | } | |
16cdcec7 MX |
3977 | } |
3978 | ||
e02119d5 CM |
3979 | mutex_lock(&BTRFS_I(inode)->log_mutex); |
3980 | ||
0870295b | 3981 | btrfs_get_logged_extents(inode, &logged_list, start, end); |
2ab28f32 | 3982 | |
e02119d5 CM |
3983 | /* |
3984 | * a brute force approach to making sure we get the most uptodate | |
3985 | * copies of everything. | |
3986 | */ | |
3987 | if (S_ISDIR(inode->i_mode)) { | |
3988 | int max_key_type = BTRFS_DIR_LOG_INDEX_KEY; | |
3989 | ||
3990 | if (inode_only == LOG_INODE_EXISTS) | |
3991 | max_key_type = BTRFS_XATTR_ITEM_KEY; | |
33345d01 | 3992 | ret = drop_objectid_items(trans, log, path, ino, max_key_type); |
e02119d5 | 3993 | } else { |
5dc562c5 JB |
3994 | if (test_and_clear_bit(BTRFS_INODE_NEEDS_FULL_SYNC, |
3995 | &BTRFS_I(inode)->runtime_flags)) { | |
e9976151 JB |
3996 | clear_bit(BTRFS_INODE_COPY_EVERYTHING, |
3997 | &BTRFS_I(inode)->runtime_flags); | |
5dc562c5 JB |
3998 | ret = btrfs_truncate_inode_items(trans, log, |
3999 | inode, 0, 0); | |
a95249b3 | 4000 | } else if (test_and_clear_bit(BTRFS_INODE_COPY_EVERYTHING, |
6cfab851 JB |
4001 | &BTRFS_I(inode)->runtime_flags) || |
4002 | inode_only == LOG_INODE_EXISTS) { | |
183f37fa LB |
4003 | if (inode_only == LOG_INODE_ALL) |
4004 | fast_search = true; | |
a95249b3 | 4005 | max_key.type = BTRFS_XATTR_ITEM_KEY; |
5dc562c5 | 4006 | ret = drop_objectid_items(trans, log, path, ino, |
e9976151 | 4007 | max_key.type); |
a95249b3 JB |
4008 | } else { |
4009 | if (inode_only == LOG_INODE_ALL) | |
4010 | fast_search = true; | |
4011 | ret = log_inode_item(trans, log, dst_path, inode); | |
4012 | if (ret) { | |
4013 | err = ret; | |
4014 | goto out_unlock; | |
4015 | } | |
4016 | goto log_extents; | |
5dc562c5 | 4017 | } |
a95249b3 | 4018 | |
e02119d5 | 4019 | } |
4a500fd1 YZ |
4020 | if (ret) { |
4021 | err = ret; | |
4022 | goto out_unlock; | |
4023 | } | |
e02119d5 | 4024 | |
d397712b | 4025 | while (1) { |
31ff1cd2 | 4026 | ins_nr = 0; |
6174d3cb | 4027 | ret = btrfs_search_forward(root, &min_key, |
de78b51a | 4028 | path, trans->transid); |
e02119d5 CM |
4029 | if (ret != 0) |
4030 | break; | |
3a5f1d45 | 4031 | again: |
31ff1cd2 | 4032 | /* note, ins_nr might be > 0 here, cleanup outside the loop */ |
33345d01 | 4033 | if (min_key.objectid != ino) |
e02119d5 CM |
4034 | break; |
4035 | if (min_key.type > max_key.type) | |
4036 | break; | |
31ff1cd2 | 4037 | |
e02119d5 | 4038 | src = path->nodes[0]; |
31ff1cd2 CM |
4039 | if (ins_nr && ins_start_slot + ins_nr == path->slots[0]) { |
4040 | ins_nr++; | |
4041 | goto next_slot; | |
4042 | } else if (!ins_nr) { | |
4043 | ins_start_slot = path->slots[0]; | |
4044 | ins_nr = 1; | |
4045 | goto next_slot; | |
e02119d5 CM |
4046 | } |
4047 | ||
16e7549f JB |
4048 | ret = copy_items(trans, inode, dst_path, path, &last_extent, |
4049 | ins_start_slot, ins_nr, inode_only); | |
4050 | if (ret < 0) { | |
4a500fd1 YZ |
4051 | err = ret; |
4052 | goto out_unlock; | |
a71db86e RV |
4053 | } |
4054 | if (ret) { | |
16e7549f JB |
4055 | ins_nr = 0; |
4056 | btrfs_release_path(path); | |
4057 | continue; | |
4a500fd1 | 4058 | } |
31ff1cd2 CM |
