xref: /openbmc/linux/fs/sync.c (revision b8bb76713ec50df2f11efee386e16f93d51e1076)
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/module.h>
9 #include <linux/sched.h>
10 #include <linux/writeback.h>
11 #include <linux/syscalls.h>
12 #include <linux/linkage.h>
13 #include <linux/pagemap.h>
14 #include <linux/quotaops.h>
15 #include <linux/buffer_head.h>
16 
17 #define VALID_FLAGS (SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE| \
18 			SYNC_FILE_RANGE_WAIT_AFTER)
19 
20 /*
21  * sync everything.  Start out by waking pdflush, because that writes back
22  * all queues in parallel.
23  */
24 static void do_sync(unsigned long wait)
25 {
26 	wakeup_pdflush(0);
27 	sync_inodes(0);		/* All mappings, inodes and their blockdevs */
28 	vfs_dq_sync(NULL);
29 	sync_supers();		/* Write the superblocks */
30 	sync_filesystems(0);	/* Start syncing the filesystems */
31 	sync_filesystems(wait);	/* Waitingly sync the filesystems */
32 	sync_inodes(wait);	/* Mappings, inodes and blockdevs, again. */
33 	if (!wait)
34 		printk("Emergency Sync complete\n");
35 	if (unlikely(laptop_mode))
36 		laptop_sync_completion();
37 }
38 
39 SYSCALL_DEFINE0(sync)
40 {
41 	do_sync(1);
42 	return 0;
43 }
44 
45 static void do_sync_work(struct work_struct *work)
46 {
47 	do_sync(0);
48 	kfree(work);
49 }
50 
51 void emergency_sync(void)
52 {
53 	struct work_struct *work;
54 
55 	work = kmalloc(sizeof(*work), GFP_ATOMIC);
56 	if (work) {
57 		INIT_WORK(work, do_sync_work);
58 		schedule_work(work);
59 	}
60 }
61 
62 /*
63  * Generic function to fsync a file.
64  *
65  * filp may be NULL if called via the msync of a vma.
66  */
67 int file_fsync(struct file *filp, struct dentry *dentry, int datasync)
68 {
69 	struct inode * inode = dentry->d_inode;
70 	struct super_block * sb;
71 	int ret, err;
72 
73 	/* sync the inode to buffers */
74 	ret = write_inode_now(inode, 0);
75 
76 	/* sync the superblock to buffers */
77 	sb = inode->i_sb;
78 	lock_super(sb);
79 	if (sb->s_dirt && sb->s_op->write_super)
80 		sb->s_op->write_super(sb);
81 	unlock_super(sb);
82 
83 	/* .. finally sync the buffers to disk */
84 	err = sync_blockdev(sb->s_bdev);
85 	if (!ret)
86 		ret = err;
87 	return ret;
88 }
89 
90 /**
91  * vfs_fsync - perform a fsync or fdatasync on a file
92  * @file:		file to sync
93  * @dentry:		dentry of @file
94  * @data:		only perform a fdatasync operation
95  *
96  * Write back data and metadata for @file to disk.  If @datasync is
97  * set only metadata needed to access modified file data is written.
98  *
99  * In case this function is called from nfsd @file may be %NULL and
100  * only @dentry is set.  This can only happen when the filesystem
101  * implements the export_operations API.
102  */
103 int vfs_fsync(struct file *file, struct dentry *dentry, int datasync)
104 {
105 	const struct file_operations *fop;
106 	struct address_space *mapping;
107 	int err, ret;
108 
109 	/*
110 	 * Get mapping and operations from the file in case we have
111 	 * as file, or get the default values for them in case we
112 	 * don't have a struct file available.  Damn nfsd..
113 	 */
114 	if (file) {
115 		mapping = file->f_mapping;
116 		fop = file->f_op;
117 	} else {
118 		mapping = dentry->d_inode->i_mapping;
119 		fop = dentry->d_inode->i_fop;
120 	}
121 
122 	if (!fop || !fop->fsync) {
123 		ret = -EINVAL;
124 		goto out;
125 	}
126 
127 	ret = filemap_fdatawrite(mapping);
128 
129 	/*
130 	 * We need to protect against concurrent writers, which could cause
131 	 * livelocks in fsync_buffers_list().
