blob: c13922206d1be59196a8c59c95f9a98bae12be74 [file] [log] [blame]
Chris Masone02119d2008-09-05 16:13:11 -04001/*
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>
20#include "ctree.h"
21#include "transaction.h"
22#include "disk-io.h"
23#include "locking.h"
24#include "print-tree.h"
25#include "compat.h"
Christoph Hellwigb2950862008-12-02 09:54:17 -050026#include "tree-log.h"
Chris Masone02119d2008-09-05 16:13:11 -040027
28/* magic values for the inode_only field in btrfs_log_inode:
29 *
30 * LOG_INODE_ALL means to log everything
31 * LOG_INODE_EXISTS means to log just enough to recreate the inode
32 * during log replay
33 */
34#define LOG_INODE_ALL 0
35#define LOG_INODE_EXISTS 1
36
37/*
Chris Mason12fcfd22009-03-24 10:24:20 -040038 * directory trouble cases
39 *
40 * 1) on rename or unlink, if the inode being unlinked isn't in the fsync
41 * log, we must force a full commit before doing an fsync of the directory
42 * where the unlink was done.
43 * ---> record transid of last unlink/rename per directory
44 *
45 * mkdir foo/some_dir
46 * normal commit
47 * rename foo/some_dir foo2/some_dir
48 * mkdir foo/some_dir
49 * fsync foo/some_dir/some_file
50 *
51 * The fsync above will unlink the original some_dir without recording
52 * it in its new location (foo2). After a crash, some_dir will be gone
53 * unless the fsync of some_file forces a full commit
54 *
55 * 2) we must log any new names for any file or dir that is in the fsync
56 * log. ---> check inode while renaming/linking.
57 *
58 * 2a) we must log any new names for any file or dir during rename
59 * when the directory they are being removed from was logged.
60 * ---> check inode and old parent dir during rename
61 *
62 * 2a is actually the more important variant. With the extra logging
63 * a crash might unlink the old name without recreating the new one
64 *
65 * 3) after a crash, we must go through any directories with a link count
66 * of zero and redo the rm -rf
67 *
68 * mkdir f1/foo
69 * normal commit
70 * rm -rf f1/foo
71 * fsync(f1)
72 *
73 * The directory f1 was fully removed from the FS, but fsync was never
74 * called on f1, only its parent dir. After a crash the rm -rf must
75 * be replayed. This must be able to recurse down the entire
76 * directory tree. The inode link count fixup code takes care of the
77 * ugly details.
78 */
79
80/*
Chris Masone02119d2008-09-05 16:13:11 -040081 * stages for the tree walking. The first
82 * stage (0) is to only pin down the blocks we find
83 * the second stage (1) is to make sure that all the inodes
84 * we find in the log are created in the subvolume.
85 *
86 * The last stage is to deal with directories and links and extents
87 * and all the other fun semantics
88 */
89#define LOG_WALK_PIN_ONLY 0
90#define LOG_WALK_REPLAY_INODES 1
91#define LOG_WALK_REPLAY_ALL 2
92
Chris Mason12fcfd22009-03-24 10:24:20 -040093static int btrfs_log_inode(struct btrfs_trans_handle *trans,
Chris Masone02119d2008-09-05 16:13:11 -040094 struct btrfs_root *root, struct inode *inode,
95 int inode_only);
Yan Zhengec051c02009-01-05 15:43:42 -050096static int link_to_fixup_dir(struct btrfs_trans_handle *trans,
97 struct btrfs_root *root,
98 struct btrfs_path *path, u64 objectid);
Chris Mason12fcfd22009-03-24 10:24:20 -040099static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans,
100 struct btrfs_root *root,
101 struct btrfs_root *log,
102 struct btrfs_path *path,
103 u64 dirid, int del_all);
Chris Masone02119d2008-09-05 16:13:11 -0400104
105/*
106 * tree logging is a special write ahead log used to make sure that
107 * fsyncs and O_SYNCs can happen without doing full tree commits.
108 *
109 * Full tree commits are expensive because they require commonly
110 * modified blocks to be recowed, creating many dirty pages in the
111 * extent tree an 4x-6x higher write load than ext3.
112 *
113 * Instead of doing a tree commit on every fsync, we use the
114 * key ranges and transaction ids to find items for a given file or directory
115 * that have changed in this transaction. Those items are copied into
116 * a special tree (one per subvolume root), that tree is written to disk
117 * and then the fsync is considered complete.
118 *
119 * After a crash, items are copied out of the log-tree back into the
120 * subvolume tree. Any file data extents found are recorded in the extent
121 * allocation tree, and the log-tree freed.
122 *
123 * The log tree is read three times, once to pin down all the extents it is
124 * using in ram and once, once to create all the inodes logged in the tree
125 * and once to do all the other items.
126 */
127
128/*
Chris Masone02119d2008-09-05 16:13:11 -0400129 * start a sub transaction and setup the log tree
130 * this increments the log tree writer count to make the people
131 * syncing the tree wait for us to finish
132 */
133static int start_log_trans(struct btrfs_trans_handle *trans,
134 struct btrfs_root *root)
135{
136 int ret;
Yan Zheng7237f182009-01-21 12:54:03 -0500137
138 mutex_lock(&root->log_mutex);
139 if (root->log_root) {
140 root->log_batch++;
141 atomic_inc(&root->log_writers);
142 mutex_unlock(&root->log_mutex);
143 return 0;
144 }
Chris Masone02119d2008-09-05 16:13:11 -0400145 mutex_lock(&root->fs_info->tree_log_mutex);
146 if (!root->fs_info->log_root_tree) {
147 ret = btrfs_init_log_root_tree(trans, root->fs_info);
148 BUG_ON(ret);
149 }
150 if (!root->log_root) {
151 ret = btrfs_add_log_tree(trans, root);
152 BUG_ON(ret);
153 }
Chris Masone02119d2008-09-05 16:13:11 -0400154 mutex_unlock(&root->fs_info->tree_log_mutex);
Yan Zheng7237f182009-01-21 12:54:03 -0500155 root->log_batch++;
156 atomic_inc(&root->log_writers);
157 mutex_unlock(&root->log_mutex);
Chris Masone02119d2008-09-05 16:13:11 -0400158 return 0;
159}
160
161/*
162 * returns 0 if there was a log transaction running and we were able
163 * to join, or returns -ENOENT if there were not transactions
164 * in progress
165 */
166static int join_running_log_trans(struct btrfs_root *root)
167{
168 int ret = -ENOENT;
169
170 smp_mb();
171 if (!root->log_root)
172 return -ENOENT;
173
Yan Zheng7237f182009-01-21 12:54:03 -0500174 mutex_lock(&root->log_mutex);
Chris Masone02119d2008-09-05 16:13:11 -0400175 if (root->log_root) {
176 ret = 0;
Yan Zheng7237f182009-01-21 12:54:03 -0500177 atomic_inc(&root->log_writers);
Chris Masone02119d2008-09-05 16:13:11 -0400178 }
Yan Zheng7237f182009-01-21 12:54:03 -0500179 mutex_unlock(&root->log_mutex);
Chris Masone02119d2008-09-05 16:13:11 -0400180 return ret;
181}
182
183/*
Chris Mason12fcfd22009-03-24 10:24:20 -0400184 * This either makes the current running log transaction wait
185 * until you call btrfs_end_log_trans() or it makes any future
186 * log transactions wait until you call btrfs_end_log_trans()
187 */
188int btrfs_pin_log_trans(struct btrfs_root *root)
189{
190 int ret = -ENOENT;
191
192 mutex_lock(&root->log_mutex);
193 atomic_inc(&root->log_writers);
194 mutex_unlock(&root->log_mutex);
195 return ret;
196}
197
198/*
Chris Masone02119d2008-09-05 16:13:11 -0400199 * indicate we're done making changes to the log tree
200 * and wake up anyone waiting to do a sync
201 */
Chris Mason12fcfd22009-03-24 10:24:20 -0400202int btrfs_end_log_trans(struct btrfs_root *root)
Chris Masone02119d2008-09-05 16:13:11 -0400203{
Yan Zheng7237f182009-01-21 12:54:03 -0500204 if (atomic_dec_and_test(&root->log_writers)) {
205 smp_mb();
206 if (waitqueue_active(&root->log_writer_wait))
207 wake_up(&root->log_writer_wait);
208 }
Chris Masone02119d2008-09-05 16:13:11 -0400209 return 0;
210}
211
212
213/*
214 * the walk control struct is used to pass state down the chain when
215 * processing the log tree. The stage field tells us which part
216 * of the log tree processing we are currently doing. The others
217 * are state fields used for that specific part
218 */
219struct walk_control {
220 /* should we free the extent on disk when done? This is used
221 * at transaction commit time while freeing a log tree
222 */
223 int free;
224
225 /* should we write out the extent buffer? This is used
226 * while flushing the log tree to disk during a sync
227 */
228 int write;
229
230 /* should we wait for the extent buffer io to finish? Also used
231 * while flushing the log tree to disk for a sync
232 */
233 int wait;
234
235 /* pin only walk, we record which extents on disk belong to the
236 * log trees
237 */
238 int pin;
239
240 /* what stage of the replay code we're currently in */
241 int stage;
242
243 /* the root we are currently replaying */
244 struct btrfs_root *replay_dest;
245
246 /* the trans handle for the current replay */
247 struct btrfs_trans_handle *trans;
248
249 /* the function that gets used to process blocks we find in the
250 * tree. Note the extent_buffer might not be up to date when it is
251 * passed in, and it must be checked or read if you need the data
252 * inside it
253 */
254 int (*process_func)(struct btrfs_root *log, struct extent_buffer *eb,
255 struct walk_control *wc, u64 gen);
256};
257
258/*
259 * process_func used to pin down extents, write them or wait on them
260 */
261static int process_one_buffer(struct btrfs_root *log,
262 struct extent_buffer *eb,
263 struct walk_control *wc, u64 gen)
264{
Josef Bacik04018de2009-04-03 10:14:18 -0400265 if (wc->pin)
Chris Masone02119d2008-09-05 16:13:11 -0400266 btrfs_update_pinned_extents(log->fs_info->extent_root,
267 eb->start, eb->len, 1);
Chris Masone02119d2008-09-05 16:13:11 -0400268
269 if (btrfs_buffer_uptodate(eb, gen)) {
270 if (wc->write)
271 btrfs_write_tree_block(eb);
272 if (wc->wait)
273 btrfs_wait_tree_block_writeback(eb);
274 }
275 return 0;
276}
277
278/*
279 * Item overwrite used by replay and tree logging. eb, slot and key all refer
280 * to the src data we are copying out.
281 *
282 * root is the tree we are copying into, and path is a scratch
283 * path for use in this function (it should be released on entry and
284 * will be released on exit).
285 *
286 * If the key is already in the destination tree the existing item is
287 * overwritten. If the existing item isn't big enough, it is extended.
288 * If it is too large, it is truncated.
289 *
290 * If the key isn't in the destination yet, a new item is inserted.
291 */
292static noinline int overwrite_item(struct btrfs_trans_handle *trans,
293 struct btrfs_root *root,
294 struct btrfs_path *path,
295 struct extent_buffer *eb, int slot,
296 struct btrfs_key *key)
297{
298 int ret;
299 u32 item_size;
300 u64 saved_i_size = 0;
301 int save_old_i_size = 0;
302 unsigned long src_ptr;
303 unsigned long dst_ptr;
304 int overwrite_root = 0;
305
306 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
307 overwrite_root = 1;
308
309 item_size = btrfs_item_size_nr(eb, slot);
310 src_ptr = btrfs_item_ptr_offset(eb, slot);
311
312 /* look for the key in the destination tree */
313 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
314 if (ret == 0) {
315 char *src_copy;
316 char *dst_copy;
317 u32 dst_size = btrfs_item_size_nr(path->nodes[0],
318 path->slots[0]);
319 if (dst_size != item_size)
320 goto insert;
321
322 if (item_size == 0) {
323 btrfs_release_path(root, path);
324 return 0;
325 }
326 dst_copy = kmalloc(item_size, GFP_NOFS);
327 src_copy = kmalloc(item_size, GFP_NOFS);
328
329 read_extent_buffer(eb, src_copy, src_ptr, item_size);
330
331 dst_ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
332 read_extent_buffer(path->nodes[0], dst_copy, dst_ptr,
333 item_size);
334 ret = memcmp(dst_copy, src_copy, item_size);
335
336 kfree(dst_copy);
337 kfree(src_copy);
338 /*
339 * they have the same contents, just return, this saves
340 * us from cowing blocks in the destination tree and doing
341 * extra writes that may not have been done by a previous
342 * sync
343 */
344 if (ret == 0) {
345 btrfs_release_path(root, path);
346 return 0;
347 }
348
349 }
350insert:
351 btrfs_release_path(root, path);
352 /* try to insert the key into the destination tree */
353 ret = btrfs_insert_empty_item(trans, root, path,
354 key, item_size);
355
356 /* make sure any existing item is the correct size */
357 if (ret == -EEXIST) {
358 u32 found_size;
359 found_size = btrfs_item_size_nr(path->nodes[0],
360 path->slots[0]);
361 if (found_size > item_size) {
362 btrfs_truncate_item(trans, root, path, item_size, 1);
363 } else if (found_size < item_size) {
Yan Zheng87b29b22008-12-17 10:21:48 -0500364 ret = btrfs_extend_item(trans, root, path,
365 item_size - found_size);
Chris Masone02119d2008-09-05 16:13:11 -0400366 BUG_ON(ret);
367 }
368 } else if (ret) {
369 BUG();
370 }
371 dst_ptr = btrfs_item_ptr_offset(path->nodes[0],
372 path->slots[0]);
373
374 /* don't overwrite an existing inode if the generation number
375 * was logged as zero. This is done when the tree logging code
376 * is just logging an inode to make sure it exists after recovery.
377 *
378 * Also, don't overwrite i_size on directories during replay.
379 * log replay inserts and removes directory items based on the
380 * state of the tree found in the subvolume, and i_size is modified
381 * as it goes
382 */
383 if (key->type == BTRFS_INODE_ITEM_KEY && ret == -EEXIST) {
384 struct btrfs_inode_item *src_item;
385 struct btrfs_inode_item *dst_item;
386
387 src_item = (struct btrfs_inode_item *)src_ptr;
388 dst_item = (struct btrfs_inode_item *)dst_ptr;
389
390 if (btrfs_inode_generation(eb, src_item) == 0)
391 goto no_copy;
392
393 if (overwrite_root &&
394 S_ISDIR(btrfs_inode_mode(eb, src_item)) &&
395 S_ISDIR(btrfs_inode_mode(path->nodes[0], dst_item))) {
396 save_old_i_size = 1;
397 saved_i_size = btrfs_inode_size(path->nodes[0],
398 dst_item);
399 }
400 }
401
402 copy_extent_buffer(path->nodes[0], eb, dst_ptr,
403 src_ptr, item_size);
404
405 if (save_old_i_size) {
406 struct btrfs_inode_item *dst_item;
407 dst_item = (struct btrfs_inode_item *)dst_ptr;
408 btrfs_set_inode_size(path->nodes[0], dst_item, saved_i_size);
409 }
410
411 /* make sure the generation is filled in */
412 if (key->type == BTRFS_INODE_ITEM_KEY) {
413 struct btrfs_inode_item *dst_item;
414 dst_item = (struct btrfs_inode_item *)dst_ptr;
415 if (btrfs_inode_generation(path->nodes[0], dst_item) == 0) {
416 btrfs_set_inode_generation(path->nodes[0], dst_item,
417 trans->transid);
418 }
419 }
420no_copy:
421 btrfs_mark_buffer_dirty(path->nodes[0]);
422 btrfs_release_path(root, path);
423 return 0;
424}
425
426/*
427 * simple helper to read an inode off the disk from a given root
428 * This can only be called for subvolume roots and not for the log
429 */
430static noinline struct inode *read_one_inode(struct btrfs_root *root,
431 u64 objectid)
432{
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400433 struct btrfs_key key;
Chris Masone02119d2008-09-05 16:13:11 -0400434 struct inode *inode;
Chris Masone02119d2008-09-05 16:13:11 -0400435
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400436 key.objectid = objectid;
437 key.type = BTRFS_INODE_ITEM_KEY;
438 key.offset = 0;
439 inode = btrfs_iget(root->fs_info->sb, &key, root);
440 if (IS_ERR(inode)) {
441 inode = NULL;
442 } else if (is_bad_inode(inode)) {
Chris Masone02119d2008-09-05 16:13:11 -0400443 iput(inode);
444 inode = NULL;
445 }
446 return inode;
447}
448
449/* replays a single extent in 'eb' at 'slot' with 'key' into the
450 * subvolume 'root'. path is released on entry and should be released
451 * on exit.
