blob: 39d83da03e0398db90428e52b98524982da469b5 [file] [log] [blame]
Chris Masone02119d2008-09-05 16:13:11 -04001/*
2 * Copyright (C) 2008 Oracle. All rights reserved.
3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
12 *
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
17 */
18
19#include <linux/sched.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090020#include <linux/slab.h>
Miao Xiec6adc9c2013-05-28 10:05:39 +000021#include <linux/blkdev.h>
Josef Bacik5dc562c2012-08-17 13:14:17 -040022#include <linux/list_sort.h>
Chris Masone02119d2008-09-05 16:13:11 -040023#include "ctree.h"
24#include "transaction.h"
25#include "disk-io.h"
26#include "locking.h"
27#include "print-tree.h"
Mark Fashehf1863732012-08-08 11:32:27 -070028#include "backref.h"
Christoph Hellwigb2950862008-12-02 09:54:17 -050029#include "tree-log.h"
Mark Fashehf1863732012-08-08 11:32:27 -070030#include "hash.h"
Chris Masone02119d2008-09-05 16:13:11 -040031
32/* magic values for the inode_only field in btrfs_log_inode:
33 *
34 * LOG_INODE_ALL means to log everything
35 * LOG_INODE_EXISTS means to log just enough to recreate the inode
36 * during log replay
37 */
38#define LOG_INODE_ALL 0
39#define LOG_INODE_EXISTS 1
40
41/*
Chris Mason12fcfd22009-03-24 10:24:20 -040042 * directory trouble cases
43 *
44 * 1) on rename or unlink, if the inode being unlinked isn't in the fsync
45 * log, we must force a full commit before doing an fsync of the directory
46 * where the unlink was done.
47 * ---> record transid of last unlink/rename per directory
48 *
49 * mkdir foo/some_dir
50 * normal commit
51 * rename foo/some_dir foo2/some_dir
52 * mkdir foo/some_dir
53 * fsync foo/some_dir/some_file
54 *
55 * The fsync above will unlink the original some_dir without recording
56 * it in its new location (foo2). After a crash, some_dir will be gone
57 * unless the fsync of some_file forces a full commit
58 *
59 * 2) we must log any new names for any file or dir that is in the fsync
60 * log. ---> check inode while renaming/linking.
61 *
62 * 2a) we must log any new names for any file or dir during rename
63 * when the directory they are being removed from was logged.
64 * ---> check inode and old parent dir during rename
65 *
66 * 2a is actually the more important variant. With the extra logging
67 * a crash might unlink the old name without recreating the new one
68 *
69 * 3) after a crash, we must go through any directories with a link count
70 * of zero and redo the rm -rf
71 *
72 * mkdir f1/foo
73 * normal commit
74 * rm -rf f1/foo
75 * fsync(f1)
76 *
77 * The directory f1 was fully removed from the FS, but fsync was never
78 * called on f1, only its parent dir. After a crash the rm -rf must
79 * be replayed. This must be able to recurse down the entire
80 * directory tree. The inode link count fixup code takes care of the
81 * ugly details.
82 */
83
84/*
Chris Masone02119d2008-09-05 16:13:11 -040085 * stages for the tree walking. The first
86 * stage (0) is to only pin down the blocks we find
87 * the second stage (1) is to make sure that all the inodes
88 * we find in the log are created in the subvolume.
89 *
90 * The last stage is to deal with directories and links and extents
91 * and all the other fun semantics
92 */
93#define LOG_WALK_PIN_ONLY 0
94#define LOG_WALK_REPLAY_INODES 1
Josef Bacikdd8e7212013-09-11 11:57:23 -040095#define LOG_WALK_REPLAY_DIR_INDEX 2
96#define LOG_WALK_REPLAY_ALL 3
Chris Masone02119d2008-09-05 16:13:11 -040097
Chris Mason12fcfd22009-03-24 10:24:20 -040098static int btrfs_log_inode(struct btrfs_trans_handle *trans,
Chris Masone02119d2008-09-05 16:13:11 -040099 struct btrfs_root *root, struct inode *inode,
100 int inode_only);
Yan Zhengec051c02009-01-05 15:43:42 -0500101static int link_to_fixup_dir(struct btrfs_trans_handle *trans,
102 struct btrfs_root *root,
103 struct btrfs_path *path, u64 objectid);
Chris Mason12fcfd22009-03-24 10:24:20 -0400104static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans,
105 struct btrfs_root *root,
106 struct btrfs_root *log,
107 struct btrfs_path *path,
108 u64 dirid, int del_all);
Chris Masone02119d2008-09-05 16:13:11 -0400109
110/*
111 * tree logging is a special write ahead log used to make sure that
112 * fsyncs and O_SYNCs can happen without doing full tree commits.
113 *
114 * Full tree commits are expensive because they require commonly
115 * modified blocks to be recowed, creating many dirty pages in the
116 * extent tree an 4x-6x higher write load than ext3.
117 *
118 * Instead of doing a tree commit on every fsync, we use the
119 * key ranges and transaction ids to find items for a given file or directory
120 * that have changed in this transaction. Those items are copied into
121 * a special tree (one per subvolume root), that tree is written to disk
122 * and then the fsync is considered complete.
123 *
124 * After a crash, items are copied out of the log-tree back into the
125 * subvolume tree. Any file data extents found are recorded in the extent
126 * allocation tree, and the log-tree freed.
127 *
128 * The log tree is read three times, once to pin down all the extents it is
129 * using in ram and once, once to create all the inodes logged in the tree
130 * and once to do all the other items.
131 */
132
133/*
Chris Masone02119d2008-09-05 16:13:11 -0400134 * start a sub transaction and setup the log tree
135 * this increments the log tree writer count to make the people
136 * syncing the tree wait for us to finish
137 */
138static int start_log_trans(struct btrfs_trans_handle *trans,
139 struct btrfs_root *root)
140{
141 int ret;
Yan, Zheng4a500fd2010-05-16 10:49:59 -0400142 int err = 0;
Yan Zheng7237f182009-01-21 12:54:03 -0500143
144 mutex_lock(&root->log_mutex);
145 if (root->log_root) {
Josef Bacikff782e02009-10-08 15:30:04 -0400146 if (!root->log_start_pid) {
147 root->log_start_pid = current->pid;
148 root->log_multiple_pids = false;
149 } else if (root->log_start_pid != current->pid) {
150 root->log_multiple_pids = true;
151 }
152
Miao Xie2ecb7922012-09-06 04:04:27 -0600153 atomic_inc(&root->log_batch);
Yan Zheng7237f182009-01-21 12:54:03 -0500154 atomic_inc(&root->log_writers);
155 mutex_unlock(&root->log_mutex);
156 return 0;
157 }
Josef Bacikff782e02009-10-08 15:30:04 -0400158 root->log_multiple_pids = false;
159 root->log_start_pid = current->pid;
Chris Masone02119d2008-09-05 16:13:11 -0400160 mutex_lock(&root->fs_info->tree_log_mutex);
161 if (!root->fs_info->log_root_tree) {
162 ret = btrfs_init_log_root_tree(trans, root->fs_info);
Yan, Zheng4a500fd2010-05-16 10:49:59 -0400163 if (ret)
164 err = ret;
Chris Masone02119d2008-09-05 16:13:11 -0400165 }
Yan, Zheng4a500fd2010-05-16 10:49:59 -0400166 if (err == 0 && !root->log_root) {
Chris Masone02119d2008-09-05 16:13:11 -0400167 ret = btrfs_add_log_tree(trans, root);
Yan, Zheng4a500fd2010-05-16 10:49:59 -0400168 if (ret)
169 err = ret;
Chris Masone02119d2008-09-05 16:13:11 -0400170 }
Chris Masone02119d2008-09-05 16:13:11 -0400171 mutex_unlock(&root->fs_info->tree_log_mutex);
Miao Xie2ecb7922012-09-06 04:04:27 -0600172 atomic_inc(&root->log_batch);
Yan Zheng7237f182009-01-21 12:54:03 -0500173 atomic_inc(&root->log_writers);
174 mutex_unlock(&root->log_mutex);
Yan, Zheng4a500fd2010-05-16 10:49:59 -0400175 return err;
Chris Masone02119d2008-09-05 16:13:11 -0400176}
177
178/*
179 * returns 0 if there was a log transaction running and we were able
180 * to join, or returns -ENOENT if there were not transactions
181 * in progress
182 */
183static int join_running_log_trans(struct btrfs_root *root)
184{
185 int ret = -ENOENT;
186
187 smp_mb();
188 if (!root->log_root)
189 return -ENOENT;
190
Yan Zheng7237f182009-01-21 12:54:03 -0500191 mutex_lock(&root->log_mutex);
Chris Masone02119d2008-09-05 16:13:11 -0400192 if (root->log_root) {
193 ret = 0;
Yan Zheng7237f182009-01-21 12:54:03 -0500194 atomic_inc(&root->log_writers);
Chris Masone02119d2008-09-05 16:13:11 -0400195 }
Yan Zheng7237f182009-01-21 12:54:03 -0500196 mutex_unlock(&root->log_mutex);
Chris Masone02119d2008-09-05 16:13:11 -0400197 return ret;
198}
199
200/*
Chris Mason12fcfd22009-03-24 10:24:20 -0400201 * This either makes the current running log transaction wait
202 * until you call btrfs_end_log_trans() or it makes any future
203 * log transactions wait until you call btrfs_end_log_trans()
204 */
205int btrfs_pin_log_trans(struct btrfs_root *root)
206{
207 int ret = -ENOENT;
208
209 mutex_lock(&root->log_mutex);
210 atomic_inc(&root->log_writers);
211 mutex_unlock(&root->log_mutex);
212 return ret;
213}
214
215/*
Chris Masone02119d2008-09-05 16:13:11 -0400216 * indicate we're done making changes to the log tree
217 * and wake up anyone waiting to do a sync
218 */
Jeff Mahoney143bede2012-03-01 14:56:26 +0100219void btrfs_end_log_trans(struct btrfs_root *root)
Chris Masone02119d2008-09-05 16:13:11 -0400220{
Yan Zheng7237f182009-01-21 12:54:03 -0500221 if (atomic_dec_and_test(&root->log_writers)) {
222 smp_mb();
223 if (waitqueue_active(&root->log_writer_wait))
224 wake_up(&root->log_writer_wait);
225 }
Chris Masone02119d2008-09-05 16:13:11 -0400226}
227
228
229/*
230 * the walk control struct is used to pass state down the chain when
231 * processing the log tree. The stage field tells us which part
232 * of the log tree processing we are currently doing. The others
233 * are state fields used for that specific part
234 */
235struct walk_control {
236 /* should we free the extent on disk when done? This is used
237 * at transaction commit time while freeing a log tree
238 */
239 int free;
240
241 /* should we write out the extent buffer? This is used
242 * while flushing the log tree to disk during a sync
243 */
244 int write;
245
246 /* should we wait for the extent buffer io to finish? Also used
247 * while flushing the log tree to disk for a sync
248 */
249 int wait;
250
251 /* pin only walk, we record which extents on disk belong to the
252 * log trees
253 */
254 int pin;
255
256 /* what stage of the replay code we're currently in */
257 int stage;
258
259 /* the root we are currently replaying */
260 struct btrfs_root *replay_dest;
261
262 /* the trans handle for the current replay */
263 struct btrfs_trans_handle *trans;
264
265 /* the function that gets used to process blocks we find in the
266 * tree. Note the extent_buffer might not be up to date when it is
267 * passed in, and it must be checked or read if you need the data
268 * inside it
269 */
270 int (*process_func)(struct btrfs_root *log, struct extent_buffer *eb,
271 struct walk_control *wc, u64 gen);
272};
273
274/*
275 * process_func used to pin down extents, write them or wait on them
276 */
277static int process_one_buffer(struct btrfs_root *log,
278 struct extent_buffer *eb,
279 struct walk_control *wc, u64 gen)
280{
Josef Bacikb50c6e22013-04-25 15:55:30 -0400281 int ret = 0;
Chris Masone02119d2008-09-05 16:13:11 -0400282
Josef Bacik8c2a1a32013-06-06 13:19:32 -0400283 /*
284 * If this fs is mixed then we need to be able to process the leaves to
285 * pin down any logged extents, so we have to read the block.
286 */
287 if (btrfs_fs_incompat(log->fs_info, MIXED_GROUPS)) {
288 ret = btrfs_read_buffer(eb, gen);
289 if (ret)
290 return ret;
291 }
292
Josef Bacikb50c6e22013-04-25 15:55:30 -0400293 if (wc->pin)
294 ret = btrfs_pin_extent_for_log_replay(log->fs_info->extent_root,
295 eb->start, eb->len);
296
297 if (!ret && btrfs_buffer_uptodate(eb, gen, 0)) {
Josef Bacik8c2a1a32013-06-06 13:19:32 -0400298 if (wc->pin && btrfs_header_level(eb) == 0)
299 ret = btrfs_exclude_logged_extents(log, eb);
Chris Masone02119d2008-09-05 16:13:11 -0400300 if (wc->write)
301 btrfs_write_tree_block(eb);
302 if (wc->wait)
303 btrfs_wait_tree_block_writeback(eb);
304 }
Josef Bacikb50c6e22013-04-25 15:55:30 -0400305 return ret;
Chris Masone02119d2008-09-05 16:13:11 -0400306}
307
308/*
309 * Item overwrite used by replay and tree logging. eb, slot and key all refer
310 * to the src data we are copying out.
311 *
312 * root is the tree we are copying into, and path is a scratch
313 * path for use in this function (it should be released on entry and
314 * will be released on exit).
315 *
316 * If the key is already in the destination tree the existing item is
317 * overwritten. If the existing item isn't big enough, it is extended.
318 * If it is too large, it is truncated.
319 *
320 * If the key isn't in the destination yet, a new item is inserted.
321 */
322static noinline int overwrite_item(struct btrfs_trans_handle *trans,
323 struct btrfs_root *root,
324 struct btrfs_path *path,
325 struct extent_buffer *eb, int slot,
326 struct btrfs_key *key)
327{
328 int ret;
329 u32 item_size;
330 u64 saved_i_size = 0;
331 int save_old_i_size = 0;
332 unsigned long src_ptr;
333 unsigned long dst_ptr;
334 int overwrite_root = 0;
Josef Bacik4bc4bee2013-04-05 20:50:09 +0000335 bool inode_item = key->type == BTRFS_INODE_ITEM_KEY;
Chris Masone02119d2008-09-05 16:13:11 -0400336
337 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
338 overwrite_root = 1;
339
340 item_size = btrfs_item_size_nr(eb, slot);
341 src_ptr = btrfs_item_ptr_offset(eb, slot);
342
343 /* look for the key in the destination tree */
344 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
Josef Bacik4bc4bee2013-04-05 20:50:09 +0000345 if (ret < 0)
346 return ret;
347
Chris Masone02119d2008-09-05 16:13:11 -0400348 if (ret == 0) {
349 char *src_copy;
350 char *dst_copy;
351 u32 dst_size = btrfs_item_size_nr(path->nodes[0],
352 path->slots[0]);
353 if (dst_size != item_size)
354 goto insert;
355
356 if (item_size == 0) {
David Sterbab3b4aa72011-04-21 01:20:15 +0200357 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -0400358 return 0;
359 }
360 dst_copy = kmalloc(item_size, GFP_NOFS);
361 src_copy = kmalloc(item_size, GFP_NOFS);
liubo2a29edc2011-01-26 06:22:08 +0000362 if (!dst_copy || !src_copy) {
David Sterbab3b4aa72011-04-21 01:20:15 +0200363 btrfs_release_path(path);
liubo2a29edc2011-01-26 06:22:08 +0000364 kfree(dst_copy);
365 kfree(src_copy);
366 return -ENOMEM;
367 }
Chris Masone02119d2008-09-05 16:13:11 -0400368
369 read_extent_buffer(eb, src_copy, src_ptr, item_size);
370
371 dst_ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
372 read_extent_buffer(path->nodes[0], dst_copy, dst_ptr,
373 item_size);
374 ret = memcmp(dst_copy, src_copy, item_size);
375
376 kfree(dst_copy);
377 kfree(src_copy);
378 /*
379 * they have the same contents, just return, this saves
380 * us from cowing blocks in the destination tree and doing
381 * extra writes that may not have been done by a previous
382 * sync
383 */
384 if (ret == 0) {
David Sterbab3b4aa72011-04-21 01:20:15 +0200385 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -0400386 return 0;
387 }
388
Josef Bacik4bc4bee2013-04-05 20:50:09 +0000389 /*
390 * We need to load the old nbytes into the inode so when we
391 * replay the extents we've logged we get the right nbytes.
392 */
393 if (inode_item) {
394 struct btrfs_inode_item *item;
395 u64 nbytes;
Josef Bacikd5554382013-09-11 14:17:00 -0400396 u32 mode;
Josef Bacik4bc4bee2013-04-05 20:50:09 +0000397
398 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
399 struct btrfs_inode_item);
400 nbytes = btrfs_inode_nbytes(path->nodes[0], item);
401 item = btrfs_item_ptr(eb, slot,
402 struct btrfs_inode_item);
403 btrfs_set_inode_nbytes(eb, item, nbytes);
Josef Bacikd5554382013-09-11 14:17:00 -0400404
405 /*
406 * If this is a directory we need to reset the i_size to
407 * 0 so that we can set it up properly when replaying
408 * the rest of the items in this log.
409 */
410 mode = btrfs_inode_mode(eb, item);
411 if (S_ISDIR(mode))
412 btrfs_set_inode_size(eb, item, 0);
Josef Bacik4bc4bee2013-04-05 20:50:09 +0000413 }
414 } else if (inode_item) {
415 struct btrfs_inode_item *item;
Josef Bacikd5554382013-09-11 14:17:00 -0400416 u32 mode;
Josef Bacik4bc4bee2013-04-05 20:50:09 +0000417
418 /*
419 * New inode, set nbytes to 0 so that the nbytes comes out
420 * properly when we replay the extents.
421 */
422 item = btrfs_item_ptr(eb, slot, struct btrfs_inode_item);
423 btrfs_set_inode_nbytes(eb, item, 0);
Josef Bacikd5554382013-09-11 14:17:00 -0400424
425 /*
426 * If this is a directory we need to reset the i_size to 0 so
427 * that we can set it up properly when replaying the rest of
428 * the items in this log.
429 */
430 mode = btrfs_inode_mode(eb, item);
431 if (S_ISDIR(mode))
432 btrfs_set_inode_size(eb, item, 0);
Chris Masone02119d2008-09-05 16:13:11 -0400433 }
434insert:
David Sterbab3b4aa72011-04-21 01:20:15 +0200435 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -0400436 /* try to insert the key into the destination tree */
437 ret = btrfs_insert_empty_item(trans, root, path,
438 key, item_size);
439
440 /* make sure any existing item is the correct size */
441 if (ret == -EEXIST) {
442 u32 found_size;
443 found_size = btrfs_item_size_nr(path->nodes[0],
444 path->slots[0]);
Jeff Mahoney143bede2012-03-01 14:56:26 +0100445 if (found_size > item_size)
Tsutomu Itohafe5fea2013-04-16 05:18:22 +0000446 btrfs_truncate_item(root, path, item_size, 1);
Jeff Mahoney143bede2012-03-01 14:56:26 +0100447 else if (found_size < item_size)
Tsutomu Itoh4b90c682013-04-16 05:18:49 +0000448 btrfs_extend_item(root, path,
Jeff Mahoney143bede2012-03-01 14:56:26 +0100449 item_size - found_size);
Chris Masone02119d2008-09-05 16:13:11 -0400450 } else if (ret) {
Yan, Zheng4a500fd2010-05-16 10:49:59 -0400451 return ret;
Chris Masone02119d2008-09-05 16:13:11 -0400452 }
453 dst_ptr = btrfs_item_ptr_offset(path->nodes[0],
454 path->slots[0]);
455
456 /* don't overwrite an existing inode if the generation number
457 * was logged as zero. This is done when the tree logging code
458 * is just logging an inode to make sure it exists after recovery.
459 *
460 * Also, don't overwrite i_size on directories during replay.
461 * log replay inserts and removes directory items based on the
462 * state of the tree found in the subvolume, and i_size is modified
463 * as it goes
464 */
465 if (key->type == BTRFS_INODE_ITEM_KEY && ret == -EEXIST) {
466 struct btrfs_inode_item *src_item;
467 struct btrfs_inode_item *dst_item;
468
469 src_item = (struct btrfs_inode_item *)src_ptr;
470 dst_item = (struct btrfs_inode_item *)dst_ptr;
471
472 if (btrfs_inode_generation(eb, src_item) == 0)
473 goto no_copy;
474
475 if (overwrite_root &&
476 S_ISDIR(btrfs_inode_mode(eb, src_item)) &&
477 S_ISDIR(btrfs_inode_mode(path->nodes[0], dst_item))) {
478 save_old_i_size = 1;
479 saved_i_size = btrfs_inode_size(path->nodes[0],
480 dst_item);
481 }
482 }
483
484 copy_extent_buffer(path->nodes[0], eb, dst_ptr,
485 src_ptr, item_size);
486
487 if (save_old_i_size) {
488 struct btrfs_inode_item *dst_item;
489 dst_item = (struct btrfs_inode_item *)dst_ptr;
490 btrfs_set_inode_size(path->nodes[0], dst_item, saved_i_size);
491 }
492
493 /* make sure the generation is filled in */
494 if (key->type == BTRFS_INODE_ITEM_KEY) {
495 struct btrfs_inode_item *dst_item;
496 dst_item = (struct btrfs_inode_item *)dst_ptr;
497 if (btrfs_inode_generation(path->nodes[0], dst_item) == 0) {
498 btrfs_set_inode_generation(path->nodes[0], dst_item,
499 trans->transid);
500 }
501 }
502no_copy:
503 btrfs_mark_buffer_dirty(path->nodes[0]);
David Sterbab3b4aa72011-04-21 01:20:15 +0200504 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -0400505 return 0;
506}
507
508/*
509 * simple helper to read an inode off the disk from a given root
510 * This can only be called for subvolume roots and not for the log
511 */
512static noinline struct inode *read_one_inode(struct btrfs_root *root,
513 u64 objectid)
514{
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400515 struct btrfs_key key;
Chris Masone02119d2008-09-05 16:13:11 -0400516 struct inode *inode;
Chris Masone02119d2008-09-05 16:13:11 -0400517
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400518 key.objectid = objectid;
519 key.type = BTRFS_INODE_ITEM_KEY;
520 key.offset = 0;
Josef Bacik73f73412009-12-04 17:38:27 +0000521 inode = btrfs_iget(root->fs_info->sb, &key, root, NULL);
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400522 if (IS_ERR(inode)) {
523 inode = NULL;
524 } else if (is_bad_inode(inode)) {
Chris Masone02119d2008-09-05 16:13:11 -0400525 iput(inode);
526 inode = NULL;
527 }
528 return inode;
529}
530
531/* replays a single extent in 'eb' at 'slot' with 'key' into the
532 * subvolume 'root'. path is released on entry and should be released
533 * on exit.
534 *
535 * extents in the log tree have not been allocated out of the extent
536 * tree yet. So, this completes the allocation, taking a reference
537 * as required if the extent already exists or creating a new extent
538 * if it isn't in the extent allocation tree yet.
539 *
540 * The extent is inserted into the file, dropping any existing extents
541 * from the file that overlap the new one.
542 */
543static noinline int replay_one_extent(struct btrfs_trans_handle *trans,
544 struct btrfs_root *root,
545 struct btrfs_path *path,
546 struct extent_buffer *eb, int slot,
547 struct btrfs_key *key)
548{
549 int found_type;
Chris Masone02119d2008-09-05 16:13:11 -0400550 u64 extent_end;
Chris Masone02119d2008-09-05 16:13:11 -0400551 u64 start = key->offset;
Josef Bacik4bc4bee2013-04-05 20:50:09 +0000552 u64 nbytes = 0;
Chris Masone02119d2008-09-05 16:13:11 -0400553 struct btrfs_file_extent_item *item;
554 struct inode *inode = NULL;
555 unsigned long size;
556 int ret = 0;
557
558 item = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
559 found_type = btrfs_file_extent_type(eb, item);
560
Yan Zhengd899e052008-10-30 14:25:28 -0400561 if (found_type == BTRFS_FILE_EXTENT_REG ||
Josef Bacik4bc4bee2013-04-05 20:50:09 +0000562 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
563 nbytes = btrfs_file_extent_num_bytes(eb, item);
564 extent_end = start + nbytes;
565
566 /*
567 * We don't add to the inodes nbytes if we are prealloc or a
568 * hole.
569 */
570 if (btrfs_file_extent_disk_bytenr(eb, item) == 0)
571 nbytes = 0;
572 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
Chris Mason514ac8a2014-01-03 21:07:00 -0800573 size = btrfs_file_extent_inline_len(eb, slot, item);
Josef Bacik4bc4bee2013-04-05 20:50:09 +0000574 nbytes = btrfs_file_extent_ram_bytes(eb, item);
Qu Wenruofda28322013-02-26 08:10:22 +0000575 extent_end = ALIGN(start + size, root->sectorsize);
Chris Masone02119d2008-09-05 16:13:11 -0400576 } else {
577 ret = 0;
578 goto out;
579 }
580
581 inode = read_one_inode(root, key->objectid);
582 if (!inode) {
583 ret = -EIO;
584 goto out;
585 }
586
587 /*
588 * first check to see if we already have this extent in the
589 * file. This must be done before the btrfs_drop_extents run
590 * so we don't try to drop this extent.
