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/export.h>
  10 #include <linux/namei.h>
  11 #include <linux/sched.h>
  12 #include <linux/writeback.h>
  13 #include <linux/syscalls.h>
  14 #include <linux/linkage.h>
  15 #include <linux/pagemap.h>
  16 #include <linux/quotaops.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         if (wait)
  33                 sync_inodes_sb(sb);
  34         else
  35                 writeback_inodes_sb(sb, WB_REASON_SYNC);
  36
  37         if (sb->s_op->sync_fs)
  38                 sb->s_op->sync_fs(sb, wait);
  39         return __sync_blockdev(sb->s_bdev, wait);
  40 }
  41
  42 /*
  43  * Write out and wait upon all dirty data associated with this
  44  * superblock.  Filesystem data as well as the underlying block
  45  * device.  Takes the superblock lock.
  46  */
  47 int sync_filesystem(struct super_block *sb)
  48 {
  49         int ret;
  50
  51         /*
  52          * We need to be protected against the filesystem going from
  53          * r/o to r/w or vice versa.
  54          */
  55         WARN_ON(!rwsem_is_locked(&sb->s_umount));
  56
  57         /*
  58          * No point in syncing out anything if the filesystem is read-only.
  59          */
  60         if (sb->s_flags & MS_RDONLY)
  61                 return 0;
  62
  63         ret = __sync_filesystem(sb, 0);
  64         if (ret < 0)
  65                 return ret;
  66         return __sync_filesystem(sb, 1);
  67 }
  68 EXPORT_SYMBOL(sync_filesystem);
  69
  70 static void sync_inodes_one_sb(struct super_block *sb, void *arg)
  71 {
  72         if (!(sb->s_flags & MS_RDONLY))
  73                 sync_inodes_sb(sb);
  74 }
  75
  76 static void sync_fs_one_sb(struct super_block *sb, void *arg)
  77 {
  78         if (!(sb->s_flags & MS_RDONLY) && sb->s_op->sync_fs)
  79                 sb->s_op->sync_fs(sb, *(int *)arg);
  80 }
  81
  82 static void fdatawrite_one_bdev(struct block_device *bdev, void *arg)
  83 {
  84         filemap_fdatawrite(bdev->bd_inode->i_mapping);
  85 }
  86
  87 static void fdatawait_one_bdev(struct block_device *bdev, void *arg)
  88 {
  89         /*
  90          * We keep the error status of individual mapping so that
  91          * applications can catch the writeback error using fsync(2).
  92          * See filemap_fdatawait_keep_errors() for details.
  93          */
  94         filemap_fdatawait_keep_errors(bdev->bd_inode->i_mapping);
  95 }
  96
  97 /*
  98  * Sync everything. We start by waking flusher threads so that most of
  99  * writeback runs on all devices in parallel. Then we sync all inodes reliably
 100  * which effectively also waits for all flusher threads to finish doing
 101  * writeback. At this point all data is on disk so metadata should be stable
 102  * and we tell filesystems to sync their metadata via ->sync_fs() calls.
 103  * Finally, we writeout all block devices because some filesystems (e.g. ext2)
 104  * just write metadata (such as inodes or bitmaps) to block device page cache
 105  * and do not sync it on their own in ->sync_fs().
