bpf: pull before calling skb_postpull_rcsum()
authorJakub Kicinski <kuba@kernel.org>
Tue, 20 Dec 2022 00:47:00 +0000 (16:47 -0800)
committerMartin KaFai Lau <martin.lau@kernel.org>
Tue, 20 Dec 2022 23:58:35 +0000 (15:58 -0800)
commit54c3f1a81421f85e60ae2eaae7be3727a09916ee
tree278a501289a2323cb1e78f7a0b3b7201152bf33f
parentcc074822465d18a2d39e0b3e2b48b6766a568db2
bpf: pull before calling skb_postpull_rcsum()

Anand hit a BUG() when pulling off headers on egress to a SW tunnel.
We get to skb_checksum_help() with an invalid checksum offset
(commit d7ea0d9df2a6 ("net: remove two BUG() from skb_checksum_help()")
converted those BUGs to WARN_ONs()).
He points out oddness in how skb_postpull_rcsum() gets used.
Indeed looks like we should pull before "postpull", otherwise
the CHECKSUM_PARTIAL fixup from skb_postpull_rcsum() will not
be able to do its job:

if (skb->ip_summed == CHECKSUM_PARTIAL &&
    skb_checksum_start_offset(skb) < 0)
skb->ip_summed = CHECKSUM_NONE;

Reported-by: Anand Parthasarathy <anpartha@meta.com>
Fixes: 6578171a7ff0 ("bpf: add bpf_skb_change_proto helper")
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Acked-by: Stanislav Fomichev <sdf@google.com>
Link: https://lore.kernel.org/r/20221220004701.402165-1-kuba@kernel.org
Signed-off-by: Martin KaFai Lau <martin.lau@kernel.org>
net/core/filter.c