xref: /openbmc/linux/fs/ext4/fsync.c (revision 2b0143b5)
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/jbd2.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  */
45 static int ext4_sync_parent(struct inode *inode)
46 {
47 	struct dentry *dentry = NULL;
48 	struct inode *next;
49 	int ret = 0;
50 
51 	if (!ext4_test_inode_state(inode, EXT4_STATE_NEWENTRY))
52 		return 0;
53 	inode = igrab(inode);
54 	while (ext4_test_inode_state(inode, EXT4_STATE_NEWENTRY)) {
55 		ext4_clear_inode_state(inode, EXT4_STATE_NEWENTRY);
56 		dentry = d_find_any_alias(inode);
57 		if (!dentry)
58 			break;
59 		next = igrab(d_inode(dentry->d_parent));
60 		dput(dentry);
61 		if (!next)
62 			break;
63 		iput(inode);
64 		inode = next;
65 		ret = sync_mapping_buffers(inode->i_mapping);
66 		if (ret)
67 			break;
68 		ret = sync_inode_metadata(inode, 1);
69 		if (ret)
70 			break;
71 	}
72 	iput(inode);
73 	return ret;
74 }
75 
76 /*
77  * akpm: A new design for ext4_sync_file().
78  *
79  * This is only called from sys_fsync(), sys_fdatasync() and sys_msync().
80  * There cannot be a transaction open by this task.
81  * Another task could have dirtied this inode.  Its data can be in any
82  * state in the journalling system.
83  *
84  * What we do is just kick off a commit and wait on it.  This will snapshot the
85  * inode to disk.
86  */
87 
88 int ext4_sync_file(struct file *file, loff_t start, loff_t end, int datasync)
89 {
90 	struct inode *inode = file->f_mapping->host;
91 	struct ext4_inode_info *ei = EXT4_I(inode);
92 	journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
93 	int ret = 0, err;
94 	tid_t commit_tid;
95 	bool needs_barrier = false;
96 
97 	J_ASSERT(ext4_journal_current_handle() == NULL);
98 
99 	trace_ext4_sync_file_enter(file, datasync);
100 
101 	if (inode->i_sb->s_flags & MS_RDONLY) {
102 		/* Make sure that we read updated s_mount_flags value */
103 		smp_rmb();
104 		if (EXT4_SB(inode->i_sb)->s_mount_flags & EXT4_MF_FS_ABORTED)
105 			ret = -EROFS;
106 		goto out;
107 	}
108 
109 	if (!journal) {
110 		ret = generic_file_fsync(file, start, end, datasync);
111 		if (!ret && !hlist_empty(&inode->i_dentry))
112 			ret = ext4_sync_parent(inode);
113 		goto out;
114 	}
115 
116 	ret = filemap_write_and_wait_range(inode->i_mapping, start, end);
117 	if (ret)
118 		return ret;
119 	/*
120 	 * data=writeback,ordered:
121 	 *  The caller's filemap_fdatawrite()/wait will sync the data.
122 	 *  Metadata is in the journal, we wait for proper transaction to
123 	 *  commit here.
124 	 *
125 	 * data=journal:
126 	 *  filemap_fdatawrite won't do anything (the buffers are clean).
127 	 *  ext4_force_commit will write the file data into the journal and
128 	 *  will wait on that.
129 	 *  filemap_fdatawait() will encounter a ton of newly-dirtied pages
130 	 *  (they were dirtied by commit).  But that's OK - the blocks are
131 	 *  safe in-journal, which is all fsync() needs to ensure.
132 	 */
133 	if (ext4_should_journal_data(inode)) {
134 		ret = ext4_force_commit(inode->i_sb);
135 		goto out;
136 	}
137 
138 	commit_tid = datasync ? ei->i_datasync_tid : ei->i_sync_tid;
139 	if (journal->j_flags & JBD2_BARRIER &&
140 	    !jbd2_trans_will_send_data_barrier(journal, commit_tid))
141 		needs_barrier = true;
142 	ret = jbd2_complete_transaction(journal, commit_tid);
143 	if (needs_barrier) {
144 		err = blkdev_issue_flush(inode->i_sb->s_bdev, GFP_KERNEL, NULL);
145 		if (!ret)
146 			ret = err;
147 	}
148 out:
149 	trace_ext4_sync_file_exit(inode, ret);
150 	return ret;
151 }
152