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 /* 81 * Sync all the data for all the filesystems (called by sys_sync() and 82 * emergency sync) 83 * 84 * This operation is careful to avoid the livelock which could easily happen 85 * if two or more filesystems are being continuously dirtied. s_need_sync 86 * is used only here. We set it against all filesystems and then clear it as 87 * we sync them. So redirtied filesystems are skipped. 88 * 89 * But if process A is currently running sync_filesystems and then process B 90 * calls sync_filesystems as well, process B will set all the s_need_sync 91 * flags again, which will cause process A to resync everything. Fix that with 92 * a local mutex. 93 */ 94 static void sync_filesystems(int wait) 95 { 96 struct super_block *sb; 97 static DEFINE_MUTEX(mutex); 98 99 mutex_lock(&mutex); /* Could be down_interruptible */ 100 spin_lock(&sb_lock); 101 list_for_each_entry(sb, &super_blocks, s_list) 102 sb->s_need_sync = 1; 103 104 restart: 105 list_for_each_entry(sb, &super_blocks, s_list) { 106 if (!sb->s_need_sync) 107 continue; 108 sb->s_need_sync = 0; 109 sb->s_count++; 110 spin_unlock(&sb_lock); 111 112 down_read(&sb->s_umount); 113 if (!(sb->s_flags & MS_RDONLY) && sb->s_root && sb->s_bdi) 114 __sync_filesystem(sb, wait); 115 up_read(&sb->s_umount); 116 117 /* restart only when sb is no longer on the list */ 118 spin_lock(&sb_lock); 119 if (__put_super_and_need_restart(sb)) 120 goto restart; 121 } 122 spin_unlock(&sb_lock); 123 mutex_unlock(&mutex); 124 } 125 126 /* 127 * sync everything. Start out by waking pdflush, because that writes back 128 * all queues in parallel. 129 */ 130 SYSCALL_DEFINE0(sync) 131 { 132 wakeup_flusher_threads(0); 133 sync_filesystems(0); 134 sync_filesystems(1); 135 if (unlikely(laptop_mode)) 136 laptop_sync_completion(); 137 return 0; 138 } 139 140 static void do_sync_work(struct work_struct *work) 141 { 142 /* 143 * Sync twice to reduce the possibility we skipped some inodes / pages 144 * because they were temporarily locked 145 */ 146 sync_filesystems(0); 147 sync_filesystems(0); 148 printk("Emergency Sync complete\n"); 149 kfree(work); 150 } 151 152 void emergency_sync(void) 153 { 154 struct work_struct *work; 155 156 work = kmalloc(sizeof(*work), GFP_ATOMIC); 157 if (work) { 158 INIT_WORK(work, do_sync_work); 159 schedule_work(work); 160 } 161 } 162 163 /* 164 * Generic function to fsync a file. 165 * 166 * filp may be NULL if called via the msync of a vma. 167 */ 168 int file_fsync(struct file *filp, struct dentry *dentry, int datasync) 169 { 170 struct inode * inode = dentry->d_inode; 171 struct super_block * sb; 172 int ret, err; 173 174 /* sync the inode to buffers */ 175 ret = write_inode_now(inode, 0); 176 177 /* sync the superblock to buffers */ 178 sb = inode->i_sb; 179 if (sb->s_dirt && sb->s_op->write_super) 180 sb->s_op->write_super(sb); 181 182 /* .. finally sync the buffers to disk */ 183 err = sync_blockdev(sb->s_bdev); 184 if (!ret) 185 ret = err; 186 return ret; 187 } 188 EXPORT_SYMBOL(file_fsync); 189 190 /** 191 * vfs_fsync_range - helper to sync a range of data & metadata to disk 192 * @file: file to sync 193 * @dentry: dentry of @file 194 * @start: offset in bytes of the beginning of data range to sync 195 * @end: offset in bytes of the end of data range (inclusive) 196 * @datasync: perform only datasync 197 * 198 * Write back data in range @start..@end and metadata for @file to disk. If 199 * @datasync is set only metadata needed to access modified file data is 200 * written. 