blob: 4487b5560dc8e4f01692c32191bf07184546e20e [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>
8#include <linux/module.h>
Al Viro914e2632006-10-18 13:55:46 -04009#include <linux/sched.h>
Andrew Mortonf79e2ab2006-03-31 02:30:42 -080010#include <linux/writeback.h>
11#include <linux/syscalls.h>
12#include <linux/linkage.h>
13#include <linux/pagemap.h>
David Howellscf9a2ae2006-08-29 19:05:54 +010014#include <linux/quotaops.h>
15#include <linux/buffer_head.h>
Jan Kara5a3e5cb2009-04-27 16:43:48 +020016#include "internal.h"
Andrew Mortonf79e2ab2006-03-31 02:30:42 -080017
18#define VALID_FLAGS (SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE| \
19 SYNC_FILE_RANGE_WAIT_AFTER)
20
Jan Karac15c54f2009-04-27 16:43:52 +020021/*
22 * Do the filesystem syncing work. For simple filesystems sync_inodes_sb(sb, 0)
23 * just dirties buffers with inodes so we have to submit IO for these buffers
24 * via __sync_blockdev(). This also speeds up the wait == 1 case since in that
25 * case write_inode() functions do sync_dirty_buffer() and thus effectively
26 * write one block at a time.
27 */
Jan Kara60b06802009-04-27 16:43:53 +020028static int __sync_filesystem(struct super_block *sb, int wait)
Jan Karac15c54f2009-04-27 16:43:52 +020029{
Jan Karac3f8a402009-04-27 16:43:55 +020030 /* Avoid doing twice syncing and cache pruning for quota sync */
31 if (!wait)
32 writeout_quota_sb(sb, -1);
33 else
34 sync_quota_sb(sb, -1);
Jan Karac15c54f2009-04-27 16:43:52 +020035 sync_inodes_sb(sb, wait);
36 lock_super(sb);
37 if (sb->s_dirt && sb->s_op->write_super)
38 sb->s_op->write_super(sb);
39 unlock_super(sb);
40 if (sb->s_op->sync_fs)
41 sb->s_op->sync_fs(sb, wait);
42 return __sync_blockdev(sb->s_bdev, wait);
43}
44
45/*
46 * Write out and wait upon all dirty data associated with this
47 * superblock. Filesystem data as well as the underlying block
48 * device. Takes the superblock lock.
49 */
Jan Kara60b06802009-04-27 16:43:53 +020050int sync_filesystem(struct super_block *sb)
Jan Karac15c54f2009-04-27 16:43:52 +020051{
52 int ret;
53
Jan Kara60b06802009-04-27 16:43:53 +020054 ret = __sync_filesystem(sb, 0);
Jan Karac15c54f2009-04-27 16:43:52 +020055 if (ret < 0)
56 return ret;
Jan Kara60b06802009-04-27 16:43:53 +020057 return __sync_filesystem(sb, 1);
Jan Karac15c54f2009-04-27 16:43:52 +020058}
Jan Kara60b06802009-04-27 16:43:53 +020059EXPORT_SYMBOL_GPL(sync_filesystem);
Jan Karac15c54f2009-04-27 16:43:52 +020060
61/*
62 * Sync all the data for all the filesystems (called by sys_sync() and
63 * emergency sync)
64 *
65 * This operation is careful to avoid the livelock which could easily happen
66 * if two or more filesystems are being continuously dirtied. s_need_sync
67 * is used only here. We set it against all filesystems and then clear it as
68 * we sync them. So redirtied filesystems are skipped.
69 *
70 * But if process A is currently running sync_filesystems and then process B
71 * calls sync_filesystems as well, process B will set all the s_need_sync
72 * flags again, which will cause process A to resync everything. Fix that with
73 * a local mutex.
