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_GPL(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         filemap_fdatawait(bdev->bd_inode->i_mapping);
  90 }
  91
  92 /*
  93  * Sync everything. We start by waking flusher threads so that most of
  94  * writeback runs on all devices in parallel. Then we sync all inodes reliably
  95  * which effectively also waits for all flusher threads to finish doing
  96  * writeback. At this point all data is on disk so metadata should be stable
  97  * and we tell filesystems to sync their metadata via ->sync_fs() calls.
  98  * Finally, we writeout all block devices because some filesystems (e.g. ext2)
  99  * just write metadata (such as inodes or bitmaps) to block device page cache
 100  * and do not sync it on their own in ->sync_fs().
 101  */
 102 SYSCALL_DEFINE0(sync)
 103 {
 104         int nowait = 0, wait = 1;
 105
 106         wakeup_flusher_threads(0, WB_REASON_SYNC);
 107         iterate_supers(sync_inodes_one_sb, NULL);
 108         iterate_supers(sync_fs_one_sb, &nowait);
 109         iterate_supers(sync_fs_one_sb, &wait);
 110         iterate_bdevs(fdatawrite_one_bdev, NULL);
 111         iterate_bdevs(fdatawait_one_bdev, NULL);
 112         if (unlikely(laptop_mode))
 113                 laptop_sync_completion();
 114         return 0;
 115 }
 116
 117 static void do_sync_work(struct work_struct *work)
 118 {
 119         int nowait = 0;
 120
 121         /*
 122          * Sync twice to reduce the possibility we skipped some inodes / pages
 123          * because they were temporarily locked
 124          */
 125         iterate_supers(sync_inodes_one_sb, &nowait);
 126         iterate_supers(sync_fs_one_sb, &nowait);
 127         iterate_bdevs(fdatawrite_one_bdev, NULL);
 128         iterate_supers(sync_inodes_one_sb, &nowait);
 129         iterate_supers(sync_fs_one_sb, &nowait);
 130         iterate_bdevs(fdatawrite_one_bdev, NULL);
 131         printk("Emergency Sync complete\n");
 132         kfree(work);
 133 }
 134
 135 void emergency_sync(void)
 136 {
 137         struct work_struct *work;
 138
 139         work = kmalloc(sizeof(*work), GFP_ATOMIC);
 140         if (work) {
 141                 INIT_WORK(work, do_sync_work);
 142                 schedule_work(work);
 143         }
 144 }
 145
 146 /*
 147  * sync a single super
 148  */
 149 SYSCALL_DEFINE1(syncfs, int, fd)
 150 {
 151         struct fd f = fdget(fd);
 152         struct super_block *sb;
 153         int ret;
 154
 155         if (!f.file)
 156                 return -EBADF;
 157         sb = f.file->f_dentry->d_sb;
 158
 159         down_read(&sb->s_umount);
 160         ret = sync_filesystem(sb);
 161         up_read(&sb->s_umount);
 162
 163         fdput(f);
 164         return ret;
 165 }
 166
 167 /**
 168  * vfs_fsync_range - helper to sync a range of data & metadata to disk
 169  * @file:               file to sync
 170  * @start:              offset in bytes of the beginning of data range to sync
 171  * @end:                offset in bytes of the end of data range (inclusive)
 172  * @datasync:           perform only datasync
 173  *
 174  * Write back data in range @start..@end and metadata for @file to disk.  If
 175  * @datasync is set only metadata needed to access modified file data is
 176  * written.
 177  */
 178 int vfs_fsync_range(struct file *file, loff_t start, loff_t end, int datasync)
 179 {
 180         if (!file->f_op || !file->f_op->fsync)
 181                 return -EINVAL;
 182         return file->f_op->fsync(file, start, end, datasync);
 183 }
 184 EXPORT_SYMBOL(vfs_fsync_range);
 185
 186 /**
 187  * vfs_fsync - perform a fsync or fdatasync on a file
 188  * @file:               file to sync
 189  * @datasync:           only perform a fdatasync operation
 190  *
 191  * Write back data and metadata for @file to disk.  If @datasync is
 192  * set only metadata needed to access modified file data is written.
 193  */
 194 int vfs_fsync(struct file *file, int datasync)
 195 {
 196         return vfs_fsync_range(file, 0, LLONG_MAX, datasync);
 197 }
 198 EXPORT_SYMBOL(vfs_fsync);
 199
 200 static int do_fsync(unsigned int fd, int datasync)
 201 {
 202         struct fd f = fdget(fd);
 203         int ret = -EBADF;
 204
 205         if (f.file) {
 206                 ret = vfs_fsync(f.file, datasync);
 207                 fdput(f);
 208         }
 209         return ret;
 210 }
 211
 212 SYSCALL_DEFINE1(fsync, unsigned int, fd)
 213 {
 214         return do_fsync(fd, 0);
 215 }
 216
 217 SYSCALL_DEFINE1(fdatasync, unsigned int, fd)
 218 {
 219         return do_fsync(fd, 1);
 220 }
 221
 222 /**
 223  * generic_write_sync - perform syncing after a write if file / inode is sync
 224  * @file:       file to which the write happened
 225  * @pos:        offset where the write started
 226  * @count:      length of the write
 227  *
 228  * This is just a simple wrapper about our general syncing function.
 229  */
 230 int generic_write_sync(struct file *file, loff_t pos, loff_t count)
 231 {
 232         if (!(file->f_flags & O_DSYNC) && !IS_SYNC(file->f_mapping->host))
 233                 return 0;
 234         return vfs_fsync_range(file, pos, pos + count - 1,
 235                                (file->f_flags & __O_SYNC) ? 0 : 1);
 236 }
 237 EXPORT_SYMBOL(generic_write_sync);
 238
 239 /*
 240  * sys_sync_file_range() permits finely controlled syncing over a segment of
 241  * a file in the range offset .. (offset+nbytes-1) inclusive.  If nbytes is
 242  * zero then sys_sync_file_range() will operate from offset out to EOF.
