blob: f10bf5213ed8a48b95a8cf3cf51dc3c2c81b8463 [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"
Qu Wenruodf2c95f2016-08-15 10:36:52 +080030#include "qgroup.h"
Chris Masone02119d2008-09-05 16:13:11 -040031
32/* magic values for the inode_only field in btrfs_log_inode:
33 *
34 * LOG_INODE_ALL means to log everything
35 * LOG_INODE_EXISTS means to log just enough to recreate the inode
36 * during log replay
37 */
38#define LOG_INODE_ALL 0
39#define LOG_INODE_EXISTS 1
40
41/*
Chris Mason12fcfd22009-03-24 10:24:20 -040042 * directory trouble cases
43 *
44 * 1) on rename or unlink, if the inode being unlinked isn't in the fsync
45 * log, we must force a full commit before doing an fsync of the directory
46 * where the unlink was done.
47 * ---> record transid of last unlink/rename per directory
48 *
49 * mkdir foo/some_dir
50 * normal commit
51 * rename foo/some_dir foo2/some_dir
52 * mkdir foo/some_dir
53 * fsync foo/some_dir/some_file
54 *
55 * The fsync above will unlink the original some_dir without recording
56 * it in its new location (foo2). After a crash, some_dir will be gone
57 * unless the fsync of some_file forces a full commit
58 *
59 * 2) we must log any new names for any file or dir that is in the fsync
60 * log. ---> check inode while renaming/linking.
61 *
62 * 2a) we must log any new names for any file or dir during rename
63 * when the directory they are being removed from was logged.
64 * ---> check inode and old parent dir during rename
65 *
66 * 2a is actually the more important variant. With the extra logging
67 * a crash might unlink the old name without recreating the new one
68 *
69 * 3) after a crash, we must go through any directories with a link count
70 * of zero and redo the rm -rf
71 *
72 * mkdir f1/foo
73 * normal commit
74 * rm -rf f1/foo
75 * fsync(f1)
76 *
77 * The directory f1 was fully removed from the FS, but fsync was never
78 * called on f1, only its parent dir. After a crash the rm -rf must
79 * be replayed. This must be able to recurse down the entire
80 * directory tree. The inode link count fixup code takes care of the
81 * ugly details.
82 */
83
84/*
Chris Masone02119d2008-09-05 16:13:11 -040085 * stages for the tree walking. The first
86 * stage (0) is to only pin down the blocks we find
87 * the second stage (1) is to make sure that all the inodes
88 * we find in the log are created in the subvolume.
89 *
90 * The last stage is to deal with directories and links and extents
91 * and all the other fun semantics
92 */
93#define LOG_WALK_PIN_ONLY 0
94#define LOG_WALK_REPLAY_INODES 1
Josef Bacikdd8e7212013-09-11 11:57:23 -040095#define LOG_WALK_REPLAY_DIR_INDEX 2
96#define LOG_WALK_REPLAY_ALL 3
Chris Masone02119d2008-09-05 16:13:11 -040097
Chris Mason12fcfd22009-03-24 10:24:20 -040098static int btrfs_log_inode(struct btrfs_trans_handle *trans,
Filipe Manana49dae1b2014-09-06 22:34:39 +010099 struct btrfs_root *root, struct inode *inode,
100 int inode_only,
101 const loff_t start,
Filipe Manana8407f552014-09-05 15:14:39 +0100102 const loff_t end,
103 struct btrfs_log_ctx *ctx);
Yan Zhengec051c02009-01-05 15:43:42 -0500104static int link_to_fixup_dir(struct btrfs_trans_handle *trans,
105 struct btrfs_root *root,
106 struct btrfs_path *path, u64 objectid);
Chris Mason12fcfd22009-03-24 10:24:20 -0400107static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans,
108 struct btrfs_root *root,
109 struct btrfs_root *log,
110 struct btrfs_path *path,
111 u64 dirid, int del_all);
Chris Masone02119d2008-09-05 16:13:11 -0400112
113/*
114 * tree logging is a special write ahead log used to make sure that
115 * fsyncs and O_SYNCs can happen without doing full tree commits.
116 *
117 * Full tree commits are expensive because they require commonly
118 * modified blocks to be recowed, creating many dirty pages in the
119 * extent tree an 4x-6x higher write load than ext3.
120 *
121 * Instead of doing a tree commit on every fsync, we use the
122 * key ranges and transaction ids to find items for a given file or directory
123 * that have changed in this transaction. Those items are copied into
124 * a special tree (one per subvolume root), that tree is written to disk
125 * and then the fsync is considered complete.
126 *
127 * After a crash, items are copied out of the log-tree back into the
128 * subvolume tree. Any file data extents found are recorded in the extent
129 * allocation tree, and the log-tree freed.
130 *
131 * The log tree is read three times, once to pin down all the extents it is
132 * using in ram and once, once to create all the inodes logged in the tree
133 * and once to do all the other items.
134 */
135
136/*
Chris Masone02119d2008-09-05 16:13:11 -0400137 * start a sub transaction and setup the log tree
138 * this increments the log tree writer count to make the people
139 * syncing the tree wait for us to finish
140 */
141static int start_log_trans(struct btrfs_trans_handle *trans,
Miao Xie8b050d32014-02-20 18:08:58 +0800142 struct btrfs_root *root,
143 struct btrfs_log_ctx *ctx)
Chris Masone02119d2008-09-05 16:13:11 -0400144{
Jeff Mahoney0b246af2016-06-22 18:54:23 -0400145 struct btrfs_fs_info *fs_info = root->fs_info;
Zhaolei34eb2a52015-08-17 18:44:45 +0800146 int ret = 0;
Yan Zheng7237f182009-01-21 12:54:03 -0500147
148 mutex_lock(&root->log_mutex);
Zhaolei34eb2a52015-08-17 18:44:45 +0800149
Yan Zheng7237f182009-01-21 12:54:03 -0500150 if (root->log_root) {
Jeff Mahoney0b246af2016-06-22 18:54:23 -0400151 if (btrfs_need_log_full_commit(fs_info, trans)) {
Miao Xie50471a32014-02-20 18:08:57 +0800152 ret = -EAGAIN;
153 goto out;
154 }
Zhaolei34eb2a52015-08-17 18:44:45 +0800155
Josef Bacikff782e02009-10-08 15:30:04 -0400156 if (!root->log_start_pid) {
Miao Xie27cdeb72014-04-02 19:51:05 +0800157 clear_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state);
Zhaolei34eb2a52015-08-17 18:44:45 +0800158 root->log_start_pid = current->pid;
Josef Bacikff782e02009-10-08 15:30:04 -0400159 } else if (root->log_start_pid != current->pid) {
Miao Xie27cdeb72014-04-02 19:51:05 +0800160 set_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state);
Josef Bacikff782e02009-10-08 15:30:04 -0400161 }
Zhaolei34eb2a52015-08-17 18:44:45 +0800162 } else {
Jeff Mahoney0b246af2016-06-22 18:54:23 -0400163 mutex_lock(&fs_info->tree_log_mutex);
164 if (!fs_info->log_root_tree)
165 ret = btrfs_init_log_root_tree(trans, fs_info);
166 mutex_unlock(&fs_info->tree_log_mutex);
Zhaolei34eb2a52015-08-17 18:44:45 +0800167 if (ret)
168 goto out;
Josef Bacikff782e02009-10-08 15:30:04 -0400169
Chris Masone02119d2008-09-05 16:13:11 -0400170 ret = btrfs_add_log_tree(trans, root);
Yan, Zheng4a500fd2010-05-16 10:49:59 -0400171 if (ret)
Miao Xiee87ac132014-02-20 18:08:53 +0800172 goto out;
Zhaolei34eb2a52015-08-17 18:44:45 +0800173
174 clear_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state);
175 root->log_start_pid = current->pid;
Chris Masone02119d2008-09-05 16:13:11 -0400176 }
Zhaolei34eb2a52015-08-17 18:44:45 +0800177
Miao Xie2ecb7922012-09-06 04:04:27 -0600178 atomic_inc(&root->log_batch);
Yan Zheng7237f182009-01-21 12:54:03 -0500179 atomic_inc(&root->log_writers);
Miao Xie8b050d32014-02-20 18:08:58 +0800180 if (ctx) {
Zhaolei34eb2a52015-08-17 18:44:45 +0800181 int index = root->log_transid % 2;
Miao Xie8b050d32014-02-20 18:08:58 +0800182 list_add_tail(&ctx->list, &root->log_ctxs[index]);
Miao Xied1433de2014-02-20 18:08:59 +0800183 ctx->log_transid = root->log_transid;
Miao Xie8b050d32014-02-20 18:08:58 +0800184 }
Zhaolei34eb2a52015-08-17 18:44:45 +0800185
Miao Xiee87ac132014-02-20 18:08:53 +0800186out:
Yan Zheng7237f182009-01-21 12:54:03 -0500187 mutex_unlock(&root->log_mutex);
Miao Xiee87ac132014-02-20 18:08:53 +0800188 return ret;
Chris Masone02119d2008-09-05 16:13:11 -0400189}
190
191/*
192 * returns 0 if there was a log transaction running and we were able
193 * to join, or returns -ENOENT if there were not transactions
194 * in progress
195 */
196static int join_running_log_trans(struct btrfs_root *root)
197{
198 int ret = -ENOENT;
199
200 smp_mb();
201 if (!root->log_root)
202 return -ENOENT;
203
Yan Zheng7237f182009-01-21 12:54:03 -0500204 mutex_lock(&root->log_mutex);
Chris Masone02119d2008-09-05 16:13:11 -0400205 if (root->log_root) {
206 ret = 0;
Yan Zheng7237f182009-01-21 12:54:03 -0500207 atomic_inc(&root->log_writers);
Chris Masone02119d2008-09-05 16:13:11 -0400208 }
Yan Zheng7237f182009-01-21 12:54:03 -0500209 mutex_unlock(&root->log_mutex);
Chris Masone02119d2008-09-05 16:13:11 -0400210 return ret;
211}
212
213/*
Chris Mason12fcfd22009-03-24 10:24:20 -0400214 * This either makes the current running log transaction wait
215 * until you call btrfs_end_log_trans() or it makes any future
216 * log transactions wait until you call btrfs_end_log_trans()
217 */
218int btrfs_pin_log_trans(struct btrfs_root *root)
219{
220 int ret = -ENOENT;
221
222 mutex_lock(&root->log_mutex);
223 atomic_inc(&root->log_writers);
224 mutex_unlock(&root->log_mutex);
225 return ret;
226}
227
228/*
Chris Masone02119d2008-09-05 16:13:11 -0400229 * indicate we're done making changes to the log tree
230 * and wake up anyone waiting to do a sync
231 */
Jeff Mahoney143bede2012-03-01 14:56:26 +0100232void btrfs_end_log_trans(struct btrfs_root *root)
Chris Masone02119d2008-09-05 16:13:11 -0400233{
Yan Zheng7237f182009-01-21 12:54:03 -0500234 if (atomic_dec_and_test(&root->log_writers)) {
David Sterba779adf02015-02-16 19:39:00 +0100235 /*
236 * Implicit memory barrier after atomic_dec_and_test
237 */
Yan Zheng7237f182009-01-21 12:54:03 -0500238 if (waitqueue_active(&root->log_writer_wait))
239 wake_up(&root->log_writer_wait);
240 }
Chris Masone02119d2008-09-05 16:13:11 -0400241}
242
243
244/*
245 * the walk control struct is used to pass state down the chain when
246 * processing the log tree. The stage field tells us which part
247 * of the log tree processing we are currently doing. The others
248 * are state fields used for that specific part
249 */
250struct walk_control {
251 /* should we free the extent on disk when done? This is used
252 * at transaction commit time while freeing a log tree
253 */
254 int free;
255
256 /* should we write out the extent buffer? This is used
257 * while flushing the log tree to disk during a sync
258 */
259 int write;
260
261 /* should we wait for the extent buffer io to finish? Also used
262 * while flushing the log tree to disk for a sync
263 */
264 int wait;
265
266 /* pin only walk, we record which extents on disk belong to the
267 * log trees
268 */
269 int pin;
270
271 /* what stage of the replay code we're currently in */
272 int stage;
273
274 /* the root we are currently replaying */
275 struct btrfs_root *replay_dest;
276
277 /* the trans handle for the current replay */
278 struct btrfs_trans_handle *trans;
279
280 /* the function that gets used to process blocks we find in the
281 * tree. Note the extent_buffer might not be up to date when it is
282 * passed in, and it must be checked or read if you need the data
283 * inside it
284 */
285 int (*process_func)(struct btrfs_root *log, struct extent_buffer *eb,
286 struct walk_control *wc, u64 gen);
287};
288
289/*
290 * process_func used to pin down extents, write them or wait on them
291 */
292static int process_one_buffer(struct btrfs_root *log,
293 struct extent_buffer *eb,
294 struct walk_control *wc, u64 gen)
295{
Jeff Mahoney0b246af2016-06-22 18:54:23 -0400296 struct btrfs_fs_info *fs_info = log->fs_info;
Josef Bacikb50c6e22013-04-25 15:55:30 -0400297 int ret = 0;
Chris Masone02119d2008-09-05 16:13:11 -0400298
Josef Bacik8c2a1a32013-06-06 13:19:32 -0400299 /*
300 * If this fs is mixed then we need to be able to process the leaves to
301 * pin down any logged extents, so we have to read the block.
302 */
Jeff Mahoney0b246af2016-06-22 18:54:23 -0400303 if (btrfs_fs_incompat(fs_info, MIXED_GROUPS)) {
Josef Bacik8c2a1a32013-06-06 13:19:32 -0400304 ret = btrfs_read_buffer(eb, gen);
305 if (ret)
306 return ret;
307 }
308
Josef Bacikb50c6e22013-04-25 15:55:30 -0400309 if (wc->pin)
Jeff Mahoney2ff7e612016-06-22 18:54:24 -0400310 ret = btrfs_pin_extent_for_log_replay(fs_info, eb->start,
311 eb->len);
Josef Bacikb50c6e22013-04-25 15:55:30 -0400312
313 if (!ret && btrfs_buffer_uptodate(eb, gen, 0)) {
Josef Bacik8c2a1a32013-06-06 13:19:32 -0400314 if (wc->pin && btrfs_header_level(eb) == 0)
Jeff Mahoney2ff7e612016-06-22 18:54:24 -0400315 ret = btrfs_exclude_logged_extents(fs_info, eb);
Chris Masone02119d2008-09-05 16:13:11 -0400316 if (wc->write)
317 btrfs_write_tree_block(eb);
318 if (wc->wait)
319 btrfs_wait_tree_block_writeback(eb);
320 }
Josef Bacikb50c6e22013-04-25 15:55:30 -0400321 return ret;
Chris Masone02119d2008-09-05 16:13:11 -0400322}
323
324/*
325 * Item overwrite used by replay and tree logging. eb, slot and key all refer
326 * to the src data we are copying out.
327 *
328 * root is the tree we are copying into, and path is a scratch
329 * path for use in this function (it should be released on entry and
330 * will be released on exit).
331 *
332 * If the key is already in the destination tree the existing item is
333 * overwritten. If the existing item isn't big enough, it is extended.
334 * If it is too large, it is truncated.
335 *
336 * If the key isn't in the destination yet, a new item is inserted.
337 */
338static noinline int overwrite_item(struct btrfs_trans_handle *trans,
339 struct btrfs_root *root,
340 struct btrfs_path *path,
341 struct extent_buffer *eb, int slot,
342 struct btrfs_key *key)
343{
Jeff Mahoney2ff7e612016-06-22 18:54:24 -0400344 struct btrfs_fs_info *fs_info = root->fs_info;
Chris Masone02119d2008-09-05 16:13:11 -0400345 int ret;
346 u32 item_size;
347 u64 saved_i_size = 0;
348 int save_old_i_size = 0;
349 unsigned long src_ptr;
350 unsigned long dst_ptr;
351 int overwrite_root = 0;
Josef Bacik4bc4bee2013-04-05 20:50:09 +0000352 bool inode_item = key->type == BTRFS_INODE_ITEM_KEY;
Chris Masone02119d2008-09-05 16:13:11 -0400353
354 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
355 overwrite_root = 1;
356
357 item_size = btrfs_item_size_nr(eb, slot);
358 src_ptr = btrfs_item_ptr_offset(eb, slot);
359
360 /* look for the key in the destination tree */
361 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
Josef Bacik4bc4bee2013-04-05 20:50:09 +0000362 if (ret < 0)
363 return ret;
364
Chris Masone02119d2008-09-05 16:13:11 -0400365 if (ret == 0) {
366 char *src_copy;
367 char *dst_copy;
368 u32 dst_size = btrfs_item_size_nr(path->nodes[0],
369 path->slots[0]);
370 if (dst_size != item_size)
371 goto insert;
372
373 if (item_size == 0) {
David Sterbab3b4aa72011-04-21 01:20:15 +0200374 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -0400375 return 0;
376 }
377 dst_copy = kmalloc(item_size, GFP_NOFS);
378 src_copy = kmalloc(item_size, GFP_NOFS);
liubo2a29edc2011-01-26 06:22:08 +0000379 if (!dst_copy || !src_copy) {
David Sterbab3b4aa72011-04-21 01:20:15 +0200380 btrfs_release_path(path);
liubo2a29edc2011-01-26 06:22:08 +0000381 kfree(dst_copy);
382 kfree(src_copy);
383 return -ENOMEM;
384 }
Chris Masone02119d2008-09-05 16:13:11 -0400385
386 read_extent_buffer(eb, src_copy, src_ptr, item_size);
387
388 dst_ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
389 read_extent_buffer(path->nodes[0], dst_copy, dst_ptr,
390 item_size);
391 ret = memcmp(dst_copy, src_copy, item_size);
392
393 kfree(dst_copy);
394 kfree(src_copy);
395 /*
396 * they have the same contents, just return, this saves
397 * us from cowing blocks in the destination tree and doing
398 * extra writes that may not have been done by a previous
399 * sync
400 */
401 if (ret == 0) {
David Sterbab3b4aa72011-04-21 01:20:15 +0200402 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -0400403 return 0;
404 }
405
Josef Bacik4bc4bee2013-04-05 20:50:09 +0000406 /*
407 * We need to load the old nbytes into the inode so when we
408 * replay the extents we've logged we get the right nbytes.
409 */
410 if (inode_item) {
411 struct btrfs_inode_item *item;
412 u64 nbytes;
Josef Bacikd5554382013-09-11 14:17:00 -0400413 u32 mode;
Josef Bacik4bc4bee2013-04-05 20:50:09 +0000414
415 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
416 struct btrfs_inode_item);
417 nbytes = btrfs_inode_nbytes(path->nodes[0], item);
418 item = btrfs_item_ptr(eb, slot,
419 struct btrfs_inode_item);
420 btrfs_set_inode_nbytes(eb, item, nbytes);
Josef Bacikd5554382013-09-11 14:17:00 -0400421
422 /*
423 * If this is a directory we need to reset the i_size to
424 * 0 so that we can set it up properly when replaying
425 * the rest of the items in this log.
426 */
427 mode = btrfs_inode_mode(eb, item);
428 if (S_ISDIR(mode))
429 btrfs_set_inode_size(eb, item, 0);
Josef Bacik4bc4bee2013-04-05 20:50:09 +0000430 }
431 } else if (inode_item) {
432 struct btrfs_inode_item *item;
Josef Bacikd5554382013-09-11 14:17:00 -0400433 u32 mode;
Josef Bacik4bc4bee2013-04-05 20:50:09 +0000434
435 /*
436 * New inode, set nbytes to 0 so that the nbytes comes out
437 * properly when we replay the extents.
438 */
439 item = btrfs_item_ptr(eb, slot, struct btrfs_inode_item);
440 btrfs_set_inode_nbytes(eb, item, 0);
Josef Bacikd5554382013-09-11 14:17:00 -0400441
442 /*
443 * If this is a directory we need to reset the i_size to 0 so
444 * that we can set it up properly when replaying the rest of
445 * the items in this log.
446 */
447 mode = btrfs_inode_mode(eb, item);
448 if (S_ISDIR(mode))
449 btrfs_set_inode_size(eb, item, 0);
Chris Masone02119d2008-09-05 16:13:11 -0400450 }
451insert:
David Sterbab3b4aa72011-04-21 01:20:15 +0200452 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -0400453 /* try to insert the key into the destination tree */
Filipe Mananadf8d1162015-01-14 01:52:25 +0000454 path->skip_release_on_error = 1;
Chris Masone02119d2008-09-05 16:13:11 -0400455 ret = btrfs_insert_empty_item(trans, root, path,
456 key, item_size);
Filipe Mananadf8d1162015-01-14 01:52:25 +0000457 path->skip_release_on_error = 0;
Chris Masone02119d2008-09-05 16:13:11 -0400458
459 /* make sure any existing item is the correct size */
Filipe Mananadf8d1162015-01-14 01:52:25 +0000460 if (ret == -EEXIST || ret == -EOVERFLOW) {
Chris Masone02119d2008-09-05 16:13:11 -0400461 u32 found_size;
462 found_size = btrfs_item_size_nr(path->nodes[0],
463 path->slots[0]);
Jeff Mahoney143bede2012-03-01 14:56:26 +0100464 if (found_size > item_size)
Jeff Mahoney2ff7e612016-06-22 18:54:24 -0400465 btrfs_truncate_item(fs_info, path, item_size, 1);
Jeff Mahoney143bede2012-03-01 14:56:26 +0100466 else if (found_size < item_size)
Jeff Mahoney2ff7e612016-06-22 18:54:24 -0400467 btrfs_extend_item(fs_info, path,
Jeff Mahoney143bede2012-03-01 14:56:26 +0100468 item_size - found_size);
Chris Masone02119d2008-09-05 16:13:11 -0400469 } else if (ret) {
Yan, Zheng4a500fd2010-05-16 10:49:59 -0400470 return ret;
Chris Masone02119d2008-09-05 16:13:11 -0400471 }
472 dst_ptr = btrfs_item_ptr_offset(path->nodes[0],
473 path->slots[0]);
474
475 /* don't overwrite an existing inode if the generation number
476 * was logged as zero. This is done when the tree logging code
477 * is just logging an inode to make sure it exists after recovery.
478 *
479 * Also, don't overwrite i_size on directories during replay.
480 * log replay inserts and removes directory items based on the
481 * state of the tree found in the subvolume, and i_size is modified
482 * as it goes
483 */
484 if (key->type == BTRFS_INODE_ITEM_KEY && ret == -EEXIST) {
485 struct btrfs_inode_item *src_item;
486 struct btrfs_inode_item *dst_item;
487
488 src_item = (struct btrfs_inode_item *)src_ptr;
489 dst_item = (struct btrfs_inode_item *)dst_ptr;
490
Filipe Manana1a4bcf42015-02-13 12:30:56 +0000491 if (btrfs_inode_generation(eb, src_item) == 0) {
492 struct extent_buffer *dst_eb = path->nodes[0];
Filipe Manana2f2ff0e2015-03-20 17:19:46 +0000493 const u64 ino_size = btrfs_inode_size(eb, src_item);
Filipe Manana1a4bcf42015-02-13 12:30:56 +0000494
Filipe Manana2f2ff0e2015-03-20 17:19:46 +0000495 /*
496 * For regular files an ino_size == 0 is used only when
497 * logging that an inode exists, as part of a directory
498 * fsync, and the inode wasn't fsynced before. In this
499 * case don't set the size of the inode in the fs/subvol
500 * tree, otherwise we would be throwing valid data away.
501 */
Filipe Manana1a4bcf42015-02-13 12:30:56 +0000502 if (S_ISREG(btrfs_inode_mode(eb, src_item)) &&
Filipe Manana2f2ff0e2015-03-20 17:19:46 +0000503 S_ISREG(btrfs_inode_mode(dst_eb, dst_item)) &&
504 ino_size != 0) {
Filipe Manana1a4bcf42015-02-13 12:30:56 +0000505 struct btrfs_map_token token;
Filipe Manana1a4bcf42015-02-13 12:30:56 +0000506
507 btrfs_init_map_token(&token);
508 btrfs_set_token_inode_size(dst_eb, dst_item,
509 ino_size, &token);
510 }
Chris Masone02119d2008-09-05 16:13:11 -0400511 goto no_copy;
Filipe Manana1a4bcf42015-02-13 12:30:56 +0000512 }
Chris Masone02119d2008-09-05 16:13:11 -0400513
514 if (overwrite_root &&
515 S_ISDIR(btrfs_inode_mode(eb, src_item)) &&
516 S_ISDIR(btrfs_inode_mode(path->nodes[0], dst_item))) {
517 save_old_i_size = 1;
518 saved_i_size = btrfs_inode_size(path->nodes[0],
519 dst_item);
520 }
521 }
522
523 copy_extent_buffer(path->nodes[0], eb, dst_ptr,
524 src_ptr, item_size);
525
526 if (save_old_i_size) {
527 struct btrfs_inode_item *dst_item;
528 dst_item = (struct btrfs_inode_item *)dst_ptr;
529 btrfs_set_inode_size(path->nodes[0], dst_item, saved_i_size);
530 }
531
532 /* make sure the generation is filled in */
533 if (key->type == BTRFS_INODE_ITEM_KEY) {
534 struct btrfs_inode_item *dst_item;
535 dst_item = (struct btrfs_inode_item *)dst_ptr;
536 if (btrfs_inode_generation(path->nodes[0], dst_item) == 0) {
537 btrfs_set_inode_generation(path->nodes[0], dst_item,
538 trans->transid);
539 }
540 }
541no_copy:
542 btrfs_mark_buffer_dirty(path->nodes[0]);
David Sterbab3b4aa72011-04-21 01:20:15 +0200543 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -0400544 return 0;
545}
546
547/*
548 * simple helper to read an inode off the disk from a given root
549 * This can only be called for subvolume roots and not for the log
550 */
551static noinline struct inode *read_one_inode(struct btrfs_root *root,
552 u64 objectid)
553{
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400554 struct btrfs_key key;
Chris Masone02119d2008-09-05 16:13:11 -0400555 struct inode *inode;
Chris Masone02119d2008-09-05 16:13:11 -0400556
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400557 key.objectid = objectid;
558 key.type = BTRFS_INODE_ITEM_KEY;
559 key.offset = 0;
Josef Bacik73f73412009-12-04 17:38:27 +0000560 inode = btrfs_iget(root->fs_info->sb, &key, root, NULL);
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400561 if (IS_ERR(inode)) {
562 inode = NULL;
563 } else if (is_bad_inode(inode)) {
Chris Masone02119d2008-09-05 16:13:11 -0400564 iput(inode);
565 inode = NULL;
566 }
567 return inode;
568}
569
570/* replays a single extent in 'eb' at 'slot' with 'key' into the
571 * subvolume 'root'. path is released on entry and should be released
572 * on exit.
573 *
574 * extents in the log tree have not been allocated out of the extent
575 * tree yet. So, this completes the allocation, taking a reference
576 * as required if the extent already exists or creating a new extent
577 * if it isn't in the extent allocation tree yet.
578 *
579 * The extent is inserted into the file, dropping any existing extents
580 * from the file that overlap the new one.
581 */
582static noinline int replay_one_extent(struct btrfs_trans_handle *trans,
583 struct btrfs_root *root,
584 struct btrfs_path *path,
585 struct extent_buffer *eb, int slot,
586 struct btrfs_key *key)
587{
Jeff Mahoney0b246af2016-06-22 18:54:23 -0400588 struct btrfs_fs_info *fs_info = root->fs_info;
Chris Masone02119d2008-09-05 16:13:11 -0400589 int found_type;
Chris Masone02119d2008-09-05 16:13:11 -0400590 u64 extent_end;
Chris Masone02119d2008-09-05 16:13:11 -0400591 u64 start = key->offset;
Josef Bacik4bc4bee2013-04-05 20:50:09 +0000592 u64 nbytes = 0;
Chris Masone02119d2008-09-05 16:13:11 -0400593 struct btrfs_file_extent_item *item;
594 struct inode *inode = NULL;
595 unsigned long size;
596 int ret = 0;
597
598 item = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
599 found_type = btrfs_file_extent_type(eb, item);
600
Yan Zhengd899e052008-10-30 14:25:28 -0400601 if (found_type == BTRFS_FILE_EXTENT_REG ||
Josef Bacik4bc4bee2013-04-05 20:50:09 +0000602 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
603 nbytes = btrfs_file_extent_num_bytes(eb, item);
604 extent_end = start + nbytes;
605
606 /*
607 * We don't add to the inodes nbytes if we are prealloc or a
608 * hole.
609 */
610 if (btrfs_file_extent_disk_bytenr(eb, item) == 0)
611 nbytes = 0;
612 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
Chris Mason514ac8a2014-01-03 21:07:00 -0800613 size = btrfs_file_extent_inline_len(eb, slot, item);
Josef Bacik4bc4bee2013-04-05 20:50:09 +0000614 nbytes = btrfs_file_extent_ram_bytes(eb, item);
Jeff Mahoneyda170662016-06-15 09:22:56 -0400615 extent_end = ALIGN(start + size,
Jeff Mahoney0b246af2016-06-22 18:54:23 -0400616 fs_info->sectorsize);
Chris Masone02119d2008-09-05 16:13:11 -0400617 } else {
618 ret = 0;
619 goto out;
620 }
621
622 inode = read_one_inode(root, key->objectid);
623 if (!inode) {
624 ret = -EIO;
625 goto out;
626 }
627
628 /*
629 * first check to see if we already have this extent in the
630 * file. This must be done before the btrfs_drop_extents run
631 * so we don't try to drop this extent.
632 */
Li Zefan33345d012011-04-20 10:31:50 +0800633 ret = btrfs_lookup_file_extent(trans, root, path, btrfs_ino(inode),
Chris Masone02119d2008-09-05 16:13:11 -0400634 start, 0);
635
Yan Zhengd899e052008-10-30 14:25:28 -0400636 if (ret == 0 &&
637 (found_type == BTRFS_FILE_EXTENT_REG ||
638 found_type == BTRFS_FILE_EXTENT_PREALLOC)) {
Chris Masone02119d2008-09-05 16:13:11 -0400639 struct btrfs_file_extent_item cmp1;
640 struct btrfs_file_extent_item cmp2;
641 struct btrfs_file_extent_item *existing;
642 struct extent_buffer *leaf;
643
644 leaf = path->nodes[0];
645 existing = btrfs_item_ptr(leaf, path->slots[0],
646 struct btrfs_file_extent_item);
647
648 read_extent_buffer(eb, &cmp1, (unsigned long)item,
649 sizeof(cmp1));
650 read_extent_buffer(leaf, &cmp2, (unsigned long)existing,
651 sizeof(cmp2));
652
653 /*
654 * we already have a pointer to this exact extent,
655 * we don't have to do anything
656 */
657 if (memcmp(&cmp1, &cmp2, sizeof(cmp1)) == 0) {
David Sterbab3b4aa72011-04-21 01:20:15 +0200658 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -0400659 goto out;
660 }
661 }
David Sterbab3b4aa72011-04-21 01:20:15 +0200662 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -0400663
664 /* drop any overlapping extents */
Josef Bacik26714852012-08-29 12:24:27 -0400665 ret = btrfs_drop_extents(trans, root, inode, start, extent_end, 1);
Josef Bacik36508602013-04-25 16:23:32 -0400666 if (ret)
667 goto out;
Chris Masone02119d2008-09-05 16:13:11 -0400668
Yan Zheng07d400a2009-01-06 11:42:00 -0500669 if (found_type == BTRFS_FILE_EXTENT_REG ||
670 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400671 u64 offset;
Yan Zheng07d400a2009-01-06 11:42:00 -0500672 unsigned long dest_offset;
673 struct btrfs_key ins;
Chris Masone02119d2008-09-05 16:13:11 -0400674
Yan Zheng07d400a2009-01-06 11:42:00 -0500675 ret = btrfs_insert_empty_item(trans, root, path, key,
676 sizeof(*item));
Josef Bacik36508602013-04-25 16:23:32 -0400677 if (ret)
678 goto out;
Yan Zheng07d400a2009-01-06 11:42:00 -0500679 dest_offset = btrfs_item_ptr_offset(path->nodes[0],
680 path->slots[0]);
681 copy_extent_buffer(path->nodes[0], eb, dest_offset,
682 (unsigned long)item, sizeof(*item));
683
684 ins.objectid = btrfs_file_extent_disk_bytenr(eb, item);
685 ins.offset = btrfs_file_extent_disk_num_bytes(eb, item);
686 ins.type = BTRFS_EXTENT_ITEM_KEY;
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400687 offset = key->offset - btrfs_file_extent_offset(eb, item);
Yan Zheng07d400a2009-01-06 11:42:00 -0500688
Qu Wenruodf2c95f2016-08-15 10:36:52 +0800689 /*
690 * Manually record dirty extent, as here we did a shallow
691 * file extent item copy and skip normal backref update,
692 * but modifying extent tree all by ourselves.
693 * So need to manually record dirty extent for qgroup,
694 * as the owner of the file extent changed from log tree
695 * (doesn't affect qgroup) to fs/file tree(affects qgroup)
696 */
Jeff Mahoney0b246af2016-06-22 18:54:23 -0400697 ret = btrfs_qgroup_trace_extent(trans, fs_info,
Qu Wenruodf2c95f2016-08-15 10:36:52 +0800698 btrfs_file_extent_disk_bytenr(eb, item),
699 btrfs_file_extent_disk_num_bytes(eb, item),
700 GFP_NOFS);
701 if (ret < 0)
702 goto out;
703
Yan Zheng07d400a2009-01-06 11:42:00 -0500704 if (ins.objectid > 0) {
705 u64 csum_start;
706 u64 csum_end;
707 LIST_HEAD(ordered_sums);
708 /*
709 * is this extent already allocated in the extent
710 * allocation tree? If so, just add a reference
711 */
Jeff Mahoney2ff7e612016-06-22 18:54:24 -0400712 ret = btrfs_lookup_data_extent(fs_info, ins.objectid,
Yan Zheng07d400a2009-01-06 11:42:00 -0500713 ins.offset);
714 if (ret == 0) {
Jeff Mahoney2ff7e612016-06-22 18:54:24 -0400715 ret = btrfs_inc_extent_ref(trans, fs_info,
Yan Zheng07d400a2009-01-06 11:42:00 -0500716 ins.objectid, ins.offset,
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400717 0, root->root_key.objectid,
Filipe Mananab06c4bf2015-10-23 07:52:54 +0100718 key->objectid, offset);
Josef Bacikb50c6e22013-04-25 15:55:30 -0400719 if (ret)
720 goto out;
Yan Zheng07d400a2009-01-06 11:42:00 -0500721 } else {
722 /*
723 * insert the extent pointer in the extent
724 * allocation tree
725 */
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400726 ret = btrfs_alloc_logged_file_extent(trans,
Jeff Mahoney2ff7e612016-06-22 18:54:24 -0400727 fs_info,
728 root->root_key.objectid,
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400729 key->objectid, offset, &ins);
Josef Bacikb50c6e22013-04-25 15:55:30 -0400730 if (ret)
731 goto out;
Yan Zheng07d400a2009-01-06 11:42:00 -0500732 }
David Sterbab3b4aa72011-04-21 01:20:15 +0200733 btrfs_release_path(path);
Yan Zheng07d400a2009-01-06 11:42:00 -0500734
735 if (btrfs_file_extent_compression(eb, item)) {
736 csum_start = ins.objectid;
737 csum_end = csum_start + ins.offset;
738 } else {
739 csum_start = ins.objectid +
740 btrfs_file_extent_offset(eb, item);
741 csum_end = csum_start +
742 btrfs_file_extent_num_bytes(eb, item);
743 }
744
745 ret = btrfs_lookup_csums_range(root->log_root,
746 csum_start, csum_end - 1,
Arne Jansena2de7332011-03-08 14:14:00 +0100747 &ordered_sums, 0);
Josef Bacik36508602013-04-25 16:23:32 -0400748 if (ret)
749 goto out;
Filipe Mananab84b8392015-08-19 11:09:40 +0100750 /*
751 * Now delete all existing cums in the csum root that
752 * cover our range. We do this because we can have an
753 * extent that is completely referenced by one file
754 * extent item and partially referenced by another
755 * file extent item (like after using the clone or
756 * extent_same ioctls). In this case if we end up doing
757 * the replay of the one that partially references the
758 * extent first, and we do not do the csum deletion
759 * below, we can get 2 csum items in the csum tree that
760 * overlap each other. For example, imagine our log has
761 * the two following file extent items:
762 *
763 * key (257 EXTENT_DATA 409600)
764 * extent data disk byte 12845056 nr 102400
765 * extent data offset 20480 nr 20480 ram 102400
766 *
767 * key (257 EXTENT_DATA 819200)
768 * extent data disk byte 12845056 nr 102400
769 * extent data offset 0 nr 102400 ram 102400
770 *
771 * Where the second one fully references the 100K extent
772 * that starts at disk byte 12845056, and the log tree
773 * has a single csum item that covers the entire range
774 * of the extent:
775 *
776 * key (EXTENT_CSUM EXTENT_CSUM 12845056) itemsize 100
777 *
778 * After the first file extent item is replayed, the
779 * csum tree gets the following csum item:
780 *
781 * key (EXTENT_CSUM EXTENT_CSUM 12865536) itemsize 20
782 *
783 * Which covers the 20K sub-range starting at offset 20K
784 * of our extent. Now when we replay the second file
785 * extent item, if we do not delete existing csum items
786 * that cover any of its blocks, we end up getting two
787 * csum items in our csum tree that overlap each other:
788 *
789 * key (EXTENT_CSUM EXTENT_CSUM 12845056) itemsize 100
790 * key (EXTENT_CSUM EXTENT_CSUM 12865536) itemsize 20
791 *
792 * Which is a problem, because after this anyone trying
793 * to lookup up for the checksum of any block of our
794 * extent starting at an offset of 40K or higher, will
795 * end up looking at the second csum item only, which
796 * does not contain the checksum for any block starting
797 * at offset 40K or higher of our extent.
