Commit | Line | Data |
---|---|---|
b2441318 | 1 | // SPDX-License-Identifier: GPL-2.0 |
f79e2abb AM |
2 | /* |
3 | * High-level sync()-related operations | |
4 | */ | |
5 | ||
6 | #include <linux/kernel.h> | |
7 | #include <linux/file.h> | |
8 | #include <linux/fs.h> | |
5a0e3ad6 | 9 | #include <linux/slab.h> |
630d9c47 | 10 | #include <linux/export.h> |
b7ed78f5 | 11 | #include <linux/namei.h> |
914e2637 | 12 | #include <linux/sched.h> |
f79e2abb AM |
13 | #include <linux/writeback.h> |
14 | #include <linux/syscalls.h> | |
15 | #include <linux/linkage.h> | |
16 | #include <linux/pagemap.h> | |
cf9a2ae8 | 17 | #include <linux/quotaops.h> |
5129a469 | 18 | #include <linux/backing-dev.h> |
5a3e5cb8 | 19 | #include "internal.h" |
f79e2abb AM |
20 | |
21 | #define VALID_FLAGS (SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE| \ | |
22 | SYNC_FILE_RANGE_WAIT_AFTER) | |
23 | ||
c15c54f5 | 24 | /* |
d8a8559c JA |
25 | * Do the filesystem syncing work. For simple filesystems |
26 | * writeback_inodes_sb(sb) just dirties buffers with inodes so we have to | |
27 | * submit IO for these buffers via __sync_blockdev(). This also speeds up the | |
28 | * wait == 1 case since in that case write_inode() functions do | |
29 | * sync_dirty_buffer() and thus effectively write one block at a time. | |
c15c54f5 | 30 | */ |
0dc83bd3 | 31 | static int __sync_filesystem(struct super_block *sb, int wait) |
c15c54f5 | 32 | { |
5fb324ad | 33 | if (wait) |
0dc83bd3 | 34 | sync_inodes_sb(sb); |
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; | |
51 | ||
5af7926f CH |
52 | /* |
53 | * We need to be protected against the filesystem going from | |
54 | * r/o to r/w or vice versa. | |
55 | */ | |
56 | WARN_ON(!rwsem_is_locked(&sb->s_umount)); | |
57 | ||
58 | /* | |
59 | * No point in syncing out anything if the filesystem is read-only. | |
60 | */ | |
bc98a42c | 61 | if (sb_rdonly(sb)) |
5af7926f CH |
62 | return 0; |
63 | ||
0dc83bd3 | 64 | ret = __sync_filesystem(sb, 0); |
c15c54f5 JK |
65 | if (ret < 0) |
66 | return ret; | |
0dc83bd3 | 67 | return __sync_filesystem(sb, 1); |
c15c54f5 | 68 | } |
10096fb1 | 69 | EXPORT_SYMBOL(sync_filesystem); |
c15c54f5 | 70 | |
b3de6531 | 71 | static void sync_inodes_one_sb(struct super_block *sb, void *arg) |
01a05b33 | 72 | { |
bc98a42c | 73 | if (!sb_rdonly(sb)) |
0dc83bd3 | 74 | sync_inodes_sb(sb); |
01a05b33 | 75 | } |
b3de6531 | 76 | |
b3de6531 JK |
77 | static void sync_fs_one_sb(struct super_block *sb, void *arg) |
78 | { | |
bc98a42c | 79 | if (!sb_rdonly(sb) && sb->s_op->sync_fs) |
b3de6531 JK |
80 | sb->s_op->sync_fs(sb, *(int *)arg); |
81 | } | |
82 | ||
d0e91b13 | 83 | static void fdatawrite_one_bdev(struct block_device *bdev, void *arg) |
b3de6531 | 84 | { |
d0e91b13 | 85 | filemap_fdatawrite(bdev->bd_inode->i_mapping); |
a8c7176b JK |
86 | } |
87 | ||
d0e91b13 | 88 | static void fdatawait_one_bdev(struct block_device *bdev, void *arg) |
a8c7176b | 89 | { |
aa750fd7 JN |
90 | /* |
91 | * We keep the error status of individual mapping so that | |
92 | * applications can catch the writeback error using fsync(2). | |
