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 long do_fsync(struct file *file, int datasync)
  79 {
  80         int ret;
  81         int err;
  82         struct address_space *mapping = file->f_mapping;
  83
  84         if (!file->f_op || !file->f_op->fsync) {
  85                 /* Why?  We can still call filemap_fdatawrite */
  86                 ret = -EINVAL;
  87                 goto out;
  88         }
  89
  90         ret = filemap_fdatawrite(mapping);
  91
  92         /*
  93          * We need to protect against concurrent writers, which could cause
  94          * livelocks in fsync_buffers_list().
  95          */
  96         mutex_lock(&mapping->host->i_mutex);
  97         err = file->f_op->fsync(file, file->f_path.dentry, datasync);
  98         if (!ret)
  99                 ret = err;
 100         mutex_unlock(&mapping->host->i_mutex);
 101         err = filemap_fdatawait(mapping);
 102         if (!ret)
 103                 ret = err;
 104 out:
 105         return ret;
 106 }
 107
 108 static long __do_fsync(unsigned int fd, int datasync)
 109 {
 110         struct file *file;
 111         int ret = -EBADF;
 112
 113         file = fget(fd);
 114         if (file) {
 115                 ret = do_fsync(file, datasync);
 116                 fput(file);
 117         }
 118         return ret;
 119 }
 120
 121 SYSCALL_DEFINE1(fsync, unsigned int, fd)
 122 {
 123         return __do_fsync(fd, 0);
 124 }
 125
 126 SYSCALL_DEFINE1(fdatasync, unsigned int, fd)
 127 {
 128         return __do_fsync(fd, 1);
 129 }
 130
 131 /*
 132  * sys_sync_file_range() permits finely controlled syncing over a segment of
 133  * a file in the range offset .. (offset+nbytes-1) inclusive.  If nbytes is
 134  * zero then sys_sync_file_range() will operate from offset out to EOF.
 135  *
 136  * The flag bits are:
 137  *
 138  * SYNC_FILE_RANGE_WAIT_BEFORE: wait upon writeout of all pages in the range
 139  * before performing the write.
 140  *
 141  * SYNC_FILE_RANGE_WRITE: initiate writeout of all those dirty pages in the
 142  * range which are not presently under writeback. Note that this may block for
 143  * significant periods due to exhaustion of disk request structures.
 144  *
 145  * SYNC_FILE_RANGE_WAIT_AFTER: wait upon writeout of all pages in the range
 146  * after performing the write.
 147  *
 148  * Useful combinations of the flag bits are:
 149  *
 150  * SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE: ensures that all pages
 151  * in the range which were dirty on entry to sys_sync_file_range() are placed
 152  * under writeout.  This is a start-write-for-data-integrity operation.
 153  *
 154  * SYNC_FILE_RANGE_WRITE: start writeout of all dirty pages in the range which
 155  * are not presently under writeout.  This is an asynchronous flush-to-disk
 156  * operation.  Not suitable for data integrity operations.
 157  *
 158  * SYNC_FILE_RANGE_WAIT_BEFORE (or SYNC_FILE_RANGE_WAIT_AFTER): wait for
 159  * completion of writeout of all pages in the range.  This will be used after an
 160  * earlier SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE operation to wait
 161  * for that operation to complete and to return the result.
 162  *
 163  * SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE|SYNC_FILE_RANGE_WAIT_AFTER:
 164  * a traditional sync() operation.  This is a write-for-data-integrity operation
 165  * which will ensure that all pages in the range which were dirty on entry to
 166  * sys_sync_file_range() are committed to disk.
 167  *
 168  *
 169  * SYNC_FILE_RANGE_WAIT_BEFORE and SYNC_FILE_RANGE_WAIT_AFTER will detect any
 170  * I/O errors or ENOSPC conditions and will return those to the caller, after
 171  * clearing the EIO and ENOSPC flags in the address_space.
 172  *
 173  * It should be noted that none of these operations write out the file's
 174  * metadata.  So unless the application is strictly performing overwrites of
 175  * already-instantiated disk blocks, there are no guarantees here that the data
 176  * will be available after a crash.