798 */
Yan Zheng07d400a2009-01-06 11:42:00 -0500799 while (!list_empty(&ordered_sums)) {
800 struct btrfs_ordered_sum *sums;
801 sums = list_entry(ordered_sums.next,
802 struct btrfs_ordered_sum,
803 list);
Josef Bacik36508602013-04-25 16:23:32 -0400804 if (!ret)
Jeff Mahoney0b246af2016-06-22 18:54:23 -0400805 ret = btrfs_del_csums(trans, fs_info,
Jeff Mahoney5b4aace2016-06-21 10:40:19 -0400806 sums->bytenr,
807 sums->len);
Filipe Mananab84b8392015-08-19 11:09:40 +0100808 if (!ret)
Josef Bacik36508602013-04-25 16:23:32 -0400809 ret = btrfs_csum_file_blocks(trans,
Jeff Mahoney0b246af2016-06-22 18:54:23 -0400810 fs_info->csum_root, sums);
Yan Zheng07d400a2009-01-06 11:42:00 -0500811 list_del(&sums->list);
812 kfree(sums);
813 }
Josef Bacik36508602013-04-25 16:23:32 -0400814 if (ret)
815 goto out;
Yan Zheng07d400a2009-01-06 11:42:00 -0500816 } else {
David Sterbab3b4aa72011-04-21 01:20:15 +0200817 btrfs_release_path(path);
Yan Zheng07d400a2009-01-06 11:42:00 -0500818 }
819 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
820 /* inline extents are easy, we just overwrite them */
821 ret = overwrite_item(trans, root, path, eb, slot, key);
Josef Bacik36508602013-04-25 16:23:32 -0400822 if (ret)
823 goto out;
Yan Zheng07d400a2009-01-06 11:42:00 -0500824 }
825
Josef Bacik4bc4bee2013-04-05 20:50:09 +0000826 inode_add_bytes(inode, nbytes);
Tsutomu Itohb9959292012-06-25 21:25:22 -0600827 ret = btrfs_update_inode(trans, root, inode);
Chris Masone02119d2008-09-05 16:13:11 -0400828out:
829 if (inode)
830 iput(inode);
831 return ret;
832}
833
834/*
835 * when cleaning up conflicts between the directory names in the
836 * subvolume, directory names in the log and directory names in the
837 * inode back references, we may have to unlink inodes from directories.
838 *
839 * This is a helper function to do the unlink of a specific directory
840 * item
841 */
842static noinline int drop_one_dir_item(struct btrfs_trans_handle *trans,
843 struct btrfs_root *root,
844 struct btrfs_path *path,
845 struct inode *dir,
846 struct btrfs_dir_item *di)
847{
Jeff Mahoney2ff7e612016-06-22 18:54:24 -0400848 struct btrfs_fs_info *fs_info = root->fs_info;
Chris Masone02119d2008-09-05 16:13:11 -0400849 struct inode *inode;
850 char *name;
851 int name_len;
852 struct extent_buffer *leaf;
853 struct btrfs_key location;
854 int ret;
855
856 leaf = path->nodes[0];
857
858 btrfs_dir_item_key_to_cpu(leaf, di, &location);
859 name_len = btrfs_dir_name_len(leaf, di);
860 name = kmalloc(name_len, GFP_NOFS);
liubo2a29edc2011-01-26 06:22:08 +0000861 if (!name)
862 return -ENOMEM;
863
Chris Masone02119d2008-09-05 16:13:11 -0400864 read_extent_buffer(leaf, name, (unsigned long)(di + 1), name_len);
David Sterbab3b4aa72011-04-21 01:20:15 +0200865 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -0400866
867 inode = read_one_inode(root, location.objectid);
Tsutomu Itohc00e9492011-04-28 09:10:23 +0000868 if (!inode) {
Josef Bacik36508602013-04-25 16:23:32 -0400869 ret = -EIO;
870 goto out;
Tsutomu Itohc00e9492011-04-28 09:10:23 +0000871 }
Chris Masone02119d2008-09-05 16:13:11 -0400872
Yan Zhengec051c02009-01-05 15:43:42 -0500873 ret = link_to_fixup_dir(trans, root, path, location.objectid);
Josef Bacik36508602013-04-25 16:23:32 -0400874 if (ret)
875 goto out;
Chris Mason12fcfd22009-03-24 10:24:20 -0400876
Chris Masone02119d2008-09-05 16:13:11 -0400877 ret = btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
Josef Bacik36508602013-04-25 16:23:32 -0400878 if (ret)
879 goto out;
Filipe David Borba Mananaada9af22013-08-05 09:25:47 +0100880 else
Jeff Mahoney2ff7e612016-06-22 18:54:24 -0400881 ret = btrfs_run_delayed_items(trans, fs_info);
Josef Bacik36508602013-04-25 16:23:32 -0400882out:
883 kfree(name);
884 iput(inode);
Chris Masone02119d2008-09-05 16:13:11 -0400885 return ret;
886}
887
888/*
889 * helper function to see if a given name and sequence number found
890 * in an inode back reference are already in a directory and correctly
891 * point to this inode
892 */
893static noinline int inode_in_dir(struct btrfs_root *root,
894 struct btrfs_path *path,
895 u64 dirid, u64 objectid, u64 index,
896 const char *name, int name_len)
897{
898 struct btrfs_dir_item *di;
899 struct btrfs_key location;
900 int match = 0;
901
902 di = btrfs_lookup_dir_index_item(NULL, root, path, dirid,
903 index, name, name_len, 0);
904 if (di && !IS_ERR(di)) {
905 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
906 if (location.objectid != objectid)
907 goto out;
908 } else
909 goto out;
David Sterbab3b4aa72011-04-21 01:20:15 +0200910 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -0400911
912 di = btrfs_lookup_dir_item(NULL, root, path, dirid, name, name_len, 0);
913 if (di && !IS_ERR(di)) {
914 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
915 if (location.objectid != objectid)
916 goto out;
917 } else
918 goto out;
919 match = 1;
920out:
David Sterbab3b4aa72011-04-21 01:20:15 +0200921 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -0400922 return match;
923}
924
925/*
926 * helper function to check a log tree for a named back reference in
927 * an inode. This is used to decide if a back reference that is
928 * found in the subvolume conflicts with what we find in the log.
929 *
930 * inode backreferences may have multiple refs in a single item,
931 * during replay we process one reference at a time, and we don't
932 * want to delete valid links to a file from the subvolume if that
933 * link is also in the log.
934 */
935static noinline int backref_in_log(struct btrfs_root *log,
936 struct btrfs_key *key,
Mark Fashehf1863732012-08-08 11:32:27 -0700937 u64 ref_objectid,
Filipe Mananadf8d1162015-01-14 01:52:25 +0000938 const char *name, int namelen)
Chris Masone02119d2008-09-05 16:13:11 -0400939{
940 struct btrfs_path *path;
941 struct btrfs_inode_ref *ref;
942 unsigned long ptr;
943 unsigned long ptr_end;
944 unsigned long name_ptr;
945 int found_name_len;
946 int item_size;
947 int ret;
948 int match = 0;
949
950 path = btrfs_alloc_path();
liubo2a29edc2011-01-26 06:22:08 +0000951 if (!path)
952 return -ENOMEM;
953
Chris Masone02119d2008-09-05 16:13:11 -0400954 ret = btrfs_search_slot(NULL, log, key, path, 0, 0);
955 if (ret != 0)
956 goto out;
957
Chris Masone02119d2008-09-05 16:13:11 -0400958 ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
Mark Fashehf1863732012-08-08 11:32:27 -0700959
960 if (key->type == BTRFS_INODE_EXTREF_KEY) {
961 if (btrfs_find_name_in_ext_backref(path, ref_objectid,
962 name, namelen, NULL))
963 match = 1;
964
965 goto out;
966 }
967
968 item_size = btrfs_item_size_nr(path->nodes[0], path->slots[0]);
Chris Masone02119d2008-09-05 16:13:11 -0400969 ptr_end = ptr + item_size;
970 while (ptr < ptr_end) {
971 ref = (struct btrfs_inode_ref *)ptr;
972 found_name_len = btrfs_inode_ref_name_len(path->nodes[0], ref);
973 if (found_name_len == namelen) {
974 name_ptr = (unsigned long)(ref + 1);
975 ret = memcmp_extent_buffer(path->nodes[0], name,
976 name_ptr, namelen);
977 if (ret == 0) {
978 match = 1;
979 goto out;
980 }
981 }
982 ptr = (unsigned long)(ref + 1) + found_name_len;
983 }
984out:
985 btrfs_free_path(path);
986 return match;
987}
988
Jan Schmidt5a1d7842012-08-17 14:04:41 -0700989static inline int __add_inode_ref(struct btrfs_trans_handle *trans,
990 struct btrfs_root *root,
991 struct btrfs_path *path,
992 struct btrfs_root *log_root,
993 struct inode *dir, struct inode *inode,
Jan Schmidt5a1d7842012-08-17 14:04:41 -0700994 struct extent_buffer *eb,
Mark Fashehf1863732012-08-08 11:32:27 -0700995 u64 inode_objectid, u64 parent_objectid,
996 u64 ref_index, char *name, int namelen,
997 int *search_done)
Jan Schmidt5a1d7842012-08-17 14:04:41 -0700998{
Jeff Mahoney2ff7e612016-06-22 18:54:24 -0400999 struct btrfs_fs_info *fs_info = root->fs_info;
Jan Schmidt5a1d7842012-08-17 14:04:41 -07001000 int ret;
Mark Fashehf1863732012-08-08 11:32:27 -07001001 char *victim_name;
1002 int victim_name_len;
1003 struct extent_buffer *leaf;
Jan Schmidt5a1d7842012-08-17 14:04:41 -07001004 struct btrfs_dir_item *di;
Mark Fashehf1863732012-08-08 11:32:27 -07001005 struct btrfs_key search_key;
1006 struct btrfs_inode_extref *extref;
Jan Schmidt5a1d7842012-08-17 14:04:41 -07001007
Mark Fashehf1863732012-08-08 11:32:27 -07001008again:
1009 /* Search old style refs */
1010 search_key.objectid = inode_objectid;
1011 search_key.type = BTRFS_INODE_REF_KEY;
1012 search_key.offset = parent_objectid;
1013 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
Jan Schmidt5a1d7842012-08-17 14:04:41 -07001014 if (ret == 0) {
Jan Schmidt5a1d7842012-08-17 14:04:41 -07001015 struct btrfs_inode_ref *victim_ref;
1016 unsigned long ptr;
1017 unsigned long ptr_end;
Mark Fashehf1863732012-08-08 11:32:27 -07001018
1019 leaf = path->nodes[0];
Jan Schmidt5a1d7842012-08-17 14:04:41 -07001020
1021 /* are we trying to overwrite a back ref for the root directory
1022 * if so, just jump out, we're done
1023 */
Mark Fashehf1863732012-08-08 11:32:27 -07001024 if (search_key.objectid == search_key.offset)
Jan Schmidt5a1d7842012-08-17 14:04:41 -07001025 return 1;
1026
1027 /* check all the names in this back reference to see
1028 * if they are in the log. if so, we allow them to stay
1029 * otherwise they must be unlinked as a conflict
1030 */
1031 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
1032 ptr_end = ptr + btrfs_item_size_nr(leaf, path->slots[0]);
1033 while (ptr < ptr_end) {
1034 victim_ref = (struct btrfs_inode_ref *)ptr;
1035 victim_name_len = btrfs_inode_ref_name_len(leaf,
1036 victim_ref);
1037 victim_name = kmalloc(victim_name_len, GFP_NOFS);
Josef Bacik36508602013-04-25 16:23:32 -04001038 if (!victim_name)
1039 return -ENOMEM;
Jan Schmidt5a1d7842012-08-17 14:04:41 -07001040
1041 read_extent_buffer(leaf, victim_name,
1042 (unsigned long)(victim_ref + 1),
1043 victim_name_len);
1044
Mark Fashehf1863732012-08-08 11:32:27 -07001045 if (!backref_in_log(log_root, &search_key,
1046 parent_objectid,
1047 victim_name,
Jan Schmidt5a1d7842012-08-17 14:04:41 -07001048 victim_name_len)) {
Zach Brown8b558c52013-10-16 12:10:34 -07001049 inc_nlink(inode);
Jan Schmidt5a1d7842012-08-17 14:04:41 -07001050 btrfs_release_path(path);
1051
1052 ret = btrfs_unlink_inode(trans, root, dir,
1053 inode, victim_name,
1054 victim_name_len);
Mark Fashehf1863732012-08-08 11:32:27 -07001055 kfree(victim_name);
Josef Bacik36508602013-04-25 16:23:32 -04001056 if (ret)
1057 return ret;
Jeff Mahoney2ff7e612016-06-22 18:54:24 -04001058 ret = btrfs_run_delayed_items(trans, fs_info);
Filipe David Borba Mananaada9af22013-08-05 09:25:47 +01001059 if (ret)
1060 return ret;
Mark Fashehf1863732012-08-08 11:32:27 -07001061 *search_done = 1;
1062 goto again;
Jan Schmidt5a1d7842012-08-17 14:04:41 -07001063 }
1064 kfree(victim_name);
Mark Fashehf1863732012-08-08 11:32:27 -07001065
Jan Schmidt5a1d7842012-08-17 14:04:41 -07001066 ptr = (unsigned long)(victim_ref + 1) + victim_name_len;
1067 }
Jan Schmidt5a1d7842012-08-17 14:04:41 -07001068
1069 /*
1070 * NOTE: we have searched root tree and checked the
Adam Buchbinderbb7ab3b2016-03-04 11:23:12 -08001071 * corresponding ref, it does not need to check again.
Jan Schmidt5a1d7842012-08-17 14:04:41 -07001072 */
1073 *search_done = 1;
1074 }
1075 btrfs_release_path(path);
1076
Mark Fashehf1863732012-08-08 11:32:27 -07001077 /* Same search but for extended refs */
1078 extref = btrfs_lookup_inode_extref(NULL, root, path, name, namelen,
1079 inode_objectid, parent_objectid, 0,
1080 0);
1081 if (!IS_ERR_OR_NULL(extref)) {
1082 u32 item_size;
1083 u32 cur_offset = 0;
1084 unsigned long base;
1085 struct inode *victim_parent;
1086
1087 leaf = path->nodes[0];
1088
1089 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1090 base = btrfs_item_ptr_offset(leaf, path->slots[0]);
1091
1092 while (cur_offset < item_size) {
Quentin Casasnovasdd9ef132015-03-03 16:31:38 +01001093 extref = (struct btrfs_inode_extref *)(base + cur_offset);
Mark Fashehf1863732012-08-08 11:32:27 -07001094
1095 victim_name_len = btrfs_inode_extref_name_len(leaf, extref);
1096
1097 if (btrfs_inode_extref_parent(leaf, extref) != parent_objectid)
1098 goto next;
1099
1100 victim_name = kmalloc(victim_name_len, GFP_NOFS);
Josef Bacik36508602013-04-25 16:23:32 -04001101 if (!victim_name)
1102 return -ENOMEM;
Mark Fashehf1863732012-08-08 11:32:27 -07001103 read_extent_buffer(leaf, victim_name, (unsigned long)&extref->name,
1104 victim_name_len);
1105
1106 search_key.objectid = inode_objectid;
1107 search_key.type = BTRFS_INODE_EXTREF_KEY;
1108 search_key.offset = btrfs_extref_hash(parent_objectid,
1109 victim_name,
1110 victim_name_len);
1111 ret = 0;
1112 if (!backref_in_log(log_root, &search_key,
1113 parent_objectid, victim_name,
1114 victim_name_len)) {
1115 ret = -ENOENT;
1116 victim_parent = read_one_inode(root,
1117 parent_objectid);
1118 if (victim_parent) {
Zach Brown8b558c52013-10-16 12:10:34 -07001119 inc_nlink(inode);
Mark Fashehf1863732012-08-08 11:32:27 -07001120 btrfs_release_path(path);
1121
1122 ret = btrfs_unlink_inode(trans, root,
1123 victim_parent,
1124 inode,
1125 victim_name,
1126 victim_name_len);
Filipe David Borba Mananaada9af22013-08-05 09:25:47 +01001127 if (!ret)
1128 ret = btrfs_run_delayed_items(
Jeff Mahoney2ff7e612016-06-22 18:54:24 -04001129 trans,
1130 fs_info);
Mark Fashehf1863732012-08-08 11:32:27 -07001131 }
Mark Fashehf1863732012-08-08 11:32:27 -07001132 iput(victim_parent);
1133 kfree(victim_name);
Josef Bacik36508602013-04-25 16:23:32 -04001134 if (ret)
1135 return ret;
Mark Fashehf1863732012-08-08 11:32:27 -07001136 *search_done = 1;
1137 goto again;
1138 }
1139 kfree(victim_name);
Josef Bacik36508602013-04-25 16:23:32 -04001140 if (ret)
1141 return ret;
Mark Fashehf1863732012-08-08 11:32:27 -07001142next:
1143 cur_offset += victim_name_len + sizeof(*extref);
1144 }
1145 *search_done = 1;
1146 }
1147 btrfs_release_path(path);
1148
Jan Schmidt5a1d7842012-08-17 14:04:41 -07001149 /* look for a conflicting sequence number */
1150 di = btrfs_lookup_dir_index_item(trans, root, path, btrfs_ino(dir),
Mark Fashehf1863732012-08-08 11:32:27 -07001151 ref_index, name, namelen, 0);
Jan Schmidt5a1d7842012-08-17 14:04:41 -07001152 if (di && !IS_ERR(di)) {
1153 ret = drop_one_dir_item(trans, root, path, dir, di);
Josef Bacik36508602013-04-25 16:23:32 -04001154 if (ret)
1155 return ret;
Jan Schmidt5a1d7842012-08-17 14:04:41 -07001156 }
1157 btrfs_release_path(path);
1158
1159 /* look for a conflicing name */
1160 di = btrfs_lookup_dir_item(trans, root, path, btrfs_ino(dir),
1161 name, namelen, 0);
1162 if (di && !IS_ERR(di)) {
1163 ret = drop_one_dir_item(trans, root, path, dir, di);
Josef Bacik36508602013-04-25 16:23:32 -04001164 if (ret)
1165 return ret;
Jan Schmidt5a1d7842012-08-17 14:04:41 -07001166 }
1167 btrfs_release_path(path);
1168
1169 return 0;
1170}
Chris Masone02119d2008-09-05 16:13:11 -04001171
Mark Fashehf1863732012-08-08 11:32:27 -07001172static int extref_get_fields(struct extent_buffer *eb, unsigned long ref_ptr,
1173 u32 *namelen, char **name, u64 *index,
1174 u64 *parent_objectid)
1175{
1176 struct btrfs_inode_extref *extref;
1177
1178 extref = (struct btrfs_inode_extref *)ref_ptr;
1179
1180 *namelen = btrfs_inode_extref_name_len(eb, extref);
1181 *name = kmalloc(*namelen, GFP_NOFS);
1182 if (*name == NULL)
1183 return -ENOMEM;
1184
1185 read_extent_buffer(eb, *name, (unsigned long)&extref->name,
1186 *namelen);
1187
1188 *index = btrfs_inode_extref_index(eb, extref);
1189 if (parent_objectid)
1190 *parent_objectid = btrfs_inode_extref_parent(eb, extref);
1191
1192 return 0;
1193}
1194
1195static int ref_get_fields(struct extent_buffer *eb, unsigned long ref_ptr,
1196 u32 *namelen, char **name, u64 *index)
1197{
1198 struct btrfs_inode_ref *ref;
1199
1200 ref = (struct btrfs_inode_ref *)ref_ptr;
1201
1202 *namelen = btrfs_inode_ref_name_len(eb, ref);
1203 *name = kmalloc(*namelen, GFP_NOFS);
1204 if (*name == NULL)
1205 return -ENOMEM;
1206
1207 read_extent_buffer(eb, *name, (unsigned long)(ref + 1), *namelen);
1208
1209 *index = btrfs_inode_ref_index(eb, ref);
1210
1211 return 0;
1212}
1213
Chris Masone02119d2008-09-05 16:13:11 -04001214/*
1215 * replay one inode back reference item found in the log tree.
1216 * eb, slot and key refer to the buffer and key found in the log tree.
1217 * root is the destination we are replaying into, and path is for temp
1218 * use by this function. (it should be released on return).
1219 */
1220static noinline int add_inode_ref(struct btrfs_trans_handle *trans,
1221 struct btrfs_root *root,
1222 struct btrfs_root *log,
1223 struct btrfs_path *path,
1224 struct extent_buffer *eb, int slot,
1225 struct btrfs_key *key)
1226{
Geyslan G. Bem03b2f082013-10-11 15:35:45 -03001227 struct inode *dir = NULL;
1228 struct inode *inode = NULL;
Chris Masone02119d2008-09-05 16:13:11 -04001229 unsigned long ref_ptr;
1230 unsigned long ref_end;
Geyslan G. Bem03b2f082013-10-11 15:35:45 -03001231 char *name = NULL;
liubo34f3e4f2011-08-06 08:35:23 +00001232 int namelen;
1233 int ret;
liuboc622ae62011-03-26 08:01:12 -04001234 int search_done = 0;
Mark Fashehf1863732012-08-08 11:32:27 -07001235 int log_ref_ver = 0;
1236 u64 parent_objectid;
1237 u64 inode_objectid;
Chris Masonf46dbe32012-10-09 11:17:20 -04001238 u64 ref_index = 0;
Mark Fashehf1863732012-08-08 11:32:27 -07001239 int ref_struct_size;
1240
1241 ref_ptr = btrfs_item_ptr_offset(eb, slot);
1242 ref_end = ref_ptr + btrfs_item_size_nr(eb, slot);
1243
1244 if (key->type == BTRFS_INODE_EXTREF_KEY) {
1245 struct btrfs_inode_extref *r;
1246
1247 ref_struct_size = sizeof(struct btrfs_inode_extref);
1248 log_ref_ver = 1;
1249 r = (struct btrfs_inode_extref *)ref_ptr;
1250 parent_objectid = btrfs_inode_extref_parent(eb, r);
1251 } else {
1252 ref_struct_size = sizeof(struct btrfs_inode_ref);
1253 parent_objectid = key->offset;
1254 }
1255 inode_objectid = key->objectid;
Chris Masone02119d2008-09-05 16:13:11 -04001256
Chris Masone02119d2008-09-05 16:13:11 -04001257 /*
1258 * it is possible that we didn't log all the parent directories
1259 * for a given inode. If we don't find the dir, just don't
1260 * copy the back ref in. The link count fixup code will take
1261 * care of the rest
1262 */
Mark Fashehf1863732012-08-08 11:32:27 -07001263 dir = read_one_inode(root, parent_objectid);
Geyslan G. Bem03b2f082013-10-11 15:35:45 -03001264 if (!dir) {
1265 ret = -ENOENT;
1266 goto out;
1267 }
Chris Masone02119d2008-09-05 16:13:11 -04001268
Mark Fashehf1863732012-08-08 11:32:27 -07001269 inode = read_one_inode(root, inode_objectid);
Tsutomu Itohc00e9492011-04-28 09:10:23 +00001270 if (!inode) {
Geyslan G. Bem03b2f082013-10-11 15:35:45 -03001271 ret = -EIO;
1272 goto out;
Tsutomu Itohc00e9492011-04-28 09:10:23 +00001273 }
Chris Masone02119d2008-09-05 16:13:11 -04001274
Jan Schmidt5a1d7842012-08-17 14:04:41 -07001275 while (ref_ptr < ref_end) {
Mark Fashehf1863732012-08-08 11:32:27 -07001276 if (log_ref_ver) {
1277 ret = extref_get_fields(eb, ref_ptr, &namelen, &name,
1278 &ref_index, &parent_objectid);
1279 /*
1280 * parent object can change from one array
1281 * item to another.
1282 */
1283 if (!dir)
1284 dir = read_one_inode(root, parent_objectid);
Geyslan G. Bem03b2f082013-10-11 15:35:45 -03001285 if (!dir) {
1286 ret = -ENOENT;
1287 goto out;
1288 }
Mark Fashehf1863732012-08-08 11:32:27 -07001289 } else {
1290 ret = ref_get_fields(eb, ref_ptr, &namelen, &name,
1291 &ref_index);
1292 }
1293 if (ret)
Geyslan G. Bem03b2f082013-10-11 15:35:45 -03001294 goto out;
Chris Masone02119d2008-09-05 16:13:11 -04001295
Jan Schmidt5a1d7842012-08-17 14:04:41 -07001296 /* if we already have a perfect match, we're done */
1297 if (!inode_in_dir(root, path, btrfs_ino(dir), btrfs_ino(inode),
Mark Fashehf1863732012-08-08 11:32:27 -07001298 ref_index, name, namelen)) {
Jan Schmidt5a1d7842012-08-17 14:04:41 -07001299 /*
1300 * look for a conflicting back reference in the
1301 * metadata. if we find one we have to unlink that name
1302 * of the file before we add our new link. Later on, we
1303 * overwrite any existing back reference, and we don't
1304 * want to create dangling pointers in the directory.
1305 */
Chris Masone02119d2008-09-05 16:13:11 -04001306
Jan Schmidt5a1d7842012-08-17 14:04:41 -07001307 if (!search_done) {
1308 ret = __add_inode_ref(trans, root, path, log,
Mark Fashehf1863732012-08-08 11:32:27 -07001309 dir, inode, eb,
1310 inode_objectid,
1311 parent_objectid,
1312 ref_index, name, namelen,
Jan Schmidt5a1d7842012-08-17 14:04:41 -07001313 &search_done);
Geyslan G. Bem03b2f082013-10-11 15:35:45 -03001314 if (ret) {
1315 if (ret == 1)
1316 ret = 0;
Jan Schmidt5a1d7842012-08-17 14:04:41 -07001317 goto out;
Josef Bacik36508602013-04-25 16:23:32 -04001318 }
Chris Masone02119d2008-09-05 16:13:11 -04001319 }
Jan Schmidt5a1d7842012-08-17 14:04:41 -07001320
1321 /* insert our name */
1322 ret = btrfs_add_link(trans, dir, inode, name, namelen,
Mark Fashehf1863732012-08-08 11:32:27 -07001323 0, ref_index);
Josef Bacik36508602013-04-25 16:23:32 -04001324 if (ret)
1325 goto out;
Jan Schmidt5a1d7842012-08-17 14:04:41 -07001326
1327 btrfs_update_inode(trans, root, inode);
Chris Masone02119d2008-09-05 16:13:11 -04001328 }
liuboc622ae62011-03-26 08:01:12 -04001329
Mark Fashehf1863732012-08-08 11:32:27 -07001330 ref_ptr = (unsigned long)(ref_ptr + ref_struct_size) + namelen;
Jan Schmidt5a1d7842012-08-17 14:04:41 -07001331 kfree(name);
Geyslan G. Bem03b2f082013-10-11 15:35:45 -03001332 name = NULL;
Mark Fashehf1863732012-08-08 11:32:27 -07001333 if (log_ref_ver) {
1334 iput(dir);
1335 dir = NULL;
1336 }
Chris Masone02119d2008-09-05 16:13:11 -04001337 }
Chris Masone02119d2008-09-05 16:13:11 -04001338
1339 /* finally write the back reference in the inode */
1340 ret = overwrite_item(trans, root, path, eb, slot, key);
Jan Schmidt5a1d7842012-08-17 14:04:41 -07001341out:
David Sterbab3b4aa72011-04-21 01:20:15 +02001342 btrfs_release_path(path);
Geyslan G. Bem03b2f082013-10-11 15:35:45 -03001343 kfree(name);
Chris Masone02119d2008-09-05 16:13:11 -04001344 iput(dir);
1345 iput(inode);
Josef Bacik36508602013-04-25 16:23:32 -04001346 return ret;
Chris Masone02119d2008-09-05 16:13:11 -04001347}
1348
Yan, Zhengc71bf092009-11-12 09:34:40 +00001349static int insert_orphan_item(struct btrfs_trans_handle *trans,
David Sterba9c4f61f2015-01-02 19:12:57 +01001350 struct btrfs_root *root, u64 ino)
Yan, Zhengc71bf092009-11-12 09:34:40 +00001351{
1352 int ret;
David Sterba381cf652015-01-02 18:45:16 +01001353
David Sterba9c4f61f2015-01-02 19:12:57 +01001354 ret = btrfs_insert_orphan_item(trans, root, ino);
1355 if (ret == -EEXIST)
1356 ret = 0;
David Sterba381cf652015-01-02 18:45:16 +01001357
Yan, Zhengc71bf092009-11-12 09:34:40 +00001358 return ret;
1359}
1360
Mark Fashehf1863732012-08-08 11:32:27 -07001361static int count_inode_extrefs(struct btrfs_root *root,
1362 struct inode *inode, struct btrfs_path *path)
Chris Masone02119d2008-09-05 16:13:11 -04001363{
Mark Fashehf1863732012-08-08 11:32:27 -07001364 int ret = 0;
1365 int name_len;
1366 unsigned int nlink = 0;
1367 u32 item_size;
1368 u32 cur_offset = 0;
1369 u64 inode_objectid = btrfs_ino(inode);
1370 u64 offset = 0;
1371 unsigned long ptr;
1372 struct btrfs_inode_extref *extref;
1373 struct extent_buffer *leaf;
1374
1375 while (1) {
1376 ret = btrfs_find_one_extref(root, inode_objectid, offset, path,
1377 &extref, &offset);
1378 if (ret)
1379 break;
1380
1381 leaf = path->nodes[0];
1382 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1383 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
Filipe Manana2c2c4522015-01-13 16:40:04 +00001384 cur_offset = 0;
Mark Fashehf1863732012-08-08 11:32:27 -07001385
1386 while (cur_offset < item_size) {
1387 extref = (struct btrfs_inode_extref *) (ptr + cur_offset);
1388 name_len = btrfs_inode_extref_name_len(leaf, extref);
1389
1390 nlink++;
1391
1392 cur_offset += name_len + sizeof(*extref);
1393 }
1394
1395 offset++;
1396 btrfs_release_path(path);
1397 }
1398 btrfs_release_path(path);
1399
Filipe Manana2c2c4522015-01-13 16:40:04 +00001400 if (ret < 0 && ret != -ENOENT)
Mark Fashehf1863732012-08-08 11:32:27 -07001401 return ret;
1402 return nlink;
1403}
1404
1405static int count_inode_refs(struct btrfs_root *root,
1406 struct inode *inode, struct btrfs_path *path)
1407{
Chris Masone02119d2008-09-05 16:13:11 -04001408 int ret;
1409 struct btrfs_key key;
Mark Fashehf1863732012-08-08 11:32:27 -07001410 unsigned int nlink = 0;
Chris Masone02119d2008-09-05 16:13:11 -04001411 unsigned long ptr;
1412 unsigned long ptr_end;
1413 int name_len;
Li Zefan33345d012011-04-20 10:31:50 +08001414 u64 ino = btrfs_ino(inode);
Chris Masone02119d2008-09-05 16:13:11 -04001415
Li Zefan33345d012011-04-20 10:31:50 +08001416 key.objectid = ino;
Chris Masone02119d2008-09-05 16:13:11 -04001417 key.type = BTRFS_INODE_REF_KEY;
1418 key.offset = (u64)-1;
1419
Chris Masond3977122009-01-05 21:25:51 -05001420 while (1) {
Chris Masone02119d2008-09-05 16:13:11 -04001421 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1422 if (ret < 0)
1423 break;
1424 if (ret > 0) {
1425 if (path->slots[0] == 0)
1426 break;
1427 path->slots[0]--;
1428 }
Filipe David Borba Mananae93ae262013-10-14 22:49:11 +01001429process_slot:
Chris Masone02119d2008-09-05 16:13:11 -04001430 btrfs_item_key_to_cpu(path->nodes[0], &key,
1431 path->slots[0]);
Li Zefan33345d012011-04-20 10:31:50 +08001432 if (key.objectid != ino ||
Chris Masone02119d2008-09-05 16:13:11 -04001433 key.type != BTRFS_INODE_REF_KEY)
1434 break;
1435 ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
1436 ptr_end = ptr + btrfs_item_size_nr(path->nodes[0],
1437 path->slots[0]);
Chris Masond3977122009-01-05 21:25:51 -05001438 while (ptr < ptr_end) {
Chris Masone02119d2008-09-05 16:13:11 -04001439 struct btrfs_inode_ref *ref;
1440
1441 ref = (struct btrfs_inode_ref *)ptr;
1442 name_len = btrfs_inode_ref_name_len(path->nodes[0],
1443 ref);
1444 ptr = (unsigned long)(ref + 1) + name_len;
1445 nlink++;
1446 }
1447
1448 if (key.offset == 0)
1449 break;
Filipe David Borba Mananae93ae262013-10-14 22:49:11 +01001450 if (path->slots[0] > 0) {
1451 path->slots[0]--;
1452 goto process_slot;
1453 }
Chris Masone02119d2008-09-05 16:13:11 -04001454 key.offset--;
David Sterbab3b4aa72011-04-21 01:20:15 +02001455 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04001456 }
David Sterbab3b4aa72011-04-21 01:20:15 +02001457 btrfs_release_path(path);
Mark Fashehf1863732012-08-08 11:32:27 -07001458
1459 return nlink;
1460}
1461
1462/*
1463 * There are a few corners where the link count of the file can't
1464 * be properly maintained during replay. So, instead of adding
1465 * lots of complexity to the log code, we just scan the backrefs
1466 * for any file that has been through replay.
1467 *
1468 * The scan will update the link count on the inode to reflect the
1469 * number of back refs found. If it goes down to zero, the iput
1470 * will free the inode.
