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 */
ext4_sync_parent(struct inode * inode)45 static 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
81 /*
82 * akpm: A new design for ext4_sync_file().
83 *
84 * This is only called from sys_fsync(), sys_fdatasync() and sys_msync().
85 * There cannot be a transaction open by this task.
86 * Another task could have dirtied this inode. Its data can be in any
87 * state in the journalling system.
88 *
89 * What we do is just kick off a commit and wait on it. This will snapshot the
90 * inode to disk.
91 */
92
ext4_sync_file(struct file * file,loff_t start,loff_t end,int datasync)93 int ext4_sync_file(struct file *file, loff_t start, loff_t end, int datasync)
94 {
95 struct inode *inode = file->f_mapping->host;
96 struct ext4_inode_info *ei = EXT4_I(inode);
97 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
98 int ret = 0, err;
99 tid_t commit_tid;
100 bool needs_barrier = false;
101
102 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
103 return -EIO;
104
105 J_ASSERT(ext4_journal_current_handle() == NULL);
106
107 trace_ext4_sync_file_enter(file, datasync);
108
109 if (sb_rdonly(inode->i_sb)) {
110 /* Make sure that we read updated s_mount_flags value */
111 smp_rmb();
112 if (EXT4_SB(inode->i_sb)->s_mount_flags & EXT4_MF_FS_ABORTED)
113 ret = -EROFS;
114 goto out;
115 }
116
117 if (!journal) {
118 ret = __generic_file_fsync(file, start, end, datasync);
119 if (!ret)
120 ret = ext4_sync_parent(inode);
121 if (test_opt(inode->i_sb, BARRIER))
122 goto issue_flush;
123 goto out;
124 }
125
126 ret = file_write_and_wait_range(file, start, end);
127 if (ret)
128 return ret;
129 /*
130 * data=writeback,ordered:
131 * The caller's filemap_fdatawrite()/wait will sync the data.
132 * Metadata is in the journal, we wait for proper transaction to
133 * commit here.
134 *
135 * data=journal:
136 * filemap_fdatawrite won't do anything (the buffers are clean).
137 * ext4_force_commit will write the file data into the journal and
138 * will wait on that.
139 * filemap_fdatawait() will encounter a ton of newly-dirtied pages
140 * (they were dirtied by commit). But that's OK - the blocks are
141 * safe in-journal, which is all fsync() needs to ensure.
142 */
143 if (ext4_should_journal_data(inode)) {
144 ret = ext4_force_commit(inode->i_sb);
145 goto out;
146 }
147
148 commit_tid = datasync ? ei->i_datasync_tid : ei->i_sync_tid;
149 if (journal->j_flags & JBD2_BARRIER &&
150 !jbd2_trans_will_send_data_barrier(journal, commit_tid))
151 needs_barrier = true;
152 ret = jbd2_complete_transaction(journal, commit_tid);
153 if (needs_barrier) {
154 issue_flush:
155 err = blkdev_issue_flush(inode->i_sb->s_bdev, GFP_KERNEL, NULL);
156 if (!ret)
157 ret = err;
158 }
159 out:
160 err = file_check_and_advance_wb_err(file);
161 if (ret == 0)
162 ret = err;
163 trace_ext4_sync_file_exit(inode, ret);
164 return ret;
165 }
166