[PATCH net-next v2 00/14] tcp: preserve receive-window accounting across ratio drift

[PATCH net-next v2 00/14] tcp: preserve receive-window accounting across ratio drift

[atwellwea]

From: Wesley Atwell <atwellwea@gmail.com>

This series keeps sender-visible TCP receive-window accounting tied to the
scaling basis that was in force when the window was advertised, even if
later receive-side truesize inflation lowers scaling_ratio or the live
receive window retracts below the largest right edge already exposed to the
sender.

After the receive-window retraction changes, the receive path needs to keep
track of two related pieces of sender-visible state:

  1. the live advertised receive window
  2. the maximum advertised right edge and the basis it was exposed with

This repost snapshots both, uses them to repair receive-buffer backing when
ratio drift would otherwise strand sender-visible space, extends
TCP_REPAIR_WINDOW so repair/restore can round-trip the new state, and adds
truesize-drift coverage through TUN packetdrill tests and netdevsim-based
selftests.

v2:
- repost to net-next and use the [PATCH net-next v2] prefix
- rebase the receive-window accounting changes on top of the retraction
  model
- split the series more finely
- snapshot both the live rwnd basis and the max advertised-window basis
- extend TCP_REPAIR_WINDOW to preserve legacy, v1, and current layouts
- add TUN RX truesize injection and packetdrill coverage for ratio drift
- split the generic netdevsim PSP extension cleanup into its own final
  patch after the peer RX truesize support
- add the requested ABI/runtime comments at the non-obvious review points

Testing:

- full runtime selftest coverage for netdevsim, tcp_ao, mptcp, and
  packetdrill; all runtime suites completed successfully
- tcp_ao completed 24/24 top-level tests, covering 803 passing checks,
  6 expected failures, 36 skips, and 0 unexpected failures
- mptcp completed 588 passing checks in aggregate, with 28 skips and
  0 unexpected failures
- packetdrill completed 219/219 runtime cases with 0 failures,
  including the new tests
- netdevsim completed 18/18 top-level runtime tests with 0 failures,
  including the peer RX truesize and related netdevsim coverage used by
  this series

Wesley Atwell (14):
  tcp: factor receive-memory accounting helpers
  tcp: snapshot advertise-time scaling for rcv_wnd
  tcp: refresh rcv_wnd snapshots at TCP write sites
  tcp: snapshot the maximum advertised receive window
  tcp: grow rcvbuf to back scaled-window quantization slack
  tcp: regrow rcvbuf when scaling_ratio drops after advertisement
  tcp: honor the maximum advertised window after live retraction
  tcp: extend TCP_REPAIR_WINDOW for live and max-window snapshots
  mptcp: refresh TCP receive-window snapshots on subflows
  tcp: expose rmem and backlog in tcp and mptcp rcvbuf_grow tracepoints
  selftests: tcp_ao: cover legacy, v1, and retracted repair windows
  tun/selftests: add RX truesize injection for TCP window tests
  netdevsim: add peer RX truesize support for selftests
  netdevsim: release pinned PSP ext on drop paths

 .../networking/net_cachelines/tcp_sock.rst    |   2 +
 drivers/net/netdevsim/netdev.c                | 156 ++++++-
 drivers/net/netdevsim/netdevsim.h             |   4 +
 drivers/net/tun.c                             |  65 +++
 include/linux/tcp.h                           |   2 +
 include/net/tcp.h                             | 118 ++++-
 include/trace/events/mptcp.h                  |  11 +-
 include/trace/events/tcp.h                    |  12 +-
 include/uapi/linux/if_tun.h                   |   4 +
 include/uapi/linux/tcp.h                      |   8 +
 net/ipv4/tcp.c                                |  75 ++-
 net/ipv4/tcp_fastopen.c                       |   2 +-
 net/ipv4/tcp_input.c                          | 160 ++++++-
 net/ipv4/tcp_minisocks.c                      |   4 +-
 net/ipv4/tcp_output.c                         |  25 +-
 net/mptcp/options.c                           |  14 +-
 net/mptcp/protocol.h                          |  14 +-
 .../selftests/drivers/net/netdevsim/Makefile  |   1 +
 .../drivers/net/netdevsim/peer-rx-truesize.sh | 426 ++++++++++++++++++
 .../tcp_rcv_neg_window_truesize.pkt           | 143 ++++++
 .../net/packetdrill/tcp_rcv_toobig.pkt        |  35 ++
 .../packetdrill/tcp_rcv_toobig_default.pkt    |  97 ++++
 .../tcp_rcv_toobig_default_truesize.pkt       | 118 +++++
 .../tcp_rcv_wnd_shrink_allowed_truesize.pkt   |  49 ++
 .../testing/selftests/net/tcp_ao/lib/aolib.h  |  83 +++-
 .../testing/selftests/net/tcp_ao/lib/repair.c |  18 +-
 .../selftests/net/tcp_ao/self-connect.c       | 201 ++++++++-
 tools/testing/selftests/net/tun.c             | 140 +++++-
 28 files changed, 1911 insertions(+), 76 deletions(-)
 create mode 100755 tools/testing/selftests/drivers/net/netdevsim/peer-rx-truesize.sh
 create mode 100644 tools/testing/selftests/net/packetdrill/tcp_rcv_neg_window_truesize.pkt
 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
 create mode 100644 tools/testing/selftests/net/packetdrill/tcp_rcv_toobig_default_truesize.pkt
 create mode 100644 tools/testing/selftests/net/packetdrill/tcp_rcv_wnd_shrink_allowed_truesize.pkt


base-commit: f807b5b9b89eb9220d034115c272c312251cbcac
-- 
2.43.0

[PATCH net-next v2 01/14] tcp: factor receive-memory accounting helpers

[atwellwea]

From: Wesley Atwell <atwellwea@gmail.com>

Factor the core receive-memory byte accounting into small helpers so
window selection, pressure checks, and prune decisions all start from
one set of quantities.

This is preparatory only. Later patches will use the same helpers when
tying sender-visible receive-window state back to hard memory admission.

Signed-off-by: Wesley Atwell <atwellwea@gmail.com>
---
 include/net/tcp.h    | 32 +++++++++++++++++++++++++++-----
 net/ipv4/tcp_input.c |  2 +-
 2 files changed, 28 insertions(+), 6 deletions(-)

diff --git a/include/net/tcp.h b/include/net/tcp.h
index f87bdacb5a69..3a0060599afe 100644
--- a/include/net/tcp.h
+++ b/include/net/tcp.h
@@ -1751,12 +1751,34 @@ static inline void tcp_scaling_ratio_init(struct sock *sk)
 	tcp_sk(sk)->scaling_ratio = TCP_DEFAULT_SCALING_RATIO;
 }
 
+/* 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)
@@ -1799,7 +1821,7 @@ 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)
@@ -1949,7 +1971,7 @@ 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_input.c b/net/ipv4/tcp_input.c
index e6b2f4be7723..b8e65e31255e 100644
--- a/net/ipv4/tcp_input.c
+++ b/net/ipv4/tcp_input.c
@@ -5959,7 +5959,7 @@ 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);
-- 
2.43.0

[PATCH net-next v2 02/14] tcp: snapshot advertise-time scaling for rcv_wnd

[atwellwea]

From: Wesley Atwell <atwellwea@gmail.com>

Track the scaling basis that was in force when tp->rcv_wnd was last
advertised, and provide helpers to refresh or interpret that snapshot.

Later patches use this live-window basis to preserve sender-visible rwnd
accounting when receive-side memory costs drift after advertisement.

Signed-off-by: Wesley Atwell <atwellwea@gmail.com>
---
 .../networking/net_cachelines/tcp_sock.rst    |  1 +
 include/linux/tcp.h                           |  1 +
 include/net/tcp.h                             | 52 ++++++++++++++++++-
 net/ipv4/tcp.c                                |  1 +
 4 files changed, 54 insertions(+), 1 deletion(-)

diff --git a/Documentation/networking/net_cachelines/tcp_sock.rst b/Documentation/networking/net_cachelines/tcp_sock.rst
index fecf61166a54..09ece1c59c2d 100644
--- a/Documentation/networking/net_cachelines/tcp_sock.rst
+++ b/Documentation/networking/net_cachelines/tcp_sock.rst
@@ -11,6 +11,7 @@ Type                          Name                    fastpath_tx_access  fastpa
 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
+u8                            rcv_wnd_scaling_ratio   read_write          read_mostly         tcp_set_rcv_wnd,tcp_can_ingest,tcp_repair_set_window,do_tcp_getsockopt
 __be32                        pred_flags              read_write          read_mostly         tcp_select_window(tx);tcp_rcv_established(rx)
 u64                           bytes_received                              read_write          tcp_rcv_nxt_update(rx)
 u32                           segs_in                                     read_write          tcp_v6_rcv(rx)
diff --git a/include/linux/tcp.h b/include/linux/tcp.h
index 6982f10e826b..2ace563d59d6 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 3a0060599afe..6fa7cdb0979e 100644
--- a/include/net/tcp.h
+++ b/include/net/tcp.h
@@ -1741,6 +1741,31 @@ 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_wnd_snapshot_valid(u8 scaling_ratio)
+{
+	return scaling_ratio != 0;
+}
+
+static inline bool tcp_space_from_wnd_snapshot(u8 scaling_ratio, int win,
+					       int *space)
+{
+	if (!tcp_wnd_snapshot_valid(scaling_ratio))
+		return false;
+
+	*space = __tcp_space_from_win(scaling_ratio, win);
+	return true;
+}
+
+/* Rebuild hard receive-memory units for data already covered by tp->rcv_wnd if
+ * the advertise-time basis is known.
+ */
+static inline bool tcp_space_from_rcv_wnd(const struct tcp_sock *tp, int win,
+					  int *space)
+{
+	return tcp_space_from_wnd_snapshot(tp->rcv_wnd_scaling_ratio, win,
+					   space);
+}
+
 /* Assume a 50% default for skb->len/skb->truesize ratio.
  * This may be adjusted later in tcp_measure_rcv_mss().
  */
@@ -1748,7 +1773,32 @@ 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. Callers can leave a zero snapshot when the
+ * advertise-time basis is unknown and refresh the pair on the next local
+ * window update.
+ */
+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
diff --git a/net/ipv4/tcp.c b/net/ipv4/tcp.c
index 516087c622ad..0383ee8d3b78 100644
--- a/net/ipv4/tcp.c
+++ b/net/ipv4/tcp.c
@@ -5275,6 +5275,7 @@ 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_mwnd_seq);
 	CACHELINE_ASSERT_GROUP_MEMBER(struct tcp_sock, tcp_sock_write_txrx, rcv_tstamp);
-- 
2.43.0

[PATCH net-next v2 03/14] tcp: refresh rcv_wnd snapshots at TCP write sites

[atwellwea]

From: Wesley Atwell <atwellwea@gmail.com>

Refresh the live rwnd snapshot whenever TCP updates tp->rcv_wnd at the
normal write sites, including child setup, tcp_select_window(), and the
initial connect-time window selection.

This keeps the live sender-visible window paired with the scaling basis
that was actually advertised.

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 d350d794a959..1c02c9cd13fe 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->rcv_mwnd_seq = newtp->rcv_wup + req->rsk_rcv_wnd;
 	newtp->rx_opt.wscale_ok = ireq->wscale_ok;
 	if (newtp->rx_opt.wscale_ok) {
diff --git a/net/ipv4/tcp_output.c b/net/ipv4/tcp_output.c
index 35c3b0ab5a0c..0b082726d7c4 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;
 		tcp_update_max_rcv_wnd_seq(tp);
 		return 0;
@@ -315,7 +315,7 @@ 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;
 	tcp_update_max_rcv_wnd_seq(tp);
 
@@ -4148,6 +4148,10 @@ 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;
-- 
2.43.0

[PATCH net-next v2 04/14] tcp: snapshot the maximum advertised receive window

[atwellwea]

From: Wesley Atwell <atwellwea@gmail.com>

Track the maximum sender-visible receive-window right edge separately
from the live rwnd, along with the scaling basis that was in force when
that larger window was advertised.

This gives later admission and restore paths enough information to
reason about retracted windows without losing the original sender-
visible bound.

Signed-off-by: Wesley Atwell <atwellwea@gmail.com>
---
 .../networking/net_cachelines/tcp_sock.rst    |  1 +
 include/linux/tcp.h                           |  1 +
 include/net/tcp.h                             | 21 ++++++++++++++++++-
 net/ipv4/tcp.c                                |  1 +
 net/ipv4/tcp_fastopen.c                       |  2 +-
 net/ipv4/tcp_input.c                          |  4 ++--
 net/ipv4/tcp_minisocks.c                      |  2 +-
 net/ipv4/tcp_output.c                         |  2 +-
 8 files changed, 28 insertions(+), 6 deletions(-)

diff --git a/Documentation/networking/net_cachelines/tcp_sock.rst b/Documentation/networking/net_cachelines/tcp_sock.rst
index 09ece1c59c2d..d58a3b1eb55d 100644
--- a/Documentation/networking/net_cachelines/tcp_sock.rst
+++ b/Documentation/networking/net_cachelines/tcp_sock.rst
@@ -11,6 +11,7 @@ Type                          Name                    fastpath_tx_access  fastpa
 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
+u8                            rcv_mwnd_scaling_ratio  read_write          read_mostly         tcp_init_max_rcv_wnd_seq,tcp_update_max_rcv_wnd_seq,tcp_repair_set_window,do_tcp_getsockopt
 u8                            rcv_wnd_scaling_ratio   read_write          read_mostly         tcp_set_rcv_wnd,tcp_can_ingest,tcp_repair_set_window,do_tcp_getsockopt
 __be32                        pred_flags              read_write          read_mostly         tcp_select_window(tx);tcp_rcv_established(rx)
 u64                           bytes_received                              read_write          tcp_rcv_nxt_update(rx)
diff --git a/include/linux/tcp.h b/include/linux/tcp.h
index 2ace563d59d6..e5d7a65ac439 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_mwnd_scaling_ratio; /* 0 if unknown, else tp->rcv_mwnd_seq basis */
 	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()) */
diff --git a/include/net/tcp.h b/include/net/tcp.h
index 6fa7cdb0979e..fc22ab6b80d5 100644
--- a/include/net/tcp.h
+++ b/include/net/tcp.h
@@ -947,13 +947,21 @@ static inline u32 tcp_max_receive_window(const struct tcp_sock *tp)
 	return (u32) win;
 }
 
+static inline void tcp_init_max_rcv_wnd_seq(struct tcp_sock *tp)
+{
+	tp->rcv_mwnd_seq = tp->rcv_wup + tp->rcv_wnd;
+	tp->rcv_mwnd_scaling_ratio = tp->rcv_wnd_scaling_ratio;
+}
+
 /* Check if we need to update the maximum receive window sequence number */
 static inline void tcp_update_max_rcv_wnd_seq(struct tcp_sock *tp)
 {
 	u32 wre = tp->rcv_wup + tp->rcv_wnd;
 
