1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * High-level sync()-related operations
4 */
5
6 #include <linux/blkdev.h>
7 #include <linux/kernel.h>
8 #include <linux/file.h>
9 #include <linux/fs.h>
10 #include <linux/slab.h>
11 #include <linux/export.h>
12 #include <linux/namei.h>
13 #include <linux/sched/xacct.h>
14 #include <linux/writeback.h>
15 #include <linux/syscalls.h>
16 #include <linux/linkage.h>
17 #include <linux/pagemap.h>
18 #include <linux/quotaops.h>
19 #include <linux/backing-dev.h>
20 #include <trace/hooks/fs.h>
21 #include "internal.h"
22
23 #define VALID_FLAGS (SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE| \
24 SYNC_FILE_RANGE_WAIT_AFTER)
25
26 /*
27 * Write out and wait upon all dirty data associated with this
28 * superblock. Filesystem data as well as the underlying block
29 * device. Takes the superblock lock.
30 */
sync_filesystem(struct super_block * sb)31 int sync_filesystem(struct super_block *sb)
32 {
33 int ret = 0;
34
35 /*
36 * We need to be protected against the filesystem going from
37 * r/o to r/w or vice versa.
38 */
39 WARN_ON(!rwsem_is_locked(&sb->s_umount));
40
41 /*
42 * No point in syncing out anything if the filesystem is read-only.
43 */
44 if (sb_rdonly(sb))
45 return 0;
46
47 /*
48 * Do the filesystem syncing work. For simple filesystems
49 * writeback_inodes_sb(sb) just dirties buffers with inodes so we have
50 * to submit I/O for these buffers via sync_blockdev(). This also
51 * speeds up the wait == 1 case since in that case write_inode()
52 * methods call sync_dirty_buffer() and thus effectively write one block
53 * at a time.
54 */
55 writeback_inodes_sb(sb, WB_REASON_SYNC);
56 if (sb->s_op->sync_fs) {
57 ret = sb->s_op->sync_fs(sb, 0);
58 if (ret)
59 return ret;
60 }
61 ret = sync_blockdev_nowait(sb->s_bdev);
62 if (ret)
63 return ret;
64
65 sync_inodes_sb(sb);
66 if (sb->s_op->sync_fs) {
67 ret = sb->s_op->sync_fs(sb, 1);
68 if (ret)
69 return ret;
70 }
71 return sync_blockdev(sb->s_bdev);
72 }
73 EXPORT_SYMBOL(sync_filesystem);
74
sync_inodes_one_sb(struct super_block * sb,void * arg)75 static void sync_inodes_one_sb(struct super_block *sb, void *arg)
76 {
77 if (!sb_rdonly(sb))
78 sync_inodes_sb(sb);
79 }
80
sync_fs_one_sb(struct super_block * sb,void * arg)81 static void sync_fs_one_sb(struct super_block *sb, void *arg)
82 {
83 if (!sb_rdonly(sb) && !(sb->s_iflags & SB_I_SKIP_SYNC) &&
84 sb->s_op->sync_fs)
85 sb->s_op->sync_fs(sb, *(int *)arg);
86 }
87
88 /*
89 * Sync everything. We start by waking flusher threads so that most of
90 * writeback runs on all devices in parallel. Then we sync all inodes reliably
91 * which effectively also waits for all flusher threads to finish doing
92 * writeback. At this point all data is on disk so metadata should be stable
93 * and we tell filesystems to sync their metadata via ->sync_fs() calls.
94 * Finally, we writeout all block devices because some filesystems (e.g. ext2)
95 * just write metadata (such as inodes or bitmaps) to block device page cache
96 * and do not sync it on their own in ->sync_fs().
