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  * 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  * @dentry:             dentry of @file
 162  * @start:              offset in bytes of the beginning of data range to sync
 163  * @end:                offset in bytes of the end of data range (inclusive)
 164  * @datasync:           perform only datasync
 165  *
 166  * Write back data in range @start..@end and metadata for @file to disk.  If
 167  * @datasync is set only metadata needed to access modified file data is
 168  * written.
 169  *
 170  * In case this function is called from nfsd @file may be %NULL and
 171  * only @dentry is set.  This can only happen when the filesystem
 172  * implements the export_operations API.
 173  */
 174 int vfs_fsync_range(struct file *file, struct dentry *dentry, loff_t start,
 175                     loff_t end, int datasync)
 176 {
 177         const struct file_operations *fop;
 178         struct address_space *mapping;
 179         int err, ret;
 180
 181         /*
 182          * Get mapping and operations from the file in case we have
 183          * as file, or get the default values for them in case we
 184          * don't have a struct file available.  Damn nfsd..
 185          */
 186         if (file) {
 187                 mapping = file->f_mapping;
 188                 fop = file->f_op;
 189         } else {
 190                 mapping = dentry->d_inode->i_mapping;
 191                 fop = dentry->d_inode->i_fop;
 192         }
 193
 194         if (!fop || !fop->fsync) {
 195                 ret = -EINVAL;
 196                 goto out;
 197         }
 198
 199         ret = filemap_write_and_wait_range(mapping, start, end);
 200
 201         /*
 202          * We need to protect against concurrent writers, which could cause
 203          * livelocks in fsync_buffers_list().
 204          */
 205         mutex_lock(&mapping->host->i_mutex);
 206         err = fop->fsync(file, dentry, datasync);
 207         if (!ret)
 208                 ret = err;
 209         mutex_unlock(&mapping->host->i_mutex);
 210
 211 out:
 212         return ret;
 213 }
 214 EXPORT_SYMBOL(vfs_fsync_range);
 215
 216 /**
 217  * vfs_fsync - perform a fsync or fdatasync on a file
 218  * @file:               file to sync
 219  * @dentry:             dentry of @file
 220  * @datasync:           only perform a fdatasync operation
 221  *
 222  * Write back data and metadata for @file to disk.  If @datasync is
 223  * set only metadata needed to access modified file data is written.
 224  *
 225  * In case this function is called from nfsd @file may be %NULL and
 226  * only @dentry is set.  This can only happen when the filesystem
 227  * implements the export_operations API.
 228  */
 229 int vfs_fsync(struct file *file, struct dentry *dentry, int datasync)
 230 {
 231         return vfs_fsync_range(file, dentry, 0, LLONG_MAX, datasync);
 232 }
 233 EXPORT_SYMBOL(vfs_fsync);
 234
 235 static int do_fsync(unsigned int fd, int datasync)
 236 {
 237         struct file *file;
 238         int ret = -EBADF;
 239
 240         file = fget(fd);
 241         if (file) {
 242                 ret = vfs_fsync(file, file->f_path.dentry, datasync);
 243                 fput(file);
 244         }
 245         return ret;
 246 }
 247
 248 SYSCALL_DEFINE1(fsync, unsigned int, fd)
 249 {
 250         return do_fsync(fd, 0);
 251 }
 252
 253 SYSCALL_DEFINE1(fdatasync, unsigned int, fd)
 254 {
 255         return do_fsync(fd, 1);
 256 }
 257
 258 /**
 259  * generic_write_sync - perform syncing after a write if file / inode is sync
 260  * @file:       file to which the write happened
 261  * @pos:        offset where the write started
 262  * @count:      length of the write
 263  *
 264  * This is just a simple wrapper about our general syncing function.
 265  */
 266 int generic_write_sync(struct file *file, loff_t pos, loff_t count)
 267 {
 268         if (!(file->f_flags & O_DSYNC) && !IS_SYNC(file->f_mapping->host))
 269                 return 0;
 270         return vfs_fsync_range(file, file->f_path.dentry, pos,
 271                                pos + count - 1,
 272                                (file->f_flags & __O_SYNC) ? 0 : 1);
 273 }
 274 EXPORT_SYMBOL(generic_write_sync);
 275
 276 /*
 277  * sys_sync_file_range() permits finely controlled syncing over a segment of
 278  * a file in the range offset .. (offset+nbytes-1) inclusive.  If nbytes is
 279  * zero then sys_sync_file_range() will operate from offset out to EOF.
 280  *
 281  * The flag bits are:
 282  *
 283  * SYNC_FILE_RANGE_WAIT_BEFORE: wait upon writeout of all pages in the range
 284  * before performing the write.
 285  *
 286  * SYNC_FILE_RANGE_WRITE: initiate writeout of all those dirty pages in the
 287  * range which are not presently under writeback. Note that this may block for
 288  * significant periods due to exhaustion of disk request structures.
 289  *
 290  * SYNC_FILE_RANGE_WAIT_AFTER: wait upon writeout of all pages in the range
 291  * after performing the write.
 292  *
 293  * Useful combinations of the flag bits are:
 294  *
 295  * SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE: ensures that all pages
 296  * in the range which were dirty on entry to sys_sync_file_range() are placed
 297  * under writeout.  This is a start-write-for-data-integrity operation.
