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_one_sb(struct super_block *sb, void *arg)
72 if (!(sb->s_flags & MS_RDONLY))
73 __sync_filesystem(sb, *(int *)arg);
76 * Sync all the data for all the filesystems (called by sys_sync() and
79 static void sync_filesystems(int wait)
81 iterate_supers(sync_one_sb, &wait);
85 * sync everything. Start out by waking pdflush, because that writes back
86 * all queues in parallel.
90 wakeup_flusher_threads(0, WB_REASON_SYNC);
93 if (unlikely(laptop_mode))
94 laptop_sync_completion();
98 static void do_sync_work(struct work_struct *work)
101 * Sync twice to reduce the possibility we skipped some inodes / pages
102 * because they were temporarily locked
106 printk("Emergency Sync complete\n");
110 void emergency_sync(void)
112 struct work_struct *work;
114 work = kmalloc(sizeof(*work), GFP_ATOMIC);
116 INIT_WORK(work, do_sync_work);
122 * sync a single super
124 SYSCALL_DEFINE1(syncfs, int, fd)
127 struct super_block *sb;
131 file = fget_light(fd, &fput_needed);
134 sb = file->f_dentry->d_sb;
136 down_read(&sb->s_umount);
137 ret = sync_filesystem(sb);
138 up_read(&sb->s_umount);
140 fput_light(file, fput_needed);
145 * vfs_fsync_range - helper to sync a range of data & metadata to disk
146 * @file: file to sync
147 * @start: offset in bytes of the beginning of data range to sync
148 * @end: offset in bytes of the end of data range (inclusive)
149 * @datasync: perform only datasync
151 * Write back data in range @start..@end and metadata for @file to disk. If
152 * @datasync is set only metadata needed to access modified file data is
155 int vfs_fsync_range(struct file *file, loff_t start, loff_t end, int datasync)
157 if (!file->f_op || !file->f_op->fsync)
159 return file->f_op->fsync(file, start, end, datasync);
161 EXPORT_SYMBOL(vfs_fsync_range);
164 * vfs_fsync - perform a fsync or fdatasync on a file
165 * @file: file to sync
166 * @datasync: only perform a fdatasync operation
168 * Write back data and metadata for @file to disk. If @datasync is
169 * set only metadata needed to access modified file data is written.
171 int vfs_fsync(struct file *file, int datasync)
173 return vfs_fsync_range(file, 0, LLONG_MAX, datasync);
175 EXPORT_SYMBOL(vfs_fsync);
177 static int do_fsync(unsigned int fd, int datasync)
183 file = fget_light(fd, &fput_needed);
185 ret = vfs_fsync(file, datasync);
186 fput_light(file, fput_needed);
191 SYSCALL_DEFINE1(fsync, unsigned int, fd)
193 return do_fsync(fd, 0);
196 SYSCALL_DEFINE1(fdatasync, unsigned int, fd)
198 return do_fsync(fd, 1);
202 * generic_write_sync - perform syncing after a write if file / inode is sync
203 * @file: file to which the write happened
204 * @pos: offset where the write started
205 * @count: length of the write
207 * This is just a simple wrapper about our general syncing function.
209 int generic_write_sync(struct file *file, loff_t pos, loff_t count)
211 if (!(file->f_flags & O_DSYNC) && !IS_SYNC(file->f_mapping->host))
213 return vfs_fsync_range(file, pos, pos + count - 1,
214 (file->f_flags & __O_SYNC) ? 0 : 1);
216 EXPORT_SYMBOL(generic_write_sync);
219 * sys_sync_file_range() permits finely controlled syncing over a segment of
220 * a file in the range offset .. (offset+nbytes-1) inclusive. If nbytes is
221 * zero then sys_sync_file_range() will operate from offset out to EOF.
225 * SYNC_FILE_RANGE_WAIT_BEFORE: wait upon writeout of all pages in the range
226 * before performing the write.
228 * SYNC_FILE_RANGE_WRITE: initiate writeout of all those dirty pages in the
229 * range which are not presently under writeback. Note that this may block for
230 * significant periods due to exhaustion of disk request structures.