-	if (after(wre, tp->rcv_mwnd_seq))
+	if (after(wre, tp->rcv_mwnd_seq)) {
 		tp->rcv_mwnd_seq = wre;
+		tp->rcv_mwnd_scaling_ratio = tp->rcv_wnd_scaling_ratio;
+	}
 }
 
 /* Choose a new window, without checks for shrinking, and without
@@ -1766,6 +1774,16 @@ static inline bool tcp_space_from_rcv_wnd(const struct tcp_sock *tp, int win,
 					   space);
 }
 
+/* Same as tcp_space_from_rcv_wnd(), but for the remembered maximum
+ * sender-visible receive window.
+ */
+static inline bool tcp_space_from_rcv_mwnd(const struct tcp_sock *tp, int win,
+					   int *space)
+{
+	return tcp_space_from_wnd_snapshot(tp->rcv_mwnd_scaling_ratio, win,
+					   space);
+}
+
 /* Assume a 50% default for skb->len/skb->truesize ratio.
  * This may be adjusted later in tcp_measure_rcv_mss().
  */
@@ -1776,6 +1794,7 @@ static inline void tcp_scaling_ratio_init(struct sock *sk)
 	struct tcp_sock *tp = tcp_sk(sk);
 
 	tp->scaling_ratio = TCP_DEFAULT_SCALING_RATIO;
+	tp->rcv_mwnd_scaling_ratio = TCP_DEFAULT_SCALING_RATIO;
 	tp->rcv_wnd_scaling_ratio = TCP_DEFAULT_SCALING_RATIO;
 }
 
diff --git a/net/ipv4/tcp.c b/net/ipv4/tcp.c
index 0383ee8d3b78..66706dbb90f5 100644
--- a/net/ipv4/tcp.c
+++ b/net/ipv4/tcp.c
@@ -5275,6 +5275,7 @@ 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_mwnd_scaling_ratio);
 	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_mwnd_seq);
diff --git a/net/ipv4/tcp_fastopen.c b/net/ipv4/tcp_fastopen.c
index 4e389d609f91..56113cf2a165 100644
--- a/net/ipv4/tcp_fastopen.c
+++ b/net/ipv4/tcp_fastopen.c
@@ -377,7 +377,7 @@ static struct sock *tcp_fastopen_create_child(struct sock *sk,
 
 	tcp_rsk(req)->rcv_nxt = tp->rcv_nxt;
 	tp->rcv_wup = tp->rcv_nxt;
-	tp->rcv_mwnd_seq = tp->rcv_wup + tp->rcv_wnd;
+	tcp_init_max_rcv_wnd_seq(tp);
 	/* tcp_conn_request() is sending the SYNACK,
 	 * and queues the child into listener accept queue.
 	 */
diff --git a/net/ipv4/tcp_input.c b/net/ipv4/tcp_input.c
index b8e65e31255e..352f814a4ff6 100644
--- a/net/ipv4/tcp_input.c
+++ b/net/ipv4/tcp_input.c
@@ -6902,7 +6902,7 @@ static int tcp_rcv_synsent_state_process(struct sock *sk, struct sk_buff *skb,
 		 */
 		WRITE_ONCE(tp->rcv_nxt, TCP_SKB_CB(skb)->seq + 1);
 		tp->rcv_wup = TCP_SKB_CB(skb)->seq + 1;
-		tp->rcv_mwnd_seq = tp->rcv_wup + tp->rcv_wnd;
+		tcp_init_max_rcv_wnd_seq(tp);
 
 		/* RFC1323: The window in SYN & SYN/ACK segments is
 		 * never scaled.
@@ -7015,7 +7015,7 @@ static int tcp_rcv_synsent_state_process(struct sock *sk, struct sk_buff *skb,
 		WRITE_ONCE(tp->rcv_nxt, TCP_SKB_CB(skb)->seq + 1);
 		WRITE_ONCE(tp->copied_seq, tp->rcv_nxt);
 		tp->rcv_wup = TCP_SKB_CB(skb)->seq + 1;
-		tp->rcv_mwnd_seq = tp->rcv_wup + tp->rcv_wnd;
+		tcp_init_max_rcv_wnd_seq(tp);
 
 		/* RFC1323: The window in SYN & SYN/ACK segments is
 		 * never scaled.
diff --git a/net/ipv4/tcp_minisocks.c b/net/ipv4/tcp_minisocks.c
index 1c02c9cd13fe..85bd9580caf9 100644
--- a/net/ipv4/tcp_minisocks.c
+++ b/net/ipv4/tcp_minisocks.c
@@ -604,7 +604,7 @@ struct sock *tcp_create_openreq_child(const struct sock *sk,
 	newtp->window_clamp = req->rsk_window_clamp;
 	newtp->rcv_ssthresh = req->rsk_rcv_wnd;
 	tcp_set_rcv_wnd(newtp, req->rsk_rcv_wnd);
-	newtp->rcv_mwnd_seq = newtp->rcv_wup + req->rsk_rcv_wnd;
+	tcp_init_max_rcv_wnd_seq(newtp);
 	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 0b082726d7c4..57a2a6daaad3 100644
--- a/net/ipv4/tcp_output.c
+++ b/net/ipv4/tcp_output.c
@@ -4171,7 +4171,7 @@ static void tcp_connect_init(struct sock *sk)
 	else
 		tp->rcv_tstamp = tcp_jiffies32;
 	tp->rcv_wup = tp->rcv_nxt;
-	tp->rcv_mwnd_seq = tp->rcv_nxt + tp->rcv_wnd;
+	tcp_init_max_rcv_wnd_seq(tp);
 	WRITE_ONCE(tp->copied_seq, tp->rcv_nxt);
 
 	inet_csk(sk)->icsk_rto = tcp_timeout_init(sk);
-- 
2.43.0

[PATCH net-next v2 05/14] tcp: grow rcvbuf to back scaled-window quantization slack

[atwellwea]

From: Wesley Atwell <atwellwea@gmail.com>

Teach TCP to grow sk_rcvbuf when scale rounding would otherwise expose
more sender-visible window than the current hard receive-memory backing
can cover.

The new helper keeps backlog and memory-pressure limits in the same
units as the rest of the receive path, while __tcp_select_window()
backs any rounding slack before advertising it.

Signed-off-by: Wesley Atwell <atwellwea@gmail.com>
---
 include/net/tcp.h     | 12 ++++++++++++
 net/ipv4/tcp_input.c  | 36 ++++++++++++++++++++++++++++++++++--
 net/ipv4/tcp_output.c | 15 +++++++++++++--
 3 files changed, 59 insertions(+), 4 deletions(-)

diff --git a/include/net/tcp.h b/include/net/tcp.h
index fc22ab6b80d5..5b479ad44f89 100644
--- a/include/net/tcp.h
+++ b/include/net/tcp.h
@@ -397,6 +397,7 @@ 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);
@@ -1844,6 +1845,17 @@ static inline int tcp_rwnd_avail(const struct sock *sk)
 	return tcp_rmem_avail(sk) - READ_ONCE(sk->sk_backlog.len);
 }
 
+/* Passive children clone the listener's sk_socket until accept() grafts
+ * their own struct socket, so only sockets that point back to themselves
+ * should autotune receive-buffer backing.
+ */
+static inline bool tcp_rcvbuf_grow_allowed(const struct sock *sk)
+{
+	struct socket *sock = READ_ONCE(sk->sk_socket);
+
+	return sock && READ_ONCE(sock->sk) == sk;
+}
+
 /* Note: caller must be prepared to deal with negative returns */
 static inline int tcp_space(const struct sock *sk)
 {
diff --git a/net/ipv4/tcp_input.c b/net/ipv4/tcp_input.c
index 352f814a4ff6..32256519a085 100644
--- a/net/ipv4/tcp_input.c
+++ b/net/ipv4/tcp_input.c
@@ -774,6 +774,38 @@ 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 sk_rmem_alloc while preserving sender-visible headroom
+ * already consumed by sk_backlog.len.
+ */
+bool tcp_try_grow_rcvbuf(struct sock *sk, int needed)
+{
+	struct net *net = sock_net(sk);
+	int backlog;
+	int rmem2;
+	int target;
+
+	needed = max(needed, 0);
+	backlog = READ_ONCE(sk->sk_backlog.len);
+	target = tcp_rmem_used(sk) + backlog + 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)
 {
@@ -785,14 +817,14 @@ 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);
 }
 
diff --git a/net/ipv4/tcp_output.c b/net/ipv4/tcp_output.c
index 57a2a6daaad3..53781cf591d2 100644
--- a/net/ipv4/tcp_output.c
+++ b/net/ipv4/tcp_output.c
@@ -3375,13 +3375,24 @@ 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_rcvbuf_grow_allowed(sk) ||
+		     !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.
-- 
2.43.0

[PATCH net-next v2 06/14] tcp: regrow rcvbuf when scaling_ratio drops after advertisement

[atwellwea]

From: Wesley Atwell <atwellwea@gmail.com>

When tcp_measure_rcv_mss() lowers scaling_ratio after a window was
already advertised, grow sk_rcvbuf so the remaining live sender-visible
window still has matching hard receive-memory backing.

This repairs the live advertised window only. Retracted-window rescue is
handled separately in a later patch.

Signed-off-by: Wesley Atwell <atwellwea@gmail.com>
---
 net/ipv4/tcp_input.c | 26 ++++++++++++++++++++++++++
 1 file changed, 26 insertions(+)

diff --git a/net/ipv4/tcp_input.c b/net/ipv4/tcp_input.c
index 32256519a085..d76e4e4c0e57 100644
--- a/net/ipv4/tcp_input.c
+++ b/net/ipv4/tcp_input.c
@@ -221,6 +221,31 @@ static __cold void tcp_gro_dev_warn(const struct sock *sk, const struct sk_buff
 	rcu_read_unlock();
 }
 
+/* If scaling_ratio drops after we already advertised tp->rcv_wnd, grow
+ * sk_rcvbuf so the remaining live window still maps back to hard memory
+ * units under the old advertise-time basis.
+ */
+static void tcp_try_grow_advertised_window(struct sock *sk,
+					   const struct sk_buff *skb)
+{
+	struct tcp_sock *tp = tcp_sk(sk);
+	int needed;
+
+	/* Keep this repair aligned with tcp_rcvbuf_grow(): do not adjust
+	 * receive-buffer backing for not-yet-accepted or orphaned sockets.
+	 */
+	if (!tcp_rcvbuf_grow_allowed(sk))
+		return;
+
+	if (!tcp_receive_window(tp))
+		return;
+
+	if (!tcp_space_from_rcv_wnd(tp, tcp_receive_window(tp), &needed))
+		return;
+
+	tcp_try_grow_rcvbuf(sk, needed);
+}
+
 /* Adapt the MSS value used to make delayed ack decision to the
  * real world.
  */
@@ -251,6 +276,7 @@ static void tcp_measure_rcv_mss(struct sock *sk, const struct sk_buff *skb)
 			if (old_ratio != tcp_sk(sk)->scaling_ratio) {
 				struct tcp_sock *tp = tcp_sk(sk);
 
+				tcp_try_grow_advertised_window(sk, skb);
 				val = tcp_win_from_space(sk, sk->sk_rcvbuf);
 				tcp_set_window_clamp(sk, val);
 
-- 
2.43.0

[PATCH net-next v2 07/14] tcp: honor the maximum advertised window after live retraction

[atwellwea]

From: Wesley Atwell <atwellwea@gmail.com>

If receive-side accounting retracts the live rwnd below a larger
sender-visible window that was already advertised, allow one in-order
skb within that historical bound to repair its backing and reach the
normal receive path.

Hard receive-memory admission is still enforced through the existing
prune and collapse path. The rescue only changes how data already
inside sender-visible sequence space is classified and backed.

Signed-off-by: Wesley Atwell <atwellwea@gmail.com>
---
 net/ipv4/tcp_input.c | 92 +++++++++++++++++++++++++++++++++++++++++---
 1 file changed, 86 insertions(+), 6 deletions(-)

diff --git a/net/ipv4/tcp_input.c b/net/ipv4/tcp_input.c
index d76e4e4c0e57..4b9309c37e99 100644
--- a/net/ipv4/tcp_input.c
+++ b/net/ipv4/tcp_input.c
@@ -5376,24 +5376,86 @@ 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. If later receive-side accounting retracts the live receive window
+ * below the maximum right edge we already advertised, allow one in-order skb
+ * which still fits inside that sender-visible bound to reach the normal
+ * receive queue path.
+ *
+ * Keep receive-memory admission itself on the legacy hard-cap path so prune
+ * and collapse behavior stay aligned with the established retracted-window
+ * handling.
+ */
+static bool tcp_skb_in_retracted_window(const struct tcp_sock *tp,
+					const struct sk_buff *skb)
+{
+	u32 live_end = tp->rcv_nxt + tcp_receive_window(tp);
+	u32 max_end = tp->rcv_nxt + tcp_max_receive_window(tp);
+
+	return after(max_end, live_end) &&
+	       after(TCP_SKB_CB(skb)->end_seq, live_end) &&
+	       !after(TCP_SKB_CB(skb)->end_seq, max_end);
+}
+
 static bool tcp_can_ingest(const struct sock *sk, const struct sk_buff *skb)
 {
-	unsigned int rmem = atomic_read(&sk->sk_rmem_alloc);
+	return tcp_rmem_used(sk) <= READ_ONCE(sk->sk_rcvbuf);
+}
+
+/* Caller already established that @skb extends into the retracted-but-still-
+ * valid sender-visible window. For in-order progress, regrow sk_rcvbuf before
+ * falling into prune/forced-mem handling.
+ *
+ * This path intentionally repairs backing for one in-order skb that is already
+ * within sender-visible sequence space, rather than treating it like ordinary
+ * receive-buffer autotuning.
+ *
+ * Keep this rescue bounded to the span accepted by this skb instead of the
+ * full historical tp->rcv_mwnd_seq. However, never grow below skb->truesize,
+ * because sk_rmem_schedule() still charges hard memory, not sender-visible
+ * window bytes.
+ */
+static void tcp_try_grow_retracted_skb(struct sock *sk,
+				       const struct sk_buff *skb)
+{
+	struct tcp_sock *tp = tcp_sk(sk);
+	int needed = skb->truesize;
+	int span_space;
+	u32 span_win;
+
+	if (TCP_SKB_CB(skb)->seq != tp->rcv_nxt)
+		return;
+
+	span_win = TCP_SKB_CB(skb)->end_seq - tp->rcv_nxt;
+	if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN)
+		span_win--;
+
+	if (tcp_space_from_rcv_mwnd(tp, span_win, &span_space))
+		needed = max_t(int, needed, span_space);
 
-	return rmem <= sk->sk_rcvbuf;
+	tcp_try_grow_rcvbuf(sk, needed);
 }
 
+/* Sender-visible window rescue does not relax hard receive-memory admission.
+ * If growth did not make room, fall back to the established prune/collapse
+ * path.
+ */
 static int tcp_try_rmem_schedule(struct sock *sk, const struct sk_buff *skb,
 				 unsigned int size)
 {
-	if (!tcp_can_ingest(sk, skb) ||
-	    !sk_rmem_schedule(sk, skb, size)) {
+	bool can_ingest = tcp_can_ingest(sk, skb);
+	bool scheduled = can_ingest && sk_rmem_schedule(sk, skb, size);
+
+	if (!scheduled) {
+		int pruned = tcp_prune_queue(sk, skb);
 
