blob: e8cbd415e50a9b468656ff0b933ca09925d4e340 [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>
Al Viro914e2632006-10-18 13:55:46 -040010#include <linux/sched.h>
Andrew Mortonf79e2ab2006-03-31 02:30:42 -080011#include <linux/writeback.h>
12#include <linux/syscalls.h>
13#include <linux/linkage.h>
14#include <linux/pagemap.h>
David Howellscf9a2ae2006-08-29 19:05:54 +010015#include <linux/quotaops.h>
16#include <linux/buffer_head.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 */
Jörn Engel5129a462010-04-25 08:54:42 +020036 if (!sb->s_bdi || 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
Jens Axboee913fc82010-05-17 12:55:07 +020045 writeback_inodes_sb_locked(sb);
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{
82 if (!(sb->s_flags & MS_RDONLY) && sb->s_bdi)
83 __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{
Jens Axboe03ba3782009-09-09 09:08:54 +0200100 wakeup_flusher_threads(0);
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
131/*
132 * Generic function to fsync a file.
133 *
134 * filp may be NULL if called via the msync of a vma.
135 */
136int file_fsync(struct file *filp, struct dentry *dentry, int datasync)
137{
138 struct inode * inode = dentry->d_inode;
139 struct super_block * sb;
140 int ret, err;
141
142 /* sync the inode to buffers */
143 ret = write_inode_now(inode, 0);
144
145 /* sync the superblock to buffers */
146 sb = inode->i_sb;
OGAWA Hirofumi762873c2008-04-29 00:59:42 -0700147 if (sb->s_dirt && sb->s_op->write_super)
David Howellscf9a2ae2006-08-29 19:05:54 +0100148 sb->s_op->write_super(sb);
David Howellscf9a2ae2006-08-29 19:05:54 +0100149
150 /* .. finally sync the buffers to disk */
151 err = sync_blockdev(sb->s_bdev);
152 if (!ret)
153 ret = err;
154 return ret;
155}
H Hartley Sweeten1fe72ea2009-09-22 16:43:51 -0700156EXPORT_SYMBOL(file_fsync);
David Howellscf9a2ae2006-08-29 19:05:54 +0100157
Christoph Hellwig4c728ef2008-12-22 21:11:15 +0100158/**
Jan Kara148f9482009-08-17 19:52:36 +0200159 * vfs_fsync_range - helper to sync a range of data & metadata to disk
Christoph Hellwig4c728ef2008-12-22 21:11:15 +0100160 * @file: file to sync
Jan Kara148f9482009-08-17 19:52:36 +0200161 * @start: offset in bytes of the beginning of data range to sync
162 * @end: offset in bytes of the end of data range (inclusive)
163 * @datasync: perform only datasync
Christoph Hellwig4c728ef2008-12-22 21:11:15 +0100164 *
Jan Kara148f9482009-08-17 19:52:36 +0200165 * Write back data in range @start..@end and metadata for @file to disk. If
166 * @datasync is set only metadata needed to access modified file data is
167 * written.
Christoph Hellwig4c728ef2008-12-22 21:11:15 +0100168 */
Christoph Hellwig8018ab02010-03-22 17:32:25 +0100169int vfs_fsync_range(struct file *file, loff_t start, loff_t end, int datasync)
David Howellscf9a2ae2006-08-29 19:05:54 +0100170{
Christoph Hellwig8018ab02010-03-22 17:32:25 +0100171 struct address_space *mapping = file->f_mapping;
Christoph Hellwig4c728ef2008-12-22 21:11:15 +0100172 int err, ret;
David Howellscf9a2ae2006-08-29 19:05:54 +0100173
Christoph Hellwig8018ab02010-03-22 17:32:25 +0100174 if (!file->f_op || !file->f_op->fsync) {
David Howellscf9a2ae2006-08-29 19:05:54 +0100175 ret = -EINVAL;
176 goto out;
177 }
178
Christoph Hellwig2daea672009-09-03 12:39:39 +0200179 ret = filemap_write_and_wait_range(mapping, start, end);
David Howellscf9a2ae2006-08-29 19:05:54 +0100180
181 /*
182 * We need to protect against concurrent writers, which could cause
183 * livelocks in fsync_buffers_list().