591 */
Li Zefan33345d012011-04-20 10:31:50 +0800592 ret = btrfs_lookup_file_extent(trans, root, path, btrfs_ino(inode),
Chris Masone02119d2008-09-05 16:13:11 -0400593 start, 0);
594
Yan Zhengd899e052008-10-30 14:25:28 -0400595 if (ret == 0 &&
596 (found_type == BTRFS_FILE_EXTENT_REG ||
597 found_type == BTRFS_FILE_EXTENT_PREALLOC)) {
Chris Masone02119d2008-09-05 16:13:11 -0400598 struct btrfs_file_extent_item cmp1;
599 struct btrfs_file_extent_item cmp2;
600 struct btrfs_file_extent_item *existing;
601 struct extent_buffer *leaf;
602
603 leaf = path->nodes[0];
604 existing = btrfs_item_ptr(leaf, path->slots[0],
605 struct btrfs_file_extent_item);
606
607 read_extent_buffer(eb, &cmp1, (unsigned long)item,
608 sizeof(cmp1));
609 read_extent_buffer(leaf, &cmp2, (unsigned long)existing,
610 sizeof(cmp2));
611
612 /*
613 * we already have a pointer to this exact extent,
614 * we don't have to do anything
615 */
616 if (memcmp(&cmp1, &cmp2, sizeof(cmp1)) == 0) {
David Sterbab3b4aa72011-04-21 01:20:15 +0200617 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -0400618 goto out;
619 }
620 }
David Sterbab3b4aa72011-04-21 01:20:15 +0200621 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -0400622
623 /* drop any overlapping extents */
Josef Bacik26714852012-08-29 12:24:27 -0400624 ret = btrfs_drop_extents(trans, root, inode, start, extent_end, 1);
Josef Bacik36508602013-04-25 16:23:32 -0400625 if (ret)
626 goto out;
Chris Masone02119d2008-09-05 16:13:11 -0400627
Yan Zheng07d400a2009-01-06 11:42:00 -0500628 if (found_type == BTRFS_FILE_EXTENT_REG ||
629 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400630 u64 offset;
Yan Zheng07d400a2009-01-06 11:42:00 -0500631 unsigned long dest_offset;
632 struct btrfs_key ins;
Chris Masone02119d2008-09-05 16:13:11 -0400633
Yan Zheng07d400a2009-01-06 11:42:00 -0500634 ret = btrfs_insert_empty_item(trans, root, path, key,
635 sizeof(*item));
Josef Bacik36508602013-04-25 16:23:32 -0400636 if (ret)
637 goto out;
Yan Zheng07d400a2009-01-06 11:42:00 -0500638 dest_offset = btrfs_item_ptr_offset(path->nodes[0],
639 path->slots[0]);
640 copy_extent_buffer(path->nodes[0], eb, dest_offset,
641 (unsigned long)item, sizeof(*item));
642
643 ins.objectid = btrfs_file_extent_disk_bytenr(eb, item);
644 ins.offset = btrfs_file_extent_disk_num_bytes(eb, item);
645 ins.type = BTRFS_EXTENT_ITEM_KEY;
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400646 offset = key->offset - btrfs_file_extent_offset(eb, item);
Yan Zheng07d400a2009-01-06 11:42:00 -0500647
648 if (ins.objectid > 0) {
649 u64 csum_start;
650 u64 csum_end;
651 LIST_HEAD(ordered_sums);
652 /*
653 * is this extent already allocated in the extent
654 * allocation tree? If so, just add a reference
655 */
656 ret = btrfs_lookup_extent(root, ins.objectid,
657 ins.offset);
658 if (ret == 0) {
659 ret = btrfs_inc_extent_ref(trans, root,
660 ins.objectid, ins.offset,
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400661 0, root->root_key.objectid,
Arne Jansen66d7e7f2011-09-12 15:26:38 +0200662 key->objectid, offset, 0);
Josef Bacikb50c6e22013-04-25 15:55:30 -0400663 if (ret)
664 goto out;
Yan Zheng07d400a2009-01-06 11:42:00 -0500665 } else {
666 /*
667 * insert the extent pointer in the extent
668 * allocation tree
669 */
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400670 ret = btrfs_alloc_logged_file_extent(trans,
671 root, root->root_key.objectid,
672 key->objectid, offset, &ins);
Josef Bacikb50c6e22013-04-25 15:55:30 -0400673 if (ret)
674 goto out;
Yan Zheng07d400a2009-01-06 11:42:00 -0500675 }
David Sterbab3b4aa72011-04-21 01:20:15 +0200676 btrfs_release_path(path);
Yan Zheng07d400a2009-01-06 11:42:00 -0500677
678 if (btrfs_file_extent_compression(eb, item)) {
679 csum_start = ins.objectid;
680 csum_end = csum_start + ins.offset;
681 } else {
682 csum_start = ins.objectid +
683 btrfs_file_extent_offset(eb, item);
684 csum_end = csum_start +
685 btrfs_file_extent_num_bytes(eb, item);
686 }
687
688 ret = btrfs_lookup_csums_range(root->log_root,
689 csum_start, csum_end - 1,
Arne Jansena2de7332011-03-08 14:14:00 +0100690 &ordered_sums, 0);
Josef Bacik36508602013-04-25 16:23:32 -0400691 if (ret)
692 goto out;
Yan Zheng07d400a2009-01-06 11:42:00 -0500693 while (!list_empty(&ordered_sums)) {
694 struct btrfs_ordered_sum *sums;
695 sums = list_entry(ordered_sums.next,
696 struct btrfs_ordered_sum,
697 list);
Josef Bacik36508602013-04-25 16:23:32 -0400698 if (!ret)
699 ret = btrfs_csum_file_blocks(trans,
Yan Zheng07d400a2009-01-06 11:42:00 -0500700 root->fs_info->csum_root,
701 sums);
Yan Zheng07d400a2009-01-06 11:42:00 -0500702 list_del(&sums->list);
703 kfree(sums);
704 }
Josef Bacik36508602013-04-25 16:23:32 -0400705 if (ret)
706 goto out;
Yan Zheng07d400a2009-01-06 11:42:00 -0500707 } else {
David Sterbab3b4aa72011-04-21 01:20:15 +0200708 btrfs_release_path(path);
Yan Zheng07d400a2009-01-06 11:42:00 -0500709 }
710 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
711 /* inline extents are easy, we just overwrite them */
712 ret = overwrite_item(trans, root, path, eb, slot, key);
Josef Bacik36508602013-04-25 16:23:32 -0400713 if (ret)
714 goto out;
Yan Zheng07d400a2009-01-06 11:42:00 -0500715 }
716
Josef Bacik4bc4bee2013-04-05 20:50:09 +0000717 inode_add_bytes(inode, nbytes);
Tsutomu Itohb9959292012-06-25 21:25:22 -0600718 ret = btrfs_update_inode(trans, root, inode);
Chris Masone02119d2008-09-05 16:13:11 -0400719out:
720 if (inode)
721 iput(inode);
722 return ret;
723}
724
725/*
726 * when cleaning up conflicts between the directory names in the
727 * subvolume, directory names in the log and directory names in the
728 * inode back references, we may have to unlink inodes from directories.
729 *
730 * This is a helper function to do the unlink of a specific directory
731 * item
732 */
733static noinline int drop_one_dir_item(struct btrfs_trans_handle *trans,
734 struct btrfs_root *root,
735 struct btrfs_path *path,
736 struct inode *dir,
737 struct btrfs_dir_item *di)
738{
739 struct inode *inode;
740 char *name;
741 int name_len;
742 struct extent_buffer *leaf;
743 struct btrfs_key location;
744 int ret;
745
746 leaf = path->nodes[0];
747
748 btrfs_dir_item_key_to_cpu(leaf, di, &location);
749 name_len = btrfs_dir_name_len(leaf, di);
750 name = kmalloc(name_len, GFP_NOFS);
liubo2a29edc2011-01-26 06:22:08 +0000751 if (!name)
752 return -ENOMEM;
753
Chris Masone02119d2008-09-05 16:13:11 -0400754 read_extent_buffer(leaf, name, (unsigned long)(di + 1), name_len);
David Sterbab3b4aa72011-04-21 01:20:15 +0200755 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -0400756
757 inode = read_one_inode(root, location.objectid);
Tsutomu Itohc00e9492011-04-28 09:10:23 +0000758 if (!inode) {
Josef Bacik36508602013-04-25 16:23:32 -0400759 ret = -EIO;
760 goto out;
Tsutomu Itohc00e9492011-04-28 09:10:23 +0000761 }
Chris Masone02119d2008-09-05 16:13:11 -0400762
Yan Zhengec051c02009-01-05 15:43:42 -0500763 ret = link_to_fixup_dir(trans, root, path, location.objectid);
Josef Bacik36508602013-04-25 16:23:32 -0400764 if (ret)
765 goto out;
Chris Mason12fcfd22009-03-24 10:24:20 -0400766
Chris Masone02119d2008-09-05 16:13:11 -0400767 ret = btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
Josef Bacik36508602013-04-25 16:23:32 -0400768 if (ret)
769 goto out;
Filipe David Borba Mananaada9af22013-08-05 09:25:47 +0100770 else
771 ret = btrfs_run_delayed_items(trans, root);
Josef Bacik36508602013-04-25 16:23:32 -0400772out:
773 kfree(name);
774 iput(inode);
Chris Masone02119d2008-09-05 16:13:11 -0400775 return ret;
776}
777
778/*
779 * helper function to see if a given name and sequence number found
780 * in an inode back reference are already in a directory and correctly
781 * point to this inode
782 */
783static noinline int inode_in_dir(struct btrfs_root *root,
784 struct btrfs_path *path,
785 u64 dirid, u64 objectid, u64 index,
786 const char *name, int name_len)
787{
788 struct btrfs_dir_item *di;
789 struct btrfs_key location;
790 int match = 0;
791
792 di = btrfs_lookup_dir_index_item(NULL, root, path, dirid,
793 index, name, name_len, 0);
794 if (di && !IS_ERR(di)) {
795 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
796 if (location.objectid != objectid)
797 goto out;
798 } else
799 goto out;
David Sterbab3b4aa72011-04-21 01:20:15 +0200800 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -0400801
802 di = btrfs_lookup_dir_item(NULL, root, path, dirid, name, name_len, 0);
803 if (di && !IS_ERR(di)) {
804 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
805 if (location.objectid != objectid)
806 goto out;
807 } else
808 goto out;
809 match = 1;
810out:
David Sterbab3b4aa72011-04-21 01:20:15 +0200811 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -0400812 return match;
813}
814
815/*
816 * helper function to check a log tree for a named back reference in
817 * an inode. This is used to decide if a back reference that is
818 * found in the subvolume conflicts with what we find in the log.
819 *
820 * inode backreferences may have multiple refs in a single item,
821 * during replay we process one reference at a time, and we don't
822 * want to delete valid links to a file from the subvolume if that
823 * link is also in the log.
824 */
825static noinline int backref_in_log(struct btrfs_root *log,
826 struct btrfs_key *key,
Mark Fashehf1863732012-08-08 11:32:27 -0700827 u64 ref_objectid,
Chris Masone02119d2008-09-05 16:13:11 -0400828 char *name, int namelen)
829{
830 struct btrfs_path *path;
831 struct btrfs_inode_ref *ref;
832 unsigned long ptr;
833 unsigned long ptr_end;
834 unsigned long name_ptr;
835 int found_name_len;
836 int item_size;
837 int ret;
838 int match = 0;
839
840 path = btrfs_alloc_path();
liubo2a29edc2011-01-26 06:22:08 +0000841 if (!path)
842 return -ENOMEM;
843
Chris Masone02119d2008-09-05 16:13:11 -0400844 ret = btrfs_search_slot(NULL, log, key, path, 0, 0);
845 if (ret != 0)
846 goto out;
847
Chris Masone02119d2008-09-05 16:13:11 -0400848 ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
Mark Fashehf1863732012-08-08 11:32:27 -0700849
850 if (key->type == BTRFS_INODE_EXTREF_KEY) {
851 if (btrfs_find_name_in_ext_backref(path, ref_objectid,
852 name, namelen, NULL))
853 match = 1;
854
855 goto out;
856 }
857
858 item_size = btrfs_item_size_nr(path->nodes[0], path->slots[0]);
Chris Masone02119d2008-09-05 16:13:11 -0400859 ptr_end = ptr + item_size;
860 while (ptr < ptr_end) {
861 ref = (struct btrfs_inode_ref *)ptr;
862 found_name_len = btrfs_inode_ref_name_len(path->nodes[0], ref);
863 if (found_name_len == namelen) {
864 name_ptr = (unsigned long)(ref + 1);
865 ret = memcmp_extent_buffer(path->nodes[0], name,
866 name_ptr, namelen);
867 if (ret == 0) {
868 match = 1;
869 goto out;
870 }
871 }
872 ptr = (unsigned long)(ref + 1) + found_name_len;
873 }
874out:
875 btrfs_free_path(path);
876 return match;
877}
878
Jan Schmidt5a1d7842012-08-17 14:04:41 -0700879static inline int __add_inode_ref(struct btrfs_trans_handle *trans,
880 struct btrfs_root *root,
881 struct btrfs_path *path,
882 struct btrfs_root *log_root,
883 struct inode *dir, struct inode *inode,
Jan Schmidt5a1d7842012-08-17 14:04:41 -0700884 struct extent_buffer *eb,
Mark Fashehf1863732012-08-08 11:32:27 -0700885 u64 inode_objectid, u64 parent_objectid,
886 u64 ref_index, char *name, int namelen,
887 int *search_done)
Jan Schmidt5a1d7842012-08-17 14:04:41 -0700888{
889 int ret;
Mark Fashehf1863732012-08-08 11:32:27 -0700890 char *victim_name;
891 int victim_name_len;
892 struct extent_buffer *leaf;
Jan Schmidt5a1d7842012-08-17 14:04:41 -0700893 struct btrfs_dir_item *di;
Mark Fashehf1863732012-08-08 11:32:27 -0700894 struct btrfs_key search_key;
895 struct btrfs_inode_extref *extref;
Jan Schmidt5a1d7842012-08-17 14:04:41 -0700896
Mark Fashehf1863732012-08-08 11:32:27 -0700897again:
898 /* Search old style refs */
899 search_key.objectid = inode_objectid;
900 search_key.type = BTRFS_INODE_REF_KEY;
901 search_key.offset = parent_objectid;
902 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
Jan Schmidt5a1d7842012-08-17 14:04:41 -0700903 if (ret == 0) {
Jan Schmidt5a1d7842012-08-17 14:04:41 -0700904 struct btrfs_inode_ref *victim_ref;
905 unsigned long ptr;
906 unsigned long ptr_end;
Mark Fashehf1863732012-08-08 11:32:27 -0700907
908 leaf = path->nodes[0];
Jan Schmidt5a1d7842012-08-17 14:04:41 -0700909
910 /* are we trying to overwrite a back ref for the root directory
911 * if so, just jump out, we're done
912 */
Mark Fashehf1863732012-08-08 11:32:27 -0700913 if (search_key.objectid == search_key.offset)
Jan Schmidt5a1d7842012-08-17 14:04:41 -0700914 return 1;
915
916 /* check all the names in this back reference to see
917 * if they are in the log. if so, we allow them to stay
918 * otherwise they must be unlinked as a conflict
919 */
920 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
921 ptr_end = ptr + btrfs_item_size_nr(leaf, path->slots[0]);
922 while (ptr < ptr_end) {
923 victim_ref = (struct btrfs_inode_ref *)ptr;
924 victim_name_len = btrfs_inode_ref_name_len(leaf,
925 victim_ref);
926 victim_name = kmalloc(victim_name_len, GFP_NOFS);
Josef Bacik36508602013-04-25 16:23:32 -0400927 if (!victim_name)
928 return -ENOMEM;
Jan Schmidt5a1d7842012-08-17 14:04:41 -0700929
930 read_extent_buffer(leaf, victim_name,
931 (unsigned long)(victim_ref + 1),
932 victim_name_len);
933
Mark Fashehf1863732012-08-08 11:32:27 -0700934 if (!backref_in_log(log_root, &search_key,
935 parent_objectid,
936 victim_name,
Jan Schmidt5a1d7842012-08-17 14:04:41 -0700937 victim_name_len)) {
Zach Brown8b558c52013-10-16 12:10:34 -0700938 inc_nlink(inode);
Jan Schmidt5a1d7842012-08-17 14:04:41 -0700939 btrfs_release_path(path);
940
941 ret = btrfs_unlink_inode(trans, root, dir,
942 inode, victim_name,
943 victim_name_len);
Mark Fashehf1863732012-08-08 11:32:27 -0700944 kfree(victim_name);
Josef Bacik36508602013-04-25 16:23:32 -0400945 if (ret)
946 return ret;
Filipe David Borba Mananaada9af22013-08-05 09:25:47 +0100947 ret = btrfs_run_delayed_items(trans, root);
948 if (ret)
949 return ret;
Mark Fashehf1863732012-08-08 11:32:27 -0700950 *search_done = 1;
951 goto again;
Jan Schmidt5a1d7842012-08-17 14:04:41 -0700952 }
953 kfree(victim_name);
Mark Fashehf1863732012-08-08 11:32:27 -0700954
Jan Schmidt5a1d7842012-08-17 14:04:41 -0700955 ptr = (unsigned long)(victim_ref + 1) + victim_name_len;
956 }
Jan Schmidt5a1d7842012-08-17 14:04:41 -0700957
958 /*
959 * NOTE: we have searched root tree and checked the
960 * coresponding ref, it does not need to check again.
961 */
962 *search_done = 1;
963 }
964 btrfs_release_path(path);
965
Mark Fashehf1863732012-08-08 11:32:27 -0700966 /* Same search but for extended refs */
967 extref = btrfs_lookup_inode_extref(NULL, root, path, name, namelen,
968 inode_objectid, parent_objectid, 0,
969 0);
970 if (!IS_ERR_OR_NULL(extref)) {
971 u32 item_size;
972 u32 cur_offset = 0;
973 unsigned long base;
974 struct inode *victim_parent;
975
976 leaf = path->nodes[0];
977
978 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
979 base = btrfs_item_ptr_offset(leaf, path->slots[0]);
980
981 while (cur_offset < item_size) {
982 extref = (struct btrfs_inode_extref *)base + cur_offset;
983
984 victim_name_len = btrfs_inode_extref_name_len(leaf, extref);
985
986 if (btrfs_inode_extref_parent(leaf, extref) != parent_objectid)
987 goto next;
988
989 victim_name = kmalloc(victim_name_len, GFP_NOFS);
Josef Bacik36508602013-04-25 16:23:32 -0400990 if (!victim_name)
991 return -ENOMEM;
Mark Fashehf1863732012-08-08 11:32:27 -0700992 read_extent_buffer(leaf, victim_name, (unsigned long)&extref->name,
993 victim_name_len);
994
995 search_key.objectid = inode_objectid;
996 search_key.type = BTRFS_INODE_EXTREF_KEY;
997 search_key.offset = btrfs_extref_hash(parent_objectid,
998 victim_name,
999 victim_name_len);
1000 ret = 0;
1001 if (!backref_in_log(log_root, &search_key,
1002 parent_objectid, victim_name,
1003 victim_name_len)) {
1004 ret = -ENOENT;
1005 victim_parent = read_one_inode(root,
1006 parent_objectid);
1007 if (victim_parent) {
Zach Brown8b558c52013-10-16 12:10:34 -07001008 inc_nlink(inode);
Mark Fashehf1863732012-08-08 11:32:27 -07001009 btrfs_release_path(path);
1010
1011 ret = btrfs_unlink_inode(trans, root,
1012 victim_parent,
1013 inode,
1014 victim_name,
1015 victim_name_len);
Filipe David Borba Mananaada9af22013-08-05 09:25:47 +01001016 if (!ret)
1017 ret = btrfs_run_delayed_items(
1018 trans, root);
Mark Fashehf1863732012-08-08 11:32:27 -07001019 }
Mark Fashehf1863732012-08-08 11:32:27 -07001020 iput(victim_parent);
1021 kfree(victim_name);
Josef Bacik36508602013-04-25 16:23:32 -04001022 if (ret)
1023 return ret;
Mark Fashehf1863732012-08-08 11:32:27 -07001024 *search_done = 1;
1025 goto again;
1026 }
1027 kfree(victim_name);
Josef Bacik36508602013-04-25 16:23:32 -04001028 if (ret)
1029 return ret;
Mark Fashehf1863732012-08-08 11:32:27 -07001030next:
1031 cur_offset += victim_name_len + sizeof(*extref);
1032 }
1033 *search_done = 1;
1034 }
1035 btrfs_release_path(path);
1036
Jan Schmidt5a1d7842012-08-17 14:04:41 -07001037 /* look for a conflicting sequence number */
1038 di = btrfs_lookup_dir_index_item(trans, root, path, btrfs_ino(dir),
Mark Fashehf1863732012-08-08 11:32:27 -07001039 ref_index, name, namelen, 0);
Jan Schmidt5a1d7842012-08-17 14:04:41 -07001040 if (di && !IS_ERR(di)) {
1041 ret = drop_one_dir_item(trans, root, path, dir, di);
Josef Bacik36508602013-04-25 16:23:32 -04001042 if (ret)
1043 return ret;
Jan Schmidt5a1d7842012-08-17 14:04:41 -07001044 }
1045 btrfs_release_path(path);
1046
1047 /* look for a conflicing name */
1048 di = btrfs_lookup_dir_item(trans, root, path, btrfs_ino(dir),
1049 name, namelen, 0);
1050 if (di && !IS_ERR(di)) {
1051 ret = drop_one_dir_item(trans, root, path, dir, di);
Josef Bacik36508602013-04-25 16:23:32 -04001052 if (ret)
1053 return ret;
Jan Schmidt5a1d7842012-08-17 14:04:41 -07001054 }
1055 btrfs_release_path(path);
1056
1057 return 0;
1058}
Chris Masone02119d2008-09-05 16:13:11 -04001059
Mark Fashehf1863732012-08-08 11:32:27 -07001060static int extref_get_fields(struct extent_buffer *eb, unsigned long ref_ptr,
1061 u32 *namelen, char **name, u64 *index,
1062 u64 *parent_objectid)
1063{
1064 struct btrfs_inode_extref *extref;
1065
1066 extref = (struct btrfs_inode_extref *)ref_ptr;
1067
1068 *namelen = btrfs_inode_extref_name_len(eb, extref);
1069 *name = kmalloc(*namelen, GFP_NOFS);
1070 if (*name == NULL)
1071 return -ENOMEM;
1072
1073 read_extent_buffer(eb, *name, (unsigned long)&extref->name,
1074 *namelen);
1075
1076 *index = btrfs_inode_extref_index(eb, extref);
1077 if (parent_objectid)
1078 *parent_objectid = btrfs_inode_extref_parent(eb, extref);
1079
1080 return 0;
1081}
1082
1083static int ref_get_fields(struct extent_buffer *eb, unsigned long ref_ptr,
1084 u32 *namelen, char **name, u64 *index)
1085{
1086 struct btrfs_inode_ref *ref;
1087
1088 ref = (struct btrfs_inode_ref *)ref_ptr;
1089
1090 *namelen = btrfs_inode_ref_name_len(eb, ref);
1091 *name = kmalloc(*namelen, GFP_NOFS);
1092 if (*name == NULL)
1093 return -ENOMEM;
1094
1095 read_extent_buffer(eb, *name, (unsigned long)(ref + 1), *namelen);
1096
1097 *index = btrfs_inode_ref_index(eb, ref);
1098
1099 return 0;
1100}
1101
Chris Masone02119d2008-09-05 16:13:11 -04001102/*
1103 * replay one inode back reference item found in the log tree.
1104 * eb, slot and key refer to the buffer and key found in the log tree.
1105 * root is the destination we are replaying into, and path is for temp
1106 * use by this function. (it should be released on return).
1107 */
1108static noinline int add_inode_ref(struct btrfs_trans_handle *trans,
1109 struct btrfs_root *root,
1110 struct btrfs_root *log,
1111 struct btrfs_path *path,
1112 struct extent_buffer *eb, int slot,
1113 struct btrfs_key *key)
1114{
Geyslan G. Bem03b2f082013-10-11 15:35:45 -03001115 struct inode *dir = NULL;
1116 struct inode *inode = NULL;
Chris Masone02119d2008-09-05 16:13:11 -04001117 unsigned long ref_ptr;
1118 unsigned long ref_end;
Geyslan G. Bem03b2f082013-10-11 15:35:45 -03001119 char *name = NULL;
liubo34f3e4f2011-08-06 08:35:23 +00001120 int namelen;
1121 int ret;
liuboc622ae62011-03-26 08:01:12 -04001122 int search_done = 0;
Mark Fashehf1863732012-08-08 11:32:27 -07001123 int log_ref_ver = 0;
1124 u64 parent_objectid;
1125 u64 inode_objectid;
Chris Masonf46dbe3de2012-10-09 11:17:20 -04001126 u64 ref_index = 0;
Mark Fashehf1863732012-08-08 11:32:27 -07001127 int ref_struct_size;
1128
1129 ref_ptr = btrfs_item_ptr_offset(eb, slot);
1130 ref_end = ref_ptr + btrfs_item_size_nr(eb, slot);
1131
1132 if (key->type == BTRFS_INODE_EXTREF_KEY) {
1133 struct btrfs_inode_extref *r;
1134
1135 ref_struct_size = sizeof(struct btrfs_inode_extref);
1136 log_ref_ver = 1;
1137 r = (struct btrfs_inode_extref *)ref_ptr;
1138 parent_objectid = btrfs_inode_extref_parent(eb, r);
1139 } else {
1140 ref_struct_size = sizeof(struct btrfs_inode_ref);
1141 parent_objectid = key->offset;
1142 }
1143 inode_objectid = key->objectid;
Chris Masone02119d2008-09-05 16:13:11 -04001144
Chris Masone02119d2008-09-05 16:13:11 -04001145 /*
1146 * it is possible that we didn't log all the parent directories
1147 * for a given inode. If we don't find the dir, just don't
1148 * copy the back ref in. The link count fixup code will take
1149 * care of the rest
1150 */
Mark Fashehf1863732012-08-08 11:32:27 -07001151 dir = read_one_inode(root, parent_objectid);
Geyslan G. Bem03b2f082013-10-11 15:35:45 -03001152 if (!dir) {
1153 ret = -ENOENT;
1154 goto out;
1155 }
Chris Masone02119d2008-09-05 16:13:11 -04001156
Mark Fashehf1863732012-08-08 11:32:27 -07001157 inode = read_one_inode(root, inode_objectid);
Tsutomu Itohc00e9492011-04-28 09:10:23 +00001158 if (!inode) {
Geyslan G. Bem03b2f082013-10-11 15:35:45 -03001159 ret = -EIO;
1160 goto out;
Tsutomu Itohc00e9492011-04-28 09:10:23 +00001161 }
Chris Masone02119d2008-09-05 16:13:11 -04001162
Jan Schmidt5a1d7842012-08-17 14:04:41 -07001163 while (ref_ptr < ref_end) {
Mark Fashehf1863732012-08-08 11:32:27 -07001164 if (log_ref_ver) {
1165 ret = extref_get_fields(eb, ref_ptr, &namelen, &name,
1166 &ref_index, &parent_objectid);
1167 /*
1168 * parent object can change from one array
1169 * item to another.
1170 */
1171 if (!dir)
1172 dir = read_one_inode(root, parent_objectid);
Geyslan G. Bem03b2f082013-10-11 15:35:45 -03001173 if (!dir) {
1174 ret = -ENOENT;
1175 goto out;
1176 }
Mark Fashehf1863732012-08-08 11:32:27 -07001177 } else {
1178 ret = ref_get_fields(eb, ref_ptr, &namelen, &name,
1179 &ref_index);
1180 }
1181 if (ret)
Geyslan G. Bem03b2f082013-10-11 15:35:45 -03001182 goto out;
Chris Masone02119d2008-09-05 16:13:11 -04001183
Jan Schmidt5a1d7842012-08-17 14:04:41 -07001184 /* if we already have a perfect match, we're done */
1185 if (!inode_in_dir(root, path, btrfs_ino(dir), btrfs_ino(inode),
Mark Fashehf1863732012-08-08 11:32:27 -07001186 ref_index, name, namelen)) {
Jan Schmidt5a1d7842012-08-17 14:04:41 -07001187 /*
1188 * look for a conflicting back reference in the
1189 * metadata. if we find one we have to unlink that name
1190 * of the file before we add our new link. Later on, we
1191 * overwrite any existing back reference, and we don't
1192 * want to create dangling pointers in the directory.
