blob: fb102a9aee9cc1ee3213d04bc1bae6fec6644f2d [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>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090020#include <linux/slab.h>
Chris Masone02119d2008-09-05 16:13:11 -040021#include "ctree.h"
22#include "transaction.h"
23#include "disk-io.h"
24#include "locking.h"
25#include "print-tree.h"
26#include "compat.h"
Christoph Hellwigb2950862008-12-02 09:54:17 -050027#include "tree-log.h"
Chris Masone02119d2008-09-05 16:13:11 -040028
29/* magic values for the inode_only field in btrfs_log_inode:
30 *
31 * LOG_INODE_ALL means to log everything
32 * LOG_INODE_EXISTS means to log just enough to recreate the inode
33 * during log replay
34 */
35#define LOG_INODE_ALL 0
36#define LOG_INODE_EXISTS 1
37
38/*
Chris Mason12fcfd22009-03-24 10:24:20 -040039 * directory trouble cases
40 *
41 * 1) on rename or unlink, if the inode being unlinked isn't in the fsync
42 * log, we must force a full commit before doing an fsync of the directory
43 * where the unlink was done.
44 * ---> record transid of last unlink/rename per directory
45 *
46 * mkdir foo/some_dir
47 * normal commit
48 * rename foo/some_dir foo2/some_dir
49 * mkdir foo/some_dir
50 * fsync foo/some_dir/some_file
51 *
52 * The fsync above will unlink the original some_dir without recording
53 * it in its new location (foo2). After a crash, some_dir will be gone
54 * unless the fsync of some_file forces a full commit
55 *
56 * 2) we must log any new names for any file or dir that is in the fsync
57 * log. ---> check inode while renaming/linking.
58 *
59 * 2a) we must log any new names for any file or dir during rename
60 * when the directory they are being removed from was logged.
61 * ---> check inode and old parent dir during rename
62 *
63 * 2a is actually the more important variant. With the extra logging
64 * a crash might unlink the old name without recreating the new one
65 *
66 * 3) after a crash, we must go through any directories with a link count
67 * of zero and redo the rm -rf
68 *
69 * mkdir f1/foo
70 * normal commit
71 * rm -rf f1/foo
72 * fsync(f1)
73 *
74 * The directory f1 was fully removed from the FS, but fsync was never
75 * called on f1, only its parent dir. After a crash the rm -rf must
76 * be replayed. This must be able to recurse down the entire
77 * directory tree. The inode link count fixup code takes care of the
78 * ugly details.
79 */
80
81/*
Chris Masone02119d2008-09-05 16:13:11 -040082 * stages for the tree walking. The first
83 * stage (0) is to only pin down the blocks we find
84 * the second stage (1) is to make sure that all the inodes
85 * we find in the log are created in the subvolume.
86 *
87 * The last stage is to deal with directories and links and extents
88 * and all the other fun semantics
89 */
90#define LOG_WALK_PIN_ONLY 0
91#define LOG_WALK_REPLAY_INODES 1
92#define LOG_WALK_REPLAY_ALL 2
93
Chris Mason12fcfd22009-03-24 10:24:20 -040094static int btrfs_log_inode(struct btrfs_trans_handle *trans,
Chris Masone02119d2008-09-05 16:13:11 -040095 struct btrfs_root *root, struct inode *inode,
96 int inode_only);
Yan Zhengec051c02009-01-05 15:43:42 -050097static int link_to_fixup_dir(struct btrfs_trans_handle *trans,
98 struct btrfs_root *root,
99 struct btrfs_path *path, u64 objectid);
Chris Mason12fcfd22009-03-24 10:24:20 -0400100static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans,
101 struct btrfs_root *root,
102 struct btrfs_root *log,
103 struct btrfs_path *path,
104 u64 dirid, int del_all);
Chris Masone02119d2008-09-05 16:13:11 -0400105
106/*
107 * tree logging is a special write ahead log used to make sure that
108 * fsyncs and O_SYNCs can happen without doing full tree commits.
109 *
110 * Full tree commits are expensive because they require commonly
111 * modified blocks to be recowed, creating many dirty pages in the
112 * extent tree an 4x-6x higher write load than ext3.
113 *
114 * Instead of doing a tree commit on every fsync, we use the
115 * key ranges and transaction ids to find items for a given file or directory
116 * that have changed in this transaction. Those items are copied into
117 * a special tree (one per subvolume root), that tree is written to disk
118 * and then the fsync is considered complete.
119 *
120 * After a crash, items are copied out of the log-tree back into the
121 * subvolume tree. Any file data extents found are recorded in the extent
122 * allocation tree, and the log-tree freed.
123 *
124 * The log tree is read three times, once to pin down all the extents it is
125 * using in ram and once, once to create all the inodes logged in the tree
126 * and once to do all the other items.
127 */
128
129/*
Chris Masone02119d2008-09-05 16:13:11 -0400130 * start a sub transaction and setup the log tree
131 * this increments the log tree writer count to make the people
132 * syncing the tree wait for us to finish
133 */
134static int start_log_trans(struct btrfs_trans_handle *trans,
135 struct btrfs_root *root)
136{
137 int ret;
Yan, Zheng4a500fd2010-05-16 10:49:59 -0400138 int err = 0;
Yan Zheng7237f182009-01-21 12:54:03 -0500139
140 mutex_lock(&root->log_mutex);
141 if (root->log_root) {
Josef Bacikff782e02009-10-08 15:30:04 -0400142 if (!root->log_start_pid) {
143 root->log_start_pid = current->pid;
144 root->log_multiple_pids = false;
145 } else if (root->log_start_pid != current->pid) {
146 root->log_multiple_pids = true;
147 }
148
Yan Zheng7237f182009-01-21 12:54:03 -0500149 root->log_batch++;
150 atomic_inc(&root->log_writers);
151 mutex_unlock(&root->log_mutex);
152 return 0;
153 }
Josef Bacikff782e02009-10-08 15:30:04 -0400154 root->log_multiple_pids = false;
155 root->log_start_pid = current->pid;
Chris Masone02119d2008-09-05 16:13:11 -0400156 mutex_lock(&root->fs_info->tree_log_mutex);
157 if (!root->fs_info->log_root_tree) {
158 ret = btrfs_init_log_root_tree(trans, root->fs_info);
Yan, Zheng4a500fd2010-05-16 10:49:59 -0400159 if (ret)
160 err = ret;
Chris Masone02119d2008-09-05 16:13:11 -0400161 }
Yan, Zheng4a500fd2010-05-16 10:49:59 -0400162 if (err == 0 && !root->log_root) {
Chris Masone02119d2008-09-05 16:13:11 -0400163 ret = btrfs_add_log_tree(trans, root);
Yan, Zheng4a500fd2010-05-16 10:49:59 -0400164 if (ret)
165 err = ret;
Chris Masone02119d2008-09-05 16:13:11 -0400166 }
Chris Masone02119d2008-09-05 16:13:11 -0400167 mutex_unlock(&root->fs_info->tree_log_mutex);
Yan Zheng7237f182009-01-21 12:54:03 -0500168 root->log_batch++;
169 atomic_inc(&root->log_writers);
170 mutex_unlock(&root->log_mutex);
Yan, Zheng4a500fd2010-05-16 10:49:59 -0400171 return err;
Chris Masone02119d2008-09-05 16:13:11 -0400172}
173
174/*
175 * returns 0 if there was a log transaction running and we were able
176 * to join, or returns -ENOENT if there were not transactions
177 * in progress
178 */
179static int join_running_log_trans(struct btrfs_root *root)
180{
181 int ret = -ENOENT;
182
183 smp_mb();
184 if (!root->log_root)
185 return -ENOENT;
186
Yan Zheng7237f182009-01-21 12:54:03 -0500187 mutex_lock(&root->log_mutex);
Chris Masone02119d2008-09-05 16:13:11 -0400188 if (root->log_root) {
189 ret = 0;
Yan Zheng7237f182009-01-21 12:54:03 -0500190 atomic_inc(&root->log_writers);
Chris Masone02119d2008-09-05 16:13:11 -0400191 }
Yan Zheng7237f182009-01-21 12:54:03 -0500192 mutex_unlock(&root->log_mutex);
Chris Masone02119d2008-09-05 16:13:11 -0400193 return ret;
194}
195
196/*
Chris Mason12fcfd22009-03-24 10:24:20 -0400197 * This either makes the current running log transaction wait
198 * until you call btrfs_end_log_trans() or it makes any future
199 * log transactions wait until you call btrfs_end_log_trans()
200 */
201int btrfs_pin_log_trans(struct btrfs_root *root)
202{
203 int ret = -ENOENT;
204
205 mutex_lock(&root->log_mutex);
206 atomic_inc(&root->log_writers);
207 mutex_unlock(&root->log_mutex);
208 return ret;
209}
210
211/*
Chris Masone02119d2008-09-05 16:13:11 -0400212 * indicate we're done making changes to the log tree
213 * and wake up anyone waiting to do a sync
214 */
Chris Mason12fcfd22009-03-24 10:24:20 -0400215int btrfs_end_log_trans(struct btrfs_root *root)
Chris Masone02119d2008-09-05 16:13:11 -0400216{
Yan Zheng7237f182009-01-21 12:54:03 -0500217 if (atomic_dec_and_test(&root->log_writers)) {
218 smp_mb();
219 if (waitqueue_active(&root->log_writer_wait))
220 wake_up(&root->log_writer_wait);
221 }
Chris Masone02119d2008-09-05 16:13:11 -0400222 return 0;
223}
224
225
226/*
227 * the walk control struct is used to pass state down the chain when
228 * processing the log tree. The stage field tells us which part
229 * of the log tree processing we are currently doing. The others
230 * are state fields used for that specific part
231 */
232struct walk_control {
233 /* should we free the extent on disk when done? This is used
234 * at transaction commit time while freeing a log tree
235 */
236 int free;
237
238 /* should we write out the extent buffer? This is used
239 * while flushing the log tree to disk during a sync
240 */
241 int write;
242
243 /* should we wait for the extent buffer io to finish? Also used
244 * while flushing the log tree to disk for a sync
245 */
246 int wait;
247
248 /* pin only walk, we record which extents on disk belong to the
249 * log trees
250 */
251 int pin;
252
253 /* what stage of the replay code we're currently in */
254 int stage;
255
256 /* the root we are currently replaying */
257 struct btrfs_root *replay_dest;
258
259 /* the trans handle for the current replay */
260 struct btrfs_trans_handle *trans;
261
262 /* the function that gets used to process blocks we find in the
263 * tree. Note the extent_buffer might not be up to date when it is
264 * passed in, and it must be checked or read if you need the data
265 * inside it
266 */
267 int (*process_func)(struct btrfs_root *log, struct extent_buffer *eb,
268 struct walk_control *wc, u64 gen);
269};
270
271/*
272 * process_func used to pin down extents, write them or wait on them
273 */
274static int process_one_buffer(struct btrfs_root *log,
275 struct extent_buffer *eb,
276 struct walk_control *wc, u64 gen)
277{
Josef Bacik04018de2009-04-03 10:14:18 -0400278 if (wc->pin)
Yan Zheng11833d62009-09-11 16:11:19 -0400279 btrfs_pin_extent(log->fs_info->extent_root,
280 eb->start, eb->len, 0);
Chris Masone02119d2008-09-05 16:13:11 -0400281
282 if (btrfs_buffer_uptodate(eb, gen)) {
283 if (wc->write)
284 btrfs_write_tree_block(eb);
285 if (wc->wait)
286 btrfs_wait_tree_block_writeback(eb);
287 }
288 return 0;
289}
290
291/*
292 * Item overwrite used by replay and tree logging. eb, slot and key all refer
293 * to the src data we are copying out.
294 *
295 * root is the tree we are copying into, and path is a scratch
296 * path for use in this function (it should be released on entry and
297 * will be released on exit).
298 *
299 * If the key is already in the destination tree the existing item is
300 * overwritten. If the existing item isn't big enough, it is extended.
301 * If it is too large, it is truncated.
302 *
303 * If the key isn't in the destination yet, a new item is inserted.
304 */
305static noinline int overwrite_item(struct btrfs_trans_handle *trans,
306 struct btrfs_root *root,
307 struct btrfs_path *path,
308 struct extent_buffer *eb, int slot,
309 struct btrfs_key *key)
310{
311 int ret;
312 u32 item_size;
313 u64 saved_i_size = 0;
314 int save_old_i_size = 0;
315 unsigned long src_ptr;
316 unsigned long dst_ptr;
317 int overwrite_root = 0;
318
319 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
320 overwrite_root = 1;
321
322 item_size = btrfs_item_size_nr(eb, slot);
323 src_ptr = btrfs_item_ptr_offset(eb, slot);
324
325 /* look for the key in the destination tree */
326 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
327 if (ret == 0) {
328 char *src_copy;
329 char *dst_copy;
330 u32 dst_size = btrfs_item_size_nr(path->nodes[0],
331 path->slots[0]);
332 if (dst_size != item_size)
333 goto insert;
334
335 if (item_size == 0) {
336 btrfs_release_path(root, path);
337 return 0;
338 }
339 dst_copy = kmalloc(item_size, GFP_NOFS);
340 src_copy = kmalloc(item_size, GFP_NOFS);
341
342 read_extent_buffer(eb, src_copy, src_ptr, item_size);
343
344 dst_ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
345 read_extent_buffer(path->nodes[0], dst_copy, dst_ptr,
346 item_size);
347 ret = memcmp(dst_copy, src_copy, item_size);
348
349 kfree(dst_copy);
350 kfree(src_copy);
351 /*
352 * they have the same contents, just return, this saves
353 * us from cowing blocks in the destination tree and doing
354 * extra writes that may not have been done by a previous
355 * sync
356 */
357 if (ret == 0) {
358 btrfs_release_path(root, path);
359 return 0;
360 }
361
362 }
363insert:
364 btrfs_release_path(root, path);
365 /* try to insert the key into the destination tree */
366 ret = btrfs_insert_empty_item(trans, root, path,
367 key, item_size);
368
369 /* make sure any existing item is the correct size */
370 if (ret == -EEXIST) {
371 u32 found_size;
372 found_size = btrfs_item_size_nr(path->nodes[0],
373 path->slots[0]);
374 if (found_size > item_size) {
375 btrfs_truncate_item(trans, root, path, item_size, 1);
376 } else if (found_size < item_size) {
Yan Zheng87b29b22008-12-17 10:21:48 -0500377 ret = btrfs_extend_item(trans, root, path,
378 item_size - found_size);
Chris Masone02119d2008-09-05 16:13:11 -0400379 BUG_ON(ret);
380 }
381 } else if (ret) {
Yan, Zheng4a500fd2010-05-16 10:49:59 -0400382 return ret;
Chris Masone02119d2008-09-05 16:13:11 -0400383 }
384 dst_ptr = btrfs_item_ptr_offset(path->nodes[0],
385 path->slots[0]);
386
387 /* don't overwrite an existing inode if the generation number
388 * was logged as zero. This is done when the tree logging code
389 * is just logging an inode to make sure it exists after recovery.
390 *
391 * Also, don't overwrite i_size on directories during replay.
392 * log replay inserts and removes directory items based on the
393 * state of the tree found in the subvolume, and i_size is modified
394 * as it goes
395 */
396 if (key->type == BTRFS_INODE_ITEM_KEY && ret == -EEXIST) {
397 struct btrfs_inode_item *src_item;
398 struct btrfs_inode_item *dst_item;
399
400 src_item = (struct btrfs_inode_item *)src_ptr;
401 dst_item = (struct btrfs_inode_item *)dst_ptr;
402
403 if (btrfs_inode_generation(eb, src_item) == 0)
404 goto no_copy;
405
406 if (overwrite_root &&
407 S_ISDIR(btrfs_inode_mode(eb, src_item)) &&
408 S_ISDIR(btrfs_inode_mode(path->nodes[0], dst_item))) {
409 save_old_i_size = 1;
410 saved_i_size = btrfs_inode_size(path->nodes[0],
411 dst_item);
412 }
413 }
414
415 copy_extent_buffer(path->nodes[0], eb, dst_ptr,
416 src_ptr, item_size);
417
418 if (save_old_i_size) {
419 struct btrfs_inode_item *dst_item;
420 dst_item = (struct btrfs_inode_item *)dst_ptr;
421 btrfs_set_inode_size(path->nodes[0], dst_item, saved_i_size);
422 }
423
424 /* make sure the generation is filled in */
425 if (key->type == BTRFS_INODE_ITEM_KEY) {
426 struct btrfs_inode_item *dst_item;
427 dst_item = (struct btrfs_inode_item *)dst_ptr;
428 if (btrfs_inode_generation(path->nodes[0], dst_item) == 0) {
429 btrfs_set_inode_generation(path->nodes[0], dst_item,
430 trans->transid);
431 }
432 }
433no_copy:
434 btrfs_mark_buffer_dirty(path->nodes[0]);
435 btrfs_release_path(root, path);
436 return 0;
437}
438
439/*
440 * simple helper to read an inode off the disk from a given root
441 * This can only be called for subvolume roots and not for the log
442 */
443static noinline struct inode *read_one_inode(struct btrfs_root *root,
444 u64 objectid)
445{
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400446 struct btrfs_key key;
Chris Masone02119d2008-09-05 16:13:11 -0400447 struct inode *inode;
Chris Masone02119d2008-09-05 16:13:11 -0400448
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400449 key.objectid = objectid;
450 key.type = BTRFS_INODE_ITEM_KEY;
451 key.offset = 0;
Josef Bacik73f73412009-12-04 17:38:27 +0000452 inode = btrfs_iget(root->fs_info->sb, &key, root, NULL);
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400453 if (IS_ERR(inode)) {
454 inode = NULL;
455 } else if (is_bad_inode(inode)) {
Chris Masone02119d2008-09-05 16:13:11 -0400456 iput(inode);
457 inode = NULL;
458 }
459 return inode;
460}
461
462/* replays a single extent in 'eb' at 'slot' with 'key' into the
463 * subvolume 'root'. path is released on entry and should be released
464 * on exit.
