blob: 5be45c12dd715acd0b674b7af3ba1858357e4d6e [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{
Miao Xie8b050d32014-02-20 18:08:58 +0800143 int index;
Chris Masone02119d2008-09-05 16:13:11 -0400144 int ret;
Yan Zheng7237f1832009-01-21 12:54:03 -0500145
146 mutex_lock(&root->log_mutex);
147 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 }
Josef Bacikff782e02009-10-08 15:30:04 -0400152 if (!root->log_start_pid) {
153 root->log_start_pid = current->pid;
Miao Xie27cdeb72014-04-02 19:51:05 +0800154 clear_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state);
Josef Bacikff782e02009-10-08 15:30:04 -0400155 } else if (root->log_start_pid != current->pid) {
Miao Xie27cdeb72014-04-02 19:51:05 +0800156 set_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state);
Josef Bacikff782e02009-10-08 15:30:04 -0400157 }
158
Miao Xie2ecb7922012-09-06 04:04:27 -0600159 atomic_inc(&root->log_batch);
Yan Zheng7237f1832009-01-21 12:54:03 -0500160 atomic_inc(&root->log_writers);
Miao Xie8b050d32014-02-20 18:08:58 +0800161 if (ctx) {
162 index = root->log_transid % 2;
163 list_add_tail(&ctx->list, &root->log_ctxs[index]);
Miao Xied1433de2014-02-20 18:08:59 +0800164 ctx->log_transid = root->log_transid;
Miao Xie8b050d32014-02-20 18:08:58 +0800165 }
Yan Zheng7237f1832009-01-21 12:54:03 -0500166 mutex_unlock(&root->log_mutex);
167 return 0;
168 }
Miao Xiee87ac132014-02-20 18:08:53 +0800169
170 ret = 0;
Chris Masone02119d2008-09-05 16:13:11 -0400171 mutex_lock(&root->fs_info->tree_log_mutex);
Miao Xiee87ac132014-02-20 18:08:53 +0800172 if (!root->fs_info->log_root_tree)
Chris Masone02119d2008-09-05 16:13:11 -0400173 ret = btrfs_init_log_root_tree(trans, root->fs_info);
Miao Xiee87ac132014-02-20 18:08:53 +0800174 mutex_unlock(&root->fs_info->tree_log_mutex);
175 if (ret)
176 goto out;
177
178 if (!root->log_root) {
Chris Masone02119d2008-09-05 16:13:11 -0400179 ret = btrfs_add_log_tree(trans, root);
Yan, Zheng4a500fd2010-05-16 10:49:59 -0400180 if (ret)
Miao Xiee87ac132014-02-20 18:08:53 +0800181 goto out;
Chris Masone02119d2008-09-05 16:13:11 -0400182 }
Miao Xie27cdeb72014-04-02 19:51:05 +0800183 clear_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state);
Miao Xiee87ac132014-02-20 18:08:53 +0800184 root->log_start_pid = current->pid;
Miao Xie2ecb7922012-09-06 04:04:27 -0600185 atomic_inc(&root->log_batch);
Yan Zheng7237f1832009-01-21 12:54:03 -0500186 atomic_inc(&root->log_writers);
Miao Xie8b050d32014-02-20 18:08:58 +0800187 if (ctx) {
188 index = root->log_transid % 2;
189 list_add_tail(&ctx->list, &root->log_ctxs[index]);
Miao Xied1433de2014-02-20 18:08:59 +0800190 ctx->log_transid = root->log_transid;
Miao Xie8b050d32014-02-20 18:08:58 +0800191 }
Miao Xiee87ac132014-02-20 18:08:53 +0800192out:
Yan Zheng7237f1832009-01-21 12:54:03 -0500193 mutex_unlock(&root->log_mutex);
Miao Xiee87ac132014-02-20 18:08:53 +0800194 return ret;
Chris Masone02119d2008-09-05 16:13:11 -0400195}
196
197/*
198 * returns 0 if there was a log transaction running and we were able
199 * to join, or returns -ENOENT if there were not transactions
200 * in progress
201 */
202static int join_running_log_trans(struct btrfs_root *root)
203{
204 int ret = -ENOENT;
205
206 smp_mb();
207 if (!root->log_root)
208 return -ENOENT;
209
Yan Zheng7237f1832009-01-21 12:54:03 -0500210 mutex_lock(&root->log_mutex);
Chris Masone02119d2008-09-05 16:13:11 -0400211 if (root->log_root) {
212 ret = 0;
Yan Zheng7237f1832009-01-21 12:54:03 -0500213 atomic_inc(&root->log_writers);
Chris Masone02119d2008-09-05 16:13:11 -0400214 }
Yan Zheng7237f1832009-01-21 12:54:03 -0500215 mutex_unlock(&root->log_mutex);
Chris Masone02119d2008-09-05 16:13:11 -0400216 return ret;
217}
218
219/*
Chris Mason12fcfd22009-03-24 10:24:20 -0400220 * This either makes the current running log transaction wait
221 * until you call btrfs_end_log_trans() or it makes any future
222 * log transactions wait until you call btrfs_end_log_trans()
223 */
224int btrfs_pin_log_trans(struct btrfs_root *root)
225{
226 int ret = -ENOENT;
227
228 mutex_lock(&root->log_mutex);
229 atomic_inc(&root->log_writers);
230 mutex_unlock(&root->log_mutex);
231 return ret;
232}
233
234/*
Chris Masone02119d2008-09-05 16:13:11 -0400235 * indicate we're done making changes to the log tree
236 * and wake up anyone waiting to do a sync
237 */
Jeff Mahoney143bede2012-03-01 14:56:26 +0100238void btrfs_end_log_trans(struct btrfs_root *root)
Chris Masone02119d2008-09-05 16:13:11 -0400239{
Yan Zheng7237f1832009-01-21 12:54:03 -0500240 if (atomic_dec_and_test(&root->log_writers)) {
241 smp_mb();
242 if (waitqueue_active(&root->log_writer_wait))
243 wake_up(&root->log_writer_wait);
244 }
Chris Masone02119d2008-09-05 16:13:11 -0400245}
246
247
248/*
249 * the walk control struct is used to pass state down the chain when
250 * processing the log tree. The stage field tells us which part
251 * of the log tree processing we are currently doing. The others
252 * are state fields used for that specific part
253 */
254struct walk_control {
255 /* should we free the extent on disk when done? This is used
256 * at transaction commit time while freeing a log tree
257 */
258 int free;
259
260 /* should we write out the extent buffer? This is used
261 * while flushing the log tree to disk during a sync
262 */
263 int write;
264
265 /* should we wait for the extent buffer io to finish? Also used
266 * while flushing the log tree to disk for a sync
267 */
268 int wait;
269
270 /* pin only walk, we record which extents on disk belong to the
271 * log trees
272 */
273 int pin;
274
275 /* what stage of the replay code we're currently in */
276 int stage;
277
278 /* the root we are currently replaying */
279 struct btrfs_root *replay_dest;
280
281 /* the trans handle for the current replay */
282 struct btrfs_trans_handle *trans;
283
284 /* the function that gets used to process blocks we find in the
285 * tree. Note the extent_buffer might not be up to date when it is
286 * passed in, and it must be checked or read if you need the data
287 * inside it
288 */
289 int (*process_func)(struct btrfs_root *log, struct extent_buffer *eb,
290 struct walk_control *wc, u64 gen);
291};
292
293/*
294 * process_func used to pin down extents, write them or wait on them
295 */
296static int process_one_buffer(struct btrfs_root *log,
297 struct extent_buffer *eb,
298 struct walk_control *wc, u64 gen)
299{
Josef Bacikb50c6e22013-04-25 15:55:30 -0400300 int ret = 0;
Chris Masone02119d2008-09-05 16:13:11 -0400301
Josef Bacik8c2a1a32013-06-06 13:19:32 -0400302 /*
303 * If this fs is mixed then we need to be able to process the leaves to
304 * pin down any logged extents, so we have to read the block.
305 */
306 if (btrfs_fs_incompat(log->fs_info, MIXED_GROUPS)) {
307 ret = btrfs_read_buffer(eb, gen);
308 if (ret)
309 return ret;
310 }
311
Josef Bacikb50c6e22013-04-25 15:55:30 -0400312 if (wc->pin)
313 ret = btrfs_pin_extent_for_log_replay(log->fs_info->extent_root,
314 eb->start, eb->len);
315
316 if (!ret && btrfs_buffer_uptodate(eb, gen, 0)) {
Josef Bacik8c2a1a32013-06-06 13:19:32 -0400317 if (wc->pin && btrfs_header_level(eb) == 0)
318 ret = btrfs_exclude_logged_extents(log, eb);
Chris Masone02119d2008-09-05 16:13:11 -0400319 if (wc->write)
320 btrfs_write_tree_block(eb);
321 if (wc->wait)
322 btrfs_wait_tree_block_writeback(eb);
323 }
Josef Bacikb50c6e22013-04-25 15:55:30 -0400324 return ret;
Chris Masone02119d2008-09-05 16:13:11 -0400325}
326
327/*
328 * Item overwrite used by replay and tree logging. eb, slot and key all refer
329 * to the src data we are copying out.
330 *
331 * root is the tree we are copying into, and path is a scratch
332 * path for use in this function (it should be released on entry and
333 * will be released on exit).
334 *
335 * If the key is already in the destination tree the existing item is
336 * overwritten. If the existing item isn't big enough, it is extended.
337 * If it is too large, it is truncated.
338 *
339 * If the key isn't in the destination yet, a new item is inserted.
340 */
341static noinline int overwrite_item(struct btrfs_trans_handle *trans,
342 struct btrfs_root *root,
343 struct btrfs_path *path,
344 struct extent_buffer *eb, int slot,
345 struct btrfs_key *key)
346{
347 int ret;
348 u32 item_size;
349 u64 saved_i_size = 0;
350 int save_old_i_size = 0;
351 unsigned long src_ptr;
352 unsigned long dst_ptr;
353 int overwrite_root = 0;
Josef Bacik4bc4bee2013-04-05 20:50:09 +0000354 bool inode_item = key->type == BTRFS_INODE_ITEM_KEY;
Chris Masone02119d2008-09-05 16:13:11 -0400355
356 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
357 overwrite_root = 1;
358
359 item_size = btrfs_item_size_nr(eb, slot);
360 src_ptr = btrfs_item_ptr_offset(eb, slot);
361
362 /* look for the key in the destination tree */
363 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
Josef Bacik4bc4bee2013-04-05 20:50:09 +0000364 if (ret < 0)
365 return ret;
366
Chris Masone02119d2008-09-05 16:13:11 -0400367 if (ret == 0) {
368 char *src_copy;
369 char *dst_copy;
370 u32 dst_size = btrfs_item_size_nr(path->nodes[0],
371 path->slots[0]);
372 if (dst_size != item_size)
373 goto insert;
374
375 if (item_size == 0) {
David Sterbab3b4aa72011-04-21 01:20:15 +0200376 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -0400377 return 0;
378 }
379 dst_copy = kmalloc(item_size, GFP_NOFS);
380 src_copy = kmalloc(item_size, GFP_NOFS);
liubo2a29edc2011-01-26 06:22:08 +0000381 if (!dst_copy || !src_copy) {
David Sterbab3b4aa72011-04-21 01:20:15 +0200382 btrfs_release_path(path);
liubo2a29edc2011-01-26 06:22:08 +0000383 kfree(dst_copy);
384 kfree(src_copy);
385 return -ENOMEM;
386 }
Chris Masone02119d2008-09-05 16:13:11 -0400387
388 read_extent_buffer(eb, src_copy, src_ptr, item_size);
389
390 dst_ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
391 read_extent_buffer(path->nodes[0], dst_copy, dst_ptr,
392 item_size);
393 ret = memcmp(dst_copy, src_copy, item_size);
394
395 kfree(dst_copy);
396 kfree(src_copy);
397 /*
398 * they have the same contents, just return, this saves
399 * us from cowing blocks in the destination tree and doing
400 * extra writes that may not have been done by a previous
401 * sync
402 */
403 if (ret == 0) {
David Sterbab3b4aa72011-04-21 01:20:15 +0200404 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -0400405 return 0;
406 }
407
Josef Bacik4bc4bee2013-04-05 20:50:09 +0000408 /*
409 * We need to load the old nbytes into the inode so when we
410 * replay the extents we've logged we get the right nbytes.
411 */
412 if (inode_item) {
413 struct btrfs_inode_item *item;
414 u64 nbytes;
Josef Bacikd5554382013-09-11 14:17:00 -0400415 u32 mode;
Josef Bacik4bc4bee2013-04-05 20:50:09 +0000416
417 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
418 struct btrfs_inode_item);
419 nbytes = btrfs_inode_nbytes(path->nodes[0], item);
420 item = btrfs_item_ptr(eb, slot,
421 struct btrfs_inode_item);
422 btrfs_set_inode_nbytes(eb, item, nbytes);
Josef Bacikd5554382013-09-11 14:17:00 -0400423
424 /*
425 * If this is a directory we need to reset the i_size to
426 * 0 so that we can set it up properly when replaying
427 * the rest of the items in this log.
428 */
429 mode = btrfs_inode_mode(eb, item);
430 if (S_ISDIR(mode))
431 btrfs_set_inode_size(eb, item, 0);
Josef Bacik4bc4bee2013-04-05 20:50:09 +0000432 }
433 } else if (inode_item) {
434 struct btrfs_inode_item *item;
Josef Bacikd5554382013-09-11 14:17:00 -0400435 u32 mode;
Josef Bacik4bc4bee2013-04-05 20:50:09 +0000436
437 /*
438 * New inode, set nbytes to 0 so that the nbytes comes out
439 * properly when we replay the extents.
440 */
441 item = btrfs_item_ptr(eb, slot, struct btrfs_inode_item);
442 btrfs_set_inode_nbytes(eb, item, 0);
Josef Bacikd5554382013-09-11 14:17:00 -0400443
444 /*
445 * If this is a directory we need to reset the i_size to 0 so
446 * that we can set it up properly when replaying the rest of
447 * the items in this log.
448 */
449 mode = btrfs_inode_mode(eb, item);
450 if (S_ISDIR(mode))
451 btrfs_set_inode_size(eb, item, 0);
Chris Masone02119d2008-09-05 16:13:11 -0400452 }
453insert:
David Sterbab3b4aa72011-04-21 01:20:15 +0200454 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -0400455 /* try to insert the key into the destination tree */
456 ret = btrfs_insert_empty_item(trans, root, path,
457 key, item_size);
458
459 /* make sure any existing item is the correct size */
460 if (ret == -EEXIST) {
461 u32 found_size;
462 found_size = btrfs_item_size_nr(path->nodes[0],
463 path->slots[0]);
Jeff Mahoney143bede2012-03-01 14:56:26 +0100464 if (found_size > item_size)
Tsutomu Itohafe5fea2013-04-16 05:18:22 +0000465 btrfs_truncate_item(root, path, item_size, 1);
Jeff Mahoney143bede2012-03-01 14:56:26 +0100466 else if (found_size < item_size)
Tsutomu Itoh4b90c682013-04-16 05:18:49 +0000467 btrfs_extend_item(root, path,
Jeff Mahoney143bede2012-03-01 14:56:26 +0100468 item_size - found_size);
Chris Masone02119d2008-09-05 16:13:11 -0400469 } else if (ret) {
Yan, Zheng4a500fd2010-05-16 10:49:59 -0400470 return ret;
Chris Masone02119d2008-09-05 16:13:11 -0400471 }
472 dst_ptr = btrfs_item_ptr_offset(path->nodes[0],
473 path->slots[0]);
474
475 /* don't overwrite an existing inode if the generation number
476 * was logged as zero. This is done when the tree logging code
477 * is just logging an inode to make sure it exists after recovery.
478 *
479 * Also, don't overwrite i_size on directories during replay.
480 * log replay inserts and removes directory items based on the
481 * state of the tree found in the subvolume, and i_size is modified
482 * as it goes
483 */
484 if (key->type == BTRFS_INODE_ITEM_KEY && ret == -EEXIST) {
485 struct btrfs_inode_item *src_item;
486 struct btrfs_inode_item *dst_item;
487
488 src_item = (struct btrfs_inode_item *)src_ptr;
489 dst_item = (struct btrfs_inode_item *)dst_ptr;
490
491 if (btrfs_inode_generation(eb, src_item) == 0)
492 goto no_copy;
493
494 if (overwrite_root &&
495 S_ISDIR(btrfs_inode_mode(eb, src_item)) &&
496 S_ISDIR(btrfs_inode_mode(path->nodes[0], dst_item))) {
497 save_old_i_size = 1;
498 saved_i_size = btrfs_inode_size(path->nodes[0],
499 dst_item);
500 }
501 }
502
503 copy_extent_buffer(path->nodes[0], eb, dst_ptr,
504 src_ptr, item_size);
505
506 if (save_old_i_size) {
507 struct btrfs_inode_item *dst_item;
508 dst_item = (struct btrfs_inode_item *)dst_ptr;
509 btrfs_set_inode_size(path->nodes[0], dst_item, saved_i_size);
510 }
511
512 /* make sure the generation is filled in */
513 if (key->type == BTRFS_INODE_ITEM_KEY) {
514 struct btrfs_inode_item *dst_item;
515 dst_item = (struct btrfs_inode_item *)dst_ptr;
516 if (btrfs_inode_generation(path->nodes[0], dst_item) == 0) {
517 btrfs_set_inode_generation(path->nodes[0], dst_item,
518 trans->transid);
519 }
520 }
521no_copy:
522 btrfs_mark_buffer_dirty(path->nodes[0]);
David Sterbab3b4aa72011-04-21 01:20:15 +0200523 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -0400524 return 0;
525}
526
527/*
528 * simple helper to read an inode off the disk from a given root
529 * This can only be called for subvolume roots and not for the log
530 */
531static noinline struct inode *read_one_inode(struct btrfs_root *root,
532 u64 objectid)
533{
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400534 struct btrfs_key key;
Chris Masone02119d2008-09-05 16:13:11 -0400535 struct inode *inode;
Chris Masone02119d2008-09-05 16:13:11 -0400536
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400537 key.objectid = objectid;
538 key.type = BTRFS_INODE_ITEM_KEY;
539 key.offset = 0;
Josef Bacik73f73412009-12-04 17:38:27 +0000540 inode = btrfs_iget(root->fs_info->sb, &key, root, NULL);
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400541 if (IS_ERR(inode)) {
542 inode = NULL;
543 } else if (is_bad_inode(inode)) {
Chris Masone02119d2008-09-05 16:13:11 -0400544 iput(inode);
545 inode = NULL;
546 }
547 return inode;
548}
549
550/* replays a single extent in 'eb' at 'slot' with 'key' into the
551 * subvolume 'root'. path is released on entry and should be released
552 * on exit.
553 *
554 * extents in the log tree have not been allocated out of the extent
555 * tree yet. So, this completes the allocation, taking a reference
556 * as required if the extent already exists or creating a new extent
557 * if it isn't in the extent allocation tree yet.
558 *
559 * The extent is inserted into the file, dropping any existing extents
560 * from the file that overlap the new one.
561 */
562static noinline int replay_one_extent(struct btrfs_trans_handle *trans,
563 struct btrfs_root *root,
564 struct btrfs_path *path,
565 struct extent_buffer *eb, int slot,
566 struct btrfs_key *key)
567{
568 int found_type;
Chris Masone02119d2008-09-05 16:13:11 -0400569 u64 extent_end;
Chris Masone02119d2008-09-05 16:13:11 -0400570 u64 start = key->offset;
Josef Bacik4bc4bee2013-04-05 20:50:09 +0000571 u64 nbytes = 0;
Chris Masone02119d2008-09-05 16:13:11 -0400572 struct btrfs_file_extent_item *item;
573 struct inode *inode = NULL;
574 unsigned long size;
575 int ret = 0;
576
577 item = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
578 found_type = btrfs_file_extent_type(eb, item);
579
Yan Zhengd899e052008-10-30 14:25:28 -0400580 if (found_type == BTRFS_FILE_EXTENT_REG ||
Josef Bacik4bc4bee2013-04-05 20:50:09 +0000581 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
582 nbytes = btrfs_file_extent_num_bytes(eb, item);
583 extent_end = start + nbytes;
584
585 /*
586 * We don't add to the inodes nbytes if we are prealloc or a
587 * hole.
588 */
589 if (btrfs_file_extent_disk_bytenr(eb, item) == 0)
590 nbytes = 0;
591 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
Chris Mason514ac8a2014-01-03 21:07:00 -0800592 size = btrfs_file_extent_inline_len(eb, slot, item);
Josef Bacik4bc4bee2013-04-05 20:50:09 +0000593 nbytes = btrfs_file_extent_ram_bytes(eb, item);
Qu Wenruofda28322013-02-26 08:10:22 +0000594 extent_end = ALIGN(start + size, root->sectorsize);
Chris Masone02119d2008-09-05 16:13:11 -0400595 } else {
596 ret = 0;
597 goto out;
598 }
599
600 inode = read_one_inode(root, key->objectid);
601 if (!inode) {
602 ret = -EIO;
603 goto out;
604 }
605
606 /*
607 * first check to see if we already have this extent in the
608 * file. This must be done before the btrfs_drop_extents run
609 * so we don't try to drop this extent.
610 */
Li Zefan33345d012011-04-20 10:31:50 +0800611 ret = btrfs_lookup_file_extent(trans, root, path, btrfs_ino(inode),
Chris Masone02119d2008-09-05 16:13:11 -0400612 start, 0);
613
Yan Zhengd899e052008-10-30 14:25:28 -0400614 if (ret == 0 &&
615 (found_type == BTRFS_FILE_EXTENT_REG ||
616 found_type == BTRFS_FILE_EXTENT_PREALLOC)) {
Chris Masone02119d2008-09-05 16:13:11 -0400617 struct btrfs_file_extent_item cmp1;
618 struct btrfs_file_extent_item cmp2;
619 struct btrfs_file_extent_item *existing;
620 struct extent_buffer *leaf;
621
622 leaf = path->nodes[0];
623 existing = btrfs_item_ptr(leaf, path->slots[0],
624 struct btrfs_file_extent_item);
625
626 read_extent_buffer(eb, &cmp1, (unsigned long)item,
627 sizeof(cmp1));
628 read_extent_buffer(leaf, &cmp2, (unsigned long)existing,
629 sizeof(cmp2));
630
631 /*
632 * we already have a pointer to this exact extent,
633 * we don't have to do anything
634 */
635 if (memcmp(&cmp1, &cmp2, sizeof(cmp1)) == 0) {
David Sterbab3b4aa72011-04-21 01:20:15 +0200636 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -0400637 goto out;
638 }
639 }
David Sterbab3b4aa72011-04-21 01:20:15 +0200640 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -0400641
642 /* drop any overlapping extents */
Josef Bacik26714852012-08-29 12:24:27 -0400643 ret = btrfs_drop_extents(trans, root, inode, start, extent_end, 1);
Josef Bacik36508602013-04-25 16:23:32 -0400644 if (ret)
645 goto out;
Chris Masone02119d2008-09-05 16:13:11 -0400646
Yan Zheng07d400a2009-01-06 11:42:00 -0500647 if (found_type == BTRFS_FILE_EXTENT_REG ||
648 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400649 u64 offset;
Yan Zheng07d400a2009-01-06 11:42:00 -0500650 unsigned long dest_offset;
651 struct btrfs_key ins;
Chris Masone02119d2008-09-05 16:13:11 -0400652
Yan Zheng07d400a2009-01-06 11:42:00 -0500653 ret = btrfs_insert_empty_item(trans, root, path, key,
654 sizeof(*item));
Josef Bacik36508602013-04-25 16:23:32 -0400655 if (ret)
656 goto out;
Yan Zheng07d400a2009-01-06 11:42:00 -0500657 dest_offset = btrfs_item_ptr_offset(path->nodes[0],
658 path->slots[0]);
659 copy_extent_buffer(path->nodes[0], eb, dest_offset,
660 (unsigned long)item, sizeof(*item));
661
662 ins.objectid = btrfs_file_extent_disk_bytenr(eb, item);
663 ins.offset = btrfs_file_extent_disk_num_bytes(eb, item);
664 ins.type = BTRFS_EXTENT_ITEM_KEY;
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400665 offset = key->offset - btrfs_file_extent_offset(eb, item);
Yan Zheng07d400a2009-01-06 11:42:00 -0500666
667 if (ins.objectid > 0) {
668 u64 csum_start;
669 u64 csum_end;
670 LIST_HEAD(ordered_sums);
671 /*
672 * is this extent already allocated in the extent
673 * allocation tree? If so, just add a reference
674 */
Filipe Manana1a4ed8f2014-10-27 10:44:24 +0000675 ret = btrfs_lookup_data_extent(root, ins.objectid,
Yan Zheng07d400a2009-01-06 11:42:00 -0500676 ins.offset);
677 if (ret == 0) {
678 ret = btrfs_inc_extent_ref(trans, root,
679 ins.objectid, ins.offset,
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400680 0, root->root_key.objectid,
Arne Jansen66d7e7f2011-09-12 15:26:38 +0200681 key->objectid, offset, 0);
Josef Bacikb50c6e22013-04-25 15:55:30 -0400682 if (ret)
683 goto out;
Yan Zheng07d400a2009-01-06 11:42:00 -0500684 } else {
685 /*
686 * insert the extent pointer in the extent
687 * allocation tree
688 */
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400689 ret = btrfs_alloc_logged_file_extent(trans,
690 root, root->root_key.objectid,
691 key->objectid, offset, &ins);
Josef Bacikb50c6e22013-04-25 15:55:30 -0400692 if (ret)
693 goto out;
Yan Zheng07d400a2009-01-06 11:42:00 -0500694 }
David Sterbab3b4aa72011-04-21 01:20:15 +0200695 btrfs_release_path(path);
Yan Zheng07d400a2009-01-06 11:42:00 -0500696
697 if (btrfs_file_extent_compression(eb, item)) {
698 csum_start = ins.objectid;
699 csum_end = csum_start + ins.offset;
700 } else {
701 csum_start = ins.objectid +
702 btrfs_file_extent_offset(eb, item);
703 csum_end = csum_start +
704 btrfs_file_extent_num_bytes(eb, item);
705 }
706
707 ret = btrfs_lookup_csums_range(root->log_root,
708 csum_start, csum_end - 1,
Arne Jansena2de7332011-03-08 14:14:00 +0100709 &ordered_sums, 0);
Josef Bacik36508602013-04-25 16:23:32 -0400710 if (ret)
711 goto out;
Yan Zheng07d400a2009-01-06 11:42:00 -0500712 while (!list_empty(&ordered_sums)) {
713 struct btrfs_ordered_sum *sums;
714 sums = list_entry(ordered_sums.next,
715 struct btrfs_ordered_sum,
716 list);
Josef Bacik36508602013-04-25 16:23:32 -0400717 if (!ret)
718 ret = btrfs_csum_file_blocks(trans,
Yan Zheng07d400a2009-01-06 11:42:00 -0500719 root->fs_info->csum_root,
720 sums);
Yan Zheng07d400a2009-01-06 11:42:00 -0500721 list_del(&sums->list);
722 kfree(sums);
723 }
Josef Bacik36508602013-04-25 16:23:32 -0400724 if (ret)
725 goto out;
Yan Zheng07d400a2009-01-06 11:42:00 -0500726 } else {
David Sterbab3b4aa72011-04-21 01:20:15 +0200727 btrfs_release_path(path);
Yan Zheng07d400a2009-01-06 11:42:00 -0500728 }
729 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
730 /* inline extents are easy, we just overwrite them */
731 ret = overwrite_item(trans, root, path, eb, slot, key);
Josef Bacik36508602013-04-25 16:23:32 -0400732 if (ret)
733 goto out;
Yan Zheng07d400a2009-01-06 11:42:00 -0500734 }
735
Josef Bacik4bc4bee2013-04-05 20:50:09 +0000736 inode_add_bytes(inode, nbytes);
Tsutomu Itohb9959292012-06-25 21:25:22 -0600737 ret = btrfs_update_inode(trans, root, inode);
Chris Masone02119d2008-09-05 16:13:11 -0400738out:
739 if (inode)
740 iput(inode);
741 return ret;
742}
743
744/*
745 * when cleaning up conflicts between the directory names in the
746 * subvolume, directory names in the log and directory names in the
747 * inode back references, we may have to unlink inodes from directories.
