blob: d31a0c4f56bed436e0eb933cceb592fdc498eb53 [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"
Anand Jainebb87652016-03-10 17:26:59 +080029#include "compression.h"
Chris Masone02119d2008-09-05 16:13:11 -040030
31/* magic values for the inode_only field in btrfs_log_inode:
32 *
33 * LOG_INODE_ALL means to log everything
34 * LOG_INODE_EXISTS means to log just enough to recreate the inode
35 * during log replay
36 */
37#define LOG_INODE_ALL 0
38#define LOG_INODE_EXISTS 1
39
40/*
Chris Mason12fcfd22009-03-24 10:24:20 -040041 * directory trouble cases
42 *
43 * 1) on rename or unlink, if the inode being unlinked isn't in the fsync
44 * log, we must force a full commit before doing an fsync of the directory
45 * where the unlink was done.
46 * ---> record transid of last unlink/rename per directory
47 *
48 * mkdir foo/some_dir
49 * normal commit
50 * rename foo/some_dir foo2/some_dir
51 * mkdir foo/some_dir
52 * fsync foo/some_dir/some_file
53 *
54 * The fsync above will unlink the original some_dir without recording
55 * it in its new location (foo2). After a crash, some_dir will be gone
56 * unless the fsync of some_file forces a full commit
57 *
58 * 2) we must log any new names for any file or dir that is in the fsync
59 * log. ---> check inode while renaming/linking.
60 *
61 * 2a) we must log any new names for any file or dir during rename
62 * when the directory they are being removed from was logged.
63 * ---> check inode and old parent dir during rename
64 *
65 * 2a is actually the more important variant. With the extra logging
66 * a crash might unlink the old name without recreating the new one
67 *
68 * 3) after a crash, we must go through any directories with a link count
69 * of zero and redo the rm -rf
70 *
71 * mkdir f1/foo
72 * normal commit
73 * rm -rf f1/foo
74 * fsync(f1)
75 *
76 * The directory f1 was fully removed from the FS, but fsync was never
77 * called on f1, only its parent dir. After a crash the rm -rf must
78 * be replayed. This must be able to recurse down the entire
79 * directory tree. The inode link count fixup code takes care of the
80 * ugly details.
81 */
82
83/*
Chris Masone02119d2008-09-05 16:13:11 -040084 * stages for the tree walking. The first
85 * stage (0) is to only pin down the blocks we find
86 * the second stage (1) is to make sure that all the inodes
87 * we find in the log are created in the subvolume.
88 *
89 * The last stage is to deal with directories and links and extents
90 * and all the other fun semantics
91 */
92#define LOG_WALK_PIN_ONLY 0
93#define LOG_WALK_REPLAY_INODES 1
Josef Bacikdd8e7212013-09-11 11:57:23 -040094#define LOG_WALK_REPLAY_DIR_INDEX 2
95#define LOG_WALK_REPLAY_ALL 3
Chris Masone02119d2008-09-05 16:13:11 -040096
Chris Mason12fcfd22009-03-24 10:24:20 -040097static int btrfs_log_inode(struct btrfs_trans_handle *trans,
Filipe Manana49dae1b2014-09-06 22:34:39 +010098 struct btrfs_root *root, struct inode *inode,
99 int inode_only,
100 const loff_t start,
Filipe Manana8407f552014-09-05 15:14:39 +0100101 const loff_t end,
102 struct btrfs_log_ctx *ctx);
Yan Zhengec051c02009-01-05 15:43:42 -0500103static int link_to_fixup_dir(struct btrfs_trans_handle *trans,
104 struct btrfs_root *root,
105 struct btrfs_path *path, u64 objectid);
Chris Mason12fcfd22009-03-24 10:24:20 -0400106static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans,
107 struct btrfs_root *root,
108 struct btrfs_root *log,
109 struct btrfs_path *path,
110 u64 dirid, int del_all);
Chris Masone02119d2008-09-05 16:13:11 -0400111
112/*
113 * tree logging is a special write ahead log used to make sure that
114 * fsyncs and O_SYNCs can happen without doing full tree commits.
115 *
116 * Full tree commits are expensive because they require commonly
117 * modified blocks to be recowed, creating many dirty pages in the
118 * extent tree an 4x-6x higher write load than ext3.
119 *
120 * Instead of doing a tree commit on every fsync, we use the
121 * key ranges and transaction ids to find items for a given file or directory
122 * that have changed in this transaction. Those items are copied into
123 * a special tree (one per subvolume root), that tree is written to disk
124 * and then the fsync is considered complete.
125 *
126 * After a crash, items are copied out of the log-tree back into the
127 * subvolume tree. Any file data extents found are recorded in the extent
128 * allocation tree, and the log-tree freed.
129 *
130 * The log tree is read three times, once to pin down all the extents it is
131 * using in ram and once, once to create all the inodes logged in the tree
132 * and once to do all the other items.
133 */
134
135/*
Chris Masone02119d2008-09-05 16:13:11 -0400136 * start a sub transaction and setup the log tree
137 * this increments the log tree writer count to make the people
138 * syncing the tree wait for us to finish
139 */
140static int start_log_trans(struct btrfs_trans_handle *trans,
Miao Xie8b050d32014-02-20 18:08:58 +0800141 struct btrfs_root *root,
142 struct btrfs_log_ctx *ctx)
Chris Masone02119d2008-09-05 16:13:11 -0400143{
Zhaolei34eb2a52015-08-17 18:44:45 +0800144 int ret = 0;
Yan Zheng7237f182009-01-21 12:54:03 -0500145
146 mutex_lock(&root->log_mutex);
Zhaolei34eb2a52015-08-17 18:44:45 +0800147
Yan Zheng7237f182009-01-21 12:54:03 -0500148 if (root->log_root) {
Miao Xie995946d2014-04-02 19:51:06 +0800149 if (btrfs_need_log_full_commit(root->fs_info, trans)) {
Miao Xie50471a32014-02-20 18:08:57 +0800150 ret = -EAGAIN;
151 goto out;
152 }
Zhaolei34eb2a52015-08-17 18:44:45 +0800153
Josef Bacikff782e02009-10-08 15:30:04 -0400154 if (!root->log_start_pid) {
Miao Xie27cdeb72014-04-02 19:51:05 +0800155 clear_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state);
Zhaolei34eb2a52015-08-17 18:44:45 +0800156 root->log_start_pid = current->pid;
Josef Bacikff782e02009-10-08 15:30:04 -0400157 } else if (root->log_start_pid != current->pid) {
Miao Xie27cdeb72014-04-02 19:51:05 +0800158 set_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state);
Josef Bacikff782e02009-10-08 15:30:04 -0400159 }
Zhaolei34eb2a52015-08-17 18:44:45 +0800160 } else {
161 mutex_lock(&root->fs_info->tree_log_mutex);
162 if (!root->fs_info->log_root_tree)
163 ret = btrfs_init_log_root_tree(trans, root->fs_info);
164 mutex_unlock(&root->fs_info->tree_log_mutex);
165 if (ret)
166 goto out;
Josef Bacikff782e02009-10-08 15:30:04 -0400167
Chris Masone02119d2008-09-05 16:13:11 -0400168 ret = btrfs_add_log_tree(trans, root);
Yan, Zheng4a500fd2010-05-16 10:49:59 -0400169 if (ret)
Miao Xiee87ac132014-02-20 18:08:53 +0800170 goto out;
Zhaolei34eb2a52015-08-17 18:44:45 +0800171
172 clear_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state);
173 root->log_start_pid = current->pid;
Chris Masone02119d2008-09-05 16:13:11 -0400174 }
Zhaolei34eb2a52015-08-17 18:44:45 +0800175
Miao Xie2ecb7922012-09-06 04:04:27 -0600176 atomic_inc(&root->log_batch);
Yan Zheng7237f182009-01-21 12:54:03 -0500177 atomic_inc(&root->log_writers);
Miao Xie8b050d32014-02-20 18:08:58 +0800178 if (ctx) {
Zhaolei34eb2a52015-08-17 18:44:45 +0800179 int index = root->log_transid % 2;
Miao Xie8b050d32014-02-20 18:08:58 +0800180 list_add_tail(&ctx->list, &root->log_ctxs[index]);
Miao Xied1433de2014-02-20 18:08:59 +0800181 ctx->log_transid = root->log_transid;
Miao Xie8b050d32014-02-20 18:08:58 +0800182 }
Zhaolei34eb2a52015-08-17 18:44:45 +0800183
Miao Xiee87ac132014-02-20 18:08:53 +0800184out:
Yan Zheng7237f182009-01-21 12:54:03 -0500185 mutex_unlock(&root->log_mutex);
Miao Xiee87ac132014-02-20 18:08:53 +0800186 return ret;
Chris Masone02119d2008-09-05 16:13:11 -0400187}
188
189/*
190 * returns 0 if there was a log transaction running and we were able
191 * to join, or returns -ENOENT if there were not transactions
192 * in progress
193 */
194static int join_running_log_trans(struct btrfs_root *root)
195{
196 int ret = -ENOENT;
197
198 smp_mb();
199 if (!root->log_root)
200 return -ENOENT;
201
Yan Zheng7237f182009-01-21 12:54:03 -0500202 mutex_lock(&root->log_mutex);
Chris Masone02119d2008-09-05 16:13:11 -0400203 if (root->log_root) {
204 ret = 0;
Yan Zheng7237f182009-01-21 12:54:03 -0500205 atomic_inc(&root->log_writers);
Chris Masone02119d2008-09-05 16:13:11 -0400206 }
Yan Zheng7237f182009-01-21 12:54:03 -0500207 mutex_unlock(&root->log_mutex);
Chris Masone02119d2008-09-05 16:13:11 -0400208 return ret;
209}
210
211/*
Chris Mason12fcfd22009-03-24 10:24:20 -0400212 * This either makes the current running log transaction wait
213 * until you call btrfs_end_log_trans() or it makes any future
214 * log transactions wait until you call btrfs_end_log_trans()
215 */
216int btrfs_pin_log_trans(struct btrfs_root *root)
217{
218 int ret = -ENOENT;
219
220 mutex_lock(&root->log_mutex);
221 atomic_inc(&root->log_writers);
222 mutex_unlock(&root->log_mutex);
223 return ret;
224}
225
226/*
Chris Masone02119d2008-09-05 16:13:11 -0400227 * indicate we're done making changes to the log tree
228 * and wake up anyone waiting to do a sync
229 */
Jeff Mahoney143bede2012-03-01 14:56:26 +0100230void btrfs_end_log_trans(struct btrfs_root *root)
Chris Masone02119d2008-09-05 16:13:11 -0400231{
Yan Zheng7237f182009-01-21 12:54:03 -0500232 if (atomic_dec_and_test(&root->log_writers)) {
David Sterba779adf02015-02-16 19:39:00 +0100233 /*
234 * Implicit memory barrier after atomic_dec_and_test
235 */
Yan Zheng7237f182009-01-21 12:54:03 -0500236 if (waitqueue_active(&root->log_writer_wait))
237 wake_up(&root->log_writer_wait);
238 }
Chris Masone02119d2008-09-05 16:13:11 -0400239}
240
241
242/*
243 * the walk control struct is used to pass state down the chain when
244 * processing the log tree. The stage field tells us which part
245 * of the log tree processing we are currently doing. The others
246 * are state fields used for that specific part
247 */
248struct walk_control {
249 /* should we free the extent on disk when done? This is used
250 * at transaction commit time while freeing a log tree
251 */
252 int free;
253
254 /* should we write out the extent buffer? This is used
255 * while flushing the log tree to disk during a sync
256 */
257 int write;
258
259 /* should we wait for the extent buffer io to finish? Also used
260 * while flushing the log tree to disk for a sync
261 */
262 int wait;
263
264 /* pin only walk, we record which extents on disk belong to the
265 * log trees
266 */
267 int pin;
268
269 /* what stage of the replay code we're currently in */
270 int stage;
271
272 /* the root we are currently replaying */
273 struct btrfs_root *replay_dest;
274
275 /* the trans handle for the current replay */
276 struct btrfs_trans_handle *trans;
277
278 /* the function that gets used to process blocks we find in the
279 * tree. Note the extent_buffer might not be up to date when it is
280 * passed in, and it must be checked or read if you need the data
281 * inside it
282 */
283 int (*process_func)(struct btrfs_root *log, struct extent_buffer *eb,
284 struct walk_control *wc, u64 gen);
285};
286
287/*
288 * process_func used to pin down extents, write them or wait on them
289 */
290static int process_one_buffer(struct btrfs_root *log,
291 struct extent_buffer *eb,
292 struct walk_control *wc, u64 gen)
293{
Josef Bacikb50c6e22013-04-25 15:55:30 -0400294 int ret = 0;
Chris Masone02119d2008-09-05 16:13:11 -0400295
Josef Bacik8c2a1a32013-06-06 13:19:32 -0400296 /*
297 * If this fs is mixed then we need to be able to process the leaves to
298 * pin down any logged extents, so we have to read the block.
299 */
300 if (btrfs_fs_incompat(log->fs_info, MIXED_GROUPS)) {
301 ret = btrfs_read_buffer(eb, gen);
302 if (ret)
303 return ret;
304 }
305
Josef Bacikb50c6e22013-04-25 15:55:30 -0400306 if (wc->pin)
307 ret = btrfs_pin_extent_for_log_replay(log->fs_info->extent_root,
308 eb->start, eb->len);
309
310 if (!ret && btrfs_buffer_uptodate(eb, gen, 0)) {
Josef Bacik8c2a1a32013-06-06 13:19:32 -0400311 if (wc->pin && btrfs_header_level(eb) == 0)
312 ret = btrfs_exclude_logged_extents(log, eb);
Chris Masone02119d2008-09-05 16:13:11 -0400313 if (wc->write)
314 btrfs_write_tree_block(eb);
315 if (wc->wait)
316 btrfs_wait_tree_block_writeback(eb);
317 }
Josef Bacikb50c6e22013-04-25 15:55:30 -0400318 return ret;
Chris Masone02119d2008-09-05 16:13:11 -0400319}
320
321/*
322 * Item overwrite used by replay and tree logging. eb, slot and key all refer
323 * to the src data we are copying out.
324 *
325 * root is the tree we are copying into, and path is a scratch
326 * path for use in this function (it should be released on entry and
327 * will be released on exit).
328 *
329 * If the key is already in the destination tree the existing item is
330 * overwritten. If the existing item isn't big enough, it is extended.
331 * If it is too large, it is truncated.
332 *
333 * If the key isn't in the destination yet, a new item is inserted.
334 */
335static noinline int overwrite_item(struct btrfs_trans_handle *trans,
336 struct btrfs_root *root,
337 struct btrfs_path *path,
338 struct extent_buffer *eb, int slot,
339 struct btrfs_key *key)
340{
341 int ret;
342 u32 item_size;
343 u64 saved_i_size = 0;
344 int save_old_i_size = 0;
345 unsigned long src_ptr;
346 unsigned long dst_ptr;
347 int overwrite_root = 0;
Josef Bacik4bc4bee2013-04-05 20:50:09 +0000348 bool inode_item = key->type == BTRFS_INODE_ITEM_KEY;
Chris Masone02119d2008-09-05 16:13:11 -0400349
350 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
351 overwrite_root = 1;
352
353 item_size = btrfs_item_size_nr(eb, slot);
354 src_ptr = btrfs_item_ptr_offset(eb, slot);
355
356 /* look for the key in the destination tree */
357 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
Josef Bacik4bc4bee2013-04-05 20:50:09 +0000358 if (ret < 0)
359 return ret;
360
Chris Masone02119d2008-09-05 16:13:11 -0400361 if (ret == 0) {
362 char *src_copy;
363 char *dst_copy;
364 u32 dst_size = btrfs_item_size_nr(path->nodes[0],
365 path->slots[0]);
366 if (dst_size != item_size)
367 goto insert;
368
369 if (item_size == 0) {
David Sterbab3b4aa72011-04-21 01:20:15 +0200370 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -0400371 return 0;
372 }
373 dst_copy = kmalloc(item_size, GFP_NOFS);
374 src_copy = kmalloc(item_size, GFP_NOFS);
liubo2a29edc2011-01-26 06:22:08 +0000375 if (!dst_copy || !src_copy) {
David Sterbab3b4aa72011-04-21 01:20:15 +0200376 btrfs_release_path(path);
liubo2a29edc2011-01-26 06:22:08 +0000377 kfree(dst_copy);
378 kfree(src_copy);
379 return -ENOMEM;
380 }
Chris Masone02119d2008-09-05 16:13:11 -0400381
382 read_extent_buffer(eb, src_copy, src_ptr, item_size);
383
384 dst_ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
385 read_extent_buffer(path->nodes[0], dst_copy, dst_ptr,
386 item_size);
387 ret = memcmp(dst_copy, src_copy, item_size);
388
389 kfree(dst_copy);
390 kfree(src_copy);
391 /*
392 * they have the same contents, just return, this saves
393 * us from cowing blocks in the destination tree and doing
394 * extra writes that may not have been done by a previous
395 * sync
396 */
397 if (ret == 0) {
David Sterbab3b4aa72011-04-21 01:20:15 +0200398 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -0400399 return 0;
400 }
401
Josef Bacik4bc4bee2013-04-05 20:50:09 +0000402 /*
403 * We need to load the old nbytes into the inode so when we
404 * replay the extents we've logged we get the right nbytes.
405 */
406 if (inode_item) {
407 struct btrfs_inode_item *item;
408 u64 nbytes;
Josef Bacikd5554382013-09-11 14:17:00 -0400409 u32 mode;
Josef Bacik4bc4bee2013-04-05 20:50:09 +0000410
411 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
412 struct btrfs_inode_item);
413 nbytes = btrfs_inode_nbytes(path->nodes[0], item);
414 item = btrfs_item_ptr(eb, slot,
415 struct btrfs_inode_item);
416 btrfs_set_inode_nbytes(eb, item, nbytes);
Josef Bacikd5554382013-09-11 14:17:00 -0400417
418 /*
419 * If this is a directory we need to reset the i_size to
420 * 0 so that we can set it up properly when replaying
421 * the rest of the items in this log.
422 */
423 mode = btrfs_inode_mode(eb, item);
424 if (S_ISDIR(mode))
425 btrfs_set_inode_size(eb, item, 0);
Josef Bacik4bc4bee2013-04-05 20:50:09 +0000426 }
427 } else if (inode_item) {
428 struct btrfs_inode_item *item;
Josef Bacikd5554382013-09-11 14:17:00 -0400429 u32 mode;
Josef Bacik4bc4bee2013-04-05 20:50:09 +0000430
431 /*
432 * New inode, set nbytes to 0 so that the nbytes comes out
433 * properly when we replay the extents.
434 */
435 item = btrfs_item_ptr(eb, slot, struct btrfs_inode_item);
436 btrfs_set_inode_nbytes(eb, item, 0);
Josef Bacikd5554382013-09-11 14:17:00 -0400437
438 /*
439 * If this is a directory we need to reset the i_size to 0 so
440 * that we can set it up properly when replaying the rest of
441 * the items in this log.
442 */
443 mode = btrfs_inode_mode(eb, item);
444 if (S_ISDIR(mode))
445 btrfs_set_inode_size(eb, item, 0);
Chris Masone02119d2008-09-05 16:13:11 -0400446 }
447insert:
David Sterbab3b4aa72011-04-21 01:20:15 +0200448 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -0400449 /* try to insert the key into the destination tree */
Filipe Mananadf8d1162015-01-14 01:52:25 +0000450 path->skip_release_on_error = 1;
Chris Masone02119d2008-09-05 16:13:11 -0400451 ret = btrfs_insert_empty_item(trans, root, path,
452 key, item_size);
Filipe Mananadf8d1162015-01-14 01:52:25 +0000453 path->skip_release_on_error = 0;
Chris Masone02119d2008-09-05 16:13:11 -0400454
455 /* make sure any existing item is the correct size */
Filipe Mananadf8d1162015-01-14 01:52:25 +0000456 if (ret == -EEXIST || ret == -EOVERFLOW) {
Chris Masone02119d2008-09-05 16:13:11 -0400457 u32 found_size;
458 found_size = btrfs_item_size_nr(path->nodes[0],
459 path->slots[0]);
Jeff Mahoney143bede2012-03-01 14:56:26 +0100460 if (found_size > item_size)
Tsutomu Itohafe5fea2013-04-16 05:18:22 +0000461 btrfs_truncate_item(root, path, item_size, 1);
Jeff Mahoney143bede2012-03-01 14:56:26 +0100462 else if (found_size < item_size)
Tsutomu Itoh4b90c682013-04-16 05:18:49 +0000463 btrfs_extend_item(root, path,
Jeff Mahoney143bede2012-03-01 14:56:26 +0100464 item_size - found_size);
Chris Masone02119d2008-09-05 16:13:11 -0400465 } else if (ret) {
Yan, Zheng4a500fd2010-05-16 10:49:59 -0400466 return ret;
Chris Masone02119d2008-09-05 16:13:11 -0400467 }
468 dst_ptr = btrfs_item_ptr_offset(path->nodes[0],
469 path->slots[0]);
470
471 /* don't overwrite an existing inode if the generation number
472 * was logged as zero. This is done when the tree logging code
473 * is just logging an inode to make sure it exists after recovery.
474 *
475 * Also, don't overwrite i_size on directories during replay.
476 * log replay inserts and removes directory items based on the
477 * state of the tree found in the subvolume, and i_size is modified
478 * as it goes
479 */
480 if (key->type == BTRFS_INODE_ITEM_KEY && ret == -EEXIST) {
481 struct btrfs_inode_item *src_item;
482 struct btrfs_inode_item *dst_item;
483
484 src_item = (struct btrfs_inode_item *)src_ptr;
485 dst_item = (struct btrfs_inode_item *)dst_ptr;
486
Filipe Manana1a4bcf42015-02-13 12:30:56 +0000487 if (btrfs_inode_generation(eb, src_item) == 0) {
488 struct extent_buffer *dst_eb = path->nodes[0];
Filipe Manana2f2ff0e2015-03-20 17:19:46 +0000489 const u64 ino_size = btrfs_inode_size(eb, src_item);
Filipe Manana1a4bcf42015-02-13 12:30:56 +0000490
Filipe Manana2f2ff0e2015-03-20 17:19:46 +0000491 /*
492 * For regular files an ino_size == 0 is used only when
493 * logging that an inode exists, as part of a directory
494 * fsync, and the inode wasn't fsynced before. In this
495 * case don't set the size of the inode in the fs/subvol
496 * tree, otherwise we would be throwing valid data away.
497 */
Filipe Manana1a4bcf42015-02-13 12:30:56 +0000498 if (S_ISREG(btrfs_inode_mode(eb, src_item)) &&
Filipe Manana2f2ff0e2015-03-20 17:19:46 +0000499 S_ISREG(btrfs_inode_mode(dst_eb, dst_item)) &&
500 ino_size != 0) {
Filipe Manana1a4bcf42015-02-13 12:30:56 +0000501 struct btrfs_map_token token;
Filipe Manana1a4bcf42015-02-13 12:30:56 +0000502
503 btrfs_init_map_token(&token);
504 btrfs_set_token_inode_size(dst_eb, dst_item,
505 ino_size, &token);
506 }
Chris Masone02119d2008-09-05 16:13:11 -0400507 goto no_copy;
Filipe Manana1a4bcf42015-02-13 12:30:56 +0000508 }
Chris Masone02119d2008-09-05 16:13:11 -0400509
510 if (overwrite_root &&
511 S_ISDIR(btrfs_inode_mode(eb, src_item)) &&
512 S_ISDIR(btrfs_inode_mode(path->nodes[0], dst_item))) {
513 save_old_i_size = 1;
514 saved_i_size = btrfs_inode_size(path->nodes[0],
515 dst_item);
516 }
517 }
518
519 copy_extent_buffer(path->nodes[0], eb, dst_ptr,
520 src_ptr, item_size);
521
522 if (save_old_i_size) {
523 struct btrfs_inode_item *dst_item;
524 dst_item = (struct btrfs_inode_item *)dst_ptr;
525 btrfs_set_inode_size(path->nodes[0], dst_item, saved_i_size);
526 }
527
528 /* make sure the generation is filled in */
529 if (key->type == BTRFS_INODE_ITEM_KEY) {
530 struct btrfs_inode_item *dst_item;
531 dst_item = (struct btrfs_inode_item *)dst_ptr;
532 if (btrfs_inode_generation(path->nodes[0], dst_item) == 0) {
533 btrfs_set_inode_generation(path->nodes[0], dst_item,
534 trans->transid);
535 }
536 }
537no_copy:
538 btrfs_mark_buffer_dirty(path->nodes[0]);
David Sterbab3b4aa72011-04-21 01:20:15 +0200539 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -0400540 return 0;
541}
542
543/*
544 * simple helper to read an inode off the disk from a given root
545 * This can only be called for subvolume roots and not for the log
546 */
547static noinline struct inode *read_one_inode(struct btrfs_root *root,
548 u64 objectid)
549{
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400550 struct btrfs_key key;
Chris Masone02119d2008-09-05 16:13:11 -0400551 struct inode *inode;
Chris Masone02119d2008-09-05 16:13:11 -0400552
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400553 key.objectid = objectid;
554 key.type = BTRFS_INODE_ITEM_KEY;
555 key.offset = 0;
Josef Bacik73f73412009-12-04 17:38:27 +0000556 inode = btrfs_iget(root->fs_info->sb, &key, root, NULL);
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400557 if (IS_ERR(inode)) {
558 inode = NULL;
559 } else if (is_bad_inode(inode)) {
Chris Masone02119d2008-09-05 16:13:11 -0400560 iput(inode);
561 inode = NULL;
562 }
563 return inode;
564}
565
566/* replays a single extent in 'eb' at 'slot' with 'key' into the
567 * subvolume 'root'. path is released on entry and should be released
568 * on exit.
569 *
570 * extents in the log tree have not been allocated out of the extent
571 * tree yet. So, this completes the allocation, taking a reference
572 * as required if the extent already exists or creating a new extent
573 * if it isn't in the extent allocation tree yet.
574 *
575 * The extent is inserted into the file, dropping any existing extents
576 * from the file that overlap the new one.
577 */
578static noinline int replay_one_extent(struct btrfs_trans_handle *trans,
579 struct btrfs_root *root,
580 struct btrfs_path *path,
581 struct extent_buffer *eb, int slot,
582 struct btrfs_key *key)
583{
584 int found_type;
Chris Masone02119d2008-09-05 16:13:11 -0400585 u64 extent_end;
Chris Masone02119d2008-09-05 16:13:11 -0400586 u64 start = key->offset;
Josef Bacik4bc4bee2013-04-05 20:50:09 +0000587 u64 nbytes = 0;
Chris Masone02119d2008-09-05 16:13:11 -0400588 struct btrfs_file_extent_item *item;
589 struct inode *inode = NULL;
590 unsigned long size;
591 int ret = 0;
592
593 item = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
594 found_type = btrfs_file_extent_type(eb, item);
595
Yan Zhengd899e052008-10-30 14:25:28 -0400596 if (found_type == BTRFS_FILE_EXTENT_REG ||
Josef Bacik4bc4bee2013-04-05 20:50:09 +0000597 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
598 nbytes = btrfs_file_extent_num_bytes(eb, item);
599 extent_end = start + nbytes;
600
601 /*
602 * We don't add to the inodes nbytes if we are prealloc or a
603 * hole.
604 */
605 if (btrfs_file_extent_disk_bytenr(eb, item) == 0)
606 nbytes = 0;
607 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
Chris Mason514ac8a2014-01-03 21:07:00 -0800608 size = btrfs_file_extent_inline_len(eb, slot, item);
Josef Bacik4bc4bee2013-04-05 20:50:09 +0000609 nbytes = btrfs_file_extent_ram_bytes(eb, item);
Qu Wenruofda28322013-02-26 08:10:22 +0000610 extent_end = ALIGN(start + size, root->sectorsize);
Chris Masone02119d2008-09-05 16:13:11 -0400611 } else {
612 ret = 0;
613 goto out;
614 }
615
616 inode = read_one_inode(root, key->objectid);
617 if (!inode) {
618 ret = -EIO;
619 goto out;
620 }
621
622 /*
623 * first check to see if we already have this extent in the
624 * file. This must be done before the btrfs_drop_extents run
625 * so we don't try to drop this extent.
626 */
Li Zefan33345d012011-04-20 10:31:50 +0800627 ret = btrfs_lookup_file_extent(trans, root, path, btrfs_ino(inode),
Chris Masone02119d2008-09-05 16:13:11 -0400628 start, 0);
629
Yan Zhengd899e052008-10-30 14:25:28 -0400630 if (ret == 0 &&
631 (found_type == BTRFS_FILE_EXTENT_REG ||
632 found_type == BTRFS_FILE_EXTENT_PREALLOC)) {
Chris Masone02119d2008-09-05 16:13:11 -0400633 struct btrfs_file_extent_item cmp1;
634 struct btrfs_file_extent_item cmp2;
635 struct btrfs_file_extent_item *existing;
636 struct extent_buffer *leaf;
637
638 leaf = path->nodes[0];
639 existing = btrfs_item_ptr(leaf, path->slots[0],
640 struct btrfs_file_extent_item);
641
642 read_extent_buffer(eb, &cmp1, (unsigned long)item,
643 sizeof(cmp1));
644 read_extent_buffer(leaf, &cmp2, (unsigned long)existing,
645 sizeof(cmp2));
646
647 /*
648 * we already have a pointer to this exact extent,
649 * we don't have to do anything
650 */
651 if (memcmp(&cmp1, &cmp2, sizeof(cmp1)) == 0) {
David Sterbab3b4aa72011-04-21 01:20:15 +0200652 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -0400653 goto out;
654 }
655 }
David Sterbab3b4aa72011-04-21 01:20:15 +0200656 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -0400657
658 /* drop any overlapping extents */
Josef Bacik26714852012-08-29 12:24:27 -0400659 ret = btrfs_drop_extents(trans, root, inode, start, extent_end, 1);
Josef Bacik36508602013-04-25 16:23:32 -0400660 if (ret)
661 goto out;
Chris Masone02119d2008-09-05 16:13:11 -0400662
Yan Zheng07d400a2009-01-06 11:42:00 -0500663 if (found_type == BTRFS_FILE_EXTENT_REG ||
664 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400665 u64 offset;
Yan Zheng07d400a2009-01-06 11:42:00 -0500666 unsigned long dest_offset;
667 struct btrfs_key ins;
Chris Masone02119d2008-09-05 16:13:11 -0400668
Yan Zheng07d400a2009-01-06 11:42:00 -0500669 ret = btrfs_insert_empty_item(trans, root, path, key,
670 sizeof(*item));
Josef Bacik36508602013-04-25 16:23:32 -0400671 if (ret)
672 goto out;
Yan Zheng07d400a2009-01-06 11:42:00 -0500673 dest_offset = btrfs_item_ptr_offset(path->nodes[0],
674 path->slots[0]);
675 copy_extent_buffer(path->nodes[0], eb, dest_offset,
676 (unsigned long)item, sizeof(*item));
677
678 ins.objectid = btrfs_file_extent_disk_bytenr(eb, item);
679 ins.offset = btrfs_file_extent_disk_num_bytes(eb, item);
680 ins.type = BTRFS_EXTENT_ITEM_KEY;
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400681 offset = key->offset - btrfs_file_extent_offset(eb, item);
Yan Zheng07d400a2009-01-06 11:42:00 -0500682
683 if (ins.objectid > 0) {
684 u64 csum_start;
685 u64 csum_end;
686 LIST_HEAD(ordered_sums);
687 /*
688 * is this extent already allocated in the extent
689 * allocation tree? If so, just add a reference
690 */
Filipe Manana1a4ed8f2014-10-27 10:44:24 +0000691 ret = btrfs_lookup_data_extent(root, ins.objectid,
Yan Zheng07d400a2009-01-06 11:42:00 -0500692 ins.offset);
693 if (ret == 0) {
694 ret = btrfs_inc_extent_ref(trans, root,
695 ins.objectid, ins.offset,
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400696 0, root->root_key.objectid,
Filipe Mananab06c4bf2015-10-23 07:52:54 +0100697 key->objectid, offset);
Josef Bacikb50c6e22013-04-25 15:55:30 -0400698 if (ret)
699 goto out;
Yan Zheng07d400a2009-01-06 11:42:00 -0500700 } else {
701 /*
702 * insert the extent pointer in the extent
703 * allocation tree
704 */
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400705 ret = btrfs_alloc_logged_file_extent(trans,
706 root, root->root_key.objectid,
707 key->objectid, offset, &ins);
Josef Bacikb50c6e22013-04-25 15:55:30 -0400708 if (ret)
709 goto out;
Yan Zheng07d400a2009-01-06 11:42:00 -0500710 }
David Sterbab3b4aa72011-04-21 01:20:15 +0200711 btrfs_release_path(path);
Yan Zheng07d400a2009-01-06 11:42:00 -0500712
713 if (btrfs_file_extent_compression(eb, item)) {
714 csum_start = ins.objectid;
715 csum_end = csum_start + ins.offset;
716 } else {
717 csum_start = ins.objectid +
718 btrfs_file_extent_offset(eb, item);
719 csum_end = csum_start +
720 btrfs_file_extent_num_bytes(eb, item);
721 }
722
723 ret = btrfs_lookup_csums_range(root->log_root,
724 csum_start, csum_end - 1,
Arne Jansena2de7332011-03-08 14:14:00 +0100725 &ordered_sums, 0);
Josef Bacik36508602013-04-25 16:23:32 -0400726 if (ret)
727 goto out;
Filipe Mananab84b8392015-08-19 11:09:40 +0100728 /*
729 * Now delete all existing cums in the csum root that
730 * cover our range. We do this because we can have an
731 * extent that is completely referenced by one file
732 * extent item and partially referenced by another
733 * file extent item (like after using the clone or
734 * extent_same ioctls). In this case if we end up doing
735 * the replay of the one that partially references the
736 * extent first, and we do not do the csum deletion
737 * below, we can get 2 csum items in the csum tree that
738 * overlap each other. For example, imagine our log has
739 * the two following file extent items:
740 *
741 * key (257 EXTENT_DATA 409600)
742 * extent data disk byte 12845056 nr 102400
743 * extent data offset 20480 nr 20480 ram 102400
744 *
745 * key (257 EXTENT_DATA 819200)
746 * extent data disk byte 12845056 nr 102400
747 * extent data offset 0 nr 102400 ram 102400
748 *
749 * Where the second one fully references the 100K extent
750 * that starts at disk byte 12845056, and the log tree
751 * has a single csum item that covers the entire range
752 * of the extent:
753 *
754 * key (EXTENT_CSUM EXTENT_CSUM 12845056) itemsize 100
755 *
756 * After the first file extent item is replayed, the
757 * csum tree gets the following csum item:
758 *
759 * key (EXTENT_CSUM EXTENT_CSUM 12865536) itemsize 20
760 *
761 * Which covers the 20K sub-range starting at offset 20K
762 * of our extent. Now when we replay the second file
763 * extent item, if we do not delete existing csum items
764 * that cover any of its blocks, we end up getting two
765 * csum items in our csum tree that overlap each other:
766 *
767 * key (EXTENT_CSUM EXTENT_CSUM 12845056) itemsize 100
768 * key (EXTENT_CSUM EXTENT_CSUM 12865536) itemsize 20
769 *
770 * Which is a problem, because after this anyone trying
771 * to lookup up for the checksum of any block of our
772 * extent starting at an offset of 40K or higher, will
773 * end up looking at the second csum item only, which
774 * does not contain the checksum for any block starting
775 * at offset 40K or higher of our extent.
