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 JK |
24 | /* |
25 | * Write out and wait upon all dirty data associated with this | |
26 | * superblock. Filesystem data as well as the underlying block | |
27 | * device. Takes the superblock lock. | |
28 | */ | |
60b0680f | 29 | int sync_filesystem(struct super_block *sb) |
c15c54f5 JK |
30 | { |
31 | int ret; | |
32 | ||
5af7926f CH |
33 | /* |
34 | * We need to be protected against the filesystem going from | |
35 | * r/o to r/w or vice versa. | |
36 | */ | |
37 | WARN_ON(!rwsem_is_locked(&sb->s_umount)); | |
38 | ||
39 | /* | |
40 | * No point in syncing out anything if the filesystem is read-only. | |
41 | */ | |
bc98a42c | 42 | if (sb_rdonly(sb)) |
5af7926f CH |
43 | return 0; |
44 | ||
9a208ba5 CH |
45 | /* |
46 | * Do the filesystem syncing work. For simple filesystems | |
47 | * writeback_inodes_sb(sb) just dirties buffers with inodes so we have | |
48 | * to submit I/O for these buffers via __sync_blockdev(). This also | |
49 | * speeds up the wait == 1 case since in that case write_inode() | |
50 | * methods call sync_dirty_buffer() and thus effectively write one block | |
51 | * at a time. | |
52 | */ | |
53 | writeback_inodes_sb(sb, WB_REASON_SYNC); | |
54 | if (sb->s_op->sync_fs) | |
55 | sb->s_op->sync_fs(sb, 0); | |
56 | ret = __sync_blockdev(sb->s_bdev, 0); | |
c15c54f5 JK |
57 | if (ret < 0) |
58 | return ret; | |
9a208ba5 CH |
59 | |
60 | sync_inodes_sb(sb); | |
61 | if (sb->s_op->sync_fs) | |
62 | sb->s_op->sync_fs(sb, 1); | |
63 | return __sync_blockdev(sb->s_bdev, 1); | |
c15c54f5 | 64 | } |
10096fb1 | 65 | EXPORT_SYMBOL(sync_filesystem); |
c15c54f5 | 66 | |
b3de6531 | 67 | static void sync_inodes_one_sb(struct super_block *sb, void *arg) |
01a05b33 | 68 | { |
bc98a42c | 69 | if (!sb_rdonly(sb)) |
0dc83bd3 | 70 | sync_inodes_sb(sb); |
01a05b33 | 71 | } |
b3de6531 | 72 | |
b3de6531 JK |
73 | static void sync_fs_one_sb(struct super_block *sb, void *arg) |
74 | { | |
32b1924b KK |
75 | if (!sb_rdonly(sb) && !(sb->s_iflags & SB_I_SKIP_SYNC) && |
76 | sb->s_op->sync_fs) | |
b3de6531 JK |
77 | sb->s_op->sync_fs(sb, *(int *)arg); |
78 | } | |
79 | ||
d0e91b13 | 80 | static void fdatawrite_one_bdev(struct block_device *bdev, void *arg) |
b3de6531 | 81 | { |
d0e91b13 | 82 | filemap_fdatawrite(bdev->bd_inode->i_mapping); |
a8c7176b JK |
83 | } |
84 | ||
d0e91b13 | 85 | static void fdatawait_one_bdev(struct block_device *bdev, void *arg) |
a8c7176b | 86 | { |
aa750fd7 JN |
87 | /* |
88 | * We keep the error status of individual mapping so that | |
89 | * applications can catch the writeback error using fsync(2). | |
90 | * See filemap_fdatawait_keep_errors() for details. | |
91 | */ | |
92 | filemap_fdatawait_keep_errors(bdev->bd_inode->i_mapping); | |
c15c54f5 JK |
93 | } |
94 | ||
3beab0b4 | 95 | /* |
4ea425b6 JK |
96 | * Sync everything. We start by waking flusher threads so that most of |
97 | * writeback runs on all devices in parallel. Then we sync all inodes reliably | |
98 | * which effectively also waits for all flusher threads to finish doing | |
99 | * writeback. At this point all data is on disk so metadata should be stable | |
100 | * and we tell filesystems to sync their metadata via ->sync_fs() calls. | |
101 | * Finally, we writeout all block devices because some filesystems (e.g. ext2) | |
102 | * just write metadata (such as inodes or bitmaps) to block device page cache | |
103 | * and do not sync it on their own in ->sync_fs(). | |
3beab0b4 | 104 | */ |
70f68ee8 | 105 | void ksys_sync(void) |
cf9a2ae8 | 106 | { |
b3de6531 JK |
107 | int nowait = 0, wait = 1; |
108 | ||