-		if (tcp_prune_queue(sk, skb) < 0)
+		if (pruned < 0)
 			return -1;
 
 		while (!sk_rmem_schedule(sk, skb, size)) {
-			if (!tcp_prune_ofo_queue(sk, skb))
+			bool pruned_ofo = tcp_prune_ofo_queue(sk, skb);
+
+			if (!pruned_ofo)
 				return -1;
 		}
 	}
@@ -5629,6 +5691,7 @@ void tcp_data_ready(struct sock *sk)
 static void tcp_data_queue(struct sock *sk, struct sk_buff *skb)
 {
 	struct tcp_sock *tp = tcp_sk(sk);
+	bool retracted;
 	enum skb_drop_reason reason;
 	bool fragstolen;
 	int eaten;
@@ -5647,6 +5710,7 @@ static void tcp_data_queue(struct sock *sk, struct sk_buff *skb)
 	}
 	tcp_cleanup_skb(skb);
 	__skb_pull(skb, tcp_hdr(skb)->doff * 4);
+	retracted = skb->len && tcp_skb_in_retracted_window(tp, skb);
 
 	reason = SKB_DROP_REASON_NOT_SPECIFIED;
 	tp->rx_opt.dsack = 0;
@@ -5667,6 +5731,9 @@ static void tcp_data_queue(struct sock *sk, struct sk_buff *skb)
 			    (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN))
 				goto queue_and_out;
 
+			if (retracted)
+				goto queue_and_out;
+
 			reason = SKB_DROP_REASON_TCP_ZEROWINDOW;
 			NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPZEROWINDOWDROP);
 			goto out_of_window;
@@ -5674,7 +5741,20 @@ static void tcp_data_queue(struct sock *sk, struct sk_buff *skb)
 
 		/* Ok. In sequence. In window. */
 queue_and_out:
+		if (unlikely(retracted))
+			tcp_try_grow_retracted_skb(sk, skb);
+
 		if (tcp_try_rmem_schedule(sk, skb, skb->truesize)) {
+			/* If the live rwnd collapsed to zero while rescuing an
+			 * skb that still fit in sender-visible sequence space,
+			 * report zero-window rather than generic proto-mem.
+			 */
+			if (unlikely(!tcp_receive_window(tp) && retracted)) {
+				reason = SKB_DROP_REASON_TCP_ZEROWINDOW;
+				NET_INC_STATS(sock_net(sk),
+					      LINUX_MIB_TCPZEROWINDOWDROP);
+				goto out_of_window;
+			}
 			/* TODO: maybe ratelimit these WIN 0 ACK ? */
 			inet_csk(sk)->icsk_ack.pending |=
 					(ICSK_ACK_NOMEM | ICSK_ACK_NOW);
-- 
2.43.0

[PATCH net-next v2 08/14] tcp: extend TCP_REPAIR_WINDOW for live and max-window snapshots

[atwellwea]

From: Wesley Atwell <atwellwea@gmail.com>

Extend TCP_REPAIR_WINDOW so repair and restore can round-trip both the
live rwnd snapshot and the remembered maximum sender-visible window.

Keep the ABI append-only by accepting the legacy and v1 prefix lengths on
both get and set, rebuilding any missing max-window state from the live
window when older userspace restores a socket.

Signed-off-by: Wesley Atwell <atwellwea@gmail.com>
---
 include/net/tcp.h        | 13 +++----
 include/uapi/linux/tcp.h |  8 +++++
 net/ipv4/tcp.c           | 73 ++++++++++++++++++++++++++++++++++++----
 3 files changed, 81 insertions(+), 13 deletions(-)

diff --git a/include/net/tcp.h b/include/net/tcp.h
index 5b479ad44f89..12e62fea2aaf 100644
--- a/include/net/tcp.h
+++ b/include/net/tcp.h
@@ -1766,13 +1766,14 @@ static inline bool tcp_space_from_wnd_snapshot(u8 scaling_ratio, int win,
 }
 
 /* Rebuild hard receive-memory units for data already covered by tp->rcv_wnd if
- * the advertise-time basis is known.
+ * 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)
 {
-	return tcp_space_from_wnd_snapshot(tp->rcv_wnd_scaling_ratio, win,
-					   space);
+	return tcp_space_from_wnd_snapshot(tp->rcv_wnd_scaling_ratio, win, space);
 }
 
 /* Same as tcp_space_from_rcv_wnd(), but for the remembered maximum
@@ -1800,9 +1801,9 @@ static inline void tcp_scaling_ratio_init(struct sock *sk)
 }
 
 /* tp->rcv_wnd is paired with the scaling_ratio that was in force when that
- * window was last advertised. Callers can leave a zero snapshot when the
- * advertise-time basis is unknown and refresh the pair on the next local
- * window update.
+ * 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)
diff --git a/include/uapi/linux/tcp.h b/include/uapi/linux/tcp.h
index 03772dd4d399..564a77f69130 100644
--- a/include/uapi/linux/tcp.h
+++ b/include/uapi/linux/tcp.h
@@ -152,6 +152,11 @@ struct tcp_repair_opt {
 	__u32	opt_val;
 };
 
+/* Append-only repair ABI.
+ * Older userspace may stop at rcv_wup or rcv_wnd_scaling_ratio.
+ * The kernel accepts those prefix lengths and rebuilds any missing
+ * receive-window snapshot state on restore.
+ */
 struct tcp_repair_window {
 	__u32	snd_wl1;
 	__u32	snd_wnd;
@@ -159,6 +164,9 @@ struct tcp_repair_window {
 
 	__u32	rcv_wnd;
 	__u32	rcv_wup;
+	__u32	rcv_wnd_scaling_ratio;  /* 0 means live-window basis unknown */
+	__u32	rcv_mwnd_seq;
+	__u32	rcv_mwnd_scaling_ratio; /* 0 means max-window basis unknown */
 };
 
 enum {
diff --git a/net/ipv4/tcp.c b/net/ipv4/tcp.c
index 66706dbb90f5..39a1265876ea 100644
--- a/net/ipv4/tcp.c
+++ b/net/ipv4/tcp.c
@@ -3533,17 +3533,31 @@ 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 inline int tcp_repair_window_v1_size(void)
+{
+	return offsetof(struct tcp_repair_window, rcv_mwnd_seq);
+}
+
 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 != tcp_repair_window_v1_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)
@@ -3559,9 +3573,47 @@ 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;
+		tcp_init_max_rcv_wnd_seq(tp);
+		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;
-	tp->rcv_mwnd_seq = opt.rcv_wup + opt.rcv_wnd;
+
+	if (len == tcp_repair_window_v1_size()) {
+		/* v1 repair can restore the live-window snapshot, but not a
+		 * retracted max-window snapshot. Rebuild it from the live pair
+		 * until a fresh local advertisement updates it again.
+		 */
+		tcp_init_max_rcv_wnd_seq(tp);
+		return 0;
+	}
+
+	if (opt.rcv_mwnd_scaling_ratio > U8_MAX)
+		return -EINVAL;
+
+	/* Userspace may repair sequence-space values after checkpoint without
+	 * also rebasing the remembered max advertised right edge. If the exact
+	 * snapshot no longer covers the restored live window, treat it like
+	 * v1 and rebuild the max-window side from the live pair.
+	 */
+	if (after(opt.rcv_wup + opt.rcv_wnd, opt.rcv_mwnd_seq)) {
+		tcp_init_max_rcv_wnd_seq(tp);
+		return 0;
+	}
+
+	tp->rcv_mwnd_seq = opt.rcv_mwnd_seq;
+	tp->rcv_mwnd_scaling_ratio = opt.rcv_mwnd_scaling_ratio;
 
 	return 0;
 }
@@ -4650,12 +4702,16 @@ 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))
+		/* Mirror the accepted set-side prefix lengths so checkpoint
+		 * tools can round-trip exactly the layout version they know.
+		 */
+		if (len != tcp_repair_window_legacy_size() &&
+		    len != tcp_repair_window_v1_size() && len != sizeof(opt))
 			return -EINVAL;
 
 		if (!tp->repair)
@@ -4666,6 +4722,9 @@ 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;
+		opt.rcv_mwnd_seq = tp->rcv_mwnd_seq;
+		opt.rcv_mwnd_scaling_ratio = tp->rcv_mwnd_scaling_ratio;
 
 		if (copy_to_sockptr(optval, &opt, len))
 			return -EFAULT;
-- 
2.43.0

[PATCH net-next v2 09/14] mptcp: refresh TCP receive-window snapshots on subflows

[atwellwea]

From: Wesley Atwell <atwellwea@gmail.com>

When MPTCP resynchronizes the per-subflow TCP shadow window from the
mptcp-level receive state, refresh the live rwnd snapshot and the
remembered maximum-window snapshot along with it.

That keeps subflow TCP bookkeeping aligned with the sender-visible
window state tracked in the core TCP patches.

Signed-off-by: Wesley Atwell <atwellwea@gmail.com>
---
 net/mptcp/options.c  | 14 +++++++++-----
 net/mptcp/protocol.h | 14 +++++++++++---
 2 files changed, 20 insertions(+), 8 deletions(-)

diff --git a/net/mptcp/options.c b/net/mptcp/options.c
index 8a1c5698983c..64cd637484a4 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));
 	tcp_update_max_rcv_wnd_seq(tp);
 	subflow->rcv_wnd_sent = mptcp_rcv_wnd;
 }
@@ -1335,12 +1338,13 @@ 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;
-		new_win = min_t(u64, win, U32_MAX);
-		tp->rcv_wnd = new_win;
+		tcp_set_rcv_wnd(tp, min_t(u64, win, U32_MAX));
+		new_win = tp->rcv_wnd;
 		tcp_update_max_rcv_wnd_seq(tp);
 
 		/* 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)
-- 
2.43.0

[PATCH net-next v2 10/14] tcp: expose rmem and backlog in tcp and mptcp rcvbuf_grow tracepoints

[atwellwea]

From: Wesley Atwell <atwellwea@gmail.com>

Extend the tcp_rcvbuf_grow and mptcp_rcvbuf_grow tracepoints with the
live receive-memory allocation and backlog occupancy that now drive the
window-growth decisions in this series.

That makes it easier to inspect sender-visible rwnd state against the
actual hard receive-memory inputs.

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),
-- 
2.43.0

[PATCH net-next v2 11/14] selftests: tcp_ao: cover legacy, v1, and retracted repair windows

[atwellwea]

From: Wesley Atwell <atwellwea@gmail.com>

Extend the tcp_ao repair selftests to exercise the legacy, v1, and
current TCP_REPAIR_WINDOW layouts, plus a synthesized retracted-window
image that preserves a larger historical right edge.

These tests validate both the append-only ABI contract and the restore-
time rebuilding of any snapshot state older userspace could not save.

Signed-off-by: Wesley Atwell <atwellwea@gmail.com>
---
 .../testing/selftests/net/tcp_ao/lib/aolib.h  |  83 +++++++-
 .../testing/selftests/net/tcp_ao/lib/repair.c |  18 +-
 .../selftests/net/tcp_ao/self-connect.c       | 201 +++++++++++++++++-
 3 files changed, 279 insertions(+), 23 deletions(-)

diff --git a/tools/testing/selftests/net/tcp_ao/lib/aolib.h b/tools/testing/selftests/net/tcp_ao/lib/aolib.h
index ebb2899c12fe..ef08db831457 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 +672,55 @@ 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_v1_size(void)
+{
+	return offsetof(struct tcp_repair_window, rcv_mwnd_seq);
+}
+
+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));
+}
+
+static inline void test_sock_checkpoint_v1(int sk,
+					   struct tcp_sock_state *state,
+					   sockaddr_af *saddr)
+{
+	__test_sock_checkpoint_opt(sk, state, test_tcp_repair_window_v1_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 +729,37 @@ 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));
+}
+
+static inline void test_sock_restore_v1(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_v1_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..a7c0f2edd351 100644
--- a/tools/testing/selftests/net/tcp_ao/self-connect.c
+++ b/tools/testing/selftests/net/tcp_ao/self-connect.c
@@ -4,6 +4,14 @@
 #include "aolib.h"
 
 static union tcp_addr local_addr;
+static bool checked_repair_window_lens;
+
+enum repair_window_mode {
+	REPAIR_WINDOW_CURRENT,
+	REPAIR_WINDOW_LEGACY,
+	REPAIR_WINDOW_V1,
+	REPAIR_WINDOW_RETRACTED,
+};
 
 static void __setup_lo_intf(const char *lo_intf,
 			    const char *addr_str, uint8_t prefix)
@@ -30,8 +38,157 @@ static void setup_lo_intf(const char *lo_intf)
 #endif
 }
 