776 */
Yan Zheng07d400a2009-01-06 11:42:00 -0500777 while (!list_empty(&ordered_sums)) {
778 struct btrfs_ordered_sum *sums;
779 sums = list_entry(ordered_sums.next,
780 struct btrfs_ordered_sum,
781 list);
Josef Bacik36508602013-04-25 16:23:32 -0400782 if (!ret)
Filipe Mananab84b8392015-08-19 11:09:40 +0100783 ret = btrfs_del_csums(trans,
784 root->fs_info->csum_root,
785 sums->bytenr,
786 sums->len);
787 if (!ret)
Josef Bacik36508602013-04-25 16:23:32 -0400788 ret = btrfs_csum_file_blocks(trans,
Yan Zheng07d400a2009-01-06 11:42:00 -0500789 root->fs_info->csum_root,
790 sums);
Yan Zheng07d400a2009-01-06 11:42:00 -0500791 list_del(&sums->list);
792 kfree(sums);
793 }
Josef Bacik36508602013-04-25 16:23:32 -0400794 if (ret)
795 goto out;
Yan Zheng07d400a2009-01-06 11:42:00 -0500796 } else {
David Sterbab3b4aa72011-04-21 01:20:15 +0200797 btrfs_release_path(path);
Yan Zheng07d400a2009-01-06 11:42:00 -0500798 }
799 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
800 /* inline extents are easy, we just overwrite them */
801 ret = overwrite_item(trans, root, path, eb, slot, key);
Josef Bacik36508602013-04-25 16:23:32 -0400802 if (ret)
803 goto out;
Yan Zheng07d400a2009-01-06 11:42:00 -0500804 }
805
Josef Bacik4bc4bee2013-04-05 20:50:09 +0000806 inode_add_bytes(inode, nbytes);
Tsutomu Itohb9959292012-06-25 21:25:22 -0600807 ret = btrfs_update_inode(trans, root, inode);
Chris Masone02119d2008-09-05 16:13:11 -0400808out:
809 if (inode)
810 iput(inode);
811 return ret;
812}
813
814/*
815 * when cleaning up conflicts between the directory names in the
816 * subvolume, directory names in the log and directory names in the
817 * inode back references, we may have to unlink inodes from directories.
818 *
819 * This is a helper function to do the unlink of a specific directory
820 * item
821 */
822static noinline int drop_one_dir_item(struct btrfs_trans_handle *trans,
823 struct btrfs_root *root,
824 struct btrfs_path *path,
825 struct inode *dir,
826 struct btrfs_dir_item *di)
827{
828 struct inode *inode;
829 char *name;
830 int name_len;
831 struct extent_buffer *leaf;
832 struct btrfs_key location;
833 int ret;
834
835 leaf = path->nodes[0];
836
837 btrfs_dir_item_key_to_cpu(leaf, di, &location);
838 name_len = btrfs_dir_name_len(leaf, di);
839 name = kmalloc(name_len, GFP_NOFS);
liubo2a29edc2011-01-26 06:22:08 +0000840 if (!name)
841 return -ENOMEM;
842
Chris Masone02119d2008-09-05 16:13:11 -0400843 read_extent_buffer(leaf, name, (unsigned long)(di + 1), name_len);
David Sterbab3b4aa72011-04-21 01:20:15 +0200844 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -0400845
846 inode = read_one_inode(root, location.objectid);
Tsutomu Itohc00e9492011-04-28 09:10:23 +0000847 if (!inode) {
Josef Bacik36508602013-04-25 16:23:32 -0400848 ret = -EIO;
849 goto out;
Tsutomu Itohc00e9492011-04-28 09:10:23 +0000850 }
Chris Masone02119d2008-09-05 16:13:11 -0400851
Yan Zhengec051c02009-01-05 15:43:42 -0500852 ret = link_to_fixup_dir(trans, root, path, location.objectid);
Josef Bacik36508602013-04-25 16:23:32 -0400853 if (ret)
854 goto out;
Chris Mason12fcfd22009-03-24 10:24:20 -0400855
Chris Masone02119d2008-09-05 16:13:11 -0400856 ret = btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
Josef Bacik36508602013-04-25 16:23:32 -0400857 if (ret)
858 goto out;
Filipe David Borba Mananaada9af22013-08-05 09:25:47 +0100859 else
860 ret = btrfs_run_delayed_items(trans, root);
Josef Bacik36508602013-04-25 16:23:32 -0400861out:
862 kfree(name);
863 iput(inode);
Chris Masone02119d2008-09-05 16:13:11 -0400864 return ret;
865}
866
867/*
868 * helper function to see if a given name and sequence number found
869 * in an inode back reference are already in a directory and correctly
870 * point to this inode
871 */
872static noinline int inode_in_dir(struct btrfs_root *root,
873 struct btrfs_path *path,
874 u64 dirid, u64 objectid, u64 index,
875 const char *name, int name_len)
876{
877 struct btrfs_dir_item *di;
878 struct btrfs_key location;
879 int match = 0;
880
881 di = btrfs_lookup_dir_index_item(NULL, root, path, dirid,
882 index, name, name_len, 0);
883 if (di && !IS_ERR(di)) {
884 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
885 if (location.objectid != objectid)
886 goto out;
887 } else
888 goto out;
David Sterbab3b4aa72011-04-21 01:20:15 +0200889 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -0400890
891 di = btrfs_lookup_dir_item(NULL, root, path, dirid, name, name_len, 0);
892 if (di && !IS_ERR(di)) {
893 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
894 if (location.objectid != objectid)
895 goto out;
896 } else
897 goto out;
898 match = 1;
899out:
David Sterbab3b4aa72011-04-21 01:20:15 +0200900 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -0400901 return match;
902}
903
904/*
905 * helper function to check a log tree for a named back reference in
906 * an inode. This is used to decide if a back reference that is
907 * found in the subvolume conflicts with what we find in the log.
908 *
909 * inode backreferences may have multiple refs in a single item,
910 * during replay we process one reference at a time, and we don't
911 * want to delete valid links to a file from the subvolume if that
912 * link is also in the log.
913 */
914static noinline int backref_in_log(struct btrfs_root *log,
915 struct btrfs_key *key,
Mark Fashehf1863732012-08-08 11:32:27 -0700916 u64 ref_objectid,
Filipe Mananadf8d1162015-01-14 01:52:25 +0000917 const char *name, int namelen)
Chris Masone02119d2008-09-05 16:13:11 -0400918{
919 struct btrfs_path *path;
920 struct btrfs_inode_ref *ref;
921 unsigned long ptr;
922 unsigned long ptr_end;
923 unsigned long name_ptr;
924 int found_name_len;
925 int item_size;
926 int ret;
927 int match = 0;
928
929 path = btrfs_alloc_path();
liubo2a29edc2011-01-26 06:22:08 +0000930 if (!path)
931 return -ENOMEM;
932
Chris Masone02119d2008-09-05 16:13:11 -0400933 ret = btrfs_search_slot(NULL, log, key, path, 0, 0);
934 if (ret != 0)
935 goto out;
936
Chris Masone02119d2008-09-05 16:13:11 -0400937 ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
Mark Fashehf1863732012-08-08 11:32:27 -0700938
939 if (key->type == BTRFS_INODE_EXTREF_KEY) {
940 if (btrfs_find_name_in_ext_backref(path, ref_objectid,
941 name, namelen, NULL))
942 match = 1;
943
944 goto out;
945 }
946
947 item_size = btrfs_item_size_nr(path->nodes[0], path->slots[0]);
Chris Masone02119d2008-09-05 16:13:11 -0400948 ptr_end = ptr + item_size;
949 while (ptr < ptr_end) {
950 ref = (struct btrfs_inode_ref *)ptr;
951 found_name_len = btrfs_inode_ref_name_len(path->nodes[0], ref);
952 if (found_name_len == namelen) {
953 name_ptr = (unsigned long)(ref + 1);
954 ret = memcmp_extent_buffer(path->nodes[0], name,
955 name_ptr, namelen);
956 if (ret == 0) {
957 match = 1;
958 goto out;
959 }
960 }
961 ptr = (unsigned long)(ref + 1) + found_name_len;
962 }
963out:
964 btrfs_free_path(path);
965 return match;
966}
967
Jan Schmidt5a1d7842012-08-17 14:04:41 -0700968static inline int __add_inode_ref(struct btrfs_trans_handle *trans,
969 struct btrfs_root *root,
970 struct btrfs_path *path,
971 struct btrfs_root *log_root,
972 struct inode *dir, struct inode *inode,
Jan Schmidt5a1d7842012-08-17 14:04:41 -0700973 struct extent_buffer *eb,
Mark Fashehf1863732012-08-08 11:32:27 -0700974 u64 inode_objectid, u64 parent_objectid,
975 u64 ref_index, char *name, int namelen,
976 int *search_done)
Jan Schmidt5a1d7842012-08-17 14:04:41 -0700977{
978 int ret;
Mark Fashehf1863732012-08-08 11:32:27 -0700979 char *victim_name;
980 int victim_name_len;
981 struct extent_buffer *leaf;
Jan Schmidt5a1d7842012-08-17 14:04:41 -0700982 struct btrfs_dir_item *di;
Mark Fashehf1863732012-08-08 11:32:27 -0700983 struct btrfs_key search_key;
984 struct btrfs_inode_extref *extref;
Jan Schmidt5a1d7842012-08-17 14:04:41 -0700985
Mark Fashehf1863732012-08-08 11:32:27 -0700986again:
987 /* Search old style refs */
988 search_key.objectid = inode_objectid;
989 search_key.type = BTRFS_INODE_REF_KEY;
990 search_key.offset = parent_objectid;
991 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
Jan Schmidt5a1d7842012-08-17 14:04:41 -0700992 if (ret == 0) {
Jan Schmidt5a1d7842012-08-17 14:04:41 -0700993 struct btrfs_inode_ref *victim_ref;
994 unsigned long ptr;
995 unsigned long ptr_end;
Mark Fashehf1863732012-08-08 11:32:27 -0700996
997 leaf = path->nodes[0];
Jan Schmidt5a1d7842012-08-17 14:04:41 -0700998
999 /* are we trying to overwrite a back ref for the root directory
1000 * if so, just jump out, we're done
1001 */
Mark Fashehf1863732012-08-08 11:32:27 -07001002 if (search_key.objectid == search_key.offset)
Jan Schmidt5a1d7842012-08-17 14:04:41 -07001003 return 1;
1004
1005 /* check all the names in this back reference to see
1006 * if they are in the log. if so, we allow them to stay
1007 * otherwise they must be unlinked as a conflict
1008 */
1009 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
1010 ptr_end = ptr + btrfs_item_size_nr(leaf, path->slots[0]);
1011 while (ptr < ptr_end) {
1012 victim_ref = (struct btrfs_inode_ref *)ptr;
1013 victim_name_len = btrfs_inode_ref_name_len(leaf,
1014 victim_ref);
1015 victim_name = kmalloc(victim_name_len, GFP_NOFS);
Josef Bacik36508602013-04-25 16:23:32 -04001016 if (!victim_name)
1017 return -ENOMEM;
Jan Schmidt5a1d7842012-08-17 14:04:41 -07001018
1019 read_extent_buffer(leaf, victim_name,
1020 (unsigned long)(victim_ref + 1),
1021 victim_name_len);
1022
Mark Fashehf1863732012-08-08 11:32:27 -07001023 if (!backref_in_log(log_root, &search_key,
1024 parent_objectid,
1025 victim_name,
Jan Schmidt5a1d7842012-08-17 14:04:41 -07001026 victim_name_len)) {
Zach Brown8b558c52013-10-16 12:10:34 -07001027 inc_nlink(inode);
Jan Schmidt5a1d7842012-08-17 14:04:41 -07001028 btrfs_release_path(path);
1029
1030 ret = btrfs_unlink_inode(trans, root, dir,
1031 inode, victim_name,
1032 victim_name_len);
Mark Fashehf1863732012-08-08 11:32:27 -07001033 kfree(victim_name);
Josef Bacik36508602013-04-25 16:23:32 -04001034 if (ret)
1035 return ret;
Filipe David Borba Mananaada9af22013-08-05 09:25:47 +01001036 ret = btrfs_run_delayed_items(trans, root);
1037 if (ret)
1038 return ret;
Mark Fashehf1863732012-08-08 11:32:27 -07001039 *search_done = 1;
1040 goto again;
Jan Schmidt5a1d7842012-08-17 14:04:41 -07001041 }
1042 kfree(victim_name);
Mark Fashehf1863732012-08-08 11:32:27 -07001043
Jan Schmidt5a1d7842012-08-17 14:04:41 -07001044 ptr = (unsigned long)(victim_ref + 1) + victim_name_len;
1045 }
Jan Schmidt5a1d7842012-08-17 14:04:41 -07001046
1047 /*
1048 * NOTE: we have searched root tree and checked the
Adam Buchbinderbb7ab3b2016-03-04 11:23:12 -08001049 * corresponding ref, it does not need to check again.
Jan Schmidt5a1d7842012-08-17 14:04:41 -07001050 */
1051 *search_done = 1;
1052 }
1053 btrfs_release_path(path);
1054
Mark Fashehf1863732012-08-08 11:32:27 -07001055 /* Same search but for extended refs */
1056 extref = btrfs_lookup_inode_extref(NULL, root, path, name, namelen,
1057 inode_objectid, parent_objectid, 0,
1058 0);
1059 if (!IS_ERR_OR_NULL(extref)) {
1060 u32 item_size;
1061 u32 cur_offset = 0;
1062 unsigned long base;
1063 struct inode *victim_parent;
1064
1065 leaf = path->nodes[0];
1066
1067 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1068 base = btrfs_item_ptr_offset(leaf, path->slots[0]);
1069
1070 while (cur_offset < item_size) {
Quentin Casasnovasdd9ef132015-03-03 16:31:38 +01001071 extref = (struct btrfs_inode_extref *)(base + cur_offset);
Mark Fashehf1863732012-08-08 11:32:27 -07001072
1073 victim_name_len = btrfs_inode_extref_name_len(leaf, extref);
1074
1075 if (btrfs_inode_extref_parent(leaf, extref) != parent_objectid)
1076 goto next;
1077
1078 victim_name = kmalloc(victim_name_len, GFP_NOFS);
Josef Bacik36508602013-04-25 16:23:32 -04001079 if (!victim_name)
1080 return -ENOMEM;
Mark Fashehf1863732012-08-08 11:32:27 -07001081 read_extent_buffer(leaf, victim_name, (unsigned long)&extref->name,
1082 victim_name_len);
1083
1084 search_key.objectid = inode_objectid;
1085 search_key.type = BTRFS_INODE_EXTREF_KEY;
1086 search_key.offset = btrfs_extref_hash(parent_objectid,
1087 victim_name,
1088 victim_name_len);
1089 ret = 0;
1090 if (!backref_in_log(log_root, &search_key,
1091 parent_objectid, victim_name,
1092 victim_name_len)) {
1093 ret = -ENOENT;
1094 victim_parent = read_one_inode(root,
1095 parent_objectid);
1096 if (victim_parent) {
Zach Brown8b558c52013-10-16 12:10:34 -07001097 inc_nlink(inode);
Mark Fashehf1863732012-08-08 11:32:27 -07001098 btrfs_release_path(path);
1099
1100 ret = btrfs_unlink_inode(trans, root,
1101 victim_parent,
1102 inode,
1103 victim_name,
1104 victim_name_len);
Filipe David Borba Mananaada9af22013-08-05 09:25:47 +01001105 if (!ret)
1106 ret = btrfs_run_delayed_items(
1107 trans, root);
Mark Fashehf1863732012-08-08 11:32:27 -07001108 }
Mark Fashehf1863732012-08-08 11:32:27 -07001109 iput(victim_parent);
1110 kfree(victim_name);
Josef Bacik36508602013-04-25 16:23:32 -04001111 if (ret)
1112 return ret;
Mark Fashehf1863732012-08-08 11:32:27 -07001113 *search_done = 1;
1114 goto again;
1115 }
1116 kfree(victim_name);
Josef Bacik36508602013-04-25 16:23:32 -04001117 if (ret)
1118 return ret;
Mark Fashehf1863732012-08-08 11:32:27 -07001119next:
1120 cur_offset += victim_name_len + sizeof(*extref);
1121 }
1122 *search_done = 1;
1123 }
1124 btrfs_release_path(path);
1125
Jan Schmidt5a1d7842012-08-17 14:04:41 -07001126 /* look for a conflicting sequence number */
1127 di = btrfs_lookup_dir_index_item(trans, root, path, btrfs_ino(dir),
Mark Fashehf1863732012-08-08 11:32:27 -07001128 ref_index, name, namelen, 0);
Jan Schmidt5a1d7842012-08-17 14:04:41 -07001129 if (di && !IS_ERR(di)) {
1130 ret = drop_one_dir_item(trans, root, path, dir, di);
Josef Bacik36508602013-04-25 16:23:32 -04001131 if (ret)
1132 return ret;
Jan Schmidt5a1d7842012-08-17 14:04:41 -07001133 }
1134 btrfs_release_path(path);
1135
1136 /* look for a conflicing name */
1137 di = btrfs_lookup_dir_item(trans, root, path, btrfs_ino(dir),
1138 name, namelen, 0);
1139 if (di && !IS_ERR(di)) {
1140 ret = drop_one_dir_item(trans, root, path, dir, di);
Josef Bacik36508602013-04-25 16:23:32 -04001141 if (ret)
1142 return ret;
Jan Schmidt5a1d7842012-08-17 14:04:41 -07001143 }
1144 btrfs_release_path(path);
1145
1146 return 0;
1147}
Chris Masone02119d2008-09-05 16:13:11 -04001148
Mark Fashehf1863732012-08-08 11:32:27 -07001149static int extref_get_fields(struct extent_buffer *eb, unsigned long ref_ptr,
1150 u32 *namelen, char **name, u64 *index,
1151 u64 *parent_objectid)
1152{
1153 struct btrfs_inode_extref *extref;
1154
1155 extref = (struct btrfs_inode_extref *)ref_ptr;
1156
1157 *namelen = btrfs_inode_extref_name_len(eb, extref);
1158 *name = kmalloc(*namelen, GFP_NOFS);
1159 if (*name == NULL)
1160 return -ENOMEM;
1161
1162 read_extent_buffer(eb, *name, (unsigned long)&extref->name,
1163 *namelen);
1164
1165 *index = btrfs_inode_extref_index(eb, extref);
1166 if (parent_objectid)
1167 *parent_objectid = btrfs_inode_extref_parent(eb, extref);
1168
1169 return 0;
1170}
1171
1172static int ref_get_fields(struct extent_buffer *eb, unsigned long ref_ptr,
1173 u32 *namelen, char **name, u64 *index)
1174{
1175 struct btrfs_inode_ref *ref;
1176
1177 ref = (struct btrfs_inode_ref *)ref_ptr;
1178
1179 *namelen = btrfs_inode_ref_name_len(eb, ref);
1180 *name = kmalloc(*namelen, GFP_NOFS);
1181 if (*name == NULL)
1182 return -ENOMEM;
1183
1184 read_extent_buffer(eb, *name, (unsigned long)(ref + 1), *namelen);
1185
1186 *index = btrfs_inode_ref_index(eb, ref);
1187
1188 return 0;
1189}
1190
Chris Masone02119d2008-09-05 16:13:11 -04001191/*
1192 * replay one inode back reference item found in the log tree.
1193 * eb, slot and key refer to the buffer and key found in the log tree.
1194 * root is the destination we are replaying into, and path is for temp
1195 * use by this function. (it should be released on return).
1196 */
1197static noinline int add_inode_ref(struct btrfs_trans_handle *trans,
1198 struct btrfs_root *root,
1199 struct btrfs_root *log,
1200 struct btrfs_path *path,
1201 struct extent_buffer *eb, int slot,
1202 struct btrfs_key *key)
1203{
Geyslan G. Bem03b2f082013-10-11 15:35:45 -03001204 struct inode *dir = NULL;
1205 struct inode *inode = NULL;
Chris Masone02119d2008-09-05 16:13:11 -04001206 unsigned long ref_ptr;
1207 unsigned long ref_end;
Geyslan G. Bem03b2f082013-10-11 15:35:45 -03001208 char *name = NULL;
liubo34f3e4f2011-08-06 08:35:23 +00001209 int namelen;
1210 int ret;
liuboc622ae62011-03-26 08:01:12 -04001211 int search_done = 0;
Mark Fashehf1863732012-08-08 11:32:27 -07001212 int log_ref_ver = 0;
1213 u64 parent_objectid;
1214 u64 inode_objectid;
Chris Masonf46dbe3de2012-10-09 11:17:20 -04001215 u64 ref_index = 0;
Mark Fashehf1863732012-08-08 11:32:27 -07001216 int ref_struct_size;
1217
1218 ref_ptr = btrfs_item_ptr_offset(eb, slot);
1219 ref_end = ref_ptr + btrfs_item_size_nr(eb, slot);
1220
1221 if (key->type == BTRFS_INODE_EXTREF_KEY) {
1222 struct btrfs_inode_extref *r;
1223
1224 ref_struct_size = sizeof(struct btrfs_inode_extref);
1225 log_ref_ver = 1;
1226 r = (struct btrfs_inode_extref *)ref_ptr;
1227 parent_objectid = btrfs_inode_extref_parent(eb, r);
1228 } else {
1229 ref_struct_size = sizeof(struct btrfs_inode_ref);
1230 parent_objectid = key->offset;
1231 }
1232 inode_objectid = key->objectid;
Chris Masone02119d2008-09-05 16:13:11 -04001233
Chris Masone02119d2008-09-05 16:13:11 -04001234 /*
1235 * it is possible that we didn't log all the parent directories
1236 * for a given inode. If we don't find the dir, just don't
1237 * copy the back ref in. The link count fixup code will take
1238 * care of the rest
1239 */
Mark Fashehf1863732012-08-08 11:32:27 -07001240 dir = read_one_inode(root, parent_objectid);
Geyslan G. Bem03b2f082013-10-11 15:35:45 -03001241 if (!dir) {
1242 ret = -ENOENT;
1243 goto out;
1244 }
Chris Masone02119d2008-09-05 16:13:11 -04001245
Mark Fashehf1863732012-08-08 11:32:27 -07001246 inode = read_one_inode(root, inode_objectid);
Tsutomu Itohc00e9492011-04-28 09:10:23 +00001247 if (!inode) {
Geyslan G. Bem03b2f082013-10-11 15:35:45 -03001248 ret = -EIO;
1249 goto out;
Tsutomu Itohc00e9492011-04-28 09:10:23 +00001250 }
Chris Masone02119d2008-09-05 16:13:11 -04001251
Jan Schmidt5a1d7842012-08-17 14:04:41 -07001252 while (ref_ptr < ref_end) {
Mark Fashehf1863732012-08-08 11:32:27 -07001253 if (log_ref_ver) {
1254 ret = extref_get_fields(eb, ref_ptr, &namelen, &name,
1255 &ref_index, &parent_objectid);
1256 /*
1257 * parent object can change from one array
1258 * item to another.
1259 */
1260 if (!dir)
1261 dir = read_one_inode(root, parent_objectid);
Geyslan G. Bem03b2f082013-10-11 15:35:45 -03001262 if (!dir) {
1263 ret = -ENOENT;
1264 goto out;
1265 }
Mark Fashehf1863732012-08-08 11:32:27 -07001266 } else {
1267 ret = ref_get_fields(eb, ref_ptr, &namelen, &name,
1268 &ref_index);
1269 }
1270 if (ret)
Geyslan G. Bem03b2f082013-10-11 15:35:45 -03001271 goto out;
Chris Masone02119d2008-09-05 16:13:11 -04001272
Jan Schmidt5a1d7842012-08-17 14:04:41 -07001273 /* if we already have a perfect match, we're done */
1274 if (!inode_in_dir(root, path, btrfs_ino(dir), btrfs_ino(inode),
Mark Fashehf1863732012-08-08 11:32:27 -07001275 ref_index, name, namelen)) {
Jan Schmidt5a1d7842012-08-17 14:04:41 -07001276 /*
1277 * look for a conflicting back reference in the
1278 * metadata. if we find one we have to unlink that name
1279 * of the file before we add our new link. Later on, we
1280 * overwrite any existing back reference, and we don't
1281 * want to create dangling pointers in the directory.
1282 */
Chris Masone02119d2008-09-05 16:13:11 -04001283
Jan Schmidt5a1d7842012-08-17 14:04:41 -07001284 if (!search_done) {
1285 ret = __add_inode_ref(trans, root, path, log,
Mark Fashehf1863732012-08-08 11:32:27 -07001286 dir, inode, eb,
1287 inode_objectid,
1288 parent_objectid,
1289 ref_index, name, namelen,
Jan Schmidt5a1d7842012-08-17 14:04:41 -07001290 &search_done);
Geyslan G. Bem03b2f082013-10-11 15:35:45 -03001291 if (ret) {
1292 if (ret == 1)
1293 ret = 0;
Jan Schmidt5a1d7842012-08-17 14:04:41 -07001294 goto out;
Josef Bacik36508602013-04-25 16:23:32 -04001295 }
Chris Masone02119d2008-09-05 16:13:11 -04001296 }
Jan Schmidt5a1d7842012-08-17 14:04:41 -07001297
1298 /* insert our name */
1299 ret = btrfs_add_link(trans, dir, inode, name, namelen,
Mark Fashehf1863732012-08-08 11:32:27 -07001300 0, ref_index);
Josef Bacik36508602013-04-25 16:23:32 -04001301 if (ret)
1302 goto out;
Jan Schmidt5a1d7842012-08-17 14:04:41 -07001303
1304 btrfs_update_inode(trans, root, inode);
Chris Masone02119d2008-09-05 16:13:11 -04001305 }
liuboc622ae62011-03-26 08:01:12 -04001306
Mark Fashehf1863732012-08-08 11:32:27 -07001307 ref_ptr = (unsigned long)(ref_ptr + ref_struct_size) + namelen;
Jan Schmidt5a1d7842012-08-17 14:04:41 -07001308 kfree(name);
Geyslan G. Bem03b2f082013-10-11 15:35:45 -03001309 name = NULL;
Mark Fashehf1863732012-08-08 11:32:27 -07001310 if (log_ref_ver) {
1311 iput(dir);
1312 dir = NULL;
1313 }
Chris Masone02119d2008-09-05 16:13:11 -04001314 }
Chris Masone02119d2008-09-05 16:13:11 -04001315
1316 /* finally write the back reference in the inode */
1317 ret = overwrite_item(trans, root, path, eb, slot, key);
Jan Schmidt5a1d7842012-08-17 14:04:41 -07001318out:
David Sterbab3b4aa72011-04-21 01:20:15 +02001319 btrfs_release_path(path);
Geyslan G. Bem03b2f082013-10-11 15:35:45 -03001320 kfree(name);
Chris Masone02119d2008-09-05 16:13:11 -04001321 iput(dir);
1322 iput(inode);
Josef Bacik36508602013-04-25 16:23:32 -04001323 return ret;
Chris Masone02119d2008-09-05 16:13:11 -04001324}
1325
Yan, Zhengc71bf092009-11-12 09:34:40 +00001326static int insert_orphan_item(struct btrfs_trans_handle *trans,
David Sterba9c4f61f2015-01-02 19:12:57 +01001327 struct btrfs_root *root, u64 ino)
Yan, Zhengc71bf092009-11-12 09:34:40 +00001328{
1329 int ret;
David Sterba381cf652015-01-02 18:45:16 +01001330
David Sterba9c4f61f2015-01-02 19:12:57 +01001331 ret = btrfs_insert_orphan_item(trans, root, ino);
1332 if (ret == -EEXIST)
1333 ret = 0;
David Sterba381cf652015-01-02 18:45:16 +01001334
Yan, Zhengc71bf092009-11-12 09:34:40 +00001335 return ret;
1336}
1337
Mark Fashehf1863732012-08-08 11:32:27 -07001338static int count_inode_extrefs(struct btrfs_root *root,
1339 struct inode *inode, struct btrfs_path *path)
Chris Masone02119d2008-09-05 16:13:11 -04001340{
Mark Fashehf1863732012-08-08 11:32:27 -07001341 int ret = 0;
1342 int name_len;
1343 unsigned int nlink = 0;
1344 u32 item_size;
1345 u32 cur_offset = 0;
1346 u64 inode_objectid = btrfs_ino(inode);
1347 u64 offset = 0;
1348 unsigned long ptr;
1349 struct btrfs_inode_extref *extref;
1350 struct extent_buffer *leaf;
1351
1352 while (1) {
1353 ret = btrfs_find_one_extref(root, inode_objectid, offset, path,
1354 &extref, &offset);
1355 if (ret)
1356 break;
1357
1358 leaf = path->nodes[0];
1359 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1360 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
Filipe Manana2c2c4522015-01-13 16:40:04 +00001361 cur_offset = 0;
Mark Fashehf1863732012-08-08 11:32:27 -07001362
1363 while (cur_offset < item_size) {
1364 extref = (struct btrfs_inode_extref *) (ptr + cur_offset);
1365 name_len = btrfs_inode_extref_name_len(leaf, extref);
1366
1367 nlink++;
1368
1369 cur_offset += name_len + sizeof(*extref);
1370 }
1371
1372 offset++;
1373 btrfs_release_path(path);
1374 }
1375 btrfs_release_path(path);
1376
Filipe Manana2c2c4522015-01-13 16:40:04 +00001377 if (ret < 0 && ret != -ENOENT)
Mark Fashehf1863732012-08-08 11:32:27 -07001378 return ret;
1379 return nlink;
1380}
1381
1382static int count_inode_refs(struct btrfs_root *root,
1383 struct inode *inode, struct btrfs_path *path)
1384{
Chris Masone02119d2008-09-05 16:13:11 -04001385 int ret;
1386 struct btrfs_key key;
Mark Fashehf1863732012-08-08 11:32:27 -07001387 unsigned int nlink = 0;
Chris Masone02119d2008-09-05 16:13:11 -04001388 unsigned long ptr;
1389 unsigned long ptr_end;
1390 int name_len;
Li Zefan33345d012011-04-20 10:31:50 +08001391 u64 ino = btrfs_ino(inode);
Chris Masone02119d2008-09-05 16:13:11 -04001392
Li Zefan33345d012011-04-20 10:31:50 +08001393 key.objectid = ino;
Chris Masone02119d2008-09-05 16:13:11 -04001394 key.type = BTRFS_INODE_REF_KEY;
1395 key.offset = (u64)-1;
1396
Chris Masond3977122009-01-05 21:25:51 -05001397 while (1) {
Chris Masone02119d2008-09-05 16:13:11 -04001398 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1399 if (ret < 0)
1400 break;
1401 if (ret > 0) {
1402 if (path->slots[0] == 0)
1403 break;
1404 path->slots[0]--;
1405 }
Filipe David Borba Mananae93ae262013-10-14 22:49:11 +01001406process_slot:
Chris Masone02119d2008-09-05 16:13:11 -04001407 btrfs_item_key_to_cpu(path->nodes[0], &key,
1408 path->slots[0]);
Li Zefan33345d012011-04-20 10:31:50 +08001409 if (key.objectid != ino ||
Chris Masone02119d2008-09-05 16:13:11 -04001410 key.type != BTRFS_INODE_REF_KEY)
1411 break;
1412 ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
1413 ptr_end = ptr + btrfs_item_size_nr(path->nodes[0],
1414 path->slots[0]);
Chris Masond3977122009-01-05 21:25:51 -05001415 while (ptr < ptr_end) {
Chris Masone02119d2008-09-05 16:13:11 -04001416 struct btrfs_inode_ref *ref;
1417
1418 ref = (struct btrfs_inode_ref *)ptr;
1419 name_len = btrfs_inode_ref_name_len(path->nodes[0],
1420 ref);
1421 ptr = (unsigned long)(ref + 1) + name_len;
1422 nlink++;
1423 }
1424
1425 if (key.offset == 0)
1426 break;
Filipe David Borba Mananae93ae262013-10-14 22:49:11 +01001427 if (path->slots[0] > 0) {
1428 path->slots[0]--;
1429 goto process_slot;
1430 }
Chris Masone02119d2008-09-05 16:13:11 -04001431 key.offset--;
David Sterbab3b4aa72011-04-21 01:20:15 +02001432 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04001433 }
David Sterbab3b4aa72011-04-21 01:20:15 +02001434 btrfs_release_path(path);
Mark Fashehf1863732012-08-08 11:32:27 -07001435
1436 return nlink;
1437}
1438
1439/*
1440 * There are a few corners where the link count of the file can't
1441 * be properly maintained during replay. So, instead of adding
1442 * lots of complexity to the log code, we just scan the backrefs
1443 * for any file that has been through replay.
1444 *
1445 * The scan will update the link count on the inode to reflect the
1446 * number of back refs found. If it goes down to zero, the iput
1447 * will free the inode.
