Merge branch 'Add new args into tcp_congestion_ops' cong_control'

Miao Xu says:

====================
This patchset attempts to add two new arguments into the hookpoint
cong_control in tcp_congestion_ops. The new arguments are inherited
from the caller tcp_cong_control and can be used by any bpf cc prog
that implements its own logic inside this hookpoint.

Please review. Thanks a lot!

Changelog
=====
v2->v3:
  - Fixed the broken selftest caused by the new arguments.
  - Renamed the selftest file name and bpf prog name.

v1->v2:
  - Split the patchset into 3 separate patches.
  - Added highlights in the selftest prog.
  - Removed the dependency on bpf_tcp_helpers.h.
====================
Signed-off-by: Martin KaFai Lau <martin.lau@kernel.org>

include/net/tcp.h

@@ -1172,7 +1172,7 @@ struct tcp_congestion_ops {
 	/* call when packets are delivered to update cwnd and pacing rate,
 	 * after all the ca_state processing. (optional)
 	 */
-	void (*cong_control)(struct sock *sk, const struct rate_sample *rs);
+	void (*cong_control)(struct sock *sk, u32 ack, int flag, const struct rate_sample *rs);
 
 
 	/* new value of cwnd after loss (required) */

net/ipv4/bpf_tcp_ca.c

@@ -107,6 +107,9 @@ static int bpf_tcp_ca_btf_struct_access(struct bpf_verifier_log *log,
 	case offsetof(struct tcp_sock, snd_cwnd_cnt):
 		end = offsetofend(struct tcp_sock, snd_cwnd_cnt);
 		break;
+	case offsetof(struct tcp_sock, snd_cwnd_stamp):
+		end = offsetofend(struct tcp_sock, snd_cwnd_stamp);
+		break;
 	case offsetof(struct tcp_sock, snd_ssthresh):
 		end = offsetofend(struct tcp_sock, snd_ssthresh);
 		break;
@@ -307,7 +310,8 @@ static u32 bpf_tcp_ca_min_tso_segs(struct sock *sk)
 	return 0;
 }
 
-static void bpf_tcp_ca_cong_control(struct sock *sk, const struct rate_sample *rs)
+static void bpf_tcp_ca_cong_control(struct sock *sk, u32 ack, int flag,
+				    const struct rate_sample *rs)
 {
 }
 

net/ipv4/tcp_bbr.c

@@ -1024,7 +1024,7 @@ static void bbr_update_model(struct sock *sk, const struct rate_sample *rs)
 	bbr_update_gains(sk);
 }
 
-__bpf_kfunc static void bbr_main(struct sock *sk, const struct rate_sample *rs)
+__bpf_kfunc static void bbr_main(struct sock *sk, u32 ack, int flag, const struct rate_sample *rs)
 {
 	struct bbr *bbr = inet_csk_ca(sk);
 	u32 bw;

net/ipv4/tcp_input.c

@@ -3541,7 +3541,7 @@ static void tcp_cong_control(struct sock *sk, u32 ack, u32 acked_sacked,
 	const struct inet_connection_sock *icsk = inet_csk(sk);
 
 	if (icsk->icsk_ca_ops->cong_control) {
-		icsk->icsk_ca_ops->cong_control(sk, rs);
+		icsk->icsk_ca_ops->cong_control(sk, ack, flag, rs);
 		return;
 	}
 

tools/testing/selftests/bpf/prog_tests/bpf_tcp_ca.c

@@ -14,6 +14,7 @@
 #include "tcp_ca_incompl_cong_ops.skel.h"
 #include "tcp_ca_unsupp_cong_op.skel.h"
 #include "tcp_ca_kfunc.skel.h"
+#include "bpf_cc_cubic.skel.h"
 
 #ifndef ENOTSUPP
 #define ENOTSUPP 524
@@ -452,6 +453,27 @@ static void test_tcp_ca_kfunc(void)
 	tcp_ca_kfunc__destroy(skel);
 }
 