132 	 */
133 	mutex_lock(&mapping->host->i_mutex);
134 	err = fop->fsync(file, dentry, datasync);
135 	if (!ret)
136 		ret = err;
137 	mutex_unlock(&mapping->host->i_mutex);
138 	err = filemap_fdatawait(mapping);
139 	if (!ret)
140 		ret = err;
141 out:
142 	return ret;
143 }
144 EXPORT_SYMBOL(vfs_fsync);
145 
146 static int do_fsync(unsigned int fd, int datasync)
147 {
148 	struct file *file;
149 	int ret = -EBADF;
150 
151 	file = fget(fd);
152 	if (file) {
153 		ret = vfs_fsync(file, file->f_path.dentry, datasync);
154 		fput(file);
155 	}
156 	return ret;
157 }
158 
159 SYSCALL_DEFINE1(fsync, unsigned int, fd)
160 {
161 	return do_fsync(fd, 0);
162 }
163 
164 SYSCALL_DEFINE1(fdatasync, unsigned int, fd)
165 {
166 	return do_fsync(fd, 1);
167 }
168 
169 /*
170  * sys_sync_file_range() permits finely controlled syncing over a segment of
171  * a file in the range offset .. (offset+nbytes-1) inclusive.  If nbytes is
172  * zero then sys_sync_file_range() will operate from offset out to EOF.
173  *
174  * The flag bits are:
175  *
176  * SYNC_FILE_RANGE_WAIT_BEFORE: wait upon writeout of all pages in the range
177  * before performing the write.
178  *
179  * SYNC_FILE_RANGE_WRITE: initiate writeout of all those dirty pages in the
180  * range which are not presently under writeback. Note that this may block for
181  * significant periods due to exhaustion of disk request structures.
182  *
183  * SYNC_FILE_RANGE_WAIT_AFTER: wait upon writeout of all pages in the range
184  * after performing the write.
185  *
186  * Useful combinations of the flag bits are:
187  *
188  * SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE: ensures that all pages
189  * in the range which were dirty on entry to sys_sync_file_range() are placed
190  * under writeout.  This is a start-write-for-data-integrity operation.
191  *
192  * SYNC_FILE_RANGE_WRITE: start writeout of all dirty pages in the range which
193  * are not presently under writeout.  This is an asynchronous flush-to-disk
194  * operation.  Not suitable for data integrity operations.
195  *
196  * SYNC_FILE_RANGE_WAIT_BEFORE (or SYNC_FILE_RANGE_WAIT_AFTER): wait for
197  * completion of writeout of all pages in the range.  This will be used after an
198  * earlier SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE operation to wait
199  * for that operation to complete and to return the result.
200  *
201  * SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE|SYNC_FILE_RANGE_WAIT_AFTER:
202  * a traditional sync() operation.  This is a write-for-data-integrity operation
203  * which will ensure that all pages in the range which were dirty on entry to
204  * sys_sync_file_range() are committed to disk.
205  *
206  *
207  * SYNC_FILE_RANGE_WAIT_BEFORE and SYNC_FILE_RANGE_WAIT_AFTER will detect any
208  * I/O errors or ENOSPC conditions and will return those to the caller, after
209  * clearing the EIO and ENOSPC flags in the address_space.
210  *
211  * It should be noted that none of these operations write out the file's
212  * metadata.  So unless the application is strictly performing overwrites of
213  * already-instantiated disk blocks, there are no guarantees here that the data
214  * will be available after a crash.