9ba4b2df | 109 | wakeup_flusher_threads(WB_REASON_SYNC); |
0dc83bd3 | 110 | iterate_supers(sync_inodes_one_sb, NULL); |
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(); | |
70f68ee8 DB |
117 | } |
118 | ||
119 | SYSCALL_DEFINE0(sync) | |
120 | { | |
121 | ksys_sync(); | |
cf9a2ae8 DH |
122 | return 0; |
123 | } | |
124 | ||
a2a9537a JA |
125 | static void do_sync_work(struct work_struct *work) |
126 | { | |
b3de6531 JK |
127 | int nowait = 0; |
128 | ||
5cee5815 JK |
129 | /* |
130 | * Sync twice to reduce the possibility we skipped some inodes / pages | |
131 | * because they were temporarily locked | |
132 | */ | |
b3de6531 JK |
133 | iterate_supers(sync_inodes_one_sb, &nowait); |
134 | iterate_supers(sync_fs_one_sb, &nowait); | |
d0e91b13 | 135 | iterate_bdevs(fdatawrite_one_bdev, NULL); |
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); |
5cee5815 | 139 | printk("Emergency Sync complete\n"); |
a2a9537a JA |
140 | kfree(work); |
141 | } | |
142 | ||
cf9a2ae8 DH |
143 | void emergency_sync(void) |
144 | { | |
a2a9537a JA |
145 | struct work_struct *work; |
146 | ||
147 | work = kmalloc(sizeof(*work), GFP_ATOMIC); | |
148 | if (work) { | |
149 | INIT_WORK(work, do_sync_work); | |
150 | schedule_work(work); | |
151 | } | |
cf9a2ae8 DH |
152 | } |
153 | ||
b7ed78f5 SW |
154 | /* |
155 | * sync a single super | |
156 | */ | |
157 | SYSCALL_DEFINE1(syncfs, int, fd) | |
158 | { | |
2903ff01 | 159 | struct fd f = fdget(fd); |
b7ed78f5 | 160 | struct super_block *sb; |
735e4ae5 | 161 | int ret, ret2; |
b7ed78f5 | 162 | |
2903ff01 | 163 | if (!f.file) |
b7ed78f5 | 164 | return -EBADF; |
b583043e | 165 | sb = f.file->f_path.dentry->d_sb; |
b7ed78f5 SW |
166 | |
167 | down_read(&sb->s_umount); | |
168 | ret = sync_filesystem(sb); | |
169 | up_read(&sb->s_umount); | |
170 | ||
735e4ae5 JL |
171 | ret2 = errseq_check_and_advance(&sb->s_wb_err, &f.file->f_sb_err); |
172 | ||
2903ff01 | 173 | fdput(f); |
735e4ae5 | 174 | return ret ? ret : ret2; |
b7ed78f5 SW |
175 | } |
176 | ||
4c728ef5 | 177 | /** |
148f948b | 178 | * vfs_fsync_range - helper to sync a range of data & metadata to disk |
4c728ef5 | 179 | * @file: file to sync |
148f948b JK |
180 | * @start: offset in bytes of the beginning of data range to sync |
181 | * @end: offset in bytes of the end of data range (inclusive) | |
182 | * @datasync: perform only datasync | |
4c728ef5 | 183 | * |
148f948b JK |
184 | * Write back data in range @start..@end and metadata for @file to disk. If |
185 | * @datasync is set only metadata needed to access modified file data is | |
186 | * written. | |
4c728ef5 | 187 | */ |
8018ab05 | 188 | int vfs_fsync_range(struct file *file, loff_t start, loff_t end, int datasync) |
cf9a2ae8 | 189 | { |
0ae45f63 TT |
190 | struct inode *inode = file->f_mapping->host; |
191 | ||
72c2d531 | 192 | if (!file->f_op->fsync) |
02c24a82 | 193 | return -EINVAL; |
0d07e557 | 194 | if (!datasync && (inode->i_state & I_DIRTY_TIME)) |
0ae45f63 | 195 | mark_inode_dirty_sync(inode); |
0f41074a | 196 | return file->f_op->fsync(file, start, end, datasync); |
cf9a2ae8 | 197 | } |
148f948b JK |
198 | EXPORT_SYMBOL(vfs_fsync_range); |
199 | ||
200 | /** | |
201 | * vfs_fsync - perform a fsync or fdatasync on a file | |
202 | * @file: file to sync | |
148f948b JK |
203 | * @datasync: only perform a fdatasync operation |
204 | * | |
205 | * Write back data and metadata for @file to disk. If @datasync is | |
206 | * set only metadata needed to access modified file data is written. | |
148f948b | 207 | */ |
8018ab05 | 208 | int vfs_fsync(struct file *file, int datasync) |
148f948b | 209 | { |
8018ab05 | 210 | return vfs_fsync_range(file, 0, LLONG_MAX, datasync); |