748 *
749 * This is a helper function to do the unlink of a specific directory
750 * item
751 */
752static noinline int drop_one_dir_item(struct btrfs_trans_handle *trans,
753 struct btrfs_root *root,
754 struct btrfs_path *path,
755 struct inode *dir,
756 struct btrfs_dir_item *di)
757{
758 struct inode *inode;
759 char *name;
760 int name_len;
761 struct extent_buffer *leaf;
762 struct btrfs_key location;
763 int ret;
764
765 leaf = path->nodes[0];
766
767 btrfs_dir_item_key_to_cpu(leaf, di, &location);
768 name_len = btrfs_dir_name_len(leaf, di);
769 name = kmalloc(name_len, GFP_NOFS);
liubo2a29edc2011-01-26 06:22:08 +0000770 if (!name)
771 return -ENOMEM;
772
Chris Masone02119d2008-09-05 16:13:11 -0400773 read_extent_buffer(leaf, name, (unsigned long)(di + 1), name_len);
David Sterbab3b4aa72011-04-21 01:20:15 +0200774 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -0400775
776 inode = read_one_inode(root, location.objectid);
Tsutomu Itohc00e9492011-04-28 09:10:23 +0000777 if (!inode) {
Josef Bacik36508602013-04-25 16:23:32 -0400778 ret = -EIO;
779 goto out;
Tsutomu Itohc00e9492011-04-28 09:10:23 +0000780 }
Chris Masone02119d2008-09-05 16:13:11 -0400781
Yan Zhengec051c02009-01-05 15:43:42 -0500782 ret = link_to_fixup_dir(trans, root, path, location.objectid);
Josef Bacik36508602013-04-25 16:23:32 -0400783 if (ret)
784 goto out;
Chris Mason12fcfd22009-03-24 10:24:20 -0400785
Chris Masone02119d2008-09-05 16:13:11 -0400786 ret = btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
Josef Bacik36508602013-04-25 16:23:32 -0400787 if (ret)
788 goto out;
Filipe David Borba Mananaada9af22013-08-05 09:25:47 +0100789 else
790 ret = btrfs_run_delayed_items(trans, root);
Josef Bacik36508602013-04-25 16:23:32 -0400791out:
792 kfree(name);
793 iput(inode);
Chris Masone02119d2008-09-05 16:13:11 -0400794 return ret;
795}
796
797/*
798 * helper function to see if a given name and sequence number found
799 * in an inode back reference are already in a directory and correctly
800 * point to this inode
801 */
802static noinline int inode_in_dir(struct btrfs_root *root,
803 struct btrfs_path *path,
804 u64 dirid, u64 objectid, u64 index,
805 const char *name, int name_len)
806{
807 struct btrfs_dir_item *di;
808 struct btrfs_key location;
809 int match = 0;
810
811 di = btrfs_lookup_dir_index_item(NULL, root, path, dirid,
812 index, name, name_len, 0);
813 if (di && !IS_ERR(di)) {
814 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
815 if (location.objectid != objectid)
816 goto out;
817 } else
818 goto out;
David Sterbab3b4aa72011-04-21 01:20:15 +0200819 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -0400820
821 di = btrfs_lookup_dir_item(NULL, root, path, dirid, name, name_len, 0);
822 if (di && !IS_ERR(di)) {
823 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
824 if (location.objectid != objectid)
825 goto out;
826 } else
827 goto out;
828 match = 1;
829out:
David Sterbab3b4aa72011-04-21 01:20:15 +0200830 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -0400831 return match;
832}
833
834/*
835 * helper function to check a log tree for a named back reference in
836 * an inode. This is used to decide if a back reference that is
837 * found in the subvolume conflicts with what we find in the log.
838 *
839 * inode backreferences may have multiple refs in a single item,
840 * during replay we process one reference at a time, and we don't
841 * want to delete valid links to a file from the subvolume if that
842 * link is also in the log.
843 */
844static noinline int backref_in_log(struct btrfs_root *log,
845 struct btrfs_key *key,
Mark Fashehf1863732012-08-08 11:32:27 -0700846 u64 ref_objectid,
Chris Masone02119d2008-09-05 16:13:11 -0400847 char *name, int namelen)
848{
849 struct btrfs_path *path;
850 struct btrfs_inode_ref *ref;
851 unsigned long ptr;
852 unsigned long ptr_end;
853 unsigned long name_ptr;
854 int found_name_len;
855 int item_size;
856 int ret;
857 int match = 0;
858
859 path = btrfs_alloc_path();
liubo2a29edc2011-01-26 06:22:08 +0000860 if (!path)
861 return -ENOMEM;
862
Chris Masone02119d2008-09-05 16:13:11 -0400863 ret = btrfs_search_slot(NULL, log, key, path, 0, 0);
864 if (ret != 0)
865 goto out;
866
Chris Masone02119d2008-09-05 16:13:11 -0400867 ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
Mark Fashehf1863732012-08-08 11:32:27 -0700868
869 if (key->type == BTRFS_INODE_EXTREF_KEY) {
870 if (btrfs_find_name_in_ext_backref(path, ref_objectid,
871 name, namelen, NULL))
872 match = 1;
873
874 goto out;
875 }
876
877 item_size = btrfs_item_size_nr(path->nodes[0], path->slots[0]);
Chris Masone02119d2008-09-05 16:13:11 -0400878 ptr_end = ptr + item_size;
879 while (ptr < ptr_end) {
880 ref = (struct btrfs_inode_ref *)ptr;
881 found_name_len = btrfs_inode_ref_name_len(path->nodes[0], ref);
882 if (found_name_len == namelen) {
883 name_ptr = (unsigned long)(ref + 1);
884 ret = memcmp_extent_buffer(path->nodes[0], name,
885 name_ptr, namelen);
886 if (ret == 0) {
887 match = 1;
888 goto out;
889 }
890 }
891 ptr = (unsigned long)(ref + 1) + found_name_len;
892 }
893out:
894 btrfs_free_path(path);
895 return match;
896}
897
Jan Schmidt5a1d7842012-08-17 14:04:41 -0700898static inline int __add_inode_ref(struct btrfs_trans_handle *trans,
899 struct btrfs_root *root,
900 struct btrfs_path *path,
901 struct btrfs_root *log_root,
902 struct inode *dir, struct inode *inode,
Jan Schmidt5a1d7842012-08-17 14:04:41 -0700903 struct extent_buffer *eb,
Mark Fashehf1863732012-08-08 11:32:27 -0700904 u64 inode_objectid, u64 parent_objectid,
905 u64 ref_index, char *name, int namelen,
906 int *search_done)
Jan Schmidt5a1d7842012-08-17 14:04:41 -0700907{
908 int ret;
Mark Fashehf1863732012-08-08 11:32:27 -0700909 char *victim_name;
910 int victim_name_len;
911 struct extent_buffer *leaf;
Jan Schmidt5a1d7842012-08-17 14:04:41 -0700912 struct btrfs_dir_item *di;
Mark Fashehf1863732012-08-08 11:32:27 -0700913 struct btrfs_key search_key;
914 struct btrfs_inode_extref *extref;
Jan Schmidt5a1d7842012-08-17 14:04:41 -0700915
Mark Fashehf1863732012-08-08 11:32:27 -0700916again:
917 /* Search old style refs */
918 search_key.objectid = inode_objectid;
919 search_key.type = BTRFS_INODE_REF_KEY;
920 search_key.offset = parent_objectid;
921 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
Jan Schmidt5a1d7842012-08-17 14:04:41 -0700922 if (ret == 0) {
Jan Schmidt5a1d7842012-08-17 14:04:41 -0700923 struct btrfs_inode_ref *victim_ref;
924 unsigned long ptr;
925 unsigned long ptr_end;
Mark Fashehf1863732012-08-08 11:32:27 -0700926
927 leaf = path->nodes[0];
Jan Schmidt5a1d7842012-08-17 14:04:41 -0700928
929 /* are we trying to overwrite a back ref for the root directory
930 * if so, just jump out, we're done
931 */
Mark Fashehf1863732012-08-08 11:32:27 -0700932 if (search_key.objectid == search_key.offset)
Jan Schmidt5a1d7842012-08-17 14:04:41 -0700933 return 1;
934
935 /* check all the names in this back reference to see
936 * if they are in the log. if so, we allow them to stay
937 * otherwise they must be unlinked as a conflict
938 */
939 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
940 ptr_end = ptr + btrfs_item_size_nr(leaf, path->slots[0]);
941 while (ptr < ptr_end) {
942 victim_ref = (struct btrfs_inode_ref *)ptr;
943 victim_name_len = btrfs_inode_ref_name_len(leaf,
944 victim_ref);
945 victim_name = kmalloc(victim_name_len, GFP_NOFS);
Josef Bacik36508602013-04-25 16:23:32 -0400946 if (!victim_name)
947 return -ENOMEM;
Jan Schmidt5a1d7842012-08-17 14:04:41 -0700948
949 read_extent_buffer(leaf, victim_name,
950 (unsigned long)(victim_ref + 1),
951 victim_name_len);
952
Mark Fashehf1863732012-08-08 11:32:27 -0700953 if (!backref_in_log(log_root, &search_key,
954 parent_objectid,
955 victim_name,
Jan Schmidt5a1d7842012-08-17 14:04:41 -0700956 victim_name_len)) {
Zach Brown8b558c52013-10-16 12:10:34 -0700957 inc_nlink(inode);
Jan Schmidt5a1d7842012-08-17 14:04:41 -0700958 btrfs_release_path(path);
959
960 ret = btrfs_unlink_inode(trans, root, dir,
961 inode, victim_name,
962 victim_name_len);
Mark Fashehf1863732012-08-08 11:32:27 -0700963 kfree(victim_name);
Josef Bacik36508602013-04-25 16:23:32 -0400964 if (ret)
965 return ret;
Filipe David Borba Mananaada9af22013-08-05 09:25:47 +0100966 ret = btrfs_run_delayed_items(trans, root);
967 if (ret)
968 return ret;
Mark Fashehf1863732012-08-08 11:32:27 -0700969 *search_done = 1;
970 goto again;
Jan Schmidt5a1d7842012-08-17 14:04:41 -0700971 }
972 kfree(victim_name);
Mark Fashehf1863732012-08-08 11:32:27 -0700973
Jan Schmidt5a1d7842012-08-17 14:04:41 -0700974 ptr = (unsigned long)(victim_ref + 1) + victim_name_len;
975 }
Jan Schmidt5a1d7842012-08-17 14:04:41 -0700976
977 /*
978 * NOTE: we have searched root tree and checked the
979 * coresponding ref, it does not need to check again.
980 */
981 *search_done = 1;
982 }
983 btrfs_release_path(path);
984
Mark Fashehf1863732012-08-08 11:32:27 -0700985 /* Same search but for extended refs */
986 extref = btrfs_lookup_inode_extref(NULL, root, path, name, namelen,
987 inode_objectid, parent_objectid, 0,
988 0);
989 if (!IS_ERR_OR_NULL(extref)) {
990 u32 item_size;
991 u32 cur_offset = 0;
992 unsigned long base;
993 struct inode *victim_parent;
994
995 leaf = path->nodes[0];
996
997 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
998 base = btrfs_item_ptr_offset(leaf, path->slots[0]);
999
1000 while (cur_offset < item_size) {
1001 extref = (struct btrfs_inode_extref *)base + cur_offset;
1002
1003 victim_name_len = btrfs_inode_extref_name_len(leaf, extref);
1004
1005 if (btrfs_inode_extref_parent(leaf, extref) != parent_objectid)
1006 goto next;
1007
1008 victim_name = kmalloc(victim_name_len, GFP_NOFS);
Josef Bacik36508602013-04-25 16:23:32 -04001009 if (!victim_name)
1010 return -ENOMEM;
Mark Fashehf1863732012-08-08 11:32:27 -07001011 read_extent_buffer(leaf, victim_name, (unsigned long)&extref->name,
1012 victim_name_len);
1013
1014 search_key.objectid = inode_objectid;
1015 search_key.type = BTRFS_INODE_EXTREF_KEY;
1016 search_key.offset = btrfs_extref_hash(parent_objectid,
1017 victim_name,
1018 victim_name_len);
1019 ret = 0;
1020 if (!backref_in_log(log_root, &search_key,
1021 parent_objectid, victim_name,
1022 victim_name_len)) {
1023 ret = -ENOENT;
1024 victim_parent = read_one_inode(root,
1025 parent_objectid);
1026 if (victim_parent) {
Zach Brown8b558c52013-10-16 12:10:34 -07001027 inc_nlink(inode);
Mark Fashehf1863732012-08-08 11:32:27 -07001028 btrfs_release_path(path);
1029
1030 ret = btrfs_unlink_inode(trans, root,
1031 victim_parent,
1032 inode,
1033 victim_name,
1034 victim_name_len);
Filipe David Borba Mananaada9af22013-08-05 09:25:47 +01001035 if (!ret)
1036 ret = btrfs_run_delayed_items(
1037 trans, root);
Mark Fashehf1863732012-08-08 11:32:27 -07001038 }
Mark Fashehf1863732012-08-08 11:32:27 -07001039 iput(victim_parent);
1040 kfree(victim_name);
Josef Bacik36508602013-04-25 16:23:32 -04001041 if (ret)
1042 return ret;
Mark Fashehf1863732012-08-08 11:32:27 -07001043 *search_done = 1;
1044 goto again;
1045 }
1046 kfree(victim_name);
Josef Bacik36508602013-04-25 16:23:32 -04001047 if (ret)
1048 return ret;
Mark Fashehf1863732012-08-08 11:32:27 -07001049next:
1050 cur_offset += victim_name_len + sizeof(*extref);
1051 }
1052 *search_done = 1;
1053 }
1054 btrfs_release_path(path);
1055
Jan Schmidt5a1d7842012-08-17 14:04:41 -07001056 /* look for a conflicting sequence number */
1057 di = btrfs_lookup_dir_index_item(trans, root, path, btrfs_ino(dir),
Mark Fashehf1863732012-08-08 11:32:27 -07001058 ref_index, name, namelen, 0);
Jan Schmidt5a1d7842012-08-17 14:04:41 -07001059 if (di && !IS_ERR(di)) {
1060 ret = drop_one_dir_item(trans, root, path, dir, di);
Josef Bacik36508602013-04-25 16:23:32 -04001061 if (ret)
1062 return ret;
Jan Schmidt5a1d7842012-08-17 14:04:41 -07001063 }
1064 btrfs_release_path(path);
1065
1066 /* look for a conflicing name */
1067 di = btrfs_lookup_dir_item(trans, root, path, btrfs_ino(dir),
1068 name, namelen, 0);
1069 if (di && !IS_ERR(di)) {
1070 ret = drop_one_dir_item(trans, root, path, dir, di);
Josef Bacik36508602013-04-25 16:23:32 -04001071 if (ret)
1072 return ret;
Jan Schmidt5a1d7842012-08-17 14:04:41 -07001073 }
1074 btrfs_release_path(path);
1075
1076 return 0;
1077}
Chris Masone02119d2008-09-05 16:13:11 -04001078
Mark Fashehf1863732012-08-08 11:32:27 -07001079static int extref_get_fields(struct extent_buffer *eb, unsigned long ref_ptr,
1080 u32 *namelen, char **name, u64 *index,
1081 u64 *parent_objectid)
1082{
1083 struct btrfs_inode_extref *extref;
1084
1085 extref = (struct btrfs_inode_extref *)ref_ptr;
1086
1087 *namelen = btrfs_inode_extref_name_len(eb, extref);
1088 *name = kmalloc(*namelen, GFP_NOFS);
1089 if (*name == NULL)
1090 return -ENOMEM;
1091
1092 read_extent_buffer(eb, *name, (unsigned long)&extref->name,
1093 *namelen);
1094
1095 *index = btrfs_inode_extref_index(eb, extref);
1096 if (parent_objectid)
1097 *parent_objectid = btrfs_inode_extref_parent(eb, extref);
1098
1099 return 0;
1100}
1101
1102static int ref_get_fields(struct extent_buffer *eb, unsigned long ref_ptr,
1103 u32 *namelen, char **name, u64 *index)
1104{
1105 struct btrfs_inode_ref *ref;
1106
1107 ref = (struct btrfs_inode_ref *)ref_ptr;
1108
1109 *namelen = btrfs_inode_ref_name_len(eb, ref);
1110 *name = kmalloc(*namelen, GFP_NOFS);
1111 if (*name == NULL)
1112 return -ENOMEM;
1113
1114 read_extent_buffer(eb, *name, (unsigned long)(ref + 1), *namelen);
1115
1116 *index = btrfs_inode_ref_index(eb, ref);
1117
1118 return 0;
1119}
1120
Chris Masone02119d2008-09-05 16:13:11 -04001121/*
1122 * replay one inode back reference item found in the log tree.
1123 * eb, slot and key refer to the buffer and key found in the log tree.
1124 * root is the destination we are replaying into, and path is for temp
1125 * use by this function. (it should be released on return).
1126 */
1127static noinline int add_inode_ref(struct btrfs_trans_handle *trans,
1128 struct btrfs_root *root,
1129 struct btrfs_root *log,
1130 struct btrfs_path *path,
1131 struct extent_buffer *eb, int slot,
1132 struct btrfs_key *key)
1133{
Geyslan G. Bem03b2f082013-10-11 15:35:45 -03001134 struct inode *dir = NULL;
1135 struct inode *inode = NULL;
Chris Masone02119d2008-09-05 16:13:11 -04001136 unsigned long ref_ptr;
1137 unsigned long ref_end;
Geyslan G. Bem03b2f082013-10-11 15:35:45 -03001138 char *name = NULL;
liubo34f3e4f2011-08-06 08:35:23 +00001139 int namelen;
1140 int ret;
liuboc622ae62011-03-26 08:01:12 -04001141 int search_done = 0;
Mark Fashehf1863732012-08-08 11:32:27 -07001142 int log_ref_ver = 0;
1143 u64 parent_objectid;
1144 u64 inode_objectid;
Chris Masonf46dbe3de2012-10-09 11:17:20 -04001145 u64 ref_index = 0;
Mark Fashehf1863732012-08-08 11:32:27 -07001146 int ref_struct_size;
1147
1148 ref_ptr = btrfs_item_ptr_offset(eb, slot);
1149 ref_end = ref_ptr + btrfs_item_size_nr(eb, slot);
1150
1151 if (key->type == BTRFS_INODE_EXTREF_KEY) {
1152 struct btrfs_inode_extref *r;
1153
1154 ref_struct_size = sizeof(struct btrfs_inode_extref);
1155 log_ref_ver = 1;
1156 r = (struct btrfs_inode_extref *)ref_ptr;
1157 parent_objectid = btrfs_inode_extref_parent(eb, r);
1158 } else {
1159 ref_struct_size = sizeof(struct btrfs_inode_ref);
1160 parent_objectid = key->offset;
1161 }
1162 inode_objectid = key->objectid;
Chris Masone02119d2008-09-05 16:13:11 -04001163
Chris Masone02119d2008-09-05 16:13:11 -04001164 /*
1165 * it is possible that we didn't log all the parent directories
1166 * for a given inode. If we don't find the dir, just don't
1167 * copy the back ref in. The link count fixup code will take
1168 * care of the rest
1169 */
Mark Fashehf1863732012-08-08 11:32:27 -07001170 dir = read_one_inode(root, parent_objectid);
Geyslan G. Bem03b2f082013-10-11 15:35:45 -03001171 if (!dir) {
1172 ret = -ENOENT;
1173 goto out;
1174 }
Chris Masone02119d2008-09-05 16:13:11 -04001175
Mark Fashehf1863732012-08-08 11:32:27 -07001176 inode = read_one_inode(root, inode_objectid);
Tsutomu Itohc00e9492011-04-28 09:10:23 +00001177 if (!inode) {
Geyslan G. Bem03b2f082013-10-11 15:35:45 -03001178 ret = -EIO;
1179 goto out;
Tsutomu Itohc00e9492011-04-28 09:10:23 +00001180 }
Chris Masone02119d2008-09-05 16:13:11 -04001181
Jan Schmidt5a1d7842012-08-17 14:04:41 -07001182 while (ref_ptr < ref_end) {
Mark Fashehf1863732012-08-08 11:32:27 -07001183 if (log_ref_ver) {
1184 ret = extref_get_fields(eb, ref_ptr, &namelen, &name,
1185 &ref_index, &parent_objectid);
1186 /*
1187 * parent object can change from one array
1188 * item to another.
1189 */
1190 if (!dir)
1191 dir = read_one_inode(root, parent_objectid);
Geyslan G. Bem03b2f082013-10-11 15:35:45 -03001192 if (!dir) {
1193 ret = -ENOENT;
1194 goto out;
1195 }
Mark Fashehf1863732012-08-08 11:32:27 -07001196 } else {
1197 ret = ref_get_fields(eb, ref_ptr, &namelen, &name,
1198 &ref_index);
1199 }
1200 if (ret)
Geyslan G. Bem03b2f082013-10-11 15:35:45 -03001201 goto out;
Chris Masone02119d2008-09-05 16:13:11 -04001202
Jan Schmidt5a1d7842012-08-17 14:04:41 -07001203 /* if we already have a perfect match, we're done */
1204 if (!inode_in_dir(root, path, btrfs_ino(dir), btrfs_ino(inode),
Mark Fashehf1863732012-08-08 11:32:27 -07001205 ref_index, name, namelen)) {
Jan Schmidt5a1d7842012-08-17 14:04:41 -07001206 /*
1207 * look for a conflicting back reference in the
1208 * metadata. if we find one we have to unlink that name
1209 * of the file before we add our new link. Later on, we
1210 * overwrite any existing back reference, and we don't
1211 * want to create dangling pointers in the directory.
1212 */
Chris Masone02119d2008-09-05 16:13:11 -04001213
Jan Schmidt5a1d7842012-08-17 14:04:41 -07001214 if (!search_done) {
1215 ret = __add_inode_ref(trans, root, path, log,
Mark Fashehf1863732012-08-08 11:32:27 -07001216 dir, inode, eb,
1217 inode_objectid,
1218 parent_objectid,
1219 ref_index, name, namelen,
Jan Schmidt5a1d7842012-08-17 14:04:41 -07001220 &search_done);
Geyslan G. Bem03b2f082013-10-11 15:35:45 -03001221 if (ret) {
1222 if (ret == 1)
1223 ret = 0;
Jan Schmidt5a1d7842012-08-17 14:04:41 -07001224 goto out;
Josef Bacik36508602013-04-25 16:23:32 -04001225 }
Chris Masone02119d2008-09-05 16:13:11 -04001226 }
Jan Schmidt5a1d7842012-08-17 14:04:41 -07001227
1228 /* insert our name */
1229 ret = btrfs_add_link(trans, dir, inode, name, namelen,
Mark Fashehf1863732012-08-08 11:32:27 -07001230 0, ref_index);
Josef Bacik36508602013-04-25 16:23:32 -04001231 if (ret)
1232 goto out;
Jan Schmidt5a1d7842012-08-17 14:04:41 -07001233
1234 btrfs_update_inode(trans, root, inode);
Chris Masone02119d2008-09-05 16:13:11 -04001235 }
liuboc622ae62011-03-26 08:01:12 -04001236
Mark Fashehf1863732012-08-08 11:32:27 -07001237 ref_ptr = (unsigned long)(ref_ptr + ref_struct_size) + namelen;
Jan Schmidt5a1d7842012-08-17 14:04:41 -07001238 kfree(name);
Geyslan G. Bem03b2f082013-10-11 15:35:45 -03001239 name = NULL;
Mark Fashehf1863732012-08-08 11:32:27 -07001240 if (log_ref_ver) {
1241 iput(dir);
1242 dir = NULL;
1243 }
Chris Masone02119d2008-09-05 16:13:11 -04001244 }
Chris Masone02119d2008-09-05 16:13:11 -04001245
1246 /* finally write the back reference in the inode */
1247 ret = overwrite_item(trans, root, path, eb, slot, key);
Jan Schmidt5a1d7842012-08-17 14:04:41 -07001248out:
David Sterbab3b4aa72011-04-21 01:20:15 +02001249 btrfs_release_path(path);
Geyslan G. Bem03b2f082013-10-11 15:35:45 -03001250 kfree(name);
Chris Masone02119d2008-09-05 16:13:11 -04001251 iput(dir);
1252 iput(inode);
Josef Bacik36508602013-04-25 16:23:32 -04001253 return ret;
Chris Masone02119d2008-09-05 16:13:11 -04001254}
1255
Yan, Zhengc71bf092009-11-12 09:34:40 +00001256static int insert_orphan_item(struct btrfs_trans_handle *trans,
1257 struct btrfs_root *root, u64 offset)
1258{
1259 int ret;
David Sterba381cf652015-01-02 18:45:16 +01001260 struct btrfs_path *path;
1261
1262 path = btrfs_alloc_path();
1263 if (!path)
1264 return -ENOMEM;
1265
1266 ret = btrfs_find_item(root, path, BTRFS_ORPHAN_OBJECTID,
Kelley Nielsen3f870c22013-11-04 19:37:39 -08001267 offset, BTRFS_ORPHAN_ITEM_KEY, NULL);
Yan, Zhengc71bf092009-11-12 09:34:40 +00001268 if (ret > 0)
1269 ret = btrfs_insert_orphan_item(trans, root, offset);
David Sterba381cf652015-01-02 18:45:16 +01001270
1271 btrfs_free_path(path);
1272
Yan, Zhengc71bf092009-11-12 09:34:40 +00001273 return ret;
1274}
1275
Mark Fashehf1863732012-08-08 11:32:27 -07001276static int count_inode_extrefs(struct btrfs_root *root,
1277 struct inode *inode, struct btrfs_path *path)
Chris Masone02119d2008-09-05 16:13:11 -04001278{
Mark Fashehf1863732012-08-08 11:32:27 -07001279 int ret = 0;
1280 int name_len;
1281 unsigned int nlink = 0;
1282 u32 item_size;
1283 u32 cur_offset = 0;
1284 u64 inode_objectid = btrfs_ino(inode);
1285 u64 offset = 0;
1286 unsigned long ptr;
1287 struct btrfs_inode_extref *extref;
1288 struct extent_buffer *leaf;
1289
1290 while (1) {
1291 ret = btrfs_find_one_extref(root, inode_objectid, offset, path,
1292 &extref, &offset);
1293 if (ret)
1294 break;
1295
1296 leaf = path->nodes[0];
1297 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1298 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
1299
1300 while (cur_offset < item_size) {
1301 extref = (struct btrfs_inode_extref *) (ptr + cur_offset);
1302 name_len = btrfs_inode_extref_name_len(leaf, extref);
1303
1304 nlink++;
1305
1306 cur_offset += name_len + sizeof(*extref);
1307 }
1308
1309 offset++;
1310 btrfs_release_path(path);
1311 }
1312 btrfs_release_path(path);
1313
1314 if (ret < 0)
1315 return ret;
1316 return nlink;
1317}
1318
1319static int count_inode_refs(struct btrfs_root *root,
1320 struct inode *inode, struct btrfs_path *path)
1321{
Chris Masone02119d2008-09-05 16:13:11 -04001322 int ret;
1323 struct btrfs_key key;
Mark Fashehf1863732012-08-08 11:32:27 -07001324 unsigned int nlink = 0;
Chris Masone02119d2008-09-05 16:13:11 -04001325 unsigned long ptr;
1326 unsigned long ptr_end;
1327 int name_len;
Li Zefan33345d012011-04-20 10:31:50 +08001328 u64 ino = btrfs_ino(inode);
Chris Masone02119d2008-09-05 16:13:11 -04001329
Li Zefan33345d012011-04-20 10:31:50 +08001330 key.objectid = ino;
Chris Masone02119d2008-09-05 16:13:11 -04001331 key.type = BTRFS_INODE_REF_KEY;
1332 key.offset = (u64)-1;
1333
Chris Masond3977122009-01-05 21:25:51 -05001334 while (1) {
Chris Masone02119d2008-09-05 16:13:11 -04001335 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1336 if (ret < 0)
1337 break;
1338 if (ret > 0) {
1339 if (path->slots[0] == 0)
1340 break;
1341 path->slots[0]--;
1342 }
Filipe David Borba Mananae93ae262013-10-14 22:49:11 +01001343process_slot:
Chris Masone02119d2008-09-05 16:13:11 -04001344 btrfs_item_key_to_cpu(path->nodes[0], &key,
1345 path->slots[0]);
Li Zefan33345d012011-04-20 10:31:50 +08001346 if (key.objectid != ino ||
Chris Masone02119d2008-09-05 16:13:11 -04001347 key.type != BTRFS_INODE_REF_KEY)
1348 break;
1349 ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
1350 ptr_end = ptr + btrfs_item_size_nr(path->nodes[0],
1351 path->slots[0]);
Chris Masond3977122009-01-05 21:25:51 -05001352 while (ptr < ptr_end) {
Chris Masone02119d2008-09-05 16:13:11 -04001353 struct btrfs_inode_ref *ref;
1354
1355 ref = (struct btrfs_inode_ref *)ptr;
1356 name_len = btrfs_inode_ref_name_len(path->nodes[0],
1357 ref);
1358 ptr = (unsigned long)(ref + 1) + name_len;
1359 nlink++;
1360 }
1361
1362 if (key.offset == 0)
1363 break;
Filipe David Borba Mananae93ae262013-10-14 22:49:11 +01001364 if (path->slots[0] > 0) {
1365 path->slots[0]--;
1366 goto process_slot;
1367 }
Chris Masone02119d2008-09-05 16:13:11 -04001368 key.offset--;
David Sterbab3b4aa72011-04-21 01:20:15 +02001369 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04001370 }
David Sterbab3b4aa72011-04-21 01:20:15 +02001371 btrfs_release_path(path);
Mark Fashehf1863732012-08-08 11:32:27 -07001372
1373 return nlink;
1374}
1375
1376/*
1377 * There are a few corners where the link count of the file can't
1378 * be properly maintained during replay. So, instead of adding
1379 * lots of complexity to the log code, we just scan the backrefs
1380 * for any file that has been through replay.
