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
  17 #define VALID_FLAGS (SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE| \
  18                         SYNC_FILE_RANGE_WAIT_AFTER)
  19
  20 /*
  21  * sync everything.  Start out by waking pdflush, because that writes back
  22  * all queues in parallel.
  23  */
  24 static void do_sync(unsigned long wait)
  25 {
  26         wakeup_pdflush(0);
  27         sync_inodes(0);         /* All mappings, inodes and their blockdevs */
  28         DQUOT_SYNC(NULL);
  29         sync_supers();          /* Write the superblocks */
  30         sync_filesystems(0);    /* Start syncing the filesystems */
  31         sync_filesystems(wait); /* Waitingly sync the filesystems */
  32         sync_inodes(wait);      /* Mappings, inodes and blockdevs, again. */
  33         if (!wait)
  34                 printk("Emergency Sync complete\n");
  35         if (unlikely(laptop_mode))
  36                 laptop_sync_completion();
  37 }
  38
  39 SYSCALL_DEFINE0(sync)
  40 {
  41         do_sync(1);
  42         return 0;
  43 }
  44
  45 void emergency_sync(void)
  46 {
  47         pdflush_operation(do_sync, 0);
  48 }
  49
  50 /*
  51  * Generic function to fsync a file.
  52  *
  53  * filp may be NULL if called via the msync of a vma.
  54  */
  55 int file_fsync(struct file *filp, struct dentry *dentry, int datasync)
  56 {
  57         struct inode * inode = dentry->d_inode;
  58         struct super_block * sb;
  59         int ret, err;
  60
  61         /* sync the inode to buffers */
  62         ret = write_inode_now(inode, 0);
  63
  64         /* sync the superblock to buffers */
  65         sb = inode->i_sb;
  66         lock_super(sb);
  67         if (sb->s_dirt && sb->s_op->write_super)
  68                 sb->s_op->write_super(sb);
  69         unlock_super(sb);
  70
  71         /* .. finally sync the buffers to disk */
  72         err = sync_blockdev(sb->s_bdev);
  73         if (!ret)
  74                 ret = err;
  75         return ret;
  76 }
  77
  78 /**
  79  * vfs_fsync - perform a fsync or fdatasync on a file
  80  * @file:               file to sync
  81  * @dentry:             dentry of @file
  82  * @data:               only perform a fdatasync operation
  83  *
  84  * Write back data and metadata for @file to disk.  If @datasync is
  85  * set only metadata needed to access modified file data is written.
  86  *
  87  * In case this function is called from nfsd @file may be %NULL and
  88  * only @dentry is set.  This can only happen when the filesystem
  89  * implements the export_operations API.
  90  */
  91 int vfs_fsync(struct file *file, struct dentry *dentry, int datasync)
  92 {
  93         const struct file_operations *fop;
  94         struct address_space *mapping;
  95         int err, ret;
  96
  97         /*
  98          * Get mapping and operations from the file in case we have
  99          * as file, or get the default values for them in case we
 100          * don't have a struct file available.  Damn nfsd..
 101          */
 102         if (file) {
 103                 mapping = file->f_mapping;
 104                 fop = file->f_op;
 105         } else {
 106                 mapping = dentry->d_inode->i_mapping;
 107                 fop = dentry->d_inode->i_fop;
 108         }
 109
 110         if (!fop || !fop->fsync) {
 111                 ret = -EINVAL;
 112                 goto out;
 113         }
 114
 115         ret = filemap_fdatawrite(mapping);
 116
 117         /*
 118          * We need to protect against concurrent writers, which could cause
 119          * livelocks in fsync_buffers_list().