452 *
453 * extents in the log tree have not been allocated out of the extent
454 * tree yet. So, this completes the allocation, taking a reference
455 * as required if the extent already exists or creating a new extent
456 * if it isn't in the extent allocation tree yet.
457 *
458 * The extent is inserted into the file, dropping any existing extents
459 * from the file that overlap the new one.
460 */
461static noinline int replay_one_extent(struct btrfs_trans_handle *trans,
462 struct btrfs_root *root,
463 struct btrfs_path *path,
464 struct extent_buffer *eb, int slot,
465 struct btrfs_key *key)
466{
467 int found_type;
468 u64 mask = root->sectorsize - 1;
469 u64 extent_end;
470 u64 alloc_hint;
471 u64 start = key->offset;
Yan Zheng07d400a2009-01-06 11:42:00 -0500472 u64 saved_nbytes;
Chris Masone02119d2008-09-05 16:13:11 -0400473 struct btrfs_file_extent_item *item;
474 struct inode *inode = NULL;
475 unsigned long size;
476 int ret = 0;
477
478 item = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
479 found_type = btrfs_file_extent_type(eb, item);
480
Yan Zhengd899e052008-10-30 14:25:28 -0400481 if (found_type == BTRFS_FILE_EXTENT_REG ||
482 found_type == BTRFS_FILE_EXTENT_PREALLOC)
Chris Masone02119d2008-09-05 16:13:11 -0400483 extent_end = start + btrfs_file_extent_num_bytes(eb, item);
484 else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
Chris Masonc8b97812008-10-29 14:49:59 -0400485 size = btrfs_file_extent_inline_len(eb, item);
Chris Masone02119d2008-09-05 16:13:11 -0400486 extent_end = (start + size + mask) & ~mask;
487 } else {
488 ret = 0;
489 goto out;
490 }
491
492 inode = read_one_inode(root, key->objectid);
493 if (!inode) {
494 ret = -EIO;
495 goto out;
496 }
497
498 /*
499 * first check to see if we already have this extent in the
500 * file. This must be done before the btrfs_drop_extents run
501 * so we don't try to drop this extent.
502 */
503 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
504 start, 0);
505
Yan Zhengd899e052008-10-30 14:25:28 -0400506 if (ret == 0 &&
507 (found_type == BTRFS_FILE_EXTENT_REG ||
508 found_type == BTRFS_FILE_EXTENT_PREALLOC)) {
Chris Masone02119d2008-09-05 16:13:11 -0400509 struct btrfs_file_extent_item cmp1;
510 struct btrfs_file_extent_item cmp2;
511 struct btrfs_file_extent_item *existing;
512 struct extent_buffer *leaf;
513
514 leaf = path->nodes[0];
515 existing = btrfs_item_ptr(leaf, path->slots[0],
516 struct btrfs_file_extent_item);
517
518 read_extent_buffer(eb, &cmp1, (unsigned long)item,
519 sizeof(cmp1));
520 read_extent_buffer(leaf, &cmp2, (unsigned long)existing,
521 sizeof(cmp2));
522
523 /*
524 * we already have a pointer to this exact extent,
525 * we don't have to do anything
526 */
527 if (memcmp(&cmp1, &cmp2, sizeof(cmp1)) == 0) {
528 btrfs_release_path(root, path);
529 goto out;
530 }
531 }
532 btrfs_release_path(root, path);
533
Yan Zheng07d400a2009-01-06 11:42:00 -0500534 saved_nbytes = inode_get_bytes(inode);
Chris Masone02119d2008-09-05 16:13:11 -0400535 /* drop any overlapping extents */
536 ret = btrfs_drop_extents(trans, root, inode,
Chris Masone980b502009-04-24 14:39:24 -0400537 start, extent_end, extent_end, start, &alloc_hint);
Chris Masone02119d2008-09-05 16:13:11 -0400538 BUG_ON(ret);
539
Yan Zheng07d400a2009-01-06 11:42:00 -0500540 if (found_type == BTRFS_FILE_EXTENT_REG ||
541 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400542 u64 offset;
Yan Zheng07d400a2009-01-06 11:42:00 -0500543 unsigned long dest_offset;
544 struct btrfs_key ins;
Chris Masone02119d2008-09-05 16:13:11 -0400545
Yan Zheng07d400a2009-01-06 11:42:00 -0500546 ret = btrfs_insert_empty_item(trans, root, path, key,
547 sizeof(*item));
548 BUG_ON(ret);
549 dest_offset = btrfs_item_ptr_offset(path->nodes[0],
550 path->slots[0]);
551 copy_extent_buffer(path->nodes[0], eb, dest_offset,
552 (unsigned long)item, sizeof(*item));
553
554 ins.objectid = btrfs_file_extent_disk_bytenr(eb, item);
555 ins.offset = btrfs_file_extent_disk_num_bytes(eb, item);
556 ins.type = BTRFS_EXTENT_ITEM_KEY;
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400557 offset = key->offset - btrfs_file_extent_offset(eb, item);
Yan Zheng07d400a2009-01-06 11:42:00 -0500558
559 if (ins.objectid > 0) {
560 u64 csum_start;
561 u64 csum_end;
562 LIST_HEAD(ordered_sums);
563 /*
564 * is this extent already allocated in the extent
565 * allocation tree? If so, just add a reference
566 */
567 ret = btrfs_lookup_extent(root, ins.objectid,
568 ins.offset);
569 if (ret == 0) {
570 ret = btrfs_inc_extent_ref(trans, root,
571 ins.objectid, ins.offset,
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400572 0, root->root_key.objectid,
573 key->objectid, offset);
Yan Zheng07d400a2009-01-06 11:42:00 -0500574 } else {
575 /*
576 * insert the extent pointer in the extent
577 * allocation tree
578 */
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400579 ret = btrfs_alloc_logged_file_extent(trans,
580 root, root->root_key.objectid,
581 key->objectid, offset, &ins);
Yan Zheng07d400a2009-01-06 11:42:00 -0500582 BUG_ON(ret);
583 }
584 btrfs_release_path(root, path);
585
586 if (btrfs_file_extent_compression(eb, item)) {
587 csum_start = ins.objectid;
588 csum_end = csum_start + ins.offset;
589 } else {
590 csum_start = ins.objectid +
591 btrfs_file_extent_offset(eb, item);
592 csum_end = csum_start +
593 btrfs_file_extent_num_bytes(eb, item);
594 }
595
596 ret = btrfs_lookup_csums_range(root->log_root,
597 csum_start, csum_end - 1,
598 &ordered_sums);
599 BUG_ON(ret);
600 while (!list_empty(&ordered_sums)) {
601 struct btrfs_ordered_sum *sums;
602 sums = list_entry(ordered_sums.next,
603 struct btrfs_ordered_sum,
604 list);
605 ret = btrfs_csum_file_blocks(trans,
606 root->fs_info->csum_root,
607 sums);
608 BUG_ON(ret);
609 list_del(&sums->list);
610 kfree(sums);
611 }
612 } else {
613 btrfs_release_path(root, path);
614 }
615 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
616 /* inline extents are easy, we just overwrite them */
617 ret = overwrite_item(trans, root, path, eb, slot, key);
618 BUG_ON(ret);
619 }
620
621 inode_set_bytes(inode, saved_nbytes);
Chris Masone02119d2008-09-05 16:13:11 -0400622 btrfs_update_inode(trans, root, inode);
623out:
624 if (inode)
625 iput(inode);
626 return ret;
627}
628
629/*
630 * when cleaning up conflicts between the directory names in the
631 * subvolume, directory names in the log and directory names in the
632 * inode back references, we may have to unlink inodes from directories.
633 *
634 * This is a helper function to do the unlink of a specific directory
635 * item
636 */
637static noinline int drop_one_dir_item(struct btrfs_trans_handle *trans,
638 struct btrfs_root *root,
639 struct btrfs_path *path,
640 struct inode *dir,
641 struct btrfs_dir_item *di)
642{
643 struct inode *inode;
644 char *name;
645 int name_len;
646 struct extent_buffer *leaf;
647 struct btrfs_key location;
648 int ret;
649
650 leaf = path->nodes[0];
651
652 btrfs_dir_item_key_to_cpu(leaf, di, &location);
653 name_len = btrfs_dir_name_len(leaf, di);
654 name = kmalloc(name_len, GFP_NOFS);
655 read_extent_buffer(leaf, name, (unsigned long)(di + 1), name_len);
656 btrfs_release_path(root, path);
657
658 inode = read_one_inode(root, location.objectid);
659 BUG_ON(!inode);
660
Yan Zhengec051c02009-01-05 15:43:42 -0500661 ret = link_to_fixup_dir(trans, root, path, location.objectid);
662 BUG_ON(ret);
Chris Mason12fcfd22009-03-24 10:24:20 -0400663
Chris Masone02119d2008-09-05 16:13:11 -0400664 ret = btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
Yan Zhengec051c02009-01-05 15:43:42 -0500665 BUG_ON(ret);
Chris Masone02119d2008-09-05 16:13:11 -0400666 kfree(name);
667
668 iput(inode);
669 return ret;
670}
671
672/*
673 * helper function to see if a given name and sequence number found
674 * in an inode back reference are already in a directory and correctly
675 * point to this inode
676 */
677static noinline int inode_in_dir(struct btrfs_root *root,
678 struct btrfs_path *path,
679 u64 dirid, u64 objectid, u64 index,
680 const char *name, int name_len)
681{
682 struct btrfs_dir_item *di;
683 struct btrfs_key location;
684 int match = 0;
685
686 di = btrfs_lookup_dir_index_item(NULL, root, path, dirid,
687 index, name, name_len, 0);
688 if (di && !IS_ERR(di)) {
689 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
690 if (location.objectid != objectid)
691 goto out;
692 } else
693 goto out;
694 btrfs_release_path(root, path);
695
696 di = btrfs_lookup_dir_item(NULL, root, path, dirid, name, name_len, 0);
697 if (di && !IS_ERR(di)) {
698 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
699 if (location.objectid != objectid)
700 goto out;
701 } else
702 goto out;
703 match = 1;
704out:
705 btrfs_release_path(root, path);
706 return match;
707}
708
709/*
710 * helper function to check a log tree for a named back reference in
711 * an inode. This is used to decide if a back reference that is
712 * found in the subvolume conflicts with what we find in the log.
713 *
714 * inode backreferences may have multiple refs in a single item,
715 * during replay we process one reference at a time, and we don't
716 * want to delete valid links to a file from the subvolume if that
717 * link is also in the log.
718 */
719static noinline int backref_in_log(struct btrfs_root *log,
720 struct btrfs_key *key,
721 char *name, int namelen)
722{
723 struct btrfs_path *path;
724 struct btrfs_inode_ref *ref;
725 unsigned long ptr;
726 unsigned long ptr_end;
727 unsigned long name_ptr;
728 int found_name_len;
729 int item_size;
730 int ret;
731 int match = 0;
732
733 path = btrfs_alloc_path();
734 ret = btrfs_search_slot(NULL, log, key, path, 0, 0);
735 if (ret != 0)
736 goto out;
737
738 item_size = btrfs_item_size_nr(path->nodes[0], path->slots[0]);
739 ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
740 ptr_end = ptr + item_size;
741 while (ptr < ptr_end) {
742 ref = (struct btrfs_inode_ref *)ptr;
743 found_name_len = btrfs_inode_ref_name_len(path->nodes[0], ref);
744 if (found_name_len == namelen) {
745 name_ptr = (unsigned long)(ref + 1);
746 ret = memcmp_extent_buffer(path->nodes[0], name,
747 name_ptr, namelen);
748 if (ret == 0) {
749 match = 1;
750 goto out;
751 }
752 }
753 ptr = (unsigned long)(ref + 1) + found_name_len;
754 }
755out:
756 btrfs_free_path(path);
757 return match;
758}
759
760
761/*
762 * replay one inode back reference item found in the log tree.
763 * eb, slot and key refer to the buffer and key found in the log tree.
764 * root is the destination we are replaying into, and path is for temp
765 * use by this function. (it should be released on return).
766 */
767static noinline int add_inode_ref(struct btrfs_trans_handle *trans,
768 struct btrfs_root *root,
769 struct btrfs_root *log,
770 struct btrfs_path *path,
771 struct extent_buffer *eb, int slot,
772 struct btrfs_key *key)
773{
774 struct inode *dir;
775 int ret;
776 struct btrfs_key location;
777 struct btrfs_inode_ref *ref;
778 struct btrfs_dir_item *di;
779 struct inode *inode;
780 char *name;
781 int namelen;
782 unsigned long ref_ptr;
783 unsigned long ref_end;
784
785 location.objectid = key->objectid;
786 location.type = BTRFS_INODE_ITEM_KEY;
787 location.offset = 0;
788
789 /*
790 * it is possible that we didn't log all the parent directories
791 * for a given inode. If we don't find the dir, just don't
792 * copy the back ref in. The link count fixup code will take
793 * care of the rest
794 */
795 dir = read_one_inode(root, key->offset);
796 if (!dir)
797 return -ENOENT;
798
799 inode = read_one_inode(root, key->objectid);
800 BUG_ON(!dir);
801
802 ref_ptr = btrfs_item_ptr_offset(eb, slot);
803 ref_end = ref_ptr + btrfs_item_size_nr(eb, slot);
804
805again:
806 ref = (struct btrfs_inode_ref *)ref_ptr;
807
808 namelen = btrfs_inode_ref_name_len(eb, ref);
809 name = kmalloc(namelen, GFP_NOFS);
810 BUG_ON(!name);
811
812 read_extent_buffer(eb, name, (unsigned long)(ref + 1), namelen);
813
814 /* if we already have a perfect match, we're done */
815 if (inode_in_dir(root, path, dir->i_ino, inode->i_ino,
816 btrfs_inode_ref_index(eb, ref),
817 name, namelen)) {
818 goto out;
819 }
820
821 /*
822 * look for a conflicting back reference in the metadata.
823 * if we find one we have to unlink that name of the file
824 * before we add our new link. Later on, we overwrite any
825 * existing back reference, and we don't want to create
826 * dangling pointers in the directory.