1381 *
1382 * The scan will update the link count on the inode to reflect the
1383 * number of back refs found. If it goes down to zero, the iput
1384 * will free the inode.
1385 */
1386static noinline int fixup_inode_link_count(struct btrfs_trans_handle *trans,
1387 struct btrfs_root *root,
1388 struct inode *inode)
1389{
1390 struct btrfs_path *path;
1391 int ret;
1392 u64 nlink = 0;
1393 u64 ino = btrfs_ino(inode);
1394
1395 path = btrfs_alloc_path();
1396 if (!path)
1397 return -ENOMEM;
1398
1399 ret = count_inode_refs(root, inode, path);
1400 if (ret < 0)
1401 goto out;
1402
1403 nlink = ret;
1404
1405 ret = count_inode_extrefs(root, inode, path);
1406 if (ret == -ENOENT)
1407 ret = 0;
1408
1409 if (ret < 0)
1410 goto out;
1411
1412 nlink += ret;
1413
1414 ret = 0;
1415
Chris Masone02119d2008-09-05 16:13:11 -04001416 if (nlink != inode->i_nlink) {
Miklos Szeredibfe86842011-10-28 14:13:29 +02001417 set_nlink(inode, nlink);
Chris Masone02119d2008-09-05 16:13:11 -04001418 btrfs_update_inode(trans, root, inode);
1419 }
Chris Mason8d5bf1c2008-09-11 15:51:21 -04001420 BTRFS_I(inode)->index_cnt = (u64)-1;
Chris Masone02119d2008-09-05 16:13:11 -04001421
Yan, Zhengc71bf092009-11-12 09:34:40 +00001422 if (inode->i_nlink == 0) {
1423 if (S_ISDIR(inode->i_mode)) {
1424 ret = replay_dir_deletes(trans, root, NULL, path,
Li Zefan33345d012011-04-20 10:31:50 +08001425 ino, 1);
Josef Bacik36508602013-04-25 16:23:32 -04001426 if (ret)
1427 goto out;
Yan, Zhengc71bf092009-11-12 09:34:40 +00001428 }
Li Zefan33345d012011-04-20 10:31:50 +08001429 ret = insert_orphan_item(trans, root, ino);
Chris Mason12fcfd22009-03-24 10:24:20 -04001430 }
Chris Mason12fcfd22009-03-24 10:24:20 -04001431
Mark Fashehf1863732012-08-08 11:32:27 -07001432out:
1433 btrfs_free_path(path);
1434 return ret;
Chris Masone02119d2008-09-05 16:13:11 -04001435}
1436
1437static noinline int fixup_inode_link_counts(struct btrfs_trans_handle *trans,
1438 struct btrfs_root *root,
1439 struct btrfs_path *path)
1440{
1441 int ret;
1442 struct btrfs_key key;
1443 struct inode *inode;
1444
1445 key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID;
1446 key.type = BTRFS_ORPHAN_ITEM_KEY;
1447 key.offset = (u64)-1;
Chris Masond3977122009-01-05 21:25:51 -05001448 while (1) {
Chris Masone02119d2008-09-05 16:13:11 -04001449 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1450 if (ret < 0)
1451 break;
1452
1453 if (ret == 1) {
1454 if (path->slots[0] == 0)
1455 break;
1456 path->slots[0]--;
1457 }
1458
1459 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1460 if (key.objectid != BTRFS_TREE_LOG_FIXUP_OBJECTID ||
1461 key.type != BTRFS_ORPHAN_ITEM_KEY)
1462 break;
1463
1464 ret = btrfs_del_item(trans, root, path);
Tsutomu Itoh65a246c2011-05-19 04:37:44 +00001465 if (ret)
1466 goto out;
Chris Masone02119d2008-09-05 16:13:11 -04001467
David Sterbab3b4aa72011-04-21 01:20:15 +02001468 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04001469 inode = read_one_inode(root, key.offset);
Tsutomu Itohc00e9492011-04-28 09:10:23 +00001470 if (!inode)
1471 return -EIO;
Chris Masone02119d2008-09-05 16:13:11 -04001472
1473 ret = fixup_inode_link_count(trans, root, inode);
Chris Masone02119d2008-09-05 16:13:11 -04001474 iput(inode);
Josef Bacik36508602013-04-25 16:23:32 -04001475 if (ret)
1476 goto out;
Chris Masone02119d2008-09-05 16:13:11 -04001477
Chris Mason12fcfd22009-03-24 10:24:20 -04001478 /*
1479 * fixup on a directory may create new entries,
1480 * make sure we always look for the highset possible
1481 * offset
1482 */
1483 key.offset = (u64)-1;
Chris Masone02119d2008-09-05 16:13:11 -04001484 }
Tsutomu Itoh65a246c2011-05-19 04:37:44 +00001485 ret = 0;
1486out:
David Sterbab3b4aa72011-04-21 01:20:15 +02001487 btrfs_release_path(path);
Tsutomu Itoh65a246c2011-05-19 04:37:44 +00001488 return ret;
Chris Masone02119d2008-09-05 16:13:11 -04001489}
1490
1491
1492/*
1493 * record a given inode in the fixup dir so we can check its link
1494 * count when replay is done. The link count is incremented here
1495 * so the inode won't go away until we check it
1496 */
1497static noinline int link_to_fixup_dir(struct btrfs_trans_handle *trans,
1498 struct btrfs_root *root,
1499 struct btrfs_path *path,
1500 u64 objectid)
1501{
1502 struct btrfs_key key;
1503 int ret = 0;
1504 struct inode *inode;
1505
1506 inode = read_one_inode(root, objectid);
Tsutomu Itohc00e9492011-04-28 09:10:23 +00001507 if (!inode)
1508 return -EIO;
Chris Masone02119d2008-09-05 16:13:11 -04001509
1510 key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID;
David Sterba962a2982014-06-04 18:41:45 +02001511 key.type = BTRFS_ORPHAN_ITEM_KEY;
Chris Masone02119d2008-09-05 16:13:11 -04001512 key.offset = objectid;
1513
1514 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1515
David Sterbab3b4aa72011-04-21 01:20:15 +02001516 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04001517 if (ret == 0) {
Josef Bacik9bf7a482013-03-01 13:35:47 -05001518 if (!inode->i_nlink)
1519 set_nlink(inode, 1);
1520 else
Zach Brown8b558c52013-10-16 12:10:34 -07001521 inc_nlink(inode);
Tsutomu Itohb9959292012-06-25 21:25:22 -06001522 ret = btrfs_update_inode(trans, root, inode);
Chris Masone02119d2008-09-05 16:13:11 -04001523 } else if (ret == -EEXIST) {
1524 ret = 0;
1525 } else {
Josef Bacik36508602013-04-25 16:23:32 -04001526 BUG(); /* Logic Error */
Chris Masone02119d2008-09-05 16:13:11 -04001527 }
1528 iput(inode);
1529
1530 return ret;
1531}
1532
1533/*
1534 * when replaying the log for a directory, we only insert names
1535 * for inodes that actually exist. This means an fsync on a directory
1536 * does not implicitly fsync all the new files in it
1537 */
1538static noinline int insert_one_name(struct btrfs_trans_handle *trans,
1539 struct btrfs_root *root,
1540 struct btrfs_path *path,
1541 u64 dirid, u64 index,
1542 char *name, int name_len, u8 type,
1543 struct btrfs_key *location)
1544{
1545 struct inode *inode;
1546 struct inode *dir;
1547 int ret;
1548
1549 inode = read_one_inode(root, location->objectid);
1550 if (!inode)
1551 return -ENOENT;
1552
1553 dir = read_one_inode(root, dirid);
1554 if (!dir) {
1555 iput(inode);
1556 return -EIO;
1557 }
Josef Bacikd5554382013-09-11 14:17:00 -04001558
Chris Masone02119d2008-09-05 16:13:11 -04001559 ret = btrfs_add_link(trans, dir, inode, name, name_len, 1, index);
1560
1561 /* FIXME, put inode into FIXUP list */
1562
1563 iput(inode);
1564 iput(dir);
1565 return ret;
1566}
1567
1568/*
1569 * take a single entry in a log directory item and replay it into
1570 * the subvolume.
1571 *
1572 * if a conflicting item exists in the subdirectory already,
1573 * the inode it points to is unlinked and put into the link count
1574 * fix up tree.
1575 *
1576 * If a name from the log points to a file or directory that does
1577 * not exist in the FS, it is skipped. fsyncs on directories
1578 * do not force down inodes inside that directory, just changes to the
1579 * names or unlinks in a directory.
1580 */
1581static noinline int replay_one_name(struct btrfs_trans_handle *trans,
1582 struct btrfs_root *root,
1583 struct btrfs_path *path,
1584 struct extent_buffer *eb,
1585 struct btrfs_dir_item *di,
1586 struct btrfs_key *key)
1587{
1588 char *name;
1589 int name_len;
1590 struct btrfs_dir_item *dst_di;
1591 struct btrfs_key found_key;
1592 struct btrfs_key log_key;
1593 struct inode *dir;
Chris Masone02119d2008-09-05 16:13:11 -04001594 u8 log_type;
Chris Mason4bef0842008-09-08 11:18:08 -04001595 int exists;
Josef Bacik36508602013-04-25 16:23:32 -04001596 int ret = 0;
Josef Bacikd5554382013-09-11 14:17:00 -04001597 bool update_size = (key->type == BTRFS_DIR_INDEX_KEY);
Chris Masone02119d2008-09-05 16:13:11 -04001598
1599 dir = read_one_inode(root, key->objectid);
Tsutomu Itohc00e9492011-04-28 09:10:23 +00001600 if (!dir)
1601 return -EIO;
Chris Masone02119d2008-09-05 16:13:11 -04001602
1603 name_len = btrfs_dir_name_len(eb, di);
1604 name = kmalloc(name_len, GFP_NOFS);
Filipe David Borba Manana2bac3252013-08-04 19:58:57 +01001605 if (!name) {
1606 ret = -ENOMEM;
1607 goto out;
1608 }
liubo2a29edc2011-01-26 06:22:08 +00001609
Chris Masone02119d2008-09-05 16:13:11 -04001610 log_type = btrfs_dir_type(eb, di);
1611 read_extent_buffer(eb, name, (unsigned long)(di + 1),
1612 name_len);
1613
1614 btrfs_dir_item_key_to_cpu(eb, di, &log_key);
Chris Mason4bef0842008-09-08 11:18:08 -04001615 exists = btrfs_lookup_inode(trans, root, path, &log_key, 0);
1616 if (exists == 0)
1617 exists = 1;
1618 else
1619 exists = 0;
David Sterbab3b4aa72011-04-21 01:20:15 +02001620 btrfs_release_path(path);
Chris Mason4bef0842008-09-08 11:18:08 -04001621
Chris Masone02119d2008-09-05 16:13:11 -04001622 if (key->type == BTRFS_DIR_ITEM_KEY) {
1623 dst_di = btrfs_lookup_dir_item(trans, root, path, key->objectid,
1624 name, name_len, 1);
Chris Masond3977122009-01-05 21:25:51 -05001625 } else if (key->type == BTRFS_DIR_INDEX_KEY) {
Chris Masone02119d2008-09-05 16:13:11 -04001626 dst_di = btrfs_lookup_dir_index_item(trans, root, path,
1627 key->objectid,
1628 key->offset, name,
1629 name_len, 1);
1630 } else {
Josef Bacik36508602013-04-25 16:23:32 -04001631 /* Corruption */
1632 ret = -EINVAL;
1633 goto out;
Chris Masone02119d2008-09-05 16:13:11 -04001634 }
David Sterbac7040052011-04-19 18:00:01 +02001635 if (IS_ERR_OR_NULL(dst_di)) {
Chris Masone02119d2008-09-05 16:13:11 -04001636 /* we need a sequence number to insert, so we only
1637 * do inserts for the BTRFS_DIR_INDEX_KEY types
1638 */
1639 if (key->type != BTRFS_DIR_INDEX_KEY)
1640 goto out;
1641 goto insert;
1642 }
1643
1644 btrfs_dir_item_key_to_cpu(path->nodes[0], dst_di, &found_key);
1645 /* the existing item matches the logged item */
1646 if (found_key.objectid == log_key.objectid &&
1647 found_key.type == log_key.type &&
1648 found_key.offset == log_key.offset &&
1649 btrfs_dir_type(path->nodes[0], dst_di) == log_type) {
Filipe Mananaa2cc11d2014-09-08 22:53:18 +01001650 update_size = false;
Chris Masone02119d2008-09-05 16:13:11 -04001651 goto out;
1652 }
1653
1654 /*
1655 * don't drop the conflicting directory entry if the inode
1656 * for the new entry doesn't exist
1657 */
Chris Mason4bef0842008-09-08 11:18:08 -04001658 if (!exists)
Chris Masone02119d2008-09-05 16:13:11 -04001659 goto out;
1660
Chris Masone02119d2008-09-05 16:13:11 -04001661 ret = drop_one_dir_item(trans, root, path, dir, dst_di);
Josef Bacik36508602013-04-25 16:23:32 -04001662 if (ret)
1663 goto out;
Chris Masone02119d2008-09-05 16:13:11 -04001664
1665 if (key->type == BTRFS_DIR_INDEX_KEY)
1666 goto insert;
1667out:
David Sterbab3b4aa72011-04-21 01:20:15 +02001668 btrfs_release_path(path);
Josef Bacikd5554382013-09-11 14:17:00 -04001669 if (!ret && update_size) {
1670 btrfs_i_size_write(dir, dir->i_size + name_len * 2);
1671 ret = btrfs_update_inode(trans, root, dir);
1672 }
Chris Masone02119d2008-09-05 16:13:11 -04001673 kfree(name);
1674 iput(dir);
Josef Bacik36508602013-04-25 16:23:32 -04001675 return ret;
Chris Masone02119d2008-09-05 16:13:11 -04001676
1677insert:
David Sterbab3b4aa72011-04-21 01:20:15 +02001678 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04001679 ret = insert_one_name(trans, root, path, key->objectid, key->offset,
1680 name, name_len, log_type, &log_key);
Josef Bacik36508602013-04-25 16:23:32 -04001681 if (ret && ret != -ENOENT)
1682 goto out;
Josef Bacikd5554382013-09-11 14:17:00 -04001683 update_size = false;
Josef Bacik36508602013-04-25 16:23:32 -04001684 ret = 0;
Chris Masone02119d2008-09-05 16:13:11 -04001685 goto out;
1686}
1687
1688/*
1689 * find all the names in a directory item and reconcile them into
1690 * the subvolume. Only BTRFS_DIR_ITEM_KEY types will have more than
1691 * one name in a directory item, but the same code gets used for
1692 * both directory index types
1693 */
1694static noinline int replay_one_dir_item(struct btrfs_trans_handle *trans,
1695 struct btrfs_root *root,
1696 struct btrfs_path *path,
1697 struct extent_buffer *eb, int slot,
1698 struct btrfs_key *key)
1699{
1700 int ret;
1701 u32 item_size = btrfs_item_size_nr(eb, slot);
1702 struct btrfs_dir_item *di;
1703 int name_len;
1704 unsigned long ptr;
1705 unsigned long ptr_end;
1706
1707 ptr = btrfs_item_ptr_offset(eb, slot);
1708 ptr_end = ptr + item_size;
Chris Masond3977122009-01-05 21:25:51 -05001709 while (ptr < ptr_end) {
Chris Masone02119d2008-09-05 16:13:11 -04001710 di = (struct btrfs_dir_item *)ptr;
Josef Bacik22a94d42011-03-16 16:47:17 -04001711 if (verify_dir_item(root, eb, di))
1712 return -EIO;
Chris Masone02119d2008-09-05 16:13:11 -04001713 name_len = btrfs_dir_name_len(eb, di);
1714 ret = replay_one_name(trans, root, path, eb, di, key);
Josef Bacik36508602013-04-25 16:23:32 -04001715 if (ret)
1716 return ret;
Chris Masone02119d2008-09-05 16:13:11 -04001717 ptr = (unsigned long)(di + 1);
1718 ptr += name_len;
1719 }
1720 return 0;
1721}
1722
1723/*
1724 * directory replay has two parts. There are the standard directory
1725 * items in the log copied from the subvolume, and range items
1726 * created in the log while the subvolume was logged.
1727 *
1728 * The range items tell us which parts of the key space the log
1729 * is authoritative for. During replay, if a key in the subvolume
1730 * directory is in a logged range item, but not actually in the log
1731 * that means it was deleted from the directory before the fsync
1732 * and should be removed.
1733 */
1734static noinline int find_dir_range(struct btrfs_root *root,
1735 struct btrfs_path *path,
1736 u64 dirid, int key_type,
1737 u64 *start_ret, u64 *end_ret)
1738{
1739 struct btrfs_key key;
1740 u64 found_end;
1741 struct btrfs_dir_log_item *item;
1742 int ret;
1743 int nritems;
1744
1745 if (*start_ret == (u64)-1)
1746 return 1;
1747
1748 key.objectid = dirid;
1749 key.type = key_type;
1750 key.offset = *start_ret;
1751
1752 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1753 if (ret < 0)
1754 goto out;
1755 if (ret > 0) {
1756 if (path->slots[0] == 0)
1757 goto out;
1758 path->slots[0]--;
1759 }
1760 if (ret != 0)
1761 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1762
1763 if (key.type != key_type || key.objectid != dirid) {
1764 ret = 1;
1765 goto next;
1766 }
1767 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
1768 struct btrfs_dir_log_item);
1769 found_end = btrfs_dir_log_end(path->nodes[0], item);
1770
1771 if (*start_ret >= key.offset && *start_ret <= found_end) {
1772 ret = 0;
1773 *start_ret = key.offset;
1774 *end_ret = found_end;
1775 goto out;
1776 }
1777 ret = 1;
1778next:
1779 /* check the next slot in the tree to see if it is a valid item */
1780 nritems = btrfs_header_nritems(path->nodes[0]);
1781 if (path->slots[0] >= nritems) {
1782 ret = btrfs_next_leaf(root, path);
1783 if (ret)
1784 goto out;
1785 } else {
1786 path->slots[0]++;
1787 }
1788
1789 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1790
1791 if (key.type != key_type || key.objectid != dirid) {
1792 ret = 1;
1793 goto out;
1794 }
1795 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
1796 struct btrfs_dir_log_item);
1797 found_end = btrfs_dir_log_end(path->nodes[0], item);
1798 *start_ret = key.offset;
1799 *end_ret = found_end;
1800 ret = 0;
1801out:
David Sterbab3b4aa72011-04-21 01:20:15 +02001802 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04001803 return ret;
1804}
1805
1806/*
1807 * this looks for a given directory item in the log. If the directory
1808 * item is not in the log, the item is removed and the inode it points
1809 * to is unlinked
1810 */
1811static noinline int check_item_in_log(struct btrfs_trans_handle *trans,
1812 struct btrfs_root *root,
1813 struct btrfs_root *log,
1814 struct btrfs_path *path,
1815 struct btrfs_path *log_path,
1816 struct inode *dir,
1817 struct btrfs_key *dir_key)
1818{
1819 int ret;
1820 struct extent_buffer *eb;
1821 int slot;
1822 u32 item_size;
1823 struct btrfs_dir_item *di;
1824 struct btrfs_dir_item *log_di;
1825 int name_len;
1826 unsigned long ptr;
1827 unsigned long ptr_end;
1828 char *name;
1829 struct inode *inode;
1830 struct btrfs_key location;
1831
1832again:
1833 eb = path->nodes[0];
1834 slot = path->slots[0];
1835 item_size = btrfs_item_size_nr(eb, slot);
1836 ptr = btrfs_item_ptr_offset(eb, slot);
1837 ptr_end = ptr + item_size;
Chris Masond3977122009-01-05 21:25:51 -05001838 while (ptr < ptr_end) {
Chris Masone02119d2008-09-05 16:13:11 -04001839 di = (struct btrfs_dir_item *)ptr;
Josef Bacik22a94d42011-03-16 16:47:17 -04001840 if (verify_dir_item(root, eb, di)) {
1841 ret = -EIO;
1842 goto out;
1843 }
1844
Chris Masone02119d2008-09-05 16:13:11 -04001845 name_len = btrfs_dir_name_len(eb, di);
1846 name = kmalloc(name_len, GFP_NOFS);
1847 if (!name) {
1848 ret = -ENOMEM;
1849 goto out;
1850 }
1851 read_extent_buffer(eb, name, (unsigned long)(di + 1),
1852 name_len);
1853 log_di = NULL;
Chris Mason12fcfd22009-03-24 10:24:20 -04001854 if (log && dir_key->type == BTRFS_DIR_ITEM_KEY) {
Chris Masone02119d2008-09-05 16:13:11 -04001855 log_di = btrfs_lookup_dir_item(trans, log, log_path,
1856 dir_key->objectid,
1857 name, name_len, 0);
Chris Mason12fcfd22009-03-24 10:24:20 -04001858 } else if (log && dir_key->type == BTRFS_DIR_INDEX_KEY) {
Chris Masone02119d2008-09-05 16:13:11 -04001859 log_di = btrfs_lookup_dir_index_item(trans, log,
1860 log_path,
1861 dir_key->objectid,
1862 dir_key->offset,
1863 name, name_len, 0);
1864 }
Filipe David Borba Manana269d0402013-10-28 17:39:21 +00001865 if (!log_di || (IS_ERR(log_di) && PTR_ERR(log_di) == -ENOENT)) {
Chris Masone02119d2008-09-05 16:13:11 -04001866 btrfs_dir_item_key_to_cpu(eb, di, &location);
David Sterbab3b4aa72011-04-21 01:20:15 +02001867 btrfs_release_path(path);
1868 btrfs_release_path(log_path);
Chris Masone02119d2008-09-05 16:13:11 -04001869 inode = read_one_inode(root, location.objectid);
Tsutomu Itohc00e9492011-04-28 09:10:23 +00001870 if (!inode) {
1871 kfree(name);
1872 return -EIO;
1873 }
Chris Masone02119d2008-09-05 16:13:11 -04001874
1875 ret = link_to_fixup_dir(trans, root,
1876 path, location.objectid);
Josef Bacik36508602013-04-25 16:23:32 -04001877 if (ret) {
1878 kfree(name);
1879 iput(inode);
1880 goto out;
1881 }
1882
Zach Brown8b558c52013-10-16 12:10:34 -07001883 inc_nlink(inode);
Chris Masone02119d2008-09-05 16:13:11 -04001884 ret = btrfs_unlink_inode(trans, root, dir, inode,
1885 name, name_len);
Josef Bacik36508602013-04-25 16:23:32 -04001886 if (!ret)
Filipe David Borba Mananaada9af22013-08-05 09:25:47 +01001887 ret = btrfs_run_delayed_items(trans, root);
Chris Masone02119d2008-09-05 16:13:11 -04001888 kfree(name);
1889 iput(inode);
Josef Bacik36508602013-04-25 16:23:32 -04001890 if (ret)
1891 goto out;
Chris Masone02119d2008-09-05 16:13:11 -04001892
1893 /* there might still be more names under this key
1894 * check and repeat if required
1895 */
1896 ret = btrfs_search_slot(NULL, root, dir_key, path,
1897 0, 0);
1898 if (ret == 0)
1899 goto again;
1900 ret = 0;
1901 goto out;
Filipe David Borba Manana269d0402013-10-28 17:39:21 +00001902 } else if (IS_ERR(log_di)) {
1903 kfree(name);
1904 return PTR_ERR(log_di);
Chris Masone02119d2008-09-05 16:13:11 -04001905 }
David Sterbab3b4aa72011-04-21 01:20:15 +02001906 btrfs_release_path(log_path);
Chris Masone02119d2008-09-05 16:13:11 -04001907 kfree(name);
1908
1909 ptr = (unsigned long)(di + 1);
1910 ptr += name_len;
1911 }
1912 ret = 0;
1913out:
David Sterbab3b4aa72011-04-21 01:20:15 +02001914 btrfs_release_path(path);
1915 btrfs_release_path(log_path);
Chris Masone02119d2008-09-05 16:13:11 -04001916 return ret;
1917}
1918
1919/*
1920 * deletion replay happens before we copy any new directory items
1921 * out of the log or out of backreferences from inodes. It
1922 * scans the log to find ranges of keys that log is authoritative for,
1923 * and then scans the directory to find items in those ranges that are
1924 * not present in the log.
1925 *
1926 * Anything we don't find in the log is unlinked and removed from the
1927 * directory.