465 *
466 * extents in the log tree have not been allocated out of the extent
467 * tree yet. So, this completes the allocation, taking a reference
468 * as required if the extent already exists or creating a new extent
469 * if it isn't in the extent allocation tree yet.
470 *
471 * The extent is inserted into the file, dropping any existing extents
472 * from the file that overlap the new one.
473 */
474static noinline int replay_one_extent(struct btrfs_trans_handle *trans,
475 struct btrfs_root *root,
476 struct btrfs_path *path,
477 struct extent_buffer *eb, int slot,
478 struct btrfs_key *key)
479{
480 int found_type;
481 u64 mask = root->sectorsize - 1;
482 u64 extent_end;
483 u64 alloc_hint;
484 u64 start = key->offset;
Yan Zheng07d400a2009-01-06 11:42:00 -0500485 u64 saved_nbytes;
Chris Masone02119d2008-09-05 16:13:11 -0400486 struct btrfs_file_extent_item *item;
487 struct inode *inode = NULL;
488 unsigned long size;
489 int ret = 0;
490
491 item = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
492 found_type = btrfs_file_extent_type(eb, item);
493
Yan Zhengd899e052008-10-30 14:25:28 -0400494 if (found_type == BTRFS_FILE_EXTENT_REG ||
495 found_type == BTRFS_FILE_EXTENT_PREALLOC)
Chris Masone02119d2008-09-05 16:13:11 -0400496 extent_end = start + btrfs_file_extent_num_bytes(eb, item);
497 else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
Chris Masonc8b97812008-10-29 14:49:59 -0400498 size = btrfs_file_extent_inline_len(eb, item);
Chris Masone02119d2008-09-05 16:13:11 -0400499 extent_end = (start + size + mask) & ~mask;
500 } else {
501 ret = 0;
502 goto out;
503 }
504
505 inode = read_one_inode(root, key->objectid);
506 if (!inode) {
507 ret = -EIO;
508 goto out;
509 }
510
511 /*
512 * first check to see if we already have this extent in the
513 * file. This must be done before the btrfs_drop_extents run
514 * so we don't try to drop this extent.
515 */
516 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
517 start, 0);
518
Yan Zhengd899e052008-10-30 14:25:28 -0400519 if (ret == 0 &&
520 (found_type == BTRFS_FILE_EXTENT_REG ||
521 found_type == BTRFS_FILE_EXTENT_PREALLOC)) {
Chris Masone02119d2008-09-05 16:13:11 -0400522 struct btrfs_file_extent_item cmp1;
523 struct btrfs_file_extent_item cmp2;
524 struct btrfs_file_extent_item *existing;
525 struct extent_buffer *leaf;
526
527 leaf = path->nodes[0];
528 existing = btrfs_item_ptr(leaf, path->slots[0],
529 struct btrfs_file_extent_item);
530
531 read_extent_buffer(eb, &cmp1, (unsigned long)item,
532 sizeof(cmp1));
533 read_extent_buffer(leaf, &cmp2, (unsigned long)existing,
534 sizeof(cmp2));
535
536 /*
537 * we already have a pointer to this exact extent,
538 * we don't have to do anything
539 */
540 if (memcmp(&cmp1, &cmp2, sizeof(cmp1)) == 0) {
541 btrfs_release_path(root, path);
542 goto out;
543 }
544 }
545 btrfs_release_path(root, path);
546
Yan Zheng07d400a2009-01-06 11:42:00 -0500547 saved_nbytes = inode_get_bytes(inode);
Chris Masone02119d2008-09-05 16:13:11 -0400548 /* drop any overlapping extents */
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000549 ret = btrfs_drop_extents(trans, inode, start, extent_end,
550 &alloc_hint, 1);
Chris Masone02119d2008-09-05 16:13:11 -0400551 BUG_ON(ret);
552
Yan Zheng07d400a2009-01-06 11:42:00 -0500553 if (found_type == BTRFS_FILE_EXTENT_REG ||
554 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400555 u64 offset;
Yan Zheng07d400a2009-01-06 11:42:00 -0500556 unsigned long dest_offset;
557 struct btrfs_key ins;
Chris Masone02119d2008-09-05 16:13:11 -0400558
Yan Zheng07d400a2009-01-06 11:42:00 -0500559 ret = btrfs_insert_empty_item(trans, root, path, key,
560 sizeof(*item));
561 BUG_ON(ret);
562 dest_offset = btrfs_item_ptr_offset(path->nodes[0],
563 path->slots[0]);
564 copy_extent_buffer(path->nodes[0], eb, dest_offset,
565 (unsigned long)item, sizeof(*item));
566
567 ins.objectid = btrfs_file_extent_disk_bytenr(eb, item);
568 ins.offset = btrfs_file_extent_disk_num_bytes(eb, item);
569 ins.type = BTRFS_EXTENT_ITEM_KEY;
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400570 offset = key->offset - btrfs_file_extent_offset(eb, item);
Yan Zheng07d400a2009-01-06 11:42:00 -0500571
572 if (ins.objectid > 0) {
573 u64 csum_start;
574 u64 csum_end;
575 LIST_HEAD(ordered_sums);
576 /*
577 * is this extent already allocated in the extent
578 * allocation tree? If so, just add a reference
579 */
580 ret = btrfs_lookup_extent(root, ins.objectid,
581 ins.offset);
582 if (ret == 0) {
583 ret = btrfs_inc_extent_ref(trans, root,
584 ins.objectid, ins.offset,
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400585 0, root->root_key.objectid,
586 key->objectid, offset);
Yan Zheng07d400a2009-01-06 11:42:00 -0500587 } else {
588 /*
589 * insert the extent pointer in the extent
590 * allocation tree
591 */
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400592 ret = btrfs_alloc_logged_file_extent(trans,
593 root, root->root_key.objectid,
594 key->objectid, offset, &ins);
Yan Zheng07d400a2009-01-06 11:42:00 -0500595 BUG_ON(ret);
596 }
597 btrfs_release_path(root, path);
598
599 if (btrfs_file_extent_compression(eb, item)) {
600 csum_start = ins.objectid;
601 csum_end = csum_start + ins.offset;
602 } else {
603 csum_start = ins.objectid +
604 btrfs_file_extent_offset(eb, item);
605 csum_end = csum_start +
606 btrfs_file_extent_num_bytes(eb, item);
607 }
608
609 ret = btrfs_lookup_csums_range(root->log_root,
610 csum_start, csum_end - 1,
611 &ordered_sums);
612 BUG_ON(ret);
613 while (!list_empty(&ordered_sums)) {
614 struct btrfs_ordered_sum *sums;
615 sums = list_entry(ordered_sums.next,
616 struct btrfs_ordered_sum,
617 list);
618 ret = btrfs_csum_file_blocks(trans,
619 root->fs_info->csum_root,
620 sums);
621 BUG_ON(ret);
622 list_del(&sums->list);
623 kfree(sums);
624 }
625 } else {
626 btrfs_release_path(root, path);
627 }
628 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
629 /* inline extents are easy, we just overwrite them */
630 ret = overwrite_item(trans, root, path, eb, slot, key);
631 BUG_ON(ret);
632 }
633
634 inode_set_bytes(inode, saved_nbytes);
Chris Masone02119d2008-09-05 16:13:11 -0400635 btrfs_update_inode(trans, root, inode);
636out:
637 if (inode)
638 iput(inode);
639 return ret;
640}
641
642/*
643 * when cleaning up conflicts between the directory names in the
644 * subvolume, directory names in the log and directory names in the
645 * inode back references, we may have to unlink inodes from directories.
646 *
647 * This is a helper function to do the unlink of a specific directory
648 * item
649 */
650static noinline int drop_one_dir_item(struct btrfs_trans_handle *trans,
651 struct btrfs_root *root,
652 struct btrfs_path *path,
653 struct inode *dir,
654 struct btrfs_dir_item *di)
655{
656 struct inode *inode;
657 char *name;
658 int name_len;
659 struct extent_buffer *leaf;
660 struct btrfs_key location;
661 int ret;
662
663 leaf = path->nodes[0];
664
665 btrfs_dir_item_key_to_cpu(leaf, di, &location);
666 name_len = btrfs_dir_name_len(leaf, di);
667 name = kmalloc(name_len, GFP_NOFS);
668 read_extent_buffer(leaf, name, (unsigned long)(di + 1), name_len);
669 btrfs_release_path(root, path);
670
671 inode = read_one_inode(root, location.objectid);
672 BUG_ON(!inode);
673
Yan Zhengec051c02009-01-05 15:43:42 -0500674 ret = link_to_fixup_dir(trans, root, path, location.objectid);
675 BUG_ON(ret);
Chris Mason12fcfd22009-03-24 10:24:20 -0400676
Chris Masone02119d2008-09-05 16:13:11 -0400677 ret = btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
Yan Zhengec051c02009-01-05 15:43:42 -0500678 BUG_ON(ret);
Chris Masone02119d2008-09-05 16:13:11 -0400679 kfree(name);
680
681 iput(inode);
682 return ret;
683}
684
685/*
686 * helper function to see if a given name and sequence number found
687 * in an inode back reference are already in a directory and correctly
688 * point to this inode
689 */
690static noinline int inode_in_dir(struct btrfs_root *root,
691 struct btrfs_path *path,
692 u64 dirid, u64 objectid, u64 index,
693 const char *name, int name_len)
694{
695 struct btrfs_dir_item *di;
696 struct btrfs_key location;
697 int match = 0;
698
699 di = btrfs_lookup_dir_index_item(NULL, root, path, dirid,
700 index, name, name_len, 0);
701 if (di && !IS_ERR(di)) {
702 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
703 if (location.objectid != objectid)
704 goto out;
705 } else
706 goto out;
707 btrfs_release_path(root, path);
708
709 di = btrfs_lookup_dir_item(NULL, root, path, dirid, name, name_len, 0);
710 if (di && !IS_ERR(di)) {
711 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
712 if (location.objectid != objectid)
713 goto out;
714 } else
715 goto out;
716 match = 1;
717out:
718 btrfs_release_path(root, path);
719 return match;
720}
721
722/*
723 * helper function to check a log tree for a named back reference in
724 * an inode. This is used to decide if a back reference that is
725 * found in the subvolume conflicts with what we find in the log.
726 *
727 * inode backreferences may have multiple refs in a single item,
728 * during replay we process one reference at a time, and we don't
729 * want to delete valid links to a file from the subvolume if that
730 * link is also in the log.
731 */
732static noinline int backref_in_log(struct btrfs_root *log,
733 struct btrfs_key *key,
734 char *name, int namelen)
735{
736 struct btrfs_path *path;
737 struct btrfs_inode_ref *ref;
738 unsigned long ptr;
739 unsigned long ptr_end;
740 unsigned long name_ptr;
741 int found_name_len;
742 int item_size;
743 int ret;
744 int match = 0;
745
746 path = btrfs_alloc_path();
747 ret = btrfs_search_slot(NULL, log, key, path, 0, 0);
748 if (ret != 0)
749 goto out;
750
751 item_size = btrfs_item_size_nr(path->nodes[0], path->slots[0]);
752 ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
753 ptr_end = ptr + item_size;
754 while (ptr < ptr_end) {
755 ref = (struct btrfs_inode_ref *)ptr;
756 found_name_len = btrfs_inode_ref_name_len(path->nodes[0], ref);
757 if (found_name_len == namelen) {
758 name_ptr = (unsigned long)(ref + 1);
759 ret = memcmp_extent_buffer(path->nodes[0], name,
760 name_ptr, namelen);
761 if (ret == 0) {
762 match = 1;
763 goto out;
764 }
765 }
766 ptr = (unsigned long)(ref + 1) + found_name_len;
767 }
768out:
769 btrfs_free_path(path);
770 return match;
771}
772
773
774/*
775 * replay one inode back reference item found in the log tree.
776 * eb, slot and key refer to the buffer and key found in the log tree.
777 * root is the destination we are replaying into, and path is for temp
778 * use by this function. (it should be released on return).
779 */
780static noinline int add_inode_ref(struct btrfs_trans_handle *trans,
781 struct btrfs_root *root,
782 struct btrfs_root *log,
783 struct btrfs_path *path,
784 struct extent_buffer *eb, int slot,
785 struct btrfs_key *key)
786{
787 struct inode *dir;
788 int ret;
789 struct btrfs_key location;
790 struct btrfs_inode_ref *ref;
791 struct btrfs_dir_item *di;
792 struct inode *inode;
793 char *name;
794 int namelen;
795 unsigned long ref_ptr;
796 unsigned long ref_end;
797
798 location.objectid = key->objectid;
799 location.type = BTRFS_INODE_ITEM_KEY;
800 location.offset = 0;
801
802 /*
803 * it is possible that we didn't log all the parent directories
804 * for a given inode. If we don't find the dir, just don't
805 * copy the back ref in. The link count fixup code will take
806 * care of the rest
807 */
808 dir = read_one_inode(root, key->offset);
809 if (!dir)
810 return -ENOENT;
811
812 inode = read_one_inode(root, key->objectid);
Julia Lawall631c07c2009-07-27 13:57:00 -0400813 BUG_ON(!inode);
Chris Masone02119d2008-09-05 16:13:11 -0400814
815 ref_ptr = btrfs_item_ptr_offset(eb, slot);
816 ref_end = ref_ptr + btrfs_item_size_nr(eb, slot);
817
818again:
819 ref = (struct btrfs_inode_ref *)ref_ptr;
820
821 namelen = btrfs_inode_ref_name_len(eb, ref);
822 name = kmalloc(namelen, GFP_NOFS);
823 BUG_ON(!name);
824
825 read_extent_buffer(eb, name, (unsigned long)(ref + 1), namelen);
826
827 /* if we already have a perfect match, we're done */
828 if (inode_in_dir(root, path, dir->i_ino, inode->i_ino,
829 btrfs_inode_ref_index(eb, ref),
830 name, namelen)) {
831 goto out;
832 }
833
834 /*
835 * look for a conflicting back reference in the metadata.
836 * if we find one we have to unlink that name of the file
837 * before we add our new link. Later on, we overwrite any
838 * existing back reference, and we don't want to create
839 * dangling pointers in the directory.
