[PATCH net 0/7] tcp: preserve advertised rwnd accounting across receive-memory decisions

[PATCH net 0/7] tcp: preserve advertised rwnd accounting across receive-memory decisions

[Wesley Atwell]

This series keeps sender-visible TCP receive-window accounting tied to the
scaling basis that was in force when the window was advertised.

Problem
-------

`tp->rcv_wnd` is an advertised promise to the sender, but later
receive-memory admission and clamping could reconstruct that promise
through the mutable live `scaling_ratio`. After ratio drift, the stack
could retain or advertise a receive window that no longer matched the
local hard rmem budget.

Fix
---

- store the advertise-time scaling basis alongside `tp->rcv_wnd`
- refresh that pair at the TCP and MPTCP receive-window write sites
- consume the snapshot in receive-memory admission, clamping, and the
  scaled-window quantization path
- preserve the snapshot across `TCP_REPAIR_WINDOW` restore when userspace
  provides it, and fall back safely when legacy userspace cannot
- expose the accounting in tracepoints and cover the ABI/runtime contract
  in selftests

Series layout
-------------

1. track the receive-window snapshot state and helpers
2. refresh the snapshot when TCP advertises or initializes windows
3. use the snapshot in receive-memory admission and clamping
4. extend `TCP_REPAIR_WINDOW` for exact restore plus legacy compatibility
5. refresh the TCP shadow window snapshot in MPTCP
6. expose rmem/backlog state in `rcvbuf_grow` tracepoints
7. cover legacy and extended repair-window layouts in selftests

Testing
-------

- `git diff --check origin/main..HEAD`
- `scripts/checkpatch.pl --strict --show-types` on patches 1-7
- `make -j8 headers`
- `make -j8 net/ipv4/tcp_input.o net/ipv4/tcp_output.o net/ipv4/tcp_minisocks.o net/ipv4/tcp.o`
- `make -j8 C=1 CF='-D__CHECK_ENDIAN__' W=1 net/ipv4/tcp_input.o net/ipv4/tcp_output.o net/ipv4/tcp_minisocks.o net/ipv4/tcp.o`
- `make SPHINXDIRS='networking/net_cachelines' htmldocs`
- `make -j8 vmlinux bzImage modules`
- `make -C tools/testing/selftests/net/tcp_ao -j8`
- `make -C tools/testing/selftests/net/mptcp -j8`
- `packetdrill --dry_run` for `tcp_rcv_toobig.pkt` and
  `tcp_rcv_toobig_default.pkt`
- `virtme-run` guest pass for both packetdrill tests
- feature-enabled guest pass for `restore_ipv4`, `self-connect_ipv4`, and
  `mptcp_sockopt.sh`

Thanks,
Wesley

---
base-commit: 908c344d5cfa0ee6efb3226d22ea661e078ebfa0
-- 
2.43.0

[PATCH net 1/7] tcp: track advertise-time scaling basis for rcv_wnd

[Wesley Atwell]

tp->rcv_wnd is an advertised window, but later receive-side accounting
needs to recover the hard memory budget that window represented when it
was exposed.

Prepare for that by storing the scaling basis alongside tp->rcv_wnd and
centralizing the helper API around the paired state. While here, make the
existing receive-memory arithmetic use the shared helper names so later
behavioral changes can build on one explicit accounting model.

This patch is groundwork only. Later patches will refresh the snapshot at
window write sites and consume it in the receive-memory paths.

Signed-off-by: Wesley Atwell <atwellwea@gmail.com>
---
 .../networking/net_cachelines/tcp_sock.rst    |  1 +
 include/linux/tcp.h                           |  1 +
 include/net/tcp.h                             | 79 +++++++++++++++++--
 net/ipv4/tcp.c                                |  1 +
 4 files changed, 76 insertions(+), 6 deletions(-)

diff --git a/Documentation/networking/net_cachelines/tcp_sock.rst b/Documentation/networking/net_cachelines/tcp_sock.rst
index 563daea10d6c..1415981b9d8a 100644
--- a/Documentation/networking/net_cachelines/tcp_sock.rst
+++ b/Documentation/networking/net_cachelines/tcp_sock.rst
@@ -12,6 +12,7 @@ struct inet_connection_sock   inet_conn
 u16                           tcp_header_len          read_mostly         read_mostly         tcp_bound_to_half_wnd,tcp_current_mss(tx);tcp_rcv_established(rx)
 u16                           gso_segs                read_mostly                             tcp_xmit_size_goal
 __be32                        pred_flags              read_write          read_mostly         tcp_select_window(tx);tcp_rcv_established(rx)
+u8                            rcv_wnd_scaling_ratio   read_write          read_mostly         tcp_set_rcv_wnd,tcp_can_ingest,tcp_clamp_window
 u64                           bytes_received                              read_write          tcp_rcv_nxt_update(rx)
 u32                           segs_in                                     read_write          tcp_v6_rcv(rx)
 u32                           data_segs_in                                read_write          tcp_v6_rcv(rx)
diff --git a/include/linux/tcp.h b/include/linux/tcp.h
index f72eef31fa23..ec6b70c1174b 100644
--- a/include/linux/tcp.h
+++ b/include/linux/tcp.h
@@ -297,6 +297,7 @@ struct tcp_sock {
 		est_ecnfield:2,/* ECN field for AccECN delivered estimates */
 		accecn_opt_demand:2,/* Demand AccECN option for n next ACKs */
 		prev_ecnfield:2; /* ECN bits from the previous segment */
+	u8	rcv_wnd_scaling_ratio; /* 0 if unknown, else tp->rcv_wnd basis */
 	__be32	pred_flags;
 	u64	tcp_clock_cache; /* cache last tcp_clock_ns() (see tcp_mstamp_refresh()) */
 	u64	tcp_mstamp;	/* most recent packet received/sent */
diff --git a/include/net/tcp.h b/include/net/tcp.h
index 978eea2d5df0..187e6d660f62 100644
--- a/include/net/tcp.h
+++ b/include/net/tcp.h
@@ -1702,6 +1702,26 @@ static inline int tcp_space_from_win(const struct sock *sk, int win)
 	return __tcp_space_from_win(tcp_sk(sk)->scaling_ratio, win);
 }
 
+static inline bool tcp_rcv_wnd_snapshot_valid(const struct tcp_sock *tp)
+{
+	return tp->rcv_wnd_scaling_ratio != 0;
+}
+
+/* Rebuild hard receive-memory units for data already covered by tp->rcv_wnd if
+ * the advertise-time basis is known. Legacy TCP_REPAIR restores can only
+ * recover tp->rcv_wnd itself; callers must fall back when the snapshot is
+ * unknown.
+ */
+static inline bool tcp_space_from_rcv_wnd(const struct tcp_sock *tp, int win,
+					  int *space)
+{
+	if (!tcp_rcv_wnd_snapshot_valid(tp))
+		return false;
+
+	*space = __tcp_space_from_win(tp->rcv_wnd_scaling_ratio, win);
+	return true;
+}
+
 /* Assume a 50% default for skb->len/skb->truesize ratio.
  * This may be adjusted later in tcp_measure_rcv_mss().
  */
@@ -1709,15 +1729,62 @@ static inline int tcp_space_from_win(const struct sock *sk, int win)
 
