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);
37 if (sb->s_dirt && sb->s_op->write_super)
38 sb->s_op->write_super(sb);
40 if (sb->s_op->sync_fs)
41 sb->s_op->sync_fs(sb, wait);
42 return __sync_blockdev(sb->s_bdev, wait);
46 * Write out and wait upon all dirty data associated with this
47 * superblock. Filesystem data as well as the underlying block
48 * device. Takes the superblock lock.
50 int sync_filesystem(struct super_block *sb)
55 * We need to be protected against the filesystem going from
56 * r/o to r/w or vice versa.
58 WARN_ON(!rwsem_is_locked(&sb->s_umount));
61 * No point in syncing out anything if the filesystem is read-only.
63 if (sb->s_flags & MS_RDONLY)
66 ret = __sync_filesystem(sb, 0);
69 return __sync_filesystem(sb, 1);
71 EXPORT_SYMBOL_GPL(sync_filesystem);
74 * Sync all the data for all the filesystems (called by sys_sync() and
77 * This operation is careful to avoid the livelock which could easily happen
78 * if two or more filesystems are being continuously dirtied. s_need_sync
79 * is used only here. We set it against all filesystems and then clear it as
80 * we sync them. So redirtied filesystems are skipped.
82 * But if process A is currently running sync_filesystems and then process B
83 * calls sync_filesystems as well, process B will set all the s_need_sync
84 * flags again, which will cause process A to resync everything. Fix that with
87 static void sync_filesystems(int wait)
89 struct super_block *sb;
90 static DEFINE_MUTEX(mutex);
92 mutex_lock(&mutex); /* Could be down_interruptible */
94 list_for_each_entry(sb, &super_blocks, s_list)
98 list_for_each_entry(sb, &super_blocks, s_list) {
103 spin_unlock(&sb_lock);
105 down_read(&sb->s_umount);
106 if (!(sb->s_flags & MS_RDONLY) && sb->s_root)
107 __sync_filesystem(sb, wait);
108 up_read(&sb->s_umount);
110 /* restart only when sb is no longer on the list */
112 if (__put_super_and_need_restart(sb))
115 spin_unlock(&sb_lock);
116 mutex_unlock(&mutex);
119 SYSCALL_DEFINE0(sync)
123 if (unlikely(laptop_mode))
124 laptop_sync_completion();
128 static void do_sync_work(struct work_struct *work)
131 * Sync twice to reduce the possibility we skipped some inodes / pages
132 * because they were temporarily locked
136 printk("Emergency Sync complete\n");
140 void emergency_sync(void)
142 struct work_struct *work;
144 work = kmalloc(sizeof(*work), GFP_ATOMIC);
146 INIT_WORK(work, do_sync_work);
152 * Generic function to fsync a file.
154 * filp may be NULL if called via the msync of a vma.
156 int file_fsync(struct file *filp, struct dentry *dentry, int datasync)
158 struct inode * inode = dentry->d_inode;
159 struct super_block * sb;
162 /* sync the inode to buffers */
163 ret = write_inode_now(inode, 0);
165 /* sync the superblock to buffers */
168 if (sb->s_dirt && sb->s_op->write_super)
169 sb->s_op->write_super(sb);
172 /* .. finally sync the buffers to disk */
173 err = sync_blockdev(sb->s_bdev);
180 * vfs_fsync - perform a fsync or fdatasync on a file
181 * @file: file to sync
182 * @dentry: dentry of @file
183 * @data: only perform a fdatasync operation
185 * Write back data and metadata for @file to disk. If @datasync is
186 * set only metadata needed to access modified file data is written.
188 * In case this function is called from nfsd @file may be %NULL and
189 * only @dentry is set. This can only happen when the filesystem
190 * implements the export_operations API.
192 int vfs_fsync(struct file *file, struct dentry *dentry, int datasync)
194 const struct file_operations *fop;
195 struct address_space *mapping;
199 * Get mapping and operations from the file in case we have
200 * as file, or get the default values for them in case we
201 * don't have a struct file available. Damn nfsd..
204 mapping = file->f_mapping;
207 mapping = dentry->d_inode->i_mapping;
208 fop = dentry->d_inode->i_fop;
211 if (!fop || !fop->fsync) {
216 ret = filemap_fdatawrite(mapping);
219 * We need to protect against concurrent writers, which could cause
220 * livelocks in fsync_buffers_list().
222 mutex_lock(&mapping->host->i_mutex);
223 err = fop->fsync(file, dentry, datasync);
226 mutex_unlock(&mapping->host->i_mutex);
227 err = filemap_fdatawait(mapping);
233 EXPORT_SYMBOL(vfs_fsync);
235 static int do_fsync(unsigned int fd, int datasync)
242 ret = vfs_fsync(file, file->f_path.dentry, datasync);
248 SYSCALL_DEFINE1(fsync, unsigned int, fd)
250 return do_fsync(fd, 0);
253 SYSCALL_DEFINE1(fdatasync, unsigned int, fd)
255 return do_fsync(fd, 1);
259 * sys_sync_file_range() permits finely controlled syncing over a segment of
260 * a file in the range offset .. (offset+nbytes-1) inclusive. If nbytes is
261 * zero then sys_sync_file_range() will operate from offset out to EOF.
