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(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
);
81 * Sync all the data for all the filesystems (called by sys_sync() and
84 * This operation is careful to avoid the livelock which could easily happen
85 * if two or more filesystems are being continuously dirtied. s_need_sync
86 * is used only here. We set it against all filesystems and then clear it as
87 * we sync them. So redirtied filesystems are skipped.
89 * But if process A is currently running sync_filesystems and then process B
90 * calls sync_filesystems as well, process B will set all the s_need_sync
91 * flags again, which will cause process A to resync everything. Fix that with
94 static void sync_filesystems(int wait
)
96 struct super_block
*sb
;
97 static DEFINE_MUTEX(mutex
);
99 mutex_lock(&mutex
); /* Could be down_interruptible */
101 list_for_each_entry(sb
, &super_blocks
, s_list
)
102 if (!list_empty(&sb
->s_instances
))
106 list_for_each_entry(sb
, &super_blocks
, s_list
) {
107 if (list_empty(&sb
->s_instances
))
109 if (!sb
->s_need_sync
)
113 spin_unlock(&sb_lock
);
115 down_read(&sb
->s_umount
);
116 if (!(sb
->s_flags
& MS_RDONLY
) && sb
->s_root
&& sb
->s_bdi
)
117 __sync_filesystem(sb
, wait
);
118 up_read(&sb
->s_umount
);
120 /* restart only when sb is no longer on the list */
122 if (__put_super_and_need_restart(sb
))
125 spin_unlock(&sb_lock
);
126 mutex_unlock(&mutex
);
130 * sync everything. Start out by waking pdflush, because that writes back
131 * all queues in parallel.
133 SYSCALL_DEFINE0(sync
)
135 wakeup_flusher_threads(0);
138 if (unlikely(laptop_mode
))
139 laptop_sync_completion();
143 static void do_sync_work(struct work_struct
*work
)
146 * Sync twice to reduce the possibility we skipped some inodes / pages
147 * because they were temporarily locked
151 printk("Emergency Sync complete\n");
155 void emergency_sync(void)
157 struct work_struct
*work
;
159 work
= kmalloc(sizeof(*work
), GFP_ATOMIC
);
161 INIT_WORK(work
, do_sync_work
);
167 * Generic function to fsync a file.
169 * filp may be NULL if called via the msync of a vma.
171 int file_fsync(struct file
*filp
, struct dentry
*dentry
, int datasync
)
173 struct inode
* inode
= dentry
->d_inode
;
174 struct super_block
* sb
;
177 /* sync the inode to buffers */
178 ret
= write_inode_now(inode
, 0);
180 /* sync the superblock to buffers */
182 if (sb
->s_dirt
&& sb
->s_op
->write_super
)
183 sb
->s_op
->write_super(sb
);
185 /* .. finally sync the buffers to disk */
186 err
= sync_blockdev(sb
->s_bdev
);
191 EXPORT_SYMBOL(file_fsync
);
194 * vfs_fsync_range - helper to sync a range of data & metadata to disk
195 * @file: file to sync
196 * @dentry: dentry of @file
197 * @start: offset in bytes of the beginning of data range to sync
198 * @end: offset in bytes of the end of data range (inclusive)
199 * @datasync: perform only datasync
201 * Write back data in range @start..@end and metadata for @file to disk. If
202 * @datasync is set only metadata needed to access modified file data is
205 * In case this function is called from nfsd @file may be %NULL and
206 * only @dentry is set. This can only happen when the filesystem
207 * implements the export_operations API.
209 int vfs_fsync_range(struct file
*file
, struct dentry
*dentry
, loff_t start
,
210 loff_t end
, int datasync
)
212 const struct file_operations
*fop
;
213 struct address_space
*mapping
;
217 * Get mapping and operations from the file in case we have
218 * as file, or get the default values for them in case we
219 * don't have a struct file available. Damn nfsd..
222 mapping
= file
->f_mapping
;
225 mapping
= dentry
->d_inode
->i_mapping
;
226 fop
= dentry
->d_inode
->i_fop
;
229 if (!fop
|| !fop
->fsync
) {
234 ret
= filemap_write_and_wait_range(mapping
, start
, end
);
237 * We need to protect against concurrent writers, which could cause
238 * livelocks in fsync_buffers_list().
240 mutex_lock(&mapping
->host
->i_mutex
);
241 err
= fop
->fsync(file
, dentry
, datasync
);
244 mutex_unlock(&mapping
->host
->i_mutex
);
249 EXPORT_SYMBOL(vfs_fsync_range
);
252 * vfs_fsync - perform a fsync or fdatasync on a file
253 * @file: file to sync
254 * @dentry: dentry of @file
255 * @datasync: only perform a fdatasync operation
257 * Write back data and metadata for @file to disk. If @datasync is
258 * set only metadata needed to access modified file data is written.
260 * In case this function is called from nfsd @file may be %NULL and
261 * only @dentry is set. This can only happen when the filesystem
262 * implements the export_operations API.
264 int vfs_fsync(struct file
*file
, struct dentry
*dentry
, int datasync
)
266 return vfs_fsync_range(file
, dentry
, 0, LLONG_MAX
, datasync
);
268 EXPORT_SYMBOL(vfs_fsync
);
270 static int do_fsync(unsigned int fd
, int datasync
)
277 ret
= vfs_fsync(file
, file
->f_path
.dentry
, datasync
);
283 SYSCALL_DEFINE1(fsync
, unsigned int, fd
)
285 return do_fsync(fd
, 0);
288 SYSCALL_DEFINE1(fdatasync
, unsigned int, fd
)
290 return do_fsync(fd
, 1);
294 * generic_write_sync - perform syncing after a write if file / inode is sync
295 * @file: file to which the write happened
296 * @pos: offset where the write started
297 * @count: length of the write
299 * This is just a simple wrapper about our general syncing function.
