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 sync_blkdev_one_sb(struct super_block *sb, void *arg)
90 if (!(sb->s_flags & MS_RDONLY))
91 __sync_blockdev(sb->s_bdev, *(int *)arg);
95 * sync everything. Start out by waking pdflush, because that writes back
96 * all queues in parallel.
100 int nowait = 0, wait = 1;
102 wakeup_flusher_threads(0, WB_REASON_SYNC);
103 iterate_supers(writeback_inodes_one_sb, NULL);
104 iterate_supers(sync_fs_one_sb, &nowait);
105 iterate_supers(sync_blkdev_one_sb, &nowait);
106 iterate_supers(sync_inodes_one_sb, NULL);
107 iterate_supers(sync_fs_one_sb, &wait);
108 iterate_supers(sync_blkdev_one_sb, &wait);
109 if (unlikely(laptop_mode))
110 laptop_sync_completion();
114 static void do_sync_work(struct work_struct *work)
119 * Sync twice to reduce the possibility we skipped some inodes / pages
120 * because they were temporarily locked
122 iterate_supers(sync_inodes_one_sb, &nowait);
123 iterate_supers(sync_fs_one_sb, &nowait);
124 iterate_supers(sync_blkdev_one_sb, &nowait);
125 iterate_supers(sync_inodes_one_sb, &nowait);
126 iterate_supers(sync_fs_one_sb, &nowait);
127 iterate_supers(sync_blkdev_one_sb, &nowait);
128 printk("Emergency Sync complete\n");
132 void emergency_sync(void)
134 struct work_struct *work;
136 work = kmalloc(sizeof(*work), GFP_ATOMIC);
138 INIT_WORK(work, do_sync_work);
144 * sync a single super
146 SYSCALL_DEFINE1(syncfs, int, fd)
149 struct super_block *sb;
153 file = fget_light(fd, &fput_needed);
156 sb = file->f_dentry->d_sb;
158 down_read(&sb->s_umount);
159 ret = sync_filesystem(sb);
160 up_read(&sb->s_umount);
162 fput_light(file, fput_needed);
167 * vfs_fsync_range - helper to sync a range of data & metadata to disk
168 * @file: file to sync
169 * @start: offset in bytes of the beginning of data range to sync
170 * @end: offset in bytes of the end of data range (inclusive)
171 * @datasync: perform only datasync
173 * Write back data in range @start..@end and metadata for @file to disk. If
174 * @datasync is set only metadata needed to access modified file data is
177 int vfs_fsync_range(struct file *file, loff_t start, loff_t end, int datasync)
179 if (!file->f_op || !file->f_op->fsync)
181 return file->f_op->fsync(file, start, end, datasync);
183 EXPORT_SYMBOL(vfs_fsync_range);
186 * vfs_fsync - perform a fsync or fdatasync on a file
187 * @file: file to sync
188 * @datasync: only perform a fdatasync operation
190 * Write back data and metadata for @file to disk. If @datasync is
191 * set only metadata needed to access modified file data is written.
193 int vfs_fsync(struct file *file, int datasync)
195 return vfs_fsync_range(file, 0, LLONG_MAX, datasync);
197 EXPORT_SYMBOL(vfs_fsync);
199 static int do_fsync(unsigned int fd, int datasync)
205 file = fget_light(fd, &fput_needed);
207 ret = vfs_fsync(file, datasync);
208 fput_light(file, fput_needed);
213 SYSCALL_DEFINE1(fsync, unsigned int, fd)
215 return do_fsync(fd, 0);
218 SYSCALL_DEFINE1(fdatasync, unsigned int, fd)
220 return do_fsync(fd, 1);
224 * generic_write_sync - perform syncing after a write if file / inode is sync
225 * @file: file to which the write happened
226 * @pos: offset where the write started
227 * @count: length of the write
229 * This is just a simple wrapper about our general syncing function.
231 int generic_write_sync(struct file *file, loff_t pos, loff_t count)
233 if (!(file->f_flags & O_DSYNC) && !IS_SYNC(file->f_mapping->host))
235 return vfs_fsync_range(file, pos, pos + count - 1,
236 (file->f_flags & __O_SYNC) ? 0 : 1);
238 EXPORT_SYMBOL(generic_write_sync);
241 * sys_sync_file_range() permits finely controlled syncing over a segment of
242 * a file in the range offset .. (offset+nbytes-1) inclusive. If nbytes is
243 * zero then sys_sync_file_range() will operate from offset out to EOF.
