Re: [PATCH bpf-next 2/2] selftests/bpf: fix "metadata marker" getting overwritten by the netstack

From: Alexander Lobakin <aleksander.lobakin@intel.com>
To: "Toke Høiland-Jørgensen" <toke@redhat.com>
Cc: Alexei Starovoitov <ast@kernel.org>,
	Daniel Borkmann <daniel@iogearbox.net>,
	Andrii Nakryiko <andrii@kernel.org>,
	Martin KaFai Lau <martin.lau@linux.dev>,
	Maciej Fijalkowski <maciej.fijalkowski@intel.com>,
	Larysa Zaremba <larysa.zaremba@intel.com>,
	Ilya Leoshkevich <iii@linux.ibm.com>, Song Liu <song@kernel.org>,
	Jesper Dangaard Brouer <hawk@kernel.org>,
	Jakub Kicinski <kuba@kernel.org>, <bpf@vger.kernel.org>,
	<netdev@vger.kernel.org>, <linux-kernel@vger.kernel.org>
Subject: Re: [PATCH bpf-next 2/2] selftests/bpf: fix "metadata marker" getting overwritten by the netstack
Date: Fri, 17 Mar 2023 14:38:44 +0100	[thread overview]
Message-ID: <cdf1ae24-eebb-a9f8-65d3-01876334a33c@intel.com> (raw)
In-Reply-To: <875yb0a25h.fsf@toke.dk>

From: Toke Høiland-Jørgensen <toke@redhat.com>
Date: Thu, 16 Mar 2023 21:10:02 +0100

> Alexander Lobakin <aleksander.lobakin@intel.com> 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 :)

I just blinked my eyes once and Ilya already came back with the detailed
stacktrace, so it's almost entirely his work <O

> 
>> Fixes: 9c94bbf9a87b ("xdp: recycle Page Pool backed skbs built from XDP frames")
>> Reported-by: Alexei Starovoitov <ast@kernel.org>
>> Link: https://lore.kernel.org/bpf/CAADnVQ+B_JOU+EpP=DKhbY9yXdN6GiRPnpTTXfEZ9sNkUeb-yQ@mail.gmail.com
>> Reported-by: Ilya Leoshkevich <iii@linux.ibm.com> # + debugging
>> Link: https://lore.kernel.org/bpf/8341c1d9f935f410438e79d3bd8a9cc50aefe105.camel@linux.ibm.com
>> Signed-off-by: Alexander Lobakin <aleksander.lobakin@intel.com>
> 
> Acked-by: Toke Høiland-Jørgensen <toke@redhat.com>
> 

Thanks! That was quick :D

Olek