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