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