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, *next;
49 if (!ext4_test_inode_state(inode, EXT4_STATE_NEWENTRY))
51 dentry = d_find_any_alias(inode);
54 while (ext4_test_inode_state(inode, EXT4_STATE_NEWENTRY)) {
55 ext4_clear_inode_state(inode, EXT4_STATE_NEWENTRY);
57 next = dget_parent(dentry);
60 inode = dentry->d_inode;
63 * The directory inode may have gone through rmdir by now. But
64 * the inode itself and its blocks are still allocated (we hold
65 * a reference to the inode via its dentry), so it didn't go
66 * through ext4_evict_inode()) and so we are safe to flush
67 * metadata blocks and the inode.
69 ret = sync_mapping_buffers(inode->i_mapping);
72 ret = sync_inode_metadata(inode, 1);
81 * akpm: A new design for ext4_sync_file().
83 * This is only called from sys_fsync(), sys_fdatasync() and sys_msync().
84 * There cannot be a transaction open by this task.
85 * Another task could have dirtied this inode. Its data can be in any
86 * state in the journalling system.
88 * What we do is just kick off a commit and wait on it. This will snapshot the
92 int ext4_sync_file(struct file *file, loff_t start, loff_t end, int datasync)
94 struct inode *inode = file->f_mapping->host;
95 struct ext4_inode_info *ei = EXT4_I(inode);
96 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
99 bool needs_barrier = false;
101 J_ASSERT(ext4_journal_current_handle() == NULL);
103 trace_ext4_sync_file_enter(file, datasync);
105 if (inode->i_sb->s_flags & MS_RDONLY) {
106 /* Make sure that we read updated s_mount_flags value */
108 if (EXT4_SB(inode->i_sb)->s_mount_flags & EXT4_MF_FS_ABORTED)
114 ret = __generic_file_fsync(file, start, end, datasync);
116 ret = ext4_sync_parent(inode);
117 if (test_opt(inode->i_sb, BARRIER))
122 ret = filemap_write_and_wait_range(inode->i_mapping, start, end);
126 * data=writeback,ordered:
127 * The caller's filemap_fdatawrite()/wait will sync the data.
128 * Metadata is in the journal, we wait for proper transaction to
132 * filemap_fdatawrite won't do anything (the buffers are clean).
133 * ext4_force_commit will write the file data into the journal and
135 * filemap_fdatawait() will encounter a ton of newly-dirtied pages
136 * (they were dirtied by commit). But that's OK - the blocks are
137 * safe in-journal, which is all fsync() needs to ensure.
139 if (ext4_should_journal_data(inode)) {
140 ret = ext4_force_commit(inode->i_sb);
144 commit_tid = datasync ? ei->i_datasync_tid : ei->i_sync_tid;
145 if (journal->j_flags & JBD2_BARRIER &&
146 !jbd2_trans_will_send_data_barrier(journal, commit_tid))
147 needs_barrier = true;
148 ret = jbd2_complete_transaction(journal, commit_tid);
151 err = blkdev_issue_flush(inode->i_sb->s_bdev, GFP_KERNEL, NULL);
156 trace_ext4_sync_file_exit(inode, ret);