blob: 592adf2e546e0726f3d0b6fd34165dd3c769a815 [file] [log] [blame]
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001/*
Mingming Cao617ba132006-10-11 01:20:53 -07002 * linux/fs/ext4/fsync.c
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003 *
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 *
Mingming Cao617ba132006-10-11 01:20:53 -070012 * ext4fs fsync primitive
Dave Kleikampac27a0e2006-10-11 01:20:50 -070013 *
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>
Mingming Caodab291a2006-10-11 01:21:01 -070029#include <linux/jbd2.h>
Eric Sandeend755fb32008-07-11 19:27:31 -040030#include <linux/blkdev.h>
Theodore Ts'o9bffad12009-06-17 11:48:11 -040031
Christoph Hellwig3dcf5452008-04-29 18:13:32 -040032#include "ext4.h"
33#include "ext4_jbd2.h"
Dave Kleikampac27a0e2006-10-11 01:20:50 -070034
Theodore Ts'o9bffad12009-06-17 11:48:11 -040035#include <trace/events/ext4.h>
36
Dave Kleikampac27a0e2006-10-11 01:20:50 -070037/*
Frank Mayhar14ece102010-05-17 08:00:00 -040038 * 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 void ext4_sync_parent(struct inode *inode)
46{
47 struct dentry *dentry = NULL;
48
49 while (inode && ext4_test_inode_state(inode, EXT4_STATE_NEWENTRY)) {
50 ext4_clear_inode_state(inode, EXT4_STATE_NEWENTRY);
51 dentry = list_entry(inode->i_dentry.next,
52 struct dentry, d_alias);
53 if (!dentry || !dentry->d_parent || !dentry->d_parent->d_inode)
54 break;
55 inode = dentry->d_parent->d_inode;
56 sync_mapping_buffers(inode->i_mapping);
57 }
58}
59
60/*
Mingming Cao617ba132006-10-11 01:20:53 -070061 * akpm: A new design for ext4_sync_file().
Dave Kleikampac27a0e2006-10-11 01:20:50 -070062 *
63 * This is only called from sys_fsync(), sys_fdatasync() and sys_msync().
64 * There cannot be a transaction open by this task.
65 * Another task could have dirtied this inode. Its data can be in any
66 * state in the journalling system.
67 *
68 * What we do is just kick off a commit and wait on it. This will snapshot the
69 * inode to disk.
Mingming Cao8d5d02e2009-09-28 15:48:29 -040070 *
71 * i_mutex lock is held when entering and exiting this function
Dave Kleikampac27a0e2006-10-11 01:20:50 -070072 */
73
Christoph Hellwig7ea80852010-05-26 17:53:25 +020074int ext4_sync_file(struct file *file, int datasync)
Dave Kleikampac27a0e2006-10-11 01:20:50 -070075{
Christoph Hellwig7ea80852010-05-26 17:53:25 +020076 struct inode *inode = file->f_mapping->host;
Jan Karab436b9b2009-12-08 23:51:10 -050077 struct ext4_inode_info *ei = EXT4_I(inode);
Eric Sandeend755fb32008-07-11 19:27:31 -040078 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
Jan Karab436b9b2009-12-08 23:51:10 -050079 int ret;
80 tid_t commit_tid;
Dave Kleikampac27a0e2006-10-11 01:20:50 -070081
Aneesh Kumar K.Vac398492007-10-16 18:38:25 -040082 J_ASSERT(ext4_journal_current_handle() == NULL);
Dave Kleikampac27a0e2006-10-11 01:20:50 -070083
Christoph Hellwig7ea80852010-05-26 17:53:25 +020084 trace_ext4_sync_file(file, datasync);
Theodore Ts'oede86cc2008-10-05 20:50:06 -040085
Jan Karab436b9b2009-12-08 23:51:10 -050086 if (inode->i_sb->s_flags & MS_RDONLY)
87 return 0;
88
Jiaying Zhangc7064ef2010-03-02 13:28:44 -050089 ret = flush_completed_IO(inode);
Mingming Cao8d5d02e2009-09-28 15:48:29 -040090 if (ret < 0)
Theodore Ts'o6b17d902009-11-23 07:24:57 -050091 return ret;
Theodore Ts'o60e66792010-05-17 07:00:00 -040092
Frank Mayhar14ece102010-05-17 08:00:00 -040093 if (!journal) {
Christoph Hellwig1b061d92010-05-26 17:53:41 +020094 ret = generic_file_fsync(file, datasync);
Frank Mayhar14ece102010-05-17 08:00:00 -040095 if (!ret && !list_empty(&inode->i_dentry))
96 ext4_sync_parent(inode);
97 return ret;
98 }
Jan Karab436b9b2009-12-08 23:51:10 -050099
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700100 /*
Jan Karab436b9b2009-12-08 23:51:10 -0500101 * data=writeback,ordered:
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700102 * The caller's filemap_fdatawrite()/wait will sync the data.
Jan Karab436b9b2009-12-08 23:51:10 -0500103 * Metadata is in the journal, we wait for proper transaction to
104 * commit here.
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700105 *
106 * data=journal:
107 * filemap_fdatawrite won't do anything (the buffers are clean).
Mingming Cao617ba132006-10-11 01:20:53 -0700108 * ext4_force_commit will write the file data into the journal and
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700109 * will wait on that.
110 * filemap_fdatawait() will encounter a ton of newly-dirtied pages
111 * (they were dirtied by commit). But that's OK - the blocks are
112 * safe in-journal, which is all fsync() needs to ensure.
113 */
Theodore Ts'o6b17d902009-11-23 07:24:57 -0500114 if (ext4_should_journal_data(inode))
115 return ext4_force_commit(inode->i_sb);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700116
Jan Karab436b9b2009-12-08 23:51:10 -0500117 commit_tid = datasync ? ei->i_datasync_tid : ei->i_sync_tid;
Theodore Ts'occ3e1be2009-12-23 06:52:08 -0500118 if (jbd2_log_start_commit(journal, commit_tid)) {
119 /*
120 * When the journal is on a different device than the
121 * fs data disk, we need to issue the barrier in
122 * writeback mode. (In ordered mode, the jbd2 layer
123 * will take care of issuing the barrier. In
124 * data=journal, all of the data blocks are written to
125 * the journal device.)
126 */
127 if (ext4_should_writeback_data(inode) &&
128 (journal->j_fs_dev != journal->j_dev) &&
129 (journal->j_flags & JBD2_BARRIER))
Dmitry Monakhovfbd9b092010-04-28 17:55:06 +0400130 blkdev_issue_flush(inode->i_sb->s_bdev, GFP_KERNEL,
131 NULL, BLKDEV_IFL_WAIT);
Dmitry Monakhov0671e702010-05-10 00:00:00 -0400132 ret = jbd2_log_wait_commit(journal, commit_tid);
Theodore Ts'occ3e1be2009-12-23 06:52:08 -0500133 } else if (journal->j_flags & JBD2_BARRIER)
Dmitry Monakhovfbd9b092010-04-28 17:55:06 +0400134 blkdev_issue_flush(inode->i_sb->s_bdev, GFP_KERNEL, NULL,
135 BLKDEV_IFL_WAIT);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700136 return ret;
137}