+static void test_cc_cubic(void)
+{
+	struct bpf_cc_cubic *cc_cubic_skel;
+	struct bpf_link *link;
+
+	cc_cubic_skel = bpf_cc_cubic__open_and_load();
+	if (!ASSERT_OK_PTR(cc_cubic_skel, "bpf_cc_cubic__open_and_load"))
+		return;
+
+	link = bpf_map__attach_struct_ops(cc_cubic_skel->maps.cc_cubic);
+	if (!ASSERT_OK_PTR(link, "bpf_map__attach_struct_ops")) {
+		bpf_cc_cubic__destroy(cc_cubic_skel);
+		return;
+	}
+
+	do_test("bpf_cc_cubic", NULL);
+
+	bpf_link__destroy(link);
+	bpf_cc_cubic__destroy(cc_cubic_skel);
+}
+
 void test_bpf_tcp_ca(void)
 {
 	if (test__start_subtest("dctcp"))
@@ -482,4 +504,6 @@ void test_bpf_tcp_ca(void)
 		test_link_replace();
 	if (test__start_subtest("tcp_ca_kfunc"))
 		test_tcp_ca_kfunc();
+	if (test__start_subtest("cc_cubic"))
+		test_cc_cubic();
 }

tools/testing/selftests/bpf/progs/bpf_cc_cubic.c

+// SPDX-License-Identifier: GPL-2.0-only
+
+/* Highlights:
+ * 1. The major difference between this bpf program and tcp_cubic.c
+ *    is that this bpf program relies on `cong_control` rather than
+ *    `cong_avoid` in the struct tcp_congestion_ops.
+ * 2. Logic such as tcp_cwnd_reduction, tcp_cong_avoid, and
+ *    tcp_update_pacing_rate is bypassed when `cong_control` is
+ *    defined, so moving these logic to `cong_control`.
+ * 3. WARNING: This bpf program is NOT the same as tcp_cubic.c.
+ *    The main purpose is to show use cases of the arguments in
+ *    `cong_control`. For simplicity's sake, it reuses tcp cubic's
+ *    kernel functions.
+ */
+
+#include "vmlinux.h"
+
+#include <bpf/bpf_helpers.h>
+#include <bpf/bpf_tracing.h>
+#include "bpf_tracing_net.h"
+
+#define BPF_STRUCT_OPS(name, args...) \
+SEC("struct_ops/"#name) \
+BPF_PROG(name, args)
+
+#define USEC_PER_SEC 1000000UL
+#define TCP_PACING_SS_RATIO (200)
+#define TCP_PACING_CA_RATIO (120)
+#define TCP_REORDERING (12)
+
+#define min(a, b) ((a) < (b) ? (a) : (b))
+#define max(a, b) ((a) > (b) ? (a) : (b))
+#define after(seq2, seq1) before(seq1, seq2)
+
+extern void cubictcp_init(struct sock *sk) __ksym;
+extern void cubictcp_cwnd_event(struct sock *sk, enum tcp_ca_event event) __ksym;
+extern __u32 cubictcp_recalc_ssthresh(struct sock *sk) __ksym;
+extern void cubictcp_state(struct sock *sk, __u8 new_state) __ksym;
+extern __u32 tcp_reno_undo_cwnd(struct sock *sk) __ksym;
+extern void cubictcp_acked(struct sock *sk, const struct ack_sample *sample) __ksym;
+extern void cubictcp_cong_avoid(struct sock *sk, __u32 ack, __u32 acked) __ksym;
+
+static struct inet_connection_sock *inet_csk(const struct sock *sk)
+{
+	return (struct inet_connection_sock *)sk;
+}
+
+static struct tcp_sock *tcp_sk(const struct sock *sk)
+{
+	return (struct tcp_sock *)sk;
+}
+
+static bool before(__u32 seq1, __u32 seq2)
+{
+	return (__s32)(seq1-seq2) < 0;
+}
+
+static __u64 div64_u64(__u64 dividend, __u64 divisor)
+{
+	return dividend / divisor;
+}
+
+static void tcp_update_pacing_rate(struct sock *sk)
+{
+	const struct tcp_sock *tp = tcp_sk(sk);
+	__u64 rate;
+
+	/* set sk_pacing_rate to 200 % of current rate (mss * cwnd / srtt) */
+	rate = (__u64)tp->mss_cache * ((USEC_PER_SEC / 100) << 3);
+
+	/* current rate is (cwnd * mss) / srtt
+	 * In Slow Start [1], set sk_pacing_rate to 200 % the current rate.
+	 * In Congestion Avoidance phase, set it to 120 % the current rate.
+	 *
+	 * [1] : Normal Slow Start condition is (tp->snd_cwnd < tp->snd_ssthresh)
+	 *	 If snd_cwnd >= (tp->snd_ssthresh / 2), we are approaching
+	 *	 end of slow start and should slow down.
+	 */
