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/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 #include "internal.h"
  17
  18 #define VALID_FLAGS (SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE| \
  19                         SYNC_FILE_RANGE_WAIT_AFTER)
  20
  21 /*
  22  * Do the filesystem syncing work. For simple filesystems
  23  * writeback_inodes_sb(sb) just dirties buffers with inodes so we have to
  24  * submit IO for these buffers via __sync_blockdev(). This also speeds up the
  25  * wait == 1 case since in that case write_inode() functions do
  26  * sync_dirty_buffer() and thus effectively write one block at a time.
  27  */
  28 static int __sync_filesystem(struct super_block *sb, int wait)
  29 {
  30         /* Avoid doing twice syncing and cache pruning for quota sync */
  31         if (!wait) {
  32                 writeout_quota_sb(sb, -1);
  33                 writeback_inodes_sb(sb);
  34         } else {
  35                 sync_quota_sb(sb, -1);
  36                 sync_inodes_sb(sb);
  37         }
  38         if (sb->s_op->sync_fs)
  39                 sb->s_op->sync_fs(sb, wait);
  40         return __sync_blockdev(sb->s_bdev, wait);
  41 }
  42
  43 /*
  44  * Write out and wait upon all dirty data associated with this
  45  * superblock.  Filesystem data as well as the underlying block
  46  * device.  Takes the superblock lock.
  47  */
  48 int sync_filesystem(struct super_block *sb)
  49 {
  50         int ret;
  51
  52         /*
  53          * We need to be protected against the filesystem going from
  54          * r/o to r/w or vice versa.
  55          */
  56         WARN_ON(!rwsem_is_locked(&sb->s_umount));
  57
  58         /*
  59          * No point in syncing out anything if the filesystem is read-only.
  60          */
  61         if (sb->s_flags & MS_RDONLY)
  62                 return 0;
  63
  64         ret = __sync_filesystem(sb, 0);
  65         if (ret < 0)
  66                 return ret;
  67         return __sync_filesystem(sb, 1);
  68 }
  69 EXPORT_SYMBOL_GPL(sync_filesystem);
  70
  71 /*
  72  * Sync all the data for all the filesystems (called by sys_sync() and
  73  * emergency sync)
  74  *
  75  * This operation is careful to avoid the livelock which could easily happen
  76  * if two or more filesystems are being continuously dirtied.  s_need_sync
  77  * is used only here.  We set it against all filesystems and then clear it as
  78  * we sync them.  So redirtied filesystems are skipped.
  79  *
  80  * But if process A is currently running sync_filesystems and then process B
  81  * calls sync_filesystems as well, process B will set all the s_need_sync
  82  * flags again, which will cause process A to resync everything.  Fix that with
  83  * a local mutex.
  84  */
  85 static void sync_filesystems(int wait)
  86 {
  87         struct super_block *sb;
  88         static DEFINE_MUTEX(mutex);
  89
  90         mutex_lock(&mutex);             /* Could be down_interruptible */
  91         spin_lock(&sb_lock);
  92         list_for_each_entry(sb, &super_blocks, s_list)
  93                 sb->s_need_sync = 1;
  94
  95 restart:
  96         list_for_each_entry(sb, &super_blocks, s_list) {
  97                 if (!sb->s_need_sync)
  98                         continue;
  99                 sb->s_need_sync = 0;
 100                 sb->s_count++;
 101                 spin_unlock(&sb_lock);
 102
 103                 down_read(&sb->s_umount);
 104                 if (!(sb->s_flags & MS_RDONLY) && sb->s_root)
 105                         __sync_filesystem(sb, wait);
 106                 up_read(&sb->s_umount);
 107
 108                 /* restart only when sb is no longer on the list */
 109                 spin_lock(&sb_lock);
 110                 if (__put_super_and_need_restart(sb))
 111                         goto restart;
 112         }
 113         spin_unlock(&sb_lock);
 114         mutex_unlock(&mutex);
 115 }
 116
 117 /*
 118  * sync everything.  Start out by waking pdflush, because that writes back
 119  * all queues in parallel.
