1/*
2 * High-level sync()-related operations
3 */
4
5#include <linux/kernel.h>
6#include <linux/file.h>
7#include <linux/fs.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>
18#include "internal.h"
19
20#define VALID_FLAGS (SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE| \
21			SYNC_FILE_RANGE_WAIT_AFTER)
22
23/*
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.
29 */
30static int __sync_filesystem(struct super_block *sb, int wait)
31{
32	if (wait)
33		sync_inodes_sb(sb);
34	else
35		writeback_inodes_sb(sb, WB_REASON_SYNC);
36
37	if (sb->s_op->sync_fs)
38		sb->s_op->sync_fs(sb, wait);
39	return __sync_blockdev(sb->s_bdev, wait);
40}
41
42/*
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.
46 */
47int sync_filesystem(struct super_block *sb)
48{
49	int ret;
50
51	/*
52	 * We need to be protected against the filesystem going from
53	 * r/o to r/w or vice versa.
54	 */
55	WARN_ON(!rwsem_is_locked(&sb->s_umount));
56
57	/*
58	 * No point in syncing out anything if the filesystem is read-only.
59	 */
60	if (sb->s_flags & MS_RDONLY)
61		return 0;
62
63	ret = __sync_filesystem(sb, 0);
64	if (ret < 0)
65		return ret;
66	return __sync_filesystem(sb, 1);
67}
68EXPORT_SYMBOL(sync_filesystem);
69
70static void sync_inodes_one_sb(struct super_block *sb, void *arg)
71{
72	if (!(sb->s_flags & MS_RDONLY))
73		sync_inodes_sb(sb);
74}
75
76static void sync_fs_one_sb(struct super_block *sb, void *arg)
77{
78	if (!(sb->s_flags & MS_RDONLY) && sb->s_op->sync_fs)
79		sb->s_op->sync_fs(sb, *(int *)arg);
80}
81
82static void fdatawrite_one_bdev(struct block_device *bdev, void *arg)
83{
84	filemap_fdatawrite(bdev->bd_inode->i_mapping);
85}
86
87static void fdatawait_one_bdev(struct block_device *bdev, void *arg)
88{
89	filemap_fdatawait(bdev->bd_inode->i_mapping);
90}
91
92/*
93 * Sync everything. We start by waking flusher threads so that most of
94 * writeback runs on all devices in parallel. Then we sync all inodes reliably
95 * which effectively also waits for all flusher threads to finish doing
96 * writeback. At this point all data is on disk so metadata should be stable
97 * and we tell filesystems to sync their metadata via ->sync_fs() calls.
98 * Finally, we writeout all block devices because some filesystems (e.g. ext2)
99 * just write metadata (such as inodes or bitmaps) to block device page cache
100 * and do not sync it on their own in ->sync_fs().
101 */
102SYSCALL_DEFINE0(sync)
103{
104	int nowait = 0, wait = 1;
105
106	wakeup_flusher_threads(0, WB_REASON_SYNC);
107	iterate_supers(sync_inodes_one_sb, NULL);
108	iterate_supers(sync_fs_one_sb, &nowait);
109	iterate_supers(sync_fs_one_sb, &wait);
110	iterate_bdevs(fdatawrite_one_bdev, NULL);
111	iterate_bdevs(fdatawait_one_bdev, NULL);
112	if (unlikely(laptop_mode))
113		laptop_sync_completion();
114	return 0;
115}
116
117static void do_sync_work(struct work_struct *work)
118{
119	int nowait = 0;
120
121	/*
122	 * Sync twice to reduce the possibility we skipped some inodes / pages
123	 * because they were temporarily locked
124	 */
125	iterate_supers(sync_inodes_one_sb, &nowait);
126	iterate_supers(sync_fs_one_sb, &nowait);
127	iterate_bdevs(fdatawrite_one_bdev, NULL);
128	iterate_supers(sync_inodes_one_sb, &nowait);
129	iterate_supers(sync_fs_one_sb, &nowait);
130	iterate_bdevs(fdatawrite_one_bdev, NULL);
131	printk("Emergency Sync complete\n");
132	kfree(work);
133}
134
135void emergency_sync(void)
136{
137	struct work_struct *work;
138
139	work = kmalloc(sizeof(*work), GFP_ATOMIC);
140	if (work) {
141		INIT_WORK(work, do_sync_work);
142		schedule_work(work);
143	}
144}
145
146/*
147 * sync a single super
148 */
149SYSCALL_DEFINE1(syncfs, int, fd)
150{
151	struct fd f = fdget(fd);
152	struct super_block *sb;
153	int ret;
154
155	if (!f.file)
156		return -EBADF;
157	sb = f.file->f_path.dentry->d_sb;
158
159	down_read(&sb->s_umount);
160	ret = sync_filesystem(sb);
161	up_read(&sb->s_umount);
162
163	fdput(f);
164	return ret;
165}
166
167/**
168 * vfs_fsync_range - helper to sync a range of data & metadata to disk
169 * @file:		file to sync
170 * @start:		offset in bytes of the beginning of data range to sync
171 * @end:		offset in bytes of the end of data range (inclusive)
172 * @datasync:		perform only datasync
173 *
174 * Write back data in range @start..@end and metadata for @file to disk.  If
175 * @datasync is set only metadata needed to access modified file data is
176 * written.
