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 "internal.h"
21
22 #define VALID_FLAGS (SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE| \
23 SYNC_FILE_RANGE_WAIT_AFTER)
24
25 /*
26 * Write out and wait upon all dirty data associated with this
27 * superblock. Filesystem data as well as the underlying block
28 * device. Takes the superblock lock.
29 */
sync_filesystem(struct super_block * sb)30 int sync_filesystem(struct super_block *sb)
31 {
32 int ret = 0;
33
34 /*
35 * We need to be protected against the filesystem going from
36 * r/o to r/w or vice versa.
37 */
38 WARN_ON(!rwsem_is_locked(&sb->s_umount));
39
40 /*
41 * No point in syncing out anything if the filesystem is read-only.
42 */
43 if (sb_rdonly(sb))
44 return 0;
45
46 /*
47 * Do the filesystem syncing work. For simple filesystems
48 * writeback_inodes_sb(sb) just dirties buffers with inodes so we have
49 * to submit I/O for these buffers via sync_blockdev(). This also
50 * speeds up the wait == 1 case since in that case write_inode()
51 * methods call sync_dirty_buffer() and thus effectively write one block
52 * at a time.
53 */
54 writeback_inodes_sb(sb, WB_REASON_SYNC);
55 if (sb->s_op->sync_fs) {
56 ret = sb->s_op->sync_fs(sb, 0);
57 if (ret)
58 return ret;
59 }
60 ret = sync_blockdev_nowait(sb->s_bdev);
61 if (ret)
62 return ret;
63
64 sync_inodes_sb(sb);
65 if (sb->s_op->sync_fs) {
66 ret = sb->s_op->sync_fs(sb, 1);
67 if (ret)
68 return ret;
69 }
70 return sync_blockdev(sb->s_bdev);
71 }
72 EXPORT_SYMBOL_NS(sync_filesystem, ANDROID_GKI_VFS_EXPORT_ONLY);
73
sync_inodes_one_sb(struct super_block * sb,void * arg)74 static void sync_inodes_one_sb(struct super_block *sb, void *arg)
75 {
76 if (!sb_rdonly(sb))
77 sync_inodes_sb(sb);
78 }
79
sync_fs_one_sb(struct super_block * sb,void * arg)80 static void sync_fs_one_sb(struct super_block *sb, void *arg)
81 {
82 if (!sb_rdonly(sb) && !(sb->s_iflags & SB_I_SKIP_SYNC) &&
83 sb->s_op->sync_fs)
84 sb->s_op->sync_fs(sb, *(int *)arg);
85 }
86
87 /*
88 * Sync everything. We start by waking flusher threads so that most of
89 * writeback runs on all devices in parallel. Then we sync all inodes reliably
90 * which effectively also waits for all flusher threads to finish doing
91 * writeback. At this point all data is on disk so metadata should be stable
92 * and we tell filesystems to sync their metadata via ->sync_fs() calls.
93 * Finally, we writeout all block devices because some filesystems (e.g. ext2)
94 * just write metadata (such as inodes or bitmaps) to block device page cache
95 * and do not sync it on their own in ->sync_fs().
96 */
ksys_sync(void)97 void ksys_sync(void)
98 {
99 int nowait = 0, wait = 1;
100
101 wakeup_flusher_threads(WB_REASON_SYNC);
102 iterate_supers(sync_inodes_one_sb, NULL);
103 iterate_supers(sync_fs_one_sb, &nowait);
104 iterate_supers(sync_fs_one_sb, &wait);
105 sync_bdevs(false);
106 sync_bdevs(true);
107 if (unlikely(laptop_mode))
108 laptop_sync_completion();
109 }
110
SYSCALL_DEFINE0(sync)111 SYSCALL_DEFINE0(sync)
112 {
113 ksys_sync();
114 return 0;
115 }
116
do_sync_work(struct work_struct * work)117 static void do_sync_work(struct work_struct *work)
118 {
119 int nowait = 0;
120
121 /*
122 * Sync twice to reduce the possibility we skipped some inodes / pages
123 * because they were temporarily locked
124 */
125 iterate_supers(sync_inodes_one_sb, &nowait);
126 iterate_supers(sync_fs_one_sb, &nowait);
127 sync_bdevs(false);
128 iterate_supers(sync_inodes_one_sb, &nowait);
129 iterate_supers(sync_fs_one_sb, &nowait);
130 sync_bdevs(false);
131 printk("Emergency Sync complete\n");
132 kfree(work);
133 }
134
emergency_sync(void)135 void emergency_sync(void)
136 {
137 struct work_struct *work;
138
139 work = kmalloc(sizeof(*work), GFP_ATOMIC);
140 if (work) {
141 INIT_WORK(work, do_sync_work);
142 schedule_work(work);
143 }
144 }
145
146 /*
147 * sync a single super
148 */
SYSCALL_DEFINE1(syncfs,int,fd)149 SYSCALL_DEFINE1(syncfs, int, fd)
150 {
151 struct fd f = fdget(fd);
152 struct super_block *sb;
153 int ret, ret2;
154
155 if (!f.file)
156 return -EBADF;
157 sb = f.file->f_path.dentry->d_sb;
158
159 down_read(&sb->s_umount);
160 ret = sync_filesystem(sb);
161 up_read(&sb->s_umount);
162
163 ret2 = errseq_check_and_advance(&sb->s_wb_err, &f.file->f_sb_err);
164
165 fdput(f);
166 return ret ? ret : ret2;
167 }
168
169 /**
170 * vfs_fsync_range - helper to sync a range of data & metadata to disk
171 * @file: file to sync
172 * @start: offset in bytes of the beginning of data range to sync
173 * @end: offset in bytes of the end of data range (inclusive)
174 * @datasync: perform only datasync
175 *
176 * Write back data in range @start..@end and metadata for @file to disk. If
177 * @datasync is set only metadata needed to access modified file data is
178 * written.
