blob: 1415f6d5863342ea35fcaf125b6484a121861b77 [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>
Miao Xie995946d2014-04-02 19:51:06 +080023#include "tree-log.h"
Chris Masone02119d2008-09-05 16:13:11 -040024#include "disk-io.h"
25#include "locking.h"
26#include "print-tree.h"
Mark Fashehf1863732012-08-08 11:32:27 -070027#include "backref.h"
Mark Fashehf1863732012-08-08 11:32:27 -070028#include "hash.h"
Chris Masone02119d2008-09-05 16:13:11 -040029
30/* magic values for the inode_only field in btrfs_log_inode:
31 *
32 * LOG_INODE_ALL means to log everything
33 * LOG_INODE_EXISTS means to log just enough to recreate the inode
34 * during log replay
35 */
36#define LOG_INODE_ALL 0
37#define LOG_INODE_EXISTS 1
38
39/*
Chris Mason12fcfd22009-03-24 10:24:20 -040040 * directory trouble cases
41 *
42 * 1) on rename or unlink, if the inode being unlinked isn't in the fsync
43 * log, we must force a full commit before doing an fsync of the directory
44 * where the unlink was done.
45 * ---> record transid of last unlink/rename per directory
46 *
47 * mkdir foo/some_dir
48 * normal commit
49 * rename foo/some_dir foo2/some_dir
50 * mkdir foo/some_dir
51 * fsync foo/some_dir/some_file
52 *
53 * The fsync above will unlink the original some_dir without recording
54 * it in its new location (foo2). After a crash, some_dir will be gone
55 * unless the fsync of some_file forces a full commit
56 *
57 * 2) we must log any new names for any file or dir that is in the fsync
58 * log. ---> check inode while renaming/linking.
59 *
60 * 2a) we must log any new names for any file or dir during rename
61 * when the directory they are being removed from was logged.
62 * ---> check inode and old parent dir during rename
63 *
64 * 2a is actually the more important variant. With the extra logging
65 * a crash might unlink the old name without recreating the new one
66 *
67 * 3) after a crash, we must go through any directories with a link count
68 * of zero and redo the rm -rf
69 *
70 * mkdir f1/foo
71 * normal commit
72 * rm -rf f1/foo
73 * fsync(f1)
74 *
75 * The directory f1 was fully removed from the FS, but fsync was never
76 * called on f1, only its parent dir. After a crash the rm -rf must
77 * be replayed. This must be able to recurse down the entire
78 * directory tree. The inode link count fixup code takes care of the
79 * ugly details.
80 */
81
82/*
Chris Masone02119d2008-09-05 16:13:11 -040083 * stages for the tree walking. The first
84 * stage (0) is to only pin down the blocks we find
85 * the second stage (1) is to make sure that all the inodes
86 * we find in the log are created in the subvolume.
87 *
88 * The last stage is to deal with directories and links and extents
89 * and all the other fun semantics
90 */
91#define LOG_WALK_PIN_ONLY 0
92#define LOG_WALK_REPLAY_INODES 1
Josef Bacikdd8e7212013-09-11 11:57:23 -040093#define LOG_WALK_REPLAY_DIR_INDEX 2
94#define LOG_WALK_REPLAY_ALL 3
Chris Masone02119d2008-09-05 16:13:11 -040095
Chris Mason12fcfd22009-03-24 10:24:20 -040096static int btrfs_log_inode(struct btrfs_trans_handle *trans,
Filipe Manana49dae1b2014-09-06 22:34:39 +010097 struct btrfs_root *root, struct inode *inode,
98 int inode_only,
99 const loff_t start,
Filipe Manana8407f552014-09-05 15:14:39 +0100100 const loff_t end,
101 struct btrfs_log_ctx *ctx);
Yan Zhengec051c02009-01-05 15:43:42 -0500102static int link_to_fixup_dir(struct btrfs_trans_handle *trans,
103 struct btrfs_root *root,
104 struct btrfs_path *path, u64 objectid);
Chris Mason12fcfd22009-03-24 10:24:20 -0400105static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans,
106 struct btrfs_root *root,
107 struct btrfs_root *log,
108 struct btrfs_path *path,
109 u64 dirid, int del_all);
Chris Masone02119d2008-09-05 16:13:11 -0400110
111/*
112 * tree logging is a special write ahead log used to make sure that
113 * fsyncs and O_SYNCs can happen without doing full tree commits.
114 *
115 * Full tree commits are expensive because they require commonly
116 * modified blocks to be recowed, creating many dirty pages in the
117 * extent tree an 4x-6x higher write load than ext3.
118 *
119 * Instead of doing a tree commit on every fsync, we use the
120 * key ranges and transaction ids to find items for a given file or directory
121 * that have changed in this transaction. Those items are copied into
122 * a special tree (one per subvolume root), that tree is written to disk
123 * and then the fsync is considered complete.
124 *
125 * After a crash, items are copied out of the log-tree back into the
126 * subvolume tree. Any file data extents found are recorded in the extent
127 * allocation tree, and the log-tree freed.
128 *
129 * The log tree is read three times, once to pin down all the extents it is
130 * using in ram and once, once to create all the inodes logged in the tree
131 * and once to do all the other items.
132 */
133
134/*
Chris Masone02119d2008-09-05 16:13:11 -0400135 * start a sub transaction and setup the log tree
136 * this increments the log tree writer count to make the people
137 * syncing the tree wait for us to finish
138 */
139static int start_log_trans(struct btrfs_trans_handle *trans,
Miao Xie8b050d32014-02-20 18:08:58 +0800140 struct btrfs_root *root,
141 struct btrfs_log_ctx *ctx)
Chris Masone02119d2008-09-05 16:13:11 -0400142{
Zhaolei34eb2a52015-08-17 18:44:45 +0800143 int ret = 0;
Yan Zheng7237f1832009-01-21 12:54:03 -0500144
145 mutex_lock(&root->log_mutex);
Zhaolei34eb2a52015-08-17 18:44:45 +0800146
Yan Zheng7237f1832009-01-21 12:54:03 -0500147 if (root->log_root) {
Miao Xie995946d2014-04-02 19:51:06 +0800148 if (btrfs_need_log_full_commit(root->fs_info, trans)) {
Miao Xie50471a32014-02-20 18:08:57 +0800149 ret = -EAGAIN;
150 goto out;
151 }
Zhaolei34eb2a52015-08-17 18:44:45 +0800152
Josef Bacikff782e02009-10-08 15:30:04 -0400153 if (!root->log_start_pid) {
Miao Xie27cdeb72014-04-02 19:51:05 +0800154 clear_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state);
Zhaolei34eb2a52015-08-17 18:44:45 +0800155 root->log_start_pid = current->pid;
Josef Bacikff782e02009-10-08 15:30:04 -0400156 } else if (root->log_start_pid != current->pid) {
Miao Xie27cdeb72014-04-02 19:51:05 +0800157 set_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state);
Josef Bacikff782e02009-10-08 15:30:04 -0400158 }
Zhaolei34eb2a52015-08-17 18:44:45 +0800159 } else {
160 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);
163 mutex_unlock(&root->fs_info->tree_log_mutex);
164 if (ret)
165 goto out;
Josef Bacikff782e02009-10-08 15:30:04 -0400166
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)
Miao Xiee87ac132014-02-20 18:08:53 +0800169 goto out;
Zhaolei34eb2a52015-08-17 18:44:45 +0800170
171 clear_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state);
172 root->log_start_pid = current->pid;
Chris Masone02119d2008-09-05 16:13:11 -0400173 }
Zhaolei34eb2a52015-08-17 18:44:45 +0800174
Miao Xie2ecb7922012-09-06 04:04:27 -0600175 atomic_inc(&root->log_batch);
Yan Zheng7237f1832009-01-21 12:54:03 -0500176 atomic_inc(&root->log_writers);
Miao Xie8b050d32014-02-20 18:08:58 +0800177 if (ctx) {
Zhaolei34eb2a52015-08-17 18:44:45 +0800178 int index = root->log_transid % 2;
Miao Xie8b050d32014-02-20 18:08:58 +0800179 list_add_tail(&ctx->list, &root->log_ctxs[index]);
Miao Xied1433de2014-02-20 18:08:59 +0800180 ctx->log_transid = root->log_transid;
Miao Xie8b050d32014-02-20 18:08:58 +0800181 }
Zhaolei34eb2a52015-08-17 18:44:45 +0800182
Miao Xiee87ac132014-02-20 18:08:53 +0800183out:
Yan Zheng7237f1832009-01-21 12:54:03 -0500184 mutex_unlock(&root->log_mutex);
Miao Xiee87ac132014-02-20 18:08:53 +0800185 return ret;
Chris Masone02119d2008-09-05 16:13:11 -0400186}
187
188/*
189 * returns 0 if there was a log transaction running and we were able
190 * to join, or returns -ENOENT if there were not transactions
191 * in progress
192 */
193static int join_running_log_trans(struct btrfs_root *root)
194{
195 int ret = -ENOENT;
196
197 smp_mb();
198 if (!root->log_root)
199 return -ENOENT;
200
Yan Zheng7237f1832009-01-21 12:54:03 -0500201 mutex_lock(&root->log_mutex);
Chris Masone02119d2008-09-05 16:13:11 -0400202 if (root->log_root) {
203 ret = 0;
Yan Zheng7237f1832009-01-21 12:54:03 -0500204 atomic_inc(&root->log_writers);
Chris Masone02119d2008-09-05 16:13:11 -0400205 }
Yan Zheng7237f1832009-01-21 12:54:03 -0500206 mutex_unlock(&root->log_mutex);
Chris Masone02119d2008-09-05 16:13:11 -0400207 return ret;
208}
209
210/*
Chris Mason12fcfd22009-03-24 10:24:20 -0400211 * This either makes the current running log transaction wait
212 * until you call btrfs_end_log_trans() or it makes any future
213 * log transactions wait until you call btrfs_end_log_trans()
214 */
215int btrfs_pin_log_trans(struct btrfs_root *root)
216{
217 int ret = -ENOENT;
218
219 mutex_lock(&root->log_mutex);
220 atomic_inc(&root->log_writers);
221 mutex_unlock(&root->log_mutex);
222 return ret;
223}
224
225/*
Chris Masone02119d2008-09-05 16:13:11 -0400226 * indicate we're done making changes to the log tree
227 * and wake up anyone waiting to do a sync
228 */
Jeff Mahoney143bede2012-03-01 14:56:26 +0100229void btrfs_end_log_trans(struct btrfs_root *root)
Chris Masone02119d2008-09-05 16:13:11 -0400230{
Yan Zheng7237f1832009-01-21 12:54:03 -0500231 if (atomic_dec_and_test(&root->log_writers)) {
David Sterba779adf02015-02-16 19:39:00 +0100232 /*
233 * Implicit memory barrier after atomic_dec_and_test
234 */
Yan Zheng7237f1832009-01-21 12:54:03 -0500235 if (waitqueue_active(&root->log_writer_wait))
236 wake_up(&root->log_writer_wait);
237 }
Chris Masone02119d2008-09-05 16:13:11 -0400238}
239
240
241/*
242 * the walk control struct is used to pass state down the chain when
243 * processing the log tree. The stage field tells us which part
244 * of the log tree processing we are currently doing. The others
245 * are state fields used for that specific part
246 */
247struct walk_control {
248 /* should we free the extent on disk when done? This is used
249 * at transaction commit time while freeing a log tree
250 */
251 int free;
252
253 /* should we write out the extent buffer? This is used
254 * while flushing the log tree to disk during a sync
255 */
256 int write;
257
258 /* should we wait for the extent buffer io to finish? Also used
259 * while flushing the log tree to disk for a sync
260 */
261 int wait;
262
263 /* pin only walk, we record which extents on disk belong to the
264 * log trees
265 */
266 int pin;
267
268 /* what stage of the replay code we're currently in */
269 int stage;
270
271 /* the root we are currently replaying */
272 struct btrfs_root *replay_dest;
273
274 /* the trans handle for the current replay */
275 struct btrfs_trans_handle *trans;
276
277 /* the function that gets used to process blocks we find in the
278 * tree. Note the extent_buffer might not be up to date when it is
279 * passed in, and it must be checked or read if you need the data
280 * inside it
281 */
282 int (*process_func)(struct btrfs_root *log, struct extent_buffer *eb,
283 struct walk_control *wc, u64 gen);
284};
285
286/*
287 * process_func used to pin down extents, write them or wait on them
288 */
289static int process_one_buffer(struct btrfs_root *log,
290 struct extent_buffer *eb,
291 struct walk_control *wc, u64 gen)
292{
Josef Bacikb50c6e22013-04-25 15:55:30 -0400293 int ret = 0;
Chris Masone02119d2008-09-05 16:13:11 -0400294
Josef Bacik8c2a1a32013-06-06 13:19:32 -0400295 /*
296 * If this fs is mixed then we need to be able to process the leaves to
297 * pin down any logged extents, so we have to read the block.
298 */
299 if (btrfs_fs_incompat(log->fs_info, MIXED_GROUPS)) {
300 ret = btrfs_read_buffer(eb, gen);
301 if (ret)
302 return ret;
303 }
304
Josef Bacikb50c6e22013-04-25 15:55:30 -0400305 if (wc->pin)
306 ret = btrfs_pin_extent_for_log_replay(log->fs_info->extent_root,
307 eb->start, eb->len);
308
309 if (!ret && btrfs_buffer_uptodate(eb, gen, 0)) {
Josef Bacik8c2a1a32013-06-06 13:19:32 -0400310 if (wc->pin && btrfs_header_level(eb) == 0)
311 ret = btrfs_exclude_logged_extents(log, eb);
Chris Masone02119d2008-09-05 16:13:11 -0400312 if (wc->write)
313 btrfs_write_tree_block(eb);
314 if (wc->wait)
315 btrfs_wait_tree_block_writeback(eb);
316 }
Josef Bacikb50c6e22013-04-25 15:55:30 -0400317 return ret;
Chris Masone02119d2008-09-05 16:13:11 -0400318}
319
320/*
321 * Item overwrite used by replay and tree logging. eb, slot and key all refer
322 * to the src data we are copying out.
323 *
324 * root is the tree we are copying into, and path is a scratch
325 * path for use in this function (it should be released on entry and
326 * will be released on exit).
327 *
328 * If the key is already in the destination tree the existing item is
329 * overwritten. If the existing item isn't big enough, it is extended.
330 * If it is too large, it is truncated.
331 *
332 * If the key isn't in the destination yet, a new item is inserted.
333 */
334static noinline int overwrite_item(struct btrfs_trans_handle *trans,
335 struct btrfs_root *root,
336 struct btrfs_path *path,
337 struct extent_buffer *eb, int slot,
338 struct btrfs_key *key)
339{
340 int ret;
341 u32 item_size;
342 u64 saved_i_size = 0;
343 int save_old_i_size = 0;
344 unsigned long src_ptr;
345 unsigned long dst_ptr;
346 int overwrite_root = 0;
Josef Bacik4bc4bee2013-04-05 20:50:09 +0000347 bool inode_item = key->type == BTRFS_INODE_ITEM_KEY;
Chris Masone02119d2008-09-05 16:13:11 -0400348
349 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
350 overwrite_root = 1;
351
352 item_size = btrfs_item_size_nr(eb, slot);
353 src_ptr = btrfs_item_ptr_offset(eb, slot);
354
355 /* look for the key in the destination tree */
356 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
Josef Bacik4bc4bee2013-04-05 20:50:09 +0000357 if (ret < 0)
358 return ret;
359
Chris Masone02119d2008-09-05 16:13:11 -0400360 if (ret == 0) {
361 char *src_copy;
362 char *dst_copy;
363 u32 dst_size = btrfs_item_size_nr(path->nodes[0],
364 path->slots[0]);
365 if (dst_size != item_size)
366 goto insert;
367
368 if (item_size == 0) {
David Sterbab3b4aa72011-04-21 01:20:15 +0200369 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -0400370 return 0;
371 }
372 dst_copy = kmalloc(item_size, GFP_NOFS);
373 src_copy = kmalloc(item_size, GFP_NOFS);
liubo2a29edc2011-01-26 06:22:08 +0000374 if (!dst_copy || !src_copy) {
David Sterbab3b4aa72011-04-21 01:20:15 +0200375 btrfs_release_path(path);
liubo2a29edc2011-01-26 06:22:08 +0000376 kfree(dst_copy);
377 kfree(src_copy);
378 return -ENOMEM;
379 }
Chris Masone02119d2008-09-05 16:13:11 -0400380
381 read_extent_buffer(eb, src_copy, src_ptr, item_size);
382
383 dst_ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
384 read_extent_buffer(path->nodes[0], dst_copy, dst_ptr,
385 item_size);
386 ret = memcmp(dst_copy, src_copy, item_size);
387
388 kfree(dst_copy);
389 kfree(src_copy);
390 /*
391 * they have the same contents, just return, this saves
392 * us from cowing blocks in the destination tree and doing
393 * extra writes that may not have been done by a previous
394 * sync
395 */
396 if (ret == 0) {
David Sterbab3b4aa72011-04-21 01:20:15 +0200397 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -0400398 return 0;
399 }
400
Josef Bacik4bc4bee2013-04-05 20:50:09 +0000401 /*
402 * We need to load the old nbytes into the inode so when we
403 * replay the extents we've logged we get the right nbytes.
404 */
405 if (inode_item) {
406 struct btrfs_inode_item *item;
407 u64 nbytes;
Josef Bacikd5554382013-09-11 14:17:00 -0400408 u32 mode;
Josef Bacik4bc4bee2013-04-05 20:50:09 +0000409
410 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
411 struct btrfs_inode_item);
412 nbytes = btrfs_inode_nbytes(path->nodes[0], item);
413 item = btrfs_item_ptr(eb, slot,
414 struct btrfs_inode_item);
415 btrfs_set_inode_nbytes(eb, item, nbytes);
Josef Bacikd5554382013-09-11 14:17:00 -0400416
417 /*
418 * If this is a directory we need to reset the i_size to
419 * 0 so that we can set it up properly when replaying
420 * the rest of the items in this log.
421 */
422 mode = btrfs_inode_mode(eb, item);
423 if (S_ISDIR(mode))
424 btrfs_set_inode_size(eb, item, 0);
Josef Bacik4bc4bee2013-04-05 20:50:09 +0000425 }
426 } else if (inode_item) {
427 struct btrfs_inode_item *item;
Josef Bacikd5554382013-09-11 14:17:00 -0400428 u32 mode;
Josef Bacik4bc4bee2013-04-05 20:50:09 +0000429
430 /*
431 * New inode, set nbytes to 0 so that the nbytes comes out
432 * properly when we replay the extents.
433 */
434 item = btrfs_item_ptr(eb, slot, struct btrfs_inode_item);
435 btrfs_set_inode_nbytes(eb, item, 0);
Josef Bacikd5554382013-09-11 14:17:00 -0400436
437 /*
438 * If this is a directory we need to reset the i_size to 0 so
439 * that we can set it up properly when replaying the rest of
440 * the items in this log.
441 */
442 mode = btrfs_inode_mode(eb, item);
443 if (S_ISDIR(mode))
444 btrfs_set_inode_size(eb, item, 0);
Chris Masone02119d2008-09-05 16:13:11 -0400445 }
446insert:
David Sterbab3b4aa72011-04-21 01:20:15 +0200447 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -0400448 /* try to insert the key into the destination tree */
Filipe Mananadf8d1162015-01-14 01:52:25 +0000449 path->skip_release_on_error = 1;
Chris Masone02119d2008-09-05 16:13:11 -0400450 ret = btrfs_insert_empty_item(trans, root, path,
451 key, item_size);
Filipe Mananadf8d1162015-01-14 01:52:25 +0000452 path->skip_release_on_error = 0;
Chris Masone02119d2008-09-05 16:13:11 -0400453
454 /* make sure any existing item is the correct size */
Filipe Mananadf8d1162015-01-14 01:52:25 +0000455 if (ret == -EEXIST || ret == -EOVERFLOW) {
Chris Masone02119d2008-09-05 16:13:11 -0400456 u32 found_size;
457 found_size = btrfs_item_size_nr(path->nodes[0],
458 path->slots[0]);
Jeff Mahoney143bede2012-03-01 14:56:26 +0100459 if (found_size > item_size)
Tsutomu Itohafe5fea2013-04-16 05:18:22 +0000460 btrfs_truncate_item(root, path, item_size, 1);
Jeff Mahoney143bede2012-03-01 14:56:26 +0100461 else if (found_size < item_size)
Tsutomu Itoh4b90c682013-04-16 05:18:49 +0000462 btrfs_extend_item(root, path,
Jeff Mahoney143bede2012-03-01 14:56:26 +0100463 item_size - found_size);
Chris Masone02119d2008-09-05 16:13:11 -0400464 } else if (ret) {
Yan, Zheng4a500fd2010-05-16 10:49:59 -0400465 return ret;
Chris Masone02119d2008-09-05 16:13:11 -0400466 }
467 dst_ptr = btrfs_item_ptr_offset(path->nodes[0],
468 path->slots[0]);
469
470 /* don't overwrite an existing inode if the generation number
471 * was logged as zero. This is done when the tree logging code
472 * is just logging an inode to make sure it exists after recovery.
473 *
474 * Also, don't overwrite i_size on directories during replay.
475 * log replay inserts and removes directory items based on the
476 * state of the tree found in the subvolume, and i_size is modified
477 * as it goes
478 */
479 if (key->type == BTRFS_INODE_ITEM_KEY && ret == -EEXIST) {
480 struct btrfs_inode_item *src_item;
481 struct btrfs_inode_item *dst_item;
482
483 src_item = (struct btrfs_inode_item *)src_ptr;
484 dst_item = (struct btrfs_inode_item *)dst_ptr;
485
Filipe Manana1a4bcf42015-02-13 12:30:56 +0000486 if (btrfs_inode_generation(eb, src_item) == 0) {
487 struct extent_buffer *dst_eb = path->nodes[0];
Filipe Manana2f2ff0e2015-03-20 17:19:46 +0000488 const u64 ino_size = btrfs_inode_size(eb, src_item);
Filipe Manana1a4bcf42015-02-13 12:30:56 +0000489
Filipe Manana2f2ff0e2015-03-20 17:19:46 +0000490 /*
491 * For regular files an ino_size == 0 is used only when
492 * logging that an inode exists, as part of a directory
493 * fsync, and the inode wasn't fsynced before. In this
494 * case don't set the size of the inode in the fs/subvol
495 * tree, otherwise we would be throwing valid data away.
496 */
Filipe Manana1a4bcf42015-02-13 12:30:56 +0000497 if (S_ISREG(btrfs_inode_mode(eb, src_item)) &&
Filipe Manana2f2ff0e2015-03-20 17:19:46 +0000498 S_ISREG(btrfs_inode_mode(dst_eb, dst_item)) &&
499 ino_size != 0) {
Filipe Manana1a4bcf42015-02-13 12:30:56 +0000500 struct btrfs_map_token token;
Filipe Manana1a4bcf42015-02-13 12:30:56 +0000501
502 btrfs_init_map_token(&token);
503 btrfs_set_token_inode_size(dst_eb, dst_item,
504 ino_size, &token);
505 }
Chris Masone02119d2008-09-05 16:13:11 -0400506 goto no_copy;
Filipe Manana1a4bcf42015-02-13 12:30:56 +0000507 }
Chris Masone02119d2008-09-05 16:13:11 -0400508
509 if (overwrite_root &&
510 S_ISDIR(btrfs_inode_mode(eb, src_item)) &&
511 S_ISDIR(btrfs_inode_mode(path->nodes[0], dst_item))) {
512 save_old_i_size = 1;
513 saved_i_size = btrfs_inode_size(path->nodes[0],
514 dst_item);
515 }
516 }
517
518 copy_extent_buffer(path->nodes[0], eb, dst_ptr,
519 src_ptr, item_size);
520
521 if (save_old_i_size) {
522 struct btrfs_inode_item *dst_item;
523 dst_item = (struct btrfs_inode_item *)dst_ptr;
524 btrfs_set_inode_size(path->nodes[0], dst_item, saved_i_size);
525 }
526
527 /* make sure the generation is filled in */
528 if (key->type == BTRFS_INODE_ITEM_KEY) {
529 struct btrfs_inode_item *dst_item;
530 dst_item = (struct btrfs_inode_item *)dst_ptr;
531 if (btrfs_inode_generation(path->nodes[0], dst_item) == 0) {
532 btrfs_set_inode_generation(path->nodes[0], dst_item,
533 trans->transid);
534 }
535 }
536no_copy:
537 btrfs_mark_buffer_dirty(path->nodes[0]);
David Sterbab3b4aa72011-04-21 01:20:15 +0200538 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -0400539 return 0;
540}
541
542/*
543 * simple helper to read an inode off the disk from a given root
544 * This can only be called for subvolume roots and not for the log
545 */
546static noinline struct inode *read_one_inode(struct btrfs_root *root,
547 u64 objectid)
548{
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400549 struct btrfs_key key;
Chris Masone02119d2008-09-05 16:13:11 -0400550 struct inode *inode;
Chris Masone02119d2008-09-05 16:13:11 -0400551
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400552 key.objectid = objectid;
553 key.type = BTRFS_INODE_ITEM_KEY;
554 key.offset = 0;
Josef Bacik73f73412009-12-04 17:38:27 +0000555 inode = btrfs_iget(root->fs_info->sb, &key, root, NULL);
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400556 if (IS_ERR(inode)) {
557 inode = NULL;
558 } else if (is_bad_inode(inode)) {
Chris Masone02119d2008-09-05 16:13:11 -0400559 iput(inode);
560 inode = NULL;
561 }
562 return inode;
563}
564
565/* replays a single extent in 'eb' at 'slot' with 'key' into the
566 * subvolume 'root'. path is released on entry and should be released
567 * on exit.
568 *
569 * extents in the log tree have not been allocated out of the extent
570 * tree yet. So, this completes the allocation, taking a reference
571 * as required if the extent already exists or creating a new extent
572 * if it isn't in the extent allocation tree yet.
573 *
574 * The extent is inserted into the file, dropping any existing extents
575 * from the file that overlap the new one.
576 */
577static noinline int replay_one_extent(struct btrfs_trans_handle *trans,
578 struct btrfs_root *root,
579 struct btrfs_path *path,
580 struct extent_buffer *eb, int slot,
581 struct btrfs_key *key)
582{
583 int found_type;
Chris Masone02119d2008-09-05 16:13:11 -0400584 u64 extent_end;
Chris Masone02119d2008-09-05 16:13:11 -0400585 u64 start = key->offset;
Josef Bacik4bc4bee2013-04-05 20:50:09 +0000586 u64 nbytes = 0;
Chris Masone02119d2008-09-05 16:13:11 -0400587 struct btrfs_file_extent_item *item;
588 struct inode *inode = NULL;
589 unsigned long size;
590 int ret = 0;
591
592 item = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
593 found_type = btrfs_file_extent_type(eb, item);
594
Yan Zhengd899e052008-10-30 14:25:28 -0400595 if (found_type == BTRFS_FILE_EXTENT_REG ||
Josef Bacik4bc4bee2013-04-05 20:50:09 +0000596 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
597 nbytes = btrfs_file_extent_num_bytes(eb, item);
598 extent_end = start + nbytes;
599
600 /*
601 * We don't add to the inodes nbytes if we are prealloc or a
602 * hole.
603 */
604 if (btrfs_file_extent_disk_bytenr(eb, item) == 0)
605 nbytes = 0;
606 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
Chris Mason514ac8a2014-01-03 21:07:00 -0800607 size = btrfs_file_extent_inline_len(eb, slot, item);
Josef Bacik4bc4bee2013-04-05 20:50:09 +0000608 nbytes = btrfs_file_extent_ram_bytes(eb, item);
Qu Wenruofda28322013-02-26 08:10:22 +0000609 extent_end = ALIGN(start + size, root->sectorsize);
Chris Masone02119d2008-09-05 16:13:11 -0400610 } else {
611 ret = 0;
612 goto out;
613 }
614
615 inode = read_one_inode(root, key->objectid);
616 if (!inode) {
617 ret = -EIO;
618 goto out;
619 }
620
621 /*
622 * first check to see if we already have this extent in the
623 * file. This must be done before the btrfs_drop_extents run
624 * so we don't try to drop this extent.
625 */
Li Zefan33345d012011-04-20 10:31:50 +0800626 ret = btrfs_lookup_file_extent(trans, root, path, btrfs_ino(inode),
Chris Masone02119d2008-09-05 16:13:11 -0400627 start, 0);
628
Yan Zhengd899e052008-10-30 14:25:28 -0400629 if (ret == 0 &&
630 (found_type == BTRFS_FILE_EXTENT_REG ||
631 found_type == BTRFS_FILE_EXTENT_PREALLOC)) {
Chris Masone02119d2008-09-05 16:13:11 -0400632 struct btrfs_file_extent_item cmp1;
633 struct btrfs_file_extent_item cmp2;
634 struct btrfs_file_extent_item *existing;
635 struct extent_buffer *leaf;
636
637 leaf = path->nodes[0];
638 existing = btrfs_item_ptr(leaf, path->slots[0],
639 struct btrfs_file_extent_item);
640
641 read_extent_buffer(eb, &cmp1, (unsigned long)item,
642 sizeof(cmp1));
643 read_extent_buffer(leaf, &cmp2, (unsigned long)existing,
644 sizeof(cmp2));
645
646 /*
647 * we already have a pointer to this exact extent,
648 * we don't have to do anything
649 */
650 if (memcmp(&cmp1, &cmp2, sizeof(cmp1)) == 0) {
David Sterbab3b4aa72011-04-21 01:20:15 +0200651 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -0400652 goto out;
653 }
654 }
David Sterbab3b4aa72011-04-21 01:20:15 +0200655 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -0400656
657 /* drop any overlapping extents */
Josef Bacik26714852012-08-29 12:24:27 -0400658 ret = btrfs_drop_extents(trans, root, inode, start, extent_end, 1);
Josef Bacik36508602013-04-25 16:23:32 -0400659 if (ret)
660 goto out;
Chris Masone02119d2008-09-05 16:13:11 -0400661
Yan Zheng07d400a2009-01-06 11:42:00 -0500662 if (found_type == BTRFS_FILE_EXTENT_REG ||
663 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400664 u64 offset;
Yan Zheng07d400a2009-01-06 11:42:00 -0500665 unsigned long dest_offset;
666 struct btrfs_key ins;
Chris Masone02119d2008-09-05 16:13:11 -0400667
Yan Zheng07d400a2009-01-06 11:42:00 -0500668 ret = btrfs_insert_empty_item(trans, root, path, key,
669 sizeof(*item));
Josef Bacik36508602013-04-25 16:23:32 -0400670 if (ret)
671 goto out;
Yan Zheng07d400a2009-01-06 11:42:00 -0500672 dest_offset = btrfs_item_ptr_offset(path->nodes[0],
673 path->slots[0]);
674 copy_extent_buffer(path->nodes[0], eb, dest_offset,
675 (unsigned long)item, sizeof(*item));
676
677 ins.objectid = btrfs_file_extent_disk_bytenr(eb, item);
678 ins.offset = btrfs_file_extent_disk_num_bytes(eb, item);
679 ins.type = BTRFS_EXTENT_ITEM_KEY;
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400680 offset = key->offset - btrfs_file_extent_offset(eb, item);
Yan Zheng07d400a2009-01-06 11:42:00 -0500681
682 if (ins.objectid > 0) {
683 u64 csum_start;
684 u64 csum_end;
685 LIST_HEAD(ordered_sums);
686 /*
687 * is this extent already allocated in the extent
688 * allocation tree? If so, just add a reference
689 */
Filipe Manana1a4ed8f2014-10-27 10:44:24 +0000690 ret = btrfs_lookup_data_extent(root, ins.objectid,
Yan Zheng07d400a2009-01-06 11:42:00 -0500691 ins.offset);
692 if (ret == 0) {
693 ret = btrfs_inc_extent_ref(trans, root,
694 ins.objectid, ins.offset,
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400695 0, root->root_key.objectid,
Filipe Mananab06c4bf2015-10-23 07:52:54 +0100696 key->objectid, offset);
Josef Bacikb50c6e22013-04-25 15:55:30 -0400697 if (ret)
698 goto out;
Yan Zheng07d400a2009-01-06 11:42:00 -0500699 } else {
700 /*
701 * insert the extent pointer in the extent
702 * allocation tree
703 */
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400704 ret = btrfs_alloc_logged_file_extent(trans,
705 root, root->root_key.objectid,
706 key->objectid, offset, &ins);
Josef Bacikb50c6e22013-04-25 15:55:30 -0400707 if (ret)
708 goto out;
Yan Zheng07d400a2009-01-06 11:42:00 -0500709 }
David Sterbab3b4aa72011-04-21 01:20:15 +0200710 btrfs_release_path(path);
Yan Zheng07d400a2009-01-06 11:42:00 -0500711
712 if (btrfs_file_extent_compression(eb, item)) {
713 csum_start = ins.objectid;
714 csum_end = csum_start + ins.offset;
715 } else {
716 csum_start = ins.objectid +
717 btrfs_file_extent_offset(eb, item);
718 csum_end = csum_start +
719 btrfs_file_extent_num_bytes(eb, item);
720 }
721
722 ret = btrfs_lookup_csums_range(root->log_root,
723 csum_start, csum_end - 1,
Arne Jansena2de7332011-03-08 14:14:00 +0100724 &ordered_sums, 0);
Josef Bacik36508602013-04-25 16:23:32 -0400725 if (ret)
726 goto out;
Filipe Mananab84b8392015-08-19 11:09:40 +0100727 /*
728 * Now delete all existing cums in the csum root that
729 * cover our range. We do this because we can have an
730 * extent that is completely referenced by one file
731 * extent item and partially referenced by another
732 * file extent item (like after using the clone or
733 * extent_same ioctls). In this case if we end up doing
734 * the replay of the one that partially references the
735 * extent first, and we do not do the csum deletion
736 * below, we can get 2 csum items in the csum tree that
737 * overlap each other. For example, imagine our log has
738 * the two following file extent items:
739 *
740 * key (257 EXTENT_DATA 409600)
741 * extent data disk byte 12845056 nr 102400
742 * extent data offset 20480 nr 20480 ram 102400
743 *
744 * key (257 EXTENT_DATA 819200)
745 * extent data disk byte 12845056 nr 102400
746 * extent data offset 0 nr 102400 ram 102400
747 *
748 * Where the second one fully references the 100K extent
749 * that starts at disk byte 12845056, and the log tree
750 * has a single csum item that covers the entire range
751 * of the extent:
752 *
753 * key (EXTENT_CSUM EXTENT_CSUM 12845056) itemsize 100
754 *
755 * After the first file extent item is replayed, the
756 * csum tree gets the following csum item:
757 *
758 * key (EXTENT_CSUM EXTENT_CSUM 12865536) itemsize 20
759 *
760 * Which covers the 20K sub-range starting at offset 20K
761 * of our extent. Now when we replay the second file
762 * extent item, if we do not delete existing csum items
763 * that cover any of its blocks, we end up getting two
764 * csum items in our csum tree that overlap each other:
765 *
766 * key (EXTENT_CSUM EXTENT_CSUM 12845056) itemsize 100
767 * key (EXTENT_CSUM EXTENT_CSUM 12865536) itemsize 20
768 *
769 * Which is a problem, because after this anyone trying
770 * to lookup up for the checksum of any block of our
771 * extent starting at an offset of 40K or higher, will
772 * end up looking at the second csum item only, which
773 * does not contain the checksum for any block starting
774 * at offset 40K or higher of our extent.
