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/module.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/buffer_head.h>
  18 #include <linux/backing-dev.h>
  19 #include "internal.h"
  20
  21 #define VALID_FLAGS (SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE| \
  22                         SYNC_FILE_RANGE_WAIT_AFTER)
  23
  24 #ifdef CONFIG_FSYNC_CONTROL
  25 extern bool fsynccontrol_fsync_enabled();
  26 #endif
  27
  28 /*
  29  * Do the filesystem syncing work. For simple filesystems
  30  * writeback_inodes_sb(sb) just dirties buffers with inodes so we have to
  31  * submit IO for these buffers via __sync_blockdev(). This also speeds up the
  32  * wait == 1 case since in that case write_inode() functions do
  33  * sync_dirty_buffer() and thus effectively write one block at a time.
  34  */
  35 static int __sync_filesystem(struct super_block *sb, int wait)
  36 {
  37         /*
  38          * This should be safe, as we require bdi backing to actually
  39          * write out data in the first place
  40          */
  41         if (sb->s_bdi == &noop_backing_dev_info)
  42                 return 0;
  43
  44         if (sb->s_qcop && sb->s_qcop->quota_sync)
  45                 sb->s_qcop->quota_sync(sb, -1, wait);
  46
  47         if (wait)
  48                 sync_inodes_sb(sb);
  49         else
  50                 writeback_inodes_sb(sb);
  51
  52         if (sb->s_op->sync_fs)
  53                 sb->s_op->sync_fs(sb, wait);
  54         return __sync_blockdev(sb->s_bdev, wait);
  55 }
  56
  57 /*
  58  * Write out and wait upon all dirty data associated with this
  59  * superblock.  Filesystem data as well as the underlying block
  60  * device.  Takes the superblock lock.
  61  */
  62 int sync_filesystem(struct super_block *sb)
  63 {
  64         int ret;
  65
  66         /*
  67          * We need to be protected against the filesystem going from
  68          * r/o to r/w or vice versa.
  69          */
  70         WARN_ON(!rwsem_is_locked(&sb->s_umount));
  71
  72         /*
  73          * No point in syncing out anything if the filesystem is read-only.
  74          */
  75         if (sb->s_flags & MS_RDONLY)
  76                 return 0;
  77
  78         ret = __sync_filesystem(sb, 0);
  79         if (ret < 0)
  80                 return ret;
  81         return __sync_filesystem(sb, 1);
  82 }
  83 EXPORT_SYMBOL_GPL(sync_filesystem);
  84
  85 static void sync_one_sb(struct super_block *sb, void *arg)
  86 {
  87         if (!(sb->s_flags & MS_RDONLY))
  88                 __sync_filesystem(sb, *(int *)arg);
  89 }
  90 /*
  91  * Sync all the data for all the filesystems (called by sys_sync() and
  92  * emergency sync)
  93  */
  94 static void sync_filesystems(int wait)
  95 {
  96         iterate_supers(sync_one_sb, &wait);
  97 }
  98
  99 /*
 100  * sync everything.  Start out by waking pdflush, because that writes back
 101  * all queues in parallel.