148f948b | 211 | } |
4c728ef5 | 212 | EXPORT_SYMBOL(vfs_fsync); |
cf9a2ae8 | 213 | |
4c728ef5 | 214 | static int do_fsync(unsigned int fd, int datasync) |
cf9a2ae8 | 215 | { |
2903ff01 | 216 | struct fd f = fdget(fd); |
cf9a2ae8 DH |
217 | int ret = -EBADF; |
218 | ||
2903ff01 AV |
219 | if (f.file) { |
220 | ret = vfs_fsync(f.file, datasync); | |
221 | fdput(f); | |
cf9a2ae8 DH |
222 | } |
223 | return ret; | |
224 | } | |
225 | ||
a5f8fa9e | 226 | SYSCALL_DEFINE1(fsync, unsigned int, fd) |
cf9a2ae8 | 227 | { |
4c728ef5 | 228 | return do_fsync(fd, 0); |
cf9a2ae8 DH |
229 | } |
230 | ||
a5f8fa9e | 231 | SYSCALL_DEFINE1(fdatasync, unsigned int, fd) |
cf9a2ae8 | 232 | { |
4c728ef5 | 233 | return do_fsync(fd, 1); |
cf9a2ae8 DH |
234 | } |
235 | ||
22f96b38 JA |
236 | int sync_file_range(struct file *file, loff_t offset, loff_t nbytes, |
237 | unsigned int flags) | |
f79e2abb AM |
238 | { |
239 | int ret; | |
7a0ad10c | 240 | struct address_space *mapping; |
f79e2abb | 241 | loff_t endbyte; /* inclusive */ |
f79e2abb AM |
242 | umode_t i_mode; |
243 | ||
244 | ret = -EINVAL; | |
245 | if (flags & ~VALID_FLAGS) | |
246 | goto out; | |
247 | ||
248 | endbyte = offset + nbytes; | |
249 | ||
250 | if ((s64)offset < 0) | |
251 | goto out; | |
252 | if ((s64)endbyte < 0) | |
253 | goto out; | |
254 | if (endbyte < offset) | |
255 | goto out; | |
256 | ||
257 | if (sizeof(pgoff_t) == 4) { | |
09cbfeaf | 258 | if (offset >= (0x100000000ULL << PAGE_SHIFT)) { |
f79e2abb AM |
259 | /* |
260 | * The range starts outside a 32 bit machine's | |
261 | * pagecache addressing capabilities. Let it "succeed" | |
262 | */ | |
263 | ret = 0; | |
264 | goto out; | |
265 | } | |
09cbfeaf | 266 | if (endbyte >= (0x100000000ULL << PAGE_SHIFT)) { |
f79e2abb AM |
267 | /* |
268 | * Out to EOF | |
269 | */ | |
270 | nbytes = 0; | |
271 | } | |
272 | } | |
273 | ||
274 | if (nbytes == 0) | |
111ebb6e | 275 | endbyte = LLONG_MAX; |
f79e2abb AM |
276 | else |
277 | endbyte--; /* inclusive */ | |
278 | ||
22f96b38 | 279 | i_mode = file_inode(file)->i_mode; |
f79e2abb AM |
280 | ret = -ESPIPE; |
281 | if (!S_ISREG(i_mode) && !S_ISBLK(i_mode) && !S_ISDIR(i_mode) && | |
282 | !S_ISLNK(i_mode)) | |
22f96b38 | 283 | goto out; |
f79e2abb | 284 | |
22f96b38 | 285 | mapping = file->f_mapping; |
7a0ad10c CH |
286 | ret = 0; |
287 | if (flags & SYNC_FILE_RANGE_WAIT_BEFORE) { | |
22f96b38 | 288 | ret = file_fdatawait_range(file, offset, endbyte); |
7a0ad10c | 289 | if (ret < 0) |
22f96b38 | 290 | goto out; |
7a0ad10c CH |
291 | } |
292 | ||
293 | if (flags & SYNC_FILE_RANGE_WRITE) { | |
c553ea4f AG |
294 | int sync_mode = WB_SYNC_NONE; |
295 | ||
296 | if ((flags & SYNC_FILE_RANGE_WRITE_AND_WAIT) == | |
297 | SYNC_FILE_RANGE_WRITE_AND_WAIT) | |
298 | sync_mode = WB_SYNC_ALL; | |
299 | ||
23d01270 | 300 | ret = __filemap_fdatawrite_range(mapping, offset, endbyte, |
c553ea4f | 301 | sync_mode); |
7a0ad10c | 302 | if (ret < 0) |
22f96b38 | 303 | goto out; |
7a0ad10c CH |
304 | } |
305 | ||
306 | if (flags & SYNC_FILE_RANGE_WAIT_AFTER) | |
22f96b38 | 307 | ret = file_fdatawait_range(file, offset, endbyte); |
7a0ad10c | 308 | |
f79e2abb AM |
309 | out: |
310 | return ret; | |
311 | } | |
312 | ||
22f96b38 | 313 | /* |
c553ea4f | 314 | * ksys_sync_file_range() permits finely controlled syncing over a segment of |
22f96b38 | 315 | * a file in the range offset .. (offset+nbytes-1) inclusive. If nbytes is |
c553ea4f | 316 | * zero then ksys_sync_file_range() will operate from offset out to EOF. |
22f96b38 JA |
317 | * |
318 | * The flag bits are: | |
319 | * | |