1928 */
1929static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans,
1930 struct btrfs_root *root,
1931 struct btrfs_root *log,
1932 struct btrfs_path *path,
Chris Mason12fcfd22009-03-24 10:24:20 -04001933 u64 dirid, int del_all)
Chris Masone02119d2008-09-05 16:13:11 -04001934{
1935 u64 range_start;
1936 u64 range_end;
1937 int key_type = BTRFS_DIR_LOG_ITEM_KEY;
1938 int ret = 0;
1939 struct btrfs_key dir_key;
1940 struct btrfs_key found_key;
1941 struct btrfs_path *log_path;
1942 struct inode *dir;
1943
1944 dir_key.objectid = dirid;
1945 dir_key.type = BTRFS_DIR_ITEM_KEY;
1946 log_path = btrfs_alloc_path();
1947 if (!log_path)
1948 return -ENOMEM;
1949
1950 dir = read_one_inode(root, dirid);
1951 /* it isn't an error if the inode isn't there, that can happen
1952 * because we replay the deletes before we copy in the inode item
1953 * from the log
1954 */
1955 if (!dir) {
1956 btrfs_free_path(log_path);
1957 return 0;
1958 }
1959again:
1960 range_start = 0;
1961 range_end = 0;
Chris Masond3977122009-01-05 21:25:51 -05001962 while (1) {
Chris Mason12fcfd22009-03-24 10:24:20 -04001963 if (del_all)
1964 range_end = (u64)-1;
1965 else {
1966 ret = find_dir_range(log, path, dirid, key_type,
1967 &range_start, &range_end);
1968 if (ret != 0)
1969 break;
1970 }
Chris Masone02119d2008-09-05 16:13:11 -04001971
1972 dir_key.offset = range_start;
Chris Masond3977122009-01-05 21:25:51 -05001973 while (1) {
Chris Masone02119d2008-09-05 16:13:11 -04001974 int nritems;
1975 ret = btrfs_search_slot(NULL, root, &dir_key, path,
1976 0, 0);
1977 if (ret < 0)
1978 goto out;
1979
1980 nritems = btrfs_header_nritems(path->nodes[0]);
1981 if (path->slots[0] >= nritems) {
1982 ret = btrfs_next_leaf(root, path);
1983 if (ret)
1984 break;
1985 }
1986 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1987 path->slots[0]);
1988 if (found_key.objectid != dirid ||
1989 found_key.type != dir_key.type)
1990 goto next_type;
1991
1992 if (found_key.offset > range_end)
1993 break;
1994
1995 ret = check_item_in_log(trans, root, log, path,
Chris Mason12fcfd22009-03-24 10:24:20 -04001996 log_path, dir,
1997 &found_key);
Josef Bacik36508602013-04-25 16:23:32 -04001998 if (ret)
1999 goto out;
Chris Masone02119d2008-09-05 16:13:11 -04002000 if (found_key.offset == (u64)-1)
2001 break;
2002 dir_key.offset = found_key.offset + 1;
2003 }
David Sterbab3b4aa72011-04-21 01:20:15 +02002004 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04002005 if (range_end == (u64)-1)
2006 break;
2007 range_start = range_end + 1;
2008 }
2009
2010next_type:
2011 ret = 0;
2012 if (key_type == BTRFS_DIR_LOG_ITEM_KEY) {
2013 key_type = BTRFS_DIR_LOG_INDEX_KEY;
2014 dir_key.type = BTRFS_DIR_INDEX_KEY;
David Sterbab3b4aa72011-04-21 01:20:15 +02002015 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04002016 goto again;
2017 }
2018out:
David Sterbab3b4aa72011-04-21 01:20:15 +02002019 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04002020 btrfs_free_path(log_path);
2021 iput(dir);
2022 return ret;
2023}
2024
2025/*
2026 * the process_func used to replay items from the log tree. This
2027 * gets called in two different stages. The first stage just looks
2028 * for inodes and makes sure they are all copied into the subvolume.
2029 *
2030 * The second stage copies all the other item types from the log into
2031 * the subvolume. The two stage approach is slower, but gets rid of
2032 * lots of complexity around inodes referencing other inodes that exist
2033 * only in the log (references come from either directory items or inode
2034 * back refs).
2035 */
2036static int replay_one_buffer(struct btrfs_root *log, struct extent_buffer *eb,
2037 struct walk_control *wc, u64 gen)
2038{
2039 int nritems;
2040 struct btrfs_path *path;
2041 struct btrfs_root *root = wc->replay_dest;
2042 struct btrfs_key key;
Chris Masone02119d2008-09-05 16:13:11 -04002043 int level;
2044 int i;
2045 int ret;
2046
Tsutomu Itoh018642a2012-05-29 18:10:13 +09002047 ret = btrfs_read_buffer(eb, gen);
2048 if (ret)
2049 return ret;
Chris Masone02119d2008-09-05 16:13:11 -04002050
2051 level = btrfs_header_level(eb);
2052
2053 if (level != 0)
2054 return 0;
2055
2056 path = btrfs_alloc_path();
Mark Fasheh1e5063d2011-07-12 10:46:06 -07002057 if (!path)
2058 return -ENOMEM;
Chris Masone02119d2008-09-05 16:13:11 -04002059
2060 nritems = btrfs_header_nritems(eb);
2061 for (i = 0; i < nritems; i++) {
2062 btrfs_item_key_to_cpu(eb, &key, i);
Chris Masone02119d2008-09-05 16:13:11 -04002063
2064 /* inode keys are done during the first stage */
2065 if (key.type == BTRFS_INODE_ITEM_KEY &&
2066 wc->stage == LOG_WALK_REPLAY_INODES) {
Chris Masone02119d2008-09-05 16:13:11 -04002067 struct btrfs_inode_item *inode_item;
2068 u32 mode;
2069
2070 inode_item = btrfs_item_ptr(eb, i,
2071 struct btrfs_inode_item);
2072 mode = btrfs_inode_mode(eb, inode_item);
2073 if (S_ISDIR(mode)) {
2074 ret = replay_dir_deletes(wc->trans,
Chris Mason12fcfd22009-03-24 10:24:20 -04002075 root, log, path, key.objectid, 0);
Josef Bacikb50c6e22013-04-25 15:55:30 -04002076 if (ret)
2077 break;
Chris Masone02119d2008-09-05 16:13:11 -04002078 }
2079 ret = overwrite_item(wc->trans, root, path,
2080 eb, i, &key);
Josef Bacikb50c6e22013-04-25 15:55:30 -04002081 if (ret)
2082 break;
Chris Masone02119d2008-09-05 16:13:11 -04002083
Yan, Zhengc71bf092009-11-12 09:34:40 +00002084 /* for regular files, make sure corresponding
2085 * orhpan item exist. extents past the new EOF
2086 * will be truncated later by orphan cleanup.
Chris Masone02119d2008-09-05 16:13:11 -04002087 */
2088 if (S_ISREG(mode)) {
Yan, Zhengc71bf092009-11-12 09:34:40 +00002089 ret = insert_orphan_item(wc->trans, root,
2090 key.objectid);
Josef Bacikb50c6e22013-04-25 15:55:30 -04002091 if (ret)
2092 break;
Chris Masone02119d2008-09-05 16:13:11 -04002093 }
Yan, Zhengc71bf092009-11-12 09:34:40 +00002094
Chris Masone02119d2008-09-05 16:13:11 -04002095 ret = link_to_fixup_dir(wc->trans, root,
2096 path, key.objectid);
Josef Bacikb50c6e22013-04-25 15:55:30 -04002097 if (ret)
2098 break;
Chris Masone02119d2008-09-05 16:13:11 -04002099 }
Josef Bacikdd8e7212013-09-11 11:57:23 -04002100
2101 if (key.type == BTRFS_DIR_INDEX_KEY &&
2102 wc->stage == LOG_WALK_REPLAY_DIR_INDEX) {
2103 ret = replay_one_dir_item(wc->trans, root, path,
2104 eb, i, &key);
2105 if (ret)
2106 break;
2107 }
2108
Chris Masone02119d2008-09-05 16:13:11 -04002109 if (wc->stage < LOG_WALK_REPLAY_ALL)
2110 continue;
2111
2112 /* these keys are simply copied */
2113 if (key.type == BTRFS_XATTR_ITEM_KEY) {
2114 ret = overwrite_item(wc->trans, root, path,
2115 eb, i, &key);
Josef Bacikb50c6e22013-04-25 15:55:30 -04002116 if (ret)
2117 break;
Liu Bo2da1c662013-05-26 13:50:29 +00002118 } else if (key.type == BTRFS_INODE_REF_KEY ||
2119 key.type == BTRFS_INODE_EXTREF_KEY) {
Mark Fashehf1863732012-08-08 11:32:27 -07002120 ret = add_inode_ref(wc->trans, root, log, path,
2121 eb, i, &key);
Josef Bacikb50c6e22013-04-25 15:55:30 -04002122 if (ret && ret != -ENOENT)
2123 break;
2124 ret = 0;
Chris Masone02119d2008-09-05 16:13:11 -04002125 } else if (key.type == BTRFS_EXTENT_DATA_KEY) {
2126 ret = replay_one_extent(wc->trans, root, path,
2127 eb, i, &key);
Josef Bacikb50c6e22013-04-25 15:55:30 -04002128 if (ret)
2129 break;
Josef Bacikdd8e7212013-09-11 11:57:23 -04002130 } else if (key.type == BTRFS_DIR_ITEM_KEY) {
Chris Masone02119d2008-09-05 16:13:11 -04002131 ret = replay_one_dir_item(wc->trans, root, path,
2132 eb, i, &key);
Josef Bacikb50c6e22013-04-25 15:55:30 -04002133 if (ret)
2134 break;
Chris Masone02119d2008-09-05 16:13:11 -04002135 }
2136 }
2137 btrfs_free_path(path);
Josef Bacikb50c6e22013-04-25 15:55:30 -04002138 return ret;
Chris Masone02119d2008-09-05 16:13:11 -04002139}
2140
Chris Masond3977122009-01-05 21:25:51 -05002141static noinline int walk_down_log_tree(struct btrfs_trans_handle *trans,
Chris Masone02119d2008-09-05 16:13:11 -04002142 struct btrfs_root *root,
2143 struct btrfs_path *path, int *level,
2144 struct walk_control *wc)
2145{
2146 u64 root_owner;
Chris Masone02119d2008-09-05 16:13:11 -04002147 u64 bytenr;
2148 u64 ptr_gen;
2149 struct extent_buffer *next;
2150 struct extent_buffer *cur;
2151 struct extent_buffer *parent;
2152 u32 blocksize;
2153 int ret = 0;
2154
2155 WARN_ON(*level < 0);
2156 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2157
Chris Masond3977122009-01-05 21:25:51 -05002158 while (*level > 0) {
Chris Masone02119d2008-09-05 16:13:11 -04002159 WARN_ON(*level < 0);
2160 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2161 cur = path->nodes[*level];
2162
Dulshani Gunawardhanafae7f212013-10-31 10:30:08 +05302163 WARN_ON(btrfs_header_level(cur) != *level);
Chris Masone02119d2008-09-05 16:13:11 -04002164
2165 if (path->slots[*level] >=
2166 btrfs_header_nritems(cur))
2167 break;
2168
2169 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
2170 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
David Sterba707e8a02014-06-04 19:22:26 +02002171 blocksize = root->nodesize;
Chris Masone02119d2008-09-05 16:13:11 -04002172
2173 parent = path->nodes[*level];
2174 root_owner = btrfs_header_owner(parent);
Chris Masone02119d2008-09-05 16:13:11 -04002175
2176 next = btrfs_find_create_tree_block(root, bytenr, blocksize);
liubo2a29edc2011-01-26 06:22:08 +00002177 if (!next)
2178 return -ENOMEM;
Chris Masone02119d2008-09-05 16:13:11 -04002179
Chris Masone02119d2008-09-05 16:13:11 -04002180 if (*level == 1) {
Mark Fasheh1e5063d2011-07-12 10:46:06 -07002181 ret = wc->process_func(root, next, wc, ptr_gen);
Josef Bacikb50c6e22013-04-25 15:55:30 -04002182 if (ret) {
2183 free_extent_buffer(next);
Mark Fasheh1e5063d2011-07-12 10:46:06 -07002184 return ret;
Josef Bacikb50c6e22013-04-25 15:55:30 -04002185 }
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002186
Chris Masone02119d2008-09-05 16:13:11 -04002187 path->slots[*level]++;
2188 if (wc->free) {
Tsutomu Itoh018642a2012-05-29 18:10:13 +09002189 ret = btrfs_read_buffer(next, ptr_gen);
2190 if (ret) {
2191 free_extent_buffer(next);
2192 return ret;
2193 }
Chris Masone02119d2008-09-05 16:13:11 -04002194
Josef Bacik681ae502013-10-07 15:11:00 -04002195 if (trans) {
2196 btrfs_tree_lock(next);
2197 btrfs_set_lock_blocking(next);
2198 clean_tree_block(trans, root, next);
2199 btrfs_wait_tree_block_writeback(next);
2200 btrfs_tree_unlock(next);
2201 }
Chris Masone02119d2008-09-05 16:13:11 -04002202
Chris Masone02119d2008-09-05 16:13:11 -04002203 WARN_ON(root_owner !=
2204 BTRFS_TREE_LOG_OBJECTID);
Chris Masone688b7252011-10-31 20:52:39 -04002205 ret = btrfs_free_and_pin_reserved_extent(root,
Chris Masond00aff02008-09-11 15:54:42 -04002206 bytenr, blocksize);
Josef Bacik36508602013-04-25 16:23:32 -04002207 if (ret) {
2208 free_extent_buffer(next);
2209 return ret;
2210 }
Chris Masone02119d2008-09-05 16:13:11 -04002211 }
2212 free_extent_buffer(next);
2213 continue;
2214 }
Tsutomu Itoh018642a2012-05-29 18:10:13 +09002215 ret = btrfs_read_buffer(next, ptr_gen);
2216 if (ret) {
2217 free_extent_buffer(next);
2218 return ret;
2219 }
Chris Masone02119d2008-09-05 16:13:11 -04002220
2221 WARN_ON(*level <= 0);
2222 if (path->nodes[*level-1])
2223 free_extent_buffer(path->nodes[*level-1]);
2224 path->nodes[*level-1] = next;
2225 *level = btrfs_header_level(next);
2226 path->slots[*level] = 0;
2227 cond_resched();
2228 }
2229 WARN_ON(*level < 0);
2230 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2231
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002232 path->slots[*level] = btrfs_header_nritems(path->nodes[*level]);
Chris Masone02119d2008-09-05 16:13:11 -04002233
2234 cond_resched();
2235 return 0;
2236}
2237
Chris Masond3977122009-01-05 21:25:51 -05002238static noinline int walk_up_log_tree(struct btrfs_trans_handle *trans,
Chris Masone02119d2008-09-05 16:13:11 -04002239 struct btrfs_root *root,
2240 struct btrfs_path *path, int *level,
2241 struct walk_control *wc)
2242{
2243 u64 root_owner;
Chris Masone02119d2008-09-05 16:13:11 -04002244 int i;
2245 int slot;
2246 int ret;
2247
Chris Masond3977122009-01-05 21:25:51 -05002248 for (i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) {
Chris Masone02119d2008-09-05 16:13:11 -04002249 slot = path->slots[i];
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002250 if (slot + 1 < btrfs_header_nritems(path->nodes[i])) {
Chris Masone02119d2008-09-05 16:13:11 -04002251 path->slots[i]++;
2252 *level = i;
2253 WARN_ON(*level == 0);
2254 return 0;
2255 } else {
Zheng Yan31840ae2008-09-23 13:14:14 -04002256 struct extent_buffer *parent;
2257 if (path->nodes[*level] == root->node)
2258 parent = path->nodes[*level];
2259 else
2260 parent = path->nodes[*level + 1];
2261
2262 root_owner = btrfs_header_owner(parent);
Mark Fasheh1e5063d2011-07-12 10:46:06 -07002263 ret = wc->process_func(root, path->nodes[*level], wc,
Chris Masone02119d2008-09-05 16:13:11 -04002264 btrfs_header_generation(path->nodes[*level]));
Mark Fasheh1e5063d2011-07-12 10:46:06 -07002265 if (ret)
2266 return ret;
2267
Chris Masone02119d2008-09-05 16:13:11 -04002268 if (wc->free) {
2269 struct extent_buffer *next;
2270
2271 next = path->nodes[*level];
2272
Josef Bacik681ae502013-10-07 15:11:00 -04002273 if (trans) {
2274 btrfs_tree_lock(next);
2275 btrfs_set_lock_blocking(next);
2276 clean_tree_block(trans, root, next);
2277 btrfs_wait_tree_block_writeback(next);
2278 btrfs_tree_unlock(next);
2279 }
Chris Masone02119d2008-09-05 16:13:11 -04002280
Chris Masone02119d2008-09-05 16:13:11 -04002281 WARN_ON(root_owner != BTRFS_TREE_LOG_OBJECTID);
Chris Masone688b7252011-10-31 20:52:39 -04002282 ret = btrfs_free_and_pin_reserved_extent(root,
Chris Masone02119d2008-09-05 16:13:11 -04002283 path->nodes[*level]->start,
Chris Masond00aff02008-09-11 15:54:42 -04002284 path->nodes[*level]->len);
Josef Bacik36508602013-04-25 16:23:32 -04002285 if (ret)
2286 return ret;
Chris Masone02119d2008-09-05 16:13:11 -04002287 }
2288 free_extent_buffer(path->nodes[*level]);
2289 path->nodes[*level] = NULL;
2290 *level = i + 1;
2291 }
2292 }
2293 return 1;
2294}
2295
2296/*
2297 * drop the reference count on the tree rooted at 'snap'. This traverses
2298 * the tree freeing any blocks that have a ref count of zero after being
2299 * decremented.
2300 */
2301static int walk_log_tree(struct btrfs_trans_handle *trans,
2302 struct btrfs_root *log, struct walk_control *wc)
2303{
2304 int ret = 0;
2305 int wret;
2306 int level;
2307 struct btrfs_path *path;
Chris Masone02119d2008-09-05 16:13:11 -04002308 int orig_level;
2309
2310 path = btrfs_alloc_path();
Tsutomu Itohdb5b4932011-03-23 08:14:16 +00002311 if (!path)
2312 return -ENOMEM;
Chris Masone02119d2008-09-05 16:13:11 -04002313
2314 level = btrfs_header_level(log->node);
2315 orig_level = level;
2316 path->nodes[level] = log->node;
2317 extent_buffer_get(log->node);
2318 path->slots[level] = 0;
2319
Chris Masond3977122009-01-05 21:25:51 -05002320 while (1) {
Chris Masone02119d2008-09-05 16:13:11 -04002321 wret = walk_down_log_tree(trans, log, path, &level, wc);
2322 if (wret > 0)
2323 break;
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002324 if (wret < 0) {
Chris Masone02119d2008-09-05 16:13:11 -04002325 ret = wret;
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002326 goto out;
2327 }
Chris Masone02119d2008-09-05 16:13:11 -04002328
2329 wret = walk_up_log_tree(trans, log, path, &level, wc);
2330 if (wret > 0)
2331 break;
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002332 if (wret < 0) {
Chris Masone02119d2008-09-05 16:13:11 -04002333 ret = wret;
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002334 goto out;
2335 }
Chris Masone02119d2008-09-05 16:13:11 -04002336 }
2337
2338 /* was the root node processed? if not, catch it here */
2339 if (path->nodes[orig_level]) {
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002340 ret = wc->process_func(log, path->nodes[orig_level], wc,
Chris Masone02119d2008-09-05 16:13:11 -04002341 btrfs_header_generation(path->nodes[orig_level]));
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002342 if (ret)
2343 goto out;
Chris Masone02119d2008-09-05 16:13:11 -04002344 if (wc->free) {
2345 struct extent_buffer *next;
2346
2347 next = path->nodes[orig_level];
2348
Josef Bacik681ae502013-10-07 15:11:00 -04002349 if (trans) {
2350 btrfs_tree_lock(next);
2351 btrfs_set_lock_blocking(next);
2352 clean_tree_block(trans, log, next);
2353 btrfs_wait_tree_block_writeback(next);
2354 btrfs_tree_unlock(next);
2355 }
Chris Masone02119d2008-09-05 16:13:11 -04002356
Chris Masone02119d2008-09-05 16:13:11 -04002357 WARN_ON(log->root_key.objectid !=
2358 BTRFS_TREE_LOG_OBJECTID);
Chris Masone688b7252011-10-31 20:52:39 -04002359 ret = btrfs_free_and_pin_reserved_extent(log, next->start,
Chris Masond00aff02008-09-11 15:54:42 -04002360 next->len);
Josef Bacik36508602013-04-25 16:23:32 -04002361 if (ret)
2362 goto out;
Chris Masone02119d2008-09-05 16:13:11 -04002363 }
2364 }
2365
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002366out:
Chris Masone02119d2008-09-05 16:13:11 -04002367 btrfs_free_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04002368 return ret;
2369}
2370
Yan Zheng7237f1832009-01-21 12:54:03 -05002371/*
2372 * helper function to update the item for a given subvolumes log root
2373 * in the tree of log roots
2374 */
2375static int update_log_root(struct btrfs_trans_handle *trans,
2376 struct btrfs_root *log)
2377{
2378 int ret;
2379
2380 if (log->log_transid == 1) {
2381 /* insert root item on the first sync */
2382 ret = btrfs_insert_root(trans, log->fs_info->log_root_tree,
2383 &log->root_key, &log->root_item);
2384 } else {
2385 ret = btrfs_update_root(trans, log->fs_info->log_root_tree,
2386 &log->root_key, &log->root_item);
2387 }
2388 return ret;
2389}
2390
Miao Xie8b050d32014-02-20 18:08:58 +08002391static void wait_log_commit(struct btrfs_trans_handle *trans,
2392 struct btrfs_root *root, int transid)
Chris Masone02119d2008-09-05 16:13:11 -04002393{
2394 DEFINE_WAIT(wait);
Yan Zheng7237f1832009-01-21 12:54:03 -05002395 int index = transid % 2;
Chris Masone02119d2008-09-05 16:13:11 -04002396
Yan Zheng7237f1832009-01-21 12:54:03 -05002397 /*
2398 * we only allow two pending log transactions at a time,
2399 * so we know that if ours is more than 2 older than the
2400 * current transaction, we're done
2401 */
Chris Masone02119d2008-09-05 16:13:11 -04002402 do {
Yan Zheng7237f1832009-01-21 12:54:03 -05002403 prepare_to_wait(&root->log_commit_wait[index],
2404 &wait, TASK_UNINTERRUPTIBLE);
2405 mutex_unlock(&root->log_mutex);
Chris Mason12fcfd22009-03-24 10:24:20 -04002406
Miao Xied1433de2014-02-20 18:08:59 +08002407 if (root->log_transid_committed < transid &&
Yan Zheng7237f1832009-01-21 12:54:03 -05002408 atomic_read(&root->log_commit[index]))
Chris Masone02119d2008-09-05 16:13:11 -04002409 schedule();
Chris Mason12fcfd22009-03-24 10:24:20 -04002410
Yan Zheng7237f1832009-01-21 12:54:03 -05002411 finish_wait(&root->log_commit_wait[index], &wait);
2412 mutex_lock(&root->log_mutex);
Miao Xied1433de2014-02-20 18:08:59 +08002413 } while (root->log_transid_committed < transid &&
Yan Zheng7237f1832009-01-21 12:54:03 -05002414 atomic_read(&root->log_commit[index]));
Yan Zheng7237f1832009-01-21 12:54:03 -05002415}
2416
Jeff Mahoney143bede2012-03-01 14:56:26 +01002417static void wait_for_writer(struct btrfs_trans_handle *trans,
2418 struct btrfs_root *root)
Yan Zheng7237f1832009-01-21 12:54:03 -05002419{
2420 DEFINE_WAIT(wait);
Miao Xie8b050d32014-02-20 18:08:58 +08002421
2422 while (atomic_read(&root->log_writers)) {
Yan Zheng7237f1832009-01-21 12:54:03 -05002423 prepare_to_wait(&root->log_writer_wait,
2424 &wait, TASK_UNINTERRUPTIBLE);
2425 mutex_unlock(&root->log_mutex);
Miao Xie8b050d32014-02-20 18:08:58 +08002426 if (atomic_read(&root->log_writers))
Yan Zheng7237f1832009-01-21 12:54:03 -05002427 schedule();
2428 mutex_lock(&root->log_mutex);
2429 finish_wait(&root->log_writer_wait, &wait);
2430 }
Chris Masone02119d2008-09-05 16:13:11 -04002431}
2432
Miao Xie8b050d32014-02-20 18:08:58 +08002433static inline void btrfs_remove_log_ctx(struct btrfs_root *root,
2434 struct btrfs_log_ctx *ctx)
2435{
2436 if (!ctx)
2437 return;
2438
2439 mutex_lock(&root->log_mutex);
2440 list_del_init(&ctx->list);
2441 mutex_unlock(&root->log_mutex);
2442}
2443
2444/*
2445 * Invoked in log mutex context, or be sure there is no other task which
2446 * can access the list.
2447 */
2448static inline void btrfs_remove_all_log_ctxs(struct btrfs_root *root,
2449 int index, int error)
2450{
2451 struct btrfs_log_ctx *ctx;
2452
2453 if (!error) {
2454 INIT_LIST_HEAD(&root->log_ctxs[index]);
2455 return;
2456 }
2457
2458 list_for_each_entry(ctx, &root->log_ctxs[index], list)
2459 ctx->log_ret = error;
2460
2461 INIT_LIST_HEAD(&root->log_ctxs[index]);
2462}
2463
Chris Masone02119d2008-09-05 16:13:11 -04002464/*
2465 * btrfs_sync_log does sends a given tree log down to the disk and
2466 * updates the super blocks to record it. When this call is done,
Chris Mason12fcfd22009-03-24 10:24:20 -04002467 * you know that any inodes previously logged are safely on disk only
2468 * if it returns 0.
2469 *
2470 * Any other return value means you need to call btrfs_commit_transaction.
2471 * Some of the edge cases for fsyncing directories that have had unlinks
2472 * or renames done in the past mean that sometimes the only safe
2473 * fsync is to commit the whole FS. When btrfs_sync_log returns -EAGAIN,
2474 * that has happened.
Chris Masone02119d2008-09-05 16:13:11 -04002475 */
2476int btrfs_sync_log(struct btrfs_trans_handle *trans,
Miao Xie8b050d32014-02-20 18:08:58 +08002477 struct btrfs_root *root, struct btrfs_log_ctx *ctx)
Chris Masone02119d2008-09-05 16:13:11 -04002478{
Yan Zheng7237f1832009-01-21 12:54:03 -05002479 int index1;
2480 int index2;
Yan, Zheng8cef4e12009-11-12 09:33:26 +00002481 int mark;
Chris Masone02119d2008-09-05 16:13:11 -04002482 int ret;
Chris Masone02119d2008-09-05 16:13:11 -04002483 struct btrfs_root *log = root->log_root;
Yan Zheng7237f1832009-01-21 12:54:03 -05002484 struct btrfs_root *log_root_tree = root->fs_info->log_root_tree;
Miao Xiebb14a592014-02-20 18:08:56 +08002485 int log_transid = 0;
Miao Xie8b050d32014-02-20 18:08:58 +08002486 struct btrfs_log_ctx root_log_ctx;
Miao Xiec6adc9c2013-05-28 10:05:39 +00002487 struct blk_plug plug;
Chris Masone02119d2008-09-05 16:13:11 -04002488
Yan Zheng7237f1832009-01-21 12:54:03 -05002489 mutex_lock(&root->log_mutex);
Miao Xied1433de2014-02-20 18:08:59 +08002490 log_transid = ctx->log_transid;
2491 if (root->log_transid_committed >= log_transid) {
Yan Zheng7237f1832009-01-21 12:54:03 -05002492 mutex_unlock(&root->log_mutex);
Miao Xie8b050d32014-02-20 18:08:58 +08002493 return ctx->log_ret;
Chris Masone02119d2008-09-05 16:13:11 -04002494 }
Miao Xied1433de2014-02-20 18:08:59 +08002495
2496 index1 = log_transid % 2;
2497 if (atomic_read(&root->log_commit[index1])) {
2498 wait_log_commit(trans, root, log_transid);
2499 mutex_unlock(&root->log_mutex);
2500 return ctx->log_ret;
2501 }
2502 ASSERT(log_transid == root->log_transid);
Yan Zheng7237f1832009-01-21 12:54:03 -05002503 atomic_set(&root->log_commit[index1], 1);
2504
2505 /* wait for previous tree log sync to complete */
2506 if (atomic_read(&root->log_commit[(index1 + 1) % 2]))
Miao Xied1433de2014-02-20 18:08:59 +08002507 wait_log_commit(trans, root, log_transid - 1);
Miao Xie48cab2e2014-02-20 18:08:52 +08002508
Yan, Zheng86df7eb2009-10-14 09:24:59 -04002509 while (1) {
Miao Xie2ecb7922012-09-06 04:04:27 -06002510 int batch = atomic_read(&root->log_batch);
Chris Masoncd354ad2011-10-20 15:45:37 -04002511 /* when we're on an ssd, just kick the log commit out */
Miao Xie27cdeb72014-04-02 19:51:05 +08002512 if (!btrfs_test_opt(root, SSD) &&
2513 test_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state)) {
Yan, Zheng86df7eb2009-10-14 09:24:59 -04002514 mutex_unlock(&root->log_mutex);
2515 schedule_timeout_uninterruptible(1);
2516 mutex_lock(&root->log_mutex);
2517 }
Chris Mason12fcfd22009-03-24 10:24:20 -04002518 wait_for_writer(trans, root);
Miao Xie2ecb7922012-09-06 04:04:27 -06002519 if (batch == atomic_read(&root->log_batch))
Chris Masone02119d2008-09-05 16:13:11 -04002520 break;
2521 }
Chris Masond0c803c2008-09-11 16:17:57 -04002522
Chris Mason12fcfd22009-03-24 10:24:20 -04002523 /* bail out if we need to do a full commit */
Miao Xie995946d2014-04-02 19:51:06 +08002524 if (btrfs_need_log_full_commit(root->fs_info, trans)) {
Chris Mason12fcfd22009-03-24 10:24:20 -04002525 ret = -EAGAIN;
Josef Bacik2ab28f32012-10-12 15:27:49 -04002526 btrfs_free_logged_extents(log, log_transid);
Chris Mason12fcfd22009-03-24 10:24:20 -04002527 mutex_unlock(&root->log_mutex);
2528 goto out;
2529 }
2530
Yan, Zheng8cef4e12009-11-12 09:33:26 +00002531 if (log_transid % 2 == 0)
2532 mark = EXTENT_DIRTY;
2533 else
2534 mark = EXTENT_NEW;
2535
Chris Mason690587d2009-10-13 13:29:19 -04002536 /* we start IO on all the marked extents here, but we don't actually
2537 * wait for them until later.
