blob: 592396c6dc47465bccf4422d6259d0dfe2885e0c [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) {
David Sterbab3b4aa72011-04-21 01:20:15 +0200336 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -0400337 return 0;
338 }
339 dst_copy = kmalloc(item_size, GFP_NOFS);
340 src_copy = kmalloc(item_size, GFP_NOFS);
liubo2a29edc2011-01-26 06:22:08 +0000341 if (!dst_copy || !src_copy) {
David Sterbab3b4aa72011-04-21 01:20:15 +0200342 btrfs_release_path(path);
liubo2a29edc2011-01-26 06:22:08 +0000343 kfree(dst_copy);
344 kfree(src_copy);
345 return -ENOMEM;
346 }
Chris Masone02119d2008-09-05 16:13:11 -0400347
348 read_extent_buffer(eb, src_copy, src_ptr, item_size);
349
350 dst_ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
351 read_extent_buffer(path->nodes[0], dst_copy, dst_ptr,
352 item_size);
353 ret = memcmp(dst_copy, src_copy, item_size);
354
355 kfree(dst_copy);
356 kfree(src_copy);
357 /*
358 * they have the same contents, just return, this saves
359 * us from cowing blocks in the destination tree and doing
360 * extra writes that may not have been done by a previous
361 * sync
362 */
363 if (ret == 0) {
David Sterbab3b4aa72011-04-21 01:20:15 +0200364 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -0400365 return 0;
366 }
367
368 }
369insert:
David Sterbab3b4aa72011-04-21 01:20:15 +0200370 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -0400371 /* try to insert the key into the destination tree */
372 ret = btrfs_insert_empty_item(trans, root, path,
373 key, item_size);
374
375 /* make sure any existing item is the correct size */
376 if (ret == -EEXIST) {
377 u32 found_size;
378 found_size = btrfs_item_size_nr(path->nodes[0],
379 path->slots[0]);
380 if (found_size > item_size) {
381 btrfs_truncate_item(trans, root, path, item_size, 1);
382 } else if (found_size < item_size) {
Yan Zheng87b29b22008-12-17 10:21:48 -0500383 ret = btrfs_extend_item(trans, root, path,
384 item_size - found_size);
Chris Masone02119d2008-09-05 16:13:11 -0400385 }
386 } else if (ret) {
Yan, Zheng4a500fd2010-05-16 10:49:59 -0400387 return ret;
Chris Masone02119d2008-09-05 16:13:11 -0400388 }
389 dst_ptr = btrfs_item_ptr_offset(path->nodes[0],
390 path->slots[0]);
391
392 /* don't overwrite an existing inode if the generation number
393 * was logged as zero. This is done when the tree logging code
394 * is just logging an inode to make sure it exists after recovery.
395 *
396 * Also, don't overwrite i_size on directories during replay.
397 * log replay inserts and removes directory items based on the
398 * state of the tree found in the subvolume, and i_size is modified
399 * as it goes
400 */
401 if (key->type == BTRFS_INODE_ITEM_KEY && ret == -EEXIST) {
402 struct btrfs_inode_item *src_item;
403 struct btrfs_inode_item *dst_item;
404
405 src_item = (struct btrfs_inode_item *)src_ptr;
406 dst_item = (struct btrfs_inode_item *)dst_ptr;
407
408 if (btrfs_inode_generation(eb, src_item) == 0)
409 goto no_copy;
410
411 if (overwrite_root &&
412 S_ISDIR(btrfs_inode_mode(eb, src_item)) &&
413 S_ISDIR(btrfs_inode_mode(path->nodes[0], dst_item))) {
414 save_old_i_size = 1;
415 saved_i_size = btrfs_inode_size(path->nodes[0],
416 dst_item);
417 }
418 }
419
420 copy_extent_buffer(path->nodes[0], eb, dst_ptr,
421 src_ptr, item_size);
422
423 if (save_old_i_size) {
424 struct btrfs_inode_item *dst_item;
425 dst_item = (struct btrfs_inode_item *)dst_ptr;
426 btrfs_set_inode_size(path->nodes[0], dst_item, saved_i_size);
427 }
428
429 /* make sure the generation is filled in */
430 if (key->type == BTRFS_INODE_ITEM_KEY) {
431 struct btrfs_inode_item *dst_item;
432 dst_item = (struct btrfs_inode_item *)dst_ptr;
433 if (btrfs_inode_generation(path->nodes[0], dst_item) == 0) {
434 btrfs_set_inode_generation(path->nodes[0], dst_item,
435 trans->transid);
436 }
437 }
438no_copy:
439 btrfs_mark_buffer_dirty(path->nodes[0]);
David Sterbab3b4aa72011-04-21 01:20:15 +0200440 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -0400441 return 0;
442}
443
444/*
445 * simple helper to read an inode off the disk from a given root
446 * This can only be called for subvolume roots and not for the log
447 */
448static noinline struct inode *read_one_inode(struct btrfs_root *root,
449 u64 objectid)
450{
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400451 struct btrfs_key key;
Chris Masone02119d2008-09-05 16:13:11 -0400452 struct inode *inode;
Chris Masone02119d2008-09-05 16:13:11 -0400453
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400454 key.objectid = objectid;
455 key.type = BTRFS_INODE_ITEM_KEY;
456 key.offset = 0;
Josef Bacik73f73412009-12-04 17:38:27 +0000457 inode = btrfs_iget(root->fs_info->sb, &key, root, NULL);
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400458 if (IS_ERR(inode)) {
459 inode = NULL;
460 } else if (is_bad_inode(inode)) {
Chris Masone02119d2008-09-05 16:13:11 -0400461 iput(inode);
462 inode = NULL;
463 }
464 return inode;
465}
466
467/* replays a single extent in 'eb' at 'slot' with 'key' into the
468 * subvolume 'root'. path is released on entry and should be released
469 * on exit.
470 *
471 * extents in the log tree have not been allocated out of the extent
472 * tree yet. So, this completes the allocation, taking a reference
473 * as required if the extent already exists or creating a new extent
474 * if it isn't in the extent allocation tree yet.
475 *
476 * The extent is inserted into the file, dropping any existing extents
477 * from the file that overlap the new one.
478 */
479static noinline int replay_one_extent(struct btrfs_trans_handle *trans,
480 struct btrfs_root *root,
481 struct btrfs_path *path,
482 struct extent_buffer *eb, int slot,
483 struct btrfs_key *key)
484{
485 int found_type;
486 u64 mask = root->sectorsize - 1;
487 u64 extent_end;
488 u64 alloc_hint;
489 u64 start = key->offset;
Yan Zheng07d400a2009-01-06 11:42:00 -0500490 u64 saved_nbytes;
Chris Masone02119d2008-09-05 16:13:11 -0400491 struct btrfs_file_extent_item *item;
492 struct inode *inode = NULL;
493 unsigned long size;
494 int ret = 0;
495
496 item = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
497 found_type = btrfs_file_extent_type(eb, item);
498
Yan Zhengd899e052008-10-30 14:25:28 -0400499 if (found_type == BTRFS_FILE_EXTENT_REG ||
500 found_type == BTRFS_FILE_EXTENT_PREALLOC)
Chris Masone02119d2008-09-05 16:13:11 -0400501 extent_end = start + btrfs_file_extent_num_bytes(eb, item);
502 else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
Chris Masonc8b97812008-10-29 14:49:59 -0400503 size = btrfs_file_extent_inline_len(eb, item);
Chris Masone02119d2008-09-05 16:13:11 -0400504 extent_end = (start + size + mask) & ~mask;
505 } else {
506 ret = 0;
507 goto out;
508 }
509
510 inode = read_one_inode(root, key->objectid);
511 if (!inode) {
512 ret = -EIO;
513 goto out;
514 }
515
516 /*
517 * first check to see if we already have this extent in the
518 * file. This must be done before the btrfs_drop_extents run
519 * so we don't try to drop this extent.
520 */
Li Zefan33345d012011-04-20 10:31:50 +0800521 ret = btrfs_lookup_file_extent(trans, root, path, btrfs_ino(inode),
Chris Masone02119d2008-09-05 16:13:11 -0400522 start, 0);
523
Yan Zhengd899e052008-10-30 14:25:28 -0400524 if (ret == 0 &&
525 (found_type == BTRFS_FILE_EXTENT_REG ||
526 found_type == BTRFS_FILE_EXTENT_PREALLOC)) {
Chris Masone02119d2008-09-05 16:13:11 -0400527 struct btrfs_file_extent_item cmp1;
528 struct btrfs_file_extent_item cmp2;
529 struct btrfs_file_extent_item *existing;
530 struct extent_buffer *leaf;
531
532 leaf = path->nodes[0];
533 existing = btrfs_item_ptr(leaf, path->slots[0],
534 struct btrfs_file_extent_item);
535
536 read_extent_buffer(eb, &cmp1, (unsigned long)item,
537 sizeof(cmp1));
538 read_extent_buffer(leaf, &cmp2, (unsigned long)existing,
539 sizeof(cmp2));
540
541 /*
542 * we already have a pointer to this exact extent,
543 * we don't have to do anything
544 */
545 if (memcmp(&cmp1, &cmp2, sizeof(cmp1)) == 0) {
David Sterbab3b4aa72011-04-21 01:20:15 +0200546 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -0400547 goto out;
548 }
549 }
David Sterbab3b4aa72011-04-21 01:20:15 +0200550 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -0400551
Yan Zheng07d400a2009-01-06 11:42:00 -0500552 saved_nbytes = inode_get_bytes(inode);
Chris Masone02119d2008-09-05 16:13:11 -0400553 /* drop any overlapping extents */
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000554 ret = btrfs_drop_extents(trans, inode, start, extent_end,
555 &alloc_hint, 1);
Chris Masone02119d2008-09-05 16:13:11 -0400556 BUG_ON(ret);
557
Yan Zheng07d400a2009-01-06 11:42:00 -0500558 if (found_type == BTRFS_FILE_EXTENT_REG ||
559 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400560 u64 offset;
Yan Zheng07d400a2009-01-06 11:42:00 -0500561 unsigned long dest_offset;
562 struct btrfs_key ins;
Chris Masone02119d2008-09-05 16:13:11 -0400563
Yan Zheng07d400a2009-01-06 11:42:00 -0500564 ret = btrfs_insert_empty_item(trans, root, path, key,
565 sizeof(*item));
566 BUG_ON(ret);
567 dest_offset = btrfs_item_ptr_offset(path->nodes[0],
568 path->slots[0]);
569 copy_extent_buffer(path->nodes[0], eb, dest_offset,
570 (unsigned long)item, sizeof(*item));
571
572 ins.objectid = btrfs_file_extent_disk_bytenr(eb, item);
573 ins.offset = btrfs_file_extent_disk_num_bytes(eb, item);
574 ins.type = BTRFS_EXTENT_ITEM_KEY;
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400575 offset = key->offset - btrfs_file_extent_offset(eb, item);
Yan Zheng07d400a2009-01-06 11:42:00 -0500576
577 if (ins.objectid > 0) {
578 u64 csum_start;
579 u64 csum_end;
580 LIST_HEAD(ordered_sums);
581 /*
582 * is this extent already allocated in the extent
583 * allocation tree? If so, just add a reference
584 */
585 ret = btrfs_lookup_extent(root, ins.objectid,
586 ins.offset);
587 if (ret == 0) {
588 ret = btrfs_inc_extent_ref(trans, root,
589 ins.objectid, ins.offset,
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400590 0, root->root_key.objectid,
591 key->objectid, offset);
Tsutomu Itoh37daa4f2011-04-28 09:18:21 +0000592 BUG_ON(ret);
Yan Zheng07d400a2009-01-06 11:42:00 -0500593 } else {
594 /*
595 * insert the extent pointer in the extent
596 * allocation tree
597 */
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400598 ret = btrfs_alloc_logged_file_extent(trans,
599 root, root->root_key.objectid,
600 key->objectid, offset, &ins);
Yan Zheng07d400a2009-01-06 11:42:00 -0500601 BUG_ON(ret);
602 }
David Sterbab3b4aa72011-04-21 01:20:15 +0200603 btrfs_release_path(path);
Yan Zheng07d400a2009-01-06 11:42:00 -0500604
605 if (btrfs_file_extent_compression(eb, item)) {
606 csum_start = ins.objectid;
607 csum_end = csum_start + ins.offset;
608 } else {
609 csum_start = ins.objectid +
610 btrfs_file_extent_offset(eb, item);
611 csum_end = csum_start +
612 btrfs_file_extent_num_bytes(eb, item);
613 }
614
615 ret = btrfs_lookup_csums_range(root->log_root,
616 csum_start, csum_end - 1,
Arne Jansena2de7332011-03-08 14:14:00 +0100617 &ordered_sums, 0);
Yan Zheng07d400a2009-01-06 11:42:00 -0500618 BUG_ON(ret);
619 while (!list_empty(&ordered_sums)) {
620 struct btrfs_ordered_sum *sums;
621 sums = list_entry(ordered_sums.next,
622 struct btrfs_ordered_sum,
623 list);
624 ret = btrfs_csum_file_blocks(trans,
625 root->fs_info->csum_root,
626 sums);
627 BUG_ON(ret);
628 list_del(&sums->list);
629 kfree(sums);
630 }
631 } else {
David Sterbab3b4aa72011-04-21 01:20:15 +0200632 btrfs_release_path(path);
Yan Zheng07d400a2009-01-06 11:42:00 -0500633 }
634 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
635 /* inline extents are easy, we just overwrite them */
636 ret = overwrite_item(trans, root, path, eb, slot, key);
637 BUG_ON(ret);
638 }
639
640 inode_set_bytes(inode, saved_nbytes);
Chris Masone02119d2008-09-05 16:13:11 -0400641 btrfs_update_inode(trans, root, inode);
642out:
643 if (inode)
644 iput(inode);
645 return ret;
646}
647
648/*
649 * when cleaning up conflicts between the directory names in the
650 * subvolume, directory names in the log and directory names in the
651 * inode back references, we may have to unlink inodes from directories.
652 *
653 * This is a helper function to do the unlink of a specific directory
654 * item
655 */
656static noinline int drop_one_dir_item(struct btrfs_trans_handle *trans,
657 struct btrfs_root *root,
658 struct btrfs_path *path,
659 struct inode *dir,
660 struct btrfs_dir_item *di)
661{
662 struct inode *inode;
663 char *name;
664 int name_len;
665 struct extent_buffer *leaf;
666 struct btrfs_key location;
667 int ret;
668
669 leaf = path->nodes[0];
670
671 btrfs_dir_item_key_to_cpu(leaf, di, &location);
672 name_len = btrfs_dir_name_len(leaf, di);
673 name = kmalloc(name_len, GFP_NOFS);
liubo2a29edc2011-01-26 06:22:08 +0000674 if (!name)
675 return -ENOMEM;
676
Chris Masone02119d2008-09-05 16:13:11 -0400677 read_extent_buffer(leaf, name, (unsigned long)(di + 1), name_len);
David Sterbab3b4aa72011-04-21 01:20:15 +0200678 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -0400679
680 inode = read_one_inode(root, location.objectid);
Tsutomu Itohc00e9492011-04-28 09:10:23 +0000681 if (!inode) {
682 kfree(name);
683 return -EIO;
684 }
Chris Masone02119d2008-09-05 16:13:11 -0400685
Yan Zhengec051c02009-01-05 15:43:42 -0500686 ret = link_to_fixup_dir(trans, root, path, location.objectid);
687 BUG_ON(ret);
Chris Mason12fcfd22009-03-24 10:24:20 -0400688
Chris Masone02119d2008-09-05 16:13:11 -0400689 ret = btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
Yan Zhengec051c02009-01-05 15:43:42 -0500690 BUG_ON(ret);
Chris Masone02119d2008-09-05 16:13:11 -0400691 kfree(name);
692
693 iput(inode);
694 return ret;
695}
696
697/*
698 * helper function to see if a given name and sequence number found
699 * in an inode back reference are already in a directory and correctly
700 * point to this inode
701 */
702static noinline int inode_in_dir(struct btrfs_root *root,
703 struct btrfs_path *path,
704 u64 dirid, u64 objectid, u64 index,
705 const char *name, int name_len)
706{
707 struct btrfs_dir_item *di;
708 struct btrfs_key location;
709 int match = 0;
710
711 di = btrfs_lookup_dir_index_item(NULL, root, path, dirid,
712 index, name, name_len, 0);
713 if (di && !IS_ERR(di)) {
714 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
715 if (location.objectid != objectid)
716 goto out;
717 } else
718 goto out;
David Sterbab3b4aa72011-04-21 01:20:15 +0200719 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -0400720
721 di = btrfs_lookup_dir_item(NULL, root, path, dirid, name, name_len, 0);
722 if (di && !IS_ERR(di)) {
723 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
724 if (location.objectid != objectid)
725 goto out;
726 } else
727 goto out;
728 match = 1;
729out:
David Sterbab3b4aa72011-04-21 01:20:15 +0200730 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -0400731 return match;
732}
733
734/*
735 * helper function to check a log tree for a named back reference in
736 * an inode. This is used to decide if a back reference that is
737 * found in the subvolume conflicts with what we find in the log.
