blob: 4a9434b622ecfc0f571c32ea3abf54df3d878123 [file] [log] [blame]
Chris Masone02119d2008-09-05 16:13:11 -04001/*
2 * Copyright (C) 2008 Oracle. All rights reserved.
3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
12 *
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
17 */
18
19#include <linux/sched.h>
20#include "ctree.h"
21#include "transaction.h"
22#include "disk-io.h"
23#include "locking.h"
24#include "print-tree.h"
25#include "compat.h"
Christoph Hellwigb2950862008-12-02 09:54:17 -050026#include "tree-log.h"
Chris Masone02119d2008-09-05 16:13:11 -040027
28/* magic values for the inode_only field in btrfs_log_inode:
29 *
30 * LOG_INODE_ALL means to log everything
31 * LOG_INODE_EXISTS means to log just enough to recreate the inode
32 * during log replay
33 */
34#define LOG_INODE_ALL 0
35#define LOG_INODE_EXISTS 1
36
37/*
Chris Mason12fcfd22009-03-24 10:24:20 -040038 * directory trouble cases
39 *
40 * 1) on rename or unlink, if the inode being unlinked isn't in the fsync
41 * log, we must force a full commit before doing an fsync of the directory
42 * where the unlink was done.
43 * ---> record transid of last unlink/rename per directory
44 *
45 * mkdir foo/some_dir
46 * normal commit
47 * rename foo/some_dir foo2/some_dir
48 * mkdir foo/some_dir
49 * fsync foo/some_dir/some_file
50 *
51 * The fsync above will unlink the original some_dir without recording
52 * it in its new location (foo2). After a crash, some_dir will be gone
53 * unless the fsync of some_file forces a full commit
54 *
55 * 2) we must log any new names for any file or dir that is in the fsync
56 * log. ---> check inode while renaming/linking.
57 *
58 * 2a) we must log any new names for any file or dir during rename
59 * when the directory they are being removed from was logged.
60 * ---> check inode and old parent dir during rename
61 *
62 * 2a is actually the more important variant. With the extra logging
63 * a crash might unlink the old name without recreating the new one
64 *
65 * 3) after a crash, we must go through any directories with a link count
66 * of zero and redo the rm -rf
67 *
68 * mkdir f1/foo
69 * normal commit
70 * rm -rf f1/foo
71 * fsync(f1)
72 *
73 * The directory f1 was fully removed from the FS, but fsync was never
74 * called on f1, only its parent dir. After a crash the rm -rf must
75 * be replayed. This must be able to recurse down the entire
76 * directory tree. The inode link count fixup code takes care of the
77 * ugly details.
78 */
79
80/*
Chris Masone02119d2008-09-05 16:13:11 -040081 * stages for the tree walking. The first
82 * stage (0) is to only pin down the blocks we find
83 * the second stage (1) is to make sure that all the inodes
84 * we find in the log are created in the subvolume.
85 *
86 * The last stage is to deal with directories and links and extents
87 * and all the other fun semantics
88 */
89#define LOG_WALK_PIN_ONLY 0
90#define LOG_WALK_REPLAY_INODES 1
91#define LOG_WALK_REPLAY_ALL 2
92
Chris Mason12fcfd22009-03-24 10:24:20 -040093static int btrfs_log_inode(struct btrfs_trans_handle *trans,
Chris Masone02119d2008-09-05 16:13:11 -040094 struct btrfs_root *root, struct inode *inode,
95 int inode_only);
Yan Zhengec051c02009-01-05 15:43:42 -050096static int link_to_fixup_dir(struct btrfs_trans_handle *trans,
97 struct btrfs_root *root,
98 struct btrfs_path *path, u64 objectid);
Chris Mason12fcfd22009-03-24 10:24:20 -040099static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans,
100 struct btrfs_root *root,
101 struct btrfs_root *log,
102 struct btrfs_path *path,
103 u64 dirid, int del_all);
Chris Masone02119d2008-09-05 16:13:11 -0400104
105/*
106 * tree logging is a special write ahead log used to make sure that
107 * fsyncs and O_SYNCs can happen without doing full tree commits.
108 *
109 * Full tree commits are expensive because they require commonly
110 * modified blocks to be recowed, creating many dirty pages in the
111 * extent tree an 4x-6x higher write load than ext3.
112 *
113 * Instead of doing a tree commit on every fsync, we use the
114 * key ranges and transaction ids to find items for a given file or directory
115 * that have changed in this transaction. Those items are copied into
116 * a special tree (one per subvolume root), that tree is written to disk
117 * and then the fsync is considered complete.
118 *
119 * After a crash, items are copied out of the log-tree back into the
120 * subvolume tree. Any file data extents found are recorded in the extent
121 * allocation tree, and the log-tree freed.
122 *
123 * The log tree is read three times, once to pin down all the extents it is
124 * using in ram and once, once to create all the inodes logged in the tree
125 * and once to do all the other items.
126 */
127
128/*
Chris Masone02119d2008-09-05 16:13:11 -0400129 * start a sub transaction and setup the log tree
130 * this increments the log tree writer count to make the people
131 * syncing the tree wait for us to finish
132 */
133static int start_log_trans(struct btrfs_trans_handle *trans,
134 struct btrfs_root *root)
135{
136 int ret;
Yan Zheng7237f182009-01-21 12:54:03 -0500137
138 mutex_lock(&root->log_mutex);
139 if (root->log_root) {
Josef Bacikff782e02009-10-08 15:30:04 -0400140 if (!root->log_start_pid) {
141 root->log_start_pid = current->pid;
142 root->log_multiple_pids = false;
143 } else if (root->log_start_pid != current->pid) {
144 root->log_multiple_pids = true;
145 }
146
Yan Zheng7237f182009-01-21 12:54:03 -0500147 root->log_batch++;
148 atomic_inc(&root->log_writers);
149 mutex_unlock(&root->log_mutex);
150 return 0;
151 }
Josef Bacikff782e02009-10-08 15:30:04 -0400152 root->log_multiple_pids = false;
153 root->log_start_pid = current->pid;
Chris Masone02119d2008-09-05 16:13:11 -0400154 mutex_lock(&root->fs_info->tree_log_mutex);
155 if (!root->fs_info->log_root_tree) {
156 ret = btrfs_init_log_root_tree(trans, root->fs_info);
157 BUG_ON(ret);
158 }
159 if (!root->log_root) {
160 ret = btrfs_add_log_tree(trans, root);
161 BUG_ON(ret);
162 }
Chris Masone02119d2008-09-05 16:13:11 -0400163 mutex_unlock(&root->fs_info->tree_log_mutex);
Yan Zheng7237f182009-01-21 12:54:03 -0500164 root->log_batch++;
165 atomic_inc(&root->log_writers);
166 mutex_unlock(&root->log_mutex);
Chris Masone02119d2008-09-05 16:13:11 -0400167 return 0;
168}
169
170/*
171 * returns 0 if there was a log transaction running and we were able
172 * to join, or returns -ENOENT if there were not transactions
173 * in progress
174 */
175static int join_running_log_trans(struct btrfs_root *root)
176{
177 int ret = -ENOENT;
178
179 smp_mb();
180 if (!root->log_root)
181 return -ENOENT;
182
Yan Zheng7237f182009-01-21 12:54:03 -0500183 mutex_lock(&root->log_mutex);
Chris Masone02119d2008-09-05 16:13:11 -0400184 if (root->log_root) {
185 ret = 0;
Yan Zheng7237f182009-01-21 12:54:03 -0500186 atomic_inc(&root->log_writers);
Chris Masone02119d2008-09-05 16:13:11 -0400187 }
Yan Zheng7237f182009-01-21 12:54:03 -0500188 mutex_unlock(&root->log_mutex);
Chris Masone02119d2008-09-05 16:13:11 -0400189 return ret;
190}
191
192/*
Chris Mason12fcfd22009-03-24 10:24:20 -0400193 * This either makes the current running log transaction wait
194 * until you call btrfs_end_log_trans() or it makes any future
195 * log transactions wait until you call btrfs_end_log_trans()
196 */
197int btrfs_pin_log_trans(struct btrfs_root *root)
198{
199 int ret = -ENOENT;
200
201 mutex_lock(&root->log_mutex);
202 atomic_inc(&root->log_writers);
203 mutex_unlock(&root->log_mutex);
204 return ret;
205}
206
207/*
Chris Masone02119d2008-09-05 16:13:11 -0400208 * indicate we're done making changes to the log tree
209 * and wake up anyone waiting to do a sync
210 */
Chris Mason12fcfd22009-03-24 10:24:20 -0400211int btrfs_end_log_trans(struct btrfs_root *root)
Chris Masone02119d2008-09-05 16:13:11 -0400212{
Yan Zheng7237f182009-01-21 12:54:03 -0500213 if (atomic_dec_and_test(&root->log_writers)) {
214 smp_mb();
215 if (waitqueue_active(&root->log_writer_wait))
216 wake_up(&root->log_writer_wait);
217 }
Chris Masone02119d2008-09-05 16:13:11 -0400218 return 0;
219}
220
221
222/*
223 * the walk control struct is used to pass state down the chain when
224 * processing the log tree. The stage field tells us which part
225 * of the log tree processing we are currently doing. The others
226 * are state fields used for that specific part
227 */
228struct walk_control {
229 /* should we free the extent on disk when done? This is used
230 * at transaction commit time while freeing a log tree
231 */
232 int free;
233
234 /* should we write out the extent buffer? This is used
235 * while flushing the log tree to disk during a sync
236 */
237 int write;
238
239 /* should we wait for the extent buffer io to finish? Also used
240 * while flushing the log tree to disk for a sync
241 */
242 int wait;
243
244 /* pin only walk, we record which extents on disk belong to the
245 * log trees
246 */
247 int pin;
248
249 /* what stage of the replay code we're currently in */
250 int stage;
251
252 /* the root we are currently replaying */
253 struct btrfs_root *replay_dest;
254
255 /* the trans handle for the current replay */
256 struct btrfs_trans_handle *trans;
257
258 /* the function that gets used to process blocks we find in the
259 * tree. Note the extent_buffer might not be up to date when it is
260 * passed in, and it must be checked or read if you need the data
261 * inside it
262 */
263 int (*process_func)(struct btrfs_root *log, struct extent_buffer *eb,
264 struct walk_control *wc, u64 gen);
265};
266
267/*
268 * process_func used to pin down extents, write them or wait on them
269 */
270static int process_one_buffer(struct btrfs_root *log,
271 struct extent_buffer *eb,
272 struct walk_control *wc, u64 gen)
273{
Josef Bacik04018de2009-04-03 10:14:18 -0400274 if (wc->pin)
Yan Zheng11833d62009-09-11 16:11:19 -0400275 btrfs_pin_extent(log->fs_info->extent_root,
276 eb->start, eb->len, 0);
Chris Masone02119d2008-09-05 16:13:11 -0400277
278 if (btrfs_buffer_uptodate(eb, gen)) {
279 if (wc->write)
280 btrfs_write_tree_block(eb);
281 if (wc->wait)
282 btrfs_wait_tree_block_writeback(eb);
283 }
284 return 0;
285}
286
287/*
288 * Item overwrite used by replay and tree logging. eb, slot and key all refer
289 * to the src data we are copying out.
290 *
291 * root is the tree we are copying into, and path is a scratch
292 * path for use in this function (it should be released on entry and
293 * will be released on exit).
294 *
295 * If the key is already in the destination tree the existing item is
296 * overwritten. If the existing item isn't big enough, it is extended.
297 * If it is too large, it is truncated.
298 *
299 * If the key isn't in the destination yet, a new item is inserted.
300 */
301static noinline int overwrite_item(struct btrfs_trans_handle *trans,
302 struct btrfs_root *root,
303 struct btrfs_path *path,
304 struct extent_buffer *eb, int slot,
305 struct btrfs_key *key)
306{
307 int ret;
308 u32 item_size;
309 u64 saved_i_size = 0;
310 int save_old_i_size = 0;
311 unsigned long src_ptr;
312 unsigned long dst_ptr;
313 int overwrite_root = 0;
314
315 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
316 overwrite_root = 1;
317
318 item_size = btrfs_item_size_nr(eb, slot);
319 src_ptr = btrfs_item_ptr_offset(eb, slot);
320
321 /* look for the key in the destination tree */
322 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
323 if (ret == 0) {
324 char *src_copy;
325 char *dst_copy;
326 u32 dst_size = btrfs_item_size_nr(path->nodes[0],
327 path->slots[0]);
328 if (dst_size != item_size)
329 goto insert;
330
331 if (item_size == 0) {
332 btrfs_release_path(root, path);
333 return 0;
334 }
335 dst_copy = kmalloc(item_size, GFP_NOFS);
336 src_copy = kmalloc(item_size, GFP_NOFS);
337
338 read_extent_buffer(eb, src_copy, src_ptr, item_size);
339
340 dst_ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
341 read_extent_buffer(path->nodes[0], dst_copy, dst_ptr,
342 item_size);
343 ret = memcmp(dst_copy, src_copy, item_size);
344
345 kfree(dst_copy);
346 kfree(src_copy);
347 /*
348 * they have the same contents, just return, this saves
349 * us from cowing blocks in the destination tree and doing
350 * extra writes that may not have been done by a previous
351 * sync
352 */
353 if (ret == 0) {
354 btrfs_release_path(root, path);
355 return 0;
356 }
357
358 }
359insert:
360 btrfs_release_path(root, path);
361 /* try to insert the key into the destination tree */
362 ret = btrfs_insert_empty_item(trans, root, path,
363 key, item_size);
364
365 /* make sure any existing item is the correct size */
366 if (ret == -EEXIST) {
367 u32 found_size;
368 found_size = btrfs_item_size_nr(path->nodes[0],
369 path->slots[0]);
370 if (found_size > item_size) {
371 btrfs_truncate_item(trans, root, path, item_size, 1);
372 } else if (found_size < item_size) {
Yan Zheng87b29b22008-12-17 10:21:48 -0500373 ret = btrfs_extend_item(trans, root, path,
374 item_size - found_size);
Chris Masone02119d2008-09-05 16:13:11 -0400375 BUG_ON(ret);
376 }
377 } else if (ret) {
378 BUG();
379 }
380 dst_ptr = btrfs_item_ptr_offset(path->nodes[0],
381 path->slots[0]);
382
383 /* don't overwrite an existing inode if the generation number
384 * was logged as zero. This is done when the tree logging code
385 * is just logging an inode to make sure it exists after recovery.
386 *
387 * Also, don't overwrite i_size on directories during replay.
388 * log replay inserts and removes directory items based on the
389 * state of the tree found in the subvolume, and i_size is modified
390 * as it goes
391 */
392 if (key->type == BTRFS_INODE_ITEM_KEY && ret == -EEXIST) {
393 struct btrfs_inode_item *src_item;
394 struct btrfs_inode_item *dst_item;
395
396 src_item = (struct btrfs_inode_item *)src_ptr;
397 dst_item = (struct btrfs_inode_item *)dst_ptr;
398
399 if (btrfs_inode_generation(eb, src_item) == 0)
400 goto no_copy;
401
402 if (overwrite_root &&
403 S_ISDIR(btrfs_inode_mode(eb, src_item)) &&
404 S_ISDIR(btrfs_inode_mode(path->nodes[0], dst_item))) {
405 save_old_i_size = 1;
406 saved_i_size = btrfs_inode_size(path->nodes[0],
407 dst_item);
408 }
409 }
410
411 copy_extent_buffer(path->nodes[0], eb, dst_ptr,
412 src_ptr, item_size);
413
414 if (save_old_i_size) {
415 struct btrfs_inode_item *dst_item;
416 dst_item = (struct btrfs_inode_item *)dst_ptr;
417 btrfs_set_inode_size(path->nodes[0], dst_item, saved_i_size);
418 }
419
420 /* make sure the generation is filled in */
421 if (key->type == BTRFS_INODE_ITEM_KEY) {
422 struct btrfs_inode_item *dst_item;
423 dst_item = (struct btrfs_inode_item *)dst_ptr;
424 if (btrfs_inode_generation(path->nodes[0], dst_item) == 0) {
425 btrfs_set_inode_generation(path->nodes[0], dst_item,
426 trans->transid);
427 }
428 }
429no_copy:
430 btrfs_mark_buffer_dirty(path->nodes[0]);
431 btrfs_release_path(root, path);
432 return 0;
433}
434
435/*
436 * simple helper to read an inode off the disk from a given root
437 * This can only be called for subvolume roots and not for the log
438 */
439static noinline struct inode *read_one_inode(struct btrfs_root *root,
440 u64 objectid)
441{
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400442 struct btrfs_key key;
Chris Masone02119d2008-09-05 16:13:11 -0400443 struct inode *inode;
Chris Masone02119d2008-09-05 16:13:11 -0400444
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400445 key.objectid = objectid;
446 key.type = BTRFS_INODE_ITEM_KEY;
447 key.offset = 0;
448 inode = btrfs_iget(root->fs_info->sb, &key, root);
449 if (IS_ERR(inode)) {
450 inode = NULL;
451 } else if (is_bad_inode(inode)) {
Chris Masone02119d2008-09-05 16:13:11 -0400452 iput(inode);
453 inode = NULL;
454 }
455 return inode;
456}
457
458/* replays a single extent in 'eb' at 'slot' with 'key' into the
459 * subvolume 'root'. path is released on entry and should be released
460 * on exit.