 static inline void tcp_scaling_ratio_init(struct sock *sk)
 {
-	tcp_sk(sk)->scaling_ratio = TCP_DEFAULT_SCALING_RATIO;
+	struct tcp_sock *tp = tcp_sk(sk);
+
+	tp->scaling_ratio = TCP_DEFAULT_SCALING_RATIO;
+	tp->rcv_wnd_scaling_ratio = TCP_DEFAULT_SCALING_RATIO;
+}
+
+/* tp->rcv_wnd is paired with the scaling_ratio that was in force when that
+ * window was last advertised. Legacy TCP_REPAIR restores can only recover the
+ * window value itself and use a zero snapshot until a fresh local window
+ * advertisement refreshes the pair.
+ */
+static inline void tcp_set_rcv_wnd_snapshot(struct tcp_sock *tp, u32 win,
+					    u8 scaling_ratio)
+{
+	tp->rcv_wnd = win;
+	tp->rcv_wnd_scaling_ratio = scaling_ratio;
+}
+
+static inline void tcp_set_rcv_wnd(struct tcp_sock *tp, u32 win)
+{
+	tcp_set_rcv_wnd_snapshot(tp, win, tp->scaling_ratio);
+}
+
+static inline void tcp_set_rcv_wnd_unknown(struct tcp_sock *tp, u32 win)
+{
+	tcp_set_rcv_wnd_snapshot(tp, win, 0);
+}
+
+/* TCP receive-side accounting reuses sk_rcvbuf as both a hard memory limit
+ * and as the source material for the advertised receive window after
+ * scaling_ratio conversion. Keep the byte accounting explicit so admission,
+ * pruning, and rwnd selection all start from the same quantities.
+ */
+static inline int tcp_rmem_used(const struct sock *sk)
+{
+	return atomic_read(&sk->sk_rmem_alloc);
+}
+
+static inline int tcp_rmem_avail(const struct sock *sk)
+{
+	return READ_ONCE(sk->sk_rcvbuf) - tcp_rmem_used(sk);
+}
+
+/* Sender-visible rwnd headroom also reserves bytes already queued on backlog.
+ * Those bytes are not free to advertise again until __release_sock() drains
+ * backlog and clears sk_backlog.len.
+ */
+static inline int tcp_rwnd_avail(const struct sock *sk)
+{
+	return tcp_rmem_avail(sk) - READ_ONCE(sk->sk_backlog.len);
 }
 
 /* Note: caller must be prepared to deal with negative returns */
 static inline int tcp_space(const struct sock *sk)
 {
-	return tcp_win_from_space(sk, READ_ONCE(sk->sk_rcvbuf) -
-				  READ_ONCE(sk->sk_backlog.len) -
-				  atomic_read(&sk->sk_rmem_alloc));
+	return tcp_win_from_space(sk, tcp_rwnd_avail(sk));
 }
 
 static inline int tcp_full_space(const struct sock *sk)
@@ -1760,7 +1827,7 @@ static inline bool tcp_rmem_pressure(const struct sock *sk)
 	rcvbuf = READ_ONCE(sk->sk_rcvbuf);
 	threshold = rcvbuf - (rcvbuf >> 3);
 
-	return atomic_read(&sk->sk_rmem_alloc) > threshold;
+	return tcp_rmem_used(sk) > threshold;
 }
 
 static inline bool tcp_epollin_ready(const struct sock *sk, int target)
@@ -1910,7 +1977,7 @@ static inline void tcp_fast_path_check(struct sock *sk)
 
 	if (RB_EMPTY_ROOT(&tp->out_of_order_queue) &&
 	    tp->rcv_wnd &&
-	    atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf &&
+	    tcp_rmem_avail(sk) > 0 &&
 	    !tp->urg_data)
 		tcp_fast_path_on(tp);
 }
diff --git a/net/ipv4/tcp.c b/net/ipv4/tcp.c
index 202a4e57a218..cec9ae1bf875 100644
--- a/net/ipv4/tcp.c
+++ b/net/ipv4/tcp.c
@@ -5238,6 +5238,7 @@ static void __init tcp_struct_check(void)
 	CACHELINE_ASSERT_GROUP_MEMBER(struct tcp_sock, tcp_sock_write_txrx, received_ce);
 	CACHELINE_ASSERT_GROUP_MEMBER(struct tcp_sock, tcp_sock_write_txrx, received_ecn_bytes);
 	CACHELINE_ASSERT_GROUP_MEMBER(struct tcp_sock, tcp_sock_write_txrx, app_limited);
+	CACHELINE_ASSERT_GROUP_MEMBER(struct tcp_sock, tcp_sock_write_txrx, rcv_wnd_scaling_ratio);
 	CACHELINE_ASSERT_GROUP_MEMBER(struct tcp_sock, tcp_sock_write_txrx, rcv_wnd);
 	CACHELINE_ASSERT_GROUP_MEMBER(struct tcp_sock, tcp_sock_write_txrx, rcv_tstamp);
 	CACHELINE_ASSERT_GROUP_MEMBER(struct tcp_sock, tcp_sock_write_txrx, rx_opt);
-- 
2.34.1

[PATCH net 2/7] tcp: preserve rcv_wnd snapshot when updating advertised windows

[Wesley Atwell]

Once tp->rcv_wnd carries paired snapshot semantics, every write of the
advertised window has to refresh the snapshot at the same time.

Convert the active-open, passive-open, and normal advertised-window
update sites to use tcp_set_rcv_wnd(). This keeps new sockets and later
window advertisements initialized with a valid advertise-time basis
before the receive-memory logic starts consuming it.

Signed-off-by: Wesley Atwell <atwellwea@gmail.com>
---
 net/ipv4/tcp_minisocks.c | 2 +-
 net/ipv4/tcp_output.c    | 8 ++++++--
 2 files changed, 7 insertions(+), 3 deletions(-)

diff --git a/net/ipv4/tcp_minisocks.c b/net/ipv4/tcp_minisocks.c
index dafb63b923d0..ae8a466b5298 100644
--- a/net/ipv4/tcp_minisocks.c
+++ b/net/ipv4/tcp_minisocks.c
@@ -603,7 +603,7 @@ struct sock *tcp_create_openreq_child(const struct sock *sk,
 	newtp->rx_opt.sack_ok = ireq->sack_ok;
 	newtp->window_clamp = req->rsk_window_clamp;
 	newtp->rcv_ssthresh = req->rsk_rcv_wnd;
-	newtp->rcv_wnd = req->rsk_rcv_wnd;
+	tcp_set_rcv_wnd(newtp, req->rsk_rcv_wnd);
 	newtp->rx_opt.wscale_ok = ireq->wscale_ok;
 	if (newtp->rx_opt.wscale_ok) {
 		newtp->rx_opt.snd_wscale = ireq->snd_wscale;
diff --git a/net/ipv4/tcp_output.c b/net/ipv4/tcp_output.c
index 326b58ff1118..c1b94d67d8fe 100644
--- a/net/ipv4/tcp_output.c
+++ b/net/ipv4/tcp_output.c
@@ -291,7 +291,7 @@ static u16 tcp_select_window(struct sock *sk)
 	 */
 	if (unlikely(inet_csk(sk)->icsk_ack.pending & ICSK_ACK_NOMEM)) {
 		tp->pred_flags = 0;
-		tp->rcv_wnd = 0;
+		tcp_set_rcv_wnd(tp, 0);
 		tp->rcv_wup = tp->rcv_nxt;
 		return 0;
 	}
@@ -314,7 +314,7 @@ static u16 tcp_select_window(struct sock *sk)
 		}
 	}
 
-	tp->rcv_wnd = new_win;
+	tcp_set_rcv_wnd(tp, new_win);
 	tp->rcv_wup = tp->rcv_nxt;
 
 	/* Make sure we do not exceed the maximum possible
@@ -4150,6 +4150,10 @@ static void tcp_connect_init(struct sock *sk)
 				  READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_window_scaling),
 				  &rcv_wscale,
 				  rcv_wnd);
+	/* tcp_select_initial_window() filled tp->rcv_wnd through its out-param,
+	 * so snapshot the scaling_ratio we will use for that initial rwnd.
+	 */
+	tcp_set_rcv_wnd(tp, tp->rcv_wnd);
 
 	tp->rx_opt.rcv_wscale = rcv_wscale;
 	tp->rcv_ssthresh = tp->rcv_wnd;
-- 
2.34.1

[PATCH net 3/7] tcp: honor advertised receive window in memory admission and clamping

[Wesley Atwell]

tp->rcv_wnd is an advertised promise to the sender, but receive-memory
accounting was still reconstructing that promise through mutable live
state.

Switch the receive-side decisions over to the advertise-time snapshot.
Use it when deciding whether a packet can be admitted, when deciding how
far to clamp future window growth, and when handling the scaled-window
quantization slack in __tcp_select_window(). If a snapshot is not
available, keep the legacy fallback behavior.

This keeps sender-visible rwnd and the local hard rmem budget in the
same unit system instead of letting ratio drift create accounting
mismatches.

