blob: f3501ef392354c9a4ae7a13a63b6aa98f6ec9b3f [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>
Andrew Mortonf79e2ab2006-03-31 02:30:42 -08009#include <linux/module.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{
Jens Axboe32a88aa2009-09-16 15:02:33 +020032 /*
33 * This should be safe, as we require bdi backing to actually
34 * write out data in the first place
35 */
Jens Axboe95f28602011-03-17 11:13:12 +010036 if (sb->s_bdi == &noop_backing_dev_info)
Jens Axboe32a88aa2009-09-16 15:02:33 +020037 return 0;
38
Christoph Hellwig5fb324a2010-02-16 03:44:52 -050039 if (sb->s_qcop && sb->s_qcop->quota_sync)
40 sb->s_qcop->quota_sync(sb, -1, wait);
41
42 if (wait)
Jens Axboed8a85592009-09-02 12:34:32 +020043 sync_inodes_sb(sb);
Christoph Hellwig5fb324a2010-02-16 03:44:52 -050044 else
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -060045 writeback_inodes_sb(sb, WB_REASON_SYNC);
Christoph Hellwig5fb324a2010-02-16 03:44:52 -050046
Jan Karac15c54f2009-04-27 16:43:52 +020047 if (sb->s_op->sync_fs)
48 sb->s_op->sync_fs(sb, wait);
49 return __sync_blockdev(sb->s_bdev, wait);
50}
51
52/*
53 * Write out and wait upon all dirty data associated with this
54 * superblock. Filesystem data as well as the underlying block
55 * device. Takes the superblock lock.
56 */
Jan Kara60b06802009-04-27 16:43:53 +020057int sync_filesystem(struct super_block *sb)
Jan Karac15c54f2009-04-27 16:43:52 +020058{
59 int ret;
60
Christoph Hellwig5af79262009-05-05 15:41:25 +020061 /*
62 * We need to be protected against the filesystem going from
63 * r/o to r/w or vice versa.
64 */
65 WARN_ON(!rwsem_is_locked(&sb->s_umount));
66
67 /*
68 * No point in syncing out anything if the filesystem is read-only.
69 */
70 if (sb->s_flags & MS_RDONLY)
71 return 0;
72
Jan Kara60b06802009-04-27 16:43:53 +020073 ret = __sync_filesystem(sb, 0);
Jan Karac15c54f2009-04-27 16:43:52 +020074 if (ret < 0)
75 return ret;
Jan Kara60b06802009-04-27 16:43:53 +020076 return __sync_filesystem(sb, 1);
Jan Karac15c54f2009-04-27 16:43:52 +020077}
Jan Kara60b06802009-04-27 16:43:53 +020078EXPORT_SYMBOL_GPL(sync_filesystem);
Jan Karac15c54f2009-04-27 16:43:52 +020079
Al Viro01a05b32010-03-23 06:06:58 -040080static void sync_one_sb(struct super_block *sb, void *arg)
81{
Jens Axboe95f28602011-03-17 11:13:12 +010082 if (!(sb->s_flags & MS_RDONLY))
Al Viro01a05b32010-03-23 06:06:58 -040083 __sync_filesystem(sb, *(int *)arg);
84}
Jan Karac15c54f2009-04-27 16:43:52 +020085/*
86 * Sync all the data for all the filesystems (called by sys_sync() and
87 * emergency sync)
Jan Karac15c54f2009-04-27 16:43:52 +020088 */
89static void sync_filesystems(int wait)
90{
Al Viro01a05b32010-03-23 06:06:58 -040091 iterate_supers(sync_one_sb, &wait);
Jan Karac15c54f2009-04-27 16:43:52 +020092}
93
Zhang, Yanmin3beab0b2009-07-05 12:08:08 -070094/*
95 * sync everything. Start out by waking pdflush, because that writes back
96 * all queues in parallel.