 243  *
 244  * The flag bits are:
 245  *
 246  * SYNC_FILE_RANGE_WAIT_BEFORE: wait upon writeout of all pages in the range
 247  * before performing the write.
 248  *
 249  * SYNC_FILE_RANGE_WRITE: initiate writeout of all those dirty pages in the
 250  * range which are not presently under writeback. Note that this may block for
 251  * significant periods due to exhaustion of disk request structures.
 252  *
 253  * SYNC_FILE_RANGE_WAIT_AFTER: wait upon writeout of all pages in the range
 254  * after performing the write.
 255  *
 256  * Useful combinations of the flag bits are:
 257  *
 258  * SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE: ensures that all pages
 259  * in the range which were dirty on entry to sys_sync_file_range() are placed
 260  * under writeout.  This is a start-write-for-data-integrity operation.
 261  *
 262  * SYNC_FILE_RANGE_WRITE: start writeout of all dirty pages in the range which
 263  * are not presently under writeout.  This is an asynchronous flush-to-disk
 264  * operation.  Not suitable for data integrity operations.
 265  *
 266  * SYNC_FILE_RANGE_WAIT_BEFORE (or SYNC_FILE_RANGE_WAIT_AFTER): wait for
 267  * completion of writeout of all pages in the range.  This will be used after an
 268  * earlier SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE operation to wait
 269  * for that operation to complete and to return the result.
 270  *
 271  * SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE|SYNC_FILE_RANGE_WAIT_AFTER:
 272  * a traditional sync() operation.  This is a write-for-data-integrity operation
 273  * which will ensure that all pages in the range which were dirty on entry to
 274  * sys_sync_file_range() are committed to disk.
 275  *
 276  *
 277  * SYNC_FILE_RANGE_WAIT_BEFORE and SYNC_FILE_RANGE_WAIT_AFTER will detect any
 278  * I/O errors or ENOSPC conditions and will return those to the caller, after
 279  * clearing the EIO and ENOSPC flags in the address_space.
 280  *
 281  * It should be noted that none of these operations write out the file's
 282  * metadata.  So unless the application is strictly performing overwrites of
 283  * already-instantiated disk blocks, there are no guarantees here that the data
 284  * will be available after a crash.
 285  */
 286 SYSCALL_DEFINE4(sync_file_range, int, fd, loff_t, offset, loff_t, nbytes,
 287                                 unsigned int, flags)
 288 {
 289         int ret;
 290         struct fd f;
 291         struct address_space *mapping;
 292         loff_t endbyte;                 /* inclusive */
 293         umode_t i_mode;
 294
 295         ret = -EINVAL;
 296         if (flags & ~VALID_FLAGS)
 297                 goto out;
 298
 299         endbyte = offset + nbytes;
 300
 301         if ((s64)offset < 0)
 302                 goto out;
 303         if ((s64)endbyte < 0)
 304                 goto out;
 305         if (endbyte < offset)
 306                 goto out;
 307
 308         if (sizeof(pgoff_t) == 4) {
 309                 if (offset >= (0x100000000ULL << PAGE_CACHE_SHIFT)) {
 310                         /*
 311                          * The range starts outside a 32 bit machine's
 312                          * pagecache addressing capabilities.  Let it "succeed"
 313                          */
 314                         ret = 0;
 315                         goto out;
 316                 }
 317                 if (endbyte >= (0x100000000ULL << PAGE_CACHE_SHIFT)) {
 318                         /*
 319                          * Out to EOF
 320                          */
 321                         nbytes = 0;
 322                 }
 323         }
 324
 325         if (nbytes == 0)
 326                 endbyte = LLONG_MAX;
 327         else
 328                 endbyte--;              /* inclusive */
 329
 330         ret = -EBADF;
 331         f = fdget(fd);
 332         if (!f.file)
 333                 goto out;
 334
 335         i_mode = file_inode(f.file)->i_mode;
 336         ret = -ESPIPE;
 337         if (!S_ISREG(i_mode) && !S_ISBLK(i_mode) && !S_ISDIR(i_mode) &&
 338                         !S_ISLNK(i_mode))
 339                 goto out_put;
 340
 341         mapping = f.file->f_mapping;
 342         if (!mapping) {
 343                 ret = -EINVAL;
 344                 goto out_put;
 345         }
 346
 347         ret = 0;
 348         if (flags & SYNC_FILE_RANGE_WAIT_BEFORE) {
 349                 ret = filemap_fdatawait_range(mapping, offset, endbyte);
 350                 if (ret < 0)
 351                         goto out_put;
 352         }
 353
 354         if (flags & SYNC_FILE_RANGE_WRITE) {
 355                 ret = filemap_fdatawrite_range(mapping, offset, endbyte);
 356                 if (ret < 0)
 357                         goto out_put;
 358         }
 359
 360         if (flags & SYNC_FILE_RANGE_WAIT_AFTER)
 361                 ret = filemap_fdatawait_range(mapping, offset, endbyte);
 362
 363 out_put:
 364         fdput(f);
 365 out:
 366         return ret;
 367 }
 368
 369 /* It would be nice if people remember that not all the world's an i386
 370    when they introduce new system calls */
 371 SYSCALL_DEFINE4(sync_file_range2, int, fd, unsigned int, flags,
 372                                  loff_t, offset, loff_t, nbytes)
 373 {
 374         return sys_sync_file_range(fd, offset, nbytes, flags);
 375 }