 120          */
 121         mutex_lock(&mapping->host->i_mutex);
 122         err = fop->fsync(file, dentry, datasync);
 123         if (!ret)
 124                 ret = err;
 125         mutex_unlock(&mapping->host->i_mutex);
 126         err = filemap_fdatawait(mapping);
 127         if (!ret)
 128                 ret = err;
 129 out:
 130         return ret;
 131 }
 132 EXPORT_SYMBOL(vfs_fsync);
 133
 134 static int do_fsync(unsigned int fd, int datasync)
 135 {
 136         struct file *file;
 137         int ret = -EBADF;
 138
 139         file = fget(fd);
 140         if (file) {
 141                 ret = vfs_fsync(file, file->f_path.dentry, datasync);
 142                 fput(file);
 143         }
 144         return ret;
 145 }
 146
 147 SYSCALL_DEFINE1(fsync, unsigned int, fd)
 148 {
 149         return do_fsync(fd, 0);
 150 }
 151
 152 SYSCALL_DEFINE1(fdatasync, unsigned int, fd)
 153 {
 154         return do_fsync(fd, 1);
 155 }
 156
 157 /*
 158  * sys_sync_file_range() permits finely controlled syncing over a segment of
 159  * a file in the range offset .. (offset+nbytes-1) inclusive.  If nbytes is
 160  * zero then sys_sync_file_range() will operate from offset out to EOF.
 161  *
 162  * The flag bits are:
 163  *
 164  * SYNC_FILE_RANGE_WAIT_BEFORE: wait upon writeout of all pages in the range
 165  * before performing the write.
 166  *
 167  * SYNC_FILE_RANGE_WRITE: initiate writeout of all those dirty pages in the
 168  * range which are not presently under writeback. Note that this may block for
 169  * significant periods due to exhaustion of disk request structures.
 170  *
 171  * SYNC_FILE_RANGE_WAIT_AFTER: wait upon writeout of all pages in the range
 172  * after performing the write.
 173  *
 174  * Useful combinations of the flag bits are:
 175  *
 176  * SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE: ensures that all pages
 177  * in the range which were dirty on entry to sys_sync_file_range() are placed
 178  * under writeout.  This is a start-write-for-data-integrity operation.
 179  *
 180  * SYNC_FILE_RANGE_WRITE: start writeout of all dirty pages in the range which
 181  * are not presently under writeout.  This is an asynchronous flush-to-disk
 182  * operation.  Not suitable for data integrity operations.
 183  *
 184  * SYNC_FILE_RANGE_WAIT_BEFORE (or SYNC_FILE_RANGE_WAIT_AFTER): wait for
 185  * completion of writeout of all pages in the range.  This will be used after an
 186  * earlier SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE operation to wait
 187  * for that operation to complete and to return the result.
 188  *
 189  * SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE|SYNC_FILE_RANGE_WAIT_AFTER:
 190  * a traditional sync() operation.  This is a write-for-data-integrity operation
 191  * which will ensure that all pages in the range which were dirty on entry to
 192  * sys_sync_file_range() are committed to disk.
 193  *
 194  *
 195  * SYNC_FILE_RANGE_WAIT_BEFORE and SYNC_FILE_RANGE_WAIT_AFTER will detect any
 196  * I/O errors or ENOSPC conditions and will return those to the caller, after
 197  * clearing the EIO and ENOSPC flags in the address_space.
 198  *
 199  * It should be noted that none of these operations write out the file's
 200  * metadata.  So unless the application is strictly performing overwrites of
 201  * already-instantiated disk blocks, there are no guarantees here that the data
 202  * will be available after a crash.