179 */
vfs_fsync_range(struct file * file,loff_t start,loff_t end,int datasync)180 int vfs_fsync_range(struct file *file, loff_t start, loff_t end, int datasync)
181 {
182 struct inode *inode = file->f_mapping->host;
183
184 if (!file->f_op->fsync)
185 return -EINVAL;
186 if (!datasync && (inode->i_state & I_DIRTY_TIME))
187 mark_inode_dirty_sync(inode);
188 return file->f_op->fsync(file, start, end, datasync);
189 }
190 EXPORT_SYMBOL(vfs_fsync_range);
191
192 /**
193 * vfs_fsync - perform a fsync or fdatasync on a file
194 * @file: file to sync
195 * @datasync: only perform a fdatasync operation
196 *
197 * Write back data and metadata for @file to disk. If @datasync is
198 * set only metadata needed to access modified file data is written.
199 */
vfs_fsync(struct file * file,int datasync)200 int vfs_fsync(struct file *file, int datasync)
201 {
202 return vfs_fsync_range(file, 0, LLONG_MAX, datasync);
203 }
204 EXPORT_SYMBOL(vfs_fsync);
205
do_fsync(unsigned int fd,int datasync)206 static int do_fsync(unsigned int fd, int datasync)
207 {
208 struct fd f = fdget(fd);
209 int ret = -EBADF;
210
211 if (f.file) {
212 ret = vfs_fsync(f.file, datasync);
213 fdput(f);
214 inc_syscfs(current);
215 }
216 return ret;
217 }
218
SYSCALL_DEFINE1(fsync,unsigned int,fd)219 SYSCALL_DEFINE1(fsync, unsigned int, fd)
220 {
221 return do_fsync(fd, 0);
222 }
223
SYSCALL_DEFINE1(fdatasync,unsigned int,fd)224 SYSCALL_DEFINE1(fdatasync, unsigned int, fd)
225 {
226 return do_fsync(fd, 1);
227 }
228
sync_file_range(struct file * file,loff_t offset,loff_t nbytes,unsigned int flags)229 int sync_file_range(struct file *file, loff_t offset, loff_t nbytes,
230 unsigned int flags)
231 {
232 int ret;
233 struct address_space *mapping;
234 loff_t endbyte; /* inclusive */
235 umode_t i_mode;
236
237 ret = -EINVAL;
238 if (flags & ~VALID_FLAGS)
239 goto out;
240
241 endbyte = offset + nbytes;
242
243 if ((s64)offset < 0)
244 goto out;
245 if ((s64)endbyte < 0)
246 goto out;
247 if (endbyte < offset)
248 goto out;
249
250 if (sizeof(pgoff_t) == 4) {
251 if (offset >= (0x100000000ULL << PAGE_SHIFT)) {
252 /*
253 * The range starts outside a 32 bit machine's
254 * pagecache addressing capabilities. Let it "succeed"
255 */
256 ret = 0;
257 goto out;
258 }
259 if (endbyte >= (0x100000000ULL << PAGE_SHIFT)) {
260 /*
261 * Out to EOF
262 */
263 nbytes = 0;
264 }
265 }
266
267 if (nbytes == 0)
268 endbyte = LLONG_MAX;
269 else
270 endbyte--; /* inclusive */
271
272 i_mode = file_inode(file)->i_mode;
273 ret = -ESPIPE;
274 if (!S_ISREG(i_mode) && !S_ISBLK(i_mode) && !S_ISDIR(i_mode) &&
275 !S_ISLNK(i_mode))
276 goto out;
277
278 mapping = file->f_mapping;
279 ret = 0;
280 if (flags & SYNC_FILE_RANGE_WAIT_BEFORE) {
281 ret = file_fdatawait_range(file, offset, endbyte);
282 if (ret < 0)
283 goto out;
284 }
285
286 if (flags & SYNC_FILE_RANGE_WRITE) {
287 int sync_mode = WB_SYNC_NONE;
288
289 if ((flags & SYNC_FILE_RANGE_WRITE_AND_WAIT) ==
290 SYNC_FILE_RANGE_WRITE_AND_WAIT)
291 sync_mode = WB_SYNC_ALL;
292
293 ret = __filemap_fdatawrite_range(mapping, offset, endbyte,
294 sync_mode);
295 if (ret < 0)
296 goto out;
297 }
298
299 if (flags & SYNC_FILE_RANGE_WAIT_AFTER)
300 ret = file_fdatawait_range(file, offset, endbyte);
301
302 out:
303 return ret;
304 }
305
306 /*
307 * ksys_sync_file_range() permits finely controlled syncing over a segment of
308 * a file in the range offset .. (offset+nbytes-1) inclusive. If nbytes is
309 * zero then ksys_sync_file_range() will operate from offset out to EOF.
