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