 298  *
 299  * SYNC_FILE_RANGE_WRITE: start writeout of all dirty pages in the range which
 300  * are not presently under writeout.  This is an asynchronous flush-to-disk
 301  * operation.  Not suitable for data integrity operations.
 302  *
 303  * SYNC_FILE_RANGE_WAIT_BEFORE (or SYNC_FILE_RANGE_WAIT_AFTER): wait for
 304  * completion of writeout of all pages in the range.  This will be used after an
 305  * earlier SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE operation to wait
 306  * for that operation to complete and to return the result.
 307  *
 308  * SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE|SYNC_FILE_RANGE_WAIT_AFTER:
 309  * a traditional sync() operation.  This is a write-for-data-integrity operation
 310  * which will ensure that all pages in the range which were dirty on entry to
 311  * sys_sync_file_range() are committed to disk.
 312  *
 313  *
 314  * SYNC_FILE_RANGE_WAIT_BEFORE and SYNC_FILE_RANGE_WAIT_AFTER will detect any
 315  * I/O errors or ENOSPC conditions and will return those to the caller, after
 316  * clearing the EIO and ENOSPC flags in the address_space.
 317  *
 318  * It should be noted that none of these operations write out the file's
 319  * metadata.  So unless the application is strictly performing overwrites of
 320  * already-instantiated disk blocks, there are no guarantees here that the data
 321  * will be available after a crash.
 322  */
 323 SYSCALL_DEFINE(sync_file_range)(int fd, loff_t offset, loff_t nbytes,
 324                                 unsigned int flags)
 325 {
 326         int ret;
 327         struct file *file;
 328         struct address_space *mapping;
 329         loff_t endbyte;                 /* inclusive */
 330         int fput_needed;
 331         umode_t i_mode;
 332
 333         ret = -EINVAL;
 334         if (flags & ~VALID_FLAGS)
 335                 goto out;
 336
 337         endbyte = offset + nbytes;
 338
 339         if ((s64)offset < 0)
 340                 goto out;
 341         if ((s64)endbyte < 0)
 342                 goto out;
 343         if (endbyte < offset)
 344                 goto out;
 345
 346         if (sizeof(pgoff_t) == 4) {
 347                 if (offset >= (0x100000000ULL << PAGE_CACHE_SHIFT)) {
 348                         /*
 349                          * The range starts outside a 32 bit machine's
 350                          * pagecache addressing capabilities.  Let it "succeed"
 351                          */
 352                         ret = 0;
 353                         goto out;
 354                 }
 355                 if (endbyte >= (0x100000000ULL << PAGE_CACHE_SHIFT)) {
 356                         /*
 357                          * Out to EOF
 358                          */
 359                         nbytes = 0;
 360                 }
 361         }
 362
 363         if (nbytes == 0)
 364                 endbyte = LLONG_MAX;
 365         else
 366                 endbyte--;              /* inclusive */
 367
 368         ret = -EBADF;
 369         file = fget_light(fd, &fput_needed);
 370         if (!file)
 371                 goto out;
 372
 373         i_mode = file->f_path.dentry->d_inode->i_mode;
 374         ret = -ESPIPE;
 375         if (!S_ISREG(i_mode) && !S_ISBLK(i_mode) && !S_ISDIR(i_mode) &&
 376                         !S_ISLNK(i_mode))
 377                 goto out_put;
 378
 379         mapping = file->f_mapping;
 380         if (!mapping) {
 381                 ret = -EINVAL;
 382                 goto out_put;
 383         }
 384
 385         ret = 0;
 386         if (flags & SYNC_FILE_RANGE_WAIT_BEFORE) {
 387                 ret = filemap_fdatawait_range(mapping, offset, endbyte);
 388                 if (ret < 0)
 389                         goto out_put;
 390         }
 391
 392         if (flags & SYNC_FILE_RANGE_WRITE) {
 393                 ret = filemap_fdatawrite_range(mapping, offset, endbyte);
 394                 if (ret < 0)
 395                         goto out_put;
 396         }
 397
 398         if (flags & SYNC_FILE_RANGE_WAIT_AFTER)
 399                 ret = filemap_fdatawait_range(mapping, offset, endbyte);
 400
 401 out_put:
 402         fput_light(file, fput_needed);
 403 out:
 404         return ret;
 405 }
 406 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
 407 asmlinkage long SyS_sync_file_range(long fd, loff_t offset, loff_t nbytes,
 408                                     long flags)
 409 {
 410         return SYSC_sync_file_range((int) fd, offset, nbytes,
 411                                     (unsigned int) flags);
 412 }
 413 SYSCALL_ALIAS(sys_sync_file_range, SyS_sync_file_range);
 414 #endif
 415
 416 /* It would be nice if people remember that not all the world's an i386
 417    when they introduce new system calls */
 418 SYSCALL_DEFINE(sync_file_range2)(int fd, unsigned int flags,
 419                                  loff_t offset, loff_t nbytes)
 420 {
 421         return sys_sync_file_range(fd, offset, nbytes, flags);
 422 }
 423 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
 424 asmlinkage long SyS_sync_file_range2(long fd, long flags,
 425                                      loff_t offset, loff_t nbytes)
 426 {
 427         return SYSC_sync_file_range2((int) fd, (unsigned int) flags,
 428                                      offset, nbytes);
 429 }
 430 SYSCALL_ALIAS(sys_sync_file_range2, SyS_sync_file_range2);
 431 #endif