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> | |
5a0e3ad6 | 8 | #include <linux/slab.h> |
630d9c47 | 9 | #include <linux/export.h> |
b7ed78f5 | 10 | #include <linux/namei.h> |
914e2637 | 11 | #include <linux/sched.h> |
f79e2abb AM |
12 | #include <linux/writeback.h> |
13 | #include <linux/syscalls.h> | |
14 | #include <linux/linkage.h> | |
15 | #include <linux/pagemap.h> | |
cf9a2ae8 | 16 | #include <linux/quotaops.h> |
5129a469 | 17 | #include <linux/backing-dev.h> |
5a3e5cb8 | 18 | #include "internal.h" |
f79e2abb AM |
19 | |
20 | #define VALID_FLAGS (SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE| \ | |
21 | SYNC_FILE_RANGE_WAIT_AFTER) | |
22 | ||
c15c54f5 | 23 | /* |
d8a8559c JA |
24 | * 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. | |
c15c54f5 | 29 | */ |
c4a391b5 JK |
30 | static int __sync_filesystem(struct super_block *sb, int wait, |
31 | unsigned long start) | |
c15c54f5 | 32 | { |
5fb324ad | 33 | if (wait) |
c4a391b5 | 34 | sync_inodes_sb(sb, start); |
5fb324ad | 35 | else |
0e175a18 | 36 | writeback_inodes_sb(sb, WB_REASON_SYNC); |
5fb324ad | 37 | |
c15c54f5 JK |
38 | if (sb->s_op->sync_fs) |
39 | sb->s_op->sync_fs(sb, wait); | |
40 | return __sync_blockdev(sb->s_bdev, wait); | |
41 | } | |
42 | ||
43 | /* | |
44 | * Write out and wait upon all dirty data associated with this | |
45 | * superblock. Filesystem data as well as the underlying block | |
46 | * device. Takes the superblock lock. | |
47 | */ | |
60b0680f | 48 | int sync_filesystem(struct super_block *sb) |
c15c54f5 JK |
49 | { |
50 | int ret; | |
c4a391b5 | 51 | unsigned long start = jiffies; |
c15c54f5 | 52 | |
5af7926f CH |
53 | /* |
54 | * We need to be protected against the filesystem going from | |
55 | * r/o to r/w or vice versa. | |
56 | */ | |
57 | WARN_ON(!rwsem_is_locked(&sb->s_umount)); | |
58 | ||
59 | /* | |
60 | * No point in syncing out anything if the filesystem is read-only. | |
61 | */ | |
62 | if (sb->s_flags & MS_RDONLY) | |
63 | return 0; | |
64 | ||
c4a391b5 | 65 | ret = __sync_filesystem(sb, 0, start); |
c15c54f5 JK |
66 | if (ret < 0) |
67 | return ret; | |
c4a391b5 | 68 | return __sync_filesystem(sb, 1, start); |
c15c54f5 | 69 | } |
60b0680f | 70 | EXPORT_SYMBOL_GPL(sync_filesystem); |
c15c54f5 | 71 | |
b3de6531 | 72 | static void sync_inodes_one_sb(struct super_block *sb, void *arg) |
01a05b33 | 73 | { |
95f28604 | 74 | if (!(sb->s_flags & MS_RDONLY)) |
c4a391b5 | 75 | sync_inodes_sb(sb, *((unsigned long *)arg)); |
01a05b33 | 76 | } |
b3de6531 | 77 | |
b3de6531 JK |
78 | static void sync_fs_one_sb(struct super_block *sb, void *arg) |
79 | { | |
80 | if (!(sb->s_flags & MS_RDONLY) && sb->s_op->sync_fs) | |
81 | sb->s_op->sync_fs(sb, *(int *)arg); | |
82 | } | |
83 | ||
d0e91b13 | 84 | static void fdatawrite_one_bdev(struct block_device *bdev, void *arg) |
b3de6531 | 85 | { |
d0e91b13 | 86 | filemap_fdatawrite(bdev->bd_inode->i_mapping); |
a8c7176b JK |
87 | } |
88 | ||
d0e91b13 | 89 | static void fdatawait_one_bdev(struct block_device *bdev, void *arg) |
a8c7176b | 90 | { |
d0e91b13 | 91 | filemap_fdatawait(bdev->bd_inode->i_mapping); |
c15c54f5 JK |
92 | } |
93 | ||
3beab0b4 | 94 | /* |
4ea425b6 JK |
95 | * Sync everything. We start by waking flusher threads so that most of |
96 | * writeback runs on all devices in parallel. Then we sync all inodes reliably | |
97 | * which effectively also waits for all flusher threads to finish doing | |
98 | * writeback. At this point all data is on disk so metadata should be stable | |