1448 */
1449static noinline int fixup_inode_link_count(struct btrfs_trans_handle *trans,
1450 struct btrfs_root *root,
1451 struct inode *inode)
1452{
1453 struct btrfs_path *path;
1454 int ret;
1455 u64 nlink = 0;
1456 u64 ino = btrfs_ino(inode);
1457
1458 path = btrfs_alloc_path();
1459 if (!path)
1460 return -ENOMEM;
1461
1462 ret = count_inode_refs(root, inode, path);
1463 if (ret < 0)
1464 goto out;
1465
1466 nlink = ret;
1467
1468 ret = count_inode_extrefs(root, inode, path);
Mark Fashehf1863732012-08-08 11:32:27 -07001469 if (ret < 0)
1470 goto out;
1471
1472 nlink += ret;
1473
1474 ret = 0;
1475
Chris Masone02119d2008-09-05 16:13:11 -04001476 if (nlink != inode->i_nlink) {
Miklos Szeredibfe86842011-10-28 14:13:29 +02001477 set_nlink(inode, nlink);
Chris Masone02119d2008-09-05 16:13:11 -04001478 btrfs_update_inode(trans, root, inode);
1479 }
Chris Mason8d5bf1c2008-09-11 15:51:21 -04001480 BTRFS_I(inode)->index_cnt = (u64)-1;
Chris Masone02119d2008-09-05 16:13:11 -04001481
Yan, Zhengc71bf092009-11-12 09:34:40 +00001482 if (inode->i_nlink == 0) {
1483 if (S_ISDIR(inode->i_mode)) {
1484 ret = replay_dir_deletes(trans, root, NULL, path,
Li Zefan33345d012011-04-20 10:31:50 +08001485 ino, 1);
Josef Bacik36508602013-04-25 16:23:32 -04001486 if (ret)
1487 goto out;
Yan, Zhengc71bf092009-11-12 09:34:40 +00001488 }
Li Zefan33345d012011-04-20 10:31:50 +08001489 ret = insert_orphan_item(trans, root, ino);
Chris Mason12fcfd22009-03-24 10:24:20 -04001490 }
Chris Mason12fcfd22009-03-24 10:24:20 -04001491
Mark Fashehf1863732012-08-08 11:32:27 -07001492out:
1493 btrfs_free_path(path);
1494 return ret;
Chris Masone02119d2008-09-05 16:13:11 -04001495}
1496
1497static noinline int fixup_inode_link_counts(struct btrfs_trans_handle *trans,
1498 struct btrfs_root *root,
1499 struct btrfs_path *path)
1500{
1501 int ret;
1502 struct btrfs_key key;
1503 struct inode *inode;
1504
1505 key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID;
1506 key.type = BTRFS_ORPHAN_ITEM_KEY;
1507 key.offset = (u64)-1;
Chris Masond3977122009-01-05 21:25:51 -05001508 while (1) {
Chris Masone02119d2008-09-05 16:13:11 -04001509 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1510 if (ret < 0)
1511 break;
1512
1513 if (ret == 1) {
1514 if (path->slots[0] == 0)
1515 break;
1516 path->slots[0]--;
1517 }
1518
1519 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1520 if (key.objectid != BTRFS_TREE_LOG_FIXUP_OBJECTID ||
1521 key.type != BTRFS_ORPHAN_ITEM_KEY)
1522 break;
1523
1524 ret = btrfs_del_item(trans, root, path);
Tsutomu Itoh65a246c2011-05-19 04:37:44 +00001525 if (ret)
1526 goto out;
Chris Masone02119d2008-09-05 16:13:11 -04001527
David Sterbab3b4aa72011-04-21 01:20:15 +02001528 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04001529 inode = read_one_inode(root, key.offset);
Tsutomu Itohc00e9492011-04-28 09:10:23 +00001530 if (!inode)
1531 return -EIO;
Chris Masone02119d2008-09-05 16:13:11 -04001532
1533 ret = fixup_inode_link_count(trans, root, inode);
Chris Masone02119d2008-09-05 16:13:11 -04001534 iput(inode);
Josef Bacik36508602013-04-25 16:23:32 -04001535 if (ret)
1536 goto out;
Chris Masone02119d2008-09-05 16:13:11 -04001537
Chris Mason12fcfd22009-03-24 10:24:20 -04001538 /*
1539 * fixup on a directory may create new entries,
1540 * make sure we always look for the highset possible
1541 * offset
1542 */
1543 key.offset = (u64)-1;
Chris Masone02119d2008-09-05 16:13:11 -04001544 }
Tsutomu Itoh65a246c2011-05-19 04:37:44 +00001545 ret = 0;
1546out:
David Sterbab3b4aa72011-04-21 01:20:15 +02001547 btrfs_release_path(path);
Tsutomu Itoh65a246c2011-05-19 04:37:44 +00001548 return ret;
Chris Masone02119d2008-09-05 16:13:11 -04001549}
1550
1551
1552/*
1553 * record a given inode in the fixup dir so we can check its link
1554 * count when replay is done. The link count is incremented here
1555 * so the inode won't go away until we check it
1556 */
1557static noinline int link_to_fixup_dir(struct btrfs_trans_handle *trans,
1558 struct btrfs_root *root,
1559 struct btrfs_path *path,
1560 u64 objectid)
1561{
1562 struct btrfs_key key;
1563 int ret = 0;
1564 struct inode *inode;
1565
1566 inode = read_one_inode(root, objectid);
Tsutomu Itohc00e9492011-04-28 09:10:23 +00001567 if (!inode)
1568 return -EIO;
Chris Masone02119d2008-09-05 16:13:11 -04001569
1570 key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID;
David Sterba962a2982014-06-04 18:41:45 +02001571 key.type = BTRFS_ORPHAN_ITEM_KEY;
Chris Masone02119d2008-09-05 16:13:11 -04001572 key.offset = objectid;
1573
1574 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1575
David Sterbab3b4aa72011-04-21 01:20:15 +02001576 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04001577 if (ret == 0) {
Josef Bacik9bf7a482013-03-01 13:35:47 -05001578 if (!inode->i_nlink)
1579 set_nlink(inode, 1);
1580 else
Zach Brown8b558c52013-10-16 12:10:34 -07001581 inc_nlink(inode);
Tsutomu Itohb9959292012-06-25 21:25:22 -06001582 ret = btrfs_update_inode(trans, root, inode);
Chris Masone02119d2008-09-05 16:13:11 -04001583 } else if (ret == -EEXIST) {
1584 ret = 0;
1585 } else {
Josef Bacik36508602013-04-25 16:23:32 -04001586 BUG(); /* Logic Error */
Chris Masone02119d2008-09-05 16:13:11 -04001587 }
1588 iput(inode);
1589
1590 return ret;
1591}
1592
1593/*
1594 * when replaying the log for a directory, we only insert names
1595 * for inodes that actually exist. This means an fsync on a directory
1596 * does not implicitly fsync all the new files in it
1597 */
1598static noinline int insert_one_name(struct btrfs_trans_handle *trans,
1599 struct btrfs_root *root,
Chris Masone02119d2008-09-05 16:13:11 -04001600 u64 dirid, u64 index,
Zhaolei60d53eb2015-08-17 18:44:46 +08001601 char *name, int name_len,
Chris Masone02119d2008-09-05 16:13:11 -04001602 struct btrfs_key *location)
1603{
1604 struct inode *inode;
1605 struct inode *dir;
1606 int ret;
1607
1608 inode = read_one_inode(root, location->objectid);
1609 if (!inode)
1610 return -ENOENT;
1611
1612 dir = read_one_inode(root, dirid);
1613 if (!dir) {
1614 iput(inode);
1615 return -EIO;
1616 }
Josef Bacikd5554382013-09-11 14:17:00 -04001617
Chris Masone02119d2008-09-05 16:13:11 -04001618 ret = btrfs_add_link(trans, dir, inode, name, name_len, 1, index);
1619
1620 /* FIXME, put inode into FIXUP list */
1621
1622 iput(inode);
1623 iput(dir);
1624 return ret;
1625}
1626
1627/*
Filipe Mananadf8d1162015-01-14 01:52:25 +00001628 * Return true if an inode reference exists in the log for the given name,
1629 * inode and parent inode.
1630 */
1631static bool name_in_log_ref(struct btrfs_root *log_root,
1632 const char *name, const int name_len,
1633 const u64 dirid, const u64 ino)
1634{
1635 struct btrfs_key search_key;
1636
1637 search_key.objectid = ino;
1638 search_key.type = BTRFS_INODE_REF_KEY;
1639 search_key.offset = dirid;
1640 if (backref_in_log(log_root, &search_key, dirid, name, name_len))
1641 return true;
1642
1643 search_key.type = BTRFS_INODE_EXTREF_KEY;
1644 search_key.offset = btrfs_extref_hash(dirid, name, name_len);
1645 if (backref_in_log(log_root, &search_key, dirid, name, name_len))
1646 return true;
1647
1648 return false;
1649}
1650
1651/*
Chris Masone02119d2008-09-05 16:13:11 -04001652 * take a single entry in a log directory item and replay it into
1653 * the subvolume.
1654 *
1655 * if a conflicting item exists in the subdirectory already,
1656 * the inode it points to is unlinked and put into the link count
1657 * fix up tree.
1658 *
1659 * If a name from the log points to a file or directory that does
1660 * not exist in the FS, it is skipped. fsyncs on directories
1661 * do not force down inodes inside that directory, just changes to the
1662 * names or unlinks in a directory.
Filipe Mananabb53eda2015-07-15 23:26:43 +01001663 *
1664 * Returns < 0 on error, 0 if the name wasn't replayed (dentry points to a
1665 * non-existing inode) and 1 if the name was replayed.
Chris Masone02119d2008-09-05 16:13:11 -04001666 */
1667static noinline int replay_one_name(struct btrfs_trans_handle *trans,
1668 struct btrfs_root *root,
1669 struct btrfs_path *path,
1670 struct extent_buffer *eb,
1671 struct btrfs_dir_item *di,
1672 struct btrfs_key *key)
1673{
1674 char *name;
1675 int name_len;
1676 struct btrfs_dir_item *dst_di;
1677 struct btrfs_key found_key;
1678 struct btrfs_key log_key;
1679 struct inode *dir;
Chris Masone02119d2008-09-05 16:13:11 -04001680 u8 log_type;
Chris Mason4bef0842008-09-08 11:18:08 -04001681 int exists;
Josef Bacik36508602013-04-25 16:23:32 -04001682 int ret = 0;
Josef Bacikd5554382013-09-11 14:17:00 -04001683 bool update_size = (key->type == BTRFS_DIR_INDEX_KEY);
Filipe Mananabb53eda2015-07-15 23:26:43 +01001684 bool name_added = false;
Chris Masone02119d2008-09-05 16:13:11 -04001685
1686 dir = read_one_inode(root, key->objectid);
Tsutomu Itohc00e9492011-04-28 09:10:23 +00001687 if (!dir)
1688 return -EIO;
Chris Masone02119d2008-09-05 16:13:11 -04001689
1690 name_len = btrfs_dir_name_len(eb, di);
1691 name = kmalloc(name_len, GFP_NOFS);
Filipe David Borba Manana2bac3252013-08-04 19:58:57 +01001692 if (!name) {
1693 ret = -ENOMEM;
1694 goto out;
1695 }
liubo2a29edc2011-01-26 06:22:08 +00001696
Chris Masone02119d2008-09-05 16:13:11 -04001697 log_type = btrfs_dir_type(eb, di);
1698 read_extent_buffer(eb, name, (unsigned long)(di + 1),
1699 name_len);
1700
1701 btrfs_dir_item_key_to_cpu(eb, di, &log_key);
Chris Mason4bef0842008-09-08 11:18:08 -04001702 exists = btrfs_lookup_inode(trans, root, path, &log_key, 0);
1703 if (exists == 0)
1704 exists = 1;
1705 else
1706 exists = 0;
David Sterbab3b4aa72011-04-21 01:20:15 +02001707 btrfs_release_path(path);
Chris Mason4bef0842008-09-08 11:18:08 -04001708
Chris Masone02119d2008-09-05 16:13:11 -04001709 if (key->type == BTRFS_DIR_ITEM_KEY) {
1710 dst_di = btrfs_lookup_dir_item(trans, root, path, key->objectid,
1711 name, name_len, 1);
Chris Masond3977122009-01-05 21:25:51 -05001712 } else if (key->type == BTRFS_DIR_INDEX_KEY) {
Chris Masone02119d2008-09-05 16:13:11 -04001713 dst_di = btrfs_lookup_dir_index_item(trans, root, path,
1714 key->objectid,
1715 key->offset, name,
1716 name_len, 1);
1717 } else {
Josef Bacik36508602013-04-25 16:23:32 -04001718 /* Corruption */
1719 ret = -EINVAL;
1720 goto out;
Chris Masone02119d2008-09-05 16:13:11 -04001721 }
David Sterbac7040052011-04-19 18:00:01 +02001722 if (IS_ERR_OR_NULL(dst_di)) {
Chris Masone02119d2008-09-05 16:13:11 -04001723 /* we need a sequence number to insert, so we only
1724 * do inserts for the BTRFS_DIR_INDEX_KEY types
1725 */
1726 if (key->type != BTRFS_DIR_INDEX_KEY)
1727 goto out;
1728 goto insert;
1729 }
1730
1731 btrfs_dir_item_key_to_cpu(path->nodes[0], dst_di, &found_key);
1732 /* the existing item matches the logged item */
1733 if (found_key.objectid == log_key.objectid &&
1734 found_key.type == log_key.type &&
1735 found_key.offset == log_key.offset &&
1736 btrfs_dir_type(path->nodes[0], dst_di) == log_type) {
Filipe Mananaa2cc11d2014-09-08 22:53:18 +01001737 update_size = false;
Chris Masone02119d2008-09-05 16:13:11 -04001738 goto out;
1739 }
1740
1741 /*
1742 * don't drop the conflicting directory entry if the inode
1743 * for the new entry doesn't exist
1744 */
Chris Mason4bef0842008-09-08 11:18:08 -04001745 if (!exists)
Chris Masone02119d2008-09-05 16:13:11 -04001746 goto out;
1747
Chris Masone02119d2008-09-05 16:13:11 -04001748 ret = drop_one_dir_item(trans, root, path, dir, dst_di);
Josef Bacik36508602013-04-25 16:23:32 -04001749 if (ret)
1750 goto out;
Chris Masone02119d2008-09-05 16:13:11 -04001751
1752 if (key->type == BTRFS_DIR_INDEX_KEY)
1753 goto insert;
1754out:
David Sterbab3b4aa72011-04-21 01:20:15 +02001755 btrfs_release_path(path);
Josef Bacikd5554382013-09-11 14:17:00 -04001756 if (!ret && update_size) {
1757 btrfs_i_size_write(dir, dir->i_size + name_len * 2);
1758 ret = btrfs_update_inode(trans, root, dir);
1759 }
Chris Masone02119d2008-09-05 16:13:11 -04001760 kfree(name);
1761 iput(dir);
Filipe Mananabb53eda2015-07-15 23:26:43 +01001762 if (!ret && name_added)
1763 ret = 1;
Josef Bacik36508602013-04-25 16:23:32 -04001764 return ret;
Chris Masone02119d2008-09-05 16:13:11 -04001765
1766insert:
Filipe Mananadf8d1162015-01-14 01:52:25 +00001767 if (name_in_log_ref(root->log_root, name, name_len,
1768 key->objectid, log_key.objectid)) {
1769 /* The dentry will be added later. */
1770 ret = 0;
1771 update_size = false;
1772 goto out;
1773 }
David Sterbab3b4aa72011-04-21 01:20:15 +02001774 btrfs_release_path(path);
Zhaolei60d53eb2015-08-17 18:44:46 +08001775 ret = insert_one_name(trans, root, key->objectid, key->offset,
1776 name, name_len, &log_key);
Filipe Mananadf8d1162015-01-14 01:52:25 +00001777 if (ret && ret != -ENOENT && ret != -EEXIST)
Josef Bacik36508602013-04-25 16:23:32 -04001778 goto out;
Filipe Mananabb53eda2015-07-15 23:26:43 +01001779 if (!ret)
1780 name_added = true;
Josef Bacikd5554382013-09-11 14:17:00 -04001781 update_size = false;
Josef Bacik36508602013-04-25 16:23:32 -04001782 ret = 0;
Chris Masone02119d2008-09-05 16:13:11 -04001783 goto out;
1784}
1785
1786/*
1787 * find all the names in a directory item and reconcile them into
1788 * the subvolume. Only BTRFS_DIR_ITEM_KEY types will have more than
1789 * one name in a directory item, but the same code gets used for
1790 * both directory index types
1791 */
1792static noinline int replay_one_dir_item(struct btrfs_trans_handle *trans,
1793 struct btrfs_root *root,
1794 struct btrfs_path *path,
1795 struct extent_buffer *eb, int slot,
1796 struct btrfs_key *key)
1797{
Filipe Mananabb53eda2015-07-15 23:26:43 +01001798 int ret = 0;
Chris Masone02119d2008-09-05 16:13:11 -04001799 u32 item_size = btrfs_item_size_nr(eb, slot);
1800 struct btrfs_dir_item *di;
1801 int name_len;
1802 unsigned long ptr;
1803 unsigned long ptr_end;
Filipe Mananabb53eda2015-07-15 23:26:43 +01001804 struct btrfs_path *fixup_path = NULL;
Chris Masone02119d2008-09-05 16:13:11 -04001805
1806 ptr = btrfs_item_ptr_offset(eb, slot);
1807 ptr_end = ptr + item_size;
Chris Masond3977122009-01-05 21:25:51 -05001808 while (ptr < ptr_end) {
Chris Masone02119d2008-09-05 16:13:11 -04001809 di = (struct btrfs_dir_item *)ptr;
Josef Bacik22a94d42011-03-16 16:47:17 -04001810 if (verify_dir_item(root, eb, di))
1811 return -EIO;
Chris Masone02119d2008-09-05 16:13:11 -04001812 name_len = btrfs_dir_name_len(eb, di);
1813 ret = replay_one_name(trans, root, path, eb, di, key);
Filipe Mananabb53eda2015-07-15 23:26:43 +01001814 if (ret < 0)
1815 break;
Chris Masone02119d2008-09-05 16:13:11 -04001816 ptr = (unsigned long)(di + 1);
1817 ptr += name_len;
Filipe Mananabb53eda2015-07-15 23:26:43 +01001818
1819 /*
1820 * If this entry refers to a non-directory (directories can not
1821 * have a link count > 1) and it was added in the transaction
1822 * that was not committed, make sure we fixup the link count of
1823 * the inode it the entry points to. Otherwise something like
1824 * the following would result in a directory pointing to an
1825 * inode with a wrong link that does not account for this dir
1826 * entry:
1827 *
1828 * mkdir testdir
1829 * touch testdir/foo
1830 * touch testdir/bar
1831 * sync
1832 *
1833 * ln testdir/bar testdir/bar_link
1834 * ln testdir/foo testdir/foo_link
1835 * xfs_io -c "fsync" testdir/bar
1836 *
1837 * <power failure>
1838 *
1839 * mount fs, log replay happens
1840 *
1841 * File foo would remain with a link count of 1 when it has two
1842 * entries pointing to it in the directory testdir. This would
1843 * make it impossible to ever delete the parent directory has
1844 * it would result in stale dentries that can never be deleted.
1845 */
1846 if (ret == 1 && btrfs_dir_type(eb, di) != BTRFS_FT_DIR) {
1847 struct btrfs_key di_key;
1848
1849 if (!fixup_path) {
1850 fixup_path = btrfs_alloc_path();
1851 if (!fixup_path) {
1852 ret = -ENOMEM;
1853 break;
1854 }
1855 }
1856
1857 btrfs_dir_item_key_to_cpu(eb, di, &di_key);
1858 ret = link_to_fixup_dir(trans, root, fixup_path,
1859 di_key.objectid);
1860 if (ret)
1861 break;
1862 }
1863 ret = 0;
Chris Masone02119d2008-09-05 16:13:11 -04001864 }
Filipe Mananabb53eda2015-07-15 23:26:43 +01001865 btrfs_free_path(fixup_path);
1866 return ret;
Chris Masone02119d2008-09-05 16:13:11 -04001867}
1868
1869/*
1870 * directory replay has two parts. There are the standard directory
1871 * items in the log copied from the subvolume, and range items
1872 * created in the log while the subvolume was logged.
1873 *
1874 * The range items tell us which parts of the key space the log
1875 * is authoritative for. During replay, if a key in the subvolume
1876 * directory is in a logged range item, but not actually in the log
1877 * that means it was deleted from the directory before the fsync
1878 * and should be removed.
1879 */
1880static noinline int find_dir_range(struct btrfs_root *root,
1881 struct btrfs_path *path,
1882 u64 dirid, int key_type,
1883 u64 *start_ret, u64 *end_ret)
1884{
1885 struct btrfs_key key;
1886 u64 found_end;
1887 struct btrfs_dir_log_item *item;
1888 int ret;
1889 int nritems;
1890
1891 if (*start_ret == (u64)-1)
1892 return 1;
1893
1894 key.objectid = dirid;
1895 key.type = key_type;
1896 key.offset = *start_ret;
1897
1898 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1899 if (ret < 0)
1900 goto out;
1901 if (ret > 0) {
1902 if (path->slots[0] == 0)
1903 goto out;
1904 path->slots[0]--;
1905 }
1906 if (ret != 0)
1907 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1908
1909 if (key.type != key_type || key.objectid != dirid) {
1910 ret = 1;
1911 goto next;
1912 }
1913 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
1914 struct btrfs_dir_log_item);
1915 found_end = btrfs_dir_log_end(path->nodes[0], item);
1916
1917 if (*start_ret >= key.offset && *start_ret <= found_end) {
1918 ret = 0;
1919 *start_ret = key.offset;
1920 *end_ret = found_end;
1921 goto out;
1922 }
1923 ret = 1;
1924next:
1925 /* check the next slot in the tree to see if it is a valid item */
1926 nritems = btrfs_header_nritems(path->nodes[0]);
1927 if (path->slots[0] >= nritems) {
1928 ret = btrfs_next_leaf(root, path);
1929 if (ret)
1930 goto out;
1931 } else {
1932 path->slots[0]++;
1933 }
1934
1935 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1936
1937 if (key.type != key_type || key.objectid != dirid) {
1938 ret = 1;
1939 goto out;
1940 }
1941 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
1942 struct btrfs_dir_log_item);
1943 found_end = btrfs_dir_log_end(path->nodes[0], item);
1944 *start_ret = key.offset;
1945 *end_ret = found_end;
1946 ret = 0;
1947out:
David Sterbab3b4aa72011-04-21 01:20:15 +02001948 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04001949 return ret;
1950}
1951
1952/*
1953 * this looks for a given directory item in the log. If the directory
1954 * item is not in the log, the item is removed and the inode it points
1955 * to is unlinked
1956 */
1957static noinline int check_item_in_log(struct btrfs_trans_handle *trans,
1958 struct btrfs_root *root,
1959 struct btrfs_root *log,
1960 struct btrfs_path *path,
1961 struct btrfs_path *log_path,
1962 struct inode *dir,
1963 struct btrfs_key *dir_key)
1964{
1965 int ret;
1966 struct extent_buffer *eb;
1967 int slot;
1968 u32 item_size;
1969 struct btrfs_dir_item *di;
1970 struct btrfs_dir_item *log_di;
1971 int name_len;
1972 unsigned long ptr;
1973 unsigned long ptr_end;
1974 char *name;
1975 struct inode *inode;
1976 struct btrfs_key location;
1977
1978again:
1979 eb = path->nodes[0];
1980 slot = path->slots[0];
1981 item_size = btrfs_item_size_nr(eb, slot);
1982 ptr = btrfs_item_ptr_offset(eb, slot);
1983 ptr_end = ptr + item_size;
Chris Masond3977122009-01-05 21:25:51 -05001984 while (ptr < ptr_end) {
Chris Masone02119d2008-09-05 16:13:11 -04001985 di = (struct btrfs_dir_item *)ptr;
Josef Bacik22a94d42011-03-16 16:47:17 -04001986 if (verify_dir_item(root, eb, di)) {
1987 ret = -EIO;
1988 goto out;
1989 }
1990
Chris Masone02119d2008-09-05 16:13:11 -04001991 name_len = btrfs_dir_name_len(eb, di);
1992 name = kmalloc(name_len, GFP_NOFS);
1993 if (!name) {
1994 ret = -ENOMEM;
1995 goto out;
1996 }
1997 read_extent_buffer(eb, name, (unsigned long)(di + 1),
1998 name_len);
1999 log_di = NULL;
Chris Mason12fcfd22009-03-24 10:24:20 -04002000 if (log && dir_key->type == BTRFS_DIR_ITEM_KEY) {
Chris Masone02119d2008-09-05 16:13:11 -04002001 log_di = btrfs_lookup_dir_item(trans, log, log_path,
2002 dir_key->objectid,
2003 name, name_len, 0);
Chris Mason12fcfd22009-03-24 10:24:20 -04002004 } else if (log && dir_key->type == BTRFS_DIR_INDEX_KEY) {
Chris Masone02119d2008-09-05 16:13:11 -04002005 log_di = btrfs_lookup_dir_index_item(trans, log,
2006 log_path,
2007 dir_key->objectid,
2008 dir_key->offset,
2009 name, name_len, 0);
2010 }
Filipe David Borba Manana269d0402013-10-28 17:39:21 +00002011 if (!log_di || (IS_ERR(log_di) && PTR_ERR(log_di) == -ENOENT)) {
Chris Masone02119d2008-09-05 16:13:11 -04002012 btrfs_dir_item_key_to_cpu(eb, di, &location);
David Sterbab3b4aa72011-04-21 01:20:15 +02002013 btrfs_release_path(path);
2014 btrfs_release_path(log_path);
Chris Masone02119d2008-09-05 16:13:11 -04002015 inode = read_one_inode(root, location.objectid);
Tsutomu Itohc00e9492011-04-28 09:10:23 +00002016 if (!inode) {
2017 kfree(name);
2018 return -EIO;
2019 }
Chris Masone02119d2008-09-05 16:13:11 -04002020
2021 ret = link_to_fixup_dir(trans, root,
2022 path, location.objectid);
Josef Bacik36508602013-04-25 16:23:32 -04002023 if (ret) {
2024 kfree(name);
2025 iput(inode);
2026 goto out;
2027 }
2028
Zach Brown8b558c52013-10-16 12:10:34 -07002029 inc_nlink(inode);
Chris Masone02119d2008-09-05 16:13:11 -04002030 ret = btrfs_unlink_inode(trans, root, dir, inode,
2031 name, name_len);
Josef Bacik36508602013-04-25 16:23:32 -04002032 if (!ret)
Filipe David Borba Mananaada9af22013-08-05 09:25:47 +01002033 ret = btrfs_run_delayed_items(trans, root);
Chris Masone02119d2008-09-05 16:13:11 -04002034 kfree(name);
2035 iput(inode);
Josef Bacik36508602013-04-25 16:23:32 -04002036 if (ret)
2037 goto out;
Chris Masone02119d2008-09-05 16:13:11 -04002038
2039 /* there might still be more names under this key
2040 * check and repeat if required
2041 */
2042 ret = btrfs_search_slot(NULL, root, dir_key, path,
2043 0, 0);
2044 if (ret == 0)
2045 goto again;
2046 ret = 0;
2047 goto out;
Filipe David Borba Manana269d0402013-10-28 17:39:21 +00002048 } else if (IS_ERR(log_di)) {
2049 kfree(name);
2050 return PTR_ERR(log_di);
Chris Masone02119d2008-09-05 16:13:11 -04002051 }
David Sterbab3b4aa72011-04-21 01:20:15 +02002052 btrfs_release_path(log_path);
Chris Masone02119d2008-09-05 16:13:11 -04002053 kfree(name);
2054
2055 ptr = (unsigned long)(di + 1);
2056 ptr += name_len;
2057 }
2058 ret = 0;
2059out:
David Sterbab3b4aa72011-04-21 01:20:15 +02002060 btrfs_release_path(path);
2061 btrfs_release_path(log_path);
Chris Masone02119d2008-09-05 16:13:11 -04002062 return ret;
2063}
2064
Filipe Manana4f764e52015-02-23 19:53:35 +00002065static int replay_xattr_deletes(struct btrfs_trans_handle *trans,
2066 struct btrfs_root *root,
2067 struct btrfs_root *log,
2068 struct btrfs_path *path,
2069 const u64 ino)
2070{
2071 struct btrfs_key search_key;
2072 struct btrfs_path *log_path;
2073 int i;
2074 int nritems;
2075 int ret;
2076
2077 log_path = btrfs_alloc_path();
2078 if (!log_path)
2079 return -ENOMEM;
2080
2081 search_key.objectid = ino;
2082 search_key.type = BTRFS_XATTR_ITEM_KEY;
2083 search_key.offset = 0;
2084again:
2085 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
2086 if (ret < 0)
2087 goto out;
2088process_leaf:
2089 nritems = btrfs_header_nritems(path->nodes[0]);
2090 for (i = path->slots[0]; i < nritems; i++) {
2091 struct btrfs_key key;
2092 struct btrfs_dir_item *di;
2093 struct btrfs_dir_item *log_di;
2094 u32 total_size;
2095 u32 cur;
2096
2097 btrfs_item_key_to_cpu(path->nodes[0], &key, i);
2098 if (key.objectid != ino || key.type != BTRFS_XATTR_ITEM_KEY) {
2099 ret = 0;
2100 goto out;
2101 }
2102
2103 di = btrfs_item_ptr(path->nodes[0], i, struct btrfs_dir_item);
2104 total_size = btrfs_item_size_nr(path->nodes[0], i);
2105 cur = 0;
2106 while (cur < total_size) {
2107 u16 name_len = btrfs_dir_name_len(path->nodes[0], di);
2108 u16 data_len = btrfs_dir_data_len(path->nodes[0], di);
2109 u32 this_len = sizeof(*di) + name_len + data_len;
2110 char *name;
2111
2112 name = kmalloc(name_len, GFP_NOFS);
2113 if (!name) {
2114 ret = -ENOMEM;
2115 goto out;
2116 }
2117 read_extent_buffer(path->nodes[0], name,
2118 (unsigned long)(di + 1), name_len);
2119
2120 log_di = btrfs_lookup_xattr(NULL, log, log_path, ino,
2121 name, name_len, 0);
2122 btrfs_release_path(log_path);
2123 if (!log_di) {
2124 /* Doesn't exist in log tree, so delete it. */
2125 btrfs_release_path(path);
2126 di = btrfs_lookup_xattr(trans, root, path, ino,
2127 name, name_len, -1);
2128 kfree(name);
2129 if (IS_ERR(di)) {
2130 ret = PTR_ERR(di);
2131 goto out;
2132 }
2133 ASSERT(di);
2134 ret = btrfs_delete_one_dir_name(trans, root,
2135 path, di);
2136 if (ret)
2137 goto out;
2138 btrfs_release_path(path);
2139 search_key = key;
2140 goto again;
2141 }
2142 kfree(name);
2143 if (IS_ERR(log_di)) {
2144 ret = PTR_ERR(log_di);
2145 goto out;
2146 }
2147 cur += this_len;
2148 di = (struct btrfs_dir_item *)((char *)di + this_len);
2149 }
2150 }
2151 ret = btrfs_next_leaf(root, path);
2152 if (ret > 0)
2153 ret = 0;
2154 else if (ret == 0)
2155 goto process_leaf;
2156out:
2157 btrfs_free_path(log_path);
2158 btrfs_release_path(path);
2159 return ret;
2160}
2161
2162
Chris Masone02119d2008-09-05 16:13:11 -04002163/*
2164 * deletion replay happens before we copy any new directory items
2165 * out of the log or out of backreferences from inodes. It
2166 * scans the log to find ranges of keys that log is authoritative for,
2167 * and then scans the directory to find items in those ranges that are
2168 * not present in the log.
2169 *
2170 * Anything we don't find in the log is unlinked and removed from the
2171 * directory.
2172 */
2173static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans,
2174 struct btrfs_root *root,
2175 struct btrfs_root *log,
2176 struct btrfs_path *path,
Chris Mason12fcfd22009-03-24 10:24:20 -04002177 u64 dirid, int del_all)
Chris Masone02119d2008-09-05 16:13:11 -04002178{
2179 u64 range_start;
2180 u64 range_end;
2181 int key_type = BTRFS_DIR_LOG_ITEM_KEY;
2182 int ret = 0;
2183 struct btrfs_key dir_key;
2184 struct btrfs_key found_key;
2185 struct btrfs_path *log_path;
2186 struct inode *dir;
2187
2188 dir_key.objectid = dirid;
2189 dir_key.type = BTRFS_DIR_ITEM_KEY;
2190 log_path = btrfs_alloc_path();
2191 if (!log_path)
2192 return -ENOMEM;
2193
2194 dir = read_one_inode(root, dirid);
2195 /* it isn't an error if the inode isn't there, that can happen
2196 * because we replay the deletes before we copy in the inode item
2197 * from the log
2198 */
2199 if (!dir) {
2200 btrfs_free_path(log_path);
2201 return 0;
2202 }
2203again:
2204 range_start = 0;
2205 range_end = 0;
Chris Masond3977122009-01-05 21:25:51 -05002206 while (1) {
Chris Mason12fcfd22009-03-24 10:24:20 -04002207 if (del_all)
2208 range_end = (u64)-1;
2209 else {
2210 ret = find_dir_range(log, path, dirid, key_type,
2211 &range_start, &range_end);
2212 if (ret != 0)
2213 break;
2214 }
Chris Masone02119d2008-09-05 16:13:11 -04002215
2216 dir_key.offset = range_start;
Chris Masond3977122009-01-05 21:25:51 -05002217 while (1) {
Chris Masone02119d2008-09-05 16:13:11 -04002218 int nritems;
2219 ret = btrfs_search_slot(NULL, root, &dir_key, path,
2220 0, 0);
2221 if (ret < 0)
2222 goto out;
2223
2224 nritems = btrfs_header_nritems(path->nodes[0]);
2225 if (path->slots[0] >= nritems) {
2226 ret = btrfs_next_leaf(root, path);
2227 if (ret)
2228 break;
2229 }
2230 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2231 path->slots[0]);
2232 if (found_key.objectid != dirid ||
2233 found_key.type != dir_key.type)
2234 goto next_type;
2235
2236 if (found_key.offset > range_end)
2237 break;
2238
2239 ret = check_item_in_log(trans, root, log, path,
Chris Mason12fcfd22009-03-24 10:24:20 -04002240 log_path, dir,
2241 &found_key);
Josef Bacik36508602013-04-25 16:23:32 -04002242 if (ret)
2243 goto out;
Chris Masone02119d2008-09-05 16:13:11 -04002244 if (found_key.offset == (u64)-1)
2245 break;
2246 dir_key.offset = found_key.offset + 1;
2247 }
David Sterbab3b4aa72011-04-21 01:20:15 +02002248 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04002249 if (range_end == (u64)-1)
2250 break;
2251 range_start = range_end + 1;
2252 }
2253
2254next_type:
2255 ret = 0;
2256 if (key_type == BTRFS_DIR_LOG_ITEM_KEY) {
2257 key_type = BTRFS_DIR_LOG_INDEX_KEY;
2258 dir_key.type = BTRFS_DIR_INDEX_KEY;
David Sterbab3b4aa72011-04-21 01:20:15 +02002259 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04002260 goto again;
2261 }
2262out:
David Sterbab3b4aa72011-04-21 01:20:15 +02002263 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04002264 btrfs_free_path(log_path);
2265 iput(dir);
2266 return ret;
2267}
2268
2269/*
2270 * the process_func used to replay items from the log tree. This
2271 * gets called in two different stages. The first stage just looks
2272 * for inodes and makes sure they are all copied into the subvolume.
2273 *
2274 * The second stage copies all the other item types from the log into
2275 * the subvolume. The two stage approach is slower, but gets rid of
2276 * lots of complexity around inodes referencing other inodes that exist
2277 * only in the log (references come from either directory items or inode
2278 * back refs).
2279 */
2280static int replay_one_buffer(struct btrfs_root *log, struct extent_buffer *eb,
2281 struct walk_control *wc, u64 gen)
2282{
2283 int nritems;
2284 struct btrfs_path *path;
2285 struct btrfs_root *root = wc->replay_dest;
2286 struct btrfs_key key;
Chris Masone02119d2008-09-05 16:13:11 -04002287 int level;
2288 int i;
2289 int ret;
2290
Tsutomu Itoh018642a2012-05-29 18:10:13 +09002291 ret = btrfs_read_buffer(eb, gen);
2292 if (ret)
2293 return ret;
Chris Masone02119d2008-09-05 16:13:11 -04002294
2295 level = btrfs_header_level(eb);
2296
2297 if (level != 0)
2298 return 0;
2299
2300 path = btrfs_alloc_path();
Mark Fasheh1e5063d2011-07-12 10:46:06 -07002301 if (!path)
2302 return -ENOMEM;
Chris Masone02119d2008-09-05 16:13:11 -04002303
2304 nritems = btrfs_header_nritems(eb);
2305 for (i = 0; i < nritems; i++) {
2306 btrfs_item_key_to_cpu(eb, &key, i);
Chris Masone02119d2008-09-05 16:13:11 -04002307
2308 /* inode keys are done during the first stage */
2309 if (key.type == BTRFS_INODE_ITEM_KEY &&
2310 wc->stage == LOG_WALK_REPLAY_INODES) {
Chris Masone02119d2008-09-05 16:13:11 -04002311 struct btrfs_inode_item *inode_item;
2312 u32 mode;
2313
2314 inode_item = btrfs_item_ptr(eb, i,
2315 struct btrfs_inode_item);
Filipe Manana4f764e52015-02-23 19:53:35 +00002316 ret = replay_xattr_deletes(wc->trans, root, log,
2317 path, key.objectid);
2318 if (ret)
2319 break;
Chris Masone02119d2008-09-05 16:13:11 -04002320 mode = btrfs_inode_mode(eb, inode_item);
2321 if (S_ISDIR(mode)) {
2322 ret = replay_dir_deletes(wc->trans,
Chris Mason12fcfd22009-03-24 10:24:20 -04002323 root, log, path, key.objectid, 0);
Josef Bacikb50c6e22013-04-25 15:55:30 -04002324 if (ret)
2325 break;
Chris Masone02119d2008-09-05 16:13:11 -04002326 }
2327 ret = overwrite_item(wc->trans, root, path,
2328 eb, i, &key);
Josef Bacikb50c6e22013-04-25 15:55:30 -04002329 if (ret)
2330 break;
Chris Masone02119d2008-09-05 16:13:11 -04002331
Yan, Zhengc71bf092009-11-12 09:34:40 +00002332 /* for regular files, make sure corresponding
Nicholas D Steeves01327612016-05-19 21:18:45 -04002333 * orphan item exist. extents past the new EOF
Yan, Zhengc71bf092009-11-12 09:34:40 +00002334 * will be truncated later by orphan cleanup.
