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_locked(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 * filp may be NULL if called via the msync of a vma.
136 int file_fsync(struct file *filp, struct dentry *dentry, int datasync)
138 struct inode * inode = dentry->d_inode;
139 struct super_block * sb;
142 /* sync the inode to buffers */
143 ret = write_inode_now(inode, 0);
145 /* sync the superblock to buffers */
147 if (sb->s_dirt && sb->s_op->write_super)
148 sb->s_op->write_super(sb);
150 /* .. finally sync the buffers to disk */
151 err = sync_blockdev(sb->s_bdev);
156 EXPORT_SYMBOL(file_fsync);
159 * vfs_fsync_range - helper to sync a range of data & metadata to disk
160 * @file: file to sync
161 * @start: offset in bytes of the beginning of data range to sync
162 * @end: offset in bytes of the end of data range (inclusive)
163 * @datasync: perform only datasync
165 * Write back data in range @start..@end and metadata for @file to disk. If
166 * @datasync is set only metadata needed to access modified file data is
169 int vfs_fsync_range(struct file *file, loff_t start, loff_t end, int datasync)
171 struct address_space *mapping = file->f_mapping;
174 if (!file->f_op || !file->f_op->fsync) {
179 ret = filemap_write_and_wait_range(mapping, start, end);
182 * We need to protect against concurrent writers, which could cause
183 * livelocks in fsync_buffers_list().
185 mutex_lock(&mapping->host->i_mutex);
186 err = file->f_op->fsync(file, file->f_path.dentry, datasync);
189 mutex_unlock(&mapping->host->i_mutex);
194 EXPORT_SYMBOL(vfs_fsync_range);
197 * vfs_fsync - perform a fsync or fdatasync on a file
198 * @file: file to sync
199 * @datasync: only perform a fdatasync operation
201 * Write back data and metadata for @file to disk. If @datasync is
202 * set only metadata needed to access modified file data is written.
204 int vfs_fsync(struct file *file, int datasync)
206 return vfs_fsync_range(file, 0, LLONG_MAX, datasync);
208 EXPORT_SYMBOL(vfs_fsync);
210 static int do_fsync(unsigned int fd, int datasync)
217 ret = vfs_fsync(file, datasync);
223 SYSCALL_DEFINE1(fsync, unsigned int, fd)
225 return do_fsync(fd, 0);
228 SYSCALL_DEFINE1(fdatasync, unsigned int, fd)
230 return do_fsync(fd, 1);
234 * generic_write_sync - perform syncing after a write if file / inode is sync
235 * @file: file to which the write happened
236 * @pos: offset where the write started
237 * @count: length of the write
239 * This is just a simple wrapper about our general syncing function.
241 int generic_write_sync(struct file *file, loff_t pos, loff_t count)
243 if (!(file->f_flags & O_DSYNC) && !IS_SYNC(file->f_mapping->host))
245 return vfs_fsync_range(file, pos, pos + count - 1,
246 (file->f_flags & __O_SYNC) ? 0 : 1);
248 EXPORT_SYMBOL(generic_write_sync);
251 * sys_sync_file_range() permits finely controlled syncing over a segment of
252 * a file in the range offset .. (offset+nbytes-1) inclusive. If nbytes is
253 * zero then sys_sync_file_range() will operate from offset out to EOF.
257 * SYNC_FILE_RANGE_WAIT_BEFORE: wait upon writeout of all pages in the range
258 * before performing the write.
260 * SYNC_FILE_RANGE_WRITE: initiate writeout of all those dirty pages in the
261 * range which are not presently under writeback. Note that this may block for
262 * significant periods due to exhaustion of disk request structures.
264 * SYNC_FILE_RANGE_WAIT_AFTER: wait upon writeout of all pages in the range
265 * after performing the write.
267 * Useful combinations of the flag bits are:
269 * SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE: ensures that all pages
270 * in the range which were dirty on entry to sys_sync_file_range() are placed
271 * under writeout. This is a start-write-for-data-integrity operation.
273 * SYNC_FILE_RANGE_WRITE: start writeout of all dirty pages in the range which
274 * are not presently under writeout. This is an asynchronous flush-to-disk
275 * operation. Not suitable for data integrity operations.
277 * SYNC_FILE_RANGE_WAIT_BEFORE (or SYNC_FILE_RANGE_WAIT_AFTER): wait for
278 * completion of writeout of all pages in the range. This will be used after an
279 * earlier SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE operation to wait
280 * for that operation to complete and to return the result.
282 * SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE|SYNC_FILE_RANGE_WAIT_AFTER:
283 * a traditional sync() operation. This is a write-for-data-integrity operation
284 * which will ensure that all pages in the range which were dirty on entry to
285 * sys_sync_file_range() are committed to disk.
288 * SYNC_FILE_RANGE_WAIT_BEFORE and SYNC_FILE_RANGE_WAIT_AFTER will detect any
289 * I/O errors or ENOSPC conditions and will return those to the caller, after
290 * clearing the EIO and ENOSPC flags in the address_space.
292 * It should be noted that none of these operations write out the file's
293 * metadata. So unless the application is strictly performing overwrites of
294 * already-instantiated disk blocks, there are no guarantees here that the data
295 * will be available after a crash.
297 SYSCALL_DEFINE(sync_file_range)(int fd, loff_t offset, loff_t nbytes,
302 struct address_space *mapping;
303 loff_t endbyte; /* inclusive */
308 if (flags & ~VALID_FLAGS)
311 endbyte = offset + nbytes;
315 if ((s64)endbyte < 0)
317 if (endbyte < offset)
320 if (sizeof(pgoff_t) == 4) {
321 if (offset >= (0x100000000ULL << PAGE_CACHE_SHIFT)) {
323 * The range starts outside a 32 bit machine's
324 * pagecache addressing capabilities. Let it "succeed"
329 if (endbyte >= (0x100000000ULL << PAGE_CACHE_SHIFT)) {
340 endbyte--; /* inclusive */
343 file = fget_light(fd, &fput_needed);
347 i_mode = file->f_path.dentry->d_inode->i_mode;
349 if (!S_ISREG(i_mode) && !S_ISBLK(i_mode) && !S_ISDIR(i_mode) &&
353 mapping = file->f_mapping;
360 if (flags & SYNC_FILE_RANGE_WAIT_BEFORE) {
361 ret = filemap_fdatawait_range(mapping, offset, endbyte);
366 if (flags & SYNC_FILE_RANGE_WRITE) {
367 ret = filemap_fdatawrite_range(mapping, offset, endbyte);
372 if (flags & SYNC_FILE_RANGE_WAIT_AFTER)
373 ret = filemap_fdatawait_range(mapping, offset, endbyte);
376 fput_light(file, fput_needed);
380 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
381 asmlinkage long SyS_sync_file_range(long fd, loff_t offset, loff_t nbytes,
384 return SYSC_sync_file_range((int) fd, offset, nbytes,
385 (unsigned int) flags);
387 SYSCALL_ALIAS(sys_sync_file_range, SyS_sync_file_range);
390 /* It would be nice if people remember that not all the world's an i386
391 when they introduce new system calls */
392 SYSCALL_DEFINE(sync_file_range2)(int fd, unsigned int flags,
393 loff_t offset, loff_t nbytes)
395 return sys_sync_file_range(fd, offset, nbytes, flags);
397 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
398 asmlinkage long SyS_sync_file_range2(long fd, long flags,
399 loff_t offset, loff_t nbytes)
401 return SYSC_sync_file_range2((int) fd, (unsigned int) flags,
404 SYSCALL_ALIAS(sys_sync_file_range2, SyS_sync_file_range2);