4059 | ins_nr = 1; |
4060 | ins_start_slot = path->slots[0]; | |
4061 | next_slot: | |
e02119d5 | 4062 | |
3a5f1d45 CM |
4063 | nritems = btrfs_header_nritems(path->nodes[0]); |
4064 | path->slots[0]++; | |
4065 | if (path->slots[0] < nritems) { | |
4066 | btrfs_item_key_to_cpu(path->nodes[0], &min_key, | |
4067 | path->slots[0]); | |
4068 | goto again; | |
4069 | } | |
31ff1cd2 | 4070 | if (ins_nr) { |
16e7549f JB |
4071 | ret = copy_items(trans, inode, dst_path, path, |
4072 | &last_extent, ins_start_slot, | |
31ff1cd2 | 4073 | ins_nr, inode_only); |
16e7549f | 4074 | if (ret < 0) { |
4a500fd1 YZ |
4075 | err = ret; |
4076 | goto out_unlock; | |
4077 | } | |
16e7549f | 4078 | ret = 0; |
31ff1cd2 CM |
4079 | ins_nr = 0; |
4080 | } | |
b3b4aa74 | 4081 | btrfs_release_path(path); |
3a5f1d45 | 4082 | |
3d41d702 | 4083 | if (min_key.offset < (u64)-1) { |
e02119d5 | 4084 | min_key.offset++; |
3d41d702 | 4085 | } else if (min_key.type < max_key.type) { |
e02119d5 | 4086 | min_key.type++; |
3d41d702 FDBM |
4087 | min_key.offset = 0; |
4088 | } else { | |
e02119d5 | 4089 | break; |
3d41d702 | 4090 | } |
e02119d5 | 4091 | } |
31ff1cd2 | 4092 | if (ins_nr) { |
16e7549f JB |
4093 | ret = copy_items(trans, inode, dst_path, path, &last_extent, |
4094 | ins_start_slot, ins_nr, inode_only); | |
4095 | if (ret < 0) { | |
4a500fd1 YZ |
4096 | err = ret; |
4097 | goto out_unlock; | |
4098 | } | |
16e7549f | 4099 | ret = 0; |
31ff1cd2 CM |
4100 | ins_nr = 0; |
4101 | } | |
5dc562c5 | 4102 | |
a95249b3 | 4103 | log_extents: |
f3b15ccd JB |
4104 | btrfs_release_path(path); |
4105 | btrfs_release_path(dst_path); | |
5dc562c5 | 4106 | if (fast_search) { |
b38ef71c FM |
4107 | /* |
4108 | * Some ordered extents started by fsync might have completed | |
4109 | * before we collected the ordered extents in logged_list, which | |
4110 | * means they're gone, not in our logged_list nor in the inode's | |
4111 | * ordered tree. We want the application/user space to know an | |
4112 | * error happened while attempting to persist file data so that | |
4113 | * it can take proper action. If such error happened, we leave | |
4114 | * without writing to the log tree and the fsync must report the | |
4115 | * file data write error and not commit the current transaction. | |
4116 | */ | |
4117 | err = btrfs_inode_check_errors(inode); | |
4118 | if (err) { | |
4119 | ctx->io_err = err; | |
4120 | goto out_unlock; | |
4121 | } | |
827463c4 | 4122 | ret = btrfs_log_changed_extents(trans, root, inode, dst_path, |
8407f553 | 4123 | &logged_list, ctx); |
5dc562c5 JB |
4124 | if (ret) { |
4125 | err = ret; | |
4126 | goto out_unlock; | |
4127 | } | |
d006a048 | 4128 | } else if (inode_only == LOG_INODE_ALL) { |
06d3d22b LB |
4129 | struct extent_map *em, *n; |
4130 | ||
49dae1bc FM |
4131 | write_lock(&em_tree->lock); |
4132 | /* | |
4133 | * We can't just remove every em if we're called for a ranged | |
4134 | * fsync - that is, one that doesn't cover the whole possible | |
4135 | * file range (0 to LLONG_MAX). This is because we can have | |
4136 | * em's that fall outside the range we're logging and therefore | |