93 | * See filemap_fdatawait_keep_errors() for details. | |
94 | */ | |
95 | filemap_fdatawait_keep_errors(bdev->bd_inode->i_mapping); | |
c15c54f5 JK |
96 | } |
97 | ||
3beab0b4 | 98 | /* |
4ea425b6 JK |
99 | * Sync everything. We start by waking flusher threads so that most of |
100 | * writeback runs on all devices in parallel. Then we sync all inodes reliably | |
101 | * which effectively also waits for all flusher threads to finish doing | |
102 | * writeback. At this point all data is on disk so metadata should be stable | |
103 | * and we tell filesystems to sync their metadata via ->sync_fs() calls. | |
104 | * Finally, we writeout all block devices because some filesystems (e.g. ext2) | |
105 | * just write metadata (such as inodes or bitmaps) to block device page cache | |
106 | * and do not sync it on their own in ->sync_fs(). | |
3beab0b4 | 107 | */ |
70f68ee8 | 108 | void ksys_sync(void) |
cf9a2ae8 | 109 | { |
b3de6531 JK |
110 | int nowait = 0, wait = 1; |
111 | ||
9ba4b2df | 112 | wakeup_flusher_threads(WB_REASON_SYNC); |
0dc83bd3 | 113 | iterate_supers(sync_inodes_one_sb, NULL); |
4ea425b6 | 114 | iterate_supers(sync_fs_one_sb, &nowait); |
b3de6531 | 115 | iterate_supers(sync_fs_one_sb, &wait); |
d0e91b13 JK |
116 | iterate_bdevs(fdatawrite_one_bdev, NULL); |
117 | iterate_bdevs(fdatawait_one_bdev, NULL); | |
cf9a2ae8 DH |
118 | if (unlikely(laptop_mode)) |
119 | laptop_sync_completion(); | |
70f68ee8 DB |
120 | } |
121 | ||
122 | SYSCALL_DEFINE0(sync) | |
123 | { | |
124 | ksys_sync(); | |
cf9a2ae8 DH |
125 | return 0; |
126 | } | |
127 | ||
a2a9537a JA |
128 | static void do_sync_work(struct work_struct *work) |
129 | { | |
b3de6531 JK |
130 | int nowait = 0; |
131 | ||
5cee5815 JK |
132 | /* |
133 | * Sync twice to reduce the possibility we skipped some inodes / pages | |
134 | * because they were temporarily locked | |
135 | */ | |
b3de6531 JK |
136 | iterate_supers(sync_inodes_one_sb, &nowait); |
137 | iterate_supers(sync_fs_one_sb, &nowait); | |
d0e91b13 | 138 | iterate_bdevs(fdatawrite_one_bdev, NULL); |
b3de6531 JK |
139 | iterate_supers(sync_inodes_one_sb, &nowait); |
140 | iterate_supers(sync_fs_one_sb, &nowait); | |
d0e91b13 | 141 | iterate_bdevs(fdatawrite_one_bdev, NULL); |
5cee5815 | 142 | printk("Emergency Sync complete\n"); |
a2a9537a JA |
143 | kfree(work); |
144 | } | |
145 | ||
cf9a2ae8 DH |
146 | void emergency_sync(void) |
147 | { | |
a2a9537a JA |
148 | struct work_struct *work; |
149 | ||
150 | work = kmalloc(sizeof(*work), GFP_ATOMIC); | |
151 | if (work) { | |
152 | INIT_WORK(work, do_sync_work); | |
153 | schedule_work(work); | |
154 | } | |
cf9a2ae8 DH |
155 | } |
156 | ||
b7ed78f5 SW |
157 | /* |
158 | * sync a single super | |
159 | */ | |
160 | SYSCALL_DEFINE1(syncfs, int, fd) | |
161 | { | |
2903ff01 | 162 | struct fd f = fdget(fd); |
b7ed78f5 | 163 | struct super_block *sb; |
735e4ae5 | 164 | int ret, ret2; |
b7ed78f5 | 165 | |
2903ff01 | 166 | if (!f.file) |
b7ed78f5 | 167 | return -EBADF; |
b583043e | 168 | sb = f.file->f_path.dentry->d_sb; |
b7ed78f5 SW |
