xfs: ensure sync write errors are returned
authorDave Chinner <dchinner@redhat.com>
Mon, 10 Jan 2011 23:13:53 +0000 (10:13 +1100)
committerDave Chinner <david@fromorbit.com>
Mon, 10 Jan 2011 23:13:53 +0000 (10:13 +1100)
xfs_file_aio_write() only returns the error from synchronous
flushing of the data and inode if error == 0. At the point where
error is being checked, it is guaranteed to be > 0. Therefore any
errors returned by the data or fsync flush will never be returned.
Fix the checks so we overwrite the current error once and only if an
error really occurred.

Signed-off-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Alex Elder <aelder@sgi.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
fs/xfs/linux-2.6/xfs_file.c

index ba8ad422a16506fdea605c7a215e572b3b3a8f6a..10b7fb4807a6db88ad65afae3eabf6367ce4926f 100644 (file)
@@ -574,7 +574,7 @@ xfs_file_aio_write(
        struct inode            *inode = mapping->host;
        struct xfs_inode        *ip = XFS_I(inode);
        struct xfs_mount        *mp = ip->i_mount;
-       ssize_t                 ret = 0, error = 0;
+       ssize_t                 ret = 0;
        int                     ioflags = 0;
        xfs_fsize_t             isize, new_size;
        int                     iolock;
@@ -590,9 +590,9 @@ xfs_file_aio_write(
        if (file->f_mode & FMODE_NOCMTIME)
                ioflags |= IO_INVIS;
 
-       error = generic_segment_checks(iovp, &nr_segs, &ocount, VERIFY_READ);
-       if (error)
-               return error;
+       ret = generic_segment_checks(iovp, &nr_segs, &ocount, VERIFY_READ);
+       if (ret)
+               return ret;
 
        count = ocount;
        if (count == 0)
@@ -616,9 +616,9 @@ relock:
        xfs_ilock(ip, XFS_ILOCK_EXCL|iolock);
 
 start:
-       error = -generic_write_checks(file, &pos, &count,
+       ret = generic_write_checks(file, &pos, &count,
                                        S_ISBLK(inode->i_mode));
-       if (error) {
+       if (ret) {
                xfs_iunlock(ip, XFS_ILOCK_EXCL|iolock);
                goto out_unlock_mutex;
        }
@@ -660,8 +660,8 @@ start:
         */
 
        if (pos > ip->i_size) {
-               error = xfs_zero_eof(ip, pos, ip->i_size);
-               if (error) {
+               ret = -xfs_zero_eof(ip, pos, ip->i_size);
+               if (ret) {
                        xfs_iunlock(ip, XFS_ILOCK_EXCL);
                        goto out_unlock_internal;
                }
@@ -674,8 +674,8 @@ start:
         * by root.  This keeps people from modifying setuid and
         * setgid binaries.
         */
-       error = -file_remove_suid(file);
-       if (unlikely(error))
+       ret = file_remove_suid(file);
+       if (unlikely(ret))
                goto out_unlock_internal;
 
        /* We can write back this queue in page reclaim */
@@ -684,10 +684,10 @@ start:
        if ((ioflags & IO_ISDIRECT)) {
                if (mapping->nrpages) {
                        WARN_ON(need_i_mutex == 0);
-                       error = xfs_flushinval_pages(ip,
+                       ret = -xfs_flushinval_pages(ip,
                                        (pos & PAGE_CACHE_MASK),
                                        -1, FI_REMAPF_LOCKED);
-                       if (error)
+                       if (ret)
                                goto out_unlock_internal;
                }
 
@@ -720,24 +720,22 @@ start:
                }
        } else {
                int enospc = 0;
-               ssize_t ret2 = 0;
 
 write_retry:
                trace_xfs_file_buffered_write(ip, count, iocb->ki_pos, ioflags);
-               ret2 = generic_file_buffered_write(iocb, iovp, nr_segs,
+               ret = generic_file_buffered_write(iocb, iovp, nr_segs,
                                pos, &iocb->ki_pos, count, ret);
                /*
                 * if we just got an ENOSPC, flush the inode now we
                 * aren't holding any page locks and retry *once*
                 */
-               if (ret2 == -ENOSPC && !enospc) {
-                       error = xfs_flush_pages(ip, 0, -1, 0, FI_NONE);
-                       if (error)
+               if (ret == -ENOSPC && !enospc) {
+                       ret = xfs_flush_pages(ip, 0, -1, 0, FI_NONE);
+                       if (ret)
                                goto out_unlock_internal;
                        enospc = 1;
                        goto write_retry;
                }
-               ret = ret2;
        }
 
        current->backing_dev_info = NULL;
@@ -753,7 +751,6 @@ write_retry:
                xfs_iunlock(ip, XFS_ILOCK_EXCL);
        }
 
-       error = -ret;
        if (ret <= 0)
                goto out_unlock_internal;
 
@@ -762,23 +759,23 @@ write_retry:
        /* Handle various SYNC-type writes */
        if ((file->f_flags & O_DSYNC) || IS_SYNC(inode)) {
                loff_t end = pos + ret - 1;
-               int error2;
+               int error, error2;
 
                xfs_iunlock(ip, iolock);
                if (need_i_mutex)
                        mutex_unlock(&inode->i_mutex);
 
-               error2 = filemap_write_and_wait_range(mapping, pos, end);
-               if (!error)
-                       error = error2;
+               error = filemap_write_and_wait_range(mapping, pos, end);
                if (need_i_mutex)
                        mutex_lock(&inode->i_mutex);
                xfs_ilock(ip, iolock);
 
                error2 = -xfs_file_fsync(file,
                                         (file->f_flags & __O_SYNC) ? 0 : 1);
-               if (!error)
-                       error = error2;
+               if (error)
+                       ret = error;
+               else if (error2)
+                       ret = error2;
        }
 
  out_unlock_internal:
@@ -800,7 +797,7 @@ write_retry:
  out_unlock_mutex:
        if (need_i_mutex)
                mutex_unlock(&inode->i_mutex);
-       return -error;
+       return ret;
 }
 
 STATIC int