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 static 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);
44 * Write out and wait upon all dirty data associated with this
45 * superblock. Filesystem data as well as the underlying block
46 * device. Takes the superblock lock.
48 int sync_filesystem(struct super_block *sb)
53 * We need to be protected against the filesystem going from
54 * r/o to r/w or vice versa.
56 WARN_ON(!rwsem_is_locked(&sb->s_umount));
59 * No point in syncing out anything if the filesystem is read-only.
64 ret = __sync_filesystem(sb, 0);
67 return __sync_filesystem(sb, 1);
69 EXPORT_SYMBOL(sync_filesystem);
71 static void sync_inodes_one_sb(struct super_block *sb, void *arg)
77 static void sync_fs_one_sb(struct super_block *sb, void *arg)
79 if (!sb_rdonly(sb) && sb->s_op->sync_fs)
80 sb->s_op->sync_fs(sb, *(int *)arg);
83 static void fdatawrite_one_bdev(struct block_device *bdev, void *arg)
85 filemap_fdatawrite(bdev->bd_inode->i_mapping);
88 static void fdatawait_one_bdev(struct block_device *bdev, void *arg)
91 * We keep the error status of individual mapping so that
92 * applications can catch the writeback error using fsync(2).
93 * See filemap_fdatawait_keep_errors() for details.
95 filemap_fdatawait_keep_errors(bdev->bd_inode->i_mapping);
99 * Sync everything. We start by waking flusher threads so that most of
100 * writeback runs on all devices in parallel. Then we sync all inodes reliably
101 * which effectively also waits for all flusher threads to finish doing
102 * writeback. At this point all data is on disk so metadata should be stable
103 * and we tell filesystems to sync their metadata via ->sync_fs() calls.
104 * Finally, we writeout all block devices because some filesystems (e.g. ext2)
105 * just write metadata (such as inodes or bitmaps) to block device page cache
106 * and do not sync it on their own in ->sync_fs().
110 int nowait = 0, wait = 1;
112 wakeup_flusher_threads(WB_REASON_SYNC);
113 iterate_supers(sync_inodes_one_sb, NULL);
114 iterate_supers(sync_fs_one_sb, &nowait);
115 iterate_supers(sync_fs_one_sb, &wait);
116 iterate_bdevs(fdatawrite_one_bdev, NULL);
117 iterate_bdevs(fdatawait_one_bdev, NULL);
118 if (unlikely(laptop_mode))
119 laptop_sync_completion();
122 SYSCALL_DEFINE0(sync)
128 static void do_sync_work(struct work_struct *work)
133 * Sync twice to reduce the possibility we skipped some inodes / pages
134 * because they were temporarily locked
136 iterate_supers(sync_inodes_one_sb, &nowait);
137 iterate_supers(sync_fs_one_sb, &nowait);
138 iterate_bdevs(fdatawrite_one_bdev, NULL);
139 iterate_supers(sync_inodes_one_sb, &nowait);
140 iterate_supers(sync_fs_one_sb, &nowait);
141 iterate_bdevs(fdatawrite_one_bdev, NULL);
142 printk("Emergency Sync complete\n");
146 void emergency_sync(void)
148 struct work_struct *work;
150 work = kmalloc(sizeof(*work), GFP_ATOMIC);
152 INIT_WORK(work, do_sync_work);
158 * sync a single super
160 SYSCALL_DEFINE1(syncfs, int, fd)
162 struct fd f = fdget(fd);
163 struct super_block *sb;
168 sb = f.file->f_path.dentry->d_sb;
170 down_read(&sb->s_umount);
171 ret = sync_filesystem(sb);
172 up_read(&sb->s_umount);
179 * vfs_fsync_range - helper to sync a range of data & metadata to disk
180 * @file: file to sync
181 * @start: offset in bytes of the beginning of data range to sync
182 * @end: offset in bytes of the end of data range (inclusive)
183 * @datasync: perform only datasync
185 * Write back data in range @start..@end and metadata for @file to disk. If
186 * @datasync is set only metadata needed to access modified file data is
189 int vfs_fsync_range(struct file *file, loff_t start, loff_t end, int datasync)
191 struct inode *inode = file->f_mapping->host;
193 if (!file->f_op->fsync)
195 if (!datasync && (inode->i_state & I_DIRTY_TIME)) {
196 spin_lock(&inode->i_lock);
197 inode->i_state &= ~I_DIRTY_TIME;
198 spin_unlock(&inode->i_lock);
199 mark_inode_dirty_sync(inode);
201 return file->f_op->fsync(file, start, end, datasync);
203 EXPORT_SYMBOL(vfs_fsync_range);
206 * vfs_fsync - perform a fsync or fdatasync on a file
207 * @file: file to sync
208 * @datasync: only perform a fdatasync operation
210 * Write back data and metadata for @file to disk. If @datasync is
211 * set only metadata needed to access modified file data is written.