Signed-off-by: Wesley Atwell <atwellwea@gmail.com>
---
 include/net/tcp.h     |  1 +
 net/ipv4/tcp_input.c  | 86 ++++++++++++++++++++++++++++++++++++++++---
 net/ipv4/tcp_output.c | 14 ++++++-
 3 files changed, 93 insertions(+), 8 deletions(-)

diff --git a/include/net/tcp.h b/include/net/tcp.h
index 187e6d660f62..88ddf7ee826e 100644
--- a/include/net/tcp.h
+++ b/include/net/tcp.h
@@ -384,6 +384,7 @@ int tcp_ioctl(struct sock *sk, int cmd, int *karg);
 enum skb_drop_reason tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb);
 void tcp_rcv_established(struct sock *sk, struct sk_buff *skb);
 void tcp_rcvbuf_grow(struct sock *sk, u32 newval);
+bool tcp_try_grow_rcvbuf(struct sock *sk, int needed);
 void tcp_rcv_space_adjust(struct sock *sk);
 int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp);
 void tcp_twsk_destructor(struct sock *sk);
diff --git a/net/ipv4/tcp_input.c b/net/ipv4/tcp_input.c
index cba89733d121..f76011fc1b7a 100644
--- a/net/ipv4/tcp_input.c
+++ b/net/ipv4/tcp_input.c
@@ -774,8 +774,37 @@ static void tcp_init_buffer_space(struct sock *sk)
 				    (u32)TCP_INIT_CWND * tp->advmss);
 }
 
+/* Try to grow sk_rcvbuf so the hard receive-memory limit covers @needed
+ * bytes beyond the memory already charged in sk_rmem_alloc.
+ */
+bool tcp_try_grow_rcvbuf(struct sock *sk, int needed)
+{
+	struct net *net = sock_net(sk);
+	int target;
+	int rmem2;
+
+	needed = max(needed, 0);
+	target = tcp_rmem_used(sk) + needed;
+
+	if (target <= READ_ONCE(sk->sk_rcvbuf))
+		return true;
+
+	rmem2 = READ_ONCE(net->ipv4.sysctl_tcp_rmem[2]);
+	if (READ_ONCE(sk->sk_rcvbuf) >= rmem2 ||
+	    (sk->sk_userlocks & SOCK_RCVBUF_LOCK) ||
+	    tcp_under_memory_pressure(sk) ||
+	    sk_memory_allocated(sk) >= sk_prot_mem_limits(sk, 0))
+		return false;
+
+	WRITE_ONCE(sk->sk_rcvbuf,
+		   min_t(int, rmem2,
+			 max_t(int, READ_ONCE(sk->sk_rcvbuf), target)));
+
+	return target <= READ_ONCE(sk->sk_rcvbuf);
+}
+
 /* 4. Recalculate window clamp after socket hit its memory bounds. */
-static void tcp_clamp_window(struct sock *sk)
+static void tcp_clamp_window_legacy(struct sock *sk)
 {
 	struct tcp_sock *tp = tcp_sk(sk);
 	struct inet_connection_sock *icsk = inet_csk(sk);
@@ -785,14 +814,42 @@ static void tcp_clamp_window(struct sock *sk)
 	icsk->icsk_ack.quick = 0;
 	rmem2 = READ_ONCE(net->ipv4.sysctl_tcp_rmem[2]);
 
-	if (sk->sk_rcvbuf < rmem2 &&
+	if (READ_ONCE(sk->sk_rcvbuf) < rmem2 &&
 	    !(sk->sk_userlocks & SOCK_RCVBUF_LOCK) &&
 	    !tcp_under_memory_pressure(sk) &&
 	    sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0)) {
 		WRITE_ONCE(sk->sk_rcvbuf,
 			   min(atomic_read(&sk->sk_rmem_alloc), rmem2));
 	}
-	if (atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf)
+	if (atomic_read(&sk->sk_rmem_alloc) > READ_ONCE(sk->sk_rcvbuf))
+		tp->rcv_ssthresh = min(tp->window_clamp, 2U * tp->advmss);
+}
+
+static void tcp_clamp_window(struct sock *sk)
+{
+	struct tcp_sock *tp = tcp_sk(sk);
+	u32 cur_rwnd = tcp_receive_window(tp);
+	int need;
+
+	if (!tcp_space_from_rcv_wnd(tp, cur_rwnd, &need)) {
+		tcp_clamp_window_legacy(sk);
+		return;
+	}
+
+	inet_csk(sk)->icsk_ack.quick = 0;
+	need = max_t(int, need, 0);
+
+	/* Keep the hard receive-memory cap large enough to honor the
+	 * remaining receive window we already exposed to the sender. Use
+	 * the scaling_ratio snapshot taken when tp->rcv_wnd was advertised,
+	 * not the mutable live ratio which may drift later in the flow.
+	 */
+	tcp_try_grow_rcvbuf(sk, need);
+
+	/* If the remaining advertised rwnd no longer fits the hard budget,
+	 * slow future window growth until the accounting converges again.
+	 */
+	if (need > tcp_rmem_avail(sk))
 		tp->rcv_ssthresh = min(tp->window_clamp, 2U * tp->advmss);
 }
 
@@ -5374,11 +5431,28 @@ static void tcp_ofo_queue(struct sock *sk)
 static bool tcp_prune_ofo_queue(struct sock *sk, const struct sk_buff *in_skb);
 static int tcp_prune_queue(struct sock *sk, const struct sk_buff *in_skb);
 
+/* Sequence checks run against the sender-visible receive window before this
+ * point. Convert the incoming payload back to the hard receive-memory budget
+ * using the scaling_ratio that was in force when tp->rcv_wnd was advertised,
+ * so admission keeps honoring the same exposed window even if the live ratio
+ * changes later in the flow. Legacy TCP_REPAIR restores do not have that
+ * advertise-time basis, so they fall back to the pre-series admission rule
+ * until a fresh local advertisement refreshes the pair.
+ *
+ * Do not subtract sk_backlog.len here. tcp_space() already reserves backlog
+ * bytes when selecting future advertised windows, and sk_backlog.len stays
+ * inflated until __release_sock() finishes draining backlog. Subtracting it
+ * again here would double count already-queued backlog packets as they move
+ * into sk_rmem_alloc.
+ */
 static bool tcp_can_ingest(const struct sock *sk, const struct sk_buff *skb)
 {
-	unsigned int rmem = atomic_read(&sk->sk_rmem_alloc);
+	int need;
+
+	if (!tcp_space_from_rcv_wnd(tcp_sk(sk), skb->len, &need))
+		return atomic_read(&sk->sk_rmem_alloc) <= READ_ONCE(sk->sk_rcvbuf);
 
-	return rmem <= sk->sk_rcvbuf;
+	return need <= tcp_rmem_avail(sk);
 }
 
 static int tcp_try_rmem_schedule(struct sock *sk, const struct sk_buff *skb,
@@ -6014,7 +6088,7 @@ static int tcp_prune_queue(struct sock *sk, const struct sk_buff *in_skb)
 	struct tcp_sock *tp = tcp_sk(sk);
 
 	/* Do nothing if our queues are empty. */
-	if (!atomic_read(&sk->sk_rmem_alloc))
+	if (!tcp_rmem_used(sk))
 		return -1;
 
 	NET_INC_STATS(sock_net(sk), LINUX_MIB_PRUNECALLED);
diff --git a/net/ipv4/tcp_output.c b/net/ipv4/tcp_output.c
index c1b94d67d8fe..5e69fc31a4da 100644
--- a/net/ipv4/tcp_output.c
+++ b/net/ipv4/tcp_output.c
@@ -3377,13 +3377,23 @@ u32 __tcp_select_window(struct sock *sk)
 	 * scaled window will not line up with the MSS boundary anyway.
 	 */
 	if (tp->rx_opt.rcv_wscale) {
+		int rcv_wscale = 1 << tp->rx_opt.rcv_wscale;
+
 		window = free_space;
 
 		/* Advertise enough space so that it won't get scaled away.
-		 * Import case: prevent zero window announcement if
+		 * Important case: prevent zero-window announcement if
 		 * 1<<rcv_wscale > mss.
 		 */
-		window = ALIGN(window, (1 << tp->rx_opt.rcv_wscale));
+		window = ALIGN(window, rcv_wscale);
+
+		/* Back any scale-quantization slack before we expose it.
+		 * Otherwise tcp_can_ingest() can reject data which is still
+		 * within the sender-visible window.
+		 */
+		if (window > free_space &&
+		    !tcp_try_grow_rcvbuf(sk, tcp_space_from_win(sk, window)))
+			window = round_down(free_space, rcv_wscale);
 	} else {
 		window = tp->rcv_wnd;
 		/* Get the largest window that is a nice multiple of mss.
-- 
2.34.1

[PATCH net 4/7] tcp: extend TCP_REPAIR_WINDOW with receive-window scaling snapshot

[Wesley Atwell]

The paired receive-window state is now part of the live TCP socket
semantics, so repair and restore need a way to preserve it.

