2 * High-level sync()-related operations
5 #include <linux/kernel.h>
6 #include <linux/file.h>
8 #include <linux/slab.h>
9 #include <linux/export.h>
10 #include <linux/namei.h>
11 #include <linux/sched.h>
12 #include <linux/writeback.h>
13 #include <linux/syscalls.h>
14 #include <linux/linkage.h>
15 #include <linux/pagemap.h>
16 #include <linux/quotaops.h>
17 #include <linux/backing-dev.h>
20 #define VALID_FLAGS (SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE| \
21 SYNC_FILE_RANGE_WAIT_AFTER)
24 * Do the filesystem syncing work. For simple filesystems
25 * writeback_inodes_sb(sb) just dirties buffers with inodes so we have to
26 * submit IO for these buffers via __sync_blockdev(). This also speeds up the
27 * wait == 1 case since in that case write_inode() functions do
28 * sync_dirty_buffer() and thus effectively write one block at a time.
30 static int __sync_filesystem(struct super_block *sb, int wait,
34 sync_inodes_sb(sb, start);
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)
51 unsigned long start = jiffies;
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.
62 if (sb->s_flags & MS_RDONLY)
65 ret = __sync_filesystem(sb, 0, start);
68 return __sync_filesystem(sb, 1, start);
70 EXPORT_SYMBOL_GPL(sync_filesystem);
72 static void sync_inodes_one_sb(struct super_block *sb, void *arg)
74 if (!(sb->s_flags & MS_RDONLY))
75 sync_inodes_sb(sb, *((unsigned long *)arg));
78 static void sync_fs_one_sb(struct super_block *sb, void *arg)
80 if (!(sb->s_flags & MS_RDONLY) && 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)
91 filemap_fdatawait(bdev->bd_inode->i_mapping);
95 * Sync everything. We start by waking flusher threads so that most of
96 * writeback runs on all devices in parallel. Then we sync all inodes reliably
97 * which effectively also waits for all flusher threads to finish doing
98 * writeback. At this point all data is on disk so metadata should be stable
99 * and we tell filesystems to sync their metadata via ->sync_fs() calls.
100 * Finally, we writeout all block devices because some filesystems (e.g. ext2)
101 * just write metadata (such as inodes or bitmaps) to block device page cache
102 * and do not sync it on their own in ->sync_fs().
104 SYSCALL_DEFINE0(sync)
106 int nowait = 0, wait = 1;
107 unsigned long start = jiffies;
109 wakeup_flusher_threads(0, WB_REASON_SYNC);
110 iterate_supers(sync_inodes_one_sb, &start);
111 iterate_supers(sync_fs_one_sb, &nowait);
112 iterate_supers(sync_fs_one_sb, &wait);
113 iterate_bdevs(fdatawrite_one_bdev, NULL);
114 iterate_bdevs(fdatawait_one_bdev, NULL);
115 if (unlikely(laptop_mode))
116 laptop_sync_completion();
120 static void do_sync_work(struct work_struct *work)
125 * Sync twice to reduce the possibility we skipped some inodes / pages
126 * because they were temporarily locked
128 iterate_supers(sync_inodes_one_sb, &nowait);
129 iterate_supers(sync_fs_one_sb, &nowait);
130 iterate_bdevs(fdatawrite_one_bdev, NULL);
131 iterate_supers(sync_inodes_one_sb, &nowait);
132 iterate_supers(sync_fs_one_sb, &nowait);
133 iterate_bdevs(fdatawrite_one_bdev, NULL);
134 printk("Emergency Sync complete\n");
138 void emergency_sync(void)
140 struct work_struct *work;
142 work = kmalloc(sizeof(*work), GFP_ATOMIC);
144 INIT_WORK(work, do_sync_work);
150 * sync a single super
152 SYSCALL_DEFINE1(syncfs, int, fd)
154 struct fd f = fdget(fd);
155 struct super_block *sb;
160 sb = f.file->f_dentry->d_sb;
162 down_read(&sb->s_umount);
163 ret = sync_filesystem(sb);
164 up_read(&sb->s_umount);
171 * vfs_fsync_range - helper to sync a range of data & metadata to disk
172 * @file: file to sync
173 * @start: offset in bytes of the beginning of data range to sync
174 * @end: offset in bytes of the end of data range (inclusive)
175 * @datasync: perform only datasync
177 * Write back data in range @start..@end and metadata for @file to disk. If
178 * @datasync is set only metadata needed to access modified file data is
181 int vfs_fsync_range(struct file *file, loff_t start, loff_t end, int datasync)
183 if (!file->f_op->fsync)
185 return file->f_op->fsync(file, start, end, datasync);
187 EXPORT_SYMBOL(vfs_fsync_range);
190 * vfs_fsync - perform a fsync or fdatasync on a file
191 * @file: file to sync
192 * @datasync: only perform a fdatasync operation
194 * Write back data and metadata for @file to disk. If @datasync is
195 * set only metadata needed to access modified file data is written.
