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/module.h>
   9 #include <linux/sched.h>
  10 #include <linux/writeback.h>
  11 #include <linux/syscalls.h>
  12 #include <linux/linkage.h>
  13 #include <linux/pagemap.h>
  14 #include <linux/quotaops.h>
  15 #include <linux/buffer_head.h>
  16 #include "internal.h"
  17
  18 #define VALID_FLAGS (SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE| \
  19                         SYNC_FILE_RANGE_WAIT_AFTER)
  20
  21 /*
  22  * Do the filesystem syncing work. For simple filesystems
  23  * writeback_inodes_sb(sb) just dirties buffers with inodes so we have to
  24  * submit IO for these buffers via __sync_blockdev(). This also speeds up the
  25  * wait == 1 case since in that case write_inode() functions do
  26  * sync_dirty_buffer() and thus effectively write one block at a time.
  27  */
  28 static int __sync_filesystem(struct super_block *sb, int wait)
  29 {
  30         /* Avoid doing twice syncing and cache pruning for quota sync */
  31         if (!wait) {
  32                 writeout_quota_sb(sb, -1);
  33                 writeback_inodes_sb(sb);
  34         } else {
  35                 sync_quota_sb(sb, -1);
  36                 sync_inodes_sb(sb);
  37         }
  38         if (sb->s_op->sync_fs)
  39                 sb->s_op->sync_fs(sb, wait);
  40         return __sync_blockdev(sb->s_bdev, wait);
  41 }
  42
  43 /*
  44  * Write out and wait upon all dirty data associated with this
  45  * superblock.  Filesystem data as well as the underlying block
  46  * device.  Takes the superblock lock.
  47  */
  48 int sync_filesystem(struct super_block *sb)
  49 {
  50         int ret;
  51
  52         /*
  53          * We need to be protected against the filesystem going from
  54          * r/o to r/w or vice versa.
  55          */
  56         WARN_ON(!rwsem_is_locked(&sb->s_umount));
  57
  58         /*
  59          * No point in syncing out anything if the filesystem is read-only.
  60          */
  61         if (sb->s_flags & MS_RDONLY)
  62                 return 0;
  63
  64         ret = __sync_filesystem(sb, 0);
  65         if (ret < 0)
  66                 return ret;
  67         return __sync_filesystem(sb, 1);
  68 }
  69 EXPORT_SYMBOL_GPL(sync_filesystem);
  70
  71 /*
  72  * Sync all the data for all the filesystems (called by sys_sync() and
  73  * emergency sync)
  74  *
  75  * This operation is careful to avoid the livelock which could easily happen
  76  * if two or more filesystems are being continuously dirtied.  s_need_sync
  77  * is used only here.  We set it against all filesystems and then clear it as
  78  * we sync them.  So redirtied filesystems are skipped.
  79  *
  80  * But if process A is currently running sync_filesystems and then process B
  81  * calls sync_filesystems as well, process B will set all the s_need_sync
  82  * flags again, which will cause process A to resync everything.  Fix that with
  83  * a local mutex.
  84  */
  85 static void sync_filesystems(int wait)
  86 {
  87         struct super_block *sb;
  88         static DEFINE_MUTEX(mutex);
  89
  90         mutex_lock(&mutex);             /* Could be down_interruptible */
  91         spin_lock(&sb_lock);
  92         list_for_each_entry(sb, &super_blocks, s_list)
  93                 sb->s_need_sync = 1;
  94
  95 restart:
  96         list_for_each_entry(sb, &super_blocks, s_list) {
  97                 if (!sb->s_need_sync)
  98                         continue;
  99                 sb->s_need_sync = 0;
 100                 sb->s_count++;
 101                 spin_unlock(&sb_lock);
 102
 103                 down_read(&sb->s_umount);
 104                 if (!(sb->s_flags & MS_RDONLY) && sb->s_root)
 105                         __sync_filesystem(sb, wait);
 106                 up_read(&sb->s_umount);
 107
 108                 /* restart only when sb is no longer on the list */
 109                 spin_lock(&sb_lock);
 110                 if (__put_super_and_need_restart(sb))
 111                         goto restart;
 112         }
 113         spin_unlock(&sb_lock);
 114         mutex_unlock(&mutex);
 115 }
 116
 117 /*
 118  * sync everything.  Start out by waking pdflush, because that writes back
 119  * all queues in parallel.