1471 */
1472static noinline int fixup_inode_link_count(struct btrfs_trans_handle *trans,
1473 struct btrfs_root *root,
1474 struct inode *inode)
1475{
1476 struct btrfs_path *path;
1477 int ret;
1478 u64 nlink = 0;
1479 u64 ino = btrfs_ino(inode);
1480
1481 path = btrfs_alloc_path();
1482 if (!path)
1483 return -ENOMEM;
1484
1485 ret = count_inode_refs(root, inode, path);
1486 if (ret < 0)
1487 goto out;
1488
1489 nlink = ret;
1490
1491 ret = count_inode_extrefs(root, inode, path);
Mark Fashehf1863732012-08-08 11:32:27 -07001492 if (ret < 0)
1493 goto out;
1494
1495 nlink += ret;
1496
1497 ret = 0;
1498
Chris Masone02119d2008-09-05 16:13:11 -04001499 if (nlink != inode->i_nlink) {
Miklos Szeredibfe86842011-10-28 14:13:29 +02001500 set_nlink(inode, nlink);
Chris Masone02119d2008-09-05 16:13:11 -04001501 btrfs_update_inode(trans, root, inode);
1502 }
Chris Mason8d5bf1c2008-09-11 15:51:21 -04001503 BTRFS_I(inode)->index_cnt = (u64)-1;
Chris Masone02119d2008-09-05 16:13:11 -04001504
Yan, Zhengc71bf092009-11-12 09:34:40 +00001505 if (inode->i_nlink == 0) {
1506 if (S_ISDIR(inode->i_mode)) {
1507 ret = replay_dir_deletes(trans, root, NULL, path,
Li Zefan33345d012011-04-20 10:31:50 +08001508 ino, 1);
Josef Bacik36508602013-04-25 16:23:32 -04001509 if (ret)
1510 goto out;
Yan, Zhengc71bf092009-11-12 09:34:40 +00001511 }
Li Zefan33345d012011-04-20 10:31:50 +08001512 ret = insert_orphan_item(trans, root, ino);
Chris Mason12fcfd22009-03-24 10:24:20 -04001513 }
Chris Mason12fcfd22009-03-24 10:24:20 -04001514
Mark Fashehf1863732012-08-08 11:32:27 -07001515out:
1516 btrfs_free_path(path);
1517 return ret;
Chris Masone02119d2008-09-05 16:13:11 -04001518}
1519
1520static noinline int fixup_inode_link_counts(struct btrfs_trans_handle *trans,
1521 struct btrfs_root *root,
1522 struct btrfs_path *path)
1523{
1524 int ret;
1525 struct btrfs_key key;
1526 struct inode *inode;
1527
1528 key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID;
1529 key.type = BTRFS_ORPHAN_ITEM_KEY;
1530 key.offset = (u64)-1;
Chris Masond3977122009-01-05 21:25:51 -05001531 while (1) {
Chris Masone02119d2008-09-05 16:13:11 -04001532 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1533 if (ret < 0)
1534 break;
1535
1536 if (ret == 1) {
1537 if (path->slots[0] == 0)
1538 break;
1539 path->slots[0]--;
1540 }
1541
1542 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1543 if (key.objectid != BTRFS_TREE_LOG_FIXUP_OBJECTID ||
1544 key.type != BTRFS_ORPHAN_ITEM_KEY)
1545 break;
1546
1547 ret = btrfs_del_item(trans, root, path);
Tsutomu Itoh65a246c2011-05-19 04:37:44 +00001548 if (ret)
1549 goto out;
Chris Masone02119d2008-09-05 16:13:11 -04001550
David Sterbab3b4aa72011-04-21 01:20:15 +02001551 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04001552 inode = read_one_inode(root, key.offset);
Tsutomu Itohc00e9492011-04-28 09:10:23 +00001553 if (!inode)
1554 return -EIO;
Chris Masone02119d2008-09-05 16:13:11 -04001555
1556 ret = fixup_inode_link_count(trans, root, inode);
Chris Masone02119d2008-09-05 16:13:11 -04001557 iput(inode);
Josef Bacik36508602013-04-25 16:23:32 -04001558 if (ret)
1559 goto out;
Chris Masone02119d2008-09-05 16:13:11 -04001560
Chris Mason12fcfd22009-03-24 10:24:20 -04001561 /*
1562 * fixup on a directory may create new entries,
1563 * make sure we always look for the highset possible
1564 * offset
1565 */
1566 key.offset = (u64)-1;
Chris Masone02119d2008-09-05 16:13:11 -04001567 }
Tsutomu Itoh65a246c2011-05-19 04:37:44 +00001568 ret = 0;
1569out:
David Sterbab3b4aa72011-04-21 01:20:15 +02001570 btrfs_release_path(path);
Tsutomu Itoh65a246c2011-05-19 04:37:44 +00001571 return ret;
Chris Masone02119d2008-09-05 16:13:11 -04001572}
1573
1574
1575/*
1576 * record a given inode in the fixup dir so we can check its link
1577 * count when replay is done. The link count is incremented here
1578 * so the inode won't go away until we check it
1579 */
1580static noinline int link_to_fixup_dir(struct btrfs_trans_handle *trans,
1581 struct btrfs_root *root,
1582 struct btrfs_path *path,
1583 u64 objectid)
1584{
1585 struct btrfs_key key;
1586 int ret = 0;
1587 struct inode *inode;
1588
1589 inode = read_one_inode(root, objectid);
Tsutomu Itohc00e9492011-04-28 09:10:23 +00001590 if (!inode)
1591 return -EIO;
Chris Masone02119d2008-09-05 16:13:11 -04001592
1593 key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID;
David Sterba962a2982014-06-04 18:41:45 +02001594 key.type = BTRFS_ORPHAN_ITEM_KEY;
Chris Masone02119d2008-09-05 16:13:11 -04001595 key.offset = objectid;
1596
1597 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1598
David Sterbab3b4aa72011-04-21 01:20:15 +02001599 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04001600 if (ret == 0) {
Josef Bacik9bf7a482013-03-01 13:35:47 -05001601 if (!inode->i_nlink)
1602 set_nlink(inode, 1);
1603 else
Zach Brown8b558c52013-10-16 12:10:34 -07001604 inc_nlink(inode);
Tsutomu Itohb9959292012-06-25 21:25:22 -06001605 ret = btrfs_update_inode(trans, root, inode);
Chris Masone02119d2008-09-05 16:13:11 -04001606 } else if (ret == -EEXIST) {
1607 ret = 0;
1608 } else {
Josef Bacik36508602013-04-25 16:23:32 -04001609 BUG(); /* Logic Error */
Chris Masone02119d2008-09-05 16:13:11 -04001610 }
1611 iput(inode);
1612
1613 return ret;
1614}
1615
1616/*
1617 * when replaying the log for a directory, we only insert names
1618 * for inodes that actually exist. This means an fsync on a directory
1619 * does not implicitly fsync all the new files in it
1620 */
1621static noinline int insert_one_name(struct btrfs_trans_handle *trans,
1622 struct btrfs_root *root,
Chris Masone02119d2008-09-05 16:13:11 -04001623 u64 dirid, u64 index,
Zhaolei60d53eb2015-08-17 18:44:46 +08001624 char *name, int name_len,
Chris Masone02119d2008-09-05 16:13:11 -04001625 struct btrfs_key *location)
1626{
1627 struct inode *inode;
1628 struct inode *dir;
1629 int ret;
1630
1631 inode = read_one_inode(root, location->objectid);
1632 if (!inode)
1633 return -ENOENT;
1634
1635 dir = read_one_inode(root, dirid);
1636 if (!dir) {
1637 iput(inode);
1638 return -EIO;
1639 }
Josef Bacikd5554382013-09-11 14:17:00 -04001640
Chris Masone02119d2008-09-05 16:13:11 -04001641 ret = btrfs_add_link(trans, dir, inode, name, name_len, 1, index);
1642
1643 /* FIXME, put inode into FIXUP list */
1644
1645 iput(inode);
1646 iput(dir);
1647 return ret;
1648}
1649
1650/*
Filipe Mananadf8d1162015-01-14 01:52:25 +00001651 * Return true if an inode reference exists in the log for the given name,
1652 * inode and parent inode.
1653 */
1654static bool name_in_log_ref(struct btrfs_root *log_root,
1655 const char *name, const int name_len,
1656 const u64 dirid, const u64 ino)
1657{
1658 struct btrfs_key search_key;
1659
1660 search_key.objectid = ino;
1661 search_key.type = BTRFS_INODE_REF_KEY;
1662 search_key.offset = dirid;
1663 if (backref_in_log(log_root, &search_key, dirid, name, name_len))
1664 return true;
1665
1666 search_key.type = BTRFS_INODE_EXTREF_KEY;
1667 search_key.offset = btrfs_extref_hash(dirid, name, name_len);
1668 if (backref_in_log(log_root, &search_key, dirid, name, name_len))
1669 return true;
1670
1671 return false;
1672}
1673
1674/*
Chris Masone02119d2008-09-05 16:13:11 -04001675 * take a single entry in a log directory item and replay it into
1676 * the subvolume.
1677 *
1678 * if a conflicting item exists in the subdirectory already,
1679 * the inode it points to is unlinked and put into the link count
1680 * fix up tree.
1681 *
1682 * If a name from the log points to a file or directory that does
1683 * not exist in the FS, it is skipped. fsyncs on directories
1684 * do not force down inodes inside that directory, just changes to the
1685 * names or unlinks in a directory.
Filipe Mananabb53eda2015-07-15 23:26:43 +01001686 *
1687 * Returns < 0 on error, 0 if the name wasn't replayed (dentry points to a
1688 * non-existing inode) and 1 if the name was replayed.
Chris Masone02119d2008-09-05 16:13:11 -04001689 */
1690static noinline int replay_one_name(struct btrfs_trans_handle *trans,
1691 struct btrfs_root *root,
1692 struct btrfs_path *path,
1693 struct extent_buffer *eb,
1694 struct btrfs_dir_item *di,
1695 struct btrfs_key *key)
1696{
1697 char *name;
1698 int name_len;
1699 struct btrfs_dir_item *dst_di;
1700 struct btrfs_key found_key;
1701 struct btrfs_key log_key;
1702 struct inode *dir;
Chris Masone02119d2008-09-05 16:13:11 -04001703 u8 log_type;
Chris Mason4bef0842008-09-08 11:18:08 -04001704 int exists;
Josef Bacik36508602013-04-25 16:23:32 -04001705 int ret = 0;
Josef Bacikd5554382013-09-11 14:17:00 -04001706 bool update_size = (key->type == BTRFS_DIR_INDEX_KEY);
Filipe Mananabb53eda2015-07-15 23:26:43 +01001707 bool name_added = false;
Chris Masone02119d2008-09-05 16:13:11 -04001708
1709 dir = read_one_inode(root, key->objectid);
Tsutomu Itohc00e9492011-04-28 09:10:23 +00001710 if (!dir)
1711 return -EIO;
Chris Masone02119d2008-09-05 16:13:11 -04001712
1713 name_len = btrfs_dir_name_len(eb, di);
1714 name = kmalloc(name_len, GFP_NOFS);
Filipe David Borba Manana2bac3252013-08-04 19:58:57 +01001715 if (!name) {
1716 ret = -ENOMEM;
1717 goto out;
1718 }
liubo2a29edc2011-01-26 06:22:08 +00001719
Chris Masone02119d2008-09-05 16:13:11 -04001720 log_type = btrfs_dir_type(eb, di);
1721 read_extent_buffer(eb, name, (unsigned long)(di + 1),
1722 name_len);
1723
1724 btrfs_dir_item_key_to_cpu(eb, di, &log_key);
Chris Mason4bef0842008-09-08 11:18:08 -04001725 exists = btrfs_lookup_inode(trans, root, path, &log_key, 0);
1726 if (exists == 0)
1727 exists = 1;
1728 else
1729 exists = 0;
David Sterbab3b4aa72011-04-21 01:20:15 +02001730 btrfs_release_path(path);
Chris Mason4bef0842008-09-08 11:18:08 -04001731
Chris Masone02119d2008-09-05 16:13:11 -04001732 if (key->type == BTRFS_DIR_ITEM_KEY) {
1733 dst_di = btrfs_lookup_dir_item(trans, root, path, key->objectid,
1734 name, name_len, 1);
Chris Masond3977122009-01-05 21:25:51 -05001735 } else if (key->type == BTRFS_DIR_INDEX_KEY) {
Chris Masone02119d2008-09-05 16:13:11 -04001736 dst_di = btrfs_lookup_dir_index_item(trans, root, path,
1737 key->objectid,
1738 key->offset, name,
1739 name_len, 1);
1740 } else {
Josef Bacik36508602013-04-25 16:23:32 -04001741 /* Corruption */
1742 ret = -EINVAL;
1743 goto out;
Chris Masone02119d2008-09-05 16:13:11 -04001744 }
David Sterbac7040052011-04-19 18:00:01 +02001745 if (IS_ERR_OR_NULL(dst_di)) {
Chris Masone02119d2008-09-05 16:13:11 -04001746 /* we need a sequence number to insert, so we only
1747 * do inserts for the BTRFS_DIR_INDEX_KEY types
1748 */
1749 if (key->type != BTRFS_DIR_INDEX_KEY)
1750 goto out;
1751 goto insert;
1752 }
1753
1754 btrfs_dir_item_key_to_cpu(path->nodes[0], dst_di, &found_key);
1755 /* the existing item matches the logged item */
1756 if (found_key.objectid == log_key.objectid &&
1757 found_key.type == log_key.type &&
1758 found_key.offset == log_key.offset &&
1759 btrfs_dir_type(path->nodes[0], dst_di) == log_type) {
Filipe Mananaa2cc11d2014-09-08 22:53:18 +01001760 update_size = false;
Chris Masone02119d2008-09-05 16:13:11 -04001761 goto out;
1762 }
1763
1764 /*
1765 * don't drop the conflicting directory entry if the inode
1766 * for the new entry doesn't exist
1767 */
Chris Mason4bef0842008-09-08 11:18:08 -04001768 if (!exists)
Chris Masone02119d2008-09-05 16:13:11 -04001769 goto out;
1770
Chris Masone02119d2008-09-05 16:13:11 -04001771 ret = drop_one_dir_item(trans, root, path, dir, dst_di);
Josef Bacik36508602013-04-25 16:23:32 -04001772 if (ret)
1773 goto out;
Chris Masone02119d2008-09-05 16:13:11 -04001774
1775 if (key->type == BTRFS_DIR_INDEX_KEY)
1776 goto insert;
1777out:
David Sterbab3b4aa72011-04-21 01:20:15 +02001778 btrfs_release_path(path);
Josef Bacikd5554382013-09-11 14:17:00 -04001779 if (!ret && update_size) {
1780 btrfs_i_size_write(dir, dir->i_size + name_len * 2);
1781 ret = btrfs_update_inode(trans, root, dir);
1782 }
Chris Masone02119d2008-09-05 16:13:11 -04001783 kfree(name);
1784 iput(dir);
Filipe Mananabb53eda2015-07-15 23:26:43 +01001785 if (!ret && name_added)
1786 ret = 1;
Josef Bacik36508602013-04-25 16:23:32 -04001787 return ret;
Chris Masone02119d2008-09-05 16:13:11 -04001788
1789insert:
Filipe Mananadf8d1162015-01-14 01:52:25 +00001790 if (name_in_log_ref(root->log_root, name, name_len,
1791 key->objectid, log_key.objectid)) {
1792 /* The dentry will be added later. */
1793 ret = 0;
1794 update_size = false;
1795 goto out;
1796 }
David Sterbab3b4aa72011-04-21 01:20:15 +02001797 btrfs_release_path(path);
Zhaolei60d53eb2015-08-17 18:44:46 +08001798 ret = insert_one_name(trans, root, key->objectid, key->offset,
1799 name, name_len, &log_key);
Filipe Mananadf8d1162015-01-14 01:52:25 +00001800 if (ret && ret != -ENOENT && ret != -EEXIST)
Josef Bacik36508602013-04-25 16:23:32 -04001801 goto out;
Filipe Mananabb53eda2015-07-15 23:26:43 +01001802 if (!ret)
1803 name_added = true;
Josef Bacikd5554382013-09-11 14:17:00 -04001804 update_size = false;
Josef Bacik36508602013-04-25 16:23:32 -04001805 ret = 0;
Chris Masone02119d2008-09-05 16:13:11 -04001806 goto out;
1807}
1808
1809/*
1810 * find all the names in a directory item and reconcile them into
1811 * the subvolume. Only BTRFS_DIR_ITEM_KEY types will have more than
1812 * one name in a directory item, but the same code gets used for
1813 * both directory index types
1814 */
1815static noinline int replay_one_dir_item(struct btrfs_trans_handle *trans,
1816 struct btrfs_root *root,
1817 struct btrfs_path *path,
1818 struct extent_buffer *eb, int slot,
1819 struct btrfs_key *key)
1820{
Jeff Mahoney2ff7e612016-06-22 18:54:24 -04001821 struct btrfs_fs_info *fs_info = root->fs_info;
Filipe Mananabb53eda2015-07-15 23:26:43 +01001822 int ret = 0;
Chris Masone02119d2008-09-05 16:13:11 -04001823 u32 item_size = btrfs_item_size_nr(eb, slot);
1824 struct btrfs_dir_item *di;
1825 int name_len;
1826 unsigned long ptr;
1827 unsigned long ptr_end;
Filipe Mananabb53eda2015-07-15 23:26:43 +01001828 struct btrfs_path *fixup_path = NULL;
Chris Masone02119d2008-09-05 16:13:11 -04001829
1830 ptr = btrfs_item_ptr_offset(eb, slot);
1831 ptr_end = ptr + item_size;
Chris Masond3977122009-01-05 21:25:51 -05001832 while (ptr < ptr_end) {
Chris Masone02119d2008-09-05 16:13:11 -04001833 di = (struct btrfs_dir_item *)ptr;
Jeff Mahoney2ff7e612016-06-22 18:54:24 -04001834 if (verify_dir_item(fs_info, eb, di))
Josef Bacik22a94d42011-03-16 16:47:17 -04001835 return -EIO;
Chris Masone02119d2008-09-05 16:13:11 -04001836 name_len = btrfs_dir_name_len(eb, di);
1837 ret = replay_one_name(trans, root, path, eb, di, key);
Filipe Mananabb53eda2015-07-15 23:26:43 +01001838 if (ret < 0)
1839 break;
Chris Masone02119d2008-09-05 16:13:11 -04001840 ptr = (unsigned long)(di + 1);
1841 ptr += name_len;
Filipe Mananabb53eda2015-07-15 23:26:43 +01001842
1843 /*
1844 * If this entry refers to a non-directory (directories can not
1845 * have a link count > 1) and it was added in the transaction
1846 * that was not committed, make sure we fixup the link count of
1847 * the inode it the entry points to. Otherwise something like
1848 * the following would result in a directory pointing to an
1849 * inode with a wrong link that does not account for this dir
1850 * entry:
1851 *
1852 * mkdir testdir
1853 * touch testdir/foo
1854 * touch testdir/bar
1855 * sync
1856 *
1857 * ln testdir/bar testdir/bar_link
1858 * ln testdir/foo testdir/foo_link
1859 * xfs_io -c "fsync" testdir/bar
1860 *
1861 * <power failure>
1862 *
1863 * mount fs, log replay happens
1864 *
1865 * File foo would remain with a link count of 1 when it has two
1866 * entries pointing to it in the directory testdir. This would
1867 * make it impossible to ever delete the parent directory has
1868 * it would result in stale dentries that can never be deleted.
1869 */
1870 if (ret == 1 && btrfs_dir_type(eb, di) != BTRFS_FT_DIR) {
1871 struct btrfs_key di_key;
1872
1873 if (!fixup_path) {
1874 fixup_path = btrfs_alloc_path();
1875 if (!fixup_path) {
1876 ret = -ENOMEM;
1877 break;
1878 }
1879 }
1880
1881 btrfs_dir_item_key_to_cpu(eb, di, &di_key);
1882 ret = link_to_fixup_dir(trans, root, fixup_path,
1883 di_key.objectid);
1884 if (ret)
1885 break;
1886 }
1887 ret = 0;
Chris Masone02119d2008-09-05 16:13:11 -04001888 }
Filipe Mananabb53eda2015-07-15 23:26:43 +01001889 btrfs_free_path(fixup_path);
1890 return ret;
Chris Masone02119d2008-09-05 16:13:11 -04001891}
1892
1893/*
1894 * directory replay has two parts. There are the standard directory
1895 * items in the log copied from the subvolume, and range items
1896 * created in the log while the subvolume was logged.
1897 *
1898 * The range items tell us which parts of the key space the log
1899 * is authoritative for. During replay, if a key in the subvolume
1900 * directory is in a logged range item, but not actually in the log
1901 * that means it was deleted from the directory before the fsync
1902 * and should be removed.
1903 */
1904static noinline int find_dir_range(struct btrfs_root *root,
1905 struct btrfs_path *path,
1906 u64 dirid, int key_type,
1907 u64 *start_ret, u64 *end_ret)
1908{
1909 struct btrfs_key key;
1910 u64 found_end;
1911 struct btrfs_dir_log_item *item;
1912 int ret;
1913 int nritems;
1914
1915 if (*start_ret == (u64)-1)
1916 return 1;
1917
1918 key.objectid = dirid;
1919 key.type = key_type;
1920 key.offset = *start_ret;
1921
1922 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1923 if (ret < 0)
1924 goto out;
1925 if (ret > 0) {
1926 if (path->slots[0] == 0)
1927 goto out;
1928 path->slots[0]--;
1929 }
1930 if (ret != 0)
1931 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1932
1933 if (key.type != key_type || key.objectid != dirid) {
1934 ret = 1;
1935 goto next;
1936 }
1937 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
1938 struct btrfs_dir_log_item);
1939 found_end = btrfs_dir_log_end(path->nodes[0], item);
1940
1941 if (*start_ret >= key.offset && *start_ret <= found_end) {
1942 ret = 0;
1943 *start_ret = key.offset;
1944 *end_ret = found_end;
1945 goto out;
1946 }
1947 ret = 1;
1948next:
1949 /* check the next slot in the tree to see if it is a valid item */
1950 nritems = btrfs_header_nritems(path->nodes[0]);
Robbie Ko2a7bf532016-10-07 17:30:47 +08001951 path->slots[0]++;
Chris Masone02119d2008-09-05 16:13:11 -04001952 if (path->slots[0] >= nritems) {
1953 ret = btrfs_next_leaf(root, path);
1954 if (ret)
1955 goto out;
Chris Masone02119d2008-09-05 16:13:11 -04001956 }
1957
1958 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1959
1960 if (key.type != key_type || key.objectid != dirid) {
1961 ret = 1;
1962 goto out;
1963 }
1964 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
1965 struct btrfs_dir_log_item);
1966 found_end = btrfs_dir_log_end(path->nodes[0], item);
1967 *start_ret = key.offset;
1968 *end_ret = found_end;
1969 ret = 0;
1970out:
David Sterbab3b4aa72011-04-21 01:20:15 +02001971 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04001972 return ret;
1973}
1974
1975/*
1976 * this looks for a given directory item in the log. If the directory
1977 * item is not in the log, the item is removed and the inode it points
1978 * to is unlinked
1979 */
1980static noinline int check_item_in_log(struct btrfs_trans_handle *trans,
1981 struct btrfs_root *root,
1982 struct btrfs_root *log,
1983 struct btrfs_path *path,
1984 struct btrfs_path *log_path,
1985 struct inode *dir,
1986 struct btrfs_key *dir_key)
1987{
Jeff Mahoney2ff7e612016-06-22 18:54:24 -04001988 struct btrfs_fs_info *fs_info = root->fs_info;
Chris Masone02119d2008-09-05 16:13:11 -04001989 int ret;
1990 struct extent_buffer *eb;
1991 int slot;
1992 u32 item_size;
1993 struct btrfs_dir_item *di;
1994 struct btrfs_dir_item *log_di;
1995 int name_len;
1996 unsigned long ptr;
1997 unsigned long ptr_end;
1998 char *name;
1999 struct inode *inode;
2000 struct btrfs_key location;
2001
2002again:
2003 eb = path->nodes[0];
2004 slot = path->slots[0];
2005 item_size = btrfs_item_size_nr(eb, slot);
2006 ptr = btrfs_item_ptr_offset(eb, slot);
2007 ptr_end = ptr + item_size;
Chris Masond3977122009-01-05 21:25:51 -05002008 while (ptr < ptr_end) {
Chris Masone02119d2008-09-05 16:13:11 -04002009 di = (struct btrfs_dir_item *)ptr;
Jeff Mahoney2ff7e612016-06-22 18:54:24 -04002010 if (verify_dir_item(fs_info, eb, di)) {
Josef Bacik22a94d42011-03-16 16:47:17 -04002011 ret = -EIO;
2012 goto out;
2013 }
2014
Chris Masone02119d2008-09-05 16:13:11 -04002015 name_len = btrfs_dir_name_len(eb, di);
2016 name = kmalloc(name_len, GFP_NOFS);
2017 if (!name) {
2018 ret = -ENOMEM;
2019 goto out;
2020 }
2021 read_extent_buffer(eb, name, (unsigned long)(di + 1),
2022 name_len);
2023 log_di = NULL;
Chris Mason12fcfd22009-03-24 10:24:20 -04002024 if (log && dir_key->type == BTRFS_DIR_ITEM_KEY) {
Chris Masone02119d2008-09-05 16:13:11 -04002025 log_di = btrfs_lookup_dir_item(trans, log, log_path,
2026 dir_key->objectid,
2027 name, name_len, 0);
Chris Mason12fcfd22009-03-24 10:24:20 -04002028 } else if (log && dir_key->type == BTRFS_DIR_INDEX_KEY) {
Chris Masone02119d2008-09-05 16:13:11 -04002029 log_di = btrfs_lookup_dir_index_item(trans, log,
2030 log_path,
2031 dir_key->objectid,
2032 dir_key->offset,
2033 name, name_len, 0);
2034 }
Filipe David Borba Manana269d0402013-10-28 17:39:21 +00002035 if (!log_di || (IS_ERR(log_di) && PTR_ERR(log_di) == -ENOENT)) {
Chris Masone02119d2008-09-05 16:13:11 -04002036 btrfs_dir_item_key_to_cpu(eb, di, &location);
David Sterbab3b4aa72011-04-21 01:20:15 +02002037 btrfs_release_path(path);
2038 btrfs_release_path(log_path);
Chris Masone02119d2008-09-05 16:13:11 -04002039 inode = read_one_inode(root, location.objectid);
Tsutomu Itohc00e9492011-04-28 09:10:23 +00002040 if (!inode) {
2041 kfree(name);
2042 return -EIO;
2043 }
Chris Masone02119d2008-09-05 16:13:11 -04002044
2045 ret = link_to_fixup_dir(trans, root,
2046 path, location.objectid);
Josef Bacik36508602013-04-25 16:23:32 -04002047 if (ret) {
2048 kfree(name);
2049 iput(inode);
2050 goto out;
2051 }
2052
Zach Brown8b558c52013-10-16 12:10:34 -07002053 inc_nlink(inode);
Chris Masone02119d2008-09-05 16:13:11 -04002054 ret = btrfs_unlink_inode(trans, root, dir, inode,
2055 name, name_len);
Josef Bacik36508602013-04-25 16:23:32 -04002056 if (!ret)
Jeff Mahoney2ff7e612016-06-22 18:54:24 -04002057 ret = btrfs_run_delayed_items(trans, fs_info);
Chris Masone02119d2008-09-05 16:13:11 -04002058 kfree(name);
2059 iput(inode);
Josef Bacik36508602013-04-25 16:23:32 -04002060 if (ret)
2061 goto out;
Chris Masone02119d2008-09-05 16:13:11 -04002062
2063 /* there might still be more names under this key
2064 * check and repeat if required
2065 */
2066 ret = btrfs_search_slot(NULL, root, dir_key, path,
2067 0, 0);
2068 if (ret == 0)
2069 goto again;
2070 ret = 0;
2071 goto out;
Filipe David Borba Manana269d0402013-10-28 17:39:21 +00002072 } else if (IS_ERR(log_di)) {
2073 kfree(name);
2074 return PTR_ERR(log_di);
Chris Masone02119d2008-09-05 16:13:11 -04002075 }
David Sterbab3b4aa72011-04-21 01:20:15 +02002076 btrfs_release_path(log_path);
Chris Masone02119d2008-09-05 16:13:11 -04002077 kfree(name);
2078
2079 ptr = (unsigned long)(di + 1);
2080 ptr += name_len;
2081 }
2082 ret = 0;
2083out:
David Sterbab3b4aa72011-04-21 01:20:15 +02002084 btrfs_release_path(path);
2085 btrfs_release_path(log_path);
Chris Masone02119d2008-09-05 16:13:11 -04002086 return ret;
2087}
2088
Filipe Manana4f764e52015-02-23 19:53:35 +00002089static int replay_xattr_deletes(struct btrfs_trans_handle *trans,
2090 struct btrfs_root *root,
2091 struct btrfs_root *log,
2092 struct btrfs_path *path,
2093 const u64 ino)
2094{
2095 struct btrfs_key search_key;
2096 struct btrfs_path *log_path;
2097 int i;
2098 int nritems;
2099 int ret;
2100
2101 log_path = btrfs_alloc_path();
2102 if (!log_path)
2103 return -ENOMEM;
2104
2105 search_key.objectid = ino;
2106 search_key.type = BTRFS_XATTR_ITEM_KEY;
2107 search_key.offset = 0;
2108again:
2109 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
2110 if (ret < 0)
2111 goto out;
2112process_leaf:
2113 nritems = btrfs_header_nritems(path->nodes[0]);
2114 for (i = path->slots[0]; i < nritems; i++) {
2115 struct btrfs_key key;
2116 struct btrfs_dir_item *di;
2117 struct btrfs_dir_item *log_di;
2118 u32 total_size;
2119 u32 cur;
2120
2121 btrfs_item_key_to_cpu(path->nodes[0], &key, i);
2122 if (key.objectid != ino || key.type != BTRFS_XATTR_ITEM_KEY) {
2123 ret = 0;
2124 goto out;
2125 }
2126
2127 di = btrfs_item_ptr(path->nodes[0], i, struct btrfs_dir_item);
2128 total_size = btrfs_item_size_nr(path->nodes[0], i);
2129 cur = 0;
2130 while (cur < total_size) {
2131 u16 name_len = btrfs_dir_name_len(path->nodes[0], di);
2132 u16 data_len = btrfs_dir_data_len(path->nodes[0], di);
2133 u32 this_len = sizeof(*di) + name_len + data_len;
2134 char *name;
2135
2136 name = kmalloc(name_len, GFP_NOFS);
2137 if (!name) {
2138 ret = -ENOMEM;
2139 goto out;
2140 }
2141 read_extent_buffer(path->nodes[0], name,
2142 (unsigned long)(di + 1), name_len);
2143
2144 log_di = btrfs_lookup_xattr(NULL, log, log_path, ino,
2145 name, name_len, 0);
2146 btrfs_release_path(log_path);
2147 if (!log_di) {
2148 /* Doesn't exist in log tree, so delete it. */
2149 btrfs_release_path(path);
2150 di = btrfs_lookup_xattr(trans, root, path, ino,
2151 name, name_len, -1);
2152 kfree(name);
2153 if (IS_ERR(di)) {
2154 ret = PTR_ERR(di);
2155 goto out;
2156 }
2157 ASSERT(di);
2158 ret = btrfs_delete_one_dir_name(trans, root,
2159 path, di);
2160 if (ret)
2161 goto out;
2162 btrfs_release_path(path);
2163 search_key = key;
2164 goto again;
2165 }
2166 kfree(name);
2167 if (IS_ERR(log_di)) {
2168 ret = PTR_ERR(log_di);
2169 goto out;
2170 }
2171 cur += this_len;
2172 di = (struct btrfs_dir_item *)((char *)di + this_len);
2173 }
2174 }
2175 ret = btrfs_next_leaf(root, path);
2176 if (ret > 0)
2177 ret = 0;
2178 else if (ret == 0)
2179 goto process_leaf;
2180out:
2181 btrfs_free_path(log_path);
2182 btrfs_release_path(path);
2183 return ret;
2184}
2185
2186
Chris Masone02119d2008-09-05 16:13:11 -04002187/*
2188 * deletion replay happens before we copy any new directory items
2189 * out of the log or out of backreferences from inodes. It
2190 * scans the log to find ranges of keys that log is authoritative for,
2191 * and then scans the directory to find items in those ranges that are
2192 * not present in the log.
2193 *
2194 * Anything we don't find in the log is unlinked and removed from the
2195 * directory.
2196 */
2197static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans,
2198 struct btrfs_root *root,
2199 struct btrfs_root *log,
2200 struct btrfs_path *path,
Chris Mason12fcfd22009-03-24 10:24:20 -04002201 u64 dirid, int del_all)
Chris Masone02119d2008-09-05 16:13:11 -04002202{
2203 u64 range_start;
2204 u64 range_end;
2205 int key_type = BTRFS_DIR_LOG_ITEM_KEY;
2206 int ret = 0;
2207 struct btrfs_key dir_key;
2208 struct btrfs_key found_key;
2209 struct btrfs_path *log_path;
2210 struct inode *dir;
2211
2212 dir_key.objectid = dirid;
2213 dir_key.type = BTRFS_DIR_ITEM_KEY;
2214 log_path = btrfs_alloc_path();
2215 if (!log_path)
2216 return -ENOMEM;
2217
2218 dir = read_one_inode(root, dirid);
2219 /* it isn't an error if the inode isn't there, that can happen
2220 * because we replay the deletes before we copy in the inode item
2221 * from the log
2222 */
2223 if (!dir) {
2224 btrfs_free_path(log_path);
2225 return 0;
2226 }
2227again:
2228 range_start = 0;
2229 range_end = 0;
Chris Masond3977122009-01-05 21:25:51 -05002230 while (1) {
Chris Mason12fcfd22009-03-24 10:24:20 -04002231 if (del_all)
2232 range_end = (u64)-1;
2233 else {
2234 ret = find_dir_range(log, path, dirid, key_type,
2235 &range_start, &range_end);
2236 if (ret != 0)
2237 break;
2238 }
Chris Masone02119d2008-09-05 16:13:11 -04002239
2240 dir_key.offset = range_start;
Chris Masond3977122009-01-05 21:25:51 -05002241 while (1) {
Chris Masone02119d2008-09-05 16:13:11 -04002242 int nritems;
2243 ret = btrfs_search_slot(NULL, root, &dir_key, path,
2244 0, 0);
2245 if (ret < 0)
2246 goto out;
2247
2248 nritems = btrfs_header_nritems(path->nodes[0]);
2249 if (path->slots[0] >= nritems) {
2250 ret = btrfs_next_leaf(root, path);
2251 if (ret)
2252 break;
2253 }
2254 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2255 path->slots[0]);
2256 if (found_key.objectid != dirid ||
2257 found_key.type != dir_key.type)
2258 goto next_type;
2259
2260 if (found_key.offset > range_end)
2261 break;
2262
2263 ret = check_item_in_log(trans, root, log, path,
Chris Mason12fcfd22009-03-24 10:24:20 -04002264 log_path, dir,
2265 &found_key);
Josef Bacik36508602013-04-25 16:23:32 -04002266 if (ret)
2267 goto out;
Chris Masone02119d2008-09-05 16:13:11 -04002268 if (found_key.offset == (u64)-1)
2269 break;
2270 dir_key.offset = found_key.offset + 1;
2271 }
David Sterbab3b4aa72011-04-21 01:20:15 +02002272 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04002273 if (range_end == (u64)-1)
2274 break;
2275 range_start = range_end + 1;
2276 }
2277
2278next_type:
2279 ret = 0;
2280 if (key_type == BTRFS_DIR_LOG_ITEM_KEY) {
2281 key_type = BTRFS_DIR_LOG_INDEX_KEY;
2282 dir_key.type = BTRFS_DIR_INDEX_KEY;
David Sterbab3b4aa72011-04-21 01:20:15 +02002283 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04002284 goto again;
2285 }
2286out:
David Sterbab3b4aa72011-04-21 01:20:15 +02002287 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04002288 btrfs_free_path(log_path);
2289 iput(dir);
2290 return ret;
2291}
2292
2293/*
2294 * the process_func used to replay items from the log tree. This
2295 * gets called in two different stages. The first stage just looks
2296 * for inodes and makes sure they are all copied into the subvolume.
2297 *
2298 * The second stage copies all the other item types from the log into
2299 * the subvolume. The two stage approach is slower, but gets rid of
2300 * lots of complexity around inodes referencing other inodes that exist
2301 * only in the log (references come from either directory items or inode
2302 * back refs).
2303 */
2304static int replay_one_buffer(struct btrfs_root *log, struct extent_buffer *eb,
2305 struct walk_control *wc, u64 gen)
2306{
2307 int nritems;
2308 struct btrfs_path *path;
2309 struct btrfs_root *root = wc->replay_dest;
2310 struct btrfs_key key;
Chris Masone02119d2008-09-05 16:13:11 -04002311 int level;
2312 int i;
2313 int ret;
2314
Tsutomu Itoh018642a2012-05-29 18:10:13 +09002315 ret = btrfs_read_buffer(eb, gen);
2316 if (ret)
2317 return ret;
Chris Masone02119d2008-09-05 16:13:11 -04002318
2319 level = btrfs_header_level(eb);
2320
2321 if (level != 0)
2322 return 0;
2323
2324 path = btrfs_alloc_path();
Mark Fasheh1e5063d2011-07-12 10:46:06 -07002325 if (!path)
2326 return -ENOMEM;
Chris Masone02119d2008-09-05 16:13:11 -04002327
2328 nritems = btrfs_header_nritems(eb);
2329 for (i = 0; i < nritems; i++) {
2330 btrfs_item_key_to_cpu(eb, &key, i);
Chris Masone02119d2008-09-05 16:13:11 -04002331
2332 /* inode keys are done during the first stage */
2333 if (key.type == BTRFS_INODE_ITEM_KEY &&
2334 wc->stage == LOG_WALK_REPLAY_INODES) {
Chris Masone02119d2008-09-05 16:13:11 -04002335 struct btrfs_inode_item *inode_item;
2336 u32 mode;
2337
2338 inode_item = btrfs_item_ptr(eb, i,
2339 struct btrfs_inode_item);
Filipe Manana4f764e52015-02-23 19:53:35 +00002340 ret = replay_xattr_deletes(wc->trans, root, log,
2341 path, key.objectid);
2342 if (ret)
2343 break;
Chris Masone02119d2008-09-05 16:13:11 -04002344 mode = btrfs_inode_mode(eb, inode_item);
2345 if (S_ISDIR(mode)) {
2346 ret = replay_dir_deletes(wc->trans,
Chris Mason12fcfd22009-03-24 10:24:20 -04002347 root, log, path, key.objectid, 0);
Josef Bacikb50c6e22013-04-25 15:55:30 -04002348 if (ret)
2349 break;
Chris Masone02119d2008-09-05 16:13:11 -04002350 }
2351 ret = overwrite_item(wc->trans, root, path,
2352 eb, i, &key);
Josef Bacikb50c6e22013-04-25 15:55:30 -04002353 if (ret)
2354 break;
Chris Masone02119d2008-09-05 16:13:11 -04002355
Yan, Zhengc71bf092009-11-12 09:34:40 +00002356 /* for regular files, make sure corresponding
Nicholas D Steeves01327612016-05-19 21:18:45 -04002357 * orphan item exist. extents past the new EOF
Yan, Zhengc71bf092009-11-12 09:34:40 +00002358 * will be truncated later by orphan cleanup.