Extend TCP_REPAIR_WINDOW with the advertise-time scaling snapshot while
keeping old userspace working. The kernel now accepts exactly the legacy
layout and the extended layout. Legacy restore leaves the snapshot
unknown so the socket falls back safely until a fresh local window
advertisement refreshes the pair, while the extended layout restores the
exact snapshot.

Signed-off-by: Wesley Atwell <atwellwea@gmail.com>
---
 include/uapi/linux/tcp.h |  1 +
 net/ipv4/tcp.c           | 34 ++++++++++++++++++++++++++++------
 2 files changed, 29 insertions(+), 6 deletions(-)

diff --git a/include/uapi/linux/tcp.h b/include/uapi/linux/tcp.h
index 03772dd4d399..3a799f4c0e1e 100644
--- a/include/uapi/linux/tcp.h
+++ b/include/uapi/linux/tcp.h
@@ -159,6 +159,7 @@ struct tcp_repair_window {
 
 	__u32	rcv_wnd;
 	__u32	rcv_wup;
+	__u32	rcv_wnd_scaling_ratio; /* 0 means advertise-time basis unknown */
 };
 
 enum {
diff --git a/net/ipv4/tcp.c b/net/ipv4/tcp.c
index cec9ae1bf875..dd2b4fe61bd8 100644
--- a/net/ipv4/tcp.c
+++ b/net/ipv4/tcp.c
@@ -3551,17 +3551,25 @@ static inline bool tcp_can_repair_sock(const struct sock *sk)
 		(sk->sk_state != TCP_LISTEN);
 }
 
+/* Keep accepting the pre-extension TCP_REPAIR_WINDOW layout so legacy
+ * userspace can restore sockets without fabricating a snapshot basis.
+ */
+static inline int tcp_repair_window_legacy_size(void)
+{
+	return offsetof(struct tcp_repair_window, rcv_wnd_scaling_ratio);
+}
+
 static int tcp_repair_set_window(struct tcp_sock *tp, sockptr_t optbuf, int len)
 {
-	struct tcp_repair_window opt;
+	struct tcp_repair_window opt = {};
 
 	if (!tp->repair)
 		return -EPERM;
 
-	if (len != sizeof(opt))
+	if (len != tcp_repair_window_legacy_size() && len != sizeof(opt))
 		return -EINVAL;
 
-	if (copy_from_sockptr(&opt, optbuf, sizeof(opt)))
+	if (copy_from_sockptr(&opt, optbuf, len))
 		return -EFAULT;
 
 	if (opt.max_window < opt.snd_wnd)
@@ -3577,7 +3585,20 @@ static int tcp_repair_set_window(struct tcp_sock *tp, sockptr_t optbuf, int len)
 	tp->snd_wnd	= opt.snd_wnd;
 	tp->max_window	= opt.max_window;
 
-	tp->rcv_wnd	= opt.rcv_wnd;
+	if (len == tcp_repair_window_legacy_size()) {
+		/* Legacy repair UAPI has no advertise-time basis for tp->rcv_wnd.
+		 * Mark the snapshot unknown until a fresh local advertisement
+		 * re-establishes the pair.
+		 */
+		tcp_set_rcv_wnd_unknown(tp, opt.rcv_wnd);
+		tp->rcv_wup	= opt.rcv_wup;
+		return 0;
+	}
+
+	if (opt.rcv_wnd_scaling_ratio > U8_MAX)
+		return -EINVAL;
+
+	tcp_set_rcv_wnd_snapshot(tp, opt.rcv_wnd, opt.rcv_wnd_scaling_ratio);
 	tp->rcv_wup	= opt.rcv_wup;
 
 	return 0;
@@ -4667,12 +4688,12 @@ int do_tcp_getsockopt(struct sock *sk, int level,
 		break;
 
 	case TCP_REPAIR_WINDOW: {
-		struct tcp_repair_window opt;
+		struct tcp_repair_window opt = {};
 
 		if (copy_from_sockptr(&len, optlen, sizeof(int)))
 			return -EFAULT;
 
-		if (len != sizeof(opt))
+		if (len != tcp_repair_window_legacy_size() && len != sizeof(opt))
 			return -EINVAL;
 
 		if (!tp->repair)
@@ -4683,6 +4704,7 @@ int do_tcp_getsockopt(struct sock *sk, int level,
 		opt.max_window	= tp->max_window;
 		opt.rcv_wnd	= tp->rcv_wnd;
 		opt.rcv_wup	= tp->rcv_wup;
+		opt.rcv_wnd_scaling_ratio = tp->rcv_wnd_scaling_ratio;
 
 		if (copy_to_sockptr(optval, &opt, len))
 			return -EFAULT;
-- 
2.34.1

[PATCH net 5/7] mptcp: refresh tcp rcv_wnd snapshot when syncing receive windows

[Wesley Atwell]

MPTCP rewrites the TCP shadow receive window on subflows when shared
receive-window state changes.

Once tp->rcv_wnd carries paired snapshot semantics, those subflow shadow
updates have to refresh the snapshot too. Convert the MPTCP window-sync
write sites to use the helper and keep the aggregate receive-space
arithmetic using the explicit rwnd-availability helper.

Signed-off-by: Wesley Atwell <atwellwea@gmail.com>
---
 net/mptcp/options.c  | 12 ++++++++----
 net/mptcp/protocol.h | 14 +++++++++++---
 2 files changed, 19 insertions(+), 7 deletions(-)

diff --git a/net/mptcp/options.c b/net/mptcp/options.c
index 43df4293f58b..6e6aa084cbfa 100644
--- a/net/mptcp/options.c
+++ b/net/mptcp/options.c
@@ -1073,9 +1073,12 @@ static void rwin_update(struct mptcp_sock *msk, struct sock *ssk,
 		return;
 
 	/* Some other subflow grew the mptcp-level rwin since rcv_wup,
-	 * resync.
+	 * resync. Keep the TCP shadow window in its advertised u32 domain
+	 * and refresh the advertise-time scaling snapshot while doing so.
 	 */
-	tp->rcv_wnd += mptcp_rcv_wnd - subflow->rcv_wnd_sent;
+	tcp_set_rcv_wnd(tp, min_t(u64, (u64)tp->rcv_wnd +
+				  (mptcp_rcv_wnd - subflow->rcv_wnd_sent),
+				  U32_MAX));
 	subflow->rcv_wnd_sent = mptcp_rcv_wnd;
 }
 
@@ -1334,11 +1337,12 @@ static void mptcp_set_rwin(struct tcp_sock *tp, struct tcphdr *th)
 	if (rcv_wnd_new != rcv_wnd_old) {
 raise_win:
 		/* The msk-level rcv wnd is after the tcp level one,
-		 * sync the latter.
+		 * sync the latter and refresh its advertise-time scaling
+		 * snapshot.
 		 */
 		rcv_wnd_new = rcv_wnd_old;
 		win = rcv_wnd_old - ack_seq;
-		tp->rcv_wnd = min_t(u64, win, U32_MAX);
+		tcp_set_rcv_wnd(tp, min_t(u64, win, U32_MAX));
 		new_win = tp->rcv_wnd;
 
 		/* Make sure we do not exceed the maximum possible
diff --git a/net/mptcp/protocol.h b/net/mptcp/protocol.h
index 0bd1ee860316..4ea95c9c0c7a 100644
--- a/net/mptcp/protocol.h
+++ b/net/mptcp/protocol.h
@@ -408,11 +408,19 @@ static inline int mptcp_space_from_win(const struct sock *sk, int win)
 	return __tcp_space_from_win(mptcp_sk(sk)->scaling_ratio, win);
 }
 
+/* MPTCP exposes window space from the mptcp-level receive queue, so it tracks
+ * a separate backlog counter from the subflow backlog embedded in struct sock.
+ */
+static inline int mptcp_rwnd_avail(const struct sock *sk)
+{
+	return READ_ONCE(sk->sk_rcvbuf) -
+	       READ_ONCE(mptcp_sk(sk)->backlog_len) -
+	       tcp_rmem_used(sk);
+}
+
 static inline int __mptcp_space(const struct sock *sk)
 {
-	return mptcp_win_from_space(sk, READ_ONCE(sk->sk_rcvbuf) -
-				    READ_ONCE(mptcp_sk(sk)->backlog_len) -
-				    sk_rmem_alloc_get(sk));
+	return mptcp_win_from_space(sk, mptcp_rwnd_avail(sk));
 }
 
 static inline struct mptcp_data_frag *mptcp_send_head(const struct sock *sk)
-- 
2.34.1

[PATCH net 6/7] tcp: expose rmem and backlog accounting in rcvbuf_grow tracepoints

[Wesley Atwell]

The receive-window work now depends on keeping sender-visible rwnd and
hard receive-memory accounting aligned.

