Commit | Line | Data |
---|---|---|
f79e2abb AM |
1 | /* |
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> | |
914e2637 | 9 | #include <linux/sched.h> |
f79e2abb AM |
10 | #include <linux/writeback.h> |
11 | #include <linux/syscalls.h> | |
12 | #include <linux/linkage.h> | |
13 | #include <linux/pagemap.h> | |
cf9a2ae8 DH |
14 | #include <linux/quotaops.h> |
15 | #include <linux/buffer_head.h> | |
5a3e5cb8 | 16 | #include "internal.h" |
f79e2abb AM |
17 | |
18 | #define VALID_FLAGS (SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE| \ | |
19 | SYNC_FILE_RANGE_WAIT_AFTER) | |
20 | ||
c15c54f5 | 21 | /* |
d8a8559c JA |
22 | * Do the filesystem syncing work. For simple filesystems |
23 | * writeback_inodes_sb(sb) just dirties buffers with inodes so we have to | |
24 | * submit IO for these buffers via __sync_blockdev(). This also speeds up the | |
25 | * wait == 1 case since in that case write_inode() functions do | |
26 | * sync_dirty_buffer() and thus effectively write one block at a time. | |
c15c54f5 | 27 | */ |
60b0680f | 28 | static int __sync_filesystem(struct super_block *sb, int wait) |
c15c54f5 | 29 | { |
32a88aa1 JA |
30 | /* |
31 | * This should be safe, as we require bdi backing to actually | |
32 | * write out data in the first place | |
33 | */ | |
34 | if (!sb->s_bdi) | |
35 | return 0; | |
36 | ||
c3f8a40c | 37 | /* Avoid doing twice syncing and cache pruning for quota sync */ |
d8a8559c | 38 | if (!wait) { |
c3f8a40c | 39 | writeout_quota_sb(sb, -1); |
d8a8559c JA |
40 | writeback_inodes_sb(sb); |
41 | } else { | |
c3f8a40c | 42 | sync_quota_sb(sb, -1); |
d8a8559c JA |
43 | sync_inodes_sb(sb); |
44 | } | |
c15c54f5 JK |
45 | if (sb->s_op->sync_fs) |
46 | sb->s_op->sync_fs(sb, wait); | |
47 | return __sync_blockdev(sb->s_bdev, wait); | |
48 | } | |
49 | ||
50 | /* | |
51 | * Write out and wait upon all dirty data associated with this | |
52 | * superblock. Filesystem data as well as the underlying block | |
53 | * device. Takes the superblock lock. | |
54 | */ | |
60b0680f | 55 | int sync_filesystem(struct super_block *sb) |
c15c54f5 JK |
56 | { |
57 | int ret; | |
58 | ||
5af7926f CH |
59 | /* |
60 | * We need to be protected against the filesystem going from | |
61 | * r/o to r/w or vice versa. | |
62 | */ | |
63 | WARN_ON(!rwsem_is_locked(&sb->s_umount)); | |
64 | ||
65 | /* | |
66 | * No point in syncing out anything if the filesystem is read-only. | |
67 | */ | |
68 | if (sb->s_flags & MS_RDONLY) | |
69 | return 0; | |
70 | ||
60b0680f | 71 | ret = __sync_filesystem(sb, 0); |
c15c54f5 JK |
72 | if (ret < 0) |
73 | return ret; | |
60b0680f | 74 | return __sync_filesystem(sb, 1); |
c15c54f5 | 75 | } |
60b0680f | 76 | EXPORT_SYMBOL_GPL(sync_filesystem); |
c15c54f5 JK |
77 | |
78 | /* | |
79 | * Sync all the data for all the filesystems (called by sys_sync() and | |
80 | * emergency sync) | |
81 | * | |
82 | * This operation is careful to avoid the livelock which could easily happen | |
83 | * if two or more filesystems are being continuously dirtied. s_need_sync | |
84 | * is used only here. We set it against all filesystems and then clear it as | |
85 | * we sync them. So redirtied filesystems are skipped. | |
86 | * | |
87 | * But if process A is currently running sync_filesystems and then process B | |
88 | * calls sync_filesystems as well, process B will set all the s_need_sync | |
89 | * flags again, which will cause process A to resync everything. Fix that with | |
90 | * a local mutex. | |
91 | */ | |
92 | static void sync_filesystems(int wait) | |
93 | { | |
94 | struct super_block *sb; | |
95 | static DEFINE_MUTEX(mutex); | |