Chris Masone02119d2008-09-05 16:13:11 -04002359 */
2360 if (S_ISREG(mode)) {
Yan, Zhengc71bf092009-11-12 09:34:40 +00002361 ret = insert_orphan_item(wc->trans, root,
2362 key.objectid);
Josef Bacikb50c6e22013-04-25 15:55:30 -04002363 if (ret)
2364 break;
Chris Masone02119d2008-09-05 16:13:11 -04002365 }
Yan, Zhengc71bf092009-11-12 09:34:40 +00002366
Chris Masone02119d2008-09-05 16:13:11 -04002367 ret = link_to_fixup_dir(wc->trans, root,
2368 path, key.objectid);
Josef Bacikb50c6e22013-04-25 15:55:30 -04002369 if (ret)
2370 break;
Chris Masone02119d2008-09-05 16:13:11 -04002371 }
Josef Bacikdd8e7212013-09-11 11:57:23 -04002372
2373 if (key.type == BTRFS_DIR_INDEX_KEY &&
2374 wc->stage == LOG_WALK_REPLAY_DIR_INDEX) {
2375 ret = replay_one_dir_item(wc->trans, root, path,
2376 eb, i, &key);
2377 if (ret)
2378 break;
2379 }
2380
Chris Masone02119d2008-09-05 16:13:11 -04002381 if (wc->stage < LOG_WALK_REPLAY_ALL)
2382 continue;
2383
2384 /* these keys are simply copied */
2385 if (key.type == BTRFS_XATTR_ITEM_KEY) {
2386 ret = overwrite_item(wc->trans, root, path,
2387 eb, i, &key);
Josef Bacikb50c6e22013-04-25 15:55:30 -04002388 if (ret)
2389 break;
Liu Bo2da1c662013-05-26 13:50:29 +00002390 } else if (key.type == BTRFS_INODE_REF_KEY ||
2391 key.type == BTRFS_INODE_EXTREF_KEY) {
Mark Fashehf1863732012-08-08 11:32:27 -07002392 ret = add_inode_ref(wc->trans, root, log, path,
2393 eb, i, &key);
Josef Bacikb50c6e22013-04-25 15:55:30 -04002394 if (ret && ret != -ENOENT)
2395 break;
2396 ret = 0;
Chris Masone02119d2008-09-05 16:13:11 -04002397 } else if (key.type == BTRFS_EXTENT_DATA_KEY) {
2398 ret = replay_one_extent(wc->trans, root, path,
2399 eb, i, &key);
Josef Bacikb50c6e22013-04-25 15:55:30 -04002400 if (ret)
2401 break;
Josef Bacikdd8e7212013-09-11 11:57:23 -04002402 } else if (key.type == BTRFS_DIR_ITEM_KEY) {
Chris Masone02119d2008-09-05 16:13:11 -04002403 ret = replay_one_dir_item(wc->trans, root, path,
2404 eb, i, &key);
Josef Bacikb50c6e22013-04-25 15:55:30 -04002405 if (ret)
2406 break;
Chris Masone02119d2008-09-05 16:13:11 -04002407 }
2408 }
2409 btrfs_free_path(path);
Josef Bacikb50c6e22013-04-25 15:55:30 -04002410 return ret;
Chris Masone02119d2008-09-05 16:13:11 -04002411}
2412
Chris Masond3977122009-01-05 21:25:51 -05002413static noinline int walk_down_log_tree(struct btrfs_trans_handle *trans,
Chris Masone02119d2008-09-05 16:13:11 -04002414 struct btrfs_root *root,
2415 struct btrfs_path *path, int *level,
2416 struct walk_control *wc)
2417{
Jeff Mahoney0b246af2016-06-22 18:54:23 -04002418 struct btrfs_fs_info *fs_info = root->fs_info;
Chris Masone02119d2008-09-05 16:13:11 -04002419 u64 root_owner;
Chris Masone02119d2008-09-05 16:13:11 -04002420 u64 bytenr;
2421 u64 ptr_gen;
2422 struct extent_buffer *next;
2423 struct extent_buffer *cur;
2424 struct extent_buffer *parent;
2425 u32 blocksize;
2426 int ret = 0;
2427
2428 WARN_ON(*level < 0);
2429 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2430
Chris Masond3977122009-01-05 21:25:51 -05002431 while (*level > 0) {
Chris Masone02119d2008-09-05 16:13:11 -04002432 WARN_ON(*level < 0);
2433 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2434 cur = path->nodes[*level];
2435
Dulshani Gunawardhanafae7f212013-10-31 10:30:08 +05302436 WARN_ON(btrfs_header_level(cur) != *level);
Chris Masone02119d2008-09-05 16:13:11 -04002437
2438 if (path->slots[*level] >=
2439 btrfs_header_nritems(cur))
2440 break;
2441
2442 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
2443 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
Jeff Mahoney0b246af2016-06-22 18:54:23 -04002444 blocksize = fs_info->nodesize;
Chris Masone02119d2008-09-05 16:13:11 -04002445
2446 parent = path->nodes[*level];
2447 root_owner = btrfs_header_owner(parent);
Chris Masone02119d2008-09-05 16:13:11 -04002448
Jeff Mahoney2ff7e612016-06-22 18:54:24 -04002449 next = btrfs_find_create_tree_block(fs_info, bytenr);
Liu Boc871b0f2016-06-06 12:01:23 -07002450 if (IS_ERR(next))
2451 return PTR_ERR(next);
Chris Masone02119d2008-09-05 16:13:11 -04002452
Chris Masone02119d2008-09-05 16:13:11 -04002453 if (*level == 1) {
Mark Fasheh1e5063d2011-07-12 10:46:06 -07002454 ret = wc->process_func(root, next, wc, ptr_gen);
Josef Bacikb50c6e22013-04-25 15:55:30 -04002455 if (ret) {
2456 free_extent_buffer(next);
Mark Fasheh1e5063d2011-07-12 10:46:06 -07002457 return ret;
Josef Bacikb50c6e22013-04-25 15:55:30 -04002458 }
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002459
Chris Masone02119d2008-09-05 16:13:11 -04002460 path->slots[*level]++;
2461 if (wc->free) {
Tsutomu Itoh018642a2012-05-29 18:10:13 +09002462 ret = btrfs_read_buffer(next, ptr_gen);
2463 if (ret) {
2464 free_extent_buffer(next);
2465 return ret;
2466 }
Chris Masone02119d2008-09-05 16:13:11 -04002467
Josef Bacik681ae502013-10-07 15:11:00 -04002468 if (trans) {
2469 btrfs_tree_lock(next);
2470 btrfs_set_lock_blocking(next);
Jeff Mahoney0b246af2016-06-22 18:54:23 -04002471 clean_tree_block(trans, fs_info, next);
Josef Bacik681ae502013-10-07 15:11:00 -04002472 btrfs_wait_tree_block_writeback(next);
2473 btrfs_tree_unlock(next);
2474 }
Chris Masone02119d2008-09-05 16:13:11 -04002475
Chris Masone02119d2008-09-05 16:13:11 -04002476 WARN_ON(root_owner !=
2477 BTRFS_TREE_LOG_OBJECTID);
Jeff Mahoney2ff7e612016-06-22 18:54:24 -04002478 ret = btrfs_free_and_pin_reserved_extent(
2479 fs_info, bytenr,
2480 blocksize);
Josef Bacik36508602013-04-25 16:23:32 -04002481 if (ret) {
2482 free_extent_buffer(next);
2483 return ret;
2484 }
Chris Masone02119d2008-09-05 16:13:11 -04002485 }
2486 free_extent_buffer(next);
2487 continue;
2488 }
Tsutomu Itoh018642a2012-05-29 18:10:13 +09002489 ret = btrfs_read_buffer(next, ptr_gen);
2490 if (ret) {
2491 free_extent_buffer(next);
2492 return ret;
2493 }
Chris Masone02119d2008-09-05 16:13:11 -04002494
2495 WARN_ON(*level <= 0);
2496 if (path->nodes[*level-1])
2497 free_extent_buffer(path->nodes[*level-1]);
2498 path->nodes[*level-1] = next;
2499 *level = btrfs_header_level(next);
2500 path->slots[*level] = 0;
2501 cond_resched();
2502 }
2503 WARN_ON(*level < 0);
2504 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2505
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002506 path->slots[*level] = btrfs_header_nritems(path->nodes[*level]);
Chris Masone02119d2008-09-05 16:13:11 -04002507
2508 cond_resched();
2509 return 0;
2510}
2511
Chris Masond3977122009-01-05 21:25:51 -05002512static noinline int walk_up_log_tree(struct btrfs_trans_handle *trans,
Chris Masone02119d2008-09-05 16:13:11 -04002513 struct btrfs_root *root,
2514 struct btrfs_path *path, int *level,
2515 struct walk_control *wc)
2516{
Jeff Mahoney0b246af2016-06-22 18:54:23 -04002517 struct btrfs_fs_info *fs_info = root->fs_info;
Chris Masone02119d2008-09-05 16:13:11 -04002518 u64 root_owner;
Chris Masone02119d2008-09-05 16:13:11 -04002519 int i;
2520 int slot;
2521 int ret;
2522
Chris Masond3977122009-01-05 21:25:51 -05002523 for (i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) {
Chris Masone02119d2008-09-05 16:13:11 -04002524 slot = path->slots[i];
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002525 if (slot + 1 < btrfs_header_nritems(path->nodes[i])) {
Chris Masone02119d2008-09-05 16:13:11 -04002526 path->slots[i]++;
2527 *level = i;
2528 WARN_ON(*level == 0);
2529 return 0;
2530 } else {
Zheng Yan31840ae2008-09-23 13:14:14 -04002531 struct extent_buffer *parent;
2532 if (path->nodes[*level] == root->node)
2533 parent = path->nodes[*level];
2534 else
2535 parent = path->nodes[*level + 1];
2536
2537 root_owner = btrfs_header_owner(parent);
Mark Fasheh1e5063d2011-07-12 10:46:06 -07002538 ret = wc->process_func(root, path->nodes[*level], wc,
Chris Masone02119d2008-09-05 16:13:11 -04002539 btrfs_header_generation(path->nodes[*level]));
Mark Fasheh1e5063d2011-07-12 10:46:06 -07002540 if (ret)
2541 return ret;
2542
Chris Masone02119d2008-09-05 16:13:11 -04002543 if (wc->free) {
2544 struct extent_buffer *next;
2545
2546 next = path->nodes[*level];
2547
Josef Bacik681ae502013-10-07 15:11:00 -04002548 if (trans) {
2549 btrfs_tree_lock(next);
2550 btrfs_set_lock_blocking(next);
Jeff Mahoney0b246af2016-06-22 18:54:23 -04002551 clean_tree_block(trans, fs_info, next);
Josef Bacik681ae502013-10-07 15:11:00 -04002552 btrfs_wait_tree_block_writeback(next);
2553 btrfs_tree_unlock(next);
2554 }
Chris Masone02119d2008-09-05 16:13:11 -04002555
Chris Masone02119d2008-09-05 16:13:11 -04002556 WARN_ON(root_owner != BTRFS_TREE_LOG_OBJECTID);
Jeff Mahoney2ff7e612016-06-22 18:54:24 -04002557 ret = btrfs_free_and_pin_reserved_extent(
2558 fs_info,
Chris Masone02119d2008-09-05 16:13:11 -04002559 path->nodes[*level]->start,
Chris Masond00aff02008-09-11 15:54:42 -04002560 path->nodes[*level]->len);
Josef Bacik36508602013-04-25 16:23:32 -04002561 if (ret)
2562 return ret;
Chris Masone02119d2008-09-05 16:13:11 -04002563 }
2564 free_extent_buffer(path->nodes[*level]);
2565 path->nodes[*level] = NULL;
2566 *level = i + 1;
2567 }
2568 }
2569 return 1;
2570}
2571
2572/*
2573 * drop the reference count on the tree rooted at 'snap'. This traverses
2574 * the tree freeing any blocks that have a ref count of zero after being
2575 * decremented.
2576 */
2577static int walk_log_tree(struct btrfs_trans_handle *trans,
2578 struct btrfs_root *log, struct walk_control *wc)
2579{
Jeff Mahoney2ff7e612016-06-22 18:54:24 -04002580 struct btrfs_fs_info *fs_info = log->fs_info;
Chris Masone02119d2008-09-05 16:13:11 -04002581 int ret = 0;
2582 int wret;
2583 int level;
2584 struct btrfs_path *path;
Chris Masone02119d2008-09-05 16:13:11 -04002585 int orig_level;
2586
2587 path = btrfs_alloc_path();
Tsutomu Itohdb5b4932011-03-23 08:14:16 +00002588 if (!path)
2589 return -ENOMEM;
Chris Masone02119d2008-09-05 16:13:11 -04002590
2591 level = btrfs_header_level(log->node);
2592 orig_level = level;
2593 path->nodes[level] = log->node;
2594 extent_buffer_get(log->node);
2595 path->slots[level] = 0;
2596
Chris Masond3977122009-01-05 21:25:51 -05002597 while (1) {
Chris Masone02119d2008-09-05 16:13:11 -04002598 wret = walk_down_log_tree(trans, log, path, &level, wc);
2599 if (wret > 0)
2600 break;
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002601 if (wret < 0) {
Chris Masone02119d2008-09-05 16:13:11 -04002602 ret = wret;
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002603 goto out;
2604 }
Chris Masone02119d2008-09-05 16:13:11 -04002605
2606 wret = walk_up_log_tree(trans, log, path, &level, wc);
2607 if (wret > 0)
2608 break;
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002609 if (wret < 0) {
Chris Masone02119d2008-09-05 16:13:11 -04002610 ret = wret;
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002611 goto out;
2612 }
Chris Masone02119d2008-09-05 16:13:11 -04002613 }
2614
2615 /* was the root node processed? if not, catch it here */
2616 if (path->nodes[orig_level]) {
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002617 ret = wc->process_func(log, path->nodes[orig_level], wc,
Chris Masone02119d2008-09-05 16:13:11 -04002618 btrfs_header_generation(path->nodes[orig_level]));
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002619 if (ret)
2620 goto out;
Chris Masone02119d2008-09-05 16:13:11 -04002621 if (wc->free) {
2622 struct extent_buffer *next;
2623
2624 next = path->nodes[orig_level];
2625
Josef Bacik681ae502013-10-07 15:11:00 -04002626 if (trans) {
2627 btrfs_tree_lock(next);
2628 btrfs_set_lock_blocking(next);
Jeff Mahoney2ff7e612016-06-22 18:54:24 -04002629 clean_tree_block(trans, fs_info, next);
Josef Bacik681ae502013-10-07 15:11:00 -04002630 btrfs_wait_tree_block_writeback(next);
2631 btrfs_tree_unlock(next);
2632 }
Chris Masone02119d2008-09-05 16:13:11 -04002633
Chris Masone02119d2008-09-05 16:13:11 -04002634 WARN_ON(log->root_key.objectid !=
2635 BTRFS_TREE_LOG_OBJECTID);
Jeff Mahoney2ff7e612016-06-22 18:54:24 -04002636 ret = btrfs_free_and_pin_reserved_extent(fs_info,
2637 next->start, next->len);
Josef Bacik36508602013-04-25 16:23:32 -04002638 if (ret)
2639 goto out;
Chris Masone02119d2008-09-05 16:13:11 -04002640 }
2641 }
2642
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002643out:
Chris Masone02119d2008-09-05 16:13:11 -04002644 btrfs_free_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04002645 return ret;
2646}
2647
Yan Zheng7237f182009-01-21 12:54:03 -05002648/*
2649 * helper function to update the item for a given subvolumes log root
2650 * in the tree of log roots
2651 */
2652static int update_log_root(struct btrfs_trans_handle *trans,
2653 struct btrfs_root *log)
2654{
Jeff Mahoney0b246af2016-06-22 18:54:23 -04002655 struct btrfs_fs_info *fs_info = log->fs_info;
Yan Zheng7237f182009-01-21 12:54:03 -05002656 int ret;
2657
2658 if (log->log_transid == 1) {
2659 /* insert root item on the first sync */
Jeff Mahoney0b246af2016-06-22 18:54:23 -04002660 ret = btrfs_insert_root(trans, fs_info->log_root_tree,
Yan Zheng7237f182009-01-21 12:54:03 -05002661 &log->root_key, &log->root_item);
2662 } else {
Jeff Mahoney0b246af2016-06-22 18:54:23 -04002663 ret = btrfs_update_root(trans, fs_info->log_root_tree,
Yan Zheng7237f182009-01-21 12:54:03 -05002664 &log->root_key, &log->root_item);
2665 }
2666 return ret;
2667}
2668
Zhaolei60d53eb2015-08-17 18:44:46 +08002669static void wait_log_commit(struct btrfs_root *root, int transid)
Chris Masone02119d2008-09-05 16:13:11 -04002670{
2671 DEFINE_WAIT(wait);
Yan Zheng7237f182009-01-21 12:54:03 -05002672 int index = transid % 2;
Chris Masone02119d2008-09-05 16:13:11 -04002673
Yan Zheng7237f182009-01-21 12:54:03 -05002674 /*
2675 * we only allow two pending log transactions at a time,
2676 * so we know that if ours is more than 2 older than the
2677 * current transaction, we're done
2678 */
Chris Masone02119d2008-09-05 16:13:11 -04002679 do {
Yan Zheng7237f182009-01-21 12:54:03 -05002680 prepare_to_wait(&root->log_commit_wait[index],
2681 &wait, TASK_UNINTERRUPTIBLE);
2682 mutex_unlock(&root->log_mutex);
Chris Mason12fcfd22009-03-24 10:24:20 -04002683
Miao Xied1433de2014-02-20 18:08:59 +08002684 if (root->log_transid_committed < transid &&
Yan Zheng7237f182009-01-21 12:54:03 -05002685 atomic_read(&root->log_commit[index]))
Chris Masone02119d2008-09-05 16:13:11 -04002686 schedule();
Chris Mason12fcfd22009-03-24 10:24:20 -04002687
Yan Zheng7237f182009-01-21 12:54:03 -05002688 finish_wait(&root->log_commit_wait[index], &wait);
2689 mutex_lock(&root->log_mutex);
Miao Xied1433de2014-02-20 18:08:59 +08002690 } while (root->log_transid_committed < transid &&
Yan Zheng7237f182009-01-21 12:54:03 -05002691 atomic_read(&root->log_commit[index]));
Yan Zheng7237f182009-01-21 12:54:03 -05002692}
2693
Zhaolei60d53eb2015-08-17 18:44:46 +08002694static void wait_for_writer(struct btrfs_root *root)
Yan Zheng7237f182009-01-21 12:54:03 -05002695{
2696 DEFINE_WAIT(wait);
Miao Xie8b050d32014-02-20 18:08:58 +08002697
2698 while (atomic_read(&root->log_writers)) {
Yan Zheng7237f182009-01-21 12:54:03 -05002699 prepare_to_wait(&root->log_writer_wait,
2700 &wait, TASK_UNINTERRUPTIBLE);
2701 mutex_unlock(&root->log_mutex);
Miao Xie8b050d32014-02-20 18:08:58 +08002702 if (atomic_read(&root->log_writers))
Yan Zheng7237f182009-01-21 12:54:03 -05002703 schedule();
Yan Zheng7237f182009-01-21 12:54:03 -05002704 finish_wait(&root->log_writer_wait, &wait);
Filipe Manana575849e2015-02-11 11:12:39 +00002705 mutex_lock(&root->log_mutex);
Yan Zheng7237f182009-01-21 12:54:03 -05002706 }
Chris Masone02119d2008-09-05 16:13:11 -04002707}
2708
Miao Xie8b050d32014-02-20 18:08:58 +08002709static inline void btrfs_remove_log_ctx(struct btrfs_root *root,
2710 struct btrfs_log_ctx *ctx)
2711{
2712 if (!ctx)
2713 return;
2714
2715 mutex_lock(&root->log_mutex);
2716 list_del_init(&ctx->list);
2717 mutex_unlock(&root->log_mutex);
2718}
2719
2720/*
2721 * Invoked in log mutex context, or be sure there is no other task which
2722 * can access the list.
2723 */
2724static inline void btrfs_remove_all_log_ctxs(struct btrfs_root *root,
2725 int index, int error)
2726{
2727 struct btrfs_log_ctx *ctx;
Chris Mason570dd452016-10-27 10:42:20 -07002728 struct btrfs_log_ctx *safe;
Miao Xie8b050d32014-02-20 18:08:58 +08002729
Chris Mason570dd452016-10-27 10:42:20 -07002730 list_for_each_entry_safe(ctx, safe, &root->log_ctxs[index], list) {
2731 list_del_init(&ctx->list);
Miao Xie8b050d32014-02-20 18:08:58 +08002732 ctx->log_ret = error;
Chris Mason570dd452016-10-27 10:42:20 -07002733 }
Miao Xie8b050d32014-02-20 18:08:58 +08002734
2735 INIT_LIST_HEAD(&root->log_ctxs[index]);
2736}
2737
Chris Masone02119d2008-09-05 16:13:11 -04002738/*
2739 * btrfs_sync_log does sends a given tree log down to the disk and
2740 * updates the super blocks to record it. When this call is done,
Chris Mason12fcfd22009-03-24 10:24:20 -04002741 * you know that any inodes previously logged are safely on disk only
2742 * if it returns 0.
2743 *
2744 * Any other return value means you need to call btrfs_commit_transaction.
2745 * Some of the edge cases for fsyncing directories that have had unlinks
2746 * or renames done in the past mean that sometimes the only safe
2747 * fsync is to commit the whole FS. When btrfs_sync_log returns -EAGAIN,
2748 * that has happened.
Chris Masone02119d2008-09-05 16:13:11 -04002749 */
2750int btrfs_sync_log(struct btrfs_trans_handle *trans,
Miao Xie8b050d32014-02-20 18:08:58 +08002751 struct btrfs_root *root, struct btrfs_log_ctx *ctx)
Chris Masone02119d2008-09-05 16:13:11 -04002752{
Yan Zheng7237f182009-01-21 12:54:03 -05002753 int index1;
2754 int index2;
Yan, Zheng8cef4e12009-11-12 09:33:26 +00002755 int mark;
Chris Masone02119d2008-09-05 16:13:11 -04002756 int ret;
Jeff Mahoney0b246af2016-06-22 18:54:23 -04002757 struct btrfs_fs_info *fs_info = root->fs_info;
Chris Masone02119d2008-09-05 16:13:11 -04002758 struct btrfs_root *log = root->log_root;
Jeff Mahoney0b246af2016-06-22 18:54:23 -04002759 struct btrfs_root *log_root_tree = fs_info->log_root_tree;
Miao Xiebb14a592014-02-20 18:08:56 +08002760 int log_transid = 0;
Miao Xie8b050d32014-02-20 18:08:58 +08002761 struct btrfs_log_ctx root_log_ctx;
Miao Xiec6adc9c2013-05-28 10:05:39 +00002762 struct blk_plug plug;
Chris Masone02119d2008-09-05 16:13:11 -04002763
Yan Zheng7237f182009-01-21 12:54:03 -05002764 mutex_lock(&root->log_mutex);
Miao Xied1433de2014-02-20 18:08:59 +08002765 log_transid = ctx->log_transid;
2766 if (root->log_transid_committed >= log_transid) {
Yan Zheng7237f182009-01-21 12:54:03 -05002767 mutex_unlock(&root->log_mutex);
Miao Xie8b050d32014-02-20 18:08:58 +08002768 return ctx->log_ret;
Chris Masone02119d2008-09-05 16:13:11 -04002769 }
Miao Xied1433de2014-02-20 18:08:59 +08002770
2771 index1 = log_transid % 2;
2772 if (atomic_read(&root->log_commit[index1])) {
Zhaolei60d53eb2015-08-17 18:44:46 +08002773 wait_log_commit(root, log_transid);
Miao Xied1433de2014-02-20 18:08:59 +08002774 mutex_unlock(&root->log_mutex);
2775 return ctx->log_ret;
2776 }
2777 ASSERT(log_transid == root->log_transid);
Yan Zheng7237f182009-01-21 12:54:03 -05002778 atomic_set(&root->log_commit[index1], 1);
2779
2780 /* wait for previous tree log sync to complete */
2781 if (atomic_read(&root->log_commit[(index1 + 1) % 2]))
Zhaolei60d53eb2015-08-17 18:44:46 +08002782 wait_log_commit(root, log_transid - 1);
Miao Xie48cab2e2014-02-20 18:08:52 +08002783
Yan, Zheng86df7eb2009-10-14 09:24:59 -04002784 while (1) {
Miao Xie2ecb7922012-09-06 04:04:27 -06002785 int batch = atomic_read(&root->log_batch);
Chris Masoncd354ad2011-10-20 15:45:37 -04002786 /* when we're on an ssd, just kick the log commit out */
Jeff Mahoney0b246af2016-06-22 18:54:23 -04002787 if (!btrfs_test_opt(fs_info, SSD) &&
Miao Xie27cdeb72014-04-02 19:51:05 +08002788 test_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state)) {
Yan, Zheng86df7eb2009-10-14 09:24:59 -04002789 mutex_unlock(&root->log_mutex);
2790 schedule_timeout_uninterruptible(1);
2791 mutex_lock(&root->log_mutex);
2792 }
Zhaolei60d53eb2015-08-17 18:44:46 +08002793 wait_for_writer(root);
Miao Xie2ecb7922012-09-06 04:04:27 -06002794 if (batch == atomic_read(&root->log_batch))
Chris Masone02119d2008-09-05 16:13:11 -04002795 break;
2796 }
Chris Masond0c803c2008-09-11 16:17:57 -04002797
Chris Mason12fcfd22009-03-24 10:24:20 -04002798 /* bail out if we need to do a full commit */
Jeff Mahoney0b246af2016-06-22 18:54:23 -04002799 if (btrfs_need_log_full_commit(fs_info, trans)) {
Chris Mason12fcfd22009-03-24 10:24:20 -04002800 ret = -EAGAIN;
Josef Bacik2ab28f32012-10-12 15:27:49 -04002801 btrfs_free_logged_extents(log, log_transid);
Chris Mason12fcfd22009-03-24 10:24:20 -04002802 mutex_unlock(&root->log_mutex);
2803 goto out;
2804 }
2805
Yan, Zheng8cef4e12009-11-12 09:33:26 +00002806 if (log_transid % 2 == 0)
2807 mark = EXTENT_DIRTY;
2808 else
2809 mark = EXTENT_NEW;
2810
Chris Mason690587d2009-10-13 13:29:19 -04002811 /* we start IO on all the marked extents here, but we don't actually
2812 * wait for them until later.
2813 */
Miao Xiec6adc9c2013-05-28 10:05:39 +00002814 blk_start_plug(&plug);
Jeff Mahoney2ff7e612016-06-22 18:54:24 -04002815 ret = btrfs_write_marked_extents(fs_info, &log->dirty_log_pages, mark);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002816 if (ret) {
Miao Xiec6adc9c2013-05-28 10:05:39 +00002817 blk_finish_plug(&plug);
Jeff Mahoney66642832016-06-10 18:19:25 -04002818 btrfs_abort_transaction(trans, ret);
Josef Bacik2ab28f32012-10-12 15:27:49 -04002819 btrfs_free_logged_extents(log, log_transid);
Jeff Mahoney0b246af2016-06-22 18:54:23 -04002820 btrfs_set_log_full_commit(fs_info, trans);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002821 mutex_unlock(&root->log_mutex);
2822 goto out;
2823 }
Yan Zheng7237f182009-01-21 12:54:03 -05002824
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002825 btrfs_set_root_node(&log->root_item, log->node);
Yan Zheng7237f182009-01-21 12:54:03 -05002826
Yan Zheng7237f182009-01-21 12:54:03 -05002827 root->log_transid++;
2828 log->log_transid = root->log_transid;
Josef Bacikff782e02009-10-08 15:30:04 -04002829 root->log_start_pid = 0;
Yan Zheng7237f182009-01-21 12:54:03 -05002830 /*
Yan, Zheng8cef4e12009-11-12 09:33:26 +00002831 * IO has been started, blocks of the log tree have WRITTEN flag set
2832 * in their headers. new modifications of the log will be written to
2833 * new positions. so it's safe to allow log writers to go in.
Yan Zheng7237f182009-01-21 12:54:03 -05002834 */
2835 mutex_unlock(&root->log_mutex);
2836
Filipe Manana28a23592016-08-23 21:13:51 +01002837 btrfs_init_log_ctx(&root_log_ctx, NULL);
Miao Xied1433de2014-02-20 18:08:59 +08002838
Yan Zheng7237f182009-01-21 12:54:03 -05002839 mutex_lock(&log_root_tree->log_mutex);
Miao Xie2ecb7922012-09-06 04:04:27 -06002840 atomic_inc(&log_root_tree->log_batch);
Yan Zheng7237f182009-01-21 12:54:03 -05002841 atomic_inc(&log_root_tree->log_writers);
Miao Xied1433de2014-02-20 18:08:59 +08002842
2843 index2 = log_root_tree->log_transid % 2;
2844 list_add_tail(&root_log_ctx.list, &log_root_tree->log_ctxs[index2]);
2845 root_log_ctx.log_transid = log_root_tree->log_transid;
2846
Yan Zheng7237f182009-01-21 12:54:03 -05002847 mutex_unlock(&log_root_tree->log_mutex);
2848
2849 ret = update_log_root(trans, log);
Yan Zheng7237f182009-01-21 12:54:03 -05002850
2851 mutex_lock(&log_root_tree->log_mutex);
2852 if (atomic_dec_and_test(&log_root_tree->log_writers)) {
David Sterba779adf02015-02-16 19:39:00 +01002853 /*
2854 * Implicit memory barrier after atomic_dec_and_test
2855 */
Yan Zheng7237f182009-01-21 12:54:03 -05002856 if (waitqueue_active(&log_root_tree->log_writer_wait))
2857 wake_up(&log_root_tree->log_writer_wait);
2858 }
2859
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002860 if (ret) {
Miao Xied1433de2014-02-20 18:08:59 +08002861 if (!list_empty(&root_log_ctx.list))
2862 list_del_init(&root_log_ctx.list);
2863
Miao Xiec6adc9c2013-05-28 10:05:39 +00002864 blk_finish_plug(&plug);
Jeff Mahoney0b246af2016-06-22 18:54:23 -04002865 btrfs_set_log_full_commit(fs_info, trans);
Miao Xie995946d2014-04-02 19:51:06 +08002866
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002867 if (ret != -ENOSPC) {
Jeff Mahoney66642832016-06-10 18:19:25 -04002868 btrfs_abort_transaction(trans, ret);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002869 mutex_unlock(&log_root_tree->log_mutex);
2870 goto out;
2871 }
Jeff Mahoneybf89d382016-09-09 20:42:44 -04002872 btrfs_wait_tree_log_extents(log, mark);
Josef Bacik2ab28f32012-10-12 15:27:49 -04002873 btrfs_free_logged_extents(log, log_transid);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002874 mutex_unlock(&log_root_tree->log_mutex);
2875 ret = -EAGAIN;
2876 goto out;
2877 }
2878
Miao Xied1433de2014-02-20 18:08:59 +08002879 if (log_root_tree->log_transid_committed >= root_log_ctx.log_transid) {
Forrest Liu3da5ab52015-01-30 19:42:12 +08002880 blk_finish_plug(&plug);
Chris Masoncbd60aa2016-09-06 05:37:40 -07002881 list_del_init(&root_log_ctx.list);
Miao Xied1433de2014-02-20 18:08:59 +08002882 mutex_unlock(&log_root_tree->log_mutex);
2883 ret = root_log_ctx.log_ret;
2884 goto out;
2885 }
Miao Xie8b050d32014-02-20 18:08:58 +08002886
Miao Xied1433de2014-02-20 18:08:59 +08002887 index2 = root_log_ctx.log_transid % 2;
Yan Zheng7237f182009-01-21 12:54:03 -05002888 if (atomic_read(&log_root_tree->log_commit[index2])) {
Miao Xiec6adc9c2013-05-28 10:05:39 +00002889 blk_finish_plug(&plug);
Jeff Mahoneybf89d382016-09-09 20:42:44 -04002890 ret = btrfs_wait_tree_log_extents(log, mark);
Josef Bacik50d9aa92014-11-21 14:52:38 -05002891 btrfs_wait_logged_extents(trans, log, log_transid);
Zhaolei60d53eb2015-08-17 18:44:46 +08002892 wait_log_commit(log_root_tree,
Miao Xied1433de2014-02-20 18:08:59 +08002893 root_log_ctx.log_transid);
Yan Zheng7237f182009-01-21 12:54:03 -05002894 mutex_unlock(&log_root_tree->log_mutex);
Filipe Manana5ab5e442014-11-13 16:59:53 +00002895 if (!ret)
2896 ret = root_log_ctx.log_ret;
Yan Zheng7237f182009-01-21 12:54:03 -05002897 goto out;
2898 }
Miao Xied1433de2014-02-20 18:08:59 +08002899 ASSERT(root_log_ctx.log_transid == log_root_tree->log_transid);
Yan Zheng7237f182009-01-21 12:54:03 -05002900 atomic_set(&log_root_tree->log_commit[index2], 1);
2901
Chris Mason12fcfd22009-03-24 10:24:20 -04002902 if (atomic_read(&log_root_tree->log_commit[(index2 + 1) % 2])) {
Zhaolei60d53eb2015-08-17 18:44:46 +08002903 wait_log_commit(log_root_tree,
Miao Xied1433de2014-02-20 18:08:59 +08002904 root_log_ctx.log_transid - 1);
Chris Mason12fcfd22009-03-24 10:24:20 -04002905 }
Yan Zheng7237f182009-01-21 12:54:03 -05002906
Zhaolei60d53eb2015-08-17 18:44:46 +08002907 wait_for_writer(log_root_tree);
Chris Mason12fcfd22009-03-24 10:24:20 -04002908
2909 /*
2910 * now that we've moved on to the tree of log tree roots,
2911 * check the full commit flag again
2912 */
Jeff Mahoney0b246af2016-06-22 18:54:23 -04002913 if (btrfs_need_log_full_commit(fs_info, trans)) {
Miao Xiec6adc9c2013-05-28 10:05:39 +00002914 blk_finish_plug(&plug);
Jeff Mahoneybf89d382016-09-09 20:42:44 -04002915 btrfs_wait_tree_log_extents(log, mark);
Josef Bacik2ab28f32012-10-12 15:27:49 -04002916 btrfs_free_logged_extents(log, log_transid);
Chris Mason12fcfd22009-03-24 10:24:20 -04002917 mutex_unlock(&log_root_tree->log_mutex);
2918 ret = -EAGAIN;
2919 goto out_wake_log_root;
2920 }
Yan Zheng7237f182009-01-21 12:54:03 -05002921
Jeff Mahoney2ff7e612016-06-22 18:54:24 -04002922 ret = btrfs_write_marked_extents(fs_info,
Miao Xiec6adc9c2013-05-28 10:05:39 +00002923 &log_root_tree->dirty_log_pages,
2924 EXTENT_DIRTY | EXTENT_NEW);
2925 blk_finish_plug(&plug);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002926 if (ret) {
Jeff Mahoney0b246af2016-06-22 18:54:23 -04002927 btrfs_set_log_full_commit(fs_info, trans);
Jeff Mahoney66642832016-06-10 18:19:25 -04002928 btrfs_abort_transaction(trans, ret);
Josef Bacik2ab28f32012-10-12 15:27:49 -04002929 btrfs_free_logged_extents(log, log_transid);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002930 mutex_unlock(&log_root_tree->log_mutex);
2931 goto out_wake_log_root;
2932 }
Jeff Mahoneybf89d382016-09-09 20:42:44 -04002933 ret = btrfs_wait_tree_log_extents(log, mark);
Filipe Manana5ab5e442014-11-13 16:59:53 +00002934 if (!ret)
Jeff Mahoneybf89d382016-09-09 20:42:44 -04002935 ret = btrfs_wait_tree_log_extents(log_root_tree,
2936 EXTENT_NEW | EXTENT_DIRTY);
Filipe Manana5ab5e442014-11-13 16:59:53 +00002937 if (ret) {
Jeff Mahoney0b246af2016-06-22 18:54:23 -04002938 btrfs_set_log_full_commit(fs_info, trans);
Filipe Manana5ab5e442014-11-13 16:59:53 +00002939 btrfs_free_logged_extents(log, log_transid);
2940 mutex_unlock(&log_root_tree->log_mutex);
2941 goto out_wake_log_root;
2942 }
Josef Bacik50d9aa92014-11-21 14:52:38 -05002943 btrfs_wait_logged_extents(trans, log, log_transid);
Chris Masone02119d2008-09-05 16:13:11 -04002944
Jeff Mahoney0b246af2016-06-22 18:54:23 -04002945 btrfs_set_super_log_root(fs_info->super_for_commit,
2946 log_root_tree->node->start);
2947 btrfs_set_super_log_root_level(fs_info->super_for_commit,
2948 btrfs_header_level(log_root_tree->node));
Chris Masone02119d2008-09-05 16:13:11 -04002949
Yan Zheng7237f182009-01-21 12:54:03 -05002950 log_root_tree->log_transid++;
Yan Zheng7237f182009-01-21 12:54:03 -05002951 mutex_unlock(&log_root_tree->log_mutex);
2952
2953 /*
2954 * nobody else is going to jump in and write the the ctree
2955 * super here because the log_commit atomic below is protecting
2956 * us. We must be called with a transaction handle pinning
2957 * the running transaction open, so a full commit can't hop
2958 * in and cause problems either.
