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/slab.h>
   9 #include <linux/export.h>
  10 #include <linux/namei.h>
  11 #include <linux/sched.h>
  12 #include <linux/writeback.h>
  13 #include <linux/syscalls.h>
  14 #include <linux/linkage.h>
  15 #include <linux/pagemap.h>
  16 #include <linux/quotaops.h>
  17 #include <linux/backing-dev.h>
  18 #include "internal.h"
  19
  20 #define VALID_FLAGS (SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE| \
  21                         SYNC_FILE_RANGE_WAIT_AFTER)
  22
  23 /*
  24  * Do the filesystem syncing work. For simple filesystems
  25  * writeback_inodes_sb(sb) just dirties buffers with inodes so we have to
  26  * submit IO for these buffers via __sync_blockdev(). This also speeds up the
  27  * wait == 1 case since in that case write_inode() functions do
  28  * sync_dirty_buffer() and thus effectively write one block at a time.
  29  */
  30 static int __sync_filesystem(struct super_block *sb, int wait)
  31 {
  32         /*
  33          * This should be safe, as we require bdi backing to actually
  34          * write out data in the first place
  35          */
  36         if (sb->s_bdi == &noop_backing_dev_info)
  37                 return 0;
  38
  39         if (sb->s_qcop && sb->s_qcop->quota_sync)
  40                 sb->s_qcop->quota_sync(sb, -1, wait);
  41
  42         if (wait)
  43                 sync_inodes_sb(sb);
  44         else
  45                 writeback_inodes_sb(sb, WB_REASON_SYNC);
  46
  47         if (sb->s_op->sync_fs)
  48                 sb->s_op->sync_fs(sb, wait);
  49         return __sync_blockdev(sb->s_bdev, wait);
  50 }
  51
  52 /*
  53  * Write out and wait upon all dirty data associated with this
  54  * superblock.  Filesystem data as well as the underlying block
  55  * device.  Takes the superblock lock.
  56  */
  57 int sync_filesystem(struct super_block *sb)
  58 {
  59         int ret;
  60
  61         /*
  62          * We need to be protected against the filesystem going from
  63          * r/o to r/w or vice versa.
  64          */
  65         WARN_ON(!rwsem_is_locked(&sb->s_umount));
  66
  67         /*
  68          * No point in syncing out anything if the filesystem is read-only.
  69          */
  70         if (sb->s_flags & MS_RDONLY)
  71                 return 0;
  72
  73         ret = __sync_filesystem(sb, 0);
  74         if (ret < 0)
  75                 return ret;
  76         return __sync_filesystem(sb, 1);
  77 }
  78 EXPORT_SYMBOL_GPL(sync_filesystem);
  79
  80 static void sync_one_sb(struct super_block *sb, void *arg)
  81 {
  82         if (!(sb->s_flags & MS_RDONLY))
  83                 __sync_filesystem(sb, *(int *)arg);
  84 }
  85 /*
  86  * Sync all the data for all the filesystems (called by sys_sync() and
  87  * emergency sync)
  88  */
  89 static void sync_filesystems(int wait)
  90 {
  91         iterate_supers(sync_one_sb, &wait);
  92 }
  93
  94 /*
  95  * sync everything.  Start out by waking pdflush, because that writes back
  96  * all queues in parallel.
  97  */
  98 SYSCALL_DEFINE0(sync)
  99 {
 100         wakeup_flusher_threads(0, WB_REASON_SYNC);
 101         sync_filesystems(0);
 102         sync_filesystems(1);
 103         if (unlikely(laptop_mode))
 104                 laptop_sync_completion();
 105         return 0;
 106 }
 107
 108 static void do_sync_work(struct work_struct *work)
 109 {
 110         /*
 111          * Sync twice to reduce the possibility we skipped some inodes / pages
 112          * because they were temporarily locked
 113          */
 114         sync_filesystems(0);
 115         sync_filesystems(0);
 116         printk("Emergency Sync complete\n");
 117         kfree(work);
 118 }
 119
 120 void emergency_sync(void)
 121 {
 122         struct work_struct *work;
 123
 124         work = kmalloc(sizeof(*work), GFP_ATOMIC);
 125         if (work) {
 126                 INIT_WORK(work, do_sync_work);
 127                 schedule_work(work);
 128         }
 129 }
 130
 131 /*
 132  * sync a single super
 133  */
 134 SYSCALL_DEFINE1(syncfs, int, fd)
 135 {
 136         struct file *file;
 137         struct super_block *sb;
 138         int ret;
 139         int fput_needed;
 140
 141         file = fget_light(fd, &fput_needed);
 142         if (!file)
 143                 return -EBADF;
 144         sb = file->f_dentry->d_sb;
 145
 146         down_read(&sb->s_umount);
 147         ret = sync_filesystem(sb);
 148         up_read(&sb->s_umount);
 149
 150         fput_light(file, fput_needed);
 151         return ret;
 152 }
 153
 154 /**
 155  * vfs_fsync_range - helper to sync a range of data & metadata to disk
 156  * @file:               file to sync
 157  * @start:              offset in bytes of the beginning of data range to sync
 158  * @end:                offset in bytes of the end of data range (inclusive)
 159  * @datasync:           perform only datasync
 160  *
 161  * Write back data in range @start..@end and metadata for @file to disk.  If
 162  * @datasync is set only metadata needed to access modified file data is
 163  * written.