827 */
828conflict_again:
829 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
830 if (ret == 0) {
831 char *victim_name;
832 int victim_name_len;
833 struct btrfs_inode_ref *victim_ref;
834 unsigned long ptr;
835 unsigned long ptr_end;
836 struct extent_buffer *leaf = path->nodes[0];
837
838 /* are we trying to overwrite a back ref for the root directory
839 * if so, just jump out, we're done
840 */
841 if (key->objectid == key->offset)
842 goto out_nowrite;
843
844 /* check all the names in this back reference to see
845 * if they are in the log. if so, we allow them to stay
846 * otherwise they must be unlinked as a conflict
847 */
848 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
849 ptr_end = ptr + btrfs_item_size_nr(leaf, path->slots[0]);
Chris Masond3977122009-01-05 21:25:51 -0500850 while (ptr < ptr_end) {
Chris Masone02119d2008-09-05 16:13:11 -0400851 victim_ref = (struct btrfs_inode_ref *)ptr;
852 victim_name_len = btrfs_inode_ref_name_len(leaf,
853 victim_ref);
854 victim_name = kmalloc(victim_name_len, GFP_NOFS);
855 BUG_ON(!victim_name);
856
857 read_extent_buffer(leaf, victim_name,
858 (unsigned long)(victim_ref + 1),
859 victim_name_len);
860
861 if (!backref_in_log(log, key, victim_name,
862 victim_name_len)) {
863 btrfs_inc_nlink(inode);
864 btrfs_release_path(root, path);
Chris Mason12fcfd22009-03-24 10:24:20 -0400865
Chris Masone02119d2008-09-05 16:13:11 -0400866 ret = btrfs_unlink_inode(trans, root, dir,
867 inode, victim_name,
868 victim_name_len);
869 kfree(victim_name);
870 btrfs_release_path(root, path);
871 goto conflict_again;
872 }
873 kfree(victim_name);
874 ptr = (unsigned long)(victim_ref + 1) + victim_name_len;
875 }
876 BUG_ON(ret);
877 }
878 btrfs_release_path(root, path);
879
880 /* look for a conflicting sequence number */
881 di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
882 btrfs_inode_ref_index(eb, ref),
883 name, namelen, 0);
884 if (di && !IS_ERR(di)) {
885 ret = drop_one_dir_item(trans, root, path, dir, di);
886 BUG_ON(ret);
887 }
888 btrfs_release_path(root, path);
889
890
891 /* look for a conflicting name */
892 di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
893 name, namelen, 0);
894 if (di && !IS_ERR(di)) {
895 ret = drop_one_dir_item(trans, root, path, dir, di);
896 BUG_ON(ret);
897 }
898 btrfs_release_path(root, path);
899
900 /* insert our name */
901 ret = btrfs_add_link(trans, dir, inode, name, namelen, 0,
902 btrfs_inode_ref_index(eb, ref));
903 BUG_ON(ret);
904
905 btrfs_update_inode(trans, root, inode);
906
907out:
908 ref_ptr = (unsigned long)(ref + 1) + namelen;
909 kfree(name);
910 if (ref_ptr < ref_end)
911 goto again;
912
913 /* finally write the back reference in the inode */
914 ret = overwrite_item(trans, root, path, eb, slot, key);
915 BUG_ON(ret);
916
917out_nowrite:
918 btrfs_release_path(root, path);
919 iput(dir);
920 iput(inode);
921 return 0;
922}
923
924/*
Chris Masone02119d2008-09-05 16:13:11 -0400925 * There are a few corners where the link count of the file can't
926 * be properly maintained during replay. So, instead of adding
927 * lots of complexity to the log code, we just scan the backrefs
928 * for any file that has been through replay.
929 *
930 * The scan will update the link count on the inode to reflect the
931 * number of back refs found. If it goes down to zero, the iput
932 * will free the inode.
933 */
934static noinline int fixup_inode_link_count(struct btrfs_trans_handle *trans,
935 struct btrfs_root *root,
936 struct inode *inode)
937{
938 struct btrfs_path *path;
939 int ret;
940 struct btrfs_key key;
941 u64 nlink = 0;
942 unsigned long ptr;
943 unsigned long ptr_end;
944 int name_len;
945
946 key.objectid = inode->i_ino;
947 key.type = BTRFS_INODE_REF_KEY;
948 key.offset = (u64)-1;
949
950 path = btrfs_alloc_path();
951
Chris Masond3977122009-01-05 21:25:51 -0500952 while (1) {
Chris Masone02119d2008-09-05 16:13:11 -0400953 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
954 if (ret < 0)
955 break;
956 if (ret > 0) {
957 if (path->slots[0] == 0)
958 break;
959 path->slots[0]--;
960 }
961 btrfs_item_key_to_cpu(path->nodes[0], &key,
962 path->slots[0]);
963 if (key.objectid != inode->i_ino ||
964 key.type != BTRFS_INODE_REF_KEY)
965 break;
966 ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
967 ptr_end = ptr + btrfs_item_size_nr(path->nodes[0],
968 path->slots[0]);
Chris Masond3977122009-01-05 21:25:51 -0500969 while (ptr < ptr_end) {
Chris Masone02119d2008-09-05 16:13:11 -0400970 struct btrfs_inode_ref *ref;
971
972 ref = (struct btrfs_inode_ref *)ptr;
973 name_len = btrfs_inode_ref_name_len(path->nodes[0],
974 ref);
975 ptr = (unsigned long)(ref + 1) + name_len;
976 nlink++;
977 }
978
979 if (key.offset == 0)
980 break;
981 key.offset--;
982 btrfs_release_path(root, path);
983 }
Chris Mason12fcfd22009-03-24 10:24:20 -0400984 btrfs_release_path(root, path);
Chris Masone02119d2008-09-05 16:13:11 -0400985 if (nlink != inode->i_nlink) {
986 inode->i_nlink = nlink;
987 btrfs_update_inode(trans, root, inode);
988 }
Chris Mason8d5bf1c2008-09-11 15:51:21 -0400989 BTRFS_I(inode)->index_cnt = (u64)-1;
Chris Masone02119d2008-09-05 16:13:11 -0400990
Chris Mason12fcfd22009-03-24 10:24:20 -0400991 if (inode->i_nlink == 0 && S_ISDIR(inode->i_mode)) {
992 ret = replay_dir_deletes(trans, root, NULL, path,
993 inode->i_ino, 1);
994 BUG_ON(ret);
995 }
996 btrfs_free_path(path);
997
Chris Masone02119d2008-09-05 16:13:11 -0400998 return 0;
999}
1000
1001static noinline int fixup_inode_link_counts(struct btrfs_trans_handle *trans,
1002 struct btrfs_root *root,
1003 struct btrfs_path *path)
1004{
1005 int ret;
1006 struct btrfs_key key;
1007 struct inode *inode;
1008
1009 key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID;
1010 key.type = BTRFS_ORPHAN_ITEM_KEY;
1011 key.offset = (u64)-1;
Chris Masond3977122009-01-05 21:25:51 -05001012 while (1) {
Chris Masone02119d2008-09-05 16:13:11 -04001013 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1014 if (ret < 0)
1015 break;
1016
1017 if (ret == 1) {
1018 if (path->slots[0] == 0)
1019 break;
1020 path->slots[0]--;
1021 }
1022
1023 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1024 if (key.objectid != BTRFS_TREE_LOG_FIXUP_OBJECTID ||
1025 key.type != BTRFS_ORPHAN_ITEM_KEY)
1026 break;
1027
1028 ret = btrfs_del_item(trans, root, path);
1029 BUG_ON(ret);
1030
1031 btrfs_release_path(root, path);
1032 inode = read_one_inode(root, key.offset);
1033 BUG_ON(!inode);
1034
1035 ret = fixup_inode_link_count(trans, root, inode);
1036 BUG_ON(ret);
1037
1038 iput(inode);
1039
Chris Mason12fcfd22009-03-24 10:24:20 -04001040 /*
1041 * fixup on a directory may create new entries,
1042 * make sure we always look for the highset possible
1043 * offset
1044 */
1045 key.offset = (u64)-1;
Chris Masone02119d2008-09-05 16:13:11 -04001046 }
1047 btrfs_release_path(root, path);
1048 return 0;
1049}
1050
1051
1052/*
1053 * record a given inode in the fixup dir so we can check its link
1054 * count when replay is done. The link count is incremented here
1055 * so the inode won't go away until we check it
1056 */
1057static noinline int link_to_fixup_dir(struct btrfs_trans_handle *trans,
1058 struct btrfs_root *root,
1059 struct btrfs_path *path,
1060 u64 objectid)
1061{
1062 struct btrfs_key key;
1063 int ret = 0;
1064 struct inode *inode;
1065
1066 inode = read_one_inode(root, objectid);
1067 BUG_ON(!inode);
1068
1069 key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID;
1070 btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
1071 key.offset = objectid;
1072
1073 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1074
1075 btrfs_release_path(root, path);
1076 if (ret == 0) {
1077 btrfs_inc_nlink(inode);
1078 btrfs_update_inode(trans, root, inode);
1079 } else if (ret == -EEXIST) {
1080 ret = 0;
1081 } else {
1082 BUG();
1083 }
1084 iput(inode);
1085
1086 return ret;
1087}
1088
1089/*
1090 * when replaying the log for a directory, we only insert names
1091 * for inodes that actually exist. This means an fsync on a directory
1092 * does not implicitly fsync all the new files in it
1093 */
1094static noinline int insert_one_name(struct btrfs_trans_handle *trans,
1095 struct btrfs_root *root,
1096 struct btrfs_path *path,
1097 u64 dirid, u64 index,
1098 char *name, int name_len, u8 type,
1099 struct btrfs_key *location)
1100{
1101 struct inode *inode;
1102 struct inode *dir;
1103 int ret;
1104
1105 inode = read_one_inode(root, location->objectid);
1106 if (!inode)
1107 return -ENOENT;
1108
1109 dir = read_one_inode(root, dirid);
1110 if (!dir) {
1111 iput(inode);
1112 return -EIO;
1113 }
1114 ret = btrfs_add_link(trans, dir, inode, name, name_len, 1, index);
1115
1116 /* FIXME, put inode into FIXUP list */
1117
1118 iput(inode);
1119 iput(dir);
1120 return ret;
1121}
1122
1123/*
1124 * take a single entry in a log directory item and replay it into
1125 * the subvolume.
1126 *
1127 * if a conflicting item exists in the subdirectory already,
1128 * the inode it points to is unlinked and put into the link count
1129 * fix up tree.
1130 *
1131 * If a name from the log points to a file or directory that does
1132 * not exist in the FS, it is skipped. fsyncs on directories
1133 * do not force down inodes inside that directory, just changes to the
1134 * names or unlinks in a directory.
1135 */
1136static noinline int replay_one_name(struct btrfs_trans_handle *trans,
1137 struct btrfs_root *root,
1138 struct btrfs_path *path,
1139 struct extent_buffer *eb,
1140 struct btrfs_dir_item *di,
1141 struct btrfs_key *key)
1142{
1143 char *name;
1144 int name_len;
1145 struct btrfs_dir_item *dst_di;
1146 struct btrfs_key found_key;
1147 struct btrfs_key log_key;
1148 struct inode *dir;
Chris Masone02119d2008-09-05 16:13:11 -04001149 u8 log_type;
Chris Mason4bef0842008-09-08 11:18:08 -04001150 int exists;
Chris Masone02119d2008-09-05 16:13:11 -04001151 int ret;
1152
1153 dir = read_one_inode(root, key->objectid);
1154 BUG_ON(!dir);
1155
1156 name_len = btrfs_dir_name_len(eb, di);
1157 name = kmalloc(name_len, GFP_NOFS);
1158 log_type = btrfs_dir_type(eb, di);
1159 read_extent_buffer(eb, name, (unsigned long)(di + 1),
1160 name_len);
1161
1162 btrfs_dir_item_key_to_cpu(eb, di, &log_key);
Chris Mason4bef0842008-09-08 11:18:08 -04001163 exists = btrfs_lookup_inode(trans, root, path, &log_key, 0);
1164 if (exists == 0)
1165 exists = 1;
1166 else
1167 exists = 0;
1168 btrfs_release_path(root, path);
1169
Chris Masone02119d2008-09-05 16:13:11 -04001170 if (key->type == BTRFS_DIR_ITEM_KEY) {
1171 dst_di = btrfs_lookup_dir_item(trans, root, path, key->objectid,
1172 name, name_len, 1);
Chris Masond3977122009-01-05 21:25:51 -05001173 } else if (key->type == BTRFS_DIR_INDEX_KEY) {
Chris Masone02119d2008-09-05 16:13:11 -04001174 dst_di = btrfs_lookup_dir_index_item(trans, root, path,
1175 key->objectid,
1176 key->offset, name,
1177 name_len, 1);
1178 } else {
1179 BUG();
1180 }
1181 if (!dst_di || IS_ERR(dst_di)) {
1182 /* we need a sequence number to insert, so we only
1183 * do inserts for the BTRFS_DIR_INDEX_KEY types
1184 */
1185 if (key->type != BTRFS_DIR_INDEX_KEY)
1186 goto out;
1187 goto insert;
1188 }
1189
1190 btrfs_dir_item_key_to_cpu(path->nodes[0], dst_di, &found_key);
1191 /* the existing item matches the logged item */
1192 if (found_key.objectid == log_key.objectid &&
1193 found_key.type == log_key.type &&
1194 found_key.offset == log_key.offset &&
1195 btrfs_dir_type(path->nodes[0], dst_di) == log_type) {
1196 goto out;
1197 }
1198
1199 /*
1200 * don't drop the conflicting directory entry if the inode
1201 * for the new entry doesn't exist
1202 */
Chris Mason4bef0842008-09-08 11:18:08 -04001203 if (!exists)
Chris Masone02119d2008-09-05 16:13:11 -04001204 goto out;
1205
Chris Masone02119d2008-09-05 16:13:11 -04001206 ret = drop_one_dir_item(trans, root, path, dir, dst_di);
1207 BUG_ON(ret);
1208
1209 if (key->type == BTRFS_DIR_INDEX_KEY)
1210 goto insert;
1211out:
1212 btrfs_release_path(root, path);
1213 kfree(name);
1214 iput(dir);
1215 return 0;
1216
1217insert:
1218 btrfs_release_path(root, path);
1219 ret = insert_one_name(trans, root, path, key->objectid, key->offset,
1220 name, name_len, log_type, &log_key);
1221
Stoyan Gaydarovc2934982009-04-02 17:05:11 -04001222 BUG_ON(ret && ret != -ENOENT);
Chris Masone02119d2008-09-05 16:13:11 -04001223 goto out;
1224}
1225
1226/*
1227 * find all the names in a directory item and reconcile them into
1228 * the subvolume. Only BTRFS_DIR_ITEM_KEY types will have more than
1229 * one name in a directory item, but the same code gets used for
1230 * both directory index types
1231 */
1232static noinline int replay_one_dir_item(struct btrfs_trans_handle *trans,
1233 struct btrfs_root *root,
1234 struct btrfs_path *path,
1235 struct extent_buffer *eb, int slot,
1236 struct btrfs_key *key)
1237{
1238 int ret;
1239 u32 item_size = btrfs_item_size_nr(eb, slot);
1240 struct btrfs_dir_item *di;
1241 int name_len;
1242 unsigned long ptr;
1243 unsigned long ptr_end;
1244
1245 ptr = btrfs_item_ptr_offset(eb, slot);
1246 ptr_end = ptr + item_size;
Chris Masond3977122009-01-05 21:25:51 -05001247 while (ptr < ptr_end) {
Chris Masone02119d2008-09-05 16:13:11 -04001248 di = (struct btrfs_dir_item *)ptr;
1249 name_len = btrfs_dir_name_len(eb, di);
1250 ret = replay_one_name(trans, root, path, eb, di, key);
1251 BUG_ON(ret);
1252 ptr = (unsigned long)(di + 1);
1253 ptr += name_len;
1254 }
1255 return 0;
1256}
1257
1258/*
1259 * directory replay has two parts. There are the standard directory
1260 * items in the log copied from the subvolume, and range items
1261 * created in the log while the subvolume was logged.
1262 *
1263 * The range items tell us which parts of the key space the log
1264 * is authoritative for. During replay, if a key in the subvolume
1265 * directory is in a logged range item, but not actually in the log
1266 * that means it was deleted from the directory before the fsync
1267 * and should be removed.