 106  */
 107 SYSCALL_DEFINE0(sync)
 108 {
 109         int nowait = 0, wait = 1;
 110
 111         wakeup_flusher_threads(0, WB_REASON_SYNC);
 112         iterate_supers(sync_inodes_one_sb, NULL);
 113         iterate_supers(sync_fs_one_sb, &nowait);
 114         iterate_supers(sync_fs_one_sb, &wait);
 115         iterate_bdevs(fdatawrite_one_bdev, NULL);
 116         iterate_bdevs(fdatawait_one_bdev, NULL);
 117         if (unlikely(laptop_mode))
 118                 laptop_sync_completion();
 119         return 0;
 120 }
 121
 122 static void do_sync_work(struct work_struct *work)
 123 {
 124         int nowait = 0;
 125
 126         /*
 127          * Sync twice to reduce the possibility we skipped some inodes / pages
 128          * because they were temporarily locked
 129          */
 130         iterate_supers(sync_inodes_one_sb, &nowait);
 131         iterate_supers(sync_fs_one_sb, &nowait);
 132         iterate_bdevs(fdatawrite_one_bdev, NULL);
 133         iterate_supers(sync_inodes_one_sb, &nowait);
 134         iterate_supers(sync_fs_one_sb, &nowait);
 135         iterate_bdevs(fdatawrite_one_bdev, NULL);
 136         printk("Emergency Sync complete\n");
 137         kfree(work);
 138 }
 139
 140 void emergency_sync(void)
 141 {
 142         struct work_struct *work;
 143
 144         work = kmalloc(sizeof(*work), GFP_ATOMIC);
 145         if (work) {
 146                 INIT_WORK(work, do_sync_work);
 147                 schedule_work(work);
 148         }
 149 }
 150
 151 /*
 152  * sync a single super
 153  */
 154 SYSCALL_DEFINE1(syncfs, int, fd)
 155 {
 156         struct fd f = fdget(fd);
 157         struct super_block *sb;
 158         int ret;
 159
 160         if (!f.file)
 161                 return -EBADF;
 162         sb = f.file->f_path.dentry->d_sb;
 163
 164         down_read(&sb->s_umount);
 165         ret = sync_filesystem(sb);
 166         up_read(&sb->s_umount);
 167
 168         fdput(f);
 169         return ret;
 170 }
 171
 172 /**
 173  * vfs_fsync_range - helper to sync a range of data & metadata to disk
 174  * @file:               file to sync
 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 int vfs_fsync_range(struct file *file, loff_t start, loff_t end, int datasync)
 184 {
 185         struct inode *inode = file->f_mapping->host;
 186
 187         if (!file->f_op->fsync)
 188                 return -EINVAL;
 189         if (!datasync && (inode->i_state & I_DIRTY_TIME)) {
 190                 spin_lock(&inode->i_lock);
 191                 inode->i_state &= ~I_DIRTY_TIME;
 192                 spin_unlock(&inode->i_lock);
 193                 mark_inode_dirty_sync(inode);
 194         }
 195         return file->f_op->fsync(file, start, end, datasync);
 196 }
 197 EXPORT_SYMBOL(vfs_fsync_range);
 198
 199 /**
 200  * vfs_fsync - perform a fsync or fdatasync on a file
 201  * @file:               file to sync
 202  * @datasync:           only perform a fdatasync operation
 203  *
 204  * Write back data and metadata for @file to disk.  If @datasync is
 205  * set only metadata needed to access modified file data is written.
 206  */
 207 int vfs_fsync(struct file *file, int datasync)
 208 {
 209         return vfs_fsync_range(file, 0, LLONG_MAX, datasync);
 210 }
 211 EXPORT_SYMBOL(vfs_fsync);
 212
 213 static int do_fsync(unsigned int fd, int datasync)
 214 {
 215         struct fd f = fdget(fd);
 216         int ret = -EBADF;
 217
 218         if (f.file) {
 219                 ret = vfs_fsync(f.file, datasync);
 220                 fdput(f);
 221         }
 222         return ret;
 223 }
 224
 225 SYSCALL_DEFINE1(fsync, unsigned int, fd)
 226 {
 227         return do_fsync(fd, 0);
 228 }
 229
 230 SYSCALL_DEFINE1(fdatasync, unsigned int, fd)
 231 {
 232         return do_fsync(fd, 1);
 233 }
 234
 235 /*
 236  * sys_sync_file_range() permits finely controlled syncing over a segment of
 237  * a file in the range offset .. (offset+nbytes-1) inclusive.  If nbytes is
 238  * zero then sys_sync_file_range() will operate from offset out to EOF.
 239  *
 240  * The flag bits are:
 241  *
 242  * SYNC_FILE_RANGE_WAIT_BEFORE: wait upon writeout of all pages in the range
 243  * before performing the write.
 244  *
 245  * SYNC_FILE_RANGE_WRITE: initiate writeout of all those dirty pages in the
 246  * range which are not presently under writeback. Note that this may block for
 247  * significant periods due to exhaustion of disk request structures.
 248  *
 249  * SYNC_FILE_RANGE_WAIT_AFTER: wait upon writeout of all pages in the range
 250  * after performing the write.
 251  *
 252  * Useful combinations of the flag bits are:
 253  *
 254  * SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE: ensures that all pages
 255  * in the range which were dirty on entry to sys_sync_file_range() are placed
 256  * under writeout.  This is a start-write-for-data-integrity operation.
 257  *
 258  * SYNC_FILE_RANGE_WRITE: start writeout of all dirty pages in the range which
 259  * are not presently under writeout.  This is an asynchronous flush-to-disk
 260  * operation.  Not suitable for data integrity operations.