+	if (tp->snd_cwnd < tp->snd_ssthresh / 2)
+		rate *= TCP_PACING_SS_RATIO;
+	else
+		rate *= TCP_PACING_CA_RATIO;
+
+	rate *= max(tp->snd_cwnd, tp->packets_out);
+
+	if (tp->srtt_us)
+		rate = div64_u64(rate, (__u64)tp->srtt_us);
+
+	sk->sk_pacing_rate = min(rate, sk->sk_max_pacing_rate);
+}
+
+static void tcp_cwnd_reduction(struct sock *sk, int newly_acked_sacked,
+			       int newly_lost, int flag)
+{
+	struct tcp_sock *tp = tcp_sk(sk);
+	int sndcnt = 0;
+	__u32 pkts_in_flight = tp->packets_out - (tp->sacked_out + tp->lost_out) + tp->retrans_out;
+	int delta = tp->snd_ssthresh - pkts_in_flight;
+
+	if (newly_acked_sacked <= 0 || !tp->prior_cwnd)
+		return;
+
+	__u32 prr_delivered = tp->prr_delivered + newly_acked_sacked;
+
+	if (delta < 0) {
+		__u64 dividend =
+			(__u64)tp->snd_ssthresh * prr_delivered + tp->prior_cwnd - 1;
+		sndcnt = (__u32)div64_u64(dividend, (__u64)tp->prior_cwnd) - tp->prr_out;
+	} else {
+		sndcnt = max(prr_delivered - tp->prr_out, newly_acked_sacked);
+		if (flag & FLAG_SND_UNA_ADVANCED && !newly_lost)
+			sndcnt++;
+		sndcnt = min(delta, sndcnt);
+	}
+	/* Force a fast retransmit upon entering fast recovery */
+	sndcnt = max(sndcnt, (tp->prr_out ? 0 : 1));
+	tp->snd_cwnd = pkts_in_flight + sndcnt;
+}
+
+/* Decide wheather to run the increase function of congestion control. */
+static bool tcp_may_raise_cwnd(const struct sock *sk, const int flag)
+{
+	if (tcp_sk(sk)->reordering > TCP_REORDERING)
+		return flag & FLAG_FORWARD_PROGRESS;
+
+	return flag & FLAG_DATA_ACKED;
+}
+
+void BPF_STRUCT_OPS(bpf_cubic_init, struct sock *sk)
+{
+	cubictcp_init(sk);
+}
+
+void BPF_STRUCT_OPS(bpf_cubic_cwnd_event, struct sock *sk, enum tcp_ca_event event)
+{
+	cubictcp_cwnd_event(sk, event);
+}
+
+void BPF_STRUCT_OPS(bpf_cubic_cong_control, struct sock *sk, __u32 ack, int flag,
+		    const struct rate_sample *rs)
+{
+	struct tcp_sock *tp = tcp_sk(sk);
+
+	if (((1<<TCP_CA_CWR) | (1<<TCP_CA_Recovery)) &
+			(1 << inet_csk(sk)->icsk_ca_state)) {
+		/* Reduce cwnd if state mandates */
+		tcp_cwnd_reduction(sk, rs->acked_sacked, rs->losses, flag);
+
+		if (!before(tp->snd_una, tp->high_seq)) {
+			/* Reset cwnd to ssthresh in CWR or Recovery (unless it's undone) */
+			if (tp->snd_ssthresh < TCP_INFINITE_SSTHRESH &&
+					inet_csk(sk)->icsk_ca_state == TCP_CA_CWR) {
+				tp->snd_cwnd = tp->snd_ssthresh;
+				tp->snd_cwnd_stamp = tcp_jiffies32;
+			}
+		}
+	} else if (tcp_may_raise_cwnd(sk, flag)) {
+		/* Advance cwnd if state allows */
+		cubictcp_cong_avoid(sk, ack, rs->acked_sacked);
+		tp->snd_cwnd_stamp = tcp_jiffies32;
+	}
+
+	tcp_update_pacing_rate(sk);
+}
+
+__u32 BPF_STRUCT_OPS(bpf_cubic_recalc_ssthresh, struct sock *sk)
+{
+	return cubictcp_recalc_ssthresh(sk);
+}
+
+void BPF_STRUCT_OPS(bpf_cubic_state, struct sock *sk, __u8 new_state)
+{
+	cubictcp_state(sk, new_state);
+}
+
+void BPF_STRUCT_OPS(bpf_cubic_acked, struct sock *sk,
+		const struct ack_sample *sample)
+{
+	cubictcp_acked(sk, sample);
+}
+
+__u32 BPF_STRUCT_OPS(bpf_cubic_undo_cwnd, struct sock *sk)
+{
+	return tcp_reno_undo_cwnd(sk);
+}
+
+SEC(".struct_ops")
+struct tcp_congestion_ops cc_cubic = {
+	.init		= (void *)bpf_cubic_init,
+	.ssthresh	= (void *)bpf_cubic_recalc_ssthresh,
+	.cong_control	= (void *)bpf_cubic_cong_control,
+	.set_state	= (void *)bpf_cubic_state,
+	.undo_cwnd	= (void *)bpf_cubic_undo_cwnd,
+	.cwnd_event	= (void *)bpf_cubic_cwnd_event,
+	.pkts_acked     = (void *)bpf_cubic_acked,
+	.name		= "bpf_cc_cubic",
+};
+
+char _license[] SEC("license") = "GPL";