775 */
Yan Zheng07d400a2009-01-06 11:42:00 -0500776 while (!list_empty(&ordered_sums)) {
777 struct btrfs_ordered_sum *sums;
778 sums = list_entry(ordered_sums.next,
779 struct btrfs_ordered_sum,
780 list);
Josef Bacik36508602013-04-25 16:23:32 -0400781 if (!ret)
Filipe Mananab84b8392015-08-19 11:09:40 +0100782 ret = btrfs_del_csums(trans,
783 root->fs_info->csum_root,
784 sums->bytenr,
785 sums->len);
786 if (!ret)
Josef Bacik36508602013-04-25 16:23:32 -0400787 ret = btrfs_csum_file_blocks(trans,
Yan Zheng07d400a2009-01-06 11:42:00 -0500788 root->fs_info->csum_root,
789 sums);
Yan Zheng07d400a2009-01-06 11:42:00 -0500790 list_del(&sums->list);
791 kfree(sums);
792 }
Josef Bacik36508602013-04-25 16:23:32 -0400793 if (ret)
794 goto out;
Yan Zheng07d400a2009-01-06 11:42:00 -0500795 } else {
David Sterbab3b4aa72011-04-21 01:20:15 +0200796 btrfs_release_path(path);
Yan Zheng07d400a2009-01-06 11:42:00 -0500797 }
798 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
799 /* inline extents are easy, we just overwrite them */
800 ret = overwrite_item(trans, root, path, eb, slot, key);
Josef Bacik36508602013-04-25 16:23:32 -0400801 if (ret)
802 goto out;
Yan Zheng07d400a2009-01-06 11:42:00 -0500803 }
804
Josef Bacik4bc4bee2013-04-05 20:50:09 +0000805 inode_add_bytes(inode, nbytes);
Tsutomu Itohb9959292012-06-25 21:25:22 -0600806 ret = btrfs_update_inode(trans, root, inode);
Chris Masone02119d2008-09-05 16:13:11 -0400807out:
808 if (inode)
809 iput(inode);
810 return ret;
811}
812
813/*
814 * when cleaning up conflicts between the directory names in the
815 * subvolume, directory names in the log and directory names in the
816 * inode back references, we may have to unlink inodes from directories.
817 *
818 * This is a helper function to do the unlink of a specific directory
819 * item
820 */
821static noinline int drop_one_dir_item(struct btrfs_trans_handle *trans,
822 struct btrfs_root *root,
823 struct btrfs_path *path,
824 struct inode *dir,
825 struct btrfs_dir_item *di)
826{
827 struct inode *inode;
828 char *name;
829 int name_len;
830 struct extent_buffer *leaf;
831 struct btrfs_key location;
832 int ret;
833
834 leaf = path->nodes[0];
835
836 btrfs_dir_item_key_to_cpu(leaf, di, &location);
837 name_len = btrfs_dir_name_len(leaf, di);
838 name = kmalloc(name_len, GFP_NOFS);
liubo2a29edc2011-01-26 06:22:08 +0000839 if (!name)
840 return -ENOMEM;
841
Chris Masone02119d2008-09-05 16:13:11 -0400842 read_extent_buffer(leaf, name, (unsigned long)(di + 1), name_len);
David Sterbab3b4aa72011-04-21 01:20:15 +0200843 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -0400844
845 inode = read_one_inode(root, location.objectid);
Tsutomu Itohc00e9492011-04-28 09:10:23 +0000846 if (!inode) {
Josef Bacik36508602013-04-25 16:23:32 -0400847 ret = -EIO;
848 goto out;
Tsutomu Itohc00e9492011-04-28 09:10:23 +0000849 }
Chris Masone02119d2008-09-05 16:13:11 -0400850
Yan Zhengec051c02009-01-05 15:43:42 -0500851 ret = link_to_fixup_dir(trans, root, path, location.objectid);
Josef Bacik36508602013-04-25 16:23:32 -0400852 if (ret)
853 goto out;
Chris Mason12fcfd22009-03-24 10:24:20 -0400854
Chris Masone02119d2008-09-05 16:13:11 -0400855 ret = btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
Josef Bacik36508602013-04-25 16:23:32 -0400856 if (ret)
857 goto out;
Filipe David Borba Mananaada9af22013-08-05 09:25:47 +0100858 else
859 ret = btrfs_run_delayed_items(trans, root);
Josef Bacik36508602013-04-25 16:23:32 -0400860out:
861 kfree(name);
862 iput(inode);
Chris Masone02119d2008-09-05 16:13:11 -0400863 return ret;
864}
865
866/*
867 * helper function to see if a given name and sequence number found
868 * in an inode back reference are already in a directory and correctly
869 * point to this inode
870 */
871static noinline int inode_in_dir(struct btrfs_root *root,
872 struct btrfs_path *path,
873 u64 dirid, u64 objectid, u64 index,
874 const char *name, int name_len)
875{
876 struct btrfs_dir_item *di;
877 struct btrfs_key location;
878 int match = 0;
879
880 di = btrfs_lookup_dir_index_item(NULL, root, path, dirid,
881 index, name, name_len, 0);
882 if (di && !IS_ERR(di)) {
883 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
884 if (location.objectid != objectid)
885 goto out;
886 } else
887 goto out;
David Sterbab3b4aa72011-04-21 01:20:15 +0200888 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -0400889
890 di = btrfs_lookup_dir_item(NULL, root, path, dirid, name, name_len, 0);
891 if (di && !IS_ERR(di)) {
892 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
893 if (location.objectid != objectid)
894 goto out;
895 } else
896 goto out;
897 match = 1;
898out:
David Sterbab3b4aa72011-04-21 01:20:15 +0200899 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -0400900 return match;
901}
902
903/*
904 * helper function to check a log tree for a named back reference in
905 * an inode. This is used to decide if a back reference that is
906 * found in the subvolume conflicts with what we find in the log.
907 *
908 * inode backreferences may have multiple refs in a single item,
909 * during replay we process one reference at a time, and we don't
910 * want to delete valid links to a file from the subvolume if that
911 * link is also in the log.
912 */
913static noinline int backref_in_log(struct btrfs_root *log,
914 struct btrfs_key *key,
Mark Fashehf1863732012-08-08 11:32:27 -0700915 u64 ref_objectid,
Filipe Mananadf8d1162015-01-14 01:52:25 +0000916 const char *name, int namelen)
Chris Masone02119d2008-09-05 16:13:11 -0400917{
918 struct btrfs_path *path;
919 struct btrfs_inode_ref *ref;
920 unsigned long ptr;
921 unsigned long ptr_end;
922 unsigned long name_ptr;
923 int found_name_len;
924 int item_size;
925 int ret;
926 int match = 0;
927
928 path = btrfs_alloc_path();
liubo2a29edc2011-01-26 06:22:08 +0000929 if (!path)
930 return -ENOMEM;
931
Chris Masone02119d2008-09-05 16:13:11 -0400932 ret = btrfs_search_slot(NULL, log, key, path, 0, 0);
933 if (ret != 0)
934 goto out;
935
Chris Masone02119d2008-09-05 16:13:11 -0400936 ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
Mark Fashehf1863732012-08-08 11:32:27 -0700937
938 if (key->type == BTRFS_INODE_EXTREF_KEY) {
939 if (btrfs_find_name_in_ext_backref(path, ref_objectid,
940 name, namelen, NULL))
941 match = 1;
942
943 goto out;
944 }
945
946 item_size = btrfs_item_size_nr(path->nodes[0], path->slots[0]);
Chris Masone02119d2008-09-05 16:13:11 -0400947 ptr_end = ptr + item_size;
948 while (ptr < ptr_end) {
949 ref = (struct btrfs_inode_ref *)ptr;
950 found_name_len = btrfs_inode_ref_name_len(path->nodes[0], ref);
951 if (found_name_len == namelen) {
952 name_ptr = (unsigned long)(ref + 1);
953 ret = memcmp_extent_buffer(path->nodes[0], name,
954 name_ptr, namelen);
955 if (ret == 0) {
956 match = 1;
957 goto out;
958 }
959 }
960 ptr = (unsigned long)(ref + 1) + found_name_len;
961 }
962out:
963 btrfs_free_path(path);
964 return match;
965}
966
Jan Schmidt5a1d7842012-08-17 14:04:41 -0700967static inline int __add_inode_ref(struct btrfs_trans_handle *trans,
968 struct btrfs_root *root,
969 struct btrfs_path *path,
970 struct btrfs_root *log_root,
971 struct inode *dir, struct inode *inode,
Jan Schmidt5a1d7842012-08-17 14:04:41 -0700972 struct extent_buffer *eb,
Mark Fashehf1863732012-08-08 11:32:27 -0700973 u64 inode_objectid, u64 parent_objectid,
974 u64 ref_index, char *name, int namelen,
975 int *search_done)
Jan Schmidt5a1d7842012-08-17 14:04:41 -0700976{
977 int ret;
Mark Fashehf1863732012-08-08 11:32:27 -0700978 char *victim_name;
979 int victim_name_len;
980 struct extent_buffer *leaf;
Jan Schmidt5a1d7842012-08-17 14:04:41 -0700981 struct btrfs_dir_item *di;
Mark Fashehf1863732012-08-08 11:32:27 -0700982 struct btrfs_key search_key;
983 struct btrfs_inode_extref *extref;
Jan Schmidt5a1d7842012-08-17 14:04:41 -0700984
Mark Fashehf1863732012-08-08 11:32:27 -0700985again:
986 /* Search old style refs */
987 search_key.objectid = inode_objectid;
988 search_key.type = BTRFS_INODE_REF_KEY;
989 search_key.offset = parent_objectid;
990 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
Jan Schmidt5a1d7842012-08-17 14:04:41 -0700991 if (ret == 0) {
Jan Schmidt5a1d7842012-08-17 14:04:41 -0700992 struct btrfs_inode_ref *victim_ref;
993 unsigned long ptr;
994 unsigned long ptr_end;
Mark Fashehf1863732012-08-08 11:32:27 -0700995
996 leaf = path->nodes[0];
Jan Schmidt5a1d7842012-08-17 14:04:41 -0700997
998 /* are we trying to overwrite a back ref for the root directory
999 * if so, just jump out, we're done
1000 */
Mark Fashehf1863732012-08-08 11:32:27 -07001001 if (search_key.objectid == search_key.offset)
Jan Schmidt5a1d7842012-08-17 14:04:41 -07001002 return 1;
1003
1004 /* check all the names in this back reference to see
1005 * if they are in the log. if so, we allow them to stay
1006 * otherwise they must be unlinked as a conflict
1007 */
1008 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
1009 ptr_end = ptr + btrfs_item_size_nr(leaf, path->slots[0]);
1010 while (ptr < ptr_end) {
1011 victim_ref = (struct btrfs_inode_ref *)ptr;
1012 victim_name_len = btrfs_inode_ref_name_len(leaf,
1013 victim_ref);
1014 victim_name = kmalloc(victim_name_len, GFP_NOFS);
Josef Bacik36508602013-04-25 16:23:32 -04001015 if (!victim_name)
1016 return -ENOMEM;
Jan Schmidt5a1d7842012-08-17 14:04:41 -07001017
1018 read_extent_buffer(leaf, victim_name,
1019 (unsigned long)(victim_ref + 1),
1020 victim_name_len);
1021
Mark Fashehf1863732012-08-08 11:32:27 -07001022 if (!backref_in_log(log_root, &search_key,
1023 parent_objectid,
1024 victim_name,
Jan Schmidt5a1d7842012-08-17 14:04:41 -07001025 victim_name_len)) {
Zach Brown8b558c52013-10-16 12:10:34 -07001026 inc_nlink(inode);
Jan Schmidt5a1d7842012-08-17 14:04:41 -07001027 btrfs_release_path(path);
1028
1029 ret = btrfs_unlink_inode(trans, root, dir,
1030 inode, victim_name,
1031 victim_name_len);
Mark Fashehf1863732012-08-08 11:32:27 -07001032 kfree(victim_name);
Josef Bacik36508602013-04-25 16:23:32 -04001033 if (ret)
1034 return ret;
Filipe David Borba Mananaada9af22013-08-05 09:25:47 +01001035 ret = btrfs_run_delayed_items(trans, root);
1036 if (ret)
1037 return ret;
Mark Fashehf1863732012-08-08 11:32:27 -07001038 *search_done = 1;
1039 goto again;
Jan Schmidt5a1d7842012-08-17 14:04:41 -07001040 }
1041 kfree(victim_name);
Mark Fashehf1863732012-08-08 11:32:27 -07001042
Jan Schmidt5a1d7842012-08-17 14:04:41 -07001043 ptr = (unsigned long)(victim_ref + 1) + victim_name_len;
1044 }
Jan Schmidt5a1d7842012-08-17 14:04:41 -07001045
1046 /*
1047 * NOTE: we have searched root tree and checked the
1048 * coresponding ref, it does not need to check again.
1049 */
1050 *search_done = 1;
1051 }
1052 btrfs_release_path(path);
1053
Mark Fashehf1863732012-08-08 11:32:27 -07001054 /* Same search but for extended refs */
1055 extref = btrfs_lookup_inode_extref(NULL, root, path, name, namelen,
1056 inode_objectid, parent_objectid, 0,
1057 0);
1058 if (!IS_ERR_OR_NULL(extref)) {
1059 u32 item_size;
1060 u32 cur_offset = 0;
1061 unsigned long base;
1062 struct inode *victim_parent;
1063
1064 leaf = path->nodes[0];
1065
1066 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1067 base = btrfs_item_ptr_offset(leaf, path->slots[0]);
1068
1069 while (cur_offset < item_size) {
Quentin Casasnovasdd9ef132015-03-03 16:31:38 +01001070 extref = (struct btrfs_inode_extref *)(base + cur_offset);
Mark Fashehf1863732012-08-08 11:32:27 -07001071
1072 victim_name_len = btrfs_inode_extref_name_len(leaf, extref);
1073
1074 if (btrfs_inode_extref_parent(leaf, extref) != parent_objectid)
1075 goto next;
1076
1077 victim_name = kmalloc(victim_name_len, GFP_NOFS);
Josef Bacik36508602013-04-25 16:23:32 -04001078 if (!victim_name)
1079 return -ENOMEM;
Mark Fashehf1863732012-08-08 11:32:27 -07001080 read_extent_buffer(leaf, victim_name, (unsigned long)&extref->name,
1081 victim_name_len);
1082
1083 search_key.objectid = inode_objectid;
1084 search_key.type = BTRFS_INODE_EXTREF_KEY;
1085 search_key.offset = btrfs_extref_hash(parent_objectid,
1086 victim_name,
1087 victim_name_len);
1088 ret = 0;
1089 if (!backref_in_log(log_root, &search_key,
1090 parent_objectid, victim_name,
1091 victim_name_len)) {
1092 ret = -ENOENT;
1093 victim_parent = read_one_inode(root,
1094 parent_objectid);
1095 if (victim_parent) {
Zach Brown8b558c52013-10-16 12:10:34 -07001096 inc_nlink(inode);
Mark Fashehf1863732012-08-08 11:32:27 -07001097 btrfs_release_path(path);
1098
1099 ret = btrfs_unlink_inode(trans, root,
1100 victim_parent,
1101 inode,
1102 victim_name,
1103 victim_name_len);
Filipe David Borba Mananaada9af22013-08-05 09:25:47 +01001104 if (!ret)
1105 ret = btrfs_run_delayed_items(
1106 trans, root);
Mark Fashehf1863732012-08-08 11:32:27 -07001107 }
Mark Fashehf1863732012-08-08 11:32:27 -07001108 iput(victim_parent);
1109 kfree(victim_name);
Josef Bacik36508602013-04-25 16:23:32 -04001110 if (ret)
1111 return ret;
Mark Fashehf1863732012-08-08 11:32:27 -07001112 *search_done = 1;
1113 goto again;
1114 }
1115 kfree(victim_name);
Josef Bacik36508602013-04-25 16:23:32 -04001116 if (ret)
1117 return ret;
Mark Fashehf1863732012-08-08 11:32:27 -07001118next:
1119 cur_offset += victim_name_len + sizeof(*extref);
1120 }
1121 *search_done = 1;
1122 }
1123 btrfs_release_path(path);
1124
Jan Schmidt5a1d7842012-08-17 14:04:41 -07001125 /* look for a conflicting sequence number */
1126 di = btrfs_lookup_dir_index_item(trans, root, path, btrfs_ino(dir),
Mark Fashehf1863732012-08-08 11:32:27 -07001127 ref_index, name, namelen, 0);
Jan Schmidt5a1d7842012-08-17 14:04:41 -07001128 if (di && !IS_ERR(di)) {
1129 ret = drop_one_dir_item(trans, root, path, dir, di);
Josef Bacik36508602013-04-25 16:23:32 -04001130 if (ret)
1131 return ret;
Jan Schmidt5a1d7842012-08-17 14:04:41 -07001132 }
1133 btrfs_release_path(path);
1134
1135 /* look for a conflicing name */
1136 di = btrfs_lookup_dir_item(trans, root, path, btrfs_ino(dir),
1137 name, namelen, 0);
1138 if (di && !IS_ERR(di)) {
1139 ret = drop_one_dir_item(trans, root, path, dir, di);
Josef Bacik36508602013-04-25 16:23:32 -04001140 if (ret)
1141 return ret;
Jan Schmidt5a1d7842012-08-17 14:04:41 -07001142 }
1143 btrfs_release_path(path);
1144
1145 return 0;
1146}
Chris Masone02119d2008-09-05 16:13:11 -04001147
Mark Fashehf1863732012-08-08 11:32:27 -07001148static int extref_get_fields(struct extent_buffer *eb, unsigned long ref_ptr,
1149 u32 *namelen, char **name, u64 *index,
1150 u64 *parent_objectid)
1151{
1152 struct btrfs_inode_extref *extref;
1153
1154 extref = (struct btrfs_inode_extref *)ref_ptr;
1155
1156 *namelen = btrfs_inode_extref_name_len(eb, extref);
1157 *name = kmalloc(*namelen, GFP_NOFS);
1158 if (*name == NULL)
1159 return -ENOMEM;
1160
1161 read_extent_buffer(eb, *name, (unsigned long)&extref->name,
1162 *namelen);
1163
1164 *index = btrfs_inode_extref_index(eb, extref);
1165 if (parent_objectid)
1166 *parent_objectid = btrfs_inode_extref_parent(eb, extref);
1167
1168 return 0;
1169}
1170
1171static int ref_get_fields(struct extent_buffer *eb, unsigned long ref_ptr,
1172 u32 *namelen, char **name, u64 *index)
1173{
1174 struct btrfs_inode_ref *ref;
1175
1176 ref = (struct btrfs_inode_ref *)ref_ptr;
1177
1178 *namelen = btrfs_inode_ref_name_len(eb, ref);
1179 *name = kmalloc(*namelen, GFP_NOFS);
1180 if (*name == NULL)
1181 return -ENOMEM;
1182
1183 read_extent_buffer(eb, *name, (unsigned long)(ref + 1), *namelen);
1184
1185 *index = btrfs_inode_ref_index(eb, ref);
1186
1187 return 0;
1188}
1189
Chris Masone02119d2008-09-05 16:13:11 -04001190/*
1191 * replay one inode back reference item found in the log tree.
1192 * eb, slot and key refer to the buffer and key found in the log tree.
1193 * root is the destination we are replaying into, and path is for temp
1194 * use by this function. (it should be released on return).
1195 */
1196static noinline int add_inode_ref(struct btrfs_trans_handle *trans,
1197 struct btrfs_root *root,
1198 struct btrfs_root *log,
1199 struct btrfs_path *path,
1200 struct extent_buffer *eb, int slot,
1201 struct btrfs_key *key)
1202{
Geyslan G. Bem03b2f082013-10-11 15:35:45 -03001203 struct inode *dir = NULL;
1204 struct inode *inode = NULL;
Chris Masone02119d2008-09-05 16:13:11 -04001205 unsigned long ref_ptr;
1206 unsigned long ref_end;
Geyslan G. Bem03b2f082013-10-11 15:35:45 -03001207 char *name = NULL;
liubo34f3e4f2011-08-06 08:35:23 +00001208 int namelen;
1209 int ret;
liuboc622ae62011-03-26 08:01:12 -04001210 int search_done = 0;
Mark Fashehf1863732012-08-08 11:32:27 -07001211 int log_ref_ver = 0;
1212 u64 parent_objectid;
1213 u64 inode_objectid;
Chris Masonf46dbe3de2012-10-09 11:17:20 -04001214 u64 ref_index = 0;
Mark Fashehf1863732012-08-08 11:32:27 -07001215 int ref_struct_size;
1216
1217 ref_ptr = btrfs_item_ptr_offset(eb, slot);
1218 ref_end = ref_ptr + btrfs_item_size_nr(eb, slot);
1219
1220 if (key->type == BTRFS_INODE_EXTREF_KEY) {
1221 struct btrfs_inode_extref *r;
1222
1223 ref_struct_size = sizeof(struct btrfs_inode_extref);
1224 log_ref_ver = 1;
1225 r = (struct btrfs_inode_extref *)ref_ptr;
1226 parent_objectid = btrfs_inode_extref_parent(eb, r);
1227 } else {
1228 ref_struct_size = sizeof(struct btrfs_inode_ref);
1229 parent_objectid = key->offset;
1230 }
1231 inode_objectid = key->objectid;
Chris Masone02119d2008-09-05 16:13:11 -04001232
Chris Masone02119d2008-09-05 16:13:11 -04001233 /*
1234 * it is possible that we didn't log all the parent directories
1235 * for a given inode. If we don't find the dir, just don't
1236 * copy the back ref in. The link count fixup code will take
1237 * care of the rest
1238 */
Mark Fashehf1863732012-08-08 11:32:27 -07001239 dir = read_one_inode(root, parent_objectid);
Geyslan G. Bem03b2f082013-10-11 15:35:45 -03001240 if (!dir) {
1241 ret = -ENOENT;
1242 goto out;
1243 }
Chris Masone02119d2008-09-05 16:13:11 -04001244
Mark Fashehf1863732012-08-08 11:32:27 -07001245 inode = read_one_inode(root, inode_objectid);
Tsutomu Itohc00e9492011-04-28 09:10:23 +00001246 if (!inode) {
Geyslan G. Bem03b2f082013-10-11 15:35:45 -03001247 ret = -EIO;
1248 goto out;
Tsutomu Itohc00e9492011-04-28 09:10:23 +00001249 }
Chris Masone02119d2008-09-05 16:13:11 -04001250
Jan Schmidt5a1d7842012-08-17 14:04:41 -07001251 while (ref_ptr < ref_end) {
Mark Fashehf1863732012-08-08 11:32:27 -07001252 if (log_ref_ver) {
1253 ret = extref_get_fields(eb, ref_ptr, &namelen, &name,
1254 &ref_index, &parent_objectid);
1255 /*
1256 * parent object can change from one array
1257 * item to another.
1258 */
1259 if (!dir)
1260 dir = read_one_inode(root, parent_objectid);
Geyslan G. Bem03b2f082013-10-11 15:35:45 -03001261 if (!dir) {
1262 ret = -ENOENT;
1263 goto out;
1264 }
Mark Fashehf1863732012-08-08 11:32:27 -07001265 } else {
1266 ret = ref_get_fields(eb, ref_ptr, &namelen, &name,
1267 &ref_index);
1268 }
1269 if (ret)
Geyslan G. Bem03b2f082013-10-11 15:35:45 -03001270 goto out;
Chris Masone02119d2008-09-05 16:13:11 -04001271
Jan Schmidt5a1d7842012-08-17 14:04:41 -07001272 /* if we already have a perfect match, we're done */
1273 if (!inode_in_dir(root, path, btrfs_ino(dir), btrfs_ino(inode),
Mark Fashehf1863732012-08-08 11:32:27 -07001274 ref_index, name, namelen)) {
Jan Schmidt5a1d7842012-08-17 14:04:41 -07001275 /*
1276 * look for a conflicting back reference in the
1277 * metadata. if we find one we have to unlink that name
1278 * of the file before we add our new link. Later on, we
1279 * overwrite any existing back reference, and we don't
1280 * want to create dangling pointers in the directory.
1281 */
Chris Masone02119d2008-09-05 16:13:11 -04001282
Jan Schmidt5a1d7842012-08-17 14:04:41 -07001283 if (!search_done) {
1284 ret = __add_inode_ref(trans, root, path, log,
Mark Fashehf1863732012-08-08 11:32:27 -07001285 dir, inode, eb,
1286 inode_objectid,
1287 parent_objectid,
1288 ref_index, name, namelen,
Jan Schmidt5a1d7842012-08-17 14:04:41 -07001289 &search_done);
Geyslan G. Bem03b2f082013-10-11 15:35:45 -03001290 if (ret) {
1291 if (ret == 1)
1292 ret = 0;
Jan Schmidt5a1d7842012-08-17 14:04:41 -07001293 goto out;
Josef Bacik36508602013-04-25 16:23:32 -04001294 }
Chris Masone02119d2008-09-05 16:13:11 -04001295 }
Jan Schmidt5a1d7842012-08-17 14:04:41 -07001296
1297 /* insert our name */
1298 ret = btrfs_add_link(trans, dir, inode, name, namelen,
Mark Fashehf1863732012-08-08 11:32:27 -07001299 0, ref_index);
Josef Bacik36508602013-04-25 16:23:32 -04001300 if (ret)
1301 goto out;
Jan Schmidt5a1d7842012-08-17 14:04:41 -07001302
1303 btrfs_update_inode(trans, root, inode);
Chris Masone02119d2008-09-05 16:13:11 -04001304 }
liuboc622ae62011-03-26 08:01:12 -04001305
Mark Fashehf1863732012-08-08 11:32:27 -07001306 ref_ptr = (unsigned long)(ref_ptr + ref_struct_size) + namelen;
Jan Schmidt5a1d7842012-08-17 14:04:41 -07001307 kfree(name);
Geyslan G. Bem03b2f082013-10-11 15:35:45 -03001308 name = NULL;
Mark Fashehf1863732012-08-08 11:32:27 -07001309 if (log_ref_ver) {
1310 iput(dir);
1311 dir = NULL;
1312 }
Chris Masone02119d2008-09-05 16:13:11 -04001313 }
Chris Masone02119d2008-09-05 16:13:11 -04001314
1315 /* finally write the back reference in the inode */
1316 ret = overwrite_item(trans, root, path, eb, slot, key);
Jan Schmidt5a1d7842012-08-17 14:04:41 -07001317out:
David Sterbab3b4aa72011-04-21 01:20:15 +02001318 btrfs_release_path(path);
Geyslan G. Bem03b2f082013-10-11 15:35:45 -03001319 kfree(name);
Chris Masone02119d2008-09-05 16:13:11 -04001320 iput(dir);
1321 iput(inode);
Josef Bacik36508602013-04-25 16:23:32 -04001322 return ret;
Chris Masone02119d2008-09-05 16:13:11 -04001323}
1324
Yan, Zhengc71bf092009-11-12 09:34:40 +00001325static int insert_orphan_item(struct btrfs_trans_handle *trans,
David Sterba9c4f61f2015-01-02 19:12:57 +01001326 struct btrfs_root *root, u64 ino)
Yan, Zhengc71bf092009-11-12 09:34:40 +00001327{
1328 int ret;
David Sterba381cf652015-01-02 18:45:16 +01001329
David Sterba9c4f61f2015-01-02 19:12:57 +01001330 ret = btrfs_insert_orphan_item(trans, root, ino);
1331 if (ret == -EEXIST)
1332 ret = 0;
David Sterba381cf652015-01-02 18:45:16 +01001333
Yan, Zhengc71bf092009-11-12 09:34:40 +00001334 return ret;
1335}
1336
Mark Fashehf1863732012-08-08 11:32:27 -07001337static int count_inode_extrefs(struct btrfs_root *root,
1338 struct inode *inode, struct btrfs_path *path)
Chris Masone02119d2008-09-05 16:13:11 -04001339{
Mark Fashehf1863732012-08-08 11:32:27 -07001340 int ret = 0;
1341 int name_len;
1342 unsigned int nlink = 0;
1343 u32 item_size;
1344 u32 cur_offset = 0;
1345 u64 inode_objectid = btrfs_ino(inode);
1346 u64 offset = 0;
1347 unsigned long ptr;
1348 struct btrfs_inode_extref *extref;
1349 struct extent_buffer *leaf;
1350
1351 while (1) {
1352 ret = btrfs_find_one_extref(root, inode_objectid, offset, path,
1353 &extref, &offset);
1354 if (ret)
1355 break;
1356
1357 leaf = path->nodes[0];
1358 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1359 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
Filipe Manana2c2c4522015-01-13 16:40:04 +00001360 cur_offset = 0;
Mark Fashehf1863732012-08-08 11:32:27 -07001361
1362 while (cur_offset < item_size) {
1363 extref = (struct btrfs_inode_extref *) (ptr + cur_offset);
1364 name_len = btrfs_inode_extref_name_len(leaf, extref);
1365
1366 nlink++;
1367
1368 cur_offset += name_len + sizeof(*extref);
1369 }
1370
1371 offset++;
1372 btrfs_release_path(path);
1373 }
1374 btrfs_release_path(path);
1375
Filipe Manana2c2c4522015-01-13 16:40:04 +00001376 if (ret < 0 && ret != -ENOENT)
Mark Fashehf1863732012-08-08 11:32:27 -07001377 return ret;
1378 return nlink;
1379}
1380
1381static int count_inode_refs(struct btrfs_root *root,
1382 struct inode *inode, struct btrfs_path *path)
1383{
Chris Masone02119d2008-09-05 16:13:11 -04001384 int ret;
1385 struct btrfs_key key;
Mark Fashehf1863732012-08-08 11:32:27 -07001386 unsigned int nlink = 0;
Chris Masone02119d2008-09-05 16:13:11 -04001387 unsigned long ptr;
1388 unsigned long ptr_end;
1389 int name_len;
Li Zefan33345d012011-04-20 10:31:50 +08001390 u64 ino = btrfs_ino(inode);
Chris Masone02119d2008-09-05 16:13:11 -04001391
Li Zefan33345d012011-04-20 10:31:50 +08001392 key.objectid = ino;
Chris Masone02119d2008-09-05 16:13:11 -04001393 key.type = BTRFS_INODE_REF_KEY;
1394 key.offset = (u64)-1;
1395
Chris Masond3977122009-01-05 21:25:51 -05001396 while (1) {
Chris Masone02119d2008-09-05 16:13:11 -04001397 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1398 if (ret < 0)
1399 break;
1400 if (ret > 0) {
1401 if (path->slots[0] == 0)
1402 break;
1403 path->slots[0]--;
1404 }
Filipe David Borba Mananae93ae262013-10-14 22:49:11 +01001405process_slot:
Chris Masone02119d2008-09-05 16:13:11 -04001406 btrfs_item_key_to_cpu(path->nodes[0], &key,
1407 path->slots[0]);
Li Zefan33345d012011-04-20 10:31:50 +08001408 if (key.objectid != ino ||
Chris Masone02119d2008-09-05 16:13:11 -04001409 key.type != BTRFS_INODE_REF_KEY)
1410 break;
1411 ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
1412 ptr_end = ptr + btrfs_item_size_nr(path->nodes[0],
1413 path->slots[0]);
Chris Masond3977122009-01-05 21:25:51 -05001414 while (ptr < ptr_end) {
Chris Masone02119d2008-09-05 16:13:11 -04001415 struct btrfs_inode_ref *ref;
1416
1417 ref = (struct btrfs_inode_ref *)ptr;
1418 name_len = btrfs_inode_ref_name_len(path->nodes[0],
1419 ref);
1420 ptr = (unsigned long)(ref + 1) + name_len;
1421 nlink++;
1422 }
1423
1424 if (key.offset == 0)
1425 break;
Filipe David Borba Mananae93ae262013-10-14 22:49:11 +01001426 if (path->slots[0] > 0) {
1427 path->slots[0]--;
1428 goto process_slot;
1429 }
Chris Masone02119d2008-09-05 16:13:11 -04001430 key.offset--;
David Sterbab3b4aa72011-04-21 01:20:15 +02001431 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04001432 }
David Sterbab3b4aa72011-04-21 01:20:15 +02001433 btrfs_release_path(path);
Mark Fashehf1863732012-08-08 11:32:27 -07001434
1435 return nlink;
1436}
1437
1438/*
1439 * There are a few corners where the link count of the file can't
1440 * be properly maintained during replay. So, instead of adding
1441 * lots of complexity to the log code, we just scan the backrefs
1442 * for any file that has been through replay.
1443 *
1444 * The scan will update the link count on the inode to reflect the
1445 * number of back refs found. If it goes down to zero, the iput
1446 * will free the inode.
