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>
17 #define VALID_FLAGS (SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE| \
18 SYNC_FILE_RANGE_WAIT_AFTER)
21 * sync everything. Start out by waking pdflush, because that writes back
22 * all queues in parallel.
24 static void do_sync(unsigned long wait)
27 sync_inodes(0); /* All mappings, inodes and their blockdevs */
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. */
34 printk("Emergency Sync complete\n");
35 if (unlikely(laptop_mode))
36 laptop_sync_completion();
39 asmlinkage long sys_sync(void)
45 void emergency_sync(void)
47 pdflush_operation(do_sync, 0);
51 * Generic function to fsync a file.
53 * filp may be NULL if called via the msync of a vma.
55 int file_fsync(struct file *filp, struct dentry *dentry, int datasync)
57 struct inode * inode = dentry->d_inode;
58 struct super_block * sb;
61 /* sync the inode to buffers */
62 ret = write_inode_now(inode, 0);
64 /* sync the superblock to buffers */
67 if (sb->s_dirt && sb->s_op->write_super)
68 sb->s_op->write_super(sb);
71 /* .. finally sync the buffers to disk */
72 err = sync_blockdev(sb->s_bdev);
79 * vfs_fsync - perform a fsync or fdatasync on a file
81 * @dentry: dentry of @file
82 * @data: only perform a fdatasync operation
84 * Write back data and metadata for @file to disk. If @datasync is
85 * set only metadata needed to access modified file data is written.
87 * In case this function is called from nfsd @file may be %NULL and
88 * only @dentry is set. This can only happen when the filesystem
89 * implements the export_operations API.
91 int vfs_fsync(struct file *file, struct dentry *dentry, int datasync)
93 const struct file_operations *fop;
94 struct address_space *mapping;
98 * Get mapping and operations from the file in case we have
99 * as file, or get the default values for them in case we
100 * don't have a struct file available. Damn nfsd..
103 mapping = file->f_mapping;
106 mapping = dentry->d_inode->i_mapping;
107 fop = dentry->d_inode->i_fop;
110 if (!fop || !fop->fsync) {
115 ret = filemap_fdatawrite(mapping);
118 * We need to protect against concurrent writers, which could cause
119 * livelocks in fsync_buffers_list().
121 mutex_lock(&mapping->host->i_mutex);
122 err = fop->fsync(file, dentry, datasync);
125 mutex_unlock(&mapping->host->i_mutex);
126 err = filemap_fdatawait(mapping);
132 EXPORT_SYMBOL(vfs_fsync);
134 static int do_fsync(unsigned int fd, int datasync)
141 ret = vfs_fsync(file, file->f_path.dentry, datasync);
147 asmlinkage long sys_fsync(unsigned int fd)
149 return do_fsync(fd, 0);
152 asmlinkage long sys_fdatasync(unsigned int fd)
154 return do_fsync(fd, 1);
158 * sys_sync_file_range() permits finely controlled syncing over a segment of
159 * a file in the range offset .. (offset+nbytes-1) inclusive. If nbytes is
160 * zero then sys_sync_file_range() will operate from offset out to EOF.
164 * SYNC_FILE_RANGE_WAIT_BEFORE: wait upon writeout of all pages in the range
165 * before performing the write.
167 * SYNC_FILE_RANGE_WRITE: initiate writeout of all those dirty pages in the
168 * range which are not presently under writeback. Note that this may block for
169 * significant periods due to exhaustion of disk request structures.
171 * SYNC_FILE_RANGE_WAIT_AFTER: wait upon writeout of all pages in the range
172 * after performing the write.
174 * Useful combinations of the flag bits are:
176 * SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE: ensures that all pages
177 * in the range which were dirty on entry to sys_sync_file_range() are placed
178 * under writeout. This is a start-write-for-data-integrity operation.
180 * SYNC_FILE_RANGE_WRITE: start writeout of all dirty pages in the range which
181 * are not presently under writeout. This is an asynchronous flush-to-disk
182 * operation. Not suitable for data integrity operations.
