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         /*
  31          * This should be safe, as we require bdi backing to actually
  32          * write out data in the first place
  33          */
  34         if (!sb->s_bdi)
  35                 return 0;
  36
  37         /* Avoid doing twice syncing and cache pruning for quota sync */
  38         if (!wait) {
  39                 writeout_quota_sb(sb, -1);
  40                 writeback_inodes_sb(sb);
  41         } else {
  42                 sync_quota_sb(sb, -1);
  43                 sync_inodes_sb(sb);
  44         }
  45         if (sb->s_op->sync_fs)
  46                 sb->s_op->sync_fs(sb, wait);
  47         return __sync_blockdev(sb->s_bdev, wait);
  48 }
  49
  50 /*
  51  * Write out and wait upon all dirty data associated with this
  52  * superblock.  Filesystem data as well as the underlying block
  53  * device.  Takes the superblock lock.
  54  */
  55 int sync_filesystem(struct super_block *sb)
  56 {
  57         int ret;
  58
  59         /*
  60          * We need to be protected against the filesystem going from
  61          * r/o to r/w or vice versa.
  62          */
  63         WARN_ON(!rwsem_is_locked(&sb->s_umount));
  64
  65         /*
  66          * No point in syncing out anything if the filesystem is read-only.
  67          */
  68         if (sb->s_flags & MS_RDONLY)
  69                 return 0;
  70
  71         ret = __sync_filesystem(sb, 0);
  72         if (ret < 0)
  73                 return ret;
  74         return __sync_filesystem(sb, 1);
  75 }
  76 EXPORT_SYMBOL_GPL(sync_filesystem);
  77
  78 /*
  79  * Sync all the data for all the filesystems (called by sys_sync() and
  80  * emergency sync)
  81  *
  82  * This operation is careful to avoid the livelock which could easily happen
  83  * if two or more filesystems are being continuously dirtied.  s_need_sync
  84  * is used only here.  We set it against all filesystems and then clear it as
  85  * we sync them.  So redirtied filesystems are skipped.
  86  *
  87  * But if process A is currently running sync_filesystems and then process B
  88  * calls sync_filesystems as well, process B will set all the s_need_sync
  89  * flags again, which will cause process A to resync everything.  Fix that with
  90  * a local mutex.
  91  */
  92 static void sync_filesystems(int wait)
  93 {
  94         struct super_block *sb;
  95         static DEFINE_MUTEX(mutex);
  96
  97         mutex_lock(&mutex);             /* Could be down_interruptible */
  98         spin_lock(&sb_lock);
  99         list_for_each_entry(sb, &super_blocks, s_list)
 100                 sb->s_need_sync = 1;
 101
 102 restart:
 103         list_for_each_entry(sb, &super_blocks, s_list) {
 104                 if (!sb->s_need_sync)
 105                         continue;
 106                 sb->s_need_sync = 0;
 107                 sb->s_count++;
 108                 spin_unlock(&sb_lock);
 109
 110                 down_read(&sb->s_umount);
 111                 if (!(sb->s_flags & MS_RDONLY) && sb->s_root && sb->s_bdi)
 112                         __sync_filesystem(sb, wait);
 113                 up_read(&sb->s_umount);
 114
 115                 /* restart only when sb is no longer on the list */
 116                 spin_lock(&sb_lock);
 117                 if (__put_super_and_need_restart(sb))
 118                         goto restart;
 119         }
 120         spin_unlock(&sb_lock);
 121         mutex_unlock(&mutex);
 122 }
 123
 124 /*
 125  * sync everything.  Start out by waking pdflush, because that writes back
 126  * all queues in parallel.
 127  */
 128 SYSCALL_DEFINE0(sync)
 129 {
 130         wakeup_flusher_threads(0);
 131         sync_filesystems(0);
 132         sync_filesystems(1);
 133         if (unlikely(laptop_mode))
 134                 laptop_sync_completion();
 135         return 0;
 136 }
 137
 138 static void do_sync_work(struct work_struct *work)
 139 {
 140         /*
 141          * Sync twice to reduce the possibility we skipped some inodes / pages
 142          * because they were temporarily locked
 143          */
 144         sync_filesystems(0);
 145         sync_filesystems(0);
 146         printk("Emergency Sync complete\n");
 147         kfree(work);
 148 }
 149
 150 void emergency_sync(void)
 151 {
 152         struct work_struct *work;
 153
 154         work = kmalloc(sizeof(*work), GFP_ATOMIC);
 155         if (work) {
 156                 INIT_WORK(work, do_sync_work);
 157                 schedule_work(work);
 158         }
 159 }
 160
 161 /*
 162  * Generic function to fsync a file.