461 *
462 * extents in the log tree have not been allocated out of the extent
463 * tree yet. So, this completes the allocation, taking a reference
464 * as required if the extent already exists or creating a new extent
465 * if it isn't in the extent allocation tree yet.
466 *
467 * The extent is inserted into the file, dropping any existing extents
468 * from the file that overlap the new one.
469 */
470static noinline int replay_one_extent(struct btrfs_trans_handle *trans,
471 struct btrfs_root *root,
472 struct btrfs_path *path,
473 struct extent_buffer *eb, int slot,
474 struct btrfs_key *key)
475{
476 int found_type;
477 u64 mask = root->sectorsize - 1;
478 u64 extent_end;
479 u64 alloc_hint;
480 u64 start = key->offset;
Yan Zheng07d400a2009-01-06 11:42:00 -0500481 u64 saved_nbytes;
Chris Masone02119d2008-09-05 16:13:11 -0400482 struct btrfs_file_extent_item *item;
483 struct inode *inode = NULL;
484 unsigned long size;
485 int ret = 0;
486
487 item = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
488 found_type = btrfs_file_extent_type(eb, item);
489
Yan Zhengd899e052008-10-30 14:25:28 -0400490 if (found_type == BTRFS_FILE_EXTENT_REG ||
491 found_type == BTRFS_FILE_EXTENT_PREALLOC)
Chris Masone02119d2008-09-05 16:13:11 -0400492 extent_end = start + btrfs_file_extent_num_bytes(eb, item);
493 else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
Chris Masonc8b97812008-10-29 14:49:59 -0400494 size = btrfs_file_extent_inline_len(eb, item);
Chris Masone02119d2008-09-05 16:13:11 -0400495 extent_end = (start + size + mask) & ~mask;
496 } else {
497 ret = 0;
498 goto out;
499 }
500
501 inode = read_one_inode(root, key->objectid);
502 if (!inode) {
503 ret = -EIO;
504 goto out;
505 }
506
507 /*
508 * first check to see if we already have this extent in the
509 * file. This must be done before the btrfs_drop_extents run
510 * so we don't try to drop this extent.
511 */
512 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
513 start, 0);
514
Yan Zhengd899e052008-10-30 14:25:28 -0400515 if (ret == 0 &&
516 (found_type == BTRFS_FILE_EXTENT_REG ||
517 found_type == BTRFS_FILE_EXTENT_PREALLOC)) {
Chris Masone02119d2008-09-05 16:13:11 -0400518 struct btrfs_file_extent_item cmp1;
519 struct btrfs_file_extent_item cmp2;
520 struct btrfs_file_extent_item *existing;
521 struct extent_buffer *leaf;
522
523 leaf = path->nodes[0];
524 existing = btrfs_item_ptr(leaf, path->slots[0],
525 struct btrfs_file_extent_item);
526
527 read_extent_buffer(eb, &cmp1, (unsigned long)item,
528 sizeof(cmp1));
529 read_extent_buffer(leaf, &cmp2, (unsigned long)existing,
530 sizeof(cmp2));
531
532 /*
533 * we already have a pointer to this exact extent,
534 * we don't have to do anything
535 */
536 if (memcmp(&cmp1, &cmp2, sizeof(cmp1)) == 0) {
537 btrfs_release_path(root, path);
538 goto out;
539 }
540 }
541 btrfs_release_path(root, path);
542
Yan Zheng07d400a2009-01-06 11:42:00 -0500543 saved_nbytes = inode_get_bytes(inode);
Chris Masone02119d2008-09-05 16:13:11 -0400544 /* drop any overlapping extents */
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000545 ret = btrfs_drop_extents(trans, inode, start, extent_end,
546 &alloc_hint, 1);
Chris Masone02119d2008-09-05 16:13:11 -0400547 BUG_ON(ret);
548
Yan Zheng07d400a2009-01-06 11:42:00 -0500549 if (found_type == BTRFS_FILE_EXTENT_REG ||
550 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400551 u64 offset;
Yan Zheng07d400a2009-01-06 11:42:00 -0500552 unsigned long dest_offset;
553 struct btrfs_key ins;
Chris Masone02119d2008-09-05 16:13:11 -0400554
Yan Zheng07d400a2009-01-06 11:42:00 -0500555 ret = btrfs_insert_empty_item(trans, root, path, key,
556 sizeof(*item));
557 BUG_ON(ret);
558 dest_offset = btrfs_item_ptr_offset(path->nodes[0],
559 path->slots[0]);
560 copy_extent_buffer(path->nodes[0], eb, dest_offset,
561 (unsigned long)item, sizeof(*item));
562
563 ins.objectid = btrfs_file_extent_disk_bytenr(eb, item);
564 ins.offset = btrfs_file_extent_disk_num_bytes(eb, item);
565 ins.type = BTRFS_EXTENT_ITEM_KEY;
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400566 offset = key->offset - btrfs_file_extent_offset(eb, item);
Yan Zheng07d400a2009-01-06 11:42:00 -0500567
568 if (ins.objectid > 0) {
569 u64 csum_start;
570 u64 csum_end;
571 LIST_HEAD(ordered_sums);
572 /*
573 * is this extent already allocated in the extent
574 * allocation tree? If so, just add a reference
575 */
576 ret = btrfs_lookup_extent(root, ins.objectid,
577 ins.offset);
578 if (ret == 0) {
579 ret = btrfs_inc_extent_ref(trans, root,
580 ins.objectid, ins.offset,
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400581 0, root->root_key.objectid,
582 key->objectid, offset);
Yan Zheng07d400a2009-01-06 11:42:00 -0500583 } else {
584 /*
585 * insert the extent pointer in the extent
586 * allocation tree
587 */
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400588 ret = btrfs_alloc_logged_file_extent(trans,
589 root, root->root_key.objectid,
590 key->objectid, offset, &ins);
Yan Zheng07d400a2009-01-06 11:42:00 -0500591 BUG_ON(ret);
592 }
593 btrfs_release_path(root, path);
594
595 if (btrfs_file_extent_compression(eb, item)) {
596 csum_start = ins.objectid;
597 csum_end = csum_start + ins.offset;
598 } else {
599 csum_start = ins.objectid +
600 btrfs_file_extent_offset(eb, item);
601 csum_end = csum_start +
602 btrfs_file_extent_num_bytes(eb, item);
603 }
604
605 ret = btrfs_lookup_csums_range(root->log_root,
606 csum_start, csum_end - 1,
607 &ordered_sums);
608 BUG_ON(ret);
609 while (!list_empty(&ordered_sums)) {
610 struct btrfs_ordered_sum *sums;
611 sums = list_entry(ordered_sums.next,
612 struct btrfs_ordered_sum,
613 list);
614 ret = btrfs_csum_file_blocks(trans,
615 root->fs_info->csum_root,
616 sums);
617 BUG_ON(ret);
618 list_del(&sums->list);
619 kfree(sums);
620 }
621 } else {
622 btrfs_release_path(root, path);
623 }
624 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
625 /* inline extents are easy, we just overwrite them */
626 ret = overwrite_item(trans, root, path, eb, slot, key);
627 BUG_ON(ret);
628 }
629
630 inode_set_bytes(inode, saved_nbytes);
Chris Masone02119d2008-09-05 16:13:11 -0400631 btrfs_update_inode(trans, root, inode);
632out:
633 if (inode)
634 iput(inode);
635 return ret;
636}
637
638/*
639 * when cleaning up conflicts between the directory names in the
640 * subvolume, directory names in the log and directory names in the
641 * inode back references, we may have to unlink inodes from directories.
642 *
643 * This is a helper function to do the unlink of a specific directory
644 * item
645 */
646static noinline int drop_one_dir_item(struct btrfs_trans_handle *trans,
647 struct btrfs_root *root,
648 struct btrfs_path *path,
649 struct inode *dir,
650 struct btrfs_dir_item *di)
651{
652 struct inode *inode;
653 char *name;
654 int name_len;
655 struct extent_buffer *leaf;
656 struct btrfs_key location;
657 int ret;
658
659 leaf = path->nodes[0];
660
661 btrfs_dir_item_key_to_cpu(leaf, di, &location);
662 name_len = btrfs_dir_name_len(leaf, di);
663 name = kmalloc(name_len, GFP_NOFS);
664 read_extent_buffer(leaf, name, (unsigned long)(di + 1), name_len);
665 btrfs_release_path(root, path);
666
667 inode = read_one_inode(root, location.objectid);
668 BUG_ON(!inode);
669
Yan Zhengec051c02009-01-05 15:43:42 -0500670 ret = link_to_fixup_dir(trans, root, path, location.objectid);
671 BUG_ON(ret);
Chris Mason12fcfd22009-03-24 10:24:20 -0400672
Chris Masone02119d2008-09-05 16:13:11 -0400673 ret = btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
Yan Zhengec051c02009-01-05 15:43:42 -0500674 BUG_ON(ret);
Chris Masone02119d2008-09-05 16:13:11 -0400675 kfree(name);
676
677 iput(inode);
678 return ret;
679}
680
681/*
682 * helper function to see if a given name and sequence number found
683 * in an inode back reference are already in a directory and correctly
684 * point to this inode
685 */
686static noinline int inode_in_dir(struct btrfs_root *root,
687 struct btrfs_path *path,
688 u64 dirid, u64 objectid, u64 index,
689 const char *name, int name_len)
690{
691 struct btrfs_dir_item *di;
692 struct btrfs_key location;
693 int match = 0;
694
695 di = btrfs_lookup_dir_index_item(NULL, root, path, dirid,
696 index, name, name_len, 0);
697 if (di && !IS_ERR(di)) {
698 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
699 if (location.objectid != objectid)
700 goto out;
701 } else
702 goto out;
703 btrfs_release_path(root, path);
704
705 di = btrfs_lookup_dir_item(NULL, root, path, dirid, name, name_len, 0);
706 if (di && !IS_ERR(di)) {
707 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
708 if (location.objectid != objectid)
709 goto out;
710 } else
711 goto out;
712 match = 1;
713out:
714 btrfs_release_path(root, path);
715 return match;
716}
717
718/*
719 * helper function to check a log tree for a named back reference in
720 * an inode. This is used to decide if a back reference that is
721 * found in the subvolume conflicts with what we find in the log.
722 *
723 * inode backreferences may have multiple refs in a single item,
724 * during replay we process one reference at a time, and we don't
725 * want to delete valid links to a file from the subvolume if that
726 * link is also in the log.
727 */
728static noinline int backref_in_log(struct btrfs_root *log,
729 struct btrfs_key *key,
730 char *name, int namelen)
731{
732 struct btrfs_path *path;
733 struct btrfs_inode_ref *ref;
734 unsigned long ptr;
735 unsigned long ptr_end;
736 unsigned long name_ptr;
737 int found_name_len;
738 int item_size;
739 int ret;
740 int match = 0;
741
742 path = btrfs_alloc_path();
743 ret = btrfs_search_slot(NULL, log, key, path, 0, 0);
744 if (ret != 0)
745 goto out;
746
747 item_size = btrfs_item_size_nr(path->nodes[0], path->slots[0]);
748 ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
749 ptr_end = ptr + item_size;
750 while (ptr < ptr_end) {
751 ref = (struct btrfs_inode_ref *)ptr;
752 found_name_len = btrfs_inode_ref_name_len(path->nodes[0], ref);
753 if (found_name_len == namelen) {
754 name_ptr = (unsigned long)(ref + 1);
755 ret = memcmp_extent_buffer(path->nodes[0], name,
756 name_ptr, namelen);
757 if (ret == 0) {
758 match = 1;
759 goto out;
760 }
761 }
762 ptr = (unsigned long)(ref + 1) + found_name_len;
763 }
764out:
765 btrfs_free_path(path);
766 return match;
767}
768
769
770/*
771 * replay one inode back reference item found in the log tree.
772 * eb, slot and key refer to the buffer and key found in the log tree.
773 * root is the destination we are replaying into, and path is for temp
774 * use by this function. (it should be released on return).
775 */
776static noinline int add_inode_ref(struct btrfs_trans_handle *trans,
777 struct btrfs_root *root,
778 struct btrfs_root *log,
779 struct btrfs_path *path,
780 struct extent_buffer *eb, int slot,
781 struct btrfs_key *key)
782{
783 struct inode *dir;
784 int ret;
785 struct btrfs_key location;
786 struct btrfs_inode_ref *ref;
787 struct btrfs_dir_item *di;
788 struct inode *inode;
789 char *name;
790 int namelen;
791 unsigned long ref_ptr;
792 unsigned long ref_end;
793
794 location.objectid = key->objectid;
795 location.type = BTRFS_INODE_ITEM_KEY;
796 location.offset = 0;
797
798 /*
799 * it is possible that we didn't log all the parent directories
800 * for a given inode. If we don't find the dir, just don't
801 * copy the back ref in. The link count fixup code will take
802 * care of the rest
803 */
804 dir = read_one_inode(root, key->offset);
805 if (!dir)
806 return -ENOENT;
807
808 inode = read_one_inode(root, key->objectid);
Julia Lawall631c07c2009-07-27 13:57:00 -0400809 BUG_ON(!inode);
Chris Masone02119d2008-09-05 16:13:11 -0400810
811 ref_ptr = btrfs_item_ptr_offset(eb, slot);
812 ref_end = ref_ptr + btrfs_item_size_nr(eb, slot);
813
814again:
815 ref = (struct btrfs_inode_ref *)ref_ptr;
816
817 namelen = btrfs_inode_ref_name_len(eb, ref);
818 name = kmalloc(namelen, GFP_NOFS);
819 BUG_ON(!name);
820
821 read_extent_buffer(eb, name, (unsigned long)(ref + 1), namelen);
822
823 /* if we already have a perfect match, we're done */
824 if (inode_in_dir(root, path, dir->i_ino, inode->i_ino,
825 btrfs_inode_ref_index(eb, ref),
826 name, namelen)) {
827 goto out;
828 }
829
830 /*
831 * look for a conflicting back reference in the metadata.
832 * if we find one we have to unlink that name of the file
833 * before we add our new link. Later on, we overwrite any
834 * existing back reference, and we don't want to create
835 * dangling pointers in the directory.
