2 * High-level sync()-related operations
5 #include <linux/kernel.h>
6 #include <linux/file.h>
8 #include <linux/slab.h>
9 #include <linux/module.h>
10 #include <linux/sched.h>
11 #include <linux/writeback.h>
12 #include <linux/syscalls.h>
13 #include <linux/linkage.h>
14 #include <linux/pagemap.h>
15 #include <linux/quotaops.h>
16 #include <linux/buffer_head.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)
33 * This should be safe, as we require bdi backing to actually
34 * write out data in the first place
36 if (!sb->s_bdi || sb->s_bdi == &noop_backing_dev_info)
39 if (sb->s_qcop && sb->s_qcop->quota_sync)
40 sb->s_qcop->quota_sync(sb, -1, wait);
45 writeback_inodes_sb(sb);
47 if (sb->s_op->sync_fs)
48 sb->s_op->sync_fs(sb, wait);
49 return __sync_blockdev(sb->s_bdev, wait);
53 * Write out and wait upon all dirty data associated with this
54 * superblock. Filesystem data as well as the underlying block
55 * device. Takes the superblock lock.
57 int sync_filesystem(struct super_block *sb)
62 * We need to be protected against the filesystem going from
63 * r/o to r/w or vice versa.
65 WARN_ON(!rwsem_is_locked(&sb->s_umount));
68 * No point in syncing out anything if the filesystem is read-only.
70 if (sb->s_flags & MS_RDONLY)
73 ret = __sync_filesystem(sb, 0);
76 return __sync_filesystem(sb, 1);
78 EXPORT_SYMBOL_GPL(sync_filesystem);
80 static void sync_one_sb(struct super_block *sb, void *arg)
82 if (!(sb->s_flags & MS_RDONLY) && sb->s_bdi)
83 __sync_filesystem(sb, *(int *)arg);
86 * Sync all the data for all the filesystems (called by sys_sync() and
89 static void sync_filesystems(int wait)
91 iterate_supers(sync_one_sb, &wait);
95 * sync everything. Start out by waking pdflush, because that writes back
96 * all queues in parallel.
100 wakeup_flusher_threads(0);
103 if (unlikely(laptop_mode))
104 laptop_sync_completion();
108 static void do_sync_work(struct work_struct *work)
111 * Sync twice to reduce the possibility we skipped some inodes / pages
112 * because they were temporarily locked
116 printk("Emergency Sync complete\n");
120 void emergency_sync(void)
122 struct work_struct *work;
124 work = kmalloc(sizeof(*work), GFP_ATOMIC);
126 INIT_WORK(work, do_sync_work);
132 * Generic function to fsync a file.
134 int file_fsync(struct file *filp, int datasync)
136 struct inode *inode = filp->f_mapping->host;
137 struct super_block * sb;
140 /* sync the inode to buffers */
141 ret = write_inode_now(inode, 0);
143 /* sync the superblock to buffers */
145 if (sb->s_dirt && sb->s_op->write_super)
146 sb->s_op->write_super(sb);
148 /* .. finally sync the buffers to disk */
149 err = sync_blockdev(sb->s_bdev);
154 EXPORT_SYMBOL(file_fsync);
157 * vfs_fsync_range - helper to sync a range of data & metadata to disk
158 * @file: file to sync
159 * @start: offset in bytes of the beginning of data range to sync
160 * @end: offset in bytes of the end of data range (inclusive)
161 * @datasync: perform only datasync
163 * Write back data in range @start..@end and metadata for @file to disk. If
164 * @datasync is set only metadata needed to access modified file data is
167 int vfs_fsync_range(struct file *file, loff_t start, loff_t end, int datasync)
169 struct address_space *mapping = file->f_mapping;
172 if (!file->f_op || !file->f_op->fsync) {
177 ret = filemap_write_and_wait_range(mapping, start, end);
180 * We need to protect against concurrent writers, which could cause
181 * livelocks in fsync_buffers_list().
183 mutex_lock(&mapping->host->i_mutex);
184 err = file->f_op->fsync(file, datasync);
187 mutex_unlock(&mapping->host->i_mutex);
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)
215 ret = vfs_fsync(file, datasync);
221 SYSCALL_DEFINE1(fsync, unsigned int, fd)
223 return do_fsync(fd, 0);
226 SYSCALL_DEFINE1(fdatasync, unsigned int, fd)
228 return do_fsync(fd, 1);
232 * generic_write_sync - perform syncing after a write if file / inode is sync
233 * @file: file to which the write happened
234 * @pos: offset where the write started
235 * @count: length of the write
237 * This is just a simple wrapper about our general syncing function.
