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 sync_inodes_sb(sb, 0)
  23  * just dirties buffers with inodes so we have to submit IO for these buffers
  24  * via __sync_blockdev(). This also speeds up the wait == 1 case since in that
  25  * case write_inode() functions do sync_dirty_buffer() and thus effectively
  26  * 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         else
  34                 sync_quota_sb(sb, -1);
  35         sync_inodes_sb(sb, wait);
  36         if (sb->s_dirt && sb->s_op->write_super)
  37                 sb->s_op->write_super(sb);
  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 SYSCALL_DEFINE0(sync)
 118 {
 119         sync_filesystems(0);
 120         sync_filesystems(1);
 121         if (unlikely(laptop_mode))
 122                 laptop_sync_completion();
 123         return 0;
 124 }
 125
 126 static void do_sync_work(struct work_struct *work)
 127 {
 128         /*
 129          * Sync twice to reduce the possibility we skipped some inodes / pages
 130          * because they were temporarily locked
 131          */
 132         sync_filesystems(0);
 133         sync_filesystems(0);
 134         printk("Emergency Sync complete\n");
 135         kfree(work);
 136 }
 137
 138 void emergency_sync(void)
 139 {
 140         struct work_struct *work;
 141
 142         work = kmalloc(sizeof(*work), GFP_ATOMIC);
 143         if (work) {
 144                 INIT_WORK(work, do_sync_work);
 145                 schedule_work(work);
 146         }
 147 }
 148
 149 /*
 150  * Generic function to fsync a file.
 151  *
 152  * filp may be NULL if called via the msync of a vma.
 153  */
 154 int file_fsync(struct file *filp, struct dentry *dentry, int datasync)
 155 {
 156         struct inode * inode = dentry->d_inode;
 157         struct super_block * sb;
 158         int ret, err;
 159
 160         /* sync the inode to buffers */
 161         ret = write_inode_now(inode, 0);
 162
 163         /* sync the superblock to buffers */
 164         sb = inode->i_sb;
 165         if (sb->s_dirt && sb->s_op->write_super)
 166                 sb->s_op->write_super(sb);
 167
 168         /* .. finally sync the buffers to disk */
 169         err = sync_blockdev(sb->s_bdev);
 170         if (!ret)
 171                 ret = err;
 172         return ret;
 173 }
 174
 175 /**
 176  * vfs_fsync - perform a fsync or fdatasync on a file
 177  * @file:               file to sync
 178  * @dentry:             dentry of @file
 179  * @data:               only perform a fdatasync operation
 180  *
 181  * Write back data and metadata for @file to disk.  If @datasync is
 182  * set only metadata needed to access modified file data is written.
 183  *
 184  * In case this function is called from nfsd @file may be %NULL and
 185  * only @dentry is set.  This can only happen when the filesystem
 186  * implements the export_operations API.
 187  */
 188 int vfs_fsync(struct file *file, struct dentry *dentry, int datasync)
 189 {
 190         const struct file_operations *fop;
 191         struct address_space *mapping;
 192         int err, ret;
 193
 194         /*
 195          * Get mapping and operations from the file in case we have
 196          * as file, or get the default values for them in case we
 197          * don't have a struct file available.  Damn nfsd..
 198          */
 199         if (file) {
 200                 mapping = file->f_mapping;
 201                 fop = file->f_op;
 202         } else {
 203                 mapping = dentry->d_inode->i_mapping;
 204                 fop = dentry->d_inode->i_fop;
 205         }
 206
 207         if (!fop || !fop->fsync) {
 208                 ret = -EINVAL;
 209                 goto out;
 210         }
 211
 212         ret = filemap_fdatawrite(mapping);
 213
 214         /*
 215          * We need to protect against concurrent writers, which could cause
 216          * livelocks in fsync_buffers_list().
