2 * High-level sync()-related operations
5 #include <linux/kernel.h>
6 #include <linux/file.h>
8 #include <linux/slab.h>
9 #include <linux/export.h>
10 #include <linux/namei.h>
11 #include <linux/sched.h>
12 #include <linux/writeback.h>
13 #include <linux/syscalls.h>
14 #include <linux/linkage.h>
15 #include <linux/pagemap.h>
16 #include <linux/quotaops.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)
35 writeback_inodes_sb(sb, WB_REASON_SYNC);
37 if (sb->s_op->sync_fs)
38 sb->s_op->sync_fs(sb, wait);
39 return __sync_blockdev(sb->s_bdev, wait);
43 * Write out and wait upon all dirty data associated with this
44 * superblock. Filesystem data as well as the underlying block
45 * device. Takes the superblock lock.
47 int sync_filesystem(struct super_block *sb)
52 * We need to be protected against the filesystem going from
53 * r/o to r/w or vice versa.
55 WARN_ON(!rwsem_is_locked(&sb->s_umount));
58 * No point in syncing out anything if the filesystem is read-only.
60 if (sb->s_flags & MS_RDONLY)
63 ret = __sync_filesystem(sb, 0);
66 return __sync_filesystem(sb, 1);
68 EXPORT_SYMBOL_GPL(sync_filesystem);
70 static void sync_inodes_one_sb(struct super_block *sb, void *arg)
72 if (!(sb->s_flags & MS_RDONLY))
76 static void writeback_inodes_one_sb(struct super_block *sb, void *arg)
78 if (!(sb->s_flags & MS_RDONLY))
79 writeback_inodes_sb(sb, WB_REASON_SYNC);
82 static void sync_fs_one_sb(struct super_block *sb, void *arg)
84 if (!(sb->s_flags & MS_RDONLY) && sb->s_op->sync_fs)
85 sb->s_op->sync_fs(sb, *(int *)arg);
88 static void flush_one_bdev(struct block_device *bdev, void *arg)
90 __sync_blockdev(bdev, 0);
93 static void sync_one_bdev(struct block_device *bdev, void *arg)
99 * sync everything. Start out by waking pdflush, because that writes back
100 * all queues in parallel.
102 SYSCALL_DEFINE0(sync)
104 int nowait = 0, wait = 1;
106 wakeup_flusher_threads(0, WB_REASON_SYNC);
107 iterate_supers(writeback_inodes_one_sb, NULL);
108 iterate_supers(sync_fs_one_sb, &nowait);
109 iterate_bdevs(flush_one_bdev, NULL);
110 iterate_supers(sync_inodes_one_sb, NULL);
111 iterate_supers(sync_fs_one_sb, &wait);
112 iterate_bdevs(sync_one_bdev, NULL);
113 if (unlikely(laptop_mode))
114 laptop_sync_completion();
118 static void do_sync_work(struct work_struct *work)
123 * Sync twice to reduce the possibility we skipped some inodes / pages
124 * because they were temporarily locked
126 iterate_supers(sync_inodes_one_sb, &nowait);
127 iterate_supers(sync_fs_one_sb, &nowait);
128 iterate_bdevs(flush_one_bdev, NULL);
129 iterate_supers(sync_inodes_one_sb, &nowait);
130 iterate_supers(sync_fs_one_sb, &nowait);
131 iterate_bdevs(flush_one_bdev, NULL);
132 printk("Emergency Sync complete\n");
136 void emergency_sync(void)
138 struct work_struct *work;
140 work = kmalloc(sizeof(*work), GFP_ATOMIC);
142 INIT_WORK(work, do_sync_work);
148 * sync a single super
150 SYSCALL_DEFINE1(syncfs, int, fd)
153 struct super_block *sb;
157 file = fget_light(fd, &fput_needed);
160 sb = file->f_dentry->d_sb;
162 down_read(&sb->s_umount);
163 ret = sync_filesystem(sb);
164 up_read(&sb->s_umount);
166 fput_light(file, fput_needed);
171 * vfs_fsync_range - helper to sync a range of data & metadata to disk
172 * @file: file to sync
173 * @start: offset in bytes of the beginning of data range to sync
174 * @end: offset in bytes of the end of data range (inclusive)
175 * @datasync: perform only datasync
177 * Write back data in range @start..@end and metadata for @file to disk. If
178 * @datasync is set only metadata needed to access modified file data is
181 int vfs_fsync_range(struct file *file, loff_t start, loff_t end, int datasync)
183 if (!file->f_op || !file->f_op->fsync)
185 return file->f_op->fsync(file, start, end, datasync);
187 EXPORT_SYMBOL(vfs_fsync_range);
190 * vfs_fsync - perform a fsync or fdatasync on a file
191 * @file: file to sync
192 * @datasync: only perform a fdatasync operation
194 * Write back data and metadata for @file to disk. If @datasync is
195 * set only metadata needed to access modified file data is written.
197 int vfs_fsync(struct file *file, int datasync)
199 return vfs_fsync_range(file, 0, LLONG_MAX, datasync);
201 EXPORT_SYMBOL(vfs_fsync);
203 static int do_fsync(unsigned int fd, int datasync)
209 file = fget_light(fd, &fput_needed);
211 ret = vfs_fsync(file, datasync);
212 fput_light(file, fput_needed);
217 SYSCALL_DEFINE1(fsync, unsigned int, fd)
219 return do_fsync(fd, 0);
222 SYSCALL_DEFINE1(fdatasync, unsigned int, fd)
224 return do_fsync(fd, 1);
228 * generic_write_sync - perform syncing after a write if file / inode is sync
229 * @file: file to which the write happened
230 * @pos: offset where the write started
231 * @count: length of the write
233 * This is just a simple wrapper about our general syncing function.