836 */
837conflict_again:
838 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
839 if (ret == 0) {
840 char *victim_name;
841 int victim_name_len;
842 struct btrfs_inode_ref *victim_ref;
843 unsigned long ptr;
844 unsigned long ptr_end;
845 struct extent_buffer *leaf = path->nodes[0];
846
847 /* are we trying to overwrite a back ref for the root directory
848 * if so, just jump out, we're done
849 */
850 if (key->objectid == key->offset)
851 goto out_nowrite;
852
853 /* check all the names in this back reference to see
854 * if they are in the log. if so, we allow them to stay
855 * otherwise they must be unlinked as a conflict
856 */
857 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
858 ptr_end = ptr + btrfs_item_size_nr(leaf, path->slots[0]);
Chris Masond3977122009-01-05 21:25:51 -0500859 while (ptr < ptr_end) {
Chris Masone02119d2008-09-05 16:13:11 -0400860 victim_ref = (struct btrfs_inode_ref *)ptr;
861 victim_name_len = btrfs_inode_ref_name_len(leaf,
862 victim_ref);
863 victim_name = kmalloc(victim_name_len, GFP_NOFS);
864 BUG_ON(!victim_name);
865
866 read_extent_buffer(leaf, victim_name,
867 (unsigned long)(victim_ref + 1),
868 victim_name_len);
869
870 if (!backref_in_log(log, key, victim_name,
871 victim_name_len)) {
872 btrfs_inc_nlink(inode);
873 btrfs_release_path(root, path);
Chris Mason12fcfd22009-03-24 10:24:20 -0400874
Chris Masone02119d2008-09-05 16:13:11 -0400875 ret = btrfs_unlink_inode(trans, root, dir,
876 inode, victim_name,
877 victim_name_len);
878 kfree(victim_name);
879 btrfs_release_path(root, path);
880 goto conflict_again;
881 }
882 kfree(victim_name);
883 ptr = (unsigned long)(victim_ref + 1) + victim_name_len;
884 }
885 BUG_ON(ret);
886 }
887 btrfs_release_path(root, path);
888
889 /* look for a conflicting sequence number */
890 di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
891 btrfs_inode_ref_index(eb, ref),
892 name, namelen, 0);
893 if (di && !IS_ERR(di)) {
894 ret = drop_one_dir_item(trans, root, path, dir, di);
895 BUG_ON(ret);
896 }
897 btrfs_release_path(root, path);
898
899
900 /* look for a conflicting name */
901 di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
902 name, namelen, 0);
903 if (di && !IS_ERR(di)) {
904 ret = drop_one_dir_item(trans, root, path, dir, di);
905 BUG_ON(ret);
906 }
907 btrfs_release_path(root, path);
908
909 /* insert our name */
910 ret = btrfs_add_link(trans, dir, inode, name, namelen, 0,
911 btrfs_inode_ref_index(eb, ref));
912 BUG_ON(ret);
913
914 btrfs_update_inode(trans, root, inode);
915
916out:
917 ref_ptr = (unsigned long)(ref + 1) + namelen;
918 kfree(name);
919 if (ref_ptr < ref_end)
920 goto again;
921
922 /* finally write the back reference in the inode */
923 ret = overwrite_item(trans, root, path, eb, slot, key);
924 BUG_ON(ret);
925
926out_nowrite:
927 btrfs_release_path(root, path);
928 iput(dir);
929 iput(inode);
930 return 0;
931}
932
Yan, Zhengc71bf092009-11-12 09:34:40 +0000933static int insert_orphan_item(struct btrfs_trans_handle *trans,
934 struct btrfs_root *root, u64 offset)
935{
936 int ret;
937 ret = btrfs_find_orphan_item(root, offset);
938 if (ret > 0)
939 ret = btrfs_insert_orphan_item(trans, root, offset);
940 return ret;
941}
942
943
Chris Masone02119d2008-09-05 16:13:11 -0400944/*
Chris Masone02119d2008-09-05 16:13:11 -0400945 * There are a few corners where the link count of the file can't
946 * be properly maintained during replay. So, instead of adding
947 * lots of complexity to the log code, we just scan the backrefs
948 * for any file that has been through replay.
949 *
950 * The scan will update the link count on the inode to reflect the
951 * number of back refs found. If it goes down to zero, the iput
952 * will free the inode.
953 */
954static noinline int fixup_inode_link_count(struct btrfs_trans_handle *trans,
955 struct btrfs_root *root,
956 struct inode *inode)
957{
958 struct btrfs_path *path;
959 int ret;
960 struct btrfs_key key;
961 u64 nlink = 0;
962 unsigned long ptr;
963 unsigned long ptr_end;
964 int name_len;
965
966 key.objectid = inode->i_ino;
967 key.type = BTRFS_INODE_REF_KEY;
968 key.offset = (u64)-1;
969
970 path = btrfs_alloc_path();
971
Chris Masond3977122009-01-05 21:25:51 -0500972 while (1) {
Chris Masone02119d2008-09-05 16:13:11 -0400973 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
974 if (ret < 0)
975 break;
976 if (ret > 0) {
977 if (path->slots[0] == 0)
978 break;
979 path->slots[0]--;
980 }
981 btrfs_item_key_to_cpu(path->nodes[0], &key,
982 path->slots[0]);
983 if (key.objectid != inode->i_ino ||
984 key.type != BTRFS_INODE_REF_KEY)
985 break;
986 ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
987 ptr_end = ptr + btrfs_item_size_nr(path->nodes[0],
988 path->slots[0]);
Chris Masond3977122009-01-05 21:25:51 -0500989 while (ptr < ptr_end) {
Chris Masone02119d2008-09-05 16:13:11 -0400990 struct btrfs_inode_ref *ref;
991
992 ref = (struct btrfs_inode_ref *)ptr;
993 name_len = btrfs_inode_ref_name_len(path->nodes[0],
994 ref);
995 ptr = (unsigned long)(ref + 1) + name_len;
996 nlink++;
997 }
998
999 if (key.offset == 0)
1000 break;
1001 key.offset--;
1002 btrfs_release_path(root, path);
1003 }
Chris Mason12fcfd22009-03-24 10:24:20 -04001004 btrfs_release_path(root, path);
Chris Masone02119d2008-09-05 16:13:11 -04001005 if (nlink != inode->i_nlink) {
1006 inode->i_nlink = nlink;
1007 btrfs_update_inode(trans, root, inode);
1008 }
Chris Mason8d5bf1c2008-09-11 15:51:21 -04001009 BTRFS_I(inode)->index_cnt = (u64)-1;
Chris Masone02119d2008-09-05 16:13:11 -04001010
Yan, Zhengc71bf092009-11-12 09:34:40 +00001011 if (inode->i_nlink == 0) {
1012 if (S_ISDIR(inode->i_mode)) {
1013 ret = replay_dir_deletes(trans, root, NULL, path,
1014 inode->i_ino, 1);
1015 BUG_ON(ret);
1016 }
1017 ret = insert_orphan_item(trans, root, inode->i_ino);
Chris Mason12fcfd22009-03-24 10:24:20 -04001018 BUG_ON(ret);
1019 }
1020 btrfs_free_path(path);
1021
Chris Masone02119d2008-09-05 16:13:11 -04001022 return 0;
1023}
1024
1025static noinline int fixup_inode_link_counts(struct btrfs_trans_handle *trans,
1026 struct btrfs_root *root,
1027 struct btrfs_path *path)
1028{
1029 int ret;
1030 struct btrfs_key key;
1031 struct inode *inode;
1032
1033 key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID;
1034 key.type = BTRFS_ORPHAN_ITEM_KEY;
1035 key.offset = (u64)-1;
Chris Masond3977122009-01-05 21:25:51 -05001036 while (1) {
Chris Masone02119d2008-09-05 16:13:11 -04001037 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1038 if (ret < 0)
1039 break;
1040
1041 if (ret == 1) {
1042 if (path->slots[0] == 0)
1043 break;
1044 path->slots[0]--;
1045 }
1046
1047 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1048 if (key.objectid != BTRFS_TREE_LOG_FIXUP_OBJECTID ||
1049 key.type != BTRFS_ORPHAN_ITEM_KEY)
1050 break;
1051
1052 ret = btrfs_del_item(trans, root, path);
1053 BUG_ON(ret);
1054
1055 btrfs_release_path(root, path);
1056 inode = read_one_inode(root, key.offset);
1057 BUG_ON(!inode);
1058
1059 ret = fixup_inode_link_count(trans, root, inode);
1060 BUG_ON(ret);
1061
1062 iput(inode);
1063
Chris Mason12fcfd22009-03-24 10:24:20 -04001064 /*
1065 * fixup on a directory may create new entries,
1066 * make sure we always look for the highset possible
1067 * offset
1068 */
1069 key.offset = (u64)-1;
Chris Masone02119d2008-09-05 16:13:11 -04001070 }
1071 btrfs_release_path(root, path);
1072 return 0;
1073}
1074
1075
1076/*
1077 * record a given inode in the fixup dir so we can check its link
1078 * count when replay is done. The link count is incremented here
1079 * so the inode won't go away until we check it
1080 */
1081static noinline int link_to_fixup_dir(struct btrfs_trans_handle *trans,
1082 struct btrfs_root *root,
1083 struct btrfs_path *path,
1084 u64 objectid)
1085{
1086 struct btrfs_key key;
1087 int ret = 0;
1088 struct inode *inode;
1089
1090 inode = read_one_inode(root, objectid);
1091 BUG_ON(!inode);
1092
1093 key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID;
1094 btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
1095 key.offset = objectid;
1096
1097 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1098
1099 btrfs_release_path(root, path);
1100 if (ret == 0) {
1101 btrfs_inc_nlink(inode);
1102 btrfs_update_inode(trans, root, inode);
1103 } else if (ret == -EEXIST) {
1104 ret = 0;
1105 } else {
1106 BUG();
1107 }
1108 iput(inode);
1109
1110 return ret;
1111}
1112
1113/*
1114 * when replaying the log for a directory, we only insert names
1115 * for inodes that actually exist. This means an fsync on a directory
1116 * does not implicitly fsync all the new files in it
1117 */
1118static noinline int insert_one_name(struct btrfs_trans_handle *trans,
1119 struct btrfs_root *root,
1120 struct btrfs_path *path,
1121 u64 dirid, u64 index,
1122 char *name, int name_len, u8 type,
1123 struct btrfs_key *location)
1124{
1125 struct inode *inode;
1126 struct inode *dir;
1127 int ret;
1128
1129 inode = read_one_inode(root, location->objectid);
1130 if (!inode)
1131 return -ENOENT;
1132
1133 dir = read_one_inode(root, dirid);
1134 if (!dir) {
1135 iput(inode);
1136 return -EIO;
1137 }
1138 ret = btrfs_add_link(trans, dir, inode, name, name_len, 1, index);
1139
1140 /* FIXME, put inode into FIXUP list */
1141
1142 iput(inode);
1143 iput(dir);
1144 return ret;
1145}
1146
1147/*
1148 * take a single entry in a log directory item and replay it into
1149 * the subvolume.
1150 *
1151 * if a conflicting item exists in the subdirectory already,
1152 * the inode it points to is unlinked and put into the link count
1153 * fix up tree.
1154 *
1155 * If a name from the log points to a file or directory that does
1156 * not exist in the FS, it is skipped. fsyncs on directories
1157 * do not force down inodes inside that directory, just changes to the
1158 * names or unlinks in a directory.
1159 */
1160static noinline int replay_one_name(struct btrfs_trans_handle *trans,
1161 struct btrfs_root *root,
1162 struct btrfs_path *path,
1163 struct extent_buffer *eb,
1164 struct btrfs_dir_item *di,
1165 struct btrfs_key *key)
1166{
1167 char *name;
1168 int name_len;
1169 struct btrfs_dir_item *dst_di;
1170 struct btrfs_key found_key;
1171 struct btrfs_key log_key;
1172 struct inode *dir;
Chris Masone02119d2008-09-05 16:13:11 -04001173 u8 log_type;
Chris Mason4bef0842008-09-08 11:18:08 -04001174 int exists;
Chris Masone02119d2008-09-05 16:13:11 -04001175 int ret;
1176
1177 dir = read_one_inode(root, key->objectid);
1178 BUG_ON(!dir);
1179
1180 name_len = btrfs_dir_name_len(eb, di);
1181 name = kmalloc(name_len, GFP_NOFS);
1182 log_type = btrfs_dir_type(eb, di);
1183 read_extent_buffer(eb, name, (unsigned long)(di + 1),
1184 name_len);
1185
1186 btrfs_dir_item_key_to_cpu(eb, di, &log_key);
Chris Mason4bef0842008-09-08 11:18:08 -04001187 exists = btrfs_lookup_inode(trans, root, path, &log_key, 0);
1188 if (exists == 0)
1189 exists = 1;
1190 else
1191 exists = 0;
1192 btrfs_release_path(root, path);
1193
Chris Masone02119d2008-09-05 16:13:11 -04001194 if (key->type == BTRFS_DIR_ITEM_KEY) {
1195 dst_di = btrfs_lookup_dir_item(trans, root, path, key->objectid,
1196 name, name_len, 1);
Chris Masond3977122009-01-05 21:25:51 -05001197 } else if (key->type == BTRFS_DIR_INDEX_KEY) {
Chris Masone02119d2008-09-05 16:13:11 -04001198 dst_di = btrfs_lookup_dir_index_item(trans, root, path,
1199 key->objectid,
1200 key->offset, name,
1201 name_len, 1);
1202 } else {
1203 BUG();
1204 }
1205 if (!dst_di || IS_ERR(dst_di)) {
1206 /* we need a sequence number to insert, so we only
1207 * do inserts for the BTRFS_DIR_INDEX_KEY types
1208 */
1209 if (key->type != BTRFS_DIR_INDEX_KEY)
1210 goto out;
1211 goto insert;
1212 }
1213
1214 btrfs_dir_item_key_to_cpu(path->nodes[0], dst_di, &found_key);
1215 /* the existing item matches the logged item */
1216 if (found_key.objectid == log_key.objectid &&
1217 found_key.type == log_key.type &&
1218 found_key.offset == log_key.offset &&
1219 btrfs_dir_type(path->nodes[0], dst_di) == log_type) {
1220 goto out;
1221 }
1222
1223 /*
1224 * don't drop the conflicting directory entry if the inode
1225 * for the new entry doesn't exist
1226 */
Chris Mason4bef0842008-09-08 11:18:08 -04001227 if (!exists)
Chris Masone02119d2008-09-05 16:13:11 -04001228 goto out;
1229
Chris Masone02119d2008-09-05 16:13:11 -04001230 ret = drop_one_dir_item(trans, root, path, dir, dst_di);
1231 BUG_ON(ret);
1232
1233 if (key->type == BTRFS_DIR_INDEX_KEY)
1234 goto insert;
1235out:
1236 btrfs_release_path(root, path);
1237 kfree(name);
1238 iput(dir);
1239 return 0;
1240
1241insert:
1242 btrfs_release_path(root, path);
1243 ret = insert_one_name(trans, root, path, key->objectid, key->offset,
1244 name, name_len, log_type, &log_key);
1245
Stoyan Gaydarovc2934982009-04-02 17:05:11 -04001246 BUG_ON(ret && ret != -ENOENT);
Chris Masone02119d2008-09-05 16:13:11 -04001247 goto out;
1248}
1249
1250/*
1251 * find all the names in a directory item and reconcile them into
1252 * the subvolume. Only BTRFS_DIR_ITEM_KEY types will have more than
1253 * one name in a directory item, but the same code gets used for
1254 * both directory index types
1255 */
1256static noinline int replay_one_dir_item(struct btrfs_trans_handle *trans,
1257 struct btrfs_root *root,
1258 struct btrfs_path *path,
1259 struct extent_buffer *eb, int slot,
1260 struct btrfs_key *key)
1261{
1262 int ret;
1263 u32 item_size = btrfs_item_size_nr(eb, slot);
1264 struct btrfs_dir_item *di;
1265 int name_len;
1266 unsigned long ptr;
1267 unsigned long ptr_end;
1268
1269 ptr = btrfs_item_ptr_offset(eb, slot);
1270 ptr_end = ptr + item_size;
Chris Masond3977122009-01-05 21:25:51 -05001271 while (ptr < ptr_end) {
Chris Masone02119d2008-09-05 16:13:11 -04001272 di = (struct btrfs_dir_item *)ptr;
1273 name_len = btrfs_dir_name_len(eb, di);
1274 ret = replay_one_name(trans, root, path, eb, di, key);
1275 BUG_ON(ret);
1276 ptr = (unsigned long)(di + 1);
1277 ptr += name_len;
1278 }
1279 return 0;
1280}
1281
1282/*
1283 * directory replay has two parts. There are the standard directory
1284 * items in the log copied from the subvolume, and range items
1285 * created in the log while the subvolume was logged.
1286 *
1287 * The range items tell us which parts of the key space the log
1288 * is authoritative for. During replay, if a key in the subvolume
1289 * directory is in a logged range item, but not actually in the log
1290 * that means it was deleted from the directory before the fsync
1291 * and should be removed.