+/* The repair ABI accepts the legacy, v1, and current 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 v1_len = test_tcp_repair_window_v1_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 = v1_len;
+	ret = getsockopt(sk, SOL_TCP, TCP_REPAIR_WINDOW, &trw, &len);
+	if (ret || len != v1_len)
+		test_fail("repair-window get accepts v1 len");
+	else
+		test_ok("repair-window get accepts v1 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");
+
+	ret = setsockopt(sk, SOL_TCP, TCP_REPAIR_WINDOW, &trw, v1_len + 1);
+	if (ret == 0 || errno != EINVAL)
+		test_fail("repair-window set rejects invalid v1+1 len");
+	else
+		test_ok("repair-window set rejects invalid v1+1 len");
+}
+
+static void test_retracted_repair_window_state(int sk,
+					       struct tcp_sock_state *img)
+{
+	struct tcp_repair_window trw = {};
+	socklen_t len = sizeof(trw);
+	int ret;
+
+	ret = getsockopt(sk, SOL_TCP, TCP_REPAIR_WINDOW, &trw, &len);
+	if (ret || len != sizeof(trw))
+		test_error("getsockopt(TCP_REPAIR_WINDOW): %d", (int)len);
+
+	if (trw.rcv_mwnd_seq != img->trw.rcv_mwnd_seq ||
+	    trw.rcv_mwnd_scaling_ratio != img->trw.rcv_mwnd_scaling_ratio ||
+	    trw.rcv_wnd != img->trw.rcv_wnd ||
+	    trw.rcv_wup != img->trw.rcv_wup ||
+	    trw.rcv_wnd_scaling_ratio != img->trw.rcv_wnd_scaling_ratio)
+		test_fail("repair-window restore preserves retracted state");
+	else
+		test_ok("repair-window restore preserves retracted state");
+}
+
+static void test_v1_repair_window_state(int sk, struct tcp_sock_state *img)
+{
+	struct tcp_repair_window trw = {};
+	socklen_t len = sizeof(trw);
+	__u32 max_right = img->trw.rcv_wup + img->trw.rcv_wnd;
+	int ret;
+
+	ret = getsockopt(sk, SOL_TCP, TCP_REPAIR_WINDOW, &trw, &len);
+	if (ret || len != sizeof(trw))
+		test_error("getsockopt(TCP_REPAIR_WINDOW): %d", (int)len);
+
+	if (trw.rcv_mwnd_seq != max_right ||
+	    trw.rcv_mwnd_scaling_ratio != img->trw.rcv_wnd_scaling_ratio ||
+	    trw.rcv_wnd != img->trw.rcv_wnd ||
+	    trw.rcv_wup != img->trw.rcv_wup ||
+	    trw.rcv_wnd_scaling_ratio != img->trw.rcv_wnd_scaling_ratio)
+		test_fail("repair-window v1 restore rebuilds max-window state");
+	else
+		test_ok("repair-window v1 restore rebuilds max-window state");
+}
+
+/* Synthesize a repair image whose live rwnd was retracted after a larger
+ * right edge had already been advertised, so restore testing can validate
+ * snapshot preservation without depending on the live receive path.
+ */
+static bool make_retracted_repair_window_state(struct tcp_sock_state *img)
+{
+	__u32 gran = 1U << img->info.tcpi_rcv_wscale;
+	__u32 max_right;
+	__u32 shrink;
+
+	if (!(img->info.tcpi_options & TCPI_OPT_WSCALE))
+		return false;
+
+	max_right = img->trw.rcv_wup + img->trw.rcv_wnd;
+	shrink = img->trw.rcv_wnd / 4;
+	if (shrink < gran)
+		shrink = gran;
+	if (shrink >= img->trw.rcv_wnd)
+		shrink = img->trw.rcv_wnd >> 1;
+	if (shrink == 0 || shrink >= img->trw.rcv_wnd)
+		return false;
+
+	img->trw.rcv_wnd -= shrink;
+	img->trw.rcv_mwnd_seq = max_right;
+	img->trw.rcv_mwnd_scaling_ratio = img->trw.rcv_wnd_scaling_ratio;
+	return true;
+}
+
+static socklen_t repair_window_len(enum repair_window_mode mode)
+{
+	switch (mode) {
+	case REPAIR_WINDOW_LEGACY:
+		return test_tcp_repair_window_legacy_size();
+	case REPAIR_WINDOW_V1:
+		return test_tcp_repair_window_v1_size();
+	case REPAIR_WINDOW_CURRENT:
+	case REPAIR_WINDOW_RETRACTED:
+		return test_tcp_repair_window_exact_size();
+	}
+
+	return test_tcp_repair_window_exact_size();
+}
+
+static void test_sock_checkpoint_mode(enum repair_window_mode mode, int sk,
+				      struct tcp_sock_state *img,
+				      sockaddr_af *addr)
+{
+	switch (mode) {
+	case REPAIR_WINDOW_LEGACY:
+		test_sock_checkpoint_legacy(sk, img, addr);
+		break;
+	case REPAIR_WINDOW_V1:
+		test_sock_checkpoint_v1(sk, img, addr);
+		break;
+	case REPAIR_WINDOW_CURRENT:
+	case REPAIR_WINDOW_RETRACTED:
+		test_sock_checkpoint(sk, img, addr);
+		break;
+	}
+}
+
 static void tcp_self_connect(const char *tst, unsigned int port,
-			     bool different_keyids, bool check_restore)
+			     bool different_keyids, bool check_restore,
+			     enum repair_window_mode repair_window_mode)
 {
 	struct tcp_counters before, after;
 	uint64_t before_aogood, after_aogood;
@@ -109,7 +266,16 @@ 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);
+	test_sock_checkpoint_mode(repair_window_mode, sk, &img, &addr);
+	if (repair_window_mode == REPAIR_WINDOW_RETRACTED &&
+	    !make_retracted_repair_window_state(&img)) {
+		test_sock_state_free(&img);
+		netstat_free(ns_before);
+		close(sk);
+		test_skip("%s: no scaled repair window to retract", tst);
+		return;
+	}
 #ifdef IPV6_TEST
 	addr.sin6_port = htons(port + 1);
 #else
@@ -123,7 +289,9 @@ 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,
+				repair_window_len(repair_window_mode),
+				&addr, &addr, sizeof(addr));
 	if (different_keyids) {
 		if (test_add_repaired_key(sk, DEFAULT_TEST_PASSWORD, 0,
 					  local_addr, -1, 7, 5))
@@ -137,6 +305,10 @@ static void tcp_self_connect(const char *tst, unsigned int port,
 			test_error("setsockopt(TCP_AO_ADD_KEY)");
 	}
 	test_ao_restore(sk, &ao_img);
+	if (repair_window_mode == REPAIR_WINDOW_V1)
+		test_v1_repair_window_state(sk, &img);
+	if (repair_window_mode == REPAIR_WINDOW_RETRACTED)
+		test_retracted_repair_window_state(sk, &img);
 	test_disable_repair(sk);
 	test_sock_state_free(&img);
 	if (test_client_verify(sk, 100, nr_packets)) {
@@ -165,20 +337,33 @@ 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,
+			 REPAIR_WINDOW_CURRENT);
 
 	/* 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,
+			 REPAIR_WINDOW_CURRENT);
+	tcp_self_connect("self-connect(restore)", port, false, true,
+			 REPAIR_WINDOW_CURRENT);
+	port += 2; /* restore test restores over different port */
+	tcp_self_connect("self-connect(restore, legacy repair window)", port,
+			 false, true, REPAIR_WINDOW_LEGACY);
+	port += 2; /* restore test restores over different port */
+	tcp_self_connect("self-connect(restore, v1 repair window)", port,
+			 false, true, REPAIR_WINDOW_V1);
+	port += 2; /* restore test restores over different port */
+	tcp_self_connect("self-connect(restore, retracted repair window)", port,
+			 false, true, REPAIR_WINDOW_RETRACTED);
 	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, REPAIR_WINDOW_CURRENT);
 	port += 2; /* restore test restores over different port */
 
 	return NULL;
@@ -186,6 +371,6 @@ static void *client_fn(void *arg)
 
 int main(int argc, char *argv[])
 {
-	test_init(5, client_fn, NULL);
+	test_init(14, client_fn, NULL);
 	return 0;
 }
-- 
2.43.0

[PATCH net-next v2 12/14] tun/selftests: add RX truesize injection for TCP window tests

[atwellwea]

From: Wesley Atwell <atwellwea@gmail.com>

Add a test-only TUN ioctl that inflates RX skb->truesize, plus the
packetdrill-side helper needed to drive that ioctl through packetdrill's
own TUN queue file descriptor.

Use that plumbing to cover the receive-window regressions where
scaling_ratio drifts after advertisement, alongside the baseline too-big
packetdrill cases that exercise the same sender-visible rwnd accounting
from the non-injected path.

Signed-off-by: Wesley Atwell <atwellwea@gmail.com>
---
 drivers/net/tun.c                             |  65 ++++++++
 include/uapi/linux/if_tun.h                   |   4 +
 .../tcp_rcv_neg_window_truesize.pkt           | 143 ++++++++++++++++++
 .../net/packetdrill/tcp_rcv_toobig.pkt        |  35 +++++
 .../packetdrill/tcp_rcv_toobig_default.pkt    |  97 ++++++++++++
 .../tcp_rcv_toobig_default_truesize.pkt       | 118 +++++++++++++++
 .../tcp_rcv_wnd_shrink_allowed_truesize.pkt   |  49 ++++++
 tools/testing/selftests/net/tun.c             | 140 ++++++++++++++++-
 8 files changed, 650 insertions(+), 1 deletion(-)
 create mode 100644 tools/testing/selftests/net/packetdrill/tcp_rcv_neg_window_truesize.pkt
 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
 create mode 100644 tools/testing/selftests/net/packetdrill/tcp_rcv_toobig_default_truesize.pkt
 create mode 100644 tools/testing/selftests/net/packetdrill/tcp_rcv_wnd_shrink_allowed_truesize.pkt

diff --git a/drivers/net/tun.c b/drivers/net/tun.c
index c492fda6fc15..2cef62cebe88 100644
--- a/drivers/net/tun.c
+++ b/drivers/net/tun.c
@@ -53,6 +53,7 @@
 #include <linux/if_ether.h>
 #include <linux/if_tun.h>
 #include <linux/if_vlan.h>
+#include <linux/overflow.h>
 #include <linux/crc32.h>
 #include <linux/math.h>
 #include <linux/nsproxy.h>
@@ -85,8 +86,13 @@
 
 #include "tun_vnet.h"
 
+struct tun_file;
+
+#define TUNSETTRUESIZE_OLD _IOW('T', 228, unsigned int)
+
 static void tun_default_link_ksettings(struct net_device *dev,
 				       struct ethtool_link_ksettings *cmd);
+static void tun_rx_update_truesize(struct tun_file *tfile, struct sk_buff *skb);
 
 #define TUN_RX_PAD (NET_IP_ALIGN + NET_SKB_PAD)
 
@@ -138,6 +144,7 @@ struct tun_file {
 		u16 queue_index;
 		unsigned int ifindex;
 	};
+	u32 rx_extra_truesize;
 	struct napi_struct napi;
 	bool napi_enabled;
 	bool napi_frags_enabled;
@@ -1817,6 +1824,7 @@ static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
 		goto free_skb;
 	}
 
+	tun_rx_update_truesize(tfile, skb);
 	switch (tun->flags & TUN_TYPE_MASK) {
 	case IFF_TUN:
 		if (tun->flags & IFF_NO_PI) {
@@ -2373,6 +2381,25 @@ static void tun_put_page(struct tun_page *tpage)
 		__page_frag_cache_drain(tpage->page, tpage->count);
 }
 
+/* Tests can inflate skb->truesize on ingress to exercise receive-memory
+ * accounting against a scaling_ratio that drifts after a window was
+ * advertised. The knob is per queue file, defaults to zero, and only changes
+ * behavior when explicitly enabled through the TUN fd.
+ */
+static void tun_rx_update_truesize(struct tun_file *tfile, struct sk_buff *skb)
+{
+	u32 extra = READ_ONCE(tfile->rx_extra_truesize);
+	unsigned int truesize;
+
+	if (!extra)
+		return;
+
+	if (check_add_overflow(skb->truesize, extra, &truesize))
+		truesize = UINT_MAX;
+
+	skb->truesize = truesize;
+}
+
 static int tun_xdp_one(struct tun_struct *tun,
 		       struct tun_file *tfile,
 		       struct xdp_buff *xdp, int *flush,
@@ -2459,6 +2486,7 @@ static int tun_xdp_one(struct tun_struct *tun,
 		goto out;
 	}
 
+	tun_rx_update_truesize(tfile, skb);
 	skb->protocol = eth_type_trans(skb, tun->dev);
 	skb_reset_network_header(skb);
 	skb_probe_transport_header(skb);
@@ -3045,6 +3073,7 @@ static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
 	struct tun_struct *tun;
 	void __user* argp = (void __user*)arg;
 	unsigned int carrier;
+	unsigned int extra_truesize;
 	struct ifreq ifr;
 	kuid_t owner;
 	kgid_t group;
@@ -3309,6 +3338,40 @@ static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
 		ret = tun_net_change_carrier(tun->dev, (bool)carrier);
 		break;
 
+	/* Support both the legacy pointer-payload form and the scalar form
+	 * used by the selftest helper when injecting truesize from
+	 * packetdrill shell commands.
+	 */
+	case TUNSETTRUESIZE:
+	case TUNSETTRUESIZE_OLD:
+		ret = -EPERM;
+		if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
+			goto unlock;
+
+		if (cmd == TUNSETTRUESIZE_OLD) {
+			ret = -EFAULT;
+			if (copy_from_user(&extra_truesize, argp,
+					   sizeof(extra_truesize))) {
+				ret = -EINVAL;
+				if (arg > U32_MAX)
+					goto unlock;
+
+				extra_truesize = arg;
+			}
+		} else {
+			ret = -EINVAL;
+			if (arg > U32_MAX)
+				goto unlock;
+
+			extra_truesize = arg;
+		}
+
+		WRITE_ONCE(tfile->rx_extra_truesize, extra_truesize);
+		netif_info(tun, drv, tun->dev,
+			   "rx extra truesize set to %u\n", extra_truesize);
+		ret = 0;
+		break;
+
 	case TUNGETDEVNETNS:
 		ret = -EPERM;
 		if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
@@ -3348,6 +3411,7 @@ static long tun_chr_compat_ioctl(struct file *file,
 	case TUNGETSNDBUF:
 	case TUNSETSNDBUF:
 	case SIOCGIFHWADDR:
+	case TUNSETTRUESIZE_OLD:
 	case SIOCSIFHWADDR:
 		arg = (unsigned long)compat_ptr(arg);
 		break;
@@ -3408,6 +3472,7 @@ static int tun_chr_open(struct inode *inode, struct file * file)
 	RCU_INIT_POINTER(tfile->tun, NULL);
 	tfile->flags = 0;
 	tfile->ifindex = 0;
+	tfile->rx_extra_truesize = 0;
 
 	init_waitqueue_head(&tfile->socket.wq.wait);
 
diff --git a/include/uapi/linux/if_tun.h b/include/uapi/linux/if_tun.h
index 79d53c7a1ebd..4be63efe6540 100644
--- a/include/uapi/linux/if_tun.h
+++ b/include/uapi/linux/if_tun.h
@@ -61,6 +61,10 @@
 #define TUNSETFILTEREBPF _IOR('T', 225, int)
 #define TUNSETCARRIER _IOW('T', 226, int)
 #define TUNGETDEVNETNS _IO('T', 227)
+/* Test-only: add scalar bytes to skb->truesize on RX after TUN allocates
+ * an skb.
+ */
+#define TUNSETTRUESIZE _IO('T', 228)
 