184 * SYNC_FILE_RANGE_WAIT_BEFORE (or SYNC_FILE_RANGE_WAIT_AFTER): wait for
185 * completion of writeout of all pages in the range. This will be used after an
186 * earlier SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE operation to wait
187 * for that operation to complete and to return the result.
189 * SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE|SYNC_FILE_RANGE_WAIT_AFTER:
190 * a traditional sync() operation. This is a write-for-data-integrity operation
191 * which will ensure that all pages in the range which were dirty on entry to
192 * sys_sync_file_range() are committed to disk.
195 * SYNC_FILE_RANGE_WAIT_BEFORE and SYNC_FILE_RANGE_WAIT_AFTER will detect any
196 * I/O errors or ENOSPC conditions and will return those to the caller, after
197 * clearing the EIO and ENOSPC flags in the address_space.
199 * It should be noted that none of these operations write out the file's
200 * metadata. So unless the application is strictly performing overwrites of
201 * already-instantiated disk blocks, there are no guarantees here that the data
202 * will be available after a crash.
204 SYSCALL_DEFINE(sync_file_range)(int fd, loff_t offset, loff_t nbytes,
209 loff_t endbyte; /* inclusive */
214 if (flags & ~VALID_FLAGS)
217 endbyte = offset + nbytes;
221 if ((s64)endbyte < 0)
223 if (endbyte < offset)
226 if (sizeof(pgoff_t) == 4) {
227 if (offset >= (0x100000000ULL << PAGE_CACHE_SHIFT)) {
229 * The range starts outside a 32 bit machine's
230 * pagecache addressing capabilities. Let it "succeed"
235 if (endbyte >= (0x100000000ULL << PAGE_CACHE_SHIFT)) {
246 endbyte--; /* inclusive */
249 file = fget_light(fd, &fput_needed);
253 i_mode = file->f_path.dentry->d_inode->i_mode;
255 if (!S_ISREG(i_mode) && !S_ISBLK(i_mode) && !S_ISDIR(i_mode) &&
259 ret = do_sync_mapping_range(file->f_mapping, offset, endbyte, flags);
261 fput_light(file, fput_needed);
265 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
266 asmlinkage long SyS_sync_file_range(long fd, loff_t offset, loff_t nbytes,
269 return SYSC_sync_file_range((int) fd, offset, nbytes,
270 (unsigned int) flags);
272 SYSCALL_ALIAS(sys_sync_file_range, SyS_sync_file_range);
275 /* It would be nice if people remember that not all the world's an i386
276 when they introduce new system calls */
277 SYSCALL_DEFINE(sync_file_range2)(int fd, unsigned int flags,
278 loff_t offset, loff_t nbytes)
280 return sys_sync_file_range(fd, offset, nbytes, flags);
282 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
283 asmlinkage long SyS_sync_file_range2(long fd, long flags,
284 loff_t offset, loff_t nbytes)
286 return SYSC_sync_file_range2((int) fd, (unsigned int) flags,
289 SYSCALL_ALIAS(sys_sync_file_range2, SyS_sync_file_range2);
293 * `endbyte' is inclusive
295 int do_sync_mapping_range(struct address_space *mapping, loff_t offset,
296 loff_t endbyte, unsigned int flags)
306 if (flags & SYNC_FILE_RANGE_WAIT_BEFORE) {
307 ret = wait_on_page_writeback_range(mapping,
308 offset >> PAGE_CACHE_SHIFT,
309 endbyte >> PAGE_CACHE_SHIFT);
314 if (flags & SYNC_FILE_RANGE_WRITE) {
315 ret = __filemap_fdatawrite_range(mapping, offset, endbyte,
321 if (flags & SYNC_FILE_RANGE_WAIT_AFTER) {
322 ret = wait_on_page_writeback_range(mapping,
323 offset >> PAGE_CACHE_SHIFT,
324 endbyte >> PAGE_CACHE_SHIFT);
329 EXPORT_SYMBOL_GPL(do_sync_mapping_range);