738 *
739 * inode backreferences may have multiple refs in a single item,
740 * during replay we process one reference at a time, and we don't
741 * want to delete valid links to a file from the subvolume if that
742 * link is also in the log.
743 */
744static noinline int backref_in_log(struct btrfs_root *log,
745 struct btrfs_key *key,
746 char *name, int namelen)
747{
748 struct btrfs_path *path;
749 struct btrfs_inode_ref *ref;
750 unsigned long ptr;
751 unsigned long ptr_end;
752 unsigned long name_ptr;
753 int found_name_len;
754 int item_size;
755 int ret;
756 int match = 0;
757
758 path = btrfs_alloc_path();
liubo2a29edc2011-01-26 06:22:08 +0000759 if (!path)
760 return -ENOMEM;
761
Chris Masone02119d2008-09-05 16:13:11 -0400762 ret = btrfs_search_slot(NULL, log, key, path, 0, 0);
763 if (ret != 0)
764 goto out;
765
766 item_size = btrfs_item_size_nr(path->nodes[0], path->slots[0]);
767 ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
768 ptr_end = ptr + item_size;
769 while (ptr < ptr_end) {
770 ref = (struct btrfs_inode_ref *)ptr;
771 found_name_len = btrfs_inode_ref_name_len(path->nodes[0], ref);
772 if (found_name_len == namelen) {
773 name_ptr = (unsigned long)(ref + 1);
774 ret = memcmp_extent_buffer(path->nodes[0], name,
775 name_ptr, namelen);
776 if (ret == 0) {
777 match = 1;
778 goto out;
779 }
780 }
781 ptr = (unsigned long)(ref + 1) + found_name_len;
782 }
783out:
784 btrfs_free_path(path);
785 return match;
786}
787
788
789/*
790 * replay one inode back reference item found in the log tree.
791 * eb, slot and key refer to the buffer and key found in the log tree.
792 * root is the destination we are replaying into, and path is for temp
793 * use by this function. (it should be released on return).
794 */
795static noinline int add_inode_ref(struct btrfs_trans_handle *trans,
796 struct btrfs_root *root,
797 struct btrfs_root *log,
798 struct btrfs_path *path,
799 struct extent_buffer *eb, int slot,
800 struct btrfs_key *key)
801{
802 struct inode *dir;
803 int ret;
Chris Masone02119d2008-09-05 16:13:11 -0400804 struct btrfs_inode_ref *ref;
Chris Masone02119d2008-09-05 16:13:11 -0400805 struct inode *inode;
806 char *name;
807 int namelen;
808 unsigned long ref_ptr;
809 unsigned long ref_end;
liuboc622ae62011-03-26 08:01:12 -0400810 int search_done = 0;
Chris Masone02119d2008-09-05 16:13:11 -0400811
Chris Masone02119d2008-09-05 16:13:11 -0400812 /*
813 * it is possible that we didn't log all the parent directories
814 * for a given inode. If we don't find the dir, just don't
815 * copy the back ref in. The link count fixup code will take
816 * care of the rest
817 */
818 dir = read_one_inode(root, key->offset);
819 if (!dir)
820 return -ENOENT;
821
822 inode = read_one_inode(root, key->objectid);
Tsutomu Itohc00e9492011-04-28 09:10:23 +0000823 if (!inode) {
824 iput(dir);
825 return -EIO;
826 }
Chris Masone02119d2008-09-05 16:13:11 -0400827
828 ref_ptr = btrfs_item_ptr_offset(eb, slot);
829 ref_end = ref_ptr + btrfs_item_size_nr(eb, slot);
830
831again:
832 ref = (struct btrfs_inode_ref *)ref_ptr;
833
834 namelen = btrfs_inode_ref_name_len(eb, ref);
835 name = kmalloc(namelen, GFP_NOFS);
836 BUG_ON(!name);
837
838 read_extent_buffer(eb, name, (unsigned long)(ref + 1), namelen);
839
840 /* if we already have a perfect match, we're done */
Li Zefan33345d012011-04-20 10:31:50 +0800841 if (inode_in_dir(root, path, btrfs_ino(dir), btrfs_ino(inode),
Chris Masone02119d2008-09-05 16:13:11 -0400842 btrfs_inode_ref_index(eb, ref),
843 name, namelen)) {
844 goto out;
845 }
846
847 /*
848 * look for a conflicting back reference in the metadata.
849 * if we find one we have to unlink that name of the file
850 * before we add our new link. Later on, we overwrite any
851 * existing back reference, and we don't want to create
852 * dangling pointers in the directory.
853 */
liuboc622ae62011-03-26 08:01:12 -0400854
855 if (search_done)
856 goto insert;
857
Chris Masone02119d2008-09-05 16:13:11 -0400858 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
859 if (ret == 0) {
860 char *victim_name;
861 int victim_name_len;
862 struct btrfs_inode_ref *victim_ref;
863 unsigned long ptr;
864 unsigned long ptr_end;
865 struct extent_buffer *leaf = path->nodes[0];
866
867 /* are we trying to overwrite a back ref for the root directory
868 * if so, just jump out, we're done
869 */
870 if (key->objectid == key->offset)
871 goto out_nowrite;
872
873 /* check all the names in this back reference to see
874 * if they are in the log. if so, we allow them to stay
875 * otherwise they must be unlinked as a conflict
876 */
877 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
878 ptr_end = ptr + btrfs_item_size_nr(leaf, path->slots[0]);
Chris Masond3977122009-01-05 21:25:51 -0500879 while (ptr < ptr_end) {
Chris Masone02119d2008-09-05 16:13:11 -0400880 victim_ref = (struct btrfs_inode_ref *)ptr;
881 victim_name_len = btrfs_inode_ref_name_len(leaf,
882 victim_ref);
883 victim_name = kmalloc(victim_name_len, GFP_NOFS);
884 BUG_ON(!victim_name);
885
886 read_extent_buffer(leaf, victim_name,
887 (unsigned long)(victim_ref + 1),
888 victim_name_len);
889
890 if (!backref_in_log(log, key, victim_name,
891 victim_name_len)) {
892 btrfs_inc_nlink(inode);
David Sterbab3b4aa72011-04-21 01:20:15 +0200893 btrfs_release_path(path);
Chris Mason12fcfd22009-03-24 10:24:20 -0400894
Chris Masone02119d2008-09-05 16:13:11 -0400895 ret = btrfs_unlink_inode(trans, root, dir,
896 inode, victim_name,
897 victim_name_len);
Chris Masone02119d2008-09-05 16:13:11 -0400898 }
899 kfree(victim_name);
900 ptr = (unsigned long)(victim_ref + 1) + victim_name_len;
901 }
902 BUG_ON(ret);
liuboc622ae62011-03-26 08:01:12 -0400903
904 /*
905 * NOTE: we have searched root tree and checked the
906 * coresponding ref, it does not need to check again.
907 */
908 search_done = 1;
Chris Masone02119d2008-09-05 16:13:11 -0400909 }
David Sterbab3b4aa72011-04-21 01:20:15 +0200910 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -0400911
liuboc622ae62011-03-26 08:01:12 -0400912insert:
Chris Masone02119d2008-09-05 16:13:11 -0400913 /* 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:
David Sterbab3b4aa72011-04-21 01:20:15 +0200931 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -0400932 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;
Li Zefan33345d012011-04-20 10:31:50 +0800969 u64 ino = btrfs_ino(inode);
Chris Masone02119d2008-09-05 16:13:11 -0400970
Li Zefan33345d012011-04-20 10:31:50 +0800971 key.objectid = ino;
Chris Masone02119d2008-09-05 16:13:11 -0400972 key.type = BTRFS_INODE_REF_KEY;
973 key.offset = (u64)-1;
974
975 path = btrfs_alloc_path();
liubo2a29edc2011-01-26 06:22:08 +0000976 if (!path)
977 return -ENOMEM;
Chris Masone02119d2008-09-05 16:13:11 -0400978
Chris Masond3977122009-01-05 21:25:51 -0500979 while (1) {
Chris Masone02119d2008-09-05 16:13:11 -0400980 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
981 if (ret < 0)
982 break;
983 if (ret > 0) {
984 if (path->slots[0] == 0)
985 break;
986 path->slots[0]--;
987 }
988 btrfs_item_key_to_cpu(path->nodes[0], &key,
989 path->slots[0]);
Li Zefan33345d012011-04-20 10:31:50 +0800990 if (key.objectid != ino ||
Chris Masone02119d2008-09-05 16:13:11 -0400991 key.type != BTRFS_INODE_REF_KEY)
992 break;
993 ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
994 ptr_end = ptr + btrfs_item_size_nr(path->nodes[0],
995 path->slots[0]);
Chris Masond3977122009-01-05 21:25:51 -0500996 while (ptr < ptr_end) {
Chris Masone02119d2008-09-05 16:13:11 -0400997 struct btrfs_inode_ref *ref;
998
999 ref = (struct btrfs_inode_ref *)ptr;
1000 name_len = btrfs_inode_ref_name_len(path->nodes[0],
1001 ref);
1002 ptr = (unsigned long)(ref + 1) + name_len;
1003 nlink++;
1004 }
1005
1006 if (key.offset == 0)
1007 break;
1008 key.offset--;
David Sterbab3b4aa72011-04-21 01:20:15 +02001009 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04001010 }
David Sterbab3b4aa72011-04-21 01:20:15 +02001011 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04001012 if (nlink != inode->i_nlink) {
1013 inode->i_nlink = nlink;
1014 btrfs_update_inode(trans, root, inode);
1015 }
Chris Mason8d5bf1c2008-09-11 15:51:21 -04001016 BTRFS_I(inode)->index_cnt = (u64)-1;
Chris Masone02119d2008-09-05 16:13:11 -04001017
Yan, Zhengc71bf092009-11-12 09:34:40 +00001018 if (inode->i_nlink == 0) {
1019 if (S_ISDIR(inode->i_mode)) {
1020 ret = replay_dir_deletes(trans, root, NULL, path,
Li Zefan33345d012011-04-20 10:31:50 +08001021 ino, 1);
Yan, Zhengc71bf092009-11-12 09:34:40 +00001022 BUG_ON(ret);
1023 }
Li Zefan33345d012011-04-20 10:31:50 +08001024 ret = insert_orphan_item(trans, root, ino);
Chris Mason12fcfd22009-03-24 10:24:20 -04001025 BUG_ON(ret);
1026 }
1027 btrfs_free_path(path);
1028
Chris Masone02119d2008-09-05 16:13:11 -04001029 return 0;
1030}
1031
1032static noinline int fixup_inode_link_counts(struct btrfs_trans_handle *trans,
1033 struct btrfs_root *root,
1034 struct btrfs_path *path)
1035{
1036 int ret;
1037 struct btrfs_key key;
1038 struct inode *inode;
1039
1040 key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID;
1041 key.type = BTRFS_ORPHAN_ITEM_KEY;
1042 key.offset = (u64)-1;
Chris Masond3977122009-01-05 21:25:51 -05001043 while (1) {
Chris Masone02119d2008-09-05 16:13:11 -04001044 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1045 if (ret < 0)
1046 break;
1047
1048 if (ret == 1) {
1049 if (path->slots[0] == 0)
1050 break;
1051 path->slots[0]--;
1052 }
1053
1054 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1055 if (key.objectid != BTRFS_TREE_LOG_FIXUP_OBJECTID ||
1056 key.type != BTRFS_ORPHAN_ITEM_KEY)
1057 break;
1058
1059 ret = btrfs_del_item(trans, root, path);
Tsutomu Itoh65a246c2011-05-19 04:37:44 +00001060 if (ret)
1061 goto out;
Chris Masone02119d2008-09-05 16:13:11 -04001062
David Sterbab3b4aa72011-04-21 01:20:15 +02001063 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04001064 inode = read_one_inode(root, key.offset);
Tsutomu Itohc00e9492011-04-28 09:10:23 +00001065 if (!inode)
1066 return -EIO;
Chris Masone02119d2008-09-05 16:13:11 -04001067
1068 ret = fixup_inode_link_count(trans, root, inode);
1069 BUG_ON(ret);
1070
1071 iput(inode);
1072
Chris Mason12fcfd22009-03-24 10:24:20 -04001073 /*
1074 * fixup on a directory may create new entries,
1075 * make sure we always look for the highset possible
1076 * offset
1077 */
1078 key.offset = (u64)-1;
Chris Masone02119d2008-09-05 16:13:11 -04001079 }
Tsutomu Itoh65a246c2011-05-19 04:37:44 +00001080 ret = 0;
1081out:
David Sterbab3b4aa72011-04-21 01:20:15 +02001082 btrfs_release_path(path);
Tsutomu Itoh65a246c2011-05-19 04:37:44 +00001083 return ret;
Chris Masone02119d2008-09-05 16:13:11 -04001084}
1085
1086
1087/*
1088 * record a given inode in the fixup dir so we can check its link
1089 * count when replay is done. The link count is incremented here
1090 * so the inode won't go away until we check it
1091 */
1092static noinline int link_to_fixup_dir(struct btrfs_trans_handle *trans,
1093 struct btrfs_root *root,
1094 struct btrfs_path *path,
1095 u64 objectid)
1096{
1097 struct btrfs_key key;
1098 int ret = 0;
1099 struct inode *inode;
1100
1101 inode = read_one_inode(root, objectid);
Tsutomu Itohc00e9492011-04-28 09:10:23 +00001102 if (!inode)
1103 return -EIO;
Chris Masone02119d2008-09-05 16:13:11 -04001104
1105 key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID;
1106 btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
1107 key.offset = objectid;
1108
1109 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1110
David Sterbab3b4aa72011-04-21 01:20:15 +02001111 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04001112 if (ret == 0) {
1113 btrfs_inc_nlink(inode);
1114 btrfs_update_inode(trans, root, inode);
1115 } else if (ret == -EEXIST) {
1116 ret = 0;
1117 } else {
1118 BUG();
1119 }
1120 iput(inode);
1121
1122 return ret;
1123}
1124
1125/*
1126 * when replaying the log for a directory, we only insert names
1127 * for inodes that actually exist. This means an fsync on a directory
1128 * does not implicitly fsync all the new files in it
1129 */
1130static noinline int insert_one_name(struct btrfs_trans_handle *trans,
1131 struct btrfs_root *root,
1132 struct btrfs_path *path,
1133 u64 dirid, u64 index,
1134 char *name, int name_len, u8 type,
1135 struct btrfs_key *location)
1136{
1137 struct inode *inode;
1138 struct inode *dir;
1139 int ret;
1140
1141 inode = read_one_inode(root, location->objectid);
1142 if (!inode)
1143 return -ENOENT;
1144
1145 dir = read_one_inode(root, dirid);
1146 if (!dir) {
1147 iput(inode);
1148 return -EIO;
1149 }
1150 ret = btrfs_add_link(trans, dir, inode, name, name_len, 1, index);
1151
1152 /* FIXME, put inode into FIXUP list */
1153
1154 iput(inode);
1155 iput(dir);
1156 return ret;
1157}
1158
1159/*
1160 * take a single entry in a log directory item and replay it into
1161 * the subvolume.
1162 *
1163 * if a conflicting item exists in the subdirectory already,
1164 * the inode it points to is unlinked and put into the link count
1165 * fix up tree.
1166 *
1167 * If a name from the log points to a file or directory that does
1168 * not exist in the FS, it is skipped. fsyncs on directories
1169 * do not force down inodes inside that directory, just changes to the
1170 * names or unlinks in a directory.