96 | ||
97 | mutex_lock(&mutex); /* Could be down_interruptible */ | |
98 | spin_lock(&sb_lock); | |
5af7926f | 99 | list_for_each_entry(sb, &super_blocks, s_list) |
c15c54f5 | 100 | sb->s_need_sync = 1; |
c15c54f5 JK |
101 | |
102 | restart: | |
103 | list_for_each_entry(sb, &super_blocks, s_list) { | |
104 | if (!sb->s_need_sync) | |
105 | continue; | |
106 | sb->s_need_sync = 0; | |
c15c54f5 JK |
107 | sb->s_count++; |
108 | spin_unlock(&sb_lock); | |
5af7926f | 109 | |
c15c54f5 | 110 | down_read(&sb->s_umount); |
32a88aa1 | 111 | if (!(sb->s_flags & MS_RDONLY) && sb->s_root && sb->s_bdi) |
60b0680f | 112 | __sync_filesystem(sb, wait); |
c15c54f5 | 113 | up_read(&sb->s_umount); |
5af7926f | 114 | |
c15c54f5 JK |
115 | /* restart only when sb is no longer on the list */ |
116 | spin_lock(&sb_lock); | |
117 | if (__put_super_and_need_restart(sb)) | |
118 | goto restart; | |
119 | } | |
120 | spin_unlock(&sb_lock); | |
121 | mutex_unlock(&mutex); | |
122 | } | |
123 | ||
3beab0b4 ZY |
124 | /* |
125 | * sync everything. Start out by waking pdflush, because that writes back | |
126 | * all queues in parallel. | |
127 | */ | |
5cee5815 | 128 | SYSCALL_DEFINE0(sync) |
cf9a2ae8 | 129 | { |
03ba3782 | 130 | wakeup_flusher_threads(0); |
5cee5815 JK |
131 | sync_filesystems(0); |
132 | sync_filesystems(1); | |
cf9a2ae8 DH |
133 | if (unlikely(laptop_mode)) |
134 | laptop_sync_completion(); | |
cf9a2ae8 DH |
135 | return 0; |
136 | } | |
137 | ||
a2a9537a JA |
138 | static void do_sync_work(struct work_struct *work) |
139 | { | |
5cee5815 JK |
140 | /* |
141 | * Sync twice to reduce the possibility we skipped some inodes / pages | |
142 | * because they were temporarily locked | |
143 | */ | |
144 | sync_filesystems(0); | |
145 | sync_filesystems(0); | |
146 | printk("Emergency Sync complete\n"); | |
a2a9537a JA |
147 | kfree(work); |
148 | } | |
149 | ||
cf9a2ae8 DH |
150 | void emergency_sync(void) |
151 | { | |
a2a9537a JA |
152 | struct work_struct *work; |
153 | ||
154 | work = kmalloc(sizeof(*work), GFP_ATOMIC); | |
155 | if (work) { | |
156 | INIT_WORK(work, do_sync_work); | |
157 | schedule_work(work); | |
158 | } | |
cf9a2ae8 DH |
159 | } |
160 | ||
161 | /* | |
162 | * Generic function to fsync a file. | |
163 | * | |
164 | * filp may be NULL if called via the msync of a vma. | |
165 | */ | |
166 | int file_fsync(struct file *filp, struct dentry *dentry, int datasync) | |
167 | { | |
168 | struct inode * inode = dentry->d_inode; | |
169 | struct super_block * sb; | |
170 | int ret, err; | |
171 | ||
172 | /* sync the inode to buffers */ | |
173 | ret = write_inode_now(inode, 0); | |
174 | ||
175 | /* sync the superblock to buffers */ | |
176 | sb = inode->i_sb; | |
762873c2 | 177 | if (sb->s_dirt && sb->s_op->write_super) |
cf9a2ae8 | 178 | sb->s_op->write_super(sb); |
cf9a2ae8 DH |
179 | |
180 | /* .. finally sync the buffers to disk */ | |
181 | err = sync_blockdev(sb->s_bdev); | |
182 | if (!ret) | |
183 | ret = err; | |
184 | return ret; | |
185 | } | |
186 | ||
4c728ef5 | 187 | /** |
148f948b | 188 | * vfs_fsync_range - helper to sync a range of data & metadata to disk |
4c728ef5 CH |
189 | * @file: file to sync |
190 | * @dentry: dentry of @file | |
148f948b JK |
191 | * @start: offset in bytes of the beginning of data range to sync |
192 | * @end: offset in bytes of the end of data range (inclusive) | |
193 | * @datasync: perform only datasync | |
4c728ef5 | 194 | * |
148f948b JK |
195 | * Write back data in range @start..@end and metadata for @file to disk. If |
196 | * @datasync is set only metadata needed to access modified file data is | |
