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)
35 writeback_inodes_sb(sb, WB_REASON_SYNC);
37 if (sb->s_op->sync_fs)
38 sb->s_op->sync_fs(sb, wait);
39 return __sync_blockdev(sb->s_bdev, wait);
43 * Write out and wait upon all dirty data associated with this
44 * superblock. Filesystem data as well as the underlying block
45 * device. Takes the superblock lock.
47 int sync_filesystem(struct super_block *sb)
52 * We need to be protected against the filesystem going from
53 * r/o to r/w or vice versa.
55 WARN_ON(!rwsem_is_locked(&sb->s_umount));
58 * No point in syncing out anything if the filesystem is read-only.
60 if (sb->s_flags & MS_RDONLY)
63 ret = __sync_filesystem(sb, 0);
66 return __sync_filesystem(sb, 1);
68 EXPORT_SYMBOL(sync_filesystem);
70 static void sync_inodes_one_sb(struct super_block *sb, void *arg)
72 if (!(sb->s_flags & MS_RDONLY))
76 static void sync_fs_one_sb(struct super_block *sb, void *arg)
78 if (!(sb->s_flags & MS_RDONLY) && sb->s_op->sync_fs)
79 sb->s_op->sync_fs(sb, *(int *)arg);
82 static void fdatawrite_one_bdev(struct block_device *bdev, void *arg)
84 filemap_fdatawrite(bdev->bd_inode->i_mapping);
87 static void fdatawait_one_bdev(struct block_device *bdev, void *arg)
89 filemap_fdatawait(bdev->bd_inode->i_mapping);
93 * Sync everything. We start by waking flusher threads so that most of
94 * writeback runs on all devices in parallel. Then we sync all inodes reliably
95 * which effectively also waits for all flusher threads to finish doing
96 * writeback. At this point all data is on disk so metadata should be stable
97 * and we tell filesystems to sync their metadata via ->sync_fs() calls.
98 * Finally, we writeout all block devices because some filesystems (e.g. ext2)
99 * just write metadata (such as inodes or bitmaps) to block device page cache
100 * and do not sync it on their own in ->sync_fs().
102 SYSCALL_DEFINE0(sync)
104 int nowait = 0, wait = 1;
106 wakeup_flusher_threads(0, WB_REASON_SYNC);
107 iterate_supers(sync_inodes_one_sb, NULL);
108 iterate_supers(sync_fs_one_sb, &nowait);
109 iterate_supers(sync_fs_one_sb, &wait);
110 iterate_bdevs(fdatawrite_one_bdev, NULL);
111 iterate_bdevs(fdatawait_one_bdev, NULL);
112 if (unlikely(laptop_mode))
113 laptop_sync_completion();
117 static void do_sync_work(struct work_struct *work)
122 * Sync twice to reduce the possibility we skipped some inodes / pages
123 * because they were temporarily locked
125 iterate_supers(sync_inodes_one_sb, &nowait);
126 iterate_supers(sync_fs_one_sb, &nowait);
127 iterate_bdevs(fdatawrite_one_bdev, NULL);
128 iterate_supers(sync_inodes_one_sb, &nowait);
129 iterate_supers(sync_fs_one_sb, &nowait);
130 iterate_bdevs(fdatawrite_one_bdev, NULL);
131 printk("Emergency Sync complete\n");
135 void emergency_sync(void)
137 struct work_struct *work;
139 work = kmalloc(sizeof(*work), GFP_ATOMIC);
141 INIT_WORK(work, do_sync_work);
147 * sync a single super
149 SYSCALL_DEFINE1(syncfs, int, fd)
151 struct fd f = fdget(fd);
152 struct super_block *sb;
157 sb = f.file->f_path.dentry->d_sb;
159 down_read(&sb->s_umount);
160 ret = sync_filesystem(sb);
161 up_read(&sb->s_umount);
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 spin_lock(&inode->i_lock);
186 inode->i_state &= ~I_DIRTY_TIME;
187 spin_unlock(&inode->i_lock);
188 mark_inode_dirty_sync(inode);
190 return file->f_op->fsync(file, start, end, datasync);
192 EXPORT_SYMBOL(vfs_fsync_range);
195 * vfs_fsync - perform a fsync or fdatasync on a file
196 * @file: file to sync
197 * @datasync: only perform a fdatasync operation
199 * Write back data and metadata for @file to disk. If @datasync is
200 * set only metadata needed to access modified file data is written.