1171 */
1172static noinline int replay_one_name(struct btrfs_trans_handle *trans,
1173 struct btrfs_root *root,
1174 struct btrfs_path *path,
1175 struct extent_buffer *eb,
1176 struct btrfs_dir_item *di,
1177 struct btrfs_key *key)
1178{
1179 char *name;
1180 int name_len;
1181 struct btrfs_dir_item *dst_di;
1182 struct btrfs_key found_key;
1183 struct btrfs_key log_key;
1184 struct inode *dir;
Chris Masone02119d2008-09-05 16:13:11 -04001185 u8 log_type;
Chris Mason4bef0842008-09-08 11:18:08 -04001186 int exists;
Chris Masone02119d2008-09-05 16:13:11 -04001187 int ret;
1188
1189 dir = read_one_inode(root, key->objectid);
Tsutomu Itohc00e9492011-04-28 09:10:23 +00001190 if (!dir)
1191 return -EIO;
Chris Masone02119d2008-09-05 16:13:11 -04001192
1193 name_len = btrfs_dir_name_len(eb, di);
1194 name = kmalloc(name_len, GFP_NOFS);
liubo2a29edc2011-01-26 06:22:08 +00001195 if (!name)
1196 return -ENOMEM;
1197
Chris Masone02119d2008-09-05 16:13:11 -04001198 log_type = btrfs_dir_type(eb, di);
1199 read_extent_buffer(eb, name, (unsigned long)(di + 1),
1200 name_len);
1201
1202 btrfs_dir_item_key_to_cpu(eb, di, &log_key);
Chris Mason4bef0842008-09-08 11:18:08 -04001203 exists = btrfs_lookup_inode(trans, root, path, &log_key, 0);
1204 if (exists == 0)
1205 exists = 1;
1206 else
1207 exists = 0;
David Sterbab3b4aa72011-04-21 01:20:15 +02001208 btrfs_release_path(path);
Chris Mason4bef0842008-09-08 11:18:08 -04001209
Chris Masone02119d2008-09-05 16:13:11 -04001210 if (key->type == BTRFS_DIR_ITEM_KEY) {
1211 dst_di = btrfs_lookup_dir_item(trans, root, path, key->objectid,
1212 name, name_len, 1);
Chris Masond3977122009-01-05 21:25:51 -05001213 } else if (key->type == BTRFS_DIR_INDEX_KEY) {
Chris Masone02119d2008-09-05 16:13:11 -04001214 dst_di = btrfs_lookup_dir_index_item(trans, root, path,
1215 key->objectid,
1216 key->offset, name,
1217 name_len, 1);
1218 } else {
1219 BUG();
1220 }
David Sterbac7040052011-04-19 18:00:01 +02001221 if (IS_ERR_OR_NULL(dst_di)) {
Chris Masone02119d2008-09-05 16:13:11 -04001222 /* we need a sequence number to insert, so we only
1223 * do inserts for the BTRFS_DIR_INDEX_KEY types
1224 */
1225 if (key->type != BTRFS_DIR_INDEX_KEY)
1226 goto out;
1227 goto insert;
1228 }
1229
1230 btrfs_dir_item_key_to_cpu(path->nodes[0], dst_di, &found_key);
1231 /* the existing item matches the logged item */
1232 if (found_key.objectid == log_key.objectid &&
1233 found_key.type == log_key.type &&
1234 found_key.offset == log_key.offset &&
1235 btrfs_dir_type(path->nodes[0], dst_di) == log_type) {
1236 goto out;
1237 }
1238
1239 /*
1240 * don't drop the conflicting directory entry if the inode
1241 * for the new entry doesn't exist
1242 */
Chris Mason4bef0842008-09-08 11:18:08 -04001243 if (!exists)
Chris Masone02119d2008-09-05 16:13:11 -04001244 goto out;
1245
Chris Masone02119d2008-09-05 16:13:11 -04001246 ret = drop_one_dir_item(trans, root, path, dir, dst_di);
1247 BUG_ON(ret);
1248
1249 if (key->type == BTRFS_DIR_INDEX_KEY)
1250 goto insert;
1251out:
David Sterbab3b4aa72011-04-21 01:20:15 +02001252 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04001253 kfree(name);
1254 iput(dir);
1255 return 0;
1256
1257insert:
David Sterbab3b4aa72011-04-21 01:20:15 +02001258 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04001259 ret = insert_one_name(trans, root, path, key->objectid, key->offset,
1260 name, name_len, log_type, &log_key);
1261
Stoyan Gaydarovc2934982009-04-02 17:05:11 -04001262 BUG_ON(ret && ret != -ENOENT);
Chris Masone02119d2008-09-05 16:13:11 -04001263 goto out;
1264}
1265
1266/*
1267 * find all the names in a directory item and reconcile them into
1268 * the subvolume. Only BTRFS_DIR_ITEM_KEY types will have more than
1269 * one name in a directory item, but the same code gets used for
1270 * both directory index types
1271 */
1272static noinline int replay_one_dir_item(struct btrfs_trans_handle *trans,
1273 struct btrfs_root *root,
1274 struct btrfs_path *path,
1275 struct extent_buffer *eb, int slot,
1276 struct btrfs_key *key)
1277{
1278 int ret;
1279 u32 item_size = btrfs_item_size_nr(eb, slot);
1280 struct btrfs_dir_item *di;
1281 int name_len;
1282 unsigned long ptr;
1283 unsigned long ptr_end;
1284
1285 ptr = btrfs_item_ptr_offset(eb, slot);
1286 ptr_end = ptr + item_size;
Chris Masond3977122009-01-05 21:25:51 -05001287 while (ptr < ptr_end) {
Chris Masone02119d2008-09-05 16:13:11 -04001288 di = (struct btrfs_dir_item *)ptr;
Josef Bacik22a94d42011-03-16 16:47:17 -04001289 if (verify_dir_item(root, eb, di))
1290 return -EIO;
Chris Masone02119d2008-09-05 16:13:11 -04001291 name_len = btrfs_dir_name_len(eb, di);
1292 ret = replay_one_name(trans, root, path, eb, di, key);
1293 BUG_ON(ret);
1294 ptr = (unsigned long)(di + 1);
1295 ptr += name_len;
1296 }
1297 return 0;
1298}
1299
1300/*
1301 * directory replay has two parts. There are the standard directory
1302 * items in the log copied from the subvolume, and range items
1303 * created in the log while the subvolume was logged.
1304 *
1305 * The range items tell us which parts of the key space the log
1306 * is authoritative for. During replay, if a key in the subvolume
1307 * directory is in a logged range item, but not actually in the log
1308 * that means it was deleted from the directory before the fsync
1309 * and should be removed.
1310 */
1311static noinline int find_dir_range(struct btrfs_root *root,
1312 struct btrfs_path *path,
1313 u64 dirid, int key_type,
1314 u64 *start_ret, u64 *end_ret)
1315{
1316 struct btrfs_key key;
1317 u64 found_end;
1318 struct btrfs_dir_log_item *item;
1319 int ret;
1320 int nritems;
1321
1322 if (*start_ret == (u64)-1)
1323 return 1;
1324
1325 key.objectid = dirid;
1326 key.type = key_type;
1327 key.offset = *start_ret;
1328
1329 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1330 if (ret < 0)
1331 goto out;
1332 if (ret > 0) {
1333 if (path->slots[0] == 0)
1334 goto out;
1335 path->slots[0]--;
1336 }
1337 if (ret != 0)
1338 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1339
1340 if (key.type != key_type || key.objectid != dirid) {
1341 ret = 1;
1342 goto next;
1343 }
1344 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
1345 struct btrfs_dir_log_item);
1346 found_end = btrfs_dir_log_end(path->nodes[0], item);
1347
1348 if (*start_ret >= key.offset && *start_ret <= found_end) {
1349 ret = 0;
1350 *start_ret = key.offset;
1351 *end_ret = found_end;
1352 goto out;
1353 }
1354 ret = 1;
1355next:
1356 /* check the next slot in the tree to see if it is a valid item */
1357 nritems = btrfs_header_nritems(path->nodes[0]);
1358 if (path->slots[0] >= nritems) {
1359 ret = btrfs_next_leaf(root, path);
1360 if (ret)
1361 goto out;
1362 } else {
1363 path->slots[0]++;
1364 }
1365
1366 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1367
1368 if (key.type != key_type || key.objectid != dirid) {
1369 ret = 1;
1370 goto out;
1371 }
1372 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
1373 struct btrfs_dir_log_item);
1374 found_end = btrfs_dir_log_end(path->nodes[0], item);
1375 *start_ret = key.offset;
1376 *end_ret = found_end;
1377 ret = 0;
1378out:
David Sterbab3b4aa72011-04-21 01:20:15 +02001379 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04001380 return ret;
1381}
1382
1383/*
1384 * this looks for a given directory item in the log. If the directory
1385 * item is not in the log, the item is removed and the inode it points
1386 * to is unlinked
1387 */
1388static noinline int check_item_in_log(struct btrfs_trans_handle *trans,
1389 struct btrfs_root *root,
1390 struct btrfs_root *log,
1391 struct btrfs_path *path,
1392 struct btrfs_path *log_path,
1393 struct inode *dir,
1394 struct btrfs_key *dir_key)
1395{
1396 int ret;
1397 struct extent_buffer *eb;
1398 int slot;
1399 u32 item_size;
1400 struct btrfs_dir_item *di;
1401 struct btrfs_dir_item *log_di;
1402 int name_len;
1403 unsigned long ptr;
1404 unsigned long ptr_end;
1405 char *name;
1406 struct inode *inode;
1407 struct btrfs_key location;
1408
1409again:
1410 eb = path->nodes[0];
1411 slot = path->slots[0];
1412 item_size = btrfs_item_size_nr(eb, slot);
1413 ptr = btrfs_item_ptr_offset(eb, slot);
1414 ptr_end = ptr + item_size;
Chris Masond3977122009-01-05 21:25:51 -05001415 while (ptr < ptr_end) {
Chris Masone02119d2008-09-05 16:13:11 -04001416 di = (struct btrfs_dir_item *)ptr;
Josef Bacik22a94d42011-03-16 16:47:17 -04001417 if (verify_dir_item(root, eb, di)) {
1418 ret = -EIO;
1419 goto out;
1420 }
1421
Chris Masone02119d2008-09-05 16:13:11 -04001422 name_len = btrfs_dir_name_len(eb, di);
1423 name = kmalloc(name_len, GFP_NOFS);
1424 if (!name) {
1425 ret = -ENOMEM;
1426 goto out;
1427 }
1428 read_extent_buffer(eb, name, (unsigned long)(di + 1),
1429 name_len);
1430 log_di = NULL;
Chris Mason12fcfd22009-03-24 10:24:20 -04001431 if (log && dir_key->type == BTRFS_DIR_ITEM_KEY) {
Chris Masone02119d2008-09-05 16:13:11 -04001432 log_di = btrfs_lookup_dir_item(trans, log, log_path,
1433 dir_key->objectid,
1434 name, name_len, 0);
Chris Mason12fcfd22009-03-24 10:24:20 -04001435 } else if (log && dir_key->type == BTRFS_DIR_INDEX_KEY) {
Chris Masone02119d2008-09-05 16:13:11 -04001436 log_di = btrfs_lookup_dir_index_item(trans, log,
1437 log_path,
1438 dir_key->objectid,
1439 dir_key->offset,
1440 name, name_len, 0);
1441 }
David Sterbac7040052011-04-19 18:00:01 +02001442 if (IS_ERR_OR_NULL(log_di)) {
Chris Masone02119d2008-09-05 16:13:11 -04001443 btrfs_dir_item_key_to_cpu(eb, di, &location);
David Sterbab3b4aa72011-04-21 01:20:15 +02001444 btrfs_release_path(path);
1445 btrfs_release_path(log_path);
Chris Masone02119d2008-09-05 16:13:11 -04001446 inode = read_one_inode(root, location.objectid);
Tsutomu Itohc00e9492011-04-28 09:10:23 +00001447 if (!inode) {
1448 kfree(name);
1449 return -EIO;
1450 }
Chris Masone02119d2008-09-05 16:13:11 -04001451
1452 ret = link_to_fixup_dir(trans, root,
1453 path, location.objectid);
1454 BUG_ON(ret);
1455 btrfs_inc_nlink(inode);
1456 ret = btrfs_unlink_inode(trans, root, dir, inode,
1457 name, name_len);
1458 BUG_ON(ret);
1459 kfree(name);
1460 iput(inode);
1461
1462 /* there might still be more names under this key
1463 * check and repeat if required
1464 */
1465 ret = btrfs_search_slot(NULL, root, dir_key, path,
1466 0, 0);
1467 if (ret == 0)
1468 goto again;
1469 ret = 0;
1470 goto out;
1471 }
David Sterbab3b4aa72011-04-21 01:20:15 +02001472 btrfs_release_path(log_path);
Chris Masone02119d2008-09-05 16:13:11 -04001473 kfree(name);
1474
1475 ptr = (unsigned long)(di + 1);
1476 ptr += name_len;
1477 }
1478 ret = 0;
1479out:
David Sterbab3b4aa72011-04-21 01:20:15 +02001480 btrfs_release_path(path);
1481 btrfs_release_path(log_path);
Chris Masone02119d2008-09-05 16:13:11 -04001482 return ret;
1483}
1484
1485/*
1486 * deletion replay happens before we copy any new directory items
1487 * out of the log or out of backreferences from inodes. It
1488 * scans the log to find ranges of keys that log is authoritative for,
1489 * and then scans the directory to find items in those ranges that are
1490 * not present in the log.
1491 *
1492 * Anything we don't find in the log is unlinked and removed from the
1493 * directory.
1494 */
1495static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans,
1496 struct btrfs_root *root,
1497 struct btrfs_root *log,
1498 struct btrfs_path *path,
Chris Mason12fcfd22009-03-24 10:24:20 -04001499 u64 dirid, int del_all)
Chris Masone02119d2008-09-05 16:13:11 -04001500{
1501 u64 range_start;
1502 u64 range_end;
1503 int key_type = BTRFS_DIR_LOG_ITEM_KEY;
1504 int ret = 0;
1505 struct btrfs_key dir_key;
1506 struct btrfs_key found_key;
1507 struct btrfs_path *log_path;
1508 struct inode *dir;
1509
1510 dir_key.objectid = dirid;
1511 dir_key.type = BTRFS_DIR_ITEM_KEY;
1512 log_path = btrfs_alloc_path();
1513 if (!log_path)
1514 return -ENOMEM;
1515
1516 dir = read_one_inode(root, dirid);
1517 /* it isn't an error if the inode isn't there, that can happen
1518 * because we replay the deletes before we copy in the inode item
1519 * from the log
1520 */
1521 if (!dir) {
1522 btrfs_free_path(log_path);
1523 return 0;
1524 }
1525again:
1526 range_start = 0;
1527 range_end = 0;
Chris Masond3977122009-01-05 21:25:51 -05001528 while (1) {
Chris Mason12fcfd22009-03-24 10:24:20 -04001529 if (del_all)
1530 range_end = (u64)-1;
1531 else {
1532 ret = find_dir_range(log, path, dirid, key_type,
1533 &range_start, &range_end);
1534 if (ret != 0)
1535 break;
1536 }
Chris Masone02119d2008-09-05 16:13:11 -04001537
1538 dir_key.offset = range_start;
Chris Masond3977122009-01-05 21:25:51 -05001539 while (1) {
Chris Masone02119d2008-09-05 16:13:11 -04001540 int nritems;
1541 ret = btrfs_search_slot(NULL, root, &dir_key, path,
1542 0, 0);
1543 if (ret < 0)
1544 goto out;
1545
1546 nritems = btrfs_header_nritems(path->nodes[0]);
1547 if (path->slots[0] >= nritems) {
1548 ret = btrfs_next_leaf(root, path);
1549 if (ret)
1550 break;
1551 }
1552 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1553 path->slots[0]);
1554 if (found_key.objectid != dirid ||
1555 found_key.type != dir_key.type)
1556 goto next_type;
1557
1558 if (found_key.offset > range_end)
1559 break;
1560
1561 ret = check_item_in_log(trans, root, log, path,
Chris Mason12fcfd22009-03-24 10:24:20 -04001562 log_path, dir,
1563 &found_key);
Chris Masone02119d2008-09-05 16:13:11 -04001564 BUG_ON(ret);
1565 if (found_key.offset == (u64)-1)
1566 break;
1567 dir_key.offset = found_key.offset + 1;
1568 }
David Sterbab3b4aa72011-04-21 01:20:15 +02001569 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04001570 if (range_end == (u64)-1)
1571 break;
1572 range_start = range_end + 1;
1573 }
1574
1575next_type:
1576 ret = 0;
1577 if (key_type == BTRFS_DIR_LOG_ITEM_KEY) {
1578 key_type = BTRFS_DIR_LOG_INDEX_KEY;
1579 dir_key.type = BTRFS_DIR_INDEX_KEY;
David Sterbab3b4aa72011-04-21 01:20:15 +02001580 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04001581 goto again;
1582 }
1583out:
David Sterbab3b4aa72011-04-21 01:20:15 +02001584 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04001585 btrfs_free_path(log_path);
1586 iput(dir);
1587 return ret;
1588}
1589
1590/*
1591 * the process_func used to replay items from the log tree. This
1592 * gets called in two different stages. The first stage just looks
1593 * for inodes and makes sure they are all copied into the subvolume.
1594 *
1595 * The second stage copies all the other item types from the log into
1596 * the subvolume. The two stage approach is slower, but gets rid of
1597 * lots of complexity around inodes referencing other inodes that exist
1598 * only in the log (references come from either directory items or inode
1599 * back refs).