 163  *
 164  * filp may be NULL if called via the msync of a vma.
 165  */
 166 int file_fsync(struct file *filp, struct dentry *dentry, int datasync)
 167 {
 168         struct inode * inode = dentry->d_inode;
 169         struct super_block * sb;
 170         int ret, err;
 171
 172         /* sync the inode to buffers */
 173         ret = write_inode_now(inode, 0);
 174
 175         /* sync the superblock to buffers */
 176         sb = inode->i_sb;
 177         if (sb->s_dirt && sb->s_op->write_super)
 178                 sb->s_op->write_super(sb);
 179
 180         /* .. finally sync the buffers to disk */
 181         err = sync_blockdev(sb->s_bdev);
 182         if (!ret)
 183                 ret = err;
 184         return ret;
 185 }
 186 EXPORT_SYMBOL(file_fsync);
 187
 188 /**
 189  * vfs_fsync_range - helper to sync a range of data & metadata to disk
 190  * @file:               file to sync
 191  * @dentry:             dentry of @file
 192  * @start:              offset in bytes of the beginning of data range to sync
 193  * @end:                offset in bytes of the end of data range (inclusive)
 194  * @datasync:           perform only datasync
 195  *
 196  * Write back data in range @start..@end and metadata for @file to disk.  If
 197  * @datasync is set only metadata needed to access modified file data is
 198  * written.
 199  *
 200  * In case this function is called from nfsd @file may be %NULL and
 201  * only @dentry is set.  This can only happen when the filesystem
 202  * implements the export_operations API.
 203  */
 204 int vfs_fsync_range(struct file *file, struct dentry *dentry, loff_t start,
 205                     loff_t end, int datasync)
 206 {
 207         const struct file_operations *fop;
 208         struct address_space *mapping;
 209         int err, ret;
 210
 211         /*
 212          * Get mapping and operations from the file in case we have
 213          * as file, or get the default values for them in case we
 214          * don't have a struct file available.  Damn nfsd..
 215          */
 216         if (file) {
 217                 mapping = file->f_mapping;
 218                 fop = file->f_op;
 219         } else {
 220                 mapping = dentry->d_inode->i_mapping;
 221                 fop = dentry->d_inode->i_fop;
 222         }
 223
 224         if (!fop || !fop->fsync) {
 225                 ret = -EINVAL;
 226                 goto out;
 227         }
 228
 229         ret = filemap_write_and_wait_range(mapping, start, end);
 230
 231         /*
 232          * We need to protect against concurrent writers, which could cause
 233          * livelocks in fsync_buffers_list().
 234          */
 235         mutex_lock(&mapping->host->i_mutex);
 236         err = fop->fsync(file, dentry, datasync);
 237         if (!ret)
 238                 ret = err;
 239         mutex_unlock(&mapping->host->i_mutex);
 240
 241 out:
 242         return ret;
 243 }
 244 EXPORT_SYMBOL(vfs_fsync_range);
 245
 246 /**
 247  * vfs_fsync - perform a fsync or fdatasync on a file
 248  * @file:               file to sync
 249  * @dentry:             dentry of @file
 250  * @datasync:           only perform a fdatasync operation
 251  *
 252  * Write back data and metadata for @file to disk.  If @datasync is
 253  * set only metadata needed to access modified file data is written.
 254  *
 255  * In case this function is called from nfsd @file may be %NULL and
 256  * only @dentry is set.  This can only happen when the filesystem
 257  * implements the export_operations API.
