Linux kernel mirror (for testing)
git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
tjh.dev
kernel
os
linux
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * linux/fs/ext4/fsync.c
4 *
5 * Copyright (C) 1993 Stephen Tweedie (sct@redhat.com)
6 * from
7 * Copyright (C) 1992 Remy Card (card@masi.ibp.fr)
8 * Laboratoire MASI - Institut Blaise Pascal
9 * Universite Pierre et Marie Curie (Paris VI)
10 * from
11 * linux/fs/minix/truncate.c Copyright (C) 1991, 1992 Linus Torvalds
12 *
13 * ext4fs fsync primitive
14 *
15 * Big-endian to little-endian byte-swapping/bitmaps by
16 * David S. Miller (davem@caip.rutgers.edu), 1995
17 *
18 * Removed unnecessary code duplication for little endian machines
19 * and excessive __inline__s.
20 * Andi Kleen, 1997
21 *
22 * Major simplications and cleanup - we only need to do the metadata, because
23 * we can depend on generic_block_fdatasync() to sync the data blocks.
24 */
25
26#include <linux/time.h>
27#include <linux/fs.h>
28#include <linux/sched.h>
29#include <linux/writeback.h>
30#include <linux/blkdev.h>
31
32#include "ext4.h"
33#include "ext4_jbd2.h"
34
35#include <trace/events/ext4.h>
36
37/*
38 * If we're not journaling and this is a just-created file, we have to
39 * sync our parent directory (if it was freshly created) since
40 * otherwise it will only be written by writeback, leaving a huge
41 * window during which a crash may lose the file. This may apply for
42 * the parent directory's parent as well, and so on recursively, if
43 * they are also freshly created.
44 */
45static int ext4_sync_parent(struct inode *inode)
46{
47 struct dentry *dentry = NULL;
48 struct inode *next;
49 int ret = 0;
50
51 if (!ext4_test_inode_state(inode, EXT4_STATE_NEWENTRY))
52 return 0;
53 inode = igrab(inode);
54 while (ext4_test_inode_state(inode, EXT4_STATE_NEWENTRY)) {
55 ext4_clear_inode_state(inode, EXT4_STATE_NEWENTRY);
56 dentry = d_find_any_alias(inode);
57 if (!dentry)
58 break;
59 next = igrab(d_inode(dentry->d_parent));
60 dput(dentry);
61 if (!next)
62 break;
63 iput(inode);
64 inode = next;
65 /*
66 * The directory inode may have gone through rmdir by now. But
67 * the inode itself and its blocks are still allocated (we hold
68 * a reference to the inode so it didn't go through
69 * ext4_evict_inode()) and so we are safe to flush metadata
70 * blocks and the inode.
71 */
72 ret = sync_mapping_buffers(inode->i_mapping);
73 if (ret)
74 break;
75 ret = sync_inode_metadata(inode, 1);
76 if (ret)
77 break;
78 }
79 iput(inode);
80 return ret;
81}
82
83static int ext4_fsync_nojournal(struct inode *inode, bool datasync,
84 bool *needs_barrier)
85{
86 int ret, err;
87
88 ret = sync_mapping_buffers(inode->i_mapping);
89 if (!(inode->i_state & I_DIRTY_ALL))
90 return ret;
91 if (datasync && !(inode->i_state & I_DIRTY_DATASYNC))
92 return ret;
93
94 err = sync_inode_metadata(inode, 1);
95 if (!ret)
96 ret = err;
97
98 if (!ret)
99 ret = ext4_sync_parent(inode);
100 if (test_opt(inode->i_sb, BARRIER))
101 *needs_barrier = true;
102
103 return ret;
104}
105
106static int ext4_fsync_journal(struct inode *inode, bool datasync,
107 bool *needs_barrier)
108{
109 struct ext4_inode_info *ei = EXT4_I(inode);
110 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
111 tid_t commit_tid = datasync ? ei->i_datasync_tid : ei->i_sync_tid;
112
113 if (journal->j_flags & JBD2_BARRIER &&
114 !jbd2_trans_will_send_data_barrier(journal, commit_tid))
115 *needs_barrier = true;
116
117 return jbd2_complete_transaction(journal, commit_tid);
118}
119
120/*
121 * akpm: A new design for ext4_sync_file().
122 *
123 * This is only called from sys_fsync(), sys_fdatasync() and sys_msync().
124 * There cannot be a transaction open by this task.
125 * Another task could have dirtied this inode. Its data can be in any
126 * state in the journalling system.
127 *
128 * What we do is just kick off a commit and wait on it. This will snapshot the
129 * inode to disk.
130 */
131int ext4_sync_file(struct file *file, loff_t start, loff_t end, int datasync)
132{
133 int ret = 0, err;
134 bool needs_barrier = false;
135 struct inode *inode = file->f_mapping->host;
136 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
137
138 if (unlikely(ext4_forced_shutdown(sbi)))
139 return -EIO;
140
141 J_ASSERT(ext4_journal_current_handle() == NULL);
142
143 trace_ext4_sync_file_enter(file, datasync);
144
145 if (sb_rdonly(inode->i_sb)) {
146 /* Make sure that we read updated s_mount_flags value */
147 smp_rmb();
148 if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED)
149 ret = -EROFS;
150 goto out;
151 }
152
153 ret = file_write_and_wait_range(file, start, end);
154 if (ret)
155 return ret;
156
157 /*
158 * data=writeback,ordered:
159 * The caller's filemap_fdatawrite()/wait will sync the data.
160 * Metadata is in the journal, we wait for proper transaction to
161 * commit here.
162 *
163 * data=journal:
164 * filemap_fdatawrite won't do anything (the buffers are clean).
165 * ext4_force_commit will write the file data into the journal and
166 * will wait on that.
167 * filemap_fdatawait() will encounter a ton of newly-dirtied pages
168 * (they were dirtied by commit). But that's OK - the blocks are
169 * safe in-journal, which is all fsync() needs to ensure.
170 */
171 if (!sbi->s_journal)
172 ret = ext4_fsync_nojournal(inode, datasync, &needs_barrier);
173 else if (ext4_should_journal_data(inode))
174 ret = ext4_force_commit(inode->i_sb);
175 else
176 ret = ext4_fsync_journal(inode, datasync, &needs_barrier);
177
178 if (needs_barrier) {
179 err = blkdev_issue_flush(inode->i_sb->s_bdev, GFP_KERNEL, NULL);
180 if (!ret)
181 ret = err;
182 }
183out:
184 err = file_check_and_advance_wb_err(file);
185 if (ret == 0)
186 ret = err;
187 trace_ext4_sync_file_exit(inode, ret);
188 return ret;
189}