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
23 * writeback_inodes_sb(sb) just dirties buffers with inodes so we have to
24 * submit IO for these buffers via __sync_blockdev(). This also speeds up the
25 * wait == 1 case since in that case write_inode() functions do
26 * sync_dirty_buffer() and thus effectively write one block at a time.
28 static int __sync_filesystem(struct super_block
*sb
, int wait
)
31 * This should be safe, as we require bdi backing to actually
32 * write out data in the first place
37 /* Avoid doing twice syncing and cache pruning for quota sync */
39 writeout_quota_sb(sb
, -1);
40 writeback_inodes_sb(sb
);
42 sync_quota_sb(sb
, -1);
45 if (sb
->s_op
->sync_fs
)
46 sb
->s_op
->sync_fs(sb
, wait
);
47 return __sync_blockdev(sb
->s_bdev
, wait
);
51 * Write out and wait upon all dirty data associated with this
52 * superblock. Filesystem data as well as the underlying block
53 * device. Takes the superblock lock.
55 int sync_filesystem(struct super_block
*sb
)
60 * We need to be protected against the filesystem going from
61 * r/o to r/w or vice versa.
63 WARN_ON(!rwsem_is_locked(&sb
->s_umount
));
66 * No point in syncing out anything if the filesystem is read-only.
68 if (sb
->s_flags
& MS_RDONLY
)
71 ret
= __sync_filesystem(sb
, 0);
74 return __sync_filesystem(sb
, 1);
76 EXPORT_SYMBOL_GPL(sync_filesystem
);
79 * Sync all the data for all the filesystems (called by sys_sync() and
82 * This operation is careful to avoid the livelock which could easily happen
83 * if two or more filesystems are being continuously dirtied. s_need_sync
84 * is used only here. We set it against all filesystems and then clear it as
85 * we sync them. So redirtied filesystems are skipped.
87 * But if process A is currently running sync_filesystems and then process B
88 * calls sync_filesystems as well, process B will set all the s_need_sync
89 * flags again, which will cause process A to resync everything. Fix that with
92 static void sync_filesystems(int wait
)
94 struct super_block
*sb
;
95 static DEFINE_MUTEX(mutex
);
97 mutex_lock(&mutex
); /* Could be down_interruptible */
99 list_for_each_entry(sb
, &super_blocks
, s_list
)
103 list_for_each_entry(sb
, &super_blocks
, s_list
) {
104 if (!sb
->s_need_sync
)
108 spin_unlock(&sb_lock
);
110 down_read(&sb
->s_umount
);
111 if (!(sb
->s_flags
& MS_RDONLY
) && sb
->s_root
&& sb
->s_bdi
)
112 __sync_filesystem(sb
, wait
);
113 up_read(&sb
->s_umount
);
115 /* restart only when sb is no longer on the list */
117 if (__put_super_and_need_restart(sb
))
120 spin_unlock(&sb_lock
);
121 mutex_unlock(&mutex
);
125 * sync everything. Start out by waking pdflush, because that writes back
126 * all queues in parallel.
128 SYSCALL_DEFINE0(sync
)
130 wakeup_flusher_threads(0);
133 if (unlikely(laptop_mode
))
134 laptop_sync_completion();
138 static void do_sync_work(struct work_struct
*work
)
141 * Sync twice to reduce the possibility we skipped some inodes / pages
142 * because they were temporarily locked
146 printk("Emergency Sync complete\n");
150 void emergency_sync(void)
152 struct work_struct
*work
;
154 work
= kmalloc(sizeof(*work
), GFP_ATOMIC
);
156 INIT_WORK(work
, do_sync_work
);
162 * Generic function to fsync a file.
164 * filp may be NULL if called via the msync of a vma.
166 int file_fsync(struct file
*filp
, struct dentry
*dentry
, int datasync
)
168 struct inode
* inode
= dentry
->d_inode
;
169 struct super_block
* sb
;
172 /* sync the inode to buffers */
173 ret
= write_inode_now(inode
, 0);
175 /* sync the superblock to buffers */
177 if (sb
->s_dirt
&& sb
->s_op
->write_super
)
178 sb
->s_op
->write_super(sb
);
180 /* .. finally sync the buffers to disk */
181 err
= sync_blockdev(sb
->s_bdev
);
186 EXPORT_SYMBOL(file_fsync
);
189 * vfs_fsync_range - helper to sync a range of data & metadata to disk
190 * @file: file to sync
191 * @dentry: dentry of @file
192 * @start: offset in bytes of the beginning of data range to sync
193 * @end: offset in bytes of the end of data range (inclusive)
194 * @datasync: perform only datasync
196 * Write back data in range @start..@end and metadata for @file to disk. If
197 * @datasync is set only metadata needed to access modified file data is
200 * In case this function is called from nfsd @file may be %NULL and
201 * only @dentry is set. This can only happen when the filesystem
202 * implements the export_operations API.
