sync.c

  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 */
 30int 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}
 72EXPORT_SYMBOL(sync_filesystem);
 73
 74static void sync_inodes_one_sb(struct super_block *sb, void *arg)
 75{
 76	if (!sb_rdonly(sb))
 77		sync_inodes_sb(sb);
 78}
 79
 80static 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 */
 97void 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
111SYSCALL_DEFINE0(sync)
112{
113	ksys_sync();
114	return 0;
115}
116
117static void do_sync_work(struct work_struct *work)
118{
119	int nowait = 0;
120	int wait = 1;
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, NULL);
127	iterate_supers(sync_fs_one_sb, &nowait);
128	sync_bdevs(false);
129	iterate_supers(sync_inodes_one_sb, NULL);
130	iterate_supers(sync_fs_one_sb, &wait);
131	sync_bdevs(false);
132	printk("Emergency Sync complete\n");
133	kfree(work);
134}
135
136void 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 */
150SYSCALL_DEFINE1(syncfs, int, fd)
151{
152	CLASS(fd, f)(fd);
153	struct super_block *sb;
154	int ret, ret2;
155
156	if (fd_empty(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	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 */
180int 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_state_read_once(inode) & I_DIRTY_TIME))
187		mark_inode_dirty_sync(inode);
188	return file->f_op->fsync(file, start, end, datasync);
189}
190EXPORT_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 */
200int vfs_fsync(struct file *file, int datasync)
201{
202	return vfs_fsync_range(file, 0, LLONG_MAX, datasync);
203}
204EXPORT_SYMBOL(vfs_fsync);
205
206static int do_fsync(unsigned int fd, int datasync)
207{
208	CLASS(fd, f)(fd);
209
210	if (fd_empty(f))
211		return -EBADF;
212
213	return vfs_fsync(fd_file(f), datasync);
214}
215
216SYSCALL_DEFINE1(fsync, unsigned int, fd)
217{
218	return do_fsync(fd, 0);
219}
220
221SYSCALL_DEFINE1(fdatasync, unsigned int, fd)
222{
223	return do_fsync(fd, 1);
224}
225
226int sync_file_range(struct file *file, loff_t offset, loff_t nbytes,
227		    unsigned int flags)
228{
229	int ret;
230	struct address_space *mapping;
231	loff_t endbyte;			/* inclusive */
232	umode_t i_mode;
233
234	ret = -EINVAL;
235	if (flags & ~VALID_FLAGS)
236		goto out;
237
238	endbyte = offset + nbytes;
239
240	if ((s64)offset < 0)
241		goto out;
242	if ((s64)endbyte < 0)
243		goto out;
244	if (endbyte < offset)
245		goto out;
246
247	if (sizeof(pgoff_t) == 4) {
248		if (offset >= (0x100000000ULL << PAGE_SHIFT)) {
249			/*
250			 * The range starts outside a 32 bit machine's
251			 * pagecache addressing capabilities.  Let it "succeed"
252			 */
253			ret = 0;
254			goto out;
255		}
256		if (endbyte >= (0x100000000ULL << PAGE_SHIFT)) {
257			/*
258			 * Out to EOF
259			 */
260			nbytes = 0;
261		}
262	}
263
264	if (nbytes == 0)
265		endbyte = LLONG_MAX;
266	else
267		endbyte--;		/* inclusive */
268
269	i_mode = file_inode(file)->i_mode;
270	ret = -ESPIPE;
271	if (!S_ISREG(i_mode) && !S_ISBLK(i_mode) && !S_ISDIR(i_mode) &&
272			!S_ISLNK(i_mode))
273		goto out;
274
275	mapping = file->f_mapping;
276	ret = 0;
277	if (flags & SYNC_FILE_RANGE_WAIT_BEFORE) {
278		ret = file_fdatawait_range(file, offset, endbyte);
279		if (ret < 0)
280			goto out;
281	}
282
283	if (flags & SYNC_FILE_RANGE_WRITE) {
284		if ((flags & SYNC_FILE_RANGE_WRITE_AND_WAIT) ==
285			     SYNC_FILE_RANGE_WRITE_AND_WAIT)
286			ret = filemap_fdatawrite_range(mapping, offset,
287					endbyte);
288		else
289			ret = filemap_flush_range(mapping, offset, endbyte);
290		if (ret < 0)
291			goto out;
292	}
293
294	if (flags & SYNC_FILE_RANGE_WAIT_AFTER)
295		ret = file_fdatawait_range(file, offset, endbyte);
296
297out:
298	return ret;
299}
300
301/*
302 * ksys_sync_file_range() permits finely controlled syncing over a segment of
303 * a file in the range offset .. (offset+nbytes-1) inclusive.  If nbytes is
304 * zero then ksys_sync_file_range() will operate from offset out to EOF.
