blob: c525a685bfa29bfad72fe57f573f387b439c6016 [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
Liu Bo781feef2016-11-30 16:20:25 -080040#define LOG_OTHER_INODE 2
Chris Masone02119d2008-09-05 16:13:11 -040041
42/*
Chris Mason12fcfd22009-03-24 10:24:20 -040043 * directory trouble cases
44 *
45 * 1) on rename or unlink, if the inode being unlinked isn't in the fsync
46 * log, we must force a full commit before doing an fsync of the directory
47 * where the unlink was done.
48 * ---> record transid of last unlink/rename per directory
49 *
50 * mkdir foo/some_dir
51 * normal commit
52 * rename foo/some_dir foo2/some_dir
53 * mkdir foo/some_dir
54 * fsync foo/some_dir/some_file
55 *
56 * The fsync above will unlink the original some_dir without recording
57 * it in its new location (foo2). After a crash, some_dir will be gone
58 * unless the fsync of some_file forces a full commit
59 *
60 * 2) we must log any new names for any file or dir that is in the fsync
61 * log. ---> check inode while renaming/linking.
62 *
63 * 2a) we must log any new names for any file or dir during rename
64 * when the directory they are being removed from was logged.
65 * ---> check inode and old parent dir during rename
66 *
67 * 2a is actually the more important variant. With the extra logging
68 * a crash might unlink the old name without recreating the new one
69 *
70 * 3) after a crash, we must go through any directories with a link count
71 * of zero and redo the rm -rf
72 *
73 * mkdir f1/foo
74 * normal commit
75 * rm -rf f1/foo
76 * fsync(f1)
77 *
78 * The directory f1 was fully removed from the FS, but fsync was never
79 * called on f1, only its parent dir. After a crash the rm -rf must
80 * be replayed. This must be able to recurse down the entire
81 * directory tree. The inode link count fixup code takes care of the
82 * ugly details.
83 */
84
85/*
Chris Masone02119d2008-09-05 16:13:11 -040086 * stages for the tree walking. The first
87 * stage (0) is to only pin down the blocks we find
88 * the second stage (1) is to make sure that all the inodes
89 * we find in the log are created in the subvolume.
90 *
91 * The last stage is to deal with directories and links and extents
92 * and all the other fun semantics
93 */
94#define LOG_WALK_PIN_ONLY 0
95#define LOG_WALK_REPLAY_INODES 1
Josef Bacikdd8e7212013-09-11 11:57:23 -040096#define LOG_WALK_REPLAY_DIR_INDEX 2
97#define LOG_WALK_REPLAY_ALL 3
Chris Masone02119d2008-09-05 16:13:11 -040098
Chris Mason12fcfd22009-03-24 10:24:20 -040099static int btrfs_log_inode(struct btrfs_trans_handle *trans,
Nikolay Borisova59108a2017-01-18 00:31:48 +0200100 struct btrfs_root *root, struct btrfs_inode *inode,
Filipe Manana49dae1b2014-09-06 22:34:39 +0100101 int inode_only,
102 const loff_t start,
Filipe Manana8407f552014-09-05 15:14:39 +0100103 const loff_t end,
104 struct btrfs_log_ctx *ctx);
Yan Zhengec051c02009-01-05 15:43:42 -0500105static int link_to_fixup_dir(struct btrfs_trans_handle *trans,
106 struct btrfs_root *root,
107 struct btrfs_path *path, u64 objectid);
Chris Mason12fcfd22009-03-24 10:24:20 -0400108static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans,
109 struct btrfs_root *root,
110 struct btrfs_root *log,
111 struct btrfs_path *path,
112 u64 dirid, int del_all);
Chris Masone02119d2008-09-05 16:13:11 -0400113
114/*
115 * tree logging is a special write ahead log used to make sure that
116 * fsyncs and O_SYNCs can happen without doing full tree commits.
117 *
118 * Full tree commits are expensive because they require commonly
119 * modified blocks to be recowed, creating many dirty pages in the
120 * extent tree an 4x-6x higher write load than ext3.
121 *
122 * Instead of doing a tree commit on every fsync, we use the
123 * key ranges and transaction ids to find items for a given file or directory
124 * that have changed in this transaction. Those items are copied into
125 * a special tree (one per subvolume root), that tree is written to disk
126 * and then the fsync is considered complete.
127 *
128 * After a crash, items are copied out of the log-tree back into the
129 * subvolume tree. Any file data extents found are recorded in the extent
130 * allocation tree, and the log-tree freed.
131 *
132 * The log tree is read three times, once to pin down all the extents it is
133 * using in ram and once, once to create all the inodes logged in the tree
134 * and once to do all the other items.
135 */
136
137/*
Chris Masone02119d2008-09-05 16:13:11 -0400138 * start a sub transaction and setup the log tree
139 * this increments the log tree writer count to make the people
140 * syncing the tree wait for us to finish
141 */
142static int start_log_trans(struct btrfs_trans_handle *trans,
Miao Xie8b050d32014-02-20 18:08:58 +0800143 struct btrfs_root *root,
144 struct btrfs_log_ctx *ctx)
Chris Masone02119d2008-09-05 16:13:11 -0400145{
Jeff Mahoney0b246af2016-06-22 18:54:23 -0400146 struct btrfs_fs_info *fs_info = root->fs_info;
Zhaolei34eb2a52015-08-17 18:44:45 +0800147 int ret = 0;
Yan Zheng7237f182009-01-21 12:54:03 -0500148
149 mutex_lock(&root->log_mutex);
Zhaolei34eb2a52015-08-17 18:44:45 +0800150
Yan Zheng7237f182009-01-21 12:54:03 -0500151 if (root->log_root) {
Jeff Mahoney0b246af2016-06-22 18:54:23 -0400152 if (btrfs_need_log_full_commit(fs_info, trans)) {
Miao Xie50471a32014-02-20 18:08:57 +0800153 ret = -EAGAIN;
154 goto out;
155 }
Zhaolei34eb2a52015-08-17 18:44:45 +0800156
Josef Bacikff782e02009-10-08 15:30:04 -0400157 if (!root->log_start_pid) {
Miao Xie27cdeb72014-04-02 19:51:05 +0800158 clear_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state);
Zhaolei34eb2a52015-08-17 18:44:45 +0800159 root->log_start_pid = current->pid;
Josef Bacikff782e02009-10-08 15:30:04 -0400160 } else if (root->log_start_pid != current->pid) {
Miao Xie27cdeb72014-04-02 19:51:05 +0800161 set_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state);
Josef Bacikff782e02009-10-08 15:30:04 -0400162 }
Zhaolei34eb2a52015-08-17 18:44:45 +0800163 } else {
Jeff Mahoney0b246af2016-06-22 18:54:23 -0400164 mutex_lock(&fs_info->tree_log_mutex);
165 if (!fs_info->log_root_tree)
166 ret = btrfs_init_log_root_tree(trans, fs_info);
167 mutex_unlock(&fs_info->tree_log_mutex);
Zhaolei34eb2a52015-08-17 18:44:45 +0800168 if (ret)
169 goto out;
Josef Bacikff782e02009-10-08 15:30:04 -0400170
Chris Masone02119d2008-09-05 16:13:11 -0400171 ret = btrfs_add_log_tree(trans, root);
Yan, Zheng4a500fd2010-05-16 10:49:59 -0400172 if (ret)
Miao Xiee87ac132014-02-20 18:08:53 +0800173 goto out;
Zhaolei34eb2a52015-08-17 18:44:45 +0800174
175 clear_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state);
176 root->log_start_pid = current->pid;
Chris Masone02119d2008-09-05 16:13:11 -0400177 }
Zhaolei34eb2a52015-08-17 18:44:45 +0800178
Miao Xie2ecb7922012-09-06 04:04:27 -0600179 atomic_inc(&root->log_batch);
Yan Zheng7237f182009-01-21 12:54:03 -0500180 atomic_inc(&root->log_writers);
Miao Xie8b050d32014-02-20 18:08:58 +0800181 if (ctx) {
Zhaolei34eb2a52015-08-17 18:44:45 +0800182 int index = root->log_transid % 2;
Miao Xie8b050d32014-02-20 18:08:58 +0800183 list_add_tail(&ctx->list, &root->log_ctxs[index]);
Miao Xied1433de2014-02-20 18:08:59 +0800184 ctx->log_transid = root->log_transid;
Miao Xie8b050d32014-02-20 18:08:58 +0800185 }
Zhaolei34eb2a52015-08-17 18:44:45 +0800186
Miao Xiee87ac132014-02-20 18:08:53 +0800187out:
Yan Zheng7237f182009-01-21 12:54:03 -0500188 mutex_unlock(&root->log_mutex);
Miao Xiee87ac132014-02-20 18:08:53 +0800189 return ret;
Chris Masone02119d2008-09-05 16:13:11 -0400190}
191
192/*
193 * returns 0 if there was a log transaction running and we were able
194 * to join, or returns -ENOENT if there were not transactions
195 * in progress
196 */
197static int join_running_log_trans(struct btrfs_root *root)
198{
199 int ret = -ENOENT;
200
201 smp_mb();
202 if (!root->log_root)
203 return -ENOENT;
204
Yan Zheng7237f182009-01-21 12:54:03 -0500205 mutex_lock(&root->log_mutex);
Chris Masone02119d2008-09-05 16:13:11 -0400206 if (root->log_root) {
207 ret = 0;
Yan Zheng7237f182009-01-21 12:54:03 -0500208 atomic_inc(&root->log_writers);
Chris Masone02119d2008-09-05 16:13:11 -0400209 }
Yan Zheng7237f182009-01-21 12:54:03 -0500210 mutex_unlock(&root->log_mutex);
Chris Masone02119d2008-09-05 16:13:11 -0400211 return ret;
212}
213
214/*
Chris Mason12fcfd22009-03-24 10:24:20 -0400215 * This either makes the current running log transaction wait
216 * until you call btrfs_end_log_trans() or it makes any future
217 * log transactions wait until you call btrfs_end_log_trans()
218 */
219int btrfs_pin_log_trans(struct btrfs_root *root)
220{
221 int ret = -ENOENT;
222
223 mutex_lock(&root->log_mutex);
224 atomic_inc(&root->log_writers);
225 mutex_unlock(&root->log_mutex);
226 return ret;
227}
228
229/*
Chris Masone02119d2008-09-05 16:13:11 -0400230 * indicate we're done making changes to the log tree
231 * and wake up anyone waiting to do a sync
232 */
Jeff Mahoney143bede2012-03-01 14:56:26 +0100233void btrfs_end_log_trans(struct btrfs_root *root)
Chris Masone02119d2008-09-05 16:13:11 -0400234{
Yan Zheng7237f182009-01-21 12:54:03 -0500235 if (atomic_dec_and_test(&root->log_writers)) {
David Sterba779adf02015-02-16 19:39:00 +0100236 /*
237 * Implicit memory barrier after atomic_dec_and_test
238 */
Yan Zheng7237f182009-01-21 12:54:03 -0500239 if (waitqueue_active(&root->log_writer_wait))
240 wake_up(&root->log_writer_wait);
241 }
Chris Masone02119d2008-09-05 16:13:11 -0400242}
243
244
245/*
246 * the walk control struct is used to pass state down the chain when
247 * processing the log tree. The stage field tells us which part
248 * of the log tree processing we are currently doing. The others
249 * are state fields used for that specific part
250 */
251struct walk_control {
252 /* should we free the extent on disk when done? This is used
253 * at transaction commit time while freeing a log tree
254 */
255 int free;
256
257 /* should we write out the extent buffer? This is used
258 * while flushing the log tree to disk during a sync
259 */
260 int write;
261
262 /* should we wait for the extent buffer io to finish? Also used
263 * while flushing the log tree to disk for a sync
264 */
265 int wait;
266
267 /* pin only walk, we record which extents on disk belong to the
268 * log trees
269 */
270 int pin;
271
272 /* what stage of the replay code we're currently in */
273 int stage;
274
275 /* the root we are currently replaying */
276 struct btrfs_root *replay_dest;
277
278 /* the trans handle for the current replay */
279 struct btrfs_trans_handle *trans;
280
281 /* the function that gets used to process blocks we find in the
282 * tree. Note the extent_buffer might not be up to date when it is
283 * passed in, and it must be checked or read if you need the data
284 * inside it
285 */
286 int (*process_func)(struct btrfs_root *log, struct extent_buffer *eb,
287 struct walk_control *wc, u64 gen);
288};
289
290/*
291 * process_func used to pin down extents, write them or wait on them
292 */
293static int process_one_buffer(struct btrfs_root *log,
294 struct extent_buffer *eb,
295 struct walk_control *wc, u64 gen)
296{
Jeff Mahoney0b246af2016-06-22 18:54:23 -0400297 struct btrfs_fs_info *fs_info = log->fs_info;
Josef Bacikb50c6e22013-04-25 15:55:30 -0400298 int ret = 0;
Chris Masone02119d2008-09-05 16:13:11 -0400299
Josef Bacik8c2a1a32013-06-06 13:19:32 -0400300 /*
301 * If this fs is mixed then we need to be able to process the leaves to
302 * pin down any logged extents, so we have to read the block.
303 */
Jeff Mahoney0b246af2016-06-22 18:54:23 -0400304 if (btrfs_fs_incompat(fs_info, MIXED_GROUPS)) {
Josef Bacik8c2a1a32013-06-06 13:19:32 -0400305 ret = btrfs_read_buffer(eb, gen);
306 if (ret)
307 return ret;
308 }
309
Josef Bacikb50c6e22013-04-25 15:55:30 -0400310 if (wc->pin)
Jeff Mahoney2ff7e612016-06-22 18:54:24 -0400311 ret = btrfs_pin_extent_for_log_replay(fs_info, eb->start,
312 eb->len);
Josef Bacikb50c6e22013-04-25 15:55:30 -0400313
314 if (!ret && btrfs_buffer_uptodate(eb, gen, 0)) {
Josef Bacik8c2a1a32013-06-06 13:19:32 -0400315 if (wc->pin && btrfs_header_level(eb) == 0)
Jeff Mahoney2ff7e612016-06-22 18:54:24 -0400316 ret = btrfs_exclude_logged_extents(fs_info, eb);
Chris Masone02119d2008-09-05 16:13:11 -0400317 if (wc->write)
318 btrfs_write_tree_block(eb);
319 if (wc->wait)
320 btrfs_wait_tree_block_writeback(eb);
321 }
Josef Bacikb50c6e22013-04-25 15:55:30 -0400322 return ret;
Chris Masone02119d2008-09-05 16:13:11 -0400323}
324
325/*
326 * Item overwrite used by replay and tree logging. eb, slot and key all refer
327 * to the src data we are copying out.
328 *
329 * root is the tree we are copying into, and path is a scratch
330 * path for use in this function (it should be released on entry and
331 * will be released on exit).
332 *
333 * If the key is already in the destination tree the existing item is
334 * overwritten. If the existing item isn't big enough, it is extended.
335 * If it is too large, it is truncated.
336 *
337 * If the key isn't in the destination yet, a new item is inserted.
338 */
339static noinline int overwrite_item(struct btrfs_trans_handle *trans,
340 struct btrfs_root *root,
341 struct btrfs_path *path,
342 struct extent_buffer *eb, int slot,
343 struct btrfs_key *key)
344{
Jeff Mahoney2ff7e612016-06-22 18:54:24 -0400345 struct btrfs_fs_info *fs_info = root->fs_info;
Chris Masone02119d2008-09-05 16:13:11 -0400346 int ret;
347 u32 item_size;
348 u64 saved_i_size = 0;
349 int save_old_i_size = 0;
350 unsigned long src_ptr;
351 unsigned long dst_ptr;
352 int overwrite_root = 0;
Josef Bacik4bc4bee2013-04-05 20:50:09 +0000353 bool inode_item = key->type == BTRFS_INODE_ITEM_KEY;
Chris Masone02119d2008-09-05 16:13:11 -0400354
355 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
356 overwrite_root = 1;
357
358 item_size = btrfs_item_size_nr(eb, slot);
359 src_ptr = btrfs_item_ptr_offset(eb, slot);
360
361 /* look for the key in the destination tree */
362 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
Josef Bacik4bc4bee2013-04-05 20:50:09 +0000363 if (ret < 0)
364 return ret;
365
Chris Masone02119d2008-09-05 16:13:11 -0400366 if (ret == 0) {
367 char *src_copy;
368 char *dst_copy;
369 u32 dst_size = btrfs_item_size_nr(path->nodes[0],
370 path->slots[0]);
371 if (dst_size != item_size)
372 goto insert;
373
374 if (item_size == 0) {
David Sterbab3b4aa72011-04-21 01:20:15 +0200375 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -0400376 return 0;
377 }
378 dst_copy = kmalloc(item_size, GFP_NOFS);
379 src_copy = kmalloc(item_size, GFP_NOFS);
liubo2a29edc2011-01-26 06:22:08 +0000380 if (!dst_copy || !src_copy) {
David Sterbab3b4aa72011-04-21 01:20:15 +0200381 btrfs_release_path(path);
liubo2a29edc2011-01-26 06:22:08 +0000382 kfree(dst_copy);
383 kfree(src_copy);
384 return -ENOMEM;
385 }
Chris Masone02119d2008-09-05 16:13:11 -0400386
387 read_extent_buffer(eb, src_copy, src_ptr, item_size);
388
389 dst_ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
390 read_extent_buffer(path->nodes[0], dst_copy, dst_ptr,
391 item_size);
392 ret = memcmp(dst_copy, src_copy, item_size);
393
394 kfree(dst_copy);
395 kfree(src_copy);
396 /*
397 * they have the same contents, just return, this saves
398 * us from cowing blocks in the destination tree and doing
399 * extra writes that may not have been done by a previous
400 * sync
401 */
402 if (ret == 0) {
David Sterbab3b4aa72011-04-21 01:20:15 +0200403 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -0400404 return 0;
405 }
406
Josef Bacik4bc4bee2013-04-05 20:50:09 +0000407 /*
408 * We need to load the old nbytes into the inode so when we
409 * replay the extents we've logged we get the right nbytes.
410 */
411 if (inode_item) {
412 struct btrfs_inode_item *item;
413 u64 nbytes;
Josef Bacikd5554382013-09-11 14:17:00 -0400414 u32 mode;
Josef Bacik4bc4bee2013-04-05 20:50:09 +0000415
416 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
417 struct btrfs_inode_item);
418 nbytes = btrfs_inode_nbytes(path->nodes[0], item);
419 item = btrfs_item_ptr(eb, slot,
420 struct btrfs_inode_item);
421 btrfs_set_inode_nbytes(eb, item, nbytes);
Josef Bacikd5554382013-09-11 14:17:00 -0400422
423 /*
424 * If this is a directory we need to reset the i_size to
425 * 0 so that we can set it up properly when replaying
426 * the rest of the items in this log.
427 */
428 mode = btrfs_inode_mode(eb, item);
429 if (S_ISDIR(mode))
430 btrfs_set_inode_size(eb, item, 0);
Josef Bacik4bc4bee2013-04-05 20:50:09 +0000431 }
432 } else if (inode_item) {
433 struct btrfs_inode_item *item;
Josef Bacikd5554382013-09-11 14:17:00 -0400434 u32 mode;
Josef Bacik4bc4bee2013-04-05 20:50:09 +0000435
436 /*
437 * New inode, set nbytes to 0 so that the nbytes comes out
438 * properly when we replay the extents.
439 */
440 item = btrfs_item_ptr(eb, slot, struct btrfs_inode_item);
441 btrfs_set_inode_nbytes(eb, item, 0);
Josef Bacikd5554382013-09-11 14:17:00 -0400442
443 /*
444 * If this is a directory we need to reset the i_size to 0 so
445 * that we can set it up properly when replaying the rest of
446 * the items in this log.
447 */
448 mode = btrfs_inode_mode(eb, item);
449 if (S_ISDIR(mode))
450 btrfs_set_inode_size(eb, item, 0);
Chris Masone02119d2008-09-05 16:13:11 -0400451 }
452insert:
David Sterbab3b4aa72011-04-21 01:20:15 +0200453 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -0400454 /* try to insert the key into the destination tree */
Filipe Mananadf8d1162015-01-14 01:52:25 +0000455 path->skip_release_on_error = 1;
Chris Masone02119d2008-09-05 16:13:11 -0400456 ret = btrfs_insert_empty_item(trans, root, path,
457 key, item_size);
Filipe Mananadf8d1162015-01-14 01:52:25 +0000458 path->skip_release_on_error = 0;
Chris Masone02119d2008-09-05 16:13:11 -0400459
460 /* make sure any existing item is the correct size */
Filipe Mananadf8d1162015-01-14 01:52:25 +0000461 if (ret == -EEXIST || ret == -EOVERFLOW) {
Chris Masone02119d2008-09-05 16:13:11 -0400462 u32 found_size;
463 found_size = btrfs_item_size_nr(path->nodes[0],
464 path->slots[0]);
Jeff Mahoney143bede2012-03-01 14:56:26 +0100465 if (found_size > item_size)
Jeff Mahoney2ff7e612016-06-22 18:54:24 -0400466 btrfs_truncate_item(fs_info, path, item_size, 1);
Jeff Mahoney143bede2012-03-01 14:56:26 +0100467 else if (found_size < item_size)
Jeff Mahoney2ff7e612016-06-22 18:54:24 -0400468 btrfs_extend_item(fs_info, path,
Jeff Mahoney143bede2012-03-01 14:56:26 +0100469 item_size - found_size);
Chris Masone02119d2008-09-05 16:13:11 -0400470 } else if (ret) {
Yan, Zheng4a500fd2010-05-16 10:49:59 -0400471 return ret;
Chris Masone02119d2008-09-05 16:13:11 -0400472 }
473 dst_ptr = btrfs_item_ptr_offset(path->nodes[0],
474 path->slots[0]);
475
476 /* don't overwrite an existing inode if the generation number
477 * was logged as zero. This is done when the tree logging code
478 * is just logging an inode to make sure it exists after recovery.
479 *
480 * Also, don't overwrite i_size on directories during replay.
481 * log replay inserts and removes directory items based on the
482 * state of the tree found in the subvolume, and i_size is modified
483 * as it goes
484 */
485 if (key->type == BTRFS_INODE_ITEM_KEY && ret == -EEXIST) {
486 struct btrfs_inode_item *src_item;
487 struct btrfs_inode_item *dst_item;
488
489 src_item = (struct btrfs_inode_item *)src_ptr;
490 dst_item = (struct btrfs_inode_item *)dst_ptr;
491
Filipe Manana1a4bcf42015-02-13 12:30:56 +0000492 if (btrfs_inode_generation(eb, src_item) == 0) {
493 struct extent_buffer *dst_eb = path->nodes[0];
Filipe Manana2f2ff0e2015-03-20 17:19:46 +0000494 const u64 ino_size = btrfs_inode_size(eb, src_item);
Filipe Manana1a4bcf42015-02-13 12:30:56 +0000495
Filipe Manana2f2ff0e2015-03-20 17:19:46 +0000496 /*
497 * For regular files an ino_size == 0 is used only when
498 * logging that an inode exists, as part of a directory
499 * fsync, and the inode wasn't fsynced before. In this
500 * case don't set the size of the inode in the fs/subvol
501 * tree, otherwise we would be throwing valid data away.
502 */
Filipe Manana1a4bcf42015-02-13 12:30:56 +0000503 if (S_ISREG(btrfs_inode_mode(eb, src_item)) &&
Filipe Manana2f2ff0e2015-03-20 17:19:46 +0000504 S_ISREG(btrfs_inode_mode(dst_eb, dst_item)) &&
505 ino_size != 0) {
Filipe Manana1a4bcf42015-02-13 12:30:56 +0000506 struct btrfs_map_token token;
Filipe Manana1a4bcf42015-02-13 12:30:56 +0000507
508 btrfs_init_map_token(&token);
509 btrfs_set_token_inode_size(dst_eb, dst_item,
510 ino_size, &token);
511 }
Chris Masone02119d2008-09-05 16:13:11 -0400512 goto no_copy;
Filipe Manana1a4bcf42015-02-13 12:30:56 +0000513 }
Chris Masone02119d2008-09-05 16:13:11 -0400514
515 if (overwrite_root &&
516 S_ISDIR(btrfs_inode_mode(eb, src_item)) &&
517 S_ISDIR(btrfs_inode_mode(path->nodes[0], dst_item))) {
518 save_old_i_size = 1;
519 saved_i_size = btrfs_inode_size(path->nodes[0],
520 dst_item);
521 }
522 }
523
524 copy_extent_buffer(path->nodes[0], eb, dst_ptr,
525 src_ptr, item_size);
526
527 if (save_old_i_size) {
528 struct btrfs_inode_item *dst_item;
529 dst_item = (struct btrfs_inode_item *)dst_ptr;
530 btrfs_set_inode_size(path->nodes[0], dst_item, saved_i_size);
531 }
532
533 /* make sure the generation is filled in */
534 if (key->type == BTRFS_INODE_ITEM_KEY) {
535 struct btrfs_inode_item *dst_item;
536 dst_item = (struct btrfs_inode_item *)dst_ptr;
537 if (btrfs_inode_generation(path->nodes[0], dst_item) == 0) {
538 btrfs_set_inode_generation(path->nodes[0], dst_item,
539 trans->transid);
540 }
541 }
542no_copy:
543 btrfs_mark_buffer_dirty(path->nodes[0]);
David Sterbab3b4aa72011-04-21 01:20:15 +0200544 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -0400545 return 0;
546}
547
548/*
549 * simple helper to read an inode off the disk from a given root
550 * This can only be called for subvolume roots and not for the log
551 */
552static noinline struct inode *read_one_inode(struct btrfs_root *root,
553 u64 objectid)
554{
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400555 struct btrfs_key key;
Chris Masone02119d2008-09-05 16:13:11 -0400556 struct inode *inode;
Chris Masone02119d2008-09-05 16:13:11 -0400557
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400558 key.objectid = objectid;
559 key.type = BTRFS_INODE_ITEM_KEY;
560 key.offset = 0;
Josef Bacik73f73412009-12-04 17:38:27 +0000561 inode = btrfs_iget(root->fs_info->sb, &key, root, NULL);
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400562 if (IS_ERR(inode)) {
563 inode = NULL;
564 } else if (is_bad_inode(inode)) {
Chris Masone02119d2008-09-05 16:13:11 -0400565 iput(inode);
566 inode = NULL;
567 }
568 return inode;
569}
570
571/* replays a single extent in 'eb' at 'slot' with 'key' into the
572 * subvolume 'root'. path is released on entry and should be released
573 * on exit.
574 *
575 * extents in the log tree have not been allocated out of the extent
576 * tree yet. So, this completes the allocation, taking a reference
577 * as required if the extent already exists or creating a new extent
578 * if it isn't in the extent allocation tree yet.
579 *
580 * The extent is inserted into the file, dropping any existing extents
581 * from the file that overlap the new one.
582 */
583static noinline int replay_one_extent(struct btrfs_trans_handle *trans,
584 struct btrfs_root *root,
585 struct btrfs_path *path,
586 struct extent_buffer *eb, int slot,
587 struct btrfs_key *key)
588{
Jeff Mahoney0b246af2016-06-22 18:54:23 -0400589 struct btrfs_fs_info *fs_info = root->fs_info;
Chris Masone02119d2008-09-05 16:13:11 -0400590 int found_type;
Chris Masone02119d2008-09-05 16:13:11 -0400591 u64 extent_end;
Chris Masone02119d2008-09-05 16:13:11 -0400592 u64 start = key->offset;
Josef Bacik4bc4bee2013-04-05 20:50:09 +0000593 u64 nbytes = 0;
Chris Masone02119d2008-09-05 16:13:11 -0400594 struct btrfs_file_extent_item *item;
595 struct inode *inode = NULL;
596 unsigned long size;
597 int ret = 0;
598
599 item = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
600 found_type = btrfs_file_extent_type(eb, item);
601
Yan Zhengd899e052008-10-30 14:25:28 -0400602 if (found_type == BTRFS_FILE_EXTENT_REG ||
Josef Bacik4bc4bee2013-04-05 20:50:09 +0000603 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
604 nbytes = btrfs_file_extent_num_bytes(eb, item);
605 extent_end = start + nbytes;
606
607 /*
608 * We don't add to the inodes nbytes if we are prealloc or a
609 * hole.
610 */
611 if (btrfs_file_extent_disk_bytenr(eb, item) == 0)
612 nbytes = 0;
613 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
Chris Mason514ac8a2014-01-03 21:07:00 -0800614 size = btrfs_file_extent_inline_len(eb, slot, item);
Josef Bacik4bc4bee2013-04-05 20:50:09 +0000615 nbytes = btrfs_file_extent_ram_bytes(eb, item);
Jeff Mahoneyda170662016-06-15 09:22:56 -0400616 extent_end = ALIGN(start + size,
Jeff Mahoney0b246af2016-06-22 18:54:23 -0400617 fs_info->sectorsize);
Chris Masone02119d2008-09-05 16:13:11 -0400618 } else {
619 ret = 0;
620 goto out;
621 }
622
623 inode = read_one_inode(root, key->objectid);
624 if (!inode) {
625 ret = -EIO;
626 goto out;
627 }
628
629 /*
630 * first check to see if we already have this extent in the
631 * file. This must be done before the btrfs_drop_extents run
632 * so we don't try to drop this extent.
633 */
Nikolay Borisov4a0cc7c2017-01-10 20:35:31 +0200634 ret = btrfs_lookup_file_extent(trans, root, path, btrfs_ino(BTRFS_I(inode)),
Chris Masone02119d2008-09-05 16:13:11 -0400635 start, 0);
636
Yan Zhengd899e052008-10-30 14:25:28 -0400637 if (ret == 0 &&
638 (found_type == BTRFS_FILE_EXTENT_REG ||
639 found_type == BTRFS_FILE_EXTENT_PREALLOC)) {
Chris Masone02119d2008-09-05 16:13:11 -0400640 struct btrfs_file_extent_item cmp1;
641 struct btrfs_file_extent_item cmp2;
642 struct btrfs_file_extent_item *existing;
643 struct extent_buffer *leaf;
644
645 leaf = path->nodes[0];
646 existing = btrfs_item_ptr(leaf, path->slots[0],
647 struct btrfs_file_extent_item);
648
649 read_extent_buffer(eb, &cmp1, (unsigned long)item,
650 sizeof(cmp1));
651 read_extent_buffer(leaf, &cmp2, (unsigned long)existing,
652 sizeof(cmp2));
653
654 /*
655 * we already have a pointer to this exact extent,
656 * we don't have to do anything
657 */
658 if (memcmp(&cmp1, &cmp2, sizeof(cmp1)) == 0) {
David Sterbab3b4aa72011-04-21 01:20:15 +0200659 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -0400660 goto out;
661 }
662 }
David Sterbab3b4aa72011-04-21 01:20:15 +0200663 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -0400664
665 /* drop any overlapping extents */
Josef Bacik26714852012-08-29 12:24:27 -0400666 ret = btrfs_drop_extents(trans, root, inode, start, extent_end, 1);
Josef Bacik36508602013-04-25 16:23:32 -0400667 if (ret)
668 goto out;
Chris Masone02119d2008-09-05 16:13:11 -0400669
Yan Zheng07d400a2009-01-06 11:42:00 -0500670 if (found_type == BTRFS_FILE_EXTENT_REG ||
671 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400672 u64 offset;
Yan Zheng07d400a2009-01-06 11:42:00 -0500673 unsigned long dest_offset;
674 struct btrfs_key ins;
Chris Masone02119d2008-09-05 16:13:11 -0400675
Yan Zheng07d400a2009-01-06 11:42:00 -0500676 ret = btrfs_insert_empty_item(trans, root, path, key,
677 sizeof(*item));
Josef Bacik36508602013-04-25 16:23:32 -0400678 if (ret)
679 goto out;
Yan Zheng07d400a2009-01-06 11:42:00 -0500680 dest_offset = btrfs_item_ptr_offset(path->nodes[0],
681 path->slots[0]);
682 copy_extent_buffer(path->nodes[0], eb, dest_offset,
683 (unsigned long)item, sizeof(*item));
684
685 ins.objectid = btrfs_file_extent_disk_bytenr(eb, item);
686 ins.offset = btrfs_file_extent_disk_num_bytes(eb, item);
687 ins.type = BTRFS_EXTENT_ITEM_KEY;
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400688 offset = key->offset - btrfs_file_extent_offset(eb, item);
Yan Zheng07d400a2009-01-06 11:42:00 -0500689
Qu Wenruodf2c95f2016-08-15 10:36:52 +0800690 /*
691 * Manually record dirty extent, as here we did a shallow
692 * file extent item copy and skip normal backref update,
693 * but modifying extent tree all by ourselves.
