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/writeback.h>
  10 #include <linux/syscalls.h>
  11 #include <linux/linkage.h>
  12 #include <linux/pagemap.h>
  13 #include <linux/quotaops.h>
  14 #include <linux/buffer_head.h>
  15
  16 #define VALID_FLAGS (SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE| \
  17                         SYNC_FILE_RANGE_WAIT_AFTER)
  18
  19 /*
  20  * sync everything.  Start out by waking pdflush, because that writes back
  21  * all queues in parallel.
  22  */
  23 static void do_sync(unsigned long wait)
  24 {
  25         wakeup_pdflush(0);
  26         sync_inodes(0);         /* All mappings, inodes and their blockdevs */
  27         DQUOT_SYNC(NULL);
  28         sync_supers();          /* Write the superblocks */
  29         sync_filesystems(0);    /* Start syncing the filesystems */
  30         sync_filesystems(wait); /* Waitingly sync the filesystems */
  31         sync_inodes(wait);      /* Mappings, inodes and blockdevs, again. */
  32         if (!wait)
  33                 printk("Emergency Sync complete\n");
  34         if (unlikely(laptop_mode))
  35                 laptop_sync_completion();
  36 }
  37
  38 asmlinkage long sys_sync(void)
  39 {
  40         do_sync(1);
  41         return 0;
  42 }
  43
  44 void emergency_sync(void)
  45 {
  46         pdflush_operation(do_sync, 0);
  47 }
  48
  49 /*
  50  * Generic function to fsync a file.
  51  *
  52  * filp may be NULL if called via the msync of a vma.
  53  */
  54 int file_fsync(struct file *filp, struct dentry *dentry, int datasync)
  55 {
  56         struct inode * inode = dentry->d_inode;
  57         struct super_block * sb;
  58         int ret, err;
  59
  60         /* sync the inode to buffers */
  61         ret = write_inode_now(inode, 0);
  62
  63         /* sync the superblock to buffers */
  64         sb = inode->i_sb;
  65         lock_super(sb);
  66         if (sb->s_op->write_super)
  67                 sb->s_op->write_super(sb);
  68         unlock_super(sb);
  69
  70         /* .. finally sync the buffers to disk */
  71         err = sync_blockdev(sb->s_bdev);
  72         if (!ret)
  73                 ret = err;
  74         return ret;
  75 }
  76
  77 long do_fsync(struct file *file, int datasync)
  78 {
  79         int ret;
  80         int err;
  81         struct address_space *mapping = file->f_mapping;
  82
  83         if (!file->f_op || !file->f_op->fsync) {
  84                 /* Why?  We can still call filemap_fdatawrite */
  85                 ret = -EINVAL;
  86                 goto out;
  87         }
  88
  89         ret = filemap_fdatawrite(mapping);
  90
  91         /*
  92          * We need to protect against concurrent writers, which could cause
  93          * livelocks in fsync_buffers_list().
  94          */
  95         mutex_lock(&mapping->host->i_mutex);
  96         err = file->f_op->fsync(file, file->f_dentry, datasync);
  97         if (!ret)
  98                 ret = err;
  99         mutex_unlock(&mapping->host->i_mutex);
 100         err = filemap_fdatawait(mapping);
 101         if (!ret)
 102                 ret = err;
 103 out:
 104         return ret;
 105 }
 106
 107 static long __do_fsync(unsigned int fd, int datasync)
 108 {
 109         struct file *file;
 110         int ret = -EBADF;
 111
 112         file = fget(fd);
 113         if (file) {
 114                 ret = do_fsync(file, datasync);
 115                 fput(file);
 116         }
 117         return ret;
 118 }
 119
 120 asmlinkage long sys_fsync(unsigned int fd)
 121 {
 122         return __do_fsync(fd, 0);
 123 }
 124
 125 asmlinkage long sys_fdatasync(unsigned int fd)
 126 {
 127         return __do_fsync(fd, 1);
 128 }
 129
 130 /*
 131  * sys_sync_file_range() permits finely controlled syncing over a segment of
 132  * a file in the range offset .. (offset+nbytes-1) inclusive.  If nbytes is
 133  * zero then sys_sync_file_range() will operate from offset out to EOF.
 134  *
 135  * The flag bits are:
 136  *
 137  * SYNC_FILE_RANGE_WAIT_BEFORE: wait upon writeout of all pages in the range
 138  * before performing the write.
 139  *
 140  * SYNC_FILE_RANGE_WRITE: initiate writeout of all those dirty pages in the
 141  * range which are not presently under writeback.
 142  *
 143  * SYNC_FILE_RANGE_WAIT_AFTER: wait upon writeout of all pages in the range
 144  * after performing the write.
 145  *
 146  * Useful combinations of the flag bits are:
 147  *
 148  * SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE: ensures that all pages
 149  * in the range which were dirty on entry to sys_sync_file_range() are placed
 150  * under writeout.  This is a start-write-for-data-integrity operation.