1292 */
1293static noinline int find_dir_range(struct btrfs_root *root,
1294 struct btrfs_path *path,
1295 u64 dirid, int key_type,
1296 u64 *start_ret, u64 *end_ret)
1297{
1298 struct btrfs_key key;
1299 u64 found_end;
1300 struct btrfs_dir_log_item *item;
1301 int ret;
1302 int nritems;
1303
1304 if (*start_ret == (u64)-1)
1305 return 1;
1306
1307 key.objectid = dirid;
1308 key.type = key_type;
1309 key.offset = *start_ret;
1310
1311 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1312 if (ret < 0)
1313 goto out;
1314 if (ret > 0) {
1315 if (path->slots[0] == 0)
1316 goto out;
1317 path->slots[0]--;
1318 }
1319 if (ret != 0)
1320 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1321
1322 if (key.type != key_type || key.objectid != dirid) {
1323 ret = 1;
1324 goto next;
1325 }
1326 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
1327 struct btrfs_dir_log_item);
1328 found_end = btrfs_dir_log_end(path->nodes[0], item);
1329
1330 if (*start_ret >= key.offset && *start_ret <= found_end) {
1331 ret = 0;
1332 *start_ret = key.offset;
1333 *end_ret = found_end;
1334 goto out;
1335 }
1336 ret = 1;
1337next:
1338 /* check the next slot in the tree to see if it is a valid item */
1339 nritems = btrfs_header_nritems(path->nodes[0]);
1340 if (path->slots[0] >= nritems) {
1341 ret = btrfs_next_leaf(root, path);
1342 if (ret)
1343 goto out;
1344 } else {
1345 path->slots[0]++;
1346 }
1347
1348 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1349
1350 if (key.type != key_type || key.objectid != dirid) {
1351 ret = 1;
1352 goto out;
1353 }
1354 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
1355 struct btrfs_dir_log_item);
1356 found_end = btrfs_dir_log_end(path->nodes[0], item);
1357 *start_ret = key.offset;
1358 *end_ret = found_end;
1359 ret = 0;
1360out:
1361 btrfs_release_path(root, path);
1362 return ret;
1363}
1364
1365/*
1366 * this looks for a given directory item in the log. If the directory
1367 * item is not in the log, the item is removed and the inode it points
1368 * to is unlinked
1369 */
1370static noinline int check_item_in_log(struct btrfs_trans_handle *trans,
1371 struct btrfs_root *root,
1372 struct btrfs_root *log,
1373 struct btrfs_path *path,
1374 struct btrfs_path *log_path,
1375 struct inode *dir,
1376 struct btrfs_key *dir_key)
1377{
1378 int ret;
1379 struct extent_buffer *eb;
1380 int slot;
1381 u32 item_size;
1382 struct btrfs_dir_item *di;
1383 struct btrfs_dir_item *log_di;
1384 int name_len;
1385 unsigned long ptr;
1386 unsigned long ptr_end;
1387 char *name;
1388 struct inode *inode;
1389 struct btrfs_key location;
1390
1391again:
1392 eb = path->nodes[0];
1393 slot = path->slots[0];
1394 item_size = btrfs_item_size_nr(eb, slot);
1395 ptr = btrfs_item_ptr_offset(eb, slot);
1396 ptr_end = ptr + item_size;
Chris Masond3977122009-01-05 21:25:51 -05001397 while (ptr < ptr_end) {
Chris Masone02119d2008-09-05 16:13:11 -04001398 di = (struct btrfs_dir_item *)ptr;
1399 name_len = btrfs_dir_name_len(eb, di);
1400 name = kmalloc(name_len, GFP_NOFS);
1401 if (!name) {
1402 ret = -ENOMEM;
1403 goto out;
1404 }
1405 read_extent_buffer(eb, name, (unsigned long)(di + 1),
1406 name_len);
1407 log_di = NULL;
Chris Mason12fcfd22009-03-24 10:24:20 -04001408 if (log && dir_key->type == BTRFS_DIR_ITEM_KEY) {
Chris Masone02119d2008-09-05 16:13:11 -04001409 log_di = btrfs_lookup_dir_item(trans, log, log_path,
1410 dir_key->objectid,
1411 name, name_len, 0);
Chris Mason12fcfd22009-03-24 10:24:20 -04001412 } else if (log && dir_key->type == BTRFS_DIR_INDEX_KEY) {
Chris Masone02119d2008-09-05 16:13:11 -04001413 log_di = btrfs_lookup_dir_index_item(trans, log,
1414 log_path,
1415 dir_key->objectid,
1416 dir_key->offset,
1417 name, name_len, 0);
1418 }
1419 if (!log_di || IS_ERR(log_di)) {
1420 btrfs_dir_item_key_to_cpu(eb, di, &location);
1421 btrfs_release_path(root, path);
1422 btrfs_release_path(log, log_path);
1423 inode = read_one_inode(root, location.objectid);
1424 BUG_ON(!inode);
1425
1426 ret = link_to_fixup_dir(trans, root,
1427 path, location.objectid);
1428 BUG_ON(ret);
1429 btrfs_inc_nlink(inode);
1430 ret = btrfs_unlink_inode(trans, root, dir, inode,
1431 name, name_len);
1432 BUG_ON(ret);
1433 kfree(name);
1434 iput(inode);
1435
1436 /* there might still be more names under this key
1437 * check and repeat if required
1438 */
1439 ret = btrfs_search_slot(NULL, root, dir_key, path,
1440 0, 0);
1441 if (ret == 0)
1442 goto again;
1443 ret = 0;
1444 goto out;
1445 }
1446 btrfs_release_path(log, log_path);
1447 kfree(name);
1448
1449 ptr = (unsigned long)(di + 1);
1450 ptr += name_len;
1451 }
1452 ret = 0;
1453out:
1454 btrfs_release_path(root, path);
1455 btrfs_release_path(log, log_path);
1456 return ret;
1457}
1458
1459/*
1460 * deletion replay happens before we copy any new directory items
1461 * out of the log or out of backreferences from inodes. It
1462 * scans the log to find ranges of keys that log is authoritative for,
1463 * and then scans the directory to find items in those ranges that are
1464 * not present in the log.
1465 *
1466 * Anything we don't find in the log is unlinked and removed from the
1467 * directory.
1468 */
1469static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans,
1470 struct btrfs_root *root,
1471 struct btrfs_root *log,
1472 struct btrfs_path *path,
Chris Mason12fcfd22009-03-24 10:24:20 -04001473 u64 dirid, int del_all)
Chris Masone02119d2008-09-05 16:13:11 -04001474{
1475 u64 range_start;
1476 u64 range_end;
1477 int key_type = BTRFS_DIR_LOG_ITEM_KEY;
1478 int ret = 0;
1479 struct btrfs_key dir_key;
1480 struct btrfs_key found_key;
1481 struct btrfs_path *log_path;
1482 struct inode *dir;
1483
1484 dir_key.objectid = dirid;
1485 dir_key.type = BTRFS_DIR_ITEM_KEY;
1486 log_path = btrfs_alloc_path();
1487 if (!log_path)
1488 return -ENOMEM;
1489
1490 dir = read_one_inode(root, dirid);
1491 /* it isn't an error if the inode isn't there, that can happen
1492 * because we replay the deletes before we copy in the inode item
1493 * from the log
1494 */
1495 if (!dir) {
1496 btrfs_free_path(log_path);
1497 return 0;
1498 }
1499again:
1500 range_start = 0;
1501 range_end = 0;
Chris Masond3977122009-01-05 21:25:51 -05001502 while (1) {
Chris Mason12fcfd22009-03-24 10:24:20 -04001503 if (del_all)
1504 range_end = (u64)-1;
1505 else {
1506 ret = find_dir_range(log, path, dirid, key_type,
1507 &range_start, &range_end);
1508 if (ret != 0)
1509 break;
1510 }
Chris Masone02119d2008-09-05 16:13:11 -04001511
1512 dir_key.offset = range_start;
Chris Masond3977122009-01-05 21:25:51 -05001513 while (1) {
Chris Masone02119d2008-09-05 16:13:11 -04001514 int nritems;
1515 ret = btrfs_search_slot(NULL, root, &dir_key, path,
1516 0, 0);
1517 if (ret < 0)
1518 goto out;
1519
1520 nritems = btrfs_header_nritems(path->nodes[0]);
1521 if (path->slots[0] >= nritems) {
1522 ret = btrfs_next_leaf(root, path);
1523 if (ret)
1524 break;
1525 }
1526 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1527 path->slots[0]);
1528 if (found_key.objectid != dirid ||
1529 found_key.type != dir_key.type)
1530 goto next_type;
1531
1532 if (found_key.offset > range_end)
1533 break;
1534
1535 ret = check_item_in_log(trans, root, log, path,
Chris Mason12fcfd22009-03-24 10:24:20 -04001536 log_path, dir,
1537 &found_key);
Chris Masone02119d2008-09-05 16:13:11 -04001538 BUG_ON(ret);
1539 if (found_key.offset == (u64)-1)
1540 break;
1541 dir_key.offset = found_key.offset + 1;
1542 }
1543 btrfs_release_path(root, path);
1544 if (range_end == (u64)-1)
1545 break;
1546 range_start = range_end + 1;
1547 }
1548
1549next_type:
1550 ret = 0;
1551 if (key_type == BTRFS_DIR_LOG_ITEM_KEY) {
1552 key_type = BTRFS_DIR_LOG_INDEX_KEY;
1553 dir_key.type = BTRFS_DIR_INDEX_KEY;
1554 btrfs_release_path(root, path);
1555 goto again;
1556 }
1557out:
1558 btrfs_release_path(root, path);
1559 btrfs_free_path(log_path);
1560 iput(dir);
1561 return ret;
1562}
1563
1564/*
1565 * the process_func used to replay items from the log tree. This
1566 * gets called in two different stages. The first stage just looks
1567 * for inodes and makes sure they are all copied into the subvolume.
1568 *
1569 * The second stage copies all the other item types from the log into
1570 * the subvolume. The two stage approach is slower, but gets rid of
1571 * lots of complexity around inodes referencing other inodes that exist
1572 * only in the log (references come from either directory items or inode
1573 * back refs).
1574 */
1575static int replay_one_buffer(struct btrfs_root *log, struct extent_buffer *eb,
1576 struct walk_control *wc, u64 gen)
1577{
1578 int nritems;
1579 struct btrfs_path *path;
1580 struct btrfs_root *root = wc->replay_dest;
1581 struct btrfs_key key;
1582 u32 item_size;
1583 int level;
1584 int i;
1585 int ret;
1586
1587 btrfs_read_buffer(eb, gen);
1588
1589 level = btrfs_header_level(eb);
1590
1591 if (level != 0)
1592 return 0;
1593
1594 path = btrfs_alloc_path();
1595 BUG_ON(!path);
1596
1597 nritems = btrfs_header_nritems(eb);
1598 for (i = 0; i < nritems; i++) {
1599 btrfs_item_key_to_cpu(eb, &key, i);
1600 item_size = btrfs_item_size_nr(eb, i);
1601
1602 /* inode keys are done during the first stage */
1603 if (key.type == BTRFS_INODE_ITEM_KEY &&
1604 wc->stage == LOG_WALK_REPLAY_INODES) {
Chris Masone02119d2008-09-05 16:13:11 -04001605 struct btrfs_inode_item *inode_item;
1606 u32 mode;
1607
1608 inode_item = btrfs_item_ptr(eb, i,
1609 struct btrfs_inode_item);
1610 mode = btrfs_inode_mode(eb, inode_item);
1611 if (S_ISDIR(mode)) {
1612 ret = replay_dir_deletes(wc->trans,
Chris Mason12fcfd22009-03-24 10:24:20 -04001613 root, log, path, key.objectid, 0);
Chris Masone02119d2008-09-05 16:13:11 -04001614 BUG_ON(ret);
1615 }
1616 ret = overwrite_item(wc->trans, root, path,
1617 eb, i, &key);
1618 BUG_ON(ret);
1619
Yan, Zhengc71bf092009-11-12 09:34:40 +00001620 /* for regular files, make sure corresponding
1621 * orhpan item exist. extents past the new EOF
1622 * will be truncated later by orphan cleanup.
Chris Masone02119d2008-09-05 16:13:11 -04001623 */
1624 if (S_ISREG(mode)) {
Yan, Zhengc71bf092009-11-12 09:34:40 +00001625 ret = insert_orphan_item(wc->trans, root,
1626 key.objectid);
Chris Masone02119d2008-09-05 16:13:11 -04001627 BUG_ON(ret);
Chris Masone02119d2008-09-05 16:13:11 -04001628 }
Yan, Zhengc71bf092009-11-12 09:34:40 +00001629
Chris Masone02119d2008-09-05 16:13:11 -04001630 ret = link_to_fixup_dir(wc->trans, root,
1631 path, key.objectid);
1632 BUG_ON(ret);
1633 }
1634 if (wc->stage < LOG_WALK_REPLAY_ALL)
1635 continue;
1636
1637 /* these keys are simply copied */
1638 if (key.type == BTRFS_XATTR_ITEM_KEY) {
1639 ret = overwrite_item(wc->trans, root, path,
1640 eb, i, &key);
1641 BUG_ON(ret);
1642 } else if (key.type == BTRFS_INODE_REF_KEY) {
1643 ret = add_inode_ref(wc->trans, root, log, path,
1644 eb, i, &key);
1645 BUG_ON(ret && ret != -ENOENT);
1646 } else if (key.type == BTRFS_EXTENT_DATA_KEY) {
1647 ret = replay_one_extent(wc->trans, root, path,
1648 eb, i, &key);
1649 BUG_ON(ret);
Chris Masone02119d2008-09-05 16:13:11 -04001650 } else if (key.type == BTRFS_DIR_ITEM_KEY ||
1651 key.type == BTRFS_DIR_INDEX_KEY) {
1652 ret = replay_one_dir_item(wc->trans, root, path,
1653 eb, i, &key);
1654 BUG_ON(ret);
1655 }
1656 }
1657 btrfs_free_path(path);
1658 return 0;
1659}
1660
Chris Masond3977122009-01-05 21:25:51 -05001661static noinline int walk_down_log_tree(struct btrfs_trans_handle *trans,
Chris Masone02119d2008-09-05 16:13:11 -04001662 struct btrfs_root *root,
1663 struct btrfs_path *path, int *level,
1664 struct walk_control *wc)
1665{
1666 u64 root_owner;
1667 u64 root_gen;
1668 u64 bytenr;
1669 u64 ptr_gen;
1670 struct extent_buffer *next;
1671 struct extent_buffer *cur;
1672 struct extent_buffer *parent;
1673 u32 blocksize;
1674 int ret = 0;
1675
1676 WARN_ON(*level < 0);
1677 WARN_ON(*level >= BTRFS_MAX_LEVEL);
1678
Chris Masond3977122009-01-05 21:25:51 -05001679 while (*level > 0) {
Chris Masone02119d2008-09-05 16:13:11 -04001680 WARN_ON(*level < 0);
1681 WARN_ON(*level >= BTRFS_MAX_LEVEL);
1682 cur = path->nodes[*level];
1683
1684 if (btrfs_header_level(cur) != *level)
1685 WARN_ON(1);
1686
1687 if (path->slots[*level] >=
1688 btrfs_header_nritems(cur))
1689 break;
1690
1691 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
1692 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
1693 blocksize = btrfs_level_size(root, *level - 1);
1694
1695 parent = path->nodes[*level];
1696 root_owner = btrfs_header_owner(parent);
1697 root_gen = btrfs_header_generation(parent);
1698
1699 next = btrfs_find_create_tree_block(root, bytenr, blocksize);
1700
1701 wc->process_func(root, next, wc, ptr_gen);
1702
1703 if (*level == 1) {
1704 path->slots[*level]++;
1705 if (wc->free) {
1706 btrfs_read_buffer(next, ptr_gen);
1707
1708 btrfs_tree_lock(next);
1709 clean_tree_block(trans, root, next);
Chris Masonb4ce94d2009-02-04 09:25:08 -05001710 btrfs_set_lock_blocking(next);
Chris Masone02119d2008-09-05 16:13:11 -04001711 btrfs_wait_tree_block_writeback(next);
1712 btrfs_tree_unlock(next);
1713
Chris Masone02119d2008-09-05 16:13:11 -04001714 WARN_ON(root_owner !=
1715 BTRFS_TREE_LOG_OBJECTID);
Chris Masond00aff02008-09-11 15:54:42 -04001716 ret = btrfs_free_reserved_extent(root,
1717 bytenr, blocksize);
Chris Masone02119d2008-09-05 16:13:11 -04001718 BUG_ON(ret);
1719 }
1720 free_extent_buffer(next);
1721 continue;
1722 }
1723 btrfs_read_buffer(next, ptr_gen);
1724
1725 WARN_ON(*level <= 0);
1726 if (path->nodes[*level-1])
1727 free_extent_buffer(path->nodes[*level-1]);
1728 path->nodes[*level-1] = next;
1729 *level = btrfs_header_level(next);
1730 path->slots[*level] = 0;
1731 cond_resched();
1732 }
1733 WARN_ON(*level < 0);
1734 WARN_ON(*level >= BTRFS_MAX_LEVEL);
1735
Chris Masond3977122009-01-05 21:25:51 -05001736 if (path->nodes[*level] == root->node)
Chris Masone02119d2008-09-05 16:13:11 -04001737 parent = path->nodes[*level];
Chris Masond3977122009-01-05 21:25:51 -05001738 else
Chris Masone02119d2008-09-05 16:13:11 -04001739 parent = path->nodes[*level + 1];
Chris Masond3977122009-01-05 21:25:51 -05001740
Chris Masone02119d2008-09-05 16:13:11 -04001741 bytenr = path->nodes[*level]->start;
1742
1743 blocksize = btrfs_level_size(root, *level);
1744 root_owner = btrfs_header_owner(parent);
1745 root_gen = btrfs_header_generation(parent);
1746
1747 wc->process_func(root, path->nodes[*level], wc,
1748 btrfs_header_generation(path->nodes[*level]));
1749
1750 if (wc->free) {
1751 next = path->nodes[*level];
1752 btrfs_tree_lock(next);
1753 clean_tree_block(trans, root, next);
Chris Masonb4ce94d2009-02-04 09:25:08 -05001754 btrfs_set_lock_blocking(next);
Chris Masone02119d2008-09-05 16:13:11 -04001755 btrfs_wait_tree_block_writeback(next);
1756 btrfs_tree_unlock(next);
1757
Chris Masone02119d2008-09-05 16:13:11 -04001758 WARN_ON(root_owner != BTRFS_TREE_LOG_OBJECTID);
Chris Masond00aff02008-09-11 15:54:42 -04001759 ret = btrfs_free_reserved_extent(root, bytenr, blocksize);
Chris Masone02119d2008-09-05 16:13:11 -04001760 BUG_ON(ret);
1761 }
1762 free_extent_buffer(path->nodes[*level]);
1763 path->nodes[*level] = NULL;
1764 *level += 1;
1765
1766 cond_resched();
1767 return 0;
1768}
1769
Chris Masond3977122009-01-05 21:25:51 -05001770static noinline int walk_up_log_tree(struct btrfs_trans_handle *trans,
Chris Masone02119d2008-09-05 16:13:11 -04001771 struct btrfs_root *root,
1772 struct btrfs_path *path, int *level,
1773 struct walk_control *wc)
1774{
1775 u64 root_owner;
1776 u64 root_gen;
1777 int i;
1778 int slot;
1779 int ret;
1780
Chris Masond3977122009-01-05 21:25:51 -05001781 for (i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) {
Chris Masone02119d2008-09-05 16:13:11 -04001782 slot = path->slots[i];
1783 if (slot < btrfs_header_nritems(path->nodes[i]) - 1) {
1784 struct extent_buffer *node;
1785 node = path->nodes[i];
1786 path->slots[i]++;
1787 *level = i;
1788 WARN_ON(*level == 0);
1789 return 0;
1790 } else {
Zheng Yan31840ae2008-09-23 13:14:14 -04001791 struct extent_buffer *parent;
1792 if (path->nodes[*level] == root->node)
1793 parent = path->nodes[*level];
1794 else
1795 parent = path->nodes[*level + 1];
1796
1797 root_owner = btrfs_header_owner(parent);
1798 root_gen = btrfs_header_generation(parent);
Chris Masone02119d2008-09-05 16:13:11 -04001799 wc->process_func(root, path->nodes[*level], wc,
1800 btrfs_header_generation(path->nodes[*level]));
1801 if (wc->free) {
1802 struct extent_buffer *next;
1803
1804 next = path->nodes[*level];
1805
1806 btrfs_tree_lock(next);
1807 clean_tree_block(trans, root, next);
Chris Masonb4ce94d2009-02-04 09:25:08 -05001808 btrfs_set_lock_blocking(next);
Chris Masone02119d2008-09-05 16:13:11 -04001809 btrfs_wait_tree_block_writeback(next);
1810 btrfs_tree_unlock(next);
1811
Chris Masone02119d2008-09-05 16:13:11 -04001812 WARN_ON(root_owner != BTRFS_TREE_LOG_OBJECTID);
Chris Masond00aff02008-09-11 15:54:42 -04001813 ret = btrfs_free_reserved_extent(root,
Chris Masone02119d2008-09-05 16:13:11 -04001814 path->nodes[*level]->start,
Chris Masond00aff02008-09-11 15:54:42 -04001815 path->nodes[*level]->len);
Chris Masone02119d2008-09-05 16:13:11 -04001816 BUG_ON(ret);
1817 }
1818 free_extent_buffer(path->nodes[*level]);
1819 path->nodes[*level] = NULL;
1820 *level = i + 1;
1821 }
1822 }
1823 return 1;
1824}
1825
1826/*
1827 * drop the reference count on the tree rooted at 'snap'. This traverses
1828 * the tree freeing any blocks that have a ref count of zero after being
1829 * decremented.
