[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <152405341177.30730.1255927134630901578.stgit@firesoul>
Date: Wed, 18 Apr 2018 14:10:11 +0200
From: Jesper Dangaard Brouer <brouer@...hat.com>
To: Daniel Borkmann <borkmann@...earbox.net>,
Alexei Starovoitov <alexei.starovoitov@...il.com>
Cc: netdev@...r.kernel.org, Jesper Dangaard Brouer <brouer@...hat.com>
Subject: [RFC net-next PATCH 1/2] bpf: avoid clear xdp_frame area again
Avoid clearing xdp_frame area if this was already done by prevous
invocations of bpf_xdp_adjust_head.
The xdp_adjust_head helper can be called multiple times by the
bpf_prog. If increasing the packet header size (with a negative
offset), kernel must assume bpf_prog store valuable information here,
and not clear this information.
In case of extending header into xdp_frame area the kernel clear this
area to avoid any info leaking.
The bug in the current implementation is that if existing xdp->data
pointer have already been moved into xdp_frame area, then memory is
cleared between new-data pointer and xdp_frame-end, which covers an
area that might contain information store by BPF-prog (as curr
xdp->data lays between those pointers).
Fixes: 6dfb970d3dbd ("xdp: avoid leaking info stored in frame data on page reuse")
Signed-off-by: Jesper Dangaard Brouer <brouer@...hat.com>
---
net/core/filter.c | 7 +++++++
1 file changed, 7 insertions(+)
diff --git a/net/core/filter.c b/net/core/filter.c
index a374b8560bc4..15e9b5477360 100644
--- a/net/core/filter.c
+++ b/net/core/filter.c
@@ -2705,6 +2705,13 @@ BPF_CALL_2(bpf_xdp_adjust_head, struct xdp_buff *, xdp, int, offset)
if (data < xdp_frame_end) {
unsigned long clearlen = xdp_frame_end - data;
+ /* Handle if prev call adjusted xdp->data into xdp_frame area */
+ if (unlikely(xdp->data < xdp_frame_end)) {
+ if (data < xdp->data)
+ clearlen = xdp->data - data;
+ else
+ clearlen = 0;
+ }
memset(data, 0, clearlen);
}
Powered by blists - more mailing lists