Alexander Lobakin <aleksander.lobakin@xxxxxxxxx> writes:
> Alexei noticed xdp_do_redirect test on BPF CI started failing on
> BE systems after skb PP recycling was enabled:
>
> test_xdp_do_redirect:PASS:prog_run 0 nsec
> test_xdp_do_redirect:PASS:pkt_count_xdp 0 nsec
> test_xdp_do_redirect:PASS:pkt_count_zero 0 nsec
> test_xdp_do_redirect:FAIL:pkt_count_tc unexpected pkt_count_tc: actual
> 220 != expected 9998
> test_max_pkt_size:PASS:prog_run_max_size 0 nsec
> test_max_pkt_size:PASS:prog_run_too_big 0 nsec
> close_netns:PASS:setns 0 nsec
> #289 xdp_do_redirect:FAIL
> Summary: 270/1674 PASSED, 30 SKIPPED, 1 FAILED
>
> and it doesn't happen on LE systems.
> Ilya then hunted it down to:
>
> #0 0x0000000000aaeee6 in neigh_hh_output (hh=0x83258df0,
> skb=0x88142200) at linux/include/net/neighbour.h:503
> #1 0x0000000000ab2cda in neigh_output (skip_cache=false,
> skb=0x88142200, n=<optimized out>) at linux/include/net/neighbour.h:544
> #2 ip6_finish_output2 (net=net@entry=0x88edba00, sk=sk@entry=0x0,
> skb=skb@entry=0x88142200) at linux/net/ipv6/ip6_output.c:134
> #3 0x0000000000ab4cbc in __ip6_finish_output (skb=0x88142200, sk=0x0,
> net=0x88edba00) at linux/net/ipv6/ip6_output.c:195
> #4 ip6_finish_output (net=0x88edba00, sk=0x0, skb=0x88142200) at
> linux/net/ipv6/ip6_output.c:206
>
> xdp_do_redirect test places a u32 marker (0x42) right before the Ethernet
> header to check it then in the XDP program and return %XDP_ABORTED if it's
> not there. Neigh xmit code likes to round up hard header length to speed
> up copying the header, so it overwrites two bytes in front of the Eth
> header. On LE systems, 0x42 is one byte at `data - 4`, while on BE it's
> `data - 1`, what explains why it happens only there.
> It didn't happen previously due to that %XDP_PASS meant the page will be
> discarded and replaced by a new one, but now it can be recycled as well,
> while bpf_test_run code doesn't reinitialize the content of recycled
> pages. This mark is limited to this particular test and its setup though,
> so there's no need to predict 1000 different possible cases. Just move
> it 4 bytes to the left, still keeping it 32 bit to match on more
> bytes.
Wow, this must have been annoying to track down - nice work :)
> Fixes: 9c94bbf9a87b ("xdp: recycle Page Pool backed skbs built from XDP frames")
> Reported-by: Alexei Starovoitov <ast@xxxxxxxxxx>
> Link: https://lore.kernel.org/bpf/CAADnVQ+B_JOU+EpP=DKhbY9yXdN6GiRPnpTTXfEZ9sNkUeb-yQ@xxxxxxxxxxxxxx
> Reported-by: Ilya Leoshkevich <iii@xxxxxxxxxxxxx> # + debugging
> Link: https://lore.kernel.org/bpf/8341c1d9f935f410438e79d3bd8a9cc50aefe105.camel@xxxxxxxxxxxxx
> Signed-off-by: Alexander Lobakin <aleksander.lobakin@xxxxxxxxx>
Acked-by: Toke Høiland-Jørgensen <toke@xxxxxxxxxx>