840 */
841conflict_again:
842 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
843 if (ret == 0) {
844 char *victim_name;
845 int victim_name_len;
846 struct btrfs_inode_ref *victim_ref;
847 unsigned long ptr;
848 unsigned long ptr_end;
849 struct extent_buffer *leaf = path->nodes[0];
850
851 /* are we trying to overwrite a back ref for the root directory
852 * if so, just jump out, we're done
853 */
854 if (key->objectid == key->offset)
855 goto out_nowrite;
856
857 /* check all the names in this back reference to see
858 * if they are in the log. if so, we allow them to stay
859 * otherwise they must be unlinked as a conflict
860 */
861 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
862 ptr_end = ptr + btrfs_item_size_nr(leaf, path->slots[0]);
Chris Masond3977122009-01-05 21:25:51 -0500863 while (ptr < ptr_end) {
Chris Masone02119d2008-09-05 16:13:11 -0400864 victim_ref = (struct btrfs_inode_ref *)ptr;
865 victim_name_len = btrfs_inode_ref_name_len(leaf,
866 victim_ref);
867 victim_name = kmalloc(victim_name_len, GFP_NOFS);
868 BUG_ON(!victim_name);
869
870 read_extent_buffer(leaf, victim_name,
871 (unsigned long)(victim_ref + 1),
872 victim_name_len);
873
874 if (!backref_in_log(log, key, victim_name,
875 victim_name_len)) {
876 btrfs_inc_nlink(inode);
877 btrfs_release_path(root, path);
Chris Mason12fcfd22009-03-24 10:24:20 -0400878
Chris Masone02119d2008-09-05 16:13:11 -0400879 ret = btrfs_unlink_inode(trans, root, dir,
880 inode, victim_name,
881 victim_name_len);
882 kfree(victim_name);
883 btrfs_release_path(root, path);
884 goto conflict_again;
885 }
886 kfree(victim_name);
887 ptr = (unsigned long)(victim_ref + 1) + victim_name_len;
888 }
889 BUG_ON(ret);
890 }
891 btrfs_release_path(root, path);
892
893 /* look for a conflicting sequence number */
894 di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
895 btrfs_inode_ref_index(eb, ref),
896 name, namelen, 0);
897 if (di && !IS_ERR(di)) {
898 ret = drop_one_dir_item(trans, root, path, dir, di);
899 BUG_ON(ret);
900 }
901 btrfs_release_path(root, path);
902
903
904 /* look for a conflicting name */
905 di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
906 name, namelen, 0);
907 if (di && !IS_ERR(di)) {
908 ret = drop_one_dir_item(trans, root, path, dir, di);
909 BUG_ON(ret);
910 }
911 btrfs_release_path(root, path);
912
913 /* insert our name */
914 ret = btrfs_add_link(trans, dir, inode, name, namelen, 0,
915 btrfs_inode_ref_index(eb, ref));
916 BUG_ON(ret);
917
918 btrfs_update_inode(trans, root, inode);
919
920out:
921 ref_ptr = (unsigned long)(ref + 1) + namelen;
922 kfree(name);
923 if (ref_ptr < ref_end)
924 goto again;
925
926 /* finally write the back reference in the inode */
927 ret = overwrite_item(trans, root, path, eb, slot, key);
928 BUG_ON(ret);
929
930out_nowrite:
931 btrfs_release_path(root, path);
932 iput(dir);
933 iput(inode);
934 return 0;
935}
936
Yan, Zhengc71bf092009-11-12 09:34:40 +0000937static int insert_orphan_item(struct btrfs_trans_handle *trans,
938 struct btrfs_root *root, u64 offset)
939{
940 int ret;
941 ret = btrfs_find_orphan_item(root, offset);
942 if (ret > 0)
943 ret = btrfs_insert_orphan_item(trans, root, offset);
944 return ret;
945}
946
947
Chris Masone02119d2008-09-05 16:13:11 -0400948/*
Chris Masone02119d2008-09-05 16:13:11 -0400949 * There are a few corners where the link count of the file can't
950 * be properly maintained during replay. So, instead of adding
951 * lots of complexity to the log code, we just scan the backrefs
952 * for any file that has been through replay.
953 *
954 * The scan will update the link count on the inode to reflect the
955 * number of back refs found. If it goes down to zero, the iput
956 * will free the inode.
957 */
958static noinline int fixup_inode_link_count(struct btrfs_trans_handle *trans,
959 struct btrfs_root *root,
960 struct inode *inode)
961{
962 struct btrfs_path *path;
963 int ret;
964 struct btrfs_key key;
965 u64 nlink = 0;
966 unsigned long ptr;
967 unsigned long ptr_end;
968 int name_len;
969
970 key.objectid = inode->i_ino;
971 key.type = BTRFS_INODE_REF_KEY;
972 key.offset = (u64)-1;
973
974 path = btrfs_alloc_path();
975
Chris Masond3977122009-01-05 21:25:51 -0500976 while (1) {
Chris Masone02119d2008-09-05 16:13:11 -0400977 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
978 if (ret < 0)
979 break;
980 if (ret > 0) {
981 if (path->slots[0] == 0)
982 break;
983 path->slots[0]--;
984 }
985 btrfs_item_key_to_cpu(path->nodes[0], &key,
986 path->slots[0]);
987 if (key.objectid != inode->i_ino ||
988 key.type != BTRFS_INODE_REF_KEY)
989 break;
990 ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
991 ptr_end = ptr + btrfs_item_size_nr(path->nodes[0],
992 path->slots[0]);
Chris Masond3977122009-01-05 21:25:51 -0500993 while (ptr < ptr_end) {
Chris Masone02119d2008-09-05 16:13:11 -0400994 struct btrfs_inode_ref *ref;
995
996 ref = (struct btrfs_inode_ref *)ptr;
997 name_len = btrfs_inode_ref_name_len(path->nodes[0],
998 ref);
999 ptr = (unsigned long)(ref + 1) + name_len;
1000 nlink++;
1001 }
1002
1003 if (key.offset == 0)
1004 break;
1005 key.offset--;
1006 btrfs_release_path(root, path);
1007 }
Chris Mason12fcfd22009-03-24 10:24:20 -04001008 btrfs_release_path(root, path);
Chris Masone02119d2008-09-05 16:13:11 -04001009 if (nlink != inode->i_nlink) {
1010 inode->i_nlink = nlink;
1011 btrfs_update_inode(trans, root, inode);
1012 }
Chris Mason8d5bf1c2008-09-11 15:51:21 -04001013 BTRFS_I(inode)->index_cnt = (u64)-1;
Chris Masone02119d2008-09-05 16:13:11 -04001014
Yan, Zhengc71bf092009-11-12 09:34:40 +00001015 if (inode->i_nlink == 0) {
1016 if (S_ISDIR(inode->i_mode)) {
1017 ret = replay_dir_deletes(trans, root, NULL, path,
1018 inode->i_ino, 1);
1019 BUG_ON(ret);
1020 }
1021 ret = insert_orphan_item(trans, root, inode->i_ino);
Chris Mason12fcfd22009-03-24 10:24:20 -04001022 BUG_ON(ret);
1023 }
1024 btrfs_free_path(path);
1025
Chris Masone02119d2008-09-05 16:13:11 -04001026 return 0;
1027}
1028
1029static noinline int fixup_inode_link_counts(struct btrfs_trans_handle *trans,
1030 struct btrfs_root *root,
1031 struct btrfs_path *path)
1032{
1033 int ret;
1034 struct btrfs_key key;
1035 struct inode *inode;
1036
1037 key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID;
1038 key.type = BTRFS_ORPHAN_ITEM_KEY;
1039 key.offset = (u64)-1;
Chris Masond3977122009-01-05 21:25:51 -05001040 while (1) {
Chris Masone02119d2008-09-05 16:13:11 -04001041 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1042 if (ret < 0)
1043 break;
1044
1045 if (ret == 1) {
1046 if (path->slots[0] == 0)
1047 break;
1048 path->slots[0]--;
1049 }
1050
1051 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1052 if (key.objectid != BTRFS_TREE_LOG_FIXUP_OBJECTID ||
1053 key.type != BTRFS_ORPHAN_ITEM_KEY)
1054 break;
1055
1056 ret = btrfs_del_item(trans, root, path);
1057 BUG_ON(ret);
1058
1059 btrfs_release_path(root, path);
1060 inode = read_one_inode(root, key.offset);
1061 BUG_ON(!inode);
1062
1063 ret = fixup_inode_link_count(trans, root, inode);
1064 BUG_ON(ret);
1065
1066 iput(inode);
1067
Chris Mason12fcfd22009-03-24 10:24:20 -04001068 /*
1069 * fixup on a directory may create new entries,
1070 * make sure we always look for the highset possible
1071 * offset
1072 */
1073 key.offset = (u64)-1;
Chris Masone02119d2008-09-05 16:13:11 -04001074 }
1075 btrfs_release_path(root, path);
1076 return 0;
1077}
1078
1079
1080/*
1081 * record a given inode in the fixup dir so we can check its link
1082 * count when replay is done. The link count is incremented here
1083 * so the inode won't go away until we check it
1084 */
1085static noinline int link_to_fixup_dir(struct btrfs_trans_handle *trans,
1086 struct btrfs_root *root,
1087 struct btrfs_path *path,
1088 u64 objectid)
1089{
1090 struct btrfs_key key;
1091 int ret = 0;
1092 struct inode *inode;
1093
1094 inode = read_one_inode(root, objectid);
1095 BUG_ON(!inode);
1096
1097 key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID;
1098 btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
1099 key.offset = objectid;
1100
1101 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1102
1103 btrfs_release_path(root, path);
1104 if (ret == 0) {
1105 btrfs_inc_nlink(inode);
1106 btrfs_update_inode(trans, root, inode);
1107 } else if (ret == -EEXIST) {
1108 ret = 0;
1109 } else {
1110 BUG();
1111 }
1112 iput(inode);
1113
1114 return ret;
1115}
1116
1117/*
1118 * when replaying the log for a directory, we only insert names
1119 * for inodes that actually exist. This means an fsync on a directory
1120 * does not implicitly fsync all the new files in it
1121 */
1122static noinline int insert_one_name(struct btrfs_trans_handle *trans,
1123 struct btrfs_root *root,
1124 struct btrfs_path *path,
1125 u64 dirid, u64 index,
1126 char *name, int name_len, u8 type,
1127 struct btrfs_key *location)
1128{
1129 struct inode *inode;
1130 struct inode *dir;
1131 int ret;
1132
1133 inode = read_one_inode(root, location->objectid);
1134 if (!inode)
1135 return -ENOENT;
1136
1137 dir = read_one_inode(root, dirid);
1138 if (!dir) {
1139 iput(inode);
1140 return -EIO;
1141 }
1142 ret = btrfs_add_link(trans, dir, inode, name, name_len, 1, index);
1143
1144 /* FIXME, put inode into FIXUP list */
1145
1146 iput(inode);
1147 iput(dir);
1148 return ret;
1149}
1150
1151/*
1152 * take a single entry in a log directory item and replay it into
1153 * the subvolume.
1154 *
1155 * if a conflicting item exists in the subdirectory already,
1156 * the inode it points to is unlinked and put into the link count
1157 * fix up tree.
1158 *
1159 * If a name from the log points to a file or directory that does
1160 * not exist in the FS, it is skipped. fsyncs on directories
1161 * do not force down inodes inside that directory, just changes to the
1162 * names or unlinks in a directory.
1163 */
1164static noinline int replay_one_name(struct btrfs_trans_handle *trans,
1165 struct btrfs_root *root,
1166 struct btrfs_path *path,
1167 struct extent_buffer *eb,
1168 struct btrfs_dir_item *di,
1169 struct btrfs_key *key)
1170{
1171 char *name;
1172 int name_len;
1173 struct btrfs_dir_item *dst_di;
1174 struct btrfs_key found_key;
1175 struct btrfs_key log_key;
1176 struct inode *dir;
Chris Masone02119d2008-09-05 16:13:11 -04001177 u8 log_type;
Chris Mason4bef0842008-09-08 11:18:08 -04001178 int exists;
Chris Masone02119d2008-09-05 16:13:11 -04001179 int ret;
1180
1181 dir = read_one_inode(root, key->objectid);
1182 BUG_ON(!dir);
1183
1184 name_len = btrfs_dir_name_len(eb, di);
1185 name = kmalloc(name_len, GFP_NOFS);
1186 log_type = btrfs_dir_type(eb, di);
1187 read_extent_buffer(eb, name, (unsigned long)(di + 1),
1188 name_len);
1189
1190 btrfs_dir_item_key_to_cpu(eb, di, &log_key);
Chris Mason4bef0842008-09-08 11:18:08 -04001191 exists = btrfs_lookup_inode(trans, root, path, &log_key, 0);
1192 if (exists == 0)
1193 exists = 1;
1194 else
1195 exists = 0;
1196 btrfs_release_path(root, path);
1197
Chris Masone02119d2008-09-05 16:13:11 -04001198 if (key->type == BTRFS_DIR_ITEM_KEY) {
1199 dst_di = btrfs_lookup_dir_item(trans, root, path, key->objectid,
1200 name, name_len, 1);
Chris Masond3977122009-01-05 21:25:51 -05001201 } else if (key->type == BTRFS_DIR_INDEX_KEY) {
Chris Masone02119d2008-09-05 16:13:11 -04001202 dst_di = btrfs_lookup_dir_index_item(trans, root, path,
1203 key->objectid,
1204 key->offset, name,
1205 name_len, 1);
1206 } else {
1207 BUG();
1208 }
1209 if (!dst_di || IS_ERR(dst_di)) {
1210 /* we need a sequence number to insert, so we only
1211 * do inserts for the BTRFS_DIR_INDEX_KEY types
1212 */
1213 if (key->type != BTRFS_DIR_INDEX_KEY)
1214 goto out;
1215 goto insert;
1216 }
1217
1218 btrfs_dir_item_key_to_cpu(path->nodes[0], dst_di, &found_key);
1219 /* the existing item matches the logged item */
1220 if (found_key.objectid == log_key.objectid &&
1221 found_key.type == log_key.type &&
1222 found_key.offset == log_key.offset &&
1223 btrfs_dir_type(path->nodes[0], dst_di) == log_type) {
1224 goto out;
1225 }
1226
1227 /*
1228 * don't drop the conflicting directory entry if the inode
1229 * for the new entry doesn't exist
1230 */
Chris Mason4bef0842008-09-08 11:18:08 -04001231 if (!exists)
Chris Masone02119d2008-09-05 16:13:11 -04001232 goto out;
1233
Chris Masone02119d2008-09-05 16:13:11 -04001234 ret = drop_one_dir_item(trans, root, path, dir, dst_di);
1235 BUG_ON(ret);
1236
1237 if (key->type == BTRFS_DIR_INDEX_KEY)
1238 goto insert;
1239out:
1240 btrfs_release_path(root, path);
1241 kfree(name);
1242 iput(dir);
1243 return 0;
1244
1245insert:
1246 btrfs_release_path(root, path);
1247 ret = insert_one_name(trans, root, path, key->objectid, key->offset,
1248 name, name_len, log_type, &log_key);
1249
Stoyan Gaydarovc2934982009-04-02 17:05:11 -04001250 BUG_ON(ret && ret != -ENOENT);
Chris Masone02119d2008-09-05 16:13:11 -04001251 goto out;
1252}
1253
1254/*
1255 * find all the names in a directory item and reconcile them into
1256 * the subvolume. Only BTRFS_DIR_ITEM_KEY types will have more than
1257 * one name in a directory item, but the same code gets used for
1258 * both directory index types
1259 */
1260static noinline int replay_one_dir_item(struct btrfs_trans_handle *trans,
1261 struct btrfs_root *root,
1262 struct btrfs_path *path,
1263 struct extent_buffer *eb, int slot,
1264 struct btrfs_key *key)
1265{
1266 int ret;
1267 u32 item_size = btrfs_item_size_nr(eb, slot);
1268 struct btrfs_dir_item *di;
1269 int name_len;
1270 unsigned long ptr;
1271 unsigned long ptr_end;
1272
1273 ptr = btrfs_item_ptr_offset(eb, slot);
1274 ptr_end = ptr + item_size;
Chris Masond3977122009-01-05 21:25:51 -05001275 while (ptr < ptr_end) {
Chris Masone02119d2008-09-05 16:13:11 -04001276 di = (struct btrfs_dir_item *)ptr;
1277 name_len = btrfs_dir_name_len(eb, di);
1278 ret = replay_one_name(trans, root, path, eb, di, key);
1279 BUG_ON(ret);
1280 ptr = (unsigned long)(di + 1);
1281 ptr += name_len;
1282 }
1283 return 0;
1284}
1285
1286/*
1287 * directory replay has two parts. There are the standard directory
1288 * items in the log copied from the subvolume, and range items
1289 * created in the log while the subvolume was logged.
1290 *
1291 * The range items tell us which parts of the key space the log
1292 * is authoritative for. During replay, if a key in the subvolume
1293 * directory is in a logged range item, but not actually in the log
1294 * that means it was deleted from the directory before the fsync
1295 * and should be removed.