Chris Masone02119d2008-09-05 16:13:11 -04002335 */
2336 if (S_ISREG(mode)) {
Yan, Zhengc71bf092009-11-12 09:34:40 +00002337 ret = insert_orphan_item(wc->trans, root,
2338 key.objectid);
Josef Bacikb50c6e22013-04-25 15:55:30 -04002339 if (ret)
2340 break;
Chris Masone02119d2008-09-05 16:13:11 -04002341 }
Yan, Zhengc71bf092009-11-12 09:34:40 +00002342
Chris Masone02119d2008-09-05 16:13:11 -04002343 ret = link_to_fixup_dir(wc->trans, root,
2344 path, key.objectid);
Josef Bacikb50c6e22013-04-25 15:55:30 -04002345 if (ret)
2346 break;
Chris Masone02119d2008-09-05 16:13:11 -04002347 }
Josef Bacikdd8e7212013-09-11 11:57:23 -04002348
2349 if (key.type == BTRFS_DIR_INDEX_KEY &&
2350 wc->stage == LOG_WALK_REPLAY_DIR_INDEX) {
2351 ret = replay_one_dir_item(wc->trans, root, path,
2352 eb, i, &key);
2353 if (ret)
2354 break;
2355 }
2356
Chris Masone02119d2008-09-05 16:13:11 -04002357 if (wc->stage < LOG_WALK_REPLAY_ALL)
2358 continue;
2359
2360 /* these keys are simply copied */
2361 if (key.type == BTRFS_XATTR_ITEM_KEY) {
2362 ret = overwrite_item(wc->trans, root, path,
2363 eb, i, &key);
Josef Bacikb50c6e22013-04-25 15:55:30 -04002364 if (ret)
2365 break;
Liu Bo2da1c662013-05-26 13:50:29 +00002366 } else if (key.type == BTRFS_INODE_REF_KEY ||
2367 key.type == BTRFS_INODE_EXTREF_KEY) {
Mark Fashehf1863732012-08-08 11:32:27 -07002368 ret = add_inode_ref(wc->trans, root, log, path,
2369 eb, i, &key);
Josef Bacikb50c6e22013-04-25 15:55:30 -04002370 if (ret && ret != -ENOENT)
2371 break;
2372 ret = 0;
Chris Masone02119d2008-09-05 16:13:11 -04002373 } else if (key.type == BTRFS_EXTENT_DATA_KEY) {
2374 ret = replay_one_extent(wc->trans, root, path,
2375 eb, i, &key);
Josef Bacikb50c6e22013-04-25 15:55:30 -04002376 if (ret)
2377 break;
Josef Bacikdd8e7212013-09-11 11:57:23 -04002378 } else if (key.type == BTRFS_DIR_ITEM_KEY) {
Chris Masone02119d2008-09-05 16:13:11 -04002379 ret = replay_one_dir_item(wc->trans, root, path,
2380 eb, i, &key);
Josef Bacikb50c6e22013-04-25 15:55:30 -04002381 if (ret)
2382 break;
Chris Masone02119d2008-09-05 16:13:11 -04002383 }
2384 }
2385 btrfs_free_path(path);
Josef Bacikb50c6e22013-04-25 15:55:30 -04002386 return ret;
Chris Masone02119d2008-09-05 16:13:11 -04002387}
2388
Chris Masond3977122009-01-05 21:25:51 -05002389static noinline int walk_down_log_tree(struct btrfs_trans_handle *trans,
Chris Masone02119d2008-09-05 16:13:11 -04002390 struct btrfs_root *root,
2391 struct btrfs_path *path, int *level,
2392 struct walk_control *wc)
2393{
2394 u64 root_owner;
Chris Masone02119d2008-09-05 16:13:11 -04002395 u64 bytenr;
2396 u64 ptr_gen;
2397 struct extent_buffer *next;
2398 struct extent_buffer *cur;
2399 struct extent_buffer *parent;
2400 u32 blocksize;
2401 int ret = 0;
2402
2403 WARN_ON(*level < 0);
2404 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2405
Chris Masond3977122009-01-05 21:25:51 -05002406 while (*level > 0) {
Chris Masone02119d2008-09-05 16:13:11 -04002407 WARN_ON(*level < 0);
2408 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2409 cur = path->nodes[*level];
2410
Dulshani Gunawardhanafae7f212013-10-31 10:30:08 +05302411 WARN_ON(btrfs_header_level(cur) != *level);
Chris Masone02119d2008-09-05 16:13:11 -04002412
2413 if (path->slots[*level] >=
2414 btrfs_header_nritems(cur))
2415 break;
2416
2417 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
2418 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
David Sterba707e8a02014-06-04 19:22:26 +02002419 blocksize = root->nodesize;
Chris Masone02119d2008-09-05 16:13:11 -04002420
2421 parent = path->nodes[*level];
2422 root_owner = btrfs_header_owner(parent);
Chris Masone02119d2008-09-05 16:13:11 -04002423
David Sterbaa83fffb2014-06-15 02:39:54 +02002424 next = btrfs_find_create_tree_block(root, bytenr);
Liu Boc871b0f2016-06-06 12:01:23 -07002425 if (IS_ERR(next))
2426 return PTR_ERR(next);
Chris Masone02119d2008-09-05 16:13:11 -04002427
Chris Masone02119d2008-09-05 16:13:11 -04002428 if (*level == 1) {
Mark Fasheh1e5063d2011-07-12 10:46:06 -07002429 ret = wc->process_func(root, next, wc, ptr_gen);
Josef Bacikb50c6e22013-04-25 15:55:30 -04002430 if (ret) {
2431 free_extent_buffer(next);
Mark Fasheh1e5063d2011-07-12 10:46:06 -07002432 return ret;
Josef Bacikb50c6e22013-04-25 15:55:30 -04002433 }
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002434
Chris Masone02119d2008-09-05 16:13:11 -04002435 path->slots[*level]++;
2436 if (wc->free) {
Tsutomu Itoh018642a2012-05-29 18:10:13 +09002437 ret = btrfs_read_buffer(next, ptr_gen);
2438 if (ret) {
2439 free_extent_buffer(next);
2440 return ret;
2441 }
Chris Masone02119d2008-09-05 16:13:11 -04002442
Josef Bacik681ae502013-10-07 15:11:00 -04002443 if (trans) {
2444 btrfs_tree_lock(next);
2445 btrfs_set_lock_blocking(next);
Daniel Dressler01d58472014-11-21 17:15:07 +09002446 clean_tree_block(trans, root->fs_info,
2447 next);
Josef Bacik681ae502013-10-07 15:11:00 -04002448 btrfs_wait_tree_block_writeback(next);
2449 btrfs_tree_unlock(next);
2450 }
Chris Masone02119d2008-09-05 16:13:11 -04002451
Chris Masone02119d2008-09-05 16:13:11 -04002452 WARN_ON(root_owner !=
2453 BTRFS_TREE_LOG_OBJECTID);
Chris Masone688b7252011-10-31 20:52:39 -04002454 ret = btrfs_free_and_pin_reserved_extent(root,
Chris Masond00aff02008-09-11 15:54:42 -04002455 bytenr, blocksize);
Josef Bacik36508602013-04-25 16:23:32 -04002456 if (ret) {
2457 free_extent_buffer(next);
2458 return ret;
2459 }
Chris Masone02119d2008-09-05 16:13:11 -04002460 }
2461 free_extent_buffer(next);
2462 continue;
2463 }
Tsutomu Itoh018642a2012-05-29 18:10:13 +09002464 ret = btrfs_read_buffer(next, ptr_gen);
2465 if (ret) {
2466 free_extent_buffer(next);
2467 return ret;
2468 }
Chris Masone02119d2008-09-05 16:13:11 -04002469
2470 WARN_ON(*level <= 0);
2471 if (path->nodes[*level-1])
2472 free_extent_buffer(path->nodes[*level-1]);
2473 path->nodes[*level-1] = next;
2474 *level = btrfs_header_level(next);
2475 path->slots[*level] = 0;
2476 cond_resched();
2477 }
2478 WARN_ON(*level < 0);
2479 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2480
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002481 path->slots[*level] = btrfs_header_nritems(path->nodes[*level]);
Chris Masone02119d2008-09-05 16:13:11 -04002482
2483 cond_resched();
2484 return 0;
2485}
2486
Chris Masond3977122009-01-05 21:25:51 -05002487static noinline int walk_up_log_tree(struct btrfs_trans_handle *trans,
Chris Masone02119d2008-09-05 16:13:11 -04002488 struct btrfs_root *root,
2489 struct btrfs_path *path, int *level,
2490 struct walk_control *wc)
2491{
2492 u64 root_owner;
Chris Masone02119d2008-09-05 16:13:11 -04002493 int i;
2494 int slot;
2495 int ret;
2496
Chris Masond3977122009-01-05 21:25:51 -05002497 for (i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) {
Chris Masone02119d2008-09-05 16:13:11 -04002498 slot = path->slots[i];
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002499 if (slot + 1 < btrfs_header_nritems(path->nodes[i])) {
Chris Masone02119d2008-09-05 16:13:11 -04002500 path->slots[i]++;
2501 *level = i;
2502 WARN_ON(*level == 0);
2503 return 0;
2504 } else {
Zheng Yan31840ae2008-09-23 13:14:14 -04002505 struct extent_buffer *parent;
2506 if (path->nodes[*level] == root->node)
2507 parent = path->nodes[*level];
2508 else
2509 parent = path->nodes[*level + 1];
2510
2511 root_owner = btrfs_header_owner(parent);
Mark Fasheh1e5063d2011-07-12 10:46:06 -07002512 ret = wc->process_func(root, path->nodes[*level], wc,
Chris Masone02119d2008-09-05 16:13:11 -04002513 btrfs_header_generation(path->nodes[*level]));
Mark Fasheh1e5063d2011-07-12 10:46:06 -07002514 if (ret)
2515 return ret;
2516
Chris Masone02119d2008-09-05 16:13:11 -04002517 if (wc->free) {
2518 struct extent_buffer *next;
2519
2520 next = path->nodes[*level];
2521
Josef Bacik681ae502013-10-07 15:11:00 -04002522 if (trans) {
2523 btrfs_tree_lock(next);
2524 btrfs_set_lock_blocking(next);
Daniel Dressler01d58472014-11-21 17:15:07 +09002525 clean_tree_block(trans, root->fs_info,
2526 next);
Josef Bacik681ae502013-10-07 15:11:00 -04002527 btrfs_wait_tree_block_writeback(next);
2528 btrfs_tree_unlock(next);
2529 }
Chris Masone02119d2008-09-05 16:13:11 -04002530
Chris Masone02119d2008-09-05 16:13:11 -04002531 WARN_ON(root_owner != BTRFS_TREE_LOG_OBJECTID);
Chris Masone688b7252011-10-31 20:52:39 -04002532 ret = btrfs_free_and_pin_reserved_extent(root,
Chris Masone02119d2008-09-05 16:13:11 -04002533 path->nodes[*level]->start,
Chris Masond00aff02008-09-11 15:54:42 -04002534 path->nodes[*level]->len);
Josef Bacik36508602013-04-25 16:23:32 -04002535 if (ret)
2536 return ret;
Chris Masone02119d2008-09-05 16:13:11 -04002537 }
2538 free_extent_buffer(path->nodes[*level]);
2539 path->nodes[*level] = NULL;
2540 *level = i + 1;
2541 }
2542 }
2543 return 1;
2544}
2545
2546/*
2547 * drop the reference count on the tree rooted at 'snap'. This traverses
2548 * the tree freeing any blocks that have a ref count of zero after being
2549 * decremented.
2550 */
2551static int walk_log_tree(struct btrfs_trans_handle *trans,
2552 struct btrfs_root *log, struct walk_control *wc)
2553{
2554 int ret = 0;
2555 int wret;
2556 int level;
2557 struct btrfs_path *path;
Chris Masone02119d2008-09-05 16:13:11 -04002558 int orig_level;
2559
2560 path = btrfs_alloc_path();
Tsutomu Itohdb5b4932011-03-23 08:14:16 +00002561 if (!path)
2562 return -ENOMEM;
Chris Masone02119d2008-09-05 16:13:11 -04002563
2564 level = btrfs_header_level(log->node);
2565 orig_level = level;
2566 path->nodes[level] = log->node;
2567 extent_buffer_get(log->node);
2568 path->slots[level] = 0;
2569
Chris Masond3977122009-01-05 21:25:51 -05002570 while (1) {
Chris Masone02119d2008-09-05 16:13:11 -04002571 wret = walk_down_log_tree(trans, log, path, &level, wc);
2572 if (wret > 0)
2573 break;
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002574 if (wret < 0) {
Chris Masone02119d2008-09-05 16:13:11 -04002575 ret = wret;
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002576 goto out;
2577 }
Chris Masone02119d2008-09-05 16:13:11 -04002578
2579 wret = walk_up_log_tree(trans, log, path, &level, wc);
2580 if (wret > 0)
2581 break;
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002582 if (wret < 0) {
Chris Masone02119d2008-09-05 16:13:11 -04002583 ret = wret;
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002584 goto out;
2585 }
Chris Masone02119d2008-09-05 16:13:11 -04002586 }
2587
2588 /* was the root node processed? if not, catch it here */
2589 if (path->nodes[orig_level]) {
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002590 ret = wc->process_func(log, path->nodes[orig_level], wc,
Chris Masone02119d2008-09-05 16:13:11 -04002591 btrfs_header_generation(path->nodes[orig_level]));
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002592 if (ret)
2593 goto out;
Chris Masone02119d2008-09-05 16:13:11 -04002594 if (wc->free) {
2595 struct extent_buffer *next;
2596
2597 next = path->nodes[orig_level];
2598
Josef Bacik681ae502013-10-07 15:11:00 -04002599 if (trans) {
2600 btrfs_tree_lock(next);
2601 btrfs_set_lock_blocking(next);
Daniel Dressler01d58472014-11-21 17:15:07 +09002602 clean_tree_block(trans, log->fs_info, next);
Josef Bacik681ae502013-10-07 15:11:00 -04002603 btrfs_wait_tree_block_writeback(next);
2604 btrfs_tree_unlock(next);
2605 }
Chris Masone02119d2008-09-05 16:13:11 -04002606
Chris Masone02119d2008-09-05 16:13:11 -04002607 WARN_ON(log->root_key.objectid !=
2608 BTRFS_TREE_LOG_OBJECTID);
Chris Masone688b7252011-10-31 20:52:39 -04002609 ret = btrfs_free_and_pin_reserved_extent(log, next->start,
Chris Masond00aff02008-09-11 15:54:42 -04002610 next->len);
Josef Bacik36508602013-04-25 16:23:32 -04002611 if (ret)
2612 goto out;
Chris Masone02119d2008-09-05 16:13:11 -04002613 }
2614 }
2615
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002616out:
Chris Masone02119d2008-09-05 16:13:11 -04002617 btrfs_free_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04002618 return ret;
2619}
2620
Yan Zheng7237f182009-01-21 12:54:03 -05002621/*
2622 * helper function to update the item for a given subvolumes log root
2623 * in the tree of log roots
2624 */
2625static int update_log_root(struct btrfs_trans_handle *trans,
2626 struct btrfs_root *log)
2627{
2628 int ret;
2629
2630 if (log->log_transid == 1) {
2631 /* insert root item on the first sync */
2632 ret = btrfs_insert_root(trans, log->fs_info->log_root_tree,
2633 &log->root_key, &log->root_item);
2634 } else {
2635 ret = btrfs_update_root(trans, log->fs_info->log_root_tree,
2636 &log->root_key, &log->root_item);
2637 }
2638 return ret;
2639}
2640
Zhaolei60d53eb2015-08-17 18:44:46 +08002641static void wait_log_commit(struct btrfs_root *root, int transid)
Chris Masone02119d2008-09-05 16:13:11 -04002642{
2643 DEFINE_WAIT(wait);
Yan Zheng7237f182009-01-21 12:54:03 -05002644 int index = transid % 2;
Chris Masone02119d2008-09-05 16:13:11 -04002645
Yan Zheng7237f182009-01-21 12:54:03 -05002646 /*
2647 * we only allow two pending log transactions at a time,
2648 * so we know that if ours is more than 2 older than the
2649 * current transaction, we're done
2650 */
Chris Masone02119d2008-09-05 16:13:11 -04002651 do {
Yan Zheng7237f182009-01-21 12:54:03 -05002652 prepare_to_wait(&root->log_commit_wait[index],
2653 &wait, TASK_UNINTERRUPTIBLE);
2654 mutex_unlock(&root->log_mutex);
Chris Mason12fcfd22009-03-24 10:24:20 -04002655
Miao Xied1433de2014-02-20 18:08:59 +08002656 if (root->log_transid_committed < transid &&
Yan Zheng7237f182009-01-21 12:54:03 -05002657 atomic_read(&root->log_commit[index]))
Chris Masone02119d2008-09-05 16:13:11 -04002658 schedule();
Chris Mason12fcfd22009-03-24 10:24:20 -04002659
Yan Zheng7237f182009-01-21 12:54:03 -05002660 finish_wait(&root->log_commit_wait[index], &wait);
2661 mutex_lock(&root->log_mutex);
Miao Xied1433de2014-02-20 18:08:59 +08002662 } while (root->log_transid_committed < transid &&
Yan Zheng7237f182009-01-21 12:54:03 -05002663 atomic_read(&root->log_commit[index]));
Yan Zheng7237f182009-01-21 12:54:03 -05002664}
2665
Zhaolei60d53eb2015-08-17 18:44:46 +08002666static void wait_for_writer(struct btrfs_root *root)
Yan Zheng7237f182009-01-21 12:54:03 -05002667{
2668 DEFINE_WAIT(wait);
Miao Xie8b050d32014-02-20 18:08:58 +08002669
2670 while (atomic_read(&root->log_writers)) {
Yan Zheng7237f182009-01-21 12:54:03 -05002671 prepare_to_wait(&root->log_writer_wait,
2672 &wait, TASK_UNINTERRUPTIBLE);
2673 mutex_unlock(&root->log_mutex);
Miao Xie8b050d32014-02-20 18:08:58 +08002674 if (atomic_read(&root->log_writers))
Yan Zheng7237f182009-01-21 12:54:03 -05002675 schedule();
Yan Zheng7237f182009-01-21 12:54:03 -05002676 finish_wait(&root->log_writer_wait, &wait);
Filipe Manana575849e2015-02-11 11:12:39 +00002677 mutex_lock(&root->log_mutex);
Yan Zheng7237f182009-01-21 12:54:03 -05002678 }
Chris Masone02119d2008-09-05 16:13:11 -04002679}
2680
Miao Xie8b050d32014-02-20 18:08:58 +08002681static inline void btrfs_remove_log_ctx(struct btrfs_root *root,
2682 struct btrfs_log_ctx *ctx)
2683{
2684 if (!ctx)
2685 return;
2686
2687 mutex_lock(&root->log_mutex);
2688 list_del_init(&ctx->list);
2689 mutex_unlock(&root->log_mutex);
2690}
2691
2692/*
2693 * Invoked in log mutex context, or be sure there is no other task which
2694 * can access the list.
2695 */
2696static inline void btrfs_remove_all_log_ctxs(struct btrfs_root *root,
2697 int index, int error)
2698{
2699 struct btrfs_log_ctx *ctx;
2700
2701 if (!error) {
2702 INIT_LIST_HEAD(&root->log_ctxs[index]);
2703 return;
2704 }
2705
2706 list_for_each_entry(ctx, &root->log_ctxs[index], list)
2707 ctx->log_ret = error;
2708
2709 INIT_LIST_HEAD(&root->log_ctxs[index]);
2710}
2711
Chris Masone02119d2008-09-05 16:13:11 -04002712/*
2713 * btrfs_sync_log does sends a given tree log down to the disk and
2714 * updates the super blocks to record it. When this call is done,
Chris Mason12fcfd22009-03-24 10:24:20 -04002715 * you know that any inodes previously logged are safely on disk only
2716 * if it returns 0.
2717 *
2718 * Any other return value means you need to call btrfs_commit_transaction.
2719 * Some of the edge cases for fsyncing directories that have had unlinks
2720 * or renames done in the past mean that sometimes the only safe
2721 * fsync is to commit the whole FS. When btrfs_sync_log returns -EAGAIN,
2722 * that has happened.
Chris Masone02119d2008-09-05 16:13:11 -04002723 */
2724int btrfs_sync_log(struct btrfs_trans_handle *trans,
Miao Xie8b050d32014-02-20 18:08:58 +08002725 struct btrfs_root *root, struct btrfs_log_ctx *ctx)
Chris Masone02119d2008-09-05 16:13:11 -04002726{
Yan Zheng7237f182009-01-21 12:54:03 -05002727 int index1;
2728 int index2;
Yan, Zheng8cef4e12009-11-12 09:33:26 +00002729 int mark;
Chris Masone02119d2008-09-05 16:13:11 -04002730 int ret;
Chris Masone02119d2008-09-05 16:13:11 -04002731 struct btrfs_root *log = root->log_root;
Yan Zheng7237f182009-01-21 12:54:03 -05002732 struct btrfs_root *log_root_tree = root->fs_info->log_root_tree;
Miao Xiebb14a592014-02-20 18:08:56 +08002733 int log_transid = 0;
Miao Xie8b050d32014-02-20 18:08:58 +08002734 struct btrfs_log_ctx root_log_ctx;
Miao Xiec6adc9c2013-05-28 10:05:39 +00002735 struct blk_plug plug;
Chris Masone02119d2008-09-05 16:13:11 -04002736
Yan Zheng7237f182009-01-21 12:54:03 -05002737 mutex_lock(&root->log_mutex);
Miao Xied1433de2014-02-20 18:08:59 +08002738 log_transid = ctx->log_transid;
2739 if (root->log_transid_committed >= log_transid) {
Yan Zheng7237f182009-01-21 12:54:03 -05002740 mutex_unlock(&root->log_mutex);
Miao Xie8b050d32014-02-20 18:08:58 +08002741 return ctx->log_ret;
Chris Masone02119d2008-09-05 16:13:11 -04002742 }
Miao Xied1433de2014-02-20 18:08:59 +08002743
2744 index1 = log_transid % 2;
2745 if (atomic_read(&root->log_commit[index1])) {
Zhaolei60d53eb2015-08-17 18:44:46 +08002746 wait_log_commit(root, log_transid);
Miao Xied1433de2014-02-20 18:08:59 +08002747 mutex_unlock(&root->log_mutex);
2748 return ctx->log_ret;
2749 }
2750 ASSERT(log_transid == root->log_transid);
Yan Zheng7237f182009-01-21 12:54:03 -05002751 atomic_set(&root->log_commit[index1], 1);
2752
2753 /* wait for previous tree log sync to complete */
2754 if (atomic_read(&root->log_commit[(index1 + 1) % 2]))
Zhaolei60d53eb2015-08-17 18:44:46 +08002755 wait_log_commit(root, log_transid - 1);
Miao Xie48cab2e2014-02-20 18:08:52 +08002756
Yan, Zheng86df7eb2009-10-14 09:24:59 -04002757 while (1) {
Miao Xie2ecb7922012-09-06 04:04:27 -06002758 int batch = atomic_read(&root->log_batch);
Chris Masoncd354ad2011-10-20 15:45:37 -04002759 /* when we're on an ssd, just kick the log commit out */
Jeff Mahoney3cdde222016-06-09 21:38:35 -04002760 if (!btrfs_test_opt(root->fs_info, SSD) &&
Miao Xie27cdeb72014-04-02 19:51:05 +08002761 test_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state)) {
Yan, Zheng86df7eb2009-10-14 09:24:59 -04002762 mutex_unlock(&root->log_mutex);
2763 schedule_timeout_uninterruptible(1);
2764 mutex_lock(&root->log_mutex);
2765 }
Zhaolei60d53eb2015-08-17 18:44:46 +08002766 wait_for_writer(root);
Miao Xie2ecb7922012-09-06 04:04:27 -06002767 if (batch == atomic_read(&root->log_batch))
Chris Masone02119d2008-09-05 16:13:11 -04002768 break;
2769 }
Chris Masond0c803c2008-09-11 16:17:57 -04002770
Chris Mason12fcfd22009-03-24 10:24:20 -04002771 /* bail out if we need to do a full commit */
Miao Xie995946d2014-04-02 19:51:06 +08002772 if (btrfs_need_log_full_commit(root->fs_info, trans)) {
Chris Mason12fcfd22009-03-24 10:24:20 -04002773 ret = -EAGAIN;
Josef Bacik2ab28f32012-10-12 15:27:49 -04002774 btrfs_free_logged_extents(log, log_transid);
Chris Mason12fcfd22009-03-24 10:24:20 -04002775 mutex_unlock(&root->log_mutex);
2776 goto out;
2777 }
2778
Yan, Zheng8cef4e12009-11-12 09:33:26 +00002779 if (log_transid % 2 == 0)
2780 mark = EXTENT_DIRTY;
2781 else
2782 mark = EXTENT_NEW;
2783
Chris Mason690587d2009-10-13 13:29:19 -04002784 /* we start IO on all the marked extents here, but we don't actually
2785 * wait for them until later.
2786 */
Miao Xiec6adc9c2013-05-28 10:05:39 +00002787 blk_start_plug(&plug);
Yan, Zheng8cef4e12009-11-12 09:33:26 +00002788 ret = btrfs_write_marked_extents(log, &log->dirty_log_pages, mark);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002789 if (ret) {
Miao Xiec6adc9c2013-05-28 10:05:39 +00002790 blk_finish_plug(&plug);
Jeff Mahoney66642832016-06-10 18:19:25 -04002791 btrfs_abort_transaction(trans, ret);
Josef Bacik2ab28f32012-10-12 15:27:49 -04002792 btrfs_free_logged_extents(log, log_transid);
Miao Xie995946d2014-04-02 19:51:06 +08002793 btrfs_set_log_full_commit(root->fs_info, trans);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002794 mutex_unlock(&root->log_mutex);
2795 goto out;
2796 }
Yan Zheng7237f182009-01-21 12:54:03 -05002797
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002798 btrfs_set_root_node(&log->root_item, log->node);
Yan Zheng7237f182009-01-21 12:54:03 -05002799
Yan Zheng7237f182009-01-21 12:54:03 -05002800 root->log_transid++;
2801 log->log_transid = root->log_transid;
Josef Bacikff782e02009-10-08 15:30:04 -04002802 root->log_start_pid = 0;
Yan Zheng7237f182009-01-21 12:54:03 -05002803 /*
Yan, Zheng8cef4e12009-11-12 09:33:26 +00002804 * IO has been started, blocks of the log tree have WRITTEN flag set
2805 * in their headers. new modifications of the log will be written to
2806 * new positions. so it's safe to allow log writers to go in.
Yan Zheng7237f182009-01-21 12:54:03 -05002807 */
2808 mutex_unlock(&root->log_mutex);
2809
Miao Xied1433de2014-02-20 18:08:59 +08002810 btrfs_init_log_ctx(&root_log_ctx);
2811
Yan Zheng7237f182009-01-21 12:54:03 -05002812 mutex_lock(&log_root_tree->log_mutex);
Miao Xie2ecb7922012-09-06 04:04:27 -06002813 atomic_inc(&log_root_tree->log_batch);
Yan Zheng7237f182009-01-21 12:54:03 -05002814 atomic_inc(&log_root_tree->log_writers);
Miao Xied1433de2014-02-20 18:08:59 +08002815
2816 index2 = log_root_tree->log_transid % 2;
2817 list_add_tail(&root_log_ctx.list, &log_root_tree->log_ctxs[index2]);
2818 root_log_ctx.log_transid = log_root_tree->log_transid;
2819
Yan Zheng7237f182009-01-21 12:54:03 -05002820 mutex_unlock(&log_root_tree->log_mutex);
2821
2822 ret = update_log_root(trans, log);
Yan Zheng7237f182009-01-21 12:54:03 -05002823
2824 mutex_lock(&log_root_tree->log_mutex);
2825 if (atomic_dec_and_test(&log_root_tree->log_writers)) {
David Sterba779adf02015-02-16 19:39:00 +01002826 /*
2827 * Implicit memory barrier after atomic_dec_and_test
2828 */
Yan Zheng7237f182009-01-21 12:54:03 -05002829 if (waitqueue_active(&log_root_tree->log_writer_wait))
2830 wake_up(&log_root_tree->log_writer_wait);
2831 }
2832
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002833 if (ret) {
Miao Xied1433de2014-02-20 18:08:59 +08002834 if (!list_empty(&root_log_ctx.list))
2835 list_del_init(&root_log_ctx.list);
2836
Miao Xiec6adc9c2013-05-28 10:05:39 +00002837 blk_finish_plug(&plug);
Miao Xie995946d2014-04-02 19:51:06 +08002838 btrfs_set_log_full_commit(root->fs_info, trans);
2839
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002840 if (ret != -ENOSPC) {
Jeff Mahoney66642832016-06-10 18:19:25 -04002841 btrfs_abort_transaction(trans, ret);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002842 mutex_unlock(&log_root_tree->log_mutex);
2843 goto out;
2844 }
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002845 btrfs_wait_marked_extents(log, &log->dirty_log_pages, mark);
Josef Bacik2ab28f32012-10-12 15:27:49 -04002846 btrfs_free_logged_extents(log, log_transid);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002847 mutex_unlock(&log_root_tree->log_mutex);
2848 ret = -EAGAIN;
2849 goto out;
2850 }
2851
Miao Xied1433de2014-02-20 18:08:59 +08002852 if (log_root_tree->log_transid_committed >= root_log_ctx.log_transid) {
Forrest Liu3da5ab52015-01-30 19:42:12 +08002853 blk_finish_plug(&plug);
Miao Xied1433de2014-02-20 18:08:59 +08002854 mutex_unlock(&log_root_tree->log_mutex);
2855 ret = root_log_ctx.log_ret;
2856 goto out;
2857 }
Miao Xie8b050d32014-02-20 18:08:58 +08002858
Miao Xied1433de2014-02-20 18:08:59 +08002859 index2 = root_log_ctx.log_transid % 2;
Yan Zheng7237f182009-01-21 12:54:03 -05002860 if (atomic_read(&log_root_tree->log_commit[index2])) {
Miao Xiec6adc9c2013-05-28 10:05:39 +00002861 blk_finish_plug(&plug);
Filipe Manana5ab5e442014-11-13 16:59:53 +00002862 ret = btrfs_wait_marked_extents(log, &log->dirty_log_pages,
2863 mark);
Josef Bacik50d9aa92014-11-21 14:52:38 -05002864 btrfs_wait_logged_extents(trans, log, log_transid);
Zhaolei60d53eb2015-08-17 18:44:46 +08002865 wait_log_commit(log_root_tree,
Miao Xied1433de2014-02-20 18:08:59 +08002866 root_log_ctx.log_transid);
Yan Zheng7237f182009-01-21 12:54:03 -05002867 mutex_unlock(&log_root_tree->log_mutex);
Filipe Manana5ab5e442014-11-13 16:59:53 +00002868 if (!ret)
2869 ret = root_log_ctx.log_ret;
Yan Zheng7237f182009-01-21 12:54:03 -05002870 goto out;
2871 }
Miao Xied1433de2014-02-20 18:08:59 +08002872 ASSERT(root_log_ctx.log_transid == log_root_tree->log_transid);
Yan Zheng7237f182009-01-21 12:54:03 -05002873 atomic_set(&log_root_tree->log_commit[index2], 1);
2874
Chris Mason12fcfd22009-03-24 10:24:20 -04002875 if (atomic_read(&log_root_tree->log_commit[(index2 + 1) % 2])) {
Zhaolei60d53eb2015-08-17 18:44:46 +08002876 wait_log_commit(log_root_tree,
Miao Xied1433de2014-02-20 18:08:59 +08002877 root_log_ctx.log_transid - 1);
Chris Mason12fcfd22009-03-24 10:24:20 -04002878 }
Yan Zheng7237f182009-01-21 12:54:03 -05002879
Zhaolei60d53eb2015-08-17 18:44:46 +08002880 wait_for_writer(log_root_tree);
Chris Mason12fcfd22009-03-24 10:24:20 -04002881
2882 /*
2883 * now that we've moved on to the tree of log tree roots,
2884 * check the full commit flag again
2885 */
Miao Xie995946d2014-04-02 19:51:06 +08002886 if (btrfs_need_log_full_commit(root->fs_info, trans)) {
Miao Xiec6adc9c2013-05-28 10:05:39 +00002887 blk_finish_plug(&plug);
Yan, Zheng8cef4e12009-11-12 09:33:26 +00002888 btrfs_wait_marked_extents(log, &log->dirty_log_pages, mark);
Josef Bacik2ab28f32012-10-12 15:27:49 -04002889 btrfs_free_logged_extents(log, log_transid);
Chris Mason12fcfd22009-03-24 10:24:20 -04002890 mutex_unlock(&log_root_tree->log_mutex);
2891 ret = -EAGAIN;
2892 goto out_wake_log_root;
2893 }
Yan Zheng7237f182009-01-21 12:54:03 -05002894
Miao Xiec6adc9c2013-05-28 10:05:39 +00002895 ret = btrfs_write_marked_extents(log_root_tree,
2896 &log_root_tree->dirty_log_pages,
2897 EXTENT_DIRTY | EXTENT_NEW);
2898 blk_finish_plug(&plug);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002899 if (ret) {
Miao Xie995946d2014-04-02 19:51:06 +08002900 btrfs_set_log_full_commit(root->fs_info, trans);
Jeff Mahoney66642832016-06-10 18:19:25 -04002901 btrfs_abort_transaction(trans, ret);
Josef Bacik2ab28f32012-10-12 15:27:49 -04002902 btrfs_free_logged_extents(log, log_transid);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002903 mutex_unlock(&log_root_tree->log_mutex);
2904 goto out_wake_log_root;
2905 }
Filipe Manana5ab5e442014-11-13 16:59:53 +00002906 ret = btrfs_wait_marked_extents(log, &log->dirty_log_pages, mark);
2907 if (!ret)
2908 ret = btrfs_wait_marked_extents(log_root_tree,
2909 &log_root_tree->dirty_log_pages,
2910 EXTENT_NEW | EXTENT_DIRTY);
2911 if (ret) {
2912 btrfs_set_log_full_commit(root->fs_info, trans);
2913 btrfs_free_logged_extents(log, log_transid);
2914 mutex_unlock(&log_root_tree->log_mutex);
2915 goto out_wake_log_root;
2916 }
Josef Bacik50d9aa92014-11-21 14:52:38 -05002917 btrfs_wait_logged_extents(trans, log, log_transid);
Chris Masone02119d2008-09-05 16:13:11 -04002918
David Sterba6c417612011-04-13 15:41:04 +02002919 btrfs_set_super_log_root(root->fs_info->super_for_commit,
Yan Zheng7237f182009-01-21 12:54:03 -05002920 log_root_tree->node->start);
David Sterba6c417612011-04-13 15:41:04 +02002921 btrfs_set_super_log_root_level(root->fs_info->super_for_commit,
Yan Zheng7237f182009-01-21 12:54:03 -05002922 btrfs_header_level(log_root_tree->node));
Chris Masone02119d2008-09-05 16:13:11 -04002923
Yan Zheng7237f182009-01-21 12:54:03 -05002924 log_root_tree->log_transid++;
Yan Zheng7237f182009-01-21 12:54:03 -05002925 mutex_unlock(&log_root_tree->log_mutex);
2926
2927 /*
2928 * nobody else is going to jump in and write the the ctree
2929 * super here because the log_commit atomic below is protecting
2930 * us. We must be called with a transaction handle pinning
2931 * the running transaction open, so a full commit can't hop
2932 * in and cause problems either.
