blob: 905f3f6b3d8567032b9d5ad6044a184cb6ac3151 [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 Kara60b06802009-04-27 16:43:53 +020030static 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)
Jens Axboed8a85592009-09-02 12:34:32 +020033 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 Kara60b06802009-04-27 16:43:53 +020063 ret = __sync_filesystem(sb, 0);
Jan Karac15c54f2009-04-27 16:43:52 +020064 if (ret < 0)
65 return ret;
Jan Kara60b06802009-04-27 16:43:53 +020066 return __sync_filesystem(sb, 1);
Jan Karac15c54f2009-04-27 16:43:52 +020067}
Jan Kara60b06802009-04-27 16:43:53 +020068EXPORT_SYMBOL_GPL(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 Karab3de6532012-07-03 16:45:30 +020073 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{
Jan Karad0e91b12012-07-03 16:45:33 +020089 filemap_fdatawait(bdev->bd_inode->i_mapping);
Jan Karac15c54f2009-04-27 16:43:52 +020090}
91
Zhang, Yanmin3beab0b2009-07-05 12:08:08 -070092/*
Jan Kara4ea425b2012-07-03 16:45:34 +020093 * 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().
Zhang, Yanmin3beab0b2009-07-05 12:08:08 -0700101 */
Heiko Carstensa5f8fa92009-01-14 14:14:11 +0100102SYSCALL_DEFINE0(sync)
David Howellscf9a2ae2006-08-29 19:05:54 +0100103{
Jan Karab3de6532012-07-03 16:45:30 +0200104 int nowait = 0, wait = 1;
105
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -0600106 wakeup_flusher_threads(0, WB_REASON_SYNC);
Jan Karab3de6532012-07-03 16:45:30 +0200107 iterate_supers(sync_inodes_one_sb, NULL);
Jan Kara4ea425b2012-07-03 16:45:34 +0200108 iterate_supers(sync_fs_one_sb, &nowait);
Jan Karab3de6532012-07-03 16:45:30 +0200109 iterate_supers(sync_fs_one_sb, &wait);
Jan Karad0e91b12012-07-03 16:45:33 +0200110 iterate_bdevs(fdatawrite_one_bdev, NULL);
111 iterate_bdevs(fdatawait_one_bdev, NULL);
Jan Kara5cee5812009-04-27 16:43:51 +0200112 if (unlikely(laptop_mode))
113 laptop_sync_completion();
David Howellscf9a2ae2006-08-29 19:05:54 +0100114 return 0;
115}
116
Jens Axboea2a95372009-03-17 09:38:40 +0100117static void do_sync_work(struct work_struct *work)
118{
Jan Karab3de6532012-07-03 16:45:30 +0200119 int nowait = 0;
120
Jan Kara5cee5812009-04-27 16:43:51 +0200121 /*
122 * Sync twice to reduce the possibility we skipped some inodes / pages
123 * because they were temporarily locked
124 */
Jan Karab3de6532012-07-03 16:45:30 +0200125 iterate_supers(sync_inodes_one_sb, &nowait);
126 iterate_supers(sync_fs_one_sb, &nowait);
Jan Karad0e91b12012-07-03 16:45:33 +0200127 iterate_bdevs(fdatawrite_one_bdev, NULL);
Jan Karab3de6532012-07-03 16:45:30 +0200128 iterate_supers(sync_inodes_one_sb, &nowait);
129 iterate_supers(sync_fs_one_sb, &nowait);
Jan Karad0e91b12012-07-03 16:45:33 +0200130 iterate_bdevs(fdatawrite_one_bdev, NULL);
Jan Kara5cee5812009-04-27 16:43:51 +0200131 printk("Emergency Sync complete\n");
Jens Axboea2a95372009-03-17 09:38:40 +0100132 kfree(work);
133}
134
David Howellscf9a2ae2006-08-29 19:05:54 +0100135void emergency_sync(void)
136{
Jens Axboea2a95372009-03-17 09:38:40 +0100137 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 }
David Howellscf9a2ae2006-08-29 19:05:54 +0100144}
145
Sage Weilb7ed78f2011-03-10 11:31:30 -0800146/*
147 * sync a single super
148 */
149SYSCALL_DEFINE1(syncfs, int, fd)
150{
Al Viro2903ff02012-08-28 12:52:22 -0400151 struct fd f = fdget(fd);
Sage Weilb7ed78f2011-03-10 11:31:30 -0800152 struct super_block *sb;
153 int ret;
Sage Weilb7ed78f2011-03-10 11:31:30 -0800154
Al Viro2903ff02012-08-28 12:52:22 -0400155 if (!f.file)
Sage Weilb7ed78f2011-03-10 11:31:30 -0800156 return -EBADF;
