blob: 5c2420c3eb17b349d02a224e543218fd127c734c [file] [log] [blame]
Andrew Mortonf79e2ab2006-03-31 02:30:42 -08001/*
2 * High-level sync()-related operations
3 */
4
5#include <linux/kernel.h>
6#include <linux/file.h>
7#include <linux/fs.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +09008#include <linux/slab.h>
Paul Gortmaker630d9c42011-11-16 23:57:37 -05009#include <linux/export.h>
Sage Weilb7ed78f2011-03-10 11:31:30 -080010#include <linux/namei.h>
Al Viro914e2632006-10-18 13:55:46 -040011#include <linux/sched.h>
Andrew Mortonf79e2ab2006-03-31 02:30:42 -080012#include <linux/writeback.h>
13#include <linux/syscalls.h>
14#include <linux/linkage.h>
15#include <linux/pagemap.h>
David Howellscf9a2ae2006-08-29 19:05:54 +010016#include <linux/quotaops.h>
Jörn Engel5129a462010-04-25 08:54:42 +020017#include <linux/backing-dev.h>
Jan Kara5a3e5cb2009-04-27 16:43:48 +020018#include "internal.h"
Andrew Mortonf79e2ab2006-03-31 02:30:42 -080019
20#define VALID_FLAGS (SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE| \
21 SYNC_FILE_RANGE_WAIT_AFTER)
22
Jan Karac15c54f2009-04-27 16:43:52 +020023/*
Jens Axboed8a85592009-09-02 12:34:32 +020024 * 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.
Jan Karac15c54f2009-04-27 16:43:52 +020029 */
Jan Kara0dc83bd2014-02-21 11:19:04 +010030static int __sync_filesystem(struct super_block *sb, int wait)
Jan Karac15c54f2009-04-27 16:43:52 +020031{
Christoph Hellwig5fb324a2010-02-16 03:44:52 -050032 if (wait)
Jan Kara0dc83bd2014-02-21 11:19:04 +010033 sync_inodes_sb(sb);
Christoph Hellwig5fb324a2010-02-16 03:44:52 -050034 else
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -060035 writeback_inodes_sb(sb, WB_REASON_SYNC);
Christoph Hellwig5fb324a2010-02-16 03:44:52 -050036
Jan Karac15c54f2009-04-27 16:43:52 +020037 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 */
Jan Kara60b06802009-04-27 16:43:53 +020047int sync_filesystem(struct super_block *sb)
Jan Karac15c54f2009-04-27 16:43:52 +020048{
49 int ret;
50
Christoph Hellwig5af79262009-05-05 15:41:25 +020051 /*
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
Jan Kara0dc83bd2014-02-21 11:19:04 +010063 ret = __sync_filesystem(sb, 0);
Jan Karac15c54f2009-04-27 16:43:52 +020064 if (ret < 0)
65 return ret;
Jan Kara0dc83bd2014-02-21 11:19:04 +010066 return __sync_filesystem(sb, 1);
Jan Karac15c54f2009-04-27 16:43:52 +020067}
Anton Altaparmakov10096fb2014-08-21 11:09:27 +010068EXPORT_SYMBOL(sync_filesystem);
Jan Karac15c54f2009-04-27 16:43:52 +020069
Jan Karab3de6532012-07-03 16:45:30 +020070static void sync_inodes_one_sb(struct super_block *sb, void *arg)
Al Viro01a05b32010-03-23 06:06:58 -040071{
Jens Axboe95f28602011-03-17 11:13:12 +010072 if (!(sb->s_flags & MS_RDONLY))
Jan Kara0dc83bd2014-02-21 11:19:04 +010073 sync_inodes_sb(sb);
Al Viro01a05b32010-03-23 06:06:58 -040074}
Jan Karab3de6532012-07-03 16:45:30 +020075
Jan Karab3de6532012-07-03 16:45:30 +020076static 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
Jan Karad0e91b12012-07-03 16:45:33 +020082static void fdatawrite_one_bdev(struct block_device *bdev, void *arg)
Jan Karab3de6532012-07-03 16:45:30 +020083{
Jan Karad0e91b12012-07-03 16:45:33 +020084 filemap_fdatawrite(bdev->bd_inode->i_mapping);
Jan Karaa8c71762012-07-03 16:45:32 +020085}
86
Jan Karad0e91b12012-07-03 16:45:33 +020087static void fdatawait_one_bdev(struct block_device *bdev, void *arg)
Jan Karaa8c71762012-07-03 16:45:32 +020088{
Junichi Nomuraaa750fd2015-11-05 18:47:23 -080089 /*
90 * We keep the error status of individual mapping so that
91 * applications can catch the writeback error using fsync(2).