1193 */
Chris Masone02119d2008-09-05 16:13:11 -04001194
Jan Schmidt5a1d7842012-08-17 14:04:41 -07001195 if (!search_done) {
1196 ret = __add_inode_ref(trans, root, path, log,
Mark Fashehf1863732012-08-08 11:32:27 -07001197 dir, inode, eb,
1198 inode_objectid,
1199 parent_objectid,
1200 ref_index, name, namelen,
Jan Schmidt5a1d7842012-08-17 14:04:41 -07001201 &search_done);
Geyslan G. Bem03b2f082013-10-11 15:35:45 -03001202 if (ret) {
1203 if (ret == 1)
1204 ret = 0;
Jan Schmidt5a1d7842012-08-17 14:04:41 -07001205 goto out;
Josef Bacik36508602013-04-25 16:23:32 -04001206 }
Chris Masone02119d2008-09-05 16:13:11 -04001207 }
Jan Schmidt5a1d7842012-08-17 14:04:41 -07001208
1209 /* insert our name */
1210 ret = btrfs_add_link(trans, dir, inode, name, namelen,
Mark Fashehf1863732012-08-08 11:32:27 -07001211 0, ref_index);
Josef Bacik36508602013-04-25 16:23:32 -04001212 if (ret)
1213 goto out;
Jan Schmidt5a1d7842012-08-17 14:04:41 -07001214
1215 btrfs_update_inode(trans, root, inode);
Chris Masone02119d2008-09-05 16:13:11 -04001216 }
liuboc622ae62011-03-26 08:01:12 -04001217
Mark Fashehf1863732012-08-08 11:32:27 -07001218 ref_ptr = (unsigned long)(ref_ptr + ref_struct_size) + namelen;
Jan Schmidt5a1d7842012-08-17 14:04:41 -07001219 kfree(name);
Geyslan G. Bem03b2f082013-10-11 15:35:45 -03001220 name = NULL;
Mark Fashehf1863732012-08-08 11:32:27 -07001221 if (log_ref_ver) {
1222 iput(dir);
1223 dir = NULL;
1224 }
Chris Masone02119d2008-09-05 16:13:11 -04001225 }
Chris Masone02119d2008-09-05 16:13:11 -04001226
1227 /* finally write the back reference in the inode */
1228 ret = overwrite_item(trans, root, path, eb, slot, key);
Jan Schmidt5a1d7842012-08-17 14:04:41 -07001229out:
David Sterbab3b4aa72011-04-21 01:20:15 +02001230 btrfs_release_path(path);
Geyslan G. Bem03b2f082013-10-11 15:35:45 -03001231 kfree(name);
Chris Masone02119d2008-09-05 16:13:11 -04001232 iput(dir);
1233 iput(inode);
Josef Bacik36508602013-04-25 16:23:32 -04001234 return ret;
Chris Masone02119d2008-09-05 16:13:11 -04001235}
1236
Yan, Zhengc71bf092009-11-12 09:34:40 +00001237static int insert_orphan_item(struct btrfs_trans_handle *trans,
1238 struct btrfs_root *root, u64 offset)
1239{
1240 int ret;
Kelley Nielsen3f870c22013-11-04 19:37:39 -08001241 ret = btrfs_find_item(root, NULL, BTRFS_ORPHAN_OBJECTID,
1242 offset, BTRFS_ORPHAN_ITEM_KEY, NULL);
Yan, Zhengc71bf092009-11-12 09:34:40 +00001243 if (ret > 0)
1244 ret = btrfs_insert_orphan_item(trans, root, offset);
1245 return ret;
1246}
1247
Mark Fashehf1863732012-08-08 11:32:27 -07001248static int count_inode_extrefs(struct btrfs_root *root,
1249 struct inode *inode, struct btrfs_path *path)
Chris Masone02119d2008-09-05 16:13:11 -04001250{
Mark Fashehf1863732012-08-08 11:32:27 -07001251 int ret = 0;
1252 int name_len;
1253 unsigned int nlink = 0;
1254 u32 item_size;
1255 u32 cur_offset = 0;
1256 u64 inode_objectid = btrfs_ino(inode);
1257 u64 offset = 0;
1258 unsigned long ptr;
1259 struct btrfs_inode_extref *extref;
1260 struct extent_buffer *leaf;
1261
1262 while (1) {
1263 ret = btrfs_find_one_extref(root, inode_objectid, offset, path,
1264 &extref, &offset);
1265 if (ret)
1266 break;
1267
1268 leaf = path->nodes[0];
1269 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1270 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
1271
1272 while (cur_offset < item_size) {
1273 extref = (struct btrfs_inode_extref *) (ptr + cur_offset);
1274 name_len = btrfs_inode_extref_name_len(leaf, extref);
1275
1276 nlink++;
1277
1278 cur_offset += name_len + sizeof(*extref);
1279 }
1280
1281 offset++;
1282 btrfs_release_path(path);
1283 }
1284 btrfs_release_path(path);
1285
1286 if (ret < 0)
1287 return ret;
1288 return nlink;
1289}
1290
1291static int count_inode_refs(struct btrfs_root *root,
1292 struct inode *inode, struct btrfs_path *path)
1293{
Chris Masone02119d2008-09-05 16:13:11 -04001294 int ret;
1295 struct btrfs_key key;
Mark Fashehf1863732012-08-08 11:32:27 -07001296 unsigned int nlink = 0;
Chris Masone02119d2008-09-05 16:13:11 -04001297 unsigned long ptr;
1298 unsigned long ptr_end;
1299 int name_len;
Li Zefan33345d012011-04-20 10:31:50 +08001300 u64 ino = btrfs_ino(inode);
Chris Masone02119d2008-09-05 16:13:11 -04001301
Li Zefan33345d012011-04-20 10:31:50 +08001302 key.objectid = ino;
Chris Masone02119d2008-09-05 16:13:11 -04001303 key.type = BTRFS_INODE_REF_KEY;
1304 key.offset = (u64)-1;
1305
Chris Masond3977122009-01-05 21:25:51 -05001306 while (1) {
Chris Masone02119d2008-09-05 16:13:11 -04001307 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1308 if (ret < 0)
1309 break;
1310 if (ret > 0) {
1311 if (path->slots[0] == 0)
1312 break;
1313 path->slots[0]--;
1314 }
Filipe David Borba Mananae93ae262013-10-14 22:49:11 +01001315process_slot:
Chris Masone02119d2008-09-05 16:13:11 -04001316 btrfs_item_key_to_cpu(path->nodes[0], &key,
1317 path->slots[0]);
Li Zefan33345d012011-04-20 10:31:50 +08001318 if (key.objectid != ino ||
Chris Masone02119d2008-09-05 16:13:11 -04001319 key.type != BTRFS_INODE_REF_KEY)
1320 break;
1321 ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
1322 ptr_end = ptr + btrfs_item_size_nr(path->nodes[0],
1323 path->slots[0]);
Chris Masond3977122009-01-05 21:25:51 -05001324 while (ptr < ptr_end) {
Chris Masone02119d2008-09-05 16:13:11 -04001325 struct btrfs_inode_ref *ref;
1326
1327 ref = (struct btrfs_inode_ref *)ptr;
1328 name_len = btrfs_inode_ref_name_len(path->nodes[0],
1329 ref);
1330 ptr = (unsigned long)(ref + 1) + name_len;
1331 nlink++;
1332 }
1333
1334 if (key.offset == 0)
1335 break;
Filipe David Borba Mananae93ae262013-10-14 22:49:11 +01001336 if (path->slots[0] > 0) {
1337 path->slots[0]--;
1338 goto process_slot;
1339 }
Chris Masone02119d2008-09-05 16:13:11 -04001340 key.offset--;
David Sterbab3b4aa72011-04-21 01:20:15 +02001341 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04001342 }
David Sterbab3b4aa72011-04-21 01:20:15 +02001343 btrfs_release_path(path);
Mark Fashehf1863732012-08-08 11:32:27 -07001344
1345 return nlink;
1346}
1347
1348/*
1349 * There are a few corners where the link count of the file can't
1350 * be properly maintained during replay. So, instead of adding
1351 * lots of complexity to the log code, we just scan the backrefs
1352 * for any file that has been through replay.
1353 *
1354 * The scan will update the link count on the inode to reflect the
1355 * number of back refs found. If it goes down to zero, the iput
1356 * will free the inode.
1357 */
1358static noinline int fixup_inode_link_count(struct btrfs_trans_handle *trans,
1359 struct btrfs_root *root,
1360 struct inode *inode)
1361{
1362 struct btrfs_path *path;
1363 int ret;
1364 u64 nlink = 0;
1365 u64 ino = btrfs_ino(inode);
1366
1367 path = btrfs_alloc_path();
1368 if (!path)
1369 return -ENOMEM;
1370
1371 ret = count_inode_refs(root, inode, path);
1372 if (ret < 0)
1373 goto out;
1374
1375 nlink = ret;
1376
1377 ret = count_inode_extrefs(root, inode, path);
1378 if (ret == -ENOENT)
1379 ret = 0;
1380
1381 if (ret < 0)
1382 goto out;
1383
1384 nlink += ret;
1385
1386 ret = 0;
1387
Chris Masone02119d2008-09-05 16:13:11 -04001388 if (nlink != inode->i_nlink) {
Miklos Szeredibfe86842011-10-28 14:13:29 +02001389 set_nlink(inode, nlink);
Chris Masone02119d2008-09-05 16:13:11 -04001390 btrfs_update_inode(trans, root, inode);
1391 }
Chris Mason8d5bf1c2008-09-11 15:51:21 -04001392 BTRFS_I(inode)->index_cnt = (u64)-1;
Chris Masone02119d2008-09-05 16:13:11 -04001393
Yan, Zhengc71bf092009-11-12 09:34:40 +00001394 if (inode->i_nlink == 0) {
1395 if (S_ISDIR(inode->i_mode)) {
1396 ret = replay_dir_deletes(trans, root, NULL, path,
Li Zefan33345d012011-04-20 10:31:50 +08001397 ino, 1);
Josef Bacik36508602013-04-25 16:23:32 -04001398 if (ret)
1399 goto out;
Yan, Zhengc71bf092009-11-12 09:34:40 +00001400 }
Li Zefan33345d012011-04-20 10:31:50 +08001401 ret = insert_orphan_item(trans, root, ino);
Chris Mason12fcfd22009-03-24 10:24:20 -04001402 }
Chris Mason12fcfd22009-03-24 10:24:20 -04001403
Mark Fashehf1863732012-08-08 11:32:27 -07001404out:
1405 btrfs_free_path(path);
1406 return ret;
Chris Masone02119d2008-09-05 16:13:11 -04001407}
1408
1409static noinline int fixup_inode_link_counts(struct btrfs_trans_handle *trans,
1410 struct btrfs_root *root,
1411 struct btrfs_path *path)
1412{
1413 int ret;
1414 struct btrfs_key key;
1415 struct inode *inode;
1416
1417 key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID;
1418 key.type = BTRFS_ORPHAN_ITEM_KEY;
1419 key.offset = (u64)-1;
Chris Masond3977122009-01-05 21:25:51 -05001420 while (1) {
Chris Masone02119d2008-09-05 16:13:11 -04001421 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1422 if (ret < 0)
1423 break;
1424
1425 if (ret == 1) {
1426 if (path->slots[0] == 0)
1427 break;
1428 path->slots[0]--;
1429 }
1430
1431 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1432 if (key.objectid != BTRFS_TREE_LOG_FIXUP_OBJECTID ||
1433 key.type != BTRFS_ORPHAN_ITEM_KEY)
1434 break;
1435
1436 ret = btrfs_del_item(trans, root, path);
Tsutomu Itoh65a246c2011-05-19 04:37:44 +00001437 if (ret)
1438 goto out;
Chris Masone02119d2008-09-05 16:13:11 -04001439
David Sterbab3b4aa72011-04-21 01:20:15 +02001440 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04001441 inode = read_one_inode(root, key.offset);
Tsutomu Itohc00e9492011-04-28 09:10:23 +00001442 if (!inode)
1443 return -EIO;
Chris Masone02119d2008-09-05 16:13:11 -04001444
1445 ret = fixup_inode_link_count(trans, root, inode);
Chris Masone02119d2008-09-05 16:13:11 -04001446 iput(inode);
Josef Bacik36508602013-04-25 16:23:32 -04001447 if (ret)
1448 goto out;
Chris Masone02119d2008-09-05 16:13:11 -04001449
Chris Mason12fcfd22009-03-24 10:24:20 -04001450 /*
1451 * fixup on a directory may create new entries,
1452 * make sure we always look for the highset possible
1453 * offset
1454 */
1455 key.offset = (u64)-1;
Chris Masone02119d2008-09-05 16:13:11 -04001456 }
Tsutomu Itoh65a246c2011-05-19 04:37:44 +00001457 ret = 0;
1458out:
David Sterbab3b4aa72011-04-21 01:20:15 +02001459 btrfs_release_path(path);
Tsutomu Itoh65a246c2011-05-19 04:37:44 +00001460 return ret;
Chris Masone02119d2008-09-05 16:13:11 -04001461}
1462
1463
1464/*
1465 * record a given inode in the fixup dir so we can check its link
1466 * count when replay is done. The link count is incremented here
1467 * so the inode won't go away until we check it
1468 */
1469static noinline int link_to_fixup_dir(struct btrfs_trans_handle *trans,
1470 struct btrfs_root *root,
1471 struct btrfs_path *path,
1472 u64 objectid)
1473{
1474 struct btrfs_key key;
1475 int ret = 0;
1476 struct inode *inode;
1477
1478 inode = read_one_inode(root, objectid);
Tsutomu Itohc00e9492011-04-28 09:10:23 +00001479 if (!inode)
1480 return -EIO;
Chris Masone02119d2008-09-05 16:13:11 -04001481
1482 key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID;
1483 btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
1484 key.offset = objectid;
1485
1486 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1487
David Sterbab3b4aa72011-04-21 01:20:15 +02001488 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04001489 if (ret == 0) {
Josef Bacik9bf7a482013-03-01 13:35:47 -05001490 if (!inode->i_nlink)
1491 set_nlink(inode, 1);
1492 else
Zach Brown8b558c52013-10-16 12:10:34 -07001493 inc_nlink(inode);
Tsutomu Itohb9959292012-06-25 21:25:22 -06001494 ret = btrfs_update_inode(trans, root, inode);
Chris Masone02119d2008-09-05 16:13:11 -04001495 } else if (ret == -EEXIST) {
1496 ret = 0;
1497 } else {
Josef Bacik36508602013-04-25 16:23:32 -04001498 BUG(); /* Logic Error */
Chris Masone02119d2008-09-05 16:13:11 -04001499 }
1500 iput(inode);
1501
1502 return ret;
1503}
1504
1505/*
1506 * when replaying the log for a directory, we only insert names
1507 * for inodes that actually exist. This means an fsync on a directory
1508 * does not implicitly fsync all the new files in it
1509 */
1510static noinline int insert_one_name(struct btrfs_trans_handle *trans,
1511 struct btrfs_root *root,
1512 struct btrfs_path *path,
1513 u64 dirid, u64 index,
1514 char *name, int name_len, u8 type,
1515 struct btrfs_key *location)
1516{
1517 struct inode *inode;
1518 struct inode *dir;
1519 int ret;
1520
1521 inode = read_one_inode(root, location->objectid);
1522 if (!inode)
1523 return -ENOENT;
1524
1525 dir = read_one_inode(root, dirid);
1526 if (!dir) {
1527 iput(inode);
1528 return -EIO;
1529 }
Josef Bacikd5554382013-09-11 14:17:00 -04001530
Chris Masone02119d2008-09-05 16:13:11 -04001531 ret = btrfs_add_link(trans, dir, inode, name, name_len, 1, index);
1532
1533 /* FIXME, put inode into FIXUP list */
1534
1535 iput(inode);
1536 iput(dir);
1537 return ret;
1538}
1539
1540/*
1541 * take a single entry in a log directory item and replay it into
1542 * the subvolume.
1543 *
1544 * if a conflicting item exists in the subdirectory already,
1545 * the inode it points to is unlinked and put into the link count
1546 * fix up tree.
1547 *
1548 * If a name from the log points to a file or directory that does
1549 * not exist in the FS, it is skipped. fsyncs on directories
1550 * do not force down inodes inside that directory, just changes to the
1551 * names or unlinks in a directory.
1552 */
1553static noinline int replay_one_name(struct btrfs_trans_handle *trans,
1554 struct btrfs_root *root,
1555 struct btrfs_path *path,
1556 struct extent_buffer *eb,
1557 struct btrfs_dir_item *di,
1558 struct btrfs_key *key)
1559{
1560 char *name;
1561 int name_len;
1562 struct btrfs_dir_item *dst_di;
1563 struct btrfs_key found_key;
1564 struct btrfs_key log_key;
1565 struct inode *dir;
Chris Masone02119d2008-09-05 16:13:11 -04001566 u8 log_type;
Chris Mason4bef0842008-09-08 11:18:08 -04001567 int exists;
Josef Bacik36508602013-04-25 16:23:32 -04001568 int ret = 0;
Josef Bacikd5554382013-09-11 14:17:00 -04001569 bool update_size = (key->type == BTRFS_DIR_INDEX_KEY);
Chris Masone02119d2008-09-05 16:13:11 -04001570
1571 dir = read_one_inode(root, key->objectid);
Tsutomu Itohc00e9492011-04-28 09:10:23 +00001572 if (!dir)
1573 return -EIO;
Chris Masone02119d2008-09-05 16:13:11 -04001574
1575 name_len = btrfs_dir_name_len(eb, di);
1576 name = kmalloc(name_len, GFP_NOFS);
Filipe David Borba Manana2bac3252013-08-04 19:58:57 +01001577 if (!name) {
1578 ret = -ENOMEM;
1579 goto out;
1580 }
liubo2a29edc2011-01-26 06:22:08 +00001581
Chris Masone02119d2008-09-05 16:13:11 -04001582 log_type = btrfs_dir_type(eb, di);
1583 read_extent_buffer(eb, name, (unsigned long)(di + 1),
1584 name_len);
1585
1586 btrfs_dir_item_key_to_cpu(eb, di, &log_key);
Chris Mason4bef0842008-09-08 11:18:08 -04001587 exists = btrfs_lookup_inode(trans, root, path, &log_key, 0);
1588 if (exists == 0)
1589 exists = 1;
1590 else
1591 exists = 0;
David Sterbab3b4aa72011-04-21 01:20:15 +02001592 btrfs_release_path(path);
Chris Mason4bef0842008-09-08 11:18:08 -04001593
Chris Masone02119d2008-09-05 16:13:11 -04001594 if (key->type == BTRFS_DIR_ITEM_KEY) {
1595 dst_di = btrfs_lookup_dir_item(trans, root, path, key->objectid,
1596 name, name_len, 1);
Chris Masond3977122009-01-05 21:25:51 -05001597 } else if (key->type == BTRFS_DIR_INDEX_KEY) {
Chris Masone02119d2008-09-05 16:13:11 -04001598 dst_di = btrfs_lookup_dir_index_item(trans, root, path,
1599 key->objectid,
1600 key->offset, name,
1601 name_len, 1);
1602 } else {
Josef Bacik36508602013-04-25 16:23:32 -04001603 /* Corruption */
1604 ret = -EINVAL;
1605 goto out;
Chris Masone02119d2008-09-05 16:13:11 -04001606 }
David Sterbac7040052011-04-19 18:00:01 +02001607 if (IS_ERR_OR_NULL(dst_di)) {
Chris Masone02119d2008-09-05 16:13:11 -04001608 /* we need a sequence number to insert, so we only
1609 * do inserts for the BTRFS_DIR_INDEX_KEY types
1610 */
1611 if (key->type != BTRFS_DIR_INDEX_KEY)
1612 goto out;
1613 goto insert;
1614 }
1615
1616 btrfs_dir_item_key_to_cpu(path->nodes[0], dst_di, &found_key);
1617 /* the existing item matches the logged item */
1618 if (found_key.objectid == log_key.objectid &&
1619 found_key.type == log_key.type &&
1620 found_key.offset == log_key.offset &&
1621 btrfs_dir_type(path->nodes[0], dst_di) == log_type) {
1622 goto out;
1623 }
1624
1625 /*
1626 * don't drop the conflicting directory entry if the inode
1627 * for the new entry doesn't exist
1628 */
Chris Mason4bef0842008-09-08 11:18:08 -04001629 if (!exists)
Chris Masone02119d2008-09-05 16:13:11 -04001630 goto out;
1631
Chris Masone02119d2008-09-05 16:13:11 -04001632 ret = drop_one_dir_item(trans, root, path, dir, dst_di);
Josef Bacik36508602013-04-25 16:23:32 -04001633 if (ret)
1634 goto out;
Chris Masone02119d2008-09-05 16:13:11 -04001635
1636 if (key->type == BTRFS_DIR_INDEX_KEY)
1637 goto insert;
1638out:
David Sterbab3b4aa72011-04-21 01:20:15 +02001639 btrfs_release_path(path);
Josef Bacikd5554382013-09-11 14:17:00 -04001640 if (!ret && update_size) {
1641 btrfs_i_size_write(dir, dir->i_size + name_len * 2);
1642 ret = btrfs_update_inode(trans, root, dir);
1643 }
Chris Masone02119d2008-09-05 16:13:11 -04001644 kfree(name);
1645 iput(dir);
Josef Bacik36508602013-04-25 16:23:32 -04001646 return ret;
Chris Masone02119d2008-09-05 16:13:11 -04001647
1648insert:
David Sterbab3b4aa72011-04-21 01:20:15 +02001649 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04001650 ret = insert_one_name(trans, root, path, key->objectid, key->offset,
1651 name, name_len, log_type, &log_key);
Josef Bacik36508602013-04-25 16:23:32 -04001652 if (ret && ret != -ENOENT)
1653 goto out;
Josef Bacikd5554382013-09-11 14:17:00 -04001654 update_size = false;
Josef Bacik36508602013-04-25 16:23:32 -04001655 ret = 0;
Chris Masone02119d2008-09-05 16:13:11 -04001656 goto out;
1657}
1658
1659/*
1660 * find all the names in a directory item and reconcile them into
1661 * the subvolume. Only BTRFS_DIR_ITEM_KEY types will have more than
1662 * one name in a directory item, but the same code gets used for
1663 * both directory index types
1664 */
1665static noinline int replay_one_dir_item(struct btrfs_trans_handle *trans,
1666 struct btrfs_root *root,
1667 struct btrfs_path *path,
1668 struct extent_buffer *eb, int slot,
1669 struct btrfs_key *key)
1670{
1671 int ret;
1672 u32 item_size = btrfs_item_size_nr(eb, slot);
1673 struct btrfs_dir_item *di;
1674 int name_len;
1675 unsigned long ptr;
1676 unsigned long ptr_end;
1677
1678 ptr = btrfs_item_ptr_offset(eb, slot);
1679 ptr_end = ptr + item_size;
Chris Masond3977122009-01-05 21:25:51 -05001680 while (ptr < ptr_end) {
Chris Masone02119d2008-09-05 16:13:11 -04001681 di = (struct btrfs_dir_item *)ptr;
Josef Bacik22a94d42011-03-16 16:47:17 -04001682 if (verify_dir_item(root, eb, di))
1683 return -EIO;
Chris Masone02119d2008-09-05 16:13:11 -04001684 name_len = btrfs_dir_name_len(eb, di);
1685 ret = replay_one_name(trans, root, path, eb, di, key);
Josef Bacik36508602013-04-25 16:23:32 -04001686 if (ret)
1687 return ret;
Chris Masone02119d2008-09-05 16:13:11 -04001688 ptr = (unsigned long)(di + 1);
1689 ptr += name_len;
1690 }
1691 return 0;
1692}
1693
1694/*
1695 * directory replay has two parts. There are the standard directory
1696 * items in the log copied from the subvolume, and range items
1697 * created in the log while the subvolume was logged.
1698 *
1699 * The range items tell us which parts of the key space the log
1700 * is authoritative for. During replay, if a key in the subvolume
1701 * directory is in a logged range item, but not actually in the log
1702 * that means it was deleted from the directory before the fsync
1703 * and should be removed.
