Loading...
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, *next;
48 int ret = 0;
49
50 if (!ext4_test_inode_state(inode, EXT4_STATE_NEWENTRY))
51 return 0;
52 dentry = d_find_any_alias(inode);
53 if (!dentry)
54 return 0;
55 while (ext4_test_inode_state(inode, EXT4_STATE_NEWENTRY)) {
56 ext4_clear_inode_state(inode, EXT4_STATE_NEWENTRY);
57
58 next = dget_parent(dentry);
59 dput(dentry);
60 dentry = next;
61 inode = dentry->d_inode;
62
63 /*
64 * The directory inode may have gone through rmdir by now. But
65 * the inode itself and its blocks are still allocated (we hold
66 * a reference to the inode via its dentry), so it didn't go
67 * through ext4_evict_inode()) and so we are safe to flush
68 * metadata blocks and the inode.
69 */
70 ret = sync_mapping_buffers(inode->i_mapping);
71 if (ret)
72 break;
73 ret = sync_inode_metadata(inode, 1);
74 if (ret)
75 break;
76 }
77 dput(dentry);
78 return ret;
79}
80
81static int ext4_fsync_nojournal(struct inode *inode, bool datasync,
82 bool *needs_barrier)
83{
84 int ret, err;
85
86 ret = sync_mapping_buffers(inode->i_mapping);
87 if (!(inode->i_state & I_DIRTY_ALL))
88 return ret;
89 if (datasync && !(inode->i_state & I_DIRTY_DATASYNC))
90 return ret;
91
92 err = sync_inode_metadata(inode, 1);
93 if (!ret)
94 ret = err;
95
96 if (!ret)
97 ret = ext4_sync_parent(inode);
98 if (test_opt(inode->i_sb, BARRIER))
99 *needs_barrier = true;
100
101 return ret;
102}
103
104static int ext4_fsync_journal(struct inode *inode, bool datasync,
105 bool *needs_barrier)
106{
107 struct ext4_inode_info *ei = EXT4_I(inode);
108 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
109 tid_t commit_tid = datasync ? ei->i_datasync_tid : ei->i_sync_tid;
110
111 if (journal->j_flags & JBD2_BARRIER &&
112 !jbd2_trans_will_send_data_barrier(journal, commit_tid))
113 *needs_barrier = true;
114
115 return ext4_fc_commit(journal, commit_tid);
116}
117
118/*
119 * akpm: A new design for ext4_sync_file().
120 *
121 * This is only called from sys_fsync(), sys_fdatasync() and sys_msync().
122 * There cannot be a transaction open by this task.
123 * Another task could have dirtied this inode. Its data can be in any
124 * state in the journalling system.
125 *
126 * What we do is just kick off a commit and wait on it. This will snapshot the
127 * inode to disk.
128 */
129int ext4_sync_file(struct file *file, loff_t start, loff_t end, int datasync)
130{
131 int ret = 0, err;
132 bool needs_barrier = false;
133 struct inode *inode = file->f_mapping->host;
134 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
135
136 if (unlikely(ext4_forced_shutdown(sbi)))
137 return -EIO;
138
139 ASSERT(ext4_journal_current_handle() == NULL);
140
141 trace_ext4_sync_file_enter(file, datasync);
142
143 if (sb_rdonly(inode->i_sb)) {
144 /* Make sure that we read updated s_mount_flags value */
145 smp_rmb();
146 if (ext4_test_mount_flag(inode->i_sb, EXT4_MF_FS_ABORTED))
147 ret = -EROFS;
148 goto out;
149 }
150
151 ret = file_write_and_wait_range(file, start, end);
152 if (ret)
153 goto out;
154
155 /*
156 * data=writeback,ordered:
157 * The caller's filemap_fdatawrite()/wait will sync the data.
158 * Metadata is in the journal, we wait for proper transaction to
159 * commit here.
160 *
161 * data=journal:
162 * filemap_fdatawrite won't do anything (the buffers are clean).
163 * ext4_force_commit will write the file data into the journal and
164 * will wait on that.
165 * filemap_fdatawait() will encounter a ton of newly-dirtied pages
166 * (they were dirtied by commit). But that's OK - the blocks are
167 * safe in-journal, which is all fsync() needs to ensure.
168 */
169 if (!sbi->s_journal)
170 ret = ext4_fsync_nojournal(inode, datasync, &needs_barrier);
171 else if (ext4_should_journal_data(inode))
172 ret = ext4_force_commit(inode->i_sb);
173 else
174 ret = ext4_fsync_journal(inode, datasync, &needs_barrier);
175
176 if (needs_barrier) {
177 err = blkdev_issue_flush(inode->i_sb->s_bdev);
178 if (!ret)
179 ret = err;
180 }
181out:
182 err = file_check_and_advance_wb_err(file);
183 if (ret == 0)
184 ret = err;
185 trace_ext4_sync_file_exit(inode, ret);
186 return ret;
187}
1/*
2 * linux/fs/ext4/fsync.c
3 *
4 * Copyright (C) 1993 Stephen Tweedie (sct@redhat.com)
5 * from
6 * Copyright (C) 1992 Remy Card (card@masi.ibp.fr)
7 * Laboratoire MASI - Institut Blaise Pascal
8 * Universite Pierre et Marie Curie (Paris VI)
9 * from
10 * linux/fs/minix/truncate.c Copyright (C) 1991, 1992 Linus Torvalds
11 *
12 * ext4fs fsync primitive
13 *
14 * Big-endian to little-endian byte-swapping/bitmaps by
15 * David S. Miller (davem@caip.rutgers.edu), 1995
16 *
17 * Removed unnecessary code duplication for little endian machines
18 * and excessive __inline__s.