 261  *
 262  * SYNC_FILE_RANGE_WAIT_BEFORE (or SYNC_FILE_RANGE_WAIT_AFTER): wait for
 263  * completion of writeout of all pages in the range.  This will be used after an
 264  * earlier SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE operation to wait
 265  * for that operation to complete and to return the result.
 266  *
 267  * SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE|SYNC_FILE_RANGE_WAIT_AFTER:
 268  * a traditional sync() operation.  This is a write-for-data-integrity operation
 269  * which will ensure that all pages in the range which were dirty on entry to
 270  * sys_sync_file_range() are committed to disk.
 271  *
 272  *
 273  * SYNC_FILE_RANGE_WAIT_BEFORE and SYNC_FILE_RANGE_WAIT_AFTER will detect any
 274  * I/O errors or ENOSPC conditions and will return those to the caller, after
 275  * clearing the EIO and ENOSPC flags in the address_space.
 276  *
 277  * It should be noted that none of these operations write out the file's
 278  * metadata.  So unless the application is strictly performing overwrites of
 279  * already-instantiated disk blocks, there are no guarantees here that the data
 280  * will be available after a crash.
 281  */
 282 SYSCALL_DEFINE4(sync_file_range, int, fd, loff_t, offset, loff_t, nbytes,
 283                                 unsigned int, flags)
 284 {
 285         int ret;
 286         struct fd f;
 287         struct address_space *mapping;
 288         loff_t endbyte;                 /* inclusive */
 289         umode_t i_mode;
 290
 291         ret = -EINVAL;
 292         if (flags & ~VALID_FLAGS)
 293                 goto out;
 294
 295         endbyte = offset + nbytes;
 296
 297         if ((s64)offset < 0)
 298                 goto out;
 299         if ((s64)endbyte < 0)
 300                 goto out;
 301         if (endbyte < offset)
 302                 goto out;
 303
 304         if (sizeof(pgoff_t) == 4) {
 305                 if (offset >= (0x100000000ULL << PAGE_CACHE_SHIFT)) {
 306                         /*
 307                          * The range starts outside a 32 bit machine's
 308                          * pagecache addressing capabilities.  Let it "succeed"
 309                          */
 310                         ret = 0;
 311                         goto out;
 312                 }
 313                 if (endbyte >= (0x100000000ULL << PAGE_CACHE_SHIFT)) {
 314                         /*
 315                          * Out to EOF
 316                          */
 317                         nbytes = 0;
 318                 }
 319         }
 320
 321         if (nbytes == 0)
 322                 endbyte = LLONG_MAX;
 323         else
 324                 endbyte--;              /* inclusive */
 325
 326         ret = -EBADF;
 327         f = fdget(fd);
 328         if (!f.file)
 329                 goto out;
 330
 331         i_mode = file_inode(f.file)->i_mode;
 332         ret = -ESPIPE;
 333         if (!S_ISREG(i_mode) && !S_ISBLK(i_mode) && !S_ISDIR(i_mode) &&
 334                         !S_ISLNK(i_mode))
 335                 goto out_put;
 336
 337         mapping = f.file->f_mapping;
 338         if (!mapping) {
 339                 ret = -EINVAL;
 340                 goto out_put;
 341         }
 342
 343         ret = 0;
 344         if (flags & SYNC_FILE_RANGE_WAIT_BEFORE) {
 345                 ret = filemap_fdatawait_range(mapping, offset, endbyte);
 346                 if (ret < 0)
 347                         goto out_put;
 348         }
 349
 350         if (flags & SYNC_FILE_RANGE_WRITE) {
 351                 ret = __filemap_fdatawrite_range(mapping, offset, endbyte,
 352                                                  WB_SYNC_NONE);
 353                 if (ret < 0)
 354                         goto out_put;
 355         }
 356
 357         if (flags & SYNC_FILE_RANGE_WAIT_AFTER)
 358                 ret = filemap_fdatawait_range(mapping, offset, endbyte);
 359
 360 out_put:
 361         fdput(f);
 362 out:
 363         return ret;
 364 }
 365
 366 /* It would be nice if people remember that not all the world's an i386
 367    when they introduce new system calls */
 368 SYSCALL_DEFINE4(sync_file_range2, int, fd, unsigned int, flags,
 369                                  loff_t, offset, loff_t, nbytes)
 370 {
 371         return sys_sync_file_range(fd, offset, nbytes, flags);
 372 }