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/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/buffer_head.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)
34 * This should be safe, as we require bdi backing to actually
35 * write out data in the first place
37 if (sb->s_bdi == &noop_backing_dev_info)
40 if (sb->s_qcop && sb->s_qcop->quota_sync)
41 sb->s_qcop->quota_sync(sb, -1, wait);
46 writeback_inodes_sb(sb);
48 if (sb->s_op->sync_fs)
49 sb->s_op->sync_fs(sb, wait);
50 return __sync_blockdev(sb->s_bdev, wait);
54 * Write out and wait upon all dirty data associated with this
55 * superblock. Filesystem data as well as the underlying block
56 * device. Takes the superblock lock.
58 int sync_filesystem(struct super_block *sb)
63 * We need to be protected against the filesystem going from
64 * r/o to r/w or vice versa.
66 WARN_ON(!rwsem_is_locked(&sb->s_umount));
69 * No point in syncing out anything if the filesystem is read-only.
71 if (sb->s_flags & MS_RDONLY)
74 ret = __sync_filesystem(sb, 0);
77 return __sync_filesystem(sb, 1);
79 EXPORT_SYMBOL_GPL(sync_filesystem);
81 static void sync_one_sb(struct super_block *sb, void *arg)
83 if (!(sb->s_flags & MS_RDONLY))
84 __sync_filesystem(sb, *(int *)arg);
87 * Sync all the data for all the filesystems (called by sys_sync() and
90 static void sync_filesystems(int wait)
92 iterate_supers(sync_one_sb, &wait);
96 * sync everything. Start out by waking pdflush, because that writes back
97 * all queues in parallel.
101 wakeup_flusher_threads(0);
104 if (unlikely(laptop_mode))
105 laptop_sync_completion();
109 static void do_sync_work(struct work_struct *work)
112 * Sync twice to reduce the possibility we skipped some inodes / pages
113 * because they were temporarily locked
117 printk("Emergency Sync complete\n");
121 void emergency_sync(void)
123 struct work_struct *work;
125 work = kmalloc(sizeof(*work), GFP_ATOMIC);
127 INIT_WORK(work, do_sync_work);
133 * sync a single super
135 SYSCALL_DEFINE1(syncfs, int, fd)
138 struct super_block *sb;
142 file = fget_light(fd, &fput_needed);
145 sb = file->f_dentry->d_sb;
147 down_read(&sb->s_umount);
148 ret = sync_filesystem(sb);
149 up_read(&sb->s_umount);
151 fput_light(file, fput_needed);
156 * vfs_fsync_range - helper to sync a range of data & metadata to disk
157 * @file: file to sync
158 * @start: offset in bytes of the beginning of data range to sync
159 * @end: offset in bytes of the end of data range (inclusive)
160 * @datasync: perform only datasync
162 * Write back data in range @start..@end and metadata for @file to disk. If
163 * @datasync is set only metadata needed to access modified file data is
166 int vfs_fsync_range(struct file *file, loff_t start, loff_t end, int datasync)
168 struct address_space *mapping = file->f_mapping;
171 if (!file->f_op || !file->f_op->fsync) {
176 ret = filemap_write_and_wait_range(mapping, start, end);
179 * We need to protect against concurrent writers, which could cause
180 * livelocks in fsync_buffers_list().
182 mutex_lock(&mapping->host->i_mutex);
183 err = file->f_op->fsync(file, datasync);
186 mutex_unlock(&mapping->host->i_mutex);
191 EXPORT_SYMBOL(vfs_fsync_range);
194 * vfs_fsync - perform a fsync or fdatasync on a file
195 * @file: file to sync
196 * @datasync: only perform a fdatasync operation
198 * Write back data and metadata for @file to disk. If @datasync is
199 * set only metadata needed to access modified file data is written.
201 int vfs_fsync(struct file *file, int datasync)
203 return vfs_fsync_range(file, 0, LLONG_MAX, datasync);
205 EXPORT_SYMBOL(vfs_fsync);
207 static int do_fsync(unsigned int fd, int datasync)
214 ret = vfs_fsync(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 * generic_write_sync - perform syncing after a write if file / inode is sync
232 * @file: file to which the write happened
233 * @pos: offset where the write started
234 * @count: length of the write
236 * This is just a simple wrapper about our general syncing function.
238 int generic_write_sync(struct file *file, loff_t pos, loff_t count)
240 if (!(file->f_flags & O_DSYNC) && !IS_SYNC(file->f_mapping->host))
242 return vfs_fsync_range(file, pos, pos + count - 1,
243 (file->f_flags & __O_SYNC) ? 0 : 1);
245 EXPORT_SYMBOL(generic_write_sync);
248 * sys_sync_file_range() permits finely controlled syncing over a segment of
249 * a file in the range offset .. (offset+nbytes-1) inclusive. If nbytes is
250 * zero then sys_sync_file_range() will operate from offset out to EOF.