 /* TUNSETIFF ifr flags */
 #define IFF_TUN		0x0001
diff --git a/tools/testing/selftests/net/packetdrill/tcp_rcv_neg_window_truesize.pkt b/tools/testing/selftests/net/packetdrill/tcp_rcv_neg_window_truesize.pkt
new file mode 100644
index 000000000000..1c5550fff509
--- /dev/null
+++ b/tools/testing/selftests/net/packetdrill/tcp_rcv_neg_window_truesize.pkt
@@ -0,0 +1,143 @@
+// SPDX-License-Identifier: GPL-2.0
+// Run the negative-window / max-advertised-window regression with inflated
+// TUN skb->truesize so scaling_ratio drifts throughout the flow. The sequence
+// checks and drop counters should remain identical to the uninflated case.
+
+--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, [1000000], 4) = 0
+   +0 bind(3, ..., ...) = 0
+   +0 listen(3, 1) = 0
+
+   +0 < S 0:0(0) win 32792 <mss 1000,nop,nop,sackOK,nop,wscale 7>
+   +0 > S. 0:0(0) ack 1 win 65535 <mss 1460,nop,nop,sackOK,nop,wscale 4>
+   +0 < . 1:1(0) ack 1 win 257
+
+   +0 accept(3, ..., ...) = 4
+
+// Put 1040000 bytes into the receive buffer.
+   +0 < P. 1:65001(65000) ack 1 win 257
+    * > .  1:1(0) ack 65001
+   +0 < P. 65001:130001(65000) ack 1 win 257
+    * > .  1:1(0) ack 130001
+   +0 < P. 130001:195001(65000) ack 1 win 257
+    * > .  1:1(0) ack 195001
+   +0 < P. 195001:260001(65000) ack 1 win 257
+    * > .  1:1(0) ack 260001
+   +0 < P. 260001:325001(65000) ack 1 win 257
+    * > .  1:1(0) ack 325001
+   +0 < P. 325001:390001(65000) ack 1 win 257
+    * > .  1:1(0) ack 390001
+   +0 < P. 390001:455001(65000) ack 1 win 257
+    * > .  1:1(0) ack 455001
+   +0 < P. 455001:520001(65000) ack 1 win 257
+    * > .  1:1(0) ack 520001
+   +0 < P. 520001:585001(65000) ack 1 win 257
+    * > .  1:1(0) ack 585001
+   +0 < P. 585001:650001(65000) ack 1 win 257
+    * > .  1:1(0) ack 650001
+   +0 < P. 650001:715001(65000) ack 1 win 257
+    * > .  1:1(0) ack 715001
+   +0 < P. 715001:780001(65000) ack 1 win 257
+    * > .  1:1(0) ack 780001
+   +0 < P. 780001:845001(65000) ack 1 win 257
+    * > .  1:1(0) ack 845001
+   +0 < P. 845001:910001(65000) ack 1 win 257
+    * > .  1:1(0) ack 910001
+   +0 < P. 910001:975001(65000) ack 1 win 257
+    * > .  1:1(0) ack 975001
+   +0 < P. 975001:1040001(65000) ack 1 win 257
+    * > .  1:1(0) ack 1040001
+
+// Start inflating future TUN skbs only after the baseline sender-visible
+// window has been established, so the negative-window checks below exercise
+// ratio drift without changing the initial max advertised window.
+   +0 `../tun --set-rx-truesize tun0 65536`
+
+// Trigger an extreme memory squeeze by shrinking SO_RCVBUF.
+   +0 setsockopt(4, SOL_SOCKET, SO_RCVBUF, [16000], 4) = 0
+
+   +0 < P. 1040001:1105001(65000) ack 1 win 257
+    * > .  1:1(0) ack 1040001 win 0
+// Check LINUX_MIB_TCPRCVQDROP has been incremented.
+   +0 `nstat -s | grep TcpExtTCPRcvQDrop | grep -q " 1 "`
+
+// RWIN == 0: rcv_wup = 1040001, rcv_wnd = 0, rcv_mwnd_seq > 1105001.
+
+// Accept pure ack with seq in max adv. window.
+   +0 write(4, ..., 1000) = 1000
+   +0 > P. 1:1001(1000) ack 1040001 win 0
+   +0 < .  1105001:1105001(0) ack 1001 win 257
+
+// In order segment, in max adv. window -> drop (SKB_DROP_REASON_TCP_ZEROWINDOW).
+   +0 < P. 1040001:1041001(1000) ack 1001 win 257
+   +0 > .  1001:1001(0) ack 1040001 win 0
+// Ooo partial segment, in max adv. window -> drop (SKB_DROP_REASON_TCP_ZEROWINDOW).
+   +0 < P. 1039001:1041001(2000) ack 1001 win 257
+   +0 > .  1001:1001(0) ack 1040001 win 0 <nop,nop,sack 1039001:1040001>
+// Check LINUX_MIB_TCPZEROWINDOWDROP has been incremented twice.
+   +0 `nstat -s | grep TcpExtTCPZeroWindowDrop | grep -q " 2 "`
+
+// Ooo segment, in max adv. window -> drop (SKB_DROP_REASON_TCP_OVERWINDOW).
+   +0 < P. 1105001:1106001(1000) ack 1001 win 257
+   +0 > .  1001:1001(0) ack 1040001 win 0
+// Ooo segment, beyond max adv. window -> drop (SKB_DROP_REASON_TCP_INVALID_SEQUENCE).
+   +0 < P. 2000001:2001001(1000) ack 1001 win 257
+   +0 > .  1001:1001(0) ack 1040001 win 0
+// Check LINUX_MIB_BEYOND_WINDOW has been incremented twice.
+   +0 `nstat -s | grep TcpExtBeyondWindow | grep -q " 2 "`
+
+// Read all data.
+   +0 read(4, ..., 2000000) = 1040000
+    * > .  1001:1001(0) ack 1040001
+
+// RWIN > 0: rcv_wup = 1040001, 0 < rcv_wnd < 32000, rcv_mwnd_seq > 1105001.
+
+// Accept pure ack with seq in max adv. window, beyond adv. window.
+   +0 write(4, ..., 1000) = 1000
+   +0 > P.  1001:2001(1000) ack 1040001
+   +0 < . 1105001:1105001(0) ack 2001 win 257
+
+// In order segment, in max adv. window, in adv. window -> accept.
+   +0 < P. 1040001:1041001(1000) ack 2001 win 257
+    * > .  2001:2001(0) ack 1041001
+
+// Ooo partial segment, in adv. window -> accept.
+   +0 < P. 1040001:1042001(2000) ack 2001 win 257
+    * > .  2001:2001(0) ack 1042001 <nop,nop,sack 1040001:1041001>
+
+// Ooo segment, in max adv. window, beyond adv. window -> drop.
+   +0 < P. 1105001:1106001(1000) ack 2001 win 257
+   +0 > .  2001:2001(0) ack 1042001
+// Ooo segment, beyond max adv. window, beyond adv. window -> drop.
+   +0 < P. 2000001:2001001(1000) ack 2001 win 257
+   +0 > .  2001:2001(0) ack 1042001
+// Check LINUX_MIB_BEYOND_WINDOW has been incremented twice more.
+   +0 `nstat -s | grep TcpExtBeyondWindow | grep -q " 4 "`
+
+// We are allowed to go beyond the window and buffer with one packet.
+   +0 < P. 1042001:1062001(20000) ack 2001 win 257
+    * > .  2001:2001(0) ack 1062001
+   +0 < P. 1062001:1082001(20000) ack 2001 win 257
+    * > .  2001:2001(0) ack 1082001 win 0
+
+// But not more: in-order segment, in max adv. window -> drop.
+   +0 < P. 1082001:1083001(1000) ack 2001 win 257
+    * > .  2001:2001(0) ack 1082001
+// Check LINUX_MIB_TCPZEROWINDOWDROP has been incremented again.
+   +0 `nstat -s | grep TcpExtTCPZeroWindowDrop | grep -q " 3 "`
+
+// Another ratio drop must not change the final zero-window decision.
+   +0 `../tun --set-rx-truesize tun0 131072`
+
+   +0 < P. 1082001:1083001(1000) ack 2001 win 257
+    * > .  2001:2001(0) ack 1082001
+// Check LINUX_MIB_TCPZEROWINDOWDROP has been incremented once more.
+   +0 `nstat -s | grep TcpExtTCPZeroWindowDrop | grep -q " 4 "`
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..837ba3633752
--- /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
+    * > .  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/packetdrill/tcp_rcv_toobig_default_truesize.pkt b/tools/testing/selftests/net/packetdrill/tcp_rcv_toobig_default_truesize.pkt
new file mode 100644
index 000000000000..c2ebe11d75f7
--- /dev/null
+++ b/tools/testing/selftests/net/packetdrill/tcp_rcv_toobig_default_truesize.pkt
@@ -0,0 +1,118 @@
+// 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. The warmup traffic
+// keeps the socket in the normal data path without changing its default
+// sk_rcvbuf. Then inflate skb->truesize on future TUN RX packets so the live
+// scaling_ratio drops after we already exposed a larger rwnd to the peer.
+// The follow-up packet should still be admitted, and tcp_clamp_window() should
+// grow sk_rcvbuf to honor the sender-visible window instead of dropping data.
+   +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 initial 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
+
+   +0 %{ base_rcvbuf = SK_MEMINFO_RCVBUF }%
+
+// Leave about 60kB queued, then make future TUN skbs look more expensive in
+// two steps. Both inflated skbs still fit the already-advertised window and
+// must be admitted, and sk_rcvbuf should keep growing as the live
+// scaling_ratio drops further.
+   +0 < P. 1040001:1102001(62000) ack 1 win 257
+   * > .  1:1(0) ack 1102001
+
+   +0 `../tun --set-rx-truesize tun0 4096`
+
+   +0 < P. 1102001:1167001(65000) ack 1 win 257
+   * > .  1:1(0) ack 1167001
+   +0 %{ assert SK_MEMINFO_RCVBUF > base_rcvbuf, (base_rcvbuf, SK_MEMINFO_RCVBUF) }%
+   +0 %{ small_rcvbuf = SK_MEMINFO_RCVBUF }%
+
+   +0 < P. 1167001:1229001(62000) ack 1 win 257
+   * > .  1:1(0) ack 1229001
+
+   +0 `../tun --set-rx-truesize tun0 65536`
+
+   +0 < P. 1229001:1294001(65000) ack 1 win 257
+   * > .  1:1(0) ack 1294001
+   +0 %{ assert SK_MEMINFO_RCVBUF > small_rcvbuf, (base_rcvbuf, small_rcvbuf, SK_MEMINFO_RCVBUF) }%
+
+   +0 < P. 1294001:1356001(62000) ack 1 win 257
+   * > .  1:1(0) ack 1356001
+   +0 read(4, ..., 254000) = 254000
diff --git a/tools/testing/selftests/net/packetdrill/tcp_rcv_wnd_shrink_allowed_truesize.pkt b/tools/testing/selftests/net/packetdrill/tcp_rcv_wnd_shrink_allowed_truesize.pkt
new file mode 100644
index 000000000000..08da5fddaa12
--- /dev/null
+++ b/tools/testing/selftests/net/packetdrill/tcp_rcv_wnd_shrink_allowed_truesize.pkt
@@ -0,0 +1,49 @@
+// SPDX-License-Identifier: GPL-2.0
+
+--mss=1000
+
+`./defaults.sh
+sysctl -q net.ipv4.tcp_shrink_window=1
+sysctl -q net.ipv4.tcp_rmem="4096 32768 $((32*1024*1024))"`
+
+   0 `nstat -n`
+
+// Establish a connection. After the first payload we know the peer has seen a
+// scaled receive window reaching sequence 25361. Inflate later TUN skbs in two
+// steps so the live scaling_ratio drops more than once, then verify that:
+//   1) a segment one byte beyond the max advertised window is still dropped,
+//   2) a segment exactly using the previously advertised max window is still
+//      accepted even though the current live ratio no longer matches that
+//      original advertisement basis.
+  +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 32792 <mss 1000,nop,wscale 7>
+  +0 > S. 0:0(0) ack 1 <mss 1460,nop,wscale 10>
+  +0 < . 1:1(0) ack 1 win 257
+
+  +0 accept(3, ..., ...) = 4
+
+  +0 < P. 1:10001(10000) ack 1 win 257
+   * > .  1:1(0) ack 10001 win 15
+
+// Max window seq advertised here is 10001 + 15*1024 = 25361.
+  +0 `../tun --set-rx-truesize tun0 4096`
+
+  +0 < P. 10001:11024(1023) ack 1 win 257
+   * > .  1:1(0) ack 11024
+
+  +0 `../tun --set-rx-truesize tun0 65536`
+
+// Segment beyond the max window stays invalid even after ratio drift.
+  +0 < P. 11024:25362(14338) ack 1 win 257
+   * > .  1:1(0) ack 11024
+
+// Segment exactly using the max window must still be accepted.
+  +0 < P. 11024:25361(14337) ack 1 win 257
+   * > .  1:1(0) ack 25361
+
+// Check LINUX_MIB_BEYOND_WINDOW has been incremented once.
+  +0 `nstat | grep TcpExtBeyondWindow | grep -q " 1 "`
diff --git a/tools/testing/selftests/net/tun.c b/tools/testing/selftests/net/tun.c
index cf106a49b55e..473992b3784d 100644
--- a/tools/testing/selftests/net/tun.c
+++ b/tools/testing/selftests/net/tun.c
@@ -2,14 +2,17 @@
 
 #define _GNU_SOURCE
 
+#include <dirent.h>
 #include <errno.h>
 #include <fcntl.h>
+#include <limits.h>
 #include <stdio.h>
 #include <stdlib.h>
 #include <string.h>
 #include <unistd.h>
 #include <linux/if_tun.h>
 #include <sys/ioctl.h>
+#include <sys/syscall.h>
 #include <sys/socket.h>
 
 #include "kselftest_harness.h"
@@ -174,6 +177,135 @@ static int tun_delete(char *dev)
 	return ip_link_del(dev);
 }
 
+static bool is_numeric_name(const char *name)
+{
+	for (; *name; name++) {
+		if (*name < '0' || *name > '9')
+			return false;
+	}
+
+	return true;
+}
+
+static int packetdrill_dup_fd(int pidfd, const char *fd_name)
+{
+	char *end;
+	unsigned long tmp;
+
+	errno = 0;
+	tmp = strtoul(fd_name, &end, 10);
+	if (errno || *end || tmp > INT_MAX) {
+		errno = EINVAL;
+		return -1;
+	}
+
+	return syscall(SYS_pidfd_getfd, pidfd, (int)tmp, 0);
+}
+
+static int open_packetdrill_tunfd(pid_t pid, const char *ifname)
+{
+	char fd_dir[PATH_MAX];
+	struct dirent *dent;
+	struct ifreq ifr = {};
+	int pidfd;
+	int saved_errno = ENOENT;
+	DIR *dir;
+
+	snprintf(fd_dir, sizeof(fd_dir), "/proc/%ld/fd", (long)pid);
+
+	pidfd = syscall(SYS_pidfd_open, pid, 0);
+	if (pidfd < 0)
+		return -1;
+
+	dir = opendir(fd_dir);
+	if (!dir) {
+		close(pidfd);
+		return -1;
+	}
+
+	while ((dent = readdir(dir))) {
+		int fd;
+
+		if (!is_numeric_name(dent->d_name))
+			continue;
+
+		/* Reopen via pidfd_getfd() so we duplicate packetdrill's attached
+		 * queue file, instead of opening a fresh /dev/net/tun instance.
+		 */
+		fd = packetdrill_dup_fd(pidfd, dent->d_name);
+		if (fd < 0) {
+			saved_errno = errno;
+			continue;
+		}
+
+		memset(&ifr, 0, sizeof(ifr));
+		if (!ioctl(fd, TUNGETIFF, &ifr) &&
+		    !strncmp(ifr.ifr_name, ifname, IFNAMSIZ)) {
+			close(pidfd);
+			closedir(dir);
+			return fd;
+		}
+
+		if (errno)
+			saved_errno = errno;
+		close(fd);
+	}
+
+	close(pidfd);
+	closedir(dir);
+	errno = saved_errno;
+	return -1;
+}
+
+/* Packetdrill owns the TUN queue fd, so drive the test ioctl through that
+ * exact file descriptor found under /proc/$PACKETDRILL_PID/fd.
+ */
+static int packetdrill_set_rx_truesize(const char *ifname, const char *value)
+{
+	char *packetdrill_pid, *end;
+	unsigned long long tmp;
+	unsigned int extra;
+	pid_t pid;
+	int fd;
+
+	packetdrill_pid = getenv("PACKETDRILL_PID");
+	if (!packetdrill_pid || !*packetdrill_pid) {
+		fprintf(stderr, "PACKETDRILL_PID is not set\n");
+		return 1;
+	}
+
+	errno = 0;
+	tmp = strtoull(packetdrill_pid, &end, 10);
+	if (errno || *end || !tmp || tmp > INT_MAX) {
+		fprintf(stderr, "invalid PACKETDRILL_PID: %s\n", packetdrill_pid);
+		return 1;
+	}
+	pid = (pid_t)tmp;
+
+	errno = 0;
+	tmp = strtoull(value, &end, 0);
+	if (errno || *end || tmp > UINT_MAX) {
+		fprintf(stderr, "invalid truesize value: %s\n", value);
+		return 1;
+	}
+	extra = (unsigned int)tmp;
+
+	fd = open_packetdrill_tunfd(pid, ifname);
+	if (fd < 0) {
+		perror("open_packetdrill_tunfd");
+		return 1;
+	}
+
+	if (ioctl(fd, TUNSETTRUESIZE, (unsigned long)extra)) {
+		perror("ioctl(TUNSETTRUESIZE)");
+		close(fd);
+		return 1;
+	}
+
+	close(fd);
+	return 0;
+}
+
 static int tun_open(char *dev, const int flags, const int hdrlen,
 		    const int features, const unsigned char *mac_addr)
 {
@@ -985,4 +1117,10 @@ XFAIL_ADD(tun_vnet_udptnl, 6in4_over_maxbytes, recv_gso_packet);
 XFAIL_ADD(tun_vnet_udptnl, 4in6_over_maxbytes, recv_gso_packet);
 XFAIL_ADD(tun_vnet_udptnl, 6in6_over_maxbytes, recv_gso_packet);
 