Expose the current rmem charge and backlog reservation in the TCP and
MPTCP rcvbuf_grow tracepoints so that later drift between advertised
window and local backing is visible during review and debugging.

Signed-off-by: Wesley Atwell <atwellwea@gmail.com>
---
 include/trace/events/mptcp.h | 11 +++++++----
 include/trace/events/tcp.h   | 12 +++++++-----
 2 files changed, 14 insertions(+), 9 deletions(-)

diff --git a/include/trace/events/mptcp.h b/include/trace/events/mptcp.h
index 269d949b2025..167970e8e0a5 100644
--- a/include/trace/events/mptcp.h
+++ b/include/trace/events/mptcp.h
@@ -199,6 +199,8 @@ TRACE_EVENT(mptcp_rcvbuf_grow,
 		__field(__u32, inq)
 		__field(__u32, space)
 		__field(__u32, ooo_space)
+		__field(__u32, rmem_alloc)
+		__field(__u32, backlog_len)
 		__field(__u32, rcvbuf)
 		__field(__u32, rcv_wnd)
 		__field(__u8, scaling_ratio)
@@ -228,6 +230,8 @@ TRACE_EVENT(mptcp_rcvbuf_grow,
 				     MPTCP_SKB_CB(msk->ooo_last_skb)->end_seq -
 				     msk->ack_seq;
 
+		__entry->rmem_alloc = tcp_rmem_used(sk);
+		__entry->backlog_len = READ_ONCE(msk->backlog_len);
 		__entry->rcvbuf = sk->sk_rcvbuf;
 		__entry->rcv_wnd = atomic64_read(&msk->rcv_wnd_sent) -
 				   msk->ack_seq;
@@ -248,12 +252,11 @@ TRACE_EVENT(mptcp_rcvbuf_grow,
 		__entry->skaddr = sk;
 	),
 
-	TP_printk("time=%u rtt_us=%u copied=%u inq=%u space=%u ooo=%u scaling_ratio=%u "
-		  "rcvbuf=%u rcv_wnd=%u family=%d sport=%hu dport=%hu saddr=%pI4 "
-		  "daddr=%pI4 saddrv6=%pI6c daddrv6=%pI6c skaddr=%p",
+	TP_printk("time=%u rtt_us=%u copied=%u inq=%u space=%u ooo=%u scaling_ratio=%u rmem_alloc=%u backlog_len=%u rcvbuf=%u rcv_wnd=%u family=%d sport=%hu dport=%hu saddr=%pI4 daddr=%pI4 saddrv6=%pI6c daddrv6=%pI6c skaddr=%p",
 		  __entry->time, __entry->rtt_us, __entry->copied,
 		  __entry->inq, __entry->space, __entry->ooo_space,
-		  __entry->scaling_ratio, __entry->rcvbuf, __entry->rcv_wnd,
+		  __entry->scaling_ratio, __entry->rmem_alloc,
+		  __entry->backlog_len, __entry->rcvbuf, __entry->rcv_wnd,
 		  __entry->family, __entry->sport, __entry->dport,
 		  __entry->saddr, __entry->daddr, __entry->saddr_v6,
 		  __entry->daddr_v6, __entry->skaddr)
diff --git a/include/trace/events/tcp.h b/include/trace/events/tcp.h
index f155f95cdb6e..92d0bd6be0ba 100644
--- a/include/trace/events/tcp.h
+++ b/include/trace/events/tcp.h
@@ -217,6 +217,8 @@ TRACE_EVENT(tcp_rcvbuf_grow,
 		__field(__u32, inq)
 		__field(__u32, space)
 		__field(__u32, ooo_space)
+		__field(__u32, rmem_alloc)
+		__field(__u32, backlog_len)
 		__field(__u32, rcvbuf)
 		__field(__u32, rcv_ssthresh)
 		__field(__u32, window_clamp)
@@ -247,6 +249,8 @@ TRACE_EVENT(tcp_rcvbuf_grow,
 				     TCP_SKB_CB(tp->ooo_last_skb)->end_seq -
 				     tp->rcv_nxt;
 
+		__entry->rmem_alloc = tcp_rmem_used(sk);
+		__entry->backlog_len = READ_ONCE(sk->sk_backlog.len);
 		__entry->rcvbuf = sk->sk_rcvbuf;
 		__entry->rcv_ssthresh = tp->rcv_ssthresh;
 		__entry->window_clamp = tp->window_clamp;
@@ -269,13 +273,11 @@ TRACE_EVENT(tcp_rcvbuf_grow,
 		__entry->sock_cookie = sock_gen_cookie(sk);
 	),
 
-	TP_printk("time=%u rtt_us=%u copied=%u inq=%u space=%u ooo=%u scaling_ratio=%u rcvbuf=%u "
-		  "rcv_ssthresh=%u window_clamp=%u rcv_wnd=%u "
-		  "family=%s sport=%hu dport=%hu saddr=%pI4 daddr=%pI4 "
-		  "saddrv6=%pI6c daddrv6=%pI6c skaddr=%p sock_cookie=%llx",
+	TP_printk("time=%u rtt_us=%u copied=%u inq=%u space=%u ooo=%u scaling_ratio=%u rmem_alloc=%u backlog_len=%u rcvbuf=%u rcv_ssthresh=%u window_clamp=%u rcv_wnd=%u family=%s sport=%hu dport=%hu saddr=%pI4 daddr=%pI4 saddrv6=%pI6c daddrv6=%pI6c skaddr=%p sock_cookie=%llx",
 		  __entry->time, __entry->rtt_us, __entry->copied,
 		  __entry->inq, __entry->space, __entry->ooo_space,
-		  __entry->scaling_ratio, __entry->rcvbuf,
+		  __entry->scaling_ratio, __entry->rmem_alloc,
+		  __entry->backlog_len, __entry->rcvbuf,
 		  __entry->rcv_ssthresh, __entry->window_clamp,
 		  __entry->rcv_wnd,
 		  show_family_name(__entry->family),
-- 
2.34.1

[PATCH net 7/7] selftests: tcp_ao: cover legacy and extended TCP_REPAIR_WINDOW layouts

[Wesley Atwell]

Extend the repair helpers and selftests so the ABI contract is pinned
down in-tree.

The TCP-AO restore coverage now exercises both the exact and legacy
TCP_REPAIR_WINDOW layouts, verifies that intermediate lengths are
rejected, and keeps the packetdrill coverage for the advertised-window
receive-memory regressions in the same net selftest series.

