Merge branch 'add TCP_BPF_SOCK_OPS_CB_FLAGS to bpf_*sockopt()'

Alan Maguire says:

====================
As previously discussed here [1], long-lived sockets can miss
a chance to set additional callbacks if a sock ops program
was not attached early in their lifetime.  Adding support
to bpf_setsockopt() to set callback flags (and bpf_getsockopt()
to retrieve them) provides other opportunities to enable callbacks,
either directly via a cgroup/setsockopt intercepted setsockopt()
or via a socket iterator.

Patch 1 adds bpf_[get|set]sockopt() support; patch 2 adds testing
for it via a sockops programs, along with verification via a
cgroup/getsockopt program.

Changes since v1 [2]:

- Removed unneeded READ_ONCE() (Martin, patch 1)
- Reworked sockopt test to leave existing tests undisturbed while adding
  test_nonstandard_opt() test to cover the TCP_BPF_SOCK_OPS_CB_FLAGS
  case; test verifies that value set via bpf_setsockopt() is what we
  expect via a call to getsockopt() which is caught by a
  cgroup/getsockopt program to provide the flags value (Martin, patch 2)
- Removed unneeded iterator test (Martin)

[1] https://lore.kernel.org/bpf/f42f157b-6e52-dd4d-3d97-9b86c84c0b00@oracle.com/
[2] https://lore.kernel.org/bpf/20240802152929.2695863-1-alan.maguire@oracle.com/
====================

Signed-off-by: Martin KaFai Lau <martin.lau@kernel.org>

1 files changed, 23 insertions, 3 deletions

diff --git a/tools/testing/selftests/bpf/progs/setget_sockopt.c b/tools/testing/selftests/bpf/progs/setget_sockopt.c
index 60518aed1ffc..6dd4318debbf 100644
--- a/tools/testing/selftests/bpf/progs/setget_sockopt.c
+++ b/tools/testing/selftests/bpf/progs/setget_sockopt.c
@@ -59,6 +59,8 @@ static const struct sockopt_test sol_tcp_tests[] = {
 	{ .opt = TCP_THIN_LINEAR_TIMEOUTS, .flip = 1, },
 	{ .opt = TCP_USER_TIMEOUT, .new = 123400, .expected = 123400, },
 	{ .opt = TCP_NOTSENT_LOWAT, .new = 1314, .expected = 1314, },
+	{ .opt = TCP_BPF_SOCK_OPS_CB_FLAGS, .new = BPF_SOCK_OPS_ALL_CB_FLAGS,
+	  .expected = BPF_SOCK_OPS_ALL_CB_FLAGS, },
 	{ .opt = 0, },
 };
 
@@ -353,11 +355,30 @@ int BPF_PROG(socket_post_create, struct socket *sock, int family,
 	return 1;
 }
 
+SEC("cgroup/getsockopt")
+int _getsockopt(struct bpf_sockopt *ctx)
+{
+	struct bpf_sock *sk = ctx->sk;
+	int *optval = ctx->optval;
+	struct tcp_sock *tp;
+
+	if (!sk || ctx->level != SOL_TCP || ctx->optname != TCP_BPF_SOCK_OPS_CB_FLAGS)
+		return 1;
+
+	tp = bpf_core_cast(sk, struct tcp_sock);
+	if (ctx->optval + sizeof(int) <= ctx->optval_end) {
+		*optval = tp->bpf_sock_ops_cb_flags;
+		ctx->retval = 0;
+	}
+	return 1;
+}
+
 SEC("sockops")
 int skops_sockopt(struct bpf_sock_ops *skops)
 {
 	struct bpf_sock *bpf_sk = skops->sk;
 	struct sock *sk;
+	int flags;
 
 	if (!bpf_sk)
 		return 1;
@@ -384,9 +405,8 @@ int skops_sockopt(struct bpf_sock_ops *skops)
 		nr_passive += !(bpf_test_sockopt(skops, sk) ||
 				test_tcp_maxseg(skops, sk) ||
 				test_tcp_saved_syn(skops, sk));
-		bpf_sock_ops_cb_flags_set(skops,
-					  skops->bpf_sock_ops_cb_flags |
-					  BPF_SOCK_OPS_STATE_CB_FLAG);
+		flags = skops->bpf_sock_ops_cb_flags | BPF_SOCK_OPS_STATE_CB_FLAG;
+		bpf_setsockopt(skops, SOL_TCP, TCP_BPF_SOCK_OPS_CB_FLAGS, &flags, sizeof(flags));
 		break;
 	case BPF_SOCK_OPS_STATE_CB:
 		if (skops->args[1] == BPF_TCP_CLOSE_WAIT)