1830 */
1831static int walk_log_tree(struct btrfs_trans_handle *trans,
1832 struct btrfs_root *log, struct walk_control *wc)
1833{
1834 int ret = 0;
1835 int wret;
1836 int level;
1837 struct btrfs_path *path;
1838 int i;
1839 int orig_level;
1840
1841 path = btrfs_alloc_path();
1842 BUG_ON(!path);
1843
1844 level = btrfs_header_level(log->node);
1845 orig_level = level;
1846 path->nodes[level] = log->node;
1847 extent_buffer_get(log->node);
1848 path->slots[level] = 0;
1849
Chris Masond3977122009-01-05 21:25:51 -05001850 while (1) {
Chris Masone02119d2008-09-05 16:13:11 -04001851 wret = walk_down_log_tree(trans, log, path, &level, wc);
1852 if (wret > 0)
1853 break;
1854 if (wret < 0)
1855 ret = wret;
1856
1857 wret = walk_up_log_tree(trans, log, path, &level, wc);
1858 if (wret > 0)
1859 break;
1860 if (wret < 0)
1861 ret = wret;
1862 }
1863
1864 /* was the root node processed? if not, catch it here */
1865 if (path->nodes[orig_level]) {
1866 wc->process_func(log, path->nodes[orig_level], wc,
1867 btrfs_header_generation(path->nodes[orig_level]));
1868 if (wc->free) {
1869 struct extent_buffer *next;
1870
1871 next = path->nodes[orig_level];
1872
1873 btrfs_tree_lock(next);
1874 clean_tree_block(trans, log, next);
Chris Masonb4ce94d2009-02-04 09:25:08 -05001875 btrfs_set_lock_blocking(next);
Chris Masone02119d2008-09-05 16:13:11 -04001876 btrfs_wait_tree_block_writeback(next);
1877 btrfs_tree_unlock(next);
1878
Chris Masone02119d2008-09-05 16:13:11 -04001879 WARN_ON(log->root_key.objectid !=
1880 BTRFS_TREE_LOG_OBJECTID);
Chris Masond00aff02008-09-11 15:54:42 -04001881 ret = btrfs_free_reserved_extent(log, next->start,
1882 next->len);
Chris Masone02119d2008-09-05 16:13:11 -04001883 BUG_ON(ret);
1884 }
1885 }
1886
1887 for (i = 0; i <= orig_level; i++) {
1888 if (path->nodes[i]) {
1889 free_extent_buffer(path->nodes[i]);
1890 path->nodes[i] = NULL;
1891 }
1892 }
1893 btrfs_free_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04001894 return ret;
1895}
1896
Yan Zheng7237f182009-01-21 12:54:03 -05001897/*
1898 * helper function to update the item for a given subvolumes log root
1899 * in the tree of log roots
1900 */
1901static int update_log_root(struct btrfs_trans_handle *trans,
1902 struct btrfs_root *log)
1903{
1904 int ret;
1905
1906 if (log->log_transid == 1) {
1907 /* insert root item on the first sync */
1908 ret = btrfs_insert_root(trans, log->fs_info->log_root_tree,
1909 &log->root_key, &log->root_item);
1910 } else {
1911 ret = btrfs_update_root(trans, log->fs_info->log_root_tree,
1912 &log->root_key, &log->root_item);
1913 }
1914 return ret;
1915}
1916
Chris Mason12fcfd22009-03-24 10:24:20 -04001917static int wait_log_commit(struct btrfs_trans_handle *trans,
1918 struct btrfs_root *root, unsigned long transid)
Chris Masone02119d2008-09-05 16:13:11 -04001919{
1920 DEFINE_WAIT(wait);
Yan Zheng7237f182009-01-21 12:54:03 -05001921 int index = transid % 2;
Chris Masone02119d2008-09-05 16:13:11 -04001922
Yan Zheng7237f182009-01-21 12:54:03 -05001923 /*
1924 * we only allow two pending log transactions at a time,
1925 * so we know that if ours is more than 2 older than the
1926 * current transaction, we're done
1927 */
Chris Masone02119d2008-09-05 16:13:11 -04001928 do {
Yan Zheng7237f182009-01-21 12:54:03 -05001929 prepare_to_wait(&root->log_commit_wait[index],
1930 &wait, TASK_UNINTERRUPTIBLE);
1931 mutex_unlock(&root->log_mutex);
Chris Mason12fcfd22009-03-24 10:24:20 -04001932
1933 if (root->fs_info->last_trans_log_full_commit !=
1934 trans->transid && root->log_transid < transid + 2 &&
Yan Zheng7237f182009-01-21 12:54:03 -05001935 atomic_read(&root->log_commit[index]))
Chris Masone02119d2008-09-05 16:13:11 -04001936 schedule();
Chris Mason12fcfd22009-03-24 10:24:20 -04001937
Yan Zheng7237f182009-01-21 12:54:03 -05001938 finish_wait(&root->log_commit_wait[index], &wait);
1939 mutex_lock(&root->log_mutex);
1940 } while (root->log_transid < transid + 2 &&
1941 atomic_read(&root->log_commit[index]));
1942 return 0;
1943}
1944
Chris Mason12fcfd22009-03-24 10:24:20 -04001945static int wait_for_writer(struct btrfs_trans_handle *trans,
1946 struct btrfs_root *root)
Yan Zheng7237f182009-01-21 12:54:03 -05001947{
1948 DEFINE_WAIT(wait);
1949 while (atomic_read(&root->log_writers)) {
1950 prepare_to_wait(&root->log_writer_wait,
1951 &wait, TASK_UNINTERRUPTIBLE);
1952 mutex_unlock(&root->log_mutex);
Chris Mason12fcfd22009-03-24 10:24:20 -04001953 if (root->fs_info->last_trans_log_full_commit !=
1954 trans->transid && atomic_read(&root->log_writers))
Yan Zheng7237f182009-01-21 12:54:03 -05001955 schedule();
1956 mutex_lock(&root->log_mutex);
1957 finish_wait(&root->log_writer_wait, &wait);
1958 }
Chris Masone02119d2008-09-05 16:13:11 -04001959 return 0;
1960}
1961
1962/*
1963 * btrfs_sync_log does sends a given tree log down to the disk and
1964 * updates the super blocks to record it. When this call is done,
Chris Mason12fcfd22009-03-24 10:24:20 -04001965 * you know that any inodes previously logged are safely on disk only
1966 * if it returns 0.
1967 *
1968 * Any other return value means you need to call btrfs_commit_transaction.
1969 * Some of the edge cases for fsyncing directories that have had unlinks
1970 * or renames done in the past mean that sometimes the only safe
1971 * fsync is to commit the whole FS. When btrfs_sync_log returns -EAGAIN,
1972 * that has happened.
Chris Masone02119d2008-09-05 16:13:11 -04001973 */
1974int btrfs_sync_log(struct btrfs_trans_handle *trans,
1975 struct btrfs_root *root)
1976{
Yan Zheng7237f182009-01-21 12:54:03 -05001977 int index1;
1978 int index2;
Yan, Zheng8cef4e12009-11-12 09:33:26 +00001979 int mark;
Chris Masone02119d2008-09-05 16:13:11 -04001980 int ret;
Chris Masone02119d2008-09-05 16:13:11 -04001981 struct btrfs_root *log = root->log_root;
Yan Zheng7237f182009-01-21 12:54:03 -05001982 struct btrfs_root *log_root_tree = root->fs_info->log_root_tree;
Yan, Zheng8cef4e12009-11-12 09:33:26 +00001983 unsigned long log_transid = 0;
Chris Masone02119d2008-09-05 16:13:11 -04001984
Yan Zheng7237f182009-01-21 12:54:03 -05001985 mutex_lock(&root->log_mutex);
1986 index1 = root->log_transid % 2;
1987 if (atomic_read(&root->log_commit[index1])) {
Chris Mason12fcfd22009-03-24 10:24:20 -04001988 wait_log_commit(trans, root, root->log_transid);
Yan Zheng7237f182009-01-21 12:54:03 -05001989 mutex_unlock(&root->log_mutex);
1990 return 0;
Chris Masone02119d2008-09-05 16:13:11 -04001991 }
Yan Zheng7237f182009-01-21 12:54:03 -05001992 atomic_set(&root->log_commit[index1], 1);
1993
1994 /* wait for previous tree log sync to complete */
1995 if (atomic_read(&root->log_commit[(index1 + 1) % 2]))
Chris Mason12fcfd22009-03-24 10:24:20 -04001996 wait_log_commit(trans, root, root->log_transid - 1);
Chris Masone02119d2008-09-05 16:13:11 -04001997
Yan, Zheng86df7eb2009-10-14 09:24:59 -04001998 while (1) {
Yan Zheng7237f182009-01-21 12:54:03 -05001999 unsigned long batch = root->log_batch;
Yan, Zheng86df7eb2009-10-14 09:24:59 -04002000 if (root->log_multiple_pids) {
2001 mutex_unlock(&root->log_mutex);
2002 schedule_timeout_uninterruptible(1);
2003 mutex_lock(&root->log_mutex);
2004 }
Chris Mason12fcfd22009-03-24 10:24:20 -04002005 wait_for_writer(trans, root);
Yan Zheng7237f182009-01-21 12:54:03 -05002006 if (batch == root->log_batch)
Chris Masone02119d2008-09-05 16:13:11 -04002007 break;
2008 }
Chris Masond0c803c2008-09-11 16:17:57 -04002009
Chris Mason12fcfd22009-03-24 10:24:20 -04002010 /* bail out if we need to do a full commit */
2011 if (root->fs_info->last_trans_log_full_commit == trans->transid) {
2012 ret = -EAGAIN;
2013 mutex_unlock(&root->log_mutex);
2014 goto out;
2015 }
2016
Yan, Zheng8cef4e12009-11-12 09:33:26 +00002017 log_transid = root->log_transid;
2018 if (log_transid % 2 == 0)
2019 mark = EXTENT_DIRTY;
2020 else
2021 mark = EXTENT_NEW;
2022
Chris Mason690587d2009-10-13 13:29:19 -04002023 /* we start IO on all the marked extents here, but we don't actually
2024 * wait for them until later.
2025 */
Yan, Zheng8cef4e12009-11-12 09:33:26 +00002026 ret = btrfs_write_marked_extents(log, &log->dirty_log_pages, mark);
Chris Masone02119d2008-09-05 16:13:11 -04002027 BUG_ON(ret);
Yan Zheng7237f182009-01-21 12:54:03 -05002028
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002029 btrfs_set_root_node(&log->root_item, log->node);
Yan Zheng7237f182009-01-21 12:54:03 -05002030
2031 root->log_batch = 0;
2032 root->log_transid++;
2033 log->log_transid = root->log_transid;
Josef Bacikff782e02009-10-08 15:30:04 -04002034 root->log_start_pid = 0;
Yan Zheng7237f182009-01-21 12:54:03 -05002035 smp_mb();
2036 /*
Yan, Zheng8cef4e12009-11-12 09:33:26 +00002037 * IO has been started, blocks of the log tree have WRITTEN flag set
2038 * in their headers. new modifications of the log will be written to
2039 * new positions. so it's safe to allow log writers to go in.