2959 */
Jeff Mahoney2ff7e612016-06-22 18:54:24 -04002960 ret = write_ctree_super(trans, fs_info, 1);
Stefan Behrens5af3e8c2012-08-01 18:56:49 +02002961 if (ret) {
Jeff Mahoney0b246af2016-06-22 18:54:23 -04002962 btrfs_set_log_full_commit(fs_info, trans);
Jeff Mahoney66642832016-06-10 18:19:25 -04002963 btrfs_abort_transaction(trans, ret);
Stefan Behrens5af3e8c2012-08-01 18:56:49 +02002964 goto out_wake_log_root;
2965 }
Yan Zheng7237f182009-01-21 12:54:03 -05002966
Chris Mason257c62e2009-10-13 13:21:08 -04002967 mutex_lock(&root->log_mutex);
2968 if (root->last_log_commit < log_transid)
2969 root->last_log_commit = log_transid;
2970 mutex_unlock(&root->log_mutex);
2971
Chris Mason12fcfd22009-03-24 10:24:20 -04002972out_wake_log_root:
Chris Mason570dd452016-10-27 10:42:20 -07002973 mutex_lock(&log_root_tree->log_mutex);
Miao Xie8b050d32014-02-20 18:08:58 +08002974 btrfs_remove_all_log_ctxs(log_root_tree, index2, ret);
2975
Miao Xied1433de2014-02-20 18:08:59 +08002976 log_root_tree->log_transid_committed++;
Yan Zheng7237f182009-01-21 12:54:03 -05002977 atomic_set(&log_root_tree->log_commit[index2], 0);
Miao Xied1433de2014-02-20 18:08:59 +08002978 mutex_unlock(&log_root_tree->log_mutex);
2979
David Sterba33a9eca2015-10-10 18:35:10 +02002980 /*
2981 * The barrier before waitqueue_active is implied by mutex_unlock
2982 */
Yan Zheng7237f182009-01-21 12:54:03 -05002983 if (waitqueue_active(&log_root_tree->log_commit_wait[index2]))
2984 wake_up(&log_root_tree->log_commit_wait[index2]);
Chris Masone02119d2008-09-05 16:13:11 -04002985out:
Miao Xied1433de2014-02-20 18:08:59 +08002986 mutex_lock(&root->log_mutex);
Chris Mason570dd452016-10-27 10:42:20 -07002987 btrfs_remove_all_log_ctxs(root, index1, ret);
Miao Xied1433de2014-02-20 18:08:59 +08002988 root->log_transid_committed++;
Yan Zheng7237f182009-01-21 12:54:03 -05002989 atomic_set(&root->log_commit[index1], 0);
Miao Xied1433de2014-02-20 18:08:59 +08002990 mutex_unlock(&root->log_mutex);
Miao Xie8b050d32014-02-20 18:08:58 +08002991
David Sterba33a9eca2015-10-10 18:35:10 +02002992 /*
2993 * The barrier before waitqueue_active is implied by mutex_unlock
2994 */
Yan Zheng7237f182009-01-21 12:54:03 -05002995 if (waitqueue_active(&root->log_commit_wait[index1]))
2996 wake_up(&root->log_commit_wait[index1]);
Chris Masonb31eabd2011-01-31 16:48:24 -05002997 return ret;
Chris Masone02119d2008-09-05 16:13:11 -04002998}
2999
Yan, Zheng4a500fd2010-05-16 10:49:59 -04003000static void free_log_tree(struct btrfs_trans_handle *trans,
3001 struct btrfs_root *log)
Chris Masone02119d2008-09-05 16:13:11 -04003002{
3003 int ret;
Chris Masond0c803c2008-09-11 16:17:57 -04003004 u64 start;
3005 u64 end;
Chris Masone02119d2008-09-05 16:13:11 -04003006 struct walk_control wc = {
3007 .free = 1,
3008 .process_func = process_one_buffer
3009 };
3010
Josef Bacik681ae502013-10-07 15:11:00 -04003011 ret = walk_log_tree(trans, log, &wc);
3012 /* I don't think this can happen but just in case */
3013 if (ret)
Jeff Mahoney66642832016-06-10 18:19:25 -04003014 btrfs_abort_transaction(trans, ret);
Chris Masone02119d2008-09-05 16:13:11 -04003015
Chris Masond3977122009-01-05 21:25:51 -05003016 while (1) {
Chris Masond0c803c2008-09-11 16:17:57 -04003017 ret = find_first_extent_bit(&log->dirty_log_pages,
Josef Bacike6138872012-09-27 17:07:30 -04003018 0, &start, &end, EXTENT_DIRTY | EXTENT_NEW,
3019 NULL);
Chris Masond0c803c2008-09-11 16:17:57 -04003020 if (ret)
3021 break;
3022
Yan, Zheng8cef4e12009-11-12 09:33:26 +00003023 clear_extent_bits(&log->dirty_log_pages, start, end,
David Sterba91166212016-04-26 23:54:39 +02003024 EXTENT_DIRTY | EXTENT_NEW);
Chris Masond0c803c2008-09-11 16:17:57 -04003025 }
3026
Josef Bacik2ab28f32012-10-12 15:27:49 -04003027 /*
3028 * We may have short-circuited the log tree with the full commit logic
3029 * and left ordered extents on our list, so clear these out to keep us
3030 * from leaking inodes and memory.
3031 */
3032 btrfs_free_logged_extents(log, 0);
3033 btrfs_free_logged_extents(log, 1);
3034
Yan Zheng7237f182009-01-21 12:54:03 -05003035 free_extent_buffer(log->node);
3036 kfree(log);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04003037}
3038
3039/*
3040 * free all the extents used by the tree log. This should be called
3041 * at commit time of the full transaction
3042 */
3043int btrfs_free_log(struct btrfs_trans_handle *trans, struct btrfs_root *root)
3044{
3045 if (root->log_root) {
3046 free_log_tree(trans, root->log_root);
3047 root->log_root = NULL;
3048 }
3049 return 0;
3050}
3051
3052int btrfs_free_log_root_tree(struct btrfs_trans_handle *trans,
3053 struct btrfs_fs_info *fs_info)
3054{
3055 if (fs_info->log_root_tree) {
3056 free_log_tree(trans, fs_info->log_root_tree);
3057 fs_info->log_root_tree = NULL;
3058 }
Chris Masone02119d2008-09-05 16:13:11 -04003059 return 0;
3060}
3061
3062/*
Chris Masone02119d2008-09-05 16:13:11 -04003063 * If both a file and directory are logged, and unlinks or renames are
3064 * mixed in, we have a few interesting corners:
3065 *
3066 * create file X in dir Y
3067 * link file X to X.link in dir Y
3068 * fsync file X
3069 * unlink file X but leave X.link
3070 * fsync dir Y
3071 *
3072 * After a crash we would expect only X.link to exist. But file X
3073 * didn't get fsync'd again so the log has back refs for X and X.link.
3074 *
3075 * We solve this by removing directory entries and inode backrefs from the
3076 * log when a file that was logged in the current transaction is
3077 * unlinked. Any later fsync will include the updated log entries, and
3078 * we'll be able to reconstruct the proper directory items from backrefs.
3079 *
3080 * This optimizations allows us to avoid relogging the entire inode
3081 * or the entire directory.
3082 */
3083int btrfs_del_dir_entries_in_log(struct btrfs_trans_handle *trans,
3084 struct btrfs_root *root,
3085 const char *name, int name_len,
3086 struct inode *dir, u64 index)
3087{
3088 struct btrfs_root *log;
3089 struct btrfs_dir_item *di;
3090 struct btrfs_path *path;
3091 int ret;
Yan, Zheng4a500fd2010-05-16 10:49:59 -04003092 int err = 0;
Chris Masone02119d2008-09-05 16:13:11 -04003093 int bytes_del = 0;
Li Zefan33345d012011-04-20 10:31:50 +08003094 u64 dir_ino = btrfs_ino(dir);
Chris Masone02119d2008-09-05 16:13:11 -04003095
Chris Mason3a5f1d42008-09-11 15:53:37 -04003096 if (BTRFS_I(dir)->logged_trans < trans->transid)
3097 return 0;
3098
Chris Masone02119d2008-09-05 16:13:11 -04003099 ret = join_running_log_trans(root);
3100 if (ret)
3101 return 0;
3102
3103 mutex_lock(&BTRFS_I(dir)->log_mutex);
3104
3105 log = root->log_root;
3106 path = btrfs_alloc_path();
Tsutomu Itoha62f44a2011-04-25 19:43:51 -04003107 if (!path) {
3108 err = -ENOMEM;
3109 goto out_unlock;
3110 }
liubo2a29edc2011-01-26 06:22:08 +00003111
Li Zefan33345d012011-04-20 10:31:50 +08003112 di = btrfs_lookup_dir_item(trans, log, path, dir_ino,
Chris Masone02119d2008-09-05 16:13:11 -04003113 name, name_len, -1);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04003114 if (IS_ERR(di)) {
3115 err = PTR_ERR(di);
3116 goto fail;
3117 }
3118 if (di) {
Chris Masone02119d2008-09-05 16:13:11 -04003119 ret = btrfs_delete_one_dir_name(trans, log, path, di);
3120 bytes_del += name_len;
Josef Bacik36508602013-04-25 16:23:32 -04003121 if (ret) {
3122 err = ret;
3123 goto fail;
3124 }
Chris Masone02119d2008-09-05 16:13:11 -04003125 }
David Sterbab3b4aa72011-04-21 01:20:15 +02003126 btrfs_release_path(path);
Li Zefan33345d012011-04-20 10:31:50 +08003127 di = btrfs_lookup_dir_index_item(trans, log, path, dir_ino,
Chris Masone02119d2008-09-05 16:13:11 -04003128 index, name, name_len, -1);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04003129 if (IS_ERR(di)) {
3130 err = PTR_ERR(di);
3131 goto fail;
3132 }
3133 if (di) {
Chris Masone02119d2008-09-05 16:13:11 -04003134 ret = btrfs_delete_one_dir_name(trans, log, path, di);
3135 bytes_del += name_len;
Josef Bacik36508602013-04-25 16:23:32 -04003136 if (ret) {
3137 err = ret;
3138 goto fail;
3139 }
Chris Masone02119d2008-09-05 16:13:11 -04003140 }
3141
3142 /* update the directory size in the log to reflect the names
3143 * we have removed
3144 */
3145 if (bytes_del) {
3146 struct btrfs_key key;
3147
Li Zefan33345d012011-04-20 10:31:50 +08003148 key.objectid = dir_ino;
Chris Masone02119d2008-09-05 16:13:11 -04003149 key.offset = 0;
3150 key.type = BTRFS_INODE_ITEM_KEY;
David Sterbab3b4aa72011-04-21 01:20:15 +02003151 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04003152
3153 ret = btrfs_search_slot(trans, log, &key, path, 0, 1);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04003154 if (ret < 0) {
3155 err = ret;
3156 goto fail;
3157 }
Chris Masone02119d2008-09-05 16:13:11 -04003158 if (ret == 0) {
3159 struct btrfs_inode_item *item;
3160 u64 i_size;
3161
3162 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3163 struct btrfs_inode_item);
3164 i_size = btrfs_inode_size(path->nodes[0], item);
3165 if (i_size > bytes_del)
3166 i_size -= bytes_del;
3167 else
3168 i_size = 0;
3169 btrfs_set_inode_size(path->nodes[0], item, i_size);
3170 btrfs_mark_buffer_dirty(path->nodes[0]);
3171 } else
3172 ret = 0;
David Sterbab3b4aa72011-04-21 01:20:15 +02003173 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04003174 }
Yan, Zheng4a500fd2010-05-16 10:49:59 -04003175fail:
Chris Masone02119d2008-09-05 16:13:11 -04003176 btrfs_free_path(path);
Tsutomu Itoha62f44a2011-04-25 19:43:51 -04003177out_unlock:
Chris Masone02119d2008-09-05 16:13:11 -04003178 mutex_unlock(&BTRFS_I(dir)->log_mutex);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04003179 if (ret == -ENOSPC) {
Miao Xie995946d2014-04-02 19:51:06 +08003180 btrfs_set_log_full_commit(root->fs_info, trans);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04003181 ret = 0;
Jeff Mahoney79787ea2012-03-12 16:03:00 +01003182 } else if (ret < 0)
Jeff Mahoney66642832016-06-10 18:19:25 -04003183 btrfs_abort_transaction(trans, ret);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01003184
Chris Mason12fcfd22009-03-24 10:24:20 -04003185 btrfs_end_log_trans(root);
Chris Masone02119d2008-09-05 16:13:11 -04003186
Andi Kleen411fc6b2010-10-29 15:14:31 -04003187 return err;
Chris Masone02119d2008-09-05 16:13:11 -04003188}
3189
3190/* see comments for btrfs_del_dir_entries_in_log */
3191int btrfs_del_inode_ref_in_log(struct btrfs_trans_handle *trans,
3192 struct btrfs_root *root,
3193 const char *name, int name_len,
3194 struct inode *inode, u64 dirid)
3195{
Jeff Mahoney0b246af2016-06-22 18:54:23 -04003196 struct btrfs_fs_info *fs_info = root->fs_info;
Chris Masone02119d2008-09-05 16:13:11 -04003197 struct btrfs_root *log;
3198 u64 index;
3199 int ret;
3200
Chris Mason3a5f1d42008-09-11 15:53:37 -04003201 if (BTRFS_I(inode)->logged_trans < trans->transid)
3202 return 0;
3203
Chris Masone02119d2008-09-05 16:13:11 -04003204 ret = join_running_log_trans(root);
3205 if (ret)
3206 return 0;
3207 log = root->log_root;
3208 mutex_lock(&BTRFS_I(inode)->log_mutex);
3209
Li Zefan33345d012011-04-20 10:31:50 +08003210 ret = btrfs_del_inode_ref(trans, log, name, name_len, btrfs_ino(inode),
Chris Masone02119d2008-09-05 16:13:11 -04003211 dirid, &index);
3212 mutex_unlock(&BTRFS_I(inode)->log_mutex);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04003213 if (ret == -ENOSPC) {
Jeff Mahoney0b246af2016-06-22 18:54:23 -04003214 btrfs_set_log_full_commit(fs_info, trans);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04003215 ret = 0;
Jeff Mahoney79787ea2012-03-12 16:03:00 +01003216 } else if (ret < 0 && ret != -ENOENT)
Jeff Mahoney66642832016-06-10 18:19:25 -04003217 btrfs_abort_transaction(trans, ret);
Chris Mason12fcfd22009-03-24 10:24:20 -04003218 btrfs_end_log_trans(root);
Chris Masone02119d2008-09-05 16:13:11 -04003219
Chris Masone02119d2008-09-05 16:13:11 -04003220 return ret;
3221}
3222
3223/*
3224 * creates a range item in the log for 'dirid'. first_offset and
3225 * last_offset tell us which parts of the key space the log should
3226 * be considered authoritative for.
3227 */
3228static noinline int insert_dir_log_key(struct btrfs_trans_handle *trans,
3229 struct btrfs_root *log,
3230 struct btrfs_path *path,
3231 int key_type, u64 dirid,
3232 u64 first_offset, u64 last_offset)
3233{
3234 int ret;
3235 struct btrfs_key key;
3236 struct btrfs_dir_log_item *item;
3237
3238 key.objectid = dirid;
3239 key.offset = first_offset;
3240 if (key_type == BTRFS_DIR_ITEM_KEY)
3241 key.type = BTRFS_DIR_LOG_ITEM_KEY;
3242 else
3243 key.type = BTRFS_DIR_LOG_INDEX_KEY;
3244 ret = btrfs_insert_empty_item(trans, log, path, &key, sizeof(*item));
Yan, Zheng4a500fd2010-05-16 10:49:59 -04003245 if (ret)
3246 return ret;
Chris Masone02119d2008-09-05 16:13:11 -04003247
3248 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3249 struct btrfs_dir_log_item);
3250 btrfs_set_dir_log_end(path->nodes[0], item, last_offset);
3251 btrfs_mark_buffer_dirty(path->nodes[0]);
David Sterbab3b4aa72011-04-21 01:20:15 +02003252 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04003253 return 0;
3254}
3255
3256/*
3257 * log all the items included in the current transaction for a given
3258 * directory. This also creates the range items in the log tree required
3259 * to replay anything deleted before the fsync
3260 */
3261static noinline int log_dir_items(struct btrfs_trans_handle *trans,
3262 struct btrfs_root *root, struct inode *inode,
3263 struct btrfs_path *path,
3264 struct btrfs_path *dst_path, int key_type,
Filipe Manana2f2ff0e2015-03-20 17:19:46 +00003265 struct btrfs_log_ctx *ctx,
Chris Masone02119d2008-09-05 16:13:11 -04003266 u64 min_offset, u64 *last_offset_ret)
3267{
3268 struct btrfs_key min_key;
Chris Masone02119d2008-09-05 16:13:11 -04003269 struct btrfs_root *log = root->log_root;
3270 struct extent_buffer *src;
Yan, Zheng4a500fd2010-05-16 10:49:59 -04003271 int err = 0;
Chris Masone02119d2008-09-05 16:13:11 -04003272 int ret;
3273 int i;
3274 int nritems;
3275 u64 first_offset = min_offset;
3276 u64 last_offset = (u64)-1;
Li Zefan33345d012011-04-20 10:31:50 +08003277 u64 ino = btrfs_ino(inode);
Chris Masone02119d2008-09-05 16:13:11 -04003278
3279 log = root->log_root;
Chris Masone02119d2008-09-05 16:13:11 -04003280
Li Zefan33345d012011-04-20 10:31:50 +08003281 min_key.objectid = ino;
Chris Masone02119d2008-09-05 16:13:11 -04003282 min_key.type = key_type;
3283 min_key.offset = min_offset;
3284
Filipe David Borba Manana6174d3c2013-10-01 16:13:42 +01003285 ret = btrfs_search_forward(root, &min_key, path, trans->transid);
Chris Masone02119d2008-09-05 16:13:11 -04003286
3287 /*
3288 * we didn't find anything from this transaction, see if there
3289 * is anything at all
3290 */
Li Zefan33345d012011-04-20 10:31:50 +08003291 if (ret != 0 || min_key.objectid != ino || min_key.type != key_type) {
3292 min_key.objectid = ino;
Chris Masone02119d2008-09-05 16:13:11 -04003293 min_key.type = key_type;
3294 min_key.offset = (u64)-1;
David Sterbab3b4aa72011-04-21 01:20:15 +02003295 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04003296 ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0);
3297 if (ret < 0) {
David Sterbab3b4aa72011-04-21 01:20:15 +02003298 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04003299 return ret;
3300 }
Li Zefan33345d012011-04-20 10:31:50 +08003301 ret = btrfs_previous_item(root, path, ino, key_type);
Chris Masone02119d2008-09-05 16:13:11 -04003302
3303 /* if ret == 0 there are items for this type,
3304 * create a range to tell us the last key of this type.
3305 * otherwise, there are no items in this directory after
3306 * *min_offset, and we create a range to indicate that.
3307 */
3308 if (ret == 0) {
3309 struct btrfs_key tmp;
3310 btrfs_item_key_to_cpu(path->nodes[0], &tmp,
3311 path->slots[0]);
Chris Masond3977122009-01-05 21:25:51 -05003312 if (key_type == tmp.type)
Chris Masone02119d2008-09-05 16:13:11 -04003313 first_offset = max(min_offset, tmp.offset) + 1;
Chris Masone02119d2008-09-05 16:13:11 -04003314 }
3315 goto done;
3316 }
3317
3318 /* go backward to find any previous key */
Li Zefan33345d012011-04-20 10:31:50 +08003319 ret = btrfs_previous_item(root, path, ino, key_type);
Chris Masone02119d2008-09-05 16:13:11 -04003320 if (ret == 0) {
3321 struct btrfs_key tmp;
3322 btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]);
3323 if (key_type == tmp.type) {
3324 first_offset = tmp.offset;
3325 ret = overwrite_item(trans, log, dst_path,
3326 path->nodes[0], path->slots[0],
3327 &tmp);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04003328 if (ret) {
3329 err = ret;
3330 goto done;
3331 }
Chris Masone02119d2008-09-05 16:13:11 -04003332 }
3333 }
David Sterbab3b4aa72011-04-21 01:20:15 +02003334 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04003335
3336 /* find the first key from this transaction again */
3337 ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0);
Dulshani Gunawardhanafae7f212013-10-31 10:30:08 +05303338 if (WARN_ON(ret != 0))
Chris Masone02119d2008-09-05 16:13:11 -04003339 goto done;
Chris Masone02119d2008-09-05 16:13:11 -04003340
3341 /*
3342 * we have a block from this transaction, log every item in it
3343 * from our directory
3344 */
Chris Masond3977122009-01-05 21:25:51 -05003345 while (1) {
Chris Masone02119d2008-09-05 16:13:11 -04003346 struct btrfs_key tmp;
3347 src = path->nodes[0];
3348 nritems = btrfs_header_nritems(src);
3349 for (i = path->slots[0]; i < nritems; i++) {
Filipe Manana2f2ff0e2015-03-20 17:19:46 +00003350 struct btrfs_dir_item *di;
3351
Chris Masone02119d2008-09-05 16:13:11 -04003352 btrfs_item_key_to_cpu(src, &min_key, i);
3353
Li Zefan33345d012011-04-20 10:31:50 +08003354 if (min_key.objectid != ino || min_key.type != key_type)
Chris Masone02119d2008-09-05 16:13:11 -04003355 goto done;
3356 ret = overwrite_item(trans, log, dst_path, src, i,
3357 &min_key);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04003358 if (ret) {
3359 err = ret;
3360 goto done;
3361 }
Filipe Manana2f2ff0e2015-03-20 17:19:46 +00003362
3363 /*
3364 * We must make sure that when we log a directory entry,
3365 * the corresponding inode, after log replay, has a
3366 * matching link count. For example:
3367 *
3368 * touch foo
3369 * mkdir mydir
3370 * sync
3371 * ln foo mydir/bar
3372 * xfs_io -c "fsync" mydir
3373 * <crash>
3374 * <mount fs and log replay>
3375 *
3376 * Would result in a fsync log that when replayed, our
3377 * file inode would have a link count of 1, but we get
3378 * two directory entries pointing to the same inode.
3379 * After removing one of the names, it would not be
3380 * possible to remove the other name, which resulted
3381 * always in stale file handle errors, and would not
3382 * be possible to rmdir the parent directory, since
3383 * its i_size could never decrement to the value
3384 * BTRFS_EMPTY_DIR_SIZE, resulting in -ENOTEMPTY errors.
3385 */
3386 di = btrfs_item_ptr(src, i, struct btrfs_dir_item);
3387 btrfs_dir_item_key_to_cpu(src, di, &tmp);
3388 if (ctx &&
3389 (btrfs_dir_transid(src, di) == trans->transid ||
3390 btrfs_dir_type(src, di) == BTRFS_FT_DIR) &&
3391 tmp.type != BTRFS_ROOT_ITEM_KEY)
3392 ctx->log_new_dentries = true;
Chris Masone02119d2008-09-05 16:13:11 -04003393 }
3394 path->slots[0] = nritems;
3395
3396 /*
3397 * look ahead to the next item and see if it is also
3398 * from this directory and from this transaction
3399 */
3400 ret = btrfs_next_leaf(root, path);
3401 if (ret == 1) {
3402 last_offset = (u64)-1;
3403 goto done;
3404 }
3405 btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]);
Li Zefan33345d012011-04-20 10:31:50 +08003406 if (tmp.objectid != ino || tmp.type != key_type) {
Chris Masone02119d2008-09-05 16:13:11 -04003407 last_offset = (u64)-1;
3408 goto done;
3409 }
3410 if (btrfs_header_generation(path->nodes[0]) != trans->transid) {
3411 ret = overwrite_item(trans, log, dst_path,
3412 path->nodes[0], path->slots[0],
3413 &tmp);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04003414 if (ret)
3415 err = ret;
3416 else
3417 last_offset = tmp.offset;
Chris Masone02119d2008-09-05 16:13:11 -04003418 goto done;
3419 }
3420 }
3421done:
David Sterbab3b4aa72011-04-21 01:20:15 +02003422 btrfs_release_path(path);
3423 btrfs_release_path(dst_path);
Chris Masone02119d2008-09-05 16:13:11 -04003424
Yan, Zheng4a500fd2010-05-16 10:49:59 -04003425 if (err == 0) {
3426 *last_offset_ret = last_offset;
3427 /*
3428 * insert the log range keys to indicate where the log
3429 * is valid
3430 */
3431 ret = insert_dir_log_key(trans, log, path, key_type,
Li Zefan33345d012011-04-20 10:31:50 +08003432 ino, first_offset, last_offset);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04003433 if (ret)
3434 err = ret;
3435 }
3436 return err;
Chris Masone02119d2008-09-05 16:13:11 -04003437}
3438
3439/*
3440 * logging directories is very similar to logging inodes, We find all the items
3441 * from the current transaction and write them to the log.
3442 *
3443 * The recovery code scans the directory in the subvolume, and if it finds a
3444 * key in the range logged that is not present in the log tree, then it means
3445 * that dir entry was unlinked during the transaction.
3446 *
3447 * In order for that scan to work, we must include one key smaller than
3448 * the smallest logged by this transaction and one key larger than the largest
3449 * key logged by this transaction.
3450 */
3451static noinline int log_directory_changes(struct btrfs_trans_handle *trans,
3452 struct btrfs_root *root, struct inode *inode,
3453 struct btrfs_path *path,
Filipe Manana2f2ff0e2015-03-20 17:19:46 +00003454 struct btrfs_path *dst_path,
3455 struct btrfs_log_ctx *ctx)
Chris Masone02119d2008-09-05 16:13:11 -04003456{
3457 u64 min_key;
3458 u64 max_key;
3459 int ret;
3460 int key_type = BTRFS_DIR_ITEM_KEY;
3461
3462again:
3463 min_key = 0;
3464 max_key = 0;
Chris Masond3977122009-01-05 21:25:51 -05003465 while (1) {
Chris Masone02119d2008-09-05 16:13:11 -04003466 ret = log_dir_items(trans, root, inode, path,
Filipe Manana2f2ff0e2015-03-20 17:19:46 +00003467 dst_path, key_type, ctx, min_key,
Chris Masone02119d2008-09-05 16:13:11 -04003468 &max_key);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04003469 if (ret)
3470 return ret;
Chris Masone02119d2008-09-05 16:13:11 -04003471 if (max_key == (u64)-1)
3472 break;
3473 min_key = max_key + 1;
3474 }
3475
3476 if (key_type == BTRFS_DIR_ITEM_KEY) {
3477 key_type = BTRFS_DIR_INDEX_KEY;
3478 goto again;
3479 }
3480 return 0;
3481}
3482
3483/*
3484 * a helper function to drop items from the log before we relog an
3485 * inode. max_key_type indicates the highest item type to remove.
3486 * This cannot be run for file data extents because it does not
3487 * free the extents they point to.
3488 */
3489static int drop_objectid_items(struct btrfs_trans_handle *trans,
3490 struct btrfs_root *log,
3491 struct btrfs_path *path,
3492 u64 objectid, int max_key_type)
3493{
3494 int ret;
3495 struct btrfs_key key;
3496 struct btrfs_key found_key;
Josef Bacik18ec90d2012-09-28 11:56:28 -04003497 int start_slot;
Chris Masone02119d2008-09-05 16:13:11 -04003498
3499 key.objectid = objectid;
3500 key.type = max_key_type;
3501 key.offset = (u64)-1;
3502
Chris Masond3977122009-01-05 21:25:51 -05003503 while (1) {
Chris Masone02119d2008-09-05 16:13:11 -04003504 ret = btrfs_search_slot(trans, log, &key, path, -1, 1);
Josef Bacik36508602013-04-25 16:23:32 -04003505 BUG_ON(ret == 0); /* Logic error */
Yan, Zheng4a500fd2010-05-16 10:49:59 -04003506 if (ret < 0)
Chris Masone02119d2008-09-05 16:13:11 -04003507 break;
3508
3509 if (path->slots[0] == 0)
3510 break;
3511
3512 path->slots[0]--;
3513 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
3514 path->slots[0]);
3515
3516 if (found_key.objectid != objectid)
3517 break;
3518
Josef Bacik18ec90d2012-09-28 11:56:28 -04003519 found_key.offset = 0;
3520 found_key.type = 0;
3521 ret = btrfs_bin_search(path->nodes[0], &found_key, 0,
3522 &start_slot);
3523
3524 ret = btrfs_del_items(trans, log, path, start_slot,
3525 path->slots[0] - start_slot + 1);
3526 /*
3527 * If start slot isn't 0 then we don't need to re-search, we've
3528 * found the last guy with the objectid in this tree.
3529 */
3530 if (ret || start_slot != 0)
Tsutomu Itoh65a246c2011-05-19 04:37:44 +00003531 break;
David Sterbab3b4aa72011-04-21 01:20:15 +02003532 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04003533 }
David Sterbab3b4aa72011-04-21 01:20:15 +02003534 btrfs_release_path(path);
Josef Bacik5bdbeb22012-05-29 16:59:49 -04003535 if (ret > 0)
3536 ret = 0;
Yan, Zheng4a500fd2010-05-16 10:49:59 -04003537 return ret;
Chris Masone02119d2008-09-05 16:13:11 -04003538}
3539
Josef Bacik94edf4a2012-09-25 14:56:25 -04003540static void fill_inode_item(struct btrfs_trans_handle *trans,
3541 struct extent_buffer *leaf,
3542 struct btrfs_inode_item *item,
Filipe Manana1a4bcf42015-02-13 12:30:56 +00003543 struct inode *inode, int log_inode_only,
3544 u64 logged_isize)
Josef Bacik94edf4a2012-09-25 14:56:25 -04003545{
Josef Bacik0b1c6cc2012-10-23 16:03:44 -04003546 struct btrfs_map_token token;
Josef Bacik94edf4a2012-09-25 14:56:25 -04003547
Josef Bacik0b1c6cc2012-10-23 16:03:44 -04003548 btrfs_init_map_token(&token);
Josef Bacik94edf4a2012-09-25 14:56:25 -04003549
3550 if (log_inode_only) {
3551 /* set the generation to zero so the recover code
3552 * can tell the difference between an logging
3553 * just to say 'this inode exists' and a logging
3554 * to say 'update this inode with these values'
3555 */
Josef Bacik0b1c6cc2012-10-23 16:03:44 -04003556 btrfs_set_token_inode_generation(leaf, item, 0, &token);
Filipe Manana1a4bcf42015-02-13 12:30:56 +00003557 btrfs_set_token_inode_size(leaf, item, logged_isize, &token);
Josef Bacik94edf4a2012-09-25 14:56:25 -04003558 } else {
Josef Bacik0b1c6cc2012-10-23 16:03:44 -04003559 btrfs_set_token_inode_generation(leaf, item,
3560 BTRFS_I(inode)->generation,
3561 &token);
3562 btrfs_set_token_inode_size(leaf, item, inode->i_size, &token);
Josef Bacik94edf4a2012-09-25 14:56:25 -04003563 }
3564
Josef Bacik0b1c6cc2012-10-23 16:03:44 -04003565 btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token);
3566 btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token);
3567 btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token);
3568 btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token);
3569
David Sterbaa937b972014-12-12 17:39:12 +01003570 btrfs_set_token_timespec_sec(leaf, &item->atime,
Josef Bacik0b1c6cc2012-10-23 16:03:44 -04003571 inode->i_atime.tv_sec, &token);
David Sterbaa937b972014-12-12 17:39:12 +01003572 btrfs_set_token_timespec_nsec(leaf, &item->atime,
Josef Bacik0b1c6cc2012-10-23 16:03:44 -04003573 inode->i_atime.tv_nsec, &token);
3574
David Sterbaa937b972014-12-12 17:39:12 +01003575 btrfs_set_token_timespec_sec(leaf, &item->mtime,
Josef Bacik0b1c6cc2012-10-23 16:03:44 -04003576 inode->i_mtime.tv_sec, &token);
David Sterbaa937b972014-12-12 17:39:12 +01003577 btrfs_set_token_timespec_nsec(leaf, &item->mtime,
Josef Bacik0b1c6cc2012-10-23 16:03:44 -04003578 inode->i_mtime.tv_nsec, &token);
3579
David Sterbaa937b972014-12-12 17:39:12 +01003580 btrfs_set_token_timespec_sec(leaf, &item->ctime,
Josef Bacik0b1c6cc2012-10-23 16:03:44 -04003581 inode->i_ctime.tv_sec, &token);
David Sterbaa937b972014-12-12 17:39:12 +01003582 btrfs_set_token_timespec_nsec(leaf, &item->ctime,
Josef Bacik0b1c6cc2012-10-23 16:03:44 -04003583 inode->i_ctime.tv_nsec, &token);
3584
3585 btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode),
3586 &token);
3587
3588 btrfs_set_token_inode_sequence(leaf, item, inode->i_version, &token);
3589 btrfs_set_token_inode_transid(leaf, item, trans->transid, &token);
3590 btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token);
3591 btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token);
3592 btrfs_set_token_inode_block_group(leaf, item, 0, &token);
Josef Bacik94edf4a2012-09-25 14:56:25 -04003593}
3594
Josef Bacika95249b2012-10-11 16:17:34 -04003595static int log_inode_item(struct btrfs_trans_handle *trans,
3596 struct btrfs_root *log, struct btrfs_path *path,
3597 struct inode *inode)
3598{
3599 struct btrfs_inode_item *inode_item;
Josef Bacika95249b2012-10-11 16:17:34 -04003600 int ret;
3601
Filipe David Borba Mananaefd0c402013-10-07 21:20:44 +01003602 ret = btrfs_insert_empty_item(trans, log, path,
3603 &BTRFS_I(inode)->location,
Josef Bacika95249b2012-10-11 16:17:34 -04003604 sizeof(*inode_item));
3605 if (ret && ret != -EEXIST)
3606 return ret;
3607 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3608 struct btrfs_inode_item);
Filipe Manana1a4bcf42015-02-13 12:30:56 +00003609 fill_inode_item(trans, path->nodes[0], inode_item, inode, 0, 0);
Josef Bacika95249b2012-10-11 16:17:34 -04003610 btrfs_release_path(path);
3611 return 0;
3612}
3613
Chris Mason31ff1cd2008-09-11 16:17:57 -04003614static noinline int copy_items(struct btrfs_trans_handle *trans,
Liu Bod2794402012-08-29 01:07:56 -06003615 struct inode *inode,
Chris Mason31ff1cd2008-09-11 16:17:57 -04003616 struct btrfs_path *dst_path,
Josef Bacik16e75492013-10-22 12:18:51 -04003617 struct btrfs_path *src_path, u64 *last_extent,
Filipe Manana1a4bcf42015-02-13 12:30:56 +00003618 int start_slot, int nr, int inode_only,
3619 u64 logged_isize)
Chris Mason31ff1cd2008-09-11 16:17:57 -04003620{
Jeff Mahoney0b246af2016-06-22 18:54:23 -04003621 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
Chris Mason31ff1cd2008-09-11 16:17:57 -04003622 unsigned long src_offset;
3623 unsigned long dst_offset;
Liu Bod2794402012-08-29 01:07:56 -06003624 struct btrfs_root *log = BTRFS_I(inode)->root->log_root;
Chris Mason31ff1cd2008-09-11 16:17:57 -04003625 struct btrfs_file_extent_item *extent;
3626 struct btrfs_inode_item *inode_item;
Josef Bacik16e75492013-10-22 12:18:51 -04003627 struct extent_buffer *src = src_path->nodes[0];
3628 struct btrfs_key first_key, last_key, key;
Chris Mason31ff1cd2008-09-11 16:17:57 -04003629 int ret;
3630 struct btrfs_key *ins_keys;
3631 u32 *ins_sizes;
3632 char *ins_data;
3633 int i;
Chris Masond20f7042008-12-08 16:58:54 -05003634 struct list_head ordered_sums;
Liu Bod2794402012-08-29 01:07:56 -06003635 int skip_csum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
Josef Bacik16e75492013-10-22 12:18:51 -04003636 bool has_extents = false;
Filipe Manana74121f7c2014-08-07 12:00:44 +01003637 bool need_find_last_extent = true;
Josef Bacik16e75492013-10-22 12:18:51 -04003638 bool done = false;
Chris Masond20f7042008-12-08 16:58:54 -05003639
3640 INIT_LIST_HEAD(&ordered_sums);
Chris Mason31ff1cd2008-09-11 16:17:57 -04003641
3642 ins_data = kmalloc(nr * sizeof(struct btrfs_key) +
3643 nr * sizeof(u32), GFP_NOFS);
liubo2a29edc2011-01-26 06:22:08 +00003644 if (!ins_data)
3645 return -ENOMEM;
3646
Josef Bacik16e75492013-10-22 12:18:51 -04003647 first_key.objectid = (u64)-1;
3648
Chris Mason31ff1cd2008-09-11 16:17:57 -04003649 ins_sizes = (u32 *)ins_data;
3650 ins_keys = (struct btrfs_key *)(ins_data + nr * sizeof(u32));
3651
3652 for (i = 0; i < nr; i++) {
3653 ins_sizes[i] = btrfs_item_size_nr(src, i + start_slot);
3654 btrfs_item_key_to_cpu(src, ins_keys + i, i + start_slot);
3655 }
3656 ret = btrfs_insert_empty_items(trans, log, dst_path,
3657 ins_keys, ins_sizes, nr);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04003658 if (ret) {
3659 kfree(ins_data);
3660 return ret;
3661 }
Chris Mason31ff1cd2008-09-11 16:17:57 -04003662
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003663 for (i = 0; i < nr; i++, dst_path->slots[0]++) {
Chris Mason31ff1cd2008-09-11 16:17:57 -04003664 dst_offset = btrfs_item_ptr_offset(dst_path->nodes[0],
3665 dst_path->slots[0]);
3666
3667 src_offset = btrfs_item_ptr_offset(src, start_slot + i);
3668
Josef Bacik16e75492013-10-22 12:18:51 -04003669 if ((i == (nr - 1)))
3670 last_key = ins_keys[i];
3671
Josef Bacik94edf4a2012-09-25 14:56:25 -04003672 if (ins_keys[i].type == BTRFS_INODE_ITEM_KEY) {
Chris Mason31ff1cd2008-09-11 16:17:57 -04003673 inode_item = btrfs_item_ptr(dst_path->nodes[0],
3674 dst_path->slots[0],
3675 struct btrfs_inode_item);
Josef Bacik94edf4a2012-09-25 14:56:25 -04003676 fill_inode_item(trans, dst_path->nodes[0], inode_item,
Filipe Manana1a4bcf42015-02-13 12:30:56 +00003677 inode, inode_only == LOG_INODE_EXISTS,
3678 logged_isize);
Josef Bacik94edf4a2012-09-25 14:56:25 -04003679 } else {
3680 copy_extent_buffer(dst_path->nodes[0], src, dst_offset,
3681 src_offset, ins_sizes[i]);
Chris Mason31ff1cd2008-09-11 16:17:57 -04003682 }
Josef Bacik94edf4a2012-09-25 14:56:25 -04003683
Josef Bacik16e75492013-10-22 12:18:51 -04003684 /*
3685 * We set need_find_last_extent here in case we know we were
3686 * processing other items and then walk into the first extent in
3687 * the inode. If we don't hit an extent then nothing changes,
3688 * we'll do the last search the next time around.