4137 | * their ordered operations haven't completed yet | |
4138 | * (btrfs_finish_ordered_io() not invoked yet). This means we | |
4139 | * didn't get their respective file extent item in the fs/subvol | |
4140 | * tree yet, and need to let the next fast fsync (one which | |
4141 | * consults the list of modified extent maps) find the em so | |
4142 | * that it logs a matching file extent item and waits for the | |
4143 | * respective ordered operation to complete (if it's still | |
4144 | * running). | |
4145 | * | |
4146 | * Removing every em outside the range we're logging would make | |
4147 | * the next fast fsync not log their matching file extent items, | |
4148 | * therefore making us lose data after a log replay. | |
4149 | */ | |
4150 | list_for_each_entry_safe(em, n, &em_tree->modified_extents, | |
4151 | list) { | |
4152 | const u64 mod_end = em->mod_start + em->mod_len - 1; | |
4153 | ||
4154 | if (em->mod_start >= start && mod_end <= end) | |
4155 | list_del_init(&em->list); | |
4156 | } | |
4157 | write_unlock(&em_tree->lock); | |
5dc562c5 JB |
4158 | } |
4159 | ||
9623f9a3 | 4160 | if (inode_only == LOG_INODE_ALL && S_ISDIR(inode->i_mode)) { |
e02119d5 | 4161 | ret = log_directory_changes(trans, root, inode, path, dst_path); |
4a500fd1 YZ |
4162 | if (ret) { |
4163 | err = ret; | |
4164 | goto out_unlock; | |
4165 | } | |
e02119d5 | 4166 | } |
49dae1bc | 4167 | |
125c4cf9 FM |
4168 | BTRFS_I(inode)->logged_trans = trans->transid; |
4169 | BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->last_sub_trans; | |
4a500fd1 | 4170 | out_unlock: |
827463c4 MX |
4171 | if (unlikely(err)) |
4172 | btrfs_put_logged_extents(&logged_list); | |
4173 | else | |
4174 | btrfs_submit_logged_extents(&logged_list, log); | |
e02119d5 CM |
4175 | mutex_unlock(&BTRFS_I(inode)->log_mutex); |
4176 | ||
4177 | btrfs_free_path(path); | |
4178 | btrfs_free_path(dst_path); | |
4a500fd1 | 4179 | return err; |
e02119d5 CM |
4180 | } |
4181 | ||
12fcfd22 CM |
4182 | /* |
4183 | * follow the dentry parent pointers up the chain and see if any | |
4184 | * of the directories in it require a full commit before they can | |
4185 | * be logged. Returns zero if nothing special needs to be done or 1 if | |
4186 | * a full commit is required. | |
4187 | */ | |
4188 | static noinline int check_parent_dirs_for_sync(struct btrfs_trans_handle *trans, | |
4189 | struct inode *inode, | |
4190 | struct dentry *parent, | |
4191 | struct super_block *sb, | |
4192 | u64 last_committed) | |
e02119d5 | 4193 | { |
12fcfd22 CM |
4194 | int ret = 0; |
4195 | struct btrfs_root *root; | |
6a912213 | 4196 | struct dentry *old_parent = NULL; |
de2b530b | 4197 | struct inode *orig_inode = inode; |
e02119d5 | 4198 | |
af4176b4 CM |
4199 | /* |
4200 | * for regular files, if its inode is already on disk, we don't | |
4201 | * have to worry about the parents at all. This is because | |
4202 | * we can use the last_unlink_trans field to record renames | |
4203 | * and other fun in this file. | |
4204 | */ | |
4205 | if (S_ISREG(inode->i_mode) && | |
4206 | BTRFS_I(inode)->generation <= last_committed && | |