239 int generic_write_sync(struct file *file, loff_t pos, loff_t count)
241 if (!(file->f_flags & O_DSYNC) && !IS_SYNC(file->f_mapping->host))
243 return vfs_fsync_range(file, pos, pos + count - 1,
244 (file->f_flags & __O_SYNC) ? 0 : 1);
246 EXPORT_SYMBOL(generic_write_sync);
249 * sys_sync_file_range() permits finely controlled syncing over a segment of
250 * a file in the range offset .. (offset+nbytes-1) inclusive. If nbytes is
251 * zero then sys_sync_file_range() will operate from offset out to EOF.
255 * SYNC_FILE_RANGE_WAIT_BEFORE: wait upon writeout of all pages in the range
256 * before performing the write.
258 * SYNC_FILE_RANGE_WRITE: initiate writeout of all those dirty pages in the
259 * range which are not presently under writeback. Note that this may block for
260 * significant periods due to exhaustion of disk request structures.
262 * SYNC_FILE_RANGE_WAIT_AFTER: wait upon writeout of all pages in the range
263 * after performing the write.
265 * Useful combinations of the flag bits are:
267 * SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE: ensures that all pages
268 * in the range which were dirty on entry to sys_sync_file_range() are placed
269 * under writeout. This is a start-write-for-data-integrity operation.
271 * SYNC_FILE_RANGE_WRITE: start writeout of all dirty pages in the range which
272 * are not presently under writeout. This is an asynchronous flush-to-disk
273 * operation. Not suitable for data integrity operations.
275 * SYNC_FILE_RANGE_WAIT_BEFORE (or SYNC_FILE_RANGE_WAIT_AFTER): wait for
276 * completion of writeout of all pages in the range. This will be used after an
277 * earlier SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE operation to wait
278 * for that operation to complete and to return the result.
280 * SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE|SYNC_FILE_RANGE_WAIT_AFTER:
281 * a traditional sync() operation. This is a write-for-data-integrity operation
282 * which will ensure that all pages in the range which were dirty on entry to
283 * sys_sync_file_range() are committed to disk.
286 * SYNC_FILE_RANGE_WAIT_BEFORE and SYNC_FILE_RANGE_WAIT_AFTER will detect any
287 * I/O errors or ENOSPC conditions and will return those to the caller, after
288 * clearing the EIO and ENOSPC flags in the address_space.
290 * It should be noted that none of these operations write out the file's
291 * metadata. So unless the application is strictly performing overwrites of
292 * already-instantiated disk blocks, there are no guarantees here that the data
293 * will be available after a crash.
295 SYSCALL_DEFINE(sync_file_range)(int fd, loff_t offset, loff_t nbytes,
300 struct address_space *mapping;
301 loff_t endbyte; /* inclusive */
306 if (flags & ~VALID_FLAGS)
309 endbyte = offset + nbytes;
313 if ((s64)endbyte < 0)
315 if (endbyte < offset)
318 if (sizeof(pgoff_t) == 4) {
319 if (offset >= (0x100000000ULL << PAGE_CACHE_SHIFT)) {
321 * The range starts outside a 32 bit machine's
322 * pagecache addressing capabilities. Let it "succeed"
327 if (endbyte >= (0x100000000ULL << PAGE_CACHE_SHIFT)) {
338 endbyte--; /* inclusive */
341 file = fget_light(fd, &fput_needed);
345 i_mode = file->f_path.dentry->d_inode->i_mode;
347 if (!S_ISREG(i_mode) && !S_ISBLK(i_mode) && !S_ISDIR(i_mode) &&
351 mapping = file->f_mapping;
358 if (flags & SYNC_FILE_RANGE_WAIT_BEFORE) {
359 ret = filemap_fdatawait_range(mapping, offset, endbyte);
364 if (flags & SYNC_FILE_RANGE_WRITE) {
365 ret = filemap_fdatawrite_range(mapping, offset, endbyte);
370 if (flags & SYNC_FILE_RANGE_WAIT_AFTER)
371 ret = filemap_fdatawait_range(mapping, offset, endbyte);
374 fput_light(file, fput_needed);
378 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
379 asmlinkage long SyS_sync_file_range(long fd, loff_t offset, loff_t nbytes,
382 return SYSC_sync_file_range((int) fd, offset, nbytes,
383 (unsigned int) flags);
385 SYSCALL_ALIAS(sys_sync_file_range, SyS_sync_file_range);
388 /* It would be nice if people remember that not all the world's an i386
389 when they introduce new system calls */
390 SYSCALL_DEFINE(sync_file_range2)(int fd, unsigned int flags,
391 loff_t offset, loff_t nbytes)
393 return sys_sync_file_range(fd, offset, nbytes, flags);
395 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
396 asmlinkage long SyS_sync_file_range2(long fd, long flags,
397 loff_t offset, loff_t nbytes)
399 return SYSC_sync_file_range2((int) fd, (unsigned int) flags,
402 SYSCALL_ALIAS(sys_sync_file_range2, SyS_sync_file_range2);