92 * See filemap_fdatawait_keep_errors() for details.
93 */
94 filemap_fdatawait_keep_errors(bdev->bd_inode->i_mapping);
Jan Karac15c54f2009-04-27 16:43:52 +020095}
96
Zhang, Yanmin3beab0b2009-07-05 12:08:08 -070097/*
Jan Kara4ea425b2012-07-03 16:45:34 +020098 * Sync everything. We start by waking flusher threads so that most of
99 * writeback runs on all devices in parallel. Then we sync all inodes reliably
100 * which effectively also waits for all flusher threads to finish doing
101 * writeback. At this point all data is on disk so metadata should be stable
102 * and we tell filesystems to sync their metadata via ->sync_fs() calls.
103 * Finally, we writeout all block devices because some filesystems (e.g. ext2)
104 * just write metadata (such as inodes or bitmaps) to block device page cache
105 * and do not sync it on their own in ->sync_fs().
Zhang, Yanmin3beab0b2009-07-05 12:08:08 -0700106 */
Heiko Carstensa5f8fa92009-01-14 14:14:11 +0100107SYSCALL_DEFINE0(sync)
David Howellscf9a2ae2006-08-29 19:05:54 +0100108{
Jan Karab3de6532012-07-03 16:45:30 +0200109 int nowait = 0, wait = 1;
110
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -0600111 wakeup_flusher_threads(0, WB_REASON_SYNC);
Jan Kara0dc83bd2014-02-21 11:19:04 +0100112 iterate_supers(sync_inodes_one_sb, NULL);
Jan Kara4ea425b2012-07-03 16:45:34 +0200113 iterate_supers(sync_fs_one_sb, &nowait);
Jan Karab3de6532012-07-03 16:45:30 +0200114 iterate_supers(sync_fs_one_sb, &wait);
Jan Karad0e91b12012-07-03 16:45:33 +0200115 iterate_bdevs(fdatawrite_one_bdev, NULL);
116 iterate_bdevs(fdatawait_one_bdev, NULL);
Jan Kara5cee5812009-04-27 16:43:51 +0200117 if (unlikely(laptop_mode))
118 laptop_sync_completion();
David Howellscf9a2ae2006-08-29 19:05:54 +0100119 return 0;
120}
121
Jens Axboea2a95372009-03-17 09:38:40 +0100122static void do_sync_work(struct work_struct *work)
123{
Jan Karab3de6532012-07-03 16:45:30 +0200124 int nowait = 0;
125
Jan Kara5cee5812009-04-27 16:43:51 +0200126 /*
127 * Sync twice to reduce the possibility we skipped some inodes / pages
128 * because they were temporarily locked
129 */
Jan Karab3de6532012-07-03 16:45:30 +0200130 iterate_supers(sync_inodes_one_sb, &nowait);
131 iterate_supers(sync_fs_one_sb, &nowait);
Jan Karad0e91b12012-07-03 16:45:33 +0200132 iterate_bdevs(fdatawrite_one_bdev, NULL);
Jan Karab3de6532012-07-03 16:45:30 +0200133 iterate_supers(sync_inodes_one_sb, &nowait);
134 iterate_supers(sync_fs_one_sb, &nowait);
Jan Karad0e91b12012-07-03 16:45:33 +0200135 iterate_bdevs(fdatawrite_one_bdev, NULL);
Jan Kara5cee5812009-04-27 16:43:51 +0200136 printk("Emergency Sync complete\n");
Jens Axboea2a95372009-03-17 09:38:40 +0100137 kfree(work);
138}
139
David Howellscf9a2ae2006-08-29 19:05:54 +0100140void emergency_sync(void)
141{
Jens Axboea2a95372009-03-17 09:38:40 +0100142 struct work_struct *work;
143
144 work = kmalloc(sizeof(*work), GFP_ATOMIC);