1447 */
1448static noinline int fixup_inode_link_count(struct btrfs_trans_handle *trans,
1449 struct btrfs_root *root,
1450 struct inode *inode)
1451{
1452 struct btrfs_path *path;
1453 int ret;
1454 u64 nlink = 0;
1455 u64 ino = btrfs_ino(inode);
1456
1457 path = btrfs_alloc_path();
1458 if (!path)
1459 return -ENOMEM;
1460
1461 ret = count_inode_refs(root, inode, path);
1462 if (ret < 0)
1463 goto out;
1464
1465 nlink = ret;
1466
1467 ret = count_inode_extrefs(root, inode, path);
Mark Fashehf1863732012-08-08 11:32:27 -07001468 if (ret < 0)
1469 goto out;
1470
1471 nlink += ret;
1472
1473 ret = 0;
1474
Chris Masone02119d2008-09-05 16:13:11 -04001475 if (nlink != inode->i_nlink) {
Miklos Szeredibfe86842011-10-28 14:13:29 +02001476 set_nlink(inode, nlink);
Chris Masone02119d2008-09-05 16:13:11 -04001477 btrfs_update_inode(trans, root, inode);
1478 }
Chris Mason8d5bf1c2008-09-11 15:51:21 -04001479 BTRFS_I(inode)->index_cnt = (u64)-1;
Chris Masone02119d2008-09-05 16:13:11 -04001480
Yan, Zhengc71bf092009-11-12 09:34:40 +00001481 if (inode->i_nlink == 0) {
1482 if (S_ISDIR(inode->i_mode)) {
1483 ret = replay_dir_deletes(trans, root, NULL, path,
Li Zefan33345d012011-04-20 10:31:50 +08001484 ino, 1);
Josef Bacik36508602013-04-25 16:23:32 -04001485 if (ret)
1486 goto out;
Yan, Zhengc71bf092009-11-12 09:34:40 +00001487 }
Li Zefan33345d012011-04-20 10:31:50 +08001488 ret = insert_orphan_item(trans, root, ino);
Chris Mason12fcfd22009-03-24 10:24:20 -04001489 }
Chris Mason12fcfd22009-03-24 10:24:20 -04001490
Mark Fashehf1863732012-08-08 11:32:27 -07001491out:
1492 btrfs_free_path(path);
1493 return ret;
Chris Masone02119d2008-09-05 16:13:11 -04001494}
1495
1496static noinline int fixup_inode_link_counts(struct btrfs_trans_handle *trans,
1497 struct btrfs_root *root,
1498 struct btrfs_path *path)
1499{
1500 int ret;
1501 struct btrfs_key key;
1502 struct inode *inode;
1503
1504 key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID;
1505 key.type = BTRFS_ORPHAN_ITEM_KEY;
1506 key.offset = (u64)-1;
Chris Masond3977122009-01-05 21:25:51 -05001507 while (1) {
Chris Masone02119d2008-09-05 16:13:11 -04001508 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1509 if (ret < 0)
1510 break;
1511
1512 if (ret == 1) {
1513 if (path->slots[0] == 0)
1514 break;
1515 path->slots[0]--;
1516 }
1517
1518 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1519 if (key.objectid != BTRFS_TREE_LOG_FIXUP_OBJECTID ||
1520 key.type != BTRFS_ORPHAN_ITEM_KEY)
1521 break;
1522
1523 ret = btrfs_del_item(trans, root, path);
Tsutomu Itoh65a246c2011-05-19 04:37:44 +00001524 if (ret)
1525 goto out;
Chris Masone02119d2008-09-05 16:13:11 -04001526
David Sterbab3b4aa72011-04-21 01:20:15 +02001527 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04001528 inode = read_one_inode(root, key.offset);
Tsutomu Itohc00e9492011-04-28 09:10:23 +00001529 if (!inode)
1530 return -EIO;
Chris Masone02119d2008-09-05 16:13:11 -04001531
1532 ret = fixup_inode_link_count(trans, root, inode);
Chris Masone02119d2008-09-05 16:13:11 -04001533 iput(inode);
Josef Bacik36508602013-04-25 16:23:32 -04001534 if (ret)
1535 goto out;
Chris Masone02119d2008-09-05 16:13:11 -04001536
Chris Mason12fcfd22009-03-24 10:24:20 -04001537 /*
1538 * fixup on a directory may create new entries,
1539 * make sure we always look for the highset possible
1540 * offset
1541 */
1542 key.offset = (u64)-1;
Chris Masone02119d2008-09-05 16:13:11 -04001543 }
Tsutomu Itoh65a246c2011-05-19 04:37:44 +00001544 ret = 0;
1545out:
David Sterbab3b4aa72011-04-21 01:20:15 +02001546 btrfs_release_path(path);
Tsutomu Itoh65a246c2011-05-19 04:37:44 +00001547 return ret;
Chris Masone02119d2008-09-05 16:13:11 -04001548}
1549
1550
1551/*
1552 * record a given inode in the fixup dir so we can check its link
1553 * count when replay is done. The link count is incremented here
1554 * so the inode won't go away until we check it
1555 */
1556static noinline int link_to_fixup_dir(struct btrfs_trans_handle *trans,
1557 struct btrfs_root *root,
1558 struct btrfs_path *path,
1559 u64 objectid)
1560{
1561 struct btrfs_key key;
1562 int ret = 0;
1563 struct inode *inode;
1564
1565 inode = read_one_inode(root, objectid);
Tsutomu Itohc00e9492011-04-28 09:10:23 +00001566 if (!inode)
1567 return -EIO;
Chris Masone02119d2008-09-05 16:13:11 -04001568
1569 key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID;
David Sterba962a2982014-06-04 18:41:45 +02001570 key.type = BTRFS_ORPHAN_ITEM_KEY;
Chris Masone02119d2008-09-05 16:13:11 -04001571 key.offset = objectid;
1572
1573 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1574
David Sterbab3b4aa72011-04-21 01:20:15 +02001575 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04001576 if (ret == 0) {
Josef Bacik9bf7a482013-03-01 13:35:47 -05001577 if (!inode->i_nlink)
1578 set_nlink(inode, 1);
1579 else
Zach Brown8b558c52013-10-16 12:10:34 -07001580 inc_nlink(inode);
Tsutomu Itohb9959292012-06-25 21:25:22 -06001581 ret = btrfs_update_inode(trans, root, inode);
Chris Masone02119d2008-09-05 16:13:11 -04001582 } else if (ret == -EEXIST) {
1583 ret = 0;
1584 } else {
Josef Bacik36508602013-04-25 16:23:32 -04001585 BUG(); /* Logic Error */
Chris Masone02119d2008-09-05 16:13:11 -04001586 }
1587 iput(inode);
1588
1589 return ret;
1590}
1591
1592/*
1593 * when replaying the log for a directory, we only insert names
1594 * for inodes that actually exist. This means an fsync on a directory
1595 * does not implicitly fsync all the new files in it
1596 */
1597static noinline int insert_one_name(struct btrfs_trans_handle *trans,
1598 struct btrfs_root *root,
Chris Masone02119d2008-09-05 16:13:11 -04001599 u64 dirid, u64 index,
Zhaolei60d53eb2015-08-17 18:44:46 +08001600 char *name, int name_len,
Chris Masone02119d2008-09-05 16:13:11 -04001601 struct btrfs_key *location)
1602{
1603 struct inode *inode;
1604 struct inode *dir;
1605 int ret;
1606
1607 inode = read_one_inode(root, location->objectid);
1608 if (!inode)
1609 return -ENOENT;
1610
1611 dir = read_one_inode(root, dirid);
1612 if (!dir) {
1613 iput(inode);
1614 return -EIO;
1615 }
Josef Bacikd5554382013-09-11 14:17:00 -04001616
Chris Masone02119d2008-09-05 16:13:11 -04001617 ret = btrfs_add_link(trans, dir, inode, name, name_len, 1, index);
1618
1619 /* FIXME, put inode into FIXUP list */
1620
1621 iput(inode);
1622 iput(dir);
1623 return ret;
1624}
1625
1626/*
Filipe Mananadf8d1162015-01-14 01:52:25 +00001627 * Return true if an inode reference exists in the log for the given name,
1628 * inode and parent inode.
1629 */
1630static bool name_in_log_ref(struct btrfs_root *log_root,
1631 const char *name, const int name_len,
1632 const u64 dirid, const u64 ino)
1633{
1634 struct btrfs_key search_key;
1635
1636 search_key.objectid = ino;
1637 search_key.type = BTRFS_INODE_REF_KEY;
1638 search_key.offset = dirid;
1639 if (backref_in_log(log_root, &search_key, dirid, name, name_len))
1640 return true;
1641
1642 search_key.type = BTRFS_INODE_EXTREF_KEY;
1643 search_key.offset = btrfs_extref_hash(dirid, name, name_len);
1644 if (backref_in_log(log_root, &search_key, dirid, name, name_len))
1645 return true;
1646
1647 return false;
1648}
1649
1650/*
Chris Masone02119d2008-09-05 16:13:11 -04001651 * take a single entry in a log directory item and replay it into
1652 * the subvolume.
1653 *
1654 * if a conflicting item exists in the subdirectory already,
1655 * the inode it points to is unlinked and put into the link count
1656 * fix up tree.
1657 *
1658 * If a name from the log points to a file or directory that does
1659 * not exist in the FS, it is skipped. fsyncs on directories
1660 * do not force down inodes inside that directory, just changes to the
1661 * names or unlinks in a directory.
Filipe Mananabb53eda2015-07-15 23:26:43 +01001662 *
1663 * Returns < 0 on error, 0 if the name wasn't replayed (dentry points to a
1664 * non-existing inode) and 1 if the name was replayed.
Chris Masone02119d2008-09-05 16:13:11 -04001665 */
1666static noinline int replay_one_name(struct btrfs_trans_handle *trans,
1667 struct btrfs_root *root,
1668 struct btrfs_path *path,
1669 struct extent_buffer *eb,
1670 struct btrfs_dir_item *di,
1671 struct btrfs_key *key)
1672{
1673 char *name;
1674 int name_len;
1675 struct btrfs_dir_item *dst_di;
1676 struct btrfs_key found_key;
1677 struct btrfs_key log_key;
1678 struct inode *dir;
Chris Masone02119d2008-09-05 16:13:11 -04001679 u8 log_type;
Chris Mason4bef0842008-09-08 11:18:08 -04001680 int exists;
Josef Bacik36508602013-04-25 16:23:32 -04001681 int ret = 0;
Josef Bacikd5554382013-09-11 14:17:00 -04001682 bool update_size = (key->type == BTRFS_DIR_INDEX_KEY);
Filipe Mananabb53eda2015-07-15 23:26:43 +01001683 bool name_added = false;
Chris Masone02119d2008-09-05 16:13:11 -04001684
1685 dir = read_one_inode(root, key->objectid);
Tsutomu Itohc00e9492011-04-28 09:10:23 +00001686 if (!dir)
1687 return -EIO;
Chris Masone02119d2008-09-05 16:13:11 -04001688
1689 name_len = btrfs_dir_name_len(eb, di);
1690 name = kmalloc(name_len, GFP_NOFS);
Filipe David Borba Manana2bac3252013-08-04 19:58:57 +01001691 if (!name) {
1692 ret = -ENOMEM;
1693 goto out;
1694 }
liubo2a29edc2011-01-26 06:22:08 +00001695
Chris Masone02119d2008-09-05 16:13:11 -04001696 log_type = btrfs_dir_type(eb, di);
1697 read_extent_buffer(eb, name, (unsigned long)(di + 1),
1698 name_len);
1699
1700 btrfs_dir_item_key_to_cpu(eb, di, &log_key);
Chris Mason4bef0842008-09-08 11:18:08 -04001701 exists = btrfs_lookup_inode(trans, root, path, &log_key, 0);
1702 if (exists == 0)
1703 exists = 1;
1704 else
1705 exists = 0;
David Sterbab3b4aa72011-04-21 01:20:15 +02001706 btrfs_release_path(path);
Chris Mason4bef0842008-09-08 11:18:08 -04001707
Chris Masone02119d2008-09-05 16:13:11 -04001708 if (key->type == BTRFS_DIR_ITEM_KEY) {
1709 dst_di = btrfs_lookup_dir_item(trans, root, path, key->objectid,
1710 name, name_len, 1);
Chris Masond3977122009-01-05 21:25:51 -05001711 } else if (key->type == BTRFS_DIR_INDEX_KEY) {
Chris Masone02119d2008-09-05 16:13:11 -04001712 dst_di = btrfs_lookup_dir_index_item(trans, root, path,
1713 key->objectid,
1714 key->offset, name,
1715 name_len, 1);
1716 } else {
Josef Bacik36508602013-04-25 16:23:32 -04001717 /* Corruption */
1718 ret = -EINVAL;
1719 goto out;
Chris Masone02119d2008-09-05 16:13:11 -04001720 }
David Sterbac7040052011-04-19 18:00:01 +02001721 if (IS_ERR_OR_NULL(dst_di)) {
Chris Masone02119d2008-09-05 16:13:11 -04001722 /* we need a sequence number to insert, so we only
1723 * do inserts for the BTRFS_DIR_INDEX_KEY types
1724 */
1725 if (key->type != BTRFS_DIR_INDEX_KEY)
1726 goto out;
1727 goto insert;
1728 }
1729
1730 btrfs_dir_item_key_to_cpu(path->nodes[0], dst_di, &found_key);
1731 /* the existing item matches the logged item */
1732 if (found_key.objectid == log_key.objectid &&
1733 found_key.type == log_key.type &&
1734 found_key.offset == log_key.offset &&
1735 btrfs_dir_type(path->nodes[0], dst_di) == log_type) {
Filipe Mananaa2cc11d2014-09-08 22:53:18 +01001736 update_size = false;
Chris Masone02119d2008-09-05 16:13:11 -04001737 goto out;
1738 }
1739
1740 /*
1741 * don't drop the conflicting directory entry if the inode
1742 * for the new entry doesn't exist
1743 */
Chris Mason4bef0842008-09-08 11:18:08 -04001744 if (!exists)
Chris Masone02119d2008-09-05 16:13:11 -04001745 goto out;
1746
Chris Masone02119d2008-09-05 16:13:11 -04001747 ret = drop_one_dir_item(trans, root, path, dir, dst_di);
Josef Bacik36508602013-04-25 16:23:32 -04001748 if (ret)
1749 goto out;
Chris Masone02119d2008-09-05 16:13:11 -04001750
1751 if (key->type == BTRFS_DIR_INDEX_KEY)
1752 goto insert;
1753out:
David Sterbab3b4aa72011-04-21 01:20:15 +02001754 btrfs_release_path(path);
Josef Bacikd5554382013-09-11 14:17:00 -04001755 if (!ret && update_size) {
1756 btrfs_i_size_write(dir, dir->i_size + name_len * 2);
1757 ret = btrfs_update_inode(trans, root, dir);
1758 }
Chris Masone02119d2008-09-05 16:13:11 -04001759 kfree(name);
1760 iput(dir);
Filipe Mananabb53eda2015-07-15 23:26:43 +01001761 if (!ret && name_added)
1762 ret = 1;
Josef Bacik36508602013-04-25 16:23:32 -04001763 return ret;
Chris Masone02119d2008-09-05 16:13:11 -04001764
1765insert:
Filipe Mananadf8d1162015-01-14 01:52:25 +00001766 if (name_in_log_ref(root->log_root, name, name_len,
1767 key->objectid, log_key.objectid)) {
1768 /* The dentry will be added later. */
1769 ret = 0;
1770 update_size = false;
1771 goto out;
1772 }
David Sterbab3b4aa72011-04-21 01:20:15 +02001773 btrfs_release_path(path);
Zhaolei60d53eb2015-08-17 18:44:46 +08001774 ret = insert_one_name(trans, root, key->objectid, key->offset,
1775 name, name_len, &log_key);
Filipe Mananadf8d1162015-01-14 01:52:25 +00001776 if (ret && ret != -ENOENT && ret != -EEXIST)
Josef Bacik36508602013-04-25 16:23:32 -04001777 goto out;
Filipe Mananabb53eda2015-07-15 23:26:43 +01001778 if (!ret)
1779 name_added = true;
Josef Bacikd5554382013-09-11 14:17:00 -04001780 update_size = false;
Josef Bacik36508602013-04-25 16:23:32 -04001781 ret = 0;
Chris Masone02119d2008-09-05 16:13:11 -04001782 goto out;
1783}
1784
1785/*
1786 * find all the names in a directory item and reconcile them into
1787 * the subvolume. Only BTRFS_DIR_ITEM_KEY types will have more than
1788 * one name in a directory item, but the same code gets used for
1789 * both directory index types
1790 */
1791static noinline int replay_one_dir_item(struct btrfs_trans_handle *trans,
1792 struct btrfs_root *root,
1793 struct btrfs_path *path,
1794 struct extent_buffer *eb, int slot,
1795 struct btrfs_key *key)
1796{
Filipe Mananabb53eda2015-07-15 23:26:43 +01001797 int ret = 0;
Chris Masone02119d2008-09-05 16:13:11 -04001798 u32 item_size = btrfs_item_size_nr(eb, slot);
1799 struct btrfs_dir_item *di;
1800 int name_len;
1801 unsigned long ptr;
1802 unsigned long ptr_end;
Filipe Mananabb53eda2015-07-15 23:26:43 +01001803 struct btrfs_path *fixup_path = NULL;
Chris Masone02119d2008-09-05 16:13:11 -04001804
1805 ptr = btrfs_item_ptr_offset(eb, slot);
1806 ptr_end = ptr + item_size;
Chris Masond3977122009-01-05 21:25:51 -05001807 while (ptr < ptr_end) {
Chris Masone02119d2008-09-05 16:13:11 -04001808 di = (struct btrfs_dir_item *)ptr;
Josef Bacik22a94d42011-03-16 16:47:17 -04001809 if (verify_dir_item(root, eb, di))
1810 return -EIO;
Chris Masone02119d2008-09-05 16:13:11 -04001811 name_len = btrfs_dir_name_len(eb, di);
1812 ret = replay_one_name(trans, root, path, eb, di, key);
Filipe Mananabb53eda2015-07-15 23:26:43 +01001813 if (ret < 0)
1814 break;
Chris Masone02119d2008-09-05 16:13:11 -04001815 ptr = (unsigned long)(di + 1);
1816 ptr += name_len;
Filipe Mananabb53eda2015-07-15 23:26:43 +01001817
1818 /*
1819 * If this entry refers to a non-directory (directories can not
1820 * have a link count > 1) and it was added in the transaction
1821 * that was not committed, make sure we fixup the link count of
1822 * the inode it the entry points to. Otherwise something like
1823 * the following would result in a directory pointing to an
1824 * inode with a wrong link that does not account for this dir
1825 * entry:
1826 *
1827 * mkdir testdir
1828 * touch testdir/foo
1829 * touch testdir/bar
1830 * sync
1831 *
1832 * ln testdir/bar testdir/bar_link
1833 * ln testdir/foo testdir/foo_link
1834 * xfs_io -c "fsync" testdir/bar
1835 *
1836 * <power failure>
1837 *
1838 * mount fs, log replay happens
1839 *
1840 * File foo would remain with a link count of 1 when it has two
1841 * entries pointing to it in the directory testdir. This would
1842 * make it impossible to ever delete the parent directory has
1843 * it would result in stale dentries that can never be deleted.
1844 */
1845 if (ret == 1 && btrfs_dir_type(eb, di) != BTRFS_FT_DIR) {
1846 struct btrfs_key di_key;
1847
1848 if (!fixup_path) {
1849 fixup_path = btrfs_alloc_path();
1850 if (!fixup_path) {
1851 ret = -ENOMEM;
1852 break;
1853 }
1854 }
1855
1856 btrfs_dir_item_key_to_cpu(eb, di, &di_key);
1857 ret = link_to_fixup_dir(trans, root, fixup_path,
1858 di_key.objectid);
1859 if (ret)
1860 break;
1861 }
1862 ret = 0;
Chris Masone02119d2008-09-05 16:13:11 -04001863 }
Filipe Mananabb53eda2015-07-15 23:26:43 +01001864 btrfs_free_path(fixup_path);
1865 return ret;
Chris Masone02119d2008-09-05 16:13:11 -04001866}
1867
1868/*
1869 * directory replay has two parts. There are the standard directory
1870 * items in the log copied from the subvolume, and range items
1871 * created in the log while the subvolume was logged.
1872 *
1873 * The range items tell us which parts of the key space the log
1874 * is authoritative for. During replay, if a key in the subvolume
1875 * directory is in a logged range item, but not actually in the log
1876 * that means it was deleted from the directory before the fsync
1877 * and should be removed.
1878 */
1879static noinline int find_dir_range(struct btrfs_root *root,
1880 struct btrfs_path *path,
1881 u64 dirid, int key_type,
1882 u64 *start_ret, u64 *end_ret)
1883{
1884 struct btrfs_key key;
1885 u64 found_end;
1886 struct btrfs_dir_log_item *item;
1887 int ret;
1888 int nritems;
1889
1890 if (*start_ret == (u64)-1)
1891 return 1;
1892
1893 key.objectid = dirid;
1894 key.type = key_type;
1895 key.offset = *start_ret;
1896
1897 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1898 if (ret < 0)
1899 goto out;
1900 if (ret > 0) {
1901 if (path->slots[0] == 0)
1902 goto out;
1903 path->slots[0]--;
1904 }
1905 if (ret != 0)
1906 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1907
1908 if (key.type != key_type || key.objectid != dirid) {
1909 ret = 1;
1910 goto next;
1911 }
1912 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
1913 struct btrfs_dir_log_item);
1914 found_end = btrfs_dir_log_end(path->nodes[0], item);
1915
1916 if (*start_ret >= key.offset && *start_ret <= found_end) {
1917 ret = 0;
1918 *start_ret = key.offset;
1919 *end_ret = found_end;
1920 goto out;
1921 }
1922 ret = 1;
1923next:
1924 /* check the next slot in the tree to see if it is a valid item */
1925 nritems = btrfs_header_nritems(path->nodes[0]);
1926 if (path->slots[0] >= nritems) {
1927 ret = btrfs_next_leaf(root, path);
1928 if (ret)
1929 goto out;
1930 } else {
1931 path->slots[0]++;
1932 }
1933
1934 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1935
1936 if (key.type != key_type || key.objectid != dirid) {
1937 ret = 1;
1938 goto out;
1939 }
1940 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
1941 struct btrfs_dir_log_item);
1942 found_end = btrfs_dir_log_end(path->nodes[0], item);
1943 *start_ret = key.offset;
1944 *end_ret = found_end;
1945 ret = 0;
1946out:
David Sterbab3b4aa72011-04-21 01:20:15 +02001947 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04001948 return ret;
1949}
1950
1951/*
1952 * this looks for a given directory item in the log. If the directory
1953 * item is not in the log, the item is removed and the inode it points
1954 * to is unlinked
1955 */
1956static noinline int check_item_in_log(struct btrfs_trans_handle *trans,
1957 struct btrfs_root *root,
1958 struct btrfs_root *log,
1959 struct btrfs_path *path,
1960 struct btrfs_path *log_path,
1961 struct inode *dir,
1962 struct btrfs_key *dir_key)
1963{
1964 int ret;
1965 struct extent_buffer *eb;
1966 int slot;
1967 u32 item_size;
1968 struct btrfs_dir_item *di;
1969 struct btrfs_dir_item *log_di;
1970 int name_len;
1971 unsigned long ptr;
1972 unsigned long ptr_end;
1973 char *name;
1974 struct inode *inode;
1975 struct btrfs_key location;
1976
1977again:
1978 eb = path->nodes[0];
1979 slot = path->slots[0];
1980 item_size = btrfs_item_size_nr(eb, slot);
1981 ptr = btrfs_item_ptr_offset(eb, slot);
1982 ptr_end = ptr + item_size;
Chris Masond3977122009-01-05 21:25:51 -05001983 while (ptr < ptr_end) {
Chris Masone02119d2008-09-05 16:13:11 -04001984 di = (struct btrfs_dir_item *)ptr;
Josef Bacik22a94d42011-03-16 16:47:17 -04001985 if (verify_dir_item(root, eb, di)) {
1986 ret = -EIO;
1987 goto out;
1988 }
1989
Chris Masone02119d2008-09-05 16:13:11 -04001990 name_len = btrfs_dir_name_len(eb, di);
1991 name = kmalloc(name_len, GFP_NOFS);
1992 if (!name) {
1993 ret = -ENOMEM;
1994 goto out;
1995 }
1996 read_extent_buffer(eb, name, (unsigned long)(di + 1),
1997 name_len);
1998 log_di = NULL;
Chris Mason12fcfd22009-03-24 10:24:20 -04001999 if (log && dir_key->type == BTRFS_DIR_ITEM_KEY) {
Chris Masone02119d2008-09-05 16:13:11 -04002000 log_di = btrfs_lookup_dir_item(trans, log, log_path,
2001 dir_key->objectid,
2002 name, name_len, 0);
Chris Mason12fcfd22009-03-24 10:24:20 -04002003 } else if (log && dir_key->type == BTRFS_DIR_INDEX_KEY) {
Chris Masone02119d2008-09-05 16:13:11 -04002004 log_di = btrfs_lookup_dir_index_item(trans, log,
2005 log_path,
2006 dir_key->objectid,
2007 dir_key->offset,
2008 name, name_len, 0);
2009 }
Filipe David Borba Manana269d0402013-10-28 17:39:21 +00002010 if (!log_di || (IS_ERR(log_di) && PTR_ERR(log_di) == -ENOENT)) {
Chris Masone02119d2008-09-05 16:13:11 -04002011 btrfs_dir_item_key_to_cpu(eb, di, &location);
David Sterbab3b4aa72011-04-21 01:20:15 +02002012 btrfs_release_path(path);
2013 btrfs_release_path(log_path);
Chris Masone02119d2008-09-05 16:13:11 -04002014 inode = read_one_inode(root, location.objectid);
Tsutomu Itohc00e9492011-04-28 09:10:23 +00002015 if (!inode) {
2016 kfree(name);
2017 return -EIO;
2018 }
Chris Masone02119d2008-09-05 16:13:11 -04002019
2020 ret = link_to_fixup_dir(trans, root,
2021 path, location.objectid);
Josef Bacik36508602013-04-25 16:23:32 -04002022 if (ret) {
2023 kfree(name);
2024 iput(inode);
2025 goto out;
2026 }
2027
Zach Brown8b558c52013-10-16 12:10:34 -07002028 inc_nlink(inode);
Chris Masone02119d2008-09-05 16:13:11 -04002029 ret = btrfs_unlink_inode(trans, root, dir, inode,
2030 name, name_len);
Josef Bacik36508602013-04-25 16:23:32 -04002031 if (!ret)
Filipe David Borba Mananaada9af22013-08-05 09:25:47 +01002032 ret = btrfs_run_delayed_items(trans, root);
Chris Masone02119d2008-09-05 16:13:11 -04002033 kfree(name);
2034 iput(inode);
Josef Bacik36508602013-04-25 16:23:32 -04002035 if (ret)
2036 goto out;
Chris Masone02119d2008-09-05 16:13:11 -04002037
2038 /* there might still be more names under this key
2039 * check and repeat if required
2040 */
2041 ret = btrfs_search_slot(NULL, root, dir_key, path,
2042 0, 0);
2043 if (ret == 0)
2044 goto again;
2045 ret = 0;
2046 goto out;
Filipe David Borba Manana269d0402013-10-28 17:39:21 +00002047 } else if (IS_ERR(log_di)) {
2048 kfree(name);
2049 return PTR_ERR(log_di);
Chris Masone02119d2008-09-05 16:13:11 -04002050 }
David Sterbab3b4aa72011-04-21 01:20:15 +02002051 btrfs_release_path(log_path);
Chris Masone02119d2008-09-05 16:13:11 -04002052 kfree(name);
2053
2054 ptr = (unsigned long)(di + 1);
2055 ptr += name_len;
2056 }
2057 ret = 0;
2058out:
David Sterbab3b4aa72011-04-21 01:20:15 +02002059 btrfs_release_path(path);
2060 btrfs_release_path(log_path);
Chris Masone02119d2008-09-05 16:13:11 -04002061 return ret;
2062}
2063
Filipe Manana4f764e52015-02-23 19:53:35 +00002064static int replay_xattr_deletes(struct btrfs_trans_handle *trans,
2065 struct btrfs_root *root,
2066 struct btrfs_root *log,
2067 struct btrfs_path *path,
2068 const u64 ino)
2069{
2070 struct btrfs_key search_key;
2071 struct btrfs_path *log_path;
2072 int i;
2073 int nritems;
2074 int ret;
2075
2076 log_path = btrfs_alloc_path();
2077 if (!log_path)
2078 return -ENOMEM;
2079
2080 search_key.objectid = ino;
2081 search_key.type = BTRFS_XATTR_ITEM_KEY;
2082 search_key.offset = 0;
2083again:
2084 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
2085 if (ret < 0)
2086 goto out;
2087process_leaf:
2088 nritems = btrfs_header_nritems(path->nodes[0]);
2089 for (i = path->slots[0]; i < nritems; i++) {
2090 struct btrfs_key key;
2091 struct btrfs_dir_item *di;
2092 struct btrfs_dir_item *log_di;
2093 u32 total_size;
2094 u32 cur;
2095
2096 btrfs_item_key_to_cpu(path->nodes[0], &key, i);
2097 if (key.objectid != ino || key.type != BTRFS_XATTR_ITEM_KEY) {
2098 ret = 0;
2099 goto out;
2100 }
2101
2102 di = btrfs_item_ptr(path->nodes[0], i, struct btrfs_dir_item);
2103 total_size = btrfs_item_size_nr(path->nodes[0], i);
2104 cur = 0;
2105 while (cur < total_size) {
2106 u16 name_len = btrfs_dir_name_len(path->nodes[0], di);
2107 u16 data_len = btrfs_dir_data_len(path->nodes[0], di);
2108 u32 this_len = sizeof(*di) + name_len + data_len;
2109 char *name;
2110
2111 name = kmalloc(name_len, GFP_NOFS);
2112 if (!name) {
2113 ret = -ENOMEM;
2114 goto out;
2115 }
2116 read_extent_buffer(path->nodes[0], name,
2117 (unsigned long)(di + 1), name_len);
2118
2119 log_di = btrfs_lookup_xattr(NULL, log, log_path, ino,
2120 name, name_len, 0);
2121 btrfs_release_path(log_path);
2122 if (!log_di) {
2123 /* Doesn't exist in log tree, so delete it. */
2124 btrfs_release_path(path);
2125 di = btrfs_lookup_xattr(trans, root, path, ino,
2126 name, name_len, -1);
2127 kfree(name);
2128 if (IS_ERR(di)) {
2129 ret = PTR_ERR(di);
2130 goto out;
2131 }
2132 ASSERT(di);
2133 ret = btrfs_delete_one_dir_name(trans, root,
2134 path, di);
2135 if (ret)
2136 goto out;
2137 btrfs_release_path(path);
2138 search_key = key;
2139 goto again;
2140 }
2141 kfree(name);
2142 if (IS_ERR(log_di)) {
2143 ret = PTR_ERR(log_di);
2144 goto out;
2145 }
2146 cur += this_len;
2147 di = (struct btrfs_dir_item *)((char *)di + this_len);
2148 }
2149 }
2150 ret = btrfs_next_leaf(root, path);
2151 if (ret > 0)
2152 ret = 0;
2153 else if (ret == 0)
2154 goto process_leaf;
2155out:
2156 btrfs_free_path(log_path);
2157 btrfs_release_path(path);
2158 return ret;
2159}
2160
2161
Chris Masone02119d2008-09-05 16:13:11 -04002162/*
2163 * deletion replay happens before we copy any new directory items
2164 * out of the log or out of backreferences from inodes. It
2165 * scans the log to find ranges of keys that log is authoritative for,
2166 * and then scans the directory to find items in those ranges that are
2167 * not present in the log.
2168 *
2169 * Anything we don't find in the log is unlinked and removed from the
2170 * directory.
2171 */
2172static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans,
2173 struct btrfs_root *root,
2174 struct btrfs_root *log,
2175 struct btrfs_path *path,
Chris Mason12fcfd22009-03-24 10:24:20 -04002176 u64 dirid, int del_all)
Chris Masone02119d2008-09-05 16:13:11 -04002177{
2178 u64 range_start;
2179 u64 range_end;
2180 int key_type = BTRFS_DIR_LOG_ITEM_KEY;
2181 int ret = 0;
2182 struct btrfs_key dir_key;
2183 struct btrfs_key found_key;
2184 struct btrfs_path *log_path;
2185 struct inode *dir;
2186
2187 dir_key.objectid = dirid;
2188 dir_key.type = BTRFS_DIR_ITEM_KEY;
2189 log_path = btrfs_alloc_path();
2190 if (!log_path)
2191 return -ENOMEM;
2192
2193 dir = read_one_inode(root, dirid);
2194 /* it isn't an error if the inode isn't there, that can happen
2195 * because we replay the deletes before we copy in the inode item
2196 * from the log
2197 */
2198 if (!dir) {
2199 btrfs_free_path(log_path);
2200 return 0;
2201 }
2202again:
2203 range_start = 0;
2204 range_end = 0;
Chris Masond3977122009-01-05 21:25:51 -05002205 while (1) {
Chris Mason12fcfd22009-03-24 10:24:20 -04002206 if (del_all)
2207 range_end = (u64)-1;
2208 else {
2209 ret = find_dir_range(log, path, dirid, key_type,
2210 &range_start, &range_end);
2211 if (ret != 0)
2212 break;
2213 }
Chris Masone02119d2008-09-05 16:13:11 -04002214
2215 dir_key.offset = range_start;
Chris Masond3977122009-01-05 21:25:51 -05002216 while (1) {
Chris Masone02119d2008-09-05 16:13:11 -04002217 int nritems;
2218 ret = btrfs_search_slot(NULL, root, &dir_key, path,
2219 0, 0);
2220 if (ret < 0)
2221 goto out;
2222
2223 nritems = btrfs_header_nritems(path->nodes[0]);
2224 if (path->slots[0] >= nritems) {
2225 ret = btrfs_next_leaf(root, path);
2226 if (ret)
2227 break;
2228 }
2229 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2230 path->slots[0]);
2231 if (found_key.objectid != dirid ||
2232 found_key.type != dir_key.type)
2233 goto next_type;
2234
2235 if (found_key.offset > range_end)
2236 break;
2237
2238 ret = check_item_in_log(trans, root, log, path,
Chris Mason12fcfd22009-03-24 10:24:20 -04002239 log_path, dir,
2240 &found_key);
Josef Bacik36508602013-04-25 16:23:32 -04002241 if (ret)
2242 goto out;
Chris Masone02119d2008-09-05 16:13:11 -04002243 if (found_key.offset == (u64)-1)
2244 break;
2245 dir_key.offset = found_key.offset + 1;
2246 }
David Sterbab3b4aa72011-04-21 01:20:15 +02002247 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04002248 if (range_end == (u64)-1)
2249 break;
2250 range_start = range_end + 1;
2251 }
2252
2253next_type:
2254 ret = 0;
2255 if (key_type == BTRFS_DIR_LOG_ITEM_KEY) {
2256 key_type = BTRFS_DIR_LOG_INDEX_KEY;
2257 dir_key.type = BTRFS_DIR_INDEX_KEY;
David Sterbab3b4aa72011-04-21 01:20:15 +02002258 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04002259 goto again;
2260 }
2261out:
David Sterbab3b4aa72011-04-21 01:20:15 +02002262 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04002263 btrfs_free_path(log_path);
2264 iput(dir);
2265 return ret;
2266}
2267
2268/*
2269 * the process_func used to replay items from the log tree. This
2270 * gets called in two different stages. The first stage just looks
2271 * for inodes and makes sure they are all copied into the subvolume.
2272 *
2273 * The second stage copies all the other item types from the log into
2274 * the subvolume. The two stage approach is slower, but gets rid of
2275 * lots of complexity around inodes referencing other inodes that exist
2276 * only in the log (references come from either directory items or inode
2277 * back refs).
