fsync.c

  1// SPDX-License-Identifier: GPL-2.0
  2/*
  3 *  linux/fs/ext4/fsync.c
  4 *
  5 *  Copyright (C) 1993  Stephen Tweedie (sct@redhat.com)
  6 *  from
  7 *  Copyright (C) 1992  Remy Card (card@masi.ibp.fr)
  8 *                      Laboratoire MASI - Institut Blaise Pascal
  9 *                      Universite Pierre et Marie Curie (Paris VI)
 10 *  from
 11 *  linux/fs/minix/truncate.c   Copyright (C) 1991, 1992  Linus Torvalds
 12 *
 13 *  ext4fs fsync primitive
 14 *
 15 *  Big-endian to little-endian byte-swapping/bitmaps by
 16 *        David S. Miller (davem@caip.rutgers.edu), 1995
 17 *
 18 *  Removed unnecessary code duplication for little endian machines
 19 *  and excessive __inline__s.
 20 *        Andi Kleen, 1997
 21 *
 22 * Major simplications and cleanup - we only need to do the metadata, because
 23 * we can depend on generic_block_fdatasync() to sync the data blocks.
 24 */
 25
 26#include <linux/time.h>
 27#include <linux/fs.h>
 28#include <linux/sched.h>
 29#include <linux/writeback.h>
 30#include <linux/blkdev.h>
 31#include <linux/buffer_head.h>
 32
 33#include "ext4.h"
 34#include "ext4_jbd2.h"
 35
 36#include <trace/events/ext4.h>
 37
 38/*
 39 * If we're not journaling and this is a just-created file, we have to
 40 * sync our parent directory (if it was freshly created) since
 41 * otherwise it will only be written by writeback, leaving a huge
 42 * window during which a crash may lose the file.  This may apply for
 43 * the parent directory's parent as well, and so on recursively, if
 44 * they are also freshly created.
 45 */
 46static int ext4_sync_parent(struct inode *inode)
 47{
 48	struct dentry *dentry, *next;
 49	int ret = 0;
 50
 51	if (!ext4_test_inode_state(inode, EXT4_STATE_NEWENTRY))
 52		return 0;
 53	dentry = d_find_any_alias(inode);
 54	if (!dentry)
 55		return 0;
 56	while (ext4_test_inode_state(inode, EXT4_STATE_NEWENTRY)) {
 57		ext4_clear_inode_state(inode, EXT4_STATE_NEWENTRY);
 58
 59		next = dget_parent(dentry);
 60		dput(dentry);
 61		dentry = next;
 62		inode = dentry->d_inode;
 63
 64		/*
 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 via its dentry), so it didn't go
 68		 * through ext4_evict_inode()) and so we are safe to flush
 69		 * metadata blocks and the inode.
 70		 */
 71		ret = sync_mapping_buffers(inode->i_mapping);
 72		if (ret)
 73			break;
 74		ret = sync_inode_metadata(inode, 1);
 75		if (ret)
 76			break;
 77	}
 78	dput(dentry);
 79	return ret;
 80}
 81
 82static int ext4_fsync_nojournal(struct file *file, loff_t start, loff_t end,
 83				int datasync, bool *needs_barrier)
 84{
 85	struct inode *inode = file->f_inode;
 86	struct writeback_control wbc = {
 87		.sync_mode = WB_SYNC_ALL,
 88		.nr_to_write = 0,
 89	};
 90	int ret;
 91
 92	ret = generic_buffers_fsync_noflush(file, start, end, datasync);
 93	if (ret)
 94		return ret;
 95
 96	/* Force writeout of inode table buffer to disk */
 97	ret = ext4_write_inode(inode, &wbc);
 98	if (ret)
 99		return ret;
100
101	ret = ext4_sync_parent(inode);
102
103	if (test_opt(inode->i_sb, BARRIER))
104		*needs_barrier = true;
105
106	return ret;
107}
108
109static int ext4_fsync_journal(struct inode *inode, bool datasync,
110			     bool *needs_barrier)
111{
112	struct ext4_inode_info *ei = EXT4_I(inode);
113	journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
114	tid_t commit_tid = datasync ? ei->i_datasync_tid : ei->i_sync_tid;
115
116	/*
117	 * Fastcommit does not really support fsync on directories or other
118	 * special files. Force a full commit.
119	 */
120	if (!S_ISREG(inode->i_mode))
121		return ext4_force_commit(inode->i_sb);
122
123	if (journal->j_flags & JBD2_BARRIER &&
124	    !jbd2_trans_will_send_data_barrier(journal, commit_tid))
125		*needs_barrier = true;
126
127	return ext4_fc_commit(journal, commit_tid);
128}
129
130/*
131 * akpm: A new design for ext4_sync_file().
132 *
133 * This is only called from sys_fsync(), sys_fdatasync() and sys_msync().
134 * There cannot be a transaction open by this task.
135 * Another task could have dirtied this inode.  Its data can be in any
136 * state in the journalling system.
137 *
138 * What we do is just kick off a commit and wait on it.  This will snapshot the
139 * inode to disk.
140 */
141int ext4_sync_file(struct file *file, loff_t start, loff_t end, int datasync)
142{
143	int ret = 0, err;
144	bool needs_barrier = false;
145	struct inode *inode = file->f_mapping->host;
146
147	ret = ext4_emergency_state(inode->i_sb);
148	if (unlikely(ret))
149		return ret;
150
151	ASSERT(ext4_journal_current_handle() == NULL);
152
153	trace_ext4_sync_file_enter(file, datasync);
154
155	if (sb_rdonly(inode->i_sb))
156		goto out;
157
158	if (!EXT4_SB(inode->i_sb)->s_journal) {
159		ret = ext4_fsync_nojournal(file, start, end, datasync,
160					   &needs_barrier);
161		if (needs_barrier)
162			goto issue_flush;
163		goto out;
164	}
165
166	ret = file_write_and_wait_range(file, start, end);
167	if (ret)
168		goto out;
169
170	/*
171	 *  The caller's filemap_fdatawrite()/wait will sync the data.
172	 *  Metadata is in the journal, we wait for proper transaction to
173	 *  commit here.
174	 */
175	ret = ext4_fsync_journal(inode, datasync, &needs_barrier);
176
177issue_flush:
178	if (needs_barrier) {
179		err = blkdev_issue_flush(inode->i_sb->s_bdev);
180		if (!ret)
181			ret = err;
182	}
183out:
184	err = file_check_and_advance_wb_err(file);
185	if (ret == 0)
186		ret = err;
187	trace_ext4_sync_file_exit(inode, ret);
188	return ret;
189}