Signed-off-by: Wesley Atwell <atwellwea@gmail.com>
---
 .../net/packetdrill/tcp_rcv_toobig.pkt        | 35 +++++++
 .../packetdrill/tcp_rcv_toobig_default.pkt    | 97 +++++++++++++++++++
 .../testing/selftests/net/tcp_ao/lib/aolib.h  | 56 +++++++++--
 .../testing/selftests/net/tcp_ao/lib/repair.c | 18 ++--
 .../selftests/net/tcp_ao/self-connect.c       | 61 ++++++++++--
 5 files changed, 244 insertions(+), 23 deletions(-)
 create mode 100644 tools/testing/selftests/net/packetdrill/tcp_rcv_toobig.pkt
 create mode 100644 tools/testing/selftests/net/packetdrill/tcp_rcv_toobig_default.pkt

diff --git a/tools/testing/selftests/net/packetdrill/tcp_rcv_toobig.pkt b/tools/testing/selftests/net/packetdrill/tcp_rcv_toobig.pkt
new file mode 100644
index 000000000000..723c739ddc32
--- /dev/null
+++ b/tools/testing/selftests/net/packetdrill/tcp_rcv_toobig.pkt
@@ -0,0 +1,35 @@
+// SPDX-License-Identifier: GPL-2.0
+
+--mss=1000
+
+`./defaults.sh`
+
+    0 `nstat -n`
+
+// Establish a connection.
+   +0 socket(..., SOCK_STREAM, IPPROTO_TCP) = 3
+   +0 setsockopt(3, SOL_SOCKET, SO_REUSEADDR, [1], 4) = 0
+   +0 setsockopt(3, SOL_SOCKET, SO_RCVBUF, [20000], 4) = 0
+   +0 bind(3, ..., ...) = 0
+   +0 listen(3, 1) = 0
+
+   +0 < S 0:0(0) win 32792 <mss 1000,nop,wscale 7>
+   +0 > S. 0:0(0) ack 1 win 18980 <mss 1460,nop,wscale 0>
+  +.1 < . 1:1(0) ack 1 win 257
+
+   +0 accept(3, ..., ...) = 4
+
+   +0 < P. 1:20001(20000) ack 1 win 257
+ +.04 > .  1:1(0) ack 20001 win 18000
+
+   +0 setsockopt(4, SOL_SOCKET, SO_RCVBUF, [12000], 4) = 0
+   +0 < P. 20001:80001(60000) ack 1 win 257
+   +0 > .  1:1(0) ack 20001 win 18000
+
+   +0 read(4, ..., 20000) = 20000
+
+// A too big packet is accepted if the receive queue is empty, but the
+// stronger admission path must not zero the receive buffer while doing so.
+   +0 < P. 20001:80001(60000) ack 1 win 257
+   +0 > .  1:1(0) ack 80001 win 0
+   +0 %{ assert SK_MEMINFO_RCVBUF > 0, SK_MEMINFO_RCVBUF }%
diff --git a/tools/testing/selftests/net/packetdrill/tcp_rcv_toobig_default.pkt b/tools/testing/selftests/net/packetdrill/tcp_rcv_toobig_default.pkt
new file mode 100644
index 000000000000..b2e4950e0b83
--- /dev/null
+++ b/tools/testing/selftests/net/packetdrill/tcp_rcv_toobig_default.pkt
@@ -0,0 +1,97 @@
+// SPDX-License-Identifier: GPL-2.0
+
+--mss=1000
+
+`./defaults.sh
+sysctl -q net.ipv4.tcp_moderate_rcvbuf=0`
+
+// Establish a connection on the default receive buffer. Leave a large skb in
+// the queue, then deliver another one which still fits the remaining rwnd.
+// We should grow sk_rcvbuf to honor the already-advertised window instead of
+// dropping the packet.
+   +0 socket(..., SOCK_STREAM, IPPROTO_TCP) = 3
+   +0 setsockopt(3, SOL_SOCKET, SO_REUSEADDR, [1], 4) = 0
+   +0 bind(3, ..., ...) = 0
+   +0 listen(3, 1) = 0
+
+   +0 < S 0:0(0) win 65535 <mss 1000,nop,nop,sackOK,nop,wscale 7>
+   +0 > S. 0:0(0) ack 1 <...>
+  +.1 < . 1:1(0) ack 1 win 257
+
+   +0 accept(3, ..., ...) = 4
+
+// Exchange enough data to get past the completely fresh-socket case while
+// still keeping the receive buffer at its 128kB default.
+   +0 < P. 1:65001(65000) ack 1 win 257
+   * > .  1:1(0) ack 65001
+   +0 read(4, ..., 65000) = 65000
+
+   +0 < P. 65001:130001(65000) ack 1 win 257
+   * > .  1:1(0) ack 130001
+   +0 read(4, ..., 65000) = 65000
+
+   +0 < P. 130001:195001(65000) ack 1 win 257
+   * > .  1:1(0) ack 195001
+   +0 read(4, ..., 65000) = 65000
+
+   +0 < P. 195001:260001(65000) ack 1 win 257
+   * > .  1:1(0) ack 260001
+   +0 read(4, ..., 65000) = 65000
+
+   +0 < P. 260001:325001(65000) ack 1 win 257
+   * > .  1:1(0) ack 325001
+   +0 read(4, ..., 65000) = 65000
+
+   +0 < P. 325001:390001(65000) ack 1 win 257
+   * > .  1:1(0) ack 390001
+   +0 read(4, ..., 65000) = 65000
+
+   +0 < P. 390001:455001(65000) ack 1 win 257
+   * > .  1:1(0) ack 455001
+   +0 read(4, ..., 65000) = 65000
+
+   +0 < P. 455001:520001(65000) ack 1 win 257
+   * > .  1:1(0) ack 520001
+   +0 read(4, ..., 65000) = 65000
+
+   +0 < P. 520001:585001(65000) ack 1 win 257
+   * > .  1:1(0) ack 585001
+   +0 read(4, ..., 65000) = 65000
+
+   +0 < P. 585001:650001(65000) ack 1 win 257
+   * > .  1:1(0) ack 650001
+   +0 read(4, ..., 65000) = 65000
+
+   +0 < P. 650001:715001(65000) ack 1 win 257
+   * > .  1:1(0) ack 715001
+   +0 read(4, ..., 65000) = 65000
+
+   +0 < P. 715001:780001(65000) ack 1 win 257
+   * > .  1:1(0) ack 780001
+   +0 read(4, ..., 65000) = 65000
+
+   +0 < P. 780001:845001(65000) ack 1 win 257
+   * > .  1:1(0) ack 845001
+   +0 read(4, ..., 65000) = 65000
+
+   +0 < P. 845001:910001(65000) ack 1 win 257
+   * > .  1:1(0) ack 910001
+   +0 read(4, ..., 65000) = 65000
+
+   +0 < P. 910001:975001(65000) ack 1 win 257
+   * > .  1:1(0) ack 975001
+   +0 read(4, ..., 65000) = 65000
+
+   +0 < P. 975001:1040001(65000) ack 1 win 257
+   * > .  1:1(0) ack 1040001
+   +0 read(4, ..., 65000) = 65000
+
+// Leave about 60kB queued, then accept another large skb which still fits
+// the rwnd we already exposed to the peer. The regression is the drop; the
+// exact sk_rcvbuf growth path is an implementation detail.
+   +0 < P. 1040001:1102001(62000) ack 1 win 257
+   * > .  1:1(0) ack 1102001
+
+   +0 < P. 1102001:1167001(65000) ack 1 win 257
+   * > .  1:1(0) ack 1167001
+   +0 read(4, ..., 127000) = 127000
diff --git a/tools/testing/selftests/net/tcp_ao/lib/aolib.h b/tools/testing/selftests/net/tcp_ao/lib/aolib.h
index ebb2899c12fe..ff259795a4a0 100644
--- a/tools/testing/selftests/net/tcp_ao/lib/aolib.h
+++ b/tools/testing/selftests/net/tcp_ao/lib/aolib.h
@@ -13,6 +13,7 @@
 #include <linux/snmp.h>
 #include <linux/tcp.h>
 #include <netinet/in.h>
+#include <stddef.h>
 #include <stdarg.h>
 #include <stdbool.h>
 #include <stdlib.h>
@@ -671,17 +672,42 @@ struct tcp_sock_state {
 	int timestamp;
 };
 