74 */
75static void sync_filesystems(int wait)
76{
77 struct super_block *sb;
78 static DEFINE_MUTEX(mutex);
79
80 mutex_lock(&mutex); /* Could be down_interruptible */
81 spin_lock(&sb_lock);
82 list_for_each_entry(sb, &super_blocks, s_list) {
83 if (sb->s_flags & MS_RDONLY)
84 continue;
85 sb->s_need_sync = 1;
86 }
87
88restart:
89 list_for_each_entry(sb, &super_blocks, s_list) {
90 if (!sb->s_need_sync)
91 continue;
92 sb->s_need_sync = 0;
93 if (sb->s_flags & MS_RDONLY)
94 continue; /* hm. Was remounted r/o meanwhile */
95 sb->s_count++;
96 spin_unlock(&sb_lock);
97 down_read(&sb->s_umount);
98 if (sb->s_root)
Jan Kara60b06802009-04-27 16:43:53 +020099 __sync_filesystem(sb, wait);
Jan Karac15c54f2009-04-27 16:43:52 +0200100 up_read(&sb->s_umount);
101 /* restart only when sb is no longer on the list */
102 spin_lock(&sb_lock);
103 if (__put_super_and_need_restart(sb))
104 goto restart;
105 }
106 spin_unlock(&sb_lock);
107 mutex_unlock(&mutex);
108}
109
Heiko Carstensa5f8fa92009-01-14 14:14:11 +0100110SYSCALL_DEFINE0(sync)
David Howellscf9a2ae2006-08-29 19:05:54 +0100111{
Jan Kara5cee5812009-04-27 16:43:51 +0200112 sync_filesystems(0);
113 sync_filesystems(1);
114 if (unlikely(laptop_mode))
115 laptop_sync_completion();
David Howellscf9a2ae2006-08-29 19:05:54 +0100116 return 0;
117}
118
Jens Axboea2a95372009-03-17 09:38:40 +0100119static void do_sync_work(struct work_struct *work)
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 */
125 sync_filesystems(0);
126 sync_filesystems(0);
127 printk("Emergency Sync complete\n");
Jens Axboea2a95372009-03-17 09:38:40 +0100128 kfree(work);
129}
130
David Howellscf9a2ae2006-08-29 19:05:54 +0100131void emergency_sync(void)
132{
Jens Axboea2a95372009-03-17 09:38:40 +0100133 struct work_struct *work;
134
135 work = kmalloc(sizeof(*work), GFP_ATOMIC);
136 if (work) {
137 INIT_WORK(work, do_sync_work);
138 schedule_work(work);
139 }
David Howellscf9a2ae2006-08-29 19:05:54 +0100140}
141
142/*
143 * Generic function to fsync a file.
144 *
145 * filp may be NULL if called via the msync of a vma.
146 */
147int file_fsync(struct file *filp, struct dentry *dentry, int datasync)
148{
149 struct inode * inode = dentry->d_inode;
150 struct super_block * sb;
151 int ret, err;
152
153 /* sync the inode to buffers */
154 ret = write_inode_now(inode, 0);
155
156 /* sync the superblock to buffers */
157 sb = inode->i_sb;
158 lock_super(sb);
OGAWA Hirofumi762873c2008-04-29 00:59:42 -0700159 if (sb->s_dirt && sb->s_op->write_super)
David Howellscf9a2ae2006-08-29 19:05:54 +0100160 sb->s_op->write_super(sb);
161 unlock_super(sb);
162
163 /* .. finally sync the buffers to disk */
164 err = sync_blockdev(sb->s_bdev);
165 if (!ret)
166 ret = err;
167 return ret;
168}
169
Christoph Hellwig4c728ef2008-12-22 21:11:15 +0100170/**
171 * vfs_fsync - perform a fsync or fdatasync on a file
172 * @file: file to sync
173 * @dentry: dentry of @file
174 * @data: only perform a fdatasync operation
175 *
176 * Write back data and metadata for @file to disk. If @datasync is
177 * set only metadata needed to access modified file data is written.
178 *
179 * In case this function is called from nfsd @file may be %NULL and
180 * only @dentry is set. This can only happen when the filesystem
181 * implements the export_operations API.
182 */
183int vfs_fsync(struct file *file, struct dentry *dentry, int datasync)
David Howellscf9a2ae2006-08-29 19:05:54 +0100184{
Christoph Hellwig4c728ef2008-12-22 21:11:15 +0100185 const struct file_operations *fop;
186 struct address_space *mapping;
187 int err, ret;
David Howellscf9a2ae2006-08-29 19:05:54 +0100188
Christoph Hellwig4c728ef2008-12-22 21:11:15 +0100189 /*
190 * Get mapping and operations from the file in case we have
191 * as file, or get the default values for them in case we
192 * don't have a struct file available. Damn nfsd..
193 */
194 if (file) {
195 mapping = file->f_mapping;
196 fop = file->f_op;
197 } else {
198 mapping = dentry->d_inode->i_mapping;
199 fop = dentry->d_inode->i_fop;
200 }
201
202 if (!fop || !fop->fsync) {
David Howellscf9a2ae2006-08-29 19:05:54 +0100203 ret = -EINVAL;
204 goto out;
205 }
206
207 ret = filemap_fdatawrite(mapping);
208
209 /*
210 * We need to protect against concurrent writers, which could cause
211 * livelocks in fsync_buffers_list().