 120  */
 121 SYSCALL_DEFINE0(sync)
 122 {
 123         wakeup_flusher_threads(0);
 124         sync_filesystems(0);
 125         sync_filesystems(1);
 126         if (unlikely(laptop_mode))
 127                 laptop_sync_completion();
 128         return 0;
 129 }
 130
 131 static void do_sync_work(struct work_struct *work)
 132 {
 133         /*
 134          * Sync twice to reduce the possibility we skipped some inodes / pages
 135          * because they were temporarily locked
 136          */
 137         sync_filesystems(0);
 138         sync_filesystems(0);
 139         printk("Emergency Sync complete\n");
 140         kfree(work);
 141 }
 142
 143 void emergency_sync(void)
 144 {
 145         struct work_struct *work;
 146
 147         work = kmalloc(sizeof(*work), GFP_ATOMIC);
 148         if (work) {
 149                 INIT_WORK(work, do_sync_work);
 150                 schedule_work(work);
 151         }
 152 }
 153
 154 /*
 155  * Generic function to fsync a file.
 156  *
 157  * filp may be NULL if called via the msync of a vma.
 158  */
 159 int file_fsync(struct file *filp, struct dentry *dentry, int datasync)
 160 {
 161         struct inode * inode = dentry->d_inode;
 162         struct super_block * sb;
 163         int ret, err;
 164
 165         /* sync the inode to buffers */
 166         ret = write_inode_now(inode, 0);
 167
 168         /* sync the superblock to buffers */
 169         sb = inode->i_sb;
 170         if (sb->s_dirt && sb->s_op->write_super)
 171                 sb->s_op->write_super(sb);
 172
 173         /* .. finally sync the buffers to disk */
 174         err = sync_blockdev(sb->s_bdev);
 175         if (!ret)
 176                 ret = err;
 177         return ret;
 178 }
 179
 180 /**
 181  * vfs_fsync_range - helper to sync a range of data & metadata to disk
 182  * @file:               file to sync
 183  * @dentry:             dentry of @file
 184  * @start:              offset in bytes of the beginning of data range to sync
 185  * @end:                offset in bytes of the end of data range (inclusive)
 186  * @datasync:           perform only datasync
 187  *
 188  * Write back data in range @start..@end and metadata for @file to disk.  If
 189  * @datasync is set only metadata needed to access modified file data is
 190  * written.
 191  *
 192  * In case this function is called from nfsd @file may be %NULL and
 193  * only @dentry is set.  This can only happen when the filesystem
 194  * implements the export_operations API.
 195  */
 196 int vfs_fsync_range(struct file *file, struct dentry *dentry, loff_t start,
 197                     loff_t end, int datasync)
 198 {
 199         const struct file_operations *fop;
 200         struct address_space *mapping;
 201         int err, ret;
 202
 203         /*
 204          * Get mapping and operations from the file in case we have
 205          * as file, or get the default values for them in case we
 206          * don't have a struct file available.  Damn nfsd..
 207          */
 208         if (file) {
 209                 mapping = file->f_mapping;
 210                 fop = file->f_op;
 211         } else {
 212                 mapping = dentry->d_inode->i_mapping;
 213                 fop = dentry->d_inode->i_fop;
 214         }
 215
 216         if (!fop || !fop->fsync) {
 217                 ret = -EINVAL;
 218                 goto out;
 219         }
 220
 221         ret = filemap_write_and_wait_range(mapping, start, end);
 222
 223         /*
 224          * We need to protect against concurrent writers, which could cause
 225          * livelocks in fsync_buffers_list().
 226          */
 227         mutex_lock(&mapping->host->i_mutex);
 228         err = fop->fsync(file, dentry, datasync);
 229         if (!ret)
 230                 ret = err;
 231         mutex_unlock(&mapping->host->i_mutex);
 232
 233 out:
 234         return ret;
 235 }
 236 EXPORT_SYMBOL(vfs_fsync_range);
 237
 238 /**
 239  * vfs_fsync - perform a fsync or fdatasync on a file
 240  * @file:               file to sync
 241  * @dentry:             dentry of @file
 242  * @datasync:           only perform a fdatasync operation
 243  *
 244  * Write back data and metadata for @file to disk.  If @datasync is
 245  * set only metadata needed to access modified file data is written.
 246  *
 247  * In case this function is called from nfsd @file may be %NULL and
 248  * only @dentry is set.  This can only happen when the filesystem
 249  * implements the export_operations API.