1268 */
1269static noinline int find_dir_range(struct btrfs_root *root,
1270 struct btrfs_path *path,
1271 u64 dirid, int key_type,
1272 u64 *start_ret, u64 *end_ret)
1273{
1274 struct btrfs_key key;
1275 u64 found_end;
1276 struct btrfs_dir_log_item *item;
1277 int ret;
1278 int nritems;
1279
1280 if (*start_ret == (u64)-1)
1281 return 1;
1282
1283 key.objectid = dirid;
1284 key.type = key_type;
1285 key.offset = *start_ret;
1286
1287 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1288 if (ret < 0)
1289 goto out;
1290 if (ret > 0) {
1291 if (path->slots[0] == 0)
1292 goto out;
1293 path->slots[0]--;
1294 }
1295 if (ret != 0)
1296 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1297
1298 if (key.type != key_type || key.objectid != dirid) {
1299 ret = 1;
1300 goto next;
1301 }
1302 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
1303 struct btrfs_dir_log_item);
1304 found_end = btrfs_dir_log_end(path->nodes[0], item);
1305
1306 if (*start_ret >= key.offset && *start_ret <= found_end) {
1307 ret = 0;
1308 *start_ret = key.offset;
1309 *end_ret = found_end;
1310 goto out;
1311 }
1312 ret = 1;
1313next:
1314 /* check the next slot in the tree to see if it is a valid item */
1315 nritems = btrfs_header_nritems(path->nodes[0]);
1316 if (path->slots[0] >= nritems) {
1317 ret = btrfs_next_leaf(root, path);
1318 if (ret)
1319 goto out;
1320 } else {
1321 path->slots[0]++;
1322 }
1323
1324 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1325
1326 if (key.type != key_type || key.objectid != dirid) {
1327 ret = 1;
1328 goto out;
1329 }
1330 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
1331 struct btrfs_dir_log_item);
1332 found_end = btrfs_dir_log_end(path->nodes[0], item);
1333 *start_ret = key.offset;
1334 *end_ret = found_end;
1335 ret = 0;
1336out:
1337 btrfs_release_path(root, path);
1338 return ret;
1339}
1340
1341/*
1342 * this looks for a given directory item in the log. If the directory
1343 * item is not in the log, the item is removed and the inode it points
1344 * to is unlinked
1345 */
1346static noinline int check_item_in_log(struct btrfs_trans_handle *trans,
1347 struct btrfs_root *root,
1348 struct btrfs_root *log,
1349 struct btrfs_path *path,
1350 struct btrfs_path *log_path,
1351 struct inode *dir,
1352 struct btrfs_key *dir_key)
1353{
1354 int ret;
1355 struct extent_buffer *eb;
1356 int slot;
1357 u32 item_size;
1358 struct btrfs_dir_item *di;
1359 struct btrfs_dir_item *log_di;
1360 int name_len;
1361 unsigned long ptr;
1362 unsigned long ptr_end;
1363 char *name;
1364 struct inode *inode;
1365 struct btrfs_key location;
1366
1367again:
1368 eb = path->nodes[0];
1369 slot = path->slots[0];
1370 item_size = btrfs_item_size_nr(eb, slot);
1371 ptr = btrfs_item_ptr_offset(eb, slot);
1372 ptr_end = ptr + item_size;
Chris Masond3977122009-01-05 21:25:51 -05001373 while (ptr < ptr_end) {
Chris Masone02119d2008-09-05 16:13:11 -04001374 di = (struct btrfs_dir_item *)ptr;
1375 name_len = btrfs_dir_name_len(eb, di);
1376 name = kmalloc(name_len, GFP_NOFS);
1377 if (!name) {
1378 ret = -ENOMEM;
1379 goto out;
1380 }
1381 read_extent_buffer(eb, name, (unsigned long)(di + 1),
1382 name_len);
1383 log_di = NULL;
Chris Mason12fcfd22009-03-24 10:24:20 -04001384 if (log && dir_key->type == BTRFS_DIR_ITEM_KEY) {
Chris Masone02119d2008-09-05 16:13:11 -04001385 log_di = btrfs_lookup_dir_item(trans, log, log_path,
1386 dir_key->objectid,
1387 name, name_len, 0);
Chris Mason12fcfd22009-03-24 10:24:20 -04001388 } else if (log && dir_key->type == BTRFS_DIR_INDEX_KEY) {
Chris Masone02119d2008-09-05 16:13:11 -04001389 log_di = btrfs_lookup_dir_index_item(trans, log,
1390 log_path,
1391 dir_key->objectid,
1392 dir_key->offset,
1393 name, name_len, 0);
1394 }
1395 if (!log_di || IS_ERR(log_di)) {
1396 btrfs_dir_item_key_to_cpu(eb, di, &location);
1397 btrfs_release_path(root, path);
1398 btrfs_release_path(log, log_path);
1399 inode = read_one_inode(root, location.objectid);
1400 BUG_ON(!inode);
1401
1402 ret = link_to_fixup_dir(trans, root,
1403 path, location.objectid);
1404 BUG_ON(ret);
1405 btrfs_inc_nlink(inode);
1406 ret = btrfs_unlink_inode(trans, root, dir, inode,
1407 name, name_len);
1408 BUG_ON(ret);
1409 kfree(name);
1410 iput(inode);
1411
1412 /* there might still be more names under this key
1413 * check and repeat if required
1414 */
1415 ret = btrfs_search_slot(NULL, root, dir_key, path,
1416 0, 0);
1417 if (ret == 0)
1418 goto again;
1419 ret = 0;
1420 goto out;
1421 }
1422 btrfs_release_path(log, log_path);
1423 kfree(name);
1424
1425 ptr = (unsigned long)(di + 1);
1426 ptr += name_len;
1427 }
1428 ret = 0;
1429out:
1430 btrfs_release_path(root, path);
1431 btrfs_release_path(log, log_path);
1432 return ret;
1433}
1434
1435/*
1436 * deletion replay happens before we copy any new directory items
1437 * out of the log or out of backreferences from inodes. It
1438 * scans the log to find ranges of keys that log is authoritative for,
1439 * and then scans the directory to find items in those ranges that are
1440 * not present in the log.
1441 *
1442 * Anything we don't find in the log is unlinked and removed from the
1443 * directory.
1444 */
1445static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans,
1446 struct btrfs_root *root,
1447 struct btrfs_root *log,
1448 struct btrfs_path *path,
Chris Mason12fcfd22009-03-24 10:24:20 -04001449 u64 dirid, int del_all)
Chris Masone02119d2008-09-05 16:13:11 -04001450{
1451 u64 range_start;
1452 u64 range_end;
1453 int key_type = BTRFS_DIR_LOG_ITEM_KEY;
1454 int ret = 0;
1455 struct btrfs_key dir_key;
1456 struct btrfs_key found_key;
1457 struct btrfs_path *log_path;
1458 struct inode *dir;
1459
1460 dir_key.objectid = dirid;
1461 dir_key.type = BTRFS_DIR_ITEM_KEY;
1462 log_path = btrfs_alloc_path();
1463 if (!log_path)
1464 return -ENOMEM;
1465
1466 dir = read_one_inode(root, dirid);
1467 /* it isn't an error if the inode isn't there, that can happen
1468 * because we replay the deletes before we copy in the inode item
1469 * from the log
1470 */
1471 if (!dir) {
1472 btrfs_free_path(log_path);
1473 return 0;
1474 }
1475again:
1476 range_start = 0;
1477 range_end = 0;
Chris Masond3977122009-01-05 21:25:51 -05001478 while (1) {
Chris Mason12fcfd22009-03-24 10:24:20 -04001479 if (del_all)
1480 range_end = (u64)-1;
1481 else {
1482 ret = find_dir_range(log, path, dirid, key_type,
1483 &range_start, &range_end);
1484 if (ret != 0)
1485 break;
1486 }
Chris Masone02119d2008-09-05 16:13:11 -04001487
1488 dir_key.offset = range_start;
Chris Masond3977122009-01-05 21:25:51 -05001489 while (1) {
Chris Masone02119d2008-09-05 16:13:11 -04001490 int nritems;
1491 ret = btrfs_search_slot(NULL, root, &dir_key, path,
1492 0, 0);
1493 if (ret < 0)
1494 goto out;
1495
1496 nritems = btrfs_header_nritems(path->nodes[0]);
1497 if (path->slots[0] >= nritems) {
1498 ret = btrfs_next_leaf(root, path);
1499 if (ret)
1500 break;
1501 }
1502 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1503 path->slots[0]);
1504 if (found_key.objectid != dirid ||
1505 found_key.type != dir_key.type)
1506 goto next_type;
1507
1508 if (found_key.offset > range_end)
1509 break;
1510
1511 ret = check_item_in_log(trans, root, log, path,
Chris Mason12fcfd22009-03-24 10:24:20 -04001512 log_path, dir,
1513 &found_key);
Chris Masone02119d2008-09-05 16:13:11 -04001514 BUG_ON(ret);
1515 if (found_key.offset == (u64)-1)
1516 break;
1517 dir_key.offset = found_key.offset + 1;
1518 }
1519 btrfs_release_path(root, path);
1520 if (range_end == (u64)-1)
1521 break;
1522 range_start = range_end + 1;
1523 }
1524
1525next_type:
1526 ret = 0;
1527 if (key_type == BTRFS_DIR_LOG_ITEM_KEY) {
1528 key_type = BTRFS_DIR_LOG_INDEX_KEY;
1529 dir_key.type = BTRFS_DIR_INDEX_KEY;
1530 btrfs_release_path(root, path);
1531 goto again;
1532 }
1533out:
1534 btrfs_release_path(root, path);
1535 btrfs_free_path(log_path);
1536 iput(dir);
1537 return ret;
1538}
1539
1540/*
1541 * the process_func used to replay items from the log tree. This
1542 * gets called in two different stages. The first stage just looks
1543 * for inodes and makes sure they are all copied into the subvolume.
1544 *
1545 * The second stage copies all the other item types from the log into
1546 * the subvolume. The two stage approach is slower, but gets rid of
1547 * lots of complexity around inodes referencing other inodes that exist
1548 * only in the log (references come from either directory items or inode
1549 * back refs).
1550 */
1551static int replay_one_buffer(struct btrfs_root *log, struct extent_buffer *eb,
1552 struct walk_control *wc, u64 gen)
1553{
1554 int nritems;
1555 struct btrfs_path *path;
1556 struct btrfs_root *root = wc->replay_dest;
1557 struct btrfs_key key;
1558 u32 item_size;
1559 int level;
1560 int i;
1561 int ret;
1562
1563 btrfs_read_buffer(eb, gen);
1564
1565 level = btrfs_header_level(eb);
1566
1567 if (level != 0)
1568 return 0;
1569
1570 path = btrfs_alloc_path();
1571 BUG_ON(!path);
1572
1573 nritems = btrfs_header_nritems(eb);
1574 for (i = 0; i < nritems; i++) {
1575 btrfs_item_key_to_cpu(eb, &key, i);
1576 item_size = btrfs_item_size_nr(eb, i);
1577
1578 /* inode keys are done during the first stage */
1579 if (key.type == BTRFS_INODE_ITEM_KEY &&
1580 wc->stage == LOG_WALK_REPLAY_INODES) {
1581 struct inode *inode;
1582 struct btrfs_inode_item *inode_item;
1583 u32 mode;
1584
1585 inode_item = btrfs_item_ptr(eb, i,
1586 struct btrfs_inode_item);
1587 mode = btrfs_inode_mode(eb, inode_item);
1588 if (S_ISDIR(mode)) {
1589 ret = replay_dir_deletes(wc->trans,
Chris Mason12fcfd22009-03-24 10:24:20 -04001590 root, log, path, key.objectid, 0);
Chris Masone02119d2008-09-05 16:13:11 -04001591 BUG_ON(ret);
1592 }
1593 ret = overwrite_item(wc->trans, root, path,
1594 eb, i, &key);
1595 BUG_ON(ret);
1596
1597 /* for regular files, truncate away
1598 * extents past the new EOF
1599 */
1600 if (S_ISREG(mode)) {
1601 inode = read_one_inode(root,
1602 key.objectid);
1603 BUG_ON(!inode);
1604
1605 ret = btrfs_truncate_inode_items(wc->trans,
1606 root, inode, inode->i_size,
1607 BTRFS_EXTENT_DATA_KEY);
1608 BUG_ON(ret);
Chris Masona74ac322009-03-24 10:24:13 -04001609
1610 /* if the nlink count is zero here, the iput
1611 * will free the inode. We bump it to make
1612 * sure it doesn't get freed until the link
1613 * count fixup is done
1614 */
1615 if (inode->i_nlink == 0) {
1616 btrfs_inc_nlink(inode);
1617 btrfs_update_inode(wc->trans,
1618 root, inode);
1619 }
Chris Masone02119d2008-09-05 16:13:11 -04001620 iput(inode);
1621 }
1622 ret = link_to_fixup_dir(wc->trans, root,
1623 path, key.objectid);
1624 BUG_ON(ret);
1625 }
1626 if (wc->stage < LOG_WALK_REPLAY_ALL)
1627 continue;
1628
1629 /* these keys are simply copied */
1630 if (key.type == BTRFS_XATTR_ITEM_KEY) {
1631 ret = overwrite_item(wc->trans, root, path,
1632 eb, i, &key);
1633 BUG_ON(ret);
1634 } else if (key.type == BTRFS_INODE_REF_KEY) {
1635 ret = add_inode_ref(wc->trans, root, log, path,
1636 eb, i, &key);
1637 BUG_ON(ret && ret != -ENOENT);
1638 } else if (key.type == BTRFS_EXTENT_DATA_KEY) {
1639 ret = replay_one_extent(wc->trans, root, path,
1640 eb, i, &key);
1641 BUG_ON(ret);
Chris Masone02119d2008-09-05 16:13:11 -04001642 } else if (key.type == BTRFS_DIR_ITEM_KEY ||
1643 key.type == BTRFS_DIR_INDEX_KEY) {
1644 ret = replay_one_dir_item(wc->trans, root, path,
1645 eb, i, &key);
1646 BUG_ON(ret);
1647 }
1648 }
1649 btrfs_free_path(path);
1650 return 0;
1651}
1652
Chris Masond3977122009-01-05 21:25:51 -05001653static noinline int walk_down_log_tree(struct btrfs_trans_handle *trans,
Chris Masone02119d2008-09-05 16:13:11 -04001654 struct btrfs_root *root,
1655 struct btrfs_path *path, int *level,
1656 struct walk_control *wc)
1657{
1658 u64 root_owner;
1659 u64 root_gen;
1660 u64 bytenr;
1661 u64 ptr_gen;
1662 struct extent_buffer *next;
1663 struct extent_buffer *cur;
1664 struct extent_buffer *parent;
1665 u32 blocksize;
1666 int ret = 0;
1667
1668 WARN_ON(*level < 0);
1669 WARN_ON(*level >= BTRFS_MAX_LEVEL);
1670
Chris Masond3977122009-01-05 21:25:51 -05001671 while (*level > 0) {
Chris Masone02119d2008-09-05 16:13:11 -04001672 WARN_ON(*level < 0);
1673 WARN_ON(*level >= BTRFS_MAX_LEVEL);
1674 cur = path->nodes[*level];
1675
1676 if (btrfs_header_level(cur) != *level)
1677 WARN_ON(1);
1678
1679 if (path->slots[*level] >=
1680 btrfs_header_nritems(cur))
1681 break;
1682
1683 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
1684 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
1685 blocksize = btrfs_level_size(root, *level - 1);
1686
1687 parent = path->nodes[*level];
1688 root_owner = btrfs_header_owner(parent);
1689 root_gen = btrfs_header_generation(parent);
1690
1691 next = btrfs_find_create_tree_block(root, bytenr, blocksize);
1692
1693 wc->process_func(root, next, wc, ptr_gen);
1694
1695 if (*level == 1) {
1696 path->slots[*level]++;
1697 if (wc->free) {
1698 btrfs_read_buffer(next, ptr_gen);