 102  */
 103 SYSCALL_DEFINE0(sync)
 104 {
 105         wakeup_flusher_threads(0);
 106         sync_filesystems(0);
 107         sync_filesystems(1);
 108         if (unlikely(laptop_mode))
 109                 laptop_sync_completion();
 110         return 0;
 111 }
 112
 113 static void do_sync_work(struct work_struct *work)
 114 {
 115         /*
 116          * Sync twice to reduce the possibility we skipped some inodes / pages
 117          * because they were temporarily locked
 118          */
 119         sync_filesystems(0);
 120         sync_filesystems(0);
 121         printk("Emergency Sync complete\n");
 122         kfree(work);
 123 }
 124
 125 void emergency_sync(void)
 126 {
 127         struct work_struct *work;
 128
 129         work = kmalloc(sizeof(*work), GFP_ATOMIC);
 130         if (work) {
 131                 INIT_WORK(work, do_sync_work);
 132                 schedule_work(work);
 133         }
 134 }
 135
 136 /*
 137  * sync a single super
 138  */
 139 SYSCALL_DEFINE1(syncfs, int, fd)
 140 {
 141         struct file *file;
 142         struct super_block *sb;
 143         int ret;
 144         int fput_needed;
 145
 146 #ifdef CONFIG_FSYNC_CONTROL
 147         if (!fsynccontrol_fsync_enabled())
 148             return 0;
 149 #endif
 150
 151         file = fget_light(fd, &fput_needed);
 152         if (!file)
 153                 return -EBADF;
 154         sb = file->f_dentry->d_sb;
 155
 156         down_read(&sb->s_umount);
 157         ret = sync_filesystem(sb);
 158         up_read(&sb->s_umount);
 159
 160         fput_light(file, fput_needed);
 161         return ret;
 162 }
 163
 164 /**
 165  * vfs_fsync_range - helper to sync a range of data & metadata to disk
 166  * @file:               file to sync
 167  * @start:              offset in bytes of the beginning of data range to sync
 168  * @end:                offset in bytes of the end of data range (inclusive)
 169  * @datasync:           perform only datasync
 170  *
 171  * Write back data in range @start..@end and metadata for @file to disk.  If
 172  * @datasync is set only metadata needed to access modified file data is
 173  * written.
 174  */
 175 int vfs_fsync_range(struct file *file, loff_t start, loff_t end, int datasync)
 176 {
 177         struct address_space *mapping = file->f_mapping;
 178         int err, ret;
 179
 180 #ifdef CONFIG_FSYNC_CONTROL
 181         if (!fsynccontrol_fsync_enabled())
 182             return 0;
 183 #endif
 184
 185         if (!file->f_op || !file->f_op->fsync) {
 186                 ret = -EINVAL;
 187                 goto out;
 188         }
 189
 190         ret = filemap_write_and_wait_range(mapping, start, end);
 191
 192         /*
 193          * We need to protect against concurrent writers, which could cause
 194          * livelocks in fsync_buffers_list().
 195          */
 196         mutex_lock(&mapping->host->i_mutex);
 197         err = file->f_op->fsync(file, datasync);
 198         if (!ret)
 199                 ret = err;
 200         mutex_unlock(&mapping->host->i_mutex);
 201
 202 out:
 203         return ret;
 204 }
 205 EXPORT_SYMBOL(vfs_fsync_range);
 206
 207 /**
 208  * vfs_fsync - perform a fsync or fdatasync on a file
 209  * @file:               file to sync
 210  * @datasync:           only perform a fdatasync operation
 211  *
 212  * Write back data and metadata for @file to disk.  If @datasync is
 213  * set only metadata needed to access modified file data is written.
 214  */
 215 int vfs_fsync(struct file *file, int datasync)
 216 {
 217 #ifdef CONFIG_FSYNC_CONTROL
 218         if (!fsynccontrol_fsync_enabled())
 219             return 0;
 220 #endif
 221
 222         return vfs_fsync_range(file, 0, LLONG_MAX, datasync);
 223 }
 224 EXPORT_SYMBOL(vfs_fsync);
 225
 226 static int do_fsync(unsigned int fd, int datasync)
 227 {
 228         struct file *file;
 229         int ret = -EBADF;
 230
 231 #ifdef CONFIG_FSYNC_CONTROL
 232         if (!fsynccontrol_fsync_enabled())
 233             return 0;
 234 #endif
 235
 236         file = fget(fd);
 237         if (file) {
 238                 ret = vfs_fsync(file, datasync);
 239                 fput(file);
 240         }
 241         return ret;
 242 }
 243
 244 SYSCALL_DEFINE1(fsync, unsigned int, fd)
 245 {
 246 #ifdef CONFIG_FSYNC_CONTROL
 247         if (!fsynccontrol_fsync_enabled())
 248             return 0;
 249 #endif
 250
 251         return do_fsync(fd, 0);
 252 }
 253
 254 SYSCALL_DEFINE1(fdatasync, unsigned int, fd)
 255 {
 256 #ifdef CONFIG_FSYNC_CONTROL
 257         if (!fsynccontrol_fsync_enabled())
 258             return 0;
 259 #endif
 260
 261         return do_fsync(fd, 1);
 262 }
 263
 264 /**
 265  * generic_write_sync - perform syncing after a write if file / inode is sync
 266  * @file:       file to which the write happened
 267  * @pos:        offset where the write started
 268  * @count:      length of the write
 269  *
 270  * This is just a simple wrapper about our general syncing function.