1600 */
1601static int replay_one_buffer(struct btrfs_root *log, struct extent_buffer *eb,
1602 struct walk_control *wc, u64 gen)
1603{
1604 int nritems;
1605 struct btrfs_path *path;
1606 struct btrfs_root *root = wc->replay_dest;
1607 struct btrfs_key key;
Chris Masone02119d2008-09-05 16:13:11 -04001608 int level;
1609 int i;
1610 int ret;
1611
1612 btrfs_read_buffer(eb, gen);
1613
1614 level = btrfs_header_level(eb);
1615
1616 if (level != 0)
1617 return 0;
1618
1619 path = btrfs_alloc_path();
1620 BUG_ON(!path);
1621
1622 nritems = btrfs_header_nritems(eb);
1623 for (i = 0; i < nritems; i++) {
1624 btrfs_item_key_to_cpu(eb, &key, i);
Chris Masone02119d2008-09-05 16:13:11 -04001625
1626 /* inode keys are done during the first stage */
1627 if (key.type == BTRFS_INODE_ITEM_KEY &&
1628 wc->stage == LOG_WALK_REPLAY_INODES) {
Chris Masone02119d2008-09-05 16:13:11 -04001629 struct btrfs_inode_item *inode_item;
1630 u32 mode;
1631
1632 inode_item = btrfs_item_ptr(eb, i,
1633 struct btrfs_inode_item);
1634 mode = btrfs_inode_mode(eb, inode_item);
1635 if (S_ISDIR(mode)) {
1636 ret = replay_dir_deletes(wc->trans,
Chris Mason12fcfd22009-03-24 10:24:20 -04001637 root, log, path, key.objectid, 0);
Chris Masone02119d2008-09-05 16:13:11 -04001638 BUG_ON(ret);
1639 }
1640 ret = overwrite_item(wc->trans, root, path,
1641 eb, i, &key);
1642 BUG_ON(ret);
1643
Yan, Zhengc71bf092009-11-12 09:34:40 +00001644 /* for regular files, make sure corresponding
1645 * orhpan item exist. extents past the new EOF
1646 * will be truncated later by orphan cleanup.
Chris Masone02119d2008-09-05 16:13:11 -04001647 */
1648 if (S_ISREG(mode)) {
Yan, Zhengc71bf092009-11-12 09:34:40 +00001649 ret = insert_orphan_item(wc->trans, root,
1650 key.objectid);
Chris Masone02119d2008-09-05 16:13:11 -04001651 BUG_ON(ret);
Chris Masone02119d2008-09-05 16:13:11 -04001652 }
Yan, Zhengc71bf092009-11-12 09:34:40 +00001653
Chris Masone02119d2008-09-05 16:13:11 -04001654 ret = link_to_fixup_dir(wc->trans, root,
1655 path, key.objectid);
1656 BUG_ON(ret);
1657 }
1658 if (wc->stage < LOG_WALK_REPLAY_ALL)
1659 continue;
1660
1661 /* these keys are simply copied */
1662 if (key.type == BTRFS_XATTR_ITEM_KEY) {
1663 ret = overwrite_item(wc->trans, root, path,
1664 eb, i, &key);
1665 BUG_ON(ret);
1666 } else if (key.type == BTRFS_INODE_REF_KEY) {
1667 ret = add_inode_ref(wc->trans, root, log, path,
1668 eb, i, &key);
1669 BUG_ON(ret && ret != -ENOENT);
1670 } else if (key.type == BTRFS_EXTENT_DATA_KEY) {
1671 ret = replay_one_extent(wc->trans, root, path,
1672 eb, i, &key);
1673 BUG_ON(ret);
Chris Masone02119d2008-09-05 16:13:11 -04001674 } else if (key.type == BTRFS_DIR_ITEM_KEY ||
1675 key.type == BTRFS_DIR_INDEX_KEY) {
1676 ret = replay_one_dir_item(wc->trans, root, path,
1677 eb, i, &key);
1678 BUG_ON(ret);
1679 }
1680 }
1681 btrfs_free_path(path);
1682 return 0;
1683}
1684
Chris Masond3977122009-01-05 21:25:51 -05001685static noinline int walk_down_log_tree(struct btrfs_trans_handle *trans,
Chris Masone02119d2008-09-05 16:13:11 -04001686 struct btrfs_root *root,
1687 struct btrfs_path *path, int *level,
1688 struct walk_control *wc)
1689{
1690 u64 root_owner;
Chris Masone02119d2008-09-05 16:13:11 -04001691 u64 bytenr;
1692 u64 ptr_gen;
1693 struct extent_buffer *next;
1694 struct extent_buffer *cur;
1695 struct extent_buffer *parent;
1696 u32 blocksize;
1697 int ret = 0;
1698
1699 WARN_ON(*level < 0);
1700 WARN_ON(*level >= BTRFS_MAX_LEVEL);
1701
Chris Masond3977122009-01-05 21:25:51 -05001702 while (*level > 0) {
Chris Masone02119d2008-09-05 16:13:11 -04001703 WARN_ON(*level < 0);
1704 WARN_ON(*level >= BTRFS_MAX_LEVEL);
1705 cur = path->nodes[*level];
1706
1707 if (btrfs_header_level(cur) != *level)
1708 WARN_ON(1);
1709
1710 if (path->slots[*level] >=
1711 btrfs_header_nritems(cur))
1712 break;
1713
1714 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
1715 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
1716 blocksize = btrfs_level_size(root, *level - 1);
1717
1718 parent = path->nodes[*level];
1719 root_owner = btrfs_header_owner(parent);
Chris Masone02119d2008-09-05 16:13:11 -04001720
1721 next = btrfs_find_create_tree_block(root, bytenr, blocksize);
liubo2a29edc2011-01-26 06:22:08 +00001722 if (!next)
1723 return -ENOMEM;
Chris Masone02119d2008-09-05 16:13:11 -04001724
Chris Masone02119d2008-09-05 16:13:11 -04001725 if (*level == 1) {
Yan, Zheng4a500fd2010-05-16 10:49:59 -04001726 wc->process_func(root, next, wc, ptr_gen);
1727
Chris Masone02119d2008-09-05 16:13:11 -04001728 path->slots[*level]++;
1729 if (wc->free) {
1730 btrfs_read_buffer(next, ptr_gen);
1731
1732 btrfs_tree_lock(next);
1733 clean_tree_block(trans, root, next);
Chris Masonb4ce94d2009-02-04 09:25:08 -05001734 btrfs_set_lock_blocking(next);
Chris Masone02119d2008-09-05 16:13:11 -04001735 btrfs_wait_tree_block_writeback(next);
1736 btrfs_tree_unlock(next);
1737
Chris Masone02119d2008-09-05 16:13:11 -04001738 WARN_ON(root_owner !=
1739 BTRFS_TREE_LOG_OBJECTID);
Chris Masond00aff02008-09-11 15:54:42 -04001740 ret = btrfs_free_reserved_extent(root,
1741 bytenr, blocksize);
Chris Masone02119d2008-09-05 16:13:11 -04001742 BUG_ON(ret);
1743 }
1744 free_extent_buffer(next);
1745 continue;
1746 }
1747 btrfs_read_buffer(next, ptr_gen);
1748
1749 WARN_ON(*level <= 0);
1750 if (path->nodes[*level-1])
1751 free_extent_buffer(path->nodes[*level-1]);
1752 path->nodes[*level-1] = next;
1753 *level = btrfs_header_level(next);
1754 path->slots[*level] = 0;
1755 cond_resched();
1756 }
1757 WARN_ON(*level < 0);
1758 WARN_ON(*level >= BTRFS_MAX_LEVEL);
1759
Yan, Zheng4a500fd2010-05-16 10:49:59 -04001760 path->slots[*level] = btrfs_header_nritems(path->nodes[*level]);
Chris Masone02119d2008-09-05 16:13:11 -04001761
1762 cond_resched();
1763 return 0;
1764}
1765
Chris Masond3977122009-01-05 21:25:51 -05001766static noinline int walk_up_log_tree(struct btrfs_trans_handle *trans,
Chris Masone02119d2008-09-05 16:13:11 -04001767 struct btrfs_root *root,
1768 struct btrfs_path *path, int *level,
1769 struct walk_control *wc)
1770{
1771 u64 root_owner;
Chris Masone02119d2008-09-05 16:13:11 -04001772 int i;
1773 int slot;
1774 int ret;
1775
Chris Masond3977122009-01-05 21:25:51 -05001776 for (i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) {
Chris Masone02119d2008-09-05 16:13:11 -04001777 slot = path->slots[i];
Yan, Zheng4a500fd2010-05-16 10:49:59 -04001778 if (slot + 1 < btrfs_header_nritems(path->nodes[i])) {
Chris Masone02119d2008-09-05 16:13:11 -04001779 path->slots[i]++;
1780 *level = i;
1781 WARN_ON(*level == 0);
1782 return 0;
1783 } else {
Zheng Yan31840ae2008-09-23 13:14:14 -04001784 struct extent_buffer *parent;
1785 if (path->nodes[*level] == root->node)
1786 parent = path->nodes[*level];
1787 else
1788 parent = path->nodes[*level + 1];
1789
1790 root_owner = btrfs_header_owner(parent);
Chris Masone02119d2008-09-05 16:13:11 -04001791 wc->process_func(root, path->nodes[*level], wc,
1792 btrfs_header_generation(path->nodes[*level]));
1793 if (wc->free) {
1794 struct extent_buffer *next;
1795
1796 next = path->nodes[*level];
1797
1798 btrfs_tree_lock(next);
1799 clean_tree_block(trans, root, next);
Chris Masonb4ce94d2009-02-04 09:25:08 -05001800 btrfs_set_lock_blocking(next);
Chris Masone02119d2008-09-05 16:13:11 -04001801 btrfs_wait_tree_block_writeback(next);
1802 btrfs_tree_unlock(next);
1803
Chris Masone02119d2008-09-05 16:13:11 -04001804 WARN_ON(root_owner != BTRFS_TREE_LOG_OBJECTID);
Chris Masond00aff02008-09-11 15:54:42 -04001805 ret = btrfs_free_reserved_extent(root,
Chris Masone02119d2008-09-05 16:13:11 -04001806 path->nodes[*level]->start,
Chris Masond00aff02008-09-11 15:54:42 -04001807 path->nodes[*level]->len);
Chris Masone02119d2008-09-05 16:13:11 -04001808 BUG_ON(ret);
1809 }
1810 free_extent_buffer(path->nodes[*level]);
1811 path->nodes[*level] = NULL;
1812 *level = i + 1;
1813 }
1814 }
1815 return 1;
1816}
1817
1818/*
1819 * drop the reference count on the tree rooted at 'snap'. This traverses
1820 * the tree freeing any blocks that have a ref count of zero after being
1821 * decremented.
1822 */
1823static int walk_log_tree(struct btrfs_trans_handle *trans,
1824 struct btrfs_root *log, struct walk_control *wc)
1825{
1826 int ret = 0;
1827 int wret;
1828 int level;
1829 struct btrfs_path *path;
1830 int i;
1831 int orig_level;
1832
1833 path = btrfs_alloc_path();
Tsutomu Itohdb5b4932011-03-23 08:14:16 +00001834 if (!path)
1835 return -ENOMEM;
Chris Masone02119d2008-09-05 16:13:11 -04001836
1837 level = btrfs_header_level(log->node);
1838 orig_level = level;
1839 path->nodes[level] = log->node;
1840 extent_buffer_get(log->node);
1841 path->slots[level] = 0;
1842
Chris Masond3977122009-01-05 21:25:51 -05001843 while (1) {
Chris Masone02119d2008-09-05 16:13:11 -04001844 wret = walk_down_log_tree(trans, log, path, &level, wc);
1845 if (wret > 0)
1846 break;
1847 if (wret < 0)
1848 ret = wret;
1849
1850 wret = walk_up_log_tree(trans, log, path, &level, wc);
1851 if (wret > 0)
1852 break;
1853 if (wret < 0)
1854 ret = wret;
1855 }
1856
1857 /* was the root node processed? if not, catch it here */
1858 if (path->nodes[orig_level]) {
1859 wc->process_func(log, path->nodes[orig_level], wc,
1860 btrfs_header_generation(path->nodes[orig_level]));
1861 if (wc->free) {
1862 struct extent_buffer *next;
1863
1864 next = path->nodes[orig_level];
1865
1866 btrfs_tree_lock(next);
1867 clean_tree_block(trans, log, next);
Chris Masonb4ce94d2009-02-04 09:25:08 -05001868 btrfs_set_lock_blocking(next);
Chris Masone02119d2008-09-05 16:13:11 -04001869 btrfs_wait_tree_block_writeback(next);
1870 btrfs_tree_unlock(next);
1871
Chris Masone02119d2008-09-05 16:13:11 -04001872 WARN_ON(log->root_key.objectid !=
1873 BTRFS_TREE_LOG_OBJECTID);
Chris Masond00aff02008-09-11 15:54:42 -04001874 ret = btrfs_free_reserved_extent(log, next->start,
1875 next->len);
Chris Masone02119d2008-09-05 16:13:11 -04001876 BUG_ON(ret);
1877 }
1878 }
1879
1880 for (i = 0; i <= orig_level; i++) {
1881 if (path->nodes[i]) {
1882 free_extent_buffer(path->nodes[i]);
1883 path->nodes[i] = NULL;
1884 }
1885 }
1886 btrfs_free_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04001887 return ret;
1888}
1889
Yan Zheng7237f182009-01-21 12:54:03 -05001890/*
1891 * helper function to update the item for a given subvolumes log root
1892 * in the tree of log roots
1893 */
1894static int update_log_root(struct btrfs_trans_handle *trans,
1895 struct btrfs_root *log)
1896{
1897 int ret;
1898
1899 if (log->log_transid == 1) {
1900 /* insert root item on the first sync */
1901 ret = btrfs_insert_root(trans, log->fs_info->log_root_tree,
1902 &log->root_key, &log->root_item);
1903 } else {
1904 ret = btrfs_update_root(trans, log->fs_info->log_root_tree,
1905 &log->root_key, &log->root_item);
1906 }
1907 return ret;
1908}
1909
Chris Mason12fcfd22009-03-24 10:24:20 -04001910static int wait_log_commit(struct btrfs_trans_handle *trans,
1911 struct btrfs_root *root, unsigned long transid)
Chris Masone02119d2008-09-05 16:13:11 -04001912{
1913 DEFINE_WAIT(wait);
Yan Zheng7237f182009-01-21 12:54:03 -05001914 int index = transid % 2;
Chris Masone02119d2008-09-05 16:13:11 -04001915
Yan Zheng7237f182009-01-21 12:54:03 -05001916 /*
1917 * we only allow two pending log transactions at a time,
1918 * so we know that if ours is more than 2 older than the
1919 * current transaction, we're done
1920 */
Chris Masone02119d2008-09-05 16:13:11 -04001921 do {
Yan Zheng7237f182009-01-21 12:54:03 -05001922 prepare_to_wait(&root->log_commit_wait[index],
1923 &wait, TASK_UNINTERRUPTIBLE);
1924 mutex_unlock(&root->log_mutex);
Chris Mason12fcfd22009-03-24 10:24:20 -04001925
1926 if (root->fs_info->last_trans_log_full_commit !=
1927 trans->transid && root->log_transid < transid + 2 &&
Yan Zheng7237f182009-01-21 12:54:03 -05001928 atomic_read(&root->log_commit[index]))
Chris Masone02119d2008-09-05 16:13:11 -04001929 schedule();
Chris Mason12fcfd22009-03-24 10:24:20 -04001930
Yan Zheng7237f182009-01-21 12:54:03 -05001931 finish_wait(&root->log_commit_wait[index], &wait);
1932 mutex_lock(&root->log_mutex);
1933 } while (root->log_transid < transid + 2 &&
1934 atomic_read(&root->log_commit[index]));
1935 return 0;
1936}
1937
Chris Mason12fcfd22009-03-24 10:24:20 -04001938static int wait_for_writer(struct btrfs_trans_handle *trans,
1939 struct btrfs_root *root)
Yan Zheng7237f182009-01-21 12:54:03 -05001940{
1941 DEFINE_WAIT(wait);
1942 while (atomic_read(&root->log_writers)) {
1943 prepare_to_wait(&root->log_writer_wait,
1944 &wait, TASK_UNINTERRUPTIBLE);
1945 mutex_unlock(&root->log_mutex);
Chris Mason12fcfd22009-03-24 10:24:20 -04001946 if (root->fs_info->last_trans_log_full_commit !=
1947 trans->transid && atomic_read(&root->log_writers))
Yan Zheng7237f182009-01-21 12:54:03 -05001948 schedule();
1949 mutex_lock(&root->log_mutex);
1950 finish_wait(&root->log_writer_wait, &wait);
1951 }
Chris Masone02119d2008-09-05 16:13:11 -04001952 return 0;
1953}
1954
1955/*
1956 * btrfs_sync_log does sends a given tree log down to the disk and
1957 * updates the super blocks to record it. When this call is done,
Chris Mason12fcfd22009-03-24 10:24:20 -04001958 * you know that any inodes previously logged are safely on disk only
1959 * if it returns 0.