2538 */
Miao Xiec6adc9c2013-05-28 10:05:39 +00002539 blk_start_plug(&plug);
Yan, Zheng8cef4e12009-11-12 09:33:26 +00002540 ret = btrfs_write_marked_extents(log, &log->dirty_log_pages, mark);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002541 if (ret) {
Miao Xiec6adc9c2013-05-28 10:05:39 +00002542 blk_finish_plug(&plug);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002543 btrfs_abort_transaction(trans, root, ret);
Josef Bacik2ab28f32012-10-12 15:27:49 -04002544 btrfs_free_logged_extents(log, log_transid);
Miao Xie995946d2014-04-02 19:51:06 +08002545 btrfs_set_log_full_commit(root->fs_info, trans);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002546 mutex_unlock(&root->log_mutex);
2547 goto out;
2548 }
Yan Zheng7237f1832009-01-21 12:54:03 -05002549
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002550 btrfs_set_root_node(&log->root_item, log->node);
Yan Zheng7237f1832009-01-21 12:54:03 -05002551
Yan Zheng7237f1832009-01-21 12:54:03 -05002552 root->log_transid++;
2553 log->log_transid = root->log_transid;
Josef Bacikff782e02009-10-08 15:30:04 -04002554 root->log_start_pid = 0;
Yan Zheng7237f1832009-01-21 12:54:03 -05002555 /*
Yan, Zheng8cef4e12009-11-12 09:33:26 +00002556 * IO has been started, blocks of the log tree have WRITTEN flag set
2557 * in their headers. new modifications of the log will be written to
2558 * new positions. so it's safe to allow log writers to go in.
Yan Zheng7237f1832009-01-21 12:54:03 -05002559 */
2560 mutex_unlock(&root->log_mutex);
2561
Miao Xied1433de2014-02-20 18:08:59 +08002562 btrfs_init_log_ctx(&root_log_ctx);
2563
Yan Zheng7237f1832009-01-21 12:54:03 -05002564 mutex_lock(&log_root_tree->log_mutex);
Miao Xie2ecb7922012-09-06 04:04:27 -06002565 atomic_inc(&log_root_tree->log_batch);
Yan Zheng7237f1832009-01-21 12:54:03 -05002566 atomic_inc(&log_root_tree->log_writers);
Miao Xied1433de2014-02-20 18:08:59 +08002567
2568 index2 = log_root_tree->log_transid % 2;
2569 list_add_tail(&root_log_ctx.list, &log_root_tree->log_ctxs[index2]);
2570 root_log_ctx.log_transid = log_root_tree->log_transid;
2571
Yan Zheng7237f1832009-01-21 12:54:03 -05002572 mutex_unlock(&log_root_tree->log_mutex);
2573
2574 ret = update_log_root(trans, log);
Yan Zheng7237f1832009-01-21 12:54:03 -05002575
2576 mutex_lock(&log_root_tree->log_mutex);
2577 if (atomic_dec_and_test(&log_root_tree->log_writers)) {
2578 smp_mb();
2579 if (waitqueue_active(&log_root_tree->log_writer_wait))
2580 wake_up(&log_root_tree->log_writer_wait);
2581 }
2582
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002583 if (ret) {
Miao Xied1433de2014-02-20 18:08:59 +08002584 if (!list_empty(&root_log_ctx.list))
2585 list_del_init(&root_log_ctx.list);
2586
Miao Xiec6adc9c2013-05-28 10:05:39 +00002587 blk_finish_plug(&plug);
Miao Xie995946d2014-04-02 19:51:06 +08002588 btrfs_set_log_full_commit(root->fs_info, trans);
2589
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002590 if (ret != -ENOSPC) {
2591 btrfs_abort_transaction(trans, root, ret);
2592 mutex_unlock(&log_root_tree->log_mutex);
2593 goto out;
2594 }
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002595 btrfs_wait_marked_extents(log, &log->dirty_log_pages, mark);
Josef Bacik2ab28f32012-10-12 15:27:49 -04002596 btrfs_free_logged_extents(log, log_transid);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002597 mutex_unlock(&log_root_tree->log_mutex);
2598 ret = -EAGAIN;
2599 goto out;
2600 }
2601
Miao Xied1433de2014-02-20 18:08:59 +08002602 if (log_root_tree->log_transid_committed >= root_log_ctx.log_transid) {
2603 mutex_unlock(&log_root_tree->log_mutex);
2604 ret = root_log_ctx.log_ret;
2605 goto out;
2606 }
Miao Xie8b050d32014-02-20 18:08:58 +08002607
Miao Xied1433de2014-02-20 18:08:59 +08002608 index2 = root_log_ctx.log_transid % 2;
Yan Zheng7237f1832009-01-21 12:54:03 -05002609 if (atomic_read(&log_root_tree->log_commit[index2])) {
Miao Xiec6adc9c2013-05-28 10:05:39 +00002610 blk_finish_plug(&plug);
Filipe Manana5ab5e442014-11-13 16:59:53 +00002611 ret = btrfs_wait_marked_extents(log, &log->dirty_log_pages,
2612 mark);
Josef Bacik50d9aa92014-11-21 14:52:38 -05002613 btrfs_wait_logged_extents(trans, log, log_transid);
Miao Xie8b050d32014-02-20 18:08:58 +08002614 wait_log_commit(trans, log_root_tree,
Miao Xied1433de2014-02-20 18:08:59 +08002615 root_log_ctx.log_transid);
Yan Zheng7237f1832009-01-21 12:54:03 -05002616 mutex_unlock(&log_root_tree->log_mutex);
Filipe Manana5ab5e442014-11-13 16:59:53 +00002617 if (!ret)
2618 ret = root_log_ctx.log_ret;
Yan Zheng7237f1832009-01-21 12:54:03 -05002619 goto out;
2620 }
Miao Xied1433de2014-02-20 18:08:59 +08002621 ASSERT(root_log_ctx.log_transid == log_root_tree->log_transid);
Yan Zheng7237f1832009-01-21 12:54:03 -05002622 atomic_set(&log_root_tree->log_commit[index2], 1);
2623
Chris Mason12fcfd22009-03-24 10:24:20 -04002624 if (atomic_read(&log_root_tree->log_commit[(index2 + 1) % 2])) {
2625 wait_log_commit(trans, log_root_tree,
Miao Xied1433de2014-02-20 18:08:59 +08002626 root_log_ctx.log_transid - 1);
Chris Mason12fcfd22009-03-24 10:24:20 -04002627 }
Yan Zheng7237f1832009-01-21 12:54:03 -05002628
Chris Mason12fcfd22009-03-24 10:24:20 -04002629 wait_for_writer(trans, log_root_tree);
2630
2631 /*
2632 * now that we've moved on to the tree of log tree roots,
2633 * check the full commit flag again
2634 */
Miao Xie995946d2014-04-02 19:51:06 +08002635 if (btrfs_need_log_full_commit(root->fs_info, trans)) {
Miao Xiec6adc9c2013-05-28 10:05:39 +00002636 blk_finish_plug(&plug);
Yan, Zheng8cef4e12009-11-12 09:33:26 +00002637 btrfs_wait_marked_extents(log, &log->dirty_log_pages, mark);
Josef Bacik2ab28f32012-10-12 15:27:49 -04002638 btrfs_free_logged_extents(log, log_transid);
Chris Mason12fcfd22009-03-24 10:24:20 -04002639 mutex_unlock(&log_root_tree->log_mutex);
2640 ret = -EAGAIN;
2641 goto out_wake_log_root;
2642 }
Yan Zheng7237f1832009-01-21 12:54:03 -05002643
Miao Xiec6adc9c2013-05-28 10:05:39 +00002644 ret = btrfs_write_marked_extents(log_root_tree,
2645 &log_root_tree->dirty_log_pages,
2646 EXTENT_DIRTY | EXTENT_NEW);
2647 blk_finish_plug(&plug);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002648 if (ret) {
Miao Xie995946d2014-04-02 19:51:06 +08002649 btrfs_set_log_full_commit(root->fs_info, trans);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002650 btrfs_abort_transaction(trans, root, ret);
Josef Bacik2ab28f32012-10-12 15:27:49 -04002651 btrfs_free_logged_extents(log, log_transid);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002652 mutex_unlock(&log_root_tree->log_mutex);
2653 goto out_wake_log_root;
2654 }
Filipe Manana5ab5e442014-11-13 16:59:53 +00002655 ret = btrfs_wait_marked_extents(log, &log->dirty_log_pages, mark);
2656 if (!ret)
2657 ret = btrfs_wait_marked_extents(log_root_tree,
2658 &log_root_tree->dirty_log_pages,
2659 EXTENT_NEW | EXTENT_DIRTY);
2660 if (ret) {
2661 btrfs_set_log_full_commit(root->fs_info, trans);
2662 btrfs_free_logged_extents(log, log_transid);
2663 mutex_unlock(&log_root_tree->log_mutex);
2664 goto out_wake_log_root;
2665 }
Josef Bacik50d9aa92014-11-21 14:52:38 -05002666 btrfs_wait_logged_extents(trans, log, log_transid);
Chris Masone02119d2008-09-05 16:13:11 -04002667
David Sterba6c417612011-04-13 15:41:04 +02002668 btrfs_set_super_log_root(root->fs_info->super_for_commit,
Yan Zheng7237f1832009-01-21 12:54:03 -05002669 log_root_tree->node->start);
David Sterba6c417612011-04-13 15:41:04 +02002670 btrfs_set_super_log_root_level(root->fs_info->super_for_commit,
Yan Zheng7237f1832009-01-21 12:54:03 -05002671 btrfs_header_level(log_root_tree->node));
Chris Masone02119d2008-09-05 16:13:11 -04002672
Yan Zheng7237f1832009-01-21 12:54:03 -05002673 log_root_tree->log_transid++;
Yan Zheng7237f1832009-01-21 12:54:03 -05002674 mutex_unlock(&log_root_tree->log_mutex);
2675
2676 /*
2677 * nobody else is going to jump in and write the the ctree
2678 * super here because the log_commit atomic below is protecting
2679 * us. We must be called with a transaction handle pinning
2680 * the running transaction open, so a full commit can't hop
2681 * in and cause problems either.
2682 */
Stefan Behrens5af3e8c2012-08-01 18:56:49 +02002683 ret = write_ctree_super(trans, root->fs_info->tree_root, 1);
Stefan Behrens5af3e8c2012-08-01 18:56:49 +02002684 if (ret) {
Miao Xie995946d2014-04-02 19:51:06 +08002685 btrfs_set_log_full_commit(root->fs_info, trans);
Stefan Behrens5af3e8c2012-08-01 18:56:49 +02002686 btrfs_abort_transaction(trans, root, ret);
2687 goto out_wake_log_root;
2688 }
Yan Zheng7237f1832009-01-21 12:54:03 -05002689
Chris Mason257c62e2009-10-13 13:21:08 -04002690 mutex_lock(&root->log_mutex);
2691 if (root->last_log_commit < log_transid)
2692 root->last_log_commit = log_transid;
2693 mutex_unlock(&root->log_mutex);
2694
Chris Mason12fcfd22009-03-24 10:24:20 -04002695out_wake_log_root:
Miao Xie8b050d32014-02-20 18:08:58 +08002696 /*
2697 * We needn't get log_mutex here because we are sure all
2698 * the other tasks are blocked.
2699 */
2700 btrfs_remove_all_log_ctxs(log_root_tree, index2, ret);
2701
Miao Xied1433de2014-02-20 18:08:59 +08002702 mutex_lock(&log_root_tree->log_mutex);
2703 log_root_tree->log_transid_committed++;
Yan Zheng7237f1832009-01-21 12:54:03 -05002704 atomic_set(&log_root_tree->log_commit[index2], 0);
Miao Xied1433de2014-02-20 18:08:59 +08002705 mutex_unlock(&log_root_tree->log_mutex);
2706
Yan Zheng7237f1832009-01-21 12:54:03 -05002707 if (waitqueue_active(&log_root_tree->log_commit_wait[index2]))
2708 wake_up(&log_root_tree->log_commit_wait[index2]);
Chris Masone02119d2008-09-05 16:13:11 -04002709out:
Miao Xie8b050d32014-02-20 18:08:58 +08002710 /* See above. */
2711 btrfs_remove_all_log_ctxs(root, index1, ret);
2712
Miao Xied1433de2014-02-20 18:08:59 +08002713 mutex_lock(&root->log_mutex);
2714 root->log_transid_committed++;
Yan Zheng7237f1832009-01-21 12:54:03 -05002715 atomic_set(&root->log_commit[index1], 0);
Miao Xied1433de2014-02-20 18:08:59 +08002716 mutex_unlock(&root->log_mutex);
Miao Xie8b050d32014-02-20 18:08:58 +08002717
Yan Zheng7237f1832009-01-21 12:54:03 -05002718 if (waitqueue_active(&root->log_commit_wait[index1]))
2719 wake_up(&root->log_commit_wait[index1]);
Chris Masonb31eabd2011-01-31 16:48:24 -05002720 return ret;
Chris Masone02119d2008-09-05 16:13:11 -04002721}
2722
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002723static void free_log_tree(struct btrfs_trans_handle *trans,
2724 struct btrfs_root *log)
Chris Masone02119d2008-09-05 16:13:11 -04002725{
2726 int ret;
Chris Masond0c803c2008-09-11 16:17:57 -04002727 u64 start;
2728 u64 end;
Chris Masone02119d2008-09-05 16:13:11 -04002729 struct walk_control wc = {
2730 .free = 1,
2731 .process_func = process_one_buffer
2732 };
2733
Josef Bacik681ae502013-10-07 15:11:00 -04002734 ret = walk_log_tree(trans, log, &wc);
2735 /* I don't think this can happen but just in case */
2736 if (ret)
2737 btrfs_abort_transaction(trans, log, ret);
Chris Masone02119d2008-09-05 16:13:11 -04002738
Chris Masond3977122009-01-05 21:25:51 -05002739 while (1) {
Chris Masond0c803c2008-09-11 16:17:57 -04002740 ret = find_first_extent_bit(&log->dirty_log_pages,
Josef Bacike6138872012-09-27 17:07:30 -04002741 0, &start, &end, EXTENT_DIRTY | EXTENT_NEW,
2742 NULL);
Chris Masond0c803c2008-09-11 16:17:57 -04002743 if (ret)
2744 break;
2745
Yan, Zheng8cef4e12009-11-12 09:33:26 +00002746 clear_extent_bits(&log->dirty_log_pages, start, end,
2747 EXTENT_DIRTY | EXTENT_NEW, GFP_NOFS);
Chris Masond0c803c2008-09-11 16:17:57 -04002748 }
2749
Josef Bacik2ab28f32012-10-12 15:27:49 -04002750 /*
2751 * We may have short-circuited the log tree with the full commit logic
2752 * and left ordered extents on our list, so clear these out to keep us
2753 * from leaking inodes and memory.
2754 */
2755 btrfs_free_logged_extents(log, 0);
2756 btrfs_free_logged_extents(log, 1);
2757
Yan Zheng7237f1832009-01-21 12:54:03 -05002758 free_extent_buffer(log->node);
2759 kfree(log);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002760}
2761
2762/*
2763 * free all the extents used by the tree log. This should be called
2764 * at commit time of the full transaction
2765 */
2766int btrfs_free_log(struct btrfs_trans_handle *trans, struct btrfs_root *root)
2767{
2768 if (root->log_root) {
2769 free_log_tree(trans, root->log_root);
2770 root->log_root = NULL;
2771 }
2772 return 0;
2773}
2774
2775int btrfs_free_log_root_tree(struct btrfs_trans_handle *trans,
2776 struct btrfs_fs_info *fs_info)
2777{
2778 if (fs_info->log_root_tree) {
2779 free_log_tree(trans, fs_info->log_root_tree);
2780 fs_info->log_root_tree = NULL;
2781 }
Chris Masone02119d2008-09-05 16:13:11 -04002782 return 0;
2783}
2784
2785/*
Chris Masone02119d2008-09-05 16:13:11 -04002786 * If both a file and directory are logged, and unlinks or renames are
2787 * mixed in, we have a few interesting corners:
2788 *
2789 * create file X in dir Y
2790 * link file X to X.link in dir Y
2791 * fsync file X
2792 * unlink file X but leave X.link
2793 * fsync dir Y
2794 *
2795 * After a crash we would expect only X.link to exist. But file X
2796 * didn't get fsync'd again so the log has back refs for X and X.link.
2797 *
2798 * We solve this by removing directory entries and inode backrefs from the
2799 * log when a file that was logged in the current transaction is
2800 * unlinked. Any later fsync will include the updated log entries, and
2801 * we'll be able to reconstruct the proper directory items from backrefs.
2802 *
2803 * This optimizations allows us to avoid relogging the entire inode
2804 * or the entire directory.
2805 */
2806int btrfs_del_dir_entries_in_log(struct btrfs_trans_handle *trans,
2807 struct btrfs_root *root,
2808 const char *name, int name_len,
2809 struct inode *dir, u64 index)
2810{
2811 struct btrfs_root *log;
2812 struct btrfs_dir_item *di;
2813 struct btrfs_path *path;
2814 int ret;
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002815 int err = 0;
Chris Masone02119d2008-09-05 16:13:11 -04002816 int bytes_del = 0;
Li Zefan33345d012011-04-20 10:31:50 +08002817 u64 dir_ino = btrfs_ino(dir);
Chris Masone02119d2008-09-05 16:13:11 -04002818
Chris Mason3a5f1d42008-09-11 15:53:37 -04002819 if (BTRFS_I(dir)->logged_trans < trans->transid)
2820 return 0;
2821
Chris Masone02119d2008-09-05 16:13:11 -04002822 ret = join_running_log_trans(root);
2823 if (ret)
2824 return 0;
2825
2826 mutex_lock(&BTRFS_I(dir)->log_mutex);
2827
2828 log = root->log_root;
2829 path = btrfs_alloc_path();
Tsutomu Itoha62f44a2011-04-25 19:43:51 -04002830 if (!path) {
2831 err = -ENOMEM;
2832 goto out_unlock;
2833 }
liubo2a29edc2011-01-26 06:22:08 +00002834
Li Zefan33345d012011-04-20 10:31:50 +08002835 di = btrfs_lookup_dir_item(trans, log, path, dir_ino,
Chris Masone02119d2008-09-05 16:13:11 -04002836 name, name_len, -1);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002837 if (IS_ERR(di)) {
2838 err = PTR_ERR(di);
2839 goto fail;
2840 }
2841 if (di) {
Chris Masone02119d2008-09-05 16:13:11 -04002842 ret = btrfs_delete_one_dir_name(trans, log, path, di);
2843 bytes_del += name_len;
Josef Bacik36508602013-04-25 16:23:32 -04002844 if (ret) {
2845 err = ret;
2846 goto fail;
2847 }
Chris Masone02119d2008-09-05 16:13:11 -04002848 }
David Sterbab3b4aa72011-04-21 01:20:15 +02002849 btrfs_release_path(path);
Li Zefan33345d012011-04-20 10:31:50 +08002850 di = btrfs_lookup_dir_index_item(trans, log, path, dir_ino,
Chris Masone02119d2008-09-05 16:13:11 -04002851 index, name, name_len, -1);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002852 if (IS_ERR(di)) {
2853 err = PTR_ERR(di);
2854 goto fail;
2855 }
2856 if (di) {
Chris Masone02119d2008-09-05 16:13:11 -04002857 ret = btrfs_delete_one_dir_name(trans, log, path, di);
2858 bytes_del += name_len;
Josef Bacik36508602013-04-25 16:23:32 -04002859 if (ret) {
2860 err = ret;
2861 goto fail;
2862 }
Chris Masone02119d2008-09-05 16:13:11 -04002863 }
2864
2865 /* update the directory size in the log to reflect the names
2866 * we have removed
2867 */
2868 if (bytes_del) {
2869 struct btrfs_key key;
2870
Li Zefan33345d012011-04-20 10:31:50 +08002871 key.objectid = dir_ino;
Chris Masone02119d2008-09-05 16:13:11 -04002872 key.offset = 0;
2873 key.type = BTRFS_INODE_ITEM_KEY;
David Sterbab3b4aa72011-04-21 01:20:15 +02002874 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04002875
2876 ret = btrfs_search_slot(trans, log, &key, path, 0, 1);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002877 if (ret < 0) {
2878 err = ret;
2879 goto fail;
2880 }
Chris Masone02119d2008-09-05 16:13:11 -04002881 if (ret == 0) {
2882 struct btrfs_inode_item *item;
2883 u64 i_size;
2884
2885 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2886 struct btrfs_inode_item);
2887 i_size = btrfs_inode_size(path->nodes[0], item);
2888 if (i_size > bytes_del)
2889 i_size -= bytes_del;
2890 else
2891 i_size = 0;
2892 btrfs_set_inode_size(path->nodes[0], item, i_size);
2893 btrfs_mark_buffer_dirty(path->nodes[0]);
2894 } else
2895 ret = 0;
David Sterbab3b4aa72011-04-21 01:20:15 +02002896 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04002897 }
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002898fail:
Chris Masone02119d2008-09-05 16:13:11 -04002899 btrfs_free_path(path);
Tsutomu Itoha62f44a2011-04-25 19:43:51 -04002900out_unlock:
Chris Masone02119d2008-09-05 16:13:11 -04002901 mutex_unlock(&BTRFS_I(dir)->log_mutex);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002902 if (ret == -ENOSPC) {
Miao Xie995946d2014-04-02 19:51:06 +08002903 btrfs_set_log_full_commit(root->fs_info, trans);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002904 ret = 0;
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002905 } else if (ret < 0)
2906 btrfs_abort_transaction(trans, root, ret);
2907
Chris Mason12fcfd22009-03-24 10:24:20 -04002908 btrfs_end_log_trans(root);
Chris Masone02119d2008-09-05 16:13:11 -04002909
Andi Kleen411fc6b2010-10-29 15:14:31 -04002910 return err;
Chris Masone02119d2008-09-05 16:13:11 -04002911}
2912
2913/* see comments for btrfs_del_dir_entries_in_log */
2914int btrfs_del_inode_ref_in_log(struct btrfs_trans_handle *trans,
2915 struct btrfs_root *root,
2916 const char *name, int name_len,
2917 struct inode *inode, u64 dirid)
2918{
2919 struct btrfs_root *log;
2920 u64 index;
2921 int ret;
2922
Chris Mason3a5f1d42008-09-11 15:53:37 -04002923 if (BTRFS_I(inode)->logged_trans < trans->transid)
2924 return 0;
2925
Chris Masone02119d2008-09-05 16:13:11 -04002926 ret = join_running_log_trans(root);
2927 if (ret)
2928 return 0;
2929 log = root->log_root;
2930 mutex_lock(&BTRFS_I(inode)->log_mutex);
2931
Li Zefan33345d012011-04-20 10:31:50 +08002932 ret = btrfs_del_inode_ref(trans, log, name, name_len, btrfs_ino(inode),
Chris Masone02119d2008-09-05 16:13:11 -04002933 dirid, &index);
2934 mutex_unlock(&BTRFS_I(inode)->log_mutex);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002935 if (ret == -ENOSPC) {
Miao Xie995946d2014-04-02 19:51:06 +08002936 btrfs_set_log_full_commit(root->fs_info, trans);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002937 ret = 0;
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002938 } else if (ret < 0 && ret != -ENOENT)
2939 btrfs_abort_transaction(trans, root, ret);
Chris Mason12fcfd22009-03-24 10:24:20 -04002940 btrfs_end_log_trans(root);
Chris Masone02119d2008-09-05 16:13:11 -04002941
Chris Masone02119d2008-09-05 16:13:11 -04002942 return ret;
2943}
2944
2945/*
2946 * creates a range item in the log for 'dirid'. first_offset and
2947 * last_offset tell us which parts of the key space the log should
2948 * be considered authoritative for.
2949 */
2950static noinline int insert_dir_log_key(struct btrfs_trans_handle *trans,
2951 struct btrfs_root *log,
2952 struct btrfs_path *path,
2953 int key_type, u64 dirid,
2954 u64 first_offset, u64 last_offset)
2955{
2956 int ret;
2957 struct btrfs_key key;
2958 struct btrfs_dir_log_item *item;
2959
2960 key.objectid = dirid;
2961 key.offset = first_offset;
2962 if (key_type == BTRFS_DIR_ITEM_KEY)
2963 key.type = BTRFS_DIR_LOG_ITEM_KEY;
2964 else
2965 key.type = BTRFS_DIR_LOG_INDEX_KEY;
2966 ret = btrfs_insert_empty_item(trans, log, path, &key, sizeof(*item));
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002967 if (ret)
2968 return ret;
Chris Masone02119d2008-09-05 16:13:11 -04002969
2970 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2971 struct btrfs_dir_log_item);
2972 btrfs_set_dir_log_end(path->nodes[0], item, last_offset);
2973 btrfs_mark_buffer_dirty(path->nodes[0]);
David Sterbab3b4aa72011-04-21 01:20:15 +02002974 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04002975 return 0;
2976}
2977
2978/*
2979 * log all the items included in the current transaction for a given
2980 * directory. This also creates the range items in the log tree required
2981 * to replay anything deleted before the fsync
2982 */
2983static noinline int log_dir_items(struct btrfs_trans_handle *trans,
2984 struct btrfs_root *root, struct inode *inode,
2985 struct btrfs_path *path,
2986 struct btrfs_path *dst_path, int key_type,
2987 u64 min_offset, u64 *last_offset_ret)
2988{
2989 struct btrfs_key min_key;
Chris Masone02119d2008-09-05 16:13:11 -04002990 struct btrfs_root *log = root->log_root;
2991 struct extent_buffer *src;
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002992 int err = 0;
Chris Masone02119d2008-09-05 16:13:11 -04002993 int ret;
2994 int i;
2995 int nritems;
2996 u64 first_offset = min_offset;
2997 u64 last_offset = (u64)-1;
Li Zefan33345d012011-04-20 10:31:50 +08002998 u64 ino = btrfs_ino(inode);
Chris Masone02119d2008-09-05 16:13:11 -04002999
3000 log = root->log_root;
Chris Masone02119d2008-09-05 16:13:11 -04003001
Li Zefan33345d012011-04-20 10:31:50 +08003002 min_key.objectid = ino;
Chris Masone02119d2008-09-05 16:13:11 -04003003 min_key.type = key_type;
3004 min_key.offset = min_offset;
3005
Filipe David Borba Manana6174d3c2013-10-01 16:13:42 +01003006 ret = btrfs_search_forward(root, &min_key, path, trans->transid);
Chris Masone02119d2008-09-05 16:13:11 -04003007
3008 /*
3009 * we didn't find anything from this transaction, see if there
3010 * is anything at all
3011 */
Li Zefan33345d012011-04-20 10:31:50 +08003012 if (ret != 0 || min_key.objectid != ino || min_key.type != key_type) {
3013 min_key.objectid = ino;
Chris Masone02119d2008-09-05 16:13:11 -04003014 min_key.type = key_type;
3015 min_key.offset = (u64)-1;
David Sterbab3b4aa72011-04-21 01:20:15 +02003016 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04003017 ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0);
3018 if (ret < 0) {
David Sterbab3b4aa72011-04-21 01:20:15 +02003019 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04003020 return ret;
3021 }
Li Zefan33345d012011-04-20 10:31:50 +08003022 ret = btrfs_previous_item(root, path, ino, key_type);
Chris Masone02119d2008-09-05 16:13:11 -04003023
3024 /* if ret == 0 there are items for this type,
3025 * create a range to tell us the last key of this type.