232 * SYNC_FILE_RANGE_WAIT_AFTER: wait upon writeout of all pages in the range
233 * after performing the write.
235 * Useful combinations of the flag bits are:
237 * SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE: ensures that all pages
238 * in the range which were dirty on entry to sys_sync_file_range() are placed
239 * under writeout. This is a start-write-for-data-integrity operation.
241 * SYNC_FILE_RANGE_WRITE: start writeout of all dirty pages in the range which
242 * are not presently under writeout. This is an asynchronous flush-to-disk
243 * operation. Not suitable for data integrity operations.
245 * SYNC_FILE_RANGE_WAIT_BEFORE (or SYNC_FILE_RANGE_WAIT_AFTER): wait for
246 * completion of writeout of all pages in the range. This will be used after an
247 * earlier SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE operation to wait
248 * for that operation to complete and to return the result.
250 * SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE|SYNC_FILE_RANGE_WAIT_AFTER:
251 * a traditional sync() operation. This is a write-for-data-integrity operation
252 * which will ensure that all pages in the range which were dirty on entry to
253 * sys_sync_file_range() are committed to disk.
256 * SYNC_FILE_RANGE_WAIT_BEFORE and SYNC_FILE_RANGE_WAIT_AFTER will detect any
257 * I/O errors or ENOSPC conditions and will return those to the caller, after
258 * clearing the EIO and ENOSPC flags in the address_space.
260 * It should be noted that none of these operations write out the file's
261 * metadata. So unless the application is strictly performing overwrites of
262 * already-instantiated disk blocks, there are no guarantees here that the data
263 * will be available after a crash.
265 SYSCALL_DEFINE(sync_file_range)(int fd, loff_t offset, loff_t nbytes,
270 struct address_space *mapping;
271 loff_t endbyte; /* inclusive */
276 if (flags & ~VALID_FLAGS)
279 endbyte = offset + nbytes;
283 if ((s64)endbyte < 0)
285 if (endbyte < offset)
288 if (sizeof(pgoff_t) == 4) {
289 if (offset >= (0x100000000ULL << PAGE_CACHE_SHIFT)) {
291 * The range starts outside a 32 bit machine's
292 * pagecache addressing capabilities. Let it "succeed"
297 if (endbyte >= (0x100000000ULL << PAGE_CACHE_SHIFT)) {
308 endbyte--; /* inclusive */
311 file = fget_light(fd, &fput_needed);
315 i_mode = file->f_path.dentry->d_inode->i_mode;
317 if (!S_ISREG(i_mode) && !S_ISBLK(i_mode) && !S_ISDIR(i_mode) &&
321 mapping = file->f_mapping;
328 if (flags & SYNC_FILE_RANGE_WAIT_BEFORE) {
329 ret = filemap_fdatawait_range(mapping, offset, endbyte);
334 if (flags & SYNC_FILE_RANGE_WRITE) {
335 ret = filemap_fdatawrite_range(mapping, offset, endbyte);
340 if (flags & SYNC_FILE_RANGE_WAIT_AFTER)
341 ret = filemap_fdatawait_range(mapping, offset, endbyte);
344 fput_light(file, fput_needed);
348 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
349 asmlinkage long SyS_sync_file_range(long fd, loff_t offset, loff_t nbytes,
352 return SYSC_sync_file_range((int) fd, offset, nbytes,
353 (unsigned int) flags);
355 SYSCALL_ALIAS(sys_sync_file_range, SyS_sync_file_range);
358 /* It would be nice if people remember that not all the world's an i386
359 when they introduce new system calls */
360 SYSCALL_DEFINE(sync_file_range2)(int fd, unsigned int flags,
361 loff_t offset, loff_t nbytes)
363 return sys_sync_file_range(fd, offset, nbytes, flags);
365 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
366 asmlinkage long SyS_sync_file_range2(long fd, long flags,
367 loff_t offset, loff_t nbytes)
369 return SYSC_sync_file_range2((int) fd, (unsigned int) flags,
372 SYSCALL_ALIAS(sys_sync_file_range2, SyS_sync_file_range2);