1 /*
2 * High-level sync()-related operations
3 */
4
5 #include <linux/kernel.h>
6 #include <linux/file.h>
7 #include <linux/fs.h>
8 #include <linux/slab.h>
9 #include <linux/export.h>
10 #include <linux/namei.h>
11 #include <linux/sched.h>
12 #include <linux/writeback.h>
13 #include <linux/syscalls.h>
14 #include <linux/linkage.h>
15 #include <linux/pagemap.h>
16 #include <linux/quotaops.h>
17 #include <linux/backing-dev.h>
18 #include "internal.h"
19
20 #define VALID_FLAGS (SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE| \
21 SYNC_FILE_RANGE_WAIT_AFTER)
22
23 /*
24 * Do the filesystem syncing work. For simple filesystems
25 * writeback_inodes_sb(sb) just dirties buffers with inodes so we have to
26 * submit IO for these buffers via __sync_blockdev(). This also speeds up the
27 * wait == 1 case since in that case write_inode() functions do
28 * sync_dirty_buffer() and thus effectively write one block at a time.
29 */
__sync_filesystem(struct super_block * sb,int wait)30 static int __sync_filesystem(struct super_block *sb, int wait)
31 {
32 if (wait)
33 sync_inodes_sb(sb);
34 else
35 writeback_inodes_sb(sb, WB_REASON_SYNC);
36
37 if (sb->s_op->sync_fs)
38 sb->s_op->sync_fs(sb, wait);
39 return __sync_blockdev(sb->s_bdev, wait);
40 }
41
42 /*
43 * Write out and wait upon all dirty data associated with this
44 * superblock. Filesystem data as well as the underlying block
45 * device. Takes the superblock lock.
46 */
sync_filesystem(struct super_block * sb)47 int sync_filesystem(struct super_block *sb)
48 {
49 int ret;
50
51 /*
52 * We need to be protected against the filesystem going from
53 * r/o to r/w or vice versa.
54 */
55 WARN_ON(!rwsem_is_locked(&sb->s_umount));
56
57 /*
58 * No point in syncing out anything if the filesystem is read-only.
59 */
60 if (sb->s_flags & MS_RDONLY)
61 return 0;
62
63 ret = __sync_filesystem(sb, 0);
64 if (ret < 0)
65 return ret;
66 return __sync_filesystem(sb, 1);
67 }
68 EXPORT_SYMBOL_GPL(sync_filesystem);
69
sync_inodes_one_sb(struct super_block * sb,void * arg)70 static void sync_inodes_one_sb(struct super_block *sb, void *arg)
71 {
72 if (!(sb->s_flags & MS_RDONLY))
73 sync_inodes_sb(sb);
74 }
75
sync_fs_one_sb(struct super_block * sb,void * arg)76 static void sync_fs_one_sb(struct super_block *sb, void *arg)
77 {
78 if (!(sb->s_flags & MS_RDONLY) && sb->s_op->sync_fs)
79 sb->s_op->sync_fs(sb, *(int *)arg);
80 }
81
fdatawrite_one_bdev(struct block_device * bdev,void * arg)82 static void fdatawrite_one_bdev(struct block_device *bdev, void *arg)
83 {
84 filemap_fdatawrite(bdev->bd_inode->i_mapping);
85 }
86
fdatawait_one_bdev(struct block_device * bdev,void * arg)87 static void fdatawait_one_bdev(struct block_device *bdev, void *arg)
88 {
89 filemap_fdatawait(bdev->bd_inode->i_mapping);
90 }
91
92 /*
93 * Sync everything. We start by waking flusher threads so that most of
94 * writeback runs on all devices in parallel. Then we sync all inodes reliably
95 * which effectively also waits for all flusher threads to finish doing
96 * writeback. At this point all data is on disk so metadata should be stable
97 * and we tell filesystems to sync their metadata via ->sync_fs() calls.
98 * Finally, we writeout all block devices because some filesystems (e.g. ext2)
99 * just write metadata (such as inodes or bitmaps) to block device page cache
100 * and do not sync it on their own in ->sync_fs().