97 */
ksys_sync(void)98 void ksys_sync(void)
99 {
100 int nowait = 0, wait = 1;
101
102 wakeup_flusher_threads(WB_REASON_SYNC);
103 iterate_supers(sync_inodes_one_sb, NULL);
104 iterate_supers(sync_fs_one_sb, &nowait);
105 iterate_supers(sync_fs_one_sb, &wait);
106 sync_bdevs(false);
107 sync_bdevs(true);
108 if (unlikely(laptop_mode))
109 laptop_sync_completion();
110 }
111
SYSCALL_DEFINE0(sync)112 SYSCALL_DEFINE0(sync)
113 {
114 ksys_sync();
115 return 0;
116 }
117
do_sync_work(struct work_struct * work)118 static void do_sync_work(struct work_struct *work)
119 {
120 int nowait = 0;
121
122 /*
123 * Sync twice to reduce the possibility we skipped some inodes / pages
124 * because they were temporarily locked
125 */
126 iterate_supers(sync_inodes_one_sb, &nowait);
127 iterate_supers(sync_fs_one_sb, &nowait);
128 sync_bdevs(false);
129 iterate_supers(sync_inodes_one_sb, &nowait);
130 iterate_supers(sync_fs_one_sb, &nowait);
131 sync_bdevs(false);
132 printk("Emergency Sync complete\n");
133 kfree(work);
134 }
135
emergency_sync(void)136 void emergency_sync(void)
137 {
138 struct work_struct *work;
139
140 work = kmalloc(sizeof(*work), GFP_ATOMIC);
141 if (work) {
142 INIT_WORK(work, do_sync_work);
143 schedule_work(work);
144 }
145 }
146
147 /*
148 * sync a single super
149 */
SYSCALL_DEFINE1(syncfs,int,fd)150 SYSCALL_DEFINE1(syncfs, int, fd)
151 {
152 struct fd f = fdget(fd);
153 struct super_block *sb;
154 int ret, ret2;
155
156 if (!fd_file(f))
157 return -EBADF;
158 sb = fd_file(f)->f_path.dentry->d_sb;
159
160 down_read(&sb->s_umount);
161 ret = sync_filesystem(sb);
162 up_read(&sb->s_umount);
163
164 ret2 = errseq_check_and_advance(&sb->s_wb_err, &fd_file(f)->f_sb_err);
165
166 fdput(f);
167 return ret ? ret : ret2;
168 }
169
170 /**
171 * vfs_fsync_range - helper to sync a range of data & metadata to disk
172 * @file: file to sync
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
176 *
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.
180 */
vfs_fsync_range(struct file * file,loff_t start,loff_t end,int datasync)181 int vfs_fsync_range(struct file *file, loff_t start, loff_t end, int datasync)
182 {
183 struct inode *inode = file->f_mapping->host;
184
185 if (!file->f_op->fsync)
186 return -EINVAL;
187 if (!datasync && (inode->i_state & I_DIRTY_TIME)) {
188 unsigned long cut_off = 0;
189 mark_inode_dirty_sync(inode);
190 trace_android_vh_vfs_fsync_range(inode, &cut_off);
191 if (cut_off)
192 return 0;
193 }
194 return file->f_op->fsync(file, start, end, datasync);
195 }
196 EXPORT_SYMBOL(vfs_fsync_range);
197
198 /**
199 * vfs_fsync - perform a fsync or fdatasync on a file
200 * @file: file to sync
201 * @datasync: only perform a fdatasync operation
202 *
203 * Write back data and metadata for @file to disk. If @datasync is
204 * set only metadata needed to access modified file data is written.