305 *
306 * The flag bits are:
307 *
308 * SYNC_FILE_RANGE_WAIT_BEFORE: wait upon writeout of all pages in the range
309 * before performing the write.
310 *
311 * SYNC_FILE_RANGE_WRITE: initiate writeout of all those dirty pages in the
312 * range which are not presently under writeback. Note that this may block for
313 * significant periods due to exhaustion of disk request structures.
314 *
315 * SYNC_FILE_RANGE_WAIT_AFTER: wait upon writeout of all pages in the range
316 * after performing the write.
317 *
318 * Useful combinations of the flag bits are:
319 *
320 * SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE: ensures that all pages
321 * in the range which were dirty on entry to ksys_sync_file_range() are placed
322 * under writeout.  This is a start-write-for-data-integrity operation.
323 *
324 * SYNC_FILE_RANGE_WRITE: start writeout of all dirty pages in the range which
325 * are not presently under writeout.  This is an asynchronous flush-to-disk
326 * operation.  Not suitable for data integrity operations.
327 *
328 * SYNC_FILE_RANGE_WAIT_BEFORE (or SYNC_FILE_RANGE_WAIT_AFTER): wait for
329 * completion of writeout of all pages in the range.  This will be used after an
330 * earlier SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE operation to wait
331 * for that operation to complete and to return the result.
332 *
333 * SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE|SYNC_FILE_RANGE_WAIT_AFTER
334 * (a.k.a. SYNC_FILE_RANGE_WRITE_AND_WAIT):
335 * a traditional sync() operation.  This is a write-for-data-integrity operation
336 * which will ensure that all pages in the range which were dirty on entry to
337 * ksys_sync_file_range() are written to disk.  It should be noted that disk
338 * caches are not flushed by this call, so there are no guarantees here that the
339 * data will be available on disk after a crash.
340 *
341 *
342 * SYNC_FILE_RANGE_WAIT_BEFORE and SYNC_FILE_RANGE_WAIT_AFTER will detect any
343 * I/O errors or ENOSPC conditions and will return those to the caller, after
344 * clearing the EIO and ENOSPC flags in the address_space.
345 *
346 * It should be noted that none of these operations write out the file's
347 * metadata.  So unless the application is strictly performing overwrites of
348 * already-instantiated disk blocks, there are no guarantees here that the data
349 * will be available after a crash.
350 */
351int ksys_sync_file_range(int fd, loff_t offset, loff_t nbytes,
352			 unsigned int flags)
353{
354	CLASS(fd, f)(fd);
355
356	if (fd_empty(f))
357		return -EBADF;
358
359	return sync_file_range(fd_file(f), offset, nbytes, flags);
360}
361
362SYSCALL_DEFINE4(sync_file_range, int, fd, loff_t, offset, loff_t, nbytes,
363				unsigned int, flags)
364{
365	return ksys_sync_file_range(fd, offset, nbytes, flags);
366}
367
368#if defined(CONFIG_COMPAT) && defined(__ARCH_WANT_COMPAT_SYNC_FILE_RANGE)
369COMPAT_SYSCALL_DEFINE6(sync_file_range, int, fd, compat_arg_u64_dual(offset),
370		       compat_arg_u64_dual(nbytes), unsigned int, flags)
371{
372	return ksys_sync_file_range(fd, compat_arg_u64_glue(offset),
373				    compat_arg_u64_glue(nbytes), flags);
374}
375#endif
376
377/* It would be nice if people remember that not all the world's an i386
378   when they introduce new system calls */
379SYSCALL_DEFINE4(sync_file_range2, int, fd, unsigned int, flags,
380				 loff_t, offset, loff_t, nbytes)
381{
382	return ksys_sync_file_range(fd, offset, nbytes, flags);
383}