]> git.proxmox.com Git - mirror_ubuntu-kernels.git/commitdiff
bpf: fix checking xdp_adjust_head on tail calls
authorDaniel Borkmann <daniel@iogearbox.net>
Mon, 17 Apr 2017 01:12:07 +0000 (03:12 +0200)
committerDavid S. Miller <davem@davemloft.net>
Mon, 17 Apr 2017 19:51:57 +0000 (15:51 -0400)
Commit 17bedab27231 ("bpf: xdp: Allow head adjustment in XDP prog")
added the xdp_adjust_head bit to the BPF prog in order to tell drivers
that the program that is to be attached requires support for the XDP
bpf_xdp_adjust_head() helper such that drivers not supporting this
helper can reject the program. There are also drivers that do support
the helper, but need to check for xdp_adjust_head bit in order to move
packet metadata prepended by the firmware away for making headroom.

For these cases, the current check for xdp_adjust_head bit is insufficient
since there can be cases where the program itself does not use the
bpf_xdp_adjust_head() helper, but tail calls into another program that
uses bpf_xdp_adjust_head(). As such, the xdp_adjust_head bit is still
set to 0. Since the first program has no control over which program it
calls into, we need to assume that bpf_xdp_adjust_head() helper is used
upon tail calls. Thus, for the very same reasons in cb_access, set the
xdp_adjust_head bit to 1 when the main program uses tail calls.

Fixes: 17bedab27231 ("bpf: xdp: Allow head adjustment in XDP prog")
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Cc: Martin KaFai Lau <kafai@fb.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
kernel/bpf/syscall.c

index ee5c969d596392051c55c7f01d0fb789a83d626a..821f9e807de5705d5b4d65e502fb13a06d3215bb 100644 (file)
@@ -623,6 +623,7 @@ static void fixup_bpf_calls(struct bpf_prog *prog)
                                 * in the program array.
                                 */
                                prog->cb_access = 1;
+                               prog->xdp_adjust_head = 1;
 
                                /* mark bpf_tail_call as different opcode
                                 * to avoid conditional branch in