3689 */
3690 if (ins_keys[i].type == BTRFS_EXTENT_DATA_KEY) {
3691 has_extents = true;
Filipe Manana74121f7c2014-08-07 12:00:44 +01003692 if (first_key.objectid == (u64)-1)
Josef Bacik16e75492013-10-22 12:18:51 -04003693 first_key = ins_keys[i];
3694 } else {
3695 need_find_last_extent = false;
3696 }
3697
Chris Mason31ff1cd2008-09-11 16:17:57 -04003698 /* take a reference on file data extents so that truncates
3699 * or deletes of this inode don't have to relog the inode
3700 * again
3701 */
David Sterba962a2982014-06-04 18:41:45 +02003702 if (ins_keys[i].type == BTRFS_EXTENT_DATA_KEY &&
Liu Bod2794402012-08-29 01:07:56 -06003703 !skip_csum) {
Chris Mason31ff1cd2008-09-11 16:17:57 -04003704 int found_type;
3705 extent = btrfs_item_ptr(src, start_slot + i,
3706 struct btrfs_file_extent_item);
3707
liubo8e531cd2011-05-06 10:36:09 +08003708 if (btrfs_file_extent_generation(src, extent) < trans->transid)
3709 continue;
3710
Chris Mason31ff1cd2008-09-11 16:17:57 -04003711 found_type = btrfs_file_extent_type(src, extent);
Josef Bacik6f1fed72012-09-26 11:07:06 -04003712 if (found_type == BTRFS_FILE_EXTENT_REG) {
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003713 u64 ds, dl, cs, cl;
3714 ds = btrfs_file_extent_disk_bytenr(src,
3715 extent);
3716 /* ds == 0 is a hole */
3717 if (ds == 0)
3718 continue;
3719
3720 dl = btrfs_file_extent_disk_num_bytes(src,
3721 extent);
3722 cs = btrfs_file_extent_offset(src, extent);
3723 cl = btrfs_file_extent_num_bytes(src,
Joe Perchesa419aef2009-08-18 11:18:35 -07003724 extent);
Chris Mason580afd72008-12-08 19:15:39 -05003725 if (btrfs_file_extent_compression(src,
3726 extent)) {
3727 cs = 0;
3728 cl = dl;
3729 }
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003730
3731 ret = btrfs_lookup_csums_range(
Jeff Mahoney0b246af2016-06-22 18:54:23 -04003732 fs_info->csum_root,
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003733 ds + cs, ds + cs + cl - 1,
Arne Jansena2de7332011-03-08 14:14:00 +01003734 &ordered_sums, 0);
Josef Bacik36508602013-04-25 16:23:32 -04003735 if (ret) {
3736 btrfs_release_path(dst_path);
3737 kfree(ins_data);
3738 return ret;
3739 }
Chris Mason31ff1cd2008-09-11 16:17:57 -04003740 }
3741 }
Chris Mason31ff1cd2008-09-11 16:17:57 -04003742 }
3743
3744 btrfs_mark_buffer_dirty(dst_path->nodes[0]);
David Sterbab3b4aa72011-04-21 01:20:15 +02003745 btrfs_release_path(dst_path);
Chris Mason31ff1cd2008-09-11 16:17:57 -04003746 kfree(ins_data);
Chris Masond20f7042008-12-08 16:58:54 -05003747
3748 /*
3749 * we have to do this after the loop above to avoid changing the
3750 * log tree while trying to change the log tree.
3751 */
Yan, Zheng4a500fd2010-05-16 10:49:59 -04003752 ret = 0;
Chris Masond3977122009-01-05 21:25:51 -05003753 while (!list_empty(&ordered_sums)) {
Chris Masond20f7042008-12-08 16:58:54 -05003754 struct btrfs_ordered_sum *sums = list_entry(ordered_sums.next,
3755 struct btrfs_ordered_sum,
3756 list);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04003757 if (!ret)
3758 ret = btrfs_csum_file_blocks(trans, log, sums);
Chris Masond20f7042008-12-08 16:58:54 -05003759 list_del(&sums->list);
3760 kfree(sums);
3761 }
Josef Bacik16e75492013-10-22 12:18:51 -04003762
3763 if (!has_extents)
3764 return ret;
3765
Filipe Manana74121f7c2014-08-07 12:00:44 +01003766 if (need_find_last_extent && *last_extent == first_key.offset) {
3767 /*
3768 * We don't have any leafs between our current one and the one
3769 * we processed before that can have file extent items for our
3770 * inode (and have a generation number smaller than our current
3771 * transaction id).
3772 */
3773 need_find_last_extent = false;
3774 }
3775
Josef Bacik16e75492013-10-22 12:18:51 -04003776 /*
3777 * Because we use btrfs_search_forward we could skip leaves that were
3778 * not modified and then assume *last_extent is valid when it really
3779 * isn't. So back up to the previous leaf and read the end of the last
3780 * extent before we go and fill in holes.
3781 */
3782 if (need_find_last_extent) {
3783 u64 len;
3784
3785 ret = btrfs_prev_leaf(BTRFS_I(inode)->root, src_path);
3786 if (ret < 0)
3787 return ret;
3788 if (ret)
3789 goto fill_holes;
3790 if (src_path->slots[0])
3791 src_path->slots[0]--;
3792 src = src_path->nodes[0];
3793 btrfs_item_key_to_cpu(src, &key, src_path->slots[0]);
3794 if (key.objectid != btrfs_ino(inode) ||
3795 key.type != BTRFS_EXTENT_DATA_KEY)
3796 goto fill_holes;
3797 extent = btrfs_item_ptr(src, src_path->slots[0],
3798 struct btrfs_file_extent_item);
3799 if (btrfs_file_extent_type(src, extent) ==
3800 BTRFS_FILE_EXTENT_INLINE) {
Chris Mason514ac8a2014-01-03 21:07:00 -08003801 len = btrfs_file_extent_inline_len(src,
3802 src_path->slots[0],
3803 extent);
Josef Bacik16e75492013-10-22 12:18:51 -04003804 *last_extent = ALIGN(key.offset + len,
Jeff Mahoney0b246af2016-06-22 18:54:23 -04003805 fs_info->sectorsize);
Josef Bacik16e75492013-10-22 12:18:51 -04003806 } else {
3807 len = btrfs_file_extent_num_bytes(src, extent);
3808 *last_extent = key.offset + len;
3809 }
3810 }
3811fill_holes:
3812 /* So we did prev_leaf, now we need to move to the next leaf, but a few
3813 * things could have happened
3814 *
3815 * 1) A merge could have happened, so we could currently be on a leaf
3816 * that holds what we were copying in the first place.
3817 * 2) A split could have happened, and now not all of the items we want
3818 * are on the same leaf.
3819 *
3820 * So we need to adjust how we search for holes, we need to drop the
3821 * path and re-search for the first extent key we found, and then walk
3822 * forward until we hit the last one we copied.
3823 */
3824 if (need_find_last_extent) {
3825 /* btrfs_prev_leaf could return 1 without releasing the path */
3826 btrfs_release_path(src_path);
3827 ret = btrfs_search_slot(NULL, BTRFS_I(inode)->root, &first_key,
3828 src_path, 0, 0);
3829 if (ret < 0)
3830 return ret;
3831 ASSERT(ret == 0);
3832 src = src_path->nodes[0];
3833 i = src_path->slots[0];
3834 } else {
3835 i = start_slot;
3836 }
3837
3838 /*
3839 * Ok so here we need to go through and fill in any holes we may have
3840 * to make sure that holes are punched for those areas in case they had
3841 * extents previously.
3842 */
3843 while (!done) {
3844 u64 offset, len;
3845 u64 extent_end;
3846
3847 if (i >= btrfs_header_nritems(src_path->nodes[0])) {
3848 ret = btrfs_next_leaf(BTRFS_I(inode)->root, src_path);
3849 if (ret < 0)
3850 return ret;
3851 ASSERT(ret == 0);
3852 src = src_path->nodes[0];
3853 i = 0;
3854 }
3855
3856 btrfs_item_key_to_cpu(src, &key, i);
3857 if (!btrfs_comp_cpu_keys(&key, &last_key))
3858 done = true;
3859 if (key.objectid != btrfs_ino(inode) ||
3860 key.type != BTRFS_EXTENT_DATA_KEY) {
3861 i++;
3862 continue;
3863 }
3864 extent = btrfs_item_ptr(src, i, struct btrfs_file_extent_item);
3865 if (btrfs_file_extent_type(src, extent) ==
3866 BTRFS_FILE_EXTENT_INLINE) {
Chris Mason514ac8a2014-01-03 21:07:00 -08003867 len = btrfs_file_extent_inline_len(src, i, extent);
Jeff Mahoneyda170662016-06-15 09:22:56 -04003868 extent_end = ALIGN(key.offset + len,
Jeff Mahoney0b246af2016-06-22 18:54:23 -04003869 fs_info->sectorsize);
Josef Bacik16e75492013-10-22 12:18:51 -04003870 } else {
3871 len = btrfs_file_extent_num_bytes(src, extent);
3872 extent_end = key.offset + len;
3873 }
3874 i++;
3875
3876 if (*last_extent == key.offset) {
3877 *last_extent = extent_end;
3878 continue;
3879 }
3880 offset = *last_extent;
3881 len = key.offset - *last_extent;
3882 ret = btrfs_insert_file_extent(trans, log, btrfs_ino(inode),
3883 offset, 0, 0, len, 0, len, 0,
3884 0, 0);
3885 if (ret)
3886 break;
Filipe Manana74121f7c2014-08-07 12:00:44 +01003887 *last_extent = extent_end;
Josef Bacik16e75492013-10-22 12:18:51 -04003888 }
3889 /*
3890 * Need to let the callers know we dropped the path so they should
3891 * re-search.
3892 */
3893 if (!ret && need_find_last_extent)
3894 ret = 1;
Yan, Zheng4a500fd2010-05-16 10:49:59 -04003895 return ret;
Chris Mason31ff1cd2008-09-11 16:17:57 -04003896}
3897
Josef Bacik5dc562c2012-08-17 13:14:17 -04003898static int extent_cmp(void *priv, struct list_head *a, struct list_head *b)
3899{
3900 struct extent_map *em1, *em2;
3901
3902 em1 = list_entry(a, struct extent_map, list);
3903 em2 = list_entry(b, struct extent_map, list);
3904
3905 if (em1->start < em2->start)
3906 return -1;
3907 else if (em1->start > em2->start)
3908 return 1;
3909 return 0;
3910}
3911
Filipe Manana8407f552014-09-05 15:14:39 +01003912static int wait_ordered_extents(struct btrfs_trans_handle *trans,
3913 struct inode *inode,
3914 struct btrfs_root *root,
3915 const struct extent_map *em,
3916 const struct list_head *logged_list,
3917 bool *ordered_io_error)
Josef Bacik5dc562c2012-08-17 13:14:17 -04003918{
Jeff Mahoney0b246af2016-06-22 18:54:23 -04003919 struct btrfs_fs_info *fs_info = root->fs_info;
Josef Bacik2ab28f32012-10-12 15:27:49 -04003920 struct btrfs_ordered_extent *ordered;
Filipe Manana8407f552014-09-05 15:14:39 +01003921 struct btrfs_root *log = root->log_root;
Josef Bacik2ab28f32012-10-12 15:27:49 -04003922 u64 mod_start = em->mod_start;
3923 u64 mod_len = em->mod_len;
Filipe Manana8407f552014-09-05 15:14:39 +01003924 const bool skip_csum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
Josef Bacik2ab28f32012-10-12 15:27:49 -04003925 u64 csum_offset;
3926 u64 csum_len;
Filipe Manana8407f552014-09-05 15:14:39 +01003927 LIST_HEAD(ordered_sums);
3928 int ret = 0;
Josef Bacik09a2a8f92013-04-05 16:51:15 -04003929
Filipe Manana8407f552014-09-05 15:14:39 +01003930 *ordered_io_error = false;
Josef Bacik70c8a912012-10-11 16:54:30 -04003931
Filipe Manana8407f552014-09-05 15:14:39 +01003932 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
3933 em->block_start == EXTENT_MAP_HOLE)
Josef Bacik70c8a912012-10-11 16:54:30 -04003934 return 0;
3935
Josef Bacik2ab28f32012-10-12 15:27:49 -04003936 /*
Filipe Manana8407f552014-09-05 15:14:39 +01003937 * Wait far any ordered extent that covers our extent map. If it
3938 * finishes without an error, first check and see if our csums are on
3939 * our outstanding ordered extents.
Josef Bacik2ab28f32012-10-12 15:27:49 -04003940 */
Miao Xie827463c2014-01-14 20:31:51 +08003941 list_for_each_entry(ordered, logged_list, log_list) {
Josef Bacik2ab28f32012-10-12 15:27:49 -04003942 struct btrfs_ordered_sum *sum;
3943
3944 if (!mod_len)
3945 break;
3946
Josef Bacik2ab28f32012-10-12 15:27:49 -04003947 if (ordered->file_offset + ordered->len <= mod_start ||
3948 mod_start + mod_len <= ordered->file_offset)
3949 continue;
3950
Filipe Manana8407f552014-09-05 15:14:39 +01003951 if (!test_bit(BTRFS_ORDERED_IO_DONE, &ordered->flags) &&
3952 !test_bit(BTRFS_ORDERED_IOERR, &ordered->flags) &&
3953 !test_bit(BTRFS_ORDERED_DIRECT, &ordered->flags)) {
3954 const u64 start = ordered->file_offset;
3955 const u64 end = ordered->file_offset + ordered->len - 1;
3956
3957 WARN_ON(ordered->inode != inode);
3958 filemap_fdatawrite_range(inode->i_mapping, start, end);
3959 }
3960
3961 wait_event(ordered->wait,
3962 (test_bit(BTRFS_ORDERED_IO_DONE, &ordered->flags) ||
3963 test_bit(BTRFS_ORDERED_IOERR, &ordered->flags)));
3964
3965 if (test_bit(BTRFS_ORDERED_IOERR, &ordered->flags)) {
Filipe Mananab38ef712014-11-13 17:01:45 +00003966 /*
3967 * Clear the AS_EIO/AS_ENOSPC flags from the inode's
3968 * i_mapping flags, so that the next fsync won't get
3969 * an outdated io error too.
3970 */
Miklos Szeredif0312212016-09-16 12:44:21 +02003971 filemap_check_errors(inode->i_mapping);
Filipe Manana8407f552014-09-05 15:14:39 +01003972 *ordered_io_error = true;
3973 break;
3974 }
Josef Bacik2ab28f32012-10-12 15:27:49 -04003975 /*
3976 * We are going to copy all the csums on this ordered extent, so
3977 * go ahead and adjust mod_start and mod_len in case this
3978 * ordered extent has already been logged.
3979 */
3980 if (ordered->file_offset > mod_start) {
3981 if (ordered->file_offset + ordered->len >=
3982 mod_start + mod_len)
3983 mod_len = ordered->file_offset - mod_start;
3984 /*
3985 * If we have this case
3986 *
3987 * |--------- logged extent ---------|
3988 * |----- ordered extent ----|
3989 *
3990 * Just don't mess with mod_start and mod_len, we'll
3991 * just end up logging more csums than we need and it
3992 * will be ok.
3993 */
3994 } else {
3995 if (ordered->file_offset + ordered->len <
3996 mod_start + mod_len) {
3997 mod_len = (mod_start + mod_len) -
3998 (ordered->file_offset + ordered->len);
3999 mod_start = ordered->file_offset +
4000 ordered->len;
4001 } else {
4002 mod_len = 0;
4003 }
4004 }
4005
Filipe Manana8407f552014-09-05 15:14:39 +01004006 if (skip_csum)
4007 continue;
4008
Josef Bacik2ab28f32012-10-12 15:27:49 -04004009 /*
4010 * To keep us from looping for the above case of an ordered
4011 * extent that falls inside of the logged extent.
4012 */
4013 if (test_and_set_bit(BTRFS_ORDERED_LOGGED_CSUM,
4014 &ordered->flags))
4015 continue;
Josef Bacik2ab28f32012-10-12 15:27:49 -04004016
Josef Bacik2ab28f32012-10-12 15:27:49 -04004017 list_for_each_entry(sum, &ordered->list, list) {
4018 ret = btrfs_csum_file_blocks(trans, log, sum);
Miao Xie827463c2014-01-14 20:31:51 +08004019 if (ret)
Filipe Manana8407f552014-09-05 15:14:39 +01004020 break;
Josef Bacik2ab28f32012-10-12 15:27:49 -04004021 }
Josef Bacik2ab28f32012-10-12 15:27:49 -04004022 }
Josef Bacik2ab28f32012-10-12 15:27:49 -04004023
Filipe Manana8407f552014-09-05 15:14:39 +01004024 if (*ordered_io_error || !mod_len || ret || skip_csum)
Josef Bacik2ab28f32012-10-12 15:27:49 -04004025 return ret;
4026
Filipe David Borba Manana488111a2013-10-28 16:30:29 +00004027 if (em->compress_type) {
4028 csum_offset = 0;
Filipe Manana8407f552014-09-05 15:14:39 +01004029 csum_len = max(em->block_len, em->orig_block_len);
Filipe David Borba Manana488111a2013-10-28 16:30:29 +00004030 } else {
4031 csum_offset = mod_start - em->start;
4032 csum_len = mod_len;
4033 }
Josef Bacik2ab28f32012-10-12 15:27:49 -04004034
Josef Bacik70c8a912012-10-11 16:54:30 -04004035 /* block start is already adjusted for the file extent offset. */
Jeff Mahoney0b246af2016-06-22 18:54:23 -04004036 ret = btrfs_lookup_csums_range(fs_info->csum_root,
Josef Bacik70c8a912012-10-11 16:54:30 -04004037 em->block_start + csum_offset,
4038 em->block_start + csum_offset +
4039 csum_len - 1, &ordered_sums, 0);
4040 if (ret)
4041 return ret;
4042
4043 while (!list_empty(&ordered_sums)) {
4044 struct btrfs_ordered_sum *sums = list_entry(ordered_sums.next,
4045 struct btrfs_ordered_sum,
4046 list);
4047 if (!ret)
4048 ret = btrfs_csum_file_blocks(trans, log, sums);
4049 list_del(&sums->list);
4050 kfree(sums);
4051 }
4052
4053 return ret;
Josef Bacik5dc562c2012-08-17 13:14:17 -04004054}
4055
Filipe Manana8407f552014-09-05 15:14:39 +01004056static int log_one_extent(struct btrfs_trans_handle *trans,
4057 struct inode *inode, struct btrfs_root *root,
4058 const struct extent_map *em,
4059 struct btrfs_path *path,
4060 const struct list_head *logged_list,
4061 struct btrfs_log_ctx *ctx)
4062{
4063 struct btrfs_root *log = root->log_root;
4064 struct btrfs_file_extent_item *fi;
4065 struct extent_buffer *leaf;
4066 struct btrfs_map_token token;
4067 struct btrfs_key key;
4068 u64 extent_offset = em->start - em->orig_start;
4069 u64 block_len;
4070 int ret;
4071 int extent_inserted = 0;
4072 bool ordered_io_err = false;
4073
4074 ret = wait_ordered_extents(trans, inode, root, em, logged_list,
4075 &ordered_io_err);
4076 if (ret)
4077 return ret;
4078
4079 if (ordered_io_err) {
4080 ctx->io_err = -EIO;
4081 return 0;
4082 }
4083
4084 btrfs_init_map_token(&token);
4085
4086 ret = __btrfs_drop_extents(trans, log, inode, path, em->start,
4087 em->start + em->len, NULL, 0, 1,
4088 sizeof(*fi), &extent_inserted);
4089 if (ret)
4090 return ret;
4091
4092 if (!extent_inserted) {
4093 key.objectid = btrfs_ino(inode);
4094 key.type = BTRFS_EXTENT_DATA_KEY;
4095 key.offset = em->start;
4096
4097 ret = btrfs_insert_empty_item(trans, log, path, &key,
4098 sizeof(*fi));
4099 if (ret)
4100 return ret;
4101 }
4102 leaf = path->nodes[0];
4103 fi = btrfs_item_ptr(leaf, path->slots[0],
4104 struct btrfs_file_extent_item);
4105
Josef Bacik50d9aa92014-11-21 14:52:38 -05004106 btrfs_set_token_file_extent_generation(leaf, fi, trans->transid,
Filipe Manana8407f552014-09-05 15:14:39 +01004107 &token);
4108 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
4109 btrfs_set_token_file_extent_type(leaf, fi,
4110 BTRFS_FILE_EXTENT_PREALLOC,
4111 &token);
4112 else
4113 btrfs_set_token_file_extent_type(leaf, fi,
4114 BTRFS_FILE_EXTENT_REG,
4115 &token);
4116
4117 block_len = max(em->block_len, em->orig_block_len);
4118 if (em->compress_type != BTRFS_COMPRESS_NONE) {
4119 btrfs_set_token_file_extent_disk_bytenr(leaf, fi,
4120 em->block_start,
4121 &token);
4122 btrfs_set_token_file_extent_disk_num_bytes(leaf, fi, block_len,
4123 &token);
4124 } else if (em->block_start < EXTENT_MAP_LAST_BYTE) {
4125 btrfs_set_token_file_extent_disk_bytenr(leaf, fi,
4126 em->block_start -
4127 extent_offset, &token);
4128 btrfs_set_token_file_extent_disk_num_bytes(leaf, fi, block_len,
4129 &token);
4130 } else {
4131 btrfs_set_token_file_extent_disk_bytenr(leaf, fi, 0, &token);
4132 btrfs_set_token_file_extent_disk_num_bytes(leaf, fi, 0,
4133 &token);
4134 }
4135
4136 btrfs_set_token_file_extent_offset(leaf, fi, extent_offset, &token);
4137 btrfs_set_token_file_extent_num_bytes(leaf, fi, em->len, &token);
4138 btrfs_set_token_file_extent_ram_bytes(leaf, fi, em->ram_bytes, &token);
4139 btrfs_set_token_file_extent_compression(leaf, fi, em->compress_type,
4140 &token);
4141 btrfs_set_token_file_extent_encryption(leaf, fi, 0, &token);
4142 btrfs_set_token_file_extent_other_encoding(leaf, fi, 0, &token);
4143 btrfs_mark_buffer_dirty(leaf);
4144
4145 btrfs_release_path(path);
4146
4147 return ret;
4148}
4149
Josef Bacik5dc562c2012-08-17 13:14:17 -04004150static int btrfs_log_changed_extents(struct btrfs_trans_handle *trans,
4151 struct btrfs_root *root,
4152 struct inode *inode,
Miao Xie827463c2014-01-14 20:31:51 +08004153 struct btrfs_path *path,
Filipe Manana8407f552014-09-05 15:14:39 +01004154 struct list_head *logged_list,
Filipe Mananade0ee0e2016-01-21 10:17:54 +00004155 struct btrfs_log_ctx *ctx,
4156 const u64 start,
4157 const u64 end)
Josef Bacik5dc562c2012-08-17 13:14:17 -04004158{
Josef Bacik5dc562c2012-08-17 13:14:17 -04004159 struct extent_map *em, *n;
4160 struct list_head extents;
4161 struct extent_map_tree *tree = &BTRFS_I(inode)->extent_tree;
4162 u64 test_gen;
4163 int ret = 0;
Josef Bacik2ab28f32012-10-12 15:27:49 -04004164 int num = 0;
Josef Bacik5dc562c2012-08-17 13:14:17 -04004165
4166 INIT_LIST_HEAD(&extents);
4167
Filipe Manana5f9a8a52016-05-12 13:53:36 +01004168 down_write(&BTRFS_I(inode)->dio_sem);
Josef Bacik5dc562c2012-08-17 13:14:17 -04004169 write_lock(&tree->lock);
4170 test_gen = root->fs_info->last_trans_committed;
4171
4172 list_for_each_entry_safe(em, n, &tree->modified_extents, list) {
4173 list_del_init(&em->list);
Josef Bacik2ab28f32012-10-12 15:27:49 -04004174
4175 /*
4176 * Just an arbitrary number, this can be really CPU intensive
4177 * once we start getting a lot of extents, and really once we
4178 * have a bunch of extents we just want to commit since it will
4179 * be faster.
4180 */
4181 if (++num > 32768) {
4182 list_del_init(&tree->modified_extents);
4183 ret = -EFBIG;
4184 goto process;
4185 }
4186
Josef Bacik5dc562c2012-08-17 13:14:17 -04004187 if (em->generation <= test_gen)
4188 continue;
Josef Bacikff44c6e2012-09-14 12:59:20 -04004189 /* Need a ref to keep it from getting evicted from cache */
4190 atomic_inc(&em->refs);
4191 set_bit(EXTENT_FLAG_LOGGING, &em->flags);
Josef Bacik5dc562c2012-08-17 13:14:17 -04004192 list_add_tail(&em->list, &extents);
Josef Bacik2ab28f32012-10-12 15:27:49 -04004193 num++;
Josef Bacik5dc562c2012-08-17 13:14:17 -04004194 }
4195
4196 list_sort(NULL, &extents, extent_cmp);
Filipe Mananade0ee0e2016-01-21 10:17:54 +00004197 btrfs_get_logged_extents(inode, logged_list, start, end);
Filipe Manana5f9a8a52016-05-12 13:53:36 +01004198 /*
4199 * Some ordered extents started by fsync might have completed
4200 * before we could collect them into the list logged_list, which
4201 * means they're gone, not in our logged_list nor in the inode's
4202 * ordered tree. We want the application/user space to know an
4203 * error happened while attempting to persist file data so that
4204 * it can take proper action. If such error happened, we leave
4205 * without writing to the log tree and the fsync must report the
4206 * file data write error and not commit the current transaction.
4207 */
Miklos Szeredif0312212016-09-16 12:44:21 +02004208 ret = filemap_check_errors(inode->i_mapping);
Filipe Manana5f9a8a52016-05-12 13:53:36 +01004209 if (ret)
4210 ctx->io_err = ret;
Josef Bacik2ab28f32012-10-12 15:27:49 -04004211process:
Josef Bacik5dc562c2012-08-17 13:14:17 -04004212 while (!list_empty(&extents)) {
4213 em = list_entry(extents.next, struct extent_map, list);
4214
4215 list_del_init(&em->list);
4216
4217 /*
4218 * If we had an error we just need to delete everybody from our
4219 * private list.
4220 */
Josef Bacikff44c6e2012-09-14 12:59:20 -04004221 if (ret) {
Josef Bacik201a9032013-01-24 12:02:07 -05004222 clear_em_logging(tree, em);
Josef Bacikff44c6e2012-09-14 12:59:20 -04004223 free_extent_map(em);
Josef Bacik5dc562c2012-08-17 13:14:17 -04004224 continue;
Josef Bacikff44c6e2012-09-14 12:59:20 -04004225 }
4226
4227 write_unlock(&tree->lock);
Josef Bacik5dc562c2012-08-17 13:14:17 -04004228
Filipe Manana8407f552014-09-05 15:14:39 +01004229 ret = log_one_extent(trans, inode, root, em, path, logged_list,
4230 ctx);
Josef Bacikff44c6e2012-09-14 12:59:20 -04004231 write_lock(&tree->lock);
Josef Bacik201a9032013-01-24 12:02:07 -05004232 clear_em_logging(tree, em);
4233 free_extent_map(em);
Josef Bacik5dc562c2012-08-17 13:14:17 -04004234 }
Josef Bacikff44c6e2012-09-14 12:59:20 -04004235 WARN_ON(!list_empty(&extents));
4236 write_unlock(&tree->lock);
Filipe Manana5f9a8a52016-05-12 13:53:36 +01004237 up_write(&BTRFS_I(inode)->dio_sem);
Josef Bacik5dc562c2012-08-17 13:14:17 -04004238
Josef Bacik5dc562c2012-08-17 13:14:17 -04004239 btrfs_release_path(path);
Josef Bacik5dc562c2012-08-17 13:14:17 -04004240 return ret;
4241}
4242
Filipe Manana1a4bcf42015-02-13 12:30:56 +00004243static int logged_inode_size(struct btrfs_root *log, struct inode *inode,
4244 struct btrfs_path *path, u64 *size_ret)
4245{
4246 struct btrfs_key key;
4247 int ret;
4248
4249 key.objectid = btrfs_ino(inode);
4250 key.type = BTRFS_INODE_ITEM_KEY;
4251 key.offset = 0;
4252
4253 ret = btrfs_search_slot(NULL, log, &key, path, 0, 0);
4254 if (ret < 0) {
4255 return ret;
4256 } else if (ret > 0) {
Filipe Manana2f2ff0e2015-03-20 17:19:46 +00004257 *size_ret = 0;
Filipe Manana1a4bcf42015-02-13 12:30:56 +00004258 } else {
4259 struct btrfs_inode_item *item;
4260
4261 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4262 struct btrfs_inode_item);
4263 *size_ret = btrfs_inode_size(path->nodes[0], item);
4264 }
4265
4266 btrfs_release_path(path);
4267 return 0;
4268}
4269
Filipe Manana36283bf2015-06-20 00:44:51 +01004270/*
4271 * At the moment we always log all xattrs. This is to figure out at log replay
4272 * time which xattrs must have their deletion replayed. If a xattr is missing
4273 * in the log tree and exists in the fs/subvol tree, we delete it. This is
4274 * because if a xattr is deleted, the inode is fsynced and a power failure
4275 * happens, causing the log to be replayed the next time the fs is mounted,
4276 * we want the xattr to not exist anymore (same behaviour as other filesystems
4277 * with a journal, ext3/4, xfs, f2fs, etc).
4278 */
4279static int btrfs_log_all_xattrs(struct btrfs_trans_handle *trans,
4280 struct btrfs_root *root,
4281 struct inode *inode,
4282 struct btrfs_path *path,
4283 struct btrfs_path *dst_path)
4284{
4285 int ret;
4286 struct btrfs_key key;
4287 const u64 ino = btrfs_ino(inode);
4288 int ins_nr = 0;
4289 int start_slot = 0;
4290
4291 key.objectid = ino;
4292 key.type = BTRFS_XATTR_ITEM_KEY;
4293 key.offset = 0;
4294
4295 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4296 if (ret < 0)
4297 return ret;
4298
4299 while (true) {
4300 int slot = path->slots[0];
4301 struct extent_buffer *leaf = path->nodes[0];
4302 int nritems = btrfs_header_nritems(leaf);
4303
4304 if (slot >= nritems) {
4305 if (ins_nr > 0) {
4306 u64 last_extent = 0;
4307
4308 ret = copy_items(trans, inode, dst_path, path,
4309 &last_extent, start_slot,
4310 ins_nr, 1, 0);
4311 /* can't be 1, extent items aren't processed */
4312 ASSERT(ret <= 0);
4313 if (ret < 0)
4314 return ret;
4315 ins_nr = 0;
4316 }
4317 ret = btrfs_next_leaf(root, path);
4318 if (ret < 0)
4319 return ret;
4320 else if (ret > 0)
4321 break;
4322 continue;
4323 }
4324
4325 btrfs_item_key_to_cpu(leaf, &key, slot);
4326 if (key.objectid != ino || key.type != BTRFS_XATTR_ITEM_KEY)
4327 break;
4328
4329 if (ins_nr == 0)
4330 start_slot = slot;
4331 ins_nr++;
4332 path->slots[0]++;
4333 cond_resched();
4334 }
4335 if (ins_nr > 0) {
4336 u64 last_extent = 0;
4337
4338 ret = copy_items(trans, inode, dst_path, path,
4339 &last_extent, start_slot,
4340 ins_nr, 1, 0);
4341 /* can't be 1, extent items aren't processed */
4342 ASSERT(ret <= 0);
4343 if (ret < 0)
4344 return ret;
4345 }
4346
4347 return 0;
4348}
4349
Filipe Mananaa89ca6f2015-06-25 04:17:46 +01004350/*
4351 * If the no holes feature is enabled we need to make sure any hole between the
4352 * last extent and the i_size of our inode is explicitly marked in the log. This
4353 * is to make sure that doing something like:
4354 *
4355 * 1) create file with 128Kb of data
4356 * 2) truncate file to 64Kb
4357 * 3) truncate file to 256Kb
4358 * 4) fsync file
4359 * 5) <crash/power failure>
4360 * 6) mount fs and trigger log replay
4361 *
4362 * Will give us a file with a size of 256Kb, the first 64Kb of data match what
4363 * the file had in its first 64Kb of data at step 1 and the last 192Kb of the
4364 * file correspond to a hole. The presence of explicit holes in a log tree is
4365 * what guarantees that log replay will remove/adjust file extent items in the
4366 * fs/subvol tree.
4367 *
4368 * Here we do not need to care about holes between extents, that is already done
4369 * by copy_items(). We also only need to do this in the full sync path, where we
4370 * lookup for extents from the fs/subvol tree only. In the fast path case, we
4371 * lookup the list of modified extent maps and if any represents a hole, we
4372 * insert a corresponding extent representing a hole in the log tree.
4373 */
4374static int btrfs_log_trailing_hole(struct btrfs_trans_handle *trans,
4375 struct btrfs_root *root,
4376 struct inode *inode,
4377 struct btrfs_path *path)
4378{
Jeff Mahoney0b246af2016-06-22 18:54:23 -04004379 struct btrfs_fs_info *fs_info = root->fs_info;
Filipe Mananaa89ca6f2015-06-25 04:17:46 +01004380 int ret;
4381 struct btrfs_key key;
4382 u64 hole_start;
4383 u64 hole_size;
4384 struct extent_buffer *leaf;
4385 struct btrfs_root *log = root->log_root;
4386 const u64 ino = btrfs_ino(inode);
4387 const u64 i_size = i_size_read(inode);
4388
Jeff Mahoney0b246af2016-06-22 18:54:23 -04004389 if (!btrfs_fs_incompat(fs_info, NO_HOLES))
Filipe Mananaa89ca6f2015-06-25 04:17:46 +01004390 return 0;
4391
4392 key.objectid = ino;
4393 key.type = BTRFS_EXTENT_DATA_KEY;
4394 key.offset = (u64)-1;
4395
4396 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4397 ASSERT(ret != 0);
4398 if (ret < 0)
4399 return ret;
4400
4401 ASSERT(path->slots[0] > 0);
4402 path->slots[0]--;
4403 leaf = path->nodes[0];
4404 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4405
4406 if (key.objectid != ino || key.type != BTRFS_EXTENT_DATA_KEY) {
4407 /* inode does not have any extents */
4408 hole_start = 0;
4409 hole_size = i_size;
4410 } else {
4411 struct btrfs_file_extent_item *extent;
4412 u64 len;
4413
4414 /*
4415 * If there's an extent beyond i_size, an explicit hole was
4416 * already inserted by copy_items().