1296 */
1297static noinline int find_dir_range(struct btrfs_root *root,
1298 struct btrfs_path *path,
1299 u64 dirid, int key_type,
1300 u64 *start_ret, u64 *end_ret)
1301{
1302 struct btrfs_key key;
1303 u64 found_end;
1304 struct btrfs_dir_log_item *item;
1305 int ret;
1306 int nritems;
1307
1308 if (*start_ret == (u64)-1)
1309 return 1;
1310
1311 key.objectid = dirid;
1312 key.type = key_type;
1313 key.offset = *start_ret;
1314
1315 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1316 if (ret < 0)
1317 goto out;
1318 if (ret > 0) {
1319 if (path->slots[0] == 0)
1320 goto out;
1321 path->slots[0]--;
1322 }
1323 if (ret != 0)
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 next;
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
1334 if (*start_ret >= key.offset && *start_ret <= found_end) {
1335 ret = 0;
1336 *start_ret = key.offset;
1337 *end_ret = found_end;
1338 goto out;
1339 }
1340 ret = 1;
1341next:
1342 /* check the next slot in the tree to see if it is a valid item */
1343 nritems = btrfs_header_nritems(path->nodes[0]);
1344 if (path->slots[0] >= nritems) {
1345 ret = btrfs_next_leaf(root, path);
1346 if (ret)
1347 goto out;
1348 } else {
1349 path->slots[0]++;
1350 }
1351
1352 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1353
1354 if (key.type != key_type || key.objectid != dirid) {
1355 ret = 1;
1356 goto out;
1357 }
1358 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
1359 struct btrfs_dir_log_item);
1360 found_end = btrfs_dir_log_end(path->nodes[0], item);
1361 *start_ret = key.offset;
1362 *end_ret = found_end;
1363 ret = 0;
1364out:
1365 btrfs_release_path(root, path);
1366 return ret;
1367}
1368
1369/*
1370 * this looks for a given directory item in the log. If the directory
1371 * item is not in the log, the item is removed and the inode it points
1372 * to is unlinked
1373 */
1374static noinline int check_item_in_log(struct btrfs_trans_handle *trans,
1375 struct btrfs_root *root,
1376 struct btrfs_root *log,
1377 struct btrfs_path *path,
1378 struct btrfs_path *log_path,
1379 struct inode *dir,
1380 struct btrfs_key *dir_key)
1381{
1382 int ret;
1383 struct extent_buffer *eb;
1384 int slot;
1385 u32 item_size;
1386 struct btrfs_dir_item *di;
1387 struct btrfs_dir_item *log_di;
1388 int name_len;
1389 unsigned long ptr;
1390 unsigned long ptr_end;
1391 char *name;
1392 struct inode *inode;
1393 struct btrfs_key location;
1394
1395again:
1396 eb = path->nodes[0];
1397 slot = path->slots[0];
1398 item_size = btrfs_item_size_nr(eb, slot);
1399 ptr = btrfs_item_ptr_offset(eb, slot);
1400 ptr_end = ptr + item_size;
Chris Masond3977122009-01-05 21:25:51 -05001401 while (ptr < ptr_end) {
Chris Masone02119d2008-09-05 16:13:11 -04001402 di = (struct btrfs_dir_item *)ptr;
1403 name_len = btrfs_dir_name_len(eb, di);
1404 name = kmalloc(name_len, GFP_NOFS);
1405 if (!name) {
1406 ret = -ENOMEM;
1407 goto out;
1408 }
1409 read_extent_buffer(eb, name, (unsigned long)(di + 1),
1410 name_len);
1411 log_di = NULL;
Chris Mason12fcfd22009-03-24 10:24:20 -04001412 if (log && dir_key->type == BTRFS_DIR_ITEM_KEY) {
Chris Masone02119d2008-09-05 16:13:11 -04001413 log_di = btrfs_lookup_dir_item(trans, log, log_path,
1414 dir_key->objectid,
1415 name, name_len, 0);
Chris Mason12fcfd22009-03-24 10:24:20 -04001416 } else if (log && dir_key->type == BTRFS_DIR_INDEX_KEY) {
Chris Masone02119d2008-09-05 16:13:11 -04001417 log_di = btrfs_lookup_dir_index_item(trans, log,
1418 log_path,
1419 dir_key->objectid,
1420 dir_key->offset,
1421 name, name_len, 0);
1422 }
1423 if (!log_di || IS_ERR(log_di)) {
1424 btrfs_dir_item_key_to_cpu(eb, di, &location);
1425 btrfs_release_path(root, path);
1426 btrfs_release_path(log, log_path);
1427 inode = read_one_inode(root, location.objectid);
1428 BUG_ON(!inode);
1429
1430 ret = link_to_fixup_dir(trans, root,
1431 path, location.objectid);
1432 BUG_ON(ret);
1433 btrfs_inc_nlink(inode);
1434 ret = btrfs_unlink_inode(trans, root, dir, inode,
1435 name, name_len);
1436 BUG_ON(ret);
1437 kfree(name);
1438 iput(inode);
1439
1440 /* there might still be more names under this key
1441 * check and repeat if required
1442 */
1443 ret = btrfs_search_slot(NULL, root, dir_key, path,
1444 0, 0);
1445 if (ret == 0)
1446 goto again;
1447 ret = 0;
1448 goto out;
1449 }
1450 btrfs_release_path(log, log_path);
1451 kfree(name);
1452
1453 ptr = (unsigned long)(di + 1);
1454 ptr += name_len;
1455 }
1456 ret = 0;
1457out:
1458 btrfs_release_path(root, path);
1459 btrfs_release_path(log, log_path);
1460 return ret;
1461}
1462
1463/*
1464 * deletion replay happens before we copy any new directory items
1465 * out of the log or out of backreferences from inodes. It
1466 * scans the log to find ranges of keys that log is authoritative for,
1467 * and then scans the directory to find items in those ranges that are
1468 * not present in the log.
1469 *
1470 * Anything we don't find in the log is unlinked and removed from the
1471 * directory.
1472 */
1473static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans,
1474 struct btrfs_root *root,
1475 struct btrfs_root *log,
1476 struct btrfs_path *path,
Chris Mason12fcfd22009-03-24 10:24:20 -04001477 u64 dirid, int del_all)
Chris Masone02119d2008-09-05 16:13:11 -04001478{
1479 u64 range_start;
1480 u64 range_end;
1481 int key_type = BTRFS_DIR_LOG_ITEM_KEY;
1482 int ret = 0;
1483 struct btrfs_key dir_key;
1484 struct btrfs_key found_key;
1485 struct btrfs_path *log_path;
1486 struct inode *dir;
1487
1488 dir_key.objectid = dirid;
1489 dir_key.type = BTRFS_DIR_ITEM_KEY;
1490 log_path = btrfs_alloc_path();
1491 if (!log_path)
1492 return -ENOMEM;
1493
1494 dir = read_one_inode(root, dirid);
1495 /* it isn't an error if the inode isn't there, that can happen
1496 * because we replay the deletes before we copy in the inode item
1497 * from the log
1498 */
1499 if (!dir) {
1500 btrfs_free_path(log_path);
1501 return 0;
1502 }
1503again:
1504 range_start = 0;
1505 range_end = 0;
Chris Masond3977122009-01-05 21:25:51 -05001506 while (1) {
Chris Mason12fcfd22009-03-24 10:24:20 -04001507 if (del_all)
1508 range_end = (u64)-1;
1509 else {
1510 ret = find_dir_range(log, path, dirid, key_type,
1511 &range_start, &range_end);
1512 if (ret != 0)
1513 break;
1514 }
Chris Masone02119d2008-09-05 16:13:11 -04001515
1516 dir_key.offset = range_start;
Chris Masond3977122009-01-05 21:25:51 -05001517 while (1) {
Chris Masone02119d2008-09-05 16:13:11 -04001518 int nritems;
1519 ret = btrfs_search_slot(NULL, root, &dir_key, path,
1520 0, 0);
1521 if (ret < 0)
1522 goto out;
1523
1524 nritems = btrfs_header_nritems(path->nodes[0]);
1525 if (path->slots[0] >= nritems) {
1526 ret = btrfs_next_leaf(root, path);
1527 if (ret)
1528 break;
1529 }
1530 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1531 path->slots[0]);
1532 if (found_key.objectid != dirid ||
1533 found_key.type != dir_key.type)
1534 goto next_type;
1535
1536 if (found_key.offset > range_end)
1537 break;
1538
1539 ret = check_item_in_log(trans, root, log, path,
Chris Mason12fcfd22009-03-24 10:24:20 -04001540 log_path, dir,
1541 &found_key);
Chris Masone02119d2008-09-05 16:13:11 -04001542 BUG_ON(ret);
1543 if (found_key.offset == (u64)-1)
1544 break;
1545 dir_key.offset = found_key.offset + 1;
1546 }
1547 btrfs_release_path(root, path);
1548 if (range_end == (u64)-1)
1549 break;
1550 range_start = range_end + 1;
1551 }
1552
1553next_type:
1554 ret = 0;
1555 if (key_type == BTRFS_DIR_LOG_ITEM_KEY) {
1556 key_type = BTRFS_DIR_LOG_INDEX_KEY;
1557 dir_key.type = BTRFS_DIR_INDEX_KEY;
1558 btrfs_release_path(root, path);
1559 goto again;
1560 }
1561out:
1562 btrfs_release_path(root, path);
1563 btrfs_free_path(log_path);
1564 iput(dir);
1565 return ret;
1566}
1567
1568/*
1569 * the process_func used to replay items from the log tree. This
1570 * gets called in two different stages. The first stage just looks
1571 * for inodes and makes sure they are all copied into the subvolume.
1572 *
1573 * The second stage copies all the other item types from the log into
1574 * the subvolume. The two stage approach is slower, but gets rid of
1575 * lots of complexity around inodes referencing other inodes that exist
1576 * only in the log (references come from either directory items or inode
1577 * back refs).
1578 */
1579static int replay_one_buffer(struct btrfs_root *log, struct extent_buffer *eb,
1580 struct walk_control *wc, u64 gen)
1581{
1582 int nritems;
1583 struct btrfs_path *path;
1584 struct btrfs_root *root = wc->replay_dest;
1585 struct btrfs_key key;
1586 u32 item_size;
1587 int level;
1588 int i;
1589 int ret;
1590
1591 btrfs_read_buffer(eb, gen);
1592
1593 level = btrfs_header_level(eb);
1594
1595 if (level != 0)
1596 return 0;
1597
1598 path = btrfs_alloc_path();
1599 BUG_ON(!path);
1600
1601 nritems = btrfs_header_nritems(eb);
1602 for (i = 0; i < nritems; i++) {
1603 btrfs_item_key_to_cpu(eb, &key, i);
1604 item_size = btrfs_item_size_nr(eb, i);
1605
1606 /* inode keys are done during the first stage */
1607 if (key.type == BTRFS_INODE_ITEM_KEY &&
1608 wc->stage == LOG_WALK_REPLAY_INODES) {
Chris Masone02119d2008-09-05 16:13:11 -04001609 struct btrfs_inode_item *inode_item;
1610 u32 mode;
1611
1612 inode_item = btrfs_item_ptr(eb, i,
1613 struct btrfs_inode_item);
1614 mode = btrfs_inode_mode(eb, inode_item);
1615 if (S_ISDIR(mode)) {
1616 ret = replay_dir_deletes(wc->trans,
Chris Mason12fcfd22009-03-24 10:24:20 -04001617 root, log, path, key.objectid, 0);
Chris Masone02119d2008-09-05 16:13:11 -04001618 BUG_ON(ret);
1619 }
1620 ret = overwrite_item(wc->trans, root, path,
1621 eb, i, &key);
1622 BUG_ON(ret);
1623
Yan, Zhengc71bf092009-11-12 09:34:40 +00001624 /* for regular files, make sure corresponding
1625 * orhpan item exist. extents past the new EOF
1626 * will be truncated later by orphan cleanup.
Chris Masone02119d2008-09-05 16:13:11 -04001627 */
1628 if (S_ISREG(mode)) {
Yan, Zhengc71bf092009-11-12 09:34:40 +00001629 ret = insert_orphan_item(wc->trans, root,
1630 key.objectid);
Chris Masone02119d2008-09-05 16:13:11 -04001631 BUG_ON(ret);
Chris Masone02119d2008-09-05 16:13:11 -04001632 }
Yan, Zhengc71bf092009-11-12 09:34:40 +00001633
Chris Masone02119d2008-09-05 16:13:11 -04001634 ret = link_to_fixup_dir(wc->trans, root,
1635 path, key.objectid);
1636 BUG_ON(ret);
1637 }
1638 if (wc->stage < LOG_WALK_REPLAY_ALL)
1639 continue;
1640
1641 /* these keys are simply copied */
1642 if (key.type == BTRFS_XATTR_ITEM_KEY) {
1643 ret = overwrite_item(wc->trans, root, path,
1644 eb, i, &key);
1645 BUG_ON(ret);
1646 } else if (key.type == BTRFS_INODE_REF_KEY) {
1647 ret = add_inode_ref(wc->trans, root, log, path,
1648 eb, i, &key);
1649 BUG_ON(ret && ret != -ENOENT);
1650 } else if (key.type == BTRFS_EXTENT_DATA_KEY) {
1651 ret = replay_one_extent(wc->trans, root, path,
1652 eb, i, &key);
1653 BUG_ON(ret);
Chris Masone02119d2008-09-05 16:13:11 -04001654 } else if (key.type == BTRFS_DIR_ITEM_KEY ||
1655 key.type == BTRFS_DIR_INDEX_KEY) {
1656 ret = replay_one_dir_item(wc->trans, root, path,
1657 eb, i, &key);
1658 BUG_ON(ret);
1659 }
1660 }
1661 btrfs_free_path(path);
1662 return 0;
1663}
1664
Chris Masond3977122009-01-05 21:25:51 -05001665static noinline int walk_down_log_tree(struct btrfs_trans_handle *trans,
Chris Masone02119d2008-09-05 16:13:11 -04001666 struct btrfs_root *root,
1667 struct btrfs_path *path, int *level,
1668 struct walk_control *wc)
1669{
1670 u64 root_owner;
1671 u64 root_gen;
1672 u64 bytenr;
1673 u64 ptr_gen;
1674 struct extent_buffer *next;
1675 struct extent_buffer *cur;
1676 struct extent_buffer *parent;
1677 u32 blocksize;
1678 int ret = 0;
1679
1680 WARN_ON(*level < 0);
1681 WARN_ON(*level >= BTRFS_MAX_LEVEL);
1682
Chris Masond3977122009-01-05 21:25:51 -05001683 while (*level > 0) {
Chris Masone02119d2008-09-05 16:13:11 -04001684 WARN_ON(*level < 0);
1685 WARN_ON(*level >= BTRFS_MAX_LEVEL);
1686 cur = path->nodes[*level];
1687
1688 if (btrfs_header_level(cur) != *level)
1689 WARN_ON(1);
1690
1691 if (path->slots[*level] >=
1692 btrfs_header_nritems(cur))
1693 break;
1694
1695 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
1696 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
1697 blocksize = btrfs_level_size(root, *level - 1);
1698
1699 parent = path->nodes[*level];
1700 root_owner = btrfs_header_owner(parent);
1701 root_gen = btrfs_header_generation(parent);
1702
1703 next = btrfs_find_create_tree_block(root, bytenr, blocksize);
1704
Chris Masone02119d2008-09-05 16:13:11 -04001705 if (*level == 1) {
Yan, Zheng4a500fd2010-05-16 10:49:59 -04001706 wc->process_func(root, next, wc, ptr_gen);
1707
Chris Masone02119d2008-09-05 16:13:11 -04001708 path->slots[*level]++;
1709 if (wc->free) {
1710 btrfs_read_buffer(next, ptr_gen);
1711
1712 btrfs_tree_lock(next);
1713 clean_tree_block(trans, root, next);
Chris Masonb4ce94d2009-02-04 09:25:08 -05001714 btrfs_set_lock_blocking(next);
Chris Masone02119d2008-09-05 16:13:11 -04001715 btrfs_wait_tree_block_writeback(next);
1716 btrfs_tree_unlock(next);
1717
Chris Masone02119d2008-09-05 16:13:11 -04001718 WARN_ON(root_owner !=
1719 BTRFS_TREE_LOG_OBJECTID);
Chris Masond00aff02008-09-11 15:54:42 -04001720 ret = btrfs_free_reserved_extent(root,
1721 bytenr, blocksize);
Chris Masone02119d2008-09-05 16:13:11 -04001722 BUG_ON(ret);
1723 }
1724 free_extent_buffer(next);
1725 continue;
1726 }
1727 btrfs_read_buffer(next, ptr_gen);
1728
1729 WARN_ON(*level <= 0);
1730 if (path->nodes[*level-1])
1731 free_extent_buffer(path->nodes[*level-1]);
1732 path->nodes[*level-1] = next;
1733 *level = btrfs_header_level(next);
1734 path->slots[*level] = 0;
1735 cond_resched();
1736 }
1737 WARN_ON(*level < 0);
1738 WARN_ON(*level >= BTRFS_MAX_LEVEL);
1739
Yan, Zheng4a500fd2010-05-16 10:49:59 -04001740 path->slots[*level] = btrfs_header_nritems(path->nodes[*level]);
Chris Masone02119d2008-09-05 16:13:11 -04001741
1742 cond_resched();
1743 return 0;
1744}
1745
Chris Masond3977122009-01-05 21:25:51 -05001746static noinline int walk_up_log_tree(struct btrfs_trans_handle *trans,
Chris Masone02119d2008-09-05 16:13:11 -04001747 struct btrfs_root *root,
1748 struct btrfs_path *path, int *level,
1749 struct walk_control *wc)
1750{
1751 u64 root_owner;
1752 u64 root_gen;
1753 int i;
1754 int slot;
1755 int ret;
1756
Chris Masond3977122009-01-05 21:25:51 -05001757 for (i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) {
Chris Masone02119d2008-09-05 16:13:11 -04001758 slot = path->slots[i];
Yan, Zheng4a500fd2010-05-16 10:49:59 -04001759 if (slot + 1 < btrfs_header_nritems(path->nodes[i])) {
Chris Masone02119d2008-09-05 16:13:11 -04001760 struct extent_buffer *node;
1761 node = path->nodes[i];
1762 path->slots[i]++;
1763 *level = i;
1764 WARN_ON(*level == 0);
1765 return 0;
1766 } else {
Zheng Yan31840ae2008-09-23 13:14:14 -04001767 struct extent_buffer *parent;
1768 if (path->nodes[*level] == root->node)
1769 parent = path->nodes[*level];
1770 else
1771 parent = path->nodes[*level + 1];
1772
1773 root_owner = btrfs_header_owner(parent);
1774 root_gen = btrfs_header_generation(parent);
Chris Masone02119d2008-09-05 16:13:11 -04001775 wc->process_func(root, path->nodes[*level], wc,
1776 btrfs_header_generation(path->nodes[*level]));
1777 if (wc->free) {
1778 struct extent_buffer *next;
1779
1780 next = path->nodes[*level];
1781
1782 btrfs_tree_lock(next);
1783 clean_tree_block(trans, root, next);
Chris Masonb4ce94d2009-02-04 09:25:08 -05001784 btrfs_set_lock_blocking(next);
Chris Masone02119d2008-09-05 16:13:11 -04001785 btrfs_wait_tree_block_writeback(next);
1786 btrfs_tree_unlock(next);
1787
Chris Masone02119d2008-09-05 16:13:11 -04001788 WARN_ON(root_owner != BTRFS_TREE_LOG_OBJECTID);
Chris Masond00aff02008-09-11 15:54:42 -04001789 ret = btrfs_free_reserved_extent(root,
Chris Masone02119d2008-09-05 16:13:11 -04001790 path->nodes[*level]->start,
Chris Masond00aff02008-09-11 15:54:42 -04001791 path->nodes[*level]->len);
Chris Masone02119d2008-09-05 16:13:11 -04001792 BUG_ON(ret);
1793 }
1794 free_extent_buffer(path->nodes[*level]);
1795 path->nodes[*level] = NULL;
1796 *level = i + 1;
1797 }
1798 }
1799 return 1;
1800}
1801
1802/*
1803 * drop the reference count on the tree rooted at 'snap'. This traverses
1804 * the tree freeing any blocks that have a ref count of zero after being
1805 * decremented.