1704 */
1705static noinline int find_dir_range(struct btrfs_root *root,
1706 struct btrfs_path *path,
1707 u64 dirid, int key_type,
1708 u64 *start_ret, u64 *end_ret)
1709{
1710 struct btrfs_key key;
1711 u64 found_end;
1712 struct btrfs_dir_log_item *item;
1713 int ret;
1714 int nritems;
1715
1716 if (*start_ret == (u64)-1)
1717 return 1;
1718
1719 key.objectid = dirid;
1720 key.type = key_type;
1721 key.offset = *start_ret;
1722
1723 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1724 if (ret < 0)
1725 goto out;
1726 if (ret > 0) {
1727 if (path->slots[0] == 0)
1728 goto out;
1729 path->slots[0]--;
1730 }
1731 if (ret != 0)
1732 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1733
1734 if (key.type != key_type || key.objectid != dirid) {
1735 ret = 1;
1736 goto next;
1737 }
1738 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
1739 struct btrfs_dir_log_item);
1740 found_end = btrfs_dir_log_end(path->nodes[0], item);
1741
1742 if (*start_ret >= key.offset && *start_ret <= found_end) {
1743 ret = 0;
1744 *start_ret = key.offset;
1745 *end_ret = found_end;
1746 goto out;
1747 }
1748 ret = 1;
1749next:
1750 /* check the next slot in the tree to see if it is a valid item */
1751 nritems = btrfs_header_nritems(path->nodes[0]);
1752 if (path->slots[0] >= nritems) {
1753 ret = btrfs_next_leaf(root, path);
1754 if (ret)
1755 goto out;
1756 } else {
1757 path->slots[0]++;
1758 }
1759
1760 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1761
1762 if (key.type != key_type || key.objectid != dirid) {
1763 ret = 1;
1764 goto out;
1765 }
1766 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
1767 struct btrfs_dir_log_item);
1768 found_end = btrfs_dir_log_end(path->nodes[0], item);
1769 *start_ret = key.offset;
1770 *end_ret = found_end;
1771 ret = 0;
1772out:
David Sterbab3b4aa72011-04-21 01:20:15 +02001773 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04001774 return ret;
1775}
1776
1777/*
1778 * this looks for a given directory item in the log. If the directory
1779 * item is not in the log, the item is removed and the inode it points
1780 * to is unlinked
1781 */
1782static noinline int check_item_in_log(struct btrfs_trans_handle *trans,
1783 struct btrfs_root *root,
1784 struct btrfs_root *log,
1785 struct btrfs_path *path,
1786 struct btrfs_path *log_path,
1787 struct inode *dir,
1788 struct btrfs_key *dir_key)
1789{
1790 int ret;
1791 struct extent_buffer *eb;
1792 int slot;
1793 u32 item_size;
1794 struct btrfs_dir_item *di;
1795 struct btrfs_dir_item *log_di;
1796 int name_len;
1797 unsigned long ptr;
1798 unsigned long ptr_end;
1799 char *name;
1800 struct inode *inode;
1801 struct btrfs_key location;
1802
1803again:
1804 eb = path->nodes[0];
1805 slot = path->slots[0];
1806 item_size = btrfs_item_size_nr(eb, slot);
1807 ptr = btrfs_item_ptr_offset(eb, slot);
1808 ptr_end = ptr + item_size;
Chris Masond3977122009-01-05 21:25:51 -05001809 while (ptr < ptr_end) {
Chris Masone02119d2008-09-05 16:13:11 -04001810 di = (struct btrfs_dir_item *)ptr;
Josef Bacik22a94d42011-03-16 16:47:17 -04001811 if (verify_dir_item(root, eb, di)) {
1812 ret = -EIO;
1813 goto out;
1814 }
1815
Chris Masone02119d2008-09-05 16:13:11 -04001816 name_len = btrfs_dir_name_len(eb, di);
1817 name = kmalloc(name_len, GFP_NOFS);
1818 if (!name) {
1819 ret = -ENOMEM;
1820 goto out;
1821 }
1822 read_extent_buffer(eb, name, (unsigned long)(di + 1),
1823 name_len);
1824 log_di = NULL;
Chris Mason12fcfd22009-03-24 10:24:20 -04001825 if (log && dir_key->type == BTRFS_DIR_ITEM_KEY) {
Chris Masone02119d2008-09-05 16:13:11 -04001826 log_di = btrfs_lookup_dir_item(trans, log, log_path,
1827 dir_key->objectid,
1828 name, name_len, 0);
Chris Mason12fcfd22009-03-24 10:24:20 -04001829 } else if (log && dir_key->type == BTRFS_DIR_INDEX_KEY) {
Chris Masone02119d2008-09-05 16:13:11 -04001830 log_di = btrfs_lookup_dir_index_item(trans, log,
1831 log_path,
1832 dir_key->objectid,
1833 dir_key->offset,
1834 name, name_len, 0);
1835 }
Filipe David Borba Manana269d0402013-10-28 17:39:21 +00001836 if (!log_di || (IS_ERR(log_di) && PTR_ERR(log_di) == -ENOENT)) {
Chris Masone02119d2008-09-05 16:13:11 -04001837 btrfs_dir_item_key_to_cpu(eb, di, &location);
David Sterbab3b4aa72011-04-21 01:20:15 +02001838 btrfs_release_path(path);
1839 btrfs_release_path(log_path);
Chris Masone02119d2008-09-05 16:13:11 -04001840 inode = read_one_inode(root, location.objectid);
Tsutomu Itohc00e9492011-04-28 09:10:23 +00001841 if (!inode) {
1842 kfree(name);
1843 return -EIO;
1844 }
Chris Masone02119d2008-09-05 16:13:11 -04001845
1846 ret = link_to_fixup_dir(trans, root,
1847 path, location.objectid);
Josef Bacik36508602013-04-25 16:23:32 -04001848 if (ret) {
1849 kfree(name);
1850 iput(inode);
1851 goto out;
1852 }
1853
Zach Brown8b558c52013-10-16 12:10:34 -07001854 inc_nlink(inode);
Chris Masone02119d2008-09-05 16:13:11 -04001855 ret = btrfs_unlink_inode(trans, root, dir, inode,
1856 name, name_len);
Josef Bacik36508602013-04-25 16:23:32 -04001857 if (!ret)
Filipe David Borba Mananaada9af22013-08-05 09:25:47 +01001858 ret = btrfs_run_delayed_items(trans, root);
Chris Masone02119d2008-09-05 16:13:11 -04001859 kfree(name);
1860 iput(inode);
Josef Bacik36508602013-04-25 16:23:32 -04001861 if (ret)
1862 goto out;
Chris Masone02119d2008-09-05 16:13:11 -04001863
1864 /* there might still be more names under this key
1865 * check and repeat if required
1866 */
1867 ret = btrfs_search_slot(NULL, root, dir_key, path,
1868 0, 0);
1869 if (ret == 0)
1870 goto again;
1871 ret = 0;
1872 goto out;
Filipe David Borba Manana269d0402013-10-28 17:39:21 +00001873 } else if (IS_ERR(log_di)) {
1874 kfree(name);
1875 return PTR_ERR(log_di);
Chris Masone02119d2008-09-05 16:13:11 -04001876 }
David Sterbab3b4aa72011-04-21 01:20:15 +02001877 btrfs_release_path(log_path);
Chris Masone02119d2008-09-05 16:13:11 -04001878 kfree(name);
1879
1880 ptr = (unsigned long)(di + 1);
1881 ptr += name_len;
1882 }
1883 ret = 0;
1884out:
David Sterbab3b4aa72011-04-21 01:20:15 +02001885 btrfs_release_path(path);
1886 btrfs_release_path(log_path);
Chris Masone02119d2008-09-05 16:13:11 -04001887 return ret;
1888}
1889
1890/*
1891 * deletion replay happens before we copy any new directory items
1892 * out of the log or out of backreferences from inodes. It
1893 * scans the log to find ranges of keys that log is authoritative for,
1894 * and then scans the directory to find items in those ranges that are
1895 * not present in the log.
1896 *
1897 * Anything we don't find in the log is unlinked and removed from the
1898 * directory.
1899 */
1900static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans,
1901 struct btrfs_root *root,
1902 struct btrfs_root *log,
1903 struct btrfs_path *path,
Chris Mason12fcfd22009-03-24 10:24:20 -04001904 u64 dirid, int del_all)
Chris Masone02119d2008-09-05 16:13:11 -04001905{
1906 u64 range_start;
1907 u64 range_end;
1908 int key_type = BTRFS_DIR_LOG_ITEM_KEY;
1909 int ret = 0;
1910 struct btrfs_key dir_key;
1911 struct btrfs_key found_key;
1912 struct btrfs_path *log_path;
1913 struct inode *dir;
1914
1915 dir_key.objectid = dirid;
1916 dir_key.type = BTRFS_DIR_ITEM_KEY;
1917 log_path = btrfs_alloc_path();
1918 if (!log_path)
1919 return -ENOMEM;
1920
1921 dir = read_one_inode(root, dirid);
1922 /* it isn't an error if the inode isn't there, that can happen
1923 * because we replay the deletes before we copy in the inode item
1924 * from the log
1925 */
1926 if (!dir) {
1927 btrfs_free_path(log_path);
1928 return 0;
1929 }
1930again:
1931 range_start = 0;
1932 range_end = 0;
Chris Masond3977122009-01-05 21:25:51 -05001933 while (1) {
Chris Mason12fcfd22009-03-24 10:24:20 -04001934 if (del_all)
1935 range_end = (u64)-1;
1936 else {
1937 ret = find_dir_range(log, path, dirid, key_type,
1938 &range_start, &range_end);
1939 if (ret != 0)
1940 break;
1941 }
Chris Masone02119d2008-09-05 16:13:11 -04001942
1943 dir_key.offset = range_start;
Chris Masond3977122009-01-05 21:25:51 -05001944 while (1) {
Chris Masone02119d2008-09-05 16:13:11 -04001945 int nritems;
1946 ret = btrfs_search_slot(NULL, root, &dir_key, path,
1947 0, 0);
1948 if (ret < 0)
1949 goto out;
1950
1951 nritems = btrfs_header_nritems(path->nodes[0]);
1952 if (path->slots[0] >= nritems) {
1953 ret = btrfs_next_leaf(root, path);
1954 if (ret)
1955 break;
1956 }
1957 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1958 path->slots[0]);
1959 if (found_key.objectid != dirid ||
1960 found_key.type != dir_key.type)
1961 goto next_type;
1962
1963 if (found_key.offset > range_end)
1964 break;
1965
1966 ret = check_item_in_log(trans, root, log, path,
Chris Mason12fcfd22009-03-24 10:24:20 -04001967 log_path, dir,
1968 &found_key);
Josef Bacik36508602013-04-25 16:23:32 -04001969 if (ret)
1970 goto out;
Chris Masone02119d2008-09-05 16:13:11 -04001971 if (found_key.offset == (u64)-1)
1972 break;
1973 dir_key.offset = found_key.offset + 1;
1974 }
David Sterbab3b4aa72011-04-21 01:20:15 +02001975 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04001976 if (range_end == (u64)-1)
1977 break;
1978 range_start = range_end + 1;
1979 }
1980
1981next_type:
1982 ret = 0;
1983 if (key_type == BTRFS_DIR_LOG_ITEM_KEY) {
1984 key_type = BTRFS_DIR_LOG_INDEX_KEY;
1985 dir_key.type = BTRFS_DIR_INDEX_KEY;
David Sterbab3b4aa72011-04-21 01:20:15 +02001986 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04001987 goto again;
1988 }
1989out:
David Sterbab3b4aa72011-04-21 01:20:15 +02001990 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04001991 btrfs_free_path(log_path);
1992 iput(dir);
1993 return ret;
1994}
1995
1996/*
1997 * the process_func used to replay items from the log tree. This
1998 * gets called in two different stages. The first stage just looks
1999 * for inodes and makes sure they are all copied into the subvolume.
2000 *
2001 * The second stage copies all the other item types from the log into
2002 * the subvolume. The two stage approach is slower, but gets rid of
2003 * lots of complexity around inodes referencing other inodes that exist
2004 * only in the log (references come from either directory items or inode
2005 * back refs).
2006 */
2007static int replay_one_buffer(struct btrfs_root *log, struct extent_buffer *eb,
2008 struct walk_control *wc, u64 gen)
2009{
2010 int nritems;
2011 struct btrfs_path *path;
2012 struct btrfs_root *root = wc->replay_dest;
2013 struct btrfs_key key;
Chris Masone02119d2008-09-05 16:13:11 -04002014 int level;
2015 int i;
2016 int ret;
2017
Tsutomu Itoh018642a2012-05-29 18:10:13 +09002018 ret = btrfs_read_buffer(eb, gen);
2019 if (ret)
2020 return ret;
Chris Masone02119d2008-09-05 16:13:11 -04002021
2022 level = btrfs_header_level(eb);
2023
2024 if (level != 0)
2025 return 0;
2026
2027 path = btrfs_alloc_path();
Mark Fasheh1e5063d2011-07-12 10:46:06 -07002028 if (!path)
2029 return -ENOMEM;
Chris Masone02119d2008-09-05 16:13:11 -04002030
2031 nritems = btrfs_header_nritems(eb);
2032 for (i = 0; i < nritems; i++) {
2033 btrfs_item_key_to_cpu(eb, &key, i);
Chris Masone02119d2008-09-05 16:13:11 -04002034
2035 /* inode keys are done during the first stage */
2036 if (key.type == BTRFS_INODE_ITEM_KEY &&
2037 wc->stage == LOG_WALK_REPLAY_INODES) {
Chris Masone02119d2008-09-05 16:13:11 -04002038 struct btrfs_inode_item *inode_item;
2039 u32 mode;
2040
2041 inode_item = btrfs_item_ptr(eb, i,
2042 struct btrfs_inode_item);
2043 mode = btrfs_inode_mode(eb, inode_item);
2044 if (S_ISDIR(mode)) {
2045 ret = replay_dir_deletes(wc->trans,
Chris Mason12fcfd22009-03-24 10:24:20 -04002046 root, log, path, key.objectid, 0);
Josef Bacikb50c6e22013-04-25 15:55:30 -04002047 if (ret)
2048 break;
Chris Masone02119d2008-09-05 16:13:11 -04002049 }
2050 ret = overwrite_item(wc->trans, root, path,
2051 eb, i, &key);
Josef Bacikb50c6e22013-04-25 15:55:30 -04002052 if (ret)
2053 break;
Chris Masone02119d2008-09-05 16:13:11 -04002054
Yan, Zhengc71bf092009-11-12 09:34:40 +00002055 /* for regular files, make sure corresponding
2056 * orhpan item exist. extents past the new EOF
2057 * will be truncated later by orphan cleanup.
Chris Masone02119d2008-09-05 16:13:11 -04002058 */
2059 if (S_ISREG(mode)) {
Yan, Zhengc71bf092009-11-12 09:34:40 +00002060 ret = insert_orphan_item(wc->trans, root,
2061 key.objectid);
Josef Bacikb50c6e22013-04-25 15:55:30 -04002062 if (ret)
2063 break;
Chris Masone02119d2008-09-05 16:13:11 -04002064 }
Yan, Zhengc71bf092009-11-12 09:34:40 +00002065
Chris Masone02119d2008-09-05 16:13:11 -04002066 ret = link_to_fixup_dir(wc->trans, root,
2067 path, key.objectid);
Josef Bacikb50c6e22013-04-25 15:55:30 -04002068 if (ret)
2069 break;
Chris Masone02119d2008-09-05 16:13:11 -04002070 }
Josef Bacikdd8e7212013-09-11 11:57:23 -04002071
2072 if (key.type == BTRFS_DIR_INDEX_KEY &&
2073 wc->stage == LOG_WALK_REPLAY_DIR_INDEX) {
2074 ret = replay_one_dir_item(wc->trans, root, path,
2075 eb, i, &key);
2076 if (ret)
2077 break;
2078 }
2079
Chris Masone02119d2008-09-05 16:13:11 -04002080 if (wc->stage < LOG_WALK_REPLAY_ALL)
2081 continue;
2082
2083 /* these keys are simply copied */
2084 if (key.type == BTRFS_XATTR_ITEM_KEY) {
2085 ret = overwrite_item(wc->trans, root, path,
2086 eb, i, &key);
Josef Bacikb50c6e22013-04-25 15:55:30 -04002087 if (ret)
2088 break;
Liu Bo2da1c662013-05-26 13:50:29 +00002089 } else if (key.type == BTRFS_INODE_REF_KEY ||
2090 key.type == BTRFS_INODE_EXTREF_KEY) {
Mark Fashehf1863732012-08-08 11:32:27 -07002091 ret = add_inode_ref(wc->trans, root, log, path,
2092 eb, i, &key);
Josef Bacikb50c6e22013-04-25 15:55:30 -04002093 if (ret && ret != -ENOENT)
2094 break;
2095 ret = 0;
Chris Masone02119d2008-09-05 16:13:11 -04002096 } else if (key.type == BTRFS_EXTENT_DATA_KEY) {
2097 ret = replay_one_extent(wc->trans, root, path,
2098 eb, i, &key);
Josef Bacikb50c6e22013-04-25 15:55:30 -04002099 if (ret)
2100 break;
Josef Bacikdd8e7212013-09-11 11:57:23 -04002101 } else if (key.type == BTRFS_DIR_ITEM_KEY) {
Chris Masone02119d2008-09-05 16:13:11 -04002102 ret = replay_one_dir_item(wc->trans, root, path,
2103 eb, i, &key);
Josef Bacikb50c6e22013-04-25 15:55:30 -04002104 if (ret)
2105 break;
Chris Masone02119d2008-09-05 16:13:11 -04002106 }
2107 }
2108 btrfs_free_path(path);
Josef Bacikb50c6e22013-04-25 15:55:30 -04002109 return ret;
Chris Masone02119d2008-09-05 16:13:11 -04002110}
2111
Chris Masond3977122009-01-05 21:25:51 -05002112static noinline int walk_down_log_tree(struct btrfs_trans_handle *trans,
Chris Masone02119d2008-09-05 16:13:11 -04002113 struct btrfs_root *root,
2114 struct btrfs_path *path, int *level,
2115 struct walk_control *wc)
2116{
2117 u64 root_owner;
Chris Masone02119d2008-09-05 16:13:11 -04002118 u64 bytenr;
2119 u64 ptr_gen;
2120 struct extent_buffer *next;
2121 struct extent_buffer *cur;
2122 struct extent_buffer *parent;
2123 u32 blocksize;
2124 int ret = 0;
2125
2126 WARN_ON(*level < 0);
2127 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2128
Chris Masond3977122009-01-05 21:25:51 -05002129 while (*level > 0) {
Chris Masone02119d2008-09-05 16:13:11 -04002130 WARN_ON(*level < 0);
2131 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2132 cur = path->nodes[*level];
2133
Dulshani Gunawardhanafae7f212013-10-31 10:30:08 +05302134 WARN_ON(btrfs_header_level(cur) != *level);
Chris Masone02119d2008-09-05 16:13:11 -04002135
2136 if (path->slots[*level] >=
2137 btrfs_header_nritems(cur))
2138 break;
2139
2140 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
2141 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
2142 blocksize = btrfs_level_size(root, *level - 1);
2143
2144 parent = path->nodes[*level];
2145 root_owner = btrfs_header_owner(parent);
Chris Masone02119d2008-09-05 16:13:11 -04002146
2147 next = btrfs_find_create_tree_block(root, bytenr, blocksize);
liubo2a29edc2011-01-26 06:22:08 +00002148 if (!next)
2149 return -ENOMEM;
Chris Masone02119d2008-09-05 16:13:11 -04002150
Chris Masone02119d2008-09-05 16:13:11 -04002151 if (*level == 1) {
Mark Fasheh1e5063d2011-07-12 10:46:06 -07002152 ret = wc->process_func(root, next, wc, ptr_gen);
Josef Bacikb50c6e22013-04-25 15:55:30 -04002153 if (ret) {
2154 free_extent_buffer(next);
Mark Fasheh1e5063d2011-07-12 10:46:06 -07002155 return ret;
Josef Bacikb50c6e22013-04-25 15:55:30 -04002156 }
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002157
Chris Masone02119d2008-09-05 16:13:11 -04002158 path->slots[*level]++;
2159 if (wc->free) {
Tsutomu Itoh018642a2012-05-29 18:10:13 +09002160 ret = btrfs_read_buffer(next, ptr_gen);
2161 if (ret) {
2162 free_extent_buffer(next);
2163 return ret;
2164 }
Chris Masone02119d2008-09-05 16:13:11 -04002165
Josef Bacik681ae502013-10-07 15:11:00 -04002166 if (trans) {
2167 btrfs_tree_lock(next);
2168 btrfs_set_lock_blocking(next);
2169 clean_tree_block(trans, root, next);
2170 btrfs_wait_tree_block_writeback(next);
2171 btrfs_tree_unlock(next);
2172 }
Chris Masone02119d2008-09-05 16:13:11 -04002173
Chris Masone02119d2008-09-05 16:13:11 -04002174 WARN_ON(root_owner !=
2175 BTRFS_TREE_LOG_OBJECTID);
Chris Masone688b7252011-10-31 20:52:39 -04002176 ret = btrfs_free_and_pin_reserved_extent(root,
Chris Masond00aff02008-09-11 15:54:42 -04002177 bytenr, blocksize);
Josef Bacik36508602013-04-25 16:23:32 -04002178 if (ret) {
2179 free_extent_buffer(next);
2180 return ret;
2181 }
Chris Masone02119d2008-09-05 16:13:11 -04002182 }
2183 free_extent_buffer(next);
2184 continue;
2185 }
Tsutomu Itoh018642a2012-05-29 18:10:13 +09002186 ret = btrfs_read_buffer(next, ptr_gen);
2187 if (ret) {
2188 free_extent_buffer(next);
2189 return ret;
2190 }
Chris Masone02119d2008-09-05 16:13:11 -04002191
2192 WARN_ON(*level <= 0);
2193 if (path->nodes[*level-1])
2194 free_extent_buffer(path->nodes[*level-1]);
2195 path->nodes[*level-1] = next;
2196 *level = btrfs_header_level(next);
2197 path->slots[*level] = 0;
2198 cond_resched();
2199 }
2200 WARN_ON(*level < 0);
2201 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2202
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002203 path->slots[*level] = btrfs_header_nritems(path->nodes[*level]);
Chris Masone02119d2008-09-05 16:13:11 -04002204
2205 cond_resched();
2206 return 0;
2207}
2208
Chris Masond3977122009-01-05 21:25:51 -05002209static noinline int walk_up_log_tree(struct btrfs_trans_handle *trans,
Chris Masone02119d2008-09-05 16:13:11 -04002210 struct btrfs_root *root,
2211 struct btrfs_path *path, int *level,
2212 struct walk_control *wc)
2213{
2214 u64 root_owner;
Chris Masone02119d2008-09-05 16:13:11 -04002215 int i;
2216 int slot;
2217 int ret;
2218
Chris Masond3977122009-01-05 21:25:51 -05002219 for (i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) {
Chris Masone02119d2008-09-05 16:13:11 -04002220 slot = path->slots[i];
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002221 if (slot + 1 < btrfs_header_nritems(path->nodes[i])) {
Chris Masone02119d2008-09-05 16:13:11 -04002222 path->slots[i]++;
2223 *level = i;
2224 WARN_ON(*level == 0);
2225 return 0;
2226 } else {
Zheng Yan31840ae2008-09-23 13:14:14 -04002227 struct extent_buffer *parent;
2228 if (path->nodes[*level] == root->node)
2229 parent = path->nodes[*level];
2230 else
2231 parent = path->nodes[*level + 1];
2232
2233 root_owner = btrfs_header_owner(parent);
Mark Fasheh1e5063d2011-07-12 10:46:06 -07002234 ret = wc->process_func(root, path->nodes[*level], wc,
Chris Masone02119d2008-09-05 16:13:11 -04002235 btrfs_header_generation(path->nodes[*level]));
Mark Fasheh1e5063d2011-07-12 10:46:06 -07002236 if (ret)
2237 return ret;
2238
Chris Masone02119d2008-09-05 16:13:11 -04002239 if (wc->free) {
2240 struct extent_buffer *next;
2241
2242 next = path->nodes[*level];
2243
Josef Bacik681ae502013-10-07 15:11:00 -04002244 if (trans) {
2245 btrfs_tree_lock(next);
2246 btrfs_set_lock_blocking(next);
2247 clean_tree_block(trans, root, next);
2248 btrfs_wait_tree_block_writeback(next);
2249 btrfs_tree_unlock(next);
2250 }
Chris Masone02119d2008-09-05 16:13:11 -04002251
Chris Masone02119d2008-09-05 16:13:11 -04002252 WARN_ON(root_owner != BTRFS_TREE_LOG_OBJECTID);
Chris Masone688b7252011-10-31 20:52:39 -04002253 ret = btrfs_free_and_pin_reserved_extent(root,
Chris Masone02119d2008-09-05 16:13:11 -04002254 path->nodes[*level]->start,
Chris Masond00aff02008-09-11 15:54:42 -04002255 path->nodes[*level]->len);
Josef Bacik36508602013-04-25 16:23:32 -04002256 if (ret)
2257 return ret;
Chris Masone02119d2008-09-05 16:13:11 -04002258 }
2259 free_extent_buffer(path->nodes[*level]);
2260 path->nodes[*level] = NULL;
2261 *level = i + 1;
2262 }
2263 }
2264 return 1;
2265}
2266
2267/*
2268 * drop the reference count on the tree rooted at 'snap'. This traverses
2269 * the tree freeing any blocks that have a ref count of zero after being
2270 * decremented.
2271 */
2272static int walk_log_tree(struct btrfs_trans_handle *trans,
2273 struct btrfs_root *log, struct walk_control *wc)
2274{
2275 int ret = 0;
2276 int wret;
2277 int level;
2278 struct btrfs_path *path;
Chris Masone02119d2008-09-05 16:13:11 -04002279 int orig_level;
2280
2281 path = btrfs_alloc_path();
Tsutomu Itohdb5b4932011-03-23 08:14:16 +00002282 if (!path)
2283 return -ENOMEM;
Chris Masone02119d2008-09-05 16:13:11 -04002284
2285 level = btrfs_header_level(log->node);
2286 orig_level = level;
2287 path->nodes[level] = log->node;
2288 extent_buffer_get(log->node);
2289 path->slots[level] = 0;
2290
Chris Masond3977122009-01-05 21:25:51 -05002291 while (1) {
Chris Masone02119d2008-09-05 16:13:11 -04002292 wret = walk_down_log_tree(trans, log, path, &level, wc);
2293 if (wret > 0)
2294 break;
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002295 if (wret < 0) {
Chris Masone02119d2008-09-05 16:13:11 -04002296 ret = wret;
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002297 goto out;
2298 }
Chris Masone02119d2008-09-05 16:13:11 -04002299
2300 wret = walk_up_log_tree(trans, log, path, &level, wc);
2301 if (wret > 0)
2302 break;
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002303 if (wret < 0) {
Chris Masone02119d2008-09-05 16:13:11 -04002304 ret = wret;
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002305 goto out;
2306 }
Chris Masone02119d2008-09-05 16:13:11 -04002307 }
2308
2309 /* was the root node processed? if not, catch it here */
2310 if (path->nodes[orig_level]) {
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002311 ret = wc->process_func(log, path->nodes[orig_level], wc,
Chris Masone02119d2008-09-05 16:13:11 -04002312 btrfs_header_generation(path->nodes[orig_level]));
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002313 if (ret)
2314 goto out;
Chris Masone02119d2008-09-05 16:13:11 -04002315 if (wc->free) {
2316 struct extent_buffer *next;
2317
2318 next = path->nodes[orig_level];
2319
Josef Bacik681ae502013-10-07 15:11:00 -04002320 if (trans) {
2321 btrfs_tree_lock(next);
2322 btrfs_set_lock_blocking(next);
2323 clean_tree_block(trans, log, next);
2324 btrfs_wait_tree_block_writeback(next);
2325 btrfs_tree_unlock(next);
2326 }
Chris Masone02119d2008-09-05 16:13:11 -04002327
Chris Masone02119d2008-09-05 16:13:11 -04002328 WARN_ON(log->root_key.objectid !=
2329 BTRFS_TREE_LOG_OBJECTID);
Chris Masone688b7252011-10-31 20:52:39 -04002330 ret = btrfs_free_and_pin_reserved_extent(log, next->start,
Chris Masond00aff02008-09-11 15:54:42 -04002331 next->len);
Josef Bacik36508602013-04-25 16:23:32 -04002332 if (ret)
2333 goto out;
Chris Masone02119d2008-09-05 16:13:11 -04002334 }
2335 }
2336
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002337out:
Chris Masone02119d2008-09-05 16:13:11 -04002338 btrfs_free_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04002339 return ret;
2340}
2341
Yan Zheng7237f182009-01-21 12:54:03 -05002342/*
2343 * helper function to update the item for a given subvolumes log root
2344 * in the tree of log roots
2345 */
2346static int update_log_root(struct btrfs_trans_handle *trans,
2347 struct btrfs_root *log)
2348{
2349 int ret;
2350
2351 if (log->log_transid == 1) {
2352 /* insert root item on the first sync */
2353 ret = btrfs_insert_root(trans, log->fs_info->log_root_tree,
2354 &log->root_key, &log->root_item);
2355 } else {
2356 ret = btrfs_update_root(trans, log->fs_info->log_root_tree,
2357 &log->root_key, &log->root_item);
2358 }
2359 return ret;
2360}
2361
Chris Mason12fcfd22009-03-24 10:24:20 -04002362static int wait_log_commit(struct btrfs_trans_handle *trans,
2363 struct btrfs_root *root, unsigned long transid)
Chris Masone02119d2008-09-05 16:13:11 -04002364{
2365 DEFINE_WAIT(wait);
Yan Zheng7237f182009-01-21 12:54:03 -05002366 int index = transid % 2;
Chris Masone02119d2008-09-05 16:13:11 -04002367
Yan Zheng7237f182009-01-21 12:54:03 -05002368 /*
2369 * we only allow two pending log transactions at a time,
2370 * so we know that if ours is more than 2 older than the
2371 * current transaction, we're done
2372 */
Chris Masone02119d2008-09-05 16:13:11 -04002373 do {
Yan Zheng7237f182009-01-21 12:54:03 -05002374 prepare_to_wait(&root->log_commit_wait[index],
2375 &wait, TASK_UNINTERRUPTIBLE);
2376 mutex_unlock(&root->log_mutex);
Chris Mason12fcfd22009-03-24 10:24:20 -04002377
2378 if (root->fs_info->last_trans_log_full_commit !=
2379 trans->transid && root->log_transid < transid + 2 &&
Yan Zheng7237f182009-01-21 12:54:03 -05002380 atomic_read(&root->log_commit[index]))
Chris Masone02119d2008-09-05 16:13:11 -04002381 schedule();
Chris Mason12fcfd22009-03-24 10:24:20 -04002382
Yan Zheng7237f182009-01-21 12:54:03 -05002383 finish_wait(&root->log_commit_wait[index], &wait);
2384 mutex_lock(&root->log_mutex);
Jan Kara6dd70ce2012-01-26 15:01:11 -05002385 } while (root->fs_info->last_trans_log_full_commit !=
2386 trans->transid && root->log_transid < transid + 2 &&
Yan Zheng7237f182009-01-21 12:54:03 -05002387 atomic_read(&root->log_commit[index]));
2388 return 0;
2389}
2390
Jeff Mahoney143bede2012-03-01 14:56:26 +01002391static void wait_for_writer(struct btrfs_trans_handle *trans,
2392 struct btrfs_root *root)
Yan Zheng7237f182009-01-21 12:54:03 -05002393{
2394 DEFINE_WAIT(wait);
Jan Kara6dd70ce2012-01-26 15:01:11 -05002395 while (root->fs_info->last_trans_log_full_commit !=
2396 trans->transid && atomic_read(&root->log_writers)) {
Yan Zheng7237f182009-01-21 12:54:03 -05002397 prepare_to_wait(&root->log_writer_wait,
2398 &wait, TASK_UNINTERRUPTIBLE);
2399 mutex_unlock(&root->log_mutex);
Chris Mason12fcfd22009-03-24 10:24:20 -04002400 if (root->fs_info->last_trans_log_full_commit !=
2401 trans->transid && atomic_read(&root->log_writers))
Yan Zheng7237f182009-01-21 12:54:03 -05002402 schedule();
2403 mutex_lock(&root->log_mutex);
2404 finish_wait(&root->log_writer_wait, &wait);
2405 }
Chris Masone02119d2008-09-05 16:13:11 -04002406}
2407
2408/*
2409 * btrfs_sync_log does sends a given tree log down to the disk and
2410 * updates the super blocks to record it. When this call is done,
Chris Mason12fcfd22009-03-24 10:24:20 -04002411 * you know that any inodes previously logged are safely on disk only
2412 * if it returns 0.
