On 1/29/21 1:59 AM, Stanislav Fomichev wrote:

On Thu, Jan 28, 2021 at 4:52 PM Daniel Borkmann [off-list ref] wrote:

quoted

On 1/28/21 12:28 AM, Stanislav Fomichev wrote:

quoted

Those hooks run as BPF_CGROUP_RUN_SA_PROG_LOCK and operate on
a locked socket.

Signed-off-by: Stanislav Fomichev <redacted>
---
   net/core/filter.c                                 | 4 ++++
   tools/testing/selftests/bpf/progs/recvmsg4_prog.c | 5 +++++
   tools/testing/selftests/bpf/progs/recvmsg6_prog.c | 5 +++++
   3 files changed, 14 insertions(+)

diff --git a/net/core/filter.c b/net/core/filter.c
index ba436b1d70c2..e15d4741719a 100644
--- a/net/core/filter.c
+++ b/net/core/filter.c

@@ -7023,6 +7023,8 @@ sock_addr_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
               case BPF_CGROUP_INET6_BIND:
               case BPF_CGROUP_INET4_CONNECT:
               case BPF_CGROUP_INET6_CONNECT:
+             case BPF_CGROUP_UDP4_RECVMSG:
+             case BPF_CGROUP_UDP6_RECVMSG:
               case BPF_CGROUP_UDP4_SENDMSG:
               case BPF_CGROUP_UDP6_SENDMSG:
               case BPF_CGROUP_INET4_GETPEERNAME:

@@ -7039,6 +7041,8 @@ sock_addr_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
               case BPF_CGROUP_INET6_BIND:
               case BPF_CGROUP_INET4_CONNECT:
               case BPF_CGROUP_INET6_CONNECT:
+             case BPF_CGROUP_UDP4_RECVMSG:
+             case BPF_CGROUP_UDP6_RECVMSG:
               case BPF_CGROUP_UDP4_SENDMSG:
               case BPF_CGROUP_UDP6_SENDMSG:
               case BPF_CGROUP_INET4_GETPEERNAME:

Looks good overall, also thanks for adding the test cases! I was about to apply, but noticed one
small nit that would be good to get resolved before that. Above you now list all the attach hooks
for sock_addr ctx, so we should just remove the whole switch that tests on prog->expected_attach_type
altogether in this last commit.

Sure, I can resend tomorrow.
But do you think it's safe and there won't ever be another sock_addr
hook that runs with an unlocked socket?

Ok, that rationale seems reasonable to keep the series as is. It probably makes sense to add a
small comment at least to the commit log to explain the reasoning, I can do so while applying.
So no need for v3, thanks!