-TEST_HARNESS_MAIN
+int main(int argc, char **argv)
+{
+	if (argc == 4 && !strcmp(argv[1], "--set-rx-truesize"))
+		return packetdrill_set_rx_truesize(argv[2], argv[3]);
+
+	return test_harness_run(argc, argv);
+}
-- 
2.43.0

[PATCH net-next v2 13/14] netdevsim: add peer RX truesize support for selftests

[atwellwea]

From: Wesley Atwell <atwellwea@gmail.com>

Add a debugfs-controlled peer RX truesize knob to netdevsim, inflate the
forwarded skb only on the peer RX side, and cover the resulting socket
memory-accounting behavior with a dedicated selftest.

This keeps the synthetic cost out of the sender-side skb geometry while
giving the selftests a second runtime vehicle for the receive-memory
accounting exercised by the TCP rwnd work.

Signed-off-by: Wesley Atwell <atwellwea@gmail.com>
---
 drivers/net/netdevsim/netdev.c                | 145 +++++-
 drivers/net/netdevsim/netdevsim.h             |   4 +
 .../selftests/drivers/net/netdevsim/Makefile  |   1 +
 .../drivers/net/netdevsim/peer-rx-truesize.sh | 426 ++++++++++++++++++
 4 files changed, 575 insertions(+), 1 deletion(-)
 create mode 100755 tools/testing/selftests/drivers/net/netdevsim/peer-rx-truesize.sh

diff --git a/drivers/net/netdevsim/netdev.c b/drivers/net/netdevsim/netdev.c
index 5ec028a00c62..22238df79b6a 100644
--- a/drivers/net/netdevsim/netdev.c
+++ b/drivers/net/netdevsim/netdev.c
@@ -17,8 +17,10 @@
 #include <linux/etherdevice.h>
 #include <linux/ethtool_netlink.h>
 #include <linux/kernel.h>
+#include <linux/kstrtox.h>
 #include <linux/module.h>
 #include <linux/netdevice.h>
+#include <linux/refcount.h>
 #include <linux/slab.h>
 #include <net/netdev_queues.h>
 #include <net/netdev_rx_queue.h>
@@ -37,6 +39,91 @@ MODULE_IMPORT_NS("NETDEV_INTERNAL");
 
 #define NSIM_RING_SIZE		256
 
+struct nsim_rx_truesize {
+	refcount_t refs;
+	u32 value;
+};
+
+static struct nsim_rx_truesize *
+nsim_rx_truesize_get(struct nsim_rx_truesize *rx_truesize)
+{
+	if (!rx_truesize)
+		return NULL;
+
+	if (!refcount_inc_not_zero(&rx_truesize->refs))
+		return NULL;
+
+	return rx_truesize;
+}
+
+static void nsim_rx_truesize_put(struct nsim_rx_truesize *rx_truesize)
+{
+	if (!rx_truesize)
+		return;
+
+	if (refcount_dec_and_test(&rx_truesize->refs))
+		kfree(rx_truesize);
+}
+
+static ssize_t nsim_rx_truesize_read(struct file *file, char __user *user_buf,
+				     size_t count, loff_t *ppos)
+{
+	struct nsim_rx_truesize *rx_truesize = file->private_data;
+	char buf[24];
+	int len;
+
+	len = scnprintf(buf, sizeof(buf), "%u\n",
+			READ_ONCE(rx_truesize->value));
+
+	return simple_read_from_buffer(user_buf, count, ppos, buf, len);
+}
+
+static ssize_t nsim_rx_truesize_write(struct file *file,
+				      const char __user *user_buf,
+				      size_t count, loff_t *ppos)
+{
+	struct nsim_rx_truesize *rx_truesize = file->private_data;
+	u32 value;
+	int err;
+
+	err = kstrtou32_from_user(user_buf, count, 0, &value);
+	if (err)
+		return err;
+
+	WRITE_ONCE(rx_truesize->value, value);
+
+	return count;
+}
+
+static int nsim_rx_truesize_open(struct inode *inode, struct file *file)
+{
+	struct nsim_rx_truesize *rx_truesize;
+
+	rx_truesize = nsim_rx_truesize_get(inode->i_private);
+	if (!rx_truesize)
+		return -ENODEV;
+
+	file->private_data = rx_truesize;
+
+	return nonseekable_open(inode, file);
+}
+
+static int nsim_rx_truesize_release(struct inode *inode, struct file *file)
+{
+	nsim_rx_truesize_put(file->private_data);
+
+	return 0;
+}
+
+static const struct file_operations nsim_rx_truesize_fops = {
+	.owner		= THIS_MODULE,
+	.open		= nsim_rx_truesize_open,
+	.read		= nsim_rx_truesize_read,
+	.write		= nsim_rx_truesize_write,
+	.release	= nsim_rx_truesize_release,
+	.llseek		= noop_llseek,
+};
+
 static void nsim_start_peer_tx_queue(struct net_device *dev, struct nsim_rq *rq)
 {
 	struct netdevsim *ns = netdev_priv(dev);
@@ -117,6 +204,28 @@ static int nsim_forward_skb(struct net_device *tx_dev,
 	return nsim_napi_rx(tx_dev, rx_dev, rq, skb);
 }
 
+/* Tests can inflate peer RX skb->truesize to exercise receiver-side TCP
+ * accounting under scaling-ratio drift without perturbing sender-side skb
+ * ownership.
+ */
+static void nsim_rx_update_truesize(struct sk_buff *skb, u32 extra)
+{
+	unsigned int truesize;
+
+	if (!extra)
+		return;
+
+	if (check_add_overflow(skb->truesize, extra, &truesize))
+		truesize = UINT_MAX;
+
+	skb->truesize = truesize;
+}
+
+static u32 nsim_rx_extra_truesize(const struct netdevsim *ns)
+{
+	return READ_ONCE(ns->rx_truesize->value);
+}
+
 static netdev_tx_t nsim_start_xmit(struct sk_buff *skb, struct net_device *dev)
 {
 	struct netdevsim *ns = netdev_priv(dev);
@@ -125,7 +234,9 @@ static netdev_tx_t nsim_start_xmit(struct sk_buff *skb, struct net_device *dev)
 	unsigned int len = skb->len;
 	struct netdevsim *peer_ns;
 	struct netdev_config *cfg;
+	struct sk_buff *nskb;
 	struct nsim_rq *rq;
+	u32 extra;
 	int rxq;
 	int dr;
 
@@ -160,7 +271,24 @@ static netdev_tx_t nsim_start_xmit(struct sk_buff *skb, struct net_device *dev)
 	      cfg->hds_thresh > len)))
 		skb_linearize(skb);
 
+	extra = nsim_rx_extra_truesize(peer_ns);
 	skb_tx_timestamp(skb);
+	if (extra) {
+		/* Clone before inflating truesize so only the peer RX path sees
+		 * the synthetic cost; sender-side skb accounting stays put.
+		 */
+		nskb = skb_clone(skb, GFP_ATOMIC);
+		if (!nskb) {
+			if (psp_ext)
+				__skb_ext_put(psp_ext);
+			goto out_drop_free;
+		}
+
+		consume_skb(skb);
+		skb = nskb;
+		nsim_rx_update_truesize(skb, extra);
+	}
+
 	if (unlikely(nsim_forward_skb(dev, peer_dev,
 				      skb, rq, psp_ext) == NET_RX_DROP))
 		goto out_drop_cnt;
@@ -1121,6 +1249,7 @@ struct netdevsim *nsim_create(struct nsim_dev *nsim_dev,
 			      u8 perm_addr[ETH_ALEN])
 {
 	struct net_device *dev;
+	struct nsim_rx_truesize *rx_truesize;
 	struct netdevsim *ns;
 	int err;
 
@@ -1140,6 +1269,13 @@ struct netdevsim *nsim_create(struct nsim_dev *nsim_dev,
 	ns->nsim_bus_dev = nsim_dev->nsim_bus_dev;
 	SET_NETDEV_DEV(dev, &ns->nsim_bus_dev->dev);
 	SET_NETDEV_DEVLINK_PORT(dev, &nsim_dev_port->devlink_port);
+	rx_truesize = kzalloc_obj(*rx_truesize);
+	if (!rx_truesize) {
+		err = -ENOMEM;
+		goto err_free_netdev;
+	}
+	refcount_set(&rx_truesize->refs, 1);
+	ns->rx_truesize = rx_truesize;
 	nsim_ethtool_init(ns);
 	if (nsim_dev_port_is_pf(nsim_dev_port))
 		err = nsim_init_netdevsim(ns);
@@ -1153,21 +1289,27 @@ struct netdevsim *nsim_create(struct nsim_dev *nsim_dev,
 	ns->qr_dfs = debugfs_create_file("queue_reset", 0200,
 					 nsim_dev_port->ddir, ns,
 					 &nsim_qreset_fops);
+	ns->rx_truesize_dfs = debugfs_create_file("rx_extra_truesize", 0600,
+						  nsim_dev_port->ddir,
+						  ns->rx_truesize,
+						  &nsim_rx_truesize_fops);
 	return ns;
 
 err_free_netdev:
+	nsim_rx_truesize_put(ns->rx_truesize);
 	free_netdev(dev);
 	return ERR_PTR(err);
 }
 
 void nsim_destroy(struct netdevsim *ns)
 {
+	struct nsim_rx_truesize *rx_truesize = ns->rx_truesize;
 	struct net_device *dev = ns->netdev;
 	struct netdevsim *peer;
 
+	debugfs_remove(ns->rx_truesize_dfs);
 	debugfs_remove(ns->qr_dfs);
 	debugfs_remove(ns->pp_dfs);
-
 	if (ns->nb.notifier_call)
 		unregister_netdevice_notifier_dev_net(ns->netdev, &ns->nb,
 						      &ns->nn);
@@ -1198,6 +1340,7 @@ void nsim_destroy(struct netdevsim *ns)
 	}
 
 	free_netdev(dev);
+	nsim_rx_truesize_put(rx_truesize);
 }
 
 bool netdev_is_nsim(struct net_device *dev)
diff --git a/drivers/net/netdevsim/netdevsim.h b/drivers/net/netdevsim/netdevsim.h
index f767fc8a7505..972ad274060e 100644
--- a/drivers/net/netdevsim/netdevsim.h
+++ b/drivers/net/netdevsim/netdevsim.h
@@ -75,6 +75,8 @@ struct nsim_macsec {
 	u8 nsim_secy_count;
 };
 
+struct nsim_rx_truesize;
+
 struct nsim_ethtool_pauseparam {
 	bool rx;
 	bool tx;
@@ -144,6 +146,8 @@ struct netdevsim {
 	} udp_ports;
 
 	struct page *page;
+	struct nsim_rx_truesize *rx_truesize;
+	struct dentry *rx_truesize_dfs;
 	struct dentry *pp_dfs;
 	struct dentry *qr_dfs;
 