97 */
Heiko Carstensa5f8fa92009-01-14 14:14:11 +010098SYSCALL_DEFINE0(sync)
David Howellscf9a2ae2006-08-29 19:05:54 +010099{
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -0600100 wakeup_flusher_threads(0, WB_REASON_SYNC);
Jan Kara5cee5812009-04-27 16:43:51 +0200101 sync_filesystems(0);
102 sync_filesystems(1);
103 if (unlikely(laptop_mode))
104 laptop_sync_completion();
David Howellscf9a2ae2006-08-29 19:05:54 +0100105 return 0;
106}
107
Jens Axboea2a95372009-03-17 09:38:40 +0100108static void do_sync_work(struct work_struct *work)
109{
Jan Kara5cee5812009-04-27 16:43:51 +0200110 /*
111 * Sync twice to reduce the possibility we skipped some inodes / pages
112 * because they were temporarily locked
113 */
114 sync_filesystems(0);
115 sync_filesystems(0);
116 printk("Emergency Sync complete\n");
Jens Axboea2a95372009-03-17 09:38:40 +0100117 kfree(work);
118}
119
David Howellscf9a2ae2006-08-29 19:05:54 +0100120void emergency_sync(void)
121{
Jens Axboea2a95372009-03-17 09:38:40 +0100122 struct work_struct *work;
123
124 work = kmalloc(sizeof(*work), GFP_ATOMIC);
125 if (work) {
126 INIT_WORK(work, do_sync_work);
127 schedule_work(work);
128 }
David Howellscf9a2ae2006-08-29 19:05:54 +0100129}
130
Sage Weilb7ed78f2011-03-10 11:31:30 -0800131/*
132 * sync a single super
133 */
134SYSCALL_DEFINE1(syncfs, int, fd)
135{
136 struct file *file;
137 struct super_block *sb;
138 int ret;
139 int fput_needed;
140
141 file = fget_light(fd, &fput_needed);
142 if (!file)
143 return -EBADF;
144 sb = file->f_dentry->d_sb;
145
146 down_read(&sb->s_umount);
147 ret = sync_filesystem(sb);
148 up_read(&sb->s_umount);
149
150 fput_light(file, fput_needed);
151 return ret;
152}
153
Christoph Hellwig4c728ef2008-12-22 21:11:15 +0100154/**
Jan Kara148f9482009-08-17 19:52:36 +0200155 * vfs_fsync_range - helper to sync a range of data & metadata to disk
Christoph Hellwig4c728ef2008-12-22 21:11:15 +0100156 * @file: file to sync
Jan Kara148f9482009-08-17 19:52:36 +0200157 * @start: offset in bytes of the beginning of data range to sync
158 * @end: offset in bytes of the end of data range (inclusive)
159 * @datasync: perform only datasync
Christoph Hellwig4c728ef2008-12-22 21:11:15 +0100160 *
Jan Kara148f9482009-08-17 19:52:36 +0200161 * Write back data in range @start..@end and metadata for @file to disk. If
162 * @datasync is set only metadata needed to access modified file data is
163 * written.
Christoph Hellwig4c728ef2008-12-22 21:11:15 +0100164 */
Christoph Hellwig8018ab02010-03-22 17:32:25 +0100165int vfs_fsync_range(struct file *file, loff_t start, loff_t end, int datasync)
David Howellscf9a2ae2006-08-29 19:05:54 +0100166{
Josef Bacik02c24a82011-07-16 20:44:56 -0400167 if (!file->f_op || !file->f_op->fsync)
168 return -EINVAL;
169 return file->f_op->fsync(file, start, end, datasync);
David Howellscf9a2ae2006-08-29 19:05:54 +0100170}
Jan Kara148f9482009-08-17 19:52:36 +0200171EXPORT_SYMBOL(vfs_fsync_range);
172
173/**
174 * vfs_fsync - perform a fsync or fdatasync on a file
175 * @file: file to sync
Jan Kara148f9482009-08-17 19:52:36 +0200176 * @datasync: only perform a fdatasync operation
177 *
178 * Write back data and metadata for @file to disk. If @datasync is
179 * set only metadata needed to access modified file data is written.