184 */
185 mutex_lock(&mapping->host->i_mutex);
Christoph Hellwig8018ab02010-03-22 17:32:25 +0100186 err = file->f_op->fsync(file, file->f_path.dentry, datasync);
David Howellscf9a2ae2006-08-29 19:05:54 +0100187 if (!ret)
188 ret = err;
189 mutex_unlock(&mapping->host->i_mutex);
Jan Kara148f9482009-08-17 19:52:36 +0200190
David Howellscf9a2ae2006-08-29 19:05:54 +0100191out:
192 return ret;
193}
Jan Kara148f9482009-08-17 19:52:36 +0200194EXPORT_SYMBOL(vfs_fsync_range);
195
196/**
197 * vfs_fsync - perform a fsync or fdatasync on a file
198 * @file: file to sync
Jan Kara148f9482009-08-17 19:52:36 +0200199 * @datasync: only perform a fdatasync operation
200 *
201 * Write back data and metadata for @file to disk. If @datasync is
202 * set only metadata needed to access modified file data is written.
Jan Kara148f9482009-08-17 19:52:36 +0200203 */
Christoph Hellwig8018ab02010-03-22 17:32:25 +0100204int vfs_fsync(struct file *file, int datasync)
Jan Kara148f9482009-08-17 19:52:36 +0200205{
Christoph Hellwig8018ab02010-03-22 17:32:25 +0100206 return vfs_fsync_range(file, 0, LLONG_MAX, datasync);
Jan Kara148f9482009-08-17 19:52:36 +0200207}
Christoph Hellwig4c728ef2008-12-22 21:11:15 +0100208EXPORT_SYMBOL(vfs_fsync);
David Howellscf9a2ae2006-08-29 19:05:54 +0100209
Christoph Hellwig4c728ef2008-12-22 21:11:15 +0100210static int do_fsync(unsigned int fd, int datasync)
David Howellscf9a2ae2006-08-29 19:05:54 +0100211{
212 struct file *file;
213 int ret = -EBADF;
214
215 file = fget(fd);
216 if (file) {
Christoph Hellwig8018ab02010-03-22 17:32:25 +0100217 ret = vfs_fsync(file, datasync);
David Howellscf9a2ae2006-08-29 19:05:54 +0100218 fput(file);
219 }
220 return ret;
221}
222
Heiko Carstensa5f8fa92009-01-14 14:14:11 +0100223SYSCALL_DEFINE1(fsync, unsigned int, fd)
David Howellscf9a2ae2006-08-29 19:05:54 +0100224{
Christoph Hellwig4c728ef2008-12-22 21:11:15 +0100225 return do_fsync(fd, 0);
David Howellscf9a2ae2006-08-29 19:05:54 +0100226}
227
Heiko Carstensa5f8fa92009-01-14 14:14:11 +0100228SYSCALL_DEFINE1(fdatasync, unsigned int, fd)
David Howellscf9a2ae2006-08-29 19:05:54 +0100229{
Christoph Hellwig4c728ef2008-12-22 21:11:15 +0100230 return do_fsync(fd, 1);
David Howellscf9a2ae2006-08-29 19:05:54 +0100231}
232
Jan Kara148f9482009-08-17 19:52:36 +0200233/**
234 * generic_write_sync - perform syncing after a write if file / inode is sync
235 * @file: file to which the write happened
236 * @pos: offset where the write started
237 * @count: length of the write
238 *
239 * This is just a simple wrapper about our general syncing function.