1960 *
1961 * Any other return value means you need to call btrfs_commit_transaction.
1962 * Some of the edge cases for fsyncing directories that have had unlinks
1963 * or renames done in the past mean that sometimes the only safe
1964 * fsync is to commit the whole FS. When btrfs_sync_log returns -EAGAIN,
1965 * that has happened.
Chris Masone02119d2008-09-05 16:13:11 -04001966 */
1967int btrfs_sync_log(struct btrfs_trans_handle *trans,
1968 struct btrfs_root *root)
1969{
Yan Zheng7237f182009-01-21 12:54:03 -05001970 int index1;
1971 int index2;
Yan, Zheng8cef4e12009-11-12 09:33:26 +00001972 int mark;
Chris Masone02119d2008-09-05 16:13:11 -04001973 int ret;
Chris Masone02119d2008-09-05 16:13:11 -04001974 struct btrfs_root *log = root->log_root;
Yan Zheng7237f182009-01-21 12:54:03 -05001975 struct btrfs_root *log_root_tree = root->fs_info->log_root_tree;
Yan, Zheng8cef4e12009-11-12 09:33:26 +00001976 unsigned long log_transid = 0;
Chris Masone02119d2008-09-05 16:13:11 -04001977
Yan Zheng7237f182009-01-21 12:54:03 -05001978 mutex_lock(&root->log_mutex);
1979 index1 = root->log_transid % 2;
1980 if (atomic_read(&root->log_commit[index1])) {
Chris Mason12fcfd22009-03-24 10:24:20 -04001981 wait_log_commit(trans, root, root->log_transid);
Yan Zheng7237f182009-01-21 12:54:03 -05001982 mutex_unlock(&root->log_mutex);
1983 return 0;
Chris Masone02119d2008-09-05 16:13:11 -04001984 }
Yan Zheng7237f182009-01-21 12:54:03 -05001985 atomic_set(&root->log_commit[index1], 1);
1986
1987 /* wait for previous tree log sync to complete */
1988 if (atomic_read(&root->log_commit[(index1 + 1) % 2]))
Chris Mason12fcfd22009-03-24 10:24:20 -04001989 wait_log_commit(trans, root, root->log_transid - 1);
Chris Masone02119d2008-09-05 16:13:11 -04001990
Yan, Zheng86df7eb2009-10-14 09:24:59 -04001991 while (1) {
Yan Zheng7237f182009-01-21 12:54:03 -05001992 unsigned long batch = root->log_batch;
Yan, Zheng86df7eb2009-10-14 09:24:59 -04001993 if (root->log_multiple_pids) {
1994 mutex_unlock(&root->log_mutex);
1995 schedule_timeout_uninterruptible(1);
1996 mutex_lock(&root->log_mutex);
1997 }
Chris Mason12fcfd22009-03-24 10:24:20 -04001998 wait_for_writer(trans, root);
Yan Zheng7237f182009-01-21 12:54:03 -05001999 if (batch == root->log_batch)
Chris Masone02119d2008-09-05 16:13:11 -04002000 break;
2001 }
Chris Masond0c803c2008-09-11 16:17:57 -04002002
Chris Mason12fcfd22009-03-24 10:24:20 -04002003 /* bail out if we need to do a full commit */
2004 if (root->fs_info->last_trans_log_full_commit == trans->transid) {
2005 ret = -EAGAIN;
2006 mutex_unlock(&root->log_mutex);
2007 goto out;
2008 }
2009
Yan, Zheng8cef4e12009-11-12 09:33:26 +00002010 log_transid = root->log_transid;
2011 if (log_transid % 2 == 0)
2012 mark = EXTENT_DIRTY;
2013 else
2014 mark = EXTENT_NEW;
2015
Chris Mason690587d2009-10-13 13:29:19 -04002016 /* we start IO on all the marked extents here, but we don't actually
2017 * wait for them until later.
2018 */
Yan, Zheng8cef4e12009-11-12 09:33:26 +00002019 ret = btrfs_write_marked_extents(log, &log->dirty_log_pages, mark);
Chris Masone02119d2008-09-05 16:13:11 -04002020 BUG_ON(ret);
Yan Zheng7237f182009-01-21 12:54:03 -05002021
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002022 btrfs_set_root_node(&log->root_item, log->node);
Yan Zheng7237f182009-01-21 12:54:03 -05002023
2024 root->log_batch = 0;
2025 root->log_transid++;
2026 log->log_transid = root->log_transid;
Josef Bacikff782e02009-10-08 15:30:04 -04002027 root->log_start_pid = 0;
Yan Zheng7237f182009-01-21 12:54:03 -05002028 smp_mb();
2029 /*
Yan, Zheng8cef4e12009-11-12 09:33:26 +00002030 * IO has been started, blocks of the log tree have WRITTEN flag set
2031 * in their headers. new modifications of the log will be written to
2032 * new positions. so it's safe to allow log writers to go in.
Yan Zheng7237f182009-01-21 12:54:03 -05002033 */
2034 mutex_unlock(&root->log_mutex);
2035
2036 mutex_lock(&log_root_tree->log_mutex);
2037 log_root_tree->log_batch++;
2038 atomic_inc(&log_root_tree->log_writers);
2039 mutex_unlock(&log_root_tree->log_mutex);
2040
2041 ret = update_log_root(trans, log);
Yan Zheng7237f182009-01-21 12:54:03 -05002042
2043 mutex_lock(&log_root_tree->log_mutex);
2044 if (atomic_dec_and_test(&log_root_tree->log_writers)) {
2045 smp_mb();
2046 if (waitqueue_active(&log_root_tree->log_writer_wait))
2047 wake_up(&log_root_tree->log_writer_wait);
2048 }
2049
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002050 if (ret) {
2051 BUG_ON(ret != -ENOSPC);
2052 root->fs_info->last_trans_log_full_commit = trans->transid;
2053 btrfs_wait_marked_extents(log, &log->dirty_log_pages, mark);
2054 mutex_unlock(&log_root_tree->log_mutex);
2055 ret = -EAGAIN;
2056 goto out;
2057 }
2058
Yan Zheng7237f182009-01-21 12:54:03 -05002059 index2 = log_root_tree->log_transid % 2;
2060 if (atomic_read(&log_root_tree->log_commit[index2])) {
Yan, Zheng8cef4e12009-11-12 09:33:26 +00002061 btrfs_wait_marked_extents(log, &log->dirty_log_pages, mark);
Chris Mason12fcfd22009-03-24 10:24:20 -04002062 wait_log_commit(trans, log_root_tree,
2063 log_root_tree->log_transid);
Yan Zheng7237f182009-01-21 12:54:03 -05002064 mutex_unlock(&log_root_tree->log_mutex);
Chris Masonb31eabd2011-01-31 16:48:24 -05002065 ret = 0;
Yan Zheng7237f182009-01-21 12:54:03 -05002066 goto out;
2067 }
2068 atomic_set(&log_root_tree->log_commit[index2], 1);
2069
Chris Mason12fcfd22009-03-24 10:24:20 -04002070 if (atomic_read(&log_root_tree->log_commit[(index2 + 1) % 2])) {
2071 wait_log_commit(trans, log_root_tree,
2072 log_root_tree->log_transid - 1);
2073 }
Yan Zheng7237f182009-01-21 12:54:03 -05002074
Chris Mason12fcfd22009-03-24 10:24:20 -04002075 wait_for_writer(trans, log_root_tree);
2076
2077 /*
2078 * now that we've moved on to the tree of log tree roots,
2079 * check the full commit flag again
2080 */
2081 if (root->fs_info->last_trans_log_full_commit == trans->transid) {
Yan, Zheng8cef4e12009-11-12 09:33:26 +00002082 btrfs_wait_marked_extents(log, &log->dirty_log_pages, mark);
Chris Mason12fcfd22009-03-24 10:24:20 -04002083 mutex_unlock(&log_root_tree->log_mutex);
2084 ret = -EAGAIN;
2085 goto out_wake_log_root;
2086 }
Yan Zheng7237f182009-01-21 12:54:03 -05002087
2088 ret = btrfs_write_and_wait_marked_extents(log_root_tree,
Yan, Zheng8cef4e12009-11-12 09:33:26 +00002089 &log_root_tree->dirty_log_pages,
2090 EXTENT_DIRTY | EXTENT_NEW);
Chris Masone02119d2008-09-05 16:13:11 -04002091 BUG_ON(ret);
Yan, Zheng8cef4e12009-11-12 09:33:26 +00002092 btrfs_wait_marked_extents(log, &log->dirty_log_pages, mark);
Chris Masone02119d2008-09-05 16:13:11 -04002093
2094 btrfs_set_super_log_root(&root->fs_info->super_for_commit,
Yan Zheng7237f182009-01-21 12:54:03 -05002095 log_root_tree->node->start);
Chris Masone02119d2008-09-05 16:13:11 -04002096 btrfs_set_super_log_root_level(&root->fs_info->super_for_commit,
Yan Zheng7237f182009-01-21 12:54:03 -05002097 btrfs_header_level(log_root_tree->node));
Chris Masone02119d2008-09-05 16:13:11 -04002098
Yan Zheng7237f182009-01-21 12:54:03 -05002099 log_root_tree->log_batch = 0;
2100 log_root_tree->log_transid++;
Chris Masone02119d2008-09-05 16:13:11 -04002101 smp_mb();
Yan Zheng7237f182009-01-21 12:54:03 -05002102
2103 mutex_unlock(&log_root_tree->log_mutex);
2104
2105 /*
2106 * nobody else is going to jump in and write the the ctree
2107 * super here because the log_commit atomic below is protecting
2108 * us. We must be called with a transaction handle pinning
2109 * the running transaction open, so a full commit can't hop
2110 * in and cause problems either.
2111 */
Arne Jansena2de7332011-03-08 14:14:00 +01002112 btrfs_scrub_pause_super(root);
Chris Mason47226072009-10-13 12:55:09 -04002113 write_ctree_super(trans, root->fs_info->tree_root, 1);
Arne Jansena2de7332011-03-08 14:14:00 +01002114 btrfs_scrub_continue_super(root);
Chris Mason12fcfd22009-03-24 10:24:20 -04002115 ret = 0;
Yan Zheng7237f182009-01-21 12:54:03 -05002116
Chris Mason257c62e2009-10-13 13:21:08 -04002117 mutex_lock(&root->log_mutex);
2118 if (root->last_log_commit < log_transid)
2119 root->last_log_commit = log_transid;
2120 mutex_unlock(&root->log_mutex);
2121
Chris Mason12fcfd22009-03-24 10:24:20 -04002122out_wake_log_root:
Yan Zheng7237f182009-01-21 12:54:03 -05002123 atomic_set(&log_root_tree->log_commit[index2], 0);
2124 smp_mb();
2125 if (waitqueue_active(&log_root_tree->log_commit_wait[index2]))
2126 wake_up(&log_root_tree->log_commit_wait[index2]);
Chris Masone02119d2008-09-05 16:13:11 -04002127out:
Yan Zheng7237f182009-01-21 12:54:03 -05002128 atomic_set(&root->log_commit[index1], 0);
2129 smp_mb();
2130 if (waitqueue_active(&root->log_commit_wait[index1]))
2131 wake_up(&root->log_commit_wait[index1]);
Chris Masonb31eabd2011-01-31 16:48:24 -05002132 return ret;
Chris Masone02119d2008-09-05 16:13:11 -04002133}
2134
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002135static void free_log_tree(struct btrfs_trans_handle *trans,
2136 struct btrfs_root *log)
Chris Masone02119d2008-09-05 16:13:11 -04002137{
2138 int ret;
Chris Masond0c803c2008-09-11 16:17:57 -04002139 u64 start;
2140 u64 end;
Chris Masone02119d2008-09-05 16:13:11 -04002141 struct walk_control wc = {
2142 .free = 1,
2143 .process_func = process_one_buffer
2144 };
2145
Chris Masone02119d2008-09-05 16:13:11 -04002146 ret = walk_log_tree(trans, log, &wc);
2147 BUG_ON(ret);
2148
Chris Masond3977122009-01-05 21:25:51 -05002149 while (1) {
Chris Masond0c803c2008-09-11 16:17:57 -04002150 ret = find_first_extent_bit(&log->dirty_log_pages,
Yan, Zheng8cef4e12009-11-12 09:33:26 +00002151 0, &start, &end, EXTENT_DIRTY | EXTENT_NEW);
Chris Masond0c803c2008-09-11 16:17:57 -04002152 if (ret)
2153 break;
2154
Yan, Zheng8cef4e12009-11-12 09:33:26 +00002155 clear_extent_bits(&log->dirty_log_pages, start, end,
2156 EXTENT_DIRTY | EXTENT_NEW, GFP_NOFS);
Chris Masond0c803c2008-09-11 16:17:57 -04002157 }
2158
Yan Zheng7237f182009-01-21 12:54:03 -05002159 free_extent_buffer(log->node);
2160 kfree(log);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002161}
2162
2163/*
2164 * free all the extents used by the tree log. This should be called
2165 * at commit time of the full transaction
2166 */
2167int btrfs_free_log(struct btrfs_trans_handle *trans, struct btrfs_root *root)
2168{
2169 if (root->log_root) {
2170 free_log_tree(trans, root->log_root);
2171 root->log_root = NULL;
2172 }
2173 return 0;
2174}
2175
2176int btrfs_free_log_root_tree(struct btrfs_trans_handle *trans,
2177 struct btrfs_fs_info *fs_info)
2178{
2179 if (fs_info->log_root_tree) {
2180 free_log_tree(trans, fs_info->log_root_tree);
2181 fs_info->log_root_tree = NULL;
2182 }
Chris Masone02119d2008-09-05 16:13:11 -04002183 return 0;
2184}
2185
2186/*
Chris Masone02119d2008-09-05 16:13:11 -04002187 * If both a file and directory are logged, and unlinks or renames are
2188 * mixed in, we have a few interesting corners:
2189 *
2190 * create file X in dir Y
2191 * link file X to X.link in dir Y
2192 * fsync file X
2193 * unlink file X but leave X.link
2194 * fsync dir Y
2195 *
2196 * After a crash we would expect only X.link to exist. But file X
2197 * didn't get fsync'd again so the log has back refs for X and X.link.
2198 *
2199 * We solve this by removing directory entries and inode backrefs from the
2200 * log when a file that was logged in the current transaction is
2201 * unlinked. Any later fsync will include the updated log entries, and
2202 * we'll be able to reconstruct the proper directory items from backrefs.
2203 *
2204 * This optimizations allows us to avoid relogging the entire inode
2205 * or the entire directory.