diff --git a/tools/testing/selftests/drivers/net/netdevsim/Makefile b/tools/testing/selftests/drivers/net/netdevsim/Makefile
index 1a228c5430f5..9e9e48d5913b 100644
--- a/tools/testing/selftests/drivers/net/netdevsim/Makefile
+++ b/tools/testing/selftests/drivers/net/netdevsim/Makefile
@@ -14,6 +14,7 @@ TEST_PROGS := \
 	macsec-offload.sh \
 	nexthop.sh \
 	peer.sh \
+	peer-rx-truesize.sh \
 	psample.sh \
 	tc-mq-visibility.sh \
 	udp_tunnel_nic.sh \
diff --git a/tools/testing/selftests/drivers/net/netdevsim/peer-rx-truesize.sh b/tools/testing/selftests/drivers/net/netdevsim/peer-rx-truesize.sh
new file mode 100755
index 000000000000..6d1101d20847
--- /dev/null
+++ b/tools/testing/selftests/drivers/net/netdevsim/peer-rx-truesize.sh
@@ -0,0 +1,426 @@
+#!/bin/bash
+# SPDX-License-Identifier: GPL-2.0-only
+
+set -euo pipefail
+
+lib_dir=$(dirname "$0")/../../../net
+source "$lib_dir"/lib.sh
+
+NSIM_SRV_ID=$((1024 + RANDOM % 1024))
+NSIM_CLI_ID=$((2048 + RANDOM % 1024))
+NSIM_SYS_LINK=/sys/bus/netdevsim/link_device
+SERVER_ADDR=192.0.2.1
+CLIENT_ADDR=192.0.2.2
+RMEM_PORT=12345
+WARM_PORT=12346
+RMEM_QUEUED_LEN=65000
+RMEM_INFLATED_LEN=65000
+RMEM_SMALL_EXTRA=4096
+RMEM_LARGE_EXTRA=65536
+WARM_WARMUP_ROUNDS=16
+WARM_WARMUP_LEN=65000
+WARM_QUEUED_LEN=62000
+WARM_INFLATED_LEN=65000
+WARM_EXTRA=65536
+
+srv_dev=
+cli_dev=
+srv_pid=
+cli_pid=
+srv_fd=
+cli_fd=
+stage_dir=
+CASE_BASE_METRIC=
+CASE_FINAL_METRIC=
+
+cleanup()
+{
+	local rc=$?
+
+	if [ -n "${srv_pid:-}" ]; then
+		kill "${srv_pid}" 2>/dev/null || true
+		wait "${srv_pid}" 2>/dev/null || true
+	fi
+
+	if [ -n "${cli_pid:-}" ]; then
+		kill "${cli_pid}" 2>/dev/null || true
+		wait "${cli_pid}" 2>/dev/null || true
+	fi
+
+	if [ -n "${srv_fd:-}" ]; then
+		eval "exec ${srv_fd}<&-"
+	fi
+
+	if [ -n "${cli_fd:-}" ]; then
+		eval "exec ${cli_fd}<&-"
+	fi
+
+	if [ -d "${stage_dir:-}" ]; then
+		rm -rf "${stage_dir}"
+	fi
+
+	cleanup_netdevsim "${NSIM_SRV_ID}" 2>/dev/null || true
+	cleanup_netdevsim "${NSIM_CLI_ID}" 2>/dev/null || true
+	cleanup_ns "${SRV:-}" "${CLI:-}" 2>/dev/null || true
+
+	exit "${rc}"
+}
+
+trap cleanup EXIT
+
+ensure_debugfs()
+{
+	if mount | grep -q 'on /sys/kernel/debug type debugfs'; then
+		return 0
+	fi
+
+	if ! mount -t debugfs none /sys/kernel/debug >/dev/null 2>&1; then
+		echo "SKIP: failed to mount debugfs"
+		exit "${ksft_skip}"
+	fi
+}
+
+ensure_netdevsim()
+{
+	if [ -w /sys/bus/netdevsim/new_device ]; then
+		return 0
+	fi
+
+	if ! modprobe netdevsim >/dev/null 2>&1; then
+		echo "SKIP: no netdevsim support"
+		exit "${ksft_skip}"
+	fi
+}
+
+create_nsim()
+{
+	local id="$1"
+	local ns="$2"
+	local addr="$3"
+	local dev
+
+	echo "${id}" | ip netns exec "${ns}" tee /sys/bus/netdevsim/new_device >/dev/null
+	udevadm settle
+
+	dev=$(ip netns exec "${ns}" ls /sys/bus/netdevsim/devices/netdevsim"${id}"/net)
+	ip -netns "${ns}" link set dev "${dev}" name "nsim${id}"
+	ip -netns "${ns}" addr add "${addr}/24" dev "nsim${id}"
+	ip -netns "${ns}" link set dev "nsim${id}" up
+
+	echo "nsim${id}"
+}
+
+link_nsim_peers()
+{
+	local srv_ifindex
+	local cli_ifindex
+
+	eval "exec {srv_fd}</var/run/netns/${SRV}"
+	eval "exec {cli_fd}</var/run/netns/${CLI}"
+
+	srv_ifindex=$(ip netns exec "${SRV}" cat /sys/class/net/"${srv_dev}"/ifindex)
+	cli_ifindex=$(ip netns exec "${CLI}" cat /sys/class/net/"${cli_dev}"/ifindex)
+
+	echo "${srv_fd}:${srv_ifindex} ${cli_fd}:${cli_ifindex}" > "${NSIM_SYS_LINK}"
+}
+
+wait_for_file()
+{
+	local path="$1"
+	local i
+
+	for i in $(seq 100); do
+		if [ -e "${path}" ]; then
+			return 0
+		fi
+		sleep 0.1
+	done
+
+	return 1
+}
+
+server_python='
+import array
+import fcntl
+import os
+import socket
+import struct
+import sys
+import time
+
+SO_MEMINFO = 55
+SK_MEMINFO_RMEM_ALLOC = 0
+TCP_MAXSEG = getattr(socket, "TCP_MAXSEG", 2)
+FIONREAD = 0x541B
+POLL_INTERVAL = 0.01
+POLL_TIMEOUT = 20.0
+
+(mode, host, port, warmup_rounds, warmup_len, queued_len, inflated_len,
+ ready_file, result_file) = sys.argv[1:]
+port = int(port)
+warmup_rounds = int(warmup_rounds)
+warmup_len = int(warmup_len)
+queued_len = int(queued_len)
+inflated_len = int(inflated_len)
+
+def queued_bytes(sock):
+    buf = array.array("I", [0])
+    fcntl.ioctl(sock.fileno(), FIONREAD, buf, True)
+    return buf[0]
+
+def wait_for_queued(sock, target):
+    deadline = time.time() + POLL_TIMEOUT
+    while time.time() < deadline:
+        if queued_bytes(sock) >= target:
+            return
+        time.sleep(POLL_INTERVAL)
+    raise SystemExit(f"timed out waiting for {target} queued bytes")
+
+def meminfo(sock):
+    raw = sock.getsockopt(socket.SOL_SOCKET, SO_MEMINFO, 9 * 4)
+    return struct.unpack("=9I", raw)
+
+def wait_for_growth(sock, idx, base):
+    deadline = time.time() + POLL_TIMEOUT
+    while time.time() < deadline:
+        cur = meminfo(sock)[idx]
+        if cur > base:
+            return cur
+        time.sleep(POLL_INTERVAL)
+    raise SystemExit(f"timed out waiting for SO_MEMINFO[{idx}] growth from {base}")
+
+def write_metric(path, value):
+    with open(path, "w", encoding="ascii") as fp:
+        fp.write(f"{value}\n")
+
+def recv_all(sock, total):
+    remaining = total
+    while remaining:
+        chunk = sock.recv(min(65536, remaining))
+        if not chunk:
+            raise SystemExit("unexpected EOF while draining receive data")
+        remaining -= len(chunk)
+
+listener = socket.socket(socket.AF_INET, socket.SOCK_STREAM)
+listener.setsockopt(socket.SOL_SOCKET, socket.SO_REUSEADDR, 1)
+listener.setsockopt(socket.IPPROTO_TCP, TCP_MAXSEG, 1000)
+listener.bind((host, port))
+listener.listen(1)
+conn, _ = listener.accept()
+
+for _ in range(warmup_rounds):
+    recv_all(conn, warmup_len)
+
+if mode == "rmem_alloc":
+    wait_for_queued(conn, queued_len)
+    base_metric = meminfo(conn)[SK_MEMINFO_RMEM_ALLOC]
+    write_metric(ready_file, base_metric)
+
+    recv_all(conn, queued_len)
+    wait_for_queued(conn, inflated_len)
+    grown_metric = meminfo(conn)[SK_MEMINFO_RMEM_ALLOC]
+    write_metric(result_file, grown_metric)
+elif mode == "rmem_alloc_warm":
+    wait_for_queued(conn, queued_len)
+    base_metric = meminfo(conn)[SK_MEMINFO_RMEM_ALLOC]
+    write_metric(ready_file, base_metric)
+
+    wait_for_queued(conn, queued_len + 1)
+    grown_metric = wait_for_growth(conn, SK_MEMINFO_RMEM_ALLOC, base_metric)
+    write_metric(result_file, grown_metric)
+elif mode == "rmem_alloc_growth":
+    # The growth cases compare against a live socket metric, so wait for
+    # observed growth instead of trusting one instantaneous post-queue sample.
+    wait_for_queued(conn, queued_len)
+    base_metric = meminfo(conn)[SK_MEMINFO_RMEM_ALLOC]
+    write_metric(ready_file, base_metric)
+
+    recv_all(conn, queued_len)
+    wait_for_queued(conn, inflated_len)
+    grown_metric = wait_for_growth(conn, SK_MEMINFO_RMEM_ALLOC, base_metric)
+    write_metric(result_file, grown_metric)
+else:
+    raise SystemExit(f"unknown mode: {mode}")
+'
+
+client_python='
+import os
+import socket
+import sys
+import time
+
+POLL_INTERVAL = 0.01
+POLL_TIMEOUT = 20.0
+
+host, port, warmup_rounds, warmup_len, queued_len, inflated_len, gate_file = sys.argv[1:]
+port = int(port)
+warmup_rounds = int(warmup_rounds)
+warmup_len = int(warmup_len)
+queued_len = int(queued_len)
+inflated_len = int(inflated_len)
+
+def send_all(sock, total):
+    payload = b"a" * min(total, 65536)
+    left = total
+    while left:
+        chunk = payload[: min(len(payload), left)]
+        sent = sock.send(chunk)
+        if sent <= 0:
+            raise SystemExit("short send")
+        left -= sent
+
+def wait_for_file(path):
+    deadline = time.time() + POLL_TIMEOUT
+    while time.time() < deadline:
+        if os.path.exists(path):
+            return
+        time.sleep(POLL_INTERVAL)
+    raise SystemExit(f"timed out waiting for {path}")
+
+cli = socket.socket(socket.AF_INET, socket.SOCK_STREAM)
+cli.setsockopt(socket.IPPROTO_TCP, socket.TCP_MAXSEG, 1000)
+cli.connect((host, port))
+for _ in range(warmup_rounds):
+    send_all(cli, warmup_len)
+send_all(cli, queued_len)
+wait_for_file(gate_file)
+send_all(cli, inflated_len)
+cli.close()
+'
+
+read_metric()
+{
+	local path="$1"
+	local value
+
+	if ! read -r value < "${path}"; then
+		echo "FAIL: unable to read metric from ${path}"
+		exit "${ksft_fail}"
+	fi
+
+	printf '%s\n' "${value}"
+}
+
+run_case()
+{
+	local case_id="$1"
+	local mode="$2"
+	local port="$3"
+	local warmups="$4"
+	local warmup_len="$5"
+	local queued_len="$6"
+	local inflated_len="$7"
+	local extra="$8"
+	local label="$9"
+	local ready_file="${stage_dir}/${case_id}.ready"
+	local result_file="${stage_dir}/${case_id}.result"
+	local gate_file="${stage_dir}/${case_id}.gate"
+
+	rm -f "${ready_file}" "${result_file}" "${gate_file}"
+	echo 0 > "${dfs_file}"
+
+	ip netns exec "${SRV}" python3 - "${mode}" "${SERVER_ADDR}" "${port}" \
+		"${warmups}" "${warmup_len}" "${queued_len}" "${inflated_len}" \
+		"${ready_file}" "${result_file}" <<PY &
+${server_python}
+PY
+	srv_pid=$!
+
+	wait_local_port_listen "${SRV}" "${port}" tcp
+
+	ip netns exec "${CLI}" python3 - "${SERVER_ADDR}" "${port}" \
+		"${warmups}" "${warmup_len}" "${queued_len}" "${inflated_len}" \
+		"${gate_file}" <<PY &
+${client_python}
+PY
+	cli_pid=$!
+
+	if ! wait_for_file "${ready_file}"; then
+		echo "FAIL: ${label}: ready marker did not appear"
+		exit "${ksft_fail}"
+	fi
+
+	echo "${extra}" > "${dfs_file}"
+	touch "${gate_file}"
+
+	wait "${cli_pid}"
+	cli_pid=
+	wait "${srv_pid}"
+	srv_pid=
+
+	CASE_BASE_METRIC=$(read_metric "${ready_file}")
+	CASE_FINAL_METRIC=$(read_metric "${result_file}")
+
+	echo "PASS: ${label}"
+}
+
+# This test only proves that injected truesize reaches socket memory
+# accounting. Packetdrill covers the sender-visible rwnd accept/drop logic.
+
+assert_no_growth()
+{
+	local label="$1"
+
+	if [ "${CASE_FINAL_METRIC}" -gt "${CASE_BASE_METRIC}" ]; then
+		echo "FAIL: ${label}: metric grew unexpectedly:" \
+		     "base=${CASE_BASE_METRIC}" \
+		     "after=${CASE_FINAL_METRIC}"
+		exit "${ksft_fail}"
+	fi
+}
+
+assert_growth()
+{
+	local label="$1"
+
+	if [ "${CASE_FINAL_METRIC}" -le "${CASE_BASE_METRIC}" ]; then
+		echo "FAIL: ${label}: metric did not grow:" \
+		     "base=${CASE_BASE_METRIC}" \
+		     "after=${CASE_FINAL_METRIC}"
+		exit "${ksft_fail}"
+	fi
+}
+
+ensure_debugfs
+ensure_netdevsim
+set +u
+setup_ns SRV CLI
+set -u
+
+srv_dev=$(create_nsim "${NSIM_SRV_ID}" "${SRV}" "${SERVER_ADDR}")
+cli_dev=$(create_nsim "${NSIM_CLI_ID}" "${CLI}" "${CLIENT_ADDR}")
+link_nsim_peers
+
+ip netns exec "${SRV}" sysctl -wq net.ipv4.tcp_moderate_rcvbuf=0
+
+stage_dir=$(mktemp -d)
+dfs_file="/sys/kernel/debug/netdevsim/netdevsim${NSIM_SRV_ID}/ports/0/rx_extra_truesize"
+
+run_case "rmem_noop" "rmem_alloc" "${RMEM_PORT}" 0 0 \
+	"${RMEM_QUEUED_LEN}" "${RMEM_INFLATED_LEN}" 0 \
+	"peer rx truesize zero no-op"
+assert_no_growth "peer rx truesize zero no-op"
+
+run_case "rmem_small" "rmem_alloc_growth" "${RMEM_PORT}" 0 0 \
+	"${RMEM_QUEUED_LEN}" "${RMEM_INFLATED_LEN}" "${RMEM_SMALL_EXTRA}" \
+	"peer rx truesize small rmem_alloc"
+assert_growth "peer rx truesize small rmem_alloc"
+small_delta=$((CASE_FINAL_METRIC - CASE_BASE_METRIC))
+
+run_case "rmem_large" "rmem_alloc_growth" "${RMEM_PORT}" 0 0 \
+	"${RMEM_QUEUED_LEN}" "${RMEM_INFLATED_LEN}" "${RMEM_LARGE_EXTRA}" \
+	"peer rx truesize large rmem_alloc"
+assert_growth "peer rx truesize large rmem_alloc"
+large_delta=$((CASE_FINAL_METRIC - CASE_BASE_METRIC))
+
+if [ "${large_delta}" -le "${small_delta}" ]; then
+	echo "FAIL: peer rx truesize stepped rmem_alloc:" \
+	     "small_delta=${small_delta}" \
+	     "large_delta=${large_delta}"
+	exit "${ksft_fail}"
+fi
+
+run_case "rmem_warm" "rmem_alloc_warm" "${WARM_PORT}" "${WARM_WARMUP_ROUNDS}" "${WARM_WARMUP_LEN}" \
+	"${WARM_QUEUED_LEN}" "${WARM_INFLATED_LEN}" "${WARM_EXTRA}" \
+	"peer rx truesize warm rmem_alloc"
+assert_growth "peer rx truesize warm rmem_alloc"
-- 
2.43.0

[PATCH net-next v2 14/14] netdevsim: release pinned PSP ext on drop paths

[atwellwea]

From: Wesley Atwell <atwellwea@gmail.com>

nsim_do_psp() can leave an extra extension reference pinned until
nsim_psp_handle_ext() reattaches it to a forwarded skb.

