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 * Write out and wait upon all dirty data associated with this
26 * superblock. Filesystem data as well as the underlying block
27 * device. Takes the superblock lock.
29 int sync_filesystem(struct super_block *sb)
34 * We need to be protected against the filesystem going from
35 * r/o to r/w or vice versa.
37 WARN_ON(!rwsem_is_locked(&sb->s_umount));
40 * No point in syncing out anything if the filesystem is read-only.
46 * Do the filesystem syncing work. For simple filesystems
47 * writeback_inodes_sb(sb) just dirties buffers with inodes so we have
48 * to submit I/O for these buffers via __sync_blockdev(). This also
49 * speeds up the wait == 1 case since in that case write_inode()
50 * methods call sync_dirty_buffer() and thus effectively write one block
53 writeback_inodes_sb(sb, WB_REASON_SYNC);
54 if (sb->s_op->sync_fs) {
55 ret = sb->s_op->sync_fs(sb, 0);
59 ret = __sync_blockdev(sb->s_bdev, 0);
64 if (sb->s_op->sync_fs) {
65 ret = sb->s_op->sync_fs(sb, 1);
69 return __sync_blockdev(sb->s_bdev, 1);
71 EXPORT_SYMBOL(sync_filesystem);
73 static void sync_inodes_one_sb(struct super_block *sb, void *arg)
79 static void sync_fs_one_sb(struct super_block *sb, void *arg)
81 if (!sb_rdonly(sb) && !(sb->s_iflags & SB_I_SKIP_SYNC) &&
83 sb->s_op->sync_fs(sb, *(int *)arg);
86 static void fdatawrite_one_bdev(struct block_device *bdev, void *arg)
88 filemap_fdatawrite(bdev->bd_inode->i_mapping);
91 static void fdatawait_one_bdev(struct block_device *bdev, void *arg)
94 * We keep the error status of individual mapping so that
95 * applications can catch the writeback error using fsync(2).
96 * See filemap_fdatawait_keep_errors() for details.
98 filemap_fdatawait_keep_errors(bdev->bd_inode->i_mapping);
102 * Sync everything. We start by waking flusher threads so that most of
103 * writeback runs on all devices in parallel. Then we sync all inodes reliably
104 * which effectively also waits for all flusher threads to finish doing
105 * writeback. At this point all data is on disk so metadata should be stable
106 * and we tell filesystems to sync their metadata via ->sync_fs() calls.
107 * Finally, we writeout all block devices because some filesystems (e.g. ext2)
108 * just write metadata (such as inodes or bitmaps) to block device page cache
109 * and do not sync it on their own in ->sync_fs().
113 int nowait = 0, wait = 1;
115 wakeup_flusher_threads(WB_REASON_SYNC);
116 iterate_supers(sync_inodes_one_sb, NULL);
117 iterate_supers(sync_fs_one_sb, &nowait);
118 iterate_supers(sync_fs_one_sb, &wait);
119 iterate_bdevs(fdatawrite_one_bdev, NULL);
120 iterate_bdevs(fdatawait_one_bdev, NULL);
121 if (unlikely(laptop_mode))
122 laptop_sync_completion();
125 SYSCALL_DEFINE0(sync)
131 static void do_sync_work(struct work_struct *work)
136 * Sync twice to reduce the possibility we skipped some inodes / pages
137 * because they were temporarily locked
139 iterate_supers(sync_inodes_one_sb, &nowait);
140 iterate_supers(sync_fs_one_sb, &nowait);
141 iterate_bdevs(fdatawrite_one_bdev, NULL);
142 iterate_supers(sync_inodes_one_sb, &nowait);
143 iterate_supers(sync_fs_one_sb, &nowait);
144 iterate_bdevs(fdatawrite_one_bdev, NULL);
145 printk("Emergency Sync complete\n");
149 void emergency_sync(void)
151 struct work_struct *work;
153 work = kmalloc(sizeof(*work), GFP_ATOMIC);
155 INIT_WORK(work, do_sync_work);
161 * sync a single super
163 SYSCALL_DEFINE1(syncfs, int, fd)
165 struct fd f = fdget(fd);
166 struct super_block *sb;
171 sb = f.file->f_path.dentry->d_sb;
173 down_read(&sb->s_umount);
174 ret = sync_filesystem(sb);
175 up_read(&sb->s_umount);
177 ret2 = errseq_check_and_advance(&sb->s_wb_err, &f.file->f_sb_err);
180 return ret ? ret : ret2;
184 * vfs_fsync_range - helper to sync a range of data & metadata to disk
185 * @file: file to sync
186 * @start: offset in bytes of the beginning of data range to sync
187 * @end: offset in bytes of the end of data range (inclusive)
188 * @datasync: perform only datasync
190 * Write back data in range @start..@end and metadata for @file to disk. If
191 * @datasync is set only metadata needed to access modified file data is
194 int vfs_fsync_range(struct file *file, loff_t start, loff_t end, int datasync)
196 struct inode *inode = file->f_mapping->host;
198 if (!file->f_op->fsync)
200 if (!datasync && (inode->i_state & I_DIRTY_TIME))
201 mark_inode_dirty_sync(inode);
202 return file->f_op->fsync(file, start, end, datasync);
204 EXPORT_SYMBOL(vfs_fsync_range);
207 * vfs_fsync - perform a fsync or fdatasync on a file
208 * @file: file to sync
209 * @datasync: only perform a fdatasync operation
211 * Write back data and metadata for @file to disk. If @datasync is
212 * set only metadata needed to access modified file data is written.