212 */
213 mutex_lock(&mapping->host->i_mutex);
Christoph Hellwig4c728ef2008-12-22 21:11:15 +0100214 err = fop->fsync(file, dentry, datasync);
David Howellscf9a2ae2006-08-29 19:05:54 +0100215 if (!ret)
216 ret = err;
217 mutex_unlock(&mapping->host->i_mutex);
218 err = filemap_fdatawait(mapping);
219 if (!ret)
220 ret = err;
221out:
222 return ret;
223}
Christoph Hellwig4c728ef2008-12-22 21:11:15 +0100224EXPORT_SYMBOL(vfs_fsync);
David Howellscf9a2ae2006-08-29 19:05:54 +0100225
Christoph Hellwig4c728ef2008-12-22 21:11:15 +0100226static int do_fsync(unsigned int fd, int datasync)
David Howellscf9a2ae2006-08-29 19:05:54 +0100227{
228 struct file *file;
229 int ret = -EBADF;
230
231 file = fget(fd);
232 if (file) {
Christoph Hellwig4c728ef2008-12-22 21:11:15 +0100233 ret = vfs_fsync(file, file->f_path.dentry, datasync);
David Howellscf9a2ae2006-08-29 19:05:54 +0100234 fput(file);
235 }
236 return ret;
237}
238
Heiko Carstensa5f8fa92009-01-14 14:14:11 +0100239SYSCALL_DEFINE1(fsync, unsigned int, fd)
David Howellscf9a2ae2006-08-29 19:05:54 +0100240{
Christoph Hellwig4c728ef2008-12-22 21:11:15 +0100241 return do_fsync(fd, 0);
David Howellscf9a2ae2006-08-29 19:05:54 +0100242}
243
Heiko Carstensa5f8fa92009-01-14 14:14:11 +0100244SYSCALL_DEFINE1(fdatasync, unsigned int, fd)
David Howellscf9a2ae2006-08-29 19:05:54 +0100245{
Christoph Hellwig4c728ef2008-12-22 21:11:15 +0100246 return do_fsync(fd, 1);
David Howellscf9a2ae2006-08-29 19:05:54 +0100247}
248
249/*
Andrew Mortonf79e2ab2006-03-31 02:30:42 -0800250 * sys_sync_file_range() permits finely controlled syncing over a segment of
251 * a file in the range offset .. (offset+nbytes-1) inclusive. If nbytes is
252 * zero then sys_sync_file_range() will operate from offset out to EOF.
253 *
254 * The flag bits are:
255 *
256 * SYNC_FILE_RANGE_WAIT_BEFORE: wait upon writeout of all pages in the range
257 * before performing the write.
258 *
259 * SYNC_FILE_RANGE_WRITE: initiate writeout of all those dirty pages in the
Pavel Machekcce77082008-07-23 21:27:36 -0700260 * range which are not presently under writeback. Note that this may block for
261 * significant periods due to exhaustion of disk request structures.
Andrew Mortonf79e2ab2006-03-31 02:30:42 -0800262 *
263 * SYNC_FILE_RANGE_WAIT_AFTER: wait upon writeout of all pages in the range
264 * after performing the write.
265 *
266 * Useful combinations of the flag bits are:
267 *
268 * SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE: ensures that all pages
269 * in the range which were dirty on entry to sys_sync_file_range() are placed
270 * under writeout. This is a start-write-for-data-integrity operation.
271 *
272 * SYNC_FILE_RANGE_WRITE: start writeout of all dirty pages in the range which
273 * are not presently under writeout. This is an asynchronous flush-to-disk
274 * operation. Not suitable for data integrity operations.
275 *
276 * SYNC_FILE_RANGE_WAIT_BEFORE (or SYNC_FILE_RANGE_WAIT_AFTER): wait for
277 * completion of writeout of all pages in the range. This will be used after an
278 * earlier SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE operation to wait
279 * for that operation to complete and to return the result.
280 *
281 * SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE|SYNC_FILE_RANGE_WAIT_AFTER:
282 * a traditional sync() operation. This is a write-for-data-integrity operation
283 * which will ensure that all pages in the range which were dirty on entry to
284 * sys_sync_file_range() are committed to disk.
285 *
286 *
287 * SYNC_FILE_RANGE_WAIT_BEFORE and SYNC_FILE_RANGE_WAIT_AFTER will detect any
288 * I/O errors or ENOSPC conditions and will return those to the caller, after
289 * clearing the EIO and ENOSPC flags in the address_space.