202 int vfs_fsync(struct file *file, int datasync)
204 return vfs_fsync_range(file, 0, LLONG_MAX, datasync);
206 EXPORT_SYMBOL(vfs_fsync);
208 static int do_fsync(unsigned int fd, int datasync)
210 struct fd f = fdget(fd);
214 ret = vfs_fsync(f.file, datasync);
220 SYSCALL_DEFINE1(fsync, unsigned int, fd)
222 return do_fsync(fd, 0);
225 SYSCALL_DEFINE1(fdatasync, unsigned int, fd)
227 return do_fsync(fd, 1);
231 * sys_sync_file_range() permits finely controlled syncing over a segment of
232 * a file in the range offset .. (offset+nbytes-1) inclusive. If nbytes is
233 * zero then sys_sync_file_range() will operate from offset out to EOF.
237 * SYNC_FILE_RANGE_WAIT_BEFORE: wait upon writeout of all pages in the range
238 * before performing the write.
240 * SYNC_FILE_RANGE_WRITE: initiate writeout of all those dirty pages in the
241 * range which are not presently under writeback. Note that this may block for
242 * significant periods due to exhaustion of disk request structures.
244 * SYNC_FILE_RANGE_WAIT_AFTER: wait upon writeout of all pages in the range
245 * after performing the write.
247 * Useful combinations of the flag bits are:
249 * SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE: ensures that all pages
250 * in the range which were dirty on entry to sys_sync_file_range() are placed
251 * under writeout. This is a start-write-for-data-integrity operation.
253 * SYNC_FILE_RANGE_WRITE: start writeout of all dirty pages in the range which
254 * are not presently under writeout. This is an asynchronous flush-to-disk
255 * operation. Not suitable for data integrity operations.
257 * SYNC_FILE_RANGE_WAIT_BEFORE (or SYNC_FILE_RANGE_WAIT_AFTER): wait for
258 * completion of writeout of all pages in the range. This will be used after an
259 * earlier SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE operation to wait
260 * for that operation to complete and to return the result.
262 * SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE|SYNC_FILE_RANGE_WAIT_AFTER:
263 * a traditional sync() operation. This is a write-for-data-integrity operation
264 * which will ensure that all pages in the range which were dirty on entry to
265 * sys_sync_file_range() are committed to disk.
268 * SYNC_FILE_RANGE_WAIT_BEFORE and SYNC_FILE_RANGE_WAIT_AFTER will detect any
269 * I/O errors or ENOSPC conditions and will return those to the caller, after
270 * clearing the EIO and ENOSPC flags in the address_space.
272 * It should be noted that none of these operations write out the file's
273 * metadata. So unless the application is strictly performing overwrites of
274 * already-instantiated disk blocks, there are no guarantees here that the data
275 * will be available after a crash.
277 SYSCALL_DEFINE4(sync_file_range, int, fd, loff_t, offset, loff_t, nbytes,
282 struct address_space *mapping;
283 loff_t endbyte; /* inclusive */
287 if (flags & ~VALID_FLAGS)
290 endbyte = offset + nbytes;
294 if ((s64)endbyte < 0)
296 if (endbyte < offset)
299 if (sizeof(pgoff_t) == 4) {
300 if (offset >= (0x100000000ULL << PAGE_CACHE_SHIFT)) {
302 * The range starts outside a 32 bit machine's
303 * pagecache addressing capabilities. Let it "succeed"
308 if (endbyte >= (0x100000000ULL << PAGE_CACHE_SHIFT)) {
319 endbyte--; /* inclusive */
326 i_mode = file_inode(f.file)->i_mode;
328 if (!S_ISREG(i_mode) && !S_ISBLK(i_mode) && !S_ISDIR(i_mode) &&
332 mapping = f.file->f_mapping;
339 if (flags & SYNC_FILE_RANGE_WAIT_BEFORE) {
340 ret = filemap_fdatawait_range(mapping, offset, endbyte);
345 if (flags & SYNC_FILE_RANGE_WRITE) {
346 ret = filemap_fdatawrite_range(mapping, offset, endbyte);
351 if (flags & SYNC_FILE_RANGE_WAIT_AFTER)
352 ret = filemap_fdatawait_range(mapping, offset, endbyte);
360 /* It would be nice if people remember that not all the world's an i386
361 when they introduce new system calls */
362 SYSCALL_DEFINE4(sync_file_range2, int, fd, unsigned int, flags,
363 loff_t, offset, loff_t, nbytes)
365 return sys_sync_file_range(fd, offset, nbytes, flags);