2206 */
2207int btrfs_del_dir_entries_in_log(struct btrfs_trans_handle *trans,
2208 struct btrfs_root *root,
2209 const char *name, int name_len,
2210 struct inode *dir, u64 index)
2211{
2212 struct btrfs_root *log;
2213 struct btrfs_dir_item *di;
2214 struct btrfs_path *path;
2215 int ret;
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002216 int err = 0;
Chris Masone02119d2008-09-05 16:13:11 -04002217 int bytes_del = 0;
Li Zefan33345d012011-04-20 10:31:50 +08002218 u64 dir_ino = btrfs_ino(dir);
Chris Masone02119d2008-09-05 16:13:11 -04002219
Chris Mason3a5f1d42008-09-11 15:53:37 -04002220 if (BTRFS_I(dir)->logged_trans < trans->transid)
2221 return 0;
2222
Chris Masone02119d2008-09-05 16:13:11 -04002223 ret = join_running_log_trans(root);
2224 if (ret)
2225 return 0;
2226
2227 mutex_lock(&BTRFS_I(dir)->log_mutex);
2228
2229 log = root->log_root;
2230 path = btrfs_alloc_path();
Tsutomu Itoha62f44a2011-04-25 19:43:51 -04002231 if (!path) {
2232 err = -ENOMEM;
2233 goto out_unlock;
2234 }
liubo2a29edc2011-01-26 06:22:08 +00002235
Li Zefan33345d012011-04-20 10:31:50 +08002236 di = btrfs_lookup_dir_item(trans, log, path, dir_ino,
Chris Masone02119d2008-09-05 16:13:11 -04002237 name, name_len, -1);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002238 if (IS_ERR(di)) {
2239 err = PTR_ERR(di);
2240 goto fail;
2241 }
2242 if (di) {
Chris Masone02119d2008-09-05 16:13:11 -04002243 ret = btrfs_delete_one_dir_name(trans, log, path, di);
2244 bytes_del += name_len;
2245 BUG_ON(ret);
2246 }
David Sterbab3b4aa72011-04-21 01:20:15 +02002247 btrfs_release_path(path);
Li Zefan33345d012011-04-20 10:31:50 +08002248 di = btrfs_lookup_dir_index_item(trans, log, path, dir_ino,
Chris Masone02119d2008-09-05 16:13:11 -04002249 index, name, name_len, -1);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002250 if (IS_ERR(di)) {
2251 err = PTR_ERR(di);
2252 goto fail;
2253 }
2254 if (di) {
Chris Masone02119d2008-09-05 16:13:11 -04002255 ret = btrfs_delete_one_dir_name(trans, log, path, di);
2256 bytes_del += name_len;
2257 BUG_ON(ret);
2258 }
2259
2260 /* update the directory size in the log to reflect the names
2261 * we have removed
2262 */
2263 if (bytes_del) {
2264 struct btrfs_key key;
2265
Li Zefan33345d012011-04-20 10:31:50 +08002266 key.objectid = dir_ino;
Chris Masone02119d2008-09-05 16:13:11 -04002267 key.offset = 0;
2268 key.type = BTRFS_INODE_ITEM_KEY;
David Sterbab3b4aa72011-04-21 01:20:15 +02002269 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04002270
2271 ret = btrfs_search_slot(trans, log, &key, path, 0, 1);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002272 if (ret < 0) {
2273 err = ret;
2274 goto fail;
2275 }
Chris Masone02119d2008-09-05 16:13:11 -04002276 if (ret == 0) {
2277 struct btrfs_inode_item *item;
2278 u64 i_size;
2279
2280 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2281 struct btrfs_inode_item);
2282 i_size = btrfs_inode_size(path->nodes[0], item);
2283 if (i_size > bytes_del)
2284 i_size -= bytes_del;
2285 else
2286 i_size = 0;
2287 btrfs_set_inode_size(path->nodes[0], item, i_size);
2288 btrfs_mark_buffer_dirty(path->nodes[0]);
2289 } else
2290 ret = 0;
David Sterbab3b4aa72011-04-21 01:20:15 +02002291 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04002292 }
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002293fail:
Chris Masone02119d2008-09-05 16:13:11 -04002294 btrfs_free_path(path);
Tsutomu Itoha62f44a2011-04-25 19:43:51 -04002295out_unlock:
Chris Masone02119d2008-09-05 16:13:11 -04002296 mutex_unlock(&BTRFS_I(dir)->log_mutex);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002297 if (ret == -ENOSPC) {
2298 root->fs_info->last_trans_log_full_commit = trans->transid;
2299 ret = 0;
2300 }
Chris Mason12fcfd22009-03-24 10:24:20 -04002301 btrfs_end_log_trans(root);
Chris Masone02119d2008-09-05 16:13:11 -04002302
Andi Kleen411fc6b2010-10-29 15:14:31 -04002303 return err;
Chris Masone02119d2008-09-05 16:13:11 -04002304}
2305
2306/* see comments for btrfs_del_dir_entries_in_log */
2307int btrfs_del_inode_ref_in_log(struct btrfs_trans_handle *trans,
2308 struct btrfs_root *root,
2309 const char *name, int name_len,
2310 struct inode *inode, u64 dirid)
2311{
2312 struct btrfs_root *log;
2313 u64 index;
2314 int ret;
2315
Chris Mason3a5f1d42008-09-11 15:53:37 -04002316 if (BTRFS_I(inode)->logged_trans < trans->transid)
2317 return 0;
2318
Chris Masone02119d2008-09-05 16:13:11 -04002319 ret = join_running_log_trans(root);
2320 if (ret)
2321 return 0;
2322 log = root->log_root;
2323 mutex_lock(&BTRFS_I(inode)->log_mutex);
2324
Li Zefan33345d012011-04-20 10:31:50 +08002325 ret = btrfs_del_inode_ref(trans, log, name, name_len, btrfs_ino(inode),
Chris Masone02119d2008-09-05 16:13:11 -04002326 dirid, &index);
2327 mutex_unlock(&BTRFS_I(inode)->log_mutex);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002328 if (ret == -ENOSPC) {
2329 root->fs_info->last_trans_log_full_commit = trans->transid;
2330 ret = 0;
2331 }
Chris Mason12fcfd22009-03-24 10:24:20 -04002332 btrfs_end_log_trans(root);
Chris Masone02119d2008-09-05 16:13:11 -04002333
Chris Masone02119d2008-09-05 16:13:11 -04002334 return ret;
2335}
2336
2337/*
2338 * creates a range item in the log for 'dirid'. first_offset and
2339 * last_offset tell us which parts of the key space the log should
2340 * be considered authoritative for.
2341 */
2342static noinline int insert_dir_log_key(struct btrfs_trans_handle *trans,
2343 struct btrfs_root *log,
2344 struct btrfs_path *path,
2345 int key_type, u64 dirid,
2346 u64 first_offset, u64 last_offset)
2347{
2348 int ret;
2349 struct btrfs_key key;
2350 struct btrfs_dir_log_item *item;
2351
2352 key.objectid = dirid;
2353 key.offset = first_offset;
2354 if (key_type == BTRFS_DIR_ITEM_KEY)
2355 key.type = BTRFS_DIR_LOG_ITEM_KEY;
2356 else
2357 key.type = BTRFS_DIR_LOG_INDEX_KEY;
2358 ret = btrfs_insert_empty_item(trans, log, path, &key, sizeof(*item));
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002359 if (ret)
2360 return ret;
Chris Masone02119d2008-09-05 16:13:11 -04002361
2362 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2363 struct btrfs_dir_log_item);
2364 btrfs_set_dir_log_end(path->nodes[0], item, last_offset);
2365 btrfs_mark_buffer_dirty(path->nodes[0]);
David Sterbab3b4aa72011-04-21 01:20:15 +02002366 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04002367 return 0;
2368}
2369
2370/*
2371 * log all the items included in the current transaction for a given
2372 * directory. This also creates the range items in the log tree required
2373 * to replay anything deleted before the fsync
2374 */
2375static noinline int log_dir_items(struct btrfs_trans_handle *trans,
2376 struct btrfs_root *root, struct inode *inode,
2377 struct btrfs_path *path,
2378 struct btrfs_path *dst_path, int key_type,
2379 u64 min_offset, u64 *last_offset_ret)
2380{
2381 struct btrfs_key min_key;
2382 struct btrfs_key max_key;
2383 struct btrfs_root *log = root->log_root;
2384 struct extent_buffer *src;
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002385 int err = 0;
Chris Masone02119d2008-09-05 16:13:11 -04002386 int ret;
2387 int i;
2388 int nritems;
2389 u64 first_offset = min_offset;
2390 u64 last_offset = (u64)-1;
Li Zefan33345d012011-04-20 10:31:50 +08002391 u64 ino = btrfs_ino(inode);
Chris Masone02119d2008-09-05 16:13:11 -04002392
2393 log = root->log_root;
Li Zefan33345d012011-04-20 10:31:50 +08002394 max_key.objectid = ino;
Chris Masone02119d2008-09-05 16:13:11 -04002395 max_key.offset = (u64)-1;
2396 max_key.type = key_type;
2397
Li Zefan33345d012011-04-20 10:31:50 +08002398 min_key.objectid = ino;
Chris Masone02119d2008-09-05 16:13:11 -04002399 min_key.type = key_type;
2400 min_key.offset = min_offset;
2401
2402 path->keep_locks = 1;
2403
2404 ret = btrfs_search_forward(root, &min_key, &max_key,
2405 path, 0, trans->transid);
2406
2407 /*
2408 * we didn't find anything from this transaction, see if there
2409 * is anything at all
2410 */
Li Zefan33345d012011-04-20 10:31:50 +08002411 if (ret != 0 || min_key.objectid != ino || min_key.type != key_type) {
2412 min_key.objectid = ino;
Chris Masone02119d2008-09-05 16:13:11 -04002413 min_key.type = key_type;
2414 min_key.offset = (u64)-1;
David Sterbab3b4aa72011-04-21 01:20:15 +02002415 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04002416 ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0);
2417 if (ret < 0) {
David Sterbab3b4aa72011-04-21 01:20:15 +02002418 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04002419 return ret;
2420 }
Li Zefan33345d012011-04-20 10:31:50 +08002421 ret = btrfs_previous_item(root, path, ino, key_type);
Chris Masone02119d2008-09-05 16:13:11 -04002422
2423 /* if ret == 0 there are items for this type,
2424 * create a range to tell us the last key of this type.
2425 * otherwise, there are no items in this directory after
2426 * *min_offset, and we create a range to indicate that.
2427 */
2428 if (ret == 0) {
2429 struct btrfs_key tmp;
2430 btrfs_item_key_to_cpu(path->nodes[0], &tmp,
2431 path->slots[0]);
Chris Masond3977122009-01-05 21:25:51 -05002432 if (key_type == tmp.type)
Chris Masone02119d2008-09-05 16:13:11 -04002433 first_offset = max(min_offset, tmp.offset) + 1;
Chris Masone02119d2008-09-05 16:13:11 -04002434 }
2435 goto done;
2436 }
2437
2438 /* go backward to find any previous key */
Li Zefan33345d012011-04-20 10:31:50 +08002439 ret = btrfs_previous_item(root, path, ino, key_type);
Chris Masone02119d2008-09-05 16:13:11 -04002440 if (ret == 0) {
2441 struct btrfs_key tmp;
2442 btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]);
2443 if (key_type == tmp.type) {
2444 first_offset = tmp.offset;
2445 ret = overwrite_item(trans, log, dst_path,
2446 path->nodes[0], path->slots[0],
2447 &tmp);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002448 if (ret) {
2449 err = ret;
2450 goto done;
2451 }
Chris Masone02119d2008-09-05 16:13:11 -04002452 }
2453 }
David Sterbab3b4aa72011-04-21 01:20:15 +02002454 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04002455
2456 /* find the first key from this transaction again */
2457 ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0);
2458 if (ret != 0) {
2459 WARN_ON(1);
2460 goto done;
2461 }
2462
2463 /*
2464 * we have a block from this transaction, log every item in it
2465 * from our directory
2466 */
Chris Masond3977122009-01-05 21:25:51 -05002467 while (1) {
Chris Masone02119d2008-09-05 16:13:11 -04002468 struct btrfs_key tmp;
2469 src = path->nodes[0];
2470 nritems = btrfs_header_nritems(src);
2471 for (i = path->slots[0]; i < nritems; i++) {
2472 btrfs_item_key_to_cpu(src, &min_key, i);
2473
Li Zefan33345d012011-04-20 10:31:50 +08002474 if (min_key.objectid != ino || min_key.type != key_type)
Chris Masone02119d2008-09-05 16:13:11 -04002475 goto done;
2476 ret = overwrite_item(trans, log, dst_path, src, i,
2477 &min_key);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002478 if (ret) {
2479 err = ret;
2480 goto done;
2481 }
Chris Masone02119d2008-09-05 16:13:11 -04002482 }
2483 path->slots[0] = nritems;
2484
2485 /*
2486 * look ahead to the next item and see if it is also
2487 * from this directory and from this transaction
2488 */
2489 ret = btrfs_next_leaf(root, path);
2490 if (ret == 1) {
2491 last_offset = (u64)-1;
2492 goto done;
2493 }
2494 btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]);
Li Zefan33345d012011-04-20 10:31:50 +08002495 if (tmp.objectid != ino || tmp.type != key_type) {
Chris Masone02119d2008-09-05 16:13:11 -04002496 last_offset = (u64)-1;
2497 goto done;
2498 }
2499 if (btrfs_header_generation(path->nodes[0]) != trans->transid) {
2500 ret = overwrite_item(trans, log, dst_path,
2501 path->nodes[0], path->slots[0],
2502 &tmp);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002503 if (ret)
2504 err = ret;
2505 else
2506 last_offset = tmp.offset;
Chris Masone02119d2008-09-05 16:13:11 -04002507 goto done;
2508 }
2509 }
2510done:
David Sterbab3b4aa72011-04-21 01:20:15 +02002511 btrfs_release_path(path);
2512 btrfs_release_path(dst_path);
Chris Masone02119d2008-09-05 16:13:11 -04002513
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002514 if (err == 0) {
2515 *last_offset_ret = last_offset;
2516 /*
2517 * insert the log range keys to indicate where the log
2518 * is valid
2519 */
2520 ret = insert_dir_log_key(trans, log, path, key_type,
Li Zefan33345d012011-04-20 10:31:50 +08002521 ino, first_offset, last_offset);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002522 if (ret)
2523 err = ret;
2524 }
2525 return err;
Chris Masone02119d2008-09-05 16:13:11 -04002526}
2527
2528/*
2529 * logging directories is very similar to logging inodes, We find all the items
2530 * from the current transaction and write them to the log.
2531 *
2532 * The recovery code scans the directory in the subvolume, and if it finds a
2533 * key in the range logged that is not present in the log tree, then it means
2534 * that dir entry was unlinked during the transaction.
2535 *
2536 * In order for that scan to work, we must include one key smaller than
2537 * the smallest logged by this transaction and one key larger than the largest
2538 * key logged by this transaction.
2539 */
2540static noinline int log_directory_changes(struct btrfs_trans_handle *trans,
2541 struct btrfs_root *root, struct inode *inode,
2542 struct btrfs_path *path,
2543 struct btrfs_path *dst_path)
2544{
2545 u64 min_key;
2546 u64 max_key;
2547 int ret;
2548 int key_type = BTRFS_DIR_ITEM_KEY;
2549
2550again:
2551 min_key = 0;
2552 max_key = 0;
Chris Masond3977122009-01-05 21:25:51 -05002553 while (1) {
Chris Masone02119d2008-09-05 16:13:11 -04002554 ret = log_dir_items(trans, root, inode, path,
2555 dst_path, key_type, min_key,
2556 &max_key);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002557 if (ret)
2558 return ret;
Chris Masone02119d2008-09-05 16:13:11 -04002559 if (max_key == (u64)-1)
2560 break;
2561 min_key = max_key + 1;
2562 }
2563
2564 if (key_type == BTRFS_DIR_ITEM_KEY) {
2565 key_type = BTRFS_DIR_INDEX_KEY;
2566 goto again;
2567 }
2568 return 0;
2569}
2570
2571/*
2572 * a helper function to drop items from the log before we relog an
2573 * inode. max_key_type indicates the highest item type to remove.
2574 * This cannot be run for file data extents because it does not
2575 * free the extents they point to.