tools/testing/selftests/bpf/progs/bpf_tracing_net.h

@@ -80,6 +80,14 @@
 #define TCP_INFINITE_SSTHRESH	0x7fffffff
 #define TCP_PINGPONG_THRESH	3
 
+#define FLAG_DATA_ACKED 0x04 /* This ACK acknowledged new data.		*/
+#define FLAG_SYN_ACKED 0x10 /* This ACK acknowledged SYN.		*/
+#define FLAG_DATA_SACKED 0x20 /* New SACK.				*/
+#define FLAG_SND_UNA_ADVANCED \
+	0x400 /* Snd_una was changed (!= FLAG_DATA_ACKED) */
+#define FLAG_ACKED (FLAG_DATA_ACKED | FLAG_SYN_ACKED)
+#define FLAG_FORWARD_PROGRESS (FLAG_ACKED | FLAG_DATA_SACKED)
+
 #define fib_nh_dev		nh_common.nhc_dev
 #define fib_nh_gw_family	nh_common.nhc_gw_family
 #define fib_nh_gw6		nh_common.nhc_gw.ipv6
@@ -119,4 +127,6 @@
 #define tw_v6_daddr		__tw_common.skc_v6_daddr
 #define tw_v6_rcv_saddr		__tw_common.skc_v6_rcv_saddr
 
+#define tcp_jiffies32 ((__u32)bpf_jiffies64())
+
 #endif

tools/testing/selftests/bpf/progs/tcp_ca_kfunc.c

@@ -5,7 +5,7 @@
 #include <bpf/bpf_tracing.h>
 
 extern void bbr_init(struct sock *sk) __ksym;
-extern void bbr_main(struct sock *sk, const struct rate_sample *rs) __ksym;
+extern void bbr_main(struct sock *sk, u32 ack, int flag, const struct rate_sample *rs) __ksym;
 extern u32 bbr_sndbuf_expand(struct sock *sk) __ksym;
 extern u32 bbr_undo_cwnd(struct sock *sk) __ksym;
 extern void bbr_cwnd_event(struct sock *sk, enum tcp_ca_event event) __ksym;
@@ -42,9 +42,9 @@ void BPF_PROG(in_ack_event, struct sock *sk, u32 flags)
 }
 
 SEC("struct_ops/cong_control")
-void BPF_PROG(cong_control, struct sock *sk, const struct rate_sample *rs)
+void BPF_PROG(cong_control, struct sock *sk, u32 ack, int flag, const struct rate_sample *rs)
 {
-	bbr_main(sk, rs);
+	bbr_main(sk, ack, flag, rs);
 }
 
 SEC("struct_ops/cong_avoid")