 250  */
 251 int vfs_fsync(struct file *file, struct dentry *dentry, int datasync)
 252 {
 253         return vfs_fsync_range(file, dentry, 0, LLONG_MAX, datasync);
 254 }
 255 EXPORT_SYMBOL(vfs_fsync);
 256
 257 static int do_fsync(unsigned int fd, int datasync)
 258 {
 259         struct file *file;
 260         int ret = -EBADF;
 261
 262         file = fget(fd);
 263         if (file) {
 264                 ret = vfs_fsync(file, file->f_path.dentry, datasync);
 265                 fput(file);
 266         }
 267         return ret;
 268 }
 269
 270 SYSCALL_DEFINE1(fsync, unsigned int, fd)
 271 {
 272         return do_fsync(fd, 0);
 273 }
 274
 275 SYSCALL_DEFINE1(fdatasync, unsigned int, fd)
 276 {
 277         return do_fsync(fd, 1);
 278 }
 279
 280 /**
 281  * generic_write_sync - perform syncing after a write if file / inode is sync
 282  * @file:       file to which the write happened
 283  * @pos:        offset where the write started
 284  * @count:      length of the write
 285  *
 286  * This is just a simple wrapper about our general syncing function.
 287  */
 288 int generic_write_sync(struct file *file, loff_t pos, loff_t count)
 289 {
 290         if (!(file->f_flags & O_SYNC) && !IS_SYNC(file->f_mapping->host))
 291                 return 0;
 292         return vfs_fsync_range(file, file->f_path.dentry, pos,
 293                                pos + count - 1, 1);
 294 }
 295 EXPORT_SYMBOL(generic_write_sync);
 296
 297 /*
 298  * sys_sync_file_range() permits finely controlled syncing over a segment of
 299  * a file in the range offset .. (offset+nbytes-1) inclusive.  If nbytes is
 300  * zero then sys_sync_file_range() will operate from offset out to EOF.
 301  *
 302  * The flag bits are:
 303  *
 304  * SYNC_FILE_RANGE_WAIT_BEFORE: wait upon writeout of all pages in the range
 305  * before performing the write.
 306  *
 307  * SYNC_FILE_RANGE_WRITE: initiate writeout of all those dirty pages in the
 308  * range which are not presently under writeback. Note that this may block for
 309  * significant periods due to exhaustion of disk request structures.
 310  *
 311  * SYNC_FILE_RANGE_WAIT_AFTER: wait upon writeout of all pages in the range
 312  * after performing the write.
 313  *
 314  * Useful combinations of the flag bits are:
 315  *
 316  * SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE: ensures that all pages
 317  * in the range which were dirty on entry to sys_sync_file_range() are placed
 318  * under writeout.  This is a start-write-for-data-integrity operation.
 319  *
 320  * SYNC_FILE_RANGE_WRITE: start writeout of all dirty pages in the range which
 321  * are not presently under writeout.  This is an asynchronous flush-to-disk
 322  * operation.  Not suitable for data integrity operations.
 323  *
 324  * SYNC_FILE_RANGE_WAIT_BEFORE (or SYNC_FILE_RANGE_WAIT_AFTER): wait for
 325  * completion of writeout of all pages in the range.  This will be used after an
 326  * earlier SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE operation to wait
 327  * for that operation to complete and to return the result.
 328  *
 329  * SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE|SYNC_FILE_RANGE_WAIT_AFTER:
 330  * a traditional sync() operation.  This is a write-for-data-integrity operation
 331  * which will ensure that all pages in the range which were dirty on entry to
 332  * sys_sync_file_range() are committed to disk.
 333  *
 334  *
 335  * SYNC_FILE_RANGE_WAIT_BEFORE and SYNC_FILE_RANGE_WAIT_AFTER will detect any
 336  * I/O errors or ENOSPC conditions and will return those to the caller, after
 337  * clearing the EIO and ENOSPC flags in the address_space.
 338  *
 339  * It should be noted that none of these operations write out the file's
 340  * metadata.  So unless the application is strictly performing overwrites of
 341  * already-instantiated disk blocks, there are no guarantees here that the data
 342  * will be available after a crash.