1699
1700 btrfs_tree_lock(next);
1701 clean_tree_block(trans, root, next);
Chris Masonb4ce94d2009-02-04 09:25:08 -05001702 btrfs_set_lock_blocking(next);
Chris Masone02119d2008-09-05 16:13:11 -04001703 btrfs_wait_tree_block_writeback(next);
1704 btrfs_tree_unlock(next);
1705
Chris Masone02119d2008-09-05 16:13:11 -04001706 WARN_ON(root_owner !=
1707 BTRFS_TREE_LOG_OBJECTID);
Chris Masond00aff02008-09-11 15:54:42 -04001708 ret = btrfs_free_reserved_extent(root,
1709 bytenr, blocksize);
Chris Masone02119d2008-09-05 16:13:11 -04001710 BUG_ON(ret);
1711 }
1712 free_extent_buffer(next);
1713 continue;
1714 }
1715 btrfs_read_buffer(next, ptr_gen);
1716
1717 WARN_ON(*level <= 0);
1718 if (path->nodes[*level-1])
1719 free_extent_buffer(path->nodes[*level-1]);
1720 path->nodes[*level-1] = next;
1721 *level = btrfs_header_level(next);
1722 path->slots[*level] = 0;
1723 cond_resched();
1724 }
1725 WARN_ON(*level < 0);
1726 WARN_ON(*level >= BTRFS_MAX_LEVEL);
1727
Chris Masond3977122009-01-05 21:25:51 -05001728 if (path->nodes[*level] == root->node)
Chris Masone02119d2008-09-05 16:13:11 -04001729 parent = path->nodes[*level];
Chris Masond3977122009-01-05 21:25:51 -05001730 else
Chris Masone02119d2008-09-05 16:13:11 -04001731 parent = path->nodes[*level + 1];
Chris Masond3977122009-01-05 21:25:51 -05001732
Chris Masone02119d2008-09-05 16:13:11 -04001733 bytenr = path->nodes[*level]->start;
1734
1735 blocksize = btrfs_level_size(root, *level);
1736 root_owner = btrfs_header_owner(parent);
1737 root_gen = btrfs_header_generation(parent);
1738
1739 wc->process_func(root, path->nodes[*level], wc,
1740 btrfs_header_generation(path->nodes[*level]));
1741
1742 if (wc->free) {
1743 next = path->nodes[*level];
1744 btrfs_tree_lock(next);
1745 clean_tree_block(trans, root, next);
Chris Masonb4ce94d2009-02-04 09:25:08 -05001746 btrfs_set_lock_blocking(next);
Chris Masone02119d2008-09-05 16:13:11 -04001747 btrfs_wait_tree_block_writeback(next);
1748 btrfs_tree_unlock(next);
1749
Chris Masone02119d2008-09-05 16:13:11 -04001750 WARN_ON(root_owner != BTRFS_TREE_LOG_OBJECTID);
Chris Masond00aff02008-09-11 15:54:42 -04001751 ret = btrfs_free_reserved_extent(root, bytenr, blocksize);
Chris Masone02119d2008-09-05 16:13:11 -04001752 BUG_ON(ret);
1753 }
1754 free_extent_buffer(path->nodes[*level]);
1755 path->nodes[*level] = NULL;
1756 *level += 1;
1757
1758 cond_resched();
1759 return 0;
1760}
1761
Chris Masond3977122009-01-05 21:25:51 -05001762static noinline int walk_up_log_tree(struct btrfs_trans_handle *trans,
Chris Masone02119d2008-09-05 16:13:11 -04001763 struct btrfs_root *root,
1764 struct btrfs_path *path, int *level,
1765 struct walk_control *wc)
1766{
1767 u64 root_owner;
1768 u64 root_gen;
1769 int i;
1770 int slot;
1771 int ret;
1772
Chris Masond3977122009-01-05 21:25:51 -05001773 for (i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) {
Chris Masone02119d2008-09-05 16:13:11 -04001774 slot = path->slots[i];
1775 if (slot < btrfs_header_nritems(path->nodes[i]) - 1) {
1776 struct extent_buffer *node;
1777 node = path->nodes[i];
1778 path->slots[i]++;
1779 *level = i;
1780 WARN_ON(*level == 0);
1781 return 0;
1782 } else {
Zheng Yan31840ae2008-09-23 13:14:14 -04001783 struct extent_buffer *parent;
1784 if (path->nodes[*level] == root->node)
1785 parent = path->nodes[*level];
1786 else
1787 parent = path->nodes[*level + 1];
1788
1789 root_owner = btrfs_header_owner(parent);
1790 root_gen = btrfs_header_generation(parent);
Chris Masone02119d2008-09-05 16:13:11 -04001791 wc->process_func(root, path->nodes[*level], wc,
1792 btrfs_header_generation(path->nodes[*level]));
1793 if (wc->free) {
1794 struct extent_buffer *next;
1795
1796 next = path->nodes[*level];
1797
1798 btrfs_tree_lock(next);
1799 clean_tree_block(trans, root, next);
Chris Masonb4ce94d2009-02-04 09:25:08 -05001800 btrfs_set_lock_blocking(next);
Chris Masone02119d2008-09-05 16:13:11 -04001801 btrfs_wait_tree_block_writeback(next);
1802 btrfs_tree_unlock(next);
1803
Chris Masone02119d2008-09-05 16:13:11 -04001804 WARN_ON(root_owner != BTRFS_TREE_LOG_OBJECTID);
Chris Masond00aff02008-09-11 15:54:42 -04001805 ret = btrfs_free_reserved_extent(root,
Chris Masone02119d2008-09-05 16:13:11 -04001806 path->nodes[*level]->start,
Chris Masond00aff02008-09-11 15:54:42 -04001807 path->nodes[*level]->len);
Chris Masone02119d2008-09-05 16:13:11 -04001808 BUG_ON(ret);
1809 }
1810 free_extent_buffer(path->nodes[*level]);
1811 path->nodes[*level] = NULL;
1812 *level = i + 1;
1813 }
1814 }
1815 return 1;
1816}
1817
1818/*
1819 * drop the reference count on the tree rooted at 'snap'. This traverses
1820 * the tree freeing any blocks that have a ref count of zero after being
1821 * decremented.
1822 */
1823static int walk_log_tree(struct btrfs_trans_handle *trans,
1824 struct btrfs_root *log, struct walk_control *wc)
1825{
1826 int ret = 0;
1827 int wret;
1828 int level;
1829 struct btrfs_path *path;
1830 int i;
1831 int orig_level;
1832
1833 path = btrfs_alloc_path();
1834 BUG_ON(!path);
1835
1836 level = btrfs_header_level(log->node);
1837 orig_level = level;
1838 path->nodes[level] = log->node;
1839 extent_buffer_get(log->node);
1840 path->slots[level] = 0;
1841
Chris Masond3977122009-01-05 21:25:51 -05001842 while (1) {
Chris Masone02119d2008-09-05 16:13:11 -04001843 wret = walk_down_log_tree(trans, log, path, &level, wc);
1844 if (wret > 0)
1845 break;
1846 if (wret < 0)
1847 ret = wret;
1848
1849 wret = walk_up_log_tree(trans, log, path, &level, wc);
1850 if (wret > 0)
1851 break;
1852 if (wret < 0)
1853 ret = wret;
1854 }
1855
1856 /* was the root node processed? if not, catch it here */
1857 if (path->nodes[orig_level]) {
1858 wc->process_func(log, path->nodes[orig_level], wc,
1859 btrfs_header_generation(path->nodes[orig_level]));
1860 if (wc->free) {
1861 struct extent_buffer *next;
1862
1863 next = path->nodes[orig_level];
1864
1865 btrfs_tree_lock(next);
1866 clean_tree_block(trans, log, next);
Chris Masonb4ce94d2009-02-04 09:25:08 -05001867 btrfs_set_lock_blocking(next);
Chris Masone02119d2008-09-05 16:13:11 -04001868 btrfs_wait_tree_block_writeback(next);
1869 btrfs_tree_unlock(next);
1870
Chris Masone02119d2008-09-05 16:13:11 -04001871 WARN_ON(log->root_key.objectid !=
1872 BTRFS_TREE_LOG_OBJECTID);
Chris Masond00aff02008-09-11 15:54:42 -04001873 ret = btrfs_free_reserved_extent(log, next->start,
1874 next->len);
Chris Masone02119d2008-09-05 16:13:11 -04001875 BUG_ON(ret);
1876 }
1877 }
1878
1879 for (i = 0; i <= orig_level; i++) {
1880 if (path->nodes[i]) {
1881 free_extent_buffer(path->nodes[i]);
1882 path->nodes[i] = NULL;
1883 }
1884 }
1885 btrfs_free_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04001886 return ret;
1887}
1888
Yan Zheng7237f182009-01-21 12:54:03 -05001889/*
1890 * helper function to update the item for a given subvolumes log root
1891 * in the tree of log roots
1892 */
1893static int update_log_root(struct btrfs_trans_handle *trans,
1894 struct btrfs_root *log)
1895{
1896 int ret;
1897
1898 if (log->log_transid == 1) {
1899 /* insert root item on the first sync */
1900 ret = btrfs_insert_root(trans, log->fs_info->log_root_tree,
1901 &log->root_key, &log->root_item);
1902 } else {
1903 ret = btrfs_update_root(trans, log->fs_info->log_root_tree,
1904 &log->root_key, &log->root_item);
1905 }
1906 return ret;
1907}
1908
Chris Mason12fcfd22009-03-24 10:24:20 -04001909static int wait_log_commit(struct btrfs_trans_handle *trans,
1910 struct btrfs_root *root, unsigned long transid)
Chris Masone02119d2008-09-05 16:13:11 -04001911{
1912 DEFINE_WAIT(wait);
Yan Zheng7237f182009-01-21 12:54:03 -05001913 int index = transid % 2;
Chris Masone02119d2008-09-05 16:13:11 -04001914
Yan Zheng7237f182009-01-21 12:54:03 -05001915 /*
1916 * we only allow two pending log transactions at a time,
1917 * so we know that if ours is more than 2 older than the
1918 * current transaction, we're done
1919 */
Chris Masone02119d2008-09-05 16:13:11 -04001920 do {
Yan Zheng7237f182009-01-21 12:54:03 -05001921 prepare_to_wait(&root->log_commit_wait[index],
1922 &wait, TASK_UNINTERRUPTIBLE);
1923 mutex_unlock(&root->log_mutex);
Chris Mason12fcfd22009-03-24 10:24:20 -04001924
1925 if (root->fs_info->last_trans_log_full_commit !=
1926 trans->transid && root->log_transid < transid + 2 &&
Yan Zheng7237f182009-01-21 12:54:03 -05001927 atomic_read(&root->log_commit[index]))
Chris Masone02119d2008-09-05 16:13:11 -04001928 schedule();
Chris Mason12fcfd22009-03-24 10:24:20 -04001929
Yan Zheng7237f182009-01-21 12:54:03 -05001930 finish_wait(&root->log_commit_wait[index], &wait);
1931 mutex_lock(&root->log_mutex);
1932 } while (root->log_transid < transid + 2 &&
1933 atomic_read(&root->log_commit[index]));
1934 return 0;
1935}
1936
Chris Mason12fcfd22009-03-24 10:24:20 -04001937static int wait_for_writer(struct btrfs_trans_handle *trans,
1938 struct btrfs_root *root)
Yan Zheng7237f182009-01-21 12:54:03 -05001939{
1940 DEFINE_WAIT(wait);
1941 while (atomic_read(&root->log_writers)) {
1942 prepare_to_wait(&root->log_writer_wait,
1943 &wait, TASK_UNINTERRUPTIBLE);
1944 mutex_unlock(&root->log_mutex);
Chris Mason12fcfd22009-03-24 10:24:20 -04001945 if (root->fs_info->last_trans_log_full_commit !=
1946 trans->transid && atomic_read(&root->log_writers))
Yan Zheng7237f182009-01-21 12:54:03 -05001947 schedule();
1948 mutex_lock(&root->log_mutex);
1949 finish_wait(&root->log_writer_wait, &wait);
1950 }
Chris Masone02119d2008-09-05 16:13:11 -04001951 return 0;
1952}
1953
1954/*
1955 * btrfs_sync_log does sends a given tree log down to the disk and
1956 * updates the super blocks to record it. When this call is done,
Chris Mason12fcfd22009-03-24 10:24:20 -04001957 * you know that any inodes previously logged are safely on disk only
1958 * if it returns 0.
1959 *
1960 * Any other return value means you need to call btrfs_commit_transaction.
1961 * Some of the edge cases for fsyncing directories that have had unlinks
1962 * or renames done in the past mean that sometimes the only safe
1963 * fsync is to commit the whole FS. When btrfs_sync_log returns -EAGAIN,
1964 * that has happened.
Chris Masone02119d2008-09-05 16:13:11 -04001965 */
1966int btrfs_sync_log(struct btrfs_trans_handle *trans,
1967 struct btrfs_root *root)
1968{
Yan Zheng7237f182009-01-21 12:54:03 -05001969 int index1;
1970 int index2;
Chris Masone02119d2008-09-05 16:13:11 -04001971 int ret;
Chris Masone02119d2008-09-05 16:13:11 -04001972 struct btrfs_root *log = root->log_root;
Yan Zheng7237f182009-01-21 12:54:03 -05001973 struct btrfs_root *log_root_tree = root->fs_info->log_root_tree;
Chris Masone02119d2008-09-05 16:13:11 -04001974
Yan Zheng7237f182009-01-21 12:54:03 -05001975 mutex_lock(&root->log_mutex);
1976 index1 = root->log_transid % 2;
1977 if (atomic_read(&root->log_commit[index1])) {
Chris Mason12fcfd22009-03-24 10:24:20 -04001978 wait_log_commit(trans, root, root->log_transid);
Yan Zheng7237f182009-01-21 12:54:03 -05001979 mutex_unlock(&root->log_mutex);
1980 return 0;
Chris Masone02119d2008-09-05 16:13:11 -04001981 }
Yan Zheng7237f182009-01-21 12:54:03 -05001982 atomic_set(&root->log_commit[index1], 1);
1983
1984 /* wait for previous tree log sync to complete */
1985 if (atomic_read(&root->log_commit[(index1 + 1) % 2]))
Chris Mason12fcfd22009-03-24 10:24:20 -04001986 wait_log_commit(trans, root, root->log_transid - 1);
Chris Masone02119d2008-09-05 16:13:11 -04001987
Chris Masond3977122009-01-05 21:25:51 -05001988 while (1) {
Yan Zheng7237f182009-01-21 12:54:03 -05001989 unsigned long batch = root->log_batch;
1990 mutex_unlock(&root->log_mutex);
Chris Masone02119d2008-09-05 16:13:11 -04001991 schedule_timeout_uninterruptible(1);
Yan Zheng7237f182009-01-21 12:54:03 -05001992 mutex_lock(&root->log_mutex);
Chris Mason12fcfd22009-03-24 10:24:20 -04001993
1994 wait_for_writer(trans, root);
Yan Zheng7237f182009-01-21 12:54:03 -05001995 if (batch == root->log_batch)
Chris Masone02119d2008-09-05 16:13:11 -04001996 break;
1997 }
Chris Masond0c803c2008-09-11 16:17:57 -04001998
Chris Mason12fcfd22009-03-24 10:24:20 -04001999 /* bail out if we need to do a full commit */
2000 if (root->fs_info->last_trans_log_full_commit == trans->transid) {
2001 ret = -EAGAIN;
2002 mutex_unlock(&root->log_mutex);
2003 goto out;
2004 }
2005
Chris Masond0c803c2008-09-11 16:17:57 -04002006 ret = btrfs_write_and_wait_marked_extents(log, &log->dirty_log_pages);
Chris Masone02119d2008-09-05 16:13:11 -04002007 BUG_ON(ret);
Yan Zheng7237f182009-01-21 12:54:03 -05002008
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002009 btrfs_set_root_node(&log->root_item, log->node);
Yan Zheng7237f182009-01-21 12:54:03 -05002010
2011 root->log_batch = 0;
2012 root->log_transid++;
2013 log->log_transid = root->log_transid;
2014 smp_mb();
2015 /*
2016 * log tree has been flushed to disk, new modifications of
2017 * the log will be written to new positions. so it's safe to
2018 * allow log writers to go in.
