bpf: Allow bpf_dynptr_from_skb() for tp_btf
authorPhilo Lu <lulie@linux.alibaba.com>
Wed, 11 Sep 2024 03:37:18 +0000 (11:37 +0800)
committerMartin KaFai Lau <martin.lau@kernel.org>
Wed, 11 Sep 2024 15:56:42 +0000 (08:56 -0700)
commitffc83860d8c09705d8e83474b8c6ec4d1d3dca41
treea06bcdff926d039aa03a50403f216b62ceb23b46
parentedd3f6f7588c713477e1299c38c84dcd91a7f148
bpf: Allow bpf_dynptr_from_skb() for tp_btf

Making tp_btf able to use bpf_dynptr_from_skb(), which is useful for skb
parsing, especially for non-linear paged skb data. This is achieved by
adding KF_TRUSTED_ARGS flag to bpf_dynptr_from_skb and registering it
for TRACING progs. With KF_TRUSTED_ARGS, args from fentry/fexit are
excluded, so that unsafe progs like fexit/__kfree_skb are not allowed.

We also need the skb dynptr to be read-only in tp_btf. Because
may_access_direct_pkt_data() returns false by default when checking
bpf_dynptr_from_skb, there is no need to add BPF_PROG_TYPE_TRACING to it
explicitly.

Suggested-by: Martin KaFai Lau <martin.lau@linux.dev>
Signed-off-by: Philo Lu <lulie@linux.alibaba.com>
Acked-by: Martin KaFai Lau <martin.lau@kernel.org>
Link: https://lore.kernel.org/r/20240911033719.91468-5-lulie@linux.alibaba.com
Signed-off-by: Martin KaFai Lau <martin.lau@kernel.org>
net/core/filter.c