2 * High-level sync()-related operations
5 #include <linux/kernel.h>
6 #include <linux/file.h>
8 #include <linux/module.h>
9 #include <linux/sched.h>
10 #include <linux/writeback.h>
11 #include <linux/syscalls.h>
12 #include <linux/linkage.h>
13 #include <linux/pagemap.h>
14 #include <linux/quotaops.h>
15 #include <linux/buffer_head.h>
18 #define VALID_FLAGS (SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE| \
19 SYNC_FILE_RANGE_WAIT_AFTER)
22 * Do the filesystem syncing work. For simple filesystems sync_inodes_sb(sb, 0)
23 * just dirties buffers with inodes so we have to submit IO for these buffers
24 * via __sync_blockdev(). This also speeds up the wait == 1 case since in that
25 * case write_inode() functions do sync_dirty_buffer() and thus effectively
26 * write one block at a time.
28 static int __sync_filesystem(struct super_block *sb, int wait)
30 /* Avoid doing twice syncing and cache pruning for quota sync */
32 writeout_quota_sb(sb, -1);
34 sync_quota_sb(sb, -1);
35 sync_inodes_sb(sb, wait);
36 if (sb->s_op->sync_fs)
37 sb->s_op->sync_fs(sb, wait);
38 return __sync_blockdev(sb->s_bdev, wait);
42 * Write out and wait upon all dirty data associated with this
43 * superblock. Filesystem data as well as the underlying block
44 * device. Takes the superblock lock.
46 int sync_filesystem(struct super_block *sb)
51 * We need to be protected against the filesystem going from
52 * r/o to r/w or vice versa.
54 WARN_ON(!rwsem_is_locked(&sb->s_umount));
57 * No point in syncing out anything if the filesystem is read-only.
59 if (sb->s_flags & MS_RDONLY)
62 ret = __sync_filesystem(sb, 0);
65 return __sync_filesystem(sb, 1);
67 EXPORT_SYMBOL_GPL(sync_filesystem);
70 * Sync all the data for all the filesystems (called by sys_sync() and
73 * This operation is careful to avoid the livelock which could easily happen
74 * if two or more filesystems are being continuously dirtied. s_need_sync
75 * is used only here. We set it against all filesystems and then clear it as
76 * we sync them. So redirtied filesystems are skipped.
78 * But if process A is currently running sync_filesystems and then process B
79 * calls sync_filesystems as well, process B will set all the s_need_sync
80 * flags again, which will cause process A to resync everything. Fix that with
83 static void sync_filesystems(int wait)
85 struct super_block *sb;
86 static DEFINE_MUTEX(mutex);
88 mutex_lock(&mutex); /* Could be down_interruptible */
90 list_for_each_entry(sb, &super_blocks, s_list)
94 list_for_each_entry(sb, &super_blocks, s_list) {
99 spin_unlock(&sb_lock);
101 down_read(&sb->s_umount);
102 if (!(sb->s_flags & MS_RDONLY) && sb->s_root)
103 __sync_filesystem(sb, wait);
104 up_read(&sb->s_umount);
106 /* restart only when sb is no longer on the list */
108 if (__put_super_and_need_restart(sb))
111 spin_unlock(&sb_lock);
112 mutex_unlock(&mutex);
116 * sync everything. Start out by waking pdflush, because that writes back
117 * all queues in parallel.
119 SYSCALL_DEFINE0(sync)
124 if (unlikely(laptop_mode))
125 laptop_sync_completion();
129 static void do_sync_work(struct work_struct *work)
132 * Sync twice to reduce the possibility we skipped some inodes / pages
133 * because they were temporarily locked
137 printk("Emergency Sync complete\n");
141 void emergency_sync(void)
143 struct work_struct *work;
145 work = kmalloc(sizeof(*work), GFP_ATOMIC);
147 INIT_WORK(work, do_sync_work);
153 * Generic function to fsync a file.
155 * filp may be NULL if called via the msync of a vma.
157 int file_fsync(struct file *filp, struct dentry *dentry, int datasync)
159 struct inode * inode = dentry->d_inode;
160 struct super_block * sb;
163 /* sync the inode to buffers */
164 ret = write_inode_now(inode, 0);
166 /* sync the superblock to buffers */
168 if (sb->s_dirt && sb->s_op->write_super)
169 sb->s_op->write_super(sb);
171 /* .. finally sync the buffers to disk */
172 err = sync_blockdev(sb->s_bdev);
179 * vfs_fsync - perform a fsync or fdatasync on a file
180 * @file: file to sync
181 * @dentry: dentry of @file
182 * @data: only perform a fdatasync operation
184 * Write back data and metadata for @file to disk. If @datasync is
185 * set only metadata needed to access modified file data is written.
187 * In case this function is called from nfsd @file may be %NULL and
188 * only @dentry is set. This can only happen when the filesystem
189 * implements the export_operations API.
191 int vfs_fsync(struct file *file, struct dentry *dentry, int datasync)
193 const struct file_operations *fop;
194 struct address_space *mapping;
198 * Get mapping and operations from the file in case we have
199 * as file, or get the default values for them in case we
200 * don't have a struct file available. Damn nfsd..
203 mapping = file->f_mapping;
206 mapping = dentry->d_inode->i_mapping;
207 fop = dentry->d_inode->i_fop;
210 if (!fop || !fop->fsync) {
215 ret = filemap_fdatawrite(mapping);
218 * We need to protect against concurrent writers, which could cause
219 * livelocks in fsync_buffers_list().