99 | * and we tell filesystems to sync their metadata via ->sync_fs() calls. | |
100 | * Finally, we writeout all block devices because some filesystems (e.g. ext2) | |
101 | * just write metadata (such as inodes or bitmaps) to block device page cache | |
102 | * and do not sync it on their own in ->sync_fs(). | |
3beab0b4 | 103 | */ |
5cee5815 | 104 | SYSCALL_DEFINE0(sync) |
cf9a2ae8 | 105 | { |
b3de6531 | 106 | int nowait = 0, wait = 1; |
c4a391b5 | 107 | unsigned long start = jiffies; |
b3de6531 | 108 | |
0e175a18 | 109 | wakeup_flusher_threads(0, WB_REASON_SYNC); |
c4a391b5 | 110 | iterate_supers(sync_inodes_one_sb, &start); |
4ea425b6 | 111 | iterate_supers(sync_fs_one_sb, &nowait); |
b3de6531 | 112 | iterate_supers(sync_fs_one_sb, &wait); |
d0e91b13 JK |
113 | iterate_bdevs(fdatawrite_one_bdev, NULL); |
114 | iterate_bdevs(fdatawait_one_bdev, NULL); | |
cf9a2ae8 DH |
115 | if (unlikely(laptop_mode)) |
116 | laptop_sync_completion(); | |
cf9a2ae8 DH |
117 | return 0; |
118 | } | |
119 | ||
a2a9537a JA |
120 | static void do_sync_work(struct work_struct *work) |
121 | { | |
b3de6531 JK |
122 | int nowait = 0; |
123 | ||
5cee5815 JK |
124 | /* |
125 | * Sync twice to reduce the possibility we skipped some inodes / pages | |
126 | * because they were temporarily locked | |
127 | */ | |
b3de6531 JK |
128 | iterate_supers(sync_inodes_one_sb, &nowait); |
129 | iterate_supers(sync_fs_one_sb, &nowait); | |
d0e91b13 | 130 | iterate_bdevs(fdatawrite_one_bdev, NULL); |
b3de6531 JK |
131 | iterate_supers(sync_inodes_one_sb, &nowait); |
132 | iterate_supers(sync_fs_one_sb, &nowait); | |
d0e91b13 | 133 | iterate_bdevs(fdatawrite_one_bdev, NULL); |
5cee5815 | 134 | printk("Emergency Sync complete\n"); |
a2a9537a JA |
135 | kfree(work); |
136 | } | |
137 | ||
cf9a2ae8 DH |
138 | void emergency_sync(void) |
139 | { | |
a2a9537a JA |
140 | struct work_struct *work; |
141 | ||
142 | work = kmalloc(sizeof(*work), GFP_ATOMIC); | |
143 | if (work) { | |
144 | INIT_WORK(work, do_sync_work); | |
145 | schedule_work(work); | |
146 | } | |
cf9a2ae8 DH |
147 | } |
148 | ||
b7ed78f5 SW |
149 | /* |
150 | * sync a single super | |
151 | */ | |
152 | SYSCALL_DEFINE1(syncfs, int, fd) | |
153 | { | |
2903ff01 | 154 | struct fd f = fdget(fd); |
b7ed78f5 SW |
155 | struct super_block *sb; |
156 | int ret; | |
b7ed78f5 | 157 | |
2903ff01 | 158 | if (!f.file) |
b7ed78f5 | 159 | return -EBADF; |
2903ff01 | 160 | sb = f.file->f_dentry->d_sb; |
b7ed78f5 SW |
161 | |
162 | down_read(&sb->s_umount); | |
163 | ret = sync_filesystem(sb); | |
164 | up_read(&sb->s_umount); | |
165 | ||
2903ff01 | 166 | fdput(f); |
b7ed78f5 SW |
167 | return ret; |
168 | } | |
169 | ||
4c728ef5 | 170 | /** |
148f948b | 171 | * vfs_fsync_range - helper to sync a range of data & metadata to disk |
4c728ef5 | 172 | * @file: file to sync |
148f948b JK |
173 | * @start: offset in bytes of the beginning of data range to sync |
174 | * @end: offset in bytes of the end of data range (inclusive) | |
175 | * @datasync: perform only datasync | |
4c728ef5 | 176 | * |
148f948b JK |
177 | * Write back data in range @start..@end and metadata for @file to disk. If |
178 | * @datasync is set only metadata needed to access modified file data is | |
179 | * written. | |
4c728ef5 | 180 | */ |
8018ab05 | 181 | int vfs_fsync_range(struct file *file, loff_t start, loff_t end, int datasync) |
cf9a2ae8 | 182 | { |
72c2d531 | 183 | if (!file->f_op->fsync) |
02c24a82 JB |