235 int generic_write_sync(struct file *file, loff_t pos, loff_t count)
237 if (!(file->f_flags & O_DSYNC) && !IS_SYNC(file->f_mapping->host))
239 return vfs_fsync_range(file, pos, pos + count - 1,
240 (file->f_flags & __O_SYNC) ? 0 : 1);
242 EXPORT_SYMBOL(generic_write_sync);
245 * sys_sync_file_range() permits finely controlled syncing over a segment of
246 * a file in the range offset .. (offset+nbytes-1) inclusive. If nbytes is
247 * zero then sys_sync_file_range() will operate from offset out to EOF.
251 * SYNC_FILE_RANGE_WAIT_BEFORE: wait upon writeout of all pages in the range
252 * before performing the write.
254 * SYNC_FILE_RANGE_WRITE: initiate writeout of all those dirty pages in the
255 * range which are not presently under writeback. Note that this may block for
256 * significant periods due to exhaustion of disk request structures.
258 * SYNC_FILE_RANGE_WAIT_AFTER: wait upon writeout of all pages in the range
259 * after performing the write.
261 * Useful combinations of the flag bits are:
263 * SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE: ensures that all pages
264 * in the range which were dirty on entry to sys_sync_file_range() are placed
265 * under writeout. This is a start-write-for-data-integrity operation.
267 * SYNC_FILE_RANGE_WRITE: start writeout of all dirty pages in the range which
268 * are not presently under writeout. This is an asynchronous flush-to-disk
269 * operation. Not suitable for data integrity operations.
271 * SYNC_FILE_RANGE_WAIT_BEFORE (or SYNC_FILE_RANGE_WAIT_AFTER): wait for
272 * completion of writeout of all pages in the range. This will be used after an
273 * earlier SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE operation to wait
274 * for that operation to complete and to return the result.
276 * SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE|SYNC_FILE_RANGE_WAIT_AFTER:
277 * a traditional sync() operation. This is a write-for-data-integrity operation
278 * which will ensure that all pages in the range which were dirty on entry to
279 * sys_sync_file_range() are committed to disk.
282 * SYNC_FILE_RANGE_WAIT_BEFORE and SYNC_FILE_RANGE_WAIT_AFTER will detect any
283 * I/O errors or ENOSPC conditions and will return those to the caller, after
284 * clearing the EIO and ENOSPC flags in the address_space.
286 * It should be noted that none of these operations write out the file's
287 * metadata. So unless the application is strictly performing overwrites of
288 * already-instantiated disk blocks, there are no guarantees here that the data
289 * will be available after a crash.
291 SYSCALL_DEFINE(sync_file_range)(int fd, loff_t offset, loff_t nbytes,
296 struct address_space *mapping;
297 loff_t endbyte; /* inclusive */
302 if (flags & ~VALID_FLAGS)
305 endbyte = offset + nbytes;
309 if ((s64)endbyte < 0)
311 if (endbyte < offset)
314 if (sizeof(pgoff_t) == 4) {
315 if (offset >= (0x100000000ULL << PAGE_CACHE_SHIFT)) {
317 * The range starts outside a 32 bit machine's
318 * pagecache addressing capabilities. Let it "succeed"
323 if (endbyte >= (0x100000000ULL << PAGE_CACHE_SHIFT)) {
334 endbyte--; /* inclusive */
337 file = fget_light(fd, &fput_needed);
341 i_mode = file->f_path.dentry->d_inode->i_mode;
343 if (!S_ISREG(i_mode) && !S_ISBLK(i_mode) && !S_ISDIR(i_mode) &&
347 mapping = file->f_mapping;
354 if (flags & SYNC_FILE_RANGE_WAIT_BEFORE) {
355 ret = filemap_fdatawait_range(mapping, offset, endbyte);
360 if (flags & SYNC_FILE_RANGE_WRITE) {
361 ret = filemap_fdatawrite_range(mapping, offset, endbyte);
366 if (flags & SYNC_FILE_RANGE_WAIT_AFTER)
367 ret = filemap_fdatawait_range(mapping, offset, endbyte);
370 fput_light(file, fput_needed);
374 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
375 asmlinkage long SyS_sync_file_range(long fd, loff_t offset, loff_t nbytes,
378 return SYSC_sync_file_range((int) fd, offset, nbytes,
379 (unsigned int) flags);
381 SYSCALL_ALIAS(sys_sync_file_range, SyS_sync_file_range);
384 /* It would be nice if people remember that not all the world's an i386
385 when they introduce new system calls */
386 SYSCALL_DEFINE(sync_file_range2)(int fd, unsigned int flags,
387 loff_t offset, loff_t nbytes)
389 return sys_sync_file_range(fd, offset, nbytes, flags);
391 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
392 asmlinkage long SyS_sync_file_range2(long fd, long flags,
393 loff_t offset, loff_t nbytes)
395 return SYSC_sync_file_range2((int) fd, (unsigned int) flags,
398 SYSCALL_ALIAS(sys_sync_file_range2, SyS_sync_file_range2);