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_iflags
& SB_I_SKIP_SYNC
) &&
82 sb
->s_op
->sync_fs(sb
, *(int *)arg
);
86 * Sync everything. We start by waking flusher threads so that most of
87 * writeback runs on all devices in parallel. Then we sync all inodes reliably
88 * which effectively also waits for all flusher threads to finish doing
89 * writeback. At this point all data is on disk so metadata should be stable
90 * and we tell filesystems to sync their metadata via ->sync_fs() calls.
91 * Finally, we writeout all block devices because some filesystems (e.g. ext2)
92 * just write metadata (such as inodes or bitmaps) to block device page cache
93 * and do not sync it on their own in ->sync_fs().
97 int nowait
= 0, wait
= 1;
99 wakeup_flusher_threads(WB_REASON_SYNC
);
100 iterate_supers(sync_inodes_one_sb
, NULL
);
101 iterate_supers(sync_fs_one_sb
, &nowait
);
102 iterate_supers(sync_fs_one_sb
, &wait
);
105 if (unlikely(laptop_mode
))
106 laptop_sync_completion();
109 SYSCALL_DEFINE0(sync
)
115 static void do_sync_work(struct work_struct
*work
)
120 * Sync twice to reduce the possibility we skipped some inodes / pages
121 * because they were temporarily locked
123 iterate_supers(sync_inodes_one_sb
, &nowait
);
124 iterate_supers(sync_fs_one_sb
, &nowait
);
126 iterate_supers(sync_inodes_one_sb
, &nowait
);
127 iterate_supers(sync_fs_one_sb
, &nowait
);
129 printk("Emergency Sync complete\n");
133 void emergency_sync(void)
135 struct work_struct
*work
;
137 work
= kmalloc(sizeof(*work
), GFP_ATOMIC
);
139 INIT_WORK(work
, do_sync_work
);
145 * sync a single super
147 SYSCALL_DEFINE1(syncfs
, int, fd
)
149 struct fd f
= fdget(fd
);
150 struct super_block
*sb
;
155 sb
= f
.file
->f_path
.dentry
->d_sb
;
157 down_read(&sb
->s_umount
);
158 ret
= sync_filesystem(sb
);
159 up_read(&sb
->s_umount
);
161 ret2
= errseq_check_and_advance(&sb
->s_wb_err
, &f
.file
->f_sb_err
);
164 return ret
? ret
: ret2
;
168 * vfs_fsync_range - helper to sync a range of data & metadata to disk
169 * @file: file to sync
170 * @start: offset in bytes of the beginning of data range to sync
171 * @end: offset in bytes of the end of data range (inclusive)
172 * @datasync: perform only datasync
174 * Write back data in range @start..@end and metadata for @file to disk. If
175 * @datasync is set only metadata needed to access modified file data is
178 int vfs_fsync_range(struct file
*file
, loff_t start
, loff_t end
, int datasync
)
180 struct inode
*inode
= file
->f_mapping
->host
;
182 if (!file
->f_op
->fsync
)
184 if (!datasync
&& (inode
->i_state
& I_DIRTY_TIME
))
185 mark_inode_dirty_sync(inode
);
186 return file
->f_op
->fsync(file
, start
, end
, datasync
);
188 EXPORT_SYMBOL(vfs_fsync_range
);
191 * vfs_fsync - perform a fsync or fdatasync on a file
192 * @file: file to sync
193 * @datasync: only perform a fdatasync operation
195 * Write back data and metadata for @file to disk. If @datasync is
196 * set only metadata needed to access modified file data is written.