694 * So need to manually record dirty extent for qgroup,
695 * as the owner of the file extent changed from log tree
696 * (doesn't affect qgroup) to fs/file tree(affects qgroup)
697 */
Jeff Mahoney0b246af2016-06-22 18:54:23 -0400698 ret = btrfs_qgroup_trace_extent(trans, fs_info,
Qu Wenruodf2c95f2016-08-15 10:36:52 +0800699 btrfs_file_extent_disk_bytenr(eb, item),
700 btrfs_file_extent_disk_num_bytes(eb, item),
701 GFP_NOFS);
702 if (ret < 0)
703 goto out;
704
Yan Zheng07d400a2009-01-06 11:42:00 -0500705 if (ins.objectid > 0) {
706 u64 csum_start;
707 u64 csum_end;
708 LIST_HEAD(ordered_sums);
709 /*
710 * is this extent already allocated in the extent
711 * allocation tree? If so, just add a reference
712 */
Jeff Mahoney2ff7e612016-06-22 18:54:24 -0400713 ret = btrfs_lookup_data_extent(fs_info, ins.objectid,
Yan Zheng07d400a2009-01-06 11:42:00 -0500714 ins.offset);
715 if (ret == 0) {
Jeff Mahoney2ff7e612016-06-22 18:54:24 -0400716 ret = btrfs_inc_extent_ref(trans, fs_info,
Yan Zheng07d400a2009-01-06 11:42:00 -0500717 ins.objectid, ins.offset,
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400718 0, root->root_key.objectid,
Filipe Mananab06c4bf2015-10-23 07:52:54 +0100719 key->objectid, offset);
Josef Bacikb50c6e22013-04-25 15:55:30 -0400720 if (ret)
721 goto out;
Yan Zheng07d400a2009-01-06 11:42:00 -0500722 } else {
723 /*
724 * insert the extent pointer in the extent
725 * allocation tree
726 */
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400727 ret = btrfs_alloc_logged_file_extent(trans,
Jeff Mahoney2ff7e612016-06-22 18:54:24 -0400728 fs_info,
729 root->root_key.objectid,
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400730 key->objectid, offset, &ins);
Josef Bacikb50c6e22013-04-25 15:55:30 -0400731 if (ret)
732 goto out;
Yan Zheng07d400a2009-01-06 11:42:00 -0500733 }
David Sterbab3b4aa72011-04-21 01:20:15 +0200734 btrfs_release_path(path);
Yan Zheng07d400a2009-01-06 11:42:00 -0500735
736 if (btrfs_file_extent_compression(eb, item)) {
737 csum_start = ins.objectid;
738 csum_end = csum_start + ins.offset;
739 } else {
740 csum_start = ins.objectid +
741 btrfs_file_extent_offset(eb, item);
742 csum_end = csum_start +
743 btrfs_file_extent_num_bytes(eb, item);
744 }
745
746 ret = btrfs_lookup_csums_range(root->log_root,
747 csum_start, csum_end - 1,
Arne Jansena2de7332011-03-08 14:14:00 +0100748 &ordered_sums, 0);
Josef Bacik36508602013-04-25 16:23:32 -0400749 if (ret)
750 goto out;
Filipe Mananab84b8392015-08-19 11:09:40 +0100751 /*
752 * Now delete all existing cums in the csum root that
753 * cover our range. We do this because we can have an
754 * extent that is completely referenced by one file
755 * extent item and partially referenced by another
756 * file extent item (like after using the clone or
757 * extent_same ioctls). In this case if we end up doing
758 * the replay of the one that partially references the
759 * extent first, and we do not do the csum deletion
760 * below, we can get 2 csum items in the csum tree that
761 * overlap each other. For example, imagine our log has
762 * the two following file extent items:
763 *
764 * key (257 EXTENT_DATA 409600)
765 * extent data disk byte 12845056 nr 102400
766 * extent data offset 20480 nr 20480 ram 102400
767 *
768 * key (257 EXTENT_DATA 819200)
769 * extent data disk byte 12845056 nr 102400
770 * extent data offset 0 nr 102400 ram 102400
771 *
772 * Where the second one fully references the 100K extent
773 * that starts at disk byte 12845056, and the log tree
774 * has a single csum item that covers the entire range
775 * of the extent:
776 *
777 * key (EXTENT_CSUM EXTENT_CSUM 12845056) itemsize 100
778 *
779 * After the first file extent item is replayed, the
780 * csum tree gets the following csum item:
781 *
782 * key (EXTENT_CSUM EXTENT_CSUM 12865536) itemsize 20
783 *
784 * Which covers the 20K sub-range starting at offset 20K
785 * of our extent. Now when we replay the second file
786 * extent item, if we do not delete existing csum items
787 * that cover any of its blocks, we end up getting two
788 * csum items in our csum tree that overlap each other:
789 *
790 * key (EXTENT_CSUM EXTENT_CSUM 12845056) itemsize 100
791 * key (EXTENT_CSUM EXTENT_CSUM 12865536) itemsize 20
792 *
793 * Which is a problem, because after this anyone trying
794 * to lookup up for the checksum of any block of our
795 * extent starting at an offset of 40K or higher, will
796 * end up looking at the second csum item only, which
797 * does not contain the checksum for any block starting
798 * at offset 40K or higher of our extent.
799 */
Yan Zheng07d400a2009-01-06 11:42:00 -0500800 while (!list_empty(&ordered_sums)) {
801 struct btrfs_ordered_sum *sums;
802 sums = list_entry(ordered_sums.next,
803 struct btrfs_ordered_sum,
804 list);
Josef Bacik36508602013-04-25 16:23:32 -0400805 if (!ret)
Jeff Mahoney0b246af2016-06-22 18:54:23 -0400806 ret = btrfs_del_csums(trans, fs_info,
Jeff Mahoney5b4aace2016-06-21 10:40:19 -0400807 sums->bytenr,
808 sums->len);
Filipe Mananab84b8392015-08-19 11:09:40 +0100809 if (!ret)
Josef Bacik36508602013-04-25 16:23:32 -0400810 ret = btrfs_csum_file_blocks(trans,
Jeff Mahoney0b246af2016-06-22 18:54:23 -0400811 fs_info->csum_root, sums);
Yan Zheng07d400a2009-01-06 11:42:00 -0500812 list_del(&sums->list);
813 kfree(sums);
814 }
Josef Bacik36508602013-04-25 16:23:32 -0400815 if (ret)
816 goto out;
Yan Zheng07d400a2009-01-06 11:42:00 -0500817 } else {
David Sterbab3b4aa72011-04-21 01:20:15 +0200818 btrfs_release_path(path);
Yan Zheng07d400a2009-01-06 11:42:00 -0500819 }
820 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
821 /* inline extents are easy, we just overwrite them */
822 ret = overwrite_item(trans, root, path, eb, slot, key);
Josef Bacik36508602013-04-25 16:23:32 -0400823 if (ret)
824 goto out;
Yan Zheng07d400a2009-01-06 11:42:00 -0500825 }
826
Josef Bacik4bc4bee2013-04-05 20:50:09 +0000827 inode_add_bytes(inode, nbytes);
Tsutomu Itohb9959292012-06-25 21:25:22 -0600828 ret = btrfs_update_inode(trans, root, inode);
Chris Masone02119d2008-09-05 16:13:11 -0400829out:
830 if (inode)
831 iput(inode);
832 return ret;
833}
834
835/*
836 * when cleaning up conflicts between the directory names in the
837 * subvolume, directory names in the log and directory names in the
838 * inode back references, we may have to unlink inodes from directories.
839 *
840 * This is a helper function to do the unlink of a specific directory
841 * item
842 */
843static noinline int drop_one_dir_item(struct btrfs_trans_handle *trans,
844 struct btrfs_root *root,
845 struct btrfs_path *path,
Nikolay Borisov207e7d92017-01-18 00:31:45 +0200846 struct btrfs_inode *dir,
Chris Masone02119d2008-09-05 16:13:11 -0400847 struct btrfs_dir_item *di)
848{
Jeff Mahoney2ff7e612016-06-22 18:54:24 -0400849 struct btrfs_fs_info *fs_info = root->fs_info;
Chris Masone02119d2008-09-05 16:13:11 -0400850 struct inode *inode;
851 char *name;
852 int name_len;
853 struct extent_buffer *leaf;
854 struct btrfs_key location;
855 int ret;
856
857 leaf = path->nodes[0];
858
859 btrfs_dir_item_key_to_cpu(leaf, di, &location);
860 name_len = btrfs_dir_name_len(leaf, di);
861 name = kmalloc(name_len, GFP_NOFS);
liubo2a29edc2011-01-26 06:22:08 +0000862 if (!name)
863 return -ENOMEM;
864
Chris Masone02119d2008-09-05 16:13:11 -0400865 read_extent_buffer(leaf, name, (unsigned long)(di + 1), name_len);
David Sterbab3b4aa72011-04-21 01:20:15 +0200866 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -0400867
868 inode = read_one_inode(root, location.objectid);
Tsutomu Itohc00e9492011-04-28 09:10:23 +0000869 if (!inode) {
Josef Bacik36508602013-04-25 16:23:32 -0400870 ret = -EIO;
871 goto out;
Tsutomu Itohc00e9492011-04-28 09:10:23 +0000872 }
Chris Masone02119d2008-09-05 16:13:11 -0400873
Yan Zhengec051c02009-01-05 15:43:42 -0500874 ret = link_to_fixup_dir(trans, root, path, location.objectid);
Josef Bacik36508602013-04-25 16:23:32 -0400875 if (ret)
876 goto out;
Chris Mason12fcfd22009-03-24 10:24:20 -0400877
Nikolay Borisov207e7d92017-01-18 00:31:45 +0200878 ret = btrfs_unlink_inode(trans, root, dir, BTRFS_I(inode), name,
879 name_len);
Josef Bacik36508602013-04-25 16:23:32 -0400880 if (ret)
881 goto out;
Filipe David Borba Mananaada9af22013-08-05 09:25:47 +0100882 else
Jeff Mahoney2ff7e612016-06-22 18:54:24 -0400883 ret = btrfs_run_delayed_items(trans, fs_info);
Josef Bacik36508602013-04-25 16:23:32 -0400884out:
885 kfree(name);
886 iput(inode);
Chris Masone02119d2008-09-05 16:13:11 -0400887 return ret;
888}
889
890/*
891 * helper function to see if a given name and sequence number found
892 * in an inode back reference are already in a directory and correctly
893 * point to this inode
894 */
895static noinline int inode_in_dir(struct btrfs_root *root,
896 struct btrfs_path *path,
897 u64 dirid, u64 objectid, u64 index,
898 const char *name, int name_len)
899{
900 struct btrfs_dir_item *di;
901 struct btrfs_key location;
902 int match = 0;
903
904 di = btrfs_lookup_dir_index_item(NULL, root, path, dirid,
905 index, name, name_len, 0);
906 if (di && !IS_ERR(di)) {
907 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
908 if (location.objectid != objectid)
909 goto out;
910 } else
911 goto out;
David Sterbab3b4aa72011-04-21 01:20:15 +0200912 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -0400913
914 di = btrfs_lookup_dir_item(NULL, root, path, dirid, name, name_len, 0);
915 if (di && !IS_ERR(di)) {
916 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
917 if (location.objectid != objectid)
918 goto out;
919 } else
920 goto out;
921 match = 1;
922out:
David Sterbab3b4aa72011-04-21 01:20:15 +0200923 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -0400924 return match;
925}
926
927/*
928 * helper function to check a log tree for a named back reference in
929 * an inode. This is used to decide if a back reference that is
930 * found in the subvolume conflicts with what we find in the log.
931 *
932 * inode backreferences may have multiple refs in a single item,
933 * during replay we process one reference at a time, and we don't
934 * want to delete valid links to a file from the subvolume if that
935 * link is also in the log.
936 */
937static noinline int backref_in_log(struct btrfs_root *log,
938 struct btrfs_key *key,
Mark Fashehf1863732012-08-08 11:32:27 -0700939 u64 ref_objectid,
Filipe Mananadf8d1162015-01-14 01:52:25 +0000940 const char *name, int namelen)
Chris Masone02119d2008-09-05 16:13:11 -0400941{
942 struct btrfs_path *path;
943 struct btrfs_inode_ref *ref;
944 unsigned long ptr;
945 unsigned long ptr_end;
946 unsigned long name_ptr;
947 int found_name_len;
948 int item_size;
949 int ret;
950 int match = 0;
951
952 path = btrfs_alloc_path();
liubo2a29edc2011-01-26 06:22:08 +0000953 if (!path)
954 return -ENOMEM;
955
Chris Masone02119d2008-09-05 16:13:11 -0400956 ret = btrfs_search_slot(NULL, log, key, path, 0, 0);
957 if (ret != 0)
958 goto out;
959
Chris Masone02119d2008-09-05 16:13:11 -0400960 ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
Mark Fashehf1863732012-08-08 11:32:27 -0700961
962 if (key->type == BTRFS_INODE_EXTREF_KEY) {
963 if (btrfs_find_name_in_ext_backref(path, ref_objectid,
964 name, namelen, NULL))
965 match = 1;
966
967 goto out;
968 }
969
970 item_size = btrfs_item_size_nr(path->nodes[0], path->slots[0]);
Chris Masone02119d2008-09-05 16:13:11 -0400971 ptr_end = ptr + item_size;
972 while (ptr < ptr_end) {
973 ref = (struct btrfs_inode_ref *)ptr;
974 found_name_len = btrfs_inode_ref_name_len(path->nodes[0], ref);
975 if (found_name_len == namelen) {
976 name_ptr = (unsigned long)(ref + 1);
977 ret = memcmp_extent_buffer(path->nodes[0], name,
978 name_ptr, namelen);
979 if (ret == 0) {
980 match = 1;
981 goto out;
982 }
983 }
984 ptr = (unsigned long)(ref + 1) + found_name_len;
985 }
986out:
987 btrfs_free_path(path);
988 return match;
989}
990
Jan Schmidt5a1d7842012-08-17 14:04:41 -0700991static inline int __add_inode_ref(struct btrfs_trans_handle *trans,
992 struct btrfs_root *root,
993 struct btrfs_path *path,
994 struct btrfs_root *log_root,
Nikolay Borisov94c91a12017-01-18 00:31:46 +0200995 struct btrfs_inode *dir,
996 struct btrfs_inode *inode,
Jan Schmidt5a1d7842012-08-17 14:04:41 -0700997 struct extent_buffer *eb,
Mark Fashehf1863732012-08-08 11:32:27 -0700998 u64 inode_objectid, u64 parent_objectid,
999 u64 ref_index, char *name, int namelen,
1000 int *search_done)
Jan Schmidt5a1d7842012-08-17 14:04:41 -07001001{
Jeff Mahoney2ff7e612016-06-22 18:54:24 -04001002 struct btrfs_fs_info *fs_info = root->fs_info;
Jan Schmidt5a1d7842012-08-17 14:04:41 -07001003 int ret;
Mark Fashehf1863732012-08-08 11:32:27 -07001004 char *victim_name;
1005 int victim_name_len;
1006 struct extent_buffer *leaf;
Jan Schmidt5a1d7842012-08-17 14:04:41 -07001007 struct btrfs_dir_item *di;
Mark Fashehf1863732012-08-08 11:32:27 -07001008 struct btrfs_key search_key;
1009 struct btrfs_inode_extref *extref;
Jan Schmidt5a1d7842012-08-17 14:04:41 -07001010
Mark Fashehf1863732012-08-08 11:32:27 -07001011again:
1012 /* Search old style refs */
1013 search_key.objectid = inode_objectid;
1014 search_key.type = BTRFS_INODE_REF_KEY;
1015 search_key.offset = parent_objectid;
1016 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
Jan Schmidt5a1d7842012-08-17 14:04:41 -07001017 if (ret == 0) {
Jan Schmidt5a1d7842012-08-17 14:04:41 -07001018 struct btrfs_inode_ref *victim_ref;
1019 unsigned long ptr;
1020 unsigned long ptr_end;
Mark Fashehf1863732012-08-08 11:32:27 -07001021
1022 leaf = path->nodes[0];
Jan Schmidt5a1d7842012-08-17 14:04:41 -07001023
1024 /* are we trying to overwrite a back ref for the root directory
1025 * if so, just jump out, we're done
1026 */
Mark Fashehf1863732012-08-08 11:32:27 -07001027 if (search_key.objectid == search_key.offset)
Jan Schmidt5a1d7842012-08-17 14:04:41 -07001028 return 1;
1029
1030 /* check all the names in this back reference to see
1031 * if they are in the log. if so, we allow them to stay
1032 * otherwise they must be unlinked as a conflict
1033 */
1034 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
1035 ptr_end = ptr + btrfs_item_size_nr(leaf, path->slots[0]);
1036 while (ptr < ptr_end) {
1037 victim_ref = (struct btrfs_inode_ref *)ptr;
1038 victim_name_len = btrfs_inode_ref_name_len(leaf,
1039 victim_ref);
1040 victim_name = kmalloc(victim_name_len, GFP_NOFS);
Josef Bacik36508602013-04-25 16:23:32 -04001041 if (!victim_name)
1042 return -ENOMEM;
Jan Schmidt5a1d7842012-08-17 14:04:41 -07001043
1044 read_extent_buffer(leaf, victim_name,
1045 (unsigned long)(victim_ref + 1),
1046 victim_name_len);
1047
Mark Fashehf1863732012-08-08 11:32:27 -07001048 if (!backref_in_log(log_root, &search_key,
1049 parent_objectid,
1050 victim_name,
Jan Schmidt5a1d7842012-08-17 14:04:41 -07001051 victim_name_len)) {
Nikolay Borisov94c91a12017-01-18 00:31:46 +02001052 inc_nlink(&inode->vfs_inode);
Jan Schmidt5a1d7842012-08-17 14:04:41 -07001053 btrfs_release_path(path);
1054
Nikolay Borisov94c91a12017-01-18 00:31:46 +02001055 ret = btrfs_unlink_inode(trans, root, dir, inode,
Nikolay Borisov4ec59342017-01-18 00:31:44 +02001056 victim_name, victim_name_len);
Mark Fashehf1863732012-08-08 11:32:27 -07001057 kfree(victim_name);
Josef Bacik36508602013-04-25 16:23:32 -04001058 if (ret)
1059 return ret;
Jeff Mahoney2ff7e612016-06-22 18:54:24 -04001060 ret = btrfs_run_delayed_items(trans, fs_info);
Filipe David Borba Mananaada9af22013-08-05 09:25:47 +01001061 if (ret)
1062 return ret;
Mark Fashehf1863732012-08-08 11:32:27 -07001063 *search_done = 1;
1064 goto again;
Jan Schmidt5a1d7842012-08-17 14:04:41 -07001065 }
1066 kfree(victim_name);
Mark Fashehf1863732012-08-08 11:32:27 -07001067
Jan Schmidt5a1d7842012-08-17 14:04:41 -07001068 ptr = (unsigned long)(victim_ref + 1) + victim_name_len;
1069 }
Jan Schmidt5a1d7842012-08-17 14:04:41 -07001070
1071 /*
1072 * NOTE: we have searched root tree and checked the
Adam Buchbinderbb7ab3b2016-03-04 11:23:12 -08001073 * corresponding ref, it does not need to check again.
Jan Schmidt5a1d7842012-08-17 14:04:41 -07001074 */
1075 *search_done = 1;
1076 }
1077 btrfs_release_path(path);
1078
Mark Fashehf1863732012-08-08 11:32:27 -07001079 /* Same search but for extended refs */
1080 extref = btrfs_lookup_inode_extref(NULL, root, path, name, namelen,
1081 inode_objectid, parent_objectid, 0,
1082 0);
1083 if (!IS_ERR_OR_NULL(extref)) {
1084 u32 item_size;
1085 u32 cur_offset = 0;
1086 unsigned long base;
1087 struct inode *victim_parent;
1088
1089 leaf = path->nodes[0];
1090
1091 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1092 base = btrfs_item_ptr_offset(leaf, path->slots[0]);
1093
1094 while (cur_offset < item_size) {
Quentin Casasnovasdd9ef132015-03-03 16:31:38 +01001095 extref = (struct btrfs_inode_extref *)(base + cur_offset);
Mark Fashehf1863732012-08-08 11:32:27 -07001096
1097 victim_name_len = btrfs_inode_extref_name_len(leaf, extref);
1098
1099 if (btrfs_inode_extref_parent(leaf, extref) != parent_objectid)
1100 goto next;
1101
1102 victim_name = kmalloc(victim_name_len, GFP_NOFS);
Josef Bacik36508602013-04-25 16:23:32 -04001103 if (!victim_name)
1104 return -ENOMEM;
Mark Fashehf1863732012-08-08 11:32:27 -07001105 read_extent_buffer(leaf, victim_name, (unsigned long)&extref->name,
1106 victim_name_len);
1107
1108 search_key.objectid = inode_objectid;
1109 search_key.type = BTRFS_INODE_EXTREF_KEY;
1110 search_key.offset = btrfs_extref_hash(parent_objectid,
1111 victim_name,
1112 victim_name_len);
1113 ret = 0;
1114 if (!backref_in_log(log_root, &search_key,
1115 parent_objectid, victim_name,
1116 victim_name_len)) {
1117 ret = -ENOENT;
1118 victim_parent = read_one_inode(root,
Nikolay Borisov94c91a12017-01-18 00:31:46 +02001119 parent_objectid);
Mark Fashehf1863732012-08-08 11:32:27 -07001120 if (victim_parent) {
Nikolay Borisov94c91a12017-01-18 00:31:46 +02001121 inc_nlink(&inode->vfs_inode);
Mark Fashehf1863732012-08-08 11:32:27 -07001122 btrfs_release_path(path);
1123
1124 ret = btrfs_unlink_inode(trans, root,
Nikolay Borisov4ec59342017-01-18 00:31:44 +02001125 BTRFS_I(victim_parent),
Nikolay Borisov94c91a12017-01-18 00:31:46 +02001126 inode,
Nikolay Borisov4ec59342017-01-18 00:31:44 +02001127 victim_name,
1128 victim_name_len);
Filipe David Borba Mananaada9af22013-08-05 09:25:47 +01001129 if (!ret)
1130 ret = btrfs_run_delayed_items(
Jeff Mahoney2ff7e612016-06-22 18:54:24 -04001131 trans,
1132 fs_info);
Mark Fashehf1863732012-08-08 11:32:27 -07001133 }
Mark Fashehf1863732012-08-08 11:32:27 -07001134 iput(victim_parent);
1135 kfree(victim_name);
Josef Bacik36508602013-04-25 16:23:32 -04001136 if (ret)
1137 return ret;
Mark Fashehf1863732012-08-08 11:32:27 -07001138 *search_done = 1;
1139 goto again;
1140 }
1141 kfree(victim_name);
Josef Bacik36508602013-04-25 16:23:32 -04001142 if (ret)
1143 return ret;
Mark Fashehf1863732012-08-08 11:32:27 -07001144next:
1145 cur_offset += victim_name_len + sizeof(*extref);
1146 }
1147 *search_done = 1;
1148 }
1149 btrfs_release_path(path);
1150
Jan Schmidt5a1d7842012-08-17 14:04:41 -07001151 /* look for a conflicting sequence number */
Nikolay Borisov94c91a12017-01-18 00:31:46 +02001152 di = btrfs_lookup_dir_index_item(trans, root, path, btrfs_ino(dir),
Mark Fashehf1863732012-08-08 11:32:27 -07001153 ref_index, name, namelen, 0);
Jan Schmidt5a1d7842012-08-17 14:04:41 -07001154 if (di && !IS_ERR(di)) {
Nikolay Borisov94c91a12017-01-18 00:31:46 +02001155 ret = drop_one_dir_item(trans, root, path, dir, di);
Josef Bacik36508602013-04-25 16:23:32 -04001156 if (ret)
1157 return ret;
Jan Schmidt5a1d7842012-08-17 14:04:41 -07001158 }
1159 btrfs_release_path(path);
1160
1161 /* look for a conflicing name */
Nikolay Borisov94c91a12017-01-18 00:31:46 +02001162 di = btrfs_lookup_dir_item(trans, root, path, btrfs_ino(dir),
Jan Schmidt5a1d7842012-08-17 14:04:41 -07001163 name, namelen, 0);
1164 if (di && !IS_ERR(di)) {
Nikolay Borisov94c91a12017-01-18 00:31:46 +02001165 ret = drop_one_dir_item(trans, root, path, dir, di);
Josef Bacik36508602013-04-25 16:23:32 -04001166 if (ret)
1167 return ret;
Jan Schmidt5a1d7842012-08-17 14:04:41 -07001168 }
1169 btrfs_release_path(path);
1170
1171 return 0;
1172}
Chris Masone02119d2008-09-05 16:13:11 -04001173
Mark Fashehf1863732012-08-08 11:32:27 -07001174static int extref_get_fields(struct extent_buffer *eb, unsigned long ref_ptr,
1175 u32 *namelen, char **name, u64 *index,
1176 u64 *parent_objectid)
1177{
1178 struct btrfs_inode_extref *extref;
1179
1180 extref = (struct btrfs_inode_extref *)ref_ptr;
1181
1182 *namelen = btrfs_inode_extref_name_len(eb, extref);
1183 *name = kmalloc(*namelen, GFP_NOFS);
1184 if (*name == NULL)
1185 return -ENOMEM;
1186
1187 read_extent_buffer(eb, *name, (unsigned long)&extref->name,
1188 *namelen);
1189
1190 *index = btrfs_inode_extref_index(eb, extref);
1191 if (parent_objectid)
1192 *parent_objectid = btrfs_inode_extref_parent(eb, extref);
1193
1194 return 0;
1195}
1196
1197static int ref_get_fields(struct extent_buffer *eb, unsigned long ref_ptr,
1198 u32 *namelen, char **name, u64 *index)
1199{
1200 struct btrfs_inode_ref *ref;
1201
1202 ref = (struct btrfs_inode_ref *)ref_ptr;
1203
1204 *namelen = btrfs_inode_ref_name_len(eb, ref);
1205 *name = kmalloc(*namelen, GFP_NOFS);
1206 if (*name == NULL)
1207 return -ENOMEM;
1208
1209 read_extent_buffer(eb, *name, (unsigned long)(ref + 1), *namelen);
1210
1211 *index = btrfs_inode_ref_index(eb, ref);
1212
1213 return 0;
1214}
1215
Chris Masone02119d2008-09-05 16:13:11 -04001216/*
1217 * replay one inode back reference item found in the log tree.
1218 * eb, slot and key refer to the buffer and key found in the log tree.
1219 * root is the destination we are replaying into, and path is for temp
1220 * use by this function. (it should be released on return).
1221 */
1222static noinline int add_inode_ref(struct btrfs_trans_handle *trans,
1223 struct btrfs_root *root,
1224 struct btrfs_root *log,
1225 struct btrfs_path *path,
1226 struct extent_buffer *eb, int slot,
1227 struct btrfs_key *key)
1228{
Geyslan G. Bem03b2f082013-10-11 15:35:45 -03001229 struct inode *dir = NULL;
1230 struct inode *inode = NULL;
Chris Masone02119d2008-09-05 16:13:11 -04001231 unsigned long ref_ptr;
1232 unsigned long ref_end;
Geyslan G. Bem03b2f082013-10-11 15:35:45 -03001233 char *name = NULL;
liubo34f3e4f2011-08-06 08:35:23 +00001234 int namelen;
1235 int ret;
liuboc622ae62011-03-26 08:01:12 -04001236 int search_done = 0;
Mark Fashehf1863732012-08-08 11:32:27 -07001237 int log_ref_ver = 0;
1238 u64 parent_objectid;
1239 u64 inode_objectid;
Chris Masonf46dbe32012-10-09 11:17:20 -04001240 u64 ref_index = 0;
Mark Fashehf1863732012-08-08 11:32:27 -07001241 int ref_struct_size;
1242
1243 ref_ptr = btrfs_item_ptr_offset(eb, slot);
1244 ref_end = ref_ptr + btrfs_item_size_nr(eb, slot);
1245
1246 if (key->type == BTRFS_INODE_EXTREF_KEY) {
1247 struct btrfs_inode_extref *r;
1248
1249 ref_struct_size = sizeof(struct btrfs_inode_extref);
1250 log_ref_ver = 1;
1251 r = (struct btrfs_inode_extref *)ref_ptr;
1252 parent_objectid = btrfs_inode_extref_parent(eb, r);
1253 } else {
1254 ref_struct_size = sizeof(struct btrfs_inode_ref);
1255 parent_objectid = key->offset;
1256 }
1257 inode_objectid = key->objectid;
Chris Masone02119d2008-09-05 16:13:11 -04001258
Chris Masone02119d2008-09-05 16:13:11 -04001259 /*
1260 * it is possible that we didn't log all the parent directories
1261 * for a given inode. If we don't find the dir, just don't
1262 * copy the back ref in. The link count fixup code will take
1263 * care of the rest
1264 */
Mark Fashehf1863732012-08-08 11:32:27 -07001265 dir = read_one_inode(root, parent_objectid);
Geyslan G. Bem03b2f082013-10-11 15:35:45 -03001266 if (!dir) {
1267 ret = -ENOENT;
1268 goto out;
1269 }
Chris Masone02119d2008-09-05 16:13:11 -04001270
Mark Fashehf1863732012-08-08 11:32:27 -07001271 inode = read_one_inode(root, inode_objectid);
Tsutomu Itohc00e9492011-04-28 09:10:23 +00001272 if (!inode) {
Geyslan G. Bem03b2f082013-10-11 15:35:45 -03001273 ret = -EIO;
1274 goto out;
Tsutomu Itohc00e9492011-04-28 09:10:23 +00001275 }
Chris Masone02119d2008-09-05 16:13:11 -04001276
Jan Schmidt5a1d7842012-08-17 14:04:41 -07001277 while (ref_ptr < ref_end) {
Mark Fashehf1863732012-08-08 11:32:27 -07001278 if (log_ref_ver) {
1279 ret = extref_get_fields(eb, ref_ptr, &namelen, &name,
1280 &ref_index, &parent_objectid);
1281 /*
1282 * parent object can change from one array
1283 * item to another.
1284 */
1285 if (!dir)
1286 dir = read_one_inode(root, parent_objectid);
Geyslan G. Bem03b2f082013-10-11 15:35:45 -03001287 if (!dir) {
1288 ret = -ENOENT;
1289 goto out;
1290 }
Mark Fashehf1863732012-08-08 11:32:27 -07001291 } else {
1292 ret = ref_get_fields(eb, ref_ptr, &namelen, &name,
1293 &ref_index);
1294 }
1295 if (ret)
Geyslan G. Bem03b2f082013-10-11 15:35:45 -03001296 goto out;
Chris Masone02119d2008-09-05 16:13:11 -04001297
Jan Schmidt5a1d7842012-08-17 14:04:41 -07001298 /* if we already have a perfect match, we're done */
Nikolay Borisov4a0cc7c2017-01-10 20:35:31 +02001299 if (!inode_in_dir(root, path, btrfs_ino(BTRFS_I(dir)), btrfs_ino(BTRFS_I(inode)),
Mark Fashehf1863732012-08-08 11:32:27 -07001300 ref_index, name, namelen)) {
Jan Schmidt5a1d7842012-08-17 14:04:41 -07001301 /*
1302 * look for a conflicting back reference in the
1303 * metadata. if we find one we have to unlink that name
1304 * of the file before we add our new link. Later on, we
1305 * overwrite any existing back reference, and we don't
1306 * want to create dangling pointers in the directory.
1307 */
Chris Masone02119d2008-09-05 16:13:11 -04001308
Jan Schmidt5a1d7842012-08-17 14:04:41 -07001309 if (!search_done) {
1310 ret = __add_inode_ref(trans, root, path, log,
Nikolay Borisov94c91a12017-01-18 00:31:46 +02001311 BTRFS_I(dir),
1312 BTRFS_I(inode), eb,
Mark Fashehf1863732012-08-08 11:32:27 -07001313 inode_objectid,
1314 parent_objectid,
1315 ref_index, name, namelen,
Jan Schmidt5a1d7842012-08-17 14:04:41 -07001316 &search_done);
Geyslan G. Bem03b2f082013-10-11 15:35:45 -03001317 if (ret) {
1318 if (ret == 1)
1319 ret = 0;
Jan Schmidt5a1d7842012-08-17 14:04:41 -07001320 goto out;
Josef Bacik36508602013-04-25 16:23:32 -04001321 }
Chris Masone02119d2008-09-05 16:13:11 -04001322 }
Jan Schmidt5a1d7842012-08-17 14:04:41 -07001323
1324 /* insert our name */
1325 ret = btrfs_add_link(trans, dir, inode, name, namelen,
Mark Fashehf1863732012-08-08 11:32:27 -07001326 0, ref_index);
Josef Bacik36508602013-04-25 16:23:32 -04001327 if (ret)
1328 goto out;
Jan Schmidt5a1d7842012-08-17 14:04:41 -07001329
1330 btrfs_update_inode(trans, root, inode);
Chris Masone02119d2008-09-05 16:13:11 -04001331 }
liuboc622ae62011-03-26 08:01:12 -04001332
Mark Fashehf1863732012-08-08 11:32:27 -07001333 ref_ptr = (unsigned long)(ref_ptr + ref_struct_size) + namelen;
Jan Schmidt5a1d7842012-08-17 14:04:41 -07001334 kfree(name);
Geyslan G. Bem03b2f082013-10-11 15:35:45 -03001335 name = NULL;
Mark Fashehf1863732012-08-08 11:32:27 -07001336 if (log_ref_ver) {
1337 iput(dir);
1338 dir = NULL;
1339 }
Chris Masone02119d2008-09-05 16:13:11 -04001340 }
Chris Masone02119d2008-09-05 16:13:11 -04001341
1342 /* finally write the back reference in the inode */
1343 ret = overwrite_item(trans, root, path, eb, slot, key);
Jan Schmidt5a1d7842012-08-17 14:04:41 -07001344out:
David Sterbab3b4aa72011-04-21 01:20:15 +02001345 btrfs_release_path(path);
Geyslan G. Bem03b2f082013-10-11 15:35:45 -03001346 kfree(name);
Chris Masone02119d2008-09-05 16:13:11 -04001347 iput(dir);
1348 iput(inode);
Josef Bacik36508602013-04-25 16:23:32 -04001349 return ret;
Chris Masone02119d2008-09-05 16:13:11 -04001350}
1351
Yan, Zhengc71bf092009-11-12 09:34:40 +00001352static int insert_orphan_item(struct btrfs_trans_handle *trans,
David Sterba9c4f61f2015-01-02 19:12:57 +01001353 struct btrfs_root *root, u64 ino)
Yan, Zhengc71bf092009-11-12 09:34:40 +00001354{
1355 int ret;
David Sterba381cf652015-01-02 18:45:16 +01001356
David Sterba9c4f61f2015-01-02 19:12:57 +01001357 ret = btrfs_insert_orphan_item(trans, root, ino);
1358 if (ret == -EEXIST)
1359 ret = 0;
David Sterba381cf652015-01-02 18:45:16 +01001360
Yan, Zhengc71bf092009-11-12 09:34:40 +00001361 return ret;
1362}
1363
Mark Fashehf1863732012-08-08 11:32:27 -07001364static int count_inode_extrefs(struct btrfs_root *root,
1365 struct inode *inode, struct btrfs_path *path)
Chris Masone02119d2008-09-05 16:13:11 -04001366{
Mark Fashehf1863732012-08-08 11:32:27 -07001367 int ret = 0;
1368 int name_len;
1369 unsigned int nlink = 0;
1370 u32 item_size;
1371 u32 cur_offset = 0;
Nikolay Borisov4a0cc7c2017-01-10 20:35:31 +02001372 u64 inode_objectid = btrfs_ino(BTRFS_I(inode));
Mark Fashehf1863732012-08-08 11:32:27 -07001373 u64 offset = 0;
1374 unsigned long ptr;
1375 struct btrfs_inode_extref *extref;
1376 struct extent_buffer *leaf;
1377
1378 while (1) {
1379 ret = btrfs_find_one_extref(root, inode_objectid, offset, path,
1380 &extref, &offset);
1381 if (ret)
1382 break;
1383
1384 leaf = path->nodes[0];
1385 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1386 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
Filipe Manana2c2c4522015-01-13 16:40:04 +00001387 cur_offset = 0;
Mark Fashehf1863732012-08-08 11:32:27 -07001388
1389 while (cur_offset < item_size) {
1390 extref = (struct btrfs_inode_extref *) (ptr + cur_offset);
1391 name_len = btrfs_inode_extref_name_len(leaf, extref);
1392
1393 nlink++;
1394
1395 cur_offset += name_len + sizeof(*extref);
1396 }
1397
1398 offset++;
1399 btrfs_release_path(path);
1400 }
1401 btrfs_release_path(path);
1402
Filipe Manana2c2c4522015-01-13 16:40:04 +00001403 if (ret < 0 && ret != -ENOENT)
Mark Fashehf1863732012-08-08 11:32:27 -07001404 return ret;
1405 return nlink;
1406}
1407
1408static int count_inode_refs(struct btrfs_root *root,
1409 struct inode *inode, struct btrfs_path *path)
1410{
Chris Masone02119d2008-09-05 16:13:11 -04001411 int ret;
1412 struct btrfs_key key;
Mark Fashehf1863732012-08-08 11:32:27 -07001413 unsigned int nlink = 0;
Chris Masone02119d2008-09-05 16:13:11 -04001414 unsigned long ptr;
1415 unsigned long ptr_end;
1416 int name_len;
Nikolay Borisov4a0cc7c2017-01-10 20:35:31 +02001417 u64 ino = btrfs_ino(BTRFS_I(inode));
Chris Masone02119d2008-09-05 16:13:11 -04001418
Li Zefan33345d012011-04-20 10:31:50 +08001419 key.objectid = ino;
Chris Masone02119d2008-09-05 16:13:11 -04001420 key.type = BTRFS_INODE_REF_KEY;
1421 key.offset = (u64)-1;
1422
Chris Masond3977122009-01-05 21:25:51 -05001423 while (1) {
Chris Masone02119d2008-09-05 16:13:11 -04001424 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1425 if (ret < 0)
1426 break;
1427 if (ret > 0) {
1428 if (path->slots[0] == 0)
1429 break;
1430 path->slots[0]--;
1431 }
Filipe David Borba Mananae93ae262013-10-14 22:49:11 +01001432process_slot:
Chris Masone02119d2008-09-05 16:13:11 -04001433 btrfs_item_key_to_cpu(path->nodes[0], &key,
1434 path->slots[0]);
Li Zefan33345d012011-04-20 10:31:50 +08001435 if (key.objectid != ino ||
Chris Masone02119d2008-09-05 16:13:11 -04001436 key.type != BTRFS_INODE_REF_KEY)
1437 break;
1438 ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
1439 ptr_end = ptr + btrfs_item_size_nr(path->nodes[0],
1440 path->slots[0]);
Chris Masond3977122009-01-05 21:25:51 -05001441 while (ptr < ptr_end) {
Chris Masone02119d2008-09-05 16:13:11 -04001442 struct btrfs_inode_ref *ref;
1443
1444 ref = (struct btrfs_inode_ref *)ptr;
1445 name_len = btrfs_inode_ref_name_len(path->nodes[0],
1446 ref);
1447 ptr = (unsigned long)(ref + 1) + name_len;
1448 nlink++;
1449 }
1450
1451 if (key.offset == 0)
1452 break;
Filipe David Borba Mananae93ae262013-10-14 22:49:11 +01001453 if (path->slots[0] > 0) {
1454 path->slots[0]--;
1455 goto process_slot;
1456 }
Chris Masone02119d2008-09-05 16:13:11 -04001457 key.offset--;
David Sterbab3b4aa72011-04-21 01:20:15 +02001458 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04001459 }
David Sterbab3b4aa72011-04-21 01:20:15 +02001460 btrfs_release_path(path);
Mark Fashehf1863732012-08-08 11:32:27 -07001461
1462 return nlink;
1463}
1464
1465/*
1466 * There are a few corners where the link count of the file can't
1467 * be properly maintained during replay. So, instead of adding
1468 * lots of complexity to the log code, we just scan the backrefs
1469 * for any file that has been through replay.