4207 | BTRFS_I(inode)->last_unlink_trans <= last_committed) | |
4208 | goto out; | |
4209 | ||
12fcfd22 CM |
4210 | if (!S_ISDIR(inode->i_mode)) { |
4211 | if (!parent || !parent->d_inode || sb != parent->d_inode->i_sb) | |
4212 | goto out; | |
4213 | inode = parent->d_inode; | |
4214 | } | |
4215 | ||
4216 | while (1) { | |
de2b530b JB |
4217 | /* |
4218 | * If we are logging a directory then we start with our inode, | |
4219 | * not our parents inode, so we need to skipp setting the | |
4220 | * logged_trans so that further down in the log code we don't | |
4221 | * think this inode has already been logged. | |
4222 | */ | |
4223 | if (inode != orig_inode) | |
4224 | BTRFS_I(inode)->logged_trans = trans->transid; | |
12fcfd22 CM |
4225 | smp_mb(); |
4226 | ||
4227 | if (BTRFS_I(inode)->last_unlink_trans > last_committed) { | |
4228 | root = BTRFS_I(inode)->root; | |
4229 | ||
4230 | /* | |
4231 | * make sure any commits to the log are forced | |
4232 | * to be full commits | |
4233 | */ | |
995946dd | 4234 | btrfs_set_log_full_commit(root->fs_info, trans); |
12fcfd22 CM |
4235 | ret = 1; |
4236 | break; | |
4237 | } | |
4238 | ||
4239 | if (!parent || !parent->d_inode || sb != parent->d_inode->i_sb) | |
4240 | break; | |
4241 | ||
76dda93c | 4242 | if (IS_ROOT(parent)) |
12fcfd22 CM |
4243 | break; |
4244 | ||
6a912213 JB |
4245 | parent = dget_parent(parent); |
4246 | dput(old_parent); | |
4247 | old_parent = parent; | |
12fcfd22 CM |
4248 | inode = parent->d_inode; |
4249 | ||
4250 | } | |
6a912213 | 4251 | dput(old_parent); |
12fcfd22 | 4252 | out: |
e02119d5 CM |
4253 | return ret; |
4254 | } | |
4255 | ||
4256 | /* | |
4257 | * helper function around btrfs_log_inode to make sure newly created | |
4258 | * parent directories also end up in the log. A minimal inode and backref | |
4259 | * only logging is done of any parent directories that are older than | |
4260 | * the last committed transaction | |
4261 | */ | |
48a3b636 ES |
4262 | static int btrfs_log_inode_parent(struct btrfs_trans_handle *trans, |
4263 | struct btrfs_root *root, struct inode *inode, | |
49dae1bc FM |
4264 | struct dentry *parent, |
4265 | const loff_t start, | |
4266 | const loff_t end, | |
4267 | int exists_only, | |
8b050d35 | 4268 | struct btrfs_log_ctx *ctx) |
e02119d5 | 4269 | { |
12fcfd22 | 4270 | int inode_only = exists_only ? LOG_INODE_EXISTS : LOG_INODE_ALL; |
e02119d5 | 4271 | struct super_block *sb; |
6a912213 | 4272 | struct dentry *old_parent = NULL; |
12fcfd22 CM |
4273 | int ret = 0; |
4274 | u64 last_committed = root->fs_info->last_trans_committed; | |
4275 | ||
4276 | sb = inode->i_sb; | |
4277 | ||
3a5e1404 SW |
4278 | if (btrfs_test_opt(root, NOTREELOG)) { |
4279 | ret = 1; | |
4280 | goto end_no_trans; | |
4281 | } | |
4282 | ||
995946dd MX |
4283 | /* |
4284 | * The prev transaction commit doesn't complete, we need do | |
4285 | * full commit by ourselves. | |
4286 | */ | |
12fcfd22 CM |
4287 | if (root->fs_info->last_trans_log_full_commit > |
4288 | root->fs_info->last_trans_committed) { | |
4289 | ret = 1; | |
4290 | goto end_no_trans; | |