198 int vfs_fsync(struct file
*file
, int datasync
)
200 return vfs_fsync_range(file
, 0, LLONG_MAX
, datasync
);
202 EXPORT_SYMBOL(vfs_fsync
);
204 static int do_fsync(unsigned int fd
, int datasync
)
206 struct fd f
= fdget(fd
);
210 ret
= vfs_fsync(f
.file
, datasync
);
216 SYSCALL_DEFINE1(fsync
, unsigned int, fd
)
218 return do_fsync(fd
, 0);
221 SYSCALL_DEFINE1(fdatasync
, unsigned int, fd
)
223 return do_fsync(fd
, 1);
226 int sync_file_range(struct file
*file
, loff_t offset
, loff_t nbytes
,
230 struct address_space
*mapping
;
231 loff_t endbyte
; /* inclusive */
235 if (flags
& ~VALID_FLAGS
)
238 endbyte
= offset
+ nbytes
;
242 if ((s64
)endbyte
< 0)
244 if (endbyte
< offset
)
247 if (sizeof(pgoff_t
) == 4) {
248 if (offset
>= (0x100000000ULL
<< PAGE_SHIFT
)) {
250 * The range starts outside a 32 bit machine's
251 * pagecache addressing capabilities. Let it "succeed"
256 if (endbyte
>= (0x100000000ULL
<< PAGE_SHIFT
)) {
267 endbyte
--; /* inclusive */
269 i_mode
= file_inode(file
)->i_mode
;
271 if (!S_ISREG(i_mode
) && !S_ISBLK(i_mode
) && !S_ISDIR(i_mode
) &&
275 mapping
= file
->f_mapping
;
277 if (flags
& SYNC_FILE_RANGE_WAIT_BEFORE
) {
278 ret
= file_fdatawait_range(file
, offset
, endbyte
);
283 if (flags
& SYNC_FILE_RANGE_WRITE
) {
284 int sync_mode
= WB_SYNC_NONE
;
286 if ((flags
& SYNC_FILE_RANGE_WRITE_AND_WAIT
) ==
287 SYNC_FILE_RANGE_WRITE_AND_WAIT
)
288 sync_mode
= WB_SYNC_ALL
;
290 ret
= __filemap_fdatawrite_range(mapping
, offset
, endbyte
,
296 if (flags
& SYNC_FILE_RANGE_WAIT_AFTER
)
297 ret
= file_fdatawait_range(file
, offset
, endbyte
);
304 * ksys_sync_file_range() permits finely controlled syncing over a segment of
305 * a file in the range offset .. (offset+nbytes-1) inclusive. If nbytes is
306 * zero then ksys_sync_file_range() will operate from offset out to EOF.
310 * SYNC_FILE_RANGE_WAIT_BEFORE: wait upon writeout of all pages in the range
311 * before performing the write.
313 * SYNC_FILE_RANGE_WRITE: initiate writeout of all those dirty pages in the
314 * range which are not presently under writeback. Note that this may block for
315 * significant periods due to exhaustion of disk request structures.
317 * SYNC_FILE_RANGE_WAIT_AFTER: wait upon writeout of all pages in the range
318 * after performing the write.
320 * Useful combinations of the flag bits are:
322 * SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE: ensures that all pages
323 * in the range which were dirty on entry to ksys_sync_file_range() are placed
324 * under writeout. This is a start-write-for-data-integrity operation.
326 * SYNC_FILE_RANGE_WRITE: start writeout of all dirty pages in the range which
327 * are not presently under writeout. This is an asynchronous flush-to-disk
328 * operation. Not suitable for data integrity operations.
330 * SYNC_FILE_RANGE_WAIT_BEFORE (or SYNC_FILE_RANGE_WAIT_AFTER): wait for
331 * completion of writeout of all pages in the range. This will be used after an
332 * earlier SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE operation to wait
333 * for that operation to complete and to return the result.
335 * SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE|SYNC_FILE_RANGE_WAIT_AFTER
336 * (a.k.a. SYNC_FILE_RANGE_WRITE_AND_WAIT):
337 * a traditional sync() operation. This is a write-for-data-integrity operation
338 * which will ensure that all pages in the range which were dirty on entry to
339 * ksys_sync_file_range() are written to disk. It should be noted that disk
340 * caches are not flushed by this call, so there are no guarantees here that the
341 * data will be available on disk after a crash.
344 * SYNC_FILE_RANGE_WAIT_BEFORE and SYNC_FILE_RANGE_WAIT_AFTER will detect any
345 * I/O errors or ENOSPC conditions and will return those to the caller, after
346 * clearing the EIO and ENOSPC flags in the address_space.
348 * It should be noted that none of these operations write out the file's
349 * metadata. So unless the application is strictly performing overwrites of
350 * already-instantiated disk blocks, there are no guarantees here that the data
351 * will be available after a crash.
353 int ksys_sync_file_range(int fd
, loff_t offset
, loff_t nbytes
,
362 ret
= sync_file_range(f
.file
, offset
, nbytes
, flags
);
368 SYSCALL_DEFINE4(sync_file_range
, int, fd
, loff_t
, offset
, loff_t
, nbytes
,
371 return ksys_sync_file_range(fd
, offset
, nbytes
, flags
);
374 /* It would be nice if people remember that not all the world's an i386
375 when they introduce new system calls */
376 SYSCALL_DEFINE4(sync_file_range2
, int, fd
, unsigned int, flags
,
377 loff_t
, offset
, loff_t
, nbytes
)
379 return ksys_sync_file_range(fd
, offset
, nbytes
, flags
);