204 int vfs_fsync_range(struct file
*file
, struct dentry
*dentry
, loff_t start
,
205 loff_t end
, int datasync
)
207 const struct file_operations
*fop
;
208 struct address_space
*mapping
;
212 * Get mapping and operations from the file in case we have
213 * as file, or get the default values for them in case we
214 * don't have a struct file available. Damn nfsd..
217 mapping
= file
->f_mapping
;
220 mapping
= dentry
->d_inode
->i_mapping
;
221 fop
= dentry
->d_inode
->i_fop
;
224 if (!fop
|| !fop
->fsync
) {
229 ret
= filemap_write_and_wait_range(mapping
, start
, end
);
232 * We need to protect against concurrent writers, which could cause
233 * livelocks in fsync_buffers_list().
235 mutex_lock(&mapping
->host
->i_mutex
);
236 err
= fop
->fsync(file
, dentry
, datasync
);
239 mutex_unlock(&mapping
->host
->i_mutex
);
244 EXPORT_SYMBOL(vfs_fsync_range
);
247 * vfs_fsync - perform a fsync or fdatasync on a file
248 * @file: file to sync
249 * @dentry: dentry of @file
250 * @datasync: only perform a fdatasync operation
252 * Write back data and metadata for @file to disk. If @datasync is
253 * set only metadata needed to access modified file data is written.
255 * In case this function is called from nfsd @file may be %NULL and
256 * only @dentry is set. This can only happen when the filesystem
257 * implements the export_operations API.
259 int vfs_fsync(struct file
*file
, struct dentry
*dentry
, int datasync
)
261 return vfs_fsync_range(file
, dentry
, 0, LLONG_MAX
, datasync
);
263 EXPORT_SYMBOL(vfs_fsync
);
265 static int do_fsync(unsigned int fd
, int datasync
)
272 ret
= vfs_fsync(file
, file
->f_path
.dentry
, datasync
);
278 SYSCALL_DEFINE1(fsync
, unsigned int, fd
)
280 return do_fsync(fd
, 0);
283 SYSCALL_DEFINE1(fdatasync
, unsigned int, fd
)
285 return do_fsync(fd
, 1);
289 * generic_write_sync - perform syncing after a write if file / inode is sync
290 * @file: file to which the write happened
291 * @pos: offset where the write started
292 * @count: length of the write
294 * This is just a simple wrapper about our general syncing function.
296 int generic_write_sync(struct file
*file
, loff_t pos
, loff_t count
)
298 if (!(file
->f_flags
& O_DSYNC
) && !IS_SYNC(file
->f_mapping
->host
))
300 return vfs_fsync_range(file
, file
->f_path
.dentry
, pos
,
302 (file
->f_flags
& __O_SYNC
) ? 0 : 1);
304 EXPORT_SYMBOL(generic_write_sync
);
307 * sys_sync_file_range() permits finely controlled syncing over a segment of
308 * a file in the range offset .. (offset+nbytes-1) inclusive. If nbytes is
309 * zero then sys_sync_file_range() will operate from offset out to EOF.
313 * SYNC_FILE_RANGE_WAIT_BEFORE: wait upon writeout of all pages in the range
314 * before performing the write.
316 * SYNC_FILE_RANGE_WRITE: initiate writeout of all those dirty pages in the
317 * range which are not presently under writeback. Note that this may block for
318 * significant periods due to exhaustion of disk request structures.
320 * SYNC_FILE_RANGE_WAIT_AFTER: wait upon writeout of all pages in the range
321 * after performing the write.
323 * Useful combinations of the flag bits are:
325 * SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE: ensures that all pages
326 * in the range which were dirty on entry to sys_sync_file_range() are placed
327 * under writeout. This is a start-write-for-data-integrity operation.