1806 */
1807static int walk_log_tree(struct btrfs_trans_handle *trans,
1808 struct btrfs_root *log, struct walk_control *wc)
1809{
1810 int ret = 0;
1811 int wret;
1812 int level;
1813 struct btrfs_path *path;
1814 int i;
1815 int orig_level;
1816
1817 path = btrfs_alloc_path();
1818 BUG_ON(!path);
1819
1820 level = btrfs_header_level(log->node);
1821 orig_level = level;
1822 path->nodes[level] = log->node;
1823 extent_buffer_get(log->node);
1824 path->slots[level] = 0;
1825
Chris Masond3977122009-01-05 21:25:51 -05001826 while (1) {
Chris Masone02119d2008-09-05 16:13:11 -04001827 wret = walk_down_log_tree(trans, log, path, &level, wc);
1828 if (wret > 0)
1829 break;
1830 if (wret < 0)
1831 ret = wret;
1832
1833 wret = walk_up_log_tree(trans, log, path, &level, wc);
1834 if (wret > 0)
1835 break;
1836 if (wret < 0)
1837 ret = wret;
1838 }
1839
1840 /* was the root node processed? if not, catch it here */
1841 if (path->nodes[orig_level]) {
1842 wc->process_func(log, path->nodes[orig_level], wc,
1843 btrfs_header_generation(path->nodes[orig_level]));
1844 if (wc->free) {
1845 struct extent_buffer *next;
1846
1847 next = path->nodes[orig_level];
1848
1849 btrfs_tree_lock(next);
1850 clean_tree_block(trans, log, next);
Chris Masonb4ce94d2009-02-04 09:25:08 -05001851 btrfs_set_lock_blocking(next);
Chris Masone02119d2008-09-05 16:13:11 -04001852 btrfs_wait_tree_block_writeback(next);
1853 btrfs_tree_unlock(next);
1854
Chris Masone02119d2008-09-05 16:13:11 -04001855 WARN_ON(log->root_key.objectid !=
1856 BTRFS_TREE_LOG_OBJECTID);
Chris Masond00aff02008-09-11 15:54:42 -04001857 ret = btrfs_free_reserved_extent(log, next->start,
1858 next->len);
Chris Masone02119d2008-09-05 16:13:11 -04001859 BUG_ON(ret);
1860 }
1861 }
1862
1863 for (i = 0; i <= orig_level; i++) {
1864 if (path->nodes[i]) {
1865 free_extent_buffer(path->nodes[i]);
1866 path->nodes[i] = NULL;
1867 }
1868 }
1869 btrfs_free_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04001870 return ret;
1871}
1872
Yan Zheng7237f182009-01-21 12:54:03 -05001873/*
1874 * helper function to update the item for a given subvolumes log root
1875 * in the tree of log roots
1876 */
1877static int update_log_root(struct btrfs_trans_handle *trans,
1878 struct btrfs_root *log)
1879{
1880 int ret;
1881
1882 if (log->log_transid == 1) {
1883 /* insert root item on the first sync */
1884 ret = btrfs_insert_root(trans, log->fs_info->log_root_tree,
1885 &log->root_key, &log->root_item);
1886 } else {
1887 ret = btrfs_update_root(trans, log->fs_info->log_root_tree,
1888 &log->root_key, &log->root_item);
1889 }
1890 return ret;
1891}
1892
Chris Mason12fcfd22009-03-24 10:24:20 -04001893static int wait_log_commit(struct btrfs_trans_handle *trans,
1894 struct btrfs_root *root, unsigned long transid)
Chris Masone02119d2008-09-05 16:13:11 -04001895{
1896 DEFINE_WAIT(wait);
Yan Zheng7237f182009-01-21 12:54:03 -05001897 int index = transid % 2;
Chris Masone02119d2008-09-05 16:13:11 -04001898
Yan Zheng7237f182009-01-21 12:54:03 -05001899 /*
1900 * we only allow two pending log transactions at a time,
1901 * so we know that if ours is more than 2 older than the
1902 * current transaction, we're done
1903 */
Chris Masone02119d2008-09-05 16:13:11 -04001904 do {
Yan Zheng7237f182009-01-21 12:54:03 -05001905 prepare_to_wait(&root->log_commit_wait[index],
1906 &wait, TASK_UNINTERRUPTIBLE);
1907 mutex_unlock(&root->log_mutex);
Chris Mason12fcfd22009-03-24 10:24:20 -04001908
1909 if (root->fs_info->last_trans_log_full_commit !=
1910 trans->transid && root->log_transid < transid + 2 &&
Yan Zheng7237f182009-01-21 12:54:03 -05001911 atomic_read(&root->log_commit[index]))
Chris Masone02119d2008-09-05 16:13:11 -04001912 schedule();
Chris Mason12fcfd22009-03-24 10:24:20 -04001913
Yan Zheng7237f182009-01-21 12:54:03 -05001914 finish_wait(&root->log_commit_wait[index], &wait);
1915 mutex_lock(&root->log_mutex);
1916 } while (root->log_transid < transid + 2 &&
1917 atomic_read(&root->log_commit[index]));
1918 return 0;
1919}
1920
Chris Mason12fcfd22009-03-24 10:24:20 -04001921static int wait_for_writer(struct btrfs_trans_handle *trans,
1922 struct btrfs_root *root)
Yan Zheng7237f182009-01-21 12:54:03 -05001923{
1924 DEFINE_WAIT(wait);
1925 while (atomic_read(&root->log_writers)) {
1926 prepare_to_wait(&root->log_writer_wait,
1927 &wait, TASK_UNINTERRUPTIBLE);
1928 mutex_unlock(&root->log_mutex);
Chris Mason12fcfd22009-03-24 10:24:20 -04001929 if (root->fs_info->last_trans_log_full_commit !=
1930 trans->transid && atomic_read(&root->log_writers))
Yan Zheng7237f182009-01-21 12:54:03 -05001931 schedule();
1932 mutex_lock(&root->log_mutex);
1933 finish_wait(&root->log_writer_wait, &wait);
1934 }
Chris Masone02119d2008-09-05 16:13:11 -04001935 return 0;
1936}
1937
1938/*
1939 * btrfs_sync_log does sends a given tree log down to the disk and
1940 * updates the super blocks to record it. When this call is done,
Chris Mason12fcfd22009-03-24 10:24:20 -04001941 * you know that any inodes previously logged are safely on disk only
1942 * if it returns 0.
1943 *
1944 * Any other return value means you need to call btrfs_commit_transaction.
1945 * Some of the edge cases for fsyncing directories that have had unlinks
1946 * or renames done in the past mean that sometimes the only safe
1947 * fsync is to commit the whole FS. When btrfs_sync_log returns -EAGAIN,
1948 * that has happened.
Chris Masone02119d2008-09-05 16:13:11 -04001949 */
1950int btrfs_sync_log(struct btrfs_trans_handle *trans,
1951 struct btrfs_root *root)
1952{
Yan Zheng7237f182009-01-21 12:54:03 -05001953 int index1;
1954 int index2;
Yan, Zheng8cef4e12009-11-12 09:33:26 +00001955 int mark;
Chris Masone02119d2008-09-05 16:13:11 -04001956 int ret;
Chris Masone02119d2008-09-05 16:13:11 -04001957 struct btrfs_root *log = root->log_root;
Yan Zheng7237f182009-01-21 12:54:03 -05001958 struct btrfs_root *log_root_tree = root->fs_info->log_root_tree;
Yan, Zheng8cef4e12009-11-12 09:33:26 +00001959 unsigned long log_transid = 0;
Chris Masone02119d2008-09-05 16:13:11 -04001960
Yan Zheng7237f182009-01-21 12:54:03 -05001961 mutex_lock(&root->log_mutex);
1962 index1 = root->log_transid % 2;
1963 if (atomic_read(&root->log_commit[index1])) {
Chris Mason12fcfd22009-03-24 10:24:20 -04001964 wait_log_commit(trans, root, root->log_transid);
Yan Zheng7237f182009-01-21 12:54:03 -05001965 mutex_unlock(&root->log_mutex);
1966 return 0;
Chris Masone02119d2008-09-05 16:13:11 -04001967 }
Yan Zheng7237f182009-01-21 12:54:03 -05001968 atomic_set(&root->log_commit[index1], 1);
1969
1970 /* wait for previous tree log sync to complete */
1971 if (atomic_read(&root->log_commit[(index1 + 1) % 2]))
Chris Mason12fcfd22009-03-24 10:24:20 -04001972 wait_log_commit(trans, root, root->log_transid - 1);
Chris Masone02119d2008-09-05 16:13:11 -04001973
Yan, Zheng86df7eb2009-10-14 09:24:59 -04001974 while (1) {
Yan Zheng7237f182009-01-21 12:54:03 -05001975 unsigned long batch = root->log_batch;
Yan, Zheng86df7eb2009-10-14 09:24:59 -04001976 if (root->log_multiple_pids) {
1977 mutex_unlock(&root->log_mutex);
1978 schedule_timeout_uninterruptible(1);
1979 mutex_lock(&root->log_mutex);
1980 }
Chris Mason12fcfd22009-03-24 10:24:20 -04001981 wait_for_writer(trans, root);
Yan Zheng7237f182009-01-21 12:54:03 -05001982 if (batch == root->log_batch)
Chris Masone02119d2008-09-05 16:13:11 -04001983 break;
1984 }
Chris Masond0c803c2008-09-11 16:17:57 -04001985
Chris Mason12fcfd22009-03-24 10:24:20 -04001986 /* bail out if we need to do a full commit */
1987 if (root->fs_info->last_trans_log_full_commit == trans->transid) {
1988 ret = -EAGAIN;
1989 mutex_unlock(&root->log_mutex);
1990 goto out;
1991 }
1992
Yan, Zheng8cef4e12009-11-12 09:33:26 +00001993 log_transid = root->log_transid;
1994 if (log_transid % 2 == 0)
1995 mark = EXTENT_DIRTY;
1996 else
1997 mark = EXTENT_NEW;
1998
Chris Mason690587d2009-10-13 13:29:19 -04001999 /* we start IO on all the marked extents here, but we don't actually
2000 * wait for them until later.
2001 */
Yan, Zheng8cef4e12009-11-12 09:33:26 +00002002 ret = btrfs_write_marked_extents(log, &log->dirty_log_pages, mark);
Chris Masone02119d2008-09-05 16:13:11 -04002003 BUG_ON(ret);
Yan Zheng7237f182009-01-21 12:54:03 -05002004
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002005 btrfs_set_root_node(&log->root_item, log->node);
Yan Zheng7237f182009-01-21 12:54:03 -05002006
2007 root->log_batch = 0;
2008 root->log_transid++;
2009 log->log_transid = root->log_transid;
Josef Bacikff782e02009-10-08 15:30:04 -04002010 root->log_start_pid = 0;
Yan Zheng7237f182009-01-21 12:54:03 -05002011 smp_mb();
2012 /*
Yan, Zheng8cef4e12009-11-12 09:33:26 +00002013 * IO has been started, blocks of the log tree have WRITTEN flag set
2014 * in their headers. new modifications of the log will be written to
2015 * new positions. so it's safe to allow log writers to go in.
Yan Zheng7237f182009-01-21 12:54:03 -05002016 */
2017 mutex_unlock(&root->log_mutex);
2018
2019 mutex_lock(&log_root_tree->log_mutex);
2020 log_root_tree->log_batch++;
2021 atomic_inc(&log_root_tree->log_writers);
2022 mutex_unlock(&log_root_tree->log_mutex);
2023
2024 ret = update_log_root(trans, log);
Yan Zheng7237f182009-01-21 12:54:03 -05002025
2026 mutex_lock(&log_root_tree->log_mutex);
2027 if (atomic_dec_and_test(&log_root_tree->log_writers)) {
2028 smp_mb();
2029 if (waitqueue_active(&log_root_tree->log_writer_wait))
2030 wake_up(&log_root_tree->log_writer_wait);
2031 }
2032
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002033 if (ret) {
2034 BUG_ON(ret != -ENOSPC);
2035 root->fs_info->last_trans_log_full_commit = trans->transid;
2036 btrfs_wait_marked_extents(log, &log->dirty_log_pages, mark);
2037 mutex_unlock(&log_root_tree->log_mutex);
2038 ret = -EAGAIN;
2039 goto out;
2040 }
2041
Yan Zheng7237f182009-01-21 12:54:03 -05002042 index2 = log_root_tree->log_transid % 2;
2043 if (atomic_read(&log_root_tree->log_commit[index2])) {
Yan, Zheng8cef4e12009-11-12 09:33:26 +00002044 btrfs_wait_marked_extents(log, &log->dirty_log_pages, mark);
Chris Mason12fcfd22009-03-24 10:24:20 -04002045 wait_log_commit(trans, log_root_tree,
2046 log_root_tree->log_transid);
Yan Zheng7237f182009-01-21 12:54:03 -05002047 mutex_unlock(&log_root_tree->log_mutex);
2048 goto out;
2049 }
2050 atomic_set(&log_root_tree->log_commit[index2], 1);
2051
Chris Mason12fcfd22009-03-24 10:24:20 -04002052 if (atomic_read(&log_root_tree->log_commit[(index2 + 1) % 2])) {
2053 wait_log_commit(trans, log_root_tree,
2054 log_root_tree->log_transid - 1);
2055 }
Yan Zheng7237f182009-01-21 12:54:03 -05002056
Chris Mason12fcfd22009-03-24 10:24:20 -04002057 wait_for_writer(trans, log_root_tree);
2058
2059 /*
2060 * now that we've moved on to the tree of log tree roots,
2061 * check the full commit flag again
2062 */
2063 if (root->fs_info->last_trans_log_full_commit == trans->transid) {
Yan, Zheng8cef4e12009-11-12 09:33:26 +00002064 btrfs_wait_marked_extents(log, &log->dirty_log_pages, mark);
Chris Mason12fcfd22009-03-24 10:24:20 -04002065 mutex_unlock(&log_root_tree->log_mutex);
2066 ret = -EAGAIN;
2067 goto out_wake_log_root;
2068 }
Yan Zheng7237f182009-01-21 12:54:03 -05002069
2070 ret = btrfs_write_and_wait_marked_extents(log_root_tree,
Yan, Zheng8cef4e12009-11-12 09:33:26 +00002071 &log_root_tree->dirty_log_pages,
2072 EXTENT_DIRTY | EXTENT_NEW);
Chris Masone02119d2008-09-05 16:13:11 -04002073 BUG_ON(ret);
Yan, Zheng8cef4e12009-11-12 09:33:26 +00002074 btrfs_wait_marked_extents(log, &log->dirty_log_pages, mark);
Chris Masone02119d2008-09-05 16:13:11 -04002075
2076 btrfs_set_super_log_root(&root->fs_info->super_for_commit,
Yan Zheng7237f182009-01-21 12:54:03 -05002077 log_root_tree->node->start);
Chris Masone02119d2008-09-05 16:13:11 -04002078 btrfs_set_super_log_root_level(&root->fs_info->super_for_commit,
Yan Zheng7237f182009-01-21 12:54:03 -05002079 btrfs_header_level(log_root_tree->node));
Chris Masone02119d2008-09-05 16:13:11 -04002080
Yan Zheng7237f182009-01-21 12:54:03 -05002081 log_root_tree->log_batch = 0;
2082 log_root_tree->log_transid++;
Chris Masone02119d2008-09-05 16:13:11 -04002083 smp_mb();
Yan Zheng7237f182009-01-21 12:54:03 -05002084
2085 mutex_unlock(&log_root_tree->log_mutex);
2086
2087 /*
2088 * nobody else is going to jump in and write the the ctree
2089 * super here because the log_commit atomic below is protecting
2090 * us. We must be called with a transaction handle pinning
2091 * the running transaction open, so a full commit can't hop
2092 * in and cause problems either.