145 if (work) {
146 INIT_WORK(work, do_sync_work);
147 schedule_work(work);
148 }
David Howellscf9a2ae2006-08-29 19:05:54 +0100149}
150
Sage Weilb7ed78f2011-03-10 11:31:30 -0800151/*
152 * sync a single super
153 */
154SYSCALL_DEFINE1(syncfs, int, fd)
155{
Al Viro2903ff02012-08-28 12:52:22 -0400156 struct fd f = fdget(fd);
Sage Weilb7ed78f2011-03-10 11:31:30 -0800157 struct super_block *sb;
158 int ret;
Sage Weilb7ed78f2011-03-10 11:31:30 -0800159
Al Viro2903ff02012-08-28 12:52:22 -0400160 if (!f.file)
Sage Weilb7ed78f2011-03-10 11:31:30 -0800161 return -EBADF;
Al Virob5830432014-10-31 01:22:04 -0400162 sb = f.file->f_path.dentry->d_sb;
Sage Weilb7ed78f2011-03-10 11:31:30 -0800163
164 down_read(&sb->s_umount);
165 ret = sync_filesystem(sb);
166 up_read(&sb->s_umount);
167
Al Viro2903ff02012-08-28 12:52:22 -0400168 fdput(f);
Sage Weilb7ed78f2011-03-10 11:31:30 -0800169 return ret;
170}
171
Christoph Hellwig4c728ef2008-12-22 21:11:15 +0100172/**
Jan Kara148f9482009-08-17 19:52:36 +0200173 * vfs_fsync_range - helper to sync a range of data & metadata to disk
Christoph Hellwig4c728ef2008-12-22 21:11:15 +0100174 * @file: file to sync
Jan Kara148f9482009-08-17 19:52:36 +0200175 * @start: offset in bytes of the beginning of data range to sync
176 * @end: offset in bytes of the end of data range (inclusive)
177 * @datasync: perform only datasync
Christoph Hellwig4c728ef2008-12-22 21:11:15 +0100178 *
Jan Kara148f9482009-08-17 19:52:36 +0200179 * Write back data in range @start..@end and metadata for @file to disk. If
180 * @datasync is set only metadata needed to access modified file data is
181 * written.
Christoph Hellwig4c728ef2008-12-22 21:11:15 +0100182 */
Christoph Hellwig8018ab02010-03-22 17:32:25 +0100183int vfs_fsync_range(struct file *file, loff_t start, loff_t end, int datasync)
David Howellscf9a2ae2006-08-29 19:05:54 +0100184{
Theodore Ts'o0ae45f62015-02-02 00:37:00 -0500185 struct inode *inode = file->f_mapping->host;
186
Al Viro72c2d532013-09-22 16:27:52 -0400187 if (!file->f_op->fsync)
Josef Bacik02c24a82011-07-16 20:44:56 -0400188 return -EINVAL;
Theodore Ts'o0ae45f62015-02-02 00:37:00 -0500189 if (!datasync && (inode->i_state & I_DIRTY_TIME)) {
190 spin_lock(&inode->i_lock);
191 inode->i_state &= ~I_DIRTY_TIME;
192 spin_unlock(&inode->i_lock);
193 mark_inode_dirty_sync(inode);
194 }
Josef Bacik02c24a82011-07-16 20:44:56 -0400195 return file->f_op->fsync(file, start, end, datasync);
David Howellscf9a2ae2006-08-29 19:05:54 +0100196}
Jan Kara148f9482009-08-17 19:52:36 +0200197EXPORT_SYMBOL(vfs_fsync_range);
198
199/**
200 * vfs_fsync - perform a fsync or fdatasync on a file
201 * @file: file to sync
Jan Kara148f9482009-08-17 19:52:36 +0200202 * @datasync: only perform a fdatasync operation
203 *
204 * Write back data and metadata for @file to disk. If @datasync is
205 * set only metadata needed to access modified file data is written.