215  */
216 SYSCALL_DEFINE(sync_file_range)(int fd, loff_t offset, loff_t nbytes,
217 				unsigned int flags)
218 {
219 	int ret;
220 	struct file *file;
221 	loff_t endbyte;			/* inclusive */
222 	int fput_needed;
223 	umode_t i_mode;
224 
225 	ret = -EINVAL;
226 	if (flags & ~VALID_FLAGS)
227 		goto out;
228 
229 	endbyte = offset + nbytes;
230 
231 	if ((s64)offset < 0)
232 		goto out;
233 	if ((s64)endbyte < 0)
234 		goto out;
235 	if (endbyte < offset)
236 		goto out;
237 
238 	if (sizeof(pgoff_t) == 4) {
239 		if (offset >= (0x100000000ULL << PAGE_CACHE_SHIFT)) {
240 			/*
241 			 * The range starts outside a 32 bit machine's
242 			 * pagecache addressing capabilities.  Let it "succeed"
243 			 */
244 			ret = 0;
245 			goto out;
246 		}
247 		if (endbyte >= (0x100000000ULL << PAGE_CACHE_SHIFT)) {
248 			/*
249 			 * Out to EOF
250 			 */
251 			nbytes = 0;
252 		}
253 	}
254 
255 	if (nbytes == 0)
256 		endbyte = LLONG_MAX;
257 	else
258 		endbyte--;		/* inclusive */
259 
260 	ret = -EBADF;
261 	file = fget_light(fd, &fput_needed);
262 	if (!file)
263 		goto out;
264 
265 	i_mode = file->f_path.dentry->d_inode->i_mode;
266 	ret = -ESPIPE;
267 	if (!S_ISREG(i_mode) && !S_ISBLK(i_mode) && !S_ISDIR(i_mode) &&
268 			!S_ISLNK(i_mode))
269 		goto out_put;
270 
271 	ret = do_sync_mapping_range(file->f_mapping, offset, endbyte, flags);
272 out_put:
273 	fput_light(file, fput_needed);
274 out:
275 	return ret;
276 }
277 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
278 asmlinkage long SyS_sync_file_range(long fd, loff_t offset, loff_t nbytes,
279 				    long flags)
280 {
281 	return SYSC_sync_file_range((int) fd, offset, nbytes,
282 				    (unsigned int) flags);
283 }
284 SYSCALL_ALIAS(sys_sync_file_range, SyS_sync_file_range);
285 #endif
286 
287 /* It would be nice if people remember that not all the world's an i386
288    when they introduce new system calls */
289 SYSCALL_DEFINE(sync_file_range2)(int fd, unsigned int flags,
290 				 loff_t offset, loff_t nbytes)
291 {
292 	return sys_sync_file_range(fd, offset, nbytes, flags);
293 }
294 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
295 asmlinkage long SyS_sync_file_range2(long fd, long flags,
296 				     loff_t offset, loff_t nbytes)
297 {
298 	return SYSC_sync_file_range2((int) fd, (unsigned int) flags,
299 				     offset, nbytes);
300 }
301 SYSCALL_ALIAS(sys_sync_file_range2, SyS_sync_file_range2);
302 #endif
303 
304 /*
305  * `endbyte' is inclusive
306  */
307 int do_sync_mapping_range(struct address_space *mapping, loff_t offset,
308 			  loff_t endbyte, unsigned int flags)
309 {
310 	int ret;
311 
312 	if (!mapping) {
313 		ret = -EINVAL;
314 		goto out;
315 	}
316 
317 	ret = 0;
318 	if (flags & SYNC_FILE_RANGE_WAIT_BEFORE) {
319 		ret = wait_on_page_writeback_range(mapping,
320 					offset >> PAGE_CACHE_SHIFT,
321 					endbyte >> PAGE_CACHE_SHIFT);
322 		if (ret < 0)
323 			goto out;
324 	}
325 
326 	if (flags & SYNC_FILE_RANGE_WRITE) {
327 		ret = __filemap_fdatawrite_range(mapping, offset, endbyte,
328 						WB_SYNC_ALL);
329 		if (ret < 0)
330 			goto out;
331 	}
332 
333 	if (flags & SYNC_FILE_RANGE_WAIT_AFTER) {
334 		ret = wait_on_page_writeback_range(mapping,
335 					offset >> PAGE_CACHE_SHIFT,
336 					endbyte >> PAGE_CACHE_SHIFT);
337 	}
338 out:
339 	return ret;
340 }
341 EXPORT_SYMBOL_GPL(do_sync_mapping_range);
342