2 * linux/fs/ext4/fsync.c
4 * Copyright (C) 1993 Stephen Tweedie (sct@redhat.com)
6 * Copyright (C) 1992 Remy Card (card@masi.ibp.fr)
7 * Laboratoire MASI - Institut Blaise Pascal
8 * Universite Pierre et Marie Curie (Paris VI)
10 * linux/fs/minix/truncate.c Copyright (C) 1991, 1992 Linus Torvalds
12 * ext4fs fsync primitive
14 * Big-endian to little-endian byte-swapping/bitmaps by
15 * David S. Miller (davem@caip.rutgers.edu), 1995
17 * Removed unnecessary code duplication for little endian machines
18 * and excessive __inline__s.
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.
25 #include <linux/time.h>
27 #include <linux/sched.h>
28 #include <linux/writeback.h>
29 #include <linux/blkdev.h>
32 #include "ext4_jbd2.h"
34 #include <trace/events/ext4.h>
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.
44 static int ext4_sync_parent(struct inode *inode)
46 struct dentry *dentry = NULL;
50 if (!ext4_test_inode_state(inode, EXT4_STATE_NEWENTRY))
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);
58 next = igrab(d_inode(dentry->d_parent));
65 * The directory inode may have gone through rmdir by now. But
66 * the inode itself and its blocks are still allocated (we hold
67 * a reference to the inode so it didn't go through
68 * ext4_evict_inode()) and so we are safe to flush metadata
69 * blocks and the inode.
71 ret = sync_mapping_buffers(inode->i_mapping);
74 ret = sync_inode_metadata(inode, 1);
83 * akpm: A new design for ext4_sync_file().
85 * This is only called from sys_fsync(), sys_fdatasync() and sys_msync().
86 * There cannot be a transaction open by this task.
87 * Another task could have dirtied this inode. Its data can be in any
88 * state in the journalling system.
90 * What we do is just kick off a commit and wait on it. This will snapshot the
94 int ext4_sync_file(struct file *file, loff_t start, loff_t end, int datasync)
96 struct inode *inode = file->f_mapping->host;
97 struct ext4_inode_info *ei = EXT4_I(inode);
98 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
101 bool needs_barrier = false;
103 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
106 J_ASSERT(ext4_journal_current_handle() == NULL);
108 trace_ext4_sync_file_enter(file, datasync);
110 if (inode->i_sb->s_flags & MS_RDONLY) {
111 /* Make sure that we read updated s_mount_flags value */
113 if (EXT4_SB(inode->i_sb)->s_mount_flags & EXT4_MF_FS_ABORTED)
119 ret = __generic_file_fsync(file, start, end, datasync);
121 ret = ext4_sync_parent(inode);
122 if (test_opt(inode->i_sb, BARRIER))
127 ret = file_write_and_wait_range(file, start, end);
131 * data=writeback,ordered:
132 * The caller's filemap_fdatawrite()/wait will sync the data.
133 * Metadata is in the journal, we wait for proper transaction to
137 * filemap_fdatawrite won't do anything (the buffers are clean).
138 * ext4_force_commit will write the file data into the journal and
140 * filemap_fdatawait() will encounter a ton of newly-dirtied pages
141 * (they were dirtied by commit). But that's OK - the blocks are
142 * safe in-journal, which is all fsync() needs to ensure.
144 if (ext4_should_journal_data(inode)) {
145 ret = ext4_force_commit(inode->i_sb);
149 commit_tid = datasync ? ei->i_datasync_tid : ei->i_sync_tid;
150 if (journal->j_flags & JBD2_BARRIER &&
151 !jbd2_trans_will_send_data_barrier(journal, commit_tid))
152 needs_barrier = true;
153 ret = jbd2_complete_transaction(journal, commit_tid);
156 err = blkdev_issue_flush(inode->i_sb->s_bdev, GFP_KERNEL, NULL);
161 trace_ext4_sync_file_exit(inode, ret);