Yan Zheng7237f182009-01-21 12:54:03 -05002040 */
2041 mutex_unlock(&root->log_mutex);
2042
2043 mutex_lock(&log_root_tree->log_mutex);
2044 log_root_tree->log_batch++;
2045 atomic_inc(&log_root_tree->log_writers);
2046 mutex_unlock(&log_root_tree->log_mutex);
2047
2048 ret = update_log_root(trans, log);
2049 BUG_ON(ret);
2050
2051 mutex_lock(&log_root_tree->log_mutex);
2052 if (atomic_dec_and_test(&log_root_tree->log_writers)) {
2053 smp_mb();
2054 if (waitqueue_active(&log_root_tree->log_writer_wait))
2055 wake_up(&log_root_tree->log_writer_wait);
2056 }
2057
2058 index2 = log_root_tree->log_transid % 2;
2059 if (atomic_read(&log_root_tree->log_commit[index2])) {
Yan, Zheng8cef4e12009-11-12 09:33:26 +00002060 btrfs_wait_marked_extents(log, &log->dirty_log_pages, mark);
Chris Mason12fcfd22009-03-24 10:24:20 -04002061 wait_log_commit(trans, log_root_tree,
2062 log_root_tree->log_transid);
Yan Zheng7237f182009-01-21 12:54:03 -05002063 mutex_unlock(&log_root_tree->log_mutex);
2064 goto out;
2065 }
2066 atomic_set(&log_root_tree->log_commit[index2], 1);
2067
Chris Mason12fcfd22009-03-24 10:24:20 -04002068 if (atomic_read(&log_root_tree->log_commit[(index2 + 1) % 2])) {
2069 wait_log_commit(trans, log_root_tree,
2070 log_root_tree->log_transid - 1);
2071 }
Yan Zheng7237f182009-01-21 12:54:03 -05002072
Chris Mason12fcfd22009-03-24 10:24:20 -04002073 wait_for_writer(trans, log_root_tree);
2074
2075 /*
2076 * now that we've moved on to the tree of log tree roots,
2077 * check the full commit flag again
2078 */
2079 if (root->fs_info->last_trans_log_full_commit == trans->transid) {
Yan, Zheng8cef4e12009-11-12 09:33:26 +00002080 btrfs_wait_marked_extents(log, &log->dirty_log_pages, mark);
Chris Mason12fcfd22009-03-24 10:24:20 -04002081 mutex_unlock(&log_root_tree->log_mutex);
2082 ret = -EAGAIN;
2083 goto out_wake_log_root;
2084 }
Yan Zheng7237f182009-01-21 12:54:03 -05002085
2086 ret = btrfs_write_and_wait_marked_extents(log_root_tree,
Yan, Zheng8cef4e12009-11-12 09:33:26 +00002087 &log_root_tree->dirty_log_pages,
2088 EXTENT_DIRTY | EXTENT_NEW);
Chris Masone02119d2008-09-05 16:13:11 -04002089 BUG_ON(ret);
Yan, Zheng8cef4e12009-11-12 09:33:26 +00002090 btrfs_wait_marked_extents(log, &log->dirty_log_pages, mark);
Chris Masone02119d2008-09-05 16:13:11 -04002091
2092 btrfs_set_super_log_root(&root->fs_info->super_for_commit,
Yan Zheng7237f182009-01-21 12:54:03 -05002093 log_root_tree->node->start);
Chris Masone02119d2008-09-05 16:13:11 -04002094 btrfs_set_super_log_root_level(&root->fs_info->super_for_commit,
Yan Zheng7237f182009-01-21 12:54:03 -05002095 btrfs_header_level(log_root_tree->node));
Chris Masone02119d2008-09-05 16:13:11 -04002096
Yan Zheng7237f182009-01-21 12:54:03 -05002097 log_root_tree->log_batch = 0;
2098 log_root_tree->log_transid++;
Chris Masone02119d2008-09-05 16:13:11 -04002099 smp_mb();
Yan Zheng7237f182009-01-21 12:54:03 -05002100
2101 mutex_unlock(&log_root_tree->log_mutex);
2102
2103 /*
2104 * nobody else is going to jump in and write the the ctree
2105 * super here because the log_commit atomic below is protecting
2106 * us. We must be called with a transaction handle pinning
2107 * the running transaction open, so a full commit can't hop
2108 * in and cause problems either.
2109 */
Chris Mason47226072009-10-13 12:55:09 -04002110 write_ctree_super(trans, root->fs_info->tree_root, 1);
Chris Mason12fcfd22009-03-24 10:24:20 -04002111 ret = 0;
Yan Zheng7237f182009-01-21 12:54:03 -05002112
Chris Mason257c62e2009-10-13 13:21:08 -04002113 mutex_lock(&root->log_mutex);
2114 if (root->last_log_commit < log_transid)
2115 root->last_log_commit = log_transid;
2116 mutex_unlock(&root->log_mutex);
2117
Chris Mason12fcfd22009-03-24 10:24:20 -04002118out_wake_log_root:
Yan Zheng7237f182009-01-21 12:54:03 -05002119 atomic_set(&log_root_tree->log_commit[index2], 0);
2120 smp_mb();
2121 if (waitqueue_active(&log_root_tree->log_commit_wait[index2]))
2122 wake_up(&log_root_tree->log_commit_wait[index2]);
Chris Masone02119d2008-09-05 16:13:11 -04002123out:
Yan Zheng7237f182009-01-21 12:54:03 -05002124 atomic_set(&root->log_commit[index1], 0);
2125 smp_mb();
2126 if (waitqueue_active(&root->log_commit_wait[index1]))
2127 wake_up(&root->log_commit_wait[index1]);
Chris Masone02119d2008-09-05 16:13:11 -04002128 return 0;
Chris Masone02119d2008-09-05 16:13:11 -04002129}
2130
Chris Mason12fcfd22009-03-24 10:24:20 -04002131/*
2132 * free all the extents used by the tree log. This should be called
Chris Masone02119d2008-09-05 16:13:11 -04002133 * at commit time of the full transaction
2134 */
2135int btrfs_free_log(struct btrfs_trans_handle *trans, struct btrfs_root *root)
2136{
2137 int ret;
2138 struct btrfs_root *log;
2139 struct key;
Chris Masond0c803c2008-09-11 16:17:57 -04002140 u64 start;
2141 u64 end;
Chris Masone02119d2008-09-05 16:13:11 -04002142 struct walk_control wc = {
2143 .free = 1,
2144 .process_func = process_one_buffer
2145 };
2146
Yan Zheng07d400a2009-01-06 11:42:00 -05002147 if (!root->log_root || root->fs_info->log_root_recovering)
Chris Masone02119d2008-09-05 16:13:11 -04002148 return 0;
2149
2150 log = root->log_root;
2151 ret = walk_log_tree(trans, log, &wc);
2152 BUG_ON(ret);
2153
Chris Masond3977122009-01-05 21:25:51 -05002154 while (1) {
Chris Masond0c803c2008-09-11 16:17:57 -04002155 ret = find_first_extent_bit(&log->dirty_log_pages,
Yan, Zheng8cef4e12009-11-12 09:33:26 +00002156 0, &start, &end, EXTENT_DIRTY | EXTENT_NEW);
Chris Masond0c803c2008-09-11 16:17:57 -04002157 if (ret)
2158 break;
2159
Yan, Zheng8cef4e12009-11-12 09:33:26 +00002160 clear_extent_bits(&log->dirty_log_pages, start, end,
2161 EXTENT_DIRTY | EXTENT_NEW, GFP_NOFS);
Chris Masond0c803c2008-09-11 16:17:57 -04002162 }
2163
Yan Zheng7237f182009-01-21 12:54:03 -05002164 if (log->log_transid > 0) {
2165 ret = btrfs_del_root(trans, root->fs_info->log_root_tree,
2166 &log->root_key);
2167 BUG_ON(ret);
2168 }
Chris Masone02119d2008-09-05 16:13:11 -04002169 root->log_root = NULL;
Yan Zheng7237f182009-01-21 12:54:03 -05002170 free_extent_buffer(log->node);
2171 kfree(log);
Chris Masone02119d2008-09-05 16:13:11 -04002172 return 0;
2173}
2174
2175/*
Chris Masone02119d2008-09-05 16:13:11 -04002176 * If both a file and directory are logged, and unlinks or renames are
2177 * mixed in, we have a few interesting corners:
2178 *
2179 * create file X in dir Y
2180 * link file X to X.link in dir Y
2181 * fsync file X
2182 * unlink file X but leave X.link
2183 * fsync dir Y
2184 *
2185 * After a crash we would expect only X.link to exist. But file X
2186 * didn't get fsync'd again so the log has back refs for X and X.link.
2187 *
2188 * We solve this by removing directory entries and inode backrefs from the
2189 * log when a file that was logged in the current transaction is
2190 * unlinked. Any later fsync will include the updated log entries, and
2191 * we'll be able to reconstruct the proper directory items from backrefs.
2192 *
2193 * This optimizations allows us to avoid relogging the entire inode
2194 * or the entire directory.
2195 */
2196int btrfs_del_dir_entries_in_log(struct btrfs_trans_handle *trans,
2197 struct btrfs_root *root,
2198 const char *name, int name_len,
2199 struct inode *dir, u64 index)
2200{
2201 struct btrfs_root *log;
2202 struct btrfs_dir_item *di;
2203 struct btrfs_path *path;
2204 int ret;
2205 int bytes_del = 0;
2206
Chris Mason3a5f1d42008-09-11 15:53:37 -04002207 if (BTRFS_I(dir)->logged_trans < trans->transid)
2208 return 0;
2209
Chris Masone02119d2008-09-05 16:13:11 -04002210 ret = join_running_log_trans(root);
2211 if (ret)
2212 return 0;
2213
2214 mutex_lock(&BTRFS_I(dir)->log_mutex);
2215
2216 log = root->log_root;
2217 path = btrfs_alloc_path();
2218 di = btrfs_lookup_dir_item(trans, log, path, dir->i_ino,
2219 name, name_len, -1);
2220 if (di && !IS_ERR(di)) {
2221 ret = btrfs_delete_one_dir_name(trans, log, path, di);
2222 bytes_del += name_len;
2223 BUG_ON(ret);
2224 }
2225 btrfs_release_path(log, path);
2226 di = btrfs_lookup_dir_index_item(trans, log, path, dir->i_ino,
2227 index, name, name_len, -1);
2228 if (di && !IS_ERR(di)) {
2229 ret = btrfs_delete_one_dir_name(trans, log, path, di);
2230 bytes_del += name_len;
2231 BUG_ON(ret);
2232 }
2233
2234 /* update the directory size in the log to reflect the names
2235 * we have removed
2236 */
2237 if (bytes_del) {
2238 struct btrfs_key key;
2239
2240 key.objectid = dir->i_ino;
2241 key.offset = 0;
2242 key.type = BTRFS_INODE_ITEM_KEY;
2243 btrfs_release_path(log, path);
2244
2245 ret = btrfs_search_slot(trans, log, &key, path, 0, 1);
2246 if (ret == 0) {
2247 struct btrfs_inode_item *item;
2248 u64 i_size;
2249
2250 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2251 struct btrfs_inode_item);
2252 i_size = btrfs_inode_size(path->nodes[0], item);
2253 if (i_size > bytes_del)
2254 i_size -= bytes_del;
2255 else
2256 i_size = 0;
2257 btrfs_set_inode_size(path->nodes[0], item, i_size);
2258 btrfs_mark_buffer_dirty(path->nodes[0]);
2259 } else
2260 ret = 0;
2261 btrfs_release_path(log, path);
2262 }
2263
2264 btrfs_free_path(path);
2265 mutex_unlock(&BTRFS_I(dir)->log_mutex);
Chris Mason12fcfd22009-03-24 10:24:20 -04002266 btrfs_end_log_trans(root);
Chris Masone02119d2008-09-05 16:13:11 -04002267
2268 return 0;
2269}
2270
2271/* see comments for btrfs_del_dir_entries_in_log */
2272int btrfs_del_inode_ref_in_log(struct btrfs_trans_handle *trans,
2273 struct btrfs_root *root,
2274 const char *name, int name_len,
2275 struct inode *inode, u64 dirid)
2276{
2277 struct btrfs_root *log;
2278 u64 index;
2279 int ret;
2280
Chris Mason3a5f1d42008-09-11 15:53:37 -04002281 if (BTRFS_I(inode)->logged_trans < trans->transid)
2282 return 0;
2283
Chris Masone02119d2008-09-05 16:13:11 -04002284 ret = join_running_log_trans(root);
2285 if (ret)
2286 return 0;
2287 log = root->log_root;
2288 mutex_lock(&BTRFS_I(inode)->log_mutex);
2289
2290 ret = btrfs_del_inode_ref(trans, log, name, name_len, inode->i_ino,
2291 dirid, &index);
2292 mutex_unlock(&BTRFS_I(inode)->log_mutex);
Chris Mason12fcfd22009-03-24 10:24:20 -04002293 btrfs_end_log_trans(root);
Chris Masone02119d2008-09-05 16:13:11 -04002294
Chris Masone02119d2008-09-05 16:13:11 -04002295 return ret;
2296}
2297
2298/*
2299 * creates a range item in the log for 'dirid'. first_offset and
2300 * last_offset tell us which parts of the key space the log should
2301 * be considered authoritative for.
2302 */
2303static noinline int insert_dir_log_key(struct btrfs_trans_handle *trans,
2304 struct btrfs_root *log,
2305 struct btrfs_path *path,
2306 int key_type, u64 dirid,
2307 u64 first_offset, u64 last_offset)
2308{
2309 int ret;
2310 struct btrfs_key key;
2311 struct btrfs_dir_log_item *item;
2312
2313 key.objectid = dirid;
2314 key.offset = first_offset;
2315 if (key_type == BTRFS_DIR_ITEM_KEY)
2316 key.type = BTRFS_DIR_LOG_ITEM_KEY;
2317 else
2318 key.type = BTRFS_DIR_LOG_INDEX_KEY;
2319 ret = btrfs_insert_empty_item(trans, log, path, &key, sizeof(*item));
2320 BUG_ON(ret);
2321
2322 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2323 struct btrfs_dir_log_item);
2324 btrfs_set_dir_log_end(path->nodes[0], item, last_offset);
2325 btrfs_mark_buffer_dirty(path->nodes[0]);
2326 btrfs_release_path(log, path);
2327 return 0;
2328}
2329
2330/*
2331 * log all the items included in the current transaction for a given
2332 * directory. This also creates the range items in the log tree required
2333 * to replay anything deleted before the fsync
2334 */
2335static noinline int log_dir_items(struct btrfs_trans_handle *trans,
2336 struct btrfs_root *root, struct inode *inode,
2337 struct btrfs_path *path,
2338 struct btrfs_path *dst_path, int key_type,
2339 u64 min_offset, u64 *last_offset_ret)
2340{
2341 struct btrfs_key min_key;
2342 struct btrfs_key max_key;
2343 struct btrfs_root *log = root->log_root;
2344 struct extent_buffer *src;
2345 int ret;
2346 int i;
2347 int nritems;
2348 u64 first_offset = min_offset;
2349 u64 last_offset = (u64)-1;
2350
2351 log = root->log_root;
2352 max_key.objectid = inode->i_ino;
2353 max_key.offset = (u64)-1;
2354 max_key.type = key_type;
2355
2356 min_key.objectid = inode->i_ino;
2357 min_key.type = key_type;
2358 min_key.offset = min_offset;
2359
2360 path->keep_locks = 1;
2361
2362 ret = btrfs_search_forward(root, &min_key, &max_key,
2363 path, 0, trans->transid);
2364
2365 /*
2366 * we didn't find anything from this transaction, see if there
2367 * is anything at all
2368 */
2369 if (ret != 0 || min_key.objectid != inode->i_ino ||
2370 min_key.type != key_type) {
2371 min_key.objectid = inode->i_ino;
2372 min_key.type = key_type;
2373 min_key.offset = (u64)-1;
2374 btrfs_release_path(root, path);
2375 ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0);
2376 if (ret < 0) {
2377 btrfs_release_path(root, path);
2378 return ret;
2379 }
2380 ret = btrfs_previous_item(root, path, inode->i_ino, key_type);
2381
2382 /* if ret == 0 there are items for this type,
2383 * create a range to tell us the last key of this type.
2384 * otherwise, there are no items in this directory after
2385 * *min_offset, and we create a range to indicate that.
2386 */
2387 if (ret == 0) {
2388 struct btrfs_key tmp;
2389 btrfs_item_key_to_cpu(path->nodes[0], &tmp,
2390 path->slots[0]);
Chris Masond3977122009-01-05 21:25:51 -05002391 if (key_type == tmp.type)
Chris Masone02119d2008-09-05 16:13:11 -04002392 first_offset = max(min_offset, tmp.offset) + 1;
Chris Masone02119d2008-09-05 16:13:11 -04002393 }
2394 goto done;
2395 }
2396
2397 /* go backward to find any previous key */
2398 ret = btrfs_previous_item(root, path, inode->i_ino, key_type);
2399 if (ret == 0) {
2400 struct btrfs_key tmp;
2401 btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]);
2402 if (key_type == tmp.type) {
2403 first_offset = tmp.offset;
2404 ret = overwrite_item(trans, log, dst_path,
2405 path->nodes[0], path->slots[0],
2406 &tmp);
2407 }
2408 }
2409 btrfs_release_path(root, path);
2410
2411 /* find the first key from this transaction again */
2412 ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0);
2413 if (ret != 0) {
2414 WARN_ON(1);
2415 goto done;
2416 }
2417
2418 /*
2419 * we have a block from this transaction, log every item in it
2420 * from our directory
2421 */
Chris Masond3977122009-01-05 21:25:51 -05002422 while (1) {
Chris Masone02119d2008-09-05 16:13:11 -04002423 struct btrfs_key tmp;
2424 src = path->nodes[0];
2425 nritems = btrfs_header_nritems(src);
2426 for (i = path->slots[0]; i < nritems; i++) {
2427 btrfs_item_key_to_cpu(src, &min_key, i);
2428
2429 if (min_key.objectid != inode->i_ino ||
2430 min_key.type != key_type)
2431 goto done;
2432 ret = overwrite_item(trans, log, dst_path, src, i,
2433 &min_key);
2434 BUG_ON(ret);
2435 }
2436 path->slots[0] = nritems;
2437
2438 /*
2439 * look ahead to the next item and see if it is also
2440 * from this directory and from this transaction
2441 */
2442 ret = btrfs_next_leaf(root, path);
2443 if (ret == 1) {
2444 last_offset = (u64)-1;
2445 goto done;
2446 }
2447 btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]);
2448 if (tmp.objectid != inode->i_ino || tmp.type != key_type) {
2449 last_offset = (u64)-1;
2450 goto done;
2451 }
2452 if (btrfs_header_generation(path->nodes[0]) != trans->transid) {
2453 ret = overwrite_item(trans, log, dst_path,
2454 path->nodes[0], path->slots[0],
2455 &tmp);
2456
2457 BUG_ON(ret);
2458 last_offset = tmp.offset;
2459 goto done;
2460 }
2461 }
2462done:
2463 *last_offset_ret = last_offset;
2464 btrfs_release_path(root, path);
2465 btrfs_release_path(log, dst_path);
2466
2467 /* insert the log range keys to indicate where the log is valid */
2468 ret = insert_dir_log_key(trans, log, path, key_type, inode->i_ino,
2469 first_offset, last_offset);
2470 BUG_ON(ret);
2471 return 0;
2472}
2473
2474/*
2475 * logging directories is very similar to logging inodes, We find all the items
2476 * from the current transaction and write them to the log.