213 int vfs_fsync(struct file *file, int datasync)
215 return vfs_fsync_range(file, 0, LLONG_MAX, datasync);
217 EXPORT_SYMBOL(vfs_fsync);
219 static int do_fsync(unsigned int fd, int datasync)
221 struct fd f = fdget(fd);
225 ret = vfs_fsync(f.file, datasync);
231 SYSCALL_DEFINE1(fsync, unsigned int, fd)
233 return do_fsync(fd, 0);
236 SYSCALL_DEFINE1(fdatasync, unsigned int, fd)
238 return do_fsync(fd, 1);
242 * sys_sync_file_range() permits finely controlled syncing over a segment of
243 * a file in the range offset .. (offset+nbytes-1) inclusive. If nbytes is
244 * zero then sys_sync_file_range() will operate from offset out to EOF.
248 * SYNC_FILE_RANGE_WAIT_BEFORE: wait upon writeout of all pages in the range
249 * before performing the write.
251 * SYNC_FILE_RANGE_WRITE: initiate writeout of all those dirty pages in the
252 * range which are not presently under writeback. Note that this may block for
253 * significant periods due to exhaustion of disk request structures.
255 * SYNC_FILE_RANGE_WAIT_AFTER: wait upon writeout of all pages in the range
256 * after performing the write.
258 * Useful combinations of the flag bits are:
260 * SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE: ensures that all pages
261 * in the range which were dirty on entry to sys_sync_file_range() are placed
262 * under writeout. This is a start-write-for-data-integrity operation.
264 * SYNC_FILE_RANGE_WRITE: start writeout of all dirty pages in the range which
265 * are not presently under writeout. This is an asynchronous flush-to-disk
266 * operation. Not suitable for data integrity operations.
268 * SYNC_FILE_RANGE_WAIT_BEFORE (or SYNC_FILE_RANGE_WAIT_AFTER): wait for
269 * completion of writeout of all pages in the range. This will be used after an
270 * earlier SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE operation to wait
271 * for that operation to complete and to return the result.
273 * SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE|SYNC_FILE_RANGE_WAIT_AFTER:
274 * a traditional sync() operation. This is a write-for-data-integrity operation
275 * which will ensure that all pages in the range which were dirty on entry to
276 * sys_sync_file_range() are committed to disk.
279 * SYNC_FILE_RANGE_WAIT_BEFORE and SYNC_FILE_RANGE_WAIT_AFTER will detect any
280 * I/O errors or ENOSPC conditions and will return those to the caller, after
281 * clearing the EIO and ENOSPC flags in the address_space.
283 * It should be noted that none of these operations write out the file's
284 * metadata. So unless the application is strictly performing overwrites of
285 * already-instantiated disk blocks, there are no guarantees here that the data
286 * will be available after a crash.
288 int ksys_sync_file_range(int fd, loff_t offset, loff_t nbytes,
293 struct address_space *mapping;
294 loff_t endbyte; /* inclusive */
298 if (flags & ~VALID_FLAGS)
301 endbyte = offset + nbytes;
305 if ((s64)endbyte < 0)
307 if (endbyte < offset)
310 if (sizeof(pgoff_t) == 4) {
311 if (offset >= (0x100000000ULL << PAGE_SHIFT)) {
313 * The range starts outside a 32 bit machine's
314 * pagecache addressing capabilities. Let it "succeed"
319 if (endbyte >= (0x100000000ULL << PAGE_SHIFT)) {
330 endbyte--; /* inclusive */
337 i_mode = file_inode(f.file)->i_mode;
339 if (!S_ISREG(i_mode) && !S_ISBLK(i_mode) && !S_ISDIR(i_mode) &&
343 mapping = f.file->f_mapping;
345 if (flags & SYNC_FILE_RANGE_WAIT_BEFORE) {
346 ret = file_fdatawait_range(f.file, offset, endbyte);
351 if (flags & SYNC_FILE_RANGE_WRITE) {
352 ret = __filemap_fdatawrite_range(mapping, offset, endbyte,
358 if (flags & SYNC_FILE_RANGE_WAIT_AFTER)
359 ret = file_fdatawait_range(f.file, offset, endbyte);
367 SYSCALL_DEFINE4(sync_file_range, int, fd, loff_t, offset, loff_t, nbytes,
370 return ksys_sync_file_range(fd, offset, nbytes, flags);
373 /* It would be nice if people remember that not all the world's an i386
374 when they introduce new system calls */
375 SYSCALL_DEFINE4(sync_file_range2, int, fd, unsigned int, flags,
376 loff_t, offset, loff_t, nbytes)
378 return ksys_sync_file_range(fd, offset, nbytes, flags);