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