 258  */
 259 int vfs_fsync(struct file *file, struct dentry *dentry, int datasync)
 260 {
 261         return vfs_fsync_range(file, dentry, 0, LLONG_MAX, datasync);
 262 }
 263 EXPORT_SYMBOL(vfs_fsync);
 264
 265 static int do_fsync(unsigned int fd, int datasync)
 266 {
 267         struct file *file;
 268         int ret = -EBADF;
 269
 270         file = fget(fd);
 271         if (file) {
 272                 ret = vfs_fsync(file, file->f_path.dentry, datasync);
 273                 fput(file);
 274         }
 275         return ret;
 276 }
 277
 278 SYSCALL_DEFINE1(fsync, unsigned int, fd)
 279 {
 280         return do_fsync(fd, 0);
 281 }
 282
 283 SYSCALL_DEFINE1(fdatasync, unsigned int, fd)
 284 {
 285         return do_fsync(fd, 1);
 286 }
 287
 288 /**
 289  * generic_write_sync - perform syncing after a write if file / inode is sync
 290  * @file:       file to which the write happened
 291  * @pos:        offset where the write started
 292  * @count:      length of the write
 293  *
 294  * This is just a simple wrapper about our general syncing function.
 295  */
 296 int generic_write_sync(struct file *file, loff_t pos, loff_t count)
 297 {
 298         if (!(file->f_flags & O_SYNC) && !IS_SYNC(file->f_mapping->host))
 299                 return 0;
 300         return vfs_fsync_range(file, file->f_path.dentry, pos,
 301                                pos + count - 1, 1);
 302 }
 303 EXPORT_SYMBOL(generic_write_sync);
 304
 305 /*
 306  * sys_sync_file_range() permits finely controlled syncing over a segment of
 307  * a file in the range offset .. (offset+nbytes-1) inclusive.  If nbytes is
 308  * zero then sys_sync_file_range() will operate from offset out to EOF.
 309  *
 310  * The flag bits are:
 311  *
 312  * SYNC_FILE_RANGE_WAIT_BEFORE: wait upon writeout of all pages in the range
 313  * before performing the write.
 314  *
 315  * SYNC_FILE_RANGE_WRITE: initiate writeout of all those dirty pages in the
 316  * range which are not presently under writeback. Note that this may block for
 317  * significant periods due to exhaustion of disk request structures.
 318  *
 319  * SYNC_FILE_RANGE_WAIT_AFTER: wait upon writeout of all pages in the range
 320  * after performing the write.
 321  *
 322  * Useful combinations of the flag bits are:
 323  *
 324  * SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE: ensures that all pages
 325  * in the range which were dirty on entry to sys_sync_file_range() are placed
 326  * under writeout.  This is a start-write-for-data-integrity operation.
 327  *
 328  * SYNC_FILE_RANGE_WRITE: start writeout of all dirty pages in the range which
 329  * are not presently under writeout.  This is an asynchronous flush-to-disk
 330  * operation.  Not suitable for data integrity operations.
 331  *
 332  * SYNC_FILE_RANGE_WAIT_BEFORE (or SYNC_FILE_RANGE_WAIT_AFTER): wait for
 333  * completion of writeout of all pages in the range.  This will be used after an
 334  * earlier SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE operation to wait
 335  * for that operation to complete and to return the result.
 336  *
 337  * SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE|SYNC_FILE_RANGE_WAIT_AFTER:
 338  * a traditional sync() operation.  This is a write-for-data-integrity operation
 339  * which will ensure that all pages in the range which were dirty on entry to
 340  * sys_sync_file_range() are committed to disk.
 341  *
 342  *
 343  * SYNC_FILE_RANGE_WAIT_BEFORE and SYNC_FILE_RANGE_WAIT_AFTER will detect any
 344  * I/O errors or ENOSPC conditions and will return those to the caller, after
 345  * clearing the EIO and ENOSPC flags in the address_space.
 346  *
 347  * It should be noted that none of these operations write out the file's
 348  * metadata.  So unless the application is strictly performing overwrites of
 349  * already-instantiated disk blocks, there are no guarantees here that the data
 350  * will be available after a crash.