2477 *
2478 * The recovery code scans the directory in the subvolume, and if it finds a
2479 * key in the range logged that is not present in the log tree, then it means
2480 * that dir entry was unlinked during the transaction.
2481 *
2482 * In order for that scan to work, we must include one key smaller than
2483 * the smallest logged by this transaction and one key larger than the largest
2484 * key logged by this transaction.
2485 */
2486static noinline int log_directory_changes(struct btrfs_trans_handle *trans,
2487 struct btrfs_root *root, struct inode *inode,
2488 struct btrfs_path *path,
2489 struct btrfs_path *dst_path)
2490{
2491 u64 min_key;
2492 u64 max_key;
2493 int ret;
2494 int key_type = BTRFS_DIR_ITEM_KEY;
2495
2496again:
2497 min_key = 0;
2498 max_key = 0;
Chris Masond3977122009-01-05 21:25:51 -05002499 while (1) {
Chris Masone02119d2008-09-05 16:13:11 -04002500 ret = log_dir_items(trans, root, inode, path,
2501 dst_path, key_type, min_key,
2502 &max_key);
2503 BUG_ON(ret);
2504 if (max_key == (u64)-1)
2505 break;
2506 min_key = max_key + 1;
2507 }
2508
2509 if (key_type == BTRFS_DIR_ITEM_KEY) {
2510 key_type = BTRFS_DIR_INDEX_KEY;
2511 goto again;
2512 }
2513 return 0;
2514}
2515
2516/*
2517 * a helper function to drop items from the log before we relog an
2518 * inode. max_key_type indicates the highest item type to remove.
2519 * This cannot be run for file data extents because it does not
2520 * free the extents they point to.
2521 */
2522static int drop_objectid_items(struct btrfs_trans_handle *trans,
2523 struct btrfs_root *log,
2524 struct btrfs_path *path,
2525 u64 objectid, int max_key_type)
2526{
2527 int ret;
2528 struct btrfs_key key;
2529 struct btrfs_key found_key;
2530
2531 key.objectid = objectid;
2532 key.type = max_key_type;
2533 key.offset = (u64)-1;
2534
Chris Masond3977122009-01-05 21:25:51 -05002535 while (1) {
Chris Masone02119d2008-09-05 16:13:11 -04002536 ret = btrfs_search_slot(trans, log, &key, path, -1, 1);
2537
2538 if (ret != 1)
2539 break;
2540
2541 if (path->slots[0] == 0)
2542 break;
2543
2544 path->slots[0]--;
2545 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2546 path->slots[0]);
2547
2548 if (found_key.objectid != objectid)
2549 break;
2550
2551 ret = btrfs_del_item(trans, log, path);
2552 BUG_ON(ret);
2553 btrfs_release_path(log, path);
2554 }
2555 btrfs_release_path(log, path);
2556 return 0;
2557}
2558
Chris Mason31ff1cd2008-09-11 16:17:57 -04002559static noinline int copy_items(struct btrfs_trans_handle *trans,
2560 struct btrfs_root *log,
2561 struct btrfs_path *dst_path,
2562 struct extent_buffer *src,
2563 int start_slot, int nr, int inode_only)
2564{
2565 unsigned long src_offset;
2566 unsigned long dst_offset;
2567 struct btrfs_file_extent_item *extent;
2568 struct btrfs_inode_item *inode_item;
2569 int ret;
2570 struct btrfs_key *ins_keys;
2571 u32 *ins_sizes;
2572 char *ins_data;
2573 int i;
Chris Masond20f7042008-12-08 16:58:54 -05002574 struct list_head ordered_sums;
2575
2576 INIT_LIST_HEAD(&ordered_sums);
Chris Mason31ff1cd2008-09-11 16:17:57 -04002577
2578 ins_data = kmalloc(nr * sizeof(struct btrfs_key) +
2579 nr * sizeof(u32), GFP_NOFS);
2580 ins_sizes = (u32 *)ins_data;
2581 ins_keys = (struct btrfs_key *)(ins_data + nr * sizeof(u32));
2582
2583 for (i = 0; i < nr; i++) {
2584 ins_sizes[i] = btrfs_item_size_nr(src, i + start_slot);
2585 btrfs_item_key_to_cpu(src, ins_keys + i, i + start_slot);
2586 }
2587 ret = btrfs_insert_empty_items(trans, log, dst_path,
2588 ins_keys, ins_sizes, nr);
2589 BUG_ON(ret);
2590
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002591 for (i = 0; i < nr; i++, dst_path->slots[0]++) {
Chris Mason31ff1cd2008-09-11 16:17:57 -04002592 dst_offset = btrfs_item_ptr_offset(dst_path->nodes[0],
2593 dst_path->slots[0]);
2594
2595 src_offset = btrfs_item_ptr_offset(src, start_slot + i);
2596
2597 copy_extent_buffer(dst_path->nodes[0], src, dst_offset,
2598 src_offset, ins_sizes[i]);
2599
2600 if (inode_only == LOG_INODE_EXISTS &&
2601 ins_keys[i].type == BTRFS_INODE_ITEM_KEY) {
2602 inode_item = btrfs_item_ptr(dst_path->nodes[0],
2603 dst_path->slots[0],
2604 struct btrfs_inode_item);
2605 btrfs_set_inode_size(dst_path->nodes[0], inode_item, 0);
2606
2607 /* set the generation to zero so the recover code
2608 * can tell the difference between an logging
2609 * just to say 'this inode exists' and a logging
2610 * to say 'update this inode with these values'
2611 */
2612 btrfs_set_inode_generation(dst_path->nodes[0],
2613 inode_item, 0);
2614 }
2615 /* take a reference on file data extents so that truncates
2616 * or deletes of this inode don't have to relog the inode
2617 * again
2618 */
2619 if (btrfs_key_type(ins_keys + i) == BTRFS_EXTENT_DATA_KEY) {
2620 int found_type;
2621 extent = btrfs_item_ptr(src, start_slot + i,
2622 struct btrfs_file_extent_item);
2623
2624 found_type = btrfs_file_extent_type(src, extent);
Yan Zhengd899e052008-10-30 14:25:28 -04002625 if (found_type == BTRFS_FILE_EXTENT_REG ||
2626 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002627 u64 ds, dl, cs, cl;
2628 ds = btrfs_file_extent_disk_bytenr(src,
2629 extent);
2630 /* ds == 0 is a hole */
2631 if (ds == 0)
2632 continue;
2633
2634 dl = btrfs_file_extent_disk_num_bytes(src,
2635 extent);
2636 cs = btrfs_file_extent_offset(src, extent);
2637 cl = btrfs_file_extent_num_bytes(src,
Joe Perchesa419aef2009-08-18 11:18:35 -07002638 extent);
Chris Mason580afd72008-12-08 19:15:39 -05002639 if (btrfs_file_extent_compression(src,
2640 extent)) {
2641 cs = 0;
2642 cl = dl;
2643 }
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002644
2645 ret = btrfs_lookup_csums_range(
2646 log->fs_info->csum_root,
2647 ds + cs, ds + cs + cl - 1,
2648 &ordered_sums);
2649 BUG_ON(ret);
Chris Mason31ff1cd2008-09-11 16:17:57 -04002650 }
2651 }
Chris Mason31ff1cd2008-09-11 16:17:57 -04002652 }
2653
2654 btrfs_mark_buffer_dirty(dst_path->nodes[0]);
2655 btrfs_release_path(log, dst_path);
2656 kfree(ins_data);
Chris Masond20f7042008-12-08 16:58:54 -05002657
2658 /*
2659 * we have to do this after the loop above to avoid changing the
2660 * log tree while trying to change the log tree.
2661 */
Chris Masond3977122009-01-05 21:25:51 -05002662 while (!list_empty(&ordered_sums)) {
Chris Masond20f7042008-12-08 16:58:54 -05002663 struct btrfs_ordered_sum *sums = list_entry(ordered_sums.next,
2664 struct btrfs_ordered_sum,
2665 list);
2666 ret = btrfs_csum_file_blocks(trans, log, sums);
2667 BUG_ON(ret);
2668 list_del(&sums->list);
2669 kfree(sums);
2670 }
Chris Mason31ff1cd2008-09-11 16:17:57 -04002671 return 0;
2672}
2673
Chris Masone02119d2008-09-05 16:13:11 -04002674/* log a single inode in the tree log.
2675 * At least one parent directory for this inode must exist in the tree
2676 * or be logged already.
2677 *
2678 * Any items from this inode changed by the current transaction are copied
2679 * to the log tree. An extra reference is taken on any extents in this
2680 * file, allowing us to avoid a whole pile of corner cases around logging
2681 * blocks that have been removed from the tree.
2682 *
2683 * See LOG_INODE_ALL and related defines for a description of what inode_only
2684 * does.
2685 *
2686 * This handles both files and directories.
2687 */
Chris Mason12fcfd22009-03-24 10:24:20 -04002688static int btrfs_log_inode(struct btrfs_trans_handle *trans,
Chris Masone02119d2008-09-05 16:13:11 -04002689 struct btrfs_root *root, struct inode *inode,
2690 int inode_only)
2691{
2692 struct btrfs_path *path;
2693 struct btrfs_path *dst_path;
2694 struct btrfs_key min_key;
2695 struct btrfs_key max_key;
2696 struct btrfs_root *log = root->log_root;
Chris Mason31ff1cd2008-09-11 16:17:57 -04002697 struct extent_buffer *src = NULL;
Chris Masone02119d2008-09-05 16:13:11 -04002698 u32 size;
2699 int ret;
Chris Mason3a5f1d42008-09-11 15:53:37 -04002700 int nritems;
Chris Mason31ff1cd2008-09-11 16:17:57 -04002701 int ins_start_slot = 0;
2702 int ins_nr;
Chris Masone02119d2008-09-05 16:13:11 -04002703
2704 log = root->log_root;
2705
2706 path = btrfs_alloc_path();
2707 dst_path = btrfs_alloc_path();
2708
2709 min_key.objectid = inode->i_ino;
2710 min_key.type = BTRFS_INODE_ITEM_KEY;
2711 min_key.offset = 0;
2712
2713 max_key.objectid = inode->i_ino;
Chris Mason12fcfd22009-03-24 10:24:20 -04002714
2715 /* today the code can only do partial logging of directories */
2716 if (!S_ISDIR(inode->i_mode))
2717 inode_only = LOG_INODE_ALL;
2718
Chris Masone02119d2008-09-05 16:13:11 -04002719 if (inode_only == LOG_INODE_EXISTS || S_ISDIR(inode->i_mode))
2720 max_key.type = BTRFS_XATTR_ITEM_KEY;
2721 else
2722 max_key.type = (u8)-1;
2723 max_key.offset = (u64)-1;
2724
Chris Masone02119d2008-09-05 16:13:11 -04002725 mutex_lock(&BTRFS_I(inode)->log_mutex);
2726
2727 /*
2728 * a brute force approach to making sure we get the most uptodate
2729 * copies of everything.
2730 */
2731 if (S_ISDIR(inode->i_mode)) {
2732 int max_key_type = BTRFS_DIR_LOG_INDEX_KEY;
2733
2734 if (inode_only == LOG_INODE_EXISTS)
2735 max_key_type = BTRFS_XATTR_ITEM_KEY;
2736 ret = drop_objectid_items(trans, log, path,
2737 inode->i_ino, max_key_type);
2738 } else {
2739 ret = btrfs_truncate_inode_items(trans, log, inode, 0, 0);
2740 }
2741 BUG_ON(ret);
2742 path->keep_locks = 1;
2743
Chris Masond3977122009-01-05 21:25:51 -05002744 while (1) {
Chris Mason31ff1cd2008-09-11 16:17:57 -04002745 ins_nr = 0;
Chris Masone02119d2008-09-05 16:13:11 -04002746 ret = btrfs_search_forward(root, &min_key, &max_key,
2747 path, 0, trans->transid);
2748 if (ret != 0)
2749 break;
Chris Mason3a5f1d42008-09-11 15:53:37 -04002750again:
Chris Mason31ff1cd2008-09-11 16:17:57 -04002751 /* note, ins_nr might be > 0 here, cleanup outside the loop */
Chris Masone02119d2008-09-05 16:13:11 -04002752 if (min_key.objectid != inode->i_ino)
2753 break;
2754 if (min_key.type > max_key.type)
2755 break;
Chris Mason31ff1cd2008-09-11 16:17:57 -04002756
Chris Masone02119d2008-09-05 16:13:11 -04002757 src = path->nodes[0];
2758 size = btrfs_item_size_nr(src, path->slots[0]);
Chris Mason31ff1cd2008-09-11 16:17:57 -04002759 if (ins_nr && ins_start_slot + ins_nr == path->slots[0]) {
2760 ins_nr++;
2761 goto next_slot;
2762 } else if (!ins_nr) {
2763 ins_start_slot = path->slots[0];
2764 ins_nr = 1;
2765 goto next_slot;
Chris Masone02119d2008-09-05 16:13:11 -04002766 }
2767
Chris Mason31ff1cd2008-09-11 16:17:57 -04002768 ret = copy_items(trans, log, dst_path, src, ins_start_slot,
2769 ins_nr, inode_only);
2770 BUG_ON(ret);
2771 ins_nr = 1;
2772 ins_start_slot = path->slots[0];
2773next_slot:
Chris Masone02119d2008-09-05 16:13:11 -04002774
Chris Mason3a5f1d42008-09-11 15:53:37 -04002775 nritems = btrfs_header_nritems(path->nodes[0]);
2776 path->slots[0]++;
2777 if (path->slots[0] < nritems) {
2778 btrfs_item_key_to_cpu(path->nodes[0], &min_key,
2779 path->slots[0]);
2780 goto again;
2781 }
Chris Mason31ff1cd2008-09-11 16:17:57 -04002782 if (ins_nr) {
2783 ret = copy_items(trans, log, dst_path, src,
2784 ins_start_slot,
2785 ins_nr, inode_only);
2786 BUG_ON(ret);
2787 ins_nr = 0;
2788 }
Chris Mason3a5f1d42008-09-11 15:53:37 -04002789 btrfs_release_path(root, path);
2790
Chris Masone02119d2008-09-05 16:13:11 -04002791 if (min_key.offset < (u64)-1)
2792 min_key.offset++;
2793 else if (min_key.type < (u8)-1)
2794 min_key.type++;
2795 else if (min_key.objectid < (u64)-1)
2796 min_key.objectid++;
2797 else
2798 break;
2799 }
Chris Mason31ff1cd2008-09-11 16:17:57 -04002800 if (ins_nr) {
2801 ret = copy_items(trans, log, dst_path, src,
2802 ins_start_slot,
2803 ins_nr, inode_only);
2804 BUG_ON(ret);
2805 ins_nr = 0;
2806 }
2807 WARN_ON(ins_nr);
Chris Mason9623f9a2008-09-11 17:42:42 -04002808 if (inode_only == LOG_INODE_ALL && S_ISDIR(inode->i_mode)) {
Chris Masone02119d2008-09-05 16:13:11 -04002809 btrfs_release_path(root, path);
2810 btrfs_release_path(log, dst_path);
2811 ret = log_directory_changes(trans, root, inode, path, dst_path);
2812 BUG_ON(ret);
2813 }
Chris Mason3a5f1d42008-09-11 15:53:37 -04002814 BTRFS_I(inode)->logged_trans = trans->transid;
Chris Masone02119d2008-09-05 16:13:11 -04002815 mutex_unlock(&BTRFS_I(inode)->log_mutex);
2816
2817 btrfs_free_path(path);
2818 btrfs_free_path(dst_path);
Chris Masone02119d2008-09-05 16:13:11 -04002819 return 0;
2820}
2821
Chris Mason12fcfd22009-03-24 10:24:20 -04002822/*
2823 * follow the dentry parent pointers up the chain and see if any
2824 * of the directories in it require a full commit before they can
2825 * be logged. Returns zero if nothing special needs to be done or 1 if
2826 * a full commit is required.
