projects/linux·blob v6.12 · refs

391 lines · 10.4 KiB

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
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
135void 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 */
149SYSCALL_DEFINE1(syncfs, int, fd)
150{
151 struct fd f = fdget(fd);
152 struct super_block *sb;
153 int ret, ret2;
154
155 if (!fd_file(f))
156 return -EBADF;
157 sb = fd_file(f)->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, &fd_file(f)->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 */
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->i_state & 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 struct fd f = fdget(fd);
209 int ret = -EBADF;
210
211 if (fd_file(f)) {
212 ret = vfs_fsync(fd_file(f), datasync);
213 fdput(f);
214 }
215 return ret;
216}
217
218SYSCALL_DEFINE1(fsync, unsigned int, fd)
219{
220 return do_fsync(fd, 0);
221}
222
223SYSCALL_DEFINE1(fdatasync, unsigned int, fd)
224{
225 return do_fsync(fd, 1);
226}
227
228int sync_file_range(struct file *file, loff_t offset, loff_t nbytes,
229 unsigned int flags)
230{
231 int ret;
232 struct address_space *mapping;
233 loff_t endbyte; /* inclusive */
234 umode_t i_mode;
235
236 ret = -EINVAL;
237 if (flags & ~VALID_FLAGS)
238 goto out;
239
240 endbyte = offset + nbytes;
241
242 if ((s64)offset < 0)
243 goto out;
244 if ((s64)endbyte < 0)
245 goto out;
246 if (endbyte < offset)
247 goto out;
248
249 if (sizeof(pgoff_t) == 4) {
250 if (offset >= (0x100000000ULL << PAGE_SHIFT)) {
251 /*
252 * The range starts outside a 32 bit machine's
253 * pagecache addressing capabilities. Let it "succeed"
254 */
255 ret = 0;
256 goto out;
257 }
258 if (endbyte >= (0x100000000ULL << PAGE_SHIFT)) {
259 /*
260 * Out to EOF
261 */
262 nbytes = 0;
263 }
264 }
265
266 if (nbytes == 0)
267 endbyte = LLONG_MAX;
268 else
269 endbyte--; /* inclusive */
270
271 i_mode = file_inode(file)->i_mode;
272 ret = -ESPIPE;
273 if (!S_ISREG(i_mode) && !S_ISBLK(i_mode) && !S_ISDIR(i_mode) &&
274 !S_ISLNK(i_mode))
275 goto out;
276
277 mapping = file->f_mapping;
278 ret = 0;
279 if (flags & SYNC_FILE_RANGE_WAIT_BEFORE) {
280 ret = file_fdatawait_range(file, offset, endbyte);
281 if (ret < 0)
282 goto out;
283 }
284
285 if (flags & SYNC_FILE_RANGE_WRITE) {
286 int sync_mode = WB_SYNC_NONE;
287
288 if ((flags & SYNC_FILE_RANGE_WRITE_AND_WAIT) ==
289 SYNC_FILE_RANGE_WRITE_AND_WAIT)
290 sync_mode = WB_SYNC_ALL;
291
292 ret = __filemap_fdatawrite_range(mapping, offset, endbyte,
293 sync_mode);
294 if (ret < 0)
295 goto out;
296 }
297
298 if (flags & SYNC_FILE_RANGE_WAIT_AFTER)
299 ret = file_fdatawait_range(file, offset, endbyte);
300
301out:
302 return ret;
303}
304
305/*
306 * ksys_sync_file_range() permits finely controlled syncing over a segment of
307 * a file in the range offset .. (offset+nbytes-1) inclusive. If nbytes is
308 * zero then ksys_sync_file_range() will operate from offset out to EOF.
309 *
310 * The flag bits are:
311 *
312 * SYNC_FILE_RANGE_WAIT_BEFORE: wait upon writeout of all pages in the range
313 * before performing the write.
314 *
315 * SYNC_FILE_RANGE_WRITE: initiate writeout of all those dirty pages in the
316 * range which are not presently under writeback. Note that this may block for
317 * significant periods due to exhaustion of disk request structures.
318 *
319 * SYNC_FILE_RANGE_WAIT_AFTER: wait upon writeout of all pages in the range
320 * after performing the write.
321 *
322 * Useful combinations of the flag bits are:
323 *
324 * SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE: ensures that all pages
325 * in the range which were dirty on entry to ksys_sync_file_range() are placed
326 * under writeout. This is a start-write-for-data-integrity operation.
327 *
328 * SYNC_FILE_RANGE_WRITE: start writeout of all dirty pages in the range which
329 * are not presently under writeout. This is an asynchronous flush-to-disk
330 * operation. Not suitable for data integrity operations.
331 *
332 * SYNC_FILE_RANGE_WAIT_BEFORE (or SYNC_FILE_RANGE_WAIT_AFTER): wait for
333 * completion of writeout of all pages in the range. This will be used after an
334 * earlier SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE operation to wait
335 * for that operation to complete and to return the result.
336 *
337 * SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE|SYNC_FILE_RANGE_WAIT_AFTER
338 * (a.k.a. SYNC_FILE_RANGE_WRITE_AND_WAIT):
339 * a traditional sync() operation. This is a write-for-data-integrity operation
340 * which will ensure that all pages in the range which were dirty on entry to
341 * ksys_sync_file_range() are written to disk. It should be noted that disk
342 * caches are not flushed by this call, so there are no guarantees here that the
343 * data will be available on disk after a crash.
344 *
345 *
346 * SYNC_FILE_RANGE_WAIT_BEFORE and SYNC_FILE_RANGE_WAIT_AFTER will detect any
347 * I/O errors or ENOSPC conditions and will return those to the caller, after
348 * clearing the EIO and ENOSPC flags in the address_space.
349 *
350 * It should be noted that none of these operations write out the file's
351 * metadata. So unless the application is strictly performing overwrites of
352 * already-instantiated disk blocks, there are no guarantees here that the data
353 * will be available after a crash.
354 */
355int ksys_sync_file_range(int fd, loff_t offset, loff_t nbytes,
356 unsigned int flags)
357{
358 int ret;
359 struct fd f;
360
361 ret = -EBADF;
362 f = fdget(fd);
363 if (fd_file(f))
364 ret = sync_file_range(fd_file(f), offset, nbytes, flags);
365
366 fdput(f);
367 return ret;
368}
369
370SYSCALL_DEFINE4(sync_file_range, int, fd, loff_t, offset, loff_t, nbytes,
371 unsigned int, flags)
372{
373 return ksys_sync_file_range(fd, offset, nbytes, flags);
374}
375
376#if defined(CONFIG_COMPAT) && defined(__ARCH_WANT_COMPAT_SYNC_FILE_RANGE)
377COMPAT_SYSCALL_DEFINE6(sync_file_range, int, fd, compat_arg_u64_dual(offset),
378 compat_arg_u64_dual(nbytes), unsigned int, flags)
379{
380 return ksys_sync_file_range(fd, compat_arg_u64_glue(offset),
381 compat_arg_u64_glue(nbytes), flags);
382}
383#endif
384
385/* It would be nice if people remember that not all the world's an i386
386 when they introduce new system calls */
387SYSCALL_DEFINE4(sync_file_range2, int, fd, unsigned int, flags,
388 loff_t, offset, loff_t, nbytes)
389{
390 return ksys_sync_file_range(fd, offset, nbytes, flags);
391}