1470 *
1471 * The scan will update the link count on the inode to reflect the
1472 * number of back refs found. If it goes down to zero, the iput
1473 * will free the inode.
1474 */
1475static noinline int fixup_inode_link_count(struct btrfs_trans_handle *trans,
1476 struct btrfs_root *root,
1477 struct inode *inode)
1478{
1479 struct btrfs_path *path;
1480 int ret;
1481 u64 nlink = 0;
Nikolay Borisov4a0cc7c2017-01-10 20:35:31 +02001482 u64 ino = btrfs_ino(BTRFS_I(inode));
Mark Fashehf1863732012-08-08 11:32:27 -07001483
1484 path = btrfs_alloc_path();
1485 if (!path)
1486 return -ENOMEM;
1487
1488 ret = count_inode_refs(root, inode, path);
1489 if (ret < 0)
1490 goto out;
1491
1492 nlink = ret;
1493
1494 ret = count_inode_extrefs(root, inode, path);
Mark Fashehf1863732012-08-08 11:32:27 -07001495 if (ret < 0)
1496 goto out;
1497
1498 nlink += ret;
1499
1500 ret = 0;
1501
Chris Masone02119d2008-09-05 16:13:11 -04001502 if (nlink != inode->i_nlink) {
Miklos Szeredibfe86842011-10-28 14:13:29 +02001503 set_nlink(inode, nlink);
Chris Masone02119d2008-09-05 16:13:11 -04001504 btrfs_update_inode(trans, root, inode);
1505 }
Chris Mason8d5bf1c2008-09-11 15:51:21 -04001506 BTRFS_I(inode)->index_cnt = (u64)-1;
Chris Masone02119d2008-09-05 16:13:11 -04001507
Yan, Zhengc71bf092009-11-12 09:34:40 +00001508 if (inode->i_nlink == 0) {
1509 if (S_ISDIR(inode->i_mode)) {
1510 ret = replay_dir_deletes(trans, root, NULL, path,
Li Zefan33345d012011-04-20 10:31:50 +08001511 ino, 1);
Josef Bacik36508602013-04-25 16:23:32 -04001512 if (ret)
1513 goto out;
Yan, Zhengc71bf092009-11-12 09:34:40 +00001514 }
Li Zefan33345d012011-04-20 10:31:50 +08001515 ret = insert_orphan_item(trans, root, ino);
Chris Mason12fcfd22009-03-24 10:24:20 -04001516 }
Chris Mason12fcfd22009-03-24 10:24:20 -04001517
Mark Fashehf1863732012-08-08 11:32:27 -07001518out:
1519 btrfs_free_path(path);
1520 return ret;
Chris Masone02119d2008-09-05 16:13:11 -04001521}
1522
1523static noinline int fixup_inode_link_counts(struct btrfs_trans_handle *trans,
1524 struct btrfs_root *root,
1525 struct btrfs_path *path)
1526{
1527 int ret;
1528 struct btrfs_key key;
1529 struct inode *inode;
1530
1531 key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID;
1532 key.type = BTRFS_ORPHAN_ITEM_KEY;
1533 key.offset = (u64)-1;
Chris Masond3977122009-01-05 21:25:51 -05001534 while (1) {
Chris Masone02119d2008-09-05 16:13:11 -04001535 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1536 if (ret < 0)
1537 break;
1538
1539 if (ret == 1) {
1540 if (path->slots[0] == 0)
1541 break;
1542 path->slots[0]--;
1543 }
1544
1545 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1546 if (key.objectid != BTRFS_TREE_LOG_FIXUP_OBJECTID ||
1547 key.type != BTRFS_ORPHAN_ITEM_KEY)
1548 break;
1549
1550 ret = btrfs_del_item(trans, root, path);
Tsutomu Itoh65a246c2011-05-19 04:37:44 +00001551 if (ret)
1552 goto out;
Chris Masone02119d2008-09-05 16:13:11 -04001553
David Sterbab3b4aa72011-04-21 01:20:15 +02001554 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04001555 inode = read_one_inode(root, key.offset);
Tsutomu Itohc00e9492011-04-28 09:10:23 +00001556 if (!inode)
1557 return -EIO;
Chris Masone02119d2008-09-05 16:13:11 -04001558
1559 ret = fixup_inode_link_count(trans, root, inode);
Chris Masone02119d2008-09-05 16:13:11 -04001560 iput(inode);
Josef Bacik36508602013-04-25 16:23:32 -04001561 if (ret)
1562 goto out;
Chris Masone02119d2008-09-05 16:13:11 -04001563
Chris Mason12fcfd22009-03-24 10:24:20 -04001564 /*
1565 * fixup on a directory may create new entries,
1566 * make sure we always look for the highset possible
1567 * offset
1568 */
1569 key.offset = (u64)-1;
Chris Masone02119d2008-09-05 16:13:11 -04001570 }
Tsutomu Itoh65a246c2011-05-19 04:37:44 +00001571 ret = 0;
1572out:
David Sterbab3b4aa72011-04-21 01:20:15 +02001573 btrfs_release_path(path);
Tsutomu Itoh65a246c2011-05-19 04:37:44 +00001574 return ret;
Chris Masone02119d2008-09-05 16:13:11 -04001575}
1576
1577
1578/*
1579 * record a given inode in the fixup dir so we can check its link
1580 * count when replay is done. The link count is incremented here
1581 * so the inode won't go away until we check it
1582 */
1583static noinline int link_to_fixup_dir(struct btrfs_trans_handle *trans,
1584 struct btrfs_root *root,
1585 struct btrfs_path *path,
1586 u64 objectid)
1587{
1588 struct btrfs_key key;
1589 int ret = 0;
1590 struct inode *inode;
1591
1592 inode = read_one_inode(root, objectid);
Tsutomu Itohc00e9492011-04-28 09:10:23 +00001593 if (!inode)
1594 return -EIO;
Chris Masone02119d2008-09-05 16:13:11 -04001595
1596 key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID;
David Sterba962a2982014-06-04 18:41:45 +02001597 key.type = BTRFS_ORPHAN_ITEM_KEY;
Chris Masone02119d2008-09-05 16:13:11 -04001598 key.offset = objectid;
1599
1600 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1601
David Sterbab3b4aa72011-04-21 01:20:15 +02001602 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04001603 if (ret == 0) {
Josef Bacik9bf7a482013-03-01 13:35:47 -05001604 if (!inode->i_nlink)
1605 set_nlink(inode, 1);
1606 else
Zach Brown8b558c52013-10-16 12:10:34 -07001607 inc_nlink(inode);
Tsutomu Itohb9959292012-06-25 21:25:22 -06001608 ret = btrfs_update_inode(trans, root, inode);
Chris Masone02119d2008-09-05 16:13:11 -04001609 } else if (ret == -EEXIST) {
1610 ret = 0;
1611 } else {
Josef Bacik36508602013-04-25 16:23:32 -04001612 BUG(); /* Logic Error */
Chris Masone02119d2008-09-05 16:13:11 -04001613 }
1614 iput(inode);
1615
1616 return ret;
1617}
1618
1619/*
1620 * when replaying the log for a directory, we only insert names
1621 * for inodes that actually exist. This means an fsync on a directory
1622 * does not implicitly fsync all the new files in it
1623 */
1624static noinline int insert_one_name(struct btrfs_trans_handle *trans,
1625 struct btrfs_root *root,
Chris Masone02119d2008-09-05 16:13:11 -04001626 u64 dirid, u64 index,
Zhaolei60d53eb2015-08-17 18:44:46 +08001627 char *name, int name_len,
Chris Masone02119d2008-09-05 16:13:11 -04001628 struct btrfs_key *location)
1629{
1630 struct inode *inode;
1631 struct inode *dir;
1632 int ret;
1633
1634 inode = read_one_inode(root, location->objectid);
1635 if (!inode)
1636 return -ENOENT;
1637
1638 dir = read_one_inode(root, dirid);
1639 if (!dir) {
1640 iput(inode);
1641 return -EIO;
1642 }
Josef Bacikd5554382013-09-11 14:17:00 -04001643
Chris Masone02119d2008-09-05 16:13:11 -04001644 ret = btrfs_add_link(trans, dir, inode, name, name_len, 1, index);
1645
1646 /* FIXME, put inode into FIXUP list */
1647
1648 iput(inode);
1649 iput(dir);
1650 return ret;
1651}
1652
1653/*
Filipe Mananadf8d1162015-01-14 01:52:25 +00001654 * Return true if an inode reference exists in the log for the given name,
1655 * inode and parent inode.
1656 */
1657static bool name_in_log_ref(struct btrfs_root *log_root,
1658 const char *name, const int name_len,
1659 const u64 dirid, const u64 ino)
1660{
1661 struct btrfs_key search_key;
1662
1663 search_key.objectid = ino;
1664 search_key.type = BTRFS_INODE_REF_KEY;
1665 search_key.offset = dirid;
1666 if (backref_in_log(log_root, &search_key, dirid, name, name_len))
1667 return true;
1668
1669 search_key.type = BTRFS_INODE_EXTREF_KEY;
1670 search_key.offset = btrfs_extref_hash(dirid, name, name_len);
1671 if (backref_in_log(log_root, &search_key, dirid, name, name_len))
1672 return true;
1673
1674 return false;
1675}
1676
1677/*
Chris Masone02119d2008-09-05 16:13:11 -04001678 * take a single entry in a log directory item and replay it into
1679 * the subvolume.
1680 *
1681 * if a conflicting item exists in the subdirectory already,
1682 * the inode it points to is unlinked and put into the link count
1683 * fix up tree.
1684 *
1685 * If a name from the log points to a file or directory that does
1686 * not exist in the FS, it is skipped. fsyncs on directories
1687 * do not force down inodes inside that directory, just changes to the
1688 * names or unlinks in a directory.
Filipe Mananabb53eda2015-07-15 23:26:43 +01001689 *
1690 * Returns < 0 on error, 0 if the name wasn't replayed (dentry points to a
1691 * non-existing inode) and 1 if the name was replayed.
Chris Masone02119d2008-09-05 16:13:11 -04001692 */
1693static noinline int replay_one_name(struct btrfs_trans_handle *trans,
1694 struct btrfs_root *root,
1695 struct btrfs_path *path,
1696 struct extent_buffer *eb,
1697 struct btrfs_dir_item *di,
1698 struct btrfs_key *key)
1699{
1700 char *name;
1701 int name_len;
1702 struct btrfs_dir_item *dst_di;
1703 struct btrfs_key found_key;
1704 struct btrfs_key log_key;
1705 struct inode *dir;
Chris Masone02119d2008-09-05 16:13:11 -04001706 u8 log_type;
Chris Mason4bef0842008-09-08 11:18:08 -04001707 int exists;
Josef Bacik36508602013-04-25 16:23:32 -04001708 int ret = 0;
Josef Bacikd5554382013-09-11 14:17:00 -04001709 bool update_size = (key->type == BTRFS_DIR_INDEX_KEY);
Filipe Mananabb53eda2015-07-15 23:26:43 +01001710 bool name_added = false;
Chris Masone02119d2008-09-05 16:13:11 -04001711
1712 dir = read_one_inode(root, key->objectid);
Tsutomu Itohc00e9492011-04-28 09:10:23 +00001713 if (!dir)
1714 return -EIO;
Chris Masone02119d2008-09-05 16:13:11 -04001715
1716 name_len = btrfs_dir_name_len(eb, di);
1717 name = kmalloc(name_len, GFP_NOFS);
Filipe David Borba Manana2bac3252013-08-04 19:58:57 +01001718 if (!name) {
1719 ret = -ENOMEM;
1720 goto out;
1721 }
liubo2a29edc2011-01-26 06:22:08 +00001722
Chris Masone02119d2008-09-05 16:13:11 -04001723 log_type = btrfs_dir_type(eb, di);
1724 read_extent_buffer(eb, name, (unsigned long)(di + 1),
1725 name_len);
1726
1727 btrfs_dir_item_key_to_cpu(eb, di, &log_key);
Chris Mason4bef0842008-09-08 11:18:08 -04001728 exists = btrfs_lookup_inode(trans, root, path, &log_key, 0);
1729 if (exists == 0)
1730 exists = 1;
1731 else
1732 exists = 0;
David Sterbab3b4aa72011-04-21 01:20:15 +02001733 btrfs_release_path(path);
Chris Mason4bef0842008-09-08 11:18:08 -04001734
Chris Masone02119d2008-09-05 16:13:11 -04001735 if (key->type == BTRFS_DIR_ITEM_KEY) {
1736 dst_di = btrfs_lookup_dir_item(trans, root, path, key->objectid,
1737 name, name_len, 1);
Chris Masond3977122009-01-05 21:25:51 -05001738 } else if (key->type == BTRFS_DIR_INDEX_KEY) {
Chris Masone02119d2008-09-05 16:13:11 -04001739 dst_di = btrfs_lookup_dir_index_item(trans, root, path,
1740 key->objectid,
1741 key->offset, name,
1742 name_len, 1);
1743 } else {
Josef Bacik36508602013-04-25 16:23:32 -04001744 /* Corruption */
1745 ret = -EINVAL;
1746 goto out;
Chris Masone02119d2008-09-05 16:13:11 -04001747 }
David Sterbac7040052011-04-19 18:00:01 +02001748 if (IS_ERR_OR_NULL(dst_di)) {
Chris Masone02119d2008-09-05 16:13:11 -04001749 /* we need a sequence number to insert, so we only
1750 * do inserts for the BTRFS_DIR_INDEX_KEY types
1751 */
1752 if (key->type != BTRFS_DIR_INDEX_KEY)
1753 goto out;
1754 goto insert;
1755 }
1756
1757 btrfs_dir_item_key_to_cpu(path->nodes[0], dst_di, &found_key);
1758 /* the existing item matches the logged item */
1759 if (found_key.objectid == log_key.objectid &&
1760 found_key.type == log_key.type &&
1761 found_key.offset == log_key.offset &&
1762 btrfs_dir_type(path->nodes[0], dst_di) == log_type) {
Filipe Mananaa2cc11d2014-09-08 22:53:18 +01001763 update_size = false;
Chris Masone02119d2008-09-05 16:13:11 -04001764 goto out;
1765 }
1766
1767 /*
1768 * don't drop the conflicting directory entry if the inode
1769 * for the new entry doesn't exist
1770 */
Chris Mason4bef0842008-09-08 11:18:08 -04001771 if (!exists)
Chris Masone02119d2008-09-05 16:13:11 -04001772 goto out;
1773
Nikolay Borisov207e7d92017-01-18 00:31:45 +02001774 ret = drop_one_dir_item(trans, root, path, BTRFS_I(dir), dst_di);
Josef Bacik36508602013-04-25 16:23:32 -04001775 if (ret)
1776 goto out;
Chris Masone02119d2008-09-05 16:13:11 -04001777
1778 if (key->type == BTRFS_DIR_INDEX_KEY)
1779 goto insert;
1780out:
David Sterbab3b4aa72011-04-21 01:20:15 +02001781 btrfs_release_path(path);
Josef Bacikd5554382013-09-11 14:17:00 -04001782 if (!ret && update_size) {
1783 btrfs_i_size_write(dir, dir->i_size + name_len * 2);
1784 ret = btrfs_update_inode(trans, root, dir);
1785 }
Chris Masone02119d2008-09-05 16:13:11 -04001786 kfree(name);
1787 iput(dir);
Filipe Mananabb53eda2015-07-15 23:26:43 +01001788 if (!ret && name_added)
1789 ret = 1;
Josef Bacik36508602013-04-25 16:23:32 -04001790 return ret;
Chris Masone02119d2008-09-05 16:13:11 -04001791
1792insert:
Filipe Mananadf8d1162015-01-14 01:52:25 +00001793 if (name_in_log_ref(root->log_root, name, name_len,
1794 key->objectid, log_key.objectid)) {
1795 /* The dentry will be added later. */
1796 ret = 0;
1797 update_size = false;
1798 goto out;
1799 }
David Sterbab3b4aa72011-04-21 01:20:15 +02001800 btrfs_release_path(path);
Zhaolei60d53eb2015-08-17 18:44:46 +08001801 ret = insert_one_name(trans, root, key->objectid, key->offset,
1802 name, name_len, &log_key);
Filipe Mananadf8d1162015-01-14 01:52:25 +00001803 if (ret && ret != -ENOENT && ret != -EEXIST)
Josef Bacik36508602013-04-25 16:23:32 -04001804 goto out;
Filipe Mananabb53eda2015-07-15 23:26:43 +01001805 if (!ret)
1806 name_added = true;
Josef Bacikd5554382013-09-11 14:17:00 -04001807 update_size = false;
Josef Bacik36508602013-04-25 16:23:32 -04001808 ret = 0;
Chris Masone02119d2008-09-05 16:13:11 -04001809 goto out;
1810}
1811
1812/*
1813 * find all the names in a directory item and reconcile them into
1814 * the subvolume. Only BTRFS_DIR_ITEM_KEY types will have more than
1815 * one name in a directory item, but the same code gets used for
1816 * both directory index types
1817 */
1818static noinline int replay_one_dir_item(struct btrfs_trans_handle *trans,
1819 struct btrfs_root *root,
1820 struct btrfs_path *path,
1821 struct extent_buffer *eb, int slot,
1822 struct btrfs_key *key)
1823{
Jeff Mahoney2ff7e612016-06-22 18:54:24 -04001824 struct btrfs_fs_info *fs_info = root->fs_info;
Filipe Mananabb53eda2015-07-15 23:26:43 +01001825 int ret = 0;
Chris Masone02119d2008-09-05 16:13:11 -04001826 u32 item_size = btrfs_item_size_nr(eb, slot);
1827 struct btrfs_dir_item *di;
1828 int name_len;
1829 unsigned long ptr;
1830 unsigned long ptr_end;
Filipe Mananabb53eda2015-07-15 23:26:43 +01001831 struct btrfs_path *fixup_path = NULL;
Chris Masone02119d2008-09-05 16:13:11 -04001832
1833 ptr = btrfs_item_ptr_offset(eb, slot);
1834 ptr_end = ptr + item_size;
Chris Masond3977122009-01-05 21:25:51 -05001835 while (ptr < ptr_end) {
Chris Masone02119d2008-09-05 16:13:11 -04001836 di = (struct btrfs_dir_item *)ptr;
Jeff Mahoney2ff7e612016-06-22 18:54:24 -04001837 if (verify_dir_item(fs_info, eb, di))
Josef Bacik22a94d42011-03-16 16:47:17 -04001838 return -EIO;
Chris Masone02119d2008-09-05 16:13:11 -04001839 name_len = btrfs_dir_name_len(eb, di);
1840 ret = replay_one_name(trans, root, path, eb, di, key);
Filipe Mananabb53eda2015-07-15 23:26:43 +01001841 if (ret < 0)
1842 break;
Chris Masone02119d2008-09-05 16:13:11 -04001843 ptr = (unsigned long)(di + 1);
1844 ptr += name_len;
Filipe Mananabb53eda2015-07-15 23:26:43 +01001845
1846 /*
1847 * If this entry refers to a non-directory (directories can not
1848 * have a link count > 1) and it was added in the transaction
1849 * that was not committed, make sure we fixup the link count of
1850 * the inode it the entry points to. Otherwise something like
1851 * the following would result in a directory pointing to an
1852 * inode with a wrong link that does not account for this dir
1853 * entry:
1854 *
1855 * mkdir testdir
1856 * touch testdir/foo
1857 * touch testdir/bar
1858 * sync
1859 *
1860 * ln testdir/bar testdir/bar_link
1861 * ln testdir/foo testdir/foo_link
1862 * xfs_io -c "fsync" testdir/bar
1863 *
1864 * <power failure>
1865 *
1866 * mount fs, log replay happens
1867 *
1868 * File foo would remain with a link count of 1 when it has two
1869 * entries pointing to it in the directory testdir. This would
1870 * make it impossible to ever delete the parent directory has
1871 * it would result in stale dentries that can never be deleted.
1872 */
1873 if (ret == 1 && btrfs_dir_type(eb, di) != BTRFS_FT_DIR) {
1874 struct btrfs_key di_key;
1875
1876 if (!fixup_path) {
1877 fixup_path = btrfs_alloc_path();
1878 if (!fixup_path) {
1879 ret = -ENOMEM;
1880 break;
1881 }
1882 }
1883
1884 btrfs_dir_item_key_to_cpu(eb, di, &di_key);
1885 ret = link_to_fixup_dir(trans, root, fixup_path,
1886 di_key.objectid);
1887 if (ret)
1888 break;
1889 }
1890 ret = 0;
Chris Masone02119d2008-09-05 16:13:11 -04001891 }
Filipe Mananabb53eda2015-07-15 23:26:43 +01001892 btrfs_free_path(fixup_path);
1893 return ret;
Chris Masone02119d2008-09-05 16:13:11 -04001894}
1895
1896/*
1897 * directory replay has two parts. There are the standard directory
1898 * items in the log copied from the subvolume, and range items
1899 * created in the log while the subvolume was logged.
1900 *
1901 * The range items tell us which parts of the key space the log
1902 * is authoritative for. During replay, if a key in the subvolume
1903 * directory is in a logged range item, but not actually in the log
1904 * that means it was deleted from the directory before the fsync
1905 * and should be removed.
1906 */
1907static noinline int find_dir_range(struct btrfs_root *root,
1908 struct btrfs_path *path,
1909 u64 dirid, int key_type,
1910 u64 *start_ret, u64 *end_ret)
1911{
1912 struct btrfs_key key;
1913 u64 found_end;
1914 struct btrfs_dir_log_item *item;
1915 int ret;
1916 int nritems;
1917
1918 if (*start_ret == (u64)-1)
1919 return 1;
1920
1921 key.objectid = dirid;
1922 key.type = key_type;
1923 key.offset = *start_ret;
1924
1925 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1926 if (ret < 0)
1927 goto out;
1928 if (ret > 0) {
1929 if (path->slots[0] == 0)
1930 goto out;
1931 path->slots[0]--;
1932 }
1933 if (ret != 0)
1934 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1935
1936 if (key.type != key_type || key.objectid != dirid) {
1937 ret = 1;
1938 goto next;
1939 }
1940 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
1941 struct btrfs_dir_log_item);
1942 found_end = btrfs_dir_log_end(path->nodes[0], item);
1943
1944 if (*start_ret >= key.offset && *start_ret <= found_end) {
1945 ret = 0;
1946 *start_ret = key.offset;
1947 *end_ret = found_end;
1948 goto out;
1949 }
1950 ret = 1;
1951next:
1952 /* check the next slot in the tree to see if it is a valid item */
1953 nritems = btrfs_header_nritems(path->nodes[0]);
Robbie Ko2a7bf532016-10-07 17:30:47 +08001954 path->slots[0]++;
Chris Masone02119d2008-09-05 16:13:11 -04001955 if (path->slots[0] >= nritems) {
1956 ret = btrfs_next_leaf(root, path);
1957 if (ret)
1958 goto out;
Chris Masone02119d2008-09-05 16:13:11 -04001959 }
1960
1961 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1962
1963 if (key.type != key_type || key.objectid != dirid) {
1964 ret = 1;
1965 goto out;
1966 }
1967 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
1968 struct btrfs_dir_log_item);
1969 found_end = btrfs_dir_log_end(path->nodes[0], item);
1970 *start_ret = key.offset;
1971 *end_ret = found_end;
1972 ret = 0;
1973out:
David Sterbab3b4aa72011-04-21 01:20:15 +02001974 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04001975 return ret;
1976}
1977
1978/*
1979 * this looks for a given directory item in the log. If the directory
1980 * item is not in the log, the item is removed and the inode it points
1981 * to is unlinked
1982 */
1983static noinline int check_item_in_log(struct btrfs_trans_handle *trans,
1984 struct btrfs_root *root,
1985 struct btrfs_root *log,
1986 struct btrfs_path *path,
1987 struct btrfs_path *log_path,
1988 struct inode *dir,
1989 struct btrfs_key *dir_key)
1990{
Jeff Mahoney2ff7e612016-06-22 18:54:24 -04001991 struct btrfs_fs_info *fs_info = root->fs_info;
Chris Masone02119d2008-09-05 16:13:11 -04001992 int ret;
1993 struct extent_buffer *eb;
1994 int slot;
1995 u32 item_size;
1996 struct btrfs_dir_item *di;
1997 struct btrfs_dir_item *log_di;
1998 int name_len;
1999 unsigned long ptr;
2000 unsigned long ptr_end;
2001 char *name;
2002 struct inode *inode;
2003 struct btrfs_key location;
2004
2005again:
2006 eb = path->nodes[0];
2007 slot = path->slots[0];
2008 item_size = btrfs_item_size_nr(eb, slot);
2009 ptr = btrfs_item_ptr_offset(eb, slot);
2010 ptr_end = ptr + item_size;
Chris Masond3977122009-01-05 21:25:51 -05002011 while (ptr < ptr_end) {
Chris Masone02119d2008-09-05 16:13:11 -04002012 di = (struct btrfs_dir_item *)ptr;
Jeff Mahoney2ff7e612016-06-22 18:54:24 -04002013 if (verify_dir_item(fs_info, eb, di)) {
Josef Bacik22a94d42011-03-16 16:47:17 -04002014 ret = -EIO;
2015 goto out;
2016 }
2017
Chris Masone02119d2008-09-05 16:13:11 -04002018 name_len = btrfs_dir_name_len(eb, di);
2019 name = kmalloc(name_len, GFP_NOFS);
2020 if (!name) {
2021 ret = -ENOMEM;
2022 goto out;
2023 }
2024 read_extent_buffer(eb, name, (unsigned long)(di + 1),
2025 name_len);
2026 log_di = NULL;
Chris Mason12fcfd22009-03-24 10:24:20 -04002027 if (log && dir_key->type == BTRFS_DIR_ITEM_KEY) {
Chris Masone02119d2008-09-05 16:13:11 -04002028 log_di = btrfs_lookup_dir_item(trans, log, log_path,
2029 dir_key->objectid,
2030 name, name_len, 0);
Chris Mason12fcfd22009-03-24 10:24:20 -04002031 } else if (log && dir_key->type == BTRFS_DIR_INDEX_KEY) {
Chris Masone02119d2008-09-05 16:13:11 -04002032 log_di = btrfs_lookup_dir_index_item(trans, log,
2033 log_path,
2034 dir_key->objectid,
2035 dir_key->offset,
2036 name, name_len, 0);
2037 }
Filipe David Borba Manana269d0402013-10-28 17:39:21 +00002038 if (!log_di || (IS_ERR(log_di) && PTR_ERR(log_di) == -ENOENT)) {
Chris Masone02119d2008-09-05 16:13:11 -04002039 btrfs_dir_item_key_to_cpu(eb, di, &location);
David Sterbab3b4aa72011-04-21 01:20:15 +02002040 btrfs_release_path(path);
2041 btrfs_release_path(log_path);
Chris Masone02119d2008-09-05 16:13:11 -04002042 inode = read_one_inode(root, location.objectid);
Tsutomu Itohc00e9492011-04-28 09:10:23 +00002043 if (!inode) {
2044 kfree(name);
2045 return -EIO;
2046 }
Chris Masone02119d2008-09-05 16:13:11 -04002047
2048 ret = link_to_fixup_dir(trans, root,
2049 path, location.objectid);
Josef Bacik36508602013-04-25 16:23:32 -04002050 if (ret) {
2051 kfree(name);
2052 iput(inode);
2053 goto out;
2054 }
2055
Zach Brown8b558c52013-10-16 12:10:34 -07002056 inc_nlink(inode);
Nikolay Borisov4ec59342017-01-18 00:31:44 +02002057 ret = btrfs_unlink_inode(trans, root, BTRFS_I(dir),
2058 BTRFS_I(inode), name, name_len);
Josef Bacik36508602013-04-25 16:23:32 -04002059 if (!ret)
Jeff Mahoney2ff7e612016-06-22 18:54:24 -04002060 ret = btrfs_run_delayed_items(trans, fs_info);
Chris Masone02119d2008-09-05 16:13:11 -04002061 kfree(name);
2062 iput(inode);
Josef Bacik36508602013-04-25 16:23:32 -04002063 if (ret)
2064 goto out;
Chris Masone02119d2008-09-05 16:13:11 -04002065
2066 /* there might still be more names under this key
2067 * check and repeat if required
2068 */
2069 ret = btrfs_search_slot(NULL, root, dir_key, path,
2070 0, 0);
2071 if (ret == 0)
2072 goto again;
2073 ret = 0;
2074 goto out;
Filipe David Borba Manana269d0402013-10-28 17:39:21 +00002075 } else if (IS_ERR(log_di)) {
2076 kfree(name);
2077 return PTR_ERR(log_di);
Chris Masone02119d2008-09-05 16:13:11 -04002078 }
David Sterbab3b4aa72011-04-21 01:20:15 +02002079 btrfs_release_path(log_path);
Chris Masone02119d2008-09-05 16:13:11 -04002080 kfree(name);
2081
2082 ptr = (unsigned long)(di + 1);
2083 ptr += name_len;
2084 }
2085 ret = 0;
2086out:
David Sterbab3b4aa72011-04-21 01:20:15 +02002087 btrfs_release_path(path);
2088 btrfs_release_path(log_path);
Chris Masone02119d2008-09-05 16:13:11 -04002089 return ret;
2090}
2091
Filipe Manana4f764e52015-02-23 19:53:35 +00002092static int replay_xattr_deletes(struct btrfs_trans_handle *trans,
2093 struct btrfs_root *root,
2094 struct btrfs_root *log,
2095 struct btrfs_path *path,
2096 const u64 ino)
2097{
2098 struct btrfs_key search_key;
2099 struct btrfs_path *log_path;
2100 int i;
2101 int nritems;
2102 int ret;
2103
2104 log_path = btrfs_alloc_path();
2105 if (!log_path)
2106 return -ENOMEM;
2107
2108 search_key.objectid = ino;
2109 search_key.type = BTRFS_XATTR_ITEM_KEY;
2110 search_key.offset = 0;
2111again:
2112 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
2113 if (ret < 0)
2114 goto out;
2115process_leaf:
2116 nritems = btrfs_header_nritems(path->nodes[0]);
2117 for (i = path->slots[0]; i < nritems; i++) {
2118 struct btrfs_key key;
2119 struct btrfs_dir_item *di;
2120 struct btrfs_dir_item *log_di;
2121 u32 total_size;
2122 u32 cur;
2123
2124 btrfs_item_key_to_cpu(path->nodes[0], &key, i);
2125 if (key.objectid != ino || key.type != BTRFS_XATTR_ITEM_KEY) {
2126 ret = 0;
2127 goto out;
2128 }
2129
2130 di = btrfs_item_ptr(path->nodes[0], i, struct btrfs_dir_item);
2131 total_size = btrfs_item_size_nr(path->nodes[0], i);
2132 cur = 0;
2133 while (cur < total_size) {
2134 u16 name_len = btrfs_dir_name_len(path->nodes[0], di);
2135 u16 data_len = btrfs_dir_data_len(path->nodes[0], di);
2136 u32 this_len = sizeof(*di) + name_len + data_len;
2137 char *name;
2138
2139 name = kmalloc(name_len, GFP_NOFS);
2140 if (!name) {
2141 ret = -ENOMEM;
2142 goto out;
2143 }
2144 read_extent_buffer(path->nodes[0], name,
2145 (unsigned long)(di + 1), name_len);
2146
2147 log_di = btrfs_lookup_xattr(NULL, log, log_path, ino,
2148 name, name_len, 0);
2149 btrfs_release_path(log_path);
2150 if (!log_di) {
2151 /* Doesn't exist in log tree, so delete it. */
2152 btrfs_release_path(path);
2153 di = btrfs_lookup_xattr(trans, root, path, ino,
2154 name, name_len, -1);
2155 kfree(name);
2156 if (IS_ERR(di)) {
2157 ret = PTR_ERR(di);
2158 goto out;
2159 }
2160 ASSERT(di);
2161 ret = btrfs_delete_one_dir_name(trans, root,
2162 path, di);
2163 if (ret)
2164 goto out;
2165 btrfs_release_path(path);
2166 search_key = key;
2167 goto again;
2168 }
2169 kfree(name);
2170 if (IS_ERR(log_di)) {
2171 ret = PTR_ERR(log_di);
2172 goto out;
2173 }
2174 cur += this_len;
2175 di = (struct btrfs_dir_item *)((char *)di + this_len);
2176 }
2177 }
2178 ret = btrfs_next_leaf(root, path);
2179 if (ret > 0)
2180 ret = 0;
2181 else if (ret == 0)
2182 goto process_leaf;
2183out:
2184 btrfs_free_path(log_path);
2185 btrfs_release_path(path);
2186 return ret;
2187}
2188
2189
Chris Masone02119d2008-09-05 16:13:11 -04002190/*
2191 * deletion replay happens before we copy any new directory items
2192 * out of the log or out of backreferences from inodes. It
2193 * scans the log to find ranges of keys that log is authoritative for,
2194 * and then scans the directory to find items in those ranges that are
2195 * not present in the log.