2827 */
2828static noinline int check_parent_dirs_for_sync(struct btrfs_trans_handle *trans,
2829 struct inode *inode,
2830 struct dentry *parent,
2831 struct super_block *sb,
2832 u64 last_committed)
Chris Masone02119d2008-09-05 16:13:11 -04002833{
Chris Mason12fcfd22009-03-24 10:24:20 -04002834 int ret = 0;
2835 struct btrfs_root *root;
Chris Masone02119d2008-09-05 16:13:11 -04002836
Chris Masonaf4176b2009-03-24 10:24:31 -04002837 /*
2838 * for regular files, if its inode is already on disk, we don't
2839 * have to worry about the parents at all. This is because
2840 * we can use the last_unlink_trans field to record renames
2841 * and other fun in this file.
2842 */
2843 if (S_ISREG(inode->i_mode) &&
2844 BTRFS_I(inode)->generation <= last_committed &&
2845 BTRFS_I(inode)->last_unlink_trans <= last_committed)
2846 goto out;
2847
Chris Mason12fcfd22009-03-24 10:24:20 -04002848 if (!S_ISDIR(inode->i_mode)) {
2849 if (!parent || !parent->d_inode || sb != parent->d_inode->i_sb)
2850 goto out;
2851 inode = parent->d_inode;
2852 }
2853
2854 while (1) {
2855 BTRFS_I(inode)->logged_trans = trans->transid;
2856 smp_mb();
2857
2858 if (BTRFS_I(inode)->last_unlink_trans > last_committed) {
2859 root = BTRFS_I(inode)->root;
2860
2861 /*
2862 * make sure any commits to the log are forced
2863 * to be full commits
2864 */
2865 root->fs_info->last_trans_log_full_commit =
2866 trans->transid;
2867 ret = 1;
2868 break;
2869 }
2870
2871 if (!parent || !parent->d_inode || sb != parent->d_inode->i_sb)
2872 break;
2873
Yan, Zheng76dda932009-09-21 16:00:26 -04002874 if (IS_ROOT(parent))
Chris Mason12fcfd22009-03-24 10:24:20 -04002875 break;
2876
2877 parent = parent->d_parent;
2878 inode = parent->d_inode;
2879
2880 }
2881out:
Chris Masone02119d2008-09-05 16:13:11 -04002882 return ret;
2883}
2884
Chris Mason257c62e2009-10-13 13:21:08 -04002885static int inode_in_log(struct btrfs_trans_handle *trans,
2886 struct inode *inode)
2887{
2888 struct btrfs_root *root = BTRFS_I(inode)->root;
2889 int ret = 0;
2890
2891 mutex_lock(&root->log_mutex);
2892 if (BTRFS_I(inode)->logged_trans == trans->transid &&
2893 BTRFS_I(inode)->last_sub_trans <= root->last_log_commit)
2894 ret = 1;
2895 mutex_unlock(&root->log_mutex);
2896 return ret;
2897}
2898
2899
Chris Masone02119d2008-09-05 16:13:11 -04002900/*
2901 * helper function around btrfs_log_inode to make sure newly created
2902 * parent directories also end up in the log. A minimal inode and backref
2903 * only logging is done of any parent directories that are older than
2904 * the last committed transaction
2905 */
Chris Mason12fcfd22009-03-24 10:24:20 -04002906int btrfs_log_inode_parent(struct btrfs_trans_handle *trans,
2907 struct btrfs_root *root, struct inode *inode,
2908 struct dentry *parent, int exists_only)
Chris Masone02119d2008-09-05 16:13:11 -04002909{
Chris Mason12fcfd22009-03-24 10:24:20 -04002910 int inode_only = exists_only ? LOG_INODE_EXISTS : LOG_INODE_ALL;
Chris Masone02119d2008-09-05 16:13:11 -04002911 struct super_block *sb;
Chris Mason12fcfd22009-03-24 10:24:20 -04002912 int ret = 0;
2913 u64 last_committed = root->fs_info->last_trans_committed;
2914
2915 sb = inode->i_sb;
2916
Sage Weil3a5e1402009-04-02 16:49:40 -04002917 if (btrfs_test_opt(root, NOTREELOG)) {
2918 ret = 1;
2919 goto end_no_trans;
2920 }
2921
Chris Mason12fcfd22009-03-24 10:24:20 -04002922 if (root->fs_info->last_trans_log_full_commit >
2923 root->fs_info->last_trans_committed) {
2924 ret = 1;
2925 goto end_no_trans;
2926 }
2927
Yan, Zheng76dda932009-09-21 16:00:26 -04002928 if (root != BTRFS_I(inode)->root ||
2929 btrfs_root_refs(&root->root_item) == 0) {
2930 ret = 1;
2931 goto end_no_trans;
2932 }
2933
Chris Mason12fcfd22009-03-24 10:24:20 -04002934 ret = check_parent_dirs_for_sync(trans, inode, parent,
2935 sb, last_committed);
2936 if (ret)
2937 goto end_no_trans;
Chris Masone02119d2008-09-05 16:13:11 -04002938
Chris Mason257c62e2009-10-13 13:21:08 -04002939 if (inode_in_log(trans, inode)) {
2940 ret = BTRFS_NO_LOG_SYNC;
2941 goto end_no_trans;
2942 }
2943
Chris Masone02119d2008-09-05 16:13:11 -04002944 start_log_trans(trans, root);
Chris Mason12fcfd22009-03-24 10:24:20 -04002945
2946 ret = btrfs_log_inode(trans, root, inode, inode_only);
2947 BUG_ON(ret);
Chris Mason12fcfd22009-03-24 10:24:20 -04002948
Chris Masonaf4176b2009-03-24 10:24:31 -04002949 /*
2950 * for regular files, if its inode is already on disk, we don't
2951 * have to worry about the parents at all. This is because
2952 * we can use the last_unlink_trans field to record renames
2953 * and other fun in this file.
2954 */
2955 if (S_ISREG(inode->i_mode) &&
2956 BTRFS_I(inode)->generation <= last_committed &&
2957 BTRFS_I(inode)->last_unlink_trans <= last_committed)
2958 goto no_parent;
2959
2960 inode_only = LOG_INODE_EXISTS;
Chris Masond3977122009-01-05 21:25:51 -05002961 while (1) {
Chris Mason12fcfd22009-03-24 10:24:20 -04002962 if (!parent || !parent->d_inode || sb != parent->d_inode->i_sb)
Chris Masone02119d2008-09-05 16:13:11 -04002963 break;
2964
Chris Mason12fcfd22009-03-24 10:24:20 -04002965 inode = parent->d_inode;
Yan, Zheng76dda932009-09-21 16:00:26 -04002966 if (root != BTRFS_I(inode)->root)
2967 break;
2968
Chris Mason12fcfd22009-03-24 10:24:20 -04002969 if (BTRFS_I(inode)->generation >
2970 root->fs_info->last_trans_committed) {
2971 ret = btrfs_log_inode(trans, root, inode, inode_only);
2972 BUG_ON(ret);
2973 }
Yan, Zheng76dda932009-09-21 16:00:26 -04002974 if (IS_ROOT(parent))
Chris Masone02119d2008-09-05 16:13:11 -04002975 break;
Chris Mason12fcfd22009-03-24 10:24:20 -04002976
2977 parent = parent->d_parent;
Chris Masone02119d2008-09-05 16:13:11 -04002978 }
Chris Masonaf4176b2009-03-24 10:24:31 -04002979no_parent:
Chris Mason12fcfd22009-03-24 10:24:20 -04002980 ret = 0;
2981 btrfs_end_log_trans(root);
2982end_no_trans:
2983 return ret;
Chris Masone02119d2008-09-05 16:13:11 -04002984}
2985
2986/*
2987 * it is not safe to log dentry if the chunk root has added new
2988 * chunks. This returns 0 if the dentry was logged, and 1 otherwise.
2989 * If this returns 1, you must commit the transaction to safely get your
2990 * data on disk.
2991 */
2992int btrfs_log_dentry_safe(struct btrfs_trans_handle *trans,
2993 struct btrfs_root *root, struct dentry *dentry)
2994{
Chris Mason12fcfd22009-03-24 10:24:20 -04002995 return btrfs_log_inode_parent(trans, root, dentry->d_inode,
2996 dentry->d_parent, 0);
Chris Masone02119d2008-09-05 16:13:11 -04002997}
2998
2999/*
3000 * should be called during mount to recover any replay any log trees
3001 * from the FS
3002 */
3003int btrfs_recover_log_trees(struct btrfs_root *log_root_tree)
3004{
3005 int ret;
3006 struct btrfs_path *path;
3007 struct btrfs_trans_handle *trans;
3008 struct btrfs_key key;
3009 struct btrfs_key found_key;
3010 struct btrfs_key tmp_key;
3011 struct btrfs_root *log;
3012 struct btrfs_fs_info *fs_info = log_root_tree->fs_info;
3013 struct walk_control wc = {
3014 .process_func = process_one_buffer,
3015 .stage = 0,
3016 };
3017
3018 fs_info->log_root_recovering = 1;
3019 path = btrfs_alloc_path();
3020 BUG_ON(!path);
3021
3022 trans = btrfs_start_transaction(fs_info->tree_root, 1);
3023
3024 wc.trans = trans;
3025 wc.pin = 1;
3026
3027 walk_log_tree(trans, log_root_tree, &wc);
3028
3029again:
3030 key.objectid = BTRFS_TREE_LOG_OBJECTID;
3031 key.offset = (u64)-1;
3032 btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
3033
Chris Masond3977122009-01-05 21:25:51 -05003034 while (1) {
Chris Masone02119d2008-09-05 16:13:11 -04003035 ret = btrfs_search_slot(NULL, log_root_tree, &key, path, 0, 0);
3036 if (ret < 0)
3037 break;
3038 if (ret > 0) {
3039 if (path->slots[0] == 0)
3040 break;
3041 path->slots[0]--;
3042 }
3043 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
3044 path->slots[0]);
3045 btrfs_release_path(log_root_tree, path);
3046 if (found_key.objectid != BTRFS_TREE_LOG_OBJECTID)
3047 break;
3048
3049 log = btrfs_read_fs_root_no_radix(log_root_tree,
3050 &found_key);
3051 BUG_ON(!log);
3052
3053
3054 tmp_key.objectid = found_key.offset;
3055 tmp_key.type = BTRFS_ROOT_ITEM_KEY;
3056 tmp_key.offset = (u64)-1;
3057
3058 wc.replay_dest = btrfs_read_fs_root_no_name(fs_info, &tmp_key);
Chris Masone02119d2008-09-05 16:13:11 -04003059 BUG_ON(!wc.replay_dest);
3060
Yan Zheng07d400a2009-01-06 11:42:00 -05003061 wc.replay_dest->log_root = log;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003062 btrfs_record_root_in_trans(trans, wc.replay_dest);
Chris Masone02119d2008-09-05 16:13:11 -04003063 ret = walk_log_tree(trans, log, &wc);
3064 BUG_ON(ret);
3065
3066 if (wc.stage == LOG_WALK_REPLAY_ALL) {
3067 ret = fixup_inode_link_counts(trans, wc.replay_dest,
3068 path);
3069 BUG_ON(ret);
3070 }
Chris Masone02119d2008-09-05 16:13:11 -04003071
3072 key.offset = found_key.offset - 1;
Yan Zheng07d400a2009-01-06 11:42:00 -05003073 wc.replay_dest->log_root = NULL;
Chris Masone02119d2008-09-05 16:13:11 -04003074 free_extent_buffer(log->node);
Chris Masonb263c2c2009-06-11 11:24:47 -04003075 free_extent_buffer(log->commit_root);
Chris Masone02119d2008-09-05 16:13:11 -04003076 kfree(log);
3077
3078 if (found_key.offset == 0)
3079 break;
3080 }
3081 btrfs_release_path(log_root_tree, path);
3082
3083 /* step one is to pin it all, step two is to replay just inodes */
3084 if (wc.pin) {
3085 wc.pin = 0;
3086 wc.process_func = replay_one_buffer;
3087 wc.stage = LOG_WALK_REPLAY_INODES;
3088 goto again;
3089 }
3090 /* step three is to replay everything */
3091 if (wc.stage < LOG_WALK_REPLAY_ALL) {
3092 wc.stage++;
3093 goto again;
3094 }
3095
3096 btrfs_free_path(path);
3097
3098 free_extent_buffer(log_root_tree->node);
3099 log_root_tree->log_root = NULL;
3100 fs_info->log_root_recovering = 0;
3101
3102 /* step 4: commit the transaction, which also unpins the blocks */
3103 btrfs_commit_transaction(trans, fs_info->tree_root);
3104
3105 kfree(log_root_tree);
3106 return 0;
3107}
Chris Mason12fcfd22009-03-24 10:24:20 -04003108
3109/*
3110 * there are some corner cases where we want to force a full
3111 * commit instead of allowing a directory to be logged.
3112 *
3113 * They revolve around files there were unlinked from the directory, and
3114 * this function updates the parent directory so that a full commit is
3115 * properly done if it is fsync'd later after the unlinks are done.
3116 */
3117void btrfs_record_unlink_dir(struct btrfs_trans_handle *trans,
3118 struct inode *dir, struct inode *inode,
3119 int for_rename)
3120{
3121 /*
Chris Masonaf4176b2009-03-24 10:24:31 -04003122 * when we're logging a file, if it hasn't been renamed
3123 * or unlinked, and its inode is fully committed on disk,
3124 * we don't have to worry about walking up the directory chain
3125 * to log its parents.
3126 *
3127 * So, we use the last_unlink_trans field to put this transid
3128 * into the file. When the file is logged we check it and
3129 * don't log the parents if the file is fully on disk.
3130 */
3131 if (S_ISREG(inode->i_mode))
3132 BTRFS_I(inode)->last_unlink_trans = trans->transid;
3133
3134 /*
Chris Mason12fcfd22009-03-24 10:24:20 -04003135 * if this directory was already logged any new
3136 * names for this file/dir will get recorded
3137 */
3138 smp_mb();
3139 if (BTRFS_I(dir)->logged_trans == trans->transid)
3140 return;
3141
3142 /*
3143 * if the inode we're about to unlink was logged,
3144 * the log will be properly updated for any new names
3145 */
3146 if (BTRFS_I(inode)->logged_trans == trans->transid)
3147 return;
3148
3149 /*
3150 * when renaming files across directories, if the directory
3151 * there we're unlinking from gets fsync'd later on, there's
3152 * no way to find the destination directory later and fsync it
3153 * properly. So, we have to be conservative and force commits
3154 * so the new name gets discovered.
3155 */
3156 if (for_rename)
3157 goto record;
3158
3159 /* we can safely do the unlink without any special recording */
3160 return;
3161
3162record:
3163 BTRFS_I(dir)->last_unlink_trans = trans->transid;
3164}
3165
3166/*
3167 * Call this after adding a new name for a file and it will properly
3168 * update the log to reflect the new name.
3169 *
3170 * It will return zero if all goes well, and it will return 1 if a
3171 * full transaction commit is required.
3172 */
3173int btrfs_log_new_name(struct btrfs_trans_handle *trans,
3174 struct inode *inode, struct inode *old_dir,
3175 struct dentry *parent)
3176{
3177 struct btrfs_root * root = BTRFS_I(inode)->root;
3178
3179 /*
Chris Masonaf4176b2009-03-24 10:24:31 -04003180 * this will force the logging code to walk the dentry chain
3181 * up for the file
3182 */
3183 if (S_ISREG(inode->i_mode))
3184 BTRFS_I(inode)->last_unlink_trans = trans->transid;
3185
3186 /*
Chris Mason12fcfd22009-03-24 10:24:20 -04003187 * if this inode hasn't been logged and directory we're renaming it
3188 * from hasn't been logged, we don't need to log it
3189 */
3190 if (BTRFS_I(inode)->logged_trans <=
3191 root->fs_info->last_trans_committed &&
3192 (!old_dir || BTRFS_I(old_dir)->logged_trans <=
3193 root->fs_info->last_trans_committed))
3194 return 0;
3195
3196 return btrfs_log_inode_parent(trans, root, inode, parent, 1);
3197}
3198