221 mutex_lock(&mapping->host->i_mutex);
222 err = fop->fsync(file, dentry, datasync);
225 mutex_unlock(&mapping->host->i_mutex);
226 err = filemap_fdatawait(mapping);
232 EXPORT_SYMBOL(vfs_fsync);
234 static int do_fsync(unsigned int fd, int datasync)
241 ret = vfs_fsync(file, file->f_path.dentry, datasync);
247 SYSCALL_DEFINE1(fsync, unsigned int, fd)
249 return do_fsync(fd, 0);
252 SYSCALL_DEFINE1(fdatasync, unsigned int, fd)
254 return do_fsync(fd, 1);
258 * sys_sync_file_range() permits finely controlled syncing over a segment of
259 * a file in the range offset .. (offset+nbytes-1) inclusive. If nbytes is
260 * zero then sys_sync_file_range() will operate from offset out to EOF.
264 * SYNC_FILE_RANGE_WAIT_BEFORE: wait upon writeout of all pages in the range
265 * before performing the write.
267 * SYNC_FILE_RANGE_WRITE: initiate writeout of all those dirty pages in the
268 * range which are not presently under writeback. Note that this may block for
269 * significant periods due to exhaustion of disk request structures.
271 * SYNC_FILE_RANGE_WAIT_AFTER: wait upon writeout of all pages in the range
272 * after performing the write.
274 * Useful combinations of the flag bits are:
276 * SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE: ensures that all pages
277 * in the range which were dirty on entry to sys_sync_file_range() are placed
278 * under writeout. This is a start-write-for-data-integrity operation.
280 * SYNC_FILE_RANGE_WRITE: start writeout of all dirty pages in the range which
281 * are not presently under writeout. This is an asynchronous flush-to-disk
282 * operation. Not suitable for data integrity operations.
284 * SYNC_FILE_RANGE_WAIT_BEFORE (or SYNC_FILE_RANGE_WAIT_AFTER): wait for
285 * completion of writeout of all pages in the range. This will be used after an
286 * earlier SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE operation to wait
287 * for that operation to complete and to return the result.
289 * SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE|SYNC_FILE_RANGE_WAIT_AFTER:
290 * a traditional sync() operation. This is a write-for-data-integrity operation
291 * which will ensure that all pages in the range which were dirty on entry to
292 * sys_sync_file_range() are committed to disk.
295 * SYNC_FILE_RANGE_WAIT_BEFORE and SYNC_FILE_RANGE_WAIT_AFTER will detect any
296 * I/O errors or ENOSPC conditions and will return those to the caller, after
297 * clearing the EIO and ENOSPC flags in the address_space.
299 * It should be noted that none of these operations write out the file's
300 * metadata. So unless the application is strictly performing overwrites of
301 * already-instantiated disk blocks, there are no guarantees here that the data
302 * will be available after a crash.
304 SYSCALL_DEFINE(sync_file_range)(int fd, loff_t offset, loff_t nbytes,
309 loff_t endbyte; /* inclusive */
314 if (flags & ~VALID_FLAGS)
317 endbyte = offset + nbytes;
321 if ((s64)endbyte < 0)
323 if (endbyte < offset)
326 if (sizeof(pgoff_t) == 4) {
327 if (offset >= (0x100000000ULL << PAGE_CACHE_SHIFT)) {
329 * The range starts outside a 32 bit machine's
330 * pagecache addressing capabilities. Let it "succeed"
335 if (endbyte >= (0x100000000ULL << PAGE_CACHE_SHIFT)) {
346 endbyte--; /* inclusive */
349 file = fget_light(fd, &fput_needed);
353 i_mode = file->f_path.dentry->d_inode->i_mode;
355 if (!S_ISREG(i_mode) && !S_ISBLK(i_mode) && !S_ISDIR(i_mode) &&
359 ret = do_sync_mapping_range(file->f_mapping, offset, endbyte, flags);
361 fput_light(file, fput_needed);
365 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
366 asmlinkage long SyS_sync_file_range(long fd, loff_t offset, loff_t nbytes,
369 return SYSC_sync_file_range((int) fd, offset, nbytes,
370 (unsigned int) flags);
372 SYSCALL_ALIAS(sys_sync_file_range, SyS_sync_file_range);
375 /* It would be nice if people remember that not all the world's an i386
376 when they introduce new system calls */
377 SYSCALL_DEFINE(sync_file_range2)(int fd, unsigned int flags,
378 loff_t offset, loff_t nbytes)
380 return sys_sync_file_range(fd, offset, nbytes, flags);
382 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
383 asmlinkage long SyS_sync_file_range2(long fd, long flags,
384 loff_t offset, loff_t nbytes)
386 return SYSC_sync_file_range2((int) fd, (unsigned int) flags,
389 SYSCALL_ALIAS(sys_sync_file_range2, SyS_sync_file_range2);
393 * `endbyte' is inclusive
395 int do_sync_mapping_range(struct address_space *mapping, loff_t offset,
396 loff_t endbyte, unsigned int flags)
406 if (flags & SYNC_FILE_RANGE_WAIT_BEFORE) {
407 ret = wait_on_page_writeback_range(mapping,
408 offset >> PAGE_CACHE_SHIFT,
409 endbyte >> PAGE_CACHE_SHIFT);
414 if (flags & SYNC_FILE_RANGE_WRITE) {
415 ret = __filemap_fdatawrite_range(mapping, offset, endbyte,
421 if (flags & SYNC_FILE_RANGE_WAIT_AFTER) {
422 ret = wait_on_page_writeback_range(mapping,
423 offset >> PAGE_CACHE_SHIFT,
424 endbyte >> PAGE_CACHE_SHIFT);
429 EXPORT_SYMBOL_GPL(do_sync_mapping_range);