201 * 202 * In case this function is called from nfsd @file may be %NULL and 203 * only @dentry is set. This can only happen when the filesystem 204 * implements the export_operations API. 205 */ 206 int vfs_fsync_range(struct file *file, struct dentry *dentry, loff_t start, 207 loff_t end, int datasync) 208 { 209 const struct file_operations *fop; 210 struct address_space *mapping; 211 int err, ret; 212 213 /* 214 * Get mapping and operations from the file in case we have 215 * as file, or get the default values for them in case we 216 * don't have a struct file available. Damn nfsd.. 217 */ 218 if (file) { 219 mapping = file->f_mapping; 220 fop = file->f_op; 221 } else { 222 mapping = dentry->d_inode->i_mapping; 223 fop = dentry->d_inode->i_fop; 224 } 225 226 if (!fop || !fop->fsync) { 227 ret = -EINVAL; 228 goto out; 229 } 230 231 ret = filemap_write_and_wait_range(mapping, start, end); 232 233 /* 234 * We need to protect against concurrent writers, which could cause 235 * livelocks in fsync_buffers_list(). 236 */ 237 mutex_lock(&mapping->host->i_mutex); 238 err = fop->fsync(file, dentry, datasync); 239 if (!ret) 240 ret = err; 241 mutex_unlock(&mapping->host->i_mutex); 242 243 out: 244 return ret; 245 } 246 EXPORT_SYMBOL(vfs_fsync_range); 247 248 /** 249 * vfs_fsync - perform a fsync or fdatasync on a file 250 * @file: file to sync 251 * @dentry: dentry of @file 252 * @datasync: only perform a fdatasync operation 253 * 254 * Write back data and metadata for @file to disk. If @datasync is 255 * set only metadata needed to access modified file data is written. 256 * 257 * In case this function is called from nfsd @file may be %NULL and 258 * only @dentry is set. This can only happen when the filesystem 259 * implements the export_operations API. 260 */ 261 int vfs_fsync(struct file *file, struct dentry *dentry, int datasync) 262 { 263 return vfs_fsync_range(file, dentry, 0, LLONG_MAX, datasync); 264 } 265 EXPORT_SYMBOL(vfs_fsync); 266 267 static int do_fsync(unsigned int fd, int datasync) 268 { 269 struct file *file; 270 int ret = -EBADF; 271 272 file = fget(fd); 273 if (file) { 274 ret = vfs_fsync(file, file->f_path.dentry, datasync); 275 fput(file); 276 } 277 return ret; 278 } 279 280 SYSCALL_DEFINE1(fsync, unsigned int, fd) 281 { 282 return do_fsync(fd, 0); 283 } 284 285 SYSCALL_DEFINE1(fdatasync, unsigned int, fd) 286 { 287 return do_fsync(fd, 1); 288 } 289 290 /** 291 * generic_write_sync - perform syncing after a write if file / inode is sync 292 * @file: file to which the write happened 293 * @pos: offset where the write started 294 * @count: length of the write 295 * 296 * This is just a simple wrapper about our general syncing function. 297 */ 298 int generic_write_sync(struct file *file, loff_t pos, loff_t count) 299 { 300 if (!(file->f_flags & O_DSYNC) && !IS_SYNC(file->f_mapping->host)) 301 return 0; 302 return vfs_fsync_range(file, file->f_path.dentry, pos, 303 pos + count - 1, 304 (file->f_flags & __O_SYNC) ? 0 : 1); 305 } 306 EXPORT_SYMBOL(generic_write_sync); 307 308 /* 309 * sys_sync_file_range() permits finely controlled syncing over a segment of 310 * a file in the range offset .. (offset+nbytes-1) inclusive. If nbytes is 311 * zero then sys_sync_file_range() will operate from offset out to EOF. 312 * 313 * The flag bits are: 314 * 315 * SYNC_FILE_RANGE_WAIT_BEFORE: wait upon writeout of all pages in the range 316 * before performing the write. 317 * 318 * SYNC_FILE_RANGE_WRITE: initiate writeout of all those dirty pages in the 319 * range which are not presently under writeback. Note that this may block for 320 * significant periods due to exhaustion of disk request structures. 321 * 322 * SYNC_FILE_RANGE_WAIT_AFTER: wait upon writeout of all pages in the range 323 * after performing the write. 324 * 325 * Useful combinations of the flag bits are: 326 * 327 * SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE: ensures that all pages 328 * in the range which were dirty on entry to sys_sync_file_range() are placed 329 * under writeout. This is a start-write-for-data-integrity operation. 330 * 331 * SYNC_FILE_RANGE_WRITE: start writeout of all dirty pages in the range which 332 * are not presently under writeout. This is an asynchronous flush-to-disk 333 * operation. Not suitable for data integrity operations. 