 164  */
 165 int vfs_fsync_range(struct file *file, loff_t start, loff_t end, int datasync)
 166 {
 167         if (!file->f_op || !file->f_op->fsync)
 168                 return -EINVAL;
 169         return file->f_op->fsync(file, start, end, datasync);
 170 }
 171 EXPORT_SYMBOL(vfs_fsync_range);
 172
 173 /**
 174  * vfs_fsync - perform a fsync or fdatasync on a file
 175  * @file:               file to sync
 176  * @datasync:           only perform a fdatasync operation
 177  *
 178  * Write back data and metadata for @file to disk.  If @datasync is
 179  * set only metadata needed to access modified file data is written.
 180  */
 181 int vfs_fsync(struct file *file, int datasync)
 182 {
 183         return vfs_fsync_range(file, 0, LLONG_MAX, datasync);
 184 }
 185 EXPORT_SYMBOL(vfs_fsync);
 186
 187 static int do_fsync(unsigned int fd, int datasync)
 188 {
 189         struct file *file;
 190         int ret = -EBADF;
 191         int fput_needed;
 192
 193         file = fget_light(fd, &fput_needed);
 194         if (file) {
 195                 ret = vfs_fsync(file, datasync);
 196                 fput_light(file, fput_needed);
 197         }
 198         return ret;
 199 }
 200
 201 SYSCALL_DEFINE1(fsync, unsigned int, fd)
 202 {
 203         return do_fsync(fd, 0);
 204 }
 205
 206 SYSCALL_DEFINE1(fdatasync, unsigned int, fd)
 207 {
 208         return do_fsync(fd, 1);
 209 }
 210
 211 /**
 212  * generic_write_sync - perform syncing after a write if file / inode is sync
 213  * @file:       file to which the write happened
 214  * @pos:        offset where the write started
 215  * @count:      length of the write
 216  *
 217  * This is just a simple wrapper about our general syncing function.
 218  */
 219 int generic_write_sync(struct file *file, loff_t pos, loff_t count)
 220 {
 221         if (!(file->f_flags & O_DSYNC) && !IS_SYNC(file->f_mapping->host))
 222                 return 0;
 223         return vfs_fsync_range(file, pos, pos + count - 1,
 224                                (file->f_flags & __O_SYNC) ? 0 : 1);
 225 }
 226 EXPORT_SYMBOL(generic_write_sync);
 227
 228 /*
 229  * sys_sync_file_range() permits finely controlled syncing over a segment of
 230  * a file in the range offset .. (offset+nbytes-1) inclusive.  If nbytes is
 231  * zero then sys_sync_file_range() will operate from offset out to EOF.
 232  *
 233  * The flag bits are:
 234  *
 235  * SYNC_FILE_RANGE_WAIT_BEFORE: wait upon writeout of all pages in the range
 236  * before performing the write.
 237  *
 238  * SYNC_FILE_RANGE_WRITE: initiate writeout of all those dirty pages in the
 239  * range which are not presently under writeback. Note that this may block for
 240  * significant periods due to exhaustion of disk request structures.
 241  *
 242  * SYNC_FILE_RANGE_WAIT_AFTER: wait upon writeout of all pages in the range
 243  * after performing the write.
 244  *
 245  * Useful combinations of the flag bits are:
 246  *
 247  * SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE: ensures that all pages
 248  * in the range which were dirty on entry to sys_sync_file_range() are placed
 249  * under writeout.  This is a start-write-for-data-integrity operation.
 250  *
 251  * SYNC_FILE_RANGE_WRITE: start writeout of all dirty pages in the range which
 252  * are not presently under writeout.  This is an asynchronous flush-to-disk
 253  * operation.  Not suitable for data integrity operations.
 254  *
 255  * SYNC_FILE_RANGE_WAIT_BEFORE (or SYNC_FILE_RANGE_WAIT_AFTER): wait for
 256  * completion of writeout of all pages in the range.  This will be used after an
 257  * earlier SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE operation to wait
 258  * for that operation to complete and to return the result.
