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