4417 */
4418 if (key.offset >= i_size)
4419 return 0;
4420
4421 extent = btrfs_item_ptr(leaf, path->slots[0],
4422 struct btrfs_file_extent_item);
4423
4424 if (btrfs_file_extent_type(leaf, extent) ==
4425 BTRFS_FILE_EXTENT_INLINE) {
4426 len = btrfs_file_extent_inline_len(leaf,
4427 path->slots[0],
4428 extent);
4429 ASSERT(len == i_size);
4430 return 0;
4431 }
4432
4433 len = btrfs_file_extent_num_bytes(leaf, extent);
4434 /* Last extent goes beyond i_size, no need to log a hole. */
4435 if (key.offset + len > i_size)
4436 return 0;
4437 hole_start = key.offset + len;
4438 hole_size = i_size - hole_start;
4439 }
4440 btrfs_release_path(path);
4441
4442 /* Last extent ends at i_size. */
4443 if (hole_size == 0)
4444 return 0;
4445
Jeff Mahoney0b246af2016-06-22 18:54:23 -04004446 hole_size = ALIGN(hole_size, fs_info->sectorsize);
Filipe Mananaa89ca6f2015-06-25 04:17:46 +01004447 ret = btrfs_insert_file_extent(trans, log, ino, hole_start, 0, 0,
4448 hole_size, 0, hole_size, 0, 0, 0);
4449 return ret;
4450}
4451
Filipe Manana56f23fd2016-03-30 23:37:21 +01004452/*
4453 * When we are logging a new inode X, check if it doesn't have a reference that
4454 * matches the reference from some other inode Y created in a past transaction
4455 * and that was renamed in the current transaction. If we don't do this, then at
4456 * log replay time we can lose inode Y (and all its files if it's a directory):
4457 *
4458 * mkdir /mnt/x
4459 * echo "hello world" > /mnt/x/foobar
4460 * sync
4461 * mv /mnt/x /mnt/y
4462 * mkdir /mnt/x # or touch /mnt/x
4463 * xfs_io -c fsync /mnt/x
4464 * <power fail>
4465 * mount fs, trigger log replay
4466 *
4467 * After the log replay procedure, we would lose the first directory and all its
4468 * files (file foobar).
4469 * For the case where inode Y is not a directory we simply end up losing it:
4470 *
4471 * echo "123" > /mnt/foo
4472 * sync
4473 * mv /mnt/foo /mnt/bar
4474 * echo "abc" > /mnt/foo
4475 * xfs_io -c fsync /mnt/foo
4476 * <power fail>
4477 *
4478 * We also need this for cases where a snapshot entry is replaced by some other
4479 * entry (file or directory) otherwise we end up with an unreplayable log due to
4480 * attempts to delete the snapshot entry (entry of type BTRFS_ROOT_ITEM_KEY) as
4481 * if it were a regular entry:
4482 *
4483 * mkdir /mnt/x
4484 * btrfs subvolume snapshot /mnt /mnt/x/snap
4485 * btrfs subvolume delete /mnt/x/snap
4486 * rmdir /mnt/x
4487 * mkdir /mnt/x
4488 * fsync /mnt/x or fsync some new file inside it
4489 * <power fail>
4490 *
4491 * The snapshot delete, rmdir of x, mkdir of a new x and the fsync all happen in
4492 * the same transaction.
4493 */
4494static int btrfs_check_ref_name_override(struct extent_buffer *eb,
4495 const int slot,
4496 const struct btrfs_key *key,
Filipe Manana44f714d2016-06-06 16:11:13 +01004497 struct inode *inode,
4498 u64 *other_ino)
Filipe Manana56f23fd2016-03-30 23:37:21 +01004499{
4500 int ret;
4501 struct btrfs_path *search_path;
4502 char *name = NULL;
4503 u32 name_len = 0;
4504 u32 item_size = btrfs_item_size_nr(eb, slot);
4505 u32 cur_offset = 0;
4506 unsigned long ptr = btrfs_item_ptr_offset(eb, slot);
4507
4508 search_path = btrfs_alloc_path();
4509 if (!search_path)
4510 return -ENOMEM;
4511 search_path->search_commit_root = 1;
4512 search_path->skip_locking = 1;
4513
4514 while (cur_offset < item_size) {
4515 u64 parent;
4516 u32 this_name_len;
4517 u32 this_len;
4518 unsigned long name_ptr;
4519 struct btrfs_dir_item *di;
4520
4521 if (key->type == BTRFS_INODE_REF_KEY) {
4522 struct btrfs_inode_ref *iref;
4523
4524 iref = (struct btrfs_inode_ref *)(ptr + cur_offset);
4525 parent = key->offset;
4526 this_name_len = btrfs_inode_ref_name_len(eb, iref);
4527 name_ptr = (unsigned long)(iref + 1);
4528 this_len = sizeof(*iref) + this_name_len;
4529 } else {
4530 struct btrfs_inode_extref *extref;
4531
4532 extref = (struct btrfs_inode_extref *)(ptr +
4533 cur_offset);
4534 parent = btrfs_inode_extref_parent(eb, extref);
4535 this_name_len = btrfs_inode_extref_name_len(eb, extref);
4536 name_ptr = (unsigned long)&extref->name;
4537 this_len = sizeof(*extref) + this_name_len;
4538 }
4539
4540 if (this_name_len > name_len) {
4541 char *new_name;
4542
4543 new_name = krealloc(name, this_name_len, GFP_NOFS);
4544 if (!new_name) {
4545 ret = -ENOMEM;
4546 goto out;
4547 }
4548 name_len = this_name_len;
4549 name = new_name;
4550 }
4551
4552 read_extent_buffer(eb, name, name_ptr, this_name_len);
4553 di = btrfs_lookup_dir_item(NULL, BTRFS_I(inode)->root,
4554 search_path, parent,
4555 name, this_name_len, 0);
4556 if (di && !IS_ERR(di)) {
Filipe Manana44f714d2016-06-06 16:11:13 +01004557 struct btrfs_key di_key;
4558
4559 btrfs_dir_item_key_to_cpu(search_path->nodes[0],
4560 di, &di_key);
4561 if (di_key.type == BTRFS_INODE_ITEM_KEY) {
4562 ret = 1;
4563 *other_ino = di_key.objectid;
4564 } else {
4565 ret = -EAGAIN;
4566 }
Filipe Manana56f23fd2016-03-30 23:37:21 +01004567 goto out;
4568 } else if (IS_ERR(di)) {
4569 ret = PTR_ERR(di);
4570 goto out;
4571 }
4572 btrfs_release_path(search_path);
4573
4574 cur_offset += this_len;
4575 }
4576 ret = 0;
4577out:
4578 btrfs_free_path(search_path);
4579 kfree(name);
4580 return ret;
4581}
4582
Chris Masone02119d2008-09-05 16:13:11 -04004583/* log a single inode in the tree log.
4584 * At least one parent directory for this inode must exist in the tree
4585 * or be logged already.
4586 *
4587 * Any items from this inode changed by the current transaction are copied
4588 * to the log tree. An extra reference is taken on any extents in this
4589 * file, allowing us to avoid a whole pile of corner cases around logging
4590 * blocks that have been removed from the tree.
4591 *
4592 * See LOG_INODE_ALL and related defines for a description of what inode_only
4593 * does.
4594 *
4595 * This handles both files and directories.
4596 */
Chris Mason12fcfd22009-03-24 10:24:20 -04004597static int btrfs_log_inode(struct btrfs_trans_handle *trans,
Filipe Manana49dae1b2014-09-06 22:34:39 +01004598 struct btrfs_root *root, struct inode *inode,
4599 int inode_only,
4600 const loff_t start,
Filipe Manana8407f552014-09-05 15:14:39 +01004601 const loff_t end,
4602 struct btrfs_log_ctx *ctx)
Chris Masone02119d2008-09-05 16:13:11 -04004603{
Jeff Mahoney0b246af2016-06-22 18:54:23 -04004604 struct btrfs_fs_info *fs_info = root->fs_info;
Chris Masone02119d2008-09-05 16:13:11 -04004605 struct btrfs_path *path;
4606 struct btrfs_path *dst_path;
4607 struct btrfs_key min_key;
4608 struct btrfs_key max_key;
4609 struct btrfs_root *log = root->log_root;
Chris Mason31ff1cd2008-09-11 16:17:57 -04004610 struct extent_buffer *src = NULL;
Miao Xie827463c2014-01-14 20:31:51 +08004611 LIST_HEAD(logged_list);
Josef Bacik16e75492013-10-22 12:18:51 -04004612 u64 last_extent = 0;
Yan, Zheng4a500fd2010-05-16 10:49:59 -04004613 int err = 0;
Chris Masone02119d2008-09-05 16:13:11 -04004614 int ret;
Chris Mason3a5f1d42008-09-11 15:53:37 -04004615 int nritems;
Chris Mason31ff1cd2008-09-11 16:17:57 -04004616 int ins_start_slot = 0;
4617 int ins_nr;
Josef Bacik5dc562c2012-08-17 13:14:17 -04004618 bool fast_search = false;
Li Zefan33345d012011-04-20 10:31:50 +08004619 u64 ino = btrfs_ino(inode);
Filipe Manana49dae1b2014-09-06 22:34:39 +01004620 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
Filipe Manana1a4bcf42015-02-13 12:30:56 +00004621 u64 logged_isize = 0;
Filipe Mananae4545de2015-06-17 12:49:23 +01004622 bool need_log_inode_item = true;
Chris Masone02119d2008-09-05 16:13:11 -04004623
Chris Masone02119d2008-09-05 16:13:11 -04004624 path = btrfs_alloc_path();
Tsutomu Itoh5df67082011-02-01 09:17:35 +00004625 if (!path)
4626 return -ENOMEM;
Chris Masone02119d2008-09-05 16:13:11 -04004627 dst_path = btrfs_alloc_path();
Tsutomu Itoh5df67082011-02-01 09:17:35 +00004628 if (!dst_path) {
4629 btrfs_free_path(path);
4630 return -ENOMEM;
4631 }
Chris Masone02119d2008-09-05 16:13:11 -04004632
Li Zefan33345d012011-04-20 10:31:50 +08004633 min_key.objectid = ino;
Chris Masone02119d2008-09-05 16:13:11 -04004634 min_key.type = BTRFS_INODE_ITEM_KEY;
4635 min_key.offset = 0;
4636
Li Zefan33345d012011-04-20 10:31:50 +08004637 max_key.objectid = ino;
Chris Mason12fcfd22009-03-24 10:24:20 -04004638
Chris Mason12fcfd22009-03-24 10:24:20 -04004639
Josef Bacik5dc562c2012-08-17 13:14:17 -04004640 /* today the code can only do partial logging of directories */
Miao Xie5269b672012-11-01 07:35:23 +00004641 if (S_ISDIR(inode->i_mode) ||
4642 (!test_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4643 &BTRFS_I(inode)->runtime_flags) &&
4644 inode_only == LOG_INODE_EXISTS))
Chris Masone02119d2008-09-05 16:13:11 -04004645 max_key.type = BTRFS_XATTR_ITEM_KEY;
4646 else
4647 max_key.type = (u8)-1;
4648 max_key.offset = (u64)-1;
4649
Filipe Manana2c2c4522015-01-13 16:40:04 +00004650 /*
4651 * Only run delayed items if we are a dir or a new file.
4652 * Otherwise commit the delayed inode only, which is needed in
4653 * order for the log replay code to mark inodes for link count
4654 * fixup (create temporary BTRFS_TREE_LOG_FIXUP_OBJECTID items).
4655 */
Josef Bacik94edf4a2012-09-25 14:56:25 -04004656 if (S_ISDIR(inode->i_mode) ||
Jeff Mahoney0b246af2016-06-22 18:54:23 -04004657 BTRFS_I(inode)->generation > fs_info->last_trans_committed)
Josef Bacik94edf4a2012-09-25 14:56:25 -04004658 ret = btrfs_commit_inode_delayed_items(trans, inode);
Filipe Manana2c2c4522015-01-13 16:40:04 +00004659 else
4660 ret = btrfs_commit_inode_delayed_inode(inode);
4661
4662 if (ret) {
4663 btrfs_free_path(path);
4664 btrfs_free_path(dst_path);
4665 return ret;
Miao Xie16cdcec2011-04-22 18:12:22 +08004666 }
4667
Chris Masone02119d2008-09-05 16:13:11 -04004668 mutex_lock(&BTRFS_I(inode)->log_mutex);
4669
Filipe Manana5e33a2b2016-02-25 23:19:38 +00004670 /*
Chris Masone02119d2008-09-05 16:13:11 -04004671 * a brute force approach to making sure we get the most uptodate
4672 * copies of everything.
4673 */
4674 if (S_ISDIR(inode->i_mode)) {
4675 int max_key_type = BTRFS_DIR_LOG_INDEX_KEY;
4676
Filipe Manana4f764e52015-02-23 19:53:35 +00004677 if (inode_only == LOG_INODE_EXISTS)
4678 max_key_type = BTRFS_XATTR_ITEM_KEY;
Li Zefan33345d012011-04-20 10:31:50 +08004679 ret = drop_objectid_items(trans, log, path, ino, max_key_type);
Chris Masone02119d2008-09-05 16:13:11 -04004680 } else {
Filipe Manana1a4bcf42015-02-13 12:30:56 +00004681 if (inode_only == LOG_INODE_EXISTS) {
4682 /*
4683 * Make sure the new inode item we write to the log has
4684 * the same isize as the current one (if it exists).
4685 * This is necessary to prevent data loss after log
4686 * replay, and also to prevent doing a wrong expanding
4687 * truncate - for e.g. create file, write 4K into offset
4688 * 0, fsync, write 4K into offset 4096, add hard link,
4689 * fsync some other file (to sync log), power fail - if
4690 * we use the inode's current i_size, after log replay
4691 * we get a 8Kb file, with the last 4Kb extent as a hole
4692 * (zeroes), as if an expanding truncate happened,
4693 * instead of getting a file of 4Kb only.
4694 */
4695 err = logged_inode_size(log, inode, path,
4696 &logged_isize);
4697 if (err)
4698 goto out_unlock;
4699 }
Filipe Mananaa7429942015-02-13 16:56:14 +00004700 if (test_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4701 &BTRFS_I(inode)->runtime_flags)) {
4702 if (inode_only == LOG_INODE_EXISTS) {
Filipe Manana4f764e52015-02-23 19:53:35 +00004703 max_key.type = BTRFS_XATTR_ITEM_KEY;
Filipe Mananaa7429942015-02-13 16:56:14 +00004704 ret = drop_objectid_items(trans, log, path, ino,
4705 max_key.type);
4706 } else {
4707 clear_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4708 &BTRFS_I(inode)->runtime_flags);
4709 clear_bit(BTRFS_INODE_COPY_EVERYTHING,
4710 &BTRFS_I(inode)->runtime_flags);
Chris Mason28ed1342014-12-17 09:41:04 -08004711 while(1) {
4712 ret = btrfs_truncate_inode_items(trans,
4713 log, inode, 0, 0);
4714 if (ret != -EAGAIN)
4715 break;
4716 }
Filipe Mananaa7429942015-02-13 16:56:14 +00004717 }
Filipe Manana4f764e52015-02-23 19:53:35 +00004718 } else if (test_and_clear_bit(BTRFS_INODE_COPY_EVERYTHING,
4719 &BTRFS_I(inode)->runtime_flags) ||
Josef Bacik6cfab852013-11-12 16:25:58 -05004720 inode_only == LOG_INODE_EXISTS) {
Filipe Manana4f764e52015-02-23 19:53:35 +00004721 if (inode_only == LOG_INODE_ALL)
Josef Bacika95249b2012-10-11 16:17:34 -04004722 fast_search = true;
Filipe Manana4f764e52015-02-23 19:53:35 +00004723 max_key.type = BTRFS_XATTR_ITEM_KEY;
Josef Bacika95249b2012-10-11 16:17:34 -04004724 ret = drop_objectid_items(trans, log, path, ino,
4725 max_key.type);
Josef Bacik5dc562c2012-08-17 13:14:17 -04004726 } else {
Liu Bo183f37f2012-11-01 06:38:47 +00004727 if (inode_only == LOG_INODE_ALL)
4728 fast_search = true;
Josef Bacika95249b2012-10-11 16:17:34 -04004729 goto log_extents;
Josef Bacik5dc562c2012-08-17 13:14:17 -04004730 }
Josef Bacika95249b2012-10-11 16:17:34 -04004731
Chris Masone02119d2008-09-05 16:13:11 -04004732 }
Yan, Zheng4a500fd2010-05-16 10:49:59 -04004733 if (ret) {
4734 err = ret;
4735 goto out_unlock;
4736 }
Chris Masone02119d2008-09-05 16:13:11 -04004737
Chris Masond3977122009-01-05 21:25:51 -05004738 while (1) {
Chris Mason31ff1cd2008-09-11 16:17:57 -04004739 ins_nr = 0;
Filipe David Borba Manana6174d3c2013-10-01 16:13:42 +01004740 ret = btrfs_search_forward(root, &min_key,
Eric Sandeende78b512013-01-31 18:21:12 +00004741 path, trans->transid);
Liu Bofb770ae2016-07-05 12:10:14 -07004742 if (ret < 0) {
4743 err = ret;
4744 goto out_unlock;
4745 }
Chris Masone02119d2008-09-05 16:13:11 -04004746 if (ret != 0)
4747 break;
Chris Mason3a5f1d42008-09-11 15:53:37 -04004748again:
Chris Mason31ff1cd2008-09-11 16:17:57 -04004749 /* note, ins_nr might be > 0 here, cleanup outside the loop */
Li Zefan33345d012011-04-20 10:31:50 +08004750 if (min_key.objectid != ino)
Chris Masone02119d2008-09-05 16:13:11 -04004751 break;
4752 if (min_key.type > max_key.type)
4753 break;
Chris Mason31ff1cd2008-09-11 16:17:57 -04004754
Filipe Mananae4545de2015-06-17 12:49:23 +01004755 if (min_key.type == BTRFS_INODE_ITEM_KEY)
4756 need_log_inode_item = false;
4757
Filipe Manana56f23fd2016-03-30 23:37:21 +01004758 if ((min_key.type == BTRFS_INODE_REF_KEY ||
4759 min_key.type == BTRFS_INODE_EXTREF_KEY) &&
4760 BTRFS_I(inode)->generation == trans->transid) {
Filipe Manana44f714d2016-06-06 16:11:13 +01004761 u64 other_ino = 0;
4762
Filipe Manana56f23fd2016-03-30 23:37:21 +01004763 ret = btrfs_check_ref_name_override(path->nodes[0],
4764 path->slots[0],
Filipe Manana44f714d2016-06-06 16:11:13 +01004765 &min_key, inode,
4766 &other_ino);
Filipe Manana56f23fd2016-03-30 23:37:21 +01004767 if (ret < 0) {
4768 err = ret;
4769 goto out_unlock;
Filipe Manana28a23592016-08-23 21:13:51 +01004770 } else if (ret > 0 && ctx &&
4771 other_ino != btrfs_ino(ctx->inode)) {
Filipe Manana44f714d2016-06-06 16:11:13 +01004772 struct btrfs_key inode_key;
4773 struct inode *other_inode;
4774
4775 if (ins_nr > 0) {
4776 ins_nr++;
4777 } else {
4778 ins_nr = 1;
4779 ins_start_slot = path->slots[0];
4780 }
4781 ret = copy_items(trans, inode, dst_path, path,
4782 &last_extent, ins_start_slot,
4783 ins_nr, inode_only,
4784 logged_isize);
4785 if (ret < 0) {
4786 err = ret;
4787 goto out_unlock;
4788 }
4789 ins_nr = 0;
4790 btrfs_release_path(path);
4791 inode_key.objectid = other_ino;
4792 inode_key.type = BTRFS_INODE_ITEM_KEY;
4793 inode_key.offset = 0;
Jeff Mahoney0b246af2016-06-22 18:54:23 -04004794 other_inode = btrfs_iget(fs_info->sb,
Filipe Manana44f714d2016-06-06 16:11:13 +01004795 &inode_key, root,
4796 NULL);
4797 /*
4798 * If the other inode that had a conflicting dir
4799 * entry was deleted in the current transaction,
4800 * we don't need to do more work nor fallback to
4801 * a transaction commit.
4802 */
4803 if (IS_ERR(other_inode) &&
4804 PTR_ERR(other_inode) == -ENOENT) {
4805 goto next_key;
4806 } else if (IS_ERR(other_inode)) {
4807 err = PTR_ERR(other_inode);
4808 goto out_unlock;
4809 }
4810 /*
4811 * We are safe logging the other inode without
4812 * acquiring its i_mutex as long as we log with
4813 * the LOG_INODE_EXISTS mode. We're safe against
4814 * concurrent renames of the other inode as well
4815 * because during a rename we pin the log and
4816 * update the log with the new name before we
4817 * unpin it.
4818 */
4819 err = btrfs_log_inode(trans, root, other_inode,
4820 LOG_INODE_EXISTS,
4821 0, LLONG_MAX, ctx);
4822 iput(other_inode);
4823 if (err)
4824 goto out_unlock;
4825 else
4826 goto next_key;
Filipe Manana56f23fd2016-03-30 23:37:21 +01004827 }
4828 }
4829
Filipe Manana36283bf2015-06-20 00:44:51 +01004830 /* Skip xattrs, we log them later with btrfs_log_all_xattrs() */
4831 if (min_key.type == BTRFS_XATTR_ITEM_KEY) {
4832 if (ins_nr == 0)
4833 goto next_slot;
4834 ret = copy_items(trans, inode, dst_path, path,
4835 &last_extent, ins_start_slot,
4836 ins_nr, inode_only, logged_isize);
4837 if (ret < 0) {
4838 err = ret;
4839 goto out_unlock;
4840 }
4841 ins_nr = 0;
4842 if (ret) {
4843 btrfs_release_path(path);
4844 continue;
4845 }
4846 goto next_slot;
4847 }
4848
Chris Masone02119d2008-09-05 16:13:11 -04004849 src = path->nodes[0];
Chris Mason31ff1cd2008-09-11 16:17:57 -04004850 if (ins_nr && ins_start_slot + ins_nr == path->slots[0]) {
4851 ins_nr++;
4852 goto next_slot;
4853 } else if (!ins_nr) {
4854 ins_start_slot = path->slots[0];
4855 ins_nr = 1;
4856 goto next_slot;
Chris Masone02119d2008-09-05 16:13:11 -04004857 }
4858
Josef Bacik16e75492013-10-22 12:18:51 -04004859 ret = copy_items(trans, inode, dst_path, path, &last_extent,
Filipe Manana1a4bcf42015-02-13 12:30:56 +00004860 ins_start_slot, ins_nr, inode_only,
4861 logged_isize);
Josef Bacik16e75492013-10-22 12:18:51 -04004862 if (ret < 0) {
Yan, Zheng4a500fd2010-05-16 10:49:59 -04004863 err = ret;
4864 goto out_unlock;
Rasmus Villemoesa71db862014-06-20 21:51:43 +02004865 }
4866 if (ret) {
Josef Bacik16e75492013-10-22 12:18:51 -04004867 ins_nr = 0;
4868 btrfs_release_path(path);
4869 continue;
Yan, Zheng4a500fd2010-05-16 10:49:59 -04004870 }
Chris Mason31ff1cd2008-09-11 16:17:57 -04004871 ins_nr = 1;
4872 ins_start_slot = path->slots[0];
4873next_slot:
Chris Masone02119d2008-09-05 16:13:11 -04004874
Chris Mason3a5f1d42008-09-11 15:53:37 -04004875 nritems = btrfs_header_nritems(path->nodes[0]);
4876 path->slots[0]++;
4877 if (path->slots[0] < nritems) {
4878 btrfs_item_key_to_cpu(path->nodes[0], &min_key,
4879 path->slots[0]);
4880 goto again;
4881 }
Chris Mason31ff1cd2008-09-11 16:17:57 -04004882 if (ins_nr) {
Josef Bacik16e75492013-10-22 12:18:51 -04004883 ret = copy_items(trans, inode, dst_path, path,
4884 &last_extent, ins_start_slot,
Filipe Manana1a4bcf42015-02-13 12:30:56 +00004885 ins_nr, inode_only, logged_isize);
Josef Bacik16e75492013-10-22 12:18:51 -04004886 if (ret < 0) {
Yan, Zheng4a500fd2010-05-16 10:49:59 -04004887 err = ret;
4888 goto out_unlock;
4889 }
Josef Bacik16e75492013-10-22 12:18:51 -04004890 ret = 0;
Chris Mason31ff1cd2008-09-11 16:17:57 -04004891 ins_nr = 0;
4892 }
David Sterbab3b4aa72011-04-21 01:20:15 +02004893 btrfs_release_path(path);
Filipe Manana44f714d2016-06-06 16:11:13 +01004894next_key:
Filipe David Borba Manana3d41d702013-10-01 17:06:53 +01004895 if (min_key.offset < (u64)-1) {
Chris Masone02119d2008-09-05 16:13:11 -04004896 min_key.offset++;
Filipe David Borba Manana3d41d702013-10-01 17:06:53 +01004897 } else if (min_key.type < max_key.type) {
Chris Masone02119d2008-09-05 16:13:11 -04004898 min_key.type++;
Filipe David Borba Manana3d41d702013-10-01 17:06:53 +01004899 min_key.offset = 0;
4900 } else {
Chris Masone02119d2008-09-05 16:13:11 -04004901 break;
Filipe David Borba Manana3d41d702013-10-01 17:06:53 +01004902 }
Chris Masone02119d2008-09-05 16:13:11 -04004903 }
Chris Mason31ff1cd2008-09-11 16:17:57 -04004904 if (ins_nr) {
Josef Bacik16e75492013-10-22 12:18:51 -04004905 ret = copy_items(trans, inode, dst_path, path, &last_extent,
Filipe Manana1a4bcf42015-02-13 12:30:56 +00004906 ins_start_slot, ins_nr, inode_only,
4907 logged_isize);
Josef Bacik16e75492013-10-22 12:18:51 -04004908 if (ret < 0) {
Yan, Zheng4a500fd2010-05-16 10:49:59 -04004909 err = ret;
4910 goto out_unlock;
4911 }
Josef Bacik16e75492013-10-22 12:18:51 -04004912 ret = 0;
Chris Mason31ff1cd2008-09-11 16:17:57 -04004913 ins_nr = 0;
4914 }
Josef Bacik5dc562c2012-08-17 13:14:17 -04004915
Filipe Manana36283bf2015-06-20 00:44:51 +01004916 btrfs_release_path(path);
4917 btrfs_release_path(dst_path);
4918 err = btrfs_log_all_xattrs(trans, root, inode, path, dst_path);
4919 if (err)
4920 goto out_unlock;
Filipe Mananaa89ca6f2015-06-25 04:17:46 +01004921 if (max_key.type >= BTRFS_EXTENT_DATA_KEY && !fast_search) {
4922 btrfs_release_path(path);
4923 btrfs_release_path(dst_path);
4924 err = btrfs_log_trailing_hole(trans, root, inode, path);
4925 if (err)
4926 goto out_unlock;
4927 }
Josef Bacika95249b2012-10-11 16:17:34 -04004928log_extents:
Josef Bacikf3b15cc2013-07-22 12:54:30 -04004929 btrfs_release_path(path);
4930 btrfs_release_path(dst_path);
Filipe Mananae4545de2015-06-17 12:49:23 +01004931 if (need_log_inode_item) {
4932 err = log_inode_item(trans, log, dst_path, inode);
4933 if (err)
4934 goto out_unlock;
4935 }
Josef Bacik5dc562c2012-08-17 13:14:17 -04004936 if (fast_search) {
Miao Xie827463c2014-01-14 20:31:51 +08004937 ret = btrfs_log_changed_extents(trans, root, inode, dst_path,
Filipe Mananade0ee0e2016-01-21 10:17:54 +00004938 &logged_list, ctx, start, end);
Josef Bacik5dc562c2012-08-17 13:14:17 -04004939 if (ret) {
4940 err = ret;
4941 goto out_unlock;
4942 }
Josef Bacikd006a042013-11-12 20:54:09 -05004943 } else if (inode_only == LOG_INODE_ALL) {
Liu Bo06d3d222012-08-27 10:52:19 -06004944 struct extent_map *em, *n;
4945
Filipe Manana49dae1b2014-09-06 22:34:39 +01004946 write_lock(&em_tree->lock);
4947 /*
4948 * We can't just remove every em if we're called for a ranged
4949 * fsync - that is, one that doesn't cover the whole possible
4950 * file range (0 to LLONG_MAX). This is because we can have
4951 * em's that fall outside the range we're logging and therefore
4952 * their ordered operations haven't completed yet
4953 * (btrfs_finish_ordered_io() not invoked yet). This means we
4954 * didn't get their respective file extent item in the fs/subvol
4955 * tree yet, and need to let the next fast fsync (one which
4956 * consults the list of modified extent maps) find the em so
4957 * that it logs a matching file extent item and waits for the
4958 * respective ordered operation to complete (if it's still
4959 * running).
4960 *
4961 * Removing every em outside the range we're logging would make
4962 * the next fast fsync not log their matching file extent items,
4963 * therefore making us lose data after a log replay.
4964 */
4965 list_for_each_entry_safe(em, n, &em_tree->modified_extents,
4966 list) {
4967 const u64 mod_end = em->mod_start + em->mod_len - 1;
4968
4969 if (em->mod_start >= start && mod_end <= end)
4970 list_del_init(&em->list);
4971 }
4972 write_unlock(&em_tree->lock);
Josef Bacik5dc562c2012-08-17 13:14:17 -04004973 }
4974
Chris Mason9623f9a2008-09-11 17:42:42 -04004975 if (inode_only == LOG_INODE_ALL && S_ISDIR(inode->i_mode)) {
Filipe Manana2f2ff0e2015-03-20 17:19:46 +00004976 ret = log_directory_changes(trans, root, inode, path, dst_path,
4977 ctx);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04004978 if (ret) {
4979 err = ret;
4980 goto out_unlock;
4981 }
Chris Masone02119d2008-09-05 16:13:11 -04004982 }
Filipe Manana49dae1b2014-09-06 22:34:39 +01004983
Filipe Manana2f2ff0e2015-03-20 17:19:46 +00004984 spin_lock(&BTRFS_I(inode)->lock);
Filipe Manana125c4cf92014-09-11 21:22:14 +01004985 BTRFS_I(inode)->logged_trans = trans->transid;
4986 BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->last_sub_trans;
Filipe Manana2f2ff0e2015-03-20 17:19:46 +00004987 spin_unlock(&BTRFS_I(inode)->lock);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04004988out_unlock:
Miao Xie827463c2014-01-14 20:31:51 +08004989 if (unlikely(err))
4990 btrfs_put_logged_extents(&logged_list);
4991 else
4992 btrfs_submit_logged_extents(&logged_list, log);
Chris Masone02119d2008-09-05 16:13:11 -04004993 mutex_unlock(&BTRFS_I(inode)->log_mutex);
4994
4995 btrfs_free_path(path);
4996 btrfs_free_path(dst_path);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04004997 return err;
Chris Masone02119d2008-09-05 16:13:11 -04004998}
4999
Chris Mason12fcfd22009-03-24 10:24:20 -04005000/*
Filipe Manana2be63d52016-02-12 11:34:23 +00005001 * Check if we must fallback to a transaction commit when logging an inode.
5002 * This must be called after logging the inode and is used only in the context
5003 * when fsyncing an inode requires the need to log some other inode - in which
5004 * case we can't lock the i_mutex of each other inode we need to log as that
5005 * can lead to deadlocks with concurrent fsync against other inodes (as we can
5006 * log inodes up or down in the hierarchy) or rename operations for example. So
5007 * we take the log_mutex of the inode after we have logged it and then check for
5008 * its last_unlink_trans value - this is safe because any task setting
5009 * last_unlink_trans must take the log_mutex and it must do this before it does
5010 * the actual unlink operation, so if we do this check before a concurrent task
5011 * sets last_unlink_trans it means we've logged a consistent version/state of
5012 * all the inode items, otherwise we are not sure and must do a transaction
Nicholas D Steeves01327612016-05-19 21:18:45 -04005013 * commit (the concurrent task might have only updated last_unlink_trans before
Filipe Manana2be63d52016-02-12 11:34:23 +00005014 * we logged the inode or it might have also done the unlink).
5015 */
5016static bool btrfs_must_commit_transaction(struct btrfs_trans_handle *trans,
5017 struct inode *inode)
5018{
5019 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
5020 bool ret = false;
5021
5022 mutex_lock(&BTRFS_I(inode)->log_mutex);
5023 if (BTRFS_I(inode)->last_unlink_trans > fs_info->last_trans_committed) {
5024 /*
5025 * Make sure any commits to the log are forced to be full
5026 * commits.
5027 */
5028 btrfs_set_log_full_commit(fs_info, trans);
5029 ret = true;
5030 }
5031 mutex_unlock(&BTRFS_I(inode)->log_mutex);
5032
5033 return ret;
5034}
5035
5036/*
Chris Mason12fcfd22009-03-24 10:24:20 -04005037 * follow the dentry parent pointers up the chain and see if any
5038 * of the directories in it require a full commit before they can
5039 * be logged. Returns zero if nothing special needs to be done or 1 if
5040 * a full commit is required.
5041 */
5042static noinline int check_parent_dirs_for_sync(struct btrfs_trans_handle *trans,
5043 struct inode *inode,
5044 struct dentry *parent,
5045 struct super_block *sb,
5046 u64 last_committed)
Chris Masone02119d2008-09-05 16:13:11 -04005047{
Chris Mason12fcfd22009-03-24 10:24:20 -04005048 int ret = 0;
Josef Bacik6a912212010-11-20 09:48:00 +00005049 struct dentry *old_parent = NULL;
Josef Bacikde2b5302013-09-11 09:36:30 -04005050 struct inode *orig_inode = inode;
Chris Masone02119d2008-09-05 16:13:11 -04005051
Chris Masonaf4176b2009-03-24 10:24:31 -04005052 /*
5053 * for regular files, if its inode is already on disk, we don't
5054 * have to worry about the parents at all. This is because
5055 * we can use the last_unlink_trans field to record renames
5056 * and other fun in this file.
5057 */
5058 if (S_ISREG(inode->i_mode) &&
5059 BTRFS_I(inode)->generation <= last_committed &&
5060 BTRFS_I(inode)->last_unlink_trans <= last_committed)
5061 goto out;
5062
Chris Mason12fcfd22009-03-24 10:24:20 -04005063 if (!S_ISDIR(inode->i_mode)) {
Al Virofc640052016-04-10 01:33:30 -04005064 if (!parent || d_really_is_negative(parent) || sb != parent->d_sb)
Chris Mason12fcfd22009-03-24 10:24:20 -04005065 goto out;
David Howells2b0143b2015-03-17 22:25:59 +00005066 inode = d_inode(parent);
Chris Mason12fcfd22009-03-24 10:24:20 -04005067 }
5068
5069 while (1) {
Josef Bacikde2b5302013-09-11 09:36:30 -04005070 /*
5071 * If we are logging a directory then we start with our inode,
Nicholas D Steeves01327612016-05-19 21:18:45 -04005072 * not our parent's inode, so we need to skip setting the
Josef Bacikde2b5302013-09-11 09:36:30 -04005073 * logged_trans so that further down in the log code we don't
5074 * think this inode has already been logged.
5075 */
5076 if (inode != orig_inode)
5077 BTRFS_I(inode)->logged_trans = trans->transid;
Chris Mason12fcfd22009-03-24 10:24:20 -04005078 smp_mb();
5079
Filipe Manana2be63d52016-02-12 11:34:23 +00005080 if (btrfs_must_commit_transaction(trans, inode)) {
Chris Mason12fcfd22009-03-24 10:24:20 -04005081 ret = 1;
5082 break;
5083 }
5084
Al Virofc640052016-04-10 01:33:30 -04005085 if (!parent || d_really_is_negative(parent) || sb != parent->d_sb)
Chris Mason12fcfd22009-03-24 10:24:20 -04005086 break;
5087
Filipe Manana44f714d2016-06-06 16:11:13 +01005088 if (IS_ROOT(parent)) {
5089 inode = d_inode(parent);
5090 if (btrfs_must_commit_transaction(trans, inode))
5091 ret = 1;
Chris Mason12fcfd22009-03-24 10:24:20 -04005092 break;
Filipe Manana44f714d2016-06-06 16:11:13 +01005093 }
Chris Mason12fcfd22009-03-24 10:24:20 -04005094
Josef Bacik6a912212010-11-20 09:48:00 +00005095 parent = dget_parent(parent);
5096 dput(old_parent);
5097 old_parent = parent;
David Howells2b0143b2015-03-17 22:25:59 +00005098 inode = d_inode(parent);
Chris Mason12fcfd22009-03-24 10:24:20 -04005099
5100 }
Josef Bacik6a912212010-11-20 09:48:00 +00005101 dput(old_parent);
Chris Mason12fcfd22009-03-24 10:24:20 -04005102out:
Chris Masone02119d2008-09-05 16:13:11 -04005103 return ret;
5104}
5105
Filipe Manana2f2ff0e2015-03-20 17:19:46 +00005106struct btrfs_dir_list {
5107 u64 ino;
5108 struct list_head list;
5109};
5110
5111/*
5112 * Log the inodes of the new dentries of a directory. See log_dir_items() for
5113 * details about the why it is needed.