2196 *
2197 * Anything we don't find in the log is unlinked and removed from the
2198 * directory.
2199 */
2200static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans,
2201 struct btrfs_root *root,
2202 struct btrfs_root *log,
2203 struct btrfs_path *path,
Chris Mason12fcfd22009-03-24 10:24:20 -04002204 u64 dirid, int del_all)
Chris Masone02119d2008-09-05 16:13:11 -04002205{
2206 u64 range_start;
2207 u64 range_end;
2208 int key_type = BTRFS_DIR_LOG_ITEM_KEY;
2209 int ret = 0;
2210 struct btrfs_key dir_key;
2211 struct btrfs_key found_key;
2212 struct btrfs_path *log_path;
2213 struct inode *dir;
2214
2215 dir_key.objectid = dirid;
2216 dir_key.type = BTRFS_DIR_ITEM_KEY;
2217 log_path = btrfs_alloc_path();
2218 if (!log_path)
2219 return -ENOMEM;
2220
2221 dir = read_one_inode(root, dirid);
2222 /* it isn't an error if the inode isn't there, that can happen
2223 * because we replay the deletes before we copy in the inode item
2224 * from the log
2225 */
2226 if (!dir) {
2227 btrfs_free_path(log_path);
2228 return 0;
2229 }
2230again:
2231 range_start = 0;
2232 range_end = 0;
Chris Masond3977122009-01-05 21:25:51 -05002233 while (1) {
Chris Mason12fcfd22009-03-24 10:24:20 -04002234 if (del_all)
2235 range_end = (u64)-1;
2236 else {
2237 ret = find_dir_range(log, path, dirid, key_type,
2238 &range_start, &range_end);
2239 if (ret != 0)
2240 break;
2241 }
Chris Masone02119d2008-09-05 16:13:11 -04002242
2243 dir_key.offset = range_start;
Chris Masond3977122009-01-05 21:25:51 -05002244 while (1) {
Chris Masone02119d2008-09-05 16:13:11 -04002245 int nritems;
2246 ret = btrfs_search_slot(NULL, root, &dir_key, path,
2247 0, 0);
2248 if (ret < 0)
2249 goto out;
2250
2251 nritems = btrfs_header_nritems(path->nodes[0]);
2252 if (path->slots[0] >= nritems) {
2253 ret = btrfs_next_leaf(root, path);
2254 if (ret)
2255 break;
2256 }
2257 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2258 path->slots[0]);
2259 if (found_key.objectid != dirid ||
2260 found_key.type != dir_key.type)
2261 goto next_type;
2262
2263 if (found_key.offset > range_end)
2264 break;
2265
2266 ret = check_item_in_log(trans, root, log, path,
Chris Mason12fcfd22009-03-24 10:24:20 -04002267 log_path, dir,
2268 &found_key);
Josef Bacik36508602013-04-25 16:23:32 -04002269 if (ret)
2270 goto out;
Chris Masone02119d2008-09-05 16:13:11 -04002271 if (found_key.offset == (u64)-1)
2272 break;
2273 dir_key.offset = found_key.offset + 1;
2274 }
David Sterbab3b4aa72011-04-21 01:20:15 +02002275 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04002276 if (range_end == (u64)-1)
2277 break;
2278 range_start = range_end + 1;
2279 }
2280
2281next_type:
2282 ret = 0;
2283 if (key_type == BTRFS_DIR_LOG_ITEM_KEY) {
2284 key_type = BTRFS_DIR_LOG_INDEX_KEY;
2285 dir_key.type = BTRFS_DIR_INDEX_KEY;
David Sterbab3b4aa72011-04-21 01:20:15 +02002286 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04002287 goto again;
2288 }
2289out:
David Sterbab3b4aa72011-04-21 01:20:15 +02002290 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04002291 btrfs_free_path(log_path);
2292 iput(dir);
2293 return ret;
2294}
2295
2296/*
2297 * the process_func used to replay items from the log tree. This
2298 * gets called in two different stages. The first stage just looks
2299 * for inodes and makes sure they are all copied into the subvolume.
2300 *
2301 * The second stage copies all the other item types from the log into
2302 * the subvolume. The two stage approach is slower, but gets rid of
2303 * lots of complexity around inodes referencing other inodes that exist
2304 * only in the log (references come from either directory items or inode
2305 * back refs).
2306 */
2307static int replay_one_buffer(struct btrfs_root *log, struct extent_buffer *eb,
2308 struct walk_control *wc, u64 gen)
2309{
2310 int nritems;
2311 struct btrfs_path *path;
2312 struct btrfs_root *root = wc->replay_dest;
2313 struct btrfs_key key;
Chris Masone02119d2008-09-05 16:13:11 -04002314 int level;
2315 int i;
2316 int ret;
2317
Tsutomu Itoh018642a2012-05-29 18:10:13 +09002318 ret = btrfs_read_buffer(eb, gen);
2319 if (ret)
2320 return ret;
Chris Masone02119d2008-09-05 16:13:11 -04002321
2322 level = btrfs_header_level(eb);
2323
2324 if (level != 0)
2325 return 0;
2326
2327 path = btrfs_alloc_path();
Mark Fasheh1e5063d2011-07-12 10:46:06 -07002328 if (!path)
2329 return -ENOMEM;
Chris Masone02119d2008-09-05 16:13:11 -04002330
2331 nritems = btrfs_header_nritems(eb);
2332 for (i = 0; i < nritems; i++) {
2333 btrfs_item_key_to_cpu(eb, &key, i);
Chris Masone02119d2008-09-05 16:13:11 -04002334
2335 /* inode keys are done during the first stage */
2336 if (key.type == BTRFS_INODE_ITEM_KEY &&
2337 wc->stage == LOG_WALK_REPLAY_INODES) {
Chris Masone02119d2008-09-05 16:13:11 -04002338 struct btrfs_inode_item *inode_item;
2339 u32 mode;
2340
2341 inode_item = btrfs_item_ptr(eb, i,
2342 struct btrfs_inode_item);
Filipe Manana4f764e52015-02-23 19:53:35 +00002343 ret = replay_xattr_deletes(wc->trans, root, log,
2344 path, key.objectid);
2345 if (ret)
2346 break;
Chris Masone02119d2008-09-05 16:13:11 -04002347 mode = btrfs_inode_mode(eb, inode_item);
2348 if (S_ISDIR(mode)) {
2349 ret = replay_dir_deletes(wc->trans,
Chris Mason12fcfd22009-03-24 10:24:20 -04002350 root, log, path, key.objectid, 0);
Josef Bacikb50c6e22013-04-25 15:55:30 -04002351 if (ret)
2352 break;
Chris Masone02119d2008-09-05 16:13:11 -04002353 }
2354 ret = overwrite_item(wc->trans, root, path,
2355 eb, i, &key);
Josef Bacikb50c6e22013-04-25 15:55:30 -04002356 if (ret)
2357 break;
Chris Masone02119d2008-09-05 16:13:11 -04002358
Yan, Zhengc71bf092009-11-12 09:34:40 +00002359 /* for regular files, make sure corresponding
Nicholas D Steeves01327612016-05-19 21:18:45 -04002360 * orphan item exist. extents past the new EOF
Yan, Zhengc71bf092009-11-12 09:34:40 +00002361 * will be truncated later by orphan cleanup.
Chris Masone02119d2008-09-05 16:13:11 -04002362 */
2363 if (S_ISREG(mode)) {
Yan, Zhengc71bf092009-11-12 09:34:40 +00002364 ret = insert_orphan_item(wc->trans, root,
2365 key.objectid);
Josef Bacikb50c6e22013-04-25 15:55:30 -04002366 if (ret)
2367 break;
Chris Masone02119d2008-09-05 16:13:11 -04002368 }
Yan, Zhengc71bf092009-11-12 09:34:40 +00002369
Chris Masone02119d2008-09-05 16:13:11 -04002370 ret = link_to_fixup_dir(wc->trans, root,
2371 path, key.objectid);
Josef Bacikb50c6e22013-04-25 15:55:30 -04002372 if (ret)
2373 break;
Chris Masone02119d2008-09-05 16:13:11 -04002374 }
Josef Bacikdd8e7212013-09-11 11:57:23 -04002375
2376 if (key.type == BTRFS_DIR_INDEX_KEY &&
2377 wc->stage == LOG_WALK_REPLAY_DIR_INDEX) {
2378 ret = replay_one_dir_item(wc->trans, root, path,
2379 eb, i, &key);
2380 if (ret)
2381 break;
2382 }
2383
Chris Masone02119d2008-09-05 16:13:11 -04002384 if (wc->stage < LOG_WALK_REPLAY_ALL)
2385 continue;
2386
2387 /* these keys are simply copied */
2388 if (key.type == BTRFS_XATTR_ITEM_KEY) {
2389 ret = overwrite_item(wc->trans, root, path,
2390 eb, i, &key);
Josef Bacikb50c6e22013-04-25 15:55:30 -04002391 if (ret)
2392 break;
Liu Bo2da1c662013-05-26 13:50:29 +00002393 } else if (key.type == BTRFS_INODE_REF_KEY ||
2394 key.type == BTRFS_INODE_EXTREF_KEY) {
Mark Fashehf1863732012-08-08 11:32:27 -07002395 ret = add_inode_ref(wc->trans, root, log, path,
2396 eb, i, &key);
Josef Bacikb50c6e22013-04-25 15:55:30 -04002397 if (ret && ret != -ENOENT)
2398 break;
2399 ret = 0;
Chris Masone02119d2008-09-05 16:13:11 -04002400 } else if (key.type == BTRFS_EXTENT_DATA_KEY) {
2401 ret = replay_one_extent(wc->trans, root, path,
2402 eb, i, &key);
Josef Bacikb50c6e22013-04-25 15:55:30 -04002403 if (ret)
2404 break;
Josef Bacikdd8e7212013-09-11 11:57:23 -04002405 } else if (key.type == BTRFS_DIR_ITEM_KEY) {
Chris Masone02119d2008-09-05 16:13:11 -04002406 ret = replay_one_dir_item(wc->trans, root, path,
2407 eb, i, &key);
Josef Bacikb50c6e22013-04-25 15:55:30 -04002408 if (ret)
2409 break;
Chris Masone02119d2008-09-05 16:13:11 -04002410 }
2411 }
2412 btrfs_free_path(path);
Josef Bacikb50c6e22013-04-25 15:55:30 -04002413 return ret;
Chris Masone02119d2008-09-05 16:13:11 -04002414}
2415
Chris Masond3977122009-01-05 21:25:51 -05002416static noinline int walk_down_log_tree(struct btrfs_trans_handle *trans,
Chris Masone02119d2008-09-05 16:13:11 -04002417 struct btrfs_root *root,
2418 struct btrfs_path *path, int *level,
2419 struct walk_control *wc)
2420{
Jeff Mahoney0b246af2016-06-22 18:54:23 -04002421 struct btrfs_fs_info *fs_info = root->fs_info;
Chris Masone02119d2008-09-05 16:13:11 -04002422 u64 root_owner;
Chris Masone02119d2008-09-05 16:13:11 -04002423 u64 bytenr;
2424 u64 ptr_gen;
2425 struct extent_buffer *next;
2426 struct extent_buffer *cur;
2427 struct extent_buffer *parent;
2428 u32 blocksize;
2429 int ret = 0;
2430
2431 WARN_ON(*level < 0);
2432 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2433
Chris Masond3977122009-01-05 21:25:51 -05002434 while (*level > 0) {
Chris Masone02119d2008-09-05 16:13:11 -04002435 WARN_ON(*level < 0);
2436 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2437 cur = path->nodes[*level];
2438
Dulshani Gunawardhanafae7f212013-10-31 10:30:08 +05302439 WARN_ON(btrfs_header_level(cur) != *level);
Chris Masone02119d2008-09-05 16:13:11 -04002440
2441 if (path->slots[*level] >=
2442 btrfs_header_nritems(cur))
2443 break;
2444
2445 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
2446 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
Jeff Mahoney0b246af2016-06-22 18:54:23 -04002447 blocksize = fs_info->nodesize;
Chris Masone02119d2008-09-05 16:13:11 -04002448
2449 parent = path->nodes[*level];
2450 root_owner = btrfs_header_owner(parent);
Chris Masone02119d2008-09-05 16:13:11 -04002451
Jeff Mahoney2ff7e612016-06-22 18:54:24 -04002452 next = btrfs_find_create_tree_block(fs_info, bytenr);
Liu Boc871b0f2016-06-06 12:01:23 -07002453 if (IS_ERR(next))
2454 return PTR_ERR(next);
Chris Masone02119d2008-09-05 16:13:11 -04002455
Chris Masone02119d2008-09-05 16:13:11 -04002456 if (*level == 1) {
Mark Fasheh1e5063d2011-07-12 10:46:06 -07002457 ret = wc->process_func(root, next, wc, ptr_gen);
Josef Bacikb50c6e22013-04-25 15:55:30 -04002458 if (ret) {
2459 free_extent_buffer(next);
Mark Fasheh1e5063d2011-07-12 10:46:06 -07002460 return ret;
Josef Bacikb50c6e22013-04-25 15:55:30 -04002461 }
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002462
Chris Masone02119d2008-09-05 16:13:11 -04002463 path->slots[*level]++;
2464 if (wc->free) {
Tsutomu Itoh018642a2012-05-29 18:10:13 +09002465 ret = btrfs_read_buffer(next, ptr_gen);
2466 if (ret) {
2467 free_extent_buffer(next);
2468 return ret;
2469 }
Chris Masone02119d2008-09-05 16:13:11 -04002470
Josef Bacik681ae502013-10-07 15:11:00 -04002471 if (trans) {
2472 btrfs_tree_lock(next);
2473 btrfs_set_lock_blocking(next);
Jeff Mahoney0b246af2016-06-22 18:54:23 -04002474 clean_tree_block(trans, fs_info, next);
Josef Bacik681ae502013-10-07 15:11:00 -04002475 btrfs_wait_tree_block_writeback(next);
2476 btrfs_tree_unlock(next);
2477 }
Chris Masone02119d2008-09-05 16:13:11 -04002478
Chris Masone02119d2008-09-05 16:13:11 -04002479 WARN_ON(root_owner !=
2480 BTRFS_TREE_LOG_OBJECTID);
Jeff Mahoney2ff7e612016-06-22 18:54:24 -04002481 ret = btrfs_free_and_pin_reserved_extent(
2482 fs_info, bytenr,
2483 blocksize);
Josef Bacik36508602013-04-25 16:23:32 -04002484 if (ret) {
2485 free_extent_buffer(next);
2486 return ret;
2487 }
Chris Masone02119d2008-09-05 16:13:11 -04002488 }
2489 free_extent_buffer(next);
2490 continue;
2491 }
Tsutomu Itoh018642a2012-05-29 18:10:13 +09002492 ret = btrfs_read_buffer(next, ptr_gen);
2493 if (ret) {
2494 free_extent_buffer(next);
2495 return ret;
2496 }
Chris Masone02119d2008-09-05 16:13:11 -04002497
2498 WARN_ON(*level <= 0);
2499 if (path->nodes[*level-1])
2500 free_extent_buffer(path->nodes[*level-1]);
2501 path->nodes[*level-1] = next;
2502 *level = btrfs_header_level(next);
2503 path->slots[*level] = 0;
2504 cond_resched();
2505 }
2506 WARN_ON(*level < 0);
2507 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2508
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002509 path->slots[*level] = btrfs_header_nritems(path->nodes[*level]);
Chris Masone02119d2008-09-05 16:13:11 -04002510
2511 cond_resched();
2512 return 0;
2513}
2514
Chris Masond3977122009-01-05 21:25:51 -05002515static noinline int walk_up_log_tree(struct btrfs_trans_handle *trans,
Chris Masone02119d2008-09-05 16:13:11 -04002516 struct btrfs_root *root,
2517 struct btrfs_path *path, int *level,
2518 struct walk_control *wc)
2519{
Jeff Mahoney0b246af2016-06-22 18:54:23 -04002520 struct btrfs_fs_info *fs_info = root->fs_info;
Chris Masone02119d2008-09-05 16:13:11 -04002521 u64 root_owner;
Chris Masone02119d2008-09-05 16:13:11 -04002522 int i;
2523 int slot;
2524 int ret;
2525
Chris Masond3977122009-01-05 21:25:51 -05002526 for (i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) {
Chris Masone02119d2008-09-05 16:13:11 -04002527 slot = path->slots[i];
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002528 if (slot + 1 < btrfs_header_nritems(path->nodes[i])) {
Chris Masone02119d2008-09-05 16:13:11 -04002529 path->slots[i]++;
2530 *level = i;
2531 WARN_ON(*level == 0);
2532 return 0;
2533 } else {
Zheng Yan31840ae2008-09-23 13:14:14 -04002534 struct extent_buffer *parent;
2535 if (path->nodes[*level] == root->node)
2536 parent = path->nodes[*level];
2537 else
2538 parent = path->nodes[*level + 1];
2539
2540 root_owner = btrfs_header_owner(parent);
Mark Fasheh1e5063d2011-07-12 10:46:06 -07002541 ret = wc->process_func(root, path->nodes[*level], wc,
Chris Masone02119d2008-09-05 16:13:11 -04002542 btrfs_header_generation(path->nodes[*level]));
Mark Fasheh1e5063d2011-07-12 10:46:06 -07002543 if (ret)
2544 return ret;
2545
Chris Masone02119d2008-09-05 16:13:11 -04002546 if (wc->free) {
2547 struct extent_buffer *next;
2548
2549 next = path->nodes[*level];
2550
Josef Bacik681ae502013-10-07 15:11:00 -04002551 if (trans) {
2552 btrfs_tree_lock(next);
2553 btrfs_set_lock_blocking(next);
Jeff Mahoney0b246af2016-06-22 18:54:23 -04002554 clean_tree_block(trans, fs_info, next);
Josef Bacik681ae502013-10-07 15:11:00 -04002555 btrfs_wait_tree_block_writeback(next);
2556 btrfs_tree_unlock(next);
2557 }
Chris Masone02119d2008-09-05 16:13:11 -04002558
Chris Masone02119d2008-09-05 16:13:11 -04002559 WARN_ON(root_owner != BTRFS_TREE_LOG_OBJECTID);
Jeff Mahoney2ff7e612016-06-22 18:54:24 -04002560 ret = btrfs_free_and_pin_reserved_extent(
2561 fs_info,
Chris Masone02119d2008-09-05 16:13:11 -04002562 path->nodes[*level]->start,
Chris Masond00aff02008-09-11 15:54:42 -04002563 path->nodes[*level]->len);
Josef Bacik36508602013-04-25 16:23:32 -04002564 if (ret)
2565 return ret;
Chris Masone02119d2008-09-05 16:13:11 -04002566 }
2567 free_extent_buffer(path->nodes[*level]);
2568 path->nodes[*level] = NULL;
2569 *level = i + 1;
2570 }
2571 }
2572 return 1;
2573}
2574
2575/*
2576 * drop the reference count on the tree rooted at 'snap'. This traverses
2577 * the tree freeing any blocks that have a ref count of zero after being
2578 * decremented.
2579 */
2580static int walk_log_tree(struct btrfs_trans_handle *trans,
2581 struct btrfs_root *log, struct walk_control *wc)
2582{
Jeff Mahoney2ff7e612016-06-22 18:54:24 -04002583 struct btrfs_fs_info *fs_info = log->fs_info;
Chris Masone02119d2008-09-05 16:13:11 -04002584 int ret = 0;
2585 int wret;
2586 int level;
2587 struct btrfs_path *path;
Chris Masone02119d2008-09-05 16:13:11 -04002588 int orig_level;
2589
2590 path = btrfs_alloc_path();
Tsutomu Itohdb5b4932011-03-23 08:14:16 +00002591 if (!path)
2592 return -ENOMEM;
Chris Masone02119d2008-09-05 16:13:11 -04002593
2594 level = btrfs_header_level(log->node);
2595 orig_level = level;
2596 path->nodes[level] = log->node;
2597 extent_buffer_get(log->node);
2598 path->slots[level] = 0;
2599
Chris Masond3977122009-01-05 21:25:51 -05002600 while (1) {
Chris Masone02119d2008-09-05 16:13:11 -04002601 wret = walk_down_log_tree(trans, log, path, &level, wc);
2602 if (wret > 0)
2603 break;
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002604 if (wret < 0) {
Chris Masone02119d2008-09-05 16:13:11 -04002605 ret = wret;
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002606 goto out;
2607 }
Chris Masone02119d2008-09-05 16:13:11 -04002608
2609 wret = walk_up_log_tree(trans, log, path, &level, wc);
2610 if (wret > 0)
2611 break;
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002612 if (wret < 0) {
Chris Masone02119d2008-09-05 16:13:11 -04002613 ret = wret;
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002614 goto out;
2615 }
Chris Masone02119d2008-09-05 16:13:11 -04002616 }
2617
2618 /* was the root node processed? if not, catch it here */
2619 if (path->nodes[orig_level]) {
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002620 ret = wc->process_func(log, path->nodes[orig_level], wc,
Chris Masone02119d2008-09-05 16:13:11 -04002621 btrfs_header_generation(path->nodes[orig_level]));
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002622 if (ret)
2623 goto out;
Chris Masone02119d2008-09-05 16:13:11 -04002624 if (wc->free) {
2625 struct extent_buffer *next;
2626
2627 next = path->nodes[orig_level];
2628
Josef Bacik681ae502013-10-07 15:11:00 -04002629 if (trans) {
2630 btrfs_tree_lock(next);
2631 btrfs_set_lock_blocking(next);
Jeff Mahoney2ff7e612016-06-22 18:54:24 -04002632 clean_tree_block(trans, fs_info, next);
Josef Bacik681ae502013-10-07 15:11:00 -04002633 btrfs_wait_tree_block_writeback(next);
2634 btrfs_tree_unlock(next);
2635 }
Chris Masone02119d2008-09-05 16:13:11 -04002636
Chris Masone02119d2008-09-05 16:13:11 -04002637 WARN_ON(log->root_key.objectid !=
2638 BTRFS_TREE_LOG_OBJECTID);
Jeff Mahoney2ff7e612016-06-22 18:54:24 -04002639 ret = btrfs_free_and_pin_reserved_extent(fs_info,
2640 next->start, next->len);
Josef Bacik36508602013-04-25 16:23:32 -04002641 if (ret)
2642 goto out;
Chris Masone02119d2008-09-05 16:13:11 -04002643 }
2644 }
2645
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002646out:
Chris Masone02119d2008-09-05 16:13:11 -04002647 btrfs_free_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04002648 return ret;
2649}
2650
Yan Zheng7237f182009-01-21 12:54:03 -05002651/*
2652 * helper function to update the item for a given subvolumes log root
2653 * in the tree of log roots
2654 */
2655static int update_log_root(struct btrfs_trans_handle *trans,
2656 struct btrfs_root *log)
2657{
Jeff Mahoney0b246af2016-06-22 18:54:23 -04002658 struct btrfs_fs_info *fs_info = log->fs_info;
Yan Zheng7237f182009-01-21 12:54:03 -05002659 int ret;
2660
2661 if (log->log_transid == 1) {
2662 /* insert root item on the first sync */
Jeff Mahoney0b246af2016-06-22 18:54:23 -04002663 ret = btrfs_insert_root(trans, fs_info->log_root_tree,
Yan Zheng7237f182009-01-21 12:54:03 -05002664 &log->root_key, &log->root_item);
2665 } else {
Jeff Mahoney0b246af2016-06-22 18:54:23 -04002666 ret = btrfs_update_root(trans, fs_info->log_root_tree,
Yan Zheng7237f182009-01-21 12:54:03 -05002667 &log->root_key, &log->root_item);
2668 }
2669 return ret;
2670}
2671
Zhaolei60d53eb2015-08-17 18:44:46 +08002672static void wait_log_commit(struct btrfs_root *root, int transid)
Chris Masone02119d2008-09-05 16:13:11 -04002673{
2674 DEFINE_WAIT(wait);
Yan Zheng7237f182009-01-21 12:54:03 -05002675 int index = transid % 2;
Chris Masone02119d2008-09-05 16:13:11 -04002676
Yan Zheng7237f182009-01-21 12:54:03 -05002677 /*
2678 * we only allow two pending log transactions at a time,
2679 * so we know that if ours is more than 2 older than the
2680 * current transaction, we're done
2681 */
Chris Masone02119d2008-09-05 16:13:11 -04002682 do {
Yan Zheng7237f182009-01-21 12:54:03 -05002683 prepare_to_wait(&root->log_commit_wait[index],
2684 &wait, TASK_UNINTERRUPTIBLE);
2685 mutex_unlock(&root->log_mutex);
Chris Mason12fcfd22009-03-24 10:24:20 -04002686
Miao Xied1433de2014-02-20 18:08:59 +08002687 if (root->log_transid_committed < transid &&
Yan Zheng7237f182009-01-21 12:54:03 -05002688 atomic_read(&root->log_commit[index]))
Chris Masone02119d2008-09-05 16:13:11 -04002689 schedule();
Chris Mason12fcfd22009-03-24 10:24:20 -04002690
Yan Zheng7237f182009-01-21 12:54:03 -05002691 finish_wait(&root->log_commit_wait[index], &wait);
2692 mutex_lock(&root->log_mutex);
Miao Xied1433de2014-02-20 18:08:59 +08002693 } while (root->log_transid_committed < transid &&
Yan Zheng7237f182009-01-21 12:54:03 -05002694 atomic_read(&root->log_commit[index]));
Yan Zheng7237f182009-01-21 12:54:03 -05002695}
2696
Zhaolei60d53eb2015-08-17 18:44:46 +08002697static void wait_for_writer(struct btrfs_root *root)
Yan Zheng7237f182009-01-21 12:54:03 -05002698{
2699 DEFINE_WAIT(wait);
Miao Xie8b050d32014-02-20 18:08:58 +08002700
2701 while (atomic_read(&root->log_writers)) {
Yan Zheng7237f182009-01-21 12:54:03 -05002702 prepare_to_wait(&root->log_writer_wait,
2703 &wait, TASK_UNINTERRUPTIBLE);
2704 mutex_unlock(&root->log_mutex);
Miao Xie8b050d32014-02-20 18:08:58 +08002705 if (atomic_read(&root->log_writers))
Yan Zheng7237f182009-01-21 12:54:03 -05002706 schedule();
Yan Zheng7237f182009-01-21 12:54:03 -05002707 finish_wait(&root->log_writer_wait, &wait);
Filipe Manana575849e2015-02-11 11:12:39 +00002708 mutex_lock(&root->log_mutex);
Yan Zheng7237f182009-01-21 12:54:03 -05002709 }
Chris Masone02119d2008-09-05 16:13:11 -04002710}
2711
Miao Xie8b050d32014-02-20 18:08:58 +08002712static inline void btrfs_remove_log_ctx(struct btrfs_root *root,
2713 struct btrfs_log_ctx *ctx)
2714{
2715 if (!ctx)
2716 return;
2717
2718 mutex_lock(&root->log_mutex);
2719 list_del_init(&ctx->list);
2720 mutex_unlock(&root->log_mutex);
2721}
2722
2723/*
2724 * Invoked in log mutex context, or be sure there is no other task which
2725 * can access the list.
2726 */
2727static inline void btrfs_remove_all_log_ctxs(struct btrfs_root *root,
2728 int index, int error)
2729{
2730 struct btrfs_log_ctx *ctx;
Chris Mason570dd452016-10-27 10:42:20 -07002731 struct btrfs_log_ctx *safe;
Miao Xie8b050d32014-02-20 18:08:58 +08002732
Chris Mason570dd452016-10-27 10:42:20 -07002733 list_for_each_entry_safe(ctx, safe, &root->log_ctxs[index], list) {
2734 list_del_init(&ctx->list);
Miao Xie8b050d32014-02-20 18:08:58 +08002735 ctx->log_ret = error;
Chris Mason570dd452016-10-27 10:42:20 -07002736 }
Miao Xie8b050d32014-02-20 18:08:58 +08002737
2738 INIT_LIST_HEAD(&root->log_ctxs[index]);
2739}
2740
Chris Masone02119d2008-09-05 16:13:11 -04002741/*
2742 * btrfs_sync_log does sends a given tree log down to the disk and
2743 * updates the super blocks to record it. When this call is done,
Chris Mason12fcfd22009-03-24 10:24:20 -04002744 * you know that any inodes previously logged are safely on disk only
2745 * if it returns 0.
2746 *
2747 * Any other return value means you need to call btrfs_commit_transaction.
2748 * Some of the edge cases for fsyncing directories that have had unlinks
2749 * or renames done in the past mean that sometimes the only safe
2750 * fsync is to commit the whole FS. When btrfs_sync_log returns -EAGAIN,
2751 * that has happened.
Chris Masone02119d2008-09-05 16:13:11 -04002752 */
2753int btrfs_sync_log(struct btrfs_trans_handle *trans,
Miao Xie8b050d32014-02-20 18:08:58 +08002754 struct btrfs_root *root, struct btrfs_log_ctx *ctx)
Chris Masone02119d2008-09-05 16:13:11 -04002755{
Yan Zheng7237f182009-01-21 12:54:03 -05002756 int index1;
2757 int index2;
Yan, Zheng8cef4e12009-11-12 09:33:26 +00002758 int mark;
Chris Masone02119d2008-09-05 16:13:11 -04002759 int ret;
Jeff Mahoney0b246af2016-06-22 18:54:23 -04002760 struct btrfs_fs_info *fs_info = root->fs_info;
Chris Masone02119d2008-09-05 16:13:11 -04002761 struct btrfs_root *log = root->log_root;
Jeff Mahoney0b246af2016-06-22 18:54:23 -04002762 struct btrfs_root *log_root_tree = fs_info->log_root_tree;
Miao Xiebb14a592014-02-20 18:08:56 +08002763 int log_transid = 0;
Miao Xie8b050d32014-02-20 18:08:58 +08002764 struct btrfs_log_ctx root_log_ctx;
Miao Xiec6adc9c2013-05-28 10:05:39 +00002765 struct blk_plug plug;
Chris Masone02119d2008-09-05 16:13:11 -04002766
Yan Zheng7237f182009-01-21 12:54:03 -05002767 mutex_lock(&root->log_mutex);
Miao Xied1433de2014-02-20 18:08:59 +08002768 log_transid = ctx->log_transid;
2769 if (root->log_transid_committed >= log_transid) {
Yan Zheng7237f182009-01-21 12:54:03 -05002770 mutex_unlock(&root->log_mutex);
Miao Xie8b050d32014-02-20 18:08:58 +08002771 return ctx->log_ret;
Chris Masone02119d2008-09-05 16:13:11 -04002772 }
Miao Xied1433de2014-02-20 18:08:59 +08002773
2774 index1 = log_transid % 2;
2775 if (atomic_read(&root->log_commit[index1])) {
Zhaolei60d53eb2015-08-17 18:44:46 +08002776 wait_log_commit(root, log_transid);
Miao Xied1433de2014-02-20 18:08:59 +08002777 mutex_unlock(&root->log_mutex);
2778 return ctx->log_ret;
2779 }
2780 ASSERT(log_transid == root->log_transid);
Yan Zheng7237f182009-01-21 12:54:03 -05002781 atomic_set(&root->log_commit[index1], 1);
2782
2783 /* wait for previous tree log sync to complete */
2784 if (atomic_read(&root->log_commit[(index1 + 1) % 2]))
Zhaolei60d53eb2015-08-17 18:44:46 +08002785 wait_log_commit(root, log_transid - 1);
Miao Xie48cab2e2014-02-20 18:08:52 +08002786
Yan, Zheng86df7eb2009-10-14 09:24:59 -04002787 while (1) {
Miao Xie2ecb7922012-09-06 04:04:27 -06002788 int batch = atomic_read(&root->log_batch);
Chris Masoncd354ad2011-10-20 15:45:37 -04002789 /* when we're on an ssd, just kick the log commit out */
Jeff Mahoney0b246af2016-06-22 18:54:23 -04002790 if (!btrfs_test_opt(fs_info, SSD) &&
Miao Xie27cdeb72014-04-02 19:51:05 +08002791 test_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state)) {
Yan, Zheng86df7eb2009-10-14 09:24:59 -04002792 mutex_unlock(&root->log_mutex);
2793 schedule_timeout_uninterruptible(1);
2794 mutex_lock(&root->log_mutex);
2795 }
Zhaolei60d53eb2015-08-17 18:44:46 +08002796 wait_for_writer(root);
Miao Xie2ecb7922012-09-06 04:04:27 -06002797 if (batch == atomic_read(&root->log_batch))
Chris Masone02119d2008-09-05 16:13:11 -04002798 break;
2799 }
Chris Masond0c803c2008-09-11 16:17:57 -04002800
Chris Mason12fcfd22009-03-24 10:24:20 -04002801 /* bail out if we need to do a full commit */
Jeff Mahoney0b246af2016-06-22 18:54:23 -04002802 if (btrfs_need_log_full_commit(fs_info, trans)) {
Chris Mason12fcfd22009-03-24 10:24:20 -04002803 ret = -EAGAIN;
Josef Bacik2ab28f32012-10-12 15:27:49 -04002804 btrfs_free_logged_extents(log, log_transid);
Chris Mason12fcfd22009-03-24 10:24:20 -04002805 mutex_unlock(&root->log_mutex);
2806 goto out;
2807 }
2808
Yan, Zheng8cef4e12009-11-12 09:33:26 +00002809 if (log_transid % 2 == 0)
2810 mark = EXTENT_DIRTY;
2811 else
2812 mark = EXTENT_NEW;
2813
Chris Mason690587d2009-10-13 13:29:19 -04002814 /* we start IO on all the marked extents here, but we don't actually
2815 * wait for them until later.