 259  *
 260  * SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE|SYNC_FILE_RANGE_WAIT_AFTER:
 261  * a traditional sync() operation.  This is a write-for-data-integrity operation
 262  * which will ensure that all pages in the range which were dirty on entry to
 263  * sys_sync_file_range() are committed to disk.
 264  *
 265  *
 266  * SYNC_FILE_RANGE_WAIT_BEFORE and SYNC_FILE_RANGE_WAIT_AFTER will detect any
 267  * I/O errors or ENOSPC conditions and will return those to the caller, after
 268  * clearing the EIO and ENOSPC flags in the address_space.
 269  *
 270  * It should be noted that none of these operations write out the file's
 271  * metadata.  So unless the application is strictly performing overwrites of
 272  * already-instantiated disk blocks, there are no guarantees here that the data
 273  * will be available after a crash.
 274  */
 275 SYSCALL_DEFINE(sync_file_range)(int fd, loff_t offset, loff_t nbytes,
 276                                 unsigned int flags)
 277 {
 278         int ret;
 279         struct file *file;
 280         struct address_space *mapping;
 281         loff_t endbyte;                 /* inclusive */
 282         int fput_needed;
 283         umode_t i_mode;
 284
 285         ret = -EINVAL;
 286         if (flags & ~VALID_FLAGS)
 287                 goto out;
 288
 289         endbyte = offset + nbytes;
 290
 291         if ((s64)offset < 0)
 292                 goto out;
 293         if ((s64)endbyte < 0)
 294                 goto out;
 295         if (endbyte < offset)
 296                 goto out;
 297
 298         if (sizeof(pgoff_t) == 4) {
 299                 if (offset >= (0x100000000ULL << PAGE_CACHE_SHIFT)) {
 300                         /*
 301                          * The range starts outside a 32 bit machine's
 302                          * pagecache addressing capabilities.  Let it "succeed"
 303                          */
 304                         ret = 0;
 305                         goto out;
 306                 }
 307                 if (endbyte >= (0x100000000ULL << PAGE_CACHE_SHIFT)) {
 308                         /*
 309                          * Out to EOF
 310                          */
 311                         nbytes = 0;
 312                 }
 313         }
 314
 315         if (nbytes == 0)
 316                 endbyte = LLONG_MAX;
 317         else
 318                 endbyte--;              /* inclusive */
 319
 320         ret = -EBADF;
 321         file = fget_light(fd, &fput_needed);
 322         if (!file)
 323                 goto out;
 324
 325         i_mode = file->f_path.dentry->d_inode->i_mode;
 326         ret = -ESPIPE;
 327         if (!S_ISREG(i_mode) && !S_ISBLK(i_mode) && !S_ISDIR(i_mode) &&
 328                         !S_ISLNK(i_mode))
 329                 goto out_put;
 330
 331         mapping = file->f_mapping;
 332         if (!mapping) {
 333                 ret = -EINVAL;
 334                 goto out_put;
 335         }
 336
 337         ret = 0;
 338         if (flags & SYNC_FILE_RANGE_WAIT_BEFORE) {
 339                 ret = filemap_fdatawait_range(mapping, offset, endbyte);
 340                 if (ret < 0)
 341                         goto out_put;
 342         }
 343
 344         if (flags & SYNC_FILE_RANGE_WRITE) {
 345                 ret = filemap_fdatawrite_range(mapping, offset, endbyte);
 346                 if (ret < 0)
 347                         goto out_put;
 348         }
 349
 350         if (flags & SYNC_FILE_RANGE_WAIT_AFTER)
 351                 ret = filemap_fdatawait_range(mapping, offset, endbyte);
 352
 353 out_put:
 354         fput_light(file, fput_needed);
 355 out:
 356         return ret;
 357 }
 358 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
 359 asmlinkage long SyS_sync_file_range(long fd, loff_t offset, loff_t nbytes,
 360                                     long flags)
 361 {
 362         return SYSC_sync_file_range((int) fd, offset, nbytes,
 363                                     (unsigned int) flags);
 364 }
 365 SYSCALL_ALIAS(sys_sync_file_range, SyS_sync_file_range);
 366 #endif
 367
 368 /* It would be nice if people remember that not all the world's an i386
 369    when they introduce new system calls */
 370 SYSCALL_DEFINE(sync_file_range2)(int fd, unsigned int flags,
 371                                  loff_t offset, loff_t nbytes)
 372 {
 373         return sys_sync_file_range(fd, offset, nbytes, flags);
 374 }
 375 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
 376 asmlinkage long SyS_sync_file_range2(long fd, long flags,
 377                                      loff_t offset, loff_t nbytes)
 378 {
 379         return SYSC_sync_file_range2((int) fd, (unsigned int) flags,
 380                                      offset, nbytes);
 381 }
 382 SYSCALL_ALIAS(sys_sync_file_range2, SyS_sync_file_range2);
 383 #endif