197 int vfs_fsync(struct file *file, int datasync)
199 return vfs_fsync_range(file, 0, LLONG_MAX, datasync);
201 EXPORT_SYMBOL(vfs_fsync);
203 static int do_fsync(unsigned int fd, int datasync)
205 struct fd f = fdget(fd);
209 ret = vfs_fsync(f.file, datasync);
215 SYSCALL_DEFINE1(fsync, unsigned int, fd)
217 return do_fsync(fd, 0);
220 SYSCALL_DEFINE1(fdatasync, unsigned int, fd)
222 return do_fsync(fd, 1);
226 * sys_sync_file_range() permits finely controlled syncing over a segment of
227 * a file in the range offset .. (offset+nbytes-1) inclusive. If nbytes is
228 * zero then sys_sync_file_range() will operate from offset out to EOF.
232 * SYNC_FILE_RANGE_WAIT_BEFORE: wait upon writeout of all pages in the range
233 * before performing the write.
235 * SYNC_FILE_RANGE_WRITE: initiate writeout of all those dirty pages in the
236 * range which are not presently under writeback. Note that this may block for
237 * significant periods due to exhaustion of disk request structures.
239 * SYNC_FILE_RANGE_WAIT_AFTER: wait upon writeout of all pages in the range
240 * after performing the write.
242 * Useful combinations of the flag bits are:
244 * SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE: ensures that all pages
245 * in the range which were dirty on entry to sys_sync_file_range() are placed
246 * under writeout. This is a start-write-for-data-integrity operation.
248 * SYNC_FILE_RANGE_WRITE: start writeout of all dirty pages in the range which
249 * are not presently under writeout. This is an asynchronous flush-to-disk
250 * operation. Not suitable for data integrity operations.
252 * SYNC_FILE_RANGE_WAIT_BEFORE (or SYNC_FILE_RANGE_WAIT_AFTER): wait for
253 * completion of writeout of all pages in the range. This will be used after an
254 * earlier SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE operation to wait
255 * for that operation to complete and to return the result.
257 * SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE|SYNC_FILE_RANGE_WAIT_AFTER:
258 * a traditional sync() operation. This is a write-for-data-integrity operation
259 * which will ensure that all pages in the range which were dirty on entry to
260 * sys_sync_file_range() are committed to disk.
263 * SYNC_FILE_RANGE_WAIT_BEFORE and SYNC_FILE_RANGE_WAIT_AFTER will detect any
264 * I/O errors or ENOSPC conditions and will return those to the caller, after
265 * clearing the EIO and ENOSPC flags in the address_space.
267 * It should be noted that none of these operations write out the file's
268 * metadata. So unless the application is strictly performing overwrites of
269 * already-instantiated disk blocks, there are no guarantees here that the data
270 * will be available after a crash.
272 SYSCALL_DEFINE4(sync_file_range, int, fd, loff_t, offset, loff_t, nbytes,
277 struct address_space *mapping;
278 loff_t endbyte; /* inclusive */
282 if (flags & ~VALID_FLAGS)
285 endbyte = offset + nbytes;
289 if ((s64)endbyte < 0)
291 if (endbyte < offset)
294 if (sizeof(pgoff_t) == 4) {
295 if (offset >= (0x100000000ULL << PAGE_CACHE_SHIFT)) {
297 * The range starts outside a 32 bit machine's
298 * pagecache addressing capabilities. Let it "succeed"
303 if (endbyte >= (0x100000000ULL << PAGE_CACHE_SHIFT)) {
314 endbyte--; /* inclusive */
321 i_mode = file_inode(f.file)->i_mode;
323 if (!S_ISREG(i_mode) && !S_ISBLK(i_mode) && !S_ISDIR(i_mode) &&
327 mapping = f.file->f_mapping;
334 if (flags & SYNC_FILE_RANGE_WAIT_BEFORE) {
335 ret = filemap_fdatawait_range(mapping, offset, endbyte);
340 if (flags & SYNC_FILE_RANGE_WRITE) {
341 ret = filemap_fdatawrite_range(mapping, offset, endbyte);
346 if (flags & SYNC_FILE_RANGE_WAIT_AFTER)
347 ret = filemap_fdatawait_range(mapping, offset, endbyte);
355 /* It would be nice if people remember that not all the world's an i386
356 when they introduce new system calls */
357 SYSCALL_DEFINE4(sync_file_range2, int, fd, unsigned int, flags,
358 loff_t, offset, loff_t, nbytes)
360 return sys_sync_file_range(fd, offset, nbytes, flags);