2816 */
Miao Xiec6adc9c2013-05-28 10:05:39 +00002817 blk_start_plug(&plug);
Jeff Mahoney2ff7e612016-06-22 18:54:24 -04002818 ret = btrfs_write_marked_extents(fs_info, &log->dirty_log_pages, mark);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002819 if (ret) {
Miao Xiec6adc9c2013-05-28 10:05:39 +00002820 blk_finish_plug(&plug);
Jeff Mahoney66642832016-06-10 18:19:25 -04002821 btrfs_abort_transaction(trans, ret);
Josef Bacik2ab28f32012-10-12 15:27:49 -04002822 btrfs_free_logged_extents(log, log_transid);
Jeff Mahoney0b246af2016-06-22 18:54:23 -04002823 btrfs_set_log_full_commit(fs_info, trans);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002824 mutex_unlock(&root->log_mutex);
2825 goto out;
2826 }
Yan Zheng7237f182009-01-21 12:54:03 -05002827
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002828 btrfs_set_root_node(&log->root_item, log->node);
Yan Zheng7237f182009-01-21 12:54:03 -05002829
Yan Zheng7237f182009-01-21 12:54:03 -05002830 root->log_transid++;
2831 log->log_transid = root->log_transid;
Josef Bacikff782e02009-10-08 15:30:04 -04002832 root->log_start_pid = 0;
Yan Zheng7237f182009-01-21 12:54:03 -05002833 /*
Yan, Zheng8cef4e12009-11-12 09:33:26 +00002834 * IO has been started, blocks of the log tree have WRITTEN flag set
2835 * in their headers. new modifications of the log will be written to
2836 * new positions. so it's safe to allow log writers to go in.
Yan Zheng7237f182009-01-21 12:54:03 -05002837 */
2838 mutex_unlock(&root->log_mutex);
2839
Filipe Manana28a23592016-08-23 21:13:51 +01002840 btrfs_init_log_ctx(&root_log_ctx, NULL);
Miao Xied1433de2014-02-20 18:08:59 +08002841
Yan Zheng7237f182009-01-21 12:54:03 -05002842 mutex_lock(&log_root_tree->log_mutex);
Miao Xie2ecb7922012-09-06 04:04:27 -06002843 atomic_inc(&log_root_tree->log_batch);
Yan Zheng7237f182009-01-21 12:54:03 -05002844 atomic_inc(&log_root_tree->log_writers);
Miao Xied1433de2014-02-20 18:08:59 +08002845
2846 index2 = log_root_tree->log_transid % 2;
2847 list_add_tail(&root_log_ctx.list, &log_root_tree->log_ctxs[index2]);
2848 root_log_ctx.log_transid = log_root_tree->log_transid;
2849
Yan Zheng7237f182009-01-21 12:54:03 -05002850 mutex_unlock(&log_root_tree->log_mutex);
2851
2852 ret = update_log_root(trans, log);
Yan Zheng7237f182009-01-21 12:54:03 -05002853
2854 mutex_lock(&log_root_tree->log_mutex);
2855 if (atomic_dec_and_test(&log_root_tree->log_writers)) {
David Sterba779adf02015-02-16 19:39:00 +01002856 /*
2857 * Implicit memory barrier after atomic_dec_and_test
2858 */
Yan Zheng7237f182009-01-21 12:54:03 -05002859 if (waitqueue_active(&log_root_tree->log_writer_wait))
2860 wake_up(&log_root_tree->log_writer_wait);
2861 }
2862
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002863 if (ret) {
Miao Xied1433de2014-02-20 18:08:59 +08002864 if (!list_empty(&root_log_ctx.list))
2865 list_del_init(&root_log_ctx.list);
2866
Miao Xiec6adc9c2013-05-28 10:05:39 +00002867 blk_finish_plug(&plug);
Jeff Mahoney0b246af2016-06-22 18:54:23 -04002868 btrfs_set_log_full_commit(fs_info, trans);
Miao Xie995946d2014-04-02 19:51:06 +08002869
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002870 if (ret != -ENOSPC) {
Jeff Mahoney66642832016-06-10 18:19:25 -04002871 btrfs_abort_transaction(trans, ret);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002872 mutex_unlock(&log_root_tree->log_mutex);
2873 goto out;
2874 }
Jeff Mahoneybf89d382016-09-09 20:42:44 -04002875 btrfs_wait_tree_log_extents(log, mark);
Josef Bacik2ab28f32012-10-12 15:27:49 -04002876 btrfs_free_logged_extents(log, log_transid);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002877 mutex_unlock(&log_root_tree->log_mutex);
2878 ret = -EAGAIN;
2879 goto out;
2880 }
2881
Miao Xied1433de2014-02-20 18:08:59 +08002882 if (log_root_tree->log_transid_committed >= root_log_ctx.log_transid) {
Forrest Liu3da5ab52015-01-30 19:42:12 +08002883 blk_finish_plug(&plug);
Chris Masoncbd60aa2016-09-06 05:37:40 -07002884 list_del_init(&root_log_ctx.list);
Miao Xied1433de2014-02-20 18:08:59 +08002885 mutex_unlock(&log_root_tree->log_mutex);
2886 ret = root_log_ctx.log_ret;
2887 goto out;
2888 }
Miao Xie8b050d32014-02-20 18:08:58 +08002889
Miao Xied1433de2014-02-20 18:08:59 +08002890 index2 = root_log_ctx.log_transid % 2;
Yan Zheng7237f182009-01-21 12:54:03 -05002891 if (atomic_read(&log_root_tree->log_commit[index2])) {
Miao Xiec6adc9c2013-05-28 10:05:39 +00002892 blk_finish_plug(&plug);
Jeff Mahoneybf89d382016-09-09 20:42:44 -04002893 ret = btrfs_wait_tree_log_extents(log, mark);
Josef Bacik50d9aa92014-11-21 14:52:38 -05002894 btrfs_wait_logged_extents(trans, log, log_transid);
Zhaolei60d53eb2015-08-17 18:44:46 +08002895 wait_log_commit(log_root_tree,
Miao Xied1433de2014-02-20 18:08:59 +08002896 root_log_ctx.log_transid);
Yan Zheng7237f182009-01-21 12:54:03 -05002897 mutex_unlock(&log_root_tree->log_mutex);
Filipe Manana5ab5e442014-11-13 16:59:53 +00002898 if (!ret)
2899 ret = root_log_ctx.log_ret;
Yan Zheng7237f182009-01-21 12:54:03 -05002900 goto out;
2901 }
Miao Xied1433de2014-02-20 18:08:59 +08002902 ASSERT(root_log_ctx.log_transid == log_root_tree->log_transid);
Yan Zheng7237f182009-01-21 12:54:03 -05002903 atomic_set(&log_root_tree->log_commit[index2], 1);
2904
Chris Mason12fcfd22009-03-24 10:24:20 -04002905 if (atomic_read(&log_root_tree->log_commit[(index2 + 1) % 2])) {
Zhaolei60d53eb2015-08-17 18:44:46 +08002906 wait_log_commit(log_root_tree,
Miao Xied1433de2014-02-20 18:08:59 +08002907 root_log_ctx.log_transid - 1);
Chris Mason12fcfd22009-03-24 10:24:20 -04002908 }
Yan Zheng7237f182009-01-21 12:54:03 -05002909
Zhaolei60d53eb2015-08-17 18:44:46 +08002910 wait_for_writer(log_root_tree);
Chris Mason12fcfd22009-03-24 10:24:20 -04002911
2912 /*
2913 * now that we've moved on to the tree of log tree roots,
2914 * check the full commit flag again
2915 */
Jeff Mahoney0b246af2016-06-22 18:54:23 -04002916 if (btrfs_need_log_full_commit(fs_info, trans)) {
Miao Xiec6adc9c2013-05-28 10:05:39 +00002917 blk_finish_plug(&plug);
Jeff Mahoneybf89d382016-09-09 20:42:44 -04002918 btrfs_wait_tree_log_extents(log, mark);
Josef Bacik2ab28f32012-10-12 15:27:49 -04002919 btrfs_free_logged_extents(log, log_transid);
Chris Mason12fcfd22009-03-24 10:24:20 -04002920 mutex_unlock(&log_root_tree->log_mutex);
2921 ret = -EAGAIN;
2922 goto out_wake_log_root;
2923 }
Yan Zheng7237f182009-01-21 12:54:03 -05002924
Jeff Mahoney2ff7e612016-06-22 18:54:24 -04002925 ret = btrfs_write_marked_extents(fs_info,
Miao Xiec6adc9c2013-05-28 10:05:39 +00002926 &log_root_tree->dirty_log_pages,
2927 EXTENT_DIRTY | EXTENT_NEW);
2928 blk_finish_plug(&plug);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002929 if (ret) {
Jeff Mahoney0b246af2016-06-22 18:54:23 -04002930 btrfs_set_log_full_commit(fs_info, trans);
Jeff Mahoney66642832016-06-10 18:19:25 -04002931 btrfs_abort_transaction(trans, ret);
Josef Bacik2ab28f32012-10-12 15:27:49 -04002932 btrfs_free_logged_extents(log, log_transid);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002933 mutex_unlock(&log_root_tree->log_mutex);
2934 goto out_wake_log_root;
2935 }
Jeff Mahoneybf89d382016-09-09 20:42:44 -04002936 ret = btrfs_wait_tree_log_extents(log, mark);
Filipe Manana5ab5e442014-11-13 16:59:53 +00002937 if (!ret)
Jeff Mahoneybf89d382016-09-09 20:42:44 -04002938 ret = btrfs_wait_tree_log_extents(log_root_tree,
2939 EXTENT_NEW | EXTENT_DIRTY);
Filipe Manana5ab5e442014-11-13 16:59:53 +00002940 if (ret) {
Jeff Mahoney0b246af2016-06-22 18:54:23 -04002941 btrfs_set_log_full_commit(fs_info, trans);
Filipe Manana5ab5e442014-11-13 16:59:53 +00002942 btrfs_free_logged_extents(log, log_transid);
2943 mutex_unlock(&log_root_tree->log_mutex);
2944 goto out_wake_log_root;
2945 }
Josef Bacik50d9aa92014-11-21 14:52:38 -05002946 btrfs_wait_logged_extents(trans, log, log_transid);
Chris Masone02119d2008-09-05 16:13:11 -04002947
Jeff Mahoney0b246af2016-06-22 18:54:23 -04002948 btrfs_set_super_log_root(fs_info->super_for_commit,
2949 log_root_tree->node->start);
2950 btrfs_set_super_log_root_level(fs_info->super_for_commit,
2951 btrfs_header_level(log_root_tree->node));
Chris Masone02119d2008-09-05 16:13:11 -04002952
Yan Zheng7237f182009-01-21 12:54:03 -05002953 log_root_tree->log_transid++;
Yan Zheng7237f182009-01-21 12:54:03 -05002954 mutex_unlock(&log_root_tree->log_mutex);
2955
2956 /*
2957 * nobody else is going to jump in and write the the ctree
2958 * super here because the log_commit atomic below is protecting
2959 * us. We must be called with a transaction handle pinning
2960 * the running transaction open, so a full commit can't hop
2961 * in and cause problems either.
2962 */
Jeff Mahoney2ff7e612016-06-22 18:54:24 -04002963 ret = write_ctree_super(trans, fs_info, 1);
Stefan Behrens5af3e8c2012-08-01 18:56:49 +02002964 if (ret) {
Jeff Mahoney0b246af2016-06-22 18:54:23 -04002965 btrfs_set_log_full_commit(fs_info, trans);
Jeff Mahoney66642832016-06-10 18:19:25 -04002966 btrfs_abort_transaction(trans, ret);
Stefan Behrens5af3e8c2012-08-01 18:56:49 +02002967 goto out_wake_log_root;
2968 }
Yan Zheng7237f182009-01-21 12:54:03 -05002969
Chris Mason257c62e2009-10-13 13:21:08 -04002970 mutex_lock(&root->log_mutex);
2971 if (root->last_log_commit < log_transid)
2972 root->last_log_commit = log_transid;
2973 mutex_unlock(&root->log_mutex);
2974
Chris Mason12fcfd22009-03-24 10:24:20 -04002975out_wake_log_root:
Chris Mason570dd452016-10-27 10:42:20 -07002976 mutex_lock(&log_root_tree->log_mutex);
Miao Xie8b050d32014-02-20 18:08:58 +08002977 btrfs_remove_all_log_ctxs(log_root_tree, index2, ret);
2978
Miao Xied1433de2014-02-20 18:08:59 +08002979 log_root_tree->log_transid_committed++;
Yan Zheng7237f182009-01-21 12:54:03 -05002980 atomic_set(&log_root_tree->log_commit[index2], 0);
Miao Xied1433de2014-02-20 18:08:59 +08002981 mutex_unlock(&log_root_tree->log_mutex);
2982
David Sterba33a9eca2015-10-10 18:35:10 +02002983 /*
2984 * The barrier before waitqueue_active is implied by mutex_unlock
2985 */
Yan Zheng7237f182009-01-21 12:54:03 -05002986 if (waitqueue_active(&log_root_tree->log_commit_wait[index2]))
2987 wake_up(&log_root_tree->log_commit_wait[index2]);
Chris Masone02119d2008-09-05 16:13:11 -04002988out:
Miao Xied1433de2014-02-20 18:08:59 +08002989 mutex_lock(&root->log_mutex);
Chris Mason570dd452016-10-27 10:42:20 -07002990 btrfs_remove_all_log_ctxs(root, index1, ret);
Miao Xied1433de2014-02-20 18:08:59 +08002991 root->log_transid_committed++;
Yan Zheng7237f182009-01-21 12:54:03 -05002992 atomic_set(&root->log_commit[index1], 0);
Miao Xied1433de2014-02-20 18:08:59 +08002993 mutex_unlock(&root->log_mutex);
Miao Xie8b050d32014-02-20 18:08:58 +08002994
David Sterba33a9eca2015-10-10 18:35:10 +02002995 /*
2996 * The barrier before waitqueue_active is implied by mutex_unlock
2997 */
Yan Zheng7237f182009-01-21 12:54:03 -05002998 if (waitqueue_active(&root->log_commit_wait[index1]))
2999 wake_up(&root->log_commit_wait[index1]);
Chris Masonb31eabd2011-01-31 16:48:24 -05003000 return ret;
Chris Masone02119d2008-09-05 16:13:11 -04003001}
3002
Yan, Zheng4a500fd2010-05-16 10:49:59 -04003003static void free_log_tree(struct btrfs_trans_handle *trans,
3004 struct btrfs_root *log)
Chris Masone02119d2008-09-05 16:13:11 -04003005{
3006 int ret;
Chris Masond0c803c2008-09-11 16:17:57 -04003007 u64 start;
3008 u64 end;
Chris Masone02119d2008-09-05 16:13:11 -04003009 struct walk_control wc = {
3010 .free = 1,
3011 .process_func = process_one_buffer
3012 };
3013
Josef Bacik681ae502013-10-07 15:11:00 -04003014 ret = walk_log_tree(trans, log, &wc);
3015 /* I don't think this can happen but just in case */
3016 if (ret)
Jeff Mahoney66642832016-06-10 18:19:25 -04003017 btrfs_abort_transaction(trans, ret);
Chris Masone02119d2008-09-05 16:13:11 -04003018
Chris Masond3977122009-01-05 21:25:51 -05003019 while (1) {
Chris Masond0c803c2008-09-11 16:17:57 -04003020 ret = find_first_extent_bit(&log->dirty_log_pages,
Josef Bacike6138872012-09-27 17:07:30 -04003021 0, &start, &end, EXTENT_DIRTY | EXTENT_NEW,
3022 NULL);
Chris Masond0c803c2008-09-11 16:17:57 -04003023 if (ret)
3024 break;
3025
Yan, Zheng8cef4e12009-11-12 09:33:26 +00003026 clear_extent_bits(&log->dirty_log_pages, start, end,
David Sterba91166212016-04-26 23:54:39 +02003027 EXTENT_DIRTY | EXTENT_NEW);
Chris Masond0c803c2008-09-11 16:17:57 -04003028 }
3029
Josef Bacik2ab28f32012-10-12 15:27:49 -04003030 /*
3031 * We may have short-circuited the log tree with the full commit logic
3032 * and left ordered extents on our list, so clear these out to keep us
3033 * from leaking inodes and memory.
3034 */
3035 btrfs_free_logged_extents(log, 0);
3036 btrfs_free_logged_extents(log, 1);
3037
Yan Zheng7237f182009-01-21 12:54:03 -05003038 free_extent_buffer(log->node);
3039 kfree(log);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04003040}
3041
3042/*
3043 * free all the extents used by the tree log. This should be called
3044 * at commit time of the full transaction
3045 */
3046int btrfs_free_log(struct btrfs_trans_handle *trans, struct btrfs_root *root)
3047{
3048 if (root->log_root) {
3049 free_log_tree(trans, root->log_root);
3050 root->log_root = NULL;
3051 }
3052 return 0;
3053}
3054
3055int btrfs_free_log_root_tree(struct btrfs_trans_handle *trans,
3056 struct btrfs_fs_info *fs_info)
3057{
3058 if (fs_info->log_root_tree) {
3059 free_log_tree(trans, fs_info->log_root_tree);
3060 fs_info->log_root_tree = NULL;
3061 }
Chris Masone02119d2008-09-05 16:13:11 -04003062 return 0;
3063}
3064
3065/*
Chris Masone02119d2008-09-05 16:13:11 -04003066 * If both a file and directory are logged, and unlinks or renames are
3067 * mixed in, we have a few interesting corners:
3068 *
3069 * create file X in dir Y
3070 * link file X to X.link in dir Y
3071 * fsync file X
3072 * unlink file X but leave X.link
3073 * fsync dir Y
3074 *
3075 * After a crash we would expect only X.link to exist. But file X
3076 * didn't get fsync'd again so the log has back refs for X and X.link.
3077 *
3078 * We solve this by removing directory entries and inode backrefs from the
3079 * log when a file that was logged in the current transaction is
3080 * unlinked. Any later fsync will include the updated log entries, and
3081 * we'll be able to reconstruct the proper directory items from backrefs.
3082 *
3083 * This optimizations allows us to avoid relogging the entire inode
3084 * or the entire directory.
3085 */
3086int btrfs_del_dir_entries_in_log(struct btrfs_trans_handle *trans,
3087 struct btrfs_root *root,
3088 const char *name, int name_len,
Nikolay Borisov49f34d12017-01-18 00:31:32 +02003089 struct btrfs_inode *dir, u64 index)
Chris Masone02119d2008-09-05 16:13:11 -04003090{
3091 struct btrfs_root *log;
3092 struct btrfs_dir_item *di;
3093 struct btrfs_path *path;
3094 int ret;
Yan, Zheng4a500fd2010-05-16 10:49:59 -04003095 int err = 0;
Chris Masone02119d2008-09-05 16:13:11 -04003096 int bytes_del = 0;
Nikolay Borisov49f34d12017-01-18 00:31:32 +02003097 u64 dir_ino = btrfs_ino(dir);
Chris Masone02119d2008-09-05 16:13:11 -04003098
Nikolay Borisov49f34d12017-01-18 00:31:32 +02003099 if (dir->logged_trans < trans->transid)
Chris Mason3a5f1d42008-09-11 15:53:37 -04003100 return 0;
3101
Chris Masone02119d2008-09-05 16:13:11 -04003102 ret = join_running_log_trans(root);
3103 if (ret)
3104 return 0;
3105
Nikolay Borisov49f34d12017-01-18 00:31:32 +02003106 mutex_lock(&dir->log_mutex);
Chris Masone02119d2008-09-05 16:13:11 -04003107
3108 log = root->log_root;
3109 path = btrfs_alloc_path();
Tsutomu Itoha62f44a2011-04-25 19:43:51 -04003110 if (!path) {
3111 err = -ENOMEM;
3112 goto out_unlock;
3113 }
liubo2a29edc2011-01-26 06:22:08 +00003114
Li Zefan33345d012011-04-20 10:31:50 +08003115 di = btrfs_lookup_dir_item(trans, log, path, dir_ino,
Chris Masone02119d2008-09-05 16:13:11 -04003116 name, name_len, -1);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04003117 if (IS_ERR(di)) {
3118 err = PTR_ERR(di);
3119 goto fail;
3120 }
3121 if (di) {
Chris Masone02119d2008-09-05 16:13:11 -04003122 ret = btrfs_delete_one_dir_name(trans, log, path, di);
3123 bytes_del += name_len;
Josef Bacik36508602013-04-25 16:23:32 -04003124 if (ret) {
3125 err = ret;
3126 goto fail;
3127 }
Chris Masone02119d2008-09-05 16:13:11 -04003128 }
David Sterbab3b4aa72011-04-21 01:20:15 +02003129 btrfs_release_path(path);
Li Zefan33345d012011-04-20 10:31:50 +08003130 di = btrfs_lookup_dir_index_item(trans, log, path, dir_ino,
Chris Masone02119d2008-09-05 16:13:11 -04003131 index, name, name_len, -1);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04003132 if (IS_ERR(di)) {
3133 err = PTR_ERR(di);
3134 goto fail;
3135 }
3136 if (di) {
Chris Masone02119d2008-09-05 16:13:11 -04003137 ret = btrfs_delete_one_dir_name(trans, log, path, di);
3138 bytes_del += name_len;
Josef Bacik36508602013-04-25 16:23:32 -04003139 if (ret) {
3140 err = ret;
3141 goto fail;
3142 }
Chris Masone02119d2008-09-05 16:13:11 -04003143 }
3144
3145 /* update the directory size in the log to reflect the names
3146 * we have removed
3147 */
3148 if (bytes_del) {
3149 struct btrfs_key key;
3150
Li Zefan33345d012011-04-20 10:31:50 +08003151 key.objectid = dir_ino;
Chris Masone02119d2008-09-05 16:13:11 -04003152 key.offset = 0;
3153 key.type = BTRFS_INODE_ITEM_KEY;
David Sterbab3b4aa72011-04-21 01:20:15 +02003154 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04003155
3156 ret = btrfs_search_slot(trans, log, &key, path, 0, 1);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04003157 if (ret < 0) {
3158 err = ret;
3159 goto fail;
3160 }
Chris Masone02119d2008-09-05 16:13:11 -04003161 if (ret == 0) {
3162 struct btrfs_inode_item *item;
3163 u64 i_size;
3164
3165 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3166 struct btrfs_inode_item);
3167 i_size = btrfs_inode_size(path->nodes[0], item);
3168 if (i_size > bytes_del)
3169 i_size -= bytes_del;
3170 else
3171 i_size = 0;
3172 btrfs_set_inode_size(path->nodes[0], item, i_size);
3173 btrfs_mark_buffer_dirty(path->nodes[0]);
3174 } else
3175 ret = 0;
David Sterbab3b4aa72011-04-21 01:20:15 +02003176 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04003177 }
Yan, Zheng4a500fd2010-05-16 10:49:59 -04003178fail:
Chris Masone02119d2008-09-05 16:13:11 -04003179 btrfs_free_path(path);
Tsutomu Itoha62f44a2011-04-25 19:43:51 -04003180out_unlock:
Nikolay Borisov49f34d12017-01-18 00:31:32 +02003181 mutex_unlock(&dir->log_mutex);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04003182 if (ret == -ENOSPC) {
Miao Xie995946d2014-04-02 19:51:06 +08003183 btrfs_set_log_full_commit(root->fs_info, trans);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04003184 ret = 0;
Jeff Mahoney79787ea2012-03-12 16:03:00 +01003185 } else if (ret < 0)
Jeff Mahoney66642832016-06-10 18:19:25 -04003186 btrfs_abort_transaction(trans, ret);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01003187
Chris Mason12fcfd22009-03-24 10:24:20 -04003188 btrfs_end_log_trans(root);
Chris Masone02119d2008-09-05 16:13:11 -04003189
Andi Kleen411fc6b2010-10-29 15:14:31 -04003190 return err;
Chris Masone02119d2008-09-05 16:13:11 -04003191}
3192
3193/* see comments for btrfs_del_dir_entries_in_log */
3194int btrfs_del_inode_ref_in_log(struct btrfs_trans_handle *trans,
3195 struct btrfs_root *root,
3196 const char *name, int name_len,
Nikolay Borisova491abb2017-01-18 00:31:33 +02003197 struct btrfs_inode *inode, u64 dirid)
Chris Masone02119d2008-09-05 16:13:11 -04003198{
Jeff Mahoney0b246af2016-06-22 18:54:23 -04003199 struct btrfs_fs_info *fs_info = root->fs_info;
Chris Masone02119d2008-09-05 16:13:11 -04003200 struct btrfs_root *log;
3201 u64 index;
3202 int ret;
3203
Nikolay Borisova491abb2017-01-18 00:31:33 +02003204 if (inode->logged_trans < trans->transid)
Chris Mason3a5f1d42008-09-11 15:53:37 -04003205 return 0;
3206
Chris Masone02119d2008-09-05 16:13:11 -04003207 ret = join_running_log_trans(root);
3208 if (ret)
3209 return 0;
3210 log = root->log_root;
Nikolay Borisova491abb2017-01-18 00:31:33 +02003211 mutex_lock(&inode->log_mutex);
Chris Masone02119d2008-09-05 16:13:11 -04003212
Nikolay Borisova491abb2017-01-18 00:31:33 +02003213 ret = btrfs_del_inode_ref(trans, log, name, name_len, btrfs_ino(inode),
Chris Masone02119d2008-09-05 16:13:11 -04003214 dirid, &index);
Nikolay Borisova491abb2017-01-18 00:31:33 +02003215 mutex_unlock(&inode->log_mutex);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04003216 if (ret == -ENOSPC) {
Jeff Mahoney0b246af2016-06-22 18:54:23 -04003217 btrfs_set_log_full_commit(fs_info, trans);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04003218 ret = 0;
Jeff Mahoney79787ea2012-03-12 16:03:00 +01003219 } else if (ret < 0 && ret != -ENOENT)
Jeff Mahoney66642832016-06-10 18:19:25 -04003220 btrfs_abort_transaction(trans, ret);
Chris Mason12fcfd22009-03-24 10:24:20 -04003221 btrfs_end_log_trans(root);
Chris Masone02119d2008-09-05 16:13:11 -04003222
Chris Masone02119d2008-09-05 16:13:11 -04003223 return ret;
3224}
3225
3226/*
3227 * creates a range item in the log for 'dirid'. first_offset and
3228 * last_offset tell us which parts of the key space the log should
3229 * be considered authoritative for.
3230 */
3231static noinline int insert_dir_log_key(struct btrfs_trans_handle *trans,
3232 struct btrfs_root *log,
3233 struct btrfs_path *path,
3234 int key_type, u64 dirid,
3235 u64 first_offset, u64 last_offset)
3236{
3237 int ret;
3238 struct btrfs_key key;
3239 struct btrfs_dir_log_item *item;
3240
3241 key.objectid = dirid;
3242 key.offset = first_offset;
3243 if (key_type == BTRFS_DIR_ITEM_KEY)
3244 key.type = BTRFS_DIR_LOG_ITEM_KEY;
3245 else
3246 key.type = BTRFS_DIR_LOG_INDEX_KEY;
3247 ret = btrfs_insert_empty_item(trans, log, path, &key, sizeof(*item));
Yan, Zheng4a500fd2010-05-16 10:49:59 -04003248 if (ret)
3249 return ret;
Chris Masone02119d2008-09-05 16:13:11 -04003250
3251 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3252 struct btrfs_dir_log_item);
3253 btrfs_set_dir_log_end(path->nodes[0], item, last_offset);
3254 btrfs_mark_buffer_dirty(path->nodes[0]);
David Sterbab3b4aa72011-04-21 01:20:15 +02003255 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04003256 return 0;
3257}
3258
3259/*
3260 * log all the items included in the current transaction for a given
3261 * directory. This also creates the range items in the log tree required
3262 * to replay anything deleted before the fsync
3263 */
3264static noinline int log_dir_items(struct btrfs_trans_handle *trans,
Nikolay Borisov684a5772017-01-18 00:31:41 +02003265 struct btrfs_root *root, struct btrfs_inode *inode,
Chris Masone02119d2008-09-05 16:13:11 -04003266 struct btrfs_path *path,
3267 struct btrfs_path *dst_path, int key_type,
Filipe Manana2f2ff0e2015-03-20 17:19:46 +00003268 struct btrfs_log_ctx *ctx,
Chris Masone02119d2008-09-05 16:13:11 -04003269 u64 min_offset, u64 *last_offset_ret)
3270{
3271 struct btrfs_key min_key;
Chris Masone02119d2008-09-05 16:13:11 -04003272 struct btrfs_root *log = root->log_root;
3273 struct extent_buffer *src;
Yan, Zheng4a500fd2010-05-16 10:49:59 -04003274 int err = 0;
Chris Masone02119d2008-09-05 16:13:11 -04003275 int ret;
3276 int i;
3277 int nritems;
3278 u64 first_offset = min_offset;
3279 u64 last_offset = (u64)-1;
Nikolay Borisov684a5772017-01-18 00:31:41 +02003280 u64 ino = btrfs_ino(inode);
Chris Masone02119d2008-09-05 16:13:11 -04003281
3282 log = root->log_root;
Chris Masone02119d2008-09-05 16:13:11 -04003283
Li Zefan33345d012011-04-20 10:31:50 +08003284 min_key.objectid = ino;
Chris Masone02119d2008-09-05 16:13:11 -04003285 min_key.type = key_type;
3286 min_key.offset = min_offset;
3287
Filipe David Borba Manana6174d3c2013-10-01 16:13:42 +01003288 ret = btrfs_search_forward(root, &min_key, path, trans->transid);
Chris Masone02119d2008-09-05 16:13:11 -04003289
3290 /*
3291 * we didn't find anything from this transaction, see if there
3292 * is anything at all
3293 */
Li Zefan33345d012011-04-20 10:31:50 +08003294 if (ret != 0 || min_key.objectid != ino || min_key.type != key_type) {
3295 min_key.objectid = ino;
Chris Masone02119d2008-09-05 16:13:11 -04003296 min_key.type = key_type;
3297 min_key.offset = (u64)-1;
David Sterbab3b4aa72011-04-21 01:20:15 +02003298 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04003299 ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0);
3300 if (ret < 0) {
David Sterbab3b4aa72011-04-21 01:20:15 +02003301 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04003302 return ret;
3303 }
Li Zefan33345d012011-04-20 10:31:50 +08003304 ret = btrfs_previous_item(root, path, ino, key_type);
Chris Masone02119d2008-09-05 16:13:11 -04003305
3306 /* if ret == 0 there are items for this type,
3307 * create a range to tell us the last key of this type.
3308 * otherwise, there are no items in this directory after
3309 * *min_offset, and we create a range to indicate that.
3310 */
3311 if (ret == 0) {
3312 struct btrfs_key tmp;
3313 btrfs_item_key_to_cpu(path->nodes[0], &tmp,
3314 path->slots[0]);
Chris Masond3977122009-01-05 21:25:51 -05003315 if (key_type == tmp.type)
Chris Masone02119d2008-09-05 16:13:11 -04003316 first_offset = max(min_offset, tmp.offset) + 1;
Chris Masone02119d2008-09-05 16:13:11 -04003317 }
3318 goto done;
3319 }
3320
3321 /* go backward to find any previous key */
Li Zefan33345d012011-04-20 10:31:50 +08003322 ret = btrfs_previous_item(root, path, ino, key_type);
Chris Masone02119d2008-09-05 16:13:11 -04003323 if (ret == 0) {
3324 struct btrfs_key tmp;
3325 btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]);
3326 if (key_type == tmp.type) {
3327 first_offset = tmp.offset;
3328 ret = overwrite_item(trans, log, dst_path,
3329 path->nodes[0], path->slots[0],
3330 &tmp);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04003331 if (ret) {
3332 err = ret;
3333 goto done;
3334 }
Chris Masone02119d2008-09-05 16:13:11 -04003335 }
3336 }
David Sterbab3b4aa72011-04-21 01:20:15 +02003337 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04003338
3339 /* find the first key from this transaction again */
3340 ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0);
Dulshani Gunawardhanafae7f212013-10-31 10:30:08 +05303341 if (WARN_ON(ret != 0))
Chris Masone02119d2008-09-05 16:13:11 -04003342 goto done;
Chris Masone02119d2008-09-05 16:13:11 -04003343
3344 /*
3345 * we have a block from this transaction, log every item in it
3346 * from our directory
3347 */
Chris Masond3977122009-01-05 21:25:51 -05003348 while (1) {
Chris Masone02119d2008-09-05 16:13:11 -04003349 struct btrfs_key tmp;
3350 src = path->nodes[0];
3351 nritems = btrfs_header_nritems(src);
3352 for (i = path->slots[0]; i < nritems; i++) {
Filipe Manana2f2ff0e2015-03-20 17:19:46 +00003353 struct btrfs_dir_item *di;
3354
Chris Masone02119d2008-09-05 16:13:11 -04003355 btrfs_item_key_to_cpu(src, &min_key, i);
3356
Li Zefan33345d012011-04-20 10:31:50 +08003357 if (min_key.objectid != ino || min_key.type != key_type)
Chris Masone02119d2008-09-05 16:13:11 -04003358 goto done;
3359 ret = overwrite_item(trans, log, dst_path, src, i,
3360 &min_key);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04003361 if (ret) {
3362 err = ret;
3363 goto done;
3364 }
Filipe Manana2f2ff0e2015-03-20 17:19:46 +00003365
3366 /*
3367 * We must make sure that when we log a directory entry,
3368 * the corresponding inode, after log replay, has a
3369 * matching link count. For example:
3370 *
3371 * touch foo
3372 * mkdir mydir
3373 * sync
3374 * ln foo mydir/bar
3375 * xfs_io -c "fsync" mydir
3376 * <crash>
3377 * <mount fs and log replay>
3378 *
3379 * Would result in a fsync log that when replayed, our
3380 * file inode would have a link count of 1, but we get
3381 * two directory entries pointing to the same inode.
3382 * After removing one of the names, it would not be
3383 * possible to remove the other name, which resulted
3384 * always in stale file handle errors, and would not
3385 * be possible to rmdir the parent directory, since
3386 * its i_size could never decrement to the value
3387 * BTRFS_EMPTY_DIR_SIZE, resulting in -ENOTEMPTY errors.
