xref: /openbmc/linux/fs/sync.c (revision 37185b33)
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  */
30 static 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  */
47 int 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 }
68 EXPORT_SYMBOL_GPL(sync_filesystem);
69 
70 static 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 
76 static 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 
82 static void fdatawrite_one_bdev(struct block_device *bdev, void *arg)
83 {
84 	filemap_fdatawrite(bdev->bd_inode->i_mapping);
85 }
86 
87 static 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  */
102 SYSCALL_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 
117 static 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 
135 void 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  */
149 SYSCALL_DEFINE1(syncfs, int, fd)
150 {
151 	struct file *file;
152 	struct super_block *sb;
153 	int ret;
154 	int fput_needed;
155 
156 	file = fget_light(fd, &fput_needed);
157 	if (!file)
158 		return -EBADF;
159 	sb = file->f_dentry->d_sb;
160 
161 	down_read(&sb->s_umount);
162 	ret = sync_filesystem(sb);
163 	up_read(&sb->s_umount);
164 
165 	fput_light(file, fput_needed);
166 	return ret;
167 }
168 
169 /**
170  * vfs_fsync_range - helper to sync a range of data & metadata to disk
171  * @file:		file to sync
172  * @start:		offset in bytes of the beginning of data range to sync
173  * @end:		offset in bytes of the end of data range (inclusive)
174  * @datasync:		perform only datasync
175  *
176  * Write back data in range @start..@end and metadata for @file to disk.  If
177  * @datasync is set only metadata needed to access modified file data is
178  * written.
179  */
180 int vfs_fsync_range(struct file *file, loff_t start, loff_t end, int datasync)
181 {
182 	if (!file->f_op || !file->f_op->fsync)
183 		return -EINVAL;
184 	return file->f_op->fsync(file, start, end, datasync);
185 }
186 EXPORT_SYMBOL(vfs_fsync_range);
187 
188 /**
189  * vfs_fsync - perform a fsync or fdatasync on a file
190  * @file:		file to sync
191  * @datasync:		only perform a fdatasync operation
192  *
193  * Write back data and metadata for @file to disk.  If @datasync is
194  * set only metadata needed to access modified file data is written.
195  */
196 int vfs_fsync(struct file *file, int datasync)
197 {
198 	return vfs_fsync_range(file, 0, LLONG_MAX, datasync);
199 }
200 EXPORT_SYMBOL(vfs_fsync);
201 
202 static int do_fsync(unsigned int fd, int datasync)
203 {
204 	struct file *file;
205 	int ret = -EBADF;
206 	int fput_needed;
207 
208 	file = fget_light(fd, &fput_needed);
209 	if (file) {
210 		ret = vfs_fsync(file, datasync);
211 		fput_light(file, fput_needed);
212 	}
213 	return ret;
214 }
215 
216 SYSCALL_DEFINE1(fsync, unsigned int, fd)
217 {
218 	return do_fsync(fd, 0);
219 }
220 
221 SYSCALL_DEFINE1(fdatasync, unsigned int, fd)
222 {
223 	return do_fsync(fd, 1);
224 }
225 
226 /**
227  * generic_write_sync - perform syncing after a write if file / inode is sync
228  * @file:	file to which the write happened
229  * @pos:	offset where the write started
230  * @count:	length of the write
231  *
232  * This is just a simple wrapper about our general syncing function.
233  */
234 int generic_write_sync(struct file *file, loff_t pos, loff_t count)
235 {
236 	if (!(file->f_flags & O_DSYNC) && !IS_SYNC(file->f_mapping->host))
237 		return 0;
238 	return vfs_fsync_range(file, pos, pos + count - 1,
239 			       (file->f_flags & __O_SYNC) ? 0 : 1);
240 }
241 EXPORT_SYMBOL(generic_write_sync);
242 
243 /*
244  * sys_sync_file_range() permits finely controlled syncing over a segment of
245  * a file in the range offset .. (offset+nbytes-1) inclusive.  If nbytes is
246  * zero then sys_sync_file_range() will operate from offset out to EOF.
247  *
248  * The flag bits are:
249  *
250  * SYNC_FILE_RANGE_WAIT_BEFORE: wait upon writeout of all pages in the range
251  * before performing the write.
252  *
253  * SYNC_FILE_RANGE_WRITE: initiate writeout of all those dirty pages in the
254  * range which are not presently under writeback. Note that this may block for
255  * significant periods due to exhaustion of disk request structures.
256  *
257  * SYNC_FILE_RANGE_WAIT_AFTER: wait upon writeout of all pages in the range
258  * after performing the write.
259  *
260  * Useful combinations of the flag bits are:
261  *
262  * SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE: ensures that all pages
263  * in the range which were dirty on entry to sys_sync_file_range() are placed
264  * under writeout.  This is a start-write-for-data-integrity operation.
265  *
266  * SYNC_FILE_RANGE_WRITE: start writeout of all dirty pages in the range which
267  * are not presently under writeout.  This is an asynchronous flush-to-disk
268  * operation.  Not suitable for data integrity operations.
