xref: /openbmc/linux/fs/ext4/fsync.c (revision 84764a41)
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 writeback_control wbc;
48 	struct dentry *dentry = NULL;
49 	struct inode *next;
50 	int ret = 0;
51 
52 	if (!ext4_test_inode_state(inode, EXT4_STATE_NEWENTRY))
53 		return 0;
54 	inode = igrab(inode);
55 	while (ext4_test_inode_state(inode, EXT4_STATE_NEWENTRY)) {
56 		ext4_clear_inode_state(inode, EXT4_STATE_NEWENTRY);
57 		dentry = d_find_any_alias(inode);
58 		if (!dentry)
59 			break;
60 		next = igrab(dentry->d_parent->d_inode);
61 		dput(dentry);
62 		if (!next)
63 			break;
64 		iput(inode);
65 		inode = next;
66 		ret = sync_mapping_buffers(inode->i_mapping);
67 		if (ret)
68 			break;
69 		memset(&wbc, 0, sizeof(wbc));
70 		wbc.sync_mode = WB_SYNC_ALL;
71 		wbc.nr_to_write = 0;         /* only write out the inode */
72 		ret = sync_inode(inode, &wbc);
73 		if (ret)
74 			break;
75 	}
76 	iput(inode);
77 	return ret;
78 }
79 
80 /**
81  * __sync_file - generic_file_fsync without the locking and filemap_write
82  * @inode:	inode to sync
83  * @datasync:	only sync essential metadata if true
84  *
85  * This is just generic_file_fsync without the locking.  This is needed for
86  * nojournal mode to make sure this inodes data/metadata makes it to disk
87  * properly.  The i_mutex should be held already.
88  */
89 static int __sync_inode(struct inode *inode, int datasync)
90 {
91 	int err;
92 	int ret;
93 
94 	ret = sync_mapping_buffers(inode->i_mapping);
95 	if (!(inode->i_state & I_DIRTY))
96 		return ret;
97 	if (datasync && !(inode->i_state & I_DIRTY_DATASYNC))
98 		return ret;
99 
100 	err = sync_inode_metadata(inode, 1);
101 	if (ret == 0)
102 		ret = err;
103 	return ret;
104 }
105 
106 /*
107  * akpm: A new design for ext4_sync_file().
108  *
109  * This is only called from sys_fsync(), sys_fdatasync() and sys_msync().
110  * There cannot be a transaction open by this task.
111  * Another task could have dirtied this inode.  Its data can be in any
112  * state in the journalling system.
113  *
114  * What we do is just kick off a commit and wait on it.  This will snapshot the
115  * inode to disk.
116  *
117  * i_mutex lock is held when entering and exiting this function
118  */
119 
120 int ext4_sync_file(struct file *file, loff_t start, loff_t end, int datasync)
121 {
122 	struct inode *inode = file->f_mapping->host;
123 	struct ext4_inode_info *ei = EXT4_I(inode);
124 	journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
125 	int ret, err;
126 	tid_t commit_tid;
127 	bool needs_barrier = false;
128 
129 	J_ASSERT(ext4_journal_current_handle() == NULL);
130 
131 	trace_ext4_sync_file_enter(file, datasync);
132 
133 	ret = filemap_write_and_wait_range(inode->i_mapping, start, end);
134 	if (ret)
135 		return ret;
136 	mutex_lock(&inode->i_mutex);
137 
138 	if (inode->i_sb->s_flags & MS_RDONLY)
139 		goto out;
140 
141 	ret = ext4_flush_unwritten_io(inode);
142 	if (ret < 0)
143 		goto out;
144 
145 	if (!journal) {
146 		ret = __sync_inode(inode, datasync);
147 		if (!ret && !hlist_empty(&inode->i_dentry))
148 			ret = ext4_sync_parent(inode);
149 		goto out;
150 	}
151 
152 	/*
153 	 * data=writeback,ordered:
154 	 *  The caller's filemap_fdatawrite()/wait will sync the data.
155 	 *  Metadata is in the journal, we wait for proper transaction to
156 	 *  commit here.
157 	 *
158 	 * data=journal:
159 	 *  filemap_fdatawrite won't do anything (the buffers are clean).
160 	 *  ext4_force_commit will write the file data into the journal and
161 	 *  will wait on that.
162 	 *  filemap_fdatawait() will encounter a ton of newly-dirtied pages
163 	 *  (they were dirtied by commit).  But that's OK - the blocks are
164 	 *  safe in-journal, which is all fsync() needs to ensure.
165 	 */
166 	if (ext4_should_journal_data(inode)) {
167 		ret = ext4_force_commit(inode->i_sb);
168 		goto out;
169 	}
170 
171 	commit_tid = datasync ? ei->i_datasync_tid : ei->i_sync_tid;
172 	if (journal->j_flags & JBD2_BARRIER &&
173 	    !jbd2_trans_will_send_data_barrier(journal, commit_tid))
174 		needs_barrier = true;
175 	jbd2_log_start_commit(journal, commit_tid);
176 	ret = jbd2_log_wait_commit(journal, commit_tid);
177 	if (needs_barrier) {
178 		err = blkdev_issue_flush(inode->i_sb->s_bdev, GFP_KERNEL, NULL);
179 		if (!ret)
180 			ret = err;
181 	}
182  out:
183 	mutex_unlock(&inode->i_mutex);
184 	trace_ext4_sync_file_exit(inode, ret);
185 	return ret;
186 }
187