197 | * written. | |
4c728ef5 CH |
198 | * |
199 | * In case this function is called from nfsd @file may be %NULL and | |
200 | * only @dentry is set. This can only happen when the filesystem | |
201 | * implements the export_operations API. | |
202 | */ | |
148f948b JK |
203 | int vfs_fsync_range(struct file *file, struct dentry *dentry, loff_t start, |
204 | loff_t end, int datasync) | |
cf9a2ae8 | 205 | { |
4c728ef5 CH |
206 | const struct file_operations *fop; |
207 | struct address_space *mapping; | |
208 | int err, ret; | |
209 | ||
210 | /* | |
211 | * Get mapping and operations from the file in case we have | |
212 | * as file, or get the default values for them in case we | |
213 | * don't have a struct file available. Damn nfsd.. | |
214 | */ | |
215 | if (file) { | |
216 | mapping = file->f_mapping; | |
217 | fop = file->f_op; | |
218 | } else { | |
219 | mapping = dentry->d_inode->i_mapping; | |
220 | fop = dentry->d_inode->i_fop; | |
221 | } | |
cf9a2ae8 | 222 | |
4c728ef5 | 223 | if (!fop || !fop->fsync) { |
cf9a2ae8 DH |
224 | ret = -EINVAL; |
225 | goto out; | |
226 | } | |
227 | ||
2daea67e | 228 | ret = filemap_write_and_wait_range(mapping, start, end); |
cf9a2ae8 DH |
229 | |
230 | /* | |
231 | * We need to protect against concurrent writers, which could cause | |
232 | * livelocks in fsync_buffers_list(). | |
233 | */ | |
234 | mutex_lock(&mapping->host->i_mutex); | |
4c728ef5 | 235 | err = fop->fsync(file, dentry, datasync); |
cf9a2ae8 DH |
236 | if (!ret) |
237 | ret = err; | |
238 | mutex_unlock(&mapping->host->i_mutex); | |
148f948b | 239 | |
cf9a2ae8 DH |
240 | out: |
241 | return ret; | |
242 | } | |
148f948b JK |
243 | EXPORT_SYMBOL(vfs_fsync_range); |
244 | ||
245 | /** | |
246 | * vfs_fsync - perform a fsync or fdatasync on a file | |
247 | * @file: file to sync | |
248 | * @dentry: dentry of @file | |
249 | * @datasync: only perform a fdatasync operation | |
250 | * | |
251 | * Write back data and metadata for @file to disk. If @datasync is | |
252 | * set only metadata needed to access modified file data is written. | |
253 | * | |
254 | * In case this function is called from nfsd @file may be %NULL and | |
255 | * only @dentry is set. This can only happen when the filesystem | |
256 | * implements the export_operations API. | |
257 | */ | |
258 | int vfs_fsync(struct file *file, struct dentry *dentry, int datasync) | |
259 | { | |
260 | return vfs_fsync_range(file, dentry, 0, LLONG_MAX, datasync); | |
261 | } | |
4c728ef5 | 262 | EXPORT_SYMBOL(vfs_fsync); |
cf9a2ae8 | 263 | |
4c728ef5 | 264 | static int do_fsync(unsigned int fd, int datasync) |
cf9a2ae8 DH |
265 | { |
266 | struct file *file; | |
267 | int ret = -EBADF; | |
268 | ||
269 | file = fget(fd); | |
270 | if (file) { | |
4c728ef5 | 271 | ret = vfs_fsync(file, file->f_path.dentry, datasync); |
cf9a2ae8 DH |
272 | fput(file); |
273 | } | |
274 | return ret; | |
275 | } | |
276 | ||
a5f8fa9e | 277 | SYSCALL_DEFINE1(fsync, unsigned int, fd) |
cf9a2ae8 | 278 | { |
4c728ef5 | 279 | return do_fsync(fd, 0); |
cf9a2ae8 DH |
280 | } |
281 | ||
a5f8fa9e | 282 | SYSCALL_DEFINE1(fdatasync, unsigned int, fd) |
cf9a2ae8 | 283 | { |
4c728ef5 | 284 | return do_fsync(fd, 1); |
cf9a2ae8 DH |
285 | } |
286 | ||
148f948b JK |
287 | /** |
288 | * generic_write_sync - perform syncing after a write if file / inode is sync | |
289 | * @file: file to which the write happened | |
290 | * @pos: offset where the write started | |
291 | * @count: length of the write | |
292 | * | |
293 | * This is just a simple wrapper about our general syncing function. | |
294 | */ | |