Al Viro2903ff02012-08-28 12:52:22 -0400157 sb = f.file->f_dentry->d_sb;
Sage Weilb7ed78f2011-03-10 11:31:30 -0800158
159 down_read(&sb->s_umount);
160 ret = sync_filesystem(sb);
161 up_read(&sb->s_umount);
162
Al Viro2903ff02012-08-28 12:52:22 -0400163 fdput(f);
Sage Weilb7ed78f2011-03-10 11:31:30 -0800164 return ret;
165}
166
Christoph Hellwig4c728ef2008-12-22 21:11:15 +0100167/**
Jan Kara148f9482009-08-17 19:52:36 +0200168 * vfs_fsync_range - helper to sync a range of data & metadata to disk
Christoph Hellwig4c728ef2008-12-22 21:11:15 +0100169 * @file: file to sync
Jan Kara148f9482009-08-17 19:52:36 +0200170 * @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
Christoph Hellwig4c728ef2008-12-22 21:11:15 +0100173 *
Jan Kara148f9482009-08-17 19:52:36 +0200174 * 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.
Christoph Hellwig4c728ef2008-12-22 21:11:15 +0100177 */
Christoph Hellwig8018ab02010-03-22 17:32:25 +0100178int vfs_fsync_range(struct file *file, loff_t start, loff_t end, int datasync)
David Howellscf9a2ae2006-08-29 19:05:54 +0100179{
Josef Bacik02c24a82011-07-16 20:44:56 -0400180 if (!file->f_op || !file->f_op->fsync)
181 return -EINVAL;
182 return file->f_op->fsync(file, start, end, datasync);
David Howellscf9a2ae2006-08-29 19:05:54 +0100183}
Jan Kara148f9482009-08-17 19:52:36 +0200184EXPORT_SYMBOL(vfs_fsync_range);
185
186/**
187 * vfs_fsync - perform a fsync or fdatasync on a file
188 * @file: file to sync
Jan Kara148f9482009-08-17 19:52:36 +0200189 * @datasync: only perform a fdatasync operation
190 *
191 * Write back data and metadata for @file to disk. If @datasync is
192 * set only metadata needed to access modified file data is written.
Jan Kara148f9482009-08-17 19:52:36 +0200193 */
Christoph Hellwig8018ab02010-03-22 17:32:25 +0100194int vfs_fsync(struct file *file, int datasync)
Jan Kara148f9482009-08-17 19:52:36 +0200195{
Christoph Hellwig8018ab02010-03-22 17:32:25 +0100196 return vfs_fsync_range(file, 0, LLONG_MAX, datasync);
Jan Kara148f9482009-08-17 19:52:36 +0200197}
Christoph Hellwig4c728ef2008-12-22 21:11:15 +0100198EXPORT_SYMBOL(vfs_fsync);
David Howellscf9a2ae2006-08-29 19:05:54 +0100199
Christoph Hellwig4c728ef2008-12-22 21:11:15 +0100200static int do_fsync(unsigned int fd, int datasync)
David Howellscf9a2ae2006-08-29 19:05:54 +0100201{
Al Viro2903ff02012-08-28 12:52:22 -0400202 struct fd f = fdget(fd);
David Howellscf9a2ae2006-08-29 19:05:54 +0100203 int ret = -EBADF;
204
Al Viro2903ff02012-08-28 12:52:22 -0400205 if (f.file) {
206 ret = vfs_fsync(f.file, datasync);
207 fdput(f);
David Howellscf9a2ae2006-08-29 19:05:54 +0100208 }
209 return ret;
210}
211
Heiko Carstensa5f8fa92009-01-14 14:14:11 +0100212SYSCALL_DEFINE1(fsync, unsigned int, fd)
David Howellscf9a2ae2006-08-29 19:05:54 +0100213{
Christoph Hellwig4c728ef2008-12-22 21:11:15 +0100214 return do_fsync(fd, 0);
David Howellscf9a2ae2006-08-29 19:05:54 +0100215}
216
Heiko Carstensa5f8fa92009-01-14 14:14:11 +0100217SYSCALL_DEFINE1(fdatasync, unsigned int, fd)
David Howellscf9a2ae2006-08-29 19:05:54 +0100218{
Christoph Hellwig4c728ef2008-12-22 21:11:15 +0100219 return do_fsync(fd, 1);
David Howellscf9a2ae2006-08-29 19:05:54 +0100220}
221
Jan Kara148f9482009-08-17 19:52:36 +0200222/**
223 * generic_write_sync - perform syncing after a write if file / inode is sync
224 * @file: file to which the write happened
225 * @pos: offset where the write started
226 * @count: length of the write
227 *
228 * This is just a simple wrapper about our general syncing function.