Jan Kara148f9482009-08-17 19:52:36 +0200206 */
Christoph Hellwig8018ab02010-03-22 17:32:25 +0100207int vfs_fsync(struct file *file, int datasync)
Jan Kara148f9482009-08-17 19:52:36 +0200208{
Christoph Hellwig8018ab02010-03-22 17:32:25 +0100209 return vfs_fsync_range(file, 0, LLONG_MAX, datasync);
Jan Kara148f9482009-08-17 19:52:36 +0200210}
Christoph Hellwig4c728ef2008-12-22 21:11:15 +0100211EXPORT_SYMBOL(vfs_fsync);
David Howellscf9a2ae2006-08-29 19:05:54 +0100212
Christoph Hellwig4c728ef2008-12-22 21:11:15 +0100213static int do_fsync(unsigned int fd, int datasync)
David Howellscf9a2ae2006-08-29 19:05:54 +0100214{
Al Viro2903ff02012-08-28 12:52:22 -0400215 struct fd f = fdget(fd);
David Howellscf9a2ae2006-08-29 19:05:54 +0100216 int ret = -EBADF;
217
Al Viro2903ff02012-08-28 12:52:22 -0400218 if (f.file) {
219 ret = vfs_fsync(f.file, datasync);
220 fdput(f);
Jin Qiane61abd82017-03-02 13:32:59 -0800221 inc_syscfs(current);
David Howellscf9a2ae2006-08-29 19:05:54 +0100222 }
223 return ret;
224}
225
Heiko Carstensa5f8fa92009-01-14 14:14:11 +0100226SYSCALL_DEFINE1(fsync, unsigned int, fd)
David Howellscf9a2ae2006-08-29 19:05:54 +0100227{
Christoph Hellwig4c728ef2008-12-22 21:11:15 +0100228 return do_fsync(fd, 0);
David Howellscf9a2ae2006-08-29 19:05:54 +0100229}
230
Heiko Carstensa5f8fa92009-01-14 14:14:11 +0100231SYSCALL_DEFINE1(fdatasync, unsigned int, fd)
David Howellscf9a2ae2006-08-29 19:05:54 +0100232{
Christoph Hellwig4c728ef2008-12-22 21:11:15 +0100233 return do_fsync(fd, 1);
David Howellscf9a2ae2006-08-29 19:05:54 +0100234}
235
236/*
Andrew Mortonf79e2ab2006-03-31 02:30:42 -0800237 * sys_sync_file_range() permits finely controlled syncing over a segment of
238 * a file in the range offset .. (offset+nbytes-1) inclusive. If nbytes is
239 * zero then sys_sync_file_range() will operate from offset out to EOF.
240 *
241 * The flag bits are:
242 *
243 * SYNC_FILE_RANGE_WAIT_BEFORE: wait upon writeout of all pages in the range
244 * before performing the write.
245 *
246 * SYNC_FILE_RANGE_WRITE: initiate writeout of all those dirty pages in the
Pavel Machekcce77082008-07-23 21:27:36 -0700247 * range which are not presently under writeback. Note that this may block for
248 * significant periods due to exhaustion of disk request structures.
Andrew Mortonf79e2ab2006-03-31 02:30:42 -0800249 *
250 * SYNC_FILE_RANGE_WAIT_AFTER: wait upon writeout of all pages in the range
251 * after performing the write.
252 *
253 * Useful combinations of the flag bits are:
254 *
255 * SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE: ensures that all pages
256 * in the range which were dirty on entry to sys_sync_file_range() are placed
257 * under writeout. This is a start-write-for-data-integrity operation.
258 *
259 * SYNC_FILE_RANGE_WRITE: start writeout of all dirty pages in the range which
260 * are not presently under writeout. This is an asynchronous flush-to-disk
261 * operation. Not suitable for data integrity operations.
262 *
263 * SYNC_FILE_RANGE_WAIT_BEFORE (or SYNC_FILE_RANGE_WAIT_AFTER): wait for
264 * completion of writeout of all pages in the range. This will be used after an
265 * earlier SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE operation to wait
266 * for that operation to complete and to return the result.
267 *
268 * SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE|SYNC_FILE_RANGE_WAIT_AFTER:
269 * a traditional sync() operation. This is a write-for-data-integrity operation
270 * which will ensure that all pages in the range which were dirty on entry to
271 * sys_sync_file_range() are committed to disk.
272 *
273 *
274 * SYNC_FILE_RANGE_WAIT_BEFORE and SYNC_FILE_RANGE_WAIT_AFTER will detect any
275 * I/O errors or ENOSPC conditions and will return those to the caller, after
276 * clearing the EIO and ENOSPC flags in the address_space.
277 *
278 * It should be noted that none of these operations write out the file's
279 * metadata. So unless the application is strictly performing overwrites of
280 * already-instantiated disk blocks, there are no guarantees here that the data
281 * will be available after a crash.
