Merge branch 'explicit-tso-segment-count'

Chia-Yu Chang says:

====================
Explicit TSO segment count

This series replaces the existing min_tso_segs() congestion control
callback with a new tso_segs() callback that allows congestion control
algorithms to provide an explicit TSO segment count for each data burst.

To support BPF congestion controls, the series also exposes
tcp_tso_autosize() as a BPF kfunc, allowing BPF implementations to
reuse the kernel TSO autosizing logic while implementing custom
tso_segs() callbacks.
====================

Link: https://patch.msgid.link/20260831165230.2696893-1-chia-yu.chang@nokia-bell-labs.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

5 files changed, 48 insertions, 21 deletions

diff --git a/include/net/tcp.h b/include/net/tcp.h
index 436495ff2271..5e5f5f9b89a3 100644
--- a/include/net/tcp.h
+++ b/include/net/tcp.h
@@ -823,6 +823,9 @@ unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu);
 unsigned int tcp_current_mss(struct sock *sk);
 u32 tcp_clamp_probe0_to_user_timeout(const struct sock *sk, u32 when);
 
+u32 tcp_tso_autosize(const struct sock *sk, unsigned int mss_now,
+		     int min_tso_segs);
+
 /* Bound MSS / TSO packet size with the half of the window */
 static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize)
 {
@@ -1360,8 +1363,16 @@ struct tcp_congestion_ops {
 	/* hook for packet ack accounting (optional) */
 	void (*pkts_acked)(struct sock *sk, const struct ack_sample *sample);
 
-	/* override sysctl_tcp_min_tso_segs (optional) */
-	u32 (*min_tso_segs)(struct sock *sk);
+	/* Override tcp_tso_autosize() (optional)
+	 *
+	 * If provided, this callback supplies the TSO segment target count
+	 * instead of using tcp_tso_autosize(). The returned value is
+	 * subsequently clamped to [1, sk->sk_gso_max_segs] by the caller.
+	 *
+	 * For the kernel callback path, mss_now originates from
+	 * tcp_current_mss() and should never be zero.
+	 */
+	u32 (*tso_segs)(struct sock *sk, u32 mss_now);
 
 	/* new value of cwnd after loss (required) */
 	u32  (*undo_cwnd)(struct sock *sk);
diff --git a/net/ipv4/bpf_tcp_ca.c b/net/ipv4/bpf_tcp_ca.c
index 791e15063237..9deed2244c2d 100644
--- a/net/ipv4/bpf_tcp_ca.c
+++ b/net/ipv4/bpf_tcp_ca.c
@@ -194,6 +194,7 @@ BTF_ID_FLAGS(func, tcp_reno_cong_avoid)
 BTF_ID_FLAGS(func, tcp_reno_undo_cwnd)
 BTF_ID_FLAGS(func, tcp_slow_start)
 BTF_ID_FLAGS(func, tcp_cong_avoid_ai)
+BTF_ID_FLAGS(func, tcp_tso_autosize)
 BTF_KFUNCS_END(bpf_tcp_ca_check_kfunc_ids)
 
 static const struct btf_kfunc_id_set bpf_tcp_ca_kfunc_set = {
@@ -284,7 +285,7 @@ static void bpf_tcp_ca_pkts_acked(struct sock *sk, const struct ack_sample *samp
 {
 }
 
