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