-extern void __test_sock_checkpoint(int sk, struct tcp_sock_state *state,
-				   void *addr, size_t addr_size);
+/* Legacy userspace stops before the snapshot field and therefore exercises
+ * the kernel's unknown-snapshot fallback path.
+ */
+static inline socklen_t test_tcp_repair_window_legacy_size(void)
+{
+	return offsetof(struct tcp_repair_window, rcv_wnd_scaling_ratio);
+}
+
+static inline socklen_t test_tcp_repair_window_exact_size(void)
+{
+	return sizeof(struct tcp_repair_window);
+}
+
+void __test_sock_checkpoint_opt(int sk, struct tcp_sock_state *state,
+				socklen_t trw_len,
+				void *addr, size_t addr_size);
 static inline void test_sock_checkpoint(int sk, struct tcp_sock_state *state,
 					sockaddr_af *saddr)
 {
-	__test_sock_checkpoint(sk, state, saddr, sizeof(*saddr));
+	__test_sock_checkpoint_opt(sk, state, test_tcp_repair_window_exact_size(),
+				   saddr, sizeof(*saddr));
+}
+
+static inline void test_sock_checkpoint_legacy(int sk,
+					       struct tcp_sock_state *state,
+					       sockaddr_af *saddr)
+{
+	__test_sock_checkpoint_opt(sk, state, test_tcp_repair_window_legacy_size(),
+				   saddr, sizeof(*saddr));
 }
 extern void test_ao_checkpoint(int sk, struct tcp_ao_repair *state);
-extern void __test_sock_restore(int sk, const char *device,
-				struct tcp_sock_state *state,
-				void *saddr, void *daddr, size_t addr_size);
+void __test_sock_restore_opt(int sk, const char *device,
+			     struct tcp_sock_state *state,
+			     socklen_t trw_len,
+			     void *saddr, void *daddr,
+			     size_t addr_size);
 static inline void test_sock_restore(int sk, struct tcp_sock_state *state,
 				     sockaddr_af *saddr,
 				     const union tcp_addr daddr,
@@ -690,7 +716,23 @@ static inline void test_sock_restore(int sk, struct tcp_sock_state *state,
 	sockaddr_af addr;
 
 	tcp_addr_to_sockaddr_in(&addr, &daddr, htons(dport));
-	__test_sock_restore(sk, veth_name, state, saddr, &addr, sizeof(addr));
+	__test_sock_restore_opt(sk, veth_name, state,
+				test_tcp_repair_window_exact_size(),
+				saddr, &addr, sizeof(addr));
+}
+
+static inline void test_sock_restore_legacy(int sk,
+					    struct tcp_sock_state *state,
+					    sockaddr_af *saddr,
+					    const union tcp_addr daddr,
+					    unsigned int dport)
+{
+	sockaddr_af addr;
+
+	tcp_addr_to_sockaddr_in(&addr, &daddr, htons(dport));
+	__test_sock_restore_opt(sk, veth_name, state,
+				test_tcp_repair_window_legacy_size(),
+				saddr, &addr, sizeof(addr));
 }
 extern void test_ao_restore(int sk, struct tcp_ao_repair *state);
 extern void test_sock_state_free(struct tcp_sock_state *state);
diff --git a/tools/testing/selftests/net/tcp_ao/lib/repair.c b/tools/testing/selftests/net/tcp_ao/lib/repair.c
index 9893b3ba69f5..befbd0f72db5 100644
--- a/tools/testing/selftests/net/tcp_ao/lib/repair.c
+++ b/tools/testing/selftests/net/tcp_ao/lib/repair.c
@@ -66,8 +66,9 @@ static void test_sock_checkpoint_queue(int sk, int queue, int qlen,
 		test_error("recv(%d): %d", qlen, ret);
 }
 
-void __test_sock_checkpoint(int sk, struct tcp_sock_state *state,
-			    void *addr, size_t addr_size)
+void __test_sock_checkpoint_opt(int sk, struct tcp_sock_state *state,
+				socklen_t trw_len,
+				void *addr, size_t addr_size)
 {
 	socklen_t len = sizeof(state->info);
 	int ret;
@@ -82,9 +83,9 @@ void __test_sock_checkpoint(int sk, struct tcp_sock_state *state,
 	if (getsockname(sk, addr, &len) || len != addr_size)
 		test_error("getsockname(): %d", (int)len);
 
-	len = sizeof(state->trw);
+	len = trw_len;
 	ret = getsockopt(sk, SOL_TCP, TCP_REPAIR_WINDOW, &state->trw, &len);
-	if (ret || len != sizeof(state->trw))
+	if (ret || len != trw_len)
 		test_error("getsockopt(TCP_REPAIR_WINDOW): %d", (int)len);
 
 	if (ioctl(sk, SIOCOUTQ, &state->outq_len))
@@ -160,9 +161,10 @@ static void test_sock_restore_queue(int sk, int queue, void *buf, int len)
 	} while (len > 0);
 }
 
-void __test_sock_restore(int sk, const char *device,
-			 struct tcp_sock_state *state,
-			 void *saddr, void *daddr, size_t addr_size)
+void __test_sock_restore_opt(int sk, const char *device,
+			     struct tcp_sock_state *state,
+			     socklen_t trw_len,
+			     void *saddr, void *daddr, size_t addr_size)
 {
 	struct tcp_repair_opt opts[4];
 	unsigned int opt_nr = 0;
@@ -215,7 +217,7 @@ void __test_sock_restore(int sk, const char *device,
 	}
 	test_sock_restore_queue(sk, TCP_RECV_QUEUE, state->in.buf, state->inq_len);
 	test_sock_restore_queue(sk, TCP_SEND_QUEUE, state->out.buf, state->outq_len);
-	if (setsockopt(sk, SOL_TCP, TCP_REPAIR_WINDOW, &state->trw, sizeof(state->trw)))
+	if (setsockopt(sk, SOL_TCP, TCP_REPAIR_WINDOW, &state->trw, trw_len))
 		test_error("setsockopt(TCP_REPAIR_WINDOW)");
 }
 
diff --git a/tools/testing/selftests/net/tcp_ao/self-connect.c b/tools/testing/selftests/net/tcp_ao/self-connect.c
index 2c73bea698a6..a7edd72ab28d 100644
--- a/tools/testing/selftests/net/tcp_ao/self-connect.c
+++ b/tools/testing/selftests/net/tcp_ao/self-connect.c
@@ -4,6 +4,7 @@
 #include "aolib.h"
 
 static union tcp_addr local_addr;
+static bool checked_repair_window_lens;
 
 static void __setup_lo_intf(const char *lo_intf,
 			    const char *addr_str, uint8_t prefix)
@@ -30,8 +31,40 @@ static void setup_lo_intf(const char *lo_intf)
 #endif
 }
 
+/* The repair ABI accepts exactly the legacy and extended layouts. */
+static void test_repair_window_len_contract(int sk)
+{
+	struct tcp_repair_window trw = {};
+	socklen_t len = test_tcp_repair_window_exact_size();
+	socklen_t bad_len = test_tcp_repair_window_legacy_size() + 1;
+	int ret;
+
+	if (checked_repair_window_lens)
+		return;
+
+	checked_repair_window_lens = true;
+
+	ret = getsockopt(sk, SOL_TCP, TCP_REPAIR_WINDOW, &trw, &len);
+	if (ret || len != test_tcp_repair_window_exact_size())
+		test_error("getsockopt(TCP_REPAIR_WINDOW): %d", (int)len);
+
+	len = bad_len;
+	ret = getsockopt(sk, SOL_TCP, TCP_REPAIR_WINDOW, &trw, &len);
+	if (ret == 0 || errno != EINVAL)
+		test_fail("repair-window get rejects invalid len");
+	else
+		test_ok("repair-window get rejects invalid len");
+
+	ret = setsockopt(sk, SOL_TCP, TCP_REPAIR_WINDOW, &trw, bad_len);
+	if (ret == 0 || errno != EINVAL)
+		test_fail("repair-window set rejects invalid len");
+	else
+		test_ok("repair-window set rejects invalid len");
+}
+
 static void tcp_self_connect(const char *tst, unsigned int port,
-			     bool different_keyids, bool check_restore)
+			     bool different_keyids, bool check_restore,
+			     bool legacy_repair_window)
 {
 	struct tcp_counters before, after;
 	uint64_t before_aogood, after_aogood;
@@ -109,7 +142,11 @@ static void tcp_self_connect(const char *tst, unsigned int port,
 	}
 
 	test_enable_repair(sk);
-	test_sock_checkpoint(sk, &img, &addr);
+	test_repair_window_len_contract(sk);
+	if (legacy_repair_window)
+		test_sock_checkpoint_legacy(sk, &img, &addr);
+	else
+		test_sock_checkpoint(sk, &img, &addr);
 #ifdef IPV6_TEST
 	addr.sin6_port = htons(port + 1);
 #else
@@ -123,7 +160,11 @@ static void tcp_self_connect(const char *tst, unsigned int port,
 		test_error("socket()");
 
 	test_enable_repair(sk);
-	__test_sock_restore(sk, "lo", &img, &addr, &addr, sizeof(addr));
+	__test_sock_restore_opt(sk, "lo", &img,
+				legacy_repair_window ?
+				test_tcp_repair_window_legacy_size() :
+				test_tcp_repair_window_exact_size(),
+				&addr, &addr, sizeof(addr));
 	if (different_keyids) {
 		if (test_add_repaired_key(sk, DEFAULT_TEST_PASSWORD, 0,
 					  local_addr, -1, 7, 5))
@@ -165,20 +206,24 @@ static void *client_fn(void *arg)
 
 	setup_lo_intf("lo");
 