254 * SYNC_FILE_RANGE_WAIT_BEFORE: wait upon writeout of all pages in the range
255 * before performing the write.
257 * SYNC_FILE_RANGE_WRITE: initiate writeout of all those dirty pages in the
258 * range which are not presently under writeback. Note that this may block for
259 * significant periods due to exhaustion of disk request structures.
261 * SYNC_FILE_RANGE_WAIT_AFTER: wait upon writeout of all pages in the range
262 * after performing the write.
264 * Useful combinations of the flag bits are:
266 * SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE: ensures that all pages
267 * in the range which were dirty on entry to sys_sync_file_range() are placed
268 * under writeout. This is a start-write-for-data-integrity operation.
270 * SYNC_FILE_RANGE_WRITE: start writeout of all dirty pages in the range which
271 * are not presently under writeout. This is an asynchronous flush-to-disk
272 * operation. Not suitable for data integrity operations.
274 * SYNC_FILE_RANGE_WAIT_BEFORE (or SYNC_FILE_RANGE_WAIT_AFTER): wait for
275 * completion of writeout of all pages in the range. This will be used after an
276 * earlier SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE operation to wait
277 * for that operation to complete and to return the result.
279 * SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE|SYNC_FILE_RANGE_WAIT_AFTER:
280 * a traditional sync() operation. This is a write-for-data-integrity operation
281 * which will ensure that all pages in the range which were dirty on entry to
282 * sys_sync_file_range() are committed to disk.
285 * SYNC_FILE_RANGE_WAIT_BEFORE and SYNC_FILE_RANGE_WAIT_AFTER will detect any
286 * I/O errors or ENOSPC conditions and will return those to the caller, after
287 * clearing the EIO and ENOSPC flags in the address_space.
289 * It should be noted that none of these operations write out the file's
290 * metadata. So unless the application is strictly performing overwrites of
291 * already-instantiated disk blocks, there are no guarantees here that the data
292 * will be available after a crash.
294 SYSCALL_DEFINE(sync_file_range)(int fd, loff_t offset, loff_t nbytes,
299 struct address_space *mapping;
300 loff_t endbyte; /* inclusive */
305 if (flags & ~VALID_FLAGS)
308 endbyte = offset + nbytes;
312 if ((s64)endbyte < 0)
314 if (endbyte < offset)
317 if (sizeof(pgoff_t) == 4) {
318 if (offset >= (0x100000000ULL << PAGE_CACHE_SHIFT)) {
320 * The range starts outside a 32 bit machine's
321 * pagecache addressing capabilities. Let it "succeed"
326 if (endbyte >= (0x100000000ULL << PAGE_CACHE_SHIFT)) {
337 endbyte--; /* inclusive */
340 file = fget_light(fd, &fput_needed);
344 i_mode = file->f_path.dentry->d_inode->i_mode;
346 if (!S_ISREG(i_mode) && !S_ISBLK(i_mode) && !S_ISDIR(i_mode) &&
350 mapping = file->f_mapping;
357 if (flags & SYNC_FILE_RANGE_WAIT_BEFORE) {
358 ret = filemap_fdatawait_range(mapping, offset, endbyte);
363 if (flags & SYNC_FILE_RANGE_WRITE) {
364 ret = filemap_fdatawrite_range(mapping, offset, endbyte);
369 if (flags & SYNC_FILE_RANGE_WAIT_AFTER)
370 ret = filemap_fdatawait_range(mapping, offset, endbyte);
373 fput_light(file, fput_needed);
377 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
378 asmlinkage long SyS_sync_file_range(long fd, loff_t offset, loff_t nbytes,
381 return SYSC_sync_file_range((int) fd, offset, nbytes,
382 (unsigned int) flags);
384 SYSCALL_ALIAS(sys_sync_file_range, SyS_sync_file_range);
387 /* It would be nice if people remember that not all the world's an i386
388 when they introduce new system calls */
389 SYSCALL_DEFINE(sync_file_range2)(int fd, unsigned int flags,
390 loff_t offset, loff_t nbytes)
392 return sys_sync_file_range(fd, offset, nbytes, flags);
394 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
395 asmlinkage long SyS_sync_file_range2(long fd, long flags,
396 loff_t offset, loff_t nbytes)
398 return SYSC_sync_file_range2((int) fd, (unsigned int) flags,
401 SYSCALL_ALIAS(sys_sync_file_range2, SyS_sync_file_range2);