2413 *
2414 * Any other return value means you need to call btrfs_commit_transaction.
2415 * Some of the edge cases for fsyncing directories that have had unlinks
2416 * or renames done in the past mean that sometimes the only safe
2417 * fsync is to commit the whole FS. When btrfs_sync_log returns -EAGAIN,
2418 * that has happened.
Chris Masone02119d2008-09-05 16:13:11 -04002419 */
2420int btrfs_sync_log(struct btrfs_trans_handle *trans,
2421 struct btrfs_root *root)
2422{
Yan Zheng7237f182009-01-21 12:54:03 -05002423 int index1;
2424 int index2;
Yan, Zheng8cef4e12009-11-12 09:33:26 +00002425 int mark;
Chris Masone02119d2008-09-05 16:13:11 -04002426 int ret;
Chris Masone02119d2008-09-05 16:13:11 -04002427 struct btrfs_root *log = root->log_root;
Yan Zheng7237f182009-01-21 12:54:03 -05002428 struct btrfs_root *log_root_tree = root->fs_info->log_root_tree;
Yan, Zheng8cef4e12009-11-12 09:33:26 +00002429 unsigned long log_transid = 0;
Miao Xiec6adc9c2013-05-28 10:05:39 +00002430 struct blk_plug plug;
Chris Masone02119d2008-09-05 16:13:11 -04002431
Yan Zheng7237f182009-01-21 12:54:03 -05002432 mutex_lock(&root->log_mutex);
Josef Bacik2ab28f32012-10-12 15:27:49 -04002433 log_transid = root->log_transid;
Yan Zheng7237f182009-01-21 12:54:03 -05002434 index1 = root->log_transid % 2;
2435 if (atomic_read(&root->log_commit[index1])) {
Chris Mason12fcfd22009-03-24 10:24:20 -04002436 wait_log_commit(trans, root, root->log_transid);
Yan Zheng7237f182009-01-21 12:54:03 -05002437 mutex_unlock(&root->log_mutex);
2438 return 0;
Chris Masone02119d2008-09-05 16:13:11 -04002439 }
Yan Zheng7237f182009-01-21 12:54:03 -05002440 atomic_set(&root->log_commit[index1], 1);
2441
2442 /* wait for previous tree log sync to complete */
2443 if (atomic_read(&root->log_commit[(index1 + 1) % 2]))
Chris Mason12fcfd22009-03-24 10:24:20 -04002444 wait_log_commit(trans, root, root->log_transid - 1);
Yan, Zheng86df7eb2009-10-14 09:24:59 -04002445 while (1) {
Miao Xie2ecb7922012-09-06 04:04:27 -06002446 int batch = atomic_read(&root->log_batch);
Chris Masoncd354ad2011-10-20 15:45:37 -04002447 /* when we're on an ssd, just kick the log commit out */
2448 if (!btrfs_test_opt(root, SSD) && root->log_multiple_pids) {
Yan, Zheng86df7eb2009-10-14 09:24:59 -04002449 mutex_unlock(&root->log_mutex);
2450 schedule_timeout_uninterruptible(1);
2451 mutex_lock(&root->log_mutex);
2452 }
Chris Mason12fcfd22009-03-24 10:24:20 -04002453 wait_for_writer(trans, root);
Miao Xie2ecb7922012-09-06 04:04:27 -06002454 if (batch == atomic_read(&root->log_batch))
Chris Masone02119d2008-09-05 16:13:11 -04002455 break;
2456 }
Chris Masond0c803c2008-09-11 16:17:57 -04002457
Chris Mason12fcfd22009-03-24 10:24:20 -04002458 /* bail out if we need to do a full commit */
2459 if (root->fs_info->last_trans_log_full_commit == trans->transid) {
2460 ret = -EAGAIN;
Josef Bacik2ab28f32012-10-12 15:27:49 -04002461 btrfs_free_logged_extents(log, log_transid);
Chris Mason12fcfd22009-03-24 10:24:20 -04002462 mutex_unlock(&root->log_mutex);
2463 goto out;
2464 }
2465
Yan, Zheng8cef4e12009-11-12 09:33:26 +00002466 if (log_transid % 2 == 0)
2467 mark = EXTENT_DIRTY;
2468 else
2469 mark = EXTENT_NEW;
2470
Chris Mason690587d2009-10-13 13:29:19 -04002471 /* we start IO on all the marked extents here, but we don't actually
2472 * wait for them until later.
2473 */
Miao Xiec6adc9c2013-05-28 10:05:39 +00002474 blk_start_plug(&plug);
Yan, Zheng8cef4e12009-11-12 09:33:26 +00002475 ret = btrfs_write_marked_extents(log, &log->dirty_log_pages, mark);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002476 if (ret) {
Miao Xiec6adc9c2013-05-28 10:05:39 +00002477 blk_finish_plug(&plug);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002478 btrfs_abort_transaction(trans, root, ret);
Josef Bacik2ab28f32012-10-12 15:27:49 -04002479 btrfs_free_logged_extents(log, log_transid);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002480 mutex_unlock(&root->log_mutex);
2481 goto out;
2482 }
Yan Zheng7237f182009-01-21 12:54:03 -05002483
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002484 btrfs_set_root_node(&log->root_item, log->node);
Yan Zheng7237f182009-01-21 12:54:03 -05002485
Yan Zheng7237f182009-01-21 12:54:03 -05002486 root->log_transid++;
2487 log->log_transid = root->log_transid;
Josef Bacikff782e02009-10-08 15:30:04 -04002488 root->log_start_pid = 0;
Yan Zheng7237f182009-01-21 12:54:03 -05002489 smp_mb();
2490 /*
Yan, Zheng8cef4e12009-11-12 09:33:26 +00002491 * IO has been started, blocks of the log tree have WRITTEN flag set
2492 * in their headers. new modifications of the log will be written to
2493 * new positions. so it's safe to allow log writers to go in.
Yan Zheng7237f182009-01-21 12:54:03 -05002494 */
2495 mutex_unlock(&root->log_mutex);
2496
2497 mutex_lock(&log_root_tree->log_mutex);
Miao Xie2ecb7922012-09-06 04:04:27 -06002498 atomic_inc(&log_root_tree->log_batch);
Yan Zheng7237f182009-01-21 12:54:03 -05002499 atomic_inc(&log_root_tree->log_writers);
2500 mutex_unlock(&log_root_tree->log_mutex);
2501
2502 ret = update_log_root(trans, log);
Yan Zheng7237f182009-01-21 12:54:03 -05002503
2504 mutex_lock(&log_root_tree->log_mutex);
2505 if (atomic_dec_and_test(&log_root_tree->log_writers)) {
2506 smp_mb();
2507 if (waitqueue_active(&log_root_tree->log_writer_wait))
2508 wake_up(&log_root_tree->log_writer_wait);
2509 }
2510
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002511 if (ret) {
Miao Xiec6adc9c2013-05-28 10:05:39 +00002512 blk_finish_plug(&plug);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002513 if (ret != -ENOSPC) {
2514 btrfs_abort_transaction(trans, root, ret);
2515 mutex_unlock(&log_root_tree->log_mutex);
2516 goto out;
2517 }
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002518 root->fs_info->last_trans_log_full_commit = trans->transid;
2519 btrfs_wait_marked_extents(log, &log->dirty_log_pages, mark);
Josef Bacik2ab28f32012-10-12 15:27:49 -04002520 btrfs_free_logged_extents(log, log_transid);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002521 mutex_unlock(&log_root_tree->log_mutex);
2522 ret = -EAGAIN;
2523 goto out;
2524 }
2525
Yan Zheng7237f182009-01-21 12:54:03 -05002526 index2 = log_root_tree->log_transid % 2;
2527 if (atomic_read(&log_root_tree->log_commit[index2])) {
Miao Xiec6adc9c2013-05-28 10:05:39 +00002528 blk_finish_plug(&plug);
Yan, Zheng8cef4e12009-11-12 09:33:26 +00002529 btrfs_wait_marked_extents(log, &log->dirty_log_pages, mark);
Chris Mason12fcfd22009-03-24 10:24:20 -04002530 wait_log_commit(trans, log_root_tree,
2531 log_root_tree->log_transid);
Josef Bacik2ab28f32012-10-12 15:27:49 -04002532 btrfs_free_logged_extents(log, log_transid);
Yan Zheng7237f182009-01-21 12:54:03 -05002533 mutex_unlock(&log_root_tree->log_mutex);
Chris Masonb31eabd2011-01-31 16:48:24 -05002534 ret = 0;
Yan Zheng7237f182009-01-21 12:54:03 -05002535 goto out;
2536 }
2537 atomic_set(&log_root_tree->log_commit[index2], 1);
2538
Chris Mason12fcfd22009-03-24 10:24:20 -04002539 if (atomic_read(&log_root_tree->log_commit[(index2 + 1) % 2])) {
2540 wait_log_commit(trans, log_root_tree,
2541 log_root_tree->log_transid - 1);
2542 }
Yan Zheng7237f182009-01-21 12:54:03 -05002543
Chris Mason12fcfd22009-03-24 10:24:20 -04002544 wait_for_writer(trans, log_root_tree);
2545
2546 /*
2547 * now that we've moved on to the tree of log tree roots,
2548 * check the full commit flag again
2549 */
2550 if (root->fs_info->last_trans_log_full_commit == trans->transid) {
Miao Xiec6adc9c2013-05-28 10:05:39 +00002551 blk_finish_plug(&plug);
Yan, Zheng8cef4e12009-11-12 09:33:26 +00002552 btrfs_wait_marked_extents(log, &log->dirty_log_pages, mark);
Josef Bacik2ab28f32012-10-12 15:27:49 -04002553 btrfs_free_logged_extents(log, log_transid);
Chris Mason12fcfd22009-03-24 10:24:20 -04002554 mutex_unlock(&log_root_tree->log_mutex);
2555 ret = -EAGAIN;
2556 goto out_wake_log_root;
2557 }
Yan Zheng7237f182009-01-21 12:54:03 -05002558
Miao Xiec6adc9c2013-05-28 10:05:39 +00002559 ret = btrfs_write_marked_extents(log_root_tree,
2560 &log_root_tree->dirty_log_pages,
2561 EXTENT_DIRTY | EXTENT_NEW);
2562 blk_finish_plug(&plug);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002563 if (ret) {
2564 btrfs_abort_transaction(trans, root, ret);
Josef Bacik2ab28f32012-10-12 15:27:49 -04002565 btrfs_free_logged_extents(log, log_transid);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002566 mutex_unlock(&log_root_tree->log_mutex);
2567 goto out_wake_log_root;
2568 }
Yan, Zheng8cef4e12009-11-12 09:33:26 +00002569 btrfs_wait_marked_extents(log, &log->dirty_log_pages, mark);
Miao Xiec6adc9c2013-05-28 10:05:39 +00002570 btrfs_wait_marked_extents(log_root_tree,
2571 &log_root_tree->dirty_log_pages,
2572 EXTENT_NEW | EXTENT_DIRTY);
Josef Bacik2ab28f32012-10-12 15:27:49 -04002573 btrfs_wait_logged_extents(log, log_transid);
Chris Masone02119d2008-09-05 16:13:11 -04002574
David Sterba6c417612011-04-13 15:41:04 +02002575 btrfs_set_super_log_root(root->fs_info->super_for_commit,
Yan Zheng7237f182009-01-21 12:54:03 -05002576 log_root_tree->node->start);
David Sterba6c417612011-04-13 15:41:04 +02002577 btrfs_set_super_log_root_level(root->fs_info->super_for_commit,
Yan Zheng7237f182009-01-21 12:54:03 -05002578 btrfs_header_level(log_root_tree->node));
Chris Masone02119d2008-09-05 16:13:11 -04002579
Yan Zheng7237f182009-01-21 12:54:03 -05002580 log_root_tree->log_transid++;
Chris Masone02119d2008-09-05 16:13:11 -04002581 smp_mb();
Yan Zheng7237f182009-01-21 12:54:03 -05002582
2583 mutex_unlock(&log_root_tree->log_mutex);
2584
2585 /*
2586 * nobody else is going to jump in and write the the ctree
2587 * super here because the log_commit atomic below is protecting
2588 * us. We must be called with a transaction handle pinning
2589 * the running transaction open, so a full commit can't hop
2590 * in and cause problems either.
2591 */
Stefan Behrens5af3e8c2012-08-01 18:56:49 +02002592 ret = write_ctree_super(trans, root->fs_info->tree_root, 1);
Stefan Behrens5af3e8c2012-08-01 18:56:49 +02002593 if (ret) {
2594 btrfs_abort_transaction(trans, root, ret);
2595 goto out_wake_log_root;
2596 }
Yan Zheng7237f182009-01-21 12:54:03 -05002597
Chris Mason257c62e2009-10-13 13:21:08 -04002598 mutex_lock(&root->log_mutex);
2599 if (root->last_log_commit < log_transid)
2600 root->last_log_commit = log_transid;
2601 mutex_unlock(&root->log_mutex);
2602
Chris Mason12fcfd22009-03-24 10:24:20 -04002603out_wake_log_root:
Yan Zheng7237f182009-01-21 12:54:03 -05002604 atomic_set(&log_root_tree->log_commit[index2], 0);
2605 smp_mb();
2606 if (waitqueue_active(&log_root_tree->log_commit_wait[index2]))
2607 wake_up(&log_root_tree->log_commit_wait[index2]);
Chris Masone02119d2008-09-05 16:13:11 -04002608out:
Yan Zheng7237f182009-01-21 12:54:03 -05002609 atomic_set(&root->log_commit[index1], 0);
2610 smp_mb();
2611 if (waitqueue_active(&root->log_commit_wait[index1]))
2612 wake_up(&root->log_commit_wait[index1]);
Chris Masonb31eabd2011-01-31 16:48:24 -05002613 return ret;
Chris Masone02119d2008-09-05 16:13:11 -04002614}
2615
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002616static void free_log_tree(struct btrfs_trans_handle *trans,
2617 struct btrfs_root *log)
Chris Masone02119d2008-09-05 16:13:11 -04002618{
2619 int ret;
Chris Masond0c803c2008-09-11 16:17:57 -04002620 u64 start;
2621 u64 end;
Chris Masone02119d2008-09-05 16:13:11 -04002622 struct walk_control wc = {
2623 .free = 1,
2624 .process_func = process_one_buffer
2625 };
2626
Josef Bacik681ae502013-10-07 15:11:00 -04002627 ret = walk_log_tree(trans, log, &wc);
2628 /* I don't think this can happen but just in case */
2629 if (ret)
2630 btrfs_abort_transaction(trans, log, ret);
Chris Masone02119d2008-09-05 16:13:11 -04002631
Chris Masond3977122009-01-05 21:25:51 -05002632 while (1) {
Chris Masond0c803c2008-09-11 16:17:57 -04002633 ret = find_first_extent_bit(&log->dirty_log_pages,
Josef Bacike6138872012-09-27 17:07:30 -04002634 0, &start, &end, EXTENT_DIRTY | EXTENT_NEW,
2635 NULL);
Chris Masond0c803c2008-09-11 16:17:57 -04002636 if (ret)
2637 break;
2638
Yan, Zheng8cef4e12009-11-12 09:33:26 +00002639 clear_extent_bits(&log->dirty_log_pages, start, end,
2640 EXTENT_DIRTY | EXTENT_NEW, GFP_NOFS);
Chris Masond0c803c2008-09-11 16:17:57 -04002641 }
2642
Josef Bacik2ab28f32012-10-12 15:27:49 -04002643 /*
2644 * We may have short-circuited the log tree with the full commit logic
2645 * and left ordered extents on our list, so clear these out to keep us
2646 * from leaking inodes and memory.
2647 */
2648 btrfs_free_logged_extents(log, 0);
2649 btrfs_free_logged_extents(log, 1);
2650
Yan Zheng7237f182009-01-21 12:54:03 -05002651 free_extent_buffer(log->node);
2652 kfree(log);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002653}
2654
2655/*
2656 * free all the extents used by the tree log. This should be called
2657 * at commit time of the full transaction
2658 */
2659int btrfs_free_log(struct btrfs_trans_handle *trans, struct btrfs_root *root)
2660{
2661 if (root->log_root) {
2662 free_log_tree(trans, root->log_root);
2663 root->log_root = NULL;
2664 }
2665 return 0;
2666}
2667
2668int btrfs_free_log_root_tree(struct btrfs_trans_handle *trans,
2669 struct btrfs_fs_info *fs_info)
2670{
2671 if (fs_info->log_root_tree) {
2672 free_log_tree(trans, fs_info->log_root_tree);
2673 fs_info->log_root_tree = NULL;
2674 }
Chris Masone02119d2008-09-05 16:13:11 -04002675 return 0;
2676}
2677
2678/*
Chris Masone02119d2008-09-05 16:13:11 -04002679 * If both a file and directory are logged, and unlinks or renames are
2680 * mixed in, we have a few interesting corners:
2681 *
2682 * create file X in dir Y
2683 * link file X to X.link in dir Y
2684 * fsync file X
2685 * unlink file X but leave X.link
2686 * fsync dir Y
2687 *
2688 * After a crash we would expect only X.link to exist. But file X
2689 * didn't get fsync'd again so the log has back refs for X and X.link.
2690 *
2691 * We solve this by removing directory entries and inode backrefs from the
2692 * log when a file that was logged in the current transaction is
2693 * unlinked. Any later fsync will include the updated log entries, and
2694 * we'll be able to reconstruct the proper directory items from backrefs.
2695 *
2696 * This optimizations allows us to avoid relogging the entire inode
2697 * or the entire directory.
2698 */
2699int btrfs_del_dir_entries_in_log(struct btrfs_trans_handle *trans,
2700 struct btrfs_root *root,
2701 const char *name, int name_len,
2702 struct inode *dir, u64 index)
2703{
2704 struct btrfs_root *log;
2705 struct btrfs_dir_item *di;
2706 struct btrfs_path *path;
2707 int ret;
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002708 int err = 0;
Chris Masone02119d2008-09-05 16:13:11 -04002709 int bytes_del = 0;
Li Zefan33345d012011-04-20 10:31:50 +08002710 u64 dir_ino = btrfs_ino(dir);
Chris Masone02119d2008-09-05 16:13:11 -04002711
Chris Mason3a5f1d42008-09-11 15:53:37 -04002712 if (BTRFS_I(dir)->logged_trans < trans->transid)
2713 return 0;
2714
Chris Masone02119d2008-09-05 16:13:11 -04002715 ret = join_running_log_trans(root);
2716 if (ret)
2717 return 0;
2718
2719 mutex_lock(&BTRFS_I(dir)->log_mutex);
2720
2721 log = root->log_root;
2722 path = btrfs_alloc_path();
Tsutomu Itoha62f44a2011-04-25 19:43:51 -04002723 if (!path) {
2724 err = -ENOMEM;
2725 goto out_unlock;
2726 }
liubo2a29edc2011-01-26 06:22:08 +00002727
Li Zefan33345d012011-04-20 10:31:50 +08002728 di = btrfs_lookup_dir_item(trans, log, path, dir_ino,
Chris Masone02119d2008-09-05 16:13:11 -04002729 name, name_len, -1);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002730 if (IS_ERR(di)) {
2731 err = PTR_ERR(di);
2732 goto fail;
2733 }
2734 if (di) {
Chris Masone02119d2008-09-05 16:13:11 -04002735 ret = btrfs_delete_one_dir_name(trans, log, path, di);
2736 bytes_del += name_len;
Josef Bacik36508602013-04-25 16:23:32 -04002737 if (ret) {
2738 err = ret;
2739 goto fail;
2740 }
Chris Masone02119d2008-09-05 16:13:11 -04002741 }
David Sterbab3b4aa72011-04-21 01:20:15 +02002742 btrfs_release_path(path);
Li Zefan33345d012011-04-20 10:31:50 +08002743 di = btrfs_lookup_dir_index_item(trans, log, path, dir_ino,
Chris Masone02119d2008-09-05 16:13:11 -04002744 index, name, name_len, -1);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002745 if (IS_ERR(di)) {
2746 err = PTR_ERR(di);
2747 goto fail;
2748 }
2749 if (di) {
Chris Masone02119d2008-09-05 16:13:11 -04002750 ret = btrfs_delete_one_dir_name(trans, log, path, di);
2751 bytes_del += name_len;
Josef Bacik36508602013-04-25 16:23:32 -04002752 if (ret) {
2753 err = ret;
2754 goto fail;
2755 }
Chris Masone02119d2008-09-05 16:13:11 -04002756 }
2757
2758 /* update the directory size in the log to reflect the names
2759 * we have removed
2760 */
2761 if (bytes_del) {
2762 struct btrfs_key key;
2763
Li Zefan33345d012011-04-20 10:31:50 +08002764 key.objectid = dir_ino;
Chris Masone02119d2008-09-05 16:13:11 -04002765 key.offset = 0;
2766 key.type = BTRFS_INODE_ITEM_KEY;
David Sterbab3b4aa72011-04-21 01:20:15 +02002767 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04002768
2769 ret = btrfs_search_slot(trans, log, &key, path, 0, 1);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002770 if (ret < 0) {
2771 err = ret;
2772 goto fail;
2773 }
Chris Masone02119d2008-09-05 16:13:11 -04002774 if (ret == 0) {
2775 struct btrfs_inode_item *item;
2776 u64 i_size;
2777
2778 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2779 struct btrfs_inode_item);
2780 i_size = btrfs_inode_size(path->nodes[0], item);
2781 if (i_size > bytes_del)
2782 i_size -= bytes_del;
2783 else
2784 i_size = 0;
2785 btrfs_set_inode_size(path->nodes[0], item, i_size);
2786 btrfs_mark_buffer_dirty(path->nodes[0]);
2787 } else
2788 ret = 0;
David Sterbab3b4aa72011-04-21 01:20:15 +02002789 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04002790 }
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002791fail:
Chris Masone02119d2008-09-05 16:13:11 -04002792 btrfs_free_path(path);
Tsutomu Itoha62f44a2011-04-25 19:43:51 -04002793out_unlock:
Chris Masone02119d2008-09-05 16:13:11 -04002794 mutex_unlock(&BTRFS_I(dir)->log_mutex);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002795 if (ret == -ENOSPC) {
2796 root->fs_info->last_trans_log_full_commit = trans->transid;
2797 ret = 0;
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002798 } else if (ret < 0)
2799 btrfs_abort_transaction(trans, root, ret);
2800
Chris Mason12fcfd22009-03-24 10:24:20 -04002801 btrfs_end_log_trans(root);
Chris Masone02119d2008-09-05 16:13:11 -04002802
Andi Kleen411fc6b2010-10-29 15:14:31 -04002803 return err;
Chris Masone02119d2008-09-05 16:13:11 -04002804}
2805
2806/* see comments for btrfs_del_dir_entries_in_log */
2807int btrfs_del_inode_ref_in_log(struct btrfs_trans_handle *trans,
2808 struct btrfs_root *root,
2809 const char *name, int name_len,
2810 struct inode *inode, u64 dirid)
2811{
2812 struct btrfs_root *log;
2813 u64 index;
2814 int ret;
2815
Chris Mason3a5f1d42008-09-11 15:53:37 -04002816 if (BTRFS_I(inode)->logged_trans < trans->transid)
2817 return 0;
2818
Chris Masone02119d2008-09-05 16:13:11 -04002819 ret = join_running_log_trans(root);
2820 if (ret)
2821 return 0;
2822 log = root->log_root;
2823 mutex_lock(&BTRFS_I(inode)->log_mutex);
2824
Li Zefan33345d012011-04-20 10:31:50 +08002825 ret = btrfs_del_inode_ref(trans, log, name, name_len, btrfs_ino(inode),
Chris Masone02119d2008-09-05 16:13:11 -04002826 dirid, &index);
2827 mutex_unlock(&BTRFS_I(inode)->log_mutex);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002828 if (ret == -ENOSPC) {
2829 root->fs_info->last_trans_log_full_commit = trans->transid;
2830 ret = 0;
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002831 } else if (ret < 0 && ret != -ENOENT)
2832 btrfs_abort_transaction(trans, root, ret);
Chris Mason12fcfd22009-03-24 10:24:20 -04002833 btrfs_end_log_trans(root);
Chris Masone02119d2008-09-05 16:13:11 -04002834
Chris Masone02119d2008-09-05 16:13:11 -04002835 return ret;
2836}
2837
2838/*
2839 * creates a range item in the log for 'dirid'. first_offset and
2840 * last_offset tell us which parts of the key space the log should
2841 * be considered authoritative for.