-	tcp_self_connect("self-connect(same keyids)", port++, false, false);
+	tcp_self_connect("self-connect(same keyids)", port++, false, false, false);
 
 	/* expecting rnext to change based on the first segment RNext != Current */
 	trace_ao_event_expect(TCP_AO_RNEXT_REQUEST, local_addr, local_addr,
 			      port, port, 0, -1, -1, -1, -1, -1, 7, 5, -1);
-	tcp_self_connect("self-connect(different keyids)", port++, true, false);
-	tcp_self_connect("self-connect(restore)", port, false, true);
+	tcp_self_connect("self-connect(different keyids)", port++, true, false, false);
+	tcp_self_connect("self-connect(restore)", port, false, true, false);
+	port += 2; /* restore test restores over different port */
+	tcp_self_connect("self-connect(restore, legacy repair window)",
+			 port, false, true, true);
 	port += 2; /* restore test restores over different port */
 	trace_ao_event_expect(TCP_AO_RNEXT_REQUEST, local_addr, local_addr,
 			      port, port, 0, -1, -1, -1, -1, -1, 7, 5, -1);
 	/* intentionally on restore they are added to the socket in different order */
 	trace_ao_event_expect(TCP_AO_RNEXT_REQUEST, local_addr, local_addr,
 			      port + 1, port + 1, 0, -1, -1, -1, -1, -1, 5, 7, -1);
-	tcp_self_connect("self-connect(restore, different keyids)", port, true, true);
+	tcp_self_connect("self-connect(restore, different keyids)",
+			 port, true, true, false);
 	port += 2; /* restore test restores over different port */
 
 	return NULL;
@@ -186,6 +231,6 @@ static void *client_fn(void *arg)
 
 int main(int argc, char *argv[])
 {
-	test_init(5, client_fn, NULL);
+	test_init(8, client_fn, NULL);
 	return 0;
 }
-- 
2.34.1

Re: [PATCH net 0/7] tcp: preserve advertised rwnd accounting across receive-memory decisions

[Eric Dumazet]

On Wed, Mar 11, 2026 at 8:56 AM Wesley Atwell <atwellwea@gmail.com> wrote:
>
> This series keeps sender-visible TCP receive-window accounting tied to the
> scaling basis that was in force when the window was advertised.
>
> Problem
> -------
>
> `tp->rcv_wnd` is an advertised promise to the sender, but later
> receive-memory admission and clamping could reconstruct that promise
> through the mutable live `scaling_ratio`. After ratio drift, the stack
> could retain or advertise a receive window that no longer matched the
> local hard rmem budget.
>
> Fix
> ---
>
> - store the advertise-time scaling basis alongside `tp->rcv_wnd`
> - refresh that pair at the TCP and MPTCP receive-window write sites
> - consume the snapshot in receive-memory admission, clamping, and the
>   scaled-window quantization path
> - preserve the snapshot across `TCP_REPAIR_WINDOW` restore when userspace
>   provides it, and fall back safely when legacy userspace cannot
> - expose the accounting in tracepoints and cover the ABI/runtime contract
>   in selftests
>

Your series will heavily conflict with Simon's one

https://patchwork.kernel.org/project/netdevbpf/list/?series=1063486&state=%2A&archive=both

I suggest you rebase/retest/resend after we merge it.

> Series layout
> -------------
>
> 1. track the receive-window snapshot state and helpers
> 2. refresh the snapshot when TCP advertises or initializes windows
> 3. use the snapshot in receive-memory admission and clamping
> 4. extend `TCP_REPAIR_WINDOW` for exact restore plus legacy compatibility
> 5. refresh the TCP shadow window snapshot in MPTCP
> 6. expose rmem/backlog state in `rcvbuf_grow` tracepoints
> 7. cover legacy and extended repair-window layouts in selftests
>
> Testing
> -------
>
> - `git diff --check origin/main..HEAD`
> - `scripts/checkpatch.pl --strict --show-types` on patches 1-7
> - `make -j8 headers`
> - `make -j8 net/ipv4/tcp_input.o net/ipv4/tcp_output.o net/ipv4/tcp_minisocks.o net/ipv4/tcp.o`
> - `make -j8 C=1 CF='-D__CHECK_ENDIAN__' W=1 net/ipv4/tcp_input.o net/ipv4/tcp_output.o net/ipv4/tcp_minisocks.o net/ipv4/tcp.o`
> - `make SPHINXDIRS='networking/net_cachelines' htmldocs`
> - `make -j8 vmlinux bzImage modules`
> - `make -C tools/testing/selftests/net/tcp_ao -j8`
> - `make -C tools/testing/selftests/net/mptcp -j8`
> - `packetdrill --dry_run` for `tcp_rcv_toobig.pkt` and
>   `tcp_rcv_toobig_default.pkt`
> - `virtme-run` guest pass for both packetdrill tests
> - feature-enabled guest pass for `restore_ipv4`, `self-connect_ipv4`, and
>   `mptcp_sockopt.sh`
>
> Thanks,
> Wesley
>
> ---
> base-commit: 908c344d5cfa0ee6efb3226d22ea661e078ebfa0
> --
> 2.43.0
>

Re: [PATCH net 0/7] tcp: preserve advertised rwnd accounting across receive-memory decisions

[Jakub Kicinski]

On Wed, 11 Mar 2026 09:34:32 +0100 Eric Dumazet wrote:
> Your series will heavily conflict with Simon's one
> 
> https://patchwork.kernel.org/project/netdevbpf/list/?series=1063486&state=%2A&archive=both
> 
> I suggest you rebase/retest/resend after we merge it.

Would it make sense to extend netdevsim and packetdrill to be able to
exercise scaling ratio a little more? Having it optionally clone the
skb and truesize += X would be trivial. IDK how many bugs this would
let us catch tho :(

Re: [PATCH net 0/7] tcp: preserve advertised rwnd accounting across receive-memory decisions

[Jakub Kicinski]

On Wed, 11 Mar 2026 01:55:53 -0600 Wesley Atwell wrote:
> Subject: [PATCH net 0/7] tcp: preserve advertised rwnd accounting across receive-memory decisions

when you repost please make sure you use "PATCH net-next v2" 
as the tag / prefix. "net" is a tree we use to fast track fixes.

Re: [PATCH net 0/7] tcp: preserve advertised rwnd accounting across receive-memory decisions

[Eric Dumazet]

On Thu, Mar 12, 2026 at 1:41 AM Jakub Kicinski <kuba@kernel.org> wrote:
>
> On Wed, 11 Mar 2026 09:34:32 +0100 Eric Dumazet wrote:
> > Your series will heavily conflict with Simon's one
> >
> > https://patchwork.kernel.org/project/netdevbpf/list/?series=1063486&state=%2A&archive=both
> >
> > I suggest you rebase/retest/resend after we merge it.
>
> Would it make sense to extend netdevsim and packetdrill to be able to
> exercise scaling ratio a little more? Having it optionally clone the
> skb and truesize += X would be trivial. IDK how many bugs this would
> let us catch tho :(

Yes, I think we mentioned this at some point.
packetdrill uses tun device.
Adding a TUN ioctl() to control how many additional bytes are added to
skb->truesize after tun allocates an skb is doable.