229 */
230int generic_write_sync(struct file *file, loff_t pos, loff_t count)
231{
Christoph Hellwig6b2f3d12009-10-27 11:05:28 +0100232 if (!(file->f_flags & O_DSYNC) && !IS_SYNC(file->f_mapping->host))
Jan Kara148f9482009-08-17 19:52:36 +0200233 return 0;
Christoph Hellwig8018ab02010-03-22 17:32:25 +0100234 return vfs_fsync_range(file, pos, pos + count - 1,
Christoph Hellwig6b2f3d12009-10-27 11:05:28 +0100235 (file->f_flags & __O_SYNC) ? 0 : 1);
Jan Kara148f9482009-08-17 19:52:36 +0200236}
237EXPORT_SYMBOL(generic_write_sync);
238
David Howellscf9a2ae2006-08-29 19:05:54 +0100239/*
Andrew Mortonf79e2ab2006-03-31 02:30:42 -0800240 * sys_sync_file_range() permits finely controlled syncing over a segment of
241 * a file in the range offset .. (offset+nbytes-1) inclusive. If nbytes is
242 * zero then sys_sync_file_range() will operate from offset out to EOF.
243 *
244 * The flag bits are:
245 *
246 * SYNC_FILE_RANGE_WAIT_BEFORE: wait upon writeout of all pages in the range
247 * before performing the write.
248 *
249 * SYNC_FILE_RANGE_WRITE: initiate writeout of all those dirty pages in the
Pavel Machekcce77082008-07-23 21:27:36 -0700250 * range which are not presently under writeback. Note that this may block for
251 * significant periods due to exhaustion of disk request structures.
Andrew Mortonf79e2ab2006-03-31 02:30:42 -0800252 *
253 * SYNC_FILE_RANGE_WAIT_AFTER: wait upon writeout of all pages in the range
254 * after performing the write.
255 *
256 * Useful combinations of the flag bits are:
257 *
258 * SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE: ensures that all pages
259 * in the range which were dirty on entry to sys_sync_file_range() are placed
260 * under writeout. This is a start-write-for-data-integrity operation.
261 *
262 * SYNC_FILE_RANGE_WRITE: start writeout of all dirty pages in the range which
263 * are not presently under writeout. This is an asynchronous flush-to-disk
264 * operation. Not suitable for data integrity operations.
265 *
266 * SYNC_FILE_RANGE_WAIT_BEFORE (or SYNC_FILE_RANGE_WAIT_AFTER): wait for
267 * completion of writeout of all pages in the range. This will be used after an
268 * earlier SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE operation to wait
269 * for that operation to complete and to return the result.
270 *
271 * SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE|SYNC_FILE_RANGE_WAIT_AFTER:
272 * a traditional sync() operation. This is a write-for-data-integrity operation
273 * which will ensure that all pages in the range which were dirty on entry to
274 * sys_sync_file_range() are committed to disk.
275 *
276 *
277 * SYNC_FILE_RANGE_WAIT_BEFORE and SYNC_FILE_RANGE_WAIT_AFTER will detect any
278 * I/O errors or ENOSPC conditions and will return those to the caller, after
279 * clearing the EIO and ENOSPC flags in the address_space.
280 *
281 * It should be noted that none of these operations write out the file's
282 * metadata. So unless the application is strictly performing overwrites of
283 * already-instantiated disk blocks, there are no guarantees here that the data
284 * will be available after a crash.
