2 * High-level sync()-related operations
5 #include <linux/kernel.h>
6 #include <linux/file.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>
18 #define VALID_FLAGS (SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE| \
19 SYNC_FILE_RANGE_WAIT_AFTER)
22 * sync everything. Start out by waking pdflush, because that writes back
23 * all queues in parallel.
25 static void do_sync(unsigned long wait
)
28 sync_inodes(0); /* All mappings, inodes and their blockdevs */
30 sync_inodes(wait
); /* Mappings, inodes and blockdevs, again. */
31 sync_supers(); /* Write the superblocks */
32 sync_filesystems(0); /* Start syncing the filesystems */
33 sync_filesystems(wait
); /* Waitingly sync the filesystems */
36 printk("Emergency Sync complete\n");
37 if (unlikely(laptop_mode
))
38 laptop_sync_completion();
47 static void do_sync_work(struct work_struct
*work
)
53 void emergency_sync(void)
55 struct work_struct
*work
;
57 work
= kmalloc(sizeof(*work
), GFP_ATOMIC
);
59 INIT_WORK(work
, do_sync_work
);
65 * Generic function to fsync a file.
67 * filp may be NULL if called via the msync of a vma.
69 int file_fsync(struct file
*filp
, struct dentry
*dentry
, int datasync
)
71 struct inode
* inode
= dentry
->d_inode
;
72 struct super_block
* sb
;
75 /* sync the inode to buffers */
76 ret
= write_inode_now(inode
, 0);
78 /* sync the superblock to buffers */
81 if (sb
->s_dirt
&& sb
->s_op
->write_super
)
82 sb
->s_op
->write_super(sb
);
85 /* .. finally sync the buffers to disk */
86 err
= sync_blockdev(sb
->s_bdev
);
93 * vfs_fsync - perform a fsync or fdatasync on a file
95 * @dentry: dentry of @file
96 * @data: only perform a fdatasync operation
98 * Write back data and metadata for @file to disk. If @datasync is
99 * set only metadata needed to access modified file data is written.
101 * In case this function is called from nfsd @file may be %NULL and
102 * only @dentry is set. This can only happen when the filesystem
103 * implements the export_operations API.
105 int vfs_fsync(struct file
*file
, struct dentry
*dentry
, int datasync
)
107 const struct file_operations
*fop
;
108 struct address_space
*mapping
;
112 * Get mapping and operations from the file in case we have
113 * as file, or get the default values for them in case we
114 * don't have a struct file available. Damn nfsd..
117 mapping
= file
->f_mapping
;
120 mapping
= dentry
->d_inode
->i_mapping
;
121 fop
= dentry
->d_inode
->i_fop
;
124 if (!fop
|| !fop
->fsync
) {
129 ret
= filemap_fdatawrite(mapping
);
132 * We need to protect against concurrent writers, which could cause
133 * livelocks in fsync_buffers_list().
135 mutex_lock(&mapping
->host
->i_mutex
);
136 err
= fop
->fsync(file
, dentry
, datasync
);
139 mutex_unlock(&mapping
->host
->i_mutex
);
140 err
= filemap_fdatawait(mapping
);
146 EXPORT_SYMBOL(vfs_fsync
);
148 static int do_fsync(unsigned int fd
, int datasync
)
155 ret
= vfs_fsync(file
, file
->f_path
.dentry
, datasync
);
161 SYSCALL_DEFINE1(fsync
, unsigned int, fd
)
163 return do_fsync(fd
, 0);
166 SYSCALL_DEFINE1(fdatasync
, unsigned int, fd
)
168 return do_fsync(fd
, 1);
172 * sys_sync_file_range() permits finely controlled syncing over a segment of
173 * a file in the range offset .. (offset+nbytes-1) inclusive. If nbytes is
174 * zero then sys_sync_file_range() will operate from offset out to EOF.
178 * SYNC_FILE_RANGE_WAIT_BEFORE: wait upon writeout of all pages in the range
179 * before performing the write.
181 * SYNC_FILE_RANGE_WRITE: initiate writeout of all those dirty pages in the
182 * range which are not presently under writeback. Note that this may block for
183 * significant periods due to exhaustion of disk request structures.
185 * SYNC_FILE_RANGE_WAIT_AFTER: wait upon writeout of all pages in the range
186 * after performing the write.
188 * Useful combinations of the flag bits are:
190 * SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE: ensures that all pages
191 * in the range which were dirty on entry to sys_sync_file_range() are placed
192 * under writeout. This is a start-write-for-data-integrity operation.
