1 // SPDX-License-Identifier: GPL-2.0
3 * High-level sync()-related operations
6 #include <linux/kernel.h>
7 #include <linux/file.h>
9 #include <linux/slab.h>
10 #include <linux/export.h>
11 #include <linux/namei.h>
12 #include <linux/sched.h>
13 #include <linux/writeback.h>
14 #include <linux/syscalls.h>
15 #include <linux/linkage.h>
16 #include <linux/pagemap.h>
17 #include <linux/quotaops.h>
18 #include <linux/backing-dev.h>
21 #define VALID_FLAGS (SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE| \
22 SYNC_FILE_RANGE_WAIT_AFTER)
25 * Do the filesystem syncing work. For simple filesystems
26 * writeback_inodes_sb(sb) just dirties buffers with inodes so we have to
27 * submit IO for these buffers via __sync_blockdev(). This also speeds up the
28 * wait == 1 case since in that case write_inode() functions do
29 * sync_dirty_buffer() and thus effectively write one block at a time.
31 int __sync_filesystem(struct super_block
*sb
, int wait
)
36 writeback_inodes_sb(sb
, WB_REASON_SYNC
);
38 if (sb
->s_op
->sync_fs
)
39 sb
->s_op
->sync_fs(sb
, wait
);
40 return __sync_blockdev(sb
->s_bdev
, wait
);
42 EXPORT_SYMBOL_GPL(__sync_filesystem
);
45 * Write out and wait upon all dirty data associated with this
46 * superblock. Filesystem data as well as the underlying block
47 * device. Takes the superblock lock.
49 int sync_filesystem(struct super_block
*sb
)
54 * We need to be protected against the filesystem going from
55 * r/o to r/w or vice versa.
57 WARN_ON(!rwsem_is_locked(&sb
->s_umount
));
60 * No point in syncing out anything if the filesystem is read-only.
65 ret
= __sync_filesystem(sb
, 0);
68 return __sync_filesystem(sb
, 1);
70 EXPORT_SYMBOL(sync_filesystem
);
72 static void sync_inodes_one_sb(struct super_block
*sb
, void *arg
)
78 static void sync_fs_one_sb(struct super_block
*sb
, void *arg
)
80 if (!sb_rdonly(sb
) && sb
->s_op
->sync_fs
)
81 sb
->s_op
->sync_fs(sb
, *(int *)arg
);
84 static void fdatawrite_one_bdev(struct block_device
*bdev
, void *arg
)
86 filemap_fdatawrite(bdev
->bd_inode
->i_mapping
);
89 static void fdatawait_one_bdev(struct block_device
*bdev
, void *arg
)
92 * We keep the error status of individual mapping so that
93 * applications can catch the writeback error using fsync(2).
94 * See filemap_fdatawait_keep_errors() for details.
96 filemap_fdatawait_keep_errors(bdev
->bd_inode
->i_mapping
);
100 * Sync everything. We start by waking flusher threads so that most of
101 * writeback runs on all devices in parallel. Then we sync all inodes reliably
102 * which effectively also waits for all flusher threads to finish doing
103 * writeback. At this point all data is on disk so metadata should be stable
104 * and we tell filesystems to sync their metadata via ->sync_fs() calls.
105 * Finally, we writeout all block devices because some filesystems (e.g. ext2)
106 * just write metadata (such as inodes or bitmaps) to block device page cache
107 * and do not sync it on their own in ->sync_fs().
109 SYSCALL_DEFINE0(sync
)
111 int nowait
= 0, wait
= 1;
113 wakeup_flusher_threads(WB_REASON_SYNC
);
114 iterate_supers(sync_inodes_one_sb
, NULL
);
115 iterate_supers(sync_fs_one_sb
, &nowait
);
116 iterate_supers(sync_fs_one_sb
, &wait
);
117 iterate_bdevs(fdatawrite_one_bdev
, NULL
);
118 iterate_bdevs(fdatawait_one_bdev
, NULL
);
119 if (unlikely(laptop_mode
))
120 laptop_sync_completion();
124 static void do_sync_work(struct work_struct
*work
)
129 * Sync twice to reduce the possibility we skipped some inodes / pages
130 * because they were temporarily locked
132 iterate_supers(sync_inodes_one_sb
, &nowait
);
133 iterate_supers(sync_fs_one_sb
, &nowait
);
134 iterate_bdevs(fdatawrite_one_bdev
, NULL
);
135 iterate_supers(sync_inodes_one_sb
, &nowait
);
136 iterate_supers(sync_fs_one_sb
, &nowait
);
137 iterate_bdevs(fdatawrite_one_bdev
, NULL
);
138 printk("Emergency Sync complete\n");
142 void emergency_sync(void)
144 struct work_struct
*work
;
146 work
= kmalloc(sizeof(*work
), GFP_ATOMIC
);
148 INIT_WORK(work
, do_sync_work
);
154 * sync a single super
156 SYSCALL_DEFINE1(syncfs
, int, fd
)
158 struct fd f
= fdget(fd
);
159 struct super_block
*sb
;
164 sb
= f
.file
->f_path
.dentry
->d_sb
;
166 down_read(&sb
->s_umount
);
167 ret
= sync_filesystem(sb
);
168 up_read(&sb
->s_umount
);
175 * vfs_fsync_range - helper to sync a range of data & metadata to disk
176 * @file: file to sync
177 * @start: offset in bytes of the beginning of data range to sync
178 * @end: offset in bytes of the end of data range (inclusive)
179 * @datasync: perform only datasync
181 * Write back data in range @start..@end and metadata for @file to disk. If
182 * @datasync is set only metadata needed to access modified file data is
185 int vfs_fsync_range(struct file
*file
, loff_t start
, loff_t end
, int datasync
)
187 struct inode
*inode
= file
->f_mapping
->host
;
189 if (!file
->f_op
->fsync
)
191 if (!datasync
&& (inode
->i_state
& I_DIRTY_TIME
)) {
192 spin_lock(&inode
->i_lock
);
193 inode
->i_state
&= ~I_DIRTY_TIME
;
194 spin_unlock(&inode
->i_lock
);
195 mark_inode_dirty_sync(inode
);
197 return file
->f_op
->fsync(file
, start
, end
, datasync
);
199 EXPORT_SYMBOL(vfs_fsync_range
);
202 * vfs_fsync - perform a fsync or fdatasync on a file
203 * @file: file to sync
204 * @datasync: only perform a fdatasync operation
206 * Write back data and metadata for @file to disk. If @datasync is
207 * set only metadata needed to access modified file data is written.