334 * 335 * SYNC_FILE_RANGE_WAIT_BEFORE (or SYNC_FILE_RANGE_WAIT_AFTER): wait for 336 * completion of writeout of all pages in the range. This will be used after an 337 * earlier SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE operation to wait 338 * for that operation to complete and to return the result. 339 * 340 * SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE|SYNC_FILE_RANGE_WAIT_AFTER: 341 * a traditional sync() operation. This is a write-for-data-integrity operation 342 * which will ensure that all pages in the range which were dirty on entry to 343 * sys_sync_file_range() are committed to disk. 344 * 345 * 346 * SYNC_FILE_RANGE_WAIT_BEFORE and SYNC_FILE_RANGE_WAIT_AFTER will detect any 347 * I/O errors or ENOSPC conditions and will return those to the caller, after 348 * clearing the EIO and ENOSPC flags in the address_space. 349 * 350 * It should be noted that none of these operations write out the file's 351 * metadata. So unless the application is strictly performing overwrites of 352 * already-instantiated disk blocks, there are no guarantees here that the data 353 * will be available after a crash. 354 */ 355 SYSCALL_DEFINE(sync_file_range)(int fd, loff_t offset, loff_t nbytes, 356 unsigned int flags) 357 { 358 int ret; 359 struct file *file; 360 struct address_space *mapping; 361 loff_t endbyte; /* inclusive */ 362 int fput_needed; 363 umode_t i_mode; 364 365 ret = -EINVAL; 366 if (flags & ~VALID_FLAGS) 367 goto out; 368 369 endbyte = offset + nbytes; 370 371 if ((s64)offset < 0) 372 goto out; 373 if ((s64)endbyte < 0) 374 goto out; 375 if (endbyte < offset) 376 goto out; 377 378 if (sizeof(pgoff_t) == 4) { 379 if (offset >= (0x100000000ULL << PAGE_CACHE_SHIFT)) { 380 /* 381 * The range starts outside a 32 bit machine's 382 * pagecache addressing capabilities. Let it "succeed" 383 */ 384 ret = 0; 385 goto out; 386 } 387 if (endbyte >= (0x100000000ULL << PAGE_CACHE_SHIFT)) { 388 /* 389 * Out to EOF 390 */ 391 nbytes = 0; 392 } 393 } 394 395 if (nbytes == 0) 396 endbyte = LLONG_MAX; 397 else 398 endbyte--; /* inclusive */ 399 400 ret = -EBADF; 401 file = fget_light(fd, &fput_needed); 402 if (!file) 403 goto out; 404 405 i_mode = file->f_path.dentry->d_inode->i_mode; 406 ret = -ESPIPE; 407 if (!S_ISREG(i_mode) && !S_ISBLK(i_mode) && !S_ISDIR(i_mode) && 408 !S_ISLNK(i_mode)) 409 goto out_put; 410 411 mapping = file->f_mapping; 412 if (!mapping) { 413 ret = -EINVAL; 414 goto out_put; 415 } 416 417 ret = 0; 418 if (flags & SYNC_FILE_RANGE_WAIT_BEFORE) { 419 ret = filemap_fdatawait_range(mapping, offset, endbyte); 420 if (ret < 0) 421 goto out_put; 422 } 423 424 if (flags & SYNC_FILE_RANGE_WRITE) { 425 ret = filemap_fdatawrite_range(mapping, offset, endbyte); 426 if (ret < 0) 427 goto out_put; 428 } 429 430 if (flags & SYNC_FILE_RANGE_WAIT_AFTER) 431 ret = filemap_fdatawait_range(mapping, offset, endbyte); 432 433 out_put: 434 fput_light(file, fput_needed); 435 out: 436 return ret; 437 } 438 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS 439 asmlinkage long SyS_sync_file_range(long fd, loff_t offset, loff_t nbytes, 440 long flags) 441 { 442 return SYSC_sync_file_range((int) fd, offset, nbytes, 443 (unsigned int) flags); 444 } 445 SYSCALL_ALIAS(sys_sync_file_range, SyS_sync_file_range); 446 #endif 447 448 /* It would be nice if people remember that not all the world's an i386 449 when they introduce new system calls */ 450 SYSCALL_DEFINE(sync_file_range2)(int fd, unsigned int flags, 451 loff_t offset, loff_t nbytes) 452 { 453 return sys_sync_file_range(fd, offset, nbytes, flags); 454 } 455 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS 456 asmlinkage long SyS_sync_file_range2(long fd, long flags, 457 loff_t offset, loff_t nbytes) 458 { 459 return SYSC_sync_file_range2((int) fd, (unsigned int) flags, 460 offset, nbytes); 461 } 462 SYSCALL_ALIAS(sys_sync_file_range2, SyS_sync_file_range2); 463 #endif 464