GFS2: Reserve journal space for quota change in do_grow

If a GFS2 file system is mounted with quotas and a file is grown
in such a way that its free blocks for the allocation are represented
in a secondary bitmap, GFS2 ran out of blocks in the transaction.
That resulted in "fatal: assertion "tr->tr_num_buf <= tr->tr_blocks".
This patch reserves extra blocks for the quota change so the
transaction has enough space.

Signed-off-by: Bob Peterson <rpeterso@redhat.com>
Signed-off-by: Steven Whitehouse <swhiteho@redhat.com>

fs/gfs2/bmap.c

@@ -1232,7 +1232,9 @@ static int do_grow(struct inode *inode, u64 size)
 		unstuff = 1;
 	}
 
-	error = gfs2_trans_begin(sdp, RES_DINODE + RES_STATFS + RES_RG_BIT, 0);
+	error = gfs2_trans_begin(sdp, RES_DINODE + RES_STATFS + RES_RG_BIT +
+				 (sdp->sd_args.ar_quota == GFS2_QUOTA_OFF ?
+				  0 : RES_QUOTA), 0);
 	if (error)
 		goto do_grow_release;