 151  *
 152  * SYNC_FILE_RANGE_WRITE: start writeout of all dirty pages in the range which
 153  * are not presently under writeout.  This is an asynchronous flush-to-disk
 154  * operation.  Not suitable for data integrity operations.
 155  *
 156  * SYNC_FILE_RANGE_WAIT_BEFORE (or SYNC_FILE_RANGE_WAIT_AFTER): wait for
 157  * completion of writeout of all pages in the range.  This will be used after an
 158  * earlier SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE operation to wait
 159  * for that operation to complete and to return the result.
 160  *
 161  * SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE|SYNC_FILE_RANGE_WAIT_AFTER:
 162  * a traditional sync() operation.  This is a write-for-data-integrity operation
 163  * which will ensure that all pages in the range which were dirty on entry to
 164  * sys_sync_file_range() are committed to disk.
 165  *
 166  *
 167  * SYNC_FILE_RANGE_WAIT_BEFORE and SYNC_FILE_RANGE_WAIT_AFTER will detect any
 168  * I/O errors or ENOSPC conditions and will return those to the caller, after
 169  * clearing the EIO and ENOSPC flags in the address_space.
 170  *
 171  * It should be noted that none of these operations write out the file's
 172  * metadata.  So unless the application is strictly performing overwrites of
 173  * already-instantiated disk blocks, there are no guarantees here that the data
 174  * will be available after a crash.
 175  */
 176 asmlinkage long sys_sync_file_range(int fd, loff_t offset, loff_t nbytes,
 177                                         unsigned int flags)
 178 {
 179         int ret;
 180         struct file *file;
 181         loff_t endbyte;                 /* inclusive */
 182         int fput_needed;
 183         umode_t i_mode;
 184
 185         ret = -EINVAL;
 186         if (flags & ~VALID_FLAGS)
 187                 goto out;
 188
 189         endbyte = offset + nbytes;
 190
 191         if ((s64)offset < 0)
 192                 goto out;
 193         if ((s64)endbyte < 0)
 194                 goto out;
 195         if (endbyte < offset)
 196                 goto out;
 197
 198         if (sizeof(pgoff_t) == 4) {
 199                 if (offset >= (0x100000000ULL << PAGE_CACHE_SHIFT)) {
 200                         /*
 201                          * The range starts outside a 32 bit machine's
 202                          * pagecache addressing capabilities.  Let it "succeed"
 203                          */
 204                         ret = 0;
 205                         goto out;
 206                 }
 207                 if (endbyte >= (0x100000000ULL << PAGE_CACHE_SHIFT)) {
 208                         /*
 209                          * Out to EOF
 210                          */
 211                         nbytes = 0;
 212                 }
 213         }
 214
 215         if (nbytes == 0)
 216                 endbyte = LLONG_MAX;
 217         else
 218                 endbyte--;              /* inclusive */
 219
 220         ret = -EBADF;
 221         file = fget_light(fd, &fput_needed);
 222         if (!file)
 223                 goto out;
 224
 225         i_mode = file->f_dentry->d_inode->i_mode;
 226         ret = -ESPIPE;
 227         if (!S_ISREG(i_mode) && !S_ISBLK(i_mode) && !S_ISDIR(i_mode) &&
 228                         !S_ISLNK(i_mode))
 229                 goto out_put;
 230
 231         ret = do_sync_file_range(file, offset, endbyte, flags);
 232 out_put:
 233         fput_light(file, fput_needed);
 234 out:
 235         return ret;
 236 }
 237
 238 /*
 239  * `endbyte' is inclusive
 240  */
 241 int do_sync_file_range(struct file *file, loff_t offset, loff_t endbyte,
 242                         unsigned int flags)
 243 {
 244         int ret;
 245         struct address_space *mapping;
 246
 247         mapping = file->f_mapping;
 248         if (!mapping) {
 249                 ret = -EINVAL;
 250                 goto out;
 251         }
 252
 253         ret = 0;
 254         if (flags & SYNC_FILE_RANGE_WAIT_BEFORE) {
 255                 ret = wait_on_page_writeback_range(mapping,
 256                                         offset >> PAGE_CACHE_SHIFT,
 257                                         endbyte >> PAGE_CACHE_SHIFT);
 258                 if (ret < 0)
 259                         goto out;
 260         }
 261
 262         if (flags & SYNC_FILE_RANGE_WRITE) {
 263                 ret = __filemap_fdatawrite_range(mapping, offset, endbyte,
 264                                                 WB_SYNC_NONE);
 265                 if (ret < 0)
 266                         goto out;
 267         }
 268
 269         if (flags & SYNC_FILE_RANGE_WAIT_AFTER) {
 270                 ret = wait_on_page_writeback_range(mapping,
 271                                         offset >> PAGE_CACHE_SHIFT,
 272                                         endbyte >> PAGE_CACHE_SHIFT);
 273         }
 274 out:
 275         return ret;
 276 }
 277 EXPORT_SYMBOL_GPL(do_sync_file_range);