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