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