 120  */
 121 SYSCALL_DEFINE0(sync)
 122 {
 123         wakeup_flusher_threads(0);
 124         sync_filesystems(0);
 125         sync_filesystems(1);
 126         if (unlikely(laptop_mode))
 127                 laptop_sync_completion();
 128         return 0;
 129 }
 130
 131 static void do_sync_work(struct work_struct *work)
 132 {
 133         /*
 134          * Sync twice to reduce the possibility we skipped some inodes / pages
 135          * because they were temporarily locked
 136          */
 137         sync_filesystems(0);
 138         sync_filesystems(0);
 139         printk("Emergency Sync complete\n");
 140         kfree(work);
 141 }
 142
 143 void emergency_sync(void)
 144 {
 145         struct work_struct *work;
 146
 147         work = kmalloc(sizeof(*work), GFP_ATOMIC);
 148         if (work) {
 149                 INIT_WORK(work, do_sync_work);
 150                 schedule_work(work);
 151         }
 152 }
 153
 154 /*
 155  * Generic function to fsync a file.
 156  *
 157  * filp may be NULL if called via the msync of a vma.
 158  */
 159 int file_fsync(struct file *filp, struct dentry *dentry, int datasync)
 160 {
 161         struct inode * inode = dentry->d_inode;
 162         struct super_block * sb;
 163         int ret, err;
 164
 165         /* sync the inode to buffers */
 166         ret = write_inode_now(inode, 0);
 167
 168         /* sync the superblock to buffers */
 169         sb = inode->i_sb;
 170         if (sb->s_dirt && sb->s_op->write_super)
 171                 sb->s_op->write_super(sb);
 172
 173         /* .. finally sync the buffers to disk */
 174         err = sync_blockdev(sb->s_bdev);
 175         if (!ret)
 176                 ret = err;
 177         return ret;
 178 }
 179
 180 /**
 181  * vfs_fsync - perform a fsync or fdatasync on a file
 182  * @file:               file to sync
 183  * @dentry:             dentry of @file
 184  * @data:               only perform a fdatasync operation
 185  *
 186  * Write back data and metadata for @file to disk.  If @datasync is
 187  * set only metadata needed to access modified file data is written.
 188  *
 189  * In case this function is called from nfsd @file may be %NULL and
 190  * only @dentry is set.  This can only happen when the filesystem
 191  * implements the export_operations API.
 192  */
 193 int vfs_fsync(struct file *file, struct dentry *dentry, int datasync)
 194 {
 195         const struct file_operations *fop;
 196         struct address_space *mapping;
 197         int err, ret;
 198
 199         /*
 200          * Get mapping and operations from the file in case we have
 201          * as file, or get the default values for them in case we
 202          * don't have a struct file available.  Damn nfsd..
 203          */
 204         if (file) {
 205                 mapping = file->f_mapping;
 206                 fop = file->f_op;
 207         } else {
 208                 mapping = dentry->d_inode->i_mapping;
 209                 fop = dentry->d_inode->i_fop;
 210         }
 211
 212         if (!fop || !fop->fsync) {
 213                 ret = -EINVAL;
 214                 goto out;
 215         }
 216
 217         ret = filemap_fdatawrite(mapping);
 218
 219         /*
 220          * We need to protect against concurrent writers, which could cause
 221          * livelocks in fsync_buffers_list().