205 */
vfs_fsync(struct file * file,int datasync)206 int vfs_fsync(struct file *file, int datasync)
207 {
208 return vfs_fsync_range(file, 0, LLONG_MAX, datasync);
209 }
210 EXPORT_SYMBOL(vfs_fsync);
211
do_fsync(unsigned int fd,int datasync)212 static int do_fsync(unsigned int fd, int datasync)
213 {
214 struct fd f = fdget(fd);
215 int ret = -EBADF;
216
217 if (fd_file(f)) {
218 ret = vfs_fsync(fd_file(f), datasync);
219 fdput(f);
220 inc_syscfs(current);
221 }
222 return ret;
223 }
224
SYSCALL_DEFINE1(fsync,unsigned int,fd)225 SYSCALL_DEFINE1(fsync, unsigned int, fd)
226 {
227 return do_fsync(fd, 0);
228 }
229
SYSCALL_DEFINE1(fdatasync,unsigned int,fd)230 SYSCALL_DEFINE1(fdatasync, unsigned int, fd)
231 {
232 return do_fsync(fd, 1);
233 }
234
sync_file_range(struct file * file,loff_t offset,loff_t nbytes,unsigned int flags)235 int sync_file_range(struct file *file, loff_t offset, loff_t nbytes,
236 unsigned int flags)
237 {
238 int ret;
239 struct address_space *mapping;
240 loff_t endbyte; /* inclusive */
241 umode_t i_mode;
242
243 ret = -EINVAL;
244 if (flags & ~VALID_FLAGS)
245 goto out;
246
247 endbyte = offset + nbytes;
248
249 if ((s64)offset < 0)
250 goto out;
251 if ((s64)endbyte < 0)
252 goto out;
253 if (endbyte < offset)
254 goto out;
255
256 if (sizeof(pgoff_t) == 4) {
257 if (offset >= (0x100000000ULL << PAGE_SHIFT)) {
258 /*
259 * The range starts outside a 32 bit machine's
260 * pagecache addressing capabilities. Let it "succeed"
261 */
262 ret = 0;
263 goto out;
264 }
265 if (endbyte >= (0x100000000ULL << PAGE_SHIFT)) {
266 /*
267 * Out to EOF
268 */
269 nbytes = 0;
270 }
271 }
272
273 if (nbytes == 0)
274 endbyte = LLONG_MAX;
275 else
276 endbyte--; /* inclusive */
277
278 i_mode = file_inode(file)->i_mode;
279 ret = -ESPIPE;
280 if (!S_ISREG(i_mode) && !S_ISBLK(i_mode) && !S_ISDIR(i_mode) &&
281 !S_ISLNK(i_mode))
282 goto out;
283
284 mapping = file->f_mapping;
285 ret = 0;
286 if (flags & SYNC_FILE_RANGE_WAIT_BEFORE) {
287 ret = file_fdatawait_range(file, offset, endbyte);
288 if (ret < 0)
289 goto out;
290 }
291
292 if (flags & SYNC_FILE_RANGE_WRITE) {
293 int sync_mode = WB_SYNC_NONE;
294
295 if ((flags & SYNC_FILE_RANGE_WRITE_AND_WAIT) ==
296 SYNC_FILE_RANGE_WRITE_AND_WAIT)
297 sync_mode = WB_SYNC_ALL;
298
299 ret = __filemap_fdatawrite_range(mapping, offset, endbyte,
300 sync_mode);
301 if (ret < 0)
302 goto out;
303 }
304
305 if (flags & SYNC_FILE_RANGE_WAIT_AFTER)
306 ret = file_fdatawait_range(file, offset, endbyte);
307
308 out:
309 return ret;
310 }
311
312 /*
313 * ksys_sync_file_range() permits finely controlled syncing over a segment of
314 * a file in the range offset .. (offset+nbytes-1) inclusive. If nbytes is
315 * zero then ksys_sync_file_range() will operate from offset out to EOF.
316 *
317 * The flag bits are:
318 *
319 * SYNC_FILE_RANGE_WAIT_BEFORE: wait upon writeout of all pages in the range
320 * before performing the write.
321 *
322 * SYNC_FILE_RANGE_WRITE: initiate writeout of all those dirty pages in the
323 * range which are not presently under writeback. Note that this may block for
324 * significant periods due to exhaustion of disk request structures.
325 *
326 * SYNC_FILE_RANGE_WAIT_AFTER: wait upon writeout of all pages in the range
327 * after performing the write.
