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