320 | * SYNC_FILE_RANGE_WAIT_BEFORE: wait upon writeout of all pages in the range | |
321 | * before performing the write. | |
322 | * | |
323 | * SYNC_FILE_RANGE_WRITE: initiate writeout of all those dirty pages in the | |
324 | * range which are not presently under writeback. Note that this may block for | |
325 | * significant periods due to exhaustion of disk request structures. | |
326 | * | |
327 | * SYNC_FILE_RANGE_WAIT_AFTER: wait upon writeout of all pages in the range | |
328 | * after performing the write. | |
329 | * | |
330 | * Useful combinations of the flag bits are: | |
331 | * | |
332 | * SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE: ensures that all pages | |
c553ea4f | 333 | * in the range which were dirty on entry to ksys_sync_file_range() are placed |
22f96b38 JA |
334 | * under writeout. This is a start-write-for-data-integrity operation. |
335 | * | |
336 | * SYNC_FILE_RANGE_WRITE: start writeout of all dirty pages in the range which | |
337 | * are not presently under writeout. This is an asynchronous flush-to-disk | |
338 | * operation. Not suitable for data integrity operations. | |
339 | * | |
340 | * SYNC_FILE_RANGE_WAIT_BEFORE (or SYNC_FILE_RANGE_WAIT_AFTER): wait for | |
341 | * completion of writeout of all pages in the range. This will be used after an | |
342 | * earlier SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE operation to wait | |
343 | * for that operation to complete and to return the result. | |
344 | * | |
c553ea4f AG |
345 | * SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE|SYNC_FILE_RANGE_WAIT_AFTER |
346 | * (a.k.a. SYNC_FILE_RANGE_WRITE_AND_WAIT): | |
22f96b38 JA |
347 | * a traditional sync() operation. This is a write-for-data-integrity operation |
348 | * which will ensure that all pages in the range which were dirty on entry to | |
c553ea4f AG |
349 | * ksys_sync_file_range() are written to disk. It should be noted that disk |
350 | * caches are not flushed by this call, so there are no guarantees here that the | |
351 | * data will be available on disk after a crash. | |
22f96b38 JA |
352 | * |
353 | * | |
354 | * SYNC_FILE_RANGE_WAIT_BEFORE and SYNC_FILE_RANGE_WAIT_AFTER will detect any | |
355 | * I/O errors or ENOSPC conditions and will return those to the caller, after | |
356 | * clearing the EIO and ENOSPC flags in the address_space. | |
357 | * | |
358 | * It should be noted that none of these operations write out the file's | |
359 | * metadata. So unless the application is strictly performing overwrites of | |
360 | * already-instantiated disk blocks, there are no guarantees here that the data | |
361 | * will be available after a crash. | |
362 | */ | |
363 | int ksys_sync_file_range(int fd, loff_t offset, loff_t nbytes, | |
364 | unsigned int flags) | |
365 | { | |
366 | int ret; | |
367 | struct fd f; | |
368 | ||
369 | ret = -EBADF; | |
370 | f = fdget(fd); | |
371 | if (f.file) | |
372 | ret = sync_file_range(f.file, offset, nbytes, flags); | |
373 | ||
374 | fdput(f); | |
375 | return ret; | |
376 | } | |
377 | ||
806cbae1 DB |
378 | SYSCALL_DEFINE4(sync_file_range, int, fd, loff_t, offset, loff_t, nbytes, |
379 | unsigned int, flags) | |
380 | { | |
381 | return ksys_sync_file_range(fd, offset, nbytes, flags); | |
382 | } | |
383 | ||
edd5cd4a DW |
384 | /* It would be nice if people remember that not all the world's an i386 |
385 | when they introduce new system calls */ | |
4a0fd5bf AV |
386 | SYSCALL_DEFINE4(sync_file_range2, int, fd, unsigned int, flags, |
387 | loff_t, offset, loff_t, nbytes) | |
edd5cd4a | 388 | { |
806cbae1 | 389 | return ksys_sync_file_range(fd, offset, nbytes, flags); |
edd5cd4a | 390 | } |