2842 */
2843static noinline int insert_dir_log_key(struct btrfs_trans_handle *trans,
2844 struct btrfs_root *log,
2845 struct btrfs_path *path,
2846 int key_type, u64 dirid,
2847 u64 first_offset, u64 last_offset)
2848{
2849 int ret;
2850 struct btrfs_key key;
2851 struct btrfs_dir_log_item *item;
2852
2853 key.objectid = dirid;
2854 key.offset = first_offset;
2855 if (key_type == BTRFS_DIR_ITEM_KEY)
2856 key.type = BTRFS_DIR_LOG_ITEM_KEY;
2857 else
2858 key.type = BTRFS_DIR_LOG_INDEX_KEY;
2859 ret = btrfs_insert_empty_item(trans, log, path, &key, sizeof(*item));
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002860 if (ret)
2861 return ret;
Chris Masone02119d2008-09-05 16:13:11 -04002862
2863 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2864 struct btrfs_dir_log_item);
2865 btrfs_set_dir_log_end(path->nodes[0], item, last_offset);
2866 btrfs_mark_buffer_dirty(path->nodes[0]);
David Sterbab3b4aa72011-04-21 01:20:15 +02002867 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04002868 return 0;
2869}
2870
2871/*
2872 * log all the items included in the current transaction for a given
2873 * directory. This also creates the range items in the log tree required
2874 * to replay anything deleted before the fsync
2875 */
2876static noinline int log_dir_items(struct btrfs_trans_handle *trans,
2877 struct btrfs_root *root, struct inode *inode,
2878 struct btrfs_path *path,
2879 struct btrfs_path *dst_path, int key_type,
2880 u64 min_offset, u64 *last_offset_ret)
2881{
2882 struct btrfs_key min_key;
Chris Masone02119d2008-09-05 16:13:11 -04002883 struct btrfs_root *log = root->log_root;
2884 struct extent_buffer *src;
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002885 int err = 0;
Chris Masone02119d2008-09-05 16:13:11 -04002886 int ret;
2887 int i;
2888 int nritems;
2889 u64 first_offset = min_offset;
2890 u64 last_offset = (u64)-1;
Li Zefan33345d012011-04-20 10:31:50 +08002891 u64 ino = btrfs_ino(inode);
Chris Masone02119d2008-09-05 16:13:11 -04002892
2893 log = root->log_root;
Chris Masone02119d2008-09-05 16:13:11 -04002894
Li Zefan33345d012011-04-20 10:31:50 +08002895 min_key.objectid = ino;
Chris Masone02119d2008-09-05 16:13:11 -04002896 min_key.type = key_type;
2897 min_key.offset = min_offset;
2898
2899 path->keep_locks = 1;
2900
Filipe David Borba Manana6174d3c2013-10-01 16:13:42 +01002901 ret = btrfs_search_forward(root, &min_key, path, trans->transid);
Chris Masone02119d2008-09-05 16:13:11 -04002902
2903 /*
2904 * we didn't find anything from this transaction, see if there
2905 * is anything at all
2906 */
Li Zefan33345d012011-04-20 10:31:50 +08002907 if (ret != 0 || min_key.objectid != ino || min_key.type != key_type) {
2908 min_key.objectid = ino;
Chris Masone02119d2008-09-05 16:13:11 -04002909 min_key.type = key_type;
2910 min_key.offset = (u64)-1;
David Sterbab3b4aa72011-04-21 01:20:15 +02002911 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04002912 ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0);
2913 if (ret < 0) {
David Sterbab3b4aa72011-04-21 01:20:15 +02002914 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04002915 return ret;
2916 }
Li Zefan33345d012011-04-20 10:31:50 +08002917 ret = btrfs_previous_item(root, path, ino, key_type);
Chris Masone02119d2008-09-05 16:13:11 -04002918
2919 /* if ret == 0 there are items for this type,
2920 * create a range to tell us the last key of this type.
2921 * otherwise, there are no items in this directory after
2922 * *min_offset, and we create a range to indicate that.
2923 */
2924 if (ret == 0) {
2925 struct btrfs_key tmp;
2926 btrfs_item_key_to_cpu(path->nodes[0], &tmp,
2927 path->slots[0]);
Chris Masond3977122009-01-05 21:25:51 -05002928 if (key_type == tmp.type)
Chris Masone02119d2008-09-05 16:13:11 -04002929 first_offset = max(min_offset, tmp.offset) + 1;
Chris Masone02119d2008-09-05 16:13:11 -04002930 }
2931 goto done;
2932 }
2933
2934 /* go backward to find any previous key */
Li Zefan33345d012011-04-20 10:31:50 +08002935 ret = btrfs_previous_item(root, path, ino, key_type);
Chris Masone02119d2008-09-05 16:13:11 -04002936 if (ret == 0) {
2937 struct btrfs_key tmp;
2938 btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]);
2939 if (key_type == tmp.type) {
2940 first_offset = tmp.offset;
2941 ret = overwrite_item(trans, log, dst_path,
2942 path->nodes[0], path->slots[0],
2943 &tmp);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002944 if (ret) {
2945 err = ret;
2946 goto done;
2947 }
Chris Masone02119d2008-09-05 16:13:11 -04002948 }
2949 }
David Sterbab3b4aa72011-04-21 01:20:15 +02002950 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04002951
2952 /* find the first key from this transaction again */
2953 ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0);
Dulshani Gunawardhanafae7f212013-10-31 10:30:08 +05302954 if (WARN_ON(ret != 0))
Chris Masone02119d2008-09-05 16:13:11 -04002955 goto done;
Chris Masone02119d2008-09-05 16:13:11 -04002956
2957 /*
2958 * we have a block from this transaction, log every item in it
2959 * from our directory
2960 */
Chris Masond3977122009-01-05 21:25:51 -05002961 while (1) {
Chris Masone02119d2008-09-05 16:13:11 -04002962 struct btrfs_key tmp;
2963 src = path->nodes[0];
2964 nritems = btrfs_header_nritems(src);
2965 for (i = path->slots[0]; i < nritems; i++) {
2966 btrfs_item_key_to_cpu(src, &min_key, i);
2967
Li Zefan33345d012011-04-20 10:31:50 +08002968 if (min_key.objectid != ino || min_key.type != key_type)
Chris Masone02119d2008-09-05 16:13:11 -04002969 goto done;
2970 ret = overwrite_item(trans, log, dst_path, src, i,
2971 &min_key);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002972 if (ret) {
2973 err = ret;
2974 goto done;
2975 }
Chris Masone02119d2008-09-05 16:13:11 -04002976 }
2977 path->slots[0] = nritems;
2978
2979 /*
2980 * look ahead to the next item and see if it is also
2981 * from this directory and from this transaction
2982 */
2983 ret = btrfs_next_leaf(root, path);
2984 if (ret == 1) {
2985 last_offset = (u64)-1;
2986 goto done;
2987 }
2988 btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]);
Li Zefan33345d012011-04-20 10:31:50 +08002989 if (tmp.objectid != ino || tmp.type != key_type) {
Chris Masone02119d2008-09-05 16:13:11 -04002990 last_offset = (u64)-1;
2991 goto done;
2992 }
2993 if (btrfs_header_generation(path->nodes[0]) != trans->transid) {
2994 ret = overwrite_item(trans, log, dst_path,
2995 path->nodes[0], path->slots[0],
2996 &tmp);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002997 if (ret)
2998 err = ret;
2999 else
3000 last_offset = tmp.offset;
Chris Masone02119d2008-09-05 16:13:11 -04003001 goto done;
3002 }
3003 }
3004done:
David Sterbab3b4aa72011-04-21 01:20:15 +02003005 btrfs_release_path(path);
3006 btrfs_release_path(dst_path);
Chris Masone02119d2008-09-05 16:13:11 -04003007
Yan, Zheng4a500fd2010-05-16 10:49:59 -04003008 if (err == 0) {
3009 *last_offset_ret = last_offset;
3010 /*
3011 * insert the log range keys to indicate where the log
3012 * is valid
3013 */
3014 ret = insert_dir_log_key(trans, log, path, key_type,
Li Zefan33345d012011-04-20 10:31:50 +08003015 ino, first_offset, last_offset);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04003016 if (ret)
3017 err = ret;
3018 }
3019 return err;
Chris Masone02119d2008-09-05 16:13:11 -04003020}
3021
3022/*
3023 * logging directories is very similar to logging inodes, We find all the items
3024 * from the current transaction and write them to the log.
3025 *
3026 * The recovery code scans the directory in the subvolume, and if it finds a
3027 * key in the range logged that is not present in the log tree, then it means
3028 * that dir entry was unlinked during the transaction.
3029 *
3030 * In order for that scan to work, we must include one key smaller than
3031 * the smallest logged by this transaction and one key larger than the largest
3032 * key logged by this transaction.
3033 */
3034static noinline int log_directory_changes(struct btrfs_trans_handle *trans,
3035 struct btrfs_root *root, struct inode *inode,
3036 struct btrfs_path *path,
3037 struct btrfs_path *dst_path)
3038{
3039 u64 min_key;
3040 u64 max_key;
3041 int ret;
3042 int key_type = BTRFS_DIR_ITEM_KEY;
3043
3044again:
3045 min_key = 0;
3046 max_key = 0;
Chris Masond3977122009-01-05 21:25:51 -05003047 while (1) {
Chris Masone02119d2008-09-05 16:13:11 -04003048 ret = log_dir_items(trans, root, inode, path,
3049 dst_path, key_type, min_key,
3050 &max_key);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04003051 if (ret)
3052 return ret;
Chris Masone02119d2008-09-05 16:13:11 -04003053 if (max_key == (u64)-1)
3054 break;
3055 min_key = max_key + 1;
3056 }
3057
3058 if (key_type == BTRFS_DIR_ITEM_KEY) {
3059 key_type = BTRFS_DIR_INDEX_KEY;
3060 goto again;
3061 }
3062 return 0;
3063}
3064
3065/*
3066 * a helper function to drop items from the log before we relog an
3067 * inode. max_key_type indicates the highest item type to remove.
3068 * This cannot be run for file data extents because it does not
3069 * free the extents they point to.
3070 */
3071static int drop_objectid_items(struct btrfs_trans_handle *trans,
3072 struct btrfs_root *log,
3073 struct btrfs_path *path,
3074 u64 objectid, int max_key_type)
3075{
3076 int ret;
3077 struct btrfs_key key;
3078 struct btrfs_key found_key;
Josef Bacik18ec90d2012-09-28 11:56:28 -04003079 int start_slot;
Chris Masone02119d2008-09-05 16:13:11 -04003080
3081 key.objectid = objectid;
3082 key.type = max_key_type;
3083 key.offset = (u64)-1;
3084
Chris Masond3977122009-01-05 21:25:51 -05003085 while (1) {
Chris Masone02119d2008-09-05 16:13:11 -04003086 ret = btrfs_search_slot(trans, log, &key, path, -1, 1);
Josef Bacik36508602013-04-25 16:23:32 -04003087 BUG_ON(ret == 0); /* Logic error */
Yan, Zheng4a500fd2010-05-16 10:49:59 -04003088 if (ret < 0)
Chris Masone02119d2008-09-05 16:13:11 -04003089 break;
3090
3091 if (path->slots[0] == 0)
3092 break;
3093
3094 path->slots[0]--;
3095 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
3096 path->slots[0]);
3097
3098 if (found_key.objectid != objectid)
3099 break;
3100
Josef Bacik18ec90d2012-09-28 11:56:28 -04003101 found_key.offset = 0;
3102 found_key.type = 0;
3103 ret = btrfs_bin_search(path->nodes[0], &found_key, 0,
3104 &start_slot);
3105
3106 ret = btrfs_del_items(trans, log, path, start_slot,
3107 path->slots[0] - start_slot + 1);
3108 /*
3109 * If start slot isn't 0 then we don't need to re-search, we've
3110 * found the last guy with the objectid in this tree.
3111 */
3112 if (ret || start_slot != 0)
Tsutomu Itoh65a246c2011-05-19 04:37:44 +00003113 break;
David Sterbab3b4aa72011-04-21 01:20:15 +02003114 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04003115 }
David Sterbab3b4aa72011-04-21 01:20:15 +02003116 btrfs_release_path(path);
Josef Bacik5bdbeb22012-05-29 16:59:49 -04003117 if (ret > 0)
3118 ret = 0;
Yan, Zheng4a500fd2010-05-16 10:49:59 -04003119 return ret;
Chris Masone02119d2008-09-05 16:13:11 -04003120}
3121
Josef Bacik94edf4a2012-09-25 14:56:25 -04003122static void fill_inode_item(struct btrfs_trans_handle *trans,
3123 struct extent_buffer *leaf,
3124 struct btrfs_inode_item *item,
3125 struct inode *inode, int log_inode_only)
3126{
Josef Bacik0b1c6cc2012-10-23 16:03:44 -04003127 struct btrfs_map_token token;
Josef Bacik94edf4a2012-09-25 14:56:25 -04003128
Josef Bacik0b1c6cc2012-10-23 16:03:44 -04003129 btrfs_init_map_token(&token);
Josef Bacik94edf4a2012-09-25 14:56:25 -04003130
3131 if (log_inode_only) {
3132 /* set the generation to zero so the recover code
3133 * can tell the difference between an logging
3134 * just to say 'this inode exists' and a logging
3135 * to say 'update this inode with these values'
3136 */
Josef Bacik0b1c6cc2012-10-23 16:03:44 -04003137 btrfs_set_token_inode_generation(leaf, item, 0, &token);
3138 btrfs_set_token_inode_size(leaf, item, 0, &token);
Josef Bacik94edf4a2012-09-25 14:56:25 -04003139 } else {
Josef Bacik0b1c6cc2012-10-23 16:03:44 -04003140 btrfs_set_token_inode_generation(leaf, item,
3141 BTRFS_I(inode)->generation,
3142 &token);
3143 btrfs_set_token_inode_size(leaf, item, inode->i_size, &token);
Josef Bacik94edf4a2012-09-25 14:56:25 -04003144 }
3145
Josef Bacik0b1c6cc2012-10-23 16:03:44 -04003146 btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token);
3147 btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token);
3148 btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token);
3149 btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token);
3150
3151 btrfs_set_token_timespec_sec(leaf, btrfs_inode_atime(item),
3152 inode->i_atime.tv_sec, &token);
3153 btrfs_set_token_timespec_nsec(leaf, btrfs_inode_atime(item),
3154 inode->i_atime.tv_nsec, &token);
3155
3156 btrfs_set_token_timespec_sec(leaf, btrfs_inode_mtime(item),
3157 inode->i_mtime.tv_sec, &token);
3158 btrfs_set_token_timespec_nsec(leaf, btrfs_inode_mtime(item),
3159 inode->i_mtime.tv_nsec, &token);
3160
3161 btrfs_set_token_timespec_sec(leaf, btrfs_inode_ctime(item),
3162 inode->i_ctime.tv_sec, &token);
3163 btrfs_set_token_timespec_nsec(leaf, btrfs_inode_ctime(item),
3164 inode->i_ctime.tv_nsec, &token);
3165
3166 btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode),
3167 &token);
3168
3169 btrfs_set_token_inode_sequence(leaf, item, inode->i_version, &token);
3170 btrfs_set_token_inode_transid(leaf, item, trans->transid, &token);
3171 btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token);
3172 btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token);
3173 btrfs_set_token_inode_block_group(leaf, item, 0, &token);
Josef Bacik94edf4a2012-09-25 14:56:25 -04003174}
3175
Josef Bacika95249b2012-10-11 16:17:34 -04003176static int log_inode_item(struct btrfs_trans_handle *trans,
3177 struct btrfs_root *log, struct btrfs_path *path,
3178 struct inode *inode)
3179{
3180 struct btrfs_inode_item *inode_item;
Josef Bacika95249b2012-10-11 16:17:34 -04003181 int ret;
3182
Filipe David Borba Mananaefd0c402013-10-07 21:20:44 +01003183 ret = btrfs_insert_empty_item(trans, log, path,
3184 &BTRFS_I(inode)->location,
Josef Bacika95249b2012-10-11 16:17:34 -04003185 sizeof(*inode_item));
3186 if (ret && ret != -EEXIST)
3187 return ret;
3188 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3189 struct btrfs_inode_item);
3190 fill_inode_item(trans, path->nodes[0], inode_item, inode, 0);
3191 btrfs_release_path(path);
3192 return 0;
3193}
3194
Chris Mason31ff1cd2008-09-11 16:17:57 -04003195static noinline int copy_items(struct btrfs_trans_handle *trans,
Liu Bod2794402012-08-29 01:07:56 -06003196 struct inode *inode,
Chris Mason31ff1cd2008-09-11 16:17:57 -04003197 struct btrfs_path *dst_path,
Josef Bacik16e75492013-10-22 12:18:51 -04003198 struct btrfs_path *src_path, u64 *last_extent,
Chris Mason31ff1cd2008-09-11 16:17:57 -04003199 int start_slot, int nr, int inode_only)
3200{
3201 unsigned long src_offset;
3202 unsigned long dst_offset;
Liu Bod2794402012-08-29 01:07:56 -06003203 struct btrfs_root *log = BTRFS_I(inode)->root->log_root;
Chris Mason31ff1cd2008-09-11 16:17:57 -04003204 struct btrfs_file_extent_item *extent;
3205 struct btrfs_inode_item *inode_item;
Josef Bacik16e75492013-10-22 12:18:51 -04003206 struct extent_buffer *src = src_path->nodes[0];
3207 struct btrfs_key first_key, last_key, key;
Chris Mason31ff1cd2008-09-11 16:17:57 -04003208 int ret;
3209 struct btrfs_key *ins_keys;
3210 u32 *ins_sizes;
3211 char *ins_data;
3212 int i;
Chris Masond20f7042008-12-08 16:58:54 -05003213 struct list_head ordered_sums;
Liu Bod2794402012-08-29 01:07:56 -06003214 int skip_csum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
Josef Bacik16e75492013-10-22 12:18:51 -04003215 bool has_extents = false;
3216 bool need_find_last_extent = (*last_extent == 0);
3217 bool done = false;
Chris Masond20f7042008-12-08 16:58:54 -05003218
3219 INIT_LIST_HEAD(&ordered_sums);
Chris Mason31ff1cd2008-09-11 16:17:57 -04003220
3221 ins_data = kmalloc(nr * sizeof(struct btrfs_key) +
3222 nr * sizeof(u32), GFP_NOFS);
liubo2a29edc2011-01-26 06:22:08 +00003223 if (!ins_data)
3224 return -ENOMEM;
3225
Josef Bacik16e75492013-10-22 12:18:51 -04003226 first_key.objectid = (u64)-1;
3227
Chris Mason31ff1cd2008-09-11 16:17:57 -04003228 ins_sizes = (u32 *)ins_data;
3229 ins_keys = (struct btrfs_key *)(ins_data + nr * sizeof(u32));
3230
3231 for (i = 0; i < nr; i++) {
3232 ins_sizes[i] = btrfs_item_size_nr(src, i + start_slot);
3233 btrfs_item_key_to_cpu(src, ins_keys + i, i + start_slot);
3234 }
3235 ret = btrfs_insert_empty_items(trans, log, dst_path,
3236 ins_keys, ins_sizes, nr);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04003237 if (ret) {
3238 kfree(ins_data);
3239 return ret;
3240 }
Chris Mason31ff1cd2008-09-11 16:17:57 -04003241
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003242 for (i = 0; i < nr; i++, dst_path->slots[0]++) {
Chris Mason31ff1cd2008-09-11 16:17:57 -04003243 dst_offset = btrfs_item_ptr_offset(dst_path->nodes[0],
3244 dst_path->slots[0]);
3245
3246 src_offset = btrfs_item_ptr_offset(src, start_slot + i);
3247
Josef Bacik16e75492013-10-22 12:18:51 -04003248 if ((i == (nr - 1)))
3249 last_key = ins_keys[i];
3250
Josef Bacik94edf4a2012-09-25 14:56:25 -04003251 if (ins_keys[i].type == BTRFS_INODE_ITEM_KEY) {
Chris Mason31ff1cd2008-09-11 16:17:57 -04003252 inode_item = btrfs_item_ptr(dst_path->nodes[0],
3253 dst_path->slots[0],
3254 struct btrfs_inode_item);
Josef Bacik94edf4a2012-09-25 14:56:25 -04003255 fill_inode_item(trans, dst_path->nodes[0], inode_item,
3256 inode, inode_only == LOG_INODE_EXISTS);
3257 } else {
3258 copy_extent_buffer(dst_path->nodes[0], src, dst_offset,
3259 src_offset, ins_sizes[i]);
Chris Mason31ff1cd2008-09-11 16:17:57 -04003260 }
Josef Bacik94edf4a2012-09-25 14:56:25 -04003261
Josef Bacik16e75492013-10-22 12:18:51 -04003262 /*
3263 * We set need_find_last_extent here in case we know we were
3264 * processing other items and then walk into the first extent in
3265 * the inode. If we don't hit an extent then nothing changes,
3266 * we'll do the last search the next time around.
3267 */
3268 if (ins_keys[i].type == BTRFS_EXTENT_DATA_KEY) {
3269 has_extents = true;
3270 if (need_find_last_extent &&
3271 first_key.objectid == (u64)-1)
3272 first_key = ins_keys[i];
3273 } else {
3274 need_find_last_extent = false;
3275 }
3276
Chris Mason31ff1cd2008-09-11 16:17:57 -04003277 /* take a reference on file data extents so that truncates
3278 * or deletes of this inode don't have to relog the inode
3279 * again
3280 */
Liu Bod2794402012-08-29 01:07:56 -06003281 if (btrfs_key_type(ins_keys + i) == BTRFS_EXTENT_DATA_KEY &&
3282 !skip_csum) {
Chris Mason31ff1cd2008-09-11 16:17:57 -04003283 int found_type;
3284 extent = btrfs_item_ptr(src, start_slot + i,
3285 struct btrfs_file_extent_item);
3286
liubo8e531cd2011-05-06 10:36:09 +08003287 if (btrfs_file_extent_generation(src, extent) < trans->transid)
3288 continue;
3289
Chris Mason31ff1cd2008-09-11 16:17:57 -04003290 found_type = btrfs_file_extent_type(src, extent);
Josef Bacik6f1fed72012-09-26 11:07:06 -04003291 if (found_type == BTRFS_FILE_EXTENT_REG) {
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003292 u64 ds, dl, cs, cl;
3293 ds = btrfs_file_extent_disk_bytenr(src,
3294 extent);
3295 /* ds == 0 is a hole */
3296 if (ds == 0)
3297 continue;
3298
3299 dl = btrfs_file_extent_disk_num_bytes(src,
3300 extent);
3301 cs = btrfs_file_extent_offset(src, extent);
3302 cl = btrfs_file_extent_num_bytes(src,
Joe Perchesa419aef2009-08-18 11:18:35 -07003303 extent);
Chris Mason580afd72008-12-08 19:15:39 -05003304 if (btrfs_file_extent_compression(src,
3305 extent)) {
3306 cs = 0;
3307 cl = dl;
3308 }
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003309
3310 ret = btrfs_lookup_csums_range(
3311 log->fs_info->csum_root,
3312 ds + cs, ds + cs + cl - 1,
Arne Jansena2de7332011-03-08 14:14:00 +01003313 &ordered_sums, 0);
Josef Bacik36508602013-04-25 16:23:32 -04003314 if (ret) {
3315 btrfs_release_path(dst_path);
3316 kfree(ins_data);
3317 return ret;
3318 }
Chris Mason31ff1cd2008-09-11 16:17:57 -04003319 }
3320 }
Chris Mason31ff1cd2008-09-11 16:17:57 -04003321 }
3322
3323 btrfs_mark_buffer_dirty(dst_path->nodes[0]);
David Sterbab3b4aa72011-04-21 01:20:15 +02003324 btrfs_release_path(dst_path);
Chris Mason31ff1cd2008-09-11 16:17:57 -04003325 kfree(ins_data);
Chris Masond20f7042008-12-08 16:58:54 -05003326
3327 /*
3328 * we have to do this after the loop above to avoid changing the
3329 * log tree while trying to change the log tree.
3330 */
Yan, Zheng4a500fd2010-05-16 10:49:59 -04003331 ret = 0;
Chris Masond3977122009-01-05 21:25:51 -05003332 while (!list_empty(&ordered_sums)) {
Chris Masond20f7042008-12-08 16:58:54 -05003333 struct btrfs_ordered_sum *sums = list_entry(ordered_sums.next,
3334 struct btrfs_ordered_sum,
3335 list);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04003336 if (!ret)
3337 ret = btrfs_csum_file_blocks(trans, log, sums);
Chris Masond20f7042008-12-08 16:58:54 -05003338 list_del(&sums->list);
3339 kfree(sums);
3340 }
Josef Bacik16e75492013-10-22 12:18:51 -04003341
3342 if (!has_extents)
3343 return ret;
3344
3345 /*
3346 * Because we use btrfs_search_forward we could skip leaves that were
3347 * not modified and then assume *last_extent is valid when it really
3348 * isn't. So back up to the previous leaf and read the end of the last
3349 * extent before we go and fill in holes.
3350 */
3351 if (need_find_last_extent) {
3352 u64 len;
3353
3354 ret = btrfs_prev_leaf(BTRFS_I(inode)->root, src_path);
3355 if (ret < 0)
3356 return ret;
3357 if (ret)
3358 goto fill_holes;
3359 if (src_path->slots[0])
3360 src_path->slots[0]--;
3361 src = src_path->nodes[0];
3362 btrfs_item_key_to_cpu(src, &key, src_path->slots[0]);
3363 if (key.objectid != btrfs_ino(inode) ||
3364 key.type != BTRFS_EXTENT_DATA_KEY)
3365 goto fill_holes;
3366 extent = btrfs_item_ptr(src, src_path->slots[0],
3367 struct btrfs_file_extent_item);
3368 if (btrfs_file_extent_type(src, extent) ==
3369 BTRFS_FILE_EXTENT_INLINE) {
Chris Mason514ac8a2014-01-03 21:07:00 -08003370 len = btrfs_file_extent_inline_len(src,
3371 src_path->slots[0],
3372 extent);
Josef Bacik16e75492013-10-22 12:18:51 -04003373 *last_extent = ALIGN(key.offset + len,
3374 log->sectorsize);
3375 } else {
3376 len = btrfs_file_extent_num_bytes(src, extent);
3377 *last_extent = key.offset + len;
3378 }
3379 }
3380fill_holes:
3381 /* So we did prev_leaf, now we need to move to the next leaf, but a few
3382 * things could have happened
3383 *
3384 * 1) A merge could have happened, so we could currently be on a leaf
3385 * that holds what we were copying in the first place.
3386 * 2) A split could have happened, and now not all of the items we want
3387 * are on the same leaf.