247 * SYNC_FILE_RANGE_WAIT_BEFORE: wait upon writeout of all pages in the range
248 * before performing the write.
250 * SYNC_FILE_RANGE_WRITE: initiate writeout of all those dirty pages in the
251 * range which are not presently under writeback. Note that this may block for
252 * significant periods due to exhaustion of disk request structures.
254 * SYNC_FILE_RANGE_WAIT_AFTER: wait upon writeout of all pages in the range
255 * after performing the write.
257 * Useful combinations of the flag bits are:
259 * SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE: ensures that all pages
260 * in the range which were dirty on entry to sys_sync_file_range() are placed
261 * under writeout. This is a start-write-for-data-integrity operation.
263 * SYNC_FILE_RANGE_WRITE: start writeout of all dirty pages in the range which
264 * are not presently under writeout. This is an asynchronous flush-to-disk
265 * operation. Not suitable for data integrity operations.
267 * SYNC_FILE_RANGE_WAIT_BEFORE (or SYNC_FILE_RANGE_WAIT_AFTER): wait for
268 * completion of writeout of all pages in the range. This will be used after an
269 * earlier SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE operation to wait
270 * for that operation to complete and to return the result.
272 * SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE|SYNC_FILE_RANGE_WAIT_AFTER:
273 * a traditional sync() operation. This is a write-for-data-integrity operation
274 * which will ensure that all pages in the range which were dirty on entry to
275 * sys_sync_file_range() are committed to disk.
278 * SYNC_FILE_RANGE_WAIT_BEFORE and SYNC_FILE_RANGE_WAIT_AFTER will detect any
279 * I/O errors or ENOSPC conditions and will return those to the caller, after
280 * clearing the EIO and ENOSPC flags in the address_space.
282 * It should be noted that none of these operations write out the file's
283 * metadata. So unless the application is strictly performing overwrites of
284 * already-instantiated disk blocks, there are no guarantees here that the data
285 * will be available after a crash.
287 SYSCALL_DEFINE(sync_file_range)(int fd, loff_t offset, loff_t nbytes,
292 struct address_space *mapping;
293 loff_t endbyte; /* inclusive */
298 if (flags & ~VALID_FLAGS)
301 endbyte = offset + nbytes;
305 if ((s64)endbyte < 0)
307 if (endbyte < offset)
310 if (sizeof(pgoff_t) == 4) {
311 if (offset >= (0x100000000ULL << PAGE_CACHE_SHIFT)) {
313 * The range starts outside a 32 bit machine's
314 * pagecache addressing capabilities. Let it "succeed"
319 if (endbyte >= (0x100000000ULL << PAGE_CACHE_SHIFT)) {
330 endbyte--; /* inclusive */
333 file = fget_light(fd, &fput_needed);
337 i_mode = file->f_path.dentry->d_inode->i_mode;
339 if (!S_ISREG(i_mode) && !S_ISBLK(i_mode) && !S_ISDIR(i_mode) &&
343 mapping = file->f_mapping;
350 if (flags & SYNC_FILE_RANGE_WAIT_BEFORE) {
351 ret = filemap_fdatawait_range(mapping, offset, endbyte);
356 if (flags & SYNC_FILE_RANGE_WRITE) {
357 ret = filemap_fdatawrite_range(mapping, offset, endbyte);
362 if (flags & SYNC_FILE_RANGE_WAIT_AFTER)
363 ret = filemap_fdatawait_range(mapping, offset, endbyte);
366 fput_light(file, fput_needed);
370 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
371 asmlinkage long SyS_sync_file_range(long fd, loff_t offset, loff_t nbytes,
374 return SYSC_sync_file_range((int) fd, offset, nbytes,
375 (unsigned int) flags);
377 SYSCALL_ALIAS(sys_sync_file_range, SyS_sync_file_range);
380 /* It would be nice if people remember that not all the world's an i386
381 when they introduce new system calls */
382 SYSCALL_DEFINE(sync_file_range2)(int fd, unsigned int flags,
383 loff_t offset, loff_t nbytes)
385 return sys_sync_file_range(fd, offset, nbytes, flags);
387 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
388 asmlinkage long SyS_sync_file_range2(long fd, long flags,
389 loff_t offset, loff_t nbytes)
391 return SYSC_sync_file_range2((int) fd, (unsigned int) flags,
394 SYSCALL_ALIAS(sys_sync_file_range2, SyS_sync_file_range2);