177 */
178int vfs_fsync_range(struct file *file, loff_t start, loff_t end, int datasync)
179{
180	struct inode *inode = file->f_mapping->host;
181
182	if (!file->f_op->fsync)
183		return -EINVAL;
184	if (!datasync && (inode->i_state & I_DIRTY_TIME)) {
185		spin_lock(&inode->i_lock);
186		inode->i_state &= ~I_DIRTY_TIME;
187		spin_unlock(&inode->i_lock);
188		mark_inode_dirty_sync(inode);
189	}
190	return file->f_op->fsync(file, start, end, datasync);
191}
192EXPORT_SYMBOL(vfs_fsync_range);
193
194/**
195 * vfs_fsync - perform a fsync or fdatasync on a file
196 * @file:		file to sync
197 * @datasync:		only perform a fdatasync operation
198 *
199 * Write back data and metadata for @file to disk.  If @datasync is
200 * set only metadata needed to access modified file data is written.
201 */
202int vfs_fsync(struct file *file, int datasync)
203{
204	return vfs_fsync_range(file, 0, LLONG_MAX, datasync);
205}
206EXPORT_SYMBOL(vfs_fsync);
207
208static int do_fsync(unsigned int fd, int datasync)
209{
210	struct fd f = fdget(fd);
211	int ret = -EBADF;
212
213	if (f.file) {
214		ret = vfs_fsync(f.file, datasync);
215		fdput(f);
216	}
217	return ret;
218}
219
220SYSCALL_DEFINE1(fsync, unsigned int, fd)
221{
222	return do_fsync(fd, 0);
223}
224
225SYSCALL_DEFINE1(fdatasync, unsigned int, fd)
226{
227	return do_fsync(fd, 1);
228}
229
230/*
231 * sys_sync_file_range() permits finely controlled syncing over a segment of
232 * a file in the range offset .. (offset+nbytes-1) inclusive.  If nbytes is
233 * zero then sys_sync_file_range() will operate from offset out to EOF.
234 *
235 * The flag bits are:
236 *
237 * SYNC_FILE_RANGE_WAIT_BEFORE: wait upon writeout of all pages in the range
238 * before performing the write.
239 *
240 * SYNC_FILE_RANGE_WRITE: initiate writeout of all those dirty pages in the
241 * range which are not presently under writeback. Note that this may block for
242 * significant periods due to exhaustion of disk request structures.
243 *
244 * SYNC_FILE_RANGE_WAIT_AFTER: wait upon writeout of all pages in the range
245 * after performing the write.
246 *
247 * Useful combinations of the flag bits are:
248 *
249 * SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE: ensures that all pages
250 * in the range which were dirty on entry to sys_sync_file_range() are placed
251 * under writeout.  This is a start-write-for-data-integrity operation.