 351  */
 352 SYSCALL_DEFINE(sync_file_range)(int fd, loff_t offset, loff_t nbytes,
 353                                 unsigned int flags)
 354 {
 355         int ret;
 356         struct file *file;
 357         loff_t endbyte;                 /* inclusive */
 358         int fput_needed;
 359         umode_t i_mode;
 360
 361         ret = -EINVAL;
 362         if (flags & ~VALID_FLAGS)
 363                 goto out;
 364
 365         endbyte = offset + nbytes;
 366
 367         if ((s64)offset < 0)
 368                 goto out;
 369         if ((s64)endbyte < 0)
 370                 goto out;
 371         if (endbyte < offset)
 372                 goto out;
 373
 374         if (sizeof(pgoff_t) == 4) {
 375                 if (offset >= (0x100000000ULL << PAGE_CACHE_SHIFT)) {
 376                         /*
 377                          * The range starts outside a 32 bit machine's
 378                          * pagecache addressing capabilities.  Let it "succeed"
 379                          */
 380                         ret = 0;
 381                         goto out;
 382                 }
 383                 if (endbyte >= (0x100000000ULL << PAGE_CACHE_SHIFT)) {
 384                         /*
 385                          * Out to EOF
 386                          */
 387                         nbytes = 0;
 388                 }
 389         }
 390
 391         if (nbytes == 0)
 392                 endbyte = LLONG_MAX;
 393         else
 394                 endbyte--;              /* inclusive */
 395
 396         ret = -EBADF;
 397         file = fget_light(fd, &fput_needed);
 398         if (!file)
 399                 goto out;
 400
 401         i_mode = file->f_path.dentry->d_inode->i_mode;
 402         ret = -ESPIPE;
 403         if (!S_ISREG(i_mode) && !S_ISBLK(i_mode) && !S_ISDIR(i_mode) &&
 404                         !S_ISLNK(i_mode))
 405                 goto out_put;
 406
 407         ret = do_sync_mapping_range(file->f_mapping, offset, endbyte, flags);
 408 out_put:
 409         fput_light(file, fput_needed);
 410 out:
 411         return ret;
 412 }
 413 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
 414 asmlinkage long SyS_sync_file_range(long fd, loff_t offset, loff_t nbytes,
 415                                     long flags)
 416 {
 417         return SYSC_sync_file_range((int) fd, offset, nbytes,
 418                                     (unsigned int) flags);
 419 }
 420 SYSCALL_ALIAS(sys_sync_file_range, SyS_sync_file_range);
 421 #endif
 422
 423 /* It would be nice if people remember that not all the world's an i386
 424    when they introduce new system calls */
 425 SYSCALL_DEFINE(sync_file_range2)(int fd, unsigned int flags,
 426                                  loff_t offset, loff_t nbytes)
 427 {
 428         return sys_sync_file_range(fd, offset, nbytes, flags);
 429 }
 430 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
 431 asmlinkage long SyS_sync_file_range2(long fd, long flags,
 432                                      loff_t offset, loff_t nbytes)
 433 {
 434         return SYSC_sync_file_range2((int) fd, (unsigned int) flags,
 435                                      offset, nbytes);
 436 }
 437 SYSCALL_ALIAS(sys_sync_file_range2, SyS_sync_file_range2);
 438 #endif
 439
 440 /*
 441  * `endbyte' is inclusive
 442  */
 443 int do_sync_mapping_range(struct address_space *mapping, loff_t offset,
 444                           loff_t endbyte, unsigned int flags)
 445 {
 446         int ret;
 447
 448         if (!mapping) {
 449                 ret = -EINVAL;
 450                 goto out;
 451         }
 452
 453         ret = 0;
 454         if (flags & SYNC_FILE_RANGE_WAIT_BEFORE) {
 455                 ret = wait_on_page_writeback_range(mapping,
 456                                         offset >> PAGE_CACHE_SHIFT,
 457                                         endbyte >> PAGE_CACHE_SHIFT);
 458                 if (ret < 0)
 459                         goto out;
 460         }
 461
 462         if (flags & SYNC_FILE_RANGE_WRITE) {
 463                 ret = __filemap_fdatawrite_range(mapping, offset, endbyte,
 464                                                 WB_SYNC_ALL);
 465                 if (ret < 0)
 466                         goto out;
 467         }
 468
 469         if (flags & SYNC_FILE_RANGE_WAIT_AFTER) {
 470                 ret = wait_on_page_writeback_range(mapping,
 471                                         offset >> PAGE_CACHE_SHIFT,
 472                                         endbyte >> PAGE_CACHE_SHIFT);
 473         }
 474 out:
 475         return ret;
 476 }
 477 EXPORT_SYMBOL_GPL(do_sync_mapping_range);