4291 | } | |
4292 | ||
76dda93c YZ |
4293 | if (root != BTRFS_I(inode)->root || |
4294 | btrfs_root_refs(&root->root_item) == 0) { | |
4295 | ret = 1; | |
4296 | goto end_no_trans; | |
4297 | } | |
4298 | ||
12fcfd22 CM |
4299 | ret = check_parent_dirs_for_sync(trans, inode, parent, |
4300 | sb, last_committed); | |
4301 | if (ret) | |
4302 | goto end_no_trans; | |
e02119d5 | 4303 | |
22ee6985 | 4304 | if (btrfs_inode_in_log(inode, trans->transid)) { |
257c62e1 CM |
4305 | ret = BTRFS_NO_LOG_SYNC; |
4306 | goto end_no_trans; | |
4307 | } | |
4308 | ||
8b050d35 | 4309 | ret = start_log_trans(trans, root, ctx); |
4a500fd1 | 4310 | if (ret) |
e87ac136 | 4311 | goto end_no_trans; |
e02119d5 | 4312 | |
8407f553 | 4313 | ret = btrfs_log_inode(trans, root, inode, inode_only, start, end, ctx); |
4a500fd1 YZ |
4314 | if (ret) |
4315 | goto end_trans; | |
12fcfd22 | 4316 | |
af4176b4 CM |
4317 | /* |
4318 | * for regular files, if its inode is already on disk, we don't | |
4319 | * have to worry about the parents at all. This is because | |
4320 | * we can use the last_unlink_trans field to record renames | |
4321 | * and other fun in this file. | |
4322 | */ | |
4323 | if (S_ISREG(inode->i_mode) && | |
4324 | BTRFS_I(inode)->generation <= last_committed && | |
4a500fd1 YZ |
4325 | BTRFS_I(inode)->last_unlink_trans <= last_committed) { |
4326 | ret = 0; | |
4327 | goto end_trans; | |
4328 | } | |
af4176b4 CM |
4329 | |
4330 | inode_only = LOG_INODE_EXISTS; | |
12fcfd22 CM |
4331 | while (1) { |
4332 | if (!parent || !parent->d_inode || sb != parent->d_inode->i_sb) | |
e02119d5 CM |
4333 | break; |
4334 | ||
12fcfd22 | 4335 | inode = parent->d_inode; |
76dda93c YZ |
4336 | if (root != BTRFS_I(inode)->root) |
4337 | break; | |
4338 | ||
12fcfd22 CM |
4339 | if (BTRFS_I(inode)->generation > |
4340 | root->fs_info->last_trans_committed) { | |
49dae1bc | 4341 | ret = btrfs_log_inode(trans, root, inode, inode_only, |
8407f553 | 4342 | 0, LLONG_MAX, ctx); |
4a500fd1 YZ |
4343 | if (ret) |
4344 | goto end_trans; | |
12fcfd22 | 4345 | } |
76dda93c | 4346 | if (IS_ROOT(parent)) |
e02119d5 | 4347 | break; |
12fcfd22 | 4348 | |
6a912213 JB |
4349 | parent = dget_parent(parent); |
4350 | dput(old_parent); | |
4351 | old_parent = parent; | |
e02119d5 | 4352 | } |
12fcfd22 | 4353 | ret = 0; |
4a500fd1 | 4354 | end_trans: |
6a912213 | 4355 | dput(old_parent); |
4a500fd1 | 4356 | if (ret < 0) { |
995946dd | 4357 | btrfs_set_log_full_commit(root->fs_info, trans); |
4a500fd1 YZ |
4358 | ret = 1; |
4359 | } | |
8b050d35 MX |
4360 | |
4361 | if (ret) | |
4362 | btrfs_remove_log_ctx(root, ctx); | |
12fcfd22 CM |
4363 | btrfs_end_log_trans(root); |
4364 | end_no_trans: | |
4365 | return ret; | |
e02119d5 CM |
4366 | } |
4367 | ||
4368 | /* | |
4369 | * it is not safe to log dentry if the chunk root has added new | |
4370 | * chunks. This returns 0 if the dentry was logged, and 1 otherwise. | |
4371 | * If this returns 1, you must commit the transaction to safely get your | |
4372 | * data on disk. | |
4373 | */ | |
4374 | int btrfs_log_dentry_safe(struct btrfs_trans_handle *trans, | |