285 */
Al Viro4a0fd5b2013-01-21 15:16:58 -0500286SYSCALL_DEFINE4(sync_file_range, int, fd, loff_t, offset, loff_t, nbytes,
287 unsigned int, flags)
Andrew Mortonf79e2ab2006-03-31 02:30:42 -0800288{
289 int ret;
Al Viro2903ff02012-08-28 12:52:22 -0400290 struct fd f;
Christoph Hellwig7a0ad102009-12-17 14:24:40 +0100291 struct address_space *mapping;
Andrew Mortonf79e2ab2006-03-31 02:30:42 -0800292 loff_t endbyte; /* inclusive */
Andrew Mortonf79e2ab2006-03-31 02:30:42 -0800293 umode_t i_mode;
294
295 ret = -EINVAL;
296 if (flags & ~VALID_FLAGS)
297 goto out;
298
299 endbyte = offset + nbytes;
300
301 if ((s64)offset < 0)
302 goto out;
303 if ((s64)endbyte < 0)
304 goto out;
305 if (endbyte < offset)
306 goto out;
307
308 if (sizeof(pgoff_t) == 4) {
309 if (offset >= (0x100000000ULL << PAGE_CACHE_SHIFT)) {
310 /*
311 * The range starts outside a 32 bit machine's
312 * pagecache addressing capabilities. Let it "succeed"
313 */
314 ret = 0;
315 goto out;
316 }
317 if (endbyte >= (0x100000000ULL << PAGE_CACHE_SHIFT)) {
318 /*
319 * Out to EOF
320 */
321 nbytes = 0;
322 }
323 }
324
325 if (nbytes == 0)
OGAWA Hirofumi111ebb62006-06-23 02:03:26 -0700326 endbyte = LLONG_MAX;
Andrew Mortonf79e2ab2006-03-31 02:30:42 -0800327 else
328 endbyte--; /* inclusive */
329
330 ret = -EBADF;
Al Viro2903ff02012-08-28 12:52:22 -0400331 f = fdget(fd);
332 if (!f.file)
Andrew Mortonf79e2ab2006-03-31 02:30:42 -0800333 goto out;
334
Al Viro496ad9a2013-01-23 17:07:38 -0500335 i_mode = file_inode(f.file)->i_mode;
Andrew Mortonf79e2ab2006-03-31 02:30:42 -0800336 ret = -ESPIPE;
337 if (!S_ISREG(i_mode) && !S_ISBLK(i_mode) && !S_ISDIR(i_mode) &&
338 !S_ISLNK(i_mode))
339 goto out_put;
340
Al Viro2903ff02012-08-28 12:52:22 -0400341 mapping = f.file->f_mapping;
Christoph Hellwig7a0ad102009-12-17 14:24:40 +0100342 if (!mapping) {
343 ret = -EINVAL;
344 goto out_put;
345 }
346
347 ret = 0;
348 if (flags & SYNC_FILE_RANGE_WAIT_BEFORE) {
349 ret = filemap_fdatawait_range(mapping, offset, endbyte);
350 if (ret < 0)
351 goto out_put;
352 }
353
354 if (flags & SYNC_FILE_RANGE_WRITE) {
355 ret = filemap_fdatawrite_range(mapping, offset, endbyte);
356 if (ret < 0)
357 goto out_put;
358 }
359
360 if (flags & SYNC_FILE_RANGE_WAIT_AFTER)
361 ret = filemap_fdatawait_range(mapping, offset, endbyte);
362
Andrew Mortonf79e2ab2006-03-31 02:30:42 -0800363out_put:
Al Viro2903ff02012-08-28 12:52:22 -0400364 fdput(f);
Andrew Mortonf79e2ab2006-03-31 02:30:42 -0800365out:
366 return ret;
367}
368
David Woodhouseedd5cd42007-06-27 14:10:09 -0700369/* It would be nice if people remember that not all the world's an i386
370 when they introduce new system calls */
Al Viro4a0fd5b2013-01-21 15:16:58 -0500371SYSCALL_DEFINE4(sync_file_range2, int, fd, unsigned int, flags,
372 loff_t, offset, loff_t, nbytes)
David Woodhouseedd5cd42007-06-27 14:10:09 -0700373{
374 return sys_sync_file_range(fd, offset, nbytes, flags);
375}