282 */
Al Viro4a0fd5b2013-01-21 15:16:58 -0500283SYSCALL_DEFINE4(sync_file_range, int, fd, loff_t, offset, loff_t, nbytes,
284 unsigned int, flags)
Andrew Mortonf79e2ab2006-03-31 02:30:42 -0800285{
286 int ret;
Al Viro2903ff02012-08-28 12:52:22 -0400287 struct fd f;
Christoph Hellwig7a0ad102009-12-17 14:24:40 +0100288 struct address_space *mapping;
Andrew Mortonf79e2ab2006-03-31 02:30:42 -0800289 loff_t endbyte; /* inclusive */
Andrew Mortonf79e2ab2006-03-31 02:30:42 -0800290 umode_t i_mode;
291
292 ret = -EINVAL;
293 if (flags & ~VALID_FLAGS)
294 goto out;
295
296 endbyte = offset + nbytes;
297
298 if ((s64)offset < 0)
299 goto out;
300 if ((s64)endbyte < 0)
301 goto out;
302 if (endbyte < offset)
303 goto out;
304
305 if (sizeof(pgoff_t) == 4) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300306 if (offset >= (0x100000000ULL << PAGE_SHIFT)) {
Andrew Mortonf79e2ab2006-03-31 02:30:42 -0800307 /*
308 * The range starts outside a 32 bit machine's
309 * pagecache addressing capabilities. Let it "succeed"
310 */
311 ret = 0;
312 goto out;
313 }
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300314 if (endbyte >= (0x100000000ULL << PAGE_SHIFT)) {
Andrew Mortonf79e2ab2006-03-31 02:30:42 -0800315 /*
316 * Out to EOF
317 */
318 nbytes = 0;
319 }
320 }
321
322 if (nbytes == 0)
OGAWA Hirofumi111ebb62006-06-23 02:03:26 -0700323 endbyte = LLONG_MAX;
Andrew Mortonf79e2ab2006-03-31 02:30:42 -0800324 else
325 endbyte--; /* inclusive */
326
327 ret = -EBADF;
Al Viro2903ff02012-08-28 12:52:22 -0400328 f = fdget(fd);
329 if (!f.file)
Andrew Mortonf79e2ab2006-03-31 02:30:42 -0800330 goto out;
331
Al Viro496ad9a2013-01-23 17:07:38 -0500332 i_mode = file_inode(f.file)->i_mode;
Andrew Mortonf79e2ab2006-03-31 02:30:42 -0800333 ret = -ESPIPE;
334 if (!S_ISREG(i_mode) && !S_ISBLK(i_mode) && !S_ISDIR(i_mode) &&
335 !S_ISLNK(i_mode))
336 goto out_put;
337
Al Viro2903ff02012-08-28 12:52:22 -0400338 mapping = f.file->f_mapping;
Christoph Hellwig7a0ad102009-12-17 14:24:40 +0100339 if (!mapping) {
340 ret = -EINVAL;
341 goto out_put;
342 }
343
344 ret = 0;
345 if (flags & SYNC_FILE_RANGE_WAIT_BEFORE) {
346 ret = filemap_fdatawait_range(mapping, offset, endbyte);
347 if (ret < 0)
348 goto out_put;
349 }
350
351 if (flags & SYNC_FILE_RANGE_WRITE) {
Jan Kara23d01272015-11-06 16:28:55 -0800352 ret = __filemap_fdatawrite_range(mapping, offset, endbyte,
353 WB_SYNC_NONE);
Christoph Hellwig7a0ad102009-12-17 14:24:40 +0100354 if (ret < 0)
355 goto out_put;
356 }
357
358 if (flags & SYNC_FILE_RANGE_WAIT_AFTER)
359 ret = filemap_fdatawait_range(mapping, offset, endbyte);
360
Andrew Mortonf79e2ab2006-03-31 02:30:42 -0800361out_put:
Al Viro2903ff02012-08-28 12:52:22 -0400362 fdput(f);
Andrew Mortonf79e2ab2006-03-31 02:30:42 -0800363out:
364 return ret;
365}
366
David Woodhouseedd5cd42007-06-27 14:10:09 -0700367/* It would be nice if people remember that not all the world's an i386
368 when they introduce new system calls */
Al Viro4a0fd5b2013-01-21 15:16:58 -0500369SYSCALL_DEFINE4(sync_file_range2, int, fd, unsigned int, flags,
370 loff_t, offset, loff_t, nbytes)
David Woodhouseedd5cd42007-06-27 14:10:09 -0700371{
372 return sys_sync_file_range(fd, offset, nbytes, flags);
373}