3026 * otherwise, there are no items in this directory after
3027 * *min_offset, and we create a range to indicate that.
3028 */
3029 if (ret == 0) {
3030 struct btrfs_key tmp;
3031 btrfs_item_key_to_cpu(path->nodes[0], &tmp,
3032 path->slots[0]);
Chris Masond3977122009-01-05 21:25:51 -05003033 if (key_type == tmp.type)
Chris Masone02119d2008-09-05 16:13:11 -04003034 first_offset = max(min_offset, tmp.offset) + 1;
Chris Masone02119d2008-09-05 16:13:11 -04003035 }
3036 goto done;
3037 }
3038
3039 /* go backward to find any previous key */
Li Zefan33345d012011-04-20 10:31:50 +08003040 ret = btrfs_previous_item(root, path, ino, key_type);
Chris Masone02119d2008-09-05 16:13:11 -04003041 if (ret == 0) {
3042 struct btrfs_key tmp;
3043 btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]);
3044 if (key_type == tmp.type) {
3045 first_offset = tmp.offset;
3046 ret = overwrite_item(trans, log, dst_path,
3047 path->nodes[0], path->slots[0],
3048 &tmp);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04003049 if (ret) {
3050 err = ret;
3051 goto done;
3052 }
Chris Masone02119d2008-09-05 16:13:11 -04003053 }
3054 }
David Sterbab3b4aa72011-04-21 01:20:15 +02003055 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04003056
3057 /* find the first key from this transaction again */
3058 ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0);
Dulshani Gunawardhanafae7f212013-10-31 10:30:08 +05303059 if (WARN_ON(ret != 0))
Chris Masone02119d2008-09-05 16:13:11 -04003060 goto done;
Chris Masone02119d2008-09-05 16:13:11 -04003061
3062 /*
3063 * we have a block from this transaction, log every item in it
3064 * from our directory
3065 */
Chris Masond3977122009-01-05 21:25:51 -05003066 while (1) {
Chris Masone02119d2008-09-05 16:13:11 -04003067 struct btrfs_key tmp;
3068 src = path->nodes[0];
3069 nritems = btrfs_header_nritems(src);
3070 for (i = path->slots[0]; i < nritems; i++) {
3071 btrfs_item_key_to_cpu(src, &min_key, i);
3072
Li Zefan33345d012011-04-20 10:31:50 +08003073 if (min_key.objectid != ino || min_key.type != key_type)
Chris Masone02119d2008-09-05 16:13:11 -04003074 goto done;
3075 ret = overwrite_item(trans, log, dst_path, src, i,
3076 &min_key);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04003077 if (ret) {
3078 err = ret;
3079 goto done;
3080 }
Chris Masone02119d2008-09-05 16:13:11 -04003081 }
3082 path->slots[0] = nritems;
3083
3084 /*
3085 * look ahead to the next item and see if it is also
3086 * from this directory and from this transaction
3087 */
3088 ret = btrfs_next_leaf(root, path);
3089 if (ret == 1) {
3090 last_offset = (u64)-1;
3091 goto done;
3092 }
3093 btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]);
Li Zefan33345d012011-04-20 10:31:50 +08003094 if (tmp.objectid != ino || tmp.type != key_type) {
Chris Masone02119d2008-09-05 16:13:11 -04003095 last_offset = (u64)-1;
3096 goto done;
3097 }
3098 if (btrfs_header_generation(path->nodes[0]) != trans->transid) {
3099 ret = overwrite_item(trans, log, dst_path,
3100 path->nodes[0], path->slots[0],
3101 &tmp);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04003102 if (ret)
3103 err = ret;
3104 else
3105 last_offset = tmp.offset;
Chris Masone02119d2008-09-05 16:13:11 -04003106 goto done;
3107 }
3108 }
3109done:
David Sterbab3b4aa72011-04-21 01:20:15 +02003110 btrfs_release_path(path);
3111 btrfs_release_path(dst_path);
Chris Masone02119d2008-09-05 16:13:11 -04003112
Yan, Zheng4a500fd2010-05-16 10:49:59 -04003113 if (err == 0) {
3114 *last_offset_ret = last_offset;
3115 /*
3116 * insert the log range keys to indicate where the log
3117 * is valid
3118 */
3119 ret = insert_dir_log_key(trans, log, path, key_type,
Li Zefan33345d012011-04-20 10:31:50 +08003120 ino, first_offset, last_offset);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04003121 if (ret)
3122 err = ret;
3123 }
3124 return err;
Chris Masone02119d2008-09-05 16:13:11 -04003125}
3126
3127/*
3128 * logging directories is very similar to logging inodes, We find all the items
3129 * from the current transaction and write them to the log.
3130 *
3131 * The recovery code scans the directory in the subvolume, and if it finds a
3132 * key in the range logged that is not present in the log tree, then it means
3133 * that dir entry was unlinked during the transaction.
3134 *
3135 * In order for that scan to work, we must include one key smaller than
3136 * the smallest logged by this transaction and one key larger than the largest
3137 * key logged by this transaction.
3138 */
3139static noinline int log_directory_changes(struct btrfs_trans_handle *trans,
3140 struct btrfs_root *root, struct inode *inode,
3141 struct btrfs_path *path,
3142 struct btrfs_path *dst_path)
3143{
3144 u64 min_key;
3145 u64 max_key;
3146 int ret;
3147 int key_type = BTRFS_DIR_ITEM_KEY;
3148
3149again:
3150 min_key = 0;
3151 max_key = 0;
Chris Masond3977122009-01-05 21:25:51 -05003152 while (1) {
Chris Masone02119d2008-09-05 16:13:11 -04003153 ret = log_dir_items(trans, root, inode, path,
3154 dst_path, key_type, min_key,
3155 &max_key);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04003156 if (ret)
3157 return ret;
Chris Masone02119d2008-09-05 16:13:11 -04003158 if (max_key == (u64)-1)
3159 break;
3160 min_key = max_key + 1;
3161 }
3162
3163 if (key_type == BTRFS_DIR_ITEM_KEY) {
3164 key_type = BTRFS_DIR_INDEX_KEY;
3165 goto again;
3166 }
3167 return 0;
3168}
3169
3170/*
3171 * a helper function to drop items from the log before we relog an
3172 * inode. max_key_type indicates the highest item type to remove.
3173 * This cannot be run for file data extents because it does not
3174 * free the extents they point to.
3175 */
3176static int drop_objectid_items(struct btrfs_trans_handle *trans,
3177 struct btrfs_root *log,
3178 struct btrfs_path *path,
3179 u64 objectid, int max_key_type)
3180{
3181 int ret;
3182 struct btrfs_key key;
3183 struct btrfs_key found_key;
Josef Bacik18ec90d2012-09-28 11:56:28 -04003184 int start_slot;
Chris Masone02119d2008-09-05 16:13:11 -04003185
3186 key.objectid = objectid;
3187 key.type = max_key_type;
3188 key.offset = (u64)-1;
3189
Chris Masond3977122009-01-05 21:25:51 -05003190 while (1) {
Chris Masone02119d2008-09-05 16:13:11 -04003191 ret = btrfs_search_slot(trans, log, &key, path, -1, 1);
Josef Bacik36508602013-04-25 16:23:32 -04003192 BUG_ON(ret == 0); /* Logic error */
Yan, Zheng4a500fd2010-05-16 10:49:59 -04003193 if (ret < 0)
Chris Masone02119d2008-09-05 16:13:11 -04003194 break;
3195
3196 if (path->slots[0] == 0)
3197 break;
3198
3199 path->slots[0]--;
3200 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
3201 path->slots[0]);
3202
3203 if (found_key.objectid != objectid)
3204 break;
3205
Josef Bacik18ec90d2012-09-28 11:56:28 -04003206 found_key.offset = 0;
3207 found_key.type = 0;
3208 ret = btrfs_bin_search(path->nodes[0], &found_key, 0,
3209 &start_slot);
3210
3211 ret = btrfs_del_items(trans, log, path, start_slot,
3212 path->slots[0] - start_slot + 1);
3213 /*
3214 * If start slot isn't 0 then we don't need to re-search, we've
3215 * found the last guy with the objectid in this tree.
3216 */
3217 if (ret || start_slot != 0)
Tsutomu Itoh65a246c2011-05-19 04:37:44 +00003218 break;
David Sterbab3b4aa72011-04-21 01:20:15 +02003219 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04003220 }
David Sterbab3b4aa72011-04-21 01:20:15 +02003221 btrfs_release_path(path);
Josef Bacik5bdbeb22012-05-29 16:59:49 -04003222 if (ret > 0)
3223 ret = 0;
Yan, Zheng4a500fd2010-05-16 10:49:59 -04003224 return ret;
Chris Masone02119d2008-09-05 16:13:11 -04003225}
3226
Josef Bacik94edf4a2012-09-25 14:56:25 -04003227static void fill_inode_item(struct btrfs_trans_handle *trans,
3228 struct extent_buffer *leaf,
3229 struct btrfs_inode_item *item,
3230 struct inode *inode, int log_inode_only)
3231{
Josef Bacik0b1c6cc2012-10-23 16:03:44 -04003232 struct btrfs_map_token token;
Josef Bacik94edf4a2012-09-25 14:56:25 -04003233
Josef Bacik0b1c6cc2012-10-23 16:03:44 -04003234 btrfs_init_map_token(&token);
Josef Bacik94edf4a2012-09-25 14:56:25 -04003235
3236 if (log_inode_only) {
3237 /* set the generation to zero so the recover code
3238 * can tell the difference between an logging
3239 * just to say 'this inode exists' and a logging
3240 * to say 'update this inode with these values'
3241 */
Josef Bacik0b1c6cc2012-10-23 16:03:44 -04003242 btrfs_set_token_inode_generation(leaf, item, 0, &token);
3243 btrfs_set_token_inode_size(leaf, item, 0, &token);
Josef Bacik94edf4a2012-09-25 14:56:25 -04003244 } else {
Josef Bacik0b1c6cc2012-10-23 16:03:44 -04003245 btrfs_set_token_inode_generation(leaf, item,
3246 BTRFS_I(inode)->generation,
3247 &token);
3248 btrfs_set_token_inode_size(leaf, item, inode->i_size, &token);
Josef Bacik94edf4a2012-09-25 14:56:25 -04003249 }
3250
Josef Bacik0b1c6cc2012-10-23 16:03:44 -04003251 btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token);
3252 btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token);
3253 btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token);
3254 btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token);
3255
3256 btrfs_set_token_timespec_sec(leaf, btrfs_inode_atime(item),
3257 inode->i_atime.tv_sec, &token);
3258 btrfs_set_token_timespec_nsec(leaf, btrfs_inode_atime(item),
3259 inode->i_atime.tv_nsec, &token);
3260
3261 btrfs_set_token_timespec_sec(leaf, btrfs_inode_mtime(item),
3262 inode->i_mtime.tv_sec, &token);
3263 btrfs_set_token_timespec_nsec(leaf, btrfs_inode_mtime(item),
3264 inode->i_mtime.tv_nsec, &token);
3265
3266 btrfs_set_token_timespec_sec(leaf, btrfs_inode_ctime(item),
3267 inode->i_ctime.tv_sec, &token);
3268 btrfs_set_token_timespec_nsec(leaf, btrfs_inode_ctime(item),
3269 inode->i_ctime.tv_nsec, &token);
3270
3271 btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode),
3272 &token);
3273
3274 btrfs_set_token_inode_sequence(leaf, item, inode->i_version, &token);
3275 btrfs_set_token_inode_transid(leaf, item, trans->transid, &token);
3276 btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token);
3277 btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token);
3278 btrfs_set_token_inode_block_group(leaf, item, 0, &token);
Josef Bacik94edf4a2012-09-25 14:56:25 -04003279}
3280
Josef Bacika95249b2012-10-11 16:17:34 -04003281static int log_inode_item(struct btrfs_trans_handle *trans,
3282 struct btrfs_root *log, struct btrfs_path *path,
3283 struct inode *inode)
3284{
3285 struct btrfs_inode_item *inode_item;
Josef Bacika95249b2012-10-11 16:17:34 -04003286 int ret;
3287
Filipe David Borba Mananaefd0c402013-10-07 21:20:44 +01003288 ret = btrfs_insert_empty_item(trans, log, path,
3289 &BTRFS_I(inode)->location,
Josef Bacika95249b2012-10-11 16:17:34 -04003290 sizeof(*inode_item));
3291 if (ret && ret != -EEXIST)
3292 return ret;
3293 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3294 struct btrfs_inode_item);
3295 fill_inode_item(trans, path->nodes[0], inode_item, inode, 0);
3296 btrfs_release_path(path);
3297 return 0;
3298}
3299
Chris Mason31ff1cd2008-09-11 16:17:57 -04003300static noinline int copy_items(struct btrfs_trans_handle *trans,
Liu Bod2794402012-08-29 01:07:56 -06003301 struct inode *inode,
Chris Mason31ff1cd2008-09-11 16:17:57 -04003302 struct btrfs_path *dst_path,
Josef Bacik16e75492013-10-22 12:18:51 -04003303 struct btrfs_path *src_path, u64 *last_extent,
Chris Mason31ff1cd2008-09-11 16:17:57 -04003304 int start_slot, int nr, int inode_only)
3305{
3306 unsigned long src_offset;
3307 unsigned long dst_offset;
Liu Bod2794402012-08-29 01:07:56 -06003308 struct btrfs_root *log = BTRFS_I(inode)->root->log_root;
Chris Mason31ff1cd2008-09-11 16:17:57 -04003309 struct btrfs_file_extent_item *extent;
3310 struct btrfs_inode_item *inode_item;
Josef Bacik16e75492013-10-22 12:18:51 -04003311 struct extent_buffer *src = src_path->nodes[0];
3312 struct btrfs_key first_key, last_key, key;
Chris Mason31ff1cd2008-09-11 16:17:57 -04003313 int ret;
3314 struct btrfs_key *ins_keys;
3315 u32 *ins_sizes;
3316 char *ins_data;
3317 int i;
Chris Masond20f7042008-12-08 16:58:54 -05003318 struct list_head ordered_sums;
Liu Bod2794402012-08-29 01:07:56 -06003319 int skip_csum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
Josef Bacik16e75492013-10-22 12:18:51 -04003320 bool has_extents = false;
Filipe Manana74121f7c2014-08-07 12:00:44 +01003321 bool need_find_last_extent = true;
Josef Bacik16e75492013-10-22 12:18:51 -04003322 bool done = false;
Chris Masond20f7042008-12-08 16:58:54 -05003323
3324 INIT_LIST_HEAD(&ordered_sums);
Chris Mason31ff1cd2008-09-11 16:17:57 -04003325
3326 ins_data = kmalloc(nr * sizeof(struct btrfs_key) +
3327 nr * sizeof(u32), GFP_NOFS);
liubo2a29edc2011-01-26 06:22:08 +00003328 if (!ins_data)
3329 return -ENOMEM;
3330
Josef Bacik16e75492013-10-22 12:18:51 -04003331 first_key.objectid = (u64)-1;
3332
Chris Mason31ff1cd2008-09-11 16:17:57 -04003333 ins_sizes = (u32 *)ins_data;
3334 ins_keys = (struct btrfs_key *)(ins_data + nr * sizeof(u32));
3335
3336 for (i = 0; i < nr; i++) {
3337 ins_sizes[i] = btrfs_item_size_nr(src, i + start_slot);
3338 btrfs_item_key_to_cpu(src, ins_keys + i, i + start_slot);
3339 }
3340 ret = btrfs_insert_empty_items(trans, log, dst_path,
3341 ins_keys, ins_sizes, nr);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04003342 if (ret) {
3343 kfree(ins_data);
3344 return ret;
3345 }
Chris Mason31ff1cd2008-09-11 16:17:57 -04003346
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003347 for (i = 0; i < nr; i++, dst_path->slots[0]++) {
Chris Mason31ff1cd2008-09-11 16:17:57 -04003348 dst_offset = btrfs_item_ptr_offset(dst_path->nodes[0],
3349 dst_path->slots[0]);
3350
3351 src_offset = btrfs_item_ptr_offset(src, start_slot + i);
3352
Josef Bacik16e75492013-10-22 12:18:51 -04003353 if ((i == (nr - 1)))
3354 last_key = ins_keys[i];
3355
Josef Bacik94edf4a2012-09-25 14:56:25 -04003356 if (ins_keys[i].type == BTRFS_INODE_ITEM_KEY) {
Chris Mason31ff1cd2008-09-11 16:17:57 -04003357 inode_item = btrfs_item_ptr(dst_path->nodes[0],
3358 dst_path->slots[0],
3359 struct btrfs_inode_item);
Josef Bacik94edf4a2012-09-25 14:56:25 -04003360 fill_inode_item(trans, dst_path->nodes[0], inode_item,
3361 inode, inode_only == LOG_INODE_EXISTS);
3362 } else {
3363 copy_extent_buffer(dst_path->nodes[0], src, dst_offset,
3364 src_offset, ins_sizes[i]);
Chris Mason31ff1cd2008-09-11 16:17:57 -04003365 }
Josef Bacik94edf4a2012-09-25 14:56:25 -04003366
Josef Bacik16e75492013-10-22 12:18:51 -04003367 /*
3368 * We set need_find_last_extent here in case we know we were
3369 * processing other items and then walk into the first extent in
3370 * the inode. If we don't hit an extent then nothing changes,
3371 * we'll do the last search the next time around.
3372 */
3373 if (ins_keys[i].type == BTRFS_EXTENT_DATA_KEY) {
3374 has_extents = true;
Filipe Manana74121f7c2014-08-07 12:00:44 +01003375 if (first_key.objectid == (u64)-1)
Josef Bacik16e75492013-10-22 12:18:51 -04003376 first_key = ins_keys[i];
3377 } else {
3378 need_find_last_extent = false;
3379 }
3380
Chris Mason31ff1cd2008-09-11 16:17:57 -04003381 /* take a reference on file data extents so that truncates
3382 * or deletes of this inode don't have to relog the inode
3383 * again
3384 */
David Sterba962a2982014-06-04 18:41:45 +02003385 if (ins_keys[i].type == BTRFS_EXTENT_DATA_KEY &&
Liu Bod2794402012-08-29 01:07:56 -06003386 !skip_csum) {
Chris Mason31ff1cd2008-09-11 16:17:57 -04003387 int found_type;
3388 extent = btrfs_item_ptr(src, start_slot + i,
3389 struct btrfs_file_extent_item);
3390
liubo8e531cd2011-05-06 10:36:09 +08003391 if (btrfs_file_extent_generation(src, extent) < trans->transid)
3392 continue;
3393
Chris Mason31ff1cd2008-09-11 16:17:57 -04003394 found_type = btrfs_file_extent_type(src, extent);
Josef Bacik6f1fed72012-09-26 11:07:06 -04003395 if (found_type == BTRFS_FILE_EXTENT_REG) {
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003396 u64 ds, dl, cs, cl;
3397 ds = btrfs_file_extent_disk_bytenr(src,
3398 extent);
3399 /* ds == 0 is a hole */
3400 if (ds == 0)
3401 continue;
3402
3403 dl = btrfs_file_extent_disk_num_bytes(src,
3404 extent);
3405 cs = btrfs_file_extent_offset(src, extent);
3406 cl = btrfs_file_extent_num_bytes(src,
Joe Perchesa419aef2009-08-18 11:18:35 -07003407 extent);
Chris Mason580afd72008-12-08 19:15:39 -05003408 if (btrfs_file_extent_compression(src,
3409 extent)) {
3410 cs = 0;
3411 cl = dl;
3412 }
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003413
3414 ret = btrfs_lookup_csums_range(
3415 log->fs_info->csum_root,
3416 ds + cs, ds + cs + cl - 1,
Arne Jansena2de7332011-03-08 14:14:00 +01003417 &ordered_sums, 0);
Josef Bacik36508602013-04-25 16:23:32 -04003418 if (ret) {
3419 btrfs_release_path(dst_path);
3420 kfree(ins_data);
3421 return ret;
3422 }
Chris Mason31ff1cd2008-09-11 16:17:57 -04003423 }
3424 }
Chris Mason31ff1cd2008-09-11 16:17:57 -04003425 }
3426
3427 btrfs_mark_buffer_dirty(dst_path->nodes[0]);
David Sterbab3b4aa72011-04-21 01:20:15 +02003428 btrfs_release_path(dst_path);
Chris Mason31ff1cd2008-09-11 16:17:57 -04003429 kfree(ins_data);
Chris Masond20f7042008-12-08 16:58:54 -05003430
3431 /*
3432 * we have to do this after the loop above to avoid changing the
3433 * log tree while trying to change the log tree.
3434 */
Yan, Zheng4a500fd2010-05-16 10:49:59 -04003435 ret = 0;
Chris Masond3977122009-01-05 21:25:51 -05003436 while (!list_empty(&ordered_sums)) {
Chris Masond20f7042008-12-08 16:58:54 -05003437 struct btrfs_ordered_sum *sums = list_entry(ordered_sums.next,
3438 struct btrfs_ordered_sum,
3439 list);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04003440 if (!ret)
3441 ret = btrfs_csum_file_blocks(trans, log, sums);
Chris Masond20f7042008-12-08 16:58:54 -05003442 list_del(&sums->list);
3443 kfree(sums);
3444 }
Josef Bacik16e75492013-10-22 12:18:51 -04003445
3446 if (!has_extents)
3447 return ret;
3448
Filipe Manana74121f7c2014-08-07 12:00:44 +01003449 if (need_find_last_extent && *last_extent == first_key.offset) {
3450 /*
3451 * We don't have any leafs between our current one and the one
3452 * we processed before that can have file extent items for our
3453 * inode (and have a generation number smaller than our current
3454 * transaction id).
3455 */
3456 need_find_last_extent = false;
3457 }
3458
Josef Bacik16e75492013-10-22 12:18:51 -04003459 /*
3460 * Because we use btrfs_search_forward we could skip leaves that were
3461 * not modified and then assume *last_extent is valid when it really
3462 * isn't. So back up to the previous leaf and read the end of the last
3463 * extent before we go and fill in holes.
3464 */
3465 if (need_find_last_extent) {
3466 u64 len;
3467
3468 ret = btrfs_prev_leaf(BTRFS_I(inode)->root, src_path);
3469 if (ret < 0)
3470 return ret;
3471 if (ret)
3472 goto fill_holes;
3473 if (src_path->slots[0])
3474 src_path->slots[0]--;
3475 src = src_path->nodes[0];
3476 btrfs_item_key_to_cpu(src, &key, src_path->slots[0]);
3477 if (key.objectid != btrfs_ino(inode) ||
3478 key.type != BTRFS_EXTENT_DATA_KEY)
3479 goto fill_holes;
3480 extent = btrfs_item_ptr(src, src_path->slots[0],
3481 struct btrfs_file_extent_item);
3482 if (btrfs_file_extent_type(src, extent) ==
3483 BTRFS_FILE_EXTENT_INLINE) {
Chris Mason514ac8a2014-01-03 21:07:00 -08003484 len = btrfs_file_extent_inline_len(src,
3485 src_path->slots[0],
3486 extent);
Josef Bacik16e75492013-10-22 12:18:51 -04003487 *last_extent = ALIGN(key.offset + len,
3488 log->sectorsize);
3489 } else {
3490 len = btrfs_file_extent_num_bytes(src, extent);
3491 *last_extent = key.offset + len;
3492 }
3493 }
3494fill_holes:
3495 /* So we did prev_leaf, now we need to move to the next leaf, but a few
3496 * things could have happened
3497 *
3498 * 1) A merge could have happened, so we could currently be on a leaf
3499 * that holds what we were copying in the first place.
3500 * 2) A split could have happened, and now not all of the items we want
3501 * are on the same leaf.
3502 *
3503 * So we need to adjust how we search for holes, we need to drop the
3504 * path and re-search for the first extent key we found, and then walk
3505 * forward until we hit the last one we copied.
3506 */
3507 if (need_find_last_extent) {
3508 /* btrfs_prev_leaf could return 1 without releasing the path */
3509 btrfs_release_path(src_path);
3510 ret = btrfs_search_slot(NULL, BTRFS_I(inode)->root, &first_key,
3511 src_path, 0, 0);
3512 if (ret < 0)
3513 return ret;
3514 ASSERT(ret == 0);
3515 src = src_path->nodes[0];
3516 i = src_path->slots[0];
3517 } else {
3518 i = start_slot;
3519 }
3520
3521 /*
3522 * Ok so here we need to go through and fill in any holes we may have
3523 * to make sure that holes are punched for those areas in case they had
3524 * extents previously.
3525 */
3526 while (!done) {
3527 u64 offset, len;
3528 u64 extent_end;
3529
3530 if (i >= btrfs_header_nritems(src_path->nodes[0])) {
3531 ret = btrfs_next_leaf(BTRFS_I(inode)->root, src_path);
3532 if (ret < 0)
3533 return ret;
3534 ASSERT(ret == 0);
3535 src = src_path->nodes[0];
3536 i = 0;
3537 }
3538
3539 btrfs_item_key_to_cpu(src, &key, i);
3540 if (!btrfs_comp_cpu_keys(&key, &last_key))
3541 done = true;
3542 if (key.objectid != btrfs_ino(inode) ||
3543 key.type != BTRFS_EXTENT_DATA_KEY) {
3544 i++;
3545 continue;
3546 }
3547 extent = btrfs_item_ptr(src, i, struct btrfs_file_extent_item);
3548 if (btrfs_file_extent_type(src, extent) ==
3549 BTRFS_FILE_EXTENT_INLINE) {
Chris Mason514ac8a2014-01-03 21:07:00 -08003550 len = btrfs_file_extent_inline_len(src, i, extent);
Josef Bacik16e75492013-10-22 12:18:51 -04003551 extent_end = ALIGN(key.offset + len, log->sectorsize);
3552 } else {
3553 len = btrfs_file_extent_num_bytes(src, extent);
3554 extent_end = key.offset + len;
3555 }
3556 i++;
3557
3558 if (*last_extent == key.offset) {
3559 *last_extent = extent_end;
3560 continue;
3561 }
3562 offset = *last_extent;
3563 len = key.offset - *last_extent;
3564 ret = btrfs_insert_file_extent(trans, log, btrfs_ino(inode),
3565 offset, 0, 0, len, 0, len, 0,
3566 0, 0);
3567 if (ret)
3568 break;
Filipe Manana74121f7c2014-08-07 12:00:44 +01003569 *last_extent = extent_end;
Josef Bacik16e75492013-10-22 12:18:51 -04003570 }
3571 /*
3572 * Need to let the callers know we dropped the path so they should
3573 * re-search.
