]> git.proxmox.com Git - mirror_ubuntu-artful-kernel.git/blob - fs/sync.c
Merge commit 'v2.6.37-rc7' into x86/security
[mirror_ubuntu-artful-kernel.git] / fs / sync.c
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/module.h>
10 #include <linux/sched.h>
11 #include <linux/writeback.h>
12 #include <linux/syscalls.h>
13 #include <linux/linkage.h>
14 #include <linux/pagemap.h>
15 #include <linux/quotaops.h>
16 #include <linux/buffer_head.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 */
30 static int __sync_filesystem(struct super_block *sb, int wait)
31 {
32 /*
33 * This should be safe, as we require bdi backing to actually
34 * write out data in the first place
35 */
36 if (!sb->s_bdi || sb->s_bdi == &noop_backing_dev_info)
37 return 0;
38
39 if (sb->s_qcop && sb->s_qcop->quota_sync)
40 sb->s_qcop->quota_sync(sb, -1, wait);
41
42 if (wait)
43 sync_inodes_sb(sb);
44 else
45 writeback_inodes_sb(sb);
46
47 if (sb->s_op->sync_fs)
48 sb->s_op->sync_fs(sb, wait);
49 return __sync_blockdev(sb->s_bdev, wait);
50 }
51
52 /*
53 * Write out and wait upon all dirty data associated with this
54 * superblock. Filesystem data as well as the underlying block
55 * device. Takes the superblock lock.
56 */
57 int sync_filesystem(struct super_block *sb)
58 {
59 int ret;
60
61 /*
62 * We need to be protected against the filesystem going from
63 * r/o to r/w or vice versa.
64 */
65 WARN_ON(!rwsem_is_locked(&sb->s_umount));
66
67 /*
68 * No point in syncing out anything if the filesystem is read-only.
69 */
70 if (sb->s_flags & MS_RDONLY)
71 return 0;
72
73 ret = __sync_filesystem(sb, 0);
74 if (ret < 0)
75 return ret;
76 return __sync_filesystem(sb, 1);
77 }
78 EXPORT_SYMBOL_GPL(sync_filesystem);
79
80 static void sync_one_sb(struct super_block *sb, void *arg)
81 {
82 if (!(sb->s_flags & MS_RDONLY) && sb->s_bdi)
83 __sync_filesystem(sb, *(int *)arg);
84 }
85 /*
86 * Sync all the data for all the filesystems (called by sys_sync() and
87 * emergency sync)
88 */
89 static void sync_filesystems(int wait)
90 {
91 iterate_supers(sync_one_sb, &wait);
92 }
93
94 /*
95 * sync everything. Start out by waking pdflush, because that writes back
96 * all queues in parallel.
97 */
98 SYSCALL_DEFINE0(sync)
99 {
100 wakeup_flusher_threads(0);
101 sync_filesystems(0);
102 sync_filesystems(1);
103 if (unlikely(laptop_mode))
104 laptop_sync_completion();
105 return 0;
106 }
107
108 static void do_sync_work(struct work_struct *work)
109 {
110 /*
111 * Sync twice to reduce the possibility we skipped some inodes / pages
112 * because they were temporarily locked
113 */
114 sync_filesystems(0);
115 sync_filesystems(0);
116 printk("Emergency Sync complete\n");
117 kfree(work);
118 }
119
120 void emergency_sync(void)
121 {
122 struct work_struct *work;
123
124 work = kmalloc(sizeof(*work), GFP_ATOMIC);
125 if (work) {
126 INIT_WORK(work, do_sync_work);
127 schedule_work(work);
128 }
129 }
130
131 /**
132 * vfs_fsync_range - helper to sync a range of data & metadata to disk
133 * @file: file to sync
134 * @start: offset in bytes of the beginning of data range to sync
135 * @end: offset in bytes of the end of data range (inclusive)
136 * @datasync: perform only datasync
137 *
138 * Write back data in range @start..@end and metadata for @file to disk. If
139 * @datasync is set only metadata needed to access modified file data is
140 * written.
