Re: [PATCH net v2] net: dsa: tag_rtl4_a: Bump min packet size

[Linus Walleij]

On Mon, Oct 30, 2023 at 11:20 PM Vladimir Oltean <olteanv@xxxxxxxxx> wrote:

> I see commit 86dd9868b878 ("net: dsa: tag_rtl4_a: Support also egress tags")
> also mentions: "Qingfang came up with the solution: we need to pad the
> ethernet frame to 60 bytes using eth_skb_pad(), then the switch will
> happily accept frames with custom tags.". So the __skb_put_padto() was
> something very empirical in the first place.
>
> Since it's all problematic, would you mind removing the __skb_put_padto()
> altogether from rtl4a_tag_xmit(), and let me know what is the output for
> the following sweep through packet sizes? I truly wonder if it's just
> for small and large packets that we see packet drops, or if it's something
> repetitive throughout the range as well.
>
> for size in $(seq 0 1476); do if ping 10.0.0.56 -s $size -W 1 -c 1 -q >/dev/null; then echo "$((size + 42)): OK"; else echo "$((size + 42)): NOK"; fi; done

The weird thing is that if I remove the __skb_put_padto()
call, ping doesn't work at all. Somehow the packets are
corrupted, because they sure get out of the switch and I
can see them arriving with tcpdump on the host.

root@OpenWrt:/# for size in $(seq 0 1476); do if ping 192.168.1.137 -s $size -W
1 -c 1 -q >/dev/null; then echo "$((size + 42)): OK"; else echo "$((size + 42)):
 NOK"; fi; done
42: NOK
43: NOK
44: NOK
45: NOK
46: NOK
47: NOK
48: NOK
49: NOK
50: NOK
51: NOK
(...)
1509: NOK
1510: NOK
1511: NOK
1512: NOK
1513: NOK
1514: NOK
1515: NOK
1516: NOK
1517: NOK
1518: NOK

This of course make no sense, since the padding function should do nothing
when the packet is bigger than 60 bytes.

So what we are seeing is some kind of side effect from the usage of
__skb_put_padto() I suppose? But I have no idea what that is, I looked
at the function and what it calls down to __skb_pad().

I'm testing skb_linearize(), which seems to be called on this path...

TCPdump on the host looks like this:
# tcpdump -i enp7s0
dropped privs to tcpdump
tcpdump: verbose output suppressed, use -v[v]... for full protocol decode
listening on enp7s0, link-type EN10MB (Ethernet), snapshot length 262144 bytes
23:28:55.184019 IP _gateway > fedora: ICMP echo request, id 2461, seq
0, length 27
23:28:56.205294 IP _gateway > fedora: ICMP echo request, id 2462, seq
0, length 28
23:28:57.226495 IP _gateway > fedora: ICMP echo request, id 2463, seq
0, length 29
23:28:58.248013 IP _gateway > fedora: ICMP echo request, id 2464, seq
0, length 30
23:28:59.269157 IP _gateway > fedora: ICMP echo request, id 2465, seq
0, length 31
23:29:00.290443 IP _gateway > fedora: ICMP echo request, id 2466, seq
0, length 32
23:29:01.698700 IP _gateway > fedora: ICMP echo request, id 2467, seq
0, length 33
23:29:02.332131 IP _gateway > fedora: ICMP echo request, id 2468, seq
0, length 34
23:29:03.352442 IP _gateway > fedora: ICMP echo request, id 2469, seq
0, length 35
(...)
23:53:33.834706 IP _gateway > fedora: ICMP echo request, id 4000, seq
0, length 1475
23:53:34.854946 IP _gateway > fedora: ICMP echo request, id 4001, seq
0, length 1476
23:53:36.258777 IP truncated-ip - 1 bytes missing! _gateway > fedora:
ICMP echo request, id 4002, seq 0, length 1477
23:53:36.896654 IP truncated-ip - 2 bytes missing! _gateway > fedora:
ICMP echo request, id 4003, seq 0, length 1478
23:53:37.918022 IP truncated-ip - 3 bytes missing! _gateway > fedora:
ICMP echo request, id 4004, seq 0, length 1479
23:53:38.938355 IP truncated-ip - 4 bytes missing! _gateway > fedora:
ICMP echo request, id 4005, seq 0, length 1480
23:53:39.958451 IP truncated-ip - 4 bytes missing! _gateway > fedora:
ICMP echo request, id 4006, seq 0, length 1480
23:53:40.978598 IP truncated-ip - 4 bytes missing! _gateway > fedora:
ICMP echo request, id 4007, seq 0, length 1480
23:53:41.998991 IP truncated-ip - 4 bytes missing! _gateway > fedora:
ICMP echo request, id 4008, seq 0, length 1480
23:53:43.020309 IP truncated-ip - 4 bytes missing! _gateway > fedora:
ICMP echo request, id 4010, seq 0, length 1480

Here you can incidentally also see what happens if we don't pad the big packets:
the packet gets truncated.

Yours,
Linus Walleij