3574 */
3575 if (!ret && need_find_last_extent)
3576 ret = 1;
Yan, Zheng4a500fd2010-05-16 10:49:59 -04003577 return ret;
Chris Mason31ff1cd2008-09-11 16:17:57 -04003578}
3579
Josef Bacik5dc562c2012-08-17 13:14:17 -04003580static int extent_cmp(void *priv, struct list_head *a, struct list_head *b)
3581{
3582 struct extent_map *em1, *em2;
3583
3584 em1 = list_entry(a, struct extent_map, list);
3585 em2 = list_entry(b, struct extent_map, list);
3586
3587 if (em1->start < em2->start)
3588 return -1;
3589 else if (em1->start > em2->start)
3590 return 1;
3591 return 0;
3592}
3593
Filipe Manana8407f552014-09-05 15:14:39 +01003594static int wait_ordered_extents(struct btrfs_trans_handle *trans,
3595 struct inode *inode,
3596 struct btrfs_root *root,
3597 const struct extent_map *em,
3598 const struct list_head *logged_list,
3599 bool *ordered_io_error)
Josef Bacik5dc562c2012-08-17 13:14:17 -04003600{
Josef Bacik2ab28f32012-10-12 15:27:49 -04003601 struct btrfs_ordered_extent *ordered;
Filipe Manana8407f552014-09-05 15:14:39 +01003602 struct btrfs_root *log = root->log_root;
Josef Bacik2ab28f32012-10-12 15:27:49 -04003603 u64 mod_start = em->mod_start;
3604 u64 mod_len = em->mod_len;
Filipe Manana8407f552014-09-05 15:14:39 +01003605 const bool skip_csum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
Josef Bacik2ab28f32012-10-12 15:27:49 -04003606 u64 csum_offset;
3607 u64 csum_len;
Filipe Manana8407f552014-09-05 15:14:39 +01003608 LIST_HEAD(ordered_sums);
3609 int ret = 0;
Josef Bacik09a2a8f92013-04-05 16:51:15 -04003610
Filipe Manana8407f552014-09-05 15:14:39 +01003611 *ordered_io_error = false;
Josef Bacik70c8a912012-10-11 16:54:30 -04003612
Filipe Manana8407f552014-09-05 15:14:39 +01003613 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
3614 em->block_start == EXTENT_MAP_HOLE)
Josef Bacik70c8a912012-10-11 16:54:30 -04003615 return 0;
3616
Josef Bacik2ab28f32012-10-12 15:27:49 -04003617 /*
Filipe Manana8407f552014-09-05 15:14:39 +01003618 * Wait far any ordered extent that covers our extent map. If it
3619 * finishes without an error, first check and see if our csums are on
3620 * our outstanding ordered extents.
Josef Bacik2ab28f32012-10-12 15:27:49 -04003621 */
Miao Xie827463c2014-01-14 20:31:51 +08003622 list_for_each_entry(ordered, logged_list, log_list) {
Josef Bacik2ab28f32012-10-12 15:27:49 -04003623 struct btrfs_ordered_sum *sum;
3624
3625 if (!mod_len)
3626 break;
3627
Josef Bacik2ab28f32012-10-12 15:27:49 -04003628 if (ordered->file_offset + ordered->len <= mod_start ||
3629 mod_start + mod_len <= ordered->file_offset)
3630 continue;
3631
Filipe Manana8407f552014-09-05 15:14:39 +01003632 if (!test_bit(BTRFS_ORDERED_IO_DONE, &ordered->flags) &&
3633 !test_bit(BTRFS_ORDERED_IOERR, &ordered->flags) &&
3634 !test_bit(BTRFS_ORDERED_DIRECT, &ordered->flags)) {
3635 const u64 start = ordered->file_offset;
3636 const u64 end = ordered->file_offset + ordered->len - 1;
3637
3638 WARN_ON(ordered->inode != inode);
3639 filemap_fdatawrite_range(inode->i_mapping, start, end);
3640 }
3641
3642 wait_event(ordered->wait,
3643 (test_bit(BTRFS_ORDERED_IO_DONE, &ordered->flags) ||
3644 test_bit(BTRFS_ORDERED_IOERR, &ordered->flags)));
3645
3646 if (test_bit(BTRFS_ORDERED_IOERR, &ordered->flags)) {
Filipe Mananab38ef712014-11-13 17:01:45 +00003647 /*
3648 * Clear the AS_EIO/AS_ENOSPC flags from the inode's
3649 * i_mapping flags, so that the next fsync won't get
3650 * an outdated io error too.
3651 */
3652 btrfs_inode_check_errors(inode);
Filipe Manana8407f552014-09-05 15:14:39 +01003653 *ordered_io_error = true;
3654 break;
3655 }
Josef Bacik2ab28f32012-10-12 15:27:49 -04003656 /*
3657 * We are going to copy all the csums on this ordered extent, so
3658 * go ahead and adjust mod_start and mod_len in case this
3659 * ordered extent has already been logged.
3660 */
3661 if (ordered->file_offset > mod_start) {
3662 if (ordered->file_offset + ordered->len >=
3663 mod_start + mod_len)
3664 mod_len = ordered->file_offset - mod_start;
3665 /*
3666 * If we have this case
3667 *
3668 * |--------- logged extent ---------|
3669 * |----- ordered extent ----|
3670 *
3671 * Just don't mess with mod_start and mod_len, we'll
3672 * just end up logging more csums than we need and it
3673 * will be ok.
3674 */
3675 } else {
3676 if (ordered->file_offset + ordered->len <
3677 mod_start + mod_len) {
3678 mod_len = (mod_start + mod_len) -
3679 (ordered->file_offset + ordered->len);
3680 mod_start = ordered->file_offset +
3681 ordered->len;
3682 } else {
3683 mod_len = 0;
3684 }
3685 }
3686
Filipe Manana8407f552014-09-05 15:14:39 +01003687 if (skip_csum)
3688 continue;
3689
Josef Bacik2ab28f32012-10-12 15:27:49 -04003690 /*
3691 * To keep us from looping for the above case of an ordered
3692 * extent that falls inside of the logged extent.
3693 */
3694 if (test_and_set_bit(BTRFS_ORDERED_LOGGED_CSUM,
3695 &ordered->flags))
3696 continue;
Josef Bacik2ab28f32012-10-12 15:27:49 -04003697
Miao Xie23c671a2014-01-14 20:31:52 +08003698 if (ordered->csum_bytes_left) {
3699 btrfs_start_ordered_extent(inode, ordered, 0);
3700 wait_event(ordered->wait,
3701 ordered->csum_bytes_left == 0);
3702 }
Josef Bacik2ab28f32012-10-12 15:27:49 -04003703
3704 list_for_each_entry(sum, &ordered->list, list) {
3705 ret = btrfs_csum_file_blocks(trans, log, sum);
Miao Xie827463c2014-01-14 20:31:51 +08003706 if (ret)
Filipe Manana8407f552014-09-05 15:14:39 +01003707 break;
Josef Bacik2ab28f32012-10-12 15:27:49 -04003708 }
Josef Bacik2ab28f32012-10-12 15:27:49 -04003709 }
Josef Bacik2ab28f32012-10-12 15:27:49 -04003710
Filipe Manana8407f552014-09-05 15:14:39 +01003711 if (*ordered_io_error || !mod_len || ret || skip_csum)
Josef Bacik2ab28f32012-10-12 15:27:49 -04003712 return ret;
3713
Filipe David Borba Manana488111a2013-10-28 16:30:29 +00003714 if (em->compress_type) {
3715 csum_offset = 0;
Filipe Manana8407f552014-09-05 15:14:39 +01003716 csum_len = max(em->block_len, em->orig_block_len);
Filipe David Borba Manana488111a2013-10-28 16:30:29 +00003717 } else {
3718 csum_offset = mod_start - em->start;
3719 csum_len = mod_len;
3720 }
Josef Bacik2ab28f32012-10-12 15:27:49 -04003721
Josef Bacik70c8a912012-10-11 16:54:30 -04003722 /* block start is already adjusted for the file extent offset. */
3723 ret = btrfs_lookup_csums_range(log->fs_info->csum_root,
3724 em->block_start + csum_offset,
3725 em->block_start + csum_offset +
3726 csum_len - 1, &ordered_sums, 0);
3727 if (ret)
3728 return ret;
3729
3730 while (!list_empty(&ordered_sums)) {
3731 struct btrfs_ordered_sum *sums = list_entry(ordered_sums.next,
3732 struct btrfs_ordered_sum,
3733 list);
3734 if (!ret)
3735 ret = btrfs_csum_file_blocks(trans, log, sums);
3736 list_del(&sums->list);
3737 kfree(sums);
3738 }
3739
3740 return ret;
Josef Bacik5dc562c2012-08-17 13:14:17 -04003741}
3742
Filipe Manana8407f552014-09-05 15:14:39 +01003743static int log_one_extent(struct btrfs_trans_handle *trans,
3744 struct inode *inode, struct btrfs_root *root,
3745 const struct extent_map *em,
3746 struct btrfs_path *path,
3747 const struct list_head *logged_list,
3748 struct btrfs_log_ctx *ctx)
3749{
3750 struct btrfs_root *log = root->log_root;
3751 struct btrfs_file_extent_item *fi;
3752 struct extent_buffer *leaf;
3753 struct btrfs_map_token token;
3754 struct btrfs_key key;
3755 u64 extent_offset = em->start - em->orig_start;
3756 u64 block_len;
3757 int ret;
3758 int extent_inserted = 0;
3759 bool ordered_io_err = false;
3760
3761 ret = wait_ordered_extents(trans, inode, root, em, logged_list,
3762 &ordered_io_err);
3763 if (ret)
3764 return ret;
3765
3766 if (ordered_io_err) {
3767 ctx->io_err = -EIO;
3768 return 0;
3769 }
3770
3771 btrfs_init_map_token(&token);
3772
3773 ret = __btrfs_drop_extents(trans, log, inode, path, em->start,
3774 em->start + em->len, NULL, 0, 1,
3775 sizeof(*fi), &extent_inserted);
3776 if (ret)
3777 return ret;
3778
3779 if (!extent_inserted) {
3780 key.objectid = btrfs_ino(inode);
3781 key.type = BTRFS_EXTENT_DATA_KEY;
3782 key.offset = em->start;
3783
3784 ret = btrfs_insert_empty_item(trans, log, path, &key,
3785 sizeof(*fi));
3786 if (ret)
3787 return ret;
3788 }
3789 leaf = path->nodes[0];
3790 fi = btrfs_item_ptr(leaf, path->slots[0],
3791 struct btrfs_file_extent_item);
3792
Josef Bacik50d9aa92014-11-21 14:52:38 -05003793 btrfs_set_token_file_extent_generation(leaf, fi, trans->transid,
Filipe Manana8407f552014-09-05 15:14:39 +01003794 &token);
3795 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
3796 btrfs_set_token_file_extent_type(leaf, fi,
3797 BTRFS_FILE_EXTENT_PREALLOC,
3798 &token);
3799 else
3800 btrfs_set_token_file_extent_type(leaf, fi,
3801 BTRFS_FILE_EXTENT_REG,
3802 &token);
3803
3804 block_len = max(em->block_len, em->orig_block_len);
3805 if (em->compress_type != BTRFS_COMPRESS_NONE) {
3806 btrfs_set_token_file_extent_disk_bytenr(leaf, fi,
3807 em->block_start,
3808 &token);
3809 btrfs_set_token_file_extent_disk_num_bytes(leaf, fi, block_len,
3810 &token);
3811 } else if (em->block_start < EXTENT_MAP_LAST_BYTE) {
3812 btrfs_set_token_file_extent_disk_bytenr(leaf, fi,
3813 em->block_start -
3814 extent_offset, &token);
3815 btrfs_set_token_file_extent_disk_num_bytes(leaf, fi, block_len,
3816 &token);
3817 } else {
3818 btrfs_set_token_file_extent_disk_bytenr(leaf, fi, 0, &token);
3819 btrfs_set_token_file_extent_disk_num_bytes(leaf, fi, 0,
3820 &token);
3821 }
3822
3823 btrfs_set_token_file_extent_offset(leaf, fi, extent_offset, &token);
3824 btrfs_set_token_file_extent_num_bytes(leaf, fi, em->len, &token);
3825 btrfs_set_token_file_extent_ram_bytes(leaf, fi, em->ram_bytes, &token);
3826 btrfs_set_token_file_extent_compression(leaf, fi, em->compress_type,
3827 &token);
3828 btrfs_set_token_file_extent_encryption(leaf, fi, 0, &token);
3829 btrfs_set_token_file_extent_other_encoding(leaf, fi, 0, &token);
3830 btrfs_mark_buffer_dirty(leaf);
3831
3832 btrfs_release_path(path);
3833
3834 return ret;
3835}
3836
Josef Bacik5dc562c2012-08-17 13:14:17 -04003837static int btrfs_log_changed_extents(struct btrfs_trans_handle *trans,
3838 struct btrfs_root *root,
3839 struct inode *inode,
Miao Xie827463c2014-01-14 20:31:51 +08003840 struct btrfs_path *path,
Filipe Manana8407f552014-09-05 15:14:39 +01003841 struct list_head *logged_list,
3842 struct btrfs_log_ctx *ctx)
Josef Bacik5dc562c2012-08-17 13:14:17 -04003843{
Josef Bacik5dc562c2012-08-17 13:14:17 -04003844 struct extent_map *em, *n;
3845 struct list_head extents;
3846 struct extent_map_tree *tree = &BTRFS_I(inode)->extent_tree;
3847 u64 test_gen;
3848 int ret = 0;
Josef Bacik2ab28f32012-10-12 15:27:49 -04003849 int num = 0;
Josef Bacik5dc562c2012-08-17 13:14:17 -04003850
3851 INIT_LIST_HEAD(&extents);
3852
Josef Bacik5dc562c2012-08-17 13:14:17 -04003853 write_lock(&tree->lock);
3854 test_gen = root->fs_info->last_trans_committed;
3855
3856 list_for_each_entry_safe(em, n, &tree->modified_extents, list) {
3857 list_del_init(&em->list);
Josef Bacik2ab28f32012-10-12 15:27:49 -04003858
3859 /*
3860 * Just an arbitrary number, this can be really CPU intensive
3861 * once we start getting a lot of extents, and really once we
3862 * have a bunch of extents we just want to commit since it will
3863 * be faster.
3864 */
3865 if (++num > 32768) {
3866 list_del_init(&tree->modified_extents);
3867 ret = -EFBIG;
3868 goto process;
3869 }
3870
Josef Bacik5dc562c2012-08-17 13:14:17 -04003871 if (em->generation <= test_gen)
3872 continue;
Josef Bacikff44c6e2012-09-14 12:59:20 -04003873 /* Need a ref to keep it from getting evicted from cache */
3874 atomic_inc(&em->refs);
3875 set_bit(EXTENT_FLAG_LOGGING, &em->flags);
Josef Bacik5dc562c2012-08-17 13:14:17 -04003876 list_add_tail(&em->list, &extents);
Josef Bacik2ab28f32012-10-12 15:27:49 -04003877 num++;
Josef Bacik5dc562c2012-08-17 13:14:17 -04003878 }
3879
3880 list_sort(NULL, &extents, extent_cmp);
3881
Josef Bacik2ab28f32012-10-12 15:27:49 -04003882process:
Josef Bacik5dc562c2012-08-17 13:14:17 -04003883 while (!list_empty(&extents)) {
3884 em = list_entry(extents.next, struct extent_map, list);
3885
3886 list_del_init(&em->list);
3887
3888 /*
3889 * If we had an error we just need to delete everybody from our
3890 * private list.
3891 */
Josef Bacikff44c6e2012-09-14 12:59:20 -04003892 if (ret) {
Josef Bacik201a9032013-01-24 12:02:07 -05003893 clear_em_logging(tree, em);
Josef Bacikff44c6e2012-09-14 12:59:20 -04003894 free_extent_map(em);
Josef Bacik5dc562c2012-08-17 13:14:17 -04003895 continue;
Josef Bacikff44c6e2012-09-14 12:59:20 -04003896 }
3897
3898 write_unlock(&tree->lock);
Josef Bacik5dc562c2012-08-17 13:14:17 -04003899
Filipe Manana8407f552014-09-05 15:14:39 +01003900 ret = log_one_extent(trans, inode, root, em, path, logged_list,
3901 ctx);
Josef Bacikff44c6e2012-09-14 12:59:20 -04003902 write_lock(&tree->lock);
Josef Bacik201a9032013-01-24 12:02:07 -05003903 clear_em_logging(tree, em);
3904 free_extent_map(em);
Josef Bacik5dc562c2012-08-17 13:14:17 -04003905 }
Josef Bacikff44c6e2012-09-14 12:59:20 -04003906 WARN_ON(!list_empty(&extents));
3907 write_unlock(&tree->lock);
Josef Bacik5dc562c2012-08-17 13:14:17 -04003908
Josef Bacik5dc562c2012-08-17 13:14:17 -04003909 btrfs_release_path(path);
Josef Bacik5dc562c2012-08-17 13:14:17 -04003910 return ret;
3911}
3912
Chris Masone02119d2008-09-05 16:13:11 -04003913/* log a single inode in the tree log.
3914 * At least one parent directory for this inode must exist in the tree
3915 * or be logged already.
3916 *
3917 * Any items from this inode changed by the current transaction are copied
3918 * to the log tree. An extra reference is taken on any extents in this
3919 * file, allowing us to avoid a whole pile of corner cases around logging
3920 * blocks that have been removed from the tree.
3921 *
3922 * See LOG_INODE_ALL and related defines for a description of what inode_only
3923 * does.
3924 *
3925 * This handles both files and directories.
3926 */
Chris Mason12fcfd22009-03-24 10:24:20 -04003927static int btrfs_log_inode(struct btrfs_trans_handle *trans,
Filipe Manana49dae1b2014-09-06 22:34:39 +01003928 struct btrfs_root *root, struct inode *inode,
3929 int inode_only,
3930 const loff_t start,
Filipe Manana8407f552014-09-05 15:14:39 +01003931 const loff_t end,
3932 struct btrfs_log_ctx *ctx)
Chris Masone02119d2008-09-05 16:13:11 -04003933{
3934 struct btrfs_path *path;
3935 struct btrfs_path *dst_path;
3936 struct btrfs_key min_key;
3937 struct btrfs_key max_key;
3938 struct btrfs_root *log = root->log_root;
Chris Mason31ff1cd2008-09-11 16:17:57 -04003939 struct extent_buffer *src = NULL;
Miao Xie827463c2014-01-14 20:31:51 +08003940 LIST_HEAD(logged_list);
Josef Bacik16e75492013-10-22 12:18:51 -04003941 u64 last_extent = 0;
Yan, Zheng4a500fd2010-05-16 10:49:59 -04003942 int err = 0;
Chris Masone02119d2008-09-05 16:13:11 -04003943 int ret;
Chris Mason3a5f1d42008-09-11 15:53:37 -04003944 int nritems;
Chris Mason31ff1cd2008-09-11 16:17:57 -04003945 int ins_start_slot = 0;
3946 int ins_nr;
Josef Bacik5dc562c2012-08-17 13:14:17 -04003947 bool fast_search = false;
Li Zefan33345d012011-04-20 10:31:50 +08003948 u64 ino = btrfs_ino(inode);
Filipe Manana49dae1b2014-09-06 22:34:39 +01003949 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
Chris Masone02119d2008-09-05 16:13:11 -04003950
Chris Masone02119d2008-09-05 16:13:11 -04003951 path = btrfs_alloc_path();
Tsutomu Itoh5df67082011-02-01 09:17:35 +00003952 if (!path)
3953 return -ENOMEM;
Chris Masone02119d2008-09-05 16:13:11 -04003954 dst_path = btrfs_alloc_path();
Tsutomu Itoh5df67082011-02-01 09:17:35 +00003955 if (!dst_path) {
3956 btrfs_free_path(path);
3957 return -ENOMEM;
3958 }
Chris Masone02119d2008-09-05 16:13:11 -04003959
Li Zefan33345d012011-04-20 10:31:50 +08003960 min_key.objectid = ino;
Chris Masone02119d2008-09-05 16:13:11 -04003961 min_key.type = BTRFS_INODE_ITEM_KEY;
3962 min_key.offset = 0;
3963
Li Zefan33345d012011-04-20 10:31:50 +08003964 max_key.objectid = ino;
Chris Mason12fcfd22009-03-24 10:24:20 -04003965
Chris Mason12fcfd22009-03-24 10:24:20 -04003966
Josef Bacik5dc562c2012-08-17 13:14:17 -04003967 /* today the code can only do partial logging of directories */
Miao Xie5269b672012-11-01 07:35:23 +00003968 if (S_ISDIR(inode->i_mode) ||
3969 (!test_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3970 &BTRFS_I(inode)->runtime_flags) &&
3971 inode_only == LOG_INODE_EXISTS))
Chris Masone02119d2008-09-05 16:13:11 -04003972 max_key.type = BTRFS_XATTR_ITEM_KEY;
3973 else
3974 max_key.type = (u8)-1;
3975 max_key.offset = (u64)-1;
3976
Josef Bacik94edf4a2012-09-25 14:56:25 -04003977 /* Only run delayed items if we are a dir or a new file */
3978 if (S_ISDIR(inode->i_mode) ||
3979 BTRFS_I(inode)->generation > root->fs_info->last_trans_committed) {
3980 ret = btrfs_commit_inode_delayed_items(trans, inode);
3981 if (ret) {
3982 btrfs_free_path(path);
3983 btrfs_free_path(dst_path);
3984 return ret;
3985 }
Miao Xie16cdcec2011-04-22 18:12:22 +08003986 }
3987
Chris Masone02119d2008-09-05 16:13:11 -04003988 mutex_lock(&BTRFS_I(inode)->log_mutex);
3989
Filipe Manana08702952014-11-13 17:00:35 +00003990 btrfs_get_logged_extents(inode, &logged_list, start, end);
Josef Bacik2ab28f32012-10-12 15:27:49 -04003991
Chris Masone02119d2008-09-05 16:13:11 -04003992 /*
3993 * a brute force approach to making sure we get the most uptodate
3994 * copies of everything.
3995 */
3996 if (S_ISDIR(inode->i_mode)) {
3997 int max_key_type = BTRFS_DIR_LOG_INDEX_KEY;
3998
3999 if (inode_only == LOG_INODE_EXISTS)
4000 max_key_type = BTRFS_XATTR_ITEM_KEY;
Li Zefan33345d012011-04-20 10:31:50 +08004001 ret = drop_objectid_items(trans, log, path, ino, max_key_type);
Chris Masone02119d2008-09-05 16:13:11 -04004002 } else {
Josef Bacik5dc562c2012-08-17 13:14:17 -04004003 if (test_and_clear_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4004 &BTRFS_I(inode)->runtime_flags)) {
Josef Bacike9976152012-10-11 15:53:56 -04004005 clear_bit(BTRFS_INODE_COPY_EVERYTHING,
4006 &BTRFS_I(inode)->runtime_flags);
Josef Bacik5dc562c2012-08-17 13:14:17 -04004007 ret = btrfs_truncate_inode_items(trans, log,
4008 inode, 0, 0);
Josef Bacika95249b2012-10-11 16:17:34 -04004009 } else if (test_and_clear_bit(BTRFS_INODE_COPY_EVERYTHING,
Josef Bacik6cfab852013-11-12 16:25:58 -05004010 &BTRFS_I(inode)->runtime_flags) ||
4011 inode_only == LOG_INODE_EXISTS) {
Josef Bacika95249b2012-10-11 16:17:34 -04004012 if (inode_only == LOG_INODE_ALL)
4013 fast_search = true;
4014 max_key.type = BTRFS_XATTR_ITEM_KEY;
4015 ret = drop_objectid_items(trans, log, path, ino,
4016 max_key.type);
Josef Bacik5dc562c2012-08-17 13:14:17 -04004017 } else {
Liu Bo183f37f2012-11-01 06:38:47 +00004018 if (inode_only == LOG_INODE_ALL)
4019 fast_search = true;
Josef Bacika95249b2012-10-11 16:17:34 -04004020 ret = log_inode_item(trans, log, dst_path, inode);
4021 if (ret) {
4022 err = ret;
4023 goto out_unlock;
4024 }
4025 goto log_extents;
Josef Bacik5dc562c2012-08-17 13:14:17 -04004026 }
Josef Bacika95249b2012-10-11 16:17:34 -04004027
Chris Masone02119d2008-09-05 16:13:11 -04004028 }
Yan, Zheng4a500fd2010-05-16 10:49:59 -04004029 if (ret) {
4030 err = ret;
4031 goto out_unlock;
4032 }
Chris Masone02119d2008-09-05 16:13:11 -04004033
Chris Masond3977122009-01-05 21:25:51 -05004034 while (1) {
Chris Mason31ff1cd2008-09-11 16:17:57 -04004035 ins_nr = 0;
Filipe David Borba Manana6174d3c2013-10-01 16:13:42 +01004036 ret = btrfs_search_forward(root, &min_key,
Eric Sandeende78b512013-01-31 18:21:12 +00004037 path, trans->transid);
Chris Masone02119d2008-09-05 16:13:11 -04004038 if (ret != 0)
4039 break;
Chris Mason3a5f1d42008-09-11 15:53:37 -04004040again:
Chris Mason31ff1cd2008-09-11 16:17:57 -04004041 /* note, ins_nr might be > 0 here, cleanup outside the loop */
Li Zefan33345d012011-04-20 10:31:50 +08004042 if (min_key.objectid != ino)
Chris Masone02119d2008-09-05 16:13:11 -04004043 break;
4044 if (min_key.type > max_key.type)
4045 break;
Chris Mason31ff1cd2008-09-11 16:17:57 -04004046
Chris Masone02119d2008-09-05 16:13:11 -04004047 src = path->nodes[0];
Chris Mason31ff1cd2008-09-11 16:17:57 -04004048 if (ins_nr && ins_start_slot + ins_nr == path->slots[0]) {
4049 ins_nr++;
4050 goto next_slot;
4051 } else if (!ins_nr) {
4052 ins_start_slot = path->slots[0];
4053 ins_nr = 1;
4054 goto next_slot;
Chris Masone02119d2008-09-05 16:13:11 -04004055 }
4056
Josef Bacik16e75492013-10-22 12:18:51 -04004057 ret = copy_items(trans, inode, dst_path, path, &last_extent,
4058 ins_start_slot, ins_nr, inode_only);
4059 if (ret < 0) {
Yan, Zheng4a500fd2010-05-16 10:49:59 -04004060 err = ret;
4061 goto out_unlock;
Rasmus Villemoesa71db862014-06-20 21:51:43 +02004062 }
4063 if (ret) {
Josef Bacik16e75492013-10-22 12:18:51 -04004064 ins_nr = 0;
4065 btrfs_release_path(path);
4066 continue;
Yan, Zheng4a500fd2010-05-16 10:49:59 -04004067 }
Chris Mason31ff1cd2008-09-11 16:17:57 -04004068 ins_nr = 1;
4069 ins_start_slot = path->slots[0];
4070next_slot:
Chris Masone02119d2008-09-05 16:13:11 -04004071
Chris Mason3a5f1d42008-09-11 15:53:37 -04004072 nritems = btrfs_header_nritems(path->nodes[0]);
4073 path->slots[0]++;
4074 if (path->slots[0] < nritems) {
4075 btrfs_item_key_to_cpu(path->nodes[0], &min_key,
4076 path->slots[0]);
4077 goto again;
4078 }
Chris Mason31ff1cd2008-09-11 16:17:57 -04004079 if (ins_nr) {
Josef Bacik16e75492013-10-22 12:18:51 -04004080 ret = copy_items(trans, inode, dst_path, path,
4081 &last_extent, ins_start_slot,
Chris Mason31ff1cd2008-09-11 16:17:57 -04004082 ins_nr, inode_only);
Josef Bacik16e75492013-10-22 12:18:51 -04004083 if (ret < 0) {
Yan, Zheng4a500fd2010-05-16 10:49:59 -04004084 err = ret;
4085 goto out_unlock;
4086 }
Josef Bacik16e75492013-10-22 12:18:51 -04004087 ret = 0;
Chris Mason31ff1cd2008-09-11 16:17:57 -04004088 ins_nr = 0;
4089 }
David Sterbab3b4aa72011-04-21 01:20:15 +02004090 btrfs_release_path(path);
Chris Mason3a5f1d42008-09-11 15:53:37 -04004091
Filipe David Borba Manana3d41d702013-10-01 17:06:53 +01004092 if (min_key.offset < (u64)-1) {
Chris Masone02119d2008-09-05 16:13:11 -04004093 min_key.offset++;
Filipe David Borba Manana3d41d702013-10-01 17:06:53 +01004094 } else if (min_key.type < max_key.type) {
Chris Masone02119d2008-09-05 16:13:11 -04004095 min_key.type++;
Filipe David Borba Manana3d41d702013-10-01 17:06:53 +01004096 min_key.offset = 0;
4097 } else {
Chris Masone02119d2008-09-05 16:13:11 -04004098 break;
Filipe David Borba Manana3d41d702013-10-01 17:06:53 +01004099 }
Chris Masone02119d2008-09-05 16:13:11 -04004100 }
Chris Mason31ff1cd2008-09-11 16:17:57 -04004101 if (ins_nr) {
Josef Bacik16e75492013-10-22 12:18:51 -04004102 ret = copy_items(trans, inode, dst_path, path, &last_extent,
4103 ins_start_slot, ins_nr, inode_only);
4104 if (ret < 0) {
Yan, Zheng4a500fd2010-05-16 10:49:59 -04004105 err = ret;
4106 goto out_unlock;
4107 }
Josef Bacik16e75492013-10-22 12:18:51 -04004108 ret = 0;
Chris Mason31ff1cd2008-09-11 16:17:57 -04004109 ins_nr = 0;
4110 }
Josef Bacik5dc562c2012-08-17 13:14:17 -04004111
Josef Bacika95249b2012-10-11 16:17:34 -04004112log_extents:
Josef Bacikf3b15cc2013-07-22 12:54:30 -04004113 btrfs_release_path(path);
4114 btrfs_release_path(dst_path);
Josef Bacik5dc562c2012-08-17 13:14:17 -04004115 if (fast_search) {
Filipe Mananab38ef712014-11-13 17:01:45 +00004116 /*
4117 * Some ordered extents started by fsync might have completed
4118 * before we collected the ordered extents in logged_list, which
4119 * means they're gone, not in our logged_list nor in the inode's
4120 * ordered tree. We want the application/user space to know an
4121 * error happened while attempting to persist file data so that
4122 * it can take proper action. If such error happened, we leave
4123 * without writing to the log tree and the fsync must report the
4124 * file data write error and not commit the current transaction.