252 *
253 * SYNC_FILE_RANGE_WRITE: start writeout of all dirty pages in the range which
254 * are not presently under writeout.  This is an asynchronous flush-to-disk
255 * operation.  Not suitable for data integrity operations.
256 *
257 * SYNC_FILE_RANGE_WAIT_BEFORE (or SYNC_FILE_RANGE_WAIT_AFTER): wait for
258 * completion of writeout of all pages in the range.  This will be used after an
259 * earlier SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE operation to wait
260 * for that operation to complete and to return the result.
261 *
262 * SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE|SYNC_FILE_RANGE_WAIT_AFTER:
263 * a traditional sync() operation.  This is a write-for-data-integrity operation
264 * which will ensure that all pages in the range which were dirty on entry to
265 * sys_sync_file_range() are committed to disk.
266 *
267 *
268 * SYNC_FILE_RANGE_WAIT_BEFORE and SYNC_FILE_RANGE_WAIT_AFTER will detect any
269 * I/O errors or ENOSPC conditions and will return those to the caller, after
270 * clearing the EIO and ENOSPC flags in the address_space.
271 *
272 * It should be noted that none of these operations write out the file's
273 * metadata.  So unless the application is strictly performing overwrites of
274 * already-instantiated disk blocks, there are no guarantees here that the data
275 * will be available after a crash.
276 */
277SYSCALL_DEFINE4(sync_file_range, int, fd, loff_t, offset, loff_t, nbytes,
278				unsigned int, flags)
279{
280	int ret;
281	struct fd f;
282	struct address_space *mapping;
283	loff_t endbyte;			/* inclusive */
284	umode_t i_mode;
285
286	ret = -EINVAL;
287	if (flags & ~VALID_FLAGS)
288		goto out;
289
290	endbyte = offset + nbytes;
291
292	if ((s64)offset < 0)
293		goto out;
294	if ((s64)endbyte < 0)
295		goto out;
296	if (endbyte < offset)
297		goto out;
298
299	if (sizeof(pgoff_t) == 4) {
300		if (offset >= (0x100000000ULL << PAGE_CACHE_SHIFT)) {
301			/*
302			 * The range starts outside a 32 bit machine's
303			 * pagecache addressing capabilities.  Let it "succeed"
304			 */
305			ret = 0;
306			goto out;
307		}
308		if (endbyte >= (0x100000000ULL << PAGE_CACHE_SHIFT)) {
309			/*
310			 * Out to EOF
311			 */
312			nbytes = 0;
313		}
314	}
315
316	if (nbytes == 0)
317		endbyte = LLONG_MAX;
318	else
319		endbyte--;		/* inclusive */
320
321	ret = -EBADF;
322	f = fdget(fd);
323	if (!f.file)
324		goto out;
325
326	i_mode = file_inode(f.file)->i_mode;
327	ret = -ESPIPE;
328	if (!S_ISREG(i_mode) && !S_ISBLK(i_mode) && !S_ISDIR(i_mode) &&
329			!S_ISLNK(i_mode))
330		goto out_put;
331
332	mapping = f.file->f_mapping;
333	if (!mapping) {
334		ret = -EINVAL;
335		goto out_put;
336	}
337
338	ret = 0;
339	if (flags & SYNC_FILE_RANGE_WAIT_BEFORE) {
340		ret = filemap_fdatawait_range(mapping, offset, endbyte);
341		if (ret < 0)
342			goto out_put;
343	}
344
345	if (flags & SYNC_FILE_RANGE_WRITE) {
346		ret = filemap_fdatawrite_range(mapping, offset, endbyte);
347		if (ret < 0)
348			goto out_put;
349	}
350
351	if (flags & SYNC_FILE_RANGE_WAIT_AFTER)
352		ret = filemap_fdatawait_range(mapping, offset, endbyte);
353
354out_put:
355	fdput(f);
356out:
357	return ret;
358}
359
360/* It would be nice if people remember that not all the world's an i386
361   when they introduce new system calls */
362SYSCALL_DEFINE4(sync_file_range2, int, fd, unsigned int, flags,
363				 loff_t, offset, loff_t, nbytes)
364{
365	return sys_sync_file_range(fd, offset, nbytes, flags);
366}
367