141 */
142 int vfs_fsync_range(struct file *file, loff_t start, loff_t end, int datasync)
143 {
144 struct address_space *mapping = file->f_mapping;
145 int err, ret;
146
147 if (!file->f_op || !file->f_op->fsync) {
148 ret = -EINVAL;
149 goto out;
150 }
151
152 ret = filemap_write_and_wait_range(mapping, start, end);
153
154 /*
155 * We need to protect against concurrent writers, which could cause
156 * livelocks in fsync_buffers_list().
157 */
158 mutex_lock(&mapping->host->i_mutex);
159 err = file->f_op->fsync(file, datasync);
160 if (!ret)
161 ret = err;
162 mutex_unlock(&mapping->host->i_mutex);
163
164 out:
165 return ret;
166 }
167 EXPORT_SYMBOL(vfs_fsync_range);
168
169 /**
170 * vfs_fsync - perform a fsync or fdatasync on a file
171 * @file: file to sync
172 * @datasync: only perform a fdatasync operation
173 *
174 * Write back data and metadata for @file to disk. If @datasync is
175 * set only metadata needed to access modified file data is written.
176 */
177 int vfs_fsync(struct file *file, int datasync)
178 {
179 return vfs_fsync_range(file, 0, LLONG_MAX, datasync);
180 }
181 EXPORT_SYMBOL(vfs_fsync);
182
183 static int do_fsync(unsigned int fd, int datasync)
184 {
185 struct file *file;
186 int ret = -EBADF;
187
188 file = fget(fd);
189 if (file) {
190 ret = vfs_fsync(file, datasync);
191 fput(file);
192 }
193 return ret;
194 }
195
196 SYSCALL_DEFINE1(fsync, unsigned int, fd)
197 {
198 return do_fsync(fd, 0);
199 }
200
201 SYSCALL_DEFINE1(fdatasync, unsigned int, fd)
202 {
203 return do_fsync(fd, 1);
204 }
205
206 /**
207 * generic_write_sync - perform syncing after a write if file / inode is sync
208 * @file: file to which the write happened
209 * @pos: offset where the write started
210 * @count: length of the write
211 *
212 * This is just a simple wrapper about our general syncing function.
213 */
214 int generic_write_sync(struct file *file, loff_t pos, loff_t count)
215 {
216 if (!(file->f_flags & O_DSYNC) && !IS_SYNC(file->f_mapping->host))
217 return 0;
218 return vfs_fsync_range(file, pos, pos + count - 1,
219 (file->f_flags & __O_SYNC) ? 0 : 1);
220 }
221 EXPORT_SYMBOL(generic_write_sync);
222
223 /*
224 * sys_sync_file_range() permits finely controlled syncing over a segment of
225 * a file in the range offset .. (offset+nbytes-1) inclusive. If nbytes is
226 * zero then sys_sync_file_range() will operate from offset out to EOF.
227 *
228 * The flag bits are:
229 *
230 * SYNC_FILE_RANGE_WAIT_BEFORE: wait upon writeout of all pages in the range
231 * before performing the write.
232 *
233 * SYNC_FILE_RANGE_WRITE: initiate writeout of all those dirty pages in the
234 * range which are not presently under writeback. Note that this may block for
235 * significant periods due to exhaustion of disk request structures.
236 *
237 * SYNC_FILE_RANGE_WAIT_AFTER: wait upon writeout of all pages in the range
238 * after performing the write.
239 *
240 * Useful combinations of the flag bits are:
241 *
242 * SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE: ensures that all pages
243 * in the range which were dirty on entry to sys_sync_file_range() are placed
244 * under writeout. This is a start-write-for-data-integrity operation.
245 *
246 * SYNC_FILE_RANGE_WRITE: start writeout of all dirty pages in the range which
247 * are not presently under writeout. This is an asynchronous flush-to-disk
248 * operation. Not suitable for data integrity operations.
249 *
250 * SYNC_FILE_RANGE_WAIT_BEFORE (or SYNC_FILE_RANGE_WAIT_AFTER): wait for
251 * completion of writeout of all pages in the range. This will be used after an
252 * earlier SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE operation to wait
253 * for that operation to complete and to return the result.
254 *
255 * SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE|SYNC_FILE_RANGE_WAIT_AFTER:
256 * a traditional sync() operation. This is a write-for-data-integrity operation
257 * which will ensure that all pages in the range which were dirty on entry to
258 * sys_sync_file_range() are committed to disk.