269  *
270  * SYNC_FILE_RANGE_WAIT_BEFORE (or SYNC_FILE_RANGE_WAIT_AFTER): wait for
271  * completion of writeout of all pages in the range.  This will be used after an
272  * earlier SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE operation to wait
273  * for that operation to complete and to return the result.
274  *
275  * SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE|SYNC_FILE_RANGE_WAIT_AFTER:
276  * a traditional sync() operation.  This is a write-for-data-integrity operation
277  * which will ensure that all pages in the range which were dirty on entry to
278  * sys_sync_file_range() are committed to disk.
279  *
280  *
281  * SYNC_FILE_RANGE_WAIT_BEFORE and SYNC_FILE_RANGE_WAIT_AFTER will detect any
282  * I/O errors or ENOSPC conditions and will return those to the caller, after
283  * clearing the EIO and ENOSPC flags in the address_space.
284  *
285  * It should be noted that none of these operations write out the file's
286  * metadata.  So unless the application is strictly performing overwrites of
287  * already-instantiated disk blocks, there are no guarantees here that the data
288  * will be available after a crash.
289  */
290 SYSCALL_DEFINE(sync_file_range)(int fd, loff_t offset, loff_t nbytes,
291 				unsigned int flags)
292 {
293 	int ret;
294 	struct file *file;
295 	struct address_space *mapping;
296 	loff_t endbyte;			/* inclusive */
297 	int fput_needed;
298 	umode_t i_mode;
299 
300 	ret = -EINVAL;
301 	if (flags & ~VALID_FLAGS)
302 		goto out;
303 
304 	endbyte = offset + nbytes;
305 
306 	if ((s64)offset < 0)
307 		goto out;
308 	if ((s64)endbyte < 0)
309 		goto out;
310 	if (endbyte < offset)
311 		goto out;
312 
313 	if (sizeof(pgoff_t) == 4) {
314 		if (offset >= (0x100000000ULL << PAGE_CACHE_SHIFT)) {
315 			/*
316 			 * The range starts outside a 32 bit machine's
317 			 * pagecache addressing capabilities.  Let it "succeed"
318 			 */
319 			ret = 0;
320 			goto out;
321 		}
322 		if (endbyte >= (0x100000000ULL << PAGE_CACHE_SHIFT)) {
323 			/*
324 			 * Out to EOF
325 			 */
326 			nbytes = 0;
327 		}
328 	}
329 
330 	if (nbytes == 0)
331 		endbyte = LLONG_MAX;
332 	else
333 		endbyte--;		/* inclusive */
334 
335 	ret = -EBADF;
336 	file = fget_light(fd, &fput_needed);
337 	if (!file)
338 		goto out;
339 
340 	i_mode = file->f_path.dentry->d_inode->i_mode;
341 	ret = -ESPIPE;
342 	if (!S_ISREG(i_mode) && !S_ISBLK(i_mode) && !S_ISDIR(i_mode) &&
343 			!S_ISLNK(i_mode))
344 		goto out_put;
345 
346 	mapping = file->f_mapping;
347 	if (!mapping) {
348 		ret = -EINVAL;
349 		goto out_put;
350 	}
351 
352 	ret = 0;
353 	if (flags & SYNC_FILE_RANGE_WAIT_BEFORE) {
354 		ret = filemap_fdatawait_range(mapping, offset, endbyte);
355 		if (ret < 0)
356 			goto out_put;
357 	}
358 
359 	if (flags & SYNC_FILE_RANGE_WRITE) {
360 		ret = filemap_fdatawrite_range(mapping, offset, endbyte);
361 		if (ret < 0)
362 			goto out_put;
363 	}
364 
365 	if (flags & SYNC_FILE_RANGE_WAIT_AFTER)
366 		ret = filemap_fdatawait_range(mapping, offset, endbyte);
367 
368 out_put:
369 	fput_light(file, fput_needed);
370 out:
371 	return ret;
372 }
373 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
374 asmlinkage long SyS_sync_file_range(long fd, loff_t offset, loff_t nbytes,
375 				    long flags)
376 {
377 	return SYSC_sync_file_range((int) fd, offset, nbytes,
378 				    (unsigned int) flags);
379 }
380 SYSCALL_ALIAS(sys_sync_file_range, SyS_sync_file_range);
381 #endif
382 
383 /* It would be nice if people remember that not all the world's an i386
384    when they introduce new system calls */
385 SYSCALL_DEFINE(sync_file_range2)(int fd, unsigned int flags,
386 				 loff_t offset, loff_t nbytes)
387 {
388 	return sys_sync_file_range(fd, offset, nbytes, flags);
389 }
390 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
391 asmlinkage long SyS_sync_file_range2(long fd, long flags,
392 				     loff_t offset, loff_t nbytes)
393 {
394 	return SYSC_sync_file_range2((int) fd, (unsigned int) flags,
395 				     offset, nbytes);
396 }
397 SYSCALL_ALIAS(sys_sync_file_range2, SyS_sync_file_range2);
398 #endif
399