184 | return -EINVAL; |
185 | return file->f_op->fsync(file, start, end, datasync); | |
cf9a2ae8 | 186 | } |
148f948b JK |
187 | EXPORT_SYMBOL(vfs_fsync_range); |
188 | ||
189 | /** | |
190 | * vfs_fsync - perform a fsync or fdatasync on a file | |
191 | * @file: file to sync | |
148f948b JK |
192 | * @datasync: only perform a fdatasync operation |
193 | * | |
194 | * Write back data and metadata for @file to disk. If @datasync is | |
195 | * set only metadata needed to access modified file data is written. | |
148f948b | 196 | */ |
8018ab05 | 197 | int vfs_fsync(struct file *file, int datasync) |
148f948b | 198 | { |
8018ab05 | 199 | return vfs_fsync_range(file, 0, LLONG_MAX, datasync); |
148f948b | 200 | } |
4c728ef5 | 201 | EXPORT_SYMBOL(vfs_fsync); |
cf9a2ae8 | 202 | |
4c728ef5 | 203 | static int do_fsync(unsigned int fd, int datasync) |
cf9a2ae8 | 204 | { |
2903ff01 | 205 | struct fd f = fdget(fd); |
cf9a2ae8 DH |
206 | int ret = -EBADF; |
207 | ||
2903ff01 AV |
208 | if (f.file) { |
209 | ret = vfs_fsync(f.file, datasync); | |
210 | fdput(f); | |
cf9a2ae8 DH |
211 | } |
212 | return ret; | |
213 | } | |
214 | ||
a5f8fa9e | 215 | SYSCALL_DEFINE1(fsync, unsigned int, fd) |
cf9a2ae8 | 216 | { |
4c728ef5 | 217 | return do_fsync(fd, 0); |
cf9a2ae8 DH |
218 | } |
219 | ||
a5f8fa9e | 220 | SYSCALL_DEFINE1(fdatasync, unsigned int, fd) |
cf9a2ae8 | 221 | { |
4c728ef5 | 222 | return do_fsync(fd, 1); |
cf9a2ae8 DH |
223 | } |
224 | ||
f79e2abb AM |
225 | /* |
226 | * sys_sync_file_range() permits finely controlled syncing over a segment of | |
227 | * a file in the range offset .. (offset+nbytes-1) inclusive. If nbytes is | |
228 | * zero then sys_sync_file_range() will operate from offset out to EOF. | |
229 | * | |
230 | * The flag bits are: | |
231 | * | |
232 | * SYNC_FILE_RANGE_WAIT_BEFORE: wait upon writeout of all pages in the range | |
233 | * before performing the write. | |
234 | * | |
235 | * SYNC_FILE_RANGE_WRITE: initiate writeout of all those dirty pages in the | |
cce77081 PM |
236 | * range which are not presently under writeback. Note that this may block for |
237 | * significant periods due to exhaustion of disk request structures. | |
f79e2abb AM |
238 | * |
239 | * SYNC_FILE_RANGE_WAIT_AFTER: wait upon writeout of all pages in the range | |
240 | * after performing the write. | |
241 | * | |
242 | * Useful combinations of the flag bits are: | |
243 | * | |
244 | * SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE: ensures that all pages | |
245 | * in the range which were dirty on entry to sys_sync_file_range() are placed | |
246 | * under writeout. This is a start-write-for-data-integrity operation. | |
247 | * | |
248 | * SYNC_FILE_RANGE_WRITE: start writeout of all dirty pages in the range which | |
249 | * are not presently under writeout. This is an asynchronous flush-to-disk | |
250 | * operation. Not suitable for data integrity operations. | |
251 | * | |
252 | * SYNC_FILE_RANGE_WAIT_BEFORE (or SYNC_FILE_RANGE_WAIT_AFTER): wait for | |
253 | * completion of writeout of all pages in the range. This will be used after an | |
254 | * earlier SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE operation to wait | |
255 | * for that operation to complete and to return the result. | |
256 | * | |
257 | * SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE|SYNC_FILE_RANGE_WAIT_AFTER: | |
258 | * a traditional sync() operation. This is a write-for-data-integrity operation | |