2278 */
2279static int replay_one_buffer(struct btrfs_root *log, struct extent_buffer *eb,
2280 struct walk_control *wc, u64 gen)
2281{
2282 int nritems;
2283 struct btrfs_path *path;
2284 struct btrfs_root *root = wc->replay_dest;
2285 struct btrfs_key key;
Chris Masone02119d2008-09-05 16:13:11 -04002286 int level;
2287 int i;
2288 int ret;
2289
Tsutomu Itoh018642a2012-05-29 18:10:13 +09002290 ret = btrfs_read_buffer(eb, gen);
2291 if (ret)
2292 return ret;
Chris Masone02119d2008-09-05 16:13:11 -04002293
2294 level = btrfs_header_level(eb);
2295
2296 if (level != 0)
2297 return 0;
2298
2299 path = btrfs_alloc_path();
Mark Fasheh1e5063d2011-07-12 10:46:06 -07002300 if (!path)
2301 return -ENOMEM;
Chris Masone02119d2008-09-05 16:13:11 -04002302
2303 nritems = btrfs_header_nritems(eb);
2304 for (i = 0; i < nritems; i++) {
2305 btrfs_item_key_to_cpu(eb, &key, i);
Chris Masone02119d2008-09-05 16:13:11 -04002306
2307 /* inode keys are done during the first stage */
2308 if (key.type == BTRFS_INODE_ITEM_KEY &&
2309 wc->stage == LOG_WALK_REPLAY_INODES) {
Chris Masone02119d2008-09-05 16:13:11 -04002310 struct btrfs_inode_item *inode_item;
2311 u32 mode;
2312
2313 inode_item = btrfs_item_ptr(eb, i,
2314 struct btrfs_inode_item);
Filipe Manana4f764e52015-02-23 19:53:35 +00002315 ret = replay_xattr_deletes(wc->trans, root, log,
2316 path, key.objectid);
2317 if (ret)
2318 break;
Chris Masone02119d2008-09-05 16:13:11 -04002319 mode = btrfs_inode_mode(eb, inode_item);
2320 if (S_ISDIR(mode)) {
2321 ret = replay_dir_deletes(wc->trans,
Chris Mason12fcfd22009-03-24 10:24:20 -04002322 root, log, path, key.objectid, 0);
Josef Bacikb50c6e22013-04-25 15:55:30 -04002323 if (ret)
2324 break;
Chris Masone02119d2008-09-05 16:13:11 -04002325 }
2326 ret = overwrite_item(wc->trans, root, path,
2327 eb, i, &key);
Josef Bacikb50c6e22013-04-25 15:55:30 -04002328 if (ret)
2329 break;
Chris Masone02119d2008-09-05 16:13:11 -04002330
Yan, Zhengc71bf092009-11-12 09:34:40 +00002331 /* for regular files, make sure corresponding
2332 * orhpan item exist. extents past the new EOF
2333 * will be truncated later by orphan cleanup.
Chris Masone02119d2008-09-05 16:13:11 -04002334 */
2335 if (S_ISREG(mode)) {
Yan, Zhengc71bf092009-11-12 09:34:40 +00002336 ret = insert_orphan_item(wc->trans, root,
2337 key.objectid);
Josef Bacikb50c6e22013-04-25 15:55:30 -04002338 if (ret)
2339 break;
Chris Masone02119d2008-09-05 16:13:11 -04002340 }
Yan, Zhengc71bf092009-11-12 09:34:40 +00002341
Chris Masone02119d2008-09-05 16:13:11 -04002342 ret = link_to_fixup_dir(wc->trans, root,
2343 path, key.objectid);
Josef Bacikb50c6e22013-04-25 15:55:30 -04002344 if (ret)
2345 break;
Chris Masone02119d2008-09-05 16:13:11 -04002346 }
Josef Bacikdd8e7212013-09-11 11:57:23 -04002347
2348 if (key.type == BTRFS_DIR_INDEX_KEY &&
2349 wc->stage == LOG_WALK_REPLAY_DIR_INDEX) {
2350 ret = replay_one_dir_item(wc->trans, root, path,
2351 eb, i, &key);
2352 if (ret)
2353 break;
2354 }
2355
Chris Masone02119d2008-09-05 16:13:11 -04002356 if (wc->stage < LOG_WALK_REPLAY_ALL)
2357 continue;
2358
2359 /* these keys are simply copied */
2360 if (key.type == BTRFS_XATTR_ITEM_KEY) {
2361 ret = overwrite_item(wc->trans, root, path,
2362 eb, i, &key);
Josef Bacikb50c6e22013-04-25 15:55:30 -04002363 if (ret)
2364 break;
Liu Bo2da1c662013-05-26 13:50:29 +00002365 } else if (key.type == BTRFS_INODE_REF_KEY ||
2366 key.type == BTRFS_INODE_EXTREF_KEY) {
Mark Fashehf1863732012-08-08 11:32:27 -07002367 ret = add_inode_ref(wc->trans, root, log, path,
2368 eb, i, &key);
Josef Bacikb50c6e22013-04-25 15:55:30 -04002369 if (ret && ret != -ENOENT)
2370 break;
2371 ret = 0;
Chris Masone02119d2008-09-05 16:13:11 -04002372 } else if (key.type == BTRFS_EXTENT_DATA_KEY) {
2373 ret = replay_one_extent(wc->trans, root, path,
2374 eb, i, &key);
Josef Bacikb50c6e22013-04-25 15:55:30 -04002375 if (ret)
2376 break;
Josef Bacikdd8e7212013-09-11 11:57:23 -04002377 } else if (key.type == BTRFS_DIR_ITEM_KEY) {
Chris Masone02119d2008-09-05 16:13:11 -04002378 ret = replay_one_dir_item(wc->trans, root, path,
2379 eb, i, &key);
Josef Bacikb50c6e22013-04-25 15:55:30 -04002380 if (ret)
2381 break;
Chris Masone02119d2008-09-05 16:13:11 -04002382 }
2383 }
2384 btrfs_free_path(path);
Josef Bacikb50c6e22013-04-25 15:55:30 -04002385 return ret;
Chris Masone02119d2008-09-05 16:13:11 -04002386}
2387
Chris Masond3977122009-01-05 21:25:51 -05002388static noinline int walk_down_log_tree(struct btrfs_trans_handle *trans,
Chris Masone02119d2008-09-05 16:13:11 -04002389 struct btrfs_root *root,
2390 struct btrfs_path *path, int *level,
2391 struct walk_control *wc)
2392{
2393 u64 root_owner;
Chris Masone02119d2008-09-05 16:13:11 -04002394 u64 bytenr;
2395 u64 ptr_gen;
2396 struct extent_buffer *next;
2397 struct extent_buffer *cur;
2398 struct extent_buffer *parent;
2399 u32 blocksize;
2400 int ret = 0;
2401
2402 WARN_ON(*level < 0);
2403 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2404
Chris Masond3977122009-01-05 21:25:51 -05002405 while (*level > 0) {
Chris Masone02119d2008-09-05 16:13:11 -04002406 WARN_ON(*level < 0);
2407 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2408 cur = path->nodes[*level];
2409
Dulshani Gunawardhanafae7f212013-10-31 10:30:08 +05302410 WARN_ON(btrfs_header_level(cur) != *level);
Chris Masone02119d2008-09-05 16:13:11 -04002411
2412 if (path->slots[*level] >=
2413 btrfs_header_nritems(cur))
2414 break;
2415
2416 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
2417 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
David Sterba707e8a02014-06-04 19:22:26 +02002418 blocksize = root->nodesize;
Chris Masone02119d2008-09-05 16:13:11 -04002419
2420 parent = path->nodes[*level];
2421 root_owner = btrfs_header_owner(parent);
Chris Masone02119d2008-09-05 16:13:11 -04002422
David Sterbaa83fffb2014-06-15 02:39:54 +02002423 next = btrfs_find_create_tree_block(root, bytenr);
liubo2a29edc2011-01-26 06:22:08 +00002424 if (!next)
2425 return -ENOMEM;
Chris Masone02119d2008-09-05 16:13:11 -04002426
Chris Masone02119d2008-09-05 16:13:11 -04002427 if (*level == 1) {
Mark Fasheh1e5063d2011-07-12 10:46:06 -07002428 ret = wc->process_func(root, next, wc, ptr_gen);
Josef Bacikb50c6e22013-04-25 15:55:30 -04002429 if (ret) {
2430 free_extent_buffer(next);
Mark Fasheh1e5063d2011-07-12 10:46:06 -07002431 return ret;
Josef Bacikb50c6e22013-04-25 15:55:30 -04002432 }
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002433
Chris Masone02119d2008-09-05 16:13:11 -04002434 path->slots[*level]++;
2435 if (wc->free) {
Tsutomu Itoh018642a2012-05-29 18:10:13 +09002436 ret = btrfs_read_buffer(next, ptr_gen);
2437 if (ret) {
2438 free_extent_buffer(next);
2439 return ret;
2440 }
Chris Masone02119d2008-09-05 16:13:11 -04002441
Josef Bacik681ae502013-10-07 15:11:00 -04002442 if (trans) {
2443 btrfs_tree_lock(next);
2444 btrfs_set_lock_blocking(next);
Daniel Dressler01d58472014-11-21 17:15:07 +09002445 clean_tree_block(trans, root->fs_info,
2446 next);
Josef Bacik681ae502013-10-07 15:11:00 -04002447 btrfs_wait_tree_block_writeback(next);
2448 btrfs_tree_unlock(next);
2449 }
Chris Masone02119d2008-09-05 16:13:11 -04002450
Chris Masone02119d2008-09-05 16:13:11 -04002451 WARN_ON(root_owner !=
2452 BTRFS_TREE_LOG_OBJECTID);
Chris Masone688b7252011-10-31 20:52:39 -04002453 ret = btrfs_free_and_pin_reserved_extent(root,
Chris Masond00aff02008-09-11 15:54:42 -04002454 bytenr, blocksize);
Josef Bacik36508602013-04-25 16:23:32 -04002455 if (ret) {
2456 free_extent_buffer(next);
2457 return ret;
2458 }
Chris Masone02119d2008-09-05 16:13:11 -04002459 }
2460 free_extent_buffer(next);
2461 continue;
2462 }
Tsutomu Itoh018642a2012-05-29 18:10:13 +09002463 ret = btrfs_read_buffer(next, ptr_gen);
2464 if (ret) {
2465 free_extent_buffer(next);
2466 return ret;
2467 }
Chris Masone02119d2008-09-05 16:13:11 -04002468
2469 WARN_ON(*level <= 0);
2470 if (path->nodes[*level-1])
2471 free_extent_buffer(path->nodes[*level-1]);
2472 path->nodes[*level-1] = next;
2473 *level = btrfs_header_level(next);
2474 path->slots[*level] = 0;
2475 cond_resched();
2476 }
2477 WARN_ON(*level < 0);
2478 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2479
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002480 path->slots[*level] = btrfs_header_nritems(path->nodes[*level]);
Chris Masone02119d2008-09-05 16:13:11 -04002481
2482 cond_resched();
2483 return 0;
2484}
2485
Chris Masond3977122009-01-05 21:25:51 -05002486static noinline int walk_up_log_tree(struct btrfs_trans_handle *trans,
Chris Masone02119d2008-09-05 16:13:11 -04002487 struct btrfs_root *root,
2488 struct btrfs_path *path, int *level,
2489 struct walk_control *wc)
2490{
2491 u64 root_owner;
Chris Masone02119d2008-09-05 16:13:11 -04002492 int i;
2493 int slot;
2494 int ret;
2495
Chris Masond3977122009-01-05 21:25:51 -05002496 for (i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) {
Chris Masone02119d2008-09-05 16:13:11 -04002497 slot = path->slots[i];
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002498 if (slot + 1 < btrfs_header_nritems(path->nodes[i])) {
Chris Masone02119d2008-09-05 16:13:11 -04002499 path->slots[i]++;
2500 *level = i;
2501 WARN_ON(*level == 0);
2502 return 0;
2503 } else {
Zheng Yan31840ae2008-09-23 13:14:14 -04002504 struct extent_buffer *parent;
2505 if (path->nodes[*level] == root->node)
2506 parent = path->nodes[*level];
2507 else
2508 parent = path->nodes[*level + 1];
2509
2510 root_owner = btrfs_header_owner(parent);
Mark Fasheh1e5063d2011-07-12 10:46:06 -07002511 ret = wc->process_func(root, path->nodes[*level], wc,
Chris Masone02119d2008-09-05 16:13:11 -04002512 btrfs_header_generation(path->nodes[*level]));
Mark Fasheh1e5063d2011-07-12 10:46:06 -07002513 if (ret)
2514 return ret;
2515
Chris Masone02119d2008-09-05 16:13:11 -04002516 if (wc->free) {
2517 struct extent_buffer *next;
2518
2519 next = path->nodes[*level];
2520
Josef Bacik681ae502013-10-07 15:11:00 -04002521 if (trans) {
2522 btrfs_tree_lock(next);
2523 btrfs_set_lock_blocking(next);
Daniel Dressler01d58472014-11-21 17:15:07 +09002524 clean_tree_block(trans, root->fs_info,
2525 next);
Josef Bacik681ae502013-10-07 15:11:00 -04002526 btrfs_wait_tree_block_writeback(next);
2527 btrfs_tree_unlock(next);
2528 }
Chris Masone02119d2008-09-05 16:13:11 -04002529
Chris Masone02119d2008-09-05 16:13:11 -04002530 WARN_ON(root_owner != BTRFS_TREE_LOG_OBJECTID);
Chris Masone688b7252011-10-31 20:52:39 -04002531 ret = btrfs_free_and_pin_reserved_extent(root,
Chris Masone02119d2008-09-05 16:13:11 -04002532 path->nodes[*level]->start,
Chris Masond00aff02008-09-11 15:54:42 -04002533 path->nodes[*level]->len);
Josef Bacik36508602013-04-25 16:23:32 -04002534 if (ret)
2535 return ret;
Chris Masone02119d2008-09-05 16:13:11 -04002536 }
2537 free_extent_buffer(path->nodes[*level]);
2538 path->nodes[*level] = NULL;
2539 *level = i + 1;
2540 }
2541 }
2542 return 1;
2543}
2544
2545/*
2546 * drop the reference count on the tree rooted at 'snap'. This traverses
2547 * the tree freeing any blocks that have a ref count of zero after being
2548 * decremented.
2549 */
2550static int walk_log_tree(struct btrfs_trans_handle *trans,
2551 struct btrfs_root *log, struct walk_control *wc)
2552{
2553 int ret = 0;
2554 int wret;
2555 int level;
2556 struct btrfs_path *path;
Chris Masone02119d2008-09-05 16:13:11 -04002557 int orig_level;
2558
2559 path = btrfs_alloc_path();
Tsutomu Itohdb5b4932011-03-23 08:14:16 +00002560 if (!path)
2561 return -ENOMEM;
Chris Masone02119d2008-09-05 16:13:11 -04002562
2563 level = btrfs_header_level(log->node);
2564 orig_level = level;
2565 path->nodes[level] = log->node;
2566 extent_buffer_get(log->node);
2567 path->slots[level] = 0;
2568
Chris Masond3977122009-01-05 21:25:51 -05002569 while (1) {
Chris Masone02119d2008-09-05 16:13:11 -04002570 wret = walk_down_log_tree(trans, log, path, &level, wc);
2571 if (wret > 0)
2572 break;
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002573 if (wret < 0) {
Chris Masone02119d2008-09-05 16:13:11 -04002574 ret = wret;
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002575 goto out;
2576 }
Chris Masone02119d2008-09-05 16:13:11 -04002577
2578 wret = walk_up_log_tree(trans, log, path, &level, wc);
2579 if (wret > 0)
2580 break;
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002581 if (wret < 0) {
Chris Masone02119d2008-09-05 16:13:11 -04002582 ret = wret;
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002583 goto out;
2584 }
Chris Masone02119d2008-09-05 16:13:11 -04002585 }
2586
2587 /* was the root node processed? if not, catch it here */
2588 if (path->nodes[orig_level]) {
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002589 ret = wc->process_func(log, path->nodes[orig_level], wc,
Chris Masone02119d2008-09-05 16:13:11 -04002590 btrfs_header_generation(path->nodes[orig_level]));
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002591 if (ret)
2592 goto out;
Chris Masone02119d2008-09-05 16:13:11 -04002593 if (wc->free) {
2594 struct extent_buffer *next;
2595
2596 next = path->nodes[orig_level];
2597
Josef Bacik681ae502013-10-07 15:11:00 -04002598 if (trans) {
2599 btrfs_tree_lock(next);
2600 btrfs_set_lock_blocking(next);
Daniel Dressler01d58472014-11-21 17:15:07 +09002601 clean_tree_block(trans, log->fs_info, next);
Josef Bacik681ae502013-10-07 15:11:00 -04002602 btrfs_wait_tree_block_writeback(next);
2603 btrfs_tree_unlock(next);
2604 }
Chris Masone02119d2008-09-05 16:13:11 -04002605
Chris Masone02119d2008-09-05 16:13:11 -04002606 WARN_ON(log->root_key.objectid !=
2607 BTRFS_TREE_LOG_OBJECTID);
Chris Masone688b7252011-10-31 20:52:39 -04002608 ret = btrfs_free_and_pin_reserved_extent(log, next->start,
Chris Masond00aff02008-09-11 15:54:42 -04002609 next->len);
Josef Bacik36508602013-04-25 16:23:32 -04002610 if (ret)
2611 goto out;
Chris Masone02119d2008-09-05 16:13:11 -04002612 }
2613 }
2614
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002615out:
Chris Masone02119d2008-09-05 16:13:11 -04002616 btrfs_free_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04002617 return ret;
2618}
2619
Yan Zheng7237f1832009-01-21 12:54:03 -05002620/*
2621 * helper function to update the item for a given subvolumes log root
2622 * in the tree of log roots
2623 */
2624static int update_log_root(struct btrfs_trans_handle *trans,
2625 struct btrfs_root *log)
2626{
2627 int ret;
2628
2629 if (log->log_transid == 1) {
2630 /* insert root item on the first sync */
2631 ret = btrfs_insert_root(trans, log->fs_info->log_root_tree,
2632 &log->root_key, &log->root_item);
2633 } else {
2634 ret = btrfs_update_root(trans, log->fs_info->log_root_tree,
2635 &log->root_key, &log->root_item);
2636 }
2637 return ret;
2638}
2639
Zhaolei60d53eb2015-08-17 18:44:46 +08002640static void wait_log_commit(struct btrfs_root *root, int transid)
Chris Masone02119d2008-09-05 16:13:11 -04002641{
2642 DEFINE_WAIT(wait);
Yan Zheng7237f1832009-01-21 12:54:03 -05002643 int index = transid % 2;
Chris Masone02119d2008-09-05 16:13:11 -04002644
Yan Zheng7237f1832009-01-21 12:54:03 -05002645 /*
2646 * we only allow two pending log transactions at a time,
2647 * so we know that if ours is more than 2 older than the
2648 * current transaction, we're done
2649 */
Chris Masone02119d2008-09-05 16:13:11 -04002650 do {
Yan Zheng7237f1832009-01-21 12:54:03 -05002651 prepare_to_wait(&root->log_commit_wait[index],
2652 &wait, TASK_UNINTERRUPTIBLE);
2653 mutex_unlock(&root->log_mutex);
Chris Mason12fcfd22009-03-24 10:24:20 -04002654
Miao Xied1433de2014-02-20 18:08:59 +08002655 if (root->log_transid_committed < transid &&
Yan Zheng7237f1832009-01-21 12:54:03 -05002656 atomic_read(&root->log_commit[index]))
Chris Masone02119d2008-09-05 16:13:11 -04002657 schedule();
Chris Mason12fcfd22009-03-24 10:24:20 -04002658
Yan Zheng7237f1832009-01-21 12:54:03 -05002659 finish_wait(&root->log_commit_wait[index], &wait);
2660 mutex_lock(&root->log_mutex);
Miao Xied1433de2014-02-20 18:08:59 +08002661 } while (root->log_transid_committed < transid &&
Yan Zheng7237f1832009-01-21 12:54:03 -05002662 atomic_read(&root->log_commit[index]));
Yan Zheng7237f1832009-01-21 12:54:03 -05002663}
2664
Zhaolei60d53eb2015-08-17 18:44:46 +08002665static void wait_for_writer(struct btrfs_root *root)
Yan Zheng7237f1832009-01-21 12:54:03 -05002666{
2667 DEFINE_WAIT(wait);
Miao Xie8b050d32014-02-20 18:08:58 +08002668
2669 while (atomic_read(&root->log_writers)) {
Yan Zheng7237f1832009-01-21 12:54:03 -05002670 prepare_to_wait(&root->log_writer_wait,
2671 &wait, TASK_UNINTERRUPTIBLE);
2672 mutex_unlock(&root->log_mutex);
Miao Xie8b050d32014-02-20 18:08:58 +08002673 if (atomic_read(&root->log_writers))
Yan Zheng7237f1832009-01-21 12:54:03 -05002674 schedule();
Yan Zheng7237f1832009-01-21 12:54:03 -05002675 finish_wait(&root->log_writer_wait, &wait);
Filipe Manana575849e2015-02-11 11:12:39 +00002676 mutex_lock(&root->log_mutex);
Yan Zheng7237f1832009-01-21 12:54:03 -05002677 }
Chris Masone02119d2008-09-05 16:13:11 -04002678}
2679
Miao Xie8b050d32014-02-20 18:08:58 +08002680static inline void btrfs_remove_log_ctx(struct btrfs_root *root,
2681 struct btrfs_log_ctx *ctx)
2682{
2683 if (!ctx)
2684 return;
2685
2686 mutex_lock(&root->log_mutex);
2687 list_del_init(&ctx->list);
2688 mutex_unlock(&root->log_mutex);
2689}
2690
2691/*
2692 * Invoked in log mutex context, or be sure there is no other task which
2693 * can access the list.
2694 */
2695static inline void btrfs_remove_all_log_ctxs(struct btrfs_root *root,
2696 int index, int error)
2697{
2698 struct btrfs_log_ctx *ctx;
2699
2700 if (!error) {
2701 INIT_LIST_HEAD(&root->log_ctxs[index]);
2702 return;
2703 }
2704
2705 list_for_each_entry(ctx, &root->log_ctxs[index], list)
2706 ctx->log_ret = error;
2707
2708 INIT_LIST_HEAD(&root->log_ctxs[index]);
2709}
2710
Chris Masone02119d2008-09-05 16:13:11 -04002711/*
2712 * btrfs_sync_log does sends a given tree log down to the disk and
2713 * updates the super blocks to record it. When this call is done,
Chris Mason12fcfd22009-03-24 10:24:20 -04002714 * you know that any inodes previously logged are safely on disk only
2715 * if it returns 0.
2716 *
2717 * Any other return value means you need to call btrfs_commit_transaction.
2718 * Some of the edge cases for fsyncing directories that have had unlinks
2719 * or renames done in the past mean that sometimes the only safe
2720 * fsync is to commit the whole FS. When btrfs_sync_log returns -EAGAIN,
2721 * that has happened.
Chris Masone02119d2008-09-05 16:13:11 -04002722 */
2723int btrfs_sync_log(struct btrfs_trans_handle *trans,
Miao Xie8b050d32014-02-20 18:08:58 +08002724 struct btrfs_root *root, struct btrfs_log_ctx *ctx)
Chris Masone02119d2008-09-05 16:13:11 -04002725{
Yan Zheng7237f1832009-01-21 12:54:03 -05002726 int index1;
2727 int index2;
Yan, Zheng8cef4e12009-11-12 09:33:26 +00002728 int mark;
Chris Masone02119d2008-09-05 16:13:11 -04002729 int ret;
Chris Masone02119d2008-09-05 16:13:11 -04002730 struct btrfs_root *log = root->log_root;
Yan Zheng7237f1832009-01-21 12:54:03 -05002731 struct btrfs_root *log_root_tree = root->fs_info->log_root_tree;
Miao Xiebb14a592014-02-20 18:08:56 +08002732 int log_transid = 0;
Miao Xie8b050d32014-02-20 18:08:58 +08002733 struct btrfs_log_ctx root_log_ctx;
Miao Xiec6adc9c2013-05-28 10:05:39 +00002734 struct blk_plug plug;
Chris Masone02119d2008-09-05 16:13:11 -04002735
Yan Zheng7237f1832009-01-21 12:54:03 -05002736 mutex_lock(&root->log_mutex);
Miao Xied1433de2014-02-20 18:08:59 +08002737 log_transid = ctx->log_transid;
2738 if (root->log_transid_committed >= log_transid) {
Yan Zheng7237f1832009-01-21 12:54:03 -05002739 mutex_unlock(&root->log_mutex);
Miao Xie8b050d32014-02-20 18:08:58 +08002740 return ctx->log_ret;
Chris Masone02119d2008-09-05 16:13:11 -04002741 }
Miao Xied1433de2014-02-20 18:08:59 +08002742
2743 index1 = log_transid % 2;
2744 if (atomic_read(&root->log_commit[index1])) {
Zhaolei60d53eb2015-08-17 18:44:46 +08002745 wait_log_commit(root, log_transid);
Miao Xied1433de2014-02-20 18:08:59 +08002746 mutex_unlock(&root->log_mutex);
2747 return ctx->log_ret;
2748 }
2749 ASSERT(log_transid == root->log_transid);
Yan Zheng7237f1832009-01-21 12:54:03 -05002750 atomic_set(&root->log_commit[index1], 1);
2751
2752 /* wait for previous tree log sync to complete */
2753 if (atomic_read(&root->log_commit[(index1 + 1) % 2]))
Zhaolei60d53eb2015-08-17 18:44:46 +08002754 wait_log_commit(root, log_transid - 1);
Miao Xie48cab2e2014-02-20 18:08:52 +08002755
Yan, Zheng86df7eb2009-10-14 09:24:59 -04002756 while (1) {
Miao Xie2ecb7922012-09-06 04:04:27 -06002757 int batch = atomic_read(&root->log_batch);
Chris Masoncd354ad2011-10-20 15:45:37 -04002758 /* when we're on an ssd, just kick the log commit out */
Miao Xie27cdeb72014-04-02 19:51:05 +08002759 if (!btrfs_test_opt(root, SSD) &&
2760 test_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state)) {
Yan, Zheng86df7eb2009-10-14 09:24:59 -04002761 mutex_unlock(&root->log_mutex);
2762 schedule_timeout_uninterruptible(1);
2763 mutex_lock(&root->log_mutex);
2764 }
Zhaolei60d53eb2015-08-17 18:44:46 +08002765 wait_for_writer(root);
Miao Xie2ecb7922012-09-06 04:04:27 -06002766 if (batch == atomic_read(&root->log_batch))
Chris Masone02119d2008-09-05 16:13:11 -04002767 break;
2768 }
Chris Masond0c803c2008-09-11 16:17:57 -04002769
Chris Mason12fcfd22009-03-24 10:24:20 -04002770 /* bail out if we need to do a full commit */
Miao Xie995946d2014-04-02 19:51:06 +08002771 if (btrfs_need_log_full_commit(root->fs_info, trans)) {
Chris Mason12fcfd22009-03-24 10:24:20 -04002772 ret = -EAGAIN;
Josef Bacik2ab28f32012-10-12 15:27:49 -04002773 btrfs_free_logged_extents(log, log_transid);
Chris Mason12fcfd22009-03-24 10:24:20 -04002774 mutex_unlock(&root->log_mutex);
2775 goto out;
2776 }
2777
Yan, Zheng8cef4e12009-11-12 09:33:26 +00002778 if (log_transid % 2 == 0)
2779 mark = EXTENT_DIRTY;
2780 else
2781 mark = EXTENT_NEW;
2782
Chris Mason690587d2009-10-13 13:29:19 -04002783 /* we start IO on all the marked extents here, but we don't actually
2784 * wait for them until later.
2785 */
Miao Xiec6adc9c2013-05-28 10:05:39 +00002786 blk_start_plug(&plug);
Yan, Zheng8cef4e12009-11-12 09:33:26 +00002787 ret = btrfs_write_marked_extents(log, &log->dirty_log_pages, mark);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002788 if (ret) {
Miao Xiec6adc9c2013-05-28 10:05:39 +00002789 blk_finish_plug(&plug);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002790 btrfs_abort_transaction(trans, root, ret);
Josef Bacik2ab28f32012-10-12 15:27:49 -04002791 btrfs_free_logged_extents(log, log_transid);
Miao Xie995946d2014-04-02 19:51:06 +08002792 btrfs_set_log_full_commit(root->fs_info, trans);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002793 mutex_unlock(&root->log_mutex);
2794 goto out;
2795 }
Yan Zheng7237f1832009-01-21 12:54:03 -05002796
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002797 btrfs_set_root_node(&log->root_item, log->node);
Yan Zheng7237f1832009-01-21 12:54:03 -05002798
Yan Zheng7237f1832009-01-21 12:54:03 -05002799 root->log_transid++;
2800 log->log_transid = root->log_transid;
Josef Bacikff782e02009-10-08 15:30:04 -04002801 root->log_start_pid = 0;
Yan Zheng7237f1832009-01-21 12:54:03 -05002802 /*
Yan, Zheng8cef4e12009-11-12 09:33:26 +00002803 * IO has been started, blocks of the log tree have WRITTEN flag set
2804 * in their headers. new modifications of the log will be written to
2805 * new positions. so it's safe to allow log writers to go in.
Yan Zheng7237f1832009-01-21 12:54:03 -05002806 */
2807 mutex_unlock(&root->log_mutex);
2808
Miao Xied1433de2014-02-20 18:08:59 +08002809 btrfs_init_log_ctx(&root_log_ctx);
2810
Yan Zheng7237f1832009-01-21 12:54:03 -05002811 mutex_lock(&log_root_tree->log_mutex);
Miao Xie2ecb7922012-09-06 04:04:27 -06002812 atomic_inc(&log_root_tree->log_batch);
Yan Zheng7237f1832009-01-21 12:54:03 -05002813 atomic_inc(&log_root_tree->log_writers);
Miao Xied1433de2014-02-20 18:08:59 +08002814
2815 index2 = log_root_tree->log_transid % 2;
2816 list_add_tail(&root_log_ctx.list, &log_root_tree->log_ctxs[index2]);
2817 root_log_ctx.log_transid = log_root_tree->log_transid;
2818
Yan Zheng7237f1832009-01-21 12:54:03 -05002819 mutex_unlock(&log_root_tree->log_mutex);
2820
2821 ret = update_log_root(trans, log);
Yan Zheng7237f1832009-01-21 12:54:03 -05002822
2823 mutex_lock(&log_root_tree->log_mutex);
2824 if (atomic_dec_and_test(&log_root_tree->log_writers)) {
David Sterba779adf02015-02-16 19:39:00 +01002825 /*
2826 * Implicit memory barrier after atomic_dec_and_test
2827 */
Yan Zheng7237f1832009-01-21 12:54:03 -05002828 if (waitqueue_active(&log_root_tree->log_writer_wait))
2829 wake_up(&log_root_tree->log_writer_wait);
2830 }
2831
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002832 if (ret) {
Miao Xied1433de2014-02-20 18:08:59 +08002833 if (!list_empty(&root_log_ctx.list))
2834 list_del_init(&root_log_ctx.list);
2835
Miao Xiec6adc9c2013-05-28 10:05:39 +00002836 blk_finish_plug(&plug);
Miao Xie995946d2014-04-02 19:51:06 +08002837 btrfs_set_log_full_commit(root->fs_info, trans);
2838
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002839 if (ret != -ENOSPC) {
2840 btrfs_abort_transaction(trans, root, ret);
2841 mutex_unlock(&log_root_tree->log_mutex);
2842 goto out;
2843 }
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002844 btrfs_wait_marked_extents(log, &log->dirty_log_pages, mark);
Josef Bacik2ab28f32012-10-12 15:27:49 -04002845 btrfs_free_logged_extents(log, log_transid);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002846 mutex_unlock(&log_root_tree->log_mutex);
2847 ret = -EAGAIN;
2848 goto out;
2849 }
2850
Miao Xied1433de2014-02-20 18:08:59 +08002851 if (log_root_tree->log_transid_committed >= root_log_ctx.log_transid) {
Forrest Liu3da5ab52015-01-30 19:42:12 +08002852 blk_finish_plug(&plug);
Chris Mason5e0286e2016-09-06 05:37:40 -07002853 list_del_init(&root_log_ctx.list);
Miao Xied1433de2014-02-20 18:08:59 +08002854 mutex_unlock(&log_root_tree->log_mutex);
2855 ret = root_log_ctx.log_ret;
2856 goto out;
2857 }
Miao Xie8b050d32014-02-20 18:08:58 +08002858
Miao Xied1433de2014-02-20 18:08:59 +08002859 index2 = root_log_ctx.log_transid % 2;
Yan Zheng7237f1832009-01-21 12:54:03 -05002860 if (atomic_read(&log_root_tree->log_commit[index2])) {
Miao Xiec6adc9c2013-05-28 10:05:39 +00002861 blk_finish_plug(&plug);
Filipe Manana5ab5e442014-11-13 16:59:53 +00002862 ret = btrfs_wait_marked_extents(log, &log->dirty_log_pages,
2863 mark);
Josef Bacik50d9aa92014-11-21 14:52:38 -05002864 btrfs_wait_logged_extents(trans, log, log_transid);
Zhaolei60d53eb2015-08-17 18:44:46 +08002865 wait_log_commit(log_root_tree,
Miao Xied1433de2014-02-20 18:08:59 +08002866 root_log_ctx.log_transid);
Yan Zheng7237f1832009-01-21 12:54:03 -05002867 mutex_unlock(&log_root_tree->log_mutex);
Filipe Manana5ab5e442014-11-13 16:59:53 +00002868 if (!ret)
2869 ret = root_log_ctx.log_ret;
Yan Zheng7237f1832009-01-21 12:54:03 -05002870 goto out;
2871 }
Miao Xied1433de2014-02-20 18:08:59 +08002872 ASSERT(root_log_ctx.log_transid == log_root_tree->log_transid);
Yan Zheng7237f1832009-01-21 12:54:03 -05002873 atomic_set(&log_root_tree->log_commit[index2], 1);
2874
Chris Mason12fcfd22009-03-24 10:24:20 -04002875 if (atomic_read(&log_root_tree->log_commit[(index2 + 1) % 2])) {
Zhaolei60d53eb2015-08-17 18:44:46 +08002876 wait_log_commit(log_root_tree,
Miao Xied1433de2014-02-20 18:08:59 +08002877 root_log_ctx.log_transid - 1);
Chris Mason12fcfd22009-03-24 10:24:20 -04002878 }
Yan Zheng7237f1832009-01-21 12:54:03 -05002879
Zhaolei60d53eb2015-08-17 18:44:46 +08002880 wait_for_writer(log_root_tree);
Chris Mason12fcfd22009-03-24 10:24:20 -04002881
2882 /*
2883 * now that we've moved on to the tree of log tree roots,
2884 * check the full commit flag again
2885 */
Miao Xie995946d2014-04-02 19:51:06 +08002886 if (btrfs_need_log_full_commit(root->fs_info, trans)) {
Miao Xiec6adc9c2013-05-28 10:05:39 +00002887 blk_finish_plug(&plug);
Yan, Zheng8cef4e12009-11-12 09:33:26 +00002888 btrfs_wait_marked_extents(log, &log->dirty_log_pages, mark);
Josef Bacik2ab28f32012-10-12 15:27:49 -04002889 btrfs_free_logged_extents(log, log_transid);
Chris Mason12fcfd22009-03-24 10:24:20 -04002890 mutex_unlock(&log_root_tree->log_mutex);
2891 ret = -EAGAIN;
2892 goto out_wake_log_root;
2893 }
Yan Zheng7237f1832009-01-21 12:54:03 -05002894
Miao Xiec6adc9c2013-05-28 10:05:39 +00002895 ret = btrfs_write_marked_extents(log_root_tree,
2896 &log_root_tree->dirty_log_pages,
2897 EXTENT_DIRTY | EXTENT_NEW);
2898 blk_finish_plug(&plug);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002899 if (ret) {
Miao Xie995946d2014-04-02 19:51:06 +08002900 btrfs_set_log_full_commit(root->fs_info, trans);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002901 btrfs_abort_transaction(trans, root, ret);
Josef Bacik2ab28f32012-10-12 15:27:49 -04002902 btrfs_free_logged_extents(log, log_transid);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002903 mutex_unlock(&log_root_tree->log_mutex);
2904 goto out_wake_log_root;
2905 }
Filipe Manana5ab5e442014-11-13 16:59:53 +00002906 ret = btrfs_wait_marked_extents(log, &log->dirty_log_pages, mark);
2907 if (!ret)
2908 ret = btrfs_wait_marked_extents(log_root_tree,
2909 &log_root_tree->dirty_log_pages,
2910 EXTENT_NEW | EXTENT_DIRTY);
2911 if (ret) {
2912 btrfs_set_log_full_commit(root->fs_info, trans);
2913 btrfs_free_logged_extents(log, log_transid);
2914 mutex_unlock(&log_root_tree->log_mutex);
2915 goto out_wake_log_root;
2916 }
Josef Bacik50d9aa92014-11-21 14:52:38 -05002917 btrfs_wait_logged_extents(trans, log, log_transid);
Chris Masone02119d2008-09-05 16:13:11 -04002918
David Sterba6c417612011-04-13 15:41:04 +02002919 btrfs_set_super_log_root(root->fs_info->super_for_commit,
Yan Zheng7237f1832009-01-21 12:54:03 -05002920 log_root_tree->node->start);
David Sterba6c417612011-04-13 15:41:04 +02002921 btrfs_set_super_log_root_level(root->fs_info->super_for_commit,
Yan Zheng7237f1832009-01-21 12:54:03 -05002922 btrfs_header_level(log_root_tree->node));
Chris Masone02119d2008-09-05 16:13:11 -04002923
Yan Zheng7237f1832009-01-21 12:54:03 -05002924 log_root_tree->log_transid++;
Yan Zheng7237f1832009-01-21 12:54:03 -05002925 mutex_unlock(&log_root_tree->log_mutex);
2926
2927 /*
2928 * nobody else is going to jump in and write the the ctree
2929 * super here because the log_commit atomic below is protecting
2930 * us. We must be called with a transaction handle pinning
2931 * the running transaction open, so a full commit can't hop
2932 * in and cause problems either.