2019 */
2020 mutex_unlock(&root->log_mutex);
2021
2022 mutex_lock(&log_root_tree->log_mutex);
2023 log_root_tree->log_batch++;
2024 atomic_inc(&log_root_tree->log_writers);
2025 mutex_unlock(&log_root_tree->log_mutex);
2026
2027 ret = update_log_root(trans, log);
2028 BUG_ON(ret);
2029
2030 mutex_lock(&log_root_tree->log_mutex);
2031 if (atomic_dec_and_test(&log_root_tree->log_writers)) {
2032 smp_mb();
2033 if (waitqueue_active(&log_root_tree->log_writer_wait))
2034 wake_up(&log_root_tree->log_writer_wait);
2035 }
2036
2037 index2 = log_root_tree->log_transid % 2;
2038 if (atomic_read(&log_root_tree->log_commit[index2])) {
Chris Mason12fcfd22009-03-24 10:24:20 -04002039 wait_log_commit(trans, log_root_tree,
2040 log_root_tree->log_transid);
Yan Zheng7237f182009-01-21 12:54:03 -05002041 mutex_unlock(&log_root_tree->log_mutex);
2042 goto out;
2043 }
2044 atomic_set(&log_root_tree->log_commit[index2], 1);
2045
Chris Mason12fcfd22009-03-24 10:24:20 -04002046 if (atomic_read(&log_root_tree->log_commit[(index2 + 1) % 2])) {
2047 wait_log_commit(trans, log_root_tree,
2048 log_root_tree->log_transid - 1);
2049 }
Yan Zheng7237f182009-01-21 12:54:03 -05002050
Chris Mason12fcfd22009-03-24 10:24:20 -04002051 wait_for_writer(trans, log_root_tree);
2052
2053 /*
2054 * now that we've moved on to the tree of log tree roots,
2055 * check the full commit flag again
2056 */
2057 if (root->fs_info->last_trans_log_full_commit == trans->transid) {
2058 mutex_unlock(&log_root_tree->log_mutex);
2059 ret = -EAGAIN;
2060 goto out_wake_log_root;
2061 }
Yan Zheng7237f182009-01-21 12:54:03 -05002062
2063 ret = btrfs_write_and_wait_marked_extents(log_root_tree,
2064 &log_root_tree->dirty_log_pages);
Chris Masone02119d2008-09-05 16:13:11 -04002065 BUG_ON(ret);
2066
2067 btrfs_set_super_log_root(&root->fs_info->super_for_commit,
Yan Zheng7237f182009-01-21 12:54:03 -05002068 log_root_tree->node->start);
Chris Masone02119d2008-09-05 16:13:11 -04002069 btrfs_set_super_log_root_level(&root->fs_info->super_for_commit,
Yan Zheng7237f182009-01-21 12:54:03 -05002070 btrfs_header_level(log_root_tree->node));
Chris Masone02119d2008-09-05 16:13:11 -04002071
Yan Zheng7237f182009-01-21 12:54:03 -05002072 log_root_tree->log_batch = 0;
2073 log_root_tree->log_transid++;
Chris Masone02119d2008-09-05 16:13:11 -04002074 smp_mb();
Yan Zheng7237f182009-01-21 12:54:03 -05002075
2076 mutex_unlock(&log_root_tree->log_mutex);
2077
2078 /*
2079 * nobody else is going to jump in and write the the ctree
2080 * super here because the log_commit atomic below is protecting
2081 * us. We must be called with a transaction handle pinning
2082 * the running transaction open, so a full commit can't hop
2083 * in and cause problems either.
2084 */
2085 write_ctree_super(trans, root->fs_info->tree_root, 2);
Chris Mason12fcfd22009-03-24 10:24:20 -04002086 ret = 0;
Yan Zheng7237f182009-01-21 12:54:03 -05002087
Chris Mason12fcfd22009-03-24 10:24:20 -04002088out_wake_log_root:
Yan Zheng7237f182009-01-21 12:54:03 -05002089 atomic_set(&log_root_tree->log_commit[index2], 0);
2090 smp_mb();
2091 if (waitqueue_active(&log_root_tree->log_commit_wait[index2]))
2092 wake_up(&log_root_tree->log_commit_wait[index2]);
Chris Masone02119d2008-09-05 16:13:11 -04002093out:
Yan Zheng7237f182009-01-21 12:54:03 -05002094 atomic_set(&root->log_commit[index1], 0);
2095 smp_mb();
2096 if (waitqueue_active(&root->log_commit_wait[index1]))
2097 wake_up(&root->log_commit_wait[index1]);
Chris Masone02119d2008-09-05 16:13:11 -04002098 return 0;
Chris Masone02119d2008-09-05 16:13:11 -04002099}
2100
Chris Mason12fcfd22009-03-24 10:24:20 -04002101/*
2102 * free all the extents used by the tree log. This should be called
Chris Masone02119d2008-09-05 16:13:11 -04002103 * at commit time of the full transaction
2104 */
2105int btrfs_free_log(struct btrfs_trans_handle *trans, struct btrfs_root *root)
2106{
2107 int ret;
2108 struct btrfs_root *log;
2109 struct key;
Chris Masond0c803c2008-09-11 16:17:57 -04002110 u64 start;
2111 u64 end;
Chris Masone02119d2008-09-05 16:13:11 -04002112 struct walk_control wc = {
2113 .free = 1,
2114 .process_func = process_one_buffer
2115 };
2116
Yan Zheng07d400a2009-01-06 11:42:00 -05002117 if (!root->log_root || root->fs_info->log_root_recovering)
Chris Masone02119d2008-09-05 16:13:11 -04002118 return 0;
2119
2120 log = root->log_root;
2121 ret = walk_log_tree(trans, log, &wc);
2122 BUG_ON(ret);
2123
Chris Masond3977122009-01-05 21:25:51 -05002124 while (1) {
Chris Masond0c803c2008-09-11 16:17:57 -04002125 ret = find_first_extent_bit(&log->dirty_log_pages,
2126 0, &start, &end, EXTENT_DIRTY);
2127 if (ret)
2128 break;
2129
2130 clear_extent_dirty(&log->dirty_log_pages,
2131 start, end, GFP_NOFS);
2132 }
2133
Yan Zheng7237f182009-01-21 12:54:03 -05002134 if (log->log_transid > 0) {
2135 ret = btrfs_del_root(trans, root->fs_info->log_root_tree,
2136 &log->root_key);
2137 BUG_ON(ret);
2138 }
Chris Masone02119d2008-09-05 16:13:11 -04002139 root->log_root = NULL;
Yan Zheng7237f182009-01-21 12:54:03 -05002140 free_extent_buffer(log->node);
2141 kfree(log);
Chris Masone02119d2008-09-05 16:13:11 -04002142 return 0;
2143}
2144
2145/*
Chris Masone02119d2008-09-05 16:13:11 -04002146 * If both a file and directory are logged, and unlinks or renames are
2147 * mixed in, we have a few interesting corners:
2148 *
2149 * create file X in dir Y
2150 * link file X to X.link in dir Y
2151 * fsync file X
2152 * unlink file X but leave X.link
2153 * fsync dir Y
2154 *
2155 * After a crash we would expect only X.link to exist. But file X
2156 * didn't get fsync'd again so the log has back refs for X and X.link.
2157 *
2158 * We solve this by removing directory entries and inode backrefs from the
2159 * log when a file that was logged in the current transaction is
2160 * unlinked. Any later fsync will include the updated log entries, and
2161 * we'll be able to reconstruct the proper directory items from backrefs.
2162 *
2163 * This optimizations allows us to avoid relogging the entire inode
2164 * or the entire directory.
2165 */
2166int btrfs_del_dir_entries_in_log(struct btrfs_trans_handle *trans,
2167 struct btrfs_root *root,
2168 const char *name, int name_len,
2169 struct inode *dir, u64 index)
2170{
2171 struct btrfs_root *log;
2172 struct btrfs_dir_item *di;
2173 struct btrfs_path *path;
2174 int ret;
2175 int bytes_del = 0;
2176
Chris Mason3a5f1d42008-09-11 15:53:37 -04002177 if (BTRFS_I(dir)->logged_trans < trans->transid)
2178 return 0;
2179
Chris Masone02119d2008-09-05 16:13:11 -04002180 ret = join_running_log_trans(root);
2181 if (ret)
2182 return 0;
2183
2184 mutex_lock(&BTRFS_I(dir)->log_mutex);
2185
2186 log = root->log_root;
2187 path = btrfs_alloc_path();
2188 di = btrfs_lookup_dir_item(trans, log, path, dir->i_ino,
2189 name, name_len, -1);
2190 if (di && !IS_ERR(di)) {
2191 ret = btrfs_delete_one_dir_name(trans, log, path, di);
2192 bytes_del += name_len;
2193 BUG_ON(ret);
2194 }
2195 btrfs_release_path(log, path);
2196 di = btrfs_lookup_dir_index_item(trans, log, path, dir->i_ino,
2197 index, name, name_len, -1);
2198 if (di && !IS_ERR(di)) {
2199 ret = btrfs_delete_one_dir_name(trans, log, path, di);
2200 bytes_del += name_len;
2201 BUG_ON(ret);
2202 }
2203
2204 /* update the directory size in the log to reflect the names
2205 * we have removed
2206 */
2207 if (bytes_del) {
2208 struct btrfs_key key;
2209
2210 key.objectid = dir->i_ino;
2211 key.offset = 0;
2212 key.type = BTRFS_INODE_ITEM_KEY;
2213 btrfs_release_path(log, path);
2214
2215 ret = btrfs_search_slot(trans, log, &key, path, 0, 1);
2216 if (ret == 0) {
2217 struct btrfs_inode_item *item;
2218 u64 i_size;
2219
2220 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2221 struct btrfs_inode_item);
2222 i_size = btrfs_inode_size(path->nodes[0], item);
2223 if (i_size > bytes_del)
2224 i_size -= bytes_del;
2225 else
2226 i_size = 0;
2227 btrfs_set_inode_size(path->nodes[0], item, i_size);
2228 btrfs_mark_buffer_dirty(path->nodes[0]);
2229 } else
2230 ret = 0;
2231 btrfs_release_path(log, path);
2232 }
2233
2234 btrfs_free_path(path);
2235 mutex_unlock(&BTRFS_I(dir)->log_mutex);
Chris Mason12fcfd22009-03-24 10:24:20 -04002236 btrfs_end_log_trans(root);
Chris Masone02119d2008-09-05 16:13:11 -04002237
2238 return 0;
2239}
2240
2241/* see comments for btrfs_del_dir_entries_in_log */
2242int btrfs_del_inode_ref_in_log(struct btrfs_trans_handle *trans,
2243 struct btrfs_root *root,
2244 const char *name, int name_len,
2245 struct inode *inode, u64 dirid)
2246{
2247 struct btrfs_root *log;
2248 u64 index;
2249 int ret;
2250
Chris Mason3a5f1d42008-09-11 15:53:37 -04002251 if (BTRFS_I(inode)->logged_trans < trans->transid)
2252 return 0;
2253
Chris Masone02119d2008-09-05 16:13:11 -04002254 ret = join_running_log_trans(root);
2255 if (ret)
2256 return 0;
2257 log = root->log_root;
2258 mutex_lock(&BTRFS_I(inode)->log_mutex);
2259
2260 ret = btrfs_del_inode_ref(trans, log, name, name_len, inode->i_ino,
2261 dirid, &index);
2262 mutex_unlock(&BTRFS_I(inode)->log_mutex);
Chris Mason12fcfd22009-03-24 10:24:20 -04002263 btrfs_end_log_trans(root);
Chris Masone02119d2008-09-05 16:13:11 -04002264
Chris Masone02119d2008-09-05 16:13:11 -04002265 return ret;
2266}
2267
2268/*
2269 * creates a range item in the log for 'dirid'. first_offset and
2270 * last_offset tell us which parts of the key space the log should
2271 * be considered authoritative for.
2272 */
2273static noinline int insert_dir_log_key(struct btrfs_trans_handle *trans,
2274 struct btrfs_root *log,
2275 struct btrfs_path *path,
2276 int key_type, u64 dirid,
2277 u64 first_offset, u64 last_offset)
2278{
2279 int ret;
2280 struct btrfs_key key;
2281 struct btrfs_dir_log_item *item;
2282
2283 key.objectid = dirid;
2284 key.offset = first_offset;
2285 if (key_type == BTRFS_DIR_ITEM_KEY)
2286 key.type = BTRFS_DIR_LOG_ITEM_KEY;
2287 else
2288 key.type = BTRFS_DIR_LOG_INDEX_KEY;
2289 ret = btrfs_insert_empty_item(trans, log, path, &key, sizeof(*item));
2290 BUG_ON(ret);
2291
2292 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2293 struct btrfs_dir_log_item);
2294 btrfs_set_dir_log_end(path->nodes[0], item, last_offset);
2295 btrfs_mark_buffer_dirty(path->nodes[0]);
2296 btrfs_release_path(log, path);
2297 return 0;
2298}
2299
2300/*
2301 * log all the items included in the current transaction for a given
2302 * directory. This also creates the range items in the log tree required
2303 * to replay anything deleted before the fsync
2304 */
2305static noinline int log_dir_items(struct btrfs_trans_handle *trans,
2306 struct btrfs_root *root, struct inode *inode,
2307 struct btrfs_path *path,
2308 struct btrfs_path *dst_path, int key_type,
2309 u64 min_offset, u64 *last_offset_ret)
2310{
2311 struct btrfs_key min_key;
2312 struct btrfs_key max_key;
2313 struct btrfs_root *log = root->log_root;
2314 struct extent_buffer *src;
2315 int ret;
2316 int i;
2317 int nritems;
2318 u64 first_offset = min_offset;
2319 u64 last_offset = (u64)-1;
2320
2321 log = root->log_root;
2322 max_key.objectid = inode->i_ino;
2323 max_key.offset = (u64)-1;
2324 max_key.type = key_type;
2325
2326 min_key.objectid = inode->i_ino;
2327 min_key.type = key_type;
2328 min_key.offset = min_offset;
2329
2330 path->keep_locks = 1;
2331
2332 ret = btrfs_search_forward(root, &min_key, &max_key,
2333 path, 0, trans->transid);
2334
2335 /*
2336 * we didn't find anything from this transaction, see if there
2337 * is anything at all
2338 */
2339 if (ret != 0 || min_key.objectid != inode->i_ino ||
2340 min_key.type != key_type) {
2341 min_key.objectid = inode->i_ino;
2342 min_key.type = key_type;
2343 min_key.offset = (u64)-1;
2344 btrfs_release_path(root, path);
2345 ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0);
2346 if (ret < 0) {
2347 btrfs_release_path(root, path);
2348 return ret;
2349 }
2350 ret = btrfs_previous_item(root, path, inode->i_ino, key_type);
2351
2352 /* if ret == 0 there are items for this type,
2353 * create a range to tell us the last key of this type.
2354 * otherwise, there are no items in this directory after
2355 * *min_offset, and we create a range to indicate that.
2356 */
2357 if (ret == 0) {
2358 struct btrfs_key tmp;
2359 btrfs_item_key_to_cpu(path->nodes[0], &tmp,
2360 path->slots[0]);
Chris Masond3977122009-01-05 21:25:51 -05002361 if (key_type == tmp.type)
Chris Masone02119d2008-09-05 16:13:11 -04002362 first_offset = max(min_offset, tmp.offset) + 1;
Chris Masone02119d2008-09-05 16:13:11 -04002363 }
2364 goto done;
2365 }
2366
2367 /* go backward to find any previous key */
2368 ret = btrfs_previous_item(root, path, inode->i_ino, key_type);
2369 if (ret == 0) {
2370 struct btrfs_key tmp;
2371 btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]);
2372 if (key_type == tmp.type) {
2373 first_offset = tmp.offset;
2374 ret = overwrite_item(trans, log, dst_path,
2375 path->nodes[0], path->slots[0],
2376 &tmp);
2377 }
2378 }
2379 btrfs_release_path(root, path);
2380
2381 /* find the first key from this transaction again */
2382 ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0);
2383 if (ret != 0) {
2384 WARN_ON(1);
2385 goto done;
2386 }
2387
2388 /*
2389 * we have a block from this transaction, log every item in it
2390 * from our directory
2391 */
Chris Masond3977122009-01-05 21:25:51 -05002392 while (1) {
Chris Masone02119d2008-09-05 16:13:11 -04002393 struct btrfs_key tmp;
2394 src = path->nodes[0];
2395 nritems = btrfs_header_nritems(src);
2396 for (i = path->slots[0]; i < nritems; i++) {
2397 btrfs_item_key_to_cpu(src, &min_key, i);
2398
2399 if (min_key.objectid != inode->i_ino ||
2400 min_key.type != key_type)
2401 goto done;
2402 ret = overwrite_item(trans, log, dst_path, src, i,
2403 &min_key);
2404 BUG_ON(ret);
2405 }
2406 path->slots[0] = nritems;
2407
2408 /*
2409 * look ahead to the next item and see if it is also
2410 * from this directory and from this transaction
2411 */
2412 ret = btrfs_next_leaf(root, path);
2413 if (ret == 1) {
2414 last_offset = (u64)-1;
2415 goto done;
2416 }
2417 btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]);
2418 if (tmp.objectid != inode->i_ino || tmp.type != key_type) {
2419 last_offset = (u64)-1;
2420 goto done;
2421 }
2422 if (btrfs_header_generation(path->nodes[0]) != trans->transid) {
2423 ret = overwrite_item(trans, log, dst_path,
2424 path->nodes[0], path->slots[0],
2425 &tmp);
2426
2427 BUG_ON(ret);
2428 last_offset = tmp.offset;
2429 goto done;
2430 }
2431 }
2432done:
2433 *last_offset_ret = last_offset;
2434 btrfs_release_path(root, path);
2435 btrfs_release_path(log, dst_path);
2436
2437 /* insert the log range keys to indicate where the log is valid */
2438 ret = insert_dir_log_key(trans, log, path, key_type, inode->i_ino,
2439 first_offset, last_offset);
2440 BUG_ON(ret);
2441 return 0;
2442}
2443
2444/*
2445 * logging directories is very similar to logging inodes, We find all the items
2446 * from the current transaction and write them to the log.