5114 * This is a recursive operation - if an existing dentry corresponds to a
5115 * directory, that directory's new entries are logged too (same behaviour as
5116 * ext3/4, xfs, f2fs, reiserfs, nilfs2). Note that when logging the inodes
5117 * the dentries point to we do not lock their i_mutex, otherwise lockdep
5118 * complains about the following circular lock dependency / possible deadlock:
5119 *
5120 * CPU0 CPU1
5121 * ---- ----
5122 * lock(&type->i_mutex_dir_key#3/2);
5123 * lock(sb_internal#2);
5124 * lock(&type->i_mutex_dir_key#3/2);
5125 * lock(&sb->s_type->i_mutex_key#14);
5126 *
5127 * Where sb_internal is the lock (a counter that works as a lock) acquired by
5128 * sb_start_intwrite() in btrfs_start_transaction().
5129 * Not locking i_mutex of the inodes is still safe because:
5130 *
5131 * 1) For regular files we log with a mode of LOG_INODE_EXISTS. It's possible
5132 * that while logging the inode new references (names) are added or removed
5133 * from the inode, leaving the logged inode item with a link count that does
5134 * not match the number of logged inode reference items. This is fine because
5135 * at log replay time we compute the real number of links and correct the
5136 * link count in the inode item (see replay_one_buffer() and
5137 * link_to_fixup_dir());
5138 *
5139 * 2) For directories we log with a mode of LOG_INODE_ALL. It's possible that
5140 * while logging the inode's items new items with keys BTRFS_DIR_ITEM_KEY and
5141 * BTRFS_DIR_INDEX_KEY are added to fs/subvol tree and the logged inode item
5142 * has a size that doesn't match the sum of the lengths of all the logged
5143 * names. This does not result in a problem because if a dir_item key is
5144 * logged but its matching dir_index key is not logged, at log replay time we
5145 * don't use it to replay the respective name (see replay_one_name()). On the
5146 * other hand if only the dir_index key ends up being logged, the respective
5147 * name is added to the fs/subvol tree with both the dir_item and dir_index
5148 * keys created (see replay_one_name()).
5149 * The directory's inode item with a wrong i_size is not a problem as well,
5150 * since we don't use it at log replay time to set the i_size in the inode
5151 * item of the fs/subvol tree (see overwrite_item()).
5152 */
5153static int log_new_dir_dentries(struct btrfs_trans_handle *trans,
5154 struct btrfs_root *root,
5155 struct inode *start_inode,
5156 struct btrfs_log_ctx *ctx)
5157{
Jeff Mahoney0b246af2016-06-22 18:54:23 -04005158 struct btrfs_fs_info *fs_info = root->fs_info;
Filipe Manana2f2ff0e2015-03-20 17:19:46 +00005159 struct btrfs_root *log = root->log_root;
5160 struct btrfs_path *path;
5161 LIST_HEAD(dir_list);
5162 struct btrfs_dir_list *dir_elem;
5163 int ret = 0;
5164
5165 path = btrfs_alloc_path();
5166 if (!path)
5167 return -ENOMEM;
5168
5169 dir_elem = kmalloc(sizeof(*dir_elem), GFP_NOFS);
5170 if (!dir_elem) {
5171 btrfs_free_path(path);
5172 return -ENOMEM;
5173 }
5174 dir_elem->ino = btrfs_ino(start_inode);
5175 list_add_tail(&dir_elem->list, &dir_list);
5176
5177 while (!list_empty(&dir_list)) {
5178 struct extent_buffer *leaf;
5179 struct btrfs_key min_key;
5180 int nritems;
5181 int i;
5182
5183 dir_elem = list_first_entry(&dir_list, struct btrfs_dir_list,
5184 list);
5185 if (ret)
5186 goto next_dir_inode;
5187
5188 min_key.objectid = dir_elem->ino;
5189 min_key.type = BTRFS_DIR_ITEM_KEY;
5190 min_key.offset = 0;
5191again:
5192 btrfs_release_path(path);
5193 ret = btrfs_search_forward(log, &min_key, path, trans->transid);
5194 if (ret < 0) {
5195 goto next_dir_inode;
5196 } else if (ret > 0) {
5197 ret = 0;
5198 goto next_dir_inode;
5199 }
5200
5201process_leaf:
5202 leaf = path->nodes[0];
5203 nritems = btrfs_header_nritems(leaf);
5204 for (i = path->slots[0]; i < nritems; i++) {
5205 struct btrfs_dir_item *di;
5206 struct btrfs_key di_key;
5207 struct inode *di_inode;
5208 struct btrfs_dir_list *new_dir_elem;
5209 int log_mode = LOG_INODE_EXISTS;
5210 int type;
5211
5212 btrfs_item_key_to_cpu(leaf, &min_key, i);
5213 if (min_key.objectid != dir_elem->ino ||
5214 min_key.type != BTRFS_DIR_ITEM_KEY)
5215 goto next_dir_inode;
5216
5217 di = btrfs_item_ptr(leaf, i, struct btrfs_dir_item);
5218 type = btrfs_dir_type(leaf, di);
5219 if (btrfs_dir_transid(leaf, di) < trans->transid &&
5220 type != BTRFS_FT_DIR)
5221 continue;
5222 btrfs_dir_item_key_to_cpu(leaf, di, &di_key);
5223 if (di_key.type == BTRFS_ROOT_ITEM_KEY)
5224 continue;
5225
Robbie Koec125cf2016-10-28 10:48:26 +08005226 btrfs_release_path(path);
Jeff Mahoney0b246af2016-06-22 18:54:23 -04005227 di_inode = btrfs_iget(fs_info->sb, &di_key, root, NULL);
Filipe Manana2f2ff0e2015-03-20 17:19:46 +00005228 if (IS_ERR(di_inode)) {
5229 ret = PTR_ERR(di_inode);
5230 goto next_dir_inode;
5231 }
5232
5233 if (btrfs_inode_in_log(di_inode, trans->transid)) {
5234 iput(di_inode);
Robbie Koec125cf2016-10-28 10:48:26 +08005235 break;
Filipe Manana2f2ff0e2015-03-20 17:19:46 +00005236 }
5237
5238 ctx->log_new_dentries = false;
Filipe Manana3f9749f2016-04-25 04:45:02 +01005239 if (type == BTRFS_FT_DIR || type == BTRFS_FT_SYMLINK)
Filipe Manana2f2ff0e2015-03-20 17:19:46 +00005240 log_mode = LOG_INODE_ALL;
Filipe Manana2f2ff0e2015-03-20 17:19:46 +00005241 ret = btrfs_log_inode(trans, root, di_inode,
5242 log_mode, 0, LLONG_MAX, ctx);
Filipe Manana2be63d52016-02-12 11:34:23 +00005243 if (!ret &&
5244 btrfs_must_commit_transaction(trans, di_inode))
5245 ret = 1;
Filipe Manana2f2ff0e2015-03-20 17:19:46 +00005246 iput(di_inode);
5247 if (ret)
5248 goto next_dir_inode;
5249 if (ctx->log_new_dentries) {
5250 new_dir_elem = kmalloc(sizeof(*new_dir_elem),
5251 GFP_NOFS);
5252 if (!new_dir_elem) {
5253 ret = -ENOMEM;
5254 goto next_dir_inode;
5255 }
5256 new_dir_elem->ino = di_key.objectid;
5257 list_add_tail(&new_dir_elem->list, &dir_list);
5258 }
5259 break;
5260 }
5261 if (i == nritems) {
5262 ret = btrfs_next_leaf(log, path);
5263 if (ret < 0) {
5264 goto next_dir_inode;
5265 } else if (ret > 0) {
5266 ret = 0;
5267 goto next_dir_inode;
5268 }
5269 goto process_leaf;
5270 }
5271 if (min_key.offset < (u64)-1) {
5272 min_key.offset++;
5273 goto again;
5274 }
5275next_dir_inode:
5276 list_del(&dir_elem->list);
5277 kfree(dir_elem);
5278 }
5279
5280 btrfs_free_path(path);
5281 return ret;
5282}
5283
Filipe Manana18aa0922015-08-05 16:49:08 +01005284static int btrfs_log_all_parents(struct btrfs_trans_handle *trans,
5285 struct inode *inode,
5286 struct btrfs_log_ctx *ctx)
5287{
Jeff Mahoney0b246af2016-06-22 18:54:23 -04005288 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
Filipe Manana18aa0922015-08-05 16:49:08 +01005289 int ret;
5290 struct btrfs_path *path;
5291 struct btrfs_key key;
5292 struct btrfs_root *root = BTRFS_I(inode)->root;
5293 const u64 ino = btrfs_ino(inode);
5294
5295 path = btrfs_alloc_path();
5296 if (!path)
5297 return -ENOMEM;
5298 path->skip_locking = 1;
5299 path->search_commit_root = 1;
5300
5301 key.objectid = ino;
5302 key.type = BTRFS_INODE_REF_KEY;
5303 key.offset = 0;
5304 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5305 if (ret < 0)
5306 goto out;
5307
5308 while (true) {
5309 struct extent_buffer *leaf = path->nodes[0];
5310 int slot = path->slots[0];
5311 u32 cur_offset = 0;
5312 u32 item_size;
5313 unsigned long ptr;
5314
5315 if (slot >= btrfs_header_nritems(leaf)) {
5316 ret = btrfs_next_leaf(root, path);
5317 if (ret < 0)
5318 goto out;
5319 else if (ret > 0)
5320 break;
5321 continue;
5322 }
5323
5324 btrfs_item_key_to_cpu(leaf, &key, slot);
5325 /* BTRFS_INODE_EXTREF_KEY is BTRFS_INODE_REF_KEY + 1 */
5326 if (key.objectid != ino || key.type > BTRFS_INODE_EXTREF_KEY)
5327 break;
5328
5329 item_size = btrfs_item_size_nr(leaf, slot);
5330 ptr = btrfs_item_ptr_offset(leaf, slot);
5331 while (cur_offset < item_size) {
5332 struct btrfs_key inode_key;
5333 struct inode *dir_inode;
5334
5335 inode_key.type = BTRFS_INODE_ITEM_KEY;
5336 inode_key.offset = 0;
5337
5338 if (key.type == BTRFS_INODE_EXTREF_KEY) {
5339 struct btrfs_inode_extref *extref;
5340
5341 extref = (struct btrfs_inode_extref *)
5342 (ptr + cur_offset);
5343 inode_key.objectid = btrfs_inode_extref_parent(
5344 leaf, extref);
5345 cur_offset += sizeof(*extref);
5346 cur_offset += btrfs_inode_extref_name_len(leaf,
5347 extref);
5348 } else {
5349 inode_key.objectid = key.offset;
5350 cur_offset = item_size;
5351 }
5352
Jeff Mahoney0b246af2016-06-22 18:54:23 -04005353 dir_inode = btrfs_iget(fs_info->sb, &inode_key,
Filipe Manana18aa0922015-08-05 16:49:08 +01005354 root, NULL);
5355 /* If parent inode was deleted, skip it. */
5356 if (IS_ERR(dir_inode))
5357 continue;
5358
Filipe Manana657ed1a2016-04-06 17:11:56 +01005359 if (ctx)
5360 ctx->log_new_dentries = false;
Filipe Manana18aa0922015-08-05 16:49:08 +01005361 ret = btrfs_log_inode(trans, root, dir_inode,
5362 LOG_INODE_ALL, 0, LLONG_MAX, ctx);
Filipe Manana2be63d52016-02-12 11:34:23 +00005363 if (!ret &&
5364 btrfs_must_commit_transaction(trans, dir_inode))
5365 ret = 1;
Filipe Manana657ed1a2016-04-06 17:11:56 +01005366 if (!ret && ctx && ctx->log_new_dentries)
5367 ret = log_new_dir_dentries(trans, root,
5368 dir_inode, ctx);
Filipe Manana18aa0922015-08-05 16:49:08 +01005369 iput(dir_inode);
5370 if (ret)
5371 goto out;
5372 }
5373 path->slots[0]++;
5374 }
5375 ret = 0;
5376out:
5377 btrfs_free_path(path);
5378 return ret;
5379}
5380
Chris Masone02119d2008-09-05 16:13:11 -04005381/*
5382 * helper function around btrfs_log_inode to make sure newly created
5383 * parent directories also end up in the log. A minimal inode and backref
5384 * only logging is done of any parent directories that are older than
5385 * the last committed transaction
5386 */
Eric Sandeen48a3b632013-04-25 20:41:01 +00005387static int btrfs_log_inode_parent(struct btrfs_trans_handle *trans,
5388 struct btrfs_root *root, struct inode *inode,
Filipe Manana49dae1b2014-09-06 22:34:39 +01005389 struct dentry *parent,
5390 const loff_t start,
5391 const loff_t end,
5392 int exists_only,
Miao Xie8b050d32014-02-20 18:08:58 +08005393 struct btrfs_log_ctx *ctx)
Chris Masone02119d2008-09-05 16:13:11 -04005394{
Jeff Mahoney0b246af2016-06-22 18:54:23 -04005395 struct btrfs_fs_info *fs_info = root->fs_info;
Chris Mason12fcfd22009-03-24 10:24:20 -04005396 int inode_only = exists_only ? LOG_INODE_EXISTS : LOG_INODE_ALL;
Chris Masone02119d2008-09-05 16:13:11 -04005397 struct super_block *sb;
Josef Bacik6a912212010-11-20 09:48:00 +00005398 struct dentry *old_parent = NULL;
Chris Mason12fcfd22009-03-24 10:24:20 -04005399 int ret = 0;
Jeff Mahoney0b246af2016-06-22 18:54:23 -04005400 u64 last_committed = fs_info->last_trans_committed;
Filipe Manana2f2ff0e2015-03-20 17:19:46 +00005401 bool log_dentries = false;
5402 struct inode *orig_inode = inode;
Chris Mason12fcfd22009-03-24 10:24:20 -04005403
5404 sb = inode->i_sb;
5405
Jeff Mahoney0b246af2016-06-22 18:54:23 -04005406 if (btrfs_test_opt(fs_info, NOTREELOG)) {
Sage Weil3a5e1402009-04-02 16:49:40 -04005407 ret = 1;
5408 goto end_no_trans;
5409 }
5410
Miao Xie995946d2014-04-02 19:51:06 +08005411 /*
5412 * The prev transaction commit doesn't complete, we need do
5413 * full commit by ourselves.
5414 */
Jeff Mahoney0b246af2016-06-22 18:54:23 -04005415 if (fs_info->last_trans_log_full_commit >
5416 fs_info->last_trans_committed) {
Chris Mason12fcfd22009-03-24 10:24:20 -04005417 ret = 1;
5418 goto end_no_trans;
5419 }
5420
Yan, Zheng76dda932009-09-21 16:00:26 -04005421 if (root != BTRFS_I(inode)->root ||
5422 btrfs_root_refs(&root->root_item) == 0) {
5423 ret = 1;
5424 goto end_no_trans;
5425 }
5426
Chris Mason12fcfd22009-03-24 10:24:20 -04005427 ret = check_parent_dirs_for_sync(trans, inode, parent,
5428 sb, last_committed);
5429 if (ret)
5430 goto end_no_trans;
Chris Masone02119d2008-09-05 16:13:11 -04005431
Josef Bacik22ee6982012-05-29 16:57:49 -04005432 if (btrfs_inode_in_log(inode, trans->transid)) {
Chris Mason257c62e2009-10-13 13:21:08 -04005433 ret = BTRFS_NO_LOG_SYNC;
5434 goto end_no_trans;
5435 }
5436
Miao Xie8b050d32014-02-20 18:08:58 +08005437 ret = start_log_trans(trans, root, ctx);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04005438 if (ret)
Miao Xiee87ac132014-02-20 18:08:53 +08005439 goto end_no_trans;
Chris Mason12fcfd22009-03-24 10:24:20 -04005440
Filipe Manana8407f552014-09-05 15:14:39 +01005441 ret = btrfs_log_inode(trans, root, inode, inode_only, start, end, ctx);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04005442 if (ret)
5443 goto end_trans;
Chris Mason12fcfd22009-03-24 10:24:20 -04005444
Chris Masonaf4176b2009-03-24 10:24:31 -04005445 /*
5446 * for regular files, if its inode is already on disk, we don't
5447 * have to worry about the parents at all. This is because
5448 * we can use the last_unlink_trans field to record renames
5449 * and other fun in this file.
5450 */
5451 if (S_ISREG(inode->i_mode) &&
5452 BTRFS_I(inode)->generation <= last_committed &&
Yan, Zheng4a500fd2010-05-16 10:49:59 -04005453 BTRFS_I(inode)->last_unlink_trans <= last_committed) {
5454 ret = 0;
5455 goto end_trans;
5456 }
Chris Masonaf4176b2009-03-24 10:24:31 -04005457
Filipe Manana2f2ff0e2015-03-20 17:19:46 +00005458 if (S_ISDIR(inode->i_mode) && ctx && ctx->log_new_dentries)
5459 log_dentries = true;
5460
Filipe Manana18aa0922015-08-05 16:49:08 +01005461 /*
Nicholas D Steeves01327612016-05-19 21:18:45 -04005462 * On unlink we must make sure all our current and old parent directory
Filipe Manana18aa0922015-08-05 16:49:08 +01005463 * inodes are fully logged. This is to prevent leaving dangling
5464 * directory index entries in directories that were our parents but are
5465 * not anymore. Not doing this results in old parent directory being
5466 * impossible to delete after log replay (rmdir will always fail with
5467 * error -ENOTEMPTY).
5468 *
5469 * Example 1:
5470 *
5471 * mkdir testdir
5472 * touch testdir/foo
5473 * ln testdir/foo testdir/bar
5474 * sync
5475 * unlink testdir/bar
5476 * xfs_io -c fsync testdir/foo
5477 * <power failure>
5478 * mount fs, triggers log replay
5479 *
5480 * If we don't log the parent directory (testdir), after log replay the
5481 * directory still has an entry pointing to the file inode using the bar
5482 * name, but a matching BTRFS_INODE_[REF|EXTREF]_KEY does not exist and
5483 * the file inode has a link count of 1.
5484 *
5485 * Example 2:
5486 *
5487 * mkdir testdir
5488 * touch foo
5489 * ln foo testdir/foo2
5490 * ln foo testdir/foo3
5491 * sync
5492 * unlink testdir/foo3
5493 * xfs_io -c fsync foo
5494 * <power failure>
5495 * mount fs, triggers log replay
5496 *
5497 * Similar as the first example, after log replay the parent directory
5498 * testdir still has an entry pointing to the inode file with name foo3
5499 * but the file inode does not have a matching BTRFS_INODE_REF_KEY item
5500 * and has a link count of 2.
5501 */
5502 if (BTRFS_I(inode)->last_unlink_trans > last_committed) {
5503 ret = btrfs_log_all_parents(trans, orig_inode, ctx);
5504 if (ret)
5505 goto end_trans;
5506 }
5507
Chris Masond3977122009-01-05 21:25:51 -05005508 while (1) {
Al Virofc640052016-04-10 01:33:30 -04005509 if (!parent || d_really_is_negative(parent) || sb != parent->d_sb)
Chris Masone02119d2008-09-05 16:13:11 -04005510 break;
5511
David Howells2b0143b2015-03-17 22:25:59 +00005512 inode = d_inode(parent);
Yan, Zheng76dda932009-09-21 16:00:26 -04005513 if (root != BTRFS_I(inode)->root)
5514 break;
5515
Filipe Manana18aa0922015-08-05 16:49:08 +01005516 if (BTRFS_I(inode)->generation > last_committed) {
5517 ret = btrfs_log_inode(trans, root, inode,
5518 LOG_INODE_EXISTS,
Filipe Manana8407f552014-09-05 15:14:39 +01005519 0, LLONG_MAX, ctx);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04005520 if (ret)
5521 goto end_trans;
Chris Mason12fcfd22009-03-24 10:24:20 -04005522 }
Yan, Zheng76dda932009-09-21 16:00:26 -04005523 if (IS_ROOT(parent))
Chris Masone02119d2008-09-05 16:13:11 -04005524 break;
Chris Mason12fcfd22009-03-24 10:24:20 -04005525
Josef Bacik6a912212010-11-20 09:48:00 +00005526 parent = dget_parent(parent);
5527 dput(old_parent);
5528 old_parent = parent;
Chris Masone02119d2008-09-05 16:13:11 -04005529 }
Filipe Manana2f2ff0e2015-03-20 17:19:46 +00005530 if (log_dentries)
5531 ret = log_new_dir_dentries(trans, root, orig_inode, ctx);
5532 else
5533 ret = 0;
Yan, Zheng4a500fd2010-05-16 10:49:59 -04005534end_trans:
Josef Bacik6a912212010-11-20 09:48:00 +00005535 dput(old_parent);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04005536 if (ret < 0) {
Jeff Mahoney0b246af2016-06-22 18:54:23 -04005537 btrfs_set_log_full_commit(fs_info, trans);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04005538 ret = 1;
5539 }
Miao Xie8b050d32014-02-20 18:08:58 +08005540
5541 if (ret)
5542 btrfs_remove_log_ctx(root, ctx);
Chris Mason12fcfd22009-03-24 10:24:20 -04005543 btrfs_end_log_trans(root);
5544end_no_trans:
5545 return ret;
Chris Masone02119d2008-09-05 16:13:11 -04005546}
5547
5548/*
5549 * it is not safe to log dentry if the chunk root has added new
5550 * chunks. This returns 0 if the dentry was logged, and 1 otherwise.
5551 * If this returns 1, you must commit the transaction to safely get your
5552 * data on disk.
5553 */
5554int btrfs_log_dentry_safe(struct btrfs_trans_handle *trans,
Miao Xie8b050d32014-02-20 18:08:58 +08005555 struct btrfs_root *root, struct dentry *dentry,
Filipe Manana49dae1b2014-09-06 22:34:39 +01005556 const loff_t start,
5557 const loff_t end,
Miao Xie8b050d32014-02-20 18:08:58 +08005558 struct btrfs_log_ctx *ctx)
Chris Masone02119d2008-09-05 16:13:11 -04005559{
Josef Bacik6a912212010-11-20 09:48:00 +00005560 struct dentry *parent = dget_parent(dentry);
5561 int ret;
5562
David Howells2b0143b2015-03-17 22:25:59 +00005563 ret = btrfs_log_inode_parent(trans, root, d_inode(dentry), parent,
Filipe Manana49dae1b2014-09-06 22:34:39 +01005564 start, end, 0, ctx);
Josef Bacik6a912212010-11-20 09:48:00 +00005565 dput(parent);
5566
5567 return ret;
Chris Masone02119d2008-09-05 16:13:11 -04005568}
5569
5570/*
5571 * should be called during mount to recover any replay any log trees
5572 * from the FS
5573 */
5574int btrfs_recover_log_trees(struct btrfs_root *log_root_tree)
5575{
5576 int ret;
5577 struct btrfs_path *path;
5578 struct btrfs_trans_handle *trans;
5579 struct btrfs_key key;
5580 struct btrfs_key found_key;
5581 struct btrfs_key tmp_key;
5582 struct btrfs_root *log;
5583 struct btrfs_fs_info *fs_info = log_root_tree->fs_info;
5584 struct walk_control wc = {
5585 .process_func = process_one_buffer,
5586 .stage = 0,
5587 };
5588
Chris Masone02119d2008-09-05 16:13:11 -04005589 path = btrfs_alloc_path();
Tsutomu Itohdb5b4932011-03-23 08:14:16 +00005590 if (!path)
5591 return -ENOMEM;
5592
Josef Bacikafcdd122016-09-02 15:40:02 -04005593 set_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags);
Chris Masone02119d2008-09-05 16:13:11 -04005594
Yan, Zheng4a500fd2010-05-16 10:49:59 -04005595 trans = btrfs_start_transaction(fs_info->tree_root, 0);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01005596 if (IS_ERR(trans)) {
5597 ret = PTR_ERR(trans);
5598 goto error;
5599 }
Chris Masone02119d2008-09-05 16:13:11 -04005600
5601 wc.trans = trans;
5602 wc.pin = 1;
5603
Tsutomu Itohdb5b4932011-03-23 08:14:16 +00005604 ret = walk_log_tree(trans, log_root_tree, &wc);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01005605 if (ret) {
Jeff Mahoney5d163e02016-09-20 10:05:00 -04005606 btrfs_handle_fs_error(fs_info, ret,
5607 "Failed to pin buffers while recovering log root tree.");
Jeff Mahoney79787ea2012-03-12 16:03:00 +01005608 goto error;
5609 }
Chris Masone02119d2008-09-05 16:13:11 -04005610
5611again:
5612 key.objectid = BTRFS_TREE_LOG_OBJECTID;
5613 key.offset = (u64)-1;
David Sterba962a2982014-06-04 18:41:45 +02005614 key.type = BTRFS_ROOT_ITEM_KEY;
Chris Masone02119d2008-09-05 16:13:11 -04005615
Chris Masond3977122009-01-05 21:25:51 -05005616 while (1) {
Chris Masone02119d2008-09-05 16:13:11 -04005617 ret = btrfs_search_slot(NULL, log_root_tree, &key, path, 0, 0);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01005618
5619 if (ret < 0) {
Anand Jain34d97002016-03-16 16:43:06 +08005620 btrfs_handle_fs_error(fs_info, ret,
Jeff Mahoney79787ea2012-03-12 16:03:00 +01005621 "Couldn't find tree log root.");
5622 goto error;
5623 }
Chris Masone02119d2008-09-05 16:13:11 -04005624 if (ret > 0) {
5625 if (path->slots[0] == 0)
5626 break;
5627 path->slots[0]--;
5628 }
5629 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
5630 path->slots[0]);
David Sterbab3b4aa72011-04-21 01:20:15 +02005631 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04005632 if (found_key.objectid != BTRFS_TREE_LOG_OBJECTID)
5633 break;
5634
Miao Xiecb517ea2013-05-15 07:48:19 +00005635 log = btrfs_read_fs_root(log_root_tree, &found_key);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01005636 if (IS_ERR(log)) {
5637 ret = PTR_ERR(log);
Anand Jain34d97002016-03-16 16:43:06 +08005638 btrfs_handle_fs_error(fs_info, ret,
Jeff Mahoney79787ea2012-03-12 16:03:00 +01005639 "Couldn't read tree log root.");
5640 goto error;
5641 }
Chris Masone02119d2008-09-05 16:13:11 -04005642
5643 tmp_key.objectid = found_key.offset;
5644 tmp_key.type = BTRFS_ROOT_ITEM_KEY;
5645 tmp_key.offset = (u64)-1;
5646
5647 wc.replay_dest = btrfs_read_fs_root_no_name(fs_info, &tmp_key);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01005648 if (IS_ERR(wc.replay_dest)) {
5649 ret = PTR_ERR(wc.replay_dest);
Josef Bacikb50c6e22013-04-25 15:55:30 -04005650 free_extent_buffer(log->node);
5651 free_extent_buffer(log->commit_root);
5652 kfree(log);
Jeff Mahoney5d163e02016-09-20 10:05:00 -04005653 btrfs_handle_fs_error(fs_info, ret,
5654 "Couldn't read target root for tree log recovery.");
Jeff Mahoney79787ea2012-03-12 16:03:00 +01005655 goto error;
5656 }
Chris Masone02119d2008-09-05 16:13:11 -04005657
Yan Zheng07d400a2009-01-06 11:42:00 -05005658 wc.replay_dest->log_root = log;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005659 btrfs_record_root_in_trans(trans, wc.replay_dest);
Chris Masone02119d2008-09-05 16:13:11 -04005660 ret = walk_log_tree(trans, log, &wc);
Chris Masone02119d2008-09-05 16:13:11 -04005661
Josef Bacikb50c6e22013-04-25 15:55:30 -04005662 if (!ret && wc.stage == LOG_WALK_REPLAY_ALL) {
Chris Masone02119d2008-09-05 16:13:11 -04005663 ret = fixup_inode_link_counts(trans, wc.replay_dest,
5664 path);
Chris Masone02119d2008-09-05 16:13:11 -04005665 }
Chris Masone02119d2008-09-05 16:13:11 -04005666
5667 key.offset = found_key.offset - 1;
Yan Zheng07d400a2009-01-06 11:42:00 -05005668 wc.replay_dest->log_root = NULL;
Chris Masone02119d2008-09-05 16:13:11 -04005669 free_extent_buffer(log->node);
Chris Masonb263c2c2009-06-11 11:24:47 -04005670 free_extent_buffer(log->commit_root);
Chris Masone02119d2008-09-05 16:13:11 -04005671 kfree(log);
5672
Josef Bacikb50c6e22013-04-25 15:55:30 -04005673 if (ret)
5674 goto error;
5675
Chris Masone02119d2008-09-05 16:13:11 -04005676 if (found_key.offset == 0)
5677 break;
5678 }
David Sterbab3b4aa72011-04-21 01:20:15 +02005679 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04005680
5681 /* step one is to pin it all, step two is to replay just inodes */
5682 if (wc.pin) {
5683 wc.pin = 0;
5684 wc.process_func = replay_one_buffer;
5685 wc.stage = LOG_WALK_REPLAY_INODES;
5686 goto again;
5687 }
5688 /* step three is to replay everything */
5689 if (wc.stage < LOG_WALK_REPLAY_ALL) {
5690 wc.stage++;
5691 goto again;
5692 }
5693
5694 btrfs_free_path(path);
5695
Josef Bacikabefa552013-04-24 16:40:05 -04005696 /* step 4: commit the transaction, which also unpins the blocks */
Jeff Mahoney3a45bb22016-09-09 21:39:03 -04005697 ret = btrfs_commit_transaction(trans);
Josef Bacikabefa552013-04-24 16:40:05 -04005698 if (ret)
5699 return ret;
5700
Chris Masone02119d2008-09-05 16:13:11 -04005701 free_extent_buffer(log_root_tree->node);
5702 log_root_tree->log_root = NULL;
Josef Bacikafcdd122016-09-02 15:40:02 -04005703 clear_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags);
Chris Masone02119d2008-09-05 16:13:11 -04005704 kfree(log_root_tree);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01005705
Josef Bacikabefa552013-04-24 16:40:05 -04005706 return 0;
Jeff Mahoney79787ea2012-03-12 16:03:00 +01005707error:
Josef Bacikb50c6e22013-04-25 15:55:30 -04005708 if (wc.trans)
Jeff Mahoney3a45bb22016-09-09 21:39:03 -04005709 btrfs_end_transaction(wc.trans);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01005710 btrfs_free_path(path);
5711 return ret;
Chris Masone02119d2008-09-05 16:13:11 -04005712}
Chris Mason12fcfd22009-03-24 10:24:20 -04005713
5714/*
5715 * there are some corner cases where we want to force a full
5716 * commit instead of allowing a directory to be logged.
5717 *
5718 * They revolve around files there were unlinked from the directory, and
5719 * this function updates the parent directory so that a full commit is
5720 * properly done if it is fsync'd later after the unlinks are done.
Filipe Manana2be63d52016-02-12 11:34:23 +00005721 *
5722 * Must be called before the unlink operations (updates to the subvolume tree,
5723 * inodes, etc) are done.
Chris Mason12fcfd22009-03-24 10:24:20 -04005724 */
5725void btrfs_record_unlink_dir(struct btrfs_trans_handle *trans,
5726 struct inode *dir, struct inode *inode,
5727 int for_rename)
5728{
5729 /*
Chris Masonaf4176b2009-03-24 10:24:31 -04005730 * when we're logging a file, if it hasn't been renamed
5731 * or unlinked, and its inode is fully committed on disk,
5732 * we don't have to worry about walking up the directory chain
5733 * to log its parents.
5734 *
5735 * So, we use the last_unlink_trans field to put this transid
5736 * into the file. When the file is logged we check it and
5737 * don't log the parents if the file is fully on disk.
5738 */
Filipe Manana657ed1a2016-04-06 17:11:56 +01005739 mutex_lock(&BTRFS_I(inode)->log_mutex);
5740 BTRFS_I(inode)->last_unlink_trans = trans->transid;
5741 mutex_unlock(&BTRFS_I(inode)->log_mutex);
Chris Masonaf4176b2009-03-24 10:24:31 -04005742
5743 /*
Chris Mason12fcfd22009-03-24 10:24:20 -04005744 * if this directory was already logged any new
5745 * names for this file/dir will get recorded
5746 */
5747 smp_mb();
5748 if (BTRFS_I(dir)->logged_trans == trans->transid)
5749 return;
5750
5751 /*
5752 * if the inode we're about to unlink was logged,
5753 * the log will be properly updated for any new names
5754 */
5755 if (BTRFS_I(inode)->logged_trans == trans->transid)
5756 return;
5757
5758 /*
5759 * when renaming files across directories, if the directory
5760 * there we're unlinking from gets fsync'd later on, there's
5761 * no way to find the destination directory later and fsync it
5762 * properly. So, we have to be conservative and force commits
5763 * so the new name gets discovered.
5764 */
5765 if (for_rename)
5766 goto record;
5767
5768 /* we can safely do the unlink without any special recording */
5769 return;
5770
5771record:
Filipe Manana2be63d52016-02-12 11:34:23 +00005772 mutex_lock(&BTRFS_I(dir)->log_mutex);
Chris Mason12fcfd22009-03-24 10:24:20 -04005773 BTRFS_I(dir)->last_unlink_trans = trans->transid;
Filipe Manana2be63d52016-02-12 11:34:23 +00005774 mutex_unlock(&BTRFS_I(dir)->log_mutex);
Chris Mason12fcfd22009-03-24 10:24:20 -04005775}
5776
5777/*
Filipe Manana1ec9a1a2016-02-10 10:42:25 +00005778 * Make sure that if someone attempts to fsync the parent directory of a deleted
5779 * snapshot, it ends up triggering a transaction commit. This is to guarantee
5780 * that after replaying the log tree of the parent directory's root we will not
5781 * see the snapshot anymore and at log replay time we will not see any log tree
5782 * corresponding to the deleted snapshot's root, which could lead to replaying
5783 * it after replaying the log tree of the parent directory (which would replay
5784 * the snapshot delete operation).
Filipe Manana2be63d52016-02-12 11:34:23 +00005785 *
5786 * Must be called before the actual snapshot destroy operation (updates to the
5787 * parent root and tree of tree roots trees, etc) are done.
Filipe Manana1ec9a1a2016-02-10 10:42:25 +00005788 */
5789void btrfs_record_snapshot_destroy(struct btrfs_trans_handle *trans,
5790 struct inode *dir)
5791{
Filipe Manana2be63d52016-02-12 11:34:23 +00005792 mutex_lock(&BTRFS_I(dir)->log_mutex);
Filipe Manana1ec9a1a2016-02-10 10:42:25 +00005793 BTRFS_I(dir)->last_unlink_trans = trans->transid;
Filipe Manana2be63d52016-02-12 11:34:23 +00005794 mutex_unlock(&BTRFS_I(dir)->log_mutex);
Filipe Manana1ec9a1a2016-02-10 10:42:25 +00005795}
5796
5797/*
Chris Mason12fcfd22009-03-24 10:24:20 -04005798 * Call this after adding a new name for a file and it will properly
5799 * update the log to reflect the new name.
5800 *
5801 * It will return zero if all goes well, and it will return 1 if a
5802 * full transaction commit is required.
5803 */
5804int btrfs_log_new_name(struct btrfs_trans_handle *trans,
5805 struct inode *inode, struct inode *old_dir,
5806 struct dentry *parent)
5807{
Jeff Mahoney0b246af2016-06-22 18:54:23 -04005808 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
Chris Mason12fcfd22009-03-24 10:24:20 -04005809 struct btrfs_root * root = BTRFS_I(inode)->root;
5810
5811 /*
Chris Masonaf4176b2009-03-24 10:24:31 -04005812 * this will force the logging code to walk the dentry chain
5813 * up for the file
5814 */
5815 if (S_ISREG(inode->i_mode))
5816 BTRFS_I(inode)->last_unlink_trans = trans->transid;
5817
5818 /*
Chris Mason12fcfd22009-03-24 10:24:20 -04005819 * if this inode hasn't been logged and directory we're renaming it
5820 * from hasn't been logged, we don't need to log it
5821 */
5822 if (BTRFS_I(inode)->logged_trans <=
Jeff Mahoney0b246af2016-06-22 18:54:23 -04005823 fs_info->last_trans_committed &&
Chris Mason12fcfd22009-03-24 10:24:20 -04005824 (!old_dir || BTRFS_I(old_dir)->logged_trans <=
Jeff Mahoney0b246af2016-06-22 18:54:23 -04005825 fs_info->last_trans_committed))
Chris Mason12fcfd22009-03-24 10:24:20 -04005826 return 0;
5827
Filipe Manana49dae1b2014-09-06 22:34:39 +01005828 return btrfs_log_inode_parent(trans, root, inode, parent, 0,
5829 LLONG_MAX, 1, NULL);
Chris Mason12fcfd22009-03-24 10:24:20 -04005830}
5831