209 int vfs_fsync(struct file
*file
, int datasync
)
211 return vfs_fsync_range(file
, 0, LLONG_MAX
, datasync
);
213 EXPORT_SYMBOL(vfs_fsync
);
215 static int do_fsync(unsigned int fd
, int datasync
)
217 struct fd f
= fdget(fd
);
221 ret
= vfs_fsync(f
.file
, datasync
);
227 SYSCALL_DEFINE1(fsync
, unsigned int, fd
)
229 return do_fsync(fd
, 0);
232 SYSCALL_DEFINE1(fdatasync
, unsigned int, fd
)
234 return do_fsync(fd
, 1);
238 * sys_sync_file_range() permits finely controlled syncing over a segment of
239 * a file in the range offset .. (offset+nbytes-1) inclusive. If nbytes is
240 * zero then sys_sync_file_range() will operate from offset out to EOF.
244 * SYNC_FILE_RANGE_WAIT_BEFORE: wait upon writeout of all pages in the range
245 * before performing the write.
247 * SYNC_FILE_RANGE_WRITE: initiate writeout of all those dirty pages in the
248 * range which are not presently under writeback. Note that this may block for
249 * significant periods due to exhaustion of disk request structures.
251 * SYNC_FILE_RANGE_WAIT_AFTER: wait upon writeout of all pages in the range
252 * after performing the write.
254 * Useful combinations of the flag bits are:
256 * SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE: ensures that all pages
257 * in the range which were dirty on entry to sys_sync_file_range() are placed
258 * under writeout. This is a start-write-for-data-integrity operation.
260 * SYNC_FILE_RANGE_WRITE: start writeout of all dirty pages in the range which
261 * are not presently under writeout. This is an asynchronous flush-to-disk
262 * operation. Not suitable for data integrity operations.
264 * SYNC_FILE_RANGE_WAIT_BEFORE (or SYNC_FILE_RANGE_WAIT_AFTER): wait for
265 * completion of writeout of all pages in the range. This will be used after an
266 * earlier SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE operation to wait
267 * for that operation to complete and to return the result.
269 * SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE|SYNC_FILE_RANGE_WAIT_AFTER:
270 * a traditional sync() operation. This is a write-for-data-integrity operation
271 * which will ensure that all pages in the range which were dirty on entry to
272 * sys_sync_file_range() are committed to disk.
275 * SYNC_FILE_RANGE_WAIT_BEFORE and SYNC_FILE_RANGE_WAIT_AFTER will detect any
276 * I/O errors or ENOSPC conditions and will return those to the caller, after
277 * clearing the EIO and ENOSPC flags in the address_space.
279 * It should be noted that none of these operations write out the file's
280 * metadata. So unless the application is strictly performing overwrites of
281 * already-instantiated disk blocks, there are no guarantees here that the data
282 * will be available after a crash.
284 SYSCALL_DEFINE4(sync_file_range
, int, fd
, loff_t
, offset
, loff_t
, nbytes
,
289 struct address_space
*mapping
;
290 loff_t endbyte
; /* inclusive */
294 if (flags
& ~VALID_FLAGS
)
297 endbyte
= offset
+ nbytes
;
301 if ((s64
)endbyte
< 0)
303 if (endbyte
< offset
)
306 if (sizeof(pgoff_t
) == 4) {
307 if (offset
>= (0x100000000ULL
<< PAGE_SHIFT
)) {
309 * The range starts outside a 32 bit machine's
310 * pagecache addressing capabilities. Let it "succeed"
315 if (endbyte
>= (0x100000000ULL
<< PAGE_SHIFT
)) {
326 endbyte
--; /* inclusive */
333 i_mode
= file_inode(f
.file
)->i_mode
;
335 if (!S_ISREG(i_mode
) && !S_ISBLK(i_mode
) && !S_ISDIR(i_mode
) &&
339 mapping
= f
.file
->f_mapping
;
341 if (flags
& SYNC_FILE_RANGE_WAIT_BEFORE
) {
342 ret
= file_fdatawait_range(f
.file
, offset
, endbyte
);
347 if (flags
& SYNC_FILE_RANGE_WRITE
) {
348 ret
= __filemap_fdatawrite_range(mapping
, offset
, endbyte
,
354 if (flags
& SYNC_FILE_RANGE_WAIT_AFTER
)
355 ret
= file_fdatawait_range(f
.file
, offset
, endbyte
);
363 /* It would be nice if people remember that not all the world's an i386
364 when they introduce new system calls */
365 SYSCALL_DEFINE4(sync_file_range2
, int, fd
, unsigned int, flags
,
366 loff_t
, offset
, loff_t
, nbytes
)
368 return sys_sync_file_range(fd
, offset
, nbytes
, flags
);