4125 */
4126 err = btrfs_inode_check_errors(inode);
4127 if (err) {
4128 ctx->io_err = err;
4129 goto out_unlock;
4130 }
Miao Xie827463c2014-01-14 20:31:51 +08004131 ret = btrfs_log_changed_extents(trans, root, inode, dst_path,
Filipe Manana8407f552014-09-05 15:14:39 +01004132 &logged_list, ctx);
Josef Bacik5dc562c2012-08-17 13:14:17 -04004133 if (ret) {
4134 err = ret;
4135 goto out_unlock;
4136 }
Josef Bacikd006a042013-11-12 20:54:09 -05004137 } else if (inode_only == LOG_INODE_ALL) {
Liu Bo06d3d222012-08-27 10:52:19 -06004138 struct extent_map *em, *n;
4139
Filipe Manana49dae1b2014-09-06 22:34:39 +01004140 write_lock(&em_tree->lock);
4141 /*
4142 * We can't just remove every em if we're called for a ranged
4143 * fsync - that is, one that doesn't cover the whole possible
4144 * file range (0 to LLONG_MAX). This is because we can have
4145 * em's that fall outside the range we're logging and therefore
4146 * their ordered operations haven't completed yet
4147 * (btrfs_finish_ordered_io() not invoked yet). This means we
4148 * didn't get their respective file extent item in the fs/subvol
4149 * tree yet, and need to let the next fast fsync (one which
4150 * consults the list of modified extent maps) find the em so
4151 * that it logs a matching file extent item and waits for the
4152 * respective ordered operation to complete (if it's still
4153 * running).
4154 *
4155 * Removing every em outside the range we're logging would make
4156 * the next fast fsync not log their matching file extent items,
4157 * therefore making us lose data after a log replay.
4158 */
4159 list_for_each_entry_safe(em, n, &em_tree->modified_extents,
4160 list) {
4161 const u64 mod_end = em->mod_start + em->mod_len - 1;
4162
4163 if (em->mod_start >= start && mod_end <= end)
4164 list_del_init(&em->list);
4165 }
4166 write_unlock(&em_tree->lock);
Josef Bacik5dc562c2012-08-17 13:14:17 -04004167 }
4168
Chris Mason9623f9a2008-09-11 17:42:42 -04004169 if (inode_only == LOG_INODE_ALL && S_ISDIR(inode->i_mode)) {
Chris Masone02119d2008-09-05 16:13:11 -04004170 ret = log_directory_changes(trans, root, inode, path, dst_path);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04004171 if (ret) {
4172 err = ret;
4173 goto out_unlock;
4174 }
Chris Masone02119d2008-09-05 16:13:11 -04004175 }
Filipe Manana49dae1b2014-09-06 22:34:39 +01004176
Filipe Manana125c4cf92014-09-11 21:22:14 +01004177 BTRFS_I(inode)->logged_trans = trans->transid;
4178 BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->last_sub_trans;
Yan, Zheng4a500fd2010-05-16 10:49:59 -04004179out_unlock:
Miao Xie827463c2014-01-14 20:31:51 +08004180 if (unlikely(err))
4181 btrfs_put_logged_extents(&logged_list);
4182 else
4183 btrfs_submit_logged_extents(&logged_list, log);
Chris Masone02119d2008-09-05 16:13:11 -04004184 mutex_unlock(&BTRFS_I(inode)->log_mutex);
4185
4186 btrfs_free_path(path);
4187 btrfs_free_path(dst_path);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04004188 return err;
Chris Masone02119d2008-09-05 16:13:11 -04004189}
4190
Chris Mason12fcfd22009-03-24 10:24:20 -04004191/*
4192 * follow the dentry parent pointers up the chain and see if any
4193 * of the directories in it require a full commit before they can
4194 * be logged. Returns zero if nothing special needs to be done or 1 if
4195 * a full commit is required.
4196 */
4197static noinline int check_parent_dirs_for_sync(struct btrfs_trans_handle *trans,
4198 struct inode *inode,
4199 struct dentry *parent,
4200 struct super_block *sb,
4201 u64 last_committed)
Chris Masone02119d2008-09-05 16:13:11 -04004202{
Chris Mason12fcfd22009-03-24 10:24:20 -04004203 int ret = 0;
4204 struct btrfs_root *root;
Josef Bacik6a912212010-11-20 09:48:00 +00004205 struct dentry *old_parent = NULL;
Josef Bacikde2b5302013-09-11 09:36:30 -04004206 struct inode *orig_inode = inode;
Chris Masone02119d2008-09-05 16:13:11 -04004207
Chris Masonaf4176b2009-03-24 10:24:31 -04004208 /*
4209 * for regular files, if its inode is already on disk, we don't
4210 * have to worry about the parents at all. This is because
4211 * we can use the last_unlink_trans field to record renames
4212 * and other fun in this file.
4213 */
4214 if (S_ISREG(inode->i_mode) &&
4215 BTRFS_I(inode)->generation <= last_committed &&
4216 BTRFS_I(inode)->last_unlink_trans <= last_committed)
4217 goto out;
4218
Chris Mason12fcfd22009-03-24 10:24:20 -04004219 if (!S_ISDIR(inode->i_mode)) {
4220 if (!parent || !parent->d_inode || sb != parent->d_inode->i_sb)
4221 goto out;
4222 inode = parent->d_inode;
4223 }
4224
4225 while (1) {
Josef Bacikde2b5302013-09-11 09:36:30 -04004226 /*
4227 * If we are logging a directory then we start with our inode,
4228 * not our parents inode, so we need to skipp setting the
4229 * logged_trans so that further down in the log code we don't
4230 * think this inode has already been logged.
4231 */
4232 if (inode != orig_inode)
4233 BTRFS_I(inode)->logged_trans = trans->transid;
Chris Mason12fcfd22009-03-24 10:24:20 -04004234 smp_mb();
4235
4236 if (BTRFS_I(inode)->last_unlink_trans > last_committed) {
4237 root = BTRFS_I(inode)->root;
4238
4239 /*
4240 * make sure any commits to the log are forced
4241 * to be full commits
4242 */
Miao Xie995946d2014-04-02 19:51:06 +08004243 btrfs_set_log_full_commit(root->fs_info, trans);
Chris Mason12fcfd22009-03-24 10:24:20 -04004244 ret = 1;
4245 break;
4246 }
4247
4248 if (!parent || !parent->d_inode || sb != parent->d_inode->i_sb)
4249 break;
4250
Yan, Zheng76dda932009-09-21 16:00:26 -04004251 if (IS_ROOT(parent))
Chris Mason12fcfd22009-03-24 10:24:20 -04004252 break;
4253
Josef Bacik6a912212010-11-20 09:48:00 +00004254 parent = dget_parent(parent);
4255 dput(old_parent);
4256 old_parent = parent;
Chris Mason12fcfd22009-03-24 10:24:20 -04004257 inode = parent->d_inode;
4258
4259 }
Josef Bacik6a912212010-11-20 09:48:00 +00004260 dput(old_parent);
Chris Mason12fcfd22009-03-24 10:24:20 -04004261out:
Chris Masone02119d2008-09-05 16:13:11 -04004262 return ret;
4263}
4264
4265/*
4266 * helper function around btrfs_log_inode to make sure newly created
4267 * parent directories also end up in the log. A minimal inode and backref
4268 * only logging is done of any parent directories that are older than
4269 * the last committed transaction
4270 */
Eric Sandeen48a3b632013-04-25 20:41:01 +00004271static int btrfs_log_inode_parent(struct btrfs_trans_handle *trans,
4272 struct btrfs_root *root, struct inode *inode,
Filipe Manana49dae1b2014-09-06 22:34:39 +01004273 struct dentry *parent,
4274 const loff_t start,
4275 const loff_t end,
4276 int exists_only,
Miao Xie8b050d32014-02-20 18:08:58 +08004277 struct btrfs_log_ctx *ctx)
Chris Masone02119d2008-09-05 16:13:11 -04004278{
Chris Mason12fcfd22009-03-24 10:24:20 -04004279 int inode_only = exists_only ? LOG_INODE_EXISTS : LOG_INODE_ALL;
Chris Masone02119d2008-09-05 16:13:11 -04004280 struct super_block *sb;
Josef Bacik6a912212010-11-20 09:48:00 +00004281 struct dentry *old_parent = NULL;
Chris Mason12fcfd22009-03-24 10:24:20 -04004282 int ret = 0;
4283 u64 last_committed = root->fs_info->last_trans_committed;
4284
4285 sb = inode->i_sb;
4286
Sage Weil3a5e1402009-04-02 16:49:40 -04004287 if (btrfs_test_opt(root, NOTREELOG)) {
4288 ret = 1;
4289 goto end_no_trans;
4290 }
4291
Miao Xie995946d2014-04-02 19:51:06 +08004292 /*
4293 * The prev transaction commit doesn't complete, we need do
4294 * full commit by ourselves.
4295 */
Chris Mason12fcfd22009-03-24 10:24:20 -04004296 if (root->fs_info->last_trans_log_full_commit >
4297 root->fs_info->last_trans_committed) {
4298 ret = 1;
4299 goto end_no_trans;
4300 }
4301
Yan, Zheng76dda932009-09-21 16:00:26 -04004302 if (root != BTRFS_I(inode)->root ||
4303 btrfs_root_refs(&root->root_item) == 0) {
4304 ret = 1;
4305 goto end_no_trans;
4306 }
4307
Chris Mason12fcfd22009-03-24 10:24:20 -04004308 ret = check_parent_dirs_for_sync(trans, inode, parent,
4309 sb, last_committed);
4310 if (ret)
4311 goto end_no_trans;
Chris Masone02119d2008-09-05 16:13:11 -04004312
Josef Bacik22ee6982012-05-29 16:57:49 -04004313 if (btrfs_inode_in_log(inode, trans->transid)) {
Chris Mason257c62e2009-10-13 13:21:08 -04004314 ret = BTRFS_NO_LOG_SYNC;
4315 goto end_no_trans;
4316 }
4317
Miao Xie8b050d32014-02-20 18:08:58 +08004318 ret = start_log_trans(trans, root, ctx);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04004319 if (ret)
Miao Xiee87ac132014-02-20 18:08:53 +08004320 goto end_no_trans;
Chris Mason12fcfd22009-03-24 10:24:20 -04004321
Filipe Manana8407f552014-09-05 15:14:39 +01004322 ret = btrfs_log_inode(trans, root, inode, inode_only, start, end, ctx);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04004323 if (ret)
4324 goto end_trans;
Chris Mason12fcfd22009-03-24 10:24:20 -04004325
Chris Masonaf4176b2009-03-24 10:24:31 -04004326 /*
4327 * for regular files, if its inode is already on disk, we don't
4328 * have to worry about the parents at all. This is because
4329 * we can use the last_unlink_trans field to record renames
4330 * and other fun in this file.
4331 */
4332 if (S_ISREG(inode->i_mode) &&
4333 BTRFS_I(inode)->generation <= last_committed &&
Yan, Zheng4a500fd2010-05-16 10:49:59 -04004334 BTRFS_I(inode)->last_unlink_trans <= last_committed) {
4335 ret = 0;
4336 goto end_trans;
4337 }
Chris Masonaf4176b2009-03-24 10:24:31 -04004338
4339 inode_only = LOG_INODE_EXISTS;
Chris Masond3977122009-01-05 21:25:51 -05004340 while (1) {
Chris Mason12fcfd22009-03-24 10:24:20 -04004341 if (!parent || !parent->d_inode || sb != parent->d_inode->i_sb)
Chris Masone02119d2008-09-05 16:13:11 -04004342 break;
4343
Chris Mason12fcfd22009-03-24 10:24:20 -04004344 inode = parent->d_inode;
Yan, Zheng76dda932009-09-21 16:00:26 -04004345 if (root != BTRFS_I(inode)->root)
4346 break;
4347
Chris Mason12fcfd22009-03-24 10:24:20 -04004348 if (BTRFS_I(inode)->generation >
4349 root->fs_info->last_trans_committed) {
Filipe Manana49dae1b2014-09-06 22:34:39 +01004350 ret = btrfs_log_inode(trans, root, inode, inode_only,
Filipe Manana8407f552014-09-05 15:14:39 +01004351 0, LLONG_MAX, ctx);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04004352 if (ret)
4353 goto end_trans;
Chris Mason12fcfd22009-03-24 10:24:20 -04004354 }
Yan, Zheng76dda932009-09-21 16:00:26 -04004355 if (IS_ROOT(parent))
Chris Masone02119d2008-09-05 16:13:11 -04004356 break;
Chris Mason12fcfd22009-03-24 10:24:20 -04004357
Josef Bacik6a912212010-11-20 09:48:00 +00004358 parent = dget_parent(parent);
4359 dput(old_parent);
4360 old_parent = parent;
Chris Masone02119d2008-09-05 16:13:11 -04004361 }
Chris Mason12fcfd22009-03-24 10:24:20 -04004362 ret = 0;
Yan, Zheng4a500fd2010-05-16 10:49:59 -04004363end_trans:
Josef Bacik6a912212010-11-20 09:48:00 +00004364 dput(old_parent);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04004365 if (ret < 0) {
Miao Xie995946d2014-04-02 19:51:06 +08004366 btrfs_set_log_full_commit(root->fs_info, trans);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04004367 ret = 1;
4368 }
Miao Xie8b050d32014-02-20 18:08:58 +08004369
4370 if (ret)
4371 btrfs_remove_log_ctx(root, ctx);
Chris Mason12fcfd22009-03-24 10:24:20 -04004372 btrfs_end_log_trans(root);
4373end_no_trans:
4374 return ret;
Chris Masone02119d2008-09-05 16:13:11 -04004375}
4376
4377/*
4378 * it is not safe to log dentry if the chunk root has added new
4379 * chunks. This returns 0 if the dentry was logged, and 1 otherwise.
4380 * If this returns 1, you must commit the transaction to safely get your
4381 * data on disk.
4382 */
4383int btrfs_log_dentry_safe(struct btrfs_trans_handle *trans,
Miao Xie8b050d32014-02-20 18:08:58 +08004384 struct btrfs_root *root, struct dentry *dentry,
Filipe Manana49dae1b2014-09-06 22:34:39 +01004385 const loff_t start,
4386 const loff_t end,
Miao Xie8b050d32014-02-20 18:08:58 +08004387 struct btrfs_log_ctx *ctx)
Chris Masone02119d2008-09-05 16:13:11 -04004388{
Josef Bacik6a912212010-11-20 09:48:00 +00004389 struct dentry *parent = dget_parent(dentry);
4390 int ret;
4391
Miao Xie8b050d32014-02-20 18:08:58 +08004392 ret = btrfs_log_inode_parent(trans, root, dentry->d_inode, parent,
Filipe Manana49dae1b2014-09-06 22:34:39 +01004393 start, end, 0, ctx);
Josef Bacik6a912212010-11-20 09:48:00 +00004394 dput(parent);
4395
4396 return ret;
Chris Masone02119d2008-09-05 16:13:11 -04004397}
4398
4399/*
4400 * should be called during mount to recover any replay any log trees
4401 * from the FS
4402 */
4403int btrfs_recover_log_trees(struct btrfs_root *log_root_tree)
4404{
4405 int ret;
4406 struct btrfs_path *path;
4407 struct btrfs_trans_handle *trans;
4408 struct btrfs_key key;
4409 struct btrfs_key found_key;
4410 struct btrfs_key tmp_key;
4411 struct btrfs_root *log;
4412 struct btrfs_fs_info *fs_info = log_root_tree->fs_info;
4413 struct walk_control wc = {
4414 .process_func = process_one_buffer,
4415 .stage = 0,
4416 };
4417
Chris Masone02119d2008-09-05 16:13:11 -04004418 path = btrfs_alloc_path();
Tsutomu Itohdb5b4932011-03-23 08:14:16 +00004419 if (!path)
4420 return -ENOMEM;
4421
4422 fs_info->log_root_recovering = 1;
Chris Masone02119d2008-09-05 16:13:11 -04004423
Yan, Zheng4a500fd2010-05-16 10:49:59 -04004424 trans = btrfs_start_transaction(fs_info->tree_root, 0);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01004425 if (IS_ERR(trans)) {
4426 ret = PTR_ERR(trans);
4427 goto error;
4428 }
Chris Masone02119d2008-09-05 16:13:11 -04004429
4430 wc.trans = trans;
4431 wc.pin = 1;
4432
Tsutomu Itohdb5b4932011-03-23 08:14:16 +00004433 ret = walk_log_tree(trans, log_root_tree, &wc);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01004434 if (ret) {
4435 btrfs_error(fs_info, ret, "Failed to pin buffers while "
4436 "recovering log root tree.");
4437 goto error;
4438 }
Chris Masone02119d2008-09-05 16:13:11 -04004439
4440again:
4441 key.objectid = BTRFS_TREE_LOG_OBJECTID;
4442 key.offset = (u64)-1;
David Sterba962a2982014-06-04 18:41:45 +02004443 key.type = BTRFS_ROOT_ITEM_KEY;
Chris Masone02119d2008-09-05 16:13:11 -04004444
Chris Masond3977122009-01-05 21:25:51 -05004445 while (1) {
Chris Masone02119d2008-09-05 16:13:11 -04004446 ret = btrfs_search_slot(NULL, log_root_tree, &key, path, 0, 0);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01004447
4448 if (ret < 0) {
4449 btrfs_error(fs_info, ret,
4450 "Couldn't find tree log root.");
4451 goto error;
4452 }
Chris Masone02119d2008-09-05 16:13:11 -04004453 if (ret > 0) {
4454 if (path->slots[0] == 0)
4455 break;
4456 path->slots[0]--;
4457 }
4458 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
4459 path->slots[0]);
David Sterbab3b4aa72011-04-21 01:20:15 +02004460 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04004461 if (found_key.objectid != BTRFS_TREE_LOG_OBJECTID)
4462 break;
4463
Miao Xiecb517ea2013-05-15 07:48:19 +00004464 log = btrfs_read_fs_root(log_root_tree, &found_key);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01004465 if (IS_ERR(log)) {
4466 ret = PTR_ERR(log);
4467 btrfs_error(fs_info, ret,
4468 "Couldn't read tree log root.");
4469 goto error;
4470 }
Chris Masone02119d2008-09-05 16:13:11 -04004471
4472 tmp_key.objectid = found_key.offset;
4473 tmp_key.type = BTRFS_ROOT_ITEM_KEY;
4474 tmp_key.offset = (u64)-1;
4475
4476 wc.replay_dest = btrfs_read_fs_root_no_name(fs_info, &tmp_key);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01004477 if (IS_ERR(wc.replay_dest)) {
4478 ret = PTR_ERR(wc.replay_dest);
Josef Bacikb50c6e22013-04-25 15:55:30 -04004479 free_extent_buffer(log->node);
4480 free_extent_buffer(log->commit_root);
4481 kfree(log);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01004482 btrfs_error(fs_info, ret, "Couldn't read target root "
4483 "for tree log recovery.");
4484 goto error;
4485 }
Chris Masone02119d2008-09-05 16:13:11 -04004486
Yan Zheng07d400a2009-01-06 11:42:00 -05004487 wc.replay_dest->log_root = log;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004488 btrfs_record_root_in_trans(trans, wc.replay_dest);
Chris Masone02119d2008-09-05 16:13:11 -04004489 ret = walk_log_tree(trans, log, &wc);
Chris Masone02119d2008-09-05 16:13:11 -04004490
Josef Bacikb50c6e22013-04-25 15:55:30 -04004491 if (!ret && wc.stage == LOG_WALK_REPLAY_ALL) {
Chris Masone02119d2008-09-05 16:13:11 -04004492 ret = fixup_inode_link_counts(trans, wc.replay_dest,
4493 path);
Chris Masone02119d2008-09-05 16:13:11 -04004494 }
Chris Masone02119d2008-09-05 16:13:11 -04004495
4496 key.offset = found_key.offset - 1;
Yan Zheng07d400a2009-01-06 11:42:00 -05004497 wc.replay_dest->log_root = NULL;
Chris Masone02119d2008-09-05 16:13:11 -04004498 free_extent_buffer(log->node);
Chris Masonb263c2c2009-06-11 11:24:47 -04004499 free_extent_buffer(log->commit_root);
Chris Masone02119d2008-09-05 16:13:11 -04004500 kfree(log);
4501
Josef Bacikb50c6e22013-04-25 15:55:30 -04004502 if (ret)
4503 goto error;
4504
Chris Masone02119d2008-09-05 16:13:11 -04004505 if (found_key.offset == 0)
4506 break;
4507 }
David Sterbab3b4aa72011-04-21 01:20:15 +02004508 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04004509
4510 /* step one is to pin it all, step two is to replay just inodes */
4511 if (wc.pin) {
4512 wc.pin = 0;
4513 wc.process_func = replay_one_buffer;
4514 wc.stage = LOG_WALK_REPLAY_INODES;
4515 goto again;
4516 }
4517 /* step three is to replay everything */
4518 if (wc.stage < LOG_WALK_REPLAY_ALL) {
4519 wc.stage++;
4520 goto again;
4521 }
4522
4523 btrfs_free_path(path);
4524
Josef Bacikabefa552013-04-24 16:40:05 -04004525 /* step 4: commit the transaction, which also unpins the blocks */
4526 ret = btrfs_commit_transaction(trans, fs_info->tree_root);
4527 if (ret)
4528 return ret;
4529
Chris Masone02119d2008-09-05 16:13:11 -04004530 free_extent_buffer(log_root_tree->node);
4531 log_root_tree->log_root = NULL;
4532 fs_info->log_root_recovering = 0;
Chris Masone02119d2008-09-05 16:13:11 -04004533 kfree(log_root_tree);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01004534
Josef Bacikabefa552013-04-24 16:40:05 -04004535 return 0;
Jeff Mahoney79787ea2012-03-12 16:03:00 +01004536error:
Josef Bacikb50c6e22013-04-25 15:55:30 -04004537 if (wc.trans)
4538 btrfs_end_transaction(wc.trans, fs_info->tree_root);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01004539 btrfs_free_path(path);
4540 return ret;
Chris Masone02119d2008-09-05 16:13:11 -04004541}
Chris Mason12fcfd22009-03-24 10:24:20 -04004542
4543/*
4544 * there are some corner cases where we want to force a full
4545 * commit instead of allowing a directory to be logged.
4546 *
4547 * They revolve around files there were unlinked from the directory, and
4548 * this function updates the parent directory so that a full commit is
4549 * properly done if it is fsync'd later after the unlinks are done.
4550 */
4551void btrfs_record_unlink_dir(struct btrfs_trans_handle *trans,
4552 struct inode *dir, struct inode *inode,
4553 int for_rename)
4554{
4555 /*
Chris Masonaf4176b2009-03-24 10:24:31 -04004556 * when we're logging a file, if it hasn't been renamed
4557 * or unlinked, and its inode is fully committed on disk,
4558 * we don't have to worry about walking up the directory chain
4559 * to log its parents.
4560 *
4561 * So, we use the last_unlink_trans field to put this transid
4562 * into the file. When the file is logged we check it and
4563 * don't log the parents if the file is fully on disk.
4564 */
4565 if (S_ISREG(inode->i_mode))
4566 BTRFS_I(inode)->last_unlink_trans = trans->transid;
4567
4568 /*
Chris Mason12fcfd22009-03-24 10:24:20 -04004569 * if this directory was already logged any new
4570 * names for this file/dir will get recorded
4571 */
4572 smp_mb();
4573 if (BTRFS_I(dir)->logged_trans == trans->transid)
4574 return;
4575
4576 /*
4577 * if the inode we're about to unlink was logged,
4578 * the log will be properly updated for any new names
4579 */
4580 if (BTRFS_I(inode)->logged_trans == trans->transid)
4581 return;
4582
4583 /*
4584 * when renaming files across directories, if the directory
4585 * there we're unlinking from gets fsync'd later on, there's
4586 * no way to find the destination directory later and fsync it
4587 * properly. So, we have to be conservative and force commits
4588 * so the new name gets discovered.
4589 */
4590 if (for_rename)
4591 goto record;
4592
4593 /* we can safely do the unlink without any special recording */
4594 return;
4595
4596record:
4597 BTRFS_I(dir)->last_unlink_trans = trans->transid;
4598}
4599
4600/*
4601 * Call this after adding a new name for a file and it will properly
4602 * update the log to reflect the new name.
4603 *
4604 * It will return zero if all goes well, and it will return 1 if a
4605 * full transaction commit is required.
4606 */
4607int btrfs_log_new_name(struct btrfs_trans_handle *trans,
4608 struct inode *inode, struct inode *old_dir,
4609 struct dentry *parent)
4610{
4611 struct btrfs_root * root = BTRFS_I(inode)->root;
4612
4613 /*
Chris Masonaf4176b2009-03-24 10:24:31 -04004614 * this will force the logging code to walk the dentry chain
4615 * up for the file
4616 */
4617 if (S_ISREG(inode->i_mode))
4618 BTRFS_I(inode)->last_unlink_trans = trans->transid;
4619
4620 /*
Chris Mason12fcfd22009-03-24 10:24:20 -04004621 * if this inode hasn't been logged and directory we're renaming it
4622 * from hasn't been logged, we don't need to log it
4623 */
4624 if (BTRFS_I(inode)->logged_trans <=
4625 root->fs_info->last_trans_committed &&
4626 (!old_dir || BTRFS_I(old_dir)->logged_trans <=
4627 root->fs_info->last_trans_committed))
4628 return 0;
4629
Filipe Manana49dae1b2014-09-06 22:34:39 +01004630 return btrfs_log_inode_parent(trans, root, inode, parent, 0,
4631 LLONG_MAX, 1, NULL);
Chris Mason12fcfd22009-03-24 10:24:20 -04004632}
4633