-static u32 bpf_tcp_ca_min_tso_segs(struct sock *sk)
+static u32 bpf_tcp_ca_tso_segs(struct sock *sk, u32 mss_now)
 {
 	return 0;
 }
@@ -320,7 +321,7 @@ static struct tcp_congestion_ops __bpf_ops_tcp_congestion_ops = {
 	.cwnd_event_tx_start = bpf_tcp_ca_cwnd_event_tx_start,
 	.in_ack_event = bpf_tcp_ca_in_ack_event,
 	.pkts_acked = bpf_tcp_ca_pkts_acked,
-	.min_tso_segs = bpf_tcp_ca_min_tso_segs,
+	.tso_segs = bpf_tcp_ca_tso_segs,
 	.cong_control = bpf_tcp_ca_cong_control,
 	.undo_cwnd = bpf_tcp_ca_undo_cwnd,
 	.sndbuf_expand = bpf_tcp_ca_sndbuf_expand,
diff --git a/net/ipv4/tcp_bbr.c b/net/ipv4/tcp_bbr.c
index 82378a2bfd1e..ecf11be46f38 100644
--- a/net/ipv4/tcp_bbr.c
+++ b/net/ipv4/tcp_bbr.c
@@ -297,11 +297,18 @@ static void bbr_set_pacing_rate(struct sock *sk, u32 bw, int gain)
 }
 
 /* override sysctl_tcp_min_tso_segs */
-__bpf_kfunc static u32 bbr_min_tso_segs(struct sock *sk)
+static u32 bbr_min_tso_segs(struct sock *sk)
 {
 	return READ_ONCE(sk->sk_pacing_rate) < (bbr_min_tso_rate >> 3) ? 1 : 2;
 }
 
+__bpf_kfunc static u32 bbr_tso_segs(struct sock *sk, u32 mss_now)
+{
+	if (unlikely(!mss_now))
+		return bbr_min_tso_segs(sk);
+	return tcp_tso_autosize(sk, mss_now, bbr_min_tso_segs(sk));
+}
+
 static u32 bbr_tso_segs_goal(struct sock *sk)
 {
 	struct tcp_sock *tp = tcp_sk(sk);
@@ -1151,7 +1158,7 @@ static struct tcp_congestion_ops tcp_bbr_cong_ops __read_mostly = {
 	.undo_cwnd	= bbr_undo_cwnd,
 	.cwnd_event_tx_start	= bbr_cwnd_event_tx_start,
 	.ssthresh	= bbr_ssthresh,
-	.min_tso_segs	= bbr_min_tso_segs,
+	.tso_segs	= bbr_tso_segs,
 	.get_info	= bbr_get_info,
 	.set_state	= bbr_set_state,
 };
@@ -1163,7 +1170,7 @@ BTF_ID_FLAGS(func, bbr_sndbuf_expand)
 BTF_ID_FLAGS(func, bbr_undo_cwnd)
 BTF_ID_FLAGS(func, bbr_cwnd_event_tx_start)
 BTF_ID_FLAGS(func, bbr_ssthresh)
-BTF_ID_FLAGS(func, bbr_min_tso_segs)
+BTF_ID_FLAGS(func, bbr_tso_segs)
 BTF_ID_FLAGS(func, bbr_set_state)
 BTF_KFUNCS_END(tcp_bbr_check_kfunc_ids)
 
diff --git a/net/ipv4/tcp_output.c b/net/ipv4/tcp_output.c
index d960e3de7d50..00417a429222 100644
--- a/net/ipv4/tcp_output.c
+++ b/net/ipv4/tcp_output.c
@@ -2252,13 +2252,21 @@ static bool tcp_nagle_check(bool partial, const struct tcp_sock *tp,
  * in bigger TSO bursts. We we cut the RTT-based allowance in half
  * for every 2^9 usec (aka 512 us) of RTT, so that the RTT-based allowance
  * is below 1500 bytes after 6 * ~500 usec = 3ms.
+ *
+ * The min_tso_segs is floored to 1 to avoid surprising conversion. Also,
+ * BPF callers may pass mss_now == 0. In that case the function returns the
+ * sanitized min_tso_segs value and skips autosizing.
  */
-static u32 tcp_tso_autosize(const struct sock *sk, unsigned int mss_now,
-			    int min_tso_segs)
+__bpf_kfunc u32 tcp_tso_autosize(const struct sock *sk, unsigned int mss_now,
+				 int min_tso_segs)
 {
+	u32 min_tso = max(min_tso_segs, 1);
 	unsigned long bytes;
 	u32 r;
 