2933 */
Stefan Behrens5af3e8c2012-08-01 18:56:49 +02002934 ret = write_ctree_super(trans, root->fs_info->tree_root, 1);
Stefan Behrens5af3e8c2012-08-01 18:56:49 +02002935 if (ret) {
Miao Xie995946d2014-04-02 19:51:06 +08002936 btrfs_set_log_full_commit(root->fs_info, trans);
Stefan Behrens5af3e8c2012-08-01 18:56:49 +02002937 btrfs_abort_transaction(trans, root, ret);
2938 goto out_wake_log_root;
2939 }
Yan Zheng7237f1832009-01-21 12:54:03 -05002940
Chris Mason257c62e2009-10-13 13:21:08 -04002941 mutex_lock(&root->log_mutex);
2942 if (root->last_log_commit < log_transid)
2943 root->last_log_commit = log_transid;
2944 mutex_unlock(&root->log_mutex);
2945
Chris Mason12fcfd22009-03-24 10:24:20 -04002946out_wake_log_root:
Miao Xie8b050d32014-02-20 18:08:58 +08002947 /*
2948 * We needn't get log_mutex here because we are sure all
2949 * the other tasks are blocked.
2950 */
2951 btrfs_remove_all_log_ctxs(log_root_tree, index2, ret);
2952
Miao Xied1433de2014-02-20 18:08:59 +08002953 mutex_lock(&log_root_tree->log_mutex);
2954 log_root_tree->log_transid_committed++;
Yan Zheng7237f1832009-01-21 12:54:03 -05002955 atomic_set(&log_root_tree->log_commit[index2], 0);
Miao Xied1433de2014-02-20 18:08:59 +08002956 mutex_unlock(&log_root_tree->log_mutex);
2957
David Sterba33a9eca2015-10-10 18:35:10 +02002958 /*
2959 * The barrier before waitqueue_active is implied by mutex_unlock
2960 */
Yan Zheng7237f1832009-01-21 12:54:03 -05002961 if (waitqueue_active(&log_root_tree->log_commit_wait[index2]))
2962 wake_up(&log_root_tree->log_commit_wait[index2]);
Chris Masone02119d2008-09-05 16:13:11 -04002963out:
Miao Xie8b050d32014-02-20 18:08:58 +08002964 /* See above. */
2965 btrfs_remove_all_log_ctxs(root, index1, ret);
2966
Miao Xied1433de2014-02-20 18:08:59 +08002967 mutex_lock(&root->log_mutex);
2968 root->log_transid_committed++;
Yan Zheng7237f1832009-01-21 12:54:03 -05002969 atomic_set(&root->log_commit[index1], 0);
Miao Xied1433de2014-02-20 18:08:59 +08002970 mutex_unlock(&root->log_mutex);
Miao Xie8b050d32014-02-20 18:08:58 +08002971
David Sterba33a9eca2015-10-10 18:35:10 +02002972 /*
2973 * The barrier before waitqueue_active is implied by mutex_unlock
2974 */
Yan Zheng7237f1832009-01-21 12:54:03 -05002975 if (waitqueue_active(&root->log_commit_wait[index1]))
2976 wake_up(&root->log_commit_wait[index1]);
Chris Masonb31eabd2011-01-31 16:48:24 -05002977 return ret;
Chris Masone02119d2008-09-05 16:13:11 -04002978}
2979
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002980static void free_log_tree(struct btrfs_trans_handle *trans,
2981 struct btrfs_root *log)
Chris Masone02119d2008-09-05 16:13:11 -04002982{
2983 int ret;
Chris Masond0c803c2008-09-11 16:17:57 -04002984 u64 start;
2985 u64 end;
Chris Masone02119d2008-09-05 16:13:11 -04002986 struct walk_control wc = {
2987 .free = 1,
2988 .process_func = process_one_buffer
2989 };
2990
Josef Bacik681ae502013-10-07 15:11:00 -04002991 ret = walk_log_tree(trans, log, &wc);
2992 /* I don't think this can happen but just in case */
2993 if (ret)
2994 btrfs_abort_transaction(trans, log, ret);
Chris Masone02119d2008-09-05 16:13:11 -04002995
Chris Masond3977122009-01-05 21:25:51 -05002996 while (1) {
Chris Masond0c803c2008-09-11 16:17:57 -04002997 ret = find_first_extent_bit(&log->dirty_log_pages,
Josef Bacike6138872012-09-27 17:07:30 -04002998 0, &start, &end, EXTENT_DIRTY | EXTENT_NEW,
2999 NULL);
Chris Masond0c803c2008-09-11 16:17:57 -04003000 if (ret)
3001 break;
3002
Yan, Zheng8cef4e12009-11-12 09:33:26 +00003003 clear_extent_bits(&log->dirty_log_pages, start, end,
3004 EXTENT_DIRTY | EXTENT_NEW, GFP_NOFS);
Chris Masond0c803c2008-09-11 16:17:57 -04003005 }
3006
Josef Bacik2ab28f32012-10-12 15:27:49 -04003007 /*
3008 * We may have short-circuited the log tree with the full commit logic
3009 * and left ordered extents on our list, so clear these out to keep us
3010 * from leaking inodes and memory.
3011 */
3012 btrfs_free_logged_extents(log, 0);
3013 btrfs_free_logged_extents(log, 1);
3014
Yan Zheng7237f1832009-01-21 12:54:03 -05003015 free_extent_buffer(log->node);
3016 kfree(log);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04003017}
3018
3019/*
3020 * free all the extents used by the tree log. This should be called
3021 * at commit time of the full transaction
3022 */
3023int btrfs_free_log(struct btrfs_trans_handle *trans, struct btrfs_root *root)
3024{
3025 if (root->log_root) {
3026 free_log_tree(trans, root->log_root);
3027 root->log_root = NULL;
3028 }
3029 return 0;
3030}
3031
3032int btrfs_free_log_root_tree(struct btrfs_trans_handle *trans,
3033 struct btrfs_fs_info *fs_info)
3034{
3035 if (fs_info->log_root_tree) {
3036 free_log_tree(trans, fs_info->log_root_tree);
3037 fs_info->log_root_tree = NULL;
3038 }
Chris Masone02119d2008-09-05 16:13:11 -04003039 return 0;
3040}
3041
3042/*
Chris Masone02119d2008-09-05 16:13:11 -04003043 * If both a file and directory are logged, and unlinks or renames are
3044 * mixed in, we have a few interesting corners:
3045 *
3046 * create file X in dir Y
3047 * link file X to X.link in dir Y
3048 * fsync file X
3049 * unlink file X but leave X.link
3050 * fsync dir Y
3051 *
3052 * After a crash we would expect only X.link to exist. But file X
3053 * didn't get fsync'd again so the log has back refs for X and X.link.
3054 *
3055 * We solve this by removing directory entries and inode backrefs from the
3056 * log when a file that was logged in the current transaction is
3057 * unlinked. Any later fsync will include the updated log entries, and
3058 * we'll be able to reconstruct the proper directory items from backrefs.
3059 *
3060 * This optimizations allows us to avoid relogging the entire inode
3061 * or the entire directory.
3062 */
3063int btrfs_del_dir_entries_in_log(struct btrfs_trans_handle *trans,
3064 struct btrfs_root *root,
3065 const char *name, int name_len,
3066 struct inode *dir, u64 index)
3067{
3068 struct btrfs_root *log;
3069 struct btrfs_dir_item *di;
3070 struct btrfs_path *path;
3071 int ret;
Yan, Zheng4a500fd2010-05-16 10:49:59 -04003072 int err = 0;
Chris Masone02119d2008-09-05 16:13:11 -04003073 int bytes_del = 0;
Li Zefan33345d012011-04-20 10:31:50 +08003074 u64 dir_ino = btrfs_ino(dir);
Chris Masone02119d2008-09-05 16:13:11 -04003075
Chris Mason3a5f1d42008-09-11 15:53:37 -04003076 if (BTRFS_I(dir)->logged_trans < trans->transid)
3077 return 0;
3078
Chris Masone02119d2008-09-05 16:13:11 -04003079 ret = join_running_log_trans(root);
3080 if (ret)
3081 return 0;
3082
3083 mutex_lock(&BTRFS_I(dir)->log_mutex);
3084
3085 log = root->log_root;
3086 path = btrfs_alloc_path();
Tsutomu Itoha62f44a2011-04-25 19:43:51 -04003087 if (!path) {
3088 err = -ENOMEM;
3089 goto out_unlock;
3090 }
liubo2a29edc2011-01-26 06:22:08 +00003091
Li Zefan33345d012011-04-20 10:31:50 +08003092 di = btrfs_lookup_dir_item(trans, log, path, dir_ino,
Chris Masone02119d2008-09-05 16:13:11 -04003093 name, name_len, -1);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04003094 if (IS_ERR(di)) {
3095 err = PTR_ERR(di);
3096 goto fail;
3097 }
3098 if (di) {
Chris Masone02119d2008-09-05 16:13:11 -04003099 ret = btrfs_delete_one_dir_name(trans, log, path, di);
3100 bytes_del += name_len;
Josef Bacik36508602013-04-25 16:23:32 -04003101 if (ret) {
3102 err = ret;
3103 goto fail;
3104 }
Chris Masone02119d2008-09-05 16:13:11 -04003105 }
David Sterbab3b4aa72011-04-21 01:20:15 +02003106 btrfs_release_path(path);
Li Zefan33345d012011-04-20 10:31:50 +08003107 di = btrfs_lookup_dir_index_item(trans, log, path, dir_ino,
Chris Masone02119d2008-09-05 16:13:11 -04003108 index, name, name_len, -1);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04003109 if (IS_ERR(di)) {
3110 err = PTR_ERR(di);
3111 goto fail;
3112 }
3113 if (di) {
Chris Masone02119d2008-09-05 16:13:11 -04003114 ret = btrfs_delete_one_dir_name(trans, log, path, di);
3115 bytes_del += name_len;
Josef Bacik36508602013-04-25 16:23:32 -04003116 if (ret) {
3117 err = ret;
3118 goto fail;
3119 }
Chris Masone02119d2008-09-05 16:13:11 -04003120 }
3121
3122 /* update the directory size in the log to reflect the names
3123 * we have removed
3124 */
3125 if (bytes_del) {
3126 struct btrfs_key key;
3127
Li Zefan33345d012011-04-20 10:31:50 +08003128 key.objectid = dir_ino;
Chris Masone02119d2008-09-05 16:13:11 -04003129 key.offset = 0;
3130 key.type = BTRFS_INODE_ITEM_KEY;
David Sterbab3b4aa72011-04-21 01:20:15 +02003131 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04003132
3133 ret = btrfs_search_slot(trans, log, &key, path, 0, 1);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04003134 if (ret < 0) {
3135 err = ret;
3136 goto fail;
3137 }
Chris Masone02119d2008-09-05 16:13:11 -04003138 if (ret == 0) {
3139 struct btrfs_inode_item *item;
3140 u64 i_size;
3141
3142 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3143 struct btrfs_inode_item);
3144 i_size = btrfs_inode_size(path->nodes[0], item);
3145 if (i_size > bytes_del)
3146 i_size -= bytes_del;
3147 else
3148 i_size = 0;
3149 btrfs_set_inode_size(path->nodes[0], item, i_size);
3150 btrfs_mark_buffer_dirty(path->nodes[0]);
3151 } else
3152 ret = 0;
David Sterbab3b4aa72011-04-21 01:20:15 +02003153 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04003154 }
Yan, Zheng4a500fd2010-05-16 10:49:59 -04003155fail:
Chris Masone02119d2008-09-05 16:13:11 -04003156 btrfs_free_path(path);
Tsutomu Itoha62f44a2011-04-25 19:43:51 -04003157out_unlock:
Chris Masone02119d2008-09-05 16:13:11 -04003158 mutex_unlock(&BTRFS_I(dir)->log_mutex);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04003159 if (ret == -ENOSPC) {
Miao Xie995946d2014-04-02 19:51:06 +08003160 btrfs_set_log_full_commit(root->fs_info, trans);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04003161 ret = 0;
Jeff Mahoney79787ea2012-03-12 16:03:00 +01003162 } else if (ret < 0)
3163 btrfs_abort_transaction(trans, root, ret);
3164
Chris Mason12fcfd22009-03-24 10:24:20 -04003165 btrfs_end_log_trans(root);
Chris Masone02119d2008-09-05 16:13:11 -04003166
Andi Kleen411fc6b2010-10-29 15:14:31 -04003167 return err;
Chris Masone02119d2008-09-05 16:13:11 -04003168}
3169
3170/* see comments for btrfs_del_dir_entries_in_log */
3171int btrfs_del_inode_ref_in_log(struct btrfs_trans_handle *trans,
3172 struct btrfs_root *root,
3173 const char *name, int name_len,
3174 struct inode *inode, u64 dirid)
3175{
3176 struct btrfs_root *log;
3177 u64 index;
3178 int ret;
3179
Chris Mason3a5f1d42008-09-11 15:53:37 -04003180 if (BTRFS_I(inode)->logged_trans < trans->transid)
3181 return 0;
3182
Chris Masone02119d2008-09-05 16:13:11 -04003183 ret = join_running_log_trans(root);
3184 if (ret)
3185 return 0;
3186 log = root->log_root;
3187 mutex_lock(&BTRFS_I(inode)->log_mutex);
3188
Li Zefan33345d012011-04-20 10:31:50 +08003189 ret = btrfs_del_inode_ref(trans, log, name, name_len, btrfs_ino(inode),
Chris Masone02119d2008-09-05 16:13:11 -04003190 dirid, &index);
3191 mutex_unlock(&BTRFS_I(inode)->log_mutex);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04003192 if (ret == -ENOSPC) {
Miao Xie995946d2014-04-02 19:51:06 +08003193 btrfs_set_log_full_commit(root->fs_info, trans);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04003194 ret = 0;
Jeff Mahoney79787ea2012-03-12 16:03:00 +01003195 } else if (ret < 0 && ret != -ENOENT)
3196 btrfs_abort_transaction(trans, root, ret);
Chris Mason12fcfd22009-03-24 10:24:20 -04003197 btrfs_end_log_trans(root);
Chris Masone02119d2008-09-05 16:13:11 -04003198
Chris Masone02119d2008-09-05 16:13:11 -04003199 return ret;
3200}
3201
3202/*
3203 * creates a range item in the log for 'dirid'. first_offset and
3204 * last_offset tell us which parts of the key space the log should
3205 * be considered authoritative for.
3206 */
3207static noinline int insert_dir_log_key(struct btrfs_trans_handle *trans,
3208 struct btrfs_root *log,
3209 struct btrfs_path *path,
3210 int key_type, u64 dirid,
3211 u64 first_offset, u64 last_offset)
3212{
3213 int ret;
3214 struct btrfs_key key;
3215 struct btrfs_dir_log_item *item;
3216
3217 key.objectid = dirid;
3218 key.offset = first_offset;
3219 if (key_type == BTRFS_DIR_ITEM_KEY)
3220 key.type = BTRFS_DIR_LOG_ITEM_KEY;
3221 else
3222 key.type = BTRFS_DIR_LOG_INDEX_KEY;
3223 ret = btrfs_insert_empty_item(trans, log, path, &key, sizeof(*item));
Yan, Zheng4a500fd2010-05-16 10:49:59 -04003224 if (ret)
3225 return ret;
Chris Masone02119d2008-09-05 16:13:11 -04003226
3227 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3228 struct btrfs_dir_log_item);
3229 btrfs_set_dir_log_end(path->nodes[0], item, last_offset);
3230 btrfs_mark_buffer_dirty(path->nodes[0]);
David Sterbab3b4aa72011-04-21 01:20:15 +02003231 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04003232 return 0;
3233}
3234
3235/*
3236 * log all the items included in the current transaction for a given
3237 * directory. This also creates the range items in the log tree required
3238 * to replay anything deleted before the fsync
3239 */
3240static noinline int log_dir_items(struct btrfs_trans_handle *trans,
3241 struct btrfs_root *root, struct inode *inode,
3242 struct btrfs_path *path,
3243 struct btrfs_path *dst_path, int key_type,
Filipe Manana2f2ff0e2015-03-20 17:19:46 +00003244 struct btrfs_log_ctx *ctx,
Chris Masone02119d2008-09-05 16:13:11 -04003245 u64 min_offset, u64 *last_offset_ret)
3246{
3247 struct btrfs_key min_key;
Chris Masone02119d2008-09-05 16:13:11 -04003248 struct btrfs_root *log = root->log_root;
3249 struct extent_buffer *src;
Yan, Zheng4a500fd2010-05-16 10:49:59 -04003250 int err = 0;
Chris Masone02119d2008-09-05 16:13:11 -04003251 int ret;
3252 int i;
3253 int nritems;
3254 u64 first_offset = min_offset;
3255 u64 last_offset = (u64)-1;
Li Zefan33345d012011-04-20 10:31:50 +08003256 u64 ino = btrfs_ino(inode);
Chris Masone02119d2008-09-05 16:13:11 -04003257
3258 log = root->log_root;
Chris Masone02119d2008-09-05 16:13:11 -04003259
Li Zefan33345d012011-04-20 10:31:50 +08003260 min_key.objectid = ino;
Chris Masone02119d2008-09-05 16:13:11 -04003261 min_key.type = key_type;
3262 min_key.offset = min_offset;
3263
Filipe David Borba Manana6174d3c2013-10-01 16:13:42 +01003264 ret = btrfs_search_forward(root, &min_key, path, trans->transid);
Chris Masone02119d2008-09-05 16:13:11 -04003265
3266 /*
3267 * we didn't find anything from this transaction, see if there
3268 * is anything at all
3269 */
Li Zefan33345d012011-04-20 10:31:50 +08003270 if (ret != 0 || min_key.objectid != ino || min_key.type != key_type) {
3271 min_key.objectid = ino;
Chris Masone02119d2008-09-05 16:13:11 -04003272 min_key.type = key_type;
3273 min_key.offset = (u64)-1;
David Sterbab3b4aa72011-04-21 01:20:15 +02003274 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04003275 ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0);
3276 if (ret < 0) {
David Sterbab3b4aa72011-04-21 01:20:15 +02003277 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04003278 return ret;
3279 }
Li Zefan33345d012011-04-20 10:31:50 +08003280 ret = btrfs_previous_item(root, path, ino, key_type);
Chris Masone02119d2008-09-05 16:13:11 -04003281
3282 /* if ret == 0 there are items for this type,
3283 * create a range to tell us the last key of this type.
3284 * otherwise, there are no items in this directory after
3285 * *min_offset, and we create a range to indicate that.
3286 */
3287 if (ret == 0) {
3288 struct btrfs_key tmp;
3289 btrfs_item_key_to_cpu(path->nodes[0], &tmp,
3290 path->slots[0]);
Chris Masond3977122009-01-05 21:25:51 -05003291 if (key_type == tmp.type)
Chris Masone02119d2008-09-05 16:13:11 -04003292 first_offset = max(min_offset, tmp.offset) + 1;
Chris Masone02119d2008-09-05 16:13:11 -04003293 }
3294 goto done;
3295 }
3296
3297 /* go backward to find any previous key */
Li Zefan33345d012011-04-20 10:31:50 +08003298 ret = btrfs_previous_item(root, path, ino, key_type);
Chris Masone02119d2008-09-05 16:13:11 -04003299 if (ret == 0) {
3300 struct btrfs_key tmp;
3301 btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]);
3302 if (key_type == tmp.type) {
3303 first_offset = tmp.offset;
3304 ret = overwrite_item(trans, log, dst_path,
3305 path->nodes[0], path->slots[0],
3306 &tmp);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04003307 if (ret) {
3308 err = ret;
3309 goto done;
3310 }
Chris Masone02119d2008-09-05 16:13:11 -04003311 }
3312 }
David Sterbab3b4aa72011-04-21 01:20:15 +02003313 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04003314
3315 /* find the first key from this transaction again */
3316 ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0);
Dulshani Gunawardhanafae7f212013-10-31 10:30:08 +05303317 if (WARN_ON(ret != 0))
Chris Masone02119d2008-09-05 16:13:11 -04003318 goto done;
Chris Masone02119d2008-09-05 16:13:11 -04003319
3320 /*
3321 * we have a block from this transaction, log every item in it
3322 * from our directory
3323 */
Chris Masond3977122009-01-05 21:25:51 -05003324 while (1) {
Chris Masone02119d2008-09-05 16:13:11 -04003325 struct btrfs_key tmp;
3326 src = path->nodes[0];
3327 nritems = btrfs_header_nritems(src);
3328 for (i = path->slots[0]; i < nritems; i++) {
Filipe Manana2f2ff0e2015-03-20 17:19:46 +00003329 struct btrfs_dir_item *di;
3330
Chris Masone02119d2008-09-05 16:13:11 -04003331 btrfs_item_key_to_cpu(src, &min_key, i);
3332
Li Zefan33345d012011-04-20 10:31:50 +08003333 if (min_key.objectid != ino || min_key.type != key_type)
Chris Masone02119d2008-09-05 16:13:11 -04003334 goto done;
3335 ret = overwrite_item(trans, log, dst_path, src, i,
3336 &min_key);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04003337 if (ret) {
3338 err = ret;
3339 goto done;
3340 }
Filipe Manana2f2ff0e2015-03-20 17:19:46 +00003341
3342 /*
3343 * We must make sure that when we log a directory entry,
3344 * the corresponding inode, after log replay, has a
3345 * matching link count. For example:
3346 *
3347 * touch foo
3348 * mkdir mydir
3349 * sync
3350 * ln foo mydir/bar
3351 * xfs_io -c "fsync" mydir
3352 * <crash>
3353 * <mount fs and log replay>
3354 *
3355 * Would result in a fsync log that when replayed, our
3356 * file inode would have a link count of 1, but we get
3357 * two directory entries pointing to the same inode.
3358 * After removing one of the names, it would not be
3359 * possible to remove the other name, which resulted
3360 * always in stale file handle errors, and would not
3361 * be possible to rmdir the parent directory, since
3362 * its i_size could never decrement to the value
3363 * BTRFS_EMPTY_DIR_SIZE, resulting in -ENOTEMPTY errors.
3364 */
3365 di = btrfs_item_ptr(src, i, struct btrfs_dir_item);
3366 btrfs_dir_item_key_to_cpu(src, di, &tmp);
3367 if (ctx &&
3368 (btrfs_dir_transid(src, di) == trans->transid ||
3369 btrfs_dir_type(src, di) == BTRFS_FT_DIR) &&
3370 tmp.type != BTRFS_ROOT_ITEM_KEY)
3371 ctx->log_new_dentries = true;
Chris Masone02119d2008-09-05 16:13:11 -04003372 }
3373 path->slots[0] = nritems;
3374
3375 /*
3376 * look ahead to the next item and see if it is also
3377 * from this directory and from this transaction
3378 */
3379 ret = btrfs_next_leaf(root, path);
3380 if (ret == 1) {
3381 last_offset = (u64)-1;
3382 goto done;
3383 }
3384 btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]);
Li Zefan33345d012011-04-20 10:31:50 +08003385 if (tmp.objectid != ino || tmp.type != key_type) {
Chris Masone02119d2008-09-05 16:13:11 -04003386 last_offset = (u64)-1;
3387 goto done;
3388 }
3389 if (btrfs_header_generation(path->nodes[0]) != trans->transid) {
3390 ret = overwrite_item(trans, log, dst_path,
3391 path->nodes[0], path->slots[0],
3392 &tmp);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04003393 if (ret)
3394 err = ret;
3395 else
3396 last_offset = tmp.offset;
Chris Masone02119d2008-09-05 16:13:11 -04003397 goto done;
3398 }
3399 }
3400done:
David Sterbab3b4aa72011-04-21 01:20:15 +02003401 btrfs_release_path(path);
3402 btrfs_release_path(dst_path);
Chris Masone02119d2008-09-05 16:13:11 -04003403
Yan, Zheng4a500fd2010-05-16 10:49:59 -04003404 if (err == 0) {
3405 *last_offset_ret = last_offset;
3406 /*
3407 * insert the log range keys to indicate where the log
3408 * is valid
3409 */
3410 ret = insert_dir_log_key(trans, log, path, key_type,
Li Zefan33345d012011-04-20 10:31:50 +08003411 ino, first_offset, last_offset);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04003412 if (ret)
3413 err = ret;
3414 }
3415 return err;
Chris Masone02119d2008-09-05 16:13:11 -04003416}
3417
3418/*
3419 * logging directories is very similar to logging inodes, We find all the items
3420 * from the current transaction and write them to the log.
3421 *
3422 * The recovery code scans the directory in the subvolume, and if it finds a
3423 * key in the range logged that is not present in the log tree, then it means
3424 * that dir entry was unlinked during the transaction.
3425 *
3426 * In order for that scan to work, we must include one key smaller than
3427 * the smallest logged by this transaction and one key larger than the largest
3428 * key logged by this transaction.
3429 */
3430static noinline int log_directory_changes(struct btrfs_trans_handle *trans,
3431 struct btrfs_root *root, struct inode *inode,
3432 struct btrfs_path *path,
Filipe Manana2f2ff0e2015-03-20 17:19:46 +00003433 struct btrfs_path *dst_path,
3434 struct btrfs_log_ctx *ctx)
Chris Masone02119d2008-09-05 16:13:11 -04003435{
3436 u64 min_key;
3437 u64 max_key;
3438 int ret;
3439 int key_type = BTRFS_DIR_ITEM_KEY;
3440
3441again:
3442 min_key = 0;
3443 max_key = 0;
Chris Masond3977122009-01-05 21:25:51 -05003444 while (1) {
Chris Masone02119d2008-09-05 16:13:11 -04003445 ret = log_dir_items(trans, root, inode, path,
Filipe Manana2f2ff0e2015-03-20 17:19:46 +00003446 dst_path, key_type, ctx, min_key,
Chris Masone02119d2008-09-05 16:13:11 -04003447 &max_key);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04003448 if (ret)
3449 return ret;
Chris Masone02119d2008-09-05 16:13:11 -04003450 if (max_key == (u64)-1)
3451 break;
3452 min_key = max_key + 1;
3453 }
3454
3455 if (key_type == BTRFS_DIR_ITEM_KEY) {
3456 key_type = BTRFS_DIR_INDEX_KEY;
3457 goto again;
3458 }
3459 return 0;
3460}
3461
3462/*
3463 * a helper function to drop items from the log before we relog an
3464 * inode. max_key_type indicates the highest item type to remove.
3465 * This cannot be run for file data extents because it does not
3466 * free the extents they point to.
3467 */
3468static int drop_objectid_items(struct btrfs_trans_handle *trans,
3469 struct btrfs_root *log,
3470 struct btrfs_path *path,
3471 u64 objectid, int max_key_type)
3472{
3473 int ret;
3474 struct btrfs_key key;
3475 struct btrfs_key found_key;
Josef Bacik18ec90d2012-09-28 11:56:28 -04003476 int start_slot;
Chris Masone02119d2008-09-05 16:13:11 -04003477
3478 key.objectid = objectid;
3479 key.type = max_key_type;
3480 key.offset = (u64)-1;
3481
Chris Masond3977122009-01-05 21:25:51 -05003482 while (1) {
Chris Masone02119d2008-09-05 16:13:11 -04003483 ret = btrfs_search_slot(trans, log, &key, path, -1, 1);
Josef Bacik36508602013-04-25 16:23:32 -04003484 BUG_ON(ret == 0); /* Logic error */
Yan, Zheng4a500fd2010-05-16 10:49:59 -04003485 if (ret < 0)
Chris Masone02119d2008-09-05 16:13:11 -04003486 break;
3487
3488 if (path->slots[0] == 0)
3489 break;
3490
3491 path->slots[0]--;
3492 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
3493 path->slots[0]);
3494
3495 if (found_key.objectid != objectid)
3496 break;
3497
Josef Bacik18ec90d2012-09-28 11:56:28 -04003498 found_key.offset = 0;
3499 found_key.type = 0;
3500 ret = btrfs_bin_search(path->nodes[0], &found_key, 0,
3501 &start_slot);
3502
3503 ret = btrfs_del_items(trans, log, path, start_slot,
3504 path->slots[0] - start_slot + 1);
3505 /*
3506 * If start slot isn't 0 then we don't need to re-search, we've
3507 * found the last guy with the objectid in this tree.