 222          */
 223         mutex_lock(&mapping->host->i_mutex);
 224         err = fop->fsync(file, dentry, datasync);
 225         if (!ret)
 226                 ret = err;
 227         mutex_unlock(&mapping->host->i_mutex);
 228         err = filemap_fdatawait(mapping);
 229         if (!ret)
 230                 ret = err;
 231 out:
 232         return ret;
 233 }
 234 EXPORT_SYMBOL(vfs_fsync);
 235
 236 static int do_fsync(unsigned int fd, int datasync)
 237 {
 238         struct file *file;
 239         int ret = -EBADF;
 240
 241         file = fget(fd);
 242         if (file) {
 243                 ret = vfs_fsync(file, file->f_path.dentry, datasync);
 244                 fput(file);
 245         }
 246         return ret;
 247 }
 248
 249 SYSCALL_DEFINE1(fsync, unsigned int, fd)
 250 {
 251         return do_fsync(fd, 0);
 252 }
 253
 254 SYSCALL_DEFINE1(fdatasync, unsigned int, fd)
 255 {
 256         return do_fsync(fd, 1);
 257 }
 258
 259 /*
 260  * sys_sync_file_range() permits finely controlled syncing over a segment of
 261  * a file in the range offset .. (offset+nbytes-1) inclusive.  If nbytes is
 262  * zero then sys_sync_file_range() will operate from offset out to EOF.
 263  *
 264  * The flag bits are:
 265  *
 266  * SYNC_FILE_RANGE_WAIT_BEFORE: wait upon writeout of all pages in the range
 267  * before performing the write.
 268  *
 269  * SYNC_FILE_RANGE_WRITE: initiate writeout of all those dirty pages in the
 270  * range which are not presently under writeback. Note that this may block for
 271  * significant periods due to exhaustion of disk request structures.
 272  *
 273  * SYNC_FILE_RANGE_WAIT_AFTER: wait upon writeout of all pages in the range
 274  * after performing the write.
 275  *
 276  * Useful combinations of the flag bits are:
 277  *
 278  * SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE: ensures that all pages
 279  * in the range which were dirty on entry to sys_sync_file_range() are placed
 280  * under writeout.  This is a start-write-for-data-integrity operation.
 281  *
 282  * SYNC_FILE_RANGE_WRITE: start writeout of all dirty pages in the range which
 283  * are not presently under writeout.  This is an asynchronous flush-to-disk
 284  * operation.  Not suitable for data integrity operations.
 285  *
 286  * SYNC_FILE_RANGE_WAIT_BEFORE (or SYNC_FILE_RANGE_WAIT_AFTER): wait for
 287  * completion of writeout of all pages in the range.  This will be used after an
 288  * earlier SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE operation to wait
 289  * for that operation to complete and to return the result.
 290  *
 291  * SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE|SYNC_FILE_RANGE_WAIT_AFTER:
 292  * a traditional sync() operation.  This is a write-for-data-integrity operation
 293  * which will ensure that all pages in the range which were dirty on entry to
 294  * sys_sync_file_range() are committed to disk.
 295  *
 296  *
 297  * SYNC_FILE_RANGE_WAIT_BEFORE and SYNC_FILE_RANGE_WAIT_AFTER will detect any
 298  * I/O errors or ENOSPC conditions and will return those to the caller, after
 299  * clearing the EIO and ENOSPC flags in the address_space.
 300  *
 301  * It should be noted that none of these operations write out the file's
 302  * metadata.  So unless the application is strictly performing overwrites of
 303  * already-instantiated disk blocks, there are no guarantees here that the data
 304  * will be available after a crash.
