xfs: ensure correct timestamp updates from truncate
authorChristoph Hellwig <hch@infradead.org>
Sun, 9 Feb 2014 23:35:22 +0000 (10:35 +1100)
committerDave Chinner <david@fromorbit.com>
Sun, 9 Feb 2014 23:35:22 +0000 (10:35 +1100)
The VFS doesn't set the proper ATTR_CTIME and ATTR_MTIME values for
truncate, so filesystems have to manually add them.  The
introduction of xfs_setattr_time accidentally broke this special
case an caused a regression in generic/313.  Fix this by removing
the local mask variable in xfs_setattr_size so that we only have a
single place to keep the attribute information.

cc: <stable@vger.kernel.org>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reported-by: Fengguang Wu <fengguang.wu@intel.com>
Reviewed-by: Brian Foster <bfoster@redhat.com>
Reviewed-by: Jie Liu <jeff.liu@oracle.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
fs/xfs/xfs_iops.c

index f35d5c9..9ddfb81 100644 (file)
@@ -705,7 +705,6 @@ xfs_setattr_size(
 {
        struct xfs_mount        *mp = ip->i_mount;
        struct inode            *inode = VFS_I(ip);
-       int                     mask = iattr->ia_valid;
        xfs_off_t               oldsize, newsize;
        struct xfs_trans        *tp;
        int                     error;
@@ -726,8 +725,8 @@ xfs_setattr_size(
 
        ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
        ASSERT(S_ISREG(ip->i_d.di_mode));
-       ASSERT((mask & (ATTR_UID|ATTR_GID|ATTR_ATIME|ATTR_ATIME_SET|
-                       ATTR_MTIME_SET|ATTR_KILL_PRIV|ATTR_TIMES_SET)) == 0);
+       ASSERT((iattr->ia_valid & (ATTR_UID|ATTR_GID|ATTR_ATIME|ATTR_ATIME_SET|
+               ATTR_MTIME_SET|ATTR_KILL_PRIV|ATTR_TIMES_SET)) == 0);
 
        oldsize = inode->i_size;
        newsize = iattr->ia_size;
@@ -736,7 +735,7 @@ xfs_setattr_size(
         * Short circuit the truncate case for zero length files.
         */
        if (newsize == 0 && oldsize == 0 && ip->i_d.di_nextents == 0) {
-               if (!(mask & (ATTR_CTIME|ATTR_MTIME)))
+               if (!(iattr->ia_valid & (ATTR_CTIME|ATTR_MTIME)))
                        return 0;
 
                /*
@@ -824,10 +823,11 @@ xfs_setattr_size(
         * these flags set.  For all other operations the VFS set these flags
         * explicitly if it wants a timestamp update.
         */
-       if (newsize != oldsize && (!(mask & (ATTR_CTIME | ATTR_MTIME)))) {
+       if (newsize != oldsize &&
+           !(iattr->ia_valid & (ATTR_CTIME | ATTR_MTIME))) {
                iattr->ia_ctime = iattr->ia_mtime =
                        current_fs_time(inode->i_sb);
-               mask |= ATTR_CTIME | ATTR_MTIME;
+               iattr->ia_valid |= ATTR_CTIME | ATTR_MTIME;
        }
 
        /*
@@ -863,9 +863,9 @@ xfs_setattr_size(
                xfs_inode_clear_eofblocks_tag(ip);
        }
 
-       if (mask & ATTR_MODE)
+       if (iattr->ia_valid & ATTR_MODE)
                xfs_setattr_mode(ip, iattr);
-       if (mask & (ATTR_ATIME|ATTR_CTIME|ATTR_MTIME))
+       if (iattr->ia_valid & (ATTR_ATIME|ATTR_CTIME|ATTR_MTIME))
                xfs_setattr_time(ip, iattr);
 
        xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);