3508 */
3509 if (ret || start_slot != 0)
Tsutomu Itoh65a246c2011-05-19 04:37:44 +00003510 break;
David Sterbab3b4aa72011-04-21 01:20:15 +02003511 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04003512 }
David Sterbab3b4aa72011-04-21 01:20:15 +02003513 btrfs_release_path(path);
Josef Bacik5bdbeb22012-05-29 16:59:49 -04003514 if (ret > 0)
3515 ret = 0;
Yan, Zheng4a500fd2010-05-16 10:49:59 -04003516 return ret;
Chris Masone02119d2008-09-05 16:13:11 -04003517}
3518
Josef Bacik94edf4a2012-09-25 14:56:25 -04003519static void fill_inode_item(struct btrfs_trans_handle *trans,
3520 struct extent_buffer *leaf,
3521 struct btrfs_inode_item *item,
Filipe Manana1a4bcf42015-02-13 12:30:56 +00003522 struct inode *inode, int log_inode_only,
3523 u64 logged_isize)
Josef Bacik94edf4a2012-09-25 14:56:25 -04003524{
Josef Bacik0b1c6cc2012-10-23 16:03:44 -04003525 struct btrfs_map_token token;
Josef Bacik94edf4a2012-09-25 14:56:25 -04003526
Josef Bacik0b1c6cc2012-10-23 16:03:44 -04003527 btrfs_init_map_token(&token);
Josef Bacik94edf4a2012-09-25 14:56:25 -04003528
3529 if (log_inode_only) {
3530 /* set the generation to zero so the recover code
3531 * can tell the difference between an logging
3532 * just to say 'this inode exists' and a logging
3533 * to say 'update this inode with these values'
3534 */
Josef Bacik0b1c6cc2012-10-23 16:03:44 -04003535 btrfs_set_token_inode_generation(leaf, item, 0, &token);
Filipe Manana1a4bcf42015-02-13 12:30:56 +00003536 btrfs_set_token_inode_size(leaf, item, logged_isize, &token);
Josef Bacik94edf4a2012-09-25 14:56:25 -04003537 } else {
Josef Bacik0b1c6cc2012-10-23 16:03:44 -04003538 btrfs_set_token_inode_generation(leaf, item,
3539 BTRFS_I(inode)->generation,
3540 &token);
3541 btrfs_set_token_inode_size(leaf, item, inode->i_size, &token);
Josef Bacik94edf4a2012-09-25 14:56:25 -04003542 }
3543
Josef Bacik0b1c6cc2012-10-23 16:03:44 -04003544 btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token);
3545 btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token);
3546 btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token);
3547 btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token);
3548
David Sterbaa937b972014-12-12 17:39:12 +01003549 btrfs_set_token_timespec_sec(leaf, &item->atime,
Josef Bacik0b1c6cc2012-10-23 16:03:44 -04003550 inode->i_atime.tv_sec, &token);
David Sterbaa937b972014-12-12 17:39:12 +01003551 btrfs_set_token_timespec_nsec(leaf, &item->atime,
Josef Bacik0b1c6cc2012-10-23 16:03:44 -04003552 inode->i_atime.tv_nsec, &token);
3553
David Sterbaa937b972014-12-12 17:39:12 +01003554 btrfs_set_token_timespec_sec(leaf, &item->mtime,
Josef Bacik0b1c6cc2012-10-23 16:03:44 -04003555 inode->i_mtime.tv_sec, &token);
David Sterbaa937b972014-12-12 17:39:12 +01003556 btrfs_set_token_timespec_nsec(leaf, &item->mtime,
Josef Bacik0b1c6cc2012-10-23 16:03:44 -04003557 inode->i_mtime.tv_nsec, &token);
3558
David Sterbaa937b972014-12-12 17:39:12 +01003559 btrfs_set_token_timespec_sec(leaf, &item->ctime,
Josef Bacik0b1c6cc2012-10-23 16:03:44 -04003560 inode->i_ctime.tv_sec, &token);
David Sterbaa937b972014-12-12 17:39:12 +01003561 btrfs_set_token_timespec_nsec(leaf, &item->ctime,
Josef Bacik0b1c6cc2012-10-23 16:03:44 -04003562 inode->i_ctime.tv_nsec, &token);
3563
3564 btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode),
3565 &token);
3566
3567 btrfs_set_token_inode_sequence(leaf, item, inode->i_version, &token);
3568 btrfs_set_token_inode_transid(leaf, item, trans->transid, &token);
3569 btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token);
3570 btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token);
3571 btrfs_set_token_inode_block_group(leaf, item, 0, &token);
Josef Bacik94edf4a2012-09-25 14:56:25 -04003572}
3573
Josef Bacika95249b2012-10-11 16:17:34 -04003574static int log_inode_item(struct btrfs_trans_handle *trans,
3575 struct btrfs_root *log, struct btrfs_path *path,
3576 struct inode *inode)
3577{
3578 struct btrfs_inode_item *inode_item;
Josef Bacika95249b2012-10-11 16:17:34 -04003579 int ret;
3580
Filipe David Borba Mananaefd0c402013-10-07 21:20:44 +01003581 ret = btrfs_insert_empty_item(trans, log, path,
3582 &BTRFS_I(inode)->location,
Josef Bacika95249b2012-10-11 16:17:34 -04003583 sizeof(*inode_item));
3584 if (ret && ret != -EEXIST)
3585 return ret;
3586 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3587 struct btrfs_inode_item);
Filipe Manana1a4bcf42015-02-13 12:30:56 +00003588 fill_inode_item(trans, path->nodes[0], inode_item, inode, 0, 0);
Josef Bacika95249b2012-10-11 16:17:34 -04003589 btrfs_release_path(path);
3590 return 0;
3591}
3592
Chris Mason31ff1cd2008-09-11 16:17:57 -04003593static noinline int copy_items(struct btrfs_trans_handle *trans,
Liu Bod2794402012-08-29 01:07:56 -06003594 struct inode *inode,
Chris Mason31ff1cd2008-09-11 16:17:57 -04003595 struct btrfs_path *dst_path,
Josef Bacik16e75492013-10-22 12:18:51 -04003596 struct btrfs_path *src_path, u64 *last_extent,
Filipe Manana1a4bcf42015-02-13 12:30:56 +00003597 int start_slot, int nr, int inode_only,
3598 u64 logged_isize)
Chris Mason31ff1cd2008-09-11 16:17:57 -04003599{
3600 unsigned long src_offset;
3601 unsigned long dst_offset;
Liu Bod2794402012-08-29 01:07:56 -06003602 struct btrfs_root *log = BTRFS_I(inode)->root->log_root;
Chris Mason31ff1cd2008-09-11 16:17:57 -04003603 struct btrfs_file_extent_item *extent;
3604 struct btrfs_inode_item *inode_item;
Josef Bacik16e75492013-10-22 12:18:51 -04003605 struct extent_buffer *src = src_path->nodes[0];
3606 struct btrfs_key first_key, last_key, key;
Chris Mason31ff1cd2008-09-11 16:17:57 -04003607 int ret;
3608 struct btrfs_key *ins_keys;
3609 u32 *ins_sizes;
3610 char *ins_data;
3611 int i;
Chris Masond20f7042008-12-08 16:58:54 -05003612 struct list_head ordered_sums;
Liu Bod2794402012-08-29 01:07:56 -06003613 int skip_csum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
Josef Bacik16e75492013-10-22 12:18:51 -04003614 bool has_extents = false;
Filipe Manana74121f7c2014-08-07 12:00:44 +01003615 bool need_find_last_extent = true;
Josef Bacik16e75492013-10-22 12:18:51 -04003616 bool done = false;
Chris Masond20f7042008-12-08 16:58:54 -05003617
3618 INIT_LIST_HEAD(&ordered_sums);
Chris Mason31ff1cd2008-09-11 16:17:57 -04003619
3620 ins_data = kmalloc(nr * sizeof(struct btrfs_key) +
3621 nr * sizeof(u32), GFP_NOFS);
liubo2a29edc2011-01-26 06:22:08 +00003622 if (!ins_data)
3623 return -ENOMEM;
3624
Josef Bacik16e75492013-10-22 12:18:51 -04003625 first_key.objectid = (u64)-1;
3626
Chris Mason31ff1cd2008-09-11 16:17:57 -04003627 ins_sizes = (u32 *)ins_data;
3628 ins_keys = (struct btrfs_key *)(ins_data + nr * sizeof(u32));
3629
3630 for (i = 0; i < nr; i++) {
3631 ins_sizes[i] = btrfs_item_size_nr(src, i + start_slot);
3632 btrfs_item_key_to_cpu(src, ins_keys + i, i + start_slot);
3633 }
3634 ret = btrfs_insert_empty_items(trans, log, dst_path,
3635 ins_keys, ins_sizes, nr);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04003636 if (ret) {
3637 kfree(ins_data);
3638 return ret;
3639 }
Chris Mason31ff1cd2008-09-11 16:17:57 -04003640
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003641 for (i = 0; i < nr; i++, dst_path->slots[0]++) {
Chris Mason31ff1cd2008-09-11 16:17:57 -04003642 dst_offset = btrfs_item_ptr_offset(dst_path->nodes[0],
3643 dst_path->slots[0]);
3644
3645 src_offset = btrfs_item_ptr_offset(src, start_slot + i);
3646
Josef Bacik16e75492013-10-22 12:18:51 -04003647 if ((i == (nr - 1)))
3648 last_key = ins_keys[i];
3649
Josef Bacik94edf4a2012-09-25 14:56:25 -04003650 if (ins_keys[i].type == BTRFS_INODE_ITEM_KEY) {
Chris Mason31ff1cd2008-09-11 16:17:57 -04003651 inode_item = btrfs_item_ptr(dst_path->nodes[0],
3652 dst_path->slots[0],
3653 struct btrfs_inode_item);
Josef Bacik94edf4a2012-09-25 14:56:25 -04003654 fill_inode_item(trans, dst_path->nodes[0], inode_item,
Filipe Manana1a4bcf42015-02-13 12:30:56 +00003655 inode, inode_only == LOG_INODE_EXISTS,
3656 logged_isize);
Josef Bacik94edf4a2012-09-25 14:56:25 -04003657 } else {
3658 copy_extent_buffer(dst_path->nodes[0], src, dst_offset,
3659 src_offset, ins_sizes[i]);
Chris Mason31ff1cd2008-09-11 16:17:57 -04003660 }
Josef Bacik94edf4a2012-09-25 14:56:25 -04003661
Josef Bacik16e75492013-10-22 12:18:51 -04003662 /*
3663 * We set need_find_last_extent here in case we know we were
3664 * processing other items and then walk into the first extent in
3665 * the inode. If we don't hit an extent then nothing changes,
3666 * we'll do the last search the next time around.
3667 */
3668 if (ins_keys[i].type == BTRFS_EXTENT_DATA_KEY) {
3669 has_extents = true;
Filipe Manana74121f7c2014-08-07 12:00:44 +01003670 if (first_key.objectid == (u64)-1)
Josef Bacik16e75492013-10-22 12:18:51 -04003671 first_key = ins_keys[i];
3672 } else {
3673 need_find_last_extent = false;
3674 }
3675
Chris Mason31ff1cd2008-09-11 16:17:57 -04003676 /* take a reference on file data extents so that truncates
3677 * or deletes of this inode don't have to relog the inode
3678 * again
3679 */
David Sterba962a2982014-06-04 18:41:45 +02003680 if (ins_keys[i].type == BTRFS_EXTENT_DATA_KEY &&
Liu Bod2794402012-08-29 01:07:56 -06003681 !skip_csum) {
Chris Mason31ff1cd2008-09-11 16:17:57 -04003682 int found_type;
3683 extent = btrfs_item_ptr(src, start_slot + i,
3684 struct btrfs_file_extent_item);
3685
liubo8e531cd2011-05-06 10:36:09 +08003686 if (btrfs_file_extent_generation(src, extent) < trans->transid)
3687 continue;
3688
Chris Mason31ff1cd2008-09-11 16:17:57 -04003689 found_type = btrfs_file_extent_type(src, extent);
Josef Bacik6f1fed72012-09-26 11:07:06 -04003690 if (found_type == BTRFS_FILE_EXTENT_REG) {
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003691 u64 ds, dl, cs, cl;
3692 ds = btrfs_file_extent_disk_bytenr(src,
3693 extent);
3694 /* ds == 0 is a hole */
3695 if (ds == 0)
3696 continue;
3697
3698 dl = btrfs_file_extent_disk_num_bytes(src,
3699 extent);
3700 cs = btrfs_file_extent_offset(src, extent);
3701 cl = btrfs_file_extent_num_bytes(src,
Joe Perchesa419aef2009-08-18 11:18:35 -07003702 extent);
Chris Mason580afd72008-12-08 19:15:39 -05003703 if (btrfs_file_extent_compression(src,
3704 extent)) {
3705 cs = 0;
3706 cl = dl;
3707 }
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003708
3709 ret = btrfs_lookup_csums_range(
3710 log->fs_info->csum_root,
3711 ds + cs, ds + cs + cl - 1,
Arne Jansena2de7332011-03-08 14:14:00 +01003712 &ordered_sums, 0);
Josef Bacik36508602013-04-25 16:23:32 -04003713 if (ret) {
3714 btrfs_release_path(dst_path);
3715 kfree(ins_data);
3716 return ret;
3717 }
Chris Mason31ff1cd2008-09-11 16:17:57 -04003718 }
3719 }
Chris Mason31ff1cd2008-09-11 16:17:57 -04003720 }
3721
3722 btrfs_mark_buffer_dirty(dst_path->nodes[0]);
David Sterbab3b4aa72011-04-21 01:20:15 +02003723 btrfs_release_path(dst_path);
Chris Mason31ff1cd2008-09-11 16:17:57 -04003724 kfree(ins_data);
Chris Masond20f7042008-12-08 16:58:54 -05003725
3726 /*
3727 * we have to do this after the loop above to avoid changing the
3728 * log tree while trying to change the log tree.
3729 */
Yan, Zheng4a500fd2010-05-16 10:49:59 -04003730 ret = 0;
Chris Masond3977122009-01-05 21:25:51 -05003731 while (!list_empty(&ordered_sums)) {
Chris Masond20f7042008-12-08 16:58:54 -05003732 struct btrfs_ordered_sum *sums = list_entry(ordered_sums.next,
3733 struct btrfs_ordered_sum,
3734 list);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04003735 if (!ret)
3736 ret = btrfs_csum_file_blocks(trans, log, sums);
Chris Masond20f7042008-12-08 16:58:54 -05003737 list_del(&sums->list);
3738 kfree(sums);
3739 }
Josef Bacik16e75492013-10-22 12:18:51 -04003740
3741 if (!has_extents)
3742 return ret;
3743
Filipe Manana74121f7c2014-08-07 12:00:44 +01003744 if (need_find_last_extent && *last_extent == first_key.offset) {
3745 /*
3746 * We don't have any leafs between our current one and the one
3747 * we processed before that can have file extent items for our
3748 * inode (and have a generation number smaller than our current
3749 * transaction id).
3750 */
3751 need_find_last_extent = false;
3752 }
3753
Josef Bacik16e75492013-10-22 12:18:51 -04003754 /*
3755 * Because we use btrfs_search_forward we could skip leaves that were
3756 * not modified and then assume *last_extent is valid when it really
3757 * isn't. So back up to the previous leaf and read the end of the last
3758 * extent before we go and fill in holes.
3759 */
3760 if (need_find_last_extent) {
3761 u64 len;
3762
3763 ret = btrfs_prev_leaf(BTRFS_I(inode)->root, src_path);
3764 if (ret < 0)
3765 return ret;
3766 if (ret)
3767 goto fill_holes;
3768 if (src_path->slots[0])
3769 src_path->slots[0]--;
3770 src = src_path->nodes[0];
3771 btrfs_item_key_to_cpu(src, &key, src_path->slots[0]);
3772 if (key.objectid != btrfs_ino(inode) ||
3773 key.type != BTRFS_EXTENT_DATA_KEY)
3774 goto fill_holes;
3775 extent = btrfs_item_ptr(src, src_path->slots[0],
3776 struct btrfs_file_extent_item);
3777 if (btrfs_file_extent_type(src, extent) ==
3778 BTRFS_FILE_EXTENT_INLINE) {
Chris Mason514ac8a2014-01-03 21:07:00 -08003779 len = btrfs_file_extent_inline_len(src,
3780 src_path->slots[0],
3781 extent);
Josef Bacik16e75492013-10-22 12:18:51 -04003782 *last_extent = ALIGN(key.offset + len,
3783 log->sectorsize);
3784 } else {
3785 len = btrfs_file_extent_num_bytes(src, extent);
3786 *last_extent = key.offset + len;
3787 }
3788 }
3789fill_holes:
3790 /* So we did prev_leaf, now we need to move to the next leaf, but a few
3791 * things could have happened
3792 *
3793 * 1) A merge could have happened, so we could currently be on a leaf
3794 * that holds what we were copying in the first place.
3795 * 2) A split could have happened, and now not all of the items we want
3796 * are on the same leaf.
3797 *
3798 * So we need to adjust how we search for holes, we need to drop the
3799 * path and re-search for the first extent key we found, and then walk
3800 * forward until we hit the last one we copied.
3801 */
3802 if (need_find_last_extent) {
3803 /* btrfs_prev_leaf could return 1 without releasing the path */
3804 btrfs_release_path(src_path);
3805 ret = btrfs_search_slot(NULL, BTRFS_I(inode)->root, &first_key,
3806 src_path, 0, 0);
3807 if (ret < 0)
3808 return ret;
3809 ASSERT(ret == 0);
3810 src = src_path->nodes[0];
3811 i = src_path->slots[0];
3812 } else {
3813 i = start_slot;
3814 }
3815
3816 /*
3817 * Ok so here we need to go through and fill in any holes we may have
3818 * to make sure that holes are punched for those areas in case they had
3819 * extents previously.
3820 */
3821 while (!done) {
3822 u64 offset, len;
3823 u64 extent_end;
3824
3825 if (i >= btrfs_header_nritems(src_path->nodes[0])) {
3826 ret = btrfs_next_leaf(BTRFS_I(inode)->root, src_path);
3827 if (ret < 0)
3828 return ret;
3829 ASSERT(ret == 0);
3830 src = src_path->nodes[0];
3831 i = 0;
3832 }
3833
3834 btrfs_item_key_to_cpu(src, &key, i);
3835 if (!btrfs_comp_cpu_keys(&key, &last_key))
3836 done = true;
3837 if (key.objectid != btrfs_ino(inode) ||
3838 key.type != BTRFS_EXTENT_DATA_KEY) {
3839 i++;
3840 continue;
3841 }
3842 extent = btrfs_item_ptr(src, i, struct btrfs_file_extent_item);
3843 if (btrfs_file_extent_type(src, extent) ==
3844 BTRFS_FILE_EXTENT_INLINE) {
Chris Mason514ac8a2014-01-03 21:07:00 -08003845 len = btrfs_file_extent_inline_len(src, i, extent);
Josef Bacik16e75492013-10-22 12:18:51 -04003846 extent_end = ALIGN(key.offset + len, log->sectorsize);
3847 } else {
3848 len = btrfs_file_extent_num_bytes(src, extent);
3849 extent_end = key.offset + len;
3850 }
3851 i++;
3852
3853 if (*last_extent == key.offset) {
3854 *last_extent = extent_end;
3855 continue;
3856 }
3857 offset = *last_extent;
3858 len = key.offset - *last_extent;
3859 ret = btrfs_insert_file_extent(trans, log, btrfs_ino(inode),
3860 offset, 0, 0, len, 0, len, 0,
3861 0, 0);
3862 if (ret)
3863 break;
Filipe Manana74121f7c2014-08-07 12:00:44 +01003864 *last_extent = extent_end;
Josef Bacik16e75492013-10-22 12:18:51 -04003865 }
3866 /*
3867 * Need to let the callers know we dropped the path so they should
3868 * re-search.
3869 */
3870 if (!ret && need_find_last_extent)
3871 ret = 1;
Yan, Zheng4a500fd2010-05-16 10:49:59 -04003872 return ret;
Chris Mason31ff1cd2008-09-11 16:17:57 -04003873}
3874
Josef Bacik5dc562c2012-08-17 13:14:17 -04003875static int extent_cmp(void *priv, struct list_head *a, struct list_head *b)
3876{
3877 struct extent_map *em1, *em2;
3878
3879 em1 = list_entry(a, struct extent_map, list);
3880 em2 = list_entry(b, struct extent_map, list);
3881
3882 if (em1->start < em2->start)
3883 return -1;
3884 else if (em1->start > em2->start)
3885 return 1;
3886 return 0;
3887}
3888
Filipe Manana8407f552014-09-05 15:14:39 +01003889static int wait_ordered_extents(struct btrfs_trans_handle *trans,
3890 struct inode *inode,
3891 struct btrfs_root *root,
3892 const struct extent_map *em,
3893 const struct list_head *logged_list,
3894 bool *ordered_io_error)
Josef Bacik5dc562c2012-08-17 13:14:17 -04003895{
Josef Bacik2ab28f32012-10-12 15:27:49 -04003896 struct btrfs_ordered_extent *ordered;
Filipe Manana8407f552014-09-05 15:14:39 +01003897 struct btrfs_root *log = root->log_root;
Josef Bacik2ab28f32012-10-12 15:27:49 -04003898 u64 mod_start = em->mod_start;
3899 u64 mod_len = em->mod_len;
Filipe Manana8407f552014-09-05 15:14:39 +01003900 const bool skip_csum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
Josef Bacik2ab28f32012-10-12 15:27:49 -04003901 u64 csum_offset;
3902 u64 csum_len;
Filipe Manana8407f552014-09-05 15:14:39 +01003903 LIST_HEAD(ordered_sums);
3904 int ret = 0;
Josef Bacik09a2a8f92013-04-05 16:51:15 -04003905
Filipe Manana8407f552014-09-05 15:14:39 +01003906 *ordered_io_error = false;
Josef Bacik70c8a912012-10-11 16:54:30 -04003907
Filipe Manana8407f552014-09-05 15:14:39 +01003908 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
3909 em->block_start == EXTENT_MAP_HOLE)
Josef Bacik70c8a912012-10-11 16:54:30 -04003910 return 0;
3911
Josef Bacik2ab28f32012-10-12 15:27:49 -04003912 /*
Filipe Manana8407f552014-09-05 15:14:39 +01003913 * Wait far any ordered extent that covers our extent map. If it
3914 * finishes without an error, first check and see if our csums are on
3915 * our outstanding ordered extents.
Josef Bacik2ab28f32012-10-12 15:27:49 -04003916 */
Miao Xie827463c2014-01-14 20:31:51 +08003917 list_for_each_entry(ordered, logged_list, log_list) {
Josef Bacik2ab28f32012-10-12 15:27:49 -04003918 struct btrfs_ordered_sum *sum;
3919
3920 if (!mod_len)
3921 break;
3922
Josef Bacik2ab28f32012-10-12 15:27:49 -04003923 if (ordered->file_offset + ordered->len <= mod_start ||
3924 mod_start + mod_len <= ordered->file_offset)
3925 continue;
3926
Filipe Manana8407f552014-09-05 15:14:39 +01003927 if (!test_bit(BTRFS_ORDERED_IO_DONE, &ordered->flags) &&
3928 !test_bit(BTRFS_ORDERED_IOERR, &ordered->flags) &&
3929 !test_bit(BTRFS_ORDERED_DIRECT, &ordered->flags)) {
3930 const u64 start = ordered->file_offset;
3931 const u64 end = ordered->file_offset + ordered->len - 1;
3932
3933 WARN_ON(ordered->inode != inode);
3934 filemap_fdatawrite_range(inode->i_mapping, start, end);
3935 }
3936
3937 wait_event(ordered->wait,
3938 (test_bit(BTRFS_ORDERED_IO_DONE, &ordered->flags) ||
3939 test_bit(BTRFS_ORDERED_IOERR, &ordered->flags)));
3940
3941 if (test_bit(BTRFS_ORDERED_IOERR, &ordered->flags)) {
Filipe Mananab38ef712014-11-13 17:01:45 +00003942 /*
3943 * Clear the AS_EIO/AS_ENOSPC flags from the inode's
3944 * i_mapping flags, so that the next fsync won't get
3945 * an outdated io error too.
3946 */
3947 btrfs_inode_check_errors(inode);
Filipe Manana8407f552014-09-05 15:14:39 +01003948 *ordered_io_error = true;
3949 break;
3950 }
Josef Bacik2ab28f32012-10-12 15:27:49 -04003951 /*
3952 * We are going to copy all the csums on this ordered extent, so
3953 * go ahead and adjust mod_start and mod_len in case this
3954 * ordered extent has already been logged.
3955 */
3956 if (ordered->file_offset > mod_start) {
3957 if (ordered->file_offset + ordered->len >=
3958 mod_start + mod_len)
3959 mod_len = ordered->file_offset - mod_start;
3960 /*
3961 * If we have this case
3962 *
3963 * |--------- logged extent ---------|
3964 * |----- ordered extent ----|
3965 *
3966 * Just don't mess with mod_start and mod_len, we'll
3967 * just end up logging more csums than we need and it
3968 * will be ok.
3969 */
3970 } else {
3971 if (ordered->file_offset + ordered->len <
3972 mod_start + mod_len) {
3973 mod_len = (mod_start + mod_len) -
3974 (ordered->file_offset + ordered->len);
3975 mod_start = ordered->file_offset +
3976 ordered->len;
3977 } else {
3978 mod_len = 0;
3979 }
3980 }
3981
Filipe Manana8407f552014-09-05 15:14:39 +01003982 if (skip_csum)
3983 continue;
3984
Josef Bacik2ab28f32012-10-12 15:27:49 -04003985 /*
3986 * To keep us from looping for the above case of an ordered
3987 * extent that falls inside of the logged extent.
3988 */
3989 if (test_and_set_bit(BTRFS_ORDERED_LOGGED_CSUM,
3990 &ordered->flags))
3991 continue;
Josef Bacik2ab28f32012-10-12 15:27:49 -04003992
Josef Bacik2ab28f32012-10-12 15:27:49 -04003993 list_for_each_entry(sum, &ordered->list, list) {
3994 ret = btrfs_csum_file_blocks(trans, log, sum);
Miao Xie827463c2014-01-14 20:31:51 +08003995 if (ret)
Filipe Manana8407f552014-09-05 15:14:39 +01003996 break;
Josef Bacik2ab28f32012-10-12 15:27:49 -04003997 }
Josef Bacik2ab28f32012-10-12 15:27:49 -04003998 }
Josef Bacik2ab28f32012-10-12 15:27:49 -04003999
Filipe Manana8407f552014-09-05 15:14:39 +01004000 if (*ordered_io_error || !mod_len || ret || skip_csum)
Josef Bacik2ab28f32012-10-12 15:27:49 -04004001 return ret;
4002
Filipe David Borba Manana488111a2013-10-28 16:30:29 +00004003 if (em->compress_type) {
4004 csum_offset = 0;
Filipe Manana8407f552014-09-05 15:14:39 +01004005 csum_len = max(em->block_len, em->orig_block_len);
Filipe David Borba Manana488111a2013-10-28 16:30:29 +00004006 } else {
4007 csum_offset = mod_start - em->start;
4008 csum_len = mod_len;
4009 }
Josef Bacik2ab28f32012-10-12 15:27:49 -04004010
Josef Bacik70c8a912012-10-11 16:54:30 -04004011 /* block start is already adjusted for the file extent offset. */
4012 ret = btrfs_lookup_csums_range(log->fs_info->csum_root,
4013 em->block_start + csum_offset,
4014 em->block_start + csum_offset +
4015 csum_len - 1, &ordered_sums, 0);
4016 if (ret)
4017 return ret;
4018
4019 while (!list_empty(&ordered_sums)) {
4020 struct btrfs_ordered_sum *sums = list_entry(ordered_sums.next,
4021 struct btrfs_ordered_sum,
4022 list);
4023 if (!ret)
4024 ret = btrfs_csum_file_blocks(trans, log, sums);
4025 list_del(&sums->list);
4026 kfree(sums);
4027 }
4028
4029 return ret;
Josef Bacik5dc562c2012-08-17 13:14:17 -04004030}
4031
Filipe Manana8407f552014-09-05 15:14:39 +01004032static int log_one_extent(struct btrfs_trans_handle *trans,
4033 struct inode *inode, struct btrfs_root *root,
4034 const struct extent_map *em,
4035 struct btrfs_path *path,
4036 const struct list_head *logged_list,
4037 struct btrfs_log_ctx *ctx)
4038{
4039 struct btrfs_root *log = root->log_root;
4040 struct btrfs_file_extent_item *fi;
4041 struct extent_buffer *leaf;
4042 struct btrfs_map_token token;
4043 struct btrfs_key key;
4044 u64 extent_offset = em->start - em->orig_start;
4045 u64 block_len;
4046 int ret;
4047 int extent_inserted = 0;
4048 bool ordered_io_err = false;
4049
4050 ret = wait_ordered_extents(trans, inode, root, em, logged_list,
4051 &ordered_io_err);
4052 if (ret)
4053 return ret;
4054
4055 if (ordered_io_err) {
4056 ctx->io_err = -EIO;
4057 return 0;
4058 }
4059
4060 btrfs_init_map_token(&token);
4061
4062 ret = __btrfs_drop_extents(trans, log, inode, path, em->start,
4063 em->start + em->len, NULL, 0, 1,
4064 sizeof(*fi), &extent_inserted);
4065 if (ret)
4066 return ret;
4067
4068 if (!extent_inserted) {
4069 key.objectid = btrfs_ino(inode);
4070 key.type = BTRFS_EXTENT_DATA_KEY;
4071 key.offset = em->start;
4072
4073 ret = btrfs_insert_empty_item(trans, log, path, &key,
4074 sizeof(*fi));
4075 if (ret)
4076 return ret;
4077 }
4078 leaf = path->nodes[0];
4079 fi = btrfs_item_ptr(leaf, path->slots[0],
4080 struct btrfs_file_extent_item);
4081
Josef Bacik50d9aa92014-11-21 14:52:38 -05004082 btrfs_set_token_file_extent_generation(leaf, fi, trans->transid,
Filipe Manana8407f552014-09-05 15:14:39 +01004083 &token);
4084 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
4085 btrfs_set_token_file_extent_type(leaf, fi,
4086 BTRFS_FILE_EXTENT_PREALLOC,
4087 &token);
4088 else
4089 btrfs_set_token_file_extent_type(leaf, fi,
4090 BTRFS_FILE_EXTENT_REG,
4091 &token);
4092
4093 block_len = max(em->block_len, em->orig_block_len);
4094 if (em->compress_type != BTRFS_COMPRESS_NONE) {
4095 btrfs_set_token_file_extent_disk_bytenr(leaf, fi,
4096 em->block_start,
4097 &token);
4098 btrfs_set_token_file_extent_disk_num_bytes(leaf, fi, block_len,
4099 &token);
4100 } else if (em->block_start < EXTENT_MAP_LAST_BYTE) {
4101 btrfs_set_token_file_extent_disk_bytenr(leaf, fi,
4102 em->block_start -
4103 extent_offset, &token);
4104 btrfs_set_token_file_extent_disk_num_bytes(leaf, fi, block_len,
4105 &token);
4106 } else {
4107 btrfs_set_token_file_extent_disk_bytenr(leaf, fi, 0, &token);
4108 btrfs_set_token_file_extent_disk_num_bytes(leaf, fi, 0,
4109 &token);
4110 }
4111
4112 btrfs_set_token_file_extent_offset(leaf, fi, extent_offset, &token);
4113 btrfs_set_token_file_extent_num_bytes(leaf, fi, em->len, &token);
4114 btrfs_set_token_file_extent_ram_bytes(leaf, fi, em->ram_bytes, &token);
4115 btrfs_set_token_file_extent_compression(leaf, fi, em->compress_type,
4116 &token);
4117 btrfs_set_token_file_extent_encryption(leaf, fi, 0, &token);
4118 btrfs_set_token_file_extent_other_encoding(leaf, fi, 0, &token);
4119 btrfs_mark_buffer_dirty(leaf);
4120
4121 btrfs_release_path(path);
4122
4123 return ret;
4124}
4125
Josef Bacik5dc562c2012-08-17 13:14:17 -04004126static int btrfs_log_changed_extents(struct btrfs_trans_handle *trans,
4127 struct btrfs_root *root,
4128 struct inode *inode,
Miao Xie827463c2014-01-14 20:31:51 +08004129 struct btrfs_path *path,
Filipe Manana8407f552014-09-05 15:14:39 +01004130 struct list_head *logged_list,
4131 struct btrfs_log_ctx *ctx)
Josef Bacik5dc562c2012-08-17 13:14:17 -04004132{
Josef Bacik5dc562c2012-08-17 13:14:17 -04004133 struct extent_map *em, *n;
4134 struct list_head extents;
4135 struct extent_map_tree *tree = &BTRFS_I(inode)->extent_tree;
4136 u64 test_gen;
4137 int ret = 0;
Josef Bacik2ab28f32012-10-12 15:27:49 -04004138 int num = 0;
Josef Bacik5dc562c2012-08-17 13:14:17 -04004139
4140 INIT_LIST_HEAD(&extents);
4141
Josef Bacik5dc562c2012-08-17 13:14:17 -04004142 write_lock(&tree->lock);
4143 test_gen = root->fs_info->last_trans_committed;
4144
4145 list_for_each_entry_safe(em, n, &tree->modified_extents, list) {
4146 list_del_init(&em->list);
Josef Bacik2ab28f32012-10-12 15:27:49 -04004147
4148 /*
4149 * Just an arbitrary number, this can be really CPU intensive
4150 * once we start getting a lot of extents, and really once we
4151 * have a bunch of extents we just want to commit since it will
4152 * be faster.
4153 */
4154 if (++num > 32768) {
4155 list_del_init(&tree->modified_extents);
4156 ret = -EFBIG;
4157 goto process;
4158 }
4159
Josef Bacik5dc562c2012-08-17 13:14:17 -04004160 if (em->generation <= test_gen)
4161 continue;
Josef Bacikff44c6e2012-09-14 12:59:20 -04004162 /* Need a ref to keep it from getting evicted from cache */
4163 atomic_inc(&em->refs);
4164 set_bit(EXTENT_FLAG_LOGGING, &em->flags);
Josef Bacik5dc562c2012-08-17 13:14:17 -04004165 list_add_tail(&em->list, &extents);
Josef Bacik2ab28f32012-10-12 15:27:49 -04004166 num++;
Josef Bacik5dc562c2012-08-17 13:14:17 -04004167 }
4168
4169 list_sort(NULL, &extents, extent_cmp);
4170
Josef Bacik2ab28f32012-10-12 15:27:49 -04004171process:
Josef Bacik5dc562c2012-08-17 13:14:17 -04004172 while (!list_empty(&extents)) {
4173 em = list_entry(extents.next, struct extent_map, list);
4174
4175 list_del_init(&em->list);
4176
4177 /*
4178 * If we had an error we just need to delete everybody from our
4179 * private list.
4180 */
Josef Bacikff44c6e2012-09-14 12:59:20 -04004181 if (ret) {
Josef Bacik201a9032013-01-24 12:02:07 -05004182 clear_em_logging(tree, em);
Josef Bacikff44c6e2012-09-14 12:59:20 -04004183 free_extent_map(em);
Josef Bacik5dc562c2012-08-17 13:14:17 -04004184 continue;
Josef Bacikff44c6e2012-09-14 12:59:20 -04004185 }
4186
4187 write_unlock(&tree->lock);
Josef Bacik5dc562c2012-08-17 13:14:17 -04004188
Filipe Manana8407f552014-09-05 15:14:39 +01004189 ret = log_one_extent(trans, inode, root, em, path, logged_list,
4190 ctx);
Josef Bacikff44c6e2012-09-14 12:59:20 -04004191 write_lock(&tree->lock);
Josef Bacik201a9032013-01-24 12:02:07 -05004192 clear_em_logging(tree, em);
4193 free_extent_map(em);
Josef Bacik5dc562c2012-08-17 13:14:17 -04004194 }
Josef Bacikff44c6e2012-09-14 12:59:20 -04004195 WARN_ON(!list_empty(&extents));
4196 write_unlock(&tree->lock);
Josef Bacik5dc562c2012-08-17 13:14:17 -04004197
Josef Bacik5dc562c2012-08-17 13:14:17 -04004198 btrfs_release_path(path);
Josef Bacik5dc562c2012-08-17 13:14:17 -04004199 return ret;
4200}
4201
Filipe Manana1a4bcf42015-02-13 12:30:56 +00004202static int logged_inode_size(struct btrfs_root *log, struct inode *inode,
4203 struct btrfs_path *path, u64 *size_ret)
4204{
4205 struct btrfs_key key;
4206 int ret;
4207
4208 key.objectid = btrfs_ino(inode);
4209 key.type = BTRFS_INODE_ITEM_KEY;
4210 key.offset = 0;
4211
4212 ret = btrfs_search_slot(NULL, log, &key, path, 0, 0);
4213 if (ret < 0) {
4214 return ret;
4215 } else if (ret > 0) {
Filipe Manana2f2ff0e2015-03-20 17:19:46 +00004216 *size_ret = 0;
Filipe Manana1a4bcf42015-02-13 12:30:56 +00004217 } else {
4218 struct btrfs_inode_item *item;
4219
4220 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4221 struct btrfs_inode_item);
4222 *size_ret = btrfs_inode_size(path->nodes[0], item);
4223 }
4224
4225 btrfs_release_path(path);
4226 return 0;
4227}
4228
Filipe Manana36283bf2015-06-20 00:44:51 +01004229/*
4230 * At the moment we always log all xattrs. This is to figure out at log replay
4231 * time which xattrs must have their deletion replayed. If a xattr is missing
4232 * in the log tree and exists in the fs/subvol tree, we delete it. This is
4233 * because if a xattr is deleted, the inode is fsynced and a power failure
4234 * happens, causing the log to be replayed the next time the fs is mounted,
4235 * we want the xattr to not exist anymore (same behaviour as other filesystems
4236 * with a journal, ext3/4, xfs, f2fs, etc).