 305  */
 306 SYSCALL_DEFINE(sync_file_range)(int fd, loff_t offset, loff_t nbytes,
 307                                 unsigned int flags)
 308 {
 309         int ret;
 310         struct file *file;
 311         loff_t endbyte;                 /* inclusive */
 312         int fput_needed;
 313         umode_t i_mode;
 314
 315         ret = -EINVAL;
 316         if (flags & ~VALID_FLAGS)
 317                 goto out;
 318
 319         endbyte = offset + nbytes;
 320
 321         if ((s64)offset < 0)
 322                 goto out;
 323         if ((s64)endbyte < 0)
 324                 goto out;
 325         if (endbyte < offset)
 326                 goto out;
 327
 328         if (sizeof(pgoff_t) == 4) {
 329                 if (offset >= (0x100000000ULL << PAGE_CACHE_SHIFT)) {
 330                         /*
 331                          * The range starts outside a 32 bit machine's
 332                          * pagecache addressing capabilities.  Let it "succeed"
 333                          */
 334                         ret = 0;
 335                         goto out;
 336                 }
 337                 if (endbyte >= (0x100000000ULL << PAGE_CACHE_SHIFT)) {
 338                         /*
 339                          * Out to EOF
 340                          */
 341                         nbytes = 0;
 342                 }
 343         }
 344
 345         if (nbytes == 0)
 346                 endbyte = LLONG_MAX;
 347         else
 348                 endbyte--;              /* inclusive */
 349
 350         ret = -EBADF;
 351         file = fget_light(fd, &fput_needed);
 352         if (!file)
 353                 goto out;
 354
 355         i_mode = file->f_path.dentry->d_inode->i_mode;
 356         ret = -ESPIPE;
 357         if (!S_ISREG(i_mode) && !S_ISBLK(i_mode) && !S_ISDIR(i_mode) &&
 358                         !S_ISLNK(i_mode))
 359                 goto out_put;
 360
 361         ret = do_sync_mapping_range(file->f_mapping, offset, endbyte, flags);
 362 out_put:
 363         fput_light(file, fput_needed);
 364 out:
 365         return ret;
 366 }
 367 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
 368 asmlinkage long SyS_sync_file_range(long fd, loff_t offset, loff_t nbytes,
 369                                     long flags)
 370 {
 371         return SYSC_sync_file_range((int) fd, offset, nbytes,
 372                                     (unsigned int) flags);
 373 }
 374 SYSCALL_ALIAS(sys_sync_file_range, SyS_sync_file_range);
 375 #endif
 376
 377 /* It would be nice if people remember that not all the world's an i386
 378    when they introduce new system calls */
 379 SYSCALL_DEFINE(sync_file_range2)(int fd, unsigned int flags,
 380                                  loff_t offset, loff_t nbytes)
 381 {
 382         return sys_sync_file_range(fd, offset, nbytes, flags);
 383 }
 384 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
 385 asmlinkage long SyS_sync_file_range2(long fd, long flags,
 386                                      loff_t offset, loff_t nbytes)
 387 {
 388         return SYSC_sync_file_range2((int) fd, (unsigned int) flags,
 389                                      offset, nbytes);
 390 }
 391 SYSCALL_ALIAS(sys_sync_file_range2, SyS_sync_file_range2);
 392 #endif
 393
 394 /*
 395  * `endbyte' is inclusive
 396  */
 397 int do_sync_mapping_range(struct address_space *mapping, loff_t offset,
 398                           loff_t endbyte, unsigned int flags)
 399 {
 400         int ret;
 401
 402         if (!mapping) {
 403                 ret = -EINVAL;
 404                 goto out;
 405         }
 406
 407         ret = 0;
 408         if (flags & SYNC_FILE_RANGE_WAIT_BEFORE) {
 409                 ret = wait_on_page_writeback_range(mapping,
 410                                         offset >> PAGE_CACHE_SHIFT,
 411                                         endbyte >> PAGE_CACHE_SHIFT);
 412                 if (ret < 0)
 413                         goto out;
 414         }
 415
 416         if (flags & SYNC_FILE_RANGE_WRITE) {
 417                 ret = __filemap_fdatawrite_range(mapping, offset, endbyte,
 418                                                 WB_SYNC_ALL);
 419                 if (ret < 0)
 420                         goto out;
 421         }
 422
 423         if (flags & SYNC_FILE_RANGE_WAIT_AFTER) {
 424                 ret = wait_on_page_writeback_range(mapping,
 425                                         offset >> PAGE_CACHE_SHIFT,
 426                                         endbyte >> PAGE_CACHE_SHIFT);
 427         }
 428 out:
 429         return ret;
 430 }
 431 EXPORT_SYMBOL_GPL(do_sync_mapping_range);