328 *
329 * Useful combinations of the flag bits are:
330 *
331 * SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE: ensures that all pages
332 * in the range which were dirty on entry to ksys_sync_file_range() are placed
333 * under writeout. This is a start-write-for-data-integrity operation.
334 *
335 * SYNC_FILE_RANGE_WRITE: start writeout of all dirty pages in the range which
336 * are not presently under writeout. This is an asynchronous flush-to-disk
337 * operation. Not suitable for data integrity operations.
338 *
339 * SYNC_FILE_RANGE_WAIT_BEFORE (or SYNC_FILE_RANGE_WAIT_AFTER): wait for
340 * completion of writeout of all pages in the range. This will be used after an
341 * earlier SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE operation to wait
342 * for that operation to complete and to return the result.
343 *
344 * SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE|SYNC_FILE_RANGE_WAIT_AFTER
345 * (a.k.a. SYNC_FILE_RANGE_WRITE_AND_WAIT):
346 * a traditional sync() operation. This is a write-for-data-integrity operation
347 * which will ensure that all pages in the range which were dirty on entry to
348 * ksys_sync_file_range() are written to disk. It should be noted that disk
349 * caches are not flushed by this call, so there are no guarantees here that the
350 * data will be available on disk after a crash.
351 *
352 *
353 * SYNC_FILE_RANGE_WAIT_BEFORE and SYNC_FILE_RANGE_WAIT_AFTER will detect any
354 * I/O errors or ENOSPC conditions and will return those to the caller, after
355 * clearing the EIO and ENOSPC flags in the address_space.
356 *
357 * It should be noted that none of these operations write out the file's
358 * metadata. So unless the application is strictly performing overwrites of
359 * already-instantiated disk blocks, there are no guarantees here that the data
360 * will be available after a crash.
361 */
ksys_sync_file_range(int fd,loff_t offset,loff_t nbytes,unsigned int flags)362 int ksys_sync_file_range(int fd, loff_t offset, loff_t nbytes,
363 unsigned int flags)
364 {
365 int ret;
366 struct fd f;
367
368 ret = -EBADF;
369 f = fdget(fd);
370 if (fd_file(f))
371 ret = sync_file_range(fd_file(f), offset, nbytes, flags);
372
373 fdput(f);
374 return ret;
375 }
376
SYSCALL_DEFINE4(sync_file_range,int,fd,loff_t,offset,loff_t,nbytes,unsigned int,flags)377 SYSCALL_DEFINE4(sync_file_range, int, fd, loff_t, offset, loff_t, nbytes,
378 unsigned int, flags)
379 {
380 return ksys_sync_file_range(fd, offset, nbytes, flags);
381 }
382
383 #if defined(CONFIG_COMPAT) && defined(__ARCH_WANT_COMPAT_SYNC_FILE_RANGE)
COMPAT_SYSCALL_DEFINE6(sync_file_range,int,fd,compat_arg_u64_dual (offset),compat_arg_u64_dual (nbytes),unsigned int,flags)384 COMPAT_SYSCALL_DEFINE6(sync_file_range, int, fd, compat_arg_u64_dual(offset),
385 compat_arg_u64_dual(nbytes), unsigned int, flags)
386 {
387 return ksys_sync_file_range(fd, compat_arg_u64_glue(offset),
388 compat_arg_u64_glue(nbytes), flags);
389 }
390 #endif
391
392 /* It would be nice if people remember that not all the world's an i386
393 when they introduce new system calls */
SYSCALL_DEFINE4(sync_file_range2,int,fd,unsigned int,flags,loff_t,offset,loff_t,nbytes)394 SYSCALL_DEFINE4(sync_file_range2, int, fd, unsigned int, flags,
395 loff_t, offset, loff_t, nbytes)
396 {
397 return ksys_sync_file_range(fd, offset, nbytes, flags);
398 }
399