3388 */
3389 di = btrfs_item_ptr(src, i, struct btrfs_dir_item);
3390 btrfs_dir_item_key_to_cpu(src, di, &tmp);
3391 if (ctx &&
3392 (btrfs_dir_transid(src, di) == trans->transid ||
3393 btrfs_dir_type(src, di) == BTRFS_FT_DIR) &&
3394 tmp.type != BTRFS_ROOT_ITEM_KEY)
3395 ctx->log_new_dentries = true;
Chris Masone02119d2008-09-05 16:13:11 -04003396 }
3397 path->slots[0] = nritems;
3398
3399 /*
3400 * look ahead to the next item and see if it is also
3401 * from this directory and from this transaction
3402 */
3403 ret = btrfs_next_leaf(root, path);
3404 if (ret == 1) {
3405 last_offset = (u64)-1;
3406 goto done;
3407 }
3408 btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]);
Li Zefan33345d012011-04-20 10:31:50 +08003409 if (tmp.objectid != ino || tmp.type != key_type) {
Chris Masone02119d2008-09-05 16:13:11 -04003410 last_offset = (u64)-1;
3411 goto done;
3412 }
3413 if (btrfs_header_generation(path->nodes[0]) != trans->transid) {
3414 ret = overwrite_item(trans, log, dst_path,
3415 path->nodes[0], path->slots[0],
3416 &tmp);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04003417 if (ret)
3418 err = ret;
3419 else
3420 last_offset = tmp.offset;
Chris Masone02119d2008-09-05 16:13:11 -04003421 goto done;
3422 }
3423 }
3424done:
David Sterbab3b4aa72011-04-21 01:20:15 +02003425 btrfs_release_path(path);
3426 btrfs_release_path(dst_path);
Chris Masone02119d2008-09-05 16:13:11 -04003427
Yan, Zheng4a500fd2010-05-16 10:49:59 -04003428 if (err == 0) {
3429 *last_offset_ret = last_offset;
3430 /*
3431 * insert the log range keys to indicate where the log
3432 * is valid
3433 */
3434 ret = insert_dir_log_key(trans, log, path, key_type,
Li Zefan33345d012011-04-20 10:31:50 +08003435 ino, first_offset, last_offset);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04003436 if (ret)
3437 err = ret;
3438 }
3439 return err;
Chris Masone02119d2008-09-05 16:13:11 -04003440}
3441
3442/*
3443 * logging directories is very similar to logging inodes, We find all the items
3444 * from the current transaction and write them to the log.
3445 *
3446 * The recovery code scans the directory in the subvolume, and if it finds a
3447 * key in the range logged that is not present in the log tree, then it means
3448 * that dir entry was unlinked during the transaction.
3449 *
3450 * In order for that scan to work, we must include one key smaller than
3451 * the smallest logged by this transaction and one key larger than the largest
3452 * key logged by this transaction.
3453 */
3454static noinline int log_directory_changes(struct btrfs_trans_handle *trans,
Nikolay Borisovdbf39ea2017-01-18 00:31:42 +02003455 struct btrfs_root *root, struct btrfs_inode *inode,
Chris Masone02119d2008-09-05 16:13:11 -04003456 struct btrfs_path *path,
Filipe Manana2f2ff0e2015-03-20 17:19:46 +00003457 struct btrfs_path *dst_path,
3458 struct btrfs_log_ctx *ctx)
Chris Masone02119d2008-09-05 16:13:11 -04003459{
3460 u64 min_key;
3461 u64 max_key;
3462 int ret;
3463 int key_type = BTRFS_DIR_ITEM_KEY;
3464
3465again:
3466 min_key = 0;
3467 max_key = 0;
Chris Masond3977122009-01-05 21:25:51 -05003468 while (1) {
Nikolay Borisovdbf39ea2017-01-18 00:31:42 +02003469 ret = log_dir_items(trans, root, inode, path, dst_path, key_type,
3470 ctx, min_key, &max_key);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04003471 if (ret)
3472 return ret;
Chris Masone02119d2008-09-05 16:13:11 -04003473 if (max_key == (u64)-1)
3474 break;
3475 min_key = max_key + 1;
3476 }
3477
3478 if (key_type == BTRFS_DIR_ITEM_KEY) {
3479 key_type = BTRFS_DIR_INDEX_KEY;
3480 goto again;
3481 }
3482 return 0;
3483}
3484
3485/*
3486 * a helper function to drop items from the log before we relog an
3487 * inode. max_key_type indicates the highest item type to remove.
3488 * This cannot be run for file data extents because it does not
3489 * free the extents they point to.
3490 */
3491static int drop_objectid_items(struct btrfs_trans_handle *trans,
3492 struct btrfs_root *log,
3493 struct btrfs_path *path,
3494 u64 objectid, int max_key_type)
3495{
3496 int ret;
3497 struct btrfs_key key;
3498 struct btrfs_key found_key;
Josef Bacik18ec90d2012-09-28 11:56:28 -04003499 int start_slot;
Chris Masone02119d2008-09-05 16:13:11 -04003500
3501 key.objectid = objectid;
3502 key.type = max_key_type;
3503 key.offset = (u64)-1;
3504
Chris Masond3977122009-01-05 21:25:51 -05003505 while (1) {
Chris Masone02119d2008-09-05 16:13:11 -04003506 ret = btrfs_search_slot(trans, log, &key, path, -1, 1);
Josef Bacik36508602013-04-25 16:23:32 -04003507 BUG_ON(ret == 0); /* Logic error */
Yan, Zheng4a500fd2010-05-16 10:49:59 -04003508 if (ret < 0)
Chris Masone02119d2008-09-05 16:13:11 -04003509 break;
3510
3511 if (path->slots[0] == 0)
3512 break;
3513
3514 path->slots[0]--;
3515 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
3516 path->slots[0]);
3517
3518 if (found_key.objectid != objectid)
3519 break;
3520
Josef Bacik18ec90d2012-09-28 11:56:28 -04003521 found_key.offset = 0;
3522 found_key.type = 0;
3523 ret = btrfs_bin_search(path->nodes[0], &found_key, 0,
3524 &start_slot);
3525
3526 ret = btrfs_del_items(trans, log, path, start_slot,
3527 path->slots[0] - start_slot + 1);
3528 /*
3529 * If start slot isn't 0 then we don't need to re-search, we've
3530 * found the last guy with the objectid in this tree.
3531 */
3532 if (ret || start_slot != 0)
Tsutomu Itoh65a246c2011-05-19 04:37:44 +00003533 break;
David Sterbab3b4aa72011-04-21 01:20:15 +02003534 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04003535 }
David Sterbab3b4aa72011-04-21 01:20:15 +02003536 btrfs_release_path(path);
Josef Bacik5bdbeb22012-05-29 16:59:49 -04003537 if (ret > 0)
3538 ret = 0;
Yan, Zheng4a500fd2010-05-16 10:49:59 -04003539 return ret;
Chris Masone02119d2008-09-05 16:13:11 -04003540}
3541
Josef Bacik94edf4a2012-09-25 14:56:25 -04003542static void fill_inode_item(struct btrfs_trans_handle *trans,
3543 struct extent_buffer *leaf,
3544 struct btrfs_inode_item *item,
Filipe Manana1a4bcf42015-02-13 12:30:56 +00003545 struct inode *inode, int log_inode_only,
3546 u64 logged_isize)
Josef Bacik94edf4a2012-09-25 14:56:25 -04003547{
Josef Bacik0b1c6cc2012-10-23 16:03:44 -04003548 struct btrfs_map_token token;
Josef Bacik94edf4a2012-09-25 14:56:25 -04003549
Josef Bacik0b1c6cc2012-10-23 16:03:44 -04003550 btrfs_init_map_token(&token);
Josef Bacik94edf4a2012-09-25 14:56:25 -04003551
3552 if (log_inode_only) {
3553 /* set the generation to zero so the recover code
3554 * can tell the difference between an logging
3555 * just to say 'this inode exists' and a logging
3556 * to say 'update this inode with these values'
3557 */
Josef Bacik0b1c6cc2012-10-23 16:03:44 -04003558 btrfs_set_token_inode_generation(leaf, item, 0, &token);
Filipe Manana1a4bcf42015-02-13 12:30:56 +00003559 btrfs_set_token_inode_size(leaf, item, logged_isize, &token);
Josef Bacik94edf4a2012-09-25 14:56:25 -04003560 } else {
Josef Bacik0b1c6cc2012-10-23 16:03:44 -04003561 btrfs_set_token_inode_generation(leaf, item,
3562 BTRFS_I(inode)->generation,
3563 &token);
3564 btrfs_set_token_inode_size(leaf, item, inode->i_size, &token);
Josef Bacik94edf4a2012-09-25 14:56:25 -04003565 }
3566
Josef Bacik0b1c6cc2012-10-23 16:03:44 -04003567 btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token);
3568 btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token);
3569 btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token);
3570 btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token);
3571
David Sterbaa937b972014-12-12 17:39:12 +01003572 btrfs_set_token_timespec_sec(leaf, &item->atime,
Josef Bacik0b1c6cc2012-10-23 16:03:44 -04003573 inode->i_atime.tv_sec, &token);
David Sterbaa937b972014-12-12 17:39:12 +01003574 btrfs_set_token_timespec_nsec(leaf, &item->atime,
Josef Bacik0b1c6cc2012-10-23 16:03:44 -04003575 inode->i_atime.tv_nsec, &token);
3576
David Sterbaa937b972014-12-12 17:39:12 +01003577 btrfs_set_token_timespec_sec(leaf, &item->mtime,
Josef Bacik0b1c6cc2012-10-23 16:03:44 -04003578 inode->i_mtime.tv_sec, &token);
David Sterbaa937b972014-12-12 17:39:12 +01003579 btrfs_set_token_timespec_nsec(leaf, &item->mtime,
Josef Bacik0b1c6cc2012-10-23 16:03:44 -04003580 inode->i_mtime.tv_nsec, &token);
3581
David Sterbaa937b972014-12-12 17:39:12 +01003582 btrfs_set_token_timespec_sec(leaf, &item->ctime,
Josef Bacik0b1c6cc2012-10-23 16:03:44 -04003583 inode->i_ctime.tv_sec, &token);
David Sterbaa937b972014-12-12 17:39:12 +01003584 btrfs_set_token_timespec_nsec(leaf, &item->ctime,
Josef Bacik0b1c6cc2012-10-23 16:03:44 -04003585 inode->i_ctime.tv_nsec, &token);
3586
3587 btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode),
3588 &token);
3589
3590 btrfs_set_token_inode_sequence(leaf, item, inode->i_version, &token);
3591 btrfs_set_token_inode_transid(leaf, item, trans->transid, &token);
3592 btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token);
3593 btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token);
3594 btrfs_set_token_inode_block_group(leaf, item, 0, &token);
Josef Bacik94edf4a2012-09-25 14:56:25 -04003595}
3596
Josef Bacika95249b2012-10-11 16:17:34 -04003597static int log_inode_item(struct btrfs_trans_handle *trans,
3598 struct btrfs_root *log, struct btrfs_path *path,
Nikolay Borisov6d889a32017-01-18 00:31:47 +02003599 struct btrfs_inode *inode)
Josef Bacika95249b2012-10-11 16:17:34 -04003600{
3601 struct btrfs_inode_item *inode_item;
Josef Bacika95249b2012-10-11 16:17:34 -04003602 int ret;
3603
Filipe David Borba Mananaefd0c402013-10-07 21:20:44 +01003604 ret = btrfs_insert_empty_item(trans, log, path,
Nikolay Borisov6d889a32017-01-18 00:31:47 +02003605 &inode->location, sizeof(*inode_item));
Josef Bacika95249b2012-10-11 16:17:34 -04003606 if (ret && ret != -EEXIST)
3607 return ret;
3608 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3609 struct btrfs_inode_item);
Nikolay Borisov6d889a32017-01-18 00:31:47 +02003610 fill_inode_item(trans, path->nodes[0], inode_item, &inode->vfs_inode,
3611 0, 0);
Josef Bacika95249b2012-10-11 16:17:34 -04003612 btrfs_release_path(path);
3613 return 0;
3614}
3615
Chris Mason31ff1cd2008-09-11 16:17:57 -04003616static noinline int copy_items(struct btrfs_trans_handle *trans,
Nikolay Borisov44d70e12017-01-18 00:31:36 +02003617 struct btrfs_inode *inode,
Chris Mason31ff1cd2008-09-11 16:17:57 -04003618 struct btrfs_path *dst_path,
Josef Bacik16e75492013-10-22 12:18:51 -04003619 struct btrfs_path *src_path, u64 *last_extent,
Filipe Manana1a4bcf42015-02-13 12:30:56 +00003620 int start_slot, int nr, int inode_only,
3621 u64 logged_isize)
Chris Mason31ff1cd2008-09-11 16:17:57 -04003622{
Nikolay Borisov44d70e12017-01-18 00:31:36 +02003623 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
Chris Mason31ff1cd2008-09-11 16:17:57 -04003624 unsigned long src_offset;
3625 unsigned long dst_offset;
Nikolay Borisov44d70e12017-01-18 00:31:36 +02003626 struct btrfs_root *log = inode->root->log_root;
Chris Mason31ff1cd2008-09-11 16:17:57 -04003627 struct btrfs_file_extent_item *extent;
3628 struct btrfs_inode_item *inode_item;
Josef Bacik16e75492013-10-22 12:18:51 -04003629 struct extent_buffer *src = src_path->nodes[0];
3630 struct btrfs_key first_key, last_key, key;
Chris Mason31ff1cd2008-09-11 16:17:57 -04003631 int ret;
3632 struct btrfs_key *ins_keys;
3633 u32 *ins_sizes;
3634 char *ins_data;
3635 int i;
Chris Masond20f7042008-12-08 16:58:54 -05003636 struct list_head ordered_sums;
Nikolay Borisov44d70e12017-01-18 00:31:36 +02003637 int skip_csum = inode->flags & BTRFS_INODE_NODATASUM;
Josef Bacik16e75492013-10-22 12:18:51 -04003638 bool has_extents = false;
Filipe Manana74121f7c2014-08-07 12:00:44 +01003639 bool need_find_last_extent = true;
Josef Bacik16e75492013-10-22 12:18:51 -04003640 bool done = false;
Chris Masond20f7042008-12-08 16:58:54 -05003641
3642 INIT_LIST_HEAD(&ordered_sums);
Chris Mason31ff1cd2008-09-11 16:17:57 -04003643
3644 ins_data = kmalloc(nr * sizeof(struct btrfs_key) +
3645 nr * sizeof(u32), GFP_NOFS);
liubo2a29edc2011-01-26 06:22:08 +00003646 if (!ins_data)
3647 return -ENOMEM;
3648
Josef Bacik16e75492013-10-22 12:18:51 -04003649 first_key.objectid = (u64)-1;
3650
Chris Mason31ff1cd2008-09-11 16:17:57 -04003651 ins_sizes = (u32 *)ins_data;
3652 ins_keys = (struct btrfs_key *)(ins_data + nr * sizeof(u32));
3653
3654 for (i = 0; i < nr; i++) {
3655 ins_sizes[i] = btrfs_item_size_nr(src, i + start_slot);
3656 btrfs_item_key_to_cpu(src, ins_keys + i, i + start_slot);
3657 }
3658 ret = btrfs_insert_empty_items(trans, log, dst_path,
3659 ins_keys, ins_sizes, nr);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04003660 if (ret) {
3661 kfree(ins_data);
3662 return ret;
3663 }
Chris Mason31ff1cd2008-09-11 16:17:57 -04003664
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003665 for (i = 0; i < nr; i++, dst_path->slots[0]++) {
Chris Mason31ff1cd2008-09-11 16:17:57 -04003666 dst_offset = btrfs_item_ptr_offset(dst_path->nodes[0],
3667 dst_path->slots[0]);
3668
3669 src_offset = btrfs_item_ptr_offset(src, start_slot + i);
3670
Josef Bacik16e75492013-10-22 12:18:51 -04003671 if ((i == (nr - 1)))
3672 last_key = ins_keys[i];
3673
Josef Bacik94edf4a2012-09-25 14:56:25 -04003674 if (ins_keys[i].type == BTRFS_INODE_ITEM_KEY) {
Chris Mason31ff1cd2008-09-11 16:17:57 -04003675 inode_item = btrfs_item_ptr(dst_path->nodes[0],
3676 dst_path->slots[0],
3677 struct btrfs_inode_item);
Josef Bacik94edf4a2012-09-25 14:56:25 -04003678 fill_inode_item(trans, dst_path->nodes[0], inode_item,
Nikolay Borisov44d70e12017-01-18 00:31:36 +02003679 &inode->vfs_inode, inode_only == LOG_INODE_EXISTS,
Filipe Manana1a4bcf42015-02-13 12:30:56 +00003680 logged_isize);
Josef Bacik94edf4a2012-09-25 14:56:25 -04003681 } else {
3682 copy_extent_buffer(dst_path->nodes[0], src, dst_offset,
3683 src_offset, ins_sizes[i]);
Chris Mason31ff1cd2008-09-11 16:17:57 -04003684 }
Josef Bacik94edf4a2012-09-25 14:56:25 -04003685
Josef Bacik16e75492013-10-22 12:18:51 -04003686 /*
3687 * We set need_find_last_extent here in case we know we were
3688 * processing other items and then walk into the first extent in
3689 * the inode. If we don't hit an extent then nothing changes,
3690 * we'll do the last search the next time around.
3691 */
3692 if (ins_keys[i].type == BTRFS_EXTENT_DATA_KEY) {
3693 has_extents = true;
Filipe Manana74121f7c2014-08-07 12:00:44 +01003694 if (first_key.objectid == (u64)-1)
Josef Bacik16e75492013-10-22 12:18:51 -04003695 first_key = ins_keys[i];
3696 } else {
3697 need_find_last_extent = false;
3698 }
3699
Chris Mason31ff1cd2008-09-11 16:17:57 -04003700 /* take a reference on file data extents so that truncates
3701 * or deletes of this inode don't have to relog the inode
3702 * again
3703 */
David Sterba962a2982014-06-04 18:41:45 +02003704 if (ins_keys[i].type == BTRFS_EXTENT_DATA_KEY &&
Liu Bod2794402012-08-29 01:07:56 -06003705 !skip_csum) {
Chris Mason31ff1cd2008-09-11 16:17:57 -04003706 int found_type;
3707 extent = btrfs_item_ptr(src, start_slot + i,
3708 struct btrfs_file_extent_item);
3709
liubo8e531cd2011-05-06 10:36:09 +08003710 if (btrfs_file_extent_generation(src, extent) < trans->transid)
3711 continue;
3712
Chris Mason31ff1cd2008-09-11 16:17:57 -04003713 found_type = btrfs_file_extent_type(src, extent);
Josef Bacik6f1fed72012-09-26 11:07:06 -04003714 if (found_type == BTRFS_FILE_EXTENT_REG) {
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003715 u64 ds, dl, cs, cl;
3716 ds = btrfs_file_extent_disk_bytenr(src,
3717 extent);
3718 /* ds == 0 is a hole */
3719 if (ds == 0)
3720 continue;
3721
3722 dl = btrfs_file_extent_disk_num_bytes(src,
3723 extent);
3724 cs = btrfs_file_extent_offset(src, extent);
3725 cl = btrfs_file_extent_num_bytes(src,
Joe Perchesa419aef2009-08-18 11:18:35 -07003726 extent);
Chris Mason580afd72008-12-08 19:15:39 -05003727 if (btrfs_file_extent_compression(src,
3728 extent)) {
3729 cs = 0;
3730 cl = dl;
3731 }
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003732
3733 ret = btrfs_lookup_csums_range(
Jeff Mahoney0b246af2016-06-22 18:54:23 -04003734 fs_info->csum_root,
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003735 ds + cs, ds + cs + cl - 1,
Arne Jansena2de7332011-03-08 14:14:00 +01003736 &ordered_sums, 0);
Josef Bacik36508602013-04-25 16:23:32 -04003737 if (ret) {
3738 btrfs_release_path(dst_path);
3739 kfree(ins_data);
3740 return ret;
3741 }
Chris Mason31ff1cd2008-09-11 16:17:57 -04003742 }
3743 }
Chris Mason31ff1cd2008-09-11 16:17:57 -04003744 }
3745
3746 btrfs_mark_buffer_dirty(dst_path->nodes[0]);
David Sterbab3b4aa72011-04-21 01:20:15 +02003747 btrfs_release_path(dst_path);
Chris Mason31ff1cd2008-09-11 16:17:57 -04003748 kfree(ins_data);
Chris Masond20f7042008-12-08 16:58:54 -05003749
3750 /*
3751 * we have to do this after the loop above to avoid changing the
3752 * log tree while trying to change the log tree.
3753 */
Yan, Zheng4a500fd2010-05-16 10:49:59 -04003754 ret = 0;
Chris Masond3977122009-01-05 21:25:51 -05003755 while (!list_empty(&ordered_sums)) {
Chris Masond20f7042008-12-08 16:58:54 -05003756 struct btrfs_ordered_sum *sums = list_entry(ordered_sums.next,
3757 struct btrfs_ordered_sum,
3758 list);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04003759 if (!ret)
3760 ret = btrfs_csum_file_blocks(trans, log, sums);
Chris Masond20f7042008-12-08 16:58:54 -05003761 list_del(&sums->list);
3762 kfree(sums);
3763 }
Josef Bacik16e75492013-10-22 12:18:51 -04003764
3765 if (!has_extents)
3766 return ret;
3767
Filipe Manana74121f7c2014-08-07 12:00:44 +01003768 if (need_find_last_extent && *last_extent == first_key.offset) {
3769 /*
3770 * We don't have any leafs between our current one and the one
3771 * we processed before that can have file extent items for our
3772 * inode (and have a generation number smaller than our current
3773 * transaction id).
3774 */
3775 need_find_last_extent = false;
3776 }
3777
Josef Bacik16e75492013-10-22 12:18:51 -04003778 /*
3779 * Because we use btrfs_search_forward we could skip leaves that were
3780 * not modified and then assume *last_extent is valid when it really
3781 * isn't. So back up to the previous leaf and read the end of the last
3782 * extent before we go and fill in holes.
3783 */
3784 if (need_find_last_extent) {
3785 u64 len;
3786
Nikolay Borisov44d70e12017-01-18 00:31:36 +02003787 ret = btrfs_prev_leaf(inode->root, src_path);
Josef Bacik16e75492013-10-22 12:18:51 -04003788 if (ret < 0)
3789 return ret;
3790 if (ret)
3791 goto fill_holes;
3792 if (src_path->slots[0])
3793 src_path->slots[0]--;
3794 src = src_path->nodes[0];
3795 btrfs_item_key_to_cpu(src, &key, src_path->slots[0]);
Nikolay Borisov44d70e12017-01-18 00:31:36 +02003796 if (key.objectid != btrfs_ino(inode) ||
Josef Bacik16e75492013-10-22 12:18:51 -04003797 key.type != BTRFS_EXTENT_DATA_KEY)
3798 goto fill_holes;
3799 extent = btrfs_item_ptr(src, src_path->slots[0],
3800 struct btrfs_file_extent_item);
3801 if (btrfs_file_extent_type(src, extent) ==
3802 BTRFS_FILE_EXTENT_INLINE) {
Chris Mason514ac8a2014-01-03 21:07:00 -08003803 len = btrfs_file_extent_inline_len(src,
3804 src_path->slots[0],
3805 extent);
Josef Bacik16e75492013-10-22 12:18:51 -04003806 *last_extent = ALIGN(key.offset + len,
Jeff Mahoney0b246af2016-06-22 18:54:23 -04003807 fs_info->sectorsize);
Josef Bacik16e75492013-10-22 12:18:51 -04003808 } else {
3809 len = btrfs_file_extent_num_bytes(src, extent);
3810 *last_extent = key.offset + len;
3811 }
3812 }
3813fill_holes:
3814 /* So we did prev_leaf, now we need to move to the next leaf, but a few
3815 * things could have happened
3816 *
3817 * 1) A merge could have happened, so we could currently be on a leaf
3818 * that holds what we were copying in the first place.
3819 * 2) A split could have happened, and now not all of the items we want
3820 * are on the same leaf.
3821 *
3822 * So we need to adjust how we search for holes, we need to drop the
3823 * path and re-search for the first extent key we found, and then walk
3824 * forward until we hit the last one we copied.
3825 */
3826 if (need_find_last_extent) {
3827 /* btrfs_prev_leaf could return 1 without releasing the path */
3828 btrfs_release_path(src_path);
Nikolay Borisov44d70e12017-01-18 00:31:36 +02003829 ret = btrfs_search_slot(NULL, inode->root, &first_key, src_path, 0, 0);
Josef Bacik16e75492013-10-22 12:18:51 -04003830 if (ret < 0)
3831 return ret;
3832 ASSERT(ret == 0);
3833 src = src_path->nodes[0];
3834 i = src_path->slots[0];
3835 } else {
3836 i = start_slot;
3837 }
3838
3839 /*
3840 * Ok so here we need to go through and fill in any holes we may have
3841 * to make sure that holes are punched for those areas in case they had
3842 * extents previously.
3843 */
3844 while (!done) {
3845 u64 offset, len;
3846 u64 extent_end;
3847
3848 if (i >= btrfs_header_nritems(src_path->nodes[0])) {
Nikolay Borisov44d70e12017-01-18 00:31:36 +02003849 ret = btrfs_next_leaf(inode->root, src_path);
Josef Bacik16e75492013-10-22 12:18:51 -04003850 if (ret < 0)
3851 return ret;
3852 ASSERT(ret == 0);
3853 src = src_path->nodes[0];
3854 i = 0;
3855 }
3856
3857 btrfs_item_key_to_cpu(src, &key, i);
3858 if (!btrfs_comp_cpu_keys(&key, &last_key))
3859 done = true;
Nikolay Borisov44d70e12017-01-18 00:31:36 +02003860 if (key.objectid != btrfs_ino(inode) ||
Josef Bacik16e75492013-10-22 12:18:51 -04003861 key.type != BTRFS_EXTENT_DATA_KEY) {
3862 i++;
3863 continue;
3864 }
3865 extent = btrfs_item_ptr(src, i, struct btrfs_file_extent_item);
3866 if (btrfs_file_extent_type(src, extent) ==
3867 BTRFS_FILE_EXTENT_INLINE) {
Chris Mason514ac8a2014-01-03 21:07:00 -08003868 len = btrfs_file_extent_inline_len(src, i, extent);
Jeff Mahoneyda170662016-06-15 09:22:56 -04003869 extent_end = ALIGN(key.offset + len,
Jeff Mahoney0b246af2016-06-22 18:54:23 -04003870 fs_info->sectorsize);
Josef Bacik16e75492013-10-22 12:18:51 -04003871 } else {
3872 len = btrfs_file_extent_num_bytes(src, extent);
3873 extent_end = key.offset + len;
3874 }
3875 i++;
3876
3877 if (*last_extent == key.offset) {
3878 *last_extent = extent_end;
3879 continue;
3880 }
3881 offset = *last_extent;
3882 len = key.offset - *last_extent;
Nikolay Borisov44d70e12017-01-18 00:31:36 +02003883 ret = btrfs_insert_file_extent(trans, log, btrfs_ino(inode),
3884 offset, 0, 0, len, 0, len, 0, 0, 0);
Josef Bacik16e75492013-10-22 12:18:51 -04003885 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,
Nikolay Borisov9d122622017-01-18 00:31:40 +02004057 struct btrfs_inode *inode, struct btrfs_root *root,
Filipe Manana8407f552014-09-05 15:14:39 +01004058 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
Nikolay Borisov9d122622017-01-18 00:31:40 +02004074 ret = wait_ordered_extents(trans, &inode->vfs_inode, root, em, logged_list,
Filipe Manana8407f552014-09-05 15:14:39 +01004075 &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
Nikolay Borisov9d122622017-01-18 00:31:40 +02004086 ret = __btrfs_drop_extents(trans, log, &inode->vfs_inode, path, em->start,
Filipe Manana8407f552014-09-05 15:14:39 +01004087 em->start + em->len, NULL, 0, 1,
4088 sizeof(*fi), &extent_inserted);
4089 if (ret)
4090 return ret;
4091
4092 if (!extent_inserted) {
Nikolay Borisov9d122622017-01-18 00:31:40 +02004093 key.objectid = btrfs_ino(inode);
Filipe Manana8407f552014-09-05 15:14:39 +01004094 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,
Nikolay Borisov9d122622017-01-18 00:31:40 +02004152 struct btrfs_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;
Nikolay Borisov9d122622017-01-18 00:31:40 +02004161 struct extent_map_tree *tree = &inode->extent_tree;
Josef Bacik5dc562c2012-08-17 13:14:17 -04004162 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
Nikolay Borisov9d122622017-01-18 00:31:40 +02004168 down_write(&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);
Nikolay Borisov9d122622017-01-18 00:31:40 +02004197 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 */
Nikolay Borisov9d122622017-01-18 00:31:40 +02004208 ret = filemap_check_errors(inode->vfs_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);
Nikolay Borisov9d122622017-01-18 00:31:40 +02004237 up_write(&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
Nikolay Borisov481b01c2017-01-18 00:31:34 +02004243static int logged_inode_size(struct btrfs_root *log, struct btrfs_inode *inode,
Filipe Manana1a4bcf42015-02-13 12:30:56 +00004244 struct btrfs_path *path, u64 *size_ret)
4245{
4246 struct btrfs_key key;
4247 int ret;
4248
Nikolay Borisov481b01c2017-01-18 00:31:34 +02004249 key.objectid = btrfs_ino(inode);
Filipe Manana1a4bcf42015-02-13 12:30:56 +00004250 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,
Nikolay Borisov1a93c362017-01-18 00:31:37 +02004281 struct btrfs_inode *inode,
Filipe Manana36283bf2015-06-20 00:44:51 +01004282 struct btrfs_path *path,
4283 struct btrfs_path *dst_path)
4284{
4285 int ret;
4286 struct btrfs_key key;
Nikolay Borisov1a93c362017-01-18 00:31:37 +02004287 const u64 ino = btrfs_ino(inode);
Filipe Manana36283bf2015-06-20 00:44:51 +01004288 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
Nikolay Borisov1a93c362017-01-18 00:31:37 +02004308 ret = copy_items(trans, inode, dst_path, path,
Filipe Manana36283bf2015-06-20 00:44:51 +01004309 &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
Nikolay Borisov1a93c362017-01-18 00:31:37 +02004338 ret = copy_items(trans, inode, dst_path, path,
Filipe Manana36283bf2015-06-20 00:44:51 +01004339 &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,
Nikolay Borisova0308dd2017-01-18 00:31:38 +02004376 struct btrfs_inode *inode,
Filipe Mananaa89ca6f2015-06-25 04:17:46 +01004377 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;
Nikolay Borisova0308dd2017-01-18 00:31:38 +02004386 const u64 ino = btrfs_ino(inode);
4387 const u64 i_size = i_size_read(&inode->vfs_inode);
Filipe Mananaa89ca6f2015-06-25 04:17:46 +01004388
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,
Nikolay Borisov4791c8f2017-01-18 00:31:35 +02004497 struct btrfs_inode *inode,
Filipe Manana44f714d2016-06-06 16:11:13 +01004498 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);
Nikolay Borisov4791c8f2017-01-18 00:31:35 +02004553 di = btrfs_lookup_dir_item(NULL, inode->root, search_path,
4554 parent, name, this_name_len, 0);
Filipe Manana56f23fd2016-03-30 23:37:21 +01004555 if (di && !IS_ERR(di)) {
Filipe Manana44f714d2016-06-06 16:11:13 +01004556 struct btrfs_key di_key;
4557
4558 btrfs_dir_item_key_to_cpu(search_path->nodes[0],
4559 di, &di_key);
4560 if (di_key.type == BTRFS_INODE_ITEM_KEY) {
4561 ret = 1;
4562 *other_ino = di_key.objectid;
4563 } else {
4564 ret = -EAGAIN;
4565 }
Filipe Manana56f23fd2016-03-30 23:37:21 +01004566 goto out;
4567 } else if (IS_ERR(di)) {
4568 ret = PTR_ERR(di);
4569 goto out;
4570 }
4571 btrfs_release_path(search_path);
4572
4573 cur_offset += this_len;
4574 }
4575 ret = 0;
4576out:
4577 btrfs_free_path(search_path);
4578 kfree(name);
4579 return ret;
4580}
4581
Chris Masone02119d2008-09-05 16:13:11 -04004582/* log a single inode in the tree log.
4583 * At least one parent directory for this inode must exist in the tree
4584 * or be logged already.
4585 *
4586 * Any items from this inode changed by the current transaction are copied
4587 * to the log tree. An extra reference is taken on any extents in this
4588 * file, allowing us to avoid a whole pile of corner cases around logging
4589 * blocks that have been removed from the tree.
4590 *
4591 * See LOG_INODE_ALL and related defines for a description of what inode_only
4592 * does.
4593 *
4594 * This handles both files and directories.
4595 */
Chris Mason12fcfd22009-03-24 10:24:20 -04004596static int btrfs_log_inode(struct btrfs_trans_handle *trans,
Nikolay Borisova59108a2017-01-18 00:31:48 +02004597 struct btrfs_root *root, struct btrfs_inode *inode,
Filipe Manana49dae1b2014-09-06 22:34:39 +01004598 int inode_only,
4599 const loff_t start,
Filipe Manana8407f552014-09-05 15:14:39 +01004600 const loff_t end,
4601 struct btrfs_log_ctx *ctx)
Chris Masone02119d2008-09-05 16:13:11 -04004602{
Jeff Mahoney0b246af2016-06-22 18:54:23 -04004603 struct btrfs_fs_info *fs_info = root->fs_info;
Chris Masone02119d2008-09-05 16:13:11 -04004604 struct btrfs_path *path;
4605 struct btrfs_path *dst_path;
4606 struct btrfs_key min_key;
4607 struct btrfs_key max_key;
4608 struct btrfs_root *log = root->log_root;
Chris Mason31ff1cd2008-09-11 16:17:57 -04004609 struct extent_buffer *src = NULL;
Miao Xie827463c2014-01-14 20:31:51 +08004610 LIST_HEAD(logged_list);
Josef Bacik16e75492013-10-22 12:18:51 -04004611 u64 last_extent = 0;
Yan, Zheng4a500fd2010-05-16 10:49:59 -04004612 int err = 0;
Chris Masone02119d2008-09-05 16:13:11 -04004613 int ret;
Chris Mason3a5f1d42008-09-11 15:53:37 -04004614 int nritems;
Chris Mason31ff1cd2008-09-11 16:17:57 -04004615 int ins_start_slot = 0;
4616 int ins_nr;
Josef Bacik5dc562c2012-08-17 13:14:17 -04004617 bool fast_search = false;
Nikolay Borisova59108a2017-01-18 00:31:48 +02004618 u64 ino = btrfs_ino(inode);
4619 struct extent_map_tree *em_tree = &inode->extent_tree;
Filipe Manana1a4bcf42015-02-13 12:30:56 +00004620 u64 logged_isize = 0;
Filipe Mananae4545de2015-06-17 12:49:23 +01004621 bool need_log_inode_item = true;
Chris Masone02119d2008-09-05 16:13:11 -04004622
Chris Masone02119d2008-09-05 16:13:11 -04004623 path = btrfs_alloc_path();
Tsutomu Itoh5df67082011-02-01 09:17:35 +00004624 if (!path)
4625 return -ENOMEM;
Chris Masone02119d2008-09-05 16:13:11 -04004626 dst_path = btrfs_alloc_path();
Tsutomu Itoh5df67082011-02-01 09:17:35 +00004627 if (!dst_path) {
4628 btrfs_free_path(path);
4629 return -ENOMEM;
4630 }
Chris Masone02119d2008-09-05 16:13:11 -04004631
Li Zefan33345d012011-04-20 10:31:50 +08004632 min_key.objectid = ino;
Chris Masone02119d2008-09-05 16:13:11 -04004633 min_key.type = BTRFS_INODE_ITEM_KEY;
4634 min_key.offset = 0;
4635
Li Zefan33345d012011-04-20 10:31:50 +08004636 max_key.objectid = ino;
Chris Mason12fcfd22009-03-24 10:24:20 -04004637
Chris Mason12fcfd22009-03-24 10:24:20 -04004638
Josef Bacik5dc562c2012-08-17 13:14:17 -04004639 /* today the code can only do partial logging of directories */
Nikolay Borisova59108a2017-01-18 00:31:48 +02004640 if (S_ISDIR(inode->vfs_inode.i_mode) ||
Miao Xie5269b672012-11-01 07:35:23 +00004641 (!test_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
Nikolay Borisova59108a2017-01-18 00:31:48 +02004642 &inode->runtime_flags) &&
Liu Bo781feef2016-11-30 16:20:25 -08004643 inode_only >= LOG_INODE_EXISTS))
Chris Masone02119d2008-09-05 16:13:11 -04004644 max_key.type = BTRFS_XATTR_ITEM_KEY;
4645 else
4646 max_key.type = (u8)-1;
4647 max_key.offset = (u64)-1;
4648
Filipe Manana2c2c4522015-01-13 16:40:04 +00004649 /*
4650 * Only run delayed items if we are a dir or a new file.