2093 */
Chris Mason47226072009-10-13 12:55:09 -04002094 write_ctree_super(trans, root->fs_info->tree_root, 1);
Chris Mason12fcfd22009-03-24 10:24:20 -04002095 ret = 0;
Yan Zheng7237f182009-01-21 12:54:03 -05002096
Chris Mason257c62e2009-10-13 13:21:08 -04002097 mutex_lock(&root->log_mutex);
2098 if (root->last_log_commit < log_transid)
2099 root->last_log_commit = log_transid;
2100 mutex_unlock(&root->log_mutex);
2101
Chris Mason12fcfd22009-03-24 10:24:20 -04002102out_wake_log_root:
Yan Zheng7237f182009-01-21 12:54:03 -05002103 atomic_set(&log_root_tree->log_commit[index2], 0);
2104 smp_mb();
2105 if (waitqueue_active(&log_root_tree->log_commit_wait[index2]))
2106 wake_up(&log_root_tree->log_commit_wait[index2]);
Chris Masone02119d2008-09-05 16:13:11 -04002107out:
Yan Zheng7237f182009-01-21 12:54:03 -05002108 atomic_set(&root->log_commit[index1], 0);
2109 smp_mb();
2110 if (waitqueue_active(&root->log_commit_wait[index1]))
2111 wake_up(&root->log_commit_wait[index1]);
Chris Masone02119d2008-09-05 16:13:11 -04002112 return 0;
Chris Masone02119d2008-09-05 16:13:11 -04002113}
2114
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002115static void free_log_tree(struct btrfs_trans_handle *trans,
2116 struct btrfs_root *log)
Chris Masone02119d2008-09-05 16:13:11 -04002117{
2118 int ret;
Chris Masond0c803c2008-09-11 16:17:57 -04002119 u64 start;
2120 u64 end;
Chris Masone02119d2008-09-05 16:13:11 -04002121 struct walk_control wc = {
2122 .free = 1,
2123 .process_func = process_one_buffer
2124 };
2125
Chris Masone02119d2008-09-05 16:13:11 -04002126 ret = walk_log_tree(trans, log, &wc);
2127 BUG_ON(ret);
2128
Chris Masond3977122009-01-05 21:25:51 -05002129 while (1) {
Chris Masond0c803c2008-09-11 16:17:57 -04002130 ret = find_first_extent_bit(&log->dirty_log_pages,
Yan, Zheng8cef4e12009-11-12 09:33:26 +00002131 0, &start, &end, EXTENT_DIRTY | EXTENT_NEW);
Chris Masond0c803c2008-09-11 16:17:57 -04002132 if (ret)
2133 break;
2134
Yan, Zheng8cef4e12009-11-12 09:33:26 +00002135 clear_extent_bits(&log->dirty_log_pages, start, end,
2136 EXTENT_DIRTY | EXTENT_NEW, GFP_NOFS);
Chris Masond0c803c2008-09-11 16:17:57 -04002137 }
2138
Yan Zheng7237f182009-01-21 12:54:03 -05002139 free_extent_buffer(log->node);
2140 kfree(log);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002141}
2142
2143/*
2144 * free all the extents used by the tree log. This should be called
2145 * at commit time of the full transaction
2146 */
2147int btrfs_free_log(struct btrfs_trans_handle *trans, struct btrfs_root *root)
2148{
2149 if (root->log_root) {
2150 free_log_tree(trans, root->log_root);
2151 root->log_root = NULL;
2152 }
2153 return 0;
2154}
2155
2156int btrfs_free_log_root_tree(struct btrfs_trans_handle *trans,
2157 struct btrfs_fs_info *fs_info)
2158{
2159 if (fs_info->log_root_tree) {
2160 free_log_tree(trans, fs_info->log_root_tree);
2161 fs_info->log_root_tree = NULL;
2162 }
Chris Masone02119d2008-09-05 16:13:11 -04002163 return 0;
2164}
2165
2166/*
Chris Masone02119d2008-09-05 16:13:11 -04002167 * If both a file and directory are logged, and unlinks or renames are
2168 * mixed in, we have a few interesting corners:
2169 *
2170 * create file X in dir Y
2171 * link file X to X.link in dir Y
2172 * fsync file X
2173 * unlink file X but leave X.link
2174 * fsync dir Y
2175 *
2176 * After a crash we would expect only X.link to exist. But file X
2177 * didn't get fsync'd again so the log has back refs for X and X.link.
2178 *
2179 * We solve this by removing directory entries and inode backrefs from the
2180 * log when a file that was logged in the current transaction is
2181 * unlinked. Any later fsync will include the updated log entries, and
2182 * we'll be able to reconstruct the proper directory items from backrefs.
2183 *
2184 * This optimizations allows us to avoid relogging the entire inode
2185 * or the entire directory.
2186 */
2187int btrfs_del_dir_entries_in_log(struct btrfs_trans_handle *trans,
2188 struct btrfs_root *root,
2189 const char *name, int name_len,
2190 struct inode *dir, u64 index)
2191{
2192 struct btrfs_root *log;
2193 struct btrfs_dir_item *di;
2194 struct btrfs_path *path;
2195 int ret;
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002196 int err = 0;
Chris Masone02119d2008-09-05 16:13:11 -04002197 int bytes_del = 0;
2198
Chris Mason3a5f1d42008-09-11 15:53:37 -04002199 if (BTRFS_I(dir)->logged_trans < trans->transid)
2200 return 0;
2201
Chris Masone02119d2008-09-05 16:13:11 -04002202 ret = join_running_log_trans(root);
2203 if (ret)
2204 return 0;
2205
2206 mutex_lock(&BTRFS_I(dir)->log_mutex);
2207
2208 log = root->log_root;
2209 path = btrfs_alloc_path();
2210 di = btrfs_lookup_dir_item(trans, log, path, dir->i_ino,
2211 name, name_len, -1);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002212 if (IS_ERR(di)) {
2213 err = PTR_ERR(di);
2214 goto fail;
2215 }
2216 if (di) {
Chris Masone02119d2008-09-05 16:13:11 -04002217 ret = btrfs_delete_one_dir_name(trans, log, path, di);
2218 bytes_del += name_len;
2219 BUG_ON(ret);
2220 }
2221 btrfs_release_path(log, path);
2222 di = btrfs_lookup_dir_index_item(trans, log, path, dir->i_ino,
2223 index, name, name_len, -1);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002224 if (IS_ERR(di)) {
2225 err = PTR_ERR(di);
2226 goto fail;
2227 }
2228 if (di) {
Chris Masone02119d2008-09-05 16:13:11 -04002229 ret = btrfs_delete_one_dir_name(trans, log, path, di);
2230 bytes_del += name_len;
2231 BUG_ON(ret);
2232 }
2233
2234 /* update the directory size in the log to reflect the names
2235 * we have removed
2236 */
2237 if (bytes_del) {
2238 struct btrfs_key key;
2239
2240 key.objectid = dir->i_ino;
2241 key.offset = 0;
2242 key.type = BTRFS_INODE_ITEM_KEY;
2243 btrfs_release_path(log, path);
2244
2245 ret = btrfs_search_slot(trans, log, &key, path, 0, 1);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002246 if (ret < 0) {
2247 err = ret;
2248 goto fail;
2249 }
Chris Masone02119d2008-09-05 16:13:11 -04002250 if (ret == 0) {
2251 struct btrfs_inode_item *item;
2252 u64 i_size;
2253
2254 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2255 struct btrfs_inode_item);
2256 i_size = btrfs_inode_size(path->nodes[0], item);
2257 if (i_size > bytes_del)
2258 i_size -= bytes_del;
2259 else
2260 i_size = 0;
2261 btrfs_set_inode_size(path->nodes[0], item, i_size);
2262 btrfs_mark_buffer_dirty(path->nodes[0]);
2263 } else
2264 ret = 0;
2265 btrfs_release_path(log, path);
2266 }
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002267fail:
Chris Masone02119d2008-09-05 16:13:11 -04002268 btrfs_free_path(path);
2269 mutex_unlock(&BTRFS_I(dir)->log_mutex);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002270 if (ret == -ENOSPC) {
2271 root->fs_info->last_trans_log_full_commit = trans->transid;
2272 ret = 0;
2273 }
Chris Mason12fcfd22009-03-24 10:24:20 -04002274 btrfs_end_log_trans(root);
Chris Masone02119d2008-09-05 16:13:11 -04002275
2276 return 0;
2277}
2278
2279/* see comments for btrfs_del_dir_entries_in_log */
2280int btrfs_del_inode_ref_in_log(struct btrfs_trans_handle *trans,
2281 struct btrfs_root *root,
2282 const char *name, int name_len,
2283 struct inode *inode, u64 dirid)
2284{
2285 struct btrfs_root *log;
2286 u64 index;
2287 int ret;
2288
Chris Mason3a5f1d42008-09-11 15:53:37 -04002289 if (BTRFS_I(inode)->logged_trans < trans->transid)
2290 return 0;
2291
Chris Masone02119d2008-09-05 16:13:11 -04002292 ret = join_running_log_trans(root);
2293 if (ret)
2294 return 0;
2295 log = root->log_root;
2296 mutex_lock(&BTRFS_I(inode)->log_mutex);
2297
2298 ret = btrfs_del_inode_ref(trans, log, name, name_len, inode->i_ino,
2299 dirid, &index);
2300 mutex_unlock(&BTRFS_I(inode)->log_mutex);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002301 if (ret == -ENOSPC) {
2302 root->fs_info->last_trans_log_full_commit = trans->transid;
2303 ret = 0;
2304 }
Chris Mason12fcfd22009-03-24 10:24:20 -04002305 btrfs_end_log_trans(root);
Chris Masone02119d2008-09-05 16:13:11 -04002306
Chris Masone02119d2008-09-05 16:13:11 -04002307 return ret;
2308}
2309
2310/*
2311 * creates a range item in the log for 'dirid'. first_offset and
2312 * last_offset tell us which parts of the key space the log should
2313 * be considered authoritative for.
2314 */
2315static noinline int insert_dir_log_key(struct btrfs_trans_handle *trans,
2316 struct btrfs_root *log,
2317 struct btrfs_path *path,
2318 int key_type, u64 dirid,
2319 u64 first_offset, u64 last_offset)
2320{
2321 int ret;
2322 struct btrfs_key key;
2323 struct btrfs_dir_log_item *item;
2324
2325 key.objectid = dirid;
2326 key.offset = first_offset;
2327 if (key_type == BTRFS_DIR_ITEM_KEY)
2328 key.type = BTRFS_DIR_LOG_ITEM_KEY;
2329 else
2330 key.type = BTRFS_DIR_LOG_INDEX_KEY;
2331 ret = btrfs_insert_empty_item(trans, log, path, &key, sizeof(*item));
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002332 if (ret)
2333 return ret;
Chris Masone02119d2008-09-05 16:13:11 -04002334
2335 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2336 struct btrfs_dir_log_item);
2337 btrfs_set_dir_log_end(path->nodes[0], item, last_offset);
2338 btrfs_mark_buffer_dirty(path->nodes[0]);
2339 btrfs_release_path(log, path);
2340 return 0;
2341}
2342
2343/*
2344 * log all the items included in the current transaction for a given
2345 * directory. This also creates the range items in the log tree required
2346 * to replay anything deleted before the fsync
2347 */
2348static noinline int log_dir_items(struct btrfs_trans_handle *trans,
2349 struct btrfs_root *root, struct inode *inode,
2350 struct btrfs_path *path,
2351 struct btrfs_path *dst_path, int key_type,
2352 u64 min_offset, u64 *last_offset_ret)
2353{
2354 struct btrfs_key min_key;
2355 struct btrfs_key max_key;
2356 struct btrfs_root *log = root->log_root;
2357 struct extent_buffer *src;
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002358 int err = 0;
Chris Masone02119d2008-09-05 16:13:11 -04002359 int ret;
2360 int i;
2361 int nritems;
2362 u64 first_offset = min_offset;
2363 u64 last_offset = (u64)-1;
2364
2365 log = root->log_root;
2366 max_key.objectid = inode->i_ino;
2367 max_key.offset = (u64)-1;
2368 max_key.type = key_type;
2369
2370 min_key.objectid = inode->i_ino;
2371 min_key.type = key_type;
2372 min_key.offset = min_offset;
2373
2374 path->keep_locks = 1;
2375
2376 ret = btrfs_search_forward(root, &min_key, &max_key,
2377 path, 0, trans->transid);
2378
2379 /*
2380 * we didn't find anything from this transaction, see if there
2381 * is anything at all
2382 */
2383 if (ret != 0 || min_key.objectid != inode->i_ino ||
2384 min_key.type != key_type) {
2385 min_key.objectid = inode->i_ino;
2386 min_key.type = key_type;
2387 min_key.offset = (u64)-1;
2388 btrfs_release_path(root, path);
2389 ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0);
2390 if (ret < 0) {
2391 btrfs_release_path(root, path);
2392 return ret;
2393 }
2394 ret = btrfs_previous_item(root, path, inode->i_ino, key_type);
2395
2396 /* if ret == 0 there are items for this type,
2397 * create a range to tell us the last key of this type.
2398 * otherwise, there are no items in this directory after
2399 * *min_offset, and we create a range to indicate that.
2400 */
2401 if (ret == 0) {
2402 struct btrfs_key tmp;
2403 btrfs_item_key_to_cpu(path->nodes[0], &tmp,
2404 path->slots[0]);
Chris Masond3977122009-01-05 21:25:51 -05002405 if (key_type == tmp.type)
Chris Masone02119d2008-09-05 16:13:11 -04002406 first_offset = max(min_offset, tmp.offset) + 1;
Chris Masone02119d2008-09-05 16:13:11 -04002407 }
2408 goto done;
2409 }
2410
2411 /* go backward to find any previous key */
2412 ret = btrfs_previous_item(root, path, inode->i_ino, key_type);
2413 if (ret == 0) {
2414 struct btrfs_key tmp;
2415 btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]);
2416 if (key_type == tmp.type) {
2417 first_offset = tmp.offset;
2418 ret = overwrite_item(trans, log, dst_path,
2419 path->nodes[0], path->slots[0],
2420 &tmp);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002421 if (ret) {
2422 err = ret;
2423 goto done;
2424 }
Chris Masone02119d2008-09-05 16:13:11 -04002425 }
2426 }
2427 btrfs_release_path(root, path);
2428
2429 /* find the first key from this transaction again */
2430 ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0);
2431 if (ret != 0) {
2432 WARN_ON(1);
2433 goto done;
2434 }
2435
2436 /*
2437 * we have a block from this transaction, log every item in it
2438 * from our directory
2439 */
Chris Masond3977122009-01-05 21:25:51 -05002440 while (1) {
Chris Masone02119d2008-09-05 16:13:11 -04002441 struct btrfs_key tmp;
2442 src = path->nodes[0];
2443 nritems = btrfs_header_nritems(src);
2444 for (i = path->slots[0]; i < nritems; i++) {
2445 btrfs_item_key_to_cpu(src, &min_key, i);
2446
2447 if (min_key.objectid != inode->i_ino ||
2448 min_key.type != key_type)
2449 goto done;
2450 ret = overwrite_item(trans, log, dst_path, src, i,
2451 &min_key);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002452 if (ret) {
2453 err = ret;
2454 goto done;
2455 }
Chris Masone02119d2008-09-05 16:13:11 -04002456 }
2457 path->slots[0] = nritems;
2458
2459 /*
2460 * look ahead to the next item and see if it is also
2461 * from this directory and from this transaction
2462 */
2463 ret = btrfs_next_leaf(root, path);
2464 if (ret == 1) {
2465 last_offset = (u64)-1;
2466 goto done;
2467 }
2468 btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]);
2469 if (tmp.objectid != inode->i_ino || tmp.type != key_type) {
2470 last_offset = (u64)-1;
2471 goto done;
2472 }
2473 if (btrfs_header_generation(path->nodes[0]) != trans->transid) {
2474 ret = overwrite_item(trans, log, dst_path,
2475 path->nodes[0], path->slots[0],
2476 &tmp);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002477 if (ret)
2478 err = ret;
2479 else
2480 last_offset = tmp.offset;
Chris Masone02119d2008-09-05 16:13:11 -04002481 goto done;
2482 }
2483 }
2484done:
Chris Masone02119d2008-09-05 16:13:11 -04002485 btrfs_release_path(root, path);
2486 btrfs_release_path(log, dst_path);
2487
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002488 if (err == 0) {
2489 *last_offset_ret = last_offset;
2490 /*
2491 * insert the log range keys to indicate where the log
2492 * is valid
2493 */
2494 ret = insert_dir_log_key(trans, log, path, key_type,
2495 inode->i_ino, first_offset,
2496 last_offset);
2497 if (ret)
2498 err = ret;
2499 }
2500 return err;
Chris Masone02119d2008-09-05 16:13:11 -04002501}
2502
2503/*
2504 * logging directories is very similar to logging inodes, We find all the items
2505 * from the current transaction and write them to the log.