Route the drop paths through a common helper and release that extra
reference when __dev_forward_skb() fails and when start_xmit() drops the
skb before it reaches the peer RX path.

This is separate from the peer RX truesize test hook itself.

Signed-off-by: Wesley Atwell <atwellwea@gmail.com>
---
 drivers/net/netdevsim/netdev.c | 19 ++++++++++++++-----
 1 file changed, 14 insertions(+), 5 deletions(-)

diff --git a/drivers/net/netdevsim/netdev.c b/drivers/net/netdevsim/netdev.c
index 22238df79b6a..c22513c523d6 100644
--- a/drivers/net/netdevsim/netdev.c
+++ b/drivers/net/netdevsim/netdev.c
@@ -187,6 +187,15 @@ static int nsim_napi_rx(struct net_device *tx_dev, struct net_device *rx_dev,
 	return NET_RX_SUCCESS;
 }
 
+/* nsim_do_psp() pins an extra extension ref until nsim_psp_handle_ext()
+ * reattaches it to a forwarded skb.
+ */
+static void nsim_psp_ext_put(struct skb_ext *psp_ext)
+{
+	if (psp_ext)
+		__skb_ext_put(psp_ext);
+}
+
 static int nsim_forward_skb(struct net_device *tx_dev,
 			    struct net_device *rx_dev,
 			    struct sk_buff *skb,
@@ -196,8 +205,10 @@ static int nsim_forward_skb(struct net_device *tx_dev,
 	int ret;
 
 	ret = __dev_forward_skb(rx_dev, skb);
-	if (ret)
+	if (ret) {
+		nsim_psp_ext_put(psp_ext);
 		return ret;
+	}
 
 	nsim_psp_handle_ext(skb, psp_ext);
 
@@ -278,11 +289,8 @@ static netdev_tx_t nsim_start_xmit(struct sk_buff *skb, struct net_device *dev)
 		 * the synthetic cost; sender-side skb accounting stays put.
 		 */
 		nskb = skb_clone(skb, GFP_ATOMIC);
-		if (!nskb) {
-			if (psp_ext)
-				__skb_ext_put(psp_ext);
+		if (!nskb)
 			goto out_drop_free;
-		}
 
 		consume_skb(skb);
 		skb = nskb;
@@ -303,6 +311,7 @@ static netdev_tx_t nsim_start_xmit(struct sk_buff *skb, struct net_device *dev)
 out_drop_any:
 	dr = SKB_DROP_REASON_NOT_SPECIFIED;
 out_drop_free:
+	nsim_psp_ext_put(psp_ext);
 	kfree_skb_reason(skb, dr);
 out_drop_cnt:
 	rcu_read_unlock();
-- 
2.43.0

Re: [PATCH net-next v2 13/14] netdevsim: add peer RX truesize support for selftests

[Jakub Kicinski]

On Sat, 14 Mar 2026 14:13:47 -0600 atwellwea@gmail.com wrote:
> diff --git a/tools/testing/selftests/drivers/net/netdevsim/Makefile b/tools/testing/selftests/drivers/net/netdevsim/Makefile
> index 1a228c5430f5..9e9e48d5913b 100644
> --- a/tools/testing/selftests/drivers/net/netdevsim/Makefile
> +++ b/tools/testing/selftests/drivers/net/netdevsim/Makefile
> @@ -14,6 +14,7 @@ TEST_PROGS := \
>  	macsec-offload.sh \
>  	nexthop.sh \
>  	peer.sh \
> +	peer-rx-truesize.sh \
>  	psample.sh \
>  	tc-mq-visibility.sh \
>  	udp_tunnel_nic.sh \

I think our order checker thinks that is not alphabetical.
You can find the scripts here:
https://github.com/linux-netdev/nipa/tree/main/tests/patch/check_selftest

It'd be good if you can try the "running locally" section from README

Re: [PATCH net-next v2 12/14] tun/selftests: add RX truesize injection for TCP window tests

[Jakub Kicinski]

On Sat, 14 Mar 2026 14:13:46 -0600 atwellwea@gmail.com wrote:
> Add a test-only TUN ioctl that inflates RX skb->truesize, plus the
> packetdrill-side helper needed to drive that ioctl through packetdrill's
> own TUN queue file descriptor.
> 
> Use that plumbing to cover the receive-window regressions where
> scaling_ratio drifts after advertisement, alongside the baseline too-big
> packetdrill cases that exercise the same sender-visible rwnd accounting
> from the non-injected path.

You missed adding the tun program to the Makefile.

  tcp-rcv-neg-window-truesize-pkt                                               
  tcp-rcv-toobig-default-truesize-pkt                                           
  tcp-rcv-wnd-shrink-allowed-truesize-pkt                                       
                                                                               
Fail with:                                                            

  sh: line 1: ../tun: No such file or directory                                 
  tcp_rcv_*_truesize.pkt:*: error executing `../tun --set-rx-truesize tun0 6553\
6` command: non-zero status 127        

https://netdev.bots.linux.dev/contest.html?pw-n=0&branch=net-next-2026-03-15--00-00&pw-n=0&pass=0

Re: [PATCH net-next v2 00/14] tcp: preserve receive-window accounting across ratio drift

[Jakub Kicinski]

On Sat, 14 Mar 2026 14:13:34 -0600 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, even if
> later receive-side truesize inflation lowers scaling_ratio or the live
> receive window retracts below the largest right edge already exposed to the
> sender.

Please wait until at least Monday before reposting this.
LLM-pocalypse is hitting us hard

Re: [PATCH net-next v2 05/14] tcp: grow rcvbuf to back scaled-window quantization slack

[Paolo Abeni]

On 3/14/26 9:13 PM, atwellwea@gmail.com wrote:
> From: Wesley Atwell <atwellwea@gmail.com>
> 
> Teach TCP to grow sk_rcvbuf when scale rounding would otherwise expose
> more sender-visible window than the current hard receive-memory backing
> can cover.
> 
> The new helper keeps backlog and memory-pressure limits in the same
> units as the rest of the receive path, while __tcp_select_window()
> backs any rounding slack before advertising it.
> 
> Signed-off-by: Wesley Atwell <atwellwea@gmail.com>
> ---
>  include/net/tcp.h     | 12 ++++++++++++
>  net/ipv4/tcp_input.c  | 36 ++++++++++++++++++++++++++++++++++--
>  net/ipv4/tcp_output.c | 15 +++++++++++++--
>  3 files changed, 59 insertions(+), 4 deletions(-)
> 
> diff --git a/include/net/tcp.h b/include/net/tcp.h
> index fc22ab6b80d5..5b479ad44f89 100644
> --- a/include/net/tcp.h
> +++ b/include/net/tcp.h
> @@ -397,6 +397,7 @@ 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);
> @@ -1844,6 +1845,17 @@ static inline int tcp_rwnd_avail(const struct sock *sk)
>  	return tcp_rmem_avail(sk) - READ_ONCE(sk->sk_backlog.len);
>  }
>  
> +/* Passive children clone the listener's sk_socket until accept() grafts
> + * their own struct socket, so only sockets that point back to themselves
> + * should autotune receive-buffer backing.
> + */
> +static inline bool tcp_rcvbuf_grow_allowed(const struct sock *sk)
> +{
> +	struct socket *sock = READ_ONCE(sk->sk_socket);
> +
> +	return sock && READ_ONCE(sock->sk) == sk;

This is executed under the sk socket lock, ONCE annotation not needed.

> +}
> +
>  /* Note: caller must be prepared to deal with negative returns */
>  static inline int tcp_space(const struct sock *sk)
>  {
> diff --git a/net/ipv4/tcp_input.c b/net/ipv4/tcp_input.c
> index 352f814a4ff6..32256519a085 100644
> --- a/net/ipv4/tcp_input.c
> +++ b/net/ipv4/tcp_input.c
> @@ -774,6 +774,38 @@ 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 sk_rmem_alloc while preserving sender-visible headroom
> + * already consumed by sk_backlog.len.
> + */
> +bool tcp_try_grow_rcvbuf(struct sock *sk, int needed)
> +{
> +	struct net *net = sock_net(sk);
> +	int backlog;
> +	int rmem2;
> +	int target;
> +
> +	needed = max(needed, 0);
> +	backlog = READ_ONCE(sk->sk_backlog.len);
> +	target = tcp_rmem_used(sk) + backlog + 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);

Same here, and more cases below.

/P

Re: [PATCH net-next v2 00/14] tcp: preserve receive-window accounting across ratio drift

[Paolo Abeni]

On 3/14/26 9:13 PM, atwellwea@gmail.com wrote:
> From: Wesley Atwell <atwellwea@gmail.com>
> 
> This series keeps sender-visible TCP receive-window accounting tied to the
> scaling basis that was in force when the window was advertised, even if
> later receive-side truesize inflation lowers scaling_ratio or the live
> receive window retracts below the largest right edge already exposed to the
> sender.
> 
> After the receive-window retraction changes, the receive path needs to keep
> track of two related pieces of sender-visible state:
> 
>   1. the live advertised receive window
>   2. the maximum advertised right edge and the basis it was exposed with
> 
> This repost snapshots both, uses them to repair receive-buffer backing when
> ratio drift would otherwise strand sender-visible space, extends
> TCP_REPAIR_WINDOW so repair/restore can round-trip the new state, and adds
> truesize-drift coverage through TUN packetdrill tests and netdevsim-based
> selftests.

The series is IMHO significantly not trivial. Can the end-user meet the
relevant condition in practice? How? What is the net benefit in
practice? Is that observable under usual conditions or require
exceptional circumstances?

I think we need a strong motivation to merge this kind of changes.

/P

Re: [PATCH net-next v2 05/14] tcp: grow rcvbuf to back scaled-window quantization slack

[Paolo Abeni]

On 3/14/26 9:13 PM, atwellwea@gmail.com wrote:
> From: Wesley Atwell <atwellwea@gmail.com>
> 
> Teach TCP to grow sk_rcvbuf when scale rounding would otherwise expose
> more sender-visible window than the current hard receive-memory backing
> can cover.
> 
> The new helper keeps backlog and memory-pressure limits in the same
> units as the rest of the receive path, while __tcp_select_window()
> backs any rounding slack before advertising it.
> 
> Signed-off-by: Wesley Atwell <atwellwea@gmail.com>
> ---
>  include/net/tcp.h     | 12 ++++++++++++
>  net/ipv4/tcp_input.c  | 36 ++++++++++++++++++++++++++++++++++--
>  net/ipv4/tcp_output.c | 15 +++++++++++++--
>  3 files changed, 59 insertions(+), 4 deletions(-)
> 
> diff --git a/include/net/tcp.h b/include/net/tcp.h
> index fc22ab6b80d5..5b479ad44f89 100644
> --- a/include/net/tcp.h
> +++ b/include/net/tcp.h
> @@ -397,6 +397,7 @@ 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);
> @@ -1844,6 +1845,17 @@ static inline int tcp_rwnd_avail(const struct sock *sk)
>  	return tcp_rmem_avail(sk) - READ_ONCE(sk->sk_backlog.len);
>  }
>  
> +/* Passive children clone the listener's sk_socket until accept() grafts
> + * their own struct socket, 

AFAICS, the above statement is false, see sk_set_socket() in sk_clone()

> so only sockets that point back to themselves
> + * should autotune receive-buffer backing.
> + */
> +static inline bool tcp_rcvbuf_grow_allowed(const struct sock *sk)
> +{
> +	struct socket *sock = READ_ONCE(sk->sk_socket);

You can just check `sk->sk_socket`. Also you could re-use this helper in
tcp_data_queue_ofo().

/P

Re: [PATCH net-next v2 05/14] tcp: grow rcvbuf to back scaled-window quantization slack

[Paolo Abeni]

On 3/14/26 9:13 PM, atwellwea@gmail.com wrote:
> diff --git a/net/ipv4/tcp_output.c b/net/ipv4/tcp_output.c
> index 57a2a6daaad3..53781cf591d2 100644
> --- a/net/ipv4/tcp_output.c
> +++ b/net/ipv4/tcp_output.c
> @@ -3375,13 +3375,24 @@ 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_rcvbuf_grow_allowed(sk) ||
> +		     !tcp_try_grow_rcvbuf(sk, tcp_space_from_win(sk, window))))
> +			window = round_down(free_space, rcv_wscale);

It looks like this can cause the advertised window to shrink even if we
are in the 'do not allow window to shrink' branch.

Also why the other branch (shrinking allowed) is not touched?

/P

Re: [PATCH net-next v2 07/14] tcp: honor the maximum advertised window after live retraction

[Paolo Abeni]

On 3/14/26 9:13 PM, atwellwea@gmail.com wrote:
> +/* Sender-visible window rescue does not relax hard receive-memory admission.
> + * If growth did not make room, fall back to the established prune/collapse
> + * path.
> + */
>  static int tcp_try_rmem_schedule(struct sock *sk, const struct sk_buff *skb,
>  				 unsigned int size)
>  {
> -	if (!tcp_can_ingest(sk, skb) ||
> -	    !sk_rmem_schedule(sk, skb, size)) {
> +	bool can_ingest = tcp_can_ingest(sk, skb);
> +	bool scheduled = can_ingest && sk_rmem_schedule(sk, skb, size);
> +
> +	if (!scheduled) {
> +		int pruned = tcp_prune_queue(sk, skb);
>  
> -		if (tcp_prune_queue(sk, skb) < 0)
> +		if (pruned < 0)
>  			return -1;
>  
>  		while (!sk_rmem_schedule(sk, skb, size)) {
> -			if (!tcp_prune_ofo_queue(sk, skb))
> +			bool pruned_ofo = tcp_prune_ofo_queue(sk, skb);
> +
> +			if (!pruned_ofo)
>  				return -1;
>  		}
>  	}

The above chunk is AFAICS pure noise. Please have a more careful local
review of this series before any next revision.

/P

Re: [PATCH net-next v2 00/14] tcp: preserve receive-window accounting across ratio drift

[Paolo Abeni]

Hi,

On 3/16/26 6:29 PM, Wesley Atwell wrote:
> The strongest real anchor here is the already documented regression
> around sender-visible rwnd diverging from hard receive-memory backing,
> rather than a general receive-accounting cleanup.

I likely missed some of the prior discussion. Could you please share a
pointer/link to the mentioned regression report?

When posting on netdev please:
- use plaintext only messages
- avoid top-posting

Thanks,

Paolo

Re: [PATCH net-next v2 00/14] tcp: preserve receive-window accounting across ratio drift

[Wesley Atwell]

Hi Paolo,

The link I meant is the netdev regression discussion that led to:

026dfef287c0 ("tcp: give up on stronger sk_rcvbuf checks (for now)")

The report/discussion thread is here:

https://lore.kernel.org/20260225122355.585fd57b@kernel.org

The revert posting itself is here:

https://patch.msgid.link/20260227003359.2391017-1-kuba@kernel.org

I should not have implied that there was a separate prior
regression report beyond that thread.

Thanks for the note on formatting. I will keep follow-ups in plaintext
and avoid top-posting.

Thanks,
Wesley