101 */
SYSCALL_DEFINE0(sync)102 SYSCALL_DEFINE0(sync)
103 {
104 int nowait = 0, wait = 1;
105
106 wakeup_flusher_threads(0, WB_REASON_SYNC);
107 iterate_supers(sync_inodes_one_sb, NULL);
108 iterate_supers(sync_fs_one_sb, &nowait);
109 iterate_supers(sync_fs_one_sb, &wait);
110 iterate_bdevs(fdatawrite_one_bdev, NULL);
111 iterate_bdevs(fdatawait_one_bdev, NULL);
112 if (unlikely(laptop_mode))
113 laptop_sync_completion();
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 iterate_bdevs(fdatawrite_one_bdev, NULL);
128 iterate_supers(sync_inodes_one_sb, &nowait);
129 iterate_supers(sync_fs_one_sb, &nowait);
130 iterate_bdevs(fdatawrite_one_bdev, NULL);
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;
154
155 if (!f.file)
156 return -EBADF;
157 sb = f.file->f_dentry->d_sb;
158
159 down_read(&sb->s_umount);
160 ret = sync_filesystem(sb);
161 up_read(&sb->s_umount);
162
163 fdput(f);
164 return ret;
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 */
vfs_fsync_range(struct file * file,loff_t start,loff_t end,int datasync)178 int vfs_fsync_range(struct file *file, loff_t start, loff_t end, int datasync)
179 {
180 if (!file->f_op || !file->f_op->fsync)
181 return -EINVAL;
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 inc_syscfs(current);
209 }
210 return ret;
211 }
212
SYSCALL_DEFINE1(fsync,unsigned int,fd)213 SYSCALL_DEFINE1(fsync, unsigned int, fd)
214 {
215 return do_fsync(fd, 0);
216 }
217
SYSCALL_DEFINE1(fdatasync,unsigned int,fd)218 SYSCALL_DEFINE1(fdatasync, unsigned int, fd)
219 {
220 return do_fsync(fd, 1);
221 }
222
223 /**
224 * generic_write_sync - perform syncing after a write if file / inode is sync
225 * @file: file to which the write happened
226 * @pos: offset where the write started
227 * @count: length of the write
228 *
229 * This is just a simple wrapper about our general syncing function.
230 */
generic_write_sync(struct file * file,loff_t pos,loff_t count)231 int generic_write_sync(struct file *file, loff_t pos, loff_t count)
232 {
233 if (!(file->f_flags & O_DSYNC) && !IS_SYNC(file->f_mapping->host))
234 return 0;
235 return vfs_fsync_range(file, pos, pos + count - 1,
236 (file->f_flags & __O_SYNC) ? 0 : 1);
237 }
238 EXPORT_SYMBOL(generic_write_sync);
239
240 /*
241 * sys_sync_file_range() permits finely controlled syncing over a segment of
242 * a file in the range offset .. (offset+nbytes-1) inclusive. If nbytes is
243 * zero then sys_sync_file_range() will operate from offset out to EOF.
244 *
245 * The flag bits are:
246 *
247 * SYNC_FILE_RANGE_WAIT_BEFORE: wait upon writeout of all pages in the range
248 * before performing the write.
249 *
250 * SYNC_FILE_RANGE_WRITE: initiate writeout of all those dirty pages in the
251 * range which are not presently under writeback. Note that this may block for
252 * significant periods due to exhaustion of disk request structures.
253 *
254 * SYNC_FILE_RANGE_WAIT_AFTER: wait upon writeout of all pages in the range
255 * after performing the write.
256 *
257 * Useful combinations of the flag bits are:
258 *
259 * SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE: ensures that all pages
260 * in the range which were dirty on entry to sys_sync_file_range() are placed
261 * under writeout. This is a start-write-for-data-integrity operation.