8b050d35 | 4375 | struct btrfs_root *root, struct dentry *dentry, |
49dae1bc FM |
4376 | const loff_t start, |
4377 | const loff_t end, | |
8b050d35 | 4378 | struct btrfs_log_ctx *ctx) |
e02119d5 | 4379 | { |
6a912213 JB |
4380 | struct dentry *parent = dget_parent(dentry); |
4381 | int ret; | |
4382 | ||
8b050d35 | 4383 | ret = btrfs_log_inode_parent(trans, root, dentry->d_inode, parent, |
49dae1bc | 4384 | start, end, 0, ctx); |
6a912213 JB |
4385 | dput(parent); |
4386 | ||
4387 | return ret; | |
e02119d5 CM |
4388 | } |
4389 | ||
4390 | /* | |
4391 | * should be called during mount to recover any replay any log trees | |
4392 | * from the FS | |
4393 | */ | |
4394 | int btrfs_recover_log_trees(struct btrfs_root *log_root_tree) | |
4395 | { | |
4396 | int ret; | |
4397 | struct btrfs_path *path; | |
4398 | struct btrfs_trans_handle *trans; | |
4399 | struct btrfs_key key; | |
4400 | struct btrfs_key found_key; | |
4401 | struct btrfs_key tmp_key; | |
4402 | struct btrfs_root *log; | |
4403 | struct btrfs_fs_info *fs_info = log_root_tree->fs_info; | |
4404 | struct walk_control wc = { | |
4405 | .process_func = process_one_buffer, | |
4406 | .stage = 0, | |
4407 | }; | |
4408 | ||
e02119d5 | 4409 | path = btrfs_alloc_path(); |
db5b493a TI |
4410 | if (!path) |
4411 | return -ENOMEM; | |
4412 | ||
4413 | fs_info->log_root_recovering = 1; | |
e02119d5 | 4414 | |
4a500fd1 | 4415 | trans = btrfs_start_transaction(fs_info->tree_root, 0); |
79787eaa JM |
4416 | if (IS_ERR(trans)) { |
4417 | ret = PTR_ERR(trans); | |
4418 | goto error; | |
4419 | } | |
e02119d5 CM |
4420 | |
4421 | wc.trans = trans; | |
4422 | wc.pin = 1; | |
4423 | ||
db5b493a | 4424 | ret = walk_log_tree(trans, log_root_tree, &wc); |
79787eaa JM |
4425 | if (ret) { |
4426 | btrfs_error(fs_info, ret, "Failed to pin buffers while " | |
4427 | "recovering log root tree."); | |
4428 | goto error; | |
4429 | } | |
e02119d5 CM |
4430 | |
4431 | again: | |
4432 | key.objectid = BTRFS_TREE_LOG_OBJECTID; | |
4433 | key.offset = (u64)-1; | |
962a298f | 4434 | key.type = BTRFS_ROOT_ITEM_KEY; |
e02119d5 | 4435 | |
d397712b | 4436 | while (1) { |
e02119d5 | 4437 | ret = btrfs_search_slot(NULL, log_root_tree, &key, path, 0, 0); |
79787eaa JM |
4438 | |
4439 | if (ret < 0) { | |
4440 | btrfs_error(fs_info, ret, | |
4441 | "Couldn't find tree log root."); | |
4442 | goto error; | |
4443 | } | |
e02119d5 CM |
4444 | if (ret > 0) { |
4445 | if (path->slots[0] == 0) | |
4446 | break; | |
4447 | path->slots[0]--; | |
4448 | } | |
4449 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, | |
4450 | path->slots[0]); | |
b3b4aa74 | 4451 | btrfs_release_path(path); |
e02119d5 CM |
4452 | if (found_key.objectid != BTRFS_TREE_LOG_OBJECTID) |
4453 | break; | |
4454 | ||
cb517eab | 4455 | log = btrfs_read_fs_root(log_root_tree, &found_key); |
79787eaa JM |
4456 | if (IS_ERR(log)) { |
4457 | ret = PTR_ERR(log); | |
4458 | btrfs_error(fs_info, ret, | |
4459 | "Couldn't read tree log root."); | |
4460 | goto error; | |
4461 | } | |
e02119d5 CM |
4462 | |
4463 | tmp_key.objectid = found_key.offset; | |
4464 | tmp_key.type = BTRFS_ROOT_ITEM_KEY; | |