3388 *
3389 * So we need to adjust how we search for holes, we need to drop the
3390 * path and re-search for the first extent key we found, and then walk
3391 * forward until we hit the last one we copied.
3392 */
3393 if (need_find_last_extent) {
3394 /* btrfs_prev_leaf could return 1 without releasing the path */
3395 btrfs_release_path(src_path);
3396 ret = btrfs_search_slot(NULL, BTRFS_I(inode)->root, &first_key,
3397 src_path, 0, 0);
3398 if (ret < 0)
3399 return ret;
3400 ASSERT(ret == 0);
3401 src = src_path->nodes[0];
3402 i = src_path->slots[0];
3403 } else {
3404 i = start_slot;
3405 }
3406
3407 /*
3408 * Ok so here we need to go through and fill in any holes we may have
3409 * to make sure that holes are punched for those areas in case they had
3410 * extents previously.
3411 */
3412 while (!done) {
3413 u64 offset, len;
3414 u64 extent_end;
3415
3416 if (i >= btrfs_header_nritems(src_path->nodes[0])) {
3417 ret = btrfs_next_leaf(BTRFS_I(inode)->root, src_path);
3418 if (ret < 0)
3419 return ret;
3420 ASSERT(ret == 0);
3421 src = src_path->nodes[0];
3422 i = 0;
3423 }
3424
3425 btrfs_item_key_to_cpu(src, &key, i);
3426 if (!btrfs_comp_cpu_keys(&key, &last_key))
3427 done = true;
3428 if (key.objectid != btrfs_ino(inode) ||
3429 key.type != BTRFS_EXTENT_DATA_KEY) {
3430 i++;
3431 continue;
3432 }
3433 extent = btrfs_item_ptr(src, i, struct btrfs_file_extent_item);
3434 if (btrfs_file_extent_type(src, extent) ==
3435 BTRFS_FILE_EXTENT_INLINE) {
Chris Mason514ac8a2014-01-03 21:07:00 -08003436 len = btrfs_file_extent_inline_len(src, i, extent);
Josef Bacik16e75492013-10-22 12:18:51 -04003437 extent_end = ALIGN(key.offset + len, log->sectorsize);
3438 } else {
3439 len = btrfs_file_extent_num_bytes(src, extent);
3440 extent_end = key.offset + len;
3441 }
3442 i++;
3443
3444 if (*last_extent == key.offset) {
3445 *last_extent = extent_end;
3446 continue;
3447 }
3448 offset = *last_extent;
3449 len = key.offset - *last_extent;
3450 ret = btrfs_insert_file_extent(trans, log, btrfs_ino(inode),
3451 offset, 0, 0, len, 0, len, 0,
3452 0, 0);
3453 if (ret)
3454 break;
3455 *last_extent = offset + len;
3456 }
3457 /*
3458 * Need to let the callers know we dropped the path so they should
3459 * re-search.
3460 */
3461 if (!ret && need_find_last_extent)
3462 ret = 1;
Yan, Zheng4a500fd2010-05-16 10:49:59 -04003463 return ret;
Chris Mason31ff1cd2008-09-11 16:17:57 -04003464}
3465
Josef Bacik5dc562c2012-08-17 13:14:17 -04003466static int extent_cmp(void *priv, struct list_head *a, struct list_head *b)
3467{
3468 struct extent_map *em1, *em2;
3469
3470 em1 = list_entry(a, struct extent_map, list);
3471 em2 = list_entry(b, struct extent_map, list);
3472
3473 if (em1->start < em2->start)
3474 return -1;
3475 else if (em1->start > em2->start)
3476 return 1;
3477 return 0;
3478}
3479
Josef Bacik5dc562c2012-08-17 13:14:17 -04003480static int log_one_extent(struct btrfs_trans_handle *trans,
3481 struct inode *inode, struct btrfs_root *root,
Josef Bacik70c8a912012-10-11 16:54:30 -04003482 struct extent_map *em, struct btrfs_path *path)
Josef Bacik5dc562c2012-08-17 13:14:17 -04003483{
3484 struct btrfs_root *log = root->log_root;
Josef Bacik70c8a912012-10-11 16:54:30 -04003485 struct btrfs_file_extent_item *fi;
3486 struct extent_buffer *leaf;
Josef Bacik2ab28f32012-10-12 15:27:49 -04003487 struct btrfs_ordered_extent *ordered;
Josef Bacik70c8a912012-10-11 16:54:30 -04003488 struct list_head ordered_sums;
Josef Bacik0b1c6cc2012-10-23 16:03:44 -04003489 struct btrfs_map_token token;
Josef Bacik5dc562c2012-08-17 13:14:17 -04003490 struct btrfs_key key;
Josef Bacik2ab28f32012-10-12 15:27:49 -04003491 u64 mod_start = em->mod_start;
3492 u64 mod_len = em->mod_len;
3493 u64 csum_offset;
3494 u64 csum_len;
Josef Bacik70c8a912012-10-11 16:54:30 -04003495 u64 extent_offset = em->start - em->orig_start;
3496 u64 block_len;
Josef Bacik5dc562c2012-08-17 13:14:17 -04003497 int ret;
Josef Bacik2ab28f32012-10-12 15:27:49 -04003498 int index = log->log_transid % 2;
Josef Bacik70c8a912012-10-11 16:54:30 -04003499 bool skip_csum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
Filipe David Borba Manana1acae572014-01-07 11:42:27 +00003500 int extent_inserted = 0;
Josef Bacik09a2a8f92013-04-05 16:51:15 -04003501
Josef Bacik70c8a912012-10-11 16:54:30 -04003502 INIT_LIST_HEAD(&ordered_sums);
Josef Bacik0b1c6cc2012-10-23 16:03:44 -04003503 btrfs_init_map_token(&token);
Josef Bacik70c8a912012-10-11 16:54:30 -04003504
Filipe David Borba Manana1acae572014-01-07 11:42:27 +00003505 ret = __btrfs_drop_extents(trans, log, inode, path, em->start,
3506 em->start + em->len, NULL, 0, 1,
3507 sizeof(*fi), &extent_inserted);
Josef Bacik09a2a8f92013-04-05 16:51:15 -04003508 if (ret)
Josef Bacik70c8a912012-10-11 16:54:30 -04003509 return ret;
Filipe David Borba Manana1acae572014-01-07 11:42:27 +00003510
3511 if (!extent_inserted) {
3512 key.objectid = btrfs_ino(inode);
3513 key.type = BTRFS_EXTENT_DATA_KEY;
3514 key.offset = em->start;
3515
3516 ret = btrfs_insert_empty_item(trans, log, path, &key,
3517 sizeof(*fi));
3518 if (ret)
3519 return ret;
3520 }
Josef Bacik70c8a912012-10-11 16:54:30 -04003521 leaf = path->nodes[0];
3522 fi = btrfs_item_ptr(leaf, path->slots[0],
3523 struct btrfs_file_extent_item);
Josef Bacik124fe662013-03-01 11:47:21 -05003524
Josef Bacik0b1c6cc2012-10-23 16:03:44 -04003525 btrfs_set_token_file_extent_generation(leaf, fi, em->generation,
3526 &token);
Josef Bacik70c8a912012-10-11 16:54:30 -04003527 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
3528 skip_csum = true;
Josef Bacik0b1c6cc2012-10-23 16:03:44 -04003529 btrfs_set_token_file_extent_type(leaf, fi,
3530 BTRFS_FILE_EXTENT_PREALLOC,
3531 &token);
Josef Bacik70c8a912012-10-11 16:54:30 -04003532 } else {
Josef Bacik0b1c6cc2012-10-23 16:03:44 -04003533 btrfs_set_token_file_extent_type(leaf, fi,
3534 BTRFS_FILE_EXTENT_REG,
3535 &token);
Josef Baciked9e8af2013-10-14 17:23:08 -04003536 if (em->block_start == EXTENT_MAP_HOLE)
Josef Bacik70c8a912012-10-11 16:54:30 -04003537 skip_csum = true;
Josef Bacik5dc562c2012-08-17 13:14:17 -04003538 }
3539
Josef Bacik70c8a912012-10-11 16:54:30 -04003540 block_len = max(em->block_len, em->orig_block_len);
3541 if (em->compress_type != BTRFS_COMPRESS_NONE) {
Josef Bacik0b1c6cc2012-10-23 16:03:44 -04003542 btrfs_set_token_file_extent_disk_bytenr(leaf, fi,
3543 em->block_start,
3544 &token);
3545 btrfs_set_token_file_extent_disk_num_bytes(leaf, fi, block_len,
3546 &token);
Josef Bacik70c8a912012-10-11 16:54:30 -04003547 } else if (em->block_start < EXTENT_MAP_LAST_BYTE) {
Josef Bacik0b1c6cc2012-10-23 16:03:44 -04003548 btrfs_set_token_file_extent_disk_bytenr(leaf, fi,
3549 em->block_start -
3550 extent_offset, &token);
3551 btrfs_set_token_file_extent_disk_num_bytes(leaf, fi, block_len,
3552 &token);
Josef Bacik70c8a912012-10-11 16:54:30 -04003553 } else {
Josef Bacik0b1c6cc2012-10-23 16:03:44 -04003554 btrfs_set_token_file_extent_disk_bytenr(leaf, fi, 0, &token);
3555 btrfs_set_token_file_extent_disk_num_bytes(leaf, fi, 0,
3556 &token);
Josef Bacik5dc562c2012-08-17 13:14:17 -04003557 }
3558
Josef Bacik0b1c6cc2012-10-23 16:03:44 -04003559 btrfs_set_token_file_extent_offset(leaf, fi,
3560 em->start - em->orig_start,
3561 &token);
3562 btrfs_set_token_file_extent_num_bytes(leaf, fi, em->len, &token);
Josef Bacikcc95bef2013-04-04 14:31:27 -04003563 btrfs_set_token_file_extent_ram_bytes(leaf, fi, em->ram_bytes, &token);
Josef Bacik0b1c6cc2012-10-23 16:03:44 -04003564 btrfs_set_token_file_extent_compression(leaf, fi, em->compress_type,
3565 &token);
3566 btrfs_set_token_file_extent_encryption(leaf, fi, 0, &token);
3567 btrfs_set_token_file_extent_other_encoding(leaf, fi, 0, &token);
Josef Bacik70c8a912012-10-11 16:54:30 -04003568 btrfs_mark_buffer_dirty(leaf);
3569
Josef Bacik70c8a912012-10-11 16:54:30 -04003570 btrfs_release_path(path);
Josef Bacik70c8a912012-10-11 16:54:30 -04003571 if (ret) {
3572 return ret;
3573 }
3574
3575 if (skip_csum)
3576 return 0;
3577
Josef Bacik2ab28f32012-10-12 15:27:49 -04003578 /*
3579 * First check and see if our csums are on our outstanding ordered
3580 * extents.
3581 */
3582again:
3583 spin_lock_irq(&log->log_extents_lock[index]);
3584 list_for_each_entry(ordered, &log->logged_list[index], log_list) {
3585 struct btrfs_ordered_sum *sum;
3586
3587 if (!mod_len)
3588 break;
3589
3590 if (ordered->inode != inode)
3591 continue;
3592
3593 if (ordered->file_offset + ordered->len <= mod_start ||
3594 mod_start + mod_len <= ordered->file_offset)
3595 continue;
3596
3597 /*
3598 * We are going to copy all the csums on this ordered extent, so
3599 * go ahead and adjust mod_start and mod_len in case this
3600 * ordered extent has already been logged.
3601 */
3602 if (ordered->file_offset > mod_start) {
3603 if (ordered->file_offset + ordered->len >=
3604 mod_start + mod_len)
3605 mod_len = ordered->file_offset - mod_start;
3606 /*
3607 * If we have this case
3608 *
3609 * |--------- logged extent ---------|
3610 * |----- ordered extent ----|
3611 *
3612 * Just don't mess with mod_start and mod_len, we'll
3613 * just end up logging more csums than we need and it
3614 * will be ok.
3615 */
3616 } else {
3617 if (ordered->file_offset + ordered->len <
3618 mod_start + mod_len) {
3619 mod_len = (mod_start + mod_len) -
3620 (ordered->file_offset + ordered->len);
3621 mod_start = ordered->file_offset +
3622 ordered->len;
3623 } else {
3624 mod_len = 0;
3625 }
3626 }
3627
3628 /*
3629 * To keep us from looping for the above case of an ordered
3630 * extent that falls inside of the logged extent.
3631 */
3632 if (test_and_set_bit(BTRFS_ORDERED_LOGGED_CSUM,
3633 &ordered->flags))
3634 continue;
3635 atomic_inc(&ordered->refs);
3636 spin_unlock_irq(&log->log_extents_lock[index]);
3637 /*
3638 * we've dropped the lock, we must either break or
3639 * start over after this.
3640 */
3641
Miao Xie23c671a2014-01-14 20:31:52 +08003642 if (ordered->csum_bytes_left) {
3643 btrfs_start_ordered_extent(inode, ordered, 0);
3644 wait_event(ordered->wait,
3645 ordered->csum_bytes_left == 0);
3646 }
Josef Bacik2ab28f32012-10-12 15:27:49 -04003647
3648 list_for_each_entry(sum, &ordered->list, list) {
3649 ret = btrfs_csum_file_blocks(trans, log, sum);
3650 if (ret) {
3651 btrfs_put_ordered_extent(ordered);
3652 goto unlocked;
3653 }
3654 }
3655 btrfs_put_ordered_extent(ordered);
3656 goto again;
3657
3658 }
3659 spin_unlock_irq(&log->log_extents_lock[index]);
3660unlocked:
3661
3662 if (!mod_len || ret)
3663 return ret;
3664
Filipe David Borba Manana488111a2013-10-28 16:30:29 +00003665 if (em->compress_type) {
3666 csum_offset = 0;
3667 csum_len = block_len;
3668 } else {
3669 csum_offset = mod_start - em->start;
3670 csum_len = mod_len;
3671 }
Josef Bacik2ab28f32012-10-12 15:27:49 -04003672
Josef Bacik70c8a912012-10-11 16:54:30 -04003673 /* block start is already adjusted for the file extent offset. */
3674 ret = btrfs_lookup_csums_range(log->fs_info->csum_root,
3675 em->block_start + csum_offset,
3676 em->block_start + csum_offset +
3677 csum_len - 1, &ordered_sums, 0);
3678 if (ret)
3679 return ret;
3680
3681 while (!list_empty(&ordered_sums)) {
3682 struct btrfs_ordered_sum *sums = list_entry(ordered_sums.next,
3683 struct btrfs_ordered_sum,
3684 list);
3685 if (!ret)
3686 ret = btrfs_csum_file_blocks(trans, log, sums);
3687 list_del(&sums->list);
3688 kfree(sums);
3689 }
3690
3691 return ret;
Josef Bacik5dc562c2012-08-17 13:14:17 -04003692}
3693
3694static int btrfs_log_changed_extents(struct btrfs_trans_handle *trans,
3695 struct btrfs_root *root,
3696 struct inode *inode,
Josef Bacik70c8a912012-10-11 16:54:30 -04003697 struct btrfs_path *path)
Josef Bacik5dc562c2012-08-17 13:14:17 -04003698{
Josef Bacik5dc562c2012-08-17 13:14:17 -04003699 struct extent_map *em, *n;
3700 struct list_head extents;
3701 struct extent_map_tree *tree = &BTRFS_I(inode)->extent_tree;
3702 u64 test_gen;
3703 int ret = 0;
Josef Bacik2ab28f32012-10-12 15:27:49 -04003704 int num = 0;
Josef Bacik5dc562c2012-08-17 13:14:17 -04003705
3706 INIT_LIST_HEAD(&extents);
3707
Josef Bacik5dc562c2012-08-17 13:14:17 -04003708 write_lock(&tree->lock);
3709 test_gen = root->fs_info->last_trans_committed;
3710
3711 list_for_each_entry_safe(em, n, &tree->modified_extents, list) {
3712 list_del_init(&em->list);
Josef Bacik2ab28f32012-10-12 15:27:49 -04003713
3714 /*
3715 * Just an arbitrary number, this can be really CPU intensive
3716 * once we start getting a lot of extents, and really once we
3717 * have a bunch of extents we just want to commit since it will
3718 * be faster.
3719 */
3720 if (++num > 32768) {
3721 list_del_init(&tree->modified_extents);
3722 ret = -EFBIG;
3723 goto process;
3724 }
3725
Josef Bacik5dc562c2012-08-17 13:14:17 -04003726 if (em->generation <= test_gen)
3727 continue;
Josef Bacikff44c6e2012-09-14 12:59:20 -04003728 /* Need a ref to keep it from getting evicted from cache */
3729 atomic_inc(&em->refs);
3730 set_bit(EXTENT_FLAG_LOGGING, &em->flags);
Josef Bacik5dc562c2012-08-17 13:14:17 -04003731 list_add_tail(&em->list, &extents);
Josef Bacik2ab28f32012-10-12 15:27:49 -04003732 num++;
Josef Bacik5dc562c2012-08-17 13:14:17 -04003733 }
3734
3735 list_sort(NULL, &extents, extent_cmp);
3736
Josef Bacik2ab28f32012-10-12 15:27:49 -04003737process:
Josef Bacik5dc562c2012-08-17 13:14:17 -04003738 while (!list_empty(&extents)) {
3739 em = list_entry(extents.next, struct extent_map, list);
3740
3741 list_del_init(&em->list);
3742
3743 /*
3744 * If we had an error we just need to delete everybody from our
3745 * private list.
3746 */
Josef Bacikff44c6e2012-09-14 12:59:20 -04003747 if (ret) {
Josef Bacik201a9032013-01-24 12:02:07 -05003748 clear_em_logging(tree, em);
Josef Bacikff44c6e2012-09-14 12:59:20 -04003749 free_extent_map(em);
Josef Bacik5dc562c2012-08-17 13:14:17 -04003750 continue;
Josef Bacikff44c6e2012-09-14 12:59:20 -04003751 }
3752
3753 write_unlock(&tree->lock);
Josef Bacik5dc562c2012-08-17 13:14:17 -04003754
Josef Bacik70c8a912012-10-11 16:54:30 -04003755 ret = log_one_extent(trans, inode, root, em, path);
Josef Bacikff44c6e2012-09-14 12:59:20 -04003756 write_lock(&tree->lock);
Josef Bacik201a9032013-01-24 12:02:07 -05003757 clear_em_logging(tree, em);
3758 free_extent_map(em);
Josef Bacik5dc562c2012-08-17 13:14:17 -04003759 }
Josef Bacikff44c6e2012-09-14 12:59:20 -04003760 WARN_ON(!list_empty(&extents));
3761 write_unlock(&tree->lock);
Josef Bacik5dc562c2012-08-17 13:14:17 -04003762
Josef Bacik5dc562c2012-08-17 13:14:17 -04003763 btrfs_release_path(path);
Josef Bacik5dc562c2012-08-17 13:14:17 -04003764 return ret;
3765}
3766
Chris Masone02119d2008-09-05 16:13:11 -04003767/* log a single inode in the tree log.
3768 * At least one parent directory for this inode must exist in the tree
3769 * or be logged already.
3770 *
3771 * Any items from this inode changed by the current transaction are copied
3772 * to the log tree. An extra reference is taken on any extents in this
3773 * file, allowing us to avoid a whole pile of corner cases around logging
3774 * blocks that have been removed from the tree.
3775 *
3776 * See LOG_INODE_ALL and related defines for a description of what inode_only
3777 * does.
3778 *
3779 * This handles both files and directories.
3780 */
Chris Mason12fcfd22009-03-24 10:24:20 -04003781static int btrfs_log_inode(struct btrfs_trans_handle *trans,
Chris Masone02119d2008-09-05 16:13:11 -04003782 struct btrfs_root *root, struct inode *inode,
3783 int inode_only)
3784{
3785 struct btrfs_path *path;
3786 struct btrfs_path *dst_path;
3787 struct btrfs_key min_key;
3788 struct btrfs_key max_key;
3789 struct btrfs_root *log = root->log_root;
Chris Mason31ff1cd2008-09-11 16:17:57 -04003790 struct extent_buffer *src = NULL;
Josef Bacik16e75492013-10-22 12:18:51 -04003791 u64 last_extent = 0;
Yan, Zheng4a500fd2010-05-16 10:49:59 -04003792 int err = 0;
Chris Masone02119d2008-09-05 16:13:11 -04003793 int ret;
Chris Mason3a5f1d42008-09-11 15:53:37 -04003794 int nritems;
Chris Mason31ff1cd2008-09-11 16:17:57 -04003795 int ins_start_slot = 0;
3796 int ins_nr;
Josef Bacik5dc562c2012-08-17 13:14:17 -04003797 bool fast_search = false;
Li Zefan33345d012011-04-20 10:31:50 +08003798 u64 ino = btrfs_ino(inode);
Chris Masone02119d2008-09-05 16:13:11 -04003799
Chris Masone02119d2008-09-05 16:13:11 -04003800 path = btrfs_alloc_path();
Tsutomu Itoh5df67082011-02-01 09:17:35 +00003801 if (!path)
3802 return -ENOMEM;
Chris Masone02119d2008-09-05 16:13:11 -04003803 dst_path = btrfs_alloc_path();
Tsutomu Itoh5df67082011-02-01 09:17:35 +00003804 if (!dst_path) {
3805 btrfs_free_path(path);
3806 return -ENOMEM;
3807 }
Chris Masone02119d2008-09-05 16:13:11 -04003808
Li Zefan33345d012011-04-20 10:31:50 +08003809 min_key.objectid = ino;
Chris Masone02119d2008-09-05 16:13:11 -04003810 min_key.type = BTRFS_INODE_ITEM_KEY;
3811 min_key.offset = 0;
3812
Li Zefan33345d012011-04-20 10:31:50 +08003813 max_key.objectid = ino;
Chris Mason12fcfd22009-03-24 10:24:20 -04003814
Chris Mason12fcfd22009-03-24 10:24:20 -04003815
Josef Bacik5dc562c2012-08-17 13:14:17 -04003816 /* today the code can only do partial logging of directories */
Miao Xie5269b672012-11-01 07:35:23 +00003817 if (S_ISDIR(inode->i_mode) ||
3818 (!test_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3819 &BTRFS_I(inode)->runtime_flags) &&
3820 inode_only == LOG_INODE_EXISTS))
Chris Masone02119d2008-09-05 16:13:11 -04003821 max_key.type = BTRFS_XATTR_ITEM_KEY;
3822 else
3823 max_key.type = (u8)-1;
3824 max_key.offset = (u64)-1;
3825
Josef Bacik94edf4a2012-09-25 14:56:25 -04003826 /* Only run delayed items if we are a dir or a new file */
3827 if (S_ISDIR(inode->i_mode) ||
3828 BTRFS_I(inode)->generation > root->fs_info->last_trans_committed) {
3829 ret = btrfs_commit_inode_delayed_items(trans, inode);
3830 if (ret) {
3831 btrfs_free_path(path);
3832 btrfs_free_path(dst_path);
3833 return ret;
3834 }
Miao Xie16cdcec2011-04-22 18:12:22 +08003835 }
3836
Chris Masone02119d2008-09-05 16:13:11 -04003837 mutex_lock(&BTRFS_I(inode)->log_mutex);
3838
Josef Bacik2ab28f32012-10-12 15:27:49 -04003839 btrfs_get_logged_extents(log, inode);
3840
Chris Masone02119d2008-09-05 16:13:11 -04003841 /*
3842 * a brute force approach to making sure we get the most uptodate
3843 * copies of everything.