262 *
263 * SYNC_FILE_RANGE_WRITE: start writeout of all dirty pages in the range which
264 * are not presently under writeout. This is an asynchronous flush-to-disk
265 * operation. Not suitable for data integrity operations.
266 *
267 * SYNC_FILE_RANGE_WAIT_BEFORE (or SYNC_FILE_RANGE_WAIT_AFTER): wait for
268 * completion of writeout of all pages in the range. This will be used after an
269 * earlier SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE operation to wait
270 * for that operation to complete and to return the result.
271 *
272 * SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE|SYNC_FILE_RANGE_WAIT_AFTER:
273 * a traditional sync() operation. This is a write-for-data-integrity operation
274 * which will ensure that all pages in the range which were dirty on entry to
275 * sys_sync_file_range() are committed to disk.
276 *
277 *
278 * SYNC_FILE_RANGE_WAIT_BEFORE and SYNC_FILE_RANGE_WAIT_AFTER will detect any
279 * I/O errors or ENOSPC conditions and will return those to the caller, after
280 * clearing the EIO and ENOSPC flags in the address_space.
281 *
282 * It should be noted that none of these operations write out the file's
283 * metadata. So unless the application is strictly performing overwrites of
284 * already-instantiated disk blocks, there are no guarantees here that the data
285 * will be available after a crash.
286 */
SYSCALL_DEFINE4(sync_file_range,int,fd,loff_t,offset,loff_t,nbytes,unsigned int,flags)287 SYSCALL_DEFINE4(sync_file_range, int, fd, loff_t, offset, loff_t, nbytes,
288 unsigned int, flags)
289 {
290 int ret;
291 struct fd f;
292 struct address_space *mapping;
293 loff_t endbyte; /* inclusive */
294 umode_t i_mode;
295
296 ret = -EINVAL;
297 if (flags & ~VALID_FLAGS)
298 goto out;
299
300 endbyte = offset + nbytes;
301
302 if ((s64)offset < 0)
303 goto out;
304 if ((s64)endbyte < 0)
305 goto out;
306 if (endbyte < offset)
307 goto out;
308
309 if (sizeof(pgoff_t) == 4) {
310 if (offset >= (0x100000000ULL << PAGE_CACHE_SHIFT)) {
311 /*
312 * The range starts outside a 32 bit machine's
313 * pagecache addressing capabilities. Let it "succeed"
314 */
315 ret = 0;
316 goto out;
317 }
318 if (endbyte >= (0x100000000ULL << PAGE_CACHE_SHIFT)) {
319 /*
320 * Out to EOF
321 */
322 nbytes = 0;
323 }
324 }
325
326 if (nbytes == 0)
327 endbyte = LLONG_MAX;
328 else
329 endbyte--; /* inclusive */
330
331 ret = -EBADF;
332 f = fdget(fd);
333 if (!f.file)
334 goto out;
335
336 i_mode = file_inode(f.file)->i_mode;
337 ret = -ESPIPE;
338 if (!S_ISREG(i_mode) && !S_ISBLK(i_mode) && !S_ISDIR(i_mode) &&
339 !S_ISLNK(i_mode))
340 goto out_put;
341
342 mapping = f.file->f_mapping;
343 if (!mapping) {
344 ret = -EINVAL;
345 goto out_put;
346 }
347
348 ret = 0;
349 if (flags & SYNC_FILE_RANGE_WAIT_BEFORE) {
350 ret = filemap_fdatawait_range(mapping, offset, endbyte);
351 if (ret < 0)
352 goto out_put;
353 }
354
355 if (flags & SYNC_FILE_RANGE_WRITE) {
356 ret = filemap_fdatawrite_range(mapping, offset, endbyte);
357 if (ret < 0)
358 goto out_put;
359 }
360
361 if (flags & SYNC_FILE_RANGE_WAIT_AFTER)
362 ret = filemap_fdatawait_range(mapping, offset, endbyte);
363
364 out_put:
365 fdput(f);
366 out:
367 return ret;
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 sys_sync_file_range(fd, offset, nbytes, flags);
376 }
377