4465 | tmp_key.offset = (u64)-1; | |
4466 | ||
4467 | wc.replay_dest = btrfs_read_fs_root_no_name(fs_info, &tmp_key); | |
79787eaa JM |
4468 | if (IS_ERR(wc.replay_dest)) { |
4469 | ret = PTR_ERR(wc.replay_dest); | |
b50c6e25 JB |
4470 | free_extent_buffer(log->node); |
4471 | free_extent_buffer(log->commit_root); | |
4472 | kfree(log); | |
79787eaa JM |
4473 | btrfs_error(fs_info, ret, "Couldn't read target root " |
4474 | "for tree log recovery."); | |
4475 | goto error; | |
4476 | } | |
e02119d5 | 4477 | |
07d400a6 | 4478 | wc.replay_dest->log_root = log; |
5d4f98a2 | 4479 | btrfs_record_root_in_trans(trans, wc.replay_dest); |
e02119d5 | 4480 | ret = walk_log_tree(trans, log, &wc); |
e02119d5 | 4481 | |
b50c6e25 | 4482 | if (!ret && wc.stage == LOG_WALK_REPLAY_ALL) { |
e02119d5 CM |
4483 | ret = fixup_inode_link_counts(trans, wc.replay_dest, |
4484 | path); | |
e02119d5 CM |
4485 | } |
4486 | ||
4487 | key.offset = found_key.offset - 1; | |
07d400a6 | 4488 | wc.replay_dest->log_root = NULL; |
e02119d5 | 4489 | free_extent_buffer(log->node); |
b263c2c8 | 4490 | free_extent_buffer(log->commit_root); |
e02119d5 CM |
4491 | kfree(log); |
4492 | ||
b50c6e25 JB |
4493 | if (ret) |
4494 | goto error; | |
4495 | ||
e02119d5 CM |
4496 | if (found_key.offset == 0) |
4497 | break; | |
4498 | } | |
b3b4aa74 | 4499 | btrfs_release_path(path); |
e02119d5 CM |
4500 | |
4501 | /* step one is to pin it all, step two is to replay just inodes */ | |
4502 | if (wc.pin) { | |
4503 | wc.pin = 0; | |
4504 | wc.process_func = replay_one_buffer; | |
4505 | wc.stage = LOG_WALK_REPLAY_INODES; | |
4506 | goto again; | |
4507 | } | |
4508 | /* step three is to replay everything */ | |
4509 | if (wc.stage < LOG_WALK_REPLAY_ALL) { | |
4510 | wc.stage++; | |
4511 | goto again; | |
4512 | } | |
4513 | ||
4514 | btrfs_free_path(path); | |
4515 | ||
abefa55a JB |
4516 | /* step 4: commit the transaction, which also unpins the blocks */ |
4517 | ret = btrfs_commit_transaction(trans, fs_info->tree_root); | |
4518 | if (ret) | |
4519 | return ret; | |
4520 | ||
e02119d5 CM |
4521 | free_extent_buffer(log_root_tree->node); |
4522 | log_root_tree->log_root = NULL; | |
4523 | fs_info->log_root_recovering = 0; | |
e02119d5 | 4524 | kfree(log_root_tree); |
79787eaa | 4525 | |
abefa55a | 4526 | return 0; |
79787eaa | 4527 | error: |
b50c6e25 JB |
4528 | if (wc.trans) |
4529 | btrfs_end_transaction(wc.trans, fs_info->tree_root); | |
79787eaa JM |
4530 | btrfs_free_path(path); |
4531 | return ret; | |
e02119d5 | 4532 | } |
12fcfd22 CM |
4533 | |
4534 | /* | |
4535 | * there are some corner cases where we want to force a full | |
4536 | * commit instead of allowing a directory to be logged. | |
4537 | * | |
4538 | * They revolve around files there were unlinked from the directory, and | |
4539 | * this function updates the parent directory so that a full commit is | |
4540 | * properly done if it is fsync'd later after the unlinks are done. | |
4541 | */ | |
4542 | void btrfs_record_unlink_dir(struct btrfs_trans_handle *trans, | |
4543 | struct inode *dir, struct inode *inode, | |
4544 | int for_rename) | |