214 int vfs_fsync(struct file *file, int datasync)
216 return vfs_fsync_range(file, 0, LLONG_MAX, datasync);
218 EXPORT_SYMBOL(vfs_fsync);
220 static int do_fsync(unsigned int fd, int datasync)
222 struct fd f = fdget(fd);
226 ret = vfs_fsync(f.file, datasync);
232 SYSCALL_DEFINE1(fsync, unsigned int, fd)
234 return do_fsync(fd, 0);
237 SYSCALL_DEFINE1(fdatasync, unsigned int, fd)
239 return do_fsync(fd, 1);
242 int sync_file_range(struct file *file, loff_t offset, loff_t nbytes,
246 struct address_space *mapping;
247 loff_t endbyte; /* inclusive */
251 if (flags & ~VALID_FLAGS)
254 endbyte = offset + nbytes;
258 if ((s64)endbyte < 0)
260 if (endbyte < offset)
263 if (sizeof(pgoff_t) == 4) {
264 if (offset >= (0x100000000ULL << PAGE_SHIFT)) {
266 * The range starts outside a 32 bit machine's
267 * pagecache addressing capabilities. Let it "succeed"
272 if (endbyte >= (0x100000000ULL << PAGE_SHIFT)) {
283 endbyte--; /* inclusive */
285 i_mode = file_inode(file)->i_mode;
287 if (!S_ISREG(i_mode) && !S_ISBLK(i_mode) && !S_ISDIR(i_mode) &&
291 mapping = file->f_mapping;
293 if (flags & SYNC_FILE_RANGE_WAIT_BEFORE) {
294 ret = file_fdatawait_range(file, offset, endbyte);
299 if (flags & SYNC_FILE_RANGE_WRITE) {
300 int sync_mode = WB_SYNC_NONE;
302 if ((flags & SYNC_FILE_RANGE_WRITE_AND_WAIT) ==
303 SYNC_FILE_RANGE_WRITE_AND_WAIT)
304 sync_mode = WB_SYNC_ALL;
306 ret = __filemap_fdatawrite_range(mapping, offset, endbyte,
312 if (flags & SYNC_FILE_RANGE_WAIT_AFTER)
313 ret = file_fdatawait_range(file, offset, endbyte);
320 * ksys_sync_file_range() permits finely controlled syncing over a segment of
321 * a file in the range offset .. (offset+nbytes-1) inclusive. If nbytes is
322 * zero then ksys_sync_file_range() will operate from offset out to EOF.
326 * SYNC_FILE_RANGE_WAIT_BEFORE: wait upon writeout of all pages in the range
327 * before performing the write.
329 * SYNC_FILE_RANGE_WRITE: initiate writeout of all those dirty pages in the
330 * range which are not presently under writeback. Note that this may block for
331 * significant periods due to exhaustion of disk request structures.
333 * SYNC_FILE_RANGE_WAIT_AFTER: wait upon writeout of all pages in the range
334 * after performing the write.
336 * Useful combinations of the flag bits are:
338 * SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE: ensures that all pages
339 * in the range which were dirty on entry to ksys_sync_file_range() are placed
340 * under writeout. This is a start-write-for-data-integrity operation.
342 * SYNC_FILE_RANGE_WRITE: start writeout of all dirty pages in the range which
343 * are not presently under writeout. This is an asynchronous flush-to-disk
344 * operation. Not suitable for data integrity operations.
346 * SYNC_FILE_RANGE_WAIT_BEFORE (or SYNC_FILE_RANGE_WAIT_AFTER): wait for
347 * completion of writeout of all pages in the range. This will be used after an
348 * earlier SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE operation to wait
349 * for that operation to complete and to return the result.
351 * SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE|SYNC_FILE_RANGE_WAIT_AFTER
352 * (a.k.a. SYNC_FILE_RANGE_WRITE_AND_WAIT):
353 * a traditional sync() operation. This is a write-for-data-integrity operation
354 * which will ensure that all pages in the range which were dirty on entry to
355 * ksys_sync_file_range() are written to disk. It should be noted that disk
356 * caches are not flushed by this call, so there are no guarantees here that the
357 * data will be available on disk after a crash.
360 * SYNC_FILE_RANGE_WAIT_BEFORE and SYNC_FILE_RANGE_WAIT_AFTER will detect any
361 * I/O errors or ENOSPC conditions and will return those to the caller, after
362 * clearing the EIO and ENOSPC flags in the address_space.
364 * It should be noted that none of these operations write out the file's
365 * metadata. So unless the application is strictly performing overwrites of
366 * already-instantiated disk blocks, there are no guarantees here that the data
367 * will be available after a crash.
369 int ksys_sync_file_range(int fd, loff_t offset, loff_t nbytes,
378 ret = sync_file_range(f.file, offset, nbytes, flags);
384 SYSCALL_DEFINE4(sync_file_range, int, fd, loff_t, offset, loff_t, nbytes,
387 return ksys_sync_file_range(fd, offset, nbytes, flags);
390 /* It would be nice if people remember that not all the world's an i386
391 when they introduce new system calls */
392 SYSCALL_DEFINE4(sync_file_range2, int, fd, unsigned int, flags,
393 loff_t, offset, loff_t, nbytes)
395 return ksys_sync_file_range(fd, offset, nbytes, flags);