4237 */
4238static int btrfs_log_all_xattrs(struct btrfs_trans_handle *trans,
4239 struct btrfs_root *root,
4240 struct inode *inode,
4241 struct btrfs_path *path,
4242 struct btrfs_path *dst_path)
4243{
4244 int ret;
4245 struct btrfs_key key;
4246 const u64 ino = btrfs_ino(inode);
4247 int ins_nr = 0;
4248 int start_slot = 0;
4249
4250 key.objectid = ino;
4251 key.type = BTRFS_XATTR_ITEM_KEY;
4252 key.offset = 0;
4253
4254 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4255 if (ret < 0)
4256 return ret;
4257
4258 while (true) {
4259 int slot = path->slots[0];
4260 struct extent_buffer *leaf = path->nodes[0];
4261 int nritems = btrfs_header_nritems(leaf);
4262
4263 if (slot >= nritems) {
4264 if (ins_nr > 0) {
4265 u64 last_extent = 0;
4266
4267 ret = copy_items(trans, inode, dst_path, path,
4268 &last_extent, start_slot,
4269 ins_nr, 1, 0);
4270 /* can't be 1, extent items aren't processed */
4271 ASSERT(ret <= 0);
4272 if (ret < 0)
4273 return ret;
4274 ins_nr = 0;
4275 }
4276 ret = btrfs_next_leaf(root, path);
4277 if (ret < 0)
4278 return ret;
4279 else if (ret > 0)
4280 break;
4281 continue;
4282 }
4283
4284 btrfs_item_key_to_cpu(leaf, &key, slot);
4285 if (key.objectid != ino || key.type != BTRFS_XATTR_ITEM_KEY)
4286 break;
4287
4288 if (ins_nr == 0)
4289 start_slot = slot;
4290 ins_nr++;
4291 path->slots[0]++;
4292 cond_resched();
4293 }
4294 if (ins_nr > 0) {
4295 u64 last_extent = 0;
4296
4297 ret = copy_items(trans, inode, dst_path, path,
4298 &last_extent, start_slot,
4299 ins_nr, 1, 0);
4300 /* can't be 1, extent items aren't processed */
4301 ASSERT(ret <= 0);
4302 if (ret < 0)
4303 return ret;
4304 }
4305
4306 return 0;
4307}
4308
Filipe Mananaa89ca6f2015-06-25 04:17:46 +01004309/*
4310 * If the no holes feature is enabled we need to make sure any hole between the
4311 * last extent and the i_size of our inode is explicitly marked in the log. This
4312 * is to make sure that doing something like:
4313 *
4314 * 1) create file with 128Kb of data
4315 * 2) truncate file to 64Kb
4316 * 3) truncate file to 256Kb
4317 * 4) fsync file
4318 * 5) <crash/power failure>
4319 * 6) mount fs and trigger log replay
4320 *
4321 * Will give us a file with a size of 256Kb, the first 64Kb of data match what
4322 * the file had in its first 64Kb of data at step 1 and the last 192Kb of the
4323 * file correspond to a hole. The presence of explicit holes in a log tree is
4324 * what guarantees that log replay will remove/adjust file extent items in the
4325 * fs/subvol tree.
4326 *
4327 * Here we do not need to care about holes between extents, that is already done
4328 * by copy_items(). We also only need to do this in the full sync path, where we
4329 * lookup for extents from the fs/subvol tree only. In the fast path case, we
4330 * lookup the list of modified extent maps and if any represents a hole, we
4331 * insert a corresponding extent representing a hole in the log tree.
4332 */
4333static int btrfs_log_trailing_hole(struct btrfs_trans_handle *trans,
4334 struct btrfs_root *root,
4335 struct inode *inode,
4336 struct btrfs_path *path)
4337{
4338 int ret;
4339 struct btrfs_key key;
4340 u64 hole_start;
4341 u64 hole_size;
4342 struct extent_buffer *leaf;
4343 struct btrfs_root *log = root->log_root;
4344 const u64 ino = btrfs_ino(inode);
4345 const u64 i_size = i_size_read(inode);
4346
4347 if (!btrfs_fs_incompat(root->fs_info, NO_HOLES))
4348 return 0;
4349
4350 key.objectid = ino;
4351 key.type = BTRFS_EXTENT_DATA_KEY;
4352 key.offset = (u64)-1;
4353
4354 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4355 ASSERT(ret != 0);
4356 if (ret < 0)
4357 return ret;
4358
4359 ASSERT(path->slots[0] > 0);
4360 path->slots[0]--;
4361 leaf = path->nodes[0];
4362 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4363
4364 if (key.objectid != ino || key.type != BTRFS_EXTENT_DATA_KEY) {
4365 /* inode does not have any extents */
4366 hole_start = 0;
4367 hole_size = i_size;
4368 } else {
4369 struct btrfs_file_extent_item *extent;
4370 u64 len;
4371
4372 /*
4373 * If there's an extent beyond i_size, an explicit hole was
4374 * already inserted by copy_items().
4375 */
4376 if (key.offset >= i_size)
4377 return 0;
4378
4379 extent = btrfs_item_ptr(leaf, path->slots[0],
4380 struct btrfs_file_extent_item);
4381
4382 if (btrfs_file_extent_type(leaf, extent) ==
4383 BTRFS_FILE_EXTENT_INLINE) {
4384 len = btrfs_file_extent_inline_len(leaf,
4385 path->slots[0],
4386 extent);
4387 ASSERT(len == i_size);
4388 return 0;
4389 }
4390
4391 len = btrfs_file_extent_num_bytes(leaf, extent);
4392 /* Last extent goes beyond i_size, no need to log a hole. */
4393 if (key.offset + len > i_size)
4394 return 0;
4395 hole_start = key.offset + len;
4396 hole_size = i_size - hole_start;
4397 }
4398 btrfs_release_path(path);
4399
4400 /* Last extent ends at i_size. */
4401 if (hole_size == 0)
4402 return 0;
4403
4404 hole_size = ALIGN(hole_size, root->sectorsize);
4405 ret = btrfs_insert_file_extent(trans, log, ino, hole_start, 0, 0,
4406 hole_size, 0, hole_size, 0, 0, 0);
4407 return ret;
4408}
4409
Filipe Manana1653a3b2016-03-30 23:37:21 +01004410/*
4411 * When we are logging a new inode X, check if it doesn't have a reference that
4412 * matches the reference from some other inode Y created in a past transaction
4413 * and that was renamed in the current transaction. If we don't do this, then at
4414 * log replay time we can lose inode Y (and all its files if it's a directory):
4415 *
4416 * mkdir /mnt/x
4417 * echo "hello world" > /mnt/x/foobar
4418 * sync
4419 * mv /mnt/x /mnt/y
4420 * mkdir /mnt/x # or touch /mnt/x
4421 * xfs_io -c fsync /mnt/x
4422 * <power fail>
4423 * mount fs, trigger log replay
4424 *
4425 * After the log replay procedure, we would lose the first directory and all its
4426 * files (file foobar).
4427 * For the case where inode Y is not a directory we simply end up losing it:
4428 *
4429 * echo "123" > /mnt/foo
4430 * sync
4431 * mv /mnt/foo /mnt/bar
4432 * echo "abc" > /mnt/foo
4433 * xfs_io -c fsync /mnt/foo
4434 * <power fail>
4435 *
4436 * We also need this for cases where a snapshot entry is replaced by some other
4437 * entry (file or directory) otherwise we end up with an unreplayable log due to
4438 * attempts to delete the snapshot entry (entry of type BTRFS_ROOT_ITEM_KEY) as
4439 * if it were a regular entry:
4440 *
4441 * mkdir /mnt/x
4442 * btrfs subvolume snapshot /mnt /mnt/x/snap
4443 * btrfs subvolume delete /mnt/x/snap
4444 * rmdir /mnt/x
4445 * mkdir /mnt/x
4446 * fsync /mnt/x or fsync some new file inside it
4447 * <power fail>
4448 *
4449 * The snapshot delete, rmdir of x, mkdir of a new x and the fsync all happen in
4450 * the same transaction.
4451 */
4452static int btrfs_check_ref_name_override(struct extent_buffer *eb,
4453 const int slot,
4454 const struct btrfs_key *key,
4455 struct inode *inode)
4456{
4457 int ret;
4458 struct btrfs_path *search_path;
4459 char *name = NULL;
4460 u32 name_len = 0;
4461 u32 item_size = btrfs_item_size_nr(eb, slot);
4462 u32 cur_offset = 0;
4463 unsigned long ptr = btrfs_item_ptr_offset(eb, slot);
4464
4465 search_path = btrfs_alloc_path();
4466 if (!search_path)
4467 return -ENOMEM;
4468 search_path->search_commit_root = 1;
4469 search_path->skip_locking = 1;
4470
4471 while (cur_offset < item_size) {
4472 u64 parent;
4473 u32 this_name_len;
4474 u32 this_len;
4475 unsigned long name_ptr;
4476 struct btrfs_dir_item *di;
4477
4478 if (key->type == BTRFS_INODE_REF_KEY) {
4479 struct btrfs_inode_ref *iref;
4480
4481 iref = (struct btrfs_inode_ref *)(ptr + cur_offset);
4482 parent = key->offset;
4483 this_name_len = btrfs_inode_ref_name_len(eb, iref);
4484 name_ptr = (unsigned long)(iref + 1);
4485 this_len = sizeof(*iref) + this_name_len;
4486 } else {
4487 struct btrfs_inode_extref *extref;
4488
4489 extref = (struct btrfs_inode_extref *)(ptr +
4490 cur_offset);
4491 parent = btrfs_inode_extref_parent(eb, extref);
4492 this_name_len = btrfs_inode_extref_name_len(eb, extref);
4493 name_ptr = (unsigned long)&extref->name;
4494 this_len = sizeof(*extref) + this_name_len;
4495 }
4496
4497 if (this_name_len > name_len) {
4498 char *new_name;
4499
4500 new_name = krealloc(name, this_name_len, GFP_NOFS);
4501 if (!new_name) {
4502 ret = -ENOMEM;
4503 goto out;
4504 }
4505 name_len = this_name_len;
4506 name = new_name;
4507 }
4508
4509 read_extent_buffer(eb, name, name_ptr, this_name_len);
4510 di = btrfs_lookup_dir_item(NULL, BTRFS_I(inode)->root,
4511 search_path, parent,
4512 name, this_name_len, 0);
4513 if (di && !IS_ERR(di)) {
4514 ret = 1;
4515 goto out;
4516 } else if (IS_ERR(di)) {
4517 ret = PTR_ERR(di);
4518 goto out;
4519 }
4520 btrfs_release_path(search_path);
4521
4522 cur_offset += this_len;
4523 }
4524 ret = 0;
4525out:
4526 btrfs_free_path(search_path);
4527 kfree(name);
4528 return ret;
4529}
4530
Chris Masone02119d2008-09-05 16:13:11 -04004531/* log a single inode in the tree log.
4532 * At least one parent directory for this inode must exist in the tree
4533 * or be logged already.
4534 *
4535 * Any items from this inode changed by the current transaction are copied
4536 * to the log tree. An extra reference is taken on any extents in this
4537 * file, allowing us to avoid a whole pile of corner cases around logging
4538 * blocks that have been removed from the tree.
4539 *
4540 * See LOG_INODE_ALL and related defines for a description of what inode_only
4541 * does.
4542 *
4543 * This handles both files and directories.
4544 */
Chris Mason12fcfd22009-03-24 10:24:20 -04004545static int btrfs_log_inode(struct btrfs_trans_handle *trans,
Filipe Manana49dae1b2014-09-06 22:34:39 +01004546 struct btrfs_root *root, struct inode *inode,
4547 int inode_only,
4548 const loff_t start,
Filipe Manana8407f552014-09-05 15:14:39 +01004549 const loff_t end,
4550 struct btrfs_log_ctx *ctx)
Chris Masone02119d2008-09-05 16:13:11 -04004551{
4552 struct btrfs_path *path;
4553 struct btrfs_path *dst_path;
4554 struct btrfs_key min_key;
4555 struct btrfs_key max_key;
4556 struct btrfs_root *log = root->log_root;
Chris Mason31ff1cd2008-09-11 16:17:57 -04004557 struct extent_buffer *src = NULL;
Miao Xie827463c2014-01-14 20:31:51 +08004558 LIST_HEAD(logged_list);
Josef Bacik16e75492013-10-22 12:18:51 -04004559 u64 last_extent = 0;
Yan, Zheng4a500fd2010-05-16 10:49:59 -04004560 int err = 0;
Chris Masone02119d2008-09-05 16:13:11 -04004561 int ret;
Chris Mason3a5f1d42008-09-11 15:53:37 -04004562 int nritems;
Chris Mason31ff1cd2008-09-11 16:17:57 -04004563 int ins_start_slot = 0;
4564 int ins_nr;
Josef Bacik5dc562c2012-08-17 13:14:17 -04004565 bool fast_search = false;
Li Zefan33345d012011-04-20 10:31:50 +08004566 u64 ino = btrfs_ino(inode);
Filipe Manana49dae1b2014-09-06 22:34:39 +01004567 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
Filipe Manana1a4bcf42015-02-13 12:30:56 +00004568 u64 logged_isize = 0;
Filipe Mananae4545de2015-06-17 12:49:23 +01004569 bool need_log_inode_item = true;
Chris Masone02119d2008-09-05 16:13:11 -04004570
Chris Masone02119d2008-09-05 16:13:11 -04004571 path = btrfs_alloc_path();
Tsutomu Itoh5df67082011-02-01 09:17:35 +00004572 if (!path)
4573 return -ENOMEM;
Chris Masone02119d2008-09-05 16:13:11 -04004574 dst_path = btrfs_alloc_path();
Tsutomu Itoh5df67082011-02-01 09:17:35 +00004575 if (!dst_path) {
4576 btrfs_free_path(path);
4577 return -ENOMEM;
4578 }
Chris Masone02119d2008-09-05 16:13:11 -04004579
Li Zefan33345d012011-04-20 10:31:50 +08004580 min_key.objectid = ino;
Chris Masone02119d2008-09-05 16:13:11 -04004581 min_key.type = BTRFS_INODE_ITEM_KEY;
4582 min_key.offset = 0;
4583
Li Zefan33345d012011-04-20 10:31:50 +08004584 max_key.objectid = ino;
Chris Mason12fcfd22009-03-24 10:24:20 -04004585
Chris Mason12fcfd22009-03-24 10:24:20 -04004586
Josef Bacik5dc562c2012-08-17 13:14:17 -04004587 /* today the code can only do partial logging of directories */
Miao Xie5269b672012-11-01 07:35:23 +00004588 if (S_ISDIR(inode->i_mode) ||
4589 (!test_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4590 &BTRFS_I(inode)->runtime_flags) &&
4591 inode_only == LOG_INODE_EXISTS))
Chris Masone02119d2008-09-05 16:13:11 -04004592 max_key.type = BTRFS_XATTR_ITEM_KEY;
4593 else
4594 max_key.type = (u8)-1;
4595 max_key.offset = (u64)-1;
4596
Filipe Manana2c2c4522015-01-13 16:40:04 +00004597 /*
4598 * Only run delayed items if we are a dir or a new file.
4599 * Otherwise commit the delayed inode only, which is needed in
4600 * order for the log replay code to mark inodes for link count
4601 * fixup (create temporary BTRFS_TREE_LOG_FIXUP_OBJECTID items).
4602 */
Josef Bacik94edf4a2012-09-25 14:56:25 -04004603 if (S_ISDIR(inode->i_mode) ||
Filipe Manana2c2c4522015-01-13 16:40:04 +00004604 BTRFS_I(inode)->generation > root->fs_info->last_trans_committed)
Josef Bacik94edf4a2012-09-25 14:56:25 -04004605 ret = btrfs_commit_inode_delayed_items(trans, inode);
Filipe Manana2c2c4522015-01-13 16:40:04 +00004606 else
4607 ret = btrfs_commit_inode_delayed_inode(inode);
4608
4609 if (ret) {
4610 btrfs_free_path(path);
4611 btrfs_free_path(dst_path);
4612 return ret;
Miao Xie16cdcec2011-04-22 18:12:22 +08004613 }
4614
Chris Masone02119d2008-09-05 16:13:11 -04004615 mutex_lock(&BTRFS_I(inode)->log_mutex);
4616
Filipe Manana08702952014-11-13 17:00:35 +00004617 btrfs_get_logged_extents(inode, &logged_list, start, end);
Josef Bacik2ab28f32012-10-12 15:27:49 -04004618
Chris Masone02119d2008-09-05 16:13:11 -04004619 /*
4620 * a brute force approach to making sure we get the most uptodate
4621 * copies of everything.
4622 */
4623 if (S_ISDIR(inode->i_mode)) {
4624 int max_key_type = BTRFS_DIR_LOG_INDEX_KEY;
4625
Filipe Manana4f764e52015-02-23 19:53:35 +00004626 if (inode_only == LOG_INODE_EXISTS)
4627 max_key_type = BTRFS_XATTR_ITEM_KEY;
Li Zefan33345d012011-04-20 10:31:50 +08004628 ret = drop_objectid_items(trans, log, path, ino, max_key_type);
Chris Masone02119d2008-09-05 16:13:11 -04004629 } else {
Filipe Manana1a4bcf42015-02-13 12:30:56 +00004630 if (inode_only == LOG_INODE_EXISTS) {
4631 /*
4632 * Make sure the new inode item we write to the log has
4633 * the same isize as the current one (if it exists).
4634 * This is necessary to prevent data loss after log
4635 * replay, and also to prevent doing a wrong expanding
4636 * truncate - for e.g. create file, write 4K into offset
4637 * 0, fsync, write 4K into offset 4096, add hard link,
4638 * fsync some other file (to sync log), power fail - if
4639 * we use the inode's current i_size, after log replay
4640 * we get a 8Kb file, with the last 4Kb extent as a hole
4641 * (zeroes), as if an expanding truncate happened,
4642 * instead of getting a file of 4Kb only.
4643 */
4644 err = logged_inode_size(log, inode, path,
4645 &logged_isize);
4646 if (err)
4647 goto out_unlock;
4648 }
Filipe Mananaa7429942015-02-13 16:56:14 +00004649 if (test_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4650 &BTRFS_I(inode)->runtime_flags)) {
4651 if (inode_only == LOG_INODE_EXISTS) {
Filipe Manana4f764e52015-02-23 19:53:35 +00004652 max_key.type = BTRFS_XATTR_ITEM_KEY;
Filipe Mananaa7429942015-02-13 16:56:14 +00004653 ret = drop_objectid_items(trans, log, path, ino,
4654 max_key.type);
4655 } else {
4656 clear_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4657 &BTRFS_I(inode)->runtime_flags);
4658 clear_bit(BTRFS_INODE_COPY_EVERYTHING,
4659 &BTRFS_I(inode)->runtime_flags);
Chris Mason28ed1342014-12-17 09:41:04 -08004660 while(1) {
4661 ret = btrfs_truncate_inode_items(trans,
4662 log, inode, 0, 0);
4663 if (ret != -EAGAIN)
4664 break;
4665 }
Filipe Mananaa7429942015-02-13 16:56:14 +00004666 }
Filipe Manana4f764e52015-02-23 19:53:35 +00004667 } else if (test_and_clear_bit(BTRFS_INODE_COPY_EVERYTHING,
4668 &BTRFS_I(inode)->runtime_flags) ||
Josef Bacik6cfab852013-11-12 16:25:58 -05004669 inode_only == LOG_INODE_EXISTS) {
Filipe Manana4f764e52015-02-23 19:53:35 +00004670 if (inode_only == LOG_INODE_ALL)
Josef Bacika95249b2012-10-11 16:17:34 -04004671 fast_search = true;
Filipe Manana4f764e52015-02-23 19:53:35 +00004672 max_key.type = BTRFS_XATTR_ITEM_KEY;
Josef Bacika95249b2012-10-11 16:17:34 -04004673 ret = drop_objectid_items(trans, log, path, ino,
4674 max_key.type);
Josef Bacik5dc562c2012-08-17 13:14:17 -04004675 } else {
Liu Bo183f37f2012-11-01 06:38:47 +00004676 if (inode_only == LOG_INODE_ALL)
4677 fast_search = true;
Josef Bacika95249b2012-10-11 16:17:34 -04004678 goto log_extents;
Josef Bacik5dc562c2012-08-17 13:14:17 -04004679 }
Josef Bacika95249b2012-10-11 16:17:34 -04004680
Chris Masone02119d2008-09-05 16:13:11 -04004681 }
Yan, Zheng4a500fd2010-05-16 10:49:59 -04004682 if (ret) {
4683 err = ret;
4684 goto out_unlock;
4685 }
Chris Masone02119d2008-09-05 16:13:11 -04004686
Chris Masond3977122009-01-05 21:25:51 -05004687 while (1) {
Chris Mason31ff1cd2008-09-11 16:17:57 -04004688 ins_nr = 0;
Filipe David Borba Manana6174d3c2013-10-01 16:13:42 +01004689 ret = btrfs_search_forward(root, &min_key,
Eric Sandeende78b512013-01-31 18:21:12 +00004690 path, trans->transid);
Chris Masone02119d2008-09-05 16:13:11 -04004691 if (ret != 0)
4692 break;
Chris Mason3a5f1d42008-09-11 15:53:37 -04004693again:
Chris Mason31ff1cd2008-09-11 16:17:57 -04004694 /* note, ins_nr might be > 0 here, cleanup outside the loop */
Li Zefan33345d012011-04-20 10:31:50 +08004695 if (min_key.objectid != ino)
Chris Masone02119d2008-09-05 16:13:11 -04004696 break;
4697 if (min_key.type > max_key.type)
4698 break;
Chris Mason31ff1cd2008-09-11 16:17:57 -04004699
Filipe Mananae4545de2015-06-17 12:49:23 +01004700 if (min_key.type == BTRFS_INODE_ITEM_KEY)
4701 need_log_inode_item = false;
4702
Filipe Manana1653a3b2016-03-30 23:37:21 +01004703 if ((min_key.type == BTRFS_INODE_REF_KEY ||
4704 min_key.type == BTRFS_INODE_EXTREF_KEY) &&
4705 BTRFS_I(inode)->generation == trans->transid) {
4706 ret = btrfs_check_ref_name_override(path->nodes[0],
4707 path->slots[0],
4708 &min_key, inode);
4709 if (ret < 0) {
4710 err = ret;
4711 goto out_unlock;
4712 } else if (ret > 0) {
4713 err = 1;
4714 btrfs_set_log_full_commit(root->fs_info, trans);
4715 goto out_unlock;
4716 }
4717 }
4718
Filipe Manana36283bf2015-06-20 00:44:51 +01004719 /* Skip xattrs, we log them later with btrfs_log_all_xattrs() */
4720 if (min_key.type == BTRFS_XATTR_ITEM_KEY) {
4721 if (ins_nr == 0)
4722 goto next_slot;
4723 ret = copy_items(trans, inode, dst_path, path,
4724 &last_extent, ins_start_slot,
4725 ins_nr, inode_only, logged_isize);
4726 if (ret < 0) {
4727 err = ret;
4728 goto out_unlock;
4729 }
4730 ins_nr = 0;
4731 if (ret) {
4732 btrfs_release_path(path);
4733 continue;
4734 }
4735 goto next_slot;
4736 }
4737
Chris Masone02119d2008-09-05 16:13:11 -04004738 src = path->nodes[0];
Chris Mason31ff1cd2008-09-11 16:17:57 -04004739 if (ins_nr && ins_start_slot + ins_nr == path->slots[0]) {
4740 ins_nr++;
4741 goto next_slot;
4742 } else if (!ins_nr) {
4743 ins_start_slot = path->slots[0];
4744 ins_nr = 1;
4745 goto next_slot;
Chris Masone02119d2008-09-05 16:13:11 -04004746 }
4747
Josef Bacik16e75492013-10-22 12:18:51 -04004748 ret = copy_items(trans, inode, dst_path, path, &last_extent,
Filipe Manana1a4bcf42015-02-13 12:30:56 +00004749 ins_start_slot, ins_nr, inode_only,
4750 logged_isize);
Josef Bacik16e75492013-10-22 12:18:51 -04004751 if (ret < 0) {
Yan, Zheng4a500fd2010-05-16 10:49:59 -04004752 err = ret;
4753 goto out_unlock;
Rasmus Villemoesa71db862014-06-20 21:51:43 +02004754 }
4755 if (ret) {
Josef Bacik16e75492013-10-22 12:18:51 -04004756 ins_nr = 0;
4757 btrfs_release_path(path);
4758 continue;
Yan, Zheng4a500fd2010-05-16 10:49:59 -04004759 }
Chris Mason31ff1cd2008-09-11 16:17:57 -04004760 ins_nr = 1;
4761 ins_start_slot = path->slots[0];
4762next_slot:
Chris Masone02119d2008-09-05 16:13:11 -04004763
Chris Mason3a5f1d42008-09-11 15:53:37 -04004764 nritems = btrfs_header_nritems(path->nodes[0]);
4765 path->slots[0]++;
4766 if (path->slots[0] < nritems) {
4767 btrfs_item_key_to_cpu(path->nodes[0], &min_key,
4768 path->slots[0]);
4769 goto again;
4770 }
Chris Mason31ff1cd2008-09-11 16:17:57 -04004771 if (ins_nr) {
Josef Bacik16e75492013-10-22 12:18:51 -04004772 ret = copy_items(trans, inode, dst_path, path,
4773 &last_extent, ins_start_slot,
Filipe Manana1a4bcf42015-02-13 12:30:56 +00004774 ins_nr, inode_only, logged_isize);
Josef Bacik16e75492013-10-22 12:18:51 -04004775 if (ret < 0) {
Yan, Zheng4a500fd2010-05-16 10:49:59 -04004776 err = ret;
4777 goto out_unlock;
4778 }
Josef Bacik16e75492013-10-22 12:18:51 -04004779 ret = 0;
Chris Mason31ff1cd2008-09-11 16:17:57 -04004780 ins_nr = 0;
4781 }
David Sterbab3b4aa72011-04-21 01:20:15 +02004782 btrfs_release_path(path);
Chris Mason3a5f1d42008-09-11 15:53:37 -04004783
Filipe David Borba Manana3d41d702013-10-01 17:06:53 +01004784 if (min_key.offset < (u64)-1) {
Chris Masone02119d2008-09-05 16:13:11 -04004785 min_key.offset++;
Filipe David Borba Manana3d41d702013-10-01 17:06:53 +01004786 } else if (min_key.type < max_key.type) {
Chris Masone02119d2008-09-05 16:13:11 -04004787 min_key.type++;
Filipe David Borba Manana3d41d702013-10-01 17:06:53 +01004788 min_key.offset = 0;
4789 } else {
Chris Masone02119d2008-09-05 16:13:11 -04004790 break;
Filipe David Borba Manana3d41d702013-10-01 17:06:53 +01004791 }
Chris Masone02119d2008-09-05 16:13:11 -04004792 }
Chris Mason31ff1cd2008-09-11 16:17:57 -04004793 if (ins_nr) {
Josef Bacik16e75492013-10-22 12:18:51 -04004794 ret = copy_items(trans, inode, dst_path, path, &last_extent,
Filipe Manana1a4bcf42015-02-13 12:30:56 +00004795 ins_start_slot, ins_nr, inode_only,
4796 logged_isize);
Josef Bacik16e75492013-10-22 12:18:51 -04004797 if (ret < 0) {
Yan, Zheng4a500fd2010-05-16 10:49:59 -04004798 err = ret;
4799 goto out_unlock;
4800 }
Josef Bacik16e75492013-10-22 12:18:51 -04004801 ret = 0;
Chris Mason31ff1cd2008-09-11 16:17:57 -04004802 ins_nr = 0;
4803 }
Josef Bacik5dc562c2012-08-17 13:14:17 -04004804
Filipe Manana36283bf2015-06-20 00:44:51 +01004805 btrfs_release_path(path);
4806 btrfs_release_path(dst_path);
4807 err = btrfs_log_all_xattrs(trans, root, inode, path, dst_path);
4808 if (err)
4809 goto out_unlock;
Filipe Mananaa89ca6f2015-06-25 04:17:46 +01004810 if (max_key.type >= BTRFS_EXTENT_DATA_KEY && !fast_search) {
4811 btrfs_release_path(path);
4812 btrfs_release_path(dst_path);
4813 err = btrfs_log_trailing_hole(trans, root, inode, path);
4814 if (err)
4815 goto out_unlock;
4816 }
Josef Bacika95249b2012-10-11 16:17:34 -04004817log_extents:
Josef Bacikf3b15cc2013-07-22 12:54:30 -04004818 btrfs_release_path(path);
4819 btrfs_release_path(dst_path);
Filipe Mananae4545de2015-06-17 12:49:23 +01004820 if (need_log_inode_item) {
4821 err = log_inode_item(trans, log, dst_path, inode);
4822 if (err)
4823 goto out_unlock;
4824 }
Josef Bacik5dc562c2012-08-17 13:14:17 -04004825 if (fast_search) {
Filipe Mananab38ef712014-11-13 17:01:45 +00004826 /*
4827 * Some ordered extents started by fsync might have completed
4828 * before we collected the ordered extents in logged_list, which
4829 * means they're gone, not in our logged_list nor in the inode's
4830 * ordered tree. We want the application/user space to know an
4831 * error happened while attempting to persist file data so that
4832 * it can take proper action. If such error happened, we leave
4833 * without writing to the log tree and the fsync must report the
4834 * file data write error and not commit the current transaction.
4835 */
4836 err = btrfs_inode_check_errors(inode);
4837 if (err) {
4838 ctx->io_err = err;
4839 goto out_unlock;
4840 }
Miao Xie827463c2014-01-14 20:31:51 +08004841 ret = btrfs_log_changed_extents(trans, root, inode, dst_path,
Filipe Manana8407f552014-09-05 15:14:39 +01004842 &logged_list, ctx);
Josef Bacik5dc562c2012-08-17 13:14:17 -04004843 if (ret) {
4844 err = ret;
4845 goto out_unlock;
4846 }
Josef Bacikd006a042013-11-12 20:54:09 -05004847 } else if (inode_only == LOG_INODE_ALL) {
Liu Bo06d3d222012-08-27 10:52:19 -06004848 struct extent_map *em, *n;
4849
Filipe Manana49dae1b2014-09-06 22:34:39 +01004850 write_lock(&em_tree->lock);
4851 /*
4852 * We can't just remove every em if we're called for a ranged
4853 * fsync - that is, one that doesn't cover the whole possible
4854 * file range (0 to LLONG_MAX). This is because we can have
4855 * em's that fall outside the range we're logging and therefore
4856 * their ordered operations haven't completed yet
4857 * (btrfs_finish_ordered_io() not invoked yet). This means we
4858 * didn't get their respective file extent item in the fs/subvol
4859 * tree yet, and need to let the next fast fsync (one which
4860 * consults the list of modified extent maps) find the em so
4861 * that it logs a matching file extent item and waits for the
4862 * respective ordered operation to complete (if it's still
4863 * running).
4864 *
4865 * Removing every em outside the range we're logging would make
4866 * the next fast fsync not log their matching file extent items,
4867 * therefore making us lose data after a log replay.
4868 */
4869 list_for_each_entry_safe(em, n, &em_tree->modified_extents,
4870 list) {
4871 const u64 mod_end = em->mod_start + em->mod_len - 1;
4872
4873 if (em->mod_start >= start && mod_end <= end)
4874 list_del_init(&em->list);
4875 }
4876 write_unlock(&em_tree->lock);
Josef Bacik5dc562c2012-08-17 13:14:17 -04004877 }
4878
Chris Mason9623f9a2008-09-11 17:42:42 -04004879 if (inode_only == LOG_INODE_ALL && S_ISDIR(inode->i_mode)) {
Filipe Manana2f2ff0e2015-03-20 17:19:46 +00004880 ret = log_directory_changes(trans, root, inode, path, dst_path,
4881 ctx);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04004882 if (ret) {
4883 err = ret;
4884 goto out_unlock;
4885 }
Chris Masone02119d2008-09-05 16:13:11 -04004886 }
Filipe Manana49dae1b2014-09-06 22:34:39 +01004887
Filipe Manana2f2ff0e2015-03-20 17:19:46 +00004888 spin_lock(&BTRFS_I(inode)->lock);
Filipe Manana125c4cf92014-09-11 21:22:14 +01004889 BTRFS_I(inode)->logged_trans = trans->transid;
4890 BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->last_sub_trans;
Filipe Manana2f2ff0e2015-03-20 17:19:46 +00004891 spin_unlock(&BTRFS_I(inode)->lock);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04004892out_unlock:
Miao Xie827463c2014-01-14 20:31:51 +08004893 if (unlikely(err))
4894 btrfs_put_logged_extents(&logged_list);
4895 else
4896 btrfs_submit_logged_extents(&logged_list, log);
Chris Masone02119d2008-09-05 16:13:11 -04004897 mutex_unlock(&BTRFS_I(inode)->log_mutex);
4898
4899 btrfs_free_path(path);
4900 btrfs_free_path(dst_path);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04004901 return err;
Chris Masone02119d2008-09-05 16:13:11 -04004902}
4903
Chris Mason12fcfd22009-03-24 10:24:20 -04004904/*
4905 * follow the dentry parent pointers up the chain and see if any
4906 * of the directories in it require a full commit before they can
4907 * be logged. Returns zero if nothing special needs to be done or 1 if
4908 * a full commit is required.
4909 */
4910static noinline int check_parent_dirs_for_sync(struct btrfs_trans_handle *trans,
4911 struct inode *inode,
4912 struct dentry *parent,
4913 struct super_block *sb,
4914 u64 last_committed)
Chris Masone02119d2008-09-05 16:13:11 -04004915{
Chris Mason12fcfd22009-03-24 10:24:20 -04004916 int ret = 0;
4917 struct btrfs_root *root;
Josef Bacik6a912212010-11-20 09:48:00 +00004918 struct dentry *old_parent = NULL;
Josef Bacikde2b5302013-09-11 09:36:30 -04004919 struct inode *orig_inode = inode;
Chris Masone02119d2008-09-05 16:13:11 -04004920
Chris Masonaf4176b2009-03-24 10:24:31 -04004921 /*
4922 * for regular files, if its inode is already on disk, we don't
4923 * have to worry about the parents at all. This is because
4924 * we can use the last_unlink_trans field to record renames
4925 * and other fun in this file.