169 | |
170 | down_read(&sb->s_umount); | |
171 | ret = sync_filesystem(sb); | |
172 | up_read(&sb->s_umount); | |
173 | ||
735e4ae5 JL |
174 | ret2 = errseq_check_and_advance(&sb->s_wb_err, &f.file->f_sb_err); |
175 | ||
2903ff01 | 176 | fdput(f); |
735e4ae5 | 177 | return ret ? ret : ret2; |
b7ed78f5 SW |
178 | } |
179 | ||
4c728ef5 | 180 | /** |
148f948b | 181 | * vfs_fsync_range - helper to sync a range of data & metadata to disk |
4c728ef5 | 182 | * @file: file to sync |
148f948b JK |
183 | * @start: offset in bytes of the beginning of data range to sync |
184 | * @end: offset in bytes of the end of data range (inclusive) | |
185 | * @datasync: perform only datasync | |
4c728ef5 | 186 | * |
148f948b JK |
187 | * Write back data in range @start..@end and metadata for @file to disk. If |
188 | * @datasync is set only metadata needed to access modified file data is | |
189 | * written. | |
4c728ef5 | 190 | */ |
8018ab05 | 191 | int vfs_fsync_range(struct file *file, loff_t start, loff_t end, int datasync) |
cf9a2ae8 | 192 | { |
0ae45f63 TT |
193 | struct inode *inode = file->f_mapping->host; |
194 | ||
72c2d531 | 195 | if (!file->f_op->fsync) |
02c24a82 | 196 | return -EINVAL; |
0d07e557 | 197 | if (!datasync && (inode->i_state & I_DIRTY_TIME)) |
0ae45f63 | 198 | mark_inode_dirty_sync(inode); |
0f41074a | 199 | return file->f_op->fsync(file, start, end, datasync); |
cf9a2ae8 | 200 | } |
148f948b JK |
201 | EXPORT_SYMBOL(vfs_fsync_range); |
202 | ||
203 | /** | |
204 | * vfs_fsync - perform a fsync or fdatasync on a file | |
205 | * @file: file to sync | |
148f948b JK |
206 | * @datasync: only perform a fdatasync operation |
207 | * | |
208 | * Write back data and metadata for @file to disk. If @datasync is | |
209 | * set only metadata needed to access modified file data is written. | |
148f948b | 210 | */ |
8018ab05 | 211 | int vfs_fsync(struct file *file, int datasync) |
148f948b | 212 | { |
8018ab05 | 213 | return vfs_fsync_range(file, 0, LLONG_MAX, datasync); |
148f948b | 214 | } |
4c728ef5 | 215 | EXPORT_SYMBOL(vfs_fsync); |
cf9a2ae8 | 216 | |
4c728ef5 | 217 | static int do_fsync(unsigned int fd, int datasync) |
cf9a2ae8 | 218 | { |
2903ff01 | 219 | struct fd f = fdget(fd); |
cf9a2ae8 DH |
220 | int ret = -EBADF; |
221 | ||
2903ff01 AV |
222 | if (f.file) { |
223 | ret = vfs_fsync(f.file, datasync); | |
224 | fdput(f); | |
cf9a2ae8 DH |
225 | } |
226 | return ret; | |
227 | } | |
228 | ||
a5f8fa9e | 229 | SYSCALL_DEFINE1(fsync, unsigned int, fd) |
cf9a2ae8 | 230 | { |
4c728ef5 | 231 | return do_fsync(fd, 0); |
cf9a2ae8 DH |
232 | } |
233 | ||
a5f8fa9e | 234 | SYSCALL_DEFINE1(fdatasync, unsigned int, fd) |
cf9a2ae8 | 235 | { |
4c728ef5 | 236 | return do_fsync(fd, 1); |
cf9a2ae8 DH |
237 | } |
238 | ||
22f96b38 JA |
239 | int sync_file_range(struct file *file, loff_t offset, loff_t nbytes, |
240 | unsigned int flags) | |
f79e2abb AM |
241 | { |
242 | int ret; | |
7a0ad10c | 243 | struct address_space *mapping; |
f79e2abb | 244 | loff_t endbyte; /* inclusive */ |
f79e2abb AM |
245 | umode_t i_mode; |
246 | ||
247 | ret = -EINVAL; | |