240 */
241int generic_write_sync(struct file *file, loff_t pos, loff_t count)
242{
Christoph Hellwig6b2f3d12009-10-27 11:05:28 +0100243 if (!(file->f_flags & O_DSYNC) && !IS_SYNC(file->f_mapping->host))
Jan Kara148f9482009-08-17 19:52:36 +0200244 return 0;
Christoph Hellwig8018ab02010-03-22 17:32:25 +0100245 return vfs_fsync_range(file, pos, pos + count - 1,
Christoph Hellwig6b2f3d12009-10-27 11:05:28 +0100246 (file->f_flags & __O_SYNC) ? 0 : 1);
Jan Kara148f9482009-08-17 19:52:36 +0200247}
248EXPORT_SYMBOL(generic_write_sync);
249
David Howellscf9a2ae2006-08-29 19:05:54 +0100250/*
Andrew Mortonf79e2ab2006-03-31 02:30:42 -0800251 * sys_sync_file_range() permits finely controlled syncing over a segment of
252 * a file in the range offset .. (offset+nbytes-1) inclusive. If nbytes is
253 * zero then sys_sync_file_range() will operate from offset out to EOF.
254 *
255 * The flag bits are:
256 *
257 * SYNC_FILE_RANGE_WAIT_BEFORE: wait upon writeout of all pages in the range
258 * before performing the write.
259 *
260 * SYNC_FILE_RANGE_WRITE: initiate writeout of all those dirty pages in the
Pavel Machekcce77082008-07-23 21:27:36 -0700261 * range which are not presently under writeback. Note that this may block for
262 * significant periods due to exhaustion of disk request structures.
Andrew Mortonf79e2ab2006-03-31 02:30:42 -0800263 *
264 * SYNC_FILE_RANGE_WAIT_AFTER: wait upon writeout of all pages in the range
265 * after performing the write.
266 *
267 * Useful combinations of the flag bits are:
268 *
269 * SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE: ensures that all pages
270 * in the range which were dirty on entry to sys_sync_file_range() are placed
271 * under writeout. This is a start-write-for-data-integrity operation.
272 *
273 * SYNC_FILE_RANGE_WRITE: start writeout of all dirty pages in the range which
274 * are not presently under writeout. This is an asynchronous flush-to-disk
275 * operation. Not suitable for data integrity operations.
276 *
277 * SYNC_FILE_RANGE_WAIT_BEFORE (or SYNC_FILE_RANGE_WAIT_AFTER): wait for
278 * completion of writeout of all pages in the range. This will be used after an
279 * earlier SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE operation to wait
280 * for that operation to complete and to return the result.
281 *
282 * SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE|SYNC_FILE_RANGE_WAIT_AFTER:
283 * a traditional sync() operation. This is a write-for-data-integrity operation
284 * which will ensure that all pages in the range which were dirty on entry to
285 * sys_sync_file_range() are committed to disk.
286 *
287 *
288 * SYNC_FILE_RANGE_WAIT_BEFORE and SYNC_FILE_RANGE_WAIT_AFTER will detect any
289 * I/O errors or ENOSPC conditions and will return those to the caller, after
290 * clearing the EIO and ENOSPC flags in the address_space.
291 *
292 * It should be noted that none of these operations write out the file's
293 * metadata. So unless the application is strictly performing overwrites of
294 * already-instantiated disk blocks, there are no guarantees here that the data
295 * will be available after a crash.