194 * SYNC_FILE_RANGE_WRITE: start writeout of all dirty pages in the range which
195 * are not presently under writeout. This is an asynchronous flush-to-disk
196 * operation. Not suitable for data integrity operations.
198 * SYNC_FILE_RANGE_WAIT_BEFORE (or SYNC_FILE_RANGE_WAIT_AFTER): wait for
199 * completion of writeout of all pages in the range. This will be used after an
200 * earlier SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE operation to wait
201 * for that operation to complete and to return the result.
203 * SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE|SYNC_FILE_RANGE_WAIT_AFTER:
204 * a traditional sync() operation. This is a write-for-data-integrity operation
205 * which will ensure that all pages in the range which were dirty on entry to
206 * sys_sync_file_range() are committed to disk.
209 * SYNC_FILE_RANGE_WAIT_BEFORE and SYNC_FILE_RANGE_WAIT_AFTER will detect any
210 * I/O errors or ENOSPC conditions and will return those to the caller, after
211 * clearing the EIO and ENOSPC flags in the address_space.
213 * It should be noted that none of these operations write out the file's
214 * metadata. So unless the application is strictly performing overwrites of
215 * already-instantiated disk blocks, there are no guarantees here that the data
216 * will be available after a crash.
218 SYSCALL_DEFINE(sync_file_range
)(int fd
, loff_t offset
, loff_t nbytes
,
223 loff_t endbyte
; /* inclusive */
228 if (flags
& ~VALID_FLAGS
)
231 endbyte
= offset
+ nbytes
;
235 if ((s64
)endbyte
< 0)
237 if (endbyte
< offset
)
240 if (sizeof(pgoff_t
) == 4) {
241 if (offset
>= (0x100000000ULL
<< PAGE_CACHE_SHIFT
)) {
243 * The range starts outside a 32 bit machine's
244 * pagecache addressing capabilities. Let it "succeed"
249 if (endbyte
>= (0x100000000ULL
<< PAGE_CACHE_SHIFT
)) {
260 endbyte
--; /* inclusive */
263 file
= fget_light(fd
, &fput_needed
);
267 i_mode
= file
->f_path
.dentry
->d_inode
->i_mode
;
269 if (!S_ISREG(i_mode
) && !S_ISBLK(i_mode
) && !S_ISDIR(i_mode
) &&
273 ret
= do_sync_mapping_range(file
->f_mapping
, offset
, endbyte
, flags
);
275 fput_light(file
, fput_needed
);
279 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
280 asmlinkage
long SyS_sync_file_range(long fd
, loff_t offset
, loff_t nbytes
,
283 return SYSC_sync_file_range((int) fd
, offset
, nbytes
,
284 (unsigned int) flags
);
286 SYSCALL_ALIAS(sys_sync_file_range
, SyS_sync_file_range
);
289 /* It would be nice if people remember that not all the world's an i386
290 when they introduce new system calls */
291 SYSCALL_DEFINE(sync_file_range2
)(int fd
, unsigned int flags
,
292 loff_t offset
, loff_t nbytes
)
294 return sys_sync_file_range(fd
, offset
, nbytes
, flags
);
296 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
297 asmlinkage
long SyS_sync_file_range2(long fd
, long flags
,
298 loff_t offset
, loff_t nbytes
)
300 return SYSC_sync_file_range2((int) fd
, (unsigned int) flags
,
303 SYSCALL_ALIAS(sys_sync_file_range2
, SyS_sync_file_range2
);
307 * `endbyte' is inclusive
309 int do_sync_mapping_range(struct address_space
*mapping
, loff_t offset
,
310 loff_t endbyte
, unsigned int flags
)
320 if (flags
& SYNC_FILE_RANGE_WAIT_BEFORE
) {
321 ret
= wait_on_page_writeback_range(mapping
,
322 offset
>> PAGE_CACHE_SHIFT
,
323 endbyte
>> PAGE_CACHE_SHIFT
);
328 if (flags
& SYNC_FILE_RANGE_WRITE
) {
329 ret
= __filemap_fdatawrite_range(mapping
, offset
, endbyte
,
335 if (flags
& SYNC_FILE_RANGE_WAIT_AFTER
) {
336 ret
= wait_on_page_writeback_range(mapping
,
337 offset
>> PAGE_CACHE_SHIFT
,
338 endbyte
>> PAGE_CACHE_SHIFT
);
343 EXPORT_SYMBOL_GPL(do_sync_mapping_range
);