248 | if (flags & ~VALID_FLAGS) | |
249 | goto out; | |
250 | ||
251 | endbyte = offset + nbytes; | |
252 | ||
253 | if ((s64)offset < 0) | |
254 | goto out; | |
255 | if ((s64)endbyte < 0) | |
256 | goto out; | |
257 | if (endbyte < offset) | |
258 | goto out; | |
259 | ||
260 | if (sizeof(pgoff_t) == 4) { | |
09cbfeaf | 261 | if (offset >= (0x100000000ULL << PAGE_SHIFT)) { |
f79e2abb AM |
262 | /* |
263 | * The range starts outside a 32 bit machine's | |
264 | * pagecache addressing capabilities. Let it "succeed" | |
265 | */ | |
266 | ret = 0; | |
267 | goto out; | |
268 | } | |
09cbfeaf | 269 | if (endbyte >= (0x100000000ULL << PAGE_SHIFT)) { |
f79e2abb AM |
270 | /* |
271 | * Out to EOF | |
272 | */ | |
273 | nbytes = 0; | |
274 | } | |
275 | } | |
276 | ||
277 | if (nbytes == 0) | |
111ebb6e | 278 | endbyte = LLONG_MAX; |
f79e2abb AM |
279 | else |
280 | endbyte--; /* inclusive */ | |
281 | ||
22f96b38 | 282 | i_mode = file_inode(file)->i_mode; |
f79e2abb AM |
283 | ret = -ESPIPE; |
284 | if (!S_ISREG(i_mode) && !S_ISBLK(i_mode) && !S_ISDIR(i_mode) && | |
285 | !S_ISLNK(i_mode)) | |
22f96b38 | 286 | goto out; |
f79e2abb | 287 | |
22f96b38 | 288 | mapping = file->f_mapping; |
7a0ad10c CH |
289 | ret = 0; |
290 | if (flags & SYNC_FILE_RANGE_WAIT_BEFORE) { | |
22f96b38 | 291 | ret = file_fdatawait_range(file, offset, endbyte); |
7a0ad10c | 292 | if (ret < 0) |
22f96b38 | 293 | goto out; |
7a0ad10c CH |
294 | } |
295 | ||
296 | if (flags & SYNC_FILE_RANGE_WRITE) { | |
c553ea4f AG |
297 | int sync_mode = WB_SYNC_NONE; |
298 | ||
299 | if ((flags & SYNC_FILE_RANGE_WRITE_AND_WAIT) == | |
300 | SYNC_FILE_RANGE_WRITE_AND_WAIT) | |
301 | sync_mode = WB_SYNC_ALL; | |
302 | ||
23d01270 | 303 | ret = __filemap_fdatawrite_range(mapping, offset, endbyte, |
c553ea4f | 304 | sync_mode); |
7a0ad10c | 305 | if (ret < 0) |
22f96b38 | 306 | goto out; |
7a0ad10c CH |
307 | } |
308 | ||
309 | if (flags & SYNC_FILE_RANGE_WAIT_AFTER) | |
22f96b38 | 310 | ret = file_fdatawait_range(file, offset, endbyte); |
7a0ad10c | 311 | |
f79e2abb AM |
312 | out: |
313 | return ret; | |
314 | } | |
315 | ||
22f96b38 | 316 | /* |
c553ea4f | 317 | * ksys_sync_file_range() permits finely controlled syncing over a segment of |
22f96b38 | 318 | * a file in the range offset .. (offset+nbytes-1) inclusive. If nbytes is |
c553ea4f | 319 | * zero then ksys_sync_file_range() will operate from offset out to EOF. |
22f96b38 JA |
320 | * |
321 | * The flag bits are: | |
322 | * | |
323 | * SYNC_FILE_RANGE_WAIT_BEFORE: wait upon writeout of all pages in the range | |
324 | * before performing the write. | |
325 | * | |
326 | * SYNC_FILE_RANGE_WRITE: initiate writeout of all those dirty pages in the | |
327 | * range which are not presently under writeback. Note that this may block for | |
328 | * significant periods due to exhaustion of disk request structures. | |
329 | * | |
330 | * SYNC_FILE_RANGE_WAIT_AFTER: wait upon writeout of all pages in the range | |
331 | * after performing the write. | |
332 | * | |
333 | * Useful combinations of the flag bits are: | |