4545 | { | |
af4176b4 CM |
4546 | /* |
4547 | * when we're logging a file, if it hasn't been renamed | |
4548 | * or unlinked, and its inode is fully committed on disk, | |
4549 | * we don't have to worry about walking up the directory chain | |
4550 | * to log its parents. | |
4551 | * | |
4552 | * So, we use the last_unlink_trans field to put this transid | |
4553 | * into the file. When the file is logged we check it and | |
4554 | * don't log the parents if the file is fully on disk. | |
4555 | */ | |
4556 | if (S_ISREG(inode->i_mode)) | |
4557 | BTRFS_I(inode)->last_unlink_trans = trans->transid; | |
4558 | ||
12fcfd22 CM |
4559 | /* |
4560 | * if this directory was already logged any new | |
4561 | * names for this file/dir will get recorded | |
4562 | */ | |
4563 | smp_mb(); | |
4564 | if (BTRFS_I(dir)->logged_trans == trans->transid) | |
4565 | return; | |
4566 | ||
4567 | /* | |
4568 | * if the inode we're about to unlink was logged, | |
4569 | * the log will be properly updated for any new names | |
4570 | */ | |
4571 | if (BTRFS_I(inode)->logged_trans == trans->transid) | |
4572 | return; | |
4573 | ||
4574 | /* | |
4575 | * when renaming files across directories, if the directory | |
4576 | * there we're unlinking from gets fsync'd later on, there's | |
4577 | * no way to find the destination directory later and fsync it | |
4578 | * properly. So, we have to be conservative and force commits | |
4579 | * so the new name gets discovered. | |
4580 | */ | |
4581 | if (for_rename) | |
4582 | goto record; | |
4583 | ||
4584 | /* we can safely do the unlink without any special recording */ | |
4585 | return; | |
4586 | ||
4587 | record: | |
4588 | BTRFS_I(dir)->last_unlink_trans = trans->transid; | |
4589 | } | |
4590 | ||
4591 | /* | |
4592 | * Call this after adding a new name for a file and it will properly | |
4593 | * update the log to reflect the new name. | |
4594 | * | |
4595 | * It will return zero if all goes well, and it will return 1 if a | |
4596 | * full transaction commit is required. | |
4597 | */ | |
4598 | int btrfs_log_new_name(struct btrfs_trans_handle *trans, | |
4599 | struct inode *inode, struct inode *old_dir, | |
4600 | struct dentry *parent) | |
4601 | { | |
4602 | struct btrfs_root * root = BTRFS_I(inode)->root; | |
4603 | ||
af4176b4 CM |
4604 | /* |
4605 | * this will force the logging code to walk the dentry chain | |
4606 | * up for the file | |
4607 | */ | |
4608 | if (S_ISREG(inode->i_mode)) | |
4609 | BTRFS_I(inode)->last_unlink_trans = trans->transid; | |
4610 | ||
12fcfd22 CM |
4611 | /* |
4612 | * if this inode hasn't been logged and directory we're renaming it | |
4613 | * from hasn't been logged, we don't need to log it | |
4614 | */ | |
4615 | if (BTRFS_I(inode)->logged_trans <= | |
4616 | root->fs_info->last_trans_committed && | |
4617 | (!old_dir || BTRFS_I(old_dir)->logged_trans <= | |
4618 | root->fs_info->last_trans_committed)) | |
4619 | return 0; | |
4620 | ||
49dae1bc FM |
4621 | return btrfs_log_inode_parent(trans, root, inode, parent, 0, |
4622 | LLONG_MAX, 1, NULL); | |
12fcfd22 CM |
4623 | } |
4624 |