+	if (unlikely(!mss_now))
+		return min_tso;
+
 	bytes = READ_ONCE(sk->sk_pacing_rate) >> READ_ONCE(sk->sk_pacing_shift);
 
 	r = tcp_min_rtt(tcp_sk(sk)) >> READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_tso_rtt_log);
@@ -2267,8 +2275,9 @@ static u32 tcp_tso_autosize(const struct sock *sk, unsigned int mss_now,
 
 	bytes = min_t(unsigned long, bytes, sk->sk_gso_max_size);
 
-	return max_t(u32, bytes / mss_now, min_tso_segs);
+	return max_t(u32, bytes / mss_now, min_tso);
 }
+EXPORT_SYMBOL_GPL(tcp_tso_autosize);
 
 /* Return the number of segments we want in the skb we are transmitting.
  * See if congestion control module wants to decide; otherwise, autosize.
@@ -2276,14 +2285,13 @@ static u32 tcp_tso_autosize(const struct sock *sk, unsigned int mss_now,
 static u32 tcp_tso_segs(struct sock *sk, unsigned int mss_now)
 {
 	const struct tcp_congestion_ops *ca_ops = inet_csk(sk)->icsk_ca_ops;
-	u32 min_tso, tso_segs;
-
-	min_tso = ca_ops->min_tso_segs ?
-			ca_ops->min_tso_segs(sk) :
-			READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_min_tso_segs);
+	u32 tso_segs;
 
-	tso_segs = tcp_tso_autosize(sk, mss_now, min_tso);
-	return min_t(u32, tso_segs, sk->sk_gso_max_segs);
+	tso_segs = ca_ops->tso_segs ?
+			ca_ops->tso_segs(sk, mss_now) :
+			tcp_tso_autosize(sk, mss_now,
+					 READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_min_tso_segs));
+	return clamp_t(u32, tso_segs, 1, sk->sk_gso_max_segs);
 }
 
 /* Returns the portion of skb which can be sent right away */
diff --git a/tools/testing/selftests/bpf/progs/tcp_ca_kfunc.c b/tools/testing/selftests/bpf/progs/tcp_ca_kfunc.c
index 0a3e9d35bf6f..58262e490336 100644
--- a/tools/testing/selftests/bpf/progs/tcp_ca_kfunc.c
+++ b/tools/testing/selftests/bpf/progs/tcp_ca_kfunc.c
@@ -10,7 +10,7 @@ extern u32 bbr_sndbuf_expand(struct sock *sk) __ksym;
 extern u32 bbr_undo_cwnd(struct sock *sk) __ksym;
 extern void bbr_cwnd_event_tx_start(struct sock *sk) __ksym;
 extern u32 bbr_ssthresh(struct sock *sk) __ksym;
-extern u32 bbr_min_tso_segs(struct sock *sk) __ksym;
+extern u32 bbr_tso_segs(struct sock *sk, u32 mss_now) __ksym;
 extern void bbr_set_state(struct sock *sk, u8 new_state) __ksym;
 
 extern void dctcp_init(struct sock *sk) __ksym;
@@ -90,9 +90,9 @@ u32 BPF_PROG(ssthresh, struct sock *sk)
 }
 
 SEC("struct_ops")
-u32 BPF_PROG(min_tso_segs, struct sock *sk)
+u32 BPF_PROG(tso_segs, struct sock *sk, u32 mss_now)
 {
-	return bbr_min_tso_segs(sk);
+	return bbr_tso_segs(sk, mss_now);
 }
 
 SEC("struct_ops")
@@ -120,7 +120,7 @@ struct tcp_congestion_ops tcp_ca_kfunc = {
 	.cwnd_event	= (void *)cwnd_event,
 	.cwnd_event_tx_start = (void *)cwnd_event_tx_start,
 	.ssthresh	= (void *)ssthresh,
-	.min_tso_segs	= (void *)min_tso_segs,
+	.tso_segs	= (void *)tso_segs,
 	.set_state	= (void *)set_state,
 	.pkts_acked     = (void *)pkts_acked,
 	.name		= "tcp_ca_kfunc",