 203  */
 204 SYSCALL_DEFINE(sync_file_range)(int fd, loff_t offset, loff_t nbytes,
 205                                 unsigned int flags)
 206 {
 207         int ret;
 208         struct file *file;
 209         loff_t endbyte;                 /* inclusive */
 210         int fput_needed;
 211         umode_t i_mode;
 212
 213         ret = -EINVAL;
 214         if (flags & ~VALID_FLAGS)
 215                 goto out;
 216
 217         endbyte = offset + nbytes;
 218
 219         if ((s64)offset < 0)
 220                 goto out;
 221         if ((s64)endbyte < 0)
 222                 goto out;
 223         if (endbyte < offset)
 224                 goto out;
 225
 226         if (sizeof(pgoff_t) == 4) {
 227                 if (offset >= (0x100000000ULL << PAGE_CACHE_SHIFT)) {
 228                         /*
 229                          * The range starts outside a 32 bit machine's
 230                          * pagecache addressing capabilities.  Let it "succeed"
 231                          */
 232                         ret = 0;
 233                         goto out;
 234                 }
 235                 if (endbyte >= (0x100000000ULL << PAGE_CACHE_SHIFT)) {
 236                         /*
 237                          * Out to EOF
 238                          */
 239                         nbytes = 0;
 240                 }
 241         }
 242
 243         if (nbytes == 0)
 244                 endbyte = LLONG_MAX;
 245         else
 246                 endbyte--;              /* inclusive */
 247
 248         ret = -EBADF;
 249         file = fget_light(fd, &fput_needed);
 250         if (!file)
 251                 goto out;
 252
 253         i_mode = file->f_path.dentry->d_inode->i_mode;
 254         ret = -ESPIPE;
 255         if (!S_ISREG(i_mode) && !S_ISBLK(i_mode) && !S_ISDIR(i_mode) &&
 256                         !S_ISLNK(i_mode))
 257                 goto out_put;
 258
 259         ret = do_sync_mapping_range(file->f_mapping, offset, endbyte, flags);
 260 out_put:
 261         fput_light(file, fput_needed);
 262 out:
 263         return ret;
 264 }
 265 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
 266 asmlinkage long SyS_sync_file_range(long fd, loff_t offset, loff_t nbytes,
 267                                     long flags)
 268 {
 269         return SYSC_sync_file_range((int) fd, offset, nbytes,
 270                                     (unsigned int) flags);
 271 }
 272 SYSCALL_ALIAS(sys_sync_file_range, SyS_sync_file_range);
 273 #endif
 274
 275 /* It would be nice if people remember that not all the world's an i386
 276    when they introduce new system calls */
 277 SYSCALL_DEFINE(sync_file_range2)(int fd, unsigned int flags,
 278                                  loff_t offset, loff_t nbytes)
 279 {
 280         return sys_sync_file_range(fd, offset, nbytes, flags);
 281 }
 282 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
 283 asmlinkage long SyS_sync_file_range2(long fd, long flags,
 284                                      loff_t offset, loff_t nbytes)
 285 {
 286         return SYSC_sync_file_range2((int) fd, (unsigned int) flags,
 287                                      offset, nbytes);
 288 }
 289 SYSCALL_ALIAS(sys_sync_file_range2, SyS_sync_file_range2);
 290 #endif
 291
 292 /*
 293  * `endbyte' is inclusive
 294  */
 295 int do_sync_mapping_range(struct address_space *mapping, loff_t offset,
 296                           loff_t endbyte, unsigned int flags)
 297 {
 298         int ret;
 299
 300         if (!mapping) {
 301                 ret = -EINVAL;
 302                 goto out;
 303         }
 304
 305         ret = 0;
 306         if (flags & SYNC_FILE_RANGE_WAIT_BEFORE) {
 307                 ret = wait_on_page_writeback_range(mapping,
 308                                         offset >> PAGE_CACHE_SHIFT,
 309                                         endbyte >> PAGE_CACHE_SHIFT);
 310                 if (ret < 0)
 311                         goto out;
 312         }
 313
 314         if (flags & SYNC_FILE_RANGE_WRITE) {
 315                 ret = __filemap_fdatawrite_range(mapping, offset, endbyte,
 316                                                 WB_SYNC_ALL);
 317                 if (ret < 0)
 318                         goto out;
 319         }
 320
 321         if (flags & SYNC_FILE_RANGE_WAIT_AFTER) {
 322                 ret = wait_on_page_writeback_range(mapping,
 323                                         offset >> PAGE_CACHE_SHIFT,
 324                                         endbyte >> PAGE_CACHE_SHIFT);
 325         }
 326 out:
 327         return ret;
 328 }
 329 EXPORT_SYMBOL_GPL(do_sync_mapping_range);