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 SYSCALL_DEFINE0(sync)
 116 {
 117         sync_filesystems(0);
 118         sync_filesystems(1);
 119         if (unlikely(laptop_mode))
 120                 laptop_sync_completion();
 121         return 0;
 122 }
 123
 124 static void do_sync_work(struct work_struct *work)
 125 {
 126         /*
 127          * Sync twice to reduce the possibility we skipped some inodes / pages
 128          * because they were temporarily locked
 129          */
 130         sync_filesystems(0);
 131         sync_filesystems(0);
 132         printk("Emergency Sync complete\n");
 133         kfree(work);
 134 }
 135
 136 void emergency_sync(void)
 137 {
 138         struct work_struct *work;
 139
 140         work = kmalloc(sizeof(*work), GFP_ATOMIC);
 141         if (work) {
 142                 INIT_WORK(work, do_sync_work);
 143                 schedule_work(work);
 144         }
 145 }
 146
 147 /*
 148  * Generic function to fsync a file.
 149  *
 150  * filp may be NULL if called via the msync of a vma.
 151  */
 152 int file_fsync(struct file *filp, struct dentry *dentry, int datasync)
 153 {
 154         struct inode * inode = dentry->d_inode;
 155         struct super_block * sb;
 156         int ret, err;
 157
 158         /* sync the inode to buffers */
 159         ret = write_inode_now(inode, 0);
 160
 161         /* sync the superblock to buffers */
 162         sb = inode->i_sb;
 163         if (sb->s_dirt && sb->s_op->write_super)
 164                 sb->s_op->write_super(sb);
 165
 166         /* .. finally sync the buffers to disk */
 167         err = sync_blockdev(sb->s_bdev);
 168         if (!ret)
 169                 ret = err;
 170         return ret;
 171 }
 172
 173 /**
 174  * vfs_fsync - perform a fsync or fdatasync on a file
 175  * @file:               file to sync
 176  * @dentry:             dentry of @file
 177  * @data:               only perform a fdatasync operation
 178  *
 179  * Write back data and metadata for @file to disk.  If @datasync is
 180  * set only metadata needed to access modified file data is written.
 181  *
 182  * In case this function is called from nfsd @file may be %NULL and
 183  * only @dentry is set.  This can only happen when the filesystem
 184  * implements the export_operations API.
 185  */
 186 int vfs_fsync(struct file *file, struct dentry *dentry, int datasync)
 187 {
 188         const struct file_operations *fop;
 189         struct address_space *mapping;
 190         int err, ret;
 191
 192         /*
 193          * Get mapping and operations from the file in case we have
 194          * as file, or get the default values for them in case we
 195          * don't have a struct file available.  Damn nfsd..
 196          */
 197         if (file) {
 198                 mapping = file->f_mapping;
 199                 fop = file->f_op;
 200         } else {
 201                 mapping = dentry->d_inode->i_mapping;
 202                 fop = dentry->d_inode->i_fop;
 203         }
 204
 205         if (!fop || !fop->fsync) {
 206                 ret = -EINVAL;
 207                 goto out;
 208         }
 209
 210         ret = filemap_fdatawrite(mapping);
 211
 212         /*
 213          * We need to protect against concurrent writers, which could cause
 214          * livelocks in fsync_buffers_list().