310 *
311 * The flag bits are:
312 *
313 * SYNC_FILE_RANGE_WAIT_BEFORE: wait upon writeout of all pages in the range
314 * before performing the write.
315 *
316 * SYNC_FILE_RANGE_WRITE: initiate writeout of all those dirty pages in the
317 * range which are not presently under writeback. Note that this may block for
318 * significant periods due to exhaustion of disk request structures.
319 *
320 * SYNC_FILE_RANGE_WAIT_AFTER: wait upon writeout of all pages in the range
321 * after performing the write.
322 *
323 * Useful combinations of the flag bits are:
324 *
325 * SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE: ensures that all pages
326 * in the range which were dirty on entry to ksys_sync_file_range() are placed
327 * under writeout. This is a start-write-for-data-integrity operation.
328 *
329 * SYNC_FILE_RANGE_WRITE: start writeout of all dirty pages in the range which
330 * are not presently under writeout. This is an asynchronous flush-to-disk
331 * operation. Not suitable for data integrity operations.
332 *
333 * SYNC_FILE_RANGE_WAIT_BEFORE (or SYNC_FILE_RANGE_WAIT_AFTER): wait for
334 * completion of writeout of all pages in the range. This will be used after an
335 * earlier SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE operation to wait
336 * for that operation to complete and to return the result.
337 *
338 * SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE|SYNC_FILE_RANGE_WAIT_AFTER
339 * (a.k.a. SYNC_FILE_RANGE_WRITE_AND_WAIT):
340 * a traditional sync() operation. This is a write-for-data-integrity operation
341 * which will ensure that all pages in the range which were dirty on entry to
342 * ksys_sync_file_range() are written to disk. It should be noted that disk
343 * caches are not flushed by this call, so there are no guarantees here that the
344 * data will be available on disk after a crash.
345 *
346 *
347 * SYNC_FILE_RANGE_WAIT_BEFORE and SYNC_FILE_RANGE_WAIT_AFTER will detect any
348 * I/O errors or ENOSPC conditions and will return those to the caller, after
349 * clearing the EIO and ENOSPC flags in the address_space.
350 *
351 * It should be noted that none of these operations write out the file's
352 * metadata. So unless the application is strictly performing overwrites of
353 * already-instantiated disk blocks, there are no guarantees here that the data
354 * will be available after a crash.
355 */
ksys_sync_file_range(int fd,loff_t offset,loff_t nbytes,unsigned int flags)356 int ksys_sync_file_range(int fd, loff_t offset, loff_t nbytes,
357 unsigned int flags)
358 {
359 int ret;
360 struct fd f;
361
362 ret = -EBADF;
363 f = fdget(fd);
364 if (f.file)
365 ret = sync_file_range(f.file, offset, nbytes, flags);
366
367 fdput(f);
368 return ret;
369 }
370
SYSCALL_DEFINE4(sync_file_range,int,fd,loff_t,offset,loff_t,nbytes,unsigned int,flags)371 SYSCALL_DEFINE4(sync_file_range, int, fd, loff_t, offset, loff_t, nbytes,
372 unsigned int, flags)
373 {
374 return ksys_sync_file_range(fd, offset, nbytes, flags);
375 }
376
377 #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)378 COMPAT_SYSCALL_DEFINE6(sync_file_range, int, fd, compat_arg_u64_dual(offset),
379 compat_arg_u64_dual(nbytes), unsigned int, flags)
380 {
381 return ksys_sync_file_range(fd, compat_arg_u64_glue(offset),
382 compat_arg_u64_glue(nbytes), flags);
383 }
384 #endif
385
386 /* It would be nice if people remember that not all the world's an i386
387 when they introduce new system calls */
SYSCALL_DEFINE4(sync_file_range2,int,fd,unsigned int,flags,loff_t,offset,loff_t,nbytes)388 SYSCALL_DEFINE4(sync_file_range2, int, fd, unsigned int, flags,
389 loff_t, offset, loff_t, nbytes)
390 {
391 return ksys_sync_file_range(fd, offset, nbytes, flags);
392 }
393