295 | int generic_write_sync(struct file *file, loff_t pos, loff_t count) | |
296 | { | |
297 | if (!(file->f_flags & O_SYNC) && !IS_SYNC(file->f_mapping->host)) | |
298 | return 0; | |
299 | return vfs_fsync_range(file, file->f_path.dentry, pos, | |
300 | pos + count - 1, 1); | |
301 | } | |
302 | EXPORT_SYMBOL(generic_write_sync); | |
303 | ||
f79e2abb AM |
304 | /* |
305 | * sys_sync_file_range() permits finely controlled syncing over a segment of | |
306 | * a file in the range offset .. (offset+nbytes-1) inclusive. If nbytes is | |
307 | * zero then sys_sync_file_range() will operate from offset out to EOF. | |
308 | * | |
309 | * The flag bits are: | |
310 | * | |
311 | * SYNC_FILE_RANGE_WAIT_BEFORE: wait upon writeout of all pages in the range | |
312 | * before performing the write. | |
313 | * | |
314 | * SYNC_FILE_RANGE_WRITE: initiate writeout of all those dirty pages in the | |
cce77081 PM |
315 | * range which are not presently under writeback. Note that this may block for |
316 | * significant periods due to exhaustion of disk request structures. | |
f79e2abb AM |
317 | * |
318 | * SYNC_FILE_RANGE_WAIT_AFTER: wait upon writeout of all pages in the range | |
319 | * after performing the write. | |
320 | * | |
321 | * Useful combinations of the flag bits are: | |
322 | * | |
323 | * SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE: ensures that all pages | |
324 | * in the range which were dirty on entry to sys_sync_file_range() are placed | |
325 | * under writeout. This is a start-write-for-data-integrity operation. | |
326 | * | |
327 | * SYNC_FILE_RANGE_WRITE: start writeout of all dirty pages in the range which | |
328 | * are not presently under writeout. This is an asynchronous flush-to-disk | |
329 | * operation. Not suitable for data integrity operations. | |
330 | * | |
331 | * SYNC_FILE_RANGE_WAIT_BEFORE (or SYNC_FILE_RANGE_WAIT_AFTER): wait for | |
332 | * completion of writeout of all pages in the range. This will be used after an | |
333 | * earlier SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE operation to wait | |
334 | * for that operation to complete and to return the result. | |
335 | * | |
336 | * SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE|SYNC_FILE_RANGE_WAIT_AFTER: | |
337 | * a traditional sync() operation. This is a write-for-data-integrity operation | |
338 | * which will ensure that all pages in the range which were dirty on entry to | |
339 | * sys_sync_file_range() are committed to disk. | |
340 | * | |
341 | * | |
342 | * SYNC_FILE_RANGE_WAIT_BEFORE and SYNC_FILE_RANGE_WAIT_AFTER will detect any | |
343 | * I/O errors or ENOSPC conditions and will return those to the caller, after | |
344 | * clearing the EIO and ENOSPC flags in the address_space. | |
345 | * | |
346 | * It should be noted that none of these operations write out the file's | |
347 | * metadata. So unless the application is strictly performing overwrites of | |
348 | * already-instantiated disk blocks, there are no guarantees here that the data | |
349 | * will be available after a crash. | |
350 | */ | |
6673e0c3 HC |
351 | SYSCALL_DEFINE(sync_file_range)(int fd, loff_t offset, loff_t nbytes, |
352 | unsigned int flags) | |
f79e2abb AM |
353 | { |
354 | int ret; | |
355 | struct file *file; | |
356 | loff_t endbyte; /* inclusive */ | |
357 | int fput_needed; | |
358 | umode_t i_mode; | |
359 | ||
360 | ret = -EINVAL; | |
361 | if (flags & ~VALID_FLAGS) | |
362 | goto out; | |
363 | ||
364 | endbyte = offset + nbytes; | |
365 | ||
366 | if ((s64)offset < 0) | |
367 | goto out; | |
368 | if ((s64)endbyte < 0) | |