265 * SYNC_FILE_RANGE_WAIT_BEFORE: wait upon writeout of all pages in the range
266 * before performing the write.
268 * SYNC_FILE_RANGE_WRITE: initiate writeout of all those dirty pages in the
269 * range which are not presently under writeback. Note that this may block for
270 * significant periods due to exhaustion of disk request structures.
272 * SYNC_FILE_RANGE_WAIT_AFTER: wait upon writeout of all pages in the range
273 * after performing the write.
275 * Useful combinations of the flag bits are:
277 * SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE: ensures that all pages
278 * in the range which were dirty on entry to sys_sync_file_range() are placed
279 * under writeout. This is a start-write-for-data-integrity operation.
281 * SYNC_FILE_RANGE_WRITE: start writeout of all dirty pages in the range which
282 * are not presently under writeout. This is an asynchronous flush-to-disk
283 * operation. Not suitable for data integrity operations.
285 * SYNC_FILE_RANGE_WAIT_BEFORE (or SYNC_FILE_RANGE_WAIT_AFTER): wait for
286 * completion of writeout of all pages in the range. This will be used after an
287 * earlier SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE operation to wait
288 * for that operation to complete and to return the result.
290 * SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE|SYNC_FILE_RANGE_WAIT_AFTER:
291 * a traditional sync() operation. This is a write-for-data-integrity operation
292 * which will ensure that all pages in the range which were dirty on entry to
293 * sys_sync_file_range() are committed to disk.
296 * SYNC_FILE_RANGE_WAIT_BEFORE and SYNC_FILE_RANGE_WAIT_AFTER will detect any
297 * I/O errors or ENOSPC conditions and will return those to the caller, after
298 * clearing the EIO and ENOSPC flags in the address_space.
300 * It should be noted that none of these operations write out the file's
301 * metadata. So unless the application is strictly performing overwrites of
302 * already-instantiated disk blocks, there are no guarantees here that the data
303 * will be available after a crash.
305 SYSCALL_DEFINE(sync_file_range)(int fd, loff_t offset, loff_t nbytes,
310 loff_t endbyte; /* inclusive */
315 if (flags & ~VALID_FLAGS)
318 endbyte = offset + nbytes;
322 if ((s64)endbyte < 0)
324 if (endbyte < offset)
327 if (sizeof(pgoff_t) == 4) {
328 if (offset >= (0x100000000ULL << PAGE_CACHE_SHIFT)) {
330 * The range starts outside a 32 bit machine's
331 * pagecache addressing capabilities. Let it "succeed"
336 if (endbyte >= (0x100000000ULL << PAGE_CACHE_SHIFT)) {
347 endbyte--; /* inclusive */
350 file = fget_light(fd, &fput_needed);
354 i_mode = file->f_path.dentry->d_inode->i_mode;
356 if (!S_ISREG(i_mode) && !S_ISBLK(i_mode) && !S_ISDIR(i_mode) &&
360 ret = do_sync_mapping_range(file->f_mapping, offset, endbyte, flags);
362 fput_light(file, fput_needed);
366 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
367 asmlinkage long SyS_sync_file_range(long fd, loff_t offset, loff_t nbytes,
370 return SYSC_sync_file_range((int) fd, offset, nbytes,
371 (unsigned int) flags);
373 SYSCALL_ALIAS(sys_sync_file_range, SyS_sync_file_range);
376 /* It would be nice if people remember that not all the world's an i386
377 when they introduce new system calls */
378 SYSCALL_DEFINE(sync_file_range2)(int fd, unsigned int flags,
379 loff_t offset, loff_t nbytes)
381 return sys_sync_file_range(fd, offset, nbytes, flags);
383 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
384 asmlinkage long SyS_sync_file_range2(long fd, long flags,
385 loff_t offset, loff_t nbytes)
387 return SYSC_sync_file_range2((int) fd, (unsigned int) flags,
390 SYSCALL_ALIAS(sys_sync_file_range2, SyS_sync_file_range2);
394 * `endbyte' is inclusive
396 int do_sync_mapping_range(struct address_space *mapping, loff_t offset,
397 loff_t endbyte, unsigned int flags)
407 if (flags & SYNC_FILE_RANGE_WAIT_BEFORE) {
408 ret = wait_on_page_writeback_range(mapping,
409 offset >> PAGE_CACHE_SHIFT,
410 endbyte >> PAGE_CACHE_SHIFT);
415 if (flags & SYNC_FILE_RANGE_WRITE) {
416 ret = __filemap_fdatawrite_range(mapping, offset, endbyte,
422 if (flags & SYNC_FILE_RANGE_WAIT_AFTER) {
423 ret = wait_on_page_writeback_range(mapping,
424 offset >> PAGE_CACHE_SHIFT,
425 endbyte >> PAGE_CACHE_SHIFT);
430 EXPORT_SYMBOL_GPL(do_sync_mapping_range);