4651 * Otherwise commit the delayed inode only, which is needed in
4652 * order for the log replay code to mark inodes for link count
4653 * fixup (create temporary BTRFS_TREE_LOG_FIXUP_OBJECTID items).
4654 */
Nikolay Borisova59108a2017-01-18 00:31:48 +02004655 if (S_ISDIR(inode->vfs_inode.i_mode) ||
4656 inode->generation > fs_info->last_trans_committed)
4657 ret = btrfs_commit_inode_delayed_items(trans, inode);
Filipe Manana2c2c4522015-01-13 16:40:04 +00004658 else
Nikolay Borisova59108a2017-01-18 00:31:48 +02004659 ret = btrfs_commit_inode_delayed_inode(inode);
Filipe Manana2c2c4522015-01-13 16:40:04 +00004660
4661 if (ret) {
4662 btrfs_free_path(path);
4663 btrfs_free_path(dst_path);
4664 return ret;
Miao Xie16cdcec2011-04-22 18:12:22 +08004665 }
4666
Liu Bo781feef2016-11-30 16:20:25 -08004667 if (inode_only == LOG_OTHER_INODE) {
4668 inode_only = LOG_INODE_EXISTS;
Nikolay Borisova59108a2017-01-18 00:31:48 +02004669 mutex_lock_nested(&inode->log_mutex, SINGLE_DEPTH_NESTING);
Liu Bo781feef2016-11-30 16:20:25 -08004670 } else {
Nikolay Borisova59108a2017-01-18 00:31:48 +02004671 mutex_lock(&inode->log_mutex);
Liu Bo781feef2016-11-30 16:20:25 -08004672 }
Chris Masone02119d2008-09-05 16:13:11 -04004673
Filipe Manana5e33a2b2016-02-25 23:19:38 +00004674 /*
Chris Masone02119d2008-09-05 16:13:11 -04004675 * a brute force approach to making sure we get the most uptodate
4676 * copies of everything.
4677 */
Nikolay Borisova59108a2017-01-18 00:31:48 +02004678 if (S_ISDIR(inode->vfs_inode.i_mode)) {
Chris Masone02119d2008-09-05 16:13:11 -04004679 int max_key_type = BTRFS_DIR_LOG_INDEX_KEY;
4680
Filipe Manana4f764e52015-02-23 19:53:35 +00004681 if (inode_only == LOG_INODE_EXISTS)
4682 max_key_type = BTRFS_XATTR_ITEM_KEY;
Li Zefan33345d012011-04-20 10:31:50 +08004683 ret = drop_objectid_items(trans, log, path, ino, max_key_type);
Chris Masone02119d2008-09-05 16:13:11 -04004684 } else {
Filipe Manana1a4bcf42015-02-13 12:30:56 +00004685 if (inode_only == LOG_INODE_EXISTS) {
4686 /*
4687 * Make sure the new inode item we write to the log has
4688 * the same isize as the current one (if it exists).
4689 * This is necessary to prevent data loss after log
4690 * replay, and also to prevent doing a wrong expanding
4691 * truncate - for e.g. create file, write 4K into offset
4692 * 0, fsync, write 4K into offset 4096, add hard link,
4693 * fsync some other file (to sync log), power fail - if
4694 * we use the inode's current i_size, after log replay
4695 * we get a 8Kb file, with the last 4Kb extent as a hole
4696 * (zeroes), as if an expanding truncate happened,
4697 * instead of getting a file of 4Kb only.
4698 */
Nikolay Borisova59108a2017-01-18 00:31:48 +02004699 err = logged_inode_size(log, inode, path, &logged_isize);
Filipe Manana1a4bcf42015-02-13 12:30:56 +00004700 if (err)
4701 goto out_unlock;
4702 }
Filipe Mananaa7429942015-02-13 16:56:14 +00004703 if (test_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
Nikolay Borisova59108a2017-01-18 00:31:48 +02004704 &inode->runtime_flags)) {
Filipe Mananaa7429942015-02-13 16:56:14 +00004705 if (inode_only == LOG_INODE_EXISTS) {
Filipe Manana4f764e52015-02-23 19:53:35 +00004706 max_key.type = BTRFS_XATTR_ITEM_KEY;
Filipe Mananaa7429942015-02-13 16:56:14 +00004707 ret = drop_objectid_items(trans, log, path, ino,
4708 max_key.type);
4709 } else {
4710 clear_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
Nikolay Borisova59108a2017-01-18 00:31:48 +02004711 &inode->runtime_flags);
Filipe Mananaa7429942015-02-13 16:56:14 +00004712 clear_bit(BTRFS_INODE_COPY_EVERYTHING,
Nikolay Borisova59108a2017-01-18 00:31:48 +02004713 &inode->runtime_flags);
Chris Mason28ed1342014-12-17 09:41:04 -08004714 while(1) {
4715 ret = btrfs_truncate_inode_items(trans,
Nikolay Borisova59108a2017-01-18 00:31:48 +02004716 log, &inode->vfs_inode, 0, 0);
Chris Mason28ed1342014-12-17 09:41:04 -08004717 if (ret != -EAGAIN)
4718 break;
4719 }
Filipe Mananaa7429942015-02-13 16:56:14 +00004720 }
Filipe Manana4f764e52015-02-23 19:53:35 +00004721 } else if (test_and_clear_bit(BTRFS_INODE_COPY_EVERYTHING,
Nikolay Borisova59108a2017-01-18 00:31:48 +02004722 &inode->runtime_flags) ||
Josef Bacik6cfab852013-11-12 16:25:58 -05004723 inode_only == LOG_INODE_EXISTS) {
Filipe Manana4f764e52015-02-23 19:53:35 +00004724 if (inode_only == LOG_INODE_ALL)
Josef Bacika95249b2012-10-11 16:17:34 -04004725 fast_search = true;
Filipe Manana4f764e52015-02-23 19:53:35 +00004726 max_key.type = BTRFS_XATTR_ITEM_KEY;
Josef Bacika95249b2012-10-11 16:17:34 -04004727 ret = drop_objectid_items(trans, log, path, ino,
4728 max_key.type);
Josef Bacik5dc562c2012-08-17 13:14:17 -04004729 } else {
Liu Bo183f37f2012-11-01 06:38:47 +00004730 if (inode_only == LOG_INODE_ALL)
4731 fast_search = true;
Josef Bacika95249b2012-10-11 16:17:34 -04004732 goto log_extents;
Josef Bacik5dc562c2012-08-17 13:14:17 -04004733 }
Josef Bacika95249b2012-10-11 16:17:34 -04004734
Chris Masone02119d2008-09-05 16:13:11 -04004735 }
Yan, Zheng4a500fd2010-05-16 10:49:59 -04004736 if (ret) {
4737 err = ret;
4738 goto out_unlock;
4739 }
Chris Masone02119d2008-09-05 16:13:11 -04004740
Chris Masond3977122009-01-05 21:25:51 -05004741 while (1) {
Chris Mason31ff1cd2008-09-11 16:17:57 -04004742 ins_nr = 0;
Filipe David Borba Manana6174d3c2013-10-01 16:13:42 +01004743 ret = btrfs_search_forward(root, &min_key,
Eric Sandeende78b512013-01-31 18:21:12 +00004744 path, trans->transid);
Liu Bofb770ae2016-07-05 12:10:14 -07004745 if (ret < 0) {
4746 err = ret;
4747 goto out_unlock;
4748 }
Chris Masone02119d2008-09-05 16:13:11 -04004749 if (ret != 0)
4750 break;
Chris Mason3a5f1d42008-09-11 15:53:37 -04004751again:
Chris Mason31ff1cd2008-09-11 16:17:57 -04004752 /* note, ins_nr might be > 0 here, cleanup outside the loop */
Li Zefan33345d012011-04-20 10:31:50 +08004753 if (min_key.objectid != ino)
Chris Masone02119d2008-09-05 16:13:11 -04004754 break;
4755 if (min_key.type > max_key.type)
4756 break;
Chris Mason31ff1cd2008-09-11 16:17:57 -04004757
Filipe Mananae4545de2015-06-17 12:49:23 +01004758 if (min_key.type == BTRFS_INODE_ITEM_KEY)
4759 need_log_inode_item = false;
4760
Filipe Manana56f23fd2016-03-30 23:37:21 +01004761 if ((min_key.type == BTRFS_INODE_REF_KEY ||
4762 min_key.type == BTRFS_INODE_EXTREF_KEY) &&
Nikolay Borisova59108a2017-01-18 00:31:48 +02004763 inode->generation == trans->transid) {
Filipe Manana44f714d2016-06-06 16:11:13 +01004764 u64 other_ino = 0;
4765
Filipe Manana56f23fd2016-03-30 23:37:21 +01004766 ret = btrfs_check_ref_name_override(path->nodes[0],
Nikolay Borisova59108a2017-01-18 00:31:48 +02004767 path->slots[0], &min_key, inode,
4768 &other_ino);
Filipe Manana56f23fd2016-03-30 23:37:21 +01004769 if (ret < 0) {
4770 err = ret;
4771 goto out_unlock;
Filipe Manana28a23592016-08-23 21:13:51 +01004772 } else if (ret > 0 && ctx &&
Nikolay Borisov4a0cc7c2017-01-10 20:35:31 +02004773 other_ino != btrfs_ino(BTRFS_I(ctx->inode))) {
Filipe Manana44f714d2016-06-06 16:11:13 +01004774 struct btrfs_key inode_key;
4775 struct inode *other_inode;
4776
4777 if (ins_nr > 0) {
4778 ins_nr++;
4779 } else {
4780 ins_nr = 1;
4781 ins_start_slot = path->slots[0];
4782 }
Nikolay Borisova59108a2017-01-18 00:31:48 +02004783 ret = copy_items(trans, inode, dst_path, path,
Filipe Manana44f714d2016-06-06 16:11:13 +01004784 &last_extent, ins_start_slot,
4785 ins_nr, inode_only,
4786 logged_isize);
4787 if (ret < 0) {
4788 err = ret;
4789 goto out_unlock;
4790 }
4791 ins_nr = 0;
4792 btrfs_release_path(path);
4793 inode_key.objectid = other_ino;
4794 inode_key.type = BTRFS_INODE_ITEM_KEY;
4795 inode_key.offset = 0;
Jeff Mahoney0b246af2016-06-22 18:54:23 -04004796 other_inode = btrfs_iget(fs_info->sb,
Filipe Manana44f714d2016-06-06 16:11:13 +01004797 &inode_key, root,
4798 NULL);
4799 /*
4800 * If the other inode that had a conflicting dir
4801 * entry was deleted in the current transaction,
4802 * we don't need to do more work nor fallback to
4803 * a transaction commit.
4804 */
4805 if (IS_ERR(other_inode) &&
4806 PTR_ERR(other_inode) == -ENOENT) {
4807 goto next_key;
4808 } else if (IS_ERR(other_inode)) {
4809 err = PTR_ERR(other_inode);
4810 goto out_unlock;
4811 }
4812 /*
4813 * We are safe logging the other inode without
4814 * acquiring its i_mutex as long as we log with
4815 * the LOG_INODE_EXISTS mode. We're safe against
4816 * concurrent renames of the other inode as well
4817 * because during a rename we pin the log and
4818 * update the log with the new name before we
4819 * unpin it.
4820 */
Nikolay Borisova59108a2017-01-18 00:31:48 +02004821 err = btrfs_log_inode(trans, root,
4822 BTRFS_I(other_inode),
4823 LOG_OTHER_INODE, 0, LLONG_MAX,
4824 ctx);
Filipe Manana44f714d2016-06-06 16:11:13 +01004825 iput(other_inode);
4826 if (err)
4827 goto out_unlock;
4828 else
4829 goto next_key;
Filipe Manana56f23fd2016-03-30 23:37:21 +01004830 }
4831 }
4832
Filipe Manana36283bf2015-06-20 00:44:51 +01004833 /* Skip xattrs, we log them later with btrfs_log_all_xattrs() */
4834 if (min_key.type == BTRFS_XATTR_ITEM_KEY) {
4835 if (ins_nr == 0)
4836 goto next_slot;
Nikolay Borisova59108a2017-01-18 00:31:48 +02004837 ret = copy_items(trans, inode, dst_path, path,
Filipe Manana36283bf2015-06-20 00:44:51 +01004838 &last_extent, ins_start_slot,
4839 ins_nr, inode_only, logged_isize);
4840 if (ret < 0) {
4841 err = ret;
4842 goto out_unlock;
4843 }
4844 ins_nr = 0;
4845 if (ret) {
4846 btrfs_release_path(path);
4847 continue;
4848 }
4849 goto next_slot;
4850 }
4851
Chris Masone02119d2008-09-05 16:13:11 -04004852 src = path->nodes[0];
Chris Mason31ff1cd2008-09-11 16:17:57 -04004853 if (ins_nr && ins_start_slot + ins_nr == path->slots[0]) {
4854 ins_nr++;
4855 goto next_slot;
4856 } else if (!ins_nr) {
4857 ins_start_slot = path->slots[0];
4858 ins_nr = 1;
4859 goto next_slot;
Chris Masone02119d2008-09-05 16:13:11 -04004860 }
4861
Nikolay Borisova59108a2017-01-18 00:31:48 +02004862 ret = copy_items(trans, inode, dst_path, path, &last_extent,
Filipe Manana1a4bcf42015-02-13 12:30:56 +00004863 ins_start_slot, ins_nr, inode_only,
4864 logged_isize);
Josef Bacik16e75492013-10-22 12:18:51 -04004865 if (ret < 0) {
Yan, Zheng4a500fd2010-05-16 10:49:59 -04004866 err = ret;
4867 goto out_unlock;
Rasmus Villemoesa71db862014-06-20 21:51:43 +02004868 }
4869 if (ret) {
Josef Bacik16e75492013-10-22 12:18:51 -04004870 ins_nr = 0;
4871 btrfs_release_path(path);
4872 continue;
Yan, Zheng4a500fd2010-05-16 10:49:59 -04004873 }
Chris Mason31ff1cd2008-09-11 16:17:57 -04004874 ins_nr = 1;
4875 ins_start_slot = path->slots[0];
4876next_slot:
Chris Masone02119d2008-09-05 16:13:11 -04004877
Chris Mason3a5f1d42008-09-11 15:53:37 -04004878 nritems = btrfs_header_nritems(path->nodes[0]);
4879 path->slots[0]++;
4880 if (path->slots[0] < nritems) {
4881 btrfs_item_key_to_cpu(path->nodes[0], &min_key,
4882 path->slots[0]);
4883 goto again;
4884 }
Chris Mason31ff1cd2008-09-11 16:17:57 -04004885 if (ins_nr) {
Nikolay Borisova59108a2017-01-18 00:31:48 +02004886 ret = copy_items(trans, inode, dst_path, path,
Josef Bacik16e75492013-10-22 12:18:51 -04004887 &last_extent, ins_start_slot,
Filipe Manana1a4bcf42015-02-13 12:30:56 +00004888 ins_nr, inode_only, logged_isize);
Josef Bacik16e75492013-10-22 12:18:51 -04004889 if (ret < 0) {
Yan, Zheng4a500fd2010-05-16 10:49:59 -04004890 err = ret;
4891 goto out_unlock;
4892 }
Josef Bacik16e75492013-10-22 12:18:51 -04004893 ret = 0;
Chris Mason31ff1cd2008-09-11 16:17:57 -04004894 ins_nr = 0;
4895 }
David Sterbab3b4aa72011-04-21 01:20:15 +02004896 btrfs_release_path(path);
Filipe Manana44f714d2016-06-06 16:11:13 +01004897next_key:
Filipe David Borba Manana3d41d702013-10-01 17:06:53 +01004898 if (min_key.offset < (u64)-1) {
Chris Masone02119d2008-09-05 16:13:11 -04004899 min_key.offset++;
Filipe David Borba Manana3d41d702013-10-01 17:06:53 +01004900 } else if (min_key.type < max_key.type) {
Chris Masone02119d2008-09-05 16:13:11 -04004901 min_key.type++;
Filipe David Borba Manana3d41d702013-10-01 17:06:53 +01004902 min_key.offset = 0;
4903 } else {
Chris Masone02119d2008-09-05 16:13:11 -04004904 break;
Filipe David Borba Manana3d41d702013-10-01 17:06:53 +01004905 }
Chris Masone02119d2008-09-05 16:13:11 -04004906 }
Chris Mason31ff1cd2008-09-11 16:17:57 -04004907 if (ins_nr) {
Nikolay Borisova59108a2017-01-18 00:31:48 +02004908 ret = copy_items(trans, inode, dst_path, path, &last_extent,
Filipe Manana1a4bcf42015-02-13 12:30:56 +00004909 ins_start_slot, ins_nr, inode_only,
4910 logged_isize);
Josef Bacik16e75492013-10-22 12:18:51 -04004911 if (ret < 0) {
Yan, Zheng4a500fd2010-05-16 10:49:59 -04004912 err = ret;
4913 goto out_unlock;
4914 }
Josef Bacik16e75492013-10-22 12:18:51 -04004915 ret = 0;
Chris Mason31ff1cd2008-09-11 16:17:57 -04004916 ins_nr = 0;
4917 }
Josef Bacik5dc562c2012-08-17 13:14:17 -04004918
Filipe Manana36283bf2015-06-20 00:44:51 +01004919 btrfs_release_path(path);
4920 btrfs_release_path(dst_path);
Nikolay Borisova59108a2017-01-18 00:31:48 +02004921 err = btrfs_log_all_xattrs(trans, root, inode, path, dst_path);
Filipe Manana36283bf2015-06-20 00:44:51 +01004922 if (err)
4923 goto out_unlock;
Filipe Mananaa89ca6f2015-06-25 04:17:46 +01004924 if (max_key.type >= BTRFS_EXTENT_DATA_KEY && !fast_search) {
4925 btrfs_release_path(path);
4926 btrfs_release_path(dst_path);
Nikolay Borisova59108a2017-01-18 00:31:48 +02004927 err = btrfs_log_trailing_hole(trans, root, inode, path);
Filipe Mananaa89ca6f2015-06-25 04:17:46 +01004928 if (err)
4929 goto out_unlock;
4930 }
Josef Bacika95249b2012-10-11 16:17:34 -04004931log_extents:
Josef Bacikf3b15cc2013-07-22 12:54:30 -04004932 btrfs_release_path(path);
4933 btrfs_release_path(dst_path);
Filipe Mananae4545de2015-06-17 12:49:23 +01004934 if (need_log_inode_item) {
Nikolay Borisova59108a2017-01-18 00:31:48 +02004935 err = log_inode_item(trans, log, dst_path, inode);
Filipe Mananae4545de2015-06-17 12:49:23 +01004936 if (err)
4937 goto out_unlock;
4938 }
Josef Bacik5dc562c2012-08-17 13:14:17 -04004939 if (fast_search) {
Nikolay Borisova59108a2017-01-18 00:31:48 +02004940 ret = btrfs_log_changed_extents(trans, root, inode, dst_path,
Filipe Mananade0ee0e2016-01-21 10:17:54 +00004941 &logged_list, ctx, start, end);
Josef Bacik5dc562c2012-08-17 13:14:17 -04004942 if (ret) {
4943 err = ret;
4944 goto out_unlock;
4945 }
Josef Bacikd006a042013-11-12 20:54:09 -05004946 } else if (inode_only == LOG_INODE_ALL) {
Liu Bo06d3d222012-08-27 10:52:19 -06004947 struct extent_map *em, *n;
4948
Filipe Manana49dae1b2014-09-06 22:34:39 +01004949 write_lock(&em_tree->lock);
4950 /*
4951 * We can't just remove every em if we're called for a ranged
4952 * fsync - that is, one that doesn't cover the whole possible
4953 * file range (0 to LLONG_MAX). This is because we can have
4954 * em's that fall outside the range we're logging and therefore
4955 * their ordered operations haven't completed yet
4956 * (btrfs_finish_ordered_io() not invoked yet). This means we
4957 * didn't get their respective file extent item in the fs/subvol
4958 * tree yet, and need to let the next fast fsync (one which
4959 * consults the list of modified extent maps) find the em so
4960 * that it logs a matching file extent item and waits for the
4961 * respective ordered operation to complete (if it's still
4962 * running).
4963 *
4964 * Removing every em outside the range we're logging would make
4965 * the next fast fsync not log their matching file extent items,
4966 * therefore making us lose data after a log replay.
4967 */
4968 list_for_each_entry_safe(em, n, &em_tree->modified_extents,
4969 list) {
4970 const u64 mod_end = em->mod_start + em->mod_len - 1;
4971
4972 if (em->mod_start >= start && mod_end <= end)
4973 list_del_init(&em->list);
4974 }
4975 write_unlock(&em_tree->lock);
Josef Bacik5dc562c2012-08-17 13:14:17 -04004976 }
4977
Nikolay Borisova59108a2017-01-18 00:31:48 +02004978 if (inode_only == LOG_INODE_ALL && S_ISDIR(inode->vfs_inode.i_mode)) {
4979 ret = log_directory_changes(trans, root, inode, path, dst_path,
4980 ctx);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04004981 if (ret) {
4982 err = ret;
4983 goto out_unlock;
4984 }
Chris Masone02119d2008-09-05 16:13:11 -04004985 }
Filipe Manana49dae1b2014-09-06 22:34:39 +01004986
Nikolay Borisova59108a2017-01-18 00:31:48 +02004987 spin_lock(&inode->lock);
4988 inode->logged_trans = trans->transid;
4989 inode->last_log_commit = inode->last_sub_trans;
4990 spin_unlock(&inode->lock);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04004991out_unlock:
Miao Xie827463c2014-01-14 20:31:51 +08004992 if (unlikely(err))
4993 btrfs_put_logged_extents(&logged_list);
4994 else
4995 btrfs_submit_logged_extents(&logged_list, log);
Nikolay Borisova59108a2017-01-18 00:31:48 +02004996 mutex_unlock(&inode->log_mutex);
Chris Masone02119d2008-09-05 16:13:11 -04004997
4998 btrfs_free_path(path);
4999 btrfs_free_path(dst_path);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04005000 return err;
Chris Masone02119d2008-09-05 16:13:11 -04005001}
5002
Chris Mason12fcfd22009-03-24 10:24:20 -04005003/*
Filipe Manana2be63d52016-02-12 11:34:23 +00005004 * Check if we must fallback to a transaction commit when logging an inode.
5005 * This must be called after logging the inode and is used only in the context
5006 * when fsyncing an inode requires the need to log some other inode - in which
5007 * case we can't lock the i_mutex of each other inode we need to log as that
5008 * can lead to deadlocks with concurrent fsync against other inodes (as we can
5009 * log inodes up or down in the hierarchy) or rename operations for example. So
5010 * we take the log_mutex of the inode after we have logged it and then check for
5011 * its last_unlink_trans value - this is safe because any task setting
5012 * last_unlink_trans must take the log_mutex and it must do this before it does
5013 * the actual unlink operation, so if we do this check before a concurrent task
5014 * sets last_unlink_trans it means we've logged a consistent version/state of
5015 * all the inode items, otherwise we are not sure and must do a transaction
Nicholas D Steeves01327612016-05-19 21:18:45 -04005016 * commit (the concurrent task might have only updated last_unlink_trans before
Filipe Manana2be63d52016-02-12 11:34:23 +00005017 * we logged the inode or it might have also done the unlink).
5018 */
5019static bool btrfs_must_commit_transaction(struct btrfs_trans_handle *trans,
Nikolay Borisovab1717b2017-01-18 00:31:27 +02005020 struct btrfs_inode *inode)
Filipe Manana2be63d52016-02-12 11:34:23 +00005021{
Nikolay Borisovab1717b2017-01-18 00:31:27 +02005022 struct btrfs_fs_info *fs_info = inode->root->fs_info;
Filipe Manana2be63d52016-02-12 11:34:23 +00005023 bool ret = false;
5024
Nikolay Borisovab1717b2017-01-18 00:31:27 +02005025 mutex_lock(&inode->log_mutex);
5026 if (inode->last_unlink_trans > fs_info->last_trans_committed) {
Filipe Manana2be63d52016-02-12 11:34:23 +00005027 /*
5028 * Make sure any commits to the log are forced to be full
5029 * commits.
5030 */
5031 btrfs_set_log_full_commit(fs_info, trans);
5032 ret = true;
5033 }
Nikolay Borisovab1717b2017-01-18 00:31:27 +02005034 mutex_unlock(&inode->log_mutex);
Filipe Manana2be63d52016-02-12 11:34:23 +00005035
5036 return ret;
5037}
5038
5039/*
Chris Mason12fcfd22009-03-24 10:24:20 -04005040 * follow the dentry parent pointers up the chain and see if any
5041 * of the directories in it require a full commit before they can
5042 * be logged. Returns zero if nothing special needs to be done or 1 if
5043 * a full commit is required.
5044 */
5045static noinline int check_parent_dirs_for_sync(struct btrfs_trans_handle *trans,
5046 struct inode *inode,
5047 struct dentry *parent,
5048 struct super_block *sb,
5049 u64 last_committed)
Chris Masone02119d2008-09-05 16:13:11 -04005050{
Chris Mason12fcfd22009-03-24 10:24:20 -04005051 int ret = 0;
Josef Bacik6a912212010-11-20 09:48:00 +00005052 struct dentry *old_parent = NULL;
Josef Bacikde2b5302013-09-11 09:36:30 -04005053 struct inode *orig_inode = inode;
Chris Masone02119d2008-09-05 16:13:11 -04005054
Chris Masonaf4176b2009-03-24 10:24:31 -04005055 /*
5056 * for regular files, if its inode is already on disk, we don't
5057 * have to worry about the parents at all. This is because
5058 * we can use the last_unlink_trans field to record renames
5059 * and other fun in this file.
5060 */
5061 if (S_ISREG(inode->i_mode) &&
5062 BTRFS_I(inode)->generation <= last_committed &&
5063 BTRFS_I(inode)->last_unlink_trans <= last_committed)
5064 goto out;
5065
Chris Mason12fcfd22009-03-24 10:24:20 -04005066 if (!S_ISDIR(inode->i_mode)) {
Al Virofc640052016-04-10 01:33:30 -04005067 if (!parent || d_really_is_negative(parent) || sb != parent->d_sb)
Chris Mason12fcfd22009-03-24 10:24:20 -04005068 goto out;
David Howells2b0143b2015-03-17 22:25:59 +00005069 inode = d_inode(parent);
Chris Mason12fcfd22009-03-24 10:24:20 -04005070 }
5071
5072 while (1) {
Josef Bacikde2b5302013-09-11 09:36:30 -04005073 /*
5074 * If we are logging a directory then we start with our inode,
Nicholas D Steeves01327612016-05-19 21:18:45 -04005075 * not our parent's inode, so we need to skip setting the
Josef Bacikde2b5302013-09-11 09:36:30 -04005076 * logged_trans so that further down in the log code we don't
5077 * think this inode has already been logged.
5078 */
5079 if (inode != orig_inode)
5080 BTRFS_I(inode)->logged_trans = trans->transid;
Chris Mason12fcfd22009-03-24 10:24:20 -04005081 smp_mb();
5082
Nikolay Borisovab1717b2017-01-18 00:31:27 +02005083 if (btrfs_must_commit_transaction(trans, BTRFS_I(inode))) {
Chris Mason12fcfd22009-03-24 10:24:20 -04005084 ret = 1;
5085 break;
5086 }
5087
Al Virofc640052016-04-10 01:33:30 -04005088 if (!parent || d_really_is_negative(parent) || sb != parent->d_sb)
Chris Mason12fcfd22009-03-24 10:24:20 -04005089 break;
5090
Filipe Manana44f714d2016-06-06 16:11:13 +01005091 if (IS_ROOT(parent)) {
5092 inode = d_inode(parent);
Nikolay Borisovab1717b2017-01-18 00:31:27 +02005093 if (btrfs_must_commit_transaction(trans, BTRFS_I(inode)))
Filipe Manana44f714d2016-06-06 16:11:13 +01005094 ret = 1;
Chris Mason12fcfd22009-03-24 10:24:20 -04005095 break;
Filipe Manana44f714d2016-06-06 16:11:13 +01005096 }
Chris Mason12fcfd22009-03-24 10:24:20 -04005097
Josef Bacik6a912212010-11-20 09:48:00 +00005098 parent = dget_parent(parent);
5099 dput(old_parent);
5100 old_parent = parent;
David Howells2b0143b2015-03-17 22:25:59 +00005101 inode = d_inode(parent);
Chris Mason12fcfd22009-03-24 10:24:20 -04005102
5103 }
Josef Bacik6a912212010-11-20 09:48:00 +00005104 dput(old_parent);
Chris Mason12fcfd22009-03-24 10:24:20 -04005105out:
Chris Masone02119d2008-09-05 16:13:11 -04005106 return ret;
5107}
5108
Filipe Manana2f2ff0e2015-03-20 17:19:46 +00005109struct btrfs_dir_list {
5110 u64 ino;
5111 struct list_head list;
5112};
5113
5114/*
5115 * Log the inodes of the new dentries of a directory. See log_dir_items() for
5116 * details about the why it is needed.
5117 * This is a recursive operation - if an existing dentry corresponds to a
5118 * directory, that directory's new entries are logged too (same behaviour as
5119 * ext3/4, xfs, f2fs, reiserfs, nilfs2). Note that when logging the inodes
5120 * the dentries point to we do not lock their i_mutex, otherwise lockdep
5121 * complains about the following circular lock dependency / possible deadlock:
5122 *
5123 * CPU0 CPU1
5124 * ---- ----
5125 * lock(&type->i_mutex_dir_key#3/2);
5126 * lock(sb_internal#2);
5127 * lock(&type->i_mutex_dir_key#3/2);
5128 * lock(&sb->s_type->i_mutex_key#14);
5129 *
5130 * Where sb_internal is the lock (a counter that works as a lock) acquired by
5131 * sb_start_intwrite() in btrfs_start_transaction().
5132 * Not locking i_mutex of the inodes is still safe because:
5133 *
5134 * 1) For regular files we log with a mode of LOG_INODE_EXISTS. It's possible
5135 * that while logging the inode new references (names) are added or removed
5136 * from the inode, leaving the logged inode item with a link count that does
5137 * not match the number of logged inode reference items. This is fine because
5138 * at log replay time we compute the real number of links and correct the
5139 * link count in the inode item (see replay_one_buffer() and
5140 * link_to_fixup_dir());
5141 *
5142 * 2) For directories we log with a mode of LOG_INODE_ALL. It's possible that
5143 * while logging the inode's items new items with keys BTRFS_DIR_ITEM_KEY and
5144 * BTRFS_DIR_INDEX_KEY are added to fs/subvol tree and the logged inode item
5145 * has a size that doesn't match the sum of the lengths of all the logged
5146 * names. This does not result in a problem because if a dir_item key is
5147 * logged but its matching dir_index key is not logged, at log replay time we
5148 * don't use it to replay the respective name (see replay_one_name()). On the
5149 * other hand if only the dir_index key ends up being logged, the respective
5150 * name is added to the fs/subvol tree with both the dir_item and dir_index
5151 * keys created (see replay_one_name()).
5152 * The directory's inode item with a wrong i_size is not a problem as well,
5153 * since we don't use it at log replay time to set the i_size in the inode
5154 * item of the fs/subvol tree (see overwrite_item()).