2506 *
2507 * The recovery code scans the directory in the subvolume, and if it finds a
2508 * key in the range logged that is not present in the log tree, then it means
2509 * that dir entry was unlinked during the transaction.
2510 *
2511 * In order for that scan to work, we must include one key smaller than
2512 * the smallest logged by this transaction and one key larger than the largest
2513 * key logged by this transaction.
2514 */
2515static noinline int log_directory_changes(struct btrfs_trans_handle *trans,
2516 struct btrfs_root *root, struct inode *inode,
2517 struct btrfs_path *path,
2518 struct btrfs_path *dst_path)
2519{
2520 u64 min_key;
2521 u64 max_key;
2522 int ret;
2523 int key_type = BTRFS_DIR_ITEM_KEY;
2524
2525again:
2526 min_key = 0;
2527 max_key = 0;
Chris Masond3977122009-01-05 21:25:51 -05002528 while (1) {
Chris Masone02119d2008-09-05 16:13:11 -04002529 ret = log_dir_items(trans, root, inode, path,
2530 dst_path, key_type, min_key,
2531 &max_key);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002532 if (ret)
2533 return ret;
Chris Masone02119d2008-09-05 16:13:11 -04002534 if (max_key == (u64)-1)
2535 break;
2536 min_key = max_key + 1;
2537 }
2538
2539 if (key_type == BTRFS_DIR_ITEM_KEY) {
2540 key_type = BTRFS_DIR_INDEX_KEY;
2541 goto again;
2542 }
2543 return 0;
2544}
2545
2546/*
2547 * a helper function to drop items from the log before we relog an
2548 * inode. max_key_type indicates the highest item type to remove.
2549 * This cannot be run for file data extents because it does not
2550 * free the extents they point to.
2551 */
2552static int drop_objectid_items(struct btrfs_trans_handle *trans,
2553 struct btrfs_root *log,
2554 struct btrfs_path *path,
2555 u64 objectid, int max_key_type)
2556{
2557 int ret;
2558 struct btrfs_key key;
2559 struct btrfs_key found_key;
2560
2561 key.objectid = objectid;
2562 key.type = max_key_type;
2563 key.offset = (u64)-1;
2564
Chris Masond3977122009-01-05 21:25:51 -05002565 while (1) {
Chris Masone02119d2008-09-05 16:13:11 -04002566 ret = btrfs_search_slot(trans, log, &key, path, -1, 1);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002567 BUG_ON(ret == 0);
2568 if (ret < 0)
Chris Masone02119d2008-09-05 16:13:11 -04002569 break;
2570
2571 if (path->slots[0] == 0)
2572 break;
2573
2574 path->slots[0]--;
2575 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2576 path->slots[0]);
2577
2578 if (found_key.objectid != objectid)
2579 break;
2580
2581 ret = btrfs_del_item(trans, log, path);
2582 BUG_ON(ret);
2583 btrfs_release_path(log, path);
2584 }
2585 btrfs_release_path(log, path);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002586 return ret;
Chris Masone02119d2008-09-05 16:13:11 -04002587}
2588
Chris Mason31ff1cd2008-09-11 16:17:57 -04002589static noinline int copy_items(struct btrfs_trans_handle *trans,
2590 struct btrfs_root *log,
2591 struct btrfs_path *dst_path,
2592 struct extent_buffer *src,
2593 int start_slot, int nr, int inode_only)
2594{
2595 unsigned long src_offset;
2596 unsigned long dst_offset;
2597 struct btrfs_file_extent_item *extent;
2598 struct btrfs_inode_item *inode_item;
2599 int ret;
2600 struct btrfs_key *ins_keys;
2601 u32 *ins_sizes;
2602 char *ins_data;
2603 int i;
Chris Masond20f7042008-12-08 16:58:54 -05002604 struct list_head ordered_sums;
2605
2606 INIT_LIST_HEAD(&ordered_sums);
Chris Mason31ff1cd2008-09-11 16:17:57 -04002607
2608 ins_data = kmalloc(nr * sizeof(struct btrfs_key) +
2609 nr * sizeof(u32), GFP_NOFS);
2610 ins_sizes = (u32 *)ins_data;
2611 ins_keys = (struct btrfs_key *)(ins_data + nr * sizeof(u32));
2612
2613 for (i = 0; i < nr; i++) {
2614 ins_sizes[i] = btrfs_item_size_nr(src, i + start_slot);
2615 btrfs_item_key_to_cpu(src, ins_keys + i, i + start_slot);
2616 }
2617 ret = btrfs_insert_empty_items(trans, log, dst_path,
2618 ins_keys, ins_sizes, nr);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002619 if (ret) {
2620 kfree(ins_data);
2621 return ret;
2622 }
Chris Mason31ff1cd2008-09-11 16:17:57 -04002623
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002624 for (i = 0; i < nr; i++, dst_path->slots[0]++) {
Chris Mason31ff1cd2008-09-11 16:17:57 -04002625 dst_offset = btrfs_item_ptr_offset(dst_path->nodes[0],
2626 dst_path->slots[0]);
2627
2628 src_offset = btrfs_item_ptr_offset(src, start_slot + i);
2629
2630 copy_extent_buffer(dst_path->nodes[0], src, dst_offset,
2631 src_offset, ins_sizes[i]);
2632
2633 if (inode_only == LOG_INODE_EXISTS &&
2634 ins_keys[i].type == BTRFS_INODE_ITEM_KEY) {
2635 inode_item = btrfs_item_ptr(dst_path->nodes[0],
2636 dst_path->slots[0],
2637 struct btrfs_inode_item);
2638 btrfs_set_inode_size(dst_path->nodes[0], inode_item, 0);
2639
2640 /* set the generation to zero so the recover code
2641 * can tell the difference between an logging
2642 * just to say 'this inode exists' and a logging
2643 * to say 'update this inode with these values'
2644 */
2645 btrfs_set_inode_generation(dst_path->nodes[0],
2646 inode_item, 0);
2647 }
2648 /* take a reference on file data extents so that truncates
2649 * or deletes of this inode don't have to relog the inode
2650 * again
2651 */
2652 if (btrfs_key_type(ins_keys + i) == BTRFS_EXTENT_DATA_KEY) {
2653 int found_type;
2654 extent = btrfs_item_ptr(src, start_slot + i,
2655 struct btrfs_file_extent_item);
2656
2657 found_type = btrfs_file_extent_type(src, extent);
Yan Zhengd899e052008-10-30 14:25:28 -04002658 if (found_type == BTRFS_FILE_EXTENT_REG ||
2659 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002660 u64 ds, dl, cs, cl;
2661 ds = btrfs_file_extent_disk_bytenr(src,
2662 extent);
2663 /* ds == 0 is a hole */
2664 if (ds == 0)
2665 continue;
2666
2667 dl = btrfs_file_extent_disk_num_bytes(src,
2668 extent);
2669 cs = btrfs_file_extent_offset(src, extent);
2670 cl = btrfs_file_extent_num_bytes(src,
Joe Perchesa419aef2009-08-18 11:18:35 -07002671 extent);
Chris Mason580afd72008-12-08 19:15:39 -05002672 if (btrfs_file_extent_compression(src,
2673 extent)) {
2674 cs = 0;
2675 cl = dl;
2676 }
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002677
2678 ret = btrfs_lookup_csums_range(
2679 log->fs_info->csum_root,
2680 ds + cs, ds + cs + cl - 1,
2681 &ordered_sums);
2682 BUG_ON(ret);
Chris Mason31ff1cd2008-09-11 16:17:57 -04002683 }
2684 }
Chris Mason31ff1cd2008-09-11 16:17:57 -04002685 }
2686
2687 btrfs_mark_buffer_dirty(dst_path->nodes[0]);
2688 btrfs_release_path(log, dst_path);
2689 kfree(ins_data);
Chris Masond20f7042008-12-08 16:58:54 -05002690
2691 /*
2692 * we have to do this after the loop above to avoid changing the
2693 * log tree while trying to change the log tree.
2694 */
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002695 ret = 0;
Chris Masond3977122009-01-05 21:25:51 -05002696 while (!list_empty(&ordered_sums)) {
Chris Masond20f7042008-12-08 16:58:54 -05002697 struct btrfs_ordered_sum *sums = list_entry(ordered_sums.next,
2698 struct btrfs_ordered_sum,
2699 list);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002700 if (!ret)
2701 ret = btrfs_csum_file_blocks(trans, log, sums);
Chris Masond20f7042008-12-08 16:58:54 -05002702 list_del(&sums->list);
2703 kfree(sums);
2704 }
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002705 return ret;
Chris Mason31ff1cd2008-09-11 16:17:57 -04002706}
2707
Chris Masone02119d2008-09-05 16:13:11 -04002708/* log a single inode in the tree log.
2709 * At least one parent directory for this inode must exist in the tree
2710 * or be logged already.
2711 *
2712 * Any items from this inode changed by the current transaction are copied
2713 * to the log tree. An extra reference is taken on any extents in this
2714 * file, allowing us to avoid a whole pile of corner cases around logging
2715 * blocks that have been removed from the tree.
2716 *
2717 * See LOG_INODE_ALL and related defines for a description of what inode_only
2718 * does.
2719 *
2720 * This handles both files and directories.
2721 */
Chris Mason12fcfd22009-03-24 10:24:20 -04002722static int btrfs_log_inode(struct btrfs_trans_handle *trans,
Chris Masone02119d2008-09-05 16:13:11 -04002723 struct btrfs_root *root, struct inode *inode,
2724 int inode_only)
2725{
2726 struct btrfs_path *path;
2727 struct btrfs_path *dst_path;
2728 struct btrfs_key min_key;
2729 struct btrfs_key max_key;
2730 struct btrfs_root *log = root->log_root;
Chris Mason31ff1cd2008-09-11 16:17:57 -04002731 struct extent_buffer *src = NULL;
Chris Masone02119d2008-09-05 16:13:11 -04002732 u32 size;
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002733 int err = 0;
Chris Masone02119d2008-09-05 16:13:11 -04002734 int ret;
Chris Mason3a5f1d42008-09-11 15:53:37 -04002735 int nritems;
Chris Mason31ff1cd2008-09-11 16:17:57 -04002736 int ins_start_slot = 0;
2737 int ins_nr;
Chris Masone02119d2008-09-05 16:13:11 -04002738
2739 log = root->log_root;
2740
2741 path = btrfs_alloc_path();
2742 dst_path = btrfs_alloc_path();
2743
2744 min_key.objectid = inode->i_ino;
2745 min_key.type = BTRFS_INODE_ITEM_KEY;
2746 min_key.offset = 0;
2747
2748 max_key.objectid = inode->i_ino;
Chris Mason12fcfd22009-03-24 10:24:20 -04002749
2750 /* today the code can only do partial logging of directories */
2751 if (!S_ISDIR(inode->i_mode))
2752 inode_only = LOG_INODE_ALL;
2753
Chris Masone02119d2008-09-05 16:13:11 -04002754 if (inode_only == LOG_INODE_EXISTS || S_ISDIR(inode->i_mode))
2755 max_key.type = BTRFS_XATTR_ITEM_KEY;
2756 else
2757 max_key.type = (u8)-1;
2758 max_key.offset = (u64)-1;
2759
Chris Masone02119d2008-09-05 16:13:11 -04002760 mutex_lock(&BTRFS_I(inode)->log_mutex);
2761
2762 /*
2763 * a brute force approach to making sure we get the most uptodate
2764 * copies of everything.
2765 */
2766 if (S_ISDIR(inode->i_mode)) {
2767 int max_key_type = BTRFS_DIR_LOG_INDEX_KEY;
2768
2769 if (inode_only == LOG_INODE_EXISTS)
2770 max_key_type = BTRFS_XATTR_ITEM_KEY;
2771 ret = drop_objectid_items(trans, log, path,
2772 inode->i_ino, max_key_type);
2773 } else {
2774 ret = btrfs_truncate_inode_items(trans, log, inode, 0, 0);
2775 }
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002776 if (ret) {
2777 err = ret;
2778 goto out_unlock;
2779 }
Chris Masone02119d2008-09-05 16:13:11 -04002780 path->keep_locks = 1;
2781
Chris Masond3977122009-01-05 21:25:51 -05002782 while (1) {
Chris Mason31ff1cd2008-09-11 16:17:57 -04002783 ins_nr = 0;
Chris Masone02119d2008-09-05 16:13:11 -04002784 ret = btrfs_search_forward(root, &min_key, &max_key,
2785 path, 0, trans->transid);
2786 if (ret != 0)
2787 break;
Chris Mason3a5f1d42008-09-11 15:53:37 -04002788again:
Chris Mason31ff1cd2008-09-11 16:17:57 -04002789 /* note, ins_nr might be > 0 here, cleanup outside the loop */
Chris Masone02119d2008-09-05 16:13:11 -04002790 if (min_key.objectid != inode->i_ino)
2791 break;
2792 if (min_key.type > max_key.type)
2793 break;
Chris Mason31ff1cd2008-09-11 16:17:57 -04002794
Chris Masone02119d2008-09-05 16:13:11 -04002795 src = path->nodes[0];
2796 size = btrfs_item_size_nr(src, path->slots[0]);
Chris Mason31ff1cd2008-09-11 16:17:57 -04002797 if (ins_nr && ins_start_slot + ins_nr == path->slots[0]) {
2798 ins_nr++;
2799 goto next_slot;
2800 } else if (!ins_nr) {
2801 ins_start_slot = path->slots[0];
2802 ins_nr = 1;
2803 goto next_slot;
Chris Masone02119d2008-09-05 16:13:11 -04002804 }
2805
Chris Mason31ff1cd2008-09-11 16:17:57 -04002806 ret = copy_items(trans, log, dst_path, src, ins_start_slot,
2807 ins_nr, inode_only);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002808 if (ret) {
2809 err = ret;
2810 goto out_unlock;
2811 }
Chris Mason31ff1cd2008-09-11 16:17:57 -04002812 ins_nr = 1;
2813 ins_start_slot = path->slots[0];
2814next_slot:
Chris Masone02119d2008-09-05 16:13:11 -04002815
Chris Mason3a5f1d42008-09-11 15:53:37 -04002816 nritems = btrfs_header_nritems(path->nodes[0]);
2817 path->slots[0]++;
2818 if (path->slots[0] < nritems) {
2819 btrfs_item_key_to_cpu(path->nodes[0], &min_key,
2820 path->slots[0]);
2821 goto again;
2822 }
Chris Mason31ff1cd2008-09-11 16:17:57 -04002823 if (ins_nr) {
2824 ret = copy_items(trans, log, dst_path, src,
2825 ins_start_slot,
2826 ins_nr, inode_only);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002827 if (ret) {
2828 err = ret;
2829 goto out_unlock;
2830 }
Chris Mason31ff1cd2008-09-11 16:17:57 -04002831 ins_nr = 0;
2832 }
Chris Mason3a5f1d42008-09-11 15:53:37 -04002833 btrfs_release_path(root, path);
2834
Chris Masone02119d2008-09-05 16:13:11 -04002835 if (min_key.offset < (u64)-1)
2836 min_key.offset++;
2837 else if (min_key.type < (u8)-1)
2838 min_key.type++;
2839 else if (min_key.objectid < (u64)-1)
2840 min_key.objectid++;
2841 else
2842 break;
2843 }
Chris Mason31ff1cd2008-09-11 16:17:57 -04002844 if (ins_nr) {
2845 ret = copy_items(trans, log, dst_path, src,
2846 ins_start_slot,
2847 ins_nr, inode_only);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002848 if (ret) {
2849 err = ret;
2850 goto out_unlock;
2851 }
Chris Mason31ff1cd2008-09-11 16:17:57 -04002852 ins_nr = 0;
2853 }
2854 WARN_ON(ins_nr);
Chris Mason9623f9a2008-09-11 17:42:42 -04002855 if (inode_only == LOG_INODE_ALL && S_ISDIR(inode->i_mode)) {
Chris Masone02119d2008-09-05 16:13:11 -04002856 btrfs_release_path(root, path);
2857 btrfs_release_path(log, dst_path);
2858 ret = log_directory_changes(trans, root, inode, path, dst_path);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002859 if (ret) {
2860 err = ret;
2861 goto out_unlock;
2862 }
Chris Masone02119d2008-09-05 16:13:11 -04002863 }
Chris Mason3a5f1d42008-09-11 15:53:37 -04002864 BTRFS_I(inode)->logged_trans = trans->transid;
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002865out_unlock:
Chris Masone02119d2008-09-05 16:13:11 -04002866 mutex_unlock(&BTRFS_I(inode)->log_mutex);
2867
2868 btrfs_free_path(path);
2869 btrfs_free_path(dst_path);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002870 return err;
Chris Masone02119d2008-09-05 16:13:11 -04002871}
2872
Chris Mason12fcfd22009-03-24 10:24:20 -04002873/*
2874 * follow the dentry parent pointers up the chain and see if any
2875 * of the directories in it require a full commit before they can
2876 * be logged. Returns zero if nothing special needs to be done or 1 if
2877 * a full commit is required.