 215          */
 216         mutex_lock(&mapping->host->i_mutex);
 217         err = fop->fsync(file, dentry, datasync);
 218         if (!ret)
 219                 ret = err;
 220         mutex_unlock(&mapping->host->i_mutex);
 221         err = filemap_fdatawait(mapping);
 222         if (!ret)
 223                 ret = err;
 224 out:
 225         return ret;
 226 }
 227 EXPORT_SYMBOL(vfs_fsync);
 228
 229 static int do_fsync(unsigned int fd, int datasync)
 230 {
 231         struct file *file;
 232         int ret = -EBADF;
 233
 234         file = fget(fd);
 235         if (file) {
 236                 ret = vfs_fsync(file, file->f_path.dentry, datasync);
 237                 fput(file);
 238         }
 239         return ret;
 240 }
 241
 242 SYSCALL_DEFINE1(fsync, unsigned int, fd)
 243 {
 244         return do_fsync(fd, 0);
 245 }
 246
 247 SYSCALL_DEFINE1(fdatasync, unsigned int, fd)
 248 {
 249         return do_fsync(fd, 1);
 250 }
 251
 252 /*
 253  * sys_sync_file_range() permits finely controlled syncing over a segment of
 254  * a file in the range offset .. (offset+nbytes-1) inclusive.  If nbytes is
 255  * zero then sys_sync_file_range() will operate from offset out to EOF.
 256  *
 257  * The flag bits are:
 258  *
 259  * SYNC_FILE_RANGE_WAIT_BEFORE: wait upon writeout of all pages in the range
 260  * before performing the write.
 261  *
 262  * SYNC_FILE_RANGE_WRITE: initiate writeout of all those dirty pages in the
 263  * range which are not presently under writeback. Note that this may block for
 264  * significant periods due to exhaustion of disk request structures.
 265  *
 266  * SYNC_FILE_RANGE_WAIT_AFTER: wait upon writeout of all pages in the range
 267  * after performing the write.
 268  *
 269  * Useful combinations of the flag bits are:
 270  *
 271  * SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE: ensures that all pages
 272  * in the range which were dirty on entry to sys_sync_file_range() are placed
 273  * under writeout.  This is a start-write-for-data-integrity operation.
 274  *
 275  * SYNC_FILE_RANGE_WRITE: start writeout of all dirty pages in the range which
 276  * are not presently under writeout.  This is an asynchronous flush-to-disk
 277  * operation.  Not suitable for data integrity operations.
 278  *
 279  * SYNC_FILE_RANGE_WAIT_BEFORE (or SYNC_FILE_RANGE_WAIT_AFTER): wait for
 280  * completion of writeout of all pages in the range.  This will be used after an
 281  * earlier SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE operation to wait
 282  * for that operation to complete and to return the result.
 283  *
 284  * SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE|SYNC_FILE_RANGE_WAIT_AFTER:
 285  * a traditional sync() operation.  This is a write-for-data-integrity operation
 286  * which will ensure that all pages in the range which were dirty on entry to
 287  * sys_sync_file_range() are committed to disk.
 288  *
 289  *
 290  * SYNC_FILE_RANGE_WAIT_BEFORE and SYNC_FILE_RANGE_WAIT_AFTER will detect any
 291  * I/O errors or ENOSPC conditions and will return those to the caller, after
 292  * clearing the EIO and ENOSPC flags in the address_space.
 293  *
 294  * It should be noted that none of these operations write out the file's
 295  * metadata.  So unless the application is strictly performing overwrites of
 296  * already-instantiated disk blocks, there are no guarantees here that the data
 297  * will be available after a crash.