301 int generic_write_sync(struct file
*file
, loff_t pos
, loff_t count
)
303 if (!(file
->f_flags
& O_DSYNC
) && !IS_SYNC(file
->f_mapping
->host
))
305 return vfs_fsync_range(file
, file
->f_path
.dentry
, pos
,
307 (file
->f_flags
& __O_SYNC
) ? 0 : 1);
309 EXPORT_SYMBOL(generic_write_sync
);
312 * sys_sync_file_range() permits finely controlled syncing over a segment of
313 * a file in the range offset .. (offset+nbytes-1) inclusive. If nbytes is
314 * zero then sys_sync_file_range() will operate from offset out to EOF.
318 * SYNC_FILE_RANGE_WAIT_BEFORE: wait upon writeout of all pages in the range
319 * before performing the write.
321 * SYNC_FILE_RANGE_WRITE: initiate writeout of all those dirty pages in the
322 * range which are not presently under writeback. Note that this may block for
323 * significant periods due to exhaustion of disk request structures.
325 * SYNC_FILE_RANGE_WAIT_AFTER: wait upon writeout of all pages in the range
326 * after performing the write.
328 * Useful combinations of the flag bits are:
330 * SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE: ensures that all pages
331 * in the range which were dirty on entry to sys_sync_file_range() are placed
332 * under writeout. This is a start-write-for-data-integrity operation.
334 * SYNC_FILE_RANGE_WRITE: start writeout of all dirty pages in the range which
335 * are not presently under writeout. This is an asynchronous flush-to-disk
336 * operation. Not suitable for data integrity operations.
338 * SYNC_FILE_RANGE_WAIT_BEFORE (or SYNC_FILE_RANGE_WAIT_AFTER): wait for
339 * completion of writeout of all pages in the range. This will be used after an
340 * earlier SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE operation to wait
341 * for that operation to complete and to return the result.
343 * SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE|SYNC_FILE_RANGE_WAIT_AFTER:
344 * a traditional sync() operation. This is a write-for-data-integrity operation
345 * which will ensure that all pages in the range which were dirty on entry to
346 * sys_sync_file_range() are committed to disk.
349 * SYNC_FILE_RANGE_WAIT_BEFORE and SYNC_FILE_RANGE_WAIT_AFTER will detect any
350 * I/O errors or ENOSPC conditions and will return those to the caller, after
351 * clearing the EIO and ENOSPC flags in the address_space.
353 * It should be noted that none of these operations write out the file's
354 * metadata. So unless the application is strictly performing overwrites of
355 * already-instantiated disk blocks, there are no guarantees here that the data
356 * will be available after a crash.
358 SYSCALL_DEFINE(sync_file_range
)(int fd
, loff_t offset
, loff_t nbytes
,
363 struct address_space
*mapping
;
364 loff_t endbyte
; /* inclusive */
369 if (flags
& ~VALID_FLAGS
)
372 endbyte
= offset
+ nbytes
;
376 if ((s64
)endbyte
< 0)
378 if (endbyte
< offset
)
381 if (sizeof(pgoff_t
) == 4) {
382 if (offset
>= (0x100000000ULL
<< PAGE_CACHE_SHIFT
)) {
384 * The range starts outside a 32 bit machine's
385 * pagecache addressing capabilities. Let it "succeed"
390 if (endbyte
>= (0x100000000ULL
<< PAGE_CACHE_SHIFT
)) {
401 endbyte
--; /* inclusive */
404 file
= fget_light(fd
, &fput_needed
);
408 i_mode
= file
->f_path
.dentry
->d_inode
->i_mode
;
410 if (!S_ISREG(i_mode
) && !S_ISBLK(i_mode
) && !S_ISDIR(i_mode
) &&
414 mapping
= file
->f_mapping
;
421 if (flags
& SYNC_FILE_RANGE_WAIT_BEFORE
) {
422 ret
= filemap_fdatawait_range(mapping
, offset
, endbyte
);
427 if (flags
& SYNC_FILE_RANGE_WRITE
) {
428 ret
= filemap_fdatawrite_range(mapping
, offset
, endbyte
);
433 if (flags
& SYNC_FILE_RANGE_WAIT_AFTER
)
434 ret
= filemap_fdatawait_range(mapping
, offset
, endbyte
);
437 fput_light(file
, fput_needed
);
441 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
442 asmlinkage
long SyS_sync_file_range(long fd
, loff_t offset
, loff_t nbytes
,
445 return SYSC_sync_file_range((int) fd
, offset
, nbytes
,
446 (unsigned int) flags
);
448 SYSCALL_ALIAS(sys_sync_file_range
, SyS_sync_file_range
);
451 /* It would be nice if people remember that not all the world's an i386
452 when they introduce new system calls */
453 SYSCALL_DEFINE(sync_file_range2
)(int fd
, unsigned int flags
,
454 loff_t offset
, loff_t nbytes
)
456 return sys_sync_file_range(fd
, offset
, nbytes
, flags
);
458 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
459 asmlinkage
long SyS_sync_file_range2(long fd
, long flags
,
460 loff_t offset
, loff_t nbytes
)
462 return SYSC_sync_file_range2((int) fd
, (unsigned int) flags
,
465 SYSCALL_ALIAS(sys_sync_file_range2
, SyS_sync_file_range2
);