 271  */
 272 int generic_write_sync(struct file *file, loff_t pos, loff_t count)
 273 {
 274 #ifdef CONFIG_FSYNC_CONTROL
 275         if (!fsynccontrol_fsync_enabled())
 276             return 0;
 277 #endif
 278
 279         if (!(file->f_flags & O_DSYNC) && !IS_SYNC(file->f_mapping->host))
 280                 return 0;
 281         return vfs_fsync_range(file, pos, pos + count - 1,
 282                                (file->f_flags & __O_SYNC) ? 0 : 1);
 283 }
 284 EXPORT_SYMBOL(generic_write_sync);
 285
 286 /*
 287  * sys_sync_file_range() permits finely controlled syncing over a segment of
 288  * a file in the range offset .. (offset+nbytes-1) inclusive.  If nbytes is
 289  * zero then sys_sync_file_range() will operate from offset out to EOF.
 290  *
 291  * The flag bits are:
 292  *
 293  * SYNC_FILE_RANGE_WAIT_BEFORE: wait upon writeout of all pages in the range
 294  * before performing the write.
 295  *
 296  * SYNC_FILE_RANGE_WRITE: initiate writeout of all those dirty pages in the
 297  * range which are not presently under writeback. Note that this may block for
 298  * significant periods due to exhaustion of disk request structures.
 299  *
 300  * SYNC_FILE_RANGE_WAIT_AFTER: wait upon writeout of all pages in the range
 301  * after performing the write.
 302  *
 303  * Useful combinations of the flag bits are:
 304  *
 305  * SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE: ensures that all pages
 306  * in the range which were dirty on entry to sys_sync_file_range() are placed
 307  * under writeout.  This is a start-write-for-data-integrity operation.
 308  *
 309  * SYNC_FILE_RANGE_WRITE: start writeout of all dirty pages in the range which
 310  * are not presently under writeout.  This is an asynchronous flush-to-disk
 311  * operation.  Not suitable for data integrity operations.
 312  *
 313  * SYNC_FILE_RANGE_WAIT_BEFORE (or SYNC_FILE_RANGE_WAIT_AFTER): wait for
 314  * completion of writeout of all pages in the range.  This will be used after an
 315  * earlier SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE operation to wait
 316  * for that operation to complete and to return the result.
 317  *
 318  * SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE|SYNC_FILE_RANGE_WAIT_AFTER:
 319  * a traditional sync() operation.  This is a write-for-data-integrity operation
 320  * which will ensure that all pages in the range which were dirty on entry to
 321  * sys_sync_file_range() are committed to disk.
 322  *
 323  *
 324  * SYNC_FILE_RANGE_WAIT_BEFORE and SYNC_FILE_RANGE_WAIT_AFTER will detect any
 325  * I/O errors or ENOSPC conditions and will return those to the caller, after
 326  * clearing the EIO and ENOSPC flags in the address_space.
 327  *
 328  * It should be noted that none of these operations write out the file's
 329  * metadata.  So unless the application is strictly performing overwrites of
 330  * already-instantiated disk blocks, there are no guarantees here that the data
 331  * will be available after a crash.