3844 */
3845 if (S_ISDIR(inode->i_mode)) {
3846 int max_key_type = BTRFS_DIR_LOG_INDEX_KEY;
3847
3848 if (inode_only == LOG_INODE_EXISTS)
3849 max_key_type = BTRFS_XATTR_ITEM_KEY;
Li Zefan33345d012011-04-20 10:31:50 +08003850 ret = drop_objectid_items(trans, log, path, ino, max_key_type);
Chris Masone02119d2008-09-05 16:13:11 -04003851 } else {
Josef Bacik5dc562c2012-08-17 13:14:17 -04003852 if (test_and_clear_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3853 &BTRFS_I(inode)->runtime_flags)) {
Josef Bacike9976152012-10-11 15:53:56 -04003854 clear_bit(BTRFS_INODE_COPY_EVERYTHING,
3855 &BTRFS_I(inode)->runtime_flags);
Josef Bacik5dc562c2012-08-17 13:14:17 -04003856 ret = btrfs_truncate_inode_items(trans, log,
3857 inode, 0, 0);
Josef Bacika95249b2012-10-11 16:17:34 -04003858 } else if (test_and_clear_bit(BTRFS_INODE_COPY_EVERYTHING,
Josef Bacik6cfab852013-11-12 16:25:58 -05003859 &BTRFS_I(inode)->runtime_flags) ||
3860 inode_only == LOG_INODE_EXISTS) {
Josef Bacika95249b2012-10-11 16:17:34 -04003861 if (inode_only == LOG_INODE_ALL)
3862 fast_search = true;
3863 max_key.type = BTRFS_XATTR_ITEM_KEY;
3864 ret = drop_objectid_items(trans, log, path, ino,
3865 max_key.type);
Josef Bacik5dc562c2012-08-17 13:14:17 -04003866 } else {
Liu Bo183f37f2012-11-01 06:38:47 +00003867 if (inode_only == LOG_INODE_ALL)
3868 fast_search = true;
Josef Bacika95249b2012-10-11 16:17:34 -04003869 ret = log_inode_item(trans, log, dst_path, inode);
3870 if (ret) {
3871 err = ret;
3872 goto out_unlock;
3873 }
3874 goto log_extents;
Josef Bacik5dc562c2012-08-17 13:14:17 -04003875 }
Josef Bacika95249b2012-10-11 16:17:34 -04003876
Chris Masone02119d2008-09-05 16:13:11 -04003877 }
Yan, Zheng4a500fd2010-05-16 10:49:59 -04003878 if (ret) {
3879 err = ret;
3880 goto out_unlock;
3881 }
Chris Masone02119d2008-09-05 16:13:11 -04003882 path->keep_locks = 1;
3883
Chris Masond3977122009-01-05 21:25:51 -05003884 while (1) {
Chris Mason31ff1cd2008-09-11 16:17:57 -04003885 ins_nr = 0;
Filipe David Borba Manana6174d3c2013-10-01 16:13:42 +01003886 ret = btrfs_search_forward(root, &min_key,
Eric Sandeende78b512013-01-31 18:21:12 +00003887 path, trans->transid);
Chris Masone02119d2008-09-05 16:13:11 -04003888 if (ret != 0)
3889 break;
Chris Mason3a5f1d42008-09-11 15:53:37 -04003890again:
Chris Mason31ff1cd2008-09-11 16:17:57 -04003891 /* note, ins_nr might be > 0 here, cleanup outside the loop */
Li Zefan33345d012011-04-20 10:31:50 +08003892 if (min_key.objectid != ino)
Chris Masone02119d2008-09-05 16:13:11 -04003893 break;
3894 if (min_key.type > max_key.type)
3895 break;
Chris Mason31ff1cd2008-09-11 16:17:57 -04003896
Chris Masone02119d2008-09-05 16:13:11 -04003897 src = path->nodes[0];
Chris Mason31ff1cd2008-09-11 16:17:57 -04003898 if (ins_nr && ins_start_slot + ins_nr == path->slots[0]) {
3899 ins_nr++;
3900 goto next_slot;
3901 } else if (!ins_nr) {
3902 ins_start_slot = path->slots[0];
3903 ins_nr = 1;
3904 goto next_slot;
Chris Masone02119d2008-09-05 16:13:11 -04003905 }
3906
Josef Bacik16e75492013-10-22 12:18:51 -04003907 ret = copy_items(trans, inode, dst_path, path, &last_extent,
3908 ins_start_slot, ins_nr, inode_only);
3909 if (ret < 0) {
Yan, Zheng4a500fd2010-05-16 10:49:59 -04003910 err = ret;
3911 goto out_unlock;
Josef Bacik16e75492013-10-22 12:18:51 -04003912 } if (ret) {
3913 ins_nr = 0;
3914 btrfs_release_path(path);
3915 continue;
Yan, Zheng4a500fd2010-05-16 10:49:59 -04003916 }
Chris Mason31ff1cd2008-09-11 16:17:57 -04003917 ins_nr = 1;
3918 ins_start_slot = path->slots[0];
3919next_slot:
Chris Masone02119d2008-09-05 16:13:11 -04003920
Chris Mason3a5f1d42008-09-11 15:53:37 -04003921 nritems = btrfs_header_nritems(path->nodes[0]);
3922 path->slots[0]++;
3923 if (path->slots[0] < nritems) {
3924 btrfs_item_key_to_cpu(path->nodes[0], &min_key,
3925 path->slots[0]);
3926 goto again;
3927 }
Chris Mason31ff1cd2008-09-11 16:17:57 -04003928 if (ins_nr) {
Josef Bacik16e75492013-10-22 12:18:51 -04003929 ret = copy_items(trans, inode, dst_path, path,
3930 &last_extent, ins_start_slot,
Chris Mason31ff1cd2008-09-11 16:17:57 -04003931 ins_nr, inode_only);
Josef Bacik16e75492013-10-22 12:18:51 -04003932 if (ret < 0) {
Yan, Zheng4a500fd2010-05-16 10:49:59 -04003933 err = ret;
3934 goto out_unlock;
3935 }
Josef Bacik16e75492013-10-22 12:18:51 -04003936 ret = 0;
Chris Mason31ff1cd2008-09-11 16:17:57 -04003937 ins_nr = 0;
3938 }
David Sterbab3b4aa72011-04-21 01:20:15 +02003939 btrfs_release_path(path);
Chris Mason3a5f1d42008-09-11 15:53:37 -04003940
Filipe David Borba Manana3d41d702013-10-01 17:06:53 +01003941 if (min_key.offset < (u64)-1) {
Chris Masone02119d2008-09-05 16:13:11 -04003942 min_key.offset++;
Filipe David Borba Manana3d41d702013-10-01 17:06:53 +01003943 } else if (min_key.type < max_key.type) {
Chris Masone02119d2008-09-05 16:13:11 -04003944 min_key.type++;
Filipe David Borba Manana3d41d702013-10-01 17:06:53 +01003945 min_key.offset = 0;
3946 } else {
Chris Masone02119d2008-09-05 16:13:11 -04003947 break;
Filipe David Borba Manana3d41d702013-10-01 17:06:53 +01003948 }
Chris Masone02119d2008-09-05 16:13:11 -04003949 }
Chris Mason31ff1cd2008-09-11 16:17:57 -04003950 if (ins_nr) {
Josef Bacik16e75492013-10-22 12:18:51 -04003951 ret = copy_items(trans, inode, dst_path, path, &last_extent,
3952 ins_start_slot, ins_nr, inode_only);
3953 if (ret < 0) {
Yan, Zheng4a500fd2010-05-16 10:49:59 -04003954 err = ret;
3955 goto out_unlock;
3956 }
Josef Bacik16e75492013-10-22 12:18:51 -04003957 ret = 0;
Chris Mason31ff1cd2008-09-11 16:17:57 -04003958 ins_nr = 0;
3959 }
Josef Bacik5dc562c2012-08-17 13:14:17 -04003960
Josef Bacika95249b2012-10-11 16:17:34 -04003961log_extents:
Josef Bacikf3b15cc2013-07-22 12:54:30 -04003962 btrfs_release_path(path);
3963 btrfs_release_path(dst_path);
Josef Bacik5dc562c2012-08-17 13:14:17 -04003964 if (fast_search) {
Josef Bacik70c8a912012-10-11 16:54:30 -04003965 ret = btrfs_log_changed_extents(trans, root, inode, dst_path);
Josef Bacik5dc562c2012-08-17 13:14:17 -04003966 if (ret) {
3967 err = ret;
3968 goto out_unlock;
3969 }
Josef Bacikd006a042013-11-12 20:54:09 -05003970 } else if (inode_only == LOG_INODE_ALL) {
Liu Bo06d3d222012-08-27 10:52:19 -06003971 struct extent_map_tree *tree = &BTRFS_I(inode)->extent_tree;
3972 struct extent_map *em, *n;
3973
Miao Xiebbe14262012-11-01 07:34:54 +00003974 write_lock(&tree->lock);
Liu Bo06d3d222012-08-27 10:52:19 -06003975 list_for_each_entry_safe(em, n, &tree->modified_extents, list)
3976 list_del_init(&em->list);
Miao Xiebbe14262012-11-01 07:34:54 +00003977 write_unlock(&tree->lock);
Josef Bacik5dc562c2012-08-17 13:14:17 -04003978 }
3979
Chris Mason9623f9a2008-09-11 17:42:42 -04003980 if (inode_only == LOG_INODE_ALL && S_ISDIR(inode->i_mode)) {
Chris Masone02119d2008-09-05 16:13:11 -04003981 ret = log_directory_changes(trans, root, inode, path, dst_path);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04003982 if (ret) {
3983 err = ret;
3984 goto out_unlock;
3985 }
Chris Masone02119d2008-09-05 16:13:11 -04003986 }
Chris Mason3a5f1d42008-09-11 15:53:37 -04003987 BTRFS_I(inode)->logged_trans = trans->transid;
Liu Bo46d8bc32012-08-29 01:07:55 -06003988 BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->last_sub_trans;
Yan, Zheng4a500fd2010-05-16 10:49:59 -04003989out_unlock:
Josef Bacik2ab28f32012-10-12 15:27:49 -04003990 if (err)
3991 btrfs_free_logged_extents(log, log->log_transid);
Chris Masone02119d2008-09-05 16:13:11 -04003992 mutex_unlock(&BTRFS_I(inode)->log_mutex);
3993
3994 btrfs_free_path(path);
3995 btrfs_free_path(dst_path);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04003996 return err;
Chris Masone02119d2008-09-05 16:13:11 -04003997}
3998
Chris Mason12fcfd22009-03-24 10:24:20 -04003999/*
4000 * follow the dentry parent pointers up the chain and see if any
4001 * of the directories in it require a full commit before they can
4002 * be logged. Returns zero if nothing special needs to be done or 1 if
4003 * a full commit is required.
4004 */
4005static noinline int check_parent_dirs_for_sync(struct btrfs_trans_handle *trans,
4006 struct inode *inode,
4007 struct dentry *parent,
4008 struct super_block *sb,
4009 u64 last_committed)
Chris Masone02119d2008-09-05 16:13:11 -04004010{
Chris Mason12fcfd22009-03-24 10:24:20 -04004011 int ret = 0;
4012 struct btrfs_root *root;
Josef Bacik6a912212010-11-20 09:48:00 +00004013 struct dentry *old_parent = NULL;
Josef Bacikde2b5302013-09-11 09:36:30 -04004014 struct inode *orig_inode = inode;
Chris Masone02119d2008-09-05 16:13:11 -04004015
Chris Masonaf4176b2009-03-24 10:24:31 -04004016 /*
4017 * for regular files, if its inode is already on disk, we don't
4018 * have to worry about the parents at all. This is because
4019 * we can use the last_unlink_trans field to record renames
4020 * and other fun in this file.
4021 */
4022 if (S_ISREG(inode->i_mode) &&
4023 BTRFS_I(inode)->generation <= last_committed &&
4024 BTRFS_I(inode)->last_unlink_trans <= last_committed)
4025 goto out;
4026
Chris Mason12fcfd22009-03-24 10:24:20 -04004027 if (!S_ISDIR(inode->i_mode)) {
4028 if (!parent || !parent->d_inode || sb != parent->d_inode->i_sb)
4029 goto out;
4030 inode = parent->d_inode;
4031 }
4032
4033 while (1) {
Josef Bacikde2b5302013-09-11 09:36:30 -04004034 /*
4035 * If we are logging a directory then we start with our inode,
4036 * not our parents inode, so we need to skipp setting the
4037 * logged_trans so that further down in the log code we don't
4038 * think this inode has already been logged.
4039 */
4040 if (inode != orig_inode)
4041 BTRFS_I(inode)->logged_trans = trans->transid;
Chris Mason12fcfd22009-03-24 10:24:20 -04004042 smp_mb();
4043
4044 if (BTRFS_I(inode)->last_unlink_trans > last_committed) {
4045 root = BTRFS_I(inode)->root;
4046
4047 /*
4048 * make sure any commits to the log are forced
4049 * to be full commits
4050 */
4051 root->fs_info->last_trans_log_full_commit =
4052 trans->transid;
4053 ret = 1;
4054 break;
4055 }
4056
4057 if (!parent || !parent->d_inode || sb != parent->d_inode->i_sb)
4058 break;
4059
Yan, Zheng76dda932009-09-21 16:00:26 -04004060 if (IS_ROOT(parent))
Chris Mason12fcfd22009-03-24 10:24:20 -04004061 break;
4062
Josef Bacik6a912212010-11-20 09:48:00 +00004063 parent = dget_parent(parent);
4064 dput(old_parent);
4065 old_parent = parent;
Chris Mason12fcfd22009-03-24 10:24:20 -04004066 inode = parent->d_inode;
4067
4068 }
Josef Bacik6a912212010-11-20 09:48:00 +00004069 dput(old_parent);
Chris Mason12fcfd22009-03-24 10:24:20 -04004070out:
Chris Masone02119d2008-09-05 16:13:11 -04004071 return ret;
4072}
4073
4074/*
4075 * helper function around btrfs_log_inode to make sure newly created
4076 * parent directories also end up in the log. A minimal inode and backref
4077 * only logging is done of any parent directories that are older than
4078 * the last committed transaction
4079 */
Eric Sandeen48a3b632013-04-25 20:41:01 +00004080static int btrfs_log_inode_parent(struct btrfs_trans_handle *trans,
4081 struct btrfs_root *root, struct inode *inode,
4082 struct dentry *parent, int exists_only)
Chris Masone02119d2008-09-05 16:13:11 -04004083{
Chris Mason12fcfd22009-03-24 10:24:20 -04004084 int inode_only = exists_only ? LOG_INODE_EXISTS : LOG_INODE_ALL;
Chris Masone02119d2008-09-05 16:13:11 -04004085 struct super_block *sb;
Josef Bacik6a912212010-11-20 09:48:00 +00004086 struct dentry *old_parent = NULL;
Chris Mason12fcfd22009-03-24 10:24:20 -04004087 int ret = 0;
4088 u64 last_committed = root->fs_info->last_trans_committed;
4089
4090 sb = inode->i_sb;
4091
Sage Weil3a5e1402009-04-02 16:49:40 -04004092 if (btrfs_test_opt(root, NOTREELOG)) {
4093 ret = 1;
4094 goto end_no_trans;
4095 }
4096
Chris Mason12fcfd22009-03-24 10:24:20 -04004097 if (root->fs_info->last_trans_log_full_commit >
4098 root->fs_info->last_trans_committed) {
4099 ret = 1;
4100 goto end_no_trans;
4101 }
4102
Yan, Zheng76dda932009-09-21 16:00:26 -04004103 if (root != BTRFS_I(inode)->root ||
4104 btrfs_root_refs(&root->root_item) == 0) {
4105 ret = 1;
4106 goto end_no_trans;
4107 }
4108
Chris Mason12fcfd22009-03-24 10:24:20 -04004109 ret = check_parent_dirs_for_sync(trans, inode, parent,
4110 sb, last_committed);
4111 if (ret)
4112 goto end_no_trans;
Chris Masone02119d2008-09-05 16:13:11 -04004113
Josef Bacik22ee6982012-05-29 16:57:49 -04004114 if (btrfs_inode_in_log(inode, trans->transid)) {
Chris Mason257c62e2009-10-13 13:21:08 -04004115 ret = BTRFS_NO_LOG_SYNC;
4116 goto end_no_trans;
4117 }
4118
Yan, Zheng4a500fd2010-05-16 10:49:59 -04004119 ret = start_log_trans(trans, root);
4120 if (ret)
4121 goto end_trans;
Chris Mason12fcfd22009-03-24 10:24:20 -04004122
4123 ret = btrfs_log_inode(trans, root, inode, inode_only);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04004124 if (ret)
4125 goto end_trans;
Chris Mason12fcfd22009-03-24 10:24:20 -04004126
Chris Masonaf4176b2009-03-24 10:24:31 -04004127 /*
4128 * for regular files, if its inode is already on disk, we don't
4129 * have to worry about the parents at all. This is because
4130 * we can use the last_unlink_trans field to record renames
4131 * and other fun in this file.
4132 */
4133 if (S_ISREG(inode->i_mode) &&
4134 BTRFS_I(inode)->generation <= last_committed &&
Yan, Zheng4a500fd2010-05-16 10:49:59 -04004135 BTRFS_I(inode)->last_unlink_trans <= last_committed) {
4136 ret = 0;
4137 goto end_trans;
4138 }
Chris Masonaf4176b2009-03-24 10:24:31 -04004139
4140 inode_only = LOG_INODE_EXISTS;
Chris Masond3977122009-01-05 21:25:51 -05004141 while (1) {
Chris Mason12fcfd22009-03-24 10:24:20 -04004142 if (!parent || !parent->d_inode || sb != parent->d_inode->i_sb)
Chris Masone02119d2008-09-05 16:13:11 -04004143 break;
4144
Chris Mason12fcfd22009-03-24 10:24:20 -04004145 inode = parent->d_inode;
Yan, Zheng76dda932009-09-21 16:00:26 -04004146 if (root != BTRFS_I(inode)->root)
4147 break;
4148
Chris Mason12fcfd22009-03-24 10:24:20 -04004149 if (BTRFS_I(inode)->generation >
4150 root->fs_info->last_trans_committed) {
4151 ret = btrfs_log_inode(trans, root, inode, inode_only);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04004152 if (ret)
4153 goto end_trans;
Chris Mason12fcfd22009-03-24 10:24:20 -04004154 }
Yan, Zheng76dda932009-09-21 16:00:26 -04004155 if (IS_ROOT(parent))
Chris Masone02119d2008-09-05 16:13:11 -04004156 break;
Chris Mason12fcfd22009-03-24 10:24:20 -04004157
Josef Bacik6a912212010-11-20 09:48:00 +00004158 parent = dget_parent(parent);
4159 dput(old_parent);
4160 old_parent = parent;
Chris Masone02119d2008-09-05 16:13:11 -04004161 }
Chris Mason12fcfd22009-03-24 10:24:20 -04004162 ret = 0;
Yan, Zheng4a500fd2010-05-16 10:49:59 -04004163end_trans:
Josef Bacik6a912212010-11-20 09:48:00 +00004164 dput(old_parent);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04004165 if (ret < 0) {
Yan, Zheng4a500fd2010-05-16 10:49:59 -04004166 root->fs_info->last_trans_log_full_commit = trans->transid;
4167 ret = 1;
4168 }
Chris Mason12fcfd22009-03-24 10:24:20 -04004169 btrfs_end_log_trans(root);
4170end_no_trans:
4171 return ret;
Chris Masone02119d2008-09-05 16:13:11 -04004172}
4173
4174/*
4175 * it is not safe to log dentry if the chunk root has added new
4176 * chunks. This returns 0 if the dentry was logged, and 1 otherwise.
4177 * If this returns 1, you must commit the transaction to safely get your
4178 * data on disk.
4179 */
4180int btrfs_log_dentry_safe(struct btrfs_trans_handle *trans,
4181 struct btrfs_root *root, struct dentry *dentry)
4182{
Josef Bacik6a912212010-11-20 09:48:00 +00004183 struct dentry *parent = dget_parent(dentry);
4184 int ret;
4185
4186 ret = btrfs_log_inode_parent(trans, root, dentry->d_inode, parent, 0);
4187 dput(parent);
4188
4189 return ret;
Chris Masone02119d2008-09-05 16:13:11 -04004190}
4191
4192/*
4193 * should be called during mount to recover any replay any log trees
4194 * from the FS
4195 */
4196int btrfs_recover_log_trees(struct btrfs_root *log_root_tree)
4197{
4198 int ret;
4199 struct btrfs_path *path;
4200 struct btrfs_trans_handle *trans;
4201 struct btrfs_key key;
4202 struct btrfs_key found_key;
4203 struct btrfs_key tmp_key;
4204 struct btrfs_root *log;
4205 struct btrfs_fs_info *fs_info = log_root_tree->fs_info;
4206 struct walk_control wc = {
4207 .process_func = process_one_buffer,
4208 .stage = 0,
4209 };
4210
Chris Masone02119d2008-09-05 16:13:11 -04004211 path = btrfs_alloc_path();
Tsutomu Itohdb5b4932011-03-23 08:14:16 +00004212 if (!path)
4213 return -ENOMEM;
4214
4215 fs_info->log_root_recovering = 1;
Chris Masone02119d2008-09-05 16:13:11 -04004216
Yan, Zheng4a500fd2010-05-16 10:49:59 -04004217 trans = btrfs_start_transaction(fs_info->tree_root, 0);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01004218 if (IS_ERR(trans)) {
4219 ret = PTR_ERR(trans);
4220 goto error;
4221 }
Chris Masone02119d2008-09-05 16:13:11 -04004222
4223 wc.trans = trans;
4224 wc.pin = 1;
4225
Tsutomu Itohdb5b4932011-03-23 08:14:16 +00004226 ret = walk_log_tree(trans, log_root_tree, &wc);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01004227 if (ret) {
4228 btrfs_error(fs_info, ret, "Failed to pin buffers while "
4229 "recovering log root tree.");
4230 goto error;
4231 }
Chris Masone02119d2008-09-05 16:13:11 -04004232
4233again:
4234 key.objectid = BTRFS_TREE_LOG_OBJECTID;
4235 key.offset = (u64)-1;
4236 btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
4237
Chris Masond3977122009-01-05 21:25:51 -05004238 while (1) {
Chris Masone02119d2008-09-05 16:13:11 -04004239 ret = btrfs_search_slot(NULL, log_root_tree, &key, path, 0, 0);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01004240
4241 if (ret < 0) {
4242 btrfs_error(fs_info, ret,
4243 "Couldn't find tree log root.");
4244 goto error;
4245 }
Chris Masone02119d2008-09-05 16:13:11 -04004246 if (ret > 0) {
4247 if (path->slots[0] == 0)
4248 break;
4249 path->slots[0]--;
4250 }
4251 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
4252 path->slots[0]);
David Sterbab3b4aa72011-04-21 01:20:15 +02004253 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04004254 if (found_key.objectid != BTRFS_TREE_LOG_OBJECTID)
4255 break;
4256
Miao Xiecb517ea2013-05-15 07:48:19 +00004257 log = btrfs_read_fs_root(log_root_tree, &found_key);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01004258 if (IS_ERR(log)) {
4259 ret = PTR_ERR(log);
4260 btrfs_error(fs_info, ret,
4261 "Couldn't read tree log root.");
4262 goto error;
4263 }
Chris Masone02119d2008-09-05 16:13:11 -04004264
4265 tmp_key.objectid = found_key.offset;
4266 tmp_key.type = BTRFS_ROOT_ITEM_KEY;
4267 tmp_key.offset = (u64)-1;
4268
4269 wc.replay_dest = btrfs_read_fs_root_no_name(fs_info, &tmp_key);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01004270 if (IS_ERR(wc.replay_dest)) {
4271 ret = PTR_ERR(wc.replay_dest);
Josef Bacikb50c6e22013-04-25 15:55:30 -04004272 free_extent_buffer(log->node);
4273 free_extent_buffer(log->commit_root);
4274 kfree(log);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01004275 btrfs_error(fs_info, ret, "Couldn't read target root "
4276 "for tree log recovery.");
4277 goto error;
4278 }
Chris Masone02119d2008-09-05 16:13:11 -04004279
Yan Zheng07d400a2009-01-06 11:42:00 -05004280 wc.replay_dest->log_root = log;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004281 btrfs_record_root_in_trans(trans, wc.replay_dest);
Chris Masone02119d2008-09-05 16:13:11 -04004282 ret = walk_log_tree(trans, log, &wc);
Chris Masone02119d2008-09-05 16:13:11 -04004283
Josef Bacikb50c6e22013-04-25 15:55:30 -04004284 if (!ret && wc.stage == LOG_WALK_REPLAY_ALL) {
Chris Masone02119d2008-09-05 16:13:11 -04004285 ret = fixup_inode_link_counts(trans, wc.replay_dest,
4286 path);
Chris Masone02119d2008-09-05 16:13:11 -04004287 }
Chris Masone02119d2008-09-05 16:13:11 -04004288
4289 key.offset = found_key.offset - 1;
Yan Zheng07d400a2009-01-06 11:42:00 -05004290 wc.replay_dest->log_root = NULL;
Chris Masone02119d2008-09-05 16:13:11 -04004291 free_extent_buffer(log->node);
Chris Masonb263c2c2009-06-11 11:24:47 -04004292 free_extent_buffer(log->commit_root);
Chris Masone02119d2008-09-05 16:13:11 -04004293 kfree(log);
4294
Josef Bacikb50c6e22013-04-25 15:55:30 -04004295 if (ret)
4296 goto error;
4297
Chris Masone02119d2008-09-05 16:13:11 -04004298 if (found_key.offset == 0)
4299 break;
4300 }
David Sterbab3b4aa72011-04-21 01:20:15 +02004301 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04004302
4303 /* step one is to pin it all, step two is to replay just inodes */
4304 if (wc.pin) {
4305 wc.pin = 0;
4306 wc.process_func = replay_one_buffer;
4307 wc.stage = LOG_WALK_REPLAY_INODES;
4308 goto again;
4309 }
4310 /* step three is to replay everything */
4311 if (wc.stage < LOG_WALK_REPLAY_ALL) {
4312 wc.stage++;
4313 goto again;
4314 }
4315
4316 btrfs_free_path(path);
4317
Josef Bacikabefa552013-04-24 16:40:05 -04004318 /* step 4: commit the transaction, which also unpins the blocks */
4319 ret = btrfs_commit_transaction(trans, fs_info->tree_root);
4320 if (ret)
4321 return ret;
4322
Chris Masone02119d2008-09-05 16:13:11 -04004323 free_extent_buffer(log_root_tree->node);
4324 log_root_tree->log_root = NULL;
4325 fs_info->log_root_recovering = 0;
Chris Masone02119d2008-09-05 16:13:11 -04004326 kfree(log_root_tree);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01004327
Josef Bacikabefa552013-04-24 16:40:05 -04004328 return 0;
Jeff Mahoney79787ea2012-03-12 16:03:00 +01004329error:
Josef Bacikb50c6e22013-04-25 15:55:30 -04004330 if (wc.trans)
4331 btrfs_end_transaction(wc.trans, fs_info->tree_root);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01004332 btrfs_free_path(path);
4333 return ret;
Chris Masone02119d2008-09-05 16:13:11 -04004334}
Chris Mason12fcfd22009-03-24 10:24:20 -04004335
4336/*
4337 * there are some corner cases where we want to force a full
4338 * commit instead of allowing a directory to be logged.
4339 *
4340 * They revolve around files there were unlinked from the directory, and
4341 * this function updates the parent directory so that a full commit is
4342 * properly done if it is fsync'd later after the unlinks are done.
4343 */
4344void btrfs_record_unlink_dir(struct btrfs_trans_handle *trans,
4345 struct inode *dir, struct inode *inode,
4346 int for_rename)
4347{
4348 /*
Chris Masonaf4176b2009-03-24 10:24:31 -04004349 * when we're logging a file, if it hasn't been renamed
4350 * or unlinked, and its inode is fully committed on disk,
4351 * we don't have to worry about walking up the directory chain
4352 * to log its parents.
4353 *
4354 * So, we use the last_unlink_trans field to put this transid
4355 * into the file. When the file is logged we check it and
4356 * don't log the parents if the file is fully on disk.
4357 */
4358 if (S_ISREG(inode->i_mode))
4359 BTRFS_I(inode)->last_unlink_trans = trans->transid;
4360
4361 /*
Chris Mason12fcfd22009-03-24 10:24:20 -04004362 * if this directory was already logged any new
4363 * names for this file/dir will get recorded
4364 */
4365 smp_mb();
4366 if (BTRFS_I(dir)->logged_trans == trans->transid)
4367 return;
4368
4369 /*
4370 * if the inode we're about to unlink was logged,
4371 * the log will be properly updated for any new names
4372 */
4373 if (BTRFS_I(inode)->logged_trans == trans->transid)
4374 return;
4375
4376 /*
4377 * when renaming files across directories, if the directory
4378 * there we're unlinking from gets fsync'd later on, there's
4379 * no way to find the destination directory later and fsync it
4380 * properly. So, we have to be conservative and force commits
4381 * so the new name gets discovered.
4382 */
4383 if (for_rename)
4384 goto record;
4385
4386 /* we can safely do the unlink without any special recording */
4387 return;
4388
4389record:
4390 BTRFS_I(dir)->last_unlink_trans = trans->transid;
4391}
4392
4393/*
4394 * Call this after adding a new name for a file and it will properly
4395 * update the log to reflect the new name.
4396 *
4397 * It will return zero if all goes well, and it will return 1 if a
4398 * full transaction commit is required.
4399 */
4400int btrfs_log_new_name(struct btrfs_trans_handle *trans,
4401 struct inode *inode, struct inode *old_dir,
4402 struct dentry *parent)
4403{
4404 struct btrfs_root * root = BTRFS_I(inode)->root;
4405
4406 /*
Chris Masonaf4176b2009-03-24 10:24:31 -04004407 * this will force the logging code to walk the dentry chain
4408 * up for the file
4409 */
4410 if (S_ISREG(inode->i_mode))
4411 BTRFS_I(inode)->last_unlink_trans = trans->transid;
4412
4413 /*
Chris Mason12fcfd22009-03-24 10:24:20 -04004414 * if this inode hasn't been logged and directory we're renaming it
4415 * from hasn't been logged, we don't need to log it
4416 */
4417 if (BTRFS_I(inode)->logged_trans <=
4418 root->fs_info->last_trans_committed &&
4419 (!old_dir || BTRFS_I(old_dir)->logged_trans <=
4420 root->fs_info->last_trans_committed))
4421 return 0;
4422
4423 return btrfs_log_inode_parent(trans, root, inode, parent, 1);
4424}
4425