296 */
Heiko Carstens6673e0c2009-01-14 14:14:02 +0100297SYSCALL_DEFINE(sync_file_range)(int fd, loff_t offset, loff_t nbytes,
298 unsigned int flags)
Andrew Mortonf79e2ab2006-03-31 02:30:42 -0800299{
300 int ret;
301 struct file *file;
Christoph Hellwig7a0ad102009-12-17 14:24:40 +0100302 struct address_space *mapping;
Andrew Mortonf79e2ab2006-03-31 02:30:42 -0800303 loff_t endbyte; /* inclusive */
304 int fput_needed;
305 umode_t i_mode;
306
307 ret = -EINVAL;
308 if (flags & ~VALID_FLAGS)
309 goto out;
310
311 endbyte = offset + nbytes;
312
313 if ((s64)offset < 0)
314 goto out;
315 if ((s64)endbyte < 0)
316 goto out;
317 if (endbyte < offset)
318 goto out;
319
320 if (sizeof(pgoff_t) == 4) {
321 if (offset >= (0x100000000ULL << PAGE_CACHE_SHIFT)) {
322 /*
323 * The range starts outside a 32 bit machine's
324 * pagecache addressing capabilities. Let it "succeed"
325 */
326 ret = 0;
327 goto out;
328 }
329 if (endbyte >= (0x100000000ULL << PAGE_CACHE_SHIFT)) {
330 /*
331 * Out to EOF
332 */
333 nbytes = 0;
334 }
335 }
336
337 if (nbytes == 0)
OGAWA Hirofumi111ebb62006-06-23 02:03:26 -0700338 endbyte = LLONG_MAX;
Andrew Mortonf79e2ab2006-03-31 02:30:42 -0800339 else
340 endbyte--; /* inclusive */
341
342 ret = -EBADF;
343 file = fget_light(fd, &fput_needed);
344 if (!file)
345 goto out;
346
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800347 i_mode = file->f_path.dentry->d_inode->i_mode;
Andrew Mortonf79e2ab2006-03-31 02:30:42 -0800348 ret = -ESPIPE;
349 if (!S_ISREG(i_mode) && !S_ISBLK(i_mode) && !S_ISDIR(i_mode) &&
350 !S_ISLNK(i_mode))
351 goto out_put;
352
Christoph Hellwig7a0ad102009-12-17 14:24:40 +0100353 mapping = file->f_mapping;
354 if (!mapping) {
355 ret = -EINVAL;
356 goto out_put;
357 }
358
359 ret = 0;
360 if (flags & SYNC_FILE_RANGE_WAIT_BEFORE) {
361 ret = filemap_fdatawait_range(mapping, offset, endbyte);
362 if (ret < 0)
363 goto out_put;
364 }
365
366 if (flags & SYNC_FILE_RANGE_WRITE) {
367 ret = filemap_fdatawrite_range(mapping, offset, endbyte);
368 if (ret < 0)
369 goto out_put;
370 }
371
372 if (flags & SYNC_FILE_RANGE_WAIT_AFTER)
373 ret = filemap_fdatawait_range(mapping, offset, endbyte);
374
Andrew Mortonf79e2ab2006-03-31 02:30:42 -0800375out_put:
376 fput_light(file, fput_needed);
377out:
378 return ret;
379}
Heiko Carstens6673e0c2009-01-14 14:14:02 +0100380#ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
381asmlinkage long SyS_sync_file_range(long fd, loff_t offset, loff_t nbytes,
382 long flags)
383{
384 return SYSC_sync_file_range((int) fd, offset, nbytes,
385 (unsigned int) flags);
386}
387SYSCALL_ALIAS(sys_sync_file_range, SyS_sync_file_range);
388#endif
Andrew Mortonf79e2ab2006-03-31 02:30:42 -0800389
David Woodhouseedd5cd42007-06-27 14:10:09 -0700390/* It would be nice if people remember that not all the world's an i386
391 when they introduce new system calls */
Heiko Carstens6673e0c2009-01-14 14:14:02 +0100392SYSCALL_DEFINE(sync_file_range2)(int fd, unsigned int flags,
393 loff_t offset, loff_t nbytes)
David Woodhouseedd5cd42007-06-27 14:10:09 -0700394{
395 return sys_sync_file_range(fd, offset, nbytes, flags);
396}
Heiko Carstens6673e0c2009-01-14 14:14:02 +0100397#ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
398asmlinkage long SyS_sync_file_range2(long fd, long flags,
399 loff_t offset, loff_t nbytes)
400{
401 return SYSC_sync_file_range2((int) fd, (unsigned int) flags,
402 offset, nbytes);
403}
404SYSCALL_ALIAS(sys_sync_file_range2, SyS_sync_file_range2);
405#endif