290 *
291 * It should be noted that none of these operations write out the file's
292 * metadata. So unless the application is strictly performing overwrites of
293 * already-instantiated disk blocks, there are no guarantees here that the data
294 * will be available after a crash.
295 */
Heiko Carstens6673e0c2009-01-14 14:14:02 +0100296SYSCALL_DEFINE(sync_file_range)(int fd, loff_t offset, loff_t nbytes,
297 unsigned int flags)
Andrew Mortonf79e2ab2006-03-31 02:30:42 -0800298{
299 int ret;
300 struct file *file;
301 loff_t endbyte; /* inclusive */
302 int fput_needed;
303 umode_t i_mode;
304
305 ret = -EINVAL;
306 if (flags & ~VALID_FLAGS)
307 goto out;
308
309 endbyte = offset + nbytes;
310
311 if ((s64)offset < 0)
312 goto out;
313 if ((s64)endbyte < 0)
314 goto out;
315 if (endbyte < offset)
316 goto out;
317
318 if (sizeof(pgoff_t) == 4) {
319 if (offset >= (0x100000000ULL << PAGE_CACHE_SHIFT)) {
320 /*
321 * The range starts outside a 32 bit machine's
322 * pagecache addressing capabilities. Let it "succeed"
323 */
324 ret = 0;
325 goto out;
326 }
327 if (endbyte >= (0x100000000ULL << PAGE_CACHE_SHIFT)) {
328 /*
329 * Out to EOF
330 */
331 nbytes = 0;
332 }
333 }
334
335 if (nbytes == 0)
OGAWA Hirofumi111ebb62006-06-23 02:03:26 -0700336 endbyte = LLONG_MAX;
Andrew Mortonf79e2ab2006-03-31 02:30:42 -0800337 else
338 endbyte--; /* inclusive */
339
340 ret = -EBADF;
341 file = fget_light(fd, &fput_needed);
342 if (!file)
343 goto out;
344
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800345 i_mode = file->f_path.dentry->d_inode->i_mode;
Andrew Mortonf79e2ab2006-03-31 02:30:42 -0800346 ret = -ESPIPE;
347 if (!S_ISREG(i_mode) && !S_ISBLK(i_mode) && !S_ISDIR(i_mode) &&
348 !S_ISLNK(i_mode))
349 goto out_put;
350
Mark Fashehef51c972007-05-08 00:27:10 -0700351 ret = do_sync_mapping_range(file->f_mapping, offset, endbyte, flags);
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
David Woodhouseedd5cd42007-06-27 14:10:09 -0700383
Andrew Mortonf79e2ab2006-03-31 02:30:42 -0800384/*
385 * `endbyte' is inclusive
386 */
Mark Fasheh5b04aa32007-03-01 11:01:55 -0800387int do_sync_mapping_range(struct address_space *mapping, loff_t offset,
388 loff_t endbyte, unsigned int flags)
Andrew Mortonf79e2ab2006-03-31 02:30:42 -0800389{
390 int ret;
Andrew Mortonf79e2ab2006-03-31 02:30:42 -0800391
Andrew Mortonf79e2ab2006-03-31 02:30:42 -0800392 if (!mapping) {
393 ret = -EINVAL;
394 goto out;
395 }
396
397 ret = 0;
398 if (flags & SYNC_FILE_RANGE_WAIT_BEFORE) {
399 ret = wait_on_page_writeback_range(mapping,
400 offset >> PAGE_CACHE_SHIFT,
401 endbyte >> PAGE_CACHE_SHIFT);
402 if (ret < 0)
403 goto out;
404 }
405
406 if (flags & SYNC_FILE_RANGE_WRITE) {
407 ret = __filemap_fdatawrite_range(mapping, offset, endbyte,
Nick Pigginee53a892009-01-06 14:39:12 -0800408 WB_SYNC_ALL);
Andrew Mortonf79e2ab2006-03-31 02:30:42 -0800409 if (ret < 0)
410 goto out;
411 }
412
413 if (flags & SYNC_FILE_RANGE_WAIT_AFTER) {
414 ret = wait_on_page_writeback_range(mapping,
415 offset >> PAGE_CACHE_SHIFT,
416 endbyte >> PAGE_CACHE_SHIFT);
417 }
418out:
419 return ret;
420}
Mark Fasheh5b04aa32007-03-01 11:01:55 -0800421EXPORT_SYMBOL_GPL(do_sync_mapping_range);