2447 *
2448 * The recovery code scans the directory in the subvolume, and if it finds a
2449 * key in the range logged that is not present in the log tree, then it means
2450 * that dir entry was unlinked during the transaction.
2451 *
2452 * In order for that scan to work, we must include one key smaller than
2453 * the smallest logged by this transaction and one key larger than the largest
2454 * key logged by this transaction.
2455 */
2456static noinline int log_directory_changes(struct btrfs_trans_handle *trans,
2457 struct btrfs_root *root, struct inode *inode,
2458 struct btrfs_path *path,
2459 struct btrfs_path *dst_path)
2460{
2461 u64 min_key;
2462 u64 max_key;
2463 int ret;
2464 int key_type = BTRFS_DIR_ITEM_KEY;
2465
2466again:
2467 min_key = 0;
2468 max_key = 0;
Chris Masond3977122009-01-05 21:25:51 -05002469 while (1) {
Chris Masone02119d2008-09-05 16:13:11 -04002470 ret = log_dir_items(trans, root, inode, path,
2471 dst_path, key_type, min_key,
2472 &max_key);
2473 BUG_ON(ret);
2474 if (max_key == (u64)-1)
2475 break;
2476 min_key = max_key + 1;
2477 }
2478
2479 if (key_type == BTRFS_DIR_ITEM_KEY) {
2480 key_type = BTRFS_DIR_INDEX_KEY;
2481 goto again;
2482 }
2483 return 0;
2484}
2485
2486/*
2487 * a helper function to drop items from the log before we relog an
2488 * inode. max_key_type indicates the highest item type to remove.
2489 * This cannot be run for file data extents because it does not
2490 * free the extents they point to.
2491 */
2492static int drop_objectid_items(struct btrfs_trans_handle *trans,
2493 struct btrfs_root *log,
2494 struct btrfs_path *path,
2495 u64 objectid, int max_key_type)
2496{
2497 int ret;
2498 struct btrfs_key key;
2499 struct btrfs_key found_key;
2500
2501 key.objectid = objectid;
2502 key.type = max_key_type;
2503 key.offset = (u64)-1;
2504
Chris Masond3977122009-01-05 21:25:51 -05002505 while (1) {
Chris Masone02119d2008-09-05 16:13:11 -04002506 ret = btrfs_search_slot(trans, log, &key, path, -1, 1);
2507
2508 if (ret != 1)
2509 break;
2510
2511 if (path->slots[0] == 0)
2512 break;
2513
2514 path->slots[0]--;
2515 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2516 path->slots[0]);
2517
2518 if (found_key.objectid != objectid)
2519 break;
2520
2521 ret = btrfs_del_item(trans, log, path);
2522 BUG_ON(ret);
2523 btrfs_release_path(log, path);
2524 }
2525 btrfs_release_path(log, path);
2526 return 0;
2527}
2528
Chris Mason31ff1cd2008-09-11 16:17:57 -04002529static noinline int copy_items(struct btrfs_trans_handle *trans,
2530 struct btrfs_root *log,
2531 struct btrfs_path *dst_path,
2532 struct extent_buffer *src,
2533 int start_slot, int nr, int inode_only)
2534{
2535 unsigned long src_offset;
2536 unsigned long dst_offset;
2537 struct btrfs_file_extent_item *extent;
2538 struct btrfs_inode_item *inode_item;
2539 int ret;
2540 struct btrfs_key *ins_keys;
2541 u32 *ins_sizes;
2542 char *ins_data;
2543 int i;
Chris Masond20f7042008-12-08 16:58:54 -05002544 struct list_head ordered_sums;
2545
2546 INIT_LIST_HEAD(&ordered_sums);
Chris Mason31ff1cd2008-09-11 16:17:57 -04002547
2548 ins_data = kmalloc(nr * sizeof(struct btrfs_key) +
2549 nr * sizeof(u32), GFP_NOFS);
2550 ins_sizes = (u32 *)ins_data;
2551 ins_keys = (struct btrfs_key *)(ins_data + nr * sizeof(u32));
2552
2553 for (i = 0; i < nr; i++) {
2554 ins_sizes[i] = btrfs_item_size_nr(src, i + start_slot);
2555 btrfs_item_key_to_cpu(src, ins_keys + i, i + start_slot);
2556 }
2557 ret = btrfs_insert_empty_items(trans, log, dst_path,
2558 ins_keys, ins_sizes, nr);
2559 BUG_ON(ret);
2560
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002561 for (i = 0; i < nr; i++, dst_path->slots[0]++) {
Chris Mason31ff1cd2008-09-11 16:17:57 -04002562 dst_offset = btrfs_item_ptr_offset(dst_path->nodes[0],
2563 dst_path->slots[0]);
2564
2565 src_offset = btrfs_item_ptr_offset(src, start_slot + i);
2566
2567 copy_extent_buffer(dst_path->nodes[0], src, dst_offset,
2568 src_offset, ins_sizes[i]);
2569
2570 if (inode_only == LOG_INODE_EXISTS &&
2571 ins_keys[i].type == BTRFS_INODE_ITEM_KEY) {
2572 inode_item = btrfs_item_ptr(dst_path->nodes[0],
2573 dst_path->slots[0],
2574 struct btrfs_inode_item);
2575 btrfs_set_inode_size(dst_path->nodes[0], inode_item, 0);
2576
2577 /* set the generation to zero so the recover code
2578 * can tell the difference between an logging
2579 * just to say 'this inode exists' and a logging
2580 * to say 'update this inode with these values'
2581 */
2582 btrfs_set_inode_generation(dst_path->nodes[0],
2583 inode_item, 0);
2584 }
2585 /* take a reference on file data extents so that truncates
2586 * or deletes of this inode don't have to relog the inode
2587 * again
2588 */
2589 if (btrfs_key_type(ins_keys + i) == BTRFS_EXTENT_DATA_KEY) {
2590 int found_type;
2591 extent = btrfs_item_ptr(src, start_slot + i,
2592 struct btrfs_file_extent_item);
2593
2594 found_type = btrfs_file_extent_type(src, extent);
Yan Zhengd899e052008-10-30 14:25:28 -04002595 if (found_type == BTRFS_FILE_EXTENT_REG ||
2596 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002597 u64 ds, dl, cs, cl;
2598 ds = btrfs_file_extent_disk_bytenr(src,
2599 extent);
2600 /* ds == 0 is a hole */
2601 if (ds == 0)
2602 continue;
2603
2604 dl = btrfs_file_extent_disk_num_bytes(src,
2605 extent);
2606 cs = btrfs_file_extent_offset(src, extent);
2607 cl = btrfs_file_extent_num_bytes(src,
2608 extent);;
Chris Mason580afd72008-12-08 19:15:39 -05002609 if (btrfs_file_extent_compression(src,
2610 extent)) {
2611 cs = 0;
2612 cl = dl;
2613 }
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002614
2615 ret = btrfs_lookup_csums_range(
2616 log->fs_info->csum_root,
2617 ds + cs, ds + cs + cl - 1,
2618 &ordered_sums);
2619 BUG_ON(ret);
Chris Mason31ff1cd2008-09-11 16:17:57 -04002620 }
2621 }
Chris Mason31ff1cd2008-09-11 16:17:57 -04002622 }
2623
2624 btrfs_mark_buffer_dirty(dst_path->nodes[0]);
2625 btrfs_release_path(log, dst_path);
2626 kfree(ins_data);
Chris Masond20f7042008-12-08 16:58:54 -05002627
2628 /*
2629 * we have to do this after the loop above to avoid changing the
2630 * log tree while trying to change the log tree.
2631 */
Chris Masond3977122009-01-05 21:25:51 -05002632 while (!list_empty(&ordered_sums)) {
Chris Masond20f7042008-12-08 16:58:54 -05002633 struct btrfs_ordered_sum *sums = list_entry(ordered_sums.next,
2634 struct btrfs_ordered_sum,
2635 list);
2636 ret = btrfs_csum_file_blocks(trans, log, sums);
2637 BUG_ON(ret);
2638 list_del(&sums->list);
2639 kfree(sums);
2640 }
Chris Mason31ff1cd2008-09-11 16:17:57 -04002641 return 0;
2642}
2643
Chris Masone02119d2008-09-05 16:13:11 -04002644/* log a single inode in the tree log.
2645 * At least one parent directory for this inode must exist in the tree
2646 * or be logged already.
2647 *
2648 * Any items from this inode changed by the current transaction are copied
2649 * to the log tree. An extra reference is taken on any extents in this
2650 * file, allowing us to avoid a whole pile of corner cases around logging
2651 * blocks that have been removed from the tree.
2652 *
2653 * See LOG_INODE_ALL and related defines for a description of what inode_only
2654 * does.
2655 *
2656 * This handles both files and directories.
2657 */
Chris Mason12fcfd22009-03-24 10:24:20 -04002658static int btrfs_log_inode(struct btrfs_trans_handle *trans,
Chris Masone02119d2008-09-05 16:13:11 -04002659 struct btrfs_root *root, struct inode *inode,
2660 int inode_only)
2661{
2662 struct btrfs_path *path;
2663 struct btrfs_path *dst_path;
2664 struct btrfs_key min_key;
2665 struct btrfs_key max_key;
2666 struct btrfs_root *log = root->log_root;
Chris Mason31ff1cd2008-09-11 16:17:57 -04002667 struct extent_buffer *src = NULL;
Chris Masone02119d2008-09-05 16:13:11 -04002668 u32 size;
2669 int ret;
Chris Mason3a5f1d42008-09-11 15:53:37 -04002670 int nritems;
Chris Mason31ff1cd2008-09-11 16:17:57 -04002671 int ins_start_slot = 0;
2672 int ins_nr;
Chris Masone02119d2008-09-05 16:13:11 -04002673
2674 log = root->log_root;
2675
2676 path = btrfs_alloc_path();
2677 dst_path = btrfs_alloc_path();
2678
2679 min_key.objectid = inode->i_ino;
2680 min_key.type = BTRFS_INODE_ITEM_KEY;
2681 min_key.offset = 0;
2682
2683 max_key.objectid = inode->i_ino;
Chris Mason12fcfd22009-03-24 10:24:20 -04002684
2685 /* today the code can only do partial logging of directories */
2686 if (!S_ISDIR(inode->i_mode))
2687 inode_only = LOG_INODE_ALL;
2688
Chris Masone02119d2008-09-05 16:13:11 -04002689 if (inode_only == LOG_INODE_EXISTS || S_ISDIR(inode->i_mode))
2690 max_key.type = BTRFS_XATTR_ITEM_KEY;
2691 else
2692 max_key.type = (u8)-1;
2693 max_key.offset = (u64)-1;
2694
Chris Masone02119d2008-09-05 16:13:11 -04002695 mutex_lock(&BTRFS_I(inode)->log_mutex);
2696
2697 /*
2698 * a brute force approach to making sure we get the most uptodate
2699 * copies of everything.
2700 */
2701 if (S_ISDIR(inode->i_mode)) {
2702 int max_key_type = BTRFS_DIR_LOG_INDEX_KEY;
2703
2704 if (inode_only == LOG_INODE_EXISTS)
2705 max_key_type = BTRFS_XATTR_ITEM_KEY;
2706 ret = drop_objectid_items(trans, log, path,
2707 inode->i_ino, max_key_type);
2708 } else {
2709 ret = btrfs_truncate_inode_items(trans, log, inode, 0, 0);
2710 }
2711 BUG_ON(ret);
2712 path->keep_locks = 1;
2713
Chris Masond3977122009-01-05 21:25:51 -05002714 while (1) {
Chris Mason31ff1cd2008-09-11 16:17:57 -04002715 ins_nr = 0;
Chris Masone02119d2008-09-05 16:13:11 -04002716 ret = btrfs_search_forward(root, &min_key, &max_key,
2717 path, 0, trans->transid);
2718 if (ret != 0)
2719 break;
Chris Mason3a5f1d42008-09-11 15:53:37 -04002720again:
Chris Mason31ff1cd2008-09-11 16:17:57 -04002721 /* note, ins_nr might be > 0 here, cleanup outside the loop */
Chris Masone02119d2008-09-05 16:13:11 -04002722 if (min_key.objectid != inode->i_ino)
2723 break;
2724 if (min_key.type > max_key.type)
2725 break;
Chris Mason31ff1cd2008-09-11 16:17:57 -04002726
Chris Masone02119d2008-09-05 16:13:11 -04002727 src = path->nodes[0];
2728 size = btrfs_item_size_nr(src, path->slots[0]);
Chris Mason31ff1cd2008-09-11 16:17:57 -04002729 if (ins_nr && ins_start_slot + ins_nr == path->slots[0]) {
2730 ins_nr++;
2731 goto next_slot;
2732 } else if (!ins_nr) {
2733 ins_start_slot = path->slots[0];
2734 ins_nr = 1;
2735 goto next_slot;
Chris Masone02119d2008-09-05 16:13:11 -04002736 }
2737
Chris Mason31ff1cd2008-09-11 16:17:57 -04002738 ret = copy_items(trans, log, dst_path, src, ins_start_slot,
2739 ins_nr, inode_only);
2740 BUG_ON(ret);
2741 ins_nr = 1;
2742 ins_start_slot = path->slots[0];
2743next_slot:
Chris Masone02119d2008-09-05 16:13:11 -04002744
Chris Mason3a5f1d42008-09-11 15:53:37 -04002745 nritems = btrfs_header_nritems(path->nodes[0]);
2746 path->slots[0]++;
2747 if (path->slots[0] < nritems) {
2748 btrfs_item_key_to_cpu(path->nodes[0], &min_key,
2749 path->slots[0]);
2750 goto again;
2751 }
Chris Mason31ff1cd2008-09-11 16:17:57 -04002752 if (ins_nr) {
2753 ret = copy_items(trans, log, dst_path, src,
2754 ins_start_slot,
2755 ins_nr, inode_only);
2756 BUG_ON(ret);
2757 ins_nr = 0;
2758 }
Chris Mason3a5f1d42008-09-11 15:53:37 -04002759 btrfs_release_path(root, path);
2760
Chris Masone02119d2008-09-05 16:13:11 -04002761 if (min_key.offset < (u64)-1)
2762 min_key.offset++;
2763 else if (min_key.type < (u8)-1)
2764 min_key.type++;
2765 else if (min_key.objectid < (u64)-1)
2766 min_key.objectid++;
2767 else
2768 break;
2769 }
Chris Mason31ff1cd2008-09-11 16:17:57 -04002770 if (ins_nr) {
2771 ret = copy_items(trans, log, dst_path, src,
2772 ins_start_slot,
2773 ins_nr, inode_only);
2774 BUG_ON(ret);
2775 ins_nr = 0;
2776 }
2777 WARN_ON(ins_nr);
Chris Mason9623f9a2008-09-11 17:42:42 -04002778 if (inode_only == LOG_INODE_ALL && S_ISDIR(inode->i_mode)) {
Chris Masone02119d2008-09-05 16:13:11 -04002779 btrfs_release_path(root, path);
2780 btrfs_release_path(log, dst_path);
2781 ret = log_directory_changes(trans, root, inode, path, dst_path);
2782 BUG_ON(ret);
2783 }
Chris Mason3a5f1d42008-09-11 15:53:37 -04002784 BTRFS_I(inode)->logged_trans = trans->transid;
Chris Masone02119d2008-09-05 16:13:11 -04002785 mutex_unlock(&BTRFS_I(inode)->log_mutex);
2786
2787 btrfs_free_path(path);
2788 btrfs_free_path(dst_path);
Chris Masone02119d2008-09-05 16:13:11 -04002789 return 0;
2790}
2791
Chris Mason12fcfd22009-03-24 10:24:20 -04002792/*
2793 * follow the dentry parent pointers up the chain and see if any
2794 * of the directories in it require a full commit before they can
2795 * be logged. Returns zero if nothing special needs to be done or 1 if
2796 * a full commit is required.