4926 */
4927 if (S_ISREG(inode->i_mode) &&
4928 BTRFS_I(inode)->generation <= last_committed &&
4929 BTRFS_I(inode)->last_unlink_trans <= last_committed)
4930 goto out;
4931
Chris Mason12fcfd22009-03-24 10:24:20 -04004932 if (!S_ISDIR(inode->i_mode)) {
David Howells2b0143b2015-03-17 22:25:59 +00004933 if (!parent || d_really_is_negative(parent) || sb != d_inode(parent)->i_sb)
Chris Mason12fcfd22009-03-24 10:24:20 -04004934 goto out;
David Howells2b0143b2015-03-17 22:25:59 +00004935 inode = d_inode(parent);
Chris Mason12fcfd22009-03-24 10:24:20 -04004936 }
4937
4938 while (1) {
Josef Bacikde2b5302013-09-11 09:36:30 -04004939 /*
4940 * If we are logging a directory then we start with our inode,
4941 * not our parents inode, so we need to skipp setting the
4942 * logged_trans so that further down in the log code we don't
4943 * think this inode has already been logged.
4944 */
4945 if (inode != orig_inode)
4946 BTRFS_I(inode)->logged_trans = trans->transid;
Chris Mason12fcfd22009-03-24 10:24:20 -04004947 smp_mb();
4948
4949 if (BTRFS_I(inode)->last_unlink_trans > last_committed) {
4950 root = BTRFS_I(inode)->root;
4951
4952 /*
4953 * make sure any commits to the log are forced
4954 * to be full commits
4955 */
Miao Xie995946d2014-04-02 19:51:06 +08004956 btrfs_set_log_full_commit(root->fs_info, trans);
Chris Mason12fcfd22009-03-24 10:24:20 -04004957 ret = 1;
4958 break;
4959 }
4960
David Howells2b0143b2015-03-17 22:25:59 +00004961 if (!parent || d_really_is_negative(parent) || sb != d_inode(parent)->i_sb)
Chris Mason12fcfd22009-03-24 10:24:20 -04004962 break;
4963
Yan, Zheng76dda932009-09-21 16:00:26 -04004964 if (IS_ROOT(parent))
Chris Mason12fcfd22009-03-24 10:24:20 -04004965 break;
4966
Josef Bacik6a912212010-11-20 09:48:00 +00004967 parent = dget_parent(parent);
4968 dput(old_parent);
4969 old_parent = parent;
David Howells2b0143b2015-03-17 22:25:59 +00004970 inode = d_inode(parent);
Chris Mason12fcfd22009-03-24 10:24:20 -04004971
4972 }
Josef Bacik6a912212010-11-20 09:48:00 +00004973 dput(old_parent);
Chris Mason12fcfd22009-03-24 10:24:20 -04004974out:
Chris Masone02119d2008-09-05 16:13:11 -04004975 return ret;
4976}
4977
Filipe Manana2f2ff0e2015-03-20 17:19:46 +00004978struct btrfs_dir_list {
4979 u64 ino;
4980 struct list_head list;
4981};
4982
4983/*
4984 * Log the inodes of the new dentries of a directory. See log_dir_items() for
4985 * details about the why it is needed.
4986 * This is a recursive operation - if an existing dentry corresponds to a
4987 * directory, that directory's new entries are logged too (same behaviour as
4988 * ext3/4, xfs, f2fs, reiserfs, nilfs2). Note that when logging the inodes
4989 * the dentries point to we do not lock their i_mutex, otherwise lockdep
4990 * complains about the following circular lock dependency / possible deadlock:
4991 *
4992 * CPU0 CPU1
4993 * ---- ----
4994 * lock(&type->i_mutex_dir_key#3/2);
4995 * lock(sb_internal#2);
4996 * lock(&type->i_mutex_dir_key#3/2);
4997 * lock(&sb->s_type->i_mutex_key#14);
4998 *
4999 * Where sb_internal is the lock (a counter that works as a lock) acquired by
5000 * sb_start_intwrite() in btrfs_start_transaction().
5001 * Not locking i_mutex of the inodes is still safe because:
5002 *
5003 * 1) For regular files we log with a mode of LOG_INODE_EXISTS. It's possible
5004 * that while logging the inode new references (names) are added or removed
5005 * from the inode, leaving the logged inode item with a link count that does
5006 * not match the number of logged inode reference items. This is fine because
5007 * at log replay time we compute the real number of links and correct the
5008 * link count in the inode item (see replay_one_buffer() and
5009 * link_to_fixup_dir());
5010 *
5011 * 2) For directories we log with a mode of LOG_INODE_ALL. It's possible that
5012 * while logging the inode's items new items with keys BTRFS_DIR_ITEM_KEY and
5013 * BTRFS_DIR_INDEX_KEY are added to fs/subvol tree and the logged inode item
5014 * has a size that doesn't match the sum of the lengths of all the logged
5015 * names. This does not result in a problem because if a dir_item key is
5016 * logged but its matching dir_index key is not logged, at log replay time we
5017 * don't use it to replay the respective name (see replay_one_name()). On the
5018 * other hand if only the dir_index key ends up being logged, the respective
5019 * name is added to the fs/subvol tree with both the dir_item and dir_index
5020 * keys created (see replay_one_name()).
5021 * The directory's inode item with a wrong i_size is not a problem as well,
5022 * since we don't use it at log replay time to set the i_size in the inode
5023 * item of the fs/subvol tree (see overwrite_item()).
5024 */
5025static int log_new_dir_dentries(struct btrfs_trans_handle *trans,
5026 struct btrfs_root *root,
5027 struct inode *start_inode,
5028 struct btrfs_log_ctx *ctx)
5029{
5030 struct btrfs_root *log = root->log_root;
5031 struct btrfs_path *path;
5032 LIST_HEAD(dir_list);
5033 struct btrfs_dir_list *dir_elem;
5034 int ret = 0;
5035
5036 path = btrfs_alloc_path();
5037 if (!path)
5038 return -ENOMEM;
5039
5040 dir_elem = kmalloc(sizeof(*dir_elem), GFP_NOFS);
5041 if (!dir_elem) {
5042 btrfs_free_path(path);
5043 return -ENOMEM;
5044 }
5045 dir_elem->ino = btrfs_ino(start_inode);
5046 list_add_tail(&dir_elem->list, &dir_list);
5047
5048 while (!list_empty(&dir_list)) {
5049 struct extent_buffer *leaf;
5050 struct btrfs_key min_key;
5051 int nritems;
5052 int i;
5053
5054 dir_elem = list_first_entry(&dir_list, struct btrfs_dir_list,
5055 list);
5056 if (ret)
5057 goto next_dir_inode;
5058
5059 min_key.objectid = dir_elem->ino;
5060 min_key.type = BTRFS_DIR_ITEM_KEY;
5061 min_key.offset = 0;
5062again:
5063 btrfs_release_path(path);
5064 ret = btrfs_search_forward(log, &min_key, path, trans->transid);
5065 if (ret < 0) {
5066 goto next_dir_inode;
5067 } else if (ret > 0) {
5068 ret = 0;
5069 goto next_dir_inode;
5070 }
5071
5072process_leaf:
5073 leaf = path->nodes[0];
5074 nritems = btrfs_header_nritems(leaf);
5075 for (i = path->slots[0]; i < nritems; i++) {
5076 struct btrfs_dir_item *di;
5077 struct btrfs_key di_key;
5078 struct inode *di_inode;
5079 struct btrfs_dir_list *new_dir_elem;
5080 int log_mode = LOG_INODE_EXISTS;
5081 int type;
5082
5083 btrfs_item_key_to_cpu(leaf, &min_key, i);
5084 if (min_key.objectid != dir_elem->ino ||
5085 min_key.type != BTRFS_DIR_ITEM_KEY)
5086 goto next_dir_inode;
5087
5088 di = btrfs_item_ptr(leaf, i, struct btrfs_dir_item);
5089 type = btrfs_dir_type(leaf, di);
5090 if (btrfs_dir_transid(leaf, di) < trans->transid &&
5091 type != BTRFS_FT_DIR)
5092 continue;
5093 btrfs_dir_item_key_to_cpu(leaf, di, &di_key);
5094 if (di_key.type == BTRFS_ROOT_ITEM_KEY)
5095 continue;
5096
5097 di_inode = btrfs_iget(root->fs_info->sb, &di_key,
5098 root, NULL);
5099 if (IS_ERR(di_inode)) {
5100 ret = PTR_ERR(di_inode);
5101 goto next_dir_inode;
5102 }
5103
5104 if (btrfs_inode_in_log(di_inode, trans->transid)) {
5105 iput(di_inode);
5106 continue;
5107 }
5108
5109 ctx->log_new_dentries = false;
5110 if (type == BTRFS_FT_DIR)
5111 log_mode = LOG_INODE_ALL;
5112 btrfs_release_path(path);
5113 ret = btrfs_log_inode(trans, root, di_inode,
5114 log_mode, 0, LLONG_MAX, ctx);
5115 iput(di_inode);
5116 if (ret)
5117 goto next_dir_inode;
5118 if (ctx->log_new_dentries) {
5119 new_dir_elem = kmalloc(sizeof(*new_dir_elem),
5120 GFP_NOFS);
5121 if (!new_dir_elem) {
5122 ret = -ENOMEM;
5123 goto next_dir_inode;
5124 }
5125 new_dir_elem->ino = di_key.objectid;
5126 list_add_tail(&new_dir_elem->list, &dir_list);
5127 }
5128 break;
5129 }
5130 if (i == nritems) {
5131 ret = btrfs_next_leaf(log, path);
5132 if (ret < 0) {
5133 goto next_dir_inode;
5134 } else if (ret > 0) {
5135 ret = 0;
5136 goto next_dir_inode;
5137 }
5138 goto process_leaf;
5139 }
5140 if (min_key.offset < (u64)-1) {
5141 min_key.offset++;
5142 goto again;
5143 }
5144next_dir_inode:
5145 list_del(&dir_elem->list);
5146 kfree(dir_elem);
5147 }
5148
5149 btrfs_free_path(path);
5150 return ret;
5151}
5152
Filipe Manana18aa0922015-08-05 16:49:08 +01005153static int btrfs_log_all_parents(struct btrfs_trans_handle *trans,
5154 struct inode *inode,
5155 struct btrfs_log_ctx *ctx)
5156{
5157 int ret;
5158 struct btrfs_path *path;
5159 struct btrfs_key key;
5160 struct btrfs_root *root = BTRFS_I(inode)->root;
5161 const u64 ino = btrfs_ino(inode);
5162
5163 path = btrfs_alloc_path();
5164 if (!path)
5165 return -ENOMEM;
5166 path->skip_locking = 1;
5167 path->search_commit_root = 1;
5168
5169 key.objectid = ino;
5170 key.type = BTRFS_INODE_REF_KEY;
5171 key.offset = 0;
5172 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5173 if (ret < 0)
5174 goto out;
5175
5176 while (true) {
5177 struct extent_buffer *leaf = path->nodes[0];
5178 int slot = path->slots[0];
5179 u32 cur_offset = 0;
5180 u32 item_size;
5181 unsigned long ptr;
5182
5183 if (slot >= btrfs_header_nritems(leaf)) {
5184 ret = btrfs_next_leaf(root, path);
5185 if (ret < 0)
5186 goto out;
5187 else if (ret > 0)
5188 break;
5189 continue;
5190 }
5191
5192 btrfs_item_key_to_cpu(leaf, &key, slot);
5193 /* BTRFS_INODE_EXTREF_KEY is BTRFS_INODE_REF_KEY + 1 */
5194 if (key.objectid != ino || key.type > BTRFS_INODE_EXTREF_KEY)
5195 break;
5196
5197 item_size = btrfs_item_size_nr(leaf, slot);
5198 ptr = btrfs_item_ptr_offset(leaf, slot);
5199 while (cur_offset < item_size) {
5200 struct btrfs_key inode_key;
5201 struct inode *dir_inode;
5202
5203 inode_key.type = BTRFS_INODE_ITEM_KEY;
5204 inode_key.offset = 0;
5205
5206 if (key.type == BTRFS_INODE_EXTREF_KEY) {
5207 struct btrfs_inode_extref *extref;
5208
5209 extref = (struct btrfs_inode_extref *)
5210 (ptr + cur_offset);
5211 inode_key.objectid = btrfs_inode_extref_parent(
5212 leaf, extref);
5213 cur_offset += sizeof(*extref);
5214 cur_offset += btrfs_inode_extref_name_len(leaf,
5215 extref);
5216 } else {
5217 inode_key.objectid = key.offset;
5218 cur_offset = item_size;
5219 }
5220
5221 dir_inode = btrfs_iget(root->fs_info->sb, &inode_key,
5222 root, NULL);
5223 /* If parent inode was deleted, skip it. */
5224 if (IS_ERR(dir_inode))
5225 continue;
5226
5227 ret = btrfs_log_inode(trans, root, dir_inode,
5228 LOG_INODE_ALL, 0, LLONG_MAX, ctx);
5229 iput(dir_inode);
5230 if (ret)
5231 goto out;
5232 }
5233 path->slots[0]++;
5234 }
5235 ret = 0;
5236out:
5237 btrfs_free_path(path);
5238 return ret;
5239}
5240
Chris Masone02119d2008-09-05 16:13:11 -04005241/*
5242 * helper function around btrfs_log_inode to make sure newly created
5243 * parent directories also end up in the log. A minimal inode and backref
5244 * only logging is done of any parent directories that are older than
5245 * the last committed transaction
5246 */
Eric Sandeen48a3b632013-04-25 20:41:01 +00005247static int btrfs_log_inode_parent(struct btrfs_trans_handle *trans,
5248 struct btrfs_root *root, struct inode *inode,
Filipe Manana49dae1b2014-09-06 22:34:39 +01005249 struct dentry *parent,
5250 const loff_t start,
5251 const loff_t end,
5252 int exists_only,
Miao Xie8b050d32014-02-20 18:08:58 +08005253 struct btrfs_log_ctx *ctx)
Chris Masone02119d2008-09-05 16:13:11 -04005254{
Chris Mason12fcfd22009-03-24 10:24:20 -04005255 int inode_only = exists_only ? LOG_INODE_EXISTS : LOG_INODE_ALL;
Chris Masone02119d2008-09-05 16:13:11 -04005256 struct super_block *sb;
Josef Bacik6a912212010-11-20 09:48:00 +00005257 struct dentry *old_parent = NULL;
Chris Mason12fcfd22009-03-24 10:24:20 -04005258 int ret = 0;
5259 u64 last_committed = root->fs_info->last_trans_committed;
Filipe Manana2f2ff0e2015-03-20 17:19:46 +00005260 bool log_dentries = false;
5261 struct inode *orig_inode = inode;
Chris Mason12fcfd22009-03-24 10:24:20 -04005262
5263 sb = inode->i_sb;
5264
Sage Weil3a5e1402009-04-02 16:49:40 -04005265 if (btrfs_test_opt(root, NOTREELOG)) {
5266 ret = 1;
5267 goto end_no_trans;
5268 }
5269
Miao Xie995946d2014-04-02 19:51:06 +08005270 /*
5271 * The prev transaction commit doesn't complete, we need do
5272 * full commit by ourselves.
5273 */
Chris Mason12fcfd22009-03-24 10:24:20 -04005274 if (root->fs_info->last_trans_log_full_commit >
5275 root->fs_info->last_trans_committed) {
5276 ret = 1;
5277 goto end_no_trans;
5278 }
5279
Yan, Zheng76dda932009-09-21 16:00:26 -04005280 if (root != BTRFS_I(inode)->root ||
5281 btrfs_root_refs(&root->root_item) == 0) {
5282 ret = 1;
5283 goto end_no_trans;
5284 }
5285
Chris Mason12fcfd22009-03-24 10:24:20 -04005286 ret = check_parent_dirs_for_sync(trans, inode, parent,
5287 sb, last_committed);
5288 if (ret)
5289 goto end_no_trans;
Chris Masone02119d2008-09-05 16:13:11 -04005290
Josef Bacik22ee6982012-05-29 16:57:49 -04005291 if (btrfs_inode_in_log(inode, trans->transid)) {
Chris Mason257c62e2009-10-13 13:21:08 -04005292 ret = BTRFS_NO_LOG_SYNC;
5293 goto end_no_trans;
5294 }
5295
Miao Xie8b050d32014-02-20 18:08:58 +08005296 ret = start_log_trans(trans, root, ctx);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04005297 if (ret)
Miao Xiee87ac132014-02-20 18:08:53 +08005298 goto end_no_trans;
Chris Mason12fcfd22009-03-24 10:24:20 -04005299
Filipe Manana8407f552014-09-05 15:14:39 +01005300 ret = btrfs_log_inode(trans, root, inode, inode_only, start, end, ctx);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04005301 if (ret)
5302 goto end_trans;
Chris Mason12fcfd22009-03-24 10:24:20 -04005303
Chris Masonaf4176b2009-03-24 10:24:31 -04005304 /*
5305 * for regular files, if its inode is already on disk, we don't
5306 * have to worry about the parents at all. This is because
5307 * we can use the last_unlink_trans field to record renames
5308 * and other fun in this file.
5309 */
5310 if (S_ISREG(inode->i_mode) &&
5311 BTRFS_I(inode)->generation <= last_committed &&
Yan, Zheng4a500fd2010-05-16 10:49:59 -04005312 BTRFS_I(inode)->last_unlink_trans <= last_committed) {
5313 ret = 0;
5314 goto end_trans;
5315 }
Chris Masonaf4176b2009-03-24 10:24:31 -04005316
Filipe Manana2f2ff0e2015-03-20 17:19:46 +00005317 if (S_ISDIR(inode->i_mode) && ctx && ctx->log_new_dentries)
5318 log_dentries = true;
5319
Filipe Manana18aa0922015-08-05 16:49:08 +01005320 /*
5321 * On unlink we must make sure all our current and old parent directores
5322 * inodes are fully logged. This is to prevent leaving dangling
5323 * directory index entries in directories that were our parents but are
5324 * not anymore. Not doing this results in old parent directory being
5325 * impossible to delete after log replay (rmdir will always fail with
5326 * error -ENOTEMPTY).
5327 *
5328 * Example 1:
5329 *
5330 * mkdir testdir
5331 * touch testdir/foo
5332 * ln testdir/foo testdir/bar
5333 * sync
5334 * unlink testdir/bar
5335 * xfs_io -c fsync testdir/foo
5336 * <power failure>
5337 * mount fs, triggers log replay
5338 *
5339 * If we don't log the parent directory (testdir), after log replay the
5340 * directory still has an entry pointing to the file inode using the bar
5341 * name, but a matching BTRFS_INODE_[REF|EXTREF]_KEY does not exist and
5342 * the file inode has a link count of 1.
5343 *
5344 * Example 2:
5345 *
5346 * mkdir testdir
5347 * touch foo
5348 * ln foo testdir/foo2
5349 * ln foo testdir/foo3
5350 * sync
5351 * unlink testdir/foo3
5352 * xfs_io -c fsync foo
5353 * <power failure>
5354 * mount fs, triggers log replay
5355 *
5356 * Similar as the first example, after log replay the parent directory
5357 * testdir still has an entry pointing to the inode file with name foo3
5358 * but the file inode does not have a matching BTRFS_INODE_REF_KEY item
5359 * and has a link count of 2.
5360 */
5361 if (BTRFS_I(inode)->last_unlink_trans > last_committed) {
5362 ret = btrfs_log_all_parents(trans, orig_inode, ctx);
5363 if (ret)
5364 goto end_trans;
5365 }
5366
Chris Masond3977122009-01-05 21:25:51 -05005367 while (1) {
David Howells2b0143b2015-03-17 22:25:59 +00005368 if (!parent || d_really_is_negative(parent) || sb != d_inode(parent)->i_sb)
Chris Masone02119d2008-09-05 16:13:11 -04005369 break;
5370
David Howells2b0143b2015-03-17 22:25:59 +00005371 inode = d_inode(parent);
Yan, Zheng76dda932009-09-21 16:00:26 -04005372 if (root != BTRFS_I(inode)->root)
5373 break;
5374
Filipe Manana18aa0922015-08-05 16:49:08 +01005375 if (BTRFS_I(inode)->generation > last_committed) {
5376 ret = btrfs_log_inode(trans, root, inode,
5377 LOG_INODE_EXISTS,
Filipe Manana8407f552014-09-05 15:14:39 +01005378 0, LLONG_MAX, ctx);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04005379 if (ret)
5380 goto end_trans;
Chris Mason12fcfd22009-03-24 10:24:20 -04005381 }
Yan, Zheng76dda932009-09-21 16:00:26 -04005382 if (IS_ROOT(parent))
Chris Masone02119d2008-09-05 16:13:11 -04005383 break;
Chris Mason12fcfd22009-03-24 10:24:20 -04005384
Josef Bacik6a912212010-11-20 09:48:00 +00005385 parent = dget_parent(parent);
5386 dput(old_parent);
5387 old_parent = parent;
Chris Masone02119d2008-09-05 16:13:11 -04005388 }
Filipe Manana2f2ff0e2015-03-20 17:19:46 +00005389 if (log_dentries)
5390 ret = log_new_dir_dentries(trans, root, orig_inode, ctx);
5391 else
5392 ret = 0;
Yan, Zheng4a500fd2010-05-16 10:49:59 -04005393end_trans:
Josef Bacik6a912212010-11-20 09:48:00 +00005394 dput(old_parent);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04005395 if (ret < 0) {
Miao Xie995946d2014-04-02 19:51:06 +08005396 btrfs_set_log_full_commit(root->fs_info, trans);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04005397 ret = 1;
5398 }
Miao Xie8b050d32014-02-20 18:08:58 +08005399
5400 if (ret)
5401 btrfs_remove_log_ctx(root, ctx);
Chris Mason12fcfd22009-03-24 10:24:20 -04005402 btrfs_end_log_trans(root);
5403end_no_trans:
5404 return ret;
Chris Masone02119d2008-09-05 16:13:11 -04005405}
5406
5407/*
5408 * it is not safe to log dentry if the chunk root has added new
5409 * chunks. This returns 0 if the dentry was logged, and 1 otherwise.
5410 * If this returns 1, you must commit the transaction to safely get your
5411 * data on disk.
5412 */
5413int btrfs_log_dentry_safe(struct btrfs_trans_handle *trans,
Miao Xie8b050d32014-02-20 18:08:58 +08005414 struct btrfs_root *root, struct dentry *dentry,
Filipe Manana49dae1b2014-09-06 22:34:39 +01005415 const loff_t start,
5416 const loff_t end,
Miao Xie8b050d32014-02-20 18:08:58 +08005417 struct btrfs_log_ctx *ctx)
Chris Masone02119d2008-09-05 16:13:11 -04005418{
Josef Bacik6a912212010-11-20 09:48:00 +00005419 struct dentry *parent = dget_parent(dentry);
5420 int ret;
5421
David Howells2b0143b2015-03-17 22:25:59 +00005422 ret = btrfs_log_inode_parent(trans, root, d_inode(dentry), parent,
Filipe Manana49dae1b2014-09-06 22:34:39 +01005423 start, end, 0, ctx);
Josef Bacik6a912212010-11-20 09:48:00 +00005424 dput(parent);
5425
5426 return ret;
Chris Masone02119d2008-09-05 16:13:11 -04005427}
5428
5429/*
5430 * should be called during mount to recover any replay any log trees
5431 * from the FS
5432 */
5433int btrfs_recover_log_trees(struct btrfs_root *log_root_tree)
5434{
5435 int ret;
5436 struct btrfs_path *path;
5437 struct btrfs_trans_handle *trans;
5438 struct btrfs_key key;
5439 struct btrfs_key found_key;
5440 struct btrfs_key tmp_key;
5441 struct btrfs_root *log;
5442 struct btrfs_fs_info *fs_info = log_root_tree->fs_info;
5443 struct walk_control wc = {
5444 .process_func = process_one_buffer,
5445 .stage = 0,
5446 };
5447
Chris Masone02119d2008-09-05 16:13:11 -04005448 path = btrfs_alloc_path();
Tsutomu Itohdb5b4932011-03-23 08:14:16 +00005449 if (!path)
5450 return -ENOMEM;
5451
5452 fs_info->log_root_recovering = 1;
Chris Masone02119d2008-09-05 16:13:11 -04005453
Yan, Zheng4a500fd2010-05-16 10:49:59 -04005454 trans = btrfs_start_transaction(fs_info->tree_root, 0);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01005455 if (IS_ERR(trans)) {
5456 ret = PTR_ERR(trans);
5457 goto error;
5458 }
Chris Masone02119d2008-09-05 16:13:11 -04005459
5460 wc.trans = trans;
5461 wc.pin = 1;
5462
Tsutomu Itohdb5b4932011-03-23 08:14:16 +00005463 ret = walk_log_tree(trans, log_root_tree, &wc);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01005464 if (ret) {
Anand Jaina4553fe2015-09-25 14:43:01 +08005465 btrfs_std_error(fs_info, ret, "Failed to pin buffers while "
Jeff Mahoney79787ea2012-03-12 16:03:00 +01005466 "recovering log root tree.");
5467 goto error;
5468 }
Chris Masone02119d2008-09-05 16:13:11 -04005469
5470again:
5471 key.objectid = BTRFS_TREE_LOG_OBJECTID;
5472 key.offset = (u64)-1;
David Sterba962a2982014-06-04 18:41:45 +02005473 key.type = BTRFS_ROOT_ITEM_KEY;
Chris Masone02119d2008-09-05 16:13:11 -04005474
Chris Masond3977122009-01-05 21:25:51 -05005475 while (1) {
Chris Masone02119d2008-09-05 16:13:11 -04005476 ret = btrfs_search_slot(NULL, log_root_tree, &key, path, 0, 0);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01005477
5478 if (ret < 0) {
Anand Jaina4553fe2015-09-25 14:43:01 +08005479 btrfs_std_error(fs_info, ret,
Jeff Mahoney79787ea2012-03-12 16:03:00 +01005480 "Couldn't find tree log root.");
5481 goto error;
5482 }
Chris Masone02119d2008-09-05 16:13:11 -04005483 if (ret > 0) {
5484 if (path->slots[0] == 0)
5485 break;
5486 path->slots[0]--;
5487 }
5488 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
5489 path->slots[0]);
David Sterbab3b4aa72011-04-21 01:20:15 +02005490 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04005491 if (found_key.objectid != BTRFS_TREE_LOG_OBJECTID)
5492 break;
5493
Miao Xiecb517ea2013-05-15 07:48:19 +00005494 log = btrfs_read_fs_root(log_root_tree, &found_key);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01005495 if (IS_ERR(log)) {
5496 ret = PTR_ERR(log);
Anand Jaina4553fe2015-09-25 14:43:01 +08005497 btrfs_std_error(fs_info, ret,
Jeff Mahoney79787ea2012-03-12 16:03:00 +01005498 "Couldn't read tree log root.");
5499 goto error;
5500 }
Chris Masone02119d2008-09-05 16:13:11 -04005501
5502 tmp_key.objectid = found_key.offset;
5503 tmp_key.type = BTRFS_ROOT_ITEM_KEY;
5504 tmp_key.offset = (u64)-1;
5505
5506 wc.replay_dest = btrfs_read_fs_root_no_name(fs_info, &tmp_key);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01005507 if (IS_ERR(wc.replay_dest)) {
5508 ret = PTR_ERR(wc.replay_dest);
Josef Bacikb50c6e22013-04-25 15:55:30 -04005509 free_extent_buffer(log->node);
5510 free_extent_buffer(log->commit_root);
5511 kfree(log);
Anand Jaina4553fe2015-09-25 14:43:01 +08005512 btrfs_std_error(fs_info, ret, "Couldn't read target root "
Jeff Mahoney79787ea2012-03-12 16:03:00 +01005513 "for tree log recovery.");
5514 goto error;
5515 }
Chris Masone02119d2008-09-05 16:13:11 -04005516
Yan Zheng07d400a2009-01-06 11:42:00 -05005517 wc.replay_dest->log_root = log;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005518 btrfs_record_root_in_trans(trans, wc.replay_dest);
Chris Masone02119d2008-09-05 16:13:11 -04005519 ret = walk_log_tree(trans, log, &wc);
Chris Masone02119d2008-09-05 16:13:11 -04005520
Josef Bacikb50c6e22013-04-25 15:55:30 -04005521 if (!ret && wc.stage == LOG_WALK_REPLAY_ALL) {
Chris Masone02119d2008-09-05 16:13:11 -04005522 ret = fixup_inode_link_counts(trans, wc.replay_dest,
5523 path);
Chris Masone02119d2008-09-05 16:13:11 -04005524 }
Chris Masone02119d2008-09-05 16:13:11 -04005525
5526 key.offset = found_key.offset - 1;
Yan Zheng07d400a2009-01-06 11:42:00 -05005527 wc.replay_dest->log_root = NULL;
Chris Masone02119d2008-09-05 16:13:11 -04005528 free_extent_buffer(log->node);
Chris Masonb263c2c2009-06-11 11:24:47 -04005529 free_extent_buffer(log->commit_root);
Chris Masone02119d2008-09-05 16:13:11 -04005530 kfree(log);
5531
Josef Bacikb50c6e22013-04-25 15:55:30 -04005532 if (ret)
5533 goto error;
5534
Chris Masone02119d2008-09-05 16:13:11 -04005535 if (found_key.offset == 0)
5536 break;
5537 }
David Sterbab3b4aa72011-04-21 01:20:15 +02005538 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04005539
5540 /* step one is to pin it all, step two is to replay just inodes */
5541 if (wc.pin) {
5542 wc.pin = 0;
5543 wc.process_func = replay_one_buffer;
5544 wc.stage = LOG_WALK_REPLAY_INODES;
5545 goto again;
5546 }
5547 /* step three is to replay everything */
5548 if (wc.stage < LOG_WALK_REPLAY_ALL) {
5549 wc.stage++;
5550 goto again;
5551 }
5552
5553 btrfs_free_path(path);
5554
Josef Bacikabefa552013-04-24 16:40:05 -04005555 /* step 4: commit the transaction, which also unpins the blocks */
5556 ret = btrfs_commit_transaction(trans, fs_info->tree_root);
5557 if (ret)
5558 return ret;
5559
Chris Masone02119d2008-09-05 16:13:11 -04005560 free_extent_buffer(log_root_tree->node);
5561 log_root_tree->log_root = NULL;
5562 fs_info->log_root_recovering = 0;
Chris Masone02119d2008-09-05 16:13:11 -04005563 kfree(log_root_tree);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01005564
Josef Bacikabefa552013-04-24 16:40:05 -04005565 return 0;
Jeff Mahoney79787ea2012-03-12 16:03:00 +01005566error:
Josef Bacikb50c6e22013-04-25 15:55:30 -04005567 if (wc.trans)
5568 btrfs_end_transaction(wc.trans, fs_info->tree_root);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01005569 btrfs_free_path(path);
5570 return ret;
Chris Masone02119d2008-09-05 16:13:11 -04005571}
Chris Mason12fcfd22009-03-24 10:24:20 -04005572
5573/*
5574 * there are some corner cases where we want to force a full
5575 * commit instead of allowing a directory to be logged.
5576 *
5577 * They revolve around files there were unlinked from the directory, and
5578 * this function updates the parent directory so that a full commit is
5579 * properly done if it is fsync'd later after the unlinks are done.
5580 */
5581void btrfs_record_unlink_dir(struct btrfs_trans_handle *trans,
5582 struct inode *dir, struct inode *inode,
5583 int for_rename)
5584{
5585 /*
Chris Masonaf4176b2009-03-24 10:24:31 -04005586 * when we're logging a file, if it hasn't been renamed
5587 * or unlinked, and its inode is fully committed on disk,
5588 * we don't have to worry about walking up the directory chain
5589 * to log its parents.
5590 *
5591 * So, we use the last_unlink_trans field to put this transid
5592 * into the file. When the file is logged we check it and
5593 * don't log the parents if the file is fully on disk.
5594 */
5595 if (S_ISREG(inode->i_mode))
5596 BTRFS_I(inode)->last_unlink_trans = trans->transid;
5597
5598 /*
Chris Mason12fcfd22009-03-24 10:24:20 -04005599 * if this directory was already logged any new
5600 * names for this file/dir will get recorded
5601 */
5602 smp_mb();
5603 if (BTRFS_I(dir)->logged_trans == trans->transid)
5604 return;
5605
5606 /*
5607 * if the inode we're about to unlink was logged,
5608 * the log will be properly updated for any new names
5609 */
5610 if (BTRFS_I(inode)->logged_trans == trans->transid)
5611 return;
5612
5613 /*
5614 * when renaming files across directories, if the directory
5615 * there we're unlinking from gets fsync'd later on, there's
5616 * no way to find the destination directory later and fsync it
5617 * properly. So, we have to be conservative and force commits
5618 * so the new name gets discovered.
5619 */
5620 if (for_rename)
5621 goto record;
5622
5623 /* we can safely do the unlink without any special recording */
5624 return;
5625
5626record:
5627 BTRFS_I(dir)->last_unlink_trans = trans->transid;
5628}
5629
5630/*
5631 * Call this after adding a new name for a file and it will properly
5632 * update the log to reflect the new name.
5633 *
5634 * It will return zero if all goes well, and it will return 1 if a
5635 * full transaction commit is required.
5636 */
5637int btrfs_log_new_name(struct btrfs_trans_handle *trans,
5638 struct inode *inode, struct inode *old_dir,
5639 struct dentry *parent)
5640{
5641 struct btrfs_root * root = BTRFS_I(inode)->root;
5642
5643 /*
Chris Masonaf4176b2009-03-24 10:24:31 -04005644 * this will force the logging code to walk the dentry chain
5645 * up for the file
5646 */
5647 if (S_ISREG(inode->i_mode))
5648 BTRFS_I(inode)->last_unlink_trans = trans->transid;
5649
5650 /*
Chris Mason12fcfd22009-03-24 10:24:20 -04005651 * if this inode hasn't been logged and directory we're renaming it
5652 * from hasn't been logged, we don't need to log it
5653 */
5654 if (BTRFS_I(inode)->logged_trans <=
5655 root->fs_info->last_trans_committed &&
5656 (!old_dir || BTRFS_I(old_dir)->logged_trans <=
5657 root->fs_info->last_trans_committed))
5658 return 0;
5659
Filipe Manana49dae1b2014-09-06 22:34:39 +01005660 return btrfs_log_inode_parent(trans, root, inode, parent, 0,
5661 LLONG_MAX, 1, NULL);
Chris Mason12fcfd22009-03-24 10:24:20 -04005662}
5663