 217          */
 218         mutex_lock(&mapping->host->i_mutex);
 219         err = fop->fsync(file, dentry, datasync);
 220         if (!ret)
 221                 ret = err;
 222         mutex_unlock(&mapping->host->i_mutex);
 223         err = filemap_fdatawait(mapping);
 224         if (!ret)
 225                 ret = err;
 226 out:
 227         return ret;
 228 }
 229 EXPORT_SYMBOL(vfs_fsync);
 230
 231 static int do_fsync(unsigned int fd, int datasync)
 232 {
 233         struct file *file;
 234         int ret = -EBADF;
 235
 236         file = fget(fd);
 237         if (file) {
 238                 ret = vfs_fsync(file, file->f_path.dentry, datasync);
 239                 fput(file);
 240         }
 241         return ret;
 242 }
 243
 244 SYSCALL_DEFINE1(fsync, unsigned int, fd)
 245 {
 246         return do_fsync(fd, 0);
 247 }
 248
 249 SYSCALL_DEFINE1(fdatasync, unsigned int, fd)
 250 {
 251         return do_fsync(fd, 1);
 252 }
 253
 254 /*
 255  * sys_sync_file_range() permits finely controlled syncing over a segment of
 256  * a file in the range offset .. (offset+nbytes-1) inclusive.  If nbytes is
 257  * zero then sys_sync_file_range() will operate from offset out to EOF.
 258  *
 259  * The flag bits are:
 260  *
 261  * SYNC_FILE_RANGE_WAIT_BEFORE: wait upon writeout of all pages in the range
 262  * before performing the write.
 263  *
 264  * SYNC_FILE_RANGE_WRITE: initiate writeout of all those dirty pages in the
 265  * range which are not presently under writeback. Note that this may block for
 266  * significant periods due to exhaustion of disk request structures.
 267  *
 268  * SYNC_FILE_RANGE_WAIT_AFTER: wait upon writeout of all pages in the range
 269  * after performing the write.
 270  *
 271  * Useful combinations of the flag bits are:
 272  *
 273  * SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE: ensures that all pages
 274  * in the range which were dirty on entry to sys_sync_file_range() are placed
 275  * under writeout.  This is a start-write-for-data-integrity operation.
 276  *
 277  * SYNC_FILE_RANGE_WRITE: start writeout of all dirty pages in the range which
 278  * are not presently under writeout.  This is an asynchronous flush-to-disk
 279  * operation.  Not suitable for data integrity operations.
 280  *
 281  * SYNC_FILE_RANGE_WAIT_BEFORE (or SYNC_FILE_RANGE_WAIT_AFTER): wait for
 282  * completion of writeout of all pages in the range.  This will be used after an
 283  * earlier SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE operation to wait
 284  * for that operation to complete and to return the result.
 285  *
 286  * SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE|SYNC_FILE_RANGE_WAIT_AFTER:
 287  * a traditional sync() operation.  This is a write-for-data-integrity operation
 288  * which will ensure that all pages in the range which were dirty on entry to
 289  * sys_sync_file_range() are committed to disk.
 290  *
 291  *
 292  * SYNC_FILE_RANGE_WAIT_BEFORE and SYNC_FILE_RANGE_WAIT_AFTER will detect any
 293  * I/O errors or ENOSPC conditions and will return those to the caller, after
 294  * clearing the EIO and ENOSPC flags in the address_space.
 295  *
 296  * It should be noted that none of these operations write out the file's
 297  * metadata.  So unless the application is strictly performing overwrites of
 298  * already-instantiated disk blocks, there are no guarantees here that the data
 299  * will be available after a crash.