259 *
260 *
261 * SYNC_FILE_RANGE_WAIT_BEFORE and SYNC_FILE_RANGE_WAIT_AFTER will detect any
262 * I/O errors or ENOSPC conditions and will return those to the caller, after
263 * clearing the EIO and ENOSPC flags in the address_space.
264 *
265 * It should be noted that none of these operations write out the file's
266 * metadata. So unless the application is strictly performing overwrites of
267 * already-instantiated disk blocks, there are no guarantees here that the data
268 * will be available after a crash.
269 */
270 SYSCALL_DEFINE(sync_file_range)(int fd, loff_t offset, loff_t nbytes,
271 unsigned int flags)
272 {
273 int ret;
274 struct file *file;
275 struct address_space *mapping;
276 loff_t endbyte; /* inclusive */
277 int fput_needed;
278 umode_t i_mode;
279
280 ret = -EINVAL;
281 if (flags & ~VALID_FLAGS)
282 goto out;
283
284 endbyte = offset + nbytes;
285
286 if ((s64)offset < 0)
287 goto out;
288 if ((s64)endbyte < 0)
289 goto out;
290 if (endbyte < offset)
291 goto out;
292
293 if (sizeof(pgoff_t) == 4) {
294 if (offset >= (0x100000000ULL << PAGE_CACHE_SHIFT)) {
295 /*
296 * The range starts outside a 32 bit machine's
297 * pagecache addressing capabilities. Let it "succeed"
298 */
299 ret = 0;
300 goto out;
301 }
302 if (endbyte >= (0x100000000ULL << PAGE_CACHE_SHIFT)) {
303 /*
304 * Out to EOF
305 */
306 nbytes = 0;
307 }
308 }
309
310 if (nbytes == 0)
311 endbyte = LLONG_MAX;
312 else
313 endbyte--; /* inclusive */
314
315 ret = -EBADF;
316 file = fget_light(fd, &fput_needed);
317 if (!file)
318 goto out;
319
320 i_mode = file->f_path.dentry->d_inode->i_mode;
321 ret = -ESPIPE;
322 if (!S_ISREG(i_mode) && !S_ISBLK(i_mode) && !S_ISDIR(i_mode) &&
323 !S_ISLNK(i_mode))
324 goto out_put;
325
326 mapping = file->f_mapping;
327 if (!mapping) {
328 ret = -EINVAL;
329 goto out_put;
330 }
331
332 ret = 0;
333 if (flags & SYNC_FILE_RANGE_WAIT_BEFORE) {
334 ret = filemap_fdatawait_range(mapping, offset, endbyte);
335 if (ret < 0)
336 goto out_put;
337 }
338
339 if (flags & SYNC_FILE_RANGE_WRITE) {
340 ret = filemap_fdatawrite_range(mapping, offset, endbyte);
341 if (ret < 0)
342 goto out_put;
343 }
344
345 if (flags & SYNC_FILE_RANGE_WAIT_AFTER)
346 ret = filemap_fdatawait_range(mapping, offset, endbyte);
347
348 out_put:
349 fput_light(file, fput_needed);
350 out:
351 return ret;
352 }
353 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
354 asmlinkage long SyS_sync_file_range(long fd, loff_t offset, loff_t nbytes,
355 long flags)
356 {
357 return SYSC_sync_file_range((int) fd, offset, nbytes,
358 (unsigned int) flags);
359 }
360 SYSCALL_ALIAS(sys_sync_file_range, SyS_sync_file_range);
361 #endif
362
363 /* It would be nice if people remember that not all the world's an i386
364 when they introduce new system calls */
365 SYSCALL_DEFINE(sync_file_range2)(int fd, unsigned int flags,
366 loff_t offset, loff_t nbytes)
367 {
368 return sys_sync_file_range(fd, offset, nbytes, flags);
369 }
370 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
371 asmlinkage long SyS_sync_file_range2(long fd, long flags,
372 loff_t offset, loff_t nbytes)
373 {
374 return SYSC_sync_file_range2((int) fd, (unsigned int) flags,
375 offset, nbytes);
376 }
377 SYSCALL_ALIAS(sys_sync_file_range2, SyS_sync_file_range2);
378 #endif