259 | * which will ensure that all pages in the range which were dirty on entry to | |
260 | * sys_sync_file_range() are committed to disk. | |
261 | * | |
262 | * | |
263 | * SYNC_FILE_RANGE_WAIT_BEFORE and SYNC_FILE_RANGE_WAIT_AFTER will detect any | |
264 | * I/O errors or ENOSPC conditions and will return those to the caller, after | |
265 | * clearing the EIO and ENOSPC flags in the address_space. | |
266 | * | |
267 | * It should be noted that none of these operations write out the file's | |
268 | * metadata. So unless the application is strictly performing overwrites of | |
269 | * already-instantiated disk blocks, there are no guarantees here that the data | |
270 | * will be available after a crash. | |
271 | */ | |
4a0fd5bf AV |
272 | SYSCALL_DEFINE4(sync_file_range, int, fd, loff_t, offset, loff_t, nbytes, |
273 | unsigned int, flags) | |
f79e2abb AM |
274 | { |
275 | int ret; | |
2903ff01 | 276 | struct fd f; |
7a0ad10c | 277 | struct address_space *mapping; |
f79e2abb | 278 | loff_t endbyte; /* inclusive */ |
f79e2abb AM |
279 | umode_t i_mode; |
280 | ||
281 | ret = -EINVAL; | |
282 | if (flags & ~VALID_FLAGS) | |
283 | goto out; | |
284 | ||
285 | endbyte = offset + nbytes; | |
286 | ||
287 | if ((s64)offset < 0) | |
288 | goto out; | |
289 | if ((s64)endbyte < 0) | |
290 | goto out; | |
291 | if (endbyte < offset) | |
292 | goto out; | |
293 | ||
294 | if (sizeof(pgoff_t) == 4) { | |
295 | if (offset >= (0x100000000ULL << PAGE_CACHE_SHIFT)) { | |
296 | /* | |
297 | * The range starts outside a 32 bit machine's | |
298 | * pagecache addressing capabilities. Let it "succeed" | |
299 | */ | |
300 | ret = 0; | |
301 | goto out; | |
302 | } | |
303 | if (endbyte >= (0x100000000ULL << PAGE_CACHE_SHIFT)) { | |
304 | /* | |
305 | * Out to EOF | |
306 | */ | |
307 | nbytes = 0; | |
308 | } | |
309 | } | |
310 | ||
311 | if (nbytes == 0) | |
111ebb6e | 312 | endbyte = LLONG_MAX; |
f79e2abb AM |
313 | else |
314 | endbyte--; /* inclusive */ | |
315 | ||
316 | ret = -EBADF; | |
2903ff01 AV |
317 | f = fdget(fd); |
318 | if (!f.file) | |
f79e2abb AM |
319 | goto out; |
320 | ||
496ad9aa | 321 | i_mode = file_inode(f.file)->i_mode; |
f79e2abb AM |
322 | ret = -ESPIPE; |
323 | if (!S_ISREG(i_mode) && !S_ISBLK(i_mode) && !S_ISDIR(i_mode) && | |
324 | !S_ISLNK(i_mode)) | |
325 | goto out_put; | |
326 | ||
2903ff01 | 327 | mapping = f.file->f_mapping; |
7a0ad10c CH |
328 | if (!mapping) { |
329 | ret = -EINVAL; | |
330 | goto out_put; | |
331 | } | |
332 | ||
333 | ret = 0; | |
334 | if (flags & SYNC_FILE_RANGE_WAIT_BEFORE) { | |
335 | ret = filemap_fdatawait_range(mapping, offset, endbyte); | |
336 | if (ret < 0) | |
337 | goto out_put; | |
338 | } | |
339 | ||
340 | if (flags & SYNC_FILE_RANGE_WRITE) { | |
341 | ret = filemap_fdatawrite_range(mapping, offset, endbyte); | |
342 | if (ret < 0) | |
343 | goto out_put; | |
344 | } | |
345 | ||
346 | if (flags & SYNC_FILE_RANGE_WAIT_AFTER) | |
347 | ret = filemap_fdatawait_range(mapping, offset, endbyte); | |
348 | ||
f79e2abb | 349 | out_put: |
2903ff01 | 350 | fdput(f); |
f79e2abb AM |
351 | out: |
352 | return ret; | |
353 | } | |
354 | ||
edd5cd4a DW |
355 | /* It would be nice if people remember that not all the world's an i386 |
356 | when they introduce new system calls */ | |
4a0fd5bf AV |
357 | SYSCALL_DEFINE4(sync_file_range2, int, fd, unsigned int, flags, |
358 | loff_t, offset, loff_t, nbytes) | |
edd5cd4a DW |
359 | { |
360 | return sys_sync_file_range(fd, offset, nbytes, flags); | |
361 | } |