Jan Kara148f9482009-08-17 19:52:36 +0200180 */
Christoph Hellwig8018ab02010-03-22 17:32:25 +0100181int vfs_fsync(struct file *file, int datasync)
Jan Kara148f9482009-08-17 19:52:36 +0200182{
Christoph Hellwig8018ab02010-03-22 17:32:25 +0100183 return vfs_fsync_range(file, 0, LLONG_MAX, datasync);
Jan Kara148f9482009-08-17 19:52:36 +0200184}
Christoph Hellwig4c728ef2008-12-22 21:11:15 +0100185EXPORT_SYMBOL(vfs_fsync);
David Howellscf9a2ae2006-08-29 19:05:54 +0100186
Christoph Hellwig4c728ef2008-12-22 21:11:15 +0100187static int do_fsync(unsigned int fd, int datasync)
David Howellscf9a2ae2006-08-29 19:05:54 +0100188{
189 struct file *file;
190 int ret = -EBADF;
191
192 file = fget(fd);
193 if (file) {
Christoph Hellwig8018ab02010-03-22 17:32:25 +0100194 ret = vfs_fsync(file, datasync);
David Howellscf9a2ae2006-08-29 19:05:54 +0100195 fput(file);
196 }
197 return ret;
198}
199
Heiko Carstensa5f8fa92009-01-14 14:14:11 +0100200SYSCALL_DEFINE1(fsync, unsigned int, fd)
David Howellscf9a2ae2006-08-29 19:05:54 +0100201{
Christoph Hellwig4c728ef2008-12-22 21:11:15 +0100202 return do_fsync(fd, 0);
David Howellscf9a2ae2006-08-29 19:05:54 +0100203}
204
Heiko Carstensa5f8fa92009-01-14 14:14:11 +0100205SYSCALL_DEFINE1(fdatasync, unsigned int, fd)
David Howellscf9a2ae2006-08-29 19:05:54 +0100206{
Christoph Hellwig4c728ef2008-12-22 21:11:15 +0100207 return do_fsync(fd, 1);
David Howellscf9a2ae2006-08-29 19:05:54 +0100208}
209
Jan Kara148f9482009-08-17 19:52:36 +0200210/**
211 * generic_write_sync - perform syncing after a write if file / inode is sync
212 * @file: file to which the write happened
213 * @pos: offset where the write started
214 * @count: length of the write
215 *
216 * This is just a simple wrapper about our general syncing function.
217 */
218int generic_write_sync(struct file *file, loff_t pos, loff_t count)
219{
Christoph Hellwig6b2f3d12009-10-27 11:05:28 +0100220 if (!(file->f_flags & O_DSYNC) && !IS_SYNC(file->f_mapping->host))
Jan Kara148f9482009-08-17 19:52:36 +0200221 return 0;
Christoph Hellwig8018ab02010-03-22 17:32:25 +0100222 return vfs_fsync_range(file, pos, pos + count - 1,
Christoph Hellwig6b2f3d12009-10-27 11:05:28 +0100223 (file->f_flags & __O_SYNC) ? 0 : 1);
Jan Kara148f9482009-08-17 19:52:36 +0200224}
225EXPORT_SYMBOL(generic_write_sync);
226
David Howellscf9a2ae2006-08-29 19:05:54 +0100227/*
Andrew Mortonf79e2ab2006-03-31 02:30:42 -0800228 * sys_sync_file_range() permits finely controlled syncing over a segment of
229 * a file in the range offset .. (offset+nbytes-1) inclusive. If nbytes is
230 * zero then sys_sync_file_range() will operate from offset out to EOF.
231 *
232 * The flag bits are:
233 *
234 * SYNC_FILE_RANGE_WAIT_BEFORE: wait upon writeout of all pages in the range
235 * before performing the write.
236 *
237 * SYNC_FILE_RANGE_WRITE: initiate writeout of all those dirty pages in the
Pavel Machekcce77082008-07-23 21:27:36 -0700238 * range which are not presently under writeback. Note that this may block for
239 * significant periods due to exhaustion of disk request structures.
Andrew Mortonf79e2ab2006-03-31 02:30:42 -0800240 *
241 * SYNC_FILE_RANGE_WAIT_AFTER: wait upon writeout of all pages in the range
242 * after performing the write.
243 *
244 * Useful combinations of the flag bits are:
245 *
246 * SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE: ensures that all pages
247 * in the range which were dirty on entry to sys_sync_file_range() are placed
248 * under writeout. This is a start-write-for-data-integrity operation.
249 *
250 * SYNC_FILE_RANGE_WRITE: start writeout of all dirty pages in the range which
251 * are not presently under writeout. This is an asynchronous flush-to-disk
252 * operation. Not suitable for data integrity operations.
253 *
254 * SYNC_FILE_RANGE_WAIT_BEFORE (or SYNC_FILE_RANGE_WAIT_AFTER): wait for
255 * completion of writeout of all pages in the range. This will be used after an
256 * earlier SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE operation to wait
257 * for that operation to complete and to return the result.
258 *
259 * SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE|SYNC_FILE_RANGE_WAIT_AFTER:
260 * a traditional sync() operation. This is a write-for-data-integrity operation
261 * which will ensure that all pages in the range which were dirty on entry to
262 * sys_sync_file_range() are committed to disk.
263 *
264 *
265 * SYNC_FILE_RANGE_WAIT_BEFORE and SYNC_FILE_RANGE_WAIT_AFTER will detect any
266 * I/O errors or ENOSPC conditions and will return those to the caller, after
267 * clearing the EIO and ENOSPC flags in the address_space.