 298  */
 299 SYSCALL_DEFINE(sync_file_range)(int fd, loff_t offset, loff_t nbytes,
 300                                 unsigned int flags)
 301 {
 302         int ret;
 303         struct file *file;
 304         loff_t endbyte;                 /* inclusive */
 305         int fput_needed;
 306         umode_t i_mode;
 307
 308         ret = -EINVAL;
 309         if (flags & ~VALID_FLAGS)
 310                 goto out;
 311
 312         endbyte = offset + nbytes;
 313
 314         if ((s64)offset < 0)
 315                 goto out;
 316         if ((s64)endbyte < 0)
 317                 goto out;
 318         if (endbyte < offset)
 319                 goto out;
 320
 321         if (sizeof(pgoff_t) == 4) {
 322                 if (offset >= (0x100000000ULL << PAGE_CACHE_SHIFT)) {
 323                         /*
 324                          * The range starts outside a 32 bit machine's
 325                          * pagecache addressing capabilities.  Let it "succeed"
 326                          */
 327                         ret = 0;
 328                         goto out;
 329                 }
 330                 if (endbyte >= (0x100000000ULL << PAGE_CACHE_SHIFT)) {
 331                         /*
 332                          * Out to EOF
 333                          */
 334                         nbytes = 0;
 335                 }
 336         }
 337
 338         if (nbytes == 0)
 339                 endbyte = LLONG_MAX;
 340         else
 341                 endbyte--;              /* inclusive */
 342
 343         ret = -EBADF;
 344         file = fget_light(fd, &fput_needed);
 345         if (!file)
 346                 goto out;
 347
 348         i_mode = file->f_path.dentry->d_inode->i_mode;
 349         ret = -ESPIPE;
 350         if (!S_ISREG(i_mode) && !S_ISBLK(i_mode) && !S_ISDIR(i_mode) &&
 351                         !S_ISLNK(i_mode))
 352                 goto out_put;
 353
 354         ret = do_sync_mapping_range(file->f_mapping, offset, endbyte, flags);
 355 out_put:
 356         fput_light(file, fput_needed);
 357 out:
 358         return ret;
 359 }
 360 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
 361 asmlinkage long SyS_sync_file_range(long fd, loff_t offset, loff_t nbytes,
 362                                     long flags)
 363 {
 364         return SYSC_sync_file_range((int) fd, offset, nbytes,
 365                                     (unsigned int) flags);
 366 }
 367 SYSCALL_ALIAS(sys_sync_file_range, SyS_sync_file_range);
 368 #endif
 369
 370 /* It would be nice if people remember that not all the world's an i386
 371    when they introduce new system calls */
 372 SYSCALL_DEFINE(sync_file_range2)(int fd, unsigned int flags,
 373                                  loff_t offset, loff_t nbytes)
 374 {
 375         return sys_sync_file_range(fd, offset, nbytes, flags);
 376 }
 377 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
 378 asmlinkage long SyS_sync_file_range2(long fd, long flags,
 379                                      loff_t offset, loff_t nbytes)
 380 {
 381         return SYSC_sync_file_range2((int) fd, (unsigned int) flags,
 382                                      offset, nbytes);
 383 }
 384 SYSCALL_ALIAS(sys_sync_file_range2, SyS_sync_file_range2);
 385 #endif
 386
 387 /*
 388  * `endbyte' is inclusive
 389  */
 390 int do_sync_mapping_range(struct address_space *mapping, loff_t offset,
 391                           loff_t endbyte, unsigned int flags)
 392 {
 393         int ret;
 394
 395         if (!mapping) {
 396                 ret = -EINVAL;
 397                 goto out;
 398         }
 399
 400         ret = 0;
 401         if (flags & SYNC_FILE_RANGE_WAIT_BEFORE) {
 402                 ret = wait_on_page_writeback_range(mapping,
 403                                         offset >> PAGE_CACHE_SHIFT,
 404                                         endbyte >> PAGE_CACHE_SHIFT);
 405                 if (ret < 0)
 406                         goto out;
 407         }
 408
 409         if (flags & SYNC_FILE_RANGE_WRITE) {
 410                 ret = __filemap_fdatawrite_range(mapping, offset, endbyte,
 411                                                 WB_SYNC_ALL);
 412                 if (ret < 0)
 413                         goto out;
 414         }
 415
 416         if (flags & SYNC_FILE_RANGE_WAIT_AFTER) {
 417                 ret = wait_on_page_writeback_range(mapping,
 418                                         offset >> PAGE_CACHE_SHIFT,
 419                                         endbyte >> PAGE_CACHE_SHIFT);
 420         }
 421 out:
 422         return ret;
 423 }
 424 EXPORT_SYMBOL_GPL(do_sync_mapping_range);