2797 */
2798static noinline int check_parent_dirs_for_sync(struct btrfs_trans_handle *trans,
2799 struct inode *inode,
2800 struct dentry *parent,
2801 struct super_block *sb,
2802 u64 last_committed)
Chris Masone02119d2008-09-05 16:13:11 -04002803{
Chris Mason12fcfd22009-03-24 10:24:20 -04002804 int ret = 0;
2805 struct btrfs_root *root;
Chris Masone02119d2008-09-05 16:13:11 -04002806
Chris Masonaf4176b2009-03-24 10:24:31 -04002807 /*
2808 * for regular files, if its inode is already on disk, we don't
2809 * have to worry about the parents at all. This is because
2810 * we can use the last_unlink_trans field to record renames
2811 * and other fun in this file.
2812 */
2813 if (S_ISREG(inode->i_mode) &&
2814 BTRFS_I(inode)->generation <= last_committed &&
2815 BTRFS_I(inode)->last_unlink_trans <= last_committed)
2816 goto out;
2817
Chris Mason12fcfd22009-03-24 10:24:20 -04002818 if (!S_ISDIR(inode->i_mode)) {
2819 if (!parent || !parent->d_inode || sb != parent->d_inode->i_sb)
2820 goto out;
2821 inode = parent->d_inode;
2822 }
2823
2824 while (1) {
2825 BTRFS_I(inode)->logged_trans = trans->transid;
2826 smp_mb();
2827
2828 if (BTRFS_I(inode)->last_unlink_trans > last_committed) {
2829 root = BTRFS_I(inode)->root;
2830
2831 /*
2832 * make sure any commits to the log are forced
2833 * to be full commits
2834 */
2835 root->fs_info->last_trans_log_full_commit =
2836 trans->transid;
2837 ret = 1;
2838 break;
2839 }
2840
2841 if (!parent || !parent->d_inode || sb != parent->d_inode->i_sb)
2842 break;
2843
2844 if (parent == sb->s_root)
2845 break;
2846
2847 parent = parent->d_parent;
2848 inode = parent->d_inode;
2849
2850 }
2851out:
Chris Masone02119d2008-09-05 16:13:11 -04002852 return ret;
2853}
2854
2855/*
2856 * helper function around btrfs_log_inode to make sure newly created
2857 * parent directories also end up in the log. A minimal inode and backref
2858 * only logging is done of any parent directories that are older than
2859 * the last committed transaction
2860 */
Chris Mason12fcfd22009-03-24 10:24:20 -04002861int btrfs_log_inode_parent(struct btrfs_trans_handle *trans,
2862 struct btrfs_root *root, struct inode *inode,
2863 struct dentry *parent, int exists_only)
Chris Masone02119d2008-09-05 16:13:11 -04002864{
Chris Mason12fcfd22009-03-24 10:24:20 -04002865 int inode_only = exists_only ? LOG_INODE_EXISTS : LOG_INODE_ALL;
Chris Masone02119d2008-09-05 16:13:11 -04002866 struct super_block *sb;
Chris Mason12fcfd22009-03-24 10:24:20 -04002867 int ret = 0;
2868 u64 last_committed = root->fs_info->last_trans_committed;
2869
2870 sb = inode->i_sb;
2871
Sage Weil3a5e1402009-04-02 16:49:40 -04002872 if (btrfs_test_opt(root, NOTREELOG)) {
2873 ret = 1;
2874 goto end_no_trans;
2875 }
2876
Chris Mason12fcfd22009-03-24 10:24:20 -04002877 if (root->fs_info->last_trans_log_full_commit >
2878 root->fs_info->last_trans_committed) {
2879 ret = 1;
2880 goto end_no_trans;
2881 }
2882
2883 ret = check_parent_dirs_for_sync(trans, inode, parent,
2884 sb, last_committed);
2885 if (ret)
2886 goto end_no_trans;
Chris Masone02119d2008-09-05 16:13:11 -04002887
2888 start_log_trans(trans, root);
Chris Mason12fcfd22009-03-24 10:24:20 -04002889
2890 ret = btrfs_log_inode(trans, root, inode, inode_only);
2891 BUG_ON(ret);
Chris Mason12fcfd22009-03-24 10:24:20 -04002892
Chris Masonaf4176b2009-03-24 10:24:31 -04002893 /*
2894 * for regular files, if its inode is already on disk, we don't
2895 * have to worry about the parents at all. This is because
2896 * we can use the last_unlink_trans field to record renames
2897 * and other fun in this file.
2898 */
2899 if (S_ISREG(inode->i_mode) &&
2900 BTRFS_I(inode)->generation <= last_committed &&
2901 BTRFS_I(inode)->last_unlink_trans <= last_committed)
2902 goto no_parent;
2903
2904 inode_only = LOG_INODE_EXISTS;
Chris Masond3977122009-01-05 21:25:51 -05002905 while (1) {
Chris Mason12fcfd22009-03-24 10:24:20 -04002906 if (!parent || !parent->d_inode || sb != parent->d_inode->i_sb)
Chris Masone02119d2008-09-05 16:13:11 -04002907 break;
2908
Chris Mason12fcfd22009-03-24 10:24:20 -04002909 inode = parent->d_inode;
2910 if (BTRFS_I(inode)->generation >
2911 root->fs_info->last_trans_committed) {
2912 ret = btrfs_log_inode(trans, root, inode, inode_only);
2913 BUG_ON(ret);
2914 }
2915 if (parent == sb->s_root)
Chris Masone02119d2008-09-05 16:13:11 -04002916 break;
Chris Mason12fcfd22009-03-24 10:24:20 -04002917
2918 parent = parent->d_parent;
Chris Masone02119d2008-09-05 16:13:11 -04002919 }
Chris Masonaf4176b2009-03-24 10:24:31 -04002920no_parent:
Chris Mason12fcfd22009-03-24 10:24:20 -04002921 ret = 0;
2922 btrfs_end_log_trans(root);
2923end_no_trans:
2924 return ret;
Chris Masone02119d2008-09-05 16:13:11 -04002925}
2926
2927/*
2928 * it is not safe to log dentry if the chunk root has added new
2929 * chunks. This returns 0 if the dentry was logged, and 1 otherwise.
2930 * If this returns 1, you must commit the transaction to safely get your
2931 * data on disk.
2932 */
2933int btrfs_log_dentry_safe(struct btrfs_trans_handle *trans,
2934 struct btrfs_root *root, struct dentry *dentry)
2935{
Chris Mason12fcfd22009-03-24 10:24:20 -04002936 return btrfs_log_inode_parent(trans, root, dentry->d_inode,
2937 dentry->d_parent, 0);
Chris Masone02119d2008-09-05 16:13:11 -04002938}
2939
2940/*
2941 * should be called during mount to recover any replay any log trees
2942 * from the FS
2943 */
2944int btrfs_recover_log_trees(struct btrfs_root *log_root_tree)
2945{
2946 int ret;
2947 struct btrfs_path *path;
2948 struct btrfs_trans_handle *trans;
2949 struct btrfs_key key;
2950 struct btrfs_key found_key;
2951 struct btrfs_key tmp_key;
2952 struct btrfs_root *log;
2953 struct btrfs_fs_info *fs_info = log_root_tree->fs_info;
Chris Mason8d5bf1c2008-09-11 15:51:21 -04002954 u64 highest_inode;
Chris Masone02119d2008-09-05 16:13:11 -04002955 struct walk_control wc = {
2956 .process_func = process_one_buffer,
2957 .stage = 0,
2958 };
2959
2960 fs_info->log_root_recovering = 1;
2961 path = btrfs_alloc_path();
2962 BUG_ON(!path);
2963
2964 trans = btrfs_start_transaction(fs_info->tree_root, 1);
2965
2966 wc.trans = trans;
2967 wc.pin = 1;
2968
2969 walk_log_tree(trans, log_root_tree, &wc);
2970
2971again:
2972 key.objectid = BTRFS_TREE_LOG_OBJECTID;
2973 key.offset = (u64)-1;
2974 btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
2975
Chris Masond3977122009-01-05 21:25:51 -05002976 while (1) {
Chris Masone02119d2008-09-05 16:13:11 -04002977 ret = btrfs_search_slot(NULL, log_root_tree, &key, path, 0, 0);
2978 if (ret < 0)
2979 break;
2980 if (ret > 0) {
2981 if (path->slots[0] == 0)
2982 break;
2983 path->slots[0]--;
2984 }
2985 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2986 path->slots[0]);
2987 btrfs_release_path(log_root_tree, path);
2988 if (found_key.objectid != BTRFS_TREE_LOG_OBJECTID)
2989 break;
2990
2991 log = btrfs_read_fs_root_no_radix(log_root_tree,
2992 &found_key);
2993 BUG_ON(!log);
2994
2995
2996 tmp_key.objectid = found_key.offset;
2997 tmp_key.type = BTRFS_ROOT_ITEM_KEY;
2998 tmp_key.offset = (u64)-1;
2999
3000 wc.replay_dest = btrfs_read_fs_root_no_name(fs_info, &tmp_key);
Chris Masone02119d2008-09-05 16:13:11 -04003001 BUG_ON(!wc.replay_dest);
3002
Yan Zheng07d400a2009-01-06 11:42:00 -05003003 wc.replay_dest->log_root = log;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003004 btrfs_record_root_in_trans(trans, wc.replay_dest);
Chris Masone02119d2008-09-05 16:13:11 -04003005 ret = walk_log_tree(trans, log, &wc);
3006 BUG_ON(ret);
3007
3008 if (wc.stage == LOG_WALK_REPLAY_ALL) {
3009 ret = fixup_inode_link_counts(trans, wc.replay_dest,
3010 path);
3011 BUG_ON(ret);
3012 }
Chris Mason8d5bf1c2008-09-11 15:51:21 -04003013 ret = btrfs_find_highest_inode(wc.replay_dest, &highest_inode);
3014 if (ret == 0) {
3015 wc.replay_dest->highest_inode = highest_inode;
3016 wc.replay_dest->last_inode_alloc = highest_inode;
3017 }
Chris Masone02119d2008-09-05 16:13:11 -04003018
3019 key.offset = found_key.offset - 1;
Yan Zheng07d400a2009-01-06 11:42:00 -05003020 wc.replay_dest->log_root = NULL;
Chris Masone02119d2008-09-05 16:13:11 -04003021 free_extent_buffer(log->node);
Chris Masonb263c2c2009-06-11 11:24:47 -04003022 free_extent_buffer(log->commit_root);
Chris Masone02119d2008-09-05 16:13:11 -04003023 kfree(log);
3024
3025 if (found_key.offset == 0)
3026 break;
3027 }
3028 btrfs_release_path(log_root_tree, path);
3029
3030 /* step one is to pin it all, step two is to replay just inodes */
3031 if (wc.pin) {
3032 wc.pin = 0;
3033 wc.process_func = replay_one_buffer;
3034 wc.stage = LOG_WALK_REPLAY_INODES;
3035 goto again;
3036 }
3037 /* step three is to replay everything */
3038 if (wc.stage < LOG_WALK_REPLAY_ALL) {
3039 wc.stage++;
3040 goto again;
3041 }
3042
3043 btrfs_free_path(path);
3044
3045 free_extent_buffer(log_root_tree->node);
3046 log_root_tree->log_root = NULL;
3047 fs_info->log_root_recovering = 0;
3048
3049 /* step 4: commit the transaction, which also unpins the blocks */
3050 btrfs_commit_transaction(trans, fs_info->tree_root);
3051
3052 kfree(log_root_tree);
3053 return 0;
3054}
Chris Mason12fcfd22009-03-24 10:24:20 -04003055
3056/*
3057 * there are some corner cases where we want to force a full
3058 * commit instead of allowing a directory to be logged.
3059 *
3060 * They revolve around files there were unlinked from the directory, and
3061 * this function updates the parent directory so that a full commit is
3062 * properly done if it is fsync'd later after the unlinks are done.
3063 */
3064void btrfs_record_unlink_dir(struct btrfs_trans_handle *trans,
3065 struct inode *dir, struct inode *inode,
3066 int for_rename)
3067{
3068 /*
Chris Masonaf4176b2009-03-24 10:24:31 -04003069 * when we're logging a file, if it hasn't been renamed
3070 * or unlinked, and its inode is fully committed on disk,
3071 * we don't have to worry about walking up the directory chain
3072 * to log its parents.
3073 *
3074 * So, we use the last_unlink_trans field to put this transid
3075 * into the file. When the file is logged we check it and
3076 * don't log the parents if the file is fully on disk.
3077 */
3078 if (S_ISREG(inode->i_mode))
3079 BTRFS_I(inode)->last_unlink_trans = trans->transid;
3080
3081 /*
Chris Mason12fcfd22009-03-24 10:24:20 -04003082 * if this directory was already logged any new
3083 * names for this file/dir will get recorded
3084 */
3085 smp_mb();
3086 if (BTRFS_I(dir)->logged_trans == trans->transid)
3087 return;
3088
3089 /*
3090 * if the inode we're about to unlink was logged,
3091 * the log will be properly updated for any new names
3092 */
3093 if (BTRFS_I(inode)->logged_trans == trans->transid)
3094 return;
3095
3096 /*
3097 * when renaming files across directories, if the directory
3098 * there we're unlinking from gets fsync'd later on, there's
3099 * no way to find the destination directory later and fsync it
3100 * properly. So, we have to be conservative and force commits
3101 * so the new name gets discovered.
3102 */
3103 if (for_rename)
3104 goto record;
3105
3106 /* we can safely do the unlink without any special recording */
3107 return;
3108
3109record:
3110 BTRFS_I(dir)->last_unlink_trans = trans->transid;
3111}
3112
3113/*
3114 * Call this after adding a new name for a file and it will properly
3115 * update the log to reflect the new name.
3116 *
3117 * It will return zero if all goes well, and it will return 1 if a
3118 * full transaction commit is required.
3119 */
3120int btrfs_log_new_name(struct btrfs_trans_handle *trans,
3121 struct inode *inode, struct inode *old_dir,
3122 struct dentry *parent)
3123{
3124 struct btrfs_root * root = BTRFS_I(inode)->root;
3125
3126 /*
Chris Masonaf4176b2009-03-24 10:24:31 -04003127 * this will force the logging code to walk the dentry chain
3128 * up for the file
3129 */
3130 if (S_ISREG(inode->i_mode))
3131 BTRFS_I(inode)->last_unlink_trans = trans->transid;
3132
3133 /*
Chris Mason12fcfd22009-03-24 10:24:20 -04003134 * if this inode hasn't been logged and directory we're renaming it
3135 * from hasn't been logged, we don't need to log it
3136 */
3137 if (BTRFS_I(inode)->logged_trans <=
3138 root->fs_info->last_trans_committed &&
3139 (!old_dir || BTRFS_I(old_dir)->logged_trans <=
3140 root->fs_info->last_trans_committed))
3141 return 0;
3142
3143 return btrfs_log_inode_parent(trans, root, inode, parent, 1);
3144}
3145