2933 */
Stefan Behrens5af3e8c2012-08-01 18:56:49 +02002934 ret = write_ctree_super(trans, root->fs_info->tree_root, 1);
Stefan Behrens5af3e8c2012-08-01 18:56:49 +02002935 if (ret) {
Miao Xie995946d2014-04-02 19:51:06 +08002936 btrfs_set_log_full_commit(root->fs_info, trans);
Jeff Mahoney66642832016-06-10 18:19:25 -04002937 btrfs_abort_transaction(trans, ret);
Stefan Behrens5af3e8c2012-08-01 18:56:49 +02002938 goto out_wake_log_root;
2939 }
Yan Zheng7237f182009-01-21 12:54:03 -05002940
Chris Mason257c62e2009-10-13 13:21:08 -04002941 mutex_lock(&root->log_mutex);
2942 if (root->last_log_commit < log_transid)
2943 root->last_log_commit = log_transid;
2944 mutex_unlock(&root->log_mutex);
2945
Chris Mason12fcfd22009-03-24 10:24:20 -04002946out_wake_log_root:
Miao Xie8b050d32014-02-20 18:08:58 +08002947 /*
2948 * We needn't get log_mutex here because we are sure all
2949 * the other tasks are blocked.
2950 */
2951 btrfs_remove_all_log_ctxs(log_root_tree, index2, ret);
2952
Miao Xied1433de2014-02-20 18:08:59 +08002953 mutex_lock(&log_root_tree->log_mutex);
2954 log_root_tree->log_transid_committed++;
Yan Zheng7237f182009-01-21 12:54:03 -05002955 atomic_set(&log_root_tree->log_commit[index2], 0);
Miao Xied1433de2014-02-20 18:08:59 +08002956 mutex_unlock(&log_root_tree->log_mutex);
2957
David Sterba33a9eca2015-10-10 18:35:10 +02002958 /*
2959 * The barrier before waitqueue_active is implied by mutex_unlock
2960 */
Yan Zheng7237f182009-01-21 12:54:03 -05002961 if (waitqueue_active(&log_root_tree->log_commit_wait[index2]))
2962 wake_up(&log_root_tree->log_commit_wait[index2]);
Chris Masone02119d2008-09-05 16:13:11 -04002963out:
Miao Xie8b050d32014-02-20 18:08:58 +08002964 /* See above. */
2965 btrfs_remove_all_log_ctxs(root, index1, ret);
2966
Miao Xied1433de2014-02-20 18:08:59 +08002967 mutex_lock(&root->log_mutex);
2968 root->log_transid_committed++;
Yan Zheng7237f182009-01-21 12:54:03 -05002969 atomic_set(&root->log_commit[index1], 0);
Miao Xied1433de2014-02-20 18:08:59 +08002970 mutex_unlock(&root->log_mutex);
Miao Xie8b050d32014-02-20 18:08:58 +08002971
David Sterba33a9eca2015-10-10 18:35:10 +02002972 /*
2973 * The barrier before waitqueue_active is implied by mutex_unlock
2974 */
Yan Zheng7237f182009-01-21 12:54:03 -05002975 if (waitqueue_active(&root->log_commit_wait[index1]))
2976 wake_up(&root->log_commit_wait[index1]);
Chris Masonb31eabd2011-01-31 16:48:24 -05002977 return ret;
Chris Masone02119d2008-09-05 16:13:11 -04002978}
2979
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002980static void free_log_tree(struct btrfs_trans_handle *trans,
2981 struct btrfs_root *log)
Chris Masone02119d2008-09-05 16:13:11 -04002982{
2983 int ret;
Chris Masond0c803c2008-09-11 16:17:57 -04002984 u64 start;
2985 u64 end;
Chris Masone02119d2008-09-05 16:13:11 -04002986 struct walk_control wc = {
2987 .free = 1,
2988 .process_func = process_one_buffer
2989 };
2990
Josef Bacik681ae502013-10-07 15:11:00 -04002991 ret = walk_log_tree(trans, log, &wc);
2992 /* I don't think this can happen but just in case */
2993 if (ret)
Jeff Mahoney66642832016-06-10 18:19:25 -04002994 btrfs_abort_transaction(trans, ret);
Chris Masone02119d2008-09-05 16:13:11 -04002995
Chris Masond3977122009-01-05 21:25:51 -05002996 while (1) {
Chris Masond0c803c2008-09-11 16:17:57 -04002997 ret = find_first_extent_bit(&log->dirty_log_pages,
Josef Bacike6138872012-09-27 17:07:30 -04002998 0, &start, &end, EXTENT_DIRTY | EXTENT_NEW,
2999 NULL);
Chris Masond0c803c2008-09-11 16:17:57 -04003000 if (ret)
3001 break;
3002
Yan, Zheng8cef4e12009-11-12 09:33:26 +00003003 clear_extent_bits(&log->dirty_log_pages, start, end,
David Sterba91166212016-04-26 23:54:39 +02003004 EXTENT_DIRTY | EXTENT_NEW);
Chris Masond0c803c2008-09-11 16:17:57 -04003005 }
3006
Josef Bacik2ab28f32012-10-12 15:27:49 -04003007 /*
3008 * We may have short-circuited the log tree with the full commit logic
3009 * and left ordered extents on our list, so clear these out to keep us
3010 * from leaking inodes and memory.
3011 */
3012 btrfs_free_logged_extents(log, 0);
3013 btrfs_free_logged_extents(log, 1);
3014
Yan Zheng7237f182009-01-21 12:54:03 -05003015 free_extent_buffer(log->node);
3016 kfree(log);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04003017}
3018
3019/*
3020 * free all the extents used by the tree log. This should be called
3021 * at commit time of the full transaction
3022 */
3023int btrfs_free_log(struct btrfs_trans_handle *trans, struct btrfs_root *root)
3024{
3025 if (root->log_root) {
3026 free_log_tree(trans, root->log_root);
3027 root->log_root = NULL;
3028 }
3029 return 0;
3030}
3031
3032int btrfs_free_log_root_tree(struct btrfs_trans_handle *trans,
3033 struct btrfs_fs_info *fs_info)
3034{
3035 if (fs_info->log_root_tree) {
3036 free_log_tree(trans, fs_info->log_root_tree);
3037 fs_info->log_root_tree = NULL;
3038 }
Chris Masone02119d2008-09-05 16:13:11 -04003039 return 0;
3040}
3041
3042/*
Chris Masone02119d2008-09-05 16:13:11 -04003043 * If both a file and directory are logged, and unlinks or renames are
3044 * mixed in, we have a few interesting corners:
3045 *
3046 * create file X in dir Y
3047 * link file X to X.link in dir Y
3048 * fsync file X
3049 * unlink file X but leave X.link
3050 * fsync dir Y
3051 *
3052 * After a crash we would expect only X.link to exist. But file X
3053 * didn't get fsync'd again so the log has back refs for X and X.link.
3054 *
3055 * We solve this by removing directory entries and inode backrefs from the
3056 * log when a file that was logged in the current transaction is
3057 * unlinked. Any later fsync will include the updated log entries, and
3058 * we'll be able to reconstruct the proper directory items from backrefs.
3059 *
3060 * This optimizations allows us to avoid relogging the entire inode
3061 * or the entire directory.
3062 */
3063int btrfs_del_dir_entries_in_log(struct btrfs_trans_handle *trans,
3064 struct btrfs_root *root,
3065 const char *name, int name_len,
3066 struct inode *dir, u64 index)
3067{
3068 struct btrfs_root *log;
3069 struct btrfs_dir_item *di;
3070 struct btrfs_path *path;
3071 int ret;
Yan, Zheng4a500fd2010-05-16 10:49:59 -04003072 int err = 0;
Chris Masone02119d2008-09-05 16:13:11 -04003073 int bytes_del = 0;
Li Zefan33345d012011-04-20 10:31:50 +08003074 u64 dir_ino = btrfs_ino(dir);
Chris Masone02119d2008-09-05 16:13:11 -04003075
Chris Mason3a5f1d42008-09-11 15:53:37 -04003076 if (BTRFS_I(dir)->logged_trans < trans->transid)
3077 return 0;
3078
Chris Masone02119d2008-09-05 16:13:11 -04003079 ret = join_running_log_trans(root);
3080 if (ret)
3081 return 0;
3082
3083 mutex_lock(&BTRFS_I(dir)->log_mutex);
3084
3085 log = root->log_root;
3086 path = btrfs_alloc_path();
Tsutomu Itoha62f44a2011-04-25 19:43:51 -04003087 if (!path) {
3088 err = -ENOMEM;
3089 goto out_unlock;
3090 }
liubo2a29edc2011-01-26 06:22:08 +00003091
Li Zefan33345d012011-04-20 10:31:50 +08003092 di = btrfs_lookup_dir_item(trans, log, path, dir_ino,
Chris Masone02119d2008-09-05 16:13:11 -04003093 name, name_len, -1);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04003094 if (IS_ERR(di)) {
3095 err = PTR_ERR(di);
3096 goto fail;
3097 }
3098 if (di) {
Chris Masone02119d2008-09-05 16:13:11 -04003099 ret = btrfs_delete_one_dir_name(trans, log, path, di);
3100 bytes_del += name_len;
Josef Bacik36508602013-04-25 16:23:32 -04003101 if (ret) {
3102 err = ret;
3103 goto fail;
3104 }
Chris Masone02119d2008-09-05 16:13:11 -04003105 }
David Sterbab3b4aa72011-04-21 01:20:15 +02003106 btrfs_release_path(path);
Li Zefan33345d012011-04-20 10:31:50 +08003107 di = btrfs_lookup_dir_index_item(trans, log, path, dir_ino,
Chris Masone02119d2008-09-05 16:13:11 -04003108 index, name, name_len, -1);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04003109 if (IS_ERR(di)) {
3110 err = PTR_ERR(di);
3111 goto fail;
3112 }
3113 if (di) {
Chris Masone02119d2008-09-05 16:13:11 -04003114 ret = btrfs_delete_one_dir_name(trans, log, path, di);
3115 bytes_del += name_len;
Josef Bacik36508602013-04-25 16:23:32 -04003116 if (ret) {
3117 err = ret;
3118 goto fail;
3119 }
Chris Masone02119d2008-09-05 16:13:11 -04003120 }
3121
3122 /* update the directory size in the log to reflect the names
3123 * we have removed
3124 */
3125 if (bytes_del) {
3126 struct btrfs_key key;
3127
Li Zefan33345d012011-04-20 10:31:50 +08003128 key.objectid = dir_ino;
Chris Masone02119d2008-09-05 16:13:11 -04003129 key.offset = 0;
3130 key.type = BTRFS_INODE_ITEM_KEY;
David Sterbab3b4aa72011-04-21 01:20:15 +02003131 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04003132
3133 ret = btrfs_search_slot(trans, log, &key, path, 0, 1);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04003134 if (ret < 0) {
3135 err = ret;
3136 goto fail;
3137 }
Chris Masone02119d2008-09-05 16:13:11 -04003138 if (ret == 0) {
3139 struct btrfs_inode_item *item;
3140 u64 i_size;
3141
3142 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3143 struct btrfs_inode_item);
3144 i_size = btrfs_inode_size(path->nodes[0], item);
3145 if (i_size > bytes_del)
3146 i_size -= bytes_del;
3147 else
3148 i_size = 0;
3149 btrfs_set_inode_size(path->nodes[0], item, i_size);
3150 btrfs_mark_buffer_dirty(path->nodes[0]);
3151 } else
3152 ret = 0;
David Sterbab3b4aa72011-04-21 01:20:15 +02003153 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04003154 }
Yan, Zheng4a500fd2010-05-16 10:49:59 -04003155fail:
Chris Masone02119d2008-09-05 16:13:11 -04003156 btrfs_free_path(path);
Tsutomu Itoha62f44a2011-04-25 19:43:51 -04003157out_unlock:
Chris Masone02119d2008-09-05 16:13:11 -04003158 mutex_unlock(&BTRFS_I(dir)->log_mutex);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04003159 if (ret == -ENOSPC) {
Miao Xie995946d2014-04-02 19:51:06 +08003160 btrfs_set_log_full_commit(root->fs_info, trans);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04003161 ret = 0;
Jeff Mahoney79787ea2012-03-12 16:03:00 +01003162 } else if (ret < 0)
Jeff Mahoney66642832016-06-10 18:19:25 -04003163 btrfs_abort_transaction(trans, ret);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01003164
Chris Mason12fcfd22009-03-24 10:24:20 -04003165 btrfs_end_log_trans(root);
Chris Masone02119d2008-09-05 16:13:11 -04003166
Andi Kleen411fc6b2010-10-29 15:14:31 -04003167 return err;
Chris Masone02119d2008-09-05 16:13:11 -04003168}
3169
3170/* see comments for btrfs_del_dir_entries_in_log */
3171int btrfs_del_inode_ref_in_log(struct btrfs_trans_handle *trans,
3172 struct btrfs_root *root,
3173 const char *name, int name_len,
3174 struct inode *inode, u64 dirid)
3175{
3176 struct btrfs_root *log;
3177 u64 index;
3178 int ret;
3179
Chris Mason3a5f1d42008-09-11 15:53:37 -04003180 if (BTRFS_I(inode)->logged_trans < trans->transid)
3181 return 0;
3182
Chris Masone02119d2008-09-05 16:13:11 -04003183 ret = join_running_log_trans(root);
3184 if (ret)
3185 return 0;
3186 log = root->log_root;
3187 mutex_lock(&BTRFS_I(inode)->log_mutex);
3188
Li Zefan33345d012011-04-20 10:31:50 +08003189 ret = btrfs_del_inode_ref(trans, log, name, name_len, btrfs_ino(inode),
Chris Masone02119d2008-09-05 16:13:11 -04003190 dirid, &index);
3191 mutex_unlock(&BTRFS_I(inode)->log_mutex);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04003192 if (ret == -ENOSPC) {
Miao Xie995946d2014-04-02 19:51:06 +08003193 btrfs_set_log_full_commit(root->fs_info, trans);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04003194 ret = 0;
Jeff Mahoney79787ea2012-03-12 16:03:00 +01003195 } else if (ret < 0 && ret != -ENOENT)
Jeff Mahoney66642832016-06-10 18:19:25 -04003196 btrfs_abort_transaction(trans, ret);
Chris Mason12fcfd22009-03-24 10:24:20 -04003197 btrfs_end_log_trans(root);
Chris Masone02119d2008-09-05 16:13:11 -04003198
Chris Masone02119d2008-09-05 16:13:11 -04003199 return ret;
3200}
3201
3202/*
3203 * creates a range item in the log for 'dirid'. first_offset and
3204 * last_offset tell us which parts of the key space the log should
3205 * be considered authoritative for.
3206 */
3207static noinline int insert_dir_log_key(struct btrfs_trans_handle *trans,
3208 struct btrfs_root *log,
3209 struct btrfs_path *path,
3210 int key_type, u64 dirid,
3211 u64 first_offset, u64 last_offset)
3212{
3213 int ret;
3214 struct btrfs_key key;
3215 struct btrfs_dir_log_item *item;
3216
3217 key.objectid = dirid;
3218 key.offset = first_offset;
3219 if (key_type == BTRFS_DIR_ITEM_KEY)
3220 key.type = BTRFS_DIR_LOG_ITEM_KEY;
3221 else
3222 key.type = BTRFS_DIR_LOG_INDEX_KEY;
3223 ret = btrfs_insert_empty_item(trans, log, path, &key, sizeof(*item));
Yan, Zheng4a500fd2010-05-16 10:49:59 -04003224 if (ret)
3225 return ret;
Chris Masone02119d2008-09-05 16:13:11 -04003226
3227 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3228 struct btrfs_dir_log_item);
3229 btrfs_set_dir_log_end(path->nodes[0], item, last_offset);
3230 btrfs_mark_buffer_dirty(path->nodes[0]);
David Sterbab3b4aa72011-04-21 01:20:15 +02003231 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04003232 return 0;
3233}
3234
3235/*
3236 * log all the items included in the current transaction for a given
3237 * directory. This also creates the range items in the log tree required
3238 * to replay anything deleted before the fsync
3239 */
3240static noinline int log_dir_items(struct btrfs_trans_handle *trans,
3241 struct btrfs_root *root, struct inode *inode,
3242 struct btrfs_path *path,
3243 struct btrfs_path *dst_path, int key_type,
Filipe Manana2f2ff0e2015-03-20 17:19:46 +00003244 struct btrfs_log_ctx *ctx,
Chris Masone02119d2008-09-05 16:13:11 -04003245 u64 min_offset, u64 *last_offset_ret)
3246{
3247 struct btrfs_key min_key;
Chris Masone02119d2008-09-05 16:13:11 -04003248 struct btrfs_root *log = root->log_root;
3249 struct extent_buffer *src;
Yan, Zheng4a500fd2010-05-16 10:49:59 -04003250 int err = 0;
Chris Masone02119d2008-09-05 16:13:11 -04003251 int ret;
3252 int i;
3253 int nritems;
3254 u64 first_offset = min_offset;
3255 u64 last_offset = (u64)-1;
Li Zefan33345d012011-04-20 10:31:50 +08003256 u64 ino = btrfs_ino(inode);
Chris Masone02119d2008-09-05 16:13:11 -04003257
3258 log = root->log_root;
Chris Masone02119d2008-09-05 16:13:11 -04003259
Li Zefan33345d012011-04-20 10:31:50 +08003260 min_key.objectid = ino;
Chris Masone02119d2008-09-05 16:13:11 -04003261 min_key.type = key_type;
3262 min_key.offset = min_offset;
3263
Filipe David Borba Manana6174d3c2013-10-01 16:13:42 +01003264 ret = btrfs_search_forward(root, &min_key, path, trans->transid);
Chris Masone02119d2008-09-05 16:13:11 -04003265
3266 /*
3267 * we didn't find anything from this transaction, see if there
3268 * is anything at all
3269 */
Li Zefan33345d012011-04-20 10:31:50 +08003270 if (ret != 0 || min_key.objectid != ino || min_key.type != key_type) {
3271 min_key.objectid = ino;
Chris Masone02119d2008-09-05 16:13:11 -04003272 min_key.type = key_type;
3273 min_key.offset = (u64)-1;
David Sterbab3b4aa72011-04-21 01:20:15 +02003274 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04003275 ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0);
3276 if (ret < 0) {
David Sterbab3b4aa72011-04-21 01:20:15 +02003277 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04003278 return ret;
3279 }
Li Zefan33345d012011-04-20 10:31:50 +08003280 ret = btrfs_previous_item(root, path, ino, key_type);
Chris Masone02119d2008-09-05 16:13:11 -04003281
3282 /* if ret == 0 there are items for this type,
3283 * create a range to tell us the last key of this type.
3284 * otherwise, there are no items in this directory after
3285 * *min_offset, and we create a range to indicate that.
3286 */
3287 if (ret == 0) {
3288 struct btrfs_key tmp;
3289 btrfs_item_key_to_cpu(path->nodes[0], &tmp,
3290 path->slots[0]);
Chris Masond3977122009-01-05 21:25:51 -05003291 if (key_type == tmp.type)
Chris Masone02119d2008-09-05 16:13:11 -04003292 first_offset = max(min_offset, tmp.offset) + 1;
Chris Masone02119d2008-09-05 16:13:11 -04003293 }
3294 goto done;
3295 }
3296
3297 /* go backward to find any previous key */
Li Zefan33345d012011-04-20 10:31:50 +08003298 ret = btrfs_previous_item(root, path, ino, key_type);
Chris Masone02119d2008-09-05 16:13:11 -04003299 if (ret == 0) {
3300 struct btrfs_key tmp;
3301 btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]);
3302 if (key_type == tmp.type) {
3303 first_offset = tmp.offset;
3304 ret = overwrite_item(trans, log, dst_path,
3305 path->nodes[0], path->slots[0],
3306 &tmp);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04003307 if (ret) {
3308 err = ret;
3309 goto done;
3310 }
Chris Masone02119d2008-09-05 16:13:11 -04003311 }
3312 }
David Sterbab3b4aa72011-04-21 01:20:15 +02003313 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04003314
3315 /* find the first key from this transaction again */
3316 ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0);
Dulshani Gunawardhanafae7f212013-10-31 10:30:08 +05303317 if (WARN_ON(ret != 0))
Chris Masone02119d2008-09-05 16:13:11 -04003318 goto done;
Chris Masone02119d2008-09-05 16:13:11 -04003319
3320 /*
3321 * we have a block from this transaction, log every item in it
3322 * from our directory
3323 */
Chris Masond3977122009-01-05 21:25:51 -05003324 while (1) {
Chris Masone02119d2008-09-05 16:13:11 -04003325 struct btrfs_key tmp;
3326 src = path->nodes[0];
3327 nritems = btrfs_header_nritems(src);
3328 for (i = path->slots[0]; i < nritems; i++) {
Filipe Manana2f2ff0e2015-03-20 17:19:46 +00003329 struct btrfs_dir_item *di;
3330
Chris Masone02119d2008-09-05 16:13:11 -04003331 btrfs_item_key_to_cpu(src, &min_key, i);
3332
Li Zefan33345d012011-04-20 10:31:50 +08003333 if (min_key.objectid != ino || min_key.type != key_type)
Chris Masone02119d2008-09-05 16:13:11 -04003334 goto done;
3335 ret = overwrite_item(trans, log, dst_path, src, i,
3336 &min_key);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04003337 if (ret) {
3338 err = ret;
3339 goto done;
3340 }
Filipe Manana2f2ff0e2015-03-20 17:19:46 +00003341
3342 /*
3343 * We must make sure that when we log a directory entry,
3344 * the corresponding inode, after log replay, has a
3345 * matching link count. For example:
3346 *
3347 * touch foo
3348 * mkdir mydir
3349 * sync
3350 * ln foo mydir/bar
3351 * xfs_io -c "fsync" mydir
3352 * <crash>
3353 * <mount fs and log replay>
3354 *
3355 * Would result in a fsync log that when replayed, our
3356 * file inode would have a link count of 1, but we get
3357 * two directory entries pointing to the same inode.
3358 * After removing one of the names, it would not be
3359 * possible to remove the other name, which resulted
3360 * always in stale file handle errors, and would not
3361 * be possible to rmdir the parent directory, since
3362 * its i_size could never decrement to the value
3363 * BTRFS_EMPTY_DIR_SIZE, resulting in -ENOTEMPTY errors.
3364 */
3365 di = btrfs_item_ptr(src, i, struct btrfs_dir_item);
3366 btrfs_dir_item_key_to_cpu(src, di, &tmp);
3367 if (ctx &&
3368 (btrfs_dir_transid(src, di) == trans->transid ||
3369 btrfs_dir_type(src, di) == BTRFS_FT_DIR) &&
3370 tmp.type != BTRFS_ROOT_ITEM_KEY)
3371 ctx->log_new_dentries = true;
Chris Masone02119d2008-09-05 16:13:11 -04003372 }
3373 path->slots[0] = nritems;
3374
3375 /*
3376 * look ahead to the next item and see if it is also
3377 * from this directory and from this transaction
3378 */
3379 ret = btrfs_next_leaf(root, path);
3380 if (ret == 1) {
3381 last_offset = (u64)-1;
3382 goto done;
3383 }
3384 btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]);
Li Zefan33345d012011-04-20 10:31:50 +08003385 if (tmp.objectid != ino || tmp.type != key_type) {
Chris Masone02119d2008-09-05 16:13:11 -04003386 last_offset = (u64)-1;
3387 goto done;
3388 }
3389 if (btrfs_header_generation(path->nodes[0]) != trans->transid) {
3390 ret = overwrite_item(trans, log, dst_path,
3391 path->nodes[0], path->slots[0],
3392 &tmp);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04003393 if (ret)
3394 err = ret;
3395 else
3396 last_offset = tmp.offset;
Chris Masone02119d2008-09-05 16:13:11 -04003397 goto done;
3398 }
3399 }
3400done:
David Sterbab3b4aa72011-04-21 01:20:15 +02003401 btrfs_release_path(path);
3402 btrfs_release_path(dst_path);
Chris Masone02119d2008-09-05 16:13:11 -04003403
Yan, Zheng4a500fd2010-05-16 10:49:59 -04003404 if (err == 0) {
3405 *last_offset_ret = last_offset;
3406 /*
3407 * insert the log range keys to indicate where the log
3408 * is valid
3409 */
3410 ret = insert_dir_log_key(trans, log, path, key_type,
Li Zefan33345d012011-04-20 10:31:50 +08003411 ino, first_offset, last_offset);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04003412 if (ret)
3413 err = ret;
3414 }
3415 return err;
Chris Masone02119d2008-09-05 16:13:11 -04003416}
3417
3418/*
3419 * logging directories is very similar to logging inodes, We find all the items
3420 * from the current transaction and write them to the log.
3421 *
3422 * The recovery code scans the directory in the subvolume, and if it finds a
3423 * key in the range logged that is not present in the log tree, then it means
3424 * that dir entry was unlinked during the transaction.
3425 *
3426 * In order for that scan to work, we must include one key smaller than
3427 * the smallest logged by this transaction and one key larger than the largest
3428 * key logged by this transaction.
3429 */
3430static noinline int log_directory_changes(struct btrfs_trans_handle *trans,
3431 struct btrfs_root *root, struct inode *inode,
3432 struct btrfs_path *path,
Filipe Manana2f2ff0e2015-03-20 17:19:46 +00003433 struct btrfs_path *dst_path,
3434 struct btrfs_log_ctx *ctx)
Chris Masone02119d2008-09-05 16:13:11 -04003435{
3436 u64 min_key;
3437 u64 max_key;
3438 int ret;
3439 int key_type = BTRFS_DIR_ITEM_KEY;
3440
3441again:
3442 min_key = 0;
3443 max_key = 0;
Chris Masond3977122009-01-05 21:25:51 -05003444 while (1) {
Chris Masone02119d2008-09-05 16:13:11 -04003445 ret = log_dir_items(trans, root, inode, path,
Filipe Manana2f2ff0e2015-03-20 17:19:46 +00003446 dst_path, key_type, ctx, min_key,
Chris Masone02119d2008-09-05 16:13:11 -04003447 &max_key);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04003448 if (ret)
3449 return ret;
Chris Masone02119d2008-09-05 16:13:11 -04003450 if (max_key == (u64)-1)
3451 break;
3452 min_key = max_key + 1;
3453 }
3454
3455 if (key_type == BTRFS_DIR_ITEM_KEY) {
3456 key_type = BTRFS_DIR_INDEX_KEY;
3457 goto again;
3458 }
3459 return 0;
3460}
3461
3462/*
3463 * a helper function to drop items from the log before we relog an
3464 * inode. max_key_type indicates the highest item type to remove.
3465 * This cannot be run for file data extents because it does not
3466 * free the extents they point to.
3467 */
3468static int drop_objectid_items(struct btrfs_trans_handle *trans,
3469 struct btrfs_root *log,
3470 struct btrfs_path *path,
3471 u64 objectid, int max_key_type)
3472{
3473 int ret;
3474 struct btrfs_key key;
3475 struct btrfs_key found_key;
Josef Bacik18ec90d2012-09-28 11:56:28 -04003476 int start_slot;
Chris Masone02119d2008-09-05 16:13:11 -04003477
3478 key.objectid = objectid;
3479 key.type = max_key_type;
3480 key.offset = (u64)-1;
3481
Chris Masond3977122009-01-05 21:25:51 -05003482 while (1) {
Chris Masone02119d2008-09-05 16:13:11 -04003483 ret = btrfs_search_slot(trans, log, &key, path, -1, 1);
Josef Bacik36508602013-04-25 16:23:32 -04003484 BUG_ON(ret == 0); /* Logic error */
Yan, Zheng4a500fd2010-05-16 10:49:59 -04003485 if (ret < 0)
Chris Masone02119d2008-09-05 16:13:11 -04003486 break;
3487
3488 if (path->slots[0] == 0)
3489 break;
3490
3491 path->slots[0]--;
3492 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
3493 path->slots[0]);
3494
3495 if (found_key.objectid != objectid)
3496 break;
3497
Josef Bacik18ec90d2012-09-28 11:56:28 -04003498 found_key.offset = 0;
3499 found_key.type = 0;
3500 ret = btrfs_bin_search(path->nodes[0], &found_key, 0,
3501 &start_slot);
3502
3503 ret = btrfs_del_items(trans, log, path, start_slot,
3504 path->slots[0] - start_slot + 1);
3505 /*
3506 * If start slot isn't 0 then we don't need to re-search, we've
3507 * found the last guy with the objectid in this tree.
3508 */
3509 if (ret || start_slot != 0)
Tsutomu Itoh65a246c2011-05-19 04:37:44 +00003510 break;
David Sterbab3b4aa72011-04-21 01:20:15 +02003511 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04003512 }
David Sterbab3b4aa72011-04-21 01:20:15 +02003513 btrfs_release_path(path);
Josef Bacik5bdbeb22012-05-29 16:59:49 -04003514 if (ret > 0)
3515 ret = 0;
Yan, Zheng4a500fd2010-05-16 10:49:59 -04003516 return ret;
Chris Masone02119d2008-09-05 16:13:11 -04003517}
3518
Josef Bacik94edf4a2012-09-25 14:56:25 -04003519static void fill_inode_item(struct btrfs_trans_handle *trans,
3520 struct extent_buffer *leaf,
3521 struct btrfs_inode_item *item,
Filipe Manana1a4bcf42015-02-13 12:30:56 +00003522 struct inode *inode, int log_inode_only,
3523 u64 logged_isize)
Josef Bacik94edf4a2012-09-25 14:56:25 -04003524{
Josef Bacik0b1c6cc2012-10-23 16:03:44 -04003525 struct btrfs_map_token token;
Josef Bacik94edf4a2012-09-25 14:56:25 -04003526
Josef Bacik0b1c6cc2012-10-23 16:03:44 -04003527 btrfs_init_map_token(&token);
Josef Bacik94edf4a2012-09-25 14:56:25 -04003528
3529 if (log_inode_only) {
3530 /* set the generation to zero so the recover code
3531 * can tell the difference between an logging
3532 * just to say 'this inode exists' and a logging
3533 * to say 'update this inode with these values'
3534 */
Josef Bacik0b1c6cc2012-10-23 16:03:44 -04003535 btrfs_set_token_inode_generation(leaf, item, 0, &token);
Filipe Manana1a4bcf42015-02-13 12:30:56 +00003536 btrfs_set_token_inode_size(leaf, item, logged_isize, &token);
Josef Bacik94edf4a2012-09-25 14:56:25 -04003537 } else {
Josef Bacik0b1c6cc2012-10-23 16:03:44 -04003538 btrfs_set_token_inode_generation(leaf, item,
3539 BTRFS_I(inode)->generation,
3540 &token);
3541 btrfs_set_token_inode_size(leaf, item, inode->i_size, &token);
Josef Bacik94edf4a2012-09-25 14:56:25 -04003542 }
3543
Josef Bacik0b1c6cc2012-10-23 16:03:44 -04003544 btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token);
3545 btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token);
3546 btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token);
3547 btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token);
3548
David Sterbaa937b972014-12-12 17:39:12 +01003549 btrfs_set_token_timespec_sec(leaf, &item->atime,
Josef Bacik0b1c6cc2012-10-23 16:03:44 -04003550 inode->i_atime.tv_sec, &token);
David Sterbaa937b972014-12-12 17:39:12 +01003551 btrfs_set_token_timespec_nsec(leaf, &item->atime,
Josef Bacik0b1c6cc2012-10-23 16:03:44 -04003552 inode->i_atime.tv_nsec, &token);
3553
David Sterbaa937b972014-12-12 17:39:12 +01003554 btrfs_set_token_timespec_sec(leaf, &item->mtime,
Josef Bacik0b1c6cc2012-10-23 16:03:44 -04003555 inode->i_mtime.tv_sec, &token);
David Sterbaa937b972014-12-12 17:39:12 +01003556 btrfs_set_token_timespec_nsec(leaf, &item->mtime,
Josef Bacik0b1c6cc2012-10-23 16:03:44 -04003557 inode->i_mtime.tv_nsec, &token);
3558
David Sterbaa937b972014-12-12 17:39:12 +01003559 btrfs_set_token_timespec_sec(leaf, &item->ctime,
Josef Bacik0b1c6cc2012-10-23 16:03:44 -04003560 inode->i_ctime.tv_sec, &token);
David Sterbaa937b972014-12-12 17:39:12 +01003561 btrfs_set_token_timespec_nsec(leaf, &item->ctime,
Josef Bacik0b1c6cc2012-10-23 16:03:44 -04003562 inode->i_ctime.tv_nsec, &token);
3563
3564 btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode),
3565 &token);
3566
3567 btrfs_set_token_inode_sequence(leaf, item, inode->i_version, &token);
3568 btrfs_set_token_inode_transid(leaf, item, trans->transid, &token);
3569 btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token);
3570 btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token);
3571 btrfs_set_token_inode_block_group(leaf, item, 0, &token);
Josef Bacik94edf4a2012-09-25 14:56:25 -04003572}
3573
Josef Bacika95249b2012-10-11 16:17:34 -04003574static int log_inode_item(struct btrfs_trans_handle *trans,
3575 struct btrfs_root *log, struct btrfs_path *path,
3576 struct inode *inode)
3577{
3578 struct btrfs_inode_item *inode_item;
Josef Bacika95249b2012-10-11 16:17:34 -04003579 int ret;
3580
Filipe David Borba Mananaefd0c402013-10-07 21:20:44 +01003581 ret = btrfs_insert_empty_item(trans, log, path,
3582 &BTRFS_I(inode)->location,
Josef Bacika95249b2012-10-11 16:17:34 -04003583 sizeof(*inode_item));
3584 if (ret && ret != -EEXIST)
3585 return ret;
3586 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3587 struct btrfs_inode_item);
Filipe Manana1a4bcf42015-02-13 12:30:56 +00003588 fill_inode_item(trans, path->nodes[0], inode_item, inode, 0, 0);
Josef Bacika95249b2012-10-11 16:17:34 -04003589 btrfs_release_path(path);
3590 return 0;
3591}
3592
Chris Mason31ff1cd2008-09-11 16:17:57 -04003593static noinline int copy_items(struct btrfs_trans_handle *trans,
Liu Bod2794402012-08-29 01:07:56 -06003594 struct inode *inode,
Chris Mason31ff1cd2008-09-11 16:17:57 -04003595 struct btrfs_path *dst_path,
Josef Bacik16e75492013-10-22 12:18:51 -04003596 struct btrfs_path *src_path, u64 *last_extent,
Filipe Manana1a4bcf42015-02-13 12:30:56 +00003597 int start_slot, int nr, int inode_only,
3598 u64 logged_isize)
Chris Mason31ff1cd2008-09-11 16:17:57 -04003599{
3600 unsigned long src_offset;
3601 unsigned long dst_offset;
Liu Bod2794402012-08-29 01:07:56 -06003602 struct btrfs_root *log = BTRFS_I(inode)->root->log_root;
Chris Mason31ff1cd2008-09-11 16:17:57 -04003603 struct btrfs_file_extent_item *extent;
3604 struct btrfs_inode_item *inode_item;
Josef Bacik16e75492013-10-22 12:18:51 -04003605 struct extent_buffer *src = src_path->nodes[0];
3606 struct btrfs_key first_key, last_key, key;
Chris Mason31ff1cd2008-09-11 16:17:57 -04003607 int ret;
3608 struct btrfs_key *ins_keys;
3609 u32 *ins_sizes;
3610 char *ins_data;
3611 int i;
Chris Masond20f7042008-12-08 16:58:54 -05003612 struct list_head ordered_sums;
Liu Bod2794402012-08-29 01:07:56 -06003613 int skip_csum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
Josef Bacik16e75492013-10-22 12:18:51 -04003614 bool has_extents = false;
Filipe Manana74121f72014-08-07 12:00:44 +01003615 bool need_find_last_extent = true;
Josef Bacik16e75492013-10-22 12:18:51 -04003616 bool done = false;
Chris Masond20f7042008-12-08 16:58:54 -05003617
3618 INIT_LIST_HEAD(&ordered_sums);
Chris Mason31ff1cd2008-09-11 16:17:57 -04003619
3620 ins_data = kmalloc(nr * sizeof(struct btrfs_key) +
3621 nr * sizeof(u32), GFP_NOFS);
liubo2a29edc2011-01-26 06:22:08 +00003622 if (!ins_data)
3623 return -ENOMEM;
3624
Josef Bacik16e75492013-10-22 12:18:51 -04003625 first_key.objectid = (u64)-1;
3626
Chris Mason31ff1cd2008-09-11 16:17:57 -04003627 ins_sizes = (u32 *)ins_data;
3628 ins_keys = (struct btrfs_key *)(ins_data + nr * sizeof(u32));
3629
3630 for (i = 0; i < nr; i++) {
3631 ins_sizes[i] = btrfs_item_size_nr(src, i + start_slot);
3632 btrfs_item_key_to_cpu(src, ins_keys + i, i + start_slot);
3633 }
3634 ret = btrfs_insert_empty_items(trans, log, dst_path,
3635 ins_keys, ins_sizes, nr);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04003636 if (ret) {
3637 kfree(ins_data);
3638 return ret;
3639 }
Chris Mason31ff1cd2008-09-11 16:17:57 -04003640
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003641 for (i = 0; i < nr; i++, dst_path->slots[0]++) {
Chris Mason31ff1cd2008-09-11 16:17:57 -04003642 dst_offset = btrfs_item_ptr_offset(dst_path->nodes[0],
3643 dst_path->slots[0]);
3644
3645 src_offset = btrfs_item_ptr_offset(src, start_slot + i);
3646
Josef Bacik16e75492013-10-22 12:18:51 -04003647 if ((i == (nr - 1)))
3648 last_key = ins_keys[i];
3649
Josef Bacik94edf4a2012-09-25 14:56:25 -04003650 if (ins_keys[i].type == BTRFS_INODE_ITEM_KEY) {
Chris Mason31ff1cd2008-09-11 16:17:57 -04003651 inode_item = btrfs_item_ptr(dst_path->nodes[0],
3652 dst_path->slots[0],
3653 struct btrfs_inode_item);
Josef Bacik94edf4a2012-09-25 14:56:25 -04003654 fill_inode_item(trans, dst_path->nodes[0], inode_item,
Filipe Manana1a4bcf42015-02-13 12:30:56 +00003655 inode, inode_only == LOG_INODE_EXISTS,
3656 logged_isize);
Josef Bacik94edf4a2012-09-25 14:56:25 -04003657 } else {
3658 copy_extent_buffer(dst_path->nodes[0], src, dst_offset,
3659 src_offset, ins_sizes[i]);
Chris Mason31ff1cd2008-09-11 16:17:57 -04003660 }
Josef Bacik94edf4a2012-09-25 14:56:25 -04003661
Josef Bacik16e75492013-10-22 12:18:51 -04003662 /*
3663 * We set need_find_last_extent here in case we know we were
3664 * processing other items and then walk into the first extent in
3665 * the inode. If we don't hit an extent then nothing changes,
3666 * we'll do the last search the next time around.