2576 */
2577static int drop_objectid_items(struct btrfs_trans_handle *trans,
2578 struct btrfs_root *log,
2579 struct btrfs_path *path,
2580 u64 objectid, int max_key_type)
2581{
2582 int ret;
2583 struct btrfs_key key;
2584 struct btrfs_key found_key;
2585
2586 key.objectid = objectid;
2587 key.type = max_key_type;
2588 key.offset = (u64)-1;
2589
Chris Masond3977122009-01-05 21:25:51 -05002590 while (1) {
Chris Masone02119d2008-09-05 16:13:11 -04002591 ret = btrfs_search_slot(trans, log, &key, path, -1, 1);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002592 BUG_ON(ret == 0);
2593 if (ret < 0)
Chris Masone02119d2008-09-05 16:13:11 -04002594 break;
2595
2596 if (path->slots[0] == 0)
2597 break;
2598
2599 path->slots[0]--;
2600 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2601 path->slots[0]);
2602
2603 if (found_key.objectid != objectid)
2604 break;
2605
2606 ret = btrfs_del_item(trans, log, path);
Tsutomu Itoh65a246c2011-05-19 04:37:44 +00002607 if (ret)
2608 break;
David Sterbab3b4aa72011-04-21 01:20:15 +02002609 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04002610 }
David Sterbab3b4aa72011-04-21 01:20:15 +02002611 btrfs_release_path(path);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002612 return ret;
Chris Masone02119d2008-09-05 16:13:11 -04002613}
2614
Chris Mason31ff1cd2008-09-11 16:17:57 -04002615static noinline int copy_items(struct btrfs_trans_handle *trans,
2616 struct btrfs_root *log,
2617 struct btrfs_path *dst_path,
2618 struct extent_buffer *src,
2619 int start_slot, int nr, int inode_only)
2620{
2621 unsigned long src_offset;
2622 unsigned long dst_offset;
2623 struct btrfs_file_extent_item *extent;
2624 struct btrfs_inode_item *inode_item;
2625 int ret;
2626 struct btrfs_key *ins_keys;
2627 u32 *ins_sizes;
2628 char *ins_data;
2629 int i;
Chris Masond20f7042008-12-08 16:58:54 -05002630 struct list_head ordered_sums;
2631
2632 INIT_LIST_HEAD(&ordered_sums);
Chris Mason31ff1cd2008-09-11 16:17:57 -04002633
2634 ins_data = kmalloc(nr * sizeof(struct btrfs_key) +
2635 nr * sizeof(u32), GFP_NOFS);
liubo2a29edc2011-01-26 06:22:08 +00002636 if (!ins_data)
2637 return -ENOMEM;
2638
Chris Mason31ff1cd2008-09-11 16:17:57 -04002639 ins_sizes = (u32 *)ins_data;
2640 ins_keys = (struct btrfs_key *)(ins_data + nr * sizeof(u32));
2641
2642 for (i = 0; i < nr; i++) {
2643 ins_sizes[i] = btrfs_item_size_nr(src, i + start_slot);
2644 btrfs_item_key_to_cpu(src, ins_keys + i, i + start_slot);
2645 }
2646 ret = btrfs_insert_empty_items(trans, log, dst_path,
2647 ins_keys, ins_sizes, nr);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002648 if (ret) {
2649 kfree(ins_data);
2650 return ret;
2651 }
Chris Mason31ff1cd2008-09-11 16:17:57 -04002652
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002653 for (i = 0; i < nr; i++, dst_path->slots[0]++) {
Chris Mason31ff1cd2008-09-11 16:17:57 -04002654 dst_offset = btrfs_item_ptr_offset(dst_path->nodes[0],
2655 dst_path->slots[0]);
2656
2657 src_offset = btrfs_item_ptr_offset(src, start_slot + i);
2658
2659 copy_extent_buffer(dst_path->nodes[0], src, dst_offset,
2660 src_offset, ins_sizes[i]);
2661
2662 if (inode_only == LOG_INODE_EXISTS &&
2663 ins_keys[i].type == BTRFS_INODE_ITEM_KEY) {
2664 inode_item = btrfs_item_ptr(dst_path->nodes[0],
2665 dst_path->slots[0],
2666 struct btrfs_inode_item);
2667 btrfs_set_inode_size(dst_path->nodes[0], inode_item, 0);
2668
2669 /* set the generation to zero so the recover code
2670 * can tell the difference between an logging
2671 * just to say 'this inode exists' and a logging
2672 * to say 'update this inode with these values'
2673 */
2674 btrfs_set_inode_generation(dst_path->nodes[0],
2675 inode_item, 0);
2676 }
2677 /* take a reference on file data extents so that truncates
2678 * or deletes of this inode don't have to relog the inode
2679 * again
2680 */
2681 if (btrfs_key_type(ins_keys + i) == BTRFS_EXTENT_DATA_KEY) {
2682 int found_type;
2683 extent = btrfs_item_ptr(src, start_slot + i,
2684 struct btrfs_file_extent_item);
2685
liubo8e531cd2011-05-06 10:36:09 +08002686 if (btrfs_file_extent_generation(src, extent) < trans->transid)
2687 continue;
2688
Chris Mason31ff1cd2008-09-11 16:17:57 -04002689 found_type = btrfs_file_extent_type(src, extent);
Yan Zhengd899e052008-10-30 14:25:28 -04002690 if (found_type == BTRFS_FILE_EXTENT_REG ||
2691 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002692 u64 ds, dl, cs, cl;
2693 ds = btrfs_file_extent_disk_bytenr(src,
2694 extent);
2695 /* ds == 0 is a hole */
2696 if (ds == 0)
2697 continue;
2698
2699 dl = btrfs_file_extent_disk_num_bytes(src,
2700 extent);
2701 cs = btrfs_file_extent_offset(src, extent);
2702 cl = btrfs_file_extent_num_bytes(src,
Joe Perchesa419aef2009-08-18 11:18:35 -07002703 extent);
Chris Mason580afd72008-12-08 19:15:39 -05002704 if (btrfs_file_extent_compression(src,
2705 extent)) {
2706 cs = 0;
2707 cl = dl;
2708 }
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002709
2710 ret = btrfs_lookup_csums_range(
2711 log->fs_info->csum_root,
2712 ds + cs, ds + cs + cl - 1,
Arne Jansena2de7332011-03-08 14:14:00 +01002713 &ordered_sums, 0);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002714 BUG_ON(ret);
Chris Mason31ff1cd2008-09-11 16:17:57 -04002715 }
2716 }
Chris Mason31ff1cd2008-09-11 16:17:57 -04002717 }
2718
2719 btrfs_mark_buffer_dirty(dst_path->nodes[0]);
David Sterbab3b4aa72011-04-21 01:20:15 +02002720 btrfs_release_path(dst_path);
Chris Mason31ff1cd2008-09-11 16:17:57 -04002721 kfree(ins_data);
Chris Masond20f7042008-12-08 16:58:54 -05002722
2723 /*
2724 * we have to do this after the loop above to avoid changing the
2725 * log tree while trying to change the log tree.
2726 */
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002727 ret = 0;
Chris Masond3977122009-01-05 21:25:51 -05002728 while (!list_empty(&ordered_sums)) {
Chris Masond20f7042008-12-08 16:58:54 -05002729 struct btrfs_ordered_sum *sums = list_entry(ordered_sums.next,
2730 struct btrfs_ordered_sum,
2731 list);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002732 if (!ret)
2733 ret = btrfs_csum_file_blocks(trans, log, sums);
Chris Masond20f7042008-12-08 16:58:54 -05002734 list_del(&sums->list);
2735 kfree(sums);
2736 }
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002737 return ret;
Chris Mason31ff1cd2008-09-11 16:17:57 -04002738}
2739
Chris Masone02119d2008-09-05 16:13:11 -04002740/* log a single inode in the tree log.
2741 * At least one parent directory for this inode must exist in the tree
2742 * or be logged already.
2743 *
2744 * Any items from this inode changed by the current transaction are copied
2745 * to the log tree. An extra reference is taken on any extents in this
2746 * file, allowing us to avoid a whole pile of corner cases around logging
2747 * blocks that have been removed from the tree.
2748 *
2749 * See LOG_INODE_ALL and related defines for a description of what inode_only
2750 * does.
2751 *
2752 * This handles both files and directories.
2753 */
Chris Mason12fcfd22009-03-24 10:24:20 -04002754static int btrfs_log_inode(struct btrfs_trans_handle *trans,
Chris Masone02119d2008-09-05 16:13:11 -04002755 struct btrfs_root *root, struct inode *inode,
2756 int inode_only)
2757{
2758 struct btrfs_path *path;
2759 struct btrfs_path *dst_path;
2760 struct btrfs_key min_key;
2761 struct btrfs_key max_key;
2762 struct btrfs_root *log = root->log_root;
Chris Mason31ff1cd2008-09-11 16:17:57 -04002763 struct extent_buffer *src = NULL;
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002764 int err = 0;
Chris Masone02119d2008-09-05 16:13:11 -04002765 int ret;
Chris Mason3a5f1d42008-09-11 15:53:37 -04002766 int nritems;
Chris Mason31ff1cd2008-09-11 16:17:57 -04002767 int ins_start_slot = 0;
2768 int ins_nr;
Li Zefan33345d012011-04-20 10:31:50 +08002769 u64 ino = btrfs_ino(inode);
Chris Masone02119d2008-09-05 16:13:11 -04002770
2771 log = root->log_root;
2772
2773 path = btrfs_alloc_path();
Tsutomu Itoh5df67082011-02-01 09:17:35 +00002774 if (!path)
2775 return -ENOMEM;
Chris Masone02119d2008-09-05 16:13:11 -04002776 dst_path = btrfs_alloc_path();
Tsutomu Itoh5df67082011-02-01 09:17:35 +00002777 if (!dst_path) {
2778 btrfs_free_path(path);
2779 return -ENOMEM;
2780 }
Chris Masone02119d2008-09-05 16:13:11 -04002781
Li Zefan33345d012011-04-20 10:31:50 +08002782 min_key.objectid = ino;
Chris Masone02119d2008-09-05 16:13:11 -04002783 min_key.type = BTRFS_INODE_ITEM_KEY;
2784 min_key.offset = 0;
2785
Li Zefan33345d012011-04-20 10:31:50 +08002786 max_key.objectid = ino;
Chris Mason12fcfd22009-03-24 10:24:20 -04002787
2788 /* today the code can only do partial logging of directories */
2789 if (!S_ISDIR(inode->i_mode))
2790 inode_only = LOG_INODE_ALL;
2791
Chris Masone02119d2008-09-05 16:13:11 -04002792 if (inode_only == LOG_INODE_EXISTS || S_ISDIR(inode->i_mode))
2793 max_key.type = BTRFS_XATTR_ITEM_KEY;
2794 else
2795 max_key.type = (u8)-1;
2796 max_key.offset = (u64)-1;
2797
Miao Xie16cdcec2011-04-22 18:12:22 +08002798 ret = btrfs_commit_inode_delayed_items(trans, inode);
2799 if (ret) {
2800 btrfs_free_path(path);
2801 btrfs_free_path(dst_path);
2802 return ret;
2803 }
2804
Chris Masone02119d2008-09-05 16:13:11 -04002805 mutex_lock(&BTRFS_I(inode)->log_mutex);
2806
2807 /*
2808 * a brute force approach to making sure we get the most uptodate
2809 * copies of everything.
2810 */
2811 if (S_ISDIR(inode->i_mode)) {
2812 int max_key_type = BTRFS_DIR_LOG_INDEX_KEY;
2813
2814 if (inode_only == LOG_INODE_EXISTS)
2815 max_key_type = BTRFS_XATTR_ITEM_KEY;
Li Zefan33345d012011-04-20 10:31:50 +08002816 ret = drop_objectid_items(trans, log, path, ino, max_key_type);
Chris Masone02119d2008-09-05 16:13:11 -04002817 } else {
2818 ret = btrfs_truncate_inode_items(trans, log, inode, 0, 0);
2819 }
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002820 if (ret) {
2821 err = ret;
2822 goto out_unlock;
2823 }
Chris Masone02119d2008-09-05 16:13:11 -04002824 path->keep_locks = 1;
2825
Chris Masond3977122009-01-05 21:25:51 -05002826 while (1) {
Chris Mason31ff1cd2008-09-11 16:17:57 -04002827 ins_nr = 0;
Chris Masone02119d2008-09-05 16:13:11 -04002828 ret = btrfs_search_forward(root, &min_key, &max_key,
2829 path, 0, trans->transid);
2830 if (ret != 0)
2831 break;
Chris Mason3a5f1d42008-09-11 15:53:37 -04002832again:
Chris Mason31ff1cd2008-09-11 16:17:57 -04002833 /* note, ins_nr might be > 0 here, cleanup outside the loop */
Li Zefan33345d012011-04-20 10:31:50 +08002834 if (min_key.objectid != ino)
Chris Masone02119d2008-09-05 16:13:11 -04002835 break;
2836 if (min_key.type > max_key.type)
2837 break;
Chris Mason31ff1cd2008-09-11 16:17:57 -04002838
Chris Masone02119d2008-09-05 16:13:11 -04002839 src = path->nodes[0];
Chris Mason31ff1cd2008-09-11 16:17:57 -04002840 if (ins_nr && ins_start_slot + ins_nr == path->slots[0]) {
2841 ins_nr++;
2842 goto next_slot;
2843 } else if (!ins_nr) {
2844 ins_start_slot = path->slots[0];
2845 ins_nr = 1;
2846 goto next_slot;
Chris Masone02119d2008-09-05 16:13:11 -04002847 }
2848
Chris Mason31ff1cd2008-09-11 16:17:57 -04002849 ret = copy_items(trans, log, dst_path, src, ins_start_slot,
2850 ins_nr, inode_only);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002851 if (ret) {
2852 err = ret;
2853 goto out_unlock;
2854 }
Chris Mason31ff1cd2008-09-11 16:17:57 -04002855 ins_nr = 1;
2856 ins_start_slot = path->slots[0];
2857next_slot:
Chris Masone02119d2008-09-05 16:13:11 -04002858
Chris Mason3a5f1d42008-09-11 15:53:37 -04002859 nritems = btrfs_header_nritems(path->nodes[0]);
2860 path->slots[0]++;
2861 if (path->slots[0] < nritems) {
2862 btrfs_item_key_to_cpu(path->nodes[0], &min_key,
2863 path->slots[0]);
2864 goto again;
2865 }
Chris Mason31ff1cd2008-09-11 16:17:57 -04002866 if (ins_nr) {
2867 ret = copy_items(trans, log, dst_path, src,
2868 ins_start_slot,
2869 ins_nr, inode_only);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002870 if (ret) {
2871 err = ret;
2872 goto out_unlock;
2873 }
Chris Mason31ff1cd2008-09-11 16:17:57 -04002874 ins_nr = 0;
2875 }
David Sterbab3b4aa72011-04-21 01:20:15 +02002876 btrfs_release_path(path);
Chris Mason3a5f1d42008-09-11 15:53:37 -04002877
Chris Masone02119d2008-09-05 16:13:11 -04002878 if (min_key.offset < (u64)-1)
2879 min_key.offset++;
2880 else if (min_key.type < (u8)-1)
2881 min_key.type++;
2882 else if (min_key.objectid < (u64)-1)
2883 min_key.objectid++;
2884 else
2885 break;
2886 }
Chris Mason31ff1cd2008-09-11 16:17:57 -04002887 if (ins_nr) {
2888 ret = copy_items(trans, log, dst_path, src,
2889 ins_start_slot,
2890 ins_nr, inode_only);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002891 if (ret) {
2892 err = ret;
2893 goto out_unlock;
2894 }
Chris Mason31ff1cd2008-09-11 16:17:57 -04002895 ins_nr = 0;
2896 }
2897 WARN_ON(ins_nr);
Chris Mason9623f9a2008-09-11 17:42:42 -04002898 if (inode_only == LOG_INODE_ALL && S_ISDIR(inode->i_mode)) {
David Sterbab3b4aa72011-04-21 01:20:15 +02002899 btrfs_release_path(path);
2900 btrfs_release_path(dst_path);
Chris Masone02119d2008-09-05 16:13:11 -04002901 ret = log_directory_changes(trans, root, inode, path, dst_path);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002902 if (ret) {
2903 err = ret;
2904 goto out_unlock;
2905 }
Chris Masone02119d2008-09-05 16:13:11 -04002906 }
Chris Mason3a5f1d42008-09-11 15:53:37 -04002907 BTRFS_I(inode)->logged_trans = trans->transid;
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002908out_unlock:
Chris Masone02119d2008-09-05 16:13:11 -04002909 mutex_unlock(&BTRFS_I(inode)->log_mutex);
2910
2911 btrfs_free_path(path);
2912 btrfs_free_path(dst_path);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002913 return err;
Chris Masone02119d2008-09-05 16:13:11 -04002914}
2915
Chris Mason12fcfd22009-03-24 10:24:20 -04002916/*
2917 * follow the dentry parent pointers up the chain and see if any
2918 * of the directories in it require a full commit before they can
2919 * be logged. Returns zero if nothing special needs to be done or 1 if
2920 * a full commit is required.
2921 */
2922static noinline int check_parent_dirs_for_sync(struct btrfs_trans_handle *trans,
2923 struct inode *inode,
2924 struct dentry *parent,
2925 struct super_block *sb,
2926 u64 last_committed)
Chris Masone02119d2008-09-05 16:13:11 -04002927{
Chris Mason12fcfd22009-03-24 10:24:20 -04002928 int ret = 0;
2929 struct btrfs_root *root;
Josef Bacik6a912212010-11-20 09:48:00 +00002930 struct dentry *old_parent = NULL;
Chris Masone02119d2008-09-05 16:13:11 -04002931
Chris Masonaf4176b2009-03-24 10:24:31 -04002932 /*
2933 * for regular files, if its inode is already on disk, we don't
2934 * have to worry about the parents at all. This is because
2935 * we can use the last_unlink_trans field to record renames
2936 * and other fun in this file.