2878 */
2879static noinline int check_parent_dirs_for_sync(struct btrfs_trans_handle *trans,
2880 struct inode *inode,
2881 struct dentry *parent,
2882 struct super_block *sb,
2883 u64 last_committed)
Chris Masone02119d2008-09-05 16:13:11 -04002884{
Chris Mason12fcfd22009-03-24 10:24:20 -04002885 int ret = 0;
2886 struct btrfs_root *root;
Chris Masone02119d2008-09-05 16:13:11 -04002887
Chris Masonaf4176b2009-03-24 10:24:31 -04002888 /*
2889 * for regular files, if its inode is already on disk, we don't
2890 * have to worry about the parents at all. This is because
2891 * we can use the last_unlink_trans field to record renames
2892 * and other fun in this file.
2893 */
2894 if (S_ISREG(inode->i_mode) &&
2895 BTRFS_I(inode)->generation <= last_committed &&
2896 BTRFS_I(inode)->last_unlink_trans <= last_committed)
2897 goto out;
2898
Chris Mason12fcfd22009-03-24 10:24:20 -04002899 if (!S_ISDIR(inode->i_mode)) {
2900 if (!parent || !parent->d_inode || sb != parent->d_inode->i_sb)
2901 goto out;
2902 inode = parent->d_inode;
2903 }
2904
2905 while (1) {
2906 BTRFS_I(inode)->logged_trans = trans->transid;
2907 smp_mb();
2908
2909 if (BTRFS_I(inode)->last_unlink_trans > last_committed) {
2910 root = BTRFS_I(inode)->root;
2911
2912 /*
2913 * make sure any commits to the log are forced
2914 * to be full commits
2915 */
2916 root->fs_info->last_trans_log_full_commit =
2917 trans->transid;
2918 ret = 1;
2919 break;
2920 }
2921
2922 if (!parent || !parent->d_inode || sb != parent->d_inode->i_sb)
2923 break;
2924
Yan, Zheng76dda932009-09-21 16:00:26 -04002925 if (IS_ROOT(parent))
Chris Mason12fcfd22009-03-24 10:24:20 -04002926 break;
2927
2928 parent = parent->d_parent;
2929 inode = parent->d_inode;
2930
2931 }
2932out:
Chris Masone02119d2008-09-05 16:13:11 -04002933 return ret;
2934}
2935
Chris Mason257c62e2009-10-13 13:21:08 -04002936static int inode_in_log(struct btrfs_trans_handle *trans,
2937 struct inode *inode)
2938{
2939 struct btrfs_root *root = BTRFS_I(inode)->root;
2940 int ret = 0;
2941
2942 mutex_lock(&root->log_mutex);
2943 if (BTRFS_I(inode)->logged_trans == trans->transid &&
2944 BTRFS_I(inode)->last_sub_trans <= root->last_log_commit)
2945 ret = 1;
2946 mutex_unlock(&root->log_mutex);
2947 return ret;
2948}
2949
2950
Chris Masone02119d2008-09-05 16:13:11 -04002951/*
2952 * helper function around btrfs_log_inode to make sure newly created
2953 * parent directories also end up in the log. A minimal inode and backref
2954 * only logging is done of any parent directories that are older than
2955 * the last committed transaction
2956 */
Chris Mason12fcfd22009-03-24 10:24:20 -04002957int btrfs_log_inode_parent(struct btrfs_trans_handle *trans,
2958 struct btrfs_root *root, struct inode *inode,
2959 struct dentry *parent, int exists_only)
Chris Masone02119d2008-09-05 16:13:11 -04002960{
Chris Mason12fcfd22009-03-24 10:24:20 -04002961 int inode_only = exists_only ? LOG_INODE_EXISTS : LOG_INODE_ALL;
Chris Masone02119d2008-09-05 16:13:11 -04002962 struct super_block *sb;
Chris Mason12fcfd22009-03-24 10:24:20 -04002963 int ret = 0;
2964 u64 last_committed = root->fs_info->last_trans_committed;
2965
2966 sb = inode->i_sb;
2967
Sage Weil3a5e1402009-04-02 16:49:40 -04002968 if (btrfs_test_opt(root, NOTREELOG)) {
2969 ret = 1;
2970 goto end_no_trans;
2971 }
2972
Chris Mason12fcfd22009-03-24 10:24:20 -04002973 if (root->fs_info->last_trans_log_full_commit >
2974 root->fs_info->last_trans_committed) {
2975 ret = 1;
2976 goto end_no_trans;
2977 }
2978
Yan, Zheng76dda932009-09-21 16:00:26 -04002979 if (root != BTRFS_I(inode)->root ||
2980 btrfs_root_refs(&root->root_item) == 0) {
2981 ret = 1;
2982 goto end_no_trans;
2983 }
2984
Chris Mason12fcfd22009-03-24 10:24:20 -04002985 ret = check_parent_dirs_for_sync(trans, inode, parent,
2986 sb, last_committed);
2987 if (ret)
2988 goto end_no_trans;
Chris Masone02119d2008-09-05 16:13:11 -04002989
Chris Mason257c62e2009-10-13 13:21:08 -04002990 if (inode_in_log(trans, inode)) {
2991 ret = BTRFS_NO_LOG_SYNC;
2992 goto end_no_trans;
2993 }
2994
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002995 ret = start_log_trans(trans, root);
2996 if (ret)
2997 goto end_trans;
Chris Mason12fcfd22009-03-24 10:24:20 -04002998
2999 ret = btrfs_log_inode(trans, root, inode, inode_only);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04003000 if (ret)
3001 goto end_trans;
Chris Mason12fcfd22009-03-24 10:24:20 -04003002
Chris Masonaf4176b2009-03-24 10:24:31 -04003003 /*
3004 * for regular files, if its inode is already on disk, we don't
3005 * have to worry about the parents at all. This is because
3006 * we can use the last_unlink_trans field to record renames
3007 * and other fun in this file.
3008 */
3009 if (S_ISREG(inode->i_mode) &&
3010 BTRFS_I(inode)->generation <= last_committed &&
Yan, Zheng4a500fd2010-05-16 10:49:59 -04003011 BTRFS_I(inode)->last_unlink_trans <= last_committed) {
3012 ret = 0;
3013 goto end_trans;
3014 }
Chris Masonaf4176b2009-03-24 10:24:31 -04003015
3016 inode_only = LOG_INODE_EXISTS;
Chris Masond3977122009-01-05 21:25:51 -05003017 while (1) {
Chris Mason12fcfd22009-03-24 10:24:20 -04003018 if (!parent || !parent->d_inode || sb != parent->d_inode->i_sb)
Chris Masone02119d2008-09-05 16:13:11 -04003019 break;
3020
Chris Mason12fcfd22009-03-24 10:24:20 -04003021 inode = parent->d_inode;
Yan, Zheng76dda932009-09-21 16:00:26 -04003022 if (root != BTRFS_I(inode)->root)
3023 break;
3024
Chris Mason12fcfd22009-03-24 10:24:20 -04003025 if (BTRFS_I(inode)->generation >
3026 root->fs_info->last_trans_committed) {
3027 ret = btrfs_log_inode(trans, root, inode, inode_only);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04003028 if (ret)
3029 goto end_trans;
Chris Mason12fcfd22009-03-24 10:24:20 -04003030 }
Yan, Zheng76dda932009-09-21 16:00:26 -04003031 if (IS_ROOT(parent))
Chris Masone02119d2008-09-05 16:13:11 -04003032 break;
Chris Mason12fcfd22009-03-24 10:24:20 -04003033
3034 parent = parent->d_parent;
Chris Masone02119d2008-09-05 16:13:11 -04003035 }
Chris Mason12fcfd22009-03-24 10:24:20 -04003036 ret = 0;
Yan, Zheng4a500fd2010-05-16 10:49:59 -04003037end_trans:
3038 if (ret < 0) {
3039 BUG_ON(ret != -ENOSPC);
3040 root->fs_info->last_trans_log_full_commit = trans->transid;
3041 ret = 1;
3042 }
Chris Mason12fcfd22009-03-24 10:24:20 -04003043 btrfs_end_log_trans(root);
3044end_no_trans:
3045 return ret;
Chris Masone02119d2008-09-05 16:13:11 -04003046}
3047
3048/*
3049 * it is not safe to log dentry if the chunk root has added new
3050 * chunks. This returns 0 if the dentry was logged, and 1 otherwise.
3051 * If this returns 1, you must commit the transaction to safely get your
3052 * data on disk.
3053 */
3054int btrfs_log_dentry_safe(struct btrfs_trans_handle *trans,
3055 struct btrfs_root *root, struct dentry *dentry)
3056{
Chris Mason12fcfd22009-03-24 10:24:20 -04003057 return btrfs_log_inode_parent(trans, root, dentry->d_inode,
3058 dentry->d_parent, 0);
Chris Masone02119d2008-09-05 16:13:11 -04003059}
3060
3061/*
3062 * should be called during mount to recover any replay any log trees
3063 * from the FS
3064 */
3065int btrfs_recover_log_trees(struct btrfs_root *log_root_tree)
3066{
3067 int ret;
3068 struct btrfs_path *path;
3069 struct btrfs_trans_handle *trans;
3070 struct btrfs_key key;
3071 struct btrfs_key found_key;
3072 struct btrfs_key tmp_key;
3073 struct btrfs_root *log;
3074 struct btrfs_fs_info *fs_info = log_root_tree->fs_info;
3075 struct walk_control wc = {
3076 .process_func = process_one_buffer,
3077 .stage = 0,
3078 };
3079
3080 fs_info->log_root_recovering = 1;
3081 path = btrfs_alloc_path();
3082 BUG_ON(!path);
3083
Yan, Zheng4a500fd2010-05-16 10:49:59 -04003084 trans = btrfs_start_transaction(fs_info->tree_root, 0);
Chris Masone02119d2008-09-05 16:13:11 -04003085
3086 wc.trans = trans;
3087 wc.pin = 1;
3088
3089 walk_log_tree(trans, log_root_tree, &wc);
3090
3091again:
3092 key.objectid = BTRFS_TREE_LOG_OBJECTID;
3093 key.offset = (u64)-1;
3094 btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
3095
Chris Masond3977122009-01-05 21:25:51 -05003096 while (1) {
Chris Masone02119d2008-09-05 16:13:11 -04003097 ret = btrfs_search_slot(NULL, log_root_tree, &key, path, 0, 0);
3098 if (ret < 0)
3099 break;
3100 if (ret > 0) {
3101 if (path->slots[0] == 0)
3102 break;
3103 path->slots[0]--;
3104 }
3105 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
3106 path->slots[0]);
3107 btrfs_release_path(log_root_tree, path);
3108 if (found_key.objectid != BTRFS_TREE_LOG_OBJECTID)
3109 break;
3110
3111 log = btrfs_read_fs_root_no_radix(log_root_tree,
3112 &found_key);
3113 BUG_ON(!log);
3114
3115
3116 tmp_key.objectid = found_key.offset;
3117 tmp_key.type = BTRFS_ROOT_ITEM_KEY;
3118 tmp_key.offset = (u64)-1;
3119
3120 wc.replay_dest = btrfs_read_fs_root_no_name(fs_info, &tmp_key);
Chris Masone02119d2008-09-05 16:13:11 -04003121 BUG_ON(!wc.replay_dest);
3122
Yan Zheng07d400a2009-01-06 11:42:00 -05003123 wc.replay_dest->log_root = log;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003124 btrfs_record_root_in_trans(trans, wc.replay_dest);
Chris Masone02119d2008-09-05 16:13:11 -04003125 ret = walk_log_tree(trans, log, &wc);
3126 BUG_ON(ret);
3127
3128 if (wc.stage == LOG_WALK_REPLAY_ALL) {
3129 ret = fixup_inode_link_counts(trans, wc.replay_dest,
3130 path);
3131 BUG_ON(ret);
3132 }
Chris Masone02119d2008-09-05 16:13:11 -04003133
3134 key.offset = found_key.offset - 1;
Yan Zheng07d400a2009-01-06 11:42:00 -05003135 wc.replay_dest->log_root = NULL;
Chris Masone02119d2008-09-05 16:13:11 -04003136 free_extent_buffer(log->node);
Chris Masonb263c2c2009-06-11 11:24:47 -04003137 free_extent_buffer(log->commit_root);
Chris Masone02119d2008-09-05 16:13:11 -04003138 kfree(log);
3139
3140 if (found_key.offset == 0)
3141 break;
3142 }
3143 btrfs_release_path(log_root_tree, path);
3144
3145 /* step one is to pin it all, step two is to replay just inodes */
3146 if (wc.pin) {
3147 wc.pin = 0;
3148 wc.process_func = replay_one_buffer;
3149 wc.stage = LOG_WALK_REPLAY_INODES;
3150 goto again;
3151 }
3152 /* step three is to replay everything */
3153 if (wc.stage < LOG_WALK_REPLAY_ALL) {
3154 wc.stage++;
3155 goto again;
3156 }
3157
3158 btrfs_free_path(path);
3159
3160 free_extent_buffer(log_root_tree->node);
3161 log_root_tree->log_root = NULL;
3162 fs_info->log_root_recovering = 0;
3163
3164 /* step 4: commit the transaction, which also unpins the blocks */
3165 btrfs_commit_transaction(trans, fs_info->tree_root);
3166
3167 kfree(log_root_tree);
3168 return 0;
3169}
Chris Mason12fcfd22009-03-24 10:24:20 -04003170
3171/*
3172 * there are some corner cases where we want to force a full
3173 * commit instead of allowing a directory to be logged.
3174 *
3175 * They revolve around files there were unlinked from the directory, and
3176 * this function updates the parent directory so that a full commit is
3177 * properly done if it is fsync'd later after the unlinks are done.
3178 */
3179void btrfs_record_unlink_dir(struct btrfs_trans_handle *trans,
3180 struct inode *dir, struct inode *inode,
3181 int for_rename)
3182{
3183 /*
Chris Masonaf4176b2009-03-24 10:24:31 -04003184 * when we're logging a file, if it hasn't been renamed
3185 * or unlinked, and its inode is fully committed on disk,
3186 * we don't have to worry about walking up the directory chain
3187 * to log its parents.
3188 *
3189 * So, we use the last_unlink_trans field to put this transid
3190 * into the file. When the file is logged we check it and
3191 * don't log the parents if the file is fully on disk.
3192 */
3193 if (S_ISREG(inode->i_mode))
3194 BTRFS_I(inode)->last_unlink_trans = trans->transid;
3195
3196 /*
Chris Mason12fcfd22009-03-24 10:24:20 -04003197 * if this directory was already logged any new
3198 * names for this file/dir will get recorded
3199 */
3200 smp_mb();
3201 if (BTRFS_I(dir)->logged_trans == trans->transid)
3202 return;
3203
3204 /*
3205 * if the inode we're about to unlink was logged,
3206 * the log will be properly updated for any new names
3207 */
3208 if (BTRFS_I(inode)->logged_trans == trans->transid)
3209 return;
3210
3211 /*
3212 * when renaming files across directories, if the directory
3213 * there we're unlinking from gets fsync'd later on, there's
3214 * no way to find the destination directory later and fsync it
3215 * properly. So, we have to be conservative and force commits
3216 * so the new name gets discovered.
3217 */
3218 if (for_rename)
3219 goto record;
3220
3221 /* we can safely do the unlink without any special recording */
3222 return;
3223
3224record:
3225 BTRFS_I(dir)->last_unlink_trans = trans->transid;
3226}
3227
3228/*
3229 * Call this after adding a new name for a file and it will properly
3230 * update the log to reflect the new name.
3231 *
3232 * It will return zero if all goes well, and it will return 1 if a
3233 * full transaction commit is required.
3234 */
3235int btrfs_log_new_name(struct btrfs_trans_handle *trans,
3236 struct inode *inode, struct inode *old_dir,
3237 struct dentry *parent)
3238{
3239 struct btrfs_root * root = BTRFS_I(inode)->root;
3240
3241 /*
Chris Masonaf4176b2009-03-24 10:24:31 -04003242 * this will force the logging code to walk the dentry chain
3243 * up for the file
3244 */
3245 if (S_ISREG(inode->i_mode))
3246 BTRFS_I(inode)->last_unlink_trans = trans->transid;
3247
3248 /*
Chris Mason12fcfd22009-03-24 10:24:20 -04003249 * if this inode hasn't been logged and directory we're renaming it
3250 * from hasn't been logged, we don't need to log it
3251 */
3252 if (BTRFS_I(inode)->logged_trans <=
3253 root->fs_info->last_trans_committed &&
3254 (!old_dir || BTRFS_I(old_dir)->logged_trans <=
3255 root->fs_info->last_trans_committed))
3256 return 0;
3257
3258 return btrfs_log_inode_parent(trans, root, inode, parent, 1);
3259}
3260