3667 */
3668 if (ins_keys[i].type == BTRFS_EXTENT_DATA_KEY) {
3669 has_extents = true;
Filipe Manana74121f72014-08-07 12:00:44 +01003670 if (first_key.objectid == (u64)-1)
Josef Bacik16e75492013-10-22 12:18:51 -04003671 first_key = ins_keys[i];
3672 } else {
3673 need_find_last_extent = false;
3674 }
3675
Chris Mason31ff1cd2008-09-11 16:17:57 -04003676 /* take a reference on file data extents so that truncates
3677 * or deletes of this inode don't have to relog the inode
3678 * again
3679 */
David Sterba962a2982014-06-04 18:41:45 +02003680 if (ins_keys[i].type == BTRFS_EXTENT_DATA_KEY &&
Liu Bod2794402012-08-29 01:07:56 -06003681 !skip_csum) {
Chris Mason31ff1cd2008-09-11 16:17:57 -04003682 int found_type;
3683 extent = btrfs_item_ptr(src, start_slot + i,
3684 struct btrfs_file_extent_item);
3685
liubo8e531cd2011-05-06 10:36:09 +08003686 if (btrfs_file_extent_generation(src, extent) < trans->transid)
3687 continue;
3688
Chris Mason31ff1cd2008-09-11 16:17:57 -04003689 found_type = btrfs_file_extent_type(src, extent);
Josef Bacik6f1fed72012-09-26 11:07:06 -04003690 if (found_type == BTRFS_FILE_EXTENT_REG) {
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003691 u64 ds, dl, cs, cl;
3692 ds = btrfs_file_extent_disk_bytenr(src,
3693 extent);
3694 /* ds == 0 is a hole */
3695 if (ds == 0)
3696 continue;
3697
3698 dl = btrfs_file_extent_disk_num_bytes(src,
3699 extent);
3700 cs = btrfs_file_extent_offset(src, extent);
3701 cl = btrfs_file_extent_num_bytes(src,
Joe Perchesa419aef2009-08-18 11:18:35 -07003702 extent);
Chris Mason580afd72008-12-08 19:15:39 -05003703 if (btrfs_file_extent_compression(src,
3704 extent)) {
3705 cs = 0;
3706 cl = dl;
3707 }
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003708
3709 ret = btrfs_lookup_csums_range(
3710 log->fs_info->csum_root,
3711 ds + cs, ds + cs + cl - 1,
Arne Jansena2de7332011-03-08 14:14:00 +01003712 &ordered_sums, 0);
Josef Bacik36508602013-04-25 16:23:32 -04003713 if (ret) {
3714 btrfs_release_path(dst_path);
3715 kfree(ins_data);
3716 return ret;
3717 }
Chris Mason31ff1cd2008-09-11 16:17:57 -04003718 }
3719 }
Chris Mason31ff1cd2008-09-11 16:17:57 -04003720 }
3721
3722 btrfs_mark_buffer_dirty(dst_path->nodes[0]);
David Sterbab3b4aa72011-04-21 01:20:15 +02003723 btrfs_release_path(dst_path);
Chris Mason31ff1cd2008-09-11 16:17:57 -04003724 kfree(ins_data);
Chris Masond20f7042008-12-08 16:58:54 -05003725
3726 /*
3727 * we have to do this after the loop above to avoid changing the
3728 * log tree while trying to change the log tree.
3729 */
Yan, Zheng4a500fd2010-05-16 10:49:59 -04003730 ret = 0;
Chris Masond3977122009-01-05 21:25:51 -05003731 while (!list_empty(&ordered_sums)) {
Chris Masond20f7042008-12-08 16:58:54 -05003732 struct btrfs_ordered_sum *sums = list_entry(ordered_sums.next,
3733 struct btrfs_ordered_sum,
3734 list);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04003735 if (!ret)
3736 ret = btrfs_csum_file_blocks(trans, log, sums);
Chris Masond20f7042008-12-08 16:58:54 -05003737 list_del(&sums->list);
3738 kfree(sums);
3739 }
Josef Bacik16e75492013-10-22 12:18:51 -04003740
3741 if (!has_extents)
3742 return ret;
3743
Filipe Manana74121f72014-08-07 12:00:44 +01003744 if (need_find_last_extent && *last_extent == first_key.offset) {
3745 /*
3746 * We don't have any leafs between our current one and the one
3747 * we processed before that can have file extent items for our
3748 * inode (and have a generation number smaller than our current
3749 * transaction id).
3750 */
3751 need_find_last_extent = false;
3752 }
3753
Josef Bacik16e75492013-10-22 12:18:51 -04003754 /*
3755 * Because we use btrfs_search_forward we could skip leaves that were
3756 * not modified and then assume *last_extent is valid when it really
3757 * isn't. So back up to the previous leaf and read the end of the last
3758 * extent before we go and fill in holes.
3759 */
3760 if (need_find_last_extent) {
3761 u64 len;
3762
3763 ret = btrfs_prev_leaf(BTRFS_I(inode)->root, src_path);
3764 if (ret < 0)
3765 return ret;
3766 if (ret)
3767 goto fill_holes;
3768 if (src_path->slots[0])
3769 src_path->slots[0]--;
3770 src = src_path->nodes[0];
3771 btrfs_item_key_to_cpu(src, &key, src_path->slots[0]);
3772 if (key.objectid != btrfs_ino(inode) ||
3773 key.type != BTRFS_EXTENT_DATA_KEY)
3774 goto fill_holes;
3775 extent = btrfs_item_ptr(src, src_path->slots[0],
3776 struct btrfs_file_extent_item);
3777 if (btrfs_file_extent_type(src, extent) ==
3778 BTRFS_FILE_EXTENT_INLINE) {
Chris Mason514ac8a2014-01-03 21:07:00 -08003779 len = btrfs_file_extent_inline_len(src,
3780 src_path->slots[0],
3781 extent);
Josef Bacik16e75492013-10-22 12:18:51 -04003782 *last_extent = ALIGN(key.offset + len,
3783 log->sectorsize);
3784 } else {
3785 len = btrfs_file_extent_num_bytes(src, extent);
3786 *last_extent = key.offset + len;
3787 }
3788 }
3789fill_holes:
3790 /* So we did prev_leaf, now we need to move to the next leaf, but a few
3791 * things could have happened
3792 *
3793 * 1) A merge could have happened, so we could currently be on a leaf
3794 * that holds what we were copying in the first place.
3795 * 2) A split could have happened, and now not all of the items we want
3796 * are on the same leaf.
3797 *
3798 * So we need to adjust how we search for holes, we need to drop the
3799 * path and re-search for the first extent key we found, and then walk
3800 * forward until we hit the last one we copied.
3801 */
3802 if (need_find_last_extent) {
3803 /* btrfs_prev_leaf could return 1 without releasing the path */
3804 btrfs_release_path(src_path);
3805 ret = btrfs_search_slot(NULL, BTRFS_I(inode)->root, &first_key,
3806 src_path, 0, 0);
3807 if (ret < 0)
3808 return ret;
3809 ASSERT(ret == 0);
3810 src = src_path->nodes[0];
3811 i = src_path->slots[0];
3812 } else {
3813 i = start_slot;
3814 }
3815
3816 /*
3817 * Ok so here we need to go through and fill in any holes we may have
3818 * to make sure that holes are punched for those areas in case they had
3819 * extents previously.
3820 */
3821 while (!done) {
3822 u64 offset, len;
3823 u64 extent_end;
3824
3825 if (i >= btrfs_header_nritems(src_path->nodes[0])) {
3826 ret = btrfs_next_leaf(BTRFS_I(inode)->root, src_path);
3827 if (ret < 0)
3828 return ret;
3829 ASSERT(ret == 0);
3830 src = src_path->nodes[0];
3831 i = 0;
3832 }
3833
3834 btrfs_item_key_to_cpu(src, &key, i);
3835 if (!btrfs_comp_cpu_keys(&key, &last_key))
3836 done = true;
3837 if (key.objectid != btrfs_ino(inode) ||
3838 key.type != BTRFS_EXTENT_DATA_KEY) {
3839 i++;
3840 continue;
3841 }
3842 extent = btrfs_item_ptr(src, i, struct btrfs_file_extent_item);
3843 if (btrfs_file_extent_type(src, extent) ==
3844 BTRFS_FILE_EXTENT_INLINE) {
Chris Mason514ac8a2014-01-03 21:07:00 -08003845 len = btrfs_file_extent_inline_len(src, i, extent);
Josef Bacik16e75492013-10-22 12:18:51 -04003846 extent_end = ALIGN(key.offset + len, log->sectorsize);
3847 } else {
3848 len = btrfs_file_extent_num_bytes(src, extent);
3849 extent_end = key.offset + len;
3850 }
3851 i++;
3852
3853 if (*last_extent == key.offset) {
3854 *last_extent = extent_end;
3855 continue;
3856 }
3857 offset = *last_extent;
3858 len = key.offset - *last_extent;
3859 ret = btrfs_insert_file_extent(trans, log, btrfs_ino(inode),
3860 offset, 0, 0, len, 0, len, 0,
3861 0, 0);
3862 if (ret)
3863 break;
Filipe Manana74121f72014-08-07 12:00:44 +01003864 *last_extent = extent_end;
Josef Bacik16e75492013-10-22 12:18:51 -04003865 }
3866 /*
3867 * Need to let the callers know we dropped the path so they should
3868 * re-search.
3869 */
3870 if (!ret && need_find_last_extent)
3871 ret = 1;
Yan, Zheng4a500fd2010-05-16 10:49:59 -04003872 return ret;
Chris Mason31ff1cd2008-09-11 16:17:57 -04003873}
3874
Josef Bacik5dc562c2012-08-17 13:14:17 -04003875static int extent_cmp(void *priv, struct list_head *a, struct list_head *b)
3876{
3877 struct extent_map *em1, *em2;
3878
3879 em1 = list_entry(a, struct extent_map, list);
3880 em2 = list_entry(b, struct extent_map, list);
3881
3882 if (em1->start < em2->start)
3883 return -1;
3884 else if (em1->start > em2->start)
3885 return 1;
3886 return 0;
3887}
3888
Filipe Manana8407f552014-09-05 15:14:39 +01003889static int wait_ordered_extents(struct btrfs_trans_handle *trans,
3890 struct inode *inode,
3891 struct btrfs_root *root,
3892 const struct extent_map *em,
3893 const struct list_head *logged_list,
3894 bool *ordered_io_error)
Josef Bacik5dc562c2012-08-17 13:14:17 -04003895{
Josef Bacik2ab28f32012-10-12 15:27:49 -04003896 struct btrfs_ordered_extent *ordered;
Filipe Manana8407f552014-09-05 15:14:39 +01003897 struct btrfs_root *log = root->log_root;
Josef Bacik2ab28f32012-10-12 15:27:49 -04003898 u64 mod_start = em->mod_start;
3899 u64 mod_len = em->mod_len;
Filipe Manana8407f552014-09-05 15:14:39 +01003900 const bool skip_csum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
Josef Bacik2ab28f32012-10-12 15:27:49 -04003901 u64 csum_offset;
3902 u64 csum_len;
Filipe Manana8407f552014-09-05 15:14:39 +01003903 LIST_HEAD(ordered_sums);
3904 int ret = 0;
Josef Bacik09a2a8f92013-04-05 16:51:15 -04003905
Filipe Manana8407f552014-09-05 15:14:39 +01003906 *ordered_io_error = false;
Josef Bacik70c8a912012-10-11 16:54:30 -04003907
Filipe Manana8407f552014-09-05 15:14:39 +01003908 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
3909 em->block_start == EXTENT_MAP_HOLE)
Josef Bacik70c8a912012-10-11 16:54:30 -04003910 return 0;
3911
Josef Bacik2ab28f32012-10-12 15:27:49 -04003912 /*
Filipe Manana8407f552014-09-05 15:14:39 +01003913 * Wait far any ordered extent that covers our extent map. If it
3914 * finishes without an error, first check and see if our csums are on
3915 * our outstanding ordered extents.
Josef Bacik2ab28f32012-10-12 15:27:49 -04003916 */
Miao Xie827463c2014-01-14 20:31:51 +08003917 list_for_each_entry(ordered, logged_list, log_list) {
Josef Bacik2ab28f32012-10-12 15:27:49 -04003918 struct btrfs_ordered_sum *sum;
3919
3920 if (!mod_len)
3921 break;
3922
Josef Bacik2ab28f32012-10-12 15:27:49 -04003923 if (ordered->file_offset + ordered->len <= mod_start ||
3924 mod_start + mod_len <= ordered->file_offset)
3925 continue;
3926
Filipe Manana8407f552014-09-05 15:14:39 +01003927 if (!test_bit(BTRFS_ORDERED_IO_DONE, &ordered->flags) &&
3928 !test_bit(BTRFS_ORDERED_IOERR, &ordered->flags) &&
3929 !test_bit(BTRFS_ORDERED_DIRECT, &ordered->flags)) {
3930 const u64 start = ordered->file_offset;
3931 const u64 end = ordered->file_offset + ordered->len - 1;
3932
3933 WARN_ON(ordered->inode != inode);
3934 filemap_fdatawrite_range(inode->i_mapping, start, end);
3935 }
3936
3937 wait_event(ordered->wait,
3938 (test_bit(BTRFS_ORDERED_IO_DONE, &ordered->flags) ||
3939 test_bit(BTRFS_ORDERED_IOERR, &ordered->flags)));
3940
3941 if (test_bit(BTRFS_ORDERED_IOERR, &ordered->flags)) {
Filipe Mananab38ef712014-11-13 17:01:45 +00003942 /*
3943 * Clear the AS_EIO/AS_ENOSPC flags from the inode's
3944 * i_mapping flags, so that the next fsync won't get
3945 * an outdated io error too.
3946 */
3947 btrfs_inode_check_errors(inode);
Filipe Manana8407f552014-09-05 15:14:39 +01003948 *ordered_io_error = true;
3949 break;
3950 }
Josef Bacik2ab28f32012-10-12 15:27:49 -04003951 /*
3952 * We are going to copy all the csums on this ordered extent, so
3953 * go ahead and adjust mod_start and mod_len in case this
3954 * ordered extent has already been logged.
3955 */
3956 if (ordered->file_offset > mod_start) {
3957 if (ordered->file_offset + ordered->len >=
3958 mod_start + mod_len)
3959 mod_len = ordered->file_offset - mod_start;
3960 /*
3961 * If we have this case
3962 *
3963 * |--------- logged extent ---------|
3964 * |----- ordered extent ----|
3965 *
3966 * Just don't mess with mod_start and mod_len, we'll
3967 * just end up logging more csums than we need and it
3968 * will be ok.
3969 */
3970 } else {
3971 if (ordered->file_offset + ordered->len <
3972 mod_start + mod_len) {
3973 mod_len = (mod_start + mod_len) -
3974 (ordered->file_offset + ordered->len);
3975 mod_start = ordered->file_offset +
3976 ordered->len;
3977 } else {
3978 mod_len = 0;
3979 }
3980 }
3981
Filipe Manana8407f552014-09-05 15:14:39 +01003982 if (skip_csum)
3983 continue;
3984
Josef Bacik2ab28f32012-10-12 15:27:49 -04003985 /*
3986 * To keep us from looping for the above case of an ordered
3987 * extent that falls inside of the logged extent.
3988 */
3989 if (test_and_set_bit(BTRFS_ORDERED_LOGGED_CSUM,
3990 &ordered->flags))
3991 continue;
Josef Bacik2ab28f32012-10-12 15:27:49 -04003992
Josef Bacik2ab28f32012-10-12 15:27:49 -04003993 list_for_each_entry(sum, &ordered->list, list) {
3994 ret = btrfs_csum_file_blocks(trans, log, sum);
Miao Xie827463c2014-01-14 20:31:51 +08003995 if (ret)
Filipe Manana8407f552014-09-05 15:14:39 +01003996 break;
Josef Bacik2ab28f32012-10-12 15:27:49 -04003997 }
Josef Bacik2ab28f32012-10-12 15:27:49 -04003998 }
Josef Bacik2ab28f32012-10-12 15:27:49 -04003999
Filipe Manana8407f552014-09-05 15:14:39 +01004000 if (*ordered_io_error || !mod_len || ret || skip_csum)
Josef Bacik2ab28f32012-10-12 15:27:49 -04004001 return ret;
4002
Filipe David Borba Manana488111a2013-10-28 16:30:29 +00004003 if (em->compress_type) {
4004 csum_offset = 0;
Filipe Manana8407f552014-09-05 15:14:39 +01004005 csum_len = max(em->block_len, em->orig_block_len);
Filipe David Borba Manana488111a2013-10-28 16:30:29 +00004006 } else {
4007 csum_offset = mod_start - em->start;
4008 csum_len = mod_len;
4009 }
Josef Bacik2ab28f32012-10-12 15:27:49 -04004010
Josef Bacik70c8a912012-10-11 16:54:30 -04004011 /* block start is already adjusted for the file extent offset. */
4012 ret = btrfs_lookup_csums_range(log->fs_info->csum_root,
4013 em->block_start + csum_offset,
4014 em->block_start + csum_offset +
4015 csum_len - 1, &ordered_sums, 0);
4016 if (ret)
4017 return ret;
4018
4019 while (!list_empty(&ordered_sums)) {
4020 struct btrfs_ordered_sum *sums = list_entry(ordered_sums.next,
4021 struct btrfs_ordered_sum,
4022 list);
4023 if (!ret)
4024 ret = btrfs_csum_file_blocks(trans, log, sums);
4025 list_del(&sums->list);
4026 kfree(sums);
4027 }
4028
4029 return ret;
Josef Bacik5dc562c2012-08-17 13:14:17 -04004030}
4031
Filipe Manana8407f552014-09-05 15:14:39 +01004032static int log_one_extent(struct btrfs_trans_handle *trans,
4033 struct inode *inode, struct btrfs_root *root,
4034 const struct extent_map *em,
4035 struct btrfs_path *path,
4036 const struct list_head *logged_list,
4037 struct btrfs_log_ctx *ctx)
4038{
4039 struct btrfs_root *log = root->log_root;
4040 struct btrfs_file_extent_item *fi;
4041 struct extent_buffer *leaf;
4042 struct btrfs_map_token token;
4043 struct btrfs_key key;
4044 u64 extent_offset = em->start - em->orig_start;
4045 u64 block_len;
4046 int ret;
4047 int extent_inserted = 0;
4048 bool ordered_io_err = false;
4049
4050 ret = wait_ordered_extents(trans, inode, root, em, logged_list,
4051 &ordered_io_err);
4052 if (ret)
4053 return ret;
4054
4055 if (ordered_io_err) {
4056 ctx->io_err = -EIO;
4057 return 0;
4058 }
4059
4060 btrfs_init_map_token(&token);
4061
4062 ret = __btrfs_drop_extents(trans, log, inode, path, em->start,
4063 em->start + em->len, NULL, 0, 1,
4064 sizeof(*fi), &extent_inserted);
4065 if (ret)
4066 return ret;
4067
4068 if (!extent_inserted) {
4069 key.objectid = btrfs_ino(inode);
4070 key.type = BTRFS_EXTENT_DATA_KEY;
4071 key.offset = em->start;
4072
4073 ret = btrfs_insert_empty_item(trans, log, path, &key,
4074 sizeof(*fi));
4075 if (ret)
4076 return ret;
4077 }
4078 leaf = path->nodes[0];
4079 fi = btrfs_item_ptr(leaf, path->slots[0],
4080 struct btrfs_file_extent_item);
4081
Josef Bacik50d9aa92014-11-21 14:52:38 -05004082 btrfs_set_token_file_extent_generation(leaf, fi, trans->transid,
Filipe Manana8407f552014-09-05 15:14:39 +01004083 &token);
4084 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
4085 btrfs_set_token_file_extent_type(leaf, fi,
4086 BTRFS_FILE_EXTENT_PREALLOC,
4087 &token);
4088 else
4089 btrfs_set_token_file_extent_type(leaf, fi,
4090 BTRFS_FILE_EXTENT_REG,
4091 &token);
4092
4093 block_len = max(em->block_len, em->orig_block_len);
4094 if (em->compress_type != BTRFS_COMPRESS_NONE) {
4095 btrfs_set_token_file_extent_disk_bytenr(leaf, fi,
4096 em->block_start,
4097 &token);
4098 btrfs_set_token_file_extent_disk_num_bytes(leaf, fi, block_len,
4099 &token);
4100 } else if (em->block_start < EXTENT_MAP_LAST_BYTE) {
4101 btrfs_set_token_file_extent_disk_bytenr(leaf, fi,
4102 em->block_start -
4103 extent_offset, &token);
4104 btrfs_set_token_file_extent_disk_num_bytes(leaf, fi, block_len,
4105 &token);
4106 } else {
4107 btrfs_set_token_file_extent_disk_bytenr(leaf, fi, 0, &token);
4108 btrfs_set_token_file_extent_disk_num_bytes(leaf, fi, 0,
4109 &token);
4110 }
4111
4112 btrfs_set_token_file_extent_offset(leaf, fi, extent_offset, &token);
4113 btrfs_set_token_file_extent_num_bytes(leaf, fi, em->len, &token);
4114 btrfs_set_token_file_extent_ram_bytes(leaf, fi, em->ram_bytes, &token);
4115 btrfs_set_token_file_extent_compression(leaf, fi, em->compress_type,
4116 &token);
4117 btrfs_set_token_file_extent_encryption(leaf, fi, 0, &token);
4118 btrfs_set_token_file_extent_other_encoding(leaf, fi, 0, &token);
4119 btrfs_mark_buffer_dirty(leaf);
4120
4121 btrfs_release_path(path);
4122
4123 return ret;
4124}
4125
Josef Bacik5dc562c2012-08-17 13:14:17 -04004126static int btrfs_log_changed_extents(struct btrfs_trans_handle *trans,
4127 struct btrfs_root *root,
4128 struct inode *inode,
Miao Xie827463c2014-01-14 20:31:51 +08004129 struct btrfs_path *path,
Filipe Manana8407f552014-09-05 15:14:39 +01004130 struct list_head *logged_list,
Filipe Mananade0ee0e2016-01-21 10:17:54 +00004131 struct btrfs_log_ctx *ctx,
4132 const u64 start,
4133 const u64 end)
Josef Bacik5dc562c2012-08-17 13:14:17 -04004134{
Josef Bacik5dc562c2012-08-17 13:14:17 -04004135 struct extent_map *em, *n;
4136 struct list_head extents;
4137 struct extent_map_tree *tree = &BTRFS_I(inode)->extent_tree;
4138 u64 test_gen;
4139 int ret = 0;
Josef Bacik2ab28f32012-10-12 15:27:49 -04004140 int num = 0;
Josef Bacik5dc562c2012-08-17 13:14:17 -04004141
4142 INIT_LIST_HEAD(&extents);
4143
Filipe Manana5f9a8a52016-05-12 13:53:36 +01004144 down_write(&BTRFS_I(inode)->dio_sem);
Josef Bacik5dc562c2012-08-17 13:14:17 -04004145 write_lock(&tree->lock);
4146 test_gen = root->fs_info->last_trans_committed;
4147
4148 list_for_each_entry_safe(em, n, &tree->modified_extents, list) {
4149 list_del_init(&em->list);
Josef Bacik2ab28f32012-10-12 15:27:49 -04004150
4151 /*
4152 * Just an arbitrary number, this can be really CPU intensive
4153 * once we start getting a lot of extents, and really once we
4154 * have a bunch of extents we just want to commit since it will
4155 * be faster.
4156 */
4157 if (++num > 32768) {
4158 list_del_init(&tree->modified_extents);
4159 ret = -EFBIG;
4160 goto process;
4161 }
4162
Josef Bacik5dc562c2012-08-17 13:14:17 -04004163 if (em->generation <= test_gen)
4164 continue;
Josef Bacikff44c6e2012-09-14 12:59:20 -04004165 /* Need a ref to keep it from getting evicted from cache */
4166 atomic_inc(&em->refs);
4167 set_bit(EXTENT_FLAG_LOGGING, &em->flags);
Josef Bacik5dc562c2012-08-17 13:14:17 -04004168 list_add_tail(&em->list, &extents);
Josef Bacik2ab28f32012-10-12 15:27:49 -04004169 num++;
Josef Bacik5dc562c2012-08-17 13:14:17 -04004170 }
4171
4172 list_sort(NULL, &extents, extent_cmp);
Filipe Mananade0ee0e2016-01-21 10:17:54 +00004173 btrfs_get_logged_extents(inode, logged_list, start, end);
Filipe Manana5f9a8a52016-05-12 13:53:36 +01004174 /*
4175 * Some ordered extents started by fsync might have completed
4176 * before we could collect them into the list logged_list, which
4177 * means they're gone, not in our logged_list nor in the inode's
4178 * ordered tree. We want the application/user space to know an
4179 * error happened while attempting to persist file data so that
4180 * it can take proper action. If such error happened, we leave
4181 * without writing to the log tree and the fsync must report the
4182 * file data write error and not commit the current transaction.
4183 */
4184 ret = btrfs_inode_check_errors(inode);
4185 if (ret)
4186 ctx->io_err = ret;
Josef Bacik2ab28f32012-10-12 15:27:49 -04004187process:
Josef Bacik5dc562c2012-08-17 13:14:17 -04004188 while (!list_empty(&extents)) {
4189 em = list_entry(extents.next, struct extent_map, list);
4190
4191 list_del_init(&em->list);
4192
4193 /*
4194 * If we had an error we just need to delete everybody from our
4195 * private list.
4196 */
Josef Bacikff44c6e2012-09-14 12:59:20 -04004197 if (ret) {
Josef Bacik201a9032013-01-24 12:02:07 -05004198 clear_em_logging(tree, em);
Josef Bacikff44c6e2012-09-14 12:59:20 -04004199 free_extent_map(em);
Josef Bacik5dc562c2012-08-17 13:14:17 -04004200 continue;
Josef Bacikff44c6e2012-09-14 12:59:20 -04004201 }
4202
4203 write_unlock(&tree->lock);
Josef Bacik5dc562c2012-08-17 13:14:17 -04004204
Filipe Manana8407f552014-09-05 15:14:39 +01004205 ret = log_one_extent(trans, inode, root, em, path, logged_list,
4206 ctx);
Josef Bacikff44c6e2012-09-14 12:59:20 -04004207 write_lock(&tree->lock);
Josef Bacik201a9032013-01-24 12:02:07 -05004208 clear_em_logging(tree, em);
4209 free_extent_map(em);
Josef Bacik5dc562c2012-08-17 13:14:17 -04004210 }
Josef Bacikff44c6e2012-09-14 12:59:20 -04004211 WARN_ON(!list_empty(&extents));
4212 write_unlock(&tree->lock);
Filipe Manana5f9a8a52016-05-12 13:53:36 +01004213 up_write(&BTRFS_I(inode)->dio_sem);
Josef Bacik5dc562c2012-08-17 13:14:17 -04004214
Josef Bacik5dc562c2012-08-17 13:14:17 -04004215 btrfs_release_path(path);
Josef Bacik5dc562c2012-08-17 13:14:17 -04004216 return ret;
4217}
4218
Filipe Manana1a4bcf42015-02-13 12:30:56 +00004219static int logged_inode_size(struct btrfs_root *log, struct inode *inode,
4220 struct btrfs_path *path, u64 *size_ret)
4221{
4222 struct btrfs_key key;
4223 int ret;
4224
4225 key.objectid = btrfs_ino(inode);
4226 key.type = BTRFS_INODE_ITEM_KEY;
4227 key.offset = 0;
4228
4229 ret = btrfs_search_slot(NULL, log, &key, path, 0, 0);
4230 if (ret < 0) {
4231 return ret;
4232 } else if (ret > 0) {
Filipe Manana2f2ff0e2015-03-20 17:19:46 +00004233 *size_ret = 0;
Filipe Manana1a4bcf42015-02-13 12:30:56 +00004234 } else {
4235 struct btrfs_inode_item *item;
4236
4237 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4238 struct btrfs_inode_item);
4239 *size_ret = btrfs_inode_size(path->nodes[0], item);
4240 }
4241
4242 btrfs_release_path(path);
4243 return 0;
4244}
4245
Filipe Manana36283bf2015-06-20 00:44:51 +01004246/*
4247 * At the moment we always log all xattrs. This is to figure out at log replay
4248 * time which xattrs must have their deletion replayed. If a xattr is missing
4249 * in the log tree and exists in the fs/subvol tree, we delete it. This is
4250 * because if a xattr is deleted, the inode is fsynced and a power failure
4251 * happens, causing the log to be replayed the next time the fs is mounted,
4252 * we want the xattr to not exist anymore (same behaviour as other filesystems
4253 * with a journal, ext3/4, xfs, f2fs, etc).
4254 */
4255static int btrfs_log_all_xattrs(struct btrfs_trans_handle *trans,
4256 struct btrfs_root *root,
4257 struct inode *inode,
4258 struct btrfs_path *path,
4259 struct btrfs_path *dst_path)
4260{
4261 int ret;
4262 struct btrfs_key key;
4263 const u64 ino = btrfs_ino(inode);
4264 int ins_nr = 0;
4265 int start_slot = 0;
4266
4267 key.objectid = ino;
4268 key.type = BTRFS_XATTR_ITEM_KEY;
4269 key.offset = 0;
4270
4271 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4272 if (ret < 0)
4273 return ret;
4274
4275 while (true) {
4276 int slot = path->slots[0];
4277 struct extent_buffer *leaf = path->nodes[0];
4278 int nritems = btrfs_header_nritems(leaf);
4279
4280 if (slot >= nritems) {
4281 if (ins_nr > 0) {
4282 u64 last_extent = 0;
4283
4284 ret = copy_items(trans, inode, dst_path, path,
4285 &last_extent, start_slot,
4286 ins_nr, 1, 0);
4287 /* can't be 1, extent items aren't processed */
4288 ASSERT(ret <= 0);
4289 if (ret < 0)
4290 return ret;
4291 ins_nr = 0;
4292 }
4293 ret = btrfs_next_leaf(root, path);
4294 if (ret < 0)
4295 return ret;
4296 else if (ret > 0)
4297 break;
4298 continue;
4299 }
4300
4301 btrfs_item_key_to_cpu(leaf, &key, slot);
4302 if (key.objectid != ino || key.type != BTRFS_XATTR_ITEM_KEY)
4303 break;
4304
4305 if (ins_nr == 0)
4306 start_slot = slot;
4307 ins_nr++;
4308 path->slots[0]++;
4309 cond_resched();
4310 }
4311 if (ins_nr > 0) {
4312 u64 last_extent = 0;
4313
4314 ret = copy_items(trans, inode, dst_path, path,
4315 &last_extent, start_slot,
4316 ins_nr, 1, 0);
4317 /* can't be 1, extent items aren't processed */
4318 ASSERT(ret <= 0);
4319 if (ret < 0)
4320 return ret;
4321 }
4322
4323 return 0;
4324}
4325
Filipe Mananaa89ca6f2015-06-25 04:17:46 +01004326/*
4327 * If the no holes feature is enabled we need to make sure any hole between the
4328 * last extent and the i_size of our inode is explicitly marked in the log. This
4329 * is to make sure that doing something like:
4330 *
4331 * 1) create file with 128Kb of data
4332 * 2) truncate file to 64Kb
4333 * 3) truncate file to 256Kb
4334 * 4) fsync file
4335 * 5) <crash/power failure>
4336 * 6) mount fs and trigger log replay
4337 *
4338 * Will give us a file with a size of 256Kb, the first 64Kb of data match what
4339 * the file had in its first 64Kb of data at step 1 and the last 192Kb of the
4340 * file correspond to a hole. The presence of explicit holes in a log tree is
4341 * what guarantees that log replay will remove/adjust file extent items in the
4342 * fs/subvol tree.
