projects/linux·blob master · refs

380 lines · 10.2 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}
108
109SYSCALL_DEFINE0(sync)
110{
111 ksys_sync();
112 return 0;
113}
114
115static void do_sync_work(struct work_struct *work)
116{
117 int nowait = 0;
118 int wait = 1;
119
120 /*
121 * Sync twice to reduce the possibility we skipped some inodes / pages
122 * because they were temporarily locked
123 */
124 iterate_supers(sync_inodes_one_sb, NULL);
125 iterate_supers(sync_fs_one_sb, &nowait);
126 sync_bdevs(false);
127 iterate_supers(sync_inodes_one_sb, NULL);
128 iterate_supers(sync_fs_one_sb, &wait);
129 sync_bdevs(false);
130 printk("Emergency Sync complete\n");
131 kfree(work);
132}
133
134void emergency_sync(void)
135{
136 struct work_struct *work;
137
138 work = kmalloc_obj(*work, GFP_ATOMIC);
139 if (work) {
140 INIT_WORK(work, do_sync_work);
141 schedule_work(work);
142 }
143}
144
145/*
146 * sync a single super
147 */
148SYSCALL_DEFINE1(syncfs, int, fd)
149{
150 CLASS(fd, f)(fd);
151 struct super_block *sb;
152 int ret, ret2;
153
154 if (fd_empty(f))
155 return -EBADF;
156 sb = fd_file(f)->f_path.dentry->d_sb;
157
158 down_read(&sb->s_umount);
159 ret = sync_filesystem(sb);
160 up_read(&sb->s_umount);
161
162 ret2 = errseq_check_and_advance(&sb->s_wb_err, &fd_file(f)->f_sb_err);
163
164 return ret ? ret : ret2;
165}
166
167/**
168 * vfs_fsync_range - helper to sync a range of data & metadata to disk
169 * @file: file to sync
170 * @start: offset in bytes of the beginning of data range to sync
171 * @end: offset in bytes of the end of data range (inclusive)
172 * @datasync: perform only datasync
173 *
174 * Write back data in range @start..@end and metadata for @file to disk. If
175 * @datasync is set only metadata needed to access modified file data is
176 * written.
177 */
178int vfs_fsync_range(struct file *file, loff_t start, loff_t end, int datasync)
179{
180 struct inode *inode = file->f_mapping->host;
181
182 if (!file->f_op->fsync)
183 return -EINVAL;
184 if (!datasync)
185 sync_lazytime(inode);
186 return file->f_op->fsync(file, start, end, datasync);
187}
188EXPORT_SYMBOL(vfs_fsync_range);
189
190/**
191 * vfs_fsync - perform a fsync or fdatasync on a file
192 * @file: file to sync
193 * @datasync: only perform a fdatasync operation
194 *
195 * Write back data and metadata for @file to disk. If @datasync is
196 * set only metadata needed to access modified file data is written.
197 */
198int vfs_fsync(struct file *file, int datasync)
199{
200 return vfs_fsync_range(file, 0, LLONG_MAX, datasync);
201}
202EXPORT_SYMBOL(vfs_fsync);
203
204static int do_fsync(unsigned int fd, int datasync)
205{
206 CLASS(fd, f)(fd);
207
208 if (fd_empty(f))
209 return -EBADF;
210
211 return vfs_fsync(fd_file(f), datasync);
212}
213
214SYSCALL_DEFINE1(fsync, unsigned int, fd)
215{
216 return do_fsync(fd, 0);
217}
218
219SYSCALL_DEFINE1(fdatasync, unsigned int, fd)
220{
221 return do_fsync(fd, 1);
222}
223
224int sync_file_range(struct file *file, loff_t offset, loff_t nbytes,
225 unsigned int flags)
226{
227 int ret;
228 struct address_space *mapping;
229 loff_t endbyte; /* inclusive */
230 umode_t i_mode;
231
232 ret = -EINVAL;
233 if (flags & ~VALID_FLAGS)
234 goto out;
235
236 endbyte = offset + nbytes;
237
238 if ((s64)offset < 0)
239 goto out;
240 if ((s64)endbyte < 0)
241 goto out;
242 if (endbyte < offset)
243 goto out;
244
245 if (sizeof(pgoff_t) == 4) {
246 if (offset >= (0x100000000ULL << PAGE_SHIFT)) {
247 /*
248 * The range starts outside a 32 bit machine's
249 * pagecache addressing capabilities. Let it "succeed"
250 */
251 ret = 0;
252 goto out;
253 }
254 if (endbyte >= (0x100000000ULL << PAGE_SHIFT)) {
255 /*
256 * Out to EOF
257 */
258 nbytes = 0;
259 }
260 }
261
262 if (nbytes == 0)
263 endbyte = LLONG_MAX;
264 else
265 endbyte--; /* inclusive */
266
267 i_mode = file_inode(file)->i_mode;
268 ret = -ESPIPE;
269 if (!S_ISREG(i_mode) && !S_ISBLK(i_mode) && !S_ISDIR(i_mode))
270 goto out;
271
272 mapping = file->f_mapping;
273 ret = 0;
274 if (flags & SYNC_FILE_RANGE_WAIT_BEFORE) {
275 ret = file_fdatawait_range(file, offset, endbyte);
276 if (ret < 0)
277 goto out;
278 }
279
280 if (flags & SYNC_FILE_RANGE_WRITE) {
281 if ((flags & SYNC_FILE_RANGE_WRITE_AND_WAIT) ==
282 SYNC_FILE_RANGE_WRITE_AND_WAIT)
283 ret = filemap_fdatawrite_range(mapping, offset,
284 endbyte);
285 else
286 ret = filemap_flush_range(mapping, offset, endbyte);
287 if (ret < 0)
288 goto out;
289 }
290
291 if (flags & SYNC_FILE_RANGE_WAIT_AFTER)
292 ret = file_fdatawait_range(file, offset, endbyte);
293
294out:
295 return ret;
296}
297
298/*
299 * ksys_sync_file_range() permits finely controlled syncing over a segment of
300 * a file in the range offset .. (offset+nbytes-1) inclusive. If nbytes is
301 * zero then ksys_sync_file_range() will operate from offset out to EOF.