369 | goto out; | |
370 | if (endbyte < offset) | |
371 | goto out; | |
372 | ||
373 | if (sizeof(pgoff_t) == 4) { | |
374 | if (offset >= (0x100000000ULL << PAGE_CACHE_SHIFT)) { | |
375 | /* | |
376 | * The range starts outside a 32 bit machine's | |
377 | * pagecache addressing capabilities. Let it "succeed" | |
378 | */ | |
379 | ret = 0; | |
380 | goto out; | |
381 | } | |
382 | if (endbyte >= (0x100000000ULL << PAGE_CACHE_SHIFT)) { | |
383 | /* | |
384 | * Out to EOF | |
385 | */ | |
386 | nbytes = 0; | |
387 | } | |
388 | } | |
389 | ||
390 | if (nbytes == 0) | |
111ebb6e | 391 | endbyte = LLONG_MAX; |
f79e2abb AM |
392 | else |
393 | endbyte--; /* inclusive */ | |
394 | ||
395 | ret = -EBADF; | |
396 | file = fget_light(fd, &fput_needed); | |
397 | if (!file) | |
398 | goto out; | |
399 | ||
0f7fc9e4 | 400 | i_mode = file->f_path.dentry->d_inode->i_mode; |
f79e2abb AM |
401 | ret = -ESPIPE; |
402 | if (!S_ISREG(i_mode) && !S_ISBLK(i_mode) && !S_ISDIR(i_mode) && | |
403 | !S_ISLNK(i_mode)) | |
404 | goto out_put; | |
405 | ||
ef51c976 | 406 | ret = do_sync_mapping_range(file->f_mapping, offset, endbyte, flags); |
f79e2abb AM |
407 | out_put: |
408 | fput_light(file, fput_needed); | |
409 | out: | |
410 | return ret; | |
411 | } | |
6673e0c3 HC |
412 | #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS |
413 | asmlinkage long SyS_sync_file_range(long fd, loff_t offset, loff_t nbytes, | |
414 | long flags) | |
415 | { | |
416 | return SYSC_sync_file_range((int) fd, offset, nbytes, | |
417 | (unsigned int) flags); | |
418 | } | |
419 | SYSCALL_ALIAS(sys_sync_file_range, SyS_sync_file_range); | |
420 | #endif | |
f79e2abb | 421 | |
edd5cd4a DW |
422 | /* It would be nice if people remember that not all the world's an i386 |
423 | when they introduce new system calls */ | |
6673e0c3 HC |
424 | SYSCALL_DEFINE(sync_file_range2)(int fd, unsigned int flags, |
425 | loff_t offset, loff_t nbytes) | |
edd5cd4a DW |
426 | { |
427 | return sys_sync_file_range(fd, offset, nbytes, flags); | |
428 | } | |
6673e0c3 HC |
429 | #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS |
430 | asmlinkage long SyS_sync_file_range2(long fd, long flags, | |
431 | loff_t offset, loff_t nbytes) | |
432 | { | |
433 | return SYSC_sync_file_range2((int) fd, (unsigned int) flags, | |
434 | offset, nbytes); | |
435 | } | |
436 | SYSCALL_ALIAS(sys_sync_file_range2, SyS_sync_file_range2); | |
437 | #endif | |
edd5cd4a | 438 | |
f79e2abb AM |
439 | /* |
440 | * `endbyte' is inclusive | |
441 | */ | |
5b04aa3a MF |
442 | int do_sync_mapping_range(struct address_space *mapping, loff_t offset, |
443 | loff_t endbyte, unsigned int flags) | |
f79e2abb AM |
444 | { |
445 | int ret; | |
f79e2abb | 446 | |
f79e2abb AM |
447 | if (!mapping) { |
448 | ret = -EINVAL; | |
449 | goto out; | |
450 | } | |
451 | ||
452 | ret = 0; | |
453 | if (flags & SYNC_FILE_RANGE_WAIT_BEFORE) { | |
454 | ret = wait_on_page_writeback_range(mapping, | |
455 | offset >> PAGE_CACHE_SHIFT, | |
456 | endbyte >> PAGE_CACHE_SHIFT); | |
457 | if (ret < 0) | |
458 | goto out; | |
459 | } | |
460 | ||
461 | if (flags & SYNC_FILE_RANGE_WRITE) { | |
462 | ret = __filemap_fdatawrite_range(mapping, offset, endbyte, | |
ee53a891 | 463 | WB_SYNC_ALL); |
f79e2abb AM |
464 | if (ret < 0) |
465 | goto out; | |
466 | } | |
467 | ||
468 | if (flags & SYNC_FILE_RANGE_WAIT_AFTER) { | |
469 | ret = wait_on_page_writeback_range(mapping, | |
470 | offset >> PAGE_CACHE_SHIFT, | |
471 | endbyte >> PAGE_CACHE_SHIFT); | |
472 | } | |
473 | out: | |
474 | return ret; | |
475 | } | |
5b04aa3a | 476 | EXPORT_SYMBOL_GPL(do_sync_mapping_range); |