268 *
269 * It should be noted that none of these operations write out the file's
270 * metadata. So unless the application is strictly performing overwrites of
271 * already-instantiated disk blocks, there are no guarantees here that the data
272 * will be available after a crash.
273 */
Heiko Carstens6673e0c2009-01-14 14:14:02 +0100274SYSCALL_DEFINE(sync_file_range)(int fd, loff_t offset, loff_t nbytes,
275 unsigned int flags)
Andrew Mortonf79e2ab2006-03-31 02:30:42 -0800276{
277 int ret;
278 struct file *file;
Christoph Hellwig7a0ad102009-12-17 14:24:40 +0100279 struct address_space *mapping;
Andrew Mortonf79e2ab2006-03-31 02:30:42 -0800280 loff_t endbyte; /* inclusive */
281 int fput_needed;
282 umode_t i_mode;
283
284 ret = -EINVAL;
285 if (flags & ~VALID_FLAGS)
286 goto out;
287
288 endbyte = offset + nbytes;
289
290 if ((s64)offset < 0)
291 goto out;
292 if ((s64)endbyte < 0)
293 goto out;
294 if (endbyte < offset)
295 goto out;
296
297 if (sizeof(pgoff_t) == 4) {
298 if (offset >= (0x100000000ULL << PAGE_CACHE_SHIFT)) {
299 /*
300 * The range starts outside a 32 bit machine's
301 * pagecache addressing capabilities. Let it "succeed"
302 */
303 ret = 0;
304 goto out;
305 }
306 if (endbyte >= (0x100000000ULL << PAGE_CACHE_SHIFT)) {
307 /*
308 * Out to EOF
309 */
310 nbytes = 0;
311 }
312 }
313
314 if (nbytes == 0)
OGAWA Hirofumi111ebb62006-06-23 02:03:26 -0700315 endbyte = LLONG_MAX;
Andrew Mortonf79e2ab2006-03-31 02:30:42 -0800316 else
317 endbyte--; /* inclusive */
318
319 ret = -EBADF;
320 file = fget_light(fd, &fput_needed);
321 if (!file)
322 goto out;
323
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800324 i_mode = file->f_path.dentry->d_inode->i_mode;
Andrew Mortonf79e2ab2006-03-31 02:30:42 -0800325 ret = -ESPIPE;
326 if (!S_ISREG(i_mode) && !S_ISBLK(i_mode) && !S_ISDIR(i_mode) &&
327 !S_ISLNK(i_mode))
328 goto out_put;
329
Christoph Hellwig7a0ad102009-12-17 14:24:40 +0100330 mapping = file->f_mapping;
331 if (!mapping) {
332 ret = -EINVAL;
333 goto out_put;
334 }
335
336 ret = 0;
337 if (flags & SYNC_FILE_RANGE_WAIT_BEFORE) {
338 ret = filemap_fdatawait_range(mapping, offset, endbyte);
339 if (ret < 0)
340 goto out_put;
341 }
342
343 if (flags & SYNC_FILE_RANGE_WRITE) {
344 ret = filemap_fdatawrite_range(mapping, offset, endbyte);
345 if (ret < 0)
346 goto out_put;
347 }
348
349 if (flags & SYNC_FILE_RANGE_WAIT_AFTER)
350 ret = filemap_fdatawait_range(mapping, offset, endbyte);
351
Andrew Mortonf79e2ab2006-03-31 02:30:42 -0800352out_put:
353 fput_light(file, fput_needed);
354out:
355 return ret;
356}
Heiko Carstens6673e0c2009-01-14 14:14:02 +0100357#ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
358asmlinkage long SyS_sync_file_range(long fd, loff_t offset, loff_t nbytes,
359 long flags)
360{
361 return SYSC_sync_file_range((int) fd, offset, nbytes,
362 (unsigned int) flags);
363}
364SYSCALL_ALIAS(sys_sync_file_range, SyS_sync_file_range);
365#endif
Andrew Mortonf79e2ab2006-03-31 02:30:42 -0800366
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 */
Heiko Carstens6673e0c2009-01-14 14:14:02 +0100369SYSCALL_DEFINE(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}
Heiko Carstens6673e0c2009-01-14 14:14:02 +0100374#ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
375asmlinkage long SyS_sync_file_range2(long fd, long flags,
376 loff_t offset, loff_t nbytes)
377{
378 return SYSC_sync_file_range2((int) fd, (unsigned int) flags,
379 offset, nbytes);
380}
381SYSCALL_ALIAS(sys_sync_file_range2, SyS_sync_file_range2);
382#endif