 332  */
 333 SYSCALL_DEFINE(sync_file_range)(int fd, loff_t offset, loff_t nbytes,
 334                                 unsigned int flags)
 335 {
 336         int ret;
 337         struct file *file;
 338         struct address_space *mapping;
 339         loff_t endbyte;                 /* inclusive */
 340         int fput_needed;
 341         umode_t i_mode;
 342
 343 #ifdef CONFIG_FSYNC_CONTROL
 344         if (!fsynccontrol_fsync_enabled())
 345             return 0;
 346 #endif
 347
 348         ret = -EINVAL;
 349         if (flags & ~VALID_FLAGS)
 350                 goto out;
 351
 352         endbyte = offset + nbytes;
 353
 354         if ((s64)offset < 0)
 355                 goto out;
 356         if ((s64)endbyte < 0)
 357                 goto out;
 358         if (endbyte < offset)
 359                 goto out;
 360
 361         if (sizeof(pgoff_t) == 4) {
 362                 if (offset >= (0x100000000ULL << PAGE_CACHE_SHIFT)) {
 363                         /*
 364                          * The range starts outside a 32 bit machine's
 365                          * pagecache addressing capabilities.  Let it "succeed"
 366                          */
 367                         ret = 0;
 368                         goto out;
 369                 }
 370                 if (endbyte >= (0x100000000ULL << PAGE_CACHE_SHIFT)) {
 371                         /*
 372                          * Out to EOF
 373                          */
 374                         nbytes = 0;
 375                 }
 376         }
 377
 378         if (nbytes == 0)
 379                 endbyte = LLONG_MAX;
 380         else
 381                 endbyte--;              /* inclusive */
 382
 383         ret = -EBADF;
 384         file = fget_light(fd, &fput_needed);
 385         if (!file)
 386                 goto out;
 387
 388         i_mode = file->f_path.dentry->d_inode->i_mode;
 389         ret = -ESPIPE;
 390         if (!S_ISREG(i_mode) && !S_ISBLK(i_mode) && !S_ISDIR(i_mode) &&
 391                         !S_ISLNK(i_mode))
 392                 goto out_put;
 393
 394         mapping = file->f_mapping;
 395         if (!mapping) {
 396                 ret = -EINVAL;
 397                 goto out_put;
 398         }
 399
 400         ret = 0;
 401         if (flags & SYNC_FILE_RANGE_WAIT_BEFORE) {
 402                 ret = filemap_fdatawait_range(mapping, offset, endbyte);
 403                 if (ret < 0)
 404                         goto out_put;
 405         }
 406
 407         if (flags & SYNC_FILE_RANGE_WRITE) {
 408                 ret = filemap_fdatawrite_range(mapping, offset, endbyte);
 409                 if (ret < 0)
 410                         goto out_put;
 411         }
 412
 413         if (flags & SYNC_FILE_RANGE_WAIT_AFTER)
 414                 ret = filemap_fdatawait_range(mapping, offset, endbyte);
 415
 416 out_put:
 417         fput_light(file, fput_needed);
 418 out:
 419         return ret;
 420 }
 421 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
 422 asmlinkage long SyS_sync_file_range(long fd, loff_t offset, loff_t nbytes,
 423                                     long flags)
 424 {
 425         return SYSC_sync_file_range((int) fd, offset, nbytes,
 426                                     (unsigned int) flags);
 427 }
 428 SYSCALL_ALIAS(sys_sync_file_range, SyS_sync_file_range);
 429 #endif
 430
 431 /* It would be nice if people remember that not all the world's an i386
 432    when they introduce new system calls */
 433 SYSCALL_DEFINE(sync_file_range2)(int fd, unsigned int flags,
 434                                  loff_t offset, loff_t nbytes)
 435 {
 436         return sys_sync_file_range(fd, offset, nbytes, flags);
 437 }
 438 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
 439 asmlinkage long SyS_sync_file_range2(long fd, long flags,
 440                                      loff_t offset, loff_t nbytes)
 441 {
 442         return SYSC_sync_file_range2((int) fd, (unsigned int) flags,
 443                                      offset, nbytes);
 444 }
 445 SYSCALL_ALIAS(sys_sync_file_range2, SyS_sync_file_range2);
 446 #endif