2937 */
2938 if (S_ISREG(inode->i_mode) &&
2939 BTRFS_I(inode)->generation <= last_committed &&
2940 BTRFS_I(inode)->last_unlink_trans <= last_committed)
2941 goto out;
2942
Chris Mason12fcfd22009-03-24 10:24:20 -04002943 if (!S_ISDIR(inode->i_mode)) {
2944 if (!parent || !parent->d_inode || sb != parent->d_inode->i_sb)
2945 goto out;
2946 inode = parent->d_inode;
2947 }
2948
2949 while (1) {
2950 BTRFS_I(inode)->logged_trans = trans->transid;
2951 smp_mb();
2952
2953 if (BTRFS_I(inode)->last_unlink_trans > last_committed) {
2954 root = BTRFS_I(inode)->root;
2955
2956 /*
2957 * make sure any commits to the log are forced
2958 * to be full commits
2959 */
2960 root->fs_info->last_trans_log_full_commit =
2961 trans->transid;
2962 ret = 1;
2963 break;
2964 }
2965
2966 if (!parent || !parent->d_inode || sb != parent->d_inode->i_sb)
2967 break;
2968
Yan, Zheng76dda932009-09-21 16:00:26 -04002969 if (IS_ROOT(parent))
Chris Mason12fcfd22009-03-24 10:24:20 -04002970 break;
2971
Josef Bacik6a912212010-11-20 09:48:00 +00002972 parent = dget_parent(parent);
2973 dput(old_parent);
2974 old_parent = parent;
Chris Mason12fcfd22009-03-24 10:24:20 -04002975 inode = parent->d_inode;
2976
2977 }
Josef Bacik6a912212010-11-20 09:48:00 +00002978 dput(old_parent);
Chris Mason12fcfd22009-03-24 10:24:20 -04002979out:
Chris Masone02119d2008-09-05 16:13:11 -04002980 return ret;
2981}
2982
Chris Mason257c62e2009-10-13 13:21:08 -04002983static int inode_in_log(struct btrfs_trans_handle *trans,
2984 struct inode *inode)
2985{
2986 struct btrfs_root *root = BTRFS_I(inode)->root;
2987 int ret = 0;
2988
2989 mutex_lock(&root->log_mutex);
2990 if (BTRFS_I(inode)->logged_trans == trans->transid &&
2991 BTRFS_I(inode)->last_sub_trans <= root->last_log_commit)
2992 ret = 1;
2993 mutex_unlock(&root->log_mutex);
2994 return ret;
2995}
2996
2997
Chris Masone02119d2008-09-05 16:13:11 -04002998/*
2999 * helper function around btrfs_log_inode to make sure newly created
3000 * parent directories also end up in the log. A minimal inode and backref
3001 * only logging is done of any parent directories that are older than
3002 * the last committed transaction
3003 */
Chris Mason12fcfd22009-03-24 10:24:20 -04003004int btrfs_log_inode_parent(struct btrfs_trans_handle *trans,
3005 struct btrfs_root *root, struct inode *inode,
3006 struct dentry *parent, int exists_only)
Chris Masone02119d2008-09-05 16:13:11 -04003007{
Chris Mason12fcfd22009-03-24 10:24:20 -04003008 int inode_only = exists_only ? LOG_INODE_EXISTS : LOG_INODE_ALL;
Chris Masone02119d2008-09-05 16:13:11 -04003009 struct super_block *sb;
Josef Bacik6a912212010-11-20 09:48:00 +00003010 struct dentry *old_parent = NULL;
Chris Mason12fcfd22009-03-24 10:24:20 -04003011 int ret = 0;
3012 u64 last_committed = root->fs_info->last_trans_committed;
3013
3014 sb = inode->i_sb;
3015
Sage Weil3a5e1402009-04-02 16:49:40 -04003016 if (btrfs_test_opt(root, NOTREELOG)) {
3017 ret = 1;
3018 goto end_no_trans;
3019 }
3020
Chris Mason12fcfd22009-03-24 10:24:20 -04003021 if (root->fs_info->last_trans_log_full_commit >
3022 root->fs_info->last_trans_committed) {
3023 ret = 1;
3024 goto end_no_trans;
3025 }
3026
Yan, Zheng76dda932009-09-21 16:00:26 -04003027 if (root != BTRFS_I(inode)->root ||
3028 btrfs_root_refs(&root->root_item) == 0) {
3029 ret = 1;
3030 goto end_no_trans;
3031 }
3032
Chris Mason12fcfd22009-03-24 10:24:20 -04003033 ret = check_parent_dirs_for_sync(trans, inode, parent,
3034 sb, last_committed);
3035 if (ret)
3036 goto end_no_trans;
Chris Masone02119d2008-09-05 16:13:11 -04003037
Chris Mason257c62e2009-10-13 13:21:08 -04003038 if (inode_in_log(trans, inode)) {
3039 ret = BTRFS_NO_LOG_SYNC;
3040 goto end_no_trans;
3041 }
3042
Yan, Zheng4a500fd2010-05-16 10:49:59 -04003043 ret = start_log_trans(trans, root);
3044 if (ret)
3045 goto end_trans;
Chris Mason12fcfd22009-03-24 10:24:20 -04003046
3047 ret = btrfs_log_inode(trans, root, inode, inode_only);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04003048 if (ret)
3049 goto end_trans;
Chris Mason12fcfd22009-03-24 10:24:20 -04003050
Chris Masonaf4176b2009-03-24 10:24:31 -04003051 /*
3052 * for regular files, if its inode is already on disk, we don't
3053 * have to worry about the parents at all. This is because
3054 * we can use the last_unlink_trans field to record renames
3055 * and other fun in this file.
3056 */
3057 if (S_ISREG(inode->i_mode) &&
3058 BTRFS_I(inode)->generation <= last_committed &&
Yan, Zheng4a500fd2010-05-16 10:49:59 -04003059 BTRFS_I(inode)->last_unlink_trans <= last_committed) {
3060 ret = 0;
3061 goto end_trans;
3062 }
Chris Masonaf4176b2009-03-24 10:24:31 -04003063
3064 inode_only = LOG_INODE_EXISTS;
Chris Masond3977122009-01-05 21:25:51 -05003065 while (1) {
Chris Mason12fcfd22009-03-24 10:24:20 -04003066 if (!parent || !parent->d_inode || sb != parent->d_inode->i_sb)
Chris Masone02119d2008-09-05 16:13:11 -04003067 break;
3068
Chris Mason12fcfd22009-03-24 10:24:20 -04003069 inode = parent->d_inode;
Yan, Zheng76dda932009-09-21 16:00:26 -04003070 if (root != BTRFS_I(inode)->root)
3071 break;
3072
Chris Mason12fcfd22009-03-24 10:24:20 -04003073 if (BTRFS_I(inode)->generation >
3074 root->fs_info->last_trans_committed) {
3075 ret = btrfs_log_inode(trans, root, inode, inode_only);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04003076 if (ret)
3077 goto end_trans;
Chris Mason12fcfd22009-03-24 10:24:20 -04003078 }
Yan, Zheng76dda932009-09-21 16:00:26 -04003079 if (IS_ROOT(parent))
Chris Masone02119d2008-09-05 16:13:11 -04003080 break;
Chris Mason12fcfd22009-03-24 10:24:20 -04003081
Josef Bacik6a912212010-11-20 09:48:00 +00003082 parent = dget_parent(parent);
3083 dput(old_parent);
3084 old_parent = parent;
Chris Masone02119d2008-09-05 16:13:11 -04003085 }
Chris Mason12fcfd22009-03-24 10:24:20 -04003086 ret = 0;
Yan, Zheng4a500fd2010-05-16 10:49:59 -04003087end_trans:
Josef Bacik6a912212010-11-20 09:48:00 +00003088 dput(old_parent);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04003089 if (ret < 0) {
3090 BUG_ON(ret != -ENOSPC);
3091 root->fs_info->last_trans_log_full_commit = trans->transid;
3092 ret = 1;
3093 }
Chris Mason12fcfd22009-03-24 10:24:20 -04003094 btrfs_end_log_trans(root);
3095end_no_trans:
3096 return ret;
Chris Masone02119d2008-09-05 16:13:11 -04003097}
3098
3099/*
3100 * it is not safe to log dentry if the chunk root has added new
3101 * chunks. This returns 0 if the dentry was logged, and 1 otherwise.
3102 * If this returns 1, you must commit the transaction to safely get your
3103 * data on disk.
3104 */
3105int btrfs_log_dentry_safe(struct btrfs_trans_handle *trans,
3106 struct btrfs_root *root, struct dentry *dentry)
3107{
Josef Bacik6a912212010-11-20 09:48:00 +00003108 struct dentry *parent = dget_parent(dentry);
3109 int ret;
3110
3111 ret = btrfs_log_inode_parent(trans, root, dentry->d_inode, parent, 0);
3112 dput(parent);
3113
3114 return ret;
Chris Masone02119d2008-09-05 16:13:11 -04003115}
3116
3117/*
3118 * should be called during mount to recover any replay any log trees
3119 * from the FS
3120 */
3121int btrfs_recover_log_trees(struct btrfs_root *log_root_tree)
3122{
3123 int ret;
3124 struct btrfs_path *path;
3125 struct btrfs_trans_handle *trans;
3126 struct btrfs_key key;
3127 struct btrfs_key found_key;
3128 struct btrfs_key tmp_key;
3129 struct btrfs_root *log;
3130 struct btrfs_fs_info *fs_info = log_root_tree->fs_info;
3131 struct walk_control wc = {
3132 .process_func = process_one_buffer,
3133 .stage = 0,
3134 };
3135
Chris Masone02119d2008-09-05 16:13:11 -04003136 path = btrfs_alloc_path();
Tsutomu Itohdb5b4932011-03-23 08:14:16 +00003137 if (!path)
3138 return -ENOMEM;
3139
3140 fs_info->log_root_recovering = 1;
Chris Masone02119d2008-09-05 16:13:11 -04003141
Yan, Zheng4a500fd2010-05-16 10:49:59 -04003142 trans = btrfs_start_transaction(fs_info->tree_root, 0);
Tsutomu Itoh98d5dc12011-01-20 06:19:37 +00003143 BUG_ON(IS_ERR(trans));
Chris Masone02119d2008-09-05 16:13:11 -04003144
3145 wc.trans = trans;
3146 wc.pin = 1;
3147
Tsutomu Itohdb5b4932011-03-23 08:14:16 +00003148 ret = walk_log_tree(trans, log_root_tree, &wc);
3149 BUG_ON(ret);
Chris Masone02119d2008-09-05 16:13:11 -04003150
3151again:
3152 key.objectid = BTRFS_TREE_LOG_OBJECTID;
3153 key.offset = (u64)-1;
3154 btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
3155
Chris Masond3977122009-01-05 21:25:51 -05003156 while (1) {
Chris Masone02119d2008-09-05 16:13:11 -04003157 ret = btrfs_search_slot(NULL, log_root_tree, &key, path, 0, 0);
3158 if (ret < 0)
3159 break;
3160 if (ret > 0) {
3161 if (path->slots[0] == 0)
3162 break;
3163 path->slots[0]--;
3164 }
3165 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
3166 path->slots[0]);
David Sterbab3b4aa72011-04-21 01:20:15 +02003167 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04003168 if (found_key.objectid != BTRFS_TREE_LOG_OBJECTID)
3169 break;
3170
3171 log = btrfs_read_fs_root_no_radix(log_root_tree,
3172 &found_key);
Tsutomu Itohdb5b4932011-03-23 08:14:16 +00003173 BUG_ON(IS_ERR(log));
Chris Masone02119d2008-09-05 16:13:11 -04003174
3175 tmp_key.objectid = found_key.offset;
3176 tmp_key.type = BTRFS_ROOT_ITEM_KEY;
3177 tmp_key.offset = (u64)-1;
3178
3179 wc.replay_dest = btrfs_read_fs_root_no_name(fs_info, &tmp_key);
Chris Masone02119d2008-09-05 16:13:11 -04003180 BUG_ON(!wc.replay_dest);
3181
Yan Zheng07d400a2009-01-06 11:42:00 -05003182 wc.replay_dest->log_root = log;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003183 btrfs_record_root_in_trans(trans, wc.replay_dest);
Chris Masone02119d2008-09-05 16:13:11 -04003184 ret = walk_log_tree(trans, log, &wc);
3185 BUG_ON(ret);
3186
3187 if (wc.stage == LOG_WALK_REPLAY_ALL) {
3188 ret = fixup_inode_link_counts(trans, wc.replay_dest,
3189 path);
3190 BUG_ON(ret);
3191 }
Chris Masone02119d2008-09-05 16:13:11 -04003192
3193 key.offset = found_key.offset - 1;
Yan Zheng07d400a2009-01-06 11:42:00 -05003194 wc.replay_dest->log_root = NULL;
Chris Masone02119d2008-09-05 16:13:11 -04003195 free_extent_buffer(log->node);
Chris Masonb263c2c2009-06-11 11:24:47 -04003196 free_extent_buffer(log->commit_root);
Chris Masone02119d2008-09-05 16:13:11 -04003197 kfree(log);
3198
3199 if (found_key.offset == 0)
3200 break;
3201 }
David Sterbab3b4aa72011-04-21 01:20:15 +02003202 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04003203
3204 /* step one is to pin it all, step two is to replay just inodes */
3205 if (wc.pin) {
3206 wc.pin = 0;
3207 wc.process_func = replay_one_buffer;
3208 wc.stage = LOG_WALK_REPLAY_INODES;
3209 goto again;
3210 }
3211 /* step three is to replay everything */
3212 if (wc.stage < LOG_WALK_REPLAY_ALL) {
3213 wc.stage++;
3214 goto again;
3215 }
3216
3217 btrfs_free_path(path);
3218
3219 free_extent_buffer(log_root_tree->node);
3220 log_root_tree->log_root = NULL;
3221 fs_info->log_root_recovering = 0;
3222
3223 /* step 4: commit the transaction, which also unpins the blocks */
3224 btrfs_commit_transaction(trans, fs_info->tree_root);
3225
3226 kfree(log_root_tree);
3227 return 0;
3228}
Chris Mason12fcfd22009-03-24 10:24:20 -04003229
3230/*
3231 * there are some corner cases where we want to force a full
3232 * commit instead of allowing a directory to be logged.
3233 *
3234 * They revolve around files there were unlinked from the directory, and
3235 * this function updates the parent directory so that a full commit is
3236 * properly done if it is fsync'd later after the unlinks are done.
3237 */
3238void btrfs_record_unlink_dir(struct btrfs_trans_handle *trans,
3239 struct inode *dir, struct inode *inode,
3240 int for_rename)
3241{
3242 /*
Chris Masonaf4176b2009-03-24 10:24:31 -04003243 * when we're logging a file, if it hasn't been renamed
3244 * or unlinked, and its inode is fully committed on disk,
3245 * we don't have to worry about walking up the directory chain
3246 * to log its parents.
3247 *
3248 * So, we use the last_unlink_trans field to put this transid
3249 * into the file. When the file is logged we check it and
3250 * don't log the parents if the file is fully on disk.
3251 */
3252 if (S_ISREG(inode->i_mode))
3253 BTRFS_I(inode)->last_unlink_trans = trans->transid;
3254
3255 /*
Chris Mason12fcfd22009-03-24 10:24:20 -04003256 * if this directory was already logged any new
3257 * names for this file/dir will get recorded
3258 */
3259 smp_mb();
3260 if (BTRFS_I(dir)->logged_trans == trans->transid)
3261 return;
3262
3263 /*
3264 * if the inode we're about to unlink was logged,
3265 * the log will be properly updated for any new names
3266 */
3267 if (BTRFS_I(inode)->logged_trans == trans->transid)
3268 return;
3269
3270 /*
3271 * when renaming files across directories, if the directory
3272 * there we're unlinking from gets fsync'd later on, there's
3273 * no way to find the destination directory later and fsync it
3274 * properly. So, we have to be conservative and force commits
3275 * so the new name gets discovered.
3276 */
3277 if (for_rename)
3278 goto record;
3279
3280 /* we can safely do the unlink without any special recording */
3281 return;
3282
3283record:
3284 BTRFS_I(dir)->last_unlink_trans = trans->transid;
3285}
3286
3287/*
3288 * Call this after adding a new name for a file and it will properly
3289 * update the log to reflect the new name.
3290 *
3291 * It will return zero if all goes well, and it will return 1 if a
3292 * full transaction commit is required.
3293 */
3294int btrfs_log_new_name(struct btrfs_trans_handle *trans,
3295 struct inode *inode, struct inode *old_dir,
3296 struct dentry *parent)
3297{
3298 struct btrfs_root * root = BTRFS_I(inode)->root;
3299
3300 /*
Chris Masonaf4176b2009-03-24 10:24:31 -04003301 * this will force the logging code to walk the dentry chain
3302 * up for the file
3303 */
3304 if (S_ISREG(inode->i_mode))
3305 BTRFS_I(inode)->last_unlink_trans = trans->transid;
3306
3307 /*
Chris Mason12fcfd22009-03-24 10:24:20 -04003308 * if this inode hasn't been logged and directory we're renaming it
3309 * from hasn't been logged, we don't need to log it
3310 */
3311 if (BTRFS_I(inode)->logged_trans <=
3312 root->fs_info->last_trans_committed &&
3313 (!old_dir || BTRFS_I(old_dir)->logged_trans <=
3314 root->fs_info->last_trans_committed))
3315 return 0;
3316
3317 return btrfs_log_inode_parent(trans, root, inode, parent, 1);
3318}
3319