4343 *
4344 * Here we do not need to care about holes between extents, that is already done
4345 * by copy_items(). We also only need to do this in the full sync path, where we
4346 * lookup for extents from the fs/subvol tree only. In the fast path case, we
4347 * lookup the list of modified extent maps and if any represents a hole, we
4348 * insert a corresponding extent representing a hole in the log tree.
4349 */
4350static int btrfs_log_trailing_hole(struct btrfs_trans_handle *trans,
4351 struct btrfs_root *root,
4352 struct inode *inode,
4353 struct btrfs_path *path)
4354{
4355 int ret;
4356 struct btrfs_key key;
4357 u64 hole_start;
4358 u64 hole_size;
4359 struct extent_buffer *leaf;
4360 struct btrfs_root *log = root->log_root;
4361 const u64 ino = btrfs_ino(inode);
4362 const u64 i_size = i_size_read(inode);
4363
4364 if (!btrfs_fs_incompat(root->fs_info, NO_HOLES))
4365 return 0;
4366
4367 key.objectid = ino;
4368 key.type = BTRFS_EXTENT_DATA_KEY;
4369 key.offset = (u64)-1;
4370
4371 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4372 ASSERT(ret != 0);
4373 if (ret < 0)
4374 return ret;
4375
4376 ASSERT(path->slots[0] > 0);
4377 path->slots[0]--;
4378 leaf = path->nodes[0];
4379 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4380
4381 if (key.objectid != ino || key.type != BTRFS_EXTENT_DATA_KEY) {
4382 /* inode does not have any extents */
4383 hole_start = 0;
4384 hole_size = i_size;
4385 } else {
4386 struct btrfs_file_extent_item *extent;
4387 u64 len;
4388
4389 /*
4390 * If there's an extent beyond i_size, an explicit hole was
4391 * already inserted by copy_items().
4392 */
4393 if (key.offset >= i_size)
4394 return 0;
4395
4396 extent = btrfs_item_ptr(leaf, path->slots[0],
4397 struct btrfs_file_extent_item);
4398
4399 if (btrfs_file_extent_type(leaf, extent) ==
4400 BTRFS_FILE_EXTENT_INLINE) {
4401 len = btrfs_file_extent_inline_len(leaf,
4402 path->slots[0],
4403 extent);
4404 ASSERT(len == i_size);
4405 return 0;
4406 }
4407
4408 len = btrfs_file_extent_num_bytes(leaf, extent);
4409 /* Last extent goes beyond i_size, no need to log a hole. */
4410 if (key.offset + len > i_size)
4411 return 0;
4412 hole_start = key.offset + len;
4413 hole_size = i_size - hole_start;
4414 }
4415 btrfs_release_path(path);
4416
4417 /* Last extent ends at i_size. */
4418 if (hole_size == 0)
4419 return 0;
4420
4421 hole_size = ALIGN(hole_size, root->sectorsize);
4422 ret = btrfs_insert_file_extent(trans, log, ino, hole_start, 0, 0,
4423 hole_size, 0, hole_size, 0, 0, 0);
4424 return ret;
4425}
4426
Filipe Manana56f23fd2016-03-30 23:37:21 +01004427/*
4428 * When we are logging a new inode X, check if it doesn't have a reference that
4429 * matches the reference from some other inode Y created in a past transaction
4430 * and that was renamed in the current transaction. If we don't do this, then at
4431 * log replay time we can lose inode Y (and all its files if it's a directory):
4432 *
4433 * mkdir /mnt/x
4434 * echo "hello world" > /mnt/x/foobar
4435 * sync
4436 * mv /mnt/x /mnt/y
4437 * mkdir /mnt/x # or touch /mnt/x
4438 * xfs_io -c fsync /mnt/x
4439 * <power fail>
4440 * mount fs, trigger log replay
4441 *
4442 * After the log replay procedure, we would lose the first directory and all its
4443 * files (file foobar).
4444 * For the case where inode Y is not a directory we simply end up losing it:
4445 *
4446 * echo "123" > /mnt/foo
4447 * sync
4448 * mv /mnt/foo /mnt/bar
4449 * echo "abc" > /mnt/foo
4450 * xfs_io -c fsync /mnt/foo
4451 * <power fail>
4452 *
4453 * We also need this for cases where a snapshot entry is replaced by some other
4454 * entry (file or directory) otherwise we end up with an unreplayable log due to
4455 * attempts to delete the snapshot entry (entry of type BTRFS_ROOT_ITEM_KEY) as
4456 * if it were a regular entry:
4457 *
4458 * mkdir /mnt/x
4459 * btrfs subvolume snapshot /mnt /mnt/x/snap
4460 * btrfs subvolume delete /mnt/x/snap
4461 * rmdir /mnt/x
4462 * mkdir /mnt/x
4463 * fsync /mnt/x or fsync some new file inside it
4464 * <power fail>
4465 *
4466 * The snapshot delete, rmdir of x, mkdir of a new x and the fsync all happen in
4467 * the same transaction.
4468 */
4469static int btrfs_check_ref_name_override(struct extent_buffer *eb,
4470 const int slot,
4471 const struct btrfs_key *key,
4472 struct inode *inode)
4473{
4474 int ret;
4475 struct btrfs_path *search_path;
4476 char *name = NULL;
4477 u32 name_len = 0;
4478 u32 item_size = btrfs_item_size_nr(eb, slot);
4479 u32 cur_offset = 0;
4480 unsigned long ptr = btrfs_item_ptr_offset(eb, slot);
4481
4482 search_path = btrfs_alloc_path();
4483 if (!search_path)
4484 return -ENOMEM;
4485 search_path->search_commit_root = 1;
4486 search_path->skip_locking = 1;
4487
4488 while (cur_offset < item_size) {
4489 u64 parent;
4490 u32 this_name_len;
4491 u32 this_len;
4492 unsigned long name_ptr;
4493 struct btrfs_dir_item *di;
4494
4495 if (key->type == BTRFS_INODE_REF_KEY) {
4496 struct btrfs_inode_ref *iref;
4497
4498 iref = (struct btrfs_inode_ref *)(ptr + cur_offset);
4499 parent = key->offset;
4500 this_name_len = btrfs_inode_ref_name_len(eb, iref);
4501 name_ptr = (unsigned long)(iref + 1);
4502 this_len = sizeof(*iref) + this_name_len;
4503 } else {
4504 struct btrfs_inode_extref *extref;
4505
4506 extref = (struct btrfs_inode_extref *)(ptr +
4507 cur_offset);
4508 parent = btrfs_inode_extref_parent(eb, extref);
4509 this_name_len = btrfs_inode_extref_name_len(eb, extref);
4510 name_ptr = (unsigned long)&extref->name;
4511 this_len = sizeof(*extref) + this_name_len;
4512 }
4513
4514 if (this_name_len > name_len) {
4515 char *new_name;
4516
4517 new_name = krealloc(name, this_name_len, GFP_NOFS);
4518 if (!new_name) {
4519 ret = -ENOMEM;
4520 goto out;
4521 }
4522 name_len = this_name_len;
4523 name = new_name;
4524 }
4525
4526 read_extent_buffer(eb, name, name_ptr, this_name_len);
4527 di = btrfs_lookup_dir_item(NULL, BTRFS_I(inode)->root,
4528 search_path, parent,
4529 name, this_name_len, 0);
4530 if (di && !IS_ERR(di)) {
4531 ret = 1;
4532 goto out;
4533 } else if (IS_ERR(di)) {
4534 ret = PTR_ERR(di);
4535 goto out;
4536 }
4537 btrfs_release_path(search_path);
4538
4539 cur_offset += this_len;
4540 }
4541 ret = 0;
4542out:
4543 btrfs_free_path(search_path);
4544 kfree(name);
4545 return ret;
4546}
4547
Chris Masone02119d2008-09-05 16:13:11 -04004548/* log a single inode in the tree log.
4549 * At least one parent directory for this inode must exist in the tree
4550 * or be logged already.
4551 *
4552 * Any items from this inode changed by the current transaction are copied
4553 * to the log tree. An extra reference is taken on any extents in this
4554 * file, allowing us to avoid a whole pile of corner cases around logging
4555 * blocks that have been removed from the tree.
4556 *
4557 * See LOG_INODE_ALL and related defines for a description of what inode_only
4558 * does.
4559 *
4560 * This handles both files and directories.
4561 */
Chris Mason12fcfd22009-03-24 10:24:20 -04004562static int btrfs_log_inode(struct btrfs_trans_handle *trans,
Filipe Manana49dae1b2014-09-06 22:34:39 +01004563 struct btrfs_root *root, struct inode *inode,
4564 int inode_only,
4565 const loff_t start,
Filipe Manana8407f552014-09-05 15:14:39 +01004566 const loff_t end,
4567 struct btrfs_log_ctx *ctx)
Chris Masone02119d2008-09-05 16:13:11 -04004568{
4569 struct btrfs_path *path;
4570 struct btrfs_path *dst_path;
4571 struct btrfs_key min_key;
4572 struct btrfs_key max_key;
4573 struct btrfs_root *log = root->log_root;
Chris Mason31ff1cd2008-09-11 16:17:57 -04004574 struct extent_buffer *src = NULL;
Miao Xie827463c2014-01-14 20:31:51 +08004575 LIST_HEAD(logged_list);
Josef Bacik16e75492013-10-22 12:18:51 -04004576 u64 last_extent = 0;
Yan, Zheng4a500fd2010-05-16 10:49:59 -04004577 int err = 0;
Chris Masone02119d2008-09-05 16:13:11 -04004578 int ret;
Chris Mason3a5f1d42008-09-11 15:53:37 -04004579 int nritems;
Chris Mason31ff1cd2008-09-11 16:17:57 -04004580 int ins_start_slot = 0;
4581 int ins_nr;
Josef Bacik5dc562c2012-08-17 13:14:17 -04004582 bool fast_search = false;
Li Zefan33345d012011-04-20 10:31:50 +08004583 u64 ino = btrfs_ino(inode);
Filipe Manana49dae1b2014-09-06 22:34:39 +01004584 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
Filipe Manana1a4bcf42015-02-13 12:30:56 +00004585 u64 logged_isize = 0;
Filipe Mananae4545de2015-06-17 12:49:23 +01004586 bool need_log_inode_item = true;
Chris Masone02119d2008-09-05 16:13:11 -04004587
Chris Masone02119d2008-09-05 16:13:11 -04004588 path = btrfs_alloc_path();
Tsutomu Itoh5df67082011-02-01 09:17:35 +00004589 if (!path)
4590 return -ENOMEM;
Chris Masone02119d2008-09-05 16:13:11 -04004591 dst_path = btrfs_alloc_path();
Tsutomu Itoh5df67082011-02-01 09:17:35 +00004592 if (!dst_path) {
4593 btrfs_free_path(path);
4594 return -ENOMEM;
4595 }
Chris Masone02119d2008-09-05 16:13:11 -04004596
Li Zefan33345d012011-04-20 10:31:50 +08004597 min_key.objectid = ino;
Chris Masone02119d2008-09-05 16:13:11 -04004598 min_key.type = BTRFS_INODE_ITEM_KEY;
4599 min_key.offset = 0;
4600
Li Zefan33345d012011-04-20 10:31:50 +08004601 max_key.objectid = ino;
Chris Mason12fcfd22009-03-24 10:24:20 -04004602
Chris Mason12fcfd22009-03-24 10:24:20 -04004603
Josef Bacik5dc562c2012-08-17 13:14:17 -04004604 /* today the code can only do partial logging of directories */
Miao Xie5269b672012-11-01 07:35:23 +00004605 if (S_ISDIR(inode->i_mode) ||
4606 (!test_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4607 &BTRFS_I(inode)->runtime_flags) &&
4608 inode_only == LOG_INODE_EXISTS))
Chris Masone02119d2008-09-05 16:13:11 -04004609 max_key.type = BTRFS_XATTR_ITEM_KEY;
4610 else
4611 max_key.type = (u8)-1;
4612 max_key.offset = (u64)-1;
4613
Filipe Manana2c2c4522015-01-13 16:40:04 +00004614 /*
4615 * Only run delayed items if we are a dir or a new file.
4616 * Otherwise commit the delayed inode only, which is needed in
4617 * order for the log replay code to mark inodes for link count
4618 * fixup (create temporary BTRFS_TREE_LOG_FIXUP_OBJECTID items).
4619 */
Josef Bacik94edf4a2012-09-25 14:56:25 -04004620 if (S_ISDIR(inode->i_mode) ||
Filipe Manana2c2c4522015-01-13 16:40:04 +00004621 BTRFS_I(inode)->generation > root->fs_info->last_trans_committed)
Josef Bacik94edf4a2012-09-25 14:56:25 -04004622 ret = btrfs_commit_inode_delayed_items(trans, inode);
Filipe Manana2c2c4522015-01-13 16:40:04 +00004623 else
4624 ret = btrfs_commit_inode_delayed_inode(inode);
4625
4626 if (ret) {
4627 btrfs_free_path(path);
4628 btrfs_free_path(dst_path);
4629 return ret;
Miao Xie16cdcec2011-04-22 18:12:22 +08004630 }
4631
Chris Masone02119d2008-09-05 16:13:11 -04004632 mutex_lock(&BTRFS_I(inode)->log_mutex);
4633
Filipe Manana5e33a2b2016-02-25 23:19:38 +00004634 /*
Chris Masone02119d2008-09-05 16:13:11 -04004635 * a brute force approach to making sure we get the most uptodate
4636 * copies of everything.
4637 */
4638 if (S_ISDIR(inode->i_mode)) {
4639 int max_key_type = BTRFS_DIR_LOG_INDEX_KEY;
4640
Filipe Manana4f764e52015-02-23 19:53:35 +00004641 if (inode_only == LOG_INODE_EXISTS)
4642 max_key_type = BTRFS_XATTR_ITEM_KEY;
Li Zefan33345d012011-04-20 10:31:50 +08004643 ret = drop_objectid_items(trans, log, path, ino, max_key_type);
Chris Masone02119d2008-09-05 16:13:11 -04004644 } else {
Filipe Manana1a4bcf42015-02-13 12:30:56 +00004645 if (inode_only == LOG_INODE_EXISTS) {
4646 /*
4647 * Make sure the new inode item we write to the log has
4648 * the same isize as the current one (if it exists).
4649 * This is necessary to prevent data loss after log
4650 * replay, and also to prevent doing a wrong expanding
4651 * truncate - for e.g. create file, write 4K into offset
4652 * 0, fsync, write 4K into offset 4096, add hard link,
4653 * fsync some other file (to sync log), power fail - if
4654 * we use the inode's current i_size, after log replay
4655 * we get a 8Kb file, with the last 4Kb extent as a hole
4656 * (zeroes), as if an expanding truncate happened,
4657 * instead of getting a file of 4Kb only.
4658 */
4659 err = logged_inode_size(log, inode, path,
4660 &logged_isize);
4661 if (err)
4662 goto out_unlock;
4663 }
Filipe Mananaa7429942015-02-13 16:56:14 +00004664 if (test_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4665 &BTRFS_I(inode)->runtime_flags)) {
4666 if (inode_only == LOG_INODE_EXISTS) {
Filipe Manana4f764e52015-02-23 19:53:35 +00004667 max_key.type = BTRFS_XATTR_ITEM_KEY;
Filipe Mananaa7429942015-02-13 16:56:14 +00004668 ret = drop_objectid_items(trans, log, path, ino,
4669 max_key.type);
4670 } else {
4671 clear_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4672 &BTRFS_I(inode)->runtime_flags);
4673 clear_bit(BTRFS_INODE_COPY_EVERYTHING,
4674 &BTRFS_I(inode)->runtime_flags);
Chris Mason28ed1342014-12-17 09:41:04 -08004675 while(1) {
4676 ret = btrfs_truncate_inode_items(trans,
4677 log, inode, 0, 0);
4678 if (ret != -EAGAIN)
4679 break;
4680 }
Filipe Mananaa7429942015-02-13 16:56:14 +00004681 }
Filipe Manana4f764e52015-02-23 19:53:35 +00004682 } else if (test_and_clear_bit(BTRFS_INODE_COPY_EVERYTHING,
4683 &BTRFS_I(inode)->runtime_flags) ||
Josef Bacik6cfab852013-11-12 16:25:58 -05004684 inode_only == LOG_INODE_EXISTS) {
Filipe Manana4f764e52015-02-23 19:53:35 +00004685 if (inode_only == LOG_INODE_ALL)
Josef Bacika95249b2012-10-11 16:17:34 -04004686 fast_search = true;
Filipe Manana4f764e52015-02-23 19:53:35 +00004687 max_key.type = BTRFS_XATTR_ITEM_KEY;
Josef Bacika95249b2012-10-11 16:17:34 -04004688 ret = drop_objectid_items(trans, log, path, ino,
4689 max_key.type);
Josef Bacik5dc562c2012-08-17 13:14:17 -04004690 } else {
Liu Bo183f37f2012-11-01 06:38:47 +00004691 if (inode_only == LOG_INODE_ALL)
4692 fast_search = true;
Josef Bacika95249b2012-10-11 16:17:34 -04004693 goto log_extents;
Josef Bacik5dc562c2012-08-17 13:14:17 -04004694 }
Josef Bacika95249b2012-10-11 16:17:34 -04004695
Chris Masone02119d2008-09-05 16:13:11 -04004696 }
Yan, Zheng4a500fd2010-05-16 10:49:59 -04004697 if (ret) {
4698 err = ret;
4699 goto out_unlock;
4700 }
Chris Masone02119d2008-09-05 16:13:11 -04004701
Chris Masond3977122009-01-05 21:25:51 -05004702 while (1) {
Chris Mason31ff1cd2008-09-11 16:17:57 -04004703 ins_nr = 0;
Filipe David Borba Manana6174d3c2013-10-01 16:13:42 +01004704 ret = btrfs_search_forward(root, &min_key,
Eric Sandeende78b512013-01-31 18:21:12 +00004705 path, trans->transid);
Liu Bofb770ae2016-07-05 12:10:14 -07004706 if (ret < 0) {
4707 err = ret;
4708 goto out_unlock;
4709 }
Chris Masone02119d2008-09-05 16:13:11 -04004710 if (ret != 0)
4711 break;
Chris Mason3a5f1d42008-09-11 15:53:37 -04004712again:
Chris Mason31ff1cd2008-09-11 16:17:57 -04004713 /* note, ins_nr might be > 0 here, cleanup outside the loop */
Li Zefan33345d012011-04-20 10:31:50 +08004714 if (min_key.objectid != ino)
Chris Masone02119d2008-09-05 16:13:11 -04004715 break;
4716 if (min_key.type > max_key.type)
4717 break;
Chris Mason31ff1cd2008-09-11 16:17:57 -04004718
Filipe Mananae4545de2015-06-17 12:49:23 +01004719 if (min_key.type == BTRFS_INODE_ITEM_KEY)
4720 need_log_inode_item = false;
4721
Filipe Manana56f23fd2016-03-30 23:37:21 +01004722 if ((min_key.type == BTRFS_INODE_REF_KEY ||
4723 min_key.type == BTRFS_INODE_EXTREF_KEY) &&
4724 BTRFS_I(inode)->generation == trans->transid) {
4725 ret = btrfs_check_ref_name_override(path->nodes[0],
4726 path->slots[0],
4727 &min_key, inode);
4728 if (ret < 0) {
4729 err = ret;
4730 goto out_unlock;
4731 } else if (ret > 0) {
4732 err = 1;
4733 btrfs_set_log_full_commit(root->fs_info, trans);
4734 goto out_unlock;
4735 }
4736 }
4737
Filipe Manana36283bf2015-06-20 00:44:51 +01004738 /* Skip xattrs, we log them later with btrfs_log_all_xattrs() */
4739 if (min_key.type == BTRFS_XATTR_ITEM_KEY) {
4740 if (ins_nr == 0)
4741 goto next_slot;
4742 ret = copy_items(trans, inode, dst_path, path,
4743 &last_extent, ins_start_slot,
4744 ins_nr, inode_only, logged_isize);
4745 if (ret < 0) {
4746 err = ret;
4747 goto out_unlock;
4748 }
4749 ins_nr = 0;
4750 if (ret) {
4751 btrfs_release_path(path);
4752 continue;
4753 }
4754 goto next_slot;
4755 }
4756
Chris Masone02119d2008-09-05 16:13:11 -04004757 src = path->nodes[0];
Chris Mason31ff1cd2008-09-11 16:17:57 -04004758 if (ins_nr && ins_start_slot + ins_nr == path->slots[0]) {
4759 ins_nr++;
4760 goto next_slot;
4761 } else if (!ins_nr) {
4762 ins_start_slot = path->slots[0];
4763 ins_nr = 1;
4764 goto next_slot;
Chris Masone02119d2008-09-05 16:13:11 -04004765 }
4766
Josef Bacik16e75492013-10-22 12:18:51 -04004767 ret = copy_items(trans, inode, dst_path, path, &last_extent,
Filipe Manana1a4bcf42015-02-13 12:30:56 +00004768 ins_start_slot, ins_nr, inode_only,
4769 logged_isize);
Josef Bacik16e75492013-10-22 12:18:51 -04004770 if (ret < 0) {
Yan, Zheng4a500fd2010-05-16 10:49:59 -04004771 err = ret;
4772 goto out_unlock;
Rasmus Villemoesa71db862014-06-20 21:51:43 +02004773 }
4774 if (ret) {
Josef Bacik16e75492013-10-22 12:18:51 -04004775 ins_nr = 0;
4776 btrfs_release_path(path);
4777 continue;
Yan, Zheng4a500fd2010-05-16 10:49:59 -04004778 }
Chris Mason31ff1cd2008-09-11 16:17:57 -04004779 ins_nr = 1;
4780 ins_start_slot = path->slots[0];
4781next_slot:
Chris Masone02119d2008-09-05 16:13:11 -04004782
Chris Mason3a5f1d42008-09-11 15:53:37 -04004783 nritems = btrfs_header_nritems(path->nodes[0]);
4784 path->slots[0]++;
4785 if (path->slots[0] < nritems) {
4786 btrfs_item_key_to_cpu(path->nodes[0], &min_key,
4787 path->slots[0]);
4788 goto again;
4789 }
Chris Mason31ff1cd2008-09-11 16:17:57 -04004790 if (ins_nr) {
Josef Bacik16e75492013-10-22 12:18:51 -04004791 ret = copy_items(trans, inode, dst_path, path,
4792 &last_extent, ins_start_slot,
Filipe Manana1a4bcf42015-02-13 12:30:56 +00004793 ins_nr, inode_only, logged_isize);
Josef Bacik16e75492013-10-22 12:18:51 -04004794 if (ret < 0) {
Yan, Zheng4a500fd2010-05-16 10:49:59 -04004795 err = ret;
4796 goto out_unlock;
4797 }
Josef Bacik16e75492013-10-22 12:18:51 -04004798 ret = 0;
Chris Mason31ff1cd2008-09-11 16:17:57 -04004799 ins_nr = 0;
4800 }
David Sterbab3b4aa72011-04-21 01:20:15 +02004801 btrfs_release_path(path);
Chris Mason3a5f1d42008-09-11 15:53:37 -04004802
Filipe David Borba Manana3d41d702013-10-01 17:06:53 +01004803 if (min_key.offset < (u64)-1) {
Chris Masone02119d2008-09-05 16:13:11 -04004804 min_key.offset++;
Filipe David Borba Manana3d41d702013-10-01 17:06:53 +01004805 } else if (min_key.type < max_key.type) {
Chris Masone02119d2008-09-05 16:13:11 -04004806 min_key.type++;
Filipe David Borba Manana3d41d702013-10-01 17:06:53 +01004807 min_key.offset = 0;
4808 } else {
Chris Masone02119d2008-09-05 16:13:11 -04004809 break;
Filipe David Borba Manana3d41d702013-10-01 17:06:53 +01004810 }
Chris Masone02119d2008-09-05 16:13:11 -04004811 }
Chris Mason31ff1cd2008-09-11 16:17:57 -04004812 if (ins_nr) {
Josef Bacik16e75492013-10-22 12:18:51 -04004813 ret = copy_items(trans, inode, dst_path, path, &last_extent,
Filipe Manana1a4bcf42015-02-13 12:30:56 +00004814 ins_start_slot, ins_nr, inode_only,
4815 logged_isize);
Josef Bacik16e75492013-10-22 12:18:51 -04004816 if (ret < 0) {
Yan, Zheng4a500fd2010-05-16 10:49:59 -04004817 err = ret;
4818 goto out_unlock;
4819 }
Josef Bacik16e75492013-10-22 12:18:51 -04004820 ret = 0;
Chris Mason31ff1cd2008-09-11 16:17:57 -04004821 ins_nr = 0;
4822 }
Josef Bacik5dc562c2012-08-17 13:14:17 -04004823
Filipe Manana36283bf2015-06-20 00:44:51 +01004824 btrfs_release_path(path);
4825 btrfs_release_path(dst_path);
4826 err = btrfs_log_all_xattrs(trans, root, inode, path, dst_path);
4827 if (err)
4828 goto out_unlock;
Filipe Mananaa89ca6f2015-06-25 04:17:46 +01004829 if (max_key.type >= BTRFS_EXTENT_DATA_KEY && !fast_search) {
4830 btrfs_release_path(path);
4831 btrfs_release_path(dst_path);
4832 err = btrfs_log_trailing_hole(trans, root, inode, path);
4833 if (err)
4834 goto out_unlock;
4835 }
Josef Bacika95249b2012-10-11 16:17:34 -04004836log_extents:
Josef Bacikf3b15cc2013-07-22 12:54:30 -04004837 btrfs_release_path(path);
4838 btrfs_release_path(dst_path);
Filipe Mananae4545de2015-06-17 12:49:23 +01004839 if (need_log_inode_item) {
4840 err = log_inode_item(trans, log, dst_path, inode);
4841 if (err)
4842 goto out_unlock;
4843 }
Josef Bacik5dc562c2012-08-17 13:14:17 -04004844 if (fast_search) {
Miao Xie827463c2014-01-14 20:31:51 +08004845 ret = btrfs_log_changed_extents(trans, root, inode, dst_path,
Filipe Mananade0ee0e2016-01-21 10:17:54 +00004846 &logged_list, ctx, start, end);
Josef Bacik5dc562c2012-08-17 13:14:17 -04004847 if (ret) {
4848 err = ret;
4849 goto out_unlock;
4850 }
Josef Bacikd006a042013-11-12 20:54:09 -05004851 } else if (inode_only == LOG_INODE_ALL) {
Liu Bo06d3d222012-08-27 10:52:19 -06004852 struct extent_map *em, *n;
4853
Filipe Manana49dae1b2014-09-06 22:34:39 +01004854 write_lock(&em_tree->lock);
4855 /*
4856 * We can't just remove every em if we're called for a ranged
4857 * fsync - that is, one that doesn't cover the whole possible
4858 * file range (0 to LLONG_MAX). This is because we can have
4859 * em's that fall outside the range we're logging and therefore
4860 * their ordered operations haven't completed yet
4861 * (btrfs_finish_ordered_io() not invoked yet). This means we
4862 * didn't get their respective file extent item in the fs/subvol
4863 * tree yet, and need to let the next fast fsync (one which
4864 * consults the list of modified extent maps) find the em so
4865 * that it logs a matching file extent item and waits for the
4866 * respective ordered operation to complete (if it's still
4867 * running).
4868 *
4869 * Removing every em outside the range we're logging would make
4870 * the next fast fsync not log their matching file extent items,
4871 * therefore making us lose data after a log replay.
4872 */
4873 list_for_each_entry_safe(em, n, &em_tree->modified_extents,
4874 list) {
4875 const u64 mod_end = em->mod_start + em->mod_len - 1;
4876
4877 if (em->mod_start >= start && mod_end <= end)
4878 list_del_init(&em->list);
4879 }
4880 write_unlock(&em_tree->lock);
Josef Bacik5dc562c2012-08-17 13:14:17 -04004881 }
4882
Chris Mason9623f9a2008-09-11 17:42:42 -04004883 if (inode_only == LOG_INODE_ALL && S_ISDIR(inode->i_mode)) {
Filipe Manana2f2ff0e2015-03-20 17:19:46 +00004884 ret = log_directory_changes(trans, root, inode, path, dst_path,
4885 ctx);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04004886 if (ret) {
4887 err = ret;
4888 goto out_unlock;
4889 }
Chris Masone02119d2008-09-05 16:13:11 -04004890 }
Filipe Manana49dae1b2014-09-06 22:34:39 +01004891
Filipe Manana2f2ff0e2015-03-20 17:19:46 +00004892 spin_lock(&BTRFS_I(inode)->lock);
Filipe Manana125c4cf92014-09-11 21:22:14 +01004893 BTRFS_I(inode)->logged_trans = trans->transid;
4894 BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->last_sub_trans;
Filipe Manana2f2ff0e2015-03-20 17:19:46 +00004895 spin_unlock(&BTRFS_I(inode)->lock);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04004896out_unlock:
Miao Xie827463c2014-01-14 20:31:51 +08004897 if (unlikely(err))
4898 btrfs_put_logged_extents(&logged_list);
4899 else
4900 btrfs_submit_logged_extents(&logged_list, log);
Chris Masone02119d2008-09-05 16:13:11 -04004901 mutex_unlock(&BTRFS_I(inode)->log_mutex);
4902
4903 btrfs_free_path(path);
4904 btrfs_free_path(dst_path);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04004905 return err;
Chris Masone02119d2008-09-05 16:13:11 -04004906}
4907
Chris Mason12fcfd22009-03-24 10:24:20 -04004908/*
Filipe Manana2be63d52016-02-12 11:34:23 +00004909 * Check if we must fallback to a transaction commit when logging an inode.
4910 * This must be called after logging the inode and is used only in the context
4911 * when fsyncing an inode requires the need to log some other inode - in which
4912 * case we can't lock the i_mutex of each other inode we need to log as that
4913 * can lead to deadlocks with concurrent fsync against other inodes (as we can
4914 * log inodes up or down in the hierarchy) or rename operations for example. So
4915 * we take the log_mutex of the inode after we have logged it and then check for
4916 * its last_unlink_trans value - this is safe because any task setting
4917 * last_unlink_trans must take the log_mutex and it must do this before it does
4918 * the actual unlink operation, so if we do this check before a concurrent task
4919 * sets last_unlink_trans it means we've logged a consistent version/state of
4920 * all the inode items, otherwise we are not sure and must do a transaction
Nicholas D Steeves01327612016-05-19 21:18:45 -04004921 * commit (the concurrent task might have only updated last_unlink_trans before
Filipe Manana2be63d52016-02-12 11:34:23 +00004922 * we logged the inode or it might have also done the unlink).
4923 */
4924static bool btrfs_must_commit_transaction(struct btrfs_trans_handle *trans,
4925 struct inode *inode)
4926{
4927 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
4928 bool ret = false;
4929
4930 mutex_lock(&BTRFS_I(inode)->log_mutex);
4931 if (BTRFS_I(inode)->last_unlink_trans > fs_info->last_trans_committed) {
4932 /*
4933 * Make sure any commits to the log are forced to be full
4934 * commits.
4935 */
4936 btrfs_set_log_full_commit(fs_info, trans);
4937 ret = true;
4938 }
4939 mutex_unlock(&BTRFS_I(inode)->log_mutex);
4940
4941 return ret;
4942}
4943
4944/*
Chris Mason12fcfd22009-03-24 10:24:20 -04004945 * follow the dentry parent pointers up the chain and see if any
4946 * of the directories in it require a full commit before they can
4947 * be logged. Returns zero if nothing special needs to be done or 1 if
4948 * a full commit is required.
4949 */
4950static noinline int check_parent_dirs_for_sync(struct btrfs_trans_handle *trans,
4951 struct inode *inode,
4952 struct dentry *parent,
4953 struct super_block *sb,
4954 u64 last_committed)
Chris Masone02119d2008-09-05 16:13:11 -04004955{
Chris Mason12fcfd22009-03-24 10:24:20 -04004956 int ret = 0;
Josef Bacik6a912212010-11-20 09:48:00 +00004957 struct dentry *old_parent = NULL;
Josef Bacikde2b5302013-09-11 09:36:30 -04004958 struct inode *orig_inode = inode;
Chris Masone02119d2008-09-05 16:13:11 -04004959
Chris Masonaf4176b2009-03-24 10:24:31 -04004960 /*
4961 * for regular files, if its inode is already on disk, we don't
4962 * have to worry about the parents at all. This is because
4963 * we can use the last_unlink_trans field to record renames
4964 * and other fun in this file.
4965 */
4966 if (S_ISREG(inode->i_mode) &&
4967 BTRFS_I(inode)->generation <= last_committed &&
4968 BTRFS_I(inode)->last_unlink_trans <= last_committed)
4969 goto out;
4970
Chris Mason12fcfd22009-03-24 10:24:20 -04004971 if (!S_ISDIR(inode->i_mode)) {
Al Virofc640052016-04-10 01:33:30 -04004972 if (!parent || d_really_is_negative(parent) || sb != parent->d_sb)
Chris Mason12fcfd22009-03-24 10:24:20 -04004973 goto out;
David Howells2b0143b2015-03-17 22:25:59 +00004974 inode = d_inode(parent);
Chris Mason12fcfd22009-03-24 10:24:20 -04004975 }
4976
4977 while (1) {
Josef Bacikde2b5302013-09-11 09:36:30 -04004978 /*
4979 * If we are logging a directory then we start with our inode,
Nicholas D Steeves01327612016-05-19 21:18:45 -04004980 * not our parent's inode, so we need to skip setting the
Josef Bacikde2b5302013-09-11 09:36:30 -04004981 * logged_trans so that further down in the log code we don't
4982 * think this inode has already been logged.
4983 */
4984 if (inode != orig_inode)
4985 BTRFS_I(inode)->logged_trans = trans->transid;
Chris Mason12fcfd22009-03-24 10:24:20 -04004986 smp_mb();
4987
Filipe Manana2be63d52016-02-12 11:34:23 +00004988 if (btrfs_must_commit_transaction(trans, inode)) {
Chris Mason12fcfd22009-03-24 10:24:20 -04004989 ret = 1;
4990 break;
4991 }
4992
Al Virofc640052016-04-10 01:33:30 -04004993 if (!parent || d_really_is_negative(parent) || sb != parent->d_sb)
Chris Mason12fcfd22009-03-24 10:24:20 -04004994 break;
4995
Yan, Zheng76dda932009-09-21 16:00:26 -04004996 if (IS_ROOT(parent))
Chris Mason12fcfd22009-03-24 10:24:20 -04004997 break;
4998
Josef Bacik6a912212010-11-20 09:48:00 +00004999 parent = dget_parent(parent);
5000 dput(old_parent);
5001 old_parent = parent;
David Howells2b0143b2015-03-17 22:25:59 +00005002 inode = d_inode(parent);
Chris Mason12fcfd22009-03-24 10:24:20 -04005003
5004 }
Josef Bacik6a912212010-11-20 09:48:00 +00005005 dput(old_parent);
Chris Mason12fcfd22009-03-24 10:24:20 -04005006out:
Chris Masone02119d2008-09-05 16:13:11 -04005007 return ret;
5008}
5009
Filipe Manana2f2ff0e2015-03-20 17:19:46 +00005010struct btrfs_dir_list {
5011 u64 ino;
5012 struct list_head list;
5013};
5014
5015/*
5016 * Log the inodes of the new dentries of a directory. See log_dir_items() for
5017 * details about the why it is needed.