5155 */
5156static int log_new_dir_dentries(struct btrfs_trans_handle *trans,
5157 struct btrfs_root *root,
Nikolay Borisov51cc0d32017-01-18 00:31:43 +02005158 struct btrfs_inode *start_inode,
Filipe Manana2f2ff0e2015-03-20 17:19:46 +00005159 struct btrfs_log_ctx *ctx)
5160{
Jeff Mahoney0b246af2016-06-22 18:54:23 -04005161 struct btrfs_fs_info *fs_info = root->fs_info;
Filipe Manana2f2ff0e2015-03-20 17:19:46 +00005162 struct btrfs_root *log = root->log_root;
5163 struct btrfs_path *path;
5164 LIST_HEAD(dir_list);
5165 struct btrfs_dir_list *dir_elem;
5166 int ret = 0;
5167
5168 path = btrfs_alloc_path();
5169 if (!path)
5170 return -ENOMEM;
5171
5172 dir_elem = kmalloc(sizeof(*dir_elem), GFP_NOFS);
5173 if (!dir_elem) {
5174 btrfs_free_path(path);
5175 return -ENOMEM;
5176 }
Nikolay Borisov51cc0d32017-01-18 00:31:43 +02005177 dir_elem->ino = btrfs_ino(start_inode);
Filipe Manana2f2ff0e2015-03-20 17:19:46 +00005178 list_add_tail(&dir_elem->list, &dir_list);
5179
5180 while (!list_empty(&dir_list)) {
5181 struct extent_buffer *leaf;
5182 struct btrfs_key min_key;
5183 int nritems;
5184 int i;
5185
5186 dir_elem = list_first_entry(&dir_list, struct btrfs_dir_list,
5187 list);
5188 if (ret)
5189 goto next_dir_inode;
5190
5191 min_key.objectid = dir_elem->ino;
5192 min_key.type = BTRFS_DIR_ITEM_KEY;
5193 min_key.offset = 0;
5194again:
5195 btrfs_release_path(path);
5196 ret = btrfs_search_forward(log, &min_key, path, trans->transid);
5197 if (ret < 0) {
5198 goto next_dir_inode;
5199 } else if (ret > 0) {
5200 ret = 0;
5201 goto next_dir_inode;
5202 }
5203
5204process_leaf:
5205 leaf = path->nodes[0];
5206 nritems = btrfs_header_nritems(leaf);
5207 for (i = path->slots[0]; i < nritems; i++) {
5208 struct btrfs_dir_item *di;
5209 struct btrfs_key di_key;
5210 struct inode *di_inode;
5211 struct btrfs_dir_list *new_dir_elem;
5212 int log_mode = LOG_INODE_EXISTS;
5213 int type;
5214
5215 btrfs_item_key_to_cpu(leaf, &min_key, i);
5216 if (min_key.objectid != dir_elem->ino ||
5217 min_key.type != BTRFS_DIR_ITEM_KEY)
5218 goto next_dir_inode;
5219
5220 di = btrfs_item_ptr(leaf, i, struct btrfs_dir_item);
5221 type = btrfs_dir_type(leaf, di);
5222 if (btrfs_dir_transid(leaf, di) < trans->transid &&
5223 type != BTRFS_FT_DIR)
5224 continue;
5225 btrfs_dir_item_key_to_cpu(leaf, di, &di_key);
5226 if (di_key.type == BTRFS_ROOT_ITEM_KEY)
5227 continue;
5228
Robbie Koec125cf2016-10-28 10:48:26 +08005229 btrfs_release_path(path);
Jeff Mahoney0b246af2016-06-22 18:54:23 -04005230 di_inode = btrfs_iget(fs_info->sb, &di_key, root, NULL);
Filipe Manana2f2ff0e2015-03-20 17:19:46 +00005231 if (IS_ERR(di_inode)) {
5232 ret = PTR_ERR(di_inode);
5233 goto next_dir_inode;
5234 }
5235
Nikolay Borisov0f8939b2017-01-18 00:31:30 +02005236 if (btrfs_inode_in_log(BTRFS_I(di_inode), trans->transid)) {
Filipe Manana2f2ff0e2015-03-20 17:19:46 +00005237 iput(di_inode);
Robbie Koec125cf2016-10-28 10:48:26 +08005238 break;
Filipe Manana2f2ff0e2015-03-20 17:19:46 +00005239 }
5240
5241 ctx->log_new_dentries = false;
Filipe Manana3f9749f2016-04-25 04:45:02 +01005242 if (type == BTRFS_FT_DIR || type == BTRFS_FT_SYMLINK)
Filipe Manana2f2ff0e2015-03-20 17:19:46 +00005243 log_mode = LOG_INODE_ALL;
Nikolay Borisova59108a2017-01-18 00:31:48 +02005244 ret = btrfs_log_inode(trans, root, BTRFS_I(di_inode),
Filipe Manana2f2ff0e2015-03-20 17:19:46 +00005245 log_mode, 0, LLONG_MAX, ctx);
Filipe Manana2be63d52016-02-12 11:34:23 +00005246 if (!ret &&
Nikolay Borisovab1717b2017-01-18 00:31:27 +02005247 btrfs_must_commit_transaction(trans, BTRFS_I(di_inode)))
Filipe Manana2be63d52016-02-12 11:34:23 +00005248 ret = 1;
Filipe Manana2f2ff0e2015-03-20 17:19:46 +00005249 iput(di_inode);
5250 if (ret)
5251 goto next_dir_inode;
5252 if (ctx->log_new_dentries) {
5253 new_dir_elem = kmalloc(sizeof(*new_dir_elem),
5254 GFP_NOFS);
5255 if (!new_dir_elem) {
5256 ret = -ENOMEM;
5257 goto next_dir_inode;
5258 }
5259 new_dir_elem->ino = di_key.objectid;
5260 list_add_tail(&new_dir_elem->list, &dir_list);
5261 }
5262 break;
5263 }
5264 if (i == nritems) {
5265 ret = btrfs_next_leaf(log, path);
5266 if (ret < 0) {
5267 goto next_dir_inode;
5268 } else if (ret > 0) {
5269 ret = 0;
5270 goto next_dir_inode;
5271 }
5272 goto process_leaf;
5273 }
5274 if (min_key.offset < (u64)-1) {
5275 min_key.offset++;
5276 goto again;
5277 }
5278next_dir_inode:
5279 list_del(&dir_elem->list);
5280 kfree(dir_elem);
5281 }
5282
5283 btrfs_free_path(path);
5284 return ret;
5285}
5286
Filipe Manana18aa0922015-08-05 16:49:08 +01005287static int btrfs_log_all_parents(struct btrfs_trans_handle *trans,
5288 struct inode *inode,
5289 struct btrfs_log_ctx *ctx)
5290{
Jeff Mahoney0b246af2016-06-22 18:54:23 -04005291 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
Filipe Manana18aa0922015-08-05 16:49:08 +01005292 int ret;
5293 struct btrfs_path *path;
5294 struct btrfs_key key;
5295 struct btrfs_root *root = BTRFS_I(inode)->root;
Nikolay Borisov4a0cc7c2017-01-10 20:35:31 +02005296 const u64 ino = btrfs_ino(BTRFS_I(inode));
Filipe Manana18aa0922015-08-05 16:49:08 +01005297
5298 path = btrfs_alloc_path();
5299 if (!path)
5300 return -ENOMEM;
5301 path->skip_locking = 1;
5302 path->search_commit_root = 1;
5303
5304 key.objectid = ino;
5305 key.type = BTRFS_INODE_REF_KEY;
5306 key.offset = 0;
5307 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5308 if (ret < 0)
5309 goto out;
5310
5311 while (true) {
5312 struct extent_buffer *leaf = path->nodes[0];
5313 int slot = path->slots[0];
5314 u32 cur_offset = 0;
5315 u32 item_size;
5316 unsigned long ptr;
5317
5318 if (slot >= btrfs_header_nritems(leaf)) {
5319 ret = btrfs_next_leaf(root, path);
5320 if (ret < 0)
5321 goto out;
5322 else if (ret > 0)
5323 break;
5324 continue;
5325 }
5326
5327 btrfs_item_key_to_cpu(leaf, &key, slot);
5328 /* BTRFS_INODE_EXTREF_KEY is BTRFS_INODE_REF_KEY + 1 */
5329 if (key.objectid != ino || key.type > BTRFS_INODE_EXTREF_KEY)
5330 break;
5331
5332 item_size = btrfs_item_size_nr(leaf, slot);
5333 ptr = btrfs_item_ptr_offset(leaf, slot);
5334 while (cur_offset < item_size) {
5335 struct btrfs_key inode_key;
5336 struct inode *dir_inode;
5337
5338 inode_key.type = BTRFS_INODE_ITEM_KEY;
5339 inode_key.offset = 0;
5340
5341 if (key.type == BTRFS_INODE_EXTREF_KEY) {
5342 struct btrfs_inode_extref *extref;
5343
5344 extref = (struct btrfs_inode_extref *)
5345 (ptr + cur_offset);
5346 inode_key.objectid = btrfs_inode_extref_parent(
5347 leaf, extref);
5348 cur_offset += sizeof(*extref);
5349 cur_offset += btrfs_inode_extref_name_len(leaf,
5350 extref);
5351 } else {
5352 inode_key.objectid = key.offset;
5353 cur_offset = item_size;
5354 }
5355
Jeff Mahoney0b246af2016-06-22 18:54:23 -04005356 dir_inode = btrfs_iget(fs_info->sb, &inode_key,
Filipe Manana18aa0922015-08-05 16:49:08 +01005357 root, NULL);
5358 /* If parent inode was deleted, skip it. */
5359 if (IS_ERR(dir_inode))
5360 continue;
5361
Filipe Manana657ed1a2016-04-06 17:11:56 +01005362 if (ctx)
5363 ctx->log_new_dentries = false;
Nikolay Borisova59108a2017-01-18 00:31:48 +02005364 ret = btrfs_log_inode(trans, root, BTRFS_I(dir_inode),
Filipe Manana18aa0922015-08-05 16:49:08 +01005365 LOG_INODE_ALL, 0, LLONG_MAX, ctx);
Filipe Manana2be63d52016-02-12 11:34:23 +00005366 if (!ret &&
Nikolay Borisovab1717b2017-01-18 00:31:27 +02005367 btrfs_must_commit_transaction(trans, BTRFS_I(dir_inode)))
Filipe Manana2be63d52016-02-12 11:34:23 +00005368 ret = 1;
Filipe Manana657ed1a2016-04-06 17:11:56 +01005369 if (!ret && ctx && ctx->log_new_dentries)
5370 ret = log_new_dir_dentries(trans, root,
Nikolay Borisov51cc0d32017-01-18 00:31:43 +02005371 BTRFS_I(dir_inode), ctx);
Filipe Manana18aa0922015-08-05 16:49:08 +01005372 iput(dir_inode);
5373 if (ret)
5374 goto out;
5375 }
5376 path->slots[0]++;
5377 }
5378 ret = 0;
5379out:
5380 btrfs_free_path(path);
5381 return ret;
5382}
5383
Chris Masone02119d2008-09-05 16:13:11 -04005384/*
5385 * helper function around btrfs_log_inode to make sure newly created
5386 * parent directories also end up in the log. A minimal inode and backref
5387 * only logging is done of any parent directories that are older than
5388 * the last committed transaction
5389 */
Eric Sandeen48a3b632013-04-25 20:41:01 +00005390static int btrfs_log_inode_parent(struct btrfs_trans_handle *trans,
5391 struct btrfs_root *root, struct inode *inode,
Filipe Manana49dae1b2014-09-06 22:34:39 +01005392 struct dentry *parent,
5393 const loff_t start,
5394 const loff_t end,
5395 int exists_only,
Miao Xie8b050d32014-02-20 18:08:58 +08005396 struct btrfs_log_ctx *ctx)
Chris Masone02119d2008-09-05 16:13:11 -04005397{
Jeff Mahoney0b246af2016-06-22 18:54:23 -04005398 struct btrfs_fs_info *fs_info = root->fs_info;
Chris Mason12fcfd22009-03-24 10:24:20 -04005399 int inode_only = exists_only ? LOG_INODE_EXISTS : LOG_INODE_ALL;
Chris Masone02119d2008-09-05 16:13:11 -04005400 struct super_block *sb;
Josef Bacik6a912212010-11-20 09:48:00 +00005401 struct dentry *old_parent = NULL;
Chris Mason12fcfd22009-03-24 10:24:20 -04005402 int ret = 0;
Jeff Mahoney0b246af2016-06-22 18:54:23 -04005403 u64 last_committed = fs_info->last_trans_committed;
Filipe Manana2f2ff0e2015-03-20 17:19:46 +00005404 bool log_dentries = false;
5405 struct inode *orig_inode = inode;
Chris Mason12fcfd22009-03-24 10:24:20 -04005406
5407 sb = inode->i_sb;
5408
Jeff Mahoney0b246af2016-06-22 18:54:23 -04005409 if (btrfs_test_opt(fs_info, NOTREELOG)) {
Sage Weil3a5e1402009-04-02 16:49:40 -04005410 ret = 1;
5411 goto end_no_trans;
5412 }
5413
Miao Xie995946d2014-04-02 19:51:06 +08005414 /*
5415 * The prev transaction commit doesn't complete, we need do
5416 * full commit by ourselves.
5417 */
Jeff Mahoney0b246af2016-06-22 18:54:23 -04005418 if (fs_info->last_trans_log_full_commit >
5419 fs_info->last_trans_committed) {
Chris Mason12fcfd22009-03-24 10:24:20 -04005420 ret = 1;
5421 goto end_no_trans;
5422 }
5423
Yan, Zheng76dda932009-09-21 16:00:26 -04005424 if (root != BTRFS_I(inode)->root ||
5425 btrfs_root_refs(&root->root_item) == 0) {
5426 ret = 1;
5427 goto end_no_trans;
5428 }
5429
Chris Mason12fcfd22009-03-24 10:24:20 -04005430 ret = check_parent_dirs_for_sync(trans, inode, parent,
5431 sb, last_committed);
5432 if (ret)
5433 goto end_no_trans;
Chris Masone02119d2008-09-05 16:13:11 -04005434
Nikolay Borisov0f8939b2017-01-18 00:31:30 +02005435 if (btrfs_inode_in_log(BTRFS_I(inode), trans->transid)) {
Chris Mason257c62e2009-10-13 13:21:08 -04005436 ret = BTRFS_NO_LOG_SYNC;
5437 goto end_no_trans;
5438 }
5439
Miao Xie8b050d32014-02-20 18:08:58 +08005440 ret = start_log_trans(trans, root, ctx);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04005441 if (ret)
Miao Xiee87ac132014-02-20 18:08:53 +08005442 goto end_no_trans;
Chris Mason12fcfd22009-03-24 10:24:20 -04005443
Nikolay Borisova59108a2017-01-18 00:31:48 +02005444 ret = btrfs_log_inode(trans, root, BTRFS_I(inode), inode_only,
5445 start, end, ctx);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04005446 if (ret)
5447 goto end_trans;
Chris Mason12fcfd22009-03-24 10:24:20 -04005448
Chris Masonaf4176b2009-03-24 10:24:31 -04005449 /*
5450 * for regular files, if its inode is already on disk, we don't
5451 * have to worry about the parents at all. This is because
5452 * we can use the last_unlink_trans field to record renames
5453 * and other fun in this file.
5454 */
5455 if (S_ISREG(inode->i_mode) &&
5456 BTRFS_I(inode)->generation <= last_committed &&
Yan, Zheng4a500fd2010-05-16 10:49:59 -04005457 BTRFS_I(inode)->last_unlink_trans <= last_committed) {
5458 ret = 0;
5459 goto end_trans;
5460 }
Chris Masonaf4176b2009-03-24 10:24:31 -04005461
Filipe Manana2f2ff0e2015-03-20 17:19:46 +00005462 if (S_ISDIR(inode->i_mode) && ctx && ctx->log_new_dentries)
5463 log_dentries = true;
5464
Filipe Manana18aa0922015-08-05 16:49:08 +01005465 /*
Nicholas D Steeves01327612016-05-19 21:18:45 -04005466 * On unlink we must make sure all our current and old parent directory
Filipe Manana18aa0922015-08-05 16:49:08 +01005467 * inodes are fully logged. This is to prevent leaving dangling
5468 * directory index entries in directories that were our parents but are
5469 * not anymore. Not doing this results in old parent directory being
5470 * impossible to delete after log replay (rmdir will always fail with
5471 * error -ENOTEMPTY).
5472 *
5473 * Example 1:
5474 *
5475 * mkdir testdir
5476 * touch testdir/foo
5477 * ln testdir/foo testdir/bar
5478 * sync
5479 * unlink testdir/bar
5480 * xfs_io -c fsync testdir/foo
5481 * <power failure>
5482 * mount fs, triggers log replay
5483 *
5484 * If we don't log the parent directory (testdir), after log replay the
5485 * directory still has an entry pointing to the file inode using the bar
5486 * name, but a matching BTRFS_INODE_[REF|EXTREF]_KEY does not exist and
5487 * the file inode has a link count of 1.
5488 *
5489 * Example 2:
5490 *
5491 * mkdir testdir
5492 * touch foo
5493 * ln foo testdir/foo2
5494 * ln foo testdir/foo3
5495 * sync
5496 * unlink testdir/foo3
5497 * xfs_io -c fsync foo
5498 * <power failure>
5499 * mount fs, triggers log replay
5500 *
5501 * Similar as the first example, after log replay the parent directory
5502 * testdir still has an entry pointing to the inode file with name foo3
5503 * but the file inode does not have a matching BTRFS_INODE_REF_KEY item
5504 * and has a link count of 2.
5505 */
5506 if (BTRFS_I(inode)->last_unlink_trans > last_committed) {
5507 ret = btrfs_log_all_parents(trans, orig_inode, ctx);
5508 if (ret)
5509 goto end_trans;
5510 }
5511
Chris Masond3977122009-01-05 21:25:51 -05005512 while (1) {
Al Virofc640052016-04-10 01:33:30 -04005513 if (!parent || d_really_is_negative(parent) || sb != parent->d_sb)
Chris Masone02119d2008-09-05 16:13:11 -04005514 break;
5515
David Howells2b0143b2015-03-17 22:25:59 +00005516 inode = d_inode(parent);
Yan, Zheng76dda932009-09-21 16:00:26 -04005517 if (root != BTRFS_I(inode)->root)
5518 break;
5519
Filipe Manana18aa0922015-08-05 16:49:08 +01005520 if (BTRFS_I(inode)->generation > last_committed) {
Nikolay Borisova59108a2017-01-18 00:31:48 +02005521 ret = btrfs_log_inode(trans, root, BTRFS_I(inode),
Filipe Manana18aa0922015-08-05 16:49:08 +01005522 LOG_INODE_EXISTS,
Filipe Manana8407f552014-09-05 15:14:39 +01005523 0, LLONG_MAX, ctx);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04005524 if (ret)
5525 goto end_trans;
Chris Mason12fcfd22009-03-24 10:24:20 -04005526 }
Yan, Zheng76dda932009-09-21 16:00:26 -04005527 if (IS_ROOT(parent))
Chris Masone02119d2008-09-05 16:13:11 -04005528 break;
Chris Mason12fcfd22009-03-24 10:24:20 -04005529
Josef Bacik6a912212010-11-20 09:48:00 +00005530 parent = dget_parent(parent);
5531 dput(old_parent);
5532 old_parent = parent;
Chris Masone02119d2008-09-05 16:13:11 -04005533 }
Filipe Manana2f2ff0e2015-03-20 17:19:46 +00005534 if (log_dentries)
Nikolay Borisov51cc0d32017-01-18 00:31:43 +02005535 ret = log_new_dir_dentries(trans, root, BTRFS_I(orig_inode), ctx);
Filipe Manana2f2ff0e2015-03-20 17:19:46 +00005536 else
5537 ret = 0;
Yan, Zheng4a500fd2010-05-16 10:49:59 -04005538end_trans:
Josef Bacik6a912212010-11-20 09:48:00 +00005539 dput(old_parent);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04005540 if (ret < 0) {
Jeff Mahoney0b246af2016-06-22 18:54:23 -04005541 btrfs_set_log_full_commit(fs_info, trans);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04005542 ret = 1;
5543 }
Miao Xie8b050d32014-02-20 18:08:58 +08005544
5545 if (ret)
5546 btrfs_remove_log_ctx(root, ctx);
Chris Mason12fcfd22009-03-24 10:24:20 -04005547 btrfs_end_log_trans(root);
5548end_no_trans:
5549 return ret;
Chris Masone02119d2008-09-05 16:13:11 -04005550}
5551
5552/*
5553 * it is not safe to log dentry if the chunk root has added new
5554 * chunks. This returns 0 if the dentry was logged, and 1 otherwise.
5555 * If this returns 1, you must commit the transaction to safely get your
5556 * data on disk.
5557 */
5558int btrfs_log_dentry_safe(struct btrfs_trans_handle *trans,
Miao Xie8b050d32014-02-20 18:08:58 +08005559 struct btrfs_root *root, struct dentry *dentry,
Filipe Manana49dae1b2014-09-06 22:34:39 +01005560 const loff_t start,
5561 const loff_t end,
Miao Xie8b050d32014-02-20 18:08:58 +08005562 struct btrfs_log_ctx *ctx)
Chris Masone02119d2008-09-05 16:13:11 -04005563{
Josef Bacik6a912212010-11-20 09:48:00 +00005564 struct dentry *parent = dget_parent(dentry);
5565 int ret;
5566
David Howells2b0143b2015-03-17 22:25:59 +00005567 ret = btrfs_log_inode_parent(trans, root, d_inode(dentry), parent,
Filipe Manana49dae1b2014-09-06 22:34:39 +01005568 start, end, 0, ctx);
Josef Bacik6a912212010-11-20 09:48:00 +00005569 dput(parent);
5570
5571 return ret;
Chris Masone02119d2008-09-05 16:13:11 -04005572}
5573
5574/*
5575 * should be called during mount to recover any replay any log trees
5576 * from the FS
5577 */
5578int btrfs_recover_log_trees(struct btrfs_root *log_root_tree)
5579{
5580 int ret;
5581 struct btrfs_path *path;
5582 struct btrfs_trans_handle *trans;
5583 struct btrfs_key key;
5584 struct btrfs_key found_key;
5585 struct btrfs_key tmp_key;
5586 struct btrfs_root *log;
5587 struct btrfs_fs_info *fs_info = log_root_tree->fs_info;
5588 struct walk_control wc = {
5589 .process_func = process_one_buffer,
5590 .stage = 0,
5591 };
5592
Chris Masone02119d2008-09-05 16:13:11 -04005593 path = btrfs_alloc_path();
Tsutomu Itohdb5b4932011-03-23 08:14:16 +00005594 if (!path)
5595 return -ENOMEM;
5596
Josef Bacikafcdd122016-09-02 15:40:02 -04005597 set_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags);
Chris Masone02119d2008-09-05 16:13:11 -04005598
Yan, Zheng4a500fd2010-05-16 10:49:59 -04005599 trans = btrfs_start_transaction(fs_info->tree_root, 0);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01005600 if (IS_ERR(trans)) {
5601 ret = PTR_ERR(trans);
5602 goto error;
5603 }
Chris Masone02119d2008-09-05 16:13:11 -04005604
5605 wc.trans = trans;
5606 wc.pin = 1;
5607
Tsutomu Itohdb5b4932011-03-23 08:14:16 +00005608 ret = walk_log_tree(trans, log_root_tree, &wc);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01005609 if (ret) {
Jeff Mahoney5d163e02016-09-20 10:05:00 -04005610 btrfs_handle_fs_error(fs_info, ret,
5611 "Failed to pin buffers while recovering log root tree.");
Jeff Mahoney79787ea2012-03-12 16:03:00 +01005612 goto error;
5613 }
Chris Masone02119d2008-09-05 16:13:11 -04005614
5615again:
5616 key.objectid = BTRFS_TREE_LOG_OBJECTID;
5617 key.offset = (u64)-1;
David Sterba962a2982014-06-04 18:41:45 +02005618 key.type = BTRFS_ROOT_ITEM_KEY;
Chris Masone02119d2008-09-05 16:13:11 -04005619
Chris Masond3977122009-01-05 21:25:51 -05005620 while (1) {
Chris Masone02119d2008-09-05 16:13:11 -04005621 ret = btrfs_search_slot(NULL, log_root_tree, &key, path, 0, 0);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01005622
5623 if (ret < 0) {
Anand Jain34d97002016-03-16 16:43:06 +08005624 btrfs_handle_fs_error(fs_info, ret,
Jeff Mahoney79787ea2012-03-12 16:03:00 +01005625 "Couldn't find tree log root.");
5626 goto error;
5627 }
Chris Masone02119d2008-09-05 16:13:11 -04005628 if (ret > 0) {
5629 if (path->slots[0] == 0)
5630 break;
5631 path->slots[0]--;
5632 }
5633 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
5634 path->slots[0]);
David Sterbab3b4aa72011-04-21 01:20:15 +02005635 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04005636 if (found_key.objectid != BTRFS_TREE_LOG_OBJECTID)
5637 break;
5638
Miao Xiecb517ea2013-05-15 07:48:19 +00005639 log = btrfs_read_fs_root(log_root_tree, &found_key);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01005640 if (IS_ERR(log)) {
5641 ret = PTR_ERR(log);
Anand Jain34d97002016-03-16 16:43:06 +08005642 btrfs_handle_fs_error(fs_info, ret,
Jeff Mahoney79787ea2012-03-12 16:03:00 +01005643 "Couldn't read tree log root.");
5644 goto error;
5645 }
Chris Masone02119d2008-09-05 16:13:11 -04005646
5647 tmp_key.objectid = found_key.offset;
5648 tmp_key.type = BTRFS_ROOT_ITEM_KEY;
5649 tmp_key.offset = (u64)-1;
5650
5651 wc.replay_dest = btrfs_read_fs_root_no_name(fs_info, &tmp_key);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01005652 if (IS_ERR(wc.replay_dest)) {
5653 ret = PTR_ERR(wc.replay_dest);
Josef Bacikb50c6e22013-04-25 15:55:30 -04005654 free_extent_buffer(log->node);
5655 free_extent_buffer(log->commit_root);
5656 kfree(log);
Jeff Mahoney5d163e02016-09-20 10:05:00 -04005657 btrfs_handle_fs_error(fs_info, ret,
5658 "Couldn't read target root for tree log recovery.");
Jeff Mahoney79787ea2012-03-12 16:03:00 +01005659 goto error;
5660 }
Chris Masone02119d2008-09-05 16:13:11 -04005661
Yan Zheng07d400a2009-01-06 11:42:00 -05005662 wc.replay_dest->log_root = log;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005663 btrfs_record_root_in_trans(trans, wc.replay_dest);
Chris Masone02119d2008-09-05 16:13:11 -04005664 ret = walk_log_tree(trans, log, &wc);
Chris Masone02119d2008-09-05 16:13:11 -04005665
Josef Bacikb50c6e22013-04-25 15:55:30 -04005666 if (!ret && wc.stage == LOG_WALK_REPLAY_ALL) {
Chris Masone02119d2008-09-05 16:13:11 -04005667 ret = fixup_inode_link_counts(trans, wc.replay_dest,
5668 path);
Chris Masone02119d2008-09-05 16:13:11 -04005669 }
Chris Masone02119d2008-09-05 16:13:11 -04005670
5671 key.offset = found_key.offset - 1;
Yan Zheng07d400a2009-01-06 11:42:00 -05005672 wc.replay_dest->log_root = NULL;
Chris Masone02119d2008-09-05 16:13:11 -04005673 free_extent_buffer(log->node);
Chris Masonb263c2c2009-06-11 11:24:47 -04005674 free_extent_buffer(log->commit_root);
Chris Masone02119d2008-09-05 16:13:11 -04005675 kfree(log);
5676
Josef Bacikb50c6e22013-04-25 15:55:30 -04005677 if (ret)
5678 goto error;
5679
Chris Masone02119d2008-09-05 16:13:11 -04005680 if (found_key.offset == 0)
5681 break;
5682 }
David Sterbab3b4aa72011-04-21 01:20:15 +02005683 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04005684
5685 /* step one is to pin it all, step two is to replay just inodes */
5686 if (wc.pin) {
5687 wc.pin = 0;
5688 wc.process_func = replay_one_buffer;
5689 wc.stage = LOG_WALK_REPLAY_INODES;
5690 goto again;
5691 }
5692 /* step three is to replay everything */
5693 if (wc.stage < LOG_WALK_REPLAY_ALL) {
5694 wc.stage++;
5695 goto again;
5696 }
5697
5698 btrfs_free_path(path);
5699
Josef Bacikabefa552013-04-24 16:40:05 -04005700 /* step 4: commit the transaction, which also unpins the blocks */
Jeff Mahoney3a45bb22016-09-09 21:39:03 -04005701 ret = btrfs_commit_transaction(trans);
Josef Bacikabefa552013-04-24 16:40:05 -04005702 if (ret)
5703 return ret;
5704
Chris Masone02119d2008-09-05 16:13:11 -04005705 free_extent_buffer(log_root_tree->node);
5706 log_root_tree->log_root = NULL;
Josef Bacikafcdd122016-09-02 15:40:02 -04005707 clear_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags);
Chris Masone02119d2008-09-05 16:13:11 -04005708 kfree(log_root_tree);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01005709
Josef Bacikabefa552013-04-24 16:40:05 -04005710 return 0;
Jeff Mahoney79787ea2012-03-12 16:03:00 +01005711error:
Josef Bacikb50c6e22013-04-25 15:55:30 -04005712 if (wc.trans)
Jeff Mahoney3a45bb22016-09-09 21:39:03 -04005713 btrfs_end_transaction(wc.trans);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01005714 btrfs_free_path(path);
5715 return ret;
Chris Masone02119d2008-09-05 16:13:11 -04005716}
Chris Mason12fcfd22009-03-24 10:24:20 -04005717
5718/*
5719 * there are some corner cases where we want to force a full
5720 * commit instead of allowing a directory to be logged.
5721 *
5722 * They revolve around files there were unlinked from the directory, and
5723 * this function updates the parent directory so that a full commit is
5724 * properly done if it is fsync'd later after the unlinks are done.
Filipe Manana2be63d52016-02-12 11:34:23 +00005725 *
5726 * Must be called before the unlink operations (updates to the subvolume tree,
5727 * inodes, etc) are done.
Chris Mason12fcfd22009-03-24 10:24:20 -04005728 */
5729void btrfs_record_unlink_dir(struct btrfs_trans_handle *trans,
Nikolay Borisov4176bdb2017-01-18 00:31:28 +02005730 struct btrfs_inode *dir, struct btrfs_inode *inode,
Chris Mason12fcfd22009-03-24 10:24:20 -04005731 int for_rename)
5732{
5733 /*
Chris Masonaf4176b2009-03-24 10:24:31 -04005734 * when we're logging a file, if it hasn't been renamed
5735 * or unlinked, and its inode is fully committed on disk,
5736 * we don't have to worry about walking up the directory chain
5737 * to log its parents.
5738 *
5739 * So, we use the last_unlink_trans field to put this transid
5740 * into the file. When the file is logged we check it and
5741 * don't log the parents if the file is fully on disk.
5742 */
Nikolay Borisov4176bdb2017-01-18 00:31:28 +02005743 mutex_lock(&inode->log_mutex);
5744 inode->last_unlink_trans = trans->transid;
5745 mutex_unlock(&inode->log_mutex);
Chris Masonaf4176b2009-03-24 10:24:31 -04005746
5747 /*
Chris Mason12fcfd22009-03-24 10:24:20 -04005748 * if this directory was already logged any new
5749 * names for this file/dir will get recorded
5750 */
5751 smp_mb();
Nikolay Borisov4176bdb2017-01-18 00:31:28 +02005752 if (dir->logged_trans == trans->transid)
Chris Mason12fcfd22009-03-24 10:24:20 -04005753 return;
5754
5755 /*
5756 * if the inode we're about to unlink was logged,
5757 * the log will be properly updated for any new names
5758 */
Nikolay Borisov4176bdb2017-01-18 00:31:28 +02005759 if (inode->logged_trans == trans->transid)
Chris Mason12fcfd22009-03-24 10:24:20 -04005760 return;
5761
5762 /*
5763 * when renaming files across directories, if the directory
5764 * there we're unlinking from gets fsync'd later on, there's
5765 * no way to find the destination directory later and fsync it
5766 * properly. So, we have to be conservative and force commits
5767 * so the new name gets discovered.
5768 */
5769 if (for_rename)
5770 goto record;
5771
5772 /* we can safely do the unlink without any special recording */
5773 return;
5774
5775record:
Nikolay Borisov4176bdb2017-01-18 00:31:28 +02005776 mutex_lock(&dir->log_mutex);
5777 dir->last_unlink_trans = trans->transid;
5778 mutex_unlock(&dir->log_mutex);
Chris Mason12fcfd22009-03-24 10:24:20 -04005779}
5780
5781/*
Filipe Manana1ec9a1a2016-02-10 10:42:25 +00005782 * Make sure that if someone attempts to fsync the parent directory of a deleted
5783 * snapshot, it ends up triggering a transaction commit. This is to guarantee
5784 * that after replaying the log tree of the parent directory's root we will not
5785 * see the snapshot anymore and at log replay time we will not see any log tree
5786 * corresponding to the deleted snapshot's root, which could lead to replaying
5787 * it after replaying the log tree of the parent directory (which would replay
5788 * the snapshot delete operation).
Filipe Manana2be63d52016-02-12 11:34:23 +00005789 *
5790 * Must be called before the actual snapshot destroy operation (updates to the
5791 * parent root and tree of tree roots trees, etc) are done.
Filipe Manana1ec9a1a2016-02-10 10:42:25 +00005792 */
5793void btrfs_record_snapshot_destroy(struct btrfs_trans_handle *trans,
Nikolay Borisov43663552017-01-18 00:31:29 +02005794 struct btrfs_inode *dir)
Filipe Manana1ec9a1a2016-02-10 10:42:25 +00005795{
Nikolay Borisov43663552017-01-18 00:31:29 +02005796 mutex_lock(&dir->log_mutex);
5797 dir->last_unlink_trans = trans->transid;
5798 mutex_unlock(&dir->log_mutex);
Filipe Manana1ec9a1a2016-02-10 10:42:25 +00005799}
5800
5801/*
Chris Mason12fcfd22009-03-24 10:24:20 -04005802 * Call this after adding a new name for a file and it will properly
5803 * update the log to reflect the new name.
5804 *
5805 * It will return zero if all goes well, and it will return 1 if a
5806 * full transaction commit is required.
5807 */
5808int btrfs_log_new_name(struct btrfs_trans_handle *trans,
Nikolay Borisov9ca5fbfb2017-01-18 00:31:31 +02005809 struct btrfs_inode *inode, struct btrfs_inode *old_dir,
Chris Mason12fcfd22009-03-24 10:24:20 -04005810 struct dentry *parent)
5811{
Nikolay Borisov9ca5fbfb2017-01-18 00:31:31 +02005812 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
5813 struct btrfs_root * root = inode->root;
Chris Mason12fcfd22009-03-24 10:24:20 -04005814
5815 /*
Chris Masonaf4176b2009-03-24 10:24:31 -04005816 * this will force the logging code to walk the dentry chain
5817 * up for the file
5818 */
Nikolay Borisov9ca5fbfb2017-01-18 00:31:31 +02005819 if (S_ISREG(inode->vfs_inode.i_mode))
5820 inode->last_unlink_trans = trans->transid;
Chris Masonaf4176b2009-03-24 10:24:31 -04005821
5822 /*
Chris Mason12fcfd22009-03-24 10:24:20 -04005823 * if this inode hasn't been logged and directory we're renaming it
5824 * from hasn't been logged, we don't need to log it
5825 */
Nikolay Borisov9ca5fbfb2017-01-18 00:31:31 +02005826 if (inode->logged_trans <= fs_info->last_trans_committed &&
5827 (!old_dir || old_dir->logged_trans <= fs_info->last_trans_committed))
Chris Mason12fcfd22009-03-24 10:24:20 -04005828 return 0;
5829
Nikolay Borisov9ca5fbfb2017-01-18 00:31:31 +02005830 return btrfs_log_inode_parent(trans, root, &inode->vfs_inode, parent, 0,
Filipe Manana49dae1b2014-09-06 22:34:39 +01005831 LLONG_MAX, 1, NULL);
Chris Mason12fcfd22009-03-24 10:24:20 -04005832}
5833