 177  */
 178 SYSCALL_DEFINE(sync_file_range)(int fd, loff_t offset, loff_t nbytes,
 179                                 unsigned int flags)
 180 {
 181         int ret;
 182         struct file *file;
 183         loff_t endbyte;                 /* inclusive */
 184         int fput_needed;
 185         umode_t i_mode;
 186
 187         ret = -EINVAL;
 188         if (flags & ~VALID_FLAGS)
 189                 goto out;
 190
 191         endbyte = offset + nbytes;
 192
 193         if ((s64)offset < 0)
 194                 goto out;
 195         if ((s64)endbyte < 0)
 196                 goto out;
 197         if (endbyte < offset)
 198                 goto out;
 199
 200         if (sizeof(pgoff_t) == 4) {
 201                 if (offset >= (0x100000000ULL << PAGE_CACHE_SHIFT)) {
 202                         /*
 203                          * The range starts outside a 32 bit machine's
 204                          * pagecache addressing capabilities.  Let it "succeed"
 205                          */
 206                         ret = 0;
 207                         goto out;
 208                 }
 209                 if (endbyte >= (0x100000000ULL << PAGE_CACHE_SHIFT)) {
 210                         /*
 211                          * Out to EOF
 212                          */
 213                         nbytes = 0;
 214                 }
 215         }
 216
 217         if (nbytes == 0)
 218                 endbyte = LLONG_MAX;
 219         else
 220                 endbyte--;              /* inclusive */
 221
 222         ret = -EBADF;
 223         file = fget_light(fd, &fput_needed);
 224         if (!file)
 225                 goto out;
 226
 227         i_mode = file->f_path.dentry->d_inode->i_mode;
 228         ret = -ESPIPE;
 229         if (!S_ISREG(i_mode) && !S_ISBLK(i_mode) && !S_ISDIR(i_mode) &&
 230                         !S_ISLNK(i_mode))
 231                 goto out_put;
 232
 233         ret = do_sync_mapping_range(file->f_mapping, offset, endbyte, flags);
 234 out_put:
 235         fput_light(file, fput_needed);
 236 out:
 237         return ret;
 238 }
 239 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
 240 asmlinkage long SyS_sync_file_range(long fd, loff_t offset, loff_t nbytes,
 241                                     long flags)
 242 {
 243         return SYSC_sync_file_range((int) fd, offset, nbytes,
 244                                     (unsigned int) flags);
 245 }
 246 SYSCALL_ALIAS(sys_sync_file_range, SyS_sync_file_range);
 247 #endif
 248
 249 /* It would be nice if people remember that not all the world's an i386
 250    when they introduce new system calls */
 251 SYSCALL_DEFINE(sync_file_range2)(int fd, unsigned int flags,
 252                                  loff_t offset, loff_t nbytes)
 253 {
 254         return sys_sync_file_range(fd, offset, nbytes, flags);
 255 }
 256 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
 257 asmlinkage long SyS_sync_file_range2(long fd, long flags,
 258                                      loff_t offset, loff_t nbytes)
 259 {
 260         return SYSC_sync_file_range2((int) fd, (unsigned int) flags,
 261                                      offset, nbytes);
 262 }
 263 SYSCALL_ALIAS(sys_sync_file_range2, SyS_sync_file_range2);
 264 #endif
 265
 266 /*
 267  * `endbyte' is inclusive
 268  */
 269 int do_sync_mapping_range(struct address_space *mapping, loff_t offset,
 270                           loff_t endbyte, unsigned int flags)
 271 {
 272         int ret;
 273
 274         if (!mapping) {
 275                 ret = -EINVAL;
 276                 goto out;
 277         }
 278
 279         ret = 0;
 280         if (flags & SYNC_FILE_RANGE_WAIT_BEFORE) {
 281                 ret = wait_on_page_writeback_range(mapping,
 282                                         offset >> PAGE_CACHE_SHIFT,
 283                                         endbyte >> PAGE_CACHE_SHIFT);
 284                 if (ret < 0)
 285                         goto out;
 286         }
 287
 288         if (flags & SYNC_FILE_RANGE_WRITE) {
 289                 ret = __filemap_fdatawrite_range(mapping, offset, endbyte,
 290                                                 WB_SYNC_ALL);
 291                 if (ret < 0)
 292                         goto out;
 293         }
 294
 295         if (flags & SYNC_FILE_RANGE_WAIT_AFTER) {
 296                 ret = wait_on_page_writeback_range(mapping,
 297                                         offset >> PAGE_CACHE_SHIFT,
 298                                         endbyte >> PAGE_CACHE_SHIFT);
 299         }
 300 out:
 301         return ret;
 302 }
 303 EXPORT_SYMBOL_GPL(do_sync_mapping_range);