302 *
303 * The flag bits are:
304 *
305 * SYNC_FILE_RANGE_WAIT_BEFORE: wait upon writeout of all pages in the range
306 * before performing the write.
307 *
308 * SYNC_FILE_RANGE_WRITE: initiate writeout of all those dirty pages in the
309 * range which are not presently under writeback. Note that this may block for
310 * significant periods due to exhaustion of disk request structures.
311 *
312 * SYNC_FILE_RANGE_WAIT_AFTER: wait upon writeout of all pages in the range
313 * after performing the write.
314 *
315 * Useful combinations of the flag bits are:
316 *
317 * SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE: ensures that all pages
318 * in the range which were dirty on entry to ksys_sync_file_range() are placed
319 * under writeout. This is a start-write-for-data-integrity operation.
320 *
321 * SYNC_FILE_RANGE_WRITE: start writeout of all dirty pages in the range which
322 * are not presently under writeout. This is an asynchronous flush-to-disk
323 * operation. Not suitable for data integrity operations.
324 *
325 * SYNC_FILE_RANGE_WAIT_BEFORE (or SYNC_FILE_RANGE_WAIT_AFTER): wait for
326 * completion of writeout of all pages in the range. This will be used after an
327 * earlier SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE operation to wait
328 * for that operation to complete and to return the result.
329 *
330 * SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE|SYNC_FILE_RANGE_WAIT_AFTER
331 * (a.k.a. SYNC_FILE_RANGE_WRITE_AND_WAIT):
332 * a traditional sync() operation. This is a write-for-data-integrity operation
333 * which will ensure that all pages in the range which were dirty on entry to
334 * ksys_sync_file_range() are written to disk. It should be noted that disk
335 * caches are not flushed by this call, so there are no guarantees here that the
336 * data will be available on disk after a crash.
337 *
338 *
339 * SYNC_FILE_RANGE_WAIT_BEFORE and SYNC_FILE_RANGE_WAIT_AFTER will detect any
340 * I/O errors or ENOSPC conditions and will return those to the caller, after
341 * clearing the EIO and ENOSPC flags in the address_space.
342 *
343 * It should be noted that none of these operations write out the file's
344 * metadata. So unless the application is strictly performing overwrites of
345 * already-instantiated disk blocks, there are no guarantees here that the data
346 * will be available after a crash.
347 */
348int ksys_sync_file_range(int fd, loff_t offset, loff_t nbytes,
349 unsigned int flags)
350{
351 CLASS(fd, f)(fd);
352
353 if (fd_empty(f))
354 return -EBADF;
355
356 return sync_file_range(fd_file(f), offset, nbytes, flags);
357}
358
359SYSCALL_DEFINE4(sync_file_range, int, fd, loff_t, offset, loff_t, nbytes,
360 unsigned int, flags)
361{
362 return ksys_sync_file_range(fd, offset, nbytes, flags);
363}
364
365#if defined(CONFIG_COMPAT) && defined(__ARCH_WANT_COMPAT_SYNC_FILE_RANGE)
366COMPAT_SYSCALL_DEFINE6(sync_file_range, int, fd, compat_arg_u64_dual(offset),
367 compat_arg_u64_dual(nbytes), unsigned int, flags)
368{
369 return ksys_sync_file_range(fd, compat_arg_u64_glue(offset),
370 compat_arg_u64_glue(nbytes), flags);
371}
372#endif
373
374/* It would be nice if people remember that not all the world's an i386
375 when they introduce new system calls */
376SYSCALL_DEFINE4(sync_file_range2, int, fd, unsigned int, flags,
377 loff_t, offset, loff_t, nbytes)
378{
379 return ksys_sync_file_range(fd, offset, nbytes, flags);
380}