334 | * | |
335 | * SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE: ensures that all pages | |
c553ea4f | 336 | * in the range which were dirty on entry to ksys_sync_file_range() are placed |
22f96b38 JA |
337 | * under writeout. This is a start-write-for-data-integrity operation. |
338 | * | |
339 | * SYNC_FILE_RANGE_WRITE: start writeout of all dirty pages in the range which | |
340 | * are not presently under writeout. This is an asynchronous flush-to-disk | |
341 | * operation. Not suitable for data integrity operations. | |
342 | * | |
343 | * SYNC_FILE_RANGE_WAIT_BEFORE (or SYNC_FILE_RANGE_WAIT_AFTER): wait for | |
344 | * completion of writeout of all pages in the range. This will be used after an | |
345 | * earlier SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE operation to wait | |
346 | * for that operation to complete and to return the result. | |
347 | * | |
c553ea4f AG |
348 | * SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE|SYNC_FILE_RANGE_WAIT_AFTER |
349 | * (a.k.a. SYNC_FILE_RANGE_WRITE_AND_WAIT): | |
22f96b38 JA |
350 | * a traditional sync() operation. This is a write-for-data-integrity operation |
351 | * which will ensure that all pages in the range which were dirty on entry to | |
c553ea4f AG |
352 | * ksys_sync_file_range() are written to disk. It should be noted that disk |
353 | * caches are not flushed by this call, so there are no guarantees here that the | |
354 | * data will be available on disk after a crash. | |
22f96b38 JA |
355 | * |
356 | * | |
357 | * SYNC_FILE_RANGE_WAIT_BEFORE and SYNC_FILE_RANGE_WAIT_AFTER will detect any | |
358 | * I/O errors or ENOSPC conditions and will return those to the caller, after | |
359 | * clearing the EIO and ENOSPC flags in the address_space. | |
360 | * | |
361 | * It should be noted that none of these operations write out the file's | |
362 | * metadata. So unless the application is strictly performing overwrites of | |
363 | * already-instantiated disk blocks, there are no guarantees here that the data | |
364 | * will be available after a crash. | |
365 | */ | |
366 | int ksys_sync_file_range(int fd, loff_t offset, loff_t nbytes, | |
367 | unsigned int flags) | |
368 | { | |
369 | int ret; | |
370 | struct fd f; | |
371 | ||
372 | ret = -EBADF; | |
373 | f = fdget(fd); | |
374 | if (f.file) | |
375 | ret = sync_file_range(f.file, offset, nbytes, flags); | |
376 | ||
377 | fdput(f); | |
378 | return ret; | |
379 | } | |
380 | ||
806cbae1 DB |
381 | SYSCALL_DEFINE4(sync_file_range, int, fd, loff_t, offset, loff_t, nbytes, |
382 | unsigned int, flags) | |
383 | { | |
384 | return ksys_sync_file_range(fd, offset, nbytes, flags); | |
385 | } | |
386 | ||
edd5cd4a DW |
387 | /* It would be nice if people remember that not all the world's an i386 |
388 | when they introduce new system calls */ | |
4a0fd5bf AV |
389 | SYSCALL_DEFINE4(sync_file_range2, int, fd, unsigned int, flags, |
390 | loff_t, offset, loff_t, nbytes) | |
edd5cd4a | 391 | { |
806cbae1 | 392 | return ksys_sync_file_range(fd, offset, nbytes, flags); |
edd5cd4a | 393 | } |