 343  */
 344 SYSCALL_DEFINE(sync_file_range)(int fd, loff_t offset, loff_t nbytes,
 345                                 unsigned int flags)
 346 {
 347         int ret;
 348         struct file *file;
 349         loff_t endbyte;                 /* inclusive */
 350         int fput_needed;
 351         umode_t i_mode;
 352
 353         ret = -EINVAL;
 354         if (flags & ~VALID_FLAGS)
 355                 goto out;
 356
 357         endbyte = offset + nbytes;
 358
 359         if ((s64)offset < 0)
 360                 goto out;
 361         if ((s64)endbyte < 0)
 362                 goto out;
 363         if (endbyte < offset)
 364                 goto out;
 365
 366         if (sizeof(pgoff_t) == 4) {
 367                 if (offset >= (0x100000000ULL << PAGE_CACHE_SHIFT)) {
 368                         /*
 369                          * The range starts outside a 32 bit machine's
 370                          * pagecache addressing capabilities.  Let it "succeed"
 371                          */
 372                         ret = 0;
 373                         goto out;
 374                 }
 375                 if (endbyte >= (0x100000000ULL << PAGE_CACHE_SHIFT)) {
 376                         /*
 377                          * Out to EOF
 378                          */
 379                         nbytes = 0;
 380                 }
 381         }
 382
 383         if (nbytes == 0)
 384                 endbyte = LLONG_MAX;
 385         else
 386                 endbyte--;              /* inclusive */
 387
 388         ret = -EBADF;
 389         file = fget_light(fd, &fput_needed);
 390         if (!file)
 391                 goto out;
 392
 393         i_mode = file->f_path.dentry->d_inode->i_mode;
 394         ret = -ESPIPE;
 395         if (!S_ISREG(i_mode) && !S_ISBLK(i_mode) && !S_ISDIR(i_mode) &&
 396                         !S_ISLNK(i_mode))
 397                 goto out_put;
 398
 399         ret = do_sync_mapping_range(file->f_mapping, offset, endbyte, flags);
 400 out_put:
 401         fput_light(file, fput_needed);
 402 out:
 403         return ret;
 404 }
 405 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
 406 asmlinkage long SyS_sync_file_range(long fd, loff_t offset, loff_t nbytes,
 407                                     long flags)
 408 {
 409         return SYSC_sync_file_range((int) fd, offset, nbytes,
 410                                     (unsigned int) flags);
 411 }
 412 SYSCALL_ALIAS(sys_sync_file_range, SyS_sync_file_range);
 413 #endif
 414
 415 /* It would be nice if people remember that not all the world's an i386
 416    when they introduce new system calls */
 417 SYSCALL_DEFINE(sync_file_range2)(int fd, unsigned int flags,
 418                                  loff_t offset, loff_t nbytes)
 419 {
 420         return sys_sync_file_range(fd, offset, nbytes, flags);
 421 }
 422 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
 423 asmlinkage long SyS_sync_file_range2(long fd, long flags,
 424                                      loff_t offset, loff_t nbytes)
 425 {
 426         return SYSC_sync_file_range2((int) fd, (unsigned int) flags,
 427                                      offset, nbytes);
 428 }
 429 SYSCALL_ALIAS(sys_sync_file_range2, SyS_sync_file_range2);
 430 #endif
 431
 432 /*
 433  * `endbyte' is inclusive
 434  */
 435 int do_sync_mapping_range(struct address_space *mapping, loff_t offset,
 436                           loff_t endbyte, unsigned int flags)
 437 {
 438         int ret;
 439
 440         if (!mapping) {
 441                 ret = -EINVAL;
 442                 goto out;
 443         }
 444
 445         ret = 0;
 446         if (flags & SYNC_FILE_RANGE_WAIT_BEFORE) {
 447                 ret = wait_on_page_writeback_range(mapping,
 448                                         offset >> PAGE_CACHE_SHIFT,
 449                                         endbyte >> PAGE_CACHE_SHIFT);
 450                 if (ret < 0)
 451                         goto out;
 452         }
 453
 454         if (flags & SYNC_FILE_RANGE_WRITE) {
 455                 ret = __filemap_fdatawrite_range(mapping, offset, endbyte,
 456                                                 WB_SYNC_ALL);
 457                 if (ret < 0)
 458                         goto out;
 459         }
 460
 461         if (flags & SYNC_FILE_RANGE_WAIT_AFTER) {
 462                 ret = wait_on_page_writeback_range(mapping,
 463                                         offset >> PAGE_CACHE_SHIFT,
 464                                         endbyte >> PAGE_CACHE_SHIFT);
 465         }
 466 out:
 467         return ret;
 468 }
 469 EXPORT_SYMBOL_GPL(do_sync_mapping_range);