329 * SYNC_FILE_RANGE_WRITE: start writeout of all dirty pages in the range which
330 * are not presently under writeout. This is an asynchronous flush-to-disk
331 * operation. Not suitable for data integrity operations.
333 * SYNC_FILE_RANGE_WAIT_BEFORE (or SYNC_FILE_RANGE_WAIT_AFTER): wait for
334 * completion of writeout of all pages in the range. This will be used after an
335 * earlier SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE operation to wait
336 * for that operation to complete and to return the result.
338 * SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE|SYNC_FILE_RANGE_WAIT_AFTER:
339 * a traditional sync() operation. This is a write-for-data-integrity operation
340 * which will ensure that all pages in the range which were dirty on entry to
341 * sys_sync_file_range() are committed to disk.
344 * SYNC_FILE_RANGE_WAIT_BEFORE and SYNC_FILE_RANGE_WAIT_AFTER will detect any
345 * I/O errors or ENOSPC conditions and will return those to the caller, after
346 * clearing the EIO and ENOSPC flags in the address_space.
348 * It should be noted that none of these operations write out the file's
349 * metadata. So unless the application is strictly performing overwrites of
350 * already-instantiated disk blocks, there are no guarantees here that the data
351 * will be available after a crash.
353 SYSCALL_DEFINE(sync_file_range
)(int fd
, loff_t offset
, loff_t nbytes
,
358 loff_t endbyte
; /* inclusive */
363 if (flags
& ~VALID_FLAGS
)
366 endbyte
= offset
+ nbytes
;
370 if ((s64
)endbyte
< 0)
372 if (endbyte
< offset
)
375 if (sizeof(pgoff_t
) == 4) {
376 if (offset
>= (0x100000000ULL
<< PAGE_CACHE_SHIFT
)) {
378 * The range starts outside a 32 bit machine's
379 * pagecache addressing capabilities. Let it "succeed"
384 if (endbyte
>= (0x100000000ULL
<< PAGE_CACHE_SHIFT
)) {
395 endbyte
--; /* inclusive */
398 file
= fget_light(fd
, &fput_needed
);
402 i_mode
= file
->f_path
.dentry
->d_inode
->i_mode
;
404 if (!S_ISREG(i_mode
) && !S_ISBLK(i_mode
) && !S_ISDIR(i_mode
) &&
408 ret
= do_sync_mapping_range(file
->f_mapping
, offset
, endbyte
, flags
);
410 fput_light(file
, fput_needed
);
414 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
415 asmlinkage
long SyS_sync_file_range(long fd
, loff_t offset
, loff_t nbytes
,
418 return SYSC_sync_file_range((int) fd
, offset
, nbytes
,
419 (unsigned int) flags
);
421 SYSCALL_ALIAS(sys_sync_file_range
, SyS_sync_file_range
);
424 /* It would be nice if people remember that not all the world's an i386
425 when they introduce new system calls */
426 SYSCALL_DEFINE(sync_file_range2
)(int fd
, unsigned int flags
,
427 loff_t offset
, loff_t nbytes
)
429 return sys_sync_file_range(fd
, offset
, nbytes
, flags
);
431 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
432 asmlinkage
long SyS_sync_file_range2(long fd
, long flags
,
433 loff_t offset
, loff_t nbytes
)
435 return SYSC_sync_file_range2((int) fd
, (unsigned int) flags
,
438 SYSCALL_ALIAS(sys_sync_file_range2
, SyS_sync_file_range2
);
442 * `endbyte' is inclusive
444 int do_sync_mapping_range(struct address_space
*mapping
, loff_t offset
,
445 loff_t endbyte
, unsigned int flags
)
455 if (flags
& SYNC_FILE_RANGE_WAIT_BEFORE
) {
456 ret
= wait_on_page_writeback_range(mapping
,
457 offset
>> PAGE_CACHE_SHIFT
,
458 endbyte
>> PAGE_CACHE_SHIFT
);
463 if (flags
& SYNC_FILE_RANGE_WRITE
) {
464 ret
= __filemap_fdatawrite_range(mapping
, offset
, endbyte
,
470 if (flags
& SYNC_FILE_RANGE_WAIT_AFTER
) {
471 ret
= wait_on_page_writeback_range(mapping
,
472 offset
>> PAGE_CACHE_SHIFT
,
473 endbyte
>> PAGE_CACHE_SHIFT
);
478 EXPORT_SYMBOL_GPL(do_sync_mapping_range
);