 300  */
 301 SYSCALL_DEFINE(sync_file_range)(int fd, loff_t offset, loff_t nbytes,
 302                                 unsigned int flags)
 303 {
 304         int ret;
 305         struct file *file;
 306         loff_t endbyte;                 /* inclusive */
 307         int fput_needed;
 308         umode_t i_mode;
 309
 310         ret = -EINVAL;
 311         if (flags & ~VALID_FLAGS)
 312                 goto out;
 313
 314         endbyte = offset + nbytes;
 315
 316         if ((s64)offset < 0)
 317                 goto out;
 318         if ((s64)endbyte < 0)
 319                 goto out;
 320         if (endbyte < offset)
 321                 goto out;
 322
 323         if (sizeof(pgoff_t) == 4) {
 324                 if (offset >= (0x100000000ULL << PAGE_CACHE_SHIFT)) {
 325                         /*
 326                          * The range starts outside a 32 bit machine's
 327                          * pagecache addressing capabilities.  Let it "succeed"
 328                          */
 329                         ret = 0;
 330                         goto out;
 331                 }
 332                 if (endbyte >= (0x100000000ULL << PAGE_CACHE_SHIFT)) {
 333                         /*
 334                          * Out to EOF
 335                          */
 336                         nbytes = 0;
 337                 }
 338         }
 339
 340         if (nbytes == 0)
 341                 endbyte = LLONG_MAX;
 342         else
 343                 endbyte--;              /* inclusive */
 344
 345         ret = -EBADF;
 346         file = fget_light(fd, &fput_needed);
 347         if (!file)
 348                 goto out;
 349
 350         i_mode = file->f_path.dentry->d_inode->i_mode;
 351         ret = -ESPIPE;
 352         if (!S_ISREG(i_mode) && !S_ISBLK(i_mode) && !S_ISDIR(i_mode) &&
 353                         !S_ISLNK(i_mode))
 354                 goto out_put;
 355
 356         ret = do_sync_mapping_range(file->f_mapping, offset, endbyte, flags);
 357 out_put:
 358         fput_light(file, fput_needed);
 359 out:
 360         return ret;
 361 }
 362 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
 363 asmlinkage long SyS_sync_file_range(long fd, loff_t offset, loff_t nbytes,
 364                                     long flags)
 365 {
 366         return SYSC_sync_file_range((int) fd, offset, nbytes,
 367                                     (unsigned int) flags);
 368 }
 369 SYSCALL_ALIAS(sys_sync_file_range, SyS_sync_file_range);
 370 #endif
 371
 372 /* It would be nice if people remember that not all the world's an i386
 373    when they introduce new system calls */
 374 SYSCALL_DEFINE(sync_file_range2)(int fd, unsigned int flags,
 375                                  loff_t offset, loff_t nbytes)
 376 {
 377         return sys_sync_file_range(fd, offset, nbytes, flags);
 378 }
 379 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
 380 asmlinkage long SyS_sync_file_range2(long fd, long flags,
 381                                      loff_t offset, loff_t nbytes)
 382 {
 383         return SYSC_sync_file_range2((int) fd, (unsigned int) flags,
 384                                      offset, nbytes);
 385 }
 386 SYSCALL_ALIAS(sys_sync_file_range2, SyS_sync_file_range2);
 387 #endif
 388
 389 /*
 390  * `endbyte' is inclusive
 391  */
 392 int do_sync_mapping_range(struct address_space *mapping, loff_t offset,
 393                           loff_t endbyte, unsigned int flags)
 394 {
 395         int ret;
 396
 397         if (!mapping) {
 398                 ret = -EINVAL;
 399                 goto out;
 400         }
 401
 402         ret = 0;
 403         if (flags & SYNC_FILE_RANGE_WAIT_BEFORE) {
 404                 ret = wait_on_page_writeback_range(mapping,
 405                                         offset >> PAGE_CACHE_SHIFT,
 406                                         endbyte >> PAGE_CACHE_SHIFT);
 407                 if (ret < 0)
 408                         goto out;
 409         }
 410
 411         if (flags & SYNC_FILE_RANGE_WRITE) {
 412                 ret = __filemap_fdatawrite_range(mapping, offset, endbyte,
 413                                                 WB_SYNC_ALL);
 414                 if (ret < 0)
 415                         goto out;
 416         }
 417
 418         if (flags & SYNC_FILE_RANGE_WAIT_AFTER) {
 419                 ret = wait_on_page_writeback_range(mapping,
 420                                         offset >> PAGE_CACHE_SHIFT,
 421                                         endbyte >> PAGE_CACHE_SHIFT);
 422         }
 423 out:
 424         return ret;
 425 }
 426 EXPORT_SYMBOL_GPL(do_sync_mapping_range);