5018 * This is a recursive operation - if an existing dentry corresponds to a
5019 * directory, that directory's new entries are logged too (same behaviour as
5020 * ext3/4, xfs, f2fs, reiserfs, nilfs2). Note that when logging the inodes
5021 * the dentries point to we do not lock their i_mutex, otherwise lockdep
5022 * complains about the following circular lock dependency / possible deadlock:
5023 *
5024 * CPU0 CPU1
5025 * ---- ----
5026 * lock(&type->i_mutex_dir_key#3/2);
5027 * lock(sb_internal#2);
5028 * lock(&type->i_mutex_dir_key#3/2);
5029 * lock(&sb->s_type->i_mutex_key#14);
5030 *
5031 * Where sb_internal is the lock (a counter that works as a lock) acquired by
5032 * sb_start_intwrite() in btrfs_start_transaction().
5033 * Not locking i_mutex of the inodes is still safe because:
5034 *
5035 * 1) For regular files we log with a mode of LOG_INODE_EXISTS. It's possible
5036 * that while logging the inode new references (names) are added or removed
5037 * from the inode, leaving the logged inode item with a link count that does
5038 * not match the number of logged inode reference items. This is fine because
5039 * at log replay time we compute the real number of links and correct the
5040 * link count in the inode item (see replay_one_buffer() and
5041 * link_to_fixup_dir());
5042 *
5043 * 2) For directories we log with a mode of LOG_INODE_ALL. It's possible that
5044 * while logging the inode's items new items with keys BTRFS_DIR_ITEM_KEY and
5045 * BTRFS_DIR_INDEX_KEY are added to fs/subvol tree and the logged inode item
5046 * has a size that doesn't match the sum of the lengths of all the logged
5047 * names. This does not result in a problem because if a dir_item key is
5048 * logged but its matching dir_index key is not logged, at log replay time we
5049 * don't use it to replay the respective name (see replay_one_name()). On the
5050 * other hand if only the dir_index key ends up being logged, the respective
5051 * name is added to the fs/subvol tree with both the dir_item and dir_index
5052 * keys created (see replay_one_name()).
5053 * The directory's inode item with a wrong i_size is not a problem as well,
5054 * since we don't use it at log replay time to set the i_size in the inode
5055 * item of the fs/subvol tree (see overwrite_item()).
5056 */
5057static int log_new_dir_dentries(struct btrfs_trans_handle *trans,
5058 struct btrfs_root *root,
5059 struct inode *start_inode,
5060 struct btrfs_log_ctx *ctx)
5061{
5062 struct btrfs_root *log = root->log_root;
5063 struct btrfs_path *path;
5064 LIST_HEAD(dir_list);
5065 struct btrfs_dir_list *dir_elem;
5066 int ret = 0;
5067
5068 path = btrfs_alloc_path();
5069 if (!path)
5070 return -ENOMEM;
5071
5072 dir_elem = kmalloc(sizeof(*dir_elem), GFP_NOFS);
5073 if (!dir_elem) {
5074 btrfs_free_path(path);
5075 return -ENOMEM;
5076 }
5077 dir_elem->ino = btrfs_ino(start_inode);
5078 list_add_tail(&dir_elem->list, &dir_list);
5079
5080 while (!list_empty(&dir_list)) {
5081 struct extent_buffer *leaf;
5082 struct btrfs_key min_key;
5083 int nritems;
5084 int i;
5085
5086 dir_elem = list_first_entry(&dir_list, struct btrfs_dir_list,
5087 list);
5088 if (ret)
5089 goto next_dir_inode;
5090
5091 min_key.objectid = dir_elem->ino;
5092 min_key.type = BTRFS_DIR_ITEM_KEY;
5093 min_key.offset = 0;
5094again:
5095 btrfs_release_path(path);
5096 ret = btrfs_search_forward(log, &min_key, path, trans->transid);
5097 if (ret < 0) {
5098 goto next_dir_inode;
5099 } else if (ret > 0) {
5100 ret = 0;
5101 goto next_dir_inode;
5102 }
5103
5104process_leaf:
5105 leaf = path->nodes[0];
5106 nritems = btrfs_header_nritems(leaf);
5107 for (i = path->slots[0]; i < nritems; i++) {
5108 struct btrfs_dir_item *di;
5109 struct btrfs_key di_key;
5110 struct inode *di_inode;
5111 struct btrfs_dir_list *new_dir_elem;
5112 int log_mode = LOG_INODE_EXISTS;
5113 int type;
5114
5115 btrfs_item_key_to_cpu(leaf, &min_key, i);
5116 if (min_key.objectid != dir_elem->ino ||
5117 min_key.type != BTRFS_DIR_ITEM_KEY)
5118 goto next_dir_inode;
5119
5120 di = btrfs_item_ptr(leaf, i, struct btrfs_dir_item);
5121 type = btrfs_dir_type(leaf, di);
5122 if (btrfs_dir_transid(leaf, di) < trans->transid &&
5123 type != BTRFS_FT_DIR)
5124 continue;
5125 btrfs_dir_item_key_to_cpu(leaf, di, &di_key);
5126 if (di_key.type == BTRFS_ROOT_ITEM_KEY)
5127 continue;
5128
5129 di_inode = btrfs_iget(root->fs_info->sb, &di_key,
5130 root, NULL);
5131 if (IS_ERR(di_inode)) {
5132 ret = PTR_ERR(di_inode);
5133 goto next_dir_inode;
5134 }
5135
5136 if (btrfs_inode_in_log(di_inode, trans->transid)) {
5137 iput(di_inode);
5138 continue;
5139 }
5140
5141 ctx->log_new_dentries = false;
Filipe Manana3f9749f2016-04-25 04:45:02 +01005142 if (type == BTRFS_FT_DIR || type == BTRFS_FT_SYMLINK)
Filipe Manana2f2ff0e2015-03-20 17:19:46 +00005143 log_mode = LOG_INODE_ALL;
5144 btrfs_release_path(path);
5145 ret = btrfs_log_inode(trans, root, di_inode,
5146 log_mode, 0, LLONG_MAX, ctx);
Filipe Manana2be63d52016-02-12 11:34:23 +00005147 if (!ret &&
5148 btrfs_must_commit_transaction(trans, di_inode))
5149 ret = 1;
Filipe Manana2f2ff0e2015-03-20 17:19:46 +00005150 iput(di_inode);
5151 if (ret)
5152 goto next_dir_inode;
5153 if (ctx->log_new_dentries) {
5154 new_dir_elem = kmalloc(sizeof(*new_dir_elem),
5155 GFP_NOFS);
5156 if (!new_dir_elem) {
5157 ret = -ENOMEM;
5158 goto next_dir_inode;
5159 }
5160 new_dir_elem->ino = di_key.objectid;
5161 list_add_tail(&new_dir_elem->list, &dir_list);
5162 }
5163 break;
5164 }
5165 if (i == nritems) {
5166 ret = btrfs_next_leaf(log, path);
5167 if (ret < 0) {
5168 goto next_dir_inode;
5169 } else if (ret > 0) {
5170 ret = 0;
5171 goto next_dir_inode;
5172 }
5173 goto process_leaf;
5174 }
5175 if (min_key.offset < (u64)-1) {
5176 min_key.offset++;
5177 goto again;
5178 }
5179next_dir_inode:
5180 list_del(&dir_elem->list);
5181 kfree(dir_elem);
5182 }
5183
5184 btrfs_free_path(path);
5185 return ret;
5186}
5187
Filipe Manana18aa0922015-08-05 16:49:08 +01005188static int btrfs_log_all_parents(struct btrfs_trans_handle *trans,
5189 struct inode *inode,
5190 struct btrfs_log_ctx *ctx)
5191{
5192 int ret;
5193 struct btrfs_path *path;
5194 struct btrfs_key key;
5195 struct btrfs_root *root = BTRFS_I(inode)->root;
5196 const u64 ino = btrfs_ino(inode);
5197
5198 path = btrfs_alloc_path();
5199 if (!path)
5200 return -ENOMEM;
5201 path->skip_locking = 1;
5202 path->search_commit_root = 1;
5203
5204 key.objectid = ino;
5205 key.type = BTRFS_INODE_REF_KEY;
5206 key.offset = 0;
5207 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5208 if (ret < 0)
5209 goto out;
5210
5211 while (true) {
5212 struct extent_buffer *leaf = path->nodes[0];
5213 int slot = path->slots[0];
5214 u32 cur_offset = 0;
5215 u32 item_size;
5216 unsigned long ptr;
5217
5218 if (slot >= btrfs_header_nritems(leaf)) {
5219 ret = btrfs_next_leaf(root, path);
5220 if (ret < 0)
5221 goto out;
5222 else if (ret > 0)
5223 break;
5224 continue;
5225 }
5226
5227 btrfs_item_key_to_cpu(leaf, &key, slot);
5228 /* BTRFS_INODE_EXTREF_KEY is BTRFS_INODE_REF_KEY + 1 */
5229 if (key.objectid != ino || key.type > BTRFS_INODE_EXTREF_KEY)
5230 break;
5231
5232 item_size = btrfs_item_size_nr(leaf, slot);
5233 ptr = btrfs_item_ptr_offset(leaf, slot);
5234 while (cur_offset < item_size) {
5235 struct btrfs_key inode_key;
5236 struct inode *dir_inode;
5237
5238 inode_key.type = BTRFS_INODE_ITEM_KEY;
5239 inode_key.offset = 0;
5240
5241 if (key.type == BTRFS_INODE_EXTREF_KEY) {
5242 struct btrfs_inode_extref *extref;
5243
5244 extref = (struct btrfs_inode_extref *)
5245 (ptr + cur_offset);
5246 inode_key.objectid = btrfs_inode_extref_parent(
5247 leaf, extref);
5248 cur_offset += sizeof(*extref);
5249 cur_offset += btrfs_inode_extref_name_len(leaf,
5250 extref);
5251 } else {
5252 inode_key.objectid = key.offset;
5253 cur_offset = item_size;
5254 }
5255
5256 dir_inode = btrfs_iget(root->fs_info->sb, &inode_key,
5257 root, NULL);
5258 /* If parent inode was deleted, skip it. */
5259 if (IS_ERR(dir_inode))
5260 continue;
5261
Filipe Manana657ed1a2016-04-06 17:11:56 +01005262 if (ctx)
5263 ctx->log_new_dentries = false;
Filipe Manana18aa0922015-08-05 16:49:08 +01005264 ret = btrfs_log_inode(trans, root, dir_inode,
5265 LOG_INODE_ALL, 0, LLONG_MAX, ctx);
Filipe Manana2be63d52016-02-12 11:34:23 +00005266 if (!ret &&
5267 btrfs_must_commit_transaction(trans, dir_inode))
5268 ret = 1;
Filipe Manana657ed1a2016-04-06 17:11:56 +01005269 if (!ret && ctx && ctx->log_new_dentries)
5270 ret = log_new_dir_dentries(trans, root,
5271 dir_inode, ctx);
Filipe Manana18aa0922015-08-05 16:49:08 +01005272 iput(dir_inode);
5273 if (ret)
5274 goto out;
5275 }
5276 path->slots[0]++;
5277 }
5278 ret = 0;
5279out:
5280 btrfs_free_path(path);
5281 return ret;
5282}
5283
Chris Masone02119d2008-09-05 16:13:11 -04005284/*
5285 * helper function around btrfs_log_inode to make sure newly created
5286 * parent directories also end up in the log. A minimal inode and backref
5287 * only logging is done of any parent directories that are older than
5288 * the last committed transaction
5289 */
Eric Sandeen48a3b632013-04-25 20:41:01 +00005290static int btrfs_log_inode_parent(struct btrfs_trans_handle *trans,
5291 struct btrfs_root *root, struct inode *inode,
Filipe Manana49dae1b2014-09-06 22:34:39 +01005292 struct dentry *parent,
5293 const loff_t start,
5294 const loff_t end,
5295 int exists_only,
Miao Xie8b050d32014-02-20 18:08:58 +08005296 struct btrfs_log_ctx *ctx)
Chris Masone02119d2008-09-05 16:13:11 -04005297{
Chris Mason12fcfd22009-03-24 10:24:20 -04005298 int inode_only = exists_only ? LOG_INODE_EXISTS : LOG_INODE_ALL;
Chris Masone02119d2008-09-05 16:13:11 -04005299 struct super_block *sb;
Josef Bacik6a912212010-11-20 09:48:00 +00005300 struct dentry *old_parent = NULL;
Chris Mason12fcfd22009-03-24 10:24:20 -04005301 int ret = 0;
5302 u64 last_committed = root->fs_info->last_trans_committed;
Filipe Manana2f2ff0e2015-03-20 17:19:46 +00005303 bool log_dentries = false;
5304 struct inode *orig_inode = inode;
Chris Mason12fcfd22009-03-24 10:24:20 -04005305
5306 sb = inode->i_sb;
5307
Jeff Mahoney3cdde222016-06-09 21:38:35 -04005308 if (btrfs_test_opt(root->fs_info, NOTREELOG)) {
Sage Weil3a5e1402009-04-02 16:49:40 -04005309 ret = 1;
5310 goto end_no_trans;
5311 }
5312
Miao Xie995946d2014-04-02 19:51:06 +08005313 /*
5314 * The prev transaction commit doesn't complete, we need do
5315 * full commit by ourselves.
5316 */
Chris Mason12fcfd22009-03-24 10:24:20 -04005317 if (root->fs_info->last_trans_log_full_commit >
5318 root->fs_info->last_trans_committed) {
5319 ret = 1;
5320 goto end_no_trans;
5321 }
5322
Yan, Zheng76dda932009-09-21 16:00:26 -04005323 if (root != BTRFS_I(inode)->root ||
5324 btrfs_root_refs(&root->root_item) == 0) {
5325 ret = 1;
5326 goto end_no_trans;
5327 }
5328
Chris Mason12fcfd22009-03-24 10:24:20 -04005329 ret = check_parent_dirs_for_sync(trans, inode, parent,
5330 sb, last_committed);
5331 if (ret)
5332 goto end_no_trans;
Chris Masone02119d2008-09-05 16:13:11 -04005333
Josef Bacik22ee6982012-05-29 16:57:49 -04005334 if (btrfs_inode_in_log(inode, trans->transid)) {
Chris Mason257c62e2009-10-13 13:21:08 -04005335 ret = BTRFS_NO_LOG_SYNC;
5336 goto end_no_trans;
5337 }
5338
Miao Xie8b050d32014-02-20 18:08:58 +08005339 ret = start_log_trans(trans, root, ctx);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04005340 if (ret)
Miao Xiee87ac132014-02-20 18:08:53 +08005341 goto end_no_trans;
Chris Mason12fcfd22009-03-24 10:24:20 -04005342
Filipe Manana8407f552014-09-05 15:14:39 +01005343 ret = btrfs_log_inode(trans, root, inode, inode_only, start, end, ctx);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04005344 if (ret)
5345 goto end_trans;
Chris Mason12fcfd22009-03-24 10:24:20 -04005346
Chris Masonaf4176b2009-03-24 10:24:31 -04005347 /*
5348 * for regular files, if its inode is already on disk, we don't
5349 * have to worry about the parents at all. This is because
5350 * we can use the last_unlink_trans field to record renames
5351 * and other fun in this file.
5352 */
5353 if (S_ISREG(inode->i_mode) &&
5354 BTRFS_I(inode)->generation <= last_committed &&
Yan, Zheng4a500fd2010-05-16 10:49:59 -04005355 BTRFS_I(inode)->last_unlink_trans <= last_committed) {
5356 ret = 0;
5357 goto end_trans;
5358 }
Chris Masonaf4176b2009-03-24 10:24:31 -04005359
Filipe Manana2f2ff0e2015-03-20 17:19:46 +00005360 if (S_ISDIR(inode->i_mode) && ctx && ctx->log_new_dentries)
5361 log_dentries = true;
5362
Filipe Manana18aa0922015-08-05 16:49:08 +01005363 /*
Nicholas D Steeves01327612016-05-19 21:18:45 -04005364 * On unlink we must make sure all our current and old parent directory
Filipe Manana18aa0922015-08-05 16:49:08 +01005365 * inodes are fully logged. This is to prevent leaving dangling
5366 * directory index entries in directories that were our parents but are
5367 * not anymore. Not doing this results in old parent directory being
5368 * impossible to delete after log replay (rmdir will always fail with
5369 * error -ENOTEMPTY).
5370 *
5371 * Example 1:
5372 *
5373 * mkdir testdir
5374 * touch testdir/foo
5375 * ln testdir/foo testdir/bar
5376 * sync
5377 * unlink testdir/bar
5378 * xfs_io -c fsync testdir/foo
5379 * <power failure>
5380 * mount fs, triggers log replay
5381 *
5382 * If we don't log the parent directory (testdir), after log replay the
5383 * directory still has an entry pointing to the file inode using the bar
5384 * name, but a matching BTRFS_INODE_[REF|EXTREF]_KEY does not exist and
5385 * the file inode has a link count of 1.
5386 *
5387 * Example 2:
5388 *
5389 * mkdir testdir
5390 * touch foo
5391 * ln foo testdir/foo2
5392 * ln foo testdir/foo3
5393 * sync
5394 * unlink testdir/foo3
5395 * xfs_io -c fsync foo
5396 * <power failure>
5397 * mount fs, triggers log replay
5398 *
5399 * Similar as the first example, after log replay the parent directory
5400 * testdir still has an entry pointing to the inode file with name foo3
5401 * but the file inode does not have a matching BTRFS_INODE_REF_KEY item
5402 * and has a link count of 2.
5403 */
5404 if (BTRFS_I(inode)->last_unlink_trans > last_committed) {
5405 ret = btrfs_log_all_parents(trans, orig_inode, ctx);
5406 if (ret)
5407 goto end_trans;
5408 }
5409
Chris Masond3977122009-01-05 21:25:51 -05005410 while (1) {
Al Virofc640052016-04-10 01:33:30 -04005411 if (!parent || d_really_is_negative(parent) || sb != parent->d_sb)
Chris Masone02119d2008-09-05 16:13:11 -04005412 break;
5413
David Howells2b0143b2015-03-17 22:25:59 +00005414 inode = d_inode(parent);
Yan, Zheng76dda932009-09-21 16:00:26 -04005415 if (root != BTRFS_I(inode)->root)
5416 break;
5417
Filipe Manana18aa0922015-08-05 16:49:08 +01005418 if (BTRFS_I(inode)->generation > last_committed) {
5419 ret = btrfs_log_inode(trans, root, inode,
5420 LOG_INODE_EXISTS,
Filipe Manana8407f552014-09-05 15:14:39 +01005421 0, LLONG_MAX, ctx);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04005422 if (ret)
5423 goto end_trans;
Chris Mason12fcfd22009-03-24 10:24:20 -04005424 }
Yan, Zheng76dda932009-09-21 16:00:26 -04005425 if (IS_ROOT(parent))
Chris Masone02119d2008-09-05 16:13:11 -04005426 break;
Chris Mason12fcfd22009-03-24 10:24:20 -04005427
Josef Bacik6a912212010-11-20 09:48:00 +00005428 parent = dget_parent(parent);
5429 dput(old_parent);
5430 old_parent = parent;
Chris Masone02119d2008-09-05 16:13:11 -04005431 }
Filipe Manana2f2ff0e2015-03-20 17:19:46 +00005432 if (log_dentries)
5433 ret = log_new_dir_dentries(trans, root, orig_inode, ctx);
5434 else
5435 ret = 0;
Yan, Zheng4a500fd2010-05-16 10:49:59 -04005436end_trans:
Josef Bacik6a912212010-11-20 09:48:00 +00005437 dput(old_parent);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04005438 if (ret < 0) {
Miao Xie995946d2014-04-02 19:51:06 +08005439 btrfs_set_log_full_commit(root->fs_info, trans);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04005440 ret = 1;
5441 }
Miao Xie8b050d32014-02-20 18:08:58 +08005442
5443 if (ret)
5444 btrfs_remove_log_ctx(root, ctx);
Chris Mason12fcfd22009-03-24 10:24:20 -04005445 btrfs_end_log_trans(root);
5446end_no_trans:
5447 return ret;
Chris Masone02119d2008-09-05 16:13:11 -04005448}
5449
5450/*
5451 * it is not safe to log dentry if the chunk root has added new
5452 * chunks. This returns 0 if the dentry was logged, and 1 otherwise.
5453 * If this returns 1, you must commit the transaction to safely get your
5454 * data on disk.
5455 */
5456int btrfs_log_dentry_safe(struct btrfs_trans_handle *trans,
Miao Xie8b050d32014-02-20 18:08:58 +08005457 struct btrfs_root *root, struct dentry *dentry,
Filipe Manana49dae1b2014-09-06 22:34:39 +01005458 const loff_t start,
5459 const loff_t end,
Miao Xie8b050d32014-02-20 18:08:58 +08005460 struct btrfs_log_ctx *ctx)
Chris Masone02119d2008-09-05 16:13:11 -04005461{
Josef Bacik6a912212010-11-20 09:48:00 +00005462 struct dentry *parent = dget_parent(dentry);
5463 int ret;
5464
David Howells2b0143b2015-03-17 22:25:59 +00005465 ret = btrfs_log_inode_parent(trans, root, d_inode(dentry), parent,
Filipe Manana49dae1b2014-09-06 22:34:39 +01005466 start, end, 0, ctx);
Josef Bacik6a912212010-11-20 09:48:00 +00005467 dput(parent);
5468
5469 return ret;
Chris Masone02119d2008-09-05 16:13:11 -04005470}
5471
5472/*
5473 * should be called during mount to recover any replay any log trees
5474 * from the FS
5475 */
5476int btrfs_recover_log_trees(struct btrfs_root *log_root_tree)
5477{
5478 int ret;
5479 struct btrfs_path *path;
5480 struct btrfs_trans_handle *trans;
5481 struct btrfs_key key;
5482 struct btrfs_key found_key;
5483 struct btrfs_key tmp_key;
5484 struct btrfs_root *log;
5485 struct btrfs_fs_info *fs_info = log_root_tree->fs_info;
5486 struct walk_control wc = {
5487 .process_func = process_one_buffer,
5488 .stage = 0,
5489 };
5490
Chris Masone02119d2008-09-05 16:13:11 -04005491 path = btrfs_alloc_path();
Tsutomu Itohdb5b4932011-03-23 08:14:16 +00005492 if (!path)
5493 return -ENOMEM;
5494
5495 fs_info->log_root_recovering = 1;
Chris Masone02119d2008-09-05 16:13:11 -04005496
Yan, Zheng4a500fd2010-05-16 10:49:59 -04005497 trans = btrfs_start_transaction(fs_info->tree_root, 0);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01005498 if (IS_ERR(trans)) {
5499 ret = PTR_ERR(trans);
5500 goto error;
5501 }
Chris Masone02119d2008-09-05 16:13:11 -04005502
5503 wc.trans = trans;
5504 wc.pin = 1;
5505
Tsutomu Itohdb5b4932011-03-23 08:14:16 +00005506 ret = walk_log_tree(trans, log_root_tree, &wc);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01005507 if (ret) {
Anand Jain34d97002016-03-16 16:43:06 +08005508 btrfs_handle_fs_error(fs_info, ret, "Failed to pin buffers while "
Jeff Mahoney79787ea2012-03-12 16:03:00 +01005509 "recovering log root tree.");
5510 goto error;
5511 }
Chris Masone02119d2008-09-05 16:13:11 -04005512
5513again:
5514 key.objectid = BTRFS_TREE_LOG_OBJECTID;
5515 key.offset = (u64)-1;
David Sterba962a2982014-06-04 18:41:45 +02005516 key.type = BTRFS_ROOT_ITEM_KEY;
Chris Masone02119d2008-09-05 16:13:11 -04005517
Chris Masond3977122009-01-05 21:25:51 -05005518 while (1) {
Chris Masone02119d2008-09-05 16:13:11 -04005519 ret = btrfs_search_slot(NULL, log_root_tree, &key, path, 0, 0);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01005520
5521 if (ret < 0) {
Anand Jain34d97002016-03-16 16:43:06 +08005522 btrfs_handle_fs_error(fs_info, ret,
Jeff Mahoney79787ea2012-03-12 16:03:00 +01005523 "Couldn't find tree log root.");
5524 goto error;
5525 }
Chris Masone02119d2008-09-05 16:13:11 -04005526 if (ret > 0) {
5527 if (path->slots[0] == 0)
5528 break;
5529 path->slots[0]--;
5530 }
5531 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
5532 path->slots[0]);
David Sterbab3b4aa72011-04-21 01:20:15 +02005533 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04005534 if (found_key.objectid != BTRFS_TREE_LOG_OBJECTID)
5535 break;
5536
Miao Xiecb517ea2013-05-15 07:48:19 +00005537 log = btrfs_read_fs_root(log_root_tree, &found_key);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01005538 if (IS_ERR(log)) {
5539 ret = PTR_ERR(log);
Anand Jain34d97002016-03-16 16:43:06 +08005540 btrfs_handle_fs_error(fs_info, ret,
Jeff Mahoney79787ea2012-03-12 16:03:00 +01005541 "Couldn't read tree log root.");
5542 goto error;
5543 }
Chris Masone02119d2008-09-05 16:13:11 -04005544
5545 tmp_key.objectid = found_key.offset;
5546 tmp_key.type = BTRFS_ROOT_ITEM_KEY;
5547 tmp_key.offset = (u64)-1;
5548
5549 wc.replay_dest = btrfs_read_fs_root_no_name(fs_info, &tmp_key);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01005550 if (IS_ERR(wc.replay_dest)) {
5551 ret = PTR_ERR(wc.replay_dest);
Josef Bacikb50c6e22013-04-25 15:55:30 -04005552 free_extent_buffer(log->node);
5553 free_extent_buffer(log->commit_root);
5554 kfree(log);
Anand Jain34d97002016-03-16 16:43:06 +08005555 btrfs_handle_fs_error(fs_info, ret, "Couldn't read target root "
Jeff Mahoney79787ea2012-03-12 16:03:00 +01005556 "for tree log recovery.");
5557 goto error;
5558 }
Chris Masone02119d2008-09-05 16:13:11 -04005559
Yan Zheng07d400a2009-01-06 11:42:00 -05005560 wc.replay_dest->log_root = log;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005561 btrfs_record_root_in_trans(trans, wc.replay_dest);
Chris Masone02119d2008-09-05 16:13:11 -04005562 ret = walk_log_tree(trans, log, &wc);
Chris Masone02119d2008-09-05 16:13:11 -04005563
Josef Bacikb50c6e22013-04-25 15:55:30 -04005564 if (!ret && wc.stage == LOG_WALK_REPLAY_ALL) {
Chris Masone02119d2008-09-05 16:13:11 -04005565 ret = fixup_inode_link_counts(trans, wc.replay_dest,
5566 path);
Chris Masone02119d2008-09-05 16:13:11 -04005567 }
Chris Masone02119d2008-09-05 16:13:11 -04005568
5569 key.offset = found_key.offset - 1;
Yan Zheng07d400a2009-01-06 11:42:00 -05005570 wc.replay_dest->log_root = NULL;
Chris Masone02119d2008-09-05 16:13:11 -04005571 free_extent_buffer(log->node);
Chris Masonb263c2c2009-06-11 11:24:47 -04005572 free_extent_buffer(log->commit_root);
Chris Masone02119d2008-09-05 16:13:11 -04005573 kfree(log);
5574
Josef Bacikb50c6e22013-04-25 15:55:30 -04005575 if (ret)
5576 goto error;
5577
Chris Masone02119d2008-09-05 16:13:11 -04005578 if (found_key.offset == 0)
5579 break;
5580 }
David Sterbab3b4aa72011-04-21 01:20:15 +02005581 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04005582
5583 /* step one is to pin it all, step two is to replay just inodes */
5584 if (wc.pin) {
5585 wc.pin = 0;
5586 wc.process_func = replay_one_buffer;
5587 wc.stage = LOG_WALK_REPLAY_INODES;
5588 goto again;
5589 }
5590 /* step three is to replay everything */
5591 if (wc.stage < LOG_WALK_REPLAY_ALL) {
5592 wc.stage++;
5593 goto again;
5594 }
5595
5596 btrfs_free_path(path);
5597
Josef Bacikabefa552013-04-24 16:40:05 -04005598 /* step 4: commit the transaction, which also unpins the blocks */
5599 ret = btrfs_commit_transaction(trans, fs_info->tree_root);
5600 if (ret)
5601 return ret;
5602
Chris Masone02119d2008-09-05 16:13:11 -04005603 free_extent_buffer(log_root_tree->node);
5604 log_root_tree->log_root = NULL;
5605 fs_info->log_root_recovering = 0;
Chris Masone02119d2008-09-05 16:13:11 -04005606 kfree(log_root_tree);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01005607
Josef Bacikabefa552013-04-24 16:40:05 -04005608 return 0;
Jeff Mahoney79787ea2012-03-12 16:03:00 +01005609error:
Josef Bacikb50c6e22013-04-25 15:55:30 -04005610 if (wc.trans)
5611 btrfs_end_transaction(wc.trans, fs_info->tree_root);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01005612 btrfs_free_path(path);
5613 return ret;
Chris Masone02119d2008-09-05 16:13:11 -04005614}
Chris Mason12fcfd22009-03-24 10:24:20 -04005615
5616/*
5617 * there are some corner cases where we want to force a full
5618 * commit instead of allowing a directory to be logged.
5619 *
5620 * They revolve around files there were unlinked from the directory, and
5621 * this function updates the parent directory so that a full commit is
5622 * properly done if it is fsync'd later after the unlinks are done.
Filipe Manana2be63d52016-02-12 11:34:23 +00005623 *
5624 * Must be called before the unlink operations (updates to the subvolume tree,
5625 * inodes, etc) are done.
Chris Mason12fcfd22009-03-24 10:24:20 -04005626 */
5627void btrfs_record_unlink_dir(struct btrfs_trans_handle *trans,
5628 struct inode *dir, struct inode *inode,
5629 int for_rename)
5630{
5631 /*
Chris Masonaf4176b2009-03-24 10:24:31 -04005632 * when we're logging a file, if it hasn't been renamed
5633 * or unlinked, and its inode is fully committed on disk,
5634 * we don't have to worry about walking up the directory chain
5635 * to log its parents.
5636 *
5637 * So, we use the last_unlink_trans field to put this transid
5638 * into the file. When the file is logged we check it and
5639 * don't log the parents if the file is fully on disk.
5640 */
Filipe Manana657ed1a2016-04-06 17:11:56 +01005641 mutex_lock(&BTRFS_I(inode)->log_mutex);
5642 BTRFS_I(inode)->last_unlink_trans = trans->transid;
5643 mutex_unlock(&BTRFS_I(inode)->log_mutex);
Chris Masonaf4176b2009-03-24 10:24:31 -04005644
5645 /*
Chris Mason12fcfd22009-03-24 10:24:20 -04005646 * if this directory was already logged any new
5647 * names for this file/dir will get recorded
5648 */
5649 smp_mb();
5650 if (BTRFS_I(dir)->logged_trans == trans->transid)
5651 return;
5652
5653 /*
5654 * if the inode we're about to unlink was logged,
5655 * the log will be properly updated for any new names
5656 */
5657 if (BTRFS_I(inode)->logged_trans == trans->transid)
5658 return;
5659
5660 /*
5661 * when renaming files across directories, if the directory
5662 * there we're unlinking from gets fsync'd later on, there's
5663 * no way to find the destination directory later and fsync it
5664 * properly. So, we have to be conservative and force commits
5665 * so the new name gets discovered.
5666 */
5667 if (for_rename)
5668 goto record;
5669
5670 /* we can safely do the unlink without any special recording */
5671 return;
5672
5673record:
Filipe Manana2be63d52016-02-12 11:34:23 +00005674 mutex_lock(&BTRFS_I(dir)->log_mutex);
Chris Mason12fcfd22009-03-24 10:24:20 -04005675 BTRFS_I(dir)->last_unlink_trans = trans->transid;
Filipe Manana2be63d52016-02-12 11:34:23 +00005676 mutex_unlock(&BTRFS_I(dir)->log_mutex);
Chris Mason12fcfd22009-03-24 10:24:20 -04005677}
5678
5679/*
Filipe Manana1ec9a1a2016-02-10 10:42:25 +00005680 * Make sure that if someone attempts to fsync the parent directory of a deleted
5681 * snapshot, it ends up triggering a transaction commit. This is to guarantee
5682 * that after replaying the log tree of the parent directory's root we will not
5683 * see the snapshot anymore and at log replay time we will not see any log tree
5684 * corresponding to the deleted snapshot's root, which could lead to replaying
5685 * it after replaying the log tree of the parent directory (which would replay
5686 * the snapshot delete operation).
Filipe Manana2be63d52016-02-12 11:34:23 +00005687 *
5688 * Must be called before the actual snapshot destroy operation (updates to the
5689 * parent root and tree of tree roots trees, etc) are done.
Filipe Manana1ec9a1a2016-02-10 10:42:25 +00005690 */
5691void btrfs_record_snapshot_destroy(struct btrfs_trans_handle *trans,
5692 struct inode *dir)
5693{
Filipe Manana2be63d52016-02-12 11:34:23 +00005694 mutex_lock(&BTRFS_I(dir)->log_mutex);
Filipe Manana1ec9a1a2016-02-10 10:42:25 +00005695 BTRFS_I(dir)->last_unlink_trans = trans->transid;
Filipe Manana2be63d52016-02-12 11:34:23 +00005696 mutex_unlock(&BTRFS_I(dir)->log_mutex);
Filipe Manana1ec9a1a2016-02-10 10:42:25 +00005697}
5698
5699/*
Chris Mason12fcfd22009-03-24 10:24:20 -04005700 * Call this after adding a new name for a file and it will properly
5701 * update the log to reflect the new name.
5702 *
5703 * It will return zero if all goes well, and it will return 1 if a
5704 * full transaction commit is required.
5705 */
5706int btrfs_log_new_name(struct btrfs_trans_handle *trans,
5707 struct inode *inode, struct inode *old_dir,
5708 struct dentry *parent)
5709{
5710 struct btrfs_root * root = BTRFS_I(inode)->root;
5711
5712 /*
Chris Masonaf4176b2009-03-24 10:24:31 -04005713 * this will force the logging code to walk the dentry chain
5714 * up for the file
5715 */
5716 if (S_ISREG(inode->i_mode))
5717 BTRFS_I(inode)->last_unlink_trans = trans->transid;
5718
5719 /*
Chris Mason12fcfd22009-03-24 10:24:20 -04005720 * if this inode hasn't been logged and directory we're renaming it
5721 * from hasn't been logged, we don't need to log it
5722 */
5723 if (BTRFS_I(inode)->logged_trans <=
5724 root->fs_info->last_trans_committed &&
5725 (!old_dir || BTRFS_I(old_dir)->logged_trans <=
5726 root->fs_info->last_trans_committed))
5727 return 0;
5728
Filipe Manana49dae1b2014-09-06 22:34:39 +01005729 return btrfs_log_inode_parent(trans, root, inode, parent, 0,
5730 LLONG_MAX, 1, NULL);
Chris Mason12fcfd22009-03-24 10:24:20 -04005731}
5732