19 * Andi Kleen, 1997
20 *
21 * Major simplications and cleanup - we only need to do the metadata, because
22 * we can depend on generic_block_fdatasync() to sync the data blocks.
23 */
24
25#include <linux/time.h>
26#include <linux/fs.h>
27#include <linux/sched.h>
28#include <linux/writeback.h>
29#include <linux/blkdev.h>
30
31#include "ext4.h"
32#include "ext4_jbd2.h"
33
34#include <trace/events/ext4.h>
35
36/*
37 * If we're not journaling and this is a just-created file, we have to
38 * sync our parent directory (if it was freshly created) since
39 * otherwise it will only be written by writeback, leaving a huge
40 * window during which a crash may lose the file. This may apply for
41 * the parent directory's parent as well, and so on recursively, if
42 * they are also freshly created.
43 */
44static int ext4_sync_parent(struct inode *inode)
45{
46 struct dentry *dentry = NULL;
47 struct inode *next;
48 int ret = 0;
49
50 if (!ext4_test_inode_state(inode, EXT4_STATE_NEWENTRY))
51 return 0;
52 inode = igrab(inode);
53 while (ext4_test_inode_state(inode, EXT4_STATE_NEWENTRY)) {
54 ext4_clear_inode_state(inode, EXT4_STATE_NEWENTRY);
55 dentry = d_find_any_alias(inode);
56 if (!dentry)
57 break;
58 next = igrab(d_inode(dentry->d_parent));
59 dput(dentry);
60 if (!next)
61 break;
62 iput(inode);
63 inode = next;
64 ret = sync_mapping_buffers(inode->i_mapping);
65 if (ret)
66 break;
67 ret = sync_inode_metadata(inode, 1);
68 if (ret)
69 break;
70 }
71 iput(inode);
72 return ret;
73}
74
75/*
76 * akpm: A new design for ext4_sync_file().
77 *
78 * This is only called from sys_fsync(), sys_fdatasync() and sys_msync().
79 * There cannot be a transaction open by this task.
80 * Another task could have dirtied this inode. Its data can be in any
81 * state in the journalling system.
82 *
83 * What we do is just kick off a commit and wait on it. This will snapshot the
84 * inode to disk.
85 */
86
87int ext4_sync_file(struct file *file, loff_t start, loff_t end, int datasync)
88{
89 struct inode *inode = file->f_mapping->host;
90 struct ext4_inode_info *ei = EXT4_I(inode);
91 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
92 int ret = 0, err;
93 tid_t commit_tid;
94 bool needs_barrier = false;
95
96 J_ASSERT(ext4_journal_current_handle() == NULL);
97
98 trace_ext4_sync_file_enter(file, datasync);
99
100 if (inode->i_sb->s_flags & MS_RDONLY) {
101 /* Make sure that we read updated s_mount_flags value */
102 smp_rmb();
103 if (EXT4_SB(inode->i_sb)->s_mount_flags & EXT4_MF_FS_ABORTED)
104 ret = -EROFS;
105 goto out;
106 }
107
108 if (!journal) {
109 ret = generic_file_fsync(file, start, end, datasync);
110 if (!ret && !hlist_empty(&inode->i_dentry))
111 ret = ext4_sync_parent(inode);
112 goto out;
113 }
114
115 ret = filemap_write_and_wait_range(inode->i_mapping, start, end);
116 if (ret)
117 return ret;
118 /*
119 * data=writeback,ordered:
120 * The caller's filemap_fdatawrite()/wait will sync the data.
121 * Metadata is in the journal, we wait for proper transaction to
122 * commit here.
123 *
124 * data=journal:
125 * filemap_fdatawrite won't do anything (the buffers are clean).
126 * ext4_force_commit will write the file data into the journal and
127 * will wait on that.
128 * filemap_fdatawait() will encounter a ton of newly-dirtied pages
129 * (they were dirtied by commit). But that's OK - the blocks are
130 * safe in-journal, which is all fsync() needs to ensure.
131 */
132 if (ext4_should_journal_data(inode)) {
133 ret = ext4_force_commit(inode->i_sb);
134 goto out;
135 }
136
137 commit_tid = datasync ? ei->i_datasync_tid : ei->i_sync_tid;
138 if (journal->j_flags & JBD2_BARRIER &&
139 !jbd2_trans_will_send_data_barrier(journal, commit_tid))
140 needs_barrier = true;
141 ret = jbd2_complete_transaction(journal, commit_tid);
142 if (needs_barrier) {
143 err = blkdev_issue_flush(inode->i_sb->s_bdev, GFP_KERNEL, NULL);
144 if (!ret)
145 ret = err;
146 }
147out:
148 trace_ext4_sync_file_exit(inode, ret);
149 return ret;
150}